QWEN3-HINT-CAST / trainer_state.json
tejeshbhalla's picture
Upload Qwen3 hint CAST LoRA checkpoint-100
1411207 verified
Raw
History Blame Contribute Delete
248 kB
{
"best_global_step": null,
"best_metric": null,
"best_model_checkpoint": null,
"epoch": 0.019443904335990667,
"eval_steps": 500,
"global_step": 100,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.06706837518140674,
"cast/masked_gap_abs_mean": 0.016313929837451724,
"cast/masked_gap_mean": -0.005374407065744435,
"cast/masked_negative_base_rate": 0.7414380516856909,
"cast/masked_positive_base_rate": 0.2585619401652366,
"cast/masked_sign_flipped_rate": 0.26113924384117126,
"cast/masked_teacher_negative_rate": 0.31567351426929235,
"cast/masked_teacher_positive_rate": 0.25186810083687305,
"cast/masked_token_adv_abs_mean": 0.8992901407182217,
"cast/masked_token_adv_max": 1.012508638203144,
"cast/masked_token_adv_mean": -0.2653462424641475,
"cast/masked_token_adv_min": -1.0473822094500065,
"cast/masked_token_weight_max": 1.149782944470644,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022766039761336287,
"cast/masked_wrong_good_token_rate": 0.19407086912542582,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0,
"clip_ratio/high_mean": 0.0,
"clip_ratio/low_mean": 0.0,
"clip_ratio/low_min": 0.0,
"clip_ratio/region_mean": 0.0,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4871.0,
"completions/max_terminated_length": 4871.0,
"completions/mean_length": 1550.90625,
"completions/mean_terminated_length": 1550.90625,
"completions/min_length": 383.0,
"completions/min_terminated_length": 383.0,
"entropy": 0.23464461648836732,
"epoch": 0.00019443904335990667,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.09508519619703293,
"learning_rate": 1e-06,
"loss": 0.17538844048976898,
"num_tokens": 109178.0,
"reward": 0.3125,
"reward_std": 0.467176616191864,
"rewards/math_answer_reward/mean": 0.3125,
"rewards/math_answer_reward/std": 0.467176616191864,
"sampling/importance_sampling_ratio/max": 2.6244285106658936,
"sampling/importance_sampling_ratio/mean": 0.7069504261016846,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.1511693000793457,
"sampling/sampling_logp_difference/mean": 0.009396504610776901,
"step": 1,
"step_time": 209.97842906991718
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.03226685058325529,
"cast/masked_gap_abs_mean": 0.018581754397018813,
"cast/masked_gap_mean": -0.0024932975559863735,
"cast/masked_negative_base_rate": 0.8672169931232929,
"cast/masked_positive_base_rate": 0.13278300035744905,
"cast/masked_sign_flipped_rate": 0.2815407491289079,
"cast/masked_teacher_negative_rate": 0.31669078301638365,
"cast/masked_teacher_positive_rate": 0.285829103551805,
"cast/masked_token_adv_abs_mean": 0.8603844037279487,
"cast/masked_token_adv_max": 1.0409232024103403,
"cast/masked_token_adv_mean": -0.20160473487339914,
"cast/masked_token_adv_min": -1.1414122991263866,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.030801058397628367,
"cast/masked_wrong_good_token_rate": 0.2492738973814994,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0016766129901952809,
"clip_ratio/high_mean": 0.0016766129901952809,
"clip_ratio/low_mean": 0.0007692674662393983,
"clip_ratio/low_min": 0.0007692674662393983,
"clip_ratio/region_mean": 0.0024458804655296262,
"completions/clipped_ratio": 0.0,
"completions/max_length": 7140.0,
"completions/max_terminated_length": 7140.0,
"completions/mean_length": 2050.6875,
"completions/mean_terminated_length": 2050.6875,
"completions/min_length": 759.0,
"completions/min_terminated_length": 759.0,
"entropy": 0.27609438938088715,
"epoch": 0.00038887808671981335,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.054419130086898804,
"learning_rate": 9.999351869855467e-07,
"loss": 0.1270785927772522,
"num_tokens": 252990.0,
"reward": 0.140625,
"reward_std": 0.3503824472427368,
"rewards/math_answer_reward/mean": 0.140625,
"rewards/math_answer_reward/std": 0.3503824472427368,
"sampling/importance_sampling_ratio/max": 2.74906849861145,
"sampling/importance_sampling_ratio/mean": 0.5171247124671936,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1105542182922363,
"sampling/sampling_logp_difference/mean": 0.009917838498950005,
"step": 2,
"step_time": 339.5407820280525
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.062333112698979676,
"cast/masked_gap_abs_mean": 0.018289597806870006,
"cast/masked_gap_mean": -0.0029588793083803466,
"cast/masked_negative_base_rate": 0.7616477459669113,
"cast/masked_positive_base_rate": 0.23835225403308868,
"cast/masked_sign_flipped_rate": 0.2517847728449851,
"cast/masked_teacher_negative_rate": 0.30535434233024716,
"cast/masked_teacher_positive_rate": 0.2538909511640668,
"cast/masked_token_adv_abs_mean": 0.8330747988075018,
"cast/masked_token_adv_max": 0.9871918242424726,
"cast/masked_token_adv_mean": -0.26372264232486486,
"cast/masked_token_adv_min": -1.112147880718112,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.0295983562245965,
"cast/masked_wrong_good_token_rate": 0.18945166072808206,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.00199096305641433,
"clip_ratio/high_mean": 0.00199096305641433,
"clip_ratio/low_mean": 0.0007739475249763927,
"clip_ratio/low_min": 0.0007739475249763927,
"clip_ratio/region_mean": 0.002764910583209712,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4727.0,
"completions/max_terminated_length": 4727.0,
"completions/mean_length": 1766.046875,
"completions/mean_terminated_length": 1766.046875,
"completions/min_length": 622.0,
"completions/min_terminated_length": 622.0,
"entropy": 0.24031613930128515,
"epoch": 0.00058331713007972,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.0843966007232666,
"learning_rate": 9.998703739710934e-07,
"loss": 0.1980438381433487,
"num_tokens": 376273.0,
"reward": 0.28125,
"reward_std": 0.4531635046005249,
"rewards/math_answer_reward/mean": 0.28125,
"rewards/math_answer_reward/std": 0.4531635046005249,
"sampling/importance_sampling_ratio/max": 2.953662872314453,
"sampling/importance_sampling_ratio/mean": 0.6725503206253052,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1528456211090088,
"sampling/sampling_logp_difference/mean": 0.008648661896586418,
"step": 3,
"step_time": 203.06000694204704
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.030673553119413555,
"cast/masked_gap_abs_mean": 0.01222511861124076,
"cast/masked_gap_mean": -0.0007789740551089608,
"cast/masked_negative_base_rate": 0.8826569858938456,
"cast/masked_positive_base_rate": 0.11734301201067865,
"cast/masked_sign_flipped_rate": 0.2769851302728057,
"cast/masked_teacher_negative_rate": 0.32738857762888074,
"cast/masked_teacher_positive_rate": 0.26550379768013954,
"cast/masked_token_adv_abs_mean": 0.9099478293210268,
"cast/masked_token_adv_max": 1.0598125457763672,
"cast/masked_token_adv_mean": -0.29554856242612004,
"cast/masked_token_adv_min": -1.19514299929142,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.025842450733762234,
"cast/masked_wrong_good_token_rate": 0.24631157680414617,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0019044730906898621,
"clip_ratio/high_mean": 0.0019044730906898621,
"clip_ratio/low_mean": 0.0006633740495090024,
"clip_ratio/low_min": 0.0006633740495090024,
"clip_ratio/region_mean": 0.002567847132013412,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5192.0,
"completions/max_terminated_length": 5192.0,
"completions/mean_length": 2206.953125,
"completions/mean_terminated_length": 2206.953125,
"completions/min_length": 642.0,
"completions/min_terminated_length": 642.0,
"entropy": 0.2553784267511219,
"epoch": 0.0007777561734396267,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.040571428835392,
"learning_rate": 9.9980556095664e-07,
"loss": 0.09751482307910919,
"num_tokens": 530342.0,
"reward": 0.171875,
"reward_std": 0.38025420904159546,
"rewards/math_answer_reward/mean": 0.171875,
"rewards/math_answer_reward/std": 0.38025420904159546,
"sampling/importance_sampling_ratio/max": 2.3473219871520996,
"sampling/importance_sampling_ratio/mean": 0.42666950821876526,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1528520584106445,
"sampling/sampling_logp_difference/mean": 0.009385108016431332,
"step": 4,
"step_time": 274.51655364289763
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.05575632944237441,
"cast/masked_gap_abs_mean": 0.01704631267057266,
"cast/masked_gap_mean": -0.0027906709407261587,
"cast/masked_negative_base_rate": 0.7582155726850033,
"cast/masked_positive_base_rate": 0.24178441986441612,
"cast/masked_sign_flipped_rate": 0.264379208907485,
"cast/masked_teacher_negative_rate": 0.2895504981279373,
"cast/masked_teacher_positive_rate": 0.26577389566227794,
"cast/masked_token_adv_abs_mean": 0.8047548364847898,
"cast/masked_token_adv_max": 1.0327837513759732,
"cast/masked_token_adv_mean": -0.1745495411887532,
"cast/masked_token_adv_min": -1.134199838154018,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.029508414503652602,
"cast/masked_wrong_good_token_rate": 0.2086228784173727,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0017162983185698977,
"clip_ratio/high_mean": 0.0017162983185698977,
"clip_ratio/low_mean": 0.0007781258227623766,
"clip_ratio/low_min": 0.0007781258227623766,
"clip_ratio/region_mean": 0.002494424157703179,
"completions/clipped_ratio": 0.046875,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 4493.0,
"completions/mean_length": 2122.015625,
"completions/mean_terminated_length": 1823.49169921875,
"completions/min_length": 367.0,
"completions/min_terminated_length": 367.0,
"entropy": 0.2263617506250739,
"epoch": 0.0009721952167995333,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.06416887044906616,
"learning_rate": 9.997407479421868e-07,
"loss": 0.13163697719573975,
"num_tokens": 677343.0,
"reward": 0.28125,
"reward_std": 0.4531635046005249,
"rewards/math_answer_reward/mean": 0.28125,
"rewards/math_answer_reward/std": 0.4531635046005249,
"sampling/importance_sampling_ratio/max": 2.781853437423706,
"sampling/importance_sampling_ratio/mean": 0.6729681491851807,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1930086612701416,
"sampling/sampling_logp_difference/mean": 0.008489049039781094,
"step": 5,
"step_time": 403.89269019200583
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.07838566676946357,
"cast/masked_gap_abs_mean": 0.021099451507325284,
"cast/masked_gap_mean": -0.0040204029587584955,
"cast/masked_negative_base_rate": 0.7048297440633178,
"cast/masked_positive_base_rate": 0.2951702519785613,
"cast/masked_sign_flipped_rate": 0.26443519443273544,
"cast/masked_teacher_negative_rate": 0.2972854902036488,
"cast/masked_teacher_positive_rate": 0.25616094563156366,
"cast/masked_token_adv_abs_mean": 0.8167687021195889,
"cast/masked_token_adv_max": 0.9936644462868571,
"cast/masked_token_adv_mean": -0.11984019191004336,
"cast/masked_token_adv_min": -1.0927594313398004,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03219947271281853,
"cast/masked_wrong_good_token_rate": 0.1860495285363868,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0016306078505294863,
"clip_ratio/high_mean": 0.0016306078505294863,
"clip_ratio/low_mean": 0.0006033903300703969,
"clip_ratio/low_min": 0.0006033903300703969,
"clip_ratio/region_mean": 0.0022339981769619044,
"completions/clipped_ratio": 0.0,
"completions/max_length": 7896.0,
"completions/max_terminated_length": 7896.0,
"completions/mean_length": 2249.171875,
"completions/mean_terminated_length": 2249.171875,
"completions/min_length": 590.0,
"completions/min_terminated_length": 590.0,
"entropy": 0.21270004753023386,
"epoch": 0.00116663426015944,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.08111836761236191,
"learning_rate": 9.996759349277335e-07,
"loss": 0.044016044586896896,
"num_tokens": 833802.0,
"reward": 0.359375,
"reward_std": 0.4836103618144989,
"rewards/math_answer_reward/mean": 0.359375,
"rewards/math_answer_reward/std": 0.4836103618144989,
"sampling/importance_sampling_ratio/max": 2.485407829284668,
"sampling/importance_sampling_ratio/mean": 0.6084213852882385,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.9799323081970215,
"sampling/sampling_logp_difference/mean": 0.007981308735907078,
"step": 6,
"step_time": 398.1688974738936
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.061362551408819854,
"cast/masked_gap_abs_mean": 0.015886812274402473,
"cast/masked_gap_mean": -0.002449131108733127,
"cast/masked_negative_base_rate": 0.7656119633466005,
"cast/masked_positive_base_rate": 0.23438803129829466,
"cast/masked_sign_flipped_rate": 0.24116594763472676,
"cast/masked_teacher_negative_rate": 0.29151507979258895,
"cast/masked_teacher_positive_rate": 0.23739769076928496,
"cast/masked_token_adv_abs_mean": 0.8658460238948464,
"cast/masked_token_adv_max": 1.0583419147878885,
"cast/masked_token_adv_mean": -0.28547400305978954,
"cast/masked_token_adv_min": -1.1827479861676693,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.027637973515084013,
"cast/masked_wrong_good_token_rate": 0.17980339680798352,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0016161244238901418,
"clip_ratio/high_mean": 0.0016161244238901418,
"clip_ratio/low_mean": 0.0005605285346064193,
"clip_ratio/low_min": 0.0005605285346064193,
"clip_ratio/region_mean": 0.0021766529644082766,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4763.0,
"completions/max_terminated_length": 4763.0,
"completions/mean_length": 2043.296875,
"completions/mean_terminated_length": 2043.296875,
"completions/min_length": 240.0,
"completions/min_terminated_length": 240.0,
"entropy": 0.21061308030039072,
"epoch": 0.0013610733035193468,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.07192060351371765,
"learning_rate": 9.996111219132802e-07,
"loss": 0.18207664787769318,
"num_tokens": 976181.0,
"reward": 0.296875,
"reward_std": 0.4604927599430084,
"rewards/math_answer_reward/mean": 0.296875,
"rewards/math_answer_reward/std": 0.4604927599430084,
"sampling/importance_sampling_ratio/max": 2.541567802429199,
"sampling/importance_sampling_ratio/mean": 0.6366204619407654,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9121077060699463,
"sampling/sampling_logp_difference/mean": 0.007895395159721375,
"step": 7,
"step_time": 240.3514125389338
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.09073963621631265,
"cast/masked_gap_abs_mean": 0.0249717844708357,
"cast/masked_gap_mean": -0.005609225301213883,
"cast/masked_negative_base_rate": 0.676128753926605,
"cast/masked_positive_base_rate": 0.323871242813766,
"cast/masked_sign_flipped_rate": 0.2843371140770614,
"cast/masked_teacher_negative_rate": 0.2969150189310312,
"cast/masked_teacher_positive_rate": 0.2735536084510386,
"cast/masked_token_adv_abs_mean": 0.8107104729861021,
"cast/masked_token_adv_max": 1.0228791739791632,
"cast/masked_token_adv_mean": -0.1128917564637959,
"cast/masked_token_adv_min": -1.1154333986341953,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03480021370342001,
"cast/masked_wrong_good_token_rate": 0.19359747716225684,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0020816404685319867,
"clip_ratio/high_mean": 0.0020816404685319867,
"clip_ratio/low_mean": 0.0007025639497442171,
"clip_ratio/low_min": 0.0007025639497442171,
"clip_ratio/region_mean": 0.002784204396448331,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5195.0,
"completions/max_terminated_length": 5195.0,
"completions/mean_length": 1421.4375,
"completions/mean_terminated_length": 1421.4375,
"completions/min_length": 679.0,
"completions/min_terminated_length": 679.0,
"entropy": 0.2042804155498743,
"epoch": 0.0015555123468792534,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.08657658100128174,
"learning_rate": 9.995463088988269e-07,
"loss": 0.05949269235134125,
"num_tokens": 1077113.0,
"reward": 0.34375,
"reward_std": 0.4787135720252991,
"rewards/math_answer_reward/mean": 0.34375,
"rewards/math_answer_reward/std": 0.4787135720252991,
"sampling/importance_sampling_ratio/max": 2.641132116317749,
"sampling/importance_sampling_ratio/mean": 0.5868303179740906,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7449855804443359,
"sampling/sampling_logp_difference/mean": 0.009150206111371517,
"step": 8,
"step_time": 251.9736781420943
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.07257443683920428,
"cast/masked_gap_abs_mean": 0.022104691874119453,
"cast/masked_gap_mean": -0.003989100499438791,
"cast/masked_negative_base_rate": 0.765513863414526,
"cast/masked_positive_base_rate": 0.2344861375167966,
"cast/masked_sign_flipped_rate": 0.29516373481601477,
"cast/masked_teacher_negative_rate": 0.34323391877114773,
"cast/masked_teacher_positive_rate": 0.2994946357794106,
"cast/masked_token_adv_abs_mean": 0.845132689923048,
"cast/masked_token_adv_max": 1.0465021124109626,
"cast/masked_token_adv_mean": -0.25338738450955134,
"cast/masked_token_adv_min": -1.1585024883970618,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03220963169587776,
"cast/masked_wrong_good_token_rate": 0.2225892967544496,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0019276899129181402,
"clip_ratio/high_mean": 0.0019276899129181402,
"clip_ratio/low_mean": 0.0006382734263752354,
"clip_ratio/low_min": 0.0006382734263752354,
"clip_ratio/region_mean": 0.002565963353845291,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4392.0,
"completions/max_terminated_length": 4392.0,
"completions/mean_length": 1962.90625,
"completions/mean_terminated_length": 1962.90625,
"completions/min_length": 611.0,
"completions/min_terminated_length": 611.0,
"entropy": 0.2864137524738908,
"epoch": 0.00174995139023916,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.0570983923971653,
"learning_rate": 9.994814958843736e-07,
"loss": 0.16365830600261688,
"num_tokens": 1215147.0,
"reward": 0.296875,
"reward_std": 0.4604927599430084,
"rewards/math_answer_reward/mean": 0.296875,
"rewards/math_answer_reward/std": 0.4604927599430084,
"sampling/importance_sampling_ratio/max": 2.1754255294799805,
"sampling/importance_sampling_ratio/mean": 0.4609132707118988,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.402822732925415,
"sampling/sampling_logp_difference/mean": 0.010235561057925224,
"step": 9,
"step_time": 236.75115443207324
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.0,
"cast/masked_correct_bad_token_rate": 0.10064402007265016,
"cast/masked_gap_abs_mean": 0.030088994244579226,
"cast/masked_gap_mean": -0.006821240114732063,
"cast/masked_negative_base_rate": 0.6098946072161198,
"cast/masked_positive_base_rate": 0.3901053888257593,
"cast/masked_sign_flipped_rate": 0.2840200802311301,
"cast/masked_teacher_negative_rate": 0.26796936709433794,
"cast/masked_teacher_positive_rate": 0.2970573240891099,
"cast/masked_token_adv_abs_mean": 0.6948650535196066,
"cast/masked_token_adv_max": 0.9002567371353507,
"cast/masked_token_adv_mean": -0.030735749867744744,
"cast/masked_token_adv_min": -0.9645791640505195,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.038468276732601225,
"cast/masked_wrong_good_token_rate": 0.18337606079876423,
"cast/mixed_group_rate": 1.0,
"clip_ratio/high_max": 0.0014356390643115446,
"clip_ratio/high_mean": 0.0014356390643115446,
"clip_ratio/low_mean": 0.0007832060146029107,
"clip_ratio/low_min": 0.0007832060146029107,
"clip_ratio/region_mean": 0.002218845096649602,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5147.0,
"completions/max_terminated_length": 5147.0,
"completions/mean_length": 1496.9375,
"completions/mean_terminated_length": 1496.9375,
"completions/min_length": 430.0,
"completions/min_terminated_length": 430.0,
"entropy": 0.23761650919914246,
"epoch": 0.0019443904335990666,
"frac_reward_zero_std": 0.0,
"grad_norm": 0.09028841555118561,
"learning_rate": 9.994166828699203e-07,
"loss": -0.0251512061804533,
"num_tokens": 1321015.0,
"reward": 0.484375,
"reward_std": 0.5037065148353577,
"rewards/math_answer_reward/mean": 0.484375,
"rewards/math_answer_reward/std": 0.5037065148353577,
"sampling/importance_sampling_ratio/max": 2.591317653656006,
"sampling/importance_sampling_ratio/mean": 0.565003514289856,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.2380824089050293,
"sampling/sampling_logp_difference/mean": 0.009557080455124378,
"step": 10,
"step_time": 315.605143566936
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.01738631201442331,
"cast/masked_gap_abs_mean": 0.019715279617230408,
"cast/masked_gap_mean": -0.004305404807610103,
"cast/masked_negative_base_rate": 0.9370690686628222,
"cast/masked_positive_base_rate": 0.0629309262149036,
"cast/masked_sign_flipped_rate": 0.27036237018182874,
"cast/masked_teacher_negative_rate": 0.32638421934098005,
"cast/masked_teacher_positive_rate": 0.2675942163914442,
"cast/masked_token_adv_abs_mean": 0.8154059993103147,
"cast/masked_token_adv_max": 0.9751813746988773,
"cast/masked_token_adv_mean": -0.3152205008082092,
"cast/masked_token_adv_min": -1.0823502130806446,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.030318328936118633,
"cast/masked_wrong_good_token_rate": 0.25297605828382075,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0016964454916887917,
"clip_ratio/high_mean": 0.0016964454916887917,
"clip_ratio/low_mean": 0.000854590963172086,
"clip_ratio/low_min": 0.000854590963172086,
"clip_ratio/region_mean": 0.0025510364393994678,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 5838.0,
"completions/mean_length": 2036.4375,
"completions/mean_terminated_length": 1938.730224609375,
"completions/min_length": 580.0,
"completions/min_terminated_length": 580.0,
"entropy": 0.24897198739927262,
"epoch": 0.0021388294769589733,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.06734070926904678,
"learning_rate": 9.99351869855467e-07,
"loss": 0.1553853303194046,
"num_tokens": 1463515.0,
"reward": 0.09375,
"reward_std": 0.29378482699394226,
"rewards/math_answer_reward/mean": 0.09375,
"rewards/math_answer_reward/std": 0.29378482699394226,
"sampling/importance_sampling_ratio/max": 2.8687195777893066,
"sampling/importance_sampling_ratio/mean": 0.5816532373428345,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.6049604415893555,
"sampling/sampling_logp_difference/mean": 0.008935092948377132,
"step": 11,
"step_time": 356.0302096590749
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.08462497038999572,
"cast/masked_gap_abs_mean": 0.021984716557199135,
"cast/masked_gap_mean": -0.005465808458211541,
"cast/masked_negative_base_rate": 0.7294434225186706,
"cast/masked_positive_base_rate": 0.2705565735232085,
"cast/masked_sign_flipped_rate": 0.24656622298061848,
"cast/masked_teacher_negative_rate": 0.29080205829814076,
"cast/masked_teacher_positive_rate": 0.2171602752059698,
"cast/masked_token_adv_abs_mean": 0.7858105516061187,
"cast/masked_token_adv_max": 1.0076016513630748,
"cast/masked_token_adv_mean": -0.17887636297382414,
"cast/masked_token_adv_min": -1.1030237218365073,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.032445462711621076,
"cast/masked_wrong_good_token_rate": 0.16194125183392316,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0015876106026553316,
"clip_ratio/high_mean": 0.0015876106026553316,
"clip_ratio/low_mean": 0.000719349601695285,
"clip_ratio/low_min": 0.000719349601695285,
"clip_ratio/region_mean": 0.0023069602066243533,
"completions/clipped_ratio": 0.0,
"completions/max_length": 6460.0,
"completions/max_terminated_length": 6460.0,
"completions/mean_length": 2193.5,
"completions/mean_terminated_length": 2193.5,
"completions/min_length": 336.0,
"completions/min_terminated_length": 336.0,
"entropy": 0.2045616654213518,
"epoch": 0.00233326852031888,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.09138477593660355,
"learning_rate": 9.992870568410137e-07,
"loss": 0.004339075181633234,
"num_tokens": 1615115.0,
"reward": 0.359375,
"reward_std": 0.4836103618144989,
"rewards/math_answer_reward/mean": 0.359375,
"rewards/math_answer_reward/std": 0.4836103618144989,
"sampling/importance_sampling_ratio/max": 2.7900450229644775,
"sampling/importance_sampling_ratio/mean": 0.6546765565872192,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9900543689727783,
"sampling/sampling_logp_difference/mean": 0.007952078245580196,
"step": 12,
"step_time": 337.5490791980119
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.04940713930409402,
"cast/masked_gap_abs_mean": 0.021393386312411167,
"cast/masked_gap_mean": -0.004285261602433366,
"cast/masked_negative_base_rate": 0.8071926310658455,
"cast/masked_positive_base_rate": 0.19280736334621906,
"cast/masked_sign_flipped_rate": 0.2803256041370332,
"cast/masked_teacher_negative_rate": 0.33702709805220366,
"cast/masked_teacher_positive_rate": 0.2868506470695138,
"cast/masked_token_adv_abs_mean": 0.9617745075374842,
"cast/masked_token_adv_max": 1.0608110204339027,
"cast/masked_token_adv_mean": -0.24292062067252118,
"cast/masked_token_adv_min": -1.164141234010458,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03146881266729906,
"cast/masked_wrong_good_token_rate": 0.23091846448369324,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0019441673530309345,
"clip_ratio/high_mean": 0.0019441673530309345,
