CAST-QWEN-3 / trainer_state.json
tejeshbhalla's picture
Upload CAST Qwen LoRA checkpoint-100
c02aa4a 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": 0.07540167623665184,
"cast/masked_gap_abs_mean": 0.0003348859263496706,
"cast/masked_gap_mean": 5.701296323934457e-06,
"cast/masked_negative_base_rate": 0.7414380516856909,
"cast/masked_positive_base_rate": 0.2585619401652366,
"cast/masked_sign_flipped_rate": 0.27001334028318524,
"cast/masked_teacher_negative_rate": 0.32948060473427176,
"cast/masked_teacher_positive_rate": 0.2492526974529028,
"cast/masked_token_adv_abs_mean": 0.8956050714477897,
"cast/masked_token_adv_max": 1.0073942635208368,
"cast/masked_token_adv_mean": -0.27101879054680467,
"cast/masked_token_adv_min": -1.006590973585844,
"cast/masked_token_weight_max": 1.0197022557258606,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.0011940337462874595,
"cast/masked_wrong_good_token_rate": 0.19461166393011808,
"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": 0.07541441917419434,
"learning_rate": 1e-06,
"loss": 0.18309621512889862,
"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": 138.9456026760745
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.04320417204871774,
"cast/masked_gap_abs_mean": 0.009557685581967235,
"cast/masked_gap_mean": 3.164942400246673e-05,
"cast/masked_negative_base_rate": 0.8672169931232929,
"cast/masked_positive_base_rate": 0.13278300035744905,
"cast/masked_sign_flipped_rate": 0.2933891713619232,
"cast/masked_teacher_negative_rate": 0.3571503348648548,
"cast/masked_teacher_positive_rate": 0.2830226211808622,
"cast/masked_token_adv_abs_mean": 0.8587681213393807,
"cast/masked_token_adv_max": 1.0409232024103403,
"cast/masked_token_adv_mean": -0.2269591999356635,
"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": 0.02249191445298493,
"cast/masked_wrong_good_token_rate": 0.25018499861471355,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0026310299235774437,
"clip_ratio/high_mean": 0.0026310299235774437,
"clip_ratio/low_mean": 0.0006287215405791358,
"clip_ratio/low_min": 0.0006287215405791358,
"clip_ratio/region_mean": 0.003259751476434758,
"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": 0.2762147618923336,
"epoch": 0.00038887808671981335,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.057483572512865067,
"learning_rate": 9.999351869855467e-07,
"loss": 0.1041565090417862,
"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": 2.8648200035095215,
"sampling/importance_sampling_ratio/mean": 0.5402711629867554,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.5866284370422363,
"sampling/sampling_logp_difference/mean": 0.00981808826327324,
"step": 2,
"step_time": 220.45476261503063
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.08295917022041976,
"cast/masked_gap_abs_mean": 0.008588890108512715,
"cast/masked_gap_mean": 0.00019422965630155886,
"cast/masked_negative_base_rate": 0.7616477459669113,
"cast/masked_positive_base_rate": 0.23835225403308868,
"cast/masked_sign_flipped_rate": 0.27720851078629494,
"cast/masked_teacher_negative_rate": 0.3351007313467562,
"cast/masked_teacher_positive_rate": 0.2553321295417845,
"cast/masked_token_adv_abs_mean": 0.8312330543994904,
"cast/masked_token_adv_max": 0.9871918242424726,
"cast/masked_token_adv_mean": -0.2791212077718228,
"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": 0.02127950987778604,
"cast/masked_wrong_good_token_rate": 0.1942493417300284,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0024300671775563387,
"clip_ratio/high_mean": 0.0024300671775563387,
"clip_ratio/low_mean": 0.0006943302632862469,
"clip_ratio/low_min": 0.0006943302632862469,
"clip_ratio/region_mean": 0.003124397411738755,
"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": 0.24014057544991374,
"epoch": 0.00058331713007972,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.05841658264398575,
"learning_rate": 9.998703739710934e-07,
"loss": 0.14965572953224182,
"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": 1.9895622730255127,
"sampling/importance_sampling_ratio/mean": 0.5416615009307861,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0082597732543945,
"sampling/sampling_logp_difference/mean": 0.0084874602034688,
"step": 3,
"step_time": 132.9880677710753
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.034150728839449584,
"cast/masked_gap_abs_mean": 0.00868575117783621,
"cast/masked_gap_mean": -3.672780536589926e-05,
"cast/masked_negative_base_rate": 0.8826569858938456,
"cast/masked_positive_base_rate": 0.11734301201067865,
"cast/masked_sign_flipped_rate": 0.2842681407928467,
"cast/masked_teacher_negative_rate": 0.34180234791710973,
"cast/masked_teacher_positive_rate": 0.2725201374851167,
"cast/masked_token_adv_abs_mean": 0.9091893546283245,
"cast/masked_token_adv_max": 1.0598125457763672,
"cast/masked_token_adv_mean": -0.29836036753840744,
"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": 0.021681181620806456,
"cast/masked_wrong_good_token_rate": 0.25011741113848984,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.002037963009570376,
"clip_ratio/high_mean": 0.002037963009570376,
"clip_ratio/low_mean": 0.000817749402813206,
"clip_ratio/low_min": 0.000817749402813206,
"clip_ratio/region_mean": 0.0028557124132930767,
"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": 0.2552597299218178,
"epoch": 0.0007777561734396267,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.048012737184762955,
"learning_rate": 9.9980556095664e-07,
"loss": 0.12857627868652344,
"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": 2.6153626441955566,
"sampling/importance_sampling_ratio/mean": 0.48284316062927246,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.387291431427002,
"sampling/sampling_logp_difference/mean": 0.009252705611288548,
"step": 4,
"step_time": 154.95612737594638
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.07174720999319106,
"cast/masked_gap_abs_mean": 0.00832445762353018,
"cast/masked_gap_mean": 0.00016178371151909232,
"cast/masked_negative_base_rate": 0.7582155726850033,
"cast/masked_positive_base_rate": 0.24178441986441612,
"cast/masked_sign_flipped_rate": 0.2778616673313081,
"cast/masked_teacher_negative_rate": 0.33694845251739025,
"cast/masked_teacher_positive_rate": 0.2619617539457977,
"cast/masked_token_adv_abs_mean": 0.803058204241097,
"cast/masked_token_adv_max": 1.0327837513759732,
"cast/masked_token_adv_mean": -0.20869818422943354,
"cast/masked_token_adv_min": -1.135774197988212,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02149996702792123,
"cast/masked_wrong_good_token_rate": 0.20611445675604045,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0023472383054468082,
"clip_ratio/high_mean": 0.0023472383054468082,
"clip_ratio/low_mean": 0.0008614326911811077,
"clip_ratio/low_min": 0.0008614326911811077,
"clip_ratio/region_mean": 0.0032086710270959884,
"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": 0.2260425193235278,
"epoch": 0.0009721952167995333,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.04281548410654068,
"learning_rate": 9.997407479421868e-07,
"loss": 0.08196026086807251,
"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": 2.8635706901550293,
"sampling/importance_sampling_ratio/mean": 0.5361987352371216,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.5466578006744385,
"sampling/sampling_logp_difference/mean": 0.008517158217728138,
"step": 5,
"step_time": 263.3327056969574
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.09845344698987901,
"cast/masked_gap_abs_mean": 0.0077184541587485,
"cast/masked_gap_mean": 9.91054717900397e-05,
"cast/masked_negative_base_rate": 0.7048297440633178,
"cast/masked_positive_base_rate": 0.2951702519785613,
"cast/masked_sign_flipped_rate": 0.28530021803453565,
"cast/masked_teacher_negative_rate": 0.35024058632552624,
"cast/masked_teacher_positive_rate": 0.258518323302269,
"cast/masked_token_adv_abs_mean": 0.8146070325747132,
"cast/masked_token_adv_max": 0.9936644462868571,
"cast/masked_token_adv_mean": -0.16159620229154825,
"cast/masked_token_adv_min": -1.0911949509754777,
"cast/masked_token_weight_max": 1.1997489035129547,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.020682332629803568,
"cast/masked_wrong_good_token_rate": 0.18684677162673324,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.00227742045535706,
"clip_ratio/high_mean": 0.00227742045535706,
"clip_ratio/low_mean": 0.0006559463158737344,
"clip_ratio/low_min": 0.0006559463158737344,
"clip_ratio/region_mean": 0.002933366795332404,
"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": 0.2123508588410914,
"epoch": 0.00116663426015944,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.029355745762586594,
"learning_rate": 9.996759349277335e-07,
"loss": 0.02806932106614113,
"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": 2.444018602371216,
"sampling/importance_sampling_ratio/mean": 0.4596092104911804,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.3499720096588135,
"sampling/sampling_logp_difference/mean": 0.007857461459934711,
"step": 6,
"step_time": 259.13701639306964
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.07273956789867952,
"cast/masked_gap_abs_mean": 0.00741772614128422,
"cast/masked_gap_mean": 4.3205557560099805e-05,
"cast/masked_negative_base_rate": 0.7656119633466005,
"cast/masked_positive_base_rate": 0.23438803129829466,
"cast/masked_sign_flipped_rate": 0.24381298944354057,
"cast/masked_teacher_negative_rate": 0.3232969786040485,
"cast/masked_teacher_positive_rate": 0.2299480144865811,
"cast/masked_token_adv_abs_mean": 0.8646273259073496,
"cast/masked_token_adv_max": 1.0583419147878885,
"cast/masked_token_adv_mean": -0.31560405157506466,
"cast/masked_token_adv_min": -1.1782502066344023,
"cast/masked_token_weight_max": 1.1991918422281742,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.020403055474162102,
"cast/masked_wrong_good_token_rate": 0.17107342206873,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.002125532879290404,
"clip_ratio/high_mean": 0.002125532879290404,
"clip_ratio/low_mean": 0.0008073912922554882,
"clip_ratio/low_min": 0.0008073912922554882,
"clip_ratio/region_mean": 0.002932924158812966,
"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": 0.21040799980983138,
"epoch": 0.0013610733035193468,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.04686853662133217,
"learning_rate": 9.996111219132802e-07,
"loss": 0.1770925521850586,
"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": 1.9720180034637451,
"sampling/importance_sampling_ratio/mean": 0.5832568407058716,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.050135374069214,
"sampling/sampling_logp_difference/mean": 0.007881351746618748,
"step": 7,
"step_time": 143.1699405350082
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.1120078347157687,
"cast/masked_gap_abs_mean": 0.008213697219616733,
"cast/masked_gap_mean": 7.563315953973415e-05,
"cast/masked_negative_base_rate": 0.676128753926605,
"cast/masked_positive_base_rate": 0.323871242813766,
"cast/masked_sign_flipped_rate": 0.29895873880013824,
"cast/masked_teacher_negative_rate": 0.35005757585167885,
"cast/masked_teacher_positive_rate": 0.2624955400824547,
"cast/masked_token_adv_abs_mean": 0.8079905230551958,
"cast/masked_token_adv_max": 1.0228791739791632,
"cast/masked_token_adv_mean": -0.15664584538899362,
"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": 0.021721332159359008,
"cast/masked_wrong_good_token_rate": 0.18695090292021632,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0026907847422990017,
"clip_ratio/high_mean": 0.0026907847422990017,
"clip_ratio/low_mean": 0.0010963139593513915,
"clip_ratio/low_min": 0.0010963139593513915,
"clip_ratio/region_mean": 0.003787098674365552,
"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": 0.20389539701864123,
"epoch": 0.0015555123468792534,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.0596182681620121,
"learning_rate": 9.995463088988269e-07,
"loss": 0.10598780959844589,
"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": 2.5764153003692627,
"sampling/importance_sampling_ratio/mean": 0.5870912671089172,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.727336049079895,
"sampling/sampling_logp_difference/mean": 0.009018292650580406,
"step": 8,
"step_time": 145.51691963203484
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.09175741765648127,
"cast/masked_gap_abs_mean": 0.009720383226522245,
"cast/masked_gap_mean": -0.0001156709182623672,
"cast/masked_negative_base_rate": 0.765513863414526,
"cast/masked_positive_base_rate": 0.2344861375167966,
"cast/masked_sign_flipped_rate": 0.30448326049372554,
"cast/masked_teacher_negative_rate": 0.38946089521050453,
"cast/masked_teacher_positive_rate": 0.28391025867313147,
"cast/masked_token_adv_abs_mean": 0.8433413524180651,
"cast/masked_token_adv_max": 1.0465021124109626,
"cast/masked_token_adv_mean": -0.29148678621277213,
"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": 0.02285274548921734,
"cast/masked_wrong_good_token_rate": 0.21272584213875234,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0023394731724692974,
"clip_ratio/high_mean": 0.0023394731724692974,
"clip_ratio/low_mean": 0.0007679452537558973,
"clip_ratio/low_min": 0.0007679452537558973,
"clip_ratio/region_mean": 0.003107418437139131,
"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": 0.2860231031663716,
"epoch": 0.00174995139023916,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.05236915871500969,
"learning_rate": 9.994814958843736e-07,
"loss": 0.18071112036705017,
"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": 2.3627138137817383,
"sampling/importance_sampling_ratio/mean": 0.43615949153900146,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1105554103851318,
"sampling/sampling_logp_difference/mean": 0.010174846276640892,
"step": 9,
"step_time": 130.84194018508424
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.0,
"cast/masked_correct_bad_token_rate": 0.1404325331095606,
"cast/masked_gap_abs_mean": 0.00899360518087633,
"cast/masked_gap_mean": -9.423056951618491e-06,
"cast/masked_negative_base_rate": 0.6098946072161198,
"cast/masked_positive_base_rate": 0.3901053888257593,
"cast/masked_sign_flipped_rate": 0.3107297644019127,
"cast/masked_teacher_negative_rate": 0.3742868402041495,
"cast/masked_teacher_positive_rate": 0.26614309288561344,
"cast/masked_token_adv_abs_mean": 0.691519851796329,
"cast/masked_token_adv_max": 0.9002567371353507,
"cast/masked_token_adv_mean": -0.10459913937665988,
"cast/masked_token_adv_min": -0.9644434070214629,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022326094564050436,
"cast/masked_wrong_good_token_rate": 0.17029723129235208,
"cast/mixed_group_rate": 1.0,
"clip_ratio/high_max": 0.0025474535032117274,
"clip_ratio/high_mean": 0.0025474535032117274,
"clip_ratio/low_mean": 0.0009466308365517762,
"clip_ratio/low_min": 0.0009466308365517762,
"clip_ratio/region_mean": 0.0034940843179356307,
"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": 0.23711516382172704,
"epoch": 0.0019443904335990666,
"frac_reward_zero_std": 0.0,
"grad_norm": 0.06807835400104523,
"learning_rate": 9.994166828699203e-07,
"loss": -0.0006617612671107054,
"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": 2.6711480617523193,
"sampling/importance_sampling_ratio/mean": 0.6817431449890137,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8482625484466553,
"sampling/sampling_logp_difference/mean": 0.009459184482693672,
"step": 10,
"step_time": 148.75796850799816
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.020839939708821476,
"cast/masked_gap_abs_mean": 0.008739019140193705,
"cast/masked_gap_mean": -5.523264837847819e-05,
"cast/masked_negative_base_rate": 0.9370690686628222,
"cast/masked_positive_base_rate": 0.0629309262149036,
"cast/masked_sign_flipped_rate": 0.2684418170247227,
"cast/masked_teacher_negative_rate": 0.3589129359461367,
"cast/masked_teacher_positive_rate": 0.2629332628566772,
"cast/masked_token_adv_abs_mean": 0.8140257280319929,
"cast/masked_token_adv_max": 0.9751813746988773,
"cast/masked_token_adv_mean": -0.3250087253691163,
"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": 0.021891983546083793,
"cast/masked_wrong_good_token_rate": 0.2476018767338246,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0024328225590579677,
"clip_ratio/high_mean": 0.0024328225590579677,
"clip_ratio/low_mean": 0.0010233560078631854,
"clip_ratio/low_min": 0.0010233560078631854,
"clip_ratio/region_mean": 0.003456178539636312,
"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": 0.24845839920453727,
"epoch": 0.0021388294769589733,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.06315355747938156,
"learning_rate": 9.99351869855467e-07,
"loss": 0.18084074556827545,
"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": 2.8548667430877686,
"sampling/importance_sampling_ratio/mean": 0.6070733666419983,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0931932926177979,
"sampling/sampling_logp_difference/mean": 0.008714194409549236,
"step": 11,
"step_time": 268.88710329300375
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.08777704817475751,
"cast/masked_gap_abs_mean": 0.007600692770211026,
"cast/masked_gap_mean": -0.00017287639974483682,
"cast/masked_negative_base_rate": 0.7294434225186706,
"cast/masked_positive_base_rate": 0.2705565735232085,
"cast/masked_sign_flipped_rate": 0.25596981309354305,
"cast/masked_teacher_negative_rate": 0.3259299984201789,
"cast/masked_teacher_positive_rate": 0.23343352461233735,
"cast/masked_token_adv_abs_mean": 0.783683268353343,
"cast/masked_token_adv_max": 1.0076016513630748,
"cast/masked_token_adv_mean": -0.19164223867846886,
"cast/masked_token_adv_min": -1.1024696184322238,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.020955989661160856,
"cast/masked_wrong_good_token_rate": 0.16819276474416256,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.002245958094135858,
"clip_ratio/high_mean": 0.002245958094135858,
"clip_ratio/low_mean": 0.0009274157137042494,
"clip_ratio/low_min": 0.0009274157137042494,
"clip_ratio/region_mean": 0.003173373799654655,
"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": 0.20406533288769424,
"epoch": 0.00233326852031888,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.048124801367521286,
"learning_rate": 9.992870568410137e-07,
"loss": 0.0334460474550724,
"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": 2.880530834197998,
"sampling/importance_sampling_ratio/mean": 0.5121333599090576,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.107926368713379,
"sampling/sampling_logp_difference/mean": 0.007823399268090725,
"step": 12,
"step_time": 202.7237523449876
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.06673253793269396,
"cast/masked_gap_abs_mean": 0.0095521788316546,
"cast/masked_gap_mean": 2.6094166400980612e-05,
"cast/masked_negative_base_rate": 0.8071926310658455,
"cast/masked_positive_base_rate": 0.19280736334621906,
"cast/masked_sign_flipped_rate": 0.3006064426153898,
"cast/masked_teacher_negative_rate": 0.3684295825660229,
"cast/masked_teacher_positive_rate": 0.2836040146648884,
"cast/masked_token_adv_abs_mean": 0.9594157431274652,
"cast/masked_token_adv_max": 1.0608110204339027,
"cast/masked_token_adv_mean": -0.2669977208133787,
"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": 0.02307386539177969,
"cast/masked_wrong_good_token_rate": 0.2338739037513733,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0026376334117230726,
"clip_ratio/high_mean": 0.0026376334117230726,
"clip_ratio/low_mean": 0.0009560298649375909,
"clip_ratio/low_min": 0.0009560298649375909,
"clip_ratio/region_mean": 0.003593663299398031,
"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": 0.2620095561724156,
"epoch": 0.0025277075636787865,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.09240326285362244,
"learning_rate": 9.992222438265604e-07,
