{ "best_global_step": null, "best_metric": null, "best_model_checkpoint": null, "episode": 119296, "epoch": 0.5110349554489376, "eval_steps": 500, "global_step": 233, "is_hyper_param_search": false, "is_local_process_zero": true, "is_world_process_zero": true, "log_history": [ { "episode": 512, "epoch": 0.0021932830705962986, "eps": 6, "loss/policy_avg": 0.16531690955162048, "lr": 3e-06, "objective/entropy": -55.32527542114258, "objective/kl": 0.2650028467178345, "objective/non_score_reward": -0.013250143267214298, "objective/rlhf_reward": 6.5853118896484375, "objective/scores": 6.598561763763428, "policy/approxkl_avg": 0.09512920677661896, "policy/clipfrac_avg": 0.294921875, "policy/entropy_avg": 0.610973596572876, "step": 0, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 6004, "val/ratio": 0.9999569058418274, "val/ratio_var": 1.1622008742051548e-06 }, { "episode": 1024, "epoch": 0.004386566141192597, "eps": 6, "loss/policy_avg": 0.22872287034988403, "lr": 2.9882812500000002e-06, "objective/entropy": -50.97515106201172, "objective/kl": 0.687268853187561, "objective/non_score_reward": -0.03436344116926193, "objective/rlhf_reward": 6.763760089874268, "objective/scores": 6.798122882843018, "policy/approxkl_avg": 0.11450275778770447, "policy/clipfrac_avg": 0.28125, "policy/entropy_avg": 0.6294225454330444, "step": 1, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 5686, "val/ratio": 0.9997917413711548, "val/ratio_var": 1.6605762311883154e-06 }, { "episode": 1536, "epoch": 0.006579849211788897, "eps": 6, "loss/policy_avg": 0.1424816995859146, "lr": 2.9765625e-06, "objective/entropy": -57.432891845703125, "objective/kl": 1.1181037425994873, "objective/non_score_reward": -0.05590519309043884, "objective/rlhf_reward": 7.33503532409668, "objective/scores": 7.390939712524414, "policy/approxkl_avg": 0.0936519205570221, "policy/clipfrac_avg": 0.322265625, "policy/entropy_avg": 0.5745395421981812, "step": 2, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 5664, "val/ratio": 1.0001490116119385, "val/ratio_var": 1.4631596059189178e-06 }, { "episode": 2048, "epoch": 0.008773132282385195, "eps": 6, "loss/policy_avg": 0.17385584115982056, "lr": 2.96484375e-06, "objective/entropy": -54.846282958984375, "objective/kl": 1.592078447341919, "objective/non_score_reward": -0.07960392534732819, "objective/rlhf_reward": 7.236883163452148, "objective/scores": 7.316486835479736, "policy/approxkl_avg": 0.11385183781385422, "policy/clipfrac_avg": 0.283203125, "policy/entropy_avg": 0.5554953813552856, "step": 3, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 5669, "val/ratio": 1.0001744031906128, "val/ratio_var": 2.2094427549745888e-06 }, { "episode": 2560, "epoch": 0.010966415352981495, "eps": 6, "loss/policy_avg": 0.13433611392974854, "lr": 2.953125e-06, "objective/entropy": -57.38214111328125, "objective/kl": 2.255763292312622, "objective/non_score_reward": -0.11278817057609558, "objective/rlhf_reward": 7.51954460144043, "objective/scores": 7.6323323249816895, "policy/approxkl_avg": 0.11168310046195984, "policy/clipfrac_avg": 0.328125, "policy/entropy_avg": 0.5522703528404236, "step": 4, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 5788, "val/ratio": 1.0000145435333252, "val/ratio_var": 1.634987597753934e-06 }, { "episode": 3072, "epoch": 0.013159698423577794, "eps": 7, "loss/policy_avg": 0.13945499062538147, "lr": 2.94140625e-06, "objective/entropy": -56.70085144042969, "objective/kl": 3.472144603729248, "objective/non_score_reward": -0.1736072301864624, "objective/rlhf_reward": 7.586057662963867, "objective/scores": 7.759665012359619, "policy/approxkl_avg": 0.10526105761528015, "policy/clipfrac_avg": 0.31640625, "policy/entropy_avg": 0.5675480365753174, "step": 5, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 5734, "val/ratio": 1.0001214742660522, "val/ratio_var": 1.3759735111307236e-06 }, { "episode": 3584, "epoch": 0.015352981494174092, "eps": 7, "loss/policy_avg": 0.09546089917421341, "lr": 2.9296875e-06, "objective/entropy": -57.61265563964844, "objective/kl": 4.814455986022949, "objective/non_score_reward": -0.24072279036045074, "objective/rlhf_reward": 8.14423942565918, "objective/scores": 8.38496208190918, "policy/approxkl_avg": 0.11149222403764725, "policy/clipfrac_avg": 0.34765625, "policy/entropy_avg": 0.5545904040336609, "step": 6, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 5490, "val/ratio": 1.0001678466796875, "val/ratio_var": 1.6421691952928086e-06 }, { "episode": 4096, "epoch": 0.01754626456477039, "eps": 7, "loss/policy_avg": 0.235539510846138, "lr": 2.91796875e-06, "objective/entropy": -52.81565856933594, "objective/kl": 7.018827438354492, "objective/non_score_reward": -0.35094135999679565, "objective/rlhf_reward": 8.318111419677734, "objective/scores": 8.669052124023438, "policy/approxkl_avg": 0.14312206208705902, "policy/clipfrac_avg": 0.2890625, "policy/entropy_avg": 0.5730656385421753, "step": 7, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 5627, "val/ratio": 0.9999744892120361, "val/ratio_var": 2.2067135887482436e-06 }, { "episode": 4608, "epoch": 0.01973954763536669, "eps": 7, "loss/policy_avg": 0.18809324502944946, "lr": 2.90625e-06, "objective/entropy": -51.69276428222656, "objective/kl": 8.87761116027832, "objective/non_score_reward": -0.4438805878162384, "objective/rlhf_reward": 8.72425651550293, "objective/scores": 9.168136596679688, "policy/approxkl_avg": 0.13404184579849243, "policy/clipfrac_avg": 0.31640625, "policy/entropy_avg": 0.5807090997695923, "step": 8, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 5006, "val/ratio": 0.9999812841415405, "val/ratio_var": 2.421862973278621e-06 }, { "episode": 5120, "epoch": 0.02193283070596299, "eps": 7, "loss/policy_avg": 0.13251258432865143, "lr": 2.89453125e-06, "objective/entropy": -52.892486572265625, "objective/kl": 11.408520698547363, "objective/non_score_reward": -0.5704259872436523, "objective/rlhf_reward": 9.237460136413574, "objective/scores": 9.807886123657227, "policy/approxkl_avg": 0.12708792090415955, "policy/clipfrac_avg": 0.345703125, "policy/entropy_avg": 0.5779529213905334, "step": 9, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 4455, "val/ratio": 1.0001035928726196, "val/ratio_var": 1.636367528590199e-06 }, { "episode": 5632, "epoch": 0.024126113776559287, "eps": 7, "loss/policy_avg": 0.18584033846855164, "lr": 2.8828125e-06, "objective/entropy": -52.1838264465332, "objective/kl": 12.914525032043457, "objective/non_score_reward": -0.645726203918457, "objective/rlhf_reward": 9.471303939819336, "objective/scores": 10.11703109741211, "policy/approxkl_avg": 0.14649230241775513, "policy/clipfrac_avg": 0.310546875, "policy/entropy_avg": 0.5547796487808228, "step": 10, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3205, "val/ratio": 1.0001304149627686, "val/ratio_var": 1.861714054030017e-06 }, { "episode": 6144, "epoch": 0.026319396847155587, "eps": 7, "loss/policy_avg": 0.29413461685180664, "lr": 2.87109375e-06, "objective/entropy": -50.931846618652344, "objective/kl": 14.186361312866211, "objective/non_score_reward": -0.7093180418014526, "objective/rlhf_reward": 9.125755310058594, "objective/scores": 9.835073471069336, "policy/approxkl_avg": 0.10939589142799377, "policy/clipfrac_avg": 0.265625, "policy/entropy_avg": 0.5228562355041504, "step": 11, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2346, "val/ratio": 1.0003429651260376, "val/ratio_var": 2.3797676931280876e-06 }, { "episode": 6656, "epoch": 0.028512679917751884, "eps": 7, "loss/policy_avg": 0.2645286023616791, "lr": 2.859375e-06, "objective/entropy": -43.677490234375, "objective/kl": 15.47945785522461, "objective/non_score_reward": -0.7739728689193726, "objective/rlhf_reward": 9.23073959350586, "objective/scores": 10.00471305847168, "policy/approxkl_avg": 0.12562136352062225, "policy/clipfrac_avg": 0.318359375, "policy/entropy_avg": 0.5210846066474915, "step": 12, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2033, "val/ratio": 