"clip_ratio/low_mean": 0.0008632063163531711,
"clip_ratio/low_min": 0.0008632063163531711,
"clip_ratio/region_mean": 0.002807373683026526,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4738.0,
"completions/max_terminated_length": 4738.0,
"completions/mean_length": 1691.59375,
"completions/mean_terminated_length": 1691.59375,
"completions/min_length": 370.0,
"completions/min_terminated_length": 370.0,
"entropy": 0.26271722093224525,
"epoch": 0.0025277075636787865,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.08375957608222961,
"learning_rate": 9.992222438265604e-07,
"loss": 0.15908388793468475,
"num_tokens": 1735761.0,
"reward": 0.234375,
"reward_std": 0.42695629596710205,
"rewards/math_answer_reward/mean": 0.234375,
"rewards/math_answer_reward/std": 0.42695629596710205,
"sampling/importance_sampling_ratio/max": 2.823263645172119,
"sampling/importance_sampling_ratio/mean": 0.5573142766952515,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.12907075881958,
"sampling/sampling_logp_difference/mean": 0.01045958697795868,
"step": 13,
"step_time": 226.32362587796524
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.0671185451792553,
"cast/masked_gap_abs_mean": 0.02639512301539071,
"cast/masked_gap_mean": -0.004975741314410698,
"cast/masked_negative_base_rate": 0.7318982388824224,
"cast/masked_positive_base_rate": 0.2681017564609647,
"cast/masked_sign_flipped_rate": 0.25213746074587107,
"cast/masked_teacher_negative_rate": 0.3173387721180916,
"cast/masked_teacher_positive_rate": 0.2679727687500417,
"cast/masked_token_adv_abs_mean": 0.8000373672693968,
"cast/masked_token_adv_max": 0.9614516906440258,
"cast/masked_token_adv_mean": -0.18809876988234464,
"cast/masked_token_adv_min": -1.072016291320324,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03405441885115579,
"cast/masked_wrong_good_token_rate": 0.18501891591586173,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.002262323286231549,
"clip_ratio/high_mean": 0.002262323286231549,
"clip_ratio/low_mean": 0.0007476541386495228,
"clip_ratio/low_min": 0.0007476541386495228,
"clip_ratio/region_mean": 0.0030099774467089446,
"completions/clipped_ratio": 0.0,
"completions/max_length": 7031.0,
"completions/max_terminated_length": 7031.0,
"completions/mean_length": 1395.640625,
"completions/mean_terminated_length": 1395.640625,
"completions/min_length": 440.0,
"completions/min_terminated_length": 440.0,
"entropy": 0.2426241203211248,
"epoch": 0.0027221466070386936,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.08422277122735977,
"learning_rate": 9.99157430812107e-07,
"loss": 0.10312047600746155,
"num_tokens": 1835090.0,
"reward": 0.34375,
"reward_std": 0.4787135720252991,
"rewards/math_answer_reward/mean": 0.34375,
"rewards/math_answer_reward/std": 0.4787135720252991,
"sampling/importance_sampling_ratio/max": 2.811018705368042,
"sampling/importance_sampling_ratio/mean": 0.6707406640052795,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7499971389770508,
"sampling/sampling_logp_difference/mean": 0.009853086434304714,
"step": 14,
"step_time": 299.11479242297355
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.0411057414021343,
"cast/masked_gap_abs_mean": 0.016717783495550975,
"cast/masked_gap_mean": -0.0018182480530413159,
"cast/masked_negative_base_rate": 0.8869069404900074,
"cast/masked_positive_base_rate": 0.11309304926544428,
"cast/masked_sign_flipped_rate": 0.29622267140075564,
"cast/masked_teacher_negative_rate": 0.3703766791149974,
"cast/masked_teacher_positive_rate": 0.28294349135831,
"cast/masked_token_adv_abs_mean": 0.9081416688859463,
"cast/masked_token_adv_max": 1.0237302472814918,
"cast/masked_token_adv_mean": -0.2713818247138988,
"cast/masked_token_adv_min": -1.1378346448764205,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02932036726269871,
"cast/masked_wrong_good_token_rate": 0.2551169303478673,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.001964338956895517,
"clip_ratio/high_mean": 0.001964338956895517,
"clip_ratio/low_mean": 0.0010387905977040646,
"clip_ratio/low_min": 0.0010387905977040646,
"clip_ratio/region_mean": 0.0030031295609660447,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4689.0,
"completions/max_terminated_length": 4689.0,
"completions/mean_length": 2008.359375,
"completions/mean_terminated_length": 2008.359375,
"completions/min_length": 524.0,
"completions/min_terminated_length": 524.0,
"entropy": 0.3075254997238517,
"epoch": 0.0029165856503986,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.0826810821890831,
"learning_rate": 9.990926177976538e-07,
"loss": 0.11570171266794205,
"num_tokens": 1977249.0,
"reward": 0.09375,
"reward_std": 0.29378482699394226,
"rewards/math_answer_reward/mean": 0.09375,
"rewards/math_answer_reward/std": 0.29378482699394226,
"sampling/importance_sampling_ratio/max": 2.7698135375976562,
"sampling/importance_sampling_ratio/mean": 0.5340856313705444,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7377318143844604,
"sampling/sampling_logp_difference/mean": 0.011122718453407288,
"step": 15,
"step_time": 212.1223155549378
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.05252719100099057,
"cast/masked_gap_abs_mean": 0.020715841790661216,
"cast/masked_gap_mean": -0.004802196254331648,
"cast/masked_negative_base_rate": 0.7396341543644667,
"cast/masked_positive_base_rate": 0.26036584423854947,
"cast/masked_sign_flipped_rate": 0.24193910788744688,
"cast/masked_teacher_negative_rate": 0.25685327034443617,
"cast/masked_teacher_positive_rate": 0.2513215499930084,
"cast/masked_token_adv_abs_mean": 0.7762469192966819,
"cast/masked_token_adv_max": 0.9771279226988554,
"cast/masked_token_adv_mean": -0.12411686254199594,
"cast/masked_token_adv_min": -1.0738605428487062,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03151404531672597,
"cast/masked_wrong_good_token_rate": 0.18941191560588777,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0014170731619742583,
"clip_ratio/high_mean": 0.0014170731619742583,
"clip_ratio/low_mean": 0.0007574393825962034,
"clip_ratio/low_min": 0.0007574393825962034,
"clip_ratio/region_mean": 0.002174512557758135,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5131.0,
"completions/max_terminated_length": 5131.0,
"completions/mean_length": 1925.8125,
"completions/mean_terminated_length": 1925.8125,
"completions/min_length": 448.0,
"completions/min_terminated_length": 448.0,
"entropy": 0.215898257214576,
"epoch": 0.003111024693758507,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.07403209060430527,
"learning_rate": 9.990278047832003e-07,
"loss": 0.07934418320655823,
"num_tokens": 2112549.0,
"reward": 0.28125,
"reward_std": 0.4531635046005249,
"rewards/math_answer_reward/mean": 0.28125,
"rewards/math_answer_reward/std": 0.4531635046005249,
"sampling/importance_sampling_ratio/max": 2.877987861633301,
"sampling/importance_sampling_ratio/mean": 0.6488768458366394,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.02785325050354,
"sampling/sampling_logp_difference/mean": 0.008358784951269627,
"step": 16,
"step_time": 281.0986711210571
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.07000820012763143,
"cast/masked_gap_abs_mean": 0.021835155959706753,
"cast/masked_gap_mean": -0.0038903999229660258,
"cast/masked_negative_base_rate": 0.7783364430069923,
"cast/masked_positive_base_rate": 0.2216635448858142,
"cast/masked_sign_flipped_rate": 0.303117117844522,
"cast/masked_teacher_negative_rate": 0.3225830919109285,
"cast/masked_teacher_positive_rate": 0.28767480980604887,
"cast/masked_token_adv_abs_mean": 0.8015914661809802,
"cast/masked_token_adv_max": 0.9985594702884555,
"cast/masked_token_adv_mean": -0.15460392436943948,
"cast/masked_token_adv_min": -1.0929966000840068,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03441973705776036,
"cast/masked_wrong_good_token_rate": 0.23310891771689057,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0017479256121077924,
"clip_ratio/high_mean": 0.0017479256121077924,
"clip_ratio/low_mean": 0.0007376973999271286,
"clip_ratio/low_min": 0.0007376973999271286,
"clip_ratio/region_mean": 0.002485623024767847,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 4684.0,
"completions/mean_length": 1549.046875,
"completions/mean_terminated_length": 1443.603271484375,
"completions/min_length": 538.0,
"completions/min_terminated_length": 538.0,
"entropy": 0.2567666624672711,
"epoch": 0.0033054637371184134,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.081749327480793,
"learning_rate": 9.989629917687472e-07,
"loss": 0.020313730463385582,
"num_tokens": 2222456.0,
"reward": 0.21875,
"reward_std": 0.4166666865348816,
"rewards/math_answer_reward/mean": 0.21875,
"rewards/math_answer_reward/std": 0.4166666865348816,
"sampling/importance_sampling_ratio/max": 2.8027782440185547,
"sampling/importance_sampling_ratio/mean": 0.6228022575378418,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.752948522567749,
"sampling/sampling_logp_difference/mean": 0.010455365292727947,
"step": 17,
"step_time": 411.2714750049345
},
{
"cast/all_correct_group_rate": 0.25,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.09096676635090262,
"cast/masked_gap_abs_mean": 0.023253026593010873,
"cast/masked_gap_mean": -0.005856357790293032,
"cast/masked_negative_base_rate": 0.6390319131314754,
"cast/masked_positive_base_rate": 0.36096808570437133,
"cast/masked_sign_flipped_rate": 0.2751821503043175,
"cast/masked_teacher_negative_rate": 0.30246221646666527,
"cast/masked_teacher_positive_rate": 0.27615788765251637,
"cast/masked_token_adv_abs_mean": 0.8868191856890917,
"cast/masked_token_adv_max": 1.034641768783331,
"cast/masked_token_adv_mean": -0.018933910801933962,
"cast/masked_token_adv_min": -1.1503049544990063,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.033482928411103785,
"cast/masked_wrong_good_token_rate": 0.1842153833131306,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0019130871919514902,
"clip_ratio/high_mean": 0.0019130871919514902,
"clip_ratio/low_mean": 0.0007378526297543431,
"clip_ratio/low_min": 0.0007378526297543431,
"clip_ratio/region_mean": 0.002650939827617549,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4657.0,
"completions/max_terminated_length": 4657.0,
"completions/mean_length": 1515.21875,
"completions/mean_terminated_length": 1515.21875,
"completions/min_length": 444.0,
"completions/min_terminated_length": 444.0,
"entropy": 0.24267979757860303,
"epoch": 0.00349990278047832,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.10959213972091675,
"learning_rate": 9.988981787542939e-07,
"loss": 0.018584363162517548,
"num_tokens": 2331982.0,
"reward": 0.40625,
"reward_std": 0.49501484632492065,
"rewards/math_answer_reward/mean": 0.40625,
"rewards/math_answer_reward/std": 0.49501484632492065,
"sampling/importance_sampling_ratio/max": 2.7390105724334717,
"sampling/importance_sampling_ratio/mean": 0.7537024021148682,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.2654191255569458,
"sampling/sampling_logp_difference/mean": 0.010372188873589039,
"step": 18,
"step_time": 270.87571251610643
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.1177772528026253,
"cast/masked_gap_abs_mean": 0.026565664797089994,
"cast/masked_gap_mean": -0.005672709508871776,
"cast/masked_negative_base_rate": 0.6128476276062429,
"cast/masked_positive_base_rate": 0.3871523691341281,
"cast/masked_sign_flipped_rate": 0.2854098607786,
"cast/masked_teacher_negative_rate": 0.33823256846517324,
"cast/masked_teacher_positive_rate": 0.277495788410306,
"cast/masked_token_adv_abs_mean": 0.8691526968032122,
"cast/masked_token_adv_max": 1.008711902424693,
"cast/masked_token_adv_mean": -0.1461679121130146,
"cast/masked_token_adv_min": -1.1099565252661705,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.0360996212111786,
"cast/masked_wrong_good_token_rate": 0.1676326049491763,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.002375374435359845,
"clip_ratio/high_mean": 0.002375374435359845,
"clip_ratio/low_mean": 0.0009999150679504965,
"clip_ratio/low_min": 0.0009999150679504965,
"clip_ratio/region_mean": 0.0033752895433281083,
"completions/clipped_ratio": 0.0,
"completions/max_length": 6059.0,
"completions/max_terminated_length": 6059.0,
"completions/mean_length": 1183.015625,
"completions/mean_terminated_length": 1183.015625,
"completions/min_length": 420.0,
"completions/min_terminated_length": 420.0,
"entropy": 0.24240582156926394,
"epoch": 0.0036943418238382267,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.08647438138723373,
"learning_rate": 9.988333657398406e-07,
"loss": 0.08147846907377243,
"num_tokens": 2418743.0,
"reward": 0.421875,
"reward_std": 0.49776285886764526,
"rewards/math_answer_reward/mean": 0.421875,
"rewards/math_answer_reward/std": 0.49776285886764526,
"sampling/importance_sampling_ratio/max": 2.397792339324951,
"sampling/importance_sampling_ratio/mean": 0.5972786545753479,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.3280539512634277,
"sampling/sampling_logp_difference/mean": 0.01074514165520668,
"step": 19,
"step_time": 323.4415142440412
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.02626719488762319,
"cast/masked_gap_abs_mean": 0.0156603154464392,
"cast/masked_gap_mean": -0.0020564324586871408,
"cast/masked_negative_base_rate": 0.88489880412817,
"cast/masked_positive_base_rate": 0.11510118655860424,
"cast/masked_sign_flipped_rate": 0.24622767791152,
"cast/masked_teacher_negative_rate": 0.28339876839891076,
"cast/masked_teacher_positive_rate": 0.25436386885121465,
"cast/masked_token_adv_abs_mean": 0.8630759622901678,
"cast/masked_token_adv_max": 1.0217759367078543,
"cast/masked_token_adv_mean": -0.3033593747531995,
"cast/masked_token_adv_min": -1.1409582905471325,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.028263551415875554,
"cast/masked_wrong_good_token_rate": 0.21996048255823553,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0017522616899441346,
"clip_ratio/high_mean": 0.0017522616899441346,
"clip_ratio/low_mean": 0.0006478665227405145,
"clip_ratio/low_min": 0.0006478665227405145,
"clip_ratio/region_mean": 0.0024001282017707126,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5592.0,
"completions/max_terminated_length": 5592.0,
"completions/mean_length": 2073.859375,
"completions/mean_terminated_length": 2073.859375,
"completions/min_length": 667.0,
"completions/min_terminated_length": 667.0,
"entropy": 0.21598226088099182,
"epoch": 0.0038887808671981333,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.05494651943445206,
"learning_rate": 9.98768552725387e-07,
"loss": 0.21258239448070526,
"num_tokens": 2563142.0,
"reward": 0.140625,
"reward_std": 0.3503824472427368,
"rewards/math_answer_reward/mean": 0.140625,
"rewards/math_answer_reward/std": 0.3503824472427368,
"sampling/importance_sampling_ratio/max": 2.7143900394439697,
"sampling/importance_sampling_ratio/mean": 0.6295812726020813,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.5600368976593018,
"sampling/sampling_logp_difference/mean": 0.008333928883075714,
"step": 20,
"step_time": 231.06047810605378
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.07069681346183643,
"cast/masked_gap_abs_mean": 0.026408043602714315,
"cast/masked_gap_mean": -0.005094292716421478,
"cast/masked_negative_base_rate": 0.7460893653333187,
"cast/masked_positive_base_rate": 0.25391063420102,
"cast/masked_sign_flipped_rate": 0.2630305583588779,
"cast/masked_teacher_negative_rate": 0.29143370455130935,
"cast/masked_teacher_positive_rate": 0.25280020106583834,
"cast/masked_token_adv_abs_mean": 0.7515196232125163,
"cast/masked_token_adv_max": 0.9924054453149438,
"cast/masked_token_adv_mean": -0.10596145631279796,
"cast/masked_token_adv_min": -1.0859634196385741,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03691637818701565,
"cast/masked_wrong_good_token_rate": 0.1923337436746806,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0018965824865517789,
"clip_ratio/high_mean": 0.0018965824865517789,
"clip_ratio/low_mean": 0.0006435436375795689,
"clip_ratio/low_min": 0.0006435436375795689,
"clip_ratio/region_mean": 0.0025401261027582223,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 5285.0,
"completions/mean_length": 1870.6875,
"completions/mean_terminated_length": 1770.349365234375,
"completions/min_length": 597.0,
"completions/min_terminated_length": 597.0,
"entropy": 0.2545768362469971,
"epoch": 0.00408321991055804,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.08671203255653381,
"learning_rate": 9.98703739710934e-07,
"loss": 0.06778078526258469,
"num_tokens": 2694538.0,
"reward": 0.265625,
"reward_std": 0.44515693187713623,
"rewards/math_answer_reward/mean": 0.265625,
"rewards/math_answer_reward/std": 0.44515693187713623,
"sampling/importance_sampling_ratio/max": 2.8222897052764893,
"sampling/importance_sampling_ratio/mean": 0.6074333190917969,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.11054527759552,
"sampling/sampling_logp_difference/mean": 0.009664753451943398,
"step": 21,
"step_time": 408.66364776797127
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.0815659822546877,
"cast/masked_gap_abs_mean": 0.02575306291691959,
"cast/masked_gap_mean": -0.00659379334138066,
"cast/masked_negative_base_rate": 0.6078154062852263,
"cast/masked_positive_base_rate": 0.39218459837138653,
"cast/masked_sign_flipped_rate": 0.23282412276603281,
"cast/masked_teacher_negative_rate": 0.257767794188112,
"cast/masked_teacher_positive_rate": 0.2300756520126015,
"cast/masked_token_adv_abs_mean": 0.8479525167495012,
"cast/masked_token_adv_max": 1.0434970706701279,
"cast/masked_token_adv_mean": -0.03890826083807042,
"cast/masked_token_adv_min": -1.1282824203372002,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03565307770622894,
"cast/masked_wrong_good_token_rate": 0.1512581406859681,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0015993175029507256,
"clip_ratio/high_mean": 0.0015993175029507256,
"clip_ratio/low_mean": 0.0007816949919288163,
"clip_ratio/low_min": 0.0007816949919288163,
"clip_ratio/region_mean": 0.002381012476689648,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4739.0,
"completions/max_terminated_length": 4739.0,
"completions/mean_length": 1344.34375,
"completions/mean_terminated_length": 1344.34375,
"completions/min_length": 613.0,
"completions/min_terminated_length": 613.0,
"entropy": 0.2049221311463043,
"epoch": 0.0042776589539179465,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.09767619520425797,
"learning_rate": 9.986389266964807e-07,
"loss": -0.019729886204004288,
"num_tokens": 2791432.0,
"reward": 0.4375,
"reward_std": 0.5,
"rewards/math_answer_reward/mean": 0.4375,
"rewards/math_answer_reward/std": 0.5,
"sampling/importance_sampling_ratio/max": 2.6556529998779297,
"sampling/importance_sampling_ratio/mean": 0.6903061866760254,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.4230399131774902,
"sampling/sampling_logp_difference/mean": 0.009304777719080448,
"step": 22,
"step_time": 269.99934676199337
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.06562431808561087,
"cast/masked_gap_abs_mean": 0.024118201254168525,
"cast/masked_gap_mean": -0.004947140088916058,
"cast/masked_negative_base_rate": 0.775532076600939,
"cast/masked_positive_base_rate": 0.22446792153641582,
"cast/masked_sign_flipped_rate": 0.31993064051494,
"cast/masked_teacher_negative_rate": 0.36050921864807606,
"cast/masked_teacher_positive_rate": 0.3109617051668465,
"cast/masked_token_adv_abs_mean": 0.9750939700752497,
"cast/masked_token_adv_max": 1.071750096976757,
"cast/masked_token_adv_mean": -0.17395510140340775,
"cast/masked_token_adv_min": -1.192714475095272,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03374552814057097,
"cast/masked_wrong_good_token_rate": 0.254306323826313,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0028481399476731895,
"clip_ratio/high_mean": 0.0028481399476731895,
"clip_ratio/low_mean": 0.0012398080771163222,
"clip_ratio/low_min": 0.0012398080771163222,
"clip_ratio/region_mean": 0.0040879480438889,
"completions/clipped_ratio": 0.0,
"completions/max_length": 3321.0,
"completions/max_terminated_length": 3321.0,
"completions/mean_length": 1105.984375,
"completions/mean_terminated_length": 1105.984375,
"completions/min_length": 358.0,
"completions/min_terminated_length": 358.0,
"entropy": 0.309428920969367,
"epoch": 0.004472097997277854,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.11801837384700775,
"learning_rate": 9.985741136820272e-07,
"loss": 0.041101954877376556,
"num_tokens": 2874783.0,
"reward": 0.25,
"reward_std": 0.4364357888698578,
"rewards/math_answer_reward/mean": 0.25,
"rewards/math_answer_reward/std": 0.4364357888698578,
"sampling/importance_sampling_ratio/max": 2.4954657554626465,
"sampling/importance_sampling_ratio/mean": 0.5566527843475342,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7273344993591309,
"sampling/sampling_logp_difference/mean": 0.01289060153067112,
"step": 23,
"step_time": 136.58293971401872
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.09531215624883771,
"cast/masked_gap_abs_mean": 0.024934635701356456,
"cast/masked_gap_mean": -0.00479325120249996,
"cast/masked_negative_base_rate": 0.6535243364050984,
"cast/masked_positive_base_rate": 0.3464756584726274,
"cast/masked_sign_flipped_rate": 0.2865803833119571,
"cast/masked_teacher_negative_rate": 0.32255886495113373,
"cast/masked_teacher_positive_rate": 0.264562975615263,
"cast/masked_token_adv_abs_mean": 0.8261113557964563,
"cast/masked_token_adv_max": 0.9499526340514421,
"cast/masked_token_adv_mean": -0.12374009017366916,
"cast/masked_token_adv_min": -1.0374459214508533,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.0345170886721462,
"cast/masked_wrong_good_token_rate": 0.19126822683028877,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0022706141216985998,
"clip_ratio/high_mean": 0.0022706141216985998,
"clip_ratio/low_mean": 0.0007457621140929405,
"clip_ratio/low_min": 0.0007457621140929405,
"clip_ratio/region_mean": 0.0030163762567099184,
"completions/clipped_ratio": 0.0,
"completions/max_length": 3038.0,
"completions/max_terminated_length": 3038.0,
"completions/mean_length": 1232.390625,
"completions/mean_terminated_length": 1232.390625,
"completions/min_length": 576.0,
"completions/min_terminated_length": 576.0,
"entropy": 0.23868330055847764,
"epoch": 0.00466653704063776,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.12313699722290039,
"learning_rate": 9.985093006675739e-07,
"loss": 0.0869496688246727,
"num_tokens": 2967016.0,
"reward": 0.40625,
"reward_std": 0.49501484632492065,
"rewards/math_answer_reward/mean": 0.40625,
"rewards/math_answer_reward/std": 0.49501484632492065,
"sampling/importance_sampling_ratio/max": 2.9839441776275635,
"sampling/importance_sampling_ratio/mean": 0.7629752159118652,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 3.0455398559570312,
"sampling/sampling_logp_difference/mean": 0.010935449972748756,
"step": 24,
"step_time": 166.99947806703858
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.03534039144869894,
"cast/masked_gap_abs_mean": 0.01609983692469541,
"cast/masked_gap_mean": -0.002354583604983418,
"cast/masked_negative_base_rate": 0.8617292605340481,
"cast/masked_positive_base_rate": 0.13827073434367776,
"cast/masked_sign_flipped_rate": 0.2682808027602732,
"cast/masked_teacher_negative_rate": 0.3175011924467981,
"cast/masked_teacher_positive_rate": 0.2638805559836328,
"cast/masked_token_adv_abs_mean": 0.8526621945202351,
"cast/masked_token_adv_max": 1.012695426121354,
"cast/masked_token_adv_mean": -0.2626360194408335,
"cast/masked_token_adv_min": -1.146652003750205,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02848807128611952,
"cast/masked_wrong_good_token_rate": 0.23294041177723557,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0019034193919651443,
"clip_ratio/high_mean": 0.0019034193919651443,
"clip_ratio/low_mean": 0.0008697298253537156,
"clip_ratio/low_min": 0.0008697298253537156,
"clip_ratio/region_mean": 0.002773149226413807,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 6248.0,
"completions/mean_length": 2030.359375,
"completions/mean_terminated_length": 1932.5556640625,
"completions/min_length": 489.0,
"completions/min_terminated_length": 489.0,
"entropy": 0.23363847634755075,
"epoch": 0.004860976083997667,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.04275371879339218,
"learning_rate": 9.984444876531208e-07,
"loss": 0.08098113536834717,
"num_tokens": 3110063.0,
"reward": 0.15625,
"reward_std": 0.36596253514289856,
"rewards/math_answer_reward/mean": 0.15625,
"rewards/math_answer_reward/std": 0.36596253514289856,
"sampling/importance_sampling_ratio/max": 2.7519514560699463,
"sampling/importance_sampling_ratio/mean": 0.46074914932250977,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7246513366699219,
"sampling/sampling_logp_difference/mean": 0.009359056130051613,
"step": 25,
"step_time": 393.44433310601744
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.07398472132626921,
"cast/masked_gap_abs_mean": 0.02273673814488575,
