Files
RLCR-2p5x-priority-bestrewa…/trainer_state.json
ModelHub XC 09b425d9b6 初始化项目,由ModelHub XC社区提供模型
Model: hector-gr/RLCR-2p5x-priority-bestreward-math
Source: Original Platform
2026-07-22 05:46:12 +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": 0.49662184969039436,
"calibration/batch_distribution_entropy": 0.24855682571268609,
"calibration/batch_entropy_100bins": 0.34211678603258494,
"calibration/batch_entropy_10bins": 0.24855682571268609,
"calibration/batch_entropy_50bins": 0.39298323694966103,
"calibration/batch_uniqueness": 0.4957406739381661,
"calibration/confidence_entropy": 0.21799519909421344,
"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": 0.24855682571268609,
"calibration/distribution_entropy_100": 0.34211678603258494,
"calibration/ece": 0.4413155346150276,
"calibration/mean_confidence": 0.922292316091292,
"calibration/unique_confidence_per_question": 0.03385416666666667,
"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": 0.01961805555555556,
"completions/max_length": 4037.2,
"completions/max_terminated_length": 4037.2,
"completions/mean_length": 492.128466796875,
"completions/mean_terminated_length": 501.96421508789064,
"completions/min_length": 0.0,
"completions/min_terminated_length": 2.6,
"epoch": 0.01201923076923077,
"grad_norm": 0.004471676889806986,
"learning_rate": 1.201923076923077e-07,
"loss": 0.0078,
"num_tokens": 8748584.0,
"reward": 0.5809432864189148,
"reward_std": 0.528658103942871,
"rewards/accuracy_reward": 0.25043402910232543,
"rewards/brier_reward": 0.3032692611217499,
"rewards/confidence_one_or_zero": 0.3365451395511627,
"rewards/format_reward": 0.6081597208976746,
"rewards/mean_confidence_reward": 0.8397148966789245,
"sampling/batch_mean_priority_error": 0.020286041666666678,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 1.0,
"sampling/error_ema_max": 0.0528124988079071,
"sampling/error_ema_mean": 1.4395399466593517e-05,
"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": 0.298681640625,
"signal/accuracy_reward/group_std_mean": 0.3567140340805054,
"signal/accuracy_reward/group_zero_std_frac": 0.11944444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1493408203125,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.1493408203125,
"signal/advantage_abs_mean": 0.4503032624721527,
"signal/advantage_pre_scale_abs_mean": 0.4503032624721527,
"signal/advantage_pre_scale_std": 0.5379110932350158,
"signal/advantage_std": 0.5379110932350158,
"signal/brier_reward/centered_abs_mean": 0.3076389491558075,
"signal/brier_reward/group_std_mean": 0.3620465755462646,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.15381947457790374,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.15381947457790374,
"signal/confidence_one_or_zero/centered_abs_mean": 0.40524630546569823,
"signal/confidence_one_or_zero/group_std_mean": 0.4523835420608521,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.00555555559694767,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.0524628275306895e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.0524628275306895e-06,
"signal/format_reward/centered_abs_mean": 0.4301974833011627,
"signal/format_reward/group_std_mean": 0.4691725194454193,
"signal/format_reward/group_zero_std_frac": 0.0,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.21509874165058135,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.21509874165058135,
"signal/mean_confidence_reward/centered_abs_mean": 0.157260724902153,
"signal/mean_confidence_reward/group_std_mean": 0.23579104840755463,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.572607152411365e-06,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 1.572607152411365e-06,
"step": 5
},
{
"calibration/aurc": 0.6212086769784675,
"calibration/batch_distribution_entropy": 0.2554635874966734,
"calibration/batch_entropy_100bins": 0.34783759217112353,
"calibration/batch_entropy_10bins": 0.2554635874966734,
"calibration/batch_entropy_50bins": 0.39758666158825434,
"calibration/batch_uniqueness": 0.5003519549953523,
"calibration/confidence_entropy": 0.20785773862853948,
"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": 0.2554635874966734,
"calibration/distribution_entropy_100": 0.34783759217112353,
"calibration/ece": 0.5789486640908559,
"calibration/mean_confidence": 0.9201863305340018,
"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": 0.02387152777777777,
"completions/max_length": 4004.0,
"completions/max_terminated_length": 4004.0,
"completions/mean_length": 515.1827270507813,
"completions/mean_terminated_length": 527.7875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 2.0,
"epoch": 0.02403846153846154,
"grad_norm": 0.004891601856797934,
"learning_rate": 2.403846153846154e-07,
"loss": 0.0047,
"num_tokens": 17765505.0,
"reward": 0.5613437056541443,
"reward_std": 0.5221718430519104,
"rewards/accuracy_reward": 0.2384548604488373,
"rewards/brier_reward": 0.2904594004154205,
"rewards/confidence_one_or_zero": 0.3299479126930237,
"rewards/format_reward": 0.59375,
"rewards/mean_confidence_reward": 0.8279155731201172,
"sampling/batch_mean_priority_error": 0.02943812500000001,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 1.0,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 4.464865050977096e-05,
"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": 0.28582356572151185,
"signal/accuracy_reward/group_std_mean": 0.34282330274581907,
"signal/accuracy_reward/group_zero_std_frac": 0.1472222238779068,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.14291178286075593,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.14291178286075593,
"signal/advantage_abs_mean": 0.44686338901519773,
"signal/advantage_pre_scale_abs_mean": 0.44686338901519773,
"signal/advantage_pre_scale_std": 0.5340252995491028,
"signal/advantage_std": 0.5340252995491028,
"signal/brier_reward/centered_abs_mean": 0.2982248544692993,
"signal/brier_reward/group_std_mean": 0.3519687294960022,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.14911242723464965,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.14911242723464965,
"signal/confidence_one_or_zero/centered_abs_mean": 0.39952799677848816,
"signal/confidence_one_or_zero/group_std_mean": 0.4472978889942169,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.01111111119389534,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.995279803348239e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.995279803348239e-06,
"signal/format_reward/centered_abs_mean": 0.43882378935813904,
"signal/format_reward/group_std_mean": 0.4742614209651947,
"signal/format_reward/group_zero_std_frac": 0.0,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.21941189467906952,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.21941189467906952,
"signal/mean_confidence_reward/centered_abs_mean": 0.16937248408794403,
"signal/mean_confidence_reward/group_std_mean": 0.24667055308818817,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.6937246755333036e-06,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 1.6937246755333036e-06,
"step": 10
},
{
"calibration/aurc": 0.5686256298557191,
"calibration/batch_distribution_entropy": 0.2613920164321266,
"calibration/batch_entropy_100bins": 0.34257888465762576,
"calibration/batch_entropy_10bins": 0.2613920164321266,
"calibration/batch_entropy_50bins": 0.3977664628482245,
"calibration/batch_uniqueness": 0.4988012462838534,
"calibration/confidence_entropy": 0.21749303316202773,
"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": 0.2613920164321266,
"calibration/distribution_entropy_100": 0.34257888465762576,
"calibration/ece": 0.5288600027898627,
"calibration/mean_confidence": 0.9190490067386252,
"calibration/unique_confidence_per_question": 0.03333333333333333,
"calibration/unique_confidences": 12.8,
"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": 0.019010416666666675,
"completions/max_length": 3969.8,
"completions/max_terminated_length": 3969.8,
"completions/mean_length": 501.7993957519531,
"completions/mean_terminated_length": 511.546826171875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 11.2,
"epoch": 0.036057692307692304,
"grad_norm": 0.005199067294597626,
"learning_rate": 3.6057692307692306e-07,
"loss": 0.0017,
"num_tokens": 26654394.0,
"reward": 0.6018155813217163,
"reward_std": 0.5264561772346497,
"rewards/accuracy_reward": 0.2533854156732559,
"rewards/brier_reward": 0.3099447190761566,
"rewards/confidence_one_or_zero": 0.32890625596046447,
"rewards/format_reward": 0.6402777791023254,
"rewards/mean_confidence_reward": 0.8362808942794799,
"sampling/batch_mean_priority_error": 0.019512222916666676,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 1.0,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 7.241204875754192e-05,
"sampling/priority_kl": 0.029195376113057135,
"sampling/priority_scale": 419434.3227580553,
"sampling/prob_entropy": 10.279756927490235,
"sampling/prob_max": 3.4387885534670204e-05,
"sampling/prob_min": 5.380612662975182e-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": 0.30558268427848817,
"signal/accuracy_reward/group_std_mean": 0.36401288509368895,
"signal/accuracy_reward/group_zero_std_frac": 0.11388889104127883,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.15279134213924409,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.15279134213924409,
"signal/advantage_abs_mean": 0.44527676701545715,
"signal/advantage_pre_scale_abs_mean": 0.44527676701545715,
"signal/advantage_pre_scale_std": 0.5341985344886779,
"signal/advantage_std": 0.5341985344886779,
"signal/brier_reward/centered_abs_mean": 0.31033373475074766,
"signal/brier_reward/group_std_mean": 0.36489441990852356,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.15516686737537383,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.15516686737537383,
"signal/confidence_one_or_zero/centered_abs_mean": 0.40199110507965086,
"signal/confidence_one_or_zero/group_std_mean": 0.45113515853881836,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.019910829811124e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.019910829811124e-06,
"signal/format_reward/centered_abs_mean": 0.4189778625965118,
"signal/format_reward/group_std_mean": 0.46179532408714297,
"signal/format_reward/group_zero_std_frac": 0.0,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.2094889312982559,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.2094889312982559,
"signal/mean_confidence_reward/centered_abs_mean": 0.16203747689723969,
"signal/mean_confidence_reward/group_std_mean": 0.2359348714351654,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.6203747463805485e-06,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 1.6203747463805485e-06,
"step": 15
},
{
"calibration/aurc": 0.5781377068007135,
"calibration/batch_distribution_entropy": 0.30133848780637856,
"calibration/batch_entropy_100bins": 0.3604450194123049,
"calibration/batch_entropy_10bins": 0.30133848780637856,
"calibration/batch_entropy_50bins": 0.4156054714933938,
"calibration/batch_uniqueness": 0.5137717719347905,
"calibration/confidence_entropy": 0.23150646493074922,
"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": 0.30133848780637856,
"calibration/distribution_entropy_100": 0.3604450194123049,
"calibration/ece": 0.5328683362672083,
"calibration/mean_confidence": 0.9074953262400243,
"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": 0.01953124999999998,
"completions/max_length": 4071.4,
"completions/max_terminated_length": 4071.4,
"completions/mean_length": 484.62213745117185,
"completions/mean_terminated_length": 494.3174987792969,
"completions/min_length": 0.0,
"completions/min_terminated_length": 10.2,
"epoch": 0.04807692307692308,
"grad_norm": 0.05662711709737778,
"learning_rate": 4.807692307692308e-07,
"loss": 0.0053,
"num_tokens": 35325977.0,
"reward": 0.6991640329360962,
"reward_std": 0.5071477174758912,
"rewards/accuracy_reward": 0.29210069179534914,
"rewards/brier_reward": 0.36349549889564514,
"rewards/confidence_one_or_zero": 0.32812500596046446,
"rewards/format_reward": 0.7427083373069763,
"rewards/mean_confidence_reward": 0.8493289351463318,
"sampling/batch_mean_priority_error": 0.011812916666667365,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9972222222222221,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 9.249055292457343e-05,
"sampling/priority_kl": 0.03000004291534424,
"sampling/priority_scale": 4.116325568780303,
"sampling/prob_entropy": 10.27895278930664,
"sampling/prob_max": 3.4690356551436706e-05,
"sampling/prob_min": 2.692946486604342e-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": 0.32266167402267454,
"signal/accuracy_reward/group_std_mean": 0.3808844566345215,
"signal/accuracy_reward/group_zero_std_frac": 0.0777777798473835,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.16133083701133727,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.16133083701133727,
"signal/advantage_abs_mean": 0.4216520845890045,
"signal/advantage_pre_scale_abs_mean": 0.4216520845890045,
"signal/advantage_pre_scale_std": 0.5141208887100219,
"signal/advantage_std": 0.5141208887100219,
"signal/brier_reward/centered_abs_mean": 0.32142401337623594,
"signal/brier_reward/group_std_mean": 0.373612767457962,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.16071200668811797,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.16071200668811797,
"signal/confidence_one_or_zero/centered_abs_mean": 0.3991102397441864,
"signal/confidence_one_or_zero/group_std_mean": 0.4487784802913666,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.9911023122840564e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.9911023122840564e-06,
"signal/format_reward/centered_abs_mean": 0.34194878935813905,
"signal/format_reward/group_std_mean": 0.4118743896484375,
"signal/format_reward/group_zero_std_frac": 0.002777777798473835,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.17097439467906952,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.17097439467906952,
"signal/mean_confidence_reward/centered_abs_mean": 0.14605502486228944,
"signal/mean_confidence_reward/group_std_mean": 0.21780959367752076,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.4605501519326936e-06,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 1.4605501519326936e-06,
"step": 20
},
{
"calibration/aurc": 0.6459646670099846,
"calibration/batch_distribution_entropy": 0.305665693826494,
"calibration/batch_entropy_100bins": 0.36347840431491163,
"calibration/batch_entropy_10bins": 0.305665693826494,
"calibration/batch_entropy_50bins": 0.4226184384984603,
"calibration/batch_uniqueness": 0.5333775639553542,
"calibration/confidence_entropy": 0.23703443276990782,
"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": 0.305665693826494,
"calibration/distribution_entropy_100": 0.36347840431491163,
"calibration/ece": 0.5938233061064975,
"calibration/mean_confidence": 0.9098523283644399,
"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": 0.018055555555555557,
"completions/max_length": 4032.4,
"completions/max_terminated_length": 4032.4,
"completions/mean_length": 451.4315185546875,
"completions/mean_terminated_length": 459.79645385742185,
"completions/min_length": 0.0,
"completions/min_terminated_length": 26.4,
"epoch": 0.06009615384615385,
"grad_norm": 0.041369203478097916,
"learning_rate": 6.009615384615385e-07,
"loss": -0.0065,
"num_tokens": 43588260.0,
"reward": 0.815249752998352,
"reward_std": 0.45611422061920165,
"rewards/accuracy_reward": 0.33463541865348817,
"rewards/brier_reward": 0.42292370200157164,
"rewards/confidence_one_or_zero": 0.31111111044883727,
"rewards/format_reward": 0.8729166746139526,
"rewards/mean_confidence_reward": 0.8755728960037231,
"sampling/batch_mean_priority_error": 0.01072958402777778,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9916666666666668,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00010534993198234588,
"sampling/priority_kl": 0.030000149086117745,
"sampling/priority_scale": 3.0989843607880174,
"sampling/prob_entropy": 10.278955841064453,
"sampling/prob_max": 3.494143529678695e-05,
"sampling/prob_min": 2.578220801296993e-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": 0.3340549051761627,
"signal/accuracy_reward/group_std_mean": 0.3912248969078064,
"signal/accuracy_reward/group_zero_std_frac": 0.06388889048248529,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.16702745258808135,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.16702745258808135,
"signal/advantage_abs_mean": 0.3729071855545044,
"signal/advantage_pre_scale_abs_mean": 0.3729071855545044,
"signal/advantage_pre_scale_std": 0.4633388459682465,
"signal/advantage_std": 0.4633388459682465,
"signal/brier_reward/centered_abs_mean": 0.3166819095611572,
"signal/brier_reward/group_std_mean": 0.3681463062763214,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.1583409547805786,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.1583409547805786,
"signal/confidence_one_or_zero/centered_abs_mean": 0.3850585877895355,
"signal/confidence_one_or_zero/group_std_mean": 0.43928542137146,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.00555555559694767,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.85058583560749e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.85058583560749e-06,
"signal/format_reward/centered_abs_mean": 0.20249566435813904,
"signal/format_reward/group_std_mean": 0.3022570312023163,
"signal/format_reward/group_zero_std_frac": 0.058333334140479566,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.10124783217906952,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.10124783217906952,
"signal/mean_confidence_reward/centered_abs_mean": 0.11333362460136413,
"signal/mean_confidence_reward/group_std_mean": 0.17320298850536348,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.13333621811762e-06,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 1.13333621811762e-06,
"step": 25
},
{
"calibration/aurc": 0.5997521391931024,
"calibration/batch_distribution_entropy": 0.30450032721554765,
"calibration/batch_entropy_100bins": 0.3638448903661742,
"calibration/batch_entropy_10bins": 0.30450032721554765,
"calibration/batch_entropy_50bins": 0.4255578176283213,
"calibration/batch_uniqueness": 0.5224223472424432,
"calibration/confidence_entropy": 0.2562116276037758,
"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": 0.30450032721554765,
"calibration/distribution_entropy_100": 0.3638448903661742,
"calibration/ece": 0.5441324652383763,
"calibration/mean_confidence": 0.9045782350307932,
"calibration/unique_confidence_per_question": 0.03385416666666667,
"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": 0.012413194444444442,
"completions/max_length": 3953.0,
"completions/max_terminated_length": 3953.0,
"completions/mean_length": 454.72821655273435,
"completions/mean_terminated_length": 460.4580322265625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 63.8,
"epoch": 0.07211538461538461,
"grad_norm": 0.002127213403582573,
"learning_rate": 7.211538461538461e-07,
"loss": -0.015,
"num_tokens": 51927145.0,
"reward": 0.9009054064750671,
"reward_std": 0.4041502892971039,
"rewards/accuracy_reward": 0.38237847089767457,
"rewards/brier_reward": 0.4791316449642181,
"rewards/confidence_one_or_zero": 0.262934023141861,
"rewards/format_reward": 0.9402777791023255,
"rewards/mean_confidence_reward": 0.8828180313110352,
"sampling/batch_mean_priority_error": 0.00947305555555556,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.986111111111111,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00011624346661847085,
"sampling/priority_kl": 0.030000174790620802,
"sampling/priority_scale": 2.581620621774346,
"sampling/prob_entropy": 10.278950500488282,
"sampling/prob_max": 3.516698052408174e-05,
"sampling/prob_min": 3.98486627091188e-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": 0.3201768696308136,
"signal/accuracy_reward/group_std_mean": 0.3793734133243561,
"signal/accuracy_reward/group_zero_std_frac": 0.07777777910232545,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1600884348154068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.1600884348154068,
"signal/advantage_abs_mean": 0.32962233424186704,
"signal/advantage_pre_scale_abs_mean": 0.32962233424186704,
"signal/advantage_pre_scale_std": 0.41350186467170713,
"signal/advantage_std": 0.41350186467170713,
"signal/brier_reward/centered_abs_mean": 0.2959588646888733,
"signal/brier_reward/group_std_mean": 0.34889656901359556,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.14797943234443664,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.14797943234443664,
"signal/confidence_one_or_zero/centered_abs_mean": 0.3444932758808136,
"signal/confidence_one_or_zero/group_std_mean": 0.41049314141273496,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.019444444589316844,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.4449325994501125e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.4449325994501125e-06,
"signal/format_reward/centered_abs_mean": 0.10245225727558135,
"signal/format_reward/group_std_mean": 0.1856452167034149,
"signal/format_reward/group_zero_std_frac": 0.2833333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.05122612863779068,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.05122612863779068,
"signal/mean_confidence_reward/centered_abs_mean": 0.09547019451856613,
"signal/mean_confidence_reward/group_std_mean": 0.14561118185520172,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 9.54701920363732e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 9.54701920363732e-07,
"step": 30
},
{
"calibration/aurc": 0.4745980647435296,
"calibration/batch_distribution_entropy": 0.33344258957053646,
"calibration/batch_entropy_100bins": 0.3649651982296054,
"calibration/batch_entropy_10bins": 0.33344258957053646,
"calibration/batch_entropy_50bins": 0.42468040346264324,
"calibration/batch_uniqueness": 0.5059170401082715,
"calibration/confidence_entropy": 0.27285792284673105,
"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": 0.33344258957053646,
"calibration/distribution_entropy_100": 0.3649651982296054,
"calibration/ece": 0.4374113852388136,
"calibration/mean_confidence": 0.8955356761082479,
"calibration/unique_confidence_per_question": 0.03854166666666666,
"calibration/unique_confidences": 14.8,
"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": 0.01258680555555558,
"completions/max_length": 3892.0,
"completions/max_terminated_length": 3892.0,
"completions/mean_length": 487.2180541992187,
"completions/mean_terminated_length": 493.463916015625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 74.0,
"epoch": 0.08413461538461539,
"grad_norm": 0.0013091641012579203,
"learning_rate": 8.41346153846154e-07,
"loss": -0.0115,
"num_tokens": 60623225.0,
"reward": 0.9715142011642456,
"reward_std": 0.367927622795105,
"rewards/accuracy_reward": 0.43394096493721007,
"rewards/brier_reward": 0.5377981483936309,
"rewards/confidence_one_or_zero": 0.2161458373069763,
"rewards/format_reward": 0.9712673544883728,
"rewards/mean_confidence_reward": 0.8788381457328797,
"sampling/batch_mean_priority_error": 0.00579201388888889,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.975,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00012595559237524868,
"sampling/priority_kl": 0.029999414458870888,
"sampling/priority_scale": 2.2547397849150004,
"sampling/prob_entropy": 10.278950309753418,
"sampling/prob_max": 3.5374295111978424e-05,
"sampling/prob_min": 5.2708895964315164e-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": 0.30965169072151183,
"signal/accuracy_reward/group_std_mean": 0.3712247610092163,
"signal/accuracy_reward/group_zero_std_frac": 0.08611111361533404,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.15482584536075591,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.15482584536075591,
"signal/advantage_abs_mean": 0.29973416328430175,
"signal/advantage_pre_scale_abs_mean": 0.29973416328430175,
"signal/advantage_pre_scale_std": 0.37970858812332153,
"signal/advantage_std": 0.37970858812332153,
"signal/brier_reward/centered_abs_mean": 0.2735893577337265,
"signal/brier_reward/group_std_mean": 0.328358393907547,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.13679467886686325,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.13679467886686325,
"signal/confidence_one_or_zero/centered_abs_mean": 0.3033962666988373,
"signal/confidence_one_or_zero/group_std_mean": 0.38285431265830994,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.01666666679084301,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.0339625936903757e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.0339625936903757e-06,
"signal/format_reward/centered_abs_mean": 0.05086263045668602,
"signal/format_reward/group_std_mean": 0.10426565259695053,
"signal/format_reward/group_zero_std_frac": 0.544444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.02543131522834301,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.02543131522834301,
"signal/mean_confidence_reward/centered_abs_mean": 0.08879880607128143,
"signal/mean_confidence_reward/group_std_mean": 0.13397251665592194,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.879880056156253e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.879880056156253e-07,
"step": 35
},
{
"calibration/aurc": 0.4944071264636566,
"calibration/batch_distribution_entropy": 0.37269805542921963,
"calibration/batch_entropy_100bins": 0.3644702705632904,
"calibration/batch_entropy_10bins": 0.37269805542921963,
"calibration/batch_entropy_50bins": 0.4263596164138678,
"calibration/batch_uniqueness": 0.4926999252218061,
"calibration/confidence_entropy": 0.3128198436922223,
"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": 0.37269805542921963,
"calibration/distribution_entropy_100": 0.3644702705632904,
"calibration/ece": 0.4117707299093361,
"calibration/mean_confidence": 0.8824584332751864,
"calibration/unique_confidence_per_question": 0.0328125,
"calibration/unique_confidences": 12.6,
"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": 0.009201388888888884,
"completions/max_length": 3999.8,
"completions/max_terminated_length": 3999.8,
"completions/mean_length": 521.1518188476563,
"completions/mean_terminated_length": 526.0406677246094,
"completions/min_length": 0.0,
"completions/min_terminated_length": 101.8,
"epoch": 0.09615384615384616,
"grad_norm": 0.0012512662215158343,
"learning_rate": 9.615384615384617e-07,
"loss": -0.0058,
"num_tokens": 69738734.0,
"reward": 1.0121073603630066,
"reward_std": 0.3443099856376648,
"rewards/accuracy_reward": 0.46319444179534913,
"rewards/brier_reward": 0.578621792793274,
"rewards/confidence_one_or_zero": 0.13663194328546524,
"rewards/format_reward": 0.9823784708976746,
"rewards/mean_confidence_reward": 0.864686906337738,
"sampling/batch_mean_priority_error": 0.006002986111111111,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9694444444444444,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.0001326042431173846,
"sampling/priority_kl": 0.030000335350632668,
"sampling/priority_scale": 2.025655818078667,
"sampling/prob_entropy": 10.27895622253418,
"sampling/prob_max": 3.5567282611737025e-05,
"sampling/prob_min": 6.42574004814378e-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": 0.3021592855453491,
"signal/accuracy_reward/group_std_mean": 0.3629056513309479,
"signal/accuracy_reward/group_zero_std_frac": 0.09444444552063942,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.15107964277267455,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.15107964277267455,
"signal/advantage_abs_mean": 0.28105672001838683,
"signal/advantage_pre_scale_abs_mean": 0.28105672001838683,
"signal/advantage_pre_scale_std": 0.36014688611030576,
"signal/advantage_std": 0.36014688611030576,
"signal/brier_reward/centered_abs_mean": 0.2543889760971069,
"signal/brier_reward/group_std_mean": 0.30701815485954287,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.12719448804855346,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.12719448804855346,
"signal/confidence_one_or_zero/centered_abs_mean": 0.21121961772441863,
"signal/confidence_one_or_zero/group_std_mean": 0.3036592543125153,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.07777777947485447,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.1121961253811607e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.1121961253811607e-06,
"signal/format_reward/centered_abs_mean": 0.03193901851773262,
"signal/format_reward/group_std_mean": 0.06820150762796402,
"signal/format_reward/group_zero_std_frac": 0.694444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01596950925886631,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01596950925886631,
"signal/mean_confidence_reward/centered_abs_mean": 0.08309155404567718,
"signal/mean_confidence_reward/group_std_mean": 0.1238383561372757,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.30915519145492e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.30915519145492e-07,
"step": 40
},
{
"calibration/aurc": 0.3610586952487986,
"calibration/batch_distribution_entropy": 0.38911473171552363,
"calibration/batch_entropy_100bins": 0.35479215071058084,
"calibration/batch_entropy_10bins": 0.38911473171552363,
"calibration/batch_entropy_50bins": 0.4150642023443708,
"calibration/batch_uniqueness": 0.47124124647406396,
"calibration/confidence_entropy": 0.33383225819897927,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.03674540682414698,
"calibration/coverage@15%": 0.05249343832020997,
"calibration/coverage@20%": 0.05249343832020997,
"calibration/coverage@25%": 0.2442367730349496,
"calibration/coverage@30%": 0.2542105263157895,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.38911473171552363,
"calibration/distribution_entropy_100": 0.35479215071058084,
"calibration/ece": 0.29194698660035917,
"calibration/mean_confidence": 0.8754128244232631,
"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": 0.011284722222222232,
"completions/max_length": 4021.8,
"completions/max_terminated_length": 4021.8,
"completions/mean_length": 558.5407104492188,
"completions/mean_terminated_length": 564.9087158203125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 119.6,
"epoch": 0.10817307692307693,
"grad_norm": 0.001072445884346962,
"learning_rate": 1.0817307692307693e-06,
"loss": -0.0059,
"num_tokens": 79291715.0,
"reward": 1.0605151176452636,
"reward_std": 0.3090484321117401,
"rewards/accuracy_reward": 0.5111979186534882,
"rewards/brier_reward": 0.627174460887909,
"rewards/confidence_one_or_zero": 0.09826388955116272,
"rewards/format_reward": 0.9826388835906983,
"rewards/mean_confidence_reward": 0.8520801186561584,
"sampling/batch_mean_priority_error": 0.005454097222222222,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9694444444444444,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00013870045950170606,
"sampling/priority_kl": 0.030001122504472733,
"sampling/priority_scale": 1.853211653465405,
"sampling/prob_entropy": 10.278968048095702,
"sampling/prob_max": 3.5752333496930076e-05,
"sampling/prob_min": 7.469276897609234e-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": 0.2584472715854645,
"signal/accuracy_reward/group_std_mean": 0.32760846614837646,
"signal/accuracy_reward/group_zero_std_frac": 0.1166666679084301,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.12922363579273224,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.12922363579273224,
"signal/advantage_abs_mean": 0.2409671425819397,
"signal/advantage_pre_scale_abs_mean": 0.2409671425819397,
"signal/advantage_pre_scale_std": 0.3296055316925049,
"signal/advantage_std": 0.3296055316925049,
"signal/brier_reward/centered_abs_mean": 0.21698220670223237,
"signal/brier_reward/group_std_mean": 0.2722037315368652,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.10849110335111618,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.10849110335111618,
"signal/confidence_one_or_zero/centered_abs_mean": 0.15730251967906952,
"signal/confidence_one_or_zero/group_std_mean": 0.24223730564117432,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.19999999850988387,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.5730251334389323e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.5730251334389323e-06,
"signal/format_reward/centered_abs_mean": 0.02931857593357563,
"signal/format_reward/group_std_mean": 0.05820476859807968,
"signal/format_reward/group_zero_std_frac": 0.7472222208976745,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.014659287966787815,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.014659287966787815,
"signal/mean_confidence_reward/centered_abs_mean": 0.08275075107812882,
"signal/mean_confidence_reward/group_std_mean": 0.1179319441318512,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.27507483336376e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.27507483336376e-07,
"step": 45
},
{
"calibration/aurc": 0.33053533787751727,
"calibration/batch_distribution_entropy": 0.512697310085846,
"calibration/batch_entropy_100bins": 0.3786039277099914,
"calibration/batch_entropy_10bins": 0.512697310085846,
"calibration/batch_entropy_50bins": 0.4452083349799242,
"calibration/batch_uniqueness": 0.5273792798695895,
"calibration/confidence_entropy": 0.3942478093536125,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.0,
"calibration/coverage@20%": 0.034557537107060214,
"calibration/coverage@25%": 0.3376609853829223,
"calibration/coverage@30%": 0.3381914893617021,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.512697310085846,
"calibration/distribution_entropy_100": 0.3786039277099914,
"calibration/ece": 0.22357226767279878,
"calibration/mean_confidence": 0.8412826863703291,
"calibration/unique_confidence_per_question": 0.03072916666666666,
"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": 0.011718749999999977,
"completions/max_length": 3947.2,
"completions/max_terminated_length": 3947.2,
"completions/mean_length": 580.1506225585938,
"completions/mean_terminated_length": 587.1082397460938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 131.6,
"epoch": 0.1201923076923077,
"grad_norm": 0.0008270725375041366,
"learning_rate": 1.201923076923077e-06,
"loss": -0.004,
"num_tokens": 89082154.0,
"reward": 1.1280964374542237,
"reward_std": 0.2733824193477631,
"rewards/accuracy_reward": 0.5832465171813965,
"rewards/brier_reward": 0.6891612410545349,
"rewards/confidence_one_or_zero": 0.05451388955116272,
"rewards/format_reward": 0.9837673544883728,
"rewards/mean_confidence_reward": 0.8320742964744567,
"sampling/batch_mean_priority_error": 0.005454166666666669,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9388888888888889,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.000144621345680207,
"sampling/priority_kl": 0.029999880492687224,
"sampling/priority_scale": 1.7189445613417775,
"sampling/prob_entropy": 10.27895565032959,
"sampling/prob_max": 3.592690773075446e-05,
"sampling/prob_min": 8.405669541389216e-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": 0.2265787750482559,
"signal/accuracy_reward/group_std_mean": 0.29308363795280457,
"signal/accuracy_reward/group_zero_std_frac": 0.1944444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.11328938752412795,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.11328938752412795,
"signal/advantage_abs_mean": 0.20797210037708283,
"signal/advantage_pre_scale_abs_mean": 0.20797210037708283,
"signal/advantage_pre_scale_std": 0.30454687476158143,
"signal/advantage_std": 0.30454687476158143,
"signal/brier_reward/centered_abs_mean": 0.18070028126239776,
"signal/brier_reward/group_std_mean": 0.23263510167598725,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.09035014063119888,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.09035014063119888,
"signal/confidence_one_or_zero/centered_abs_mean": 0.09245876669883728,
"signal/confidence_one_or_zero/group_std_mean": 0.1641049787402153,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.37222222685813905,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 9.245876526620123e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 9.245876526620123e-07,
"signal/format_reward/centered_abs_mean": 0.0284559465944767,
"signal/format_reward/group_std_mean": 0.056744760274887084,
"signal/format_reward/group_zero_std_frac": 0.7555555701255798,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01422797329723835,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01422797329723835,
"signal/mean_confidence_reward/centered_abs_mean": 0.08219865411520004,
"signal/mean_confidence_reward/group_std_mean": 0.11515908986330033,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.219864980674174e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.219864980674174e-07,
"step": 50
},
{
"epoch": 0.1201923076923077,
"eval_calibration/aurc": 0.3777325068473874,
"eval_calibration/batch_distribution_entropy": 0.5340746979902073,
"eval_calibration/batch_entropy_100bins": 0.36490768829677633,
"eval_calibration/batch_entropy_10bins": 0.5340746979902073,
"eval_calibration/batch_entropy_50bins": 0.4295634265060346,
"eval_calibration/batch_uniqueness": 0.5068985753349399,
"eval_calibration/confidence_entropy": 0.42309028695130074,
"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%": 0.026501766784452298,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.5340746979902073,
"eval_calibration/distribution_entropy_100": 0.36490768829677633,
"eval_calibration/ece": 0.23831272084805666,
"eval_calibration/mean_confidence": 0.8301855123674912,
"eval_calibration/unique_confidence_per_question": 0.010416666666666666,
"eval_calibration/unique_confidences": 12,
"eval_completions/clipped_ratio": 0.012152777777777771,
"eval_completions/max_length": 2896.3333333333335,
"eval_completions/max_terminated_length": 2896.3333333333335,
"eval_completions/mean_length": 610.5784810384115,
"eval_completions/mean_terminated_length": 618.1752726236979,
"eval_completions/min_length": 44.0,
"eval_completions/min_terminated_length": 183.83333333333334,
"eval_loss": 0.0,
"eval_num_tokens": 89082154.0,
"eval_reward": 1.1268197298049927,
"eval_reward_std": 0.429857795437177,
"eval_rewards/accuracy_reward": 0.581597218910853,
"eval_rewards/brier_reward": 0.689386377731959,
"eval_rewards/confidence_one_or_zero": 0.026909722092871863,
"eval_rewards/format_reward": 0.9826388855775198,
"eval_rewards/mean_confidence_reward": 0.816510409116745,
"eval_runtime": 203.2672,
"eval_samples_per_second": 4.92,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4685329894224803,
"eval_signal/accuracy_reward/group_std_mean": 0.4912516276041667,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.23426649471124014,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.23426649471124014,
"eval_signal/advantage_abs_mean": 0.39997173845767975,
"eval_signal/advantage_pre_scale_abs_mean": 0.39997173845767975,
"eval_signal/advantage_pre_scale_std": 0.42536558707555133,
"eval_signal/advantage_std": 0.42536558707555133,
"eval_signal/brier_reward/centered_abs_mean": 0.3129906157652537,
"eval_signal/brier_reward/group_std_mean": 0.3424776792526245,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.15649530788262686,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.15649530788262686,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.05094400964056452,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.12426702585071325,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 0.3888889029622078,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.094400895207704e-07,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.094400895207704e-07,
"eval_signal/format_reward/centered_abs_mean": 0.033311632151405014,
"eval_signal/format_reward/group_std_mean": 0.08924104770024617,
"eval_signal/format_reward/group_zero_std_frac": 0.5277777860562006,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.016655816075702507,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.016655816075702507,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.08704317485292752,
"eval_signal/mean_confidence_reward/group_std_mean": 0.13417261466383934,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.704317148537181e-07,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 8.704317148537181e-07,
"eval_steps_per_second": 0.03,
"step": 50
},
{
"epoch": 0.1201923076923077,
"step": 50,
"train_probe_calibration/aurc": 0.3112787978069011,
"train_probe_calibration/batch_distribution_entropy": 0.518454152532018,
"train_probe_calibration/batch_entropy_100bins": 0.3640552531542229,
"train_probe_calibration/batch_entropy_10bins": 0.518454152532018,
"train_probe_calibration/batch_entropy_50bins": 0.42855995364850563,
"train_probe_calibration/batch_uniqueness": 0.4955930864197531,
"train_probe_calibration/confidence_entropy": 0.41655487998101076,
"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%": 0.0,
"train_probe_calibration/coverage@30%": 0.528,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.518454152532018,
"train_probe_calibration/distribution_entropy_100": 0.3640552531542229,
"train_probe_calibration/ece": 0.20810666666666672,
"train_probe_calibration/mean_confidence": 0.8338844444444444,
"train_probe_calibration/unique_confidence_per_question": 0.011284722222222222,
"train_probe_calibration/unique_confidences": 13,
"train_probe_completions/clipped_ratio": 0.017361111111111105,
"train_probe_completions/max_length": 2972.0,
"train_probe_completions/max_terminated_length": 2972.0,
"train_probe_completions/mean_length": 631.3243204752604,
"train_probe_completions/mean_terminated_length": 642.1988016764323,
"train_probe_completions/min_length": 23.666666666666668,
"train_probe_completions/min_terminated_length": 173.83333333333334,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 89082154.0,
"train_probe_reward": 1.151051640510559,
"train_probe_reward_std": 0.4297498017549515,
"train_probe_rewards/accuracy_reward": 0.6111111144224802,
"train_probe_rewards/brier_reward": 0.7144128779570261,
"train_probe_rewards/confidence_one_or_zero": 0.026041666356225807,
"train_probe_rewards/format_reward": 0.9765625,
"train_probe_rewards/mean_confidence_reward": 0.8143402536710104,
"train_probe_runtime": 210.8945,
"train_probe_samples_per_second": 4.742,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.458984375,
"train_probe_signal/accuracy_reward/group_std_mean": 0.48603397111097973,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.2294921875,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.2294921875,
"train_probe_signal/advantage_abs_mean": 0.39387212693691254,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.39387212693691254,
"train_probe_signal/advantage_pre_scale_std": 0.42617770036061603,
"train_probe_signal/advantage_std": 0.42617770036061603,
"train_probe_signal/brier_reward/centered_abs_mean": 0.30468609432379407,
"train_probe_signal/brier_reward/group_std_mean": 0.3368734121322632,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.15234304716189703,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.15234304716189703,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.04980468718955914,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.13184053326646486,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 0.3055555646618207,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.980468446547093e-07,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.980468446547093e-07,
"train_probe_signal/format_reward/centered_abs_mean": 0.04421658099939426,
"train_probe_signal/format_reward/group_std_mean": 0.10708824172616005,
"train_probe_signal/format_reward/group_zero_std_frac": 0.4722222263614337,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.02210829049969713,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.02210829049969713,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.09213974326848984,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.14700974648197493,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 9.213974484130935e-07,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 9.213974484130935e-07,
"train_probe_steps_per_second": 0.028
},
{
"calibration/aurc": 0.38612993561216935,
"calibration/batch_distribution_entropy": 0.5405270158216778,
"calibration/batch_entropy_100bins": 0.3653346702300958,
"calibration/batch_entropy_10bins": 0.5405270158216778,
"calibration/batch_entropy_50bins": 0.42908534837434686,
"calibration/batch_uniqueness": 0.49659988221929785,
"calibration/confidence_entropy": 0.4437692389330308,
"calibration/coverage@0%": 0.003172045997444586,
"calibration/coverage@1%": 0.003172045997444586,
"calibration/coverage@10%": 0.003172045997444586,
"calibration/coverage@15%": 0.007427365146380757,
"calibration/coverage@20%": 0.12019332259318927,
"calibration/coverage@25%": 0.2031720459974446,
"calibration/coverage@30%": 0.20787178490083885,
"calibration/coverage@5%": 0.003172045997444586,
"calibration/distribution_entropy_10": 0.5405270158216778,
"calibration/distribution_entropy_100": 0.3653346702300958,
"calibration/ece": 0.22770160682164314,
"calibration/mean_confidence": 0.8139400842135398,
"calibration/unique_confidence_per_question": 0.028645833333333336,
"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": 0.014930555555555558,
"completions/max_length": 4055.0,
"completions/max_terminated_length": 4055.0,
"completions/mean_length": 632.4847290039063,
"completions/mean_terminated_length": 642.06376953125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 150.2,
"epoch": 0.13221153846153846,
"grad_norm": 0.0007234508520923555,
"learning_rate": 1.3221153846153848e-06,
"loss": -0.0063,
"num_tokens": 99434298.0,
"reward": 1.1398391246795654,
"reward_std": 0.2665547639131546,
"rewards/accuracy_reward": 0.5956597089767456,
"rewards/brier_reward": 0.7043145775794983,
"rewards/confidence_one_or_zero": 0.021006944589316845,
"rewards/format_reward": 0.9796875,
"rewards/mean_confidence_reward": 0.8053324580192566,
"sampling/batch_mean_priority_error": 0.004031666666666667,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9555555555555555,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00015117116854526102,
"sampling/priority_kl": 0.03000043034553528,
"sampling/priority_scale": 1.6105974317062646,
"sampling/prob_entropy": 10.278947639465333,
"sampling/prob_max": 3.609474879340269e-05,
"sampling/prob_min": 8.840246482577641e-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": 0.23159722089767457,
"signal/accuracy_reward/group_std_mean": 0.290811088681221,
"signal/accuracy_reward/group_zero_std_frac": 0.22777778208255767,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.11579861044883728,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.11579861044883728,
"signal/advantage_abs_mean": 0.20904322266578673,
"signal/advantage_pre_scale_abs_mean": 0.20904322266578673,
"signal/advantage_pre_scale_std": 0.30396526455879214,
"signal/advantage_std": 0.30396526455879214,
"signal/brier_reward/centered_abs_mean": 0.1729253500699997,
"signal/brier_reward/group_std_mean": 0.21837244629859925,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08646267503499985,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.08646267503499985,
"signal/confidence_one_or_zero/centered_abs_mean": 0.03847656212747097,
"signal/confidence_one_or_zero/group_std_mean": 0.08330218568444252,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.6250000059604645,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.847656159905455e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.847656159905455e-07,
"signal/format_reward/centered_abs_mean": 0.03191189244389534,
"signal/format_reward/group_std_mean": 0.05313037186861038,
"signal/format_reward/group_zero_std_frac": 0.7972222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01595594622194767,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01595594622194767,
"signal/mean_confidence_reward/centered_abs_mean": 0.08242994993925094,
"signal/mean_confidence_reward/group_std_mean": 0.11044336557388305,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.242994795182313e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.242994795182313e-07,
"step": 55
},
{
"calibration/aurc": 0.3336939658670478,
"calibration/batch_distribution_entropy": 0.5553734722247106,
"calibration/batch_entropy_100bins": 0.3494255823263553,
"calibration/batch_entropy_10bins": 0.5553734722247106,
"calibration/batch_entropy_50bins": 0.4113381418560308,
"calibration/batch_uniqueness": 0.4833634475353441,
"calibration/confidence_entropy": 0.48042285577547605,
"calibration/coverage@0%": 0.003222263714138264,
"calibration/coverage@1%": 0.003222263714138264,
"calibration/coverage@10%": 0.003222263714138264,
"calibration/coverage@15%": 0.003222263714138264,
"calibration/coverage@20%": 0.11631561781737232,
"calibration/coverage@25%": 0.27990400831869156,
"calibration/coverage@30%": 0.27990400831869156,
"calibration/coverage@5%": 0.003222263714138264,
"calibration/distribution_entropy_10": 0.5553734722247106,
"calibration/distribution_entropy_100": 0.3494255823263553,
"calibration/ece": 0.1802489469086288,
"calibration/mean_confidence": 0.7959282239222343,
"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": 0.011545138888888907,
"completions/max_length": 4024.4,
"completions/max_terminated_length": 4024.4,
"completions/mean_length": 628.0548706054688,
"completions/mean_terminated_length": 635.40615234375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 128.4,
"epoch": 0.14423076923076922,
"grad_norm": 0.0007936802576296031,
"learning_rate": 1.4423076923076922e-06,
"loss": -0.0038,
"num_tokens": 109729394.0,
"reward": 1.173889660835266,
"reward_std": 0.24853478670120238,
"rewards/accuracy_reward": 0.6305555582046509,
"rewards/brier_reward": 0.7328330159187317,
"rewards/confidence_one_or_zero": 0.006944444682449103,
"rewards/format_reward": 0.9843749880790711,
"rewards/mean_confidence_reward": 0.7818385362625122,
"sampling/batch_mean_priority_error": 0.004377708333333334,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9388888888888889,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00015656949544791133,
"sampling/priority_kl": 0.02999980337917805,
"sampling/priority_scale": 1.5206903575453907,
"sampling/prob_entropy": 10.27895908355713,
"sampling/prob_max": 3.625823956099339e-05,
"sampling/prob_min": 7.926992839202285e-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": 0.21881510317325592,
"signal/accuracy_reward/group_std_mean": 0.2803986519575119,
"signal/accuracy_reward/group_zero_std_frac": 0.23888889253139495,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.10940755158662796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.10940755158662796,
"signal/advantage_abs_mean": 0.1892775774002075,
"signal/advantage_pre_scale_abs_mean": 0.1892775774002075,
"signal/advantage_pre_scale_std": 0.28426212072372437,
"signal/advantage_std": 0.28426212072372437,
"signal/brier_reward/centered_abs_mean": 0.15051650404930114,
"signal/brier_reward/group_std_mean": 0.19497915208339692,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07525825202465057,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.07525825202465057,
"signal/confidence_one_or_zero/centered_abs_mean": 0.013183593563735486,
"signal/confidence_one_or_zero/group_std_mean": 0.0330656249076128,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.8333333492279053,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.3183593381427272e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.3183593381427272e-07,
"signal/format_reward/centered_abs_mean": 0.02564019113779068,
"signal/format_reward/group_std_mean": 0.050878601521253584,
"signal/format_reward/group_zero_std_frac": 0.7805555582046508,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01282009556889534,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01282009556889534,
"signal/mean_confidence_reward/centered_abs_mean": 0.0763712003827095,
"signal/mean_confidence_reward/group_std_mean": 0.10629059821367264,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.637119892933697e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.637119892933697e-07,
"step": 60
},
{
"calibration/aurc": 0.37586397587645826,
"calibration/batch_distribution_entropy": 0.5549776555813649,
"calibration/batch_entropy_100bins": 0.31981659342569346,
"calibration/batch_entropy_10bins": 0.5549776555813649,
"calibration/batch_entropy_50bins": 0.3764829191915983,
"calibration/batch_uniqueness": 0.40589435403374124,
"calibration/confidence_entropy": 0.5348350374864965,
"calibration/coverage@0%": 0.0026654322500107444,
"calibration/coverage@1%": 0.0026654322500107444,
"calibration/coverage@10%": 0.0026654322500107444,
"calibration/coverage@15%": 0.04246310706235907,
"calibration/coverage@20%": 0.17721111767243863,
"calibration/coverage@25%": 0.20479732456899033,
"calibration/coverage@30%": 0.20479732456899033,
"calibration/coverage@5%": 0.0026654322500107444,
"calibration/distribution_entropy_10": 0.5549776555813649,
"calibration/distribution_entropy_100": 0.31981659342569346,
"calibration/ece": 0.16123688431669309,
"calibration/mean_confidence": 0.7563851204504758,
"calibration/unique_confidence_per_question": 0.024479166666666666,
"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": 0.01866319444444444,
"completions/max_length": 4017.2,
"completions/max_terminated_length": 4017.2,
"completions/mean_length": 665.1203979492187,
"completions/mean_terminated_length": 677.7648315429688,
"completions/min_length": 0.0,
"completions/min_terminated_length": 195.2,
"epoch": 0.15625,
"grad_norm": 0.0006811345228925347,
"learning_rate": 1.5625e-06,
"loss": -0.0077,
"num_tokens": 120519677.0,
"reward": 1.1617473363876343,
"reward_std": 0.2451097071170807,
"rewards/accuracy_reward": 0.6177951335906983,
"rewards/brier_reward": 0.7292956352233887,
"rewards/confidence_one_or_zero": 0.0018229167035315186,
"rewards/format_reward": 0.9763888955116272,
"rewards/mean_confidence_reward": 0.7422178864479065,
"sampling/batch_mean_priority_error": 0.005695416666666666,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.95,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00016235393995884807,
"sampling/priority_kl": 0.03000013642013073,
"sampling/priority_scale": 1.4449297070968896,
"sampling/prob_entropy": 10.278961563110352,
"sampling/prob_max": 3.641730290837586e-05,
"sampling/prob_min": 8.587727643316612e-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": 0.21707356870174407,
"signal/accuracy_reward/group_std_mean": 0.27858763337135317,
"signal/accuracy_reward/group_zero_std_frac": 0.2361111134290695,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.10853678435087204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.10853678435087204,
"signal/advantage_abs_mean": 0.1863380938768387,
"signal/advantage_pre_scale_abs_mean": 0.1863380938768387,
"signal/advantage_pre_scale_std": 0.2836464047431946,
"signal/advantage_std": 0.2836464047431946,
"signal/brier_reward/centered_abs_mean": 0.14009126722812654,
"signal/brier_reward/group_std_mean": 0.1813989222049713,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07004563361406327,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.07004563361406327,
"signal/confidence_one_or_zero/centered_abs_mean": 0.003521050268318504,
"signal/confidence_one_or_zero/group_std_mean": 0.010013032704591751,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9444444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.5210501181381915e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.5210501181381915e-08,
"signal/format_reward/centered_abs_mean": 0.03613281212747097,
"signal/format_reward/group_std_mean": 0.062471595406532285,
"signal/format_reward/group_zero_std_frac": 0.7583333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.018066406063735486,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.018066406063735486,
"signal/mean_confidence_reward/centered_abs_mean": 0.08147167861461639,
"signal/mean_confidence_reward/group_std_mean": 0.11126352846622467,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.147167591232574e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.147167591232574e-07,
"step": 65
},
{
"calibration/aurc": 0.3543039655168289,
"calibration/batch_distribution_entropy": 0.5730260289511547,
"calibration/batch_entropy_100bins": 0.31485441323741503,
"calibration/batch_entropy_10bins": 0.5730260289511547,
"calibration/batch_entropy_50bins": 0.37064152096135977,
"calibration/batch_uniqueness": 0.425199174266254,
"calibration/confidence_entropy": 0.5637355662717745,
"calibration/coverage@0%": 0.0015804016757562183,
"calibration/coverage@1%": 0.0015804016757562183,
"calibration/coverage@10%": 0.0015804016757562183,
"calibration/coverage@15%": 0.0015804016757562183,
"calibration/coverage@20%": 0.028442691684597947,
"calibration/coverage@25%": 0.10854879248035391,
"calibration/coverage@30%": 0.4010498976969764,
"calibration/coverage@5%": 0.0015804016757562183,
"calibration/distribution_entropy_10": 0.5730260289511547,
"calibration/distribution_entropy_100": 0.31485441323741503,
"calibration/ece": 0.1173558645266779,
"calibration/mean_confidence": 0.7295070335410896,
"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": 0.01597222222222221,
"completions/max_length": 4049.2,
"completions/max_terminated_length": 4049.2,
"completions/mean_length": 684.5787475585937,
"completions/mean_terminated_length": 695.9264892578125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 201.6,
"epoch": 0.16826923076923078,
"grad_norm": 0.0007994287298060954,
"learning_rate": 1.682692307692308e-06,
"loss": -0.0082,
"num_tokens": 131515944.0,
"reward": 1.1738406658172607,
"reward_std": 0.2270677924156189,
"rewards/accuracy_reward": 0.6275173664093018,
"rewards/brier_reward": 0.7403752088546753,
"rewards/confidence_one_or_zero": 0.001909722265554592,
"rewards/format_reward": 0.9797743082046508,
"rewards/mean_confidence_reward": 0.7171657919883728,
"sampling/batch_mean_priority_error": 0.006220486111111113,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9305555555555556,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00016924338706303388,
"sampling/priority_kl": 0.030000567063689233,
"sampling/priority_scale": 1.3802560209762305,
"sampling/prob_entropy": 10.278955459594727,
"sampling/prob_max": 3.657195411506109e-05,
"sampling/prob_min": 9.199866326525807e-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": 0.20630967915058135,
"signal/accuracy_reward/group_std_mean": 0.2702312171459198,
"signal/accuracy_reward/group_zero_std_frac": 0.2416666716337204,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.10315483957529067,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.10315483957529067,
"signal/advantage_abs_mean": 0.17068581879138947,
"signal/advantage_pre_scale_abs_mean": 0.17068581879138947,
"signal/advantage_pre_scale_std": 0.26389814019203184,
"signal/advantage_std": 0.26389814019203184,
"signal/brier_reward/centered_abs_mean": 0.12424491196870804,
"signal/brier_reward/group_std_mean": 0.16228446662425994,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.06212245598435402,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.06212245598435402,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0036349825211800637,
"signal/confidence_one_or_zero/group_std_mean": 0.009772197715938092,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9472222328186035,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.634982306266465e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.634982306266465e-08,
"signal/format_reward/centered_abs_mean": 0.029291448928415776,
"signal/format_reward/group_std_mean": 0.05109502263367176,
"signal/format_reward/group_zero_std_frac": 0.8027777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.014645724464207888,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.014645724464207888,
"signal/mean_confidence_reward/centered_abs_mean": 0.07622585743665695,
"signal/mean_confidence_reward/group_std_mean": 0.10273666977882386,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.622585485478339e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.622585485478339e-07,
"step": 70
},
{
"calibration/aurc": 0.2858884438881962,
"calibration/batch_distribution_entropy": 0.5609883010680576,
"calibration/batch_entropy_100bins": 0.31588468551229204,
"calibration/batch_entropy_10bins": 0.5609883010680576,
"calibration/batch_entropy_50bins": 0.37185434081367924,
"calibration/batch_uniqueness": 0.4256885939815541,
"calibration/confidence_entropy": 0.5814513279306728,
"calibration/coverage@0%": 0.00585995756396415,
"calibration/coverage@1%": 0.00585995756396415,
"calibration/coverage@10%": 0.010634493372982716,
"calibration/coverage@15%": 0.022864303181662667,
"calibration/coverage@20%": 0.02446004786251373,
"calibration/coverage@25%": 0.4076161182910999,
"calibration/coverage@30%": 0.7452000677239121,
"calibration/coverage@5%": 0.00585995756396415,
"calibration/distribution_entropy_10": 0.5609883010680576,
"calibration/distribution_entropy_100": 0.31588468551229204,
"calibration/ece": 0.10178788260454422,
"calibration/mean_confidence": 0.7036365611088247,
"calibration/unique_confidence_per_question": 0.0265625,
"calibration/unique_confidences": 10.2,
"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": 0.014756944444444442,
"completions/max_length": 3767.4,
"completions/max_terminated_length": 3767.4,
"completions/mean_length": 663.8887939453125,
"completions/mean_terminated_length": 673.8903564453125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 212.4,
"epoch": 0.18028846153846154,
"grad_norm": 0.0005497903912328184,
"learning_rate": 1.8028846153846156e-06,
"loss": -0.0113,
"num_tokens": 142235079.0,
"reward": 1.2008941411972045,
"reward_std": 0.21300730109214783,
"rewards/accuracy_reward": 0.6611979246139527,
"rewards/brier_reward": 0.7575034737586975,
"rewards/confidence_one_or_zero": 0.0012152778159361332,
"rewards/format_reward": 0.9830729126930237,
"rewards/mean_confidence_reward": 0.6975086808204651,
"sampling/batch_mean_priority_error": 0.006739583333333335,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9277777777777777,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00017708493396639824,
"sampling/priority_kl": 0.029999949410557745,
"sampling/priority_scale": 1.324010860780254,
"sampling/prob_entropy": 10.278962707519531,
"sampling/prob_max": 3.672436432680115e-05,
"sampling/prob_min": 9.772332668944728e-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": 0.19198676347732543,
"signal/accuracy_reward/group_std_mean": 0.25041328370571136,
"signal/accuracy_reward/group_zero_std_frac": 0.3000000029802322,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.09599338173866272,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.09599338173866272,
"signal/advantage_abs_mean": 0.15800958573818208,
"signal/advantage_pre_scale_abs_mean": 0.15800958573818208,
"signal/advantage_pre_scale_std": 0.2527189999818802,
"signal/advantage_std": 0.2527189999818802,
"signal/brier_reward/centered_abs_mean": 0.11541662514209747,
"signal/brier_reward/group_std_mean": 0.15058891773223876,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.057708312571048734,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.057708312571048734,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0022460937150754036,
"signal/confidence_one_or_zero/group_std_mean": 0.005444145016372204,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222089767456,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.2460936577317624e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.2460936577317624e-08,
"signal/format_reward/centered_abs_mean": 0.02916124165058136,
"signal/format_reward/group_std_mean": 0.05634359717369079,
"signal/format_reward/group_zero_std_frac": 0.7638888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01458062082529068,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01458062082529068,
"signal/mean_confidence_reward/centered_abs_mean": 0.07778618484735489,
"signal/mean_confidence_reward/group_std_mean": 0.1058467149734497,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.778618396514502e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.778618396514502e-07,
"step": 75
},
{
"calibration/aurc": 0.4032866542071951,
"calibration/batch_distribution_entropy": 0.519863369326861,
"calibration/batch_entropy_100bins": 0.28161969711614626,
"calibration/batch_entropy_10bins": 0.519863369326861,
"calibration/batch_entropy_50bins": 0.33151815087660375,
"calibration/batch_uniqueness": 0.36394117551198735,
"calibration/confidence_entropy": 0.6014145393968999,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.0,
"calibration/coverage@20%": 0.0032,
"calibration/coverage@25%": 0.005310817941952506,
"calibration/coverage@30%": 0.2021108179419525,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.519863369326861,
"calibration/distribution_entropy_100": 0.28161969711614626,
"calibration/ece": 0.14039651033528555,
"calibration/mean_confidence": 0.6901999298390014,
"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": 0.012586805555555558,
"completions/max_length": 3871.8,
"completions/max_terminated_length": 3871.8,
"completions/mean_length": 687.1427368164062,
"completions/mean_terminated_length": 695.88330078125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 191.4,
"epoch": 0.19230769230769232,
"grad_norm": 0.0006263877148739994,
"learning_rate": 1.9230769230769234e-06,
"loss": -0.0087,
"num_tokens": 153266035.0,
"reward": 1.18265540599823,
"reward_std": 0.20969317853450775,
"rewards/accuracy_reward": 0.6322916626930237,
"rewards/brier_reward": 0.747154700756073,
"rewards/confidence_one_or_zero": 0.0008680555794853718,
"rewards/format_reward": 0.9858506917953491,
"rewards/mean_confidence_reward": 0.6823298692703247,
"sampling/batch_mean_priority_error": 0.009656250000000002,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9166666666666667,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00018732520984485746,
"sampling/priority_kl": 0.029999008402228355,
"sampling/priority_scale": 1.2748031734023244,
"sampling/prob_entropy": 10.278940010070801,
"sampling/prob_max": 3.6873606586596e-05,
"sampling/prob_min": 1.0306679178029298e-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": 0.20177951455116272,
"signal/accuracy_reward/group_std_mean": 0.25801017582416536,
"signal/accuracy_reward/group_zero_std_frac": 0.28888889253139494,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.10088975727558136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.10088975727558136,
"signal/advantage_abs_mean": 0.1581605225801468,
"signal/advantage_pre_scale_abs_mean": 0.1581605225801468,
"signal/advantage_pre_scale_std": 0.2474273145198822,
"signal/advantage_std": 0.2474273145198822,
"signal/brier_reward/centered_abs_mean": 0.10887281149625778,
"signal/brier_reward/group_std_mean": 0.1416696459054947,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.05443640574812889,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.05443640574812889,
"signal/confidence_one_or_zero/centered_abs_mean": 0.001649305538740009,
"signal/confidence_one_or_zero/group_std_mean": 0.004259948246181011,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9777777671813965,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6493054744159964e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6493054744159964e-08,
"signal/format_reward/centered_abs_mean": 0.02550455778837204,
"signal/format_reward/group_std_mean": 0.052160313725471495,
"signal/format_reward/group_zero_std_frac": 0.7750000119209289,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01275227889418602,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01275227889418602,
"signal/mean_confidence_reward/centered_abs_mean": 0.07146369963884354,
"signal/mean_confidence_reward/group_std_mean": 0.09777705222368241,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.146369853217039e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.146369853217039e-07,
"step": 80
},
{
"calibration/aurc": 0.3661781722935645,
"calibration/batch_distribution_entropy": 0.4992257819234746,
"calibration/batch_entropy_100bins": 0.2705582175171825,
"calibration/batch_entropy_10bins": 0.4992257819234746,
"calibration/batch_entropy_50bins": 0.31849675606594413,
"calibration/batch_uniqueness": 0.33240218609144073,
"calibration/confidence_entropy": 0.610284052868314,
"calibration/coverage@0%": 0.002702702702702703,
"calibration/coverage@1%": 0.002702702702702703,
"calibration/coverage@10%": 0.002702702702702703,
"calibration/coverage@15%": 0.002702702702702703,
"calibration/coverage@20%": 0.002702702702702703,
"calibration/coverage@25%": 0.20270270270270271,
"calibration/coverage@30%": 0.20270270270270271,
"calibration/coverage@5%": 0.002702702702702703,
"calibration/distribution_entropy_10": 0.4992257819234746,
"calibration/distribution_entropy_100": 0.2705582175171825,
"calibration/ece": 0.12068481756481757,
"calibration/mean_confidence": 0.6806887197314783,
"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": 0.015625,
"completions/max_length": 4002.4,
"completions/max_terminated_length": 4002.4,
"completions/mean_length": 665.6905395507813,
"completions/mean_terminated_length": 676.254638671875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 173.6,
"epoch": 0.20432692307692307,
"grad_norm": 0.0009465516777709126,
"learning_rate": 2.043269230769231e-06,
"loss": -0.011,
"num_tokens": 164055622.0,
"reward": 1.1916949510574342,
"reward_std": 0.21416583359241487,
"rewards/accuracy_reward": 0.6455729126930236,
"rewards/brier_reward": 0.754991102218628,
"rewards/confidence_one_or_zero": 0.0003472222248092294,
"rewards/format_reward": 0.9828125,
"rewards/mean_confidence_reward": 0.6632769107818604,
"sampling/batch_mean_priority_error": 0.011204861111111113,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.875,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00019981984514743089,
"sampling/priority_kl": 0.02999875508248806,
"sampling/priority_scale": 1.231435882905498,
"sampling/prob_entropy": 10.27894401550293,
"sampling/prob_max": 3.701617824845016e-05,
"sampling/prob_min": 1.0804863813973498e-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": 0.20201280117034912,
"signal/accuracy_reward/group_std_mean": 0.2671117514371872,
"signal/accuracy_reward/group_zero_std_frac": 0.25,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.10100640058517456,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.10100640058517456,
"signal/advantage_abs_mean": 0.15618274211883545,
"signal/advantage_pre_scale_abs_mean": 0.15618274211883545,
"signal/advantage_pre_scale_std": 0.2473001003265381,
"signal/advantage_std": 0.2473001003265381,
"signal/brier_reward/centered_abs_mean": 0.10262170583009719,
"signal/brier_reward/group_std_mean": 0.13768472969532014,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.051310852915048596,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.051310852915048596,
"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": 0.03017578162252903,
"signal/format_reward/group_std_mean": 0.05734373852610588,
"signal/format_reward/group_zero_std_frac": 0.7666666746139527,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.015087890811264515,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.015087890811264515,
"signal/mean_confidence_reward/centered_abs_mean": 0.07156276404857635,
"signal/mean_confidence_reward/group_std_mean": 0.09738876074552535,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.156276069508749e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.156276069508749e-07,
"step": 85
},
{
"calibration/aurc": 0.37116374332762697,
"calibration/batch_distribution_entropy": 0.4757943579865319,
"calibration/batch_entropy_100bins": 0.26224067309847543,
"calibration/batch_entropy_10bins": 0.4757943579865319,
"calibration/batch_entropy_50bins": 0.3087054773529833,
"calibration/batch_uniqueness": 0.28793524936717196,
"calibration/confidence_entropy": 0.6222985334875628,
"calibration/coverage@0%": 0.0005494505494505495,
"calibration/coverage@1%": 0.0005494505494505495,
"calibration/coverage@10%": 0.0005494505494505495,
"calibration/coverage@15%": 0.0005494505494505495,
"calibration/coverage@20%": 0.0005494505494505495,
"calibration/coverage@25%": 0.04475997686524002,
"calibration/coverage@30%": 0.20054945054945056,
"calibration/coverage@5%": 0.0005494505494505495,
"calibration/distribution_entropy_10": 0.4757943579865319,
"calibration/distribution_entropy_100": 0.26224067309847543,
"calibration/ece": 0.0960165185039298,
"calibration/mean_confidence": 0.6651432014026919,
"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": 0.015364583333333348,
"completions/max_length": 3960.0,
"completions/max_terminated_length": 3960.0,
"completions/mean_length": 679.5826416015625,
"completions/mean_terminated_length": 690.2464477539063,
"completions/min_length": 0.0,
"completions/min_terminated_length": 179.6,
"epoch": 0.21634615384615385,
"grad_norm": 0.0005822376697324216,
"learning_rate": 2.1634615384615387e-06,
"loss": -0.0117,
"num_tokens": 174998654.0,
"reward": 1.1882888317108153,
"reward_std": 0.2097031891345978,
"rewards/accuracy_reward": 0.637586796283722,
"rewards/brier_reward": 0.7558179020881652,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9831597328186035,
"rewards/mean_confidence_reward": 0.6578081369400024,
"sampling/batch_mean_priority_error": 0.011411458333333336,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8833333333333334,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00021357860241550953,
"sampling/priority_kl": 0.02999962829053402,
"sampling/priority_scale": 1.1929583669174462,
"sampling/prob_entropy": 10.278938102722169,
"sampling/prob_max": 3.715199272846803e-05,
"sampling/prob_min": 1.1269766037003138e-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": 0.2021755665540695,
"signal/accuracy_reward/group_std_mean": 0.26527499556541445,
"signal/accuracy_reward/group_zero_std_frac": 0.2555555641651154,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.10108778327703476,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.10108778327703476,
"signal/advantage_abs_mean": 0.15595764219760894,
"signal/advantage_pre_scale_abs_mean": 0.15595764219760894,
"signal/advantage_pre_scale_std": 0.24449576139450074,
"signal/advantage_std": 0.24449576139450074,
"signal/brier_reward/centered_abs_mean": 0.10028855502605438,
"signal/brier_reward/group_std_mean": 0.13360466957092285,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.05014427751302719,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.05014427751302719,
"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": 0.02744140550494194,
"signal/format_reward/group_std_mean": 0.050358953326940535,
"signal/format_reward/group_zero_std_frac": 0.7916666865348816,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01372070275247097,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01372070275247097,
"signal/mean_confidence_reward/centered_abs_mean": 0.06903888806700706,
"signal/mean_confidence_reward/group_std_mean": 0.0928740844130516,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.903888788656331e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.903888788656331e-07,
"step": 90
},
{
"calibration/aurc": 0.3758242948585995,
"calibration/batch_distribution_entropy": 0.4563553410376876,
"calibration/batch_entropy_100bins": 0.24585762197374333,
"calibration/batch_entropy_10bins": 0.4563553410376876,
"calibration/batch_entropy_50bins": 0.28941961464449506,
"calibration/batch_uniqueness": 0.24668494033689709,
"calibration/confidence_entropy": 0.619717511469672,
"calibration/coverage@0%": 0.002688172043010753,
"calibration/coverage@1%": 0.002688172043010753,
"calibration/coverage@10%": 0.027419354838709675,
"calibration/coverage@15%": 0.05049822665090357,
"calibration/coverage@20%": 0.05470324180659384,
"calibration/coverage@25%": 0.13965527496313093,
"calibration/coverage@30%": 0.13965527496313093,
"calibration/coverage@5%": 0.002688172043010753,
"calibration/distribution_entropy_10": 0.4563553410376876,
"calibration/distribution_entropy_100": 0.24585762197374333,
"calibration/ece": 0.10688778213917556,
"calibration/mean_confidence": 0.6715519727450985,
"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": 0.010416666666666675,
"completions/max_length": 4000.8,
"completions/max_terminated_length": 4000.8,
"completions/mean_length": 644.25087890625,
"completions/mean_terminated_length": 651.0717529296875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 198.4,
"epoch": 0.2283653846153846,
"grad_norm": 0.0006384830921888351,
"learning_rate": 2.283653846153846e-06,
"loss": -0.0086,
"num_tokens": 185514024.0,
"reward": 1.1923712730407714,
"reward_std": 0.1970378041267395,
"rewards/accuracy_reward": 0.6372395873069763,
"rewards/brier_reward": 0.7592951536178589,
"rewards/confidence_one_or_zero": 0.000260416668606922,
"rewards/format_reward": 0.9881944417953491,
"rewards/mean_confidence_reward": 0.6610156297683716,
"sampling/batch_mean_priority_error": 0.012123263888888892,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9027777777777779,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00022780775325372815,
"sampling/priority_kl": 0.030000927299261092,
"sampling/priority_scale": 1.1581266285385936,
"sampling/prob_entropy": 10.278959465026855,
"sampling/prob_max": 3.729468226083554e-05,
"sampling/prob_min": 1.1713916137523483e-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": 0.19498155415058135,
"signal/accuracy_reward/group_std_mean": 0.2563039720058441,
"signal/accuracy_reward/group_zero_std_frac": 0.275,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.09749077707529068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.09749077707529068,
"signal/advantage_abs_mean": 0.14624694287776946,
"signal/advantage_pre_scale_abs_mean": 0.14624694287776946,
"signal/advantage_pre_scale_std": 0.23240326046943666,
"signal/advantage_std": 0.23240326046943666,
"signal/brier_reward/centered_abs_mean": 0.09518856704235076,
"signal/brier_reward/group_std_mean": 0.12528109848499297,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04759428352117538,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04759428352117538,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0005045572877861559,
"signal/confidence_one_or_zero/group_std_mean": 0.0014731390401721,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/format_reward/centered_abs_mean": 0.01911892332136631,
"signal/format_reward/group_std_mean": 0.037534280493855474,
"signal/format_reward/group_zero_std_frac": 0.8361111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009559461660683156,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009559461660683156,
"signal/mean_confidence_reward/centered_abs_mean": 0.06400986313819886,
"signal/mean_confidence_reward/group_std_mean": 0.08521138578653335,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.400986194421421e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.400986194421421e-07,
"step": 95
},
{
"calibration/aurc": 0.3678310838070479,
"calibration/batch_distribution_entropy": 0.46524390786738545,
"calibration/batch_entropy_100bins": 0.2504651973371159,
"calibration/batch_entropy_10bins": 0.46524390786738545,
"calibration/batch_entropy_50bins": 0.2948435778123125,
"calibration/batch_uniqueness": 0.2624876520204674,
"calibration/confidence_entropy": 0.6178975760044858,
"calibration/coverage@0%": 0.0026709523267721014,
"calibration/coverage@1%": 0.0026709523267721014,
"calibration/coverage@10%": 0.0026709523267721014,
"calibration/coverage@15%": 0.04598646034816247,
"calibration/coverage@20%": 0.06513539651837524,
"calibration/coverage@25%": 0.09243198444488443,
"calibration/coverage@30%": 0.36099780535738873,
"calibration/coverage@5%": 0.0026709523267721014,
"calibration/distribution_entropy_10": 0.46524390786738545,
"calibration/distribution_entropy_100": 0.2504651973371159,
"calibration/ece": 0.12128469451627992,
"calibration/mean_confidence": 0.6711969216252879,
"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": 0.01493055555555558,
"completions/max_length": 3220.6,
"completions/max_terminated_length": 3220.6,
"completions/mean_length": 642.8724853515625,
"completions/mean_terminated_length": 652.767138671875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 207.6,
"epoch": 0.2403846153846154,
"grad_norm": 0.0006411863723769784,
"learning_rate": 2.403846153846154e-06,
"loss": -0.0114,
"num_tokens": 196021611.0,
"reward": 1.1830902338027953,
"reward_std": 0.1911737948656082,
"rewards/accuracy_reward": 0.6262152671813965,
"rewards/brier_reward": 0.7555767059326172,
"rewards/confidence_one_or_zero": 0.0006076388934161514,
"rewards/format_reward": 0.9843749880790711,
"rewards/mean_confidence_reward": 0.6688211679458618,
"sampling/batch_mean_priority_error": 0.011630208333333334,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.888888888888889,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00024215736775659024,
"sampling/priority_kl": 0.029999716952443124,
"sampling/priority_scale": 1.126824200199917,
"sampling/prob_entropy": 10.278961944580079,
"sampling/prob_max": 3.743151755770668e-05,
"sampling/prob_min": 1.2130623326811474e-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": 0.17940538227558137,
"signal/accuracy_reward/group_std_mean": 0.23320743143558503,
"signal/accuracy_reward/group_zero_std_frac": 0.34722222089767457,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08970269113779068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08970269113779068,
"signal/advantage_abs_mean": 0.14159118086099626,
"signal/advantage_pre_scale_abs_mean": 0.14159118086099626,
"signal/advantage_pre_scale_std": 0.2317925751209259,
"signal/advantage_std": 0.2317925751209259,
"signal/brier_reward/centered_abs_mean": 0.09510144293308258,
"signal/brier_reward/group_std_mean": 0.12647282779216767,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04755072146654129,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04755072146654129,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0011773003381676972,
"signal/confidence_one_or_zero/group_std_mean": 0.0034373244270682335,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9805555462837219,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.1773002484005702e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.1773002484005702e-08,
"signal/format_reward/centered_abs_mean": 0.02516275979578495,
"signal/format_reward/group_std_mean": 0.04855161756277084,
"signal/format_reward/group_zero_std_frac": 0.7972222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.012581379897892476,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.012581379897892476,
"signal/mean_confidence_reward/centered_abs_mean": 0.06360747069120407,
"signal/mean_confidence_reward/group_std_mean": 0.08851810693740844,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.360746851896692e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.360746851896692e-07,
"step": 100
},
{
"epoch": 0.2403846153846154,
"eval_calibration/aurc": 0.3215100260380746,
"eval_calibration/batch_distribution_entropy": 0.4526846744320586,
"eval_calibration/batch_entropy_100bins": 0.23913880523342096,
"eval_calibration/batch_entropy_10bins": 0.4526846744320586,
"eval_calibration/batch_entropy_50bins": 0.28151033228733163,
"eval_calibration/batch_uniqueness": 0.22393147108541367,
"eval_calibration/confidence_entropy": 0.6107200486093004,
"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%": 0.17032484635645304,
"eval_calibration/coverage@30%": 0.17032484635645304,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.4526846744320586,
"eval_calibration/distribution_entropy_100": 0.23913880523342096,
"eval_calibration/ece": 0.06097453906935911,
"eval_calibration/mean_confidence": 0.6855575065847235,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.009548611111111105,
"eval_completions/max_length": 2007.0,
"eval_completions/max_terminated_length": 2007.0,
"eval_completions/mean_length": 630.4380289713541,
"eval_completions/mean_terminated_length": 636.6086832682291,
"eval_completions/min_length": 53.666666666666664,
"eval_completions/min_terminated_length": 231.83333333333334,
"eval_loss": 0.0,
"eval_num_tokens": 196021611.0,
"eval_reward": 1.2005070646603901,
"eval_reward_std": 0.34502998491128284,
"eval_rewards/accuracy_reward": 0.6475694477558136,
"eval_rewards/brier_reward": 0.7647157112757365,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9887152711550394,
"eval_rewards/mean_confidence_reward": 0.6778211891651154,
"eval_runtime": 206.6087,
"eval_samples_per_second": 4.84,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4409722189108531,
"eval_signal/accuracy_reward/group_std_mean": 0.47597145040829975,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.22048610945542654,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.22048610945542654,
"eval_signal/advantage_abs_mean": 0.31116824348767597,
"eval_signal/advantage_pre_scale_abs_mean": 0.31116824348767597,
"eval_signal/advantage_pre_scale_std": 0.34272391100724536,
"eval_signal/advantage_std": 0.34272391100724536,
"eval_signal/brier_reward/centered_abs_mean": 0.16785739362239838,
"eval_signal/brier_reward/group_std_mean": 0.19804431994756064,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08392869681119919,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08392869681119919,
"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": 0.02153862826526165,
"eval_signal/format_reward/group_std_mean": 0.057330875347057976,
"eval_signal/format_reward/group_zero_std_frac": 0.6944444676240286,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.010769314132630825,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.010769314132630825,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.06253797250489394,
"eval_signal/mean_confidence_reward/group_std_mean": 0.09316463147600491,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.253797171969685e-07,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 6.253797171969685e-07,
"eval_steps_per_second": 0.029,
"step": 100
},
{
"epoch": 0.2403846153846154,
"step": 100,
"train_probe_calibration/aurc": 0.2912553971506111,
"train_probe_calibration/batch_distribution_entropy": 0.4504727867061737,
"train_probe_calibration/batch_entropy_100bins": 0.23646639818152504,
"train_probe_calibration/batch_entropy_10bins": 0.4504727867061737,
"train_probe_calibration/batch_entropy_50bins": 0.2783644179450236,
"train_probe_calibration/batch_uniqueness": 0.21059908012963577,
"train_probe_calibration/confidence_entropy": 0.6113843826161691,
"train_probe_calibration/coverage@0%": 0.003524229074889868,
"train_probe_calibration/coverage@1%": 0.003524229074889868,
"train_probe_calibration/coverage@10%": 0.003524229074889868,
"train_probe_calibration/coverage@15%": 0.003524229074889868,
"train_probe_calibration/coverage@20%": 0.15859030837004406,
"train_probe_calibration/coverage@25%": 0.16299559471365638,
"train_probe_calibration/coverage@30%": 0.16299559471365638,
"train_probe_calibration/coverage@5%": 0.003524229074889868,
"train_probe_calibration/distribution_entropy_10": 0.4504727867061737,
"train_probe_calibration/distribution_entropy_100": 0.23646639818152504,
"train_probe_calibration/ece": 0.024537444933920727,
"train_probe_calibration/mean_confidence": 0.6848017621145375,
"train_probe_calibration/unique_confidence_per_question": 0.008680555555555556,
"train_probe_calibration/unique_confidences": 10,
"train_probe_completions/clipped_ratio": 0.013020833333333334,
"train_probe_completions/max_length": 1784.3333333333333,
"train_probe_completions/max_terminated_length": 1784.3333333333333,
"train_probe_completions/mean_length": 630.1005452473959,
"train_probe_completions/mean_terminated_length": 638.4372863769531,
"train_probe_completions/min_length": 31.0,
"train_probe_completions/min_terminated_length": 199.0,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 196021611.0,
"train_probe_reward": 1.2102249264717102,
"train_probe_reward_std": 0.3475934366385142,
"train_probe_rewards/accuracy_reward": 0.6631944477558136,
"train_probe_rewards/brier_reward": 0.7719986935456594,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9852430522441864,
"train_probe_rewards/mean_confidence_reward": 0.6746961772441864,
"train_probe_runtime": 205.5697,
"train_probe_samples_per_second": 4.865,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.4342447866996129,
"train_probe_signal/accuracy_reward/group_std_mean": 0.47248346110184986,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.21712239334980646,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.21712239334980646,
"train_probe_signal/advantage_abs_mean": 0.30893393854300183,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.30893393854300183,
"train_probe_signal/advantage_pre_scale_std": 0.3456893265247345,
"train_probe_signal/advantage_std": 0.3456893265247345,
"train_probe_signal/brier_reward/centered_abs_mean": 0.16557522614796957,
"train_probe_signal/brier_reward/group_std_mean": 0.1985683018962542,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08278761307398479,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.08278761307398479,
"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": 0.028157552083333332,
"train_probe_signal/format_reward/group_std_mean": 0.0715202484279871,
"train_probe_signal/format_reward/group_zero_std_frac": 0.6388889054457346,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.014078776041666666,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.014078776041666666,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.06591253789762656,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.09998770679036777,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.59125400185682e-07,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 6.59125400185682e-07,
"train_probe_steps_per_second": 0.029
},
{
"calibration/aurc": 0.26389656392228356,
"calibration/batch_distribution_entropy": 0.45371543209103865,
"calibration/batch_entropy_100bins": 0.23856088528938027,
"calibration/batch_entropy_10bins": 0.45371543209103865,
"calibration/batch_entropy_50bins": 0.28083001427987325,
"calibration/batch_uniqueness": 0.20227893860033813,
"calibration/confidence_entropy": 0.6076272761171853,
"calibration/coverage@0%": 0.003161061382141975,
"calibration/coverage@1%": 0.003161061382141975,
"calibration/coverage@10%": 0.03545272804880864,
"calibration/coverage@15%": 0.066505359627756,
"calibration/coverage@20%": 0.1984970699244401,
"calibration/coverage@25%": 0.442561798088433,
"calibration/coverage@30%": 0.5094067178745292,
"calibration/coverage@5%": 0.003161061382141975,
"calibration/distribution_entropy_10": 0.45371543209103865,
"calibration/distribution_entropy_100": 0.23856088528938027,
"calibration/ece": 0.07547122674488144,
"calibration/mean_confidence": 0.6852840604168908,
"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": 0.011979166666666674,
"completions/max_length": 3383.4,
"completions/max_terminated_length": 3383.4,
"completions/mean_length": 634.3682373046875,
"completions/mean_terminated_length": 642.1746948242187,
"completions/min_length": 0.0,
"completions/min_terminated_length": 176.6,
"epoch": 0.25240384615384615,
"grad_norm": 0.0006580045446753502,
"learning_rate": 2.5240384615384618e-06,
"loss": -0.0119,
"num_tokens": 206421181.0,
"reward": 1.2218866109848023,
"reward_std": 0.19819111227989197,
"rewards/accuracy_reward": 0.6784722208976746,
"rewards/brier_reward": 0.7780476689338685,
"rewards/confidence_one_or_zero": 0.0006076388992369175,
"rewards/format_reward": 0.9872395873069764,
"rewards/mean_confidence_reward": 0.6806759357452392,
"sampling/batch_mean_priority_error": 0.011409722222222226,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.0002556914696469903,
"sampling/priority_kl": 0.030000254139304162,
"sampling/priority_scale": 1.098377120634541,
"sampling/prob_entropy": 10.278963851928712,
"sampling/prob_max": 3.757354352273978e-05,
"sampling/prob_min": 1.2527930266514886e-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": 0.191796875,
"signal/accuracy_reward/group_std_mean": 0.2503849893808365,
"signal/accuracy_reward/group_zero_std_frac": 0.3000000059604645,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0958984375,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0958984375,
"signal/advantage_abs_mean": 0.14796263575553895,
"signal/advantage_pre_scale_abs_mean": 0.14796263575553895,
"signal/advantage_pre_scale_std": 0.23866211771965026,
"signal/advantage_std": 0.23866211771965026,
"signal/brier_reward/centered_abs_mean": 0.09807889461517334,
"signal/brier_reward/group_std_mean": 0.12925267368555068,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04903944730758667,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04903944730758667,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0011338975746184587,
"signal/confidence_one_or_zero/group_std_mean": 0.002487868396565318,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9888888835906983,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.133897491456537e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.133897491456537e-08,
"signal/format_reward/centered_abs_mean": 0.021641710214316844,
"signal/format_reward/group_std_mean": 0.0401812169700861,
"signal/format_reward/group_zero_std_frac": 0.8361111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010820855107158422,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010820855107158422,
"signal/mean_confidence_reward/centered_abs_mean": 0.0599391832947731,
"signal/mean_confidence_reward/group_std_mean": 0.08279892951250076,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.993918307467538e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.993918307467538e-07,
"step": 105
},
{
"calibration/aurc": 0.26641149139303594,
"calibration/batch_distribution_entropy": 0.4376444450609268,
"calibration/batch_entropy_100bins": 0.22824844943430406,
"calibration/batch_entropy_10bins": 0.4376444450609268,
"calibration/batch_entropy_50bins": 0.26869038164509174,
"calibration/batch_uniqueness": 0.15933272420322098,
"calibration/confidence_entropy": 0.5984371837654384,
"calibration/coverage@0%": 0.0031290796344647515,
"calibration/coverage@1%": 0.0031290796344647515,
"calibration/coverage@10%": 0.0031290796344647515,
"calibration/coverage@15%": 0.1655629420683272,
"calibration/coverage@20%": 0.3696741775111553,
"calibration/coverage@25%": 0.5557291666666667,
"calibration/coverage@30%": 0.6296296296296295,
"calibration/coverage@5%": 0.0031290796344647515,
"calibration/distribution_entropy_10": 0.4376444450609268,
"calibration/distribution_entropy_100": 0.22824844943430406,
"calibration/ece": 0.10459396096814946,
"calibration/mean_confidence": 0.6990674712288072,
"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": 0.012934027777777768,
"completions/max_length": 3801.6,
"completions/max_terminated_length": 3801.6,
"completions/mean_length": 622.4458251953125,
"completions/mean_terminated_length": 630.5509155273437,
"completions/min_length": 0.0,
"completions/min_terminated_length": 186.0,
"epoch": 0.2644230769230769,
"grad_norm": 0.0007942788652144372,
"learning_rate": 2.6442307692307696e-06,
"loss": -0.0128,
"num_tokens": 216648717.0,
"reward": 1.2171909093856812,
"reward_std": 0.19215842187404633,
"rewards/accuracy_reward": 0.670312511920929,
"rewards/brier_reward": 0.777683699131012,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9863715291023254,
"rewards/mean_confidence_reward": 0.6971328139305115,
"sampling/batch_mean_priority_error": 0.010491319444444447,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8666666666666666,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00026847636327147483,
"sampling/priority_kl": 0.029999426752328872,
"sampling/priority_scale": 1.0725965857040136,
"sampling/prob_entropy": 10.278949737548828,
"sampling/prob_max": 3.771020783460699e-05,
"sampling/prob_min": 1.2901787158625666e-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": 0.17727864384651185,
"signal/accuracy_reward/group_std_mean": 0.23305801749229432,
"signal/accuracy_reward/group_zero_std_frac": 0.34722222089767457,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08863932192325592,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08863932192325592,
"signal/advantage_abs_mean": 0.14164598286151886,
"signal/advantage_pre_scale_abs_mean": 0.14164598286151886,
"signal/advantage_pre_scale_std": 0.2351542145013809,
"signal/advantage_std": 0.2351542145013809,
"signal/brier_reward/centered_abs_mean": 0.09551279246807098,
"signal/brier_reward/group_std_mean": 0.12876519858837127,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04775639623403549,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04775639623403549,
"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": 0.023551432229578496,
"signal/format_reward/group_std_mean": 0.04473615065217018,
"signal/format_reward/group_zero_std_frac": 0.8166666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.011775716114789248,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.011775716114789248,
"signal/mean_confidence_reward/centered_abs_mean": 0.057459690421819684,
"signal/mean_confidence_reward/group_std_mean": 0.0822980672121048,
"signal/mean_confidence_reward/group_zero_std_frac": 0.002777777798473835,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.745968792325584e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.745968792325584e-07,
"step": 110
},
{
"calibration/aurc": 0.24452289289161647,
"calibration/batch_distribution_entropy": 0.3926279866114429,
"calibration/batch_entropy_100bins": 0.20270615798497577,
"calibration/batch_entropy_10bins": 0.3926279866114429,
"calibration/batch_entropy_50bins": 0.23862240942175567,
"calibration/batch_uniqueness": 0.027939809524415986,
"calibration/confidence_entropy": 0.5924180543458887,
"calibration/coverage@0%": 0.0015950943282864696,
"calibration/coverage@1%": 0.0015950943282864696,
"calibration/coverage@10%": 0.049489831170391727,
"calibration/coverage@15%": 0.049489831170391727,
"calibration/coverage@20%": 0.26600106589988565,
"calibration/coverage@25%": 0.6319309239288837,
"calibration/coverage@30%": 0.6708350335179248,
"calibration/coverage@5%": 0.049489831170391727,
"calibration/distribution_entropy_10": 0.3926279866114429,
"calibration/distribution_entropy_100": 0.20270615798497577,
"calibration/ece": 0.07019371319484866,
"calibration/mean_confidence": 0.7079467487920142,
"calibration/unique_confidence_per_question": 0.01614583333333333,
"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": 0.019010416666666675,
"completions/max_length": 3956.6,
"completions/max_terminated_length": 3956.6,
"completions/mean_length": 638.0852416992187,
"completions/mean_terminated_length": 650.6302490234375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 160.8,
"epoch": 0.2764423076923077,
"grad_norm": 0.0006896952399984002,
"learning_rate": 2.7644230769230775e-06,
"loss": -0.0187,
"num_tokens": 227081667.0,
"reward": 1.2036991119384766,
"reward_std": 0.19953635036945344,
"rewards/accuracy_reward": 0.6560763955116272,
"rewards/brier_reward": 0.7710996150970459,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9802083253860474,
"rewards/mean_confidence_reward": 0.6872612833976746,
"sampling/batch_mean_priority_error": 0.011953125000000004,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8805555555555555,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00028192729223519564,
"sampling/priority_kl": 0.029999881237745284,
"sampling/priority_scale": 1.0489887594711036,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 3.784664804697968e-05,
"sampling/prob_min": 1.3257731006888207e-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": 0.18827039897441863,
"signal/accuracy_reward/group_std_mean": 0.24098731875419616,
"signal/accuracy_reward/group_zero_std_frac": 0.3416666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.09413519948720932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.09413519948720932,
"signal/advantage_abs_mean": 0.15093255639076233,
"signal/advantage_pre_scale_abs_mean": 0.15093255639076233,
"signal/advantage_pre_scale_std": 0.24881745576858522,
"signal/advantage_std": 0.24881745576858522,
"signal/brier_reward/centered_abs_mean": 0.09960789233446121,
"signal/brier_reward/group_std_mean": 0.131089448928833,
"signal/brier_reward/group_zero_std_frac": 0.002777777798473835,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.049803946167230606,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.049803946167230606,
"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": 0.03161892369389534,
"signal/format_reward/group_std_mean": 0.05357316210865974,
"signal/format_reward/group_zero_std_frac": 0.8,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01580946184694767,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01580946184694767,
"signal/mean_confidence_reward/centered_abs_mean": 0.05904405862092972,
"signal/mean_confidence_reward/group_std_mean": 0.08394580483436584,
"signal/mean_confidence_reward/group_zero_std_frac": 0.00555555559694767,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.904405384171696e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.904405384171696e-07,
"step": 115
},
{
"calibration/aurc": 0.29914037716218794,
"calibration/batch_distribution_entropy": 0.3724276398058252,
"calibration/batch_entropy_100bins": 0.1926613136182875,
"calibration/batch_entropy_10bins": 0.3724276398058252,
"calibration/batch_entropy_50bins": 0.22679778115751045,
"calibration/batch_uniqueness": -0.008368241406920666,
"calibration/confidence_entropy": 0.6073866082278495,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.02888283378746594,
"calibration/coverage@15%": 0.07471616712079927,
"calibration/coverage@20%": 0.1263931959166108,
"calibration/coverage@25%": 0.1390916086150235,
"calibration/coverage@30%": 0.4805212748275576,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.3724276398058252,
"calibration/distribution_entropy_100": 0.1926613136182875,
"calibration/ece": 0.06616926435315493,
"calibration/mean_confidence": 0.6935359204897231,
"calibration/unique_confidence_per_question": 0.011979166666666666,
"calibration/unique_confidences": 4.6,
"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": 0.01440972222222221,
"completions/max_length": 3441.8,
"completions/max_terminated_length": 3441.8,
"completions/mean_length": 615.9207641601563,
"completions/mean_terminated_length": 625.0067504882812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 181.8,
"epoch": 0.28846153846153844,
"grad_norm": 0.0009943849872797728,
"learning_rate": 2.8846153846153845e-06,
"loss": -0.0133,
"num_tokens": 237272690.0,
"reward": 1.2263082265853882,
"reward_std": 0.18344857096672057,
"rewards/accuracy_reward": 0.6831597208976745,
"rewards/brier_reward": 0.78411306142807,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9853298544883728,
"rewards/mean_confidence_reward": 0.6838411569595337,
"sampling/batch_mean_priority_error": 0.013315972222222222,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8416666666666668,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.0002971566456835717,
"sampling/priority_kl": 0.029999834671616554,
"sampling/priority_scale": 1.0274231314193458,
"sampling/prob_entropy": 10.278933334350587,
"sampling/prob_max": 3.79780241928529e-05,
"sampling/prob_min": 1.2208350563014392e-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": 0.17330729067325593,
"signal/accuracy_reward/group_std_mean": 0.22918421924114227,
"signal/accuracy_reward/group_zero_std_frac": 0.347222226858139,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08665364533662796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08665364533662796,
"signal/advantage_abs_mean": 0.13476323038339616,
"signal/advantage_pre_scale_abs_mean": 0.13476323038339616,
"signal/advantage_pre_scale_std": 0.22975938022136688,
"signal/advantage_std": 0.22975938022136688,
"signal/brier_reward/centered_abs_mean": 0.08760800659656524,
"signal/brier_reward/group_std_mean": 0.11845918893814086,
"signal/brier_reward/group_zero_std_frac": 0.002777777798473835,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04380400329828262,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04380400329828262,
"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": 0.02306315116584301,
"signal/format_reward/group_std_mean": 0.03991239666938782,
"signal/format_reward/group_zero_std_frac": 0.8472222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.011531575582921505,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.011531575582921505,
"signal/mean_confidence_reward/centered_abs_mean": 0.051366379112005235,
"signal/mean_confidence_reward/group_std_mean": 0.07370935529470443,
"signal/mean_confidence_reward/group_zero_std_frac": 0.002777777798473835,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.136637696523394e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.136637696523394e-07,
"step": 120
},
{
"calibration/aurc": 0.28633294767991674,
"calibration/batch_distribution_entropy": 0.4130528797346272,
"calibration/batch_entropy_100bins": 0.21892998572305739,
"calibration/batch_entropy_10bins": 0.4130528797346272,
"calibration/batch_entropy_50bins": 0.2577208369356918,
"calibration/batch_uniqueness": 0.10911798642188608,
"calibration/confidence_entropy": 0.6022333939945146,
"calibration/coverage@0%": 0.0005208333333333333,
"calibration/coverage@1%": 0.0005208333333333333,
"calibration/coverage@10%": 0.022452948215839864,
"calibration/coverage@15%": 0.0445582113737346,
"calibration/coverage@20%": 0.1449467695747928,
"calibration/coverage@25%": 0.43465486633249795,
"calibration/coverage@30%": 0.46481359649122805,
"calibration/coverage@5%": 0.0005208333333333333,
"calibration/distribution_entropy_10": 0.4130528797346272,
"calibration/distribution_entropy_100": 0.21892998572305739,
"calibration/ece": 0.08499723916874016,
"calibration/mean_confidence": 0.696090644907764,
"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": 0.013888888888888885,
"completions/max_length": 3665.8,
"completions/max_terminated_length": 3665.8,
"completions/mean_length": 659.197314453125,
"completions/mean_terminated_length": 668.5060913085938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 202.4,
"epoch": 0.3004807692307692,
"grad_norm": 0.0009211119031533599,
"learning_rate": 3.0048076923076923e-06,
"loss": -0.0125,
"num_tokens": 248003059.0,
"reward": 1.1965364456176757,
"reward_std": 0.19677658081054689,
"rewards/accuracy_reward": 0.6400173664093017,
"rewards/brier_reward": 0.7674516320228577,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9855902671813965,
"rewards/mean_confidence_reward": 0.6775911569595336,
"sampling/batch_mean_priority_error": 0.01447222222222222,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8083333333333333,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.0003136721788905561,
"sampling/priority_kl": 0.02999979183077812,
"sampling/priority_scale": 1.0074425875907764,
"sampling/prob_entropy": 10.27892837524414,
"sampling/prob_max": 3.810692724073306e-05,
"sampling/prob_min": 1.2523106124717743e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.29520000219345094,
"sampling/prompt_draws_total": 8856.0,
"sampling/seen_fraction": 0.27564666271209715,
"sampling/unseen_fraction": 0.7243533372879029,
"signal/accuracy_reward/centered_abs_mean": 0.19204101264476775,
"signal/accuracy_reward/group_std_mean": 0.24852104783058165,
"signal/accuracy_reward/group_zero_std_frac": 0.3083333343267441,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.09602050632238388,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.09602050632238388,
"signal/advantage_abs_mean": 0.14804043769836425,
"signal/advantage_pre_scale_abs_mean": 0.14804043769836425,
"signal/advantage_pre_scale_std": 0.23920418024063111,
"signal/advantage_std": 0.23920418024063111,
"signal/brier_reward/centered_abs_mean": 0.0960723802447319,
"signal/brier_reward/group_std_mean": 0.1261401355266571,
"signal/brier_reward/group_zero_std_frac": 0.008333333395421505,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04803619012236595,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04803619012236595,
"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": 0.023676214925944804,
"signal/format_reward/group_std_mean": 0.042564719542860986,
"signal/format_reward/group_zero_std_frac": 0.8305555701255798,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.011838107462972402,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.011838107462972402,
"signal/mean_confidence_reward/centered_abs_mean": 0.05564128011465073,
"signal/mean_confidence_reward/group_std_mean": 0.07778785824775696,
"signal/mean_confidence_reward/group_zero_std_frac": 0.013888888992369175,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.564127889101656e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.564127889101656e-07,
"step": 125
},
{
"calibration/aurc": 0.2770300314859839,
"calibration/batch_distribution_entropy": 0.4395422044627269,
"calibration/batch_entropy_100bins": 0.24320209729782186,
"calibration/batch_entropy_10bins": 0.4395422044627269,
"calibration/batch_entropy_50bins": 0.2862935739620296,
"calibration/batch_uniqueness": 0.21519033964356288,
"calibration/confidence_entropy": 0.6144014297973173,
"calibration/coverage@0%": 0.0005305039787798408,
"calibration/coverage@1%": 0.0005305039787798408,
"calibration/coverage@10%": 0.056487505235236636,
"calibration/coverage@15%": 0.18665173868483526,
"calibration/coverage@20%": 0.36770437026378266,
"calibration/coverage@25%": 0.4574792010415144,
"calibration/coverage@30%": 0.7263157894736842,
"calibration/coverage@5%": 0.035267346084043,
"calibration/distribution_entropy_10": 0.4395422044627269,
"calibration/distribution_entropy_100": 0.24320209729782186,
"calibration/ece": 0.1087024526647045,
"calibration/mean_confidence": 0.6762298586141063,
"calibration/unique_confidence_per_question": 0.015104166666666665,
"calibration/unique_confidences": 5.8,
"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": 0.01857638888888886,
"completions/max_length": 3667.8,
"completions/max_terminated_length": 3667.8,
"completions/mean_length": 665.5612915039062,
"completions/mean_terminated_length": 678.1058227539063,
"completions/min_length": 0.0,
"completions/min_terminated_length": 205.8,
"epoch": 0.3125,
"grad_norm": 0.0007662621792405844,
"learning_rate": 3.125e-06,
"loss": -0.0174,
"num_tokens": 258767733.0,
"reward": 1.2184621334075927,
"reward_std": 0.18495941460132598,
"rewards/accuracy_reward": 0.6723090171813965,
"rewards/brier_reward": 0.783264982700348,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.981336796283722,
"rewards/mean_confidence_reward": 0.664318585395813,
"sampling/batch_mean_priority_error": 0.015236111111111112,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8055555555555556,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00033118717256002127,
"sampling/priority_kl": 0.03000081516802311,
"sampling/priority_scale": 0.9889760912163184,
"sampling/prob_entropy": 10.27894229888916,
"sampling/prob_max": 3.823187798843719e-05,
"sampling/prob_min": 1.2823050019505899e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.3072000026702881,
"sampling/prompt_draws_total": 9216.0,
"sampling/seen_fraction": 0.28531333804130554,
"sampling/unseen_fraction": 0.7146866619586945,
"signal/accuracy_reward/centered_abs_mean": 0.17028537392616272,
"signal/accuracy_reward/group_std_mean": 0.22526440620422364,
"signal/accuracy_reward/group_zero_std_frac": 0.3555555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08514268696308136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08514268696308136,
"signal/advantage_abs_mean": 0.13408068269491197,
"signal/advantage_pre_scale_abs_mean": 0.13408068269491197,
"signal/advantage_pre_scale_std": 0.22951894402503967,
"signal/advantage_std": 0.22951894402503967,
"signal/brier_reward/centered_abs_mean": 0.08635973781347275,
"signal/brier_reward/group_std_mean": 0.11832628399133682,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04317986890673638,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04317986890673638,
"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": 0.03133138045668602,
"signal/format_reward/group_std_mean": 0.055094408243894576,
"signal/format_reward/group_zero_std_frac": 0.7888888835906982,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01566569022834301,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01566569022834301,
"signal/mean_confidence_reward/centered_abs_mean": 0.06074300110340118,
"signal/mean_confidence_reward/group_std_mean": 0.0846795991063118,
"signal/mean_confidence_reward/group_zero_std_frac": 0.00555555559694767,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.074300131331257e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.074300131331257e-07,
"step": 130
},
{
"calibration/aurc": 0.3045633905704531,
"calibration/batch_distribution_entropy": 0.4904286109595712,
"calibration/batch_entropy_100bins": 0.27985768204534994,
"calibration/batch_entropy_10bins": 0.4904286109595712,
"calibration/batch_entropy_50bins": 0.3294439352444273,
"calibration/batch_uniqueness": 0.3409336534741945,
"calibration/confidence_entropy": 0.612664591868816,
"calibration/coverage@0%": 0.0010471275946903505,
"calibration/coverage@1%": 0.0010471275946903505,
"calibration/coverage@10%": 0.04461668140046463,
"calibration/coverage@15%": 0.12446196389804717,
"calibration/coverage@20%": 0.25337443074154853,
"calibration/coverage@25%": 0.3707428517941801,
"calibration/coverage@30%": 0.5226702799638021,
"calibration/coverage@5%": 0.0010471275946903505,
"calibration/distribution_entropy_10": 0.4904286109595712,
"calibration/distribution_entropy_100": 0.27985768204534994,
"calibration/ece": 0.1163251270119496,
"calibration/mean_confidence": 0.6686119593374754,
"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": 0.018576388888888885,
"completions/max_length": 3678.8,
"completions/max_terminated_length": 3678.8,
"completions/mean_length": 694.5222412109375,
"completions/mean_terminated_length": 707.6533081054688,
"completions/min_length": 0.0,
"completions/min_terminated_length": 206.8,
"epoch": 0.3245192307692308,
"grad_norm": 0.0012281141243875027,
"learning_rate": 3.245192307692308e-06,
"loss": -0.0183,
"num_tokens": 269858293.0,
"reward": 1.2133918523788452,
"reward_std": 0.19259853959083556,
"rewards/accuracy_reward": 0.6637152910232544,
"rewards/brier_reward": 0.7818053007125855,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.981249988079071,
"rewards/mean_confidence_reward": 0.6537803769111633,
"sampling/batch_mean_priority_error": 0.014315972222222223,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8055555555555556,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00034877662546932696,
"sampling/priority_kl": 0.030000032857060432,
"sampling/priority_scale": 0.9717868029838428,
"sampling/prob_entropy": 10.278975677490234,
"sampling/prob_max": 3.8358993333531546e-05,
"sampling/prob_min": 1.3112282431393396e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.31919999718666076,
"sampling/prompt_draws_total": 9576.0,
"sampling/seen_fraction": 0.29510666728019713,
"sampling/unseen_fraction": 0.7048933327198028,
"signal/accuracy_reward/centered_abs_mean": 0.17950303554534913,
"signal/accuracy_reward/group_std_mean": 0.23662585020065308,
"signal/accuracy_reward/group_zero_std_frac": 0.33055556416511533,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08975151777267457,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08975151777267457,
"signal/advantage_abs_mean": 0.14017547965049743,
"signal/advantage_pre_scale_abs_mean": 0.14017547965049743,
"signal/advantage_pre_scale_std": 0.23439024090766908,
"signal/advantage_std": 0.23439024090766908,
"signal/brier_reward/centered_abs_mean": 0.08922516703605651,
"signal/brier_reward/group_std_mean": 0.12257051169872284,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.044612583518028257,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.044612583518028257,
"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": 0.03119574598968029,
"signal/format_reward/group_std_mean": 0.05646200627088547,
"signal/format_reward/group_zero_std_frac": 0.7750000119209289,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.015597872994840145,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.015597872994840145,
"signal/mean_confidence_reward/centered_abs_mean": 0.0654410794377327,
"signal/mean_confidence_reward/group_std_mean": 0.09104419052600861,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.544107804984378e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.544107804984378e-07,
"step": 135
},
{
"calibration/aurc": 0.18749490988916556,
"calibration/batch_distribution_entropy": 0.5415738700817532,
"calibration/batch_entropy_100bins": 0.32182948863917904,
"calibration/batch_entropy_10bins": 0.5415738700817532,
"calibration/batch_entropy_50bins": 0.37885246686854196,
"calibration/batch_uniqueness": 0.41862065299611595,
"calibration/confidence_entropy": 0.5883915406218511,
"calibration/coverage@0%": 0.008034688346883468,
"calibration/coverage@1%": 0.008034688346883468,
"calibration/coverage@10%": 0.1927743034213184,
"calibration/coverage@15%": 0.2530638476572433,
"calibration/coverage@20%": 0.5535869100406655,
"calibration/coverage@25%": 0.7756670084830667,
"calibration/coverage@30%": 0.9135496445544167,
"calibration/coverage@5%": 0.09550135501355014,
"calibration/distribution_entropy_10": 0.5415738700817532,
"calibration/distribution_entropy_100": 0.32182948863917904,
"calibration/ece": 0.07815413711451906,
"calibration/mean_confidence": 0.6764555789425304,
"calibration/unique_confidence_per_question": 0.02395833333333333,
"calibration/unique_confidences": 9.2,
"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": 0.028038194444444463,
"completions/max_length": 3858.4,
"completions/max_terminated_length": 3858.4,
"completions/mean_length": 683.644970703125,
"completions/mean_terminated_length": 703.34462890625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 201.8,
"epoch": 0.33653846153846156,
"grad_norm": 0.0006961016333661973,
"learning_rate": 3.365384615384616e-06,
"loss": -0.0251,
"num_tokens": 280792859.0,
"reward": 1.236612319946289,
"reward_std": 0.19623805284500123,
"rewards/accuracy_reward": 0.7059027791023255,
"rewards/brier_reward": 0.7958673954010009,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9714409589767456,
"rewards/mean_confidence_reward": 0.6710633635520935,
"sampling/batch_mean_priority_error": 0.011197916666666667,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.825,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.00036346286069601776,
"sampling/priority_kl": 0.030000948533415794,
"sampling/priority_scale": 0.9559410989983007,
"sampling/prob_entropy": 10.278955841064453,
"sampling/prob_max": 3.8486441917484625e-05,
"sampling/prob_min": 1.3388117440626957e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.3312000036239624,
"sampling/prompt_draws_total": 9936.0,
"sampling/seen_fraction": 0.30485999584198,
"sampling/unseen_fraction": 0.69514000415802,
"signal/accuracy_reward/centered_abs_mean": 0.17124565839767455,
"signal/accuracy_reward/group_std_mean": 0.22862593531608583,
"signal/accuracy_reward/group_zero_std_frac": 0.3444444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08562282919883728,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08562282919883728,
"signal/advantage_abs_mean": 0.14275099337100983,
"signal/advantage_pre_scale_abs_mean": 0.14275099337100983,
"signal/advantage_pre_scale_std": 0.24261143207550048,
"signal/advantage_std": 0.24261143207550048,
"signal/brier_reward/centered_abs_mean": 0.09853724390268326,
"signal/brier_reward/group_std_mean": 0.1336524322628975,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04926862195134163,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04926862195134163,
"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": 0.04031575620174408,
"signal/format_reward/group_std_mean": 0.06712066158652305,
"signal/format_reward/group_zero_std_frac": 0.7472222328186036,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.02015787810087204,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.02015787810087204,
"signal/mean_confidence_reward/centered_abs_mean": 0.07758951485157013,
"signal/mean_confidence_reward/group_std_mean": 0.10319010466337204,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.758951369396527e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.758951369396527e-07,
"step": 140
},
{
"calibration/aurc": 0.22960033807496577,
"calibration/batch_distribution_entropy": 0.4959596321391365,
"calibration/batch_entropy_100bins": 0.2828537847475544,
"calibration/batch_entropy_10bins": 0.4959596321391365,
"calibration/batch_entropy_50bins": 0.3329708988689269,
"calibration/batch_uniqueness": 0.29531464718269096,
"calibration/confidence_entropy": 0.5656490610797936,
"calibration/coverage@0%": 0.013919308526411784,
"calibration/coverage@1%": 0.013919308526411784,
"calibration/coverage@10%": 0.013919308526411784,
"calibration/coverage@15%": 0.21862042889332853,
"calibration/coverage@20%": 0.45811751394931627,
"calibration/coverage@25%": 0.793972602739726,
"calibration/coverage@30%": 0.8,
"calibration/coverage@5%": 0.013919308526411784,
"calibration/distribution_entropy_10": 0.4959596321391365,
"calibration/distribution_entropy_100": 0.2828537847475544,
"calibration/ece": 0.07045841533199215,
"calibration/mean_confidence": 0.7150147551275411,
"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": 0.02161458333333337,
"completions/max_length": 3232.6,
"completions/max_terminated_length": 3232.6,
"completions/mean_length": 680.2881103515625,
"completions/mean_terminated_length": 695.2589965820313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 211.4,
"epoch": 0.3485576923076923,
"grad_norm": 0.0007520325016230345,
"learning_rate": 3.4855769230769233e-06,
"loss": -0.0219,
"num_tokens": 291734930.0,
"reward": 1.227646541595459,
"reward_std": 0.1878492623567581,
"rewards/accuracy_reward": 0.6833333373069763,
"rewards/brier_reward": 0.7935603260993958,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9783854126930237,
"rewards/mean_confidence_reward": 0.697022533416748,
"sampling/batch_mean_priority_error": 0.010859374999999998,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8111111111111111,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.0003761312167625874,
"sampling/priority_kl": 0.02999988608062267,
"sampling/priority_scale": 0.94111793639604,
"sampling/prob_entropy": 10.278964805603028,
"sampling/prob_max": 3.8614498043898494e-05,
"sampling/prob_min": 1.3654396025231107e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.3431999981403351,
"sampling/prompt_draws_total": 10296.0,
"sampling/seen_fraction": 0.31462000012397767,
"sampling/unseen_fraction": 0.6853799998760224,
"signal/accuracy_reward/centered_abs_mean": 0.16524522602558137,
"signal/accuracy_reward/group_std_mean": 0.2168790102005005,
"signal/accuracy_reward/group_zero_std_frac": 0.3888888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08262261301279068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08262261301279068,
