{ "best_global_step": null, "best_metric": null, "best_model_checkpoint": null, "epoch": 1.0, "eval_steps": 100, "global_step": 681, "is_hyper_param_search": false, "is_local_process_zero": true, "is_world_process_zero": true, "log_history": [ { "beta_dpo/beta": 0.1005711555480957, "beta_dpo/beta_margin_grad_mean": -0.49939221143722534, "beta_dpo/beta_margin_grad_std": 0.007166189141571522, "beta_dpo/beta_margin_mean": 0.002432748908177018, "beta_dpo/beta_margin_std": 0.028674542903900146, "beta_dpo/beta_used": 0.1005711555480957, "beta_dpo/beta_used_raw": 0.1005711555480957, "beta_dpo/gap_mean": 0.0036213158164173365, "beta_dpo/gap_std": 0.04173510894179344, "beta_dpo/loss_margin_mean": 0.024216145277023315, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.0014684287812041115, "grad_norm": 150.13343811035156, "learning_rate": 0.0, "logits/chosen": -3.1338672637939453, "logits/rejected": -3.1045522689819336, "loss": 1.385, "step": 1 }, { "beta_dpo/beta": 0.10121786594390869, "beta_dpo/beta_margin_grad_mean": -0.5002189874649048, "beta_dpo/beta_margin_grad_std": 0.006193624343723059, "beta_dpo/beta_margin_mean": -0.0008768404368311167, "beta_dpo/beta_margin_std": 0.02477947063744068, "beta_dpo/beta_used": 0.10121786594390869, "beta_dpo/beta_used_raw": 0.10121786594390869, "beta_dpo/gap_mean": 0.0025136363692581654, "beta_dpo/gap_std": 0.08383626490831375, "beta_dpo/loss_margin_mean": -0.008541211485862732, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.002936857562408223, "grad_norm": 156.69430541992188, "learning_rate": 7.246376811594203e-09, "logits/chosen": -3.149817943572998, "logits/rejected": -3.133805751800537, "loss": 1.3838, "step": 2 }, { "beta_dpo/beta": 0.10055717825889587, "beta_dpo/beta_margin_grad_mean": -0.4997645616531372, "beta_dpo/beta_margin_grad_std": 0.006223098374903202, "beta_dpo/beta_margin_mean": 0.0009417525725439191, "beta_dpo/beta_margin_std": 0.024896962568163872, "beta_dpo/beta_used": 0.10055717825889587, "beta_dpo/beta_used_raw": 0.10055717825889587, "beta_dpo/gap_mean": 0.0029192278161644936, "beta_dpo/gap_std": 0.113316610455513, "beta_dpo/loss_margin_mean": 0.009323984384536743, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.004405286343612335, "grad_norm": 169.9915771484375, "learning_rate": 1.4492753623188406e-08, "logits/chosen": -3.1339285373687744, "logits/rejected": -3.0974249839782715, "loss": 1.3851, "step": 3 }, { "beta_dpo/beta": 0.10269482433795929, "beta_dpo/beta_margin_grad_mean": -0.49976256489753723, "beta_dpo/beta_margin_grad_std": 0.006724101025611162, "beta_dpo/beta_margin_mean": 0.0009488743380643427, "beta_dpo/beta_margin_std": 0.026902994140982628, "beta_dpo/beta_used": 0.10269482433795929, "beta_dpo/beta_used_raw": 0.10269482433795929, "beta_dpo/gap_mean": 0.0034820283763110638, "beta_dpo/gap_std": 0.14156511425971985, "beta_dpo/loss_margin_mean": 0.00913846492767334, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.005873715124816446, "grad_norm": 153.58120727539062, "learning_rate": 2.1739130434782606e-08, "logits/chosen": -3.124056816101074, "logits/rejected": -3.1173763275146484, "loss": 1.3814, "step": 4 }, { "beta_dpo/beta": 0.09814765304327011, "beta_dpo/beta_margin_grad_mean": -0.5008248090744019, "beta_dpo/beta_margin_grad_std": 0.005996009334921837, "beta_dpo/beta_margin_mean": -0.00329973129555583, "beta_dpo/beta_margin_std": 0.02398756332695484, "beta_dpo/beta_used": 0.09814765304327011, "beta_dpo/beta_used_raw": 0.09814765304327011, "beta_dpo/gap_mean": 0.0016051515704020858, "beta_dpo/gap_std": 0.1615678071975708, "beta_dpo/loss_margin_mean": -0.03306567668914795, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.007342143906020558, "grad_norm": 179.1089324951172, "learning_rate": 2.898550724637681e-08, "logits/chosen": -3.109281063079834, "logits/rejected": -3.089329481124878, "loss": 1.3892, "step": 5 }, { "beta_dpo/beta": 0.09695610404014587, "beta_dpo/beta_margin_grad_mean": -0.500086784362793, "beta_dpo/beta_margin_grad_std": 0.006598716601729393, "beta_dpo/beta_margin_mean": -0.00034629780566319823, "beta_dpo/beta_margin_std": 0.02640008181333542, "beta_dpo/beta_used": 0.09695610404014587, "beta_dpo/beta_used_raw": 0.09695610404014587, "beta_dpo/gap_mean": -0.002422480145469308, "beta_dpo/gap_std": 0.18281012773513794, "beta_dpo/loss_margin_mean": -0.00355510413646698, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.00881057268722467, "grad_norm": 171.80946350097656, "learning_rate": 3.6231884057971014e-08, "logits/chosen": -3.1225814819335938, "logits/rejected": -3.0773398876190186, "loss": 1.3915, "step": 6 }, { "beta_dpo/beta": 0.10045399516820908, "beta_dpo/beta_margin_grad_mean": -0.49887970089912415, "beta_dpo/beta_margin_grad_std": 0.005884926300495863, "beta_dpo/beta_margin_mean": 0.004481919575482607, "beta_dpo/beta_margin_std": 0.02354278229176998, "beta_dpo/beta_used": 0.10045399516820908, "beta_dpo/beta_used_raw": 0.10045399516820908, "beta_dpo/gap_mean": 0.004135521594434977, "beta_dpo/gap_std": 0.1954803466796875, "beta_dpo/loss_margin_mean": 0.04471883177757263, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.010279001468428781, "grad_norm": 165.5727081298828, "learning_rate": 4.347826086956521e-08, "logits/chosen": -3.1420326232910156, "logits/rejected": -3.1474711894989014, "loss": 1.3852, "step": 7 }, { "beta_dpo/beta": 0.09605514258146286, "beta_dpo/beta_margin_grad_mean": -0.5012519359588623, "beta_dpo/beta_margin_grad_std": 0.006095360033214092, "beta_dpo/beta_margin_mean": -0.005007945001125336, "beta_dpo/beta_margin_std": 0.02438473515212536, "beta_dpo/beta_used": 0.09605514258146286, "beta_dpo/beta_used_raw": 0.09605514258146286, "beta_dpo/gap_mean": -0.0014673962723463774, "beta_dpo/gap_std": 0.20564857125282288, "beta_dpo/loss_margin_mean": -0.052922144532203674, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.011747430249632892, "grad_norm": 163.31930541992188, "learning_rate": 5.0724637681159424e-08, "logits/chosen": -3.198502540588379, "logits/rejected": -3.1550021171569824, "loss": 1.3927, "step": 8 }, { "beta_dpo/beta": 0.10017342120409012, "beta_dpo/beta_margin_grad_mean": -0.5000122785568237, "beta_dpo/beta_margin_grad_std": 0.006785591598600149, "beta_dpo/beta_margin_mean": -4.894528683507815e-05, "beta_dpo/beta_margin_std": 0.02714758738875389, "beta_dpo/beta_used": 0.10017342120409012, "beta_dpo/beta_used_raw": 0.10017342120409012, "beta_dpo/gap_mean": -0.007042712531983852, "beta_dpo/gap_std": 0.21777689456939697, "beta_dpo/loss_margin_mean": -0.0016328543424606323, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.013215859030837005, "grad_norm": 181.66322326660156, "learning_rate": 5.797101449275362e-08, "logits/chosen": -3.1416659355163574, "logits/rejected": -3.1250786781311035, "loss": 1.3867, "step": 9 }, { "beta_dpo/beta": 0.09914623945951462, "beta_dpo/beta_margin_grad_mean": -0.5005456805229187, "beta_dpo/beta_margin_grad_std": 0.005261780694127083, "beta_dpo/beta_margin_mean": -0.0021829898469150066, "beta_dpo/beta_margin_std": 0.021049022674560547, "beta_dpo/beta_used": 0.09914623945951462, "beta_dpo/beta_used_raw": 0.09914623945951462, "beta_dpo/gap_mean": -0.006503046490252018, "beta_dpo/gap_std": 0.21944838762283325, "beta_dpo/loss_margin_mean": -0.02211102843284607, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.014684287812041116, "grad_norm": 168.22305297851562, "learning_rate": 6.521739130434782e-08, "logits/chosen": -3.10261869430542, "logits/rejected": -3.080986499786377, "loss": 1.3884, "step": 10 }, { "beta_dpo/beta": 0.10375332087278366, "beta_dpo/beta_margin_grad_mean": -0.49827736616134644, "beta_dpo/beta_margin_grad_std": 0.006540252361446619, "beta_dpo/beta_margin_mean": 0.00689197052270174, "beta_dpo/beta_margin_std": 0.026165250688791275, "beta_dpo/beta_used": 0.10375332087278366, "beta_dpo/beta_used_raw": 0.10375332087278366, "beta_dpo/gap_mean": 0.002864137524738908, "beta_dpo/gap_std": 0.22579887509346008, "beta_dpo/loss_margin_mean": 0.06643152236938477, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.016152716593245228, "grad_norm": 186.03494262695312, "learning_rate": 7.246376811594203e-08, "logits/chosen": -3.155540943145752, "logits/rejected": -3.142815113067627, "loss": 1.3796, "step": 11 }, { "beta_dpo/beta": 0.10144880414009094, "beta_dpo/beta_margin_grad_mean": -0.49903959035873413, "beta_dpo/beta_margin_grad_std": 0.006939120590686798, "beta_dpo/beta_margin_mean": 0.003841817146167159, "beta_dpo/beta_margin_std": 0.02776145376265049, "beta_dpo/beta_used": 0.10144880414009094, "beta_dpo/beta_used_raw": 0.10144880414009094, "beta_dpo/gap_mean": 0.01069832406938076, "beta_dpo/gap_std": 0.23313455283641815, "beta_dpo/loss_margin_mean": 0.037982091307640076, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.01762114537444934, "grad_norm": 202.00311279296875, "learning_rate": 7.971014492753623e-08, "logits/chosen": -3.1653075218200684, "logits/rejected": -3.153486967086792, "loss": 1.3828, "step": 12 }, { "beta_dpo/beta": 0.10425792634487152, "beta_dpo/beta_margin_grad_mean": -0.4985834062099457, "beta_dpo/beta_margin_grad_std": 0.007317332550883293, "beta_dpo/beta_margin_mean": 0.005666609387844801, "beta_dpo/beta_margin_std": 0.02927694283425808, "beta_dpo/beta_used": 0.10425792634487152, "beta_dpo/beta_used_raw": 0.10425792634487152, "beta_dpo/gap_mean": 0.01822437345981598, "beta_dpo/gap_std": 0.24303942918777466, "beta_dpo/loss_margin_mean": 0.054477810859680176, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.01908957415565345, "grad_norm": 165.83859252929688, "learning_rate": 8.695652173913042e-08, "logits/chosen": -3.167670726776123, "logits/rejected": -3.135024070739746, "loss": 1.377, "step": 13 }, { "beta_dpo/beta": 0.10415012389421463, "beta_dpo/beta_margin_grad_mean": -0.4957675039768219, "beta_dpo/beta_margin_grad_std": 0.00820655096322298, "beta_dpo/beta_margin_mean": 0.016939442604780197, "beta_dpo/beta_margin_std": 0.03285452350974083, "beta_dpo/beta_used": 0.10415012389421463, "beta_dpo/beta_used_raw": 0.10415012389421463, "beta_dpo/gap_mean": 0.039918702095746994, "beta_dpo/gap_std": 0.2563033699989319, "beta_dpo/loss_margin_mean": 0.16253972053527832, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.020558002936857563, "grad_norm": 203.8751983642578, "learning_rate": 9.420289855072464e-08, "logits/chosen": -3.132795572280884, "logits/rejected": -3.1252737045288086, "loss": 1.3751, "step": 14 }, { "beta_dpo/beta": 0.10431154817342758, "beta_dpo/beta_margin_grad_mean": -0.4954010248184204, "beta_dpo/beta_margin_grad_std": 0.006738570984452963, "beta_dpo/beta_margin_mean": 0.018400877714157104, "beta_dpo/beta_margin_std": 0.026965487748384476, "beta_dpo/beta_used": 0.10431154817342758, "beta_dpo/beta_used_raw": 0.10431154817342758, "beta_dpo/gap_mean": 0.06402108818292618, "beta_dpo/gap_std": 0.2579445540904999, "beta_dpo/loss_margin_mean": 0.17562128603458405, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.022026431718061675, "grad_norm": 158.30233764648438, "learning_rate": 1.0144927536231885e-07, "logits/chosen": -3.1343517303466797, "logits/rejected": -3.124833106994629, "loss": 1.3722, "step": 15 }, { "beta_dpo/beta": 0.10335851460695267, "beta_dpo/beta_margin_grad_mean": -0.4952828586101532, "beta_dpo/beta_margin_grad_std": 0.007043875753879547, "beta_dpo/beta_margin_mean": 0.018874231725931168, "beta_dpo/beta_margin_std": 0.028187856078147888, "beta_dpo/beta_used": 0.10335851460695267, "beta_dpo/beta_used_raw": 0.10335851460695267, "beta_dpo/gap_mean": 0.08750247955322266, "beta_dpo/gap_std": 0.25985032320022583, "beta_dpo/loss_margin_mean": 0.1823047697544098, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.023494860499265784, "grad_norm": 162.9868621826172, "learning_rate": 1.0869565217391303e-07, "logits/chosen": -3.1242918968200684, "logits/rejected": -3.090561628341675, "loss": 1.3716, "step": 16 }, { "beta_dpo/beta": 0.10424471646547318, "beta_dpo/beta_margin_grad_mean": -0.4921805262565613, "beta_dpo/beta_margin_grad_std": 0.009317115880548954, "beta_dpo/beta_margin_mean": 0.031297143548727036, "beta_dpo/beta_margin_std": 0.037318408489227295, "beta_dpo/beta_used": 0.10424471646547318, "beta_dpo/beta_used_raw": 0.10424471646547318, "beta_dpo/gap_mean": 0.12156766653060913, "beta_dpo/gap_std": 0.273958683013916, "beta_dpo/loss_margin_mean": 0.2991063594818115, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.024963289280469897, "grad_norm": 171.33746337890625, "learning_rate": 1.1594202898550725e-07, "logits/chosen": -3.1508283615112305, "logits/rejected": -3.0991227626800537, "loss": 1.3663, "step": 17 }, { "beta_dpo/beta": 0.10741108655929565, "beta_dpo/beta_margin_grad_mean": -0.4899873435497284, "beta_dpo/beta_margin_grad_std": 0.009942273609340191, "beta_dpo/beta_margin_mean": 0.040072713047266006, "beta_dpo/beta_margin_std": 0.039800975471735, "beta_dpo/beta_used": 0.10741108655929565, "beta_dpo/beta_used_raw": 0.10741108655929565, "beta_dpo/gap_mean": 0.1662050485610962, "beta_dpo/gap_std": 0.2930639386177063, "beta_dpo/loss_margin_mean": 0.37321096658706665, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.02643171806167401, "grad_norm": 172.73681640625, "learning_rate": 1.2318840579710146e-07, "logits/chosen": -3.1059744358062744, "logits/rejected": -3.0668582916259766, "loss": 1.3556, "step": 18 }, { "beta_dpo/beta": 0.10548347979784012, "beta_dpo/beta_margin_grad_mean": -0.4869752824306488, "beta_dpo/beta_margin_grad_std": 0.01236420962959528, "beta_dpo/beta_margin_mean": 0.05215656757354736, "beta_dpo/beta_margin_std": 0.04959164187312126, "beta_dpo/beta_used": 0.10548347979784012, "beta_dpo/beta_used_raw": 0.10548347979784012, "beta_dpo/gap_mean": 0.21926948428153992, "beta_dpo/gap_std": 0.317290723323822, "beta_dpo/loss_margin_mean": 0.49080538749694824, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.027900146842878122, "grad_norm": 154.02589416503906, "learning_rate": 1.3043478260869563e-07, "logits/chosen": -3.137395143508911, "logits/rejected": -3.095036506652832, "loss": 1.3537, "step": 19 }, { "beta_dpo/beta": 0.10729610919952393, "beta_dpo/beta_margin_grad_mean": -0.4847802519798279, "beta_dpo/beta_margin_grad_std": 0.01249635312706232, "beta_dpo/beta_margin_mean": 0.06094222515821457, "beta_dpo/beta_margin_std": 0.05008528009057045, "beta_dpo/beta_used": 0.10729610919952393, "beta_dpo/beta_used_raw": 0.10729610919952393, "beta_dpo/gap_mean": 0.28570470213890076, "beta_dpo/gap_std": 0.34619662165641785, "beta_dpo/loss_margin_mean": 0.5685427784919739, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.02936857562408223, "grad_norm": 151.9236602783203, "learning_rate": 1.3768115942028986e-07, "logits/chosen": -3.151339530944824, "logits/rejected": -3.1141912937164307, "loss": 1.3433, "step": 20 }, { "beta_dpo/beta": 0.1047445610165596, "beta_dpo/beta_margin_grad_mean": -0.4856947958469391, "beta_dpo/beta_margin_grad_std": 0.015066784806549549, "beta_dpo/beta_margin_mean": 0.057291969656944275, "beta_dpo/beta_margin_std": 0.0604330375790596, "beta_dpo/beta_used": 0.1047445610165596, "beta_dpo/beta_used_raw": 0.1047445610165596, "beta_dpo/gap_mean": 0.3368791937828064, "beta_dpo/gap_std": 0.3872354030609131, "beta_dpo/loss_margin_mean": 0.5463607311248779, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.030837004405286344, "grad_norm": 139.0661163330078, "learning_rate": 1.4492753623188405e-07, "logits/chosen": -3.0985090732574463, "logits/rejected": -3.0805447101593018, "loss": 1.3432, "step": 21 }, { "beta_dpo/beta": 0.10737676173448563, "beta_dpo/beta_margin_grad_mean": -0.4813934862613678, "beta_dpo/beta_margin_grad_std": 0.013753235340118408, "beta_dpo/beta_margin_mean": 0.074525848031044, "beta_dpo/beta_margin_std": 0.055176492780447006, "beta_dpo/beta_used": 0.10737676173448563, "beta_dpo/beta_used_raw": 0.10737676173448563, "beta_dpo/gap_mean": 0.40355396270751953, "beta_dpo/gap_std": 0.4153197705745697, "beta_dpo/loss_margin_mean": 0.6897796988487244, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.032305433186490456, "grad_norm": 172.8441162109375, "learning_rate": 1.5217391304347825e-07, "logits/chosen": -3.1092402935028076, "logits/rejected": -3.09182071685791, "loss": 1.3307, "step": 22 }, { "beta_dpo/beta": 0.10904830694198608, "beta_dpo/beta_margin_grad_mean": -0.4761696755886078, "beta_dpo/beta_margin_grad_std": 0.021913398057222366, "beta_dpo/beta_margin_mean": 0.09563583135604858, "beta_dpo/beta_margin_std": 0.08828449249267578, "beta_dpo/beta_used": 0.10904830694198608, "beta_dpo/beta_used_raw": 0.10904830694198608, "beta_dpo/gap_mean": 0.4720858931541443, "beta_dpo/gap_std": 0.4659385085105896, "beta_dpo/loss_margin_mean": 0.8740031123161316, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.033773861967694566, "grad_norm": 156.1903076171875, "learning_rate": 1.5942028985507245e-07, "logits/chosen": -3.1103243827819824, "logits/rejected": -3.088109016418457, "loss": 1.3201, "step": 23 }, { "beta_dpo/beta": 0.11075541377067566, "beta_dpo/beta_margin_grad_mean": -0.471403568983078, "beta_dpo/beta_margin_grad_std": 0.021366121247410774, "beta_dpo/beta_margin_mean": 0.11474256962537766, "beta_dpo/beta_margin_std": 0.08595109730958939, "beta_dpo/beta_used": 0.11075541377067566, "beta_dpo/beta_used_raw": 0.11075541377067566, "beta_dpo/gap_mean": 0.578666090965271, "beta_dpo/gap_std": 0.5306879281997681, "beta_dpo/loss_margin_mean": 1.0362656116485596, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.03524229074889868, "grad_norm": 177.53704833984375, "learning_rate": 1.6666666666666665e-07, "logits/chosen": -3.085505723953247, "logits/rejected": -3.097721815109253, "loss": 1.305, "step": 24 }, { "beta_dpo/beta": 0.10202641040086746, "beta_dpo/beta_margin_grad_mean": -0.47696781158447266, "beta_dpo/beta_margin_grad_std": 0.02352500520646572, "beta_dpo/beta_margin_mean": 0.09244953840970993, "beta_dpo/beta_margin_std": 0.09470401704311371, "beta_dpo/beta_used": 0.10202641040086746, "beta_dpo/beta_used_raw": 0.10202641040086746, "beta_dpo/gap_mean": 0.6436024904251099, "beta_dpo/gap_std": 0.5977407693862915, "beta_dpo/loss_margin_mean": 0.9032971858978271, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.03671071953010279, "grad_norm": 133.51351928710938, "learning_rate": 1.7391304347826085e-07, "logits/chosen": -3.1361374855041504, "logits/rejected": -3.1234917640686035, "loss": 1.319, "step": 25 }, { "beta_dpo/beta": 0.11549623310565948, "beta_dpo/beta_margin_grad_mean": -0.453942209482193, "beta_dpo/beta_margin_grad_std": 0.04830946400761604, "beta_dpo/beta_margin_mean": 0.18718303740024567, "beta_dpo/beta_margin_std": 0.20004823803901672, "beta_dpo/beta_used": 0.11549623310565948, "beta_dpo/beta_used_raw": 0.11549623310565948, "beta_dpo/gap_mean": 0.7885938882827759, "beta_dpo/gap_std": 0.7514917850494385, "beta_dpo/loss_margin_mean": 1.6038861274719238, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.0381791483113069, "grad_norm": 163.3068389892578, "learning_rate": 1.8115942028985507e-07, "logits/chosen": -3.134460687637329, "logits/rejected": -3.140258550643921, "loss": 1.27, "step": 26 }, { "beta_dpo/beta": 0.1283809393644333, "beta_dpo/beta_margin_grad_mean": -0.4395935833454132, "beta_dpo/beta_margin_grad_std": 0.03977083042263985, "beta_dpo/beta_margin_mean": 0.24452631175518036, "beta_dpo/beta_margin_std": 0.16269095242023468, "beta_dpo/beta_used": 0.1283809393644333, "beta_dpo/beta_used_raw": 0.1283809393644333, "beta_dpo/gap_mean": 1.0029573440551758, "beta_dpo/gap_std": 0.8620645999908447, "beta_dpo/loss_margin_mean": 1.8610539436340332, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.039647577092511016, "grad_norm": 196.32107543945312, "learning_rate": 1.8840579710144927e-07, "logits/chosen": -3.0956711769104004, "logits/rejected": -3.0962371826171875, "loss": 1.2053, "step": 27 }, { "beta_dpo/beta": 0.12041008472442627, "beta_dpo/beta_margin_grad_mean": -0.44316551089286804, "beta_dpo/beta_margin_grad_std": 0.03816642612218857, "beta_dpo/beta_margin_mean": 0.2298518419265747, "beta_dpo/beta_margin_std": 0.15634086728096008, "beta_dpo/beta_used": 0.12041008472442627, "beta_dpo/beta_used_raw": 0.12041008472442627, "beta_dpo/gap_mean": 1.1611372232437134, "beta_dpo/gap_std": 0.9274260997772217, "beta_dpo/loss_margin_mean": 1.9079362154006958, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.041116005873715125, "grad_norm": 173.4256134033203, "learning_rate": 1.9565217391304347e-07, "logits/chosen": -3.1474990844726562, "logits/rejected": -3.1038711071014404, "loss": 1.2161, "step": 28 }, { "beta_dpo/beta": 0.12488596141338348, "beta_dpo/beta_margin_grad_mean": -0.43044498562812805, "beta_dpo/beta_margin_grad_std": 0.04702915996313095, "beta_dpo/beta_margin_mean": 0.2829776108264923, "beta_dpo/beta_margin_std": 0.19504757225513458, "beta_dpo/beta_used": 0.12488596141338348, "beta_dpo/beta_used_raw": 0.12488596141338348, "beta_dpo/gap_mean": 1.347472906112671, "beta_dpo/gap_std": 1.0383023023605347, "beta_dpo/loss_margin_mean": 2.274510145187378, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.042584434654919234, "grad_norm": 177.54214477539062, "learning_rate": 2.028985507246377e-07, "logits/chosen": -3.128300666809082, "logits/rejected": -3.105536937713623, "loss": 1.1833, "step": 29 }, { "beta_dpo/beta": 0.12252659350633621, "beta_dpo/beta_margin_grad_mean": -0.4280206561088562, "beta_dpo/beta_margin_grad_std": 0.05462159961462021, "beta_dpo/beta_margin_mean": 0.29425692558288574, "beta_dpo/beta_margin_std": 0.22929762303829193, "beta_dpo/beta_used": 0.12252659350633621, "beta_dpo/beta_used_raw": 0.12252659350633621, "beta_dpo/gap_mean": 1.5580029487609863, "beta_dpo/gap_std": 1.1984624862670898, "beta_dpo/loss_margin_mean": 2.384845018386841, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.04405286343612335, "grad_norm": 179.45391845703125, "learning_rate": 2.1014492753623187e-07, "logits/chosen": -3.109475612640381, "logits/rejected": -3.0789778232574463, "loss": 1.1672, "step": 30 }, { "beta_dpo/beta": 0.12765327095985413, "beta_dpo/beta_margin_grad_mean": -0.4269483983516693, "beta_dpo/beta_margin_grad_std": 0.06185346469283104, "beta_dpo/beta_margin_mean": 0.3004482090473175, "beta_dpo/beta_margin_std": 0.2656671404838562, "beta_dpo/beta_used": 0.12765327095985413, "beta_dpo/beta_used_raw": 0.12765327095985413, "beta_dpo/gap_mean": 1.7079854011535645, "beta_dpo/gap_std": 1.3484312295913696, "beta_dpo/loss_margin_mean": 2.3504257202148438, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.04552129221732746, "grad_norm": 166.8577880859375, "learning_rate": 2.1739130434782607e-07, "logits/chosen": -3.1314797401428223, "logits/rejected": -3.0868959426879883, "loss": 1.1353, "step": 31 }, { "beta_dpo/beta": 0.10552072525024414, "beta_dpo/beta_margin_grad_mean": -0.4227868914604187, "beta_dpo/beta_margin_grad_std": 0.06014800816774368, "beta_dpo/beta_margin_mean": 0.31740450859069824, "beta_dpo/beta_margin_std": 0.25505444407463074, "beta_dpo/beta_used": 0.10552072525024414, "beta_dpo/beta_used_raw": 0.10552072525024414, "beta_dpo/gap_mean": 1.9182875156402588, "beta_dpo/gap_std": 1.5188119411468506, "beta_dpo/loss_margin_mean": 2.994873046875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.04698972099853157, "grad_norm": 143.5096435546875, "learning_rate": 2.2463768115942027e-07, "logits/chosen": -3.1662778854370117, "logits/rejected": -3.149886131286621, "loss": 1.1878, "step": 32 }, { "beta_dpo/beta": 0.12314489483833313, "beta_dpo/beta_margin_grad_mean": -0.41513192653656006, "beta_dpo/beta_margin_grad_std": 0.07031543552875519, "beta_dpo/beta_margin_mean": 0.3495492935180664, "beta_dpo/beta_margin_std": 0.30640801787376404, "beta_dpo/beta_used": 0.12314489483833313, "beta_dpo/beta_used_raw": 0.12314489483833313, "beta_dpo/gap_mean": 2.081599235534668, "beta_dpo/gap_std": 1.7187418937683105, "beta_dpo/loss_margin_mean": 2.8416450023651123, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.048458149779735685, "grad_norm": 178.19117736816406, "learning_rate": 2.318840579710145e-07, "logits/chosen": -3.126152276992798, "logits/rejected": -3.110962152481079, "loss": 1.1123, "step": 33 }, { "beta_dpo/beta": 0.10388785600662231, "beta_dpo/beta_margin_grad_mean": -0.4212900698184967, "beta_dpo/beta_margin_grad_std": 0.0669008269906044, "beta_dpo/beta_margin_mean": 0.3246135115623474, "beta_dpo/beta_margin_std": 0.28224509954452515, "beta_dpo/beta_used": 0.10388785600662231, "beta_dpo/beta_used_raw": 0.10388785600662231, "beta_dpo/gap_mean": 2.2918877601623535, "beta_dpo/gap_std": 1.891815185546875, "beta_dpo/loss_margin_mean": 3.123857021331787, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.049926578560939794, "grad_norm": 142.10772705078125, "learning_rate": 2.391304347826087e-07, "logits/chosen": -3.1223511695861816, "logits/rejected": -3.1378636360168457, "loss": 1.1663, "step": 34 }, { "beta_dpo/beta": 0.1047118604183197, "beta_dpo/beta_margin_grad_mean": -0.40423837304115295, "beta_dpo/beta_margin_grad_std": 0.07508349418640137, "beta_dpo/beta_margin_mean": 0.39983880519866943, "beta_dpo/beta_margin_std": 0.32720947265625, "beta_dpo/beta_used": 0.1047118604183197, "beta_dpo/beta_used_raw": 0.1047118604183197, "beta_dpo/gap_mean": 2.542179584503174, "beta_dpo/gap_std": 2.094670295715332, "beta_dpo/loss_margin_mean": 3.8020293712615967, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.0513950073421439, "grad_norm": 136.2096710205078, "learning_rate": 2.463768115942029e-07, "logits/chosen": -3.1520180702209473, "logits/rejected": -3.1538925170898438, "loss": 1.1385, "step": 35 }, { "beta_dpo/beta": 0.12119206041097641, "beta_dpo/beta_margin_grad_mean": -0.3750362694263458, "beta_dpo/beta_margin_grad_std": 0.11262792348861694, "beta_dpo/beta_margin_mean": 0.5494691729545593, "beta_dpo/beta_margin_std": 0.5312042832374573, "beta_dpo/beta_used": 0.12119206041097641, "beta_dpo/beta_used_raw": 0.12119206041097641, "beta_dpo/gap_mean": 2.822805881500244, "beta_dpo/gap_std": 2.430403709411621, "beta_dpo/loss_margin_mean": 4.381009578704834, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.05286343612334802, "grad_norm": 132.07125854492188, "learning_rate": 2.536231884057971e-07, "logits/chosen": -3.1347692012786865, "logits/rejected": -3.133504629135132, "loss": 1.062, "step": 36 }, { "beta_dpo/beta": 0.12719795107841492, "beta_dpo/beta_margin_grad_mean": -0.368235319852829, "beta_dpo/beta_margin_grad_std": 0.10667005181312561, "beta_dpo/beta_margin_mean": 0.579422652721405, "beta_dpo/beta_margin_std": 0.5222266912460327, "beta_dpo/beta_used": 0.12719795107841492, "beta_dpo/beta_used_raw": 0.12719795107841492, "beta_dpo/gap_mean": 3.1904327869415283, "beta_dpo/gap_std": 2.7459893226623535, "beta_dpo/loss_margin_mean": 4.56009578704834, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.05433186490455213, "grad_norm": 143.18426513671875, "learning_rate": 2.6086956521739126e-07, "logits/chosen": -3.1382253170013428, "logits/rejected": -3.1056854724884033, "loss": 1.0, "step": 37 }, { "beta_dpo/beta": 0.14245614409446716, "beta_dpo/beta_margin_grad_mean": -0.31650885939598083, "beta_dpo/beta_margin_grad_std": 0.14402246475219727, "beta_dpo/beta_margin_mean": 0.8959528803825378, "beta_dpo/beta_margin_std": 0.8295892477035522, "beta_dpo/beta_used": 0.14245614409446716, "beta_dpo/beta_used_raw": 0.14245614409446716, "beta_dpo/gap_mean": 3.688507556915283, "beta_dpo/gap_std": 3.2539007663726807, "beta_dpo/loss_margin_mean": 6.270021438598633, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.055800293685756244, "grad_norm": 143.8856201171875, "learning_rate": 2.681159420289855e-07, "logits/chosen": -3.1051831245422363, "logits/rejected": -3.11987042427063, "loss": 0.9095, "step": 38 }, { "beta_dpo/beta": 0.1562386155128479, "beta_dpo/beta_margin_grad_mean": -0.28196191787719727, "beta_dpo/beta_margin_grad_std": 0.12678173184394836, "beta_dpo/beta_margin_mean": 1.0526187419891357, "beta_dpo/beta_margin_std": 0.7619715929031372, "beta_dpo/beta_used": 0.1562386155128479, "beta_dpo/beta_used_raw": 0.1562386155128479, "beta_dpo/gap_mean": 4.267126560211182, "beta_dpo/gap_std": 3.6263532638549805, "beta_dpo/loss_margin_mean": 6.750538349151611, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.05726872246696035, "grad_norm": 157.51174926757812, "learning_rate": 2.753623188405797e-07, "logits/chosen": -3.107410430908203, "logits/rejected": -3.0857043266296387, "loss": 0.7892, "step": 39 }, { "beta_dpo/beta": 0.1138225644826889, "beta_dpo/beta_margin_grad_mean": -0.3327762484550476, "beta_dpo/beta_margin_grad_std": 0.1230630874633789, "beta_dpo/beta_margin_mean": 0.7856088876724243, "beta_dpo/beta_margin_std": 0.7016869783401489, "beta_dpo/beta_used": 0.1138225644826889, "beta_dpo/beta_used_raw": 0.1138225644826889, "beta_dpo/gap_mean": 4.675539016723633, "beta_dpo/gap_std": 3.911595582962036, "beta_dpo/loss_margin_mean": 6.719384670257568, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.05873715124816446, "grad_norm": 121.4354248046875, "learning_rate": 2.8260869565217386e-07, "logits/chosen": -3.154458999633789, "logits/rejected": -3.135817050933838, "loss": 0.942, "step": 40 }, { "beta_dpo/beta": 0.17786243557929993, "beta_dpo/beta_margin_grad_mean": -0.2580529749393463, "beta_dpo/beta_margin_grad_std": 0.18610239028930664, "beta_dpo/beta_margin_mean": 1.4926997423171997, "beta_dpo/beta_margin_std": 1.711531162261963, "beta_dpo/beta_used": 0.17786243557929993, "beta_dpo/beta_used_raw": 0.17786243557929993, "beta_dpo/gap_mean": 5.223352432250977, "beta_dpo/gap_std": 4.3833465576171875, "beta_dpo/loss_margin_mean": 8.071098327636719, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.06020558002936858, "grad_norm": 153.85923767089844, "learning_rate": 2.898550724637681e-07, "logits/chosen": -3.1468467712402344, "logits/rejected": -3.1274843215942383, "loss": 0.6769, "step": 41 }, { "beta_dpo/beta": 0.1443341076374054, "beta_dpo/beta_margin_grad_mean": -0.28167596459388733, "beta_dpo/beta_margin_grad_std": 0.18971994519233704, "beta_dpo/beta_margin_mean": 1.44333815574646, "beta_dpo/beta_margin_std": 1.6360913515090942, "beta_dpo/beta_used": 0.1443341076374054, "beta_dpo/beta_used_raw": 0.1443341076374054, "beta_dpo/gap_mean": 5.9276628494262695, "beta_dpo/gap_std": 5.334522724151611, "beta_dpo/loss_margin_mean": 9.092480659484863, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.06167400881057269, "grad_norm": 157.87091064453125, "learning_rate": 2.971014492753623e-07, "logits/chosen": -3.130484104156494, "logits/rejected": -3.144623279571533, "loss": 0.8067, "step": 42 }, { "beta_dpo/beta": 0.18256396055221558, "beta_dpo/beta_margin_grad_mean": -0.20482036471366882, "beta_dpo/beta_margin_grad_std": 0.17849688231945038, "beta_dpo/beta_margin_mean": 1.8994702100753784, "beta_dpo/beta_margin_std": 1.475931167602539, "beta_dpo/beta_used": 0.18256396055221558, "beta_dpo/beta_used_raw": 0.18256396055221558, "beta_dpo/gap_mean": 6.808272361755371, "beta_dpo/gap_std": 5.8962321281433105, "beta_dpo/loss_margin_mean": 10.406832695007324, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.0631424375917768, "grad_norm": 143.8301239013672, "learning_rate": 3.043478260869565e-07, "logits/chosen": -3.1377265453338623, "logits/rejected": -3.1264193058013916, "loss": 0.5973, "step": 43 }, { "beta_dpo/beta": 0.24507826566696167, "beta_dpo/beta_margin_grad_mean": -0.14221227169036865, "beta_dpo/beta_margin_grad_std": 0.19724884629249573, "beta_dpo/beta_margin_mean": 3.0058138370513916, "beta_dpo/beta_margin_std": 2.3838889598846436, "beta_dpo/beta_used": 0.24507826566696167, "beta_dpo/beta_used_raw": 0.24507826566696167, "beta_dpo/gap_mean": 7.831865310668945, "beta_dpo/gap_std": 6.489755153656006, "beta_dpo/loss_margin_mean": 12.265984535217285, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.06461086637298091, "grad_norm": 115.14469146728516, "learning_rate": 3.115942028985507e-07, "logits/chosen": -3.0733070373535156, "logits/rejected": -3.0947670936584473, "loss": 0.3312, "step": 44 }, { "beta_dpo/beta": 0.038562677800655365, "beta_dpo/beta_margin_grad_mean": -0.4185938835144043, "beta_dpo/beta_margin_grad_std": 0.0780901312828064, "beta_dpo/beta_margin_mean": 0.3413236737251282, "beta_dpo/beta_margin_std": 0.34545838832855225, "beta_dpo/beta_used": 0.038562677800655365, "beta_dpo/beta_used_raw": 0.038562677800655365, "beta_dpo/gap_mean": 8.08301067352295, "beta_dpo/gap_std": 6.940834045410156, "beta_dpo/loss_margin_mean": 8.900261878967285, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.06607929515418502, "grad_norm": 59.289710998535156, "learning_rate": 3.188405797101449e-07, "logits/chosen": -3.0954208374023438, "logits/rejected": -3.097357749938965, "loss": 1.1443, "step": 45 }, { "beta_dpo/beta": 0.11945217847824097, "beta_dpo/beta_margin_grad_mean": -0.2516638934612274, "beta_dpo/beta_margin_grad_std": 0.16823652386665344, "beta_dpo/beta_margin_mean": 1.4459818601608276, "beta_dpo/beta_margin_std": 1.3947581052780151, "beta_dpo/beta_used": 0.11945217847824097, "beta_dpo/beta_used_raw": 0.11945217847824097, "beta_dpo/gap_mean": 8.649866104125977, "beta_dpo/gap_std": 7.57880973815918, "beta_dpo/loss_margin_mean": 11.921540260314941, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.06754772393538913, "grad_norm": 121.44014739990234, "learning_rate": 3.260869565217391e-07, "logits/chosen": -3.073902130126953, "logits/rejected": -3.071622371673584, "loss": 0.6997, "step": 46 }, { "beta_dpo/beta": 0.09873463958501816, "beta_dpo/beta_margin_grad_mean": -0.30619868636131287, "beta_dpo/beta_margin_grad_std": 0.17535436153411865, "beta_dpo/beta_margin_mean": 1.112424612045288, "beta_dpo/beta_margin_std": 1.307705283164978, "beta_dpo/beta_used": 0.09873463958501816, "beta_dpo/beta_used_raw": 0.09873463958501816, "beta_dpo/gap_mean": 8.923969268798828, "beta_dpo/gap_std": 8.19182014465332, "beta_dpo/loss_margin_mean": 10.096163749694824, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.06901615271659324, "grad_norm": 100.65027618408203, "learning_rate": 3.333333333333333e-07, "logits/chosen": -3.120870351791382, "logits/rejected": -3.112800121307373, "loss": 0.7862, "step": 47 }, { "beta_dpo/beta": 0.04364791885018349, "beta_dpo/beta_margin_grad_mean": -0.3776269853115082, "beta_dpo/beta_margin_grad_std": 0.10796888172626495, "beta_dpo/beta_margin_mean": 0.5390008687973022, "beta_dpo/beta_margin_std": 0.5236583948135376, "beta_dpo/beta_used": 0.04364791885018349, "beta_dpo/beta_used_raw": 0.04364791885018349, "beta_dpo/gap_mean": 9.664556503295898, "beta_dpo/gap_std": 8.99693489074707, "beta_dpo/loss_margin_mean": 12.434694290161133, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.07048458149779736, "grad_norm": 72.16007232666016, "learning_rate": 3.4057971014492755e-07, "logits/chosen": -3.109793186187744, "logits/rejected": -3.0930979251861572, "loss": 1.067, "step": 48 }, { "beta_dpo/beta": 0.1467229723930359, "beta_dpo/beta_margin_grad_mean": -0.2280292958021164, "beta_dpo/beta_margin_grad_std": 0.1954279989004135, "beta_dpo/beta_margin_mean": 2.1265885829925537, "beta_dpo/beta_margin_std": 2.2465431690216064, "beta_dpo/beta_used": 0.1467229723930359, "beta_dpo/beta_used_raw": 0.1467229723930359, "beta_dpo/gap_mean": 10.395795822143555, "beta_dpo/gap_std": 9.3504638671875, "beta_dpo/loss_margin_mean": 13.992044448852539, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.07195301027900147, "grad_norm": 86.9437484741211, "learning_rate": 3.478260869565217e-07, "logits/chosen": -3.131073474884033, "logits/rejected": -3.105538845062256, "loss": 0.6067, "step": 49 }, { "beta_dpo/beta": 0.18687647581100464, "beta_dpo/beta_margin_grad_mean": -0.22798167169094086, "beta_dpo/beta_margin_grad_std": 0.24264167249202728, "beta_dpo/beta_margin_mean": 2.600717306137085, "beta_dpo/beta_margin_std": 3.0297393798828125, "beta_dpo/beta_used": 0.18687647581100464, "beta_dpo/beta_used_raw": 0.18687647581100464, "beta_dpo/gap_mean": 11.018354415893555, "beta_dpo/gap_std": 10.152142524719238, "beta_dpo/loss_margin_mean": 13.667196273803711, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.07342143906020558, "grad_norm": 170.4370574951172, "learning_rate": 3.5507246376811595e-07, "logits/chosen": -3.116657257080078, "logits/rejected": -3.1149511337280273, "loss": 0.5317, "step": 50 }, { "beta_dpo/beta": 0.0934867113828659, "beta_dpo/beta_margin_grad_mean": -0.3069649636745453, "beta_dpo/beta_margin_grad_std": 0.1902894377708435, "beta_dpo/beta_margin_mean": 1.3371574878692627, "beta_dpo/beta_margin_std": 1.9382656812667847, "beta_dpo/beta_used": 0.0934867113828659, "beta_dpo/beta_used_raw": 0.0934867113828659, "beta_dpo/gap_mean": 11.582159996032715, "beta_dpo/gap_std": 11.08337688446045, "beta_dpo/loss_margin_mean": 14.272186279296875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.07488986784140969, "grad_norm": 96.85399627685547, "learning_rate": 3.6231884057971015e-07, "logits/chosen": -3.06636643409729, "logits/rejected": -3.0568933486938477, "loss": 0.7809, "step": 51 }, { "beta_dpo/beta": 0.18484123051166534, "beta_dpo/beta_margin_grad_mean": -0.21149781346321106, "beta_dpo/beta_margin_grad_std": 0.19529053568840027, "beta_dpo/beta_margin_mean": 3.6904523372650146, "beta_dpo/beta_margin_std": 4.510212421417236, "beta_dpo/beta_used": 0.18484123051166534, "beta_dpo/beta_used_raw": 0.18484123051166534, "beta_dpo/gap_mean": 12.820097923278809, "beta_dpo/gap_std": 11.931066513061523, "beta_dpo/loss_margin_mean": 18.8443546295166, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.0763582966226138, "grad_norm": 73.05899810791016, "learning_rate": 3.695652173913043e-07, "logits/chosen": -3.0794599056243896, "logits/rejected": -3.0700924396514893, "loss": 0.5789, "step": 52 }, { "beta_dpo/beta": 0.28637969493865967, "beta_dpo/beta_margin_grad_mean": -0.12516933679580688, "beta_dpo/beta_margin_grad_std": 0.2069372832775116, "beta_dpo/beta_margin_mean": 6.572714328765869, "beta_dpo/beta_margin_std": 7.698074817657471, "beta_dpo/beta_used": 0.28637969493865967, "beta_dpo/beta_used_raw": 0.28637969493865967, "beta_dpo/gap_mean": 13.912508010864258, "beta_dpo/gap_std": 12.778947830200195, "beta_dpo/loss_margin_mean": 20.386211395263672, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.07782672540381791, "grad_norm": 187.00868225097656, "learning_rate": 3.7681159420289855e-07, "logits/chosen": -3.0854148864746094, "logits/rejected": -3.0860137939453125, "loss": 0.2734, "step": 53 }, { "beta_dpo/beta": 0.06887197494506836, "beta_dpo/beta_margin_grad_mean": -0.3133368492126465, "beta_dpo/beta_margin_grad_std": 0.20501112937927246, "beta_dpo/beta_margin_mean": 1.3893598318099976, "beta_dpo/beta_margin_std": 1.9802852869033813, "beta_dpo/beta_used": 0.06887197494506836, "beta_dpo/beta_used_raw": 0.06331352889537811, "beta_dpo/gap_mean": 15.25615119934082, "beta_dpo/gap_std": 13.717507362365723, "beta_dpo/loss_margin_mean": 19.47393226623535, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.07929515418502203, "grad_norm": 57.23472213745117, "learning_rate": 3.8405797101449274e-07, "logits/chosen": -3.093733310699463, "logits/rejected": -3.0969040393829346, "loss": 0.8662, "step": 54 }, { "beta_dpo/beta": 0.06593922525644302, "beta_dpo/beta_margin_grad_mean": -0.3532218635082245, "beta_dpo/beta_margin_grad_std": 0.2077246904373169, "beta_dpo/beta_margin_mean": 1.131407380104065, "beta_dpo/beta_margin_std": 1.861047387123108, "beta_dpo/beta_used": 0.06593922525644302, "beta_dpo/beta_used_raw": 0.06507368385791779, "beta_dpo/gap_mean": 16.030441284179688, "beta_dpo/gap_std": 15.258533477783203, "beta_dpo/loss_margin_mean": 18.9444522857666, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.08076358296622614, "grad_norm": 94.53569030761719, "learning_rate": 3.9130434782608694e-07, "logits/chosen": -3.090271472930908, "logits/rejected": -3.0747573375701904, "loss": 0.9575, "step": 55 }, { "beta_dpo/beta": 0.3255024254322052, "beta_dpo/beta_margin_grad_mean": -0.16037730872631073, "beta_dpo/beta_margin_grad_std": 0.3197742998600006, "beta_dpo/beta_margin_mean": 7.480155944824219, "beta_dpo/beta_margin_std": 8.26274585723877, "beta_dpo/beta_used": 0.3255024254322052, "beta_dpo/beta_used_raw": 0.3255024254322052, "beta_dpo/gap_mean": 17.08382797241211, "beta_dpo/gap_std": 16.365951538085938, "beta_dpo/loss_margin_mean": 22.620737075805664, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.08223201174743025, "grad_norm": 467.24908447265625, "learning_rate": 3.9855072463768114e-07, "logits/chosen": -3.066891670227051, "logits/rejected": -3.0782933235168457, "loss": 0.4909, "step": 56 }, { "beta_dpo/beta": 0.13110555708408356, "beta_dpo/beta_margin_grad_mean": -0.16527362167835236, "beta_dpo/beta_margin_grad_std": 0.23161666095256805, "beta_dpo/beta_margin_mean": 3.3455374240875244, "beta_dpo/beta_margin_std": 3.1305649280548096, "beta_dpo/beta_used": 0.13110555708408356, "beta_dpo/beta_used_raw": 0.13110555708408356, "beta_dpo/gap_mean": 18.27362823486328, "beta_dpo/gap_std": 17.460128784179688, "beta_dpo/loss_margin_mean": 25.040273666381836, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.08370044052863436, "grad_norm": 155.8965606689453, "learning_rate": 4.057971014492754e-07, "logits/chosen": -3.085702419281006, "logits/rejected": -3.0846567153930664, "loss": 0.5031, "step": 57 }, { "beta_dpo/beta": 0.0819333866238594, "beta_dpo/beta_margin_grad_mean": -0.1960979849100113, "beta_dpo/beta_margin_grad_std": 0.16590198874473572, "beta_dpo/beta_margin_mean": 2.0763485431671143, "beta_dpo/beta_margin_std": 1.8784925937652588, "beta_dpo/beta_used": 0.0819333866238594, "beta_dpo/beta_used_raw": 0.0819333866238594, "beta_dpo/gap_mean": 19.728099822998047, "beta_dpo/gap_std": 18.534076690673828, "beta_dpo/loss_margin_mean": 25.43709373474121, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.08516886930983847, "grad_norm": 105.10377502441406, "learning_rate": 4.1304347826086954e-07, "logits/chosen": -3.0557703971862793, "logits/rejected": -3.044924259185791, "loss": 0.5304, "step": 58 }, { "beta_dpo/beta": 0.19281019270420074, "beta_dpo/beta_margin_grad_mean": -0.28002673387527466, "beta_dpo/beta_margin_grad_std": 0.255830854177475, "beta_dpo/beta_margin_mean": 5.561700344085693, "beta_dpo/beta_margin_std": 8.050152778625488, "beta_dpo/beta_used": 0.19281019270420074, "beta_dpo/beta_used_raw": 0.06114870309829712, "beta_dpo/gap_mean": 21.075639724731445, "beta_dpo/gap_std": 19.727760314941406, "beta_dpo/loss_margin_mean": 25.989961624145508, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.08663729809104258, "grad_norm": 139.58779907226562, "learning_rate": 4.2028985507246374e-07, "logits/chosen": -3.0765440464019775, "logits/rejected": -3.0723140239715576, "loss": 0.7404, "step": 59 }, { "beta_dpo/beta": 0.3099629878997803, "beta_dpo/beta_margin_grad_mean": -0.16709139943122864, "beta_dpo/beta_margin_grad_std": 0.31567007303237915, "beta_dpo/beta_margin_mean": 7.2313055992126465, "beta_dpo/beta_margin_std": 9.950211524963379, "beta_dpo/beta_used": 0.3099629878997803, "beta_dpo/beta_used_raw": 0.3099629878997803, "beta_dpo/gap_mean": 21.46399688720703, "beta_dpo/gap_std": 21.259231567382812, "beta_dpo/loss_margin_mean": 21.761423110961914, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.0881057268722467, "grad_norm": 332.7749328613281, "learning_rate": 4.2753623188405794e-07, "logits/chosen": -3.051283836364746, "logits/rejected": -3.03759503364563, "loss": 0.2553, "step": 60 }, { "beta_dpo/beta": 0.0911061018705368, "beta_dpo/beta_margin_grad_mean": -0.31341421604156494, "beta_dpo/beta_margin_grad_std": 0.25546589493751526, "beta_dpo/beta_margin_mean": 2.5058257579803467, "beta_dpo/beta_margin_std": 4.593942642211914, "beta_dpo/beta_used": 0.0911061018705368, "beta_dpo/beta_used_raw": 0.0911061018705368, "beta_dpo/gap_mean": 22.091663360595703, "beta_dpo/gap_std": 22.40526580810547, "beta_dpo/loss_margin_mean": 26.832210540771484, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.08957415565345081, "grad_norm": 144.1743927001953, "learning_rate": 4.3478260869565214e-07, "logits/chosen": -3.0594305992126465, "logits/rejected": -3.0812244415283203, "loss": 0.8416, "step": 61 }, { "beta_dpo/beta": 0.055423423647880554, "beta_dpo/beta_margin_grad_mean": -0.3313938081264496, "beta_dpo/beta_margin_grad_std": 0.21602469682693481, "beta_dpo/beta_margin_mean": 1.2362252473831177, "beta_dpo/beta_margin_std": 1.8517836332321167, "beta_dpo/beta_used": 0.055423423647880554, "beta_dpo/beta_used_raw": -0.009761884808540344, "beta_dpo/gap_mean": 22.36334228515625, "beta_dpo/gap_std": 22.1772403717041, "beta_dpo/loss_margin_mean": 22.912548065185547, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.09104258443465492, "grad_norm": 53.500511169433594, "learning_rate": 4.420289855072464e-07, "logits/chosen": -3.0460290908813477, "logits/rejected": -3.0380520820617676, "loss": 0.8303, "step": 62 }, { "beta_dpo/beta": 0.027289319783449173, "beta_dpo/beta_margin_grad_mean": -0.39005380868911743, "beta_dpo/beta_margin_grad_std": 0.17873267829418182, "beta_dpo/beta_margin_mean": 0.7003328204154968, "beta_dpo/beta_margin_std": 1.3609882593154907, "beta_dpo/beta_used": 0.027289319783449173, "beta_dpo/beta_used_raw": 0.017231859266757965, "beta_dpo/gap_mean": 23.183303833007812, "beta_dpo/gap_std": 23.588115692138672, "beta_dpo/loss_margin_mean": 27.170320510864258, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.09251101321585903, "grad_norm": 58.18537521362305, "learning_rate": 4.4927536231884053e-07, "logits/chosen": -3.0674667358398438, "logits/rejected": -3.0683016777038574, "loss": 1.0376, "step": 63 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4925476312637329, "beta_dpo/beta_margin_grad_std": 0.00753110833466053, "beta_dpo/beta_margin_mean": 0.029822714626789093, "beta_dpo/beta_margin_std": 0.030163373798131943, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.11208631843328476, "beta_dpo/gap_mean": 24.51315689086914, "beta_dpo/gap_std": 25.101673126220703, "beta_dpo/loss_margin_mean": 29.822711944580078, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.09397944199706314, "grad_norm": 3.092960834503174, "learning_rate": 4.5652173913043473e-07, "logits/chosen": -3.0635104179382324, "logits/rejected": -3.0607059001922607, "loss": 1.3655, "step": 64 }, { "beta_dpo/beta": 0.2129736691713333, "beta_dpo/beta_margin_grad_mean": -0.2961460053920746, "beta_dpo/beta_margin_grad_std": 0.2763463854789734, "beta_dpo/beta_margin_mean": 7.6759843826293945, "beta_dpo/beta_margin_std": 12.091070175170898, "beta_dpo/beta_used": 0.2129736691713333, "beta_dpo/beta_used_raw": 0.18820346891880035, "beta_dpo/gap_mean": 25.80198097229004, "beta_dpo/gap_std": 26.017024993896484, "beta_dpo/loss_margin_mean": 32.39765167236328, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.09544787077826726, "grad_norm": 288.397705078125, "learning_rate": 4.63768115942029e-07, "logits/chosen": -3.04362416267395, "logits/rejected": -3.0484161376953125, "loss": 0.8591, "step": 65 }, { "beta_dpo/beta": 0.02076115272939205, "beta_dpo/beta_margin_grad_mean": -0.37283429503440857, "beta_dpo/beta_margin_grad_std": 0.1666054129600525, "beta_dpo/beta_margin_mean": 0.713555097579956, "beta_dpo/beta_margin_std": 1.0996434688568115, "beta_dpo/beta_used": 0.02076115272939205, "beta_dpo/beta_used_raw": -0.16835710406303406, "beta_dpo/gap_mean": 26.47984504699707, "beta_dpo/gap_std": 27.09063148498535, "beta_dpo/loss_margin_mean": 27.242469787597656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.09691629955947137, "grad_norm": 32.42694091796875, "learning_rate": 4.7101449275362313e-07, "logits/chosen": -3.029677391052246, "logits/rejected": -3.0298986434936523, "loss": 1.0385, "step": 66 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4936298727989197, "beta_dpo/beta_margin_grad_std": 0.007120794616639614, "beta_dpo/beta_margin_mean": 0.025488976389169693, "beta_dpo/beta_margin_std": 0.02850145660340786, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.2273913323879242, "beta_dpo/gap_mean": 25.91208267211914, "beta_dpo/gap_std": 27.34076499938965, "beta_dpo/loss_margin_mean": 25.488975524902344, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.09838472834067548, "grad_norm": 3.0101864337921143, "learning_rate": 4.782608695652174e-07, "logits/chosen": -2.986940860748291, "logits/rejected": -2.976597547531128, "loss": 1.366, "step": 67 }, { "beta_dpo/beta": 0.14753571152687073, "beta_dpo/beta_margin_grad_mean": -0.2845630645751953, "beta_dpo/beta_margin_grad_std": 0.2652169167995453, "beta_dpo/beta_margin_mean": 5.07893180847168, "beta_dpo/beta_margin_std": 6.9394354820251465, "beta_dpo/beta_used": 0.14753571152687073, "beta_dpo/beta_used_raw": 0.058987803757190704, "beta_dpo/gap_mean": 26.669593811035156, "beta_dpo/gap_std": 27.67668914794922, "beta_dpo/loss_margin_mean": 31.38116455078125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.09985315712187959, "grad_norm": 293.31982421875, "learning_rate": 4.855072463768116e-07, "logits/chosen": -3.0039591789245605, "logits/rejected": -2.98842191696167, "loss": 0.88, "step": 68 }, { "beta_dpo/beta": 0.2622404098510742, "beta_dpo/beta_margin_grad_mean": -0.30106690526008606, "beta_dpo/beta_margin_grad_std": 0.28205162286758423, "beta_dpo/beta_margin_mean": 10.098217010498047, "beta_dpo/beta_margin_std": 17.523113250732422, "beta_dpo/beta_used": 0.2622404098510742, "beta_dpo/beta_used_raw": 0.10589639842510223, "beta_dpo/gap_mean": 28.50058364868164, "beta_dpo/gap_std": 28.632122039794922, "beta_dpo/loss_margin_mean": 36.14906311035156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1013215859030837, "grad_norm": 702.201416015625, "learning_rate": 4.927536231884058e-07, "logits/chosen": -3.010495662689209, "logits/rejected": -3.016476631164551, "loss": 0.8454, "step": 69 }, { "beta_dpo/beta": 0.05058491602540016, "beta_dpo/beta_margin_grad_mean": -0.314628541469574, "beta_dpo/beta_margin_grad_std": 0.21072426438331604, "beta_dpo/beta_margin_mean": 1.6962952613830566, "beta_dpo/beta_margin_std": 2.6675915718078613, "beta_dpo/beta_used": 0.05058491602540016, "beta_dpo/beta_used_raw": -0.013440538197755814, "beta_dpo/gap_mean": 29.656864166259766, "beta_dpo/gap_std": 29.526466369628906, "beta_dpo/loss_margin_mean": 34.522823333740234, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1027900146842878, "grad_norm": 88.88278198242188, "learning_rate": 5e-07, "logits/chosen": -2.9925642013549805, "logits/rejected": -2.9963603019714355, "loss": 0.8527, "step": 70 }, { "beta_dpo/beta": 0.0875883623957634, "beta_dpo/beta_margin_grad_mean": -0.2784726917743683, "beta_dpo/beta_margin_grad_std": 0.24897883832454681, "beta_dpo/beta_margin_mean": 4.192900657653809, "beta_dpo/beta_margin_std": 6.242672443389893, "beta_dpo/beta_used": 0.0875883623957634, "beta_dpo/beta_used_raw": 0.009463444352149963, "beta_dpo/gap_mean": 31.732118606567383, "beta_dpo/gap_std": 30.914306640625, "beta_dpo/loss_margin_mean": 40.439083099365234, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.10425844346549193, "grad_norm": 84.22056579589844, "learning_rate": 4.999967061337492e-07, "logits/chosen": -2.9915552139282227, "logits/rejected": -3.0017127990722656, "loss": 0.8371, "step": 71 }, { "beta_dpo/beta": 0.08755672723054886, "beta_dpo/beta_margin_grad_mean": -0.22227872908115387, "beta_dpo/beta_margin_grad_std": 0.20551539957523346, "beta_dpo/beta_margin_mean": 3.4542064666748047, "beta_dpo/beta_margin_std": 4.351135730743408, "beta_dpo/beta_used": 0.08755672723054886, "beta_dpo/beta_used_raw": 0.08755672723054886, "beta_dpo/gap_mean": 33.012107849121094, "beta_dpo/gap_std": 31.48851776123047, "beta_dpo/loss_margin_mean": 40.631866455078125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.10572687224669604, "grad_norm": 129.3617401123047, "learning_rate": 4.999868246217933e-07, "logits/chosen": -3.0176591873168945, "logits/rejected": -3.012908935546875, "loss": 0.6462, "step": 72 }, { "beta_dpo/beta": 0.03140610456466675, "beta_dpo/beta_margin_grad_mean": -0.3549874424934387, "beta_dpo/beta_margin_grad_std": 0.22458156943321228, "beta_dpo/beta_margin_mean": 1.2976723909378052, "beta_dpo/beta_margin_std": 2.2271358966827393, "beta_dpo/beta_used": 0.03140610456466675, "beta_dpo/beta_used_raw": 0.02235252410173416, "beta_dpo/gap_mean": 34.28472137451172, "beta_dpo/gap_std": 33.075035095214844, "beta_dpo/loss_margin_mean": 39.549720764160156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.10719530102790015, "grad_norm": 72.28749084472656, "learning_rate": 4.999703557245192e-07, "logits/chosen": -3.002220630645752, "logits/rejected": -3.003396987915039, "loss": 1.0018, "step": 73 }, { "beta_dpo/beta": 0.2684296667575836, "beta_dpo/beta_margin_grad_mean": -0.11206705868244171, "beta_dpo/beta_margin_grad_std": 0.24692973494529724, "beta_dpo/beta_margin_mean": 12.898226737976074, "beta_dpo/beta_margin_std": 19.086931228637695, "beta_dpo/beta_used": 0.2684296667575836, "beta_dpo/beta_used_raw": 0.2684296667575836, "beta_dpo/gap_mean": 36.407623291015625, "beta_dpo/gap_std": 34.9376220703125, "beta_dpo/loss_margin_mean": 47.21061325073242, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.10866372980910426, "grad_norm": 242.15377807617188, "learning_rate": 4.999472998758977e-07, "logits/chosen": -2.9633467197418213, "logits/rejected": -2.9860920906066895, "loss": 0.2967, "step": 74 }, { "beta_dpo/beta": 0.051932115107774734, "beta_dpo/beta_margin_grad_mean": -0.2812905013561249, "beta_dpo/beta_margin_grad_std": 0.22937369346618652, "beta_dpo/beta_margin_mean": 2.8386423587799072, "beta_dpo/beta_margin_std": 4.261453151702881, "beta_dpo/beta_used": 0.051932115107774734, "beta_dpo/beta_used_raw": -0.15860339999198914, "beta_dpo/gap_mean": 38.04350280761719, "beta_dpo/gap_std": 36.71551513671875, "beta_dpo/loss_margin_mean": 46.44905471801758, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.11013215859030837, "grad_norm": 59.50148391723633, "learning_rate": 4.999176576834721e-07, "logits/chosen": -2.9508605003356934, "logits/rejected": -2.9791858196258545, "loss": 0.792, "step": 75 }, { "beta_dpo/beta": 0.0431789867579937, "beta_dpo/beta_margin_grad_mean": -0.32823190093040466, "beta_dpo/beta_margin_grad_std": 0.24035446345806122, "beta_dpo/beta_margin_mean": 1.565407156944275, "beta_dpo/beta_margin_std": 2.5664196014404297, "beta_dpo/beta_used": 0.0431789867579937, "beta_dpo/beta_used_raw": -0.04499024525284767, "beta_dpo/gap_mean": 38.48039245605469, "beta_dpo/gap_std": 36.71884536743164, "beta_dpo/loss_margin_mean": 35.94696044921875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.11160058737151249, "grad_norm": 67.32833099365234, "learning_rate": 4.998814299283415e-07, "logits/chosen": -2.9746932983398438, "logits/rejected": -2.968583822250366, "loss": 0.8447, "step": 76 }, { "beta_dpo/beta": 0.34446391463279724, "beta_dpo/beta_margin_grad_mean": -0.08970285952091217, "beta_dpo/beta_margin_grad_std": 0.22384564578533173, "beta_dpo/beta_margin_mean": 14.092086791992188, "beta_dpo/beta_margin_std": 18.515459060668945, "beta_dpo/beta_used": 0.34446391463279724, "beta_dpo/beta_used_raw": 0.34446391463279724, "beta_dpo/gap_mean": 39.179561614990234, "beta_dpo/gap_std": 37.16301727294922, "beta_dpo/loss_margin_mean": 42.844810485839844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1130690161527166, "grad_norm": 293.4575500488281, "learning_rate": 4.998386175651409e-07, "logits/chosen": -2.947664499282837, "logits/rejected": -2.959230899810791, "loss": 0.1321, "step": 77 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4911826550960541, "beta_dpo/beta_margin_grad_std": 0.008943051099777222, "beta_dpo/beta_margin_mean": 0.035288915038108826, "beta_dpo/beta_margin_std": 0.035820893943309784, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.23828990757465363, "beta_dpo/gap_mean": 38.626731872558594, "beta_dpo/gap_std": 37.65977478027344, "beta_dpo/loss_margin_mean": 35.28891372680664, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1145374449339207, "grad_norm": 3.816556692123413, "learning_rate": 4.997892217220159e-07, "logits/chosen": -2.9786555767059326, "logits/rejected": -2.973895788192749, "loss": 1.3538, "step": 78 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.48892611265182495, "beta_dpo/beta_margin_grad_std": 0.010049103759229183, "beta_dpo/beta_margin_mean": 0.044323720037937164, "beta_dpo/beta_margin_std": 0.04024270921945572, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.21949611604213715, "beta_dpo/gap_mean": 39.299583435058594, "beta_dpo/gap_std": 37.97666931152344, "beta_dpo/loss_margin_mean": 44.3237190246582, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.11600587371512482, "grad_norm": 3.6560094356536865, "learning_rate": 4.997332437005931e-07, "logits/chosen": -2.9516005516052246, "logits/rejected": -2.9511117935180664, "loss": 1.3528, "step": 79 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4893885552883148, "beta_dpo/beta_margin_grad_std": 0.012339146807789803, "beta_dpo/beta_margin_mean": 0.04248877242207527, "beta_dpo/beta_margin_std": 0.049461763352155685, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.5071743726730347, "beta_dpo/gap_mean": 40.222755432128906, "beta_dpo/gap_std": 39.75575256347656, "beta_dpo/loss_margin_mean": 42.48876953125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.11747430249632893, "grad_norm": 3.7161951065063477, "learning_rate": 4.996706849759452e-07, "logits/chosen": -2.991058826446533, "logits/rejected": -2.986891269683838, "loss": 1.3566, "step": 80 }, { "beta_dpo/beta": 0.00884567853063345, "beta_dpo/beta_margin_grad_mean": -0.40446242690086365, "beta_dpo/beta_margin_grad_std": 0.1263885498046875, "beta_dpo/beta_margin_mean": 0.4591708183288574, "beta_dpo/beta_margin_std": 0.710827648639679, "beta_dpo/beta_used": 0.00884567853063345, "beta_dpo/beta_used_raw": -0.16323187947273254, "beta_dpo/gap_mean": 41.52234649658203, "beta_dpo/gap_std": 41.37035369873047, "beta_dpo/loss_margin_mean": 49.2286491394043, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.11894273127753303, "grad_norm": 29.591167449951172, "learning_rate": 4.996015471965529e-07, "logits/chosen": -2.9471867084503174, "logits/rejected": -2.958087921142578, "loss": 1.1115, "step": 81 }, { "beta_dpo/beta": 0.17800916731357574, "beta_dpo/beta_margin_grad_mean": -0.3327975273132324, "beta_dpo/beta_margin_grad_std": 0.30031806230545044, "beta_dpo/beta_margin_mean": 6.430704116821289, "beta_dpo/beta_margin_std": 13.450794219970703, "beta_dpo/beta_used": 0.17800916731357574, "beta_dpo/beta_used_raw": 0.1494421511888504, "beta_dpo/gap_mean": 41.84193420410156, "beta_dpo/gap_std": 42.8049201965332, "beta_dpo/loss_margin_mean": 45.061553955078125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.12041116005873716, "grad_norm": 565.0650634765625, "learning_rate": 4.995258321842611e-07, "logits/chosen": -2.938267230987549, "logits/rejected": -2.9391956329345703, "loss": 1.6144, "step": 82 }, { "beta_dpo/beta": 0.18749435245990753, "beta_dpo/beta_margin_grad_mean": -0.271941602230072, "beta_dpo/beta_margin_grad_std": 0.2540948987007141, "beta_dpo/beta_margin_mean": 9.913958549499512, "beta_dpo/beta_margin_std": 13.994511604309082, "beta_dpo/beta_used": 0.18749435245990753, "beta_dpo/beta_used_raw": 0.03422471880912781, "beta_dpo/gap_mean": 43.76371765136719, "beta_dpo/gap_std": 42.843292236328125, "beta_dpo/loss_margin_mean": 50.59626770019531, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.12187958883994127, "grad_norm": 307.31402587890625, "learning_rate": 4.994435419342304e-07, "logits/chosen": -2.9267525672912598, "logits/rejected": -2.9482855796813965, "loss": 1.123, "step": 83 }, { "beta_dpo/beta": 0.43998458981513977, "beta_dpo/beta_margin_grad_mean": -0.26115646958351135, "beta_dpo/beta_margin_grad_std": 0.2607704699039459, "beta_dpo/beta_margin_mean": 23.472091674804688, "beta_dpo/beta_margin_std": 