"cast/masked_gap_mean": -0.0034077027030434692,
"cast/masked_negative_base_rate": 0.6875515312422067,
"cast/masked_positive_base_rate": 0.31244845874607563,
"cast/masked_sign_flipped_rate": 0.2620743999723345,
"cast/masked_teacher_negative_rate": 0.2885669849347323,
"cast/masked_teacher_positive_rate": 0.2768934080377221,
"cast/masked_token_adv_abs_mean": 0.6361680449917912,
"cast/masked_token_adv_max": 0.8498547561466694,
"cast/masked_token_adv_mean": -0.023735223410767503,
"cast/masked_token_adv_min": -0.9337626304477453,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03456381516298279,
"cast/masked_wrong_good_token_rate": 0.1880896770162508,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0015294112135961768,
"clip_ratio/high_mean": 0.0015294112135961768,
"clip_ratio/low_mean": 0.0006209741341081099,
"clip_ratio/low_min": 0.0006209741341081099,
"clip_ratio/region_mean": 0.0021503853540707496,
"completions/clipped_ratio": 0.0,
"completions/max_length": 6418.0,
"completions/max_terminated_length": 6418.0,
"completions/mean_length": 2001.15625,
"completions/mean_terminated_length": 2001.15625,
"completions/min_length": 430.0,
"completions/min_terminated_length": 430.0,
"entropy": 0.2665193718858063,
"epoch": 0.005055415127357573,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.059411853551864624,
"learning_rate": 9.983796746386673e-07,
"loss": 0.030272316187620163,
"num_tokens": 3249993.0,
"reward": 0.34375,
"reward_std": 0.4787135720252991,
"rewards/math_answer_reward/mean": 0.34375,
"rewards/math_answer_reward/std": 0.4787135720252991,
"sampling/importance_sampling_ratio/max": 2.456583261489868,
"sampling/importance_sampling_ratio/mean": 0.5125008225440979,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1502207517623901,
"sampling/sampling_logp_difference/mean": 0.009593631140887737,
"step": 26,
"step_time": 330.3176479819522
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.0,
"cast/masked_correct_bad_token_rate": 0.10907223040703684,
"cast/masked_gap_abs_mean": 0.026925828045932576,
"cast/masked_gap_mean": -0.005886789907890488,
"cast/masked_negative_base_rate": 0.5548847913742065,
"cast/masked_positive_base_rate": 0.44511520490050316,
"cast/masked_sign_flipped_rate": 0.2792712734080851,
"cast/masked_teacher_negative_rate": 0.2581384121440351,
"cast/masked_teacher_positive_rate": 0.2887932010926306,
"cast/masked_token_adv_abs_mean": 0.7691338201984763,
"cast/masked_token_adv_max": 0.9419733565300703,
"cast/masked_token_adv_mean": 0.045428414159687236,
"cast/masked_token_adv_min": -1.0068981647491455,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03689752705395222,
"cast/masked_wrong_good_token_rate": 0.17019904241897166,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0013767619320788071,
"clip_ratio/high_mean": 0.0013767619320788071,
"clip_ratio/low_mean": 0.0005736043158321991,
"clip_ratio/low_min": 0.0005736043158321991,
"clip_ratio/region_mean": 0.0019503662460920168,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5296.0,
"completions/max_terminated_length": 5296.0,
"completions/mean_length": 2162.921875,
"completions/mean_terminated_length": 2162.921875,
"completions/min_length": 766.0,
"completions/min_terminated_length": 766.0,
"entropy": 0.26483233738690615,
"epoch": 0.00524985417071748,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.06950706243515015,
"learning_rate": 9.98314861624214e-07,
"loss": -0.043015770614147186,
"num_tokens": 3401060.0,
"reward": 0.484375,
"reward_std": 0.5037065148353577,
"rewards/math_answer_reward/mean": 0.484375,
"rewards/math_answer_reward/std": 0.5037065148353577,
"sampling/importance_sampling_ratio/max": 2.5665640830993652,
"sampling/importance_sampling_ratio/mean": 0.47309350967407227,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.0036513805389404,
"sampling/sampling_logp_difference/mean": 0.010576708242297173,
"step": 27,
"step_time": 381.8457279529248
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.04867532686330378,
"cast/masked_gap_abs_mean": 0.013869739093934186,
"cast/masked_gap_mean": -0.001515975440611328,
"cast/masked_negative_base_rate": 0.7873802538961172,
"cast/masked_positive_base_rate": 0.21261974330991507,
"cast/masked_sign_flipped_rate": 0.2874332475475967,
"cast/masked_teacher_negative_rate": 0.3345515700057149,
"cast/masked_teacher_positive_rate": 0.29111008485779166,
"cast/masked_token_adv_abs_mean": 0.9210309470072389,
"cast/masked_token_adv_max": 1.0317463986575603,
"cast/masked_token_adv_mean": -0.25008471075852867,
"cast/masked_token_adv_min": -1.1630674004554749,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.026352063927333802,
"cast/masked_wrong_good_token_rate": 0.2387579206842929,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.002146960881873383,
"clip_ratio/high_mean": 0.002146960881873383,
"clip_ratio/low_mean": 0.0008487871564284433,
"clip_ratio/low_min": 0.0008487871564284433,
"clip_ratio/region_mean": 0.0029957480292068794,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4474.0,
"completions/max_terminated_length": 4474.0,
"completions/mean_length": 1836.15625,
"completions/mean_terminated_length": 1836.15625,
"completions/min_length": 603.0,
"completions/min_terminated_length": 603.0,
"entropy": 0.26293937349691987,
"epoch": 0.005444293214077387,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.052892591804265976,
"learning_rate": 9.98250048609761e-07,
"loss": 0.0760813057422638,
"num_tokens": 3532702.0,
"reward": 0.265625,
"reward_std": 0.44515693187713623,
"rewards/math_answer_reward/mean": 0.265625,
"rewards/math_answer_reward/std": 0.44515693187713623,
"sampling/importance_sampling_ratio/max": 2.5329079627990723,
"sampling/importance_sampling_ratio/mean": 0.42758452892303467,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.6528494358062744,
"sampling/sampling_logp_difference/mean": 0.010517977178096771,
"step": 28,
"step_time": 235.36455207897234
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.054088532691821456,
"cast/masked_gap_abs_mean": 0.015389430467621423,
"cast/masked_gap_mean": -0.0019956981338964397,
"cast/masked_negative_base_rate": 0.76144780497998,
"cast/masked_positive_base_rate": 0.23855218943208456,
"cast/masked_sign_flipped_rate": 0.2739150314591825,
"cast/masked_teacher_negative_rate": 0.3083249037154019,
"cast/masked_teacher_positive_rate": 0.2891133395023644,
"cast/masked_token_adv_abs_mean": 0.8587933368980885,
"cast/masked_token_adv_max": 0.9961292427033186,
"cast/masked_token_adv_mean": -0.17750786512624472,
"cast/masked_token_adv_min": -1.1170049291104078,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.027538908587303013,
"cast/masked_wrong_good_token_rate": 0.2198264985345304,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.001630727514566388,
"clip_ratio/high_mean": 0.001630727514566388,
"clip_ratio/low_mean": 0.0005790184704892454,
"clip_ratio/low_min": 0.0005790184704892454,
"clip_ratio/region_mean": 0.002209745962318266,
"completions/clipped_ratio": 0.0,
"completions/max_length": 6767.0,
"completions/max_terminated_length": 6767.0,
"completions/mean_length": 2742.59375,
"completions/mean_terminated_length": 2742.59375,
"completions/min_length": 716.0,
"completions/min_terminated_length": 716.0,
"entropy": 0.24118876690045,
"epoch": 0.005638732257437293,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.03594682738184929,
"learning_rate": 9.981852355953076e-07,
"loss": 0.05427195504307747,
"num_tokens": 3721388.0,
"reward": 0.25,
"reward_std": 0.4364357888698578,
"rewards/math_answer_reward/mean": 0.25,
"rewards/math_answer_reward/std": 0.4364357888698578,
"sampling/importance_sampling_ratio/max": 2.1207809448242188,
"sampling/importance_sampling_ratio/mean": 0.39664334058761597,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1394009590148926,
"sampling/sampling_logp_difference/mean": 0.009079433977603912,
"step": 29,
"step_time": 378.98389111016877
},
{
"cast/all_correct_group_rate": 0.25,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.06046384829096496,
"cast/masked_gap_abs_mean": 0.019732925386051647,
"cast/masked_gap_mean": -0.004485454310270143,
"cast/masked_negative_base_rate": 0.7130365455523133,
"cast/masked_positive_base_rate": 0.28696345374919474,
"cast/masked_sign_flipped_rate": 0.2577768908813596,
"cast/masked_teacher_negative_rate": 0.27996830688789487,
"cast/masked_teacher_positive_rate": 0.26843130867928267,
"cast/masked_token_adv_abs_mean": 0.810046523809433,
"cast/masked_token_adv_max": 1.017344051040709,
"cast/masked_token_adv_mean": -0.09888326679356396,
"cast/masked_token_adv_min": -1.1412504268810153,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.031226314487867057,
"cast/masked_wrong_good_token_rate": 0.19731304375454783,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0017577257967786863,
"clip_ratio/high_mean": 0.0017577257967786863,
"clip_ratio/low_mean": 0.0006281403857428813,
"clip_ratio/low_min": 0.0006281403857428813,
"clip_ratio/region_mean": 0.0023858661916165147,
"completions/clipped_ratio": 0.0,
"completions/max_length": 3448.0,
"completions/max_terminated_length": 3448.0,
"completions/mean_length": 1528.171875,
"completions/mean_terminated_length": 1528.171875,
"completions/min_length": 242.0,
"completions/min_terminated_length": 242.0,
"entropy": 0.2174106165766716,
"epoch": 0.0058331713007972,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.06475689262151718,
"learning_rate": 9.98120422580854e-07,
"loss": 0.07153639942407608,
"num_tokens": 3832671.0,
"reward": 0.375,
"reward_std": 0.48795005679130554,
"rewards/math_answer_reward/mean": 0.375,
"rewards/math_answer_reward/std": 0.48795005679130554,
"sampling/importance_sampling_ratio/max": 2.7186005115509033,
"sampling/importance_sampling_ratio/mean": 0.6006669998168945,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8138862252235413,
"sampling/sampling_logp_difference/mean": 0.009872965514659882,
"step": 30,
"step_time": 215.18504892112105
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.0970024477574043,
"cast/masked_gap_abs_mean": 0.021859852480702102,
"cast/masked_gap_mean": -0.004030376592851326,
"cast/masked_negative_base_rate": 0.6390382051467896,
"cast/masked_positive_base_rate": 0.3609617929905653,
"cast/masked_sign_flipped_rate": 0.29817443154752254,
"cast/masked_teacher_negative_rate": 0.2887126631103456,
"cast/masked_teacher_positive_rate": 0.3056431105360389,
"cast/masked_token_adv_abs_mean": 0.799266048707068,
"cast/masked_token_adv_max": 0.962659690529108,
"cast/masked_token_adv_mean": -0.09826143219834194,
"cast/masked_token_adv_min": -1.0519682727754116,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.033709332754369825,
"cast/masked_wrong_good_token_rate": 0.201171982800588,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.001542965107546479,
"clip_ratio/high_mean": 0.001542965107546479,
"clip_ratio/low_mean": 0.0006019352958901436,
"clip_ratio/low_min": 0.0006019352958901436,
"clip_ratio/region_mean": 0.002144900414350559,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 6939.0,
"completions/mean_length": 2425.15625,
"completions/mean_terminated_length": 2333.619140625,
"completions/min_length": 607.0,
"completions/min_terminated_length": 607.0,
"entropy": 0.25060989940539,
"epoch": 0.0060276103441571065,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.06223725900053978,
"learning_rate": 9.980556095664008e-07,
"loss": 0.012041570618748665,
"num_tokens": 4000217.0,
"reward": 0.421875,
"reward_std": 0.49776285886764526,
"rewards/math_answer_reward/mean": 0.421875,
"rewards/math_answer_reward/std": 0.49776285886764526,
"sampling/importance_sampling_ratio/max": 2.654982328414917,
"sampling/importance_sampling_ratio/mean": 0.4486572742462158,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.378830909729004,
"sampling/sampling_logp_difference/mean": 0.010598539374768734,
"step": 31,
"step_time": 425.0822707420157
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.11369886435568333,
"cast/masked_gap_abs_mean": 0.02839366244734265,
"cast/masked_gap_mean": -0.006042019156666356,
"cast/masked_negative_base_rate": 0.551108623854816,
"cast/masked_positive_base_rate": 0.4488913742825389,
"cast/masked_sign_flipped_rate": 0.2760483385063708,
"cast/masked_teacher_negative_rate": 0.29658035887405276,
"cast/masked_teacher_positive_rate": 0.27369685494340956,
"cast/masked_token_adv_abs_mean": 0.8625110182911158,
"cast/masked_token_adv_max": 1.0221386216580868,
"cast/masked_token_adv_mean": -0.004203182179480791,
"cast/masked_token_adv_min": -1.0931584723293781,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03781335899839178,
"cast/masked_wrong_good_token_rate": 0.16234947217162699,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0017260055283259135,
"clip_ratio/high_mean": 0.0017260055283259135,
"clip_ratio/low_mean": 0.0008768961679379572,
"clip_ratio/low_min": 0.0008768961679379572,
"clip_ratio/region_mean": 0.0026029016908069025,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4592.0,
"completions/max_terminated_length": 4592.0,
"completions/mean_length": 1357.40625,
"completions/mean_terminated_length": 1357.40625,
"completions/min_length": 524.0,
"completions/min_terminated_length": 524.0,
"entropy": 0.24189870269037783,
"epoch": 0.006222049387517014,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.10549221932888031,
"learning_rate": 9.979907965519477e-07,
"loss": -0.013073958456516266,
"num_tokens": 4097755.0,
"reward": 0.484375,
"reward_std": 0.5037065148353577,
"rewards/math_answer_reward/mean": 0.484375,
"rewards/math_answer_reward/std": 0.5037065148353577,
"sampling/importance_sampling_ratio/max": 2.7206714153289795,
"sampling/importance_sampling_ratio/mean": 0.6539725065231323,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.6759878993034363,
"sampling/sampling_logp_difference/mean": 0.01012218277901411,
"step": 32,
"step_time": 276.0073487381451
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.03293630562257022,
"cast/masked_gap_abs_mean": 0.017464402539189905,
"cast/masked_gap_mean": -0.0025544972233717544,
"cast/masked_negative_base_rate": 0.795351724140346,
"cast/masked_positive_base_rate": 0.2046482702717185,
"cast/masked_sign_flipped_rate": 0.25216002529487014,
"cast/masked_teacher_negative_rate": 0.25961318239569664,
"cast/masked_teacher_positive_rate": 0.2937126765027642,
"cast/masked_token_adv_abs_mean": 0.7628101045265794,
"cast/masked_token_adv_max": 1.03647277969867,
"cast/masked_token_adv_mean": -0.09147478174418211,
"cast/masked_token_adv_min": -1.1684689773246646,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.0307426939252764,
"cast/masked_wrong_good_token_rate": 0.2192237200215459,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0018214052997791441,
"clip_ratio/high_mean": 0.0018214052997791441,
"clip_ratio/low_mean": 0.0007294006077245285,
"clip_ratio/low_min": 0.0007294006077245285,
"clip_ratio/region_mean": 0.002550805895225494,
"completions/clipped_ratio": 0.0,
"completions/max_length": 3605.0,
"completions/max_terminated_length": 3605.0,
"completions/mean_length": 1701.359375,
"completions/mean_terminated_length": 1701.359375,
"completions/min_length": 710.0,
"completions/min_terminated_length": 710.0,
"entropy": 0.23444979498162866,
"epoch": 0.00641648843087692,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.06196843460202217,
"learning_rate": 9.979259835374942e-07,
"loss": 0.027137164026498795,
"num_tokens": 4217914.0,
"reward": 0.234375,
"reward_std": 0.42695629596710205,
"rewards/math_answer_reward/mean": 0.234375,
"rewards/math_answer_reward/std": 0.42695629596710205,
"sampling/importance_sampling_ratio/max": 2.4452130794525146,
"sampling/importance_sampling_ratio/mean": 0.5065360069274902,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 3.1611125469207764,
"sampling/sampling_logp_difference/mean": 0.009801158681511879,
"step": 33,
"step_time": 196.44994001099258
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.07250226207543164,
"cast/masked_gap_abs_mean": 0.02004892130207736,
"cast/masked_gap_mean": -0.003906553527798451,
"cast/masked_negative_base_rate": 0.676145825535059,
"cast/masked_positive_base_rate": 0.323854167945683,
"cast/masked_sign_flipped_rate": 0.25755096855573356,
"cast/masked_teacher_negative_rate": 0.26940466836094856,
"cast/masked_teacher_positive_rate": 0.26148498663678765,
"cast/masked_token_adv_abs_mean": 0.6970880832523108,
"cast/masked_token_adv_max": 0.848291234113276,
"cast/masked_token_adv_mean": -0.10008010317687877,
"cast/masked_token_adv_min": -0.9266186011955142,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.031243318924680352,
"cast/masked_wrong_good_token_rate": 0.18504870729520917,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.001709435093744105,
"clip_ratio/high_mean": 0.001709435093744105,
"clip_ratio/low_mean": 0.0007297870265574602,
"clip_ratio/low_min": 0.0007297870265574602,
"clip_ratio/region_mean": 0.002439222093016724,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 8106.0,
"completions/mean_length": 2269.5,
"completions/mean_terminated_length": 2175.4921875,
"completions/min_length": 676.0,
"completions/min_terminated_length": 676.0,
"entropy": 0.22634972585365176,
"epoch": 0.006610927474236827,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.04952499642968178,
"learning_rate": 9.97861170523041e-07,
"loss": 0.07481169700622559,
"num_tokens": 4376162.0,
"reward": 0.328125,
"reward_std": 0.4732423722743988,
"rewards/math_answer_reward/mean": 0.328125,
"rewards/math_answer_reward/std": 0.4732423722743988,
"sampling/importance_sampling_ratio/max": 2.9703896045684814,
"sampling/importance_sampling_ratio/mean": 0.5641449689865112,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9830443859100342,
"sampling/sampling_logp_difference/mean": 0.008612075820565224,
"step": 34,
"step_time": 398.4752207309357
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.11350122140720487,
"cast/masked_gap_abs_mean": 0.027373665187042207,
"cast/masked_gap_mean": -0.006000477617476463,
"cast/masked_negative_base_rate": 0.5005701640620828,
"cast/masked_positive_base_rate": 0.49942983826622367,
"cast/masked_sign_flipped_rate": 0.25855090748518705,
"cast/masked_teacher_negative_rate": 0.2792816683650017,
"cast/masked_teacher_positive_rate": 0.2673387941904366,
"cast/masked_token_adv_abs_mean": 0.6753621427342296,
"cast/masked_token_adv_max": 0.8819929575547576,
"cast/masked_token_adv_mean": 0.031560365881887265,
"cast/masked_token_adv_min": -0.981206295080483,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03561179165262729,
"cast/masked_wrong_good_token_rate": 0.1450496874167584,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0018721546684901114,
"clip_ratio/high_mean": 0.0018721546684901114,
"clip_ratio/low_mean": 0.0005859621205672738,
"clip_ratio/low_min": 0.0005859621205672738,
"clip_ratio/region_mean": 0.0024581168090662686,
"completions/clipped_ratio": 0.0,
"completions/max_length": 3340.0,
"completions/max_terminated_length": 3340.0,
"completions/mean_length": 1368.546875,
"completions/mean_terminated_length": 1368.546875,
"completions/min_length": 339.0,
"completions/min_terminated_length": 339.0,
"entropy": 0.24085174757055938,
"epoch": 0.006805366517596733,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.06529594957828522,
"learning_rate": 9.977963575085876e-07,
"loss": -0.028898751363158226,
"num_tokens": 4478893.0,
"reward": 0.53125,
"reward_std": 0.5029674172401428,
"rewards/math_answer_reward/mean": 0.53125,
"rewards/math_answer_reward/std": 0.5029674172401428,
"sampling/importance_sampling_ratio/max": 2.8871357440948486,
"sampling/importance_sampling_ratio/mean": 0.5544576644897461,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7373254299163818,
"sampling/sampling_logp_difference/mean": 0.011182185262441635,
"step": 35,
"step_time": 205.5380978989706
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.08872954820981249,
"cast/masked_gap_abs_mean": 0.023777092079399154,
"cast/masked_gap_mean": -0.004890330256557718,
"cast/masked_negative_base_rate": 0.6214258726686239,
"cast/masked_positive_base_rate": 0.37857412127777934,
"cast/masked_sign_flipped_rate": 0.2624443802051246,
"cast/masked_teacher_negative_rate": 0.29063388984650373,
"cast/masked_teacher_positive_rate": 0.27814409462735057,
"cast/masked_token_adv_abs_mean": 0.8270317735150456,
"cast/masked_token_adv_max": 1.0239122174680233,
"cast/masked_token_adv_mean": -0.0283673913218081,
"cast/masked_token_adv_min": -1.1326854601502419,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03409180592279881,
"cast/masked_wrong_good_token_rate": 0.1737148310057819,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0015291757381419302,
"clip_ratio/high_mean": 0.0015291757381419302,
"clip_ratio/low_mean": 0.0006507235898425279,
"clip_ratio/low_min": 0.0006507235898425279,
"clip_ratio/region_mean": 0.002179899323891732,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5517.0,
"completions/max_terminated_length": 5517.0,
"completions/mean_length": 1818.390625,
"completions/mean_terminated_length": 1818.390625,
"completions/min_length": 489.0,
"completions/min_terminated_length": 489.0,
"entropy": 0.2577267964370549,
"epoch": 0.00699980556095664,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.07322128117084503,
"learning_rate": 9.977315444941343e-07,
"loss": 0.04012518748641014,
"num_tokens": 4607886.0,
"reward": 0.40625,
"reward_std": 0.49501484632492065,
"rewards/math_answer_reward/mean": 0.40625,
"rewards/math_answer_reward/std": 0.49501484632492065,
"sampling/importance_sampling_ratio/max": 2.6711678504943848,
"sampling/importance_sampling_ratio/mean": 0.5081104636192322,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0078680515289307,
"sampling/sampling_logp_difference/mean": 0.010502639226615429,
"step": 36,
"step_time": 263.28461235386203
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.07490448467433453,
"cast/masked_gap_abs_mean": 0.015320657039410435,
"cast/masked_gap_mean": -0.0019661675319184724,
"cast/masked_negative_base_rate": 0.7441395493224263,
"cast/masked_positive_base_rate": 0.25586043833754957,
"cast/masked_sign_flipped_rate": 0.24956873338669538,
"cast/masked_teacher_negative_rate": 0.3079958772286773,
"cast/masked_teacher_positive_rate": 0.23495387891307473,
"cast/masked_token_adv_abs_mean": 0.9878798369318247,
"cast/masked_token_adv_max": 1.0781249701976776,
"cast/masked_token_adv_mean": -0.28953927190741524,
"cast/masked_token_adv_min": -1.2000000476837158,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.027064162772148848,
"cast/masked_wrong_good_token_rate": 0.17466424801386893,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0018939281198981917,
"clip_ratio/high_mean": 0.0018939281198981917,
"clip_ratio/low_mean": 0.000609281511970039,
"clip_ratio/low_min": 0.000609281511970039,
"clip_ratio/region_mean": 0.002503209640053683,
"completions/clipped_ratio": 0.046875,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 6785.0,
"completions/mean_length": 2384.515625,
"completions/mean_terminated_length": 2098.901611328125,
"completions/min_length": 381.0,
"completions/min_terminated_length": 381.0,
"entropy": 0.19611806562170386,
"epoch": 0.007194244604316547,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.04399775341153145,
"learning_rate": 9.97666731479681e-07,
"loss": 0.11257432401180267,
"num_tokens": 4771783.0,
"reward": 0.3125,
"reward_std": 0.467176616191864,
"rewards/math_answer_reward/mean": 0.3125,
"rewards/math_answer_reward/std": 0.467176616191864,
"sampling/importance_sampling_ratio/max": 2.716902256011963,
"sampling/importance_sampling_ratio/mean": 0.4982875883579254,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.6463074684143066,
"sampling/sampling_logp_difference/mean": 0.008233757689595222,
"step": 37,
"step_time": 439.662980426976
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.06128438748419285,
"cast/masked_gap_abs_mean": 0.021976040021399967,
"cast/masked_gap_mean": -0.0051507197201772215,
"cast/masked_negative_base_rate": 0.7987304572016001,
"cast/masked_positive_base_rate": 0.20126953721046448,
"cast/masked_sign_flipped_rate": 0.2656559655442834,
"cast/masked_teacher_negative_rate": 0.29980143439024687,
"cast/masked_teacher_positive_rate": 0.25360310124233365,
"cast/masked_token_adv_abs_mean": 0.7855046652257442,
"cast/masked_token_adv_max": 0.9774105465039611,
"cast/masked_token_adv_mean": -0.19279355183243752,
"cast/masked_token_adv_min": -1.0686580454930663,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03205488598905504,
"cast/masked_wrong_good_token_rate": 0.20437157608103007,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0015974896496118163,
"clip_ratio/high_mean": 0.0015974896496118163,
"clip_ratio/low_mean": 0.0006451320950873196,
"clip_ratio/low_min": 0.0006451320950873196,
"clip_ratio/region_mean": 0.0022426217401516624,
"completions/clipped_ratio": 0.0,
"completions/max_length": 6385.0,
"completions/max_terminated_length": 6385.0,
"completions/mean_length": 2173.5,
"completions/mean_terminated_length": 2173.5,