0.9997715950012207, "val/ratio_var": 2.0162181044725003e-06 }, { "episode": 7168, "epoch": 0.030705962988348184, "eps": 7, "loss/policy_avg": 0.11199663579463959, "lr": 2.84765625e-06, "objective/entropy": -48.57129669189453, "objective/kl": 15.832266807556152, "objective/non_score_reward": -0.7916133403778076, "objective/rlhf_reward": 9.829407691955566, "objective/scores": 10.621020317077637, "policy/approxkl_avg": 0.11009921133518219, "policy/clipfrac_avg": 0.33203125, "policy/entropy_avg": 0.5088958740234375, "step": 13, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2090, "val/ratio": 0.9999546408653259, "val/ratio_var": 1.2889486242784187e-06 }, { "episode": 7680, "epoch": 0.03289924605894448, "eps": 7, "loss/policy_avg": 0.2023422122001648, "lr": 2.8359375e-06, "objective/entropy": -44.59040451049805, "objective/kl": 16.71160888671875, "objective/non_score_reward": -0.8355805277824402, "objective/rlhf_reward": 9.781827926635742, "objective/scores": 10.61740779876709, "policy/approxkl_avg": 0.1277025043964386, "policy/clipfrac_avg": 0.3203125, "policy/entropy_avg": 0.5323245525360107, "step": 14, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2203, "val/ratio": 1.0000615119934082, "val/ratio_var": 1.983828269658261e-06 }, { "episode": 8192, "epoch": 0.03509252912954078, "eps": 7, "loss/policy_avg": 0.11938837170600891, "lr": 2.82421875e-06, "objective/entropy": -42.89122772216797, "objective/kl": 17.852779388427734, "objective/non_score_reward": -0.8926390409469604, "objective/rlhf_reward": 9.708269119262695, "objective/scores": 10.600908279418945, "policy/approxkl_avg": 0.13053694367408752, "policy/clipfrac_avg": 0.326171875, "policy/entropy_avg": 0.5566774606704712, "step": 15, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2344, "val/ratio": 0.9997885823249817, "val/ratio_var": 1.6801355968709686e-06 }, { "episode": 8704, "epoch": 0.03728581220013708, "eps": 7, "loss/policy_avg": 0.13911566138267517, "lr": 2.8125e-06, "objective/entropy": -41.679447174072266, "objective/kl": 19.221725463867188, "objective/non_score_reward": -0.9610862731933594, "objective/rlhf_reward": 10.175975799560547, "objective/scores": 11.137063026428223, "policy/approxkl_avg": 0.13960514962673187, "policy/clipfrac_avg": 0.341796875, "policy/entropy_avg": 0.5391991138458252, "step": 16, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2362, "val/ratio": 1.0002028942108154, "val/ratio_var": 1.7096700730689918e-06 }, { "episode": 9216, "epoch": 0.03947909527073338, "eps": 7, "loss/policy_avg": 0.1141306459903717, "lr": 2.80078125e-06, "objective/entropy": -38.35896682739258, "objective/kl": 21.127605438232422, "objective/non_score_reward": -1.0563803911209106, "objective/rlhf_reward": 10.05744743347168, "objective/scores": 11.113826751708984, "policy/approxkl_avg": 0.16503070294857025, "policy/clipfrac_avg": 0.357421875, "policy/entropy_avg": 0.5556284189224243, "step": 17, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2512, "val/ratio": 0.9997155070304871, "val/ratio_var": 2.5818217181949876e-06 }, { "episode": 9728, "epoch": 0.041672378341329676, "eps": 7, "loss/policy_avg": 0.1537306159734726, "lr": 2.7890625e-06, "objective/entropy": -34.704803466796875, "objective/kl": 21.983102798461914, "objective/non_score_reward": -1.0991551876068115, "objective/rlhf_reward": 9.975513458251953, "objective/scores": 11.074668884277344, "policy/approxkl_avg": 0.16006678342819214, "policy/clipfrac_avg": 0.369140625, "policy/entropy_avg": 0.5572025775909424, "step": 18, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2321, "val/ratio": 0.9999374747276306, "val/ratio_var": 2.024344212259166e-06 }, { "episode": 10240, "epoch": 0.04386566141192598, "eps": 7, "loss/policy_avg": 0.18893875181674957, "lr": 2.77734375e-06, "objective/entropy": -33.0050163269043, "objective/kl": 23.7343807220459, "objective/non_score_reward": -1.1867190599441528, "objective/rlhf_reward": 9.882341384887695, "objective/scores": 11.069059371948242, "policy/approxkl_avg": 0.16320574283599854, "policy/clipfrac_avg": 0.3203125, "policy/entropy_avg": 0.5377587676048279, "step": 19, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2204, "val/ratio": 0.9999250173568726, "val/ratio_var": 1.6979042811726686e-06 }, { "episode": 10752, "epoch": 0.04605894448252228, "eps": 7, "loss/policy_avg": 0.1287260800600052, "lr": 2.765625e-06, "objective/entropy": -30.667198181152344, "objective/kl": 25.033987045288086, "objective/non_score_reward": -1.251699447631836, "objective/rlhf_reward": 10.062000274658203, "objective/scores": 11.313698768615723, "policy/approxkl_avg": 0.1715945601463318, "policy/clipfrac_avg": 0.357421875, "policy/entropy_avg": 0.5366133451461792, "step": 20, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2182, "val/ratio": 0.9997314214706421, "val/ratio_var": 2.0495133412623545e-06 }, { "episode": 11264, "epoch": 0.048252227553118573, "eps": 7, "loss/policy_avg": 0.19849294424057007, "lr": 2.75390625e-06, "objective/entropy": -33.668846130371094, "objective/kl": 24.064254760742188, "objective/non_score_reward": -1.2032127380371094, "objective/rlhf_reward": 10.199529647827148, "objective/scores": 11.402742385864258, "policy/approxkl_avg": 0.16391384601593018, "policy/clipfrac_avg": 0.310546875, "policy/entropy_avg": 0.5062953233718872, "step": 21, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2180, "val/ratio": 0.9998014569282532, "val/ratio_var": 2.044889470198541e-06 }, { "episode": 11776, "epoch": 0.05044551062371487, "eps": 7, "loss/policy_avg": 0.10195901989936829, "lr": 2.7421875e-06, "objective/entropy": -39.06690979003906, "objective/kl": 22.586650848388672, "objective/non_score_reward": -1.1293325424194336, "objective/rlhf_reward": 10.262170791625977, "objective/scores": 11.391502380371094, "policy/approxkl_avg": 0.1463184356689453, "policy/clipfrac_avg": 0.3671875, "policy/entropy_avg": 0.4969005286693573, "step": 22, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2233, "val/ratio": 0.999759316444397, "val/ratio_var": 2.41930774791399e-06 }, { "episode": 12288, "epoch": 0.052638793694311174, "eps": 7, "loss/policy_avg": 0.10310226678848267, "lr": 2.73046875e-06, "objective/entropy": -46.28221893310547, "objective/kl": 22.036273956298828, "objective/non_score_reward": -1.101813793182373, "objective/rlhf_reward": 10.483089447021484, "objective/scores": 11.5849027633667, "policy/approxkl_avg": 0.1329042762517929, "policy/clipfrac_avg": 0.326171875, "policy/entropy_avg": 0.4535849094390869, "step": 23, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2392, "val/ratio": 0.9999419450759888, "val/ratio_var": 1.8799750023390516e-06 }, { "episode": 12800, "epoch": 0.05483207676490747, "eps": 7, "loss/policy_avg": 0.11269126832485199, "lr": 2.71875e-06, "objective/entropy": -43.433353424072266, "objective/kl": 22.60041046142578, "objective/non_score_reward": -1.1300204992294312, "objective/rlhf_reward": 10.324277877807617, "objective/scores": 11.45429801940918, "policy/approxkl_avg": 0.12673968076705933, "policy/clipfrac_avg": 0.287109375, "policy/entropy_avg": 0.4514559209346771, "step": 24, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2316, "val/ratio": 0.9997400045394897, "val/ratio_var": 1.5890127542661503e-06 }, { "episode": 13312, "epoch": 0.05702535983550377, "eps": 7, "loss/policy_avg": 0.2015887349843979, "lr": 2.70703125e-06, "objective/entropy": -46.94512176513672, "objective/kl": 21.664478302001953, "objective/non_score_reward": -1.0832240581512451, "objective/rlhf_reward": 10.445198059082031, "objective/scores": 11.528423309326172, "policy/approxkl_avg": 0.12853267788887024, "policy/clipfrac_avg": 0.3203125, "policy/entropy_avg": 0.39970842003822327, "step": 25, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2327, "val/ratio": 1.000229835510254, "val/ratio_var": 1.6951470342974062e-06 }, { "episode": 13824, "epoch": 0.05921864290610007, "eps": 7, "loss/policy_avg": 0.07959149777889252, "lr": 2.6953125e-06, "objective/entropy": -52.64186096191406, "objective/kl": 