32.818458557128906, "beta_dpo/beta_used": 0.43998458981513977, "beta_dpo/beta_used_raw": 0.12223160266876221, "beta_dpo/gap_mean": 44.48904800415039, "beta_dpo/gap_std": 41.942283630371094, "beta_dpo/loss_margin_mean": 44.652137756347656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.12334801762114538, "grad_norm": 2.4324653148651123, "learning_rate": 4.993546786148857e-07, "logits/chosen": -2.9470911026000977, "logits/rejected": -2.941443920135498, "loss": 0.6775, "step": 84 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.49015551805496216, "beta_dpo/beta_margin_grad_std": 0.009804553352296352, "beta_dpo/beta_margin_mean": 0.03939983248710632, "beta_dpo/beta_margin_std": 0.03925201669335365, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.16602811217308044, "beta_dpo/gap_mean": 43.268497467041016, "beta_dpo/gap_std": 41.009002685546875, "beta_dpo/loss_margin_mean": 39.39983367919922, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.12481644640234948, "grad_norm": 4.367675304412842, "learning_rate": 4.992592445678582e-07, "logits/chosen": -2.961709976196289, "logits/rejected": -2.935671806335449, "loss": 1.3481, "step": 85 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4886876940727234, "beta_dpo/beta_margin_grad_std": 0.01566956751048565, "beta_dpo/beta_margin_mean": 0.04532717540860176, "beta_dpo/beta_margin_std": 0.06287750601768494, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.7040857076644897, "beta_dpo/gap_mean": 43.424720764160156, "beta_dpo/gap_std": 43.841766357421875, "beta_dpo/loss_margin_mean": 45.32717514038086, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1262848751835536, "grad_norm": 3.6461679935455322, "learning_rate": 4.991572423079235e-07, "logits/chosen": -2.920724391937256, "logits/rejected": -2.9186758995056152, "loss": 1.3568, "step": 86 }, { "beta_dpo/beta": 0.04776458814740181, "beta_dpo/beta_margin_grad_mean": -0.23744255304336548, "beta_dpo/beta_margin_grad_std": 0.22597795724868774, "beta_dpo/beta_margin_mean": 2.473987340927124, "beta_dpo/beta_margin_std": 3.041849374771118, "beta_dpo/beta_used": 0.04776458814740181, "beta_dpo/beta_used_raw": 0.04776458814740181, "beta_dpo/gap_mean": 44.674156188964844, "beta_dpo/gap_std": 46.15048599243164, "beta_dpo/loss_margin_mean": 51.3042106628418, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1277533039647577, "grad_norm": 171.111083984375, "learning_rate": 4.990486745229364e-07, "logits/chosen": -2.9184794425964355, "logits/rejected": -2.9294486045837402, "loss": 0.7438, "step": 87 }, { "beta_dpo/beta": 0.2957174479961395, "beta_dpo/beta_margin_grad_mean": -0.12127989530563354, "beta_dpo/beta_margin_grad_std": 0.29749196767807007, "beta_dpo/beta_margin_mean": 18.622495651245117, "beta_dpo/beta_margin_std": 17.863021850585938, "beta_dpo/beta_used": 0.2957174479961395, "beta_dpo/beta_used_raw": 0.2957174479961395, "beta_dpo/gap_mean": 47.820945739746094, "beta_dpo/gap_std": 48.210693359375, "beta_dpo/loss_margin_mean": 62.004920959472656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.12922173274596183, "grad_norm": 559.2757568359375, "learning_rate": 4.989335440737586e-07, "logits/chosen": -2.9339442253112793, "logits/rejected": -2.943288564682007, "loss": 0.7024, "step": 88 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4874421954154968, "beta_dpo/beta_margin_grad_std": 0.013273374177515507, "beta_dpo/beta_margin_mean": 0.05029499903321266, "beta_dpo/beta_margin_std": 0.053264521062374115, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.4985317587852478, "beta_dpo/gap_mean": 48.497169494628906, "beta_dpo/gap_std": 49.19569396972656, "beta_dpo/loss_margin_mean": 50.29499435424805, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.13069016152716592, "grad_norm": 4.085449695587158, "learning_rate": 4.988118539941847e-07, "logits/chosen": -2.910324811935425, "logits/rejected": -2.9079842567443848, "loss": 1.3484, "step": 89 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4830739498138428, "beta_dpo/beta_margin_grad_std": 0.019520951434969902, "beta_dpo/beta_margin_mean": 0.06789281219244003, "beta_dpo/beta_margin_std": 0.07855159044265747, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.08338849246501923, "beta_dpo/gap_mean": 50.64426040649414, "beta_dpo/gap_std": 53.176513671875, "beta_dpo/loss_margin_mean": 67.89280700683594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.13215859030837004, "grad_norm": 5.08858585357666, "learning_rate": 4.986836074908615e-07, "logits/chosen": -2.881624221801758, "logits/rejected": -2.917466878890991, "loss": 1.3397, "step": 90 }, { "beta_dpo/beta": 0.046673670411109924, "beta_dpo/beta_margin_grad_mean": -0.28842246532440186, "beta_dpo/beta_margin_grad_std": 0.22475126385688782, "beta_dpo/beta_margin_mean": 2.8846089839935303, "beta_dpo/beta_margin_std": 4.184475898742676, "beta_dpo/beta_used": 0.046673670411109924, "beta_dpo/beta_used_raw": -0.17408108711242676, "beta_dpo/gap_mean": 52.84674835205078, "beta_dpo/gap_std": 54.52159118652344, "beta_dpo/loss_margin_mean": 56.735435485839844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.13362701908957417, "grad_norm": 93.58074951171875, "learning_rate": 4.985488079432037e-07, "logits/chosen": -2.899129629135132, "logits/rejected": -2.9001574516296387, "loss": 0.7869, "step": 91 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4871750771999359, "beta_dpo/beta_margin_grad_std": 0.014737357385456562, "beta_dpo/beta_margin_mean": 0.051369354128837585, "beta_dpo/beta_margin_std": 0.059110358357429504, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.3833653926849365, "beta_dpo/gap_mean": 53.06073760986328, "beta_dpo/gap_std": 55.06708526611328, "beta_dpo/loss_margin_mean": 51.36935043334961, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.13509544787077826, "grad_norm": 4.548574924468994, "learning_rate": 4.984074589033043e-07, "logits/chosen": -2.8663196563720703, "logits/rejected": -2.865586757659912, "loss": 1.3421, "step": 92 }, { "beta_dpo/beta": 0.0362042672932148, "beta_dpo/beta_margin_grad_mean": -0.33124664425849915, "beta_dpo/beta_margin_grad_std": 0.2518288195133209, "beta_dpo/beta_margin_mean": 1.77645742893219, "beta_dpo/beta_margin_std": 2.907167911529541, "beta_dpo/beta_used": 0.0362042672932148, "beta_dpo/beta_used_raw": -0.0553596131503582, "beta_dpo/gap_mean": 53.00493240356445, "beta_dpo/gap_std": 55.078792572021484, "beta_dpo/loss_margin_mean": 55.34242630004883, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.13656387665198239, "grad_norm": 138.0321807861328, "learning_rate": 4.982595640958425e-07, "logits/chosen": -2.8780605792999268, "logits/rejected": -2.865324020385742, "loss": 0.8654, "step": 93 }, { "beta_dpo/beta": 0.09114255011081696, "beta_dpo/beta_margin_grad_mean": -0.289016455411911, "beta_dpo/beta_margin_grad_std": 0.25749266147613525, "beta_dpo/beta_margin_mean": 6.2431559562683105, "beta_dpo/beta_margin_std": 10.020861625671387, "beta_dpo/beta_used": 0.09114255011081696, "beta_dpo/beta_used_raw": -0.06984854489564896, "beta_dpo/gap_mean": 55.642555236816406, "beta_dpo/gap_std": 57.18549346923828, "beta_dpo/loss_margin_mean": 68.45297241210938, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.13803230543318648, "grad_norm": 414.5119934082031, "learning_rate": 4.98105127417984e-07, "logits/chosen": -2.9009084701538086, "logits/rejected": -2.900761127471924, "loss": 0.8078, "step": 94 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4851512312889099, "beta_dpo/beta_margin_grad_std": 0.015362189151346684, "beta_dpo/beta_margin_mean": 0.059510547667741776, "beta_dpo/beta_margin_std": 0.061817847192287445, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.7674299478530884, "beta_dpo/gap_mean": 57.458091735839844, "beta_dpo/gap_std": 58.81836700439453, "beta_dpo/loss_margin_mean": 59.51054763793945, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1395007342143906, "grad_norm": 4.450385570526123, "learning_rate": 4.979441529392784e-07, "logits/chosen": -2.855530261993408, "logits/rejected": -2.8571863174438477, "loss": 1.344, "step": 95 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4837089478969574, "beta_dpo/beta_margin_grad_std": 0.01984335109591484, "beta_dpo/beta_margin_mean": 0.06533493846654892, "beta_dpo/beta_margin_std": 0.07974996417760849, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.7220409512519836, "beta_dpo/gap_mean": 58.48717498779297, "beta_dpo/gap_std": 60.73028564453125, "beta_dpo/loss_margin_mean": 65.3349380493164, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.14096916299559473, "grad_norm": 4.7258100509643555, "learning_rate": 4.977766449015534e-07, "logits/chosen": -2.904493570327759, "logits/rejected": -2.9026832580566406, "loss": 1.3426, "step": 96 }, { "beta_dpo/beta": 0.11098214983940125, "beta_dpo/beta_margin_grad_mean": -0.2807328701019287, "beta_dpo/beta_margin_grad_std": 0.2636745870113373, "beta_dpo/beta_margin_mean": 8.650201797485352, "beta_dpo/beta_margin_std": 13.248647689819336, "beta_dpo/beta_used": 0.11098214983940125, "beta_dpo/beta_used_raw": -0.17043042182922363, "beta_dpo/gap_mean": 59.17014694213867, "beta_dpo/gap_std": 62.05863952636719, "beta_dpo/loss_margin_mean": 67.58939361572266, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.14243759177679882, "grad_norm": 179.77455139160156, "learning_rate": 4.976026077188012e-07, "logits/chosen": -2.905517578125, "logits/rejected": -2.897564649581909, "loss": 0.9284, "step": 97 }, { "beta_dpo/beta": 0.49735867977142334, "beta_dpo/beta_margin_grad_mean": -0.13786101341247559, "beta_dpo/beta_margin_grad_std": 0.32877954840660095, "beta_dpo/beta_margin_mean": 29.118545532226562, "beta_dpo/beta_margin_std": 44.36186599731445, "beta_dpo/beta_used": 0.49735867977142334, "beta_dpo/beta_used_raw": 0.49735867977142334, "beta_dpo/gap_mean": 60.218910217285156, "beta_dpo/gap_std": 62.831886291503906, "beta_dpo/loss_margin_mean": 62.66773223876953, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.14390602055800295, "grad_norm": 1460.3056640625, "learning_rate": 4.974220459770639e-07, "logits/chosen": -2.874201774597168, "logits/rejected": -2.884645462036133, "loss": 1.6259, "step": 98 }, { "beta_dpo/beta": 0.27869686484336853, "beta_dpo/beta_margin_grad_mean": -0.07528817653656006, "beta_dpo/beta_margin_grad_std": 0.24065373837947845, "beta_dpo/beta_margin_mean": 21.852785110473633, "beta_dpo/beta_margin_std": 22.130836486816406, "beta_dpo/beta_used": 0.27869686484336853, "beta_dpo/beta_used_raw": 0.27869686484336853, "beta_dpo/gap_mean": 63.37742614746094, "beta_dpo/gap_std": 64.15032958984375, "beta_dpo/loss_margin_mean": 77.13097381591797, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.14537444933920704, "grad_norm": 384.5785217285156, "learning_rate": 4.972349644343108e-07, "logits/chosen": -2.845553398132324, "logits/rejected": -2.8533787727355957, "loss": 0.4593, "step": 99 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4882340431213379, "beta_dpo/beta_margin_grad_std": 0.015319556929171085, "beta_dpo/beta_margin_mean": 0.047130510210990906, "beta_dpo/beta_margin_std": 0.06143619865179062, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.3977210819721222, "beta_dpo/gap_mean": 61.136131286621094, "beta_dpo/gap_std": 64.02367401123047, "beta_dpo/loss_margin_mean": 47.13050842285156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.14684287812041116, "grad_norm": 5.179396629333496, "learning_rate": 4.970413680203148e-07, "logits/chosen": -2.8722152709960938, "logits/rejected": -2.8694002628326416, "loss": 1.3346, "step": 100 }, { "epoch": 0.14684287812041116, "eval_beta_dpo/beta": 0.021090898662805557, "eval_beta_dpo/beta_margin_grad_mean": -0.4726831018924713, "eval_beta_dpo/beta_margin_grad_std": 0.04030865803360939, "eval_beta_dpo/beta_margin_mean": 1.4684983491897583, "eval_beta_dpo/beta_margin_std": 2.0474720001220703, "eval_beta_dpo/beta_used": 0.021090898662805557, "eval_beta_dpo/beta_used_raw": -1.2172808647155762, "eval_beta_dpo/gap_mean": 60.651268005371094, "eval_beta_dpo/gap_std": 63.85259246826172, "eval_beta_dpo/loss_margin_mean": 38.69658660888672, "eval_beta_dpo/mask_keep_frac": 1.0, "eval_logits/chosen": -2.9129130840301514, "eval_logits/rejected": -2.903272867202759, "eval_loss": 0.7824556827545166, "eval_runtime": 37.8912, "eval_samples_per_second": 61.729, "eval_steps_per_second": 1.953, "step": 100 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4872121512889862, "beta_dpo/beta_margin_grad_std": 0.0187575314193964, "beta_dpo/beta_margin_mean": 0.05126000568270683, "beta_dpo/beta_margin_std": 0.07528804987668991, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.9154506325721741, "beta_dpo/gap_mean": 59.639793395996094, "beta_dpo/gap_std": 64.89390563964844, "beta_dpo/loss_margin_mean": 51.26000213623047, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.14831130690161526, "grad_norm": 4.951231002807617, "learning_rate": 4.968412618365215e-07, "logits/chosen": -2.889308452606201, "logits/rejected": -2.883979320526123, "loss": 1.3445, "step": 101 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.48730704188346863, "beta_dpo/beta_margin_grad_std": 0.01587892696261406, "beta_dpo/beta_margin_mean": 0.050841983407735825, "beta_dpo/beta_margin_std": 0.06364595144987106, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.11553104221820831, "beta_dpo/gap_mean": 57.58677673339844, "beta_dpo/gap_std": 65.677001953125, "beta_dpo/loss_margin_mean": 50.84197998046875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.14977973568281938, "grad_norm": 5.152194023132324, "learning_rate": 4.966346511559149e-07, "logits/chosen": -2.8980913162231445, "logits/rejected": -2.883908987045288, "loss": 1.3336, "step": 102 }, { "beta_dpo/beta": 0.9354996681213379, "beta_dpo/beta_margin_grad_mean": -0.15053774416446686, "beta_dpo/beta_margin_grad_std": 0.3507133424282074, "beta_dpo/beta_margin_mean": 78.73807525634766, "beta_dpo/beta_margin_std": 91.95064544677734, "beta_dpo/beta_used": 0.9354996681213379, "beta_dpo/beta_used_raw": 0.9354996681213379, "beta_dpo/gap_mean": 61.26586151123047, "beta_dpo/gap_std": 69.46250915527344, "beta_dpo/loss_margin_mean": 81.90780639648438, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1512481644640235, "grad_norm": 3375.601318359375, "learning_rate": 4.964215414228785e-07, "logits/chosen": -2.829054117202759, "logits/rejected": -2.808652639389038, "loss": 1.8738, "step": 103 }, { "beta_dpo/beta": 0.84881192445755, "beta_dpo/beta_margin_grad_mean": -0.10420799255371094, "beta_dpo/beta_margin_grad_std": 0.2966291308403015, "beta_dpo/beta_margin_mean": 72.4658432006836, "beta_dpo/beta_margin_std": 76.15467834472656, "beta_dpo/beta_used": 0.84881192445755, "beta_dpo/beta_used_raw": 0.84881192445755, "beta_dpo/gap_mean": 65.47351837158203, "beta_dpo/gap_std": 72.57354736328125, "beta_dpo/loss_margin_mean": 84.90675354003906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1527165932452276, "grad_norm": 844.726806640625, "learning_rate": 4.96201938253052e-07, "logits/chosen": -2.822493553161621, "logits/rejected": -2.817017078399658, "loss": 0.5082, "step": 104 }, { "beta_dpo/beta": 0.607758104801178, "beta_dpo/beta_margin_grad_mean": -0.07786455005407333, "beta_dpo/beta_margin_grad_std": 0.25590455532073975, "beta_dpo/beta_margin_mean": 51.5904541015625, "beta_dpo/beta_margin_std": 54.65573501586914, "beta_dpo/beta_used": 0.607758104801178, "beta_dpo/beta_used_raw": 0.607758104801178, "beta_dpo/gap_mean": 68.90283203125, "beta_dpo/gap_std": 75.64898681640625, "beta_dpo/loss_margin_mean": 87.5045394897461, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.15418502202643172, "grad_norm": 1462.4896240234375, "learning_rate": 4.959758474331832e-07, "logits/chosen": -2.8367412090301514, "logits/rejected": -2.831176280975342, "loss": 0.8486, "step": 105 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.483272522687912, "beta_dpo/beta_margin_grad_std": 0.01717246323823929, "beta_dpo/beta_margin_mean": 0.06702879071235657, "beta_dpo/beta_margin_std": 0.06890414655208588, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.6237323880195618, "beta_dpo/gap_mean": 69.55685424804688, "beta_dpo/gap_std": 74.42953491210938, "beta_dpo/loss_margin_mean": 67.0287857055664, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.15565345080763582, "grad_norm": 5.48669958114624, "learning_rate": 4.957432749209755e-07, "logits/chosen": -2.825288772583008, "logits/rejected": -2.8292627334594727, "loss": 1.3305, "step": 106 }, { "beta_dpo/beta": 0.19774837791919708, "beta_dpo/beta_margin_grad_mean": -0.2614811658859253, "beta_dpo/beta_margin_grad_std": 0.25258755683898926, "beta_dpo/beta_margin_mean": 17.963912963867188, "beta_dpo/beta_margin_std": 28.16929054260254, "beta_dpo/beta_used": 0.19774837791919708, "beta_dpo/beta_used_raw": 0.027784347534179688, "beta_dpo/gap_mean": 72.93892669677734, "beta_dpo/gap_std": 76.43939208984375, "beta_dpo/loss_margin_mean": 91.99864959716797, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.15712187958883994, "grad_norm": 554.6088256835938, "learning_rate": 4.955042268449307e-07, "logits/chosen": -2.838146448135376, "logits/rejected": -2.829848289489746, "loss": 0.9258, "step": 107 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.47412681579589844, "beta_dpo/beta_margin_grad_std": 0.02628808096051216, "beta_dpo/beta_margin_mean": 0.10398299247026443, "beta_dpo/beta_margin_std": 0.10614117234945297, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.3238527774810791, "beta_dpo/gap_mean": 78.68049621582031, "beta_dpo/gap_std": 82.48477172851562, "beta_dpo/loss_margin_mean": 103.98298645019531, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.15859030837004406, "grad_norm": 5.760761260986328, "learning_rate": 4.952587095041881e-07, "logits/chosen": -2.799206256866455, "logits/rejected": -2.7892160415649414, "loss": 1.317, "step": 108 }, { "beta_dpo/beta": 0.07913082838058472, "beta_dpo/beta_margin_grad_mean": -0.2913077771663666, "beta_dpo/beta_margin_grad_std": 0.26053932309150696, "beta_dpo/beta_margin_mean": 7.3264055252075195, "beta_dpo/beta_margin_std": 12.864703178405762, "beta_dpo/beta_used": 0.07913082838058472, "beta_dpo/beta_used_raw": 0.018090281635522842, "beta_dpo/gap_mean": 80.97541809082031, "beta_dpo/gap_std": 86.63990783691406, "beta_dpo/loss_margin_mean": 87.2574234008789, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.16005873715124816, "grad_norm": 48.93942642211914, "learning_rate": 4.95006729368358e-07, "logits/chosen": -2.8221731185913086, "logits/rejected": -2.8154239654541016, "loss": 0.7203, "step": 109 }, { "beta_dpo/beta": 0.13006778061389923, "beta_dpo/beta_margin_grad_mean": -0.29096195101737976, "beta_dpo/beta_margin_grad_std": 0.2567855417728424, "beta_dpo/beta_margin_mean": 11.700284957885742, "beta_dpo/beta_margin_std": 19.173776626586914, "beta_dpo/beta_used": 0.13006778061389923, "beta_dpo/beta_used_raw": -0.47315120697021484, "beta_dpo/gap_mean": 80.23991394042969, "beta_dpo/gap_std": 88.77531433105469, "beta_dpo/loss_margin_mean": 69.10839080810547, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.16152716593245228, "grad_norm": 311.8143310546875, "learning_rate": 4.947482930773511e-07, "logits/chosen": -2.800227165222168, "logits/rejected": -2.790717601776123, "loss": 1.1166, "step": 110 }, { "beta_dpo/beta": 0.23672765493392944, "beta_dpo/beta_margin_grad_mean": -0.32680338621139526, "beta_dpo/beta_margin_grad_std": 0.30344852805137634, "beta_dpo/beta_margin_mean": 21.393625259399414, "beta_dpo/beta_margin_std": 45.63887023925781, "beta_dpo/beta_used": 0.23672765493392944, "beta_dpo/beta_used_raw": 0.19301117956638336, "beta_dpo/gap_mean": 80.5330810546875, "beta_dpo/gap_std": 92.93572998046875, "beta_dpo/loss_margin_mean": 91.79338073730469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.16299559471365638, "grad_norm": 1797.791015625, "learning_rate": 4.944834074412042e-07, "logits/chosen": -2.82328724861145, "logits/rejected": -2.8216376304626465, "loss": 5.3229, "step": 111 }, { "beta_dpo/beta": 0.3619656264781952, "beta_dpo/beta_margin_grad_mean": -0.3380275368690491, "beta_dpo/beta_margin_grad_std": 0.3120715320110321, "beta_dpo/beta_margin_mean": 25.55278968811035, "beta_dpo/beta_margin_std": 48.8296012878418, "beta_dpo/beta_used": 0.3619656264781952, "beta_dpo/beta_used_raw": -0.08405748009681702, "beta_dpo/gap_mean": 78.98956298828125, "beta_dpo/gap_std": 91.90986633300781, "beta_dpo/loss_margin_mean": 65.13848114013672, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1644640234948605, "grad_norm": 1818.7918701171875, "learning_rate": 4.942120794399002e-07, "logits/chosen": -2.7841193675994873, "logits/rejected": -2.7855634689331055, "loss": 2.5001, "step": 112 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.48207101225852966, "beta_dpo/beta_margin_grad_std": 0.01914754882454872, "beta_dpo/beta_margin_mean": 0.07188614457845688, "beta_dpo/beta_margin_std": 0.07695025205612183, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.5030999779701233, "beta_dpo/gap_mean": 77.31889343261719, "beta_dpo/gap_std": 88.37618255615234, "beta_dpo/loss_margin_mean": 71.88613891601562, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.16593245227606462, "grad_norm": 6.126889705657959, "learning_rate": 4.939343162231841e-07, "logits/chosen": -2.798726797103882, "logits/rejected": -2.7898099422454834, "loss": 1.3211, "step": 113 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4782230257987976, "beta_dpo/beta_margin_grad_std": 0.02653990499675274, "beta_dpo/beta_margin_mean": 0.08755876123905182, "beta_dpo/beta_margin_std": 0.10747722536325455, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.9737315773963928, "beta_dpo/gap_mean": 78.52099609375, "beta_dpo/gap_std": 91.858642578125, "beta_dpo/loss_margin_mean": 87.55876159667969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.16740088105726872, "grad_norm": 6.043005466461182, "learning_rate": 4.936501251103751e-07, "logits/chosen": -2.7921528816223145, "logits/rejected": -2.7812323570251465, "loss": 1.3277, "step": 114 }, { "beta_dpo/beta": 0.09022793918848038, "beta_dpo/beta_margin_grad_mean": -0.3093552589416504, "beta_dpo/beta_margin_grad_std": 0.27903029322624207, "beta_dpo/beta_margin_mean": 9.817399024963379, "beta_dpo/beta_margin_std": 18.483304977416992, "beta_dpo/beta_used": 0.09022793918848038, "beta_dpo/beta_used_raw": -0.45726335048675537, "beta_dpo/gap_mean": 80.5860824584961, "beta_dpo/gap_std": 93.43727111816406, "beta_dpo/loss_margin_mean": 95.46105194091797, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.16886930983847284, "grad_norm": 466.8434753417969, "learning_rate": 4.933595135901732e-07, "logits/chosen": -2.8372421264648438, "logits/rejected": -2.8418526649475098, "loss": 1.2961, "step": 115 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.47841930389404297, "beta_dpo/beta_margin_grad_std": 0.022723974660038948, "beta_dpo/beta_margin_mean": 0.08669696003198624, "beta_dpo/beta_margin_std": 0.09222324937582016, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.2080257534980774, "beta_dpo/gap_mean": 82.31340026855469, "beta_dpo/gap_std": 93.35620880126953, "beta_dpo/loss_margin_mean": 86.69695281982422, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.17033773861967694, "grad_norm": 6.324546813964844, "learning_rate": 4.930624893204624e-07, "logits/chosen": -2.785477876663208, "logits/rejected": -2.792661190032959, "loss": 1.3117, "step": 116 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4851381480693817, "beta_dpo/beta_margin_grad_std": 0.022359393537044525, "beta_dpo/beta_margin_mean": 0.05954824388027191, "beta_dpo/beta_margin_std": 0.08978039026260376, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.7292430400848389, "beta_dpo/gap_mean": 79.99589538574219, "beta_dpo/gap_std": 94.4775619506836, "beta_dpo/loss_margin_mean": 59.548240661621094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.17180616740088106, "grad_norm": 6.232842445373535, "learning_rate": 4.927590601281083e-07, "logits/chosen": -2.785600185394287, "logits/rejected": -2.775449752807617, "loss": 1.322, "step": 117 }, { "beta_dpo/beta": 0.38719773292541504, "beta_dpo/beta_margin_grad_mean": -0.12059462815523148, "beta_dpo/beta_margin_grad_std": 0.2853429317474365, "beta_dpo/beta_margin_mean": 31.55919647216797, "beta_dpo/beta_margin_std": 51.115875244140625, "beta_dpo/beta_used": 0.38719773292541504, "beta_dpo/beta_used_raw": 0.38719773292541504, "beta_dpo/gap_mean": 79.46942138671875, "beta_dpo/gap_std": 92.74491882324219, "beta_dpo/loss_margin_mean": 87.44821166992188, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.17327459618208516, "grad_norm": 740.7146606445312, "learning_rate": 4.924492340087524e-07, "logits/chosen": -2.816103935241699, "logits/rejected": -2.816561222076416, "loss": 1.2551, "step": 118 }, { "beta_dpo/beta": 0.6606523394584656, "beta_dpo/beta_margin_grad_mean": -0.3558006286621094, "beta_dpo/beta_margin_grad_std": 0.3170710504055023, "beta_dpo/beta_margin_mean": 56.66292190551758, "beta_dpo/beta_margin_std": 127.7380599975586, "beta_dpo/beta_used": 0.6606523394584656, "beta_dpo/beta_used_raw": 0.13558882474899292, "beta_dpo/gap_mean": 84.13648986816406, "beta_dpo/gap_std": 96.36943817138672, "beta_dpo/loss_margin_mean": 101.86002349853516, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.17474302496328928, "grad_norm": 4218.83154296875, "learning_rate": 4.92133019126601e-07, "logits/chosen": -2.8366334438323975, "logits/rejected": -2.8262441158294678, "loss": 4.2288, "step": 119 }, { "beta_dpo/beta": 1.5045459270477295, "beta_dpo/beta_margin_grad_mean": -0.12338397651910782, "beta_dpo/beta_margin_grad_std": 0.3060937523841858, "beta_dpo/beta_margin_mean": 165.48951721191406, "beta_dpo/beta_margin_std": 231.87692260742188, "beta_dpo/beta_used": 1.5045459270477295, "beta_dpo/beta_used_raw": 1.5045459270477295, "beta_dpo/gap_mean": 87.93060302734375, "beta_dpo/gap_std": 102.44084167480469, "beta_dpo/loss_margin_mean": 109.52928924560547, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1762114537444934, "grad_norm": 2128.182373046875, "learning_rate": 4.918104238142103e-07, "logits/chosen": -2.851163864135742, "logits/rejected": -2.836759090423584, "loss": 1.3594, "step": 120 }, { "beta_dpo/beta": 0.9340792894363403, "beta_dpo/beta_margin_grad_mean": -0.06197686493396759, "beta_dpo/beta_margin_grad_std": 0.23998677730560303, "beta_dpo/beta_margin_mean": 120.16432189941406, "beta_dpo/beta_margin_std": 121.8327865600586, "beta_dpo/beta_used": 0.9340792894363403, "beta_dpo/beta_used_raw": 0.9340792894363403, "beta_dpo/gap_mean": 95.33363342285156, "beta_dpo/gap_std": 105.84669494628906, "beta_dpo/loss_margin_mean": 130.99896240234375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1776798825256975, "grad_norm": 790.7160034179688, "learning_rate": 4.91481456572267e-07, "logits/chosen": -2.795456886291504, "logits/rejected": -2.79376220703125, "loss": 1.2803, "step": 121 }, { "beta_dpo/beta": 0.41401243209838867, "beta_dpo/beta_margin_grad_mean": -0.26532647013664246, "beta_dpo/beta_margin_grad_std": 0.2661411166191101, "beta_dpo/beta_margin_mean": 51.30147171020508, "beta_dpo/beta_margin_std": 88.02983856201172, "beta_dpo/beta_used": 0.41401243209838867, "beta_dpo/beta_used_raw": 0.17554566264152527, "beta_dpo/gap_mean": 101.07794952392578, "beta_dpo/gap_std": 108.308349609375, "beta_dpo/loss_margin_mean": 126.12928009033203, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.17914831130690162, "grad_norm": 3183.692138671875, "learning_rate": 4.911461260693638e-07, "logits/chosen": -2.7693471908569336, "logits/rejected": -2.7797422409057617, "loss": 5.005, "step": 122 }, { "beta_dpo/beta": 0.8816313743591309, "beta_dpo/beta_margin_grad_mean": -0.30386897921562195, "beta_dpo/beta_margin_grad_std": 0.29489797353744507, "beta_dpo/beta_margin_mean": 100.96495056152344, "beta_dpo/beta_margin_std": 171.0264434814453, "beta_dpo/beta_used": 0.8816313743591309, "beta_dpo/beta_used_raw": 0.12362289428710938, "beta_dpo/gap_mean": 101.99317932128906, "beta_dpo/gap_std": 108.65696716308594, "beta_dpo/loss_margin_mean": 91.89662170410156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.18061674008810572, "grad_norm": 4966.56640625, "learning_rate": 4.908044411417711e-07, "logits/chosen": -2.8212316036224365, "logits/rejected": -2.8233444690704346, "loss": 5.5543, "step": 123 }, { "beta_dpo/beta": 0.2012283205986023, "beta_dpo/beta_margin_grad_mean": -0.29456260800361633, "beta_dpo/beta_margin_grad_std": 0.2699277400970459, "beta_dpo/beta_margin_mean": 25.60476303100586, "beta_dpo/beta_margin_std": 42.12246322631836, "beta_dpo/beta_used": 0.2012283205986023, "beta_dpo/beta_used_raw": 0.014203429222106934, "beta_dpo/gap_mean": 104.23336791992188, "beta_dpo/gap_std": 109.76822662353516, "beta_dpo/loss_margin_mean": 127.76998901367188, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.18208516886930984, "grad_norm": 713.97998046875, "learning_rate": 4.904564107932048e-07, "logits/chosen": -2.8201169967651367, "logits/rejected": -2.839149236679077, "loss": 1.1241, "step": 124 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4756603538990021, "beta_dpo/beta_margin_grad_std": 0.025525817647576332, "beta_dpo/beta_margin_mean": 0.09777088463306427, "beta_dpo/beta_margin_std": 0.10292989015579224, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.8198720216751099, "beta_dpo/gap_mean": 103.85806274414062, "beta_dpo/gap_std": 108.71501159667969, "beta_dpo/loss_margin_mean": 97.7708740234375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.18355359765051396, "grad_norm": 7.420654296875, "learning_rate": 4.90102044194588e-07, "logits/chosen": -2.781724452972412, "logits/rejected": -2.7918858528137207, "loss": 1.3016, "step": 125 }, { "beta_dpo/beta": 0.04341450706124306, "beta_dpo/beta_margin_grad_mean": -0.3020491302013397, "beta_dpo/beta_margin_grad_std": 0.25825148820877075, "beta_dpo/beta_margin_mean": 5.03449010848999, "beta_dpo/beta_margin_std": 8.020267486572266, "beta_dpo/beta_used": 0.04341450706124306, "beta_dpo/beta_used_raw": -0.4769749641418457, "beta_dpo/gap_mean": 104.70053100585938, "beta_dpo/gap_std": 109.29353332519531, "beta_dpo/loss_margin_mean": 104.23768615722656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.18502202643171806, "grad_norm": 176.98025512695312, "learning_rate": 4.897413506838102e-07, "logits/chosen": -2.812808036804199, "logits/rejected": -2.8147010803222656, "loss": 0.8898, "step": 126 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4809127449989319, "beta_dpo/beta_margin_grad_std": 0.018567463383078575, "beta_dpo/beta_margin_mean": 0.07651624083518982, "beta_dpo/beta_margin_std": 0.0745573341846466, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.2902711629867554, "beta_dpo/gap_mean": 100.42000579833984, "beta_dpo/gap_std": 105.15016174316406, "beta_dpo/loss_margin_mean": 76.5162353515625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.18649045521292218, "grad_norm": 6.344285488128662, "learning_rate": 4.89374339765481e-07, "logits/chosen": -2.828252077102661, "logits/rejected": -2.825937271118164, "loss": 1.3116, "step": 127 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.47901013493537903, "beta_dpo/beta_margin_grad_std": 0.026977377012372017, "beta_dpo/beta_margin_mean": 0.08433213084936142, "beta_dpo/beta_margin_std": 0.10864510387182236, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.2069345712661743, "beta_dpo/gap_mean": 96.0081558227539, "beta_dpo/gap_std": 103.26679229736328, "beta_dpo/loss_margin_mean": 84.3321304321289, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.18795888399412627, "grad_norm": 6.212590217590332, "learning_rate": 4.890010211106795e-07, "logits/chosen": -2.8211920261383057, "logits/rejected": -2.818962574005127, "loss": 1.3148, "step": 128 }, { "beta_dpo/beta": 0.10110783576965332, "beta_dpo/beta_margin_grad_mean": -0.33091118931770325, "beta_dpo/beta_margin_grad_std": 0.2910465598106384, "beta_dpo/beta_margin_mean": 10.52313232421875, "beta_dpo/beta_margin_std": 18.802413940429688, "beta_dpo/beta_used": 0.10110783576965332, "beta_dpo/beta_used_raw": 0.025442659854888916, "beta_dpo/gap_mean": 95.70207214355469, "beta_dpo/gap_std": 105.74406433105469, "beta_dpo/loss_margin_mean": 88.13529205322266, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.1894273127753304, "grad_norm": 417.793212890625, "learning_rate": 4.88621404556699e-07, "logits/chosen": -2.8037643432617188, "logits/rejected": -2.787684440612793, "loss": 1.4084, "step": 129 }, { "beta_dpo/beta": 0.23459400236606598, "beta_dpo/beta_margin_grad_mean": -0.2973823845386505, "beta_dpo/beta_margin_grad_std": 0.28484469652175903, "beta_dpo/beta_margin_mean": 29.821685791015625, "beta_dpo/beta_margin_std": 45.19208526611328, "beta_dpo/beta_used": 0.23459400236606598, "beta_dpo/beta_used_raw": -0.19728092849254608, "beta_dpo/gap_mean": 95.61531066894531, "beta_dpo/gap_std": 104.48291015625, "beta_dpo/loss_margin_mean": 108.31153869628906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.19089574155653452, "grad_norm": 1495.203125, "learning_rate": 4.882355001067891e-07, "logits/chosen": -2.765773057937622, "logits/rejected": -2.762284278869629, "loss": 2.5681, "step": 130 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4765341281890869, "beta_dpo/beta_margin_grad_std": 0.02474939450621605, "beta_dpo/beta_margin_mean": 0.09421201795339584, "beta_dpo/beta_margin_std": 0.09967362135648727, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.4088224470615387, "beta_dpo/gap_mean": 96.90589904785156, "beta_dpo/gap_std": 104.24870300292969, "beta_dpo/loss_margin_mean": 94.2120132446289, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.19236417033773862, "grad_norm": 7.157093048095703, "learning_rate": 4.878433179298909e-07, "logits/chosen": -2.7631754875183105, "logits/rejected": -2.775458574295044, "loss": 1.301, "step": 131 }, { "beta_dpo/beta": 0.017363857477903366, "beta_dpo/beta_margin_grad_mean": -0.3200368583202362, "beta_dpo/beta_margin_grad_std": 0.22496141493320465, "beta_dpo/beta_margin_mean": 1.8344775438308716, "beta_dpo/beta_margin_std": 3.168743371963501, "beta_dpo/beta_used": 0.017363857477903366, "beta_dpo/beta_used_raw": -1.0121409893035889, "beta_dpo/gap_mean": 95.9954833984375, "beta_dpo/gap_std": 105.45658874511719, "beta_dpo/loss_margin_mean": 85.24566650390625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.19383259911894274, "grad_norm": 54.672882080078125, "learning_rate": 4.874448683603694e-07, "logits/chosen": -2.77361798286438, "logits/rejected": -2.784986972808838, "loss": 0.8296, "step": 132 }, { "beta_dpo/beta": 0.44009122252464294, "beta_dpo/beta_margin_grad_mean": -0.2839052081108093, "beta_dpo/beta_margin_grad_std": 0.28118520975112915, "beta_dpo/beta_margin_mean": 48.52168273925781, "beta_dpo/beta_margin_std": 70.87135314941406, "beta_dpo/beta_used": 0.44009122252464294, "beta_dpo/beta_used_raw": 0.10653135180473328, "beta_dpo/gap_mean": 96.31338500976562, "beta_dpo/gap_std": 104.94859313964844, "beta_dpo/loss_margin_mean": 107.33547973632812, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.19530102790014683, "grad_norm": 1724.233642578125, "learning_rate": 4.870401618977415e-07, "logits/chosen": -2.7555744647979736, "logits/rejected": -2.765404224395752, "loss": 2.2334, "step": 133 }, { "beta_dpo/beta": 0.45935988426208496, "beta_dpo/beta_margin_grad_mean": -0.306671679019928, "beta_dpo/beta_margin_grad_std": 0.29106321930885315, "beta_dpo/beta_margin_mean": 41.79468536376953, "beta_dpo/beta_margin_std": 67.19827270507812, "beta_dpo/beta_used": 0.45935988426208496, "beta_dpo/beta_used_raw": 0.19374334812164307, "beta_dpo/gap_mean": 95.26974487304688, "beta_dpo/gap_std": 103.39216613769531, "beta_dpo/loss_margin_mean": 87.6424560546875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.19676945668135096, "grad_norm": 1905.6392822265625, "learning_rate": 4.866292092063986e-07, "logits/chosen": -2.7540433406829834, "logits/rejected": -2.7586114406585693, "loss": 2.5789, "step": 134 }, { "beta_dpo/beta": 0.3179980516433716, "beta_dpo/beta_margin_grad_mean": -0.0668000727891922, "beta_dpo/beta_margin_grad_std": 0.24074798822402954, "beta_dpo/beta_margin_mean": 37.95073318481445, "beta_dpo/beta_margin_std": 41.624820709228516, "beta_dpo/beta_used": 0.3179980516433716, "beta_dpo/beta_used_raw": 0.3179980516433716, "beta_dpo/gap_mean": 98.79299926757812, "beta_dpo/gap_std": 105.39846801757812, "beta_dpo/loss_margin_mean": 123.87374114990234, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.19823788546255505, "grad_norm": 400.1125183105469, "learning_rate": 4.862120211153265e-07, "logits/chosen": -2.7113471031188965, "logits/rejected": -2.7307372093200684, "loss": 0.6672, "step": 135 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4736141860485077, "beta_dpo/beta_margin_grad_std": 0.03395112603902817, "beta_dpo/beta_margin_mean": 0.10649430751800537, "beta_dpo/beta_margin_std": 0.13846558332443237, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.6484628915786743, "beta_dpo/gap_mean": 102.35041809082031, "beta_dpo/gap_std": 113.31024169921875, "beta_dpo/loss_margin_mean": 106.49430847167969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.19970631424375918, "grad_norm": 6.583731651306152, "learning_rate": 4.857886086178193e-07, "logits/chosen": -2.7721986770629883, "logits/rejected": -2.7757081985473633, "loss": 1.3157, "step": 136 }, { "beta_dpo/beta": 0.8055553436279297, "beta_dpo/beta_margin_grad_mean": -0.28008514642715454, "beta_dpo/beta_margin_grad_std": 0.2794659733772278, "beta_dpo/beta_margin_mean": 109.63809967041016, "beta_dpo/beta_margin_std": 171.1824188232422, "beta_dpo/beta_used": 0.8055553436279297, "beta_dpo/beta_used_raw": -0.391549289226532, "beta_dpo/gap_mean": 106.54948425292969, "beta_dpo/gap_std": 117.57719421386719, "beta_dpo/loss_margin_mean": 136.09088134765625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2011747430249633, "grad_norm": 694.3067626953125, "learning_rate": 4.853589828711902e-07, "logits/chosen": -2.7071657180786133, "logits/rejected": -2.72584867477417, "loss": 0.9732, "step": 137 }, { "beta_dpo/beta": 0.631058394908905, "beta_dpo/beta_margin_grad_mean": -0.27331289649009705, "beta_dpo/beta_margin_grad_std": 0.26683762669563293, "beta_dpo/beta_margin_mean": 88.83894348144531, "beta_dpo/beta_margin_std": 135.77273559570312, "beta_dpo/beta_used": 0.631058394908905, "beta_dpo/beta_used_raw": 0.5807164311408997, "beta_dpo/gap_mean": 108.60539245605469, "beta_dpo/gap_std": 118.61262512207031, "beta_dpo/loss_margin_mean": 118.49124145507812, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2026431718061674, "grad_norm": 1751.6961669921875, "learning_rate": 4.849231551964771e-07, "logits/chosen": -2.6887621879577637, "logits/rejected": -2.6839466094970703, "loss": 2.7148, "step": 138 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.47053390741348267, "beta_dpo/beta_margin_grad_std": 0.02328990399837494, "beta_dpo/beta_margin_mean": 0.1183198019862175, "beta_dpo/beta_margin_std": 0.09397386014461517, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.4085111618041992, "beta_dpo/gap_mean": 110.96134948730469, "beta_dpo/gap_std": 114.99424743652344, "beta_dpo/loss_margin_mean": 118.31979370117188, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.20411160058737152, "grad_norm": 7.088428020477295, "learning_rate": 4.844811370781446e-07, "logits/chosen": -2.668765068054199, "logits/rejected": -2.664041757583618, "loss": 1.2878, "step": 139 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4751591980457306, "beta_dpo/beta_margin_grad_std": 0.024874050170183182, "beta_dpo/beta_margin_mean": 0.09969595074653625, "beta_dpo/beta_margin_std": 0.1000821441411972, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.47669804096221924, "beta_dpo/gap_mean": 110.13583374023438, "beta_dpo/gap_std": 112.09087371826172, "beta_dpo/loss_margin_mean": 99.6959457397461, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2055800293685756, "grad_norm": 7.575645446777344, "learning_rate": 4.840329401637809e-07, "logits/chosen": -2.6947057247161865, "logits/rejected": -2.6926677227020264, "loss": 1.2898, "step": 140 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.47280803322792053, "beta_dpo/beta_margin_grad_std": 0.036303650587797165, "beta_dpo/beta_margin_mean": 0.10958551615476608, "beta_dpo/beta_margin_std": 0.14771895110607147, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.6926910877227783, "beta_dpo/gap_mean": 108.89583587646484, "beta_dpo/gap_std": 115.7603759765625, "beta_dpo/loss_margin_mean": 109.58551025390625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.20704845814977973, "grad_norm": 7.19344425201416, "learning_rate": 4.83578576263792e-07, "logits/chosen": -2.7385268211364746, "logits/rejected": -2.7286226749420166, "loss": 1.2944, "step": 141 }, { "beta_dpo/beta": 0.18145199120044708, "beta_dpo/beta_margin_grad_mean": -0.2992917001247406, "beta_dpo/beta_margin_grad_std": 0.2895396947860718, "beta_dpo/beta_margin_mean": 23.18352508544922, "beta_dpo/beta_margin_std": 34.87459945678711, "beta_dpo/beta_used": 0.18145199120044708, "beta_dpo/beta_used_raw": -0.9675522446632385, "beta_dpo/gap_mean": 111.3647232055664, "beta_dpo/gap_std": 120.15746307373047, "beta_dpo/loss_margin_mean": 122.43708801269531, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.20851688693098386, "grad_norm": 1052.642822265625, "learning_rate": 4.83118057351089e-07, "logits/chosen": -2.668506622314453, "logits/rejected": -2.6709957122802734, "loss": 1.5849, "step": 142 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4766398072242737, "beta_dpo/beta_margin_grad_std": 0.030281823128461838, "beta_dpo/beta_margin_mean": 0.09403061121702194, "beta_dpo/beta_margin_std": 0.1224864274263382, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.4053492546081543, "beta_dpo/gap_mean": 110.77619171142578, "beta_dpo/gap_std": 120.38482666015625, "beta_dpo/loss_margin_mean": 94.03060913085938, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.20998531571218795, "grad_norm": 7.156264781951904, "learning_rate": 4.826513955607734e-07, "logits/chosen": -2.6267662048339844, "logits/rejected": -2.6181480884552, "loss": 1.3042, "step": 143 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.47435808181762695, "beta_dpo/beta_margin_grad_std": 0.02727697044610977, "beta_dpo/beta_margin_mean": 0.10298500955104828, "beta_dpo/beta_margin_std": 0.10976456105709076, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.3952561616897583, "beta_dpo/gap_mean": 107.01061248779297, "beta_dpo/gap_std": 117.98554992675781, "beta_dpo/loss_margin_mean": 102.98500061035156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.21145374449339208, "grad_norm": 7.684736251831055, "learning_rate": 4.821786031898176e-07, "logits/chosen": -2.6427910327911377, "logits/rejected": -2.642993211746216, "loss": 1.2919, "step": 144 }, { "beta_dpo/beta": 0.3937012255191803, "beta_dpo/beta_margin_grad_mean": -0.2757228910923004, "beta_dpo/beta_margin_grad_std": 0.26662591099739075, "beta_dpo/beta_margin_mean": 40.2667236328125, "beta_dpo/beta_margin_std": 75.03095245361328, "beta_dpo/beta_used": 0.3937012255191803, "beta_dpo/beta_used_raw": 0.18985168635845184, "beta_dpo/gap_mean": 107.73323822021484, "beta_dpo/gap_std": 117.16415405273438, "beta_dpo/loss_margin_mean": 111.22066497802734, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.21292217327459617, "grad_norm": 894.4259033203125, "learning_rate": 4.816996926967401e-07, "logits/chosen": -2.630441665649414, "logits/rejected": -2.6299538612365723, "loss": 0.6781, "step": 145 }, { "beta_dpo/beta": 0.4095514714717865, "beta_dpo/beta_margin_grad_mean": -0.12312614917755127, "beta_dpo/beta_margin_grad_std": 0.3112068176269531, "beta_dpo/beta_margin_mean": 43.80937957763672, "beta_dpo/beta_margin_std": 50.46400833129883, "beta_dpo/beta_used": 0.4095514714717865, "beta_dpo/beta_used_raw": 0.4095514714717865, "beta_dpo/gap_mean": 108.25813293457031, "beta_dpo/gap_std": 116.30998992919922, "beta_dpo/loss_margin_mean": 108.27872467041016, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2143906020558003, "grad_norm": 1232.63330078125, "learning_rate": 4.812146767012779e-07, "logits/chosen": -2.6039223670959473, "logits/rejected": -2.601820707321167, "loss": 1.0263, "step": 146 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.47363805770874023, "beta_dpo/beta_margin_grad_std": 0.029436958953738213, "beta_dpo/beta_margin_mean": 0.1060258075594902, "beta_dpo/beta_margin_std": 0.11908337473869324, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.618476152420044, "beta_dpo/gap_mean": 107.93118286132812, "beta_dpo/gap_std": 118.50152587890625, "beta_dpo/loss_margin_mean": 106.02580261230469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.21585903083700442, "grad_norm": 7.791094779968262, "learning_rate": 4.807235679840536e-07, "logits/chosen": -2.580909490585327, "logits/rejected": -2.580644369125366, "loss": 1.2941, "step": 147 }, { "beta_dpo/beta": 0.021544385701417923, "beta_dpo/beta_margin_grad_mean": -0.3400629460811615, "beta_dpo/beta_margin_grad_std": 0.26523733139038086, "beta_dpo/beta_margin_mean": 3.093792676925659, "beta_dpo/beta_margin_std": 6.440648078918457, "beta_dpo/beta_used": 0.021544385701417923, "beta_dpo/beta_used_raw": -0.8347901105880737, "beta_dpo/gap_mean": 104.83203125, "beta_dpo/gap_std": 118.19056701660156, "beta_dpo/loss_margin_mean": 100.39215850830078, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2173274596182085, "grad_norm": 217.89871215820312, "learning_rate": 4.802263794862384e-07, "logits/chosen": -2.570409059524536, "logits/rejected": -2.576443672180176, "loss": 1.2305, "step": 148 }, { "beta_dpo/beta": 0.199530228972435, "beta_dpo/beta_margin_grad_mean": -0.3008774518966675, "beta_dpo/beta_margin_grad_std": 0.2911630868911743, "beta_dpo/beta_margin_mean": 34.24274826049805, "beta_dpo/beta_margin_std": 55.22300720214844, "beta_dpo/beta_used": 0.199530228972435, "beta_dpo/beta_used_raw": -0.7738866806030273, "beta_dpo/gap_mean": 112.32363891601562, "beta_dpo/gap_std": 125.47573852539062, "beta_dpo/loss_margin_mean": 131.21102905273438, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.21879588839941264, "grad_norm": 2060.334228515625, "learning_rate": 4.797231243092118e-07, "logits/chosen": -2.5769052505493164, "logits/rejected": -2.581183910369873, "loss": 3.0704, "step": 149 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.46728765964508057, "beta_dpo/beta_margin_grad_std": 0.03217744454741478, "beta_dpo/beta_margin_mean": 0.1318235695362091, "beta_dpo/beta_margin_std": 0.1308116763830185, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.7136021852493286, "beta_dpo/gap_mean": 114.93362426757812, "beta_dpo/gap_std": 126.79879760742188, "beta_dpo/loss_margin_mean": 131.8235626220703, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.22026431718061673, "grad_norm": 8.056601524353027, "learning_rate": 4.792138157142157e-07, "logits/chosen": -2.5865211486816406, "logits/rejected": -2.5941879749298096, "loss": 1.2894, "step": 150 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4679759740829468, "beta_dpo/beta_margin_grad_std": 0.03260992467403412, "beta_dpo/beta_margin_mean": 0.12910982966423035, "beta_dpo/beta_margin_std": 0.13299117982387543, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.161845862865448, "beta_dpo/gap_mean": 117.1995849609375, "beta_dpo/gap_std": 127.47024536132812, "beta_dpo/loss_margin_mean": 129.1098175048828, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.22173274596182085, "grad_norm": 8.165633201599121, "learning_rate": 4.786984671220053e-07, "logits/chosen": -2.597113847732544, "logits/rejected": -2.6131339073181152, "loss": 1.2784, "step": 151 }, { "beta_dpo/beta": 0.35655510425567627, "beta_dpo/beta_margin_grad_mean": -0.293893039226532, "beta_dpo/beta_margin_grad_std": 0.28947779536247253, "beta_dpo/beta_margin_mean": 48.276973724365234, "beta_dpo/beta_margin_std": 90.70513916015625, "beta_dpo/beta_used": 0.35655510425567627, "beta_dpo/beta_used_raw": 0.10587197542190552, "beta_dpo/gap_mean": 121.5870590209961, "beta_dpo/gap_std": 130.81991577148438, "beta_dpo/loss_margin_mean": 142.88314819335938, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.22320117474302498, "grad_norm": 2186.73388671875, "learning_rate": 4.78177092112495e-07, "logits/chosen": -2.5673489570617676, "logits/rejected": -2.573914051055908, "loss": 1.5166, "step": 152 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.46919021010398865, "beta_dpo/beta_margin_grad_std": 0.04115996137261391, "beta_dpo/beta_margin_mean": 0.12469355762004852, "beta_dpo/beta_margin_std": 0.16790282726287842, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.7168277502059937, "beta_dpo/gap_mean": 122.23858642578125, "beta_dpo/gap_std": 137.7699737548828, "beta_dpo/loss_margin_mean": 124.69355773925781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.22466960352422907, "grad_norm": 7.622184753417969, "learning_rate": 4.776497044244016e-07, "logits/chosen": -2.5728824138641357, "logits/rejected": -2.5736918449401855, "loss": 1.2987, "step": 153 }, { "beta_dpo/beta": 1.0757020711898804, "beta_dpo/beta_margin_grad_mean": -0.28336623311042786, "beta_dpo/beta_margin_grad_std": 0.2811123728752136, "beta_dpo/beta_margin_mean": 181.07421875, "beta_dpo/beta_margin_std": 271.3897705078125, "beta_dpo/beta_used": 1.0757020711898804, "beta_dpo/beta_used_raw": 0.654621958732605, "beta_dpo/gap_mean": 123.79661560058594, "beta_dpo/gap_std": 139.25714111328125, "beta_dpo/loss_margin_mean": 138.57278442382812, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2261380323054332, "grad_norm": 9847.91015625, "learning_rate": 4.771163179548808e-07, "logits/chosen": -2.5638508796691895, "logits/rejected": -2.5663564205169678, "loss": 2.0635, "step": 154 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.47398510575294495, "beta_dpo/beta_margin_grad_std": 0.026723386719822884, "beta_dpo/beta_margin_mean": 0.10449974238872528, "beta_dpo/beta_margin_std": 0.10794886201620102, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.4330192804336548, "beta_dpo/gap_mean": 123.12739562988281, "beta_dpo/gap_std": 134.49668884277344, "beta_dpo/loss_margin_mean": 104.49974060058594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2276064610866373, "grad_norm": 8.36424446105957, "learning_rate": 4.7657694675916247e-07, "logits/chosen": -2.595834255218506, "logits/rejected": -2.5887231826782227, "loss": 1.2924, "step": 155 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.47591692209243774, "beta_dpo/beta_margin_grad_std": 0.03880538046360016, "beta_dpo/beta_margin_mean": 0.09696371853351593, "beta_dpo/beta_margin_std": 0.15683069825172424, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.1627048254013062, "beta_dpo/gap_mean": 119.81661224365234, "beta_dpo/gap_std": 137.55975341796875, "beta_dpo/loss_margin_mean": 96.96371459960938, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2290748898678414, "grad_norm": 8.904071807861328, "learning_rate": 4.7603160505017893e-07, "logits/chosen": -2.5694189071655273, "logits/rejected": -2.5704991817474365, "loss": 1.2922, "step": 156 }, { "beta_dpo/beta": 1.0197237730026245, "beta_dpo/beta_margin_grad_mean": -0.26143965125083923, "beta_dpo/beta_margin_grad_std": 0.26461806893348694, "beta_dpo/beta_margin_mean": 192.70529174804688, "beta_dpo/beta_margin_std": 259.1637268066406, "beta_dpo/beta_used": 1.0197237730026245, "beta_dpo/beta_used_raw": 0.7222856283187866, "beta_dpo/gap_mean": 125.96708679199219, "beta_dpo/gap_std": 138.84896850585938, "beta_dpo/loss_margin_mean": 174.92384338378906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2305433186490455, "grad_norm": 6987.59423828125, "learning_rate": 4.7548030719819154e-07, "logits/chosen": -2.514146327972412, "logits/rejected": -2.5194973945617676, "loss": 2.6962, "step": 157 }, { "beta_dpo/beta": 0.9159993529319763, "beta_dpo/beta_margin_grad_mean": -0.2739473581314087, "beta_dpo/beta_margin_grad_std": 0.26920560002326965, "beta_dpo/beta_margin_mean": 162.5165252685547, "beta_dpo/beta_margin_std": 260.4980773925781, "beta_dpo/beta_used": 0.9159993529319763, "beta_dpo/beta_used_raw": 0.8316487669944763, "beta_dpo/gap_mean": 135.50296020507812, "beta_dpo/gap_std": 145.57350158691406, "beta_dpo/loss_margin_mean": 182.47454833984375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.23201174743024963, "grad_norm": 5398.83154296875, "learning_rate": 4.7492306773041136e-07, "logits/chosen": -2.5161776542663574, "logits/rejected": -2.5338191986083984, "loss": 5.0943, "step": 158 }, { "beta_dpo/beta": 0.36463621258735657, "beta_dpo/beta_margin_grad_mean": -0.3097352981567383, "beta_dpo/beta_margin_grad_std": 0.3000122308731079, "beta_dpo/beta_margin_mean": 49.08452224731445, "beta_dpo/beta_margin_std": 94.18856811523438, "beta_dpo/beta_used": 0.36463621258735657, "beta_dpo/beta_used_raw": -0.35086220502853394, "beta_dpo/gap_mean": 138.55572509765625, "beta_dpo/gap_std": 149.52554321289062, "beta_dpo/loss_margin_mean": 143.02760314941406, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.23348017621145375, "grad_norm": 3679.53076171875, "learning_rate": 4.743599013306165e-07, "logits/chosen": -2.5350537300109863, "logits/rejected": -2.527538299560547, "loss": 5.0817, "step": 159 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4628070592880249, "beta_dpo/beta_margin_grad_std": 0.04738787189126015, "beta_dpo/beta_margin_mean": 0.15084336698055267, "beta_dpo/beta_margin_std": 0.19365453720092773, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.107983946800232, "beta_dpo/gap_mean": 142.3861083984375, "beta_dpo/gap_std": 158.1815185546875, "beta_dpo/loss_margin_mean": 150.84335327148438, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.23494860499265785, "grad_norm": 9.592857360839844, "learning_rate": 4.737908228387656e-07, "logits/chosen": -2.50848388671875, "logits/rejected": -2.5071470737457275, "loss": 1.2723, "step": 160 }, { "beta_dpo/beta": 0.5015167593955994, "beta_dpo/beta_margin_grad_mean": -0.3111899793148041, "beta_dpo/beta_margin_grad_std": 0.29640379548072815, "beta_dpo/beta_margin_mean": 87.97203063964844, "beta_dpo/beta_margin_std": 136.19284057617188, "beta_dpo/beta_used": 0.5015167593955994, "beta_dpo/beta_used_raw": -0.03423714637756348, "beta_dpo/gap_mean": 142.8735809326172, "beta_dpo/gap_std": 155.4887237548828, "beta_dpo/loss_margin_mean": 143.24179077148438, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.23641703377386197, "grad_norm": 4087.4951171875, "learning_rate": 4.7321584725060594e-07, "logits/chosen": -2.4902877807617188, "logits/rejected": -2.498593330383301, "loss": 5.4119, "step": 161 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.463805228471756, "beta_dpo/beta_margin_grad_std": 0.03792537748813629, "beta_dpo/beta_margin_mean": 0.1462424099445343, "beta_dpo/beta_margin_std": 0.1551448106765747, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.097806692123413, "beta_dpo/gap_mean": 143.18414306640625, "beta_dpo/gap_std": 155.8494873046875, "beta_dpo/loss_margin_mean": 146.24240112304688, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.23788546255506607, "grad_norm": 8.91688346862793, "learning_rate": 4.7263498971727905e-07, "logits/chosen": -2.5275540351867676, "logits/rejected": -2.533555030822754, "loss": 1.2697, "step": 162 }, { "beta_dpo/beta": 0.14172588288784027, "beta_dpo/beta_margin_grad_mean": -0.29044896364212036, "beta_dpo/beta_margin_grad_std": 0.2808626890182495, "beta_dpo/beta_margin_mean": 20.941862106323242, "beta_dpo/beta_margin_std": 32.66022872924805, "beta_dpo/beta_used": 0.14172588288784027, "beta_dpo/beta_used_raw": -0.5184394717216492, "beta_dpo/gap_mean": 141.9095916748047, "beta_dpo/gap_std": 151.08145141601562, "beta_dpo/loss_margin_mean": 143.454345703125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2393538913362702, "grad_norm": 561.4199829101562, "learning_rate": 4.720482655449212e-07, "logits/chosen": -2.4482381343841553, "logits/rejected": -2.4447622299194336, "loss": 1.4418, "step": 163 }, { "beta_dpo/beta": 0.23951520025730133, "beta_dpo/beta_margin_grad_mean": -0.2740349769592285, "beta_dpo/beta_margin_grad_std": 0.2672988772392273, "beta_dpo/beta_margin_mean": 33.166099548339844, "beta_dpo/beta_margin_std": 47.26707458496094, "beta_dpo/beta_used": 0.23951520025730133, "beta_dpo/beta_used_raw": 0.11707229167222977, "beta_dpo/gap_mean": 142.1658935546875, "beta_dpo/gap_std": 142.63870239257812, "beta_dpo/loss_margin_mean": 141.0481414794922, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.24082232011747431, "grad_norm": 1184.495849609375, "learning_rate": 4.714556901942599e-07, "logits/chosen": -2.453787326812744, "logits/rejected": -2.461193799972534, "loss": 0.6742, "step": 164 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4727182984352112, "beta_dpo/beta_margin_grad_std": 0.033597320318222046, "beta_dpo/beta_margin_mean": 0.10977371782064438, "beta_dpo/beta_margin_std": 0.13549566268920898, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.5455293655395508, "beta_dpo/gap_mean": 138.08554077148438, "beta_dpo/gap_std": 140.043701171875, "beta_dpo/loss_margin_mean": 109.77371215820312, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2422907488986784, "grad_norm": 9.582477569580078, "learning_rate": 4.708572792802069e-07, "logits/chosen": -2.466679096221924, "logits/rejected": -2.4749114513397217, "loss": 1.2815, "step": 165 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.46377497911453247, "beta_dpo/beta_margin_grad_std": 0.05350736901164055, "beta_dpo/beta_margin_mean": 0.14767391979694366, "beta_dpo/beta_margin_std": 0.22187907993793488, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.600358784198761, "beta_dpo/gap_mean": 137.3470916748047, "beta_dpo/gap_std": 151.5578155517578, "beta_dpo/loss_margin_mean": 147.6739044189453, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.24375917767988253, "grad_norm": 9.58260440826416, "learning_rate": 4.702530485714461e-07, "logits/chosen": -2.4487009048461914, "logits/rejected": -2.4638147354125977, "loss": 1.2681, "step": 166 }, { "beta_dpo/beta": 0.7830126881599426, "beta_dpo/beta_margin_grad_mean": -0.10973574221134186, "beta_dpo/beta_margin_grad_std": 0.3119940757751465, "beta_dpo/beta_margin_mean": 129.75360107421875, "beta_dpo/beta_margin_std": 127.3109359741211, "beta_dpo/beta_used": 0.7830126881599426, "beta_dpo/beta_used_raw": 0.7830126881599426, "beta_dpo/gap_mean": 141.41099548339844, "beta_dpo/gap_std": 154.25009155273438, "beta_dpo/loss_margin_mean": 165.0268096923828, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.24522760646108663, "grad_norm": 6046.77734375, "learning_rate": 4.6964301399001877e-07, "logits/chosen": -2.447455883026123, "logits/rejected": -2.469198703765869, "loss": 7.7956, "step": 167 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4695003628730774, "beta_dpo/beta_margin_grad_std": 0.03412729129195213, "beta_dpo/beta_margin_mean": 0.12282736599445343, "beta_dpo/beta_margin_std": 0.13787633180618286, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.081223487854004, "beta_dpo/gap_mean": 140.6132354736328, "beta_dpo/gap_std": 153.59951782226562, "beta_dpo/loss_margin_mean": 122.82736206054688, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.24669603524229075, "grad_norm": 9.591252326965332, "learning_rate": 4.690271916109034e-07, "logits/chosen": -2.4348056316375732, "logits/rejected": -2.4319422245025635, "loss": 1.2879, "step": 168 }, { "beta_dpo/beta": 0.2029220461845398, "beta_dpo/beta_margin_grad_mean": -0.3294808864593506, "beta_dpo/beta_margin_grad_std": 0.3101998269557953, "beta_dpo/beta_margin_mean": 32.48569869995117, "beta_dpo/beta_margin_std": 