"completions/min_length": 821.0,
"completions/min_terminated_length": 821.0,
"entropy": 0.2633713469840586,
"epoch": 0.007388683647676453,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.06283463537693024,
"learning_rate": 9.976019184652277e-07,
"loss": 0.07560800015926361,
"num_tokens": 4923727.0,
"reward": 0.171875,
"reward_std": 0.38025420904159546,
"rewards/math_answer_reward/mean": 0.171875,
"rewards/math_answer_reward/std": 0.38025420904159546,
"sampling/importance_sampling_ratio/max": 2.995880126953125,
"sampling/importance_sampling_ratio/mean": 0.40209323167800903,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0365272760391235,
"sampling/sampling_logp_difference/mean": 0.010598253458738327,
"step": 38,
"step_time": 300.99947757000336
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.08526245265966281,
"cast/masked_gap_abs_mean": 0.02161799991154112,
"cast/masked_gap_mean": -0.004182812929684587,
"cast/masked_negative_base_rate": 0.6814862163737416,
"cast/masked_positive_base_rate": 0.31851377384737134,
"cast/masked_sign_flipped_rate": 0.2759870602749288,
"cast/masked_teacher_negative_rate": 0.3112457701936364,
"cast/masked_teacher_positive_rate": 0.271091231610626,
"cast/masked_token_adv_abs_mean": 0.7308783968910575,
"cast/masked_token_adv_max": 0.8604638613760471,
"cast/masked_token_adv_mean": -0.14485811790655134,
"cast/masked_token_adv_min": -0.940530177205801,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03320777596672997,
"cast/masked_wrong_good_token_rate": 0.19072460662573576,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0019102787173324032,
"clip_ratio/high_mean": 0.0019102787173324032,
"clip_ratio/low_mean": 0.0007682078899051703,
"clip_ratio/low_min": 0.0007682078899051703,
"clip_ratio/region_mean": 0.0026784866022353526,
"completions/clipped_ratio": 0.0,
"completions/max_length": 7171.0,
"completions/max_terminated_length": 7171.0,
"completions/mean_length": 1983.359375,
"completions/mean_terminated_length": 1983.359375,
"completions/min_length": 414.0,
"completions/min_terminated_length": 414.0,
"entropy": 0.2311276299878955,
"epoch": 0.00758312269103636,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.052613355219364166,
"learning_rate": 9.975371054507744e-07,
"loss": 0.04312083125114441,
"num_tokens": 5063726.0,
"reward": 0.390625,
"reward_std": 0.4917473793029785,
"rewards/math_answer_reward/mean": 0.390625,
"rewards/math_answer_reward/std": 0.4917473793029785,
"sampling/importance_sampling_ratio/max": 2.4745662212371826,
"sampling/importance_sampling_ratio/mean": 0.4978492856025696,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0369445085525513,
"sampling/sampling_logp_difference/mean": 0.010587261989712715,
"step": 39,
"step_time": 347.17950499698054
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.06981995317619294,
"cast/masked_gap_abs_mean": 0.022856797455460764,
"cast/masked_gap_mean": -0.004166832875284854,
"cast/masked_negative_base_rate": 0.7591868117451668,
"cast/masked_positive_base_rate": 0.24081317894160748,
"cast/masked_sign_flipped_rate": 0.2680515693500638,
"cast/masked_teacher_negative_rate": 0.31132067972794175,
"cast/masked_teacher_positive_rate": 0.2572633158415556,
"cast/masked_token_adv_abs_mean": 0.7222169870510697,
"cast/masked_token_adv_max": 0.9283391162753105,
"cast/masked_token_adv_mean": -0.17160409659845755,
"cast/masked_token_adv_min": -1.0231995228677988,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03386510215932503,
"cast/masked_wrong_good_token_rate": 0.19823161722160876,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.00161303075037722,
"clip_ratio/high_mean": 0.00161303075037722,
"clip_ratio/low_mean": 0.0007940274795146252,
"clip_ratio/low_min": 0.0007940274795146252,
"clip_ratio/region_mean": 0.0024070582330750767,
"completions/clipped_ratio": 0.0,
"completions/max_length": 6490.0,
"completions/max_terminated_length": 6490.0,
"completions/mean_length": 1911.203125,
"completions/mean_terminated_length": 1911.203125,
"completions/min_length": 425.0,
"completions/min_terminated_length": 425.0,
"entropy": 0.24882917990908027,
"epoch": 0.0077775617343962666,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.04977863281965256,
"learning_rate": 9.974722924363211e-07,
"loss": 0.07515259087085724,
"num_tokens": 5199147.0,
"reward": 0.265625,
"reward_std": 0.44515693187713623,
"rewards/math_answer_reward/mean": 0.265625,
"rewards/math_answer_reward/std": 0.44515693187713623,
"sampling/importance_sampling_ratio/max": 2.631476879119873,
"sampling/importance_sampling_ratio/mean": 0.41724735498428345,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0841344594955444,
"sampling/sampling_logp_difference/mean": 0.010334123857319355,
"step": 40,
"step_time": 302.33100661594654
},
{
"cast/all_correct_group_rate": 0.25,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.06916080508381128,
"cast/masked_gap_abs_mean": 0.014911695790942758,
"cast/masked_gap_mean": -0.0012580934677544064,
"cast/masked_negative_base_rate": 0.7151693273335695,
"cast/masked_positive_base_rate": 0.28483065916225314,
"cast/masked_sign_flipped_rate": 0.2790391896851361,
"cast/masked_teacher_negative_rate": 0.32306996546685696,
"cast/masked_teacher_positive_rate": 0.2912965421564877,
"cast/masked_token_adv_abs_mean": 0.9488120581954718,
"cast/masked_token_adv_max": 1.0546874552965164,
"cast/masked_token_adv_mean": -0.20035551459295675,
"cast/masked_token_adv_min": -1.2000000476837158,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.027575060434173793,
"cast/masked_wrong_good_token_rate": 0.2098783862311393,
"cast/mixed_group_rate": 0.125,
"clip_ratio/high_max": 0.0020611212621588493,
"clip_ratio/high_mean": 0.0020611212621588493,
"clip_ratio/low_mean": 0.0007487212369596818,
"clip_ratio/low_min": 0.0007487212369596818,
"clip_ratio/region_mean": 0.0028098424954805523,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5073.0,
"completions/max_terminated_length": 5073.0,
"completions/mean_length": 1776.671875,
"completions/mean_terminated_length": 1776.671875,
"completions/min_length": 515.0,
"completions/min_terminated_length": 515.0,
"entropy": 0.25791268702596426,
"epoch": 0.007972000777756174,
"frac_reward_zero_std": 0.875,
"grad_norm": 0.07107025384902954,
"learning_rate": 9.974074794218678e-07,
"loss": 0.068091481924057,
"num_tokens": 5324854.0,
"reward": 0.359375,
"reward_std": 0.4836103618144989,
"rewards/math_answer_reward/mean": 0.359375,
"rewards/math_answer_reward/std": 0.4836103618144989,
"sampling/importance_sampling_ratio/max": 2.021629810333252,
"sampling/importance_sampling_ratio/mean": 0.49875715374946594,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8505187034606934,
"sampling/sampling_logp_difference/mean": 0.01156819798052311,
"step": 41,
"step_time": 220.73192399091204
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.875,
"cast/masked_correct_bad_token_rate": 0.02900956175290048,
"cast/masked_gap_abs_mean": 0.01278109461418353,
"cast/masked_gap_mean": -0.0014367601090725657,
"cast/masked_negative_base_rate": 0.8930426947772503,
"cast/masked_positive_base_rate": 0.10695729777216911,
"cast/masked_sign_flipped_rate": 0.2794750318862498,
"cast/masked_teacher_negative_rate": 0.3437515404075384,
"cast/masked_teacher_positive_rate": 0.2772607929073274,
"cast/masked_token_adv_abs_mean": 0.9416233850643039,
"cast/masked_token_adv_max": 1.033472646959126,
"cast/masked_token_adv_mean": -0.3786330249567982,
"cast/masked_token_adv_min": -1.1757545499131083,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.025391928502358496,
"cast/masked_wrong_good_token_rate": 0.2504654701333493,
"cast/mixed_group_rate": 0.125,
"clip_ratio/high_max": 0.002045947687292937,
"clip_ratio/high_mean": 0.002045947687292937,
"clip_ratio/low_mean": 0.0008769638516241685,
"clip_ratio/low_min": 0.0008769638516241685,
"clip_ratio/region_mean": 0.0029229115207272116,
"completions/clipped_ratio": 0.0,
"completions/max_length": 6679.0,
"completions/max_terminated_length": 6679.0,
"completions/mean_length": 1824.046875,
"completions/mean_terminated_length": 1824.046875,
"completions/min_length": 364.0,
"completions/min_terminated_length": 364.0,
"entropy": 0.25340477377176285,
"epoch": 0.00816643982111608,
"frac_reward_zero_std": 0.875,
"grad_norm": 0.06542278081178665,
"learning_rate": 9.973426664074145e-07,
"loss": 0.2097892165184021,
"num_tokens": 5452161.0,
"reward": 0.109375,
"reward_std": 0.3145764470100403,
"rewards/math_answer_reward/mean": 0.109375,
"rewards/math_answer_reward/std": 0.3145764470100403,
"sampling/importance_sampling_ratio/max": 2.8988795280456543,
"sampling/importance_sampling_ratio/mean": 0.6327698826789856,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.15824556350708,
"sampling/sampling_logp_difference/mean": 0.011026299558579922,
"step": 42,
"step_time": 259.5605158561084
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.07230629539117217,
"cast/masked_gap_abs_mean": 0.022521272810990922,
"cast/masked_gap_mean": -0.005178376764547465,
"cast/masked_negative_base_rate": 0.7462949650362134,
"cast/masked_positive_base_rate": 0.2537050321698189,
"cast/masked_sign_flipped_rate": 0.29132723715156317,
"cast/masked_teacher_negative_rate": 0.33171342546120286,
"cast/masked_teacher_positive_rate": 0.2876292294822633,
"cast/masked_token_adv_abs_mean": 0.8103791661560535,
"cast/masked_token_adv_max": 1.02059980109334,
"cast/masked_token_adv_mean": -0.19625127501785755,
"cast/masked_token_adv_min": -1.1342569906264544,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03277414554031566,
"cast/masked_wrong_good_token_rate": 0.2190209401305765,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.002270306933496613,
"clip_ratio/high_mean": 0.002270306933496613,
"clip_ratio/low_mean": 0.0008293706550830393,
"clip_ratio/low_min": 0.0008293706550830393,
"clip_ratio/region_mean": 0.0030996776149549987,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4236.0,
"completions/max_terminated_length": 4236.0,
"completions/mean_length": 1701.859375,
"completions/mean_terminated_length": 1701.859375,
"completions/min_length": 578.0,
"completions/min_terminated_length": 578.0,
"entropy": 0.27594127878546715,
"epoch": 0.008360878864475986,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.05762952193617821,
"learning_rate": 9.972778533929612e-07,
"loss": 0.09932228177785873,
"num_tokens": 5571808.0,
"reward": 0.25,
"reward_std": 0.4364357888698578,
"rewards/math_answer_reward/mean": 0.25,
"rewards/math_answer_reward/std": 0.4364357888698578,
"sampling/importance_sampling_ratio/max": 2.6748857498168945,
"sampling/importance_sampling_ratio/mean": 0.41678786277770996,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.885716199874878,
"sampling/sampling_logp_difference/mean": 0.012231443077325821,
"step": 43,
"step_time": 260.91357902495656
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.056054261745885015,
"cast/masked_gap_abs_mean": 0.019705080572748557,
"cast/masked_gap_mean": -0.0035702796677696824,
"cast/masked_negative_base_rate": 0.7693555569276214,
"cast/masked_positive_base_rate": 0.2306444412097335,
"cast/masked_sign_flipped_rate": 0.2874224968254566,
"cast/masked_teacher_negative_rate": 0.32890443969517946,
"cast/masked_teacher_positive_rate": 0.280002293176949,
"cast/masked_token_adv_abs_mean": 0.7796087265014648,
"cast/masked_token_adv_max": 1.0007508611306548,
"cast/masked_token_adv_mean": -0.19633605575654656,
"cast/masked_token_adv_min": -1.1169296326115727,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.031929607095662504,
"cast/masked_wrong_good_token_rate": 0.2313682340318337,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.002173602320908685,
"clip_ratio/high_mean": 0.002173602320908685,
"clip_ratio/low_mean": 0.0009068159733942593,
"clip_ratio/low_min": 0.0009068159733942593,
"clip_ratio/region_mean": 0.0030804182906649658,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5432.0,
"completions/max_terminated_length": 5432.0,
"completions/mean_length": 1683.03125,
"completions/mean_terminated_length": 1683.03125,
"completions/min_length": 281.0,
"completions/min_terminated_length": 281.0,
"entropy": 0.2808148446492851,
"epoch": 0.008555317907835893,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.07006601244211197,
"learning_rate": 9.97213040378508e-07,
"loss": 0.09357066452503204,
"num_tokens": 5691210.0,
"reward": 0.21875,
"reward_std": 0.4166666865348816,
"rewards/math_answer_reward/mean": 0.21875,
"rewards/math_answer_reward/std": 0.4166666865348816,
"sampling/importance_sampling_ratio/max": 2.213256597518921,
"sampling/importance_sampling_ratio/mean": 0.45158281922340393,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.2390527725219727,
"sampling/sampling_logp_difference/mean": 0.012214520014822483,
"step": 44,
"step_time": 278.9529197560041
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.05483709799591452,
"cast/masked_gap_abs_mean": 0.017569850228028372,
"cast/masked_gap_mean": -0.002894192718855493,
"cast/masked_negative_base_rate": 0.7822737013921142,
"cast/masked_positive_base_rate": 0.21772629534825683,
"cast/masked_sign_flipped_rate": 0.2667382131330669,
"cast/masked_teacher_negative_rate": 0.3285700702108443,
"cast/masked_teacher_positive_rate": 0.2619658843614161,
"cast/masked_token_adv_abs_mean": 0.8665894009172916,
"cast/masked_token_adv_max": 1.0260647302493453,
"cast/masked_token_adv_mean": -0.2720907983602956,
"cast/masked_token_adv_min": -1.149531508795917,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.029608766024466604,
"cast/masked_wrong_good_token_rate": 0.21190111548639834,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0019314425571792526,
"clip_ratio/high_mean": 0.0019314425571792526,
"clip_ratio/low_mean": 0.0007958761780173518,
"clip_ratio/low_min": 0.0007958761780173518,
"clip_ratio/region_mean": 0.0027273187315586256,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5174.0,
"completions/max_terminated_length": 5174.0,
"completions/mean_length": 1878.421875,
"completions/mean_terminated_length": 1878.421875,
"completions/min_length": 686.0,
"completions/min_terminated_length": 686.0,
"entropy": 0.24939055973663926,
"epoch": 0.0087497569511958,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.03968026861548424,
"learning_rate": 9.971482273640546e-07,
"loss": 0.06260300427675247,
"num_tokens": 5823005.0,
"reward": 0.28125,
"reward_std": 0.4531635046005249,
"rewards/math_answer_reward/mean": 0.28125,
"rewards/math_answer_reward/std": 0.4531635046005249,
"sampling/importance_sampling_ratio/max": 1.8932844400405884,
"sampling/importance_sampling_ratio/mean": 0.3252776265144348,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9672952890396118,
"sampling/sampling_logp_difference/mean": 0.01098933257162571,
"step": 45,
"step_time": 229.8125469549268
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.04760466847801581,
"cast/masked_gap_abs_mean": 0.01832439858117141,
"cast/masked_gap_mean": -0.0022738266414989994,
"cast/masked_negative_base_rate": 0.8100231778807938,
"cast/masked_positive_base_rate": 0.18997682351619005,
"cast/masked_sign_flipped_rate": 0.29491065302863717,
"cast/masked_teacher_negative_rate": 0.3426437973976135,
"cast/masked_teacher_positive_rate": 0.29817007295787334,
"cast/masked_token_adv_abs_mean": 0.8778185723349452,
"cast/masked_token_adv_max": 1.0328885139897466,
"cast/masked_token_adv_mean": -0.17654119385406375,
"cast/masked_token_adv_min": -1.1429440854117274,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.030793300829827785,
"cast/masked_wrong_good_token_rate": 0.24730598519090563,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0017252627003472298,
"clip_ratio/high_mean": 0.0017252627003472298,
"clip_ratio/low_mean": 0.0006544677839883661,
"clip_ratio/low_min": 0.0006544677839883661,
"clip_ratio/region_mean": 0.0023797304820618592,
"completions/clipped_ratio": 0.0,
"completions/max_length": 7127.0,
"completions/max_terminated_length": 7127.0,
"completions/mean_length": 2425.484375,
"completions/mean_terminated_length": 2425.484375,
"completions/min_length": 688.0,
"completions/min_terminated_length": 688.0,
"entropy": 0.2886524763889611,
"epoch": 0.008944195994555707,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.04130927845835686,
"learning_rate": 9.970834143496013e-07,
"loss": 0.10161659866571426,
"num_tokens": 5991996.0,
"reward": 0.21875,
"reward_std": 0.4166666865348816,
"rewards/math_answer_reward/mean": 0.21875,
"rewards/math_answer_reward/std": 0.4166666865348816,
"sampling/importance_sampling_ratio/max": 1.7813085317611694,
"sampling/importance_sampling_ratio/mean": 0.33861660957336426,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 4.677785873413086,
"sampling/sampling_logp_difference/mean": 0.01197443064302206,
"step": 46,
"step_time": 365.22155502802343
},
{
"cast/all_correct_group_rate": 0.25,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.1056745310779661,
"cast/masked_gap_abs_mean": 0.036080819758353755,
"cast/masked_gap_mean": -0.007912280263781213,
"cast/masked_negative_base_rate": 0.6405656998977065,
"cast/masked_positive_base_rate": 0.3594342996366322,
"cast/masked_sign_flipped_rate": 0.3013916821219027,
"cast/masked_teacher_negative_rate": 0.32708696462213993,
"cast/masked_teacher_positive_rate": 0.30123277055099607,
"cast/masked_token_adv_abs_mean": 0.8158738790079951,
"cast/masked_token_adv_max": 0.9858372788876295,
"cast/masked_token_adv_mean": -0.011611061170697212,
"cast/masked_token_adv_min": -1.0945283863693476,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.04031826287973672,
"cast/masked_wrong_good_token_rate": 0.19571715244092047,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0015854206621952471,
"clip_ratio/high_mean": 0.0015854206621952471,
"clip_ratio/low_mean": 0.000977830962256121,
"clip_ratio/low_min": 0.000977830962256121,
"clip_ratio/region_mean": 0.0025632516353653045,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5460.0,
"completions/max_terminated_length": 5460.0,
"completions/mean_length": 1645.640625,
"completions/mean_terminated_length": 1645.640625,
"completions/min_length": 602.0,
"completions/min_terminated_length": 602.0,
"entropy": 0.2945717656984925,
"epoch": 0.009138635037915614,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.055282898247241974,
"learning_rate": 9.97018601335148e-07,
"loss": 0.010913747362792492,
"num_tokens": 6108821.0,
"reward": 0.421875,
"reward_std": 0.49776285886764526,
"rewards/math_answer_reward/mean": 0.421875,
"rewards/math_answer_reward/std": 0.49776285886764526,
"sampling/importance_sampling_ratio/max": 1.4192065000534058,
"sampling/importance_sampling_ratio/mean": 0.31854960322380066,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1254727840423584,
"sampling/sampling_logp_difference/mean": 0.013029222376644611,
"step": 47,
"step_time": 309.93392714107176
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.0,
"cast/masked_correct_bad_token_rate": 0.14885492203757167,
"cast/masked_gap_abs_mean": 0.03327723901020363,
"cast/masked_gap_mean": -0.00916987011078163,
"cast/masked_negative_base_rate": 0.4675047677010298,
"cast/masked_positive_base_rate": 0.5324952304363251,
"cast/masked_sign_flipped_rate": 0.2693721977993846,
"cast/masked_teacher_negative_rate": 0.2878989465534687,
"cast/masked_teacher_positive_rate": 0.2609111345373094,
"cast/masked_token_adv_abs_mean": 0.8754882458597422,
"cast/masked_token_adv_max": 0.9999091066420078,
"cast/masked_token_adv_mean": 0.04822459147544578,
"cast/masked_token_adv_min": -1.0731104537844658,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.040023430716246367,
"cast/masked_wrong_good_token_rate": 0.12051727727521211,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0015029124369903002,
"clip_ratio/high_mean": 0.0015029124369903002,
"clip_ratio/low_mean": 0.0006955884464332485,
"clip_ratio/low_min": 0.0006955884464332485,
"clip_ratio/region_mean": 0.002198500882514054,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 4730.0,
"completions/mean_length": 1259.015625,
"completions/mean_terminated_length": 1148.96826171875,
"completions/min_length": 460.0,
"completions/min_terminated_length": 460.0,
"entropy": 0.2170436200685799,
"epoch": 0.00933307408127552,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.1171441450715065,
"learning_rate": 9.969537883206947e-07,
"loss": -0.03215210139751434,
"num_tokens": 6199134.0,
"reward": 0.5,
"reward_std": 0.5039526224136353,
"rewards/math_answer_reward/mean": 0.5,
"rewards/math_answer_reward/std": 0.5039526224136353,
"sampling/importance_sampling_ratio/max": 2.7270658016204834,
"sampling/importance_sampling_ratio/mean": 0.6389820575714111,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.3418593406677246,
"sampling/sampling_logp_difference/mean": 0.011402653530240059,
"step": 48,
"step_time": 440.403916420124
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.042307715280912817,
"cast/masked_gap_abs_mean": 0.01699508825549856,
"cast/masked_gap_mean": -0.002787197065543978,
"cast/masked_negative_base_rate": 0.8146085143089294,
"cast/masked_positive_base_rate": 0.18539148615673184,
"cast/masked_sign_flipped_rate": 0.2807110813446343,
"cast/masked_teacher_negative_rate": 0.3220081776380539,
"cast/masked_teacher_positive_rate": 0.2797797406092286,
"cast/masked_token_adv_abs_mean": 0.8990280823782086,
"cast/masked_token_adv_max": 1.0737702921032906,
"cast/masked_token_adv_mean": -0.17650969326496124,
"cast/masked_token_adv_min": -1.1896661259233952,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02874106157105416,
"cast/masked_wrong_good_token_rate": 0.2384033661801368,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.001804048730264185,
"clip_ratio/high_mean": 0.001804048730264185,
"clip_ratio/low_mean": 0.0007947874619276263,
"clip_ratio/low_min": 0.0007947874619276263,
"clip_ratio/region_mean": 0.002598836155812023,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4457.0,
"completions/max_terminated_length": 4457.0,
"completions/mean_length": 1450.21875,
"completions/mean_terminated_length": 1450.21875,
"completions/min_length": 418.0,
"completions/min_terminated_length": 418.0,
"entropy": 0.2747834974434227,
"epoch": 0.009527513124635427,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.0751279816031456,
"learning_rate": 9.968889753062414e-07,
"loss": 0.08474267274141312,
"num_tokens": 6304180.0,
"reward": 0.234375,
"reward_std": 0.42695629596710205,
"rewards/math_answer_reward/mean": 0.234375,
"rewards/math_answer_reward/std": 0.42695629596710205,
"sampling/importance_sampling_ratio/max": 2.974398612976074,
"sampling/importance_sampling_ratio/mean": 0.5958175659179688,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9760918617248535,
"sampling/sampling_logp_difference/mean": 0.011843171902000904,
"step": 49,
"step_time": 263.82589706999715
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.06777688505826518,
"cast/masked_gap_abs_mean": 0.023731585330097005,
"cast/masked_gap_mean": -0.004118772136251891,
"cast/masked_negative_base_rate": 0.7226162496954203,
"cast/masked_positive_base_rate": 0.27738374238833785,
"cast/masked_sign_flipped_rate": 0.2831896136049181,
"cast/masked_teacher_negative_rate": 0.2818638854660094,
"cast/masked_teacher_positive_rate": 0.2920646029524505,
"cast/masked_token_adv_abs_mean": 0.6907868506386876,
"cast/masked_token_adv_max": 0.972032692283392,
"cast/masked_token_adv_mean": -0.06179959391010925,
"cast/masked_token_adv_min": -1.0412516947835684,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.035381800204049796,
"cast/masked_wrong_good_token_rate": 0.21541272813919932,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0012113352527194365,
"clip_ratio/high_mean": 0.0012113352527194365,
"clip_ratio/low_mean": 0.0008272698264590872,
"clip_ratio/low_min": 0.0008272698264590872,
"clip_ratio/region_mean": 0.0020386050764500396,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5003.0,
"completions/max_terminated_length": 5003.0,
"completions/mean_length": 1838.140625,
"completions/mean_terminated_length": 1838.140625,
"completions/min_length": 475.0,
"completions/min_terminated_length": 475.0,
"entropy": 0.2787684313952923,
"epoch": 0.009721952167995334,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.05889909341931343,
"learning_rate": 9.968241622917881e-07,
"loss": -0.022936755791306496,
"num_tokens": 6432429.0,
"reward": 0.3125,
"reward_std": 0.467176616191864,
"rewards/math_answer_reward/mean": 0.3125,
"rewards/math_answer_reward/std": 0.467176616191864,
"sampling/importance_sampling_ratio/max": 2.9716458320617676,
"sampling/importance_sampling_ratio/mean": 0.4343165159225464,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.9255127906799316,
"sampling/sampling_logp_difference/mean": 0.012797052040696144,
"step": 50,
"step_time": 322.24122984099085
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.09208891319576651,
"cast/masked_gap_abs_mean": 0.01981040278042201,
"cast/masked_gap_mean": -0.003532831242324619,
"cast/masked_negative_base_rate": 0.6648754612542689,