"loss": 0.1675257831811905,
"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": 2.744572162628174,
"sampling/importance_sampling_ratio/mean": 0.6999609470367432,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.4346261024475098,
"sampling/sampling_logp_difference/mean": 0.010185331106185913,
"step": 13,
"step_time": 137.2018680109759
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.09497179719619453,
"cast/masked_gap_abs_mean": 0.009602549660485238,
"cast/masked_gap_mean": 4.1113443899121194e-05,
"cast/masked_negative_base_rate": 0.7318982388824224,
"cast/masked_positive_base_rate": 0.2681017564609647,
"cast/masked_sign_flipped_rate": 0.2886825115419924,
"cast/masked_teacher_negative_rate": 0.3557600877247751,
"cast/masked_teacher_positive_rate": 0.26532260049134493,
"cast/masked_token_adv_abs_mean": 0.7977913692593575,
"cast/masked_token_adv_max": 0.9614516906440258,
"cast/masked_token_adv_mean": -0.22434303152840585,
"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": 0.02322837890824303,
"cast/masked_wrong_good_token_rate": 0.1937107127159834,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.003281567227531923,
"clip_ratio/high_mean": 0.003281567227531923,
"clip_ratio/low_mean": 0.0010501039887458319,
"clip_ratio/low_min": 0.0010501039887458319,
"clip_ratio/region_mean": 0.004331671178078977,
"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": 0.24176476034335792,
"epoch": 0.0027221466070386936,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.07015682011842728,
"learning_rate": 9.99157430812107e-07,
"loss": 0.13378775119781494,
"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": 2.804028034210205,
"sampling/importance_sampling_ratio/mean": 0.6044740676879883,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.852097749710083,
"sampling/sampling_logp_difference/mean": 0.009744569659233093,
"step": 14,
"step_time": 208.76064723401214
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.04178183525800705,
"cast/masked_gap_abs_mean": 0.010153447525226511,
"cast/masked_gap_mean": -7.594389262521872e-05,
"cast/masked_negative_base_rate": 0.8869069404900074,
"cast/masked_positive_base_rate": 0.11309304926544428,
"cast/masked_sign_flipped_rate": 0.3074562381953001,
"cast/masked_teacher_negative_rate": 0.3817735342308879,
"cast/masked_teacher_positive_rate": 0.3013722589239478,
"cast/masked_token_adv_abs_mean": 0.9071930330246687,
"cast/masked_token_adv_max": 1.0237302472814918,
"cast/masked_token_adv_mean": -0.2636474344290036,
"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": 0.023323369619902223,
"cast/masked_wrong_good_token_rate": 0.26567440410144627,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0025091055294979014,
"clip_ratio/high_mean": 0.0025091055294979014,
"clip_ratio/low_mean": 0.0010040268316515721,
"clip_ratio/low_min": 0.0010040268316515721,
"clip_ratio/region_mean": 0.0035131323820678517,
"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": 0.3067493666894734,
"epoch": 0.0029165856503986,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.06742899119853973,
"learning_rate": 9.990926177976538e-07,
"loss": 0.11106016486883163,
"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": 2.839901924133301,
"sampling/importance_sampling_ratio/mean": 0.5847718715667725,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7030641436576843,
"sampling/sampling_logp_difference/mean": 0.010903000831604004,
"step": 15,
"step_time": 141.06610155795352
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.08012342534493655,
"cast/masked_gap_abs_mean": 0.007652370317373425,
"cast/masked_gap_mean": 4.0899534823779504e-05,
"cast/masked_negative_base_rate": 0.7396341543644667,
"cast/masked_positive_base_rate": 0.26036584423854947,
"cast/masked_sign_flipped_rate": 0.25769047951325774,
"cast/masked_teacher_negative_rate": 0.32991626393049955,
"cast/masked_teacher_positive_rate": 0.23241939302533865,
"cast/masked_token_adv_abs_mean": 0.7743849288672209,
"cast/masked_token_adv_max": 0.9771279226988554,
"cast/masked_token_adv_mean": -0.19062347065846552,
"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": 0.020722884917631745,
"cast/masked_wrong_good_token_rate": 0.17756705358624458,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0023868031521487865,
"clip_ratio/high_mean": 0.0023868031521487865,
"clip_ratio/low_mean": 0.000907718198504881,
"clip_ratio/low_min": 0.000907718198504881,
"clip_ratio/region_mean": 0.0032945213388302363,
"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": 0.21527321520261467,
"epoch": 0.003111024693758507,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.05322488024830818,
"learning_rate": 9.990278047832003e-07,
"loss": 0.04187319800257683,
"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": 2.6855854988098145,
"sampling/importance_sampling_ratio/mean": 0.4917927384376526,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.392490863800049,
"sampling/sampling_logp_difference/mean": 0.00840782094746828,
"step": 16,
"step_time": 152.4657335340744
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.07932704500854015,
"cast/masked_gap_abs_mean": 0.009863192943157628,
"cast/masked_gap_mean": -3.1241944355997475e-06,
"cast/masked_negative_base_rate": 0.7783364430069923,
"cast/masked_positive_base_rate": 0.2216635448858142,
"cast/masked_sign_flipped_rate": 0.3084926069714129,
"cast/masked_teacher_negative_rate": 0.371010048314929,
"cast/masked_teacher_positive_rate": 0.28728161146864295,
"cast/masked_token_adv_abs_mean": 0.7994884252548218,
"cast/masked_token_adv_max": 0.9985594702884555,
"cast/masked_token_adv_mean": -0.17401909017644357,
"cast/masked_token_adv_min": -1.0923998234793544,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.023400130798108876,
"cast/masked_wrong_good_token_rate": 0.22916556056588888,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0032399507326772436,
"clip_ratio/high_mean": 0.0032399507326772436,
"clip_ratio/low_mean": 0.001099741066354909,
"clip_ratio/low_min": 0.001099741066354909,
"clip_ratio/region_mean": 0.004339691789937206,
"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": 0.2560529443435371,
"epoch": 0.0033054637371184134,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.079947330057621,
"learning_rate": 9.989629917687472e-07,
"loss": 0.09603219479322433,
"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": 2.8774523735046387,
"sampling/importance_sampling_ratio/mean": 0.578902542591095,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.3328344821929932,
"sampling/sampling_logp_difference/mean": 0.010229099541902542,
"step": 17,
"step_time": 258.8720124180545
},
{
"cast/all_correct_group_rate": 0.25,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.12071756337536499,
"cast/masked_gap_abs_mean": 0.008904059592168778,
"cast/masked_gap_mean": -0.00013484146961673105,
"cast/masked_negative_base_rate": 0.6390319131314754,
"cast/masked_positive_base_rate": 0.36096808570437133,
"cast/masked_sign_flipped_rate": 0.3008006578311324,
"cast/masked_teacher_negative_rate": 0.3643172774463892,
"cast/masked_teacher_positive_rate": 0.2653203080408275,
"cast/masked_token_adv_abs_mean": 0.8837857190519571,
"cast/masked_token_adv_max": 1.034641768783331,
"cast/masked_token_adv_mean": -0.08514962811022997,
"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": 0.022618042246904224,
"cast/masked_wrong_good_token_rate": 0.18008309474680573,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.002968822453112807,
"clip_ratio/high_mean": 0.002968822453112807,
"clip_ratio/low_mean": 0.0010269033909935388,
"clip_ratio/low_min": 0.0010269033909935388,
"clip_ratio/region_mean": 0.003995725823187968,
"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": 0.24180993880145252,
"epoch": 0.00349990278047832,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.0698453038930893,
"learning_rate": 9.988981787542939e-07,
"loss": 0.018959034234285355,
"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": 2.7645645141601562,
"sampling/importance_sampling_ratio/mean": 0.6989774703979492,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9544506072998047,
"sampling/sampling_logp_difference/mean": 0.009967838414013386,
"step": 18,
"step_time": 137.3097658501065
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.1387669900432229,
"cast/masked_gap_abs_mean": 0.009982270348700695,
"cast/masked_gap_mean": 0.0001262845229348386,
"cast/masked_negative_base_rate": 0.6128476276062429,
"cast/masked_positive_base_rate": 0.3871523691341281,
"cast/masked_sign_flipped_rate": 0.32565534161403775,
"cast/masked_teacher_negative_rate": 0.3623216338455677,
"cast/masked_teacher_positive_rate": 0.29738293308764696,
"cast/masked_token_adv_abs_mean": 0.8666639495640993,
"cast/masked_token_adv_max": 1.008711902424693,
"cast/masked_token_adv_mean": -0.13876563915982842,
"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": 0.024220241408329457,
"cast/masked_wrong_good_token_rate": 0.18688835389912128,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0031790565353730926,
"clip_ratio/high_mean": 0.0031790565353730926,
"clip_ratio/low_mean": 0.001251715378202789,
"clip_ratio/low_min": 0.001251715378202789,
"clip_ratio/region_mean": 0.0044307718926575035,
"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": 0.24176941346377134,
"epoch": 0.0036943418238382267,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.07012743502855301,
"learning_rate": 9.988333657398406e-07,
"loss": 0.07597429305315018,
"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": 2.6462721824645996,
"sampling/importance_sampling_ratio/mean": 0.5790227651596069,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.5455565452575684,
"sampling/sampling_logp_difference/mean": 0.010294925421476364,
"step": 19,
"step_time": 175.419133702002
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.04049463733099401,
"cast/masked_gap_abs_mean": 0.007777481194352731,
"cast/masked_gap_mean": -2.163747097849722e-05,
"cast/masked_negative_base_rate": 0.88489880412817,
"cast/masked_positive_base_rate": 0.11510118655860424,
"cast/masked_sign_flipped_rate": 0.24733365047723055,
"cast/masked_teacher_negative_rate": 0.33732436830177903,
"cast/masked_teacher_positive_rate": 0.234545414801687,
"cast/masked_token_adv_abs_mean": 0.8616737853735685,
"cast/masked_token_adv_max": 1.0217759367078543,
"cast/masked_token_adv_mean": -0.35416852321759507,
"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": 0.020690129575086758,
"cast/masked_wrong_good_token_rate": 0.20683901174925268,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0019666152802528813,
"clip_ratio/high_mean": 0.0019666152802528813,
"clip_ratio/low_mean": 0.0009174044362225686,
"clip_ratio/low_min": 0.0009174044362225686,
"clip_ratio/region_mean": 0.00288401970101404,
"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": 0.21517048333771527,
"epoch": 0.0038887808671981333,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.05178552865982056,
"learning_rate": 9.98768552725387e-07,
"loss": 0.2131439596414566,
"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": 2.557992458343506,
"sampling/importance_sampling_ratio/mean": 0.5836626887321472,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.6569418907165527,
"sampling/sampling_logp_difference/mean": 0.00793016143143177,
"step": 20,
"step_time": 166.78154101598193
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.08413377136457711,
"cast/masked_gap_abs_mean": 0.009144431649474427,
"cast/masked_gap_mean": 7.569711240051902e-05,
"cast/masked_negative_base_rate": 0.7460893653333187,
"cast/masked_positive_base_rate": 0.25391063420102,
"cast/masked_sign_flipped_rate": 0.2821589158847928,
"cast/masked_teacher_negative_rate": 0.3456883323378861,
"cast/masked_teacher_positive_rate": 0.26328433072194457,
"cast/masked_token_adv_abs_mean": 0.7481773504987359,
"cast/masked_token_adv_max": 0.9924054453149438,
"cast/masked_token_adv_mean": -0.11834585507313022,
"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": 0.022558800090337172,
"cast/masked_wrong_good_token_rate": 0.1980251451022923,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0028771635461453116,
"clip_ratio/high_mean": 0.0028771635461453116,
"clip_ratio/low_mean": 0.0010889112081713392,
"clip_ratio/low_min": 0.0010889112081713392,
"clip_ratio/region_mean": 0.003966074766140082,
"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": 0.25360747473314404,
"epoch": 0.00408321991055804,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.03531507030129433,
"learning_rate": 9.98703739710934e-07,
"loss": 0.053934887051582336,
"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": 2.2331087589263916,
"sampling/importance_sampling_ratio/mean": 0.49800044298171997,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7942287921905518,
"sampling/sampling_logp_difference/mean": 0.009310731664299965,
"step": 21,
"step_time": 266.4142180309864
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.11139943904709071,
"cast/masked_gap_abs_mean": 0.008074681034486275,
"cast/masked_gap_mean": -2.2128803863097346e-05,
"cast/masked_negative_base_rate": 0.6078154062852263,
"cast/masked_positive_base_rate": 0.39218459837138653,
"cast/masked_sign_flipped_rate": 0.26587387593463063,
"cast/masked_teacher_negative_rate": 0.3155854516662657,
"cast/masked_teacher_positive_rate": 0.2292364665772766,
"cast/masked_token_adv_abs_mean": 0.8446133816614747,
"cast/masked_token_adv_max": 1.0434970706701279,
"cast/masked_token_adv_mean": -0.08807034817436943,
"cast/masked_token_adv_min": -1.126993490383029,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.021578985935775563,
"cast/masked_wrong_good_token_rate": 0.15447443793527782,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0027351849694241537,
"clip_ratio/high_mean": 0.0027351849694241537,
"clip_ratio/low_mean": 0.000981997225608211,
"clip_ratio/low_min": 0.000981997225608211,
"clip_ratio/region_mean": 0.003717182164109545,
"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": 0.2039946608711034,
"epoch": 0.0042776589539179465,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.0894847884774208,
"learning_rate": 9.986389266964807e-07,
"loss": 0.04481421411037445,
"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": 2.8325042724609375,
"sampling/importance_sampling_ratio/mean": 0.7766112089157104,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1795439720153809,
"sampling/sampling_logp_difference/mean": 0.009053087793290615,
"step": 22,
"step_time": 132.43258942797547
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.07550492242444307,
"cast/masked_gap_abs_mean": 0.0114825579803437,
"cast/masked_gap_mean": -2.25851349568984e-05,
"cast/masked_negative_base_rate": 0.775532076600939,
"cast/masked_positive_base_rate": 0.22446792153641582,
"cast/masked_sign_flipped_rate": 0.32797486474737525,
"cast/masked_teacher_negative_rate": 0.3859280599281192,
"cast/masked_teacher_positive_rate": 0.3085369518958032,
"cast/masked_token_adv_abs_mean": 0.9723059516400099,
"cast/masked_token_adv_max": 1.071750096976757,
"cast/masked_token_adv_mean": -0.19849501688440796,
"cast/masked_token_adv_min": -1.1899301037192345,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02475263416999951,
"cast/masked_wrong_good_token_rate": 0.25246994104236364,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0034776003667502664,
"clip_ratio/high_mean": 0.0034776003667502664,
"clip_ratio/low_mean": 0.0011503628611535532,
"clip_ratio/low_min": 0.0011503628611535532,
"clip_ratio/region_mean": 0.004627963251550682,
"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": 0.3083141460083425,
"epoch": 0.004472097997277854,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.11863382905721664,
"learning_rate": 9.985741136820272e-07,
"loss": 0.10658173263072968,
"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": 2.792290210723877,
"sampling/importance_sampling_ratio/mean": 0.5663627982139587,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7320728302001953,
"sampling/sampling_logp_difference/mean": 0.012518594041466713,
"step": 23,
"step_time": 92.5146557810076
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.10811224370263517,
"cast/masked_gap_abs_mean": 0.009166414645733312,
"cast/masked_gap_mean": 1.7740516966568975e-05,
"cast/masked_negative_base_rate": 0.6535243364050984,
"cast/masked_positive_base_rate": 0.3464756584726274,
"cast/masked_sign_flipped_rate": 0.30607651406899095,
"cast/masked_teacher_negative_rate": 0.35560536105185747,
"cast/masked_teacher_positive_rate": 0.2767373351380229,
"cast/masked_token_adv_abs_mean": 0.8234735429286957,
"cast/masked_token_adv_max": 0.9499526340514421,
"cast/masked_token_adv_mean": -0.13675118240644224,
"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": 0.022760097810532898,
"cast/masked_wrong_good_token_rate": 0.19796427059918642,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.003153099001792725,
"clip_ratio/high_mean": 0.003153099001792725,
"clip_ratio/low_mean": 0.0010413205081931665,
"clip_ratio/low_min": 0.0010413205081931665,
"clip_ratio/region_mean": 0.004194419496343471,
"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": 0.23751422902569175,
"epoch": 0.00466653704063776,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.09378629177808762,
"learning_rate": 9.985093006675739e-07,
"loss": 0.10850650072097778,
"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": 2.689601182937622,
"sampling/importance_sampling_ratio/mean": 0.8244835734367371,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1105525493621826,
"sampling/sampling_logp_difference/mean": 0.010338903404772282,
"step": 24,
"step_time": 90.36057408104534
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.04543603537604213,
"cast/masked_gap_abs_mean": 0.0083692412590608,
"cast/masked_gap_mean": -4.818679096274536e-05,
"cast/masked_negative_base_rate": 0.8617292605340481,
"cast/masked_positive_base_rate": 0.13827073434367776,
"cast/masked_sign_flipped_rate": 0.27676752489060163,
"cast/masked_teacher_negative_rate": 0.3561823694035411,
"cast/masked_teacher_positive_rate": 0.26209727488458157,
"cast/masked_token_adv_abs_mean": 0.8512601722031832,
"cast/masked_token_adv_max": 1.012695426121354,
"cast/masked_token_adv_mean": -0.2906405671965331,
"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": 0.021685353247448802,
"cast/masked_wrong_good_token_rate": 0.23133148765191436,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0024775474612397375,
"clip_ratio/high_mean": 0.0024775474612397375,
"clip_ratio/low_mean": 0.0008606515475548804,
"clip_ratio/low_min": 0.0008606515475548804,
"clip_ratio/region_mean": 0.0033381989760528086,
"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": 0.23236830974929035,
"epoch": 0.004860976083997667,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.047416675835847855,
"learning_rate": 9.984444876531208e-07,
"loss": 0.12369062006473541,
"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": 2.6448144912719727,
"sampling/importance_sampling_ratio/mean": 0.5508453845977783,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.744603157043457,
"sampling/sampling_logp_difference/mean": 0.008908877149224281,
"step": 25,
"step_time": 270.59058677600115
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.10509100242052227,
"cast/masked_gap_abs_mean": 0.009026272004120983,
"cast/masked_gap_mean": 1.429499889127328e-06,
"cast/masked_negative_base_rate": 0.6875515312422067,
"cast/masked_positive_base_rate": 0.31244845874607563,
"cast/masked_sign_flipped_rate": 0.2949011782184243,
"cast/masked_teacher_negative_rate": 0.36529657058417797,
"cast/masked_teacher_positive_rate": 0.2669433746486902,
"cast/masked_token_adv_abs_mean": 0.6338496059179306,
"cast/masked_token_adv_max": 0.8498547561466694,
"cast/masked_token_adv_mean": -0.07118164359781076,
"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": 0.021857958810869604,