21.013965606689453, "objective/non_score_reward": -1.0506982803344727, "objective/rlhf_reward": 10.545175552368164, "objective/scores": 11.595874786376953, "policy/approxkl_avg": 0.08630259335041046, "policy/clipfrac_avg": 0.26953125, "policy/entropy_avg": 0.3702251613140106, "step": 26, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2337, "val/ratio": 0.999913215637207, "val/ratio_var": 9.910673952617799e-07 }, { "episode": 14336, "epoch": 0.06141192597669637, "eps": 7, "loss/policy_avg": 0.133376345038414, "lr": 2.68359375e-06, "objective/entropy": -52.62764358520508, "objective/kl": 21.873783111572266, "objective/non_score_reward": -1.0936890840530396, "objective/rlhf_reward": 10.531662940979004, "objective/scores": 11.62535285949707, "policy/approxkl_avg": 0.09903587400913239, "policy/clipfrac_avg": 0.259765625, "policy/entropy_avg": 0.3591485619544983, "step": 27, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2342, "val/ratio": 1.0000910758972168, "val/ratio_var": 1.4223566040527658e-06 }, { "episode": 14848, "epoch": 0.06360520904729267, "eps": 7, "loss/policy_avg": 0.09103123843669891, "lr": 2.671875e-06, "objective/entropy": -53.78799057006836, "objective/kl": 22.003097534179688, "objective/non_score_reward": -1.1001547574996948, "objective/rlhf_reward": 10.68588924407959, "objective/scores": 11.786044120788574, "policy/approxkl_avg": 0.09524992108345032, "policy/clipfrac_avg": 0.2734375, "policy/entropy_avg": 0.34278032183647156, "step": 28, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2247, "val/ratio": 0.9998263120651245, "val/ratio_var": 1.1149676311106305e-06 }, { "episode": 15360, "epoch": 0.06579849211788896, "eps": 7, "loss/policy_avg": 0.11379580199718475, "lr": 2.66015625e-06, "objective/entropy": -51.830589294433594, "objective/kl": 22.123140335083008, "objective/non_score_reward": -1.1061570644378662, "objective/rlhf_reward": 10.770975112915039, "objective/scores": 11.877132415771484, "policy/approxkl_avg": 0.0886557325720787, "policy/clipfrac_avg": 0.275390625, "policy/entropy_avg": 0.32301610708236694, "step": 29, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2108, "val/ratio": 0.9998676776885986, "val/ratio_var": 9.721951528263162e-07 }, { "episode": 15872, "epoch": 0.06799177518848526, "eps": 7, "loss/policy_avg": 0.1263660490512848, "lr": 2.6484375e-06, "objective/entropy": -52.30486297607422, "objective/kl": 22.81517219543457, "objective/non_score_reward": -1.1407586336135864, "objective/rlhf_reward": 10.94261360168457, "objective/scores": 12.083372116088867, "policy/approxkl_avg": 0.08135354518890381, "policy/clipfrac_avg": 0.2578125, "policy/entropy_avg": 0.3432938754558563, "step": 30, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2136, "val/ratio": 0.9998652935028076, "val/ratio_var": 1.0441638096381212e-06 }, { "episode": 16384, "epoch": 0.07018505825908156, "eps": 7, "loss/policy_avg": 0.11862729489803314, "lr": 2.63671875e-06, "objective/entropy": -52.75566101074219, "objective/kl": 23.3867130279541, "objective/non_score_reward": -1.1693357229232788, "objective/rlhf_reward": 10.69216251373291, "objective/scores": 11.86149787902832, "policy/approxkl_avg": 0.08456984162330627, "policy/clipfrac_avg": 0.25390625, "policy/entropy_avg": 0.28824326395988464, "step": 31, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2113, "val/ratio": 0.9998213052749634, "val/ratio_var": 1.4351767276821192e-06 }, { "episode": 16896, "epoch": 0.07237834132967787, "eps": 7, "loss/policy_avg": 0.1567326784133911, "lr": 2.6250000000000003e-06, "objective/entropy": -54.1901741027832, "objective/kl": 22.347511291503906, "objective/non_score_reward": -1.1173756122589111, "objective/rlhf_reward": 10.90484619140625, "objective/scores": 12.022221565246582, "policy/approxkl_avg": 0.07176832109689713, "policy/clipfrac_avg": 0.240234375, "policy/entropy_avg": 0.29196977615356445, "step": 32, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2180, "val/ratio": 1.0000624656677246, "val/ratio_var": 1.1582534398257849e-06 }, { "episode": 17408, "epoch": 0.07457162440027416, "eps": 7, "loss/policy_avg": 0.08574818074703217, "lr": 2.61328125e-06, "objective/entropy": -53.509254455566406, "objective/kl": 23.09099578857422, "objective/non_score_reward": -1.1545497179031372, "objective/rlhf_reward": 10.845325469970703, "objective/scores": 11.99987506866455, "policy/approxkl_avg": 0.07496441900730133, "policy/clipfrac_avg": 0.265625, "policy/entropy_avg": 0.2839745581150055, "step": 33, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2102, "val/ratio": 0.999615490436554, "val/ratio_var": 1.1524506362547982e-06 }, { "episode": 17920, "epoch": 0.07676490747087046, "eps": 7, "loss/policy_avg": 0.0927099734544754, "lr": 2.6015625e-06, "objective/entropy": -52.29348373413086, "objective/kl": 23.082698822021484, "objective/non_score_reward": -1.15413498878479, "objective/rlhf_reward": 10.935807228088379, "objective/scores": 12.08994197845459, "policy/approxkl_avg": 0.15793153643608093, "policy/clipfrac_avg": 0.265625, "policy/entropy_avg": 0.2997421622276306, "step": 34, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2219, "val/ratio": 1.0001107454299927, "val/ratio_var": 9.742823294800473e-07 }, { "episode": 18432, "epoch": 0.07895819054146676, "eps": 7, "loss/policy_avg": 0.033941492438316345, "lr": 2.5898437500000003e-06, "objective/entropy": -53.88836669921875, "objective/kl": 22.903392791748047, "objective/non_score_reward": -1.1451694965362549, "objective/rlhf_reward": 11.03926944732666, "objective/scores": 12.184438705444336, "policy/approxkl_avg": 0.09045101702213287, "policy/clipfrac_avg": 0.298828125, "policy/entropy_avg": 0.31827011704444885, "step": 35, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2304, "val/ratio": 0.9999386072158813, "val/ratio_var": 1.0199778444075491e-06 }, { "episode": 18944, "epoch": 0.08115147361206305, "eps": 7, "loss/policy_avg": 0.08208812773227692, "lr": 2.578125e-06, "objective/entropy": -53.41987609863281, "objective/kl": 23.544723510742188, "objective/non_score_reward": -1.1772363185882568, "objective/rlhf_reward": 11.115821838378906, "objective/scores": 12.293057441711426, "policy/approxkl_avg": 0.08877880871295929, "policy/clipfrac_avg": 0.296875, "policy/entropy_avg": 0.31178680062294006, "step": 36, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2276, "val/ratio": 1.000001072883606, "val/ratio_var": 1.1726771163012017e-06 }, { "episode": 19456, "epoch": 0.08334475668265935, "eps": 7, "loss/policy_avg": 0.11828020215034485, "lr": 2.56640625e-06, "objective/entropy": -54.274024963378906, "objective/kl": 23.336807250976562, "objective/non_score_reward": -1.1668403148651123, "objective/rlhf_reward": 11.09507942199707, "objective/scores": 12.261919975280762, "policy/approxkl_avg": 0.09653887152671814, "policy/clipfrac_avg": 0.294921875, "policy/entropy_avg": 0.32997897267341614, "step": 37, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2241, "val/ratio": 1.000211477279663, "val/ratio_var": 1.642234337850823e-06 }, { "episode": 19968, "epoch": 0.08553803975325565, "eps": 7, "loss/policy_avg": 0.11273995041847229, "lr": 2.5546875000000003e-06, "objective/entropy": -53.54825210571289, "objective/kl": 23.62361717224121, "objective/non_score_reward": -1.1811809539794922, "objective/rlhf_reward": 11.013565063476562, "objective/scores": 12.194746017456055, "policy/approxkl_avg": 0.100711889564991, "policy/clipfrac_avg": 0.291015625, "policy/entropy_avg": 0.34581631422042847, "step": 38, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2229, "val/ratio": 0.9998159408569336, "val/ratio_var": 1.3141981298758765e-06 }, { "episode": 20480, "epoch": 0.08773132282385196, "eps": 7, "loss/policy_avg": 0.05108005180954933, "lr": 2.54296875e-06, "objective/entropy": -51.924232482910156, "objective/kl": 23.902057647705078, "objective/non_score_reward": -1.1951029300689697, "objective/rlhf_reward": 11.001468658447266, "objective/scores": 12.196571350097656, "policy/approxkl_avg": 