"cast/masked_positive_base_rate": 0.3351245285011828,
"cast/masked_sign_flipped_rate": 0.2682434576563537,
"cast/masked_teacher_negative_rate": 0.2951231310144067,
"cast/masked_teacher_positive_rate": 0.26542914332821965,
"cast/masked_token_adv_abs_mean": 0.8754056096076965,
"cast/masked_token_adv_max": 1.015033669769764,
"cast/masked_token_adv_mean": -0.16062757272447925,
"cast/masked_token_adv_min": -1.0957528129220009,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.031855312059633434,
"cast/masked_wrong_good_token_rate": 0.17615454434417188,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.001702809801827243,
"clip_ratio/high_mean": 0.001702809801827243,
"clip_ratio/low_mean": 0.0006894682246638695,
"clip_ratio/low_min": 0.0006894682246638695,
"clip_ratio/region_mean": 0.0023922780364955543,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5957.0,
"completions/max_terminated_length": 5957.0,
"completions/mean_length": 2311.796875,
"completions/mean_terminated_length": 2311.796875,
"completions/min_length": 440.0,
"completions/min_terminated_length": 440.0,
"entropy": 0.228669211268425,
"epoch": 0.00991639121135524,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.06941656023263931,
"learning_rate": 9.967593492773348e-07,
"loss": 0.1421968936920166,
"num_tokens": 6591384.0,
"reward": 0.375,
"reward_std": 0.48795005679130554,
"rewards/math_answer_reward/mean": 0.375,
"rewards/math_answer_reward/std": 0.48795005679130554,
"sampling/importance_sampling_ratio/max": 2.4207799434661865,
"sampling/importance_sampling_ratio/mean": 0.4641863703727722,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.170398473739624,
"sampling/sampling_logp_difference/mean": 0.010683245956897736,
"step": 51,
"step_time": 290.55334921600297
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.03696847567334771,
"cast/masked_gap_abs_mean": 0.021121723708347417,
"cast/masked_gap_mean": -0.004264076424533414,
"cast/masked_negative_base_rate": 0.8417554656043649,
"cast/masked_positive_base_rate": 0.15824452973902225,
"cast/masked_sign_flipped_rate": 0.29367287224158645,
"cast/masked_teacher_negative_rate": 0.34636165807023644,
"cast/masked_teacher_positive_rate": 0.3056874074973166,
"cast/masked_token_adv_abs_mean": 0.9330242592841387,
"cast/masked_token_adv_max": 1.0720829218626022,
"cast/masked_token_adv_mean": -0.1986540666257497,
"cast/masked_token_adv_min": -1.1823805533349514,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03201783279655501,
"cast/masked_wrong_good_token_rate": 0.2567043963354081,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.00193609668440331,
"clip_ratio/high_mean": 0.00193609668440331,
"clip_ratio/low_mean": 0.001020538365082757,
"clip_ratio/low_min": 0.001020538365082757,
"clip_ratio/region_mean": 0.0029566350640379824,
"completions/clipped_ratio": 0.0,
"completions/max_length": 3922.0,
"completions/max_terminated_length": 3922.0,
"completions/mean_length": 1880.09375,
"completions/mean_terminated_length": 1880.09375,
"completions/min_length": 595.0,
"completions/min_terminated_length": 595.0,
"entropy": 0.3004783168435097,
"epoch": 0.010110830254715146,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.07148437201976776,
"learning_rate": 9.966945362628815e-07,
"loss": 0.13325995206832886,
"num_tokens": 6724862.0,
"reward": 0.1875,
"reward_std": 0.39339789748191833,
"rewards/math_answer_reward/mean": 0.1875,
"rewards/math_answer_reward/std": 0.39339789748191833,
"sampling/importance_sampling_ratio/max": 2.7840282917022705,
"sampling/importance_sampling_ratio/mean": 0.4518299698829651,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.4987658262252808,
"sampling/sampling_logp_difference/mean": 0.013252981938421726,
"step": 52,
"step_time": 225.0795294759737
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.11199058697093278,
"cast/masked_gap_abs_mean": 0.026572534174192697,
"cast/masked_gap_mean": -0.007330546360435619,
"cast/masked_negative_base_rate": 0.5921763726510108,
"cast/masked_positive_base_rate": 0.4078236264176667,
"cast/masked_sign_flipped_rate": 0.2704781428910792,
"cast/masked_teacher_negative_rate": 0.3059976422227919,
"cast/masked_teacher_positive_rate": 0.2570579508319497,
"cast/masked_token_adv_abs_mean": 0.8392893318086863,
"cast/masked_token_adv_max": 1.0285062808543444,
"cast/masked_token_adv_mean": -0.11436766646511387,
"cast/masked_token_adv_min": -1.1165072321891785,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03617795306490734,
"cast/masked_wrong_good_token_rate": 0.15848755568731576,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0018035783505183645,
"clip_ratio/high_mean": 0.0018035783505183645,
"clip_ratio/low_mean": 0.0007119353936104744,
"clip_ratio/low_min": 0.0007119353936104744,
"clip_ratio/region_mean": 0.00251551373366965,
"completions/clipped_ratio": 0.0,
"completions/max_length": 3948.0,
"completions/max_terminated_length": 3948.0,
"completions/mean_length": 1587.5,
"completions/mean_terminated_length": 1587.5,
"completions/min_length": 613.0,
"completions/min_terminated_length": 613.0,
"entropy": 0.2325567991938442,
"epoch": 0.010305269298075053,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.10605103522539139,
"learning_rate": 9.966297232484283e-07,
"loss": 0.07153032720088959,
"num_tokens": 6837958.0,
"reward": 0.40625,
"reward_std": 0.49501484632492065,
"rewards/math_answer_reward/mean": 0.40625,
"rewards/math_answer_reward/std": 0.49501484632492065,
"sampling/importance_sampling_ratio/max": 2.951457977294922,
"sampling/importance_sampling_ratio/mean": 0.48967236280441284,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9377610683441162,
"sampling/sampling_logp_difference/mean": 0.012454560026526451,
"step": 53,
"step_time": 213.33191583488951
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.09282333170995116,
"cast/masked_gap_abs_mean": 0.029956402257084846,
"cast/masked_gap_mean": -0.007526466523131603,
"cast/masked_negative_base_rate": 0.6553607988171279,
"cast/masked_positive_base_rate": 0.3446392035111785,
"cast/masked_sign_flipped_rate": 0.2613898394629359,
"cast/masked_teacher_negative_rate": 0.32088689412921667,
"cast/masked_teacher_positive_rate": 0.24872201564721763,
"cast/masked_token_adv_abs_mean": 0.7068346161395311,
"cast/masked_token_adv_max": 0.9115561936050653,
"cast/masked_token_adv_mean": -0.09235530221485533,
"cast/masked_token_adv_min": -0.9989214148372412,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03711638739332557,
"cast/masked_wrong_good_token_rate": 0.16856650763656944,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0017756512825144455,
"clip_ratio/high_mean": 0.0017756512825144455,
"clip_ratio/low_mean": 0.0007182965709944256,
"clip_ratio/low_min": 0.0007182965709944256,
"clip_ratio/region_mean": 0.0024939478280430194,
"completions/clipped_ratio": 0.0,
"completions/max_length": 6695.0,
"completions/max_terminated_length": 6695.0,
"completions/mean_length": 1901.84375,
"completions/mean_terminated_length": 1901.84375,
"completions/min_length": 488.0,
"completions/min_terminated_length": 488.0,
"entropy": 0.24618721334263682,
"epoch": 0.01049970834143496,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.056419309228658676,
"learning_rate": 9.96564910233975e-07,
"loss": 0.005234166514128447,
"num_tokens": 6970524.0,
"reward": 0.359375,
"reward_std": 0.4836103618144989,
"rewards/math_answer_reward/mean": 0.359375,
"rewards/math_answer_reward/std": 0.4836103618144989,
"sampling/importance_sampling_ratio/max": 2.939702033996582,
"sampling/importance_sampling_ratio/mean": 0.47305911779403687,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1576030254364014,
"sampling/sampling_logp_difference/mean": 0.01211445964872837,
"step": 54,
"step_time": 378.4294174578972
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.75,
"cast/masked_correct_bad_token_rate": 0.017509002005681396,
"cast/masked_gap_abs_mean": 0.011163174334797077,
"cast/masked_gap_mean": -0.0009164843554572144,
"cast/masked_negative_base_rate": 0.9120821105316281,
"cast/masked_positive_base_rate": 0.08791788201779127,
"cast/masked_sign_flipped_rate": 0.2634035078808665,
"cast/masked_teacher_negative_rate": 0.3202851084060967,
"cast/masked_teacher_positive_rate": 0.2657169969752431,
"cast/masked_token_adv_abs_mean": 0.9450232982635498,
"cast/masked_token_adv_max": 1.064395286142826,
"cast/masked_token_adv_mean": -0.31971472236909904,
"cast/masked_token_adv_min": -1.1896661259233952,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.024674072221387178,
"cast/masked_wrong_good_token_rate": 0.2458945068065077,
"cast/mixed_group_rate": 0.25,
"clip_ratio/high_max": 0.002104801002133172,
"clip_ratio/high_mean": 0.002104801002133172,
"clip_ratio/low_mean": 0.0007490105158467486,
"clip_ratio/low_min": 0.0007490105158467486,
"clip_ratio/region_mean": 0.0028538115293486044,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 4902.0,
"completions/mean_length": 2252.6875,
"completions/mean_terminated_length": 2158.412841796875,
"completions/min_length": 897.0,
"completions/min_terminated_length": 897.0,
"entropy": 0.23907504929229617,
"epoch": 0.010694147384794867,
"frac_reward_zero_std": 0.75,
"grad_norm": 0.05300338566303253,
"learning_rate": 9.965000972195217e-07,
"loss": 0.17585761845111847,
"num_tokens": 7126200.0,
"reward": 0.078125,
"reward_std": 0.27048972249031067,
"rewards/math_answer_reward/mean": 0.078125,
"rewards/math_answer_reward/std": 0.27048972249031067,
"sampling/importance_sampling_ratio/max": 2.470280408859253,
"sampling/importance_sampling_ratio/mean": 0.48446863889694214,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.6462562084197998,
"sampling/sampling_logp_difference/mean": 0.011493071913719177,
"step": 55,
"step_time": 367.7506803739525
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.12328574899584055,
"cast/masked_gap_abs_mean": 0.02932522215996869,
"cast/masked_gap_mean": -0.007450343357504607,
"cast/masked_negative_base_rate": 0.5606556981801987,
"cast/masked_positive_base_rate": 0.4393442999571562,
"cast/masked_sign_flipped_rate": 0.27966641867533326,
"cast/masked_teacher_negative_rate": 0.31829622527584434,
"cast/masked_teacher_positive_rate": 0.2730967542156577,
"cast/masked_token_adv_abs_mean": 0.8080292409285903,
"cast/masked_token_adv_max": 0.9984474936500192,
"cast/masked_token_adv_mean": -0.11656789459084393,
"cast/masked_token_adv_min": -1.1035829139873385,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03687110013561323,
"cast/masked_wrong_good_token_rate": 0.15638066944666207,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0022842572561785346,
"clip_ratio/high_mean": 0.0022842572561785346,
"clip_ratio/low_mean": 0.0010095049110532273,
"clip_ratio/low_min": 0.0010095049110532273,
"clip_ratio/region_mean": 0.0032937621772362036,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4866.0,
"completions/max_terminated_length": 4866.0,
"completions/mean_length": 1615.65625,
"completions/mean_terminated_length": 1615.65625,
"completions/min_length": 563.0,
"completions/min_terminated_length": 563.0,
"entropy": 0.2733059129677713,
"epoch": 0.010888586428154774,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.09659390896558762,
"learning_rate": 9.964352842050684e-07,
"loss": 0.14960622787475586,
"num_tokens": 7240242.0,
"reward": 0.453125,
"reward_std": 0.501733124256134,
"rewards/math_answer_reward/mean": 0.453125,
"rewards/math_answer_reward/std": 0.501733124256134,
"sampling/importance_sampling_ratio/max": 2.705918788909912,
"sampling/importance_sampling_ratio/mean": 0.606426477432251,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0501103401184082,
"sampling/sampling_logp_difference/mean": 0.013655967079102993,
"step": 56,
"step_time": 299.6405909799505
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.08434560487512499,
"cast/masked_gap_abs_mean": 0.016597209658357315,
"cast/masked_gap_mean": -0.0030561784104179424,
"cast/masked_negative_base_rate": 0.7419142033904791,
"cast/masked_positive_base_rate": 0.25808579847216606,
"cast/masked_sign_flipped_rate": 0.2748896135017276,
"cast/masked_teacher_negative_rate": 0.3281150823459029,
"cast/masked_teacher_positive_rate": 0.24863105453550816,
"cast/masked_token_adv_abs_mean": 0.9379336182028055,
"cast/masked_token_adv_max": 1.0687499642372131,
"cast/masked_token_adv_mean": -0.2984958670567721,
"cast/masked_token_adv_min": -1.2000000476837158,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.028588214132469147,
"cast/masked_wrong_good_token_rate": 0.19054401014000177,
"cast/mixed_group_rate": 0.25,
"clip_ratio/high_max": 0.0022413638107536826,
"clip_ratio/high_mean": 0.0022413638107536826,
"clip_ratio/low_mean": 0.0008522675698259263,
"clip_ratio/low_min": 0.0008522675698259263,
"clip_ratio/region_mean": 0.003093631365118199,
"completions/clipped_ratio": 0.0,
"completions/max_length": 7495.0,
"completions/max_terminated_length": 7495.0,
"completions/mean_length": 1870.296875,
"completions/mean_terminated_length": 1870.296875,
"completions/min_length": 630.0,
"completions/min_terminated_length": 630.0,
"entropy": 0.23637179704383016,
"epoch": 0.01108302547151468,
"frac_reward_zero_std": 0.75,
"grad_norm": 0.06726193428039551,
"learning_rate": 9.96370471190615e-07,
"loss": 0.18551820516586304,
"num_tokens": 7374253.0,
"reward": 0.28125,
"reward_std": 0.4531635046005249,
"rewards/math_answer_reward/mean": 0.28125,
"rewards/math_answer_reward/std": 0.4531635046005249,
"sampling/importance_sampling_ratio/max": 2.5953357219696045,
"sampling/importance_sampling_ratio/mean": 0.517195463180542,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.7040975093841553,
"sampling/sampling_logp_difference/mean": 0.011444337666034698,
"step": 57,
"step_time": 318.27715014101705
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.06928757351124659,
"cast/masked_gap_abs_mean": 0.018555309216026217,
"cast/masked_gap_mean": -0.0030105494645340514,
"cast/masked_negative_base_rate": 0.7362182773649693,
"cast/masked_positive_base_rate": 0.26378172682598233,
"cast/masked_sign_flipped_rate": 0.2385275810956955,
"cast/masked_teacher_negative_rate": 0.28994466457515955,
"cast/masked_teacher_positive_rate": 0.2261885805055499,
"cast/masked_token_adv_abs_mean": 0.8518080543726683,
"cast/masked_token_adv_max": 1.0109032932668924,
"cast/masked_token_adv_mean": -0.17812304547987878,
"cast/masked_token_adv_min": -1.1338895577937365,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.030808804149273783,
"cast/masked_wrong_good_token_rate": 0.16924000752624124,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0020663385330408346,
"clip_ratio/high_mean": 0.0020663385330408346,
"clip_ratio/low_mean": 0.00080614720354788,
"clip_ratio/low_min": 0.00080614720354788,
"clip_ratio/region_mean": 0.0028724857311317464,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 5522.0,
"completions/mean_length": 1883.96875,
"completions/mean_terminated_length": 1783.8414306640625,
"completions/min_length": 611.0,
"completions/min_terminated_length": 611.0,
"entropy": 0.2036246198695153,
"epoch": 0.011277464514874587,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.05646713823080063,
"learning_rate": 9.963056581761618e-07,
"loss": 0.07500893622636795,
"num_tokens": 7507619.0,
"reward": 0.296875,
"reward_std": 0.4604927599430084,
"rewards/math_answer_reward/mean": 0.296875,
"rewards/math_answer_reward/std": 0.4604927599430084,
"sampling/importance_sampling_ratio/max": 2.8970704078674316,
"sampling/importance_sampling_ratio/mean": 0.5250893831253052,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0285167694091797,
"sampling/sampling_logp_difference/mean": 0.011293395422399044,
"step": 58,
"step_time": 382.33948635600973
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.1079303806181997,
"cast/masked_gap_abs_mean": 0.01911267059040256,
"cast/masked_gap_mean": -0.0033657551639407757,
"cast/masked_negative_base_rate": 0.49799581803381443,
"cast/masked_positive_base_rate": 0.502004174515605,
"cast/masked_sign_flipped_rate": 0.23979624453932047,
"cast/masked_teacher_negative_rate": 0.2789880386553705,
"cast/masked_teacher_positive_rate": 0.2592534013092518,
"cast/masked_token_adv_abs_mean": 0.8758938629180193,
"cast/masked_token_adv_max": 1.0266915019601583,
"cast/masked_token_adv_mean": -0.014265447534853593,
"cast/masked_token_adv_min": -1.125147495418787,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03295723331393674,
"cast/masked_wrong_good_token_rate": 0.13186586229130626,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.001883819484646665,
"clip_ratio/high_mean": 0.001883819484646665,
"clip_ratio/low_mean": 0.0007344352716245339,
"clip_ratio/low_min": 0.0007344352716245339,
"clip_ratio/region_mean": 0.00261825476809463,
"completions/clipped_ratio": 0.0,
"completions/max_length": 3341.0,
"completions/max_terminated_length": 3341.0,
"completions/mean_length": 1228.171875,
"completions/mean_terminated_length": 1228.171875,
"completions/min_length": 518.0,
"completions/min_terminated_length": 518.0,
"entropy": 0.2185242553241551,
"epoch": 0.011471903558234494,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.07078153640031815,
"learning_rate": 9.962408451617085e-07,
"loss": -0.008466268889605999,
"num_tokens": 7597702.0,
"reward": 0.484375,
"reward_std": 0.5037065148353577,
"rewards/math_answer_reward/mean": 0.484375,
"rewards/math_answer_reward/std": 0.5037065148353577,
"sampling/importance_sampling_ratio/max": 2.8606247901916504,
"sampling/importance_sampling_ratio/mean": 0.5620474815368652,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.2288920879364014,
"sampling/sampling_logp_difference/mean": 0.012892697006464005,
"step": 59,
"step_time": 220.07719951399486
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.019532027246896178,
"cast/masked_gap_abs_mean": 0.013052289010374807,
"cast/masked_gap_mean": -0.0014783646991531896,
"cast/masked_negative_base_rate": 0.9013398252427578,
"cast/masked_positive_base_rate": 0.09866016590967774,
"cast/masked_sign_flipped_rate": 0.26775365229696035,
"cast/masked_teacher_negative_rate": 0.322916594799608,
"cast/masked_teacher_positive_rate": 0.27042751340195537,
"cast/masked_token_adv_abs_mean": 0.8941772785037756,
"cast/masked_token_adv_max": 1.0576875880360603,
"cast/masked_token_adv_mean": -0.2769501215261698,
"cast/masked_token_adv_min": -1.192714475095272,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02612652478273958,
"cast/masked_wrong_good_token_rate": 0.2482216247008182,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0020760128491019714,
"clip_ratio/high_mean": 0.0020760128491019714,
"clip_ratio/low_mean": 0.0007436294081344386,
"clip_ratio/low_min": 0.0007436294081344386,
"clip_ratio/region_mean": 0.0028196422736073146,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5453.0,
"completions/max_terminated_length": 5453.0,
"completions/mean_length": 1981.046875,
"completions/mean_terminated_length": 1981.046875,
"completions/min_length": 283.0,
"completions/min_terminated_length": 283.0,
"entropy": 0.2528722523711622,
"epoch": 0.0116663426015944,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.0570807084441185,
"learning_rate": 9.961760321472552e-07,
"loss": 0.06658700108528137,
"num_tokens": 7736689.0,
"reward": 0.109375,
"reward_std": 0.3145764470100403,
"rewards/math_answer_reward/mean": 0.109375,
"rewards/math_answer_reward/std": 0.3145764470100403,
"sampling/importance_sampling_ratio/max": 2.6603035926818848,
"sampling/importance_sampling_ratio/mean": 0.34670162200927734,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 3.0436863899230957,
"sampling/sampling_logp_difference/mean": 0.012413431890308857,
"step": 60,
"step_time": 235.69594370003324
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.026989707082975656,
"cast/masked_gap_abs_mean": 0.015091136883711442,
"cast/masked_gap_mean": -0.0016078577507414593,
"cast/masked_negative_base_rate": 0.8654961185529828,
"cast/masked_positive_base_rate": 0.1345038777217269,
"cast/masked_sign_flipped_rate": 0.2819924377836287,
"cast/masked_teacher_negative_rate": 0.3544535553082824,
"cast/masked_teacher_positive_rate": 0.2952809897251427,
"cast/masked_token_adv_abs_mean": 0.8955369619652629,
"cast/masked_token_adv_max": 1.0281508099287748,
"cast/masked_token_adv_mean": -0.2805901729152538,
"cast/masked_token_adv_min": -1.1482438631355762,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.027334542537573725,
"cast/masked_wrong_good_token_rate": 0.25500273145735264,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.002337996455025859,
"clip_ratio/high_mean": 0.002337996455025859,
"clip_ratio/low_mean": 0.0008602124285062018,
"clip_ratio/low_min": 0.0008602124285062018,
"clip_ratio/region_mean": 0.003198208887624787,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4379.0,
"completions/max_terminated_length": 4379.0,
"completions/mean_length": 1810.828125,
"completions/mean_terminated_length": 1810.828125,
"completions/min_length": 426.0,
"completions/min_terminated_length": 426.0,
"entropy": 0.27206060383468866,
"epoch": 0.011860781644954306,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.06254898756742477,
"learning_rate": 9.961112191328019e-07,
"loss": 0.08755838125944138,
"num_tokens": 7865814.0,
"reward": 0.203125,
"reward_std": 0.40550529956817627,
"rewards/math_answer_reward/mean": 0.203125,
"rewards/math_answer_reward/std": 0.40550529956817627,
"sampling/importance_sampling_ratio/max": 2.7333719730377197,
"sampling/importance_sampling_ratio/mean": 0.5067349672317505,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8911075592041016,
"sampling/sampling_logp_difference/mean": 0.013327207416296005,
"step": 61,
"step_time": 190.55285292401095
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.0,
"cast/masked_correct_bad_token_rate": 0.14521082304418087,
"cast/masked_gap_abs_mean": 0.025562124283169396,
"cast/masked_gap_mean": -0.0063721620131218515,
"cast/masked_negative_base_rate": 0.4759402684867382,
"cast/masked_positive_base_rate": 0.5240597368683666,
"cast/masked_sign_flipped_rate": 0.2665015975944698,
"cast/masked_teacher_negative_rate": 0.25399826699867845,
"cast/masked_teacher_positive_rate": 0.25417220406234264,
"cast/masked_token_adv_abs_mean": 0.8416986782103777,
"cast/masked_token_adv_max": 0.9869773900136352,
"cast/masked_token_adv_mean": 0.02753257768927142,
"cast/masked_token_adv_min": -1.063688493333757,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03640865546185523,
"cast/masked_wrong_good_token_rate": 0.12129077315330505,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0013267741214804118,
"clip_ratio/high_mean": 0.0013267741214804118,
"clip_ratio/low_mean": 0.0007108516579137358,
"clip_ratio/low_min": 0.0007108516579137358,
"clip_ratio/region_mean": 0.002037625774391927,
"completions/clipped_ratio": 0.0,
"completions/max_length": 7194.0,
"completions/max_terminated_length": 7194.0,
"completions/mean_length": 1911.421875,
"completions/mean_terminated_length": 1911.421875,
"completions/min_length": 594.0,
"completions/min_terminated_length": 594.0,
"entropy": 0.20195830473676324,
"epoch": 0.012055220688314213,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.07982209324836731,
"learning_rate": 9.960464061183486e-07,
"loss": 0.050944987684488297,
"num_tokens": 7999401.0,
"reward": 0.546875,
"reward_std": 0.501733124256134,
"rewards/math_answer_reward/mean": 0.546875,
"rewards/math_answer_reward/std": 0.501733124256134,
"sampling/importance_sampling_ratio/max": 2.9339542388916016,
"sampling/importance_sampling_ratio/mean": 0.47482383251190186,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 4.631350517272949,
"sampling/sampling_logp_difference/mean": 0.0107650775462389,
"step": 62,
"step_time": 375.3122961970803
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.04242762224748731,
"cast/masked_gap_abs_mean": 0.018927806566352956,
"cast/masked_gap_mean": -0.0023916548578597485,
"cast/masked_negative_base_rate": 0.8577438807114959,
"cast/masked_positive_base_rate": 0.14225611742585897,
"cast/masked_sign_flipped_rate": 0.2812508810311556,
"cast/masked_teacher_negative_rate": 0.31554174143821,
"cast/masked_teacher_positive_rate": 0.2777915042825043,
"cast/masked_token_adv_abs_mean": 0.7889521298930049,
"cast/masked_token_adv_max": 0.9946974897757173,
"cast/masked_token_adv_mean": -0.2658546452294104,
"cast/masked_token_adv_min": -1.1100114928558469,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.031498800264671445,
"cast/masked_wrong_good_token_rate": 0.23882325878366828,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0018231169979117112,
"clip_ratio/high_mean": 0.0018231169979117112,
"clip_ratio/low_mean": 0.0007673868240090087,
"clip_ratio/low_min": 0.0007673868240090087,
"clip_ratio/region_mean": 0.0025905038019118365,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4681.0,
"completions/max_terminated_length": 4681.0,
"completions/mean_length": 1796.375,
"completions/mean_terminated_length": 1796.375,
"completions/min_length": 395.0,
"completions/min_terminated_length": 395.0,