"signal/advantage_abs_mean": 0.13840331137180328,
"signal/advantage_pre_scale_abs_mean": 0.13840331137180328,
"signal/advantage_pre_scale_std": 0.23685503900051116,
"signal/advantage_std": 0.23685503900051116,
"signal/brier_reward/centered_abs_mean": 0.099394890666008,
"signal/brier_reward/group_std_mean": 0.13434166312217713,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.049697445333004,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.049697445333004,
"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": 0.03282877579331398,
"signal/format_reward/group_std_mean": 0.057800959795713425,
"signal/format_reward/group_zero_std_frac": 0.7722222208976746,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01641438789665699,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01641438789665699,
"signal/mean_confidence_reward/centered_abs_mean": 0.07129204124212266,
"signal/mean_confidence_reward/group_std_mean": 0.09770053625106812,
"signal/mean_confidence_reward/group_zero_std_frac": 0.002777777798473835,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.129204050215776e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.129204050215776e-07,
"step": 145
},
{
"calibration/aurc": 0.22222502440396444,
"calibration/batch_distribution_entropy": 0.44819059016372265,
"calibration/batch_entropy_100bins": 0.26214575431917136,
"calibration/batch_entropy_10bins": 0.44819059016372265,
"calibration/batch_entropy_50bins": 0.3085937405017596,
"calibration/batch_uniqueness": 0.12955742353336192,
"calibration/confidence_entropy": 0.5445938599239366,
"calibration/coverage@0%": 0.003794037940379404,
"calibration/coverage@1%": 0.003794037940379404,
"calibration/coverage@10%": 0.01745031027333772,
"calibration/coverage@15%": 0.16208621165564271,
"calibration/coverage@20%": 0.1759586285469275,
"calibration/coverage@25%": 0.7342802841846832,
"calibration/coverage@30%": 0.8715341334626141,
"calibration/coverage@5%": 0.003794037940379404,
"calibration/distribution_entropy_10": 0.44819059016372265,
"calibration/distribution_entropy_100": 0.26214575431917136,
"calibration/ece": 0.06575164506779053,
"calibration/mean_confidence": 0.7346546095304988,
"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": 0.024131944444444442,
"completions/max_length": 3942.8,
"completions/max_terminated_length": 3942.8,
"completions/mean_length": 710.7791748046875,
"completions/mean_terminated_length": 728.4054809570313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 218.8,
"epoch": 0.3605769230769231,
"grad_norm": 0.0008258689776994288,
"learning_rate": 3.605769230769231e-06,
"loss": -0.0233,
"num_tokens": 303037570.0,
"reward": 1.235294795036316,
"reward_std": 0.18425678908824922,
"rewards/accuracy_reward": 0.6968750119209289,
"rewards/brier_reward": 0.7978322505950928,
"rewards/confidence_one_or_zero": 0.00026041667442768814,
"rewards/format_reward": 0.9758680462837219,
"rewards/mean_confidence_reward": 0.7146310806274414,
"sampling/batch_mean_priority_error": 0.008072916666666667,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7638888888888888,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.0003874113957863301,
"sampling/priority_kl": 0.030000269412994385,
"sampling/priority_scale": 0.9272584259742871,
"sampling/prob_entropy": 10.278953742980956,
"sampling/prob_max": 3.8738126750104127e-05,
"sampling/prob_min": 1.3909414155932609e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.35520000457763673,
"sampling/prompt_draws_total": 10656.0,
"sampling/seen_fraction": 0.3240066647529602,
"sampling/unseen_fraction": 0.6759933352470398,
"signal/accuracy_reward/centered_abs_mean": 0.15432942807674407,
"signal/accuracy_reward/group_std_mean": 0.20061970353126526,
"signal/accuracy_reward/group_zero_std_frac": 0.4444444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07716471403837204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07716471403837204,
"signal/advantage_abs_mean": 0.13734771311283112,
"signal/advantage_pre_scale_abs_mean": 0.13734771311283112,
"signal/advantage_pre_scale_std": 0.24063621163368226,
"signal/advantage_std": 0.24063621163368226,
"signal/brier_reward/centered_abs_mean": 0.10788649320602417,
"signal/brier_reward/group_std_mean": 0.14216929376125337,
"signal/brier_reward/group_zero_std_frac": 0.008333333395421505,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.053943246603012085,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.053943246603012085,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0005045572761446238,
"signal/confidence_one_or_zero/group_std_mean": 0.0014731390401721,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/format_reward/centered_abs_mean": 0.03764105886220932,
"signal/format_reward/group_std_mean": 0.059189148247241974,
"signal/format_reward/group_zero_std_frac": 0.794444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01882052943110466,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01882052943110466,
"signal/mean_confidence_reward/centered_abs_mean": 0.07287055253982544,
"signal/mean_confidence_reward/group_std_mean": 0.0999222919344902,
"signal/mean_confidence_reward/group_zero_std_frac": 0.008333333395421505,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.28705470010027e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.28705470010027e-07,
"step": 150
},
{
"epoch": 0.3605769230769231,
"eval_calibration/aurc": 0.26528970355274367,
"eval_calibration/batch_distribution_entropy": 0.45167656638554243,
"eval_calibration/batch_entropy_100bins": 0.2686325582166219,
"eval_calibration/batch_entropy_10bins": 0.45167656638554243,
"eval_calibration/batch_entropy_50bins": 0.3162299010942306,
"eval_calibration/batch_uniqueness": 0.07833556234382769,
"eval_calibration/confidence_entropy": 0.5360695238200753,
"eval_calibration/coverage@0%": 0.0008787346221441124,
"eval_calibration/coverage@1%": 0.0008787346221441124,
"eval_calibration/coverage@10%": 0.0008787346221441124,
"eval_calibration/coverage@15%": 0.0008787346221441124,
"eval_calibration/coverage@20%": 0.0008787346221441124,
"eval_calibration/coverage@25%": 0.04569420035149385,
"eval_calibration/coverage@30%": 0.8611599297012302,
"eval_calibration/coverage@5%": 0.0008787346221441124,
"eval_calibration/distribution_entropy_10": 0.45167656638554243,
"eval_calibration/distribution_entropy_100": 0.2686325582166219,
"eval_calibration/ece": 0.06616871704745168,
"eval_calibration/mean_confidence": 0.7304920913884007,
"eval_calibration/unique_confidence_per_question": 0.010416666666666666,
"eval_calibration/unique_confidences": 12,
"eval_completions/clipped_ratio": 0.01128472222222221,
"eval_completions/max_length": 2444.5,
"eval_completions/max_terminated_length": 2444.5,
"eval_completions/mean_length": 721.1014099121094,
"eval_completions/mean_terminated_length": 729.3672587076823,
"eval_completions/min_length": 61.333333333333336,
"eval_completions/min_terminated_length": 276.1666666666667,
"eval_loss": 0.0,
"eval_num_tokens": 303037570.0,
"eval_reward": 1.213831086953481,
"eval_reward_std": 0.362030953168869,
"eval_rewards/accuracy_reward": 0.6562500099341074,
"eval_rewards/brier_reward": 0.783550351858139,
"eval_rewards/confidence_one_or_zero": 0.0008680555814256271,
"eval_rewards/format_reward": 0.987847218910853,
"eval_rewards/mean_confidence_reward": 0.7216145892937978,
"eval_runtime": 204.1612,
"eval_samples_per_second": 4.898,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4382595469554265,
"eval_signal/accuracy_reward/group_std_mean": 0.4749273806810379,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.21912977347771326,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.21912977347771326,
"eval_signal/advantage_abs_mean": 0.32130875686804455,
"eval_signal/advantage_pre_scale_abs_mean": 0.32130875686804455,
"eval_signal/advantage_pre_scale_std": 0.3592623174190521,
"eval_signal/advantage_std": 0.3592623174190521,
"eval_signal/brier_reward/centered_abs_mean": 0.20470946778853735,
"eval_signal/brier_reward/group_std_mean": 0.24981755266586939,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.10235473389426868,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.10235473389426868,
"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": 0.023328993003815413,
"eval_signal/format_reward/group_std_mean": 0.06276767483601968,
"eval_signal/format_reward/group_zero_std_frac": 0.6666666865348816,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.011664496501907706,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.011664496501907706,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.11676974222064018,
"eval_signal/mean_confidence_reward/group_std_mean": 0.15584261218706766,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.1676973485919007e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.1676973485919007e-06,
"eval_steps_per_second": 0.029,
"step": 150
},
{
"epoch": 0.3605769230769231,
"step": 150,
"train_probe_calibration/aurc": 0.22968415288321048,
"train_probe_calibration/batch_distribution_entropy": 0.4644610670848097,
"train_probe_calibration/batch_entropy_100bins": 0.2657689272124461,
"train_probe_calibration/batch_entropy_10bins": 0.4644610670848097,
"train_probe_calibration/batch_entropy_50bins": 0.31285888100927634,
"train_probe_calibration/batch_uniqueness": 0.0961154339441112,
"train_probe_calibration/confidence_entropy": 0.5351932847465817,
"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%": 0.782608695652174,
"train_probe_calibration/coverage@30%": 0.9804791481810116,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.4644610670848097,
"train_probe_calibration/distribution_entropy_100": 0.2657689272124461,
"train_probe_calibration/ece": 0.047249334516415234,
"train_probe_calibration/mean_confidence": 0.7368677905944986,
"train_probe_calibration/unique_confidence_per_question": 0.009548611111111112,
"train_probe_calibration/unique_confidences": 11,
"train_probe_completions/clipped_ratio": 0.019965277777777773,
"train_probe_completions/max_length": 2151.1666666666665,
"train_probe_completions/max_terminated_length": 2151.1666666666665,
"train_probe_completions/mean_length": 720.7236124674479,
"train_probe_completions/mean_terminated_length": 735.2484944661459,
"train_probe_completions/min_length": 41.333333333333336,
"train_probe_completions/min_terminated_length": 247.5,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 303037570.0,
"train_probe_reward": 1.2241229216257732,
"train_probe_reward_std": 0.37155655523141223,
"train_probe_rewards/accuracy_reward": 0.6805555522441864,
"train_probe_rewards/brier_reward": 0.7893771429856619,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9782986044883728,
"train_probe_rewards/mean_confidence_reward": 0.7208767135938009,
"train_probe_runtime": 206.6309,
"train_probe_samples_per_second": 4.84,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.4214409738779068,
"train_probe_signal/accuracy_reward/group_std_mean": 0.46496439973513287,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.2107204869389534,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.2107204869389534,
"train_probe_signal/advantage_abs_mean": 0.31965141991774243,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.31965141991774243,
"train_probe_signal/advantage_pre_scale_std": 0.36928872764110565,
"train_probe_signal/advantage_std": 0.36928872764110565,
"train_probe_signal/brier_reward/centered_abs_mean": 0.20193182677030563,
"train_probe_signal/brier_reward/group_std_mean": 0.2502300019065539,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.10096591338515282,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.10096591338515282,
"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": 0.04150390625,
"train_probe_signal/format_reward/group_std_mean": 0.1078145479162534,
"train_probe_signal/format_reward/group_zero_std_frac": 0.4444444552063942,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.020751953125,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.020751953125,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.11885850255688031,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.16483317812283835,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.1885849744430743e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.1885849744430743e-06,
"train_probe_steps_per_second": 0.029
},
{
"calibration/aurc": 0.2557523680414625,
"calibration/batch_distribution_entropy": 0.4446285204684316,
"calibration/batch_entropy_100bins": 0.2549144802425309,
"calibration/batch_entropy_10bins": 0.4446285204684316,
"calibration/batch_entropy_50bins": 0.30008120165977276,
"calibration/batch_uniqueness": 0.06078989133921503,
"calibration/confidence_entropy": 0.539813055658638,
"calibration/coverage@0%": 0.009716893674677317,
"calibration/coverage@1%": 0.009716893674677317,
"calibration/coverage@10%": 0.009716893674677317,
"calibration/coverage@15%": 0.018093857025462658,
"calibration/coverage@20%": 0.3081061517111193,
"calibration/coverage@25%": 0.48860219392747817,
"calibration/coverage@30%": 0.8685701872786454,
"calibration/coverage@5%": 0.009716893674677317,
"calibration/distribution_entropy_10": 0.4446285204684316,
"calibration/distribution_entropy_100": 0.2549144802425309,
"calibration/ece": 0.08582490386189781,
"calibration/mean_confidence": 0.7317781102423115,
"calibration/unique_confidence_per_question": 0.024479166666666666,
"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": 0.020920138888888905,
"completions/max_length": 4018.8,
"completions/max_terminated_length": 4018.8,
"completions/mean_length": 730.2631958007812,
"completions/mean_terminated_length": 745.8842163085938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 220.2,
"epoch": 0.37259615384615385,
"grad_norm": 0.0007210660842247307,
"learning_rate": 3.725961538461539e-06,
"loss": -0.022,
"num_tokens": 314521082.0,
"reward": 1.2058180570602417,
"reward_std": 0.1962680220603943,
"rewards/accuracy_reward": 0.6491319298744201,
"rewards/brier_reward": 0.7835837721824646,
"rewards/confidence_one_or_zero": 0.000260416668606922,
"rewards/format_reward": 0.97890625,
"rewards/mean_confidence_reward": 0.7079079747200012,
"sampling/batch_mean_priority_error": 0.010718749999999999,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7972222222222223,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.0003989488526713103,
"sampling/priority_kl": 0.030000166222453117,
"sampling/priority_scale": 0.9142814338905737,
"sampling/prob_entropy": 10.278959274291992,
"sampling/prob_max": 3.8861997745698317e-05,
"sampling/prob_min": 1.4154319433146157e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.3671999990940094,
"sampling/prompt_draws_total": 11016.0,
"sampling/seen_fraction": 0.3333600044250488,
"sampling/unseen_fraction": 0.6666399955749511,
"signal/accuracy_reward/centered_abs_mean": 0.17177734375,
"signal/accuracy_reward/group_std_mean": 0.22341654002666472,
"signal/accuracy_reward/group_zero_std_frac": 0.38055555820465087,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.085888671875,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.085888671875,
"signal/advantage_abs_mean": 0.14426402002573013,
"signal/advantage_pre_scale_abs_mean": 0.14426402002573013,
"signal/advantage_pre_scale_std": 0.2449123114347458,
"signal/advantage_std": 0.2449123114347458,
"signal/brier_reward/centered_abs_mean": 0.11227346062660218,
"signal/brier_reward/group_std_mean": 0.1499951809644699,
"signal/brier_reward/group_zero_std_frac": 0.036111111380159856,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.05613673031330109,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.05613673031330109,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0005045572877861559,
"signal/confidence_one_or_zero/group_std_mean": 0.0014731390401721,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/format_reward/centered_abs_mean": 0.03428276889026165,
"signal/format_reward/group_std_mean": 0.06242974400520325,
"signal/format_reward/group_zero_std_frac": 0.75,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.017141384445130824,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.017141384445130824,
"signal/mean_confidence_reward/centered_abs_mean": 0.07776611000299453,
"signal/mean_confidence_reward/group_std_mean": 0.1076754778623581,
"signal/mean_confidence_reward/group_zero_std_frac": 0.04166666753590107,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.776610914334014e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.776610914334014e-07,
"step": 155
},
{
"calibration/aurc": 0.2439373889828011,
"calibration/batch_distribution_entropy": 0.5504421513946065,
"calibration/batch_entropy_100bins": 0.33446785407269986,
"calibration/batch_entropy_10bins": 0.5504421513946065,
"calibration/batch_entropy_50bins": 0.39373014616984314,
"calibration/batch_uniqueness": 0.4654293862688167,
"calibration/confidence_entropy": 0.5986497568225043,
"calibration/coverage@0%": 0.001610020467583384,
"calibration/coverage@1%": 0.001610020467583384,
"calibration/coverage@10%": 0.14383160308307236,
"calibration/coverage@15%": 0.19684997578648447,
"calibration/coverage@20%": 0.30920282176116654,
"calibration/coverage@25%": 0.531721487620661,
"calibration/coverage@30%": 0.7082227931036897,
"calibration/coverage@5%": 0.1370430913337251,
"calibration/distribution_entropy_10": 0.5504421513946065,
"calibration/distribution_entropy_100": 0.33446785407269986,
"calibration/ece": 0.1376500234882277,
"calibration/mean_confidence": 0.64873141349647,
"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": 0.02465277777777779,
"completions/max_length": 2934.0,
"completions/max_terminated_length": 2934.0,
"completions/mean_length": 711.8350708007813,
"completions/mean_terminated_length": 729.8376831054687,
"completions/min_length": 0.0,
"completions/min_terminated_length": 223.2,
"epoch": 0.38461538461538464,
"grad_norm": 0.0005352473235689104,
"learning_rate": 3.846153846153847e-06,
"loss": -0.0249,
"num_tokens": 325805742.0,
"reward": 1.2114521980285644,
"reward_std": 0.1843463361263275,
"rewards/accuracy_reward": 0.6668402671813964,
"rewards/brier_reward": 0.7807913780212402,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9752604126930237,
"rewards/mean_confidence_reward": 0.614529812335968,
"sampling/batch_mean_priority_error": 0.01641840277777778,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8166666666666667,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.0004144439706578851,
"sampling/priority_kl": 0.030000307038426398,
"sampling/priority_scale": 0.9023418128257618,
"sampling/prob_entropy": 10.278948974609374,
"sampling/prob_max": 3.8993890484562144e-05,
"sampling/prob_min": 1.43893730637501e-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": 0.16278212070465087,
"signal/accuracy_reward/group_std_mean": 0.21243431568145751,
"signal/accuracy_reward/group_zero_std_frac": 0.4055555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08139106035232543,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08139106035232543,
"signal/advantage_abs_mean": 0.13364365547895432,
"signal/advantage_pre_scale_abs_mean": 0.13364365547895432,
"signal/advantage_pre_scale_std": 0.23043625950813293,
"signal/advantage_std": 0.23043625950813293,
"signal/brier_reward/centered_abs_mean": 0.10205348432064057,
"signal/brier_reward/group_std_mean": 0.1377951443195343,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.051026742160320285,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.051026742160320285,
"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": 0.041357421875,
"signal/format_reward/group_std_mean": 0.07086744979023933,
"signal/format_reward/group_zero_std_frac": 0.7361111164093017,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0206787109375,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0206787109375,
"signal/mean_confidence_reward/centered_abs_mean": 0.09281512051820755,
"signal/mean_confidence_reward/group_std_mean": 0.12052757143974305,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 9.281511893277639e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 9.281511893277639e-07,
"step": 160
},
{
"calibration/aurc": 0.2216550039950434,
"calibration/batch_distribution_entropy": 0.5601830934410311,
"calibration/batch_entropy_100bins": 0.3656942587139888,
"calibration/batch_entropy_10bins": 0.5601830934410311,
"calibration/batch_entropy_50bins": 0.4304893644745743,
"calibration/batch_uniqueness": 0.5556512416662471,
"calibration/confidence_entropy": 0.6361458858573619,
"calibration/coverage@0%": 0.005390835579514825,
"calibration/coverage@1%": 0.005390835579514825,
"calibration/coverage@10%": 0.18005116166647134,
"calibration/coverage@15%": 0.2898822023363278,
"calibration/coverage@20%": 0.3762952458145887,
"calibration/coverage@25%": 0.482056475647318,
"calibration/coverage@30%": 0.7994447425005069,
"calibration/coverage@5%": 0.12570333557951482,
"calibration/distribution_entropy_10": 0.5601830934410311,
"calibration/distribution_entropy_100": 0.3656942587139888,
"calibration/ece": 0.13297860585793592,
"calibration/mean_confidence": 0.5568156688622672,
"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": 0.02031250000000002,
"completions/max_length": 3688.0,
"completions/max_terminated_length": 3688.0,
"completions/mean_length": 720.4741333007812,
"completions/mean_terminated_length": 735.3773803710938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 225.4,
"epoch": 0.39663461538461536,
"grad_norm": 0.0006447676569223404,
"learning_rate": 3.966346153846154e-06,
"loss": -0.0217,
"num_tokens": 337217700.0,
"reward": 1.2038058042526245,
"reward_std": 0.17525742650032045,
"rewards/accuracy_reward": 0.650868046283722,
"rewards/brier_reward": 0.7770451307296753,
"rewards/confidence_one_or_zero": 0.0003472222248092294,
"rewards/format_reward": 0.979687511920929,
"rewards/mean_confidence_reward": 0.5470833301544189,
"sampling/batch_mean_priority_error": 0.02227604166666667,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7833333333333333,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.0004395937954541296,
"sampling/priority_kl": 0.029999757558107375,
"sampling/priority_scale": 0.8911275923484936,
"sampling/prob_entropy": 10.278950691223145,
"sampling/prob_max": 3.912305910489522e-05,
"sampling/prob_min": 1.461659157939721e-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": 0.17914496660232543,
"signal/accuracy_reward/group_std_mean": 0.23302085399627687,
"signal/accuracy_reward/group_zero_std_frac": 0.3444444537162781,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08957248330116271,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08957248330116271,
"signal/advantage_abs_mean": 0.12544142454862595,
"signal/advantage_pre_scale_abs_mean": 0.12544142454862595,
"signal/advantage_pre_scale_std": 0.21346945464611053,
"signal/advantage_std": 0.21346945464611053,
"signal/brier_reward/centered_abs_mean": 0.09407158941030502,
"signal/brier_reward/group_std_mean": 0.12839447557926179,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04703579470515251,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04703579470515251,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0006618923507630825,
"signal/confidence_one_or_zero/group_std_mean": 0.0016652445774525404,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 6.618923009682476e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 6.618923009682476e-09,
"signal/format_reward/centered_abs_mean": 0.033984375,
"signal/format_reward/group_std_mean": 0.06068957597017288,
"signal/format_reward/group_zero_std_frac": 0.7611111164093017,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0169921875,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0169921875,
"signal/mean_confidence_reward/centered_abs_mean": 0.08409613519906997,
"signal/mean_confidence_reward/group_std_mean": 0.10996766984462739,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.409613201365573e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.409613201365573e-07,
"step": 165
},
{
"calibration/aurc": 0.20885690512845118,
"calibration/batch_distribution_entropy": 0.5220946730173782,
"calibration/batch_entropy_100bins": 0.31367912429843525,
"calibration/batch_entropy_10bins": 0.5220946730173782,
"calibration/batch_entropy_50bins": 0.3692579898384084,
"calibration/batch_uniqueness": 0.445031632020122,
"calibration/confidence_entropy": 0.6104902665786043,
"calibration/coverage@0%": 0.0010627337520287912,
"calibration/coverage@1%": 0.0010627337520287912,
"calibration/coverage@10%": 0.09353585203159868,
"calibration/coverage@15%": 0.29444203202348196,
"calibration/coverage@20%": 0.5902344275452739,
"calibration/coverage@25%": 0.6245352191020022,
"calibration/coverage@30%": 0.7654614101797976,
"calibration/coverage@5%": 0.0010627337520287912,
"calibration/distribution_entropy_10": 0.5220946730173782,
"calibration/distribution_entropy_100": 0.31367912429843525,
"calibration/ece": 0.100633464513873,
"calibration/mean_confidence": 0.6394755651544398,
"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": 0.02309027777777777,
"completions/max_length": 3179.2,
"completions/max_terminated_length": 3179.2,
"completions/mean_length": 700.7806640625,
"completions/mean_terminated_length": 717.4763305664062,
"completions/min_length": 0.0,
"completions/min_terminated_length": 207.2,
"epoch": 0.40865384615384615,
"grad_norm": 0.0006071875686757267,
"learning_rate": 4.086538461538462e-06,
"loss": -0.0234,
"num_tokens": 348414821.0,
"reward": 1.2222771883010863,
"reward_std": 0.18217334151268005,
"rewards/accuracy_reward": 0.6748263955116272,
"rewards/brier_reward": 0.7928924798965454,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9768229246139526,
"rewards/mean_confidence_reward": 0.6243949770927429,
"sampling/batch_mean_priority_error": 0.01674305555555556,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7277777777777777,
"sampling/error_ema_max": 0.05624999850988388,
"sampling/error_ema_mean": 0.0004621386586222798,
"sampling/priority_kl": 0.029999804869294166,
"sampling/priority_scale": 0.8803985654609278,
"sampling/prob_entropy": 10.278963851928712,
"sampling/prob_max": 3.924263146473095e-05,
"sampling/prob_min": 1.4835979709459934e-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": 0.1626953125,
"signal/accuracy_reward/group_std_mean": 0.21717799603939056,
"signal/accuracy_reward/group_zero_std_frac": 0.375,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08134765625,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08134765625,
"signal/advantage_abs_mean": 0.1283343553543091,
"signal/advantage_pre_scale_abs_mean": 0.1283343553543091,
"signal/advantage_pre_scale_std": 0.22712890207767486,
"signal/advantage_std": 0.22712890207767486,
"signal/brier_reward/centered_abs_mean": 0.09392926543951034,
"signal/brier_reward/group_std_mean": 0.13066150546073912,
"signal/brier_reward/group_zero_std_frac": 0.016666667349636555,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04696463271975517,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04696463271975517,
"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": 0.03892686627805233,
"signal/format_reward/group_std_mean": 0.07041922360658645,
"signal/format_reward/group_zero_std_frac": 0.7222222208976745,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.019463433139026164,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.019463433139026164,
"signal/mean_confidence_reward/centered_abs_mean": 0.08394259884953499,
"signal/mean_confidence_reward/group_std_mean": 0.11203955709934235,
"signal/mean_confidence_reward/group_zero_std_frac": 0.016666667349636555,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.394259793931269e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.394259793931269e-07,
"step": 170
},
{
"calibration/aurc": 0.2306798394707507,
"calibration/batch_distribution_entropy": 0.38562760462846135,
"calibration/batch_entropy_100bins": 0.20862008032824364,
"calibration/batch_entropy_10bins": 0.38562760462846135,
"calibration/batch_entropy_50bins": 0.2455841831177888,
"calibration/batch_uniqueness": -0.08259410038297318,
"calibration/confidence_entropy": 0.5396309137914621,
"calibration/coverage@0%": 0.004212333055377086,
"calibration/coverage@1%": 0.004212333055377086,
"calibration/coverage@10%": 0.20264575342091234,
"calibration/coverage@15%": 0.47179227748508346,
"calibration/coverage@20%": 0.5140382667898963,
"calibration/coverage@25%": 0.5668867516383812,
"calibration/coverage@30%": 0.6010457534209123,
"calibration/coverage@5%": 0.004212333055377086,
"calibration/distribution_entropy_10": 0.38562760462846135,
"calibration/distribution_entropy_100": 0.20862008032824364,
"calibration/ece": 0.15894475234322572,
"calibration/mean_confidence": 0.7424218898846368,
"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": 0.015625,
"completions/max_length": 3438.0,
"completions/max_terminated_length": 3438.0,
"completions/mean_length": 675.3561645507813,
"completions/mean_terminated_length": 686.0961181640625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 194.2,
"epoch": 0.4206730769230769,
"grad_norm": 0.0007575862691737711,
"learning_rate": 4.20673076923077e-06,
"loss": -0.0157,
"num_tokens": 359284844.0,
"reward": 1.2133283615112305,
"reward_std": 0.18129928410053253,
"rewards/accuracy_reward": 0.6555555582046508,
"rewards/brier_reward": 0.786798620223999,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9842882037162781,
"rewards/mean_confidence_reward": 0.716137146949768,
"sampling/batch_mean_priority_error": 0.016479166666666673,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7444444444444444,
"sampling/error_ema_max": 0.07174999564886093,
"sampling/error_ema_mean": 0.0004800788010470569,
"sampling/priority_kl": 0.03000015616416931,
"sampling/priority_scale": 0.8703835427528247,
"sampling/prob_entropy": 10.278958511352538,
"sampling/prob_max": 3.936297362088226e-05,
"sampling/prob_min": 1.5047048691485543e-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": 0.16691623330116273,
"signal/accuracy_reward/group_std_mean": 0.21467551290988923,
"signal/accuracy_reward/group_zero_std_frac": 0.40277778506278994,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08345811665058137,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08345811665058137,
"signal/advantage_abs_mean": 0.1352637842297554,
"signal/advantage_pre_scale_abs_mean": 0.1352637842297554,
"signal/advantage_pre_scale_std": 0.23468086123466492,
"signal/advantage_std": 0.23468086123466492,
"signal/brier_reward/centered_abs_mean": 0.10355425029993057,
"signal/brier_reward/group_std_mean": 0.13606458008289338,
"signal/brier_reward/group_zero_std_frac": 0.13611111268401146,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.051777125149965283,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.051777125149965283,
"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": 0.02566731721162796,
"signal/format_reward/group_std_mean": 0.04588991403579712,
"signal/format_reward/group_zero_std_frac": 0.819444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01283365860581398,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01283365860581398,
"signal/mean_confidence_reward/centered_abs_mean": 0.06004774793982506,
"signal/mean_confidence_reward/group_std_mean": 0.08248940110206604,
"signal/mean_confidence_reward/group_zero_std_frac": 0.20000000521540642,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.004774718348927e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.004774718348927e-07,
"step": 175
},
{
"calibration/aurc": 0.23788836281119957,
"calibration/batch_distribution_entropy": 0.3135210491423183,
"calibration/batch_entropy_100bins": 0.1708648779490961,
"calibration/batch_entropy_10bins": 0.3135210491423183,
"calibration/batch_entropy_50bins": 0.2011393697511128,
"calibration/batch_uniqueness": -0.24657734863529396,
"calibration/confidence_entropy": 0.5349542704968283,
"calibration/coverage@0%": 0.0005277044854881266,
"calibration/coverage@1%": 0.0005277044854881266,
"calibration/coverage@10%": 0.1397947201922944,
"calibration/coverage@15%": 0.1722554531765876,
"calibration/coverage@20%": 0.20052770448548812,
"calibration/coverage@25%": 0.6988382212767134,
"calibration/coverage@30%": 0.7187335092348285,
"calibration/coverage@5%": 0.0005277044854881266,
"calibration/distribution_entropy_10": 0.3135210491423183,
"calibration/distribution_entropy_100": 0.1708648779490961,
"calibration/ece": 0.09756549754912114,
"calibration/mean_confidence": 0.7477576020125064,
"calibration/unique_confidence_per_question": 0.01614583333333333,
"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": 0.008767361111111139,
"completions/max_length": 2763.6,
"completions/max_terminated_length": 2763.6,
"completions/mean_length": 635.5590209960938,
"completions/mean_terminated_length": 641.1690063476562,
"completions/min_length": 0.0,
"completions/min_terminated_length": 198.2,
"epoch": 0.4326923076923077,
"grad_norm": 0.0008080592378973961,
"learning_rate": 4.326923076923077e-06,
"loss": -0.0088,
"num_tokens": 369703348.0,
"reward": 1.2152099132537841,
"reward_std": 0.17329217493534088,
"rewards/accuracy_reward": 0.6556423544883728,
"rewards/brier_reward": 0.7836167335510253,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9911458253860473,
"rewards/mean_confidence_reward": 0.7386631965637207,
"sampling/batch_mean_priority_error": 0.01854861111111112,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7194444444444446,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0005015615141019225,
"sampling/priority_kl": 0.030001365393400193,
"sampling/priority_scale": 0.8609562695259229,
"sampling/prob_entropy": 10.278947448730468,
"sampling/prob_max": 3.948486191802658e-05,
"sampling/prob_min": 1.5251663899107371e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.4272000014781952,
"sampling/prompt_draws_total": 12816.0,
"sampling/seen_fraction": 0.37929333448410035,
"sampling/unseen_fraction": 0.6207066655158997,
"signal/accuracy_reward/centered_abs_mean": 0.16618380844593048,
"signal/accuracy_reward/group_std_mean": 0.21578561067581176,
"signal/accuracy_reward/group_zero_std_frac": 0.40277778506278994,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08309190422296524,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08309190422296524,
"signal/advantage_abs_mean": 0.1297558218240738,
"signal/advantage_pre_scale_abs_mean": 0.1297558218240738,
"signal/advantage_pre_scale_std": 0.22660800516605378,
"signal/advantage_std": 0.22660800516605378,
"signal/brier_reward/centered_abs_mean": 0.09994956403970719,
"signal/brier_reward/group_std_mean": 0.12978233247995377,
"signal/brier_reward/group_zero_std_frac": 0.25,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04997478201985359,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04997478201985359,
"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": 0.015700954757630826,
"signal/format_reward/group_std_mean": 0.0326525878161192,
"signal/format_reward/group_zero_std_frac": 0.8527777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007850477378815413,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007850477378815413,
"signal/mean_confidence_reward/centered_abs_mean": 0.0490852989256382,
"signal/mean_confidence_reward/group_std_mean": 0.06961074024438858,
"signal/mean_confidence_reward/group_zero_std_frac": 0.347222226858139,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.908529831482156e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.908529831482156e-07,
"step": 180
},
{
"calibration/aurc": 0.2195309944937937,
"calibration/batch_distribution_entropy": 0.46614281882144226,
"calibration/batch_entropy_100bins": 0.24709190776634332,
"calibration/batch_entropy_10bins": 0.46614281882144226,
"calibration/batch_entropy_50bins": 0.29087259590896697,
"calibration/batch_uniqueness": 0.11431428212023334,
"calibration/confidence_entropy": 0.5664291168624079,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.21468253968253967,
"calibration/coverage@15%": 0.2621847211834123,
"calibration/coverage@20%": 0.37131170531039637,
"calibration/coverage@25%": 0.6566805328150284,
"calibration/coverage@30%": 0.7982800231056182,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.46614281882144226,
"calibration/distribution_entropy_100": 0.24709190776634332,
"calibration/ece": 0.11462966079621192,
"calibration/mean_confidence": 0.7002868308990463,
"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": 0.010069444444444442,
"completions/max_length": 2952.8,
"completions/max_terminated_length": 2952.8,
"completions/mean_length": 627.3454956054687,
"completions/mean_terminated_length": 633.7,
"completions/min_length": 0.0,
"completions/min_terminated_length": 189.0,
"epoch": 0.44471153846153844,
"grad_norm": 0.0006965293432585895,
"learning_rate": 4.447115384615385e-06,
"loss": -0.0111,
"num_tokens": 380053184.0,
"reward": 1.223191213607788,
"reward_std": 0.16192339062690736,
"rewards/accuracy_reward": 0.65625,
"rewards/brier_reward": 0.8004485845565796,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9896701455116272,
"rewards/mean_confidence_reward": 0.6838238000869751,
"sampling/batch_mean_priority_error": 0.017381944444444446,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7138888888888889,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0005218912032432854,
"sampling/priority_kl": 0.029999763518571854,
"sampling/priority_scale": 0.852015310456045,
"sampling/prob_entropy": 10.278967094421386,
"sampling/prob_max": 3.960507456213236e-05,
"sampling/prob_min": 1.54499604832381e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.43920000195503234,
"sampling/prompt_draws_total": 13176.0,
"sampling/seen_fraction": 0.3879866719245911,
"sampling/unseen_fraction": 0.612013328075409,
"signal/accuracy_reward/centered_abs_mean": 0.15327690839767455,
"signal/accuracy_reward/group_std_mean": 0.2048683673143387,
"signal/accuracy_reward/group_zero_std_frac": 0.4166666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07663845419883727,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07663845419883727,
"signal/advantage_abs_mean": 0.11860896944999695,
"signal/advantage_pre_scale_abs_mean": 0.11860896944999695,
"signal/advantage_pre_scale_std": 0.2088211804628372,
"signal/advantage_std": 0.2088211804628372,
"signal/brier_reward/centered_abs_mean": 0.09137874394655228,
"signal/brier_reward/group_std_mean": 0.12084674090147018,
"signal/brier_reward/group_zero_std_frac": 0.13333333544433118,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04568937197327614,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04568937197327614,
"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": 0.017301432229578494,
"signal/format_reward/group_std_mean": 0.03283517360687256,
"signal/format_reward/group_zero_std_frac": 0.8638888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008650716114789247,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008650716114789247,
"signal/mean_confidence_reward/centered_abs_mean": 0.060839305818080905,
"signal/mean_confidence_reward/group_std_mean": 0.08073554188013077,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1666666679084301,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.083930202294141e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.083930202294141e-07,
"step": 185
},
{
"calibration/aurc": 0.21473587964092697,
"calibration/batch_distribution_entropy": 0.6310474175894865,
"calibration/batch_entropy_100bins": 0.35102006180287754,
"calibration/batch_entropy_10bins": 0.6310474175894865,
"calibration/batch_entropy_50bins": 0.41321513729733095,
"calibration/batch_uniqueness": 0.4533243875178625,
"calibration/confidence_entropy": 0.5994709288007358,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.07847561271199939,
"calibration/coverage@15%": 0.07847561271199939,
"calibration/coverage@20%": 0.45539064069003243,
"calibration/coverage@25%": 0.8117419117341532,
"calibration/coverage@30%": 0.9213213582677164,
"calibration/coverage@5%": 0.07847561271199939,
"calibration/distribution_entropy_10": 0.6310474175894865,
"calibration/distribution_entropy_100": 0.35102006180287754,
"calibration/ece": 0.139439512544235,
"calibration/mean_confidence": 0.595194758209025,
"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": 0.012152777777777768,
"completions/max_length": 3434.0,
"completions/max_terminated_length": 3434.0,
"completions/mean_length": 632.26484375,
"completions/mean_terminated_length": 640.1283569335938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 182.2,
"epoch": 0.4567307692307692,
"grad_norm": 0.0005936509114690125,
"learning_rate": 4.567307692307692e-06,
"loss": -0.0125,
"num_tokens": 390448267.0,
"reward": 1.2223145484924316,
"reward_std": 0.15787692070007325,
"rewards/accuracy_reward": 0.6589409828186035,
"rewards/brier_reward": 0.7980021953582763,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9876736164093017,
"rewards/mean_confidence_reward": 0.6120138883590698,
"sampling/batch_mean_priority_error": 0.021647569444444445,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6944444444444444,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0005465433234348894,
"sampling/priority_kl": 0.03000170923769474,
"sampling/priority_scale": 0.8435857356758788,
"sampling/prob_entropy": 10.278957557678222,
"sampling/prob_max": 3.9722044311929495e-05,
"sampling/prob_min": 1.5640574201825074e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.451199996471405,
"sampling/prompt_draws_total": 13536.0,
"sampling/seen_fraction": 0.3963599979877472,
"sampling/unseen_fraction": 0.6036400020122528,
"signal/accuracy_reward/centered_abs_mean": 0.1682779908180237,
"signal/accuracy_reward/group_std_mean": 0.22021067440509795,
"signal/accuracy_reward/group_zero_std_frac": 0.37777777910232546,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08413899540901185,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08413899540901185,
"signal/advantage_abs_mean": 0.11482741385698318,
"signal/advantage_pre_scale_abs_mean": 0.11482741385698318,
"signal/advantage_pre_scale_std": 0.19687965214252473,
"signal/advantage_std": 0.19687965214252473,
"signal/brier_reward/centered_abs_mean": 0.09163180440664291,
"signal/brier_reward/group_std_mean": 0.12160668224096298,
"signal/brier_reward/group_zero_std_frac": 0.03333333395421505,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.045815902203321456,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.045815902203321456,
"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": 0.020030382089316845,
"signal/format_reward/group_std_mean": 0.03791286274790764,
"signal/format_reward/group_zero_std_frac": 0.8416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010015191044658422,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010015191044658422,
"signal/mean_confidence_reward/centered_abs_mean": 0.07467366904020309,
"signal/mean_confidence_reward/group_std_mean": 0.09739873111248017,
"signal/mean_confidence_reward/group_zero_std_frac": 0.05000000111758709,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.4673666858871e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.4673666858871e-07,
"step": 190
},
{
"calibration/aurc": 0.20660881131868614,
"calibration/batch_distribution_entropy": 0.4472838898131556,
"calibration/batch_entropy_100bins": 0.24029866810382172,
"calibration/batch_entropy_10bins": 0.4472838898131556,
"calibration/batch_entropy_50bins": 0.28287570409194335,
"calibration/batch_uniqueness": -0.012685817494700501,
"calibration/confidence_entropy": 0.5425070079135119,
"calibration/coverage@0%": 0.00263863235518769,
"calibration/coverage@1%": 0.00263863235518769,
"calibration/coverage@10%": 0.2060695032310762,
"calibration/coverage@15%": 0.25063183744858286,
"calibration/coverage@20%": 0.5009042063459346,
"calibration/coverage@25%": 0.7282959128636407,
"calibration/coverage@30%": 0.7795261848813401,
"calibration/coverage@5%": 0.07927905230269425,
"calibration/distribution_entropy_10": 0.4472838898131556,
"calibration/distribution_entropy_100": 0.24029866810382172,
"calibration/ece": 0.10933349429072714,
"calibration/mean_confidence": 0.7094343831929872,
"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": 0.014236111111111116,
"completions/max_length": 3384.8,
"completions/max_terminated_length": 3384.8,
"completions/mean_length": 604.729345703125,
"completions/mean_terminated_length": 613.4677368164063,
"completions/min_length": 0.0,
"completions/min_terminated_length": 181.4,
"epoch": 0.46875,
"grad_norm": 0.0006664813263341784,
"learning_rate": 4.6875000000000004e-06,
"loss": -0.0163,
"num_tokens": 400524029.0,
"reward": 1.2562130212783813,
"reward_std": 0.16529837548732756,
"rewards/accuracy_reward": 0.7055555582046509,
"rewards/brier_reward": 0.8210926651954651,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9857638955116272,
"rewards/mean_confidence_reward": 0.6951171875,
"sampling/batch_mean_priority_error": 0.017201388888888898,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.711111111111111,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0005688484525308013,
"sampling/priority_kl": 0.0300009872764349,
"sampling/priority_scale": 0.8355700433952734,
"sampling/prob_entropy": 10.27896671295166,
"sampling/prob_max": 3.9841223770054055e-05,
"sampling/prob_min": 1.5827041352167726e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.46320000290870667,
"sampling/prompt_draws_total": 13896.0,
"sampling/seen_fraction": 0.4048066735267639,
"sampling/unseen_fraction": 0.5951933264732361,
"signal/accuracy_reward/centered_abs_mean": 0.15091145932674407,
"signal/accuracy_reward/group_std_mean": 0.20120739936828613,
"signal/accuracy_reward/group_zero_std_frac": 0.41666666865348817,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07545572966337204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07545572966337204,
"signal/advantage_abs_mean": 0.12027461379766465,
"signal/advantage_pre_scale_abs_mean": 0.12027461379766465,
"signal/advantage_pre_scale_std": 0.2162972718477249,
"signal/advantage_std": 0.2162972718477249,
"signal/brier_reward/centered_abs_mean": 0.09864553064107895,
"signal/brier_reward/group_std_mean": 0.13091023713350297,
"signal/brier_reward/group_zero_std_frac": 0.20277777761220933,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.049322765320539474,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.049322765320539474,
"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": 0.023763021267950536,
"signal/format_reward/group_std_mean": 0.04480189755558968,
"signal/format_reward/group_zero_std_frac": 0.8138889074325562,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.011881510633975268,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.011881510633975268,
"signal/mean_confidence_reward/centered_abs_mean": 0.06588922217488288,
"signal/mean_confidence_reward/group_std_mean": 0.08890577405691147,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2833333373069763,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.588921792172187e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.588921792172187e-07,
"step": 195
},
{
"calibration/aurc": 0.23729910544861918,
"calibration/batch_distribution_entropy": 0.4545196357256331,
"calibration/batch_entropy_100bins": 0.24545220588673908,
"calibration/batch_entropy_10bins": 0.4545196357256331,
"calibration/batch_entropy_50bins": 0.288942365386451,
"calibration/batch_uniqueness": -0.03244423971310495,
"calibration/confidence_entropy": 0.5201880994887154,
"calibration/coverage@0%": 0.03456750156587587,
"calibration/coverage@1%": 0.03456750156587587,
"calibration/coverage@10%": 0.24514294203790565,
"calibration/coverage@15%": 0.41627155096179014,
"calibration/coverage@20%": 0.431494648074651,
"calibration/coverage@25%": 0.5556250906734358,
"calibration/coverage@30%": 0.5726113920432987,
"calibration/coverage@5%": 0.03456750156587587,
"calibration/distribution_entropy_10": 0.4545196357256331,
"calibration/distribution_entropy_100": 0.24545220588673908,
"calibration/ece": 0.15331469183532745,
"calibration/mean_confidence": 0.7156514431371688,
"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": 0.013368055555555581,
"completions/max_length": 3403.8,
"completions/max_terminated_length": 3403.8,
"completions/mean_length": 616.7714599609375,
"completions/mean_terminated_length": 625.16923828125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 144.0,
"epoch": 0.4807692307692308,
"grad_norm": 0.0007234844961203635,
"learning_rate": 4.807692307692308e-06,
"loss": -0.0135,
"num_tokens": 410753716.0,
"reward": 1.2360211133956909,
"reward_std": 0.16374669075012208,
"rewards/accuracy_reward": 0.6888020753860473,
"rewards/brier_reward": 0.7968539476394654,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9863715291023254,
"rewards/mean_confidence_reward": 0.7336067795753479,
"sampling/batch_mean_priority_error": 0.01865972222222222,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.675,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0005898760398849844,
"sampling/priority_kl": 0.029998957365751266,
"sampling/priority_scale": 0.8279964744346217,
"sampling/prob_entropy": 10.27893295288086,
"sampling/prob_max": 3.9958462730282916e-05,
"sampling/prob_min": 1.6006992518668993e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.47519999742507935,
"sampling/prompt_draws_total": 14256.0,
"sampling/seen_fraction": 0.4130400002002716,
"sampling/unseen_fraction": 0.5869599997997283,
"signal/accuracy_reward/centered_abs_mean": 0.14013129323720933,
"signal/accuracy_reward/group_std_mean": 0.1888304203748703,
"signal/accuracy_reward/group_zero_std_frac": 0.4527777791023254,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07006564661860466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07006564661860466,
"signal/advantage_abs_mean": 0.11893169432878495,
"signal/advantage_pre_scale_abs_mean": 0.11893169432878495,
"signal/advantage_pre_scale_std": 0.21923663318157197,
"signal/advantage_std": 0.21923663318157197,
"signal/brier_reward/centered_abs_mean": 0.10698548257350922,
"signal/brier_reward/group_std_mean": 0.13870647549629211,
"signal/brier_reward/group_zero_std_frac": 0.13055555894970894,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.05349274128675461,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.05349274128675461,
"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": 0.02221679650247097,
"signal/format_reward/group_std_mean": 0.04155225083231926,
"signal/format_reward/group_zero_std_frac": 0.8250000238418579,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.011108398251235486,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.011108398251235486,
"signal/mean_confidence_reward/centered_abs_mean": 0.06812093928456306,
"signal/mean_confidence_reward/group_std_mean": 0.09376206994056702,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1750000037252903,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.812093602093228e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.812093602093228e-07,
"step": 200
},
{
"epoch": 0.4807692307692308,
"eval_calibration/aurc": 0.24002113019857146,
"eval_calibration/batch_distribution_entropy": 0.48160177072803806,
"eval_calibration/batch_entropy_100bins": 0.25873340992629923,
"eval_calibration/batch_entropy_10bins": 0.48160177072803806,
"eval_calibration/batch_entropy_50bins": 0.30457678389374016,
"eval_calibration/batch_uniqueness": 0.01719030076599557,
"eval_calibration/confidence_entropy": 0.49903381393986235,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.0,
"eval_calibration/coverage@15%": 0.1235657546337158,
"eval_calibration/coverage@20%": 0.1235657546337158,
"eval_calibration/coverage@25%": 0.8093556928508385,
"eval_calibration/coverage@30%": 0.9196822594880847,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.48160177072803806,
"eval_calibration/distribution_entropy_100": 0.25873340992629923,
"eval_calibration/ece": 0.06690203000882607,
"eval_calibration/mean_confidence": 0.738217122683142,
"eval_calibration/unique_confidence_per_question": 0.008680555555555556,
"eval_calibration/unique_confidences": 10,
"eval_completions/clipped_ratio": 0.012152777777777771,
"eval_completions/max_length": 2276.1666666666665,
"eval_completions/max_terminated_length": 2276.1666666666665,
"eval_completions/mean_length": 586.3701680501302,
"eval_completions/mean_terminated_length": 593.6424357096354,
"eval_completions/min_length": 44.5,
"eval_completions/min_terminated_length": 192.33333333333334,
"eval_loss": 0.0,
"eval_num_tokens": 410753716.0,
"eval_reward": 1.2204153537750244,
"eval_reward_std": 0.37025384108225506,
"eval_rewards/accuracy_reward": 0.6675347288449606,
"eval_rewards/brier_reward": 0.7897742787996928,
"eval_rewards/confidence_one_or_zero": 0.0008680555814256271,
"eval_rewards/format_reward": 0.9835069378217062,
"eval_rewards/mean_confidence_reward": 0.7268229126930237,
"eval_runtime": 207.2024,
"eval_samples_per_second": 4.826,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4338650157054265,
"eval_signal/accuracy_reward/group_std_mean": 0.47237643599510193,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.21693250785271326,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.21693250785271326,
"eval_signal/advantage_abs_mean": 0.3235609481732051,
"eval_signal/advantage_pre_scale_abs_mean": 0.3235609481732051,
"eval_signal/advantage_pre_scale_std": 0.36790668964385986,
"eval_signal/advantage_std": 0.36790668964385986,
"eval_signal/brier_reward/centered_abs_mean": 0.22734156996011734,
"eval_signal/brier_reward/group_std_mean": 0.27066074560085934,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.11367078498005867,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.11367078498005867,
"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": 0.03152126741285125,
"eval_signal/format_reward/group_std_mean": 0.08380424821128447,
"eval_signal/format_reward/group_zero_std_frac": 0.5555555745959282,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.015760633706425626,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.015760633706425626,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.13929034397006035,
"eval_signal/mean_confidence_reward/group_std_mean": 0.2003222107887268,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.3929034328915197e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.3929034328915197e-06,
"eval_steps_per_second": 0.029,
"step": 200
},
{
"epoch": 0.4807692307692308,
"step": 200,
"train_probe_calibration/aurc": 0.2113365019677367,
"train_probe_calibration/batch_distribution_entropy": 0.4955069873679676,
"train_probe_calibration/batch_entropy_100bins": 0.26813957141588457,
"train_probe_calibration/batch_entropy_10bins": 0.4955069873679676,
"train_probe_calibration/batch_entropy_50bins": 0.3156495650088622,
"train_probe_calibration/batch_uniqueness": 0.0653170297438859,
"train_probe_calibration/confidence_entropy": 0.49672003313055696,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.1411042944785276,
"train_probe_calibration/coverage@15%": 0.1411042944785276,
"train_probe_calibration/coverage@20%": 0.1411042944785276,
"train_probe_calibration/coverage@25%": 0.8773006134969326,
"train_probe_calibration/coverage@30%": 0.9570552147239264,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.4955069873679676,
"train_probe_calibration/distribution_entropy_100": 0.26813957141588457,
"train_probe_calibration/ece": 0.046099912357580895,
"train_probe_calibration/mean_confidence": 0.738737949167397,
"train_probe_calibration/unique_confidence_per_question": 0.010416666666666666,
"train_probe_calibration/unique_confidences": 12,
"train_probe_completions/clipped_ratio": 0.010243055555555566,
"train_probe_completions/max_length": 2403.3333333333335,
"train_probe_completions/max_terminated_length": 2403.3333333333335,
"train_probe_completions/mean_length": 601.1067911783854,
"train_probe_completions/mean_terminated_length": 607.4044189453125,
"train_probe_completions/min_length": 32.833333333333336,
"train_probe_completions/min_terminated_length": 177.83333333333334,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 410753716.0,
"train_probe_reward": 1.2459991375605266,
"train_probe_reward_std": 0.35098352034886676,
"train_probe_rewards/accuracy_reward": 0.6901041666666666,
"train_probe_rewards/brier_reward": 0.8114279508590698,
"train_probe_rewards/confidence_one_or_zero": 0.0008680555814256271,
"train_probe_rewards/format_reward": 0.9904513955116272,
"train_probe_rewards/mean_confidence_reward": 0.7316840390364329,
"train_probe_runtime": 182.4753,
"train_probe_samples_per_second": 5.48,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.4162868907054265,
"train_probe_signal/accuracy_reward/group_std_mean": 0.46246403952439624,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20814344535271326,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.20814344535271326,
"train_probe_signal/advantage_abs_mean": 0.30224540332953137,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.30224540332953137,
"train_probe_signal/advantage_pre_scale_std": 0.3477505644162496,
"train_probe_signal/advantage_std": 0.3477505644162496,
"train_probe_signal/brier_reward/centered_abs_mean": 0.2085253670811653,
"train_probe_signal/brier_reward/group_std_mean": 0.2562006364266078,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.10426268354058266,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.10426268354058266,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0016818575871487458,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0049104637776811915,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222288449606,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151002e-08,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151002e-08,
"train_probe_signal/format_reward/centered_abs_mean": 0.018391926772892475,
"train_probe_signal/format_reward/group_std_mean": 0.051025692373514175,
"train_probe_signal/format_reward/group_zero_std_frac": 0.7222222437461218,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.009195963386446238,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.009195963386446238,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.13380803540349007,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.1927381455898285,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.3380803428238626e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.3380803428238626e-06,
"train_probe_steps_per_second": 0.033
},
{
"calibration/aurc": 0.22641020059736908,
"calibration/batch_distribution_entropy": 0.586543433079215,
"calibration/batch_entropy_100bins": 0.31469597338663524,
"calibration/batch_entropy_10bins": 0.586543433079215,
"calibration/batch_entropy_50bins": 0.3704550081325571,
"calibration/batch_uniqueness": 0.3057555199387029,
"calibration/confidence_entropy": 0.48789312654324435,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.11884628551865799,
"calibration/coverage@15%": 0.11884628551865799,
"calibration/coverage@20%": 0.31884628551865796,
"calibration/coverage@25%": 0.728710047005235,
"calibration/coverage@30%": 0.9123436701244794,
"calibration/coverage@5%": 0.03695652173913043,
"calibration/distribution_entropy_10": 0.586543433079215,
"calibration/distribution_entropy_100": 0.31469597338663524,
"calibration/ece": 0.12346906503340056,
"calibration/mean_confidence": 0.720077070835353,
"calibration/unique_confidence_per_question": 0.023958333333333335,
"calibration/unique_confidences": 9.2,
"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": 0.01675347222222221,
"completions/max_length": 3890.2,
"completions/max_terminated_length": 3890.2,
"completions/mean_length": 571.915625,
"completions/mean_terminated_length": 581.7823852539062,
"completions/min_length": 0.0,
"completions/min_terminated_length": 149.0,
"epoch": 0.49278846153846156,
"grad_norm": 0.0007555597112514079,
"learning_rate": 4.927884615384616e-06,
"loss": -0.0201,
"num_tokens": 420431336.0,
"reward": 1.2604899883270264,
"reward_std": 0.1605339229106903,
"rewards/accuracy_reward": 0.7164062619209289,
"rewards/brier_reward": 0.8213125944137574,
"rewards/confidence_one_or_zero": 0.0008680555736646056,
"rewards/format_reward": 0.9832465171813964,
"rewards/mean_confidence_reward": 0.7297421932220459,
"sampling/batch_mean_priority_error": 0.013986111111111105,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6888888888888889,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0006075228797271848,
"sampling/priority_kl": 0.029999784752726556,
"sampling/priority_scale": 0.8207675993675366,
"sampling/prob_entropy": 10.278958702087403,
"sampling/prob_max": 4.0076706500258294e-05,
"sampling/prob_min": 1.61830939759966e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.487200003862381,
"sampling/prompt_draws_total": 14616.0,
"sampling/seen_fraction": 0.42125333547592164,
"sampling/unseen_fraction": 0.5787466645240784,
"signal/accuracy_reward/centered_abs_mean": 0.1313639312982559,
"signal/accuracy_reward/group_std_mean": 0.1768402189016342,
"signal/accuracy_reward/group_zero_std_frac": 0.4833333432674408,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06568196564912795,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06568196564912795,
"signal/advantage_abs_mean": 0.11555989980697631,
"signal/advantage_pre_scale_abs_mean": 0.11555989980697631,
"signal/advantage_pre_scale_std": 0.21790286600589753,
"signal/advantage_std": 0.21790286600589753,
"signal/brier_reward/centered_abs_mean": 0.1100591003894806,
"signal/brier_reward/group_std_mean": 0.14190090894699098,
"signal/brier_reward/group_zero_std_frac": 0.033333333767950536,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0550295501947403,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0550295501947403,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0016276041511446237,
"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": 1.6276040781804114e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6276040781804114e-08,
"signal/format_reward/centered_abs_mean": 0.02888997383415699,
"signal/format_reward/group_std_mean": 0.052112925052642825,
"signal/format_reward/group_zero_std_frac": 0.7972222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.014444986917078495,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.014444986917078495,
"signal/mean_confidence_reward/centered_abs_mean": 0.08421467393636703,
"signal/mean_confidence_reward/group_std_mean": 0.10899612158536912,
"signal/mean_confidence_reward/group_zero_std_frac": 0.05277777872979641,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.421466986874293e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.421466986874293e-07,
"step": 205
},
{
"calibration/aurc": 0.1461755373917801,
"calibration/batch_distribution_entropy": 0.6707544144938835,
"calibration/batch_entropy_100bins": 0.37178587247237216,
"calibration/batch_entropy_10bins": 0.6707544144938835,
"calibration/batch_entropy_50bins": 0.4376603136294584,
"calibration/batch_uniqueness": 0.5310754998290197,
"calibration/confidence_entropy": 0.4849550254131734,
"calibration/coverage@0%": 0.002102507635094426,
"calibration/coverage@1%": 0.14442151395239228,
"calibration/coverage@10%": 0.4801659934974561,
"calibration/coverage@15%": 0.4953812431926138,
"calibration/coverage@20%": 0.6357792125051241,
"calibration/coverage@25%": 0.8384257306194505,
"calibration/coverage@30%": 0.9014771570440239,
"calibration/coverage@5%": 0.2594849378129214,
"calibration/distribution_entropy_10": 0.6707544144938835,
"calibration/distribution_entropy_100": 0.37178587247237216,
"calibration/ece": 0.09125336785763163,
"calibration/mean_confidence": 0.6740264282946639,
"calibration/unique_confidence_per_question": 0.023958333333333335,
"calibration/unique_confidences": 9.2,
"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": 3380.0,
"completions/max_terminated_length": 3380.0,
"completions/mean_length": 601.1892578125,
"completions/mean_terminated_length": 607.6563232421875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 153.6,
"epoch": 0.5048076923076923,
"grad_norm": 0.0007844443898648024,
"learning_rate": 4.987980769230769e-06,
"loss": -0.0119,
"num_tokens": 430468588.0,
"reward": 1.2284770011901855,
"reward_std": 0.1580641061067581,
"rewards/accuracy_reward": 0.6583333373069763,
"rewards/brier_reward": 0.8092842698097229,
"rewards/confidence_one_or_zero": 0.0006944444496184588,
"rewards/format_reward": 0.9893229246139527,
"rewards/mean_confidence_reward": 0.6731336712837219,
"sampling/batch_mean_priority_error": 0.019003472222222213,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6805555555555556,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0006265622796490788,
"sampling/priority_kl": 0.0299996230751276,
"sampling/priority_scale": 0.8139940678840503,
"sampling/prob_entropy": 10.278954696655273,
"sampling/prob_max": 4.019671468995512e-05,
"sampling/prob_min": 1.6353473620256408e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.49920000433921813,
"sampling/prompt_draws_total": 14976.0,
"sampling/seen_fraction": 0.4294800043106079,
"sampling/unseen_fraction": 0.570519995689392,
"signal/accuracy_reward/centered_abs_mean": 0.14608289897441865,
"signal/accuracy_reward/group_std_mean": 0.1931061327457428,
"signal/accuracy_reward/group_zero_std_frac": 0.45,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07304144948720932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07304144948720932,
"signal/advantage_abs_mean": 0.11451363265514374,
"signal/advantage_pre_scale_abs_mean": 0.11451363265514374,
"signal/advantage_pre_scale_std": 0.2036235064268112,
"signal/advantage_std": 0.2036235064268112,
"signal/brier_reward/centered_abs_mean": 0.11961710602045059,
"signal/brier_reward/group_std_mean": 0.15400556921958924,
"signal/brier_reward/group_zero_std_frac": 0.013888888992369175,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.05980855301022529,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.05980855301022529,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0013129340019077062,
"signal/confidence_one_or_zero/group_std_mean": 0.0032778555527329446,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9833333253860473,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.3129339038187026e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.3129339038187026e-08,
"signal/format_reward/centered_abs_mean": 0.018994140811264514,
"signal/format_reward/group_std_mean": 0.036902287229895595,
"signal/format_reward/group_zero_std_frac": 0.8472222208976745,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009497070405632257,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009497070405632257,
"signal/mean_confidence_reward/centered_abs_mean": 0.09470810443162918,
"signal/mean_confidence_reward/group_std_mean": 0.1206357717514038,
"signal/mean_confidence_reward/group_zero_std_frac": 0.013888888992369175,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 9.470810482525849e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 9.470810482525849e-07,
"step": 210
},
{
"calibration/aurc": 0.14871075118082513,
"calibration/batch_distribution_entropy": 0.6749137696780482,
"calibration/batch_entropy_100bins": 0.3727244117417784,
"calibration/batch_entropy_10bins": 0.6749137696780482,
"calibration/batch_entropy_50bins": 0.43876514687196533,
"calibration/batch_uniqueness": 0.5322232887996369,
"calibration/confidence_entropy": 0.48361457584113304,
"calibration/coverage@0%": 0.09635040020423015,
"calibration/coverage@1%": 0.09635040020423015,
"calibration/coverage@10%": 0.39986044205220733,
"calibration/coverage@15%": 0.6437249317552453,
"calibration/coverage@20%": 0.7111178626707388,
"calibration/coverage@25%": 0.7794437929301391,
"calibration/coverage@30%": 0.926595744680851,
"calibration/coverage@5%": 0.2446842144983858,
"calibration/distribution_entropy_10": 0.6749137696780482,
"calibration/distribution_entropy_100": 0.3727244117417784,
"calibration/ece": 0.11950171895871634,
"calibration/mean_confidence": 0.6841320759702054,
"calibration/unique_confidence_per_question": 0.024479166666666666,
"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": 0.01059027777777779,
"completions/max_length": 2960.2,
"completions/max_terminated_length": 2960.2,
"completions/mean_length": 544.8087768554688,
"completions/mean_terminated_length": 550.6374694824219,
"completions/min_length": 0.0,
"completions/min_terminated_length": 144.8,
"epoch": 0.5168269230769231,
"grad_norm": 0.0010941576911136508,
"learning_rate": 4.957932692307692e-06,
"loss": -0.0102,
"num_tokens": 439827985.0,
"reward": 1.2625808477401734,
"reward_std": 0.15077738016843795,
"rewards/accuracy_reward": 0.718836784362793,
"rewards/brier_reward": 0.8174222230911254,
"rewards/confidence_one_or_zero": 0.0047743055620230734,
"rewards/format_reward": 0.9888888955116272,
"rewards/mean_confidence_reward": 0.6864542722702026,
"sampling/batch_mean_priority_error": 0.014708333333333315,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7027777777777777,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0006463405909016729,
"sampling/priority_kl": 0.030000553280115128,
"sampling/priority_scale": 0.8077474296791479,
"sampling/prob_entropy": 10.278956031799316,
"sampling/prob_max": 4.031913995277137e-05,
"sampling/prob_min": 1.6515195602551104e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5111999988555909,
"sampling/prompt_draws_total": 15336.0,
"sampling/seen_fraction": 0.437773334980011,
"sampling/unseen_fraction": 0.562226665019989,
"signal/accuracy_reward/centered_abs_mean": 0.1398925766348839,
"signal/accuracy_reward/group_std_mean": 0.18380642533302308,
"signal/accuracy_reward/group_zero_std_frac": 0.4777777850627899,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06994628831744194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06994628831744194,
"signal/advantage_abs_mean": 0.10771286487579346,
"signal/advantage_pre_scale_abs_mean": 0.10771286487579346,
"signal/advantage_pre_scale_std": 0.1967930257320404,
"signal/advantage_std": 0.1967930257320404,
"signal/brier_reward/centered_abs_mean": 0.10499090403318405,
"signal/brier_reward/group_std_mean": 0.14038468152284622,
"signal/brier_reward/group_zero_std_frac": 0.033333334140479565,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.052495452016592024,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.052495452016592024,
"signal/confidence_one_or_zero/centered_abs_mean": 0.008479817654006183,
"signal/confidence_one_or_zero/group_std_mean": 0.016532309353351593,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9305555701255799,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 8.479817523721067e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.479817523721067e-08,
"signal/format_reward/centered_abs_mean": 0.01947699673473835,
"signal/format_reward/group_std_mean": 0.037950745224952696,
"signal/format_reward/group_zero_std_frac": 0.8416666626930237,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009738498367369175,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009738498367369175,
"signal/mean_confidence_reward/centered_abs_mean": 0.09180866181850433,
"signal/mean_confidence_reward/group_std_mean": 0.1205675944685936,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0361111119389534,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 9.180865731650556e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 9.180865731650556e-07,
"step": 215
},
{
"calibration/aurc": 0.23897279996570475,
"calibration/batch_distribution_entropy": 0.6549670881134286,
"calibration/batch_entropy_100bins": 0.3524090104914431,
"calibration/batch_entropy_10bins": 0.6549670881134286,
"calibration/batch_entropy_50bins": 0.4148501852205088,
"calibration/batch_uniqueness": 0.48453062977196665,
"calibration/confidence_entropy": 0.5208603656696392,
"calibration/coverage@0%": 0.032112316565714005,
"calibration/coverage@1%": 0.08569321842247793,
"calibration/coverage@10%": 0.08569321842247793,
"calibration/coverage@15%": 0.4360667565516437,
"calibration/coverage@20%": 0.4980136219467849,
"calibration/coverage@25%": 0.5319574558906188,
"calibration/coverage@30%": 0.6400633783411409,
"calibration/coverage@5%": 0.08569321842247793,
"calibration/distribution_entropy_10": 0.6549670881134286,
"calibration/distribution_entropy_100": 0.3524090104914431,
"calibration/ece": 0.12067979587505484,
"calibration/mean_confidence": 0.7013421687699827,
"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": 0.008854166666666673,
"completions/max_length": 2822.8,
"completions/max_terminated_length": 2822.8,
"completions/mean_length": 574.1637939453125,
"completions/mean_terminated_length": 579.2975830078125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 132.4,
"epoch": 0.5288461538461539,
"grad_norm": 0.0008874834165908396,
"learning_rate": 4.927884615384616e-06,
"loss": -0.0088,
"num_tokens": 449520944.0,
"reward": 1.2366694450378417,
"reward_std": 0.14118025302886963,
"rewards/accuracy_reward": 0.6623263835906983,
"rewards/brier_reward": 0.8200264930725097,
"rewards/confidence_one_or_zero": 0.0019965277635492383,
"rewards/format_reward": 0.9909722208976746,
"rewards/mean_confidence_reward": 0.6875000238418579,
"sampling/batch_mean_priority_error": 0.016652777777777773,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6777777777777778,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0006648902199231088,
"sampling/priority_kl": 0.02999868765473366,
"sampling/priority_scale": 0.8017917334800586,
"sampling/prob_entropy": 10.278938484191894,
"sampling/prob_max": 4.0442708268528804e-05,
"sampling/prob_min": 1.667476535658352e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.523199999332428,
"sampling/prompt_draws_total": 15696.0,
"sampling/seen_fraction": 0.4460400104522705,
"sampling/unseen_fraction": 0.5539599895477295,
"signal/accuracy_reward/centered_abs_mean": 0.13765191435813903,
"signal/accuracy_reward/group_std_mean": 0.179826021194458,
"signal/accuracy_reward/group_zero_std_frac": 0.49444445967674255,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06882595717906952,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06882595717906952,
"signal/advantage_abs_mean": 0.10393836796283722,
"signal/advantage_pre_scale_abs_mean": 0.10393836796283722,
"signal/advantage_pre_scale_std": 0.18965348303318025,
"signal/advantage_std": 0.18965348303318025,
"signal/brier_reward/centered_abs_mean": 0.08609333783388137,
"signal/brier_reward/group_std_mean": 0.11452355682849884,
"signal/brier_reward/group_zero_std_frac": 0.06388889122754335,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.043046668916940686,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.043046668916940686,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0036295572528615592,
"signal/confidence_one_or_zero/group_std_mean": 0.00784936579875648,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9638888835906982,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.629557205897527e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.629557205897527e-08,
"signal/format_reward/centered_abs_mean": 0.01592881940305233,
"signal/format_reward/group_std_mean": 0.030093761160969734,
"signal/format_reward/group_zero_std_frac": 0.8777777791023255,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007964409701526166,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007964409701526166,
"signal/mean_confidence_reward/centered_abs_mean": 0.0710023283958435,
"signal/mean_confidence_reward/group_std_mean": 0.09437335133552552,
"signal/mean_confidence_reward/group_zero_std_frac": 0.07222222406417131,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.100233005985501e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.100233005985501e-07,
"step": 220
},
{
"calibration/aurc": 0.14094224662459304,
"calibration/batch_distribution_entropy": 0.566623231366884,
"calibration/batch_entropy_100bins": 0.30633370359683026,
"calibration/batch_entropy_10bins": 0.566623231366884,
"calibration/batch_entropy_50bins": 0.36061107943639037,
"calibration/batch_uniqueness": 0.3183684648717434,
"calibration/confidence_entropy": 0.5458095920829429,
"calibration/coverage@0%": 0.08241021163558863,
"calibration/coverage@1%": 0.08241021163558863,
"calibration/coverage@10%": 0.3662256485440952,
"calibration/coverage@15%": 0.6514003532685047,
"calibration/coverage@20%": 0.7386676534884836,
"calibration/coverage@25%": 0.85249343832021,
"calibration/coverage@30%": 0.8771653543307087,
"calibration/coverage@5%": 0.20669219596979227,
"calibration/distribution_entropy_10": 0.566623231366884,
"calibration/distribution_entropy_100": 0.30633370359683026,
"calibration/ece": 0.12998478649698814,
"calibration/mean_confidence": 0.7048597419529478,
"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": 0.007031250000000022,
"completions/max_length": 3238.0,
"completions/max_terminated_length": 3238.0,
"completions/mean_length": 582.3898681640625,
"completions/mean_terminated_length": 586.5300170898438,
"completions/min_length": 0.0,
"completions/min_terminated_length": 140.4,
"epoch": 0.5408653846153846,
"grad_norm": 0.000878737133461982,
"learning_rate": 4.897836538461539e-06,
"loss": -0.0071,
"num_tokens": 459297339.0,
"reward": 1.268056082725525,
"reward_std": 0.14799101948738097,
"rewards/accuracy_reward": 0.7123263955116272,
"rewards/brier_reward": 0.8308030605316162,
"rewards/confidence_one_or_zero": 0.0003472222248092294,
"rewards/format_reward": 0.992968761920929,
"rewards/mean_confidence_reward": 0.6942693829536438,
"sampling/batch_mean_priority_error": 0.015236111111111112,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6083333333333334,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0006837929598987102,
"sampling/priority_kl": 0.029999540001153947,
"sampling/priority_scale": 0.7958873092429712,
"sampling/prob_entropy": 10.278943252563476,
"sampling/prob_max": 4.055707686347887e-05,
"sampling/prob_min": 1.6831377070047893e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5351999998092651,
"sampling/prompt_draws_total": 16056.0,
"sampling/seen_fraction": 0.4536533296108246,
"sampling/unseen_fraction": 0.5463466703891754,
"signal/accuracy_reward/centered_abs_mean": 0.14823133498430252,
"signal/accuracy_reward/group_std_mean": 0.19535186886787415,
"signal/accuracy_reward/group_zero_std_frac": 0.4444444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07411566749215126,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07411566749215126,
"signal/advantage_abs_mean": 0.1074238434433937,
"signal/advantage_pre_scale_abs_mean": 0.1074238434433937,
"signal/advantage_pre_scale_std": 0.19590988755226135,
"signal/advantage_std": 0.19590988755226135,
"signal/brier_reward/centered_abs_mean": 0.08263461142778397,
"signal/brier_reward/group_std_mean": 0.11035842001438141,
"signal/brier_reward/group_zero_std_frac": 0.10000000223517418,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.041317305713891986,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.041317305713891986,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0006510416511446238,
"signal/confidence_one_or_zero/group_std_mean": 0.0013663037680089474,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 6.510416739047286e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 6.510416739047286e-09,
"signal/format_reward/centered_abs_mean": 0.012972005270421505,
"signal/format_reward/group_std_mean": 0.029025893285870553,
"signal/format_reward/group_zero_std_frac": 0.8638888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.006486002635210752,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.006486002635210752,
"signal/mean_confidence_reward/centered_abs_mean": 0.06351888552308083,
"signal/mean_confidence_reward/group_std_mean": 0.08457329869270325,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1361111119389534,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.351887918754074e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.351887918754074e-07,
"step": 225
},
{
"calibration/aurc": 0.27389093170088574,
"calibration/batch_distribution_entropy": 0.6147581307636807,
"calibration/batch_entropy_100bins": 0.35248461902091915,
"calibration/batch_entropy_10bins": 0.6147581307636807,
"calibration/batch_entropy_50bins": 0.4149391903580725,
"calibration/batch_uniqueness": 0.53075158986908,
"calibration/confidence_entropy": 0.6116111945484981,
"calibration/coverage@0%": 0.16600427324666853,
"calibration/coverage@1%": 0.16600427324666853,
"calibration/coverage@10%": 0.27494610286396765,
"calibration/coverage@15%": 0.3732418053072486,
"calibration/coverage@20%": 0.525133256149047,
"calibration/coverage@25%": 0.5403165022223455,
"calibration/coverage@30%": 0.6074272857188354,
"calibration/coverage@5%": 0.2184861334049799,
"calibration/distribution_entropy_10": 0.6147581307636807,
"calibration/distribution_entropy_100": 0.35248461902091915,
"calibration/ece": 0.1604331230059492,
"calibration/mean_confidence": 0.6082048790180217,
"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": 0.009114583333333325,
"completions/max_length": 3444.4,
"completions/max_terminated_length": 3444.4,
"completions/mean_length": 609.4268188476562,
"completions/mean_terminated_length": 615.0386962890625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 168.0,
"epoch": 0.5528846153846154,
"grad_norm": 0.0006518294103443623,
"learning_rate": 4.867788461538462e-06,
"loss": -0.009,
"num_tokens": 469448784.0,
"reward": 1.2289212703704835,
"reward_std": 0.1348082423210144,
"rewards/accuracy_reward": 0.6572916746139527,
"rewards/brier_reward": 0.8097397804260253,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9907986044883728,
"rewards/mean_confidence_reward": 0.6189192771911621,
"sampling/batch_mean_priority_error": 0.018409722222222223,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6194444444444444,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0007028119522146881,
"sampling/priority_kl": 0.029999968782067298,
"sampling/priority_scale": 0.7902504027122632,
"sampling/prob_entropy": 10.27895679473877,
"sampling/prob_max": 4.067026457050815e-05,
"sampling/prob_min": 1.6983808018267153e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5472000002861023,
"sampling/prompt_draws_total": 16416.0,
"sampling/seen_fraction": 0.4611199975013733,
"sampling/unseen_fraction": 0.5388800024986267,
"signal/accuracy_reward/centered_abs_mean": 0.1420247435569763,
"signal/accuracy_reward/group_std_mean": 0.18404721319675446,
"signal/accuracy_reward/group_zero_std_frac": 0.4805555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07101237177848815,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07101237177848815,
"signal/advantage_abs_mean": 0.09798679202795028,
"signal/advantage_pre_scale_abs_mean": 0.09798679202795028,
"signal/advantage_pre_scale_std": 0.1785242438316345,
"signal/advantage_std": 0.1785242438316345,
"signal/brier_reward/centered_abs_mean": 0.07310768216848373,
"signal/brier_reward/group_std_mean": 0.09774392247200012,
"signal/brier_reward/group_zero_std_frac": 0.04722222276031971,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03655384108424187,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03655384108424187,
"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": 0.016503906436264516,
"signal/format_reward/group_std_mean": 0.0324388038367033,
"signal/format_reward/group_zero_std_frac": 0.8638888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008251953218132258,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008251953218132258,
"signal/mean_confidence_reward/centered_abs_mean": 0.06303982138633728,
"signal/mean_confidence_reward/group_std_mean": 0.08338831216096879,
"signal/mean_confidence_reward/group_zero_std_frac": 0.05000000055879354,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.30398199064075e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.30398199064075e-07,
"step": 230
},
{
"calibration/aurc": 0.1102212275482241,
"calibration/batch_distribution_entropy": 0.6097776666063451,
"calibration/batch_entropy_100bins": 0.3411749959336878,
"calibration/batch_entropy_10bins": 0.6097776666063451,
"calibration/batch_entropy_50bins": 0.401625685047952,
"calibration/batch_uniqueness": 0.4715644556634209,
"calibration/confidence_entropy": 0.5820979355769902,
"calibration/coverage@0%": 0.026701570680628273,
"calibration/coverage@1%": 0.026701570680628273,
"calibration/coverage@10%": 0.5605327597168789,
"calibration/coverage@15%": 0.6912604228899767,
"calibration/coverage@20%": 0.856343607976538,
"calibration/coverage@25%": 0.9007216618645737,
"calibration/coverage@30%": 0.9732620320855615,
"calibration/coverage@5%": 0.3460591668717388,
"calibration/distribution_entropy_10": 0.6097776666063451,
"calibration/distribution_entropy_100": 0.3411749959336878,
"calibration/ece": 0.13206102299034156,
"calibration/mean_confidence": 0.6639744295390552,
"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": 0.008854166666666673,
"completions/max_length": 3477.6,
"completions/max_terminated_length": 3477.6,
"completions/mean_length": 614.6240356445312,
"completions/mean_terminated_length": 620.1178588867188,
"completions/min_length": 0.0,
"completions/min_terminated_length": 176.2,
"epoch": 0.5649038461538461,
"grad_norm": 0.0005726331728510559,
"learning_rate": 4.837740384615385e-06,
"loss": -0.0107,
"num_tokens": 479633509.0,
"reward": 1.2486154317855835,
"reward_std": 0.13345372974872588,
"rewards/accuracy_reward": 0.6847222089767456,
"rewards/brier_reward": 0.8213493824005127,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9911458253860473,
"rewards/mean_confidence_reward": 0.6699392318725585,
"sampling/batch_mean_priority_error": 0.02002777777777778,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6583333333333334,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0007279346697032452,
"sampling/priority_kl": 0.030000114068388938,
"sampling/priority_scale": 0.7850202381843701,
"sampling/prob_entropy": 10.278967475891113,
"sampling/prob_max": 4.078907732036896e-05,
"sampling/prob_min": 1.713214842311572e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5592000126838684,
"sampling/prompt_draws_total": 16776.0,
"sampling/seen_fraction": 0.46881999969482424,
"sampling/unseen_fraction": 0.5311800003051758,
"signal/accuracy_reward/centered_abs_mean": 0.13311631977558136,
"signal/accuracy_reward/group_std_mean": 0.17697255909442902,
"signal/accuracy_reward/group_zero_std_frac": 0.49444445967674255,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06655815988779068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06655815988779068,
"signal/advantage_abs_mean": 0.09503975510597229,
"signal/advantage_pre_scale_abs_mean": 0.09503975510597229,
"signal/advantage_pre_scale_std": 0.18311896324157714,
"signal/advantage_std": 0.18311896324157714,
"signal/brier_reward/centered_abs_mean": 0.06759776845574379,
"signal/brier_reward/group_std_mean": 0.09351931512355804,
"signal/brier_reward/group_zero_std_frac": 0.1083333358168602,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.033798884227871896,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.033798884227871896,
"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": 0.015733507089316846,
"signal/format_reward/group_std_mean": 0.03246893137693405,
"signal/format_reward/group_zero_std_frac": 0.8555555462837219,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007866753544658423,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007866753544658423,
"signal/mean_confidence_reward/centered_abs_mean": 0.0571609228849411,
"signal/mean_confidence_reward/group_std_mean": 0.07677199840545654,
"signal/mean_confidence_reward/group_zero_std_frac": 0.11944444626569747,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.716092005059181e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.716092005059181e-07,
"step": 235
},
{
"calibration/aurc": 0.1632135963960027,
"calibration/batch_distribution_entropy": 0.4861958787539655,
"calibration/batch_entropy_100bins": 0.26268981924080526,
"calibration/batch_entropy_10bins": 0.4861958787539655,
"calibration/batch_entropy_50bins": 0.3092342049246103,
"calibration/batch_uniqueness": 0.09928411464403394,
"calibration/confidence_entropy": 0.5340064855582761,
"calibration/coverage@0%": 0.012550408048883846,
"calibration/coverage@1%": 0.012550408048883846,
"calibration/coverage@10%": 0.28228749390998803,
"calibration/coverage@15%": 0.5240212071014891,
"calibration/coverage@20%": 0.6545321803971482,
"calibration/coverage@25%": 0.8324212202344337,
"calibration/coverage@30%": 0.9084603502248711,
"calibration/coverage@5%": 0.05803197436651038,
"calibration/distribution_entropy_10": 0.4861958787539655,
"calibration/distribution_entropy_100": 0.26268981924080526,
"calibration/ece": 0.12853953979598606,
"calibration/mean_confidence": 0.7273770341281883,
"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": 0.007638888888888884,
"completions/max_length": 3554.2,
"completions/max_terminated_length": 3554.2,
"completions/mean_length": 640.9253662109375,
"completions/mean_terminated_length": 645.88857421875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 175.4,
"epoch": 0.5769230769230769,
"grad_norm": 0.0005463262787088752,
"learning_rate": 4.807692307692308e-06,
"loss": -0.0075,
"num_tokens": 490094857.0,
"reward": 1.2511736631393433,
"reward_std": 0.14524880647659302,
"rewards/accuracy_reward": 0.6892361044883728,
"rewards/brier_reward": 0.8207356572151184,
"rewards/confidence_one_or_zero": 0.0006076389050576836,
"rewards/format_reward": 0.9923611164093018,
"rewards/mean_confidence_reward": 0.7208593845367431,
"sampling/batch_mean_priority_error": 0.018305555555555554,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6611111111111111,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0007488534669391811,
"sampling/priority_kl": 0.03000020459294319,
"sampling/priority_scale": 0.7801542580360546,
"sampling/prob_entropy": 10.278952980041504,
"sampling/prob_max": 4.091102891834453e-05,
"sampling/prob_min": 1.7276007201871835e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5711999893188476,
"sampling/prompt_draws_total": 17136.0,
"sampling/seen_fraction": 0.4765933334827423,
"sampling/unseen_fraction": 0.5234066665172576,
"signal/accuracy_reward/centered_abs_mean": 0.1394205719232559,
"signal/accuracy_reward/group_std_mean": 0.18355306684970857,
"signal/accuracy_reward/group_zero_std_frac": 0.48055556416511536,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06971028596162795,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06971028596162795,
"signal/advantage_abs_mean": 0.10699738562107086,
"signal/advantage_pre_scale_abs_mean": 0.10699738562107086,
"signal/advantage_pre_scale_std": 0.1997263699769974,
"signal/advantage_std": 0.1997263699769974,
"signal/brier_reward/centered_abs_mean": 0.08055151253938675,
"signal/brier_reward/group_std_mean": 0.10744158029556275,
"signal/brier_reward/group_zero_std_frac": 0.15555555671453475,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.040275756269693375,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.040275756269693375,
"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": 1.0145398476879563e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.0145398476879563e-08,
"signal/format_reward/centered_abs_mean": 0.01343315988779068,
"signal/format_reward/group_std_mean": 0.027276124432682992,
"signal/format_reward/group_zero_std_frac": 0.8805555701255798,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00671657994389534,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00671657994389534,
"signal/mean_confidence_reward/centered_abs_mean": 0.053337682783603665,
"signal/mean_confidence_reward/group_std_mean": 0.07268032133579254,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1916666656732559,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.333767774118314e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.333767774118314e-07,
"step": 240
},
{
"calibration/aurc": 0.09553962138261952,
"calibration/batch_distribution_entropy": 0.6930865139911127,
"calibration/batch_entropy_100bins": 0.3858514920175591,
"calibration/batch_entropy_10bins": 0.6930865139911127,
"calibration/batch_entropy_50bins": 0.45421813337823486,
"calibration/batch_uniqueness": 0.5950436333597635,
"calibration/confidence_entropy": 0.5898984752654866,
"calibration/coverage@0%": 0.09646038175943286,
"calibration/coverage@1%": 0.21339159869064978,
"calibration/coverage@10%": 0.592138100501477,
"calibration/coverage@15%": 0.7390341236991794,
"calibration/coverage@20%": 0.850719112347446,
"calibration/coverage@25%": 0.9280562517404622,
"calibration/coverage@30%": 0.962962962962963,
"calibration/coverage@5%": 0.48448592927246625,
"calibration/distribution_entropy_10": 0.6930865139911127,
"calibration/distribution_entropy_100": 0.3858514920175591,
"calibration/ece": 0.14864530817334146,
"calibration/mean_confidence": 0.6301486821970961,
"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": 0.011805555555555559,
"completions/max_length": 3296.6,
"completions/max_terminated_length": 3296.6,
"completions/mean_length": 649.9940307617187,
"completions/mean_terminated_length": 657.7092407226562,
"completions/min_length": 0.0,
"completions/min_terminated_length": 177.4,
"epoch": 0.5889423076923077,
"grad_norm": 0.00048271266859956086,
"learning_rate": 4.777644230769231e-06,
"loss": -0.0127,
"num_tokens": 500675460.0,
"reward": 1.2657977104187013,
"reward_std": 0.13289434760808944,
"rewards/accuracy_reward": 0.7050347089767456,
"rewards/brier_reward": 0.8383535146713257,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9881944298744202,
"rewards/mean_confidence_reward": 0.6343446135520935,
"sampling/batch_mean_priority_error": 0.020006944444444442,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6694444444444445,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0007706952514126897,
"sampling/priority_kl": 0.02999957129359245,
"sampling/priority_scale": 0.7756844341056421,
"sampling/prob_entropy": 10.278955841064453,
"sampling/prob_max": 4.103910687263124e-05,
"sampling/prob_min": 1.7415898764738812e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5832000017166138,
"sampling/prompt_draws_total": 17496.0,
"sampling/seen_fraction": 0.4846266686916351,
"sampling/unseen_fraction": 0.5153733313083648,
"signal/accuracy_reward/centered_abs_mean": 0.13336588442325592,
"signal/accuracy_reward/group_std_mean": 0.17644108831882477,
"signal/accuracy_reward/group_zero_std_frac": 0.4944444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06668294221162796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06668294221162796,
"signal/advantage_abs_mean": 0.09448128640651703,
"signal/advantage_pre_scale_abs_mean": 0.09448128640651703,
"signal/advantage_pre_scale_std": 0.17919619977474213,
"signal/advantage_std": 0.17919619977474213,
"signal/brier_reward/centered_abs_mean": 0.07075202167034149,
"signal/brier_reward/group_std_mean": 0.09702449142932892,
"signal/brier_reward/group_zero_std_frac": 0.030555556155741216,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.035376010835170744,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.035376010835170744,
"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": 0.02027994804084301,
"signal/format_reward/group_std_mean": 0.03983944281935692,
"signal/format_reward/group_zero_std_frac": 0.8305555701255798,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010139974020421504,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010139974020421504,
"signal/mean_confidence_reward/centered_abs_mean": 0.06342665180563926,
"signal/mean_confidence_reward/group_std_mean": 0.08341833353042602,
"signal/mean_confidence_reward/group_zero_std_frac": 0.030555556155741216,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.342664960357069e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.342664960357069e-07,
"step": 245
},
{
"calibration/aurc": 0.16545562957720988,
"calibration/batch_distribution_entropy": 0.6516094600536402,
"calibration/batch_entropy_100bins": 0.3457829111167018,
"calibration/batch_entropy_10bins": 0.6516094600536402,
"calibration/batch_entropy_50bins": 0.40705004824595303,
"calibration/batch_uniqueness": 0.46248375969187416,
"calibration/confidence_entropy": 0.5245775845540271,
"calibration/coverage@0%": 0.007853403141361256,
"calibration/coverage@1%": 0.007853403141361256,
"calibration/coverage@10%": 0.35422205268554696,
"calibration/coverage@15%": 0.5339711789098691,
"calibration/coverage@20%": 0.634497870918797,
"calibration/coverage@25%": 0.6957461049419,
"calibration/coverage@30%": 0.9298429319371728,
"calibration/coverage@5%": 0.13723436933694882,
"calibration/distribution_entropy_10": 0.6516094600536402,
"calibration/distribution_entropy_100": 0.3457829111167018,
"calibration/ece": 0.11143773320785884,
"calibration/mean_confidence": 0.7164784939762043,
"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": 0.01215277777777779,
"completions/max_length": 3935.2,
"completions/max_terminated_length": 3935.2,
"completions/mean_length": 682.9305541992187,
"completions/mean_terminated_length": 691.3621459960938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 178.2,
"epoch": 0.6009615384615384,
"grad_norm": 0.0005736547173000872,
"learning_rate": 4.747596153846154e-06,
"loss": -0.0129,
"num_tokens": 511674948.0,
"reward": 1.2450485229492188,
"reward_std": 0.15984397232532502,
"rewards/accuracy_reward": 0.6749131798744201,
"rewards/brier_reward": 0.8274962782859803,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9876736164093017,
"rewards/mean_confidence_reward": 0.6966883778572083,
"sampling/batch_mean_priority_error": 0.014520833333333325,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.638888888888889,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0007915808353573083,
"sampling/priority_kl": 0.029998910427093507,
"sampling/priority_scale": 0.7715063988929615,
"sampling/prob_entropy": 10.27894859313965,
"sampling/prob_max": 4.116657073609531e-05,
"sampling/prob_min": 1.7544802904012614e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5952000021934509,
"sampling/prompt_draws_total": 17856.0,
"sampling/seen_fraction": 0.4925266742706299,
"sampling/unseen_fraction": 0.5074733257293701,
"signal/accuracy_reward/centered_abs_mean": 0.15193684995174409,
"signal/accuracy_reward/group_std_mean": 0.1979072332382202,
"signal/accuracy_reward/group_zero_std_frac": 0.4416666805744171,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07596842497587204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07596842497587204,
"signal/advantage_abs_mean": 0.11579143851995469,
"signal/advantage_pre_scale_abs_mean": 0.11579143851995469,
"signal/advantage_pre_scale_std": 0.20864003598690034,
"signal/advantage_std": 0.20864003598690034,
"signal/brier_reward/centered_abs_mean": 0.09156632274389268,
"signal/brier_reward/group_std_mean": 0.12338759154081344,
"signal/brier_reward/group_zero_std_frac": 0.06388888973742723,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04578316137194634,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04578316137194634,
"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": 0.02147352434694767,
"signal/format_reward/group_std_mean": 0.0425883911550045,
"signal/format_reward/group_zero_std_frac": 0.8166666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010736762173473835,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010736762173473835,
"signal/mean_confidence_reward/centered_abs_mean": 0.07210828363895416,
"signal/mean_confidence_reward/group_std_mean": 0.0960259661078453,
"signal/mean_confidence_reward/group_zero_std_frac": 0.08333333432674409,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.210828130155278e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.210828130155278e-07,
"step": 250
},
{
"epoch": 0.6009615384615384,
"eval_calibration/aurc": 0.1658275768643548,
"eval_calibration/batch_distribution_entropy": 0.653935370011339,
"eval_calibration/batch_entropy_100bins": 0.3449556613935705,
"eval_calibration/batch_entropy_10bins": 0.653935370011339,
"eval_calibration/batch_entropy_50bins": 0.40607622325667125,
"eval_calibration/batch_uniqueness": 0.4734441366574332,
"eval_calibration/confidence_entropy": 0.4784185724453361,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.35175438596491226,
"eval_calibration/coverage@15%": 0.35175438596491226,
"eval_calibration/coverage@20%": 0.6894736842105263,
"eval_calibration/coverage@25%": 0.8719298245614036,
"eval_calibration/coverage@30%": 0.9780701754385965,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.653935370011339,
"eval_calibration/distribution_entropy_100": 0.3449556613935705,
"eval_calibration/ece": 0.07307017543859642,
"eval_calibration/mean_confidence": 0.7571052631578947,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.009548611111111105,
"eval_completions/max_length": 2102.1666666666665,
"eval_completions/max_terminated_length": 2102.1666666666665,
"eval_completions/mean_length": 644.3861287434896,
"eval_completions/mean_terminated_length": 650.6084696451823,
"eval_completions/min_length": 54.5,
"eval_completions/min_terminated_length": 229.16666666666666,
"eval_loss": 0.0,
"eval_num_tokens": 511674948.0,
"eval_reward": 1.2462879220644634,
"eval_reward_std": 0.348220556974411,
"eval_rewards/accuracy_reward": 0.6822916666666666,
"eval_rewards/brier_reward": 0.8206857641537985,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9895833333333334,
"eval_rewards/mean_confidence_reward": 0.7492187718550364,
"eval_runtime": 203.1518,
"eval_samples_per_second": 4.922,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4181857605775197,
"eval_signal/accuracy_reward/group_std_mean": 0.4631023903687795,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20909288028875986,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.20909288028875986,
"eval_signal/advantage_abs_mean": 0.30280476808547974,
"eval_signal/advantage_pre_scale_abs_mean": 0.30280476808547974,
"eval_signal/advantage_pre_scale_std": 0.34675098955631256,
"eval_signal/advantage_std": 0.34675098955631256,
"eval_signal/brier_reward/centered_abs_mean": 0.19453559319178262,
"eval_signal/brier_reward/group_std_mean": 0.24338269233703613,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.09726779659589131,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.09726779659589131,
"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": 0.020182291356225807,
"eval_signal/format_reward/group_std_mean": 0.05892556471129259,
"eval_signal/format_reward/group_zero_std_frac": 0.6666666865348816,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.010091145678112904,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.010091145678112904,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.13532442972064018,
"eval_signal/mean_confidence_reward/group_std_mean": 0.18789123992125192,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.3532442532474913e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.3532442532474913e-06,
"eval_steps_per_second": 0.03,
"step": 250
},
{
"epoch": 0.6009615384615384,
"step": 250,
"train_probe_calibration/aurc": 0.13297189810899127,
"train_probe_calibration/batch_distribution_entropy": 0.6515029645632585,
"train_probe_calibration/batch_entropy_100bins": 0.3455663791755371,
"train_probe_calibration/batch_entropy_10bins": 0.6515029645632585,
"train_probe_calibration/batch_entropy_50bins": 0.40679515034827146,
"train_probe_calibration/batch_uniqueness": 0.4740345275528414,
"train_probe_calibration/confidence_entropy": 0.4782895608513383,
"train_probe_calibration/coverage@0%": 0.0008710801393728223,
"train_probe_calibration/coverage@1%": 0.0008710801393728223,
"train_probe_calibration/coverage@10%": 0.35714285714285715,
"train_probe_calibration/coverage@15%": 0.6933797909407665,
"train_probe_calibration/coverage@20%": 0.8179442508710801,
"train_probe_calibration/coverage@25%": 0.8945993031358885,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0008710801393728223,
"train_probe_calibration/distribution_entropy_10": 0.6515029645632585,
"train_probe_calibration/distribution_entropy_100": 0.3455663791755371,
"train_probe_calibration/ece": 0.06977351916376293,
"train_probe_calibration/mean_confidence": 0.7554878048780488,
"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": 2169.3333333333335,
"train_probe_completions/max_terminated_length": 2169.3333333333335,
"train_probe_completions/mean_length": 656.5750223795573,
"train_probe_completions/mean_terminated_length": 658.8591105143229,
"train_probe_completions/min_length": 58.833333333333336,
"train_probe_completions/min_terminated_length": 201.83333333333334,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 511674948.0,
"train_probe_reward": 1.277138610680898,
"train_probe_reward_std": 0.3227684100468953,
"train_probe_rewards/accuracy_reward": 0.7161458333333334,
"train_probe_rewards/brier_reward": 0.8415885468324026,
"train_probe_rewards/confidence_one_or_zero": 0.0008680555814256271,
"train_probe_rewards/format_reward": 0.9965277910232544,
"train_probe_rewards/mean_confidence_reward": 0.7528645694255829,
"train_probe_runtime": 166.5968,
"train_probe_samples_per_second": 6.003,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3948025157054265,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4493370900551478,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19740125785271326,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.19740125785271326,
"train_probe_signal/advantage_abs_mean": 0.27557384471098584,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.27557384471098584,
"train_probe_signal/advantage_pre_scale_std": 0.32040198644002277,
"train_probe_signal/advantage_std": 0.32040198644002277,
"train_probe_signal/brier_reward/centered_abs_mean": 0.17354112366835275,
"train_probe_signal/brier_reward/group_std_mean": 0.22265213479598364,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08677056183417638,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.08677056183417638,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0016818575871487458,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0049104637776811915,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222288449606,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151002e-08,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151002e-08,
"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": 0.1332410884400209,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.17862537999947867,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.332410799174492e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.332410799174492e-06,
"train_probe_steps_per_second": 0.036
},
{
"calibration/aurc": 0.11053866199233855,
"calibration/batch_distribution_entropy": 0.6553593655311809,
"calibration/batch_entropy_100bins": 0.3494508559530446,
"calibration/batch_entropy_10bins": 0.6553593655311809,
"calibration/batch_entropy_50bins": 0.41136789356044556,
"calibration/batch_uniqueness": 0.47738488620607616,
"calibration/confidence_entropy": 0.49347215678980705,
"calibration/coverage@0%": 0.0010498687664041995,
"calibration/coverage@1%": 0.0010498687664041995,
"calibration/coverage@10%": 0.6203634476385791,
"calibration/coverage@15%": 0.7494999706104414,
"calibration/coverage@20%": 0.7756438133090919,
"calibration/coverage@25%": 0.8393311222141723,
"calibration/coverage@30%": 0.9272251308900523,
"calibration/coverage@5%": 0.39120135439401754,
"calibration/distribution_entropy_10": 0.6553593655311809,
"calibration/distribution_entropy_100": 0.3494508559530446,
"calibration/ece": 0.15268235447490713,
"calibration/mean_confidence": 0.7408424220083726,
"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": 0.00894097222222221,
"completions/max_length": 3785.2,
"completions/max_terminated_length": 3785.2,
"completions/mean_length": 660.906689453125,
"completions/mean_terminated_length": 666.9714111328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 168.6,
"epoch": 0.6129807692307693,
"grad_norm": 0.00047295857802964747,
"learning_rate": 4.7175480769230775e-06,
"loss": -0.0093,
"num_tokens": 522387505.0,
"reward": 1.2695069313049316,
"reward_std": 0.14805976301431656,
"rewards/accuracy_reward": 0.7131944537162781,
"rewards/brier_reward": 0.8347456574440002,
"rewards/confidence_one_or_zero": 0.0005208333546761424,
"rewards/format_reward": 0.9910590410232544,
"rewards/mean_confidence_reward": 0.7380295157432556,
"sampling/batch_mean_priority_error": 0.011499999999999993,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6277777777777779,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0008047144394367934,
"sampling/priority_kl": 0.03000001311302185,
"sampling/priority_scale": 0.7673150241142139,
"sampling/prob_entropy": 10.278948783874512,
"sampling/prob_max": 4.1286813939223066e-05,
"sampling/prob_min": 1.7676256538834424e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6072000026702881,
"sampling/prompt_draws_total": 18216.0,
"sampling/seen_fraction": 0.49989998936653135,
"sampling/unseen_fraction": 0.5001000106334687,
"signal/accuracy_reward/centered_abs_mean": 0.1382269948720932,
"signal/accuracy_reward/group_std_mean": 0.1822678416967392,
"signal/accuracy_reward/group_zero_std_frac": 0.48055556416511536,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0691134974360466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0691134974360466,
"signal/advantage_abs_mean": 0.10847936570644379,
"signal/advantage_pre_scale_abs_mean": 0.10847936570644379,
"signal/advantage_pre_scale_std": 0.2013978600502014,
"signal/advantage_std": 0.2013978600502014,
"signal/brier_reward/centered_abs_mean": 0.08796882182359696,
"signal/brier_reward/group_std_mean": 0.11736268252134323,
"signal/brier_reward/group_zero_std_frac": 0.06666666846722365,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04398441091179848,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04398441091179848,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0009874132345430553,
"signal/confidence_one_or_zero/group_std_mean": 0.0023483963683247565,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9888888835906983,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 9.874132445020223e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 9.874132445020223e-09,
"signal/format_reward/centered_abs_mean": 0.01550021693110466,
"signal/format_reward/group_std_mean": 0.030670232325792312,
"signal/format_reward/group_zero_std_frac": 0.8666666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00775010846555233,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00775010846555233,
"signal/mean_confidence_reward/centered_abs_mean": 0.06791991218924523,
"signal/mean_confidence_reward/group_std_mean": 0.0899539515376091,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0750000024214387,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.791991154386778e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.791991154386778e-07,
"step": 255
},
{
"calibration/aurc": 0.13061344417676773,
"calibration/batch_distribution_entropy": 0.7119939536306921,
"calibration/batch_entropy_100bins": 0.3947636348685257,
"calibration/batch_entropy_10bins": 0.7119939536306921,
"calibration/batch_entropy_50bins": 0.4647093637451295,
"calibration/batch_uniqueness": 0.6090274589011522,
"calibration/confidence_entropy": 0.5537789059431736,
"calibration/coverage@0%": 0.0994236422574895,
"calibration/coverage@1%": 0.0994236422574895,
"calibration/coverage@10%": 0.390270058015489,
"calibration/coverage@15%": 0.6097525669281451,
"calibration/coverage@20%": 0.7780259450593018,
"calibration/coverage@25%": 0.8862583121109224,
"calibration/coverage@30%": 0.9415506508205999,
"calibration/coverage@5%": 0.31689589896770565,
"calibration/distribution_entropy_10": 0.7119939536306921,
"calibration/distribution_entropy_100": 0.3947636348685257,
"calibration/ece": 0.1114447297112239,
"calibration/mean_confidence": 0.6725977867976177,
"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": 0.008072916666666697,
"completions/max_length": 2881.2,
"completions/max_terminated_length": 2881.2,
"completions/mean_length": 658.5054809570313,
"completions/mean_terminated_length": 663.83076171875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 183.0,
"epoch": 0.625,
"grad_norm": 0.0004797276924364269,
"learning_rate": 4.6875000000000004e-06,
"loss": -0.0062,
"num_tokens": 533054992.0,
"reward": 1.279056715965271,
"reward_std": 0.14249440133571625,
"rewards/accuracy_reward": 0.7252604126930237,
"rewards/brier_reward": 0.8409123301506043,
"rewards/confidence_one_or_zero": 0.0003472222248092294,
"rewards/format_reward": 0.9919270753860474,
"rewards/mean_confidence_reward": 0.6728732585906982,
"sampling/batch_mean_priority_error": 0.013777777777777772,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6083333333333334,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0008190826396457851,
"sampling/priority_kl": 0.029998597875237464,
"sampling/priority_scale": 0.7635116636054591,
"sampling/prob_entropy": 10.27892894744873,
"sampling/prob_max": 4.141414974583313e-05,
"sampling/prob_min": 1.780548664100934e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6192000031471252,
"sampling/prompt_draws_total": 18576.0,
"sampling/seen_fraction": 0.5075666785240174,
"sampling/unseen_fraction": 0.4924333214759827,
"signal/accuracy_reward/centered_abs_mean": 0.14213866889476776,
"signal/accuracy_reward/group_std_mean": 0.19297046065330506,
"signal/accuracy_reward/group_zero_std_frac": 0.42777777910232545,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07106933444738388,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07106933444738388,
"signal/advantage_abs_mean": 0.10037857741117477,
"signal/advantage_pre_scale_abs_mean": 0.10037857741117477,
"signal/advantage_pre_scale_std": 0.18658272922039032,
"signal/advantage_std": 0.18658272922039032,
"signal/brier_reward/centered_abs_mean": 0.08256456702947616,
"signal/brier_reward/group_std_mean": 0.11133366227149963,
"signal/brier_reward/group_zero_std_frac": 0.01666666679084301,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04128228351473808,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04128228351473808,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0006618923507630825,
"signal/confidence_one_or_zero/group_std_mean": 0.0016652445774525404,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 6.618923364953844e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 6.618923364953844e-09,
"signal/format_reward/centered_abs_mean": 0.014415147714316845,
"signal/format_reward/group_std_mean": 0.02902185283601284,
"signal/format_reward/group_zero_std_frac": 0.875,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0072075738571584225,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0072075738571584225,
"signal/mean_confidence_reward/centered_abs_mean": 0.07456976771354676,
"signal/mean_confidence_reward/group_std_mean": 0.09588464945554734,
"signal/mean_confidence_reward/group_zero_std_frac": 0.01666666679084301,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.456976504727209e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.456976504727209e-07,
"step": 260
},
{
"calibration/aurc": 0.18852691153159343,
"calibration/batch_distribution_entropy": 0.6839669052669889,
"calibration/batch_entropy_100bins": 0.3899278269620705,
"calibration/batch_entropy_10bins": 0.6839669052669889,
"calibration/batch_entropy_50bins": 0.459016728920367,
"calibration/batch_uniqueness": 0.5566917367060518,
"calibration/confidence_entropy": 0.508166540858328,
"calibration/coverage@0%": 0.007313596491228071,
"calibration/coverage@1%": 0.007313596491228071,
"calibration/coverage@10%": 0.2226528216755989,
"calibration/coverage@15%": 0.4099509467330068,
"calibration/coverage@20%": 0.5052252704677362,
"calibration/coverage@25%": 0.7828199512629264,
"calibration/coverage@30%": 0.8094723100178646,
"calibration/coverage@5%": 0.17213198834226556,
"calibration/distribution_entropy_10": 0.6839669052669889,
"calibration/distribution_entropy_100": 0.3899278269620705,
"calibration/ece": 0.12156555954972767,
"calibration/mean_confidence": 0.6929760522571534,
"calibration/unique_confidence_per_question": 0.023958333333333335,
"calibration/unique_confidences": 9.2,
"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": 0.0078125,
"completions/max_length": 3458.6,
"completions/max_terminated_length": 3458.6,
"completions/mean_length": 680.2973999023437,
"completions/mean_terminated_length": 685.6692993164063,
"completions/min_length": 0.0,
"completions/min_terminated_length": 210.2,
"epoch": 0.6370192307692307,
"grad_norm": 0.00045690243132412434,
"learning_rate": 4.657451923076923e-06,
"loss": -0.0076,
"num_tokens": 543990706.0,
"reward": 1.2635553359985352,
"reward_std": 0.13125103265047072,
"rewards/accuracy_reward": 0.6978298544883728,
"rewards/brier_reward": 0.8370790004730224,
"rewards/confidence_one_or_zero": 0.009027777757728473,
"rewards/format_reward": 0.9921875,
"rewards/mean_confidence_reward": 0.7020920038223266,
"sampling/batch_mean_priority_error": 0.011541666666666662,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6444444444444445,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0008350071730092168,
"sampling/priority_kl": 0.029999512434005737,
"sampling/priority_scale": 0.7598387658363208,
"sampling/prob_entropy": 10.278968620300294,
"sampling/prob_max": 4.154030175413936e-05,
"sampling/prob_min": 1.7932389891939237e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6311999917030334,
"sampling/prompt_draws_total": 18936.0,
"sampling/seen_fraction": 0.5150799989700318,
"sampling/unseen_fraction": 0.48492000102996824,
"signal/accuracy_reward/centered_abs_mean": 0.12266167551279068,
"signal/accuracy_reward/group_std_mean": 0.16602103412151337,
"signal/accuracy_reward/group_zero_std_frac": 0.5166666746139527,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06133083775639534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06133083775639534,
"signal/advantage_abs_mean": 0.0934511199593544,
"signal/advantage_pre_scale_abs_mean": 0.0934511199593544,
"signal/advantage_pre_scale_std": 0.17902643084526063,
"signal/advantage_std": 0.17902643084526063,
"signal/brier_reward/centered_abs_mean": 0.0809135839343071,
"signal/brier_reward/group_std_mean": 0.10942707508802414,
"signal/brier_reward/group_zero_std_frac": 0.02500000037252903,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04045679196715355,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04045679196715355,
"signal/confidence_one_or_zero/centered_abs_mean": 0.01451822918606922,
"signal/confidence_one_or_zero/group_std_mean": 0.02576328720897436,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.8972222208976746,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.451822868148156e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.451822868148156e-07,
"signal/format_reward/centered_abs_mean": 0.013216145616024733,
"signal/format_reward/group_std_mean": 0.026454249024391176,
"signal/format_reward/group_zero_std_frac": 0.8833333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.006608072808012367,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.006608072808012367,
"signal/mean_confidence_reward/centered_abs_mean": 0.07293184846639633,
"signal/mean_confidence_reward/group_std_mean": 0.09473260641098022,
"signal/mean_confidence_reward/group_zero_std_frac": 0.02500000037252903,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.293184808077058e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.293184808077058e-07,
"step": 265
},
{
"calibration/aurc": 0.1961220920497752,
"calibration/batch_distribution_entropy": 0.5809251673671703,
"calibration/batch_entropy_100bins": 0.34474063958986223,
"calibration/batch_entropy_10bins": 0.5809251673671703,
"calibration/batch_entropy_50bins": 0.4058231030683695,
"calibration/batch_uniqueness": 0.3972966344560953,
"calibration/confidence_entropy": 0.438974753112911,
"calibration/coverage@0%": 0.011518324607329843,
"calibration/coverage@1%": 0.11727748691099475,
"calibration/coverage@10%": 0.1649214659685864,
"calibration/coverage@15%": 0.4003130425273058,
"calibration/coverage@20%": 0.5895827851102029,
"calibration/coverage@25%": 0.6401852823306403,
"calibration/coverage@30%": 0.8300192746708952,
"calibration/coverage@5%": 0.1324607329842932,
"calibration/distribution_entropy_10": 0.5809251673671703,
"calibration/distribution_entropy_100": 0.34474063958986223,
"calibration/ece": 0.13466166017912237,
"calibration/mean_confidence": 0.759952178567523,
"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": 0.006076388888888884,
"completions/max_length": 3589.8,
"completions/max_terminated_length": 3589.8,
"completions/mean_length": 707.28369140625,
"completions/mean_terminated_length": 711.668896484375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 252.8,
"epoch": 0.6490384615384616,
"grad_norm": 0.0005051839398220181,
"learning_rate": 4.627403846153847e-06,
"loss": -0.0071,
"num_tokens": 555252822.0,
"reward": 1.2621881484985351,
"reward_std": 0.14973460435867308,
"rewards/accuracy_reward": 0.6929687619209289,
"rewards/brier_reward": 0.8376414775848389,
"rewards/confidence_one_or_zero": 0.05598958358168602,
"rewards/format_reward": 0.99375,
"rewards/mean_confidence_reward": 0.7462586879730224,
"sampling/batch_mean_priority_error": 0.016298611111111104,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5972222222222222,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0008497483097016812,
"sampling/priority_kl": 0.030001790821552278,
"sampling/priority_scale": 0.7562869013054296,
"sampling/prob_entropy": 10.278965950012207,
"sampling/prob_max": 4.166323342360556e-05,
"sampling/prob_min": 1.8056220142170787e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6432000041007996,
"sampling/prompt_draws_total": 19296.0,
"sampling/seen_fraction": 0.5222799897193908,
"sampling/unseen_fraction": 0.4777200102806091,
"signal/accuracy_reward/centered_abs_mean": 0.14156358540058137,
"signal/accuracy_reward/group_std_mean": 0.19307936429977418,
"signal/accuracy_reward/group_zero_std_frac": 0.43055555820465086,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07078179270029068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07078179270029068,
"signal/advantage_abs_mean": 0.10628217607736587,
"signal/advantage_pre_scale_abs_mean": 0.10628217607736587,
"signal/advantage_pre_scale_std": 0.19850011467933654,
"signal/advantage_std": 0.19850011467933654,
"signal/brier_reward/centered_abs_mean": 0.09157379269599915,
"signal/brier_reward/group_std_mean": 0.12708270996809007,
"signal/brier_reward/group_zero_std_frac": 0.0305555559694767,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.045786896347999574,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.045786896347999574,
"signal/confidence_one_or_zero/centered_abs_mean": 0.08119032084941864,
"signal/confidence_one_or_zero/group_std_mean": 0.11917831152677535,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.6194444537162781,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 8.119032031572714e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.119032031572714e-07,