66.5421371459961, "beta_dpo/beta_used": 0.2029220461845398, "beta_dpo/beta_used_raw": -0.917236328125, "beta_dpo/gap_mean": 139.76837158203125, "beta_dpo/gap_std": 157.51992797851562, "beta_dpo/loss_margin_mean": 134.8309326171875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.24816446402349487, "grad_norm": 1549.650146484375, "learning_rate": 4.6840559766159235e-07, "logits/chosen": -2.412907361984253, "logits/rejected": -2.4144067764282227, "loss": 1.9765, "step": 169 }, { "beta_dpo/beta": 0.14525021612644196, "beta_dpo/beta_margin_grad_mean": -0.2680424451828003, "beta_dpo/beta_margin_grad_std": 0.2648860812187195, "beta_dpo/beta_margin_mean": 20.61962890625, "beta_dpo/beta_margin_std": 32.76333236694336, "beta_dpo/beta_used": 0.14525021612644196, "beta_dpo/beta_used_raw": -0.05808502435684204, "beta_dpo/gap_mean": 137.88375854492188, "beta_dpo/gap_std": 152.96636962890625, "beta_dpo/loss_margin_mean": 133.55470275878906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.24963289280469897, "grad_norm": 42.14877700805664, "learning_rate": 4.6777824852166437e-07, "logits/chosen": -2.4181408882141113, "logits/rejected": -2.4320292472839355, "loss": 0.6382, "step": 170 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4662558138370514, "beta_dpo/beta_margin_grad_std": 0.0472683347761631, "beta_dpo/beta_margin_mean": 0.1371498852968216, "beta_dpo/beta_margin_std": 0.19521786272525787, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.053070068359375, "beta_dpo/gap_mean": 136.4842987060547, "beta_dpo/gap_std": 157.8990478515625, "beta_dpo/loss_margin_mean": 137.14987182617188, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2511013215859031, "grad_norm": 10.087845802307129, "learning_rate": 4.6714516072235273e-07, "logits/chosen": -2.4267027378082275, "logits/rejected": -2.4207305908203125, "loss": 1.2906, "step": 171 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4663008153438568, "beta_dpo/beta_margin_grad_std": 0.034671008586883545, "beta_dpo/beta_margin_mean": 0.1361597627401352, "beta_dpo/beta_margin_std": 0.1432521492242813, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.8160032629966736, "beta_dpo/gap_mean": 136.5276336669922, "beta_dpo/gap_std": 157.35430908203125, "beta_dpo/loss_margin_mean": 136.15975952148438, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2525697503671072, "grad_norm": 9.957985877990723, "learning_rate": 4.6650635094610966e-07, "logits/chosen": -2.444511890411377, "logits/rejected": -2.4361934661865234, "loss": 1.2706, "step": 172 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.47176581621170044, "beta_dpo/beta_margin_grad_std": 0.035718757659196854, "beta_dpo/beta_margin_mean": 0.11384298652410507, "beta_dpo/beta_margin_std": 0.14550843834877014, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.0770390033721924, "beta_dpo/gap_mean": 133.58248901367188, "beta_dpo/gap_std": 156.02670288085938, "beta_dpo/loss_margin_mean": 113.84297943115234, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2540381791483113, "grad_norm": 9.779240608215332, "learning_rate": 4.6586183602616687e-07, "logits/chosen": -2.432872772216797, "logits/rejected": -2.435973882675171, "loss": 1.2937, "step": 173 }, { "beta_dpo/beta": 0.3803338408470154, "beta_dpo/beta_margin_grad_mean": -0.30775538086891174, "beta_dpo/beta_margin_grad_std": 0.29982587695121765, "beta_dpo/beta_margin_mean": 62.02728271484375, "beta_dpo/beta_margin_std": 124.6036605834961, "beta_dpo/beta_used": 0.3803338408470154, "beta_dpo/beta_used_raw": 0.21193695068359375, "beta_dpo/gap_mean": 137.41583251953125, "beta_dpo/gap_std": 162.42489624023438, "beta_dpo/loss_margin_mean": 164.56101989746094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2555066079295154, "grad_norm": 3307.25830078125, "learning_rate": 4.652116329460919e-07, "logits/chosen": -2.427743434906006, "logits/rejected": -2.437385082244873, "loss": 4.3908, "step": 174 }, { "beta_dpo/beta": 0.29845502972602844, "beta_dpo/beta_margin_grad_mean": -0.29286932945251465, "beta_dpo/beta_margin_grad_std": 0.2887791693210602, "beta_dpo/beta_margin_mean": 52.850128173828125, "beta_dpo/beta_margin_std": 88.22852325439453, "beta_dpo/beta_used": 0.29845502972602844, "beta_dpo/beta_used_raw": 0.1955851912498474, "beta_dpo/gap_mean": 142.67230224609375, "beta_dpo/gap_std": 161.5905303955078, "beta_dpo/loss_margin_mean": 167.7036590576172, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.25697503671071953, "grad_norm": 3898.72021484375, "learning_rate": 4.645557588393406e-07, "logits/chosen": -2.424152374267578, "logits/rejected": -2.4236903190612793, "loss": 9.4346, "step": 175 }, { "beta_dpo/beta": 0.5137513875961304, "beta_dpo/beta_margin_grad_mean": -0.2328094094991684, "beta_dpo/beta_margin_grad_std": 0.23490336537361145, "beta_dpo/beta_margin_mean": 96.55097198486328, "beta_dpo/beta_margin_std": 138.6733856201172, "beta_dpo/beta_used": 0.5137513875961304, "beta_dpo/beta_used_raw": -0.3851640224456787, "beta_dpo/gap_mean": 148.3616943359375, "beta_dpo/gap_std": 161.01882934570312, "beta_dpo/loss_margin_mean": 163.68028259277344, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.25844346549192365, "grad_norm": 6.7193403244018555, "learning_rate": 4.638942309888058e-07, "logits/chosen": -2.455569267272949, "logits/rejected": -2.4624271392822266, "loss": 0.6383, "step": 176 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4615153968334198, "beta_dpo/beta_margin_grad_std": 0.045166175812482834, "beta_dpo/beta_margin_mean": 0.15591758489608765, "beta_dpo/beta_margin_std": 0.18478171527385712, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.3913229703903198, "beta_dpo/gap_mean": 148.726806640625, "beta_dpo/gap_std": 166.01434326171875, "beta_dpo/loss_margin_mean": 155.91757202148438, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2599118942731278, "grad_norm": 9.99582576751709, "learning_rate": 4.6322706682636137e-07, "logits/chosen": -2.4966821670532227, "logits/rejected": -2.503170967102051, "loss": 1.2699, "step": 177 }, { "beta_dpo/beta": 0.6906166076660156, "beta_dpo/beta_margin_grad_mean": -0.24617867171764374, "beta_dpo/beta_margin_grad_std": 0.243428036570549, "beta_dpo/beta_margin_mean": 125.11310577392578, "beta_dpo/beta_margin_std": 196.64479064941406, "beta_dpo/beta_used": 0.6906166076660156, "beta_dpo/beta_used_raw": 0.5849722027778625, "beta_dpo/gap_mean": 156.6367645263672, "beta_dpo/gap_std": 169.2756805419922, "beta_dpo/loss_margin_mean": 194.61502075195312, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.26138032305433184, "grad_norm": 1372.7340087890625, "learning_rate": 4.6255428393240354e-07, "logits/chosen": -2.419686794281006, "logits/rejected": -2.419022560119629, "loss": 0.643, "step": 178 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.46199724078178406, "beta_dpo/beta_margin_grad_std": 0.0452277697622776, "beta_dpo/beta_margin_mean": 0.15396226942539215, "beta_dpo/beta_margin_std": 0.18460071086883545, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.5662036538124084, "beta_dpo/gap_mean": 158.12814331054688, "beta_dpo/gap_std": 171.18023681640625, "beta_dpo/loss_margin_mean": 153.96226501464844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.26284875183553597, "grad_norm": 10.445080757141113, "learning_rate": 4.6187590003538724e-07, "logits/chosen": -2.497511863708496, "logits/rejected": -2.508460283279419, "loss": 1.249, "step": 179 }, { "beta_dpo/beta": 0.10046719759702682, "beta_dpo/beta_margin_grad_mean": -0.3064371645450592, "beta_dpo/beta_margin_grad_std": 0.29360902309417725, "beta_dpo/beta_margin_mean": 18.78900146484375, "beta_dpo/beta_margin_std": 31.81175422668457, "beta_dpo/beta_used": 0.10046719759702682, "beta_dpo/beta_used_raw": -0.13104525208473206, "beta_dpo/gap_mean": 157.84716796875, "beta_dpo/gap_std": 173.5079345703125, "beta_dpo/loss_margin_mean": 168.60824584960938, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2643171806167401, "grad_norm": 1487.51171875, "learning_rate": 4.611919330113591e-07, "logits/chosen": -2.4725236892700195, "logits/rejected": -2.4876785278320312, "loss": 2.5514, "step": 180 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.468180775642395, "beta_dpo/beta_margin_grad_std": 0.03856905922293663, "beta_dpo/beta_margin_mean": 0.12825387716293335, "beta_dpo/beta_margin_std": 0.15621516108512878, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.8151113390922546, "beta_dpo/gap_mean": 154.2423095703125, "beta_dpo/gap_std": 171.04672241210938, "beta_dpo/loss_margin_mean": 128.25387573242188, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2657856093979442, "grad_norm": 9.46091365814209, "learning_rate": 4.605024008834863e-07, "logits/chosen": -2.453889846801758, "logits/rejected": -2.453495979309082, "loss": 1.2548, "step": 181 }, { "beta_dpo/beta": 1.074347972869873, "beta_dpo/beta_margin_grad_mean": -0.24131838977336884, "beta_dpo/beta_margin_grad_std": 0.24388866126537323, "beta_dpo/beta_margin_mean": 206.86056518554688, "beta_dpo/beta_margin_std": 286.9940490722656, "beta_dpo/beta_used": 1.074347972869873, "beta_dpo/beta_used_raw": 0.8183754086494446, "beta_dpo/gap_mean": 159.13601684570312, "beta_dpo/gap_std": 169.01043701171875, "beta_dpo/loss_margin_mean": 191.2897491455078, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.26725403817914833, "grad_norm": 6.596961975097656, "learning_rate": 4.598073218215817e-07, "logits/chosen": -2.4499645233154297, "logits/rejected": -2.452929973602295, "loss": 0.6226, "step": 182 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.46204835176467896, "beta_dpo/beta_margin_grad_std": 0.037284255027770996, "beta_dpo/beta_margin_mean": 0.15333117544651031, "beta_dpo/beta_margin_std": 0.152970090508461, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.040143370628357, "beta_dpo/gap_mean": 160.8644561767578, "beta_dpo/gap_std": 167.21820068359375, "beta_dpo/loss_margin_mean": 153.33116149902344, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2687224669603524, "grad_norm": 10.506147384643555, "learning_rate": 4.5910671414162484e-07, "logits/chosen": -2.4273524284362793, "logits/rejected": -2.4347071647644043, "loss": 1.2515, "step": 183 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4735415279865265, "beta_dpo/beta_margin_grad_std": 0.03694302588701248, "beta_dpo/beta_margin_mean": 0.10642834007740021, "beta_dpo/beta_margin_std": 0.14879478514194489, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.0169718265533447, "beta_dpo/gap_mean": 150.88897705078125, "beta_dpo/gap_std": 165.56298828125, "beta_dpo/loss_margin_mean": 106.42832946777344, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2701908957415565, "grad_norm": 10.852279663085938, "learning_rate": 4.5840059630527985e-07, "logits/chosen": -2.424748420715332, "logits/rejected": -2.420259952545166, "loss": 1.2617, "step": 184 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4680558145046234, "beta_dpo/beta_margin_grad_std": 0.03759468346834183, "beta_dpo/beta_margin_mean": 0.12906821072101593, "beta_dpo/beta_margin_std": 0.15372078120708466, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.9787791967391968, "beta_dpo/gap_mean": 146.270263671875, "beta_dpo/gap_std": 162.84954833984375, "beta_dpo/loss_margin_mean": 129.06820678710938, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.27165932452276065, "grad_norm": 10.52782917022705, "learning_rate": 4.5768898691940836e-07, "logits/chosen": -2.3389992713928223, "logits/rejected": -2.340506076812744, "loss": 1.265, "step": 185 }, { "beta_dpo/beta": 0.196396142244339, "beta_dpo/beta_margin_grad_mean": -0.259600967168808, "beta_dpo/beta_margin_grad_std": 0.25825586915016174, "beta_dpo/beta_margin_mean": 36.263214111328125, "beta_dpo/beta_margin_std": 56.00040817260742, "beta_dpo/beta_used": 0.196396142244339, "beta_dpo/beta_used_raw": 0.1395997554063797, "beta_dpo/gap_mean": 151.24038696289062, "beta_dpo/gap_std": 161.67489624023438, "beta_dpo/loss_margin_mean": 186.27423095703125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.27312775330396477, "grad_norm": 1914.20361328125, "learning_rate": 4.5697190473557947e-07, "logits/chosen": -2.3745052814483643, "logits/rejected": -2.37973690032959, "loss": 0.7652, "step": 186 }, { "beta_dpo/beta": 0.9558632373809814, "beta_dpo/beta_margin_grad_mean": -0.28026333451271057, "beta_dpo/beta_margin_grad_std": 0.2789249122142792, "beta_dpo/beta_margin_mean": 181.9654083251953, "beta_dpo/beta_margin_std": 302.32769775390625, "beta_dpo/beta_used": 0.9558632373809814, "beta_dpo/beta_used_raw": 0.7472833395004272, "beta_dpo/gap_mean": 153.18206787109375, "beta_dpo/gap_std": 159.89886474609375, "beta_dpo/loss_margin_mean": 162.0480194091797, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2745961820851689, "grad_norm": 5661.1728515625, "learning_rate": 4.5624936864957555e-07, "logits/chosen": -2.2961974143981934, "logits/rejected": -2.298018217086792, "loss": 6.1816, "step": 187 }, { "beta_dpo/beta": 0.2359977513551712, "beta_dpo/beta_margin_grad_mean": -0.2686719000339508, "beta_dpo/beta_margin_grad_std": 0.2627968490123749, "beta_dpo/beta_margin_mean": 45.696014404296875, "beta_dpo/beta_margin_std": 83.678955078125, "beta_dpo/beta_used": 0.2359977513551712, "beta_dpo/beta_used_raw": 0.019898220896720886, "beta_dpo/gap_mean": 157.99917602539062, "beta_dpo/gap_std": 162.6558074951172, "beta_dpo/loss_margin_mean": 179.31515502929688, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.27606461086637296, "grad_norm": 1707.17724609375, "learning_rate": 4.5552139770089454e-07, "logits/chosen": -2.2876267433166504, "logits/rejected": -2.297708034515381, "loss": 0.92, "step": 188 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.456310510635376, "beta_dpo/beta_margin_grad_std": 0.04195929691195488, "beta_dpo/beta_margin_mean": 0.17710651457309723, "beta_dpo/beta_margin_std": 0.17350885272026062, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.4375971555709839, "beta_dpo/gap_mean": 161.61734008789062, "beta_dpo/gap_std": 166.82327270507812, "beta_dpo/loss_margin_mean": 177.10650634765625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2775330396475771, "grad_norm": 12.050735473632812, "learning_rate": 4.5478801107224794e-07, "logits/chosen": -2.290733814239502, "logits/rejected": -2.3053641319274902, "loss": 1.2584, "step": 189 }, { "beta_dpo/beta": 0.6101582646369934, "beta_dpo/beta_margin_grad_mean": -0.24652166664600372, "beta_dpo/beta_margin_grad_std": 0.2449052780866623, "beta_dpo/beta_margin_mean": 159.73216247558594, "beta_dpo/beta_margin_std": 244.68267822265625, "beta_dpo/beta_used": 0.6101582646369934, "beta_dpo/beta_used_raw": -0.6181744337081909, "beta_dpo/gap_mean": 165.5086669921875, "beta_dpo/gap_std": 169.28228759765625, "beta_dpo/loss_margin_mean": 198.18690490722656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2790014684287812, "grad_norm": 2421.037841796875, "learning_rate": 4.5404922808905543e-07, "logits/chosen": -2.3267829418182373, "logits/rejected": -2.345957040786743, "loss": 0.7114, "step": 190 }, { "beta_dpo/beta": 0.9563454389572144, "beta_dpo/beta_margin_grad_mean": -0.07813985645771027, "beta_dpo/beta_margin_grad_std": 0.268363893032074, "beta_dpo/beta_margin_mean": 211.70956420898438, "beta_dpo/beta_margin_std": 206.75030517578125, "beta_dpo/beta_used": 0.9563454389572144, "beta_dpo/beta_used_raw": 0.9563454389572144, "beta_dpo/gap_mean": 179.21434020996094, "beta_dpo/gap_std": 175.45181274414062, "beta_dpo/loss_margin_mean": 225.3310089111328, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.28046989720998533, "grad_norm": 3832.57177734375, "learning_rate": 4.5330506821893565e-07, "logits/chosen": -2.335681915283203, "logits/rejected": -2.3289167881011963, "loss": 6.3921, "step": 191 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.45723795890808105, "beta_dpo/beta_margin_grad_std": 0.04705416411161423, "beta_dpo/beta_margin_mean": 0.17367665469646454, "beta_dpo/beta_margin_std": 0.19400545954704285, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.6269910335540771, "beta_dpo/gap_mean": 177.93748474121094, "beta_dpo/gap_std": 179.75079345703125, "beta_dpo/loss_margin_mean": 173.67665100097656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.28193832599118945, "grad_norm": 13.0220365524292, "learning_rate": 4.5255555107119336e-07, "logits/chosen": -2.293377161026001, "logits/rejected": -2.295954465866089, "loss": 1.246, "step": 192 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4627629220485687, "beta_dpo/beta_margin_grad_std": 0.0443439707159996, "beta_dpo/beta_margin_mean": 0.1510872095823288, "beta_dpo/beta_margin_std": 0.18337495625019073, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.660752773284912, "beta_dpo/gap_mean": 175.26287841796875, "beta_dpo/gap_std": 181.79058837890625, "beta_dpo/loss_margin_mean": 151.0872039794922, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2834067547723935, "grad_norm": 11.858062744140625, "learning_rate": 4.5180069639630236e-07, "logits/chosen": -2.3066320419311523, "logits/rejected": -2.331085681915283, "loss": 1.2649, "step": 193 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.46572333574295044, "beta_dpo/beta_margin_grad_std": 0.03027997724711895, "beta_dpo/beta_margin_mean": 0.1380530595779419, "beta_dpo/beta_margin_std": 0.12353204935789108, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.2002902030944824, "beta_dpo/gap_mean": 170.08328247070312, "beta_dpo/gap_std": 173.58062744140625, "beta_dpo/loss_margin_mean": 138.0530548095703, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.28487518355359764, "grad_norm": 11.269763946533203, "learning_rate": 4.510405240853854e-07, "logits/chosen": -2.218247890472412, "logits/rejected": -2.219308376312256, "loss": 1.261, "step": 194 }, { "beta_dpo/beta": 0.7751163244247437, "beta_dpo/beta_margin_grad_mean": -0.09672369062900543, "beta_dpo/beta_margin_grad_std": 0.2833358645439148, "beta_dpo/beta_margin_mean": 158.73081970214844, "beta_dpo/beta_margin_std": 144.5747528076172, "beta_dpo/beta_used": 0.7751163244247437, "beta_dpo/beta_used_raw": 0.7751163244247437, "beta_dpo/gap_mean": 173.27928161621094, "beta_dpo/gap_std": 173.402587890625, "beta_dpo/loss_margin_mean": 206.54397583007812, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.28634361233480177, "grad_norm": 6923.83349609375, "learning_rate": 4.5027505416968985e-07, "logits/chosen": -2.3047163486480713, "logits/rejected": -2.3096752166748047, "loss": 5.0274, "step": 195 }, { "beta_dpo/beta": 0.239418163895607, "beta_dpo/beta_margin_grad_mean": -0.2947882413864136, "beta_dpo/beta_margin_grad_std": 0.2903721034526825, "beta_dpo/beta_margin_mean": 50.04564666748047, "beta_dpo/beta_margin_std": 79.44770050048828, "beta_dpo/beta_used": 0.239418163895607, "beta_dpo/beta_used_raw": -0.7751691341400146, "beta_dpo/gap_mean": 176.44851684570312, "beta_dpo/gap_std": 173.1087646484375, "beta_dpo/loss_margin_mean": 175.94964599609375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2878120411160059, "grad_norm": 2369.5244140625, "learning_rate": 4.495043068200599e-07, "logits/chosen": -2.306786060333252, "logits/rejected": -2.316213607788086, "loss": 3.7479, "step": 196 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.461849182844162, "beta_dpo/beta_margin_grad_std": 0.033273905515670776, "beta_dpo/beta_margin_mean": 0.15370342135429382, "beta_dpo/beta_margin_std": 0.13502883911132812, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.0447132587432861, "beta_dpo/gap_mean": 170.2903594970703, "beta_dpo/gap_std": 166.9288787841797, "beta_dpo/loss_margin_mean": 153.7034149169922, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.28928046989721, "grad_norm": 12.06312084197998, "learning_rate": 4.4872830234640493e-07, "logits/chosen": -2.2932043075561523, "logits/rejected": -2.2925753593444824, "loss": 1.2436, "step": 197 }, { "beta_dpo/beta": 0.40779924392700195, "beta_dpo/beta_margin_grad_mean": -0.2927795648574829, "beta_dpo/beta_margin_grad_std": 0.2834474742412567, "beta_dpo/beta_margin_mean": 91.81517028808594, "beta_dpo/beta_margin_std": 143.10154724121094, "beta_dpo/beta_used": 0.40779924392700195, "beta_dpo/beta_used_raw": -0.3979314863681793, "beta_dpo/gap_mean": 179.01412963867188, "beta_dpo/gap_std": 170.71841430664062, "beta_dpo/loss_margin_mean": 226.88037109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2907488986784141, "grad_norm": 5604.81884765625, "learning_rate": 4.479470611971645e-07, "logits/chosen": -2.319179058074951, "logits/rejected": -2.3143765926361084, "loss": 9.2093, "step": 198 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4550706148147583, "beta_dpo/beta_margin_grad_std": 0.04198264703154564, "beta_dpo/beta_margin_mean": 0.18165776133537292, "beta_dpo/beta_margin_std": 0.1720964014530182, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.750559687614441, "beta_dpo/gap_mean": 181.78192138671875, "beta_dpo/gap_std": 174.14816284179688, "beta_dpo/loss_margin_mean": 181.6577606201172, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2922173274596182, "grad_norm": 11.930063247680664, "learning_rate": 4.471606039587695e-07, "logits/chosen": -2.3009300231933594, "logits/rejected": -2.294772148132324, "loss": 1.2447, "step": 199 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.46233460307121277, "beta_dpo/beta_margin_grad_std": 0.056460920721292496, "beta_dpo/beta_margin_mean": 0.15314218401908875, "beta_dpo/beta_margin_std": 0.23340220749378204, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.7534046173095703, "beta_dpo/gap_mean": 178.69091796875, "beta_dpo/gap_std": 183.65151977539062, "beta_dpo/loss_margin_mean": 153.14218139648438, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2936857562408223, "grad_norm": 12.505084037780762, "learning_rate": 4.4636895135509966e-07, "logits/chosen": -2.274620532989502, "logits/rejected": -2.277723789215088, "loss": 1.265, "step": 200 }, { "epoch": 0.2936857562408223, "eval_beta_dpo/beta": 0.041580744087696075, "eval_beta_dpo/beta_margin_grad_mean": -0.4593558609485626, "eval_beta_dpo/beta_margin_grad_std": 0.06084528937935829, "eval_beta_dpo/beta_margin_mean": 8.074563980102539, "eval_beta_dpo/beta_margin_std": 10.459110260009766, "eval_beta_dpo/beta_used": 0.041580744087696075, "eval_beta_dpo/beta_used_raw": -3.920811414718628, "eval_beta_dpo/gap_mean": 175.91966247558594, "eval_beta_dpo/gap_std": 183.71018981933594, "eval_beta_dpo/loss_margin_mean": 108.90614318847656, "eval_beta_dpo/mask_keep_frac": 1.0, "eval_logits/chosen": -2.311591863632202, "eval_logits/rejected": -2.305934429168701, "eval_loss": 1.2116435766220093, "eval_runtime": 37.7898, "eval_samples_per_second": 61.895, "eval_steps_per_second": 1.958, "step": 200 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.45610305666923523, "beta_dpo/beta_margin_grad_std": 0.05195591226220131, "beta_dpo/beta_margin_mean": 0.17831824719905853, "beta_dpo/beta_margin_std": 0.21344201266765594, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.0059877634048462, "beta_dpo/gap_mean": 175.1689453125, "beta_dpo/gap_std": 188.05361938476562, "beta_dpo/loss_margin_mean": 178.3182373046875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.29515418502202645, "grad_norm": 14.165757179260254, "learning_rate": 4.455721242469372e-07, "logits/chosen": -2.3472347259521484, "logits/rejected": -2.3590919971466064, "loss": 1.2403, "step": 201 }, { "beta_dpo/beta": 0.16364361345767975, "beta_dpo/beta_margin_grad_mean": -0.28894707560539246, "beta_dpo/beta_margin_grad_std": 0.28408119082450867, "beta_dpo/beta_margin_mean": 30.87427520751953, "beta_dpo/beta_margin_std": 56.27296447753906, "beta_dpo/beta_used": 0.16364361345767975, "beta_dpo/beta_used_raw": -1.1798573732376099, "beta_dpo/gap_mean": 177.89190673828125, "beta_dpo/gap_std": 191.12405395507812, "beta_dpo/loss_margin_mean": 183.90859985351562, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.2966226138032305, "grad_norm": 2054.255126953125, "learning_rate": 4.4477014363141755e-07, "logits/chosen": -2.2713704109191895, "logits/rejected": -2.2855608463287354, "loss": 2.1346, "step": 202 }, { "beta_dpo/beta": 0.19677655398845673, "beta_dpo/beta_margin_grad_mean": -0.30733898282051086, "beta_dpo/beta_margin_grad_std": 0.29845014214515686, "beta_dpo/beta_margin_mean": 41.14481735229492, "beta_dpo/beta_margin_std": 74.64393615722656, "beta_dpo/beta_used": 0.19677655398845673, "beta_dpo/beta_used_raw": -0.3622090816497803, "beta_dpo/gap_mean": 180.37454223632812, "beta_dpo/gap_std": 190.1531524658203, "beta_dpo/loss_margin_mean": 188.07620239257812, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.29809104258443464, "grad_norm": 2106.6591796875, "learning_rate": 4.439630306414758e-07, "logits/chosen": -2.2831432819366455, "logits/rejected": -2.294023275375366, "loss": 1.9171, "step": 203 }, { "beta_dpo/beta": 0.037158019840717316, "beta_dpo/beta_margin_grad_mean": -0.2885894477367401, "beta_dpo/beta_margin_grad_std": 0.28164830803871155, "beta_dpo/beta_margin_mean": 6.227684497833252, "beta_dpo/beta_margin_std": 10.331753730773926, "beta_dpo/beta_used": 0.037158019840717316, "beta_dpo/beta_used_raw": -0.3006288409233093, "beta_dpo/gap_mean": 178.19146728515625, "beta_dpo/gap_std": 192.77317810058594, "beta_dpo/loss_margin_mean": 169.02920532226562, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.29955947136563876, "grad_norm": 258.5797119140625, "learning_rate": 4.431508065452897e-07, "logits/chosen": -2.3215532302856445, "logits/rejected": -2.3090624809265137, "loss": 0.9138, "step": 204 }, { "beta_dpo/beta": 1.7806830406188965, "beta_dpo/beta_margin_grad_mean": -0.07821886986494064, "beta_dpo/beta_margin_grad_std": 0.26834189891815186, "beta_dpo/beta_margin_mean": 394.4228210449219, "beta_dpo/beta_margin_std": 418.83489990234375, "beta_dpo/beta_used": 1.7806830406188965, "beta_dpo/beta_used_raw": 1.7806830406188965, "beta_dpo/gap_mean": 183.717529296875, "beta_dpo/gap_std": 191.0102081298828, "beta_dpo/loss_margin_mean": 210.12306213378906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3010279001468429, "grad_norm": 6468.51318359375, "learning_rate": 4.4233349274571974e-07, "logits/chosen": -2.2471261024475098, "logits/rejected": -2.2409067153930664, "loss": 7.2147, "step": 205 }, { "beta_dpo/beta": 0.5732086300849915, "beta_dpo/beta_margin_grad_mean": -0.14021912217140198, "beta_dpo/beta_margin_grad_std": 0.3377551734447479, "beta_dpo/beta_margin_mean": 113.2298583984375, "beta_dpo/beta_margin_std": 139.0940704345703, "beta_dpo/beta_used": 0.5732086300849915, "beta_dpo/beta_used_raw": 0.5732086300849915, "beta_dpo/gap_mean": 186.09320068359375, "beta_dpo/gap_std": 190.4658966064453, "beta_dpo/loss_margin_mean": 205.77224731445312, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.302496328928047, "grad_norm": 7734.55419921875, "learning_rate": 4.415111107797445e-07, "logits/chosen": -2.221261501312256, "logits/rejected": -2.21230411529541, "loss": 11.641, "step": 206 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.44699206948280334, "beta_dpo/beta_margin_grad_std": 0.04681679978966713, "beta_dpo/beta_margin_mean": 0.21536417305469513, "beta_dpo/beta_margin_std": 0.19422101974487305, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.9365102052688599, "beta_dpo/gap_mean": 191.98245239257812, "beta_dpo/gap_std": 192.3529052734375, "beta_dpo/loss_margin_mean": 215.36416625976562, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3039647577092511, "grad_norm": 13.226594924926758, "learning_rate": 4.4068368231789365e-07, "logits/chosen": -2.2933871746063232, "logits/rejected": -2.283078193664551, "loss": 1.2246, "step": 207 }, { "beta_dpo/beta": 0.8147650957107544, "beta_dpo/beta_margin_grad_mean": -0.09077098220586777, "beta_dpo/beta_margin_grad_std": 0.28280818462371826, "beta_dpo/beta_margin_mean": 174.40185546875, "beta_dpo/beta_margin_std": 171.0436248779297, "beta_dpo/beta_used": 0.8147650957107544, "beta_dpo/beta_used_raw": 0.8147650957107544, "beta_dpo/gap_mean": 196.13491821289062, "beta_dpo/gap_std": 191.99342346191406, "beta_dpo/loss_margin_mean": 211.70095825195312, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3054331864904552, "grad_norm": 5424.021484375, "learning_rate": 4.398512291636768e-07, "logits/chosen": -2.30989408493042, "logits/rejected": -2.319075584411621, "loss": 5.355, "step": 208 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.45670098066329956, "beta_dpo/beta_margin_grad_std": 0.041713543236255646, "beta_dpo/beta_margin_mean": 0.17479273676872253, "beta_dpo/beta_margin_std": 0.16900895535945892, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.6124803423881531, "beta_dpo/gap_mean": 193.11166381835938, "beta_dpo/gap_std": 188.4140625, "beta_dpo/loss_margin_mean": 174.792724609375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3069016152716593, "grad_norm": 12.918232917785645, "learning_rate": 4.3901377325300857e-07, "logits/chosen": -2.2634706497192383, "logits/rejected": -2.2491159439086914, "loss": 1.2183, "step": 209 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.45144328474998474, "beta_dpo/beta_margin_grad_std": 0.04742031916975975, "beta_dpo/beta_margin_mean": 0.19709433615207672, "beta_dpo/beta_margin_std": 0.19505690038204193, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.2010389566421509, "beta_dpo/gap_mean": 191.64120483398438, "beta_dpo/gap_std": 186.86257934570312, "beta_dpo/loss_margin_mean": 197.09432983398438, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.30837004405286345, "grad_norm": 13.08560848236084, "learning_rate": 4.381713366536311e-07, "logits/chosen": -2.2852721214294434, "logits/rejected": -2.291003704071045, "loss": 1.229, "step": 210 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4518587291240692, "beta_dpo/beta_margin_grad_std": 0.046131283044815063, "beta_dpo/beta_margin_mean": 0.19535738229751587, "beta_dpo/beta_margin_std": 0.1901794821023941, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.092252492904663, "beta_dpo/gap_mean": 192.7459716796875, "beta_dpo/gap_std": 187.72750854492188, "beta_dpo/loss_margin_mean": 195.3573760986328, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.30983847283406757, "grad_norm": 14.05884075164795, "learning_rate": 4.373239415645323e-07, "logits/chosen": -2.2702016830444336, "logits/rejected": -2.2552332878112793, "loss": 1.2252, "step": 211 }, { "beta_dpo/beta": 0.8706912398338318, "beta_dpo/beta_margin_grad_mean": -0.26792824268341064, "beta_dpo/beta_margin_grad_std": 0.27179181575775146, "beta_dpo/beta_margin_mean": 268.76666259765625, "beta_dpo/beta_margin_std": 430.33172607421875, "beta_dpo/beta_used": 0.8706912398338318, "beta_dpo/beta_used_raw": 0.801848292350769, "beta_dpo/gap_mean": 207.16200256347656, "beta_dpo/gap_std": 197.4569549560547, "beta_dpo/loss_margin_mean": 271.86383056640625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.31130690161527164, "grad_norm": 12608.0732421875, "learning_rate": 4.3647161031536086e-07, "logits/chosen": -2.2498250007629395, "logits/rejected": -2.2473154067993164, "loss": 28.9501, "step": 212 }, { "beta_dpo/beta": 0.3058314323425293, "beta_dpo/beta_margin_grad_mean": -0.30717548727989197, "beta_dpo/beta_margin_grad_std": 0.29951781034469604, "beta_dpo/beta_margin_mean": 63.14212417602539, "beta_dpo/beta_margin_std": 126.7613754272461, "beta_dpo/beta_used": 0.3058314323425293, "beta_dpo/beta_used_raw": -1.3592239618301392, "beta_dpo/gap_mean": 206.8724822998047, "beta_dpo/gap_std": 205.34677124023438, "beta_dpo/loss_margin_mean": 192.0325164794922, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.31277533039647576, "grad_norm": 3455.124267578125, "learning_rate": 4.3561436536583774e-07, "logits/chosen": -2.2697277069091797, "logits/rejected": -2.2519259452819824, "loss": 2.4062, "step": 213 }, { "beta_dpo/beta": 0.16119687259197235, "beta_dpo/beta_margin_grad_mean": -0.30010974407196045, "beta_dpo/beta_margin_grad_std": 0.28895044326782227, "beta_dpo/beta_margin_mean": 32.903358459472656, "beta_dpo/beta_margin_std": 53.89445877075195, "beta_dpo/beta_used": 0.16119687259197235, "beta_dpo/beta_used_raw": -0.6572102904319763, "beta_dpo/gap_mean": 200.41761779785156, "beta_dpo/gap_std": 208.61569213867188, "beta_dpo/loss_margin_mean": 179.22732543945312, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3142437591776799, "grad_norm": 883.9422607421875, "learning_rate": 4.3475222930516473e-07, "logits/chosen": -2.2414190769195557, "logits/rejected": -2.2436366081237793, "loss": 1.1557, "step": 214 }, { "beta_dpo/beta": 0.7592843770980835, "beta_dpo/beta_margin_grad_mean": -0.2566761374473572, "beta_dpo/beta_margin_grad_std": 0.26009878516197205, "beta_dpo/beta_margin_mean": 188.02699279785156, "beta_dpo/beta_margin_std": 280.02783203125, "beta_dpo/beta_used": 0.7592843770980835, "beta_dpo/beta_used_raw": 0.24615520238876343, "beta_dpo/gap_mean": 204.9820556640625, "beta_dpo/gap_std": 203.79104614257812, "beta_dpo/loss_margin_mean": 223.06326293945312, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.315712187958884, "grad_norm": 9.2144136428833, "learning_rate": 4.3388522485142885e-07, "logits/chosen": -2.2476065158843994, "logits/rejected": -2.237344741821289, "loss": 0.6064, "step": 215 }, { "beta_dpo/beta": 0.5309363007545471, "beta_dpo/beta_margin_grad_mean": -0.27667322754859924, "beta_dpo/beta_margin_grad_std": 0.27724382281303406, "beta_dpo/beta_margin_mean": 122.65149688720703, "beta_dpo/beta_margin_std": 183.43125915527344, "beta_dpo/beta_used": 0.5309363007545471, "beta_dpo/beta_used_raw": -1.3396328687667847, "beta_dpo/gap_mean": 202.26290893554688, "beta_dpo/gap_std": 199.85479736328125, "beta_dpo/loss_margin_mean": 198.3951873779297, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.31718061674008813, "grad_norm": 4384.37841796875, "learning_rate": 4.330133748510036e-07, "logits/chosen": -2.247248888015747, "logits/rejected": -2.2399697303771973, "loss": 2.7083, "step": 216 }, { "beta_dpo/beta": 0.16512493789196014, "beta_dpo/beta_margin_grad_mean": -0.28015637397766113, "beta_dpo/beta_margin_grad_std": 0.2829425632953644, "beta_dpo/beta_margin_mean": 38.849979400634766, "beta_dpo/beta_margin_std": 62.65520477294922, "beta_dpo/beta_used": 0.16512493789196014, "beta_dpo/beta_used_raw": -0.6432375311851501, "beta_dpo/gap_mean": 211.07345581054688, "beta_dpo/gap_std": 202.76217651367188, "beta_dpo/loss_margin_mean": 249.60948181152344, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3186490455212922, "grad_norm": 2745.58251953125, "learning_rate": 4.3213670227794757e-07, "logits/chosen": -2.2299795150756836, "logits/rejected": -2.2205770015716553, "loss": 5.3863, "step": 217 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4506628215312958, "beta_dpo/beta_margin_grad_std": 0.0487155057489872, "beta_dpo/beta_margin_mean": 0.20028528571128845, "beta_dpo/beta_margin_std": 0.2005169838666916, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.0722289085388184, "beta_dpo/gap_mean": 211.20944213867188, "beta_dpo/gap_std": 202.60032653808594, "beta_dpo/loss_margin_mean": 200.2852783203125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3201174743024963, "grad_norm": 13.857083320617676, "learning_rate": 4.3125523023339815e-07, "logits/chosen": -2.257091999053955, "logits/rejected": -2.25131893157959, "loss": 1.2226, "step": 218 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4540587365627289, "beta_dpo/beta_margin_grad_std": 0.060583606362342834, "beta_dpo/beta_margin_mean": 0.18906962871551514, "beta_dpo/beta_margin_std": 0.25983956456184387, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.0623416900634766, "beta_dpo/gap_mean": 204.9937744140625, "beta_dpo/gap_std": 208.16064453125, "beta_dpo/loss_margin_mean": 189.06961059570312, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.32158590308370044, "grad_norm": 13.54708194732666, "learning_rate": 4.303689819449636e-07, "logits/chosen": -2.2969722747802734, "logits/rejected": -2.297567367553711, "loss": 1.2297, "step": 219 }, { "beta_dpo/beta": 0.08469678461551666, "beta_dpo/beta_margin_grad_mean": -0.3169372081756592, "beta_dpo/beta_margin_grad_std": 0.3025418817996979, "beta_dpo/beta_margin_mean": 16.391508102416992, "beta_dpo/beta_margin_std": 27.10245132446289, "beta_dpo/beta_used": 0.08469678461551666, "beta_dpo/beta_used_raw": -0.6565529108047485, "beta_dpo/gap_mean": 204.47842407226562, "beta_dpo/gap_std": 211.26084899902344, "beta_dpo/loss_margin_mean": 193.17495727539062, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.32305433186490456, "grad_norm": 924.0001831054688, "learning_rate": 4.2947798076611047e-07, "logits/chosen": -2.285961151123047, "logits/rejected": -2.2831854820251465, "loss": 1.8538, "step": 220 }, { "beta_dpo/beta": 1.2386034727096558, "beta_dpo/beta_margin_grad_mean": -0.09203282743692398, "beta_dpo/beta_margin_grad_std": 0.28641635179519653, "beta_dpo/beta_margin_mean": 341.2391052246094, "beta_dpo/beta_margin_std": 419.7342834472656, "beta_dpo/beta_used": 1.2386034727096558, "beta_dpo/beta_used_raw": 1.2386034727096558, "beta_dpo/gap_mean": 214.00718688964844, "beta_dpo/gap_std": 216.42800903320312, "beta_dpo/loss_margin_mean": 272.1820373535156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3245227606461087, "grad_norm": 8449.5234375, "learning_rate": 4.285822501755485e-07, "logits/chosen": -2.263484239578247, "logits/rejected": -2.2643935680389404, "loss": 11.7807, "step": 221 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4465690553188324, "beta_dpo/beta_margin_grad_std": 0.05451636388897896, "beta_dpo/beta_margin_mean": 0.2180844247341156, "beta_dpo/beta_margin_std": 0.2271333932876587, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.0969059467315674, "beta_dpo/gap_mean": 215.2980499267578, "beta_dpo/gap_std": 219.42501831054688, "beta_dpo/loss_margin_mean": 218.08441162109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.32599118942731276, "grad_norm": 13.580673217773438, "learning_rate": 4.276818137766118e-07, "logits/chosen": -2.291238784790039, "logits/rejected": -2.294947624206543, "loss": 1.2209, "step": 222 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.44978615641593933, "beta_dpo/beta_margin_grad_std": 0.05852412432432175, "beta_dpo/beta_margin_mean": 0.20669816434383392, "beta_dpo/beta_margin_std": 0.2520856559276581, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.296399474143982, "beta_dpo/gap_mean": 216.46047973632812, "beta_dpo/gap_std": 226.12139892578125, "beta_dpo/loss_margin_mean": 206.69815063476562, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3274596182085169, "grad_norm": 13.991449356079102, "learning_rate": 4.2677669529663686e-07, "logits/chosen": -2.237861156463623, "logits/rejected": -2.2309041023254395, "loss": 1.2099, "step": 223 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.44978049397468567, "beta_dpo/beta_margin_grad_std": 0.05648142471909523, "beta_dpo/beta_margin_mean": 0.20587411522865295, "beta_dpo/beta_margin_std": 0.24117736518383026, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.2211096286773682, "beta_dpo/gap_mean": 214.611083984375, "beta_dpo/gap_std": 228.7374267578125, "beta_dpo/loss_margin_mean": 205.87411499023438, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.328928046989721, "grad_norm": 14.716062545776367, "learning_rate": 4.2586691858633747e-07, "logits/chosen": -2.2692012786865234, "logits/rejected": -2.2605040073394775, "loss": 1.2091, "step": 224 }, { "beta_dpo/beta": 0.866571843624115, "beta_dpo/beta_margin_grad_mean": -0.28175389766693115, "beta_dpo/beta_margin_grad_std": 0.284710556268692, "beta_dpo/beta_margin_mean": 206.28526306152344, "beta_dpo/beta_margin_std": 341.5513610839844, "beta_dpo/beta_used": 0.866571843624115, "beta_dpo/beta_used_raw": -0.3526184558868408, "beta_dpo/gap_mean": 217.59930419921875, "beta_dpo/gap_std": 227.94232177734375, "beta_dpo/loss_margin_mean": 245.15667724609375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3303964757709251, "grad_norm": 11778.642578125, "learning_rate": 4.249525076191759e-07, "logits/chosen": -2.27790904045105, "logits/rejected": -2.2668089866638184, "loss": 6.3533, "step": 225 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.45367875695228577, "beta_dpo/beta_margin_grad_std": 0.04829113930463791, "beta_dpo/beta_margin_mean": 0.18813456594944, "beta_dpo/beta_margin_std": 0.1991308182477951, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.7146835327148438, "beta_dpo/gap_mean": 216.79244995117188, "beta_dpo/gap_std": 225.35797119140625, "beta_dpo/loss_margin_mean": 188.1345672607422, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.33186490455212925, "grad_norm": 14.342082977294922, "learning_rate": 4.2403348649073167e-07, "logits/chosen": -2.3111863136291504, "logits/rejected": -2.2828011512756348, "loss": 1.2303, "step": 226 }, { "beta_dpo/beta": 0.3605048954486847, "beta_dpo/beta_margin_grad_mean": -0.28487730026245117, "beta_dpo/beta_margin_grad_std": 0.2841986417770386, "beta_dpo/beta_margin_mean": 109.1982421875, "beta_dpo/beta_margin_std": 195.55567932128906, "beta_dpo/beta_used": 0.3605048954486847, "beta_dpo/beta_used_raw": -0.9117497205734253, "beta_dpo/gap_mean": 221.8527069091797, "beta_dpo/gap_std": 232.91197204589844, "beta_dpo/loss_margin_mean": 260.5398864746094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3333333333333333, "grad_norm": 5748.3427734375, "learning_rate": 4.2310987941806615e-07, "logits/chosen": -2.3186283111572266, "logits/rejected": -2.2997145652770996, "loss": 6.4073, "step": 227 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4516187608242035, "beta_dpo/beta_margin_grad_std": 0.057563863694667816, "beta_dpo/beta_margin_mean": 0.1965341418981552, "beta_dpo/beta_margin_std": 0.23873193562030792, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.252966284751892, "beta_dpo/gap_mean": 217.8734130859375, "beta_dpo/gap_std": 229.7576141357422, "beta_dpo/loss_margin_mean": 196.5341339111328, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.33480176211453744, "grad_norm": 15.525106430053711, "learning_rate": 4.2218171073908463e-07, "logits/chosen": -2.3476057052612305, "logits/rejected": -2.339545726776123, "loss": 1.2061, "step": 228 }, { "beta_dpo/beta": 0.1788349598646164, "beta_dpo/beta_margin_grad_mean": -0.29465344548225403, "beta_dpo/beta_margin_grad_std": 0.28662359714508057, "beta_dpo/beta_margin_mean": 40.58890151977539, "beta_dpo/beta_margin_std": 64.144775390625, "beta_dpo/beta_used": 0.1788349598646164, "beta_dpo/beta_used_raw": -0.0582791268825531, "beta_dpo/gap_mean": 218.74749755859375, "beta_dpo/gap_std": 236.47198486328125, "beta_dpo/loss_margin_mean": 228.95513916015625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.33627019089574156, "grad_norm": 2251.75146484375, "learning_rate": 4.212490049118951e-07, "logits/chosen": -2.3584940433502197, "logits/rejected": -2.3436903953552246, "loss": 2.3098, "step": 229 }, { "beta_dpo/beta": 0.795495331287384, "beta_dpo/beta_margin_grad_mean": -0.2751595079898834, "beta_dpo/beta_margin_grad_std": 0.2707229256629944, "beta_dpo/beta_margin_mean": 202.500732421875, "beta_dpo/beta_margin_std": 299.9369201660156, "beta_dpo/beta_used": 0.795495331287384, "beta_dpo/beta_used_raw": 0.5070621967315674, "beta_dpo/gap_mean": 219.34637451171875, "beta_dpo/gap_std": 225.2493896484375, "beta_dpo/loss_margin_mean": 228.44595336914062, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3377386196769457, "grad_norm": 9665.818359375, "learning_rate": 4.203117865141635e-07, "logits/chosen": -2.2865986824035645, "logits/rejected": -2.294416904449463, "loss": 7.6845, "step": 230 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4476536810398102, "beta_dpo/beta_margin_grad_std": 0.04368142783641815, "beta_dpo/beta_margin_mean": 0.21227702498435974, "beta_dpo/beta_margin_std": 0.18027953803539276, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.8700859546661377, "beta_dpo/gap_mean": 220.33432006835938, "beta_dpo/gap_std": 214.647705078125, "beta_dpo/loss_margin_mean": 212.2770233154297, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3392070484581498, "grad_norm": 14.92269229888916, "learning_rate": 4.1937008024246625e-07, "logits/chosen": -2.3519227504730225, "logits/rejected": -2.328807830810547, "loss": 1.1986, "step": 231 }, { "beta_dpo/beta": 0.23005230724811554, "beta_dpo/beta_margin_grad_mean": -0.2875669300556183, "beta_dpo/beta_margin_grad_std": 0.2871084213256836, "beta_dpo/beta_margin_mean": 49.535980224609375, "beta_dpo/beta_margin_std": 79.29191589355469, "beta_dpo/beta_used": 0.23005230724811554, "beta_dpo/beta_used_raw": 0.0414634644985199, "beta_dpo/gap_mean": 218.3118133544922, "beta_dpo/gap_std": 214.149169921875, "beta_dpo/loss_margin_mean": 210.24900817871094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3406754772393539, "grad_norm": 2202.3125, "learning_rate": 4.1842391091163933e-07, "logits/chosen": -2.3467612266540527, "logits/rejected": -2.3430676460266113, "loss": 2.6911, "step": 232 }, { "beta_dpo/beta": 0.6968293786048889, "beta_dpo/beta_margin_grad_mean": -0.24002498388290405, "beta_dpo/beta_margin_grad_std": 0.2530527412891388, "beta_dpo/beta_margin_mean": 181.13375854492188, "beta_dpo/beta_margin_std": 268.4140930175781, "beta_dpo/beta_used": 0.6968293786048889, "beta_dpo/beta_used_raw": 0.5653680562973022, "beta_dpo/gap_mean": 228.13232421875, "beta_dpo/gap_std": 221.7039794921875, "beta_dpo/loss_margin_mean": 303.7350158691406, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.342143906020558, "grad_norm": 6852.927734375, "learning_rate": 4.174733034541245e-07, "logits/chosen": -2.345778226852417, "logits/rejected": -2.349656105041504, "loss": 9.6398, "step": 233 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4342879354953766, "beta_dpo/beta_margin_grad_std": 0.06734412163496017, "beta_dpo/beta_margin_mean": 0.2719317376613617, "beta_dpo/beta_margin_std": 0.2893328368663788, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.735541820526123, "beta_dpo/gap_mean": 242.57504272460938, "beta_dpo/gap_std": 239.12721252441406, "beta_dpo/loss_margin_mean": 271.9317321777344, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3436123348017621, "grad_norm": 14.259328842163086, "learning_rate": 4.165182829193126e-07, "logits/chosen": -2.3349199295043945, "logits/rejected": -2.356750965118408, "loss": 1.1938, "step": 234 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.44918081164360046, "beta_dpo/beta_margin_grad_std": 0.06389525532722473, "beta_dpo/beta_margin_mean": 0.20874598622322083, "beta_dpo/beta_margin_std": 0.26976871490478516, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.224062204360962, "beta_dpo/gap_mean": 234.73019409179688, "beta_dpo/gap_std": 243.65020751953125, "beta_dpo/loss_margin_mean": 208.74598693847656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.34508076358296624, "grad_norm": 15.3160400390625, "learning_rate": 4.1555887447288255e-07, "logits/chosen": -2.3654861450195312, "logits/rejected": -2.3537168502807617, "loss": 1.2086, "step": 235 }, { "beta_dpo/beta": 0.22831876575946808, "beta_dpo/beta_margin_grad_mean": -0.29526790976524353, "beta_dpo/beta_margin_grad_std": 0.2938448190689087, "beta_dpo/beta_margin_mean": 50.824398040771484, "beta_dpo/beta_margin_std": 78.6937255859375, "beta_dpo/beta_used": 0.22831876575946808, "beta_dpo/beta_used_raw": -0.8176271319389343, "beta_dpo/gap_mean": 235.43679809570312, "beta_dpo/gap_std": 244.1046905517578, "beta_dpo/loss_margin_mean": 243.5421142578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3465491923641703, "grad_norm": 3292.40771484375, "learning_rate": 4.1459510339613946e-07, "logits/chosen": -2.356201648712158, "logits/rejected": -2.3701233863830566, "loss": 4.0937, "step": 236 }, { "beta_dpo/beta": 1.0147018432617188, "beta_dpo/beta_margin_grad_mean": -0.25070029497146606, "beta_dpo/beta_margin_grad_std": 0.25762683153152466, "beta_dpo/beta_margin_mean": 303.3717956542969, "beta_dpo/beta_margin_std": 481.146240234375, "beta_dpo/beta_used": 1.0147018432617188, "beta_dpo/beta_used_raw": -0.4895740747451782, "beta_dpo/gap_mean": 241.46392822265625, "beta_dpo/gap_std": 249.50332641601562, "beta_dpo/loss_margin_mean": 275.3121032714844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.34801762114537443, "grad_norm": 11534.1455078125, "learning_rate": 4.136269950853473e-07, "logits/chosen": -2.380739212036133, "logits/rejected": -2.381621837615967, "loss": 14.4849, "step": 237 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4481091797351837, "beta_dpo/beta_margin_grad_std": 0.05575034022331238, "beta_dpo/beta_margin_mean": 0.21202851831912994, "beta_dpo/beta_margin_std": 0.23184596002101898, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.4691739082336426, "beta_dpo/gap_mean": 241.19456481933594, "beta_dpo/gap_std": 248.67156982421875, "beta_dpo/loss_margin_mean": 212.02850341796875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.34948604992657856, "grad_norm": 14.532193183898926, "learning_rate": 4.126545750510605e-07, "logits/chosen": -2.375584602355957, "logits/rejected": -2.3742737770080566, "loss": 1.2202, "step": 238 }, { "beta_dpo/beta": 1.912995457649231, "beta_dpo/beta_margin_grad_mean": -0.29015278816223145, "beta_dpo/beta_margin_grad_std": 0.2879817485809326, "beta_dpo/beta_margin_mean": 527.4285278320312, "beta_dpo/beta_margin_std": 877.66650390625, "beta_dpo/beta_used": 1.912995457649231, "beta_dpo/beta_used_raw": -0.6588293313980103, "beta_dpo/gap_mean": 238.7420654296875, "beta_dpo/gap_std": 245.08407592773438, "beta_dpo/loss_margin_mean": 228.3488311767578, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3509544787077827, "grad_norm": 22015.724609375, "learning_rate": 4.116778689174514e-07, "logits/chosen": -2.369143486022949, "logits/rejected": -2.358672857284546, "loss": 39.3869, "step": 239 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4522637724876404, "beta_dpo/beta_margin_grad_std": 0.04363469406962395, "beta_dpo/beta_margin_mean": 0.19304384291172028, "beta_dpo/beta_margin_std": 0.17740407586097717, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.5935862064361572, "beta_dpo/gap_mean": 231.2987060546875, "beta_dpo/gap_std": 231.66029357910156, "beta_dpo/loss_margin_mean": 193.04383850097656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3524229074889868, "grad_norm": 15.185770988464355, "learning_rate": 4.106969024216348e-07, "logits/chosen": -2.439711093902588, "logits/rejected": -2.4188616275787354, "loss": 1.2157, "step": 240 }, { "beta_dpo/beta": 0.7368063926696777, "beta_dpo/beta_margin_grad_mean": -0.2956547141075134, "beta_dpo/beta_margin_grad_std": 0.2953491806983948, "beta_dpo/beta_margin_mean": 150.56866455078125, "beta_dpo/beta_margin_std": 260.19439697265625, "beta_dpo/beta_used": 0.7368063926696777, "beta_dpo/beta_used_raw": -0.016419827938079834, "beta_dpo/gap_mean": 226.96133422851562, "beta_dpo/gap_std": 228.65875244140625, "beta_dpo/loss_margin_mean": 235.113037109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.35389133627019087, "grad_norm": 11925.3466796875, "learning_rate": 4.097117014129903e-07, "logits/chosen": -2.4186618328094482, "logits/rejected": -2.3823959827423096, "loss": 7.4844, "step": 241 }, { "beta_dpo/beta": 1.0837945938110352, "beta_dpo/beta_margin_grad_mean": -0.10944601148366928, "beta_dpo/beta_margin_grad_std": 0.3120846450328827, "beta_dpo/beta_margin_mean": 249.89736938476562, "beta_dpo/beta_margin_std": 290.0773010253906, "beta_dpo/beta_used": 1.0837945938110352, "beta_dpo/beta_used_raw": 1.0837945938110352, "beta_dpo/gap_mean": 229.24334716796875, "beta_dpo/gap_std": 231.86746215820312, "beta_dpo/loss_margin_mean": 231.20904541015625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.355359765051395, "grad_norm": 14227.9248046875, "learning_rate": 4.087222918524807e-07, "logits/chosen": -2.3967206478118896, "logits/rejected": -2.3583226203918457, "loss": 12.149, "step": 242 }, { "beta_dpo/beta": 0.29786860942840576, "beta_dpo/beta_margin_grad_mean": -0.23357638716697693, "beta_dpo/beta_margin_grad_std": 0.2237424999475479, "beta_dpo/beta_margin_mean": 74.67481231689453, "beta_dpo/beta_margin_std": 106.87940216064453, "beta_dpo/beta_used": 0.29786860942840576, "beta_dpo/beta_used_raw": 0.10270766913890839, "beta_dpo/gap_mean": 229.03512573242188, "beta_dpo/gap_std": 227.02081298828125, "beta_dpo/loss_margin_mean": 232.3207550048828, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3568281938325991, "grad_norm": 473.8674621582031, "learning_rate": 4.07728699811968e-07, "logits/chosen": -2.393955945968628, "logits/rejected": -2.363004684448242, "loss": 0.6125, "step": 243 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.44421160221099854, "beta_dpo/beta_margin_grad_std": 0.05300465598702431, "beta_dpo/beta_margin_mean": 0.227519690990448, "beta_dpo/beta_margin_std": 0.22171121835708618, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.1679985523223877, "beta_dpo/gap_mean": 230.13543701171875, "beta_dpo/gap_std": 226.2316131591797, "beta_dpo/loss_margin_mean": 227.51968383789062, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.35829662261380324, "grad_norm": 16.667314529418945, "learning_rate": 4.067309514735267e-07, "logits/chosen": -2.446463108062744, "logits/rejected": -2.438051223754883, "loss": 1.1932, "step": 244 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.44525429606437683, "beta_dpo/beta_margin_grad_std": 0.05737599730491638, "beta_dpo/beta_margin_mean": 0.2234031707048416, "beta_dpo/beta_margin_std": 0.2370459884405136, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.410043239593506, "beta_dpo/gap_mean": 229.35089111328125, "beta_dpo/gap_std": 226.95535278320312, "beta_dpo/loss_margin_mean": 223.40316772460938, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.35976505139500736, "grad_norm": 15.008759498596191, "learning_rate": 4.057290731287531e-07, "logits/chosen": -2.4402174949645996, "logits/rejected": -2.3967413902282715, "loss": 1.2316, "step": 245 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.44665589928627014, "beta_dpo/beta_margin_grad_std": 0.059832848608493805, "beta_dpo/beta_margin_mean": 0.21825571358203888, "beta_dpo/beta_margin_std": 0.2485092133283615, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.323253631591797, "beta_dpo/gap_mean": 227.4248046875, "beta_dpo/gap_std": 229.33834838867188, "beta_dpo/loss_margin_mean": 218.25570678710938, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.36123348017621143, "grad_norm": 14.934514999389648, "learning_rate": 4.047230911780736e-07, "logits/chosen": -2.4563374519348145, "logits/rejected": -2.428318500518799, "loss": 1.2305, "step": 246 }, { "beta_dpo/beta": 0.007803923450410366, "beta_dpo/beta_margin_grad_mean": -0.32257363200187683, "beta_dpo/beta_margin_grad_std": 0.2511254549026489, "beta_dpo/beta_margin_mean": 2.3317739963531494, "beta_dpo/beta_margin_std": 4.475132465362549, "beta_dpo/beta_used": 0.007803923450410366, "beta_dpo/beta_used_raw": -0.14297455549240112, "beta_dpo/gap_mean": 234.86566162109375, "beta_dpo/gap_std": 243.15460205078125, "beta_dpo/loss_margin_mean": 278.5243225097656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.36270190895741555, "grad_norm": 101.28236389160156, "learning_rate": 4.0371303213004814e-07, "logits/chosen": -2.4293994903564453, "logits/rejected": -2.430069923400879, "loss": 0.886, "step": 247 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.43503713607788086, "beta_dpo/beta_margin_grad_std": 0.05003669857978821, "beta_dpo/beta_margin_mean": 0.2657236158847809, "beta_dpo/beta_margin_std": 0.2158864438533783, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.8810434341430664, "beta_dpo/gap_mean": 241.89364624023438, "beta_dpo/gap_std": 237.35562133789062, "beta_dpo/loss_margin_mean": 265.7236022949219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3641703377386197, "grad_norm": 14.336139678955078, "learning_rate": 4.0269892260067197e-07, "logits/chosen": -2.402165174484253, "logits/rejected": -2.41436505317688, "loss": 1.1786, "step": 248 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.45203158259391785, "beta_dpo/beta_margin_grad_std": 0.06264989078044891, "beta_dpo/beta_margin_mean": 0.19835467636585236, "beta_dpo/beta_margin_std": 0.2707987129688263, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.373201847076416, "beta_dpo/gap_mean": 237.4274139404297, "beta_dpo/gap_std": 237.676513671875, "beta_dpo/loss_margin_mean": 198.35467529296875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3656387665198238, "grad_norm": 14.760684967041016, "learning_rate": 4.0168078931267426e-07, "logits/chosen": -2.412367343902588, "logits/rejected": -2.3927431106567383, "loss": 1.2226, "step": 249 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4453158974647522, "beta_dpo/beta_margin_grad_std": 0.05328512191772461, "beta_dpo/beta_margin_mean": 0.22280624508857727, "beta_dpo/beta_margin_std": 0.22093412280082703, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.016047477722168, "beta_dpo/gap_mean": 230.994140625, "beta_dpo/gap_std": 236.17393493652344, "beta_dpo/loss_margin_mean": 222.8062286376953, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3671071953010279, "grad_norm": 14.936260223388672, "learning_rate": 4.006586590948141e-07, "logits/chosen": -2.403975009918213, "logits/rejected": -2.3519978523254395, "loss": 1.2061, "step": 250 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4486519694328308, "beta_dpo/beta_margin_grad_std": 0.05090898275375366, "beta_dpo/beta_margin_mean": 0.20870457589626312, "beta_dpo/beta_margin_std": 0.2092035412788391, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.0009233951568604, "beta_dpo/gap_mean": 228.9009552001953, "beta_dpo/gap_std": 233.10757446289062, "beta_dpo/loss_margin_mean": 208.70457458496094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.368575624082232, "grad_norm": 14.669971466064453, "learning_rate": 3.9963255888117325e-07, "logits/chosen": -2.398078203201294, "logits/rejected": -2.362529754638672, "loss": 1.2071, "step": 251 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4456026554107666, "beta_dpo/beta_margin_grad_std": 0.0523262694478035, "beta_dpo/beta_margin_mean": 0.22128848731517792, "beta_dpo/beta_margin_std": 0.21624049544334412, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.0428235530853271, "beta_dpo/gap_mean": 224.8851318359375, "beta_dpo/gap_std": 229.4764404296875, "beta_dpo/loss_margin_mean": 221.28848266601562, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3700440528634361, "grad_norm": 16.351465225219727, "learning_rate": 3.9860251571044666e-07, "logits/chosen": -2.4359705448150635, "logits/rejected": -2.3960788249969482, "loss": 1.1988, "step": 252 }, { "beta_dpo/beta": 0.6227975487709045, "beta_dpo/beta_margin_grad_mean": -0.29470059275627136, "beta_dpo/beta_margin_grad_std": 0.29031190276145935, "beta_dpo/beta_margin_mean": 181.15090942382812, "beta_dpo/beta_margin_std": 269.0624084472656, "beta_dpo/beta_used": 0.6227975487709045, "beta_dpo/beta_used_raw": -1.4962077140808105, "beta_dpo/gap_mean": 220.82672119140625, "beta_dpo/gap_std": 225.36129760742188, "beta_dpo/loss_margin_mean": 216.6586456298828, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.37151248164464024, "grad_norm": 11820.3720703125, "learning_rate": 3.9756855672522986e-07, "logits/chosen": -2.4164085388183594, "logits/rejected": -2.4211513996124268, "loss": 4.0804, "step": 253 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.45039060711860657, "beta_dpo/beta_margin_grad_std": 0.05692875757813454, "beta_dpo/beta_margin_mean": 0.20209303498268127, "beta_dpo/beta_margin_std": 0.23470279574394226, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.512616753578186, "beta_dpo/gap_mean": 221.57920837402344, "beta_dpo/gap_std": 227.94493103027344, "beta_dpo/loss_margin_mean": 