"entropy": 0.25615110993385315,
"epoch": 0.01224965973167412,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.03897292912006378,
"learning_rate": 9.959815931038953e-07,
"loss": 0.07000210881233215,
"num_tokens": 8124409.0,
"reward": 0.140625,
"reward_std": 0.3503824472427368,
"rewards/math_answer_reward/mean": 0.140625,
"rewards/math_answer_reward/std": 0.3503824472427368,
"sampling/importance_sampling_ratio/max": 2.911404609680176,
"sampling/importance_sampling_ratio/mean": 0.36925214529037476,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.1038291454315186,
"sampling/sampling_logp_difference/mean": 0.013307884335517883,
"step": 63,
"step_time": 236.71527594403597
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.10839362745173275,
"cast/masked_gap_abs_mean": 0.023499861446907744,
"cast/masked_gap_mean": -0.005923997097227129,
"cast/masked_negative_base_rate": 0.5954568218439817,
"cast/masked_positive_base_rate": 0.4045431804843247,
"cast/masked_sign_flipped_rate": 0.2623394909314811,
"cast/masked_teacher_negative_rate": 0.2923989621922374,
"cast/masked_teacher_positive_rate": 0.25746095832437277,
"cast/masked_token_adv_abs_mean": 0.8260899586603045,
"cast/masked_token_adv_max": 1.0342155043035746,
"cast/masked_token_adv_mean": -0.09250241701374762,
"cast/masked_token_adv_min": -1.1444606482982635,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.033916386135388166,
"cast/masked_wrong_good_token_rate": 0.15394586289767176,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.002740088679274777,
"clip_ratio/high_mean": 0.002740088679274777,
"clip_ratio/low_mean": 0.00068346854277479,
"clip_ratio/low_min": 0.00068346854277479,
"clip_ratio/region_mean": 0.0034235572420584504,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4578.0,
"completions/max_terminated_length": 4578.0,
"completions/mean_length": 1170.828125,
"completions/mean_terminated_length": 1170.828125,
"completions/min_length": 286.0,
"completions/min_terminated_length": 286.0,
"entropy": 0.21401460003107786,
"epoch": 0.012444098775034027,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.09905269742012024,
"learning_rate": 9.95916780089442e-07,
"loss": 0.07141934335231781,
"num_tokens": 8209990.0,
"reward": 0.484375,
"reward_std": 0.5037065148353577,
"rewards/math_answer_reward/mean": 0.484375,
"rewards/math_answer_reward/std": 0.5037065148353577,
"sampling/importance_sampling_ratio/max": 2.2050797939300537,
"sampling/importance_sampling_ratio/mean": 0.48872649669647217,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.819434404373169,
"sampling/sampling_logp_difference/mean": 0.013465678319334984,
"step": 64,
"step_time": 264.209801022982
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.11225565912900493,
"cast/masked_gap_abs_mean": 0.02540155572933145,
"cast/masked_gap_mean": -0.005897237154158574,
"cast/masked_negative_base_rate": 0.5918788006529212,
"cast/masked_positive_base_rate": 0.40812119911424816,
"cast/masked_sign_flipped_rate": 0.28453885903581977,
"cast/masked_teacher_negative_rate": 0.31561617739498615,
"cast/masked_teacher_positive_rate": 0.280298191588372,
"cast/masked_token_adv_abs_mean": 0.7988204956054688,
"cast/masked_token_adv_max": 1.0008167456835508,
"cast/masked_token_adv_mean": -0.11498698440846056,
"cast/masked_token_adv_min": -1.1170049291104078,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03482303611235693,
"cast/masked_wrong_good_token_rate": 0.17228320008143783,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0022056146035538404,
"clip_ratio/high_mean": 0.0022056146035538404,
"clip_ratio/low_mean": 0.0007513090095017105,
"clip_ratio/low_min": 0.0007513090095017105,
"clip_ratio/region_mean": 0.0029569236166935298,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5249.0,
"completions/max_terminated_length": 5249.0,
"completions/mean_length": 1363.0625,
"completions/mean_terminated_length": 1363.0625,
"completions/min_length": 374.0,
"completions/min_terminated_length": 374.0,
"entropy": 0.2280926201492548,
"epoch": 0.012638537818393934,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.05448943376541138,
"learning_rate": 9.958519670749887e-07,
"loss": 0.006279608700424433,
"num_tokens": 8311130.0,
"reward": 0.453125,
"reward_std": 0.501733124256134,
"rewards/math_answer_reward/mean": 0.453125,
"rewards/math_answer_reward/std": 0.501733124256134,
"sampling/importance_sampling_ratio/max": 2.593477964401245,
"sampling/importance_sampling_ratio/mean": 0.4806937277317047,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.9117140769958496,
"sampling/sampling_logp_difference/mean": 0.013647285290062428,
"step": 65,
"step_time": 277.2148892811092
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.09737660363316536,
"cast/masked_gap_abs_mean": 0.015963705140165985,
"cast/masked_gap_mean": -0.0026694372365909658,
"cast/masked_negative_base_rate": 0.616733786650002,
"cast/masked_positive_base_rate": 0.3832662121858448,
"cast/masked_sign_flipped_rate": 0.26677480014041066,
"cast/masked_teacher_negative_rate": 0.2773253880441189,
"cast/masked_teacher_positive_rate": 0.2535180998966098,
"cast/masked_token_adv_abs_mean": 0.8718180013820529,
"cast/masked_token_adv_max": 1.022943602874875,
"cast/masked_token_adv_mean": -0.09932920822757296,
"cast/masked_token_adv_min": -1.1262213289737701,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.029681418731343,
"cast/masked_wrong_good_token_rate": 0.1693981965072453,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0016855551639309851,
"clip_ratio/high_mean": 0.0016855551639309851,
"clip_ratio/low_mean": 0.0007554739408988098,
"clip_ratio/low_min": 0.0007554739408988098,
"clip_ratio/region_mean": 0.002441029118926963,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 6014.0,
"completions/mean_length": 2345.984375,
"completions/mean_terminated_length": 2253.190673828125,
"completions/min_length": 760.0,
"completions/min_terminated_length": 760.0,
"entropy": 0.20751451514661312,
"epoch": 0.01283297686175384,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.036826781928539276,
"learning_rate": 9.957871540605352e-07,
"loss": 4.718080163002014e-06,
"num_tokens": 8473081.0,
"reward": 0.421875,
"reward_std": 0.49776285886764526,
"rewards/math_answer_reward/mean": 0.421875,
"rewards/math_answer_reward/std": 0.49776285886764526,
"sampling/importance_sampling_ratio/max": 2.276228427886963,
"sampling/importance_sampling_ratio/mean": 0.3609461188316345,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.496710181236267,
"sampling/sampling_logp_difference/mean": 0.01173495128750801,
"step": 66,
"step_time": 393.2689535920799
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.10645500500686467,
"cast/masked_gap_abs_mean": 0.02305358281591907,
"cast/masked_gap_mean": -0.005851109866853221,
"cast/masked_negative_base_rate": 0.5626683072187006,
"cast/masked_positive_base_rate": 0.4373316867277026,
"cast/masked_sign_flipped_rate": 0.2543840822763741,
"cast/masked_teacher_negative_rate": 0.2684559137560427,
"cast/masked_teacher_positive_rate": 0.2403659182600677,
"cast/masked_token_adv_abs_mean": 0.8278956320136786,
"cast/masked_token_adv_max": 1.0085157044231892,
"cast/masked_token_adv_mean": 0.025351319694891572,
"cast/masked_token_adv_min": -1.0990180019289255,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03436723753111437,
"cast/masked_wrong_good_token_rate": 0.14792907796800137,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0015293284186554956,
"clip_ratio/high_mean": 0.0015293284186554956,
"clip_ratio/low_mean": 0.0007841594060664647,
"clip_ratio/low_min": 0.0007841594060664647,
"clip_ratio/region_mean": 0.0023134878365453915,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4839.0,
"completions/max_terminated_length": 4839.0,
"completions/mean_length": 1451.34375,
"completions/mean_terminated_length": 1451.34375,
"completions/min_length": 321.0,
"completions/min_terminated_length": 321.0,
"entropy": 0.1822846969589591,
"epoch": 0.013027415905113747,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.0753243938088417,
"learning_rate": 9.95722341046082e-07,
"loss": -0.031574442982673645,
"num_tokens": 8575823.0,
"reward": 0.46875,
"reward_std": 0.5029674172401428,
"rewards/math_answer_reward/mean": 0.46875,
"rewards/math_answer_reward/std": 0.5029674172401428,
"sampling/importance_sampling_ratio/max": 2.848837375640869,
"sampling/importance_sampling_ratio/mean": 0.4853495657444,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0046765804290771,
"sampling/sampling_logp_difference/mean": 0.011719238013029099,
"step": 67,
"step_time": 257.20295863688807
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.75,
"cast/masked_correct_bad_token_rate": 0.008674063603393734,
"cast/masked_gap_abs_mean": 0.013911930305766873,
"cast/masked_gap_mean": -0.0014125155831639091,
"cast/masked_negative_base_rate": 0.9606668297201395,
"cast/masked_positive_base_rate": 0.03933315351605415,
"cast/masked_sign_flipped_rate": 0.2949646865017712,
"cast/masked_teacher_negative_rate": 0.33340252889320254,
"cast/masked_teacher_positive_rate": 0.2987431837245822,
"cast/masked_token_adv_abs_mean": 0.9105384051799774,
"cast/masked_token_adv_max": 1.0294507937505841,
"cast/masked_token_adv_mean": -0.31510631809942424,
"cast/masked_token_adv_min": -1.155086706392467,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.026344164100009948,
"cast/masked_wrong_good_token_rate": 0.28629062231630087,
"cast/mixed_group_rate": 0.25,
"clip_ratio/high_max": 0.002195772829509224,
"clip_ratio/high_mean": 0.002195772829509224,
"clip_ratio/low_mean": 0.0007392541228909977,
"clip_ratio/low_min": 0.0007392541228909977,
"clip_ratio/region_mean": 0.0029350269560382003,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 6134.0,
"completions/mean_length": 2119.203125,
"completions/mean_terminated_length": 2022.8096923828125,
"completions/min_length": 500.0,
"completions/min_terminated_length": 500.0,
"entropy": 0.263271220959723,
"epoch": 0.013221854948473654,
"frac_reward_zero_std": 0.75,
"grad_norm": 0.04589544236660004,
"learning_rate": 9.956575280316288e-07,
"loss": 0.07096798717975616,
"num_tokens": 8724428.0,
"reward": 0.0625,
"reward_std": 0.24397502839565277,
"rewards/math_answer_reward/mean": 0.0625,
"rewards/math_answer_reward/std": 0.24397502839565277,
"sampling/importance_sampling_ratio/max": 2.8011250495910645,
"sampling/importance_sampling_ratio/mean": 0.30385684967041016,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0155611038208008,
"sampling/sampling_logp_difference/mean": 0.014232408255338669,
"step": 68,
"step_time": 379.25852832201053
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.12017002189531922,
"cast/masked_gap_abs_mean": 0.030150309874443337,
"cast/masked_gap_mean": -0.008745835582658401,
"cast/masked_negative_base_rate": 0.4573919107206166,
"cast/masked_positive_base_rate": 0.542608086951077,
"cast/masked_sign_flipped_rate": 0.2491881176829338,
"cast/masked_teacher_negative_rate": 0.25276980036869645,
"cast/masked_teacher_positive_rate": 0.28622876247391105,
"cast/masked_token_adv_abs_mean": 0.8207677733153105,
"cast/masked_token_adv_max": 0.9811808913946152,
"cast/masked_token_adv_mean": -0.01509105326840654,
"cast/masked_token_adv_min": -1.094563938677311,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03725522255990654,
"cast/masked_wrong_good_token_rate": 0.12901809671893716,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0020089418430870865,
"clip_ratio/high_mean": 0.0020089418430870865,
"clip_ratio/low_mean": 0.0007389941392830224,
"clip_ratio/low_min": 0.0007389941392830224,
"clip_ratio/region_mean": 0.002747936012383434,
"completions/clipped_ratio": 0.03125,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 5667.0,
"completions/mean_length": 1760.859375,
"completions/mean_terminated_length": 1553.4031982421875,
"completions/min_length": 371.0,
"completions/min_terminated_length": 371.0,
"entropy": 0.19980755262076855,
"epoch": 0.01341629399183356,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.051032654941082,
"learning_rate": 9.955927150171755e-07,
"loss": 0.0203288234770298,
"num_tokens": 8850147.0,
"reward": 0.59375,
"reward_std": 0.49501484632492065,
"rewards/math_answer_reward/mean": 0.59375,
"rewards/math_answer_reward/std": 0.49501484632492065,
"sampling/importance_sampling_ratio/max": 2.2339940071105957,
"sampling/importance_sampling_ratio/mean": 0.4149836599826813,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.3100368976593018,
"sampling/sampling_logp_difference/mean": 0.01230557356029749,
"step": 69,
"step_time": 457.4533055240172
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.75,
"cast/masked_correct_bad_token_rate": 0.02939909475389868,
"cast/masked_gap_abs_mean": 0.01309524021053221,
"cast/masked_gap_mean": -0.0008285172083901671,
"cast/masked_negative_base_rate": 0.8901497293263674,
"cast/masked_positive_base_rate": 0.10985027113929391,
"cast/masked_sign_flipped_rate": 0.26481343200430274,
"cast/masked_teacher_negative_rate": 0.3278226535767317,
"cast/masked_teacher_positive_rate": 0.2642390700057149,
"cast/masked_token_adv_abs_mean": 0.9821283873170614,
"cast/masked_token_adv_max": 1.0687499642372131,
"cast/masked_token_adv_mean": -0.39625441865064204,
"cast/masked_token_adv_min": -1.2000000476837158,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02600766409887001,
"cast/masked_wrong_good_token_rate": 0.23541433713398874,
"cast/mixed_group_rate": 0.25,
"clip_ratio/high_max": 0.002550583875290613,
"clip_ratio/high_mean": 0.002550583875290613,
"clip_ratio/low_mean": 0.0008424592324445257,
"clip_ratio/low_min": 0.0008424592324445257,
"clip_ratio/region_mean": 0.00339304311091837,
"completions/clipped_ratio": 0.03125,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 4978.0,
"completions/mean_length": 2122.3125,
"completions/mean_terminated_length": 1926.51611328125,
"completions/min_length": 570.0,
"completions/min_terminated_length": 570.0,
"entropy": 0.23224759381264448,
"epoch": 0.013610733035193466,
"frac_reward_zero_std": 0.75,
"grad_norm": 0.05179428309202194,
"learning_rate": 9.95527902002722e-07,
"loss": 0.11830363422632217,
"num_tokens": 9000951.0,
"reward": 0.171875,
"reward_std": 0.38025420904159546,
"rewards/math_answer_reward/mean": 0.171875,
"rewards/math_answer_reward/std": 0.38025420904159546,
"sampling/importance_sampling_ratio/max": 2.0307788848876953,
"sampling/importance_sampling_ratio/mean": 0.40306612849235535,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.5474915504455566,
"sampling/sampling_logp_difference/mean": 0.013015633448958397,
"step": 70,
"step_time": 341.34962554089725
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.875,
"cast/masked_correct_bad_token_rate": 0.006036931648850441,
"cast/masked_gap_abs_mean": 0.010942415588942822,
"cast/masked_gap_mean": -0.0008490576410622452,
"cast/masked_negative_base_rate": 0.9807054847478867,
"cast/masked_positive_base_rate": 0.019294507801532745,
"cast/masked_sign_flipped_rate": 0.2671844174619764,
"cast/masked_teacher_negative_rate": 0.32581498147919774,
"cast/masked_teacher_positive_rate": 0.268001178978011,
"cast/masked_token_adv_abs_mean": 0.9506133440881968,
"cast/masked_token_adv_max": 1.0546874552965164,
"cast/masked_token_adv_mean": -0.431663254275918,
"cast/masked_token_adv_min": -1.2000000476837158,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02425901172682643,
"cast/masked_wrong_good_token_rate": 0.26114748581312597,
"cast/mixed_group_rate": 0.125,
"clip_ratio/high_max": 0.002212261995737208,
"clip_ratio/high_mean": 0.002212261995737208,
"clip_ratio/low_mean": 0.0008620840249022876,
"clip_ratio/low_min": 0.0008620840249022876,
"clip_ratio/region_mean": 0.003074345990171423,
"completions/clipped_ratio": 0.0,
"completions/max_length": 7915.0,
"completions/max_terminated_length": 7915.0,
"completions/mean_length": 2161.59375,
"completions/mean_terminated_length": 2161.59375,
"completions/min_length": 861.0,
"completions/min_terminated_length": 861.0,
"entropy": 0.25158591486979276,
"epoch": 0.013805172078553373,
"frac_reward_zero_std": 0.875,
"grad_norm": 0.06258294731378555,
"learning_rate": 9.954630889882689e-07,
"loss": 0.13555867969989777,
"num_tokens": 9152837.0,
"reward": 0.015625,
"reward_std": 0.125,
"rewards/math_answer_reward/mean": 0.015625,
"rewards/math_answer_reward/std": 0.125,
"sampling/importance_sampling_ratio/max": 2.781864643096924,
"sampling/importance_sampling_ratio/mean": 0.49871695041656494,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.550980567932129,
"sampling/sampling_logp_difference/mean": 0.012764372862875462,
"step": 71,
"step_time": 316.67405992618296
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.10016521340003237,
"cast/masked_gap_abs_mean": 0.026885384926572442,
"cast/masked_gap_mean": -0.007052542971337061,
"cast/masked_negative_base_rate": 0.5765360384248197,
"cast/masked_positive_base_rate": 0.4234639548230916,
"cast/masked_sign_flipped_rate": 0.2437310479581356,
"cast/masked_teacher_negative_rate": 0.28136274591088295,
"cast/masked_teacher_positive_rate": 0.2590452693402767,
"cast/masked_token_adv_abs_mean": 0.9463209789246321,
"cast/masked_token_adv_max": 1.0451844409108162,
"cast/masked_token_adv_mean": -0.06526732444763184,
"cast/masked_token_adv_min": -1.135567992925644,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.036387732601724565,
"cast/masked_wrong_good_token_rate": 0.14356583496555686,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0018866450882342178,
"clip_ratio/high_mean": 0.0018866450882342178,
"clip_ratio/low_mean": 0.000619549577095313,
"clip_ratio/low_min": 0.000619549577095313,
"clip_ratio/region_mean": 0.0025061946544155944,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5217.0,
"completions/max_terminated_length": 5217.0,
"completions/mean_length": 1403.328125,
"completions/mean_terminated_length": 1403.328125,
"completions/min_length": 331.0,
"completions/min_terminated_length": 331.0,
"entropy": 0.2215897182468325,
"epoch": 0.01399961112191328,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.08350963890552521,
"learning_rate": 9.953982759738156e-07,
"loss": -0.03314840421080589,
"num_tokens": 9253722.0,
"reward": 0.515625,
"reward_std": 0.5037065148353577,
"rewards/math_answer_reward/mean": 0.515625,
"rewards/math_answer_reward/std": 0.5037065148353577,
"sampling/importance_sampling_ratio/max": 2.6949048042297363,
"sampling/importance_sampling_ratio/mean": 0.48076295852661133,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.110546588897705,
"sampling/sampling_logp_difference/mean": 0.013868484646081924,
"step": 72,
"step_time": 306.8905190828955
},
{
"cast/all_correct_group_rate": 0.25,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.1514857456786558,
"cast/masked_gap_abs_mean": 0.035065659845713526,
"cast/masked_gap_mean": -0.010233993294377797,
"cast/masked_negative_base_rate": 0.34614958614110947,
"cast/masked_positive_base_rate": 0.6538504157215357,
"cast/masked_sign_flipped_rate": 0.2647973841521889,
"cast/masked_teacher_negative_rate": 0.2690280079841614,
"cast/masked_teacher_positive_rate": 0.2899699779227376,
"cast/masked_token_adv_abs_mean": 0.8604907561093569,
"cast/masked_token_adv_max": 1.0147856147959828,
"cast/masked_token_adv_mean": 0.13014899592963047,
"cast/masked_token_adv_min": -1.1484026154503226,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03968079580226913,
"cast/masked_wrong_good_token_rate": 0.11331163812428713,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0017965268325497163,
"clip_ratio/high_mean": 0.0017965268325497163,
"clip_ratio/low_mean": 0.0008288508997793542,
"clip_ratio/low_min": 0.0008288508997793542,
"clip_ratio/region_mean": 0.002625377746880986,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4824.0,
"completions/max_terminated_length": 4824.0,
"completions/mean_length": 980.9375,
"completions/mean_terminated_length": 980.9375,
"completions/min_length": 335.0,
"completions/min_terminated_length": 335.0,
"entropy": 0.23632271797396243,
"epoch": 0.014194050165273187,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.14383070170879364,
"learning_rate": 9.95333462959362e-07,
"loss": -0.07561352849006653,
"num_tokens": 9326158.0,
"reward": 0.640625,
"reward_std": 0.4836103618144989,
"rewards/math_answer_reward/mean": 0.640625,
"rewards/math_answer_reward/std": 0.4836103618144989,
"sampling/importance_sampling_ratio/max": 2.587641954421997,
"sampling/importance_sampling_ratio/mean": 0.6466450095176697,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0842843055725098,
"sampling/sampling_logp_difference/mean": 0.014172445051372051,
"step": 73,
"step_time": 305.9208747880475
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.05797164258547127,
"cast/masked_gap_abs_mean": 0.0199999504693551,
"cast/masked_gap_mean": -0.004150158810716675,
"cast/masked_negative_base_rate": 0.7446632799692452,
"cast/masked_positive_base_rate": 0.2553367121145129,
"cast/masked_sign_flipped_rate": 0.24192516272887588,
"cast/masked_teacher_negative_rate": 0.3173753055743873,
"cast/masked_teacher_positive_rate": 0.2609021901153028,
"cast/masked_token_adv_abs_mean": 0.7440438792109489,
"cast/masked_token_adv_max": 0.931053064763546,
"cast/masked_token_adv_mean": -0.21117777231847867,
"cast/masked_token_adv_min": -1.0372749995440245,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.031725778768304735,
"cast/masked_wrong_good_token_rate": 0.18395352025981992,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0020972715519747,
"clip_ratio/high_mean": 0.0020972715519747,
"clip_ratio/low_mean": 0.0008252259422079078,
"clip_ratio/low_min": 0.0008252259422079078,
"clip_ratio/region_mean": 0.0029224974787211977,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4083.0,
"completions/max_terminated_length": 4083.0,
"completions/mean_length": 1474.625,
"completions/mean_terminated_length": 1474.625,
"completions/min_length": 385.0,
"completions/min_terminated_length": 385.0,
"entropy": 0.2396159884519875,
"epoch": 0.014388489208633094,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.0571376271545887,
"learning_rate": 9.95268649944909e-07,
"loss": 0.12479601055383682,
"num_tokens": 9432022.0,
"reward": 0.28125,
"reward_std": 0.4531635046005249,
"rewards/math_answer_reward/mean": 0.28125,
"rewards/math_answer_reward/std": 0.4531635046005249,
"sampling/importance_sampling_ratio/max": 2.9566938877105713,
"sampling/importance_sampling_ratio/mean": 0.5436280369758606,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.3833870887756348,
"sampling/sampling_logp_difference/mean": 0.014271045103669167,
"step": 74,
"step_time": 225.10347176599316
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.05054876324720681,
"cast/masked_gap_abs_mean": 0.01711568341124803,
"cast/masked_gap_mean": -0.002931514001176083,
"cast/masked_negative_base_rate": 0.7442568298429251,
"cast/masked_positive_base_rate": 0.2557431608438492,
"cast/masked_sign_flipped_rate": 0.2751299487426877,
"cast/masked_teacher_negative_rate": 0.30559394834563136,
"cast/masked_teacher_positive_rate": 0.28028195491060615,
"cast/masked_token_adv_abs_mean": 0.7448254972696304,
"cast/masked_token_adv_max": 0.9918404491618276,
"cast/masked_token_adv_mean": -0.16040567882737378,
"cast/masked_token_adv_min": -1.106619986705482,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.030186190560925752,
"cast/masked_wrong_good_token_rate": 0.22458118491340429,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.001931318232891499,
"clip_ratio/high_mean": 0.001931318232891499,
"clip_ratio/low_mean": 0.0006769709125364898,
"clip_ratio/low_min": 0.0006769709125364898,
"clip_ratio/region_mean": 0.002608289150884957,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4790.0,
"completions/max_terminated_length": 4790.0,
"completions/mean_length": 2143.078125,
"completions/mean_terminated_length": 2143.078125,
"completions/min_length": 458.0,
"completions/min_terminated_length": 458.0,
"entropy": 0.2605021698400378,
"epoch": 0.014582928251993,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.059629373252391815,
"learning_rate": 9.952038369304557e-07,
"loss": 0.09710555523633957,
"num_tokens": 9579699.0,
"reward": 0.265625,
"reward_std": 0.44515693187713623,
"rewards/math_answer_reward/mean": 0.265625,
"rewards/math_answer_reward/std": 0.44515693187713623,
"sampling/importance_sampling_ratio/max": 2.705278158187866,
"sampling/importance_sampling_ratio/mean": 0.4724620282649994,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.3831663131713867,
"sampling/sampling_logp_difference/mean": 0.014079036191105843,
"step": 75,
"step_time": 236.8925766638713
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.04897405137307942,
"cast/masked_gap_abs_mean": 0.015598101002979092,
"cast/masked_gap_mean": -0.0021794492742515104,
"cast/masked_negative_base_rate": 0.797152501065284,
"cast/masked_positive_base_rate": 0.2028474914841354,
"cast/masked_sign_flipped_rate": 0.22239531273953617,
"cast/masked_teacher_negative_rate": 0.2814150496851653,
"cast/masked_teacher_positive_rate": 0.2209913933183998,