"signal/format_reward/centered_abs_mean": 0.011154513899236917,
"signal/format_reward/group_std_mean": 0.023835764080286027,
"signal/format_reward/group_zero_std_frac": 0.8916666746139527,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005577256949618459,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005577256949618459,
"signal/mean_confidence_reward/centered_abs_mean": 0.07271484434604644,
"signal/mean_confidence_reward/group_std_mean": 0.09719483703374862,
"signal/mean_confidence_reward/group_zero_std_frac": 0.03611111212521791,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.271484378179593e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.271484378179593e-07,
"step": 270
},
{
"calibration/aurc": 0.1864150533610327,
"calibration/batch_distribution_entropy": 0.6408848797337117,
"calibration/batch_entropy_100bins": 0.4092029903128368,
"calibration/batch_entropy_10bins": 0.6408848797337117,
"calibration/batch_entropy_50bins": 0.4817071393473589,
"calibration/batch_uniqueness": 0.6019178475709618,
"calibration/confidence_entropy": 0.42906083533825345,
"calibration/coverage@0%": 0.031413612565445025,
"calibration/coverage@1%": 0.031413612565445025,
"calibration/coverage@10%": 0.27362383769329013,
"calibration/coverage@15%": 0.4249930657811958,
"calibration/coverage@20%": 0.5929327449426574,
"calibration/coverage@25%": 0.798945107189404,
"calibration/coverage@30%": 0.878995504394491,
"calibration/coverage@5%": 0.14640315539772328,
"calibration/distribution_entropy_10": 0.6408848797337117,
"calibration/distribution_entropy_100": 0.4092029903128368,
"calibration/ece": 0.10553314892878747,
"calibration/mean_confidence": 0.7237165907374583,
"calibration/unique_confidence_per_question": 0.025520833333333336,
"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": 0.008940972222222232,
"completions/max_length": 3354.2,
"completions/max_terminated_length": 3354.2,
"completions/mean_length": 684.2703125,
"completions/mean_terminated_length": 690.500439453125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 180.6,
"epoch": 0.6610576923076923,
"grad_norm": 0.0005132072838023305,
"learning_rate": 4.597355769230769e-06,
"loss": -0.0086,
"num_tokens": 566232160.0,
"reward": 1.2669503688812256,
"reward_std": 0.13826109766960143,
"rewards/accuracy_reward": 0.7004340291023254,
"rewards/brier_reward": 0.8423903942108154,
"rewards/confidence_one_or_zero": 0.12612847462296486,
"rewards/format_reward": 0.9910590171813964,
"rewards/mean_confidence_reward": 0.7350477337837219,
"sampling/batch_mean_priority_error": 0.015659722222222217,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6027777777777779,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0008683778694830834,
"sampling/priority_kl": 0.029999760910868645,
"sampling/priority_scale": 0.7529768407111987,
"sampling/prob_entropy": 10.2789737701416,
"sampling/prob_max": 4.178994277026504e-05,
"sampling/prob_min": 1.8177531092078426e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6552000045776367,
"sampling/prompt_draws_total": 19656.0,
"sampling/seen_fraction": 0.5296066761016845,
"sampling/unseen_fraction": 0.47039332389831545,
"signal/accuracy_reward/centered_abs_mean": 0.12946506291627885,
"signal/accuracy_reward/group_std_mean": 0.1760338395833969,
"signal/accuracy_reward/group_zero_std_frac": 0.4833333432674408,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06473253145813943,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06473253145813943,
"signal/advantage_abs_mean": 0.09839854985475541,
"signal/advantage_pre_scale_abs_mean": 0.09839854985475541,
"signal/advantage_pre_scale_std": 0.19160001575946808,
"signal/advantage_std": 0.19160001575946808,
"signal/brier_reward/centered_abs_mean": 0.08714145421981812,
"signal/brier_reward/group_std_mean": 0.11808212995529174,
"signal/brier_reward/group_zero_std_frac": 0.027777778543531896,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04357072710990906,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04357072710990906,
"signal/confidence_one_or_zero/centered_abs_mean": 0.12776150181889534,
"signal/confidence_one_or_zero/group_std_mean": 0.1739050790667534,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.4805555582046509,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.2776150015270105e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.2776150015270105e-06,
"signal/format_reward/centered_abs_mean": 0.01463216133415699,
"signal/format_reward/group_std_mean": 0.028245595470070838,
"signal/format_reward/group_zero_std_frac": 0.8805555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007316080667078495,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007316080667078495,
"signal/mean_confidence_reward/centered_abs_mean": 0.07807481437921523,
"signal/mean_confidence_reward/group_std_mean": 0.10356819778680801,
"signal/mean_confidence_reward/group_zero_std_frac": 0.027777778543531896,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.807481210875266e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.807481210875266e-07,
"step": 275
},
{
"calibration/aurc": 0.16869080575234072,
"calibration/batch_distribution_entropy": 0.664091431699852,
"calibration/batch_entropy_100bins": 0.3700299330201189,
"calibration/batch_entropy_10bins": 0.664091431699852,
"calibration/batch_entropy_50bins": 0.43559325011712813,
"calibration/batch_uniqueness": 0.48766579592898707,
"calibration/confidence_entropy": 0.4723369159996705,
"calibration/coverage@0%": 0.06598825401202599,
"calibration/coverage@1%": 0.06598825401202599,
"calibration/coverage@10%": 0.26032505422653046,
"calibration/coverage@15%": 0.4990277088435562,
"calibration/coverage@20%": 0.5279106395236692,
"calibration/coverage@25%": 0.8020478480261879,
"calibration/coverage@30%": 0.9015880044183131,
"calibration/coverage@5%": 0.08221862050417258,
"calibration/distribution_entropy_10": 0.664091431699852,
"calibration/distribution_entropy_100": 0.3700299330201189,
"calibration/ece": 0.11553050746975288,
"calibration/mean_confidence": 0.686896090546837,
"calibration/unique_confidence_per_question": 0.025520833333333336,
"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": 0.00694444444444442,
"completions/max_length": 3185.0,
"completions/max_terminated_length": 3185.0,
"completions/mean_length": 674.0809936523438,
"completions/mean_terminated_length": 678.8185791015625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 198.4,
"epoch": 0.6730769230769231,
"grad_norm": 0.0005486437003128231,
"learning_rate": 4.567307692307692e-06,
"loss": -0.0059,
"num_tokens": 577073221.0,
"reward": 1.272284698486328,
"reward_std": 0.13358466923236847,
"rewards/accuracy_reward": 0.7074652671813965,
"rewards/brier_reward": 0.8440346002578736,
"rewards/confidence_one_or_zero": 0.0027777778159361333,
"rewards/format_reward": 0.9930555582046509,
"rewards/mean_confidence_reward": 0.6953281283378601,
"sampling/batch_mean_priority_error": 0.015076388888888872,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6222222222222221,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0008863226044923067,
"sampling/priority_kl": 0.03000061586499214,
"sampling/priority_scale": 0.7499105036491528,
"sampling/prob_entropy": 10.278962135314941,
"sampling/prob_max": 4.191710904706269e-05,
"sampling/prob_min": 1.829475695558358e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6672000050544739,
"sampling/prompt_draws_total": 20016.0,
"sampling/seen_fraction": 0.5368266701698303,
"sampling/unseen_fraction": 0.4631733298301697,
"signal/accuracy_reward/centered_abs_mean": 0.12867838591337205,
"signal/accuracy_reward/group_std_mean": 0.17682747542858124,
"signal/accuracy_reward/group_zero_std_frac": 0.4694444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06433919295668603,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06433919295668603,
"signal/advantage_abs_mean": 0.09517031162977219,
"signal/advantage_pre_scale_abs_mean": 0.09517031162977219,
"signal/advantage_pre_scale_std": 0.18133316040039063,
"signal/advantage_std": 0.18133316040039063,
"signal/brier_reward/centered_abs_mean": 0.08287988752126693,
"signal/brier_reward/group_std_mean": 0.11331825703382492,
"signal/brier_reward/group_zero_std_frac": 0.14722222685813904,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.041439943760633466,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.041439943760633466,
"signal/confidence_one_or_zero/centered_abs_mean": 0.005208333174232393,
"signal/confidence_one_or_zero/group_std_mean": 0.012185867689549924,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9416666746139526,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.2083332491292825e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.2083332491292825e-08,
"signal/format_reward/centered_abs_mean": 0.01195746548473835,
"signal/format_reward/group_std_mean": 0.022025031968951225,
"signal/format_reward/group_zero_std_frac": 0.9111111283302307,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005978732742369175,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005978732742369175,
"signal/mean_confidence_reward/centered_abs_mean": 0.07131211757659912,
"signal/mean_confidence_reward/group_std_mean": 0.09345167428255081,
"signal/mean_confidence_reward/group_zero_std_frac": 0.16111111044883727,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.13121221451729e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.13121221451729e-07,
"step": 280
},
{
"calibration/aurc": 0.10583889998251635,
"calibration/batch_distribution_entropy": 0.6460838299377851,
"calibration/batch_entropy_100bins": 0.35334087685404303,
"calibration/batch_entropy_10bins": 0.6460838299377851,
"calibration/batch_entropy_50bins": 0.4159471632250901,
"calibration/batch_uniqueness": 0.4410605124714365,
"calibration/confidence_entropy": 0.4774449305819912,
"calibration/coverage@0%": 0.1712793733681462,
"calibration/coverage@1%": 0.28541270670147956,
"calibration/coverage@10%": 0.5194118255004352,
"calibration/coverage@15%": 0.7913901762402088,
"calibration/coverage@20%": 0.8342494342906877,
"calibration/coverage@25%": 0.9118363794604004,
"calibration/coverage@30%": 0.9326697127937337,
"calibration/coverage@5%": 0.447759954308094,
"calibration/distribution_entropy_10": 0.6460838299377851,
"calibration/distribution_entropy_100": 0.35334087685404303,
"calibration/ece": 0.13417588936031327,
"calibration/mean_confidence": 0.7112051397954744,
"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": 0.00590277777777779,
"completions/max_length": 3611.0,
"completions/max_terminated_length": 3611.0,
"completions/mean_length": 694.354443359375,
"completions/mean_terminated_length": 698.5254760742188,
"completions/min_length": 0.0,
"completions/min_terminated_length": 226.4,
"epoch": 0.6850961538461539,
"grad_norm": 0.00047999402158893645,
"learning_rate": 4.537259615384616e-06,
"loss": -0.0058,
"num_tokens": 588172472.0,
"reward": 1.298223352432251,
"reward_std": 0.11750643104314804,
"rewards/accuracy_reward": 0.7395833253860473,
"rewards/brier_reward": 0.8629256367683411,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9939236164093017,
"rewards/mean_confidence_reward": 0.7073857069015503,
"sampling/batch_mean_priority_error": 0.014631944444444428,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5805555555555556,
"sampling/error_ema_max": 0.0756249949336052,
"sampling/error_ema_mean": 0.0009036242845468223,
"sampling/priority_kl": 0.029998960718512536,
"sampling/priority_scale": 0.7471439658896998,
"sampling/prob_entropy": 10.278946495056152,
"sampling/prob_max": 4.204997385386378e-05,
"sampling/prob_min": 1.8408967298455538e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6791999936103821,
"sampling/prompt_draws_total": 20376.0,
"sampling/seen_fraction": 0.5442400097846984,
"sampling/unseen_fraction": 0.4557599902153015,
"signal/accuracy_reward/centered_abs_mean": 0.11642795205116271,
"signal/accuracy_reward/group_std_mean": 0.15990724563598632,
"signal/accuracy_reward/group_zero_std_frac": 0.5277777850627899,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05821397602558136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05821397602558136,
"signal/advantage_abs_mean": 0.08154689669609069,
"signal/advantage_pre_scale_abs_mean": 0.08154689669609069,
"signal/advantage_pre_scale_std": 0.16533636152744294,
"signal/advantage_std": 0.16533636152744294,
"signal/brier_reward/centered_abs_mean": 0.07062470018863679,
"signal/brier_reward/group_std_mean": 0.09852024167776108,
"signal/brier_reward/group_zero_std_frac": 0.14444444626569747,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03531235009431839,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03531235009431839,
"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": 0.011013455037027598,
"signal/format_reward/group_std_mean": 0.022193879634141923,
"signal/format_reward/group_zero_std_frac": 0.9055555701255799,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005506727518513799,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005506727518513799,
"signal/mean_confidence_reward/centered_abs_mean": 0.06535256057977676,
"signal/mean_confidence_reward/group_std_mean": 0.0859237179160118,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1611111119389534,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.53525580673886e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.53525580673886e-07,
"step": 285
},
{
"calibration/aurc": 0.17162131598213481,
"calibration/batch_distribution_entropy": 0.6749536064273652,
"calibration/batch_entropy_100bins": 0.3733026957392366,
"calibration/batch_entropy_10bins": 0.6749536064273652,
"calibration/batch_entropy_50bins": 0.4394458934372163,
"calibration/batch_uniqueness": 0.5114593856578805,
"calibration/confidence_entropy": 0.49050545259714856,
"calibration/coverage@0%": 0.0643979057591623,
"calibration/coverage@1%": 0.0643979057591623,
"calibration/coverage@10%": 0.5205018787496098,
"calibration/coverage@15%": 0.5777854982460253,
"calibration/coverage@20%": 0.6465968586387435,
"calibration/coverage@25%": 0.6732984293193718,
"calibration/coverage@30%": 0.7230366492146596,
"calibration/coverage@5%": 0.30811697924868053,
"calibration/distribution_entropy_10": 0.6749536064273652,
"calibration/distribution_entropy_100": 0.3733026957392366,
"calibration/ece": 0.11309064965504387,
"calibration/mean_confidence": 0.6855380969615151,
"calibration/unique_confidence_per_question": 0.024479166666666666,
"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": 0.004427083333333348,
"completions/max_length": 3676.2,
"completions/max_terminated_length": 3676.2,
"completions/mean_length": 689.5286499023438,
"completions/mean_terminated_length": 692.6000366210938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 202.4,
"epoch": 0.6971153846153846,
"grad_norm": 0.0005578218260779977,
"learning_rate": 4.507211538461539e-06,
"loss": -0.0039,
"num_tokens": 599223810.0,
"reward": 1.2588875055313111,
"reward_std": 0.11392218917608261,
"rewards/accuracy_reward": 0.6797742962837219,
"rewards/brier_reward": 0.8425005674362183,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9954861164093017,
"rewards/mean_confidence_reward": 0.7026808500289917,
"sampling/batch_mean_priority_error": 0.020199652777777757,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5555555555555556,
"sampling/error_ema_max": 0.07812499850988389,
"sampling/error_ema_mean": 0.0009238900383934379,
"sampling/priority_kl": 0.030000124126672745,
"sampling/priority_scale": 0.7441923797829076,
"sampling/prob_entropy": 10.278951454162598,
"sampling/prob_max": 4.216997185721993e-05,
"sampling/prob_min": 1.7492775805294512e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.6911999940872192,
"sampling/prompt_draws_total": 20736.0,
"sampling/seen_fraction": 0.5508933305740357,
"sampling/unseen_fraction": 0.44910666942596433,
"signal/accuracy_reward/centered_abs_mean": 0.11195204108953476,
"signal/accuracy_reward/group_std_mean": 0.1539290338754654,
"signal/accuracy_reward/group_zero_std_frac": 0.5361111164093018,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05597602054476738,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05597602054476738,
"signal/advantage_abs_mean": 0.08156270831823349,
"signal/advantage_pre_scale_abs_mean": 0.08156270831823349,
"signal/advantage_pre_scale_std": 0.16118119955062865,
"signal/advantage_std": 0.16118119955062865,
"signal/brier_reward/centered_abs_mean": 0.07383809015154838,
"signal/brier_reward/group_std_mean": 0.09918788969516754,
"signal/brier_reward/group_zero_std_frac": 0.17777777910232545,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03691904507577419,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03691904507577419,
"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": 0.00814887140877545,
"signal/format_reward/group_std_mean": 0.01690137628465891,
"signal/format_reward/group_zero_std_frac": 0.925000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004074435704387725,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004074435704387725,
"signal/mean_confidence_reward/centered_abs_mean": 0.06701208576560021,
"signal/mean_confidence_reward/group_std_mean": 0.08645372539758682,
"signal/mean_confidence_reward/group_zero_std_frac": 0.19444444179534912,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.70120812173991e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.70120812173991e-07,
"step": 290
},
{
"calibration/aurc": 0.1382071528975812,
"calibration/batch_distribution_entropy": 0.6787802551891892,
"calibration/batch_entropy_100bins": 0.36278431485319834,
"calibration/batch_entropy_10bins": 0.6787802551891892,
"calibration/batch_entropy_50bins": 0.427063825644149,
"calibration/batch_uniqueness": 0.5060428668027408,
"calibration/confidence_entropy": 0.4827378502143943,
"calibration/coverage@0%": 0.12598425196850394,
"calibration/coverage@1%": 0.2199105971128609,
"calibration/coverage@10%": 0.5906996629131781,
"calibration/coverage@15%": 0.6597831464379947,
"calibration/coverage@20%": 0.7319604771328057,
"calibration/coverage@25%": 0.7403693931398416,
"calibration/coverage@30%": 0.7403693931398416,
"calibration/coverage@5%": 0.5577887377163278,
"calibration/distribution_entropy_10": 0.6787802551891892,
"calibration/distribution_entropy_100": 0.36278431485319834,
"calibration/ece": 0.1557866114345061,
"calibration/mean_confidence": 0.7255701798802711,
"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": 0.003385416666666674,
"completions/max_length": 3645.8,
"completions/max_terminated_length": 3645.8,
"completions/mean_length": 704.8776977539062,
"completions/mean_terminated_length": 707.2697143554688,
"completions/min_length": 0.0,
"completions/min_terminated_length": 209.2,
"epoch": 0.7091346153846154,
"grad_norm": 0.000640033686067909,
"learning_rate": 4.477163461538462e-06,
"loss": -0.0014,
"num_tokens": 610431169.0,
"reward": 1.2582273960113526,
"reward_std": 0.12031746208667755,
"rewards/accuracy_reward": 0.6776909708976746,
"rewards/brier_reward": 0.8424825191497802,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9962673544883728,
"rewards/mean_confidence_reward": 0.6952007293701172,
"sampling/batch_mean_priority_error": 0.018067708333333325,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5194444444444445,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0009439987828955054,
"sampling/priority_kl": 0.030000235140323638,
"sampling/priority_scale": 0.7413066328270361,
"sampling/prob_entropy": 10.278957557678222,
"sampling/prob_max": 4.2287943506380544e-05,
"sampling/prob_min": 1.7601668878342026e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7032000064849854,
"sampling/prompt_draws_total": 21096.0,
"sampling/seen_fraction": 0.5573600053787231,
"sampling/unseen_fraction": 0.44263999462127684,
"signal/accuracy_reward/centered_abs_mean": 0.12797851413488387,
"signal/accuracy_reward/group_std_mean": 0.16741083860397338,
"signal/accuracy_reward/group_zero_std_frac": 0.5333333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06398925706744193,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06398925706744193,
"signal/advantage_abs_mean": 0.0889437347650528,
"signal/advantage_pre_scale_abs_mean": 0.0889437347650528,
"signal/advantage_pre_scale_std": 0.16828711926937104,
"signal/advantage_std": 0.16828711926937104,
"signal/brier_reward/centered_abs_mean": 0.07769880592823028,
"signal/brier_reward/group_std_mean": 0.10278659611940384,
"signal/brier_reward/group_zero_std_frac": 0.16388889104127885,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03884940296411514,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03884940296411514,
"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": 0.00691731758415699,
"signal/format_reward/group_std_mean": 0.01580982506275177,
"signal/format_reward/group_zero_std_frac": 0.9250000238418579,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003458658792078495,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003458658792078495,
"signal/mean_confidence_reward/centered_abs_mean": 0.06594567447900772,
"signal/mean_confidence_reward/group_std_mean": 0.08560345023870468,
"signal/mean_confidence_reward/group_zero_std_frac": 0.16944444328546523,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.594567025786091e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.594567025786091e-07,
"step": 295
},
{
"calibration/aurc": 0.1393076428472104,
"calibration/batch_distribution_entropy": 0.6548606113464862,
"calibration/batch_entropy_100bins": 0.3476336789716639,
"calibration/batch_entropy_10bins": 0.6548606113464862,
"calibration/batch_entropy_50bins": 0.40922874221964145,
"calibration/batch_uniqueness": 0.46888306010664504,
"calibration/confidence_entropy": 0.4854339588086577,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.2814272469458988,
"calibration/coverage@15%": 0.6440798371911454,
"calibration/coverage@20%": 0.8484375,
"calibration/coverage@25%": 0.8744791666666668,
"calibration/coverage@30%": 0.9484375,
"calibration/coverage@5%": 0.1928855802792321,
"calibration/distribution_entropy_10": 0.6548606113464862,
"calibration/distribution_entropy_100": 0.3476336789716639,
"calibration/ece": 0.09438192770046852,
"calibration/mean_confidence": 0.737763256923395,
"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": 0.002951388888888884,
"completions/max_length": 3239.0,
"completions/max_terminated_length": 3239.0,
"completions/mean_length": 714.7509765625,
"completions/mean_terminated_length": 716.9332641601562,
"completions/min_length": 0.0,
"completions/min_terminated_length": 182.8,
"epoch": 0.7211538461538461,
"grad_norm": 0.0006034580292180181,
"learning_rate": 4.447115384615385e-06,
"loss": -0.0022,
"num_tokens": 621738700.0,
"reward": 1.275432014465332,
"reward_std": 0.11760217696428299,
"rewards/accuracy_reward": 0.7190972208976746,
"rewards/brier_reward": 0.8347038507461548,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9970486044883728,
"rewards/mean_confidence_reward": 0.7160360455513001,
"sampling/batch_mean_priority_error": 0.02022395833333332,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5833333333333333,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.000968804070726037,
"sampling/priority_kl": 0.02999955527484417,
"sampling/priority_scale": 0.7386725247139111,
"sampling/prob_entropy": 10.278958892822265,
"sampling/prob_max": 4.241131755406968e-05,
"sampling/prob_min": 1.7708083396428266e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7152000069618225,
"sampling/prompt_draws_total": 21456.0,
"sampling/seen_fraction": 0.5639866590499878,
"sampling/unseen_fraction": 0.4360133409500122,
"signal/accuracy_reward/centered_abs_mean": 0.11738281100988388,
"signal/accuracy_reward/group_std_mean": 0.1607626795768738,
"signal/accuracy_reward/group_zero_std_frac": 0.5222222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05869140550494194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05869140550494194,
"signal/advantage_abs_mean": 0.08313068300485611,
"signal/advantage_pre_scale_abs_mean": 0.08313068300485611,
"signal/advantage_pre_scale_std": 0.16150366961956025,
"signal/advantage_std": 0.16150366961956025,
"signal/brier_reward/centered_abs_mean": 0.07401107624173164,
"signal/brier_reward/group_std_mean": 0.09984385818243027,
"signal/brier_reward/group_zero_std_frac": 0.15277777910232543,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03700553812086582,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03700553812086582,
"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": 0.005425347259733826,
"signal/format_reward/group_std_mean": 0.012002479285001755,
"signal/format_reward/group_zero_std_frac": 0.944444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002712673629866913,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002712673629866913,
"signal/mean_confidence_reward/centered_abs_mean": 0.06404319405555725,
"signal/mean_confidence_reward/group_std_mean": 0.08435939252376556,
"signal/mean_confidence_reward/group_zero_std_frac": 0.17222222238779067,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.404319151442906e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.404319151442906e-07,
"step": 300
},
{
"epoch": 0.7211538461538461,
"eval_calibration/aurc": 0.17038049269262773,
"eval_calibration/batch_distribution_entropy": 0.6199671776505913,
"eval_calibration/batch_entropy_100bins": 0.3268388624035065,
"eval_calibration/batch_entropy_10bins": 0.6199671776505913,
"eval_calibration/batch_entropy_50bins": 0.3847494206129197,
"eval_calibration/batch_uniqueness": 0.3539211074677266,
"eval_calibration/confidence_entropy": 0.45050842980070266,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.0,
"eval_calibration/coverage@15%": 0.516100957354221,
"eval_calibration/coverage@20%": 0.7728459530026109,
"eval_calibration/coverage@25%": 0.8677110530896431,
"eval_calibration/coverage@30%": 0.9843342036553525,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.6199671776505913,
"eval_calibration/distribution_entropy_100": 0.3268388624035065,
"eval_calibration/ece": 0.07167101827676242,
"eval_calibration/mean_confidence": 0.7650565709312447,
"eval_calibration/unique_confidence_per_question": 0.009548611111111112,
"eval_calibration/unique_confidences": 11,
"eval_completions/clipped_ratio": 0.0026041666666666665,
"eval_completions/max_length": 2238.3333333333335,
"eval_completions/max_terminated_length": 2238.3333333333335,
"eval_completions/mean_length": 717.2819620768229,
"eval_completions/mean_terminated_length": 719.1739807128906,
"eval_completions/min_length": 183.33333333333334,
"eval_completions/min_terminated_length": 260.8333333333333,
"eval_loss": 0.0,
"eval_num_tokens": 621738700.0,
"eval_reward": 1.260056495666504,
"eval_reward_std": 0.3358425050973892,
"eval_rewards/accuracy_reward": 0.6927083333333334,
"eval_rewards/brier_reward": 0.8299935062726339,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9973958333333334,
"eval_rewards/mean_confidence_reward": 0.7630642155806223,
"eval_runtime": 145.5855,
"eval_samples_per_second": 6.869,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4129774272441864,
"eval_signal/accuracy_reward/group_std_mean": 0.46044932802518207,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.2064887136220932,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.2064887136220932,
"eval_signal/advantage_abs_mean": 0.29073385894298553,
"eval_signal/advantage_pre_scale_abs_mean": 0.29073385894298553,
"eval_signal/advantage_pre_scale_std": 0.3327233244975408,
"eval_signal/advantage_std": 0.3327233244975408,
"eval_signal/brier_reward/centered_abs_mean": 0.1913610895474752,
"eval_signal/brier_reward/group_std_mean": 0.2476587469379107,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0956805447737376,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.0956805447737376,
"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": 0.9166666766007742,
"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": 0.1503743181626002,
"eval_signal/mean_confidence_reward/group_std_mean": 0.19400002310673395,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.5037432300838798e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.5037432300838798e-06,
"eval_steps_per_second": 0.041,
"step": 300
},
{
"epoch": 0.7211538461538461,
"step": 300,
"train_probe_calibration/aurc": 0.13103238168185818,
"train_probe_calibration/batch_distribution_entropy": 0.6086489322342906,
"train_probe_calibration/batch_entropy_100bins": 0.32399377529298606,
"train_probe_calibration/batch_entropy_10bins": 0.6086489322342906,
"train_probe_calibration/batch_entropy_50bins": 0.38140023009953883,
"train_probe_calibration/batch_uniqueness": 0.3492309125655866,
"train_probe_calibration/confidence_entropy": 0.4497829731002935,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.515230635335074,
"train_probe_calibration/coverage@15%": 0.7737162750217581,
"train_probe_calibration/coverage@20%": 0.8598781549173194,
"train_probe_calibration/coverage@25%": 0.9512619669277633,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.6086489322342906,
"train_probe_calibration/distribution_entropy_100": 0.32399377529298606,
"train_probe_calibration/ece": 0.038424717145343686,
"train_probe_calibration/mean_confidence": 0.7651436031331593,
"train_probe_calibration/unique_confidence_per_question": 0.010416666666666666,
"train_probe_calibration/unique_confidences": 12,
"train_probe_completions/clipped_ratio": 0.001736111111111105,
"train_probe_completions/max_length": 2880.1666666666665,
"train_probe_completions/max_terminated_length": 2880.1666666666665,
"train_probe_completions/mean_length": 747.1415100097656,
"train_probe_completions/mean_terminated_length": 748.4186096191406,
"train_probe_completions/min_length": 203.83333333333334,
"train_probe_completions/min_terminated_length": 243.0,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 621738700.0,
"train_probe_reward": 1.2938086589177449,
"train_probe_reward_std": 0.31472263236840564,
"train_probe_rewards/accuracy_reward": 0.7352430522441864,
"train_probe_rewards/brier_reward": 0.8549631237983704,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9973958333333334,
"train_probe_rewards/mean_confidence_reward": 0.7631510297457377,
"train_probe_runtime": 165.7718,
"train_probe_samples_per_second": 6.032,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3800455729166667,
"train_probe_signal/accuracy_reward/group_std_mean": 0.44128966828187305,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19002278645833334,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.19002278645833334,
"train_probe_signal/advantage_abs_mean": 0.26244449118773144,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.26244449118773144,
"train_probe_signal/advantage_pre_scale_std": 0.3124504089355469,
"train_probe_signal/advantage_std": 0.3124504089355469,
"train_probe_signal/brier_reward/centered_abs_mean": 0.1670305853088697,
"train_probe_signal/brier_reward/group_std_mean": 0.22494085133075714,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08351529265443484,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.08351529265443484,
"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": 0.9166666766007742,
"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": 0.15324705094099045,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.1951539988319079,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.5324703970994353e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.5324703970994353e-06,
"train_probe_steps_per_second": 0.036
},
{
"calibration/aurc": 0.13092579680447874,
"calibration/batch_distribution_entropy": 0.6276670537021948,
"calibration/batch_entropy_100bins": 0.3316895599320206,
"calibration/batch_entropy_10bins": 0.6276670537021948,
"calibration/batch_entropy_50bins": 0.3904595832598581,
"calibration/batch_uniqueness": 0.38209639015063,
"calibration/confidence_entropy": 0.4591777053056802,
"calibration/coverage@0%": 0.0010526315789473684,
"calibration/coverage@1%": 0.0010526315789473684,
"calibration/coverage@10%": 0.41942434210526314,
"calibration/coverage@15%": 0.5662409246026293,
"calibration/coverage@20%": 0.7830812859691332,
"calibration/coverage@25%": 0.9054183811099111,
"calibration/coverage@30%": 0.9398232984293194,
"calibration/coverage@5%": 0.3413212719298246,
"calibration/distribution_entropy_10": 0.6276670537021948,
"calibration/distribution_entropy_100": 0.3316895599320206,
"calibration/ece": 0.08329600836702859,
"calibration/mean_confidence": 0.749807476925117,
"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": 0.005381944444444442,
"completions/max_length": 3670.6,
"completions/max_terminated_length": 3670.6,
"completions/mean_length": 771.6765747070312,
"completions/mean_terminated_length": 775.8438354492188,
"completions/min_length": 0.0,
"completions/min_terminated_length": 222.0,
"epoch": 0.7331730769230769,
"grad_norm": 0.0007620741962455213,
"learning_rate": 4.4170673076923085e-06,
"loss": -0.0037,
"num_tokens": 633738398.0,
"reward": 1.2771973609924316,
"reward_std": 0.12964553087949754,
"rewards/accuracy_reward": 0.7176215171813964,
"rewards/brier_reward": 0.8421401739120483,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9946180582046509,
"rewards/mean_confidence_reward": 0.7522072315216064,
"sampling/batch_mean_priority_error": 0.013479166666666658,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5555555555555556,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0009865233907476068,
"sampling/priority_kl": 0.029999896138906478,
"sampling/priority_scale": 0.736354249692522,
"sampling/prob_entropy": 10.278947830200195,
"sampling/prob_max": 4.253947699908167e-05,
"sampling/prob_min": 1.781034661689773e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7271999955177307,
"sampling/prompt_draws_total": 21816.0,
"sampling/seen_fraction": 0.570739996433258,
"sampling/unseen_fraction": 0.42926000356674193,
"signal/accuracy_reward/centered_abs_mean": 0.12093641608953476,
"signal/accuracy_reward/group_std_mean": 0.162637859582901,
"signal/accuracy_reward/group_zero_std_frac": 0.522222226858139,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06046820804476738,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06046820804476738,
"signal/advantage_abs_mean": 0.09479492604732513,
"signal/advantage_pre_scale_abs_mean": 0.09479492604732513,
"signal/advantage_pre_scale_std": 0.18216970562934875,
"signal/advantage_std": 0.18216970562934875,
"signal/brier_reward/centered_abs_mean": 0.0808374285697937,
"signal/brier_reward/group_std_mean": 0.10848785042762757,
"signal/brier_reward/group_zero_std_frac": 0.17500000298023224,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04041871428489685,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04041871428489685,
"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": 0.007291666604578495,
"signal/format_reward/group_std_mean": 0.016734511591494083,
"signal/format_reward/group_zero_std_frac": 0.919444465637207,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0036458333022892477,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0036458333022892477,
"signal/mean_confidence_reward/centered_abs_mean": 0.06553446054458618,
"signal/mean_confidence_reward/group_std_mean": 0.08630905747413635,
"signal/mean_confidence_reward/group_zero_std_frac": 0.18611111342906952,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.553445928147994e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.553445928147994e-07,
"step": 305
},
{
"calibration/aurc": 0.13837794223912891,
"calibration/batch_distribution_entropy": 0.7199264324569175,
"calibration/batch_entropy_100bins": 0.3920106756730617,
"calibration/batch_entropy_10bins": 0.7199264324569175,
"calibration/batch_entropy_50bins": 0.46146862472273603,
"calibration/batch_uniqueness": 0.5973591612208066,
"calibration/confidence_entropy": 0.5280266050339674,
"calibration/coverage@0%": 0.1471692683938273,
"calibration/coverage@1%": 0.1471692683938273,
"calibration/coverage@10%": 0.501302979724963,
"calibration/coverage@15%": 0.5405839022139196,
"calibration/coverage@20%": 0.6710906688447321,
"calibration/coverage@25%": 0.9076763769086709,
"calibration/coverage@30%": 0.95,
"calibration/coverage@5%": 0.4018550029124053,
"calibration/distribution_entropy_10": 0.7199264324569175,
"calibration/distribution_entropy_100": 0.3920106756730617,
"calibration/ece": 0.14846240057694993,
"calibration/mean_confidence": 0.687106785034867,
"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": 0.00399305555555558,
"completions/max_length": 3469.8,
"completions/max_terminated_length": 3469.8,
"completions/mean_length": 777.2681518554688,
"completions/mean_terminated_length": 780.4172485351562,
"completions/min_length": 47.4,
"completions/min_terminated_length": 246.8,
"epoch": 0.7451923076923077,
"grad_norm": 0.0005892643239349127,
"learning_rate": 4.3870192307692315e-06,
"loss": -0.003,
"num_tokens": 645810607.0,
"reward": 1.279511857032776,
"reward_std": 0.11136721819639206,
"rewards/accuracy_reward": 0.7183159708976745,
"rewards/brier_reward": 0.8445999145507812,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.99609375,
"rewards/mean_confidence_reward": 0.7028311610221862,
"sampling/batch_mean_priority_error": 0.01767881944444443,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5083333333333333,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.001006615860387683,
"sampling/priority_kl": 0.030000757426023483,
"sampling/priority_scale": 0.7339580834144727,
"sampling/prob_entropy": 10.278960227966309,
"sampling/prob_max": 4.2663155909394845e-05,
"sampling/prob_min": 1.7912808471010066e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7391999959945679,
"sampling/prompt_draws_total": 22176.0,
"sampling/seen_fraction": 0.5771666646003724,
"sampling/unseen_fraction": 0.4228333353996277,
"signal/accuracy_reward/centered_abs_mean": 0.114306640625,
"signal/accuracy_reward/group_std_mean": 0.1513543099164963,
"signal/accuracy_reward/group_zero_std_frac": 0.5694444417953491,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0571533203125,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0571533203125,
"signal/advantage_abs_mean": 0.08048739284276962,
"signal/advantage_pre_scale_abs_mean": 0.08048739284276962,
"signal/advantage_pre_scale_std": 0.1592124491930008,
"signal/advantage_std": 0.1592124491930008,
"signal/brier_reward/centered_abs_mean": 0.07133363783359528,
"signal/brier_reward/group_std_mean": 0.09470874518156051,
"signal/brier_reward/group_zero_std_frac": 0.1166666679084301,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03566681891679764,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03566681891679764,
"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": 0.01631898358464241,
"signal/format_reward/group_zero_std_frac": 0.925000011920929,
"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": 0.06672917604446411,
"signal/mean_confidence_reward/group_std_mean": 0.08671197891235352,
"signal/mean_confidence_reward/group_zero_std_frac": 0.12222222313284874,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.672917265859723e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.672917265859723e-07,
"step": 310
},
{
"calibration/aurc": 0.13686577318136903,
"calibration/batch_distribution_entropy": 0.6244593056933125,
"calibration/batch_entropy_100bins": 0.34029339548266907,
"calibration/batch_entropy_10bins": 0.6244593056933125,
"calibration/batch_entropy_50bins": 0.40058787926118855,
"calibration/batch_uniqueness": 0.40405619414598126,
"calibration/confidence_entropy": 0.47154955375434,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5093171296296296,
"calibration/coverage@15%": 0.5838955026455026,
"calibration/coverage@20%": 0.7709579672454998,
"calibration/coverage@25%": 0.8654971714534379,
"calibration/coverage@30%": 0.9467798085291557,
"calibration/coverage@5%": 0.19791666666666666,
"calibration/distribution_entropy_10": 0.6244593056933125,
"calibration/distribution_entropy_100": 0.34029339548266907,
"calibration/ece": 0.11483152296855784,
"calibration/mean_confidence": 0.7404223662829652,
"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": 0.004253472222222232,
"completions/max_length": 3368.0,
"completions/max_terminated_length": 3368.0,
"completions/mean_length": 790.6902954101563,
"completions/mean_terminated_length": 794.0724731445313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 238.2,
"epoch": 0.7572115384615384,
"grad_norm": 0.0007672872743569314,
"learning_rate": 4.356971153846154e-06,
"loss": -0.0027,
"num_tokens": 658014783.0,
"reward": 1.290112805366516,
"reward_std": 0.12518732845783234,
"rewards/accuracy_reward": 0.7293402910232544,
"rewards/brier_reward": 0.8555580139160156,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9953125,
"rewards/mean_confidence_reward": 0.7429038882255554,
"sampling/batch_mean_priority_error": 0.014930555555555548,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5194444444444444,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0010217264294624328,
"sampling/priority_kl": 0.029999467357993127,
"sampling/priority_scale": 0.7315961777931079,
"sampling/prob_entropy": 10.278939628601075,
"sampling/prob_max": 4.2783647950273006e-05,
"sampling/prob_min": 1.8013632143265568e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.751199996471405,
"sampling/prompt_draws_total": 22536.0,
"sampling/seen_fraction": 0.5833466529846192,
"sampling/unseen_fraction": 0.41665334701538087,
"signal/accuracy_reward/centered_abs_mean": 0.11973741352558136,
"signal/accuracy_reward/group_std_mean": 0.16205187290906906,
"signal/accuracy_reward/group_zero_std_frac": 0.5250000119209289,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05986870676279068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05986870676279068,
"signal/advantage_abs_mean": 0.09079453721642494,
"signal/advantage_pre_scale_abs_mean": 0.09079453721642494,
"signal/advantage_pre_scale_std": 0.17762331664562225,
"signal/advantage_std": 0.17762331664562225,
"signal/brier_reward/centered_abs_mean": 0.0763278879225254,
"signal/brier_reward/group_std_mean": 0.10317817330360413,
"signal/brier_reward/group_zero_std_frac": 0.13611111082136632,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0381639439612627,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0381639439612627,
"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": 0.00849609375,
"signal/format_reward/group_std_mean": 0.0171542102470994,
"signal/format_reward/group_zero_std_frac": 0.9277777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004248046875,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004248046875,
"signal/mean_confidence_reward/centered_abs_mean": 0.0643157720565796,
"signal/mean_confidence_reward/group_std_mean": 0.08580757826566696,
"signal/mean_confidence_reward/group_zero_std_frac": 0.15277777947485446,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.431576821341878e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.431576821341878e-07,
"step": 315
},
{
"calibration/aurc": 0.16717747584400544,
"calibration/batch_distribution_entropy": 0.602479901250202,
"calibration/batch_entropy_100bins": 0.324314097132174,
"calibration/batch_entropy_10bins": 0.602479901250202,
"calibration/batch_entropy_50bins": 0.38177730778586694,
"calibration/batch_uniqueness": 0.32554403739572646,
"calibration/confidence_entropy": 0.4465209078155052,
"calibration/coverage@0%": 0.0016101816101816101,
"calibration/coverage@1%": 0.0016101816101816101,
"calibration/coverage@10%": 0.2815870954907109,
"calibration/coverage@15%": 0.46984229412824813,
"calibration/coverage@20%": 0.6756455747056269,
"calibration/coverage@25%": 0.8430287430287431,
"calibration/coverage@30%": 0.90006292006292,
"calibration/coverage@5%": 0.1304059931285062,
"calibration/distribution_entropy_10": 0.602479901250202,
"calibration/distribution_entropy_100": 0.324314097132174,
"calibration/ece": 0.11514510673588034,
"calibration/mean_confidence": 0.7590287926881031,
"calibration/unique_confidence_per_question": 0.027083333333333338,
"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": 0.011284722222222232,
"completions/max_length": 3799.6,
"completions/max_terminated_length": 3799.6,
"completions/mean_length": 851.075341796875,
"completions/mean_terminated_length": 861.0564086914062,
"completions/min_length": 0.0,
"completions/min_terminated_length": 247.4,
"epoch": 0.7692307692307693,
"grad_norm": 0.0007334538386203349,
"learning_rate": 4.326923076923077e-06,
"loss": -0.0122,
"num_tokens": 670893411.0,
"reward": 1.2725468873977661,
"reward_std": 0.145135360956192,
"rewards/accuracy_reward": 0.72265625,
"rewards/brier_reward": 0.834314513206482,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9881076335906982,
"rewards/mean_confidence_reward": 0.7703335165977478,
"sampling/batch_mean_priority_error": 0.019050540123456784,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.49444444444444446,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0010432714829221369,
"sampling/priority_kl": 0.030000781640410423,
"sampling/priority_scale": 0.7293619335396215,
"sampling/prob_entropy": 10.278955459594727,
"sampling/prob_max": 4.290501019568183e-05,
"sampling/prob_min": 1.811251713661477e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7632000088691712,
"sampling/prompt_draws_total": 22896.0,
"sampling/seen_fraction": 0.5894733190536499,
"sampling/unseen_fraction": 0.4105266809463501,
"signal/accuracy_reward/centered_abs_mean": 0.13217773288488388,
"signal/accuracy_reward/group_std_mean": 0.17295660972595214,
"signal/accuracy_reward/group_zero_std_frac": 0.5166666626930236,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06608886644244194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06608886644244194,
"signal/advantage_abs_mean": 0.1076438531279564,
"signal/advantage_pre_scale_abs_mean": 0.1076438531279564,
"signal/advantage_pre_scale_std": 0.20443538427352906,
"signal/advantage_std": 0.20443538427352906,
"signal/brier_reward/centered_abs_mean": 0.09236456230282783,
"signal/brier_reward/group_std_mean": 0.12330232709646224,
"signal/brier_reward/group_zero_std_frac": 0.225,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.046182281151413915,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.046182281151413915,
"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": 0.018429904524236918,
"signal/format_reward/group_std_mean": 0.03126038983464241,
"signal/format_reward/group_zero_std_frac": 0.8805555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009214952262118459,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009214952262118459,
"signal/mean_confidence_reward/centered_abs_mean": 0.06820750385522842,
"signal/mean_confidence_reward/group_std_mean": 0.09028767794370651,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2500000029802322,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.820750058977865e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.820750058977865e-07,
"step": 320
},
{
"calibration/aurc": 0.10863199741676803,
"calibration/batch_distribution_entropy": 0.6347477303485826,
"calibration/batch_entropy_100bins": 0.34220862177900163,
"calibration/batch_entropy_10bins": 0.6347477303485826,
"calibration/batch_entropy_50bins": 0.40284245266912944,
"calibration/batch_uniqueness": 0.40180875509585673,
"calibration/confidence_entropy": 0.46986021781998544,
"calibration/coverage@0%": 0.061942257217847775,
"calibration/coverage@1%": 0.061942257217847775,
"calibration/coverage@10%": 0.477210801653874,
"calibration/coverage@15%": 0.7232488607091858,
"calibration/coverage@20%": 0.8684191024109602,
"calibration/coverage@25%": 0.9326315789473684,
"calibration/coverage@30%": 0.9547368421052631,
"calibration/coverage@5%": 0.309322237655811,
"calibration/distribution_entropy_10": 0.6347477303485826,
"calibration/distribution_entropy_100": 0.34220862177900163,
"calibration/ece": 0.1094783425165194,
"calibration/mean_confidence": 0.7294889187832778,
"calibration/unique_confidence_per_question": 0.0265625,
"calibration/unique_confidences": 10.2,
"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": 0.011024305555555558,
"completions/max_length": 3871.2,
"completions/max_terminated_length": 3871.2,
"completions/mean_length": 931.85478515625,
"completions/mean_terminated_length": 942.4100463867187,
"completions/min_length": 0.0,
"completions/min_terminated_length": 252.6,
"epoch": 0.78125,
"grad_norm": 0.000653931638225913,
"learning_rate": 4.296875e-06,
"loss": -0.0099,
"num_tokens": 684696186.0,
"reward": 1.258277130126953,
"reward_std": 0.15136132538318633,
"rewards/accuracy_reward": 0.703906238079071,
"rewards/brier_reward": 0.8240050315856934,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9886284589767456,
"rewards/mean_confidence_reward": 0.7231320858001709,
"sampling/batch_mean_priority_error": 0.01323225308641975,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5083333333333333,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0010600588750094176,
"sampling/priority_kl": 0.0299997728317976,
"sampling/priority_scale": 0.7272611201042309,
"sampling/prob_entropy": 10.27895393371582,
"sampling/prob_max": 4.3025674676755445e-05,
"sampling/prob_min": 1.8203360014013015e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7751999974250794,
"sampling/prompt_draws_total": 23256.0,
"sampling/seen_fraction": 0.5954999923706055,
"sampling/unseen_fraction": 0.40450000762939453,
"signal/accuracy_reward/centered_abs_mean": 0.14385308176279069,
"signal/accuracy_reward/group_std_mean": 0.19072539806365968,
"signal/accuracy_reward/group_zero_std_frac": 0.450000011920929,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07192654088139534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07192654088139534,
"signal/advantage_abs_mean": 0.10959743559360505,
"signal/advantage_pre_scale_abs_mean": 0.10959743559360505,
"signal/advantage_pre_scale_std": 0.2033509284257889,
"signal/advantage_std": 0.2033509284257889,
"signal/brier_reward/centered_abs_mean": 0.09024559706449509,
"signal/brier_reward/group_std_mean": 0.12317982912063599,
"signal/brier_reward/group_zero_std_frac": 0.10833333600312471,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04512279853224754,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04512279853224754,
"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": 0.019015842117369175,
"signal/format_reward/group_std_mean": 0.03431776948273182,
"signal/format_reward/group_zero_std_frac": 0.8638888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009507921058684587,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009507921058684587,
"signal/mean_confidence_reward/centered_abs_mean": 0.07643213719129563,
"signal/mean_confidence_reward/group_std_mean": 0.10056560337543488,
"signal/mean_confidence_reward/group_zero_std_frac": 0.12222222611308098,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.643213734809251e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.643213734809251e-07,
"step": 325
},
{
"calibration/aurc": 0.11886209120977206,
"calibration/batch_distribution_entropy": 0.7440964345869737,
"calibration/batch_entropy_100bins": 0.40832423826981046,
"calibration/batch_entropy_10bins": 0.7440964345869737,
"calibration/batch_entropy_50bins": 0.4806726866603972,
"calibration/batch_uniqueness": 0.5958464567420566,
"calibration/confidence_entropy": 0.5508633686135129,
"calibration/coverage@0%": 0.13636579655279554,
"calibration/coverage@1%": 0.13636579655279554,
"calibration/coverage@10%": 0.3909505278628627,
"calibration/coverage@15%": 0.7179516933647312,
"calibration/coverage@20%": 0.8417117793387895,
"calibration/coverage@25%": 0.9079508984453867,
"calibration/coverage@30%": 0.9534564102564103,
"calibration/coverage@5%": 0.26832501380613044,
"calibration/distribution_entropy_10": 0.7440964345869737,
"calibration/distribution_entropy_100": 0.40832423826981046,
"calibration/ece": 0.14901436740330448,
"calibration/mean_confidence": 0.633296158815817,
"calibration/unique_confidence_per_question": 0.02760416666666667,
"calibration/unique_confidences": 10.6,
"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": 0.014236111111111116,
"completions/max_length": 3919.2,
"completions/max_terminated_length": 3919.2,
"completions/mean_length": 923.294970703125,
"completions/mean_terminated_length": 936.7350952148438,
"completions/min_length": 0.0,
"completions/min_terminated_length": 255.0,
"epoch": 0.7932692307692307,
"grad_norm": 0.0006061934400349855,
"learning_rate": 4.266826923076923e-06,
"loss": -0.0167,
"num_tokens": 698460000.0,
"reward": 1.250968337059021,
"reward_std": 0.12344784587621689,
"rewards/accuracy_reward": 0.6842013955116272,
"rewards/brier_reward": 0.8321323394775391,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9855902791023254,
"rewards/mean_confidence_reward": 0.6281623840332031,
"sampling/batch_mean_priority_error": 0.020135886873267825,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.49444444444444435,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.001079965545795858,
"sampling/priority_kl": 0.029999606683850287,
"sampling/priority_scale": 0.7252804696327075,
"sampling/prob_entropy": 10.278935813903809,
"sampling/prob_max": 4.3146366806467995e-05,
"sampling/prob_min": 1.8293721950612962e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7871999979019165,
"sampling/prompt_draws_total": 23616.0,
"sampling/seen_fraction": 0.601419997215271,
"sampling/unseen_fraction": 0.398580002784729,
"signal/accuracy_reward/centered_abs_mean": 0.11248915046453475,
"signal/accuracy_reward/group_std_mean": 0.15042957961559295,
"signal/accuracy_reward/group_zero_std_frac": 0.5555555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05624457523226738,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05624457523226738,
"signal/advantage_abs_mean": 0.08699684739112853,
"signal/advantage_pre_scale_abs_mean": 0.08699684739112853,
"signal/advantage_pre_scale_std": 0.1748094767332077,
"signal/advantage_std": 0.1748094767332077,
"signal/brier_reward/centered_abs_mean": 0.07624856233596802,
"signal/brier_reward/group_std_mean": 0.10298327952623368,
"signal/brier_reward/group_zero_std_frac": 0.06111111249774694,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03812428116798401,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03812428116798401,
"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": 0.02356770820915699,
"signal/format_reward/group_std_mean": 0.0421368557959795,
"signal/format_reward/group_zero_std_frac": 0.8333333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.011783854104578495,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.011783854104578495,
"signal/mean_confidence_reward/centered_abs_mean": 0.06578600704669953,
"signal/mean_confidence_reward/group_std_mean": 0.08733049482107162,
"signal/mean_confidence_reward/group_zero_std_frac": 0.06388889048248529,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.578600618922792e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.578600618922792e-07,
"step": 330
},
{
"calibration/aurc": 0.08823840005491149,
"calibration/batch_distribution_entropy": 0.6795859117610638,
"calibration/batch_entropy_100bins": 0.3769886039645168,
"calibration/batch_entropy_10bins": 0.6795859117610638,
"calibration/batch_entropy_50bins": 0.44378488496252083,
"calibration/batch_uniqueness": 0.5419092140454492,
"calibration/confidence_entropy": 0.5797361103491635,
"calibration/coverage@0%": 0.07404611229393553,
"calibration/coverage@1%": 0.32532828569161565,
"calibration/coverage@10%": 0.62353149111474,
"calibration/coverage@15%": 0.8074700707412464,
"calibration/coverage@20%": 0.8522986812087805,
"calibration/coverage@25%": 0.9249272262817085,
"calibration/coverage@30%": 0.9376344086021506,
"calibration/coverage@5%": 0.47406383457698664,
"calibration/distribution_entropy_10": 0.6795859117610638,
"calibration/distribution_entropy_100": 0.3769886039645168,
"calibration/ece": 0.1873793358632409,
"calibration/mean_confidence": 0.617761095095165,
"calibration/unique_confidence_per_question": 0.026041666666666664,
"calibration/unique_confidences": 10.0,
"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": 0.013107638888888884,
"completions/max_length": 3929.0,
"completions/max_terminated_length": 3929.0,
"completions/mean_length": 956.8102416992188,
"completions/mean_terminated_length": 969.6889038085938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 243.0,
"epoch": 0.8052884615384616,
"grad_norm": 0.0007015041192062199,
"learning_rate": 4.236778846153847e-06,
"loss": -0.0143,
"num_tokens": 712620406.0,
"reward": 1.2798907995223998,
"reward_std": 0.11801934242248535,
"rewards/accuracy_reward": 0.7401909708976746,
"rewards/brier_reward": 0.8328598380088806,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.986718761920929,
"rewards/mean_confidence_reward": 0.5951099038124085,
"sampling/batch_mean_priority_error": 0.025409692578360808,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.48888888888888893,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0011080709984526039,
"sampling/priority_kl": 0.029999785125255585,
"sampling/priority_scale": 0.7234982907539234,
"sampling/prob_entropy": 10.27892837524414,
"sampling/prob_max": 4.3270392779959366e-05,
"sampling/prob_min": 1.8385029761702754e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7991999983787537,
"sampling/prompt_draws_total": 23976.0,
"sampling/seen_fraction": 0.6074000000953674,
"sampling/unseen_fraction": 0.39259999990463257,
"signal/accuracy_reward/centered_abs_mean": 0.12392035871744156,
"signal/accuracy_reward/group_std_mean": 0.1636717587709427,
"signal/accuracy_reward/group_zero_std_frac": 0.5333333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06196017935872078,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06196017935872078,
"signal/advantage_abs_mean": 0.08390375822782517,
"signal/advantage_pre_scale_abs_mean": 0.08390375822782517,
"signal/advantage_pre_scale_std": 0.16726355850696564,
"signal/advantage_std": 0.16726355850696564,
"signal/brier_reward/centered_abs_mean": 0.07296821475028992,
"signal/brier_reward/group_std_mean": 0.09915484488010406,
"signal/brier_reward/group_zero_std_frac": 0.05555555708706379,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03648410737514496,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03648410737514496,
"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": 0.022119140625,
"signal/format_reward/group_std_mean": 0.04018164724111557,
"signal/format_reward/group_zero_std_frac": 0.8388889074325562,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0110595703125,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0110595703125,
"signal/mean_confidence_reward/centered_abs_mean": 0.06477264314889908,
"signal/mean_confidence_reward/group_std_mean": 0.08538633137941361,
"signal/mean_confidence_reward/group_zero_std_frac": 0.05555555708706379,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.477264264503902e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.477264264503902e-07,
"step": 335
},
{
"calibration/aurc": 0.13969130786823786,
"calibration/batch_distribution_entropy": 0.6532072460044838,
"calibration/batch_entropy_100bins": 0.35603898661564687,
"calibration/batch_entropy_10bins": 0.6532072460044838,
"calibration/batch_entropy_50bins": 0.4191233343814,
"calibration/batch_uniqueness": 0.4666210398887169,
"calibration/confidence_entropy": 0.5266185188445387,
"calibration/coverage@0%": 0.04281984334203655,
"calibration/coverage@1%": 0.12428198433420365,
"calibration/coverage@10%": 0.4448985862018387,
"calibration/coverage@15%": 0.5834735535647629,
"calibration/coverage@20%": 0.6554527202314298,
"calibration/coverage@25%": 0.7479446061963418,
"calibration/coverage@30%": 0.8989081746920494,
"calibration/coverage@5%": 0.39751933088268976,
"calibration/distribution_entropy_10": 0.6532072460044838,
"calibration/distribution_entropy_100": 0.35603898661564687,
"calibration/ece": 0.11965576739251063,
"calibration/mean_confidence": 0.7021747605627129,
"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": 0.011284722222222232,
"completions/max_length": 4009.8,
"completions/max_terminated_length": 4009.8,
"completions/mean_length": 956.4407104492187,
"completions/mean_terminated_length": 967.5976806640625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 290.6,
"epoch": 0.8173076923076923,
"grad_norm": 0.0009944693883880973,
"learning_rate": 4.20673076923077e-06,
"loss": -0.0111,
"num_tokens": 726726955.0,
"reward": 1.2707728147506714,
"reward_std": 0.12885977178812028,
"rewards/accuracy_reward": 0.7121527791023254,
"rewards/brier_reward": 0.8406639814376831,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9887152791023255,
"rewards/mean_confidence_reward": 0.6793183445930481,
"sampling/batch_mean_priority_error": 0.01789025548696844,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.48611111111111105,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0011320403078570961,
"sampling/priority_kl": 0.030000582709908484,
"sampling/priority_scale": 0.721624392340891,
"sampling/prob_entropy": 10.27894229888916,
"sampling/prob_max": 4.339007573435083e-05,
"sampling/prob_min": 1.8476794139132835e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.8111999988555908,
"sampling/prompt_draws_total": 24336.0,
"sampling/seen_fraction": 0.6130866646766663,
"sampling/unseen_fraction": 0.3869133353233337,
"signal/accuracy_reward/centered_abs_mean": 0.12188584953546525,
"signal/accuracy_reward/group_std_mean": 0.16384209394454957,
"signal/accuracy_reward/group_zero_std_frac": 0.5250000178813934,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06094292476773262,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06094292476773262,
"signal/advantage_abs_mean": 0.09127618968486786,
"signal/advantage_pre_scale_abs_mean": 0.09127618968486786,
"signal/advantage_pre_scale_std": 0.18212750256061555,
"signal/advantage_std": 0.18212750256061555,
"signal/brier_reward/centered_abs_mean": 0.07710054516792297,
"signal/brier_reward/group_std_mean": 0.10524273961782456,
"signal/brier_reward/group_zero_std_frac": 0.09722222462296486,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03855027258396149,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03855027258396149,
"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": 0.01925998292863369,
"signal/format_reward/group_std_mean": 0.03434903621673584,
"signal/format_reward/group_zero_std_frac": 0.8666666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009629991464316845,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009629991464316845,
"signal/mean_confidence_reward/centered_abs_mean": 0.06390919610857963,
"signal/mean_confidence_reward/group_std_mean": 0.08523715138435364,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1027777798473835,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.390919111254334e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.390919111254334e-07,
"step": 340
},
{
"calibration/aurc": 0.09080647825464169,
"calibration/batch_distribution_entropy": 0.5828487105108893,
"calibration/batch_entropy_100bins": 0.30621682881224266,
"calibration/batch_entropy_10bins": 0.5828487105108893,
"calibration/batch_entropy_50bins": 0.36047349633099196,
"calibration/batch_uniqueness": 0.32978425497608505,
"calibration/confidence_entropy": 0.4501530039067473,
"calibration/coverage@0%": 0.17561666048288466,
"calibration/coverage@1%": 0.28143676630299047,
"calibration/coverage@10%": 0.667472317927639,
"calibration/coverage@15%": 0.7943842026331502,
"calibration/coverage@20%": 0.8553810501040052,
"calibration/coverage@25%": 0.8659630606860158,
"calibration/coverage@30%": 0.9324538258575199,
"calibration/coverage@5%": 0.4009431340756205,
"calibration/distribution_entropy_10": 0.5828487105108893,
"calibration/distribution_entropy_100": 0.30621682881224266,
"calibration/ece": 0.13318603844096294,
"calibration/mean_confidence": 0.7800264505674386,
"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": 0.007812500000000023,
"completions/max_length": 3928.8,
"completions/max_terminated_length": 3928.8,
"completions/mean_length": 894.5711059570312,
"completions/mean_terminated_length": 901.781787109375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 221.2,
"epoch": 0.8293269230769231,
"grad_norm": 0.0008186151972040534,
"learning_rate": 4.176682692307693e-06,
"loss": -0.0084,
"num_tokens": 740097726.0,
"reward": 1.2895998001098632,
"reward_std": 0.1312810465693474,
"rewards/accuracy_reward": 0.7419270992279052,
"rewards/brier_reward": 0.8452433705329895,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9920138835906982,
"rewards/mean_confidence_reward": 0.7631656527519226,
"sampling/batch_mean_priority_error": 0.014798804012345674,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4916666666666667,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.001147703337483108,
"sampling/priority_kl": 0.03000086471438408,
"sampling/priority_scale": 0.7201250732643529,
"sampling/prob_entropy": 10.278956031799316,
"sampling/prob_max": 4.351890675025061e-05,
"sampling/prob_min": 1.8564553465694188e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.823200011253357,
"sampling/prompt_draws_total": 24696.0,
"sampling/seen_fraction": 0.6190933346748352,
"sampling/unseen_fraction": 0.3809066653251648,
"signal/accuracy_reward/centered_abs_mean": 0.1175509974360466,
"signal/accuracy_reward/group_std_mean": 0.15728657841682434,
"signal/accuracy_reward/group_zero_std_frac": 0.5388888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0587754987180233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0587754987180233,
"signal/advantage_abs_mean": 0.093961463868618,
"signal/advantage_pre_scale_abs_mean": 0.093961463868618,
"signal/advantage_pre_scale_std": 0.1893182247877121,
"signal/advantage_std": 0.1893182247877121,
"signal/brier_reward/centered_abs_mean": 0.07490430921316146,
"signal/brier_reward/group_std_mean": 0.10306931734085083,
"signal/brier_reward/group_zero_std_frac": 0.16944444477558135,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03745215460658073,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03745215460658073,
"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": 0.014811197947710752,
"signal/format_reward/group_std_mean": 0.03360258936882019,
"signal/format_reward/group_zero_std_frac": 0.8416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007405598973855376,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007405598973855376,
"signal/mean_confidence_reward/centered_abs_mean": 0.059064753353595734,
"signal/mean_confidence_reward/group_std_mean": 0.07968888133764267,
"signal/mean_confidence_reward/group_zero_std_frac": 0.17777778208255768,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.906475166739256e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.906475166739256e-07,
"step": 345
},
{
"calibration/aurc": 0.17821393399430568,
"calibration/batch_distribution_entropy": 0.5281294051861017,
"calibration/batch_entropy_100bins": 0.2789663542218751,
"calibration/batch_entropy_10bins": 0.5281294051861017,
"calibration/batch_entropy_50bins": 0.3283946785521962,
"calibration/batch_uniqueness": 0.2076128794208711,
"calibration/confidence_entropy": 0.4363416647275226,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.36962873167834004,
"calibration/coverage@15%": 0.5259358465608466,
"calibration/coverage@20%": 0.676443308413955,
"calibration/coverage@25%": 0.784228192926247,
"calibration/coverage@30%": 0.8718103195543829,
"calibration/coverage@5%": 0.13733681462140993,
"calibration/distribution_entropy_10": 0.5281294051861017,
"calibration/distribution_entropy_100": 0.2789663542218751,
"calibration/ece": 0.1380739637193096,
"calibration/mean_confidence": 0.7937734840831027,
"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": 0.012239583333333326,
"completions/max_length": 3893.8,
"completions/max_terminated_length": 3893.8,
"completions/mean_length": 956.0777099609375,
"completions/mean_terminated_length": 967.9925659179687,
"completions/min_length": 0.0,
"completions/min_terminated_length": 256.2,
"epoch": 0.8413461538461539,
"grad_norm": 0.0008964097942225635,
"learning_rate": 4.146634615384616e-06,
"loss": -0.0127,
"num_tokens": 754233501.0,
"reward": 1.2726234674453736,
"reward_std": 0.15154092907905578,
"rewards/accuracy_reward": 0.7247395753860474,
"rewards/brier_reward": 0.8327310800552368,
"rewards/confidence_one_or_zero": 0.000260416668606922,
"rewards/format_reward": 0.9877604126930237,
"rewards/mean_confidence_reward": 0.7958799719810485,
"sampling/batch_mean_priority_error": 0.014881944444444437,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4722222222222222,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0011662805685773493,
"sampling/priority_kl": 0.030000032857060432,
"sampling/priority_scale": 0.7186060607200488,
"sampling/prob_entropy": 10.278964614868164,
"sampling/prob_max": 4.364315245766193e-05,
"sampling/prob_min": 1.8648596233106217e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.8351999878883362,
"sampling/prompt_draws_total": 25056.0,
"sampling/seen_fraction": 0.6248266577720643,
"sampling/unseen_fraction": 0.3751733422279358,
"signal/accuracy_reward/centered_abs_mean": 0.1272081181406975,
"signal/accuracy_reward/group_std_mean": 0.1718853384256363,
"signal/accuracy_reward/group_zero_std_frac": 0.49722222685813905,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06360405907034875,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06360405907034875,
"signal/advantage_abs_mean": 0.10877088755369187,
"signal/advantage_pre_scale_abs_mean": 0.10877088755369187,
"signal/advantage_pre_scale_std": 0.21029807925224303,
"signal/advantage_std": 0.21029807925224303,
"signal/brier_reward/centered_abs_mean": 0.08816509544849396,
"signal/brier_reward/group_std_mean": 0.12144324630498886,
"signal/brier_reward/group_zero_std_frac": 0.20000000149011612,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04408254772424698,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04408254772424698,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0005045572877861559,
"signal/confidence_one_or_zero/group_std_mean": 0.0014731390401721,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/format_reward/centered_abs_mean": 0.019775390811264514,
"signal/format_reward/group_std_mean": 0.03549950160086155,
"signal/format_reward/group_zero_std_frac": 0.8583333373069764,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009887695405632257,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009887695405632257,
"signal/mean_confidence_reward/centered_abs_mean": 0.05625826939940452,
"signal/mean_confidence_reward/group_std_mean": 0.07565618604421616,
"signal/mean_confidence_reward/group_zero_std_frac": 0.23888888955116272,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.625826929644973e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.625826929644973e-07,
"step": 350
},
{
"epoch": 0.8413461538461539,
"eval_calibration/aurc": 0.16606825973650804,
"eval_calibration/batch_distribution_entropy": 0.5848651349995244,
"eval_calibration/batch_entropy_100bins": 0.3121246426668811,
"eval_calibration/batch_entropy_10bins": 0.5848651349995244,
"eval_calibration/batch_entropy_50bins": 0.367428079213045,
"eval_calibration/batch_uniqueness": 0.3537960150186423,
"eval_calibration/confidence_entropy": 0.4525167738181449,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.0,
"eval_calibration/coverage@15%": 0.5004389815627743,
"eval_calibration/coverage@20%": 0.7471466198419666,
"eval_calibration/coverage@25%": 0.8805970149253731,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.5848651349995244,
"eval_calibration/distribution_entropy_100": 0.3121246426668811,
"eval_calibration/ece": 0.07936786654960498,
"eval_calibration/mean_confidence": 0.7861281826163302,
"eval_calibration/unique_confidence_per_question": 0.010416666666666666,
"eval_calibration/unique_confidences": 12,
"eval_completions/clipped_ratio": 0.008680555555555544,
"eval_completions/max_length": 3213.5,
"eval_completions/max_terminated_length": 3213.5,
"eval_completions/mean_length": 1018.7047322591146,
"eval_completions/mean_terminated_length": 1027.7530924479167,
"eval_completions/min_length": 129.83333333333334,
"eval_completions/min_terminated_length": 324.8333333333333,
"eval_loss": 0.0,
"eval_num_tokens": 754233501.0,
"eval_reward": 1.2533830801645915,
"eval_reward_std": 0.35429341594378155,
"eval_rewards/accuracy_reward": 0.6987847288449606,
"eval_rewards/brier_reward": 0.8192505836486816,
"eval_rewards/confidence_one_or_zero": 0.0008680555814256271,
"eval_rewards/format_reward": 0.9887152711550394,
"eval_rewards/mean_confidence_reward": 0.7776041527589163,
"eval_runtime": 207.5115,
"eval_samples_per_second": 4.819,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4018554737170537,
"eval_signal/accuracy_reward/group_std_mean": 0.45326830446720123,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20092773685852686,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.20092773685852686,
"eval_signal/advantage_abs_mean": 0.30226531128088635,
"eval_signal/advantage_pre_scale_abs_mean": 0.30226531128088635,
"eval_signal/advantage_pre_scale_std": 0.3529137223958969,
"eval_signal/advantage_std": 0.3529137223958969,
"eval_signal/brier_reward/centered_abs_mean": 0.2024331366022428,
"eval_signal/brier_reward/group_std_mean": 0.2561540404955546,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.1012165683011214,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.1012165683011214,
"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": 0.021755641947189968,
"eval_signal/format_reward/group_std_mean": 0.06084661930799484,
"eval_signal/format_reward/group_zero_std_frac": 0.6666666915019354,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.010877820973594984,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.010877820973594984,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.12927334010601044,
"eval_signal/mean_confidence_reward/group_std_mean": 0.1735323245326678,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.2927333538452028e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.2927333538452028e-06,
"eval_steps_per_second": 0.029,
"step": 350
},
{
"epoch": 0.8413461538461539,
"step": 350,
"train_probe_calibration/aurc": 0.13379297260752054,
"train_probe_calibration/batch_distribution_entropy": 0.5711268571988026,
"train_probe_calibration/batch_entropy_100bins": 0.301687839849546,
"train_probe_calibration/batch_entropy_10bins": 0.5711268571988026,
"train_probe_calibration/batch_entropy_50bins": 0.35514204380253295,
"train_probe_calibration/batch_uniqueness": 0.32175586312125304,
"train_probe_calibration/confidence_entropy": 0.44695025580987513,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.5123022847100176,
"train_probe_calibration/coverage@15%": 0.5123022847100176,
"train_probe_calibration/coverage@20%": 0.8321616871704746,
"train_probe_calibration/coverage@25%": 0.9226713532513181,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.5711268571988026,
"train_probe_calibration/distribution_entropy_100": 0.301687839849546,
"train_probe_calibration/ece": 0.06252196836555338,
"train_probe_calibration/mean_confidence": 0.7897627416520212,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.011284722222222229,
"train_probe_completions/max_length": 3109.1666666666665,
"train_probe_completions/max_terminated_length": 3109.1666666666665,
"train_probe_completions/mean_length": 1027.9149475097656,
"train_probe_completions/mean_terminated_length": 1039.559346516927,
"train_probe_completions/min_length": 97.5,
"train_probe_completions/min_terminated_length": 289.6666666666667,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 754233501.0,
"train_probe_reward": 1.2771899302800496,
"train_probe_reward_std": 0.3486747443675995,
"train_probe_rewards/accuracy_reward": 0.7326388955116272,
"train_probe_rewards/brier_reward": 0.8338780403137207,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9878472288449606,
"train_probe_rewards/mean_confidence_reward": 0.7801649272441864,
"train_probe_runtime": 210.7138,
"train_probe_samples_per_second": 4.746,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3829210102558136,
"train_probe_signal/accuracy_reward/group_std_mean": 0.44330157339572906,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1914605051279068,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.1914605051279068,
"train_probe_signal/advantage_abs_mean": 0.28897509972254437,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.28897509972254437,
"train_probe_signal/advantage_pre_scale_std": 0.346636101603508,
"train_probe_signal/advantage_std": 0.346636101603508,
"train_probe_signal/brier_reward/centered_abs_mean": 0.19374607255061468,
"train_probe_signal/brier_reward/group_std_mean": 0.2492145225405693,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.09687303627530734,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.09687303627530734,
"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": 0.02332899331425627,
"train_probe_signal/format_reward/group_std_mean": 0.06276767483601968,
"train_probe_signal/format_reward/group_zero_std_frac": 0.6666666766007742,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.011664496657128135,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.011664496657128135,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.1300916684170564,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.17594958345095316,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.3009165892678236e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.3009165892678236e-06,
"train_probe_steps_per_second": 0.028
},
{
"calibration/aurc": 0.21392597337683425,
"calibration/batch_distribution_entropy": 0.5947614532869031,
"calibration/batch_entropy_100bins": 0.32152541550615227,
"calibration/batch_entropy_10bins": 0.5947614532869031,
"calibration/batch_entropy_50bins": 0.37665393610920955,
"calibration/batch_uniqueness": 0.39637935511209765,
"calibration/confidence_entropy": 0.47112190410813054,
"calibration/coverage@0%": 0.0005347593582887701,
"calibration/coverage@1%": 0.0005347593582887701,
"calibration/coverage@10%": 0.03333899216252158,
"calibration/coverage@15%": 0.20688396570749515,
"calibration/coverage@20%": 0.5179778252432606,
"calibration/coverage@25%": 0.745892119333808,
"calibration/coverage@30%": 0.8624386627335564,
"calibration/coverage@5%": 0.03333899216252158,
"calibration/distribution_entropy_10": 0.5947614532869031,
"calibration/distribution_entropy_100": 0.32152541550615227,
"calibration/ece": 0.12702372029237322,
"calibration/mean_confidence": 0.768511211859987,
"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": 0.011111111111111138,
"completions/max_length": 3950.6,
"completions/max_terminated_length": 3950.6,
"completions/mean_length": 1083.3509521484375,
"completions/mean_terminated_length": 1095.9435668945312,
"completions/min_length": 0.0,
"completions/min_terminated_length": 284.8,
"epoch": 0.8533653846153846,
"grad_norm": 0.0007383286720141768,
"learning_rate": 4.116586538461539e-06,
"loss": -0.0118,
"num_tokens": 769812904.0,
"reward": 1.2695351600646974,
"reward_std": 0.1539124518632889,
"rewards/accuracy_reward": 0.7167534708976746,
"rewards/brier_reward": 0.8334994554519654,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9888020992279053,
"rewards/mean_confidence_reward": 0.767276120185852,
"sampling/batch_mean_priority_error": 0.009087106963734565,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5166666666666666,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0011780404252931475,
"sampling/priority_kl": 0.029998865351080893,
"sampling/priority_scale": 0.7173613130347803,
"sampling/prob_entropy": 10.278946495056152,
"sampling/prob_max": 4.3774918594863266e-05,
"sampling/prob_min": 1.872878747235518e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.8472000002861023,
"sampling/prompt_draws_total": 25416.0,
"sampling/seen_fraction": 0.6307400107383728,
"sampling/unseen_fraction": 0.3692599892616272,
"signal/accuracy_reward/centered_abs_mean": 0.136865234375,
"signal/accuracy_reward/group_std_mean": 0.1819414019584656,
"signal/accuracy_reward/group_zero_std_frac": 0.46944445967674253,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0684326171875,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0684326171875,
"signal/advantage_abs_mean": 0.11170721650123597,
"signal/advantage_pre_scale_abs_mean": 0.11170721650123597,
"signal/advantage_pre_scale_std": 0.20967997014522552,
"signal/advantage_std": 0.20967997014522552,
"signal/brier_reward/centered_abs_mean": 0.08867686092853547,
"signal/brier_reward/group_std_mean": 0.12051396965980529,
"signal/brier_reward/group_zero_std_frac": 0.061111112125217915,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.044338430464267733,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.044338430464267733,
"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": 0.01946072056889534,
"signal/format_reward/group_std_mean": 0.03850421905517578,
"signal/format_reward/group_zero_std_frac": 0.8333333253860473,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00973036028444767,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00973036028444767,
"signal/mean_confidence_reward/centered_abs_mean": 0.06412911564111709,
"signal/mean_confidence_reward/group_std_mean": 0.08581863492727279,
"signal/mean_confidence_reward/group_zero_std_frac": 0.06388888973742723,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.41291137526423e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.41291137526423e-07,
"step": 355
},
{
"calibration/aurc": 0.13148102052335414,
"calibration/batch_distribution_entropy": 0.6065593156304296,
"calibration/batch_entropy_100bins": 0.3349155943149549,
"calibration/batch_entropy_10bins": 0.6065593156304296,
"calibration/batch_entropy_50bins": 0.39367936060083886,
"calibration/batch_uniqueness": 0.4486734367925983,
"calibration/confidence_entropy": 0.49709864708157836,
"calibration/coverage@0%": 0.0010582677165354329,
"calibration/coverage@1%": 0.0010582677165354329,
"calibration/coverage@10%": 0.39493326306021215,
"calibration/coverage@15%": 0.5546346030242957,
"calibration/coverage@20%": 0.7751250022507082,
"calibration/coverage@25%": 0.9252238588909896,
"calibration/coverage@30%": 0.9725593667546175,
"calibration/coverage@5%": 0.20250454482663352,
"calibration/distribution_entropy_10": 0.6065593156304296,
"calibration/distribution_entropy_100": 0.3349155943149549,
"calibration/ece": 0.08411662682927092,
"calibration/mean_confidence": 0.7514742164045105,
"calibration/unique_confidence_per_question": 0.0265625,
"calibration/unique_confidences": 10.2,
"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": 0.012586805555555558,
"completions/max_length": 3931.8,
"completions/max_terminated_length": 3931.8,
"completions/mean_length": 1107.271875,
"completions/mean_terminated_length": 1121.4744506835937,
"completions/min_length": 0.0,
"completions/min_terminated_length": 287.6,
"epoch": 0.8653846153846154,
"grad_norm": 0.0007773222750984132,
"learning_rate": 4.086538461538462e-06,
"loss": -0.0146,
"num_tokens": 785668036.0,
"reward": 1.2651597738265992,
"reward_std": 0.13495090901851653,
"rewards/accuracy_reward": 0.7113715291023255,
"rewards/brier_reward": 0.8316938519477844,
"rewards/confidence_one_or_zero": 0.0006076388992369175,
"rewards/format_reward": 0.9872395753860473,
"rewards/mean_confidence_reward": 0.7259296298027038,
"sampling/batch_mean_priority_error": 0.012178544557823126,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.47777777777777775,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0011908516753464937,
"sampling/priority_kl": 0.029999721422791482,
"sampling/priority_scale": 0.7162323415046558,
"sampling/prob_entropy": 10.278964042663574,
"sampling/prob_max": 4.390918838907965e-05,
"sampling/prob_min": 1.881135249277577e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.8592000007629395,
"sampling/prompt_draws_total": 25776.0,
"sampling/seen_fraction": 0.6366600155830383,
"sampling/unseen_fraction": 0.3633399844169617,
"signal/accuracy_reward/centered_abs_mean": 0.11122504472732545,
"signal/accuracy_reward/group_std_mean": 0.1490047350525856,
"signal/accuracy_reward/group_zero_std_frac": 0.5583333432674408,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05561252236366272,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05561252236366272,
"signal/advantage_abs_mean": 0.09449863135814666,
"signal/advantage_pre_scale_abs_mean": 0.09449863135814666,
"signal/advantage_pre_scale_std": 0.18920642435550689,
"signal/advantage_std": 0.18920642435550689,
"signal/brier_reward/centered_abs_mean": 0.07522271126508713,
"signal/brier_reward/group_std_mean": 0.10542623102664947,
"signal/brier_reward/group_zero_std_frac": 0.02222222276031971,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.037611355632543565,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.037611355632543565,
"signal/confidence_one_or_zero/centered_abs_mean": 0.001177300326526165,
"signal/confidence_one_or_zero/group_std_mean": 0.0034373244270682335,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9805555462837219,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.1773002484005702e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.1773002484005702e-08,
"signal/format_reward/centered_abs_mean": 0.022216796875,
"signal/format_reward/group_std_mean": 0.04519350677728653,
"signal/format_reward/group_zero_std_frac": 0.800000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0111083984375,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0111083984375,
"signal/mean_confidence_reward/centered_abs_mean": 0.06626706123352051,
"signal/mean_confidence_reward/group_std_mean": 0.09146434217691421,
"signal/mean_confidence_reward/group_zero_std_frac": 0.02222222276031971,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.626705726375803e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.626705726375803e-07,
"step": 360
},
{
"calibration/aurc": 0.11277692746806756,
"calibration/batch_distribution_entropy": 0.6456395996278379,
"calibration/batch_entropy_100bins": 0.35729776936261837,
"calibration/batch_entropy_10bins": 0.6456395996278379,
"calibration/batch_entropy_50bins": 0.4198155415901105,
"calibration/batch_uniqueness": 0.473133505996256,
"calibration/confidence_entropy": 0.5311463132708747,
"calibration/coverage@0%": 0.06946531022399341,
"calibration/coverage@1%": 0.06946531022399341,
"calibration/coverage@10%": 0.5981245716041369,
"calibration/coverage@15%": 0.7243289521094971,
"calibration/coverage@20%": 0.8575133118963988,
"calibration/coverage@25%": 0.9006357537602036,
"calibration/coverage@30%": 0.941569895397018,
"calibration/coverage@5%": 0.35447750809190104,
"calibration/distribution_entropy_10": 0.6456395996278379,
"calibration/distribution_entropy_100": 0.35729776936261837,
"calibration/ece": 0.12188690030338205,
"calibration/mean_confidence": 0.7149970556060781,
"calibration/unique_confidence_per_question": 0.0296875,
"calibration/unique_confidences": 11.4,
"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": 0.0140625,
"completions/max_length": 3901.2,
"completions/max_terminated_length": 3901.2,
"completions/mean_length": 1143.3982177734374,
"completions/mean_terminated_length": 1159.8265380859375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 300.6,
"epoch": 0.8774038461538461,
"grad_norm": 0.0007361032767221332,
"learning_rate": 4.0564903846153846e-06,
"loss": -0.0169,
"num_tokens": 801933423.0,
"reward": 1.277792716026306,
"reward_std": 0.13029044717550278,
"rewards/accuracy_reward": 0.7333333373069764,
"rewards/brier_reward": 0.8366477251052856,
"rewards/confidence_one_or_zero": 0.0011284722480922938,
"rewards/format_reward": 0.9855902791023254,
"rewards/mean_confidence_reward": 0.7020756244659424,
"sampling/batch_mean_priority_error": 0.013198098471161029,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.48888888888888893,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0012046080781146884,
"sampling/priority_kl": 0.0300000824034214,
"sampling/priority_scale": 0.7152359307045117,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 4.404441424412653e-05,
"sampling/prob_min": 1.8891548461397178e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.8712000012397766,
"sampling/prompt_draws_total": 26136.0,
"sampling/seen_fraction": 0.6425066709518432,
"sampling/unseen_fraction": 0.3574933290481567,
"signal/accuracy_reward/centered_abs_mean": 0.10416666716337204,
"signal/accuracy_reward/group_std_mean": 0.14664508551359176,
"signal/accuracy_reward/group_zero_std_frac": 0.5500000059604645,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05208333358168602,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05208333358168602,
"signal/advantage_abs_mean": 0.08776064664125442,
"signal/advantage_pre_scale_abs_mean": 0.08776064664125442,
"signal/advantage_pre_scale_std": 0.18522117137908936,
"signal/advantage_std": 0.18522117137908936,
"signal/brier_reward/centered_abs_mean": 0.06859320253133774,
"signal/brier_reward/group_std_mean": 0.09766077995300293,
"signal/brier_reward/group_zero_std_frac": 0.013888888992369175,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03429660126566887,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03429660126566887,
"signal/confidence_one_or_zero/centered_abs_mean": 0.002164713526144624,
"signal/confidence_one_or_zero/group_std_mean": 0.005785720981657505,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9694444417953492,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.1647135284297293e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.1647135284297293e-08,
"signal/format_reward/centered_abs_mean": 0.024555121548473835,
"signal/format_reward/group_std_mean": 0.0473395012319088,
"signal/format_reward/group_zero_std_frac": 0.800000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.012277560774236917,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.012277560774236917,
"signal/mean_confidence_reward/centered_abs_mean": 0.06469639614224434,
"signal/mean_confidence_reward/group_std_mean": 0.08819877356290817,
"signal/mean_confidence_reward/group_zero_std_frac": 0.013888888992369175,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.469639174611075e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.469639174611075e-07,
"step": 365
},
{
"calibration/aurc": 0.1386283079594242,
"calibration/batch_distribution_entropy": 0.6931053922664379,
"calibration/batch_entropy_100bins": 0.38151320223980084,
"calibration/batch_entropy_10bins": 0.6931053922664379,
"calibration/batch_entropy_50bins": 0.4491111688447984,
"calibration/batch_uniqueness": 0.5672203195561937,
"calibration/confidence_entropy": 0.538125630240651,
"calibration/coverage@0%": 0.06229394364436791,
"calibration/coverage@1%": 0.1340216923354674,
"calibration/coverage@10%": 0.44910862347893205,
"calibration/coverage@15%": 0.5674081734008574,
"calibration/coverage@20%": 0.7405481357036434,
"calibration/coverage@25%": 0.8005214966111949,
"calibration/coverage@30%": 0.919051442164734,
"calibration/coverage@5%": 0.2802612486748973,
"calibration/distribution_entropy_10": 0.6931053922664379,
"calibration/distribution_entropy_100": 0.38151320223980084,
"calibration/ece": 0.13417186403782377,
"calibration/mean_confidence": 0.7066106638231785,
"calibration/unique_confidence_per_question": 0.0296875,
"calibration/unique_confidences": 11.4,
"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": 0.022829861111111117,
"completions/max_length": 4059.0,
"completions/max_terminated_length": 4059.0,
"completions/mean_length": 1201.0027099609374,
"completions/mean_terminated_length": 1229.2183837890625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 341.8,
"epoch": 0.8894230769230769,
"grad_norm": 0.0007796495920047164,
"learning_rate": 4.026442307692308e-06,
"loss": -0.0246,
"num_tokens": 818856078.0,
"reward": 1.2549877882003784,
"reward_std": 0.15138355791568756,
"rewards/accuracy_reward": 0.7102430582046508,
"rewards/brier_reward": 0.8225485563278199,
"rewards/confidence_one_or_zero": 0.0004340277810115367,
"rewards/format_reward": 0.9771701335906983,
"rewards/mean_confidence_reward": 0.6898022055625915,
"sampling/batch_mean_priority_error": 0.016525694816846175,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5111111111111111,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0012221216456964612,
"sampling/priority_kl": 0.030001383274793625,
"sampling/priority_scale": 0.714642256568186,
"sampling/prob_entropy": 10.278974151611328,
"sampling/prob_max": 4.4190864718984814e-05,
"sampling/prob_min": 1.8967684445669876e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.8832000017166137,
"sampling/prompt_draws_total": 26496.0,
"sampling/seen_fraction": 0.6487133264541626,
"sampling/unseen_fraction": 0.3512866735458374,
"signal/accuracy_reward/centered_abs_mean": 0.12760416567325591,
"signal/accuracy_reward/group_std_mean": 0.1682043880224228,
"signal/accuracy_reward/group_zero_std_frac": 0.5194444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06380208283662796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06380208283662796,
"signal/advantage_abs_mean": 0.10744879245758057,
"signal/advantage_pre_scale_abs_mean": 0.10744879245758057,
"signal/advantage_pre_scale_std": 0.21186884343624116,
"signal/advantage_std": 0.21186884343624116,
"signal/brier_reward/centered_abs_mean": 0.07936687171459197,
"signal/brier_reward/group_std_mean": 0.11007404923439026,
"signal/brier_reward/group_zero_std_frac": 0.00555555559694767,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03968343585729599,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03968343585729599,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0008409288129769266,
"signal/confidence_one_or_zero/group_std_mean": 0.0024552317336201668,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 8.409287488575501e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.409287488575501e-09,
"signal/format_reward/centered_abs_mean": 0.03474934808909893,
"signal/format_reward/group_std_mean": 0.062300239503383634,
"signal/format_reward/group_zero_std_frac": 0.7500000119209289,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.017374674044549464,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.017374674044549464,
"signal/mean_confidence_reward/centered_abs_mean": 0.07127341628074646,
"signal/mean_confidence_reward/group_std_mean": 0.0958697572350502,
"signal/mean_confidence_reward/group_zero_std_frac": 0.00555555559694767,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.127341632440221e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.127341632440221e-07,
"step": 370
},
{
"calibration/aurc": 0.10029335766168149,
"calibration/batch_distribution_entropy": 0.6902467942968492,
"calibration/batch_entropy_100bins": 0.3845658940121653,
"calibration/batch_entropy_10bins": 0.6902467942968492,
"calibration/batch_entropy_50bins": 0.45234008291154665,
"calibration/batch_uniqueness": 0.5422519454806408,
"calibration/confidence_entropy": 0.5160662325119072,
"calibration/coverage@0%": 0.10886992645820467,
"calibration/coverage@1%": 0.18280609667097064,
"calibration/coverage@10%": 0.538345710678048,
"calibration/coverage@15%": 0.6977265965974512,
"calibration/coverage@20%": 0.8670921079320463,
"calibration/coverage@25%": 0.9028599732154458,
"calibration/coverage@30%": 0.9792349726775956,
"calibration/coverage@5%": 0.38965790086026597,
"calibration/distribution_entropy_10": 0.6902467942968492,
"calibration/distribution_entropy_100": 0.3845658940121653,
"calibration/ece": 0.12545484203132018,
"calibration/mean_confidence": 0.7241922030059907,
"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": 0.01684027777777781,
"completions/max_length": 4045.0,
"completions/max_terminated_length": 4045.0,
"completions/mean_length": 1237.6139892578126,
"completions/mean_terminated_length": 1258.8419921875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 356.2,
"epoch": 0.9014423076923077,
"grad_norm": 0.0007148521253839135,
"learning_rate": 3.996394230769231e-06,
"loss": -0.0189,
"num_tokens": 836198671.0,
"reward": 1.2773468732833861,
"reward_std": 0.149992173910141,
"rewards/accuracy_reward": 0.7309895873069763,
"rewards/brier_reward": 0.8406170248985291,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9830729246139527,
"rewards/mean_confidence_reward": 0.7109095573425293,
"sampling/batch_mean_priority_error": 0.013326693225758213,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.425,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0012390354182571172,
"sampling/priority_kl": 0.030000606551766396,
"sampling/priority_scale": 0.713545459578745,
"sampling/prob_entropy": 10.278951835632324,
"sampling/prob_max": 4.431868874235079e-05,
"sampling/prob_min": 1.9047261230298317e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.895199990272522,
"sampling/prompt_draws_total": 26856.0,
"sampling/seen_fraction": 0.6540533304214478,
"sampling/unseen_fraction": 0.34594666957855225,
"signal/accuracy_reward/centered_abs_mean": 0.138232421875,
"signal/accuracy_reward/group_std_mean": 0.17746042907238008,
"signal/accuracy_reward/group_zero_std_frac": 0.5055555641651154,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0691162109375,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0691162109375,
"signal/advantage_abs_mean": 0.10759708732366562,
"signal/advantage_pre_scale_abs_mean": 0.10759708732366562,
"signal/advantage_pre_scale_std": 0.20929722487926483,
"signal/advantage_std": 0.20929722487926483,
"signal/brier_reward/centered_abs_mean": 0.0766251802444458,
"signal/brier_reward/group_std_mean": 0.10730268806219101,
"signal/brier_reward/group_zero_std_frac": 0.05000000093132258,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0383125901222229,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0383125901222229,
"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": 0.02868381068110466,
"signal/format_reward/group_std_mean": 0.05322592183947563,
"signal/format_reward/group_zero_std_frac": 0.7833333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01434190534055233,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01434190534055233,
"signal/mean_confidence_reward/centered_abs_mean": 0.06712552011013032,
"signal/mean_confidence_reward/group_std_mean": 0.09156746119260788,
"signal/mean_confidence_reward/group_zero_std_frac": 0.05000000093132258,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.712551794407773e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.712551794407773e-07,
"step": 375
},
{
"calibration/aurc": 0.06687940158463018,
"calibration/batch_distribution_entropy": 0.6765535702173531,
"calibration/batch_entropy_100bins": 0.3947297330988418,
"calibration/batch_entropy_10bins": 0.6765535702173531,
"calibration/batch_entropy_50bins": 0.4615792721378192,
"calibration/batch_uniqueness": 0.5223129261859903,
"calibration/confidence_entropy": 0.48673361828445605,
"calibration/coverage@0%": 0.10752003188823844,
"calibration/coverage@1%": 0.38548056959488225,
"calibration/coverage@10%": 0.7209976575948486,
"calibration/coverage@15%": 0.8182014845492833,
"calibration/coverage@20%": 0.8610435822792895,
"calibration/coverage@25%": 0.919575639140328,
"calibration/coverage@30%": 0.9627777777777778,
"calibration/coverage@5%": 0.6677153378203822,
"calibration/distribution_entropy_10": 0.6765535702173531,
"calibration/distribution_entropy_100": 0.3947297330988418,
"calibration/ece": 0.1527699526413858,
"calibration/mean_confidence": 0.7355369041664158,
"calibration/unique_confidence_per_question": 0.04114583333333334,
"calibration/unique_confidences": 15.8,
"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": 0.029079861111111115,
"completions/max_length": 4048.0,
"completions/max_terminated_length": 4048.0,
"completions/mean_length": 1363.7867431640625,
"completions/mean_terminated_length": 1404.6439697265625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 416.0,
"epoch": 0.9134615384615384,
"grad_norm": 0.0005625736084766686,
"learning_rate": 3.966346153846154e-06,
"loss": -0.0313,
"num_tokens": 855025046.0,
"reward": 1.2572867155075074,
"reward_std": 0.15128274112939835,
"rewards/accuracy_reward": 0.7096354007720947,
"rewards/brier_reward": 0.8341776728630066,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9707465171813965,
"rewards/mean_confidence_reward": 0.6915950655937195,
"sampling/batch_mean_priority_error": 0.015527221156608076,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.43888888888888894,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0012547509046271442,
"sampling/priority_kl": 0.029999660328030586,
"sampling/priority_scale": 0.7124824702506884,
"sampling/prob_entropy": 10.278944396972657,
"sampling/prob_max": 4.444640653673559e-05,
"sampling/prob_min": 1.9126197730656714e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9072000026702881,
"sampling/prompt_draws_total": 27216.0,
"sampling/seen_fraction": 0.6593066573143005,
"sampling/unseen_fraction": 0.34069334268569945,
"signal/accuracy_reward/centered_abs_mean": 0.12245551496744156,
"signal/accuracy_reward/group_std_mean": 0.16243197470903398,
"signal/accuracy_reward/group_zero_std_frac": 0.5250000119209289,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06122775748372078,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06122775748372078,
"signal/advantage_abs_mean": 0.10677067637443542,
"signal/advantage_pre_scale_abs_mean": 0.10677067637443542,
"signal/advantage_pre_scale_std": 0.21411246061325073,
"signal/advantage_std": 0.21411246061325073,
"signal/brier_reward/centered_abs_mean": 0.08327719122171402,
"signal/brier_reward/group_std_mean": 0.11531828492879867,
"signal/brier_reward/group_zero_std_frac": 0.07777777835726737,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04163859561085701,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04163859561085701,
"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": 0.04182400144636631,
"signal/format_reward/group_std_mean": 0.07100048959255219,
"signal/format_reward/group_zero_std_frac": 0.7305555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.020912000723183156,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.020912000723183156,
"signal/mean_confidence_reward/centered_abs_mean": 0.07033055424690246,
"signal/mean_confidence_reward/group_std_mean": 0.0946671724319458,
"signal/mean_confidence_reward/group_zero_std_frac": 0.07777777835726737,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.033055226202123e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.033055226202123e-07,
"step": 380
},
{
"calibration/aurc": 0.09305614350965419,
"calibration/batch_distribution_entropy": 0.6883623876554645,
"calibration/batch_entropy_100bins": 0.3986153066388806,
"calibration/batch_entropy_10bins": 0.6883623876554645,
"calibration/batch_entropy_50bins": 0.4679957218573433,
"calibration/batch_uniqueness": 0.5225835025994161,
"calibration/confidence_entropy": 0.48683141606517133,
"calibration/coverage@0%": 0.2549390269850611,
"calibration/coverage@1%": 0.31022357983058957,
"calibration/coverage@10%": 0.5844218566775878,
"calibration/coverage@15%": 0.711092937754674,
"calibration/coverage@20%": 0.9079099138819948,
"calibration/coverage@25%": 0.9586206896551724,
"calibration/coverage@30%": 0.9586206896551724,
"calibration/coverage@5%": 0.4897845827164402,
"calibration/distribution_entropy_10": 0.6883623876554645,
"calibration/distribution_entropy_100": 0.3986153066388806,
"calibration/ece": 0.117542802957425,
"calibration/mean_confidence": 0.7463551204519809,
"calibration/unique_confidence_per_question": 0.049479166666666664,
"calibration/unique_confidences": 19.0,
"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": 0.024826388888888905,
"completions/max_length": 4044.2,
"completions/max_terminated_length": 4044.2,
"completions/mean_length": 1388.8342041015626,
"completions/mean_terminated_length": 1424.3485107421875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 400.2,
"epoch": 0.9254807692307693,
"grad_norm": 0.0005671771941706538,
"learning_rate": 3.936298076923077e-06,
"loss": -0.0298,
"num_tokens": 874080512.0,
"reward": 1.2843957901000977,
"reward_std": 0.16392693817615508,
"rewards/accuracy_reward": 0.7434895992279053,
"rewards/brier_reward": 0.8502007126808167,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9750868082046509,
"rewards/mean_confidence_reward": 0.7203326463699341,
"sampling/batch_mean_priority_error": 0.009188816231087813,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4416666666666667,
"sampling/error_ema_max": 0.07874999940395355,
"sampling/error_ema_mean": 0.0012672836193814874,
"sampling/priority_kl": 0.030001260340213776,
"sampling/priority_scale": 0.7116214931709692,
"sampling/prob_entropy": 10.278961372375488,
"sampling/prob_max": 4.457824543351308e-05,
"sampling/prob_min": 1.9202485782443544e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9192000031471252,
"sampling/prompt_draws_total": 27576.0,
"sampling/seen_fraction": 0.6646333336830139,
"sampling/unseen_fraction": 0.3353666663169861,
"signal/accuracy_reward/centered_abs_mean": 0.12014431208372116,
"signal/accuracy_reward/group_std_mean": 0.1654144525527954,
"signal/accuracy_reward/group_zero_std_frac": 0.49722222089767454,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06007215604186058,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06007215604186058,
"signal/advantage_abs_mean": 0.11156907230615616,
"signal/advantage_pre_scale_abs_mean": 0.11156907230615616,
"signal/advantage_pre_scale_std": 0.22239162027835846,
"signal/advantage_std": 0.22239162027835846,
"signal/brier_reward/centered_abs_mean": 0.0873231202363968,
"signal/brier_reward/group_std_mean": 0.1248858317732811,
"signal/brier_reward/group_zero_std_frac": 0.10000000298023223,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0436615601181984,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0436615601181984,
"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": 0.04041341133415699,
"signal/format_reward/group_std_mean": 0.07462992146611214,
"signal/format_reward/group_zero_std_frac": 0.6916666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.020206705667078495,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.020206705667078495,
"signal/mean_confidence_reward/centered_abs_mean": 0.07641240209341049,
"signal/mean_confidence_reward/group_std_mean": 0.10640424340963364,
"signal/mean_confidence_reward/group_zero_std_frac": 0.10000000298023223,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.641240131306404e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.641240131306404e-07,
"step": 385
},
{
"calibration/aurc": 0.13190790475605713,
"calibration/batch_distribution_entropy": 0.6945369113155875,
"calibration/batch_entropy_100bins": 0.3849626963468917,
"calibration/batch_entropy_10bins": 0.6945369113155875,
"calibration/batch_entropy_50bins": 0.4531718574953187,
"calibration/batch_uniqueness": 0.5077238360997742,
"calibration/confidence_entropy": 0.48116454635255723,
"calibration/coverage@0%": 0.12612137203166227,
"calibration/coverage@1%": 0.21842906433935458,
"calibration/coverage@10%": 0.42993735385232174,
"calibration/coverage@15%": 0.5684094677216365,
"calibration/coverage@20%": 0.653749367578372,
"calibration/coverage@25%": 0.887906932118014,
"calibration/coverage@30%": 0.9514267379679143,
"calibration/coverage@5%": 0.3843560040766311,
"calibration/distribution_entropy_10": 0.6945369113155875,
"calibration/distribution_entropy_100": 0.3849626963468917,
"calibration/ece": 0.10824475809618166,
"calibration/mean_confidence": 0.73170941520179,
"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": 0.025173611111111095,
"completions/max_length": 4080.2,
"completions/max_terminated_length": 4080.2,
"completions/mean_length": 1396.7363037109376,
"completions/mean_terminated_length": 1433.1280517578125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 430.2,
"epoch": 0.9375,
"grad_norm": 0.000409966945881024,
"learning_rate": 3.90625e-06,
"loss": -0.0266,
"num_tokens": 893290818.0,
"reward": 1.27172749042511,
"reward_std": 0.15499115586280823,
"rewards/accuracy_reward": 0.7238715171813965,
"rewards/brier_reward": 0.8447428822517395,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9748263835906983,
"rewards/mean_confidence_reward": 0.7116460919380188,
"sampling/batch_mean_priority_error": 0.012741569140127442,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.47777777777777786,
"sampling/error_ema_max": 0.10777499675750732,
"sampling/error_ema_mean": 0.001280206791125238,
"sampling/priority_kl": 0.029999075457453728,
"sampling/priority_scale": 0.7110501586692408,
"sampling/prob_entropy": 10.27893943786621,
"sampling/prob_max": 4.471761203603819e-05,
"sampling/prob_min": 1.9275546219432726e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9312000036239624,
"sampling/prompt_draws_total": 27936.0,
"sampling/seen_fraction": 0.6701466679573059,
"sampling/unseen_fraction": 0.32985333204269407,
"signal/accuracy_reward/centered_abs_mean": 0.11694335788488389,
"signal/accuracy_reward/group_std_mean": 0.16504150331020356,
"signal/accuracy_reward/group_zero_std_frac": 0.48611111640930177,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05847167894244194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05847167894244194,
"signal/advantage_abs_mean": 0.1060485914349556,
"signal/advantage_pre_scale_abs_mean": 0.1060485914349556,
"signal/advantage_pre_scale_std": 0.21162693500518798,
"signal/advantage_std": 0.21162693500518798,
"signal/brier_reward/centered_abs_mean": 0.08900733143091202,
"signal/brier_reward/group_std_mean": 0.12304618060588837,
"signal/brier_reward/group_zero_std_frac": 0.1194444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04450366571545601,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04450366571545601,
"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": 0.03811848983168602,
"signal/format_reward/group_std_mean": 0.06721950992941857,
"signal/format_reward/group_zero_std_frac": 0.7361111164093017,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01905924491584301,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01905924491584301,
"signal/mean_confidence_reward/centered_abs_mean": 0.07885867208242417,
"signal/mean_confidence_reward/group_std_mean": 0.10565297603607178,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1250000014901161,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.885867375989619e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.885867375989619e-07,
"step": 390
},
{
"calibration/aurc": 0.13851231496365907,
"calibration/batch_distribution_entropy": 0.6436947726909951,
"calibration/batch_entropy_100bins": 0.3657471373198543,
"calibration/batch_entropy_10bins": 0.6436947726909951,
"calibration/batch_entropy_50bins": 0.42894801110205344,
"calibration/batch_uniqueness": 0.4334929718033763,
"calibration/confidence_entropy": 0.4724938880811342,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.05520833333333334,
"calibration/coverage@10%": 0.6129555591818974,
"calibration/coverage@15%": 0.6619791666666667,
"calibration/coverage@20%": 0.7166666666666667,
"calibration/coverage@25%": 0.7223958333333333,
"calibration/coverage@30%": 0.9573684210526316,
"calibration/coverage@5%": 0.3250827582991421,
"calibration/distribution_entropy_10": 0.6436947726909951,
"calibration/distribution_entropy_100": 0.3657471373198543,
"calibration/ece": 0.13354448480012385,
"calibration/mean_confidence": 0.7630965564173401,
"calibration/unique_confidence_per_question": 0.04114583333333333,
"calibration/unique_confidences": 15.8,
"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": 0.014236111111111116,
"completions/max_length": 4012.6,
"completions/max_terminated_length": 4012.6,
"completions/mean_length": 1314.4784912109376,
"completions/mean_terminated_length": 1333.518115234375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 404.8,
"epoch": 0.9495192307692307,
"grad_norm": 0.0004194934736005962,
"learning_rate": 3.876201923076923e-06,
"loss": -0.0172,
"num_tokens": 911520586.0,
"reward": 1.2895013809204101,
"reward_std": 0.13368285447359085,
"rewards/accuracy_reward": 0.7513888955116272,
"rewards/brier_reward": 0.8418354272842408,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9857638955116272,
"rewards/mean_confidence_reward": 0.728717315196991,
"sampling/batch_mean_priority_error": 0.01850263710004697,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4416666666666666,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.001296161743812263,
"sampling/priority_kl": 0.029999866709113122,
"sampling/priority_scale": 0.7105536759132519,
"sampling/prob_entropy": 10.278941345214843,
"sampling/prob_max": 4.4857734610559416e-05,
"sampling/prob_min": 1.934731080837082e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9431999921798706,
"sampling/prompt_draws_total": 28296.0,
"sampling/seen_fraction": 0.6755799889564514,
"sampling/unseen_fraction": 0.32442001104354856,
"signal/accuracy_reward/centered_abs_mean": 0.1146484375,
"signal/accuracy_reward/group_std_mean": 0.15858114361763,
"signal/accuracy_reward/group_zero_std_frac": 0.5194444596767426,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05732421875,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05732421875,
"signal/advantage_abs_mean": 0.09345841109752655,
"signal/advantage_pre_scale_abs_mean": 0.09345841109752655,
"signal/advantage_pre_scale_std": 0.18798796832561493,
"signal/advantage_std": 0.18798796832561493,
"signal/brier_reward/centered_abs_mean": 0.0772012397646904,
"signal/brier_reward/group_std_mean": 0.10745720118284226,
"signal/brier_reward/group_zero_std_frac": 0.16944444477558135,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0386006198823452,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0386006198823452,
"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": 0.02316623255610466,
"signal/format_reward/group_std_mean": 0.03984398618340492,
"signal/format_reward/group_zero_std_frac": 0.8472222208976745,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01158311627805233,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01158311627805233,
"signal/mean_confidence_reward/centered_abs_mean": 0.07465712130069732,
"signal/mean_confidence_reward/group_std_mean": 0.09905266910791397,
"signal/mean_confidence_reward/group_zero_std_frac": 0.17777777910232545,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.465711860277224e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.465711860277224e-07,
"step": 395
},
{
"calibration/aurc": 0.17381065033169904,
"calibration/batch_distribution_entropy": 0.7264575998314206,
"calibration/batch_entropy_100bins": 0.41238367312173424,
"calibration/batch_entropy_10bins": 0.7264575998314206,
"calibration/batch_entropy_50bins": 0.48482897701067085,
"calibration/batch_uniqueness": 0.553584006751787,
"calibration/confidence_entropy": 0.5046126707223293,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.28964513992561025,
"calibration/coverage@15%": 0.4644649068420569,
"calibration/coverage@20%": 0.6632898498571449,
"calibration/coverage@25%": 0.7977037298323743,
"calibration/coverage@30%": 0.9234206058439817,
"calibration/coverage@5%": 0.12178938666589316,
"calibration/distribution_entropy_10": 0.7264575998314206,
"calibration/distribution_entropy_100": 0.41238367312173424,
"calibration/ece": 0.09958374316931348,
"calibration/mean_confidence": 0.6901195510198486,
"calibration/unique_confidence_per_question": 0.03958333333333334,
"calibration/unique_confidences": 15.2,
"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": 0.019704861111111117,
"completions/max_length": 4046.8,
"completions/max_terminated_length": 4046.8,
"completions/mean_length": 1286.089697265625,
"completions/mean_terminated_length": 1311.7608154296875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 412.8,
"epoch": 0.9615384615384616,
"grad_norm": 0.000408906169468537,
"learning_rate": 3.846153846153847e-06,
"loss": -0.0214,
"num_tokens": 929409907.0,
"reward": 1.2701274394989013,
"reward_std": 0.13115532398223878,
"rewards/accuracy_reward": 0.7190972447395325,
"rewards/brier_reward": 0.8408486485481262,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9802951335906982,
"rewards/mean_confidence_reward": 0.6962047338485717,
"sampling/batch_mean_priority_error": 0.01543325384199716,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.44722222222222224,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0013169086771085859,
"sampling/priority_kl": 0.029999712482094765,
"sampling/priority_scale": 0.7101320088142529,
"sampling/prob_entropy": 10.278942680358886,
"sampling/prob_max": 4.4999313831795004e-05,
"sampling/prob_min": 1.9418062584009023e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9551999926567077,
"sampling/prompt_draws_total": 28656.0,
"sampling/seen_fraction": 0.6809599995613098,
"sampling/unseen_fraction": 0.3190400004386902,
"signal/accuracy_reward/centered_abs_mean": 0.1167534738779068,
"signal/accuracy_reward/group_std_mean": 0.15651549100875856,
"signal/accuracy_reward/group_zero_std_frac": 0.544444453716278,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0583767369389534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0583767369389534,
"signal/advantage_abs_mean": 0.0936752513051033,
"signal/advantage_pre_scale_abs_mean": 0.0936752513051033,
"signal/advantage_pre_scale_std": 0.18921864926815032,
"signal/advantage_std": 0.18921864926815032,
"signal/brier_reward/centered_abs_mean": 0.0845369666814804,
"signal/brier_reward/group_std_mean": 0.1135342001914978,
"signal/brier_reward/group_zero_std_frac": 0.16388889104127885,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0422684833407402,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0422684833407402,
"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": 0.02873806469142437,
"signal/format_reward/group_std_mean": 0.04568745791912079,
"signal/format_reward/group_zero_std_frac": 0.8388888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.014369032345712185,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.014369032345712185,
"signal/mean_confidence_reward/centered_abs_mean": 0.07661729902029038,
"signal/mean_confidence_reward/group_std_mean": 0.100196972489357,
"signal/mean_confidence_reward/group_zero_std_frac": 0.16666666865348817,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.661729910068971e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.661729910068971e-07,
"step": 400
},
{
"epoch": 0.9615384615384616,
"eval_calibration/aurc": 0.14373158731161875,
"eval_calibration/batch_distribution_entropy": 0.7376132763277105,
"eval_calibration/batch_entropy_100bins": 0.4222256539447724,
"eval_calibration/batch_entropy_10bins": 0.7376132763277105,
"eval_calibration/batch_entropy_50bins": 0.49703720826994124,
"eval_calibration/batch_uniqueness": 0.5167880426479297,
"eval_calibration/confidence_entropy": 0.4762156650786053,
"eval_calibration/coverage@0%": 0.0008718395815170009,
"eval_calibration/coverage@1%": 0.0008718395815170009,
"eval_calibration/coverage@10%": 0.45422842197035745,
"eval_calibration/coverage@15%": 0.6687009590235397,
"eval_calibration/coverage@20%": 0.7707061900610288,
"eval_calibration/coverage@25%": 0.8857890148212729,
"eval_calibration/coverage@30%": 0.981691368788143,
"eval_calibration/coverage@5%": 0.0008718395815170009,
"eval_calibration/distribution_entropy_10": 0.7376132763277105,
"eval_calibration/distribution_entropy_100": 0.4222256539447724,
"eval_calibration/ece": 0.023292783412285037,
"eval_calibration/mean_confidence": 0.6979463351440592,
"eval_calibration/unique_confidence_per_question": 0.026041666666666668,
"eval_calibration/unique_confidences": 30,
"eval_completions/clipped_ratio": 0.00434027777777779,
"eval_completions/max_length": 3106.3333333333335,
"eval_completions/max_terminated_length": 3106.3333333333335,
"eval_completions/mean_length": 1229.538594563802,
"eval_completions/mean_terminated_length": 1234.8570149739583,
"eval_completions/min_length": 170.66666666666666,
"eval_completions/min_terminated_length": 471.8333333333333,
"eval_loss": 0.0,
"eval_num_tokens": 929409907.0,
"eval_reward": 1.267914394537608,
"eval_reward_std": 0.3055635889371236,
"eval_rewards/accuracy_reward": 0.6935763855775198,
"eval_rewards/brier_reward": 0.8465787768363953,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9956597288449606,
"eval_rewards/mean_confidence_reward": 0.6949170331160227,
"eval_runtime": 207.1959,
"eval_samples_per_second": 4.826,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4138997296492259,
"eval_signal/accuracy_reward/group_std_mean": 0.46077505747477215,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20694986482461294,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.20694986482461294,
"eval_signal/advantage_abs_mean": 0.26225854456424713,
"eval_signal/advantage_pre_scale_abs_mean": 0.26225854456424713,
"eval_signal/advantage_pre_scale_std": 0.3032806118329366,
"eval_signal/advantage_std": 0.3032806118329366,
"eval_signal/brier_reward/centered_abs_mean": 0.15997044493754706,
"eval_signal/brier_reward/group_std_mean": 0.21410523851712546,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07998522246877353,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.07998522246877353,
"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": 0.2043088103334109,
"eval_signal/mean_confidence_reward/group_std_mean": 0.2400745451450348,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.043088102254842e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.043088102254842e-06,
"eval_steps_per_second": 0.029,
"step": 400
},
{
"epoch": 0.9615384615384616,
"step": 400,
"train_probe_calibration/aurc": 0.1189485048532275,
"train_probe_calibration/batch_distribution_entropy": 0.7404175967039297,
"train_probe_calibration/batch_entropy_100bins": 0.4319272238513574,
"train_probe_calibration/batch_entropy_10bins": 0.7404175967039297,
"train_probe_calibration/batch_entropy_50bins": 0.5063624573463197,
"train_probe_calibration/batch_uniqueness": 0.5271480740354949,
"train_probe_calibration/confidence_entropy": 0.4770563278287817,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.5663176265270506,
"train_probe_calibration/coverage@15%": 0.725130890052356,
"train_probe_calibration/coverage@20%": 0.7827225130890052,
"train_probe_calibration/coverage@25%": 0.956369982547993,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.7404175967039297,
"train_probe_calibration/distribution_entropy_100": 0.4319272238513574,
"train_probe_calibration/ece": 0.03721958610977619,
"train_probe_calibration/mean_confidence": 0.7000523817904639,
"train_probe_calibration/unique_confidence_per_question": 0.026909722222222224,
"train_probe_calibration/unique_confidences": 31,
"train_probe_completions/clipped_ratio": 0.007638888888888899,
"train_probe_completions/max_length": 3379.5,
"train_probe_completions/max_terminated_length": 3379.5,
"train_probe_completions/mean_length": 1260.2481486002605,