0.0894564539194107, "policy/clipfrac_avg": 0.28515625, "policy/entropy_avg": 0.3434683382511139, "step": 39, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2254, "val/ratio": 1.0003117322921753, "val/ratio_var": 1.1836557405331405e-06 }, { "episode": 20992, "epoch": 0.08992460589444826, "eps": 7, "loss/policy_avg": 0.09136471152305603, "lr": 2.53125e-06, "objective/entropy": -52.43147277832031, "objective/kl": 25.310583114624023, "objective/non_score_reward": -1.2655291557312012, "objective/rlhf_reward": 10.961225509643555, "objective/scores": 12.226754188537598, "policy/approxkl_avg": 0.10527636855840683, "policy/clipfrac_avg": 0.296875, "policy/entropy_avg": 0.32169109582901, "step": 40, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2151, "val/ratio": 0.9999606013298035, "val/ratio_var": 1.430940642421774e-06 }, { "episode": 21504, "epoch": 0.09211788896504455, "eps": 7, "loss/policy_avg": 0.11661476641893387, "lr": 2.5195312500000003e-06, "objective/entropy": -46.344642639160156, "objective/kl": 27.143455505371094, "objective/non_score_reward": -1.3571728467941284, "objective/rlhf_reward": 10.886428833007812, "objective/scores": 12.243600845336914, "policy/approxkl_avg": 0.11585399508476257, "policy/clipfrac_avg": 0.28515625, "policy/entropy_avg": 0.3539811074733734, "step": 41, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2093, "val/ratio": 0.999966025352478, "val/ratio_var": 2.110434706992237e-06 }, { "episode": 22016, "epoch": 0.09431117203564085, "eps": 7, "loss/policy_avg": 0.1412871778011322, "lr": 2.5078125e-06, "objective/entropy": -48.18741989135742, "objective/kl": 26.137479782104492, "objective/non_score_reward": -1.3068740367889404, "objective/rlhf_reward": 11.087350845336914, "objective/scores": 12.394224166870117, "policy/approxkl_avg": 0.10230039805173874, "policy/clipfrac_avg": 0.30859375, "policy/entropy_avg": 0.35482996702194214, "step": 42, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 1995, "val/ratio": 1.0001914501190186, "val/ratio_var": 1.8418863874103408e-06 }, { "episode": 22528, "epoch": 0.09650445510623715, "eps": 7, "loss/policy_avg": 0.14972682297229767, "lr": 2.49609375e-06, "objective/entropy": -47.18873977661133, "objective/kl": 27.248594284057617, "objective/non_score_reward": -1.3624296188354492, "objective/rlhf_reward": 10.733315467834473, "objective/scores": 12.095744132995605, "policy/approxkl_avg": 0.11158671975135803, "policy/clipfrac_avg": 0.2890625, "policy/entropy_avg": 0.3710693418979645, "step": 43, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2105, "val/ratio": 1.000127911567688, "val/ratio_var": 1.7979554058911162e-06 }, { "episode": 23040, "epoch": 0.09869773817683344, "eps": 7, "loss/policy_avg": 0.1013510599732399, "lr": 2.4843750000000002e-06, "objective/entropy": -48.7869758605957, "objective/kl": 26.826696395874023, "objective/non_score_reward": -1.3413348197937012, "objective/rlhf_reward": 10.954305648803711, "objective/scores": 12.295639038085938, "policy/approxkl_avg": 0.09727650135755539, "policy/clipfrac_avg": 0.267578125, "policy/entropy_avg": 0.341769278049469, "step": 44, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 1957, "val/ratio": 0.9999566078186035, "val/ratio_var": 9.611487712390954e-07 }, { "episode": 23552, "epoch": 0.10089102124742974, "eps": 7, "loss/policy_avg": 0.09324927628040314, "lr": 2.47265625e-06, "objective/entropy": -48.091217041015625, "objective/kl": 27.1876163482666, "objective/non_score_reward": -1.3593807220458984, "objective/rlhf_reward": 11.09292984008789, "objective/scores": 12.452309608459473, "policy/approxkl_avg": 0.09258678555488586, "policy/clipfrac_avg": 0.294921875, "policy/entropy_avg": 0.3322508931159973, "step": 45, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 1981, "val/ratio": 0.9999479055404663, "val/ratio_var": 1.1157736707900767e-06 }, { "episode": 24064, "epoch": 0.10308430431802605, "eps": 7, "loss/policy_avg": 0.05657796561717987, "lr": 2.4609375e-06, "objective/entropy": -50.28386688232422, "objective/kl": 26.767160415649414, "objective/non_score_reward": -1.338357925415039, "objective/rlhf_reward": 11.471373558044434, "objective/scores": 12.809732437133789, "policy/approxkl_avg": 0.08614060282707214, "policy/clipfrac_avg": 0.28125, "policy/entropy_avg": 0.32660749554634094, "step": 46, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 1974, "val/ratio": 1.000014305114746, "val/ratio_var": 1.3160678236090462e-06 }, { "episode": 24576, "epoch": 0.10527758738862235, "eps": 7, "loss/policy_avg": 0.11685071885585785, "lr": 2.4492187500000002e-06, "objective/entropy": -47.901275634765625, "objective/kl": 28.288379669189453, "objective/non_score_reward": -1.414419174194336, "objective/rlhf_reward": 11.333412170410156, "objective/scores": 12.747832298278809, "policy/approxkl_avg": 0.10291337966918945, "policy/clipfrac_avg": 0.25390625, "policy/entropy_avg": 0.31352245807647705, "step": 47, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 1895, "val/ratio": 1.0002663135528564, "val/ratio_var": 1.1478540500320378e-06 }, { "episode": 25088, "epoch": 0.10747087045921865, "eps": 7, "loss/policy_avg": 0.09044268727302551, "lr": 2.4375e-06, "objective/entropy": -46.25014877319336, "objective/kl": 29.569900512695312, "objective/non_score_reward": -1.4784950017929077, "objective/rlhf_reward": 11.207091331481934, "objective/scores": 12.685586929321289, "policy/approxkl_avg": 0.08932553231716156, "policy/clipfrac_avg": 0.2578125, "policy/entropy_avg": 0.30406129360198975, "step": 48, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 1779, "val/ratio": 0.9998706579208374, "val/ratio_var": 9.449911431147484e-07 }, { "episode": 25600, "epoch": 0.10966415352981494, "eps": 7, "loss/policy_avg": 0.0890183225274086, "lr": 2.42578125e-06, "objective/entropy": -48.26085662841797, "objective/kl": 30.082365036010742, "objective/non_score_reward": -1.5041183233261108, "objective/rlhf_reward": 11.227607727050781, "objective/scores": 12.731725692749023, "policy/approxkl_avg": 0.08349761366844177, "policy/clipfrac_avg": 0.302734375, "policy/entropy_avg": 0.2885536849498749, "step": 49, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 1885, "val/ratio": 1.000079870223999, "val/ratio_var": 1.2881323527835775e-06 }, { "episode": 26112, "epoch": 0.11185743660041124, "eps": 7, "loss/policy_avg": 0.1166183277964592, "lr": 2.4140625000000002e-06, "objective/entropy": -45.06092071533203, "objective/kl": 30.306961059570312, "objective/non_score_reward": -1.515347957611084, "objective/rlhf_reward": 11.126758575439453, "objective/scores": 12.642105102539062, "policy/approxkl_avg": 0.08703693002462387, "policy/clipfrac_avg": 0.279296875, "policy/entropy_avg": 0.281158983707428, "step": 50, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 1857, "val/ratio": 1.000382423400879, "val/ratio_var": 1.3761283526037005e-06 }, { "episode": 26624, "epoch": 0.11405071967100754, "eps": 7, "loss/policy_avg": 0.09397272765636444, "lr": 2.40234375e-06, "objective/entropy": -47.06073760986328, "objective/kl": 30.048595428466797, "objective/non_score_reward": -1.502429723739624, "objective/rlhf_reward": 11.373811721801758, "objective/scores": 12.876239776611328, "policy/approxkl_avg": 0.07360626757144928, "policy/clipfrac_avg": 0.23828125, "policy/entropy_avg": 0.25442755222320557, "step": 51, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 1919, "val/ratio": 1.0000107288360596, "val/ratio_var": 1.3316806644070311e-06 }, { "episode": 27136, "epoch": 0.11624400274160383, "eps": 7, "loss/policy_avg": 0.05725743621587753, "lr": 2.390625e-06, "objective/entropy": -50.653358459472656, "objective/kl": 28.8651123046875, "objective/non_score_reward": -1.4432556629180908, "objective/rlhf_reward": 11.409313201904297, "objective/scores": 12.852569580078125, "policy/approxkl_avg": 0.0559498593211174, "policy/clipfrac_avg": 0.23828125, "policy/entropy_avg": 0.2538541257381439, "step": 52, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2038, "val/ratio": 