202.093017578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.37298091042584436, "grad_norm": 14.783519744873047, "learning_rate": 3.965307091713037e-07, "logits/chosen": -2.4098727703094482, "logits/rejected": -2.3902459144592285, "loss": 1.2072, "step": 254 }, { "beta_dpo/beta": 1.7512093782424927, "beta_dpo/beta_margin_grad_mean": -0.3025147020816803, "beta_dpo/beta_margin_grad_std": 0.2974975109100342, "beta_dpo/beta_margin_mean": 497.9840087890625, "beta_dpo/beta_margin_std": 947.763427734375, "beta_dpo/beta_used": 1.7512093782424927, "beta_dpo/beta_used_raw": 1.4928070306777954, "beta_dpo/gap_mean": 225.62283325195312, "beta_dpo/gap_std": 236.36817932128906, "beta_dpo/loss_margin_mean": 245.6376953125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3744493392070485, "grad_norm": 11111.490234375, "learning_rate": 3.954890003969163e-07, "logits/chosen": -2.3819503784179688, "logits/rejected": -2.3786773681640625, "loss": 0.9409, "step": 255 }, { "beta_dpo/beta": 1.2602046728134155, "beta_dpo/beta_margin_grad_mean": -0.2357739359140396, "beta_dpo/beta_margin_grad_std": 0.24245071411132812, "beta_dpo/beta_margin_mean": 373.910888671875, "beta_dpo/beta_margin_std": 559.2975463867188, "beta_dpo/beta_used": 1.2602046728134155, "beta_dpo/beta_used_raw": 0.8002493977546692, "beta_dpo/gap_mean": 230.1458740234375, "beta_dpo/gap_std": 236.79115295410156, "beta_dpo/loss_margin_mean": 270.7061767578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.37591776798825255, "grad_norm": 9.83940601348877, "learning_rate": 3.944434578520628e-07, "logits/chosen": -2.378755807876587, "logits/rejected": -2.3856723308563232, "loss": 0.597, "step": 256 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4384312331676483, "beta_dpo/beta_margin_grad_std": 0.06231582164764404, "beta_dpo/beta_margin_mean": 0.25216352939605713, "beta_dpo/beta_margin_std": 0.2593878209590912, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.7316312193870544, "beta_dpo/gap_mean": 237.4232177734375, "beta_dpo/gap_std": 242.20278930664062, "beta_dpo/loss_margin_mean": 252.16351318359375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.37738619676945667, "grad_norm": 15.36467456817627, "learning_rate": 3.933941090877615e-07, "logits/chosen": -2.391475200653076, "logits/rejected": -2.376940965652466, "loss": 1.1852, "step": 257 }, { "beta_dpo/beta": 2.6202783584594727, "beta_dpo/beta_margin_grad_mean": -0.12499994784593582, "beta_dpo/beta_margin_grad_std": 0.3307187855243683, "beta_dpo/beta_margin_mean": 712.9149169921875, "beta_dpo/beta_margin_std": 668.4194946289062, "beta_dpo/beta_used": 2.6202783584594727, "beta_dpo/beta_used_raw": 2.6202783584594727, "beta_dpo/gap_mean": 240.86416625976562, "beta_dpo/gap_std": 243.09378051757812, "beta_dpo/loss_margin_mean": 272.5574035644531, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3788546255506608, "grad_norm": 16114.0732421875, "learning_rate": 3.923409817553284e-07, "logits/chosen": -2.3731837272644043, "logits/rejected": -2.371438980102539, "loss": 16.2011, "step": 258 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4404573440551758, "beta_dpo/beta_margin_grad_std": 0.05528492480516434, "beta_dpo/beta_margin_mean": 0.24297261238098145, "beta_dpo/beta_margin_std": 0.22916458547115326, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.3427672386169434, "beta_dpo/gap_mean": 244.031982421875, "beta_dpo/gap_std": 241.36285400390625, "beta_dpo/loss_margin_mean": 242.97259521484375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3803230543318649, "grad_norm": 16.28805160522461, "learning_rate": 3.9128410360564793e-07, "logits/chosen": -2.386549949645996, "logits/rejected": -2.3812201023101807, "loss": 1.2012, "step": 259 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4323839843273163, "beta_dpo/beta_margin_grad_std": 0.0644708201289177, "beta_dpo/beta_margin_mean": 0.27984052896499634, "beta_dpo/beta_margin_std": 0.2821098566055298, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.7506465911865234, "beta_dpo/gap_mean": 248.44383239746094, "beta_dpo/gap_std": 245.54461669921875, "beta_dpo/loss_margin_mean": 279.84051513671875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.38179148311306904, "grad_norm": 17.50973892211914, "learning_rate": 3.9022350248844246e-07, "logits/chosen": -2.427947521209717, "logits/rejected": -2.4491896629333496, "loss": 1.1866, "step": 260 }, { "beta_dpo/beta": 0.3739962875843048, "beta_dpo/beta_margin_grad_mean": -0.0983000099658966, "beta_dpo/beta_margin_grad_std": 0.2856021821498871, "beta_dpo/beta_margin_mean": 113.49102020263672, "beta_dpo/beta_margin_std": 196.81541442871094, "beta_dpo/beta_used": 0.3739962875843048, "beta_dpo/beta_used_raw": 0.3739962875843048, "beta_dpo/gap_mean": 256.7048034667969, "beta_dpo/gap_std": 256.45654296875, "beta_dpo/loss_margin_mean": 285.4291076660156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3832599118942731, "grad_norm": 6361.6826171875, "learning_rate": 3.891592063515376e-07, "logits/chosen": -2.3608932495117188, "logits/rejected": -2.3539364337921143, "loss": 6.9612, "step": 261 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.44041234254837036, "beta_dpo/beta_margin_grad_std": 0.052026648074388504, "beta_dpo/beta_margin_mean": 0.24300675094127655, "beta_dpo/beta_margin_std": 0.21662083268165588, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.2271695137023926, "beta_dpo/gap_mean": 253.71353149414062, "beta_dpo/gap_std": 248.6503143310547, "beta_dpo/loss_margin_mean": 243.00674438476562, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.38472834067547723, "grad_norm": 17.864093780517578, "learning_rate": 3.880912432401264e-07, "logits/chosen": -2.4036097526550293, "logits/rejected": -2.3713953495025635, "loss": 1.1904, "step": 262 }, { "beta_dpo/beta": 1.3717470169067383, "beta_dpo/beta_margin_grad_mean": -0.2676565945148468, "beta_dpo/beta_margin_grad_std": 0.27245068550109863, "beta_dpo/beta_margin_mean": 446.4638366699219, "beta_dpo/beta_margin_std": 755.0382690429688, "beta_dpo/beta_used": 1.3717470169067383, "beta_dpo/beta_used_raw": -0.1793736219406128, "beta_dpo/gap_mean": 260.64508056640625, "beta_dpo/gap_std": 252.19851684570312, "beta_dpo/loss_margin_mean": 282.16845703125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.38619676945668135, "grad_norm": 10648.0771484375, "learning_rate": 3.870196412960302e-07, "logits/chosen": -2.421666145324707, "logits/rejected": -2.383852005004883, "loss": 4.4638, "step": 263 }, { "beta_dpo/beta": 1.3721956014633179, "beta_dpo/beta_margin_grad_mean": -0.2683139741420746, "beta_dpo/beta_margin_grad_std": 0.2728201150894165, "beta_dpo/beta_margin_mean": 423.73870849609375, "beta_dpo/beta_margin_std": 758.1284790039062, "beta_dpo/beta_used": 1.3721956014633179, "beta_dpo/beta_used_raw": 0.22174859046936035, "beta_dpo/gap_mean": 262.8730163574219, "beta_dpo/gap_std": 257.6596374511719, "beta_dpo/loss_margin_mean": 271.05548095703125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3876651982378855, "grad_norm": 5079.34765625, "learning_rate": 3.8594442875695665e-07, "logits/chosen": -2.4400243759155273, "logits/rejected": -2.442418098449707, "loss": 3.497, "step": 264 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4414041340351105, "beta_dpo/beta_margin_grad_std": 0.05608292669057846, "beta_dpo/beta_margin_mean": 0.23940885066986084, "beta_dpo/beta_margin_std": 0.23345738649368286, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.7446627616882324, "beta_dpo/gap_mean": 259.660888671875, "beta_dpo/gap_std": 252.59515380859375, "beta_dpo/loss_margin_mean": 239.40884399414062, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3891336270190896, "grad_norm": 18.96255111694336, "learning_rate": 3.848656339557562e-07, "logits/chosen": -2.427466869354248, "logits/rejected": -2.411646842956543, "loss": 1.1797, "step": 265 }, { "beta_dpo/beta": 0.16673444211483002, "beta_dpo/beta_margin_grad_mean": -0.27364593744277954, "beta_dpo/beta_margin_grad_std": 0.27338290214538574, "beta_dpo/beta_margin_mean": 39.5683479309082, "beta_dpo/beta_margin_std": 59.80198287963867, "beta_dpo/beta_used": 0.16673444211483002, "beta_dpo/beta_used_raw": -1.1203869581222534, "beta_dpo/gap_mean": 247.86456298828125, "beta_dpo/gap_std": 249.63990783691406, "beta_dpo/loss_margin_mean": 205.1346435546875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.39060205580029367, "grad_norm": 1790.510009765625, "learning_rate": 3.8378328531967507e-07, "logits/chosen": -2.421651840209961, "logits/rejected": -2.381192207336426, "loss": 2.2698, "step": 266 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4446525573730469, "beta_dpo/beta_margin_grad_std": 0.06158663332462311, "beta_dpo/beta_margin_mean": 0.22752374410629272, "beta_dpo/beta_margin_std": 0.2612149715423584, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.89312744140625, "beta_dpo/gap_mean": 245.2734375, "beta_dpo/gap_std": 250.494384765625, "beta_dpo/loss_margin_mean": 227.52374267578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3920704845814978, "grad_norm": 17.995609283447266, "learning_rate": 3.8269741136960646e-07, "logits/chosen": -2.4637513160705566, "logits/rejected": -2.419002056121826, "loss": 1.2079, "step": 267 }, { "beta_dpo/beta": 0.0013819551095366478, "beta_dpo/beta_margin_grad_mean": -0.4197249114513397, "beta_dpo/beta_margin_grad_std": 0.10671903938055038, "beta_dpo/beta_margin_mean": 0.35537663102149963, "beta_dpo/beta_margin_std": 0.5258194208145142, "beta_dpo/beta_used": 0.0013819551095366478, "beta_dpo/beta_used_raw": -0.5071589946746826, "beta_dpo/gap_mean": 242.955078125, "beta_dpo/gap_std": 258.65301513671875, "beta_dpo/loss_margin_mean": 249.29928588867188, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3935389133627019, "grad_norm": 23.119298934936523, "learning_rate": 3.8160804071933894e-07, "logits/chosen": -2.447445869445801, "logits/rejected": -2.4397740364074707, "loss": 1.1212, "step": 268 }, { "beta_dpo/beta": 2.2243704795837402, "beta_dpo/beta_margin_grad_mean": -0.2187625616788864, "beta_dpo/beta_margin_grad_std": 0.41339200735092163, "beta_dpo/beta_margin_mean": 724.3359985351562, "beta_dpo/beta_margin_std": 1003.6710205078125, "beta_dpo/beta_used": 2.2243704795837402, "beta_dpo/beta_used_raw": 2.2243704795837402, "beta_dpo/gap_mean": 251.21771240234375, "beta_dpo/gap_std": 269.38458251953125, "beta_dpo/loss_margin_mean": 301.1477966308594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.39500734214390604, "grad_norm": 53991.6875, "learning_rate": 3.8051520207480204e-07, "logits/chosen": -2.4700229167938232, "logits/rejected": -2.456906318664551, "loss": 69.1583, "step": 269 }, { "beta_dpo/beta": 2.0886917114257812, "beta_dpo/beta_margin_grad_mean": -0.10937510430812836, "beta_dpo/beta_margin_grad_std": 0.3121090829372406, "beta_dpo/beta_margin_mean": 528.6803588867188, "beta_dpo/beta_margin_std": 832.2965087890625, "beta_dpo/beta_used": 2.0886917114257812, "beta_dpo/beta_used_raw": 2.0886917114257812, "beta_dpo/gap_mean": 256.38360595703125, "beta_dpo/gap_std": 277.2386474609375, "beta_dpo/loss_margin_mean": 266.50457763671875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.3964757709251101, "grad_norm": 28465.341796875, "learning_rate": 3.794189242333106e-07, "logits/chosen": -2.5167593955993652, "logits/rejected": -2.494492530822754, "loss": 34.1063, "step": 270 }, { "beta_dpo/beta": 0.8449544310569763, "beta_dpo/beta_margin_grad_mean": -0.26794520020484924, "beta_dpo/beta_margin_grad_std": 0.2721104919910431, "beta_dpo/beta_margin_mean": 248.46165466308594, "beta_dpo/beta_margin_std": 399.88555908203125, "beta_dpo/beta_used": 0.8449544310569763, "beta_dpo/beta_used_raw": 0.3260602355003357, "beta_dpo/gap_mean": 260.4082336425781, "beta_dpo/gap_std": 271.07489013671875, "beta_dpo/loss_margin_mean": 264.7347412109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.39794419970631423, "grad_norm": 8032.013671875, "learning_rate": 3.7831923608280514e-07, "logits/chosen": -2.4501864910125732, "logits/rejected": -2.4311776161193848, "loss": 9.2538, "step": 271 }, { "beta_dpo/beta": 2.243833541870117, "beta_dpo/beta_margin_grad_mean": -0.0751146674156189, "beta_dpo/beta_margin_grad_std": 0.25838276743888855, "beta_dpo/beta_margin_mean": 774.5305786132812, "beta_dpo/beta_margin_std": 909.3794555664062, "beta_dpo/beta_used": 2.243833541870117, "beta_dpo/beta_used_raw": 2.243833541870117, "beta_dpo/gap_mean": 272.1531982421875, "beta_dpo/gap_std": 264.27728271484375, "beta_dpo/loss_margin_mean": 351.419677734375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.39941262848751835, "grad_norm": 15859.11328125, "learning_rate": 3.772161666010912e-07, "logits/chosen": -2.4539542198181152, "logits/rejected": -2.4497289657592773, "loss": 6.7279, "step": 272 }, { "beta_dpo/beta": 0.32312482595443726, "beta_dpo/beta_margin_grad_mean": -0.23232722282409668, "beta_dpo/beta_margin_grad_std": 0.2386544793844223, "beta_dpo/beta_margin_mean": 110.91842651367188, "beta_dpo/beta_margin_std": 163.00204467773438, "beta_dpo/beta_used": 0.32312482595443726, "beta_dpo/beta_used_raw": -0.6652578115463257, "beta_dpo/gap_mean": 282.7765808105469, "beta_dpo/gap_std": 264.1724853515625, "beta_dpo/loss_margin_mean": 307.410400390625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4008810572687225, "grad_norm": 4223.08251953125, "learning_rate": 3.761097448550755e-07, "logits/chosen": -2.448002815246582, "logits/rejected": -2.430415630340576, "loss": 5.6684, "step": 273 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.436676561832428, "beta_dpo/beta_margin_grad_std": 0.050382133573293686, "beta_dpo/beta_margin_mean": 0.25775137543678284, "beta_dpo/beta_margin_std": 0.2082599699497223, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.682006597518921, "beta_dpo/gap_mean": 278.31329345703125, "beta_dpo/gap_std": 254.832763671875, "beta_dpo/loss_margin_mean": 257.7513427734375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4023494860499266, "grad_norm": 16.958667755126953, "learning_rate": 3.75e-07, "logits/chosen": -2.454503297805786, "logits/rejected": -2.440640926361084, "loss": 1.1892, "step": 274 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4363858103752136, "beta_dpo/beta_margin_grad_std": 0.06395233422517776, "beta_dpo/beta_margin_mean": 0.26140275597572327, "beta_dpo/beta_margin_std": 0.2691759765148163, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.7606722116470337, "beta_dpo/gap_mean": 273.5849914550781, "beta_dpo/gap_std": 255.31951904296875, "beta_dpo/loss_margin_mean": 261.4027404785156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.40381791483113066, "grad_norm": 15.986796379089355, "learning_rate": 3.738869612786737e-07, "logits/chosen": -2.4930272102355957, "logits/rejected": -2.4809093475341797, "loss": 1.1688, "step": 275 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.45231759548187256, "beta_dpo/beta_margin_grad_std": 0.05670015886425972, "beta_dpo/beta_margin_mean": 0.19459043443202972, "beta_dpo/beta_margin_std": 0.2346203327178955, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.28926682472229, "beta_dpo/gap_mean": 264.0999755859375, "beta_dpo/gap_std": 254.43577575683594, "beta_dpo/loss_margin_mean": 194.59042358398438, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4052863436123348, "grad_norm": 16.828632354736328, "learning_rate": 3.7277065802070204e-07, "logits/chosen": -2.4924230575561523, "logits/rejected": -2.454463005065918, "loss": 1.1837, "step": 276 }, { "beta_dpo/beta": 2.115616798400879, "beta_dpo/beta_margin_grad_mean": -0.10937541723251343, "beta_dpo/beta_margin_grad_std": 0.31210899353027344, "beta_dpo/beta_margin_mean": 563.19384765625, "beta_dpo/beta_margin_std": 618.1759643554688, "beta_dpo/beta_used": 2.115616798400879, "beta_dpo/beta_used_raw": 2.115616798400879, "beta_dpo/gap_mean": 261.0350646972656, "beta_dpo/gap_std": 248.45208740234375, "beta_dpo/loss_margin_mean": 268.9210510253906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4067547723935389, "grad_norm": 6979.076171875, "learning_rate": 3.71651119641714e-07, "logits/chosen": -2.4732658863067627, "logits/rejected": -2.4552745819091797, "loss": 4.5947, "step": 277 }, { "beta_dpo/beta": 0.25115522742271423, "beta_dpo/beta_margin_grad_mean": -0.26586809754371643, "beta_dpo/beta_margin_grad_std": 0.27073895931243896, "beta_dpo/beta_margin_mean": 77.13081359863281, "beta_dpo/beta_margin_std": 126.1852798461914, "beta_dpo/beta_used": 0.25115522742271423, "beta_dpo/beta_used_raw": -1.3573105335235596, "beta_dpo/gap_mean": 265.0778503417969, "beta_dpo/gap_std": 248.96527099609375, "beta_dpo/loss_margin_mean": 279.1919250488281, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.40822320117474303, "grad_norm": 2251.078857421875, "learning_rate": 3.705283756425872e-07, "logits/chosen": -2.524477481842041, "logits/rejected": -2.508835792541504, "loss": 2.4695, "step": 278 }, { "beta_dpo/beta": 0.060165222734212875, "beta_dpo/beta_margin_grad_mean": -0.24480833113193512, "beta_dpo/beta_margin_grad_std": 0.24604347348213196, "beta_dpo/beta_margin_mean": 20.768430709838867, "beta_dpo/beta_margin_std": 33.017086029052734, "beta_dpo/beta_used": 0.060165222734212875, "beta_dpo/beta_used_raw": -1.1045668125152588, "beta_dpo/gap_mean": 271.9744873046875, "beta_dpo/gap_std": 254.4259490966797, "beta_dpo/loss_margin_mean": 303.9395751953125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.40969162995594716, "grad_norm": 422.1612243652344, "learning_rate": 3.6940245560867e-07, "logits/chosen": -2.4369864463806152, "logits/rejected": -2.438380718231201, "loss": 0.6394, "step": 279 }, { "beta_dpo/beta": 1.2358628511428833, "beta_dpo/beta_margin_grad_mean": -0.12507064640522003, "beta_dpo/beta_margin_grad_std": 0.3306926190853119, "beta_dpo/beta_margin_mean": 338.0630187988281, "beta_dpo/beta_margin_std": 364.5479431152344, "beta_dpo/beta_used": 1.2358628511428833, "beta_dpo/beta_used_raw": 1.2358628511428833, "beta_dpo/gap_mean": 274.9402770996094, "beta_dpo/gap_std": 252.64535522460938, "beta_dpo/loss_margin_mean": 287.3036804199219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4111600587371512, "grad_norm": 10145.9853515625, "learning_rate": 3.6827338920900253e-07, "logits/chosen": -2.4949872493743896, "logits/rejected": -2.48579478263855, "loss": 9.5901, "step": 280 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4418678879737854, "beta_dpo/beta_margin_grad_std": 0.05114522576332092, "beta_dpo/beta_margin_mean": 0.23657526075839996, "beta_dpo/beta_margin_std": 0.21189118921756744, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.398056745529175, "beta_dpo/gap_mean": 269.671630859375, "beta_dpo/gap_std": 247.16676330566406, "beta_dpo/loss_margin_mean": 236.5752410888672, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.41262848751835535, "grad_norm": 14.78842544555664, "learning_rate": 3.6714120619553435e-07, "logits/chosen": -2.5062947273254395, "logits/rejected": -2.4647984504699707, "loss": 1.1774, "step": 281 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.43719446659088135, "beta_dpo/beta_margin_grad_std": 0.06138541176915169, "beta_dpo/beta_margin_mean": 0.2587386667728424, "beta_dpo/beta_margin_std": 0.26557499170303345, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.9040853977203369, "beta_dpo/gap_mean": 265.92706298828125, "beta_dpo/gap_std": 249.60317993164062, "beta_dpo/loss_margin_mean": 258.7386474609375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.41409691629955947, "grad_norm": 16.898693084716797, "learning_rate": 3.660059364023408e-07, "logits/chosen": -2.499563217163086, "logits/rejected": -2.4865164756774902, "loss": 1.1584, "step": 282 }, { "beta_dpo/beta": 1.9231171607971191, "beta_dpo/beta_margin_grad_mean": -0.07813180983066559, "beta_dpo/beta_margin_grad_std": 0.2683661878108978, "beta_dpo/beta_margin_mean": 601.787841796875, "beta_dpo/beta_margin_std": 425.41082763671875, "beta_dpo/beta_used": 1.9231171607971191, "beta_dpo/beta_used_raw": 1.9231171607971191, "beta_dpo/gap_mean": 271.40765380859375, "beta_dpo/gap_std": 244.66708374023438, "beta_dpo/loss_margin_mean": 318.6204528808594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4155653450807636, "grad_norm": 14138.8076171875, "learning_rate": 3.6486760974483685e-07, "logits/chosen": -2.5169782638549805, "logits/rejected": -2.4992263317108154, "loss": 8.8542, "step": 283 }, { "beta_dpo/beta": 0.9466410875320435, "beta_dpo/beta_margin_grad_mean": -0.25939232110977173, "beta_dpo/beta_margin_grad_std": 0.26850318908691406, "beta_dpo/beta_margin_mean": 256.5771484375, "beta_dpo/beta_margin_std": 401.2411804199219, "beta_dpo/beta_used": 0.9466410875320435, "beta_dpo/beta_used_raw": 0.6345901489257812, "beta_dpo/gap_mean": 276.92742919921875, "beta_dpo/gap_std": 244.3406982421875, "beta_dpo/loss_margin_mean": 289.7724609375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4170337738619677, "grad_norm": 7450.4580078125, "learning_rate": 3.6372625621898863e-07, "logits/chosen": -2.505162239074707, "logits/rejected": -2.4776804447174072, "loss": 7.3208, "step": 284 }, { "beta_dpo/beta": 1.4398752450942993, "beta_dpo/beta_margin_grad_mean": -0.078125, "beta_dpo/beta_margin_grad_std": 0.2683681845664978, "beta_dpo/beta_margin_mean": 387.22125244140625, "beta_dpo/beta_margin_std": 445.4748229980469, "beta_dpo/beta_used": 1.4398752450942993, "beta_dpo/beta_used_raw": 1.4398752450942993, "beta_dpo/gap_mean": 278.3759765625, "beta_dpo/gap_std": 243.76156616210938, "beta_dpo/loss_margin_mean": 285.8123779296875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4185022026431718, "grad_norm": 20105.54296875, "learning_rate": 3.625819059005228e-07, "logits/chosen": -2.5132524967193604, "logits/rejected": -2.4971303939819336, "loss": 20.075, "step": 285 }, { "beta_dpo/beta": 1.556891679763794, "beta_dpo/beta_margin_grad_mean": -0.12499526143074036, "beta_dpo/beta_margin_grad_std": 0.330706387758255, "beta_dpo/beta_margin_mean": 447.1572570800781, "beta_dpo/beta_margin_std": 412.15057373046875, "beta_dpo/beta_used": 1.556891679763794, "beta_dpo/beta_used_raw": 1.556891679763794, "beta_dpo/gap_mean": 282.3958740234375, "beta_dpo/gap_std": 245.08740234375, "beta_dpo/loss_margin_mean": 289.4398193359375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4199706314243759, "grad_norm": 10238.416015625, "learning_rate": 3.614345889441346e-07, "logits/chosen": -2.477703094482422, "logits/rejected": -2.4490466117858887, "loss": 1.207, "step": 286 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.450316458940506, "beta_dpo/beta_margin_grad_std": 0.058557696640491486, "beta_dpo/beta_margin_mean": 0.20236189663410187, "beta_dpo/beta_margin_std": 0.24117599427700043, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.1759114265441895, "beta_dpo/gap_mean": 271.0549621582031, "beta_dpo/gap_std": 246.55422973632812, "beta_dpo/loss_margin_mean": 202.36187744140625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.42143906020558003, "grad_norm": 15.727797508239746, "learning_rate": 3.6028433558269275e-07, "logits/chosen": -2.454252004623413, "logits/rejected": -2.4215316772460938, "loss": 1.1749, "step": 287 }, { "beta_dpo/beta": 1.5804258584976196, "beta_dpo/beta_margin_grad_mean": -0.22239087522029877, "beta_dpo/beta_margin_grad_std": 0.22534386813640594, "beta_dpo/beta_margin_mean": 554.5604858398438, "beta_dpo/beta_margin_std": 696.531982421875, "beta_dpo/beta_used": 1.5804258584976196, "beta_dpo/beta_used_raw": 0.8364652991294861, "beta_dpo/gap_mean": 272.4178161621094, "beta_dpo/gap_std": 239.0502471923828, "beta_dpo/loss_margin_mean": 287.3144226074219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.42290748898678415, "grad_norm": 12.860248565673828, "learning_rate": 3.5913117612644327e-07, "logits/chosen": -2.441218137741089, "logits/rejected": -2.4160122871398926, "loss": 0.5814, "step": 288 }, { "beta_dpo/beta": 0.12335401028394699, "beta_dpo/beta_margin_grad_mean": -0.2808164358139038, "beta_dpo/beta_margin_grad_std": 0.2781689763069153, "beta_dpo/beta_margin_mean": 38.710533142089844, "beta_dpo/beta_margin_std": 60.30266189575195, "beta_dpo/beta_used": 0.12335401028394699, "beta_dpo/beta_used_raw": -2.816128730773926, "beta_dpo/gap_mean": 267.7777404785156, "beta_dpo/gap_std": 242.87863159179688, "beta_dpo/loss_margin_mean": 269.4267883300781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4243759177679883, "grad_norm": 1399.761962890625, "learning_rate": 3.5797514096221024e-07, "logits/chosen": -2.432061195373535, "logits/rejected": -2.416738986968994, "loss": 2.1497, "step": 289 }, { "beta_dpo/beta": 0.4147535264492035, "beta_dpo/beta_margin_grad_mean": -0.257093608379364, "beta_dpo/beta_margin_grad_std": 0.2680981159210205, "beta_dpo/beta_margin_mean": 118.29973602294922, "beta_dpo/beta_margin_std": 170.95254516601562, "beta_dpo/beta_used": 0.4147535264492035, "beta_dpo/beta_used_raw": -1.4733829498291016, "beta_dpo/gap_mean": 274.8150634765625, "beta_dpo/gap_std": 246.7188720703125, "beta_dpo/loss_margin_mean": 309.44354248046875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.42584434654919234, "grad_norm": 4860.09814453125, "learning_rate": 3.568162605525952e-07, "logits/chosen": -2.382627010345459, "logits/rejected": -2.3980727195739746, "loss": 0.8021, "step": 290 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.44023028016090393, "beta_dpo/beta_margin_grad_std": 0.05959730222821236, "beta_dpo/beta_margin_mean": 0.24392713606357574, "beta_dpo/beta_margin_std": 0.24562755227088928, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.8497285842895508, "beta_dpo/gap_mean": 271.6788635253906, "beta_dpo/gap_std": 250.40158081054688, "beta_dpo/loss_margin_mean": 243.9271240234375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.42731277533039647, "grad_norm": 17.112838745117188, "learning_rate": 3.5565456543517485e-07, "logits/chosen": -2.4613046646118164, "logits/rejected": -2.4365696907043457, "loss": 1.1703, "step": 291 }, { "beta_dpo/beta": 0.21685589849948883, "beta_dpo/beta_margin_grad_mean": -0.3011167347431183, "beta_dpo/beta_margin_grad_std": 0.28959137201309204, "beta_dpo/beta_margin_mean": 62.53536605834961, "beta_dpo/beta_margin_std": 106.66192626953125, "beta_dpo/beta_used": 0.21685589849948883, "beta_dpo/beta_used_raw": 0.09813689440488815, "beta_dpo/gap_mean": 274.6377868652344, "beta_dpo/gap_std": 246.9325408935547, "beta_dpo/loss_margin_mean": 290.8916015625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4287812041116006, "grad_norm": 5888.60498046875, "learning_rate": 3.5449008622169583e-07, "logits/chosen": -2.3982367515563965, "logits/rejected": -2.3866944313049316, "loss": 4.748, "step": 292 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4417594075202942, "beta_dpo/beta_margin_grad_std": 0.06032756716012955, "beta_dpo/beta_margin_mean": 0.23812739551067352, "beta_dpo/beta_margin_std": 0.24970707297325134, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.6080701351165771, "beta_dpo/gap_mean": 269.470703125, "beta_dpo/gap_std": 247.3360137939453, "beta_dpo/loss_margin_mean": 238.1273956298828, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4302496328928047, "grad_norm": 15.048643112182617, "learning_rate": 3.5332285359726846e-07, "logits/chosen": -2.4533298015594482, "logits/rejected": -2.4286036491394043, "loss": 1.1695, "step": 293 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4407859146595001, "beta_dpo/beta_margin_grad_std": 0.053464584052562714, "beta_dpo/beta_margin_mean": 0.2411770522594452, "beta_dpo/beta_margin_std": 0.22201335430145264, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.280418872833252, "beta_dpo/gap_mean": 265.29278564453125, "beta_dpo/gap_std": 245.94830322265625, "beta_dpo/loss_margin_mean": 241.1770477294922, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.43171806167400884, "grad_norm": 17.06812286376953, "learning_rate": 3.5215289831955786e-07, "logits/chosen": -2.3770015239715576, "logits/rejected": -2.384579658508301, "loss": 1.1791, "step": 294 }, { "beta_dpo/beta": 2.3084752559661865, "beta_dpo/beta_margin_grad_mean": -0.2641025185585022, "beta_dpo/beta_margin_grad_std": 0.2718868553638458, "beta_dpo/beta_margin_mean": 686.6762084960938, "beta_dpo/beta_margin_std": 1093.60888671875, "beta_dpo/beta_used": 2.3084752559661865, "beta_dpo/beta_used_raw": 0.6230998039245605, "beta_dpo/gap_mean": 266.80670166015625, "beta_dpo/gap_std": 250.25885009765625, "beta_dpo/loss_margin_mean": 285.25213623046875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4331864904552129, "grad_norm": 11.919480323791504, "learning_rate": 3.509802512179737e-07, "logits/chosen": -2.370579719543457, "logits/rejected": -2.3648414611816406, "loss": 0.5964, "step": 295 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.44102582335472107, "beta_dpo/beta_margin_grad_std": 0.06035232171416283, "beta_dpo/beta_margin_mean": 0.24229387938976288, "beta_dpo/beta_margin_std": 0.2558496296405792, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.7434802055358887, "beta_dpo/gap_mean": 264.58294677734375, "beta_dpo/gap_std": 255.38644409179688, "beta_dpo/loss_margin_mean": 242.2938690185547, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.434654919236417, "grad_norm": 16.63273811340332, "learning_rate": 3.498049431928577e-07, "logits/chosen": -2.4266653060913086, "logits/rejected": -2.3969693183898926, "loss": 1.1781, "step": 296 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.43483033776283264, "beta_dpo/beta_margin_grad_std": 0.05599094182252884, "beta_dpo/beta_margin_mean": 0.266786128282547, "beta_dpo/beta_margin_std": 0.2364174723625183, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.5507869720458984, "beta_dpo/gap_mean": 262.3899841308594, "beta_dpo/gap_std": 251.85415649414062, "beta_dpo/loss_margin_mean": 266.7861022949219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.43612334801762115, "grad_norm": 17.059518814086914, "learning_rate": 3.486270052146694e-07, "logits/chosen": -2.384669303894043, "logits/rejected": -2.378746509552002, "loss": 1.1732, "step": 297 }, { "beta_dpo/beta": 2.586780071258545, "beta_dpo/beta_margin_grad_mean": -0.2367078959941864, "beta_dpo/beta_margin_grad_std": 0.25049635767936707, "beta_dpo/beta_margin_mean": 772.3213500976562, "beta_dpo/beta_margin_std": 1197.1170654296875, "beta_dpo/beta_used": 2.586780071258545, "beta_dpo/beta_used_raw": 2.4882137775421143, "beta_dpo/gap_mean": 272.2360534667969, "beta_dpo/gap_std": 257.9064636230469, "beta_dpo/loss_margin_mean": 338.0078430175781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.43759177679882527, "grad_norm": 11.470409393310547, "learning_rate": 3.474464683231698e-07, "logits/chosen": -2.401449680328369, "logits/rejected": -2.4212589263916016, "loss": 0.571, "step": 298 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4342046082019806, "beta_dpo/beta_margin_grad_std": 0.055606916546821594, "beta_dpo/beta_margin_mean": 0.26877033710479736, "beta_dpo/beta_margin_std": 0.2329106479883194, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.441830039024353, "beta_dpo/gap_mean": 274.23779296875, "beta_dpo/gap_std": 258.99835205078125, "beta_dpo/loss_margin_mean": 268.77032470703125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4390602055800294, "grad_norm": 15.817339897155762, "learning_rate": 3.462633636266041e-07, "logits/chosen": -2.361654281616211, "logits/rejected": -2.373199224472046, "loss": 1.1586, "step": 299 }, { "beta_dpo/beta": 1.0433894395828247, "beta_dpo/beta_margin_grad_mean": -0.2306303083896637, "beta_dpo/beta_margin_grad_std": 0.24043621122837067, "beta_dpo/beta_margin_mean": 355.8514709472656, "beta_dpo/beta_margin_std": 520.4441528320312, "beta_dpo/beta_used": 1.0433894395828247, "beta_dpo/beta_used_raw": -0.06062328815460205, "beta_dpo/gap_mean": 281.2528991699219, "beta_dpo/gap_std": 265.603759765625, "beta_dpo/loss_margin_mean": 316.542236328125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.44052863436123346, "grad_norm": 14.615754127502441, "learning_rate": 3.4507772230088147e-07, "logits/chosen": -2.371137857437134, "logits/rejected": -2.378791570663452, "loss": 0.5857, "step": 300 }, { "epoch": 0.44052863436123346, "eval_beta_dpo/beta": 0.0032366011291742325, "eval_beta_dpo/beta_margin_grad_mean": -0.4553416073322296, "eval_beta_dpo/beta_margin_grad_std": 0.07153664529323578, "eval_beta_dpo/beta_margin_mean": 0.8038776516914368, "eval_beta_dpo/beta_margin_std": 1.0105534791946411, "eval_beta_dpo/beta_used": 0.0032366011291742325, "eval_beta_dpo/beta_used_raw": -7.0407514572143555, "eval_beta_dpo/gap_mean": 284.4014892578125, "eval_beta_dpo/gap_std": 265.404052734375, "eval_beta_dpo/loss_margin_mean": 165.3889923095703, "eval_beta_dpo/mask_keep_frac": 1.0, "eval_logits/chosen": -2.395129442214966, "eval_logits/rejected": -2.3756070137023926, "eval_loss": 0.6708105206489563, "eval_runtime": 37.7646, "eval_samples_per_second": 61.936, "eval_steps_per_second": 1.96, "step": 300 }, { "beta_dpo/beta": 0.4013200104236603, "beta_dpo/beta_margin_grad_mean": -0.24900613725185394, "beta_dpo/beta_margin_grad_std": 0.25779998302459717, "beta_dpo/beta_margin_mean": 136.3580780029297, "beta_dpo/beta_margin_std": 207.45651245117188, "beta_dpo/beta_used": 0.4013200104236603, "beta_dpo/beta_used_raw": -0.9006747007369995, "beta_dpo/gap_mean": 288.820556640625, "beta_dpo/gap_std": 268.8150634765625, "beta_dpo/loss_margin_mean": 303.63372802734375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4419970631424376, "grad_norm": 4913.37890625, "learning_rate": 3.4388957558875316e-07, "logits/chosen": -2.354623794555664, "logits/rejected": -2.343824863433838, "loss": 2.6589, "step": 301 }, { "beta_dpo/beta": 1.2232584953308105, "beta_dpo/beta_margin_grad_mean": -0.24484674632549286, "beta_dpo/beta_margin_grad_std": 0.2557476758956909, "beta_dpo/beta_margin_mean": 401.79315185546875, "beta_dpo/beta_margin_std": 578.2942504882812, "beta_dpo/beta_used": 1.2232584953308105, "beta_dpo/beta_used_raw": 1.0286716222763062, "beta_dpo/gap_mean": 292.2723388671875, "beta_dpo/gap_std": 271.0513916015625, "beta_dpo/loss_margin_mean": 315.1329345703125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4434654919236417, "grad_norm": 8826.119140625, "learning_rate": 3.426989547989902e-07, "logits/chosen": -2.3831279277801514, "logits/rejected": -2.3882064819335938, "loss": 1.4867, "step": 302 }, { "beta_dpo/beta": 1.5807337760925293, "beta_dpo/beta_margin_grad_mean": -0.15623971819877625, "beta_dpo/beta_margin_grad_std": 0.3630678653717041, "beta_dpo/beta_margin_mean": 455.2434387207031, "beta_dpo/beta_margin_std": 776.2095336914062, "beta_dpo/beta_used": 1.5807337760925293, "beta_dpo/beta_used_raw": 1.5807337760925293, "beta_dpo/gap_mean": 291.7049560546875, "beta_dpo/gap_std": 282.81158447265625, "beta_dpo/loss_margin_mean": 283.4997253417969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.44493392070484583, "grad_norm": 7618.3427734375, "learning_rate": 3.4150589130555773e-07, "logits/chosen": -2.403137445449829, "logits/rejected": -2.3835113048553467, "loss": 9.1621, "step": 303 }, { "beta_dpo/beta": 1.647692322731018, "beta_dpo/beta_margin_grad_mean": -0.2932804524898529, "beta_dpo/beta_margin_grad_std": 0.292863667011261, "beta_dpo/beta_margin_mean": 490.46783447265625, "beta_dpo/beta_margin_std": 817.7069091796875, "beta_dpo/beta_used": 1.647692322731018, "beta_dpo/beta_used_raw": 1.154199481010437, "beta_dpo/gap_mean": 291.1902160644531, "beta_dpo/gap_std": 278.3885498046875, "beta_dpo/loss_margin_mean": 290.32611083984375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.44640234948604995, "grad_norm": 24986.185546875, "learning_rate": 3.403104165467883e-07, "logits/chosen": -2.42034649848938, "logits/rejected": -2.406019687652588, "loss": 30.0877, "step": 304 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.43377500772476196, "beta_dpo/beta_margin_grad_std": 0.06665903329849243, "beta_dpo/beta_margin_mean": 0.2723696529865265, "beta_dpo/beta_margin_std": 0.27916526794433594, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.7424488067626953, "beta_dpo/gap_mean": 286.82373046875, "beta_dpo/gap_std": 275.3907775878906, "beta_dpo/loss_margin_mean": 272.36962890625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.447870778267254, "grad_norm": 19.328815460205078, "learning_rate": 3.391125620245535e-07, "logits/chosen": -2.408841848373413, "logits/rejected": -2.3937630653381348, "loss": 1.1722, "step": 305 }, { "beta_dpo/beta": 0.8187703490257263, "beta_dpo/beta_margin_grad_mean": -0.2615983486175537, "beta_dpo/beta_margin_grad_std": 0.2689959704875946, "beta_dpo/beta_margin_mean": 221.95233154296875, "beta_dpo/beta_margin_std": 310.6951599121094, "beta_dpo/beta_used": 0.8187703490257263, "beta_dpo/beta_used_raw": 0.23448729515075684, "beta_dpo/gap_mean": 286.04840087890625, "beta_dpo/gap_std": 266.01214599609375, "beta_dpo/loss_margin_mean": 279.9851989746094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.44933920704845814, "grad_norm": 12.146223068237305, "learning_rate": 3.3791235930343417e-07, "logits/chosen": -2.376699686050415, "logits/rejected": -2.353483200073242, "loss": 0.5756, "step": 306 }, { "beta_dpo/beta": 0.02242352068424225, "beta_dpo/beta_margin_grad_mean": -0.322419136762619, "beta_dpo/beta_margin_grad_std": 0.2971925735473633, "beta_dpo/beta_margin_mean": 5.65758752822876, "beta_dpo/beta_margin_std": 10.40542984008789, "beta_dpo/beta_used": 0.02242352068424225, "beta_dpo/beta_used_raw": -1.8627017736434937, "beta_dpo/gap_mean": 283.6302185058594, "beta_dpo/gap_std": 267.28631591796875, "beta_dpo/loss_margin_mean": 264.5188293457031, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.45080763582966227, "grad_norm": 351.361083984375, "learning_rate": 3.367098400098881e-07, "logits/chosen": -2.3703196048736572, "logits/rejected": -2.361326217651367, "loss": 1.543, "step": 307 }, { "beta_dpo/beta": 0.610724925994873, "beta_dpo/beta_margin_grad_mean": -0.3044406473636627, "beta_dpo/beta_margin_grad_std": 0.2935749292373657, "beta_dpo/beta_margin_mean": 177.69180297851562, "beta_dpo/beta_margin_std": 322.2966613769531, "beta_dpo/beta_used": 0.610724925994873, "beta_dpo/beta_used_raw": 0.18854334950447083, "beta_dpo/gap_mean": 280.91876220703125, "beta_dpo/gap_std": 270.7321472167969, "beta_dpo/loss_margin_mean": 267.55584716796875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4522760646108664, "grad_norm": 5888.60400390625, "learning_rate": 3.355050358314172e-07, "logits/chosen": -2.3571863174438477, "logits/rejected": -2.3389384746551514, "loss": 11.2154, "step": 308 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.43760332465171814, "beta_dpo/beta_margin_grad_std": 0.056733082979917526, "beta_dpo/beta_margin_mean": 0.254102885723114, "beta_dpo/beta_margin_std": 0.23493850231170654, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.7044029235839844, "beta_dpo/gap_mean": 273.9566650390625, "beta_dpo/gap_std": 264.2417297363281, "beta_dpo/loss_margin_mean": 254.1028594970703, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.45374449339207046, "grad_norm": 19.618959426879883, "learning_rate": 3.3429797851573183e-07, "logits/chosen": -2.3406341075897217, "logits/rejected": -2.3255674839019775, "loss": 1.1658, "step": 309 }, { "beta_dpo/beta": 0.6274186968803406, "beta_dpo/beta_margin_grad_mean": -0.3148829936981201, "beta_dpo/beta_margin_grad_std": 0.30497488379478455, "beta_dpo/beta_margin_mean": 173.7240447998047, "beta_dpo/beta_margin_std": 294.6714172363281, "beta_dpo/beta_used": 0.6274186968803406, "beta_dpo/beta_used_raw": -0.6384754180908203, "beta_dpo/gap_mean": 270.5137939453125, "beta_dpo/gap_std": 256.47113037109375, "beta_dpo/loss_margin_mean": 253.77883911132812, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4552129221732746, "grad_norm": 15305.533203125, "learning_rate": 3.3308869986991487e-07, "logits/chosen": -2.35772442817688, "logits/rejected": -2.3493471145629883, "loss": 22.3664, "step": 310 }, { "beta_dpo/beta": 1.5589945316314697, "beta_dpo/beta_margin_grad_mean": -0.10382533818483353, "beta_dpo/beta_margin_grad_std": 0.2990655303001404, "beta_dpo/beta_margin_mean": 463.5744323730469, "beta_dpo/beta_margin_std": 523.1697387695312, "beta_dpo/beta_used": 1.5589945316314697, "beta_dpo/beta_used_raw": 1.5589945316314697, "beta_dpo/gap_mean": 278.9131164550781, "beta_dpo/gap_std": 256.4837341308594, "beta_dpo/loss_margin_mean": 317.3361511230469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4566813509544787, "grad_norm": 17659.806640625, "learning_rate": 3.3187723175958346e-07, "logits/chosen": -2.3346574306488037, "logits/rejected": -2.313831090927124, "loss": 10.112, "step": 311 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4359610378742218, "beta_dpo/beta_margin_grad_std": 0.06931335479021072, "beta_dpo/beta_margin_mean": 0.26454854011535645, "beta_dpo/beta_margin_std": 0.29412809014320374, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.6447594165802002, "beta_dpo/gap_mean": 275.85595703125, "beta_dpo/gap_std": 261.27203369140625, "beta_dpo/loss_margin_mean": 264.5485534667969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4581497797356828, "grad_norm": 19.33686065673828, "learning_rate": 3.306636061080487e-07, "logits/chosen": -2.301133155822754, "logits/rejected": -2.2737810611724854, "loss": 1.1643, "step": 312 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4278359115123749, "beta_dpo/beta_margin_grad_std": 0.06270962953567505, "beta_dpo/beta_margin_mean": 0.29759153723716736, "beta_dpo/beta_margin_std": 0.26974210143089294, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.4672694206237793, "beta_dpo/gap_mean": 277.2411804199219, "beta_dpo/gap_std": 261.7774963378906, "beta_dpo/loss_margin_mean": 297.5915222167969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.45961820851688695, "grad_norm": 18.111801147460938, "learning_rate": 3.2944785489547537e-07, "logits/chosen": -2.3542749881744385, "logits/rejected": -2.3362579345703125, "loss": 1.1587, "step": 313 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.42816469073295593, "beta_dpo/beta_margin_grad_std": 0.06579755246639252, "beta_dpo/beta_margin_mean": 0.29621294140815735, "beta_dpo/beta_margin_std": 0.27990517020225525, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.222659111022949, "beta_dpo/gap_mean": 284.3181457519531, "beta_dpo/gap_std": 268.36395263671875, "beta_dpo/loss_margin_mean": 296.2129211425781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.461086637298091, "grad_norm": 20.57044792175293, "learning_rate": 3.2823001015803857e-07, "logits/chosen": -2.3612356185913086, "logits/rejected": -2.352372169494629, "loss": 1.1678, "step": 314 }, { "beta_dpo/beta": 0.1255430281162262, "beta_dpo/beta_margin_grad_mean": -0.2888107895851135, "beta_dpo/beta_margin_grad_std": 0.27396243810653687, "beta_dpo/beta_margin_mean": 33.68264389038086, "beta_dpo/beta_margin_std": 54.088321685791016, "beta_dpo/beta_used": 0.1255430281162262, "beta_dpo/beta_used_raw": -1.2729917764663696, "beta_dpo/gap_mean": 279.51483154296875, "beta_dpo/gap_std": 260.94415283203125, "beta_dpo/loss_margin_mean": 245.77093505859375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.46255506607929514, "grad_norm": 1714.6759033203125, "learning_rate": 3.270101039870797e-07, "logits/chosen": -2.284019947052002, "logits/rejected": -2.2785418033599854, "loss": 2.0985, "step": 315 }, { "beta_dpo/beta": 0.617267370223999, "beta_dpo/beta_margin_grad_mean": -0.23152686655521393, "beta_dpo/beta_margin_grad_std": 0.24057447910308838, "beta_dpo/beta_margin_mean": 237.53465270996094, "beta_dpo/beta_margin_std": 373.554443359375, "beta_dpo/beta_used": 0.617267370223999, "beta_dpo/beta_used_raw": -0.2918035387992859, "beta_dpo/gap_mean": 287.0285949707031, "beta_dpo/gap_std": 267.6070556640625, "beta_dpo/loss_margin_mean": 332.6310729980469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.46402349486049926, "grad_norm": 13.612666130065918, "learning_rate": 3.2578816852826086e-07, "logits/chosen": -2.326328754425049, "logits/rejected": -2.3280577659606934, "loss": 0.5785, "step": 316 }, { "beta_dpo/beta": 0.3495140075683594, "beta_dpo/beta_margin_grad_mean": -0.2776326537132263, "beta_dpo/beta_margin_grad_std": 0.2784407138824463, "beta_dpo/beta_margin_mean": 142.2916717529297, "beta_dpo/beta_margin_std": 227.7239990234375, "beta_dpo/beta_used": 0.3495140075683594, "beta_dpo/beta_used_raw": -0.26205989718437195, "beta_dpo/gap_mean": 290.8403625488281, "beta_dpo/gap_std": 272.17449951171875, "beta_dpo/loss_margin_mean": 337.1584167480469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4654919236417034, "grad_norm": 4855.1845703125, "learning_rate": 3.2456423598071783e-07, "logits/chosen": -2.3395438194274902, "logits/rejected": -2.323230743408203, "loss": 5.2066, "step": 317 }, { "beta_dpo/beta": 0.2519617974758148, "beta_dpo/beta_margin_grad_mean": -0.30234721302986145, "beta_dpo/beta_margin_grad_std": 0.29342392086982727, "beta_dpo/beta_margin_mean": 63.16868591308594, "beta_dpo/beta_margin_std": 115.97257995605469, "beta_dpo/beta_used": 0.2519617974758148, "beta_dpo/beta_used_raw": -0.1758996844291687, "beta_dpo/gap_mean": 292.19378662109375, "beta_dpo/gap_std": 275.5003662109375, "beta_dpo/loss_margin_mean": 260.8631591796875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4669603524229075, "grad_norm": 1530.6976318359375, "learning_rate": 3.233383385962115e-07, "logits/chosen": -2.356491804122925, "logits/rejected": -2.3217403888702393, "loss": 1.5612, "step": 318 }, { "beta_dpo/beta": 1.8876981735229492, "beta_dpo/beta_margin_grad_mean": -0.0937500074505806, "beta_dpo/beta_margin_grad_std": 0.2914806008338928, "beta_dpo/beta_margin_mean": 687.0026245117188, "beta_dpo/beta_margin_std": 757.8516845703125, "beta_dpo/beta_used": 1.8876981735229492, "beta_dpo/beta_used_raw": 1.8876981735229492, "beta_dpo/gap_mean": 300.7230529785156, "beta_dpo/gap_std": 288.64886474609375, "beta_dpo/loss_margin_mean": 364.14324951171875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4684287812041116, "grad_norm": 9876.3720703125, "learning_rate": 3.2211050867827805e-07, "logits/chosen": -2.335461139678955, "logits/rejected": -2.3492789268493652, "loss": 18.2557, "step": 319 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.420613557100296, "beta_dpo/beta_margin_grad_std": 0.06336997449398041, "beta_dpo/beta_margin_mean": 0.3268977701663971, "beta_dpo/beta_margin_std": 0.26918449997901917, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.7549973130226135, "beta_dpo/gap_mean": 309.19366455078125, "beta_dpo/gap_std": 285.6737060546875, "beta_dpo/loss_margin_mean": 326.89776611328125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4698972099853157, "grad_norm": 18.470361709594727, "learning_rate": 3.208807785813777e-07, "logits/chosen": -2.343623638153076, "logits/rejected": -2.3438785076141357, "loss": 1.1248, "step": 320 }, { "beta_dpo/beta": 1.1016801595687866, "beta_dpo/beta_margin_grad_mean": -0.23686403036117554, "beta_dpo/beta_margin_grad_std": 0.24444517493247986, "beta_dpo/beta_margin_mean": 430.26971435546875, "beta_dpo/beta_margin_std": 625.2273559570312, "beta_dpo/beta_used": 1.1016801595687866, "beta_dpo/beta_used_raw": -0.355446457862854, "beta_dpo/gap_mean": 312.37139892578125, "beta_dpo/gap_std": 286.314453125, "beta_dpo/loss_margin_mean": 314.2262268066406, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4713656387665198, "grad_norm": 17.7462158203125, "learning_rate": 3.1964918071004217e-07, "logits/chosen": -2.3021278381347656, "logits/rejected": -2.2860074043273926, "loss": 0.5814, "step": 321 }, { "beta_dpo/beta": 1.7430135011672974, "beta_dpo/beta_margin_grad_mean": -0.2846636474132538, "beta_dpo/beta_margin_grad_std": 0.28654375672340393, "beta_dpo/beta_margin_mean": 607.9188232421875, "beta_dpo/beta_margin_std": 957.2604370117188, "beta_dpo/beta_used": 1.7430135011672974, "beta_dpo/beta_used_raw": 0.8375918865203857, "beta_dpo/gap_mean": 308.1527404785156, "beta_dpo/gap_std": 287.0491943359375, "beta_dpo/loss_margin_mean": 287.3334045410156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.47283406754772395, "grad_norm": 24231.912109375, "learning_rate": 3.184157475180207e-07, "logits/chosen": -2.332392454147339, "logits/rejected": -2.3149595260620117, "loss": 23.491, "step": 322 }, { "beta_dpo/beta": 0.23395885527133942, "beta_dpo/beta_margin_grad_mean": -0.3341367244720459, "beta_dpo/beta_margin_grad_std": 0.3162374496459961, "beta_dpo/beta_margin_mean": 62.195438385009766, "beta_dpo/beta_margin_std": 114.66217041015625, "beta_dpo/beta_used": 0.23395885527133942, "beta_dpo/beta_used_raw": -3.1040267944335938, "beta_dpo/gap_mean": 295.2753601074219, "beta_dpo/gap_std": 282.6934814453125, "beta_dpo/loss_margin_mean": 235.39443969726562, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.47430249632892807, "grad_norm": 3386.258544921875, "learning_rate": 3.171805115074251e-07, "logits/chosen": -2.3514816761016846, "logits/rejected": -2.3433778285980225, "loss": 4.9578, "step": 323 }, { "beta_dpo/beta": 0.3184950053691864, "beta_dpo/beta_margin_grad_mean": -0.3133268654346466, "beta_dpo/beta_margin_grad_std": 0.3128097951412201, "beta_dpo/beta_margin_mean": 79.24275207519531, "beta_dpo/beta_margin_std": 142.1294403076172, "beta_dpo/beta_used": 0.3184950053691864, "beta_dpo/beta_used_raw": 0.007871389389038086, "beta_dpo/gap_mean": 291.29779052734375, "beta_dpo/gap_std": 282.01904296875, "beta_dpo/loss_margin_mean": 317.21417236328125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.47577092511013214, "grad_norm": 4426.65576171875, "learning_rate": 3.1594350522787295e-07, "logits/chosen": -2.34005069732666, "logits/rejected": -2.3183207511901855, "loss": 2.4952, "step": 324 }, { "beta_dpo/beta": 0.4672297239303589, "beta_dpo/beta_margin_grad_mean": -0.2523164451122284, "beta_dpo/beta_margin_grad_std": 0.2602192461490631, "beta_dpo/beta_margin_mean": 114.00592803955078, "beta_dpo/beta_margin_std": 195.69052124023438, "beta_dpo/beta_used": 0.4672297239303589, "beta_dpo/beta_used_raw": -2.8321478366851807, "beta_dpo/gap_mean": 288.2220458984375, "beta_dpo/gap_std": 285.31488037109375, "beta_dpo/loss_margin_mean": 241.5183563232422, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.47723935389133626, "grad_norm": 3548.100830078125, "learning_rate": 3.147047612756302e-07, "logits/chosen": -2.3933932781219482, "logits/rejected": -2.3574209213256836, "loss": 4.5265, "step": 325 }, { "beta_dpo/beta": 0.42803052067756653, "beta_dpo/beta_margin_grad_mean": -0.24226514995098114, "beta_dpo/beta_margin_grad_std": 0.25276702642440796, "beta_dpo/beta_margin_mean": 128.09483337402344, "beta_dpo/beta_margin_std": 189.2570343017578, "beta_dpo/beta_used": 0.42803052067756653, "beta_dpo/beta_used_raw": 0.20111998915672302, "beta_dpo/gap_mean": 288.9057312011719, "beta_dpo/gap_std": 278.6829833984375, "beta_dpo/loss_margin_mean": 309.9962158203125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4787077826725404, "grad_norm": 12.956615447998047, "learning_rate": 3.134643122927519e-07, "logits/chosen": -2.4037997722625732, "logits/rejected": -2.368350028991699, "loss": 0.5699, "step": 326 }, { "beta_dpo/beta": 2.713379383087158, "beta_dpo/beta_margin_grad_mean": -0.09374991059303284, "beta_dpo/beta_margin_grad_std": 0.29148030281066895, "beta_dpo/beta_margin_mean": 989.2396850585938, "beta_dpo/beta_margin_std": 727.1015625, "beta_dpo/beta_used": 2.713379383087158, "beta_dpo/beta_used_raw": 2.713379383087158, "beta_dpo/gap_mean": 302.56488037109375, "beta_dpo/gap_std": 276.96759033203125, "beta_dpo/loss_margin_mean": 361.91253662109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4801762114537445, "grad_norm": 30234.04296875, "learning_rate": 3.1222219096622264e-07, "logits/chosen": -2.3508028984069824, "logits/rejected": -2.3310558795928955, "loss": 8.5624, "step": 327 }, { "beta_dpo/beta": 2.522514820098877, "beta_dpo/beta_margin_grad_mean": -0.08467871695756912, "beta_dpo/beta_margin_grad_std": 0.2693813741207123, "beta_dpo/beta_margin_mean": 884.0071411132812, "beta_dpo/beta_margin_std": 907.109130859375, "beta_dpo/beta_used": 2.522514820098877, "beta_dpo/beta_used_raw": 2.522514820098877, "beta_dpo/gap_mean": 310.073486328125, "beta_dpo/gap_std": 278.1096496582031, "beta_dpo/loss_margin_mean": 338.466796875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.48164464023494863, "grad_norm": 10285.1142578125, "learning_rate": 3.1097843002709427e-07, "logits/chosen": -2.3761606216430664, "logits/rejected": -2.3547372817993164, "loss": 10.5089, "step": 328 }, { "beta_dpo/beta": 2.3934481143951416, "beta_dpo/beta_margin_grad_mean": -0.2540913224220276, "beta_dpo/beta_margin_grad_std": 0.2650849223136902, "beta_dpo/beta_margin_mean": 809.6107788085938, "beta_dpo/beta_margin_std": 1348.3310546875, "beta_dpo/beta_used": 2.3934481143951416, "beta_dpo/beta_used_raw": 1.0050034523010254, "beta_dpo/gap_mean": 315.6600646972656, "beta_dpo/gap_std": 280.899658203125, "beta_dpo/loss_margin_mean": 346.1232604980469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4831130690161527, "grad_norm": 12.77830696105957, "learning_rate": 3.0973306224962437e-07, "logits/chosen": -2.3475561141967773, "logits/rejected": -2.316133975982666, "loss": 0.5727, "step": 329 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41881564259529114, "beta_dpo/beta_margin_grad_std": 0.0694441944360733, "beta_dpo/beta_margin_mean": 0.3351494073867798, "beta_dpo/beta_margin_std": 0.2930092513561249, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.9319770336151123, "beta_dpo/gap_mean": 319.4997863769531, "beta_dpo/gap_std": 284.83905029296875, "beta_dpo/loss_margin_mean": 335.1493835449219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4845814977973568, "grad_norm": 18.973018646240234, "learning_rate": 3.084861204504122e-07, "logits/chosen": -2.369523048400879, "logits/rejected": -2.3711061477661133, "loss": 1.1196, "step": 330 }, { "beta_dpo/beta": 0.2725904583930969, "beta_dpo/beta_margin_grad_mean": -0.23779310286045074, "beta_dpo/beta_margin_grad_std": 0.2499072253704071, "beta_dpo/beta_margin_mean": 81.00366973876953, "beta_dpo/beta_margin_std": 118.57772064208984, "beta_dpo/beta_used": 0.2725904583930969, "beta_dpo/beta_used_raw": -0.001759946346282959, "beta_dpo/gap_mean": 322.59796142578125, "beta_dpo/gap_std": 284.0390319824219, "beta_dpo/loss_margin_mean": 326.7667541503906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.48604992657856094, "grad_norm": 13.570107460021973, "learning_rate": 3.072376374875335e-07, "logits/chosen": -2.355848789215088, "logits/rejected": -2.340954303741455, "loss": 0.553, "step": 331 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4359738230705261, "beta_dpo/beta_margin_grad_std": 0.06110256910324097, "beta_dpo/beta_margin_mean": 0.2621835172176361, "beta_dpo/beta_margin_std": 0.25429460406303406, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.248039245605469, "beta_dpo/gap_mean": 312.61383056640625, "beta_dpo/gap_std": 277.851318359375, "beta_dpo/loss_margin_mean": 262.1835021972656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.48751835535976507, "grad_norm": 19.325729370117188, "learning_rate": 3.059876462596758e-07, "logits/chosen": -2.3538053035736084, "logits/rejected": -2.339722156524658, "loss": 1.1728, "step": 332 }, { "beta_dpo/beta": 2.793745279312134, "beta_dpo/beta_margin_grad_mean": -0.14066281914710999, "beta_dpo/beta_margin_grad_std": 0.3476191461086273, "beta_dpo/beta_margin_mean": 984.294677734375, "beta_dpo/beta_margin_std": 1002.9381713867188, "beta_dpo/beta_used": 2.793745279312134, "beta_dpo/beta_used_raw": 2.793745279312134, "beta_dpo/gap_mean": 315.35302734375, "beta_dpo/gap_std": 276.364990234375, "beta_dpo/loss_margin_mean": 338.3910217285156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4889867841409692, "grad_norm": 13925.4560546875, "learning_rate": 3.0473617970527015e-07, "logits/chosen": -2.367403268814087, "logits/rejected": -2.354949474334717, "loss": 17.0446, "step": 333 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4284491240978241, "beta_dpo/beta_margin_grad_std": 0.0682436004281044, "beta_dpo/beta_margin_mean": 0.29454606771469116, "beta_dpo/beta_margin_std": 0.28985467553138733, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.959665298461914, "beta_dpo/gap_mean": 313.328369140625, "beta_dpo/gap_std": 281.157470703125, "beta_dpo/loss_margin_mean": 294.5460510253906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.49045521292217326, "grad_norm": 18.42644500732422, "learning_rate": 3.034832708016243e-07, "logits/chosen": -2.3879494667053223, "logits/rejected": -2.3688955307006836, "loss": 1.1501, "step": 334 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4463410973548889, "beta_dpo/beta_margin_grad_std": 0.062163643538951874, "beta_dpo/beta_margin_mean": 0.22004252672195435, "beta_dpo/beta_margin_std": 0.2587723433971405, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.695589065551758, "beta_dpo/gap_mean": 296.7751159667969, "beta_dpo/gap_std": 274.739501953125, "beta_dpo/loss_margin_mean": 220.04251098632812, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4919236417033774, "grad_norm": 19.224916458129883, "learning_rate": 3.022289525640531e-07, "logits/chosen": -2.3774776458740234, "logits/rejected": -2.358548164367676, "loss": 1.1783, "step": 335 }, { "beta_dpo/beta": 1.3821998834609985, "beta_dpo/beta_margin_grad_mean": -0.09547650814056396, "beta_dpo/beta_margin_grad_std": 0.28482678532600403, "beta_dpo/beta_margin_mean": 575.6812133789062, "beta_dpo/beta_margin_std": 781.006591796875, "beta_dpo/beta_used": 1.3821998834609985, "beta_dpo/beta_used_raw": 1.3821998834609985, "beta_dpo/gap_mean": 304.68438720703125, "beta_dpo/gap_std": 286.78021240234375, "beta_dpo/loss_margin_mean": 382.9951171875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4933920704845815, "grad_norm": 26202.302734375, "learning_rate": 3.009732580450086e-07, "logits/chosen": -2.341956853866577, "logits/rejected": -2.3473258018493652, "loss": 10.1207, "step": 336 }, { "beta_dpo/beta": 0.8355345726013184, "beta_dpo/beta_margin_grad_mean": -0.301108717918396, "beta_dpo/beta_margin_grad_std": 0.3031361401081085, "beta_dpo/beta_margin_mean": 238.27578735351562, "beta_dpo/beta_margin_std": 417.9895324707031, "beta_dpo/beta_used": 0.8355345726013184, "beta_dpo/beta_used_raw": 0.20989900827407837, "beta_dpo/gap_mean": 314.0046081542969, "beta_dpo/gap_std": 295.2994079589844, "beta_dpo/loss_margin_mean": 319.1441650390625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4948604992657856, "grad_norm": 15146.6640625, "learning_rate": 2.9971622033320914e-07, "logits/chosen": -2.3654699325561523, "logits/rejected": -2.3615758419036865, "loss": 13.694, "step": 337 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.42984750866889954, "beta_dpo/beta_margin_grad_std": 0.06152024492621422, "beta_dpo/beta_margin_mean": 0.28707996010780334, "beta_dpo/beta_margin_std": 0.2556547224521637, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.1681759357452393, "beta_dpo/gap_mean": 309.56866455078125, "beta_dpo/gap_std": 290.0804138183594, "beta_dpo/loss_margin_mean": 287.0799560546875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.49632892804698975, "grad_norm": 18.67024803161621, "learning_rate": 2.984578725527675e-07, "logits/chosen": -2.3676528930664062, "logits/rejected": -2.366528034210205, "loss": 1.1585, "step": 338 }, { "beta_dpo/beta": 1.054119348526001, "beta_dpo/beta_margin_grad_mean": -0.2663116455078125, "beta_dpo/beta_margin_grad_std": 0.2718449532985687, "beta_dpo/beta_margin_mean": 411.3683776855469, "beta_dpo/beta_margin_std": 620.0492553710938, "beta_dpo/beta_used": 1.054119348526001, "beta_dpo/beta_used_raw": -0.3248262405395508, "beta_dpo/gap_mean": 306.15435791015625, "beta_dpo/gap_std": 