"train_probe_completions/mean_terminated_length": 1269.918924967448,
"train_probe_completions/min_length": 143.5,
"train_probe_completions/min_terminated_length": 424.6666666666667,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 929409907.0,
"train_probe_reward": 1.2899209260940552,
"train_probe_reward_std": 0.29275282720724743,
"train_probe_rewards/accuracy_reward": 0.7282986044883728,
"train_probe_rewards/brier_reward": 0.8567376335461935,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9947916666666666,
"train_probe_rewards/mean_confidence_reward": 0.6964062650998434,
"train_probe_runtime": 206.9266,
"train_probe_samples_per_second": 4.833,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3830837706724803,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4421002119779587,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19154188533624014,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.19154188533624014,
"train_probe_signal/advantage_abs_mean": 0.24344467123349509,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.24344467123349509,
"train_probe_signal/advantage_pre_scale_std": 0.29259032507737476,
"train_probe_signal/advantage_std": 0.29259032507737476,
"train_probe_signal/brier_reward/centered_abs_mean": 0.14901887873808542,
"train_probe_signal/brier_reward/group_std_mean": 0.20320318390925726,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07450943936904271,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07450943936904271,
"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": 0.20346981287002563,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.24091133226950964,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.034698032578793e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.034698032578793e-06,
"train_probe_steps_per_second": 0.029
},
{
"calibration/aurc": 0.08449489142619329,
"calibration/batch_distribution_entropy": 0.6227200143289963,
"calibration/batch_entropy_100bins": 0.36338874779102365,
"calibration/batch_entropy_10bins": 0.6227200143289963,
"calibration/batch_entropy_50bins": 0.4252669096203041,
"calibration/batch_uniqueness": 0.3437965242623367,
"calibration/confidence_entropy": 0.4589139405558756,
"calibration/coverage@0%": 0.1618964947089947,
"calibration/coverage@1%": 0.18094411375661373,
"calibration/coverage@10%": 0.639110911662315,
"calibration/coverage@15%": 0.9039912280701753,
"calibration/coverage@20%": 0.976842105263158,
"calibration/coverage@25%": 0.9873684210526316,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.35717879677980857,
"calibration/distribution_entropy_10": 0.6227200143289963,
"calibration/distribution_entropy_100": 0.36338874779102365,
"calibration/ece": 0.1369168523797913,
"calibration/mean_confidence": 0.7638626715936297,
"calibration/unique_confidence_per_question": 0.0453125,
"calibration/unique_confidences": 17.4,
"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": 0.007291666666666674,
"completions/max_length": 3925.2,
"completions/max_terminated_length": 3925.2,
"completions/mean_length": 1233.3153076171875,
"completions/mean_terminated_length": 1242.411474609375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 377.4,
"epoch": 0.9735576923076923,
"grad_norm": 0.0005183282773941755,
"learning_rate": 3.81610576923077e-06,
"loss": -0.0065,
"num_tokens": 946737955.0,
"reward": 1.2722141742706299,
"reward_std": 0.13236766010522844,
"rewards/accuracy_reward": 0.7138020753860473,
"rewards/brier_reward": 0.8379905343055725,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9926215171813965,
"rewards/mean_confidence_reward": 0.7111494064331054,
"sampling/batch_mean_priority_error": 0.017268427815570656,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4083333333333333,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0013354042312130332,
"sampling/priority_kl": 0.030000364780426024,
"sampling/priority_scale": 0.7095950543647632,
"sampling/prob_entropy": 10.278936386108398,
"sampling/prob_max": 4.51350053481292e-05,
"sampling/prob_min": 1.948899807757698e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9672000050544739,
"sampling/prompt_draws_total": 29016.0,
"sampling/seen_fraction": 0.686026668548584,
"sampling/unseen_fraction": 0.31397333145141604,
"signal/accuracy_reward/centered_abs_mean": 0.13479274809360503,
"signal/accuracy_reward/group_std_mean": 0.17868539094924926,
"signal/accuracy_reward/group_zero_std_frac": 0.49166667461395264,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06739637404680252,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06739637404680252,
"signal/advantage_abs_mean": 0.09586342722177506,
"signal/advantage_pre_scale_abs_mean": 0.09586342722177506,
"signal/advantage_pre_scale_std": 0.18204142153263092,
"signal/advantage_std": 0.18204142153263092,
"signal/brier_reward/centered_abs_mean": 0.08206898719072342,
"signal/brier_reward/group_std_mean": 0.11170048117637635,
"signal/brier_reward/group_zero_std_frac": 0.19722222685813903,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04103449359536171,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04103449359536171,
"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": 0.012939453218132257,
"signal/format_reward/group_std_mean": 0.024650372192263605,
"signal/format_reward/group_zero_std_frac": 0.9,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0064697266090661286,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0064697266090661286,
"signal/mean_confidence_reward/centered_abs_mean": 0.07400602400302887,
"signal/mean_confidence_reward/group_std_mean": 0.09739783853292465,
"signal/mean_confidence_reward/group_zero_std_frac": 0.20555555820465088,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.400601930385164e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.400601930385164e-07,
"step": 405
},
{
"calibration/aurc": 0.15484554277953202,
"calibration/batch_distribution_entropy": 0.6140492358734193,
"calibration/batch_entropy_100bins": 0.3570470771855673,
"calibration/batch_entropy_10bins": 0.6140492358734193,
"calibration/batch_entropy_50bins": 0.41687620588852903,
"calibration/batch_uniqueness": 0.31276141098895854,
"calibration/confidence_entropy": 0.4564292639629343,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.11302083333333332,
"calibration/coverage@10%": 0.44270833333333337,
"calibration/coverage@15%": 0.5036458333333333,
"calibration/coverage@20%": 0.6453125,
"calibration/coverage@25%": 0.8445162521815008,
"calibration/coverage@30%": 0.9244791666666667,
"calibration/coverage@5%": 0.14583333333333331,
"calibration/distribution_entropy_10": 0.6140492358734193,
"calibration/distribution_entropy_100": 0.3570470771855673,
"calibration/ece": 0.11358587200609367,
"calibration/mean_confidence": 0.7408229315798783,
"calibration/unique_confidence_per_question": 0.05,
"calibration/unique_confidences": 19.2,
"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": 3927.8,
"completions/max_terminated_length": 3927.8,
"completions/mean_length": 1141.2799560546875,
"completions/mean_terminated_length": 1145.9073486328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 334.2,
"epoch": 0.9855769230769231,
"grad_norm": 0.00041302453610114753,
"learning_rate": 3.7860576923076927e-06,
"loss": -0.0047,
"num_tokens": 962938460.0,
"reward": 1.295286440849304,
"reward_std": 0.10948915630578995,
"rewards/accuracy_reward": 0.7405382037162781,
"rewards/brier_reward": 0.8540129423141479,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9960069298744202,
"rewards/mean_confidence_reward": 0.7417306542396546,
"sampling/batch_mean_priority_error": 0.015546130820105933,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4416666666666666,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0013538436032831669,
"sampling/priority_kl": 0.03000061847269535,
"sampling/priority_scale": 0.7092480838997289,
"sampling/prob_entropy": 10.278968238830567,
"sampling/prob_max": 4.527691780822352e-05,
"sampling/prob_min": 1.955830375663936e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.979200005531311,
"sampling/prompt_draws_total": 29376.0,
"sampling/seen_fraction": 0.6912333369255066,
"sampling/unseen_fraction": 0.3087666630744934,
"signal/accuracy_reward/centered_abs_mean": 0.11591254472732544,
"signal/accuracy_reward/group_std_mean": 0.15239314138889312,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111164093018,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05795627236366272,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05795627236366272,
"signal/advantage_abs_mean": 0.08225717097520828,
"signal/advantage_pre_scale_abs_mean": 0.08225717097520828,
"signal/advantage_pre_scale_std": 0.16391776800155639,
"signal/advantage_std": 0.16391776800155639,
"signal/brier_reward/centered_abs_mean": 0.07231278643012047,
"signal/brier_reward/group_std_mean": 0.09499142915010453,
"signal/brier_reward/group_zero_std_frac": 0.31388888955116273,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.036156393215060235,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.036156393215060235,
"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": 0.0058593748603016135,
"signal/format_reward/group_std_mean": 0.010942479129880667,
"signal/format_reward/group_zero_std_frac": 0.9555555701255798,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0029296874301508067,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0029296874301508067,
"signal/mean_confidence_reward/centered_abs_mean": 0.06354806199669838,
"signal/mean_confidence_reward/group_std_mean": 0.08368089199066162,
"signal/mean_confidence_reward/group_zero_std_frac": 0.32777778506278993,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.354805918817874e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.354805918817874e-07,
"step": 410
},
{
"calibration/aurc": 0.20428814046532634,
"calibration/batch_distribution_entropy": 0.6639657042279807,
"calibration/batch_entropy_100bins": 0.36625001474860847,
"calibration/batch_entropy_10bins": 0.6639657042279807,
"calibration/batch_entropy_50bins": 0.4311435914864713,
"calibration/batch_uniqueness": 0.43971812448499553,
"calibration/confidence_entropy": 0.4823920956258697,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.3251643245435366,
"calibration/coverage@15%": 0.3486671698927818,
"calibration/coverage@20%": 0.5319422349391003,
"calibration/coverage@25%": 0.5914233410340884,
"calibration/coverage@30%": 0.6817874865008953,
"calibration/coverage@5%": 0.1489891050264514,
"calibration/distribution_entropy_10": 0.6639657042279807,
"calibration/distribution_entropy_100": 0.36625001474860847,
"calibration/ece": 0.12688027308188327,
"calibration/mean_confidence": 0.7077171406744632,
"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": 0.0035590277777777677,
"completions/max_length": 3899.6,
"completions/max_terminated_length": 3899.6,
"completions/mean_length": 1136.95966796875,
"completions/mean_terminated_length": 1141.050439453125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 376.8,
"epoch": 0.9975961538461539,
"grad_norm": 0.0003201451036147773,
"learning_rate": 3.756009615384616e-06,
"loss": -0.0029,
"num_tokens": 979157035.0,
"reward": 1.2846659660339355,
"reward_std": 0.11020902842283249,
"rewards/accuracy_reward": 0.7144965410232544,
"rewards/brier_reward": 0.858467161655426,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9963541507720948,
"rewards/mean_confidence_reward": 0.7005656719207763,
"sampling/batch_mean_priority_error": 0.015297752478049573,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4416666666666666,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0013716007117182017,
"sampling/priority_kl": 0.030000818893313408,
"sampling/priority_scale": 0.7091951430076733,
"sampling/prob_entropy": 10.278961944580079,
"sampling/prob_max": 4.542603783193044e-05,
"sampling/prob_min": 1.9623946718638764e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9911999940872193,
"sampling/prompt_draws_total": 29736.0,
"sampling/seen_fraction": 0.6965666651725769,
"sampling/unseen_fraction": 0.3034333348274231,
"signal/accuracy_reward/centered_abs_mean": 0.11788736879825593,
"signal/accuracy_reward/group_std_mean": 0.15433158874511718,
"signal/accuracy_reward/group_zero_std_frac": 0.5666666626930237,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05894368439912796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05894368439912796,
"signal/advantage_abs_mean": 0.08145553916692734,
"signal/advantage_pre_scale_abs_mean": 0.08145553916692734,
"signal/advantage_pre_scale_std": 0.1603451281785965,
"signal/advantage_std": 0.1603451281785965,
"signal/brier_reward/centered_abs_mean": 0.06856870576739311,
"signal/brier_reward/group_std_mean": 0.09311682730913162,
"signal/brier_reward/group_zero_std_frac": 0.2194444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03428435288369656,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03428435288369656,
"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": 0.006564670125953853,
"signal/format_reward/group_std_mean": 0.014062346518039703,
"signal/format_reward/group_zero_std_frac": 0.9361111283302307,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0032823350629769266,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0032823350629769266,
"signal/mean_confidence_reward/centered_abs_mean": 0.06457249075174332,
"signal/mean_confidence_reward/group_std_mean": 0.08391422182321548,
"signal/mean_confidence_reward/group_zero_std_frac": 0.22777777910232544,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.45724924197566e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.45724924197566e-07,
"step": 415
},
{
"calibration/aurc": 0.15826559352185168,
"calibration/batch_distribution_entropy": 0.6645688676194069,
"calibration/batch_entropy_100bins": 0.3655125923463937,
"calibration/batch_entropy_10bins": 0.6645688676194069,
"calibration/batch_entropy_50bins": 0.42807686348036145,
"calibration/batch_uniqueness": 0.43058323792640574,
"calibration/confidence_entropy": 0.49414698113893907,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5363545554067046,
"calibration/coverage@15%": 0.6613748966207811,
"calibration/coverage@20%": 0.7334378801963009,
"calibration/coverage@25%": 0.7501672650130548,
"calibration/coverage@30%": 0.7892950391644908,
"calibration/coverage@5%": 0.44908787774709624,
"calibration/distribution_entropy_10": 0.6645688676194069,
"calibration/distribution_entropy_100": 0.3655125923463937,
"calibration/ece": 0.10076544504268907,
"calibration/mean_confidence": 0.6712129246549783,
"calibration/unique_confidence_per_question": 0.035416666666666666,
"calibration/unique_confidences": 13.6,
"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": 3455.4,
"completions/max_terminated_length": 3455.4,
"completions/mean_length": 1117.3010498046874,
"completions/mean_terminated_length": 1119.234912109375,
"completions/min_length": 106.2,
"completions/min_terminated_length": 362.4,
"epoch": 1.0096153846153846,
"grad_norm": 0.0004065485845785588,
"learning_rate": 3.725961538461539e-06,
"loss": -0.0,
"num_tokens": 995089724.0,
"reward": 1.2725646257400514,
"reward_std": 0.11338547170162201,
"rewards/accuracy_reward": 0.694531238079071,
"rewards/brier_reward": 0.8528412699699401,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9977430582046509,
"rewards/mean_confidence_reward": 0.6812376379966736,
"sampling/batch_mean_priority_error": 0.01960522136886875,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4083333333333334,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0013906704727560282,
"sampling/priority_kl": 0.029997918754816055,
"sampling/priority_scale": 0.708841937663965,
"sampling/prob_entropy": 10.278946876525879,
"sampling/prob_max": 4.5564852916868405e-05,
"sampling/prob_min": 1.9692146815941668e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 1.0032000064849853,
"sampling/prompt_draws_total": 30096.0,
"sampling/seen_fraction": 0.7014533281326294,
"sampling/unseen_fraction": 0.2985466718673706,
"signal/accuracy_reward/centered_abs_mean": 0.128759765625,
"signal/accuracy_reward/group_std_mean": 0.1751173049211502,
"signal/accuracy_reward/group_zero_std_frac": 0.4750000059604645,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0643798828125,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0643798828125,
"signal/advantage_abs_mean": 0.08181026428937913,
"signal/advantage_pre_scale_abs_mean": 0.08181026428937913,
"signal/advantage_pre_scale_std": 0.15451219081878662,
"signal/advantage_std": 0.15451219081878662,
"signal/brier_reward/centered_abs_mean": 0.06762193292379379,
"signal/brier_reward/group_std_mean": 0.0920287773013115,
"signal/brier_reward/group_zero_std_frac": 0.14444444626569747,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.033810966461896894,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.033810966461896894,
"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": 0.004242621548473835,
"signal/format_reward/group_std_mean": 0.010189045500010253,
"signal/format_reward/group_zero_std_frac": 0.950000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0021213107742369176,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0021213107742369176,
"signal/mean_confidence_reward/centered_abs_mean": 0.06683679148554802,
"signal/mean_confidence_reward/group_std_mean": 0.08758924454450608,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1500000014901161,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.683678748231614e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.683678748231614e-07,
"step": 420
},
{
"calibration/aurc": 0.05998141922513901,
"calibration/batch_distribution_entropy": 0.6505124534342099,
"calibration/batch_entropy_100bins": 0.3638051185873787,
"calibration/batch_entropy_10bins": 0.6505124534342099,
"calibration/batch_entropy_50bins": 0.42808093280425225,
"calibration/batch_uniqueness": 0.41772793882104836,
"calibration/confidence_entropy": 0.4729893772005923,
"calibration/coverage@0%": 0.2644555047867711,
"calibration/coverage@1%": 0.4766000153687816,
"calibration/coverage@10%": 0.6602030146987719,
"calibration/coverage@15%": 0.9266926381808889,
"calibration/coverage@20%": 0.9624020887728459,
"calibration/coverage@25%": 1.0,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.600641020659787,
"calibration/distribution_entropy_10": 0.6505124534342099,
"calibration/distribution_entropy_100": 0.3638051185873787,
"calibration/ece": 0.15267265542144076,
"calibration/mean_confidence": 0.7118193110200474,
"calibration/unique_confidence_per_question": 0.0390625,
"calibration/unique_confidences": 15.0,
"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": 0.0043402777777777676,
"completions/max_length": 3774.8,
"completions/max_terminated_length": 3774.8,
"completions/mean_length": 1087.99921875,
"completions/mean_terminated_length": 1092.8842041015625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 337.6,
"epoch": 1.0216346153846154,
"grad_norm": 0.000375980423996225,
"learning_rate": 3.695913461538462e-06,
"loss": -0.004,
"num_tokens": 1010726195.0,
"reward": 1.284477162361145,
"reward_std": 0.10648652762174607,
"rewards/accuracy_reward": 0.712413203716278,
"rewards/brier_reward": 0.8608673214912415,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9956597208976745,
"rewards/mean_confidence_reward": 0.7011353492736816,
"sampling/batch_mean_priority_error": 0.018059459333132105,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4333333333333333,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0014110021525993942,
"sampling/priority_kl": 0.030000055208802223,
"sampling/priority_scale": 0.7089194476371631,
"sampling/prob_entropy": 10.27895622253418,
"sampling/prob_max": 4.5716890599578616e-05,
"sampling/prob_min": 1.9756038454943337e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 1.0152000188827515,
"sampling/prompt_draws_total": 30456.0,
"sampling/seen_fraction": 0.7066600084304809,
"sampling/unseen_fraction": 0.29333999156951907,
"signal/accuracy_reward/centered_abs_mean": 0.1138400599360466,
"signal/accuracy_reward/group_std_mean": 0.15474726110696793,
"signal/accuracy_reward/group_zero_std_frac": 0.5416666746139527,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0569200299680233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0569200299680233,
"signal/advantage_abs_mean": 0.07521377429366112,
"signal/advantage_pre_scale_abs_mean": 0.07521377429366112,
"signal/advantage_pre_scale_std": 0.15179350972175598,
"signal/advantage_std": 0.15179350972175598,
"signal/brier_reward/centered_abs_mean": 0.06321329697966575,
"signal/brier_reward/group_std_mean": 0.08789711743593216,
"signal/brier_reward/group_zero_std_frac": 0.1944444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03160664848983288,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03160664848983288,
"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": 0.00712890645954758,
"signal/format_reward/group_std_mean": 0.014370427234098315,
"signal/format_reward/group_zero_std_frac": 0.9361111283302307,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00356445322977379,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00356445322977379,
"signal/mean_confidence_reward/centered_abs_mean": 0.06138454973697662,
"signal/mean_confidence_reward/group_std_mean": 0.08237865716218948,
"signal/mean_confidence_reward/group_zero_std_frac": 0.19722222685813903,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.138454864412779e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.138454864412779e-07,
"step": 425
},
{
"calibration/aurc": 0.1623026411211937,
"calibration/batch_distribution_entropy": 0.5908128190664673,
"calibration/batch_entropy_100bins": 0.3205678919351199,
"calibration/batch_entropy_10bins": 0.5908128190664673,
"calibration/batch_entropy_50bins": 0.37684439145325405,
"calibration/batch_uniqueness": 0.3022808775676606,
"calibration/confidence_entropy": 0.46237970751720275,
"calibration/coverage@0%": 0.09869451697127937,
"calibration/coverage@1%": 0.10966057441253263,
"calibration/coverage@10%": 0.2581171127067015,
"calibration/coverage@15%": 0.5452921018276763,
"calibration/coverage@20%": 0.694095409051349,
"calibration/coverage@25%": 0.800100630983464,
"calibration/coverage@30%": 0.8564485422106178,
"calibration/coverage@5%": 0.2393671127067015,
"calibration/distribution_entropy_10": 0.5908128190664673,
"calibration/distribution_entropy_100": 0.3205678919351199,
"calibration/ece": 0.11328370918962101,
"calibration/mean_confidence": 0.7135982632765554,
"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": 0.002517361111111116,
"completions/max_length": 3556.4,
"completions/max_terminated_length": 3556.4,
"completions/mean_length": 1057.6259765625,
"completions/mean_terminated_length": 1060.273388671875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 344.4,
"epoch": 1.0336538461538463,
"grad_norm": 0.0003816092503257096,
"learning_rate": 3.665865384615385e-06,
"loss": -0.0019,
"num_tokens": 1026012798.0,
"reward": 1.2823591470718383,
"reward_std": 0.10721758604049683,
"rewards/accuracy_reward": 0.7147569417953491,
"rewards/brier_reward": 0.8524640798568726,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9974826455116272,
"rewards/mean_confidence_reward": 0.7300627708435059,
"sampling/batch_mean_priority_error": 0.014562219208853807,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4138888888888889,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0014308568555861712,
"sampling/priority_kl": 0.02999901808798313,
"sampling/priority_scale": 0.7088448344962671,
"sampling/prob_entropy": 10.278940200805664,
"sampling/prob_max": 4.5860721729695794e-05,
"sampling/prob_min": 1.9819942463072947e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 1.0271999835968018,
"sampling/prompt_draws_total": 30816.0,
"sampling/seen_fraction": 0.7115333437919616,
"sampling/unseen_fraction": 0.28846665620803835,
"signal/accuracy_reward/centered_abs_mean": 0.11008029580116271,
"signal/accuracy_reward/group_std_mean": 0.15001013576984407,
"signal/accuracy_reward/group_zero_std_frac": 0.5583333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05504014790058136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05504014790058136,
"signal/advantage_abs_mean": 0.07600466758012772,
"signal/advantage_pre_scale_abs_mean": 0.07600466758012772,
"signal/advantage_pre_scale_std": 0.15484246909618377,
"signal/advantage_std": 0.15484246909618377,
"signal/brier_reward/centered_abs_mean": 0.06448182314634324,
"signal/brier_reward/group_std_mean": 0.08896796703338623,
"signal/brier_reward/group_zero_std_frac": 0.2583333343267441,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03224091157317162,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03224091157317162,
"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": 0.004833984328433872,
"signal/format_reward/group_std_mean": 0.013044581189751625,
"signal/format_reward/group_zero_std_frac": 0.9305555701255799,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002416992164216936,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002416992164216936,
"signal/mean_confidence_reward/centered_abs_mean": 0.057430972903966905,
"signal/mean_confidence_reward/group_std_mean": 0.07837551683187485,
"signal/mean_confidence_reward/group_zero_std_frac": 0.27777778506278994,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.743097062804736e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.743097062804736e-07,
"step": 430
},
{
"calibration/aurc": 0.1205467953686877,
"calibration/batch_distribution_entropy": 0.5210485050832704,
"calibration/batch_entropy_100bins": 0.28943236080843737,
"calibration/batch_entropy_10bins": 0.5210485050832704,
"calibration/batch_entropy_50bins": 0.34000717947510306,
"calibration/batch_uniqueness": 0.10985326071123118,
"calibration/confidence_entropy": 0.4302312746231255,
"calibration/coverage@0%": 0.15156522687609075,
"calibration/coverage@1%": 0.15469022687609074,
"calibration/coverage@10%": 0.5183311166469629,
"calibration/coverage@15%": 0.6411402916174069,
"calibration/coverage@20%": 0.8527174639700501,
"calibration/coverage@25%": 0.9147707840736363,
"calibration/coverage@30%": 0.9346197151382085,
"calibration/coverage@5%": 0.2399388827900692,
"calibration/distribution_entropy_10": 0.5210485050832704,
"calibration/distribution_entropy_100": 0.28943236080843737,
"calibration/ece": 0.10173666440002979,
"calibration/mean_confidence": 0.7815555736898057,
"calibration/unique_confidence_per_question": 0.03645833333333333,
"calibration/unique_confidences": 14.0,
"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": 0.00425347222222221,
"completions/max_length": 3720.4,
"completions/max_terminated_length": 3720.4,
"completions/mean_length": 1028.5506958007813,
"completions/mean_terminated_length": 1033.01962890625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 328.2,
"epoch": 1.0456730769230769,
"grad_norm": 0.0004367251240182668,
"learning_rate": 3.635817307692308e-06,
"loss": -0.0043,
"num_tokens": 1040972006.0,
"reward": 1.2838960409164428,
"reward_std": 0.12252383381128311,
"rewards/accuracy_reward": 0.7300347208976745,
"rewards/brier_reward": 0.8419958472251892,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9957465171813965,
"rewards/mean_confidence_reward": 0.7535378098487854,
"sampling/batch_mean_priority_error": 0.014893905477138685,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0014472598442807794,
"sampling/priority_kl": 0.03000083900988102,
"sampling/priority_scale": 0.7090173423988745,
"sampling/prob_entropy": 10.278966331481934,
"sampling/prob_max": 4.6011665108380837e-05,
"sampling/prob_min": 1.9881115804309957e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 1.039199995994568,
"sampling/prompt_draws_total": 31176.0,
"sampling/seen_fraction": 0.7165466666221618,
"sampling/unseen_fraction": 0.2834533333778381,
"signal/accuracy_reward/centered_abs_mean": 0.12941623032093047,
"signal/accuracy_reward/group_std_mean": 0.16836948990821837,
"signal/accuracy_reward/group_zero_std_frac": 0.5277777731418609,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06470811516046523,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06470811516046523,
"signal/advantage_abs_mean": 0.09151800125837325,
"signal/advantage_pre_scale_abs_mean": 0.09151800125837325,
"signal/advantage_pre_scale_std": 0.17786999344825744,
"signal/advantage_std": 0.17786999344825744,
"signal/brier_reward/centered_abs_mean": 0.07523893266916275,
"signal/brier_reward/group_std_mean": 0.0994583934545517,
"signal/brier_reward/group_zero_std_frac": 0.27500000298023225,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.037619466334581374,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.037619466334581374,
"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": 0.007373047061264515,
"signal/format_reward/group_std_mean": 0.01584246177226305,
"signal/format_reward/group_zero_std_frac": 0.9277777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0036865235306322575,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0036865235306322575,
"signal/mean_confidence_reward/centered_abs_mean": 0.06189207062125206,
"signal/mean_confidence_reward/group_std_mean": 0.08114207834005356,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2916666716337204,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.189206828821626e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.189206828821626e-07,
"step": 435
},
{
"calibration/aurc": 0.10714301687641507,
"calibration/batch_distribution_entropy": 0.5955961633333724,
"calibration/batch_entropy_100bins": 0.34357802421851297,
"calibration/batch_entropy_10bins": 0.5955961633333724,
"calibration/batch_entropy_50bins": 0.4022762759967714,
"calibration/batch_uniqueness": 0.2600934386043134,
"calibration/confidence_entropy": 0.44799529131081606,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.2546138743455497,
"calibration/coverage@10%": 0.5192162958115183,
"calibration/coverage@15%": 0.7130208333333334,
"calibration/coverage@20%": 0.8552028795811518,
"calibration/coverage@25%": 0.9090095986038396,
"calibration/coverage@30%": 0.9408376963350786,
"calibration/coverage@5%": 0.46792375654450263,
"calibration/distribution_entropy_10": 0.5955961633333724,
"calibration/distribution_entropy_100": 0.34357802421851297,
"calibration/ece": 0.11402197692917163,
"calibration/mean_confidence": 0.7536124024710045,
"calibration/unique_confidence_per_question": 0.0453125,
"calibration/unique_confidences": 17.4,
"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": 3729.4,
"completions/max_terminated_length": 3729.4,
"completions/mean_length": 1017.2171997070312,
"completions/mean_terminated_length": 1020.3733520507812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 311.2,
"epoch": 1.0576923076923077,
"grad_norm": 0.00039360576192848384,
"learning_rate": 3.605769230769231e-06,
"loss": -0.0023,
"num_tokens": 1055791948.0,
"reward": 1.2873583793640138,
"reward_std": 0.1034932404756546,
"rewards/accuracy_reward": 0.71875,
"rewards/brier_reward": 0.8590771198272705,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.996875,
"rewards/mean_confidence_reward": 0.7335354685783386,
"sampling/batch_mean_priority_error": 0.014799672360578007,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3805555555555556,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0014623381663113832,
"sampling/priority_kl": 0.03000078722834587,
"sampling/priority_scale": 0.7089662969810888,
"sampling/prob_entropy": 10.278951263427734,
"sampling/prob_max": 4.6152883442118764e-05,
"sampling/prob_min": 1.99433427042095e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 1.051199984550476,
"sampling/prompt_draws_total": 31536.0,
"sampling/seen_fraction": 0.7211666703224182,
"sampling/unseen_fraction": 0.2788333296775818,
"signal/accuracy_reward/centered_abs_mean": 0.10870225876569747,
"signal/accuracy_reward/group_std_mean": 0.14185952842235566,
"signal/accuracy_reward/group_zero_std_frac": 0.600000011920929,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05435112938284874,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05435112938284874,
"signal/advantage_abs_mean": 0.07555654495954514,
"signal/advantage_pre_scale_abs_mean": 0.07555654495954514,
"signal/advantage_pre_scale_std": 0.1532256066799164,
"signal/advantage_std": 0.1532256066799164,
"signal/brier_reward/centered_abs_mean": 0.06442730501294136,
"signal/brier_reward/group_std_mean": 0.08673342168331147,
"signal/brier_reward/group_zero_std_frac": 0.23055555820465087,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03221365250647068,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03221365250647068,
"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": 0.005946180550381541,
"signal/format_reward/group_std_mean": 0.015451083704829217,
"signal/format_reward/group_zero_std_frac": 0.919444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0029730902751907706,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0029730902751907706,
"signal/mean_confidence_reward/centered_abs_mean": 0.061560317128896716,
"signal/mean_confidence_reward/group_std_mean": 0.08096567541360855,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2361111134290695,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.156031759019242e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.156031759019242e-07,
"step": 440
},
{
"calibration/aurc": 0.15670651247509326,
"calibration/batch_distribution_entropy": 0.6795985620528495,
"calibration/batch_entropy_100bins": 0.39968524286734586,
"calibration/batch_entropy_10bins": 0.6795985620528495,
"calibration/batch_entropy_50bins": 0.47013045656830627,
"calibration/batch_uniqueness": 0.4595951769139866,
"calibration/confidence_entropy": 0.48444565435591047,
"calibration/coverage@0%": 0.10548302872062662,
"calibration/coverage@1%": 0.12532637075718017,
"calibration/coverage@10%": 0.38832387030820453,
"calibration/coverage@15%": 0.5937442798776023,
"calibration/coverage@20%": 0.6182411249602829,
"calibration/coverage@25%": 0.8065749029521875,
"calibration/coverage@30%": 0.8878541727105695,
"calibration/coverage@5%": 0.1869341180046141,
"calibration/distribution_entropy_10": 0.6795985620528495,
"calibration/distribution_entropy_100": 0.39968524286734586,
"calibration/ece": 0.11142151772778952,
"calibration/mean_confidence": 0.6838090146706527,
"calibration/unique_confidence_per_question": 0.04791666666666666,
"calibration/unique_confidences": 18.4,
"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": 0.002951388888888884,
"completions/max_length": 3774.6,
"completions/max_terminated_length": 3774.6,
"completions/mean_length": 1003.1759521484375,
"completions/mean_terminated_length": 1006.1692138671875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 313.0,
"epoch": 1.0697115384615385,
"grad_norm": 0.00040377015830017626,
"learning_rate": 3.575721153846154e-06,
"loss": -0.0024,
"num_tokens": 1070414551.0,
"reward": 1.3242327213287353,
"reward_std": 0.0985756278038025,
"rewards/accuracy_reward": 0.7759548544883728,
"rewards/brier_reward": 0.8754475831985473,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9970486164093018,
"rewards/mean_confidence_reward": 0.7241761922836304,
"sampling/batch_mean_priority_error": 0.011732953869047609,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3472222222222222,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0014753591967746615,
"sampling/priority_kl": 0.030000388249754907,
"sampling/priority_scale": 0.7085408985847608,
"sampling/prob_entropy": 10.278962707519531,
"sampling/prob_max": 4.628084570867941e-05,
"sampling/prob_min": 1.9537291882443243e-05,
"sampling/prompt_draws_max": 5.6,
"sampling/prompt_draws_mean": 1.0631999969482422,
"sampling/prompt_draws_total": 31896.0,
"sampling/seen_fraction": 0.7253000020980835,
"sampling/unseen_fraction": 0.2746999979019165,
"signal/accuracy_reward/centered_abs_mean": 0.10059136152267456,
"signal/accuracy_reward/group_std_mean": 0.13695140928030014,
"signal/accuracy_reward/group_zero_std_frac": 0.5916666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05029568076133728,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05029568076133728,
"signal/advantage_abs_mean": 0.06849844828248024,
"signal/advantage_pre_scale_abs_mean": 0.06849844828248024,
"signal/advantage_pre_scale_std": 0.1458652526140213,
"signal/advantage_std": 0.1458652526140213,
"signal/brier_reward/centered_abs_mean": 0.05644915029406548,
"signal/brier_reward/group_std_mean": 0.07987928092479706,
"signal/brier_reward/group_zero_std_frac": 0.15000000298023225,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02822457514703274,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02822457514703274,
"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": 0.005598958255723119,
"signal/format_reward/group_std_mean": 0.013899841345846653,
"signal/format_reward/group_zero_std_frac": 0.9305555701255799,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0027994791278615596,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0027994791278615596,
"signal/mean_confidence_reward/centered_abs_mean": 0.05799039080739021,
"signal/mean_confidence_reward/group_std_mean": 0.07693097293376923,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1583333358168602,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.79903905872925e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.79903905872925e-07,
"step": 445
},
{
"calibration/aurc": 0.19829493130072745,
"calibration/batch_distribution_entropy": 0.6906534768349407,
"calibration/batch_entropy_100bins": 0.4124322883013575,
"calibration/batch_entropy_10bins": 0.6906534768349407,
"calibration/batch_entropy_50bins": 0.48449631588350217,
"calibration/batch_uniqueness": 0.4989087204350981,
"calibration/confidence_entropy": 0.49707835632558944,
"calibration/coverage@0%": 0.03172123015873016,
"calibration/coverage@1%": 0.03172123015873016,
"calibration/coverage@10%": 0.26872519841269843,
"calibration/coverage@15%": 0.5488925212473441,
"calibration/coverage@20%": 0.5915760529933758,
"calibration/coverage@25%": 0.5942215556388785,
"calibration/coverage@30%": 0.7055076448777237,
"calibration/coverage@5%": 0.2476686507936508,
"calibration/distribution_entropy_10": 0.6906534768349407,
"calibration/distribution_entropy_100": 0.4124322883013575,
"calibration/ece": 0.11379584590896452,
"calibration/mean_confidence": 0.6684599442887945,
"calibration/unique_confidence_per_question": 0.04687500000000001,
"calibration/unique_confidences": 18.0,
"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": 0.006076388888888906,
"completions/max_length": 3719.4,
"completions/max_terminated_length": 3719.4,
"completions/mean_length": 1025.4586059570313,
"completions/mean_terminated_length": 1031.8718627929688,
"completions/min_length": 0.0,
"completions/min_terminated_length": 332.6,
"epoch": 1.0817307692307692,
"grad_norm": 0.00046888578799553216,
"learning_rate": 3.5456730769230774e-06,
"loss": -0.0058,
"num_tokens": 1085298394.0,
"reward": 1.3057713031768798,
"reward_std": 0.11162674725055695,
"rewards/accuracy_reward": 0.7561631917953491,
"rewards/brier_reward": 0.8614419937133789,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9939236164093017,
"rewards/mean_confidence_reward": 0.6888800263404846,
"sampling/batch_mean_priority_error": 0.01921851300705466,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3777777777777778,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0014933944679796697,
"sampling/priority_kl": 0.030000264942646026,
"sampling/priority_scale": 0.7086831629509106,
"sampling/prob_entropy": 10.278955841064453,
"sampling/prob_max": 4.6429499343503265e-05,
"sampling/prob_min": 1.9281814093119465e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.0752000093460083,
"sampling/prompt_draws_total": 32256.0,
"sampling/seen_fraction": 0.7299266695976258,
"sampling/unseen_fraction": 0.27007333040237425,
"signal/accuracy_reward/centered_abs_mean": 0.11822374314069747,
"signal/accuracy_reward/group_std_mean": 0.15489685237407685,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111223697662,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05911187157034874,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05911187157034874,
"signal/advantage_abs_mean": 0.07847933173179626,
"signal/advantage_pre_scale_abs_mean": 0.07847933173179626,
"signal/advantage_pre_scale_std": 0.16222485899925232,
"signal/advantage_std": 0.16222485899925232,
"signal/brier_reward/centered_abs_mean": 0.06763920560479164,
"signal/brier_reward/group_std_mean": 0.09310382902622223,
"signal/brier_reward/group_zero_std_frac": 0.11388889253139496,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03381960280239582,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03381960280239582,
"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": 0.011338975839316845,
"signal/format_reward/group_std_mean": 0.026788859441876412,
"signal/format_reward/group_zero_std_frac": 0.8694444537162781,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005669487919658422,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005669487919658422,
"signal/mean_confidence_reward/centered_abs_mean": 0.06197234690189361,
"signal/mean_confidence_reward/group_std_mean": 0.08344518542289733,
"signal/mean_confidence_reward/group_zero_std_frac": 0.11388889253139496,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.197234370119986e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.197234370119986e-07,
"step": 450
},
{
"epoch": 1.0817307692307692,
"eval_calibration/aurc": 0.13411705755388098,
"eval_calibration/batch_distribution_entropy": 0.7462482530200102,
"eval_calibration/batch_entropy_100bins": 0.42241674770712895,
"eval_calibration/batch_entropy_10bins": 0.7462482530200102,
"eval_calibration/batch_entropy_50bins": 0.496558810090405,
"eval_calibration/batch_uniqueness": 0.600966601402125,
"eval_calibration/confidence_entropy": 0.4965210712316667,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.544259421560035,
"eval_calibration/coverage@15%": 0.6985100788781771,
"eval_calibration/coverage@20%": 0.7572304995617879,
"eval_calibration/coverage@25%": 0.8632778264680105,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.7462482530200102,
"eval_calibration/distribution_entropy_100": 0.42241674770712895,
"eval_calibration/ece": 0.041625847129790935,
"eval_calibration/mean_confidence": 0.6837050591827103,
"eval_calibration/unique_confidence_per_question": 0.016493055555555556,
"eval_calibration/unique_confidences": 19,
"eval_completions/clipped_ratio": 0.006944444444444438,
"eval_completions/max_length": 2698.0,
"eval_completions/max_terminated_length": 2698.0,
"eval_completions/mean_length": 1044.328837076823,
"eval_completions/mean_terminated_length": 1051.6819254557292,
"eval_completions/min_length": 155.16666666666666,
"eval_completions/min_terminated_length": 400.6666666666667,
"eval_loss": 0.0,
"eval_num_tokens": 1085298394.0,
"eval_reward": 1.2724303007125854,
"eval_reward_std": 0.3061056633790334,
"eval_rewards/accuracy_reward": 0.7048611044883728,
"eval_rewards/brier_reward": 0.8495345413684845,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9904513855775198,
"eval_rewards/mean_confidence_reward": 0.6771766245365143,
"eval_runtime": 201.6774,
"eval_samples_per_second": 4.958,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4029947866996129,
"eval_signal/accuracy_reward/group_std_mean": 0.45439985394477844,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20149739334980646,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.20149739334980646,
"eval_signal/advantage_abs_mean": 0.2554394875963529,
"eval_signal/advantage_pre_scale_abs_mean": 0.2554394875963529,
"eval_signal/advantage_pre_scale_std": 0.3054951975742976,
"eval_signal/advantage_std": 0.3054951975742976,
"eval_signal/brier_reward/centered_abs_mean": 0.15039577335119247,
"eval_signal/brier_reward/group_std_mean": 0.20635540535052618,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07519788667559624,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.07519788667559624,
"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": 0.018391926928112905,
"eval_signal/format_reward/group_std_mean": 0.051025692063073315,
"eval_signal/format_reward/group_zero_std_frac": 0.722222238779068,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.009195963464056453,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.009195963464056453,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.20515539248784384,
"eval_signal/mean_confidence_reward/group_std_mean": 0.2403454432884852,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.05155390631262e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.05155390631262e-06,
"eval_steps_per_second": 0.03,
"step": 450
},
{
"epoch": 1.0817307692307692,
"step": 450,
"train_probe_calibration/aurc": 0.1033654966199605,
"train_probe_calibration/batch_distribution_entropy": 0.7499117856336898,
"train_probe_calibration/batch_entropy_100bins": 0.42785477507255537,
"train_probe_calibration/batch_entropy_10bins": 0.7499117856336898,
"train_probe_calibration/batch_entropy_50bins": 0.5023441304265334,
"train_probe_calibration/batch_uniqueness": 0.5832717759310557,
"train_probe_calibration/confidence_entropy": 0.4965396364266109,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.6482456140350877,
"train_probe_calibration/coverage@15%": 0.7666666666666667,
"train_probe_calibration/coverage@20%": 0.8780701754385964,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.39649122807017545,
"train_probe_calibration/distribution_entropy_10": 0.7499117856336898,
"train_probe_calibration/distribution_entropy_100": 0.42785477507255537,
"train_probe_calibration/ece": 0.07051012742617996,
"train_probe_calibration/mean_confidence": 0.6876272994744047,
"train_probe_calibration/unique_confidence_per_question": 0.024305555555555556,
"train_probe_calibration/unique_confidences": 28,
"train_probe_completions/clipped_ratio": 0.007812500000000019,
"train_probe_completions/max_length": 2930.0,
"train_probe_completions/max_terminated_length": 2930.0,
"train_probe_completions/mean_length": 1057.2571411132812,
"train_probe_completions/mean_terminated_length": 1065.5167032877605,
"train_probe_completions/min_length": 54.166666666666664,
"train_probe_completions/min_terminated_length": 351.0,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1085298394.0,
"train_probe_reward": 1.2937801281611125,
"train_probe_reward_std": 0.29190226395924884,
"train_probe_rewards/accuracy_reward": 0.7465277711550394,
"train_probe_rewards/brier_reward": 0.8514354924360911,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9895833432674408,
"train_probe_rewards/mean_confidence_reward": 0.6804644962151846,
"train_probe_runtime": 211.8545,
"train_probe_samples_per_second": 4.72,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3678385416666667,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4330555697282155,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.18391927083333334,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.18391927083333334,
"train_probe_signal/advantage_abs_mean": 0.23508351047833762,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.23508351047833762,
"train_probe_signal/advantage_pre_scale_std": 0.2930772602558136,
"train_probe_signal/advantage_std": 0.2930772602558136,
"train_probe_signal/brier_reward/centered_abs_mean": 0.14737369120121002,
"train_probe_signal/brier_reward/group_std_mean": 0.19736657788356146,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07368684560060501,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07368684560060501,
"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": 0.019748263837148745,
"train_probe_signal/format_reward/group_std_mean": 0.049431003319720425,
"train_probe_signal/format_reward/group_zero_std_frac": 0.7500000298023224,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.009874131918574372,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.009874131918574372,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.20315797378619513,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.23725442836682,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.031579716306927e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.031579716306927e-06,
"train_probe_steps_per_second": 0.028
},
{
"calibration/aurc": 0.13629810162538877,
"calibration/batch_distribution_entropy": 0.7171101453541321,
"calibration/batch_entropy_100bins": 0.40900396610625045,
"calibration/batch_entropy_10bins": 0.7171101453541321,
"calibration/batch_entropy_50bins": 0.4814728512715501,
"calibration/batch_uniqueness": 0.5450898760297638,
"calibration/confidence_entropy": 0.49317485953668533,
"calibration/coverage@0%": 0.051458885941644564,
"calibration/coverage@1%": 0.0843501326259947,
"calibration/coverage@10%": 0.4297356031906254,
"calibration/coverage@15%": 0.6599059517453104,
"calibration/coverage@20%": 0.7393530701754385,
"calibration/coverage@25%": 0.9118708006718925,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.14801061007957558,
"calibration/distribution_entropy_10": 0.7171101453541321,
"calibration/distribution_entropy_100": 0.40900396610625045,
"calibration/ece": 0.12500761844623415,
"calibration/mean_confidence": 0.6974867715920466,
"calibration/unique_confidence_per_question": 0.03802083333333333,
"calibration/unique_confidences": 14.6,
"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": 0.0109375,
"completions/max_length": 3788.0,
"completions/max_terminated_length": 3788.0,
"completions/mean_length": 1065.3917724609375,
"completions/mean_terminated_length": 1077.2124877929687,
"completions/min_length": 0.0,
"completions/min_terminated_length": 335.6,
"epoch": 1.09375,
"grad_norm": 0.0005689400713890791,
"learning_rate": 3.5156250000000003e-06,
"loss": -0.0114,
"num_tokens": 1100698651.0,
"reward": 1.282351565361023,
"reward_std": 0.14021986573934556,
"rewards/accuracy_reward": 0.732812511920929,
"rewards/brier_reward": 0.842814815044403,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.989062511920929,
"rewards/mean_confidence_reward": 0.6714834690093994,
"sampling/batch_mean_priority_error": 0.016140534315776323,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3555555555555555,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0015142235439270734,
"sampling/priority_kl": 0.03000020533800125,
"sampling/priority_scale": 0.7086563527351245,
"sampling/prob_entropy": 10.278951644897461,
"sampling/prob_max": 4.657171957660467e-05,
"sampling/prob_min": 1.9341520965099334e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.0871999979019165,
"sampling/prompt_draws_total": 32616.0,
"sampling/seen_fraction": 0.7342666625976563,
"sampling/unseen_fraction": 0.26573333740234373,
"signal/accuracy_reward/centered_abs_mean": 0.1413736969232559,
"signal/accuracy_reward/group_std_mean": 0.1889444947242737,
"signal/accuracy_reward/group_zero_std_frac": 0.450000011920929,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07068684846162795,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07068684846162795,
"signal/advantage_abs_mean": 0.09725596755743027,
"signal/advantage_pre_scale_abs_mean": 0.09725596755743027,
"signal/advantage_pre_scale_std": 0.18758226335048675,
"signal/advantage_std": 0.18758226335048675,
"signal/brier_reward/centered_abs_mean": 0.08304826766252518,
"signal/brier_reward/group_std_mean": 0.11446874588727951,
"signal/brier_reward/group_zero_std_frac": 0.0972222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04152413383126259,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04152413383126259,
"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": 0.01982421875,
"signal/format_reward/group_std_mean": 0.04075228720903397,
"signal/format_reward/group_zero_std_frac": 0.825000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009912109375,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009912109375,
"signal/mean_confidence_reward/centered_abs_mean": 0.072690649330616,
"signal/mean_confidence_reward/group_std_mean": 0.09616097360849381,
"signal/mean_confidence_reward/group_zero_std_frac": 0.10555555671453476,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.269064440151851e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.269064440151851e-07,
"step": 455
},
{
"calibration/aurc": 0.10542953162029697,
"calibration/batch_distribution_entropy": 0.6119959945602249,
"calibration/batch_entropy_100bins": 0.34331494201870555,
"calibration/batch_entropy_10bins": 0.6119959945602249,
"calibration/batch_entropy_50bins": 0.4041447949551973,
"calibration/batch_uniqueness": 0.38240274912033423,
"calibration/confidence_entropy": 0.46975072875067064,
"calibration/coverage@0%": 0.11705626640419946,
"calibration/coverage@1%": 0.2493479330708661,
"calibration/coverage@10%": 0.6614818330000416,
"calibration/coverage@15%": 0.7343451860184145,
"calibration/coverage@20%": 0.7878556847060784,
"calibration/coverage@25%": 0.8256509602966297,
"calibration/coverage@30%": 0.8790442861309004,
"calibration/coverage@5%": 0.5737981320043328,
"calibration/distribution_entropy_10": 0.6119959945602249,
"calibration/distribution_entropy_100": 0.34331494201870555,
"calibration/ece": 0.1093204766721867,
"calibration/mean_confidence": 0.6956372109085324,
"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": 0.0065972222222221875,
"completions/max_length": 3885.8,
"completions/max_terminated_length": 3885.8,
"completions/mean_length": 1045.6061767578126,
"completions/mean_terminated_length": 1052.6762451171876,
"completions/min_length": 0.0,
"completions/min_terminated_length": 356.4,
"epoch": 1.1057692307692308,
"grad_norm": 0.0004963566898368299,
"learning_rate": 3.4855769230769233e-06,
"loss": -0.0065,
"num_tokens": 1115864770.0,
"reward": 1.297843074798584,
"reward_std": 0.11444427967071533,
"rewards/accuracy_reward": 0.7450520753860473,
"rewards/brier_reward": 0.8572170615196228,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9934027791023254,
"rewards/mean_confidence_reward": 0.7170493602752686,
"sampling/batch_mean_priority_error": 0.01892611882716048,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3472222222222222,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.001534794899635017,
"sampling/priority_kl": 0.029999419301748275,
"sampling/priority_scale": 0.7087399541633204,
"sampling/prob_entropy": 10.278960227966309,
"sampling/prob_max": 4.671617498388514e-05,
"sampling/prob_min": 1.9399503071326762e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.0992000102996826,
"sampling/prompt_draws_total": 32976.0,
"sampling/seen_fraction": 0.7386133313179016,
"sampling/unseen_fraction": 0.26138666868209837,
"signal/accuracy_reward/centered_abs_mean": 0.10742730051279067,
"signal/accuracy_reward/group_std_mean": 0.14884938895702363,
"signal/accuracy_reward/group_zero_std_frac": 0.547222226858139,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05371365025639534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05371365025639534,
"signal/advantage_abs_mean": 0.07775596380233765,
"signal/advantage_pre_scale_abs_mean": 0.07775596380233765,
"signal/advantage_pre_scale_std": 0.16309164762496947,
"signal/advantage_std": 0.16309164762496947,
"signal/brier_reward/centered_abs_mean": 0.07044132426381111,
"signal/brier_reward/group_std_mean": 0.09729270190000534,
"signal/brier_reward/group_zero_std_frac": 0.1833333343267441,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.035220662131905556,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.035220662131905556,
"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": 0.012337239738553763,
"signal/format_reward/group_std_mean": 0.028919679298996925,
"signal/format_reward/group_zero_std_frac": 0.8611111164093017,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0061686198692768816,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0061686198692768816,
"signal/mean_confidence_reward/centered_abs_mean": 0.06461131945252419,
"signal/mean_confidence_reward/group_std_mean": 0.08675065785646438,
"signal/mean_confidence_reward/group_zero_std_frac": 0.19166666865348816,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.461131533797016e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.461131533797016e-07,
"step": 460
},
{
"calibration/aurc": 0.09359492378663284,
"calibration/batch_distribution_entropy": 0.6444820041232762,
"calibration/batch_entropy_100bins": 0.34567299261889833,
"calibration/batch_entropy_10bins": 0.6444820041232762,
"calibration/batch_entropy_50bins": 0.4069206539688053,
"calibration/batch_uniqueness": 0.4058231469905339,
"calibration/confidence_entropy": 0.46850383484094343,
"calibration/coverage@0%": 0.08489583333333334,
"calibration/coverage@1%": 0.25885416666666666,
"calibration/coverage@10%": 0.5697916666666666,
"calibration/coverage@15%": 0.9046960159003872,
"calibration/coverage@20%": 0.9368139668557263,
"calibration/coverage@25%": 0.9630606860158311,
"calibration/coverage@30%": 0.9889182058047494,
"calibration/coverage@5%": 0.5152302631578947,
"calibration/distribution_entropy_10": 0.6444820041232762,
"calibration/distribution_entropy_100": 0.34567299261889833,
"calibration/ece": 0.1616714965172559,
"calibration/mean_confidence": 0.7261069838474573,
"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": 0.004079861111111116,
"completions/max_length": 3468.8,
"completions/max_terminated_length": 3468.8,
"completions/mean_length": 979.820849609375,
"completions/mean_terminated_length": 983.8570434570313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 310.4,
"epoch": 1.1177884615384615,
"grad_norm": 0.0004011022101622075,
"learning_rate": 3.4555288461538466e-06,
"loss": -0.0027,
"num_tokens": 1130241874.0,
"reward": 1.30813148021698,
"reward_std": 0.098023721575737,
"rewards/accuracy_reward": 0.7587673664093018,
"rewards/brier_reward": 0.8615612864494324,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9959201335906982,
"rewards/mean_confidence_reward": 0.707803475856781,
"sampling/batch_mean_priority_error": 0.019616319444444426,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.32499999999999996,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0015541198663413525,
"sampling/priority_kl": 0.029999390617012976,
"sampling/priority_scale": 0.7084668695228175,
"sampling/prob_entropy": 10.278964424133301,
"sampling/prob_max": 4.684552040998824e-05,
"sampling/prob_min": 1.945980329765007e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.1112000226974488,
"sampling/prompt_draws_total": 33336.0,
"sampling/seen_fraction": 0.7424600005149842,
"sampling/unseen_fraction": 0.2575399994850159,
"signal/accuracy_reward/centered_abs_mean": 0.10092773586511612,
"signal/accuracy_reward/group_std_mean": 0.13783197551965715,
"signal/accuracy_reward/group_zero_std_frac": 0.5916666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05046386793255806,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05046386793255806,
"signal/advantage_abs_mean": 0.06758801788091659,
"signal/advantage_pre_scale_abs_mean": 0.06758801788091659,
"signal/advantage_pre_scale_std": 0.1427416533231735,
"signal/advantage_std": 0.1427416533231735,
"signal/brier_reward/centered_abs_mean": 0.06408771499991417,
"signal/brier_reward/group_std_mean": 0.08897604048252106,
"signal/brier_reward/group_zero_std_frac": 0.18055555522441863,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.032043857499957085,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.032043857499957085,
"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": 0.007687717024236917,
"signal/format_reward/group_std_mean": 0.018355799838900566,
"signal/format_reward/group_zero_std_frac": 0.9111111283302307,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0038438585121184587,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0038438585121184587,
"signal/mean_confidence_reward/centered_abs_mean": 0.06441164910793304,
"signal/mean_confidence_reward/group_std_mean": 0.08494757264852523,
"signal/mean_confidence_reward/group_zero_std_frac": 0.18333333134651184,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.44116460080113e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.44116460080113e-07,
"step": 465
},
{
"calibration/aurc": 0.14959051893570235,
"calibration/batch_distribution_entropy": 0.6709319069566796,
"calibration/batch_entropy_100bins": 0.3686111766059616,
"calibration/batch_entropy_10bins": 0.6709319069566796,
"calibration/batch_entropy_50bins": 0.43392311302166864,
"calibration/batch_uniqueness": 0.47061380252948803,
"calibration/confidence_entropy": 0.4698123180967967,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.11279373368146214,
"calibration/coverage@10%": 0.5033365397828775,
"calibration/coverage@15%": 0.5956836608492511,
"calibration/coverage@20%": 0.62553112546379,
"calibration/coverage@25%": 0.7751461988304094,
"calibration/coverage@30%": 0.8230409356725147,
"calibration/coverage@5%": 0.3579304658513124,
"calibration/distribution_entropy_10": 0.6709319069566796,
"calibration/distribution_entropy_100": 0.3686111766059616,
"calibration/ece": 0.10774334189113621,
"calibration/mean_confidence": 0.6822683751007925,
"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": 0.005642361111111094,
"completions/max_length": 3636.0,
"completions/max_terminated_length": 3636.0,
"completions/mean_length": 1004.15703125,
"completions/mean_terminated_length": 1009.826318359375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 334.2,
"epoch": 1.1298076923076923,
"grad_norm": 0.0005025005084462464,
"learning_rate": 3.4254807692307695e-06,
"loss": -0.0059,
"num_tokens": 1144918851.0,
"reward": 1.2935538053512574,
"reward_std": 0.10897820442914963,
"rewards/accuracy_reward": 0.7296875,
"rewards/brier_reward": 0.8633089423179626,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9940972089767456,
"rewards/mean_confidence_reward": 0.6967671632766723,
"sampling/batch_mean_priority_error": 0.018480902777777763,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.38333333333333336,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0015776523388922215,
"sampling/priority_kl": 0.029999713972210885,
"sampling/priority_scale": 0.7086878955131397,
"sampling/prob_entropy": 10.27896728515625,
"sampling/prob_max": 4.699049750342965e-05,
"sampling/prob_min": 1.9505477393977343e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.123200011253357,
"sampling/prompt_draws_total": 33696.0,
"sampling/seen_fraction": 0.7466799855232239,
"sampling/unseen_fraction": 0.2533200144767761,
"signal/accuracy_reward/centered_abs_mean": 0.1121744766831398,
"signal/accuracy_reward/group_std_mean": 0.14886936098337172,
"signal/accuracy_reward/group_zero_std_frac": 0.5666666626930237,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0560872383415699,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0560872383415699,
"signal/advantage_abs_mean": 0.07866804748773575,
"signal/advantage_pre_scale_abs_mean": 0.07866804748773575,
"signal/advantage_pre_scale_std": 0.1617803692817688,
"signal/advantage_std": 0.1617803692817688,
"signal/brier_reward/centered_abs_mean": 0.07295527458190917,
"signal/brier_reward/group_std_mean": 0.09744260013103485,
"signal/brier_reward/group_zero_std_frac": 0.20833333134651183,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03647763729095459,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03647763729095459,
"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": 0.01085069440305233,
"signal/format_reward/group_std_mean": 0.023369645327329637,
"signal/format_reward/group_zero_std_frac": 0.8944444537162781,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005425347201526165,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005425347201526165,
"signal/mean_confidence_reward/centered_abs_mean": 0.0644960232079029,
"signal/mean_confidence_reward/group_std_mean": 0.08588441014289856,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2194444477558136,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.449602551583667e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.449602551583667e-07,
"step": 470
},
{
"calibration/aurc": 0.20161512188607883,
"calibration/batch_distribution_entropy": 0.5796123879186574,
"calibration/batch_entropy_100bins": 0.3144117230666262,
"calibration/batch_entropy_10bins": 0.5796123879186574,
"calibration/batch_entropy_50bins": 0.37012039325497403,
"calibration/batch_uniqueness": 0.25974887218320325,
"calibration/confidence_entropy": 0.4370606776018837,
"calibration/coverage@0%": 0.0005249343832020997,
"calibration/coverage@1%": 0.0005249343832020997,
"calibration/coverage@10%": 0.1614624343832021,
"calibration/coverage@15%": 0.4038601372813059,
"calibration/coverage@20%": 0.6582004721668278,
"calibration/coverage@25%": 0.7330276926872391,
"calibration/coverage@30%": 0.7796344647519582,
"calibration/coverage@5%": 0.13750410104986877,
"calibration/distribution_entropy_10": 0.5796123879186574,
"calibration/distribution_entropy_100": 0.3144117230666262,
"calibration/ece": 0.14319296054480074,
"calibration/mean_confidence": 0.7470921170078164,
"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": 0.007638888888888906,
"completions/max_length": 3197.4,
"completions/max_terminated_length": 3197.4,
"completions/mean_length": 962.5202270507813,
"completions/mean_terminated_length": 969.9363037109375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 334.6,
"epoch": 1.1418269230769231,
"grad_norm": 0.000619587255641818,
"learning_rate": 3.3954326923076925e-06,
"loss": -0.0078,
"num_tokens": 1159110892.0,
"reward": 1.2515861749649049,
"reward_std": 0.1271716073155403,
"rewards/accuracy_reward": 0.6802951335906983,
"rewards/brier_reward": 0.8305015563964844,
"rewards/confidence_one_or_zero": 0.0007812500232830643,
"rewards/format_reward": 0.9923611164093018,
"rewards/mean_confidence_reward": 0.7246870160102844,
"sampling/batch_mean_priority_error": 0.01913888888888888,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.33888888888888885,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0015997483395040035,
"sampling/priority_kl": 0.03000010587275028,
"sampling/priority_scale": 0.7091093718772754,
"sampling/prob_entropy": 10.278954887390137,
"sampling/prob_max": 4.7144032578216866e-05,
"sampling/prob_min": 1.9522143702488393e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.1351999759674072,
"sampling/prompt_draws_total": 34056.0,
"sampling/seen_fraction": 0.7509999871253967,
"sampling/unseen_fraction": 0.24900001287460327,
"signal/accuracy_reward/centered_abs_mean": 0.11660698801279068,
"signal/accuracy_reward/group_std_mean": 0.16022705137729645,
"signal/accuracy_reward/group_zero_std_frac": 0.5194444417953491,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05830349400639534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05830349400639534,
"signal/advantage_abs_mean": 0.08898119479417801,
"signal/advantage_pre_scale_abs_mean": 0.08898119479417801,
"signal/advantage_pre_scale_std": 0.1809540420770645,
"signal/advantage_std": 0.1809540420770645,
"signal/brier_reward/centered_abs_mean": 0.08139911442995071,
"signal/brier_reward/group_std_mean": 0.10895915925502778,
"signal/brier_reward/group_zero_std_frac": 0.23055555522441865,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.040699557214975354,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.040699557214975354,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0014485677238553762,
"signal/confidence_one_or_zero/group_std_mean": 0.0031183868646621703,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111164093017,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.448567594763972e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.448567594763972e-08,
"signal/format_reward/centered_abs_mean": 0.014257812313735485,
"signal/format_reward/group_std_mean": 0.033097638934850696,
"signal/format_reward/group_zero_std_frac": 0.8416666865348816,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007128906156867743,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007128906156867743,
"signal/mean_confidence_reward/centered_abs_mean": 0.06632542088627816,
"signal/mean_confidence_reward/group_std_mean": 0.08892321586608887,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2416666716337204,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.632541840190243e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.632541840190243e-07,
"step": 475
},
{
"calibration/aurc": 0.1505559280296241,
"calibration/batch_distribution_entropy": 0.7007643129699044,
"calibration/batch_entropy_100bins": 0.3944490515473666,
"calibration/batch_entropy_10bins": 0.7007643129699044,
"calibration/batch_entropy_50bins": 0.46433904134937665,
"calibration/batch_uniqueness": 0.5243341324771948,
"calibration/confidence_entropy": 0.47274465400556076,
"calibration/coverage@0%": 0.03875456576162474,
"calibration/coverage@1%": 0.03875456576162474,
"calibration/coverage@10%": 0.4747473679080457,
"calibration/coverage@15%": 0.5356127446310162,
"calibration/coverage@20%": 0.6051703213524231,
"calibration/coverage@25%": 0.8049892598336099,
"calibration/coverage@30%": 0.8858638743455497,
"calibration/coverage@5%": 0.2838690084455597,
"calibration/distribution_entropy_10": 0.7007643129699044,
"calibration/distribution_entropy_100": 0.3944490515473666,
"calibration/ece": 0.10062478087275668,
"calibration/mean_confidence": 0.6347946440611388,
"calibration/unique_confidence_per_question": 0.025520833333333336,
"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": 0.004253472222222188,
"completions/max_length": 3582.4,
"completions/max_terminated_length": 3582.4,
"completions/mean_length": 957.6535766601562,
"completions/mean_terminated_length": 961.7112426757812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 292.0,
"epoch": 1.1538461538461537,
"grad_norm": 0.0004493494925554842,
"learning_rate": 3.365384615384616e-06,
"loss": -0.0035,
"num_tokens": 1173217301.0,
"reward": 1.29926335811615,
"reward_std": 0.11280029714107513,
"rewards/accuracy_reward": 0.7427951335906983,
"rewards/brier_reward": 0.8599711418151855,
"rewards/confidence_one_or_zero": 0.002777777798473835,
"rewards/format_reward": 0.9957465171813965,
"rewards/mean_confidence_reward": 0.6893273830413819,
"sampling/batch_mean_priority_error": 0.018572631173611093,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3416666666666667,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0016196536598727107,
"sampling/priority_kl": 0.03000083565711975,
"sampling/priority_scale": 0.7094456852180884,
"sampling/prob_entropy": 10.27894229888916,
"sampling/prob_max": 4.72917192382738e-05,
"sampling/prob_min": 1.9575109763536603e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.1471999883651733,
"sampling/prompt_draws_total": 34416.0,
"sampling/seen_fraction": 0.7551133275032044,
"sampling/unseen_fraction": 0.24488667249679566,
"signal/accuracy_reward/centered_abs_mean": 0.11629231721162796,
"signal/accuracy_reward/group_std_mean": 0.15797212719917297,
"signal/accuracy_reward/group_zero_std_frac": 0.5333333313465118,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05814615860581398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05814615860581398,
"signal/advantage_abs_mean": 0.07877274304628372,
"signal/advantage_pre_scale_abs_mean": 0.07877274304628372,
"signal/advantage_pre_scale_std": 0.1593018352985382,
"signal/advantage_std": 0.1593018352985382,
"signal/brier_reward/centered_abs_mean": 0.07354676201939583,
"signal/brier_reward/group_std_mean": 0.10252981781959533,
"signal/brier_reward/group_zero_std_frac": 0.16944444477558135,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03677338100969792,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03677338100969792,
"signal/confidence_one_or_zero/centered_abs_mean": 0.004372829850763082,
"signal/confidence_one_or_zero/group_std_mean": 0.005886570177972316,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9833333253860473,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.372829707222081e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.372829707222081e-08,
"signal/format_reward/centered_abs_mean": 0.00796983502805233,
"signal/format_reward/group_std_mean": 0.019131885096430778,
"signal/format_reward/group_zero_std_frac": 0.9055555582046508,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003984917514026165,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003984917514026165,
"signal/mean_confidence_reward/centered_abs_mean": 0.0666200652718544,
"signal/mean_confidence_reward/group_std_mean": 0.09021197110414506,
"signal/mean_confidence_reward/group_zero_std_frac": 0.18055555820465088,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.66200674004358e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.66200674004358e-07,
"step": 480
},
{
"calibration/aurc": 0.13403767836577124,
"calibration/batch_distribution_entropy": 0.6798431874057342,
"calibration/batch_entropy_100bins": 0.3820831852568066,
"calibration/batch_entropy_10bins": 0.6798431874057342,
"calibration/batch_entropy_50bins": 0.4497821436301697,
"calibration/batch_uniqueness": 0.5025837024384686,
"calibration/confidence_entropy": 0.4837146415950676,
"calibration/coverage@0%": 0.04804177545691906,
"calibration/coverage@1%": 0.23603133159268933,
"calibration/coverage@10%": 0.44020887728459535,
"calibration/coverage@15%": 0.4981723237597911,
"calibration/coverage@20%": 0.7363672214969539,
"calibration/coverage@25%": 0.8303062445604874,
"calibration/coverage@30%": 0.9237597911227156,
"calibration/coverage@5%": 0.37650130548302874,
"calibration/distribution_entropy_10": 0.6798431874057342,
"calibration/distribution_entropy_100": 0.3820831852568066,
"calibration/ece": 0.11172792917754566,
"calibration/mean_confidence": 0.6669808665143604,
"calibration/unique_confidence_per_question": 0.025520833333333336,
"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": 0.0014756944444444641,
"completions/max_length": 3134.2,
"completions/max_terminated_length": 3134.2,
"completions/mean_length": 949.7003662109375,
"completions/mean_terminated_length": 951.1052368164062,
"completions/min_length": 0.0,
"completions/min_terminated_length": 310.2,
"epoch": 1.1658653846153846,
"grad_norm": 0.0005296074668876827,
"learning_rate": 3.3353365384615388e-06,
"loss": -0.0007,
"num_tokens": 1187263033.0,
"reward": 1.2922637701034545,
"reward_std": 0.10782433152198792,
"rewards/accuracy_reward": 0.7294270753860473,
"rewards/brier_reward": 0.856648850440979,
"rewards/confidence_one_or_zero": 0.0008680555794853718,
"rewards/format_reward": 0.9984375,
"rewards/mean_confidence_reward": 0.7005642294883728,
"sampling/batch_mean_priority_error": 0.015800347222222205,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.38888888888888884,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.001638196362182498,
"sampling/priority_kl": 0.030000192299485206,
"sampling/priority_scale": 0.7103035271400586,
"sampling/prob_entropy": 10.27894287109375,
"sampling/prob_max": 4.745964688481763e-05,
"sampling/prob_min": 1.9623675325419753e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.1592000007629395,
"sampling/prompt_draws_total": 34776.0,
"sampling/seen_fraction": 0.7596666693687439,
"sampling/unseen_fraction": 0.24033333063125611,
"signal/accuracy_reward/centered_abs_mean": 0.11507704108953476,
"signal/accuracy_reward/group_std_mean": 0.15593255162239075,
"signal/accuracy_reward/group_zero_std_frac": 0.5416666746139527,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05753852054476738,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05753852054476738,
"signal/advantage_abs_mean": 0.0767291359603405,
"signal/advantage_pre_scale_abs_mean": 0.0767291359603405,
"signal/advantage_pre_scale_std": 0.15259878933429719,
"signal/advantage_std": 0.15259878933429719,
"signal/brier_reward/centered_abs_mean": 0.07359779626131058,
"signal/brier_reward/group_std_mean": 0.10185133069753646,
"signal/brier_reward/group_zero_std_frac": 0.20000000298023224,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03679889813065529,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03679889813065529,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0016710068914107979,
"signal/confidence_one_or_zero/group_std_mean": 0.004611522844061256,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.975,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6710067995973077e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6710067995973077e-08,
"signal/format_reward/centered_abs_mean": 0.0030164929921738803,
"signal/format_reward/group_std_mean": 0.008539893664419651,
"signal/format_reward/group_zero_std_frac": 0.9527777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0015082464960869402,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0015082464960869402,
"signal/mean_confidence_reward/centered_abs_mean": 0.06438992619514465,
"signal/mean_confidence_reward/group_std_mean": 0.08493052572011947,
"signal/mean_confidence_reward/group_zero_std_frac": 0.22500000298023223,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.438992159019108e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.438992159019108e-07,
"step": 485
},
{
"calibration/aurc": 0.10957053794746666,
"calibration/batch_distribution_entropy": 0.47153568446971417,
"calibration/batch_entropy_100bins": 0.2574468096316364,
"calibration/batch_entropy_10bins": 0.47153568446971417,
"calibration/batch_entropy_50bins": 0.30306221884388146,
"calibration/batch_uniqueness": 0.01182822670089628,
"calibration/confidence_entropy": 0.41203219808780067,
"calibration/coverage@0%": 0.003645833333333333,
"calibration/coverage@1%": 0.003645833333333333,
"calibration/coverage@10%": 0.5036458333333333,
"calibration/coverage@15%": 0.6979166666666667,
"calibration/coverage@20%": 0.9109375,
"calibration/coverage@25%": 0.9651041666666668,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.40816742819843344,
"calibration/distribution_entropy_10": 0.47153568446971417,
"calibration/distribution_entropy_100": 0.2574468096316364,
"calibration/ece": 0.106922930401436,
"calibration/mean_confidence": 0.7968985885824631,
"calibration/unique_confidence_per_question": 0.025520833333333336,
"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": 0.0013020833333333482,
"completions/max_length": 3576.4,
"completions/max_terminated_length": 3576.4,
"completions/mean_length": 944.9557373046875,
"completions/mean_terminated_length": 946.2088623046875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 279.4,
"epoch": 1.1778846153846154,
"grad_norm": 0.000669217377435416,
"learning_rate": 3.3052884615384617e-06,
"loss": -0.001,
"num_tokens": 1201210939.0,
"reward": 1.292154812812805,
"reward_std": 0.1198092296719551,
"rewards/accuracy_reward": 0.729687488079071,
"rewards/brier_reward": 0.8559090256690979,
"rewards/confidence_one_or_zero": 0.0005208333430346101,
"rewards/format_reward": 0.9986979007720947,
"rewards/mean_confidence_reward": 0.764504897594452,
"sampling/batch_mean_priority_error": 0.012920138888888873,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.28055555555555556,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.001652483455836773,
"sampling/priority_kl": 0.03000018261373043,
"sampling/priority_scale": 0.710580509970896,
"sampling/prob_entropy": 10.278962326049804,
"sampling/prob_max": 4.760355150210671e-05,
"sampling/prob_min": 1.9676406736834906e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.1711999893188476,
"sampling/prompt_draws_total": 35136.0,
"sampling/seen_fraction": 0.7635533332824707,
"sampling/unseen_fraction": 0.23644666671752929,
"signal/accuracy_reward/centered_abs_mean": 0.12069227546453476,
"signal/accuracy_reward/group_std_mean": 0.16156412959098815,
"signal/accuracy_reward/group_zero_std_frac": 0.5277777850627899,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06034613773226738,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06034613773226738,
"signal/advantage_abs_mean": 0.08784503787755966,
"signal/advantage_pre_scale_abs_mean": 0.08784503787755966,
"signal/advantage_pre_scale_std": 0.17135231494903563,
"signal/advantage_std": 0.17135231494903563,
"signal/brier_reward/centered_abs_mean": 0.07915262952446937,
"signal/brier_reward/group_std_mean": 0.10748916268348693,
"signal/brier_reward/group_zero_std_frac": 0.3000000029802322,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.039576314762234686,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.039576314762234686,
"signal/confidence_one_or_zero/centered_abs_mean": 0.000998263864312321,
"signal/confidence_one_or_zero/group_std_mean": 0.0026473373174667357,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 9.982638005112676e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 9.982638005112676e-09,
"signal/format_reward/centered_abs_mean": 0.002479383663740009,
"signal/format_reward/group_std_mean": 0.006169932056218385,
"signal/format_reward/group_zero_std_frac": 0.9694444417953492,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0012396918318700045,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0012396918318700045,
"signal/mean_confidence_reward/centered_abs_mean": 0.06347360759973526,
"signal/mean_confidence_reward/group_std_mean": 0.08472321182489395,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3388888955116272,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.347360454128647e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.347360454128647e-07,
"step": 490
},
{
"calibration/aurc": 0.09521846205223616,
"calibration/batch_distribution_entropy": 0.36980064712623495,
"calibration/batch_entropy_100bins": 0.19743328193770213,
"calibration/batch_entropy_10bins": 0.36980064712623495,
"calibration/batch_entropy_50bins": 0.23241526505332483,
"calibration/batch_uniqueness": -0.24817345815044733,
"calibration/confidence_entropy": 0.38735411680887133,
"calibration/coverage@0%": 0.0026191895725828007,
"calibration/coverage@1%": 0.0026191895725828007,
"calibration/coverage@10%": 0.5172412847871823,
"calibration/coverage@15%": 0.9143114969024759,
"calibration/coverage@20%": 0.9509227362204724,
"calibration/coverage@25%": 0.9661458333333334,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.3112806068954175,
"calibration/distribution_entropy_10": 0.36980064712623495,
"calibration/distribution_entropy_100": 0.19743328193770213,
"calibration/ece": 0.07073480712716064,
"calibration/mean_confidence": 0.8144085704304098,
"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": 0.001388888888888884,
"completions/max_length": 3515.2,
"completions/max_terminated_length": 3515.2,
"completions/mean_length": 971.435595703125,
"completions/mean_terminated_length": 972.7840454101563,
"completions/min_length": 0.0,
"completions/min_terminated_length": 312.8,
"epoch": 1.1899038461538463,
"grad_norm": 0.0005312844878062606,
"learning_rate": 3.2752403846153846e-06,
"loss": -0.001,
"num_tokens": 1215500757.0,
"reward": 1.297963547706604,
"reward_std": 0.11535521298646927,
"rewards/accuracy_reward": 0.7472222208976745,
"rewards/brier_reward": 0.8500782132148743,
"rewards/confidence_one_or_zero": 0.0032986111473292112,
"rewards/format_reward": 0.9986110925674438,
"rewards/mean_confidence_reward": 0.773796558380127,
"sampling/batch_mean_priority_error": 0.018286666666666646,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3388888888888889,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0016717145452275872,
"sampling/priority_kl": 0.02999982200562954,
"sampling/priority_scale": 0.7107961237663403,
"sampling/prob_entropy": 10.278962326049804,
"sampling/prob_max": 4.7745810297783466e-05,
"sampling/prob_min": 1.9729911946342328e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.1832000017166138,
"sampling/prompt_draws_total": 35496.0,
"sampling/seen_fraction": 0.7672733306884766,
"sampling/unseen_fraction": 0.23272666931152344,
"signal/accuracy_reward/centered_abs_mean": 0.11678602248430252,
"signal/accuracy_reward/group_std_mean": 0.15678493082523345,
"signal/accuracy_reward/group_zero_std_frac": 0.544444453716278,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05839301124215126,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05839301124215126,
"signal/advantage_abs_mean": 0.08481334894895554,
"signal/advantage_pre_scale_abs_mean": 0.08481334894895554,
"signal/advantage_pre_scale_std": 0.1691618800163269,
"signal/advantage_std": 0.1691618800163269,
"signal/brier_reward/centered_abs_mean": 0.0775853805243969,
"signal/brier_reward/group_std_mean": 0.10400516092777252,
"signal/brier_reward/group_zero_std_frac": 0.3111111164093018,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03879269026219845,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03879269026219845,
"signal/confidence_one_or_zero/centered_abs_mean": 0.00521918386220932,
"signal/confidence_one_or_zero/group_std_mean": 0.00827118419110775,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.2191836630299804e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.2191836630299804e-08,
"signal/format_reward/centered_abs_mean": 0.002615017397329211,
"signal/format_reward/group_std_mean": 0.006256770342588425,
"signal/format_reward/group_zero_std_frac": 0.9694444417953492,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0013075086986646055,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0013075086986646055,
"signal/mean_confidence_reward/centered_abs_mean": 0.060755544900894166,
"signal/mean_confidence_reward/group_std_mean": 0.08054654747247696,
"signal/mean_confidence_reward/group_zero_std_frac": 0.35277777910232544,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.07555386977765e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.07555386977765e-07,
"step": 495
},
{
"calibration/aurc": 0.1642228836953578,
"calibration/batch_distribution_entropy": 0.5205450125344753,
"calibration/batch_entropy_100bins": 0.2780713068077269,
"calibration/batch_entropy_10bins": 0.5205450125344753,
"calibration/batch_entropy_50bins": 0.3273410432180067,
"calibration/batch_uniqueness": 0.11008029513888892,
"calibration/confidence_entropy": 0.4279427060637147,
"calibration/coverage@0%": 0.0005208333333333333,
"calibration/coverage@1%": 0.0005208333333333333,
"calibration/coverage@10%": 0.31302083333333336,
"calibration/coverage@15%": 0.34635416666666663,
"calibration/coverage@20%": 0.6489583333333334,
"calibration/coverage@25%": 0.8666666666666666,
"calibration/coverage@30%": 0.9427083333333333,
"calibration/coverage@5%": 0.1515625,
"calibration/distribution_entropy_10": 0.5205450125344753,
"calibration/distribution_entropy_100": 0.2780713068077269,
"calibration/ece": 0.11427083333333339,
"calibration/mean_confidence": 0.7689583333333335,
"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": 0.002690972222222232,
"completions/max_length": 3503.4,
"completions/max_terminated_length": 3503.4,
"completions/mean_length": 976.9451293945312,
"completions/mean_terminated_length": 979.6044799804688,
"completions/min_length": 73.4,
"completions/min_terminated_length": 329.2,
"epoch": 1.2019230769230769,
"grad_norm": 0.0005742677603848279,
"learning_rate": 3.245192307692308e-06,
"loss": -0.0012,
"num_tokens": 1229878109.0,
"reward": 1.2855183839797975,
"reward_std": 0.10680401921272278,
"rewards/accuracy_reward": 0.7275173664093018,
"rewards/brier_reward": 0.8461951017379761,
"rewards/confidence_one_or_zero": 0.0024305555154569445,
"rewards/format_reward": 0.9973090291023254,
"rewards/mean_confidence_reward": 0.7703506827354432,
"sampling/batch_mean_priority_error": 0.02147395833333332,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.32499999999999996,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0016937416279688478,
"sampling/priority_kl": 0.02999955452978611,
"sampling/priority_scale": 0.7113852322334424,
"sampling/prob_entropy": 10.278951644897461,
"sampling/prob_max": 4.7900483332341536e-05,
"sampling/prob_min": 1.9779334616032428e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.19520001411438,
"sampling/prompt_draws_total": 35856.0,
"sampling/seen_fraction": 0.771233332157135,
"sampling/unseen_fraction": 0.22876666784286498,
"signal/accuracy_reward/centered_abs_mean": 0.10395507663488388,
"signal/accuracy_reward/group_std_mean": 0.1416932612657547,
"signal/accuracy_reward/group_zero_std_frac": 0.5805555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05197753831744194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05197753831744194,
"signal/advantage_abs_mean": 0.07803195416927337,
"signal/advantage_pre_scale_abs_mean": 0.07803195416927337,
"signal/advantage_pre_scale_std": 0.15995383262634277,
"signal/advantage_std": 0.15995383262634277,
"signal/brier_reward/centered_abs_mean": 0.073187355697155,
"signal/brier_reward/group_std_mean": 0.09856675863265991,
"signal/brier_reward/group_zero_std_frac": 0.3388888955116272,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0365936778485775,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0365936778485775,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0036132814129814506,
"signal/confidence_one_or_zero/group_std_mean": 0.005107206478714943,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9833333373069764,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.613281194247975e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.613281194247975e-08,
"signal/format_reward/centered_abs_mean": 0.0033365884679369628,
"signal/format_reward/group_std_mean": 0.006680760718882084,
"signal/format_reward/group_zero_std_frac": 0.9694444417953492,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0016682942339684814,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0016682942339684814,
"signal/mean_confidence_reward/centered_abs_mean": 0.059506846219301225,
"signal/mean_confidence_reward/group_std_mean": 0.07812358886003494,
"signal/mean_confidence_reward/group_zero_std_frac": 0.37222222089767454,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.950684339950385e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.950684339950385e-07,
"step": 500
},
{
"epoch": 1.2019230769230769,
"eval_calibration/aurc": 0.1517025700700997,
"eval_calibration/batch_distribution_entropy": 0.5561453822123794,
"eval_calibration/batch_entropy_100bins": 0.30538964238624244,
"eval_calibration/batch_entropy_10bins": 0.5561453822123794,
"eval_calibration/batch_entropy_50bins": 0.35949974585406874,
"eval_calibration/batch_uniqueness": 0.1863500945179584,
"eval_calibration/confidence_entropy": 0.4351659852744771,
"eval_calibration/coverage@0%": 0.004347826086956522,
"eval_calibration/coverage@1%": 0.004347826086956522,
"eval_calibration/coverage@10%": 0.004347826086956522,
"eval_calibration/coverage@15%": 0.717391304347826,
"eval_calibration/coverage@20%": 0.7947826086956522,
"eval_calibration/coverage@25%": 0.9034782608695652,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.004347826086956522,
"eval_calibration/distribution_entropy_10": 0.5561453822123794,
"eval_calibration/distribution_entropy_100": 0.30538964238624244,
"eval_calibration/ece": 0.04030434782608717,
"eval_calibration/mean_confidence": 0.7420434782608697,
"eval_calibration/unique_confidence_per_question": 0.009548611111111112,
"eval_calibration/unique_confidences": 11,
"eval_completions/clipped_ratio": 0.0017361111111111234,
"eval_completions/max_length": 2621.0,
"eval_completions/max_terminated_length": 2621.0,
"eval_completions/mean_length": 976.9887288411459,
"eval_completions/mean_terminated_length": 978.6986490885416,
"eval_completions/min_length": 218.16666666666666,
"eval_completions/min_terminated_length": 329.5,
"eval_loss": 0.0,
"eval_num_tokens": 1229878109.0,
"eval_reward": 1.2767446438471477,
"eval_reward_std": 0.3123304198185603,
"eval_rewards/accuracy_reward": 0.7005208333333334,
"eval_rewards/brier_reward": 0.8546896775563558,
"eval_rewards/confidence_one_or_zero": 0.0017361111628512542,
"eval_rewards/format_reward": 0.9982638955116272,
"eval_rewards/mean_confidence_reward": 0.740755170583725,
"eval_runtime": 166.5302,
"eval_samples_per_second": 6.005,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4050021668275197,
"eval_signal/accuracy_reward/group_std_mean": 0.4556471159060796,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20250108341375986,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.20250108341375986,
"eval_signal/advantage_abs_mean": 0.2665083234508832,
"eval_signal/advantage_pre_scale_abs_mean": 0.2665083234508832,
"eval_signal/advantage_pre_scale_std": 0.31057239572207135,
"eval_signal/advantage_std": 0.31057239572207135,
"eval_signal/brier_reward/centered_abs_mean": 0.16707724332809448,
"eval_signal/brier_reward/group_std_mean": 0.2277213086684545,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08353862166404724,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08353862166404724,
"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": 0.0033637151742974916,
"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": 0.1944471299648285,
"eval_signal/mean_confidence_reward/group_std_mean": 0.2322454775373141,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.94447125068109e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.94447125068109e-06,
"eval_steps_per_second": 0.036,
"step": 500
},
{
"epoch": 1.2019230769230769,
"step": 500,
"train_probe_calibration/aurc": 0.10092870521186632,
"train_probe_calibration/batch_distribution_entropy": 0.5295895610072994,
"train_probe_calibration/batch_entropy_100bins": 0.292901367615838,
"train_probe_calibration/batch_entropy_10bins": 0.5295895610072994,
"train_probe_calibration/batch_entropy_50bins": 0.3447987508529416,
"train_probe_calibration/batch_uniqueness": 0.12208538290895064,
"train_probe_calibration/confidence_entropy": 0.4257298528067076,
"train_probe_calibration/coverage@0%": 0.006076388888888889,
"train_probe_calibration/coverage@1%": 0.006076388888888889,
"train_probe_calibration/coverage@10%": 0.7395833333333334,
"train_probe_calibration/coverage@15%": 0.828125,
"train_probe_calibration/coverage@20%": 0.9210069444444444,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.006076388888888889,
"train_probe_calibration/distribution_entropy_10": 0.5295895610072994,
"train_probe_calibration/distribution_entropy_100": 0.292901367615838,
"train_probe_calibration/ece": 0.04473090277777759,
"train_probe_calibration/mean_confidence": 0.757873263888889,
"train_probe_calibration/unique_confidence_per_question": 0.010416666666666666,
"train_probe_calibration/unique_confidences": 12,
"train_probe_completions/clipped_ratio": 0.0,
"train_probe_completions/max_length": 2861.6666666666665,
"train_probe_completions/max_terminated_length": 2861.6666666666665,
"train_probe_completions/mean_length": 996.1976013183594,
"train_probe_completions/mean_terminated_length": 996.1976013183594,
"train_probe_completions/min_length": 311.8333333333333,
"train_probe_completions/min_terminated_length": 311.8333333333333,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1229878109.0,
"train_probe_reward": 1.3183907469113667,
"train_probe_reward_std": 0.2839902589718501,
"train_probe_rewards/accuracy_reward": 0.7526041567325592,
"train_probe_rewards/brier_reward": 0.8841620783011118,
"train_probe_rewards/confidence_one_or_zero": 0.0017361111628512542,
"train_probe_rewards/format_reward": 1.0,
"train_probe_rewards/mean_confidence_reward": 0.7578732271989187,
"train_probe_runtime": 150.5336,
"train_probe_samples_per_second": 6.643,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3594292551279068,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4275974631309509,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1797146275639534,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.1797146275639534,
"train_probe_signal/advantage_abs_mean": 0.23350405196348825,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.23350405196348825,
"train_probe_signal/advantage_pre_scale_std": 0.28261765340964,
"train_probe_signal/advantage_std": 0.28261765340964,
"train_probe_signal/brier_reward/centered_abs_mean": 0.13781099766492844,
"train_probe_signal/brier_reward/group_std_mean": 0.195537269115448,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.06890549883246422,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.06890549883246422,
"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": 0.0,
"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": 0.18608286480108896,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.22383885085582733,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.8608286040944222e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.8608286040944222e-06,
"train_probe_steps_per_second": 0.04
},
{
"calibration/aurc": 0.14251614829725212,
"calibration/batch_distribution_entropy": 0.5263690608949052,
"calibration/batch_entropy_100bins": 0.2825064938167437,
"calibration/batch_entropy_10bins": 0.5263690608949052,
"calibration/batch_entropy_50bins": 0.33256207360429674,
"calibration/batch_uniqueness": 0.1299214529303327,
"calibration/confidence_entropy": 0.4352356192038183,
"calibration/coverage@0%": 0.0026041666666666665,
"calibration/coverage@1%": 0.0026041666666666665,
"calibration/coverage@10%": 0.42893684725848563,
"calibration/coverage@15%": 0.637861727589208,
"calibration/coverage@20%": 0.7889074738903394,
"calibration/coverage@25%": 0.8707966166231506,
"calibration/coverage@30%": 0.915625,
"calibration/coverage@5%": 0.13054150348128807,
"calibration/distribution_entropy_10": 0.5263690608949052,
"calibration/distribution_entropy_100": 0.2825064938167437,
"calibration/ece": 0.08594960291557871,
"calibration/mean_confidence": 0.7611633757615319,
"calibration/unique_confidence_per_question": 0.023958333333333335,
"calibration/unique_confidences": 9.2,
"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": 0.001475694444444442,
"completions/max_length": 3047.6,
"completions/max_terminated_length": 3047.6,
"completions/mean_length": 951.081591796875,
"completions/mean_terminated_length": 952.4546752929688,
"completions/min_length": 0.0,
"completions/min_terminated_length": 297.6,
"epoch": 1.2139423076923077,
"grad_norm": 0.0005359905189834535,
"learning_rate": 3.215144230769231e-06,
"loss": 0.0,
"num_tokens": 1243919369.0,
"reward": 1.2760514259338378,
"reward_std": 0.10757787972688675,
"rewards/accuracy_reward": 0.7075520873069763,
"rewards/brier_reward": 0.8460114002227783,
"rewards/confidence_one_or_zero": 0.00303819450782612,
"rewards/format_reward": 0.998524296283722,
"rewards/mean_confidence_reward": 0.7550451397895813,
"sampling/batch_mean_priority_error": 0.02132645833333332,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.32222222222222224,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0017179204151034355,
"sampling/priority_kl": 0.030000147596001624,
"sampling/priority_scale": 0.7120238006813452,
"sampling/prob_entropy": 10.278959465026855,
"sampling/prob_max": 4.8056802188511935e-05,
"sampling/prob_min": 1.9828132644761352e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.207200002670288,
"sampling/prompt_draws_total": 36216.0,
"sampling/seen_fraction": 0.7751600027084351,
"sampling/unseen_fraction": 0.22483999729156495,
"signal/accuracy_reward/centered_abs_mean": 0.112841796875,
"signal/accuracy_reward/group_std_mean": 0.1478495255112648,
"signal/accuracy_reward/group_zero_std_frac": 0.5777777791023254,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0564208984375,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0564208984375,
"signal/advantage_abs_mean": 0.08059335947036743,
"signal/advantage_pre_scale_abs_mean": 0.08059335947036743,
"signal/advantage_pre_scale_std": 0.16366566717624664,
"signal/advantage_std": 0.16366566717624664,
"signal/brier_reward/centered_abs_mean": 0.0721733033657074,
"signal/brier_reward/group_std_mean": 0.09534063339233398,
"signal/brier_reward/group_zero_std_frac": 0.30555555522441863,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0360866516828537,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0360866516828537,
"signal/confidence_one_or_zero/centered_abs_mean": 0.004823133698664606,
"signal/confidence_one_or_zero/group_std_mean": 0.007467509433627129,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.975,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.8231336791104697e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.8231336791104697e-08,
"signal/format_reward/centered_abs_mean": 0.002728949720039964,
"signal/format_reward/group_std_mean": 0.006286143884062767,
"signal/format_reward/group_zero_std_frac": 0.9694444417953492,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.001364474860019982,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.001364474860019982,
"signal/mean_confidence_reward/centered_abs_mean": 0.05904992073774338,
"signal/mean_confidence_reward/group_std_mean": 0.07791801393032075,
"signal/mean_confidence_reward/group_zero_std_frac": 0.325,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.904991667193826e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.904991667193826e-07,
"step": 505
},
{
"calibration/aurc": 0.15843764969117557,
"calibration/batch_distribution_entropy": 0.5827104619682373,
"calibration/batch_entropy_100bins": 0.31697319022488957,
"calibration/batch_entropy_10bins": 0.5827104619682373,
"calibration/batch_entropy_50bins": 0.3731357109494903,
"calibration/batch_uniqueness": 0.24571379969595064,
"calibration/confidence_entropy": 0.44576822093219964,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.3182042152419016,
"calibration/coverage@15%": 0.5514867442168618,
"calibration/coverage@20%": 0.7391858641388367,
"calibration/coverage@25%": 0.8683088043313957,
"calibration/coverage@30%": 0.936482939632546,
"calibration/coverage@5%": 0.05511811023622047,
"calibration/distribution_entropy_10": 0.5827104619682373,
"calibration/distribution_entropy_100": 0.31697319022488957,
"calibration/ece": 0.09253316062817325,
"calibration/mean_confidence": 0.6995938054322008,
"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": 0.004600694444444419,
"completions/max_length": 3544.0,
"completions/max_terminated_length": 3544.0,
"completions/mean_length": 1037.9000122070313,
"completions/mean_terminated_length": 1042.7189819335938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 296.4,
"epoch": 1.2259615384615385,
"grad_norm": 0.00042660924373194575,
"learning_rate": 3.185096153846154e-06,
"loss": -0.0038,
"num_tokens": 1259024649.0,
"reward": 1.284938645362854,
"reward_std": 0.10022608935832977,
"rewards/accuracy_reward": 0.7225694537162781,
"rewards/brier_reward": 0.851894474029541,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9953993201255799,
"rewards/mean_confidence_reward": 0.7036577582359314,
"sampling/batch_mean_priority_error": 0.019823124999999987,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.35000000000000003,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.00173952616751194,
"sampling/priority_kl": 0.029999304562807083,
"sampling/priority_scale": 0.7129302441840991,
"sampling/prob_entropy": 10.278946113586425,
"sampling/prob_max": 4.822285627597012e-05,
"sampling/prob_min": 1.987423274840694e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.2192000150680542,
"sampling/prompt_draws_total": 36576.0,
"sampling/seen_fraction": 0.7792199850082397,
"sampling/unseen_fraction": 0.22078001499176025,
"signal/accuracy_reward/centered_abs_mean": 0.10558810830116272,
"signal/accuracy_reward/group_std_mean": 0.14366630464792252,
"signal/accuracy_reward/group_zero_std_frac": 0.569444453716278,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05279405415058136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05279405415058136,
"signal/advantage_abs_mean": 0.07034819722175598,
"signal/advantage_pre_scale_abs_mean": 0.07034819722175598,
"signal/advantage_pre_scale_std": 0.1494814023375511,
"signal/advantage_std": 0.1494814023375511,
"signal/brier_reward/centered_abs_mean": 0.06675559654831886,
"signal/brier_reward/group_std_mean": 0.0913299024105072,
"signal/brier_reward/group_zero_std_frac": 0.2527777820825577,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03337779827415943,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03337779827415943,
"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": 0.00739474818110466,
"signal/format_reward/group_std_mean": 0.01648670230060816,
"signal/format_reward/group_zero_std_frac": 0.919444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00369737409055233,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00369737409055233,
"signal/mean_confidence_reward/centered_abs_mean": 0.05924249067902565,
"signal/mean_confidence_reward/group_std_mean": 0.07888469099998474,
"signal/mean_confidence_reward/group_zero_std_frac": 0.27222222089767456,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.924248853261816e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.924248853261816e-07,
"step": 510
},
{
"calibration/aurc": 0.14249268358674544,
"calibration/batch_distribution_entropy": 0.7107646528720579,
"calibration/batch_entropy_100bins": 0.3949343638565591,
"calibration/batch_entropy_10bins": 0.7107646528720579,
"calibration/batch_entropy_50bins": 0.4649103431474706,
"calibration/batch_uniqueness": 0.5531128800054119,
"calibration/confidence_entropy": 0.49678547413664287,
"calibration/coverage@0%": 0.12320026178010472,
"calibration/coverage@1%": 0.1399269197207679,
"calibration/coverage@10%": 0.5555655541012217,
"calibration/coverage@15%": 0.6448489310645724,
"calibration/coverage@20%": 0.7060291230366492,
"calibration/coverage@25%": 0.7723958333333333,
"calibration/coverage@30%": 0.7817708333333333,
"calibration/coverage@5%": 0.4678037739965095,
"calibration/distribution_entropy_10": 0.7107646528720579,
"calibration/distribution_entropy_100": 0.3949343638565591,
"calibration/ece": 0.1659895833333333,
"calibration/mean_confidence": 0.6266151287085516,
"calibration/unique_confidence_per_question": 0.023958333333333335,
"calibration/unique_confidences": 9.2,
"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": 0.003645833333333326,
"completions/max_length": 3360.6,
"completions/max_terminated_length": 3360.6,
"completions/mean_length": 1012.97119140625,
"completions/mean_terminated_length": 1016.7093627929687,
"completions/min_length": 0.0,
"completions/min_terminated_length": 297.8,
"epoch": 1.2379807692307692,
"grad_norm": 0.00041118654189631343,
"learning_rate": 3.1550480769230772e-06,
"loss": -0.0021,
"num_tokens": 1273792253.0,
"reward": 1.2771016120910645,
"reward_std": 0.09434493482112885,
"rewards/accuracy_reward": 0.70390625,
"rewards/brier_reward": 0.8539300918579101,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9963541626930237,
"rewards/mean_confidence_reward": 0.6344878196716308,
"sampling/batch_mean_priority_error": 0.023963541666666654,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.36666666666666664,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0017642594641074539,
"sampling/priority_kl": 0.030000543966889382,
"sampling/priority_scale": 0.7140926540596411,
"sampling/prob_entropy": 10.278964042663574,
"sampling/prob_max": 4.839725079364143e-05,
"sampling/prob_min": 1.9917297686333767e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.2312000036239623,
"sampling/prompt_draws_total": 36936.0,
"sampling/seen_fraction": 0.7833999991416931,
"sampling/unseen_fraction": 0.21660000085830688,
"signal/accuracy_reward/centered_abs_mean": 0.10814344584941864,
"signal/accuracy_reward/group_std_mean": 0.14549138844013215,
"signal/accuracy_reward/group_zero_std_frac": 0.5777778029441833,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05407172292470932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05407172292470932,
"signal/advantage_abs_mean": 0.06500188633799553,
"signal/advantage_pre_scale_abs_mean": 0.06500188633799553,
"signal/advantage_pre_scale_std": 0.13270943313837053,
"signal/advantage_std": 0.13270943313837053,
"signal/brier_reward/centered_abs_mean": 0.069306131452322,
"signal/brier_reward/group_std_mean": 0.09529425203800201,
"signal/brier_reward/group_zero_std_frac": 0.18611111342906952,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.034653065726161,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.034653065726161,
"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": 0.006901041604578495,
"signal/format_reward/group_std_mean": 0.016902656853199007,
"signal/format_reward/group_zero_std_frac": 0.9166666865348816,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0034505208022892475,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0034505208022892475,
"signal/mean_confidence_reward/centered_abs_mean": 0.0655409075319767,
"signal/mean_confidence_reward/group_std_mean": 0.08637888133525848,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1972222238779068,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.554090873578388e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.554090873578388e-07,
"step": 515
},
{
"calibration/aurc": 0.11034997527151753,
"calibration/batch_distribution_entropy": 0.6606235870600051,
"calibration/batch_entropy_100bins": 0.37894630253316874,
"calibration/batch_entropy_10bins": 0.6606235870600051,
"calibration/batch_entropy_50bins": 0.44608945604224043,
"calibration/batch_uniqueness": 0.48710002437172895,
"calibration/confidence_entropy": 0.46217896173955075,
"calibration/coverage@0%": 0.326305045202683,
"calibration/coverage@1%": 0.3531308586426697,
"calibration/coverage@10%": 0.5324284289943687,
"calibration/coverage@15%": 0.7396593314142887,
"calibration/coverage@20%": 0.8100803062793416,
"calibration/coverage@25%": 0.845740115818856,
"calibration/coverage@30%": 0.9000499937507811,
"calibration/coverage@5%": 0.4778884532976136,
"calibration/distribution_entropy_10": 0.6606235870600051,
"calibration/distribution_entropy_100": 0.37894630253316874,
"calibration/ece": 0.15638544680169772,
"calibration/mean_confidence": 0.6244671800405404,
"calibration/unique_confidence_per_question": 0.026041666666666664,
"calibration/unique_confidences": 10.0,
"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": 0.008506944444444442,
"completions/max_length": 3761.6,
"completions/max_terminated_length": 3761.6,
"completions/mean_length": 1067.0155395507813,
"completions/mean_terminated_length": 1076.108544921875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 321.8,
"epoch": 1.25,
"grad_norm": 0.00043587424443103373,
"learning_rate": 3.125e-06,
"loss": -0.0086,
"num_tokens": 1289181520.0,
"reward": 1.2708593368530274,
"reward_std": 0.10695169419050217,
"rewards/accuracy_reward": 0.711718761920929,
"rewards/brier_reward": 0.8384949564933777,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9914930582046508,
"rewards/mean_confidence_reward": 0.5937291979789734,
"sampling/batch_mean_priority_error": 0.03257645833333332,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.33333333333333337,
"sampling/error_ema_max": 0.12712499499320984,
"sampling/error_ema_mean": 0.0017990589840337634,
"sampling/priority_kl": 0.030000123009085656,
"sampling/priority_scale": 0.7153894484275952,
"sampling/prob_entropy": 10.27896556854248,
"sampling/prob_max": 4.8576216795481744e-05,
"sampling/prob_min": 1.9846079521812498e-05,
"sampling/prompt_draws_max": 6.2,
"sampling/prompt_draws_mean": 1.2431999921798706,
"sampling/prompt_draws_total": 37296.0,
"sampling/seen_fraction": 0.7875866651535034,
"sampling/unseen_fraction": 0.21241333484649658,
"signal/accuracy_reward/centered_abs_mean": 0.12175021916627884,
"signal/accuracy_reward/group_std_mean": 0.16610720455646516,
"signal/accuracy_reward/group_zero_std_frac": 0.5027777850627899,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06087510958313942,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06087510958313942,
"signal/advantage_abs_mean": 0.07055862694978714,
"signal/advantage_pre_scale_abs_mean": 0.07055862694978714,
"signal/advantage_pre_scale_std": 0.14613762199878694,
"signal/advantage_std": 0.14613762199878694,
"signal/brier_reward/centered_abs_mean": 0.07722499519586563,
"signal/brier_reward/group_std_mean": 0.10501042306423188,
"signal/brier_reward/group_zero_std_frac": 0.14166666716337203,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03861249759793282,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03861249759793282,
"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": 0.015755208395421504,
"signal/format_reward/group_std_mean": 0.034674161300063136,
"signal/format_reward/group_zero_std_frac": 0.8416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007877604197710752,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007877604197710752,
"signal/mean_confidence_reward/centered_abs_mean": 0.06463623270392418,
"signal/mean_confidence_reward/group_std_mean": 0.08526600450277329,
"signal/mean_confidence_reward/group_zero_std_frac": 0.15,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.463623094532522e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.463623094532522e-07,
"step": 520
},
{
"calibration/aurc": 0.12669337083961585,
"calibration/batch_distribution_entropy": 0.5590922243032654,
"calibration/batch_entropy_100bins": 0.3075703534005839,
"calibration/batch_entropy_10bins": 0.5590922243032654,
"calibration/batch_entropy_50bins": 0.36206684357654284,
"calibration/batch_uniqueness": 0.2584583537271246,
"calibration/confidence_entropy": 0.4585718308112806,
"calibration/coverage@0%": 0.0777363900621045,
"calibration/coverage@1%": 0.0777363900621045,
"calibration/coverage@10%": 0.36548442155816746,
"calibration/coverage@15%": 0.6758849441412209,
"calibration/coverage@20%": 0.7699068744066566,
"calibration/coverage@25%": 0.8686922655944602,
"calibration/coverage@30%": 0.9425866979393532,
"calibration/coverage@5%": 0.34415108822483415,
"calibration/distribution_entropy_10": 0.5590922243032654,
"calibration/distribution_entropy_100": 0.3075703534005839,
"calibration/ece": 0.09078399273439078,
"calibration/mean_confidence": 0.6898426965050423,
"calibration/unique_confidence_per_question": 0.0265625,
"calibration/unique_confidences": 10.2,
"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": 0.013020833333333304,
"completions/max_length": 3502.8,
"completions/max_terminated_length": 3502.8,
"completions/mean_length": 1041.5444580078124,
"completions/mean_terminated_length": 1055.194970703125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 310.6,
"epoch": 1.2620192307692308,
"grad_norm": 0.0005399513756856322,
"learning_rate": 3.094951923076923e-06,
"loss": -0.014,
"num_tokens": 1304277552.0,
"reward": 1.2725529432296754,
"reward_std": 0.12192474752664566,
"rewards/accuracy_reward": 0.7072916746139526,
"rewards/brier_reward": 0.8508217453956604,
"rewards/confidence_one_or_zero": 0.003125000064028427,
"rewards/format_reward": 0.9869791507720947,
"rewards/mean_confidence_reward": 0.6684785842895508,
"sampling/batch_mean_priority_error": 0.021243124999999984,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.32222222222222224,
"sampling/error_ema_max": 0.13218749463558196,
"sampling/error_ema_mean": 0.0018275268841534853,
"sampling/priority_kl": 0.029999518766999245,
"sampling/priority_scale": 0.7165736616356299,
"sampling/prob_entropy": 10.278942108154297,
"sampling/prob_max": 4.875140512012876e-05,
"sampling/prob_min": 1.94367141375551e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.2551999807357788,
"sampling/prompt_draws_total": 37656.0,
"sampling/seen_fraction": 0.7915599942207336,
"sampling/unseen_fraction": 0.20844000577926636,
"signal/accuracy_reward/centered_abs_mean": 0.12051866203546524,
"signal/accuracy_reward/group_std_mean": 0.15819574296474456,
"signal/accuracy_reward/group_zero_std_frac": 0.5555555641651153,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06025933101773262,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06025933101773262,
"signal/advantage_abs_mean": 0.08612038418650628,
"signal/advantage_pre_scale_abs_mean": 0.08612038418650628,
"signal/advantage_pre_scale_std": 0.1748889595270157,
"signal/advantage_std": 0.1748889595270157,
"signal/brier_reward/centered_abs_mean": 0.07994573190808296,
"signal/brier_reward/group_std_mean": 0.10857147574424744,
"signal/brier_reward/group_zero_std_frac": 0.1944444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03997286595404148,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03997286595404148,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0013020833604969085,
"signal/confidence_one_or_zero/group_std_mean": 0.00297891478985548,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.3020832767551838e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.3020832767551838e-08,
"signal/format_reward/centered_abs_mean": 0.02253689169883728,
"signal/format_reward/group_std_mean": 0.043699033185839654,
"signal/format_reward/group_zero_std_frac": 0.8166666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01126844584941864,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01126844584941864,
"signal/mean_confidence_reward/centered_abs_mean": 0.06638413742184639,
"signal/mean_confidence_reward/group_std_mean": 0.0887540727853775,
"signal/mean_confidence_reward/group_zero_std_frac": 0.19722222685813903,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.638413196924375e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.638413196924375e-07,
"step": 525
},
{
"calibration/aurc": 0.1575017173793545,
"calibration/batch_distribution_entropy": 0.4244637415882807,
"calibration/batch_entropy_100bins": 0.25155728112065345,
"calibration/batch_entropy_10bins": 0.4244637415882807,
"calibration/batch_entropy_50bins": 0.29612916117252525,
"calibration/batch_uniqueness": -0.01812680104952442,
"calibration/confidence_entropy": 0.39954801937750983,
"calibration/coverage@0%": 0.030234880010291933,
"calibration/coverage@1%": 0.1964617929390518,
"calibration/coverage@10%": 0.35107920718707286,
"calibration/coverage@15%": 0.4797605073611032,
"calibration/coverage@20%": 0.49194402501674783,
"calibration/coverage@25%": 0.8504267548121826,
"calibration/coverage@30%": 0.9523887048784596,
"calibration/coverage@5%": 0.21493144993113625,
"calibration/distribution_entropy_10": 0.4244637415882807,
"calibration/distribution_entropy_100": 0.25155728112065345,
"calibration/ece": 0.11915949920399335,
"calibration/mean_confidence": 0.778276141262144,
"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": 0.01171875,
"completions/max_length": 3533.2,
"completions/max_terminated_length": 3533.2,
"completions/mean_length": 1003.426123046875,
"completions/mean_terminated_length": 1015.3042358398437,
"completions/min_length": 0.0,
"completions/min_terminated_length": 316.8,
"epoch": 1.2740384615384617,
"grad_norm": 0.0005864729755558074,
"learning_rate": 3.0649038461538464e-06,
"loss": -0.0134,
"num_tokens": 1318960189.0,
"reward": 1.295579433441162,
"reward_std": 0.12813606709241868,
"rewards/accuracy_reward": 0.74765625,
"rewards/brier_reward": 0.8552927732467651,
"rewards/confidence_one_or_zero": 0.01319444440305233,
"rewards/format_reward": 0.9881944537162781,
"rewards/mean_confidence_reward": 0.7612300276756286,
"sampling/batch_mean_priority_error": 0.012533472222222208,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3194444444444445,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0018447082722559572,
"sampling/priority_kl": 0.03000018410384655,
"sampling/priority_scale": 0.7177888571983203,
"sampling/prob_entropy": 10.278960990905762,
"sampling/prob_max": 4.892680808552541e-05,
"sampling/prob_min": 1.9476248780847527e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.267199993133545,
"sampling/prompt_draws_total": 38016.0,
"sampling/seen_fraction": 0.7954866647720337,
"sampling/unseen_fraction": 0.2045133352279663,
"signal/accuracy_reward/centered_abs_mean": 0.10601671040058136,
"signal/accuracy_reward/group_std_mean": 0.14474707543849946,
"signal/accuracy_reward/group_zero_std_frac": 0.569444453716278,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05300835520029068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05300835520029068,
"signal/advantage_abs_mean": 0.0865313470363617,
"signal/advantage_pre_scale_abs_mean": 0.0865313470363617,
"signal/advantage_pre_scale_std": 0.18588004112243653,
"signal/advantage_std": 0.18588004112243653,
"signal/brier_reward/centered_abs_mean": 0.0755378246307373,
"signal/brier_reward/group_std_mean": 0.10764126926660537,
"signal/brier_reward/group_zero_std_frac": 0.225,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03776891231536865,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03776891231536865,
"signal/confidence_one_or_zero/centered_abs_mean": 0.016254340205341576,
"signal/confidence_one_or_zero/group_std_mean": 0.021729598753154278,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9361111164093018,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6254339669785621e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6254339669785621e-07,
"signal/format_reward/centered_abs_mean": 0.02116970419883728,
"signal/format_reward/group_std_mean": 0.044343719631433486,
"signal/format_reward/group_zero_std_frac": 0.8027777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01058485209941864,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01058485209941864,
"signal/mean_confidence_reward/centered_abs_mean": 0.06626470685005188,
"signal/mean_confidence_reward/group_std_mean": 0.0918770357966423,
"signal/mean_confidence_reward/group_zero_std_frac": 0.23333333134651185,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.626470508308557e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.626470508308557e-07,
"step": 530
},
{
"calibration/aurc": 0.1569989778980939,
"calibration/batch_distribution_entropy": 0.43016315371581937,
"calibration/batch_entropy_100bins": 0.24929648735262097,
"calibration/batch_entropy_10bins": 0.43016315371581937,
"calibration/batch_entropy_50bins": 0.2934677913281341,
"calibration/batch_uniqueness": -0.014269761773715925,
"calibration/confidence_entropy": 0.4131327922802903,
"calibration/coverage@0%": 0.021600163948871234,
"calibration/coverage@1%": 0.021600163948871234,
"calibration/coverage@10%": 0.33457047537762075,
"calibration/coverage@15%": 0.5210108246875189,
"calibration/coverage@20%": 0.543657795926932,
"calibration/coverage@25%": 0.7147992623067101,
"calibration/coverage@30%": 0.9004071499503474,
"calibration/coverage@5%": 0.32035569549687365,
"calibration/distribution_entropy_10": 0.43016315371581937,
"calibration/distribution_entropy_100": 0.24929648735262097,
"calibration/ece": 0.11942447875570532,
"calibration/mean_confidence": 0.7816901375303423,
"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": 0.016319444444444442,
"completions/max_length": 3661.6,
"completions/max_terminated_length": 3661.6,
"completions/mean_length": 1042.116064453125,
"completions/mean_terminated_length": 1059.5912841796876,
"completions/min_length": 0.0,
"completions/min_terminated_length": 318.2,
"epoch": 1.2860576923076923,
"grad_norm": 0.000569651136174798,
"learning_rate": 3.0348557692307694e-06,
"loss": -0.0164,
"num_tokens": 1334048246.0,
"reward": 1.2901008605957032,
"reward_std": 0.13372049778699874,
"rewards/accuracy_reward": 0.7421006798744202,
"rewards/brier_reward": 0.8544050455093384,
"rewards/confidence_one_or_zero": 0.012847222294658423,