1.0001037120819092, "val/ratio_var": 1.2117743608541787e-06 }, { "episode": 27648, "epoch": 0.11843728581220014, "eps": 7, "loss/policy_avg": 0.04747297987341881, "lr": 2.3789062500000002e-06, "objective/entropy": -49.83374786376953, "objective/kl": 29.47016716003418, "objective/non_score_reward": -1.473508358001709, "objective/rlhf_reward": 11.507312774658203, "objective/scores": 12.98082160949707, "policy/approxkl_avg": 0.07053554803133011, "policy/clipfrac_avg": 0.244140625, "policy/entropy_avg": 0.23207688331604004, "step": 53, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2084, "val/ratio": 0.9998562335968018, "val/ratio_var": 7.929276648610539e-07 }, { "episode": 28160, "epoch": 0.12063056888279644, "eps": 7, "loss/policy_avg": 0.11368705332279205, "lr": 2.3671875e-06, "objective/entropy": -51.63511657714844, "objective/kl": 28.794090270996094, "objective/non_score_reward": -1.4397046566009521, "objective/rlhf_reward": 11.756919860839844, "objective/scores": 13.196623802185059, "policy/approxkl_avg": 0.06422245502471924, "policy/clipfrac_avg": 0.203125, "policy/entropy_avg": 0.21511417627334595, "step": 54, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2029, "val/ratio": 0.9999565482139587, "val/ratio_var": 9.939893743649009e-07 }, { "episode": 28672, "epoch": 0.12282385195339274, "eps": 7, "loss/policy_avg": 0.0857730507850647, "lr": 2.35546875e-06, "objective/entropy": -48.501441955566406, "objective/kl": 29.051284790039062, "objective/non_score_reward": -1.4525642395019531, "objective/rlhf_reward": 11.569185256958008, "objective/scores": 13.021749496459961, "policy/approxkl_avg": 0.05940084159374237, "policy/clipfrac_avg": 0.193359375, "policy/entropy_avg": 0.20681166648864746, "step": 55, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2137, "val/ratio": 0.9998863935470581, "val/ratio_var": 7.905163101895596e-07 }, { "episode": 29184, "epoch": 0.12501713502398903, "eps": 7, "loss/policy_avg": 0.06318055093288422, "lr": 2.3437500000000002e-06, "objective/entropy": -52.91703414916992, "objective/kl": 28.871070861816406, "objective/non_score_reward": -1.4435535669326782, "objective/rlhf_reward": 11.726110458374023, "objective/scores": 13.169663429260254, "policy/approxkl_avg": 0.0693005621433258, "policy/clipfrac_avg": 0.236328125, "policy/entropy_avg": 0.2014627903699875, "step": 56, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2150, "val/ratio": 0.9999423027038574, "val/ratio_var": 9.004759249364724e-07 }, { "episode": 29696, "epoch": 0.12721041809458533, "eps": 7, "loss/policy_avg": 0.025315172970294952, "lr": 2.33203125e-06, "objective/entropy": -51.94420623779297, "objective/kl": 28.651988983154297, "objective/non_score_reward": -1.432599425315857, "objective/rlhf_reward": 11.433904647827148, "objective/scores": 12.86650276184082, "policy/approxkl_avg": 0.055910706520080566, "policy/clipfrac_avg": 0.203125, "policy/entropy_avg": 0.21284107863903046, "step": 57, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2193, "val/ratio": 1.000023603439331, "val/ratio_var": 7.501536742893222e-07 }, { "episode": 30208, "epoch": 0.12940370116518163, "eps": 7, "loss/policy_avg": 0.04747464135289192, "lr": 2.3203125e-06, "objective/entropy": -54.25482940673828, "objective/kl": 28.553482055664062, "objective/non_score_reward": -1.4276741743087769, "objective/rlhf_reward": 11.46591567993164, "objective/scores": 12.893589973449707, "policy/approxkl_avg": 0.04430193454027176, "policy/clipfrac_avg": 0.181640625, "policy/entropy_avg": 0.197004497051239, "step": 58, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2182, "val/ratio": 1.0000498294830322, "val/ratio_var": 5.298446126289491e-07 }, { "episode": 30720, "epoch": 0.13159698423577793, "eps": 7, "loss/policy_avg": 0.07828865945339203, "lr": 2.30859375e-06, "objective/entropy": -55.69086837768555, "objective/kl": 27.18333625793457, "objective/non_score_reward": -1.3591668605804443, "objective/rlhf_reward": 11.24705696105957, "objective/scores": 12.606224060058594, "policy/approxkl_avg": 0.050338346511125565, "policy/clipfrac_avg": 0.177734375, "policy/entropy_avg": 0.19780844449996948, "step": 59, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2173, "val/ratio": 0.9999244213104248, "val/ratio_var": 8.30796864192962e-07 }, { "episode": 31232, "epoch": 0.13379026730637422, "eps": 7, "loss/policy_avg": 0.06892979145050049, "lr": 2.296875e-06, "objective/entropy": -55.50807189941406, "objective/kl": 27.259765625, "objective/non_score_reward": -1.3629883527755737, "objective/rlhf_reward": 11.379904747009277, "objective/scores": 12.74289321899414, "policy/approxkl_avg": 0.05562983453273773, "policy/clipfrac_avg": 0.1953125, "policy/entropy_avg": 0.2033534049987793, "step": 60, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2124, "val/ratio": 0.9998317956924438, "val/ratio_var": 7.12839153038658e-07 }, { "episode": 31744, "epoch": 0.13598355037697052, "eps": 7, "loss/policy_avg": 0.06427884846925735, "lr": 2.28515625e-06, "objective/entropy": -53.272911071777344, "objective/kl": 28.615087509155273, "objective/non_score_reward": -1.4307544231414795, "objective/rlhf_reward": 11.401360511779785, "objective/scores": 12.832115173339844, "policy/approxkl_avg": 0.05360804870724678, "policy/clipfrac_avg": 0.181640625, "policy/entropy_avg": 0.20671643316745758, "step": 61, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2243, "val/ratio": 0.9998288750648499, "val/ratio_var": 7.739292300357192e-07 }, { "episode": 32256, "epoch": 0.13817683344756682, "eps": 7, "loss/policy_avg": 0.036871928721666336, "lr": 2.2734375e-06, "objective/entropy": -56.196807861328125, "objective/kl": 26.758060455322266, "objective/non_score_reward": -1.3379030227661133, "objective/rlhf_reward": 11.421424865722656, "objective/scores": 12.759328842163086, "policy/approxkl_avg": 0.06217839941382408, "policy/clipfrac_avg": 0.224609375, "policy/entropy_avg": 0.20198120176792145, "step": 62, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2232, "val/ratio": 0.9998797178268433, "val/ratio_var": 9.791126558411634e-07 }, { "episode": 32768, "epoch": 0.1403701165181631, "eps": 7, "loss/policy_avg": 0.05493534356355667, "lr": 2.26171875e-06, "objective/entropy": -56.74695587158203, "objective/kl": 27.563007354736328, "objective/non_score_reward": -1.378150463104248, "objective/rlhf_reward": 11.451597213745117, "objective/scores": 12.829747200012207, "policy/approxkl_avg": 0.0605655238032341, "policy/clipfrac_avg": 0.205078125, "policy/entropy_avg": 0.21349653601646423, "step": 63, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2429, "val/ratio": 1.0000948905944824, "val/ratio_var": 8.029837204048818e-07 }, { "episode": 33280, "epoch": 0.14256339958875944, "eps": 7, "loss/policy_avg": 0.06948225200176239, "lr": 2.25e-06, "objective/entropy": -55.68669128417969, "objective/kl": 27.24061393737793, "objective/non_score_reward": -1.3620307445526123, "objective/rlhf_reward": 11.590509414672852, "objective/scores": 12.95254135131836, "policy/approxkl_avg": 0.058622099459171295, "policy/clipfrac_avg": 0.2109375, "policy/entropy_avg": 0.20613785088062286, "step": 64, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2500, "val/ratio": 1.0000636577606201, "val/ratio_var": 1.0106417676070123e-06 }, { "episode": 33792, "epoch": 0.14475668265935573, "eps": 7, "loss/policy_avg": 0.0674080029129982, "lr": 2.23828125e-06, "objective/entropy": -56.052215576171875, "objective/kl": 27.84102439880371, "objective/non_score_reward": -1.3920512199401855, "objective/rlhf_reward": 11.380613327026367, "objective/scores": 12.772663116455078, "policy/approxkl_avg": 0.05588747560977936, "policy/clipfrac_avg": 0.1875, "policy/entropy_avg": 0.21730531752109528, "step": 65, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2533, "val/ratio": 1.000044822692871, "val/ratio_var": 1.0719528518166044e-06 }, { "episode": 34304, "epoch": 0.14694996572995203, "eps": 7, "loss/policy_avg": 0.1223275363445282, "lr": 