285.44110107421875, "beta_dpo/loss_margin_mean": 320.0867614746094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.4977973568281938, "grad_norm": 24645.341796875, "learning_rate": 2.9719824786231796e-07, "logits/chosen": -2.3688817024230957, "logits/rejected": -2.361849784851074, "loss": 6.4518, "step": 339 }, { "beta_dpo/beta": 2.684582233428955, "beta_dpo/beta_margin_grad_mean": -0.26743197441101074, "beta_dpo/beta_margin_grad_std": 0.27371782064437866, "beta_dpo/beta_margin_mean": 901.1436767578125, "beta_dpo/beta_margin_std": 1441.9869384765625, "beta_dpo/beta_used": 2.684582233428955, "beta_dpo/beta_used_raw": 2.532560110092163, "beta_dpo/gap_mean": 309.4892578125, "beta_dpo/gap_std": 288.36248779296875, "beta_dpo/loss_margin_mean": 288.3397521972656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.49926578560939794, "grad_norm": 13.370800018310547, "learning_rate": 2.959373794541426e-07, "logits/chosen": -2.3758139610290527, "logits/rejected": -2.346534490585327, "loss": 0.5592, "step": 340 }, { "beta_dpo/beta": 2.065169334411621, "beta_dpo/beta_margin_grad_mean": -0.23790918290615082, "beta_dpo/beta_margin_grad_std": 0.25097015500068665, "beta_dpo/beta_margin_mean": 729.3997192382812, "beta_dpo/beta_margin_std": 1149.5667724609375, "beta_dpo/beta_used": 2.065169334411621, "beta_dpo/beta_used_raw": 1.1916053295135498, "beta_dpo/gap_mean": 313.42578125, "beta_dpo/gap_std": 293.43829345703125, "beta_dpo/loss_margin_mean": 356.7818603515625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5007342143906021, "grad_norm": 13.331892013549805, "learning_rate": 2.946753005532965e-07, "logits/chosen": -2.3476555347442627, "logits/rejected": -2.346323013305664, "loss": 0.5701, "step": 341 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4230160415172577, "beta_dpo/beta_margin_grad_std": 0.07352254539728165, "beta_dpo/beta_margin_mean": 0.3177441358566284, "beta_dpo/beta_margin_std": 0.3101942837238312, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.5433149337768555, "beta_dpo/gap_mean": 312.7081298828125, "beta_dpo/gap_std": 296.92974853515625, "beta_dpo/loss_margin_mean": 317.7441101074219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5022026431718062, "grad_norm": 20.234586715698242, "learning_rate": 2.934120444167326e-07, "logits/chosen": -2.400458335876465, "logits/rejected": -2.365809917449951, "loss": 1.1335, "step": 342 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4237779676914215, "beta_dpo/beta_margin_grad_std": 0.05391445755958557, "beta_dpo/beta_margin_mean": 0.3117174208164215, "beta_dpo/beta_margin_std": 0.22616280615329742, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.4838824272155762, "beta_dpo/gap_mean": 314.9137268066406, "beta_dpo/gap_std": 284.4820556640625, "beta_dpo/loss_margin_mean": 311.7174072265625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5036710719530103, "grad_norm": 21.022735595703125, "learning_rate": 2.9214764433242476e-07, "logits/chosen": -2.385115623474121, "logits/rejected": -2.39156436920166, "loss": 1.1268, "step": 343 }, { "beta_dpo/beta": 1.2279607057571411, "beta_dpo/beta_margin_grad_mean": -0.25877782702445984, "beta_dpo/beta_margin_grad_std": 0.2643044590950012, "beta_dpo/beta_margin_mean": 545.8978271484375, "beta_dpo/beta_margin_std": 796.9879760742188, "beta_dpo/beta_used": 1.2279607057571411, "beta_dpo/beta_used_raw": -2.5525197982788086, "beta_dpo/gap_mean": 309.91912841796875, "beta_dpo/gap_std": 284.9899597167969, "beta_dpo/loss_margin_mean": 314.7950439453125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5051395007342144, "grad_norm": 15404.0849609375, "learning_rate": 2.9088213361849126e-07, "logits/chosen": -2.3805313110351562, "logits/rejected": -2.383361339569092, "loss": 3.8253, "step": 344 }, { "beta_dpo/beta": 3.3728437423706055, "beta_dpo/beta_margin_grad_mean": -0.09375, "beta_dpo/beta_margin_grad_std": 0.2914806008338928, "beta_dpo/beta_margin_mean": 1225.3790283203125, "beta_dpo/beta_margin_std": 1016.9912719726562, "beta_dpo/beta_used": 3.3728437423706055, "beta_dpo/beta_used_raw": 3.3728437423706055, "beta_dpo/gap_mean": 323.52496337890625, "beta_dpo/gap_std": 289.0306701660156, "beta_dpo/loss_margin_mean": 365.1717529296875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5066079295154186, "grad_norm": 20076.240234375, "learning_rate": 2.896155456223163e-07, "logits/chosen": -2.40063214302063, "logits/rejected": -2.3850648403167725, "loss": 10.5342, "step": 345 }, { "beta_dpo/beta": 0.6880494952201843, "beta_dpo/beta_margin_grad_mean": -0.25628435611724854, "beta_dpo/beta_margin_grad_std": 0.2660585939884186, "beta_dpo/beta_margin_mean": 223.1298065185547, "beta_dpo/beta_margin_std": 299.98797607421875, "beta_dpo/beta_used": 0.6880494952201843, "beta_dpo/beta_used_raw": 0.48999911546707153, "beta_dpo/gap_mean": 326.8367919921875, "beta_dpo/gap_std": 281.955810546875, "beta_dpo/loss_margin_mean": 328.8027038574219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5080763582966226, "grad_norm": 8900.7021484375, "learning_rate": 2.883479137196714e-07, "logits/chosen": -2.3676557540893555, "logits/rejected": -2.3624587059020996, "loss": 6.0288, "step": 346 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.44006967544555664, "beta_dpo/beta_margin_grad_std": 0.0717720165848732, "beta_dpo/beta_margin_mean": 0.24584244191646576, "beta_dpo/beta_margin_std": 0.2963031232357025, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.904984474182129, "beta_dpo/gap_mean": 316.7066650390625, "beta_dpo/gap_std": 282.1290283203125, "beta_dpo/loss_margin_mean": 245.84243774414062, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5095447870778267, "grad_norm": 21.67890739440918, "learning_rate": 2.8707927131383614e-07, "logits/chosen": -2.3673133850097656, "logits/rejected": -2.3626620769500732, "loss": 1.1499, "step": 347 }, { "beta_dpo/beta": 1.4570677280426025, "beta_dpo/beta_margin_grad_mean": -0.23459888994693756, "beta_dpo/beta_margin_grad_std": 0.24054688215255737, "beta_dpo/beta_margin_mean": 557.3722534179688, "beta_dpo/beta_margin_std": 804.013671875, "beta_dpo/beta_used": 1.4570677280426025, "beta_dpo/beta_used_raw": 1.1090539693832397, "beta_dpo/gap_mean": 314.82440185546875, "beta_dpo/gap_std": 278.79046630859375, "beta_dpo/loss_margin_mean": 317.42047119140625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5110132158590308, "grad_norm": 11.807612419128418, "learning_rate": 2.858096518347179e-07, "logits/chosen": -2.3952393531799316, "logits/rejected": -2.3867907524108887, "loss": 0.5596, "step": 348 }, { "beta_dpo/beta": 0.08341504633426666, "beta_dpo/beta_margin_grad_mean": -0.25735506415367126, "beta_dpo/beta_margin_grad_std": 0.26621419191360474, "beta_dpo/beta_margin_mean": 28.167213439941406, "beta_dpo/beta_margin_std": 44.491722106933594, "beta_dpo/beta_used": 0.08341504633426666, "beta_dpo/beta_used_raw": -0.4417477548122406, "beta_dpo/gap_mean": 314.21697998046875, "beta_dpo/gap_std": 281.6492004394531, "beta_dpo/loss_margin_mean": 326.5581359863281, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5124816446402349, "grad_norm": 1751.5216064453125, "learning_rate": 2.845390887379706e-07, "logits/chosen": -2.3977246284484863, "logits/rejected": -2.3966641426086426, "loss": 2.7493, "step": 349 }, { "beta_dpo/beta": 1.8887107372283936, "beta_dpo/beta_margin_grad_mean": -0.140625, "beta_dpo/beta_margin_grad_std": 0.34763428568840027, "beta_dpo/beta_margin_mean": 593.6013793945312, "beta_dpo/beta_margin_std": 579.1753540039062, "beta_dpo/beta_used": 1.8887107372283936, "beta_dpo/beta_used_raw": 1.8887107372283936, "beta_dpo/gap_mean": 314.50628662109375, "beta_dpo/gap_std": 285.41729736328125, "beta_dpo/loss_margin_mean": 308.5635986328125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5139500734214391, "grad_norm": 25647.646484375, "learning_rate": 2.8326761550411346e-07, "logits/chosen": -2.409909725189209, "logits/rejected": -2.4100773334503174, "loss": 28.0702, "step": 350 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4200161397457123, "beta_dpo/beta_margin_grad_std": 0.07670863717794418, "beta_dpo/beta_margin_mean": 0.33481884002685547, "beta_dpo/beta_margin_std": 0.3401849567890167, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.0178706645965576, "beta_dpo/gap_mean": 318.64599609375, "beta_dpo/gap_std": 295.79644775390625, "beta_dpo/loss_margin_mean": 334.8188171386719, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5154185022026432, "grad_norm": 20.02607536315918, "learning_rate": 2.819952656376487e-07, "logits/chosen": -2.3981099128723145, "logits/rejected": -2.4018301963806152, "loss": 1.1514, "step": 351 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.43255868554115295, "beta_dpo/beta_margin_grad_std": 0.06271059066057205, "beta_dpo/beta_margin_mean": 0.2771955132484436, "beta_dpo/beta_margin_std": 0.26465079188346863, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.255035400390625, "beta_dpo/gap_mean": 310.7604675292969, "beta_dpo/gap_std": 290.19696044921875, "beta_dpo/loss_margin_mean": 277.19549560546875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5168869309838473, "grad_norm": 18.892972946166992, "learning_rate": 2.8072207266617854e-07, "logits/chosen": -2.4018521308898926, "logits/rejected": -2.383004903793335, "loss": 1.1557, "step": 352 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4298509657382965, "beta_dpo/beta_margin_grad_std": 0.07284195721149445, "beta_dpo/beta_margin_mean": 0.2901245057582855, "beta_dpo/beta_margin_std": 0.30724653601646423, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.7248914241790771, "beta_dpo/gap_mean": 307.8128356933594, "beta_dpo/gap_std": 293.14794921875, "beta_dpo/loss_margin_mean": 290.12451171875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5183553597650514, "grad_norm": 20.134214401245117, "learning_rate": 2.794480701395219e-07, "logits/chosen": -2.4077649116516113, "logits/rejected": -2.4009876251220703, "loss": 1.1411, "step": 353 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.42897459864616394, "beta_dpo/beta_margin_grad_std": 0.058552030473947525, "beta_dpo/beta_margin_mean": 0.29052889347076416, "beta_dpo/beta_margin_std": 0.2439156472682953, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.4506525993347168, "beta_dpo/gap_mean": 302.3212585449219, "beta_dpo/gap_std": 285.1046447753906, "beta_dpo/loss_margin_mean": 290.52886962890625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5198237885462555, "grad_norm": 23.041349411010742, "learning_rate": 2.781732916288303e-07, "logits/chosen": -2.4216408729553223, "logits/rejected": -2.4001893997192383, "loss": 1.138, "step": 354 }, { "beta_dpo/beta": 0.6849542856216431, "beta_dpo/beta_margin_grad_mean": -0.10933418571949005, "beta_dpo/beta_margin_grad_std": 0.3119906485080719, "beta_dpo/beta_margin_mean": 228.0188446044922, "beta_dpo/beta_margin_std": 165.11636352539062, "beta_dpo/beta_used": 0.6849542856216431, "beta_dpo/beta_used_raw": 0.6849542856216431, "beta_dpo/gap_mean": 305.43170166015625, "beta_dpo/gap_std": 274.4084777832031, "beta_dpo/loss_margin_mean": 328.5459289550781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5212922173274597, "grad_norm": 6968.03076171875, "learning_rate": 2.7689777072570284e-07, "logits/chosen": -2.445467948913574, "logits/rejected": -2.4192585945129395, "loss": 5.7505, "step": 355 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4357198476791382, "beta_dpo/beta_margin_grad_std": 0.06848623603582382, "beta_dpo/beta_margin_mean": 0.2656390070915222, "beta_dpo/beta_margin_std": 0.2923762798309326, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.9701589345932007, "beta_dpo/gap_mean": 300.7390441894531, "beta_dpo/gap_std": 275.934326171875, "beta_dpo/loss_margin_mean": 265.6390075683594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5227606461086637, "grad_norm": 20.48623275756836, "learning_rate": 2.7562154104130176e-07, "logits/chosen": -2.409104824066162, "logits/rejected": -2.3870463371276855, "loss": 1.1484, "step": 356 }, { "beta_dpo/beta": 1.1052085161209106, "beta_dpo/beta_margin_grad_mean": -0.25593680143356323, "beta_dpo/beta_margin_grad_std": 0.2634008824825287, "beta_dpo/beta_margin_mean": 405.9659729003906, "beta_dpo/beta_margin_std": 574.3975830078125, "beta_dpo/beta_used": 1.1052085161209106, "beta_dpo/beta_used_raw": -1.2659064531326294, "beta_dpo/gap_mean": 301.724609375, "beta_dpo/gap_std": 276.3421325683594, "beta_dpo/loss_margin_mean": 287.9976806640625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5242290748898678, "grad_norm": 9097.716796875, "learning_rate": 2.7434463620546594e-07, "logits/chosen": -2.4178619384765625, "logits/rejected": -2.4092202186584473, "loss": 3.893, "step": 357 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4275015890598297, "beta_dpo/beta_margin_grad_std": 0.0712355524301529, "beta_dpo/beta_margin_mean": 0.29972508549690247, "beta_dpo/beta_margin_std": 0.3031264543533325, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.7463812828063965, "beta_dpo/gap_mean": 296.3010559082031, "beta_dpo/gap_std": 281.1275939941406, "beta_dpo/loss_margin_mean": 299.7250671386719, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5256975036710719, "grad_norm": 19.981340408325195, "learning_rate": 2.730670898658255e-07, "logits/chosen": -2.4385619163513184, "logits/rejected": -2.4157814979553223, "loss": 1.1319, "step": 358 }, { "beta_dpo/beta": 1.1340972185134888, "beta_dpo/beta_margin_grad_mean": -0.3279348611831665, "beta_dpo/beta_margin_grad_std": 0.31545180082321167, "beta_dpo/beta_margin_mean": 409.3197937011719, "beta_dpo/beta_margin_std": 649.3925170898438, "beta_dpo/beta_used": 1.1340972185134888, "beta_dpo/beta_used_raw": 0.9882941842079163, "beta_dpo/gap_mean": 301.54461669921875, "beta_dpo/gap_std": 286.58502197265625, "beta_dpo/loss_margin_mean": 308.6645812988281, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.527165932452276, "grad_norm": 32715.87109375, "learning_rate": 2.717889356869146e-07, "logits/chosen": -2.3811445236206055, "logits/rejected": -2.3567399978637695, "loss": 38.7465, "step": 359 }, { "beta_dpo/beta": 0.06283307075500488, "beta_dpo/beta_margin_grad_mean": -0.2699156701564789, "beta_dpo/beta_margin_grad_std": 0.27412280440330505, "beta_dpo/beta_margin_mean": 18.395671844482422, "beta_dpo/beta_margin_std": 25.289981842041016, "beta_dpo/beta_used": 0.06283307075500488, "beta_dpo/beta_used_raw": -3.388972043991089, "beta_dpo/gap_mean": 292.21331787109375, "beta_dpo/gap_std": 284.0353088378906, "beta_dpo/loss_margin_mean": 261.1766357421875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5286343612334802, "grad_norm": 1064.697265625, "learning_rate": 2.7051020734928443e-07, "logits/chosen": -2.3837945461273193, "logits/rejected": -2.376451015472412, "loss": 1.0703, "step": 360 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4298721253871918, "beta_dpo/beta_margin_grad_std": 0.07070347666740417, "beta_dpo/beta_margin_mean": 0.28980541229248047, "beta_dpo/beta_margin_std": 0.2996097505092621, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.996284544467926, "beta_dpo/gap_mean": 288.74700927734375, "beta_dpo/gap_std": 280.52886962890625, "beta_dpo/loss_margin_mean": 289.8053894042969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5301027900146843, "grad_norm": 25.262163162231445, "learning_rate": 2.6923093854861593e-07, "logits/chosen": -2.4027366638183594, "logits/rejected": -2.4041662216186523, "loss": 1.1471, "step": 361 }, { "beta_dpo/beta": 3.8285510540008545, "beta_dpo/beta_margin_grad_mean": -0.25968730449676514, "beta_dpo/beta_margin_grad_std": 0.2681870758533478, "beta_dpo/beta_margin_mean": 1458.04541015625, "beta_dpo/beta_margin_std": 2284.134521484375, "beta_dpo/beta_used": 3.8285510540008545, "beta_dpo/beta_used_raw": 2.4841794967651367, "beta_dpo/gap_mean": 301.5517883300781, "beta_dpo/gap_std": 286.40460205078125, "beta_dpo/loss_margin_mean": 343.222412109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5315712187958884, "grad_norm": 17.703359603881836, "learning_rate": 2.679511629948319e-07, "logits/chosen": -2.412851333618164, "logits/rejected": -2.408217430114746, "loss": 0.5786, "step": 362 }, { "beta_dpo/beta": 1.8720965385437012, "beta_dpo/beta_margin_grad_mean": -0.22648286819458008, "beta_dpo/beta_margin_grad_std": 0.2343624234199524, "beta_dpo/beta_margin_mean": 804.5293579101562, "beta_dpo/beta_margin_std": 1079.2637939453125, "beta_dpo/beta_used": 1.8720965385437012, "beta_dpo/beta_used_raw": 0.9251274466514587, "beta_dpo/gap_mean": 309.974365234375, "beta_dpo/gap_std": 281.13372802734375, "beta_dpo/loss_margin_mean": 375.29144287109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5330396475770925, "grad_norm": 13.820899963378906, "learning_rate": 2.6667091441120816e-07, "logits/chosen": -2.3960952758789062, "logits/rejected": -2.389695167541504, "loss": 0.5714, "step": 363 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4259724020957947, "beta_dpo/beta_margin_grad_std": 0.06279082596302032, "beta_dpo/beta_margin_mean": 0.304084450006485, "beta_dpo/beta_margin_std": 0.2647766172885895, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.1112940311431885, "beta_dpo/gap_mean": 312.33929443359375, "beta_dpo/gap_std": 278.56072998046875, "beta_dpo/loss_margin_mean": 304.08441162109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5345080763582967, "grad_norm": 23.981868743896484, "learning_rate": 2.6539022653348575e-07, "logits/chosen": -2.423187255859375, "logits/rejected": -2.4290435314178467, "loss": 1.1401, "step": 364 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41367998719215393, "beta_dpo/beta_margin_grad_std": 0.06623264402151108, "beta_dpo/beta_margin_mean": 0.3594275116920471, "beta_dpo/beta_margin_std": 0.30056267976760864, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.2951538562774658, "beta_dpo/gap_mean": 319.5771484375, "beta_dpo/gap_std": 276.10504150390625, "beta_dpo/loss_margin_mean": 359.427490234375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5359765051395007, "grad_norm": 19.053972244262695, "learning_rate": 2.641091331089811e-07, "logits/chosen": -2.439345121383667, "logits/rejected": -2.4456310272216797, "loss": 1.1205, "step": 365 }, { "beta_dpo/beta": 1.3038524389266968, "beta_dpo/beta_margin_grad_mean": -0.26780223846435547, "beta_dpo/beta_margin_grad_std": 0.2728710174560547, "beta_dpo/beta_margin_mean": 548.268310546875, "beta_dpo/beta_margin_std": 849.1015014648438, "beta_dpo/beta_used": 1.3038524389266968, "beta_dpo/beta_used_raw": -1.0728014707565308, "beta_dpo/gap_mean": 328.21820068359375, "beta_dpo/gap_std": 286.6083984375, "beta_dpo/loss_margin_mean": 328.7030334472656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5374449339207048, "grad_norm": 12437.0556640625, "learning_rate": 2.6282766789569736e-07, "logits/chosen": -2.40187406539917, "logits/rejected": -2.413430690765381, "loss": 3.6027, "step": 366 }, { "beta_dpo/beta": 0.2638746500015259, "beta_dpo/beta_margin_grad_mean": -0.2950782775878906, "beta_dpo/beta_margin_grad_std": 0.2939762771129608, "beta_dpo/beta_margin_mean": 86.77686309814453, "beta_dpo/beta_margin_std": 148.9335174560547, "beta_dpo/beta_used": 0.2638746500015259, "beta_dpo/beta_used_raw": -0.3957434594631195, "beta_dpo/gap_mean": 318.469482421875, "beta_dpo/gap_std": 279.47314453125, "beta_dpo/loss_margin_mean": 298.730712890625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5389133627019089, "grad_norm": 4519.6572265625, "learning_rate": 2.615458646614349e-07, "logits/chosen": -2.404602289199829, "logits/rejected": -2.3893771171569824, "loss": 3.52, "step": 367 }, { "beta_dpo/beta": 2.132652759552002, "beta_dpo/beta_margin_grad_mean": -0.2524746060371399, "beta_dpo/beta_margin_grad_std": 0.2634090185165405, "beta_dpo/beta_margin_mean": 816.8623657226562, "beta_dpo/beta_margin_std": 1151.4071044921875, "beta_dpo/beta_used": 2.132652759552002, "beta_dpo/beta_used_raw": 1.5166137218475342, "beta_dpo/gap_mean": 326.68768310546875, "beta_dpo/gap_std": 279.3883056640625, "beta_dpo/loss_margin_mean": 374.5268859863281, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.540381791483113, "grad_norm": 4498.89404296875, "learning_rate": 2.6026375718290083e-07, "logits/chosen": -2.4052700996398926, "logits/rejected": -2.4133400917053223, "loss": 4.8171, "step": 368 }, { "beta_dpo/beta": 0.07468457520008087, "beta_dpo/beta_margin_grad_mean": -0.23666079342365265, "beta_dpo/beta_margin_grad_std": 0.23870179057121277, "beta_dpo/beta_margin_mean": 24.3631534576416, "beta_dpo/beta_margin_std": 35.491668701171875, "beta_dpo/beta_used": 0.07468457520008087, "beta_dpo/beta_used_raw": -0.6055254936218262, "beta_dpo/gap_mean": 326.35491943359375, "beta_dpo/gap_std": 280.87725830078125, "beta_dpo/loss_margin_mean": 311.52703857421875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5418502202643172, "grad_norm": 11.01858139038086, "learning_rate": 2.589813792448196e-07, "logits/chosen": -2.4168224334716797, "logits/rejected": -2.4043703079223633, "loss": 0.5587, "step": 369 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.43343597650527954, "beta_dpo/beta_margin_grad_std": 0.06827400624752045, "beta_dpo/beta_margin_mean": 0.27436593174934387, "beta_dpo/beta_margin_std": 0.2874927818775177, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.793357253074646, "beta_dpo/gap_mean": 316.9214782714844, "beta_dpo/gap_std": 280.57403564453125, "beta_dpo/loss_margin_mean": 274.36590576171875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5433186490455213, "grad_norm": 21.518030166625977, "learning_rate": 2.5769876463904263e-07, "logits/chosen": -2.4265782833099365, "logits/rejected": -2.425027847290039, "loss": 1.1337, "step": 370 }, { "beta_dpo/beta": 0.5957542657852173, "beta_dpo/beta_margin_grad_mean": -0.25642988085746765, "beta_dpo/beta_margin_grad_std": 0.26836514472961426, "beta_dpo/beta_margin_mean": 196.8231964111328, "beta_dpo/beta_margin_std": 344.92559814453125, "beta_dpo/beta_used": 0.5957542657852173, "beta_dpo/beta_used_raw": -2.339658498764038, "beta_dpo/gap_mean": 319.158203125, "beta_dpo/gap_std": 289.86175537109375, "beta_dpo/loss_margin_mean": 333.40985107421875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5447870778267254, "grad_norm": 8407.552734375, "learning_rate": 2.5641594716365744e-07, "logits/chosen": -2.393846035003662, "logits/rejected": -2.3929755687713623, "loss": 21.0427, "step": 371 }, { "beta_dpo/beta": 3.440232038497925, "beta_dpo/beta_margin_grad_mean": -0.09375, "beta_dpo/beta_margin_grad_std": 0.2914806008338928, "beta_dpo/beta_margin_mean": 1281.364501953125, "beta_dpo/beta_margin_std": 1292.0177001953125, "beta_dpo/beta_used": 3.440232038497925, "beta_dpo/beta_used_raw": 3.440232038497925, "beta_dpo/gap_mean": 320.1593017578125, "beta_dpo/gap_std": 292.0980224609375, "beta_dpo/loss_margin_mean": 341.0747985839844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5462555066079295, "grad_norm": 31112.7265625, "learning_rate": 2.551329606220976e-07, "logits/chosen": -2.3891334533691406, "logits/rejected": -2.360813617706299, "loss": 9.5999, "step": 372 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41610556840896606, "beta_dpo/beta_margin_grad_std": 0.05844424292445183, "beta_dpo/beta_margin_mean": 0.34435054659843445, "beta_dpo/beta_margin_std": 0.2457403987646103, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.9711276292800903, "beta_dpo/gap_mean": 326.219970703125, "beta_dpo/gap_std": 283.1405334472656, "beta_dpo/loss_margin_mean": 344.35052490234375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5477239353891337, "grad_norm": 23.21306800842285, "learning_rate": 2.538498388222517e-07, "logits/chosen": -2.3829376697540283, "logits/rejected": -2.3680813312530518, "loss": 1.1126, "step": 373 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.42734894156455994, "beta_dpo/beta_margin_grad_std": 0.06432215124368668, "beta_dpo/beta_margin_mean": 0.2995753586292267, "beta_dpo/beta_margin_std": 0.2753455340862274, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.3080596923828125, "beta_dpo/gap_mean": 326.4972229003906, "beta_dpo/gap_std": 282.38189697265625, "beta_dpo/loss_margin_mean": 299.5753479003906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5491923641703378, "grad_norm": 22.25040054321289, "learning_rate": 2.525666155755725e-07, "logits/chosen": -2.427107810974121, "logits/rejected": -2.4131693840026855, "loss": 1.1579, "step": 374 }, { "beta_dpo/beta": 2.3931167125701904, "beta_dpo/beta_margin_grad_mean": -0.20865072309970856, "beta_dpo/beta_margin_grad_std": 0.216493159532547, "beta_dpo/beta_margin_mean": 1006.3369140625, "beta_dpo/beta_margin_std": 1276.470947265625, "beta_dpo/beta_used": 2.3931167125701904, "beta_dpo/beta_used_raw": 0.9554816484451294, "beta_dpo/gap_mean": 333.14459228515625, "beta_dpo/gap_std": 280.5745849609375, "beta_dpo/loss_margin_mean": 382.5895690917969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5506607929515418, "grad_norm": 13.200187683105469, "learning_rate": 2.512833246961859e-07, "logits/chosen": -2.409682273864746, "logits/rejected": -2.407869815826416, "loss": 0.5667, "step": 375 }, { "beta_dpo/beta": 0.5595067143440247, "beta_dpo/beta_margin_grad_mean": -0.2484658807516098, "beta_dpo/beta_margin_grad_std": 0.26340359449386597, "beta_dpo/beta_margin_mean": 214.9127960205078, "beta_dpo/beta_margin_std": 298.3186950683594, "beta_dpo/beta_used": 0.5595067143440247, "beta_dpo/beta_used_raw": -0.2936529517173767, "beta_dpo/gap_mean": 342.8294677734375, "beta_dpo/gap_std": 288.7122802734375, "beta_dpo/loss_margin_mean": 393.45977783203125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5521292217327459, "grad_norm": 8034.6337890625, "learning_rate": 2.5e-07, "logits/chosen": -2.4047350883483887, "logits/rejected": -2.384918451309204, "loss": 9.2557, "step": 376 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4221385717391968, "beta_dpo/beta_margin_grad_std": 0.0715164914727211, "beta_dpo/beta_margin_mean": 0.32525014877319336, "beta_dpo/beta_margin_std": 0.3200809955596924, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -5.223406791687012, "beta_dpo/gap_mean": 342.27374267578125, "beta_dpo/gap_std": 292.09466552734375, "beta_dpo/loss_margin_mean": 325.2501525878906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.55359765051395, "grad_norm": 21.198095321655273, "learning_rate": 2.487166753038141e-07, "logits/chosen": -2.341383695602417, "logits/rejected": -2.343954086303711, "loss": 1.1565, "step": 377 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4200609624385834, "beta_dpo/beta_margin_grad_std": 0.06614904850721359, "beta_dpo/beta_margin_mean": 0.3278721570968628, "beta_dpo/beta_margin_std": 0.2774437367916107, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.6767090559005737, "beta_dpo/gap_mean": 338.7351379394531, "beta_dpo/gap_std": 294.2265625, "beta_dpo/loss_margin_mean": 327.87213134765625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5550660792951542, "grad_norm": 20.219329833984375, "learning_rate": 2.4743338442442754e-07, "logits/chosen": -2.350855827331543, "logits/rejected": -2.344240665435791, "loss": 1.1113, "step": 378 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41115090250968933, "beta_dpo/beta_margin_grad_std": 0.08131575584411621, "beta_dpo/beta_margin_mean": 0.37218692898750305, "beta_dpo/beta_margin_std": 0.3555106520652771, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.8637235164642334, "beta_dpo/gap_mean": 345.63275146484375, "beta_dpo/gap_std": 302.9630126953125, "beta_dpo/loss_margin_mean": 372.1868896484375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5565345080763583, "grad_norm": 21.625322341918945, "learning_rate": 2.461501611777483e-07, "logits/chosen": -2.3660264015197754, "logits/rejected": -2.391849994659424, "loss": 1.096, "step": 379 }, { "beta_dpo/beta": 0.558686375617981, "beta_dpo/beta_margin_grad_mean": -0.24460071325302124, "beta_dpo/beta_margin_grad_std": 0.26073068380355835, "beta_dpo/beta_margin_mean": 214.0121307373047, "beta_dpo/beta_margin_std": 353.10638427734375, "beta_dpo/beta_used": 0.558686375617981, "beta_dpo/beta_used_raw": 0.4444287121295929, "beta_dpo/gap_mean": 352.9265441894531, "beta_dpo/gap_std": 309.6590270996094, "beta_dpo/loss_margin_mean": 409.6791076660156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5580029368575624, "grad_norm": 13815.2509765625, "learning_rate": 2.4486703937790243e-07, "logits/chosen": -2.35787034034729, "logits/rejected": -2.374066114425659, "loss": 2.8621, "step": 380 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4248752295970917, "beta_dpo/beta_margin_grad_std": 0.07637570053339005, "beta_dpo/beta_margin_mean": 0.3129652142524719, "beta_dpo/beta_margin_std": 0.3291749358177185, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.8646886348724365, "beta_dpo/gap_mean": 352.447998046875, "beta_dpo/gap_std": 314.13433837890625, "beta_dpo/loss_margin_mean": 312.9652099609375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5594713656387665, "grad_norm": 20.60698890686035, "learning_rate": 2.435840528363426e-07, "logits/chosen": -2.350996494293213, "logits/rejected": -2.340618133544922, "loss": 1.1346, "step": 381 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4127730131149292, "beta_dpo/beta_margin_grad_std": 0.0697304978966713, "beta_dpo/beta_margin_mean": 0.3608226776123047, "beta_dpo/beta_margin_std": 0.29635754227638245, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.334915280342102, "beta_dpo/gap_mean": 351.69036865234375, "beta_dpo/gap_std": 313.22735595703125, "beta_dpo/loss_margin_mean": 360.8226623535156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5609397944199707, "grad_norm": 20.6890811920166, "learning_rate": 2.4230123536095745e-07, "logits/chosen": -2.3812966346740723, "logits/rejected": -2.3885483741760254, "loss": 1.0979, "step": 382 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4209439754486084, "beta_dpo/beta_margin_grad_std": 0.07652373611927032, "beta_dpo/beta_margin_mean": 0.32738256454467773, "beta_dpo/beta_margin_std": 0.3242851793766022, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.201818585395813, "beta_dpo/gap_mean": 349.58636474609375, "beta_dpo/gap_std": 314.92181396484375, "beta_dpo/loss_margin_mean": 327.3825378417969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5624082232011748, "grad_norm": 24.684696197509766, "learning_rate": 2.4101862075518037e-07, "logits/chosen": -2.3738174438476562, "logits/rejected": -2.3857202529907227, "loss": 1.0963, "step": 383 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4266614019870758, "beta_dpo/beta_margin_grad_std": 0.0517151914536953, "beta_dpo/beta_margin_mean": 0.2990947663784027, "beta_dpo/beta_margin_std": 0.214844211935997, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.5872749090194702, "beta_dpo/gap_mean": 338.76934814453125, "beta_dpo/gap_std": 299.67193603515625, "beta_dpo/loss_margin_mean": 299.0947570800781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5638766519823789, "grad_norm": 21.041105270385742, "learning_rate": 2.397362428170992e-07, "logits/chosen": -2.406465768814087, "logits/rejected": -2.4182257652282715, "loss": 1.1106, "step": 384 }, { "beta_dpo/beta": 0.5529597401618958, "beta_dpo/beta_margin_grad_mean": -0.23210173845291138, "beta_dpo/beta_margin_grad_std": 0.24000734090805054, "beta_dpo/beta_margin_mean": 192.6968536376953, "beta_dpo/beta_margin_std": 267.4351501464844, "beta_dpo/beta_used": 0.5529597401618958, "beta_dpo/beta_used_raw": 0.1586916744709015, "beta_dpo/gap_mean": 334.78759765625, "beta_dpo/gap_std": 289.0450744628906, "beta_dpo/loss_margin_mean": 311.2054138183594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5653450807635829, "grad_norm": 6151.4521484375, "learning_rate": 2.3845413533856514e-07, "logits/chosen": -2.377863883972168, "logits/rejected": -2.3569583892822266, "loss": 3.7171, "step": 385 }, { "beta_dpo/beta": 0.328173965215683, "beta_dpo/beta_margin_grad_mean": -0.24216587841510773, "beta_dpo/beta_margin_grad_std": 0.25181594491004944, "beta_dpo/beta_margin_mean": 135.31890869140625, "beta_dpo/beta_margin_std": 188.71702575683594, "beta_dpo/beta_used": 0.328173965215683, "beta_dpo/beta_used_raw": -2.265230894088745, "beta_dpo/gap_mean": 334.6175231933594, "beta_dpo/gap_std": 285.37115478515625, "beta_dpo/loss_margin_mean": 366.7230224609375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.566813509544787, "grad_norm": 7890.92236328125, "learning_rate": 2.3717233210430254e-07, "logits/chosen": -2.3772597312927246, "logits/rejected": -2.3809022903442383, "loss": 1.637, "step": 386 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4257817566394806, "beta_dpo/beta_margin_grad_std": 0.07243922352790833, "beta_dpo/beta_margin_mean": 0.3068787157535553, "beta_dpo/beta_margin_std": 0.30547431111335754, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.440854072570801, "beta_dpo/gap_mean": 335.63763427734375, "beta_dpo/gap_std": 289.9078369140625, "beta_dpo/loss_margin_mean": 306.8786926269531, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5682819383259912, "grad_norm": 26.328731536865234, "learning_rate": 2.3589086689101889e-07, "logits/chosen": -2.427687168121338, "logits/rejected": -2.4119105339050293, "loss": 1.1311, "step": 387 }, { "beta_dpo/beta": 1.6118403673171997, "beta_dpo/beta_margin_grad_mean": -0.22462284564971924, "beta_dpo/beta_margin_grad_std": 0.23515048623085022, "beta_dpo/beta_margin_mean": 676.41845703125, "beta_dpo/beta_margin_std": 1056.8880615234375, "beta_dpo/beta_used": 1.6118403673171997, "beta_dpo/beta_used_raw": -0.10103893280029297, "beta_dpo/gap_mean": 341.0576477050781, "beta_dpo/gap_std": 298.05126953125, "beta_dpo/loss_margin_mean": 380.3453063964844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5697503671071953, "grad_norm": 17.152055740356445, "learning_rate": 2.3460977346651428e-07, "logits/chosen": -2.349907159805298, "logits/rejected": -2.363679885864258, "loss": 0.5682, "step": 388 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4282563030719757, "beta_dpo/beta_margin_grad_std": 0.08526238799095154, "beta_dpo/beta_margin_mean": 0.2955423891544342, "beta_dpo/beta_margin_std": 0.36567941308021545, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.094040870666504, "beta_dpo/gap_mean": 335.10137939453125, "beta_dpo/gap_std": 306.0793762207031, "beta_dpo/loss_margin_mean": 295.5423583984375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5712187958883994, "grad_norm": 23.256101608276367, "learning_rate": 2.3332908558879177e-07, "logits/chosen": -2.379899024963379, "logits/rejected": -2.373753309249878, "loss": 1.1102, "step": 389 }, { "beta_dpo/beta": 0.6165112257003784, "beta_dpo/beta_margin_grad_mean": -0.0921190157532692, "beta_dpo/beta_margin_grad_std": 0.28663116693496704, "beta_dpo/beta_margin_mean": 242.41787719726562, "beta_dpo/beta_margin_std": 333.33306884765625, "beta_dpo/beta_used": 0.6165112257003784, "beta_dpo/beta_used_raw": 0.6165112257003784, "beta_dpo/gap_mean": 338.8628845214844, "beta_dpo/gap_std": 316.99127197265625, "beta_dpo/loss_margin_mean": 362.915771484375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5726872246696035, "grad_norm": 15604.0400390625, "learning_rate": 2.320488370051681e-07, "logits/chosen": -2.3529043197631836, "logits/rejected": -2.3543856143951416, "loss": 6.7076, "step": 390 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4340205192565918, "beta_dpo/beta_margin_grad_std": 0.06161868944764137, "beta_dpo/beta_margin_mean": 0.26973778009414673, "beta_dpo/beta_margin_std": 0.25571221113204956, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.148557186126709, "beta_dpo/gap_mean": 328.48919677734375, "beta_dpo/gap_std": 299.50042724609375, "beta_dpo/loss_margin_mean": 269.7377624511719, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5741556534508077, "grad_norm": 21.558528900146484, "learning_rate": 2.3076906145138405e-07, "logits/chosen": -2.3723154067993164, "logits/rejected": -2.3779938220977783, "loss": 1.1553, "step": 391 }, { "beta_dpo/beta": 1.9164685010910034, "beta_dpo/beta_margin_grad_mean": -0.06281369924545288, "beta_dpo/beta_margin_grad_std": 0.2419932633638382, "beta_dpo/beta_margin_mean": 839.9161376953125, "beta_dpo/beta_margin_std": 943.3709716796875, "beta_dpo/beta_used": 1.9164685010910034, "beta_dpo/beta_used_raw": 1.9164685010910034, "beta_dpo/gap_mean": 332.79083251953125, "beta_dpo/gap_std": 301.228271484375, "beta_dpo/loss_margin_mean": 401.9228515625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5756240822320118, "grad_norm": 8398.5751953125, "learning_rate": 2.294897926507156e-07, "logits/chosen": -2.405207633972168, "logits/rejected": -2.4119091033935547, "loss": 11.169, "step": 392 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4241882562637329, "beta_dpo/beta_margin_grad_std": 0.07016981393098831, "beta_dpo/beta_margin_mean": 0.313728928565979, "beta_dpo/beta_margin_std": 0.30085715651512146, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.3607499599456787, "beta_dpo/gap_mean": 336.00286865234375, "beta_dpo/gap_std": 306.16607666015625, "beta_dpo/loss_margin_mean": 313.7289123535156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5770925110132159, "grad_norm": 26.408864974975586, "learning_rate": 2.2821106431308543e-07, "logits/chosen": -2.376577854156494, "logits/rejected": -2.37333345413208, "loss": 1.1245, "step": 393 }, { "beta_dpo/beta": 1.376448392868042, "beta_dpo/beta_margin_grad_mean": -0.062322188168764114, "beta_dpo/beta_margin_grad_std": 0.24126726388931274, "beta_dpo/beta_margin_mean": 556.2864990234375, "beta_dpo/beta_margin_std": 614.0519409179688, "beta_dpo/beta_used": 1.376448392868042, "beta_dpo/beta_used_raw": 1.376448392868042, "beta_dpo/gap_mean": 340.67095947265625, "beta_dpo/gap_std": 299.90325927734375, "beta_dpo/loss_margin_mean": 384.258544921875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.57856093979442, "grad_norm": 19108.568359375, "learning_rate": 2.2693291013417452e-07, "logits/chosen": -2.3865888118743896, "logits/rejected": -2.3961262702941895, "loss": 3.1294, "step": 394 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.42161938548088074, "beta_dpo/beta_margin_grad_std": 0.07712607830762863, "beta_dpo/beta_margin_mean": 0.32536518573760986, "beta_dpo/beta_margin_std": 0.3256779909133911, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.202303886413574, "beta_dpo/gap_mean": 342.54486083984375, "beta_dpo/gap_std": 305.68255615234375, "beta_dpo/loss_margin_mean": 325.36517333984375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.580029368575624, "grad_norm": 23.2067813873291, "learning_rate": 2.2565536379453404e-07, "logits/chosen": -2.386712074279785, "logits/rejected": -2.381330966949463, "loss": 1.153, "step": 395 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4167252480983734, "beta_dpo/beta_margin_grad_std": 0.07602323591709137, "beta_dpo/beta_margin_mean": 0.3457351326942444, "beta_dpo/beta_margin_std": 0.32599034905433655, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.017214775085449, "beta_dpo/gap_mean": 341.43377685546875, "beta_dpo/gap_std": 311.3189697265625, "beta_dpo/loss_margin_mean": 345.735107421875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5814977973568282, "grad_norm": 22.437475204467773, "learning_rate": 2.2437845895869825e-07, "logits/chosen": -2.412064552307129, "logits/rejected": -2.3995161056518555, "loss": 1.112, "step": 396 }, { "beta_dpo/beta": 0.7200398445129395, "beta_dpo/beta_margin_grad_mean": -0.22126896679401398, "beta_dpo/beta_margin_grad_std": 0.2345595508813858, "beta_dpo/beta_margin_mean": 340.8731689453125, "beta_dpo/beta_margin_std": 545.7304077148438, "beta_dpo/beta_used": 0.7200398445129395, "beta_dpo/beta_used_raw": 0.6792468428611755, "beta_dpo/gap_mean": 355.73583984375, "beta_dpo/gap_std": 323.22137451171875, "beta_dpo/loss_margin_mean": 421.983154296875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5829662261380323, "grad_norm": 17.3627872467041, "learning_rate": 2.2310222927429716e-07, "logits/chosen": -2.3999505043029785, "logits/rejected": -2.3841006755828857, "loss": 0.5404, "step": 397 }, { "beta_dpo/beta": 1.0242469310760498, "beta_dpo/beta_margin_grad_mean": -0.14056435227394104, "beta_dpo/beta_margin_grad_std": 0.34748461842536926, "beta_dpo/beta_margin_mean": 392.9497985839844, "beta_dpo/beta_margin_std": 412.2462463378906, "beta_dpo/beta_used": 1.0242469310760498, "beta_dpo/beta_used_raw": 1.0242469310760498, "beta_dpo/gap_mean": 359.8270263671875, "beta_dpo/gap_std": 337.1595153808594, "beta_dpo/loss_margin_mean": 382.2912292480469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5844346549192364, "grad_norm": 17113.068359375, "learning_rate": 2.2182670837116972e-07, "logits/chosen": -2.438488483428955, "logits/rejected": -2.441718101501465, "loss": 22.2563, "step": 398 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41422000527381897, "beta_dpo/beta_margin_grad_std": 0.07931336015462875, "beta_dpo/beta_margin_mean": 0.3558643162250519, "beta_dpo/beta_margin_std": 0.3342893421649933, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.7503660917282104, "beta_dpo/gap_mean": 361.33551025390625, "beta_dpo/gap_std": 341.2704162597656, "beta_dpo/loss_margin_mean": 355.8642883300781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5859030837004405, "grad_norm": 25.74508285522461, "learning_rate": 2.2055192986047804e-07, "logits/chosen": -2.4001259803771973, "logits/rejected": -2.3805899620056152, "loss": 1.1024, "step": 399 }, { "beta_dpo/beta": 0.27362868189811707, "beta_dpo/beta_margin_grad_mean": -0.26691824197769165, "beta_dpo/beta_margin_grad_std": 0.2769628167152405, "beta_dpo/beta_margin_mean": 99.65248107910156, "beta_dpo/beta_margin_std": 166.322265625, "beta_dpo/beta_used": 0.27362868189811707, "beta_dpo/beta_used_raw": 0.03512483835220337, "beta_dpo/gap_mean": 361.3894348144531, "beta_dpo/gap_std": 336.9342956542969, "beta_dpo/loss_margin_mean": 369.5531921386719, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5873715124816447, "grad_norm": 8530.029296875, "learning_rate": 2.192779273338215e-07, "logits/chosen": -2.4256482124328613, "logits/rejected": -2.4234414100646973, "loss": 3.7878, "step": 400 }, { "epoch": 0.5873715124816447, "eval_beta_dpo/beta": 0.0010000000474974513, "eval_beta_dpo/beta_margin_grad_mean": -0.45058003067970276, "eval_beta_dpo/beta_margin_grad_std": 0.08448585122823715, "eval_beta_dpo/beta_margin_mean": 0.2054125964641571, "eval_beta_dpo/beta_margin_std": 0.35709553956985474, "eval_beta_dpo/beta_used": 0.0010000000474974513, "eval_beta_dpo/beta_used_raw": -9.301395416259766, "eval_beta_dpo/gap_mean": 362.1024475097656, "eval_beta_dpo/gap_std": 333.2911682128906, "eval_beta_dpo/loss_margin_mean": 205.41256713867188, "eval_beta_dpo/mask_keep_frac": 1.0, "eval_logits/chosen": -2.443066358566284, "eval_logits/rejected": -2.4332497119903564, "eval_loss": 0.6121687293052673, "eval_runtime": 37.7883, "eval_samples_per_second": 61.897, "eval_steps_per_second": 1.958, "step": 400 }, { "beta_dpo/beta": 1.0822142362594604, "beta_dpo/beta_margin_grad_mean": -0.27038997411727905, "beta_dpo/beta_margin_grad_std": 0.27789729833602905, "beta_dpo/beta_margin_mean": 469.0853576660156, "beta_dpo/beta_margin_std": 778.7431030273438, "beta_dpo/beta_used": 1.0822142362594604, "beta_dpo/beta_used_raw": -1.9135266542434692, "beta_dpo/gap_mean": 361.82781982421875, "beta_dpo/gap_std": 342.1406555175781, "beta_dpo/loss_margin_mean": 327.20556640625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5888399412628488, "grad_norm": 3323.7548828125, "learning_rate": 2.1800473436235136e-07, "logits/chosen": -2.4136600494384766, "logits/rejected": -2.4133119583129883, "loss": 3.0841, "step": 401 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41412436962127686, "beta_dpo/beta_margin_grad_std": 0.0651504173874855, "beta_dpo/beta_margin_mean": 0.35367244482040405, "beta_dpo/beta_margin_std": 0.2750694453716278, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.9857521057128906, "beta_dpo/gap_mean": 353.83746337890625, "beta_dpo/gap_std": 333.810302734375, "beta_dpo/loss_margin_mean": 353.67242431640625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5903083700440529, "grad_norm": 22.399593353271484, "learning_rate": 2.1673238449588665e-07, "logits/chosen": -2.428457736968994, "logits/rejected": -2.423470973968506, "loss": 1.1155, "step": 402 }, { "beta_dpo/beta": 2.470536470413208, "beta_dpo/beta_margin_grad_mean": -0.2379431277513504, "beta_dpo/beta_margin_grad_std": 0.2529750466346741, "beta_dpo/beta_margin_mean": 895.9714965820312, "beta_dpo/beta_margin_std": 1318.6358642578125, "beta_dpo/beta_used": 2.470536470413208, "beta_dpo/beta_used_raw": 0.7853410243988037, "beta_dpo/gap_mean": 355.718994140625, "beta_dpo/gap_std": 329.26959228515625, "beta_dpo/loss_margin_mean": 364.1114807128906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.591776798825257, "grad_norm": 11.740818977355957, "learning_rate": 2.154609112620295e-07, "logits/chosen": -2.408144474029541, "logits/rejected": -2.403205394744873, "loss": 0.5649, "step": 403 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4196135401725769, "beta_dpo/beta_margin_grad_std": 0.08152054995298386, "beta_dpo/beta_margin_mean": 0.3342151939868927, "beta_dpo/beta_margin_std": 0.3467388153076172, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.5054094791412354, "beta_dpo/gap_mean": 351.60748291015625, "beta_dpo/gap_std": 336.61431884765625, "beta_dpo/loss_margin_mean": 334.2151794433594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.593245227606461, "grad_norm": 22.172229766845703, "learning_rate": 2.1419034816528218e-07, "logits/chosen": -2.3768763542175293, "logits/rejected": -2.3775620460510254, "loss": 1.1159, "step": 404 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4178047478199005, "beta_dpo/beta_margin_grad_std": 0.08822597563266754, "beta_dpo/beta_margin_mean": 0.34523844718933105, "beta_dpo/beta_margin_std": 0.38265952467918396, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -5.212499618530273, "beta_dpo/gap_mean": 349.62945556640625, "beta_dpo/gap_std": 341.0721435546875, "beta_dpo/loss_margin_mean": 345.2384338378906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5947136563876652, "grad_norm": 24.03404426574707, "learning_rate": 2.129207286861638e-07, "logits/chosen": -2.394247531890869, "logits/rejected": -2.390904426574707, "loss": 1.1631, "step": 405 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40800565481185913, "beta_dpo/beta_margin_grad_std": 0.06867733597755432, "beta_dpo/beta_margin_mean": 0.3816227912902832, "beta_dpo/beta_margin_std": 0.29813575744628906, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.0216950178146362, "beta_dpo/gap_mean": 353.78961181640625, "beta_dpo/gap_std": 335.23284912109375, "beta_dpo/loss_margin_mean": 381.6227722167969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5961820851688693, "grad_norm": 22.132526397705078, "learning_rate": 2.1165208628032861e-07, "logits/chosen": -2.3941879272460938, "logits/rejected": -2.3966739177703857, "loss": 1.0938, "step": 406 }, { "beta_dpo/beta": 1.3681471347808838, "beta_dpo/beta_margin_grad_mean": -0.2872449457645416, "beta_dpo/beta_margin_grad_std": 0.2897031307220459, "beta_dpo/beta_margin_mean": 529.73486328125, "beta_dpo/beta_margin_std": 880.0147094726562, "beta_dpo/beta_used": 1.3681471347808838, "beta_dpo/beta_used_raw": -1.0896592140197754, "beta_dpo/gap_mean": 344.74383544921875, "beta_dpo/gap_std": 335.9972229003906, "beta_dpo/loss_margin_mean": 297.04425048828125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5976505139500734, "grad_norm": 15591.009765625, "learning_rate": 2.1038445437768375e-07, "logits/chosen": -2.4210031032562256, "logits/rejected": -2.4093196392059326, "loss": 7.8766, "step": 407 }, { "beta_dpo/beta": 0.12347578257322311, "beta_dpo/beta_margin_grad_mean": -0.28025034070014954, "beta_dpo/beta_margin_grad_std": 0.2851478159427643, "beta_dpo/beta_margin_mean": 42.36891174316406, "beta_dpo/beta_margin_std": 62.56731414794922, "beta_dpo/beta_used": 0.12347578257322311, "beta_dpo/beta_used_raw": -1.5166099071502686, "beta_dpo/gap_mean": 342.59320068359375, "beta_dpo/gap_std": 329.6495361328125, "beta_dpo/loss_margin_mean": 304.6890563964844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.5991189427312775, "grad_norm": 2112.055419921875, "learning_rate": 2.0911786638150872e-07, "logits/chosen": -2.4404101371765137, "logits/rejected": -2.420276641845703, "loss": 2.3999, "step": 408 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4296059310436249, "beta_dpo/beta_margin_grad_std": 0.08227775245904922, "beta_dpo/beta_margin_mean": 0.29228663444519043, "beta_dpo/beta_margin_std": 0.34818896651268005, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.490858793258667, "beta_dpo/gap_mean": 331.255126953125, "beta_dpo/gap_std": 332.9500732421875, "beta_dpo/loss_margin_mean": 292.28662109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6005873715124816, "grad_norm": 20.568885803222656, "learning_rate": 2.0785235566757517e-07, "logits/chosen": -2.4326343536376953, "logits/rejected": -2.4223175048828125, "loss": 1.1187, "step": 409 }, { "beta_dpo/beta": 1.8076552152633667, "beta_dpo/beta_margin_grad_mean": -0.24894118309020996, "beta_dpo/beta_margin_grad_std": 0.2569856345653534, "beta_dpo/beta_margin_mean": 749.4814453125, "beta_dpo/beta_margin_std": 1033.7783203125, "beta_dpo/beta_used": 1.8076552152633667, "beta_dpo/beta_used_raw": 0.11705267429351807, "beta_dpo/gap_mean": 334.65582275390625, "beta_dpo/gap_std": 324.4977111816406, "beta_dpo/loss_margin_mean": 339.3305358886719, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6020558002936858, "grad_norm": 28801.537109375, "learning_rate": 2.065879555832674e-07, "logits/chosen": -2.426015853881836, "logits/rejected": -2.4410088062286377, "loss": 1.872, "step": 410 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4142020046710968, "beta_dpo/beta_margin_grad_std": 0.06422904133796692, "beta_dpo/beta_margin_mean": 0.35382363200187683, "beta_dpo/beta_margin_std": 0.27380073070526123, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.1547770500183105, "beta_dpo/gap_mean": 335.4295654296875, "beta_dpo/gap_std": 315.85736083984375, "beta_dpo/loss_margin_mean": 353.8236083984375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6035242290748899, "grad_norm": 19.184236526489258, "learning_rate": 2.0532469944670343e-07, "logits/chosen": -2.3894524574279785, "logits/rejected": -2.401761293411255, "loss": 1.1076, "step": 411 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4059164226055145, "beta_dpo/beta_margin_grad_std": 0.07964363694190979, "beta_dpo/beta_margin_mean": 0.39483025670051575, "beta_dpo/beta_margin_std": 0.3526383638381958, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.261457920074463, "beta_dpo/gap_mean": 340.3641052246094, "beta_dpo/gap_std": 316.4858093261719, "beta_dpo/loss_margin_mean": 394.83026123046875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.604992657856094, "grad_norm": 22.225645065307617, "learning_rate": 2.0406262054585738e-07, "logits/chosen": -2.419179677963257, "logits/rejected": -2.4454574584960938, "loss": 1.1086, "step": 412 }, { "beta_dpo/beta": 0.9109219908714294, "beta_dpo/beta_margin_grad_mean": -0.09896637499332428, "beta_dpo/beta_margin_grad_std": 0.2925999164581299, "beta_dpo/beta_margin_mean": 362.6519775390625, "beta_dpo/beta_margin_std": 495.7064514160156, "beta_dpo/beta_used": 0.9109219908714294, "beta_dpo/beta_used_raw": 0.9109219908714294, "beta_dpo/gap_mean": 349.16204833984375, "beta_dpo/gap_std": 321.62481689453125, "beta_dpo/loss_margin_mean": 370.2857666015625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6064610866372981, "grad_norm": 12787.8896484375, "learning_rate": 2.0280175213768205e-07, "logits/chosen": -2.421027660369873, "logits/rejected": -2.4294235706329346, "loss": 7.296, "step": 413 }, { "beta_dpo/beta": 2.4083163738250732, "beta_dpo/beta_margin_grad_mean": -0.23511956632137299, "beta_dpo/beta_margin_grad_std": 0.24318645894527435, "beta_dpo/beta_margin_mean": 1103.0013427734375, "beta_dpo/beta_margin_std": 1460.6190185546875, "beta_dpo/beta_used": 2.4083163738250732, "beta_dpo/beta_used_raw": -0.48015403747558594, "beta_dpo/gap_mean": 347.5014343261719, "beta_dpo/gap_std": 315.90374755859375, "beta_dpo/loss_margin_mean": 354.2364501953125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6079295154185022, "grad_norm": 15.694453239440918, "learning_rate": 2.0154212744723247e-07, "logits/chosen": -2.4023854732513428, "logits/rejected": -2.420386552810669, "loss": 0.5842, "step": 414 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41483768820762634, "beta_dpo/beta_margin_grad_std": 0.07904361933469772, "beta_dpo/beta_margin_mean": 0.35519102215766907, "beta_dpo/beta_margin_std": 0.3411670923233032, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.372532367706299, "beta_dpo/gap_mean": 357.978271484375, "beta_dpo/gap_std": 318.06103515625, "beta_dpo/loss_margin_mean": 355.1910095214844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6093979441997063, "grad_norm": 24.613069534301758, "learning_rate": 2.002837796667909e-07, "logits/chosen": -2.423779010772705, "logits/rejected": -2.4402318000793457, "loss": 1.1358, "step": 415 }, { "beta_dpo/beta": 2.4572019577026367, "beta_dpo/beta_margin_grad_mean": -0.22583799064159393, "beta_dpo/beta_margin_grad_std": 0.2351226955652237, "beta_dpo/beta_margin_mean": 1054.186767578125, "beta_dpo/beta_margin_std": 1374.137451171875, "beta_dpo/beta_used": 2.4572019577026367, "beta_dpo/beta_used_raw": -8.606910705566406e-05, "beta_dpo/gap_mean": 354.178955078125, "beta_dpo/gap_std": 311.4892272949219, "beta_dpo/loss_margin_mean": 378.5653991699219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6108663729809104, "grad_norm": 20.630220413208008, "learning_rate": 1.990267419549914e-07, "logits/chosen": -2.422940969467163, "logits/rejected": -2.435427188873291, "loss": 0.5757, "step": 416 }, { "beta_dpo/beta": 0.37648269534111023, "beta_dpo/beta_margin_grad_mean": -0.27004316449165344, "beta_dpo/beta_margin_grad_std": 0.277245432138443, "beta_dpo/beta_margin_mean": 133.46795654296875, "beta_dpo/beta_margin_std": 200.62649536132812, "beta_dpo/beta_used": 0.37648269534111023, "beta_dpo/beta_used_raw": -0.30419254302978516, "beta_dpo/gap_mean": 356.74456787109375, "beta_dpo/gap_std": 307.15887451171875, "beta_dpo/loss_margin_mean": 345.98773193359375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6123348017621145, "grad_norm": 27230.5078125, "learning_rate": 1.9777104743594686e-07, "logits/chosen": -2.422550678253174, "logits/rejected": -2.4108152389526367, "loss": 12.2488, "step": 417 }, { "beta_dpo/beta": 3.925933837890625, "beta_dpo/beta_margin_grad_mean": -0.21461881697177887, "beta_dpo/beta_margin_grad_std": 0.22274993360042572, "beta_dpo/beta_margin_mean": 1944.69580078125, "beta_dpo/beta_margin_std": 2413.48388671875, "beta_dpo/beta_used": 3.925933837890625, "beta_dpo/beta_used_raw": 2.9994614124298096, "beta_dpo/gap_mean": 356.8074645996094, "beta_dpo/gap_std": 310.31988525390625, "beta_dpo/loss_margin_mean": 394.4806213378906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6138032305433186, "grad_norm": 15.336505889892578, "learning_rate": 1.965167291983757e-07, "logits/chosen": -2.4672048091888428, "logits/rejected": -2.467080593109131, "loss": 0.5562, "step": 418 }, { "beta_dpo/beta": 0.24180491268634796, "beta_dpo/beta_margin_grad_mean": -0.24379843473434448, "beta_dpo/beta_margin_grad_std": 0.24562664330005646, "beta_dpo/beta_margin_mean": 92.98143005371094, "beta_dpo/beta_margin_std": 130.87518310546875, "beta_dpo/beta_used": 0.24180491268634796, "beta_dpo/beta_used_raw": -0.4406244158744812, "beta_dpo/gap_mean": 366.676513671875, "beta_dpo/gap_std": 303.807373046875, "beta_dpo/loss_margin_mean": 380.661376953125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6152716593245228, "grad_norm": 3546.734375, "learning_rate": 1.9526382029472988e-07, "logits/chosen": -2.4470913410186768, "logits/rejected": -2.453092336654663, "loss": 0.6635, "step": 419 }, { "beta_dpo/beta": 1.377697229385376, "beta_dpo/beta_margin_grad_mean": -0.12197641283273697, "beta_dpo/beta_margin_grad_std": 0.32349058985710144, "beta_dpo/beta_margin_mean": 463.7217102050781, "beta_dpo/beta_margin_std": 575.2052612304688, "beta_dpo/beta_used": 1.377697229385376, "beta_dpo/beta_used_raw": 1.377697229385376, "beta_dpo/gap_mean": 362.47210693359375, "beta_dpo/gap_std": 298.56646728515625, "beta_dpo/loss_margin_mean": 335.21160888671875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6167400881057269, "grad_norm": 19327.3125, "learning_rate": 1.9401235374032425e-07, "logits/chosen": -2.411374092102051, "logits/rejected": -2.393552780151367, "loss": 14.9604, "step": 420 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.411777138710022, "beta_dpo/beta_margin_grad_std": 0.07005602866411209, "beta_dpo/beta_margin_mean": 0.36520177125930786, "beta_dpo/beta_margin_std": 0.29842692613601685, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.6941306591033936, "beta_dpo/gap_mean": 360.69580078125, "beta_dpo/gap_std": 296.3153076171875, "beta_dpo/loss_margin_mean": 365.2017517089844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.618208516886931, "grad_norm": 20.577306747436523, "learning_rate": 1.9276236251246653e-07, "logits/chosen": -2.4650721549987793, "logits/rejected": -2.475637912750244, "loss": 1.0969, "step": 421 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4262285530567169, "beta_dpo/beta_margin_grad_std": 0.08390183001756668, "beta_dpo/beta_margin_mean": 0.30507245659828186, "beta_dpo/beta_margin_std": 0.3505830466747284, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.5142788887023926, "beta_dpo/gap_mean": 356.1475524902344, "beta_dpo/gap_std": 305.0068664550781, "beta_dpo/loss_margin_mean": 305.07244873046875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6196769456681351, "grad_norm": 27.27686309814453, "learning_rate": 1.9151387954958792e-07, "logits/chosen": -2.433502674102783, "logits/rejected": -2.4503097534179688, "loss": 1.1011, "step": 422 }, { "beta_dpo/beta": 2.667612075805664, "beta_dpo/beta_margin_grad_mean": -0.09069596976041794, "beta_dpo/beta_margin_grad_std": 0.2812235653400421, "beta_dpo/beta_margin_mean": 1046.2611083984375, "beta_dpo/beta_margin_std": 1394.6136474609375, "beta_dpo/beta_used": 2.667612075805664, "beta_dpo/beta_used_raw": 2.667612075805664, "beta_dpo/gap_mean": 355.60418701171875, "beta_dpo/gap_std": 305.63763427734375, "beta_dpo/loss_margin_mean": 379.52520751953125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6211453744493393, "grad_norm": 49413.26953125, "learning_rate": 1.902669377503756e-07, "logits/chosen": -2.4104795455932617, "logits/rejected": -2.43206787109375, "loss": 27.8684, "step": 423 }, { "beta_dpo/beta": 0.14746440947055817, "beta_dpo/beta_margin_grad_mean": -0.2809794247150421, "beta_dpo/beta_margin_grad_std": 0.2841648757457733, "beta_dpo/beta_margin_mean": 50.7834587097168, "beta_dpo/beta_margin_std": 84.9849624633789, "beta_dpo/beta_used": 0.14746440947055817, "beta_dpo/beta_used_raw": -1.0671793222427368, "beta_dpo/gap_mean": 350.1261291503906, "beta_dpo/gap_std": 303.9034729003906, "beta_dpo/loss_margin_mean": 301.35723876953125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6226138032305433, "grad_norm": 2292.713134765625, "learning_rate": 1.890215699729057e-07, "logits/chosen": -2.436124086380005, "logits/rejected": -2.434342861175537, "loss": 1.9952, "step": 424 }, { "beta_dpo/beta": 0.3063303828239441, "beta_dpo/beta_margin_grad_mean": -0.2702360153198242, "beta_dpo/beta_margin_grad_std": 0.28220611810684204, "beta_dpo/beta_margin_mean": 