"rewards/format_reward": 0.9836805582046508,
"rewards/mean_confidence_reward": 0.7595043301582336,
"sampling/batch_mean_priority_error": 0.014679791666666655,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.33888888888888885,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.001859647035598755,
"sampling/priority_kl": 0.029999933019280434,
"sampling/priority_scale": 0.7192020476097241,
"sampling/prob_entropy": 10.278958702087403,
"sampling/prob_max": 4.91113263706211e-05,
"sampling/prob_min": 1.9514276573318057e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.279200005531311,
"sampling/prompt_draws_total": 38376.0,
"sampling/seen_fraction": 0.7994733214378357,
"sampling/unseen_fraction": 0.2005266785621643,
"signal/accuracy_reward/centered_abs_mean": 0.11075846254825591,
"signal/accuracy_reward/group_std_mean": 0.15180006623268127,
"signal/accuracy_reward/group_zero_std_frac": 0.5472222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05537923127412796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05537923127412796,
"signal/advantage_abs_mean": 0.09288178682327271,
"signal/advantage_pre_scale_abs_mean": 0.09288178682327271,
"signal/advantage_pre_scale_std": 0.19578463435173035,
"signal/advantage_std": 0.19578463435173035,
"signal/brier_reward/centered_abs_mean": 0.08117109388113022,
"signal/brier_reward/group_std_mean": 0.11099212318658828,
"signal/brier_reward/group_zero_std_frac": 0.22777777910232544,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04058554694056511,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04058554694056511,
"signal/confidence_one_or_zero/centered_abs_mean": 0.01097005195915699,
"signal/confidence_one_or_zero/group_std_mean": 0.01581360977143049,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.95,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.097005167594034e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.097005167594034e-07,
"signal/format_reward/centered_abs_mean": 0.0247178815305233,
"signal/format_reward/group_std_mean": 0.04667534232139588,
"signal/format_reward/group_zero_std_frac": 0.8111111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01235894076526165,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01235894076526165,
"signal/mean_confidence_reward/centered_abs_mean": 0.0684033289551735,
"signal/mean_confidence_reward/group_std_mean": 0.09226205348968505,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2611111134290695,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.840332844149089e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.840332844149089e-07,
"step": 535
},
{
"calibration/aurc": 0.10487956573505355,
"calibration/batch_distribution_entropy": 0.4847926145012039,
"calibration/batch_entropy_100bins": 0.2802343493743758,
"calibration/batch_entropy_10bins": 0.4847926145012039,
"calibration/batch_entropy_50bins": 0.32988734192973024,
"calibration/batch_uniqueness": 0.10339296265997477,
"calibration/confidence_entropy": 0.4206212969895116,
"calibration/coverage@0%": 0.025190412070028183,
"calibration/coverage@1%": 0.025190412070028183,
"calibration/coverage@10%": 0.6500353826812535,
"calibration/coverage@15%": 0.8218298884514436,
"calibration/coverage@20%": 0.8951689632545932,
"calibration/coverage@25%": 0.9317585301837269,
"calibration/coverage@30%": 0.9653543307086615,
"calibration/coverage@5%": 0.18086323528902554,
"calibration/distribution_entropy_10": 0.4847926145012039,
"calibration/distribution_entropy_100": 0.2802343493743758,
"calibration/ece": 0.06519686971041702,
"calibration/mean_confidence": 0.7616189501538526,
"calibration/unique_confidence_per_question": 0.023958333333333335,
"calibration/unique_confidences": 9.2,
"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": 0.012499999999999978,
"completions/max_length": 3479.0,
"completions/max_terminated_length": 3479.0,
"completions/mean_length": 998.4607788085938,
"completions/mean_terminated_length": 1011.1102172851563,
"completions/min_length": 0.0,
"completions/min_terminated_length": 303.6,
"epoch": 1.2980769230769231,
"grad_norm": 0.0005864489357918501,
"learning_rate": 3.0048076923076923e-06,
"loss": -0.0134,
"num_tokens": 1348647378.0,
"reward": 1.3082192659378051,
"reward_std": 0.12426633834838867,
"rewards/accuracy_reward": 0.7584201455116272,
"rewards/brier_reward": 0.8705028891563416,
"rewards/confidence_one_or_zero": 0.014496527798473834,
"rewards/format_reward": 0.9875,
"rewards/mean_confidence_reward": 0.7618720889091491,
"sampling/batch_mean_priority_error": 0.013787291666666656,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.275,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0018759463913738728,
"sampling/priority_kl": 0.03000033609569073,
"sampling/priority_scale": 0.7201821506721899,
"sampling/prob_entropy": 10.278959465026855,
"sampling/prob_max": 4.927778500132263e-05,
"sampling/prob_min": 1.955580955836922e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.2911999940872192,
"sampling/prompt_draws_total": 38736.0,
"sampling/seen_fraction": 0.8030200004577637,
"sampling/unseen_fraction": 0.19697999954223633,
"signal/accuracy_reward/centered_abs_mean": 0.10099283754825591,
"signal/accuracy_reward/group_std_mean": 0.14111926555633544,
"signal/accuracy_reward/group_zero_std_frac": 0.5722222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05049641877412796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05049641877412796,
"signal/advantage_abs_mean": 0.08462517410516739,
"signal/advantage_pre_scale_abs_mean": 0.08462517410516739,
"signal/advantage_pre_scale_std": 0.1842319041490555,
"signal/advantage_std": 0.1842319041490555,
"signal/brier_reward/centered_abs_mean": 0.07206552177667618,
"signal/brier_reward/group_std_mean": 0.10260828882455826,
"signal/brier_reward/group_zero_std_frac": 0.24722222089767457,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03603276088833809,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03603276088833809,
"signal/confidence_one_or_zero/centered_abs_mean": 0.013037109375,
"signal/confidence_one_or_zero/group_std_mean": 0.017157964035868646,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.950000011920929,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.3037108601565705e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.3037108601565705e-07,
"signal/format_reward/centered_abs_mean": 0.02145182304084301,
"signal/format_reward/group_std_mean": 0.039249108731746675,
"signal/format_reward/group_zero_std_frac": 0.8444444417953492,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010725911520421506,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010725911520421506,
"signal/mean_confidence_reward/centered_abs_mean": 0.06385495141148567,
"signal/mean_confidence_reward/group_std_mean": 0.08670918941497803,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2666666626930237,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.385494771166122e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.385494771166122e-07,
"step": 540
},
{
"calibration/aurc": 0.0925614115676943,
"calibration/batch_distribution_entropy": 0.5153764623631993,
"calibration/batch_entropy_100bins": 0.2838094451090586,
"calibration/batch_entropy_10bins": 0.5153764623631993,
"calibration/batch_entropy_50bins": 0.3340958867840345,
"calibration/batch_uniqueness": 0.1330143627366373,
"calibration/confidence_entropy": 0.4356818326638935,
"calibration/coverage@0%": 0.006799534297347009,
"calibration/coverage@1%": 0.10496226395613965,
"calibration/coverage@10%": 0.7552523354340004,
"calibration/coverage@15%": 0.8411124730780685,
"calibration/coverage@20%": 0.860109834555641,
"calibration/coverage@25%": 0.9328083989501312,
"calibration/coverage@30%": 0.9328083989501312,
"calibration/coverage@5%": 0.5268695781056467,
"calibration/distribution_entropy_10": 0.5153764623631993,
"calibration/distribution_entropy_100": 0.2838094451090586,
"calibration/ece": 0.11489496489971199,
"calibration/mean_confidence": 0.740481303681016,
"calibration/unique_confidence_per_question": 0.024479166666666666,
"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": 0.01137152777777779,
"completions/max_length": 3503.0,
"completions/max_terminated_length": 3503.0,
"completions/mean_length": 1055.1447875976562,
"completions/mean_terminated_length": 1067.214404296875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 291.2,
"epoch": 1.3100961538461537,
"grad_norm": 0.00045707475510425866,
"learning_rate": 2.974759615384616e-06,
"loss": -0.0128,
"num_tokens": 1363906294.0,
"reward": 1.2843178749084472,
"reward_std": 0.11411627978086472,
"rewards/accuracy_reward": 0.717881953716278,
"rewards/brier_reward": 0.8621111154556275,
"rewards/confidence_one_or_zero": 0.006857638875953853,
"rewards/format_reward": 0.9886284708976746,
"rewards/mean_confidence_reward": 0.7079152345657349,
"sampling/batch_mean_priority_error": 0.02168749999999999,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2972222222222222,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0018944249954074622,
"sampling/priority_kl": 0.030000030994415283,
"sampling/priority_scale": 0.721052998281084,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 4.943923995597288e-05,
"sampling/prob_min": 1.959796609298792e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3032000064849854,
"sampling/prompt_draws_total": 39096.0,
"sampling/seen_fraction": 0.8064000010490417,
"sampling/unseen_fraction": 0.19359999895095825,
"signal/accuracy_reward/centered_abs_mean": 0.09945746511220932,
"signal/accuracy_reward/group_std_mean": 0.14147261381149293,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111164093018,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04972873255610466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04972873255610466,
"signal/advantage_abs_mean": 0.07555333971977234,
"signal/advantage_pre_scale_abs_mean": 0.07555333971977234,
"signal/advantage_pre_scale_std": 0.1664302110671997,
"signal/advantage_std": 0.1664302110671997,
"signal/brier_reward/centered_abs_mean": 0.07284339591860771,
"signal/brier_reward/group_std_mean": 0.0998562067747116,
"signal/brier_reward/group_zero_std_frac": 0.2777777761220932,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.036421697959303855,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.036421697959303855,
"signal/confidence_one_or_zero/centered_abs_mean": 0.006461588526144623,
"signal/confidence_one_or_zero/group_std_mean": 0.009746908582746983,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9666666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 6.461588242245853e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 6.461588242245853e-08,
"signal/format_reward/centered_abs_mean": 0.020111762173473834,
"signal/format_reward/group_std_mean": 0.0422509741038084,
"signal/format_reward/group_zero_std_frac": 0.8083333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010055881086736917,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010055881086736917,
"signal/mean_confidence_reward/centered_abs_mean": 0.06666777282953262,
"signal/mean_confidence_reward/group_std_mean": 0.0885131299495697,
"signal/mean_confidence_reward/group_zero_std_frac": 0.28611111342906953,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.666776812380704e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.666776812380704e-07,
"step": 545
},
{
"calibration/aurc": 0.16459503744651127,
"calibration/batch_distribution_entropy": 0.5923094750073634,
"calibration/batch_entropy_100bins": 0.3324064309706861,
"calibration/batch_entropy_10bins": 0.5923094750073634,
"calibration/batch_entropy_50bins": 0.3913034722476989,
"calibration/batch_uniqueness": 0.3446194186815835,
"calibration/confidence_entropy": 0.47171471433285006,
"calibration/coverage@0%": 0.146875,
"calibration/coverage@1%": 0.17760416666666667,
"calibration/coverage@10%": 0.5500827271430608,
"calibration/coverage@15%": 0.5785037797746398,
"calibration/coverage@20%": 0.6137341320781629,
"calibration/coverage@25%": 0.6347867636571103,
"calibration/coverage@30%": 0.7637341320781628,
"calibration/coverage@5%": 0.48028352767080296,
"calibration/distribution_entropy_10": 0.5923094750073634,
"calibration/distribution_entropy_100": 0.3324064309706861,
"calibration/ece": 0.11403086741667849,
"calibration/mean_confidence": 0.6745715935969667,
"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": 0.009027777777777768,
"completions/max_length": 3682.8,
"completions/max_terminated_length": 3682.8,
"completions/mean_length": 1073.9698120117187,
"completions/mean_terminated_length": 1084.0141479492188,
"completions/min_length": 0.0,
"completions/min_terminated_length": 306.4,
"epoch": 1.3221153846153846,
"grad_norm": 0.00048103288281708956,
"learning_rate": 2.9447115384615386e-06,
"loss": -0.0103,
"num_tokens": 1379368762.0,
"reward": 1.2940167665481568,
"reward_std": 0.11286754608154297,
"rewards/accuracy_reward": 0.7332465410232544,
"rewards/brier_reward": 0.863887631893158,
"rewards/confidence_one_or_zero": 0.003125000064028427,
"rewards/format_reward": 0.9908854007720947,
"rewards/mean_confidence_reward": 0.6962173104286193,
"sampling/batch_mean_priority_error": 0.017588541666666645,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.24722222222222223,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0019167169695720077,
"sampling/priority_kl": 0.030000125616788866,
"sampling/priority_scale": 0.7219337046379224,
"sampling/prob_entropy": 10.278951835632324,
"sampling/prob_max": 4.9601865612203255e-05,
"sampling/prob_min": 1.964022740139626e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3152000188827515,
"sampling/prompt_draws_total": 39456.0,
"sampling/seen_fraction": 0.8097066640853882,
"sampling/unseen_fraction": 0.1902933359146118,
"signal/accuracy_reward/centered_abs_mean": 0.10252278447151184,
"signal/accuracy_reward/group_std_mean": 0.14087741523981095,
"signal/accuracy_reward/group_zero_std_frac": 0.575000011920929,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05126139223575592,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05126139223575592,
"signal/advantage_abs_mean": 0.07567564025521278,
"signal/advantage_pre_scale_abs_mean": 0.07567564025521278,
"signal/advantage_pre_scale_std": 0.16293342113494874,
"signal/advantage_std": 0.16293342113494874,
"signal/brier_reward/centered_abs_mean": 0.07278087660670281,
"signal/brier_reward/group_std_mean": 0.10256119221448898,
"signal/brier_reward/group_zero_std_frac": 0.2250000059604645,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.036390438303351404,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.036390438303351404,
"signal/confidence_one_or_zero/centered_abs_mean": 0.004318576550576836,
"signal/confidence_one_or_zero/group_std_mean": 0.00713206510990858,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.3185764297959395e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.3185764297959395e-08,
"signal/format_reward/centered_abs_mean": 0.01661783866584301,
"signal/format_reward/group_std_mean": 0.035409700125455856,
"signal/format_reward/group_zero_std_frac": 0.8416666626930237,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008308919332921504,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008308919332921504,
"signal/mean_confidence_reward/centered_abs_mean": 0.06690583825111389,
"signal/mean_confidence_reward/group_std_mean": 0.08986445367336274,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2416666716337204,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.690583632007474e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.690583632007474e-07,
"step": 550
},
{
"epoch": 1.3221153846153846,
"eval_calibration/aurc": 0.14916881488210604,
"eval_calibration/batch_distribution_entropy": 0.5969406634321193,
"eval_calibration/batch_entropy_100bins": 0.33255700130558596,
"eval_calibration/batch_entropy_10bins": 0.5969406634321193,
"eval_calibration/batch_entropy_50bins": 0.39148072120973537,
"eval_calibration/batch_uniqueness": 0.30708720276119833,
"eval_calibration/confidence_entropy": 0.44911964238701935,
"eval_calibration/coverage@0%": 0.004366812227074236,
"eval_calibration/coverage@1%": 0.004366812227074236,
"eval_calibration/coverage@10%": 0.004366812227074236,
"eval_calibration/coverage@15%": 0.7013100436681222,
"eval_calibration/coverage@20%": 0.7449781659388647,
"eval_calibration/coverage@25%": 0.8698689956331878,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.004366812227074236,
"eval_calibration/distribution_entropy_10": 0.5969406634321193,
"eval_calibration/distribution_entropy_100": 0.33255700130558596,
"eval_calibration/ece": 0.0260611353711793,
"eval_calibration/mean_confidence": 0.7135720524017469,
"eval_calibration/unique_confidence_per_question": 0.010416666666666666,
"eval_calibration/unique_confidences": 12,
"eval_completions/clipped_ratio": 0.005208333333333352,
"eval_completions/max_length": 3089.1666666666665,
"eval_completions/max_terminated_length": 3089.1666666666665,
"eval_completions/mean_length": 1039.5005696614583,
"eval_completions/mean_terminated_length": 1044.9644978841145,
"eval_completions/min_length": 121.83333333333333,
"eval_completions/min_terminated_length": 337.0,
"eval_loss": 0.0,
"eval_num_tokens": 1379368762.0,
"eval_reward": 1.2735520402590434,
"eval_reward_std": 0.3102180262406667,
"eval_rewards/accuracy_reward": 0.7048611044883728,
"eval_rewards/brier_reward": 0.8483050564924876,
"eval_rewards/confidence_one_or_zero": 0.004340277907128136,
"eval_rewards/format_reward": 0.9939236243565878,
"eval_rewards/mean_confidence_reward": 0.7092360655466715,
"eval_runtime": 206.7898,
"eval_samples_per_second": 4.836,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4040798594554265,
"eval_signal/accuracy_reward/group_std_mean": 0.454540491104126,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20203992972771326,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.20203992972771326,
"eval_signal/advantage_abs_mean": 0.26199281215667725,
"eval_signal/advantage_pre_scale_abs_mean": 0.26199281215667725,
"eval_signal/advantage_pre_scale_std": 0.3097561299800873,
"eval_signal/advantage_std": 0.3097561299800873,
"eval_signal/brier_reward/centered_abs_mean": 0.16876979420582452,
"eval_signal/brier_reward/group_std_mean": 0.22740193704764047,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08438489710291226,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08438489710291226,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.008409287935743729,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.02455231888840596,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 0.8611111342906952,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 8.409287488575501e-08,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.409287488575501e-08,
"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": 0.8333333631356558,
"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": 0.21962020794550577,
"eval_signal/mean_confidence_reward/group_std_mean": 0.25181200603644055,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.196201952150053e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.196201952150053e-06,
"eval_steps_per_second": 0.029,
"step": 550
},
{
"epoch": 1.3221153846153846,
"step": 550,
"train_probe_calibration/aurc": 0.10523588155297994,
"train_probe_calibration/batch_distribution_entropy": 0.6093510187951117,
"train_probe_calibration/batch_entropy_100bins": 0.34401648735790014,
"train_probe_calibration/batch_entropy_10bins": 0.6093510187951117,
"train_probe_calibration/batch_entropy_50bins": 0.40497064277758876,
"train_probe_calibration/batch_uniqueness": 0.33258799228299335,
"train_probe_calibration/confidence_entropy": 0.44647008666728183,
"train_probe_calibration/coverage@0%": 0.013054830287206266,
"train_probe_calibration/coverage@1%": 0.013054830287206266,
"train_probe_calibration/coverage@10%": 0.6736292428198434,
"train_probe_calibration/coverage@15%": 0.763272410791993,
"train_probe_calibration/coverage@20%": 0.8790252393385553,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.013054830287206266,
"train_probe_calibration/distribution_entropy_10": 0.6093510187951117,
"train_probe_calibration/distribution_entropy_100": 0.34401648735790014,
"train_probe_calibration/ece": 0.041583986074847604,
"train_probe_calibration/mean_confidence": 0.7203829416884248,
"train_probe_calibration/unique_confidence_per_question": 0.011284722222222222,
"train_probe_calibration/unique_confidences": 13,
"train_probe_completions/clipped_ratio": 0.0026041666666666665,
"train_probe_completions/max_length": 2821.5,
"train_probe_completions/max_terminated_length": 2821.5,
"train_probe_completions/mean_length": 1065.775899251302,
"train_probe_completions/mean_terminated_length": 1068.367899576823,
"train_probe_completions/min_length": 219.33333333333334,
"train_probe_completions/min_terminated_length": 329.3333333333333,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1379368762.0,
"train_probe_reward": 1.3099865317344666,
"train_probe_reward_std": 0.2795550574858983,
"train_probe_rewards/accuracy_reward": 0.7500000099341074,
"train_probe_rewards/brier_reward": 0.8725626766681671,
"train_probe_rewards/confidence_one_or_zero": 0.008680555736646056,
"train_probe_rewards/format_reward": 0.9973958333333334,
"train_probe_rewards/mean_confidence_reward": 0.7185069024562836,
"train_probe_runtime": 173.9441,
"train_probe_samples_per_second": 5.749,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3633897503217061,
"train_probe_signal/accuracy_reward/group_std_mean": 0.42989715933799744,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.18169487516085306,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.18169487516085306,
"train_probe_signal/advantage_abs_mean": 0.22928122927745184,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.22928122927745184,
"train_probe_signal/advantage_pre_scale_std": 0.2797047644853592,
"train_probe_signal/advantage_std": 0.2797047644853592,
"train_probe_signal/brier_reward/centered_abs_mean": 0.14238324761390686,
"train_probe_signal/brier_reward/group_std_mean": 0.19753756870826086,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07119162380695343,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07119162380695343,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.01671006918574373,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.04611522859583298,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 0.750000019868215,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6710068469668235e-07,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6710068469668235e-07,
"train_probe_signal/format_reward/centered_abs_mean": 0.0049370659204820795,
"train_probe_signal/format_reward/group_std_mean": 0.011741982462505499,
"train_probe_signal/format_reward/group_zero_std_frac": 0.944444457689921,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0024685329602410397,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.0024685329602410397,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.21267250428597131,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.24548349777857462,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.1267249697605925e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.1267249697605925e-06,
"train_probe_steps_per_second": 0.034
},
{
"calibration/aurc": 0.13244701628255368,
"calibration/batch_distribution_entropy": 0.5633908185400199,
"calibration/batch_entropy_100bins": 0.32076954253274204,
"calibration/batch_entropy_10bins": 0.5633908185400199,
"calibration/batch_entropy_50bins": 0.3776047154618286,
"calibration/batch_uniqueness": 0.2638795656491243,
"calibration/confidence_entropy": 0.4354364276359979,
"calibration/coverage@0%": 0.01723647697291977,
"calibration/coverage@1%": 0.01723647697291977,
"calibration/coverage@10%": 0.45358358508878655,
"calibration/coverage@15%": 0.6231473760474622,
"calibration/coverage@20%": 0.6979631730756086,
"calibration/coverage@25%": 0.8934800392670157,
"calibration/coverage@30%": 0.9400169066317627,
"calibration/coverage@5%": 0.3031577652317745,
"calibration/distribution_entropy_10": 0.5633908185400199,
"calibration/distribution_entropy_100": 0.32076954253274204,
"calibration/ece": 0.10251768482381678,
"calibration/mean_confidence": 0.7301818809265057,
"calibration/unique_confidence_per_question": 0.024479166666666666,
"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": 0.005468749999999978,
"completions/max_length": 3548.2,
"completions/max_terminated_length": 3548.2,
"completions/mean_length": 1017.2588500976562,
"completions/mean_terminated_length": 1022.870556640625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 303.4,
"epoch": 1.3341346153846154,
"grad_norm": 0.0005966307362541556,
"learning_rate": 2.9146634615384615e-06,
"loss": -0.0048,
"num_tokens": 1394187200.0,
"reward": 1.3012691497802735,
"reward_std": 0.11076888740062714,
"rewards/accuracy_reward": 0.746874988079071,
"rewards/brier_reward": 0.8611177206039429,
"rewards/confidence_one_or_zero": 0.00494791658129543,
"rewards/format_reward": 0.994531261920929,
"rewards/mean_confidence_reward": 0.713351571559906,
"sampling/batch_mean_priority_error": 0.01391326388888887,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.24999999999999994,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0019340271828696132,
"sampling/priority_kl": 0.03000037595629692,
"sampling/priority_scale": 0.7224328219657764,
"sampling/prob_entropy": 10.278961563110352,
"sampling/prob_max": 4.974793046130799e-05,
"sampling/prob_min": 1.9638664525700733e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3272000074386596,
"sampling/prompt_draws_total": 39816.0,
"sampling/seen_fraction": 0.8126399993896485,
"sampling/unseen_fraction": 0.18736000061035157,
"signal/accuracy_reward/centered_abs_mean": 0.11321614533662797,
"signal/accuracy_reward/group_std_mean": 0.1554545447230339,
"signal/accuracy_reward/group_zero_std_frac": 0.5333333313465118,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05660807266831398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05660807266831398,
"signal/advantage_abs_mean": 0.07871722057461739,
"signal/advantage_pre_scale_abs_mean": 0.07871722057461739,
"signal/advantage_pre_scale_std": 0.15928139686584472,
"signal/advantage_std": 0.15928139686584472,
"signal/brier_reward/centered_abs_mean": 0.07254311144351959,
"signal/brier_reward/group_std_mean": 0.09697409868240356,
"signal/brier_reward/group_zero_std_frac": 0.2527777761220932,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.036271555721759795,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.036271555721759795,
"signal/confidence_one_or_zero/centered_abs_mean": 0.006168619776144624,
"signal/confidence_one_or_zero/group_std_mean": 0.008891820348799229,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222328186035,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 6.168619535174002e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 6.168619535174002e-08,
"signal/format_reward/centered_abs_mean": 0.009749348927289247,
"signal/format_reward/group_std_mean": 0.021487992629408835,
"signal/format_reward/group_zero_std_frac": 0.8972222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004874674463644623,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004874674463644623,
"signal/mean_confidence_reward/centered_abs_mean": 0.06694743260741234,
"signal/mean_confidence_reward/group_std_mean": 0.0881627306342125,
"signal/mean_confidence_reward/group_zero_std_frac": 0.26388889253139497,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.694743092339195e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.694743092339195e-07,
"step": 555
},
{
"calibration/aurc": 0.18615071715847942,
"calibration/batch_distribution_entropy": 0.5391185474680029,
"calibration/batch_entropy_100bins": 0.3057328655456327,
"calibration/batch_entropy_10bins": 0.5391185474680029,
"calibration/batch_entropy_50bins": 0.3599037826039988,
"calibration/batch_uniqueness": 0.22894520104889549,
"calibration/confidence_entropy": 0.43995140877076694,
"calibration/coverage@0%": 0.0203604967718753,
"calibration/coverage@1%": 0.0203604967718753,
"calibration/coverage@10%": 0.28963133010520864,
"calibration/coverage@15%": 0.570893208075795,
"calibration/coverage@20%": 0.5835928841258398,
"calibration/coverage@25%": 0.8432911523191866,
"calibration/coverage@30%": 0.8639158847184987,
"calibration/coverage@5%": 0.27608966343854197,
"calibration/distribution_entropy_10": 0.5391185474680029,
"calibration/distribution_entropy_100": 0.3057328655456327,
"calibration/ece": 0.14419091037754006,
"calibration/mean_confidence": 0.731739201801835,
"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": 0.004600694444444419,
"completions/max_length": 3374.8,
"completions/max_terminated_length": 3374.8,
"completions/mean_length": 983.3083251953125,
"completions/mean_terminated_length": 987.8928466796875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 287.2,
"epoch": 1.3461538461538463,
"grad_norm": 0.0004465328238438815,
"learning_rate": 2.8846153846153845e-06,
"loss": -0.0045,
"num_tokens": 1408643552.0,
"reward": 1.2871931076049805,
"reward_std": 0.10639832764863968,
"rewards/accuracy_reward": 0.7247395753860474,
"rewards/brier_reward": 0.8542326092720032,
"rewards/confidence_one_or_zero": 0.00512152787996456,
"rewards/format_reward": 0.9953993082046508,
"rewards/mean_confidence_reward": 0.734346342086792,
"sampling/batch_mean_priority_error": 0.018213541666666656,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.30833333333333335,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0019512236351147294,
"sampling/priority_kl": 0.03000052720308304,
"sampling/priority_scale": 0.7237809360725805,
"sampling/prob_entropy": 10.278968620300294,
"sampling/prob_max": 4.992713656974957e-05,
"sampling/prob_min": 1.960492882062681e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.339199995994568,
"sampling/prompt_draws_total": 40176.0,
"sampling/seen_fraction": 0.816160011291504,
"sampling/unseen_fraction": 0.1838399887084961,
"signal/accuracy_reward/centered_abs_mean": 0.10935872346162796,
"signal/accuracy_reward/group_std_mean": 0.14367188811302184,
"signal/accuracy_reward/group_zero_std_frac": 0.5888888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05467936173081398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05467936173081398,
"signal/advantage_abs_mean": 0.07600348442792892,
"signal/advantage_pre_scale_abs_mean": 0.07600348442792892,
"signal/advantage_pre_scale_std": 0.15933072566986084,
"signal/advantage_std": 0.15933072566986084,
"signal/brier_reward/centered_abs_mean": 0.06570380330085754,
"signal/brier_reward/group_std_mean": 0.09007837623357773,
"signal/brier_reward/group_zero_std_frac": 0.27777778208255766,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03285190165042877,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03285190165042877,
"signal/confidence_one_or_zero/centered_abs_mean": 0.00596245659980923,
"signal/confidence_one_or_zero/group_std_mean": 0.009446198446676135,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9666666746139526,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.962456164354535e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.962456164354535e-08,
"signal/format_reward/centered_abs_mean": 0.008490668423473836,
"signal/format_reward/group_std_mean": 0.019875646382570267,
"signal/format_reward/group_zero_std_frac": 0.9027777791023255,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004245334211736918,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004245334211736918,
"signal/mean_confidence_reward/centered_abs_mean": 0.05294933393597603,
"signal/mean_confidence_reward/group_std_mean": 0.0712207317352295,
"signal/mean_confidence_reward/group_zero_std_frac": 0.30833333134651186,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.294933203003893e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.294933203003893e-07,
"step": 560
},
{
"calibration/aurc": 0.12142111368921756,
"calibration/batch_distribution_entropy": 0.5238570877525583,
"calibration/batch_entropy_100bins": 0.28641818336889885,
"calibration/batch_entropy_10bins": 0.5238570877525583,
"calibration/batch_entropy_50bins": 0.3371668512544869,
"calibration/batch_uniqueness": 0.17762239623692094,
"calibration/confidence_entropy": 0.44057883362318667,
"calibration/coverage@0%": 0.010473959768531277,
"calibration/coverage@1%": 0.010473959768531277,
"calibration/coverage@10%": 0.5173849545329292,
"calibration/coverage@15%": 0.7562441443923946,
"calibration/coverage@20%": 0.8324249104436483,
"calibration/coverage@25%": 0.8717277486910995,
"calibration/coverage@30%": 0.9293193717277486,
"calibration/coverage@5%": 0.2633535409203637,
"calibration/distribution_entropy_10": 0.5238570877525583,
"calibration/distribution_entropy_100": 0.28641818336889885,
"calibration/ece": 0.13695384403416913,
"calibration/mean_confidence": 0.7534154036924774,
"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": 0.00512152777777779,
"completions/max_length": 3400.0,
"completions/max_terminated_length": 3400.0,
"completions/mean_length": 970.3908081054688,
"completions/mean_terminated_length": 975.3807373046875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 275.6,
"epoch": 1.3581730769230769,
"grad_norm": 0.0005166183691471815,
"learning_rate": 2.8545673076923082e-06,
"loss": -0.0057,
"num_tokens": 1422888790.0,
"reward": 1.2914889097213744,
"reward_std": 0.10451765954494477,
"rewards/accuracy_reward": 0.7174479246139527,
"rewards/brier_reward": 0.870636236667633,
"rewards/confidence_one_or_zero": 0.008940972248092293,
"rewards/format_reward": 0.9948784828186035,
"rewards/mean_confidence_reward": 0.750803804397583,
"sampling/batch_mean_priority_error": 0.015555694444444431,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2861111111111111,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0019691059831529857,
"sampling/priority_kl": 0.03000061921775341,
"sampling/priority_scale": 0.7250464618904516,
"sampling/prob_entropy": 10.278957557678222,
"sampling/prob_max": 5.010353270336054e-05,
"sampling/prob_min": 1.9642066035885364e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.351199984550476,
"sampling/prompt_draws_total": 40536.0,
"sampling/seen_fraction": 0.8195199966430664,
"sampling/unseen_fraction": 0.1804800033569336,
"signal/accuracy_reward/centered_abs_mean": 0.09441731572151184,
"signal/accuracy_reward/group_std_mean": 0.1311457559466362,
"signal/accuracy_reward/group_zero_std_frac": 0.6027777791023254,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04720865786075592,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04720865786075592,
"signal/advantage_abs_mean": 0.07170857340097428,
"signal/advantage_pre_scale_abs_mean": 0.07170857340097428,
"signal/advantage_pre_scale_std": 0.15759305655956268,
"signal/advantage_std": 0.15759305655956268,
"signal/brier_reward/centered_abs_mean": 0.06307352632284165,
"signal/brier_reward/group_std_mean": 0.08904340118169785,
"signal/brier_reward/group_zero_std_frac": 0.2805555582046509,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.031536763161420824,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.031536763161420824,
"signal/confidence_one_or_zero/centered_abs_mean": 0.010639105923473835,
"signal/confidence_one_or_zero/group_std_mean": 0.01555493324995041,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.95,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.0639105525456216e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.0639105525456216e-07,
"signal/format_reward/centered_abs_mean": 0.009662543423473835,
"signal/format_reward/group_std_mean": 0.02305259294807911,
"signal/format_reward/group_zero_std_frac": 0.8888888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0048312717117369175,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0048312717117369175,
"signal/mean_confidence_reward/centered_abs_mean": 0.057669497281312945,
"signal/mean_confidence_reward/group_std_mean": 0.07707741409540177,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3027777850627899,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.766949357166595e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.766949357166595e-07,
"step": 565
},
{
"calibration/aurc": 0.11921036526172504,
"calibration/batch_distribution_entropy": 0.5480813620266501,
"calibration/batch_entropy_100bins": 0.30107017435370487,
"calibration/batch_entropy_10bins": 0.5480813620266501,
"calibration/batch_entropy_50bins": 0.35441493797457274,
"calibration/batch_uniqueness": 0.22293690516241052,
"calibration/confidence_entropy": 0.44491889967539644,
"calibration/coverage@0%": 0.00939947780678851,
"calibration/coverage@1%": 0.1324361270214482,
"calibration/coverage@10%": 0.5075941861577788,
"calibration/coverage@15%": 0.576529271982466,
"calibration/coverage@20%": 0.7843186540538324,
"calibration/coverage@25%": 0.9754745974760661,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.27188905444752776,
"calibration/distribution_entropy_10": 0.5480813620266501,
"calibration/distribution_entropy_100": 0.30107017435370487,
"calibration/ece": 0.09685982504135167,
"calibration/mean_confidence": 0.7530256108487235,
"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": 0.002604166666666674,
"completions/max_length": 3331.2,
"completions/max_terminated_length": 3331.2,
"completions/mean_length": 920.540380859375,
"completions/mean_terminated_length": 922.9418823242188,
"completions/min_length": 0.0,
"completions/min_terminated_length": 261.8,
"epoch": 1.3701923076923077,
"grad_norm": 0.0005127664771862328,
"learning_rate": 2.8245192307692307e-06,
"loss": -0.0021,
"num_tokens": 1436592839.0,
"reward": 1.3091936588287354,
"reward_std": 0.10138077139854432,
"rewards/accuracy_reward": 0.7527777671813964,
"rewards/brier_reward": 0.868198013305664,
"rewards/confidence_one_or_zero": 0.014236111333593727,
"rewards/format_reward": 0.9973958373069763,
"rewards/mean_confidence_reward": 0.7676869392395019,
"sampling/batch_mean_priority_error": 0.017774374999999985,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.28888888888888886,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0019868580624461175,
"sampling/priority_kl": 0.030000395700335504,
"sampling/priority_scale": 0.7266766966087743,
"sampling/prob_entropy": 10.27896671295166,
"sampling/prob_max": 5.02947841596324e-05,
"sampling/prob_min": 1.9675568910315633e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3631999969482422,
"sampling/prompt_draws_total": 40896.0,
"sampling/seen_fraction": 0.8230933308601379,
"sampling/unseen_fraction": 0.17690666913986205,
"signal/accuracy_reward/centered_abs_mean": 0.09973958432674408,
"signal/accuracy_reward/group_std_mean": 0.13625123351812363,
"signal/accuracy_reward/group_zero_std_frac": 0.5944444537162781,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04986979216337204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04986979216337204,
"signal/advantage_abs_mean": 0.07208630740642548,
"signal/advantage_pre_scale_abs_mean": 0.07208630740642548,
"signal/advantage_pre_scale_std": 0.15612260699272157,
"signal/advantage_std": 0.15612260699272157,
"signal/brier_reward/centered_abs_mean": 0.06025089025497436,
"signal/brier_reward/group_std_mean": 0.08231211751699448,
"signal/brier_reward/group_zero_std_frac": 0.32222222685813906,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03012544512748718,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03012544512748718,
"signal/confidence_one_or_zero/centered_abs_mean": 0.013237847201526164,
"signal/confidence_one_or_zero/group_std_mean": 0.01707735490053892,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9527777791023254,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.3237846872016234e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.3237846872016234e-07,
"signal/format_reward/centered_abs_mean": 0.0049153645522892475,
"signal/format_reward/group_std_mean": 0.012703005410730838,
"signal/format_reward/group_zero_std_frac": 0.9333333492279052,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0024576822761446238,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0024576822761446238,
"signal/mean_confidence_reward/centered_abs_mean": 0.05037656500935554,
"signal/mean_confidence_reward/group_std_mean": 0.0677335023880005,
"signal/mean_confidence_reward/group_zero_std_frac": 0.33611111640930175,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.037656308104487e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.037656308104487e-07,
"step": 570
},
{
"calibration/aurc": 0.14634278778879556,
"calibration/batch_distribution_entropy": 0.47070895536337715,
"calibration/batch_entropy_100bins": 0.2672472458527859,
"calibration/batch_entropy_10bins": 0.47070895536337715,
"calibration/batch_entropy_50bins": 0.31459913379368903,
"calibration/batch_uniqueness": 0.05658321692658895,
"calibration/confidence_entropy": 0.41867838259744794,
"calibration/coverage@0%": 0.11620439353136577,
"calibration/coverage@1%": 0.11620439353136577,
"calibration/coverage@10%": 0.32988513087866067,
"calibration/coverage@15%": 0.5988134457916969,
"calibration/coverage@20%": 0.7043593564173719,
"calibration/coverage@25%": 0.9228826697127938,
"calibration/coverage@30%": 0.9359375,
"calibration/coverage@5%": 0.13348188044236053,
"calibration/distribution_entropy_10": 0.47070895536337715,
"calibration/distribution_entropy_100": 0.2672472458527859,
"calibration/ece": 0.12027931569792079,
"calibration/mean_confidence": 0.7783694586745135,
"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": 3425.6,
"completions/max_terminated_length": 3425.6,
"completions/mean_length": 931.355126953125,
"completions/mean_terminated_length": 934.3935180664063,
"completions/min_length": 0.0,
"completions/min_terminated_length": 263.4,
"epoch": 1.3822115384615383,
"grad_norm": 0.0005164265749044716,
"learning_rate": 2.7944711538461537e-06,
"loss": -0.0033,
"num_tokens": 1450395042.0,
"reward": 1.2989335775375366,
"reward_std": 0.10441743582487106,
"rewards/accuracy_reward": 0.7419270873069763,
"rewards/brier_reward": 0.8592235565185546,
"rewards/confidence_one_or_zero": 0.006423611292848364,
"rewards/format_reward": 0.9967013955116272,
"rewards/mean_confidence_reward": 0.7564403891563416,
"sampling/batch_mean_priority_error": 0.0193909722222222,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.24722222222222223,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0020075502339750527,
"sampling/priority_kl": 0.02999947778880596,
"sampling/priority_scale": 0.7279814063804224,
"sampling/prob_entropy": 10.278944969177246,
"sampling/prob_max": 5.0473969895392654e-05,
"sampling/prob_min": 1.9712427456397563e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3751999855041503,
"sampling/prompt_draws_total": 41256.0,
"sampling/seen_fraction": 0.8263200044631958,
"sampling/unseen_fraction": 0.1736799955368042,
"signal/accuracy_reward/centered_abs_mean": 0.1043891042470932,
"signal/accuracy_reward/group_std_mean": 0.1397392600774765,
"signal/accuracy_reward/group_zero_std_frac": 0.5944444537162781,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0521945521235466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0521945521235466,
"signal/advantage_abs_mean": 0.07473682761192321,
"signal/advantage_pre_scale_abs_mean": 0.07473682761192321,
"signal/advantage_pre_scale_std": 0.159325709939003,
"signal/advantage_std": 0.159325709939003,
"signal/brier_reward/centered_abs_mean": 0.065924521535635,
"signal/brier_reward/group_std_mean": 0.09040133357048034,
"signal/brier_reward/group_zero_std_frac": 0.29444444477558135,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0329622607678175,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0329622607678175,
"signal/confidence_one_or_zero/centered_abs_mean": 0.005121527786832303,
"signal/confidence_one_or_zero/group_std_mean": 0.007448598276823759,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9749999880790711,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.121527379969848e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.121527379969848e-08,
"signal/format_reward/centered_abs_mean": 0.006043836660683155,
"signal/format_reward/group_std_mean": 0.013839713856577873,
"signal/format_reward/group_zero_std_frac": 0.9333333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0030219183303415776,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0030219183303415776,
"signal/mean_confidence_reward/centered_abs_mean": 0.054480834305286406,
"signal/mean_confidence_reward/group_std_mean": 0.07193099856376647,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3111111164093018,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.448083584269625e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.448083584269625e-07,
"step": 575
},
{
"calibration/aurc": 0.08932079073471824,
"calibration/batch_distribution_entropy": 0.5381646310222294,
"calibration/batch_entropy_100bins": 0.2897085398638714,
"calibration/batch_entropy_10bins": 0.5381646310222294,
"calibration/batch_entropy_50bins": 0.34104020568284665,
"calibration/batch_uniqueness": 0.21316519945185397,
"calibration/confidence_entropy": 0.4485409395001082,
"calibration/coverage@0%": 0.09574185902138946,
"calibration/coverage@1%": 0.40310812781007205,
"calibration/coverage@10%": 0.6712954711336622,
"calibration/coverage@15%": 0.790268415175699,
"calibration/coverage@20%": 0.8702820072348633,
"calibration/coverage@25%": 0.9377659574468085,
"calibration/coverage@30%": 0.9377659574468085,
"calibration/coverage@5%": 0.5254277448234277,
"calibration/distribution_entropy_10": 0.5381646310222294,
"calibration/distribution_entropy_100": 0.2897085398638714,
"calibration/ece": 0.12022005527506718,
"calibration/mean_confidence": 0.7487579877428203,
"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": 0.006510416666666674,
"completions/max_length": 3385.4,
"completions/max_terminated_length": 3385.4,
"completions/mean_length": 913.8025146484375,
"completions/mean_terminated_length": 919.8221069335938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 242.6,
"epoch": 1.3942307692307692,
"grad_norm": 0.0005168268107809126,
"learning_rate": 2.7644230769230775e-06,
"loss": -0.0064,
"num_tokens": 1464024639.0,
"reward": 1.2927096128463744,
"reward_std": 0.10726402699947357,
"rewards/accuracy_reward": 0.7303819537162781,
"rewards/brier_reward": 0.8615329027175903,
"rewards/confidence_one_or_zero": 0.0038194443448446693,
"rewards/format_reward": 0.9934895753860473,
"rewards/mean_confidence_reward": 0.737480640411377,
"sampling/batch_mean_priority_error": 0.0157292361111111,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.21388888888888888,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0020283636171370746,
"sampling/priority_kl": 0.03000028133392334,
"sampling/priority_scale": 0.7286482155555859,
"sampling/prob_entropy": 10.278956604003906,
"sampling/prob_max": 5.0625135918380694e-05,
"sampling/prob_min": 1.9754447203013113e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3871999979019165,
"sampling/prompt_draws_total": 41616.0,
"sampling/seen_fraction": 0.8290400028228759,
"sampling/unseen_fraction": 0.170959997177124,
"signal/accuracy_reward/centered_abs_mean": 0.10659722238779068,
"signal/accuracy_reward/group_std_mean": 0.14086600244045258,
"signal/accuracy_reward/group_zero_std_frac": 0.5916666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05329861119389534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05329861119389534,
"signal/advantage_abs_mean": 0.07496702522039414,
"signal/advantage_pre_scale_abs_mean": 0.07496702522039414,
"signal/advantage_pre_scale_std": 0.16025392711162567,
"signal/advantage_std": 0.16025392711162567,
"signal/brier_reward/centered_abs_mean": 0.06117345690727234,
"signal/brier_reward/group_std_mean": 0.08554955422878266,
"signal/brier_reward/group_zero_std_frac": 0.286111119389534,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03058672845363617,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03058672845363617,
"signal/confidence_one_or_zero/centered_abs_mean": 0.004839409724809229,
"signal/confidence_one_or_zero/group_std_mean": 0.006505797291174531,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9805555462837219,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.839409761814295e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.839409761814295e-08,
"signal/format_reward/centered_abs_mean": 0.01199544258415699,
"signal/format_reward/group_std_mean": 0.027937114983797074,
"signal/format_reward/group_zero_std_frac": 0.8638888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005997721292078495,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005997721292078495,
"signal/mean_confidence_reward/centered_abs_mean": 0.05558514967560768,
"signal/mean_confidence_reward/group_std_mean": 0.07502751946449279,
"signal/mean_confidence_reward/group_zero_std_frac": 0.29444445073604586,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.55851465833257e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.55851465833257e-07,
"step": 580
},
{
"calibration/aurc": 0.13578598524695468,
"calibration/batch_distribution_entropy": 0.6923252720826605,
"calibration/batch_entropy_100bins": 0.38076604303053696,
"calibration/batch_entropy_10bins": 0.6923252720826605,
"calibration/batch_entropy_50bins": 0.44823162511258763,
"calibration/batch_uniqueness": 0.5211813855646842,
"calibration/confidence_entropy": 0.4883225648832853,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5262217544051508,
"calibration/coverage@15%": 0.6181509985919922,
"calibration/coverage@20%": 0.7440733804491955,
"calibration/coverage@25%": 0.8506527415143603,
"calibration/coverage@30%": 0.8908616187989556,
"calibration/coverage@5%": 0.1427678974204749,
"calibration/distribution_entropy_10": 0.6923252720826605,
"calibration/distribution_entropy_100": 0.38076604303053696,
"calibration/ece": 0.11659514989132358,
"calibration/mean_confidence": 0.6826111711071315,
"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": 0.00434027777777779,
"completions/max_length": 3558.6,
"completions/max_terminated_length": 3558.6,
"completions/mean_length": 931.40166015625,
"completions/mean_terminated_length": 935.4742553710937,
"completions/min_length": 0.0,
"completions/min_terminated_length": 292.0,
"epoch": 1.40625,
"grad_norm": 0.0004743568133562803,
"learning_rate": 2.7343750000000004e-06,
"loss": -0.0033,
"num_tokens": 1477845746.0,
"reward": 1.300478959083557,
"reward_std": 0.10303066670894623,
"rewards/accuracy_reward": 0.7401909589767456,
"rewards/brier_reward": 0.8650924324989319,
"rewards/confidence_one_or_zero": 0.001996527798473835,
"rewards/format_reward": 0.9956597328186035,
"rewards/mean_confidence_reward": 0.7386232614517212,
"sampling/batch_mean_priority_error": 0.017812638888888878,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.25555555555555554,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.00204639732837677,
"sampling/priority_kl": 0.029999741166830064,
"sampling/priority_scale": 0.7295177638297901,
"sampling/prob_entropy": 10.278957939147949,
"sampling/prob_max": 5.07840006321203e-05,
"sampling/prob_min": 1.9794189211097545e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3992000102996827,
"sampling/prompt_draws_total": 41976.0,
"sampling/seen_fraction": 0.8318466663360595,
"sampling/unseen_fraction": 0.16815333366394042,
"signal/accuracy_reward/centered_abs_mean": 0.097607421875,
"signal/accuracy_reward/group_std_mean": 0.13638707101345063,
"signal/accuracy_reward/group_zero_std_frac": 0.5833333492279053,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0488037109375,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0488037109375,
"signal/advantage_abs_mean": 0.07051034420728683,
"signal/advantage_pre_scale_abs_mean": 0.07051034420728683,
"signal/advantage_pre_scale_std": 0.15343783795833588,
"signal/advantage_std": 0.15343783795833588,
"signal/brier_reward/centered_abs_mean": 0.05911755934357643,
"signal/brier_reward/group_std_mean": 0.08391907587647437,
"signal/brier_reward/group_zero_std_frac": 0.27777778208255766,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029558779671788216,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029558779671788216,
"signal/confidence_one_or_zero/centered_abs_mean": 0.001578775979578495,
"signal/confidence_one_or_zero/group_std_mean": 0.0020235927775502207,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.5787759366503452e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.5787759366503452e-08,
"signal/format_reward/centered_abs_mean": 0.008083767257630825,
"signal/format_reward/group_std_mean": 0.01864474155008793,
"signal/format_reward/group_zero_std_frac": 0.9111111283302307,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0040418836288154125,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0040418836288154125,
"signal/mean_confidence_reward/centered_abs_mean": 0.05392475575208664,
"signal/mean_confidence_reward/group_std_mean": 0.07254645675420761,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2916666716337204,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.392475145526987e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.392475145526987e-07,
"step": 585
},
{
"calibration/aurc": 0.12619754771127717,
"calibration/batch_distribution_entropy": 0.5520588005609969,
"calibration/batch_entropy_100bins": 0.3022191708991097,
"calibration/batch_entropy_10bins": 0.5520588005609969,
"calibration/batch_entropy_50bins": 0.355767518117215,
"calibration/batch_uniqueness": 0.25986767018800117,
"calibration/confidence_entropy": 0.44241364927549753,
"calibration/coverage@0%": 0.1192790139616056,
"calibration/coverage@1%": 0.1192790139616056,
"calibration/coverage@10%": 0.39546247818499125,
"calibration/coverage@15%": 0.609375,
"calibration/coverage@20%": 0.7259080497382199,
"calibration/coverage@25%": 0.8811709205933681,
"calibration/coverage@30%": 0.9625,
"calibration/coverage@5%": 0.3761480148342059,
"calibration/distribution_entropy_10": 0.5520588005609969,
"calibration/distribution_entropy_100": 0.3022191708991097,
"calibration/ece": 0.10985983856893541,
"calibration/mean_confidence": 0.7686387434554974,
"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": 0.004947916666666674,
"completions/max_length": 3448.6,
"completions/max_terminated_length": 3448.6,
"completions/mean_length": 858.2862915039062,
"completions/mean_terminated_length": 862.65830078125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 247.0,
"epoch": 1.4182692307692308,
"grad_norm": 0.0004786239587701857,
"learning_rate": 2.7043269230769233e-06,
"loss": -0.0049,
"num_tokens": 1490827028.0,
"reward": 1.3150073766708374,
"reward_std": 0.09183973670005799,
"rewards/accuracy_reward": 0.7597222208976746,
"rewards/brier_reward": 0.8752251267433167,
"rewards/confidence_one_or_zero": 0.006163194461259991,
"rewards/format_reward": 0.9950520873069764,
"rewards/mean_confidence_reward": 0.7613700985908508,
"sampling/batch_mean_priority_error": 0.018032986111111104,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2555555555555556,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.002064537163823843,
"sampling/priority_kl": 0.030000126734375954,
"sampling/priority_scale": 0.7309222042793408,
"sampling/prob_entropy": 10.278956604003906,
"sampling/prob_max": 5.0964941328857095e-05,
"sampling/prob_min": 1.9828715812764132e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.4111999988555908,
"sampling/prompt_draws_total": 42336.0,
"sampling/seen_fraction": 0.8349533319473267,
"sampling/unseen_fraction": 0.16504666805267335,
"signal/accuracy_reward/centered_abs_mean": 0.09001736044883728,
"signal/accuracy_reward/group_std_mean": 0.12046184241771699,
"signal/accuracy_reward/group_zero_std_frac": 0.6444444417953491,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04500868022441864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04500868022441864,
"signal/advantage_abs_mean": 0.06484012752771377,
"signal/advantage_pre_scale_abs_mean": 0.06484012752771377,
"signal/advantage_pre_scale_std": 0.14825278222560884,
"signal/advantage_std": 0.14825278222560884,
"signal/brier_reward/centered_abs_mean": 0.05299278721213341,
"signal/brier_reward/group_std_mean": 0.07479698657989502,
"signal/brier_reward/group_zero_std_frac": 0.2833333402872086,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.026496393606066704,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.026496393606066704,
"signal/confidence_one_or_zero/centered_abs_mean": 0.008360460097901524,
"signal/confidence_one_or_zero/group_std_mean": 0.012945984303951264,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9555555582046509,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 8.360459915479624e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.360459915479624e-08,
"signal/format_reward/centered_abs_mean": 0.008816189365461469,
"signal/format_reward/group_std_mean": 0.018386806827038527,
"signal/format_reward/group_zero_std_frac": 0.919444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004408094682730734,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004408094682730734,
"signal/mean_confidence_reward/centered_abs_mean": 0.04662701785564423,
"signal/mean_confidence_reward/group_std_mean": 0.06395412832498551,
"signal/mean_confidence_reward/group_zero_std_frac": 0.30277778804302213,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.66270154220183e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.66270154220183e-07,
"step": 590
},
{
"calibration/aurc": 0.13849873769615206,
"calibration/batch_distribution_entropy": 0.48137985436237296,
"calibration/batch_entropy_100bins": 0.2622942038657148,
"calibration/batch_entropy_10bins": 0.48137985436237296,
"calibration/batch_entropy_50bins": 0.3087684929060569,
"calibration/batch_uniqueness": 0.08955665497088608,
"calibration/confidence_entropy": 0.41628624550681853,
"calibration/coverage@0%": 0.12970925805047867,
"calibration/coverage@1%": 0.12970925805047867,
"calibration/coverage@10%": 0.44916095517841603,
"calibration/coverage@15%": 0.6550220300261097,
"calibration/coverage@20%": 0.7295827893820713,
"calibration/coverage@25%": 0.7760307876414274,
"calibration/coverage@30%": 0.9441253263707573,
"calibration/coverage@5%": 0.169292591383812,
"calibration/distribution_entropy_10": 0.48137985436237296,
"calibration/distribution_entropy_100": 0.2622942038657148,
"calibration/ece": 0.10869756309834637,
"calibration/mean_confidence": 0.802804395126197,
"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": 0.001822916666666674,
"completions/max_length": 3234.6,
"completions/max_terminated_length": 3234.6,
"completions/mean_length": 858.982373046875,
"completions/mean_terminated_length": 860.55595703125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 248.8,
"epoch": 1.4302884615384617,
"grad_norm": 0.0005619844305329025,
"learning_rate": 2.6742788461538467e-06,
"loss": -0.0012,
"num_tokens": 1503825673.0,
"reward": 1.3214678764343262,
"reward_std": 0.09593596607446671,
"rewards/accuracy_reward": 0.7690972208976745,
"rewards/brier_reward": 0.8758194446563721,
"rewards/confidence_one_or_zero": 0.008159722306299955,
"rewards/format_reward": 0.9980034708976746,
"rewards/mean_confidence_reward": 0.7763272404670716,
"sampling/batch_mean_priority_error": 0.013152847222222213,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2555555555555556,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.002083018142729998,
"sampling/priority_kl": 0.03000060245394707,
"sampling/priority_scale": 0.7321809828514233,
"sampling/prob_entropy": 10.278948402404785,
"sampling/prob_max": 5.113998267916031e-05,
"sampling/prob_min": 1.9864497153321282e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.423200011253357,
"sampling/prompt_draws_total": 42696.0,
"sampling/seen_fraction": 0.8378800034523011,
"sampling/unseen_fraction": 0.16211999654769899,
"signal/accuracy_reward/centered_abs_mean": 0.09106987863779067,
"signal/accuracy_reward/group_std_mean": 0.12727132737636565,
"signal/accuracy_reward/group_zero_std_frac": 0.6138888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04553493931889534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04553493931889534,
"signal/advantage_abs_mean": 0.06666973233222961,
"signal/advantage_pre_scale_abs_mean": 0.06666973233222961,
"signal/advantage_pre_scale_std": 0.1487389624118805,
"signal/advantage_std": 0.1487389624118805,
"signal/brier_reward/centered_abs_mean": 0.05577497556805611,
"signal/brier_reward/group_std_mean": 0.07818025052547455,
"signal/brier_reward/group_zero_std_frac": 0.2833333373069763,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.027887487784028053,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.027887487784028053,
"signal/confidence_one_or_zero/centered_abs_mean": 0.010763888782821595,
"signal/confidence_one_or_zero/group_std_mean": 0.015660046227276327,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9472222208976746,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.0763889122245018e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.0763889122245018e-07,
"signal/format_reward/centered_abs_mean": 0.0037706162431277333,
"signal/format_reward/group_std_mean": 0.00909621324390173,
"signal/format_reward/group_zero_std_frac": 0.9555555701255798,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0018853081215638666,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0018853081215638666,
"signal/mean_confidence_reward/centered_abs_mean": 0.04750293791294098,
"signal/mean_confidence_reward/group_std_mean": 0.06490557044744491,
"signal/mean_confidence_reward/group_zero_std_frac": 0.31388888955116273,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.750293499000691e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.750293499000691e-07,
"step": 595
},
{
"calibration/aurc": 0.11944654806083935,
"calibration/batch_distribution_entropy": 0.5310314355245698,
"calibration/batch_entropy_100bins": 0.30088690024296083,
"calibration/batch_entropy_10bins": 0.5310314355245698,
"calibration/batch_entropy_50bins": 0.35419919065675515,
"calibration/batch_uniqueness": 0.21813118524641628,
"calibration/confidence_entropy": 0.42552355286541294,
"calibration/coverage@0%": 0.008333333333333333,
"calibration/coverage@1%": 0.008333333333333333,
"calibration/coverage@10%": 0.5651928877142514,
"calibration/coverage@15%": 0.7183434071393817,
"calibration/coverage@20%": 0.7904855643044619,
"calibration/coverage@25%": 0.8083333333333332,
"calibration/coverage@30%": 0.9651041666666668,
"calibration/coverage@5%": 0.2861527943823776,
"calibration/distribution_entropy_10": 0.5310314355245698,
"calibration/distribution_entropy_100": 0.30088690024296083,
"calibration/ece": 0.11940171993680018,
"calibration/mean_confidence": 0.7663529327918157,
"calibration/unique_confidence_per_question": 0.02395833333333333,
"calibration/unique_confidences": 9.2,
"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": 0.0021701388888888838,
"completions/max_length": 3184.2,
"completions/max_terminated_length": 3184.2,
"completions/mean_length": 854.9868896484375,
"completions/mean_terminated_length": 856.8657470703125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 217.4,
"epoch": 1.4423076923076923,
"grad_norm": 0.0005304485675878823,
"learning_rate": 2.6442307692307696e-06,
"loss": -0.0011,
"num_tokens": 1516804210.0,
"reward": 1.3223771572113037,
"reward_std": 0.10662513226270676,
"rewards/accuracy_reward": 0.7762152671813964,
"rewards/brier_reward": 0.8709542870521545,
"rewards/confidence_one_or_zero": 0.006423611100763083,
"rewards/format_reward": 0.997569453716278,
"rewards/mean_confidence_reward": 0.7678810715675354,
"sampling/batch_mean_priority_error": 0.011822916666666653,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.28055555555555556,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0020938485860824583,
"sampling/priority_kl": 0.02999958209693432,
"sampling/priority_scale": 0.7341154158348218,
"sampling/prob_entropy": 10.278950881958007,
"sampling/prob_max": 5.1341670769033954e-05,
"sampling/prob_min": 1.9893073113053105e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.435200023651123,
"sampling/prompt_draws_total": 43056.0,
"sampling/seen_fraction": 0.8412066698074341,
"sampling/unseen_fraction": 0.15879333019256592,
"signal/accuracy_reward/centered_abs_mean": 0.10538194328546524,
"signal/accuracy_reward/group_std_mean": 0.14465567022562026,
"signal/accuracy_reward/group_zero_std_frac": 0.5666666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05269097164273262,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05269097164273262,
"signal/advantage_abs_mean": 0.07484711110591888,
"signal/advantage_pre_scale_abs_mean": 0.07484711110591888,
"signal/advantage_pre_scale_std": 0.16025152802467346,
"signal/advantage_std": 0.16025152802467346,
"signal/brier_reward/centered_abs_mean": 0.06213294193148613,
"signal/brier_reward/group_std_mean": 0.08648014217615127,
"signal/brier_reward/group_zero_std_frac": 0.2972222208976746,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.031066470965743064,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.031066470965743064,
"signal/confidence_one_or_zero/centered_abs_mean": 0.010112847248092294,
"signal/confidence_one_or_zero/group_std_mean": 0.01544865956529975,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.950000011920929,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.0112846808851827e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.0112846808851827e-07,
"signal/format_reward/centered_abs_mean": 0.00461154505610466,
"signal/format_reward/group_std_mean": 0.011821653135120869,
"signal/format_reward/group_zero_std_frac": 0.9388888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00230577252805233,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00230577252805233,
"signal/mean_confidence_reward/centered_abs_mean": 0.04797801896929741,
"signal/mean_confidence_reward/group_std_mean": 0.0646023727953434,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3277777791023254,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.797801636868826e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.797801636868826e-07,
"step": 600
},
{
"epoch": 1.4423076923076923,
"eval_calibration/aurc": 0.1346761021889891,
"eval_calibration/batch_distribution_entropy": 0.5733805972211355,
"eval_calibration/batch_entropy_100bins": 0.3126081544974916,
"eval_calibration/batch_entropy_10bins": 0.5733805972211355,
"eval_calibration/batch_entropy_50bins": 0.3679972615168838,
"eval_calibration/batch_uniqueness": 0.2611101591861721,
"eval_calibration/confidence_entropy": 0.4430562863986335,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.5864465682015638,
"eval_calibration/coverage@15%": 0.7497827975673328,
"eval_calibration/coverage@20%": 0.8583840139009556,
"eval_calibration/coverage@25%": 0.8583840139009556,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.5733805972211355,
"eval_calibration/distribution_entropy_100": 0.3126081544974916,
"eval_calibration/ece": 0.04639443961772359,
"eval_calibration/mean_confidence": 0.7647263249348393,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.0,
"eval_completions/max_length": 2566.5,
"eval_completions/max_terminated_length": 2566.5,
"eval_completions/mean_length": 882.0298665364584,
"eval_completions/mean_terminated_length": 882.0298665364584,
"eval_completions/min_length": 304.3333333333333,
"eval_completions/min_terminated_length": 304.3333333333333,
"eval_loss": 0.0,
"eval_num_tokens": 1516804210.0,
"eval_reward": 1.2891743381818135,
"eval_reward_std": 0.31100672483444214,
"eval_rewards/accuracy_reward": 0.723090281089147,
"eval_rewards/brier_reward": 0.8561111291249593,
"eval_rewards/confidence_one_or_zero": 0.005208333333333333,
"eval_rewards/format_reward": 0.9991319477558136,
"eval_rewards/mean_confidence_reward": 0.7640624543031057,
"eval_runtime": 151.9223,
"eval_samples_per_second": 6.582,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3875325520833333,
"eval_signal/accuracy_reward/group_std_mean": 0.445096492767334,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19376627604166666,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19376627604166666,
"eval_signal/advantage_abs_mean": 0.26332706212997437,
"eval_signal/advantage_pre_scale_abs_mean": 0.26332706212997437,
"eval_signal/advantage_pre_scale_std": 0.30966611703236896,
"eval_signal/advantage_std": 0.30966611703236896,
"eval_signal/brier_reward/centered_abs_mean": 0.16171767810980478,
"eval_signal/brier_reward/group_std_mean": 0.22035816063483557,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08085883905490239,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08085883905490239,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.009874132151405016,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.023483964304129284,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 0.8888888955116272,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 9.874131497629908e-08,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 9.874131497629908e-08,
"eval_signal/format_reward/centered_abs_mean": 0.0016818575871487458,
"eval_signal/format_reward/group_std_mean": 0.0049104637776811915,
"eval_signal/format_reward/group_zero_std_frac": 0.9722222288449606,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0008409287935743729,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.0008409287935743729,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.16384546707073847,
"eval_signal/mean_confidence_reward/group_std_mean": 0.19949180136124292,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.6384545915570925e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.6384545915570925e-06,
"eval_steps_per_second": 0.039,
"step": 600
},
{
"epoch": 1.4423076923076923,
"step": 600,
"train_probe_calibration/aurc": 0.10479481723911278,
"train_probe_calibration/batch_distribution_entropy": 0.5860862023827814,
"train_probe_calibration/batch_entropy_100bins": 0.3205852918546655,
"train_probe_calibration/batch_entropy_10bins": 0.5860862023827814,
"train_probe_calibration/batch_entropy_50bins": 0.3773878185447479,
"train_probe_calibration/batch_uniqueness": 0.2788052930056711,
"train_probe_calibration/confidence_entropy": 0.43982697754627376,
"train_probe_calibration/coverage@0%": 0.008695652173913044,
"train_probe_calibration/coverage@1%": 0.008695652173913044,
"train_probe_calibration/coverage@10%": 0.7052173913043478,
"train_probe_calibration/coverage@15%": 0.7939130434782609,
"train_probe_calibration/coverage@20%": 0.8486956521739131,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.008695652173913044,
"train_probe_calibration/distribution_entropy_10": 0.5860862023827814,
"train_probe_calibration/distribution_entropy_100": 0.3205852918546655,
"train_probe_calibration/ece": 0.03444637681159412,
"train_probe_calibration/mean_confidence": 0.7627420289855074,
"train_probe_calibration/unique_confidence_per_question": 0.011284722222222222,
"train_probe_calibration/unique_confidences": 13,
"train_probe_completions/clipped_ratio": 0.0017361111111111234,
"train_probe_completions/max_length": 3118.3333333333335,
"train_probe_completions/max_terminated_length": 3118.3333333333335,
"train_probe_completions/mean_length": 913.1567789713541,
"train_probe_completions/mean_terminated_length": 914.722666422526,
"train_probe_completions/min_length": 178.5,
"train_probe_completions/min_terminated_length": 271.3333333333333,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1516804210.0,
"train_probe_reward": 1.315513809521993,
"train_probe_reward_std": 0.29222285747528076,
"train_probe_rewards/accuracy_reward": 0.7578124900658926,
"train_probe_rewards/brier_reward": 0.8749358355998993,
"train_probe_rewards/confidence_one_or_zero": 0.006076389069979389,
"train_probe_rewards/format_reward": 0.9982638955116272,
"train_probe_rewards/mean_confidence_reward": 0.7614177962144216,
"train_probe_runtime": 171.292,
"train_probe_samples_per_second": 5.838,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3546549479166667,
"train_probe_signal/accuracy_reward/group_std_mean": 0.42514685293038684,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.17732747395833334,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.17732747395833334,
"train_probe_signal/advantage_abs_mean": 0.23652945458889008,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.23652945458889008,
"train_probe_signal/advantage_pre_scale_std": 0.29100040594736737,
"train_probe_signal/advantage_std": 0.29100040594736737,
"train_probe_signal/brier_reward/centered_abs_mean": 0.14432349801063538,
"train_probe_signal/brier_reward/group_std_mean": 0.2020617425441742,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07216174900531769,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07216174900531769,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.011773003110041222,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.034373246443768345,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 0.8055555820465088,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.1773002484005701e-07,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.1773002484005701e-07,
"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": 0.1672744701306025,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.20480586091677347,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.6727445692292047e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.6727445692292047e-06,
"train_probe_steps_per_second": 0.035
},
{
"calibration/aurc": 0.2025863559052315,
"calibration/batch_distribution_entropy": 0.6094061078151227,
"calibration/batch_entropy_100bins": 0.32643694694181385,
"calibration/batch_entropy_10bins": 0.6094061078151227,
"calibration/batch_entropy_50bins": 0.38427629223435283,
"calibration/batch_uniqueness": 0.35345220359531254,
"calibration/confidence_entropy": 0.4623222893283905,
"calibration/coverage@0%": 0.0020942841199528784,
"calibration/coverage@1%": 0.0020942841199528784,
"calibration/coverage@10%": 0.16511511745328622,
"calibration/coverage@15%": 0.19219845078661957,
"calibration/coverage@20%": 0.5574318453467625,
"calibration/coverage@25%": 0.8025488314663892,
"calibration/coverage@30%": 0.8855078430882883,
"calibration/coverage@5%": 0.0020942841199528784,
"calibration/distribution_entropy_10": 0.6094061078151227,
"calibration/distribution_entropy_100": 0.32643694694181385,
"calibration/ece": 0.11574754381809685,
"calibration/mean_confidence": 0.7419782801931942,
"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": 0.0019097222222222321,
"completions/max_length": 3675.2,
"completions/max_terminated_length": 3675.2,
"completions/mean_length": 899.375439453125,
"completions/mean_terminated_length": 901.0966918945312,
"completions/min_length": 0.0,
"completions/min_terminated_length": 254.6,
"epoch": 1.4543269230769231,
"grad_norm": 0.0005201512249186635,
"learning_rate": 2.6141826923076926e-06,
"loss": -0.0007,
"num_tokens": 1530249463.0,
"reward": 1.3011293888092041,
"reward_std": 0.10355799049139022,
"rewards/accuracy_reward": 0.7501736044883728,
"rewards/brier_reward": 0.8539797186851501,
"rewards/confidence_one_or_zero": 0.003211805532919243,
"rewards/format_reward": 0.9980902791023254,
"rewards/mean_confidence_reward": 0.7592641592025757,
"sampling/batch_mean_priority_error": 0.022168472222222204,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2638888888888889,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0021110890433192254,
"sampling/priority_kl": 0.029999648779630662,
"sampling/priority_scale": 0.7359204589622095,
"sampling/prob_entropy": 10.27893238067627,
"sampling/prob_max": 5.153878591954708e-05,
"sampling/prob_min": 1.9922950014006346e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.4471999883651734,
"sampling/prompt_draws_total": 43416.0,
"sampling/seen_fraction": 0.8443333506584167,
"sampling/unseen_fraction": 0.15566664934158325,
"signal/accuracy_reward/centered_abs_mean": 0.10117187350988388,
"signal/accuracy_reward/group_std_mean": 0.14249942302703858,
"signal/accuracy_reward/group_zero_std_frac": 0.5555555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05058593675494194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05058593675494194,
"signal/advantage_abs_mean": 0.07119562327861786,
"signal/advantage_pre_scale_abs_mean": 0.07119562327861786,
"signal/advantage_pre_scale_std": 0.1525435507297516,
"signal/advantage_std": 0.1525435507297516,
"signal/brier_reward/centered_abs_mean": 0.05812448561191559,
"signal/brier_reward/group_std_mean": 0.08289218097925186,
"signal/brier_reward/group_zero_std_frac": 0.2472222179174423,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029062242805957796,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029062242805957796,
"signal/confidence_one_or_zero/centered_abs_mean": 0.00451931421412155,
"signal/confidence_one_or_zero/group_std_mean": 0.007009075395762921,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.975,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.519313883122322e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.519313883122322e-08,
"signal/format_reward/centered_abs_mean": 0.003656683978624642,
"signal/format_reward/group_std_mean": 0.009607256762683392,
"signal/format_reward/group_zero_std_frac": 0.95,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.001828341989312321,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.001828341989312321,
"signal/mean_confidence_reward/centered_abs_mean": 0.05031117424368858,
"signal/mean_confidence_reward/group_std_mean": 0.06716113239526748,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2666666656732559,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.031117325415834e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.031117325415834e-07,
"step": 605
},
{
"calibration/aurc": 0.18925906471339116,
"calibration/batch_distribution_entropy": 0.5951494301130872,
"calibration/batch_entropy_100bins": 0.31397400838451495,
"calibration/batch_entropy_10bins": 0.5951494301130872,
"calibration/batch_entropy_50bins": 0.36960512261335693,
"calibration/batch_uniqueness": 0.34783256793587974,
"calibration/confidence_entropy": 0.46064776300794225,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.13490813648293962,
"calibration/coverage@10%": 0.1942257217847769,
"calibration/coverage@15%": 0.2793733681462141,
"calibration/coverage@20%": 0.608043679286336,
"calibration/coverage@25%": 0.772106179286336,
"calibration/coverage@30%": 0.8325228459530025,
"calibration/coverage@5%": 0.13490813648293962,
"calibration/distribution_entropy_10": 0.5951494301130872,
"calibration/distribution_entropy_100": 0.31397400838451495,
"calibration/ece": 0.1528422618435751,
"calibration/mean_confidence": 0.7713018239927909,
"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": 0.0021701388888888838,
"completions/max_length": 3768.2,
"completions/max_terminated_length": 3768.2,
"completions/mean_length": 932.9999145507812,
"completions/mean_terminated_length": 935.0447998046875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 247.4,
"epoch": 1.4663461538461537,
"grad_norm": 0.0006306435097940266,
"learning_rate": 2.584134615384616e-06,
"loss": -0.0024,
"num_tokens": 1544096950.0,
"reward": 1.2903532981872559,
"reward_std": 0.1076474979519844,
"rewards/accuracy_reward": 0.7316840291023254,
"rewards/brier_reward": 0.8511776804924012,
"rewards/confidence_one_or_zero": 0.0012152777577284723,
"rewards/format_reward": 0.9978298544883728,
"rewards/mean_confidence_reward": 0.7503623723983764,
"sampling/batch_mean_priority_error": 0.01682638888888887,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2722222222222222,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.002135100308805704,
"sampling/priority_kl": 0.03000096157193184,
"sampling/priority_scale": 0.7381196321221069,
"sampling/prob_entropy": 10.278960800170898,
"sampling/prob_max": 5.1753289881162345e-05,
"sampling/prob_min": 1.994912381633185e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.4591999769210815,
"sampling/prompt_draws_total": 43776.0,
"sampling/seen_fraction": 0.8476733326911926,
"sampling/unseen_fraction": 0.15232666730880737,
"signal/accuracy_reward/centered_abs_mean": 0.10917426198720932,
"signal/accuracy_reward/group_std_mean": 0.15016919672489165,
"signal/accuracy_reward/group_zero_std_frac": 0.5472222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05458713099360466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05458713099360466,
"signal/advantage_abs_mean": 0.0764501854777336,
"signal/advantage_pre_scale_abs_mean": 0.0764501854777336,
"signal/advantage_pre_scale_std": 0.15649850368499757,
"signal/advantage_std": 0.15649850368499757,
"signal/brier_reward/centered_abs_mean": 0.06043119207024574,
"signal/brier_reward/group_std_mean": 0.082360278069973,
"signal/brier_reward/group_zero_std_frac": 0.25,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03021559603512287,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03021559603512287,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0015082465368323028,
"signal/confidence_one_or_zero/group_std_mean": 0.0018771322444081306,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.5082464699389676e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.5082464699389676e-08,
"signal/format_reward/centered_abs_mean": 0.004139539937023073,
"signal/format_reward/group_std_mean": 0.010728820972144604,
"signal/format_reward/group_zero_std_frac": 0.944444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0020697699685115365,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0020697699685115365,
"signal/mean_confidence_reward/centered_abs_mean": 0.0506988450884819,
"signal/mean_confidence_reward/group_std_mean": 0.06803653538227081,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2694444477558136,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.069884593922325e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.069884593922325e-07,
"step": 610
},
{
"calibration/aurc": 0.11789342149139752,
"calibration/batch_distribution_entropy": 0.6047390118271716,
"calibration/batch_entropy_100bins": 0.3246738368739022,
"calibration/batch_entropy_10bins": 0.6047390118271716,
"calibration/batch_entropy_50bins": 0.3822007875892891,
"calibration/batch_uniqueness": 0.38138563368055556,
"calibration/confidence_entropy": 0.4734669462340119,
"calibration/coverage@0%": 0.06666666666666667,
"calibration/coverage@1%": 0.19583333333333336,
"calibration/coverage@10%": 0.6197916666666666,
"calibration/coverage@15%": 0.6838541666666667,
"calibration/coverage@20%": 0.7208333333333333,
"calibration/coverage@25%": 0.8546875,
"calibration/coverage@30%": 0.8953125,
"calibration/coverage@5%": 0.3583333333333334,
"calibration/distribution_entropy_10": 0.6047390118271716,
"calibration/distribution_entropy_100": 0.3246738368739022,
"calibration/ece": 0.11421875,
"calibration/mean_confidence": 0.7418229166666668,
"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": 0.0014756944444444641,
"completions/max_length": 3789.0,
"completions/max_terminated_length": 3789.0,
"completions/mean_length": 950.9626831054687,
"completions/mean_terminated_length": 952.3809326171875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 252.2,
"epoch": 1.4783653846153846,
"grad_norm": 0.0005810891161672771,
"learning_rate": 2.554086538461539e-06,
"loss": -0.0,
"num_tokens": 1558176616.0,
"reward": 1.3133844137191772,
"reward_std": 0.10787461549043656,
"rewards/accuracy_reward": 0.7603298664093018,
"rewards/brier_reward": 0.8678995728492737,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.998524296283722,
"rewards/mean_confidence_reward": 0.7534403920173645,
"sampling/batch_mean_priority_error": 0.01978472222222221,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.23333333333333334,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0021563940681517124,
"sampling/priority_kl": 0.029999713599681854,
"sampling/priority_scale": 0.7398386180168017,
"sampling/prob_entropy": 10.278945922851562,
"sampling/prob_max": 5.1947306928923354e-05,
"sampling/prob_min": 1.9979511489509604e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.4711999893188477,
"sampling/prompt_draws_total": 44136.0,
"sampling/seen_fraction": 0.8506266713142395,
"sampling/unseen_fraction": 0.1493733286857605,
"signal/accuracy_reward/centered_abs_mean": 0.11985134780406952,
"signal/accuracy_reward/group_std_mean": 0.15811396539211273,
"signal/accuracy_reward/group_zero_std_frac": 0.55,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05992567390203476,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05992567390203476,
"signal/advantage_abs_mean": 0.08039970993995667,
"signal/advantage_pre_scale_abs_mean": 0.08039970993995667,
"signal/advantage_pre_scale_std": 0.16128072142601013,
"signal/advantage_std": 0.16128072142601013,
"signal/brier_reward/centered_abs_mean": 0.058254283666610715,
"signal/brier_reward/group_std_mean": 0.07793922573328019,
"signal/brier_reward/group_zero_std_frac": 0.2638888955116272,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029127141833305357,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029127141833305357,
"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": 0.002533637161832303,
"signal/format_reward/group_std_mean": 0.005043594865128398,
"signal/format_reward/group_zero_std_frac": 0.9777777671813965,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0012668185809161514,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0012668185809161514,
"signal/mean_confidence_reward/centered_abs_mean": 0.04812428429722786,
"signal/mean_confidence_reward/group_std_mean": 0.062294195592403415,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2694444447755814,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.812428301192995e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.812428301192995e-07,
"step": 615
},
{
"calibration/aurc": 0.1058172588479028,
"calibration/batch_distribution_entropy": 0.6578273898301509,
"calibration/batch_entropy_100bins": 0.35653299906657165,
"calibration/batch_entropy_10bins": 0.6578273898301509,
"calibration/batch_entropy_50bins": 0.41970487784557414,
"calibration/batch_uniqueness": 0.5193359375,
"calibration/confidence_entropy": 0.4997947054499211,
"calibration/coverage@0%": 0.065625,
"calibration/coverage@1%": 0.1296875,
"calibration/coverage@10%": 0.5354166666666667,
"calibration/coverage@15%": 0.6609375,
"calibration/coverage@20%": 0.8901041666666668,
"calibration/coverage@25%": 0.9182291666666668,
"calibration/coverage@30%": 0.9848958333333332,
"calibration/coverage@5%": 0.3864583333333333,
"calibration/distribution_entropy_10": 0.6578273898301509,
"calibration/distribution_entropy_100": 0.35653299906657165,
"calibration/ece": 0.11013020833333327,
"calibration/mean_confidence": 0.7221614583333333,
"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": 0.0021701388888888617,
"completions/max_length": 3434.0,
"completions/max_terminated_length": 3434.0,
"completions/mean_length": 941.9648681640625,
"completions/mean_terminated_length": 944.02392578125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 269.6,
"epoch": 1.4903846153846154,
"grad_norm": 0.00043419600115157664,
"learning_rate": 2.5240384615384618e-06,
"loss": -0.0009,
"num_tokens": 1572119859.0,
"reward": 1.323990035057068,
"reward_std": 0.08798304051160813,
"rewards/accuracy_reward": 0.7714409828186035,
"rewards/brier_reward": 0.8786945819854737,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9978298544883728,
"rewards/mean_confidence_reward": 0.7352184534072876,
"sampling/batch_mean_priority_error": 0.013222222222222213,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.19166666666666668,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0021701503079384564,
"sampling/priority_kl": 0.029999665170907974,
"sampling/priority_scale": 0.7408775746589527,
"sampling/prob_entropy": 10.278956604003906,
"sampling/prob_max": 5.211559619056061e-05,
"sampling/prob_min": 2.0017361021018586e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.4832000017166138,
"sampling/prompt_draws_total": 44496.0,
"sampling/seen_fraction": 0.8530933260917664,
"sampling/unseen_fraction": 0.14690667390823364,
"signal/accuracy_reward/centered_abs_mean": 0.08047417402267457,
"signal/accuracy_reward/group_std_mean": 0.11685203611850739,
"signal/accuracy_reward/group_zero_std_frac": 0.6194444537162781,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04023708701133728,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04023708701133728,
"signal/advantage_abs_mean": 0.05966672003269195,
"signal/advantage_pre_scale_abs_mean": 0.05966672003269195,
"signal/advantage_pre_scale_std": 0.1359117567539215,
"signal/advantage_std": 0.1359117567539215,
"signal/brier_reward/centered_abs_mean": 0.0505802758038044,
"signal/brier_reward/group_std_mean": 0.07184310257434845,
"signal/brier_reward/group_zero_std_frac": 0.1722222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0252901379019022,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0252901379019022,
"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": 3.2552080142522754e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.2552080142522754e-09,
"signal/format_reward/centered_abs_mean": 0.0040961371967568995,
"signal/format_reward/group_std_mean": 0.009779365314170719,
"signal/format_reward/group_zero_std_frac": 0.9527777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0020480685983784498,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0020480685983784498,
"signal/mean_confidence_reward/centered_abs_mean": 0.049345605075359344,
"signal/mean_confidence_reward/group_std_mean": 0.06616655513644218,
"signal/mean_confidence_reward/group_zero_std_frac": 0.18333333134651184,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.93456042249818e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.93456042249818e-07,
"step": 620
},
{
"calibration/aurc": 0.15336518701159388,
"calibration/batch_distribution_entropy": 0.6800076074158324,
"calibration/batch_entropy_100bins": 0.3622864830464606,
"calibration/batch_entropy_10bins": 0.6800076074158324,
"calibration/batch_entropy_50bins": 0.4264777860961084,
"calibration/batch_uniqueness": 0.5138020531346246,
"calibration/confidence_entropy": 0.5045679523521033,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.06631853785900783,
"calibration/coverage@10%": 0.3418826731392584,
"calibration/coverage@15%": 0.5271101989576695,
"calibration/coverage@20%": 0.5976153626398852,
"calibration/coverage@25%": 0.8201260253695442,
"calibration/coverage@30%": 0.9593709203655353,
"calibration/coverage@5%": 0.17774225319175183,
"calibration/distribution_entropy_10": 0.6800076074158324,
"calibration/distribution_entropy_100": 0.3622864830464606,
"calibration/ece": 0.09629617928490819,
"calibration/mean_confidence": 0.6966044761549471,
"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": 0.002343749999999978,
"completions/max_length": 3773.8,
"completions/max_terminated_length": 3773.8,
"completions/mean_length": 1026.7007934570313,
"completions/mean_terminated_length": 1029.1406860351562,
"completions/min_length": 0.0,
"completions/min_terminated_length": 326.8,
"epoch": 1.5024038461538463,
"grad_norm": 0.00045041058911010623,
"learning_rate": 2.4939903846153847e-06,
"loss": -0.0015,
"num_tokens": 1587079484.0,
"reward": 1.3066246509552002,
"reward_std": 0.0912088081240654,
"rewards/accuracy_reward": 0.7471354246139527,
"rewards/brier_reward": 0.8684430360794068,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.997656238079071,
"rewards/mean_confidence_reward": 0.728438937664032,
"sampling/batch_mean_priority_error": 0.018874999999999982,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.24722222222222223,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.0021893881261348725,
"sampling/priority_kl": 0.030000680685043336,
"sampling/priority_scale": 0.7427042663795873,
"sampling/prob_entropy": 10.278961181640625,
"sampling/prob_max": 5.23165290360339e-05,
"sampling/prob_min": 2.0047180805704555e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.495199990272522,
"sampling/prompt_draws_total": 44856.0,
"sampling/seen_fraction": 0.8559533357620239,
"sampling/unseen_fraction": 0.14404666423797607,
"signal/accuracy_reward/centered_abs_mean": 0.1004937082529068,
"signal/accuracy_reward/group_std_mean": 0.1350753501057625,
"signal/accuracy_reward/group_zero_std_frac": 0.6055555701255798,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0502468541264534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0502468541264534,
"signal/advantage_abs_mean": 0.06540203243494033,
"signal/advantage_pre_scale_abs_mean": 0.06540203243494033,
"signal/advantage_pre_scale_std": 0.13887231945991516,
"signal/advantage_std": 0.13887231945991516,
"signal/brier_reward/centered_abs_mean": 0.05391479507088661,
"signal/brier_reward/group_std_mean": 0.07308482676744461,
"signal/brier_reward/group_zero_std_frac": 0.19722222685813903,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.026957397535443305,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.026957397535443305,
"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": 0.004258897609543055,
"signal/format_reward/group_std_mean": 0.009096213802695274,
"signal/format_reward/group_zero_std_frac": 0.9583333492279053,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0021294488047715276,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0021294488047715276,
"signal/mean_confidence_reward/centered_abs_mean": 0.0523322694003582,
"signal/mean_confidence_reward/group_std_mean": 0.06916438788175583,
"signal/mean_confidence_reward/group_zero_std_frac": 0.19722222685813903,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.233226659129286e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.233226659129286e-07,
"step": 625
},
{
"calibration/aurc": 0.12000735705395464,
"calibration/batch_distribution_entropy": 0.6255764616425743,
"calibration/batch_entropy_100bins": 0.3354601150908449,
"calibration/batch_entropy_10bins": 0.6255764616425743,
"calibration/batch_entropy_50bins": 0.3948982197857547,
"calibration/batch_uniqueness": 0.4280289172596209,
"calibration/confidence_entropy": 0.47212258337113344,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.21232513398378453,
"calibration/coverage@10%": 0.38557459862122667,
"calibration/coverage@15%": 0.48163925438596494,
"calibration/coverage@20%": 0.8848291993037425,
"calibration/coverage@25%": 0.9450713148733453,
"calibration/coverage@30%": 0.9932291666666668,
"calibration/coverage@5%": 0.3062012304750126,
"calibration/distribution_entropy_10": 0.6255764616425743,
"calibration/distribution_entropy_100": 0.3354601150908449,
"calibration/ece": 0.10917860213998729,
"calibration/mean_confidence": 0.7486235025901614,
"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": 0.002777777777777768,
"completions/max_length": 3631.8,
"completions/max_terminated_length": 3631.8,
"completions/mean_length": 1021.059814453125,
"completions/mean_terminated_length": 1023.94638671875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 284.0,
"epoch": 1.5144230769230769,
"grad_norm": 0.0005333871231414378,
"learning_rate": 2.463942307692308e-06,
"loss": -0.0021,
"num_tokens": 1601977741.0,
"reward": 1.3167559146881103,
"reward_std": 0.0949985533952713,
"rewards/accuracy_reward": 0.7634548664093017,
"rewards/brier_reward": 0.8730803966522217,
"rewards/confidence_one_or_zero": 0.0004340277810115367,
"rewards/format_reward": 0.9969618082046509,
"rewards/mean_confidence_reward": 0.7371064901351929,
"sampling/batch_mean_priority_error": 0.014197530864197522,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.22777777777777777,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.002204936416819692,
"sampling/priority_kl": 0.02999979294836521,
"sampling/priority_scale": 0.7445269406074658,
"sampling/prob_entropy": 10.278949737548828,
"sampling/prob_max": 5.25149138411507e-05,
"sampling/prob_min": 2.0075886277481912e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.507200002670288,
"sampling/prompt_draws_total": 45216.0,
"sampling/seen_fraction": 0.8587599992752075,
"sampling/unseen_fraction": 0.14124000072479248,
"signal/accuracy_reward/centered_abs_mean": 0.09511176496744156,
"signal/accuracy_reward/group_std_mean": 0.12550814002752303,
"signal/accuracy_reward/group_zero_std_frac": 0.6333333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04755588248372078,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04755588248372078,
"signal/advantage_abs_mean": 0.06778244376182556,
"signal/advantage_pre_scale_abs_mean": 0.06778244376182556,
"signal/advantage_pre_scale_std": 0.14562630653381348,
"signal/advantage_std": 0.14562630653381348,
"signal/brier_reward/centered_abs_mean": 0.055177373439073564,
"signal/brier_reward/group_std_mean": 0.07687650173902512,
"signal/brier_reward/group_zero_std_frac": 0.2,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.027588686719536782,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.027588686719536782,
"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": 0.005821397621184587,
"signal/format_reward/group_std_mean": 0.015639285184443,
"signal/format_reward/group_zero_std_frac": 0.9166666746139527,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0029106988105922936,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0029106988105922936,
"signal/mean_confidence_reward/centered_abs_mean": 0.053772156685590745,
"signal/mean_confidence_reward/group_std_mean": 0.0706539012491703,
"signal/mean_confidence_reward/group_zero_std_frac": 0.20555555820465088,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.377215302360127e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.377215302360127e-07,
"step": 630
},
{
"calibration/aurc": 0.06758576087491405,
"calibration/batch_distribution_entropy": 0.6739458670298898,
"calibration/batch_entropy_100bins": 0.36148430969206863,
"calibration/batch_entropy_10bins": 0.6739458670298898,
"calibration/batch_entropy_50bins": 0.42553348060237434,
"calibration/batch_uniqueness": 0.507114894490273,
"calibration/confidence_entropy": 0.4857174340818884,
"calibration/coverage@0%": 0.12276725871313672,
"calibration/coverage@1%": 0.2607880920464701,
"calibration/coverage@10%": 0.7569473367679553,
"calibration/coverage@15%": 0.8394321061326202,
"calibration/coverage@20%": 0.9129786771105308,
"calibration/coverage@25%": 0.9682291666666668,
"calibration/coverage@30%": 0.9682291666666668,
"calibration/coverage@5%": 0.5899399641137999,
"calibration/distribution_entropy_10": 0.6739458670298898,
"calibration/distribution_entropy_100": 0.36148430969206863,
"calibration/ece": 0.14519666110659998,
"calibration/mean_confidence": 0.7185335590463714,
"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": 0.003906249999999978,
"completions/max_length": 3880.8,
"completions/max_terminated_length": 3880.8,
"completions/mean_length": 1042.0560668945313,
"completions/mean_terminated_length": 1046.1614990234375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 307.2,
"epoch": 1.5264423076923077,
"grad_norm": 0.0005072118365205824,
"learning_rate": 2.433894230769231e-06,
"loss": -0.0038,
"num_tokens": 1617074483.0,
"reward": 1.3064142227172852,
"reward_std": 0.0989925280213356,
"rewards/accuracy_reward": 0.7481770873069763,
"rewards/brier_reward": 0.868543004989624,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9960937619209289,
"rewards/mean_confidence_reward": 0.729508101940155,
"sampling/batch_mean_priority_error": 0.016961419753086404,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2416666666666667,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.002223148616030812,
"sampling/priority_kl": 0.030000782012939452,
"sampling/priority_scale": 0.7464619219535962,
"sampling/prob_entropy": 10.2789737701416,
"sampling/prob_max": 5.2718666120199485e-05,
"sampling/prob_min": 2.0103476708754896e-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": 0.09907226413488388,
"signal/accuracy_reward/group_std_mean": 0.1344002977013588,
"signal/accuracy_reward/group_zero_std_frac": 0.597222238779068,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04953613206744194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04953613206744194,
"signal/advantage_abs_mean": 0.06925319880247116,
"signal/advantage_pre_scale_abs_mean": 0.06925319880247116,
"signal/advantage_pre_scale_std": 0.14847098886966706,
"signal/advantage_std": 0.14847098886966706,
"signal/brier_reward/centered_abs_mean": 0.05832105129957199,
"signal/brier_reward/group_std_mean": 0.08077382892370225,
"signal/brier_reward/group_zero_std_frac": 0.1972222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029160525649785995,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029160525649785995,
"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": 0.00697157122194767,
"signal/format_reward/group_std_mean": 0.01599796488881111,
"signal/format_reward/group_zero_std_frac": 0.9222222447395325,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003485785610973835,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003485785610973835,
"signal/mean_confidence_reward/centered_abs_mean": 0.05409794896841049,
"signal/mean_confidence_reward/group_std_mean": 0.07281152158975601,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.409794880506525e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.409794880506525e-07,
"step": 635
},
{
"calibration/aurc": 0.1493596396212617,
"calibration/batch_distribution_entropy": 0.6154120129307172,
"calibration/batch_entropy_100bins": 0.327493360904968,
"calibration/batch_entropy_10bins": 0.6154120129307172,
"calibration/batch_entropy_50bins": 0.3855198856591432,
"calibration/batch_uniqueness": 0.38064147099576423,
"calibration/confidence_entropy": 0.4642811581171026,
"calibration/coverage@0%": 0.0005208333333333333,
"calibration/coverage@1%": 0.0005208333333333333,
"calibration/coverage@10%": 0.39853397241148086,
"calibration/coverage@15%": 0.4210599270674741,
"calibration/coverage@20%": 0.6583795691906005,
"calibration/coverage@25%": 0.7992833442123586,
"calibration/coverage@30%": 0.9686684073107049,
"calibration/coverage@5%": 0.23565933853804272,
"calibration/distribution_entropy_10": 0.6154120129307172,
"calibration/distribution_entropy_100": 0.327493360904968,
"calibration/ece": 0.11085883086933453,
"calibration/mean_confidence": 0.7440802817524015,
"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": 0.004166666666666674,
"completions/max_length": 3869.0,
"completions/max_terminated_length": 3869.0,
"completions/mean_length": 1044.1888916015625,
"completions/mean_terminated_length": 1048.6419189453125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 344.0,
"epoch": 1.5384615384615383,
"grad_norm": 0.0005268683307804167,
"learning_rate": 2.403846153846154e-06,
"loss": -0.0053,
"num_tokens": 1632200083.0,
"reward": 1.2998849391937255,
"reward_std": 0.10648185312747956,
"rewards/accuracy_reward": 0.7403645873069763,
"rewards/brier_reward": 0.863817822933197,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9955729126930237,
"rewards/mean_confidence_reward": 0.7254003286361694,
"sampling/batch_mean_priority_error": 0.015239513888888875,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.23055555555555554,
"sampling/error_ema_max": 0.1524374932050705,
"sampling/error_ema_mean": 0.002240751124918461,
"sampling/priority_kl": 0.030000269040465356,
"sampling/priority_scale": 0.7487170279258862,
"sampling/prob_entropy": 10.27895679473877,
"sampling/prob_max": 5.293711947160773e-05,
"sampling/prob_min": 2.012807090068236e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.5312000036239624,
"sampling/prompt_draws_total": 45936.0,
"sampling/seen_fraction": 0.8645133256912232,
"sampling/unseen_fraction": 0.13548667430877687,
"signal/accuracy_reward/centered_abs_mean": 0.1005154088139534,
"signal/accuracy_reward/group_std_mean": 0.13976921439170836,
"signal/accuracy_reward/group_zero_std_frac": 0.5777777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0502577044069767,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0502577044069767,
"signal/advantage_abs_mean": 0.07303778156638145,
"signal/advantage_pre_scale_abs_mean": 0.07303778156638145,
"signal/advantage_pre_scale_std": 0.15646542310714723,
"signal/advantage_std": 0.15646542310714723,
"signal/brier_reward/centered_abs_mean": 0.06378994956612587,
"signal/brier_reward/group_std_mean": 0.08879801481962205,
"signal/brier_reward/group_zero_std_frac": 0.2194444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.031894974783062933,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.031894974783062933,
"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": 0.00820855046622455,
"signal/format_reward/group_std_mean": 0.01877419650554657,
"signal/format_reward/group_zero_std_frac": 0.9111111283302307,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004104275233112275,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004104275233112275,
"signal/mean_confidence_reward/centered_abs_mean": 0.05598941519856453,
"signal/mean_confidence_reward/group_std_mean": 0.07363492101430893,
"signal/mean_confidence_reward/group_zero_std_frac": 0.23333333432674408,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.598941356765863e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.598941356765863e-07,
"step": 640
},
{
"calibration/aurc": 0.10121633485801577,
"calibration/batch_distribution_entropy": 0.647102852218405,
"calibration/batch_entropy_100bins": 0.3475252170040523,
"calibration/batch_entropy_10bins": 0.647102852218405,
"calibration/batch_entropy_50bins": 0.40910106254626905,
"calibration/batch_uniqueness": 0.4184239848201855,
"calibration/confidence_entropy": 0.47056694558071294,
"calibration/coverage@0%": 0.12291666666666667,
"calibration/coverage@1%": 0.2486727589208007,
"calibration/coverage@10%": 0.615986727589208,
"calibration/coverage@15%": 0.6795583115752828,
"calibration/coverage@20%": 0.8546154264577893,
"calibration/coverage@25%": 0.8957612597911228,
"calibration/coverage@30%": 0.9583333333333333,
"calibration/coverage@5%": 0.5100195822454309,
"calibration/distribution_entropy_10": 0.647102852218405,
"calibration/distribution_entropy_100": 0.3475252170040523,
"calibration/ece": 0.11775878481288057,
"calibration/mean_confidence": 0.7193620811575283,
"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": 0.005642361111111116,
"completions/max_length": 3893.4,
"completions/max_terminated_length": 3893.4,
"completions/mean_length": 1168.6723388671876,
"completions/mean_terminated_length": 1175.3999267578124,
"completions/min_length": 0.0,
"completions/min_terminated_length": 342.4,
"epoch": 1.5504807692307692,
"grad_norm": 0.0004638670652639121,
"learning_rate": 2.373798076923077e-06,
"loss": -0.0056,
"num_tokens": 1648768276.0,
"reward": 1.2972732782363892,
"reward_std": 0.10741431713104248,
"rewards/accuracy_reward": 0.7346354246139526,
"rewards/brier_reward": 0.8657999157905578,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9940972328186035,
"rewards/mean_confidence_reward": 0.703185749053955,
"sampling/batch_mean_priority_error": 0.01578472222222221,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.21944444444444441,
"sampling/error_ema_max": 0.15972749292850494,
"sampling/error_ema_mean": 0.002255667420104146,
"sampling/priority_kl": 0.02999972328543663,
"sampling/priority_scale": 0.7503741204505786,
"sampling/prob_entropy": 10.278932762145995,
"sampling/prob_max": 5.313203946570866e-05,
"sampling/prob_min": 2.015899335674476e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.5432000160217285,
"sampling/prompt_draws_total": 46296.0,
"sampling/seen_fraction": 0.8670133352279663,
"sampling/unseen_fraction": 0.1329866647720337,
"signal/accuracy_reward/centered_abs_mean": 0.10655924528837205,
"signal/accuracy_reward/group_std_mean": 0.14632892608642578,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111164093018,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05327962264418602,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05327962264418602,
"signal/advantage_abs_mean": 0.07435599118471145,
"signal/advantage_pre_scale_abs_mean": 0.07435599118471145,
"signal/advantage_pre_scale_std": 0.156350177526474,
"signal/advantage_std": 0.156350177526474,
"signal/brier_reward/centered_abs_mean": 0.06651555225253106,
"signal/brier_reward/group_std_mean": 0.09179123044013977,
"signal/brier_reward/group_zero_std_frac": 0.21111110746860504,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03325777612626553,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03325777612626553,
"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": 0.010601128358393908,
"signal/format_reward/group_std_mean": 0.022472953796386717,
"signal/format_reward/group_zero_std_frac": 0.8972222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005300564179196954,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005300564179196954,
"signal/mean_confidence_reward/centered_abs_mean": 0.062262913584709166,
"signal/mean_confidence_reward/group_std_mean": 0.08256838023662567,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2194444417953491,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.226291361599578e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.226291361599578e-07,
"step": 645
},
{
"calibration/aurc": 0.13022814826365287,
"calibration/batch_distribution_entropy": 0.6004951907444687,
"calibration/batch_entropy_100bins": 0.3225284149216801,
"calibration/batch_entropy_10bins": 0.6004951907444687,
"calibration/batch_entropy_50bins": 0.3796752315797697,
"calibration/batch_uniqueness": 0.33640241705561535,
"calibration/confidence_entropy": 0.4576013413890502,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.16230366492146597,
"calibration/coverage@10%": 0.4286458333333334,
"calibration/coverage@15%": 0.5036458333333333,
"calibration/coverage@20%": 0.7052083333333334,
"calibration/coverage@25%": 0.9473113001745201,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.40098985602094245,
"calibration/distribution_entropy_10": 0.6004951907444687,
"calibration/distribution_entropy_100": 0.3225284149216801,
"calibration/ece": 0.1170646542321117,
"calibration/mean_confidence": 0.754144169938918,
"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": 0.006249999999999978,
"completions/max_length": 3870.4,
"completions/max_terminated_length": 3870.4,
"completions/mean_length": 1101.4054077148437,
"completions/mean_terminated_length": 1108.5246948242188,
"completions/min_length": 0.0,
"completions/min_terminated_length": 342.4,
"epoch": 1.5625,
"grad_norm": 0.0004739153664559126,
"learning_rate": 2.3437500000000002e-06,
"loss": -0.0062,
"num_tokens": 1664519922.0,
"reward": 1.3281028270721436,
"reward_std": 0.09923821091651916,
"rewards/accuracy_reward": 0.7807291626930237,
"rewards/brier_reward": 0.8817115902900696,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.99375,
"rewards/mean_confidence_reward": 0.7447699666023254,
"sampling/batch_mean_priority_error": 0.015923611111111097,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.20277777777777778,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0022721522953361275,
"sampling/priority_kl": 0.029999472200870514,
"sampling/priority_scale": 0.7521035611396656,
"sampling/prob_entropy": 10.278953170776367,
"sampling/prob_max": 5.332994842319749e-05,
"sampling/prob_min": 2.0186370602459646e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.5552000045776366,
"sampling/prompt_draws_total": 46656.0,
"sampling/seen_fraction": 0.8695200085639954,
"sampling/unseen_fraction": 0.13047999143600464,
"signal/accuracy_reward/centered_abs_mean": 0.09236111044883728,
"signal/accuracy_reward/group_std_mean": 0.12668517976999283,
"signal/accuracy_reward/group_zero_std_frac": 0.6166666626930237,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04618055522441864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04618055522441864,
"signal/advantage_abs_mean": 0.0662669561803341,
"signal/advantage_pre_scale_abs_mean": 0.0662669561803341,
"signal/advantage_pre_scale_std": 0.15250126719474794,
"signal/advantage_std": 0.15250126719474794,
"signal/brier_reward/centered_abs_mean": 0.061259324103593825,
"signal/brier_reward/group_std_mean": 0.08748518377542495,
"signal/brier_reward/group_zero_std_frac": 0.25833334028720856,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.030629662051796912,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.030629662051796912,
"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": 0.011512586940079927,
"signal/format_reward/group_std_mean": 0.025058790668845178,
"signal/format_reward/group_zero_std_frac": 0.8861111164093017,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0057562934700399635,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0057562934700399635,
"signal/mean_confidence_reward/centered_abs_mean": 0.05550645664334297,
"signal/mean_confidence_reward/group_std_mean": 0.07544975653290749,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2694444447755814,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.550645596485992e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.550645596485992e-07,
"step": 650
},
{
"epoch": 1.5625,
"eval_calibration/aurc": 0.1271157797067916,
"eval_calibration/batch_distribution_entropy": 0.6324314643752337,
"eval_calibration/batch_entropy_100bins": 0.3458577555472716,
"eval_calibration/batch_entropy_10bins": 0.6324314643752337,
"eval_calibration/batch_entropy_50bins": 0.4071381538986471,
"eval_calibration/batch_uniqueness": 0.38409579930558946,
"eval_calibration/confidence_entropy": 0.45110729568686725,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.5148601398601399,
"eval_calibration/coverage@15%": 0.7508741258741258,
"eval_calibration/coverage@20%": 0.791083916083916,
"eval_calibration/coverage@25%": 0.9475524475524476,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.6324314643752337,
"eval_calibration/distribution_entropy_100": 0.3458577555472716,
"eval_calibration/ece": 0.026005244755244586,
"eval_calibration/mean_confidence": 0.7260052447552449,
"eval_calibration/unique_confidence_per_question": 0.008680555555555556,
"eval_calibration/unique_confidences": 10,
"eval_completions/clipped_ratio": 0.006076388888888895,
"eval_completions/max_length": 3015.1666666666665,
"eval_completions/max_terminated_length": 3015.1666666666665,
"eval_completions/mean_length": 1108.1156412760417,
"eval_completions/mean_terminated_length": 1114.9664103190105,
"eval_completions/min_length": 92.0,
"eval_completions/min_terminated_length": 413.1666666666667,
"eval_loss": 0.0,
"eval_num_tokens": 1664519922.0,
"eval_reward": 1.2852014501889546,
"eval_reward_std": 0.30990825096766156,
"eval_rewards/accuracy_reward": 0.723090281089147,
"eval_rewards/brier_reward": 0.8542426228523254,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9930555621782938,
"eval_rewards/mean_confidence_reward": 0.7209634979565939,
"eval_runtime": 211.902,
"eval_samples_per_second": 4.719,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3867730001608531,
"eval_signal/accuracy_reward/group_std_mean": 0.44401001930236816,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19338650008042654,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19338650008042654,
"eval_signal/advantage_abs_mean": 0.25779375185569126,
"eval_signal/advantage_pre_scale_abs_mean": 0.25779375185569126,
"eval_signal/advantage_pre_scale_std": 0.30922891199588776,
"eval_signal/advantage_std": 0.30922891199588776,
"eval_signal/brier_reward/centered_abs_mean": 0.16802789767583212,
"eval_signal/brier_reward/group_std_mean": 0.2282408873240153,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08401394883791606,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08401394883791606,
"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": 0.013454860697189966,
"eval_signal/format_reward/group_std_mean": 0.03928371022144953,
"eval_signal/format_reward/group_zero_std_frac": 0.7777778108914694,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.006727430348594983,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.006727430348594983,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.20234372218449911,
"eval_signal/mean_confidence_reward/group_std_mean": 0.24267150461673737,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.0234372755112418e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.0234372755112418e-06,
"eval_steps_per_second": 0.028,
"step": 650
},
{
"epoch": 1.5625,
"step": 650,
"train_probe_calibration/aurc": 0.09629671541450747,
"train_probe_calibration/batch_distribution_entropy": 0.6276142254740842,
"train_probe_calibration/batch_entropy_100bins": 0.34791016850026435,
"train_probe_calibration/batch_entropy_10bins": 0.6276142254740842,
"train_probe_calibration/batch_entropy_50bins": 0.4095542212191468,
"train_probe_calibration/batch_uniqueness": 0.36119736906450195,
"train_probe_calibration/confidence_entropy": 0.4449701838043957,
"train_probe_calibration/coverage@0%": 0.004370629370629371,
"train_probe_calibration/coverage@1%": 0.004370629370629371,
"train_probe_calibration/coverage@10%": 0.6704545454545454,
"train_probe_calibration/coverage@15%": 0.8041958041958042,
"train_probe_calibration/coverage@20%": 0.8767482517482518,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.004370629370629371,
"train_probe_calibration/distribution_entropy_10": 0.6276142254740842,
"train_probe_calibration/distribution_entropy_100": 0.34791016850026435,
"train_probe_calibration/ece": 0.052840909090908994,
"train_probe_calibration/mean_confidence": 0.7326486013986014,
"train_probe_calibration/unique_confidence_per_question": 0.010416666666666666,
"train_probe_calibration/unique_confidences": 12,
"train_probe_completions/clipped_ratio": 0.006944444444444438,
"train_probe_completions/max_length": 3299.8333333333335,
"train_probe_completions/max_terminated_length": 3299.8333333333335,
"train_probe_completions/mean_length": 1149.763203938802,
"train_probe_completions/mean_terminated_length": 1157.6914672851562,
"train_probe_completions/min_length": 59.0,
"train_probe_completions/min_terminated_length": 355.5,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1664519922.0,
"train_probe_reward": 1.307287057240804,
"train_probe_reward_std": 0.29700932403405506,
"train_probe_rewards/accuracy_reward": 0.7491319278875986,
"train_probe_rewards/brier_reward": 0.8723719517389933,
"train_probe_rewards/confidence_one_or_zero": 0.0026041667442768812,
"train_probe_rewards/format_reward": 0.9930555522441864,
"train_probe_rewards/mean_confidence_reward": 0.7275607585906982,
"train_probe_runtime": 209.4521,
"train_probe_samples_per_second": 4.774,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3669162342945735,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4330385575691859,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.18345811714728674,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.18345811714728674,
"train_probe_signal/advantage_abs_mean": 0.24020937830209732,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.24020937830209732,
"train_probe_signal/advantage_pre_scale_std": 0.29617064197858173,
"train_probe_signal/advantage_std": 0.29617064197858173,
"train_probe_signal/brier_reward/centered_abs_mean": 0.14779583364725113,
"train_probe_signal/brier_reward/group_std_mean": 0.20587429155906042,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07389791682362556,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07389791682362556,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0050455727614462376,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.014731391333043575,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 0.9166666766007742,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.0455724931453005e-08,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.0455724931453005e-08,
"train_probe_signal/format_reward/centered_abs_mean": 0.013454860697189966,
"train_probe_signal/format_reward/group_std_mean": 0.03928371022144953,
"train_probe_signal/format_reward/group_zero_std_frac": 0.7777778009573618,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.006727430348594983,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.006727430348594983,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.19929740081230798,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.24239403754472733,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.992973921005614e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.992973921005614e-06,
"train_probe_steps_per_second": 0.029
},
{
"calibration/aurc": 0.10075111118953009,
"calibration/batch_distribution_entropy": 0.5685699983151254,
"calibration/batch_entropy_100bins": 0.31157743001517346,
"calibration/batch_entropy_10bins": 0.5685699983151254,
"calibration/batch_entropy_50bins": 0.3667839093332813,
"calibration/batch_uniqueness": 0.2724217686970425,
"calibration/confidence_entropy": 0.4463080918393615,
"calibration/coverage@0%": 0.11823734729493893,
"calibration/coverage@1%": 0.11823734729493893,
"calibration/coverage@10%": 0.5741792102966841,
"calibration/coverage@15%": 0.8676347076788831,
"calibration/coverage@20%": 0.9130890052356021,
"calibration/coverage@25%": 0.9455497382198953,
"calibration/coverage@30%": 0.9748691099476441,
"calibration/coverage@5%": 0.2265570462478185,
"calibration/distribution_entropy_10": 0.5685699983151254,
"calibration/distribution_entropy_100": 0.31157743001517346,
"calibration/ece": 0.10682209860383933,
"calibration/mean_confidence": 0.7507646160558464,
"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": 0.0052951388888889065,
"completions/max_length": 4049.8,
"completions/max_terminated_length": 4049.8,
"completions/mean_length": 1132.967041015625,
"completions/mean_terminated_length": 1139.044775390625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 344.8,
"epoch": 1.5745192307692308,
"grad_norm": 0.0005549096968024969,
"learning_rate": 2.3137019230769236e-06,
"loss": -0.006,
"num_tokens": 1680651318.0,
"reward": 1.3247023105621338,
"reward_std": 0.11366159170866012,
"rewards/accuracy_reward": 0.7790798664093017,
"rewards/brier_reward": 0.875691819190979,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9946180582046509,
"rewards/mean_confidence_reward": 0.7438610911369323,
"sampling/batch_mean_priority_error": 0.010328194444444435,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.25555555555555554,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0022848639637231827,
"sampling/priority_kl": 0.030000000447034835,
"sampling/priority_scale": 0.7543738663429395,
"sampling/prob_entropy": 10.278957939147949,
"sampling/prob_max": 5.354871318559162e-05,
"sampling/prob_min": 2.0208099886076526e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.567199993133545,
"sampling/prompt_draws_total": 47016.0,
"sampling/seen_fraction": 0.8722799897193909,
"sampling/unseen_fraction": 0.12772001028060914,
"signal/accuracy_reward/centered_abs_mean": 0.11072590947151184,
"signal/accuracy_reward/group_std_mean": 0.1466132655739784,