"cast/masked_wrong_good_token_rate": 0.18981017591431737,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0021171679809413035,
"clip_ratio/high_mean": 0.0021171679809413035,
"clip_ratio/low_mean": 0.0010359731249991455,
"clip_ratio/low_min": 0.0010359731249991455,
"clip_ratio/region_mean": 0.0031531411077594385,
"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": 0.2651766624767333,
"epoch": 0.005055415127357573,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.05345948413014412,
"learning_rate": 9.983796746386673e-07,
"loss": 0.06908059120178223,
"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": 2.950369119644165,
"sampling/importance_sampling_ratio/mean": 0.6947933435440063,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.6418895721435547,
"sampling/sampling_logp_difference/mean": 0.009199452586472034,
"step": 26,
"step_time": 200.46420910488814
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.0,
"cast/masked_correct_bad_token_rate": 0.14863623259589076,
"cast/masked_gap_abs_mean": 0.009099123228224926,
"cast/masked_gap_mean": -7.119418228285213e-07,
"cast/masked_negative_base_rate": 0.5548847913742065,
"cast/masked_positive_base_rate": 0.44511520490050316,
"cast/masked_sign_flipped_rate": 0.30581009900197387,
"cast/masked_teacher_negative_rate": 0.3552693617530167,
"cast/masked_teacher_positive_rate": 0.2629146655090153,
"cast/masked_token_adv_abs_mean": 0.7660624859854579,
"cast/masked_token_adv_max": 0.9419733565300703,
"cast/masked_token_adv_mean": -0.033631134749157354,
"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": 0.021994095761328936,
"cast/masked_wrong_good_token_rate": 0.15717386617325246,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0021265009763737908,
"clip_ratio/high_mean": 0.0021265009763737908,
"clip_ratio/low_mean": 0.00086753362120362,
"clip_ratio/low_min": 0.00086753362120362,
"clip_ratio/region_mean": 0.002994034584844485,
"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": 0.2633260260336101,
"epoch": 0.00524985417071748,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.036070872098207474,
"learning_rate": 9.98314861624214e-07,
"loss": -0.010548461228609085,
"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": 1.7247287034988403,
"sampling/importance_sampling_ratio/mean": 0.4214469790458679,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.223006010055542,
"sampling/sampling_logp_difference/mean": 0.009877452626824379,
"step": 27,
"step_time": 164.48860610890551
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.07096311612986028,
"cast/masked_gap_abs_mean": 0.00889401719905436,
"cast/masked_gap_mean": -3.9098749084587325e-05,
"cast/masked_negative_base_rate": 0.7873802538961172,
"cast/masked_positive_base_rate": 0.21261974330991507,
"cast/masked_sign_flipped_rate": 0.2894062218256295,
"cast/masked_teacher_negative_rate": 0.38274661358445883,
"cast/masked_teacher_positive_rate": 0.26637217681854963,
"cast/masked_token_adv_abs_mean": 0.9200695818290114,
"cast/masked_token_adv_max": 1.0317463986575603,
"cast/masked_token_adv_mean": -0.31544924886839,
"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": 0.022095281223300844,
"cast/masked_wrong_good_token_rate": 0.21844310546293855,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0020639216108975234,
"clip_ratio/high_mean": 0.0020639216108975234,
"clip_ratio/low_mean": 0.0007886513185439981,
"clip_ratio/low_min": 0.0007886513185439981,
"clip_ratio/region_mean": 0.0028525729176180903,
"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": 0.26123066851869226,
"epoch": 0.005444293214077387,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.08036497235298157,
"learning_rate": 9.98250048609761e-07,
"loss": 0.20769448578357697,
"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": 2.9116907119750977,
"sampling/importance_sampling_ratio/mean": 0.6061539649963379,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.2273352146148682,
"sampling/sampling_logp_difference/mean": 0.009839124046266079,
"step": 28,
"step_time": 133.88668746408075
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.07612499222159386,
"cast/masked_gap_abs_mean": 0.008191849177819677,
"cast/masked_gap_mean": -4.116357752081967e-05,
"cast/masked_negative_base_rate": 0.76144780497998,
"cast/masked_positive_base_rate": 0.23855218943208456,
"cast/masked_sign_flipped_rate": 0.2851045480929315,
"cast/masked_teacher_negative_rate": 0.3573023257777095,
"cast/masked_teacher_positive_rate": 0.26766452193260193,
"cast/masked_token_adv_abs_mean": 0.8574392730370164,
"cast/masked_token_adv_max": 0.9961292427033186,
"cast/masked_token_adv_mean": -0.23575025201716926,
"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": 0.021191389474552125,
"cast/masked_wrong_good_token_rate": 0.208979555638507,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0019827648629870964,
"clip_ratio/high_mean": 0.0019827648629870964,
"clip_ratio/low_mean": 0.000820489603029273,
"clip_ratio/low_min": 0.000820489603029273,
"clip_ratio/region_mean": 0.0028032544687448535,
"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": 0.23967273393645883,
"epoch": 0.005638732257437293,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.04513034597039223,
"learning_rate": 9.981852355953076e-07,
"loss": 0.1420724093914032,
"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": 2.6845016479492188,
"sampling/importance_sampling_ratio/mean": 0.5668693780899048,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7337391376495361,
"sampling/sampling_logp_difference/mean": 0.008446319960057735,
"step": 29,
"step_time": 221.9570908251044
},
{
"cast/all_correct_group_rate": 0.25,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.08766021393239498,
"cast/masked_gap_abs_mean": 0.00822867569513619,
"cast/masked_gap_mean": -3.203051591071926e-05,
"cast/masked_negative_base_rate": 0.7130365455523133,
"cast/masked_positive_base_rate": 0.28696345374919474,
"cast/masked_sign_flipped_rate": 0.27854283479973674,
"cast/masked_teacher_negative_rate": 0.33700960222631693,
"cast/masked_teacher_positive_rate": 0.25356326159089804,
"cast/masked_token_adv_abs_mean": 0.807909781113267,
"cast/masked_token_adv_max": 1.01687787566334,
"cast/masked_token_adv_mean": -0.16693260002648458,
"cast/masked_token_adv_min": -1.1406629150733352,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02176000166218728,
"cast/masked_wrong_good_token_rate": 0.19088262179866433,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0023666970373596996,
"clip_ratio/high_mean": 0.0023666970373596996,
"clip_ratio/low_mean": 0.000992647152088466,
"clip_ratio/low_min": 0.000992647152088466,
"clip_ratio/region_mean": 0.0033593441876291763,
"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": 0.2159935098607093,
"epoch": 0.0058331713007972,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.06755436956882477,
"learning_rate": 9.98120422580854e-07,
"loss": 0.10049697011709213,
"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": 2.414320230484009,
"sampling/importance_sampling_ratio/mean": 0.6758387684822083,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.6919755339622498,
"sampling/sampling_logp_difference/mean": 0.009192313998937607,
"step": 30,
"step_time": 97.02242440383998
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.12645897280890495,
"cast/masked_gap_abs_mean": 0.009138892739429139,
"cast/masked_gap_mean": 2.659996794762698e-05,
"cast/masked_negative_base_rate": 0.6390382051467896,
"cast/masked_positive_base_rate": 0.3609617929905653,
"cast/masked_sign_flipped_rate": 0.31465368857607245,
"cast/masked_teacher_negative_rate": 0.3624859172850847,
"cast/masked_teacher_positive_rate": 0.2880529845133424,
"cast/masked_token_adv_abs_mean": 0.7970142662525177,
"cast/masked_token_adv_max": 0.962659690529108,
"cast/masked_token_adv_mean": -0.15270360745489597,
"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": 0.022450102434959263,
"cast/masked_wrong_good_token_rate": 0.18819471588358283,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.002487419365934329,
"clip_ratio/high_mean": 0.002487419365934329,
"clip_ratio/low_mean": 0.000820154576558707,
"clip_ratio/low_min": 0.000820154576558707,
"clip_ratio/region_mean": 0.0033075739484047517,
"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": 0.24896306218579412,
"epoch": 0.0060276103441571065,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.04412037134170532,
"learning_rate": 9.980556095664008e-07,
"loss": 0.06061670184135437,
"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": 2.6014437675476074,
"sampling/importance_sampling_ratio/mean": 0.48437875509262085,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1886683702468872,
"sampling/sampling_logp_difference/mean": 0.009623178280889988,
"step": 31,
"step_time": 273.26218054204946
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.14690247573889792,
"cast/masked_gap_abs_mean": 0.009348065570520703,
"cast/masked_gap_mean": -3.8172927304458426e-05,
"cast/masked_negative_base_rate": 0.551108623854816,
"cast/masked_positive_base_rate": 0.4488913742825389,
"cast/masked_sign_flipped_rate": 0.31190154841169715,
"cast/masked_teacher_negative_rate": 0.35330006619915366,
"cast/masked_teacher_positive_rate": 0.2687120595946908,
"cast/masked_token_adv_abs_mean": 0.859165208414197,
"cast/masked_token_adv_max": 1.0221386216580868,
"cast/masked_token_adv_mean": -0.05791156552731991,
"cast/masked_token_adv_min": -1.0929433107376099,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022878628282342106,
"cast/masked_wrong_good_token_rate": 0.16499907278921455,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.002517100236218539,
"clip_ratio/high_mean": 0.002517100236218539,
"clip_ratio/low_mean": 0.0009702857450974989,
"clip_ratio/low_min": 0.0009702857450974989,
"clip_ratio/region_mean": 0.0034873859949584585,
"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": 0.23995936149731278,
"epoch": 0.006222049387517014,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.07325555384159088,
"learning_rate": 9.979907965519477e-07,
"loss": 0.030916104093194008,
"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": 2.8343751430511475,
"sampling/importance_sampling_ratio/mean": 0.6320965886116028,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7489891052246094,
"sampling/sampling_logp_difference/mean": 0.009270422160625458,
"step": 32,
"step_time": 126.23043232192867
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.06595868151634932,
"cast/masked_gap_abs_mean": 0.008449329718132503,
"cast/masked_gap_mean": -1.0985751998759952e-05,
"cast/masked_negative_base_rate": 0.795351724140346,
"cast/masked_positive_base_rate": 0.2046482702717185,
"cast/masked_sign_flipped_rate": 0.27869987627491355,
"cast/masked_teacher_negative_rate": 0.32989889895543456,
"cast/masked_teacher_positive_rate": 0.26397630106657743,
"cast/masked_token_adv_abs_mean": 0.7609101636335254,
"cast/masked_token_adv_max": 1.03647277969867,
"cast/masked_token_adv_mean": -0.1580037391395308,
"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": 0.021870541851967573,
"cast/masked_wrong_good_token_rate": 0.21274119475856423,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.002395163093751762,
"clip_ratio/high_mean": 0.002395163093751762,
"clip_ratio/low_mean": 0.00090010477197211,
"clip_ratio/low_min": 0.00090010477197211,
"clip_ratio/region_mean": 0.003295267866633367,
"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": 0.23269863333553076,
"epoch": 0.00641648843087692,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.05682654306292534,
"learning_rate": 9.979259835374942e-07,
"loss": 0.0754956603050232,
"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": 2.592017889022827,
"sampling/importance_sampling_ratio/mean": 0.5750653743743896,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 3.4028193950653076,
"sampling/sampling_logp_difference/mean": 0.008798387832939625,
"step": 33,
"step_time": 102.4500099470315
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.09768366464413702,
"cast/masked_gap_abs_mean": 0.008351872333150823,
"cast/masked_gap_mean": 2.1177997084009803e-05,
"cast/masked_negative_base_rate": 0.676145825535059,
"cast/masked_positive_base_rate": 0.323854167945683,
"cast/masked_sign_flipped_rate": 0.2817387371324003,
"cast/masked_teacher_negative_rate": 0.3234522365964949,
"cast/masked_teacher_positive_rate": 0.25673570251092315,
"cast/masked_token_adv_abs_mean": 0.6955733606591821,
"cast/masked_token_adv_max": 0.848291234113276,
"cast/masked_token_adv_mean": -0.13650225091259927,
"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": 0.021623743668897077,
"cast/masked_wrong_good_token_rate": 0.18405507458373904,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.002251298968985793,
"clip_ratio/high_mean": 0.002251298968985793,
"clip_ratio/low_mean": 0.0009938501079886919,
"clip_ratio/low_min": 0.0009938501079886919,
"clip_ratio/region_mean": 0.003245149104259326,
"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": 0.2247882557567209,
"epoch": 0.006610927474236827,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.04699010029435158,
"learning_rate": 9.97861170523041e-07,
"loss": 0.06462706625461578,
"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": 2.5518386363983154,
"sampling/importance_sampling_ratio/mean": 0.5705206394195557,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.9767348766326904,
"sampling/sampling_logp_difference/mean": 0.007753041107207537,
"step": 34,
"step_time": 273.35180958904675
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.15132633142638952,
"cast/masked_gap_abs_mean": 0.009412902625626884,
"cast/masked_gap_mean": 1.5541676532393467e-05,
"cast/masked_negative_base_rate": 0.5005701640620828,
"cast/masked_positive_base_rate": 0.49942983826622367,
"cast/masked_sign_flipped_rate": 0.30049426574259996,
"cast/masked_teacher_negative_rate": 0.3400684189982712,
"cast/masked_teacher_positive_rate": 0.2662795218639076,
"cast/masked_token_adv_abs_mean": 0.6729429429396987,
"cast/masked_token_adv_max": 0.8819929575547576,
"cast/masked_token_adv_mean": -0.021039977960754186,
"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": 0.022688008757540956,
"cast/masked_wrong_good_token_rate": 0.14916793489828706,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.002429949538054643,
"clip_ratio/high_mean": 0.002429949538054643,
"clip_ratio/low_mean": 0.0009645233121773344,
"clip_ratio/low_min": 0.0009645233121773344,
"clip_ratio/region_mean": 0.0033944728384085465,
"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": 0.2388327862136066,
"epoch": 0.006805366517596733,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.07543313503265381,
"learning_rate": 9.977963575085876e-07,
"loss": 0.014049575664103031,
"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": 2.809549570083618,
"sampling/importance_sampling_ratio/mean": 0.6907652616500854,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.6398296356201172,
"sampling/sampling_logp_difference/mean": 0.009891439229249954,
"step": 35,
"step_time": 98.48211741592968
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.12588576029520482,
"cast/masked_gap_abs_mean": 0.009561845537973568,
"cast/masked_gap_mean": -4.929784381602076e-06,
"cast/masked_negative_base_rate": 0.6214258726686239,
"cast/masked_positive_base_rate": 0.37857412127777934,
"cast/masked_sign_flipped_rate": 0.3050325242802501,
"cast/masked_teacher_negative_rate": 0.34942934196442366,
"cast/masked_teacher_positive_rate": 0.2712551769800484,
"cast/masked_token_adv_abs_mean": 0.8243454648181796,
"cast/masked_token_adv_max": 1.0239122174680233,
"cast/masked_token_adv_mean": -0.08144258486572653,
"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": 0.022822067083325237,
"cast/masked_wrong_good_token_rate": 0.17914676223881543,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0022627567286690464,
"clip_ratio/high_mean": 0.0022627567286690464,
"clip_ratio/low_mean": 0.0009094812280636688,
"clip_ratio/low_min": 0.0009094812280636688,
"clip_ratio/region_mean": 0.0031722379608254414,
"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": 0.2557943395804614,
"epoch": 0.00699980556095664,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.07301393151283264,
"learning_rate": 9.977315444941343e-07,
"loss": 0.06232655420899391,
"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": 2.9090027809143066,
"sampling/importance_sampling_ratio/mean": 0.6266592741012573,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.1183149814605713,
"sampling/sampling_logp_difference/mean": 0.009372740983963013,
"step": 36,
"step_time": 165.15147567610256
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.08441135857719928,
"cast/masked_gap_abs_mean": 0.00728035943757277,
"cast/masked_gap_mean": -8.835240036475511e-05,
"cast/masked_negative_base_rate": 0.7441395493224263,
"cast/masked_positive_base_rate": 0.25586043833754957,
"cast/masked_sign_flipped_rate": 0.2564804144203663,
"cast/masked_teacher_negative_rate": 0.3279013242572546,
"cast/masked_teacher_positive_rate": 0.23793868301436305,
"cast/masked_token_adv_abs_mean": 0.9863313976675272,
"cast/masked_token_adv_max": 1.0781249701976776,
"cast/masked_token_adv_mean": -0.3130458614905365,
"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": 0.02049118559807539,
"cast/masked_wrong_good_token_rate": 0.17206905642524362,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.002196720233769156,
"clip_ratio/high_mean": 0.002196720233769156,
"clip_ratio/low_mean": 0.000776385459175799,
"clip_ratio/low_min": 0.000776385459175799,
"clip_ratio/region_mean": 0.0029731056638411246,
"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": 0.19456078601069748,
"epoch": 0.007194244604316547,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.05932265520095825,
"learning_rate": 9.97666731479681e-07,
"loss": 0.16373787820339203,
"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": 2.9397852420806885,
"sampling/importance_sampling_ratio/mean": 0.6662572622299194,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1150591373443604,
"sampling/sampling_logp_difference/mean": 0.007384068798273802,
"step": 37,
"step_time": 274.0738996420114
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.06796297058463097,
"cast/masked_gap_abs_mean": 0.009061138858669437,
"cast/masked_gap_mean": 2.4379328332102546e-05,
"cast/masked_negative_base_rate": 0.7987304572016001,
"cast/masked_positive_base_rate": 0.20126953721046448,
"cast/masked_sign_flipped_rate": 0.26950891641899943,
"cast/masked_teacher_negative_rate": 0.3446820182725787,
"cast/masked_teacher_positive_rate": 0.2561068432405591,
"cast/masked_token_adv_abs_mean": 0.7837593564763665,
"cast/masked_token_adv_max": 0.9774105465039611,
"cast/masked_token_adv_mean": -0.21665471360029187,
"cast/masked_token_adv_min": -1.0688263988122344,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022155535931233317,
"cast/masked_wrong_good_token_rate": 0.20154594560153782,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.002283074827573728,
"clip_ratio/high_mean": 0.002283074827573728,
"clip_ratio/low_mean": 0.0009569754838594235,
"clip_ratio/low_min": 0.0009569754838594235,
"clip_ratio/region_mean": 0.0032400503259850666,
"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": 0.26127393264323473,
"epoch": 0.007388683647676453,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.0367390401661396,
"learning_rate": 9.976019184652277e-07,
"loss": 0.10068871825933456,
"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": 2.9355547428131104,
"sampling/importance_sampling_ratio/mean": 0.42828333377838135,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.462102174758911,
"sampling/sampling_logp_difference/mean": 0.00934706162661314,
"step": 38,
"step_time": 193.55105158506194
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.11091284768190235,