115.15734100341797, "beta_dpo/beta_margin_std": 173.1321258544922, "beta_dpo/beta_used": 0.3063303828239441, "beta_dpo/beta_used_raw": -0.04413789510726929, "beta_dpo/gap_mean": 347.618896484375, "beta_dpo/gap_std": 309.3476257324219, "beta_dpo/loss_margin_mean": 364.8583068847656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6240822320117474, "grad_norm": 3245.521240234375, "learning_rate": 1.8777780903377732e-07, "logits/chosen": -2.4158806800842285, "logits/rejected": -2.452991008758545, "loss": 2.1748, "step": 425 }, { "beta_dpo/beta": 2.3221449851989746, "beta_dpo/beta_margin_grad_mean": -0.23410384356975555, "beta_dpo/beta_margin_grad_std": 0.2420140653848648, "beta_dpo/beta_margin_mean": 937.5267944335938, "beta_dpo/beta_margin_std": 1492.68994140625, "beta_dpo/beta_used": 2.3221449851989746, "beta_dpo/beta_used_raw": 0.11249232292175293, "beta_dpo/gap_mean": 343.1509704589844, "beta_dpo/gap_std": 309.1755676269531, "beta_dpo/loss_margin_mean": 330.6883850097656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6255506607929515, "grad_norm": 20.291976928710938, "learning_rate": 1.8653568770724803e-07, "logits/chosen": -2.4429891109466553, "logits/rejected": -2.435697078704834, "loss": 0.5759, "step": 426 }, { "beta_dpo/beta": 0.05136517435312271, "beta_dpo/beta_margin_grad_mean": -0.29277148842811584, "beta_dpo/beta_margin_grad_std": 0.290822833776474, "beta_dpo/beta_margin_mean": 14.553664207458496, "beta_dpo/beta_margin_std": 25.774181365966797, "beta_dpo/beta_used": 0.05136517435312271, "beta_dpo/beta_used_raw": -2.2542648315429688, "beta_dpo/gap_mean": 339.93310546875, "beta_dpo/gap_std": 314.22222900390625, "beta_dpo/loss_margin_mean": 297.6253662109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6270190895741556, "grad_norm": 1669.3848876953125, "learning_rate": 1.8529523872436977e-07, "logits/chosen": -2.484468460083008, "logits/rejected": -2.4828639030456543, "loss": 1.797, "step": 427 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4063303768634796, "beta_dpo/beta_margin_grad_std": 0.08325894176959991, "beta_dpo/beta_margin_mean": 0.3934994637966156, "beta_dpo/beta_margin_std": 0.36640259623527527, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.5234259366989136, "beta_dpo/gap_mean": 344.0052490234375, "beta_dpo/gap_std": 324.0714416503906, "beta_dpo/loss_margin_mean": 393.49945068359375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6284875183553598, "grad_norm": 29.064952850341797, "learning_rate": 1.8405649477212697e-07, "logits/chosen": -2.44685697555542, "logits/rejected": -2.4712371826171875, "loss": 1.1086, "step": 428 }, { "beta_dpo/beta": 1.4477723836898804, "beta_dpo/beta_margin_grad_mean": -0.2723529636859894, "beta_dpo/beta_margin_grad_std": 0.28279608488082886, "beta_dpo/beta_margin_mean": 609.0072631835938, "beta_dpo/beta_margin_std": 894.19482421875, "beta_dpo/beta_used": 1.4477723836898804, "beta_dpo/beta_used_raw": 0.639708936214447, "beta_dpo/gap_mean": 350.4301452636719, "beta_dpo/gap_std": 327.36663818359375, "beta_dpo/loss_margin_mean": 378.28228759765625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6299559471365639, "grad_norm": 38852.13671875, "learning_rate": 1.828194884925749e-07, "logits/chosen": -2.4718470573425293, "logits/rejected": -2.4727447032928467, "loss": 23.4925, "step": 429 }, { "beta_dpo/beta": 1.2469240427017212, "beta_dpo/beta_margin_grad_mean": -0.27254435420036316, "beta_dpo/beta_margin_grad_std": 0.2825857698917389, "beta_dpo/beta_margin_mean": 510.65069580078125, "beta_dpo/beta_margin_std": 836.8815307617188, "beta_dpo/beta_used": 1.2469240427017212, "beta_dpo/beta_used_raw": 0.10372543334960938, "beta_dpo/gap_mean": 358.3047180175781, "beta_dpo/gap_std": 334.221923828125, "beta_dpo/loss_margin_mean": 370.95501708984375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.631424375917768, "grad_norm": 50662.0078125, "learning_rate": 1.8158425248197928e-07, "logits/chosen": -2.4730615615844727, "logits/rejected": -2.498109817504883, "loss": 15.693, "step": 430 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4051775336265564, "beta_dpo/beta_margin_grad_std": 0.07288450002670288, "beta_dpo/beta_margin_mean": 0.3952929973602295, "beta_dpo/beta_margin_std": 0.3199489414691925, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.4916090965270996, "beta_dpo/gap_mean": 365.2705383300781, "beta_dpo/gap_std": 334.9329833984375, "beta_dpo/loss_margin_mean": 395.29296875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6328928046989721, "grad_norm": 24.509140014648438, "learning_rate": 1.8035081928995788e-07, "logits/chosen": -2.456411361694336, "logits/rejected": -2.4808506965637207, "loss": 1.1022, "step": 431 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41120636463165283, "beta_dpo/beta_margin_grad_std": 0.066596120595932, "beta_dpo/beta_margin_mean": 0.36704325675964355, "beta_dpo/beta_margin_std": 0.28722265362739563, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.3621997833251953, "beta_dpo/gap_mean": 362.98553466796875, "beta_dpo/gap_std": 325.4151611328125, "beta_dpo/loss_margin_mean": 367.0432434082031, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6343612334801763, "grad_norm": 25.281761169433594, "learning_rate": 1.791192214186223e-07, "logits/chosen": -2.482973098754883, "logits/rejected": -2.493764877319336, "loss": 1.1166, "step": 432 }, { "beta_dpo/beta": 2.1224658489227295, "beta_dpo/beta_margin_grad_mean": -0.2876777946949005, "beta_dpo/beta_margin_grad_std": 0.292548805475235, "beta_dpo/beta_margin_mean": 841.04052734375, "beta_dpo/beta_margin_std": 1226.0413818359375, "beta_dpo/beta_used": 2.1224658489227295, "beta_dpo/beta_used_raw": 0.5382475852966309, "beta_dpo/gap_mean": 363.06561279296875, "beta_dpo/gap_std": 322.05609130859375, "beta_dpo/loss_margin_mean": 364.42138671875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6358296622613803, "grad_norm": 59117.21484375, "learning_rate": 1.7788949132172193e-07, "logits/chosen": -2.4168553352355957, "logits/rejected": -2.4499008655548096, "loss": 13.9108, "step": 433 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4250311851501465, "beta_dpo/beta_margin_grad_std": 0.07213795185089111, "beta_dpo/beta_margin_mean": 0.317569375038147, "beta_dpo/beta_margin_std": 0.3578694462776184, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.1827993392944336, "beta_dpo/gap_mean": 360.7991943359375, "beta_dpo/gap_std": 328.626220703125, "beta_dpo/loss_margin_mean": 317.5693664550781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6372980910425844, "grad_norm": 21.91067123413086, "learning_rate": 1.7666166140378853e-07, "logits/chosen": -2.5135700702667236, "logits/rejected": -2.504258155822754, "loss": 1.1136, "step": 434 }, { "beta_dpo/beta": 0.6958755850791931, "beta_dpo/beta_margin_grad_mean": -0.046905096620321274, "beta_dpo/beta_margin_grad_std": 0.2113645374774933, "beta_dpo/beta_margin_mean": 274.05462646484375, "beta_dpo/beta_margin_std": 276.3533935546875, "beta_dpo/beta_used": 0.6958755850791931, "beta_dpo/beta_used_raw": 0.6958755850791931, "beta_dpo/gap_mean": 360.8138427734375, "beta_dpo/gap_std": 327.732177734375, "beta_dpo/loss_margin_mean": 387.7349548339844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6387665198237885, "grad_norm": 10641.4619140625, "learning_rate": 1.7543576401928218e-07, "logits/chosen": -2.479780673980713, "logits/rejected": -2.4990386962890625, "loss": 14.9256, "step": 435 }, { "beta_dpo/beta": 1.1166325807571411, "beta_dpo/beta_margin_grad_mean": -0.22967371344566345, "beta_dpo/beta_margin_grad_std": 0.2383747398853302, "beta_dpo/beta_margin_mean": 423.3056335449219, "beta_dpo/beta_margin_std": 537.5963134765625, "beta_dpo/beta_used": 1.1166325807571411, "beta_dpo/beta_used_raw": -0.7592837810516357, "beta_dpo/gap_mean": 359.64361572265625, "beta_dpo/gap_std": 311.7923583984375, "beta_dpo/loss_margin_mean": 337.6617736816406, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6402349486049926, "grad_norm": 35.73384094238281, "learning_rate": 1.742118314717391e-07, "logits/chosen": -2.4742934703826904, "logits/rejected": -2.4709324836730957, "loss": 0.5629, "step": 436 }, { "beta_dpo/beta": 0.42722299695014954, "beta_dpo/beta_margin_grad_mean": -0.26437872648239136, "beta_dpo/beta_margin_grad_std": 0.2714844048023224, "beta_dpo/beta_margin_mean": 150.46022033691406, "beta_dpo/beta_margin_std": 217.8881378173828, "beta_dpo/beta_used": 0.42722299695014954, "beta_dpo/beta_used_raw": 0.32361477613449097, "beta_dpo/gap_mean": 351.33404541015625, "beta_dpo/gap_std": 304.4229431152344, "beta_dpo/loss_margin_mean": 308.41162109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6417033773861968, "grad_norm": 7131.47998046875, "learning_rate": 1.7298989601292036e-07, "logits/chosen": -2.4637017250061035, "logits/rejected": -2.454047203063965, "loss": 1.8349, "step": 437 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4128943979740143, "beta_dpo/beta_margin_grad_std": 0.06244545429944992, "beta_dpo/beta_margin_mean": 0.35861438512802124, "beta_dpo/beta_margin_std": 0.26432162523269653, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.8020297884941101, "beta_dpo/gap_mean": 347.9464111328125, "beta_dpo/gap_std": 297.93682861328125, "beta_dpo/loss_margin_mean": 358.6143798828125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6431718061674009, "grad_norm": 25.952686309814453, "learning_rate": 1.7176998984196144e-07, "logits/chosen": -2.4292826652526855, "logits/rejected": -2.4409079551696777, "loss": 1.0941, "step": 438 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.42735329270362854, "beta_dpo/beta_margin_grad_std": 0.0616798959672451, "beta_dpo/beta_margin_mean": 0.2987552881240845, "beta_dpo/beta_margin_std": 0.26535582542419434, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.115419864654541, "beta_dpo/gap_mean": 343.51715087890625, "beta_dpo/gap_std": 292.8160400390625, "beta_dpo/loss_margin_mean": 298.7552490234375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.644640234948605, "grad_norm": 27.892019271850586, "learning_rate": 1.7055214510452458e-07, "logits/chosen": -2.435819149017334, "logits/rejected": -2.4549179077148438, "loss": 1.1119, "step": 439 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4165194034576416, "beta_dpo/beta_margin_grad_std": 0.07028567790985107, "beta_dpo/beta_margin_mean": 0.34535932540893555, "beta_dpo/beta_margin_std": 0.2985744774341583, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.8327579498291016, "beta_dpo/gap_mean": 340.54168701171875, "beta_dpo/gap_std": 291.77471923828125, "beta_dpo/loss_margin_mean": 345.3592834472656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6461086637298091, "grad_norm": 34.85447692871094, "learning_rate": 1.6933639389195134e-07, "logits/chosen": -2.4840898513793945, "logits/rejected": -2.5059878826141357, "loss": 1.12, "step": 440 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4149302840232849, "beta_dpo/beta_margin_grad_std": 0.06381641328334808, "beta_dpo/beta_margin_mean": 0.3505040109157562, "beta_dpo/beta_margin_std": 0.2713123559951782, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.8756400346755981, "beta_dpo/gap_mean": 340.91888427734375, "beta_dpo/gap_std": 290.773681640625, "beta_dpo/loss_margin_mean": 350.50396728515625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6475770925110133, "grad_norm": 23.49580192565918, "learning_rate": 1.681227682404166e-07, "logits/chosen": -2.4471378326416016, "logits/rejected": -2.469273805618286, "loss": 1.0976, "step": 441 }, { "beta_dpo/beta": 2.011472702026367, "beta_dpo/beta_margin_grad_mean": -0.06538188457489014, "beta_dpo/beta_margin_grad_std": 0.24239790439605713, "beta_dpo/beta_margin_mean": 782.205322265625, "beta_dpo/beta_margin_std": 524.3056030273438, "beta_dpo/beta_used": 2.011472702026367, "beta_dpo/beta_used_raw": 2.011472702026367, "beta_dpo/gap_mean": 348.65679931640625, "beta_dpo/gap_std": 282.76776123046875, "beta_dpo/loss_margin_mean": 388.67510986328125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6490455212922174, "grad_norm": 25934.462890625, "learning_rate": 1.669113001300851e-07, "logits/chosen": -2.450514316558838, "logits/rejected": -2.4609298706054688, "loss": 6.0554, "step": 442 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.42931288480758667, "beta_dpo/beta_margin_grad_std": 0.07746709883213043, "beta_dpo/beta_margin_mean": 0.2938655614852905, "beta_dpo/beta_margin_std": 0.33218199014663696, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.1781232357025146, "beta_dpo/gap_mean": 344.43426513671875, "beta_dpo/gap_std": 289.60614013671875, "beta_dpo/loss_margin_mean": 293.86553955078125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6505139500734214, "grad_norm": 22.906946182250977, "learning_rate": 1.6570202148426815e-07, "logits/chosen": -2.47257661819458, "logits/rejected": -2.4808216094970703, "loss": 1.1169, "step": 443 }, { "beta_dpo/beta": 0.8307179808616638, "beta_dpo/beta_margin_grad_mean": -0.20021703839302063, "beta_dpo/beta_margin_grad_std": 0.20530231297016144, "beta_dpo/beta_margin_mean": 372.78521728515625, "beta_dpo/beta_margin_std": 491.99072265625, "beta_dpo/beta_used": 0.8307179808616638, "beta_dpo/beta_used_raw": 0.622127115726471, "beta_dpo/gap_mean": 353.92706298828125, "beta_dpo/gap_std": 291.21728515625, "beta_dpo/loss_margin_mean": 429.93499755859375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6519823788546255, "grad_norm": 21.445575714111328, "learning_rate": 1.6449496416858282e-07, "logits/chosen": -2.4195876121520996, "logits/rejected": -2.456974983215332, "loss": 0.5403, "step": 444 }, { "beta_dpo/beta": 0.6511656641960144, "beta_dpo/beta_margin_grad_mean": -0.292864054441452, "beta_dpo/beta_margin_grad_std": 0.29051047563552856, "beta_dpo/beta_margin_mean": 210.04498291015625, "beta_dpo/beta_margin_std": 451.1952819824219, "beta_dpo/beta_used": 0.6511656641960144, "beta_dpo/beta_used_raw": -0.10653191804885864, "beta_dpo/gap_mean": 353.89892578125, "beta_dpo/gap_std": 299.44140625, "beta_dpo/loss_margin_mean": 332.807861328125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6534508076358296, "grad_norm": 12001.302734375, "learning_rate": 1.6329015999011182e-07, "logits/chosen": -2.465177059173584, "logits/rejected": -2.4750633239746094, "loss": 23.2322, "step": 445 }, { "beta_dpo/beta": 0.5344843864440918, "beta_dpo/beta_margin_grad_mean": -0.2088708132505417, "beta_dpo/beta_margin_grad_std": 0.21212056279182434, "beta_dpo/beta_margin_mean": 277.230224609375, "beta_dpo/beta_margin_std": 368.5066223144531, "beta_dpo/beta_used": 0.5344843864440918, "beta_dpo/beta_used_raw": -0.5909568667411804, "beta_dpo/gap_mean": 359.41986083984375, "beta_dpo/gap_std": 302.03582763671875, "beta_dpo/loss_margin_mean": 429.4493713378906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6549192364170338, "grad_norm": 36.90497970581055, "learning_rate": 1.6208764069656578e-07, "logits/chosen": -2.422713041305542, "logits/rejected": -2.4601097106933594, "loss": 0.5524, "step": 446 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4114223122596741, "beta_dpo/beta_margin_grad_std": 0.07890864461660385, "beta_dpo/beta_margin_mean": 0.3711691200733185, "beta_dpo/beta_margin_std": 0.34903499484062195, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.839432716369629, "beta_dpo/gap_mean": 369.88690185546875, "beta_dpo/gap_std": 312.835205078125, "beta_dpo/loss_margin_mean": 371.16912841796875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6563876651982379, "grad_norm": 23.881223678588867, "learning_rate": 1.608874379754465e-07, "logits/chosen": -2.469653606414795, "logits/rejected": -2.506166458129883, "loss": 1.1173, "step": 447 }, { "beta_dpo/beta": 0.9949462413787842, "beta_dpo/beta_margin_grad_mean": -0.22382524609565735, "beta_dpo/beta_margin_grad_std": 0.23410984873771667, "beta_dpo/beta_margin_mean": 429.1389465332031, "beta_dpo/beta_margin_std": 562.9327392578125, "beta_dpo/beta_used": 0.9949462413787842, "beta_dpo/beta_used_raw": -2.4666709899902344, "beta_dpo/gap_mean": 369.296142578125, "beta_dpo/gap_std": 308.22808837890625, "beta_dpo/loss_margin_mean": 389.3961181640625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.657856093979442, "grad_norm": 19.88318634033203, "learning_rate": 1.5968958345321177e-07, "logits/chosen": -2.43017578125, "logits/rejected": -2.461536407470703, "loss": 0.5812, "step": 448 }, { "beta_dpo/beta": 2.169326066970825, "beta_dpo/beta_margin_grad_mean": -0.2861442565917969, "beta_dpo/beta_margin_grad_std": 0.2906011939048767, "beta_dpo/beta_margin_mean": 866.29150390625, "beta_dpo/beta_margin_std": 1363.672119140625, "beta_dpo/beta_used": 2.169326066970825, "beta_dpo/beta_used_raw": 1.4625353813171387, "beta_dpo/gap_mean": 373.68212890625, "beta_dpo/gap_std": 307.01202392578125, "beta_dpo/loss_margin_mean": 371.7033996582031, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6593245227606461, "grad_norm": 50823.05078125, "learning_rate": 1.584941086944423e-07, "logits/chosen": -2.4476749897003174, "logits/rejected": -2.4613916873931885, "loss": 54.4296, "step": 449 }, { "beta_dpo/beta": 1.47364342212677, "beta_dpo/beta_margin_grad_mean": -0.21337755024433136, "beta_dpo/beta_margin_grad_std": 0.2242012321949005, "beta_dpo/beta_margin_mean": 637.6218872070312, "beta_dpo/beta_margin_std": 821.07470703125, "beta_dpo/beta_used": 1.47364342212677, "beta_dpo/beta_used_raw": 1.1894080638885498, "beta_dpo/gap_mean": 380.8287353515625, "beta_dpo/gap_std": 297.178955078125, "beta_dpo/loss_margin_mean": 431.9281311035156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6607929515418502, "grad_norm": 20.199665069580078, "learning_rate": 1.573010452010098e-07, "logits/chosen": -2.45393705368042, "logits/rejected": -2.5022172927856445, "loss": 0.5291, "step": 450 }, { "beta_dpo/beta": 1.6201918125152588, "beta_dpo/beta_margin_grad_mean": -0.2785087823867798, "beta_dpo/beta_margin_grad_std": 0.283831387758255, "beta_dpo/beta_margin_mean": 560.2830810546875, "beta_dpo/beta_margin_std": 958.1199340820312, "beta_dpo/beta_used": 1.6201918125152588, "beta_dpo/beta_used_raw": -0.5515909194946289, "beta_dpo/gap_mean": 371.70465087890625, "beta_dpo/gap_std": 297.33319091796875, "beta_dpo/loss_margin_mean": 312.59564208984375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6622613803230544, "grad_norm": 25548.22265625, "learning_rate": 1.5611042441124687e-07, "logits/chosen": -2.4564037322998047, "logits/rejected": -2.468679428100586, "loss": 14.8356, "step": 451 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4107057452201843, "beta_dpo/beta_margin_grad_std": 0.06496448814868927, "beta_dpo/beta_margin_mean": 0.36832571029663086, "beta_dpo/beta_margin_std": 0.27488240599632263, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.28033387660980225, "beta_dpo/gap_mean": 369.5523376464844, "beta_dpo/gap_std": 295.28765869140625, "beta_dpo/loss_margin_mean": 368.3257141113281, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6637298091042585, "grad_norm": 38.57242202758789, "learning_rate": 1.549222776991186e-07, "logits/chosen": -2.464397668838501, "logits/rejected": -2.5154190063476562, "loss": 1.0722, "step": 452 }, { "beta_dpo/beta": 0.47413191199302673, "beta_dpo/beta_margin_grad_mean": -0.23885342478752136, "beta_dpo/beta_margin_grad_std": 0.25014886260032654, "beta_dpo/beta_margin_mean": 179.6550750732422, "beta_dpo/beta_margin_std": 257.9681396484375, "beta_dpo/beta_used": 0.47413191199302673, "beta_dpo/beta_used_raw": -0.7867467999458313, "beta_dpo/gap_mean": 369.88800048828125, "beta_dpo/gap_std": 293.82562255859375, "beta_dpo/loss_margin_mean": 364.253173828125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6651982378854625, "grad_norm": 7813.66796875, "learning_rate": 1.5373663637339584e-07, "logits/chosen": -2.4826607704162598, "logits/rejected": -2.501310348510742, "loss": 9.4541, "step": 453 }, { "beta_dpo/beta": 0.6042494773864746, "beta_dpo/beta_margin_grad_mean": -0.2383243590593338, "beta_dpo/beta_margin_grad_std": 0.25166648626327515, "beta_dpo/beta_margin_mean": 243.75340270996094, "beta_dpo/beta_margin_std": 361.9959716796875, "beta_dpo/beta_used": 0.6042494773864746, "beta_dpo/beta_used_raw": -0.8156985640525818, "beta_dpo/gap_mean": 371.50396728515625, "beta_dpo/gap_std": 297.9644775390625, "beta_dpo/loss_margin_mean": 380.0973815917969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6666666666666666, "grad_norm": 7821.49853515625, "learning_rate": 1.5255353167683017e-07, "logits/chosen": -2.459460735321045, "logits/rejected": -2.4877805709838867, "loss": 12.3928, "step": 454 }, { "beta_dpo/beta": 0.558521568775177, "beta_dpo/beta_margin_grad_mean": -0.229967400431633, "beta_dpo/beta_margin_grad_std": 0.24533946812152863, "beta_dpo/beta_margin_mean": 263.99224853515625, "beta_dpo/beta_margin_std": 370.77044677734375, "beta_dpo/beta_used": 0.558521568775177, "beta_dpo/beta_used_raw": 0.07860961556434631, "beta_dpo/gap_mean": 381.37921142578125, "beta_dpo/gap_std": 305.9591064453125, "beta_dpo/loss_margin_mean": 450.9393615722656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6681350954478708, "grad_norm": 5878.1142578125, "learning_rate": 1.5137299478533064e-07, "logits/chosen": -2.476715087890625, "logits/rejected": -2.544955253601074, "loss": 4.0629, "step": 455 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.401821494102478, "beta_dpo/beta_margin_grad_std": 0.07234375923871994, "beta_dpo/beta_margin_mean": 0.4079440236091614, "beta_dpo/beta_margin_std": 0.31001952290534973, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.8649553060531616, "beta_dpo/gap_mean": 387.3846435546875, "beta_dpo/gap_std": 308.45947265625, "beta_dpo/loss_margin_mean": 407.9440002441406, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6696035242290749, "grad_norm": 28.53789520263672, "learning_rate": 1.5019505680714232e-07, "logits/chosen": -2.4815030097961426, "logits/rejected": -2.5309150218963623, "loss": 1.0786, "step": 456 }, { "beta_dpo/beta": 0.10005475580692291, "beta_dpo/beta_margin_grad_mean": -0.25547903776168823, "beta_dpo/beta_margin_grad_std": 0.26753029227256775, "beta_dpo/beta_margin_mean": 35.68285369873047, "beta_dpo/beta_margin_std": 50.96905517578125, "beta_dpo/beta_used": 0.10005475580692291, "beta_dpo/beta_used_raw": -0.9987051486968994, "beta_dpo/gap_mean": 384.96356201171875, "beta_dpo/gap_std": 305.0810546875, "beta_dpo/loss_margin_mean": 348.1621398925781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.671071953010279, "grad_norm": 23.370983123779297, "learning_rate": 1.4901974878202627e-07, "logits/chosen": -2.480140209197998, "logits/rejected": -2.486316680908203, "loss": 0.5419, "step": 457 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4038095474243164, "beta_dpo/beta_margin_grad_std": 0.06412005424499512, "beta_dpo/beta_margin_mean": 0.3973011076450348, "beta_dpo/beta_margin_std": 0.2737996578216553, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.0926374197006226, "beta_dpo/gap_mean": 384.5351257324219, "beta_dpo/gap_std": 302.91986083984375, "beta_dpo/loss_margin_mean": 397.30108642578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6725403817914831, "grad_norm": 28.026363372802734, "learning_rate": 1.4784710168044212e-07, "logits/chosen": -2.47013783454895, "logits/rejected": -2.502309799194336, "loss": 1.0673, "step": 458 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4167326092720032, "beta_dpo/beta_margin_grad_std": 0.06269286572933197, "beta_dpo/beta_margin_mean": 0.3419567346572876, "beta_dpo/beta_margin_std": 0.26270684599876404, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.917838454246521, "beta_dpo/gap_mean": 380.58148193359375, "beta_dpo/gap_std": 296.5938720703125, "beta_dpo/loss_margin_mean": 341.9566955566406, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6740088105726872, "grad_norm": 31.750755310058594, "learning_rate": 1.466771464027316e-07, "logits/chosen": -2.46304988861084, "logits/rejected": -2.501253128051758, "loss": 1.0655, "step": 459 }, { "beta_dpo/beta": 1.6334787607192993, "beta_dpo/beta_margin_grad_mean": -0.22203302383422852, "beta_dpo/beta_margin_grad_std": 0.23113414645195007, "beta_dpo/beta_margin_mean": 714.5479736328125, "beta_dpo/beta_margin_std": 986.8130493164062, "beta_dpo/beta_used": 1.6334787607192993, "beta_dpo/beta_used_raw": -0.6699074506759644, "beta_dpo/gap_mean": 378.25775146484375, "beta_dpo/gap_std": 291.2427978515625, "beta_dpo/loss_margin_mean": 398.70556640625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6754772393538914, "grad_norm": 17868.126953125, "learning_rate": 1.4550991377830423e-07, "logits/chosen": -2.5045523643493652, "logits/rejected": -2.5729987621307373, "loss": 4.6404, "step": 460 }, { "beta_dpo/beta": 0.8932273983955383, "beta_dpo/beta_margin_grad_mean": -0.2543882429599762, "beta_dpo/beta_margin_grad_std": 0.2651003301143646, "beta_dpo/beta_margin_mean": 307.47015380859375, "beta_dpo/beta_margin_std": 430.9241027832031, "beta_dpo/beta_used": 0.8932273983955383, "beta_dpo/beta_used_raw": 0.08712393045425415, "beta_dpo/gap_mean": 376.22406005859375, "beta_dpo/gap_std": 286.36328125, "beta_dpo/loss_margin_mean": 346.59515380859375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6769456681350955, "grad_norm": 32094.228515625, "learning_rate": 1.4434543456482518e-07, "logits/chosen": -2.499390125274658, "logits/rejected": -2.5347375869750977, "loss": 0.7427, "step": 461 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.418192595243454, "beta_dpo/beta_margin_grad_std": 0.07179278135299683, "beta_dpo/beta_margin_mean": 0.3387749493122101, "beta_dpo/beta_margin_std": 0.30487060546875, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.4746923446655273, "beta_dpo/gap_mean": 369.9292907714844, "beta_dpo/gap_std": 287.8791809082031, "beta_dpo/loss_margin_mean": 338.7749328613281, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6784140969162996, "grad_norm": 23.766319274902344, "learning_rate": 1.4318373944740484e-07, "logits/chosen": -2.5215415954589844, "logits/rejected": -2.5495944023132324, "loss": 1.1155, "step": 462 }, { "beta_dpo/beta": 0.6219136714935303, "beta_dpo/beta_margin_grad_mean": -0.20181849598884583, "beta_dpo/beta_margin_grad_std": 0.2130599319934845, "beta_dpo/beta_margin_mean": 281.9193420410156, "beta_dpo/beta_margin_std": 356.6446228027344, "beta_dpo/beta_used": 0.6219136714935303, "beta_dpo/beta_used_raw": -1.7789533138275146, "beta_dpo/gap_mean": 376.6002197265625, "beta_dpo/gap_std": 303.69659423828125, "beta_dpo/loss_margin_mean": 434.07806396484375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6798825256975036, "grad_norm": 18.568052291870117, "learning_rate": 1.4202485903778976e-07, "logits/chosen": -2.4930672645568848, "logits/rejected": -2.5247983932495117, "loss": 0.5632, "step": 463 }, { "beta_dpo/beta": 5.593118667602539, "beta_dpo/beta_margin_grad_mean": -0.078125, "beta_dpo/beta_margin_grad_std": 0.2683681845664978, "beta_dpo/beta_margin_mean": 2755.906005859375, "beta_dpo/beta_margin_std": 1905.5625, "beta_dpo/beta_used": 5.593118667602539, "beta_dpo/beta_used_raw": 5.593118667602539, "beta_dpo/gap_mean": 396.98193359375, "beta_dpo/gap_std": 305.0847473144531, "beta_dpo/loss_margin_mean": 491.942138671875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6813509544787077, "grad_norm": 58260.62890625, "learning_rate": 1.4086882387355658e-07, "logits/chosen": -2.4916696548461914, "logits/rejected": -2.550224781036377, "loss": 9.9301, "step": 464 }, { "beta_dpo/beta": 1.9029580354690552, "beta_dpo/beta_margin_grad_mean": -0.24977104365825653, "beta_dpo/beta_margin_grad_std": 0.2622261047363281, "beta_dpo/beta_margin_mean": 899.8712768554688, "beta_dpo/beta_margin_std": 1295.013916015625, "beta_dpo/beta_used": 1.9029580354690552, "beta_dpo/beta_used_raw": 1.2743040323257446, "beta_dpo/gap_mean": 404.0155029296875, "beta_dpo/gap_std": 305.983154296875, "beta_dpo/loss_margin_mean": 430.5644226074219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6828193832599119, "grad_norm": 30087.87109375, "learning_rate": 1.3971566441730714e-07, "logits/chosen": -2.516110897064209, "logits/rejected": -2.569148540496826, "loss": 0.7493, "step": 465 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.402956485748291, "beta_dpo/beta_margin_grad_std": 0.07183395326137543, "beta_dpo/beta_margin_mean": 0.40361106395721436, "beta_dpo/beta_margin_std": 0.3109658658504486, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.1695146560668945, "beta_dpo/gap_mean": 405.3373107910156, "beta_dpo/gap_std": 308.2467041015625, "beta_dpo/loss_margin_mean": 403.6110534667969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.684287812041116, "grad_norm": 22.92670249938965, "learning_rate": 1.3856541105586545e-07, "logits/chosen": -2.4996213912963867, "logits/rejected": -2.5383553504943848, "loss": 1.0607, "step": 466 }, { "beta_dpo/beta": 4.274341583251953, "beta_dpo/beta_margin_grad_mean": -0.27562493085861206, "beta_dpo/beta_margin_grad_std": 0.2861670255661011, "beta_dpo/beta_margin_mean": 2275.166259765625, "beta_dpo/beta_margin_std": 4076.872802734375, "beta_dpo/beta_used": 4.274341583251953, "beta_dpo/beta_used_raw": 2.634734630584717, "beta_dpo/gap_mean": 413.9611511230469, "beta_dpo/gap_std": 339.21044921875, "beta_dpo/loss_margin_mean": 418.0201110839844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6857562408223201, "grad_norm": 106529.4140625, "learning_rate": 1.3741809409947729e-07, "logits/chosen": -2.5394535064697266, "logits/rejected": -2.5817573070526123, "loss": 2.1242, "step": 467 }, { "beta_dpo/beta": 1.4308552742004395, "beta_dpo/beta_margin_grad_mean": -0.2482784539461136, "beta_dpo/beta_margin_grad_std": 0.2650541663169861, "beta_dpo/beta_margin_mean": 655.9033203125, "beta_dpo/beta_margin_std": 1054.0318603515625, "beta_dpo/beta_used": 1.4308552742004395, "beta_dpo/beta_used_raw": 0.6839801669120789, "beta_dpo/gap_mean": 410.94439697265625, "beta_dpo/gap_std": 351.870361328125, "beta_dpo/loss_margin_mean": 434.7200012207031, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6872246696035242, "grad_norm": 32999.328125, "learning_rate": 1.362737437810114e-07, "logits/chosen": -2.5492682456970215, "logits/rejected": -2.587491750717163, "loss": 27.315, "step": 468 }, { "beta_dpo/beta": 1.8693115711212158, "beta_dpo/beta_margin_grad_mean": -0.046875, "beta_dpo/beta_margin_grad_std": 0.21137107908725739, "beta_dpo/beta_margin_mean": 848.8661499023438, "beta_dpo/beta_margin_std": 818.3424072265625, "beta_dpo/beta_used": 1.8693115711212158, "beta_dpo/beta_used_raw": 1.8693115711212158, "beta_dpo/gap_mean": 419.2552490234375, "beta_dpo/gap_std": 346.93157958984375, "beta_dpo/loss_margin_mean": 452.9244384765625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6886930983847284, "grad_norm": 8297.3203125, "learning_rate": 1.351323902551631e-07, "logits/chosen": -2.5490849018096924, "logits/rejected": -2.5901732444763184, "loss": 0.8058, "step": 469 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.3999681770801544, "beta_dpo/beta_margin_grad_std": 0.07137025147676468, "beta_dpo/beta_margin_mean": 0.4155758321285248, "beta_dpo/beta_margin_std": 0.30809733271598816, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.152496337890625, "beta_dpo/gap_mean": 421.21563720703125, "beta_dpo/gap_std": 342.18878173828125, "beta_dpo/loss_margin_mean": 415.5758361816406, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6901615271659325, "grad_norm": 20.7974853515625, "learning_rate": 1.339940635976592e-07, "logits/chosen": -2.4830265045166016, "logits/rejected": -2.5244758129119873, "loss": 1.0437, "step": 470 }, { "beta_dpo/beta": 1.478458046913147, "beta_dpo/beta_margin_grad_mean": -0.23011885583400726, "beta_dpo/beta_margin_grad_std": 0.23751112818717957, "beta_dpo/beta_margin_mean": 747.5370483398438, "beta_dpo/beta_margin_std": 983.1398315429688, "beta_dpo/beta_used": 1.478458046913147, "beta_dpo/beta_used_raw": -1.9127386808395386, "beta_dpo/gap_mean": 421.38037109375, "beta_dpo/gap_std": 329.10821533203125, "beta_dpo/loss_margin_mean": 398.04986572265625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6916299559471366, "grad_norm": 19.789304733276367, "learning_rate": 1.3285879380446563e-07, "logits/chosen": -2.537778377532959, "logits/rejected": -2.556882858276367, "loss": 0.5604, "step": 471 }, { "beta_dpo/beta": 0.560537576675415, "beta_dpo/beta_margin_grad_mean": -0.2782280147075653, "beta_dpo/beta_margin_grad_std": 0.28962796926498413, "beta_dpo/beta_margin_mean": 254.7558135986328, "beta_dpo/beta_margin_std": 468.81280517578125, "beta_dpo/beta_used": 0.560537576675415, "beta_dpo/beta_used_raw": -2.741248846054077, "beta_dpo/gap_mean": 418.555908203125, "beta_dpo/gap_std": 343.5157165527344, "beta_dpo/loss_margin_mean": 437.1629943847656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6930983847283406, "grad_norm": 11883.9248046875, "learning_rate": 1.317266107909975e-07, "logits/chosen": -2.5896711349487305, "logits/rejected": -2.6116137504577637, "loss": 14.0955, "step": 472 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.42513933777809143, "beta_dpo/beta_margin_grad_std": 0.07835263758897781, "beta_dpo/beta_margin_mean": 0.3119952082633972, "beta_dpo/beta_margin_std": 0.3434739410877228, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.782205581665039, "beta_dpo/gap_mean": 403.55072021484375, "beta_dpo/gap_std": 345.77349853515625, "beta_dpo/loss_margin_mean": 311.9952087402344, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6945668135095447, "grad_norm": 28.633747100830078, "learning_rate": 1.3059754439133002e-07, "logits/chosen": -2.514747381210327, "logits/rejected": -2.5092692375183105, "loss": 1.1064, "step": 473 }, { "beta_dpo/beta": 0.020935971289873123, "beta_dpo/beta_margin_grad_mean": -0.08262869715690613, "beta_dpo/beta_margin_grad_std": 0.23250208795070648, "beta_dpo/beta_margin_mean": 8.29855728149414, "beta_dpo/beta_margin_std": 6.226876258850098, "beta_dpo/beta_used": 0.020935971289873123, "beta_dpo/beta_used_raw": 0.020935971289873123, "beta_dpo/gap_mean": 397.82635498046875, "beta_dpo/gap_std": 337.72735595703125, "beta_dpo/loss_margin_mean": 396.6646728515625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6960352422907489, "grad_norm": 188.52386474609375, "learning_rate": 1.2947162435741277e-07, "logits/chosen": -2.512279987335205, "logits/rejected": -2.5492238998413086, "loss": 0.13, "step": 474 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4154300093650818, "beta_dpo/beta_margin_grad_std": 0.0788368210196495, "beta_dpo/beta_margin_mean": 0.3546939194202423, "beta_dpo/beta_margin_std": 0.35067981481552124, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.2628698348999023, "beta_dpo/gap_mean": 390.2703857421875, "beta_dpo/gap_std": 338.8818359375, "beta_dpo/loss_margin_mean": 354.69390869140625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.697503671071953, "grad_norm": 24.54264259338379, "learning_rate": 1.2834888035828596e-07, "logits/chosen": -2.537545680999756, "logits/rejected": -2.5808987617492676, "loss": 1.0798, "step": 475 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41025692224502563, "beta_dpo/beta_margin_grad_std": 0.0702584832906723, "beta_dpo/beta_margin_mean": 0.37235116958618164, "beta_dpo/beta_margin_std": 0.3026776611804962, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.833922863006592, "beta_dpo/gap_mean": 388.89324951171875, "beta_dpo/gap_std": 336.2122497558594, "beta_dpo/loss_margin_mean": 372.3511657714844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.6989720998531571, "grad_norm": 33.014305114746094, "learning_rate": 1.2722934197929802e-07, "logits/chosen": -2.5620553493499756, "logits/rejected": -2.5980238914489746, "loss": 1.0909, "step": 476 }, { "beta_dpo/beta": 1.2425551414489746, "beta_dpo/beta_margin_grad_mean": -0.28471827507019043, "beta_dpo/beta_margin_grad_std": 0.29025498032569885, "beta_dpo/beta_margin_mean": 494.6365661621094, "beta_dpo/beta_margin_std": 732.710693359375, "beta_dpo/beta_used": 1.2425551414489746, "beta_dpo/beta_used_raw": 0.43174052238464355, "beta_dpo/gap_mean": 384.2132568359375, "beta_dpo/gap_std": 325.43865966796875, "beta_dpo/loss_margin_mean": 377.49932861328125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7004405286343612, "grad_norm": 9603.7666015625, "learning_rate": 1.2611303872132631e-07, "logits/chosen": -2.513644218444824, "logits/rejected": -2.5187504291534424, "loss": 4.2004, "step": 477 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4097580313682556, "beta_dpo/beta_margin_grad_std": 0.06587263941764832, "beta_dpo/beta_margin_mean": 0.3733612298965454, "beta_dpo/beta_margin_std": 0.28482842445373535, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.8485658168792725, "beta_dpo/gap_mean": 381.9994812011719, "beta_dpo/gap_std": 319.3729248046875, "beta_dpo/loss_margin_mean": 373.3612060546875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7019089574155654, "grad_norm": 30.411699295043945, "learning_rate": 1.2500000000000005e-07, "logits/chosen": -2.530264139175415, "logits/rejected": -2.573782205581665, "loss": 1.094, "step": 478 }, { "beta_dpo/beta": 0.9422978758811951, "beta_dpo/beta_margin_grad_mean": -0.2220228761434555, "beta_dpo/beta_margin_grad_std": 0.23495404422283173, "beta_dpo/beta_margin_mean": 474.04681396484375, "beta_dpo/beta_margin_std": 649.1632080078125, "beta_dpo/beta_used": 0.9422978758811951, "beta_dpo/beta_used_raw": -1.9218789339065552, "beta_dpo/gap_mean": 387.78704833984375, "beta_dpo/gap_std": 326.0636291503906, "beta_dpo/loss_margin_mean": 436.54107666015625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7033773861967695, "grad_norm": 22.1717472076416, "learning_rate": 1.2389025514492456e-07, "logits/chosen": -2.5045852661132812, "logits/rejected": -2.568833112716675, "loss": 0.5685, "step": 479 }, { "beta_dpo/beta": 1.2717182636260986, "beta_dpo/beta_margin_grad_mean": -0.2628205120563507, "beta_dpo/beta_margin_grad_std": 0.26975834369659424, "beta_dpo/beta_margin_mean": 558.4786376953125, "beta_dpo/beta_margin_std": 861.1541748046875, "beta_dpo/beta_used": 1.2717182636260986, "beta_dpo/beta_used_raw": -0.5218298435211182, "beta_dpo/gap_mean": 385.14337158203125, "beta_dpo/gap_std": 321.2519836425781, "beta_dpo/loss_margin_mean": 358.7318420410156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7048458149779736, "grad_norm": 34886.80859375, "learning_rate": 1.227838333989088e-07, "logits/chosen": -2.5319626331329346, "logits/rejected": -2.5537118911743164, "loss": 41.8083, "step": 480 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40268126130104065, "beta_dpo/beta_margin_grad_std": 0.08109249919652939, "beta_dpo/beta_margin_mean": 0.4087076485157013, "beta_dpo/beta_margin_std": 0.3610389828681946, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.2213141918182373, "beta_dpo/gap_mean": 391.42706298828125, "beta_dpo/gap_std": 332.341552734375, "beta_dpo/loss_margin_mean": 408.7076416015625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7063142437591777, "grad_norm": 25.161544799804688, "learning_rate": 1.2168076391719489e-07, "logits/chosen": -2.5190436840057373, "logits/rejected": -2.566678285598755, "loss": 1.0755, "step": 481 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4173520803451538, "beta_dpo/beta_margin_grad_std": 0.07615692913532257, "beta_dpo/beta_margin_mean": 0.34368330240249634, "beta_dpo/beta_margin_std": 0.32845625281333923, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.8063464164733887, "beta_dpo/gap_mean": 385.44921875, "beta_dpo/gap_std": 331.2301025390625, "beta_dpo/loss_margin_mean": 343.68328857421875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7077826725403817, "grad_norm": 26.447629928588867, "learning_rate": 1.2058107576668938e-07, "logits/chosen": -2.5346310138702393, "logits/rejected": -2.5641980171203613, "loss": 1.0925, "step": 482 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.39801958203315735, "beta_dpo/beta_margin_grad_std": 0.08198197931051254, "beta_dpo/beta_margin_mean": 0.4265040457248688, "beta_dpo/beta_margin_std": 0.35432857275009155, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.8292474746704102, "beta_dpo/gap_mean": 391.0637512207031, "beta_dpo/gap_std": 335.705322265625, "beta_dpo/loss_margin_mean": 426.5040283203125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7092511013215859, "grad_norm": 27.4080810546875, "learning_rate": 1.194847979251979e-07, "logits/chosen": -2.523871898651123, "logits/rejected": -2.5619630813598633, "loss": 1.0621, "step": 483 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4196456968784332, "beta_dpo/beta_margin_grad_std": 0.08170211315155029, "beta_dpo/beta_margin_mean": 0.33604294061660767, "beta_dpo/beta_margin_std": 0.3569084405899048, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.1099014282226562, "beta_dpo/gap_mean": 378.6564636230469, "beta_dpo/gap_std": 334.283935546875, "beta_dpo/loss_margin_mean": 336.04290771484375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.71071953010279, "grad_norm": 27.6585636138916, "learning_rate": 1.1839195928066101e-07, "logits/chosen": -2.5312819480895996, "logits/rejected": -2.561755657196045, "loss": 1.1056, "step": 484 }, { "beta_dpo/beta": 1.6319665908813477, "beta_dpo/beta_margin_grad_mean": -0.12443296611309052, "beta_dpo/beta_margin_grad_std": 0.3292010724544525, "beta_dpo/beta_margin_mean": 740.8789672851562, "beta_dpo/beta_margin_std": 888.0823364257812, "beta_dpo/beta_used": 1.6319665908813477, "beta_dpo/beta_used_raw": 1.6319665908813477, "beta_dpo/gap_mean": 391.543212890625, "beta_dpo/gap_std": 351.833740234375, "beta_dpo/loss_margin_mean": 453.990478515625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7121879588839941, "grad_norm": 43631.0625, "learning_rate": 1.1730258863039347e-07, "logits/chosen": -2.533259153366089, "logits/rejected": -2.598179340362549, "loss": 48.0713, "step": 485 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40132319927215576, "beta_dpo/beta_margin_grad_std": 0.08668551594018936, "beta_dpo/beta_margin_mean": 0.41604912281036377, "beta_dpo/beta_margin_std": 0.3828808665275574, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.8486566543579102, "beta_dpo/gap_mean": 397.93218994140625, "beta_dpo/gap_std": 359.19091796875, "beta_dpo/loss_margin_mean": 416.0491027832031, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7136563876651982, "grad_norm": 32.61040115356445, "learning_rate": 1.1621671468032493e-07, "logits/chosen": -2.5248947143554688, "logits/rejected": -2.5619006156921387, "loss": 1.0617, "step": 486 }, { "beta_dpo/beta": 3.366377353668213, "beta_dpo/beta_margin_grad_mean": -0.078125, "beta_dpo/beta_margin_grad_std": 0.2683681845664978, "beta_dpo/beta_margin_mean": 1427.7965087890625, "beta_dpo/beta_margin_std": 970.3147583007812, "beta_dpo/beta_used": 3.366377353668213, "beta_dpo/beta_used_raw": 3.366377353668213, "beta_dpo/gap_mean": 402.7051696777344, "beta_dpo/gap_std": 350.80084228515625, "beta_dpo/loss_margin_mean": 424.222900390625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7151248164464024, "grad_norm": 0.0, "learning_rate": 1.1513436604424378e-07, "logits/chosen": -2.5087289810180664, "logits/rejected": -2.566622257232666, "loss": 0.0, "step": 487 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41341128945350647, "beta_dpo/beta_margin_grad_std": 0.07426728308200836, "beta_dpo/beta_margin_mean": 0.3592289388179779, "beta_dpo/beta_margin_std": 0.314900279045105, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.6400840282440186, "beta_dpo/gap_mean": 395.1749267578125, "beta_dpo/gap_std": 344.27288818359375, "beta_dpo/loss_margin_mean": 359.2289123535156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7165932452276065, "grad_norm": 28.298213958740234, "learning_rate": 1.1405557124304335e-07, "logits/chosen": -2.5287461280822754, "logits/rejected": -2.5617752075195312, "loss": 1.1019, "step": 488 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41154804825782776, "beta_dpo/beta_margin_grad_std": 0.06765101850032806, "beta_dpo/beta_margin_mean": 0.3656145930290222, "beta_dpo/beta_margin_std": 0.2879979908466339, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.769493579864502, "beta_dpo/gap_mean": 391.515625, "beta_dpo/gap_std": 335.818359375, "beta_dpo/loss_margin_mean": 365.6145935058594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7180616740088106, "grad_norm": 24.00615692138672, "learning_rate": 1.1298035870396985e-07, "logits/chosen": -2.5616259574890137, "logits/rejected": -2.594806432723999, "loss": 1.1112, "step": 489 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4056692123413086, "beta_dpo/beta_margin_grad_std": 0.10516832768917084, "beta_dpo/beta_margin_mean": 0.40864163637161255, "beta_dpo/beta_margin_std": 0.48387810587882996, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -5.209046363830566, "beta_dpo/gap_mean": 392.92205810546875, "beta_dpo/gap_std": 358.7939453125, "beta_dpo/loss_margin_mean": 408.6416015625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7195301027900147, "grad_norm": 26.91084098815918, "learning_rate": 1.1190875675987355e-07, "logits/chosen": -2.6055257320404053, "logits/rejected": -2.6725831031799316, "loss": 1.1313, "step": 490 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41814425587654114, "beta_dpo/beta_margin_grad_std": 0.07073578983545303, "beta_dpo/beta_margin_mean": 0.3384140133857727, "beta_dpo/beta_margin_std": 0.2998770475387573, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.0293617248535156, "beta_dpo/gap_mean": 385.5333251953125, "beta_dpo/gap_std": 351.7672119140625, "beta_dpo/loss_margin_mean": 338.41400146484375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7209985315712188, "grad_norm": 27.26293182373047, "learning_rate": 1.1084079364846241e-07, "logits/chosen": -2.5373287200927734, "logits/rejected": -2.5686709880828857, "loss": 1.0702, "step": 491 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.420479416847229, "beta_dpo/beta_margin_grad_std": 0.072646863758564, "beta_dpo/beta_margin_mean": 0.3299318253993988, "beta_dpo/beta_margin_std": 0.31079521775245667, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.52386474609375, "beta_dpo/gap_mean": 376.2781677246094, "beta_dpo/gap_std": 346.640380859375, "beta_dpo/loss_margin_mean": 329.93182373046875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7224669603524229, "grad_norm": 26.98293685913086, "learning_rate": 1.097764975115576e-07, "logits/chosen": -2.5452327728271484, "logits/rejected": -2.5773203372955322, "loss": 1.1255, "step": 492 }, { "beta_dpo/beta": 0.19389446079730988, "beta_dpo/beta_margin_grad_mean": -0.30418387055397034, "beta_dpo/beta_margin_grad_std": 0.3022761046886444, "beta_dpo/beta_margin_mean": 63.774085998535156, "beta_dpo/beta_margin_std": 104.5694351196289, "beta_dpo/beta_used": 0.19389446079730988, "beta_dpo/beta_used_raw": -0.4551483392715454, "beta_dpo/gap_mean": 366.77899169921875, "beta_dpo/gap_std": 340.6854553222656, "beta_dpo/loss_margin_mean": 331.5319519042969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.723935389133627, "grad_norm": 4613.84375, "learning_rate": 1.0871589639435203e-07, "logits/chosen": -2.5896708965301514, "logits/rejected": -2.6086065769195557, "loss": 3.1786, "step": 493 }, { "beta_dpo/beta": 3.5493483543395996, "beta_dpo/beta_margin_grad_mean": -0.09375, "beta_dpo/beta_margin_grad_std": 0.2914806008338928, "beta_dpo/beta_margin_mean": 1450.2022705078125, "beta_dpo/beta_margin_std": 1306.51318359375, "beta_dpo/beta_used": 3.5493483543395996, "beta_dpo/beta_used_raw": 3.5493483543395996, "beta_dpo/gap_mean": 373.76806640625, "beta_dpo/gap_std": 341.1934814453125, "beta_dpo/loss_margin_mean": 413.81048583984375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7254038179148311, "grad_norm": 1680.6007080078125, "learning_rate": 1.0765901824467166e-07, "logits/chosen": -2.5582916736602783, "logits/rejected": -2.609689474105835, "loss": 0.9905, "step": 494 }, { "beta_dpo/beta": 4.066619873046875, "beta_dpo/beta_margin_grad_mean": -0.21904002130031586, "beta_dpo/beta_margin_grad_std": 0.23028062283992767, "beta_dpo/beta_margin_mean": 1956.0379638671875, "beta_dpo/beta_margin_std": 2361.65234375, "beta_dpo/beta_used": 4.066619873046875, "beta_dpo/beta_used_raw": 3.1382861137390137, "beta_dpo/gap_mean": 380.26708984375, "beta_dpo/gap_std": 335.47235107421875, "beta_dpo/loss_margin_mean": 433.75335693359375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7268722466960352, "grad_norm": 22.30237579345703, "learning_rate": 1.0660589091223854e-07, "logits/chosen": -2.5691702365875244, "logits/rejected": -2.6163458824157715, "loss": 0.5429, "step": 495 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.409112811088562, "beta_dpo/beta_margin_grad_std": 0.06478798389434814, "beta_dpo/beta_margin_mean": 0.37426888942718506, "beta_dpo/beta_margin_std": 0.27219101786613464, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.4662165641784668, "beta_dpo/gap_mean": 383.7269287109375, "beta_dpo/gap_std": 323.4496765136719, "beta_dpo/loss_margin_mean": 374.26885986328125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7283406754772394, "grad_norm": 27.432092666625977, "learning_rate": 1.0555654214793722e-07, "logits/chosen": -2.5857880115509033, "logits/rejected": -2.6112606525421143, "loss": 1.0738, "step": 496 }, { "beta_dpo/beta": 0.4332939088344574, "beta_dpo/beta_margin_grad_mean": -0.2696186900138855, "beta_dpo/beta_margin_grad_std": 0.27868661284446716, "beta_dpo/beta_margin_mean": 175.2734375, "beta_dpo/beta_margin_std": 240.9627685546875, "beta_dpo/beta_used": 0.4332939088344574, "beta_dpo/beta_used_raw": -1.5341229438781738, "beta_dpo/gap_mean": 384.20770263671875, "beta_dpo/gap_std": 314.7350158691406, "beta_dpo/loss_margin_mean": 381.3740234375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7298091042584435, "grad_norm": 12841.9072265625, "learning_rate": 1.0451099960308374e-07, "logits/chosen": -2.5560591220855713, "logits/rejected": -2.60593318939209, "loss": 3.4472, "step": 497 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41190412640571594, "beta_dpo/beta_margin_grad_std": 0.08984668552875519, "beta_dpo/beta_margin_mean": 0.37027707695961, "beta_dpo/beta_margin_std": 0.39214304089546204, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.132285237312317, "beta_dpo/gap_mean": 380.50390625, "beta_dpo/gap_std": 325.7791442871094, "beta_dpo/loss_margin_mean": 370.2770690917969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7312775330396476, "grad_norm": 28.40172004699707, "learning_rate": 1.0346929082869641e-07, "logits/chosen": -2.5905566215515137, "logits/rejected": -2.6150288581848145, "loss": 1.0755, "step": 498 }, { "beta_dpo/beta": 2.1377925872802734, "beta_dpo/beta_margin_grad_mean": -0.25142624974250793, "beta_dpo/beta_margin_grad_std": 0.2667090594768524, "beta_dpo/beta_margin_mean": 950.3165283203125, "beta_dpo/beta_margin_std": 1399.9261474609375, "beta_dpo/beta_used": 2.1377925872802734, "beta_dpo/beta_used_raw": -0.2245168685913086, "beta_dpo/gap_mean": 386.89794921875, "beta_dpo/gap_std": 335.3507080078125, "beta_dpo/loss_margin_mean": 415.10186767578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7327459618208517, "grad_norm": 27508.541015625, "learning_rate": 1.0243144327477013e-07, "logits/chosen": -2.5723462104797363, "logits/rejected": -2.636953830718994, "loss": 25.3095, "step": 499 }, { "beta_dpo/beta": 0.8406103849411011, "beta_dpo/beta_margin_grad_mean": -0.21563133597373962, "beta_dpo/beta_margin_grad_std": 0.27382031083106995, "beta_dpo/beta_margin_mean": 361.0441589355469, "beta_dpo/beta_margin_std": 578.6198120117188, "beta_dpo/beta_used": 0.8406103849411011, "beta_dpo/beta_used_raw": 0.8406103849411011, "beta_dpo/gap_mean": 390.808837890625, "beta_dpo/gap_std": 344.6997375488281, "beta_dpo/loss_margin_mean": 407.3626403808594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7342143906020558, "grad_norm": 12742.4658203125, "learning_rate": 1.0139748428955333e-07, "logits/chosen": -2.5479164123535156, "logits/rejected": -2.6190683841705322, "loss": 6.7444, "step": 500 }, { "epoch": 0.7342143906020558, "eval_beta_dpo/beta": 0.0010000000474974513, "eval_beta_dpo/beta_margin_grad_mean": -0.4440717399120331, "eval_beta_dpo/beta_margin_grad_std": 0.09194090217351913, "eval_beta_dpo/beta_margin_mean": 0.23392271995544434, "eval_beta_dpo/beta_margin_std": 0.3909527361392975, "eval_beta_dpo/beta_used": 0.0010000000474974513, "eval_beta_dpo/beta_used_raw": -9.295312881469727, "eval_beta_dpo/gap_mean": 390.51129150390625, "eval_beta_dpo/gap_std": 345.8570556640625, "eval_beta_dpo/loss_margin_mean": 233.9227294921875, "eval_beta_dpo/mask_keep_frac": 1.0, "eval_logits/chosen": -2.6421258449554443, "eval_logits/rejected": -2.6563594341278076, "eval_loss": 0.602572500705719, "eval_runtime": 37.7828, "eval_samples_per_second": 61.907, "eval_steps_per_second": 1.959, "step": 500 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40555235743522644, "beta_dpo/beta_margin_grad_std": 0.0859164372086525, "beta_dpo/beta_margin_mean": 0.39426204562187195, "beta_dpo/beta_margin_std": 0.3646033704280853, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.6879664659500122, "beta_dpo/gap_mean": 391.73577880859375, "beta_dpo/gap_std": 350.0826416015625, "beta_dpo/loss_margin_mean": 394.26202392578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.73568281938326, "grad_norm": 30.496671676635742, "learning_rate": 1.0036744111882672e-07, "logits/chosen": -2.519293785095215, "logits/rejected": -2.5676636695861816, "loss": 1.0742, "step": 501 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4322647154331207, "beta_dpo/beta_margin_grad_std": 0.07655175775289536, "beta_dpo/beta_margin_mean": 0.27862733602523804, "beta_dpo/beta_margin_std": 0.3172290623188019, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.7983059883117676, "beta_dpo/gap_mean": 375.173095703125, "beta_dpo/gap_std": 344.82647705078125, "beta_dpo/loss_margin_mean": 278.6273193359375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.737151248164464, "grad_norm": 28.89243507385254, "learning_rate": 9.934134090518592e-08, "logits/chosen": -2.5925819873809814, "logits/rejected": -2.604139804840088, "loss": 1.1023, "step": 502 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4080732762813568, "beta_dpo/beta_margin_grad_std": 0.07349276542663574, "beta_dpo/beta_margin_mean": 0.3811602294445038, "beta_dpo/beta_margin_std": 0.31262895464897156, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.2302193641662598, "beta_dpo/gap_mean": 371.81048583984375, "beta_dpo/gap_std": 340.92041015625, "beta_dpo/loss_margin_mean": 381.16021728515625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7386196769456681, "grad_norm": 29.317602157592773, "learning_rate": 9.831921068732571e-08, "logits/chosen": -2.562480926513672, "logits/rejected": -2.6049766540527344, "loss": 1.0943, "step": 503 }, { "beta_dpo/beta": 3.8556745052337646, "beta_dpo/beta_margin_grad_mean": -0.07770384848117828, "beta_dpo/beta_margin_grad_std": 0.2669384777545929, "beta_dpo/beta_margin_mean": 1844.1893310546875, "beta_dpo/beta_margin_std": 1516.7772216796875, "beta_dpo/beta_used": 3.8556745052337646, "beta_dpo/beta_used_raw": 3.8556745052337646, "beta_dpo/gap_mean": 387.630126953125, "beta_dpo/gap_std": 345.894775390625, "beta_dpo/loss_margin_mean": 475.9586181640625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7400881057268722, "grad_norm": 30736.17578125, "learning_rate": 9.730107739932805e-08, "logits/chosen": -2.5559229850769043, "logits/rejected": -2.630270481109619, "loss": 0.1445, "step": 504 }, { "beta_dpo/beta": 0.16585735976696014, "beta_dpo/beta_margin_grad_mean": -0.2709311544895172, "beta_dpo/beta_margin_grad_std": 0.27986839413642883, "beta_dpo/beta_margin_mean": 50.6553840637207, "beta_dpo/beta_margin_std": 122.78081512451172, "beta_dpo/beta_used": 0.16585735976696014, "beta_dpo/beta_used_raw": 0.008618459105491638, "beta_dpo/gap_mean": 381.24481201171875, "beta_dpo/gap_std": 361.23394775390625, "beta_dpo/loss_margin_mean": 317.9100036621094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7415565345080763, "grad_norm": 4374.5302734375, "learning_rate": 9.628696786995188e-08, "logits/chosen": -2.6019692420959473, "logits/rejected": -2.6386215686798096, "loss": 5.8766, "step": 505 }, { "beta_dpo/beta": 2.0099172592163086, "beta_dpo/beta_margin_grad_mean": -0.267021119594574, "beta_dpo/beta_margin_grad_std": 0.27634257078170776, "beta_dpo/beta_margin_mean": 805.3701171875, "beta_dpo/beta_margin_std": 1344.735107421875, "beta_dpo/beta_used": 2.0099172592163086, "beta_dpo/beta_used_raw": 1.2173224687576294, "beta_dpo/gap_mean": 379.2687072753906, "beta_dpo/gap_std": 362.82806396484375, "beta_dpo/loss_margin_mean": 386.8839416503906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7430249632892805, "grad_norm": 4789.49853515625, "learning_rate": 9.527690882192635e-08, "logits/chosen": -2.577932357788086, "logits/rejected": -2.627098560333252, "loss": 5.6618, "step": 506 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4179416000843048, "beta_dpo/beta_margin_grad_std": 0.08015478402376175, "beta_dpo/beta_margin_mean": 0.34152480959892273, "beta_dpo/beta_margin_std": 0.33954671025276184, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.6754093170166016, "beta_dpo/gap_mean": 374.45867919921875, "beta_dpo/gap_std": 354.2435607910156, "beta_dpo/loss_margin_mean": 341.5248107910156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7444933920704846, "grad_norm": 29.211889266967773, "learning_rate": 9.427092687124691e-08, "logits/chosen": -2.6004366874694824, "logits/rejected": -2.6304922103881836, "loss": 1.1038, "step": 507 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4112531542778015, "beta_dpo/beta_margin_grad_std": 0.09117571264505386, "beta_dpo/beta_margin_mean": 0.3736902177333832, "beta_dpo/beta_margin_std": 0.3952173888683319, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.8720993995666504, "beta_dpo/gap_mean": 369.85711669921875, "beta_dpo/gap_std": 362.80743408203125, "beta_dpo/loss_margin_mean": 373.6902160644531, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7459618208516887, "grad_norm": 28.369312286376953, "learning_rate": 9.326904852647344e-08, "logits/chosen": -2.6036999225616455, "logits/rejected": -2.642367362976074, "loss": 1.1269, "step": 508 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41093164682388306, "beta_dpo/beta_margin_grad_std": 0.07275278866291046, "beta_dpo/beta_margin_mean": 0.36885690689086914, "beta_dpo/beta_margin_std": 0.30931705236434937, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.5401596426963806, "beta_dpo/gap_mean": 372.418212890625, "beta_dpo/gap_std": 355.0517272949219, "beta_dpo/loss_margin_mean": 368.8569030761719, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7474302496328928, "grad_norm": 26.87523651123047, "learning_rate": 9.227130018803195e-08, "logits/chosen": -2.550929307937622, "logits/rejected": -2.5829358100891113, "loss": 1.0726, "step": 509 }, { "beta_dpo/beta": 1.996654748916626, "beta_dpo/beta_margin_grad_mean": -0.21876604855060577, "beta_dpo/beta_margin_grad_std": 0.2280726283788681, "beta_dpo/beta_margin_mean": 924.012939453125, "beta_dpo/beta_margin_std": 1230.9832763671875, "beta_dpo/beta_used": 1.996654748916626, "beta_dpo/beta_used_raw": 0.9159380197525024, "beta_dpo/gap_mean": 381.06939697265625, "beta_dpo/gap_std": 340.1996154785156, "beta_dpo/loss_margin_mean": 424.53790283203125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.748898678414097, "grad_norm": 15.459576606750488, "learning_rate": 9.127770814751932e-08, "logits/chosen": -2.52266788482666, "logits/rejected": -2.6167397499084473, "loss": 0.5411, "step": 