"signal/accuracy_reward/group_zero_std_frac": 0.5805555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05536295473575592,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05536295473575592,
"signal/advantage_abs_mean": 0.08102467954158783,
"signal/advantage_pre_scale_abs_mean": 0.08102467954158783,
"signal/advantage_pre_scale_std": 0.1670161157846451,
"signal/advantage_std": 0.1670161157846451,
"signal/brier_reward/centered_abs_mean": 0.07082972005009651,
"signal/brier_reward/group_std_mean": 0.09686807245016098,
"signal/brier_reward/group_zero_std_frac": 0.25555555820465087,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.035414860025048255,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.035414860025048255,
"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": 0.009939236287027597,
"signal/format_reward/group_std_mean": 0.022142794728279114,
"signal/format_reward/group_zero_std_frac": 0.8972222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004969618143513798,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004969618143513798,
"signal/mean_confidence_reward/centered_abs_mean": 0.062287604063749315,
"signal/mean_confidence_reward/group_std_mean": 0.08283003866672516,
"signal/mean_confidence_reward/group_zero_std_frac": 0.27222222089767456,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.22875973022019e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.22875973022019e-07,
"step": 655
},
{
"calibration/aurc": 0.0863774517347887,
"calibration/batch_distribution_entropy": 0.5152594103145469,
"calibration/batch_entropy_100bins": 0.2818539806355876,
"calibration/batch_entropy_10bins": 0.5152594103145469,
"calibration/batch_entropy_50bins": 0.3317939456450146,
"calibration/batch_uniqueness": 0.12971007215826963,
"calibration/confidence_entropy": 0.4175701939998537,
"calibration/coverage@0%": 0.004699753142956087,
"calibration/coverage@1%": 0.2383308997922163,
"calibration/coverage@10%": 0.6815367990148502,
"calibration/coverage@15%": 0.7087168189730201,
"calibration/coverage@20%": 0.8961352262837249,
"calibration/coverage@25%": 0.9514360313315928,
"calibration/coverage@30%": 0.9791122715404701,
"calibration/coverage@5%": 0.47374620658072814,
"calibration/distribution_entropy_10": 0.5152594103145469,
"calibration/distribution_entropy_100": 0.2818539806355876,
"calibration/ece": 0.10784367922937763,
"calibration/mean_confidence": 0.7653679821743472,
"calibration/unique_confidence_per_question": 0.025520833333333336,
"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": 0.00546875,
"completions/max_length": 3719.0,
"completions/max_terminated_length": 3719.0,
"completions/mean_length": 1113.911572265625,
"completions/mean_terminated_length": 1120.09814453125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 338.6,
"epoch": 1.5865384615384617,
"grad_norm": 0.000522863061632961,
"learning_rate": 2.283653846153846e-06,
"loss": -0.0064,
"num_tokens": 1696576859.0,
"reward": 1.3091982364654542,
"reward_std": 0.10364175438880921,
"rewards/accuracy_reward": 0.7514756917953491,
"rewards/brier_reward": 0.872548270225525,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9943576455116272,
"rewards/mean_confidence_reward": 0.7404010176658631,
"sampling/batch_mean_priority_error": 0.018774722222222213,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.20833333333333334,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0023031047079712153,
"sampling/priority_kl": 0.029999990016222,
"sampling/priority_scale": 0.756784874224104,
"sampling/prob_entropy": 10.278948020935058,
"sampling/prob_max": 5.3775295964442196e-05,
"sampling/prob_min": 2.0230506925145165e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.579199981689453,
"sampling/prompt_draws_total": 47376.0,
"sampling/seen_fraction": 0.8750333428382874,
"sampling/unseen_fraction": 0.12496665716171265,
"signal/accuracy_reward/centered_abs_mean": 0.0965549036860466,
"signal/accuracy_reward/group_std_mean": 0.1316729962825775,
"signal/accuracy_reward/group_zero_std_frac": 0.6027777791023254,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0482774518430233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0482774518430233,
"signal/advantage_abs_mean": 0.07157290279865265,
"signal/advantage_pre_scale_abs_mean": 0.07157290279865265,
"signal/advantage_pre_scale_std": 0.15894106924533843,
"signal/advantage_std": 0.15894106924533843,
"signal/brier_reward/centered_abs_mean": 0.06766726225614547,
"signal/brier_reward/group_std_mean": 0.09276285767555237,
"signal/brier_reward/group_zero_std_frac": 0.29722222983837127,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.033833631128072736,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.033833631128072736,
"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": 0.010508897621184588,
"signal/format_reward/group_std_mean": 0.024353000894188882,
"signal/format_reward/group_zero_std_frac": 0.8833333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005254448810592294,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005254448810592294,
"signal/mean_confidence_reward/centered_abs_mean": 0.057074010372161865,
"signal/mean_confidence_reward/group_std_mean": 0.07592665776610374,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3027777820825577,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.707401101062714e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.707401101062714e-07,
"step": 660
},
{
"calibration/aurc": 0.12276055299354609,
"calibration/batch_distribution_entropy": 0.5013685251115868,
"calibration/batch_entropy_100bins": 0.27919717568735347,
"calibration/batch_entropy_10bins": 0.5013685251115868,
"calibration/batch_entropy_50bins": 0.3286663978466972,
"calibration/batch_uniqueness": 0.1017443592414641,
"calibration/confidence_entropy": 0.41964784067908434,
"calibration/coverage@0%": 0.006303522885504853,
"calibration/coverage@1%": 0.24174120324829168,
"calibration/coverage@10%": 0.42533609125158983,
"calibration/coverage@15%": 0.6526996167428637,
"calibration/coverage@20%": 0.7504232809130803,
"calibration/coverage@25%": 0.8795194169072108,
"calibration/coverage@30%": 0.954172378872531,
"calibration/coverage@5%": 0.3723595899461068,
"calibration/distribution_entropy_10": 0.5013685251115868,
"calibration/distribution_entropy_100": 0.27919717568735347,
"calibration/ece": 0.12758788380365582,
"calibration/mean_confidence": 0.7620131446358779,
"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": 0.0046875,
"completions/max_length": 3796.8,
"completions/max_terminated_length": 3796.8,
"completions/mean_length": 1092.1130493164062,
"completions/mean_terminated_length": 1097.2896240234375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 324.4,
"epoch": 1.5985576923076923,
"grad_norm": 0.0005642919568344951,
"learning_rate": 2.2536057692307694e-06,
"loss": -0.0051,
"num_tokens": 1712260593.0,
"reward": 1.3006659269332885,
"reward_std": 0.12148597985506057,
"rewards/accuracy_reward": 0.7433159828186036,
"rewards/brier_reward": 0.8627753138542176,
"rewards/confidence_one_or_zero": 0.0004340277810115367,
"rewards/format_reward": 0.995225703716278,
"rewards/mean_confidence_reward": 0.7463836789131164,
"sampling/batch_mean_priority_error": 0.014619930555555546,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.18055555555555552,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0023194745648652316,
"sampling/priority_kl": 0.030000966414809226,
"sampling/priority_scale": 0.7585157931083814,
"sampling/prob_entropy": 10.278953742980956,
"sampling/prob_max": 5.397429194999859e-05,
"sampling/prob_min": 2.026002111961134e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.5911999940872192,
"sampling/prompt_draws_total": 47736.0,
"sampling/seen_fraction": 0.8773533344268799,
"sampling/unseen_fraction": 0.12264666557312012,
"signal/accuracy_reward/centered_abs_mean": 0.1187445729970932,
"signal/accuracy_reward/group_std_mean": 0.15946634411811828,
"signal/accuracy_reward/group_zero_std_frac": 0.5333333313465118,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0593722864985466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0593722864985466,
"signal/advantage_abs_mean": 0.0857117235660553,
"signal/advantage_pre_scale_abs_mean": 0.0857117235660553,
"signal/advantage_pre_scale_std": 0.1751396745443344,
"signal/advantage_std": 0.1751396745443344,
"signal/brier_reward/centered_abs_mean": 0.07548005133867264,
"signal/brier_reward/group_std_mean": 0.10238111168146133,
"signal/brier_reward/group_zero_std_frac": 0.22222222685813903,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03774002566933632,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03774002566933632,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0008409288129769266,
"signal/confidence_one_or_zero/group_std_mean": 0.0024552317336201668,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 8.409287488575501e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.409287488575501e-09,
"signal/format_reward/centered_abs_mean": 0.008805338572710753,
"signal/format_reward/group_std_mean": 0.021177830919623376,
"signal/format_reward/group_zero_std_frac": 0.8944444537162781,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004402669286355377,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004402669286355377,
"signal/mean_confidence_reward/centered_abs_mean": 0.06224251911044121,
"signal/mean_confidence_reward/group_std_mean": 0.0842223346233368,
"signal/mean_confidence_reward/group_zero_std_frac": 0.23611111640930177,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.224251819730853e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.224251819730853e-07,
"step": 665
},
{
"calibration/aurc": 0.14450476197851136,
"calibration/batch_distribution_entropy": 0.5765642673767126,
"calibration/batch_entropy_100bins": 0.32623106838479715,
"calibration/batch_entropy_10bins": 0.5765642673767126,
"calibration/batch_entropy_50bins": 0.3840339353281199,
"calibration/batch_uniqueness": 0.3014844645350637,
"calibration/confidence_entropy": 0.44685345164390433,
"calibration/coverage@0%": 0.10156524122777513,
"calibration/coverage@1%": 0.24797121854016604,
"calibration/coverage@10%": 0.39848095112594883,
"calibration/coverage@15%": 0.526121066747444,
"calibration/coverage@20%": 0.70951778967666,
"calibration/coverage@25%": 0.736237360957404,
"calibration/coverage@30%": 0.9067455000069071,
"calibration/coverage@5%": 0.26001310335691996,
"calibration/distribution_entropy_10": 0.5765642673767126,
"calibration/distribution_entropy_100": 0.32623106838479715,
"calibration/ece": 0.12060073742183959,
"calibration/mean_confidence": 0.7040622262912196,
"calibration/unique_confidence_per_question": 0.026041666666666668,
"calibration/unique_confidences": 10.0,
"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": 0.008159722222222231,
"completions/max_length": 3857.2,
"completions/max_terminated_length": 3857.2,
"completions/mean_length": 1086.3653076171875,
"completions/mean_terminated_length": 1095.41630859375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 307.0,
"epoch": 1.6105769230769231,
"grad_norm": 0.0005110831116326153,
"learning_rate": 2.2235576923076924e-06,
"loss": -0.0096,
"num_tokens": 1727855969.0,
"reward": 1.2902185916900635,
"reward_std": 0.11893133968114852,
"rewards/accuracy_reward": 0.7259548664093017,
"rewards/brier_reward": 0.8628010869026184,
"rewards/confidence_one_or_zero": 0.0009548611182253808,
"rewards/format_reward": 0.9916666626930237,
"rewards/mean_confidence_reward": 0.7324783086776734,
"sampling/batch_mean_priority_error": 0.013338541666666653,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.20833333333333334,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0023346453439444304,
"sampling/priority_kl": 0.030000185966491698,
"sampling/priority_scale": 0.760347467684187,
"sampling/prob_entropy": 10.278972244262695,
"sampling/prob_max": 5.417778957053088e-05,
"sampling/prob_min": 2.02883380552521e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6032000064849854,
"sampling/prompt_draws_total": 48096.0,
"sampling/seen_fraction": 0.8797000050544739,
"sampling/unseen_fraction": 0.12029999494552612,
"signal/accuracy_reward/centered_abs_mean": 0.11508788913488388,
"signal/accuracy_reward/group_std_mean": 0.15065041184425354,
"signal/accuracy_reward/group_zero_std_frac": 0.5777777850627899,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05754394456744194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05754394456744194,
"signal/advantage_abs_mean": 0.08531344085931777,
"signal/advantage_pre_scale_abs_mean": 0.08531344085931777,
"signal/advantage_pre_scale_std": 0.17639075517654418,
"signal/advantage_std": 0.17639075517654418,
"signal/brier_reward/centered_abs_mean": 0.0753116026520729,
"signal/brier_reward/group_std_mean": 0.10187897086143494,
"signal/brier_reward/group_zero_std_frac": 0.2527777820825577,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03765580132603645,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03765580132603645,
"signal/confidence_one_or_zero/centered_abs_mean": 0.001719835086259991,
"signal/confidence_one_or_zero/group_std_mean": 0.0033398654311895372,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.7198350121816475e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.7198350121816475e-08,
"signal/format_reward/centered_abs_mean": 0.014876302052289247,
"signal/format_reward/group_std_mean": 0.029834812879562377,
"signal/format_reward/group_zero_std_frac": 0.8722222328186036,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007438151026144623,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007438151026144623,
"signal/mean_confidence_reward/centered_abs_mean": 0.06751834452152253,
"signal/mean_confidence_reward/group_std_mean": 0.08781284093856812,
"signal/mean_confidence_reward/group_zero_std_frac": 0.26111111640930174,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.751834575879911e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.751834575879911e-07,
"step": 670
},
{
"calibration/aurc": 0.10970071896599873,
"calibration/batch_distribution_entropy": 0.568013647591677,
"calibration/batch_entropy_100bins": 0.3102414092567368,
"calibration/batch_entropy_10bins": 0.568013647591677,
"calibration/batch_entropy_50bins": 0.3652111673130844,
"calibration/batch_uniqueness": 0.22665178946390013,
"calibration/confidence_entropy": 0.4394298617068798,
"calibration/coverage@0%": 0.0010443864229765013,
"calibration/coverage@1%": 0.11925019117231687,
"calibration/coverage@10%": 0.5128259821504217,
"calibration/coverage@15%": 0.7708662648180452,
"calibration/coverage@20%": 0.8658396625012432,
"calibration/coverage@25%": 0.9119364882506528,
"calibration/coverage@30%": 0.9583999999999999,
"calibration/coverage@5%": 0.27560069804636217,
"calibration/distribution_entropy_10": 0.568013647591677,
"calibration/distribution_entropy_100": 0.3102414092567368,
"calibration/ece": 0.07774484295633795,
"calibration/mean_confidence": 0.736116044646728,
"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": 0.011805555555555559,
"completions/max_length": 3540.6,
"completions/max_terminated_length": 3540.6,
"completions/mean_length": 1091.9043701171875,
"completions/mean_terminated_length": 1105.053369140625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 341.6,
"epoch": 1.6225961538461537,
"grad_norm": 0.0005459992680698633,
"learning_rate": 2.1935096153846157e-06,
"loss": -0.0141,
"num_tokens": 1743536499.0,
"reward": 1.2934834003448485,
"reward_std": 0.12105322629213333,
"rewards/accuracy_reward": 0.7386284708976746,
"rewards/brier_reward": 0.8602157592773437,
"rewards/confidence_one_or_zero": 0.0006076389050576836,
"rewards/format_reward": 0.9881076335906982,
"rewards/mean_confidence_reward": 0.7474018931388855,
"sampling/batch_mean_priority_error": 0.017633680555555545,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.16388888888888892,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0023507375735789537,
"sampling/priority_kl": 0.03000086210668087,
"sampling/priority_scale": 0.7620475710136816,
"sampling/prob_entropy": 10.27895736694336,
"sampling/prob_max": 5.437384315882809e-05,
"sampling/prob_min": 2.031473413808271e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6151999950408935,
"sampling/prompt_draws_total": 48456.0,
"sampling/seen_fraction": 0.8819133400917053,
"sampling/unseen_fraction": 0.11808665990829467,
"signal/accuracy_reward/centered_abs_mean": 0.10015733391046525,
"signal/accuracy_reward/group_std_mean": 0.13727897256612778,
"signal/accuracy_reward/group_zero_std_frac": 0.5833333492279053,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05007866695523262,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05007866695523262,
"signal/advantage_abs_mean": 0.07990130856633186,
"signal/advantage_pre_scale_abs_mean": 0.07990130856633186,
"signal/advantage_pre_scale_std": 0.1813565194606781,
"signal/advantage_std": 0.1813565194606781,
"signal/brier_reward/centered_abs_mean": 0.06943394243717194,
"signal/brier_reward/group_std_mean": 0.09994765222072602,
"signal/brier_reward/group_zero_std_frac": 0.2611111134290695,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03471697121858597,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03471697121858597,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0011555989156477153,
"signal/confidence_one_or_zero/group_std_mean": 0.002839442901313305,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.1555989232192587e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.1555989232192587e-08,
"signal/format_reward/centered_abs_mean": 0.021370442770421504,
"signal/format_reward/group_std_mean": 0.04681904092431068,
"signal/format_reward/group_zero_std_frac": 0.7805555582046508,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010685221385210752,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010685221385210752,
"signal/mean_confidence_reward/centered_abs_mean": 0.061398386210203174,
"signal/mean_confidence_reward/group_std_mean": 0.08617035299539566,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2722222238779068,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.139838546914688e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.139838546914688e-07,
"step": 675
},
{
"calibration/aurc": 0.09587670960100644,
"calibration/batch_distribution_entropy": 0.6175683596353149,
"calibration/batch_entropy_100bins": 0.3329512515331028,
"calibration/batch_entropy_10bins": 0.6175683596353149,
"calibration/batch_entropy_50bins": 0.39194482619865295,
"calibration/batch_uniqueness": 0.3871926237126157,
"calibration/confidence_entropy": 0.46478713643106306,
"calibration/coverage@0%": 0.0843501326259947,
"calibration/coverage@1%": 0.0843501326259947,
"calibration/coverage@10%": 0.5346998344195839,
"calibration/coverage@15%": 0.7661888774463907,
"calibration/coverage@20%": 0.892814975451509,
"calibration/coverage@25%": 0.9522546419098143,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.40982914163222484,
"calibration/distribution_entropy_10": 0.6175683596353149,
"calibration/distribution_entropy_100": 0.3329512515331028,
"calibration/ece": 0.10635281530775749,
"calibration/mean_confidence": 0.738215510844212,
"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": 0.013541666666666697,
"completions/max_length": 3670.6,
"completions/max_terminated_length": 3670.6,
"completions/mean_length": 1083.8117431640626,
"completions/mean_terminated_length": 1098.7698486328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 303.6,
"epoch": 1.6346153846153846,
"grad_norm": 0.0005681436159648001,
"learning_rate": 2.1634615384615387e-06,
"loss": -0.0154,
"num_tokens": 1759137466.0,
"reward": 1.2990227937698364,
"reward_std": 0.12866669595241548,
"rewards/accuracy_reward": 0.7521701455116272,
"rewards/brier_reward": 0.8594894886016846,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9863715291023254,
"rewards/mean_confidence_reward": 0.724100124835968,
"sampling/batch_mean_priority_error": 0.014114583333333323,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.18611111111111112,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0023681627586483955,
"sampling/priority_kl": 0.03000106252729893,
"sampling/priority_scale": 0.7635146678658202,
"sampling/prob_entropy": 10.278950500488282,
"sampling/prob_max": 5.456006110762246e-05,
"sampling/prob_min": 2.0344129006844013e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6272000074386597,
"sampling/prompt_draws_total": 48816.0,
"sampling/seen_fraction": 0.8839533448219299,
"sampling/unseen_fraction": 0.11604665517807007,
"signal/accuracy_reward/centered_abs_mean": 0.10556098073720932,
"signal/accuracy_reward/group_std_mean": 0.1469675600528717,
"signal/accuracy_reward/group_zero_std_frac": 0.5499999940395355,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05278049036860466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05278049036860466,
"signal/advantage_abs_mean": 0.08514032810926438,
"signal/advantage_pre_scale_abs_mean": 0.08514032810926438,
"signal/advantage_pre_scale_std": 0.18456737399101258,
"signal/advantage_std": 0.18456737399101258,
"signal/brier_reward/centered_abs_mean": 0.0730242095887661,
"signal/brier_reward/group_std_mean": 0.10391690582036972,
"signal/brier_reward/group_zero_std_frac": 0.2472222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03651210479438305,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03651210479438305,
"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": 0.0242567278444767,
"signal/format_reward/group_std_mean": 0.04920388534665108,
"signal/format_reward/group_zero_std_frac": 0.7861111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01212836392223835,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01212836392223835,
"signal/mean_confidence_reward/centered_abs_mean": 0.06222810372710228,
"signal/mean_confidence_reward/group_std_mean": 0.08487941175699235,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2555555611848831,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.222810156941705e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.222810156941705e-07,
"step": 680
},
{
"calibration/aurc": 0.11139059240130803,
"calibration/batch_distribution_entropy": 0.5826805325067836,
"calibration/batch_entropy_100bins": 0.31370010142444527,
"calibration/batch_entropy_10bins": 0.5826805325067836,
"calibration/batch_entropy_50bins": 0.3692826837717405,
"calibration/batch_uniqueness": 0.26415515433613046,
"calibration/confidence_entropy": 0.44579667814925283,
"calibration/coverage@0%": 0.0015789473684210526,
"calibration/coverage@1%": 0.21343152310484434,
"calibration/coverage@10%": 0.6413132971561079,
"calibration/coverage@15%": 0.6920358264160477,
"calibration/coverage@20%": 0.7286971880429265,
"calibration/coverage@25%": 0.9074074074074074,
"calibration/coverage@30%": 0.9465608465608465,
"calibration/coverage@5%": 0.2941253664665643,
"calibration/distribution_entropy_10": 0.5826805325067836,
"calibration/distribution_entropy_100": 0.31370010142444527,
"calibration/ece": 0.0968422284319643,
"calibration/mean_confidence": 0.7433893348079172,
"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": 0.01909722222222223,
"completions/max_length": 3731.2,
"completions/max_terminated_length": 3731.2,
"completions/mean_length": 1044.6437622070312,
"completions/mean_terminated_length": 1065.1299072265624,
"completions/min_length": 0.0,
"completions/min_terminated_length": 362.4,
"epoch": 1.6466346153846154,
"grad_norm": 0.0005210732924751937,
"learning_rate": 2.1334134615384616e-06,
"loss": -0.0232,
"num_tokens": 1774231634.0,
"reward": 1.3034531831741334,
"reward_std": 0.14097131937742233,
"rewards/accuracy_reward": 0.7689236164093017,
"rewards/brier_reward": 0.8570653319358825,
"rewards/confidence_one_or_zero": 0.0004340277868323028,
"rewards/format_reward": 0.9809027671813965,
"rewards/mean_confidence_reward": 0.7295277953147888,
"sampling/batch_mean_priority_error": 0.0184565972222222,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.16111111111111112,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.002385252341628075,
"sampling/priority_kl": 0.029998932778835297,
"sampling/priority_scale": 0.7653319059638306,
"sampling/prob_entropy": 10.278940200805664,
"sampling/prob_max": 5.4764158267062155e-05,
"sampling/prob_min": 2.0372347353259102e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6392000198364258,
"sampling/prompt_draws_total": 49176.0,
"sampling/seen_fraction": 0.8861200094223023,
"sampling/unseen_fraction": 0.11387999057769775,
"signal/accuracy_reward/centered_abs_mean": 0.10553385317325592,
"signal/accuracy_reward/group_std_mean": 0.14716241806745528,
"signal/accuracy_reward/group_zero_std_frac": 0.550000011920929,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05276692658662796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05276692658662796,
"signal/advantage_abs_mean": 0.09143907949328423,
"signal/advantage_pre_scale_abs_mean": 0.09143907949328423,
"signal/advantage_pre_scale_std": 0.19914956390857697,
"signal/advantage_std": 0.19914956390857697,
"signal/brier_reward/centered_abs_mean": 0.07749231234192848,
"signal/brier_reward/group_std_mean": 0.11150548756122589,
"signal/brier_reward/group_zero_std_frac": 0.21944444477558137,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03874615617096424,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03874615617096424,
"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": 8.192273526219651e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.192273526219651e-09,
"signal/format_reward/centered_abs_mean": 0.03379991315305233,
"signal/format_reward/group_std_mean": 0.06723003908991813,
"signal/format_reward/group_zero_std_frac": 0.7111111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.016899956576526164,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.016899956576526164,
"signal/mean_confidence_reward/centered_abs_mean": 0.06801529824733735,
"signal/mean_confidence_reward/group_std_mean": 0.09337150603532791,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2333333373069763,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.80152959375846e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.80152959375846e-07,
"step": 685
},
{
"calibration/aurc": 0.15859837238043628,
"calibration/batch_distribution_entropy": 0.5560990964924017,
"calibration/batch_entropy_100bins": 0.30055202226479294,
"calibration/batch_entropy_10bins": 0.5560990964924017,
"calibration/batch_entropy_50bins": 0.3538049777191362,
"calibration/batch_uniqueness": 0.23692587593867795,
"calibration/confidence_entropy": 0.4395761004833318,
"calibration/coverage@0%": 0.10421626984126986,
"calibration/coverage@1%": 0.10421626984126986,
"calibration/coverage@10%": 0.4268589877471457,
"calibration/coverage@15%": 0.5054053536619326,
"calibration/coverage@20%": 0.6296827137287664,
"calibration/coverage@25%": 0.6529631370091896,
"calibration/coverage@30%": 0.8776994569757728,
"calibration/coverage@5%": 0.2671792328042328,
"calibration/distribution_entropy_10": 0.5560990964924017,
"calibration/distribution_entropy_100": 0.30055202226479294,
"calibration/ece": 0.1268294085909217,
"calibration/mean_confidence": 0.7558809610832637,
"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": 0.012326388888888883,
"completions/max_length": 3738.0,
"completions/max_terminated_length": 3738.0,
"completions/mean_length": 1039.7527099609374,
"completions/mean_terminated_length": 1052.7323120117187,
"completions/min_length": 0.0,
"completions/min_terminated_length": 337.0,
"epoch": 1.6586538461538463,
"grad_norm": 0.0005724552902393043,
"learning_rate": 2.103365384615385e-06,
"loss": -0.0163,
"num_tokens": 1789305713.0,
"reward": 1.2902822971343995,
"reward_std": 0.12336039692163467,
"rewards/accuracy_reward": 0.7328124880790711,
"rewards/brier_reward": 0.8601503968238831,
"rewards/confidence_one_or_zero": 0.0006076388934161514,
"rewards/format_reward": 0.9875868082046508,
"rewards/mean_confidence_reward": 0.7491449594497681,
"sampling/batch_mean_priority_error": 0.016593749999999984,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.18055555555555555,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0024040149059146644,
"sampling/priority_kl": 0.029999926313757898,
"sampling/priority_scale": 0.7672343671089038,
"sampling/prob_entropy": 10.27895393371582,
"sampling/prob_max": 5.49718904949259e-05,
"sampling/prob_min": 2.039941209659446e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.651200008392334,
"sampling/prompt_draws_total": 49536.0,
"sampling/seen_fraction": 0.8882933378219604,
"sampling/unseen_fraction": 0.11170666217803955,
"signal/accuracy_reward/centered_abs_mean": 0.0985568568110466,
"signal/accuracy_reward/group_std_mean": 0.13777171075344086,
"signal/accuracy_reward/group_zero_std_frac": 0.5722222268581391,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0492784284055233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0492784284055233,
"signal/advantage_abs_mean": 0.08106912076473236,
"signal/advantage_pre_scale_abs_mean": 0.08106912076473236,
"signal/advantage_pre_scale_std": 0.1815045118331909,
"signal/advantage_std": 0.1815045118331909,
"signal/brier_reward/centered_abs_mean": 0.06963968127965928,
"signal/brier_reward/group_std_mean": 0.10002299994230271,
"signal/brier_reward/group_zero_std_frac": 0.2666666656732559,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03481984063982964,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03481984063982964,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0011555989389307798,
"signal/confidence_one_or_zero/group_std_mean": 0.0028394428081810474,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.1555989232192587e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.1555989232192587e-08,
"signal/format_reward/centered_abs_mean": 0.02232530377805233,
"signal/format_reward/group_std_mean": 0.047450629994273186,
"signal/format_reward/group_zero_std_frac": 0.7861111044883728,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.011162651889026164,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.011162651889026164,
"signal/mean_confidence_reward/centered_abs_mean": 0.0594577394425869,
"signal/mean_confidence_reward/group_std_mean": 0.08299815356731415,
"signal/mean_confidence_reward/group_zero_std_frac": 0.275,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.945773523308162e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.945773523308162e-07,
"step": 690
},
{
"calibration/aurc": 0.14306273310915993,
"calibration/batch_distribution_entropy": 0.5412362104100029,
"calibration/batch_entropy_100bins": 0.28788372913303284,
"calibration/batch_entropy_10bins": 0.5412362104100029,
"calibration/batch_entropy_50bins": 0.3388920680156938,
"calibration/batch_uniqueness": 0.19993184384829876,
"calibration/confidence_entropy": 0.43030819920007846,
"calibration/coverage@0%": 0.0015679466324005852,
"calibration/coverage@1%": 0.0015679466324005852,
"calibration/coverage@10%": 0.29376102142085764,
"calibration/coverage@15%": 0.5923256735882328,
"calibration/coverage@20%": 0.6827225130890052,
"calibration/coverage@25%": 0.9539267015706805,
"calibration/coverage@30%": 0.9952879581151833,
"calibration/coverage@5%": 0.12271677169767474,
"calibration/distribution_entropy_10": 0.5412362104100029,
"calibration/distribution_entropy_100": 0.28788372913303284,
"calibration/ece": 0.10015638456385936,
"calibration/mean_confidence": 0.7878304375760394,
"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": 0.011805555555555559,
"completions/max_length": 3752.0,
"completions/max_terminated_length": 3752.0,
"completions/mean_length": 1083.738134765625,
"completions/mean_terminated_length": 1096.7212158203124,
"completions/min_length": 0.0,
"completions/min_terminated_length": 365.8,
"epoch": 1.6706730769230769,
"grad_norm": 0.0005091908969916403,
"learning_rate": 2.073317307692308e-06,
"loss": -0.014,
"num_tokens": 1804906792.0,
"reward": 1.2935439109802247,
"reward_std": 0.13235266357660294,
"rewards/accuracy_reward": 0.7440104126930237,
"rewards/brier_reward": 0.8550411939620972,
"rewards/confidence_one_or_zero": 0.0004340277810115367,
"rewards/format_reward": 0.9880208134651184,
"rewards/mean_confidence_reward": 0.7656136989593506,
"sampling/batch_mean_priority_error": 0.015585138888888872,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.19444444444444448,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0024218444246798753,
"sampling/priority_kl": 0.030001531541347503,
"sampling/priority_scale": 0.7690952242119238,
"sampling/prob_entropy": 10.278961181640625,
"sampling/prob_max": 5.517598328879103e-05,
"sampling/prob_min": 2.0425973343662917e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6631999969482423,
"sampling/prompt_draws_total": 49896.0,
"sampling/seen_fraction": 0.8903999924659729,
"sampling/unseen_fraction": 0.1096000075340271,
"signal/accuracy_reward/centered_abs_mean": 0.10761176198720931,
"signal/accuracy_reward/group_std_mean": 0.14982850253582,
"signal/accuracy_reward/group_zero_std_frac": 0.5416666805744171,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05380588099360466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05380588099360466,
"signal/advantage_abs_mean": 0.0892701119184494,
"signal/advantage_pre_scale_abs_mean": 0.0892701119184494,
"signal/advantage_pre_scale_std": 0.19025329649448394,
"signal/advantage_std": 0.19025329649448394,
"signal/brier_reward/centered_abs_mean": 0.0762951672077179,
"signal/brier_reward/group_std_mean": 0.10774291157722474,
"signal/brier_reward/group_zero_std_frac": 0.28333333134651184,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03814758360385895,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03814758360385895,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0008409288129769266,
"signal/confidence_one_or_zero/group_std_mean": 0.0024552317336201668,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 8.409287488575501e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.409287488575501e-09,
"signal/format_reward/centered_abs_mean": 0.02054036445915699,
"signal/format_reward/group_std_mean": 0.04146440476179123,
"signal/format_reward/group_zero_std_frac": 0.8222222208976746,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010270182229578494,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010270182229578494,
"signal/mean_confidence_reward/centered_abs_mean": 0.05742166191339493,
"signal/mean_confidence_reward/group_std_mean": 0.08025620728731156,
"signal/mean_confidence_reward/group_zero_std_frac": 0.31388888955116273,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.74216630866431e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.74216630866431e-07,
"step": 695
},
{
"calibration/aurc": 0.17708395332896149,
"calibration/batch_distribution_entropy": 0.47968307546433503,
"calibration/batch_entropy_100bins": 0.2581500156911111,
"calibration/batch_entropy_10bins": 0.47968307546433503,
"calibration/batch_entropy_50bins": 0.30389002163931644,
"calibration/batch_uniqueness": 0.0036444937827355916,
"calibration/confidence_entropy": 0.41149710071485873,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.1581151832460733,
"calibration/coverage@10%": 0.18272251308900525,
"calibration/coverage@15%": 0.3733681462140992,
"calibration/coverage@20%": 0.6550717906875544,
"calibration/coverage@25%": 0.7416890339425588,
"calibration/coverage@30%": 0.8869976764968722,
"calibration/coverage@5%": 0.17172774869109947,
"calibration/distribution_entropy_10": 0.47968307546433503,
"calibration/distribution_entropy_100": 0.2581500156911111,
"calibration/ece": 0.11511028309375874,
"calibration/mean_confidence": 0.779884540079836,
"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": 0.0140625,
"completions/max_length": 4008.0,
"completions/max_terminated_length": 4008.0,
"completions/mean_length": 1111.742431640625,
"completions/mean_terminated_length": 1127.698046875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 359.6,
"epoch": 1.6826923076923077,
"grad_norm": 0.0005041222320869565,
"learning_rate": 2.043269230769231e-06,
"loss": -0.0171,
"num_tokens": 1820841809.0,
"reward": 1.272655439376831,
"reward_std": 0.13738891631364822,
"rewards/accuracy_reward": 0.7166666626930237,
"rewards/brier_reward": 0.8430387616157532,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9855902791023254,
"rewards/mean_confidence_reward": 0.7627676367759705,
"sampling/batch_mean_priority_error": 0.02263194444444442,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.17777777777777776,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0024414270650595427,
"sampling/priority_kl": 0.030000188574194907,
"sampling/priority_scale": 0.7714008987648413,
"sampling/prob_entropy": 10.278949165344239,
"sampling/prob_max": 5.540331112570129e-05,
"sampling/prob_min": 2.044780703727156e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6751999855041504,
"sampling/prompt_draws_total": 50256.0,
"sampling/seen_fraction": 0.8926600098609925,
"sampling/unseen_fraction": 0.10733999013900757,
"signal/accuracy_reward/centered_abs_mean": 0.11134982705116273,
"signal/accuracy_reward/group_std_mean": 0.15158891975879668,
"signal/accuracy_reward/group_zero_std_frac": 0.5500000178813934,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05567491352558136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05567491352558136,
"signal/advantage_abs_mean": 0.09317022860050202,
"signal/advantage_pre_scale_abs_mean": 0.09317022860050202,
"signal/advantage_pre_scale_std": 0.19810806214809418,
"signal/advantage_std": 0.19810806214809418,
"signal/brier_reward/centered_abs_mean": 0.07885801494121551,
"signal/brier_reward/group_std_mean": 0.11172184944152833,
"signal/brier_reward/group_zero_std_frac": 0.30555555820465086,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.039429007470607756,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.039429007470607756,
"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": 0.02542317658662796,
"signal/format_reward/group_std_mean": 0.05069509223103523,
"signal/format_reward/group_zero_std_frac": 0.7833333492279053,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01271158829331398,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01271158829331398,
"signal/mean_confidence_reward/centered_abs_mean": 0.059342540055513385,
"signal/mean_confidence_reward/group_std_mean": 0.08301161825656891,
"signal/mean_confidence_reward/group_zero_std_frac": 0.338888892531395,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.934253863415507e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.934253863415507e-07,
"step": 700
},
{
"epoch": 1.6826923076923077,
"eval_calibration/aurc": 0.14504904838655935,
"eval_calibration/batch_distribution_entropy": 0.5285209554550367,
"eval_calibration/batch_entropy_100bins": 0.28874062383944954,
"eval_calibration/batch_entropy_10bins": 0.5285209554550367,
"eval_calibration/batch_entropy_50bins": 0.33990079059964734,
"eval_calibration/batch_uniqueness": 0.12871531778070863,
"eval_calibration/confidence_entropy": 0.42277103690494733,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.0,
"eval_calibration/coverage@15%": 0.7372583479789103,
"eval_calibration/coverage@20%": 0.8295254833040422,
"eval_calibration/coverage@25%": 0.9332161687170475,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.5285209554550367,
"eval_calibration/distribution_entropy_100": 0.28874062383944954,
"eval_calibration/ece": 0.05492091388400712,
"eval_calibration/mean_confidence": 0.779701230228471,
"eval_calibration/unique_confidence_per_question": 0.009548611111111112,
"eval_calibration/unique_confidences": 11,
"eval_completions/clipped_ratio": 0.009548611111111105,
"eval_completions/max_length": 2959.8333333333335,
"eval_completions/max_terminated_length": 2959.8333333333335,
"eval_completions/mean_length": 1100.4611206054688,
"eval_completions/mean_terminated_length": 1111.1649169921875,
"eval_completions/min_length": 157.33333333333334,
"eval_completions/min_terminated_length": 408.8333333333333,
"eval_loss": 0.0,
"eval_num_tokens": 1820841809.0,
"eval_reward": 1.2722798387209575,
"eval_reward_std": 0.3444438676039378,
"eval_rewards/accuracy_reward": 0.7170138855775198,
"eval_rewards/brier_reward": 0.8396831452846527,
"eval_rewards/confidence_one_or_zero": 0.0008680555814256271,
"eval_rewards/format_reward": 0.987847218910853,
"eval_rewards/mean_confidence_reward": 0.7702256739139557,
"eval_runtime": 206.6268,
"eval_samples_per_second": 4.84,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3889973958333333,
"eval_signal/accuracy_reward/group_std_mean": 0.4453851133584976,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19449869791666666,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19449869791666666,
"eval_signal/advantage_abs_mean": 0.2865140189727147,
"eval_signal/advantage_pre_scale_abs_mean": 0.2865140189727147,
"eval_signal/advantage_pre_scale_std": 0.343355471889178,
"eval_signal/advantage_std": 0.343355471889178,
"eval_signal/brier_reward/centered_abs_mean": 0.19361031303803125,
"eval_signal/brier_reward/group_std_mean": 0.2558669224381447,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.09680515651901563,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.09680515651901563,
"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": 0.023111979166666668,
"eval_signal/format_reward/group_std_mean": 0.05925192994376024,
"eval_signal/format_reward/group_zero_std_frac": 0.694444457689921,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.011555989583333334,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.011555989583333334,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.1663682535290718,
"eval_signal/mean_confidence_reward/group_std_mean": 0.2147673244277636,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.6636825156031894e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.6636825156031894e-06,
"eval_steps_per_second": 0.029,
"step": 700
},
{
"epoch": 1.6826923076923077,
"step": 700,
"train_probe_calibration/aurc": 0.09761829923854591,
"train_probe_calibration/batch_distribution_entropy": 0.5113301053234623,
"train_probe_calibration/batch_entropy_100bins": 0.2753161203855543,
"train_probe_calibration/batch_entropy_10bins": 0.5113301053234623,
"train_probe_calibration/batch_entropy_50bins": 0.3240976823403679,
"train_probe_calibration/batch_uniqueness": 0.0860380738460258,
"train_probe_calibration/confidence_entropy": 0.4179497070733501,
"train_probe_calibration/coverage@0%": 0.000877963125548727,
"train_probe_calibration/coverage@1%": 0.000877963125548727,
"train_probe_calibration/coverage@10%": 0.6611062335381914,
"train_probe_calibration/coverage@15%": 0.8674275680421423,
"train_probe_calibration/coverage@20%": 0.9429323968393327,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.000877963125548727,
"train_probe_calibration/distribution_entropy_10": 0.5113301053234623,
"train_probe_calibration/distribution_entropy_100": 0.2753161203855543,
"train_probe_calibration/ece": 0.04934152765583824,
"train_probe_calibration/mean_confidence": 0.7838454784899035,
"train_probe_calibration/unique_confidence_per_question": 0.008680555555555556,
"train_probe_calibration/unique_confidences": 10,
"train_probe_completions/clipped_ratio": 0.012847222222222213,
"train_probe_completions/max_length": 3267.8333333333335,
"train_probe_completions/max_terminated_length": 3267.8333333333335,
"train_probe_completions/mean_length": 1117.0877278645833,
"train_probe_completions/mean_terminated_length": 1131.6324666341145,
"train_probe_completions/min_length": 68.83333333333333,
"train_probe_completions/min_terminated_length": 401.6666666666667,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1820841809.0,
"train_probe_reward": 1.3121604919433594,
"train_probe_reward_std": 0.3140617559353511,
"train_probe_rewards/accuracy_reward": 0.7647569477558136,
"train_probe_rewards/brier_reward": 0.8708333671092987,
"train_probe_rewards/confidence_one_or_zero": 0.0008680555814256271,
"train_probe_rewards/format_reward": 0.9887152711550394,
"train_probe_rewards/mean_confidence_reward": 0.7749999562899271,
"train_probe_runtime": 215.9823,
"train_probe_samples_per_second": 4.63,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3469509482383728,
"train_probe_signal/accuracy_reward/group_std_mean": 0.41993160049120587,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1734754741191864,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.1734754741191864,
"train_probe_signal/advantage_abs_mean": 0.2481620361407598,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.2481620361407598,
"train_probe_signal/advantage_pre_scale_std": 0.31520261367162067,
"train_probe_signal/advantage_std": 0.31520261367162067,
"train_probe_signal/brier_reward/centered_abs_mean": 0.15899253884951273,
"train_probe_signal/brier_reward/group_std_mean": 0.2224351391196251,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07949626942475636,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07949626942475636,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0016818575871487458,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0049104637776811915,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222288449606,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151002e-08,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151002e-08,
"train_probe_signal/format_reward/centered_abs_mean": 0.021647135261446238,
"train_probe_signal/format_reward/group_std_mean": 0.057857212610542774,
"train_probe_signal/format_reward/group_zero_std_frac": 0.6944444626569748,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.010823567630723119,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.010823567630723119,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.1647297888994217,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.21364916115999222,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.6472978738117188e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.6472978738117188e-06,
"train_probe_steps_per_second": 0.028
},
{
"calibration/aurc": 0.09527763748953584,
"calibration/batch_distribution_entropy": 0.4666990231139505,
"calibration/batch_entropy_100bins": 0.2543638325193935,
"calibration/batch_entropy_10bins": 0.4666990231139505,
"calibration/batch_entropy_50bins": 0.2994329880695009,
"calibration/batch_uniqueness": -0.0006402490523223392,
"calibration/confidence_entropy": 0.41268119792488084,
"calibration/coverage@0%": 0.001591715136898851,
"calibration/coverage@1%": 0.11693262422780795,
"calibration/coverage@10%": 0.6053837453253181,
"calibration/coverage@15%": 0.6613588502464217,
"calibration/coverage@20%": 0.9261853853413919,
"calibration/coverage@25%": 0.9798408488063661,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.4151094610454094,
"calibration/distribution_entropy_10": 0.4666990231139505,
"calibration/distribution_entropy_100": 0.2543638325193935,
"calibration/ece": 0.10833089271311573,
"calibration/mean_confidence": 0.7738942866147162,
"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": 0.02387152777777777,
"completions/max_length": 3863.6,
"completions/max_terminated_length": 3863.6,
"completions/mean_length": 1091.7198974609375,
"completions/mean_terminated_length": 1118.4887451171876,
"completions/min_length": 0.0,
"completions/min_terminated_length": 347.2,
"epoch": 1.6947115384615383,
"grad_norm": 0.0005031170439906418,
"learning_rate": 2.013221153846154e-06,
"loss": -0.0285,
"num_tokens": 1836506294.0,
"reward": 1.2860944509506225,
"reward_std": 0.15281510651111602,
"rewards/accuracy_reward": 0.7423611164093018,
"rewards/brier_reward": 0.8538578391075134,
"rewards/confidence_one_or_zero": 0.0006944444554392249,
"rewards/format_reward": 0.9759548664093017,
"rewards/mean_confidence_reward": 0.7574574589729309,
"sampling/batch_mean_priority_error": 0.015291874999999988,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.20555555555555555,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0024599439930170774,
"sampling/priority_kl": 0.029999907687306406,
"sampling/priority_scale": 0.7738102614646778,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 5.563581798924133e-05,
"sampling/prob_min": 2.0469454102567398e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6871999979019165,
"sampling/prompt_draws_total": 50616.0,
"sampling/seen_fraction": 0.894920003414154,
"sampling/unseen_fraction": 0.10507999658584595,
"signal/accuracy_reward/centered_abs_mean": 0.11489800363779068,
"signal/accuracy_reward/group_std_mean": 0.15308040380477905,
"signal/accuracy_reward/group_zero_std_frac": 0.5527777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05744900181889534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05744900181889534,
"signal/advantage_abs_mean": 0.1047695130109787,
"signal/advantage_pre_scale_abs_mean": 0.1047695130109787,
"signal/advantage_pre_scale_std": 0.22305024564266204,
"signal/advantage_std": 0.22305024564266204,
"signal/brier_reward/centered_abs_mean": 0.08341319859027863,
"signal/brier_reward/group_std_mean": 0.11793318986892701,
"signal/brier_reward/group_zero_std_frac": 0.3083333373069763,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04170659929513931,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04170659929513931,
"signal/confidence_one_or_zero/centered_abs_mean": 0.001312933990266174,
"signal/confidence_one_or_zero/group_std_mean": 0.0030315483920276163,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.3129339393458395e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.3129339393458395e-08,
"signal/format_reward/centered_abs_mean": 0.04051106758415699,
"signal/format_reward/group_std_mean": 0.07489528208971023,
"signal/format_reward/group_zero_std_frac": 0.7,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.020255533792078496,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.020255533792078496,
"signal/mean_confidence_reward/centered_abs_mean": 0.0620111808180809,
"signal/mean_confidence_reward/group_std_mean": 0.08973864912986755,
"signal/mean_confidence_reward/group_zero_std_frac": 0.325000011920929,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.201117571436044e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.201117571436044e-07,
"step": 705
},
{
"calibration/aurc": 0.05065273865740427,
"calibration/batch_distribution_entropy": 0.42453929252108147,
"calibration/batch_entropy_100bins": 0.22995059355568942,
"calibration/batch_entropy_10bins": 0.42453929252108147,
"calibration/batch_entropy_50bins": 0.2706941181643254,
"calibration/batch_uniqueness": -0.09111012720397868,
"calibration/confidence_entropy": 0.4049054598791278,
"calibration/coverage@0%": 0.0005390835579514825,
"calibration/coverage@1%": 0.3773087855908114,
"calibration/coverage@10%": 0.8881542699724518,
"calibration/coverage@15%": 0.8881542699724518,
"calibration/coverage@20%": 0.9079889807162533,
"calibration/coverage@25%": 0.9212121212121213,
"calibration/coverage@30%": 0.9592286501377411,
"calibration/coverage@5%": 0.7564973755142573,
"calibration/distribution_entropy_10": 0.42453929252108147,
"calibration/distribution_entropy_100": 0.22995059355568942,
"calibration/ece": 0.12564958761058773,
"calibration/mean_confidence": 0.806028914134358,
"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": 0.033333333333333326,
"completions/max_length": 3931.6,
"completions/max_terminated_length": 3931.6,
"completions/mean_length": 1060.0455810546875,
"completions/mean_terminated_length": 1096.588037109375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 344.4,
"epoch": 1.7067307692307692,
"grad_norm": 0.0005156388506293297,
"learning_rate": 1.983173076923077e-06,
"loss": -0.0423,
"num_tokens": 1851793635.0,
"reward": 1.2908846139907837,
"reward_std": 0.17143570482730866,
"rewards/accuracy_reward": 0.7593749761581421,
"rewards/brier_reward": 0.8557994365692139,
"rewards/confidence_one_or_zero": 0.0004340277810115367,
"rewards/format_reward": 0.9665798544883728,
"rewards/mean_confidence_reward": 0.7478854060173035,
"sampling/batch_mean_priority_error": 0.016651180555555548,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.18055555555555555,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0024777584709227084,
"sampling/priority_kl": 0.030000355467200278,
"sampling/priority_scale": 0.7764478266471997,
"sampling/prob_entropy": 10.278940582275391,
"sampling/prob_max": 5.5878546845633537e-05,
"sampling/prob_min": 2.0488816153374502e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6991999864578247,
"sampling/prompt_draws_total": 50976.0,
"sampling/seen_fraction": 0.8972199916839599,
"sampling/unseen_fraction": 0.10278000831604003,
"signal/accuracy_reward/centered_abs_mean": 0.11018880158662796,
"signal/accuracy_reward/group_std_mean": 0.15256147384643554,
"signal/accuracy_reward/group_zero_std_frac": 0.5333333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05509440079331398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05509440079331398,
"signal/advantage_abs_mean": 0.11043548285961151,
"signal/advantage_pre_scale_abs_mean": 0.11043548285961151,
"signal/advantage_pre_scale_std": 0.23827863931655885,
"signal/advantage_std": 0.23827863931655885,
"signal/brier_reward/centered_abs_mean": 0.08768158704042435,
"signal/brier_reward/group_std_mean": 0.1298682451248169,
"signal/brier_reward/group_zero_std_frac": 0.2888888955116272,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.043840793520212175,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.043840793520212175,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0007758246618323028,
"signal/confidence_one_or_zero/group_std_mean": 0.001424409542232752,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 7.758245601507952e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 7.758245601507952e-09,
"signal/format_reward/centered_abs_mean": 0.05783962681889534,
"signal/format_reward/group_std_mean": 0.10528798550367355,
"signal/format_reward/group_zero_std_frac": 0.5888888835906982,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.02891981340944767,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.02891981340944767,
"signal/mean_confidence_reward/centered_abs_mean": 0.07172510176897048,
"signal/mean_confidence_reward/group_std_mean": 0.10591092556715012,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3000000059604645,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.172509754127532e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.172509754127532e-07,
"step": 710
},
{
"calibration/aurc": 0.12651208488571192,
"calibration/batch_distribution_entropy": 0.4406230705036201,
"calibration/batch_entropy_100bins": 0.24175972802918233,
"calibration/batch_entropy_10bins": 0.4406230705036201,
"calibration/batch_entropy_50bins": 0.2845956401963263,
"calibration/batch_uniqueness": -0.04112757521393523,
"calibration/confidence_entropy": 0.4019345807333458,
"calibration/coverage@0%": 0.0032071301247771833,
"calibration/coverage@1%": 0.0032071301247771833,
"calibration/coverage@10%": 0.5005391301827457,
"calibration/coverage@15%": 0.6837769353175447,
"calibration/coverage@20%": 0.7536103363758805,
"calibration/coverage@25%": 0.8005333333333333,
"calibration/coverage@30%": 0.9648,
"calibration/coverage@5%": 0.25429476399576834,
"calibration/distribution_entropy_10": 0.4406230705036201,
"calibration/distribution_entropy_100": 0.24175972802918233,
"calibration/ece": 0.11199074373692071,
"calibration/mean_confidence": 0.7841360414479299,
"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": 0.026215277777777768,
"completions/max_length": 3809.6,
"completions/max_terminated_length": 3809.6,
"completions/mean_length": 1087.0016845703126,
"completions/mean_terminated_length": 1116.36162109375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 349.6,
"epoch": 1.71875,
"grad_norm": 0.0004386329383123666,
"learning_rate": 1.953125e-06,
"loss": -0.0338,
"num_tokens": 1867390998.0,
"reward": 1.2846812963485719,
"reward_std": 0.16560640335083007,
"rewards/accuracy_reward": 0.7434027791023254,
"rewards/brier_reward": 0.8521600246429444,
"rewards/confidence_one_or_zero": 0.0011284722480922938,
"rewards/format_reward": 0.9737847208976745,
"rewards/mean_confidence_reward": 0.7552181720733643,
"sampling/batch_mean_priority_error": 0.021241319444444427,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.17222222222222222,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.002497523045167327,
"sampling/priority_kl": 0.029999979585409165,
"sampling/priority_scale": 0.7792064726585523,
"sampling/prob_entropy": 10.278952026367188,
"sampling/prob_max": 5.612889508483931e-05,
"sampling/prob_min": 2.0507384033408015e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.7111999988555908,
"sampling/prompt_draws_total": 51336.0,
"sampling/seen_fraction": 0.8995266675949096,
"sampling/unseen_fraction": 0.10047333240509033,
"signal/accuracy_reward/centered_abs_mean": 0.11961805522441864,
"signal/accuracy_reward/group_std_mean": 0.1661587804555893,
"signal/accuracy_reward/group_zero_std_frac": 0.4888888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05980902761220932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05980902761220932,
"signal/advantage_abs_mean": 0.10705532133579254,
"signal/advantage_pre_scale_abs_mean": 0.10705532133579254,
"signal/advantage_pre_scale_std": 0.23213987052440643,
"signal/advantage_std": 0.23213987052440643,
"signal/brier_reward/centered_abs_mean": 0.08633466958999633,
"signal/brier_reward/group_std_mean": 0.1285327583551407,
"signal/brier_reward/group_zero_std_frac": 0.272222226858139,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.043167334794998166,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.043167334794998166,
"signal/confidence_one_or_zero/centered_abs_mean": 0.002099609375,
"signal/confidence_one_or_zero/group_std_mean": 0.004754898836836219,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9777777671813965,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.0996092331415638e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.0996092331415638e-08,
"signal/format_reward/centered_abs_mean": 0.04519314244389534,
"signal/format_reward/group_std_mean": 0.08304037302732467,
"signal/format_reward/group_zero_std_frac": 0.669444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.02259657122194767,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.02259657122194767,
"signal/mean_confidence_reward/centered_abs_mean": 0.06462962925434113,
"signal/mean_confidence_reward/group_std_mean": 0.09531620144844055,
"signal/mean_confidence_reward/group_zero_std_frac": 0.29722222685813904,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.462962801379035e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.462962801379035e-07,
"step": 715
},
{
"calibration/aurc": 0.12515534149725946,
"calibration/batch_distribution_entropy": 0.6140921867201559,
"calibration/batch_entropy_100bins": 0.33031808133393403,
"calibration/batch_entropy_10bins": 0.6140921867201559,
"calibration/batch_entropy_50bins": 0.38884510084452834,
"calibration/batch_uniqueness": 0.3432815524821929,
"calibration/confidence_entropy": 0.45580034763085564,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.4892404793508831,
"calibration/coverage@15%": 0.7023772313197574,
"calibration/coverage@20%": 0.776623220299957,
"calibration/coverage@25%": 0.8490904818341354,
"calibration/coverage@30%": 0.9391746551321021,
"calibration/coverage@5%": 0.3691450261025064,
"calibration/distribution_entropy_10": 0.6140921867201559,
"calibration/distribution_entropy_100": 0.33031808133393403,
"calibration/ece": 0.09690335220922594,
"calibration/mean_confidence": 0.7311622409754512,
"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": 0.021875,
"completions/max_length": 4014.8,
"completions/max_terminated_length": 4014.8,
"completions/mean_length": 1126.20751953125,
"completions/mean_terminated_length": 1151.4857177734375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 335.2,
"epoch": 1.7307692307692308,
"grad_norm": 0.00038971329922787845,
"learning_rate": 1.9230769230769234e-06,
"loss": -0.0253,
"num_tokens": 1883474380.0,
"reward": 1.2772084712982177,
"reward_std": 0.16071378737688063,
"rewards/accuracy_reward": 0.7256076335906982,
"rewards/brier_reward": 0.8507566690444947,
"rewards/confidence_one_or_zero": 0.0003472222248092294,
"rewards/format_reward": 0.9780381917953491,
"rewards/mean_confidence_reward": 0.7236449599266053,
"sampling/batch_mean_priority_error": 0.02141319444444443,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.16666666666666669,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0025197225622832775,
"sampling/priority_kl": 0.02999928630888462,
"sampling/priority_scale": 0.7810052573448047,
"sampling/prob_entropy": 10.278952980041504,
"sampling/prob_max": 5.633040564134717e-05,
"sampling/prob_min": 2.053321950370446e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.723200011253357,
"sampling/prompt_draws_total": 51696.0,
"sampling/seen_fraction": 0.9013800024986267,
"sampling/unseen_fraction": 0.09861999750137329,
"signal/accuracy_reward/centered_abs_mean": 0.12566731870174408,
"signal/accuracy_reward/group_std_mean": 0.17436597347259522,
"signal/accuracy_reward/group_zero_std_frac": 0.4694444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06283365935087204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06283365935087204,
"signal/advantage_abs_mean": 0.10567668825387955,
"signal/advantage_pre_scale_abs_mean": 0.10567668825387955,
"signal/advantage_pre_scale_std": 0.21769486963748932,
"signal/advantage_std": 0.21769486963748932,
"signal/brier_reward/centered_abs_mean": 0.08949927836656571,
"signal/brier_reward/group_std_mean": 0.1280209854245186,
"signal/brier_reward/group_zero_std_frac": 0.24722222089767457,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.044749639183282855,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.044749639183282855,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0006618923507630825,
"signal/confidence_one_or_zero/group_std_mean": 0.0016652445774525404,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 6.618923364953844e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 6.618923364953844e-09,
"signal/format_reward/centered_abs_mean": 0.03660481795668602,
"signal/format_reward/group_std_mean": 0.06963877379894257,
"signal/format_reward/group_zero_std_frac": 0.7111111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01830240897834301,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01830240897834301,
"signal/mean_confidence_reward/centered_abs_mean": 0.06871153563261032,
"signal/mean_confidence_reward/group_std_mean": 0.09722689241170883,
"signal/mean_confidence_reward/group_zero_std_frac": 0.26388889253139497,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.871153459542256e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.871153459542256e-07,
"step": 720
},
{
"calibration/aurc": 0.03142397230761359,
"calibration/batch_distribution_entropy": 0.6257232630459896,
"calibration/batch_entropy_100bins": 0.3396185513087483,
"calibration/batch_entropy_10bins": 0.6257232630459896,
"calibration/batch_entropy_50bins": 0.39979346361853624,
"calibration/batch_uniqueness": 0.36902582625996616,
"calibration/confidence_entropy": 0.45576618081231,
"calibration/coverage@0%": 0.2551352944478752,
"calibration/coverage@1%": 0.554943884024782,
"calibration/coverage@10%": 0.909504708137898,
"calibration/coverage@15%": 0.9487870619946092,
"calibration/coverage@20%": 0.9681940700808624,
"calibration/coverage@25%": 0.984366576819407,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.8229580442487784,
"calibration/distribution_entropy_10": 0.6257232630459896,
"calibration/distribution_entropy_100": 0.3396185513087483,
"calibration/ece": 0.1760231815956147,
"calibration/mean_confidence": 0.7242368288686128,
"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": 0.018055555555555557,
"completions/max_length": 3828.0,
"completions/max_terminated_length": 3828.0,
"completions/mean_length": 1194.933251953125,
"completions/mean_terminated_length": 1217.0438232421875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 366.2,
"epoch": 1.7427884615384617,
"grad_norm": 0.0004138158110436052,
"learning_rate": 1.8930288461538463e-06,
"loss": -0.021,
"num_tokens": 1900354091.0,
"reward": 1.317106509208679,
"reward_std": 0.14283051192760468,
"rewards/accuracy_reward": 0.7918402910232544,
"rewards/brier_reward": 0.8604142665863037,
"rewards/confidence_one_or_zero": 0.000260416668606922,
"rewards/format_reward": 0.9819444417953491,
"rewards/mean_confidence_reward": 0.6967925429344177,
"sampling/batch_mean_priority_error": 0.02047395833333332,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.15555555555555553,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.002539826137945056,
"sampling/priority_kl": 0.029999739304184914,
"sampling/priority_scale": 0.7832678138511255,
"sampling/prob_entropy": 10.278940773010254,
"sampling/prob_max": 5.655627537635155e-05,
"sampling/prob_min": 2.0555373703246004e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.735199999809265,
"sampling/prompt_draws_total": 52056.0,
"sampling/seen_fraction": 0.9033733367919922,
"sampling/unseen_fraction": 0.09662666320800781,
"signal/accuracy_reward/centered_abs_mean": 0.11361762136220932,
"signal/accuracy_reward/group_std_mean": 0.16370243728160858,
"signal/accuracy_reward/group_zero_std_frac": 0.47500001192092894,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05680881068110466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05680881068110466,
"signal/advantage_abs_mean": 0.08912541419267654,
"signal/advantage_pre_scale_abs_mean": 0.08912541419267654,
"signal/advantage_pre_scale_std": 0.19675522148609162,
"signal/advantage_std": 0.19675522148609162,
"signal/brier_reward/centered_abs_mean": 0.07736555710434914,
"signal/brier_reward/group_std_mean": 0.11556062549352646,
"signal/brier_reward/group_zero_std_frac": 0.19444444626569748,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03868277855217457,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03868277855217457,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0004937065881676972,
"signal/confidence_one_or_zero/group_std_mean": 0.001174198230728507,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.9370658672387435e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.9370658672387435e-09,
"signal/format_reward/centered_abs_mean": 0.03257378488779068,
"signal/format_reward/group_std_mean": 0.06698863953351974,
"signal/format_reward/group_zero_std_frac": 0.7083333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01628689244389534,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01628689244389534,
"signal/mean_confidence_reward/centered_abs_mean": 0.06848578304052352,
"signal/mean_confidence_reward/group_std_mean": 0.09661134332418442,
"signal/mean_confidence_reward/group_zero_std_frac": 0.20277778059244156,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.848578095741686e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.848578095741686e-07,
"step": 725
},
{
"calibration/aurc": 0.14456499448653962,
"calibration/batch_distribution_entropy": 0.6292655368791673,
"calibration/batch_entropy_100bins": 0.347850189209286,
"calibration/batch_entropy_10bins": 0.6292655368791673,
"calibration/batch_entropy_50bins": 0.40948361456826393,
"calibration/batch_uniqueness": 0.4494320384125043,
"calibration/confidence_entropy": 0.47106112366511493,
"calibration/coverage@0%": 0.1103448275862069,
"calibration/coverage@1%": 0.1103448275862069,
"calibration/coverage@10%": 0.47038486034027843,
"calibration/coverage@15%": 0.6424399455239072,
"calibration/coverage@20%": 0.7320380864849364,
"calibration/coverage@25%": 0.7943072157725353,
"calibration/coverage@30%": 0.9107367566470469,
"calibration/coverage@5%": 0.19433942456415387,
"calibration/distribution_entropy_10": 0.6292655368791673,
"calibration/distribution_entropy_100": 0.347850189209286,
"calibration/ece": 0.11105866402149867,
"calibration/mean_confidence": 0.6971300219725503,
"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": 0.015538194444444464,
"completions/max_length": 3970.4,
"completions/max_terminated_length": 3970.4,
"completions/mean_length": 1178.22333984375,
"completions/mean_terminated_length": 1196.9110595703125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 353.8,
"epoch": 1.7548076923076923,
"grad_norm": 0.0003167448448948562,
"learning_rate": 1.8629807692307695e-06,
"loss": -0.0197,
"num_tokens": 1917059544.0,
"reward": 1.2899388551712037,
"reward_std": 0.1306004211306572,
"rewards/accuracy_reward": 0.7422743201255798,
"rewards/brier_reward": 0.8531275868415833,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9844618201255798,
"rewards/mean_confidence_reward": 0.6970572829246521,
"sampling/batch_mean_priority_error": 0.018291666666666647,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.17222222222222222,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0025618880987167358,
"sampling/priority_kl": 0.029999687150120736,
"sampling/priority_scale": 0.7856972634559497,
"sampling/prob_entropy": 10.27893886566162,
"sampling/prob_max": 5.6790837697917594e-05,
"sampling/prob_min": 2.0575926464516668e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.7472000122070312,
"sampling/prompt_draws_total": 52416.0,
"sampling/seen_fraction": 0.9053866744041443,
"sampling/unseen_fraction": 0.09461332559585571,
"signal/accuracy_reward/centered_abs_mean": 0.10812174528837204,
"signal/accuracy_reward/group_std_mean": 0.1530183345079422,
"signal/accuracy_reward/group_zero_std_frac": 0.522222238779068,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05406087264418602,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05406087264418602,
"signal/advantage_abs_mean": 0.0842123880982399,
"signal/advantage_pre_scale_abs_mean": 0.0842123880982399,
"signal/advantage_pre_scale_std": 0.1876286417245865,
"signal/advantage_std": 0.1876286417245865,
"signal/brier_reward/centered_abs_mean": 0.0749520257115364,
"signal/brier_reward/group_std_mean": 0.10771744549274445,
"signal/brier_reward/group_zero_std_frac": 0.18611111640930175,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0374760128557682,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0374760128557682,
"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": 0.02612304650247097,
"signal/format_reward/group_std_mean": 0.051665914803743364,
"signal/format_reward/group_zero_std_frac": 0.7777777791023255,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.013061523251235486,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.013061523251235486,
"signal/mean_confidence_reward/centered_abs_mean": 0.0662548802793026,
"signal/mean_confidence_reward/group_std_mean": 0.09253504127264023,
"signal/mean_confidence_reward/group_zero_std_frac": 0.18888889253139496,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.625487912970129e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.625487912970129e-07,
"step": 730
},
{
"calibration/aurc": 0.1911213039384451,
"calibration/batch_distribution_entropy": 0.5815784623872695,
"calibration/batch_entropy_100bins": 0.3156432238254143,
"calibration/batch_entropy_10bins": 0.5815784623872695,
"calibration/batch_entropy_50bins": 0.3715700960227042,
"calibration/batch_uniqueness": 0.2773596382466226,
"calibration/confidence_entropy": 0.4474254359773086,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.1801047120418848,
"calibration/coverage@15%": 0.4702755641247623,
"calibration/coverage@20%": 0.49230769230769234,
"calibration/coverage@25%": 0.737332625994695,
"calibration/coverage@30%": 0.9899204244031831,
"calibration/coverage@5%": 0.1293193717277487,
"calibration/distribution_entropy_10": 0.5815784623872695,
"calibration/distribution_entropy_100": 0.3156432238254143,
"calibration/ece": 0.14364294131946279,
"calibration/mean_confidence": 0.7393319764923701,
"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": 0.014236111111111139,
"completions/max_length": 3983.0,
"completions/max_terminated_length": 3983.0,
"completions/mean_length": 1152.609423828125,
"completions/mean_terminated_length": 1169.4090576171875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 397.0,
"epoch": 1.7668269230769231,
"grad_norm": 0.0003992510319221765,
"learning_rate": 1.8329326923076924e-06,
"loss": -0.0169,
"num_tokens": 1933420836.0,
"reward": 1.2813831567764282,
"reward_std": 0.13340601325035095,
"rewards/accuracy_reward": 0.7334201335906982,
"rewards/brier_reward": 0.8441761016845704,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.98515625,
"rewards/mean_confidence_reward": 0.6882100462913513,
"sampling/batch_mean_priority_error": 0.02233506944444443,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1388888888888889,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.002584920870140195,
"sampling/priority_kl": 0.030000413581728936,
"sampling/priority_scale": 0.7875803768867626,
"sampling/prob_entropy": 10.278951263427734,
"sampling/prob_max": 5.6999971275217834e-05,
"sampling/prob_min": 2.0601512733264827e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.7592000246047974,
"sampling/prompt_draws_total": 52776.0,
"sampling/seen_fraction": 0.907146668434143,
"sampling/unseen_fraction": 0.09285333156585693,
"signal/accuracy_reward/centered_abs_mean": 0.11967773735523224,
"signal/accuracy_reward/group_std_mean": 0.16079054325819014,
"signal/accuracy_reward/group_zero_std_frac": 0.522222226858139,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05983886867761612,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05983886867761612,
"signal/advantage_abs_mean": 0.08656046390533448,
"signal/advantage_pre_scale_abs_mean": 0.08656046390533448,
"signal/advantage_pre_scale_std": 0.18703484237194062,
"signal/advantage_std": 0.18703484237194062,
"signal/brier_reward/centered_abs_mean": 0.07910026758909225,
"signal/brier_reward/group_std_mean": 0.11268624216318131,
"signal/brier_reward/group_zero_std_frac": 0.17500000298023224,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.039550133794546125,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.039550133794546125,
"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": 3.2552080142522754e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.2552080142522754e-09,
"signal/format_reward/centered_abs_mean": 0.02683919295668602,
"signal/format_reward/group_std_mean": 0.05439011082053184,
"signal/format_reward/group_zero_std_frac": 0.7666666865348816,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01341959647834301,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01341959647834301,
"signal/mean_confidence_reward/centered_abs_mean": 0.06722780838608741,
"signal/mean_confidence_reward/group_std_mean": 0.09363017976284027,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1833333343267441,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.7227805402581e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.7227805402581e-07,
"step": 735
},
{
"calibration/aurc": 0.07135814786399183,
"calibration/batch_distribution_entropy": 0.5937620416887921,
"calibration/batch_entropy_100bins": 0.32064189271944316,
"calibration/batch_entropy_10bins": 0.5937620416887921,
"calibration/batch_entropy_50bins": 0.3774544481669698,
"calibration/batch_uniqueness": 0.32722568282924336,
"calibration/confidence_entropy": 0.44936877206762604,
"calibration/coverage@0%": 0.2635238016710642,
"calibration/coverage@1%": 0.35501313366391496,
"calibration/coverage@10%": 0.7962148920012707,
"calibration/coverage@15%": 0.8760531857395459,
"calibration/coverage@20%": 0.8953349265772423,
"calibration/coverage@25%": 0.9400240368011715,
"calibration/coverage@30%": 0.9630606860158311,
"calibration/coverage@5%": 0.5747872021644158,
"calibration/distribution_entropy_10": 0.5937620416887921,
"calibration/distribution_entropy_100": 0.32064189271944316,
"calibration/ece": 0.1270965697419163,
"calibration/mean_confidence": 0.7460525301778327,
"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": 0.010329861111111116,
"completions/max_length": 3854.0,
"completions/max_terminated_length": 3854.0,
"completions/mean_length": 1144.8279541015625,
"completions/mean_terminated_length": 1156.9118408203126,
"completions/min_length": 0.0,
"completions/min_terminated_length": 350.6,
"epoch": 1.7788461538461537,
"grad_norm": 0.00032714667031541467,
"learning_rate": 1.8028846153846156e-06,
"loss": -0.0117,
"num_tokens": 1949681606.0,
"reward": 1.2939432859420776,
"reward_std": 0.12169390916824341,
"rewards/accuracy_reward": 0.7418402910232544,
"rewards/brier_reward": 0.8563619971275329,
"rewards/confidence_one_or_zero": 0.0005208333546761424,
"rewards/format_reward": 0.9896701335906982,
"rewards/mean_confidence_reward": 0.7095920205116272,
"sampling/batch_mean_priority_error": 0.019717013888888878,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.16111111111111112,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0026072211097925902,
"sampling/priority_kl": 0.030000557750463487,
"sampling/priority_scale": 0.7897889913292602,
"sampling/prob_entropy": 10.278955268859864,
"sampling/prob_max": 5.7226319040637466e-05,
"sampling/prob_min": 2.0624370517907664e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.7711999893188477,
"sampling/prompt_draws_total": 53136.0,
"sampling/seen_fraction": 0.9089933156967163,
"sampling/unseen_fraction": 0.0910066843032837,
"signal/accuracy_reward/centered_abs_mean": 0.11880425363779068,
"signal/accuracy_reward/group_std_mean": 0.15754524171352385,
"signal/accuracy_reward/group_zero_std_frac": 0.5472222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05940212681889534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05940212681889534,
"signal/advantage_abs_mean": 0.08523241132497787,
"signal/advantage_pre_scale_abs_mean": 0.08523241132497787,
"signal/advantage_pre_scale_std": 0.17800681591033934,
"signal/advantage_std": 0.17800681591033934,
"signal/brier_reward/centered_abs_mean": 0.07595813944935799,
"signal/brier_reward/group_std_mean": 0.10386582911014557,
"signal/brier_reward/group_zero_std_frac": 0.19444445073604583,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.037979069724678995,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.037979069724678995,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0009006076375953853,
"signal/confidence_one_or_zero/group_std_mean": 0.001515774242579937,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 9.006076595596824e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 9.006076595596824e-09,
"signal/format_reward/centered_abs_mean": 0.017822265625,
"signal/format_reward/group_std_mean": 0.03541161455214024,
"signal/format_reward/group_zero_std_frac": 0.8444444417953492,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0089111328125,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0089111328125,
"signal/mean_confidence_reward/centered_abs_mean": 0.06562201455235481,
"signal/mean_confidence_reward/group_std_mean": 0.08824090212583542,
"signal/mean_confidence_reward/group_zero_std_frac": 0.20833333432674409,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.562201065207773e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.562201065207773e-07,
"step": 740
},
{
"calibration/aurc": 0.044976670240617245,
"calibration/batch_distribution_entropy": 0.4711153755218615,
"calibration/batch_entropy_100bins": 0.25180322817921036,
"calibration/batch_entropy_10bins": 0.4711153755218615,
"calibration/batch_entropy_50bins": 0.2964186860704684,
"calibration/batch_uniqueness": 0.039686466266251816,
"calibration/confidence_entropy": 0.41058957351063724,
"calibration/coverage@0%": 0.31086406657963445,
"calibration/coverage@1%": 0.3129528394255875,
"calibration/coverage@10%": 0.922811956048738,
"calibration/coverage@15%": 0.9473006418624891,
"calibration/coverage@20%": 0.9738903394255874,
"calibration/coverage@25%": 1.0,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.6403679830287206,
"calibration/distribution_entropy_10": 0.4711153755218615,
"calibration/distribution_entropy_100": 0.25180322817921036,
"calibration/ece": 0.10536104765013041,
"calibration/mean_confidence": 0.7969752230200176,
"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": 0.00546875,
"completions/max_length": 4029.2,
"completions/max_terminated_length": 4029.2,
"completions/mean_length": 1128.099853515625,
"completions/mean_terminated_length": 1134.4380615234375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 394.2,
"epoch": 1.7908653846153846,
"grad_norm": 0.00034849203075282276,
"learning_rate": 1.7728365384615387e-06,
"loss": -0.005,
"num_tokens": 1965781412.0,
"reward": 1.3357252836227418,
"reward_std": 0.0930047869682312,
"rewards/accuracy_reward": 0.7889756798744202,
"rewards/brier_reward": 0.887928593158722,
"rewards/confidence_one_or_zero": 0.0006944444554392249,
"rewards/format_reward": 0.9945312380790711,
"rewards/mean_confidence_reward": 0.7552994608879089,
"sampling/batch_mean_priority_error": 0.014748263888888868,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.175,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0026240932755172254,
"sampling/priority_kl": 0.02999989651143551,
"sampling/priority_scale": 0.7923730551963672,
"sampling/prob_entropy": 10.27895565032959,
"sampling/prob_max": 5.7472297339700165e-05,
"sampling/prob_min": 2.0643966854549946e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.7831999778747558,
"sampling/prompt_draws_total": 53496.0,
"sampling/seen_fraction": 0.9109466791152954,
"sampling/unseen_fraction": 0.08905332088470459,
"signal/accuracy_reward/centered_abs_mean": 0.08330620601773261,
"signal/accuracy_reward/group_std_mean": 0.11775979846715927,
"signal/accuracy_reward/group_zero_std_frac": 0.6305555701255798,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04165310300886631,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04165310300886631,
"signal/advantage_abs_mean": 0.06172395497560501,
"signal/advantage_pre_scale_abs_mean": 0.06172395497560501,
"signal/advantage_pre_scale_std": 0.1456386312842369,
"signal/advantage_std": 0.1456386312842369,
"signal/brier_reward/centered_abs_mean": 0.05597181022167206,
"signal/brier_reward/group_std_mean": 0.08043636679649353,
"signal/brier_reward/group_zero_std_frac": 0.30555556416511537,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02798590511083603,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02798590511083603,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0013454860891215503,
"signal/confidence_one_or_zero/group_std_mean": 0.003928370773792267,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9777777671813965,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.3454859981720801e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.3454859981720801e-08,
"signal/format_reward/centered_abs_mean": 0.009792751679196954,
"signal/format_reward/group_std_mean": 0.02300018724054098,
"signal/format_reward/group_zero_std_frac": 0.8861111164093017,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004896375839598477,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004896375839598477,
"signal/mean_confidence_reward/centered_abs_mean": 0.05380534529685974,
"signal/mean_confidence_reward/group_std_mean": 0.07323706895112991,
"signal/mean_confidence_reward/group_zero_std_frac": 0.31111111044883727,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.38053438958741e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.38053438958741e-07,
"step": 745
},
{
"calibration/aurc": 0.11353951064295462,
"calibration/batch_distribution_entropy": 0.5589911152512703,
"calibration/batch_entropy_100bins": 0.3068109774965393,
"calibration/batch_entropy_10bins": 0.5589911152512703,
"calibration/batch_entropy_50bins": 0.3611729185489008,
"calibration/batch_uniqueness": 0.21741473871920394,
"calibration/confidence_entropy": 0.4441730678233955,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.16335078534031414,
"calibration/coverage@10%": 0.5125457796204305,
"calibration/coverage@15%": 0.7319539847079646,
"calibration/coverage@20%": 0.811123733562073,
"calibration/coverage@25%": 0.8920657185711904,
"calibration/coverage@30%": 0.9080231653797011,
"calibration/coverage@5%": 0.3494851216817216,
"calibration/distribution_entropy_10": 0.5589911152512703,
"calibration/distribution_entropy_100": 0.3068109774965393,
"calibration/ece": 0.10185885955664371,
"calibration/mean_confidence": 0.7363732055069778,
"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": 0.007204861111111094,
"completions/max_length": 3862.2,
"completions/max_terminated_length": 3862.2,
"completions/mean_length": 1151.9552978515626,
"completions/mean_terminated_length": 1160.3269287109374,
"completions/min_length": 0.0,
"completions/min_terminated_length": 391.4,
"epoch": 1.8028846153846154,
"grad_norm": 0.0003831425274256617,
"learning_rate": 1.7427884615384616e-06,
"loss": -0.008,
"num_tokens": 1982158337.0,
"reward": 1.3099550485610962,
"reward_std": 0.11404931545257568,
"rewards/accuracy_reward": 0.7524305582046509,
"rewards/brier_reward": 0.87466961145401,
"rewards/confidence_one_or_zero": 0.00034722223062999547,
"rewards/format_reward": 0.9927951455116272,
"rewards/mean_confidence_reward": 0.7468029499053955,
"sampling/batch_mean_priority_error": 0.013510416666666653,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.17777777777777776,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.002636962477117777,
"sampling/priority_kl": 0.029998420923948287,
"sampling/priority_scale": 0.7955237090354785,
"sampling/prob_entropy": 10.278936958312988,
"sampling/prob_max": 5.7744530931813645e-05,
"sampling/prob_min": 2.0657485219999216e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.795199990272522,
"sampling/prompt_draws_total": 53856.0,
"sampling/seen_fraction": 0.9130800008773804,
"sampling/unseen_fraction": 0.08691999912261963,
"signal/accuracy_reward/centered_abs_mean": 0.1026584193110466,
"signal/accuracy_reward/group_std_mean": 0.14243891835212708,
"signal/accuracy_reward/group_zero_std_frac": 0.569444453716278,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0513292096555233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0513292096555233,
"signal/advantage_abs_mean": 0.07915246114134789,
"signal/advantage_pre_scale_abs_mean": 0.07915246114134789,
"signal/advantage_pre_scale_std": 0.16908468604087828,
"signal/advantage_std": 0.16908468604087828,
"signal/brier_reward/centered_abs_mean": 0.07080509662628173,
"signal/brier_reward/group_std_mean": 0.09794733822345733,
"signal/brier_reward/group_zero_std_frac": 0.27777778208255766,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.035402548313140866,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.035402548313140866,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0006727430387400091,
"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": 0.012559678684920073,
"signal/format_reward/group_std_mean": 0.02510856781154871,
"signal/format_reward/group_zero_std_frac": 0.8916666626930236,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.006279839342460037,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.006279839342460037,
"signal/mean_confidence_reward/centered_abs_mean": 0.060461973398923875,
"signal/mean_confidence_reward/group_std_mean": 0.08233949095010758,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2916666716337204,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.04619685873331e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.04619685873331e-07,
"step": 750
},
{
"epoch": 1.8028846153846154,
"eval_calibration/aurc": 0.12918669119756246,
"eval_calibration/batch_distribution_entropy": 0.570947120401169,
"eval_calibration/batch_entropy_100bins": 0.31176983812988257,
"eval_calibration/batch_entropy_10bins": 0.570947120401169,
"eval_calibration/batch_entropy_50bins": 0.36701040905277965,
"eval_calibration/batch_uniqueness": 0.22771267764757755,
"eval_calibration/confidence_entropy": 0.4321299611272379,
"eval_calibration/coverage@0%": 0.0008748906386701663,
"eval_calibration/coverage@1%": 0.0008748906386701663,
"eval_calibration/coverage@10%": 0.600174978127734,
"eval_calibration/coverage@15%": 0.7961504811898513,
"eval_calibration/coverage@20%": 0.8958880139982502,
"eval_calibration/coverage@25%": 0.9667541557305337,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0008748906386701663,
"eval_calibration/distribution_entropy_10": 0.570947120401169,
"eval_calibration/distribution_entropy_100": 0.31176983812988257,
"eval_calibration/ece": 0.023053368328958688,
"eval_calibration/mean_confidence": 0.7554243219597551,
"eval_calibration/unique_confidence_per_question": 0.008680555555555556,
"eval_calibration/unique_confidences": 10,
"eval_completions/clipped_ratio": 0.0078125,
"eval_completions/max_length": 3036.3333333333335,
"eval_completions/max_terminated_length": 3036.3333333333335,
"eval_completions/mean_length": 1114.5510864257812,
"eval_completions/mean_terminated_length": 1123.3021850585938,
"eval_completions/min_length": 98.16666666666667,
"eval_completions/min_terminated_length": 455.0,
"eval_loss": 0.0,
"eval_num_tokens": 1982158337.0,
"eval_reward": 1.2921201586723328,
"eval_reward_std": 0.3189353197813034,
"eval_rewards/accuracy_reward": 0.7343750099341074,
"eval_rewards/brier_reward": 0.8576627671718597,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9921875,
"eval_rewards/mean_confidence_reward": 0.7495225568612417,
"eval_runtime": 203.2359,
"eval_samples_per_second": 4.92,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3766276041666667,
"eval_signal/accuracy_reward/group_std_mean": 0.43860721588134766,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.18831380208333334,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.18831380208333334,
"eval_signal/advantage_abs_mean": 0.26045594612757367,
"eval_signal/advantage_pre_scale_abs_mean": 0.26045594612757367,
"eval_signal/advantage_pre_scale_std": 0.31735892593860626,
"eval_signal/advantage_std": 0.31735892593860626,
"eval_signal/brier_reward/centered_abs_mean": 0.1701806883017222,
"eval_signal/brier_reward/group_std_mean": 0.23481983691453934,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0850903441508611,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.0850903441508611,
"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": 0.015136718439559141,
"eval_signal/format_reward/group_std_mean": 0.044194173688689865,
"eval_signal/format_reward/group_zero_std_frac": 0.750000019868215,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.007568359219779571,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.007568359219779571,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.18361271917819977,
"eval_signal/mean_confidence_reward/group_std_mean": 0.23008321225643158,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.8361271637938141e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.8361271637938141e-06,
"eval_steps_per_second": 0.03,
"step": 750
},
{
"epoch": 1.8028846153846154,
"step": 750,
"train_probe_calibration/aurc": 0.09515691675342279,
"train_probe_calibration/batch_distribution_entropy": 0.5559239476940845,
"train_probe_calibration/batch_entropy_100bins": 0.30124399218193226,
"train_probe_calibration/batch_entropy_10bins": 0.5559239476940845,
"train_probe_calibration/batch_entropy_50bins": 0.3546195535096131,
"train_probe_calibration/batch_uniqueness": 0.18609991806998472,
"train_probe_calibration/confidence_entropy": 0.42859308955033887,
"train_probe_calibration/coverage@0%": 0.0008726003490401396,
"train_probe_calibration/coverage@1%": 0.0008726003490401396,
"train_probe_calibration/coverage@10%": 0.7705061082024433,
"train_probe_calibration/coverage@15%": 0.8403141361256544,
"train_probe_calibration/coverage@20%": 0.9101221640488656,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0008726003490401396,
"train_probe_calibration/distribution_entropy_10": 0.5559239476940845,
"train_probe_calibration/distribution_entropy_100": 0.30124399218193226,
"train_probe_calibration/ece": 0.026527050610820127,
"train_probe_calibration/mean_confidence": 0.7626527050610821,
"train_probe_calibration/unique_confidence_per_question": 0.008680555555555556,
"train_probe_calibration/unique_confidences": 10,
"train_probe_completions/clipped_ratio": 0.007638888888888899,
"train_probe_completions/max_length": 3220.3333333333335,
"train_probe_completions/max_terminated_length": 3220.3333333333335,
"train_probe_completions/mean_length": 1125.8380330403645,
"train_probe_completions/mean_terminated_length": 1134.5219319661458,
"train_probe_completions/min_length": 124.33333333333333,
"train_probe_completions/min_terminated_length": 399.1666666666667,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1982158337.0,
"train_probe_reward": 1.3254329760869343,
"train_probe_reward_std": 0.2898659110069275,
"train_probe_rewards/accuracy_reward": 0.7777777711550394,
"train_probe_rewards/brier_reward": 0.8782812555631002,
"train_probe_rewards/confidence_one_or_zero": 0.0008680555814256271,
"train_probe_rewards/format_reward": 0.9947916666666666,
"train_probe_rewards/mean_confidence_reward": 0.7586805323759714,
"train_probe_runtime": 198.345,
"train_probe_samples_per_second": 5.042,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3351779480775197,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4128282864888509,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.16758897403875986,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.16758897403875986,
"train_probe_signal/advantage_abs_mean": 0.2279451290766398,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.2279451290766398,
"train_probe_signal/advantage_pre_scale_std": 0.2899184872706731,
"train_probe_signal/advantage_std": 0.2899184872706731,
"train_probe_signal/brier_reward/centered_abs_mean": 0.14683595051368079,
"train_probe_signal/brier_reward/group_std_mean": 0.2066178321838379,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07341797525684039,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07341797525684039,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0016818575871487458,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0049104637776811915,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222288449606,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151002e-08,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151002e-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": 0.8333333532015482,
"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": 0.17816295226415,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.2219566653172175,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.7816294833513287e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.7816294833513287e-06,
"train_probe_steps_per_second": 0.03
},
{
"calibration/aurc": 0.08235913041108442,
"calibration/batch_distribution_entropy": 0.5500182521390489,
"calibration/batch_entropy_100bins": 0.2986105969772843,
"calibration/batch_entropy_10bins": 0.5500182521390489,
"calibration/batch_entropy_50bins": 0.351519563282679,
"calibration/batch_uniqueness": 0.19279256086586521,
"calibration/confidence_entropy": 0.4438144496211079,
"calibration/coverage@0%": 0.11076115485564304,
"calibration/coverage@1%": 0.11076115485564304,
"calibration/coverage@10%": 0.7177205038646448,
"calibration/coverage@15%": 0.8222280028529407,
"calibration/coverage@20%": 0.926595744680851,
"calibration/coverage@25%": 0.9627659574468085,
"calibration/coverage@30%": 0.9627659574468085,
"calibration/coverage@5%": 0.3648423271930728,
"calibration/distribution_entropy_10": 0.5500182521390489,
"calibration/distribution_entropy_100": 0.2986105969772843,
"calibration/ece": 0.09704431335458163,
"calibration/mean_confidence": 0.7434555824273797,
"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": 0.00746527777777779,
"completions/max_length": 3774.0,
"completions/max_terminated_length": 3774.0,
"completions/mean_length": 1096.2972412109375,
"completions/mean_terminated_length": 1104.5652099609374,
"completions/min_length": 0.0,
"completions/min_terminated_length": 383.8,
"epoch": 1.8149038461538463,
"grad_norm": 0.000413256959291175,
"learning_rate": 1.7127403846153848e-06,
"loss": -0.0082,
"num_tokens": 1997896449.0,
"reward": 1.324518609046936,
"reward_std": 0.11593645066022873,
"rewards/accuracy_reward": 0.7758680582046509,
"rewards/brier_reward": 0.8807059168815613,
"rewards/confidence_one_or_zero": 0.0006944444496184588,
"rewards/format_reward": 0.9924479007720948,
"rewards/mean_confidence_reward": 0.7667656183242798,
"sampling/batch_mean_priority_error": 0.013126805555555541,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.14166666666666666,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0026503289118409157,
"sampling/priority_kl": 0.03000011146068573,
"sampling/priority_scale": 0.7983648717170582,
"sampling/prob_entropy": 10.278961563110352,
"sampling/prob_max": 5.800310827908106e-05,
"sampling/prob_min": 2.0673956169048324e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.8072000026702881,
"sampling/prompt_draws_total": 54216.0,
"sampling/seen_fraction": 0.9150466680526733,
"sampling/unseen_fraction": 0.08495333194732665,
"signal/accuracy_reward/centered_abs_mean": 0.10064018815755844,
"signal/accuracy_reward/group_std_mean": 0.13902317732572556,
"signal/accuracy_reward/group_zero_std_frac": 0.5777777791023254,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05032009407877922,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05032009407877922,
"signal/advantage_abs_mean": 0.07687721401453018,
"signal/advantage_pre_scale_abs_mean": 0.07687721401453018,
"signal/advantage_pre_scale_std": 0.1724386215209961,
"signal/advantage_std": 0.1724386215209961,
"signal/brier_reward/centered_abs_mean": 0.06619351655244828,
"signal/brier_reward/group_std_mean": 0.09618951678276062,
"signal/brier_reward/group_zero_std_frac": 0.3027777820825577,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03309675827622414,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03309675827622414,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0012803819496184587,
"signal/confidence_one_or_zero/group_std_mean": 0.002897548582404852,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.2803818094653252e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.2803818094653252e-08,
"signal/format_reward/centered_abs_mean": 0.013926866464316845,
"signal/format_reward/group_std_mean": 0.03264928236603737,
"signal/format_reward/group_zero_std_frac": 0.8416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0069634332321584225,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0069634332321584225,
"signal/mean_confidence_reward/centered_abs_mean": 0.054513295739889146,
"signal/mean_confidence_reward/group_std_mean": 0.07755149900913239,
"signal/mean_confidence_reward/group_zero_std_frac": 0.31666667461395265,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.451329343486577e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.451329343486577e-07,
"step": 755
},
{
"calibration/aurc": 0.15240919905327432,
"calibration/batch_distribution_entropy": 0.4942784512127184,
"calibration/batch_entropy_100bins": 0.26749138967220476,
"calibration/batch_entropy_10bins": 0.4942784512127184,
"calibration/batch_entropy_50bins": 0.3148865359476951,
"calibration/batch_uniqueness": 0.09131136876063979,
"calibration/confidence_entropy": 0.4264967519358572,
"calibration/coverage@0%": 0.19791259415198284,
"calibration/coverage@1%": 0.2005235602094241,
"calibration/coverage@10%": 0.4353681520038709,
"calibration/coverage@15%": 0.5631992935300282,
"calibration/coverage@20%": 0.6284375861118765,
"calibration/coverage@25%": 0.6861228988702123,
"calibration/coverage@30%": 0.7128244695508406,
"calibration/coverage@5%": 0.4264442674894352,
"calibration/distribution_entropy_10": 0.4942784512127184,
"calibration/distribution_entropy_100": 0.26749138967220476,
"calibration/ece": 0.1297135731642174,
"calibration/mean_confidence": 0.7767179014679333,
"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": 0.010156249999999978,
"completions/max_length": 3959.8,
"completions/max_terminated_length": 3959.8,
"completions/mean_length": 1118.2989501953125,
"completions/mean_terminated_length": 1129.805322265625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 319.8,
"epoch": 1.8269230769230769,
"grad_norm": 0.000439549854490906,
"learning_rate": 1.682692307692308e-06,
"loss": -0.0094,
"num_tokens": 2013854613.0,
"reward": 1.2982789754867554,
"reward_std": 0.1340230643749237,
"rewards/accuracy_reward": 0.7386284828186035,
"rewards/brier_reward": 0.868070662021637,
"rewards/confidence_one_or_zero": 0.0008680555678438395,
"rewards/format_reward": 0.9898437619209289,
"rewards/mean_confidence_reward": 0.7592884659767151,
"sampling/batch_mean_priority_error": 0.01165805555555555,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1388888888888889,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.002661566250026226,
"sampling/priority_kl": 0.030000675842165946,
"sampling/priority_scale": 0.8004713356727734,
"sampling/prob_entropy": 10.27895393371582,
"sampling/prob_max": 5.822636740049347e-05,
"sampling/prob_min": 2.069712099910248e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.8191999912261962,
"sampling/prompt_draws_total": 54576.0,
"sampling/seen_fraction": 0.916700005531311,
"sampling/unseen_fraction": 0.08329999446868896,
"signal/accuracy_reward/centered_abs_mean": 0.1171820729970932,
"signal/accuracy_reward/group_std_mean": 0.15496748685836792,
"signal/accuracy_reward/group_zero_std_frac": 0.5583333492279052,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0585910364985466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0585910364985466,
"signal/advantage_abs_mean": 0.09350767433643341,
"signal/advantage_pre_scale_abs_mean": 0.09350767433643341,
"signal/advantage_pre_scale_std": 0.19366811215877533,
"signal/advantage_std": 0.19366811215877533,
"signal/brier_reward/centered_abs_mean": 0.07623469531536102,
"signal/brier_reward/group_std_mean": 0.10736920237541199,
"signal/brier_reward/group_zero_std_frac": 0.31666667461395265,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03811734765768051,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03811734765768051,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0016493055154569447,
"signal/confidence_one_or_zero/group_std_mean": 0.004259948246181011,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9777777671813965,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6493054033617227e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6493054033617227e-08,
"signal/format_reward/centered_abs_mean": 0.01809353344142437,
"signal/format_reward/group_std_mean": 0.04034832753241062,
"signal/format_reward/group_zero_std_frac": 0.8083333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009046766720712186,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009046766720712186,
"signal/mean_confidence_reward/centered_abs_mean": 0.06380566284060478,
"signal/mean_confidence_reward/group_std_mean": 0.08879230618476867,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3388888895511627,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.38056599200354e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.38056599200354e-07,
"step": 760
},
{
"calibration/aurc": 0.14281510974146783,
"calibration/batch_distribution_entropy": 0.47071555151756134,
"calibration/batch_entropy_100bins": 0.2578550493311225,
"calibration/batch_entropy_10bins": 0.47071555151756134,
"calibration/batch_entropy_50bins": 0.30354279201285356,
"calibration/batch_uniqueness": 0.03149784874336764,
"calibration/confidence_entropy": 0.42275016873700616,
"calibration/coverage@0%": 0.0015665796344647518,
"calibration/coverage@1%": 0.0015665796344647518,
"calibration/coverage@10%": 0.4061340976251519,
"calibration/coverage@15%": 0.599683554376951,
"calibration/coverage@20%": 0.6860280201828867,
"calibration/coverage@25%": 0.9121581419441547,
"calibration/coverage@30%": 0.929390517923267,
"calibration/coverage@5%": 0.3643477282163753,
"calibration/distribution_entropy_10": 0.47071555151756134,
"calibration/distribution_entropy_100": 0.2578550493311225,
"calibration/ece": 0.10513159295486174,
"calibration/mean_confidence": 0.7598361444108246,
"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": 0.012760416666666673,
"completions/max_length": 3986.8,
"completions/max_terminated_length": 3986.8,
"completions/mean_length": 1143.3353515625,
"completions/mean_terminated_length": 1158.1183837890626,
"completions/min_length": 0.0,
"completions/min_terminated_length": 365.6,
"epoch": 1.8389423076923077,
"grad_norm": 0.00036109599750488997,
"learning_rate": 1.6526442307692309e-06,
"loss": -0.0147,
"num_tokens": 2030106252.0,
"reward": 1.3092787742614747,
"reward_std": 0.12436873316764832,
"rewards/accuracy_reward": 0.7537326335906982,
"rewards/brier_reward": 0.8776569604873657,
"rewards/confidence_one_or_zero": 0.0014756944845430553,
"rewards/format_reward": 0.9871527910232544,
"rewards/mean_confidence_reward": 0.7626976013183594,
"sampling/batch_mean_priority_error": 0.0149636111111111,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.17222222222222222,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0026781358756124973,
"sampling/priority_kl": 0.029999176412820815,
"sampling/priority_scale": 0.803554934146814,
"sampling/prob_entropy": 10.27893524169922,
"sampling/prob_max": 5.850200104760006e-05,
"sampling/prob_min": 2.0712401601485907e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.8312000036239624,
"sampling/prompt_draws_total": 54936.0,
"sampling/seen_fraction": 0.9186333417892456,
"sampling/unseen_fraction": 0.0813666582107544,
"signal/accuracy_reward/centered_abs_mean": 0.0984754778444767,
"signal/accuracy_reward/group_std_mean": 0.1380874663591385,
"signal/accuracy_reward/group_zero_std_frac": 0.575,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04923773892223835,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04923773892223835,
"signal/advantage_abs_mean": 0.08254304528236389,
"signal/advantage_pre_scale_abs_mean": 0.08254304528236389,
"signal/advantage_pre_scale_std": 0.1854351818561554,
"signal/advantage_std": 0.1854351818561554,
"signal/brier_reward/centered_abs_mean": 0.07168641835451126,
"signal/brier_reward/group_std_mean": 0.1029394656419754,
"signal/brier_reward/group_zero_std_frac": 0.325,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03584320917725563,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03584320917725563,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0027506509562954307,
"signal/confidence_one_or_zero/group_std_mean": 0.006121202511712909,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222089767456,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.7506509070462926e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.7506509070462926e-08,
"signal/format_reward/centered_abs_mean": 0.0229600690305233,
"signal/format_reward/group_std_mean": 0.0469165489077568,
"signal/format_reward/group_zero_std_frac": 0.7972222208976746,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01148003451526165,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01148003451526165,
"signal/mean_confidence_reward/centered_abs_mean": 0.060264953225851056,
"signal/mean_confidence_reward/group_std_mean": 0.08513574749231338,
"signal/mean_confidence_reward/group_zero_std_frac": 0.34722222089767457,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.026494929756154e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.026494929756154e-07,
"step": 765
},
{
"calibration/aurc": 0.0969113642472206,
"calibration/batch_distribution_entropy": 0.5355198771027292,
"calibration/batch_entropy_100bins": 0.2940913247107916,
"calibration/batch_entropy_10bins": 0.5355198771027292,
"calibration/batch_entropy_50bins": 0.3461995490917764,
"calibration/batch_uniqueness": 0.17352555787323637,
"calibration/confidence_entropy": 0.4251987915992827,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.39703934791792184,
"calibration/coverage@10%": 0.5134270126524422,
"calibration/coverage@15%": 0.7232000000000001,
"calibration/coverage@20%": 0.8891553133514986,
"calibration/coverage@25%": 0.9139211625794733,
"calibration/coverage@30%": 0.9462466848319708,
"calibration/coverage@5%": 0.48544393797740987,
"calibration/distribution_entropy_10": 0.5355198771027292,
"calibration/distribution_entropy_100": 0.2940913247107916,
"calibration/ece": 0.12311490454704932,
"calibration/mean_confidence": 0.7432522071893854,
"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": 0.01918402777777779,
"completions/max_length": 3894.0,
"completions/max_terminated_length": 3894.0,
"completions/mean_length": 1133.996337890625,
"completions/mean_terminated_length": 1156.170947265625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 332.4,
"epoch": 1.8509615384615383,
"grad_norm": 0.0003580257180146873,
"learning_rate": 1.622596153846154e-06,
"loss": -0.024,
"num_tokens": 2046281922.0,
"reward": 1.2793505191802979,
"reward_std": 0.13908308297395705,
"rewards/accuracy_reward": 0.739062488079071,
"rewards/brier_reward": 0.8388078927993774,
"rewards/confidence_one_or_zero": 0.000520833337213844,
"rewards/format_reward": 0.9808159708976746,
"rewards/mean_confidence_reward": 0.739722204208374,
"sampling/batch_mean_priority_error": 0.029795208333333316,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.14444444444444446,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0026988604106009005,
"sampling/priority_kl": 0.03000030443072319,
"sampling/priority_scale": 0.8066103636985644,
"sampling/prob_entropy": 10.278947448730468,
"sampling/prob_max": 5.877538642380386e-05,
"sampling/prob_min": 2.0727198352687994e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.8432000160217286,
"sampling/prompt_draws_total": 55296.0,
"sampling/seen_fraction": 0.9205333352088928,
"sampling/unseen_fraction": 0.07946666479110717,
"signal/accuracy_reward/centered_abs_mean": 0.09299045205116271,
"signal/accuracy_reward/group_std_mean": 0.14082872718572617,
"signal/accuracy_reward/group_zero_std_frac": 0.5194444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04649522602558136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04649522602558136,
"signal/advantage_abs_mean": 0.08546854555606842,
"signal/advantage_pre_scale_abs_mean": 0.08546854555606842,
"signal/advantage_pre_scale_std": 0.19609870314598082,
"signal/advantage_std": 0.19609870314598082,
"signal/brier_reward/centered_abs_mean": 0.0788871854543686,
"signal/brier_reward/group_std_mean": 0.11534269303083419,
"signal/brier_reward/group_zero_std_frac": 0.29166666567325594,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0394435927271843,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0394435927271843,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0009765624767169356,
"signal/confidence_one_or_zero/group_std_mean": 0.002295762859284878,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9888888835906983,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 9.765624042756826e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 9.765624042756826e-09,
"signal/format_reward/centered_abs_mean": 0.03418511338531971,
"signal/format_reward/group_std_mean": 0.0701390728354454,
"signal/format_reward/group_zero_std_frac": 0.6916666865348816,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.017092556692659855,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.017092556692659855,
"signal/mean_confidence_reward/centered_abs_mean": 0.06734917759895324,
"signal/mean_confidence_reward/group_std_mean": 0.09585417956113815,
"signal/mean_confidence_reward/group_zero_std_frac": 0.31111110746860504,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.734917519679584e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.734917519679584e-07,
"step": 770
},
{
"calibration/aurc": 0.12452181176719923,
"calibration/batch_distribution_entropy": 0.5645438819251132,
"calibration/batch_entropy_100bins": 0.30598169572049205,
"calibration/batch_entropy_10bins": 0.5645438819251132,
"calibration/batch_entropy_50bins": 0.36019670145980476,
"calibration/batch_uniqueness": 0.21593060217032994,
"calibration/confidence_entropy": 0.43755199598394123,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5315212310817679,
"calibration/coverage@15%": 0.6500954204779696,
"calibration/coverage@20%": 0.7159204687677779,
"calibration/coverage@25%": 0.7367021276595744,
"calibration/coverage@30%": 0.9347236294044805,
"calibration/coverage@5%": 0.3420548763580603,
"calibration/distribution_entropy_10": 0.5645438819251132,
"calibration/distribution_entropy_100": 0.30598169572049205,
"calibration/ece": 0.1157229115376667,
"calibration/mean_confidence": 0.7470467694516723,
"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": 0.020572916666666698,
"completions/max_length": 4037.2,
"completions/max_terminated_length": 4037.2,
"completions/mean_length": 1137.8390869140626,
"completions/mean_terminated_length": 1161.6483642578125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 346.0,
"epoch": 1.8629807692307692,
"grad_norm": 0.0003609874111134559,
"learning_rate": 1.592548076923077e-06,
"loss": -0.0248,
"num_tokens": 2062517284.0,
"reward": 1.2978382110595703,
"reward_std": 0.16978729665279388,
"rewards/accuracy_reward": 0.757118034362793,
"rewards/brier_reward": 0.859116506576538,
"rewards/confidence_one_or_zero": 0.0005208333430346101,
"rewards/format_reward": 0.9794270873069764,
"rewards/mean_confidence_reward": 0.7398263812065125,
"sampling/batch_mean_priority_error": 0.015230972222222208,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1388888888888889,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.002722617890685797,
"sampling/priority_kl": 0.030000732839107515,
"sampling/priority_scale": 0.8091420352226123,
"sampling/prob_entropy": 10.278940010070801,
"sampling/prob_max": 5.902362318011001e-05,
"sampling/prob_min": 2.07467583095422e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.8552000045776367,
"sampling/prompt_draws_total": 55656.0,
"sampling/seen_fraction": 0.9221866846084594,
"sampling/unseen_fraction": 0.07781331539154053,
"signal/accuracy_reward/centered_abs_mean": 0.13017577975988387,
"signal/accuracy_reward/group_std_mean": 0.1837555378675461,
"signal/accuracy_reward/group_zero_std_frac": 0.42222222685813904,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06508788987994193,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06508788987994193,
"signal/advantage_abs_mean": 0.10840746015310287,
"signal/advantage_pre_scale_abs_mean": 0.10840746015310287,
"signal/advantage_pre_scale_std": 0.22453723847866058,
"signal/advantage_std": 0.22453723847866058,
"signal/brier_reward/centered_abs_mean": 0.0910664215683937,
"signal/brier_reward/group_std_mean": 0.13270987272262574,
"signal/brier_reward/group_zero_std_frac": 0.2333333373069763,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04553321078419685,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04553321078419685,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0009657118120230734,
"signal/confidence_one_or_zero/group_std_mean": 0.001996822049841285,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 9.6571170615789e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 9.6571170615789e-09,
"signal/format_reward/centered_abs_mean": 0.0372775599360466,
"signal/format_reward/group_std_mean": 0.07882281616330147,
"signal/format_reward/group_zero_std_frac": 0.6472222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0186387799680233,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0186387799680233,
"signal/mean_confidence_reward/centered_abs_mean": 0.07244938015937805,
"signal/mean_confidence_reward/group_std_mean": 0.1059411883354187,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2472222238779068,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.244937705763732e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.244937705763732e-07,
"step": 775
},
{
"calibration/aurc": 0.11927318188747289,
"calibration/batch_distribution_entropy": 0.5832886490653397,
"calibration/batch_entropy_100bins": 0.32257344192265663,
"calibration/batch_entropy_10bins": 0.5832886490653397,
"calibration/batch_entropy_50bins": 0.37972823663678856,
"calibration/batch_uniqueness": 0.2869097012867606,
"calibration/confidence_entropy": 0.4442802120811054,
"calibration/coverage@0%": 0.0026913712699970275,
"calibration/coverage@1%": 0.0026913712699970275,
"calibration/coverage@10%": 0.6345159195735076,
"calibration/coverage@15%": 0.7238751797743481,
"calibration/coverage@20%": 0.7616195582547809,
"calibration/coverage@25%": 0.8136705538510954,
"calibration/coverage@30%": 0.8939209882708932,
"calibration/coverage@5%": 0.398541891975399,
"calibration/distribution_entropy_10": 0.5832886490653397,
"calibration/distribution_entropy_100": 0.32257344192265663,
"calibration/ece": 0.124093627094425,
"calibration/mean_confidence": 0.7332632525470724,
"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": 0.02274305555555558,
"completions/max_length": 3808.8,
"completions/max_terminated_length": 3808.8,
"completions/mean_length": 1098.0076538085937,
"completions/mean_terminated_length": 1123.6128173828124,
"completions/min_length": 0.0,
"completions/min_terminated_length": 338.4,
"epoch": 1.875,
"grad_norm": 0.00035296386340633035,
"learning_rate": 1.5625e-06,
"loss": -0.0281,
"num_tokens": 2078260412.0,
"reward": 1.2954104661941528,
"reward_std": 0.16067642271518706,
"rewards/accuracy_reward": 0.7480902791023254,
"rewards/brier_reward": 0.8654589772224426,
"rewards/confidence_one_or_zero": 0.0014756944379769265,
"rewards/format_reward": 0.9772569537162781,
"rewards/mean_confidence_reward": 0.739305567741394,
"sampling/batch_mean_priority_error": 0.016687638888888877,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.11944444444444444,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0027397376950830223,
"sampling/priority_kl": 0.029999836534261703,
"sampling/priority_scale": 0.8112623870139941,
"sampling/prob_entropy": 10.27894859313965,
"sampling/prob_max": 5.925482328166254e-05,
"sampling/prob_min": 2.0771039271494373e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.8672000169754028,
"sampling/prompt_draws_total": 56016.0,
"sampling/seen_fraction": 0.9236533403396606,
"sampling/unseen_fraction": 0.07634665966033935,
"signal/accuracy_reward/centered_abs_mean": 0.10810547024011612,
"signal/accuracy_reward/group_std_mean": 0.15696669220924378,
"signal/accuracy_reward/group_zero_std_frac": 0.4888888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05405273512005806,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05405273512005806,
"signal/advantage_abs_mean": 0.09703816920518875,
"signal/advantage_pre_scale_abs_mean": 0.09703816920518875,
"signal/advantage_pre_scale_std": 0.21690228283405305,
"signal/advantage_std": 0.21690228283405305,
"signal/brier_reward/centered_abs_mean": 0.08053490072488785,
"signal/brier_reward/group_std_mean": 0.1245546817779541,
"signal/brier_reward/group_zero_std_frac": 0.22777777910232544,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.040267450362443925,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.040267450362443925,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0024576822528615595,
"signal/confidence_one_or_zero/group_std_mean": 0.004898328334093094,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9777777671813965,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.4576820578658953e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.4576820578658953e-08,
"signal/format_reward/centered_abs_mean": 0.04134114608168602,
"signal/format_reward/group_std_mean": 0.08592917323112488,
"signal/format_reward/group_zero_std_frac": 0.6250000119209289,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.02067057304084301,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.02067057304084301,
"signal/mean_confidence_reward/centered_abs_mean": 0.06903146356344222,
"signal/mean_confidence_reward/group_std_mean": 0.1041269913315773,
"signal/mean_confidence_reward/group_zero_std_frac": 0.23611111044883729,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.903146299919172e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.903146299919172e-07,
"step": 780
},
{
"calibration/aurc": 0.0742543345201262,
"calibration/batch_distribution_entropy": 0.589344371392418,
"calibration/batch_entropy_100bins": 0.32672389653760925,
"calibration/batch_entropy_10bins": 0.589344371392418,
"calibration/batch_entropy_50bins": 0.38461408465571767,
"calibration/batch_uniqueness": 0.31280304906023604,
"calibration/confidence_entropy": 0.44947390020650085,
"calibration/coverage@0%": 0.45254355752482356,
"calibration/coverage@1%": 0.4780291397957228,
"calibration/coverage@10%": 0.7283491565285981,
"calibration/coverage@15%": 0.7909260581440964,
"calibration/coverage@20%": 0.8653551912568307,
"calibration/coverage@25%": 0.8749551912568307,
"calibration/coverage@30%": 0.9204109289617486,
"calibration/coverage@5%": 0.6444980282635661,
"calibration/distribution_entropy_10": 0.589344371392418,
"calibration/distribution_entropy_100": 0.32672389653760925,
"calibration/ece": 0.1305743347675779,
"calibration/mean_confidence": 0.7089313602672528,
"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": 0.03116319444444444,
"completions/max_length": 3744.0,
"completions/max_terminated_length": 3744.0,
"completions/mean_length": 1160.1212890625,
"completions/mean_terminated_length": 1197.4378662109375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 397.8,
"epoch": 1.8870192307692308,
"grad_norm": 0.00033553142566233873,
"learning_rate": 1.5324519230769232e-06,
"loss": -0.0351,
"num_tokens": 2094745393.0,
"reward": 1.2564839124679565,
"reward_std": 0.17908986806869506,
"rewards/accuracy_reward": 0.6998263955116272,
"rewards/brier_reward": 0.8443773746490478,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9687499880790711,
"rewards/mean_confidence_reward": 0.7083628416061402,
"sampling/batch_mean_priority_error": 0.021218819444444433,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.11666666666666667,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.002760501578450203,
"sampling/priority_kl": 0.029999128356575967,
"sampling/priority_scale": 0.8130677400855347,
"sampling/prob_entropy": 10.278962326049804,
"sampling/prob_max": 5.946964593022131e-05,
"sampling/prob_min": 2.066538181679789e-05,
"sampling/prompt_draws_max": 7.2,
"sampling/prompt_draws_mean": 1.8791999816894531,
"sampling/prompt_draws_total": 56376.0,
"sampling/seen_fraction": 0.925,
"sampling/unseen_fraction": 0.075,
"signal/accuracy_reward/centered_abs_mean": 0.11258680373430252,
"signal/accuracy_reward/group_std_mean": 0.16565240919589996,
"signal/accuracy_reward/group_zero_std_frac": 0.46111111640930175,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05629340186715126,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05629340186715126,
"signal/advantage_abs_mean": 0.11116953194141388,
"signal/advantage_pre_scale_abs_mean": 0.11116953194141388,
"signal/advantage_pre_scale_std": 0.23421459197998046,