"cast/masked_gap_abs_mean": 0.008861826150678098,
"cast/masked_gap_mean": -6.241568266496245e-05,
"cast/masked_negative_base_rate": 0.6814862163737416,
"cast/masked_positive_base_rate": 0.31851377384737134,
"cast/masked_sign_flipped_rate": 0.3093612571246922,
"cast/masked_teacher_negative_rate": 0.35753704607486725,
"cast/masked_teacher_positive_rate": 0.275733336340636,
"cast/masked_token_adv_abs_mean": 0.7293221689760685,
"cast/masked_token_adv_max": 0.8604638613760471,
"cast/masked_token_adv_mean": -0.16477992315776646,
"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": 0.022392690181732178,
"cast/masked_wrong_good_token_rate": 0.19844840932637453,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0024044210331339855,
"clip_ratio/high_mean": 0.0024044210331339855,
"clip_ratio/low_mean": 0.0010037233096227283,
"clip_ratio/low_min": 0.0010037233096227283,
"clip_ratio/region_mean": 0.0034081443518516608,
"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": 0.22908123792149127,
"epoch": 0.00758312269103636,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.06288203597068787,
"learning_rate": 9.975371054507744e-07,
"loss": 0.10303903371095657,
"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": 2.7447242736816406,
"sampling/importance_sampling_ratio/mean": 0.6373331546783447,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1105506420135498,
"sampling/sampling_logp_difference/mean": 0.009182456880807877,
"step": 39,
"step_time": 231.25372937388602
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.0869808669667691,
"cast/masked_gap_abs_mean": 0.008994543924927711,
"cast/masked_gap_mean": 4.0166060216506594e-05,
"cast/masked_negative_base_rate": 0.7591868117451668,
"cast/masked_positive_base_rate": 0.24081317894160748,
"cast/masked_sign_flipped_rate": 0.288746387232095,
"cast/masked_teacher_negative_rate": 0.3668307298794389,
"cast/masked_teacher_positive_rate": 0.2626654556952417,
"cast/masked_token_adv_abs_mean": 0.7197249559685588,
"cast/masked_token_adv_max": 0.9283391162753105,
"cast/masked_token_adv_mean": -0.19197966833598912,
"cast/masked_token_adv_min": -1.0230584423989058,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02232081675902009,
"cast/masked_wrong_good_token_rate": 0.20176551956683397,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0023291566503758077,
"clip_ratio/high_mean": 0.0023291566503758077,
"clip_ratio/low_mean": 0.000732218083612679,
"clip_ratio/low_min": 0.000732218083612679,
"clip_ratio/region_mean": 0.0030613747403549496,
"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": 0.2466601738706231,
"epoch": 0.0077775617343962666,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.058035291731357574,
"learning_rate": 9.974722924363211e-07,
"loss": 0.1709366887807846,
"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": 2.7949984073638916,
"sampling/importance_sampling_ratio/mean": 0.5943577885627747,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9051461219787598,
"sampling/sampling_logp_difference/mean": 0.00886913388967514,
"step": 40,
"step_time": 195.46495445200708
},
{
"cast/all_correct_group_rate": 0.25,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.09728774975519627,
"cast/masked_gap_abs_mean": 0.009159208959317766,
"cast/masked_gap_mean": -0.00011376830394738136,
"cast/masked_negative_base_rate": 0.7151693273335695,
"cast/masked_positive_base_rate": 0.28483065916225314,
"cast/masked_sign_flipped_rate": 0.29898134153336287,
"cast/masked_teacher_negative_rate": 0.3611158477142453,
"cast/masked_teacher_positive_rate": 0.2703435714356601,
"cast/masked_token_adv_abs_mean": 0.9469579942524433,
"cast/masked_token_adv_max": 1.0546874552965164,
"cast/masked_token_adv_mean": -0.24952742899768054,
"cast/masked_token_adv_min": -1.1997498981654644,
"cast/masked_token_weight_max": 1.1997498981654644,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022706614225171506,
"cast/masked_wrong_good_token_rate": 0.20169359119608998,
"cast/mixed_group_rate": 0.125,
"clip_ratio/high_max": 0.002162655469874153,
"clip_ratio/high_mean": 0.002162655469874153,
"clip_ratio/low_mean": 0.0009493868901699898,
"clip_ratio/low_min": 0.0009493868901699898,
"clip_ratio/region_mean": 0.003112042355496669,
"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": 0.2554546883329749,
"epoch": 0.007972000777756174,
"frac_reward_zero_std": 0.875,
"grad_norm": 0.05790575221180916,
"learning_rate": 9.974074794218678e-07,
"loss": 0.08802931755781174,
"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": 2.602684736251831,
"sampling/importance_sampling_ratio/mean": 0.5129495859146118,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8390898704528809,
"sampling/sampling_logp_difference/mean": 0.009790458716452122,
"step": 41,
"step_time": 147.30263931193622
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.875,
"cast/masked_correct_bad_token_rate": 0.03805837780237198,
"cast/masked_gap_abs_mean": 0.009114296743064187,
"cast/masked_gap_mean": -0.00023764620823385485,
"cast/masked_negative_base_rate": 0.8930426947772503,
"cast/masked_positive_base_rate": 0.10695729777216911,
"cast/masked_sign_flipped_rate": 0.2852556095458567,
"cast/masked_teacher_negative_rate": 0.355336288921535,
"cast/masked_teacher_positive_rate": 0.27298082364723086,
"cast/masked_token_adv_abs_mean": 0.9412195328623056,
"cast/masked_token_adv_max": 1.033472646959126,
"cast/masked_token_adv_mean": -0.3910328111378476,
"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": 0.022836856078356504,
"cast/masked_wrong_good_token_rate": 0.24719723174348474,
"cast/mixed_group_rate": 0.125,
"clip_ratio/high_max": 0.0024077063253571396,
"clip_ratio/high_mean": 0.0024077063253571396,
"clip_ratio/low_mean": 0.0009662687862146413,
"clip_ratio/low_min": 0.0009662687862146413,
"clip_ratio/region_mean": 0.003373975105205318,
"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": 0.2509133405983448,
"epoch": 0.00816643982111608,
"frac_reward_zero_std": 0.875,
"grad_norm": 0.08223497867584229,
"learning_rate": 9.973426664074145e-07,
"loss": 0.27267447113990784,
"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": 2.9711296558380127,
"sampling/importance_sampling_ratio/mean": 0.7263649106025696,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7273352146148682,
"sampling/sampling_logp_difference/mean": 0.00955159217119217,
"step": 42,
"step_time": 203.69333430021652
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.09040886908769608,
"cast/masked_gap_abs_mean": 0.009696896930108778,
"cast/masked_gap_mean": -9.58390318714919e-05,
"cast/masked_negative_base_rate": 0.7462949650362134,
"cast/masked_positive_base_rate": 0.2537050321698189,
"cast/masked_sign_flipped_rate": 0.3029978242702782,
"cast/masked_teacher_negative_rate": 0.3656229367479682,
"cast/masked_teacher_positive_rate": 0.2789644212462008,
"cast/masked_token_adv_abs_mean": 0.8087759530171752,
"cast/masked_token_adv_max": 1.02059980109334,
"cast/masked_token_adv_mean": -0.22510353242978454,
"cast/masked_token_adv_min": -1.134043661877513,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02300074289087206,
"cast/masked_wrong_good_token_rate": 0.21258895355276763,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.00244624223705614,
"clip_ratio/high_mean": 0.00244624223705614,
"clip_ratio/low_mean": 0.0009396896934958932,
"clip_ratio/low_min": 0.0009396896934958932,
"clip_ratio/region_mean": 0.003385931962839095,
"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": 0.2731356513686478,
"epoch": 0.008360878864475986,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.06264642626047134,
"learning_rate": 9.972778533929612e-07,
"loss": 0.13891656696796417,
"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": 2.754892587661743,
"sampling/importance_sampling_ratio/mean": 0.5070384740829468,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.4984270334243774,
"sampling/sampling_logp_difference/mean": 0.010324406437575817,
"step": 43,
"step_time": 121.362138421071
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.07347885007038713,
"cast/masked_gap_abs_mean": 0.009790325653739274,
"cast/masked_gap_mean": 0.0001377832815521174,
"cast/masked_negative_base_rate": 0.7693555569276214,
"cast/masked_positive_base_rate": 0.2306444412097335,
"cast/masked_sign_flipped_rate": 0.30503204790875316,
"cast/masked_teacher_negative_rate": 0.36554758809506893,
"cast/masked_teacher_positive_rate": 0.28177780006080866,
"cast/masked_token_adv_abs_mean": 0.7782454742118716,
"cast/masked_token_adv_max": 1.0007508611306548,
"cast/masked_token_adv_mean": -0.21499018224858446,
"cast/masked_token_adv_min": -1.116360572166741,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02282586181536317,
"cast/masked_wrong_good_token_rate": 0.23155319853685796,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0028966893514734693,
"clip_ratio/high_mean": 0.0028966893514734693,
"clip_ratio/low_mean": 0.0009003291784210887,
"clip_ratio/low_min": 0.0009003291784210887,
"clip_ratio/region_mean": 0.0037970185476297047,
"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": 0.27769548119977117,
"epoch": 0.008555317907835893,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.06184614822268486,
"learning_rate": 9.97213040378508e-07,
"loss": 0.10782088339328766,
"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": 2.460670232772827,
"sampling/importance_sampling_ratio/mean": 0.5162792205810547,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 4.44534969329834,
"sampling/sampling_logp_difference/mean": 0.010349072515964508,
"step": 44,
"step_time": 161.39731844194466
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.07212227385025471,
"cast/masked_gap_abs_mean": 0.00878744876536075,
"cast/masked_gap_mean": 3.536397838388439e-05,
"cast/masked_negative_base_rate": 0.7822737013921142,
"cast/masked_positive_base_rate": 0.21772629534825683,
"cast/masked_sign_flipped_rate": 0.280248514842242,
"cast/masked_teacher_negative_rate": 0.36749067064374685,
"cast/masked_teacher_positive_rate": 0.25757518969476223,
"cast/masked_token_adv_abs_mean": 0.8648540489375591,
"cast/masked_token_adv_max": 1.0260647302493453,
"cast/masked_token_adv_mean": -0.30054123559966683,
"cast/masked_token_adv_min": -1.151216915808618,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.021832135156728327,
"cast/masked_wrong_good_token_rate": 0.20812623901292682,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0024155761384463403,
"clip_ratio/high_mean": 0.0024155761384463403,
"clip_ratio/low_mean": 0.0011783851464315376,
"clip_ratio/low_min": 0.0011783851464315376,
"clip_ratio/region_mean": 0.0035939613117079716,
"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": 0.2467019958421588,
"epoch": 0.0087497569511958,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.04465532302856445,
"learning_rate": 9.971482273640546e-07,
"loss": 0.11927051842212677,
"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": 2.612135648727417,
"sampling/importance_sampling_ratio/mean": 0.4354548454284668,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0235925912857056,
"sampling/sampling_logp_difference/mean": 0.00921543687582016,
"step": 45,
"step_time": 154.84717604101752
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.06595766544342041,
"cast/masked_gap_abs_mean": 0.009326171901193447,
"cast/masked_gap_mean": 0.00010998268334105887,
"cast/masked_negative_base_rate": 0.8100231778807938,
"cast/masked_positive_base_rate": 0.18997682351619005,
"cast/masked_sign_flipped_rate": 0.317003165371716,
"cast/masked_teacher_negative_rate": 0.3899944545701146,
"cast/masked_teacher_positive_rate": 0.30171894654631615,
"cast/masked_token_adv_abs_mean": 0.8762097069993615,
"cast/masked_token_adv_max": 1.0328885139897466,
"cast/masked_token_adv_mean": -0.21540844184346497,
"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": 0.02213648206088692,
"cast/masked_wrong_good_token_rate": 0.25104550062678754,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.002238386481622001,
"clip_ratio/high_mean": 0.002238386481622001,
"clip_ratio/low_mean": 0.0008133577493936173,
"clip_ratio/low_min": 0.0008133577493936173,
"clip_ratio/region_mean": 0.003051744195545325,
"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": 0.285619530826807,
"epoch": 0.008944195994555707,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.032319582998752594,
"learning_rate": 9.970834143496013e-07,
"loss": 0.07124408334493637,
"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": 2.099602699279785,
"sampling/importance_sampling_ratio/mean": 0.39266473054885864,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 5.176179885864258,
"sampling/sampling_logp_difference/mean": 0.009971310384571552,
"step": 46,
"step_time": 229.55913826296455
},
{
"cast/all_correct_group_rate": 0.25,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.13336130511015654,
"cast/masked_gap_abs_mean": 0.010533620632486418,
"cast/masked_gap_mean": -0.00014370381040862412,
"cast/masked_negative_base_rate": 0.6405656998977065,
"cast/masked_positive_base_rate": 0.3594342996366322,
"cast/masked_sign_flipped_rate": 0.3255960368551314,
"cast/masked_teacher_negative_rate": 0.39104455802589655,
"cast/masked_teacher_positive_rate": 0.28882898623123765,
"cast/masked_token_adv_abs_mean": 0.8100866433233023,
"cast/masked_token_adv_max": 0.9858372788876295,
"cast/masked_token_adv_mean": -0.07501125750422943,
"cast/masked_token_adv_min": -1.09337511472404,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02382063475670293,
"cast/masked_wrong_good_token_rate": 0.19223473267629743,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.002621103874844266,
"clip_ratio/high_mean": 0.002621103874844266,
"clip_ratio/low_mean": 0.0010420654425615794,
"clip_ratio/low_min": 0.0010420654425615794,
"clip_ratio/region_mean": 0.003663169325591298,
"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": 0.2910246164537966,
"epoch": 0.009138635037915614,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.07035275548696518,
"learning_rate": 9.97018601335148e-07,
"loss": 0.13869012892246246,
"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": 2.627039909362793,
"sampling/importance_sampling_ratio/mean": 0.6216158866882324,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7819712162017822,
"sampling/sampling_logp_difference/mean": 0.01048360113054514,
"step": 47,
"step_time": 158.7220798259077
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.0,
"cast/masked_correct_bad_token_rate": 0.1819662859197706,
"cast/masked_gap_abs_mean": 0.008690150134498253,
"cast/masked_gap_mean": 0.0002072433610464941,
"cast/masked_negative_base_rate": 0.4675047677010298,
"cast/masked_positive_base_rate": 0.5324952304363251,
"cast/masked_sign_flipped_rate": 0.3027907791547477,
"cast/masked_teacher_negative_rate": 0.35083322413265705,
"cast/masked_teacher_positive_rate": 0.2605379153974354,
"cast/masked_token_adv_abs_mean": 0.8712713401764631,
"cast/masked_token_adv_max": 0.9999091066420078,
"cast/masked_token_adv_mean": -0.00890017765777884,
"cast/masked_token_adv_min": -1.0709234792739153,
"cast/masked_token_weight_max": 1.199919380247593,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.021993522066622972,
"cast/masked_wrong_good_token_rate": 0.12082449381705374,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.003064125972741749,
"clip_ratio/high_mean": 0.003064125972741749,
"clip_ratio/low_mean": 0.0011575405142139061,
"clip_ratio/low_min": 0.0011575405142139061,
"clip_ratio/region_mean": 0.004221666473313235,
"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": 0.2142971227876842,
"epoch": 0.00933307408127552,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.08334919065237045,
"learning_rate": 9.969537883206947e-07,
"loss": 0.042960312217473984,
"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": 2.931452512741089,
"sampling/importance_sampling_ratio/mean": 0.800648033618927,
"sampling/importance_sampling_ratio/min": 0.000695361290127039,
"sampling/sampling_logp_difference/max": 2.434075355529785,
"sampling/sampling_logp_difference/mean": 0.009055417962372303,
"step": 48,
"step_time": 256.9930206818972
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.056139033287763596,
"cast/masked_gap_abs_mean": 0.00897680057096295,
"cast/masked_gap_mean": -2.67826628714829e-05,
"cast/masked_negative_base_rate": 0.8146085143089294,
"cast/masked_positive_base_rate": 0.18539148615673184,
"cast/masked_sign_flipped_rate": 0.2827273551374674,
"cast/masked_teacher_negative_rate": 0.3649442698806524,
"cast/masked_teacher_positive_rate": 0.26541501795873046,
"cast/masked_token_adv_abs_mean": 0.8977016480639577,
"cast/masked_token_adv_max": 1.0737702921032906,
"cast/masked_token_adv_mean": -0.2214975090464577,
"cast/masked_token_adv_min": -1.1887310519814491,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022105299693066627,
"cast/masked_wrong_good_token_rate": 0.22658832115121186,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.002224740841484163,
"clip_ratio/high_mean": 0.002224740841484163,
"clip_ratio/low_mean": 0.0009301207355747465,
"clip_ratio/low_min": 0.0009301207355747465,
"clip_ratio/region_mean": 0.0031548615588690154,
"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": 0.2712486379314214,
"epoch": 0.009527513124635427,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.06614314019680023,
"learning_rate": 9.968889753062414e-07,
"loss": 0.14364849030971527,
"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": 2.541189193725586,
"sampling/importance_sampling_ratio/mean": 0.6740047931671143,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.722158670425415,
"sampling/sampling_logp_difference/mean": 0.009428436867892742,
"step": 49,
"step_time": 123.9688211340108
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.09154471143847331,
"cast/masked_gap_abs_mean": 0.009754821134265512,
"cast/masked_gap_mean": 6.932815483651211e-05,
"cast/masked_negative_base_rate": 0.7226162496954203,
"cast/masked_positive_base_rate": 0.27738374238833785,
"cast/masked_sign_flipped_rate": 0.2992625432088971,
"cast/masked_teacher_negative_rate": 0.3671647747978568,
"cast/masked_teacher_positive_rate": 0.2713055252097547,
"cast/masked_token_adv_abs_mean": 0.6886172760277987,
"cast/masked_token_adv_max": 0.972032692283392,
"cast/masked_token_adv_mean": -0.10807860917702783,
"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": 0.02271361800376326,
"cast/masked_wrong_good_token_rate": 0.20771783194504678,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0027401930565247312,
"clip_ratio/high_mean": 0.0027401930565247312,
"clip_ratio/low_mean": 0.000850930661727034,
"clip_ratio/low_min": 0.000850930661727034,
"clip_ratio/region_mean": 0.0035911237137042917,
"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": 0.2758284162264317,
"epoch": 0.009721952167995334,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.05226795747876167,
"learning_rate": 9.968241622917881e-07,
"loss": 0.01283115241676569,
"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": 2.618908643722534,
"sampling/importance_sampling_ratio/mean": 0.47958293557167053,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8236097097396851,
"sampling/sampling_logp_difference/mean": 0.010377779603004456,
"step": 50,
"step_time": 147.82380417900276
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.11087374435737729,
"cast/masked_gap_abs_mean": 0.008257852052338421,
"cast/masked_gap_mean": -5.3988800800652825e-05,
"cast/masked_negative_base_rate": 0.6648754612542689,
"cast/masked_positive_base_rate": 0.3351245285011828,
"cast/masked_sign_flipped_rate": 0.2891837786883116,
"cast/masked_teacher_negative_rate": 0.34325721487402916,
"cast/masked_teacher_positive_rate": 0.26669878978282213,
"cast/masked_token_adv_abs_mean": 0.872913908213377,
"cast/masked_token_adv_max": 1.015033669769764,