2.2265625e-06, "objective/entropy": -55.07699966430664, "objective/kl": 28.410842895507812, "objective/non_score_reward": -1.4205421209335327, "objective/rlhf_reward": 11.466856002807617, "objective/scores": 12.887397766113281, "policy/approxkl_avg": 0.07928226888179779, "policy/clipfrac_avg": 0.189453125, "policy/entropy_avg": 0.22597652673721313, "step": 66, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2586, "val/ratio": 1.000163197517395, "val/ratio_var": 1.0074684269056888e-06 }, { "episode": 34816, "epoch": 0.14914324880054833, "eps": 7, "loss/policy_avg": 0.06255296617746353, "lr": 2.21484375e-06, "objective/entropy": -56.23228454589844, "objective/kl": 27.560832977294922, "objective/non_score_reward": -1.3780417442321777, "objective/rlhf_reward": 11.549421310424805, "objective/scores": 12.927461624145508, "policy/approxkl_avg": 0.07416369020938873, "policy/clipfrac_avg": 0.2109375, "policy/entropy_avg": 0.22490039467811584, "step": 67, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2561, "val/ratio": 0.9998071193695068, "val/ratio_var": 7.868264901844668e-07 }, { "episode": 35328, "epoch": 0.15133653187114463, "eps": 7, "loss/policy_avg": 0.07749070227146149, "lr": 2.203125e-06, "objective/entropy": -58.02006912231445, "objective/kl": 28.338050842285156, "objective/non_score_reward": -1.4169025421142578, "objective/rlhf_reward": 11.502689361572266, "objective/scores": 12.919590950012207, "policy/approxkl_avg": 0.050264619290828705, "policy/clipfrac_avg": 0.17578125, "policy/entropy_avg": 0.22556829452514648, "step": 68, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2776, "val/ratio": 0.9998827576637268, "val/ratio_var": 6.7324697283766e-07 }, { "episode": 35840, "epoch": 0.15352981494174092, "eps": 7, "loss/policy_avg": 0.04901011660695076, "lr": 2.19140625e-06, "objective/entropy": -57.22145462036133, "objective/kl": 28.202184677124023, "objective/non_score_reward": -1.410109281539917, "objective/rlhf_reward": 11.595409393310547, "objective/scores": 13.005517959594727, "policy/approxkl_avg": 0.05339674651622772, "policy/clipfrac_avg": 0.197265625, "policy/entropy_avg": 0.2128574401140213, "step": 69, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2659, "val/ratio": 0.9999070167541504, "val/ratio_var": 6.147514000076626e-07 }, { "episode": 36352, "epoch": 0.15572309801233722, "eps": 7, "loss/policy_avg": 0.04001376032829285, "lr": 2.1796875e-06, "objective/entropy": -55.01459503173828, "objective/kl": 29.15834617614746, "objective/non_score_reward": -1.4579174518585205, "objective/rlhf_reward": 11.658880233764648, "objective/scores": 13.11679744720459, "policy/approxkl_avg": 0.0480184480547905, "policy/clipfrac_avg": 0.208984375, "policy/entropy_avg": 0.20534545183181763, "step": 70, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2526, "val/ratio": 1.0001248121261597, "val/ratio_var": 7.817943696863949e-07 }, { "episode": 36864, "epoch": 0.15791638108293352, "eps": 7, "loss/policy_avg": 0.0806807279586792, "lr": 2.16796875e-06, "objective/entropy": -54.07442092895508, "objective/kl": 28.838031768798828, "objective/non_score_reward": -1.4419015645980835, "objective/rlhf_reward": 11.803167343139648, "objective/scores": 13.24506950378418, "policy/approxkl_avg": 0.05658765137195587, "policy/clipfrac_avg": 0.16796875, "policy/entropy_avg": 0.22964796423912048, "step": 71, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2643, "val/ratio": 1.000180721282959, "val/ratio_var": 8.247698701779882e-07 }, { "episode": 37376, "epoch": 0.1601096641535298, "eps": 7, "loss/policy_avg": 0.051692895591259, "lr": 2.15625e-06, "objective/entropy": -54.76805114746094, "objective/kl": 29.931236267089844, "objective/non_score_reward": -1.4965617656707764, "objective/rlhf_reward": 11.834163665771484, "objective/scores": 13.330726623535156, "policy/approxkl_avg": 0.0646805614233017, "policy/clipfrac_avg": 0.19140625, "policy/entropy_avg": 0.2366558015346527, "step": 72, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2777, "val/ratio": 0.9999984502792358, "val/ratio_var": 1.2423626003510435e-06 }, { "episode": 37888, "epoch": 0.1623029472241261, "eps": 7, "loss/policy_avg": 0.05296178162097931, "lr": 2.14453125e-06, "objective/entropy": -52.60663604736328, "objective/kl": 29.616146087646484, "objective/non_score_reward": -1.4808073043823242, "objective/rlhf_reward": 11.590347290039062, "objective/scores": 13.071154594421387, "policy/approxkl_avg": 0.05852410942316055, "policy/clipfrac_avg": 0.203125, "policy/entropy_avg": 0.22921660542488098, "step": 73, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2599, "val/ratio": 0.9999311566352844, "val/ratio_var": 9.6972451046895e-07 }, { "episode": 38400, "epoch": 0.1644962302947224, "eps": 7, "loss/policy_avg": 0.05771177262067795, "lr": 2.1328125e-06, "objective/entropy": -50.87690734863281, "objective/kl": 31.555557250976562, "objective/non_score_reward": -1.5777779817581177, "objective/rlhf_reward": 11.474925994873047, "objective/scores": 13.052703857421875, "policy/approxkl_avg": 0.06892445683479309, "policy/clipfrac_avg": 0.23828125, "policy/entropy_avg": 0.2290726602077484, "step": 74, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2545, "val/ratio": 1.0000228881835938, "val/ratio_var": 8.888110301086272e-07 }, { "episode": 38912, "epoch": 0.1666895133653187, "eps": 7, "loss/policy_avg": 0.08321105688810349, "lr": 2.12109375e-06, "objective/entropy": -49.94180679321289, "objective/kl": 31.343868255615234, "objective/non_score_reward": -1.5671935081481934, "objective/rlhf_reward": 11.681865692138672, "objective/scores": 13.249059677124023, "policy/approxkl_avg": 0.06739464402198792, "policy/clipfrac_avg": 0.205078125, "policy/entropy_avg": 0.22561725974082947, "step": 75, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2453, "val/ratio": 1.0000121593475342, "val/ratio_var": 1.293211198571953e-06 }, { "episode": 39424, "epoch": 0.168882796435915, "eps": 7, "loss/policy_avg": 0.04158315062522888, "lr": 2.109375e-06, "objective/entropy": -51.920650482177734, "objective/kl": 30.741943359375, "objective/non_score_reward": -1.5370972156524658, "objective/rlhf_reward": 11.58804988861084, "objective/scores": 13.125146865844727, "policy/approxkl_avg": 0.06920449435710907, "policy/clipfrac_avg": 0.2265625, "policy/entropy_avg": 0.23201118409633636, "step": 76, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2530, "val/ratio": 1.000053882598877, "val/ratio_var": 1.389662315887108e-06 }, { "episode": 39936, "epoch": 0.1710760795065113, "eps": 7, "loss/policy_avg": 0.053401652723550797, "lr": 2.09765625e-06, "objective/entropy": -54.08863830566406, "objective/kl": 29.898433685302734, "objective/non_score_reward": -1.4949218034744263, "objective/rlhf_reward": 11.703346252441406, "objective/scores": 13.19826889038086, "policy/approxkl_avg": 0.05423176288604736, "policy/clipfrac_avg": 0.185546875, "policy/entropy_avg": 0.23785756528377533, "step": 77, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2552, "val/ratio": 1.000126838684082, "val/ratio_var": 9.200340400639107e-07 }, { "episode": 40448, "epoch": 0.17326936257710762, "eps": 7, "loss/policy_avg": 0.016359925270080566, "lr": 2.0859375e-06, "objective/entropy": -52.485328674316406, "objective/kl": 31.477571487426758, "objective/non_score_reward": -1.5738786458969116, "objective/rlhf_reward": 11.632445335388184, "objective/scores": 13.206323623657227, "policy/approxkl_avg": 0.05758652836084366, "policy/clipfrac_avg": 0.244140625, "policy/entropy_avg": 0.22791028022766113, "step": 78, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2451, "val/ratio": 0.9999771118164062, "val/ratio_var": 1.3394280813372461e-06 }, { "episode": 40960, "epoch": 0.17546264564770392, "eps": 7, "loss/policy_avg": 0.04443100094795227, "lr": 2.07421875e-06, "objective/entropy": -52.247257232666016, "objective/kl": 31.95646095275879, "objective/non_score_reward": -1.597823143005371, "objective/rlhf_reward": 11.39157485961914, "objective/scores": 