510 }, { "beta_dpo/beta": 0.6738762259483337, "beta_dpo/beta_margin_grad_mean": -0.26766470074653625, "beta_dpo/beta_margin_grad_std": 0.2765524685382843, "beta_dpo/beta_margin_mean": 276.3068542480469, "beta_dpo/beta_margin_std": 440.9131164550781, "beta_dpo/beta_used": 0.6738762259483337, "beta_dpo/beta_used_raw": -0.26501893997192383, "beta_dpo/gap_mean": 381.2438659667969, "beta_dpo/gap_std": 331.12408447265625, "beta_dpo/loss_margin_mean": 387.6205139160156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.750367107195301, "grad_norm": 6812.96630859375, "learning_rate": 9.028829858700973e-08, "logits/chosen": -2.5795559883117676, "logits/rejected": -2.6349778175354004, "loss": 0.6808, "step": 511 }, { "beta_dpo/beta": 0.29571211338043213, "beta_dpo/beta_margin_grad_mean": -0.21608670055866241, "beta_dpo/beta_margin_grad_std": 0.2315717190504074, "beta_dpo/beta_margin_mean": 118.46554565429688, "beta_dpo/beta_margin_std": 172.0571746826172, "beta_dpo/beta_used": 0.29571211338043213, "beta_dpo/beta_used_raw": -0.0050897300243377686, "beta_dpo/gap_mean": 386.8934020996094, "beta_dpo/gap_std": 338.73162841796875, "beta_dpo/loss_margin_mean": 408.7341003417969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7518355359765051, "grad_norm": 8849.5224609375, "learning_rate": 8.930309757836516e-08, "logits/chosen": -2.5242481231689453, "logits/rejected": -2.5922141075134277, "loss": 4.6625, "step": 512 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41406044363975525, "beta_dpo/beta_margin_grad_std": 0.07585693150758743, "beta_dpo/beta_margin_mean": 0.35993170738220215, "beta_dpo/beta_margin_std": 0.3399631977081299, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.504025936126709, "beta_dpo/gap_mean": 387.1397705078125, "beta_dpo/gap_std": 338.95806884765625, "beta_dpo/loss_margin_mean": 359.9316711425781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7533039647577092, "grad_norm": 31.581235885620117, "learning_rate": 8.832213108254863e-08, "logits/chosen": -2.545290231704712, "logits/rejected": -2.597968578338623, "loss": 1.0993, "step": 513 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4150505065917969, "beta_dpo/beta_margin_grad_std": 0.07808467000722885, "beta_dpo/beta_margin_mean": 0.3535368740558624, "beta_dpo/beta_margin_std": 0.3357268273830414, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.028075695037842, "beta_dpo/gap_mean": 375.724365234375, "beta_dpo/gap_std": 334.5943908691406, "beta_dpo/loss_margin_mean": 353.536865234375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7547723935389133, "grad_norm": 32.19744873046875, "learning_rate": 8.734542494893954e-08, "logits/chosen": -2.555717706680298, "logits/rejected": -2.603783369064331, "loss": 1.0886, "step": 514 }, { "beta_dpo/beta": 3.737210988998413, "beta_dpo/beta_margin_grad_mean": -0.14027747511863708, "beta_dpo/beta_margin_grad_std": 0.3467850983142853, "beta_dpo/beta_margin_mean": 1422.5458984375, "beta_dpo/beta_margin_std": 1979.079345703125, "beta_dpo/beta_used": 3.737210988998413, "beta_dpo/beta_used_raw": 3.737210988998413, "beta_dpo/gap_mean": 374.7799987792969, "beta_dpo/gap_std": 340.4820556640625, "beta_dpo/loss_margin_mean": 349.0910949707031, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7562408223201175, "grad_norm": 7053.50537109375, "learning_rate": 8.637300491465272e-08, "logits/chosen": -2.5503506660461426, "logits/rejected": -2.6203575134277344, "loss": 3.1348, "step": 515 }, { "beta_dpo/beta": 0.4134178161621094, "beta_dpo/beta_margin_grad_mean": -0.279133677482605, "beta_dpo/beta_margin_grad_std": 0.2908896803855896, "beta_dpo/beta_margin_mean": 205.34104919433594, "beta_dpo/beta_margin_std": 399.5401611328125, "beta_dpo/beta_used": 0.4134178161621094, "beta_dpo/beta_used_raw": -0.5360516309738159, "beta_dpo/gap_mean": 384.8233642578125, "beta_dpo/gap_std": 366.90264892578125, "beta_dpo/loss_margin_mean": 453.2371520996094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7577092511013216, "grad_norm": 11149.91796875, "learning_rate": 8.540489660386064e-08, "logits/chosen": -2.595862627029419, "logits/rejected": -2.688004493713379, "loss": 8.3798, "step": 516 }, { "beta_dpo/beta": 1.5258046388626099, "beta_dpo/beta_margin_grad_mean": -0.2544478476047516, "beta_dpo/beta_margin_grad_std": 0.26925018429756165, "beta_dpo/beta_margin_mean": 720.7048950195312, "beta_dpo/beta_margin_std": 1094.427001953125, "beta_dpo/beta_used": 1.5258046388626099, "beta_dpo/beta_used_raw": 1.4570624828338623, "beta_dpo/gap_mean": 392.5709533691406, "beta_dpo/gap_std": 370.2559509277344, "beta_dpo/loss_margin_mean": 411.7127685546875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7591776798825257, "grad_norm": 21861.8203125, "learning_rate": 8.444112552711752e-08, "logits/chosen": -2.5828640460968018, "logits/rejected": -2.6403748989105225, "loss": 18.2512, "step": 517 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4097965955734253, "beta_dpo/beta_margin_grad_std": 0.0833662748336792, "beta_dpo/beta_margin_mean": 0.3765363395214081, "beta_dpo/beta_margin_std": 0.3559281527996063, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.791685104370117, "beta_dpo/gap_mean": 390.30194091796875, "beta_dpo/gap_std": 370.6991882324219, "beta_dpo/loss_margin_mean": 376.53631591796875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7606461086637298, "grad_norm": 35.12870788574219, "learning_rate": 8.348171708068747e-08, "logits/chosen": -2.565255880355835, "logits/rejected": -2.629263162612915, "loss": 1.0891, "step": 518 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4017196595668793, "beta_dpo/beta_margin_grad_std": 0.0956096351146698, "beta_dpo/beta_margin_mean": 0.4176686108112335, "beta_dpo/beta_margin_std": 0.4266796112060547, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.772631883621216, "beta_dpo/gap_mean": 392.6380615234375, "beta_dpo/gap_std": 373.5154113769531, "beta_dpo/loss_margin_mean": 417.6685791015625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.762114537444934, "grad_norm": 29.76219367980957, "learning_rate": 8.25266965458755e-08, "logits/chosen": -2.607625961303711, "logits/rejected": -2.6605143547058105, "loss": 1.0877, "step": 519 }, { "beta_dpo/beta": 1.1038464307785034, "beta_dpo/beta_margin_grad_mean": -0.07864061743021011, "beta_dpo/beta_margin_grad_std": 0.26824915409088135, "beta_dpo/beta_margin_mean": 507.0424499511719, "beta_dpo/beta_margin_std": 722.092041015625, "beta_dpo/beta_used": 1.1038464307785034, "beta_dpo/beta_used_raw": 1.1038464307785034, "beta_dpo/gap_mean": 399.25439453125, "beta_dpo/gap_std": 378.07611083984375, "beta_dpo/loss_margin_mean": 422.9557189941406, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7635829662261381, "grad_norm": 16670.017578125, "learning_rate": 8.15760890883607e-08, "logits/chosen": -2.5519027709960938, "logits/rejected": -2.6170506477355957, "loss": 6.8461, "step": 520 }, { "beta_dpo/beta": 1.2119355201721191, "beta_dpo/beta_margin_grad_mean": -0.1504676640033722, "beta_dpo/beta_margin_grad_std": 0.35239845514297485, "beta_dpo/beta_margin_mean": 429.1769104003906, "beta_dpo/beta_margin_std": 439.9748229980469, "beta_dpo/beta_used": 1.2119355201721191, "beta_dpo/beta_used_raw": 1.2119355201721191, "beta_dpo/gap_mean": 391.2038879394531, "beta_dpo/gap_std": 371.0228271484375, "beta_dpo/loss_margin_mean": 357.31500244140625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7650513950073421, "grad_norm": 20681.9140625, "learning_rate": 8.062991975753378e-08, "logits/chosen": -2.595400333404541, "logits/rejected": -2.645524024963379, "loss": 13.2254, "step": 521 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.42150411009788513, "beta_dpo/beta_margin_grad_std": 0.09007870405912399, "beta_dpo/beta_margin_mean": 0.3268641233444214, "beta_dpo/beta_margin_std": 0.3883510231971741, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.027829647064209, "beta_dpo/gap_mean": 379.1689147949219, "beta_dpo/gap_std": 368.09771728515625, "beta_dpo/loss_margin_mean": 326.8641052246094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7665198237885462, "grad_norm": 25.660398483276367, "learning_rate": 7.968821348583643e-08, "logits/chosen": -2.570929527282715, "logits/rejected": -2.6126561164855957, "loss": 1.1038, "step": 522 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4145914912223816, "beta_dpo/beta_margin_grad_std": 0.08707987517118454, "beta_dpo/beta_margin_mean": 0.3594570457935333, "beta_dpo/beta_margin_std": 0.3804139494895935, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.161890029907227, "beta_dpo/gap_mean": 377.38458251953125, "beta_dpo/gap_std": 372.1881103515625, "beta_dpo/loss_margin_mean": 359.45703125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7679882525697503, "grad_norm": 29.40875244140625, "learning_rate": 7.875099508810484e-08, "logits/chosen": -2.5962986946105957, "logits/rejected": -2.658698081970215, "loss": 1.1252, "step": 523 }, { "beta_dpo/beta": 2.29657244682312, "beta_dpo/beta_margin_grad_mean": -0.24990545213222504, "beta_dpo/beta_margin_grad_std": 0.2622232139110565, "beta_dpo/beta_margin_mean": 990.7221069335938, "beta_dpo/beta_margin_std": 1473.548583984375, "beta_dpo/beta_used": 2.29657244682312, "beta_dpo/beta_used_raw": 1.6613166332244873, "beta_dpo/gap_mean": 380.9888916015625, "beta_dpo/gap_std": 365.5908203125, "beta_dpo/loss_margin_mean": 409.36083984375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7694566813509545, "grad_norm": 44968.94140625, "learning_rate": 7.781828926091535e-08, "logits/chosen": -2.635100841522217, "logits/rejected": -2.6668195724487305, "loss": 25.215, "step": 524 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.3882511854171753, "beta_dpo/beta_margin_grad_std": 0.08840624243021011, "beta_dpo/beta_margin_mean": 0.4734336733818054, "beta_dpo/beta_margin_std": 0.39495328068733215, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.465731620788574, "beta_dpo/gap_mean": 395.54022216796875, "beta_dpo/gap_std": 364.6204833984375, "beta_dpo/loss_margin_mean": 473.43365478515625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7709251101321586, "grad_norm": 29.38068199157715, "learning_rate": 7.689012058193384e-08, "logits/chosen": -2.5895333290100098, "logits/rejected": -2.678103446960449, "loss": 1.0849, "step": 525 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.39801308512687683, "beta_dpo/beta_margin_grad_std": 0.069210484623909, "beta_dpo/beta_margin_mean": 0.42354100942611694, "beta_dpo/beta_margin_std": 0.29874640703201294, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.39645516872406, "beta_dpo/gap_mean": 402.06280517578125, "beta_dpo/gap_std": 358.2450256347656, "beta_dpo/loss_margin_mean": 423.5409851074219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7723935389133627, "grad_norm": 25.767820358276367, "learning_rate": 7.596651350926836e-08, "logits/chosen": -2.546776294708252, "logits/rejected": -2.6113803386688232, "loss": 1.0586, "step": 526 }, { "beta_dpo/beta": 1.2439494132995605, "beta_dpo/beta_margin_grad_mean": -0.2362092137336731, "beta_dpo/beta_margin_grad_std": 0.24737012386322021, "beta_dpo/beta_margin_mean": 454.62713623046875, "beta_dpo/beta_margin_std": 778.6525268554688, "beta_dpo/beta_used": 1.2439494132995605, "beta_dpo/beta_used_raw": 1.1723434925079346, "beta_dpo/gap_mean": 398.12164306640625, "beta_dpo/gap_std": 351.50225830078125, "beta_dpo/loss_margin_mean": 367.6454772949219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7738619676945668, "grad_norm": 16.68294334411621, "learning_rate": 7.504749238082414e-08, "logits/chosen": -2.6491198539733887, "logits/rejected": -2.690948009490967, "loss": 0.5226, "step": 527 }, { "beta_dpo/beta": 0.8667228817939758, "beta_dpo/beta_margin_grad_mean": -0.26016074419021606, "beta_dpo/beta_margin_grad_std": 0.27118903398513794, "beta_dpo/beta_margin_mean": 354.2501220703125, "beta_dpo/beta_margin_std": 623.5271606445312, "beta_dpo/beta_used": 0.8667228817939758, "beta_dpo/beta_used_raw": 0.277756929397583, "beta_dpo/gap_mean": 398.49981689453125, "beta_dpo/gap_std": 356.4637451171875, "beta_dpo/loss_margin_mean": 409.8684997558594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.775330396475771, "grad_norm": 1461.6524658203125, "learning_rate": 7.413308141366254e-08, "logits/chosen": -2.6200203895568848, "logits/rejected": -2.6761021614074707, "loss": 0.5897, "step": 528 }, { "beta_dpo/beta": 0.020879942923784256, "beta_dpo/beta_margin_grad_mean": -0.297647625207901, "beta_dpo/beta_margin_grad_std": 0.2874828577041626, "beta_dpo/beta_margin_mean": 7.774229049682617, "beta_dpo/beta_margin_std": 13.420893669128418, "beta_dpo/beta_used": 0.020879942923784256, "beta_dpo/beta_used_raw": -1.6348989009857178, "beta_dpo/gap_mean": 395.13067626953125, "beta_dpo/gap_std": 357.2676696777344, "beta_dpo/loss_margin_mean": 370.92144775390625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7767988252569751, "grad_norm": 690.0531616210938, "learning_rate": 7.322330470336313e-08, "logits/chosen": -2.5948238372802734, "logits/rejected": -2.6651344299316406, "loss": 1.5042, "step": 529 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4054359197616577, "beta_dpo/beta_margin_grad_std": 0.08558761328458786, "beta_dpo/beta_margin_mean": 0.39957693219184875, "beta_dpo/beta_margin_std": 0.38682180643081665, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.191973090171814, "beta_dpo/gap_mean": 393.1609802246094, "beta_dpo/gap_std": 356.60321044921875, "beta_dpo/loss_margin_mean": 399.576904296875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7782672540381792, "grad_norm": 31.403076171875, "learning_rate": 7.231818622338822e-08, "logits/chosen": -2.56978440284729, "logits/rejected": -2.6270089149475098, "loss": 1.0674, "step": 530 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4135720729827881, "beta_dpo/beta_margin_grad_std": 0.06900368630886078, "beta_dpo/beta_margin_mean": 0.35677656531333923, "beta_dpo/beta_margin_std": 0.29285889863967896, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.2725701332092285, "beta_dpo/gap_mean": 389.8656005859375, "beta_dpo/gap_std": 350.780517578125, "beta_dpo/loss_margin_mean": 356.77655029296875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7797356828193832, "grad_norm": 32.01457595825195, "learning_rate": 7.141774982445147e-08, "logits/chosen": -2.617753028869629, "logits/rejected": -2.669069290161133, "loss": 1.0808, "step": 531 }, { "beta_dpo/beta": 0.38395339250564575, "beta_dpo/beta_margin_grad_mean": -0.279400497674942, "beta_dpo/beta_margin_grad_std": 0.2882375121116638, "beta_dpo/beta_margin_mean": 182.82447814941406, "beta_dpo/beta_margin_std": 285.89324951171875, "beta_dpo/beta_used": 0.38395339250564575, "beta_dpo/beta_used_raw": -0.1769140660762787, "beta_dpo/gap_mean": 390.6937255859375, "beta_dpo/gap_std": 356.8406066894531, "beta_dpo/loss_margin_mean": 376.4180908203125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7812041116005873, "grad_norm": 17459.1015625, "learning_rate": 7.052201923388953e-08, "logits/chosen": -2.6145548820495605, "logits/rejected": -2.652617931365967, "loss": 15.0837, "step": 532 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4183679223060608, "beta_dpo/beta_margin_grad_std": 0.08154841512441635, "beta_dpo/beta_margin_mean": 0.34022700786590576, "beta_dpo/beta_margin_std": 0.35058268904685974, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.292230129241943, "beta_dpo/gap_mean": 374.2669677734375, "beta_dpo/gap_std": 358.24761962890625, "beta_dpo/loss_margin_mean": 340.22698974609375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7826725403817915, "grad_norm": 28.039445877075195, "learning_rate": 6.963101805503646e-08, "logits/chosen": -2.640665054321289, "logits/rejected": -2.666677474975586, "loss": 1.1201, "step": 533 }, { "beta_dpo/beta": 2.3452858924865723, "beta_dpo/beta_margin_grad_mean": -0.23946087062358856, "beta_dpo/beta_margin_grad_std": 0.25099116563796997, "beta_dpo/beta_margin_mean": 1004.8426513671875, "beta_dpo/beta_margin_std": 1333.9495849609375, "beta_dpo/beta_used": 2.3452858924865723, "beta_dpo/beta_used_raw": 1.948282241821289, "beta_dpo/gap_mean": 380.38348388671875, "beta_dpo/gap_std": 344.8291015625, "beta_dpo/loss_margin_mean": 386.6946105957031, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7841409691629956, "grad_norm": 23.792999267578125, "learning_rate": 6.874476976660184e-08, "logits/chosen": -2.6000583171844482, "logits/rejected": -2.6584179401397705, "loss": 0.5378, "step": 534 }, { "beta_dpo/beta": 0.4307881295681, "beta_dpo/beta_margin_grad_mean": -0.2031729370355606, "beta_dpo/beta_margin_grad_std": 0.2144741714000702, "beta_dpo/beta_margin_mean": 199.35324096679688, "beta_dpo/beta_margin_std": 241.94654846191406, "beta_dpo/beta_used": 0.4307881295681, "beta_dpo/beta_used_raw": -1.0114219188690186, "beta_dpo/gap_mean": 384.7388610839844, "beta_dpo/gap_std": 346.4287109375, "beta_dpo/loss_margin_mean": 431.2344970703125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7856093979441997, "grad_norm": 19.05939292907715, "learning_rate": 6.786329772205246e-08, "logits/chosen": -2.6111888885498047, "logits/rejected": -2.6767916679382324, "loss": 0.5469, "step": 535 }, { "beta_dpo/beta": 1.0523020029067993, "beta_dpo/beta_margin_grad_mean": -0.2505837380886078, "beta_dpo/beta_margin_grad_std": 0.2626131772994995, "beta_dpo/beta_margin_mean": 488.0250244140625, "beta_dpo/beta_margin_std": 711.1327514648438, "beta_dpo/beta_used": 1.0523020029067993, "beta_dpo/beta_used_raw": -0.6562622785568237, "beta_dpo/gap_mean": 391.91632080078125, "beta_dpo/gap_std": 333.91339111328125, "beta_dpo/loss_margin_mean": 422.7450256347656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7870778267254038, "grad_norm": 29718.033203125, "learning_rate": 6.698662514899638e-08, "logits/chosen": -2.6316137313842773, "logits/rejected": -2.7104220390319824, "loss": 17.1121, "step": 536 }, { "beta_dpo/beta": 0.34959638118743896, "beta_dpo/beta_margin_grad_mean": -0.22656311094760895, "beta_dpo/beta_margin_grad_std": 0.2316565066576004, "beta_dpo/beta_margin_mean": 147.22267150878906, "beta_dpo/beta_margin_std": 205.2282257080078, "beta_dpo/beta_used": 0.34959638118743896, "beta_dpo/beta_used_raw": -1.9871588945388794, "beta_dpo/gap_mean": 391.814208984375, "beta_dpo/gap_std": 327.1514892578125, "beta_dpo/loss_margin_mean": 381.2959899902344, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.788546255506608, "grad_norm": 6313.296875, "learning_rate": 6.611477514857114e-08, "logits/chosen": -2.599724531173706, "logits/rejected": -2.6477694511413574, "loss": 8.0099, "step": 537 }, { "beta_dpo/beta": 1.2048133611679077, "beta_dpo/beta_margin_grad_mean": -0.2943846881389618, "beta_dpo/beta_margin_grad_std": 0.2998571991920471, "beta_dpo/beta_margin_mean": 482.0516662597656, "beta_dpo/beta_margin_std": 875.5813598632812, "beta_dpo/beta_used": 1.2048133611679077, "beta_dpo/beta_used_raw": 1.1537326574325562, "beta_dpo/gap_mean": 395.1236572265625, "beta_dpo/gap_std": 327.88568115234375, "beta_dpo/loss_margin_mean": 404.2270202636719, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7900146842878121, "grad_norm": 41323.671875, "learning_rate": 6.524777069483525e-08, "logits/chosen": -2.661776542663574, "logits/rejected": -2.7106385231018066, "loss": 14.8493, "step": 538 }, { "beta_dpo/beta": 1.1724320650100708, "beta_dpo/beta_margin_grad_mean": -0.25635582208633423, "beta_dpo/beta_margin_grad_std": 0.2684703767299652, "beta_dpo/beta_margin_mean": 555.3789672851562, "beta_dpo/beta_margin_std": 792.5721435546875, "beta_dpo/beta_used": 1.1724320650100708, "beta_dpo/beta_used_raw": -1.5928860902786255, "beta_dpo/gap_mean": 399.99566650390625, "beta_dpo/gap_std": 335.73211669921875, "beta_dpo/loss_margin_mean": 406.2193298339844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7914831130690162, "grad_norm": 28027.939453125, "learning_rate": 6.438563463416221e-08, "logits/chosen": -2.6445555686950684, "logits/rejected": -2.700587511062622, "loss": 4.6975, "step": 539 }, { "beta_dpo/beta": 2.8483712673187256, "beta_dpo/beta_margin_grad_mean": -0.15959151089191437, "beta_dpo/beta_margin_grad_std": 0.36039698123931885, "beta_dpo/beta_margin_mean": 1546.208740234375, "beta_dpo/beta_margin_std": 2491.53271484375, "beta_dpo/beta_used": 2.8483712673187256, "beta_dpo/beta_used_raw": 2.8483712673187256, "beta_dpo/gap_mean": 402.3629150390625, "beta_dpo/gap_std": 344.58099365234375, "beta_dpo/loss_margin_mean": 461.46783447265625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7929515418502202, "grad_norm": 125367.3359375, "learning_rate": 6.352838968463919e-08, "logits/chosen": -2.636195659637451, "logits/rejected": -2.7234175205230713, "loss": 106.2179, "step": 540 }, { "beta_dpo/beta": 0.7488507628440857, "beta_dpo/beta_margin_grad_mean": -0.09352385252714157, "beta_dpo/beta_margin_grad_std": 0.2907818555831909, "beta_dpo/beta_margin_mean": 317.4271545410156, "beta_dpo/beta_margin_std": 262.6924133300781, "beta_dpo/beta_used": 0.7488507628440857, "beta_dpo/beta_used_raw": 0.7488507628440857, "beta_dpo/gap_mean": 415.33734130859375, "beta_dpo/gap_std": 352.0528869628906, "beta_dpo/loss_margin_mean": 423.2877502441406, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7944199706314243, "grad_norm": 10829.640625, "learning_rate": 6.267605843546767e-08, "logits/chosen": -2.6342930793762207, "logits/rejected": -2.6951427459716797, "loss": 11.3344, "step": 541 }, { "beta_dpo/beta": 0.9092043042182922, "beta_dpo/beta_margin_grad_mean": -0.20870338380336761, "beta_dpo/beta_margin_grad_std": 0.21450397372245789, "beta_dpo/beta_margin_mean": 514.0123291015625, "beta_dpo/beta_margin_std": 732.8487548828125, "beta_dpo/beta_used": 0.9092043042182922, "beta_dpo/beta_used_raw": 0.6870288848876953, "beta_dpo/gap_mean": 422.68658447265625, "beta_dpo/gap_std": 355.011474609375, "beta_dpo/loss_margin_mean": 452.62939453125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7958883994126285, "grad_norm": 18.31541633605957, "learning_rate": 6.182866334636888e-08, "logits/chosen": -2.609865188598633, "logits/rejected": -2.7070038318634033, "loss": 0.5163, "step": 542 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4074610471725464, "beta_dpo/beta_margin_grad_std": 0.09081309288740158, "beta_dpo/beta_margin_mean": 0.3921317458152771, "beta_dpo/beta_margin_std": 0.4030529856681824, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.4958415031433105, "beta_dpo/gap_mean": 415.68060302734375, "beta_dpo/gap_std": 361.8984375, "beta_dpo/loss_margin_mean": 392.1317443847656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7973568281938326, "grad_norm": 31.554765701293945, "learning_rate": 6.098622674699147e-08, "logits/chosen": -2.6560518741607666, "logits/rejected": -2.747661828994751, "loss": 1.0673, "step": 543 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.39683008193969727, "beta_dpo/beta_margin_grad_std": 0.09721162915229797, "beta_dpo/beta_margin_mean": 0.4478076100349426, "beta_dpo/beta_margin_std": 0.4668709635734558, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.288872003555298, "beta_dpo/gap_mean": 416.6801452636719, "beta_dpo/gap_std": 379.8859558105469, "beta_dpo/loss_margin_mean": 447.8075866699219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.7988252569750367, "grad_norm": 30.542661666870117, "learning_rate": 6.01487708363232e-08, "logits/chosen": -2.613583564758301, "logits/rejected": -2.71563458442688, "loss": 1.0614, "step": 544 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.401055246591568, "beta_dpo/beta_margin_grad_std": 0.08540057390928268, "beta_dpo/beta_margin_mean": 0.4177744388580322, "beta_dpo/beta_margin_std": 0.38359367847442627, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.404287576675415, "beta_dpo/gap_mean": 419.52044677734375, "beta_dpo/gap_std": 381.7796630859375, "beta_dpo/loss_margin_mean": 417.7744140625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8002936857562408, "grad_norm": 28.606706619262695, "learning_rate": 5.9316317682106294e-08, "logits/chosen": -2.57968807220459, "logits/rejected": -2.652435302734375, "loss": 1.0454, "step": 545 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4062955677509308, "beta_dpo/beta_margin_grad_std": 0.08303419500589371, "beta_dpo/beta_margin_mean": 0.39024630188941956, "beta_dpo/beta_margin_std": 0.3554038107395172, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.077838897705078, "beta_dpo/gap_mean": 417.3631591796875, "beta_dpo/gap_std": 376.4284362792969, "beta_dpo/loss_margin_mean": 390.24627685546875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.801762114537445, "grad_norm": 27.95216941833496, "learning_rate": 5.848888922025552e-08, "logits/chosen": -2.5938525199890137, "logits/rejected": -2.6582956314086914, "loss": 1.0875, "step": 546 }, { "beta_dpo/beta": 2.1992263793945312, "beta_dpo/beta_margin_grad_mean": -0.2521184980869293, "beta_dpo/beta_margin_grad_std": 0.264556348323822, "beta_dpo/beta_margin_mean": 901.9774169921875, "beta_dpo/beta_margin_std": 1424.427490234375, "beta_dpo/beta_used": 2.1992263793945312, "beta_dpo/beta_used_raw": 0.5545767545700073, "beta_dpo/gap_mean": 409.3026123046875, "beta_dpo/gap_std": 369.95867919921875, "beta_dpo/loss_margin_mean": 390.19146728515625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8032305433186491, "grad_norm": 20749.9375, "learning_rate": 5.7666507254280265e-08, "logits/chosen": -2.629915714263916, "logits/rejected": -2.6946840286254883, "loss": 15.5121, "step": 547 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.3962554335594177, "beta_dpo/beta_margin_grad_std": 0.07508466392755508, "beta_dpo/beta_margin_mean": 0.4344782829284668, "beta_dpo/beta_margin_std": 0.3340926766395569, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.715212345123291, "beta_dpo/gap_mean": 413.58660888671875, "beta_dpo/gap_std": 367.1708984375, "beta_dpo/loss_margin_mean": 434.478271484375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8046989720998532, "grad_norm": 31.523386001586914, "learning_rate": 5.684919345471029e-08, "logits/chosen": -2.636038064956665, "logits/rejected": -2.7150168418884277, "loss": 1.0832, "step": 548 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41863563656806946, "beta_dpo/beta_margin_grad_std": 0.07815483957529068, "beta_dpo/beta_margin_mean": 0.3384191393852234, "beta_dpo/beta_margin_std": 0.33370572328567505, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.80178689956665, "beta_dpo/gap_mean": 403.1805419921875, "beta_dpo/gap_std": 359.09490966796875, "beta_dpo/loss_margin_mean": 338.41912841796875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8061674008810573, "grad_norm": 27.266035079956055, "learning_rate": 5.603696935852426e-08, "logits/chosen": -2.6075119972229004, "logits/rejected": -2.672619342803955, "loss": 1.1066, "step": 549 }, { "beta_dpo/beta": 1.0455665588378906, "beta_dpo/beta_margin_grad_mean": -0.2663308084011078, "beta_dpo/beta_margin_grad_std": 0.2776106595993042, "beta_dpo/beta_margin_mean": 477.6100769042969, "beta_dpo/beta_margin_std": 690.04931640625, "beta_dpo/beta_used": 1.0455665588378906, "beta_dpo/beta_used_raw": 0.7292245626449585, "beta_dpo/gap_mean": 404.16845703125, "beta_dpo/gap_std": 352.69635009765625, "beta_dpo/loss_margin_mean": 419.5482177734375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8076358296622613, "grad_norm": 23331.697265625, "learning_rate": 5.5229856368582376e-08, "logits/chosen": -2.585789680480957, "logits/rejected": -2.6722607612609863, "loss": 21.6382, "step": 550 }, { "beta_dpo/beta": 2.5201354026794434, "beta_dpo/beta_margin_grad_mean": -0.078125, "beta_dpo/beta_margin_grad_std": 0.2683681845664978, "beta_dpo/beta_margin_mean": 1171.402099609375, "beta_dpo/beta_margin_std": 1031.3978271484375, "beta_dpo/beta_used": 2.5201354026794434, "beta_dpo/beta_used_raw": 2.5201354026794434, "beta_dpo/gap_mean": 412.00518798828125, "beta_dpo/gap_std": 351.75830078125, "beta_dpo/loss_margin_mean": 468.22454833984375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8091042584434655, "grad_norm": 50141.44140625, "learning_rate": 5.4427875753062734e-08, "logits/chosen": -2.585766553878784, "logits/rejected": -2.6833677291870117, "loss": 40.5927, "step": 551 }, { "beta_dpo/beta": 2.441541910171509, "beta_dpo/beta_margin_grad_mean": -0.0625, "beta_dpo/beta_margin_grad_std": 0.24206146597862244, "beta_dpo/beta_margin_mean": 1086.489013671875, "beta_dpo/beta_margin_std": 1131.8115234375, "beta_dpo/beta_used": 2.441541910171509, "beta_dpo/beta_used_raw": 2.441541910171509, "beta_dpo/gap_mean": 425.93841552734375, "beta_dpo/gap_std": 354.64801025390625, "beta_dpo/loss_margin_mean": 469.59014892578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8105726872246696, "grad_norm": 5.776141833273076e-18, "learning_rate": 5.363104864490034e-08, "logits/chosen": -2.6029000282287598, "logits/rejected": -2.7190818786621094, "loss": 0.0, "step": 552 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4171546697616577, "beta_dpo/beta_margin_grad_std": 0.08783596754074097, "beta_dpo/beta_margin_mean": 0.35056599974632263, "beta_dpo/beta_margin_std": 0.3956577479839325, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -5.028049945831299, "beta_dpo/gap_mean": 416.62200927734375, "beta_dpo/gap_std": 366.66143798828125, "beta_dpo/loss_margin_mean": 350.56597900390625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8120411160058737, "grad_norm": 38.534873962402344, "learning_rate": 5.2839396041230415e-08, "logits/chosen": -2.6273856163024902, "logits/rejected": -2.6760740280151367, "loss": 1.0993, "step": 553 }, { "beta_dpo/beta": 1.2997931241989136, "beta_dpo/beta_margin_grad_mean": -0.109375, "beta_dpo/beta_margin_grad_std": 0.3121091425418854, "beta_dpo/beta_margin_mean": 508.1010437011719, "beta_dpo/beta_margin_std": 419.82977294921875, "beta_dpo/beta_used": 1.2997931241989136, "beta_dpo/beta_used_raw": 1.2997931241989136, "beta_dpo/gap_mean": 408.51751708984375, "beta_dpo/gap_std": 352.59368896484375, "beta_dpo/loss_margin_mean": 392.65533447265625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8135095447870778, "grad_norm": 1126.18505859375, "learning_rate": 5.205293880283551e-08, "logits/chosen": -2.5999350547790527, "logits/rejected": -2.641538143157959, "loss": 1.1067, "step": 554 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.38637006282806396, "beta_dpo/beta_margin_grad_std": 0.08780043572187424, "beta_dpo/beta_margin_mean": 0.4845529794692993, "beta_dpo/beta_margin_std": 0.4055006504058838, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.4782108068466187, "beta_dpo/gap_mean": 418.3115234375, "beta_dpo/gap_std": 359.09735107421875, "beta_dpo/loss_margin_mean": 484.5529479980469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8149779735682819, "grad_norm": 34.80107116699219, "learning_rate": 5.127169765359515e-08, "logits/chosen": -2.5728185176849365, "logits/rejected": -2.6831090450286865, "loss": 1.0492, "step": 555 }, { "beta_dpo/beta": 0.1738041341304779, "beta_dpo/beta_margin_grad_mean": -0.2653172016143799, "beta_dpo/beta_margin_grad_std": 0.27693215012550354, "beta_dpo/beta_margin_mean": 74.29708862304688, "beta_dpo/beta_margin_std": 117.45230102539062, "beta_dpo/beta_used": 0.1738041341304779, "beta_dpo/beta_used_raw": -0.14080658555030823, "beta_dpo/gap_mean": 420.77178955078125, "beta_dpo/gap_std": 361.7735595703125, "beta_dpo/loss_margin_mean": 409.55499267578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8164464023494861, "grad_norm": 9489.583984375, "learning_rate": 5.049569317994012e-08, "logits/chosen": -2.5879225730895996, "logits/rejected": -2.687049388885498, "loss": 0.8618, "step": 556 }, { "beta_dpo/beta": 1.4377208948135376, "beta_dpo/beta_margin_grad_mean": -0.2533913552761078, "beta_dpo/beta_margin_grad_std": 0.26882457733154297, "beta_dpo/beta_margin_mean": 662.1341552734375, "beta_dpo/beta_margin_std": 1010.3662109375, "beta_dpo/beta_used": 1.4377208948135376, "beta_dpo/beta_used_raw": -2.254744052886963, "beta_dpo/gap_mean": 420.5530090332031, "beta_dpo/gap_std": 363.3053283691406, "beta_dpo/loss_margin_mean": 413.83026123046875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8179148311306902, "grad_norm": 39387.4296875, "learning_rate": 4.9724945830310144e-08, "logits/chosen": -2.614232063293457, "logits/rejected": -2.698105812072754, "loss": 1.1225, "step": 557 }, { "beta_dpo/beta": 2.3792362213134766, "beta_dpo/beta_margin_grad_mean": -0.2171439528465271, "beta_dpo/beta_margin_grad_std": 0.23049600422382355, "beta_dpo/beta_margin_mean": 1273.4932861328125, "beta_dpo/beta_margin_std": 1737.8341064453125, "beta_dpo/beta_used": 2.3792362213134766, "beta_dpo/beta_used_raw": 2.243035316467285, "beta_dpo/gap_mean": 433.89971923828125, "beta_dpo/gap_std": 363.2912902832031, "beta_dpo/loss_margin_mean": 526.4646606445312, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8193832599118943, "grad_norm": 42691.87109375, "learning_rate": 4.8959475914614554e-08, "logits/chosen": -2.583843231201172, "logits/rejected": -2.682178497314453, "loss": 0.5778, "step": 558 }, { "beta_dpo/beta": 0.7704762816429138, "beta_dpo/beta_margin_grad_mean": -0.26277267932891846, "beta_dpo/beta_margin_grad_std": 0.2761915922164917, "beta_dpo/beta_margin_mean": 319.6239013671875, "beta_dpo/beta_margin_std": 505.88055419921875, "beta_dpo/beta_used": 0.7704762816429138, "beta_dpo/beta_used_raw": -1.242068886756897, "beta_dpo/gap_mean": 435.0184326171875, "beta_dpo/gap_std": 358.98150634765625, "beta_dpo/loss_margin_mean": 415.4324645996094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8208516886930984, "grad_norm": 10755.6123046875, "learning_rate": 4.8199303603697614e-08, "logits/chosen": -2.5926883220672607, "logits/rejected": -2.6858973503112793, "loss": 19.0269, "step": 559 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41588789224624634, "beta_dpo/beta_margin_grad_std": 0.07136394083499908, "beta_dpo/beta_margin_mean": 0.3474079966545105, "beta_dpo/beta_margin_std": 0.30195000767707825, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -5.194782257080078, "beta_dpo/gap_mean": 420.5264587402344, "beta_dpo/gap_std": 349.8380126953125, "beta_dpo/loss_margin_mean": 347.4079895019531, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8223201174743024, "grad_norm": 30.960620880126953, "learning_rate": 4.7444448928806615e-08, "logits/chosen": -2.601827621459961, "logits/rejected": -2.647773265838623, "loss": 1.0969, "step": 560 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4009363651275635, "beta_dpo/beta_margin_grad_std": 0.08546130359172821, "beta_dpo/beta_margin_mean": 0.41753634810447693, "beta_dpo/beta_margin_std": 0.37727662920951843, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.459141254425049, "beta_dpo/gap_mean": 415.37933349609375, "beta_dpo/gap_std": 348.2698974609375, "beta_dpo/loss_margin_mean": 417.53631591796875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8237885462555066, "grad_norm": 37.56061935424805, "learning_rate": 4.669493178106432e-08, "logits/chosen": -2.5700759887695312, "logits/rejected": -2.675487995147705, "loss": 1.0767, "step": 561 }, { "beta_dpo/beta": 0.2502177059650421, "beta_dpo/beta_margin_grad_mean": -0.21572551131248474, "beta_dpo/beta_margin_grad_std": 0.2213025540113449, "beta_dpo/beta_margin_mean": 141.0299835205078, "beta_dpo/beta_margin_std": 179.341552734375, "beta_dpo/beta_used": 0.2502177059650421, "beta_dpo/beta_used_raw": -3.710040330886841, "beta_dpo/gap_mean": 425.248779296875, "beta_dpo/gap_std": 348.01812744140625, "beta_dpo/loss_margin_mean": 423.1015930175781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8252569750367107, "grad_norm": 24.252979278564453, "learning_rate": 4.5950771910944596e-08, "logits/chosen": -2.571836233139038, "logits/rejected": -2.639535427093506, "loss": 0.5656, "step": 562 }, { "beta_dpo/beta": 0.2640678286552429, "beta_dpo/beta_margin_grad_mean": -0.29870668053627014, "beta_dpo/beta_margin_grad_std": 0.29806166887283325, "beta_dpo/beta_margin_mean": 106.81857299804688, "beta_dpo/beta_margin_std": 167.0647430419922, "beta_dpo/beta_used": 0.2640678286552429, "beta_dpo/beta_used_raw": -3.6815316677093506, "beta_dpo/gap_mean": 400.4122314453125, "beta_dpo/gap_std": 342.2979736328125, "beta_dpo/loss_margin_mean": 323.61669921875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8267254038179148, "grad_norm": 2278.085205078125, "learning_rate": 4.521198892775202e-08, "logits/chosen": -2.5583033561706543, "logits/rejected": -2.625527858734131, "loss": 1.711, "step": 563 }, { "beta_dpo/beta": 1.4716358184814453, "beta_dpo/beta_margin_grad_mean": -0.06238251551985741, "beta_dpo/beta_margin_grad_std": 0.2377287596464157, "beta_dpo/beta_margin_mean": 640.9151611328125, "beta_dpo/beta_margin_std": 831.071044921875, "beta_dpo/beta_used": 1.4716358184814453, "beta_dpo/beta_used_raw": 1.4716358184814453, "beta_dpo/gap_mean": 404.2149658203125, "beta_dpo/gap_std": 335.1101989746094, "beta_dpo/loss_margin_mean": 427.81158447265625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8281938325991189, "grad_norm": 40077.27734375, "learning_rate": 4.447860229910544e-08, "logits/chosen": -2.6260743141174316, "logits/rejected": -2.693694591522217, "loss": 9.7535, "step": 564 }, { "beta_dpo/beta": 1.3616304397583008, "beta_dpo/beta_margin_grad_mean": -0.2510206699371338, "beta_dpo/beta_margin_grad_std": 0.26485851407051086, "beta_dpo/beta_margin_mean": 526.1046752929688, "beta_dpo/beta_margin_std": 818.9513549804688, "beta_dpo/beta_used": 1.3616304397583008, "beta_dpo/beta_used_raw": 0.21367371082305908, "beta_dpo/gap_mean": 402.90283203125, "beta_dpo/gap_std": 332.8077392578125, "beta_dpo/loss_margin_mean": 384.8162841796875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8296622613803231, "grad_norm": 25.107654571533203, "learning_rate": 4.375063135042445e-08, "logits/chosen": -2.5877442359924316, "logits/rejected": -2.6708731651306152, "loss": 0.5359, "step": 565 }, { "beta_dpo/beta": 1.1117289066314697, "beta_dpo/beta_margin_grad_mean": -0.20614933967590332, "beta_dpo/beta_margin_grad_std": 0.21492846310138702, "beta_dpo/beta_margin_mean": 586.4969482421875, "beta_dpo/beta_margin_std": 798.0072631835938, "beta_dpo/beta_used": 1.1117289066314697, "beta_dpo/beta_used_raw": -2.902581214904785, "beta_dpo/gap_mean": 405.11932373046875, "beta_dpo/gap_std": 338.6856994628906, "beta_dpo/loss_margin_mean": 448.7620544433594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8311306901615272, "grad_norm": 25.039459228515625, "learning_rate": 4.3028095264420525e-08, "logits/chosen": -2.568953037261963, "logits/rejected": -2.6582655906677246, "loss": 0.5761, "step": 566 }, { "beta_dpo/beta": 0.7229585647583008, "beta_dpo/beta_margin_grad_mean": -0.2360886186361313, "beta_dpo/beta_margin_grad_std": 0.24904760718345642, "beta_dpo/beta_margin_mean": 289.8537902832031, "beta_dpo/beta_margin_std": 402.62164306640625, "beta_dpo/beta_used": 0.7229585647583008, "beta_dpo/beta_used_raw": -2.865267753601074, "beta_dpo/gap_mean": 407.5924072265625, "beta_dpo/gap_std": 333.9209289550781, "beta_dpo/loss_margin_mean": 388.1324768066406, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8325991189427313, "grad_norm": 17.503690719604492, "learning_rate": 4.231101308059165e-08, "logits/chosen": -2.597644329071045, "logits/rejected": -2.676164150238037, "loss": 0.5655, "step": 567 }, { "beta_dpo/beta": 1.4946538209915161, "beta_dpo/beta_margin_grad_mean": -0.2185162454843521, "beta_dpo/beta_margin_grad_std": 0.23013083636760712, "beta_dpo/beta_margin_mean": 696.1226196289062, "beta_dpo/beta_margin_std": 929.7108154296875, "beta_dpo/beta_used": 1.4946538209915161, "beta_dpo/beta_used_raw": 0.43433964252471924, "beta_dpo/gap_mean": 412.0642395019531, "beta_dpo/gap_std": 327.5071105957031, "beta_dpo/loss_margin_mean": 429.1242370605469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8340675477239354, "grad_norm": 25.224254608154297, "learning_rate": 4.1599403694720145e-08, "logits/chosen": -2.5394160747528076, "logits/rejected": -2.6383776664733887, "loss": 0.5296, "step": 568 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4124916195869446, "beta_dpo/beta_margin_grad_std": 0.07660473138093948, "beta_dpo/beta_margin_mean": 0.36489877104759216, "beta_dpo/beta_margin_std": 0.3362376391887665, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.51121187210083, "beta_dpo/gap_mean": 404.7514953613281, "beta_dpo/gap_std": 331.2776794433594, "beta_dpo/loss_margin_mean": 364.8987731933594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8355359765051396, "grad_norm": 38.54376983642578, "learning_rate": 4.089328585837512e-08, "logits/chosen": -2.577268123626709, "logits/rejected": -2.6315019130706787, "loss": 1.0865, "step": 569 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4066663384437561, "beta_dpo/beta_margin_grad_std": 0.07038101553916931, "beta_dpo/beta_margin_mean": 0.38760435581207275, "beta_dpo/beta_margin_std": 0.30543091893196106, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.0776772499084473, "beta_dpo/gap_mean": 402.11566162109375, "beta_dpo/gap_std": 327.2163391113281, "beta_dpo/loss_margin_mean": 387.6043395996094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8370044052863436, "grad_norm": 30.418283462524414, "learning_rate": 4.019267817841834e-08, "logits/chosen": -2.6355581283569336, "logits/rejected": -2.698420286178589, "loss": 1.0645, "step": 570 }, { "beta_dpo/beta": 0.9894961714744568, "beta_dpo/beta_margin_grad_mean": -0.015625763684511185, "beta_dpo/beta_margin_grad_std": 0.12401949614286423, "beta_dpo/beta_margin_mean": 509.4083251953125, "beta_dpo/beta_margin_std": 569.6787719726562, "beta_dpo/beta_used": 0.9894961714744568, "beta_dpo/beta_used_raw": 0.9894961714744568, "beta_dpo/gap_mean": 413.31024169921875, "beta_dpo/gap_std": 324.23162841796875, "beta_dpo/loss_margin_mean": 487.08660888671875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8384728340675477, "grad_norm": 0.0953197032213211, "learning_rate": 3.9497599116513705e-08, "logits/chosen": -2.5648531913757324, "logits/rejected": -2.6703522205352783, "loss": 0.0, "step": 571 }, { "beta_dpo/beta": 2.384432792663574, "beta_dpo/beta_margin_grad_mean": -0.26681461930274963, "beta_dpo/beta_margin_grad_std": 0.2802523076534271, "beta_dpo/beta_margin_mean": 1058.479736328125, "beta_dpo/beta_margin_std": 1517.7403564453125, "beta_dpo/beta_used": 2.384432792663574, "beta_dpo/beta_used_raw": 1.8693180084228516, "beta_dpo/gap_mean": 416.3902587890625, "beta_dpo/gap_std": 328.5089111328125, "beta_dpo/loss_margin_mean": 412.9160461425781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8399412628487518, "grad_norm": 131773.4375, "learning_rate": 3.880806698864086e-08, "logits/chosen": -2.548858642578125, "logits/rejected": -2.6395535469055176, "loss": 10.9273, "step": 572 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.410361111164093, "beta_dpo/beta_margin_grad_std": 0.0751841589808464, "beta_dpo/beta_margin_mean": 0.37288060784339905, "beta_dpo/beta_margin_std": 0.32199469208717346, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.7826108932495117, "beta_dpo/gap_mean": 408.19134521484375, "beta_dpo/gap_std": 329.8096008300781, "beta_dpo/loss_margin_mean": 372.8805847167969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8414096916299559, "grad_norm": 31.714269638061523, "learning_rate": 3.812409996461275e-08, "logits/chosen": -2.6157898902893066, "logits/rejected": -2.681966781616211, "loss": 1.064, "step": 573 }, { "beta_dpo/beta": 3.0706517696380615, "beta_dpo/beta_margin_grad_mean": -0.23131898045539856, "beta_dpo/beta_margin_grad_std": 0.2463950663805008, "beta_dpo/beta_margin_mean": 1551.0074462890625, "beta_dpo/beta_margin_std": 2098.472900390625, "beta_dpo/beta_used": 3.0706517696380615, "beta_dpo/beta_used_raw": 2.929409980773926, "beta_dpo/gap_mean": 412.53070068359375, "beta_dpo/gap_std": 330.5635986328125, "beta_dpo/loss_margin_mean": 460.5701904296875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8428781204111601, "grad_norm": 21.132007598876953, "learning_rate": 3.74457160675965e-08, "logits/chosen": -2.6238203048706055, "logits/rejected": -2.720996856689453, "loss": 0.5214, "step": 574 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.38933318853378296, "beta_dpo/beta_margin_grad_std": 0.0830337256193161, "beta_dpo/beta_margin_mean": 0.4672733247280121, "beta_dpo/beta_margin_std": 0.37139537930488586, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.8129258155822754, "beta_dpo/gap_mean": 423.2646484375, "beta_dpo/gap_std": 333.3087463378906, "beta_dpo/loss_margin_mean": 467.2732849121094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8443465491923642, "grad_norm": 36.160743713378906, "learning_rate": 3.677293317363864e-08, "logits/chosen": -2.5244140625, "logits/rejected": -2.6159071922302246, "loss": 1.0514, "step": 575 }, { "beta_dpo/beta": 0.7232588529586792, "beta_dpo/beta_margin_grad_mean": -0.2747720777988434, "beta_dpo/beta_margin_grad_std": 0.2860935628414154, "beta_dpo/beta_margin_mean": 359.212646484375, "beta_dpo/beta_margin_std": 551.1680297851562, "beta_dpo/beta_used": 0.7232588529586792, "beta_dpo/beta_used_raw": -2.2491068840026855, "beta_dpo/gap_mean": 428.517578125, "beta_dpo/gap_std": 351.6317443847656, "beta_dpo/loss_margin_mean": 409.20550537109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8458149779735683, "grad_norm": 22067.658203125, "learning_rate": 3.6105769011194224e-08, "logits/chosen": -2.6149516105651855, "logits/rejected": -2.7341456413269043, "loss": 7.0247, "step": 576 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4043275713920593, "beta_dpo/beta_margin_grad_std": 0.06338826566934586, "beta_dpo/beta_margin_mean": 0.3954284191131592, "beta_dpo/beta_margin_std": 0.27391013503074646, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.11868953704834, "beta_dpo/gap_mean": 419.32061767578125, "beta_dpo/gap_std": 346.9593505859375, "beta_dpo/loss_margin_mean": 395.42840576171875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8472834067547724, "grad_norm": 23.889705657958984, "learning_rate": 3.5444241160659304e-08, "logits/chosen": -2.5963869094848633, "logits/rejected": -2.666391372680664, "loss": 1.0657, "step": 577 }, { "beta_dpo/beta": 1.4147895574569702, "beta_dpo/beta_margin_grad_mean": -0.23017671704292297, "beta_dpo/beta_margin_grad_std": 0.24086488783359528, "beta_dpo/beta_margin_mean": 697.8129272460938, "beta_dpo/beta_margin_std": 934.8630981445312, "beta_dpo/beta_used": 1.4147895574569702, "beta_dpo/beta_used_raw": 0.5740838050842285, "beta_dpo/gap_mean": 418.50299072265625, "beta_dpo/gap_std": 338.3753662109375, "beta_dpo/loss_margin_mean": 392.7145690917969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8487518355359766, "grad_norm": 17.308809280395508, "learning_rate": 3.478836705390808e-08, "logits/chosen": -2.5874085426330566, "logits/rejected": -2.6705245971679688, "loss": 0.5284, "step": 578 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4038708806037903, "beta_dpo/beta_margin_grad_std": 0.06529239565134048, "beta_dpo/beta_margin_mean": 0.39867639541625977, "beta_dpo/beta_margin_std": 0.2885708808898926, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.684361457824707, "beta_dpo/gap_mean": 410.989501953125, "beta_dpo/gap_std": 332.41058349609375, "beta_dpo/loss_margin_mean": 398.6763610839844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8502202643171806, "grad_norm": 27.025897979736328, "learning_rate": 3.41381639738331e-08, "logits/chosen": -2.5982184410095215, "logits/rejected": -2.6704506874084473, "loss": 1.064, "step": 579 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4143833518028259, "beta_dpo/beta_margin_grad_std": 0.09059581905603409, "beta_dpo/beta_margin_mean": 0.36092129349708557, "beta_dpo/beta_margin_std": 0.3993350863456726, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.0408852100372314, "beta_dpo/gap_mean": 404.3271484375, "beta_dpo/gap_std": 339.910888671875, "beta_dpo/loss_margin_mean": 360.9212646484375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8516886930983847, "grad_norm": 35.905391693115234, "learning_rate": 3.349364905389032e-08, "logits/chosen": -2.6305480003356934, "logits/rejected": -2.701788902282715, "loss": 1.0811, "step": 580 }, { "beta_dpo/beta": 2.7418603897094727, "beta_dpo/beta_margin_grad_mean": -0.24978549778461456, "beta_dpo/beta_margin_grad_std": 0.26566624641418457, "beta_dpo/beta_margin_mean": 1087.9808349609375, "beta_dpo/beta_margin_std": 1690.84423828125, "beta_dpo/beta_used": 2.7418603897094727, "beta_dpo/beta_used_raw": 1.8317091464996338, "beta_dpo/gap_mean": 399.22796630859375, "beta_dpo/gap_std": 348.1463928222656, "beta_dpo/loss_margin_mean": 397.565673828125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8531571218795888, "grad_norm": 18.974281311035156, "learning_rate": 3.285483927764726e-08, "logits/chosen": -2.6319580078125, "logits/rejected": -2.72849702835083, "loss": 0.5351, "step": 581 }, { "beta_dpo/beta": 4.0522236824035645, "beta_dpo/beta_margin_grad_mean": -0.23803767561912537, "beta_dpo/beta_margin_grad_std": 0.25662490725517273, "beta_dpo/beta_margin_mean": 2030.54248046875, "beta_dpo/beta_margin_std": 2940.37939453125, "beta_dpo/beta_used": 4.0522236824035645, "beta_dpo/beta_used_raw": 2.1622235774993896, "beta_dpo/gap_mean": 401.3326416015625, "beta_dpo/gap_std": 360.9919738769531, "beta_dpo/loss_margin_mean": 435.0095520019531, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8546255506607929, "grad_norm": 60133.34765625, "learning_rate": 3.222175147833556e-08, "logits/chosen": -2.6616077423095703, "logits/rejected": -2.7627522945404053, "loss": 28.4446, "step": 582 }, { "beta_dpo/beta": 2.0656654834747314, "beta_dpo/beta_margin_grad_mean": -0.25362730026245117, "beta_dpo/beta_margin_grad_std": 0.26769015192985535, "beta_dpo/beta_margin_mean": 971.8246459960938, "beta_dpo/beta_margin_std": 1321.0806884765625, "beta_dpo/beta_used": 2.0656654834747314, "beta_dpo/beta_used_raw": -2.4300012588500977, "beta_dpo/gap_mean": 404.298828125, "beta_dpo/gap_std": 363.2397155761719, "beta_dpo/loss_margin_mean": 354.9598693847656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.856093979441997, "grad_norm": 24.137685775756836, "learning_rate": 3.159440233840763e-08, "logits/chosen": -2.6524295806884766, "logits/rejected": -2.708967685699463, "loss": 0.589, "step": 583 }, { "beta_dpo/beta": 3.7489781379699707, "beta_dpo/beta_margin_grad_mean": -0.03125, "beta_dpo/beta_margin_grad_std": 0.17399263381958008, "beta_dpo/beta_margin_mean": 1806.1541748046875, "beta_dpo/beta_margin_std": 1009.447021484375, "beta_dpo/beta_used": 3.7489781379699707, "beta_dpo/beta_used_raw": 3.7489781379699707, "beta_dpo/gap_mean": 409.24481201171875, "beta_dpo/gap_std": 357.03033447265625, "beta_dpo/loss_margin_mean": 484.4200134277344, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8575624082232012, "grad_norm": 42800.1015625, "learning_rate": 3.0972808389096635e-08, "logits/chosen": -2.6379153728485107, "logits/rejected": -2.73630428314209, "loss": 29.1861, "step": 584 }, { "beta_dpo/beta": 1.4875493049621582, "beta_dpo/beta_margin_grad_mean": -0.21340592205524445, "beta_dpo/beta_margin_grad_std": 0.22890949249267578, "beta_dpo/beta_margin_mean": 652.2798461914062, "beta_dpo/beta_margin_std": 899.9537963867188, "beta_dpo/beta_used": 1.4875493049621582, "beta_dpo/beta_used_raw": 1.2730743885040283, "beta_dpo/gap_mean": 416.50323486328125, "beta_dpo/gap_std": 353.55279541015625, "beta_dpo/loss_margin_mean": 440.5109558105469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8590308370044053, "grad_norm": 190.09495544433594, "learning_rate": 3.035698600998121e-08, "logits/chosen": -2.620150566101074, "logits/rejected": -2.7122976779937744, "loss": 0.5226, "step": 585 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40226635336875916, "beta_dpo/beta_margin_grad_std": 0.08389244973659515, "beta_dpo/beta_margin_mean": 0.41126564145088196, "beta_dpo/beta_margin_std": 0.3721611201763153, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.403745174407959, "beta_dpo/gap_mean": 414.5293884277344, "beta_dpo/gap_std": 352.2566833496094, "beta_dpo/loss_margin_mean": 411.265625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8604992657856094, "grad_norm": 30.519203186035156, "learning_rate": 2.974695142855388e-08, "logits/chosen": -2.6606011390686035, "logits/rejected": -2.7575459480285645, "loss": 1.0402, "step": 586 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4073527753353119, "beta_dpo/beta_margin_grad_std": 0.08573532849550247, "beta_dpo/beta_margin_mean": 0.3906094431877136, "beta_dpo/beta_margin_std": 0.37780997157096863, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.9087142944335938, "beta_dpo/gap_mean": 414.48919677734375, "beta_dpo/gap_std": 359.83526611328125, "beta_dpo/loss_margin_mean": 390.6094055175781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8619676945668135, "grad_norm": 31.185678482055664, "learning_rate": 2.9142720719793122e-08, "logits/chosen": -2.6660566329956055, "logits/rejected": -2.7582545280456543, "loss": 1.0746, "step": 587 }, { "beta_dpo/beta": 1.2841373682022095, "beta_dpo/beta_margin_grad_mean": -0.24702578783035278, "beta_dpo/beta_margin_grad_std": 0.26337432861328125, "beta_dpo/beta_margin_mean": 613.1337890625, "beta_dpo/beta_margin_std": 1012.4978637695312, "beta_dpo/beta_used": 1.2841373682022095, "beta_dpo/beta_used_raw": 0.5301622748374939, "beta_dpo/gap_mean": 418.21356201171875, "beta_dpo/gap_std": 369.077392578125, "beta_dpo/loss_margin_mean": 447.8412170410156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8634361233480177, "grad_norm": 156.63604736328125, "learning_rate": 2.8544309805740018e-08, "logits/chosen": -2.622619152069092, "logits/rejected": -2.741431951522827, "loss": 0.5288, "step": 588 }, { "beta_dpo/beta": 2.8131837844848633, "beta_dpo/beta_margin_grad_mean": -0.15622615814208984, "beta_dpo/beta_margin_grad_std": 0.3630368411540985, "beta_dpo/beta_margin_mean": 1222.51513671875, "beta_dpo/beta_margin_std": 1587.3143310546875, "beta_dpo/beta_used": 2.8131837844848633, "beta_dpo/beta_used_raw": 2.8131837844848633, "beta_dpo/gap_mean": 424.3757629394531, "beta_dpo/gap_std": 373.8643798828125, "beta_dpo/loss_margin_mean": 476.1635437011719, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8649045521292217, "grad_norm": 21344.30859375, "learning_rate": 2.7951734455078786e-08, "logits/chosen": -2.665593147277832, "logits/rejected": -2.7899739742279053, "loss": 19.9603, "step": 589 }, { "beta_dpo/beta": 0.3272712528705597, "beta_dpo/beta_margin_grad_mean": -0.24900765717029572, "beta_dpo/beta_margin_grad_std": 0.27214398980140686, "beta_dpo/beta_margin_mean": 128.85853576660156, "beta_dpo/beta_margin_std": 206.27391052246094, "beta_dpo/beta_used": 0.3272712528705597, "beta_dpo/beta_used_raw": -0.018486618995666504, "beta_dpo/gap_mean": 432.82989501953125, "beta_dpo/gap_std": 383.5555725097656, "beta_dpo/loss_margin_mean": 477.1200256347656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8663729809104258, "grad_norm": 12794.5634765625, "learning_rate": 2.736501028272095e-08, "logits/chosen": -2.6336288452148438, "logits/rejected": -2.7674968242645264, "loss": 4.8078, "step": 590 }, { "beta_dpo/beta": 1.0653685331344604, "beta_dpo/beta_margin_grad_mean": -0.24658845365047455, "beta_dpo/beta_margin_grad_std": 0.2630532681941986, "beta_dpo/beta_margin_mean": 457.1188049316406, "beta_dpo/beta_margin_std": 694.888916015625, "beta_dpo/beta_used": 1.0653685331344604, "beta_dpo/beta_used_raw": 0.8892130255699158, "beta_dpo/gap_mean": 437.3492126464844, "beta_dpo/gap_std": 384.0495300292969, "beta_dpo/loss_margin_mean": 425.42724609375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8678414096916299, "grad_norm": 21.35018539428711, "learning_rate": 2.678415274939408e-08, "logits/chosen": -2.6317532062530518, "logits/rejected": -2.708348512649536, "loss": 0.5106, "step": 591 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4068998694419861, "beta_dpo/beta_margin_grad_std": 0.07823016494512558, "beta_dpo/beta_margin_mean": 0.3857780396938324, "beta_dpo/beta_margin_std": 0.328925758600235, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.6076627969741821, "beta_dpo/gap_mean": 429.1395568847656, "beta_dpo/gap_std": 375.99981689453125, "beta_dpo/loss_margin_mean": 385.77801513671875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.869309838472834, "grad_norm": 33.770408630371094, "learning_rate": 2.6209177161234442e-08, "logits/chosen": -2.661647319793701, "logits/rejected": -2.7471630573272705, "loss": 1.0442, "step": 592 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4293622672557831, "beta_dpo/beta_margin_grad_std": 0.08562355488538742, "beta_dpo/beta_margin_mean": 0.2922646403312683, "beta_dpo/beta_margin_std": 0.3565959632396698, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.2156394720077515, "beta_dpo/gap_mean": 407.1551513671875, "beta_dpo/gap_std": 372.3304443359375, "beta_dpo/loss_margin_mean": 292.2646179199219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8707782672540382, "grad_norm": 36.524818420410156, "learning_rate": 2.564009866938349e-08, "logits/chosen": -2.611865997314453, "logits/rejected": -2.6518046855926514, "loss": 1.0559, "step": 593 }, { "beta_dpo/beta": 0.8479436039924622, "beta_dpo/beta_margin_grad_mean": -0.2311505377292633, "beta_dpo/beta_margin_grad_std": 0.2402058094739914, "beta_dpo/beta_margin_mean": 418.7370910644531, "beta_dpo/beta_margin_std": 590.7848510742188, "beta_dpo/beta_used": 0.8479436039924622, "beta_dpo/beta_used_raw": -2.808560848236084, "beta_dpo/gap_mean": 395.17657470703125, "beta_dpo/gap_std": 363.5038146972656, "beta_dpo/loss_margin_mean": 389.0055847167969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8722466960352423, "grad_norm": 34.29524612426758, "learning_rate": 2.5076932269588708e-08, "logits/chosen": -2.6745779514312744, "logits/rejected": -2.7460973262786865, "loss": 0.5787, "step": 594 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40659037232398987, "beta_dpo/beta_margin_grad_std": 0.08061845600605011, "beta_dpo/beta_margin_mean": 0.3967190682888031, "beta_dpo/beta_margin_std": 0.38002651929855347, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.7765114307403564, "beta_dpo/gap_mean": 399.8134765625, "beta_dpo/gap_std": 366.6833190917969, "beta_dpo/loss_margin_mean": 396.71905517578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8737151248164464, "grad_norm": 24.00165557861328, "learning_rate": 2.451969280180849e-08, "logits/chosen": -2.631199359893799, "logits/rejected": -2.704611301422119, "loss": 1.092, "step": 595 }, { "beta_dpo/beta": 0.2762553095817566, "beta_dpo/beta_margin_grad_mean": -0.24379974603652954, "beta_dpo/beta_margin_grad_std": 0.25649240612983704, "beta_dpo/beta_margin_mean": 115.0361557006836, "beta_dpo/beta_margin_std": 168.74607849121094, "beta_dpo/beta_used": 0.2762553095817566, "beta_dpo/beta_used_raw": -2.552454948425293, "beta_dpo/gap_mean": 392.48553466796875, "beta_dpo/gap_std": 366.8974304199219, "beta_dpo/loss_margin_mean": 365.7869873046875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8751835535976505, "grad_norm": 7005.2744140625, "learning_rate": 2.396839494982103e-08, "logits/chosen": -2.6771159172058105, "logits/rejected": -2.754608631134033, "loss": 1.0769, "step": 596 }, { "beta_dpo/beta": 0.9559941291809082, "beta_dpo/beta_margin_grad_mean": -0.2519912123680115, "beta_dpo/beta_margin_grad_std": 0.26672062277793884, "beta_dpo/beta_margin_mean": 528.5262451171875, "beta_dpo/beta_margin_std": 805.1343383789062, "beta_dpo/beta_used": 0.9559941291809082, "beta_dpo/beta_used_raw": -0.41350656747817993, "beta_dpo/gap_mean": 407.9326171875, "beta_dpo/gap_std": 