"cast/masked_token_adv_mean": -0.18416843877639621,
"cast/masked_token_adv_min": -1.0935717411339283,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.021525257907342166,
"cast/masked_wrong_good_token_rate": 0.17831003409810364,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.002193793387050391,
"clip_ratio/high_mean": 0.002193793387050391,
"clip_ratio/low_mean": 0.0007622628886565508,
"clip_ratio/low_min": 0.0007622628886565508,
"clip_ratio/region_mean": 0.002956056319817435,
"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": 0.2262600506655872,
"epoch": 0.00991639121135524,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.048856221139431,
"learning_rate": 9.967593492773348e-07,
"loss": 0.10438315570354462,
"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": 2.8710594177246094,
"sampling/importance_sampling_ratio/mean": 0.705763578414917,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8248758912086487,
"sampling/sampling_logp_difference/mean": 0.008419380523264408,
"step": 51,
"step_time": 183.19165293505648
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.05430146516300738,
"cast/masked_gap_abs_mean": 0.010204099889961071,
"cast/masked_gap_mean": -0.00013953898911722717,
"cast/masked_negative_base_rate": 0.8417554656043649,
"cast/masked_positive_base_rate": 0.15824452973902225,
"cast/masked_sign_flipped_rate": 0.31040558870881796,
"cast/masked_teacher_negative_rate": 0.39350080117583275,
"cast/masked_teacher_positive_rate": 0.2959583792835474,
"cast/masked_token_adv_abs_mean": 0.9307650774717331,
"cast/masked_token_adv_max": 1.0720829218626022,
"cast/masked_token_adv_mean": -0.23820831818738952,
"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": 0.023536924505606294,
"cast/masked_wrong_good_token_rate": 0.25610412494279444,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.002449403398713912,
"clip_ratio/high_mean": 0.002449403398713912,
"clip_ratio/low_mean": 0.0008622400455351453,
"clip_ratio/low_min": 0.0008622400455351453,
"clip_ratio/region_mean": 0.0033116434715338983,
"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": 0.2975070171523839,
"epoch": 0.010110830254715146,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.08816739171743393,
"learning_rate": 9.966945362628815e-07,
"loss": 0.15404991805553436,
"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": 2.9045135974884033,
"sampling/importance_sampling_ratio/mean": 0.6386955976486206,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9250767827033997,
"sampling/sampling_logp_difference/mean": 0.010824895463883877,
"step": 52,
"step_time": 115.04979335595272
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.13103906298056245,
"cast/masked_gap_abs_mean": 0.00867158173059579,
"cast/masked_gap_mean": 8.520549022250634e-05,
"cast/masked_negative_base_rate": 0.5921763726510108,
"cast/masked_positive_base_rate": 0.4078236264176667,
"cast/masked_sign_flipped_rate": 0.2958722645416856,
"cast/masked_teacher_negative_rate": 0.34344194270670414,
"cast/masked_teacher_positive_rate": 0.26881610229611397,
"cast/masked_token_adv_abs_mean": 0.8359079034999013,
"cast/masked_token_adv_max": 1.0285062808543444,
"cast/masked_token_adv_mean": -0.13531865987170022,
"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": 0.022305827355012298,
"cast/masked_wrong_good_token_rate": 0.16483320004772395,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0027284732277621515,
"clip_ratio/high_mean": 0.0027284732277621515,
"clip_ratio/low_mean": 0.0010108187616424402,
"clip_ratio/low_min": 0.0010108187616424402,
"clip_ratio/region_mean": 0.0037392919621197507,
"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": 0.2297525075264275,
"epoch": 0.010305269298075053,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.0757998526096344,
"learning_rate": 9.966297232484283e-07,
"loss": 0.10435373336076736,
"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": 2.9255056381225586,
"sampling/importance_sampling_ratio/mean": 0.6371498107910156,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8585796356201172,
"sampling/sampling_logp_difference/mean": 0.009534546174108982,
"step": 53,
"step_time": 109.03995435204706
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.11193778668530285,
"cast/masked_gap_abs_mean": 0.009238735045073554,
"cast/masked_gap_mean": -2.1865243979846127e-05,
"cast/masked_negative_base_rate": 0.6553607988171279,
"cast/masked_positive_base_rate": 0.3446392035111785,
"cast/masked_sign_flipped_rate": 0.29842907935380936,
"cast/masked_teacher_negative_rate": 0.3556557856500149,
"cast/masked_teacher_positive_rate": 0.2669450039975345,
"cast/masked_token_adv_abs_mean": 0.7042175699025393,
"cast/masked_token_adv_max": 0.9115561936050653,
"cast/masked_token_adv_mean": -0.0964602081803605,
"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": 0.022769812669139355,
"cast/masked_wrong_good_token_rate": 0.18649129173718393,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0026978229734595516,
"clip_ratio/high_mean": 0.0026978229734595516,
"clip_ratio/low_mean": 0.0008825387731121737,
"clip_ratio/low_min": 0.0008825387731121737,
"clip_ratio/region_mean": 0.0035803617465717252,
"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": 0.2438266526442021,
"epoch": 0.01049970834143496,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.046803612262010574,
"learning_rate": 9.96564910233975e-07,
"loss": 0.013173217885196209,
"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": 2.99257230758667,
"sampling/importance_sampling_ratio/mean": 0.5788929462432861,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.199986219406128,
"sampling/sampling_logp_difference/mean": 0.009375724010169506,
"step": 54,
"step_time": 203.37697717212723
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.75,
"cast/masked_correct_bad_token_rate": 0.025258404202759266,
"cast/masked_gap_abs_mean": 0.008525458790245466,
"cast/masked_gap_mean": 2.568952883308384e-05,
"cast/masked_negative_base_rate": 0.9120821105316281,
"cast/masked_positive_base_rate": 0.08791788201779127,
"cast/masked_sign_flipped_rate": 0.2608451754786074,
"cast/masked_teacher_negative_rate": 0.3493344997987151,
"cast/masked_teacher_positive_rate": 0.255140638910234,
"cast/masked_token_adv_abs_mean": 0.9445765241980553,
"cast/masked_token_adv_max": 1.064395286142826,
"cast/masked_token_adv_mean": -0.3555865193484351,
"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": 0.021619874343741685,
"cast/masked_wrong_good_token_rate": 0.23558677011169493,
"cast/mixed_group_rate": 0.25,
"clip_ratio/high_max": 0.002227030836365884,
"clip_ratio/high_mean": 0.002227030836365884,
"clip_ratio/low_mean": 0.0007343955971919058,
"clip_ratio/low_min": 0.0007343955971919058,
"clip_ratio/region_mean": 0.0029614264312840533,
"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": 0.23636674229055643,
"epoch": 0.010694147384794867,
"frac_reward_zero_std": 0.75,
"grad_norm": 0.05278799682855606,
"learning_rate": 9.965000972195217e-07,
"loss": 0.21401697397232056,
"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": 2.735874652862549,
"sampling/importance_sampling_ratio/mean": 0.5522472858428955,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7373205423355103,
"sampling/sampling_logp_difference/mean": 0.008717945776879787,
"step": 55,
"step_time": 267.44089655095013
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.15523835399653763,
"cast/masked_gap_abs_mean": 0.00977662108198274,
"cast/masked_gap_mean": 0.00026457849881467155,
"cast/masked_negative_base_rate": 0.5606556981801987,
"cast/masked_positive_base_rate": 0.4393442999571562,
"cast/masked_sign_flipped_rate": 0.31416797125712037,
"cast/masked_teacher_negative_rate": 0.3649888327345252,
"cast/masked_teacher_positive_rate": 0.27489376440644264,
"cast/masked_token_adv_abs_mean": 0.8041507983580232,
"cast/masked_token_adv_max": 0.9984474936500192,
"cast/masked_token_adv_mean": -0.14618788729421794,
"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": 0.022755273093935102,
"cast/masked_wrong_good_token_rate": 0.15892961854115129,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.002641144434164744,
"clip_ratio/high_mean": 0.002641144434164744,
"clip_ratio/low_mean": 0.0009398416450494551,
"clip_ratio/low_min": 0.0009398416450494551,
"clip_ratio/region_mean": 0.0035809860710287467,
"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": 0.2704219617880881,
"epoch": 0.010888586428154774,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.05931501463055611,
"learning_rate": 9.964352842050684e-07,
"loss": 0.056634243577718735,
"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": 2.899747371673584,
"sampling/importance_sampling_ratio/mean": 0.5644305944442749,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.9032444953918457,
"sampling/sampling_logp_difference/mean": 0.010586690157651901,
"step": 56,
"step_time": 139.8945430839376
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.09202963695861399,
"cast/masked_gap_abs_mean": 0.008649816016259138,
"cast/masked_gap_mean": 0.00014146899853528794,
"cast/masked_negative_base_rate": 0.7419142033904791,
"cast/masked_positive_base_rate": 0.25808579847216606,
"cast/masked_sign_flipped_rate": 0.2760803955607116,
"cast/masked_teacher_negative_rate": 0.34562700102105737,
"cast/masked_teacher_positive_rate": 0.2518073320388794,
"cast/masked_token_adv_abs_mean": 0.9360907152295113,
"cast/masked_token_adv_max": 1.0687499642372131,
"cast/masked_token_adv_mean": -0.320831737248227,
"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": 0.02153065722086467,
"cast/masked_wrong_good_token_rate": 0.18405075767077506,
"cast/mixed_group_rate": 0.25,
"clip_ratio/high_max": 0.0023762480977893574,
"clip_ratio/high_mean": 0.0023762480977893574,
"clip_ratio/low_mean": 0.0007844120655136066,
"clip_ratio/low_min": 0.0007844120655136066,
"clip_ratio/region_mean": 0.003160660147841554,
"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": 0.23360277828760445,
"epoch": 0.01108302547151468,
"frac_reward_zero_std": 0.75,
"grad_norm": 0.06469165533781052,
"learning_rate": 9.96370471190615e-07,
"loss": 0.30016282200813293,
"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": 2.931509256362915,
"sampling/importance_sampling_ratio/mean": 0.6179847717285156,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 7.276133060455322,
"sampling/sampling_logp_difference/mean": 0.008552690036594868,
"step": 57,
"step_time": 241.38528090092586
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.08040050300769508,
"cast/masked_gap_abs_mean": 0.00787402005516924,
"cast/masked_gap_mean": 0.0001125245040327627,
"cast/masked_negative_base_rate": 0.7362182773649693,
"cast/masked_positive_base_rate": 0.26378172682598233,
"cast/masked_sign_flipped_rate": 0.25853768829256296,
"cast/masked_teacher_negative_rate": 0.3134927027858794,
"cast/masked_teacher_positive_rate": 0.2390769673511386,
"cast/masked_token_adv_abs_mean": 0.8492175210267305,
"cast/masked_token_adv_max": 1.0109032932668924,
"cast/masked_token_adv_mean": -0.1898872099845903,
"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": 0.021331840893253684,
"cast/masked_wrong_good_token_rate": 0.1781371859833598,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.002467530297508347,
"clip_ratio/high_mean": 0.002467530297508347,
"clip_ratio/low_mean": 0.0009266034580832638,
"clip_ratio/low_min": 0.0009266034580832638,
"clip_ratio/region_mean": 0.003394133749679895,
"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": 0.20118912169709802,
"epoch": 0.011277464514874587,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.05418000370264053,
"learning_rate": 9.963056581761618e-07,
"loss": 0.08751223981380463,
"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": 2.865302085876465,
"sampling/importance_sampling_ratio/mean": 0.547175943851471,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9017418622970581,
"sampling/sampling_logp_difference/mean": 0.008321200497448444,
"step": 58,
"step_time": 271.0389580830815
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.15163059742189944,
"cast/masked_gap_abs_mean": 0.008740335091715679,
"cast/masked_gap_mean": -6.444084459644728e-05,
"cast/masked_negative_base_rate": 0.49799581803381443,
"cast/masked_positive_base_rate": 0.502004174515605,
"cast/masked_sign_flipped_rate": 0.2873364519327879,
"cast/masked_teacher_negative_rate": 0.3353893496096134,
"cast/masked_teacher_positive_rate": 0.24948588199913502,
"cast/masked_token_adv_abs_mean": 0.8731774259358644,
"cast/masked_token_adv_max": 1.0266915019601583,
"cast/masked_token_adv_mean": -0.07645641986164264,
"cast/masked_token_adv_min": -1.1249661780893803,
"cast/masked_token_weight_max": 1.199749756604433,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022584606369491667,
"cast/masked_wrong_good_token_rate": 0.13570585462730378,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0027949359573540278,
"clip_ratio/high_mean": 0.0027949359573540278,
"clip_ratio/low_mean": 0.0009574840414643404,
"clip_ratio/low_min": 0.0009574840414643404,
"clip_ratio/region_mean": 0.003752419954253128,
"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": 0.21598330466076732,
"epoch": 0.011471903558234494,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.06529859453439713,
"learning_rate": 9.962408451617085e-07,
"loss": -0.006746533792465925,
"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": 2.5795373916625977,
"sampling/importance_sampling_ratio/mean": 0.589619517326355,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9853978157043457,
"sampling/sampling_logp_difference/mean": 0.009253195486962795,
"step": 59,
"step_time": 93.15257632400608
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.02909607137553394,
"cast/masked_gap_abs_mean": 0.008529116210411303,
"cast/masked_gap_mean": 4.7473757128102534e-05,
"cast/masked_negative_base_rate": 0.9013398252427578,
"cast/masked_positive_base_rate": 0.09866016590967774,
"cast/masked_sign_flipped_rate": 0.2710537910461426,
"cast/masked_teacher_negative_rate": 0.35837032459676266,
"cast/masked_teacher_positive_rate": 0.2610379825346172,
"cast/masked_token_adv_abs_mean": 0.8932753372937441,
"cast/masked_token_adv_max": 1.0576875880360603,
"cast/masked_token_adv_mean": -0.3102522486296948,
"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": 0.021439569944050163,
"cast/masked_wrong_good_token_rate": 0.24195771990343928,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0021610846561088692,
"clip_ratio/high_mean": 0.0021610846561088692,
"clip_ratio/low_mean": 0.0009267101750083384,
"clip_ratio/low_min": 0.0009267101750083384,
"clip_ratio/region_mean": 0.0030877948011038825,
"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": 0.24979626783169806,
"epoch": 0.0116663426015944,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.05580515414476395,
"learning_rate": 9.961760321472552e-07,
"loss": 0.1445513516664505,
"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": 2.824261426925659,
"sampling/importance_sampling_ratio/mean": 0.5477856993675232,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.5039867162704468,
"sampling/sampling_logp_difference/mean": 0.009042279794812202,
"step": 60,
"step_time": 161.62706496598548
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.044694975134916604,
"cast/masked_gap_abs_mean": 0.009549252776196226,
"cast/masked_gap_mean": 0.00010694018988033349,
"cast/masked_negative_base_rate": 0.8654961185529828,
"cast/masked_positive_base_rate": 0.1345038777217269,
"cast/masked_sign_flipped_rate": 0.3124868618324399,
"cast/masked_teacher_negative_rate": 0.3704560436308384,
"cast/masked_teacher_positive_rate": 0.2984726568683982,
"cast/masked_token_adv_abs_mean": 0.8948290171101689,
"cast/masked_token_adv_max": 1.0281508099287748,
"cast/masked_token_adv_mean": -0.28328741260338575,
"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": 0.022813796473201364,
"cast/masked_wrong_good_token_rate": 0.26779188588261604,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.002394408909822232,
"clip_ratio/high_mean": 0.002394408909822232,
"clip_ratio/low_mean": 0.0009521286519884598,
"clip_ratio/low_min": 0.0009521286519884598,
"clip_ratio/region_mean": 0.0033465376109234057,
"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": 0.26891204551793635,
"epoch": 0.011860781644954306,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.07637074589729309,
"learning_rate": 9.961112191328019e-07,
"loss": 0.19085629284381866,
"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": 2.9192190170288086,
"sampling/importance_sampling_ratio/mean": 0.7080554962158203,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7364048957824707,
"sampling/sampling_logp_difference/mean": 0.009949129074811935,
"step": 61,
"step_time": 133.95986241998617
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.0,
"cast/masked_correct_bad_token_rate": 0.17658105061855167,
"cast/masked_gap_abs_mean": 0.0076156123614055105,
"cast/masked_gap_mean": 5.220790956173005e-05,
"cast/masked_negative_base_rate": 0.4759402684867382,
"cast/masked_positive_base_rate": 0.5240597368683666,
"cast/masked_sign_flipped_rate": 0.2899611033499241,
"cast/masked_teacher_negative_rate": 0.33154874527826905,
"cast/masked_teacher_positive_rate": 0.24630169617012143,
"cast/masked_token_adv_abs_mean": 0.8379851672798395,
"cast/masked_token_adv_max": 0.9869773900136352,
"cast/masked_token_adv_mean": -0.03442883532261476,
"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": 0.020715738704893738,
"cast/masked_wrong_good_token_rate": 0.11338005354627967,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0023573103162561893,
"clip_ratio/high_mean": 0.0023573103162561893,
"clip_ratio/low_mean": 0.0009461390391152236,
"clip_ratio/low_min": 0.0009461390391152236,
"clip_ratio/region_mean": 0.00330344936446636,
"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": 0.1994137023575604,
"epoch": 0.012055220688314213,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.0442718043923378,
"learning_rate": 9.960464061183486e-07,
"loss": 0.023740656673908234,
"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": 2.799053430557251,
"sampling/importance_sampling_ratio/mean": 0.5237549543380737,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8392471075057983,
"sampling/sampling_logp_difference/mean": 0.007702337112277746,
"step": 62,
"step_time": 224.97681845285115
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.04981245496310294,
"cast/masked_gap_abs_mean": 0.009298108619987033,
"cast/masked_gap_mean": 5.824428399137105e-06,
"cast/masked_negative_base_rate": 0.8577438807114959,
"cast/masked_positive_base_rate": 0.14225611742585897,
"cast/masked_sign_flipped_rate": 0.27648847736418247,
"cast/masked_teacher_negative_rate": 0.3576759956777096,
"cast/masked_teacher_positive_rate": 0.26809552777558565,
"cast/masked_token_adv_abs_mean": 0.7875767527148128,
"cast/masked_token_adv_max": 0.9946974897757173,
"cast/masked_token_adv_mean": -0.29076882428489625,
"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": 0.022479382110759616,
"cast/masked_wrong_good_token_rate": 0.22667602193541825,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0024881332283257507,
"clip_ratio/high_mean": 0.0024881332283257507,
"clip_ratio/low_mean": 0.0009588701987013337,
"clip_ratio/low_min": 0.0009588701987013337,
"clip_ratio/region_mean": 0.0034470034443074837,
"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": 0.2531194854527712,
"epoch": 0.01224965973167412,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.050841402262449265,
"learning_rate": 9.959815931038953e-07,
"loss": 0.14473453164100647,
"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": 2.9445018768310547,
"sampling/importance_sampling_ratio/mean": 0.5162006616592407,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.8846280574798584,