12.989398002624512, "policy/approxkl_avg": 0.062236685305833817, "policy/clipfrac_avg": 0.2421875, "policy/entropy_avg": 0.2292611300945282, "step": 79, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2326, "val/ratio": 0.9999898672103882, "val/ratio_var": 1.2004873042315012e-06 }, { "episode": 41472, "epoch": 0.17765592871830022, "eps": 7, "loss/policy_avg": 0.06968129426240921, "lr": 2.0625e-06, "objective/entropy": -53.13126754760742, "objective/kl": 31.035659790039062, "objective/non_score_reward": -1.5517830848693848, "objective/rlhf_reward": 11.682025909423828, "objective/scores": 13.233808517456055, "policy/approxkl_avg": 0.05690852180123329, "policy/clipfrac_avg": 0.20703125, "policy/entropy_avg": 0.22946974635124207, "step": 80, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2409, "val/ratio": 0.9998325109481812, "val/ratio_var": 6.52003791401512e-07 }, { "episode": 41984, "epoch": 0.1798492117888965, "eps": 7, "loss/policy_avg": 0.05149594694375992, "lr": 2.05078125e-06, "objective/entropy": -53.41246032714844, "objective/kl": 30.750011444091797, "objective/non_score_reward": -1.5375006198883057, "objective/rlhf_reward": 11.637171745300293, "objective/scores": 13.17467212677002, "policy/approxkl_avg": 0.07379427552223206, "policy/clipfrac_avg": 0.224609375, "policy/entropy_avg": 0.23126959800720215, "step": 81, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2298, "val/ratio": 1.0000648498535156, "val/ratio_var": 1.0951293916150462e-06 }, { "episode": 42496, "epoch": 0.1820424948594928, "eps": 7, "loss/policy_avg": 0.1227312758564949, "lr": 2.0390625e-06, "objective/entropy": -51.953941345214844, "objective/kl": 31.699325561523438, "objective/non_score_reward": -1.5849663019180298, "objective/rlhf_reward": 11.240765571594238, "objective/scores": 12.82573127746582, "policy/approxkl_avg": 0.08355057239532471, "policy/clipfrac_avg": 0.224609375, "policy/entropy_avg": 0.21501347422599792, "step": 82, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2209, "val/ratio": 1.0003668069839478, "val/ratio_var": 1.6002859410946257e-05 }, { "episode": 43008, "epoch": 0.1842357779300891, "eps": 7, "loss/policy_avg": 0.035134680569171906, "lr": 2.02734375e-06, "objective/entropy": -51.80181121826172, "objective/kl": 31.36351776123047, "objective/non_score_reward": -1.568176031112671, "objective/rlhf_reward": 11.594715118408203, "objective/scores": 13.162890434265137, "policy/approxkl_avg": 0.05513919144868851, "policy/clipfrac_avg": 0.212890625, "policy/entropy_avg": 0.2203870713710785, "step": 83, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2286, "val/ratio": 0.999825119972229, "val/ratio_var": 7.327400339818269e-07 }, { "episode": 43520, "epoch": 0.1864290610006854, "eps": 7, "loss/policy_avg": 0.047856032848358154, "lr": 2.015625e-06, "objective/entropy": -50.72874450683594, "objective/kl": 31.98059844970703, "objective/non_score_reward": -1.5990298986434937, "objective/rlhf_reward": 11.84218978881836, "objective/scores": 13.4412202835083, "policy/approxkl_avg": 0.06686994433403015, "policy/clipfrac_avg": 0.20703125, "policy/entropy_avg": 0.208661288022995, "step": 84, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2292, "val/ratio": 1.0000088214874268, "val/ratio_var": 8.189380196199636e-07 }, { "episode": 44032, "epoch": 0.1886223440712817, "eps": 7, "loss/policy_avg": 0.06779906898736954, "lr": 2.00390625e-06, "objective/entropy": -53.01400375366211, "objective/kl": 31.55933380126953, "objective/non_score_reward": -1.5779666900634766, "objective/rlhf_reward": 11.665364265441895, "objective/scores": 13.243330955505371, "policy/approxkl_avg": 0.052923060953617096, "policy/clipfrac_avg": 0.177734375, "policy/entropy_avg": 0.2091614007949829, "step": 85, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2185, "val/ratio": 0.9999080896377563, "val/ratio_var": 7.456224579982518e-07 }, { "episode": 44544, "epoch": 0.190815627141878, "eps": 7, "loss/policy_avg": 0.05880901217460632, "lr": 1.9921875e-06, "objective/entropy": -49.737144470214844, "objective/kl": 31.93135643005371, "objective/non_score_reward": -1.596567988395691, "objective/rlhf_reward": 11.870726585388184, "objective/scores": 13.467294692993164, "policy/approxkl_avg": 0.055838484317064285, "policy/clipfrac_avg": 0.20703125, "policy/entropy_avg": 0.2014000117778778, "step": 86, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2277, "val/ratio": 1.0000534057617188, "val/ratio_var": 1.0206571232629358e-06 }, { "episode": 45056, "epoch": 0.1930089102124743, "eps": 7, "loss/policy_avg": 0.04301924630999565, "lr": 1.98046875e-06, "objective/entropy": -49.397239685058594, "objective/kl": 32.40669250488281, "objective/non_score_reward": -1.6203346252441406, "objective/rlhf_reward": 11.55942153930664, "objective/scores": 13.179756164550781, "policy/approxkl_avg": 0.056858062744140625, "policy/clipfrac_avg": 0.2109375, "policy/entropy_avg": 0.18657711148262024, "step": 87, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2092, "val/ratio": 0.9999588131904602, "val/ratio_var": 9.056823273567716e-07 }, { "episode": 45568, "epoch": 0.1952021932830706, "eps": 7, "loss/policy_avg": 0.04341074824333191, "lr": 1.96875e-06, "objective/entropy": -50.30104064941406, "objective/kl": 31.91788673400879, "objective/non_score_reward": -1.5958943367004395, "objective/rlhf_reward": 11.957464218139648, "objective/scores": 13.553359985351562, "policy/approxkl_avg": 0.054110683500766754, "policy/clipfrac_avg": 0.193359375, "policy/entropy_avg": 0.177315354347229, "step": 88, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2108, "val/ratio": 1.0000953674316406, "val/ratio_var": 8.026260616134095e-07 }, { "episode": 46080, "epoch": 0.1973954763536669, "eps": 7, "loss/policy_avg": 0.060787640511989594, "lr": 1.95703125e-06, "objective/entropy": -45.673580169677734, "objective/kl": 32.983306884765625, "objective/non_score_reward": -1.649165391921997, "objective/rlhf_reward": 11.565816879272461, "objective/scores": 13.214982986450195, "policy/approxkl_avg": 0.061830729246139526, "policy/clipfrac_avg": 0.21875, "policy/entropy_avg": 0.19462166726589203, "step": 89, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2096, "val/ratio": 1.000004768371582, "val/ratio_var": 7.845932259442634e-07 }, { "episode": 46592, "epoch": 0.19958875942426318, "eps": 7, "loss/policy_avg": 0.06810497492551804, "lr": 1.9453125e-06, "objective/entropy": -49.181297302246094, "objective/kl": 32.277130126953125, "objective/non_score_reward": -1.6138564348220825, "objective/rlhf_reward": 11.766349792480469, "objective/scores": 13.380206108093262, "policy/approxkl_avg": 0.05695688724517822, "policy/clipfrac_avg": 0.203125, "policy/entropy_avg": 0.17930074036121368, "step": 90, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2083, "val/ratio": 0.9998986721038818, "val/ratio_var": 8.221533107644063e-07 }, { "episode": 47104, "epoch": 0.20178204249485948, "eps": 7, "loss/policy_avg": 0.055701229721307755, "lr": 1.93359375e-06, "objective/entropy": -45.867530822753906, "objective/kl": 31.220354080200195, "objective/non_score_reward": -1.5610177516937256, "objective/rlhf_reward": 11.842193603515625, "objective/scores": 13.40321159362793, "policy/approxkl_avg": 0.05533070117235184, "policy/clipfrac_avg": 0.173828125, "policy/entropy_avg": 0.18107563257217407, "step": 91, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2156, "val/ratio": 1.0001351833343506, "val/ratio_var": 8.131133313327155e-07 }, { "episode": 47616, "epoch": 0.20397532556545578, "eps": 7, "loss/policy_avg": 0.05227220430970192, "lr": 1.921875e-06, "objective/entropy": -48.75312805175781, "objective/kl": 32.26356506347656, "objective/non_score_reward": -1.6131783723831177, "objective/rlhf_reward": 11.66547966003418, "objective/scores": 13.278657913208008, "policy/approxkl_avg": 0.052241113036870956, "policy/clipfrac_avg": 0.185546875, "policy/entropy_avg": 0.193963885307312, "step": 92, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2168, "val/ratio": 