"signal/advantage_std": 0.23421459197998046,
"signal/brier_reward/centered_abs_mean": 0.09295912683010102,
"signal/brier_reward/group_std_mean": 0.1409787118434906,
"signal/brier_reward/group_zero_std_frac": 0.21666666567325593,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04647956341505051,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04647956341505051,
"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": 0.05329861119389534,
"signal/format_reward/group_std_mean": 0.10066336393356323,
"signal/format_reward/group_zero_std_frac": 0.5888888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.02664930559694767,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.02664930559694767,
"signal/mean_confidence_reward/centered_abs_mean": 0.07824294120073319,
"signal/mean_confidence_reward/group_std_mean": 0.11478137820959092,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2222222238779068,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.824294129932241e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.824294129932241e-07,
"step": 785
},
{
"calibration/aurc": 0.06961830567263416,
"calibration/batch_distribution_entropy": 0.5126674195657974,
"calibration/batch_entropy_100bins": 0.28559755646404594,
"calibration/batch_entropy_10bins": 0.5126674195657974,
"calibration/batch_entropy_50bins": 0.3362008225397263,
"calibration/batch_uniqueness": 0.13931816006468628,
"calibration/confidence_entropy": 0.42045557884474016,
"calibration/coverage@0%": 0.1750028368794326,
"calibration/coverage@1%": 0.4881517730496454,
"calibration/coverage@10%": 0.6925202687569989,
"calibration/coverage@15%": 0.8557704804304482,
"calibration/coverage@20%": 0.9347311367850519,
"calibration/coverage@25%": 0.9788617886178862,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.5073219858156028,
"calibration/distribution_entropy_10": 0.5126674195657974,
"calibration/distribution_entropy_100": 0.28559755646404594,
"calibration/ece": 0.14474426545519645,
"calibration/mean_confidence": 0.7469527089649397,
"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": 0.022048611111111116,
"completions/max_length": 3817.6,
"completions/max_terminated_length": 3817.6,
"completions/mean_length": 1154.2109619140624,
"completions/mean_terminated_length": 1180.3725341796876,
"completions/min_length": 0.0,
"completions/min_terminated_length": 356.6,
"epoch": 1.8990384615384617,
"grad_norm": 0.00026942059048451483,
"learning_rate": 1.5024038461538462e-06,
"loss": -0.0259,
"num_tokens": 2111142735.0,
"reward": 1.2929217100143433,
"reward_std": 0.1408931255340576,
"rewards/accuracy_reward": 0.7424479126930237,
"rewards/brier_reward": 0.8654301285743713,
"rewards/confidence_one_or_zero": 0.000260416668606922,
"rewards/format_reward": 0.9779513955116272,
"rewards/mean_confidence_reward": 0.7021756172180176,
"sampling/batch_mean_priority_error": 0.02039062499999999,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.16666666666666669,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0027823722921311855,
"sampling/priority_kl": 0.03000016063451767,
"sampling/priority_scale": 0.8168170273536817,
"sampling/prob_entropy": 10.278951454162598,
"sampling/prob_max": 5.9781748859677464e-05,
"sampling/prob_min": 2.014181300182827e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.8911999940872193,
"sampling/prompt_draws_total": 56736.0,
"sampling/seen_fraction": 0.9269533276557922,
"sampling/unseen_fraction": 0.07304667234420777,
"signal/accuracy_reward/centered_abs_mean": 0.09775933176279068,
"signal/accuracy_reward/group_std_mean": 0.14343787282705306,
"signal/accuracy_reward/group_zero_std_frac": 0.5361111342906952,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04887966588139534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04887966588139534,
"signal/advantage_abs_mean": 0.08402638137340546,
"signal/advantage_pre_scale_abs_mean": 0.08402638137340546,
"signal/advantage_pre_scale_std": 0.19729985892772675,
"signal/advantage_std": 0.19729985892772675,
"signal/brier_reward/centered_abs_mean": 0.07575573846697807,
"signal/brier_reward/group_std_mean": 0.1154901534318924,
"signal/brier_reward/group_zero_std_frac": 0.26666666865348815,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.037877869233489035,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.037877869233489035,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0004937065881676972,
"signal/confidence_one_or_zero/group_std_mean": 0.001174198230728507,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.9370658672387435e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.9370658672387435e-09,
"signal/format_reward/centered_abs_mean": 0.03956163190305233,
"signal/format_reward/group_std_mean": 0.07983514145016671,
"signal/format_reward/group_zero_std_frac": 0.6611111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.019780815951526164,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.019780815951526164,
"signal/mean_confidence_reward/centered_abs_mean": 0.06639526039361954,
"signal/mean_confidence_reward/group_std_mean": 0.09458223283290863,
"signal/mean_confidence_reward/group_zero_std_frac": 0.28611111640930176,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.639525736318319e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.639525736318319e-07,
"step": 790
},
{
"calibration/aurc": 0.10809945506791305,
"calibration/batch_distribution_entropy": 0.51655803839253,
"calibration/batch_entropy_100bins": 0.2764817265817528,
"calibration/batch_entropy_10bins": 0.51655803839253,
"calibration/batch_entropy_50bins": 0.3254698150951826,
"calibration/batch_uniqueness": 0.10905104035191324,
"calibration/confidence_entropy": 0.4123756179040833,
"calibration/coverage@0%": 0.16886543535620052,
"calibration/coverage@1%": 0.18258575197889182,
"calibration/coverage@10%": 0.5745128779395297,
"calibration/coverage@15%": 0.7417426545086119,
"calibration/coverage@20%": 0.7608465608465609,
"calibration/coverage@25%": 0.9472339954015346,
"calibration/coverage@30%": 0.9883597883597883,
"calibration/coverage@5%": 0.509484882418813,
"calibration/distribution_entropy_10": 0.51655803839253,
"calibration/distribution_entropy_100": 0.2764817265817528,
"calibration/ece": 0.11734529183564894,
"calibration/mean_confidence": 0.7646222394616007,
"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": 0.018142361111111095,
"completions/max_length": 3878.4,
"completions/max_terminated_length": 3878.4,
"completions/mean_length": 1139.3298095703126,
"completions/mean_terminated_length": 1160.3458251953125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 348.0,
"epoch": 1.9110576923076923,
"grad_norm": 0.0003292847832199186,
"learning_rate": 1.4723557692307693e-06,
"loss": -0.021,
"num_tokens": 2127333094.0,
"reward": 1.3033838033676148,
"reward_std": 0.13692439794540406,
"rewards/accuracy_reward": 0.7607638835906982,
"rewards/brier_reward": 0.8641319513320923,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9818576455116272,
"rewards/mean_confidence_reward": 0.7087239623069763,
"sampling/batch_mean_priority_error": 0.01860763888888887,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1388888888888889,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0028058627154678105,
"sampling/priority_kl": 0.02999955378472805,
"sampling/priority_scale": 0.8200449288124219,
"sampling/prob_entropy": 10.278947067260741,
"sampling/prob_max": 6.007041083648801e-05,
"sampling/prob_min": 2.0152244542259722e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9031999826431274,
"sampling/prompt_draws_total": 57096.0,
"sampling/seen_fraction": 0.9286933422088623,
"sampling/unseen_fraction": 0.07130665779113769,
"signal/accuracy_reward/centered_abs_mean": 0.10312500149011612,
"signal/accuracy_reward/group_std_mean": 0.1497005671262741,
"signal/accuracy_reward/group_zero_std_frac": 0.5250000178813934,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05156250074505806,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05156250074505806,
"signal/advantage_abs_mean": 0.08454073220491409,
"signal/advantage_pre_scale_abs_mean": 0.08454073220491409,
"signal/advantage_pre_scale_std": 0.19431009292602539,
"signal/advantage_std": 0.19431009292602539,
"signal/brier_reward/centered_abs_mean": 0.07752134352922439,
"signal/brier_reward/group_std_mean": 0.11462950706481934,
"signal/brier_reward/group_zero_std_frac": 0.25833333730697633,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.038760671764612196,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.038760671764612196,
"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": 0.032470703125,
"signal/format_reward/group_std_mean": 0.06638929843902588,
"signal/format_reward/group_zero_std_frac": 0.7138888835906982,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0162353515625,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0162353515625,
"signal/mean_confidence_reward/centered_abs_mean": 0.0648546002805233,
"signal/mean_confidence_reward/group_std_mean": 0.09345088601112365,
"signal/mean_confidence_reward/group_zero_std_frac": 0.26944445371627807,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.485460176008928e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.485460176008928e-07,
"step": 795
},
{
"calibration/aurc": 0.08310682378745267,
"calibration/batch_distribution_entropy": 0.5474554567180643,
"calibration/batch_entropy_100bins": 0.2976009201485093,
"calibration/batch_entropy_10bins": 0.5474554567180643,
"calibration/batch_entropy_50bins": 0.3503309880562792,
"calibration/batch_uniqueness": 0.1914895892966749,
"calibration/confidence_entropy": 0.4297139293492359,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.1569553805774278,
"calibration/coverage@10%": 0.5246396339494996,
"calibration/coverage@15%": 0.900323914456959,
"calibration/coverage@20%": 0.9470198799550718,
"calibration/coverage@25%": 0.9659520529071368,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.47196373180625156,
"calibration/distribution_entropy_10": 0.5474554567180643,
"calibration/distribution_entropy_100": 0.2976009201485093,
"calibration/ece": 0.09881079840529157,
"calibration/mean_confidence": 0.7574989102430648,
"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": 0.015538194444444441,
"completions/max_length": 3858.8,
"completions/max_terminated_length": 3858.8,
"completions/mean_length": 1155.186572265625,
"completions/mean_terminated_length": 1173.2910888671875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 392.2,
"epoch": 1.9230769230769231,
"grad_norm": 0.00032893981551751494,
"learning_rate": 1.4423076923076922e-06,
"loss": -0.0176,
"num_tokens": 2143736747.0,
"reward": 1.2860599279403686,
"reward_std": 0.12449116706848144,
"rewards/accuracy_reward": 0.7293402791023255,
"rewards/brier_reward": 0.8583038091659546,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9844618082046509,
"rewards/mean_confidence_reward": 0.6920920133590698,
"sampling/batch_mean_priority_error": 0.02531944444444443,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1222222222222222,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.002828038949519396,
"sampling/priority_kl": 0.030001144856214523,
"sampling/priority_scale": 0.8224454701179639,
"sampling/prob_entropy": 10.278958320617676,
"sampling/prob_max": 6.031659504515119e-05,
"sampling/prob_min": 2.0170245261397212e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9151999950408936,
"sampling/prompt_draws_total": 57456.0,
"sampling/seen_fraction": 0.930133330821991,
"sampling/unseen_fraction": 0.06986666917800903,
"signal/accuracy_reward/centered_abs_mean": 0.09819878488779069,
"signal/accuracy_reward/group_std_mean": 0.13898843824863433,
"signal/accuracy_reward/group_zero_std_frac": 0.5638889074325562,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04909939244389534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04909939244389534,
"signal/advantage_abs_mean": 0.08012338802218437,
"signal/advantage_pre_scale_abs_mean": 0.08012338802218437,
"signal/advantage_pre_scale_std": 0.18103047609329223,
"signal/advantage_std": 0.18103047609329223,
"signal/brier_reward/centered_abs_mean": 0.07400564849376678,
"signal/brier_reward/group_std_mean": 0.10794279575347901,
"signal/brier_reward/group_zero_std_frac": 0.25000000596046446,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03700282424688339,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03700282424688339,
"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": 0.02664388008415699,
"signal/format_reward/group_std_mean": 0.053740817308425906,
"signal/format_reward/group_zero_std_frac": 0.7666666746139527,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.013321940042078494,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.013321940042078494,
"signal/mean_confidence_reward/centered_abs_mean": 0.06345649063587189,
"signal/mean_confidence_reward/group_std_mean": 0.088161501288414,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2583333343267441,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.345648898786749e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.345648898786749e-07,
"step": 800
},
{
"epoch": 1.9230769230769231,
"eval_calibration/aurc": 0.11880398434664954,
"eval_calibration/batch_distribution_entropy": 0.6147144311896361,
"eval_calibration/batch_entropy_100bins": 0.3377655202084739,
"eval_calibration/batch_entropy_10bins": 0.6147144311896361,
"eval_calibration/batch_entropy_50bins": 0.39761210538908526,
"eval_calibration/batch_uniqueness": 0.3144897799357819,
"eval_calibration/confidence_entropy": 0.44134527960603626,
"eval_calibration/coverage@0%": 0.0008849557522123894,
"eval_calibration/coverage@1%": 0.0008849557522123894,
"eval_calibration/coverage@10%": 0.5610619469026549,
"eval_calibration/coverage@15%": 0.7911504424778761,
"eval_calibration/coverage@20%": 0.8734513274336283,
"eval_calibration/coverage@25%": 0.9362831858407079,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0008849557522123894,
"eval_calibration/distribution_entropy_10": 0.6147144311896361,
"eval_calibration/distribution_entropy_100": 0.3377655202084739,
"eval_calibration/ece": 0.0228318584070795,
"eval_calibration/mean_confidence": 0.7276106194690266,
"eval_calibration/unique_confidence_per_question": 0.008680555555555556,
"eval_calibration/unique_confidences": 10,
"eval_completions/clipped_ratio": 0.016493055555555563,
"eval_completions/max_length": 3315.8333333333335,
"eval_completions/max_terminated_length": 3315.8333333333335,
"eval_completions/mean_length": 1101.993428548177,
"eval_completions/mean_terminated_length": 1120.671407063802,
"eval_completions/min_length": 95.0,
"eval_completions/min_terminated_length": 414.0,
"eval_loss": 0.0,
"eval_num_tokens": 2143736747.0,
"eval_reward": 1.2803370555241902,
"eval_reward_std": 0.3288910339275996,
"eval_rewards/accuracy_reward": 0.7282986144224802,
"eval_rewards/brier_reward": 0.851458340883255,
"eval_rewards/confidence_one_or_zero": 0.0008680555814256271,
"eval_rewards/format_reward": 0.980902781089147,
"eval_rewards/mean_confidence_reward": 0.713715265194575,
"eval_runtime": 215.3204,
"eval_samples_per_second": 4.644,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3815646668275197,
"eval_signal/accuracy_reward/group_std_mean": 0.44144123792648315,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19078233341375986,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19078233341375986,
"eval_signal/advantage_abs_mean": 0.2663094103336334,
"eval_signal/advantage_pre_scale_abs_mean": 0.2663094103336334,
"eval_signal/advantage_pre_scale_std": 0.3311290790637334,
"eval_signal/advantage_std": 0.3311290790637334,
"eval_signal/brier_reward/centered_abs_mean": 0.1741362934311231,
"eval_signal/brier_reward/group_std_mean": 0.23898712793986002,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08706814671556155,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08706814671556155,
"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": 0.0360243059694767,
"eval_signal/format_reward/group_std_mean": 0.08358859581251939,
"eval_signal/format_reward/group_zero_std_frac": 0.6111111293236414,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.01801215298473835,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.01801215298473835,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.21578231950600943,
"eval_signal/mean_confidence_reward/group_std_mean": 0.2589796607693036,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.1578231326202513e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.1578231326202513e-06,
"eval_steps_per_second": 0.028,
"step": 800
},
{
"epoch": 1.9230769230769231,
"step": 800,
"train_probe_calibration/aurc": 0.07785864721043029,
"train_probe_calibration/batch_distribution_entropy": 0.6141568816793782,
"train_probe_calibration/batch_entropy_100bins": 0.33691340500194467,
"train_probe_calibration/batch_entropy_10bins": 0.6141568816793782,
"train_probe_calibration/batch_entropy_50bins": 0.3966090091550676,
"train_probe_calibration/batch_uniqueness": 0.3150661881499335,
"train_probe_calibration/confidence_entropy": 0.4417113423189894,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.7555163283318623,
"train_probe_calibration/coverage@15%": 0.8711385701676964,
"train_probe_calibration/coverage@20%": 0.940864960282436,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.559576345984113,
"train_probe_calibration/distribution_entropy_10": 0.6141568816793782,
"train_probe_calibration/distribution_entropy_100": 0.33691340500194467,
"train_probe_calibration/ece": 0.05710503089143854,
"train_probe_calibration/mean_confidence": 0.7289496910856135,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.013888888888888895,
"train_probe_completions/max_length": 3508.1666666666665,
"train_probe_completions/max_terminated_length": 3508.1666666666665,
"train_probe_completions/mean_length": 1159.2106526692708,
"train_probe_completions/mean_terminated_length": 1175.3173014322917,
"train_probe_completions/min_length": 67.5,
"train_probe_completions/min_terminated_length": 381.0,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 2143736747.0,
"train_probe_reward": 1.3131625652313232,
"train_probe_reward_std": 0.3047017107407252,
"train_probe_rewards/accuracy_reward": 0.7725694477558136,
"train_probe_rewards/brier_reward": 0.8702343702316284,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9835069477558136,
"train_probe_rewards/mean_confidence_reward": 0.7169270614782969,
"train_probe_runtime": 219.8338,
"train_probe_samples_per_second": 4.549,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3395182341337204,
"train_probe_signal/accuracy_reward/group_std_mean": 0.41524408757686615,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1697591170668602,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.1697591170668602,
"train_probe_signal/advantage_abs_mean": 0.2342825730641683,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.2342825730641683,
"train_probe_signal/advantage_pre_scale_std": 0.30573564767837524,
"train_probe_signal/advantage_std": 0.30573564767837524,
"train_probe_signal/brier_reward/centered_abs_mean": 0.1541151429216067,
"train_probe_signal/brier_reward/group_std_mean": 0.21736273914575577,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07705757146080335,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07705757146080335,
"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": 0.03173828125,
"train_probe_signal/format_reward/group_std_mean": 0.08731999310354392,
"train_probe_signal/format_reward/group_zero_std_frac": 0.5277777860562006,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.015869140625,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.015869140625,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.21296656876802444,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.256157249212265,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.1296655935051e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.1296655935051e-06,
"train_probe_steps_per_second": 0.027
},
{
"calibration/aurc": 0.14018359887347748,
"calibration/batch_distribution_entropy": 0.5496416672348073,
"calibration/batch_entropy_100bins": 0.30464898744357305,
"calibration/batch_entropy_10bins": 0.5496416672348073,
"calibration/batch_entropy_50bins": 0.3586278588392549,
"calibration/batch_uniqueness": 0.21255569752260492,
"calibration/confidence_entropy": 0.43085971630365183,
"calibration/coverage@0%": 0.10600522193211488,
"calibration/coverage@1%": 0.10600522193211488,
"calibration/coverage@10%": 0.4817272836293558,
"calibration/coverage@15%": 0.6566745855658296,
"calibration/coverage@20%": 0.681375514946289,
"calibration/coverage@25%": 0.7320855614973262,
"calibration/coverage@30%": 0.9173718776152853,
"calibration/coverage@5%": 0.3815107334653093,
"calibration/distribution_entropy_10": 0.5496416672348073,
"calibration/distribution_entropy_100": 0.30464898744357305,
"calibration/ece": 0.12319044946411499,
"calibration/mean_confidence": 0.7137513905267816,
"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": 0.013368055555555581,
"completions/max_length": 3851.6,
"completions/max_terminated_length": 3851.6,
"completions/mean_length": 1107.5060913085938,
"completions/mean_terminated_length": 1122.5991943359375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 374.4,
"epoch": 1.9350961538461537,
"grad_norm": 0.00039220356848090887,
"learning_rate": 1.4122596153846154e-06,
"loss": -0.0167,
"num_tokens": 2159599793.0,
"reward": 1.3191929578781127,
"reward_std": 0.12780744731426238,
"rewards/accuracy_reward": 0.7817708373069763,
"rewards/brier_reward": 0.8699685335159302,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9866319417953491,
"rewards/mean_confidence_reward": 0.7295781254768372,
"sampling/batch_mean_priority_error": 0.015404583333333314,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.10833333333333332,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0028499219566583633,
"sampling/priority_kl": 0.02999933548271656,
"sampling/priority_scale": 0.8254779875511303,
"sampling/prob_entropy": 10.278951835632324,
"sampling/prob_max": 6.059883380657993e-05,
"sampling/prob_min": 2.0182952357572504e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9272000074386597,
"sampling/prompt_draws_total": 57816.0,
"sampling/seen_fraction": 0.9317333340644837,
"sampling/unseen_fraction": 0.06826666593551636,
"signal/accuracy_reward/centered_abs_mean": 0.11254340261220933,
"signal/accuracy_reward/group_std_mean": 0.15324022769927978,
"signal/accuracy_reward/group_zero_std_frac": 0.547222226858139,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05627170130610466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05627170130610466,
"signal/advantage_abs_mean": 0.08396558910608291,
"signal/advantage_pre_scale_abs_mean": 0.08396558910608291,
"signal/advantage_pre_scale_std": 0.186562117934227,
"signal/advantage_std": 0.186562117934227,
"signal/brier_reward/centered_abs_mean": 0.07254556939005852,
"signal/brier_reward/group_std_mean": 0.10383311659097672,
"signal/brier_reward/group_zero_std_frac": 0.2750000059604645,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03627278469502926,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03627278469502926,
"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": 0.02421874962747097,
"signal/format_reward/group_std_mean": 0.05017288029193878,
"signal/format_reward/group_zero_std_frac": 0.7805555582046508,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.012109374813735484,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.012109374813735484,
"signal/mean_confidence_reward/centered_abs_mean": 0.06297873258590699,
"signal/mean_confidence_reward/group_std_mean": 0.08698192089796067,
"signal/mean_confidence_reward/group_zero_std_frac": 0.27777777910232543,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.297873142102617e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.297873142102617e-07,
"step": 805
},
{
"calibration/aurc": 0.09635417618896773,
"calibration/batch_distribution_entropy": 0.6069344126434967,
"calibration/batch_entropy_100bins": 0.33837644445400744,
"calibration/batch_entropy_10bins": 0.6069344126434967,
"calibration/batch_entropy_50bins": 0.39833127552625586,
"calibration/batch_uniqueness": 0.3299621599214643,
"calibration/confidence_entropy": 0.4488800064944396,
"calibration/coverage@0%": 0.2592541216380225,
"calibration/coverage@1%": 0.35666938755149735,
"calibration/coverage@10%": 0.6201582130176267,
"calibration/coverage@15%": 0.7854698715538,
"calibration/coverage@20%": 0.8520324225968146,
"calibration/coverage@25%": 0.9184785179743304,
"calibration/coverage@30%": 0.9486772486772488,
"calibration/coverage@5%": 0.5282726669246759,
"calibration/distribution_entropy_10": 0.6069344126434967,
"calibration/distribution_entropy_100": 0.33837644445400744,
"calibration/ece": 0.14434275009587091,
"calibration/mean_confidence": 0.7124749362295653,
"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": 0.010937499999999978,
"completions/max_length": 3784.4,
"completions/max_terminated_length": 3784.4,
"completions/mean_length": 1093.8130126953124,
"completions/mean_terminated_length": 1106.06572265625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 320.4,
"epoch": 1.9471153846153846,
"grad_norm": 0.00026345642982050776,
"learning_rate": 1.3822115384615387e-06,
"loss": -0.0131,
"num_tokens": 2175325671.0,
"reward": 1.3192132711410522,
"reward_std": 0.11112102270126342,
"rewards/accuracy_reward": 0.7668402791023254,
"rewards/brier_reward": 0.8825091242790222,
"rewards/confidence_one_or_zero": 0.000260416668606922,
"rewards/format_reward": 0.9890625,
"rewards/mean_confidence_reward": 0.7354774475097656,
"sampling/batch_mean_priority_error": 0.014980902777777763,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.12222222222222223,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0028622327372431755,
"sampling/priority_kl": 0.02999916300177574,
"sampling/priority_scale": 0.8274989903206006,
"sampling/prob_entropy": 10.27893295288086,
"sampling/prob_max": 6.0829608264612035e-05,
"sampling/prob_min": 2.0205341570544988e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9391999959945678,
"sampling/prompt_draws_total": 58176.0,
"sampling/seen_fraction": 0.9329733371734619,
"sampling/unseen_fraction": 0.06702666282653809,
"signal/accuracy_reward/centered_abs_mean": 0.08981119841337204,
"signal/accuracy_reward/group_std_mean": 0.125412817299366,
"signal/accuracy_reward/group_zero_std_frac": 0.619444465637207,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04490559920668602,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04490559920668602,
"signal/advantage_abs_mean": 0.07182730138301849,
"signal/advantage_pre_scale_abs_mean": 0.07182730138301849,
"signal/advantage_pre_scale_std": 0.16563299298286438,
"signal/advantage_std": 0.16563299298286438,
"signal/brier_reward/centered_abs_mean": 0.06471699699759484,
"signal/brier_reward/group_std_mean": 0.09428556412458419,
"signal/brier_reward/group_zero_std_frac": 0.3027777761220932,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03235849849879742,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03235849849879742,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0005045572877861559,
"signal/confidence_one_or_zero/group_std_mean": 0.0014731390401721,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/format_reward/centered_abs_mean": 0.020095486007630826,
"signal/format_reward/group_std_mean": 0.04359339289367199,
"signal/format_reward/group_zero_std_frac": 0.8027777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010047743003815413,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010047743003815413,
"signal/mean_confidence_reward/centered_abs_mean": 0.057460668683052066,
"signal/mean_confidence_reward/group_std_mean": 0.08030145466327668,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3111111134290695,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.746066847223119e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.746066847223119e-07,
"step": 810
},
{
"calibration/aurc": 0.11943973394907752,
"calibration/batch_distribution_entropy": 0.6097454318555073,
"calibration/batch_entropy_100bins": 0.3330386406506821,
"calibration/batch_entropy_10bins": 0.6097454318555073,
"calibration/batch_entropy_50bins": 0.3920476992539233,
"calibration/batch_uniqueness": 0.35021879727501143,
"calibration/confidence_entropy": 0.46230057877054287,
"calibration/coverage@0%": 0.0005208333333333333,
"calibration/coverage@1%": 0.11968093832020996,
"calibration/coverage@10%": 0.4596600877192983,
"calibration/coverage@15%": 0.6889399384355663,
"calibration/coverage@20%": 0.7673364872512395,
"calibration/coverage@25%": 0.8386753920915095,
"calibration/coverage@30%": 0.928375872600349,
"calibration/coverage@5%": 0.44544956140350883,
"calibration/distribution_entropy_10": 0.6097454318555073,
"calibration/distribution_entropy_100": 0.3330386406506821,
"calibration/ece": 0.12871242879192818,
"calibration/mean_confidence": 0.7380755399174433,
"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": 0.010763888888888906,
"completions/max_length": 3856.6,
"completions/max_terminated_length": 3856.6,
"completions/mean_length": 1123.5369384765625,
"completions/mean_terminated_length": 1135.761328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 347.8,
"epoch": 1.9591346153846154,
"grad_norm": 0.00031497349846176803,
"learning_rate": 1.3521634615384617e-06,
"loss": -0.0125,
"num_tokens": 2191361008.0,
"reward": 1.2978993654251099,
"reward_std": 0.11660217195749283,
"rewards/accuracy_reward": 0.7500867962837219,
"rewards/brier_reward": 0.8564614892005921,
"rewards/confidence_one_or_zero": 0.0003472222248092294,
"rewards/format_reward": 0.9892361164093018,
"rewards/mean_confidence_reward": 0.7145020127296448,
"sampling/batch_mean_priority_error": 0.01364756944444443,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.125,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.002879917761310935,
"sampling/priority_kl": 0.030000254139304162,
"sampling/priority_scale": 0.8304974854225293,
"sampling/prob_entropy": 10.278964805603028,
"sampling/prob_max": 6.111022739787585e-05,
"sampling/prob_min": 2.0217650671838783e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.951200008392334,
"sampling/prompt_draws_total": 58536.0,
"sampling/seen_fraction": 0.9345066666603088,
"sampling/unseen_fraction": 0.06549333333969116,
"signal/accuracy_reward/centered_abs_mean": 0.09825846254825592,
"signal/accuracy_reward/group_std_mean": 0.13935578167438506,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111283302307,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04912923127412796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04912923127412796,
"signal/advantage_abs_mean": 0.07806268632411957,
"signal/advantage_pre_scale_abs_mean": 0.07806268632411957,
"signal/advantage_pre_scale_std": 0.16955201625823973,
"signal/advantage_std": 0.16955201625823973,
"signal/brier_reward/centered_abs_mean": 0.0731019839644432,
"signal/brier_reward/group_std_mean": 0.1011770486831665,
"signal/brier_reward/group_zero_std_frac": 0.2722222298383713,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0365509919822216,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0365509919822216,
"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": 0.01690538190305233,
"signal/format_reward/group_std_mean": 0.03581056408584118,
"signal/format_reward/group_zero_std_frac": 0.8361111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008452690951526166,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008452690951526166,
"signal/mean_confidence_reward/centered_abs_mean": 0.0661589540541172,
"signal/mean_confidence_reward/group_std_mean": 0.08866492062807083,
"signal/mean_confidence_reward/group_zero_std_frac": 0.28333333432674407,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.615895244976855e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.615895244976855e-07,
"step": 815
},
{
"calibration/aurc": 0.08312325960839878,
"calibration/batch_distribution_entropy": 0.5998779297506034,
"calibration/batch_entropy_100bins": 0.3265478904334089,
"calibration/batch_entropy_10bins": 0.5998779297506034,
"calibration/batch_entropy_50bins": 0.38440689311760784,
"calibration/batch_uniqueness": 0.34809773650721676,
"calibration/confidence_entropy": 0.4625567678265304,
"calibration/coverage@0%": 0.13159268929503914,
"calibration/coverage@1%": 0.2706551892950392,
"calibration/coverage@10%": 0.7121716413964393,
"calibration/coverage@15%": 0.7550979655032367,
"calibration/coverage@20%": 0.8651255174136491,
"calibration/coverage@25%": 0.9336898395721924,
"calibration/coverage@30%": 0.9336898395721924,
"calibration/coverage@5%": 0.5316787422227396,
"calibration/distribution_entropy_10": 0.5998779297506034,
"calibration/distribution_entropy_100": 0.3265478904334089,
"calibration/ece": 0.11884740100683082,
"calibration/mean_confidence": 0.7191161629178507,
"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": 3874.8,
"completions/max_terminated_length": 3874.8,
"completions/mean_length": 1103.8129638671876,
"completions/mean_terminated_length": 1113.31953125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 335.4,
"epoch": 1.9711538461538463,
"grad_norm": 0.00028875298448838294,
"learning_rate": 1.3221153846153848e-06,
"loss": -0.0115,
"num_tokens": 2207179365.0,
"reward": 1.3053389310836792,
"reward_std": 0.10983250439167022,
"rewards/accuracy_reward": 0.7494791626930237,
"rewards/brier_reward": 0.8696907758712769,
"rewards/confidence_one_or_zero": 0.0005208333488553763,
"rewards/format_reward": 0.9914930582046508,
"rewards/mean_confidence_reward": 0.7453255295753479,
"sampling/batch_mean_priority_error": 0.017092013888888875,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.09166666666666666,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0028944870922714473,
"sampling/priority_kl": 0.03000035136938095,
"sampling/priority_scale": 0.8323657929664477,
"sampling/prob_entropy": 10.278962516784668,
"sampling/prob_max": 6.1331047618296e-05,
"sampling/prob_min": 2.0240285448380745e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9632000207901001,
"sampling/prompt_draws_total": 58896.0,
"sampling/seen_fraction": 0.935646653175354,
"sampling/unseen_fraction": 0.064353346824646,
"signal/accuracy_reward/centered_abs_mean": 0.08990885466337203,
"signal/accuracy_reward/group_std_mean": 0.12569952458143235,
"signal/accuracy_reward/group_zero_std_frac": 0.6111111164093017,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04495442733168602,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04495442733168602,
"signal/advantage_abs_mean": 0.07215002328157424,
"signal/advantage_pre_scale_abs_mean": 0.07215002328157424,
"signal/advantage_pre_scale_std": 0.16824093759059905,
"signal/advantage_std": 0.16824093759059905,
"signal/brier_reward/centered_abs_mean": 0.06411735936999322,
"signal/brier_reward/group_std_mean": 0.09321332573890687,
"signal/brier_reward/group_zero_std_frac": 0.286111107468605,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03205867968499661,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03205867968499661,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0009440104477107525,
"signal/confidence_one_or_zero/group_std_mean": 0.0016452476382255554,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 9.44010345449442e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 9.44010345449442e-09,
"signal/format_reward/centered_abs_mean": 0.01577690988779068,
"signal/format_reward/group_std_mean": 0.035082895308732986,
"signal/format_reward/group_zero_std_frac": 0.8388888835906982,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00788845494389534,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00788845494389534,
"signal/mean_confidence_reward/centered_abs_mean": 0.05973497703671456,
"signal/mean_confidence_reward/group_std_mean": 0.08283800929784775,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2972222238779068,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.973497536615469e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.973497536615469e-07,
"step": 820
},
{
"calibration/aurc": 0.134517132188666,
"calibration/batch_distribution_entropy": 0.5765132471956196,
"calibration/batch_entropy_100bins": 0.3098934592550557,
"calibration/batch_entropy_10bins": 0.5765132471956196,
"calibration/batch_entropy_50bins": 0.3648015662008893,
"calibration/batch_uniqueness": 0.2773345030487901,
"calibration/confidence_entropy": 0.4444361845612733,
"calibration/coverage@0%": 0.08743455497382199,
"calibration/coverage@1%": 0.1130890052356021,
"calibration/coverage@10%": 0.473882286361064,
"calibration/coverage@15%": 0.6896555209298408,
"calibration/coverage@20%": 0.7025587467362925,
"calibration/coverage@25%": 0.7706706060189639,
"calibration/coverage@30%": 0.8,
"calibration/coverage@5%": 0.4251553643439793,
"calibration/distribution_entropy_10": 0.5765132471956196,
"calibration/distribution_entropy_100": 0.3098934592550557,
"calibration/ece": 0.14405314137800523,
"calibration/mean_confidence": 0.7510510679057971,
"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": 0.009809027777777767,
"completions/max_length": 3832.2,
"completions/max_terminated_length": 3832.2,
"completions/mean_length": 1109.3079833984375,
"completions/mean_terminated_length": 1120.2862060546875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 319.8,
"epoch": 1.9831730769230769,
"grad_norm": 0.0003624038945417851,
"learning_rate": 1.292067307692308e-06,
"loss": -0.0095,
"num_tokens": 2223047681.0,
"reward": 1.2855605363845826,
"reward_std": 0.12681028991937637,
"rewards/accuracy_reward": 0.725,
"rewards/brier_reward": 0.8560023784637452,
"rewards/confidence_one_or_zero": 0.0014756944321561605,
"rewards/format_reward": 0.9901041746139526,
"rewards/mean_confidence_reward": 0.7217751502990722,
"sampling/batch_mean_priority_error": 0.020626736111111096,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.125,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0029144018888473512,
"sampling/priority_kl": 0.029999225586652755,
"sampling/priority_scale": 0.8351738035911694,
"sampling/prob_entropy": 10.278949737548828,
"sampling/prob_max": 6.160250341054052e-05,
"sampling/prob_min": 2.0253971888450907e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9752000093460083,
"sampling/prompt_draws_total": 59256.0,
"sampling/seen_fraction": 0.9370333194732666,
"sampling/unseen_fraction": 0.0629666805267334,
"signal/accuracy_reward/centered_abs_mean": 0.11440972238779068,
"signal/accuracy_reward/group_std_mean": 0.15993244349956512,
"signal/accuracy_reward/group_zero_std_frac": 0.5111111164093017,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05720486119389534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05720486119389534,
"signal/advantage_abs_mean": 0.084007428586483,
"signal/advantage_pre_scale_abs_mean": 0.084007428586483,
"signal/advantage_pre_scale_std": 0.17651504874229432,
"signal/advantage_std": 0.17651504874229432,
"signal/brier_reward/centered_abs_mean": 0.0703900247812271,
"signal/brier_reward/group_std_mean": 0.10124405175447464,
"signal/brier_reward/group_zero_std_frac": 0.21666666567325593,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03519501239061355,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03519501239061355,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0020128038129769267,
"signal/confidence_one_or_zero/group_std_mean": 0.0033502712845802306,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.0128037192534975e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.0128037192534975e-08,
"signal/format_reward/centered_abs_mean": 0.01800130195915699,
"signal/format_reward/group_std_mean": 0.04087369553744793,
"signal/format_reward/group_zero_std_frac": 0.8055555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009000650979578495,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009000650979578495,
"signal/mean_confidence_reward/centered_abs_mean": 0.06586859598755837,
"signal/mean_confidence_reward/group_std_mean": 0.09096481949090958,
"signal/mean_confidence_reward/group_zero_std_frac": 0.23055555522441865,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.586859740309591e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.586859740309591e-07,
"step": 825
},
{
"calibration/aurc": 0.05599930548950739,
"calibration/batch_distribution_entropy": 0.596382853068725,
"calibration/batch_entropy_100bins": 0.3233534373407135,
"calibration/batch_entropy_10bins": 0.596382853068725,
"calibration/batch_entropy_50bins": 0.3806464346227048,
"calibration/batch_uniqueness": 0.3026821286182817,
"calibration/confidence_entropy": 0.44498857508984246,
"calibration/coverage@0%": 0.3928695921741526,
"calibration/coverage@1%": 0.5114820990811839,
"calibration/coverage@10%": 0.7170144628173889,
"calibration/coverage@15%": 0.8434561514934973,
"calibration/coverage@20%": 0.9599111234550757,
"calibration/coverage@25%": 0.9810026385224274,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.633519389230383,
"calibration/distribution_entropy_10": 0.596382853068725,
"calibration/distribution_entropy_100": 0.3233534373407135,
"calibration/ece": 0.11589207315098178,
"calibration/mean_confidence": 0.7559549854800441,
"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": 0.009288194444444443,
"completions/max_length": 3944.6,
"completions/max_terminated_length": 3944.6,
"completions/mean_length": 1123.5111328125,
"completions/mean_terminated_length": 1134.081689453125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 343.4,
"epoch": 1.9951923076923077,
"grad_norm": 0.00034137690090574324,
"learning_rate": 1.2620192307692309e-06,
"loss": -0.0091,
"num_tokens": 2239053249.0,
"reward": 1.3294086456298828,
"reward_std": 0.11628881245851516,
"rewards/accuracy_reward": 0.7828125,
"rewards/brier_reward": 0.8852781653404236,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9907118201255798,
"rewards/mean_confidence_reward": 0.7438917636871338,
"sampling/batch_mean_priority_error": 0.015333333333333319,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.11111111111111112,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0029324951581656935,
"sampling/priority_kl": 0.029998801276087762,
"sampling/priority_scale": 0.8383028804557398,
"sampling/prob_entropy": 10.278944396972657,
"sampling/prob_max": 6.189564446685835e-05,
"sampling/prob_min": 2.0265720377210527e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9871999740600585,
"sampling/prompt_draws_total": 59616.0,
"sampling/seen_fraction": 0.9384733319282532,
"sampling/unseen_fraction": 0.06152666807174682,
"signal/accuracy_reward/centered_abs_mean": 0.10640190988779068,
"signal/accuracy_reward/group_std_mean": 0.1437681049108505,
"signal/accuracy_reward/group_zero_std_frac": 0.5777777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05320095494389534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05320095494389534,
"signal/advantage_abs_mean": 0.07790218591690064,
"signal/advantage_pre_scale_abs_mean": 0.07790218591690064,
"signal/advantage_pre_scale_std": 0.17275477647781373,
"signal/advantage_std": 0.17275477647781373,
"signal/brier_reward/centered_abs_mean": 0.06396772861480712,
"signal/brier_reward/group_std_mean": 0.09234273433685303,
"signal/brier_reward/group_zero_std_frac": 0.3000000059604645,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03198386430740356,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03198386430740356,
"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": 0.01687825508415699,
"signal/format_reward/group_std_mean": 0.038504977524280545,
"signal/format_reward/group_zero_std_frac": 0.8166666865348816,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008439127542078494,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008439127542078494,
"signal/mean_confidence_reward/centered_abs_mean": 0.05897117778658867,
"signal/mean_confidence_reward/group_std_mean": 0.08022859096527099,
"signal/mean_confidence_reward/group_zero_std_frac": 0.30555556416511537,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.897117375752714e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.897117375752714e-07,
"step": 830
},
{
"calibration/aurc": 0.09532195250897704,
"calibration/batch_distribution_entropy": 0.5789066551694979,
"calibration/batch_entropy_100bins": 0.3114747096174578,
"calibration/batch_entropy_10bins": 0.5789066551694979,
"calibration/batch_entropy_50bins": 0.36666298854309254,
"calibration/batch_uniqueness": 0.28124029620299623,
"calibration/confidence_entropy": 0.4461880173219065,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.09018567639257294,
"calibration/coverage@10%": 0.6630341494576916,
"calibration/coverage@15%": 0.7041357090282889,
"calibration/coverage@20%": 0.7343744358187398,
"calibration/coverage@25%": 0.9196472857536511,
"calibration/coverage@30%": 0.954282171397079,
"calibration/coverage@5%": 0.5344841473745607,
"calibration/distribution_entropy_10": 0.5789066551694979,
"calibration/distribution_entropy_100": 0.3114747096174578,
"calibration/ece": 0.10387739165931135,
"calibration/mean_confidence": 0.7439021672650166,
"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": 0.00581597222222221,
"completions/max_length": 3222.0,
"completions/max_terminated_length": 3222.0,
"completions/mean_length": 1013.2931518554688,
"completions/mean_terminated_length": 1019.6262084960938,
"completions/min_length": 109.0,
"completions/min_terminated_length": 390.0,
"epoch": 2.0072115384615383,
"grad_norm": 0.00037373026134446263,
"learning_rate": 1.231971153846154e-06,
"loss": -0.0107,
"num_tokens": 2254636269.0,
"reward": 1.3175593137741088,
"reward_std": 0.11637879014015198,
"rewards/accuracy_reward": 0.7723090291023255,
"rewards/brier_reward": 0.8709544539451599,
"rewards/confidence_one_or_zero": 0.0003472222248092294,
"rewards/format_reward": 0.9918402791023254,
"rewards/mean_confidence_reward": 0.7433246612548828,
"sampling/batch_mean_priority_error": 0.018690972222222206,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.09722222222222224,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0029484704602509735,
"sampling/priority_kl": 0.030000633746385574,
"sampling/priority_scale": 0.8412145436042919,
"sampling/prob_entropy": 10.278964805603028,
"sampling/prob_max": 6.217749323695898e-05,
"sampling/prob_min": 2.027921473199967e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9991999864578247,
"sampling/prompt_draws_total": 59976.0,
"sampling/seen_fraction": 0.9398399949073791,
"sampling/unseen_fraction": 0.06016000509262085,
"signal/accuracy_reward/centered_abs_mean": 0.1052788645029068,
"signal/accuracy_reward/group_std_mean": 0.14525774121284485,
"signal/accuracy_reward/group_zero_std_frac": 0.5583333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0526394322514534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0526394322514534,
"signal/advantage_abs_mean": 0.07736062556505204,
"signal/advantage_pre_scale_abs_mean": 0.07736062556505204,
"signal/advantage_pre_scale_std": 0.17310604453086853,
"signal/advantage_std": 0.17310604453086853,
"signal/brier_reward/centered_abs_mean": 0.06680887416005135,
"signal/brier_reward/group_std_mean": 0.09623768329620361,
"signal/brier_reward/group_zero_std_frac": 0.2916666716337204,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.033404437080025676,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.033404437080025676,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0006076388992369175,
"signal/confidence_one_or_zero/group_std_mean": 0.0009333631955087185,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 6.076388103792851e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 6.076388103792851e-09,
"signal/format_reward/centered_abs_mean": 0.01511501744389534,
"signal/format_reward/group_std_mean": 0.03533230200409889,
"signal/format_reward/group_zero_std_frac": 0.8277777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00755750872194767,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00755750872194767,
"signal/mean_confidence_reward/centered_abs_mean": 0.055697161704301834,
"signal/mean_confidence_reward/group_std_mean": 0.07788751721382141,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3083333373069763,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.569716222453281e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.569716222453281e-07,
"step": 835
},
{
"calibration/aurc": 0.14482306548752338,
"calibration/batch_distribution_entropy": 0.600104045211634,
"calibration/batch_entropy_100bins": 0.3210407699158253,
"calibration/batch_entropy_10bins": 0.600104045211634,
"calibration/batch_entropy_50bins": 0.3779239999487719,
"calibration/batch_uniqueness": 0.3226214688891113,
"calibration/confidence_entropy": 0.45247366041500064,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.3567177980852915,
"calibration/coverage@15%": 0.5672233964662163,
"calibration/coverage@20%": 0.6812894580518601,
"calibration/coverage@25%": 0.8855333139928441,
"calibration/coverage@30%": 0.9259632254410322,
"calibration/coverage@5%": 0.140625,
"calibration/distribution_entropy_10": 0.600104045211634,
"calibration/distribution_entropy_100": 0.3210407699158253,
"calibration/ece": 0.08676930940638516,
"calibration/mean_confidence": 0.758886644010665,
"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": 0.006076388888888884,
"completions/max_length": 3938.6,
"completions/max_terminated_length": 3938.6,
"completions/mean_length": 1111.3046875,
"completions/mean_terminated_length": 1118.1972412109376,
"completions/min_length": 0.0,
"completions/min_terminated_length": 314.0,
"epoch": 2.019230769230769,
"grad_norm": 0.0003290143795311451,
"learning_rate": 1.201923076923077e-06,
"loss": -0.0074,
"num_tokens": 2270521187.0,
"reward": 1.3089183807373046,
"reward_std": 0.11291085630655288,
"rewards/accuracy_reward": 0.7555555462837219,
"rewards/brier_reward": 0.8684296846389771,
"rewards/confidence_one_or_zero": 0.0010416666569653898,
"rewards/format_reward": 0.9938368082046509,
"rewards/mean_confidence_reward": 0.7367968678474426,
"sampling/batch_mean_priority_error": 0.018819444444444423,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.09166666666666667,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0029684344306588175,
"sampling/priority_kl": 0.030000444129109384,
"sampling/priority_scale": 0.8429635942680761,
"sampling/prob_entropy": 10.27894229888916,
"sampling/prob_max": 6.239795475266874e-05,
"sampling/prob_min": 2.030375944741536e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.0112000465393067,
"sampling/prompt_draws_total": 60336.0,
"sampling/seen_fraction": 0.9408199906349182,
"sampling/unseen_fraction": 0.05918000936508179,
"signal/accuracy_reward/centered_abs_mean": 0.10628255158662796,
"signal/accuracy_reward/group_std_mean": 0.14591727107763292,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111164093018,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05314127579331398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05314127579331398,
"signal/advantage_abs_mean": 0.07720207050442696,
"signal/advantage_pre_scale_abs_mean": 0.07720207050442696,
"signal/advantage_pre_scale_std": 0.16867663860321044,
"signal/advantage_std": 0.16867663860321044,
"signal/brier_reward/centered_abs_mean": 0.06350096985697747,
"signal/brier_reward/group_std_mean": 0.09030406326055526,
"signal/brier_reward/group_zero_std_frac": 0.28888889253139494,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03175048492848873,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03175048492848873,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0014214409398846327,
"signal/confidence_one_or_zero/group_std_mean": 0.0018314871937036513,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.4214408849966275e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.4214408849966275e-08,
"signal/format_reward/centered_abs_mean": 0.011496310774236917,
"signal/format_reward/group_std_mean": 0.026431189477443696,
"signal/format_reward/group_zero_std_frac": 0.875,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005748155387118458,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005748155387118458,
"signal/mean_confidence_reward/centered_abs_mean": 0.058305129408836365,
"signal/mean_confidence_reward/group_std_mean": 0.0791845440864563,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3000000029802322,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.830512634474872e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.830512634474872e-07,
"step": 840
},
{
"calibration/aurc": 0.06706975557753085,
"calibration/batch_distribution_entropy": 0.5721016540996581,
"calibration/batch_entropy_100bins": 0.3115216554832395,
"calibration/batch_entropy_10bins": 0.5721016540996581,
"calibration/batch_entropy_50bins": 0.36671825245671313,
"calibration/batch_uniqueness": 0.23765986424922766,
"calibration/confidence_entropy": 0.4423131495801164,
"calibration/coverage@0%": 0.20417194819509507,
"calibration/coverage@1%": 0.3680449049909621,
"calibration/coverage@10%": 0.6030081601539127,
"calibration/coverage@15%": 0.906186369660654,
"calibration/coverage@20%": 0.9657121772245313,
"calibration/coverage@25%": 1.0,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.4381744026348315,
"calibration/distribution_entropy_10": 0.5721016540996581,
"calibration/distribution_entropy_100": 0.3115216554832395,
"calibration/ece": 0.1172741215185497,
"calibration/mean_confidence": 0.7487478732080494,
"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": 0.008940972222222232,
"completions/max_length": 3910.4,
"completions/max_terminated_length": 3910.4,
"completions/mean_length": 1185.9043212890624,
"completions/mean_terminated_length": 1196.602978515625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 345.0,
"epoch": 2.03125,
"grad_norm": 0.0003298562951385975,
"learning_rate": 1.1718750000000001e-06,
"loss": -0.0107,
"num_tokens": 2287292437.0,
"reward": 1.2747956275939942,
"reward_std": 0.11422278136014938,
"rewards/accuracy_reward": 0.7032986044883728,
"rewards/brier_reward": 0.8552193760871887,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9910590291023255,
"rewards/mean_confidence_reward": 0.7114800333976745,
"sampling/batch_mean_priority_error": 0.022694444444444427,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08055555555555556,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.002991753304377198,
"sampling/priority_kl": 0.029999999701976775,
"sampling/priority_scale": 0.8449041426414624,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 6.262821407290175e-05,
"sampling/prob_min": 2.0326085723354483e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.023200035095215,
"sampling/prompt_draws_total": 60696.0,
"sampling/seen_fraction": 0.941866660118103,
"sampling/unseen_fraction": 0.05813333988189697,
"signal/accuracy_reward/centered_abs_mean": 0.1039605051279068,
"signal/accuracy_reward/group_std_mean": 0.1419745147228241,
"signal/accuracy_reward/group_zero_std_frac": 0.5805555701255798,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0519802525639534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0519802525639534,
"signal/advantage_abs_mean": 0.0762131504714489,
"signal/advantage_pre_scale_abs_mean": 0.0762131504714489,
"signal/advantage_pre_scale_std": 0.16904608607292176,
"signal/advantage_std": 0.16904608607292176,
"signal/brier_reward/centered_abs_mean": 0.06413592025637627,
"signal/brier_reward/group_std_mean": 0.0929721713066101,
"signal/brier_reward/group_zero_std_frac": 0.2583333313465118,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03206796012818813,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03206796012818813,
"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": 0.016704644449055193,
"signal/format_reward/group_std_mean": 0.0384040080010891,
"signal/format_reward/group_zero_std_frac": 0.8194444656372071,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008352322224527597,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008352322224527597,
"signal/mean_confidence_reward/centered_abs_mean": 0.06115180775523186,
"signal/mean_confidence_reward/group_std_mean": 0.08378626108169555,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2611111104488373,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.115180099186545e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.115180099186545e-07,
"step": 845
},
{
"calibration/aurc": 0.09126816611407072,
"calibration/batch_distribution_entropy": 0.516295811065282,
"calibration/batch_entropy_100bins": 0.27987268448332003,
"calibration/batch_entropy_10bins": 0.516295811065282,
"calibration/batch_entropy_50bins": 0.32946159587166834,
"calibration/batch_uniqueness": 0.10043751990069968,
"calibration/confidence_entropy": 0.4204486071838677,
"calibration/coverage@0%": 0.1371714221667972,
"calibration/coverage@1%": 0.2718972707307659,
"calibration/coverage@10%": 0.6968140041355322,
"calibration/coverage@15%": 0.7426602756256365,
"calibration/coverage@20%": 0.7659504662980957,
"calibration/coverage@25%": 0.8975470486401967,
"calibration/coverage@30%": 0.9748031496062992,
"calibration/coverage@5%": 0.5180043382809688,
"calibration/distribution_entropy_10": 0.516295811065282,
"calibration/distribution_entropy_100": 0.27987268448332003,
"calibration/ece": 0.09604928174517917,
"calibration/mean_confidence": 0.7789018314999862,
"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": 0.010763888888888906,
"completions/max_length": 3898.6,
"completions/max_terminated_length": 3898.6,
"completions/mean_length": 1155.826318359375,
"completions/mean_terminated_length": 1168.4384765625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 356.0,
"epoch": 2.043269230769231,
"grad_norm": 0.00034717281232587993,
"learning_rate": 1.141826923076923e-06,
"loss": -0.0113,
"num_tokens": 2303699396.0,
"reward": 1.2955496788024903,
"reward_std": 0.10840608775615693,
"rewards/accuracy_reward": 0.73359375,
"rewards/brier_reward": 0.8682549834251404,
"rewards/confidence_one_or_zero": 0.000260416668606922,
"rewards/format_reward": 0.9892361164093018,
"rewards/mean_confidence_reward": 0.7277135252952576,
"sampling/batch_mean_priority_error": 0.015092013888888877,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08888888888888888,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0030097827315330506,
"sampling/priority_kl": 0.030000165477395056,
"sampling/priority_scale": 0.8467580139869824,
"sampling/prob_entropy": 10.278955841064453,
"sampling/prob_max": 6.285476847551763e-05,
"sampling/prob_min": 2.0349307305878028e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.035200023651123,
"sampling/prompt_draws_total": 61056.0,
"sampling/seen_fraction": 0.9428733229637146,
"sampling/unseen_fraction": 0.0571266770362854,
"signal/accuracy_reward/centered_abs_mean": 0.08740776926279067,
"signal/accuracy_reward/group_std_mean": 0.12122353315353393,
"signal/accuracy_reward/group_zero_std_frac": 0.6277777791023255,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04370388463139534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04370388463139534,
"signal/advantage_abs_mean": 0.071315798163414,
"signal/advantage_pre_scale_abs_mean": 0.071315798163414,
"signal/advantage_pre_scale_std": 0.1623822569847107,
"signal/advantage_std": 0.1623822569847107,
"signal/brier_reward/centered_abs_mean": 0.06464180201292039,
"signal/brier_reward/group_std_mean": 0.09374651610851288,
"signal/brier_reward/group_zero_std_frac": 0.2611111134290695,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03232090100646019,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03232090100646019,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0005045572877861559,
"signal/confidence_one_or_zero/group_std_mean": 0.0014731390401721,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/format_reward/centered_abs_mean": 0.01946614608168602,
"signal/format_reward/group_std_mean": 0.04095918200910091,
"signal/format_reward/group_zero_std_frac": 0.8166666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00973307304084301,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00973307304084301,
"signal/mean_confidence_reward/centered_abs_mean": 0.05734120234847069,
"signal/mean_confidence_reward/group_std_mean": 0.0792054533958435,
"signal/mean_confidence_reward/group_zero_std_frac": 0.27500000298023225,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.7341202364114e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.7341202364114e-07,
"step": 850
},
{
"epoch": 2.043269230769231,
"eval_calibration/aurc": 0.12376605829554335,
"eval_calibration/batch_distribution_entropy": 0.6170005564464504,
"eval_calibration/batch_entropy_100bins": 0.33893845519545,
"eval_calibration/batch_entropy_10bins": 0.6170005564464504,
"eval_calibration/batch_entropy_50bins": 0.3989928654778244,
"eval_calibration/batch_uniqueness": 0.327932433642645,
"eval_calibration/confidence_entropy": 0.4474724073814369,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.5520559930008749,
"eval_calibration/coverage@15%": 0.7637795275590551,
"eval_calibration/coverage@20%": 0.8162729658792651,
"eval_calibration/coverage@25%": 0.9588801399825022,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.6170005564464504,
"eval_calibration/distribution_entropy_100": 0.33893845519545,
"eval_calibration/ece": 0.025065616797900182,
"eval_calibration/mean_confidence": 0.7349518810148731,
"eval_calibration/unique_confidence_per_question": 0.008680555555555556,
"eval_calibration/unique_confidences": 10,
"eval_completions/clipped_ratio": 0.011111111111111127,
"eval_completions/max_length": 3113.3333333333335,
"eval_completions/max_terminated_length": 3113.3333333333335,
"eval_completions/mean_length": 1146.4441324869792,
"eval_completions/mean_terminated_length": 1159.1681518554688,
"eval_completions/min_length": 0.0,
"eval_completions/min_terminated_length": 466.6666666666667,
"eval_loss": 0.0,
"eval_num_tokens": 2303699396.0,
"eval_reward": 1.2894940972328186,
"eval_reward_std": 0.31144089500109357,
"eval_rewards/accuracy_reward": 0.725694457689921,
"eval_rewards/brier_reward": 0.8610915839672089,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9921875099341074,
"eval_rewards/mean_confidence_reward": 0.7292100787162781,
"eval_runtime": 213.4681,
"eval_samples_per_second": 4.685,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3849826355775197,
"eval_signal/accuracy_reward/group_std_mean": 0.4438539495070775,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19249131778875986,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19249131778875986,
"eval_signal/advantage_abs_mean": 0.2564364845554034,
"eval_signal/advantage_pre_scale_abs_mean": 0.2564364845554034,
"eval_signal/advantage_pre_scale_std": 0.3103178143501282,
"eval_signal/advantage_std": 0.3103178143501282,
"eval_signal/brier_reward/centered_abs_mean": 0.1578616127371788,
"eval_signal/brier_reward/group_std_mean": 0.22026897221803665,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0789308063685894,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.0789308063685894,
"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": 0.015028211598594984,
"eval_signal/format_reward/group_std_mean": 0.041204764818151794,
"eval_signal/format_reward/group_zero_std_frac": 0.7777778108914694,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.007514105799297492,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.007514105799297492,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.1957817698518435,
"eval_signal/mean_confidence_reward/group_std_mean": 0.23651453604300818,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.9578177254212883e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.9578177254212883e-06,
"eval_steps_per_second": 0.028,
"step": 850
},
{
"epoch": 2.043269230769231,
"step": 850,
"train_probe_calibration/aurc": 0.07444353586731828,
"train_probe_calibration/batch_distribution_entropy": 0.5981090318149916,
"train_probe_calibration/batch_entropy_100bins": 0.3278047570973816,
"train_probe_calibration/batch_entropy_10bins": 0.5981090318149916,
"train_probe_calibration/batch_entropy_50bins": 0.3858864562185043,
"train_probe_calibration/batch_uniqueness": 0.28846979706331144,
"train_probe_calibration/confidence_entropy": 0.4422734115458898,
"train_probe_calibration/coverage@0%": 0.002617801047120419,
"train_probe_calibration/coverage@1%": 0.002617801047120419,
"train_probe_calibration/coverage@10%": 0.7853403141361257,
"train_probe_calibration/coverage@15%": 0.8359511343804538,
"train_probe_calibration/coverage@20%": 0.9720767888307156,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.5732984293193717,
"train_probe_calibration/distribution_entropy_10": 0.5981090318149916,
"train_probe_calibration/distribution_entropy_100": 0.3278047570973816,
"train_probe_calibration/ece": 0.043935427574170946,
"train_probe_calibration/mean_confidence": 0.7504799301919721,
"train_probe_calibration/unique_confidence_per_question": 0.009548611111111112,
"train_probe_calibration/unique_confidences": 11,
"train_probe_completions/clipped_ratio": 0.00434027777777779,
"train_probe_completions/max_length": 3625.8333333333335,
"train_probe_completions/max_terminated_length": 3625.8333333333335,
"train_probe_completions/mean_length": 1171.7156575520833,
"train_probe_completions/mean_terminated_length": 1176.8045654296875,
"train_probe_completions/min_length": 142.5,
"train_probe_completions/min_terminated_length": 381.3333333333333,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 2303699396.0,
"train_probe_reward": 1.3320420583089192,
"train_probe_reward_std": 0.2776293953259786,
"train_probe_rewards/accuracy_reward": 0.7838541666666666,
"train_probe_rewards/brier_reward": 0.8854231834411621,
"train_probe_rewards/confidence_one_or_zero": 0.0017361111628512542,
"train_probe_rewards/format_reward": 0.9947916766007742,
"train_probe_rewards/mean_confidence_reward": 0.7465711534023285,
"train_probe_runtime": 208.723,
"train_probe_samples_per_second": 4.791,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3287217865387599,
"train_probe_signal/accuracy_reward/group_std_mean": 0.40822718540827435,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.16436089326937994,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.16436089326937994,
"train_probe_signal/advantage_abs_mean": 0.216099684437116,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.216099684437116,
"train_probe_signal/advantage_pre_scale_std": 0.27747392157713574,
"train_probe_signal/advantage_std": 0.27747392157713574,
"train_probe_signal/brier_reward/centered_abs_mean": 0.13258452092607817,
"train_probe_signal/brier_reward/group_std_mean": 0.18689996004104614,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.06629226046303908,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.06629226046303908,
"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": 0.1832817792892456,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.22317108511924744,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.832817683104319e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.832817683104319e-06,
"train_probe_steps_per_second": 0.029
},
{
"calibration/aurc": 0.09007888032879449,
"calibration/batch_distribution_entropy": 0.6190782522581999,
"calibration/batch_entropy_100bins": 0.339766442064793,
"calibration/batch_entropy_10bins": 0.6190782522581999,
"calibration/batch_entropy_50bins": 0.3999675582236997,
"calibration/batch_uniqueness": 0.3461391671099906,
"calibration/confidence_entropy": 0.45763288876498853,
"calibration/coverage@0%": 0.11989392096017934,
"calibration/coverage@1%": 0.15130753352562437,
"calibration/coverage@10%": 0.7290703941733264,
"calibration/coverage@15%": 0.7876857827380854,
"calibration/coverage@20%": 0.84802688708862,
"calibration/coverage@25%": 0.8917965838078045,
"calibration/coverage@30%": 0.9583613387029579,
"calibration/coverage@5%": 0.5147525764057768,
"calibration/distribution_entropy_10": 0.6190782522581999,
"calibration/distribution_entropy_100": 0.339766442064793,
"calibration/ece": 0.106227240554117,
"calibration/mean_confidence": 0.7179123295037517,
"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": 0.009895833333333326,
"completions/max_length": 4016.8,
"completions/max_terminated_length": 4016.8,
"completions/mean_length": 1155.4071044921875,
"completions/mean_terminated_length": 1166.94560546875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 386.6,
"epoch": 2.0552884615384617,
"grad_norm": 0.00033060292480513453,
"learning_rate": 1.1117788461538462e-06,
"loss": -0.0106,
"num_tokens": 2320095190.0,
"reward": 1.322322916984558,
"reward_std": 0.12814645171165467,
"rewards/accuracy_reward": 0.778125,
"rewards/brier_reward": 0.8764886498451233,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9900173544883728,
"rewards/mean_confidence_reward": 0.7386140108108521,
"sampling/batch_mean_priority_error": 0.014572916666666647,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.09444444444444446,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.003024717140942812,
"sampling/priority_kl": 0.029999865218997,
"sampling/priority_scale": 0.8490281641716138,
"sampling/prob_entropy": 10.278955459594727,
"sampling/prob_max": 6.310095050139352e-05,
"sampling/prob_min": 2.036733421846293e-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": 0.11383463591337203,
"signal/accuracy_reward/group_std_mean": 0.158145971596241,
"signal/accuracy_reward/group_zero_std_frac": 0.5111111104488373,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05691731795668602,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05691731795668602,
"signal/advantage_abs_mean": 0.08426995873451233,
"signal/advantage_pre_scale_abs_mean": 0.08426995873451233,
"signal/advantage_pre_scale_std": 0.18189284801483155,
"signal/advantage_std": 0.18189284801483155,
"signal/brier_reward/centered_abs_mean": 0.07048432901501656,
"signal/brier_reward/group_std_mean": 0.10308757573366165,
"signal/brier_reward/group_zero_std_frac": 0.24722222089767457,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03524216450750828,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03524216450750828,
"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": 0.018017578125,
"signal/format_reward/group_std_mean": 0.04189658388495445,
"signal/format_reward/group_zero_std_frac": 0.7972222208976746,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0090087890625,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0090087890625,
"signal/mean_confidence_reward/centered_abs_mean": 0.0627426914870739,
"signal/mean_confidence_reward/group_std_mean": 0.08627793490886689,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2555555522441864,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.274268912420666e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.274268912420666e-07,
"step": 855
},
{
"calibration/aurc": 0.06585142036589152,
"calibration/batch_distribution_entropy": 0.5993146527254817,
"calibration/batch_entropy_100bins": 0.3219695299662035,
"calibration/batch_entropy_10bins": 0.5993146527254817,
"calibration/batch_entropy_50bins": 0.37901732125286536,
"calibration/batch_uniqueness": 0.3283448677885148,
"calibration/confidence_entropy": 0.4534555868322087,
"calibration/coverage@0%": 0.0020833333333333333,
"calibration/coverage@1%": 0.21415682414698164,
"calibration/coverage@10%": 0.7858256473448325,
"calibration/coverage@15%": 0.8751886482939633,
"calibration/coverage@20%": 0.9480314960629922,
"calibration/coverage@25%": 0.9585301837270341,
"calibration/coverage@30%": 0.9826771653543307,
"calibration/coverage@5%": 0.509247578346114,
"calibration/distribution_entropy_10": 0.5993146527254817,
"calibration/distribution_entropy_100": 0.3219695299662035,
"calibration/ece": 0.10910423112703256,
"calibration/mean_confidence": 0.7536399901403481,
"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": 0.013454861111111115,
"completions/max_length": 3936.8,
"completions/max_terminated_length": 3936.8,
"completions/mean_length": 1172.1067626953125,
"completions/mean_terminated_length": 1188.252099609375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 399.0,
"epoch": 2.0673076923076925,
"grad_norm": 0.00032517011277377605,
"learning_rate": 1.0817307692307693e-06,
"loss": -0.0157,
"num_tokens": 2336673156.0,
"reward": 1.2997122764587403,
"reward_std": 0.1217604398727417,
"rewards/accuracy_reward": 0.753125011920929,
"rewards/brier_reward": 0.8597398161888122,
"rewards/confidence_one_or_zero": 0.0010416667035315185,
"rewards/format_reward": 0.9865451455116272,
"rewards/mean_confidence_reward": 0.7358984351158142,
"sampling/batch_mean_priority_error": 0.01943229166666666,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08888888888888888,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.0030421236995607616,
"sampling/priority_kl": 0.030000335723161697,
"sampling/priority_scale": 0.8511217415565625,
"sampling/prob_entropy": 10.27894630432129,
"sampling/prob_max": 6.33396688499488e-05,
"sampling/prob_min": 2.038745406025555e-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": 0.1017795130610466,
"signal/accuracy_reward/group_std_mean": 0.14184551537036896,
"signal/accuracy_reward/group_zero_std_frac": 0.5638888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0508897565305233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0508897565305233,
"signal/advantage_abs_mean": 0.08031857311725617,
"signal/advantage_pre_scale_abs_mean": 0.08031857311725617,
"signal/advantage_pre_scale_std": 0.18018272817134856,
"signal/advantage_std": 0.18018272817134856,
"signal/brier_reward/centered_abs_mean": 0.06887489780783654,
"signal/brier_reward/group_std_mean": 0.09927848726511002,
"signal/brier_reward/group_zero_std_frac": 0.2638888955116272,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03443744890391827,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03443744890391827,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0019314235891215504,
"signal/confidence_one_or_zero/group_std_mean": 0.004263852536678314,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9805555462837219,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.9314235188971908e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.9314235188971908e-08,
"signal/format_reward/centered_abs_mean": 0.02167426198720932,
"signal/format_reward/group_std_mean": 0.04529851377010345,
"signal/format_reward/group_zero_std_frac": 0.7916666746139527,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01083713099360466,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01083713099360466,
"signal/mean_confidence_reward/centered_abs_mean": 0.0583306223154068,
"signal/mean_confidence_reward/group_std_mean": 0.08277992159128189,
"signal/mean_confidence_reward/group_zero_std_frac": 0.27777778506278994,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.833062118654198e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.833062118654198e-07,
"step": 860
},
{
"calibration/aurc": 0.11603837439787193,
"calibration/batch_distribution_entropy": 0.6149790592596738,
"calibration/batch_entropy_100bins": 0.3336354259414588,
"calibration/batch_entropy_10bins": 0.6149790592596738,
"calibration/batch_entropy_50bins": 0.3927502252423204,
"calibration/batch_uniqueness": 0.3461293343834476,
"calibration/confidence_entropy": 0.45388902139774084,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.6761772873094903,
"calibration/coverage@15%": 0.7397942336205208,
"calibration/coverage@20%": 0.8286319299555995,
"calibration/coverage@25%": 0.8827070729755906,
"calibration/coverage@30%": 0.9983914209115282,
"calibration/coverage@5%": 0.12650918635170602,
"calibration/distribution_entropy_10": 0.6149790592596738,
"calibration/distribution_entropy_100": 0.3336354259414588,
"calibration/ece": 0.1132342967072548,
"calibration/mean_confidence": 0.7376037727930024,
"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": 0.011805555555555559,
"completions/max_length": 3987.8,
"completions/max_terminated_length": 3987.8,
"completions/mean_length": 1180.26572265625,
"completions/mean_terminated_length": 1194.537158203125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 374.2,
"epoch": 2.079326923076923,
"grad_norm": 0.00033040158450603485,
"learning_rate": 1.0516826923076925e-06,
"loss": -0.0155,
"num_tokens": 2353379865.0,
"reward": 1.3009061336517334,
"reward_std": 0.1203909158706665,
"rewards/accuracy_reward": 0.7534722208976745,
"rewards/brier_reward": 0.8601308584213256,
"rewards/confidence_one_or_zero": 0.001128472265554592,
"rewards/format_reward": 0.9881944417953491,
"rewards/mean_confidence_reward": 0.7362447857856751,
"sampling/batch_mean_priority_error": 0.019236180555555545,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08888888888888888,
"sampling/error_ema_max": 0.17066249251365662,
"sampling/error_ema_mean": 0.003059811145067215,
"sampling/priority_kl": 0.030000845715403558,
"sampling/priority_scale": 0.853543382906355,
"sampling/prob_entropy": 10.278954887390137,
"sampling/prob_max": 6.359893741318956e-05,
"sampling/prob_min": 2.040498839050997e-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": 0.09594184011220933,
"signal/accuracy_reward/group_std_mean": 0.13271133005619049,
"signal/accuracy_reward/group_zero_std_frac": 0.5944444537162781,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04797092005610466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04797092005610466,
"signal/advantage_abs_mean": 0.07722088620066643,
"signal/advantage_pre_scale_abs_mean": 0.07722088620066643,
"signal/advantage_pre_scale_std": 0.17718519866466523,
"signal/advantage_std": 0.17718519866466523,
"signal/brier_reward/centered_abs_mean": 0.06708870679140091,
"signal/brier_reward/group_std_mean": 0.09765137284994126,
"signal/brier_reward/group_zero_std_frac": 0.2750000059604645,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03354435339570046,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03354435339570046,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0019802517374046145,
"signal/confidence_one_or_zero/group_std_mean": 0.0035894179251044988,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.98025162490012e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.98025162490012e-08,
"signal/format_reward/centered_abs_mean": 0.02185329906642437,
"signal/format_reward/group_std_mean": 0.048659897595644,
"signal/format_reward/group_zero_std_frac": 0.774999988079071,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010926649533212185,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010926649533212185,
"signal/mean_confidence_reward/centered_abs_mean": 0.060614097863435745,
"signal/mean_confidence_reward/group_std_mean": 0.08568830490112304,
"signal/mean_confidence_reward/group_zero_std_frac": 0.27777777910232543,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.061409635549353e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.061409635549353e-07,
"step": 865
},
{
"calibration/aurc": 0.0948379996016497,
"calibration/batch_distribution_entropy": 0.5135562906550475,
"calibration/batch_entropy_100bins": 0.2770817450004746,
"calibration/batch_entropy_10bins": 0.5135562906550475,
"calibration/batch_entropy_50bins": 0.32617614706948517,
"calibration/batch_uniqueness": 0.10571137929105578,
"calibration/confidence_entropy": 0.421645673634799,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.24772940203912927,
"calibration/coverage@10%": 0.6397247472177792,
"calibration/coverage@15%": 0.698163496533323,
"calibration/coverage@20%": 0.7657894736842106,
"calibration/coverage@25%": 0.9245362099766388,
"calibration/coverage@30%": 0.9586684073107049,
"calibration/coverage@5%": 0.577195633961183,
"calibration/distribution_entropy_10": 0.5135562906550475,
"calibration/distribution_entropy_100": 0.2770817450004746,
"calibration/ece": 0.12980076800239265,
"calibration/mean_confidence": 0.7757306178170007,
"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": 0.01137152777777779,
"completions/max_length": 4021.6,
"completions/max_terminated_length": 4021.6,
"completions/mean_length": 1178.4822265625,
"completions/mean_terminated_length": 1192.223583984375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 308.0,
"epoch": 2.0913461538461537,
"grad_norm": 0.0003403700247872621,
"learning_rate": 1.0216346153846154e-06,
"loss": -0.0138,
"num_tokens": 2370032364.0,
"reward": 1.2729515790939332,
"reward_std": 0.11964078694581985,
"rewards/accuracy_reward": 0.7024305582046508,
"rewards/brier_reward": 0.8548296451568603,
"rewards/confidence_one_or_zero": 0.0004340277810115367,
"rewards/format_reward": 0.9886284828186035,
"rewards/mean_confidence_reward": 0.7271978974342346,
"sampling/batch_mean_priority_error": 0.025902777777777757,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.0861111111111111,
"sampling/error_ema_max": 0.1780436158180237,
"sampling/error_ema_mean": 0.00308342189528048,
"sampling/priority_kl": 0.030000101029872894,
"sampling/priority_scale": 0.856144517636858,
"sampling/prob_entropy": 10.278953170776367,
"sampling/prob_max": 6.386785244103522e-05,
"sampling/prob_min": 2.0420402506715618e-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": 0.1059787318110466,
"signal/accuracy_reward/group_std_mean": 0.14246926009654998,
"signal/accuracy_reward/group_zero_std_frac": 0.5777777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0529893659055233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0529893659055233,
"signal/advantage_abs_mean": 0.07931408137083054,
"signal/advantage_pre_scale_abs_mean": 0.07931408137083054,
"signal/advantage_pre_scale_std": 0.17501305937767028,
"signal/advantage_std": 0.17501305937767028,
"signal/brier_reward/centered_abs_mean": 0.06485896408557892,
"signal/brier_reward/group_std_mean": 0.09463429301977158,
"signal/brier_reward/group_zero_std_frac": 0.311111119389534,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03242948204278946,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03242948204278946,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0007758246618323028,
"signal/confidence_one_or_zero/group_std_mean": 0.001424409542232752,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 7.758245601507952e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 7.758245601507952e-09,
"signal/format_reward/centered_abs_mean": 0.02083875872194767,
"signal/format_reward/group_std_mean": 0.04646492563188076,
"signal/format_reward/group_zero_std_frac": 0.7833333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010419379360973835,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010419379360973835,
"signal/mean_confidence_reward/centered_abs_mean": 0.05682915970683098,
"signal/mean_confidence_reward/group_std_mean": 0.0796852469444275,
"signal/mean_confidence_reward/group_zero_std_frac": 0.32222222685813906,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.682916025762097e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.682916025762097e-07,
"step": 870
},
{
"calibration/aurc": 0.1326259607295234,
"calibration/batch_distribution_entropy": 0.5703000664198925,
"calibration/batch_entropy_100bins": 0.3109658784851812,
"calibration/batch_entropy_10bins": 0.5703000664198925,
"calibration/batch_entropy_50bins": 0.36606400076699525,
"calibration/batch_uniqueness": 0.26676141664879094,
"calibration/confidence_entropy": 0.4490646421766165,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.13263707571801567,
"calibration/coverage@10%": 0.2908921726570746,
"calibration/coverage@15%": 0.6724807127374366,
"calibration/coverage@20%": 0.8730004205402586,
"calibration/coverage@25%": 0.9331179101221639,
"calibration/coverage@30%": 0.9471804101221639,
"calibration/coverage@5%": 0.2830592744847508,
"calibration/distribution_entropy_10": 0.5703000664198925,
"calibration/distribution_entropy_100": 0.3109658784851812,
"calibration/ece": 0.11515947318894779,
"calibration/mean_confidence": 0.7536589003345024,
"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": 0.010677083333333327,
"completions/max_length": 3974.4,
"completions/max_terminated_length": 3974.4,
"completions/mean_length": 1170.606982421875,
"completions/mean_terminated_length": 1183.2097412109374,
"completions/min_length": 0.0,
"completions/min_terminated_length": 372.2,
"epoch": 2.1033653846153846,
"grad_norm": 0.00039038530667312443,
"learning_rate": 9.915865384615386e-07,
"loss": -0.0127,
"num_tokens": 2386605948.0,
"reward": 1.2966346979141234,
"reward_std": 0.11709458529949188,
"rewards/accuracy_reward": 0.7470486044883728,
"rewards/brier_reward": 0.8568825960159302,
"rewards/confidence_one_or_zero": 0.0019097222539130598,
"rewards/format_reward": 0.9893229007720947,
"rewards/mean_confidence_reward": 0.7596822857856751,
"sampling/batch_mean_priority_error": 0.022411527777777763,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.003110817261040211,
"sampling/priority_kl": 0.0300005417317152,
"sampling/priority_scale": 0.8590299667092041,
"sampling/prob_entropy": 10.278952026367188,
"sampling/prob_max": 6.415506068151444e-05,
"sampling/prob_min": 2.0433545432752e-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": 0.09424912929534912,
"signal/accuracy_reward/group_std_mean": 0.1341444432735443,
"signal/accuracy_reward/group_zero_std_frac": 0.5805555701255798,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04712456464767456,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04712456464767456,
"signal/advantage_abs_mean": 0.07558977454900742,
"signal/advantage_pre_scale_abs_mean": 0.07558977454900742,
"signal/advantage_pre_scale_std": 0.17589679658412932,
"signal/advantage_std": 0.17589679658412932,
"signal/brier_reward/centered_abs_mean": 0.06083095222711563,
"signal/brier_reward/group_std_mean": 0.09111705869436264,
"signal/brier_reward/group_zero_std_frac": 0.2805555611848831,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.030415476113557816,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.030415476113557816,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0034722222364507614,
"signal/confidence_one_or_zero/group_std_mean": 0.006823088601231575,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222089767456,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.472221976608125e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.472221976608125e-08,
"signal/format_reward/centered_abs_mean": 0.019514974392950534,
"signal/format_reward/group_std_mean": 0.04219624921679497,
"signal/format_reward/group_zero_std_frac": 0.8083333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009757487196475267,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009757487196475267,
"signal/mean_confidence_reward/centered_abs_mean": 0.05601774528622627,
"signal/mean_confidence_reward/group_std_mean": 0.0795501410961151,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3000000029802322,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.601774432761886e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.601774432761886e-07,
"step": 875
},
{
"calibration/aurc": 0.15871351254864058,
"calibration/batch_distribution_entropy": 0.5979251893860597,
"calibration/batch_entropy_100bins": 0.32784764455624166,
"calibration/batch_entropy_10bins": 0.5979251893860597,
"calibration/batch_entropy_50bins": 0.3859369426411611,
"calibration/batch_uniqueness": 0.321340491986652,
"calibration/confidence_entropy": 0.44463021494639604,
"calibration/coverage@0%": 0.0036635759955593017,
"calibration/coverage@1%": 0.0036635759955593017,
"calibration/coverage@10%": 0.2672878219176986,
"calibration/coverage@15%": 0.62364241134756,
"calibration/coverage@20%": 0.7431073626128978,
"calibration/coverage@25%": 0.8104618719351233,
"calibration/coverage@30%": 0.8886401039120709,
"calibration/coverage@5%": 0.2076019580433531,
"calibration/distribution_entropy_10": 0.5979251893860597,
"calibration/distribution_entropy_100": 0.32784764455624166,
"calibration/ece": 0.1257229745901393,
"calibration/mean_confidence": 0.7385559006044398,
"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": 0.011979166666666652,
"completions/max_length": 4006.4,
"completions/max_terminated_length": 4006.4,
"completions/mean_length": 1191.8174560546875,
"completions/mean_terminated_length": 1206.124072265625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 375.8,
"epoch": 2.1153846153846154,
"grad_norm": 0.0003725996648427099,
"learning_rate": 9.615384615384617e-07,
"loss": -0.0146,
"num_tokens": 2403446789.0,
"reward": 1.2804926872253417,
"reward_std": 0.12390426397323609,
"rewards/accuracy_reward": 0.7283854126930237,
"rewards/brier_reward": 0.8445640206336975,
"rewards/confidence_one_or_zero": 0.001215277798473835,
"rewards/format_reward": 0.9880208373069763,
"rewards/mean_confidence_reward": 0.7525034785270691,
"sampling/batch_mean_priority_error": 0.018138958333333323,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08888888888888888,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.0031300032045692206,
"sampling/priority_kl": 0.030000030994415283,
"sampling/priority_scale": 0.86199485657271,
"sampling/prob_entropy": 10.278945159912109,
"sampling/prob_max": 6.444982864195481e-05,
"sampling/prob_min": 2.044653374468908e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.107200002670288,
"sampling/prompt_draws_total": 63216.0,
"sampling/seen_fraction": 0.9495066642761231,
"sampling/unseen_fraction": 0.05049333572387695,
"signal/accuracy_reward/centered_abs_mean": 0.10213216096162796,
"signal/accuracy_reward/group_std_mean": 0.14350362718105317,
"signal/accuracy_reward/group_zero_std_frac": 0.550000011920929,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05106608048081398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05106608048081398,
"signal/advantage_abs_mean": 0.08228934407234192,
"signal/advantage_pre_scale_abs_mean": 0.08228934407234192,
"signal/advantage_pre_scale_std": 0.18068719208240508,
"signal/advantage_std": 0.18068719208240508,
"signal/brier_reward/centered_abs_mean": 0.07063276022672653,
"signal/brier_reward/group_std_mean": 0.1014021709561348,
"signal/brier_reward/group_zero_std_frac": 0.2472222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.035316380113363265,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.035316380113363265,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0021592881996184586,
"signal/confidence_one_or_zero/group_std_mean": 0.0035119739361107348,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9888888835906983,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.1592879306808755e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.1592879306808755e-08,
"signal/format_reward/centered_abs_mean": 0.020681423507630824,
"signal/format_reward/group_std_mean": 0.04035768248140812,
"signal/format_reward/group_zero_std_frac": 0.8277777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010340711753815412,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010340711753815412,
"signal/mean_confidence_reward/centered_abs_mean": 0.057702585309743884,
"signal/mean_confidence_reward/group_std_mean": 0.08118895143270492,
"signal/mean_confidence_reward/group_zero_std_frac": 0.26111111640930174,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.770258098891645e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.770258098891645e-07,
"step": 880
},
{
"calibration/aurc": 0.06402128082310694,
"calibration/batch_distribution_entropy": 0.5680207879108812,
"calibration/batch_entropy_100bins": 0.30439327731708266,
"calibration/batch_entropy_10bins": 0.5680207879108812,
"calibration/batch_entropy_50bins": 0.35832684101570556,
"calibration/batch_uniqueness": 0.2564380669528326,
"calibration/confidence_entropy": 0.4407376487347753,
"calibration/coverage@0%": 0.1181074878729164,
"calibration/coverage@1%": 0.1181074878729164,
"calibration/coverage@10%": 0.7444454004999521,
"calibration/coverage@15%": 0.8812842511410043,
"calibration/coverage@20%": 0.9736083672879692,
"calibration/coverage@25%": 0.9923705722070844,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.5450419304602916,
"calibration/distribution_entropy_10": 0.5680207879108812,
"calibration/distribution_entropy_100": 0.30439327731708266,
"calibration/ece": 0.10531805626526428,
"calibration/mean_confidence": 0.7626525515430307,
"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": 0.009635416666666697,
"completions/max_length": 3861.8,
"completions/max_terminated_length": 3861.8,
"completions/mean_length": 1187.0781494140624,
"completions/mean_terminated_length": 1198.6516845703125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 378.2,
"epoch": 2.1274038461538463,
"grad_norm": 0.0003827160981018096,
"learning_rate": 9.314903846153847e-07,
"loss": -0.0097,
"num_tokens": 2420203177.0,
"reward": 1.3280963182449341,
"reward_std": 0.11590573638677597,
"rewards/accuracy_reward": 0.7860243082046509,
"rewards/brier_reward": 0.8797884583473206,
"rewards/confidence_one_or_zero": 0.0010416666977107526,
"rewards/format_reward": 0.9903645873069763,
"rewards/mean_confidence_reward": 0.7674363493919373,
"sampling/batch_mean_priority_error": 0.016940972222222215,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08055555555555555,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.003148121014237404,
"sampling/priority_kl": 0.030001012980937956,
"sampling/priority_scale": 0.864310544799082,
"sampling/prob_entropy": 10.27896728515625,
"sampling/prob_max": 6.470941589213908e-05,
"sampling/prob_min": 2.0465672423597426e-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": 0.09685872495174408,
"signal/accuracy_reward/group_std_mean": 0.13430885076522828,
"signal/accuracy_reward/group_zero_std_frac": 0.5888888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04842936247587204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04842936247587204,
"signal/advantage_abs_mean": 0.07655163407325745,
"signal/advantage_pre_scale_abs_mean": 0.07655163407325745,
"signal/advantage_pre_scale_std": 0.1744627386331558,
"signal/advantage_std": 0.1744627386331558,
"signal/brier_reward/centered_abs_mean": 0.061053629219532016,
"signal/brier_reward/group_std_mean": 0.09024021923542022,
"signal/brier_reward/group_zero_std_frac": 0.2944444537162781,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.030526814609766008,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.030526814609766008,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0019205729477107525,
"signal/confidence_one_or_zero/group_std_mean": 0.0036947034299373628,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.9205727141979877e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.9205727141979877e-08,
"signal/format_reward/centered_abs_mean": 0.017789713852107526,
"signal/format_reward/group_std_mean": 0.04016635827720165,
"signal/format_reward/group_zero_std_frac": 0.8111111283302307,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008894856926053763,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008894856926053763,
"signal/mean_confidence_reward/centered_abs_mean": 0.054922057688236235,
"signal/mean_confidence_reward/group_std_mean": 0.07804816663265228,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3083333373069763,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.492205900736736e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.492205900736736e-07,
"step": 885
},
{
"calibration/aurc": 0.08135021556057123,
"calibration/batch_distribution_entropy": 0.524748091346437,
"calibration/batch_entropy_100bins": 0.2862168308453696,
"calibration/batch_entropy_10bins": 0.524748091346437,
"calibration/batch_entropy_50bins": 0.3369298223216449,
"calibration/batch_uniqueness": 0.16690429657296696,
"calibration/confidence_entropy": 0.42909584582563703,
"calibration/coverage@0%": 0.2532940378039133,
"calibration/coverage@1%": 0.2622650140572114,
"calibration/coverage@10%": 0.5535564940030449,
"calibration/coverage@15%": 0.8217641250795115,
"calibration/coverage@20%": 0.8642814403829415,
"calibration/coverage@25%": 0.9159268929503916,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.5171448845043642,
"calibration/distribution_entropy_10": 0.524748091346437,
"calibration/distribution_entropy_100": 0.2862168308453696,
"calibration/ece": 0.12077254211531878,
"calibration/mean_confidence": 0.769572349046679,
"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": 0.010156249999999978,
"completions/max_length": 3911.6,
"completions/max_terminated_length": 3911.6,
"completions/mean_length": 1136.7791748046875,
"completions/mean_terminated_length": 1148.4474609375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 323.8,
"epoch": 2.139423076923077,
"grad_norm": 0.0003794643562287092,
"learning_rate": 9.014423076923078e-07,
"loss": -0.0122,
"num_tokens": 2436393785.0,
"reward": 1.2959206342697143,
"reward_std": 0.12398441880941391,
"rewards/accuracy_reward": 0.7411458373069764,
"rewards/brier_reward": 0.8608364820480346,
"rewards/confidence_one_or_zero": 0.0021701388992369174,
"rewards/format_reward": 0.98984375,
"rewards/mean_confidence_reward": 0.7559620976448059,
"sampling/batch_mean_priority_error": 0.01584027777777776,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.0031612716615200044,
"sampling/priority_kl": 0.030000754073262213,
"sampling/priority_scale": 0.8672416865592822,
"sampling/prob_entropy": 10.278964042663574,
"sampling/prob_max": 6.500642193714156e-05,
"sampling/prob_min": 2.0479490194702522e-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": 0.10128038078546524,
"signal/accuracy_reward/group_std_mean": 0.1370842784643173,
"signal/accuracy_reward/group_zero_std_frac": 0.5916666626930237,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05064019039273262,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05064019039273262,
"signal/advantage_abs_mean": 0.08475142419338226,
"signal/advantage_pre_scale_abs_mean": 0.08475142419338226,
"signal/advantage_pre_scale_std": 0.18326860070228576,
"signal/advantage_std": 0.18326860070228576,
"signal/brier_reward/centered_abs_mean": 0.06973624005913734,
"signal/brier_reward/group_std_mean": 0.09959090650081634,
"signal/brier_reward/group_zero_std_frac": 0.25277777910232546,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03486812002956867,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03486812002956867,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0037923176772892476,
"signal/confidence_one_or_zero/group_std_mean": 0.007589569129049778,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9666666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.79231739344732e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.79231739344732e-08,
"signal/format_reward/centered_abs_mean": 0.018202039785683154,
"signal/format_reward/group_std_mean": 0.03888495191931725,
"signal/format_reward/group_zero_std_frac": 0.8222222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009101019892841577,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009101019892841577,
"signal/mean_confidence_reward/centered_abs_mean": 0.05642659738659859,
"signal/mean_confidence_reward/group_std_mean": 0.07863938361406327,
"signal/mean_confidence_reward/group_zero_std_frac": 0.26111111640930174,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.642659402838035e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.642659402838035e-07,
"step": 890
},
{
"calibration/aurc": 0.0672622035824589,
"calibration/batch_distribution_entropy": 0.5153431049136842,
"calibration/batch_entropy_100bins": 0.27666561559182135,
"calibration/batch_entropy_10bins": 0.5153431049136842,
"calibration/batch_entropy_50bins": 0.325686286262536,
"calibration/batch_uniqueness": 0.13510791260886815,
"calibration/confidence_entropy": 0.4249389713275911,
"calibration/coverage@0%": 0.23533583556747092,
"calibration/coverage@1%": 0.23533583556747092,
"calibration/coverage@10%": 0.6751500670241286,
"calibration/coverage@15%": 0.890733604463142,
"calibration/coverage@20%": 0.9260365535248042,
"calibration/coverage@25%": 0.9811119234116624,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.6207139909790096,
"calibration/distribution_entropy_10": 0.5153431049136842,
"calibration/distribution_entropy_100": 0.27666561559182135,
"calibration/ece": 0.11191498549810834,
"calibration/mean_confidence": 0.7828315720120509,
"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": 0.017708333333333326,
"completions/max_length": 4015.2,
"completions/max_terminated_length": 4015.2,
"completions/mean_length": 1184.7015625,
"completions/mean_terminated_length": 1206.0575927734376,
"completions/min_length": 0.0,
"completions/min_terminated_length": 380.8,
"epoch": 2.1514423076923075,
"grad_norm": 0.000362895370926708,
"learning_rate": 8.713942307692308e-07,
"loss": -0.0209,
"num_tokens": 2453134187.0,
"reward": 1.2979624271392822,
"reward_std": 0.13720027953386307,
"rewards/accuracy_reward": 0.7483506917953491,
"rewards/brier_reward": 0.8652673482894897,
"rewards/confidence_one_or_zero": 0.000260416668606922,
"rewards/format_reward": 0.9822916746139526,
"rewards/mean_confidence_reward": 0.75567706823349,
"sampling/batch_mean_priority_error": 0.01643749999999999,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1388888888888889,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.0031778300646692513,
"sampling/priority_kl": 0.02999963201582432,
"sampling/priority_scale": 0.8716409623390063,
"sampling/prob_entropy": 10.278945159912109,
"sampling/prob_max": 6.538898887811228e-05,
"sampling/prob_min": 2.0479632075876e-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": 0.1037814661860466,
"signal/accuracy_reward/group_std_mean": 0.14674909710884093,
"signal/accuracy_reward/group_zero_std_frac": 0.5444444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0518907330930233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0518907330930233,
"signal/advantage_abs_mean": 0.09052008986473084,
"signal/advantage_pre_scale_abs_mean": 0.09052008986473084,
"signal/advantage_pre_scale_std": 0.19969316124916076,
"signal/advantage_std": 0.19969316124916076,
"signal/brier_reward/centered_abs_mean": 0.07278569489717483,
"signal/brier_reward/group_std_mean": 0.1068168118596077,
"signal/brier_reward/group_zero_std_frac": 0.2638888955116272,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.036392847448587416,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.036392847448587416,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0005045572877861559,
"signal/confidence_one_or_zero/group_std_mean": 0.0014731390401721,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.0455724931453e-09,
"signal/format_reward/centered_abs_mean": 0.02977430522441864,
"signal/format_reward/group_std_mean": 0.05782720670104027,
"signal/format_reward/group_zero_std_frac": 0.7527777791023255,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01488715261220932,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01488715261220932,
"signal/mean_confidence_reward/centered_abs_mean": 0.059619140625,
"signal/mean_confidence_reward/group_std_mean": 0.0842504397034645,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2916666716337204,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.961913984720013e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.961913984720013e-07,
"step": 895
},
{
"calibration/aurc": 0.1151792000974751,
"calibration/batch_distribution_entropy": 0.6237504339432944,
"calibration/batch_entropy_100bins": 0.3416187716182247,
"calibration/batch_entropy_10bins": 0.6237504339432944,
"calibration/batch_entropy_50bins": 0.4021480906035585,
"calibration/batch_uniqueness": 0.36068083610505663,
"calibration/confidence_entropy": 0.45672126327320495,
"calibration/coverage@0%": 0.0016217158800279005,
"calibration/coverage@1%": 0.31676063706114854,
"calibration/coverage@10%": 0.6080577514273167,
"calibration/coverage@15%": 0.6460583456043814,
"calibration/coverage@20%": 0.676589746570565,
"calibration/coverage@25%": 0.8457617975419833,
"calibration/coverage@30%": 0.8878378209337962,
"calibration/coverage@5%": 0.43888339920948616,
"calibration/distribution_entropy_10": 0.6237504339432944,
"calibration/distribution_entropy_100": 0.3416187716182247,
"calibration/ece": 0.13816415671046162,
"calibration/mean_confidence": 0.7257923929019635,
"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": 0.01710069444444442,
"completions/max_length": 3986.4,
"completions/max_terminated_length": 3986.4,
"completions/mean_length": 1174.2949951171875,
"completions/mean_terminated_length": 1194.865283203125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 339.8,
"epoch": 2.1634615384615383,
"grad_norm": 0.00030595791758969426,
"learning_rate": 8.41346153846154e-07,
"loss": -0.0217,
"num_tokens": 2469756945.0,
"reward": 1.2932031631469727,
"reward_std": 0.13037577122449875,
"rewards/accuracy_reward": 0.7412326335906982,
"rewards/brier_reward": 0.862172544002533,
"rewards/confidence_one_or_zero": 0.0016493055794853718,
"rewards/format_reward": 0.9829861044883728,
"rewards/mean_confidence_reward": 0.7530859351158142,
"sampling/batch_mean_priority_error": 0.017564236111111096,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.10277777777777777,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.003195428755134344,
"sampling/priority_kl": 0.029999273642897607,
"sampling/priority_scale": 0.876445037056692,
"sampling/prob_entropy": 10.278936004638672,
"sampling/prob_max": 6.5792647365015e-05,
"sampling/prob_min": 2.0472642790991814e-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": 0.09095594584941864,
"signal/accuracy_reward/group_std_mean": 0.13292890042066574,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111044883728,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04547797292470932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04547797292470932,
"signal/advantage_abs_mean": 0.08368964940309524,
"signal/advantage_pre_scale_abs_mean": 0.08368964940309524,
"signal/advantage_pre_scale_std": 0.19153147637844087,
"signal/advantage_std": 0.19153147637844087,
"signal/brier_reward/centered_abs_mean": 0.07003483921289444,
"signal/brier_reward/group_std_mean": 0.10422394722700119,
"signal/brier_reward/group_zero_std_frac": 0.2416666716337204,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03501741960644722,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03501741960644722,
"signal/confidence_one_or_zero/centered_abs_mean": 0.002891710109543055,
"signal/confidence_one_or_zero/group_std_mean": 0.0054592888336628675,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9777777671813965,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.8917099115233213e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.8917099115233213e-08,
"signal/format_reward/centered_abs_mean": 0.03011067695915699,
"signal/format_reward/group_std_mean": 0.05745980739593506,
"signal/format_reward/group_zero_std_frac": 0.7638888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.015055338479578495,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.015055338479578495,
"signal/mean_confidence_reward/centered_abs_mean": 0.05772487968206406,
"signal/mean_confidence_reward/group_std_mean": 0.08457386791706085,
"signal/mean_confidence_reward/group_zero_std_frac": 0.24444445073604584,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.772487611466204e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.772487611466204e-07,
"step": 900
},
{
"epoch": 2.1634615384615383,
"eval_calibration/aurc": 0.12391385848990848,
"eval_calibration/batch_distribution_entropy": 0.5881655655004824,
"eval_calibration/batch_entropy_100bins": 0.3189768360923287,
"eval_calibration/batch_entropy_10bins": 0.5881655655004824,
"eval_calibration/batch_entropy_50bins": 0.37549437044592116,
"eval_calibration/batch_uniqueness": 0.26470764035785677,
"eval_calibration/confidence_entropy": 0.43872068387637836,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.5814977973568282,
"eval_calibration/coverage@15%": 0.7629955947136564,
"eval_calibration/coverage@20%": 0.8607929515418502,
"eval_calibration/coverage@25%": 0.9286343612334802,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.5881655655004824,
"eval_calibration/distribution_entropy_100": 0.3189768360923287,
"eval_calibration/ece": 0.04660792951541838,
"eval_calibration/mean_confidence": 0.7623788546255507,
"eval_calibration/unique_confidence_per_question": 0.008680555555555556,
"eval_calibration/unique_confidences": 10,
"eval_completions/clipped_ratio": 0.013020833333333351,
"eval_completions/max_length": 3145.1666666666665,
"eval_completions/max_terminated_length": 3145.1666666666665,
"eval_completions/mean_length": 1151.9165445963542,
"eval_completions/mean_terminated_length": 1167.2333984375,
"eval_completions/min_length": 84.0,
"eval_completions/min_terminated_length": 382.1666666666667,
"eval_loss": 0.0,
"eval_num_tokens": 2469756945.0,
"eval_reward": 1.278518795967102,
"eval_reward_std": 0.3357190191745758,
"eval_rewards/accuracy_reward": 0.7248263855775198,
"eval_rewards/brier_reward": 0.8469531536102295,
"eval_rewards/confidence_one_or_zero": 0.0008680555814256271,
"eval_rewards/format_reward": 0.9852430621782938,
"eval_rewards/mean_confidence_reward": 0.7511284351348877,
"eval_runtime": 214.2565,
"eval_samples_per_second": 4.667,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3844943592945735,
"eval_signal/accuracy_reward/group_std_mean": 0.44289974868297577,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19224717964728674,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19224717964728674,
"eval_signal/advantage_abs_mean": 0.2765105913082759,
"eval_signal/advantage_pre_scale_abs_mean": 0.2765105913082759,
"eval_signal/advantage_pre_scale_std": 0.3355718404054642,
"eval_signal/advantage_std": 0.3355718404054642,
"eval_signal/brier_reward/centered_abs_mean": 0.17802898089090982,
"eval_signal/brier_reward/group_std_mean": 0.2396557405591011,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08901449044545491,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08901449044545491,
"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": 0.028374566230922937,
"eval_signal/format_reward/group_std_mean": 0.07749906585862239,
"eval_signal/format_reward/group_zero_std_frac": 0.5833333432674408,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.014187283115461469,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.014187283115461469,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.17690427352984747,
"eval_signal/mean_confidence_reward/group_std_mean": 0.22398289293050766,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.76904264511298e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.76904264511298e-06,
"eval_steps_per_second": 0.028,
"step": 900
},
{
"epoch": 2.1634615384615383,
"step": 900,
"train_probe_calibration/aurc": 0.08825829287780615,
"train_probe_calibration/batch_distribution_entropy": 0.5748713882314315,
"train_probe_calibration/batch_entropy_100bins": 0.3128677793937022,
"train_probe_calibration/batch_entropy_10bins": 0.5748713882314315,
"train_probe_calibration/batch_entropy_50bins": 0.3683028877440073,
"train_probe_calibration/batch_uniqueness": 0.2436643079308503,
"train_probe_calibration/confidence_entropy": 0.43504627966835785,
"train_probe_calibration/coverage@0%": 0.0017574692442882249,
"train_probe_calibration/coverage@1%": 0.0017574692442882249,
"train_probe_calibration/coverage@10%": 0.7135325131810193,
"train_probe_calibration/coverage@15%": 0.8637961335676626,
"train_probe_calibration/coverage@20%": 0.9288224956063269,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0017574692442882249,
"train_probe_calibration/distribution_entropy_10": 0.5748713882314315,
"train_probe_calibration/distribution_entropy_100": 0.3128677793937022,
"train_probe_calibration/ece": 0.037170474516695806,
"train_probe_calibration/mean_confidence": 0.7673989455184536,
"train_probe_calibration/unique_confidence_per_question": 0.008680555555555556,
"train_probe_calibration/unique_confidences": 10,
"train_probe_completions/clipped_ratio": 0.014583333333333337,
"train_probe_completions/max_length": 3627.5,
"train_probe_completions/max_terminated_length": 3627.5,
"train_probe_completions/mean_length": 1189.9757283528645,
"train_probe_completions/mean_terminated_length": 1207.6895345052083,
"train_probe_completions/min_length": 0.0,
"train_probe_completions/min_terminated_length": 403.5,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 2469756945.0,
"train_probe_reward": 1.31611434618632,
"train_probe_reward_std": 0.30907075603802997,
"train_probe_rewards/accuracy_reward": 0.7725694477558136,
"train_probe_rewards/brier_reward": 0.8717969159285227,
"train_probe_rewards/confidence_one_or_zero": 0.0017361111628512542,
"train_probe_rewards/format_reward": 0.987847218910853,
"train_probe_rewards/mean_confidence_reward": 0.7580729226271311,
"train_probe_runtime": 219.1252,
"train_probe_samples_per_second": 4.564,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3404947916666667,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4156750937302907,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.17024739583333334,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.17024739583333334,
"train_probe_signal/advantage_abs_mean": 0.2414021318157514,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.2414021318157514,
"train_probe_signal/advantage_pre_scale_std": 0.3106756657361984,
"train_probe_signal/advantage_std": 0.3106756657361984,
"train_probe_signal/brier_reward/centered_abs_mean": 0.15325956543286642,
"train_probe_signal/brier_reward/group_std_mean": 0.217030035952727,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07662978271643321,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07662978271643321,
"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": 0.023220485852410395,
"train_probe_signal/format_reward/group_std_mean": 0.0597782659654816,
"train_probe_signal/format_reward/group_zero_std_frac": 0.6944444676240286,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.011610242926205197,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.011610242926205197,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.17286781469980875,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.2182989940047264,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.7286780765365013e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.7286780765365013e-06,
"train_probe_steps_per_second": 0.027
},
{
"calibration/aurc": 0.11285190153335967,
"calibration/batch_distribution_entropy": 0.5125113264155846,
"calibration/batch_entropy_100bins": 0.28051964622634007,
"calibration/batch_entropy_10bins": 0.5125113264155846,
"calibration/batch_entropy_50bins": 0.33022318876779816,
"calibration/batch_uniqueness": 0.11970550029297737,
"calibration/confidence_entropy": 0.4248528232860238,
"calibration/coverage@0%": 0.0010471204188481676,
"calibration/coverage@1%": 0.15916230366492146,
"calibration/coverage@10%": 0.475216599372537,
"calibration/coverage@15%": 0.6550666859718866,
"calibration/coverage@20%": 0.8777354111405836,
"calibration/coverage@25%": 0.9380374115826703,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.41120545975349587,
"calibration/distribution_entropy_10": 0.5125113264155846,
"calibration/distribution_entropy_100": 0.28051964622634007,
"calibration/ece": 0.1134556583005788,
"calibration/mean_confidence": 0.7646870969901838,
"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": 0.012065972222222231,
"completions/max_length": 3955.0,
"completions/max_terminated_length": 3955.0,
"completions/mean_length": 1146.1902099609374,
"completions/mean_terminated_length": 1160.305859375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 370.2,
"epoch": 2.175480769230769,
"grad_norm": 0.0003137659514322877,
"learning_rate": 8.11298076923077e-07,
"loss": -0.0156,
"num_tokens": 2486076896.0,
"reward": 1.3248236179351807,
"reward_std": 0.12037934958934784,
"rewards/accuracy_reward": 0.7915798664093018,
"rewards/brier_reward": 0.8702048897743225,
"rewards/confidence_one_or_zero": 0.0006076388992369175,
"rewards/format_reward": 0.9878472328186035,
"rewards/mean_confidence_reward": 0.7613367795944214,
"sampling/batch_mean_priority_error": 0.01599999999999998,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.09722222222222222,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.0032106395810842514,
"sampling/priority_kl": 0.029998685419559478,
"sampling/priority_scale": 0.8803082048194483,
"sampling/prob_entropy": 10.278942108154297,
"sampling/prob_max": 6.614316371269524e-05,
"sampling/prob_min": 2.047557645710185e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.167199993133545,
"sampling/prompt_draws_total": 65016.0,
"sampling/seen_fraction": 0.9556066632270813,
"sampling/unseen_fraction": 0.0443933367729187,
"signal/accuracy_reward/centered_abs_mean": 0.10290256142616272,
"signal/accuracy_reward/group_std_mean": 0.14169478863477708,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111283302307,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05145128071308136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05145128071308136,
"signal/advantage_abs_mean": 0.08037637546658516,
"signal/advantage_pre_scale_abs_mean": 0.08037637546658516,
"signal/advantage_pre_scale_std": 0.18246108591556548,
"signal/advantage_std": 0.18246108591556548,
"signal/brier_reward/centered_abs_mean": 0.06332063004374504,
"signal/brier_reward/group_std_mean": 0.09189697206020356,
"signal/brier_reward/group_zero_std_frac": 0.2972222208976746,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03166031502187252,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03166031502187252,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0011664496269077063,
"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": 1.1664495858099145e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.1664495858099145e-08,
"signal/format_reward/centered_abs_mean": 0.02082248292863369,
"signal/format_reward/group_std_mean": 0.04095887430012226,
"signal/format_reward/group_zero_std_frac": 0.8222222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010411241464316845,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010411241464316845,
"signal/mean_confidence_reward/centered_abs_mean": 0.05295898914337158,
"signal/mean_confidence_reward/group_std_mean": 0.07509669959545136,
"signal/mean_confidence_reward/group_zero_std_frac": 0.31388888955116273,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.295898745316663e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.295898745316663e-07,
"step": 905
},
{
"calibration/aurc": 0.13006745286642976,
"calibration/batch_distribution_entropy": 0.5223763623022502,
"calibration/batch_entropy_100bins": 0.282748995104773,
"calibration/batch_entropy_10bins": 0.5223763623022502,
"calibration/batch_entropy_50bins": 0.33284754219692675,
"calibration/batch_uniqueness": 0.15022688073890672,
"calibration/confidence_entropy": 0.43531714143483746,
"calibration/coverage@0%": 0.14986118137011228,
"calibration/coverage@1%": 0.14986118137011228,
"calibration/coverage@10%": 0.17308017873158985,
"calibration/coverage@15%": 0.6216689548354984,
"calibration/coverage@20%": 0.9512319467919914,
"calibration/coverage@25%": 0.9937007874015749,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.16727542939122048,
"calibration/distribution_entropy_10": 0.5223763623022502,
"calibration/distribution_entropy_100": 0.282748995104773,
"calibration/ece": 0.09975166442201888,
"calibration/mean_confidence": 0.7799321903281939,
"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": 0.014236111111111093,
"completions/max_length": 3917.4,
"completions/max_terminated_length": 3917.4,
"completions/mean_length": 1192.1790283203125,
"completions/mean_terminated_length": 1209.365478515625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 353.4,
"epoch": 2.1875,
"grad_norm": 0.0003341422707308084,
"learning_rate": 7.8125e-07,
"loss": -0.0176,
"num_tokens": 2502874798.0,
"reward": 1.2831647396087646,
"reward_std": 0.13368360996246337,
"rewards/accuracy_reward": 0.7313368082046509,
"rewards/brier_reward": 0.849213969707489,
"rewards/confidence_one_or_zero": 0.0006944444554392249,
"rewards/format_reward": 0.9857638835906982,
"rewards/mean_confidence_reward": 0.7395833253860473,
"sampling/batch_mean_priority_error": 0.017489583333333322,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.06666666666666668,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.003227849630638957,
"sampling/priority_kl": 0.029999754205346108,
"sampling/priority_scale": 0.8832817733054981,
"sampling/prob_entropy": 10.278969383239746,
"sampling/prob_max": 6.644506938755513e-05,
"sampling/prob_min": 2.0487737128860316e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.179200029373169,
"sampling/prompt_draws_total": 65376.0,
"sampling/seen_fraction": 0.9565999984741211,
"sampling/unseen_fraction": 0.043400001525878903,
"signal/accuracy_reward/centered_abs_mean": 0.10817599892616273,
"signal/accuracy_reward/group_std_mean": 0.15262871086597443,
"signal/accuracy_reward/group_zero_std_frac": 0.5277777850627899,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05408799946308136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05408799946308136,
"signal/advantage_abs_mean": 0.08564771115779876,
"signal/advantage_pre_scale_abs_mean": 0.08564771115779876,
"signal/advantage_pre_scale_std": 0.19176922142505645,
"signal/advantage_std": 0.19176922142505645,
"signal/brier_reward/centered_abs_mean": 0.06797481775283813,
"signal/brier_reward/group_std_mean": 0.10257073640823364,
"signal/brier_reward/group_zero_std_frac": 0.2638888955116272,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.033987408876419066,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.033987408876419066,
"signal/confidence_one_or_zero/centered_abs_mean": 0.001302083337213844,
"signal/confidence_one_or_zero/group_std_mean": 0.0029789147432893516,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.302083241228047e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.302083241228047e-08,
"signal/format_reward/centered_abs_mean": 0.02587890625,
"signal/format_reward/group_std_mean": 0.05563742592930794,
"signal/format_reward/group_zero_std_frac": 0.7472222328186036,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.012939453125,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.012939453125,
"signal/mean_confidence_reward/centered_abs_mean": 0.058856339752674104,
"signal/mean_confidence_reward/group_std_mean": 0.08640173077583313,
"signal/mean_confidence_reward/group_zero_std_frac": 0.27499999701976774,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.88563352721394e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.88563352721394e-07,
"step": 910
},
{
"calibration/aurc": 0.10734249873858166,
"calibration/batch_distribution_entropy": 0.5846067326248493,
"calibration/batch_entropy_100bins": 0.3120906730568165,
"calibration/batch_entropy_10bins": 0.5846067326248493,
"calibration/batch_entropy_50bins": 0.3673880907377006,
"calibration/batch_uniqueness": 0.2989200019476256,