"sampling/sampling_logp_difference/mean": 0.009547882713377476,
"step": 63,
"step_time": 133.51274775306229
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.14192295377142727,
"cast/masked_gap_abs_mean": 0.00886192062171176,
"cast/masked_gap_mean": 0.00014121197830263554,
"cast/masked_negative_base_rate": 0.5954568218439817,
"cast/masked_positive_base_rate": 0.4045431804843247,
"cast/masked_sign_flipped_rate": 0.29256354458630085,
"cast/masked_teacher_negative_rate": 0.34497988130897284,
"cast/masked_teacher_positive_rate": 0.24663081532344222,
"cast/masked_token_adv_abs_mean": 0.8228231808170676,
"cast/masked_token_adv_max": 1.0342155043035746,
"cast/masked_token_adv_mean": -0.14669496702845208,
"cast/masked_token_adv_min": -1.1433943659067154,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022482828178908676,
"cast/masked_wrong_good_token_rate": 0.15064058976713568,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0030606143564000377,
"clip_ratio/high_mean": 0.0030606143564000377,
"clip_ratio/low_mean": 0.0009602876716598985,
"clip_ratio/low_min": 0.0009602876716598985,
"clip_ratio/region_mean": 0.004020901960757328,
"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": 0.21123783686198294,
"epoch": 0.012444098775034027,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.07010306417942047,
"learning_rate": 9.95916780089442e-07,
"loss": 0.05579278618097305,
"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": 2.0307540893554688,
"sampling/importance_sampling_ratio/mean": 0.7110783457756042,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.9064671993255615,
"sampling/sampling_logp_difference/mean": 0.0092029869556427,
"step": 64,
"step_time": 125.0495309070102
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.13659233751241118,
"cast/masked_gap_abs_mean": 0.00890085834544152,
"cast/masked_gap_mean": 6.569908693876414e-05,
"cast/masked_negative_base_rate": 0.5918788006529212,
"cast/masked_positive_base_rate": 0.40812119911424816,
"cast/masked_sign_flipped_rate": 0.2999548213556409,
"cast/masked_teacher_negative_rate": 0.36068171821534634,
"cast/masked_teacher_positive_rate": 0.26736877113580704,
"cast/masked_token_adv_abs_mean": 0.7960978802293539,
"cast/masked_token_adv_max": 1.0008167456835508,
"cast/masked_token_adv_mean": -0.16283198143355548,
"cast/masked_token_adv_min": -1.1163788568228483,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022436604893300682,
"cast/masked_wrong_good_token_rate": 0.16336248512379825,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0028876953292638063,
"clip_ratio/high_mean": 0.0028876953292638063,
"clip_ratio/low_mean": 0.0011397473272154457,
"clip_ratio/low_min": 0.0011397473272154457,
"clip_ratio/region_mean": 0.0040274426246469375,
"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": 0.22429389273747802,
"epoch": 0.012638537818393934,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.07370628416538239,
"learning_rate": 9.958519670749887e-07,
"loss": 0.1308983415365219,
"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": 2.8778700828552246,
"sampling/importance_sampling_ratio/mean": 0.6838222742080688,
"sampling/importance_sampling_ratio/min": 0.014866217039525509,
"sampling/sampling_logp_difference/max": 0.8087811470031738,
"sampling/sampling_logp_difference/mean": 0.009463171474635601,
"step": 65,
"step_time": 158.2861882550642
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.12225427385419607,
"cast/masked_gap_abs_mean": 0.007693660285440274,
"cast/masked_gap_mean": 2.4201688574976288e-05,
"cast/masked_negative_base_rate": 0.616733786650002,
"cast/masked_positive_base_rate": 0.3832662121858448,
"cast/masked_sign_flipped_rate": 0.28792133275419474,
"cast/masked_teacher_negative_rate": 0.332840318325907,
"cast/masked_teacher_positive_rate": 0.253705064766109,
"cast/masked_token_adv_abs_mean": 0.8695665365085006,
"cast/masked_token_adv_max": 1.022943602874875,
"cast/masked_token_adv_mean": -0.14215529020293616,
"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": 0.02087765946635045,
"cast/masked_wrong_good_token_rate": 0.16566705820150673,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0022255709300225135,
"clip_ratio/high_mean": 0.0022255709300225135,
"clip_ratio/low_mean": 0.0007638969000254292,
"clip_ratio/low_min": 0.0007638969000254292,
"clip_ratio/region_mean": 0.0029894678555137943,
"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": 0.20489980187267065,
"epoch": 0.01283297686175384,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.04067128524184227,
"learning_rate": 9.957871540605352e-07,
"loss": 0.023002970963716507,
"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": 2.767570734024048,
"sampling/importance_sampling_ratio/mean": 0.5531458854675293,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.6621971130371094,
"sampling/sampling_logp_difference/mean": 0.008254320360720158,
"step": 66,
"step_time": 272.52449508893187
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.12953835155349225,
"cast/masked_gap_abs_mean": 0.007566865155240521,
"cast/masked_gap_mean": -0.00010381544018400746,
"cast/masked_negative_base_rate": 0.5626683072187006,
"cast/masked_positive_base_rate": 0.4373316867277026,
"cast/masked_sign_flipped_rate": 0.27733563631772995,
"cast/masked_teacher_negative_rate": 0.32386042084544897,
"cast/masked_teacher_positive_rate": 0.24402588373050094,
"cast/masked_token_adv_abs_mean": 0.8249164242297411,
"cast/masked_token_adv_max": 1.0085157044231892,
"cast/masked_token_adv_mean": -0.007779439678415656,
"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": 0.021370380534790456,
"cast/masked_wrong_good_token_rate": 0.147797285229899,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.002277240764669841,
"clip_ratio/high_mean": 0.002277240764669841,
"clip_ratio/low_mean": 0.0011595253008636064,
"clip_ratio/low_min": 0.0011595253008636064,
"clip_ratio/region_mean": 0.003436766070080921,
"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": 0.179686373565346,
"epoch": 0.013027415905113747,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.06525473296642303,
"learning_rate": 9.95722341046082e-07,
"loss": -0.01239592395722866,
"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": 2.7989084720611572,
"sampling/importance_sampling_ratio/mean": 0.6963121891021729,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8481130599975586,
"sampling/sampling_logp_difference/mean": 0.00786517933011055,
"step": 67,
"step_time": 133.5403149400372
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.75,
"cast/masked_correct_bad_token_rate": 0.013547386508435011,
"cast/masked_gap_abs_mean": 0.009112865154747851,
"cast/masked_gap_mean": -6.235810587895685e-06,
"cast/masked_negative_base_rate": 0.9606668297201395,
"cast/masked_positive_base_rate": 0.03933315351605415,
"cast/masked_sign_flipped_rate": 0.2851053839549422,
"cast/masked_teacher_negative_rate": 0.36467608343809843,
"cast/masked_teacher_positive_rate": 0.2814084505662322,
"cast/masked_token_adv_abs_mean": 0.909663469530642,
"cast/masked_token_adv_max": 1.0294507937505841,
"cast/masked_token_adv_mean": -0.3549479671055451,
"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": 0.02224936365382746,
"cast/masked_wrong_good_token_rate": 0.27155799558386207,
"cast/mixed_group_rate": 0.25,
"clip_ratio/high_max": 0.002212746381701436,
"clip_ratio/high_mean": 0.002212746381701436,
"clip_ratio/low_mean": 0.0009415811373401084,
"clip_ratio/low_min": 0.0009415811373401084,
"clip_ratio/region_mean": 0.0031543275363219436,
"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": 0.2598267216235399,
"epoch": 0.013221854948473654,
"frac_reward_zero_std": 0.75,
"grad_norm": 0.05840139091014862,
"learning_rate": 9.956575280316288e-07,
"loss": 0.16540491580963135,
"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": 2.687131881713867,
"sampling/importance_sampling_ratio/mean": 0.5638353228569031,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0763484239578247,
"sampling/sampling_logp_difference/mean": 0.009698573499917984,
"step": 68,
"step_time": 265.96112499092123
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.17675346287433058,
"cast/masked_gap_abs_mean": 0.008053057827055454,
"cast/masked_gap_mean": 0.0001276730834236517,
"cast/masked_negative_base_rate": 0.4573919107206166,
"cast/masked_positive_base_rate": 0.542608086951077,
"cast/masked_sign_flipped_rate": 0.2988571422174573,
"cast/masked_teacher_negative_rate": 0.33770093601197004,
"cast/masked_teacher_positive_rate": 0.24968277709558606,
"cast/masked_token_adv_abs_mean": 0.8167366469278932,
"cast/masked_token_adv_max": 0.9811808913946152,
"cast/masked_token_adv_mean": -0.0956248736474663,
"cast/masked_token_adv_min": -1.0942393466830254,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.021751200663857162,
"cast/masked_wrong_good_token_rate": 0.12210367887746543,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.0025298396685684565,
"clip_ratio/high_mean": 0.0025298396685684565,
"clip_ratio/low_mean": 0.0007958099699862942,
"clip_ratio/low_min": 0.0007958099699862942,
"clip_ratio/region_mean": 0.003325649639009498,
"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": 0.19675025390461087,
"epoch": 0.01341629399183356,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.03981938585639,
"learning_rate": 9.955927150171755e-07,
"loss": -0.014707633294165134,
"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": 2.759721040725708,
"sampling/importance_sampling_ratio/mean": 0.6216585636138916,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.4554905891418457,
"sampling/sampling_logp_difference/mean": 0.008401052094995975,
"step": 69,
"step_time": 267.9205508690793
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.75,
"cast/masked_correct_bad_token_rate": 0.03669861506205052,
"cast/masked_gap_abs_mean": 0.00879837793763727,
"cast/masked_gap_mean": -5.9446612766578255e-05,
"cast/masked_negative_base_rate": 0.8901497293263674,
"cast/masked_positive_base_rate": 0.10985027113929391,
"cast/masked_sign_flipped_rate": 0.26450858218595386,
"cast/masked_teacher_negative_rate": 0.3460956485942006,
"cast/masked_teacher_positive_rate": 0.25487704295665026,
"cast/masked_token_adv_abs_mean": 0.9812994580715895,
"cast/masked_token_adv_max": 1.0687499642372131,
"cast/masked_token_adv_mean": -0.42017630208283663,
"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": 0.022204925189726055,
"cast/masked_wrong_good_token_rate": 0.22780996817164123,
"cast/mixed_group_rate": 0.25,
"clip_ratio/high_max": 0.0026018485023087123,
"clip_ratio/high_mean": 0.0026018485023087123,
"clip_ratio/low_mean": 0.0007670721543036052,
"clip_ratio/low_min": 0.0007670721543036052,
"clip_ratio/region_mean": 0.00336892066479777,
"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": 0.2286684620194137,
"epoch": 0.013610733035193466,
"frac_reward_zero_std": 0.75,
"grad_norm": 0.04467220604419708,
"learning_rate": 9.95527902002722e-07,
"loss": 0.14497093856334686,
"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": 2.271578311920166,
"sampling/importance_sampling_ratio/mean": 0.4394351541996002,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.255159854888916,
"sampling/sampling_logp_difference/mean": 0.008971469476819038,
"step": 70,
"step_time": 271.663609089941
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.875,
"cast/masked_correct_bad_token_rate": 0.008274147287011147,
"cast/masked_gap_abs_mean": 0.008543972682673484,
"cast/masked_gap_mean": 2.120507065228594e-05,
"cast/masked_negative_base_rate": 0.9807054847478867,
"cast/masked_positive_base_rate": 0.019294507801532745,
"cast/masked_sign_flipped_rate": 0.2578729984816164,
"cast/masked_teacher_negative_rate": 0.34596965136006474,
"cast/masked_teacher_positive_rate": 0.2556357828434557,
"cast/masked_token_adv_abs_mean": 0.9501910023391247,
"cast/masked_token_adv_max": 1.0546874552965164,
"cast/masked_token_adv_mean": -0.45947257766965777,
"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": 0.02138325513806194,
"cast/masked_wrong_good_token_rate": 0.2495988509617746,
"cast/mixed_group_rate": 0.125,
"clip_ratio/high_max": 0.0022139067623356823,
"clip_ratio/high_mean": 0.0022139067623356823,
"clip_ratio/low_mean": 0.0008400341334890982,
"clip_ratio/low_min": 0.0008400341334890982,
"clip_ratio/region_mean": 0.003053940956306178,
"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": 0.24780980590730906,
"epoch": 0.013805172078553373,
"frac_reward_zero_std": 0.875,
"grad_norm": 0.05653403699398041,
"learning_rate": 9.954630889882689e-07,
"loss": 0.2082698494195938,
"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": 2.903388738632202,
"sampling/importance_sampling_ratio/mean": 0.5065118074417114,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.3762526512145996,
"sampling/sampling_logp_difference/mean": 0.008795135654509068,
"step": 71,
"step_time": 254.44281946704723
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.1368188970373012,
"cast/masked_gap_abs_mean": 0.008358282488188706,
"cast/masked_gap_mean": 0.0001285133533883709,
"cast/masked_negative_base_rate": 0.5765360384248197,
"cast/masked_positive_base_rate": 0.4234639548230916,
"cast/masked_sign_flipped_rate": 0.28893421357497573,
"cast/masked_teacher_negative_rate": 0.33595342095941305,
"cast/masked_teacher_positive_rate": 0.2552150972187519,
"cast/masked_token_adv_abs_mean": 0.9428302999585867,
"cast/masked_token_adv_max": 1.0451844409108162,
"cast/masked_token_adv_mean": -0.11629320865904447,
"cast/masked_token_adv_min": -1.1302027627825737,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.021385911386460066,
"cast/masked_wrong_good_token_rate": 0.15211531589739025,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0029135878758097533,
"clip_ratio/high_mean": 0.0029135878758097533,
"clip_ratio/low_mean": 0.0009128504070758936,
"clip_ratio/low_min": 0.0009128504070758936,
"clip_ratio/region_mean": 0.0038264382928900886,
"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": 0.2178162147756666,
"epoch": 0.01399961112191328,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.059951942414045334,
"learning_rate": 9.953982759738156e-07,
"loss": 0.0711783766746521,
"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": 2.000028610229492,
"sampling/importance_sampling_ratio/mean": 0.6002368927001953,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.1528666019439697,
"sampling/sampling_logp_difference/mean": 0.008993521332740784,
"step": 72,
"step_time": 146.95701860598638
},
{
"cast/all_correct_group_rate": 0.25,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.19993111118674278,
"cast/masked_gap_abs_mean": 0.00864142736827489,
"cast/masked_gap_mean": 8.885131029501281e-05,
"cast/masked_negative_base_rate": 0.34614958614110947,
"cast/masked_positive_base_rate": 0.6538504157215357,
"cast/masked_sign_flipped_rate": 0.3064462919719517,
"cast/masked_teacher_negative_rate": 0.33517175912857056,
"cast/masked_teacher_positive_rate": 0.2635028753429651,
"cast/masked_token_adv_abs_mean": 0.8530241530388594,
"cast/masked_token_adv_max": 1.0147856147959828,
"cast/masked_token_adv_mean": 0.0647778382699471,
"cast/masked_token_adv_min": -1.1434519281610847,
"cast/masked_token_weight_max": 1.199692964553833,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.021970175730530173,
"cast/masked_wrong_good_token_rate": 0.10651517997030169,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.002766826457445859,
"clip_ratio/high_mean": 0.002766826457445859,
"clip_ratio/low_mean": 0.0008209049237848376,
"clip_ratio/low_min": 0.0008209049237848376,
"clip_ratio/region_mean": 0.0035877313866876648,
"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": 0.23213600972667336,
"epoch": 0.014194050165273187,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.10185817629098892,
"learning_rate": 9.95333462959362e-07,
"loss": -0.04344337061047554,
"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": 2.8607027530670166,
"sampling/importance_sampling_ratio/mean": 0.8319095969200134,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.49673891067504883,
"sampling/sampling_logp_difference/mean": 0.008946876972913742,
"step": 73,
"step_time": 128.6689785530325
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.08219140314031392,
"cast/masked_gap_abs_mean": 0.008759267191635445,
"cast/masked_gap_mean": 0.00014644248864215115,
"cast/masked_negative_base_rate": 0.7446632799692452,
"cast/masked_positive_base_rate": 0.2553367121145129,
"cast/masked_sign_flipped_rate": 0.2859932999126613,
"cast/masked_teacher_negative_rate": 0.337968610227108,
"cast/masked_teacher_positive_rate": 0.27238682797178626,
"cast/masked_token_adv_abs_mean": 0.7426272900775075,
"cast/masked_token_adv_max": 0.931053064763546,
"cast/masked_token_adv_mean": -0.21710029180394486,
"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": 0.022117521555628628,
"cast/masked_wrong_good_token_rate": 0.20380189653951675,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.002652097027748823,
"clip_ratio/high_mean": 0.002652097027748823,
"clip_ratio/low_mean": 0.00103096733437269,
"clip_ratio/low_min": 0.00103096733437269,
"clip_ratio/region_mean": 0.003683064365759492,
"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": 0.2365065033081919,
"epoch": 0.014388489208633094,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.05233754962682724,
"learning_rate": 9.95268649944909e-07,
"loss": 0.1030249074101448,
"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": 2.486335515975952,
"sampling/importance_sampling_ratio/mean": 0.6504868268966675,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9900574684143066,
"sampling/sampling_logp_difference/mean": 0.00942291971296072,
"step": 74,
"step_time": 113.65382324895472
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.06888874911237508,
"cast/masked_gap_abs_mean": 0.009134963125688955,
"cast/masked_gap_mean": -0.00017147174196452397,
"cast/masked_negative_base_rate": 0.7442568298429251,
"cast/masked_positive_base_rate": 0.2557431608438492,
"cast/masked_sign_flipped_rate": 0.2897478216327727,
"cast/masked_teacher_negative_rate": 0.3530923407524824,
"cast/masked_teacher_positive_rate": 0.2714514322578907,
"cast/masked_token_adv_abs_mean": 0.743761164136231,
"cast/masked_token_adv_max": 0.9918404491618276,
"cast/masked_token_adv_mean": -0.1871636698488146,
"cast/masked_token_adv_min": -1.1049812389537692,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022411125130020082,
"cast/masked_wrong_good_token_rate": 0.2208590725203976,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.0023090958638931625,
"clip_ratio/high_mean": 0.0023090958638931625,
"clip_ratio/low_mean": 0.0010169207516810275,
"clip_ratio/low_min": 0.0010169207516810275,
"clip_ratio/region_mean": 0.0033260165982937906,
"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": 0.25680991169065237,
"epoch": 0.014582928251993,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.04877878352999687,
"learning_rate": 9.952038369304557e-07,
"loss": 0.08954093605279922,
"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": 2.801825523376465,
"sampling/importance_sampling_ratio/mean": 0.5158376097679138,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.6454229354858398,
"sampling/sampling_logp_difference/mean": 0.009610062465071678,
"step": 75,
"step_time": 143.616194680857
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.06732630264014006,
"cast/masked_gap_abs_mean": 0.007581979320093524,
"cast/masked_gap_mean": -8.436288322855034e-05,
"cast/masked_negative_base_rate": 0.797152501065284,
"cast/masked_positive_base_rate": 0.2028474914841354,
"cast/masked_sign_flipped_rate": 0.24984461208805442,
"cast/masked_teacher_negative_rate": 0.3219556990079582,
"cast/masked_teacher_positive_rate": 0.22851746180094779,
"cast/masked_token_adv_abs_mean": 0.8028777418658137,
"cast/masked_token_adv_max": 1.0010836860165,