373.6595764160156, "beta_dpo/loss_margin_mean": 463.8984680175781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8766519823788547, "grad_norm": 22720.951171875, "learning_rate": 2.3423053240837514e-08, "logits/chosen": -2.669362783432007, "logits/rejected": -2.770700693130493, "loss": 10.1259, "step": 597 }, { "beta_dpo/beta": 3.963083267211914, "beta_dpo/beta_margin_grad_mean": -0.24513758718967438, "beta_dpo/beta_margin_grad_std": 0.25738582015037537, "beta_dpo/beta_margin_mean": 2212.566650390625, "beta_dpo/beta_margin_std": 3305.775634765625, "beta_dpo/beta_used": 3.963083267211914, "beta_dpo/beta_used_raw": 3.8194849491119385, "beta_dpo/gap_mean": 415.28436279296875, "beta_dpo/gap_std": 379.65228271484375, "beta_dpo/loss_margin_mean": 428.660400390625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8781204111600588, "grad_norm": 26.368377685546875, "learning_rate": 2.2883682045119062e-08, "logits/chosen": -2.6603126525878906, "logits/rejected": -2.7383084297180176, "loss": 0.5217, "step": 598 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40531477332115173, "beta_dpo/beta_margin_grad_std": 0.07651591300964355, "beta_dpo/beta_margin_mean": 0.3960907757282257, "beta_dpo/beta_margin_std": 0.3420655429363251, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.9786413908004761, "beta_dpo/gap_mean": 409.6473388671875, "beta_dpo/gap_std": 375.72576904296875, "beta_dpo/loss_margin_mean": 396.09075927734375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8795888399412628, "grad_norm": 25.34820556640625, "learning_rate": 2.2350295575598367e-08, "logits/chosen": -2.662172317504883, "logits/rejected": -2.7468762397766113, "loss": 1.0483, "step": 599 }, { "beta_dpo/beta": 0.6657280325889587, "beta_dpo/beta_margin_grad_mean": -0.2325766235589981, "beta_dpo/beta_margin_grad_std": 0.24632975459098816, "beta_dpo/beta_margin_mean": 258.60589599609375, "beta_dpo/beta_margin_std": 383.517333984375, "beta_dpo/beta_used": 0.6657280325889587, "beta_dpo/beta_used_raw": -0.8385080099105835, "beta_dpo/gap_mean": 404.39202880859375, "beta_dpo/gap_std": 360.17938232421875, "beta_dpo/loss_margin_mean": 396.5282287597656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8810572687224669, "grad_norm": 21.900676727294922, "learning_rate": 2.1822907887504932e-08, "logits/chosen": -2.691469669342041, "logits/rejected": -2.772876262664795, "loss": 0.5388, "step": 600 }, { "epoch": 0.8810572687224669, "eval_beta_dpo/beta": 0.0010000000474974513, "eval_beta_dpo/beta_margin_grad_mean": -0.44217321276664734, "eval_beta_dpo/beta_margin_grad_std": 0.09827572852373123, "eval_beta_dpo/beta_margin_mean": 0.24340428411960602, "eval_beta_dpo/beta_margin_std": 0.4217238128185272, "eval_beta_dpo/beta_used": 0.0010000000474974513, "eval_beta_dpo/beta_used_raw": -9.559965133666992, "eval_beta_dpo/gap_mean": 404.4036560058594, "eval_beta_dpo/gap_std": 357.40692138671875, "eval_beta_dpo/loss_margin_mean": 243.40426635742188, "eval_beta_dpo/mask_keep_frac": 1.0, "eval_logits/chosen": -2.781342029571533, "eval_logits/rejected": -2.8108348846435547, "eval_loss": 0.6014874577522278, "eval_runtime": 37.9176, "eval_samples_per_second": 61.686, "eval_steps_per_second": 1.952, "step": 600 }, { "beta_dpo/beta": 3.2707443237304688, "beta_dpo/beta_margin_grad_mean": -0.1272067278623581, "beta_dpo/beta_margin_grad_std": 0.3303484618663788, "beta_dpo/beta_margin_mean": 1377.1533203125, "beta_dpo/beta_margin_std": 1483.408203125, "beta_dpo/beta_used": 3.2707443237304688, "beta_dpo/beta_used_raw": 3.2707443237304688, "beta_dpo/gap_mean": 407.78009033203125, "beta_dpo/gap_std": 361.7393798828125, "beta_dpo/loss_margin_mean": 442.43011474609375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.882525697503671, "grad_norm": 115072.109375, "learning_rate": 2.1301532877994742e-08, "logits/chosen": -2.629629611968994, "logits/rejected": -2.7460412979125977, "loss": 73.7058, "step": 601 }, { "beta_dpo/beta": 0.6206492781639099, "beta_dpo/beta_margin_grad_mean": -0.23634099960327148, "beta_dpo/beta_margin_grad_std": 0.25701799988746643, "beta_dpo/beta_margin_mean": 263.51983642578125, "beta_dpo/beta_margin_std": 366.13446044921875, "beta_dpo/beta_used": 0.6206492781639099, "beta_dpo/beta_used_raw": 0.1565786600112915, "beta_dpo/gap_mean": 418.2040100097656, "beta_dpo/gap_std": 355.448486328125, "beta_dpo/loss_margin_mean": 467.5210876464844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8839941262848752, "grad_norm": 17.351716995239258, "learning_rate": 2.0786184285784298e-08, "logits/chosen": -2.7129225730895996, "logits/rejected": -2.8077144622802734, "loss": 0.5206, "step": 602 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40567564964294434, "beta_dpo/beta_margin_grad_std": 0.08365360647439957, "beta_dpo/beta_margin_mean": 0.39521723985671997, "beta_dpo/beta_margin_std": 0.362171471118927, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.66328501701355, "beta_dpo/gap_mean": 419.609375, "beta_dpo/gap_std": 357.5216064453125, "beta_dpo/loss_margin_mean": 395.21722412109375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8854625550660793, "grad_norm": 28.13165283203125, "learning_rate": 2.0276875690788204e-08, "logits/chosen": -2.724299430847168, "logits/rejected": -2.7903506755828857, "loss": 1.0665, "step": 603 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.39339369535446167, "beta_dpo/beta_margin_grad_std": 0.08459168672561646, "beta_dpo/beta_margin_mean": 0.45500051975250244, "beta_dpo/beta_margin_std": 0.4009777903556824, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.187976598739624, "beta_dpo/gap_mean": 422.8662109375, "beta_dpo/gap_std": 361.38897705078125, "beta_dpo/loss_margin_mean": 455.00048828125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8869309838472834, "grad_norm": 29.62197494506836, "learning_rate": 1.977362051376158e-08, "logits/chosen": -2.675366163253784, "logits/rejected": -2.7828691005706787, "loss": 1.0659, "step": 604 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.3992781937122345, "beta_dpo/beta_margin_grad_std": 0.09205874055624008, "beta_dpo/beta_margin_mean": 0.4294624328613281, "beta_dpo/beta_margin_std": 0.4166560769081116, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.5361804962158203, "beta_dpo/gap_mean": 425.74029541015625, "beta_dpo/gap_std": 377.2355651855469, "beta_dpo/loss_margin_mean": 429.46240234375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8883994126284875, "grad_norm": 26.918848037719727, "learning_rate": 1.9276432015946446e-08, "logits/chosen": -2.671755790710449, "logits/rejected": -2.7664794921875, "loss": 1.061, "step": 605 }, { "beta_dpo/beta": 0.22745679318904877, "beta_dpo/beta_margin_grad_mean": -0.2854154407978058, "beta_dpo/beta_margin_grad_std": 0.29264795780181885, "beta_dpo/beta_margin_mean": 94.71522521972656, "beta_dpo/beta_margin_std": 168.86326599121094, "beta_dpo/beta_used": 0.22745679318904877, "beta_dpo/beta_used_raw": -0.09920136630535126, "beta_dpo/gap_mean": 419.8639221191406, "beta_dpo/gap_std": 381.78704833984375, "beta_dpo/loss_margin_mean": 393.2106018066406, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8898678414096917, "grad_norm": 2858.609375, "learning_rate": 1.8785323298722093e-08, "logits/chosen": -2.6562769412994385, "logits/rejected": -2.7424442768096924, "loss": 2.9883, "step": 606 }, { "beta_dpo/beta": 1.7348406314849854, "beta_dpo/beta_margin_grad_mean": -0.27311357855796814, "beta_dpo/beta_margin_grad_std": 0.28208261728286743, "beta_dpo/beta_margin_mean": 661.6439819335938, "beta_dpo/beta_margin_std": 1191.2918701171875, "beta_dpo/beta_used": 1.7348406314849854, "beta_dpo/beta_used_raw": -0.1976158618927002, "beta_dpo/gap_mean": 408.81640625, "beta_dpo/gap_std": 380.6749572753906, "beta_dpo/loss_margin_mean": 353.4035949707031, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8913362701908958, "grad_norm": 22962.126953125, "learning_rate": 1.8300307303259904e-08, "logits/chosen": -2.70379900932312, "logits/rejected": -2.7581634521484375, "loss": 13.9436, "step": 607 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.39080411195755005, "beta_dpo/beta_margin_grad_std": 0.07455727458000183, "beta_dpo/beta_margin_mean": 0.4569474458694458, "beta_dpo/beta_margin_std": 0.328748881816864, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.0561343431472778, "beta_dpo/gap_mean": 413.77947998046875, "beta_dpo/gap_std": 373.3118896484375, "beta_dpo/loss_margin_mean": 456.9474182128906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8928046989720999, "grad_norm": 33.10684585571289, "learning_rate": 1.7821396810182437e-08, "logits/chosen": -2.6324472427368164, "logits/rejected": -2.7429165840148926, "loss": 1.0452, "step": 608 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.39637741446495056, "beta_dpo/beta_margin_grad_std": 0.07506369054317474, "beta_dpo/beta_margin_mean": 0.43244999647140503, "beta_dpo/beta_margin_std": 0.3281441330909729, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.2443511486053467, "beta_dpo/gap_mean": 418.79693603515625, "beta_dpo/gap_std": 365.2125244140625, "beta_dpo/loss_margin_mean": 432.449951171875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8942731277533039, "grad_norm": 32.35358810424805, "learning_rate": 1.7348604439226617e-08, "logits/chosen": -2.6647348403930664, "logits/rejected": -2.755319595336914, "loss": 1.0498, "step": 609 }, { "beta_dpo/beta": 0.5392435193061829, "beta_dpo/beta_margin_grad_mean": -0.13582731783390045, "beta_dpo/beta_margin_grad_std": 0.33346015214920044, "beta_dpo/beta_margin_mean": 214.91415405273438, "beta_dpo/beta_margin_std": 323.2533874511719, "beta_dpo/beta_used": 0.5392435193061829, "beta_dpo/beta_used_raw": 0.5392435193061829, "beta_dpo/gap_mean": 417.583740234375, "beta_dpo/gap_std": 368.72650146484375, "beta_dpo/loss_margin_mean": 401.9151916503906, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.895741556534508, "grad_norm": 11989.935546875, "learning_rate": 1.6881942648911074e-08, "logits/chosen": -2.679978370666504, "logits/rejected": -2.7630457878112793, "loss": 2.1845, "step": 610 }, { "beta_dpo/beta": 0.9980809688568115, "beta_dpo/beta_margin_grad_mean": -0.24803996086120605, "beta_dpo/beta_margin_grad_std": 0.261429101228714, "beta_dpo/beta_margin_mean": 542.123046875, "beta_dpo/beta_margin_std": 784.00927734375, "beta_dpo/beta_used": 0.9980809688568115, "beta_dpo/beta_used_raw": -1.0866873264312744, "beta_dpo/gap_mean": 409.5821838378906, "beta_dpo/gap_std": 377.1724853515625, "beta_dpo/loss_margin_mean": 410.0389709472656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8972099853157122, "grad_norm": 32357.361328125, "learning_rate": 1.6421423736208e-08, "logits/chosen": -2.6723227500915527, "logits/rejected": -2.769993305206299, "loss": 16.8196, "step": 611 }, { "beta_dpo/beta": 2.3185596466064453, "beta_dpo/beta_margin_grad_mean": -0.23413825035095215, "beta_dpo/beta_margin_grad_std": 0.24901816248893738, "beta_dpo/beta_margin_mean": 1160.555908203125, "beta_dpo/beta_margin_std": 1747.4896240234375, "beta_dpo/beta_used": 2.3185596466064453, "beta_dpo/beta_used_raw": -0.5901319980621338, "beta_dpo/gap_mean": 418.91546630859375, "beta_dpo/gap_std": 378.8981628417969, "beta_dpo/loss_margin_mean": 448.5855407714844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.8986784140969163, "grad_norm": 61496.91796875, "learning_rate": 1.5967059836219042e-08, "logits/chosen": -2.6705970764160156, "logits/rejected": -2.763137102127075, "loss": 96.682, "step": 612 }, { "beta_dpo/beta": 2.020970344543457, "beta_dpo/beta_margin_grad_mean": -0.19436858594417572, "beta_dpo/beta_margin_grad_std": 0.20417682826519012, "beta_dpo/beta_margin_mean": 1071.8077392578125, "beta_dpo/beta_margin_std": 1328.2359619140625, "beta_dpo/beta_used": 2.020970344543457, "beta_dpo/beta_used_raw": 1.5588773488998413, "beta_dpo/gap_mean": 433.0877685546875, "beta_dpo/gap_std": 376.868408203125, "beta_dpo/loss_margin_mean": 500.5608825683594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9001468428781204, "grad_norm": 18.278871536254883, "learning_rate": 1.551886292185553e-08, "logits/chosen": -2.6771979331970215, "logits/rejected": -2.7864818572998047, "loss": 0.5219, "step": 613 }, { "beta_dpo/beta": 0.24590833485126495, "beta_dpo/beta_margin_grad_mean": -0.2588002383708954, "beta_dpo/beta_margin_grad_std": 0.2748665511608124, "beta_dpo/beta_margin_mean": 115.03885650634766, "beta_dpo/beta_margin_std": 184.78355407714844, "beta_dpo/beta_used": 0.24590833485126495, "beta_dpo/beta_used_raw": -0.8744754195213318, "beta_dpo/gap_mean": 440.6515197753906, "beta_dpo/gap_std": 371.03680419921875, "beta_dpo/loss_margin_mean": 458.093017578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9016152716593245, "grad_norm": 4608.9931640625, "learning_rate": 1.507684480352292e-08, "logits/chosen": -2.6669607162475586, "logits/rejected": -2.793607234954834, "loss": 4.2601, "step": 614 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40423691272735596, "beta_dpo/beta_margin_grad_std": 0.09169073402881622, "beta_dpo/beta_margin_mean": 0.405151903629303, "beta_dpo/beta_margin_std": 0.404344379901886, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.9845547676086426, "beta_dpo/gap_mean": 434.8990173339844, "beta_dpo/gap_std": 375.1392822265625, "beta_dpo/loss_margin_mean": 405.1518859863281, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9030837004405287, "grad_norm": 32.88970184326172, "learning_rate": 1.4641017128809801e-08, "logits/chosen": -2.6823935508728027, "logits/rejected": -2.760178327560425, "loss": 1.0597, "step": 615 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4224950969219208, "beta_dpo/beta_margin_grad_std": 0.09631507843732834, "beta_dpo/beta_margin_mean": 0.32495495676994324, "beta_dpo/beta_margin_std": 0.4116884171962738, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -6.70911169052124, "beta_dpo/gap_mean": 418.6253662109375, "beta_dpo/gap_std": 380.8779296875, "beta_dpo/loss_margin_mean": 324.9549255371094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9045521292217328, "grad_norm": 30.849700927734375, "learning_rate": 1.4211391382180637e-08, "logits/chosen": -2.723806381225586, "logits/rejected": -2.776388168334961, "loss": 1.125, "step": 616 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.43244972825050354, "beta_dpo/beta_margin_grad_std": 0.0854559987783432, "beta_dpo/beta_margin_mean": 0.28103265166282654, "beta_dpo/beta_margin_std": 0.3670780658721924, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -5.807326793670654, "beta_dpo/gap_mean": 394.8181457519531, "beta_dpo/gap_std": 384.7374267578125, "beta_dpo/loss_margin_mean": 281.0326232910156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9060205580029369, "grad_norm": 39.271549224853516, "learning_rate": 1.378797888467345e-08, "logits/chosen": -2.678978443145752, "logits/rejected": -2.696861982345581, "loss": 1.1238, "step": 617 }, { "beta_dpo/beta": 3.199589967727661, "beta_dpo/beta_margin_grad_mean": -0.255416601896286, "beta_dpo/beta_margin_grad_std": 0.2696244418621063, "beta_dpo/beta_margin_mean": 1545.2900390625, "beta_dpo/beta_margin_std": 2523.5302734375, "beta_dpo/beta_used": 3.199589967727661, "beta_dpo/beta_used_raw": 2.9987692832946777, "beta_dpo/gap_mean": 389.8605041503906, "beta_dpo/gap_std": 387.74383544921875, "beta_dpo/loss_margin_mean": 414.118408203125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9074889867841409, "grad_norm": 48797.57421875, "learning_rate": 1.3370790793601371e-08, "logits/chosen": -2.6983931064605713, "logits/rejected": -2.777221918106079, "loss": 29.3627, "step": 618 }, { "beta_dpo/beta": 1.616031527519226, "beta_dpo/beta_margin_grad_mean": -0.21139781177043915, "beta_dpo/beta_margin_grad_std": 0.22920767962932587, "beta_dpo/beta_margin_mean": 684.386962890625, "beta_dpo/beta_margin_std": 936.7298583984375, "beta_dpo/beta_used": 1.616031527519226, "beta_dpo/beta_used_raw": 1.383022665977478, "beta_dpo/gap_mean": 402.82708740234375, "beta_dpo/gap_std": 391.6029052734375, "beta_dpo/loss_margin_mean": 444.0133972167969, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.908957415565345, "grad_norm": 21.482271194458008, "learning_rate": 1.2959838102258535e-08, "logits/chosen": -2.6790366172790527, "logits/rejected": -2.7796411514282227, "loss": 0.5328, "step": 619 }, { "beta_dpo/beta": 0.5794578790664673, "beta_dpo/beta_margin_grad_mean": -0.27765581011772156, "beta_dpo/beta_margin_grad_std": 0.2895503342151642, "beta_dpo/beta_margin_mean": 221.4326934814453, "beta_dpo/beta_margin_std": 396.32781982421875, "beta_dpo/beta_used": 0.5794578790664673, "beta_dpo/beta_used_raw": -1.1368510723114014, "beta_dpo/gap_mean": 396.19281005859375, "beta_dpo/gap_std": 392.14300537109375, "beta_dpo/loss_margin_mean": 336.462646484375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9104258443465492, "grad_norm": 11751.0576171875, "learning_rate": 1.2555131639630567e-08, "logits/chosen": -2.7585866451263428, "logits/rejected": -2.8155956268310547, "loss": 1.5936, "step": 620 }, { "beta_dpo/beta": 1.6563297510147095, "beta_dpo/beta_margin_grad_mean": -0.12106994539499283, "beta_dpo/beta_margin_grad_std": 0.32161012291908264, "beta_dpo/beta_margin_mean": 764.5799560546875, "beta_dpo/beta_margin_std": 776.7243041992188, "beta_dpo/beta_used": 1.6563297510147095, "beta_dpo/beta_used_raw": 1.6563297510147095, "beta_dpo/gap_mean": 396.5146789550781, "beta_dpo/gap_std": 397.0887451171875, "beta_dpo/loss_margin_mean": 447.8368225097656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9118942731277533, "grad_norm": 31736.255859375, "learning_rate": 1.2156682070109086e-08, "logits/chosen": -2.6869962215423584, "logits/rejected": -2.7847015857696533, "loss": 14.073, "step": 621 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.41488850116729736, "beta_dpo/beta_margin_grad_std": 0.07999890297651291, "beta_dpo/beta_margin_mean": 0.355562299489975, "beta_dpo/beta_margin_std": 0.34979644417762756, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.0101447105407715, "beta_dpo/gap_mean": 397.34625244140625, "beta_dpo/gap_std": 393.946533203125, "beta_dpo/loss_margin_mean": 355.5622863769531, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9133627019089574, "grad_norm": 31.479806900024414, "learning_rate": 1.1764499893210878e-08, "logits/chosen": -2.645836353302002, "logits/rejected": -2.7034900188446045, "loss": 1.1033, "step": 622 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4140608608722687, "beta_dpo/beta_margin_grad_std": 0.09136771410703659, "beta_dpo/beta_margin_mean": 0.3609658181667328, "beta_dpo/beta_margin_std": 0.39438700675964355, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.0260062217712402, "beta_dpo/gap_mean": 391.07568359375, "beta_dpo/gap_std": 391.4143981933594, "beta_dpo/loss_margin_mean": 360.9658203125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9148311306901615, "grad_norm": 28.536666870117188, "learning_rate": 1.1378595443300998e-08, "logits/chosen": -2.684234619140625, "logits/rejected": -2.754396677017212, "loss": 1.0963, "step": 623 }, { "beta_dpo/beta": 2.6343820095062256, "beta_dpo/beta_margin_grad_mean": -0.24445195496082306, "beta_dpo/beta_margin_grad_std": 0.26650404930114746, "beta_dpo/beta_margin_mean": 1267.75537109375, "beta_dpo/beta_margin_std": 1948.750732421875, "beta_dpo/beta_used": 2.6343820095062256, "beta_dpo/beta_used_raw": 0.8192785978317261, "beta_dpo/gap_mean": 396.8379821777344, "beta_dpo/gap_std": 400.57037353515625, "beta_dpo/loss_margin_mean": 465.2864990234375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9162995594713657, "grad_norm": 90469.3125, "learning_rate": 1.0998978889320582e-08, "logits/chosen": -2.734670400619507, "logits/rejected": -2.8197638988494873, "loss": 3.3046, "step": 624 }, { "beta_dpo/beta": 1.0586000680923462, "beta_dpo/beta_margin_grad_mean": -0.23547331988811493, "beta_dpo/beta_margin_grad_std": 0.24736563861370087, "beta_dpo/beta_margin_mean": 472.75970458984375, "beta_dpo/beta_margin_std": 665.16015625, "beta_dpo/beta_used": 1.0586000680923462, "beta_dpo/beta_used_raw": -0.7660439014434814, "beta_dpo/gap_mean": 401.3667297363281, "beta_dpo/gap_std": 384.7314453125, "beta_dpo/loss_margin_mean": 412.4743957519531, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9177679882525698, "grad_norm": 26121.599609375, "learning_rate": 1.0625660234518913e-08, "logits/chosen": -2.6922459602355957, "logits/rejected": -2.767746925354004, "loss": 17.1082, "step": 625 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40667426586151123, "beta_dpo/beta_margin_grad_std": 0.09596937894821167, "beta_dpo/beta_margin_mean": 0.39498090744018555, "beta_dpo/beta_margin_std": 0.4217798709869385, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.5139915943145752, "beta_dpo/gap_mean": 402.25384521484375, "beta_dpo/gap_std": 386.7181091308594, "beta_dpo/loss_margin_mean": 394.98089599609375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9192364170337739, "grad_norm": 34.07730484008789, "learning_rate": 1.0258649316189721e-08, "logits/chosen": -2.66159725189209, "logits/rejected": -2.7381627559661865, "loss": 1.0654, "step": 626 }, { "beta_dpo/beta": 3.29160737991333, "beta_dpo/beta_margin_grad_mean": -0.07827726006507874, "beta_dpo/beta_margin_grad_std": 0.26832348108291626, "beta_dpo/beta_margin_mean": 1679.4234619140625, "beta_dpo/beta_margin_std": 2345.806884765625, "beta_dpo/beta_used": 3.29160737991333, "beta_dpo/beta_used_raw": 3.29160737991333, "beta_dpo/gap_mean": 412.205810546875, "beta_dpo/gap_std": 392.6907958984375, "beta_dpo/loss_margin_mean": 481.80487060546875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.920704845814978, "grad_norm": 38.47300720214844, "learning_rate": 9.897955805412e-09, "logits/chosen": -2.662278413772583, "logits/rejected": -2.77905011177063, "loss": 0.0004, "step": 627 }, { "beta_dpo/beta": 0.16923931241035461, "beta_dpo/beta_margin_grad_mean": -0.25576457381248474, "beta_dpo/beta_margin_grad_std": 0.2738649845123291, "beta_dpo/beta_margin_mean": 74.43370819091797, "beta_dpo/beta_margin_std": 140.65432739257812, "beta_dpo/beta_used": 0.16923931241035461, "beta_dpo/beta_used_raw": -0.19239750504493713, "beta_dpo/gap_mean": 422.8117370605469, "beta_dpo/gap_std": 404.112548828125, "beta_dpo/loss_margin_mean": 459.7860107421875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.922173274596182, "grad_norm": 5785.26611328125, "learning_rate": 9.543589206795238e-09, "logits/chosen": -2.7000410556793213, "logits/rejected": -2.7991576194763184, "loss": 6.3099, "step": 628 }, { "beta_dpo/beta": 0.24494610726833344, "beta_dpo/beta_margin_grad_mean": -0.25123071670532227, "beta_dpo/beta_margin_grad_std": 0.26800966262817383, "beta_dpo/beta_margin_mean": 103.84109497070312, "beta_dpo/beta_margin_std": 145.34765625, "beta_dpo/beta_used": 0.24494610726833344, "beta_dpo/beta_used_raw": -1.8342968225479126, "beta_dpo/gap_mean": 422.31903076171875, "beta_dpo/gap_std": 400.2264404296875, "beta_dpo/loss_margin_mean": 406.9854736328125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9236417033773862, "grad_norm": 3084.765380859375, "learning_rate": 9.19555885822887e-09, "logits/chosen": -2.698057174682617, "logits/rejected": -2.772855281829834, "loss": 1.5217, "step": 629 }, { "beta_dpo/beta": 0.010254332795739174, "beta_dpo/beta_margin_grad_mean": -0.2662602663040161, "beta_dpo/beta_margin_grad_std": 0.2255883365869522, "beta_dpo/beta_margin_mean": 4.899608135223389, "beta_dpo/beta_margin_std": 7.351168155670166, "beta_dpo/beta_used": 0.010254332795739174, "beta_dpo/beta_used_raw": -1.358794927597046, "beta_dpo/gap_mean": 418.43365478515625, "beta_dpo/gap_std": 380.6859436035156, "beta_dpo/loss_margin_mean": 416.1902160644531, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9251101321585903, "grad_norm": 128.09698486328125, "learning_rate": 8.85387393063622e-09, "logits/chosen": -2.66994571685791, "logits/rejected": -2.7564034461975098, "loss": 0.6538, "step": 630 }, { "beta_dpo/beta": 0.572502613067627, "beta_dpo/beta_margin_grad_mean": -0.25285497307777405, "beta_dpo/beta_margin_grad_std": 0.25762128829956055, "beta_dpo/beta_margin_mean": 263.9234619140625, "beta_dpo/beta_margin_std": 379.5833435058594, "beta_dpo/beta_used": 0.572502613067627, "beta_dpo/beta_used_raw": -2.5124354362487793, "beta_dpo/gap_mean": 411.2584228515625, "beta_dpo/gap_std": 380.47344970703125, "beta_dpo/loss_margin_mean": 375.1467590332031, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9265785609397944, "grad_norm": 6237.07177734375, "learning_rate": 8.518543427732949e-09, "logits/chosen": -2.7391209602355957, "logits/rejected": -2.8207530975341797, "loss": 0.5817, "step": 631 }, { "beta_dpo/beta": 0.6863118410110474, "beta_dpo/beta_margin_grad_mean": -0.2555917203426361, "beta_dpo/beta_margin_grad_std": 0.2693924009799957, "beta_dpo/beta_margin_mean": 278.4541931152344, "beta_dpo/beta_margin_std": 425.826416015625, "beta_dpo/beta_used": 0.6863118410110474, "beta_dpo/beta_used_raw": -3.893587350845337, "beta_dpo/gap_mean": 409.9851989746094, "beta_dpo/gap_std": 375.94549560546875, "beta_dpo/loss_margin_mean": 378.54730224609375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9280469897209985, "grad_norm": 13932.6142578125, "learning_rate": 8.189576185789637e-09, "logits/chosen": -2.712128162384033, "logits/rejected": -2.77974009513855, "loss": 20.7515, "step": 632 }, { "beta_dpo/beta": 0.8220680356025696, "beta_dpo/beta_margin_grad_mean": -0.2414667308330536, "beta_dpo/beta_margin_grad_std": 0.25375181436538696, "beta_dpo/beta_margin_mean": 374.5078125, "beta_dpo/beta_margin_std": 545.6969604492188, "beta_dpo/beta_used": 0.8220680356025696, "beta_dpo/beta_used_raw": -0.9592093229293823, "beta_dpo/gap_mean": 402.13140869140625, "beta_dpo/gap_std": 372.6382141113281, "beta_dpo/loss_margin_mean": 393.3555908203125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9295154185022027, "grad_norm": 38270.734375, "learning_rate": 7.866980873399015e-09, "logits/chosen": -2.7510485649108887, "logits/rejected": -2.832095146179199, "loss": 33.6173, "step": 633 }, { "beta_dpo/beta": 1.9843794107437134, "beta_dpo/beta_margin_grad_mean": -0.23467524349689484, "beta_dpo/beta_margin_grad_std": 0.25170350074768066, "beta_dpo/beta_margin_mean": 1058.7847900390625, "beta_dpo/beta_margin_std": 1698.30029296875, "beta_dpo/beta_used": 1.9843794107437134, "beta_dpo/beta_used_raw": -1.2706066370010376, "beta_dpo/gap_mean": 416.826171875, "beta_dpo/gap_std": 384.57305908203125, "beta_dpo/loss_margin_mean": 465.8406982421875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9309838472834068, "grad_norm": 23538.87109375, "learning_rate": 7.550765991247654e-09, "logits/chosen": -2.7076034545898438, "logits/rejected": -2.8257126808166504, "loss": 3.3576, "step": 634 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4157167077064514, "beta_dpo/beta_margin_grad_std": 0.08389006555080414, "beta_dpo/beta_margin_mean": 0.3560367524623871, "beta_dpo/beta_margin_std": 0.37797629833221436, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -5.190537452697754, "beta_dpo/gap_mean": 407.5589904785156, "beta_dpo/gap_std": 385.02099609375, "beta_dpo/loss_margin_mean": 356.0367431640625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9324522760646109, "grad_norm": 27.796987533569336, "learning_rate": 7.240939871891699e-09, "logits/chosen": -2.733865261077881, "logits/rejected": -2.7787771224975586, "loss": 1.1092, "step": 635 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4153151214122772, "beta_dpo/beta_margin_grad_std": 0.08367358148097992, "beta_dpo/beta_margin_mean": 0.3526773154735565, "beta_dpo/beta_margin_std": 0.35728541016578674, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.042080402374268, "beta_dpo/gap_mean": 395.2179870605469, "beta_dpo/gap_std": 383.86175537109375, "beta_dpo/loss_margin_mean": 352.67730712890625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.933920704845815, "grad_norm": 32.076175689697266, "learning_rate": 6.937510679537628e-09, "logits/chosen": -2.7010793685913086, "logits/rejected": -2.7489147186279297, "loss": 1.1087, "step": 636 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.39836955070495605, "beta_dpo/beta_margin_grad_std": 0.09108960628509521, "beta_dpo/beta_margin_mean": 0.43129873275756836, "beta_dpo/beta_margin_std": 0.4063197672367096, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.6715822219848633, "beta_dpo/gap_mean": 400.29400634765625, "beta_dpo/gap_std": 386.3330993652344, "beta_dpo/loss_margin_mean": 431.2987060546875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9353891336270191, "grad_norm": 30.193471908569336, "learning_rate": 6.640486409826785e-09, "logits/chosen": -2.697937250137329, "logits/rejected": -2.79691219329834, "loss": 1.098, "step": 637 }, { "beta_dpo/beta": 1.0422019958496094, "beta_dpo/beta_margin_grad_mean": -0.2520017921924591, "beta_dpo/beta_margin_grad_std": 0.26553109288215637, "beta_dpo/beta_margin_mean": 457.24200439453125, "beta_dpo/beta_margin_std": 703.8070678710938, "beta_dpo/beta_used": 1.0422019958496094, "beta_dpo/beta_used_raw": -1.928234577178955, "beta_dpo/gap_mean": 401.11383056640625, "beta_dpo/gap_std": 381.48553466796875, "beta_dpo/loss_margin_mean": 406.8208923339844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9368575624082232, "grad_norm": 4143.97802734375, "learning_rate": 6.349874889624962e-09, "logits/chosen": -2.663649320602417, "logits/rejected": -2.739919662475586, "loss": 2.9598, "step": 638 }, { "beta_dpo/beta": 0.8174912929534912, "beta_dpo/beta_margin_grad_mean": -0.27413713932037354, "beta_dpo/beta_margin_grad_std": 0.2775159478187561, "beta_dpo/beta_margin_mean": 371.14276123046875, "beta_dpo/beta_margin_std": 529.0587768554688, "beta_dpo/beta_used": 0.8174912929534912, "beta_dpo/beta_used_raw": -3.1657114028930664, "beta_dpo/gap_mean": 384.8955078125, "beta_dpo/gap_std": 367.0290832519531, "beta_dpo/loss_margin_mean": 319.404052734375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9383259911894273, "grad_norm": 12486.169921875, "learning_rate": 6.065683776815933e-09, "logits/chosen": -2.637770652770996, "logits/rejected": -2.6620218753814697, "loss": 5.479, "step": 639 }, { "beta_dpo/beta": 0.4668891727924347, "beta_dpo/beta_margin_grad_mean": -0.19809961318969727, "beta_dpo/beta_margin_grad_std": 0.21003539860248566, "beta_dpo/beta_margin_mean": 256.6402893066406, "beta_dpo/beta_margin_std": 346.9462890625, "beta_dpo/beta_used": 0.4668891727924347, "beta_dpo/beta_used_raw": 0.036036670207977295, "beta_dpo/gap_mean": 402.61187744140625, "beta_dpo/gap_std": 376.2877197265625, "beta_dpo/loss_margin_mean": 500.78619384765625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9397944199706314, "grad_norm": 27.00104522705078, "learning_rate": 5.7879205600998296e-09, "logits/chosen": -2.6643662452697754, "logits/rejected": -2.7640390396118164, "loss": 0.5343, "step": 640 }, { "beta_dpo/beta": 1.7024071216583252, "beta_dpo/beta_margin_grad_mean": -0.2545595169067383, "beta_dpo/beta_margin_grad_std": 0.2657167315483093, "beta_dpo/beta_margin_mean": 737.9752197265625, "beta_dpo/beta_margin_std": 1113.3966064453125, "beta_dpo/beta_used": 1.7024071216583252, "beta_dpo/beta_used_raw": -0.05958735942840576, "beta_dpo/gap_mean": 413.01031494140625, "beta_dpo/gap_std": 378.18206787109375, "beta_dpo/loss_margin_mean": 426.4712829589844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9412628487518355, "grad_norm": 37319.4296875, "learning_rate": 5.516592558795746e-09, "logits/chosen": -2.7087669372558594, "logits/rejected": -2.7869772911071777, "loss": 33.9035, "step": 641 }, { "beta_dpo/beta": 2.8232624530792236, "beta_dpo/beta_margin_grad_mean": -0.20511434972286224, "beta_dpo/beta_margin_grad_std": 0.21880197525024414, "beta_dpo/beta_margin_mean": 1559.873291015625, "beta_dpo/beta_margin_std": 2155.26123046875, "beta_dpo/beta_used": 2.8232624530792236, "beta_dpo/beta_used_raw": 0.7488954067230225, "beta_dpo/gap_mean": 417.9560852050781, "beta_dpo/gap_std": 394.83203125, "beta_dpo/loss_margin_mean": 470.5616455078125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9427312775330396, "grad_norm": 23.94333839416504, "learning_rate": 5.251706922648868e-09, "logits/chosen": -2.6790223121643066, "logits/rejected": -2.778899908065796, "loss": 0.554, "step": 642 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40317976474761963, "beta_dpo/beta_margin_grad_std": 0.07390031218528748, "beta_dpo/beta_margin_mean": 0.40420055389404297, "beta_dpo/beta_margin_std": 0.3269351124763489, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.0951257944107056, "beta_dpo/gap_mean": 421.8905944824219, "beta_dpo/gap_std": 384.7071228027344, "beta_dpo/loss_margin_mean": 404.2005310058594, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9441997063142438, "grad_norm": 28.286523818969727, "learning_rate": 4.993270631642038e-09, "logits/chosen": -2.6818013191223145, "logits/rejected": -2.7871358394622803, "loss": 1.0385, "step": 643 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4079779386520386, "beta_dpo/beta_margin_grad_std": 0.08079829812049866, "beta_dpo/beta_margin_mean": 0.38570505380630493, "beta_dpo/beta_margin_std": 0.35576584935188293, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -1.0808398723602295, "beta_dpo/gap_mean": 418.0806884765625, "beta_dpo/gap_std": 381.2316589355469, "beta_dpo/loss_margin_mean": 385.7050476074219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9456681350954479, "grad_norm": 33.19130325317383, "learning_rate": 4.741290495811873e-09, "logits/chosen": -2.66062068939209, "logits/rejected": -2.7422003746032715, "loss": 1.0446, "step": 644 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4123542308807373, "beta_dpo/beta_margin_grad_std": 0.09075028449296951, "beta_dpo/beta_margin_mean": 0.37113770842552185, "beta_dpo/beta_margin_std": 0.40058884024620056, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.672241687774658, "beta_dpo/gap_mean": 409.40924072265625, "beta_dpo/gap_std": 381.2076416015625, "beta_dpo/loss_margin_mean": 371.1376953125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.947136563876652, "grad_norm": 28.928796768188477, "learning_rate": 4.495773155069299e-09, "logits/chosen": -2.667562484741211, "logits/rejected": -2.756563663482666, "loss": 1.1056, "step": 645 }, { "beta_dpo/beta": 0.956828236579895, "beta_dpo/beta_margin_grad_mean": -0.24536260962486267, "beta_dpo/beta_margin_grad_std": 0.25645798444747925, "beta_dpo/beta_margin_mean": 412.48236083984375, "beta_dpo/beta_margin_std": 568.2828979492188, "beta_dpo/beta_used": 0.956828236579895, "beta_dpo/beta_used_raw": -0.878277599811554, "beta_dpo/gap_mean": 402.3372802734375, "beta_dpo/gap_std": 367.37091064453125, "beta_dpo/loss_margin_mean": 363.10498046875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9486049926578561, "grad_norm": 9614.8828125, "learning_rate": 4.256725079024553e-09, "logits/chosen": -2.677600383758545, "logits/rejected": -2.7275893688201904, "loss": 6.6603, "step": 646 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4117124080657959, "beta_dpo/beta_margin_grad_std": 0.06875015795230865, "beta_dpo/beta_margin_mean": 0.3639964163303375, "beta_dpo/beta_margin_std": 0.28941068053245544, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.700502395629883, "beta_dpo/gap_mean": 389.188232421875, "beta_dpo/gap_std": 354.70404052734375, "beta_dpo/loss_margin_mean": 363.99639892578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9500734214390602, "grad_norm": 31.953645706176758, "learning_rate": 4.024152566816791e-09, "logits/chosen": -2.6596388816833496, "logits/rejected": -2.7517409324645996, "loss": 1.1028, "step": 647 }, { "beta_dpo/beta": 0.5703181624412537, "beta_dpo/beta_margin_grad_mean": -0.24799123406410217, "beta_dpo/beta_margin_grad_std": 0.264789342880249, "beta_dpo/beta_margin_mean": 255.6763916015625, "beta_dpo/beta_margin_std": 416.8722229003906, "beta_dpo/beta_used": 0.5703181624412537, "beta_dpo/beta_used_raw": -1.4812275171279907, "beta_dpo/gap_mean": 395.81396484375, "beta_dpo/gap_std": 356.5727233886719, "beta_dpo/loss_margin_mean": 429.5019836425781, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9515418502202643, "grad_norm": 27103.84375, "learning_rate": 3.798061746947995e-09, "logits/chosen": -2.6732091903686523, "logits/rejected": -2.8048720359802246, "loss": 31.5114, "step": 648 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4284921884536743, "beta_dpo/beta_margin_grad_std": 0.09621818363666534, "beta_dpo/beta_margin_mean": 0.3019213378429413, "beta_dpo/beta_margin_std": 0.4140036404132843, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.014761924743652, "beta_dpo/gap_mean": 385.2996826171875, "beta_dpo/gap_std": 365.2505798339844, "beta_dpo/loss_margin_mean": 301.92132568359375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9530102790014684, "grad_norm": 31.426589965820312, "learning_rate": 3.5784585771215235e-09, "logits/chosen": -2.712721347808838, "logits/rejected": -2.750772476196289, "loss": 1.1226, "step": 649 }, { "beta_dpo/beta": 2.429518222808838, "beta_dpo/beta_margin_grad_mean": -0.15641333162784576, "beta_dpo/beta_margin_grad_std": 0.3630194067955017, "beta_dpo/beta_margin_mean": 1180.8328857421875, "beta_dpo/beta_margin_std": 1849.820556640625, "beta_dpo/beta_used": 2.429518222808838, "beta_dpo/beta_used_raw": 2.429518222808838, "beta_dpo/gap_mean": 391.6793212890625, "beta_dpo/gap_std": 385.2804870605469, "beta_dpo/loss_margin_mean": 466.9042663574219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9544787077826725, "grad_norm": 76912.0078125, "learning_rate": 3.3653488440851253e-09, "logits/chosen": -2.6398544311523438, "logits/rejected": -2.746957540512085, "loss": 46.2267, "step": 650 }, { "beta_dpo/beta": 2.192887306213379, "beta_dpo/beta_margin_grad_mean": -0.14100220799446106, "beta_dpo/beta_margin_grad_std": 0.33823278546333313, "beta_dpo/beta_margin_mean": 1079.196533203125, "beta_dpo/beta_margin_std": 1474.2933349609375, "beta_dpo/beta_used": 2.192887306213379, "beta_dpo/beta_used_raw": 2.192887306213379, "beta_dpo/gap_mean": 405.810546875, "beta_dpo/gap_std": 392.9246826171875, "beta_dpo/loss_margin_mean": 439.4105529785156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9559471365638766, "grad_norm": 72114.8203125, "learning_rate": 3.158738163478475e-09, "logits/chosen": -2.726027011871338, "logits/rejected": -2.8322134017944336, "loss": 14.6975, "step": 651 }, { "beta_dpo/beta": 3.001312732696533, "beta_dpo/beta_margin_grad_mean": -0.15693050622940063, "beta_dpo/beta_margin_grad_std": 0.3628394305706024, "beta_dpo/beta_margin_mean": 1289.045654296875, "beta_dpo/beta_margin_std": 2198.432861328125, "beta_dpo/beta_used": 3.001312732696533, "beta_dpo/beta_used_raw": 3.001312732696533, "beta_dpo/gap_mean": 408.07373046875, "beta_dpo/gap_std": 401.121337890625, "beta_dpo/loss_margin_mean": 430.41156005859375, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9574155653450808, "grad_norm": 58857.55859375, "learning_rate": 2.9586319796851555e-09, "logits/chosen": -2.7092862129211426, "logits/rejected": -2.830409049987793, "loss": 64.9825, "step": 652 }, { "beta_dpo/beta": 1.4299049377441406, "beta_dpo/beta_margin_grad_mean": -0.23352642357349396, "beta_dpo/beta_margin_grad_std": 0.25692373514175415, "beta_dpo/beta_margin_mean": 643.51318359375, "beta_dpo/beta_margin_std": 985.338623046875, "beta_dpo/beta_used": 1.4299049377441406, "beta_dpo/beta_used_raw": 0.3514312505722046, "beta_dpo/gap_mean": 420.1622314453125, "beta_dpo/gap_std": 404.7218933105469, "beta_dpo/loss_margin_mean": 500.4862365722656, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9588839941262849, "grad_norm": 35622.0390625, "learning_rate": 2.7650355656892166e-09, "logits/chosen": -2.677021026611328, "logits/rejected": -2.806358814239502, "loss": 23.9097, "step": 653 }, { "beta_dpo/beta": 0.9755141735076904, "beta_dpo/beta_margin_grad_mean": -0.2645564079284668, "beta_dpo/beta_margin_grad_std": 0.2780976891517639, "beta_dpo/beta_margin_mean": 399.7411804199219, "beta_dpo/beta_margin_std": 707.6503295898438, "beta_dpo/beta_used": 0.9755141735076904, "beta_dpo/beta_used_raw": -0.04932612180709839, "beta_dpo/gap_mean": 423.97796630859375, "beta_dpo/gap_std": 405.8636474609375, "beta_dpo/loss_margin_mean": 406.8052062988281, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.960352422907489, "grad_norm": 18767.052734375, "learning_rate": 2.577954022936174e-09, "logits/chosen": -2.706268787384033, "logits/rejected": -2.816140651702881, "loss": 4.9214, "step": 654 }, { "beta_dpo/beta": 1.0244628190994263, "beta_dpo/beta_margin_grad_mean": -0.25356224179267883, "beta_dpo/beta_margin_grad_std": 0.2714630663394928, "beta_dpo/beta_margin_mean": 510.8900451660156, "beta_dpo/beta_margin_std": 796.9216918945312, "beta_dpo/beta_used": 1.0244628190994263, "beta_dpo/beta_used_raw": 0.2029227614402771, "beta_dpo/gap_mean": 427.8386535644531, "beta_dpo/gap_std": 412.89520263671875, "beta_dpo/loss_margin_mean": 435.32635498046875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9618208516886931, "grad_norm": 18119.2421875, "learning_rate": 2.397392281198729e-09, "logits/chosen": -2.680917263031006, "logits/rejected": -2.7932283878326416, "loss": 3.7187, "step": 655 }, { "beta_dpo/beta": 2.7679896354675293, "beta_dpo/beta_margin_grad_mean": -0.08084268122911453, "beta_dpo/beta_margin_grad_std": 0.26834091544151306, "beta_dpo/beta_margin_mean": 1555.604248046875, "beta_dpo/beta_margin_std": 1787.474853515625, "beta_dpo/beta_used": 2.7679896354675293, "beta_dpo/beta_used_raw": 2.7679896354675293, "beta_dpo/gap_mean": 440.5277404785156, "beta_dpo/gap_std": 414.5347900390625, "beta_dpo/loss_margin_mean": 540.3606567382812, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9632892804698973, "grad_norm": 6175.9521484375, "learning_rate": 2.223355098446622e-09, "logits/chosen": -2.654201030731201, "logits/rejected": -2.7860169410705566, "loss": 3.1373, "step": 656 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.38836878538131714, "beta_dpo/beta_margin_grad_std": 0.08058323711156845, "beta_dpo/beta_margin_mean": 0.4706077575683594, "beta_dpo/beta_margin_std": 0.3598029613494873, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.769383430480957, "beta_dpo/gap_mean": 452.3384094238281, "beta_dpo/gap_std": 401.4930725097656, "beta_dpo/loss_margin_mean": 470.60772705078125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9647577092511013, "grad_norm": 29.89715003967285, "learning_rate": 2.055847060721566e-09, "logits/chosen": -2.7141106128692627, "logits/rejected": -2.818988084793091, "loss": 1.0684, "step": 657 }, { "beta_dpo/beta": 0.3125349283218384, "beta_dpo/beta_margin_grad_mean": -0.2526739537715912, "beta_dpo/beta_margin_grad_std": 0.2634952664375305, "beta_dpo/beta_margin_mean": 149.250732421875, "beta_dpo/beta_margin_std": 212.36988830566406, "beta_dpo/beta_used": 0.3125349283218384, "beta_dpo/beta_used_raw": -3.3986754417419434, "beta_dpo/gap_mean": 444.14215087890625, "beta_dpo/gap_std": 383.74365234375, "beta_dpo/loss_margin_mean": 420.6098937988281, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9662261380323054, "grad_norm": 15924.490234375, "learning_rate": 1.8948725820160662e-09, "logits/chosen": -2.6853952407836914, "logits/rejected": -2.801093578338623, "loss": 8.8263, "step": 658 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4057837426662445, "beta_dpo/beta_margin_grad_std": 0.08504177629947662, "beta_dpo/beta_margin_mean": 0.3961847126483917, "beta_dpo/beta_margin_std": 0.37236061692237854, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.6913902759552, "beta_dpo/gap_mean": 439.3583679199219, "beta_dpo/gap_std": 378.472412109375, "beta_dpo/loss_margin_mean": 396.1846923828125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9676945668135095, "grad_norm": 33.702049255371094, "learning_rate": 1.7404359041573723e-09, "logits/chosen": -2.692453145980835, "logits/rejected": -2.7410359382629395, "loss": 1.0658, "step": 659 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.3935113251209259, "beta_dpo/beta_margin_grad_std": 0.07827742397785187, "beta_dpo/beta_margin_mean": 0.4469809830188751, "beta_dpo/beta_margin_std": 0.3453811705112457, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.6651883125305176, "beta_dpo/gap_mean": 438.81976318359375, "beta_dpo/gap_std": 376.5857849121094, "beta_dpo/loss_margin_mean": 446.9809875488281, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9691629955947136, "grad_norm": 25.069368362426758, "learning_rate": 1.592541096695571e-09, "logits/chosen": -2.6554760932922363, "logits/rejected": -2.734593391418457, "loss": 1.0505, "step": 660 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40231382846832275, "beta_dpo/beta_margin_grad_std": 0.0794859230518341, "beta_dpo/beta_margin_mean": 0.4129953980445862, "beta_dpo/beta_margin_std": 0.3699604570865631, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.9038610458374023, "beta_dpo/gap_mean": 431.3531494140625, "beta_dpo/gap_std": 369.5039367675781, "beta_dpo/loss_margin_mean": 412.9953918457031, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9706314243759178, "grad_norm": 28.522525787353516, "learning_rate": 1.4511920567963908e-09, "logits/chosen": -2.6922972202301025, "logits/rejected": -2.764153242111206, "loss": 1.0573, "step": 661 }, { "beta_dpo/beta": 1.1851645708084106, "beta_dpo/beta_margin_grad_mean": -0.10670406371355057, "beta_dpo/beta_margin_grad_std": 0.29752376675605774, "beta_dpo/beta_margin_mean": 517.0401000976562, "beta_dpo/beta_margin_std": 781.3206787109375, "beta_dpo/beta_used": 1.1851645708084106, "beta_dpo/beta_used_raw": 1.1851645708084106, "beta_dpo/gap_mean": 432.52001953125, "beta_dpo/gap_std": 375.65093994140625, "beta_dpo/loss_margin_mean": 417.3674011230469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9720998531571219, "grad_norm": 10333.798828125, "learning_rate": 1.3163925091384532e-09, "logits/chosen": -2.6792619228363037, "logits/rejected": -2.7577905654907227, "loss": 4.327, "step": 662 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.39322903752326965, "beta_dpo/beta_margin_grad_std": 0.10028216242790222, "beta_dpo/beta_margin_mean": 0.46100181341171265, "beta_dpo/beta_margin_std": 0.46238890290260315, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.2035021781921387, "beta_dpo/gap_mean": 433.63494873046875, "beta_dpo/gap_std": 389.35064697265625, "beta_dpo/loss_margin_mean": 461.0018005371094, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.973568281938326, "grad_norm": 36.53968811035156, "learning_rate": 1.1881460058152382e-09, "logits/chosen": -2.7297868728637695, "logits/rejected": -2.8309314250946045, "loss": 1.0669, "step": 663 }, { "beta_dpo/beta": 0.8343223929405212, "beta_dpo/beta_margin_grad_mean": -0.20197950303554535, "beta_dpo/beta_margin_grad_std": 0.22016191482543945, "beta_dpo/beta_margin_mean": 431.966064453125, "beta_dpo/beta_margin_std": 605.8634643554688, "beta_dpo/beta_used": 0.8343223929405212, "beta_dpo/beta_used_raw": -0.03142118453979492, "beta_dpo/gap_mean": 449.9844970703125, "beta_dpo/gap_std": 394.67535400390625, "beta_dpo/loss_margin_mean": 532.53369140625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9750367107195301, "grad_norm": 37235.65625, "learning_rate": 1.066455926241383e-09, "logits/chosen": -2.619297504425049, "logits/rejected": -2.753797769546509, "loss": 5.8712, "step": 664 }, { "beta_dpo/beta": 0.3265571594238281, "beta_dpo/beta_margin_grad_mean": -0.2155420482158661, "beta_dpo/beta_margin_grad_std": 0.23008538782596588, "beta_dpo/beta_margin_mean": 137.3500518798828, "beta_dpo/beta_margin_std": 180.5250701904297, "beta_dpo/beta_used": 0.3265571594238281, "beta_dpo/beta_used_raw": -1.14801025390625, "beta_dpo/gap_mean": 450.2773742675781, "beta_dpo/gap_std": 383.91949462890625, "beta_dpo/loss_margin_mean": 422.32843017578125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9765051395007343, "grad_norm": 31.23434829711914, "learning_rate": 9.513254770636137e-10, "logits/chosen": -2.6968719959259033, "logits/rejected": -2.7933096885681152, "loss": 0.5228, "step": 665 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40940189361572266, "beta_dpo/beta_margin_grad_std": 0.08813069015741348, "beta_dpo/beta_margin_mean": 0.3812684416770935, "beta_dpo/beta_margin_std": 0.38156983256340027, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.0323309898376465, "beta_dpo/gap_mean": 441.28741455078125, "beta_dpo/gap_std": 383.9112548828125, "beta_dpo/loss_margin_mean": 381.2684326171875, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9779735682819384, "grad_norm": 33.36760330200195, "learning_rate": 8.427576920763956e-10, "logits/chosen": -2.680263042449951, "logits/rejected": -2.7376909255981445, "loss": 1.0748, "step": 666 }, { "beta_dpo/beta": 0.045063599944114685, "beta_dpo/beta_margin_grad_mean": -0.26865431666374207, "beta_dpo/beta_margin_grad_std": 0.2749733626842499, "beta_dpo/beta_margin_mean": 20.753093719482422, "beta_dpo/beta_margin_std": 34.60969543457031, "beta_dpo/beta_used": 0.045063599944114685, "beta_dpo/beta_used_raw": -0.0253823921084404, "beta_dpo/gap_mean": 433.1988525390625, "beta_dpo/gap_std": 388.9822998046875, "beta_dpo/loss_margin_mean": 400.0951843261719, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9794419970631424, "grad_norm": 1199.2330322265625, "learning_rate": 7.407554321417764e-10, "logits/chosen": -2.7143611907958984, "logits/rejected": -2.784897804260254, "loss": 0.5765, "step": 667 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4063694179058075, "beta_dpo/beta_margin_grad_std": 0.0748078003525734, "beta_dpo/beta_margin_mean": 0.3895391523838043, "beta_dpo/beta_margin_std": 0.3211991488933563, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.1987266540527344, "beta_dpo/gap_mean": 420.1002197265625, "beta_dpo/gap_std": 376.9314880371094, "beta_dpo/loss_margin_mean": 389.53912353515625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9809104258443465, "grad_norm": 25.63221549987793, "learning_rate": 6.453213851142225e-10, "logits/chosen": -2.7238945960998535, "logits/rejected": -2.78692889213562, "loss": 1.07, "step": 668 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.40521949529647827, "beta_dpo/beta_margin_grad_std": 0.08128622174263, "beta_dpo/beta_margin_mean": 0.3963834047317505, "beta_dpo/beta_margin_std": 0.35128095746040344, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.542426109313965, "beta_dpo/gap_mean": 417.46697998046875, "beta_dpo/gap_std": 373.0389404296875, "beta_dpo/loss_margin_mean": 396.3833923339844, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9823788546255506, "grad_norm": 26.949857711791992, "learning_rate": 5.564580657695939e-10, "logits/chosen": -2.7172703742980957, "logits/rejected": -2.7912344932556152, "loss": 1.0688, "step": 669 }, { "beta_dpo/beta": 0.05918378010392189, "beta_dpo/beta_margin_grad_mean": -0.2792292535305023, "beta_dpo/beta_margin_grad_std": 0.29437491297721863, "beta_dpo/beta_margin_mean": 25.40052604675293, "beta_dpo/beta_margin_std": 43.54753112792969, "beta_dpo/beta_used": 0.05918378010392189, "beta_dpo/beta_used_raw": -2.199822187423706, "beta_dpo/gap_mean": 417.28021240234375, "beta_dpo/gap_std": 383.91400146484375, "beta_dpo/loss_margin_mean": 422.6147155761719, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9838472834067548, "grad_norm": 3242.469482421875, "learning_rate": 4.741678157389739e-10, "logits/chosen": -2.631437063217163, "logits/rejected": -2.709677219390869, "loss": 3.7298, "step": 670 }, { "beta_dpo/beta": 0.2660801410675049, "beta_dpo/beta_margin_grad_mean": -0.2764192223548889, "beta_dpo/beta_margin_grad_std": 0.29103344678878784, "beta_dpo/beta_margin_mean": 98.8425064086914, "beta_dpo/beta_margin_std": 195.78741455078125, "beta_dpo/beta_used": 0.2660801410675049, "beta_dpo/beta_used_raw": -1.3270782232284546, "beta_dpo/gap_mean": 414.47650146484375, "beta_dpo/gap_std": 394.97772216796875, "beta_dpo/loss_margin_mean": 404.5050964355469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9853157121879589, "grad_norm": 9340.4814453125, "learning_rate": 3.9845280344705245e-10, "logits/chosen": -2.6861400604248047, "logits/rejected": -2.7837982177734375, "loss": 6.7776, "step": 671 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4036671817302704, "beta_dpo/beta_margin_grad_std": 0.09355795383453369, "beta_dpo/beta_margin_mean": 0.40639743208885193, "beta_dpo/beta_margin_std": 0.4057061970233917, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -2.5519256591796875, "beta_dpo/gap_mean": 413.4635009765625, "beta_dpo/gap_std": 395.70111083984375, "beta_dpo/loss_margin_mean": 406.3974304199219, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.986784140969163, "grad_norm": 35.250362396240234, "learning_rate": 3.293150240547549e-10, "logits/chosen": -2.658691167831421, "logits/rejected": -2.7492518424987793, "loss": 1.077, "step": 672 }, { "beta_dpo/beta": 0.9086623191833496, "beta_dpo/beta_margin_grad_mean": -0.2269849181175232, "beta_dpo/beta_margin_grad_std": 0.24361680448055267, "beta_dpo/beta_margin_mean": 426.99517822265625, "beta_dpo/beta_margin_std": 589.7709350585938, "beta_dpo/beta_used": 0.9086623191833496, "beta_dpo/beta_used_raw": -1.017942190170288, "beta_dpo/gap_mean": 421.73614501953125, "beta_dpo/gap_std": 390.53887939453125, "beta_dpo/loss_margin_mean": 465.11651611328125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9882525697503671, "grad_norm": 38753.8828125, "learning_rate": 2.6675629940689504e-10, "logits/chosen": -2.6246771812438965, "logits/rejected": -2.71297287940979, "loss": 3.2363, "step": 673 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4113142192363739, "beta_dpo/beta_margin_grad_std": 0.0917758122086525, "beta_dpo/beta_margin_mean": 0.3718118667602539, "beta_dpo/beta_margin_std": 0.39119818806648254, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -0.3440747559070587, "beta_dpo/gap_mean": 413.60247802734375, "beta_dpo/gap_std": 390.3453674316406, "beta_dpo/loss_margin_mean": 371.81182861328125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9897209985315712, "grad_norm": 26.38853645324707, "learning_rate": 2.1077827798404725e-10, "logits/chosen": -2.6841392517089844, "logits/rejected": -2.7508318424224854, "loss": 1.0447, "step": 674 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.39887651801109314, "beta_dpo/beta_margin_grad_std": 0.08341260254383087, "beta_dpo/beta_margin_mean": 0.42533355951309204, "beta_dpo/beta_margin_std": 0.3670670986175537, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -3.083004951477051, "beta_dpo/gap_mean": 418.37017822265625, "beta_dpo/gap_std": 389.00335693359375, "beta_dpo/loss_margin_mean": 425.3335266113281, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9911894273127754, "grad_norm": 22.43824005126953, "learning_rate": 1.6138243485910863e-10, "logits/chosen": -2.626479148864746, "logits/rejected": -2.709594249725342, "loss": 1.0741, "step": 675 }, { "beta_dpo/beta": 1.3021284341812134, "beta_dpo/beta_margin_grad_mean": -0.27147722244262695, "beta_dpo/beta_margin_grad_std": 0.28175440430641174, "beta_dpo/beta_margin_mean": 523.4214477539062, "beta_dpo/beta_margin_std": 893.8363647460938, "beta_dpo/beta_used": 1.3021284341812134, "beta_dpo/beta_used_raw": 0.7559030055999756, "beta_dpo/gap_mean": 408.2674255371094, "beta_dpo/gap_std": 382.36383056640625, "beta_dpo/loss_margin_mean": 371.8644714355469, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9926578560939795, "grad_norm": 25749.53515625, "learning_rate": 1.1857007165852472e-10, "logits/chosen": -2.6717138290405273, "logits/rejected": -2.7089462280273438, "loss": 3.2861, "step": 676 }, { "beta_dpo/beta": 2.582223653793335, "beta_dpo/beta_margin_grad_mean": -0.047779832035303116, "beta_dpo/beta_margin_grad_std": 0.21129241585731506, "beta_dpo/beta_margin_mean": 1187.6051025390625, "beta_dpo/beta_margin_std": 1076.4876708984375, "beta_dpo/beta_used": 2.582223653793335, "beta_dpo/beta_used_raw": 2.582223653793335, "beta_dpo/gap_mean": 415.4775695800781, "beta_dpo/gap_std": 378.42694091796875, "beta_dpo/loss_margin_mean": 460.754150390625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9941262848751835, "grad_norm": 50163.39453125, "learning_rate": 8.23423165278725e-11, "logits/chosen": -2.6218011379241943, "logits/rejected": -2.706326961517334, "loss": 5.7215, "step": 677 }, { "beta_dpo/beta": 0.4946085810661316, "beta_dpo/beta_margin_grad_mean": -0.264606773853302, "beta_dpo/beta_margin_grad_std": 0.2809548079967499, "beta_dpo/beta_margin_mean": 201.0238800048828, "beta_dpo/beta_margin_std": 325.8772277832031, "beta_dpo/beta_used": 0.4946085810661316, "beta_dpo/beta_used_raw": -0.259298712015152, "beta_dpo/gap_mean": 415.98052978515625, "beta_dpo/gap_std": 376.99005126953125, "beta_dpo/loss_margin_mean": 404.6373291015625, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9955947136563876, "grad_norm": 6914.888671875, "learning_rate": 5.270012410216185e-11, "logits/chosen": -2.6387507915496826, "logits/rejected": -2.7189407348632812, "loss": 4.2516, "step": 678 }, { "beta_dpo/beta": 0.0010000000474974513, "beta_dpo/beta_margin_grad_mean": -0.4169081151485443, "beta_dpo/beta_margin_grad_std": 0.07902750372886658, "beta_dpo/beta_margin_mean": 0.34554827213287354, "beta_dpo/beta_margin_std": 0.3380792737007141, "beta_dpo/beta_used": 0.0010000000474974513, "beta_dpo/beta_used_raw": -4.045163631439209, "beta_dpo/gap_mean": 405.5008850097656, "beta_dpo/gap_std": 373.1779479980469, "beta_dpo/loss_margin_mean": 345.5482482910156, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9970631424375918, "grad_norm": 32.21836471557617, "learning_rate": 2.9644275480772416e-11, "logits/chosen": -2.7105813026428223, "logits/rejected": -2.7737274169921875, "loss": 1.0921, "step": 679 }, { "beta_dpo/beta": 2.237010955810547, "beta_dpo/beta_margin_grad_mean": -0.25544458627700806, "beta_dpo/beta_margin_grad_std": 0.27421900629997253, "beta_dpo/beta_margin_mean": 1121.7413330078125, "beta_dpo/beta_margin_std": 1776.8070068359375, "beta_dpo/beta_used": 2.237010955810547, "beta_dpo/beta_used_raw": 1.5813672542572021, "beta_dpo/gap_mean": 415.71636962890625, "beta_dpo/gap_std": 377.8543395996094, "beta_dpo/loss_margin_mean": 493.08428955078125, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 0.9985315712187959, "grad_norm": 100621.9609375, "learning_rate": 1.31753782067201e-11, "logits/chosen": -2.689682722091675, "logits/rejected": -2.7969093322753906, "loss": 49.8028, "step": 680 }, { "beta_dpo/beta": 0.9950853586196899, "beta_dpo/beta_margin_grad_mean": -0.2544940412044525, "beta_dpo/beta_margin_grad_std": 0.2666052281856537, "beta_dpo/beta_margin_mean": 497.1185607910156, "beta_dpo/beta_margin_std": 809.5509643554688, "beta_dpo/beta_used": 0.9950853586196899, "beta_dpo/beta_used_raw": -0.9259518980979919, "beta_dpo/gap_mean": 416.84112548828125, "beta_dpo/gap_std": 377.1964416503906, "beta_dpo/loss_margin_mean": 425.1723327636719, "beta_dpo/mask_keep_frac": 0.78125, "epoch": 1.0, "grad_norm": 16219.0615234375, "learning_rate": 3.2938662507808745e-12, "logits/chosen": -2.7180047035217285, "logits/rejected": -2.8164992332458496, "loss": 11.4723, "step": 681 }, { "epoch": 1.0, "step": 681, "total_flos": 0.0, "train_loss": 4.511853932745809, "train_runtime": 2770.5528, "train_samples_per_second": 15.736, "train_steps_per_second": 0.246 } ], "logging_steps": 1, "max_steps": 681, "num_input_tokens_seen": 0, "num_train_epochs": 1, "save_steps": 200, "stateful_callbacks": { "TrainerControl": { "args": { "should_epoch_stop": false, "should_evaluate": false, "should_log": false, "should_save": true, "should_training_stop": true }, "attributes": {} } }, "total_flos": 0.0, "train_batch_size": 8, "trial_name": null, "trial_params": null }