"calibration/confidence_entropy": 0.4400907935995416,
"calibration/coverage@0%": 0.11556728232189975,
"calibration/coverage@1%": 0.11556728232189975,
"calibration/coverage@10%": 0.5006087125384353,
"calibration/coverage@15%": 0.701646585434703,
"calibration/coverage@20%": 0.7472763925729443,
"calibration/coverage@25%": 0.9019538873994637,
"calibration/coverage@30%": 0.9163538873994639,
"calibration/coverage@5%": 0.4473732813107564,
"calibration/distribution_entropy_10": 0.5846067326248493,
"calibration/distribution_entropy_100": 0.3120906730568165,
"calibration/ece": 0.13515245517625668,
"calibration/mean_confidence": 0.7708814941117261,
"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": 0.015885416666666673,
"completions/max_length": 3968.0,
"completions/max_terminated_length": 3968.0,
"completions/mean_length": 1211.677880859375,
"completions/mean_terminated_length": 1231.2791748046875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 379.4,
"epoch": 2.199519230769231,
"grad_norm": 0.0003999628243036568,
"learning_rate": 7.512019230769231e-07,
"loss": -0.0198,
"num_tokens": 2519922959.0,
"reward": 1.3061288833618163,
"reward_std": 0.13455301523208618,
"rewards/accuracy_reward": 0.7528645873069764,
"rewards/brier_reward": 0.8754372000694275,
"rewards/confidence_one_or_zero": 0.000260416668606922,
"rewards/format_reward": 0.9839409589767456,
"rewards/mean_confidence_reward": 0.7495818614959717,
"sampling/batch_mean_priority_error": 0.011972222222222214,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08611111111111111,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.0032408578786998986,
"sampling/priority_kl": 0.029999370872974395,
"sampling/priority_scale": 0.8859618365531787,
"sampling/prob_entropy": 10.278959465026855,
"sampling/prob_max": 6.67307889671065e-05,
"sampling/prob_min": 2.0502503321040422e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.191200017929077,
"sampling/prompt_draws_total": 65736.0,
"sampling/seen_fraction": 0.9575066685676574,
"sampling/unseen_fraction": 0.04249333143234253,
"signal/accuracy_reward/centered_abs_mean": 0.098388671875,
"signal/accuracy_reward/group_std_mean": 0.13771490454673768,
"signal/accuracy_reward/group_zero_std_frac": 0.569444453716278,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0491943359375,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0491943359375,
"signal/advantage_abs_mean": 0.08455693274736405,
"signal/advantage_pre_scale_abs_mean": 0.08455693274736405,
"signal/advantage_pre_scale_std": 0.19366905689239503,
"signal/advantage_std": 0.19366905689239503,
"signal/brier_reward/centered_abs_mean": 0.06902645826339722,
"signal/brier_reward/group_std_mean": 0.10544530004262924,
"signal/brier_reward/group_zero_std_frac": 0.2888888895511627,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03451322913169861,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03451322913169861,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0004937065881676972,
"signal/confidence_one_or_zero/group_std_mean": 0.001174198230728507,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.9370658672387435e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.9370658672387435e-09,
"signal/format_reward/centered_abs_mean": 0.029150390625,
"signal/format_reward/group_std_mean": 0.06167853176593781,
"signal/format_reward/group_zero_std_frac": 0.725,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0145751953125,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0145751953125,
"signal/mean_confidence_reward/centered_abs_mean": 0.06032561212778091,
"signal/mean_confidence_reward/group_std_mean": 0.08825595080852508,
"signal/mean_confidence_reward/group_zero_std_frac": 0.30555555820465086,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.03256125941698e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.03256125941698e-07,
"step": 915
},
{
"calibration/aurc": 0.05147939346039525,
"calibration/batch_distribution_entropy": 0.5554592712525979,
"calibration/batch_entropy_100bins": 0.3054071326847551,
"calibration/batch_entropy_10bins": 0.5554592712525979,
"calibration/batch_entropy_50bins": 0.3595203351504873,
"calibration/batch_uniqueness": 0.2235803159489635,
"calibration/confidence_entropy": 0.43550255930955634,
"calibration/coverage@0%": 0.12471754581242238,
"calibration/coverage@1%": 0.32483220253296385,
"calibration/coverage@10%": 0.8518813929328273,
"calibration/coverage@15%": 0.9233775087260035,
"calibration/coverage@20%": 0.9572916666666668,
"calibration/coverage@25%": 0.9828125,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.6240863469354763,
"calibration/distribution_entropy_10": 0.5554592712525979,
"calibration/distribution_entropy_100": 0.3054071326847551,
"calibration/ece": 0.11478028141041932,
"calibration/mean_confidence": 0.7626465657632444,
"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": 0.012326388888888883,
"completions/max_length": 4030.2,
"completions/max_terminated_length": 4030.2,
"completions/mean_length": 1243.1653076171874,
"completions/mean_terminated_length": 1258.8077880859375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 377.6,
"epoch": 2.2115384615384617,
"grad_norm": 0.00030230532865971327,
"learning_rate": 7.211538461538461e-07,
"loss": -0.015,
"num_tokens": 2537358143.0,
"reward": 1.3202105522155763,
"reward_std": 0.11819398552179336,
"rewards/accuracy_reward": 0.7690104126930237,
"rewards/brier_reward": 0.8837220311164856,
"rewards/confidence_one_or_zero": 0.0014756945020053536,
"rewards/format_reward": 0.9876736044883728,
"rewards/mean_confidence_reward": 0.7527213454246521,
"sampling/batch_mean_priority_error": 0.01480034722222221,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08055555555555556,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.0032544697634875773,
"sampling/priority_kl": 0.02999933548271656,
"sampling/priority_scale": 0.8895649968879298,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 6.707530410494655e-05,
"sampling/prob_min": 2.0509682144620456e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.2032000064849853,
"sampling/prompt_draws_total": 66096.0,
"sampling/seen_fraction": 0.9585599899291992,
"sampling/unseen_fraction": 0.04144001007080078,
"signal/accuracy_reward/centered_abs_mean": 0.09224717915058137,
"signal/accuracy_reward/group_std_mean": 0.13409704864025115,
"signal/accuracy_reward/group_zero_std_frac": 0.5666666686534881,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04612358957529068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04612358957529068,
"signal/advantage_abs_mean": 0.07567234933376313,
"signal/advantage_pre_scale_abs_mean": 0.07567234933376313,
"signal/advantage_pre_scale_std": 0.17569505870342256,
"signal/advantage_std": 0.17569505870342256,
"signal/brier_reward/centered_abs_mean": 0.0633341558277607,
"signal/brier_reward/group_std_mean": 0.09425111263990402,
"signal/brier_reward/group_zero_std_frac": 0.2666666716337204,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03166707791388035,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03166707791388035,
"signal/confidence_one_or_zero/centered_abs_mean": 0.002381727460306138,
"signal/confidence_one_or_zero/group_std_mean": 0.0034978067502379417,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9888888835906983,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.3817273842041685e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.3817273842041685e-08,
"signal/format_reward/centered_abs_mean": 0.02136501707136631,
"signal/format_reward/group_std_mean": 0.04142211228609085,
"signal/format_reward/group_zero_std_frac": 0.825000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010682508535683154,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010682508535683154,
"signal/mean_confidence_reward/centered_abs_mean": 0.05657335072755813,
"signal/mean_confidence_reward/group_std_mean": 0.07980925291776657,
"signal/mean_confidence_reward/group_zero_std_frac": 0.28611111640930176,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.657335009345843e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.657335009345843e-07,
"step": 920
},
{
"calibration/aurc": 0.0879748932949886,
"calibration/batch_distribution_entropy": 0.5164573947181831,
"calibration/batch_entropy_100bins": 0.27981525602896584,
"calibration/batch_entropy_10bins": 0.5164573947181831,
"calibration/batch_entropy_50bins": 0.32939399202438713,
"calibration/batch_uniqueness": 0.13016719252669615,
"calibration/confidence_entropy": 0.4321006951580036,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.17060367454068243,
"calibration/coverage@10%": 0.594761664136493,
"calibration/coverage@15%": 0.847351488641174,
"calibration/coverage@20%": 0.8934965429659941,
"calibration/coverage@25%": 0.9884210526315791,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.3633532753575516,
"calibration/distribution_entropy_10": 0.5164573947181831,
"calibration/distribution_entropy_100": 0.27981525602896584,
"calibration/ece": 0.11073895301248071,
"calibration/mean_confidence": 0.7676560881506441,
"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": 0.014236111111111139,
"completions/max_length": 4039.2,
"completions/max_terminated_length": 4039.2,
"completions/mean_length": 1223.8078857421874,
"completions/mean_terminated_length": 1241.5063720703124,
"completions/min_length": 0.0,
"completions/min_terminated_length": 395.0,
"epoch": 2.2235576923076925,
"grad_norm": 0.0003381640126463026,
"learning_rate": 6.911057692307694e-07,
"loss": -0.0185,
"num_tokens": 2554543098.0,
"reward": 1.3357814073562622,
"reward_std": 0.12255254238843918,
"rewards/accuracy_reward": 0.8040798783302308,
"rewards/brier_reward": 0.8818773984909057,
"rewards/confidence_one_or_zero": 0.0003472222248092294,
"rewards/format_reward": 0.9855902671813965,
"rewards/mean_confidence_reward": 0.7671050190925598,
"sampling/batch_mean_priority_error": 0.014381944444444428,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.07222222222222222,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.003269455023109913,
"sampling/priority_kl": 0.030000782385468484,
"sampling/priority_scale": 0.8927285850746557,
"sampling/prob_entropy": 10.278964614868164,
"sampling/prob_max": 6.739153468515724e-05,
"sampling/prob_min": 2.051973424386233e-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": 0.08742946833372116,
"signal/accuracy_reward/group_std_mean": 0.13191601932048796,
"signal/accuracy_reward/group_zero_std_frac": 0.5500000059604645,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04371473416686058,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04371473416686058,
"signal/advantage_abs_mean": 0.0757093921303749,
"signal/advantage_pre_scale_abs_mean": 0.0757093921303749,
"signal/advantage_pre_scale_std": 0.18412608802318572,
"signal/advantage_std": 0.18412608802318572,
"signal/brier_reward/centered_abs_mean": 0.06260225027799607,
"signal/brier_reward/group_std_mean": 0.09538311511278152,
"signal/brier_reward/group_zero_std_frac": 0.25,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.031301125138998034,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.031301125138998034,
"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": 0.02479383647441864,
"signal/format_reward/group_std_mean": 0.0480035699903965,
"signal/format_reward/group_zero_std_frac": 0.7972222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01239691823720932,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01239691823720932,
"signal/mean_confidence_reward/centered_abs_mean": 0.05885334834456444,
"signal/mean_confidence_reward/group_std_mean": 0.08518647253513337,
"signal/mean_confidence_reward/group_zero_std_frac": 0.25555555820465087,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.88533464451757e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.88533464451757e-07,
"step": 925
},
{
"calibration/aurc": 0.04662198307011302,
"calibration/batch_distribution_entropy": 0.516365256651883,
"calibration/batch_entropy_100bins": 0.27891007924346534,
"calibration/batch_entropy_10bins": 0.516365256651883,
"calibration/batch_entropy_50bins": 0.3283284325581336,
"calibration/batch_uniqueness": 0.14541062928179097,
"calibration/confidence_entropy": 0.4226501508386008,
"calibration/coverage@0%": 0.003685752774035858,
"calibration/coverage@1%": 0.26561503772239636,
"calibration/coverage@10%": 0.8407537880959353,
"calibration/coverage@15%": 0.9285548718823253,
"calibration/coverage@20%": 0.9957559681697614,
"calibration/coverage@25%": 1.0,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.5987353612990702,
"calibration/distribution_entropy_10": 0.516365256651883,
"calibration/distribution_entropy_100": 0.27891007924346534,
"calibration/ece": 0.10740721265309265,
"calibration/mean_confidence": 0.7776348798779866,
"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": 0.015104166666666651,
"completions/max_length": 3987.2,
"completions/max_terminated_length": 3987.2,
"completions/mean_length": 1215.920947265625,
"completions/mean_terminated_length": 1234.5720458984374,
"completions/min_length": 0.0,
"completions/min_terminated_length": 381.8,
"epoch": 2.235576923076923,
"grad_norm": 0.00032682056189514697,
"learning_rate": 6.610576923076924e-07,
"loss": -0.0176,
"num_tokens": 2571639243.0,
"reward": 1.3099323987960816,
"reward_std": 0.12759646475315095,
"rewards/accuracy_reward": 0.7669270992279053,
"rewards/brier_reward": 0.867940080165863,
"rewards/confidence_one_or_zero": 0.0013888889225199819,
"rewards/format_reward": 0.9849826455116272,
"rewards/mean_confidence_reward": 0.7527873277664184,
"sampling/batch_mean_priority_error": 0.02201562499999999,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08055555555555556,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.0032882534898817537,
"sampling/priority_kl": 0.029999984428286552,
"sampling/priority_scale": 0.8953011691337451,
"sampling/prob_entropy": 10.278952026367188,
"sampling/prob_max": 6.767441664123908e-05,
"sampling/prob_min": 2.0523589410004205e-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": 0.10499674379825592,
"signal/accuracy_reward/group_std_mean": 0.14623960852622986,
"signal/accuracy_reward/group_zero_std_frac": 0.5472222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05249837189912796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05249837189912796,
"signal/advantage_abs_mean": 0.08412635624408722,
"signal/advantage_pre_scale_abs_mean": 0.08412635624408722,
"signal/advantage_pre_scale_std": 0.18825245201587676,
"signal/advantage_std": 0.18825245201587676,
"signal/brier_reward/centered_abs_mean": 0.06787939220666886,
"signal/brier_reward/group_std_mean": 0.09967477023601531,
"signal/brier_reward/group_zero_std_frac": 0.27777778208255766,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03393969610333443,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03393969610333443,
"signal/confidence_one_or_zero/centered_abs_mean": 0.002473958348855376,
"signal/confidence_one_or_zero/group_std_mean": 0.004445959627628326,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9833333253860473,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.4739582471511314e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.4739582471511314e-08,
"signal/format_reward/centered_abs_mean": 0.02551540769636631,
"signal/format_reward/group_std_mean": 0.050245880335569384,
"signal/format_reward/group_zero_std_frac": 0.7833333492279053,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.012757703848183154,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.012757703848183154,
"signal/mean_confidence_reward/centered_abs_mean": 0.060628201067447665,
"signal/mean_confidence_reward/group_std_mean": 0.08692292124032974,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2805555611848831,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.062819920771289e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.062819920771289e-07,
"step": 930
},
{
"calibration/aurc": 0.09865934087702413,
"calibration/batch_distribution_entropy": 0.5613819194359139,
"calibration/batch_entropy_100bins": 0.30502884058675644,
"calibration/batch_entropy_10bins": 0.5613819194359139,
"calibration/batch_entropy_50bins": 0.35907501581343804,
"calibration/batch_uniqueness": 0.21185733367966794,
"calibration/confidence_entropy": 0.4414002995727424,
"calibration/coverage@0%": 0.11916010498687664,
"calibration/coverage@1%": 0.22384760498687664,
"calibration/coverage@10%": 0.6438190081681894,
"calibration/coverage@15%": 0.715426839846917,
"calibration/coverage@20%": 0.8546681971527119,
"calibration/coverage@25%": 0.9043575479930193,
"calibration/coverage@30%": 0.9801210732984293,
"calibration/coverage@5%": 0.38435554410545286,
"calibration/distribution_entropy_10": 0.5613819194359139,
"calibration/distribution_entropy_100": 0.30502884058675644,
"calibration/ece": 0.10591542545003192,
"calibration/mean_confidence": 0.7539848946683242,
"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": 0.008333333333333326,
"completions/max_length": 3919.2,
"completions/max_terminated_length": 3919.2,
"completions/mean_length": 1211.381689453125,
"completions/mean_terminated_length": 1221.6566162109375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 394.0,
"epoch": 2.2475961538461537,
"grad_norm": 0.0003420446300879121,
"learning_rate": 6.310096153846154e-07,
"loss": -0.009,
"num_tokens": 2588697464.0,
"reward": 1.3223564863204955,
"reward_std": 0.11324650794267654,
"rewards/accuracy_reward": 0.77890625,
"rewards/brier_reward": 0.8741249918937684,
"rewards/confidence_one_or_zero": 0.0013888888817746193,
"rewards/format_reward": 0.9916666626930237,
"rewards/mean_confidence_reward": 0.7576215147972107,
"sampling/batch_mean_priority_error": 0.015006944444444434,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08611111111111111,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.0033051112201064824,
"sampling/priority_kl": 0.029999464750289917,
"sampling/priority_scale": 0.8995260178809985,
"sampling/prob_entropy": 10.27895565032959,
"sampling/prob_max": 6.806203018641099e-05,
"sampling/prob_min": 2.052525269391481e-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": 0.09180229902267456,
"signal/accuracy_reward/group_std_mean": 0.1310224562883377,
"signal/accuracy_reward/group_zero_std_frac": 0.5916666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04590114951133728,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04590114951133728,
"signal/advantage_abs_mean": 0.07339992076158523,
"signal/advantage_pre_scale_abs_mean": 0.07339992076158523,
"signal/advantage_pre_scale_std": 0.16779326498508454,
"signal/advantage_std": 0.16779326498508454,
"signal/brier_reward/centered_abs_mean": 0.06400536745786667,
"signal/brier_reward/group_std_mean": 0.0935688391327858,
"signal/brier_reward/group_zero_std_frac": 0.3111111164093018,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.032002683728933334,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.032002683728933334,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0023111979360692205,
"signal/confidence_one_or_zero/group_std_mean": 0.0038206671364605428,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.311197775384244e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.311197775384244e-08,
"signal/format_reward/centered_abs_mean": 0.01534288190305233,
"signal/format_reward/group_std_mean": 0.03594543486833572,
"signal/format_reward/group_zero_std_frac": 0.825000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007671440951526165,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007671440951526165,
"signal/mean_confidence_reward/centered_abs_mean": 0.05464844331145287,
"signal/mean_confidence_reward/group_std_mean": 0.07704180479049683,
"signal/mean_confidence_reward/group_zero_std_frac": 0.325,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.46484409369441e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.46484409369441e-07,
"step": 935
},
{
"calibration/aurc": 0.20805025517207953,
"calibration/batch_distribution_entropy": 0.5147650665098341,
"calibration/batch_entropy_100bins": 0.2828651513916462,
"calibration/batch_entropy_10bins": 0.5147650665098341,
"calibration/batch_entropy_50bins": 0.332984279498441,
"calibration/batch_uniqueness": 0.13400243013850727,
"calibration/confidence_entropy": 0.4250079763322886,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.19530026109660575,
"calibration/coverage@15%": 0.2,
"calibration/coverage@20%": 0.3621052631578947,
"calibration/coverage@25%": 0.6842105263157895,
"calibration/coverage@30%": 0.8644356955380577,
"calibration/coverage@5%": 0.17806788511749347,
"calibration/distribution_entropy_10": 0.5147650665098341,
"calibration/distribution_entropy_100": 0.2828651513916462,
"calibration/ece": 0.13677418047081075,
"calibration/mean_confidence": 0.7845539229954371,
"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": 0.010416666666666652,
"completions/max_length": 4023.2,
"completions/max_terminated_length": 4023.2,
"completions/mean_length": 1196.5845947265625,
"completions/mean_terminated_length": 1209.4766357421875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 357.0,
"epoch": 2.2596153846153846,
"grad_norm": 0.00033075769897550344,
"learning_rate": 6.009615384615385e-07,
"loss": -0.0139,
"num_tokens": 2605555558.0,
"reward": 1.3028273344039918,
"reward_std": 0.12005721032619476,
"rewards/accuracy_reward": 0.7556423544883728,
"rewards/brier_reward": 0.8603268980979919,
"rewards/confidence_one_or_zero": 0.0035590277751907707,
"rewards/format_reward": 0.9896701455116272,
"rewards/mean_confidence_reward": 0.7611631751060486,
"sampling/batch_mean_priority_error": 0.015236111111111094,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.11111111111111112,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.003319880459457636,
"sampling/priority_kl": 0.030000341683626176,
"sampling/priority_scale": 0.9037518800469115,
"sampling/prob_entropy": 10.27896785736084,
"sampling/prob_max": 6.845462194178253e-05,
"sampling/prob_min": 2.0527686865534632e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.251200008392334,
"sampling/prompt_draws_total": 67536.0,
"sampling/seen_fraction": 0.9625133395195007,
"sampling/unseen_fraction": 0.03748666048049927,
"signal/accuracy_reward/centered_abs_mean": 0.09748806357383728,
"signal/accuracy_reward/group_std_mean": 0.13585958927869796,
"signal/accuracy_reward/group_zero_std_frac": 0.5805555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04874403178691864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04874403178691864,
"signal/advantage_abs_mean": 0.07968758940696716,
"signal/advantage_pre_scale_abs_mean": 0.07968758940696716,
"signal/advantage_pre_scale_std": 0.1783791333436966,
"signal/advantage_std": 0.1783791333436966,
"signal/brier_reward/centered_abs_mean": 0.06847795248031616,
"signal/brier_reward/group_std_mean": 0.09856122732162476,
"signal/brier_reward/group_zero_std_frac": 0.272222226858139,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03423897624015808,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03423897624015808,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0056152343284338714,
"signal/confidence_one_or_zero/group_std_mean": 0.009219340793788434,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9666666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.615234499600774e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.615234499600774e-08,
"signal/format_reward/centered_abs_mean": 0.01794162318110466,
"signal/format_reward/group_std_mean": 0.03940186202526093,
"signal/format_reward/group_zero_std_frac": 0.8138888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00897081159055233,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00897081159055233,
"signal/mean_confidence_reward/centered_abs_mean": 0.05832188725471497,
"signal/mean_confidence_reward/group_std_mean": 0.0822516843676567,
"signal/mean_confidence_reward/group_zero_std_frac": 0.28611111342906953,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.832188662679983e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.832188662679983e-07,
"step": 940
},
{
"calibration/aurc": 0.08275949176278782,
"calibration/batch_distribution_entropy": 0.5367319900238796,
"calibration/batch_entropy_100bins": 0.29038277184202704,
"calibration/batch_entropy_10bins": 0.5367319900238796,
"calibration/batch_entropy_50bins": 0.34183390065854946,
"calibration/batch_uniqueness": 0.15807961196908696,
"calibration/confidence_entropy": 0.4283059576867362,
"calibration/coverage@0%": 0.0020929242417979613,
"calibration/coverage@1%": 0.1342078067483254,
"calibration/coverage@10%": 0.5568662763841568,
"calibration/coverage@15%": 0.7838610378188214,
"calibration/coverage@20%": 0.9398416886543537,
"calibration/coverage@25%": 0.9398416886543537,
"calibration/coverage@30%": 0.9889182058047494,
"calibration/coverage@5%": 0.4866640950946416,
"calibration/distribution_entropy_10": 0.5367319900238796,
"calibration/distribution_entropy_100": 0.29038277184202704,
"calibration/ece": 0.0891567783594712,
"calibration/mean_confidence": 0.7716399624464515,
"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": 0.0073784722222222324,
"completions/max_length": 3970.2,
"completions/max_terminated_length": 3970.2,
"completions/mean_length": 1171.341259765625,
"completions/mean_terminated_length": 1180.1314208984375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 366.8,
"epoch": 2.2716346153846154,
"grad_norm": 0.0003024759062100202,
"learning_rate": 5.709134615384615e-07,
"loss": -0.0076,
"num_tokens": 2622145377.0,
"reward": 1.3309629917144776,
"reward_std": 0.10374844670295716,
"rewards/accuracy_reward": 0.7901041746139527,
"rewards/brier_reward": 0.8792714834213257,
"rewards/confidence_one_or_zero": 0.0013888889399822802,
"rewards/format_reward": 0.9925347208976746,
"rewards/mean_confidence_reward": 0.7810798525810242,
"sampling/batch_mean_priority_error": 0.02181430555555554,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08611111111111111,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.003340022638440132,
"sampling/priority_kl": 0.029998552426695822,
"sampling/priority_scale": 0.9088814555900171,
"sampling/prob_entropy": 10.278948783874512,
"sampling/prob_max": 6.890085205668584e-05,
"sampling/prob_min": 2.052076088148169e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.2631999492645263,
"sampling/prompt_draws_total": 67896.0,
"sampling/seen_fraction": 0.9637333393096924,
"sampling/unseen_fraction": 0.036266660690307616,
"signal/accuracy_reward/centered_abs_mean": 0.09259982705116272,
"signal/accuracy_reward/group_std_mean": 0.12963927239179612,
"signal/accuracy_reward/group_zero_std_frac": 0.5972222208976745,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04629991352558136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04629991352558136,
"signal/advantage_abs_mean": 0.06978111416101455,
"signal/advantage_pre_scale_abs_mean": 0.06978111416101455,
"signal/advantage_pre_scale_std": 0.1631460428237915,
"signal/advantage_std": 0.1631460428237915,
"signal/brier_reward/centered_abs_mean": 0.05654947534203529,
"signal/brier_reward/group_std_mean": 0.0824481189250946,
"signal/brier_reward/group_zero_std_frac": 0.32222222089767455,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.028274737671017645,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.028274737671017645,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0022894965135492385,
"signal/confidence_one_or_zero/group_std_mean": 0.0044072819873690605,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9805555462837219,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.2894964502029323e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.2894964502029323e-08,
"signal/format_reward/centered_abs_mean": 0.01277126707136631,
"signal/format_reward/group_std_mean": 0.02690165415406227,
"signal/format_reward/group_zero_std_frac": 0.8777777791023255,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.006385633535683155,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.006385633535683155,
"signal/mean_confidence_reward/centered_abs_mean": 0.05080154687166214,
"signal/mean_confidence_reward/group_std_mean": 0.07160323113203049,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3277777791023254,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.080154494407907e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.080154494407907e-07,
"step": 945
},
{
"calibration/aurc": 0.09588108721796844,
"calibration/batch_distribution_entropy": 0.5394386918957037,
"calibration/batch_entropy_100bins": 0.297117790418978,
"calibration/batch_entropy_10bins": 0.5394386918957037,
"calibration/batch_entropy_50bins": 0.34976225555564866,
"calibration/batch_uniqueness": 0.19306110715988942,
"calibration/confidence_entropy": 0.4337806183847327,
"calibration/coverage@0%": 0.0015764260249554365,
"calibration/coverage@1%": 0.0015764260249554365,
"calibration/coverage@10%": 0.4995804428607521,
"calibration/coverage@15%": 0.7345839123050795,
"calibration/coverage@20%": 0.9384760498687663,
"calibration/coverage@25%": 0.9557291666666666,
"calibration/coverage@30%": 0.9838541666666666,
"calibration/coverage@5%": 0.4185485880107233,
"calibration/distribution_entropy_10": 0.5394386918957037,
"calibration/distribution_entropy_100": 0.297117790418978,
"calibration/ece": 0.11819121550731962,
"calibration/mean_confidence": 0.7558088564255337,
"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": 0.009375,
"completions/max_length": 3963.4,
"completions/max_terminated_length": 3963.4,
"completions/mean_length": 1201.931005859375,
"completions/mean_terminated_length": 1213.7183837890625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 353.8,
"epoch": 2.2836538461538463,
"grad_norm": 0.00037571621942333877,
"learning_rate": 5.408653846153847e-07,
"loss": -0.0107,
"num_tokens": 2639093542.0,
"reward": 1.31483793258667,
"reward_std": 0.12279156893491745,
"rewards/accuracy_reward": 0.7750868201255798,
"rewards/brier_reward": 0.8638620734214782,
"rewards/confidence_one_or_zero": 0.0012152777868323027,
"rewards/format_reward": 0.9907118082046509,
"rewards/mean_confidence_reward": 0.7566597104072571,
"sampling/batch_mean_priority_error": 0.016486180555555542,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08888888888888888,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.003358766110613942,
"sampling/priority_kl": 0.02999931052327156,
"sampling/priority_scale": 0.9136200845008716,
"sampling/prob_entropy": 10.27894992828369,
"sampling/prob_max": 6.932988035259768e-05,
"sampling/prob_min": 2.0518549717962742e-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": 0.10987955629825592,
"signal/accuracy_reward/group_std_mean": 0.15317733138799666,
"signal/accuracy_reward/group_zero_std_frac": 0.5305555582046508,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05493977814912796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05493977814912796,
"signal/advantage_abs_mean": 0.08325956612825394,
"signal/advantage_pre_scale_abs_mean": 0.08325956612825394,
"signal/advantage_pre_scale_std": 0.18065005242824556,
"signal/advantage_std": 0.18065005242824556,
"signal/brier_reward/centered_abs_mean": 0.06960290074348449,
"signal/brier_reward/group_std_mean": 0.098955138027668,
"signal/brier_reward/group_zero_std_frac": 0.28611111342906953,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.034801450371742246,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.034801450371742246,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0022677951259538532,
"signal/confidence_one_or_zero/group_std_mean": 0.0049757368862628935,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9777777791023254,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.2677950539673474e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.2677950539673474e-08,
"signal/format_reward/centered_abs_mean": 0.016129557462409138,
"signal/format_reward/group_std_mean": 0.03346684370189905,
"signal/format_reward/group_zero_std_frac": 0.850000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008064778731204569,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008064778731204569,
"signal/mean_confidence_reward/centered_abs_mean": 0.057556481659412385,
"signal/mean_confidence_reward/group_std_mean": 0.07936900556087494,
"signal/mean_confidence_reward/group_zero_std_frac": 0.29722222983837127,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.755647976002365e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.755647976002365e-07,
"step": 950
},
{
"epoch": 2.2836538461538463,
"eval_calibration/aurc": 0.1323624339907363,
"eval_calibration/batch_distribution_entropy": 0.5527244664044493,
"eval_calibration/batch_entropy_100bins": 0.3033107718398318,
"eval_calibration/batch_entropy_10bins": 0.5527244664044493,
"eval_calibration/batch_entropy_50bins": 0.35705253308267776,
"eval_calibration/batch_uniqueness": 0.1874871369537685,
"eval_calibration/confidence_entropy": 0.4302961789890096,
"eval_calibration/coverage@0%": 0.001755926251097454,
"eval_calibration/coverage@1%": 0.001755926251097454,
"eval_calibration/coverage@10%": 0.001755926251097454,
"eval_calibration/coverage@15%": 0.7647058823529411,
"eval_calibration/coverage@20%": 0.8595258999122037,
"eval_calibration/coverage@25%": 0.9727831431079894,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.001755926251097454,
"eval_calibration/distribution_entropy_10": 0.5527244664044493,
"eval_calibration/distribution_entropy_100": 0.3033107718398318,
"eval_calibration/ece": 0.03213345039508359,
"eval_calibration/mean_confidence": 0.766988586479368,
"eval_calibration/unique_confidence_per_question": 0.008680555555555556,
"eval_calibration/unique_confidences": 10,
"eval_completions/clipped_ratio": 0.010416666666666666,
"eval_completions/max_length": 3121.0,
"eval_completions/max_terminated_length": 3121.0,
"eval_completions/mean_length": 1174.4849446614583,
"eval_completions/mean_terminated_length": 1187.0264282226562,
"eval_completions/min_length": 169.5,
"eval_completions/min_terminated_length": 426.6666666666667,
"eval_loss": 0.0,
"eval_num_tokens": 2639093542.0,
"eval_reward": 1.2817957599957783,
"eval_reward_std": 0.33336834609508514,
"eval_rewards/accuracy_reward": 0.7265624900658926,
"eval_rewards/brier_reward": 0.8482986390590668,
"eval_rewards/confidence_one_or_zero": 0.0017361111628512542,
"eval_rewards/format_reward": 0.988715281089147,
"eval_rewards/mean_confidence_reward": 0.7583333154519399,
"eval_runtime": 205.7147,
"eval_samples_per_second": 4.861,
"eval_signal/accuracy_reward/centered_abs_mean": 0.38330078125,
"eval_signal/accuracy_reward/group_std_mean": 0.4423869450887044,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.191650390625,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.191650390625,
"eval_signal/advantage_abs_mean": 0.2746702879667282,
"eval_signal/advantage_pre_scale_abs_mean": 0.2746702879667282,
"eval_signal/advantage_pre_scale_std": 0.3323294669389725,
"eval_signal/advantage_std": 0.3323294669389725,
"eval_signal/brier_reward/centered_abs_mean": 0.18063857903083166,
"eval_signal/brier_reward/group_std_mean": 0.24481445302565893,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.09031928951541583,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.09031928951541583,
"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": 0.021647135571887095,
"eval_signal/format_reward/group_std_mean": 0.05785721136877934,
"eval_signal/format_reward/group_zero_std_frac": 0.694444457689921,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.010823567785943547,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.010823567785943547,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.17593313256899515,
"eval_signal/mean_confidence_reward/group_std_mean": 0.22150815278291702,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.7593313259567367e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.7593313259567367e-06,
"eval_steps_per_second": 0.029,
"step": 950
},
{
"epoch": 2.2836538461538463,
"step": 950,
"train_probe_calibration/aurc": 0.08694457521157567,
"train_probe_calibration/batch_distribution_entropy": 0.5292848064008367,
"train_probe_calibration/batch_entropy_100bins": 0.2957835531105676,
"train_probe_calibration/batch_entropy_10bins": 0.5292848064008367,
"train_probe_calibration/batch_entropy_50bins": 0.3481916129839666,
"train_probe_calibration/batch_uniqueness": 0.14519625865892766,
"train_probe_calibration/confidence_entropy": 0.42145287989154906,
"train_probe_calibration/coverage@0%": 0.007936507936507936,
"train_probe_calibration/coverage@1%": 0.007936507936507936,
"train_probe_calibration/coverage@10%": 0.736331569664903,
"train_probe_calibration/coverage@15%": 0.8694885361552028,
"train_probe_calibration/coverage@20%": 0.935626102292769,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.007936507936507936,
"train_probe_calibration/distribution_entropy_10": 0.5292848064008367,
"train_probe_calibration/distribution_entropy_100": 0.2957835531105676,
"train_probe_calibration/ece": 0.0483245149911815,
"train_probe_calibration/mean_confidence": 0.7779541446208115,
"train_probe_calibration/unique_confidence_per_question": 0.008680555555555556,
"train_probe_calibration/unique_confidences": 10,
"train_probe_completions/clipped_ratio": 0.022916666666666658,
"train_probe_completions/max_length": 3290.1666666666665,
"train_probe_completions/max_terminated_length": 3290.1666666666665,
"train_probe_completions/mean_length": 1182.226094563802,
"train_probe_completions/mean_terminated_length": 1209.64501953125,
"train_probe_completions/min_length": 0.0,
"train_probe_completions/min_terminated_length": 403.8333333333333,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 2639093542.0,
"train_probe_reward": 1.311934808890025,
"train_probe_reward_std": 0.3206389049688975,
"train_probe_rewards/accuracy_reward": 0.7708333432674408,
"train_probe_rewards/brier_reward": 0.8686458369096121,
"train_probe_rewards/confidence_one_or_zero": 0.007812500155220429,
"train_probe_rewards/format_reward": 0.9843749900658926,
"train_probe_rewards/mean_confidence_reward": 0.7664062082767487,
"train_probe_runtime": 216.4321,
"train_probe_samples_per_second": 4.62,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3417968700329463,
"train_probe_signal/accuracy_reward/group_std_mean": 0.41666914025942486,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.17089843501647314,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.17089843501647314,
"train_probe_signal/advantage_abs_mean": 0.24869356552759805,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.24869356552759805,
"train_probe_signal/advantage_pre_scale_std": 0.32127704719702405,
"train_probe_signal/advantage_std": 0.32127704719702405,
"train_probe_signal/brier_reward/centered_abs_mean": 0.16127172112464905,
"train_probe_signal/brier_reward/group_std_mean": 0.22831493616104126,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08063586056232452,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.08063586056232452,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.015136718439559141,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.044194173688689865,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 0.750000019868215,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.5136717479435902e-07,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.5136717479435902e-07,
"train_probe_signal/format_reward/centered_abs_mean": 0.02983940951526165,
"train_probe_signal/format_reward/group_std_mean": 0.07643071251610915,
"train_probe_signal/format_reward/group_zero_std_frac": 0.6111111243565878,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.014919704757630825,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.014919704757630825,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.1732530097166697,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.21967636048793793,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.7325300518677977e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.7325300518677977e-06,
"train_probe_steps_per_second": 0.028
},
{
"calibration/aurc": 0.07333166444541113,
"calibration/batch_distribution_entropy": 0.4691600399329013,
"calibration/batch_entropy_100bins": 0.259992130435832,
"calibration/batch_entropy_10bins": 0.4691600399329013,
"calibration/batch_entropy_50bins": 0.30605852931163485,
"calibration/batch_uniqueness": 0.02508672457747697,
"calibration/confidence_entropy": 0.41341911692850275,
"calibration/coverage@0%": 0.1371778412351656,
"calibration/coverage@1%": 0.29612520965621825,
"calibration/coverage@10%": 0.6635685193374894,
"calibration/coverage@15%": 0.8417596776609555,
"calibration/coverage@20%": 0.913203132801689,
"calibration/coverage@25%": 0.943324250681199,
"calibration/coverage@30%": 0.9662125340599456,
"calibration/coverage@5%": 0.613810694638307,
"calibration/distribution_entropy_10": 0.4691600399329013,
"calibration/distribution_entropy_100": 0.259992130435832,
"calibration/ece": 0.1054306746835024,
"calibration/mean_confidence": 0.7830915209381537,
"calibration/unique_confidence_per_question": 0.02291666666666667,
"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": 0.008072916666666674,
"completions/max_length": 3900.2,
"completions/max_terminated_length": 3900.2,
"completions/mean_length": 1190.8841064453125,
"completions/mean_terminated_length": 1200.72021484375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 359.0,
"epoch": 2.295673076923077,
"grad_norm": 0.00038597677485086024,
"learning_rate": 5.108173076923077e-07,
"loss": -0.0088,
"num_tokens": 2655893167.0,
"reward": 1.3142507553100586,
"reward_std": 0.12009938955307006,
"rewards/accuracy_reward": 0.7626736164093018,
"rewards/brier_reward": 0.8738854169845581,
"rewards/confidence_one_or_zero": 0.006510416604578495,
"rewards/format_reward": 0.9919270873069763,
"rewards/mean_confidence_reward": 0.7650685906410217,
"sampling/batch_mean_priority_error": 0.014256944444444437,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.0611111111111111,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.0033722519874572754,
"sampling/priority_kl": 0.03000014051795006,
"sampling/priority_scale": 0.9171127259498462,
"sampling/prob_entropy": 10.278940582275391,
"sampling/prob_max": 6.968646630411968e-05,
"sampling/prob_min": 2.0528324967017397e-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": 0.1023220494389534,
"signal/accuracy_reward/group_std_mean": 0.14170491993427276,
"signal/accuracy_reward/group_zero_std_frac": 0.5666666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0511610247194767,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0511610247194767,
"signal/advantage_abs_mean": 0.0812824696302414,
"signal/advantage_pre_scale_abs_mean": 0.0812824696302414,
"signal/advantage_pre_scale_std": 0.17862094044685364,
"signal/advantage_std": 0.17862094044685364,
"signal/brier_reward/centered_abs_mean": 0.06971872299909591,
"signal/brier_reward/group_std_mean": 0.09945357143878937,
"signal/brier_reward/group_zero_std_frac": 0.2611111104488373,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03485936149954796,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03485936149954796,
"signal/confidence_one_or_zero/centered_abs_mean": 0.010997178964316845,
"signal/confidence_one_or_zero/group_std_mean": 0.01997890342026949,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.919444453716278,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.0997178350180547e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.0997178350180547e-07,
"signal/format_reward/centered_abs_mean": 0.014317491091787815,
"signal/format_reward/group_std_mean": 0.03275507166981697,
"signal/format_reward/group_zero_std_frac": 0.8416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0071587455458939075,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0071587455458939075,
"signal/mean_confidence_reward/centered_abs_mean": 0.05922640487551689,
"signal/mean_confidence_reward/group_std_mean": 0.08267918974161148,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2805555611848831,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.922640070821217e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.922640070821217e-07,
"step": 955
},
{
"calibration/aurc": 0.15454163559109715,
"calibration/batch_distribution_entropy": 0.5259110373168256,
"calibration/batch_entropy_100bins": 0.2926179457862498,
"calibration/batch_entropy_10bins": 0.5259110373168256,
"calibration/batch_entropy_50bins": 0.34446511126087703,
"calibration/batch_uniqueness": 0.18261172039536924,
"calibration/confidence_entropy": 0.4419454509507362,
"calibration/coverage@0%": 0.0031620826368647173,
"calibration/coverage@1%": 0.0031620826368647173,
"calibration/coverage@10%": 0.1914038723973695,
"calibration/coverage@15%": 0.6911513946873098,
"calibration/coverage@20%": 0.7164730321142104,
"calibration/coverage@25%": 0.7899632365839073,
"calibration/coverage@30%": 0.9133823932449123,
"calibration/coverage@5%": 0.08895155632107524,
"calibration/distribution_entropy_10": 0.5259110373168256,
"calibration/distribution_entropy_100": 0.2926179457862498,
"calibration/ece": 0.07072003946950807,
"calibration/mean_confidence": 0.7394015883287092,
"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": 0.009635416666666674,
"completions/max_length": 4009.6,
"completions/max_terminated_length": 4009.6,
"completions/mean_length": 1197.4577392578126,
"completions/mean_terminated_length": 1209.1400390625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 299.4,
"epoch": 2.3076923076923075,
"grad_norm": 0.00035911460872739553,
"learning_rate": 4.807692307692308e-07,
"loss": -0.0132,
"num_tokens": 2672770984.0,
"reward": 1.2979086875915526,
"reward_std": 0.115200474858284,
"rewards/accuracy_reward": 0.7491319417953491,
"rewards/brier_reward": 0.8563055396080017,
"rewards/confidence_one_or_zero": 0.00486111119389534,
"rewards/format_reward": 0.9903645753860474,
"rewards/mean_confidence_reward": 0.7599635362625122,
"sampling/batch_mean_priority_error": 0.02080034722222221,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.06666666666666667,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.003393551055341959,
"sampling/priority_kl": 0.02999952659010887,
"sampling/priority_scale": 0.9202498614555225,
"sampling/prob_entropy": 10.278947830200195,
"sampling/prob_max": 7.002019556239248e-05,
"sampling/prob_min": 2.0540579134831206e-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": 0.10064019113779069,
"signal/accuracy_reward/group_std_mean": 0.1408391058444977,
"signal/accuracy_reward/group_zero_std_frac": 0.5666666865348816,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05032009556889534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05032009556889534,
"signal/advantage_abs_mean": 0.07774469703435898,
"signal/advantage_pre_scale_abs_mean": 0.07774469703435898,
"signal/advantage_pre_scale_std": 0.1746388465166092,
"signal/advantage_std": 0.1746388465166092,
"signal/brier_reward/centered_abs_mean": 0.0663982793688774,
"signal/brier_reward/group_std_mean": 0.09372083991765975,
"signal/brier_reward/group_zero_std_frac": 0.30277777910232545,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0331991396844387,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0331991396844387,
"signal/confidence_one_or_zero/centered_abs_mean": 0.008007812686264515,
"signal/confidence_one_or_zero/group_std_mean": 0.013047228381037713,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9527777910232544,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 8.007812297705641e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.007812297705641e-08,
"signal/format_reward/centered_abs_mean": 0.01668294258415699,
"signal/format_reward/group_std_mean": 0.03328734934329987,
"signal/format_reward/group_zero_std_frac": 0.8555555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008341471292078495,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008341471292078495,
"signal/mean_confidence_reward/centered_abs_mean": 0.054650018364191054,
"signal/mean_confidence_reward/group_std_mean": 0.07545189410448075,
"signal/mean_confidence_reward/group_zero_std_frac": 0.31944444179534914,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.465001663651492e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.465001663651492e-07,
"step": 960
},
{
"calibration/aurc": 0.09435869514788404,
"calibration/batch_distribution_entropy": 0.5417798515703095,
"calibration/batch_entropy_100bins": 0.2991118932828889,
"calibration/batch_entropy_10bins": 0.5417798515703095,
"calibration/batch_entropy_50bins": 0.3521096811827305,
"calibration/batch_uniqueness": 0.195715446392127,
"calibration/confidence_entropy": 0.4327788700734908,
"calibration/coverage@0%": 0.00842801821291399,
"calibration/coverage@1%": 0.07491878338441794,
"calibration/coverage@10%": 0.510230192019455,
"calibration/coverage@15%": 0.8736405967921123,
"calibration/coverage@20%": 0.9055978336342175,
"calibration/coverage@25%": 0.9472295514511874,
"calibration/coverage@30%": 0.9804749340369394,
"calibration/coverage@5%": 0.42329687799096083,
"calibration/distribution_entropy_10": 0.5417798515703095,
"calibration/distribution_entropy_100": 0.2991118932828889,
"calibration/ece": 0.08588226087876341,
"calibration/mean_confidence": 0.7723784778485797,
"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": 0.009201388888888884,
"completions/max_length": 3968.0,
"completions/max_terminated_length": 3968.0,
"completions/mean_length": 1192.00224609375,
"completions/mean_terminated_length": 1203.069677734375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 362.2,
"epoch": 2.3197115384615383,
"grad_norm": 0.0003501853789202869,
"learning_rate": 4.507211538461539e-07,
"loss": -0.0082,
"num_tokens": 2689622018.0,
"reward": 1.3000178813934327,
"reward_std": 0.12227881848812103,
"rewards/accuracy_reward": 0.7453125,
"rewards/brier_reward": 0.8639963388442993,
"rewards/confidence_one_or_zero": 0.0023437499767169356,
"rewards/format_reward": 0.990711796283722,
"rewards/mean_confidence_reward": 0.753797733783722,
"sampling/batch_mean_priority_error": 0.02541145833333331,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.058333333333333334,
"sampling/error_ema_max": 0.20756810903549194,
"sampling/error_ema_mean": 0.0034138909075409176,
"sampling/priority_kl": 0.029999924823641777,
"sampling/priority_scale": 0.9233753741020336,
"sampling/prob_entropy": 10.278951263427734,
"sampling/prob_max": 7.034816226223484e-05,
"sampling/prob_min": 2.0551019042613915e-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": 0.10756293386220932,
"signal/accuracy_reward/group_std_mean": 0.1473521500825882,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111223697662,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05378146693110466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05378146693110466,
"signal/advantage_abs_mean": 0.08300841152668,
"signal/advantage_pre_scale_abs_mean": 0.08300841152668,
"signal/advantage_pre_scale_std": 0.1792859762907028,
"signal/advantage_std": 0.1792859762907028,
"signal/brier_reward/centered_abs_mean": 0.06932666748762131,
"signal/brier_reward/group_std_mean": 0.09935078769922256,
"signal/brier_reward/group_zero_std_frac": 0.2972222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.034663333743810656,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.034663333743810656,
"signal/confidence_one_or_zero/centered_abs_mean": 0.003640407882630825,
"signal/confidence_one_or_zero/group_std_mean": 0.0070123381447046995,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9694444417953492,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.6404079040153195e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.6404079040153195e-08,
"signal/format_reward/centered_abs_mean": 0.015728081949055193,
"signal/format_reward/group_std_mean": 0.0342374749481678,
"signal/format_reward/group_zero_std_frac": 0.8416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007864040974527597,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007864040974527597,
"signal/mean_confidence_reward/centered_abs_mean": 0.058252228796482085,
"signal/mean_confidence_reward/group_std_mean": 0.08077710419893265,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3111111164093018,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.825222729072265e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.825222729072265e-07,
"step": 965
},
{
"calibration/aurc": 0.1283695774898897,
"calibration/batch_distribution_entropy": 0.4090666755266697,
"calibration/batch_entropy_100bins": 0.2291534308213528,
"calibration/batch_entropy_10bins": 0.4090666755266697,
"calibration/batch_entropy_50bins": 0.26975571109144936,
"calibration/batch_uniqueness": -0.11918714253546994,
"calibration/confidence_entropy": 0.3987882773655907,
"calibration/coverage@0%": 0.006338832649334719,
"calibration/coverage@1%": 0.006338832649334719,
"calibration/coverage@10%": 0.33537109071385085,
"calibration/coverage@15%": 0.6480689601414199,
"calibration/coverage@20%": 0.8397426788405131,
"calibration/coverage@25%": 0.9216353890922493,
"calibration/coverage@30%": 0.9573457688029435,
"calibration/coverage@5%": 0.31146316060632395,
"calibration/distribution_entropy_10": 0.4090666755266697,
"calibration/distribution_entropy_100": 0.2291534308213528,
"calibration/ece": 0.10992615151961782,
"calibration/mean_confidence": 0.7916775578849042,
"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": 0.011371527777777769,
"completions/max_length": 3996.6,
"completions/max_terminated_length": 3996.6,
"completions/mean_length": 1157.7234619140625,
"completions/mean_terminated_length": 1171.0572021484375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 340.0,
"epoch": 2.331730769230769,
"grad_norm": 0.0003276313655078411,
"learning_rate": 4.20673076923077e-07,
"loss": -0.0134,
"num_tokens": 2706052112.0,
"reward": 1.2972458362579347,
"reward_std": 0.1148213267326355,
"rewards/accuracy_reward": 0.7394965291023254,
"rewards/brier_reward": 0.8667853713035584,
"rewards/confidence_one_or_zero": 0.007812499930150807,
"rewards/format_reward": 0.9881944417953491,
"rewards/mean_confidence_reward": 0.7607065796852112,
"sampling/batch_mean_priority_error": 0.01752951388888887,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.075,
"sampling/error_ema_max": 0.214666748046875,
"sampling/error_ema_mean": 0.0034340286627411844,
"sampling/priority_kl": 0.029998504742980003,
"sampling/priority_scale": 0.9262121139792725,
"sampling/prob_entropy": 10.278935623168945,
"sampling/prob_max": 7.06596183590591e-05,
"sampling/prob_min": 2.0564094666042364e-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": 0.09217122495174408,
"signal/accuracy_reward/group_std_mean": 0.13487593978643417,
"signal/accuracy_reward/group_zero_std_frac": 0.5666666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04608561247587204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04608561247587204,
"signal/advantage_abs_mean": 0.0762186512351036,
"signal/advantage_pre_scale_abs_mean": 0.0762186512351036,
"signal/advantage_pre_scale_std": 0.17515968084335326,
"signal/advantage_std": 0.17515968084335326,
"signal/brier_reward/centered_abs_mean": 0.06812559142708778,
"signal/brier_reward/group_std_mean": 0.0956755131483078,
"signal/brier_reward/group_zero_std_frac": 0.325,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03406279571354389,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03406279571354389,
"signal/confidence_one_or_zero/centered_abs_mean": 0.011111111100763082,
"signal/confidence_one_or_zero/group_std_mean": 0.016250798664987087,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9472222328186035,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.1111110467254549e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.1111110467254549e-07,
"signal/format_reward/centered_abs_mean": 0.019780815951526164,
"signal/format_reward/group_std_mean": 0.03664836846292019,
"signal/format_reward/group_zero_std_frac": 0.850000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009890407975763082,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009890407975763082,
"signal/mean_confidence_reward/centered_abs_mean": 0.05831093341112137,
"signal/mean_confidence_reward/group_std_mean": 0.07978967428207398,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3305555582046509,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.831093289998535e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.831093289998535e-07,
"step": 970
},
{
"calibration/aurc": 0.07742137009449009,
"calibration/batch_distribution_entropy": 0.4473659611295971,
"calibration/batch_entropy_100bins": 0.2406289326352915,
"calibration/batch_entropy_10bins": 0.4473659611295971,
"calibration/batch_entropy_50bins": 0.2832644861547543,
"calibration/batch_uniqueness": -0.030094964409126422,
"calibration/confidence_entropy": 0.40054666523372673,
"calibration/coverage@0%": 0.004720330152492104,
"calibration/coverage@1%": 0.19002216742283334,
"calibration/coverage@10%": 0.564974172832612,
"calibration/coverage@15%": 0.8678512720113531,
"calibration/coverage@20%": 0.9352835711953708,
"calibration/coverage@25%": 0.968969470921398,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.5474675415328771,
"calibration/distribution_entropy_10": 0.4473659611295971,
"calibration/distribution_entropy_100": 0.2406289326352915,
"calibration/ece": 0.09206130155147745,
"calibration/mean_confidence": 0.8047266900851249,
"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": 0.00980902777777779,
"completions/max_length": 4012.4,
"completions/max_terminated_length": 4012.4,
"completions/mean_length": 1177.9795166015624,
"completions/mean_terminated_length": 1189.7114501953124,
"completions/min_length": 0.0,
"completions/min_terminated_length": 375.6,
"epoch": 2.34375,
"grad_norm": 0.0003336299560032785,
"learning_rate": 3.90625e-07,
"loss": -0.011,
"num_tokens": 2722713668.0,
"reward": 1.3038686752319335,
"reward_std": 0.12683693766593934,
"rewards/accuracy_reward": 0.7487847328186035,
"rewards/brier_reward": 0.8687464833259583,
"rewards/confidence_one_or_zero": 0.006163194449618459,
"rewards/format_reward": 0.9901909708976746,
"rewards/mean_confidence_reward": 0.7557253837585449,
"sampling/batch_mean_priority_error": 0.017069444444444432,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.041666666666666664,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.0034511482808738947,
"sampling/priority_kl": 0.02999942600727081,
"sampling/priority_scale": 0.9295170843834057,
"sampling/prob_entropy": 10.27896728515625,
"sampling/prob_max": 7.100841903593392e-05,
"sampling/prob_min": 2.057480924122501e-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": 0.10644531100988389,
"signal/accuracy_reward/group_std_mean": 0.1484037458896637,
"signal/accuracy_reward/group_zero_std_frac": 0.5416666626930237,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05322265550494194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05322265550494194,
"signal/advantage_abs_mean": 0.08296125084161758,
"signal/advantage_pre_scale_abs_mean": 0.08296125084161758,
"signal/advantage_pre_scale_std": 0.18262528777122497,
"signal/advantage_std": 0.18262528777122497,
"signal/brier_reward/centered_abs_mean": 0.06895918846130371,
"signal/brier_reward/group_std_mean": 0.10257805585861206,
"signal/brier_reward/group_zero_std_frac": 0.27500000298023225,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.034479594230651854,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.034479594230651854,
"signal/confidence_one_or_zero/centered_abs_mean": 0.00921766497194767,
"signal/confidence_one_or_zero/group_std_mean": 0.013176932744681836,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9583333373069763,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 9.217664711513863e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 9.217664711513863e-08,
"signal/format_reward/centered_abs_mean": 0.01813693568110466,
"signal/format_reward/group_std_mean": 0.041134567186236384,
"signal/format_reward/group_zero_std_frac": 0.8055555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00906846784055233,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00906846784055233,
"signal/mean_confidence_reward/centered_abs_mean": 0.05696495622396469,
"signal/mean_confidence_reward/group_std_mean": 0.08042566478252411,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3000000059604645,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.69649512272008e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.69649512272008e-07,
"step": 975
},
{
"calibration/aurc": 0.114188890279602,
"calibration/batch_distribution_entropy": 0.509483514083351,
"calibration/batch_entropy_100bins": 0.27925843126992217,
"calibration/batch_entropy_10bins": 0.509483514083351,
"calibration/batch_entropy_50bins": 0.3287385069273901,
"calibration/batch_uniqueness": 0.10562402381743607,
"calibration/confidence_entropy": 0.4224627921340617,
"calibration/coverage@0%": 0.004688859878154917,
"calibration/coverage@1%": 0.004688859878154917,
"calibration/coverage@10%": 0.48668718904833363,
"calibration/coverage@15%": 0.8230040970112507,
"calibration/coverage@20%": 0.8691468565882465,
"calibration/coverage@25%": 0.9188481675392669,
"calibration/coverage@30%": 0.9785340314136125,
"calibration/coverage@5%": 0.13314838990426459,
"calibration/distribution_entropy_10": 0.509483514083351,
"calibration/distribution_entropy_100": 0.27925843126992217,
"calibration/ece": 0.08088774396367386,
"calibration/mean_confidence": 0.7717182568799339,
"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": 0.009288194444444443,
"completions/max_length": 3904.8,
"completions/max_terminated_length": 3904.8,
"completions/mean_length": 1206.857666015625,
"completions/mean_terminated_length": 1218.132958984375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 376.0,
"epoch": 2.355769230769231,
"grad_norm": 0.0003652828745543957,
"learning_rate": 3.6057692307692306e-07,
"loss": -0.0112,
"num_tokens": 2739742364.0,
"reward": 1.2970961809158326,
"reward_std": 0.12603443413972854,
"rewards/accuracy_reward": 0.7352430582046509,
"rewards/brier_reward": 0.868222451210022,
"rewards/confidence_one_or_zero": 0.006076388992369175,
"rewards/format_reward": 0.9907118082046509,
"rewards/mean_confidence_reward": 0.7478255152702331,
"sampling/batch_mean_priority_error": 0.012765624999999978,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.0638888888888889,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.0034659527707844974,
"sampling/priority_kl": 0.03000069931149483,
"sampling/priority_scale": 0.9330740154488012,
"sampling/prob_entropy": 10.278945541381836,
"sampling/prob_max": 7.137841748772189e-05,
"sampling/prob_min": 2.058420177490916e-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": 0.10789930522441864,
"signal/accuracy_reward/group_std_mean": 0.15151995718479155,
"signal/accuracy_reward/group_zero_std_frac": 0.5305555582046508,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05394965261220932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05394965261220932,
"signal/advantage_abs_mean": 0.08410682827234268,
"signal/advantage_pre_scale_abs_mean": 0.08410682827234268,
"signal/advantage_pre_scale_std": 0.18120644092559815,
"signal/advantage_std": 0.18120644092559815,
"signal/brier_reward/centered_abs_mean": 0.07255458235740661,
"signal/brier_reward/group_std_mean": 0.10437788367271424,
"signal/brier_reward/group_zero_std_frac": 0.2472222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.036277291178703305,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.036277291178703305,
"signal/confidence_one_or_zero/centered_abs_mean": 0.008268229104578495,
"signal/confidence_one_or_zero/group_std_mean": 0.01167407687753439,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9638888716697693,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 8.268229265695481e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.268229265695481e-08,
"signal/format_reward/centered_abs_mean": 0.01691080741584301,
"signal/format_reward/group_std_mean": 0.03558777719736099,
"signal/format_reward/group_zero_std_frac": 0.8416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008455403707921505,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008455403707921505,
"signal/mean_confidence_reward/centered_abs_mean": 0.06018635854125023,
"signal/mean_confidence_reward/group_std_mean": 0.0855872243642807,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2555555522441864,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.018635644977622e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.018635644977622e-07,
"step": 980
},
{
"calibration/aurc": 0.09821316486708462,
"calibration/batch_distribution_entropy": 0.4763177103621107,
"calibration/batch_entropy_100bins": 0.2620128684370206,
"calibration/batch_entropy_10bins": 0.4763177103621107,
"calibration/batch_entropy_50bins": 0.30843730939136715,
"calibration/batch_uniqueness": 0.03894753407550304,
"calibration/confidence_entropy": 0.41985677353875783,
"calibration/coverage@0%": 0.0010443935427574171,
"calibration/coverage@1%": 0.3315063263525305,
"calibration/coverage@10%": 0.5141361256544503,
"calibration/coverage@15%": 0.6827754260888939,
"calibration/coverage@20%": 0.8849085343704444,
"calibration/coverage@25%": 0.9062789084181313,
"calibration/coverage@30%": 0.9482308048103608,
"calibration/coverage@5%": 0.3743946335078534,
"calibration/distribution_entropy_10": 0.4763177103621107,
"calibration/distribution_entropy_100": 0.2620128684370206,
"calibration/ece": 0.10219283998519561,
"calibration/mean_confidence": 0.768914136498992,
"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": 0.010850694444444442,
"completions/max_length": 4002.2,
"completions/max_terminated_length": 4002.2,
"completions/mean_length": 1192.258154296875,
"completions/mean_terminated_length": 1205.37119140625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 324.6,
"epoch": 2.3677884615384617,
"grad_norm": 0.00036673585418611765,
"learning_rate": 3.305288461538462e-07,
"loss": -0.0131,
"num_tokens": 2756580154.0,
"reward": 1.3056840419769287,
"reward_std": 0.12273909598588943,
"rewards/accuracy_reward": 0.7542534708976746,
"rewards/brier_reward": 0.8679502487182618,
"rewards/confidence_one_or_zero": 0.0036458333488553762,
"rewards/format_reward": 0.9891493082046509,
"rewards/mean_confidence_reward": 0.7502586841583252,
"sampling/batch_mean_priority_error": 0.015890624999999985,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.07222222222222223,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.003479621047154069,
"sampling/priority_kl": 0.03000035434961319,
"sampling/priority_scale": 0.9369861424202099,
"sampling/prob_entropy": 10.278960037231446,
"sampling/prob_max": 7.176910003181547e-05,
"sampling/prob_min": 2.0589240375556982e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.3592000007629395,
"sampling/prompt_draws_total": 70776.0,
"sampling/seen_fraction": 0.9701933264732361,
"sampling/unseen_fraction": 0.029806673526763916,
"signal/accuracy_reward/centered_abs_mean": 0.11163736879825592,
"signal/accuracy_reward/group_std_mean": 0.15212086737155914,
"signal/accuracy_reward/group_zero_std_frac": 0.5444444596767426,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05581868439912796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05581868439912796,
"signal/advantage_abs_mean": 0.08209312856197357,
"signal/advantage_pre_scale_abs_mean": 0.08209312856197357,
"signal/advantage_pre_scale_std": 0.1826746553182602,
"signal/advantage_std": 0.1826746553182602,
"signal/brier_reward/centered_abs_mean": 0.06789509430527688,
"signal/brier_reward/group_std_mean": 0.09828180372714997,
"signal/brier_reward/group_zero_std_frac": 0.2916666746139526,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03394754715263844,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03394754715263844,
"signal/confidence_one_or_zero/centered_abs_mean": 0.005815972248092294,
"signal/confidence_one_or_zero/group_std_mean": 0.00891453055664897,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9694444417953492,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.815972059508567e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.815972059508567e-08,
"signal/format_reward/centered_abs_mean": 0.01933051198720932,
"signal/format_reward/group_std_mean": 0.04063206985592842,
"signal/format_reward/group_zero_std_frac": 0.8166666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00966525599360466,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00966525599360466,
"signal/mean_confidence_reward/centered_abs_mean": 0.05723828449845314,
"signal/mean_confidence_reward/group_std_mean": 0.08099256753921509,
"signal/mean_confidence_reward/group_zero_std_frac": 0.31666667461395265,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.723828166992462e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.723828166992462e-07,
"step": 985
},
{
"calibration/aurc": 0.12899346784143634,
"calibration/batch_distribution_entropy": 0.6010847128653459,
"calibration/batch_entropy_100bins": 0.335635511985207,
"calibration/batch_entropy_10bins": 0.6010847128653459,
"calibration/batch_entropy_50bins": 0.39510469417189975,
"calibration/batch_uniqueness": 0.32869682683431645,
"calibration/confidence_entropy": 0.4531429864179647,
"calibration/coverage@0%": 0.0027100271002710027,
"calibration/coverage@1%": 0.1635455362386522,
"calibration/coverage@10%": 0.5229942110518833,
"calibration/coverage@15%": 0.6890601465083612,
"calibration/coverage@20%": 0.7557347769854298,
"calibration/coverage@25%": 0.8328105328297377,
"calibration/coverage@30%": 0.8637517871241706,
"calibration/coverage@5%": 0.3896951780608053,
"calibration/distribution_entropy_10": 0.6010847128653459,
"calibration/distribution_entropy_100": 0.335635511985207,
"calibration/ece": 0.1153655501319042,
"calibration/mean_confidence": 0.7248026823679614,
"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": 0.013368055555555536,
"completions/max_length": 4035.0,
"completions/max_terminated_length": 4035.0,
"completions/mean_length": 1169.717041015625,
"completions/mean_terminated_length": 1185.7587158203125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 349.0,
"epoch": 2.3798076923076925,
"grad_norm": 0.00033228841493837535,
"learning_rate": 3.0048076923076924e-07,
"loss": -0.0134,
"num_tokens": 2773171422.0,
"reward": 1.3092052459716796,
"reward_std": 0.11055757701396943,
"rewards/accuracy_reward": 0.7555555701255798,
"rewards/brier_reward": 0.8762081742286683,
"rewards/confidence_one_or_zero": 0.002951388928340748,
"rewards/format_reward": 0.9866319298744202,
"rewards/mean_confidence_reward": 0.7412042856216431,
"sampling/batch_mean_priority_error": 0.014855972222222204,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.06666666666666665,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.003495658095926046,
"sampling/priority_kl": 0.029999981075525282,
"sampling/priority_scale": 0.9406435906654224,
"sampling/prob_entropy": 10.27895278930664,
"sampling/prob_max": 7.214548531919718e-05,
"sampling/prob_min": 2.0596579270204528e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.3711999893188476,
"sampling/prompt_draws_total": 71136.0,
"sampling/seen_fraction": 0.9709399938583374,
"sampling/unseen_fraction": 0.029060006141662598,
"signal/accuracy_reward/centered_abs_mean": 0.08971354067325592,
"signal/accuracy_reward/group_std_mean": 0.131047847867012,
"signal/accuracy_reward/group_zero_std_frac": 0.5777777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04485677033662796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04485677033662796,
"signal/advantage_abs_mean": 0.07056974172592163,
"signal/advantage_pre_scale_abs_mean": 0.07056974172592163,
"signal/advantage_pre_scale_std": 0.16588001251220702,
"signal/advantage_std": 0.16588001251220702,
"signal/brier_reward/centered_abs_mean": 0.0633417621254921,
"signal/brier_reward/group_std_mean": 0.0932878315448761,
"signal/brier_reward/group_zero_std_frac": 0.29722222983837127,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03167088106274605,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03167088106274605,
"signal/confidence_one_or_zero/centered_abs_mean": 0.00477430549217388,
"signal/confidence_one_or_zero/group_std_mean": 0.007662679813802242,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222089767456,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.774305395471856e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.774305395471856e-08,
"signal/format_reward/centered_abs_mean": 0.018674045242369174,
"signal/format_reward/group_std_mean": 0.03792152032256126,
"signal/format_reward/group_zero_std_frac": 0.8305555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009337022621184587,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009337022621184587,
"signal/mean_confidence_reward/centered_abs_mean": 0.05140403136610985,
"signal/mean_confidence_reward/group_std_mean": 0.07264350578188897,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3305555611848831,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.140403288805828e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.140403288805828e-07,
"step": 990
},
{
"calibration/aurc": 0.12118961159305712,
"calibration/batch_distribution_entropy": 0.5614932309347664,
"calibration/batch_entropy_100bins": 0.312126605886408,
"calibration/batch_entropy_10bins": 0.5614932309347664,
"calibration/batch_entropy_50bins": 0.3674303902833075,
"calibration/batch_uniqueness": 0.24348487145883607,
"calibration/confidence_entropy": 0.45600217907986407,
"calibration/coverage@0%": 0.11927083333333333,
"calibration/coverage@1%": 0.11927083333333333,
"calibration/coverage@10%": 0.6682374557913351,
"calibration/coverage@15%": 0.7743050950486294,
"calibration/coverage@20%": 0.8369791666666668,
"calibration/coverage@25%": 0.865625,
"calibration/coverage@30%": 0.8802083333333334,
"calibration/coverage@5%": 0.396875,
"calibration/distribution_entropy_10": 0.5614932309347664,
"calibration/distribution_entropy_100": 0.312126605886408,
"calibration/ece": 0.08444297082228104,
"calibration/mean_confidence": 0.7185261936339524,
"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": 0.007725694444444442,
"completions/max_length": 3925.6,
"completions/max_terminated_length": 3925.6,
"completions/mean_length": 1177.221826171875,
"completions/mean_terminated_length": 1186.37841796875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 351.0,
"epoch": 2.391826923076923,
"grad_norm": 0.0003700850938912481,
"learning_rate": 2.7043269230769233e-07,
"loss": -0.0092,
"num_tokens": 2789834041.0,
"reward": 1.311913752555847,
"reward_std": 0.11160127222537994,
"rewards/accuracy_reward": 0.7585069417953492,
"rewards/brier_reward": 0.8730313658714295,
"rewards/confidence_one_or_zero": 0.0027777777751907705,
"rewards/format_reward": 0.9922743082046509,
"rewards/mean_confidence_reward": 0.7425931811332702,
"sampling/batch_mean_priority_error": 0.014538194444444439,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.0638888888888889,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.0035095302388072013,
"sampling/priority_kl": 0.03000053986907005,
"sampling/priority_scale": 0.945136696123518,
"sampling/prob_entropy": 10.278954315185548,
"sampling/prob_max": 7.25828402210027e-05,
"sampling/prob_min": 2.0597735419869424e-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": 0.09546440988779067,
"signal/accuracy_reward/group_std_mean": 0.13176625818014145,
"signal/accuracy_reward/group_zero_std_frac": 0.6083333373069764,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04773220494389534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04773220494389534,
"signal/advantage_abs_mean": 0.07565358579158783,
"signal/advantage_pre_scale_abs_mean": 0.07565358579158783,
"signal/advantage_pre_scale_std": 0.17233096063137054,
"signal/advantage_std": 0.17233096063137054,
"signal/brier_reward/centered_abs_mean": 0.062432841211557386,
"signal/brier_reward/group_std_mean": 0.08957359939813614,
"signal/brier_reward/group_zero_std_frac": 0.2944444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.031216420605778693,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.031216420605778693,
"signal/confidence_one_or_zero/centered_abs_mean": 0.004915364505723119,
"signal/confidence_one_or_zero/group_std_mean": 0.008829268533736468,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9666666507720947,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.915364328894611e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.915364328894611e-08,
"signal/format_reward/centered_abs_mean": 0.014274088852107524,
"signal/format_reward/group_std_mean": 0.03213131241500378,
"signal/format_reward/group_zero_std_frac": 0.85,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007137044426053762,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007137044426053762,
"signal/mean_confidence_reward/centered_abs_mean": 0.05760388597846031,
"signal/mean_confidence_reward/group_std_mean": 0.07929947078227997,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3083333343267441,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.760388432918262e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.760388432918262e-07,
"step": 995
},
{
"calibration/aurc": 0.1082374968660619,
"calibration/batch_distribution_entropy": 0.5258876466513132,
"calibration/batch_entropy_100bins": 0.2916314963423535,
"calibration/batch_entropy_10bins": 0.5258876466513132,
"calibration/batch_entropy_50bins": 0.3433038789361407,
"calibration/batch_uniqueness": 0.1551850622828222,
"calibration/confidence_entropy": 0.4261764350444436,
"calibration/coverage@0%": 0.1582286656589254,
"calibration/coverage@1%": 0.18799367871375566,
"calibration/coverage@10%": 0.6456064650845373,
"calibration/coverage@15%": 0.7005630981791973,
"calibration/coverage@20%": 0.7442322830248527,
"calibration/coverage@25%": 0.90729044834308,
"calibration/coverage@30%": 0.9736842105263157,
"calibration/coverage@5%": 0.29036588798198704,
"calibration/distribution_entropy_10": 0.5258876466513132,
"calibration/distribution_entropy_100": 0.2916314963423535,
"calibration/ece": 0.10422118197110612,
"calibration/mean_confidence": 0.7664825552288578,
"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": 0.01206597222222221,
"completions/max_length": 4046.0,
"completions/max_terminated_length": 4046.0,
"completions/mean_length": 1195.227685546875,
"completions/mean_terminated_length": 1209.8916748046875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 318.4,
"epoch": 2.4038461538461537,
"grad_norm": 0.0003808954788837582,
"learning_rate": 2.403846153846154e-07,
"loss": -0.0137,
"num_tokens": 2806684056.0,
"reward": 1.3327399015426635,
"reward_std": 0.1265719160437584,
"rewards/accuracy_reward": 0.7993055582046509,
"rewards/brier_reward": 0.8782251477241516,
"rewards/confidence_one_or_zero": 0.007378472317941487,
"rewards/format_reward": 0.9879340291023254,
"rewards/mean_confidence_reward": 0.7483344674110413,
"sampling/batch_mean_priority_error": 0.015536458333333319,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.05277777777777778,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.003524286998435855,
"sampling/priority_kl": 0.029999404773116113,
"sampling/priority_scale": 0.9481643735663965,
"sampling/prob_entropy": 10.278943824768067,
"sampling/prob_max": 7.291929505299777e-05,
"sampling/prob_min": 2.060989645542577e-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": 0.11416015625,
"signal/accuracy_reward/group_std_mean": 0.154748073220253,
"signal/accuracy_reward/group_zero_std_frac": 0.5444444417953491,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.057080078125,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.057080078125,
"signal/advantage_abs_mean": 0.08525396138429642,
"signal/advantage_pre_scale_abs_mean": 0.08525396138429642,
"signal/advantage_pre_scale_std": 0.18693980872631072,
"signal/advantage_std": 0.18693980872631072,
"signal/brier_reward/centered_abs_mean": 0.07114665731787681,
"signal/brier_reward/group_std_mean": 0.10072609037160873,
"signal/brier_reward/group_zero_std_frac": 0.2611111134290695,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.035573328658938405,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.035573328658938405,
"signal/confidence_one_or_zero/centered_abs_mean": 0.010801866371184587,
"signal/confidence_one_or_zero/group_std_mean": 0.01483048414811492,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9555555582046509,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.0801865997223104e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.0801865997223104e-07,
"signal/format_reward/centered_abs_mean": 0.021066622994840144,
"signal/format_reward/group_std_mean": 0.044086290150880815,
"signal/format_reward/group_zero_std_frac": 0.8027777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010533311497420072,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010533311497420072,
"signal/mean_confidence_reward/centered_abs_mean": 0.060543371737003325,
"signal/mean_confidence_reward/group_std_mean": 0.08616852164268493,
"signal/mean_confidence_reward/group_zero_std_frac": 0.26388889253139497,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.054337063687854e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.054337063687854e-07,
"step": 1000
},
{
"epoch": 2.4038461538461537,
"eval_calibration/aurc": 0.13409442568496843,
"eval_calibration/batch_distribution_entropy": 0.568377548088194,
"eval_calibration/batch_entropy_100bins": 0.31589399351948105,
"eval_calibration/batch_entropy_10bins": 0.568377548088194,
"eval_calibration/batch_entropy_50bins": 0.3718652980491398,
"eval_calibration/batch_uniqueness": 0.2256740864367637,
"eval_calibration/confidence_entropy": 0.43497557334362497,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.6070484581497797,
"eval_calibration/coverage@15%": 0.7436123348017621,
"eval_calibration/coverage@20%": 0.8440528634361234,
"eval_calibration/coverage@25%": 0.9198237885462555,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.568377548088194,
"eval_calibration/distribution_entropy_100": 0.31589399351948105,
"eval_calibration/ece": 0.030616740088105585,
"eval_calibration/mean_confidence": 0.7574008810572688,
"eval_calibration/unique_confidence_per_question": 0.009548611111111112,
"eval_calibration/unique_confidences": 11,
"eval_completions/clipped_ratio": 0.013020833333333334,
"eval_completions/max_length": 3174.1666666666665,
"eval_completions/max_terminated_length": 3174.1666666666665,
"eval_completions/mean_length": 1153.855448404948,
"eval_completions/mean_terminated_length": 1169.1691080729167,
"eval_completions/min_length": 92.5,
"eval_completions/min_terminated_length": 431.0,
"eval_loss": 0.0,
"eval_num_tokens": 2806684056.0,
"eval_reward": 1.2744812965393066,
"eval_reward_std": 0.33850590387980145,
"eval_rewards/accuracy_reward": 0.7196180522441864,
"eval_rewards/brier_reward": 0.8440863788127899,
"eval_rewards/confidence_one_or_zero": 0.0034722223257025084,
"eval_rewards/format_reward": 0.9852430522441864,
"eval_rewards/mean_confidence_reward": 0.7462239265441895,
"eval_runtime": 214.7505,
"eval_samples_per_second": 4.657,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3906792551279068,
"eval_signal/accuracy_reward/group_std_mean": 0.4469083696603775,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1953396275639534,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.1953396275639534,
"eval_signal/advantage_abs_mean": 0.27881846328576404,
"eval_signal/advantage_pre_scale_abs_mean": 0.27881846328576404,
"eval_signal/advantage_pre_scale_std": 0.3381160447994868,
"eval_signal/advantage_std": 0.3381160447994868,
"eval_signal/brier_reward/centered_abs_mean": 0.18172378093004227,
"eval_signal/brier_reward/group_std_mean": 0.2443449025352796,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.09086189046502113,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.09086189046502113,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.006727430348594983,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.019641855110724766,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 0.8888888955116272,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 6.72743022770798e-08,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 6.72743022770798e-08,
"eval_signal/format_reward/centered_abs_mean": 0.028591579447189968,
"eval_signal/format_reward/group_std_mean": 0.08347788328925769,
"eval_signal/format_reward/group_zero_std_frac": 0.5277777860562006,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.014295789723594984,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.014295789723594984,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.18783635646104813,
"eval_signal/mean_confidence_reward/group_std_mean": 0.23139711221059164,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.8783634156231226e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.8783634156231226e-06,
"eval_steps_per_second": 0.028,
"step": 1000
},
{
"epoch": 2.4038461538461537,
"step": 1000,
"train_probe_calibration/aurc": 0.08278325344477341,
"train_probe_calibration/batch_distribution_entropy": 0.5657352345973774,
"train_probe_calibration/batch_entropy_100bins": 0.3148932591301308,
"train_probe_calibration/batch_entropy_10bins": 0.5657352345973774,
"train_probe_calibration/batch_entropy_50bins": 0.37068724971774347,
"train_probe_calibration/batch_uniqueness": 0.22923492591324754,
"train_probe_calibration/confidence_entropy": 0.4316876402690614,
"train_probe_calibration/coverage@0%": 0.006993006993006993,
"train_probe_calibration/coverage@1%": 0.006993006993006993,
"train_probe_calibration/coverage@10%": 0.756993006993007,
"train_probe_calibration/coverage@15%": 0.8487762237762237,
"train_probe_calibration/coverage@20%": 0.9755244755244755,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.006993006993006993,
"train_probe_calibration/distribution_entropy_10": 0.5657352345973774,
"train_probe_calibration/distribution_entropy_100": 0.3148932591301308,
"train_probe_calibration/ece": 0.04641608391608378,
"train_probe_calibration/mean_confidence": 0.7628496503496504,
"train_probe_calibration/unique_confidence_per_question": 0.008680555555555556,
"train_probe_calibration/unique_confidences": 10,
"train_probe_completions/clipped_ratio": 0.010243055555555566,
"train_probe_completions/max_length": 3426.6666666666665,
"train_probe_completions/max_terminated_length": 3426.6666666666665,
"train_probe_completions/mean_length": 1182.697245279948,
"train_probe_completions/mean_terminated_length": 1195.1104532877605,
"train_probe_completions/min_length": 0.0,
"train_probe_completions/min_terminated_length": 391.3333333333333,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 2806684056.0,
"train_probe_reward": 1.3287967244784038,
"train_probe_reward_std": 0.2908843557039897,
"train_probe_rewards/accuracy_reward": 0.7838541766007742,
"train_probe_rewards/brier_reward": 0.880668431520462,
"train_probe_rewards/confidence_one_or_zero": 0.006944444651405017,
"train_probe_rewards/format_reward": 0.9930555621782938,
"train_probe_rewards/mean_confidence_reward": 0.7575520773728689,
"train_probe_runtime": 210.6731,
"train_probe_samples_per_second": 4.747,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3299153695503871,
"train_probe_signal/accuracy_reward/group_std_mean": 0.40984514355659485,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.16495768477519354,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.16495768477519354,
"train_probe_signal/advantage_abs_mean": 0.22570937126874924,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.22570937126874924,
"train_probe_signal/advantage_pre_scale_std": 0.2904963940382004,
"train_probe_signal/advantage_std": 0.2904963940382004,
"train_probe_signal/brier_reward/centered_abs_mean": 0.1419260154167811,
"train_probe_signal/brier_reward/group_std_mean": 0.20473192383845648,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07096300770839055,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07096300770839055,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.013454860697189966,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.03928371022144953,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 0.7777778108914694,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.3454859981720801e-07,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.3454859981720801e-07,
"train_probe_signal/format_reward/centered_abs_mean": 0.013454860697189966,
"train_probe_signal/format_reward/group_std_mean": 0.03928371022144953,
"train_probe_signal/format_reward/group_zero_std_frac": 0.7777778108914694,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.006727430348594983,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.006727430348594983,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.17776689926783243,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.21993445108334223,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.7776690318290396e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.7776690318290396e-06,
"train_probe_steps_per_second": 0.028
},
{
"calibration/aurc": 0.08983140347452864,
"calibration/batch_distribution_entropy": 0.4862920985635162,
"calibration/batch_entropy_100bins": 0.26166824612290984,
"calibration/batch_entropy_10bins": 0.4862920985635162,
"calibration/batch_entropy_50bins": 0.3080316255791384,
"calibration/batch_uniqueness": 0.058742779627740674,
"calibration/confidence_entropy": 0.4193815918272631,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.7181727908967728,
"calibration/coverage@15%": 0.8119686864028626,
"calibration/coverage@20%": 0.8841824854644381,
"calibration/coverage@25%": 0.9353757635253054,
"calibration/coverage@30%": 0.9875,
"calibration/coverage@5%": 0.5207319418073775,
"calibration/distribution_entropy_10": 0.4862920985635162,
"calibration/distribution_entropy_100": 0.26166824612290984,
"calibration/ece": 0.1097107746709409,
"calibration/mean_confidence": 0.779109564129616,
"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": 0.009635416666666674,
"completions/max_length": 4010.0,
"completions/max_terminated_length": 4010.0,
"completions/mean_length": 1164.0114501953126,
"completions/mean_terminated_length": 1175.432861328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 329.6,
"epoch": 2.4158653846153846,
"grad_norm": 0.00032606004970148206,
"learning_rate": 2.103365384615385e-07,
"loss": -0.01,
"num_tokens": 2823159004.0,
"reward": 1.3159709930419923,
"reward_std": 0.11136936545372009,
"rewards/accuracy_reward": 0.7671875,
"rewards/brier_reward": 0.8743745565414429,
"rewards/confidence_one_or_zero": 0.002864583331393078,
"rewards/format_reward": 0.9903645873069763,
"rewards/mean_confidence_reward": 0.7671875119209289,
"sampling/batch_mean_priority_error": 0.01744618055555554,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.061111111111111116,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.0035401007160544397,
"sampling/priority_kl": 0.0300002608448267,
"sampling/priority_scale": 0.9519850552314892,
"sampling/prob_entropy": 10.278963851928712,
"sampling/prob_max": 7.331292872549965e-05,
"sampling/prob_min": 2.0615904577425682e-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": 0.09576822966337203,
"signal/accuracy_reward/group_std_mean": 0.1338461920619011,
"signal/accuracy_reward/group_zero_std_frac": 0.5805555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04788411483168602,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04788411483168602,
"signal/advantage_abs_mean": 0.07476246803998947,
"signal/advantage_pre_scale_abs_mean": 0.07476246803998947,
"signal/advantage_pre_scale_std": 0.16831520795822144,
"signal/advantage_std": 0.16831520795822144,
"signal/brier_reward/centered_abs_mean": 0.060066214948892596,
"signal/brier_reward/group_std_mean": 0.08783653080463409,
"signal/brier_reward/group_zero_std_frac": 0.3194444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.030033107474446298,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.030033107474446298,
"signal/confidence_one_or_zero/centered_abs_mean": 0.004714626760687679,
"signal/confidence_one_or_zero/group_std_mean": 0.007716905884444713,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.7146265558239975e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.7146265558239975e-08,
"signal/format_reward/centered_abs_mean": 0.016531033255159856,
"signal/format_reward/group_std_mean": 0.03449917696416378,
"signal/format_reward/group_zero_std_frac": 0.8416666626930237,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008265516627579928,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008265516627579928,
"signal/mean_confidence_reward/centered_abs_mean": 0.0552905336022377,
"signal/mean_confidence_reward/group_std_mean": 0.0773746132850647,
"signal/mean_confidence_reward/group_zero_std_frac": 0.32777778506278993,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.529053282771202e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.529053282771202e-07,
"step": 1005
},
{
"calibration/aurc": 0.21375878198864506,
"calibration/batch_distribution_entropy": 0.5777321783955085,
"calibration/batch_entropy_100bins": 0.31826796586466405,
"calibration/batch_entropy_10bins": 0.5777321783955085,
"calibration/batch_entropy_50bins": 0.3746598998833386,
"calibration/batch_uniqueness": 0.2724663195034364,
"calibration/confidence_entropy": 0.4344012037873786,
"calibration/coverage@0%": 0.005265935286765727,
"calibration/coverage@1%": 0.005265935286765727,
"calibration/coverage@10%": 0.23780679094163043,
"calibration/coverage@15%": 0.4462542287974838,
"calibration/coverage@20%": 0.4462542287974838,
"calibration/coverage@25%": 0.6007957518989147,
"calibration/coverage@30%": 0.6465743777587609,
"calibration/coverage@5%": 0.22250336086247474,
"calibration/distribution_entropy_10": 0.5777321783955085,
"calibration/distribution_entropy_100": 0.31826796586466405,
"calibration/ece": 0.14821627013700106,
"calibration/mean_confidence": 0.7294570024087697,
"calibration/unique_confidence_per_question": 0.024479166666666666,
"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": 0.010069444444444419,
"completions/max_length": 3955.6,
"completions/max_terminated_length": 3955.6,
"completions/mean_length": 1155.7947021484374,
"completions/mean_terminated_length": 1167.70869140625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 309.4,
"epoch": 2.4278846153846154,
"grad_norm": 0.00034449115628376603,
"learning_rate": 1.8028846153846153e-07,
"loss": -0.0115,
"num_tokens": 2839577855.0,
"reward": 1.3057581901550293,
"reward_std": 0.11530058681964875,
"rewards/accuracy_reward": 0.7427083373069763,
"rewards/brier_reward": 0.8788626194000244,
"rewards/confidence_one_or_zero": 0.0037326389632653446,
"rewards/format_reward": 0.9899305582046509,
"rewards/mean_confidence_reward": 0.7416854858398437,
"sampling/batch_mean_priority_error": 0.014730902777777763,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08333333333333334,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.0035560850985348223,
"sampling/priority_kl": 0.02999974824488163,
"sampling/priority_scale": 0.9572434485191479,
"sampling/prob_entropy": 10.27894992828369,
"sampling/prob_max": 7.380725000984966e-05,
"sampling/prob_min": 2.060995284409728e-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": 0.09190538227558136,
"signal/accuracy_reward/group_std_mean": 0.13165127336978913,
"signal/accuracy_reward/group_zero_std_frac": 0.5777777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04595269113779068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04595269113779068,
"signal/advantage_abs_mean": 0.07484828606247902,
"signal/advantage_pre_scale_abs_mean": 0.07484828606247902,
"signal/advantage_pre_scale_std": 0.17094075083732604,
"signal/advantage_std": 0.17094075083732604,
"signal/brier_reward/centered_abs_mean": 0.06410380974411964,
"signal/brier_reward/group_std_mean": 0.09359335005283356,
"signal/brier_reward/group_zero_std_frac": 0.2833333373069763,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03205190487205982,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03205190487205982,
"signal/confidence_one_or_zero/centered_abs_mean": 0.006060112884733826,
"signal/confidence_one_or_zero/group_std_mean": 0.01077386476099491,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9583333373069763,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 6.060112909267445e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 6.060112909267445e-08,
"signal/format_reward/centered_abs_mean": 0.018250868655741215,
"signal/format_reward/group_std_mean": 0.039764121547341345,
"signal/format_reward/group_zero_std_frac": 0.8166666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009125434327870607,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009125434327870607,
"signal/mean_confidence_reward/centered_abs_mean": 0.05567202791571617,
"signal/mean_confidence_reward/group_std_mean": 0.07979700714349747,
"signal/mean_confidence_reward/group_zero_std_frac": 0.30277777910232545,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.567202720158093e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.567202720158093e-07,
"step": 1010
},
{
"calibration/aurc": 0.10069863430886601,
"calibration/batch_distribution_entropy": 0.5385201084042638,
"calibration/batch_entropy_100bins": 0.295772450904362,
"calibration/batch_entropy_10bins": 0.5385201084042638,
"calibration/batch_entropy_50bins": 0.34817854364645356,
"calibration/batch_uniqueness": 0.18474362211384246,
"calibration/confidence_entropy": 0.4417251183958671,
"calibration/coverage@0%": 0.1368421052631579,
"calibration/coverage@1%": 0.2763157894736842,
"calibration/coverage@10%": 0.6819344535481182,
"calibration/coverage@15%": 0.7071976114428551,
"calibration/coverage@20%": 0.7562380224968754,
"calibration/coverage@25%": 0.8,
"calibration/coverage@30%": 0.8908136482939633,
"calibration/coverage@5%": 0.5855627809056849,
"calibration/distribution_entropy_10": 0.5385201084042638,
"calibration/distribution_entropy_100": 0.295772450904362,
"calibration/ece": 0.13028352871666612,
"calibration/mean_confidence": 0.736554621132017,
"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": 0.010243055555555557,
"completions/max_length": 3879.6,
"completions/max_terminated_length": 3879.6,
"completions/mean_length": 1163.62529296875,
"completions/mean_terminated_length": 1175.6705322265625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 369.0,
"epoch": 2.4399038461538463,
"grad_norm": 0.0003656467597465962,
"learning_rate": 1.5024038461538462e-07,
"loss": -0.0113,
"num_tokens": 2856091874.0,
"reward": 1.3181779623031615,
"reward_std": 0.11705977618694305,
"rewards/accuracy_reward": 0.7717881798744202,
"rewards/brier_reward": 0.8749693989753723,
"rewards/confidence_one_or_zero": 0.002690972265554592,
"rewards/format_reward": 0.9895833373069763,
"rewards/mean_confidence_reward": 0.7427213311195373,
"sampling/batch_mean_priority_error": 0.014354166666666657,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.061111111111111116,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.003569081239402294,
"sampling/priority_kl": 0.02999938577413559,
"sampling/priority_scale": 0.9616323411231861,
"sampling/prob_entropy": 10.278946685791016,
"sampling/prob_max": 7.424467476084829e-05,
"sampling/prob_min": 2.0611210493370892e-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": 0.1018825963139534,
"signal/accuracy_reward/group_std_mean": 0.14123183190822602,
"signal/accuracy_reward/group_zero_std_frac": 0.569444453716278,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0509412981569767,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0509412981569767,
"signal/advantage_abs_mean": 0.07748131155967712,
"signal/advantage_pre_scale_abs_mean": 0.07748131155967712,
"signal/advantage_pre_scale_std": 0.17561177611351014,
"signal/advantage_std": 0.17561177611351014,
"signal/brier_reward/centered_abs_mean": 0.0639573760330677,
"signal/brier_reward/group_std_mean": 0.09368740767240524,
"signal/brier_reward/group_zero_std_frac": 0.2722222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03197868801653385,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03197868801653385,
"signal/confidence_one_or_zero/centered_abs_mean": 0.004356553789693862,
"signal/confidence_one_or_zero/group_std_mean": 0.007231736136600375,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222328186035,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.35655380215394e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.35655380215394e-08,
"signal/format_reward/centered_abs_mean": 0.018543837033212184,
"signal/format_reward/group_std_mean": 0.040525621920824054,
"signal/format_reward/group_zero_std_frac": 0.8111111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009271918516606092,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009271918516606092,
"signal/mean_confidence_reward/centered_abs_mean": 0.05968533381819725,
"signal/mean_confidence_reward/group_std_mean": 0.08249245584011078,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2805555522441864,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.968533400846354e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.968533400846354e-07,
"step": 1015
},
{
"calibration/aurc": 0.07653606272396447,
"calibration/batch_distribution_entropy": 0.6288167408136538,
"calibration/batch_entropy_100bins": 0.3500217781385155,
"calibration/batch_entropy_10bins": 0.6288167408136538,
"calibration/batch_entropy_50bins": 0.4120399739197384,
"calibration/batch_uniqueness": 0.39337521507351514,
"calibration/confidence_entropy": 0.455890063812096,
"calibration/coverage@0%": 0.3141049782428512,
"calibration/coverage@1%": 0.46896834852880237,
"calibration/coverage@10%": 0.7289743490122945,
"calibration/coverage@15%": 0.7870126657687526,
"calibration/coverage@20%": 0.8191179289266473,
"calibration/coverage@25%": 0.8948741193535017,
"calibration/coverage@30%": 0.9583333333333333,
"calibration/coverage@5%": 0.5418569553805774,
"calibration/distribution_entropy_10": 0.6288167408136538,
"calibration/distribution_entropy_100": 0.3500217781385155,
"calibration/ece": 0.14023634997927878,
"calibration/mean_confidence": 0.7137680877883685,
"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": 0.009809027777777767,
"completions/max_length": 3967.6,
"completions/max_terminated_length": 3967.6,
"completions/mean_length": 1186.3420166015626,
"completions/mean_terminated_length": 1198.19521484375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 382.8,
"epoch": 2.451923076923077,
"grad_norm": 0.0003450646181590855,
"learning_rate": 1.201923076923077e-07,
"loss": -0.0112,
"num_tokens": 2872886630.0,
"reward": 1.3077133417129516,
"reward_std": 0.1105430081486702,
"rewards/accuracy_reward": 0.7447916626930237,
"rewards/brier_reward": 0.8804292678833008,
"rewards/confidence_one_or_zero": 0.004253472259733826,
"rewards/format_reward": 0.9901909708976746,
"rewards/mean_confidence_reward": 0.7359375,
"sampling/batch_mean_priority_error": 0.019133680555555543,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.05,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.0035847860854119062,
"sampling/priority_kl": 0.029998553916811943,
"sampling/priority_scale": 0.9654334722785279,
"sampling/prob_entropy": 10.27893943786621,
"sampling/prob_max": 7.464435184374451e-05,
"sampling/prob_min": 2.0355882952571847e-05,
"sampling/prompt_draws_max": 8.4,
"sampling/prompt_draws_mean": 2.4432000160217284,
"sampling/prompt_draws_total": 73296.0,
"sampling/seen_fraction": 0.9754599928855896,
"sampling/unseen_fraction": 0.0245400071144104,
"signal/accuracy_reward/centered_abs_mean": 0.09067925289273263,
"signal/accuracy_reward/group_std_mean": 0.12974078506231307,
"signal/accuracy_reward/group_zero_std_frac": 0.5972222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04533962644636631,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04533962644636631,
"signal/advantage_abs_mean": 0.07144246324896812,
"signal/advantage_pre_scale_abs_mean": 0.07144246324896812,
"signal/advantage_pre_scale_std": 0.16677022576332093,
"signal/advantage_std": 0.16677022576332093,
"signal/brier_reward/centered_abs_mean": 0.06360199972987175,
"signal/brier_reward/group_std_mean": 0.093186055123806,
"signal/brier_reward/group_zero_std_frac": 0.27777778506278994,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.031800999864935876,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.031800999864935876,
"signal/confidence_one_or_zero/centered_abs_mean": 0.006819661497138441,
"signal/confidence_one_or_zero/group_std_mean": 0.010803023539483547,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9611111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 6.819661280133005e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 6.819661280133005e-08,
"signal/format_reward/centered_abs_mean": 0.01769205741584301,
"signal/format_reward/group_std_mean": 0.038584979623556136,
"signal/format_reward/group_zero_std_frac": 0.8222222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008846028707921505,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008846028707921505,
"signal/mean_confidence_reward/centered_abs_mean": 0.05903239473700524,
"signal/mean_confidence_reward/group_std_mean": 0.0819184049963951,
"signal/mean_confidence_reward/group_zero_std_frac": 0.28055556416511535,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.903239184590347e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.903239184590347e-07,
"step": 1020
},
{
"calibration/aurc": 0.07430514258786,
"calibration/batch_distribution_entropy": 0.5098176087204103,
"calibration/batch_entropy_100bins": 0.2840145574128153,
"calibration/batch_entropy_10bins": 0.5098176087204103,
"calibration/batch_entropy_50bins": 0.3343373416693276,
"calibration/batch_uniqueness": 0.1409113739591033,
"calibration/confidence_entropy": 0.4284641227215422,
"calibration/coverage@0%": 0.13056060585348517,
"calibration/coverage@1%": 0.31228384345139637,
"calibration/coverage@10%": 0.8268208768494343,
"calibration/coverage@15%": 0.8967770887728459,
"calibration/coverage@20%": 0.926019908616188,
"calibration/coverage@25%": 0.9536458333333332,
"calibration/coverage@30%": 0.9536458333333332,
"calibration/coverage@5%": 0.49786913500927277,
"calibration/distribution_entropy_10": 0.5098176087204103,
"calibration/distribution_entropy_100": 0.2840145574128153,
"calibration/ece": 0.11457569452038867,
"calibration/mean_confidence": 0.7698891784741114,
"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": 0.009374999999999977,
"completions/max_length": 3959.8,
"completions/max_terminated_length": 3959.8,
"completions/mean_length": 1172.4079833984374,
"completions/mean_terminated_length": 1183.5778564453126,
"completions/min_length": 0.0,
"completions/min_terminated_length": 331.4,
"epoch": 2.4639423076923075,
"grad_norm": 0.0004397702869027853,
"learning_rate": 9.014423076923076e-08,
"loss": -0.01,
"num_tokens": 2889486818.0,
"reward": 1.3201935291290283,
"reward_std": 0.11437996476888657,
"rewards/accuracy_reward": 0.7700520753860474,
"rewards/brier_reward": 0.8796947836875916,
"rewards/confidence_one_or_zero": 0.0035590278333984316,
"rewards/format_reward": 0.990625,
"rewards/mean_confidence_reward": 0.7564895629882813,
"sampling/batch_mean_priority_error": 0.013347291666666655,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.05,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.0035991712007671594,
"sampling/priority_kl": 0.02999994195997715,
"sampling/priority_scale": 0.969786411547102,
"sampling/prob_entropy": 10.278949737548828,
"sampling/prob_max": 7.508765411330387e-05,
"sampling/prob_min": 1.9963686645496638e-05,
"sampling/prompt_draws_max": 9.0,
"sampling/prompt_draws_mean": 2.4552000522613526,
"sampling/prompt_draws_total": 73656.0,
"sampling/seen_fraction": 0.9761600017547607,
"sampling/unseen_fraction": 0.023839998245239257,
"signal/accuracy_reward/centered_abs_mean": 0.0878634974360466,
"signal/accuracy_reward/group_std_mean": 0.1292222797870636,
"signal/accuracy_reward/group_zero_std_frac": 0.5722222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0439317487180233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0439317487180233,
"signal/advantage_abs_mean": 0.07393695563077926,
"signal/advantage_pre_scale_abs_mean": 0.07393695563077926,
"signal/advantage_pre_scale_std": 0.1688719093799591,
"signal/advantage_std": 0.1688719093799591,
"signal/brier_reward/centered_abs_mean": 0.06520559415221214,
"signal/brier_reward/group_std_mean": 0.09605432748794555,
"signal/brier_reward/group_zero_std_frac": 0.2666666656732559,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03260279707610607,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03260279707610607,
"signal/confidence_one_or_zero/centered_abs_mean": 0.005040147551335394,
"signal/confidence_one_or_zero/group_std_mean": 0.007720847195014357,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.0401475704120455e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.0401475704120455e-08,
"signal/format_reward/centered_abs_mean": 0.01625434048473835,
"signal/format_reward/group_std_mean": 0.03423248454928398,
"signal/format_reward/group_zero_std_frac": 0.8444444537162781,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008127170242369175,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008127170242369175,
"signal/mean_confidence_reward/centered_abs_mean": 0.05742301493883133,
"signal/mean_confidence_reward/group_std_mean": 0.08127100467681884,
"signal/mean_confidence_reward/group_zero_std_frac": 0.27777777910232543,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.742301368627523e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.742301368627523e-07,
"step": 1025
},
{
"calibration/aurc": 0.190737357158577,
"calibration/batch_distribution_entropy": 0.6315100168513081,
"calibration/batch_entropy_100bins": 0.3447852289048042,
"calibration/batch_entropy_10bins": 0.6315100168513081,
"calibration/batch_entropy_50bins": 0.40587559288847,
"calibration/batch_uniqueness": 0.3688717344268504,
"calibration/confidence_entropy": 0.45521653229510173,
"calibration/coverage@0%": 0.0005249343832020997,
"calibration/coverage@1%": 0.0005249343832020997,
"calibration/coverage@10%": 0.21187835813717001,
"calibration/coverage@15%": 0.5564304461942258,
"calibration/coverage@20%": 0.6035728518228416,
"calibration/coverage@25%": 0.7324071402069505,
"calibration/coverage@30%": 0.7968586387434555,
"calibration/coverage@5%": 0.10551181102362203,
"calibration/distribution_entropy_10": 0.6315100168513081,
"calibration/distribution_entropy_100": 0.3447852289048042,
"calibration/ece": 0.10167360074755058,
"calibration/mean_confidence": 0.7203472767998242,
"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": 0.008940972222222232,
"completions/max_length": 3926.6,
"completions/max_terminated_length": 3926.6,
"completions/mean_length": 1155.531884765625,
"completions/mean_terminated_length": 1166.0040283203125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 369.6,
"epoch": 2.4759615384615383,
"grad_norm": 0.00048662739573046565,
"learning_rate": 6.009615384615386e-08,
"loss": -0.0097,
"num_tokens": 2905924817.0,
"reward": 1.3112340927124024,
"reward_std": 0.11016112565994263,
"rewards/accuracy_reward": 0.7549479126930236,
"rewards/brier_reward": 0.876445984840393,
"rewards/confidence_one_or_zero": 0.0024305555562023073,
"rewards/format_reward": 0.9910590291023255,
"rewards/mean_confidence_reward": 0.7597517251968384,
"sampling/batch_mean_priority_error": 0.018508680555555542,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.03888888888888889,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.003616081550717354,
"sampling/priority_kl": 0.030000020191073416,
"sampling/priority_scale": 0.9718193710548804,
"sampling/prob_entropy": 10.278950881958007,
"sampling/prob_max": 7.537075434811413e-05,
"sampling/prob_min": 1.998338666453492e-05,
"sampling/prompt_draws_max": 9.0,
"sampling/prompt_draws_mean": 2.467199945449829,
"sampling/prompt_draws_total": 74016.0,
"sampling/seen_fraction": 0.9765533447265625,
"sampling/unseen_fraction": 0.0234466552734375,
"signal/accuracy_reward/centered_abs_mean": 0.09672851413488388,
"signal/accuracy_reward/group_std_mean": 0.13473242670297622,
"signal/accuracy_reward/group_zero_std_frac": 0.5861111164093018,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04836425706744194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04836425706744194,
"signal/advantage_abs_mean": 0.07378808557987213,
"signal/advantage_pre_scale_abs_mean": 0.07378808557987213,
"signal/advantage_pre_scale_std": 0.16930631101131438,
"signal/advantage_std": 0.16930631101131438,
"signal/brier_reward/centered_abs_mean": 0.06198948621749878,
"signal/brier_reward/group_std_mean": 0.08934446573257446,
"signal/brier_reward/group_zero_std_frac": 0.3194444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03099474310874939,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03099474310874939,
"signal/confidence_one_or_zero/centered_abs_mean": 0.004220920044463128,
"signal/confidence_one_or_zero/group_std_mean": 0.008400925993919372,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9638888835906982,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.220919791464439e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.220919791464439e-08,
"signal/format_reward/centered_abs_mean": 0.01584743931889534,
"signal/format_reward/group_std_mean": 0.03291997648775578,
"signal/format_reward/group_zero_std_frac": 0.8527777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00792371965944767,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00792371965944767,
"signal/mean_confidence_reward/centered_abs_mean": 0.05436865985393524,
"signal/mean_confidence_reward/group_std_mean": 0.07493493854999542,
"signal/mean_confidence_reward/group_zero_std_frac": 0.33333333730697634,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.43686576293112e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.43686576293112e-07,
"step": 1030
},
{
"calibration/aurc": 0.09454584045054779,
"calibration/batch_distribution_entropy": 0.5348206829324342,
"calibration/batch_entropy_100bins": 0.2949276423704754,
"calibration/batch_entropy_10bins": 0.5348206829324342,
"calibration/batch_entropy_50bins": 0.34718404870924985,
"calibration/batch_uniqueness": 0.15781809362199886,
"calibration/confidence_entropy": 0.4316328098559209,
"calibration/coverage@0%": 0.2692378857342571,
"calibration/coverage@1%": 0.2750121639494802,
"calibration/coverage@10%": 0.49451697127937333,
"calibration/coverage@15%": 0.8351814077458659,
"calibration/coverage@20%": 0.8727412423846823,
"calibration/coverage@25%": 0.8931067776327242,
"calibration/coverage@30%": 0.9379582789382072,
"calibration/coverage@5%": 0.48296841484892716,
"calibration/distribution_entropy_10": 0.5348206829324342,
"calibration/distribution_entropy_100": 0.2949276423704754,
"calibration/ece": 0.14140299592935995,
"calibration/mean_confidence": 0.7589548983025295,
"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": 0.007118055555555558,
"completions/max_length": 4002.2,
"completions/max_terminated_length": 4002.2,
"completions/mean_length": 1200.6076416015626,
"completions/mean_terminated_length": 1209.26611328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 324.4,
"epoch": 2.487980769230769,
"grad_norm": 0.00033582025207579136,
"learning_rate": 3.004807692307693e-08,
"loss": -0.0079,
"num_tokens": 2922816713.0,
"reward": 1.3218994140625,
"reward_std": 0.1117009773850441,
"rewards/accuracy_reward": 0.7754340291023254,
"rewards/brier_reward": 0.8754678845405579,
"rewards/confidence_one_or_zero": 0.003645833331393078,
"rewards/format_reward": 0.9928819537162781,
"rewards/mean_confidence_reward": 0.7542786240577698,
"sampling/batch_mean_priority_error": 0.015368055555555543,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.0611111111111111,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.003631386300548911,
"sampling/priority_kl": 0.030000073090195655,
"sampling/priority_scale": 0.9759026825660839,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 7.579407974844798e-05,
"sampling/prob_min": 1.9983825040981174e-05,
"sampling/prompt_draws_max": 9.0,
"sampling/prompt_draws_mean": 2.479199981689453,
"sampling/prompt_draws_total": 74376.0,
"sampling/seen_fraction": 0.9772333383560181,
"sampling/unseen_fraction": 0.022766661643981934,
"signal/accuracy_reward/centered_abs_mean": 0.09267035722732545,
"signal/accuracy_reward/group_std_mean": 0.1369502991437912,
"signal/accuracy_reward/group_zero_std_frac": 0.5472222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04633517861366272,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04633517861366272,
"signal/advantage_abs_mean": 0.07139594107866287,
"signal/advantage_pre_scale_abs_mean": 0.07139594107866287,
"signal/advantage_pre_scale_std": 0.1638278156518936,
"signal/advantage_std": 0.1638278156518936,
"signal/brier_reward/centered_abs_mean": 0.06340188980102539,
"signal/brier_reward/group_std_mean": 0.09231583029031754,
"signal/brier_reward/group_zero_std_frac": 0.2944444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.031700944900512694,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.031700944900512694,
"signal/confidence_one_or_zero/centered_abs_mean": 0.005414496490266174,
"signal/confidence_one_or_zero/group_std_mean": 0.008160660043358803,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222089767456,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.4144963712587925e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.4144963712587925e-08,
"signal/format_reward/centered_abs_mean": 0.013075086567550898,
"signal/format_reward/group_std_mean": 0.029334478825330735,
"signal/format_reward/group_zero_std_frac": 0.8611111164093017,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.006537543283775449,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.006537543283775449,
"signal/mean_confidence_reward/centered_abs_mean": 0.05938764438033104,
"signal/mean_confidence_reward/group_std_mean": 0.08216754347085953,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3055555522441864,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.938764388702112e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.938764388702112e-07,
"step": 1035
},
{
"calibration/aurc": 0.09311068544688798,
"calibration/batch_distribution_entropy": 0.49783279102907524,
"calibration/batch_entropy_100bins": 0.27292808943936137,
"calibration/batch_entropy_10bins": 0.49783279102907524,
"calibration/batch_entropy_50bins": 0.32128653094852655,
"calibration/batch_uniqueness": 0.08343998356816631,
"calibration/confidence_entropy": 0.4269823165432417,
"calibration/coverage@0%": 0.0005221932114882506,
"calibration/coverage@1%": 0.0005221932114882506,
"calibration/coverage@10%": 0.7010835291108992,
"calibration/coverage@15%": 0.7568982758052714,
"calibration/coverage@20%": 0.8080768529044391,
"calibration/coverage@25%": 0.9448275862068967,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.27050854970185123,
"calibration/distribution_entropy_10": 0.49783279102907524,
"calibration/distribution_entropy_100": 0.27292808943936137,
"calibration/ece": 0.10161390043986591,
"calibration/mean_confidence": 0.7690658569311672,
"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": 0.007638888888888906,
"completions/max_length": 3960.6,
"completions/max_terminated_length": 3960.6,
"completions/mean_length": 1175.173974609375,
"completions/mean_terminated_length": 1184.4427734375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 362.0,
"epoch": 2.5,
"grad_norm": 0.000330591865349561,
"learning_rate": 0.0,
"loss": -0.008,
"num_tokens": 2939416413.0,
"reward": 1.3254358530044557,
"reward_std": 0.10986914932727813,
"rewards/accuracy_reward": 0.7760416626930237,
"rewards/brier_reward": 0.8824537873268128,
"rewards/confidence_one_or_zero": 0.003559027856681496,
"rewards/format_reward": 0.9923611164093018,
"rewards/mean_confidence_reward": 0.7549175500869751,
"sampling/batch_mean_priority_error": 0.011868055555555548,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.05277777777777778,
"sampling/error_ema_max": 0.24306130409240723,
"sampling/error_ema_mean": 0.003644486190751195,
"sampling/priority_kl": 0.03000090755522251,
"sampling/priority_scale": 0.9802652895683422,
"sampling/prob_entropy": 10.278954124450683,
"sampling/prob_max": 7.624288991792127e-05,
"sampling/prob_min": 1.998272418859415e-05,
"sampling/prompt_draws_max": 9.0,
"sampling/prompt_draws_mean": 2.4911999702453613,
"sampling/prompt_draws_total": 74736.0,
"sampling/seen_fraction": 0.9778799891471863,
"sampling/unseen_fraction": 0.02212001085281372,
"signal/accuracy_reward/centered_abs_mean": 0.09396701455116271,
"signal/accuracy_reward/group_std_mean": 0.13123033344745635,
"signal/accuracy_reward/group_zero_std_frac": 0.5972222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04698350727558136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04698350727558136,
"signal/advantage_abs_mean": 0.07230262234807014,
"signal/advantage_pre_scale_abs_mean": 0.07230262234807014,
"signal/advantage_pre_scale_std": 0.1636380672454834,
"signal/advantage_std": 0.1636380672454834,
"signal/brier_reward/centered_abs_mean": 0.0633222870528698,
"signal/brier_reward/group_std_mean": 0.09267549812793732,
"signal/brier_reward/group_zero_std_frac": 0.26944445073604584,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0316611435264349,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0316611435264349,
"signal/confidence_one_or_zero/centered_abs_mean": 0.005799696198664606,
"signal/confidence_one_or_zero/group_std_mean": 0.009705995582044125,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9638888835906982,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.799696225494699e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.799696225494699e-08,
"signal/format_reward/centered_abs_mean": 0.014149305317550898,
"signal/format_reward/group_std_mean": 0.032183651253581046,
"signal/format_reward/group_zero_std_frac": 0.8472222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007074652658775449,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007074652658775449,
"signal/mean_confidence_reward/centered_abs_mean": 0.05784342586994171,
"signal/mean_confidence_reward/group_std_mean": 0.08098789751529693,
"signal/mean_confidence_reward/group_zero_std_frac": 0.28888889253139494,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.784342533843301e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.784342533843301e-07,
"step": 1040
},
{
"epoch": 2.5,
"step": 1040,
"total_flos": 0.0,
"train_loss": -0.011179522975817235,
"train_runtime": 247236.2535,
"train_samples_per_second": 0.303,
"train_steps_per_second": 0.004
}
],
"logging_steps": 5,
"max_steps": 1040,
"num_input_tokens_seen": 2939416413,
"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
}