"cast/masked_token_adv_mean": -0.2722798086469993,
"cast/masked_token_adv_min": -1.106328514404595,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.020313128392444924,
"cast/masked_wrong_good_token_rate": 0.18251830840017647,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.00243456621774385,
"clip_ratio/high_mean": 0.00243456621774385,
"clip_ratio/low_mean": 0.0008908547079045093,
"clip_ratio/low_min": 0.0008908547079045093,
"clip_ratio/region_mean": 0.003325420932014822,
"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": 0.20028587663546205,
"epoch": 0.014777367295352907,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.04738288000226021,
"learning_rate": 9.951390239160022e-07,
"loss": 0.16080540418624878,
"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": 2.759819269180298,
"sampling/importance_sampling_ratio/mean": 0.6307873725891113,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 4.617574691772461,
"sampling/sampling_logp_difference/mean": 0.007333536632359028,
"step": 76,
"step_time": 271.9024003529921
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.047377214999869466,
"cast/masked_gap_abs_mean": 0.007624619553098455,
"cast/masked_gap_mean": 6.018949579811306e-05,
"cast/masked_negative_base_rate": 0.8387785786762834,
"cast/masked_positive_base_rate": 0.16122141014784575,
"cast/masked_sign_flipped_rate": 0.2474751346744597,
"cast/masked_teacher_negative_rate": 0.3222675574943423,
"cast/masked_teacher_positive_rate": 0.23635475989431143,
"cast/masked_token_adv_abs_mean": 0.9636716470122337,
"cast/masked_token_adv_max": 1.0622703284025192,
"cast/masked_token_adv_mean": -0.36609130166471004,
"cast/masked_token_adv_min": -1.1872376017272472,
"cast/masked_token_weight_max": 1.1994064822793007,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02046177556621842,
"cast/masked_wrong_good_token_rate": 0.20009791827760637,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0020960777947038878,
"clip_ratio/high_mean": 0.0020960777947038878,
"clip_ratio/low_mean": 0.0007666799474463915,
"clip_ratio/low_min": 0.0007666799474463915,
"clip_ratio/region_mean": 0.002862757733964827,
"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": 0.22068954235874116,
"epoch": 0.014971806338712814,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.053100086748600006,
"learning_rate": 9.950742109015489e-07,
"loss": 0.24104353785514832,
"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": 2.748882532119751,
"sampling/importance_sampling_ratio/mean": 0.5797144174575806,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.02785325050354,
"sampling/sampling_logp_difference/mean": 0.008147772401571274,
"step": 77,
"step_time": 232.88329445305862
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.14439467317424715,
"cast/masked_gap_abs_mean": 0.00881847902201116,
"cast/masked_gap_mean": 1.2829746765419259e-05,
"cast/masked_negative_base_rate": 0.5452797124162316,
"cast/masked_positive_base_rate": 0.4547202857211232,
"cast/masked_sign_flipped_rate": 0.2964954930357635,
"cast/masked_teacher_negative_rate": 0.3464332539588213,
"cast/masked_teacher_positive_rate": 0.2610924867913127,
"cast/masked_token_adv_abs_mean": 0.7739748489111662,
"cast/masked_token_adv_max": 0.9589521251618862,
"cast/masked_token_adv_mean": -0.10457766917534173,
"cast/masked_token_adv_min": -1.04742581397295,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022327924147248268,
"cast/masked_wrong_good_token_rate": 0.1521008196286857,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.002677722561202245,
"clip_ratio/high_mean": 0.002677722561202245,
"clip_ratio/low_mean": 0.0008462278033221082,
"clip_ratio/low_min": 0.0008462278033221082,
"clip_ratio/region_mean": 0.0035239503704360686,
"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": 0.22767411870881915,
"epoch": 0.01516624538207272,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.06734936684370041,
"learning_rate": 9.950093978870958e-07,
"loss": 0.0039851549081504345,
"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": 2.9778473377227783,
"sampling/importance_sampling_ratio/mean": 0.6693481206893921,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.3005167245864868,
"sampling/sampling_logp_difference/mean": 0.009208445437252522,
"step": 78,
"step_time": 108.59823362293537
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.03173699788749218,
"cast/masked_gap_abs_mean": 0.009998481662478298,
"cast/masked_gap_mean": 4.937579103980738e-05,
"cast/masked_negative_base_rate": 0.909321703016758,
"cast/masked_positive_base_rate": 0.09067829605191946,
"cast/masked_sign_flipped_rate": 0.30219610314816236,
"cast/masked_teacher_negative_rate": 0.3765377616509795,
"cast/masked_teacher_positive_rate": 0.29267304949462414,
"cast/masked_token_adv_abs_mean": 0.8496329430490732,
"cast/masked_token_adv_max": 1.0432463623583317,
"cast/masked_token_adv_mean": -0.2545151373487897,
"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": 0.022686188982333988,
"cast/masked_wrong_good_token_rate": 0.2704591052606702,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0024409614961768966,
"clip_ratio/high_mean": 0.0024409614961768966,
"clip_ratio/low_mean": 0.0008353567250196647,
"clip_ratio/low_min": 0.0008353567250196647,
"clip_ratio/region_mean": 0.00327631820255192,
"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": 0.3034679889678955,
"epoch": 0.015360684425432626,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.02220337837934494,
"learning_rate": 9.949445848726425e-07,
"loss": 0.0686374381184578,
"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": 2.4576046466827393,
"sampling/importance_sampling_ratio/mean": 0.3383052349090576,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.772713840007782,
"sampling/sampling_logp_difference/mean": 0.010557887144386768,
"step": 79,
"step_time": 279.4779279269278
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.1675086496397853,
"cast/masked_gap_abs_mean": 0.00941903458442539,
"cast/masked_gap_mean": -6.0625032745065255e-05,
"cast/masked_negative_base_rate": 0.5152651919052005,
"cast/masked_positive_base_rate": 0.4847348006442189,
"cast/masked_sign_flipped_rate": 0.31583967758342624,
"cast/masked_teacher_negative_rate": 0.3676381604745984,
"cast/masked_teacher_positive_rate": 0.2677987697534263,
"cast/masked_token_adv_abs_mean": 0.8648247169330716,
"cast/masked_token_adv_max": 1.0449450556188822,
"cast/masked_token_adv_mean": -0.13407735613873228,
"cast/masked_token_adv_min": -1.162080142647028,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022852317546494305,
"cast/masked_wrong_good_token_rate": 0.14833102794364095,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.002806874055750086,
"clip_ratio/high_mean": 0.002806874055750086,
"clip_ratio/low_mean": 0.0009480019350576185,
"clip_ratio/low_min": 0.0009480019350576185,
"clip_ratio/region_mean": 0.0037548760028585093,
"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": 0.28650818578898907,
"epoch": 0.015555123468792533,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.07289958000183105,
"learning_rate": 9.94879771858189e-07,
"loss": 0.04264841973781586,
"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": 2.8133151531219482,
"sampling/importance_sampling_ratio/mean": 0.747952938079834,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.5940090417861938,
"sampling/sampling_logp_difference/mean": 0.010329053737223148,
"step": 80,
"step_time": 244.95380383296288
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.07283499056939036,
"cast/masked_gap_abs_mean": 0.009302366510382853,
"cast/masked_gap_mean": -5.971104292257223e-05,
"cast/masked_negative_base_rate": 0.7859457723388914,
"cast/masked_positive_base_rate": 0.21405422291718423,
"cast/masked_sign_flipped_rate": 0.28763003647327423,
"cast/masked_teacher_negative_rate": 0.34227897180244327,
"cast/masked_teacher_positive_rate": 0.27475858200341463,
"cast/masked_token_adv_abs_mean": 0.8463705014437437,
"cast/masked_token_adv_max": 1.0573088712990284,
"cast/masked_token_adv_mean": -0.18615006457548589,
"cast/masked_token_adv_min": -1.1654959246516228,
"cast/masked_token_weight_max": 1.1987980045378208,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022681213449686766,
"cast/masked_wrong_good_token_rate": 0.2147950456055696,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.002675135743629653,
"clip_ratio/high_mean": 0.002675135743629653,
"clip_ratio/low_mean": 0.0011399455106584355,
"clip_ratio/low_min": 0.0011399455106584355,
"clip_ratio/region_mean": 0.0038150811960804276,
"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": 0.24644522997550666,
"epoch": 0.01574956251215244,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.0866413563489914,
"learning_rate": 9.948149588437357e-07,
"loss": 0.07519197463989258,
"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": 2.7801618576049805,
"sampling/importance_sampling_ratio/mean": 0.7156411409378052,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9900568723678589,
"sampling/sampling_logp_difference/mean": 0.009937168098986149,
"step": 81,
"step_time": 115.73673583604977
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.12986552459187806,
"cast/masked_gap_abs_mean": 0.008134154792060144,
"cast/masked_gap_mean": -0.00015043806882886201,
"cast/masked_negative_base_rate": 0.6183662274852395,
"cast/masked_positive_base_rate": 0.38163377391174436,
"cast/masked_sign_flipped_rate": 0.301291489508003,
"cast/masked_teacher_negative_rate": 0.34412652533501387,
"cast/masked_teacher_positive_rate": 0.2619450972415507,
"cast/masked_token_adv_abs_mean": 0.854621322825551,
"cast/masked_token_adv_max": 1.002469346858561,
"cast/masked_token_adv_mean": -0.1668843172956258,
"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": 0.021985574683640152,
"cast/masked_wrong_good_token_rate": 0.17142596514895558,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0024618923998787068,
"clip_ratio/high_mean": 0.0024618923998787068,
"clip_ratio/low_mean": 0.0009330611674158718,
"clip_ratio/low_min": 0.0009330611674158718,
"clip_ratio/region_mean": 0.003394953575480031,
"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": 0.19996826257556677,
"epoch": 0.015944001555512347,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.0727379322052002,
"learning_rate": 9.947501458292826e-07,
"loss": 0.07563764601945877,
"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": 2.4103689193725586,
"sampling/importance_sampling_ratio/mean": 0.8064315319061279,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 3.003965377807617,
"sampling/sampling_logp_difference/mean": 0.00892983004450798,
"step": 82,
"step_time": 125.53383913001744
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.09966595028527081,
"cast/masked_gap_abs_mean": 0.007695627587963827,
"cast/masked_gap_mean": -0.00017380794847099423,
"cast/masked_negative_base_rate": 0.6978404768742621,
"cast/masked_positive_base_rate": 0.302159515209496,
"cast/masked_sign_flipped_rate": 0.2710584672167897,
"cast/masked_teacher_negative_rate": 0.3403177550062537,
"cast/masked_teacher_positive_rate": 0.23810330452397466,
"cast/masked_token_adv_abs_mean": 0.7324405713006854,
"cast/masked_token_adv_max": 0.9246781915426254,
"cast/masked_token_adv_mean": -0.14958274649688974,
"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": 0.021040223189629614,
"cast/masked_wrong_good_token_rate": 0.1713925173971802,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0021545221425185446,
"clip_ratio/high_mean": 0.0021545221425185446,
"clip_ratio/low_mean": 0.0008370203918275365,
"clip_ratio/low_min": 0.0008370203918275365,
"clip_ratio/region_mean": 0.0029915425475337543,
"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": 0.21577559784054756,
"epoch": 0.016138440598872254,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.04885944351553917,
"learning_rate": 9.94685332814829e-07,
"loss": 0.09736268222332001,
"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": 2.288393259048462,
"sampling/importance_sampling_ratio/mean": 0.6499333381652832,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8774776458740234,
"sampling/sampling_logp_difference/mean": 0.008600446395576,
"step": 83,
"step_time": 137.0459316468914
},
{
"cast/all_correct_group_rate": 0.25,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.1280041901045479,
"cast/masked_gap_abs_mean": 0.007529600712587126,
"cast/masked_gap_mean": -0.00010177547972034517,
"cast/masked_negative_base_rate": 0.6462241159752011,
"cast/masked_positive_base_rate": 0.35377588169649243,
"cast/masked_sign_flipped_rate": 0.2691315021365881,
"cast/masked_teacher_negative_rate": 0.32864926708862185,
"cast/masked_teacher_positive_rate": 0.23776511615142226,
"cast/masked_token_adv_abs_mean": 0.8885950930416584,
"cast/masked_token_adv_max": 0.9936055755242705,
"cast/masked_token_adv_mean": -0.23230237641837448,
"cast/masked_token_adv_min": -1.1248350301757455,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.020717678649816662,
"cast/masked_wrong_good_token_rate": 0.14112731022760272,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0019175790257577319,
"clip_ratio/high_mean": 0.0019175790257577319,
"clip_ratio/low_mean": 0.0007932095409159956,
"clip_ratio/low_min": 0.0007932095409159956,
"clip_ratio/region_mean": 0.0027107885616715066,
"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": 0.20280549372546375,
"epoch": 0.01633287964223216,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.04943402111530304,
"learning_rate": 9.946205198003758e-07,
"loss": 0.12903805077075958,
"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": 2.7823150157928467,
"sampling/importance_sampling_ratio/mean": 0.6069948673248291,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.3776627779006958,
"sampling/sampling_logp_difference/mean": 0.0077428012154996395,
"step": 84,
"step_time": 254.07884085198748
},
{
"cast/all_correct_group_rate": 0.375,
"cast/all_wrong_group_rate": 0.0,
"cast/masked_correct_bad_token_rate": 0.1914862059056759,
"cast/masked_gap_abs_mean": 0.00689992900879588,
"cast/masked_gap_mean": -0.00010100559751435867,
"cast/masked_negative_base_rate": 0.3721053935587406,
"cast/masked_positive_base_rate": 0.6278946027159691,
"cast/masked_sign_flipped_rate": 0.277372756972909,
"cast/masked_teacher_negative_rate": 0.3218452497385442,
"cast/masked_teacher_positive_rate": 0.2195048974826932,
"cast/masked_token_adv_abs_mean": 0.8029569331556559,
"cast/masked_token_adv_max": 1.0249658022075891,
"cast/masked_token_adv_mean": 0.08385192209971137,
"cast/masked_token_adv_min": -1.1315984036773443,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.020771792449522763,
"cast/masked_wrong_good_token_rate": 0.08588655013591051,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.002236386244476307,
"clip_ratio/high_mean": 0.002236386244476307,
"clip_ratio/low_mean": 0.0009047812936842092,
"clip_ratio/low_min": 0.0009047812936842092,
"clip_ratio/region_mean": 0.0031411675536219263,
"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": 0.15093764127232134,
"epoch": 0.01652731868559207,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.073886938393116,
"learning_rate": 9.945557067859225e-07,
"loss": -0.10491490364074707,
"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": 2.9576961994171143,
"sampling/importance_sampling_ratio/mean": 0.6921924352645874,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9984359741210938,
"sampling/sampling_logp_difference/mean": 0.007354133762419224,
"step": 85,
"step_time": 90.54505383208743
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.05779461399652064,
"cast/masked_gap_abs_mean": 0.009267302113585174,
"cast/masked_gap_mean": -0.00015966579360338073,
"cast/masked_negative_base_rate": 0.8346325857564807,
"cast/masked_positive_base_rate": 0.16536741191521287,
"cast/masked_sign_flipped_rate": 0.2916068392805755,
"cast/masked_teacher_negative_rate": 0.36398901604115963,
"cast/masked_teacher_positive_rate": 0.2736351229250431,
"cast/masked_token_adv_abs_mean": 0.866229085251689,
"cast/masked_token_adv_max": 1.0388257997110486,
"cast/masked_token_adv_mean": -0.25493213103618473,
"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": 0.022663925250526518,
"cast/masked_wrong_good_token_rate": 0.233812223887071,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0026864374685828807,
"clip_ratio/high_mean": 0.0026864374685828807,
"clip_ratio/low_mean": 0.0008685597786097787,
"clip_ratio/low_min": 0.0008685597786097787,
"clip_ratio/region_mean": 0.00355499724537367,
"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": 0.26525850617326796,
"epoch": 0.016721757728951972,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.06789598613977432,
"learning_rate": 9.944908937714692e-07,
"loss": 0.1334284096956253,
"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": 2.4271950721740723,
"sampling/importance_sampling_ratio/mean": 0.717940628528595,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.188699722290039,
"sampling/sampling_logp_difference/mean": 0.009960900992155075,
"step": 86,
"step_time": 260.51356751893763
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.0,
"cast/masked_correct_bad_token_rate": 0.08579519297927618,
"cast/masked_gap_abs_mean": 0.008579092216677964,
"cast/masked_gap_mean": 0.0001988101680865384,
"cast/masked_negative_base_rate": 0.7286968003027141,
"cast/masked_positive_base_rate": 0.2713031987659633,
"cast/masked_sign_flipped_rate": 0.2685903627425432,
"cast/masked_teacher_negative_rate": 0.32758610090240836,
"cast/masked_teacher_positive_rate": 0.245617866050452,
"cast/masked_token_adv_abs_mean": 0.6674297749996185,
"cast/masked_token_adv_max": 0.9270736500620842,
"cast/masked_token_adv_mean": -0.08793177627376281,
"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": 0.021911724732490256,
"cast/masked_wrong_good_token_rate": 0.18279516999609768,
"cast/mixed_group_rate": 1.0,
"clip_ratio/high_max": 0.002619505328766536,
"clip_ratio/high_mean": 0.002619505328766536,
"clip_ratio/low_mean": 0.0008974117029083573,
"clip_ratio/low_min": 0.0008974117029083573,
"clip_ratio/region_mean": 0.0035169170369044878,
"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": 0.21606561169028282,
"epoch": 0.01691619677231188,
"frac_reward_zero_std": 0.0,
"grad_norm": 0.04831128939986229,
"learning_rate": 9.94426080757016e-07,
"loss": 0.002982978243380785,
"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": 2.493037462234497,
"sampling/importance_sampling_ratio/mean": 0.5078309774398804,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.0082676410675049,
"sampling/sampling_logp_difference/mean": 0.008792135864496231,
"step": 87,
"step_time": 168.73832181989565
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.06076537445187569,
"cast/masked_gap_abs_mean": 0.009185772389173508,
"cast/masked_gap_mean": 0.00021652561065366172,
"cast/masked_negative_base_rate": 0.7918969644233584,
"cast/masked_positive_base_rate": 0.2081030271947384,
"cast/masked_sign_flipped_rate": 0.28980998508632183,
"cast/masked_teacher_negative_rate": 0.35932608414441347,
"cast/masked_teacher_positive_rate": 0.27482148772105575,
"cast/masked_token_adv_abs_mean": 0.8111578412353992,
"cast/masked_token_adv_max": 1.0428476529195905,
"cast/masked_token_adv_mean": -0.2148764255689457,
"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": 0.02221666823606938,
"cast/masked_wrong_good_token_rate": 0.2290446124970913,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0024757690262049437,
"clip_ratio/high_mean": 0.0024757690262049437,
"clip_ratio/low_mean": 0.0006759468146810832,
"clip_ratio/low_min": 0.0006759468146810832,
"clip_ratio/region_mean": 0.0031517158349743113,
"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": 0.27089800825342536,
"epoch": 0.017110635815671786,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.05411742627620697,