"cast/masked_token_adv_abs_mean": 0.8042309721931815,
"cast/masked_token_adv_max": 1.0010836860165,
"cast/masked_token_adv_mean": -0.23475803004112095,
"cast/masked_token_adv_min": -1.106595710851252,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.028443500719731674,
"cast/masked_wrong_good_token_rate": 0.1734212611336261,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.001539375047741487,
"clip_ratio/high_mean": 0.001539375047741487,
"clip_ratio/low_mean": 0.0006915497474437871,
"clip_ratio/low_min": 0.0006915497474437871,
"clip_ratio/region_mean": 0.002230924788818811,
"completions/clipped_ratio": 0.03125,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 7870.0,
"completions/mean_length": 2183.703125,
"completions/mean_terminated_length": 1989.8870849609375,
"completions/min_length": 572.0,
"completions/min_terminated_length": 572.0,
"entropy": 0.20294540445320308,
"epoch": 0.014777367295352907,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.049406785517930984,
"learning_rate": 9.951390239160022e-07,
"loss": 0.18492041528224945,
"num_tokens": 9730720.0,
"reward": 0.265625,
"reward_std": 0.44515693187713623,
"rewards/math_answer_reward/mean": 0.265625,
"rewards/math_answer_reward/std": 0.44515693187713623,
"sampling/importance_sampling_ratio/max": 2.5651984214782715,
"sampling/importance_sampling_ratio/mean": 0.48095637559890747,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 3.6176815032958984,
"sampling/sampling_logp_difference/mean": 0.011136957444250584,
"step": 76,
"step_time": 378.82580151516595
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.030908936052583158,
"cast/masked_gap_abs_mean": 0.014691615433548577,
"cast/masked_gap_mean": -0.0033976965175526175,
"cast/masked_negative_base_rate": 0.8387785786762834,
"cast/masked_positive_base_rate": 0.16122141014784575,
"cast/masked_sign_flipped_rate": 0.22749048029072583,
"cast/masked_teacher_negative_rate": 0.29803017410449684,
"cast/masked_teacher_positive_rate": 0.24106493405997753,
"cast/masked_token_adv_abs_mean": 0.9647163432091475,
"cast/masked_token_adv_max": 1.0622703284025192,
"cast/masked_token_adv_mean": -0.34536741478950717,
"cast/masked_token_adv_min": -1.1868721097707748,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02554172178497538,
"cast/masked_wrong_good_token_rate": 0.19658154365606606,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.001799006140572601,
"clip_ratio/high_mean": 0.001799006140572601,
"clip_ratio/low_mean": 0.0006596053217435838,
"clip_ratio/low_min": 0.0006596053217435838,
"clip_ratio/region_mean": 0.0024586114632256795,
"completions/clipped_ratio": 0.0,
"completions/max_length": 7096.0,
"completions/max_terminated_length": 7096.0,
"completions/mean_length": 2531.390625,
"completions/mean_terminated_length": 2531.390625,
"completions/min_length": 456.0,
"completions/min_terminated_length": 456.0,
"entropy": 0.2237861747853458,
"epoch": 0.014971806338712814,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.031633295118808746,
"learning_rate": 9.950742109015489e-07,
"loss": 0.046409815549850464,
"num_tokens": 9906377.0,
"reward": 0.203125,
"reward_std": 0.40550529956817627,
"rewards/math_answer_reward/mean": 0.203125,
"rewards/math_answer_reward/std": 0.40550529956817627,
"sampling/importance_sampling_ratio/max": 2.6325724124908447,
"sampling/importance_sampling_ratio/mean": 0.2830667197704315,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.6528494358062744,
"sampling/sampling_logp_difference/mean": 0.012069109827280045,
"step": 77,
"step_time": 317.6759743329894
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.12085383641533554,
"cast/masked_gap_abs_mean": 0.03017867126618512,
"cast/masked_gap_mean": -0.007563625571947341,
"cast/masked_negative_base_rate": 0.5452797124162316,
"cast/masked_positive_base_rate": 0.4547202857211232,
"cast/masked_sign_flipped_rate": 0.26768231159076095,
"cast/masked_teacher_negative_rate": 0.3005653638392687,
"cast/masked_teacher_positive_rate": 0.2554940767586231,
"cast/masked_token_adv_abs_mean": 0.7780925128608942,
"cast/masked_token_adv_max": 0.9589521251618862,
"cast/masked_token_adv_mean": -0.09146550108562224,
"cast/masked_token_adv_min": -1.0477310493588448,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03812834282871336,
"cast/masked_wrong_good_token_rate": 0.14682847331278026,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.001917329366733611,
"clip_ratio/high_mean": 0.001917329366733611,
"clip_ratio/low_mean": 0.0006580319432032411,
"clip_ratio/low_min": 0.0006580319432032411,
"clip_ratio/region_mean": 0.0025753613153938204,
"completions/clipped_ratio": 0.0,
"completions/max_length": 3938.0,
"completions/max_terminated_length": 3938.0,
"completions/mean_length": 1592.3125,
"completions/mean_terminated_length": 1592.3125,
"completions/min_length": 584.0,
"completions/min_terminated_length": 584.0,
"entropy": 0.23090350488200784,
"epoch": 0.01516624538207272,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.03928327560424805,
"learning_rate": 9.950093978870958e-07,
"loss": 0.011212719604372978,
"num_tokens": 10018397.0,
"reward": 0.515625,
"reward_std": 0.5037065148353577,
"rewards/math_answer_reward/mean": 0.515625,
"rewards/math_answer_reward/std": 0.5037065148353577,
"sampling/importance_sampling_ratio/max": 1.0548644065856934,
"sampling/importance_sampling_ratio/mean": 0.19413438439369202,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.6105523109436035,
"sampling/sampling_logp_difference/mean": 0.014733578078448772,
"step": 78,
"step_time": 245.80363385099918
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.024690338410437107,
"cast/masked_gap_abs_mean": 0.014988324153819121,
"cast/masked_gap_mean": -0.00148137242069879,
"cast/masked_negative_base_rate": 0.909321703016758,
"cast/masked_positive_base_rate": 0.09067829605191946,
"cast/masked_sign_flipped_rate": 0.2967385263182223,
"cast/masked_teacher_negative_rate": 0.3381611993536353,
"cast/masked_teacher_positive_rate": 0.29326310148462653,
"cast/masked_token_adv_abs_mean": 0.8504461217671633,
"cast/masked_token_adv_max": 1.0432463623583317,
"cast/masked_token_adv_mean": -0.24284473748411983,
"cast/masked_token_adv_min": -1.1654959246516228,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.028072442684788257,
"cast/masked_wrong_good_token_rate": 0.2720481881406158,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0018893184733315138,
"clip_ratio/high_mean": 0.0018893184733315138,
"clip_ratio/low_mean": 0.0007722442969679832,
"clip_ratio/low_min": 0.0007722442969679832,
"clip_ratio/region_mean": 0.0026615627593855606,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 7089.0,
"completions/mean_length": 2954.90625,
"completions/mean_terminated_length": 2871.77783203125,
"completions/min_length": 803.0,
"completions/min_terminated_length": 803.0,
"entropy": 0.3078423058614135,
"epoch": 0.015360684425432626,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.023594539612531662,
"learning_rate": 9.949445848726425e-07,
"loss": 0.03818678483366966,
"num_tokens": 10220175.0,
"reward": 0.078125,
"reward_std": 0.27048972249031067,
"rewards/math_answer_reward/mean": 0.078125,
"rewards/math_answer_reward/std": 0.27048972249031067,
"sampling/importance_sampling_ratio/max": 2.892029285430908,
"sampling/importance_sampling_ratio/mean": 0.22207501530647278,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.0310425758361816,
"sampling/sampling_logp_difference/mean": 0.015710119158029556,
"step": 79,
"step_time": 390.1158590890409
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.12122780550271273,
"cast/masked_gap_abs_mean": 0.021819481189595535,
"cast/masked_gap_mean": -0.004141138899058205,
"cast/masked_negative_base_rate": 0.5152651919052005,
"cast/masked_positive_base_rate": 0.4847348006442189,
"cast/masked_sign_flipped_rate": 0.2702502580359578,
"cast/masked_teacher_negative_rate": 0.30549766262993217,
"cast/masked_teacher_positive_rate": 0.2834485829807818,
"cast/masked_token_adv_abs_mean": 0.8675577649846673,
"cast/masked_token_adv_max": 1.0449450556188822,
"cast/masked_token_adv_mean": -0.07457406462344807,
"cast/masked_token_adv_min": -1.1609497591853142,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03233132173772901,
"cast/masked_wrong_good_token_rate": 0.1490224525332451,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0022158010210659995,
"clip_ratio/high_mean": 0.0022158010210659995,
"clip_ratio/low_mean": 0.0009097748441035947,
"clip_ratio/low_min": 0.0009097748441035947,
"clip_ratio/region_mean": 0.0031255758740371675,
"completions/clipped_ratio": 0.0,
"completions/max_length": 7847.0,
"completions/max_terminated_length": 7847.0,
"completions/mean_length": 1606.5,
"completions/mean_terminated_length": 1606.5,
"completions/min_length": 493.0,
"completions/min_terminated_length": 493.0,
"entropy": 0.28932539047673345,
"epoch": 0.015555123468792533,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.0562044233083725,
"learning_rate": 9.94879771858189e-07,
"loss": 0.04391856864094734,
"num_tokens": 10333183.0,
"reward": 0.40625,
"reward_std": 0.49501484632492065,
"rewards/math_answer_reward/mean": 0.40625,
"rewards/math_answer_reward/std": 0.49501484632492065,
"sampling/importance_sampling_ratio/max": 2.965697765350342,
"sampling/importance_sampling_ratio/mean": 0.42001599073410034,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.5908066034317017,
"sampling/sampling_logp_difference/mean": 0.015711892396211624,
"step": 80,
"step_time": 398.5504199399147
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.05759149952791631,
"cast/masked_gap_abs_mean": 0.02271694573573768,
"cast/masked_gap_mean": -0.004219836017746559,
"cast/masked_negative_base_rate": 0.7859457723388914,
"cast/masked_positive_base_rate": 0.21405422291718423,
"cast/masked_sign_flipped_rate": 0.25855478923767805,
"cast/masked_teacher_negative_rate": 0.3204103233292699,
"cast/masked_teacher_positive_rate": 0.2685479912906885,
"cast/masked_token_adv_abs_mean": 0.8487129472196102,
"cast/masked_token_adv_max": 1.0573088712990284,
"cast/masked_token_adv_mean": -0.16897118429187685,
"cast/masked_token_adv_min": -1.1654959246516228,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.032401872507762164,
"cast/masked_wrong_good_token_rate": 0.20096328976069344,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0020989369331800845,
"clip_ratio/high_mean": 0.0020989369331800845,
"clip_ratio/low_mean": 0.000902929652511375,
"clip_ratio/low_min": 0.000902929652511375,
"clip_ratio/region_mean": 0.003001866592967417,
"completions/clipped_ratio": 0.0,
"completions/max_length": 3976.0,
"completions/max_terminated_length": 3976.0,
"completions/mean_length": 1605.25,
"completions/mean_terminated_length": 1605.25,
"completions/min_length": 3.0,
"completions/min_terminated_length": 3.0,
"entropy": 0.24997127801179886,
"epoch": 0.01574956251215244,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.06247047707438469,
"learning_rate": 9.948149588437357e-07,
"loss": 0.05979541689157486,
"num_tokens": 10447879.0,
"reward": 0.25,
"reward_std": 0.4364357888698578,
"rewards/math_answer_reward/mean": 0.25,
"rewards/math_answer_reward/std": 0.4364357888698578,
"sampling/importance_sampling_ratio/max": 2.644139528274536,
"sampling/importance_sampling_ratio/mean": 0.45634377002716064,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.0415730476379395,
"sampling/sampling_logp_difference/mean": 0.015406670048832893,
"step": 81,
"step_time": 266.9202053799527
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.08788202831055969,
"cast/masked_gap_abs_mean": 0.02519994265458081,
"cast/masked_gap_mean": -0.006422155444397504,
"cast/masked_negative_base_rate": 0.6183662274852395,
"cast/masked_positive_base_rate": 0.38163377391174436,
"cast/masked_sign_flipped_rate": 0.25764738069847226,
"cast/masked_teacher_negative_rate": 0.2890194724313915,
"cast/masked_teacher_positive_rate": 0.2783924285322428,
"cast/masked_token_adv_abs_mean": 0.8580179996788502,
"cast/masked_token_adv_max": 1.002469346858561,
"cast/masked_token_adv_mean": -0.10933808155823499,
"cast/masked_token_adv_min": -1.1242507575079799,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03398823260795325,
"cast/masked_wrong_good_token_rate": 0.16976535227149725,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.001943971260516264,
"clip_ratio/high_mean": 0.001943971260516264,
"clip_ratio/low_mean": 0.0009723725183903298,
"clip_ratio/low_min": 0.0009723725183903298,
"clip_ratio/region_mean": 0.0029163437502575107,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4409.0,
"completions/max_terminated_length": 4409.0,
"completions/mean_length": 1396.15625,
"completions/mean_terminated_length": 1396.15625,
"completions/min_length": 418.0,
"completions/min_terminated_length": 418.0,
"entropy": 0.2028041712474078,
"epoch": 0.015944001555512347,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.05737616494297981,
"learning_rate": 9.947501458292826e-07,
"loss": 0.013246686197817326,
"num_tokens": 10548873.0,
"reward": 0.5,
"reward_std": 0.5039526224136353,
"rewards/math_answer_reward/mean": 0.5,
"rewards/math_answer_reward/std": 0.5039526224136353,
"sampling/importance_sampling_ratio/max": 2.9951415061950684,
"sampling/importance_sampling_ratio/mean": 0.440087229013443,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.3778164386749268,
"sampling/sampling_logp_difference/mean": 0.01472086738795042,
"step": 82,
"step_time": 264.34695948002627
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.07736812811344862,
"cast/masked_gap_abs_mean": 0.01985525555210188,
"cast/masked_gap_mean": -0.004138016566002989,
"cast/masked_negative_base_rate": 0.6978404768742621,
"cast/masked_positive_base_rate": 0.302159515209496,
"cast/masked_sign_flipped_rate": 0.24450479540973902,
"cast/masked_teacher_negative_rate": 0.28325431048870087,
"cast/masked_teacher_positive_rate": 0.23665263038128614,
"cast/masked_token_adv_abs_mean": 0.7351079285144806,
"cast/masked_token_adv_max": 0.9246781915426254,
"cast/masked_token_adv_mean": -0.1105121897708159,
"cast/masked_token_adv_min": -1.0299894269555807,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03260745090665296,
"cast/masked_wrong_good_token_rate": 0.16713666578289121,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0017104882726926007,
"clip_ratio/high_mean": 0.0017104882726926007,
"clip_ratio/low_mean": 0.000624500170488318,
"clip_ratio/low_min": 0.000624500170488318,
"clip_ratio/region_mean": 0.0023349884349954664,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4784.0,
"completions/max_terminated_length": 4784.0,
"completions/mean_length": 1696.328125,
"completions/mean_terminated_length": 1696.328125,
"completions/min_length": 457.0,
"completions/min_terminated_length": 457.0,
"entropy": 0.21901856688782573,
"epoch": 0.016138440598872254,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.046357929706573486,
"learning_rate": 9.94685332814829e-07,
"loss": 0.02634461410343647,
"num_tokens": 10667814.0,
"reward": 0.328125,
"reward_std": 0.4732423722743988,
"rewards/math_answer_reward/mean": 0.328125,
"rewards/math_answer_reward/std": 0.4732423722743988,
"sampling/importance_sampling_ratio/max": 2.9045209884643555,
"sampling/importance_sampling_ratio/mean": 0.4222012758255005,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.2713241577148438,
"sampling/sampling_logp_difference/mean": 0.013769680634140968,
"step": 83,
"step_time": 274.507219907071
},
{
"cast/all_correct_group_rate": 0.25,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.09175263144425116,
"cast/masked_gap_abs_mean": 0.021454502377309836,
"cast/masked_gap_mean": -0.005908557388238478,
"cast/masked_negative_base_rate": 0.6462241159752011,
"cast/masked_positive_base_rate": 0.35377588169649243,
"cast/masked_sign_flipped_rate": 0.22995626833289862,
"cast/masked_teacher_negative_rate": 0.28106815135106444,
"cast/masked_teacher_positive_rate": 0.24886219576001167,
"cast/masked_token_adv_abs_mean": 0.891908023506403,
"cast/masked_token_adv_max": 0.9936055755242705,
"cast/masked_token_adv_mean": -0.16229717918031383,
"cast/masked_token_adv_min": -1.1239190706983209,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03159984922967851,
"cast/masked_wrong_good_token_rate": 0.13820363522972912,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0016552390802644368,
"clip_ratio/high_mean": 0.0016552390802644368,
"clip_ratio/low_mean": 0.0007516237324125541,
"clip_ratio/low_min": 0.0007516237324125541,
"clip_ratio/region_mean": 0.002406862811767496,
"completions/clipped_ratio": 0.0,
"completions/max_length": 8031.0,
"completions/max_terminated_length": 8031.0,
"completions/mean_length": 1921.09375,
"completions/mean_terminated_length": 1921.09375,
"completions/min_length": 493.0,
"completions/min_terminated_length": 493.0,
"entropy": 0.20612602564506233,
"epoch": 0.01633287964223216,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.0412062406539917,
"learning_rate": 9.946205198003758e-07,
"loss": 0.033581722527742386,
"num_tokens": 10801732.0,
"reward": 0.4375,
"reward_std": 0.5,
"rewards/math_answer_reward/mean": 0.4375,
"rewards/math_answer_reward/std": 0.5,
"sampling/importance_sampling_ratio/max": 2.4278311729431152,
"sampling/importance_sampling_ratio/mean": 0.24214141070842743,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.5728163719177246,
"sampling/sampling_logp_difference/mean": 0.012617121450603008,
"step": 84,
"step_time": 400.57771207304904
},
{
"cast/all_correct_group_rate": 0.375,
"cast/all_wrong_group_rate": 0.0,
"cast/masked_correct_bad_token_rate": 0.132009778637439,
"cast/masked_gap_abs_mean": 0.027908734686207026,
"cast/masked_gap_mean": -0.00941459901878261,
"cast/masked_negative_base_rate": 0.3721053935587406,
"cast/masked_positive_base_rate": 0.6278946027159691,
"cast/masked_sign_flipped_rate": 0.235485153272748,
"cast/masked_teacher_negative_rate": 0.21508114179596305,
"cast/masked_teacher_positive_rate": 0.2529021059162915,
"cast/masked_token_adv_abs_mean": 0.8072278508916497,
"cast/masked_token_adv_max": 1.0249658022075891,
"cast/masked_token_adv_mean": 0.21155625305254944,
"cast/masked_token_adv_min": -1.1338895577937365,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03634776937542483,
"cast/masked_wrong_good_token_rate": 0.10347537533380091,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0016802964637463447,
"clip_ratio/high_mean": 0.0016802964637463447,
"clip_ratio/low_mean": 0.0007258121404447593,
"clip_ratio/low_min": 0.0007258121404447593,
"clip_ratio/region_mean": 0.002406108604191104,
"completions/clipped_ratio": 0.0,
"completions/max_length": 3403.0,
"completions/max_terminated_length": 3403.0,
"completions/mean_length": 1138.0625,
"completions/mean_terminated_length": 1138.0625,
"completions/min_length": 543.0,
"completions/min_terminated_length": 543.0,
"entropy": 0.15323513862676919,
"epoch": 0.01652731868559207,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.05815603956580162,
"learning_rate": 9.945557067859225e-07,
"loss": -0.04857257008552551,
"num_tokens": 10885512.0,
"reward": 0.640625,
"reward_std": 0.4836103618144989,
"rewards/math_answer_reward/mean": 0.640625,
"rewards/math_answer_reward/std": 0.4836103618144989,
"sampling/importance_sampling_ratio/max": 1.8030431270599365,
"sampling/importance_sampling_ratio/mean": 0.3217121958732605,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.3074233531951904,
"sampling/sampling_logp_difference/mean": 0.013205742463469505,
"step": 85,
"step_time": 261.39173070093966
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.040811124606989324,
"cast/masked_gap_abs_mean": 0.01928760936425533,
"cast/masked_gap_mean": -0.00336475643302947,
"cast/masked_negative_base_rate": 0.8346325857564807,
"cast/masked_positive_base_rate": 0.16536741191521287,
"cast/masked_sign_flipped_rate": 0.27764033898711205,
"cast/masked_teacher_negative_rate": 0.3198752417229116,
"cast/masked_teacher_positive_rate": 0.2834220724180341,
"cast/masked_token_adv_abs_mean": 0.8680205950513482,
"cast/masked_token_adv_max": 1.0388257997110486,
"cast/masked_token_adv_mean": -0.2108814464590978,
"cast/masked_token_adv_min": -1.1550737610086799,
"cast/masked_token_weight_max": 1.1999871022999287,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03150649002054706,
"cast/masked_wrong_good_token_rate": 0.2368292142637074,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0019260895087427343,
"clip_ratio/high_mean": 0.0019260895087427343,
"clip_ratio/low_mean": 0.0007491348646908591,
"clip_ratio/low_min": 0.0007491348646908591,
"clip_ratio/region_mean": 0.0026752243375085527,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 5644.0,
"completions/mean_length": 1718.125,
"completions/mean_terminated_length": 1615.3651123046875,
"completions/min_length": 393.0,
"completions/min_terminated_length": 393.0,
"entropy": 0.26927599986083806,
"epoch": 0.016721757728951972,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.0369802750647068,
"learning_rate": 9.944908937714692e-07,
"loss": 0.021631160750985146,
"num_tokens": 11007520.0,
"reward": 0.21875,
"reward_std": 0.4166666865348816,
"rewards/math_answer_reward/mean": 0.21875,
"rewards/math_answer_reward/std": 0.4166666865348816,
"sampling/importance_sampling_ratio/max": 2.2676024436950684,
"sampling/importance_sampling_ratio/mean": 0.31913572549819946,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.2294950485229492,
"sampling/sampling_logp_difference/mean": 0.016522735357284546,
"step": 86,
"step_time": 378.2398617079307
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.0,
"cast/masked_correct_bad_token_rate": 0.06431408814387396,
"cast/masked_gap_abs_mean": 0.029850320948753506,
"cast/masked_gap_mean": -0.007346165675699012,
"cast/masked_negative_base_rate": 0.7286968003027141,
"cast/masked_positive_base_rate": 0.2713031987659633,
"cast/masked_sign_flipped_rate": 0.2433698638342321,
"cast/masked_teacher_negative_rate": 0.2556144269183278,
"cast/masked_teacher_positive_rate": 0.2477066209539771,
"cast/masked_token_adv_abs_mean": 0.6711854431778193,
"cast/masked_token_adv_max": 0.9270736500620842,
"cast/masked_token_adv_mean": -0.05769701569715835,
"cast/masked_token_adv_min": -1.005941353738308,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03782528982264921,
"cast/masked_wrong_good_token_rate": 0.17905577446799725,
"cast/mixed_group_rate": 1.0,
"clip_ratio/high_max": 0.0014527291837111989,
"clip_ratio/high_mean": 0.0014527291837111989,
"clip_ratio/low_mean": 0.0006799751195103454,
"clip_ratio/low_min": 0.0006799751195103454,
"clip_ratio/region_mean": 0.0021327043032215443,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5524.0,
"completions/max_terminated_length": 5524.0,
"completions/mean_length": 1676.375,
"completions/mean_terminated_length": 1676.375,
"completions/min_length": 534.0,
"completions/min_terminated_length": 534.0,
"entropy": 0.21897555189207196,
"epoch": 0.01691619677231188,
"frac_reward_zero_std": 0.0,
"grad_norm": 0.03431493788957596,
"learning_rate": 9.94426080757016e-07,
"loss": 0.007225247099995613,
"num_tokens": 11128448.0,
"reward": 0.3125,
"reward_std": 0.467176616191864,
"rewards/math_answer_reward/mean": 0.3125,
"rewards/math_answer_reward/std": 0.467176616191864,
"sampling/importance_sampling_ratio/max": 2.045844078063965,
"sampling/importance_sampling_ratio/mean": 0.19253623485565186,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1528491973876953,
"sampling/sampling_logp_difference/mean": 0.014867645688354969,
"step": 87,
"step_time": 299.78823803315754
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.04985179880168289,
"cast/masked_gap_abs_mean": 0.017136048700194806,
"cast/masked_gap_mean": -0.0025915791787838316,
"cast/masked_negative_base_rate": 0.7918969644233584,
"cast/masked_positive_base_rate": 0.2081030271947384,
"cast/masked_sign_flipped_rate": 0.2821108754724264,
"cast/masked_teacher_negative_rate": 0.3302389266900718,
"cast/masked_teacher_positive_rate": 0.2752064820379019,
"cast/masked_token_adv_abs_mean": 0.8124342765659094,
"cast/masked_token_adv_max": 1.0428476529195905,
"cast/masked_token_adv_mean": -0.18800109007861465,
"cast/masked_token_adv_min": -1.1757545499131083,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.029371694545261562,
"cast/masked_wrong_good_token_rate": 0.23225907795131207,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.00198361142520298,
"clip_ratio/high_mean": 0.00198361142520298,
"clip_ratio/low_mean": 0.0007612491162944934,
"clip_ratio/low_min": 0.0007612491162944934,
"clip_ratio/region_mean": 0.0027448605114841484,
"completions/clipped_ratio": 0.0,
"completions/max_length": 7061.0,
"completions/max_terminated_length": 7061.0,
"completions/mean_length": 1723.8125,
"completions/mean_terminated_length": 1723.8125,
"completions/min_length": 435.0,
"completions/min_terminated_length": 435.0,
"entropy": 0.27482923050411046,
"epoch": 0.017110635815671786,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.027105549350380898,