"learning_rate": 9.943612677425626e-07,
"loss": 0.0969361886382103,
"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": 2.43574857711792,
"sampling/importance_sampling_ratio/mean": 0.5363961458206177,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.08536958694458,
"sampling/sampling_logp_difference/mean": 0.009938384406268597,
"step": 88,
"step_time": 219.712468052865
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.11339306831359863,
"cast/masked_gap_abs_mean": 0.01000705080514308,
"cast/masked_gap_mean": -1.3994663689231857e-05,
"cast/masked_negative_base_rate": 0.639074582606554,
"cast/masked_positive_base_rate": 0.36092541459947824,
"cast/masked_sign_flipped_rate": 0.31359641067683697,
"cast/masked_teacher_negative_rate": 0.3593245632946491,
"cast/masked_teacher_positive_rate": 0.28477634070441127,
"cast/masked_token_adv_abs_mean": 0.901396943256259,
"cast/masked_token_adv_max": 1.0538752600550652,
"cast/masked_token_adv_mean": -0.10411886778911139,
"cast/masked_token_adv_min": -1.1810476630926132,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02348976588109508,
"cast/masked_wrong_good_token_rate": 0.20020334143191576,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.00259397590343724,
"clip_ratio/high_mean": 0.00259397590343724,
"clip_ratio/low_mean": 0.0011644536898529623,
"clip_ratio/low_min": 0.0011644536898529623,
"clip_ratio/region_mean": 0.0037584295860142447,
"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": 0.2688576770015061,
"epoch": 0.017305074859031693,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.06850066035985947,
"learning_rate": 9.942964547281093e-07,
"loss": 0.04414565861225128,
"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": 2.849586009979248,
"sampling/importance_sampling_ratio/mean": 0.6507673263549805,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.7456774711608887,
"sampling/sampling_logp_difference/mean": 0.010913350619375706,
"step": 89,
"step_time": 91.29102069692453
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.12687632953748107,
"cast/masked_gap_abs_mean": 0.008886618357792031,
"cast/masked_gap_mean": -5.071717936289133e-06,
"cast/masked_negative_base_rate": 0.5812701564282179,
"cast/masked_positive_base_rate": 0.4187298398464918,
"cast/masked_sign_flipped_rate": 0.3068250613287091,
"cast/masked_teacher_negative_rate": 0.3493947950191796,
"cast/masked_teacher_positive_rate": 0.26596179069019854,
"cast/masked_token_adv_abs_mean": 0.7255755281075835,
"cast/masked_token_adv_max": 0.946481442078948,
"cast/masked_token_adv_mean": -0.03937654163019033,
"cast/masked_token_adv_min": -1.0425376687198877,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.021762722899438813,
"cast/masked_wrong_good_token_rate": 0.17994873179122806,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0024196046879296773,
"clip_ratio/high_mean": 0.0024196046879296773,
"clip_ratio/low_mean": 0.0007985762917996908,
"clip_ratio/low_min": 0.0007985762917996908,
"clip_ratio/region_mean": 0.003218181002921483,
"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": 0.2483805363299325,
"epoch": 0.0174995139023916,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.07030467689037323,
"learning_rate": 9.94231641713656e-07,
"loss": 0.04425030201673508,
"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": 2.773536443710327,
"sampling/importance_sampling_ratio/mean": 0.6696163415908813,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.4419617652893066,
"sampling/sampling_logp_difference/mean": 0.008943975903093815,
"step": 90,
"step_time": 260.0567855599511
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.09866937703918666,
"cast/masked_gap_abs_mean": 0.00924397402559407,
"cast/masked_gap_mean": -0.00010538753258515499,
"cast/masked_negative_base_rate": 0.710356798954308,
"cast/masked_positive_base_rate": 0.2896431968547404,
"cast/masked_sign_flipped_rate": 0.2900167121551931,
"cast/masked_teacher_negative_rate": 0.3465316016227007,
"cast/masked_teacher_positive_rate": 0.2641274812631309,
"cast/masked_token_adv_abs_mean": 0.9161017686128616,
"cast/masked_token_adv_max": 1.0278681861236691,
"cast/masked_token_adv_mean": -0.21711354769649915,
"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": 0.022664154646918178,
"cast/masked_wrong_good_token_rate": 0.19134733523242176,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.002849064185284078,
"clip_ratio/high_mean": 0.002849064185284078,
"clip_ratio/low_mean": 0.0011314368684907095,
"clip_ratio/low_min": 0.0011314368684907095,
"clip_ratio/region_mean": 0.003980501063779229,
"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": 0.24702822184190154,
"epoch": 0.017693952945751507,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.09132314473390579,
"learning_rate": 9.941668286992027e-07,
"loss": 0.15841123461723328,
"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": 2.945580244064331,
"sampling/importance_sampling_ratio/mean": 0.633141279220581,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.5736541748046875,
"sampling/sampling_logp_difference/mean": 0.010127313435077667,
"step": 91,
"step_time": 142.7750892371114
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.5,
"cast/masked_correct_bad_token_rate": 0.038458642084151506,
"cast/masked_gap_abs_mean": 0.008925680434913374,
"cast/masked_gap_mean": 9.999083420098032e-05,
"cast/masked_negative_base_rate": 0.8886058144271374,
"cast/masked_positive_base_rate": 0.11139418696984649,
"cast/masked_sign_flipped_rate": 0.27386286575347185,
"cast/masked_teacher_negative_rate": 0.361261117272079,
"cast/masked_teacher_positive_rate": 0.2655253130942583,
"cast/masked_token_adv_abs_mean": 0.8419493464753032,
"cast/masked_token_adv_max": 1.0489782318472862,
"cast/masked_token_adv_mean": -0.320329992857296,
"cast/masked_token_adv_min": -1.1693841591477394,
"cast/masked_token_weight_max": 1.1997498981654644,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.0219340932671912,
"cast/masked_wrong_good_token_rate": 0.23540422460064292,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.002197670484747505,
"clip_ratio/high_mean": 0.002197670484747505,
"clip_ratio/low_mean": 0.000982873527391348,
"clip_ratio/low_min": 0.000982873527391348,
"clip_ratio/region_mean": 0.0031805440194148105,
"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": 0.26104152761399746,
"epoch": 0.017888391989111414,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.04497107118368149,
"learning_rate": 9.941020156847494e-07,
"loss": 0.16532301902770996,
"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": 2.2265524864196777,
"sampling/importance_sampling_ratio/mean": 0.5417255163192749,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 3.218289852142334,
"sampling/sampling_logp_difference/mean": 0.009679955430328846,
"step": 92,
"step_time": 268.42871981399367
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.092274428345263,
"cast/masked_gap_abs_mean": 0.009401339819305576,
"cast/masked_gap_mean": 2.123871081494144e-05,
"cast/masked_negative_base_rate": 0.7363050412386656,
"cast/masked_positive_base_rate": 0.2636949513107538,
"cast/masked_sign_flipped_rate": 0.3053069170564413,
"cast/masked_teacher_negative_rate": 0.3749164789915085,
"cast/masked_teacher_positive_rate": 0.27714605862274766,
"cast/masked_token_adv_abs_mean": 0.8207710944116116,
"cast/masked_token_adv_max": 1.029025984928012,
"cast/masked_token_adv_mean": -0.21107484656386077,
"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": 0.022795315308030695,
"cast/masked_wrong_good_token_rate": 0.21303248964250088,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0028924201578774955,
"clip_ratio/high_mean": 0.0028924201578774955,
"clip_ratio/low_mean": 0.0013454050258587813,
"clip_ratio/low_min": 0.0013454050258587813,
"clip_ratio/region_mean": 0.00423782517464133,
"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": 0.25561659410595894,
"epoch": 0.01808283103247132,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.058427274227142334,
"learning_rate": 9.940372026702961e-07,
"loss": 0.15061025321483612,
"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": 2.4688496589660645,
"sampling/importance_sampling_ratio/mean": 0.6942985653877258,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8087838888168335,
"sampling/sampling_logp_difference/mean": 0.010002934373915195,
"step": 93,
"step_time": 261.94198310686625
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.11378976819105446,
"cast/masked_gap_abs_mean": 0.009574561670888215,
"cast/masked_gap_mean": 0.0002064939017216716,
"cast/masked_negative_base_rate": 0.6694277804344893,
"cast/masked_positive_base_rate": 0.33057222329080105,
"cast/masked_sign_flipped_rate": 0.289237376768142,
"cast/masked_teacher_negative_rate": 0.35453967191278934,
"cast/masked_teacher_positive_rate": 0.2580291721969843,
"cast/masked_token_adv_abs_mean": 0.7310415310785174,
"cast/masked_token_adv_max": 0.9746227925643325,
"cast/masked_token_adv_mean": -0.170814230704309,
"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": 0.023375735501758754,
"cast/masked_wrong_good_token_rate": 0.17544761067256331,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.003184773302564281,
"clip_ratio/high_mean": 0.003184773302564281,
"clip_ratio/low_mean": 0.0010147853554371977,
"clip_ratio/low_min": 0.0010147853554371977,
"clip_ratio/region_mean": 0.004199558672553394,
"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": 0.22595580387860537,
"epoch": 0.01827727007583123,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.07323954254388809,
"learning_rate": 9.939723896558428e-07,
"loss": 0.11927018314599991,
"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": 2.7372963428497314,
"sampling/importance_sampling_ratio/mean": 0.6177743673324585,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 3.4453470706939697,
"sampling/sampling_logp_difference/mean": 0.010276490822434425,
"step": 94,
"step_time": 89.95019250496989
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.07152324414346367,
"cast/masked_gap_abs_mean": 0.007471306424122304,
"cast/masked_gap_mean": -0.00016201107001734272,
"cast/masked_negative_base_rate": 0.7589047974906862,
"cast/masked_positive_base_rate": 0.24109520157799125,
"cast/masked_sign_flipped_rate": 0.25055271480232477,
"cast/masked_teacher_negative_rate": 0.3359938617795706,
"cast/masked_teacher_positive_rate": 0.22934252815321088,
"cast/masked_token_adv_abs_mean": 0.9699990749359131,
"cast/masked_token_adv_max": 1.0741031169891357,
"cast/masked_token_adv_mean": -0.3136908619198948,
"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": 0.020754976198077202,
"cast/masked_wrong_good_token_rate": 0.179029471706599,
"cast/mixed_group_rate": 0.375,
"clip_ratio/high_max": 0.0019598948365455726,
"clip_ratio/high_mean": 0.0019598948365455726,
"clip_ratio/low_mean": 0.0010666817506717052,
"clip_ratio/low_min": 0.0010666817506717052,
"clip_ratio/region_mean": 0.003026576559932437,
"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": 0.1995403692126274,
"epoch": 0.018471709119191132,
"frac_reward_zero_std": 0.625,
"grad_norm": 0.06696436554193497,
"learning_rate": 9.939075766413895e-07,
"loss": 0.10931354761123657,
"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": 2.9910106658935547,
"sampling/importance_sampling_ratio/mean": 0.7083587646484375,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.2288601398468018,
"sampling/sampling_logp_difference/mean": 0.008089183829724789,
"step": 95,
"step_time": 144.69437986711273
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.07979841227643192,
"cast/masked_gap_abs_mean": 0.008250523344031535,
"cast/masked_gap_mean": 0.00014424092503872998,
"cast/masked_negative_base_rate": 0.7598645752295852,
"cast/masked_positive_base_rate": 0.24013542803004384,
"cast/masked_sign_flipped_rate": 0.26457509212195873,
"cast/masked_teacher_negative_rate": 0.34668746404349804,
"cast/masked_teacher_positive_rate": 0.23722122889012098,
"cast/masked_token_adv_abs_mean": 0.7133157830685377,
"cast/masked_token_adv_max": 0.9309405395761132,
"cast/masked_token_adv_mean": -0.19488177332095802,
"cast/masked_token_adv_min": -1.0261680008843541,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.02180574054364115,
"cast/masked_wrong_good_token_rate": 0.1847766803111881,
"cast/mixed_group_rate": 0.75,
"clip_ratio/high_max": 0.0028706349676212994,
"clip_ratio/high_mean": 0.0028706349676212994,
"clip_ratio/low_mean": 0.0010538939695834415,
"clip_ratio/low_min": 0.0010538939695834415,
"clip_ratio/region_mean": 0.003924528929928783,
"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": 0.20041491044685245,
"epoch": 0.01866614816255104,
"frac_reward_zero_std": 0.25,
"grad_norm": 0.04806852713227272,
"learning_rate": 9.938427636269362e-07,
"loss": 0.11060790717601776,
"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": 2.549257755279541,
"sampling/importance_sampling_ratio/mean": 0.5646318197250366,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.9684344530105591,
"sampling/sampling_logp_difference/mean": 0.008750040084123611,
"step": 96,
"step_time": 111.29640703392215
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.375,
"cast/masked_correct_bad_token_rate": 0.08199123607482761,
"cast/masked_gap_abs_mean": 0.008848114768625237,
"cast/masked_gap_mean": 4.0174430978368036e-05,
"cast/masked_negative_base_rate": 0.7210455322638154,
"cast/masked_positive_base_rate": 0.2789544726256281,
"cast/masked_sign_flipped_rate": 0.2801753976382315,
"cast/masked_teacher_negative_rate": 0.35620984621345997,
"cast/masked_teacher_positive_rate": 0.25605650758370757,
"cast/masked_token_adv_abs_mean": 0.8406227910891175,
"cast/masked_token_adv_max": 1.0085688102990389,
"cast/masked_token_adv_mean": -0.206000157922972,
"cast/masked_token_adv_min": -1.119497599080205,
"cast/masked_token_weight_max": 1.2000000476837158,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.022327656857669353,
"cast/masked_wrong_good_token_rate": 0.19818416028283536,
"cast/mixed_group_rate": 0.5,
"clip_ratio/high_max": 0.002668595563591225,
"clip_ratio/high_mean": 0.002668595563591225,
"clip_ratio/low_mean": 0.0009567429406160954,
"clip_ratio/low_min": 0.0009567429406160954,
"clip_ratio/region_mean": 0.0036253385005693417,
"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": 0.23030728101730347,
"epoch": 0.018860587205910946,
"frac_reward_zero_std": 0.5,
"grad_norm": 0.08365492522716522,
"learning_rate": 9.93777950612483e-07,
"loss": 0.15895135700702667,
"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": 2.928277015686035,
"sampling/importance_sampling_ratio/mean": 0.6390383839607239,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 1.2170898914337158,
"sampling/sampling_logp_difference/mean": 0.01011976320296526,
"step": 97,
"step_time": 221.53683824714972
},
{
"cast/all_correct_group_rate": 0.0,
"cast/all_wrong_group_rate": 0.125,
"cast/masked_correct_bad_token_rate": 0.09774420084431767,
"cast/masked_gap_abs_mean": 0.007168954485678114,
"cast/masked_gap_mean": 3.66045542818938e-05,
"cast/masked_negative_base_rate": 0.6617059875279665,
"cast/masked_positive_base_rate": 0.3382940050214529,
"cast/masked_sign_flipped_rate": 0.2535724211484194,
"cast/masked_teacher_negative_rate": 0.3184695988893509,
"cast/masked_teacher_positive_rate": 0.21595858503133059,
"cast/masked_token_adv_abs_mean": 0.76718576438725,
"cast/masked_token_adv_max": 0.9555053869262338,
"cast/masked_token_adv_mean": -0.12650088325608522,
"cast/masked_token_adv_min": -1.0169959915801883,
"cast/masked_token_weight_max": 1.199985958635807,
"cast/masked_token_weight_min": 1.0,
"cast/masked_token_weight_std": 0.019987784326076508,
"cast/masked_wrong_good_token_rate": 0.15582822007127106,
"cast/mixed_group_rate": 0.875,
"clip_ratio/high_max": 0.002396813204541104,
"clip_ratio/high_mean": 0.002396813204541104,
"clip_ratio/low_mean": 0.0009007182866298535,
"clip_ratio/low_min": 0.0009007182866298535,
"clip_ratio/region_mean": 0.0032975315043586306,
"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": 0.1882087243720889,
"epoch": 0.019055026249270853,
"frac_reward_zero_std": 0.125,
"grad_norm": 0.03311840072274208,
"learning_rate": 9.937131375980296e-07,
"loss": 0.0457211472094059,
"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": 2.353341579437256,
"sampling/importance_sampling_ratio/mean": 0.7049669623374939,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 2.762234926223755,
"sampling/sampling_logp_difference/mean": 0.00709116505458951,
"step": 98,
"step_time": 274.6682433609676
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.25,
"cast/masked_correct_bad_token_rate": 0.11644318234175444,
"cast/masked_gap_abs_mean": 0.007959328388096765,
"cast/masked_gap_mean": 2.3248185698321322e-05,
"cast/masked_negative_base_rate": 0.6527670696377754,
"cast/masked_positive_base_rate": 0.347232922911644,
"cast/masked_sign_flipped_rate": 0.27401354257017374,
"cast/masked_teacher_negative_rate": 0.342509794048965,
"cast/masked_teacher_positive_rate": 0.23756515700370073,
"cast/masked_token_adv_abs_mean": 0.7814900306984782,
"cast/masked_token_adv_max": 0.9604845317080617,
"cast/masked_token_adv_mean": -0.1378151071839966,
"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": 0.021254867082461715,
"cast/masked_wrong_good_token_rate": 0.15757036046124995,
"cast/mixed_group_rate": 0.625,
"clip_ratio/high_max": 0.0022138517815619707,
"clip_ratio/high_mean": 0.0022138517815619707,
"clip_ratio/low_mean": 0.0007013471295067575,
"clip_ratio/low_min": 0.0007013471295067575,
"clip_ratio/region_mean": 0.0029151989256206434,
"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": 0.21688769524917006,
"epoch": 0.01924946529263076,
"frac_reward_zero_std": 0.375,
"grad_norm": 0.04235367849469185,
"learning_rate": 9.936483245835763e-07,
"loss": 0.09898544102907181,
"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": 2.9712250232696533,
"sampling/importance_sampling_ratio/mean": 0.5888824462890625,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.7436556816101074,
"sampling/sampling_logp_difference/mean": 0.008226566016674042,
"step": 99,
"step_time": 239.51719569493434
},
{
"cast/all_correct_group_rate": 0.125,
"cast/all_wrong_group_rate": 0.625,
"cast/masked_correct_bad_token_rate": 0.10193813988007605,
"cast/masked_gap_abs_mean": 0.010380031919339672,
"cast/masked_gap_mean": 0.00011079045833639611,
"cast/masked_negative_base_rate": 0.7036592029035091,
"cast/masked_positive_base_rate": 0.29634079057723284,
"cast/masked_sign_flipped_rate": 0.3128499398007989,
"cast/masked_teacher_negative_rate": 0.3767198156565428,
"cast/masked_teacher_positive_rate": 0.2836825675331056,
"cast/masked_token_adv_abs_mean": 0.8960793456062675,
"cast/masked_token_adv_max": 1.0169453416019678,
"cast/masked_token_adv_mean": -0.23079512373078614,
"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": 0.02410770155256614,
"cast/masked_wrong_good_token_rate": 0.21091179666109383,
"cast/mixed_group_rate": 0.25,
"clip_ratio/high_max": 0.0032207840340561233,
"clip_ratio/high_mean": 0.0032207840340561233,
"clip_ratio/low_mean": 0.0010769990331027657,
"clip_ratio/low_min": 0.0010769990331027657,
"clip_ratio/region_mean": 0.004297783052606974,
"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": 0.2598504894413054,
"epoch": 0.019443904335990667,
"frac_reward_zero_std": 0.75,
"grad_norm": 0.08796551823616028,
"learning_rate": 9.93583511569123e-07,
"loss": 0.11268557608127594,
"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": 2.4312832355499268,
"sampling/importance_sampling_ratio/mean": 0.6172981262207031,
"sampling/importance_sampling_ratio/min": 0.0,
"sampling/sampling_logp_difference/max": 0.8719643354415894,
"sampling/sampling_logp_difference/mean": 0.011135343462228775,
"step": 100,
"step_time": 78.66995777995908
}
],
"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
}