"learning_rate": 9.943612677425626e-07,
"loss": 0.015880821272730827,
"num_tokens": 11251260.0,
"reward": 0.265625,
"reward_std": 0.44515693187713623,
"rewards/math_answer_reward/mean": 0.265625,
"rewards/math_answer_reward/std": 0.44515693187713623,
"sampling/importance_sampling_ratio/max": 2.9907636642456055,
"sampling/importance_sampling_ratio/mean": 0.29343587160110474,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.1060853004455566,
"sampling/sampling_logp_difference/mean": 0.01606837473809719,
"step": 88,
"step_time": 369.7049394810165
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.08957548823673278,
"cast/masked_gap_abs_mean": 0.02285469106573146,
"cast/masked_gap_mean": -0.0045531419012334595,
"cast/masked_negative_base_rate": 0.639074582606554,
"cast/masked_positive_base_rate": 0.36092541459947824,
"cast/masked_sign_flipped_rate": 0.2821482648141682,
"cast/masked_teacher_negative_rate": 0.3204978075809777,
"cast/masked_teacher_positive_rate": 0.27800292568281293,
"cast/masked_token_adv_abs_mean": 0.9050066191703081,
"cast/masked_token_adv_max": 1.0538752600550652,
"cast/masked_token_adv_mean": -0.08741019140870776,
"cast/masked_token_adv_min": -1.1827499456703663,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03409111639484763,
"cast/masked_wrong_good_token_rate": 0.19257277820724994,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0021475402854775894,
"clip_ratio/high_mean": 0.0021475402854775894,
"clip_ratio/low_mean": 0.0007379850212601013,
"clip_ratio/low_min": 0.0007379850212601013,
"clip_ratio/region_mean": 0.0028855253076471854,
"completions/clipped_ratio": 0.0,
"completions/max_length": 3144.0,
"completions/max_terminated_length": 3144.0,
"completions/mean_length": 1399.453125,
"completions/mean_terminated_length": 1399.453125,
"completions/min_length": 420.0,
"completions/min_terminated_length": 420.0,
"entropy": 0.2729101004078984,
"epoch": 0.017305074859031693,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.06256688386201859,
"learning_rate": 9.942964547281093e-07,
"loss": 0.0029808301478624344,
"num_tokens": 11354241.0,
"reward": 0.375,
"reward_std": 0.48795005679130554,
"rewards/math_answer_reward/mean": 0.375,
"rewards/math_answer_reward/std": 0.48795005679130554,
"sampling/importance_sampling_ratio/max": 2.0112452507019043,
"sampling/importance_sampling_ratio/mean": 0.2966873347759247,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.9950289726257324,
"sampling/sampling_logp_difference/mean": 0.018590325489640236,
"step": 89,
"step_time": 173.57327941304538
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.0865205394802615,
"cast/masked_gap_abs_mean": 0.023229667858686298,
"cast/masked_gap_mean": -0.005730504243274481,
"cast/masked_negative_base_rate": 0.5812701564282179,
"cast/masked_positive_base_rate": 0.4187298398464918,
"cast/masked_sign_flipped_rate": 0.2717050106730312,
"cast/masked_teacher_negative_rate": 0.28039802168495953,
"cast/masked_teacher_positive_rate": 0.28455156343989074,
"cast/masked_token_adv_abs_mean": 0.7277112053707242,
"cast/masked_token_adv_max": 0.946481442078948,
"cast/masked_token_adv_mean": 0.03296627485542558,
"cast/masked_token_adv_min": -1.0441931392997503,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03410495526622981,
"cast/masked_wrong_good_token_rate": 0.1851844706106931,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0016635788715575472,
"clip_ratio/high_mean": 0.0016635788715575472,
"clip_ratio/low_mean": 0.0007879398326622322,
"clip_ratio/low_min": 0.0007879398326622322,
"clip_ratio/region_mean": 0.0024515187033102848,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 6232.0,
"completions/mean_length": 1763.96875,
"completions/mean_terminated_length": 1661.9366455078125,
"completions/min_length": 679.0,
"completions/min_terminated_length": 679.0,
"entropy": 0.2519535430474207,
"epoch": 0.0174995139023916,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.030814219266176224,
"learning_rate": 9.94231641713656e-07,
"loss": -0.002478471025824547,
"num_tokens": 11477167.0,
"reward": 0.421875,
"reward_std": 0.49776285886764526,
"rewards/math_answer_reward/mean": 0.421875,
"rewards/math_answer_reward/std": 0.49776285886764526,
"sampling/importance_sampling_ratio/max": 1.5927011966705322,
"sampling/importance_sampling_ratio/mean": 0.2119845151901245,
"sampling/importance_sampling_ratio/min": 2.9625076422234997e-05,
"sampling/sampling_logp_difference/max": 2.392782688140869,
"sampling/sampling_logp_difference/mean": 0.01582634262740612,
"step": 90,
"step_time": 451.58230069698766
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.07624906185083091,
"cast/masked_gap_abs_mean": 0.02778148047218565,
"cast/masked_gap_mean": -0.008772584634471059,
"cast/masked_negative_base_rate": 0.710356798954308,
"cast/masked_positive_base_rate": 0.2896431968547404,
"cast/masked_sign_flipped_rate": 0.2703374261036515,
"cast/masked_teacher_negative_rate": 0.3022001311182976,
"cast/masked_teacher_positive_rate": 0.2735400521196425,
"cast/masked_token_adv_abs_mean": 0.9200334325432777,
"cast/masked_token_adv_max": 1.0278681861236691,
"cast/masked_token_adv_mean": -0.1792027880437672,
"cast/masked_token_adv_min": -1.153278007172048,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.034609479014761746,
"cast/masked_wrong_good_token_rate": 0.19408836518414319,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.00210537462044158,
"clip_ratio/high_mean": 0.00210537462044158,
"clip_ratio/low_mean": 0.0009064154810403124,
"clip_ratio/low_min": 0.0009064154810403124,
"clip_ratio/region_mean": 0.0030117901260382496,
"completions/clipped_ratio": 0.0,
"completions/max_length": 5100.0,
"completions/max_terminated_length": 5100.0,
"completions/mean_length": 1483.875,
"completions/mean_terminated_length": 1483.875,
"completions/min_length": 473.0,
"completions/min_terminated_length": 473.0,
"entropy": 0.25050475797615945,
"epoch": 0.017693952945751507,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.060515180230140686,
"learning_rate": 9.941668286992027e-07,
"loss": 0.004905788227915764,
"num_tokens": 11582183.0,
"reward": 0.34375,
"reward_std": 0.4787135720252991,
"rewards/math_answer_reward/mean": 0.34375,
"rewards/math_answer_reward/std": 0.4787135720252991,
"sampling/importance_sampling_ratio/max": 2.5512518882751465,
"sampling/importance_sampling_ratio/mean": 0.24665144085884094,
"sampling/importance_sampling_ratio/min": 7.396880391752347e-06,
"sampling/sampling_logp_difference/max": 7.937036037445068,
"sampling/sampling_logp_difference/mean": 0.01770530454814434,
"step": 91,
"step_time": 253.5507039950171
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.0299047171138227,
"cast/masked_gap_abs_mean": 0.018860315598431043,
"cast/masked_gap_mean": -0.002511902475134775,
"cast/masked_negative_base_rate": 0.8886058144271374,
"cast/masked_positive_base_rate": 0.11139418696984649,
"cast/masked_sign_flipped_rate": 0.27728556701913476,
"cast/masked_teacher_negative_rate": 0.3145010876469314,
"cast/masked_teacher_positive_rate": 0.27852561697363853,
"cast/masked_token_adv_abs_mean": 0.8439469151198864,
"cast/masked_token_adv_max": 1.0489782318472862,
"cast/masked_token_adv_mean": -0.28254990495042875,
"cast/masked_token_adv_min": -1.1720466315746307,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03112269623670727,
"cast/masked_wrong_good_token_rate": 0.247380850603804,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0019507912074914202,
"clip_ratio/high_mean": 0.0019507912074914202,
"clip_ratio/low_mean": 0.000874479178492038,
"clip_ratio/low_min": 0.000874479178492038,
"clip_ratio/region_mean": 0.002825270383254974,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 4596.0,
"completions/mean_length": 2155.375,
"completions/mean_terminated_length": 2059.5556640625,
"completions/min_length": 604.0,
"completions/min_terminated_length": 604.0,
"entropy": 0.26544094318524003,
"epoch": 0.017888391989111414,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.036917440593242645,
"learning_rate": 9.941020156847494e-07,
"loss": 0.0665392130613327,
"num_tokens": 11731815.0,
"reward": 0.140625,
"reward_std": 0.3503824472427368,
"rewards/math_answer_reward/mean": 0.140625,
"rewards/math_answer_reward/std": 0.3503824472427368,
"sampling/importance_sampling_ratio/max": 2.587423086166382,
"sampling/importance_sampling_ratio/mean": 0.2584162652492523,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 3.4972968101501465,
"sampling/sampling_logp_difference/mean": 0.016141941770911217,
"step": 92,
"step_time": 358.93073036300484
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.06487572670448571,
"cast/masked_gap_abs_mean": 0.022869028442073613,
"cast/masked_gap_mean": -0.004203884293019655,
"cast/masked_negative_base_rate": 0.7363050412386656,
"cast/masked_positive_base_rate": 0.2636949513107538,
"cast/masked_sign_flipped_rate": 0.2843236392363906,
"cast/masked_teacher_negative_rate": 0.3157469085417688,
"cast/masked_teacher_positive_rate": 0.29326131427660584,
"cast/masked_token_adv_abs_mean": 0.8234654720872641,
"cast/masked_token_adv_max": 1.029025984928012,
"cast/masked_token_adv_mean": -0.16409112699329853,
"cast/masked_token_adv_min": -1.1385297663509846,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03415700059849769,
"cast/masked_wrong_good_token_rate": 0.21944791381247342,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.001865464564616559,
"clip_ratio/high_mean": 0.001865464564616559,
"clip_ratio/low_mean": 0.0007974347768140433,
"clip_ratio/low_min": 0.0007974347768140433,
"clip_ratio/region_mean": 0.0026628993455233285,
"completions/clipped_ratio": 0.015625,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 5161.0,
"completions/mean_length": 1733.234375,
"completions/mean_terminated_length": 1630.71435546875,
"completions/min_length": 672.0,
"completions/min_terminated_length": 672.0,
"entropy": 0.25970415212213993,
"epoch": 0.01808283103247132,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.03738826885819435,
"learning_rate": 9.940372026702961e-07,
"loss": 0.02899312973022461,
"num_tokens": 11854326.0,
"reward": 0.328125,
"reward_std": 0.4732423722743988,
"rewards/math_answer_reward/mean": 0.328125,
"rewards/math_answer_reward/std": 0.4732423722743988,
"sampling/importance_sampling_ratio/max": 1.7990796566009521,
"sampling/importance_sampling_ratio/mean": 0.2758469581604004,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.645143985748291,
"sampling/sampling_logp_difference/mean": 0.017507607117295265,
"step": 93,
"step_time": 374.6775522270764
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.09703488508239388,
"cast/masked_gap_abs_mean": 0.027122816041810438,
"cast/masked_gap_mean": -0.005961320091955713,
"cast/masked_negative_base_rate": 0.6694277804344893,
"cast/masked_positive_base_rate": 0.33057222329080105,
"cast/masked_sign_flipped_rate": 0.2707845140248537,
"cast/masked_teacher_negative_rate": 0.3142662616446614,
"cast/masked_teacher_positive_rate": 0.2536355056799948,
"cast/masked_token_adv_abs_mean": 0.7333928802981973,
"cast/masked_token_adv_max": 0.9746227925643325,
"cast/masked_token_adv_mean": -0.13653752719983459,
"cast/masked_token_adv_min": -1.0549261020496488,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.037191618350334466,
"cast/masked_wrong_good_token_rate": 0.17374962894245982,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.002418015312287025,
"clip_ratio/high_mean": 0.002418015312287025,
"clip_ratio/low_mean": 0.0009051786937561701,
"clip_ratio/low_min": 0.0009051786937561701,
"clip_ratio/region_mean": 0.0033231940178666264,
"completions/clipped_ratio": 0.0,
"completions/max_length": 2977.0,
"completions/max_terminated_length": 2977.0,
"completions/mean_length": 1142.46875,
"completions/mean_terminated_length": 1142.46875,
"completions/min_length": 439.0,
"completions/min_terminated_length": 439.0,
"entropy": 0.22979120537638664,
"epoch": 0.01827727007583123,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.035961274057626724,
"learning_rate": 9.939723896558428e-07,
"loss": 0.04122205078601837,
"num_tokens": 11943420.0,
"reward": 0.34375,
"reward_std": 0.4787135720252991,
"rewards/math_answer_reward/mean": 0.34375,
"rewards/math_answer_reward/std": 0.4787135720252991,
"sampling/importance_sampling_ratio/max": 2.3252973556518555,
"sampling/importance_sampling_ratio/mean": 0.22391439974308014,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.46209979057312,
"sampling/sampling_logp_difference/mean": 0.018632803112268448,
"step": 94,
"step_time": 276.2551781919901
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.04739067272748798,
"cast/masked_gap_abs_mean": 0.014457467274041846,
"cast/masked_gap_mean": -0.0021993336779644324,
"cast/masked_negative_base_rate": 0.7589047974906862,
"cast/masked_positive_base_rate": 0.24109520157799125,
"cast/masked_sign_flipped_rate": 0.2404210097156465,
"cast/masked_teacher_negative_rate": 0.27736339066177607,
"cast/masked_teacher_positive_rate": 0.25059274677187204,
"cast/masked_token_adv_abs_mean": 0.9713997188955545,
"cast/masked_token_adv_max": 1.0741031169891357,
"cast/masked_token_adv_mean": -0.24538661845144816,
"cast/masked_token_adv_min": -1.1793322041630745,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.028705874690786004,
"cast/masked_wrong_good_token_rate": 0.19303033757023513,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0019553809015633306,
"clip_ratio/high_mean": 0.0019553809015633306,
"clip_ratio/low_mean": 0.000971936736277712,
"clip_ratio/low_min": 0.000971936736277712,
"clip_ratio/region_mean": 0.0029273176296555903,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4895.0,
"completions/max_terminated_length": 4895.0,
"completions/mean_length": 1726.484375,
"completions/mean_terminated_length": 1726.484375,
"completions/min_length": 275.0,
"completions/min_terminated_length": 275.0,
"entropy": 0.20287742163054645,
"epoch": 0.018471709119191132,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.051848772913217545,
"learning_rate": 9.939075766413895e-07,
"loss": 0.09750928729772568,
"num_tokens": 12065419.0,
"reward": 0.203125,
"reward_std": 0.40550529956817627,
"rewards/math_answer_reward/mean": 0.203125,
"rewards/math_answer_reward/std": 0.40550529956817627,
"sampling/importance_sampling_ratio/max": 2.7641475200653076,
"sampling/importance_sampling_ratio/mean": 0.343007355928421,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.5866260528564453,
"sampling/sampling_logp_difference/mean": 0.014510354027152061,
"step": 95,
"step_time": 230.15794552199077
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.06144008110277355,
"cast/masked_gap_abs_mean": 0.023941663443110883,
"cast/masked_gap_mean": -0.005867482273515634,
"cast/masked_negative_base_rate": 0.7598645752295852,
"cast/masked_positive_base_rate": 0.24013542803004384,
"cast/masked_sign_flipped_rate": 0.24026437802240252,
"cast/masked_teacher_negative_rate": 0.2982555637136102,
"cast/masked_teacher_positive_rate": 0.23409539135172963,
"cast/masked_token_adv_abs_mean": 0.7156187929213047,
"cast/masked_token_adv_max": 0.9309405395761132,
"cast/masked_token_adv_mean": -0.16872837109258398,
"cast/masked_token_adv_min": -1.0264321016147733,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03482074831845239,
"cast/masked_wrong_good_token_rate": 0.1788242975017056,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0016977922769001452,
"clip_ratio/high_mean": 0.0016977922769001452,
"clip_ratio/low_mean": 0.0007691188357057399,
"clip_ratio/low_min": 0.0007691188357057399,
"clip_ratio/region_mean": 0.002466911122610327,
"completions/clipped_ratio": 0.0,
"completions/max_length": 4057.0,
"completions/max_terminated_length": 4057.0,
"completions/mean_length": 1490.25,
"completions/mean_terminated_length": 1490.25,
"completions/min_length": 514.0,
"completions/min_terminated_length": 514.0,
"entropy": 0.20394511800259352,
"epoch": 0.01866614816255104,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.03808789327740669,
"learning_rate": 9.938427636269362e-07,
"loss": 0.05374879390001297,
"num_tokens": 12171715.0,
"reward": 0.265625,
"reward_std": 0.44515693187713623,
"rewards/math_answer_reward/mean": 0.265625,
"rewards/math_answer_reward/std": 0.44515693187713623,
"sampling/importance_sampling_ratio/max": 2.834563732147217,
"sampling/importance_sampling_ratio/mean": 0.2880350351333618,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.5944104194641113,
"sampling/sampling_logp_difference/mean": 0.015960652381181717,
"step": 96,
"step_time": 228.10980090600788
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.06289492751238868,
"cast/masked_gap_abs_mean": 0.023624615772860125,
"cast/masked_gap_mean": -0.005356822563953756,
"cast/masked_negative_base_rate": 0.7210455322638154,
"cast/masked_positive_base_rate": 0.2789544726256281,
"cast/masked_sign_flipped_rate": 0.266345561016351,
"cast/masked_teacher_negative_rate": 0.32170065492391586,
"cast/masked_teacher_positive_rate": 0.2670367108657956,
"cast/masked_token_adv_abs_mean": 0.842876547947526,
"cast/masked_token_adv_max": 1.0085688102990389,
"cast/masked_token_adv_mean": -0.17072315217228606,
"cast/masked_token_adv_min": -1.1205072868615389,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.0326214304077439,
"cast/masked_wrong_good_token_rate": 0.20345063204877079,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0022516692079079803,
"clip_ratio/high_mean": 0.0022516692079079803,
"clip_ratio/low_mean": 0.0008149260615937237,
"clip_ratio/low_min": 0.0008149260615937237,
"clip_ratio/region_mean": 0.0030665952726849355,
"completions/clipped_ratio": 0.0,
"completions/max_length": 6945.0,
"completions/max_terminated_length": 6945.0,
"completions/mean_length": 1702.3125,
"completions/mean_terminated_length": 1702.3125,
"completions/min_length": 262.0,
"completions/min_terminated_length": 262.0,
"entropy": 0.23367834673263133,
"epoch": 0.018860587205910946,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.04988474026322365,
"learning_rate": 9.93777950612483e-07,
"loss": 0.040662772953510284,
"num_tokens": 12292679.0,
"reward": 0.40625,
"reward_std": 0.49501484632492065,
"rewards/math_answer_reward/mean": 0.40625,
"rewards/math_answer_reward/std": 0.49501484632492065,
"sampling/importance_sampling_ratio/max": 2.4588866233825684,
"sampling/importance_sampling_ratio/mean": 0.3378726541996002,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.309671401977539,
"sampling/sampling_logp_difference/mean": 0.017672615125775337,
"step": 97,
"step_time": 354.1849966599257
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.06251373147824779,
"cast/masked_gap_abs_mean": 0.01880450756289065,
"cast/masked_gap_mean": -0.0038095033596619032,
"cast/masked_negative_base_rate": 0.6617059875279665,
"cast/masked_positive_base_rate": 0.3382940050214529,
"cast/masked_sign_flipped_rate": 0.23201879137195647,
"cast/masked_teacher_negative_rate": 0.23387025552801788,
"cast/masked_teacher_positive_rate": 0.24339983356185257,
"cast/masked_token_adv_abs_mean": 0.7689817324280739,
"cast/masked_token_adv_max": 0.9555053869262338,
"cast/masked_token_adv_mean": -0.05404926661776699,
"cast/masked_token_adv_min": -1.017006197012961,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.031576262263115495,
"cast/masked_wrong_good_token_rate": 0.16950506065040827,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0014910192508068576,
"clip_ratio/high_mean": 0.0014910192508068576,
"clip_ratio/low_mean": 0.0008149026007231441,
"clip_ratio/low_min": 0.0008149026007231441,
"clip_ratio/region_mean": 0.0023059218592607067,
"completions/clipped_ratio": 0.078125,
"completions/max_length": 8192.0,
"completions/max_terminated_length": 7837.0,
"completions/mean_length": 2206.578125,
"completions/mean_terminated_length": 1699.3389892578125,
"completions/min_length": 513.0,
"completions/min_terminated_length": 513.0,
"entropy": 0.1912753116339445,
"epoch": 0.019055026249270853,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.05575968325138092,
"learning_rate": 9.937131375980296e-07,
"loss": 0.013404985889792442,
"num_tokens": 12444132.0,
"reward": 0.34375,
"reward_std": 0.4787135720252991,
"rewards/math_answer_reward/mean": 0.34375,
"rewards/math_answer_reward/std": 0.4787135720252991,
"sampling/importance_sampling_ratio/max": 2.8906028270721436,
"sampling/importance_sampling_ratio/mean": 0.4677075445652008,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.271317720413208,
"sampling/sampling_logp_difference/mean": 0.013372254557907581,
"step": 98,
"step_time": 420.09760803499375
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.08942654822021723,
"cast/masked_gap_abs_mean": 0.020104088267544284,
"cast/masked_gap_mean": -0.004382861532747029,
"cast/masked_negative_base_rate": 0.6527670696377754,
"cast/masked_positive_base_rate": 0.347232922911644,
"cast/masked_sign_flipped_rate": 0.25120771769434214,
"cast/masked_teacher_negative_rate": 0.28208068013191223,
"cast/masked_teacher_positive_rate": 0.24447826901450753,
"cast/masked_token_adv_abs_mean": 0.7838798118755221,
"cast/masked_token_adv_max": 0.9604845317080617,
"cast/masked_token_adv_mean": -0.08640847538481466,
"cast/masked_token_adv_min": -1.0548395263031125,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.033055191568564624,
"cast/masked_wrong_good_token_rate": 0.1617811699397862,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.001453101971947035,
"clip_ratio/high_mean": 0.001453101971947035,
"clip_ratio/low_mean": 0.0007677661105844891,
"clip_ratio/low_min": 0.0007677661105844891,
"clip_ratio/region_mean": 0.002220868100266671,
"completions/clipped_ratio": 0.0,
"completions/max_length": 7549.0,
"completions/max_terminated_length": 7549.0,
"completions/mean_length": 2121.65625,
"completions/mean_terminated_length": 2121.65625,
"completions/min_length": 841.0,
"completions/min_terminated_length": 841.0,
"entropy": 0.22135206637904048,
"epoch": 0.01924946529263076,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.04119051247835159,
"learning_rate": 9.936483245835763e-07,
"loss": 0.022711534053087234,
"num_tokens": 12591574.0,
"reward": 0.359375,
"reward_std": 0.4836103618144989,
"rewards/math_answer_reward/mean": 0.359375,
"rewards/math_answer_reward/std": 0.4836103618144989,
"sampling/importance_sampling_ratio/max": 2.126614809036255,
"sampling/importance_sampling_ratio/mean": 0.3352319002151489,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.314540982246399,
"sampling/sampling_logp_difference/mean": 0.015252893790602684,
"step": 99,
"step_time": 345.72251328098355
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.0742945036618039,
"cast/masked_gap_abs_mean": 0.023961322265677154,
"cast/masked_gap_mean": -0.0067796403266271454,
"cast/masked_negative_base_rate": 0.7036592029035091,
"cast/masked_positive_base_rate": 0.29634079057723284,
"cast/masked_sign_flipped_rate": 0.29103928757831454,
"cast/masked_teacher_negative_rate": 0.34329995745792985,
"cast/masked_teacher_positive_rate": 0.3000688459724188,
"cast/masked_token_adv_abs_mean": 0.8981147473677993,
"cast/masked_token_adv_max": 1.0169453416019678,
"cast/masked_token_adv_mean": -0.1828393561881967,
"cast/masked_token_adv_min": -1.1515090521425009,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.03196637466317043,
"cast/masked_wrong_good_token_rate": 0.2167447826359421,
"cast/mixed_group_rate": 0.25,
"clip_ratio/high_max": 0.002694927839911543,
"clip_ratio/high_mean": 0.002694927839911543,
"clip_ratio/low_mean": 0.001234820144418336,
"clip_ratio/low_min": 0.001234820144418336,
"clip_ratio/region_mean": 0.003929748005248257,
"completions/clipped_ratio": 0.0,
"completions/max_length": 2824.0,
"completions/max_terminated_length": 2824.0,
"completions/mean_length": 1376.203125,
"completions/mean_terminated_length": 1376.203125,
"completions/min_length": 451.0,
"completions/min_terminated_length": 451.0,
"entropy": 0.2637432408519089,
"epoch": 0.019443904335990667,
"frac_reward_zero_std": 0.75,
"grad_norm": 0.054625071585178375,
"learning_rate": 9.93583511569123e-07,
"loss": 0.02553533762693405,
"num_tokens": 12691043.0,
"reward": 0.34375,
"reward_std": 0.4787135720252991,
"rewards/math_answer_reward/mean": 0.34375,
"rewards/math_answer_reward/std": 0.4787135720252991,
"sampling/importance_sampling_ratio/max": 2.372849941253662,
"sampling/importance_sampling_ratio/mean": 0.2609149217605591,
"sampling/importance_sampling_ratio/min": 0.00012522871838882565,
"sampling/sampling_logp_difference/max": 1.677259922027588,
"sampling/sampling_logp_difference/mean": 0.02041148766875267,
"step": 100,
"step_time": 153.14356127602514
}
],
"logging_steps": 1,
"max_steps": 15429,
"num_input_tokens_seen": 12691043,
"num_train_epochs": 3,
"save_steps": 100,
"stateful_callbacks": {
"TrainerControl": {
"args": {
"should_epoch_stop": false,
"should_evaluate": false,
"should_log": false,
"should_save": true,
"should_training_stop": false
},
"attributes": {}
}
},
"total_flos": 0.0,
"train_batch_size": 2,
"trial_name": null,
"trial_params": null
}