0.999929666519165, "val/ratio_var": 7.763456437714922e-07 }, { "episode": 48128, "epoch": 0.2061686086360521, "eps": 7, "loss/policy_avg": 0.1222270280122757, "lr": 1.91015625e-06, "objective/entropy": -51.514556884765625, "objective/kl": 30.764408111572266, "objective/non_score_reward": -1.5382204055786133, "objective/rlhf_reward": 11.856256484985352, "objective/scores": 13.394476890563965, "policy/approxkl_avg": 0.05982975289225578, "policy/clipfrac_avg": 0.1875, "policy/entropy_avg": 0.1786767691373825, "step": 93, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2209, "val/ratio": 1.000001072883606, "val/ratio_var": 1.080703782463388e-06 }, { "episode": 48640, "epoch": 0.2083618917066484, "eps": 7, "loss/policy_avg": 0.05552385002374649, "lr": 1.8984375e-06, "objective/entropy": -50.57196044921875, "objective/kl": 31.101146697998047, "objective/non_score_reward": -1.555057406425476, "objective/rlhf_reward": 11.832839965820312, "objective/scores": 13.387897491455078, "policy/approxkl_avg": 0.056622885167598724, "policy/clipfrac_avg": 0.203125, "policy/entropy_avg": 0.1913204789161682, "step": 94, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2301, "val/ratio": 1.000207781791687, "val/ratio_var": 1.146110207628226e-06 }, { "episode": 49152, "epoch": 0.2105551747772447, "eps": 7, "loss/policy_avg": 0.06297577172517776, "lr": 1.8867187500000001e-06, "objective/entropy": -46.39762878417969, "objective/kl": 31.752248764038086, "objective/non_score_reward": -1.587612509727478, "objective/rlhf_reward": 11.969014167785645, "objective/scores": 13.556626319885254, "policy/approxkl_avg": 0.05632961913943291, "policy/clipfrac_avg": 0.2109375, "policy/entropy_avg": 0.18543684482574463, "step": 95, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2353, "val/ratio": 1.0002882480621338, "val/ratio_var": 1.0207173772869282e-06 }, { "episode": 49664, "epoch": 0.212748457847841, "eps": 7, "loss/policy_avg": 0.1230374351143837, "lr": 1.875e-06, "objective/entropy": -49.59297180175781, "objective/kl": 31.085838317871094, "objective/non_score_reward": -1.554291844367981, "objective/rlhf_reward": 11.586030960083008, "objective/scores": 13.140321731567383, "policy/approxkl_avg": 0.05181390047073364, "policy/clipfrac_avg": 0.15625, "policy/entropy_avg": 0.2025001049041748, "step": 96, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2391, "val/ratio": 0.9999078512191772, "val/ratio_var": 8.279580470116343e-07 }, { "episode": 50176, "epoch": 0.2149417409184373, "eps": 7, "loss/policy_avg": 0.09635881334543228, "lr": 1.86328125e-06, "objective/entropy": -49.964263916015625, "objective/kl": 31.440513610839844, "objective/non_score_reward": -1.5720256567001343, "objective/rlhf_reward": 11.82887077331543, "objective/scores": 13.400896072387695, "policy/approxkl_avg": 0.08038289844989777, "policy/clipfrac_avg": 0.1953125, "policy/entropy_avg": 0.19088959693908691, "step": 97, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2472, "val/ratio": 0.9999905228614807, "val/ratio_var": 5.620386787086318e-07 }, { "episode": 50688, "epoch": 0.2171350239890336, "eps": 7, "loss/policy_avg": 0.06588736176490784, "lr": 1.8515625000000001e-06, "objective/entropy": -47.49836730957031, "objective/kl": 32.423866271972656, "objective/non_score_reward": -1.6211934089660645, "objective/rlhf_reward": 11.744735717773438, "objective/scores": 13.365928649902344, "policy/approxkl_avg": 0.05758340656757355, "policy/clipfrac_avg": 0.21875, "policy/entropy_avg": 0.2113109529018402, "step": 98, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2630, "val/ratio": 1.0000933408737183, "val/ratio_var": 6.861681072223291e-07 }, { "episode": 51200, "epoch": 0.21932830705962988, "eps": 7, "loss/policy_avg": 0.04643254354596138, "lr": 1.83984375e-06, "objective/entropy": -48.634422302246094, "objective/kl": 30.64898681640625, "objective/non_score_reward": -1.5324492454528809, "objective/rlhf_reward": 12.014122009277344, "objective/scores": 13.54656982421875, "policy/approxkl_avg": 0.049847401678562164, "policy/clipfrac_avg": 0.197265625, "policy/entropy_avg": 0.20576263964176178, "step": 99, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2796, "val/ratio": 0.9998978972434998, "val/ratio_var": 7.073323331496795e-07 }, { "episode": 51712, "epoch": 0.22152159013022618, "eps": 7, "loss/policy_avg": 0.062006328254938126, "lr": 1.828125e-06, "objective/entropy": -44.873626708984375, "objective/kl": 32.74094772338867, "objective/non_score_reward": -1.637047529220581, "objective/rlhf_reward": 11.676469802856445, "objective/scores": 13.313518524169922, "policy/approxkl_avg": 0.05966964364051819, "policy/clipfrac_avg": 0.2109375, "policy/entropy_avg": 0.20745235681533813, "step": 100, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2748, "val/ratio": 1.0000770092010498, "val/ratio_var": 8.392998438466748e-07 }, { "episode": 52224, "epoch": 0.22371487320082248, "eps": 7, "loss/policy_avg": 0.07344357669353485, "lr": 1.81640625e-06, "objective/entropy": -48.71068572998047, "objective/kl": 31.158803939819336, "objective/non_score_reward": -1.5579402446746826, "objective/rlhf_reward": 12.125553131103516, "objective/scores": 13.683492660522461, "policy/approxkl_avg": 0.07100490480661392, "policy/clipfrac_avg": 0.189453125, "policy/entropy_avg": 0.21341751515865326, "step": 101, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2885, "val/ratio": 1.00006103515625, "val/ratio_var": 7.003912401160051e-07 }, { "episode": 52736, "epoch": 0.22590815627141878, "eps": 7, "loss/policy_avg": 0.06474415212869644, "lr": 1.8046875e-06, "objective/entropy": -48.79777908325195, "objective/kl": 31.459522247314453, "objective/non_score_reward": -1.5729761123657227, "objective/rlhf_reward": 11.911396026611328, "objective/scores": 13.48437213897705, "policy/approxkl_avg": 0.05279037356376648, "policy/clipfrac_avg": 0.1875, "policy/entropy_avg": 0.21712711453437805, "step": 102, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2853, "val/ratio": 0.999960720539093, "val/ratio_var": 6.893490649417799e-07 }, { "episode": 53248, "epoch": 0.22810143934201507, "eps": 7, "loss/policy_avg": 0.02789628878235817, "lr": 1.79296875e-06, "objective/entropy": -48.6159553527832, "objective/kl": 32.19043731689453, "objective/non_score_reward": -1.6095221042633057, "objective/rlhf_reward": 11.970061302185059, "objective/scores": 13.579584121704102, "policy/approxkl_avg": 0.06396406888961792, "policy/clipfrac_avg": 0.20703125, "policy/entropy_avg": 0.21447110176086426, "step": 103, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2813, "val/ratio": 0.9999403357505798, "val/ratio_var": 7.448740575455304e-07 }, { "episode": 53760, "epoch": 0.23029472241261137, "eps": 7, "loss/policy_avg": 0.04925355315208435, "lr": 1.78125e-06, "objective/entropy": -48.03702926635742, "objective/kl": 31.010663986206055, "objective/non_score_reward": -1.550533413887024, "objective/rlhf_reward": 11.89893627166748, "objective/scores": 13.449468612670898, "policy/approxkl_avg": 0.04896912723779678, "policy/clipfrac_avg": 0.2109375, "policy/entropy_avg": 0.21880042552947998, "step": 104, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2957, "val/ratio": 1.000004529953003, "val/ratio_var": 8.116120966406015e-07 }, { "episode": 54272, "epoch": 0.23248800548320767, "eps": 7, "loss/policy_avg": 0.06632091104984283, "lr": 1.76953125e-06, "objective/entropy": -47.25006866455078, "objective/kl": 31.43054962158203, "objective/non_score_reward": -1.5715274810791016, "objective/rlhf_reward": 11.807076454162598, "objective/scores": 13.3786039352417, "policy/approxkl_avg": 0.066850945353508, "policy/clipfrac_avg": 0.208984375, "policy/entropy_avg": 0.21371319890022278, "step": 105, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2918, "val/ratio": 0.999915361404419, "val/ratio_var": 1.0229657618765486e-06 }, { "episode": 54784, "epoch": 0.23468128855380396, "eps": 7, "loss/policy_avg": 0.06832720339298248, "lr": 1.7578125e-06, "objective/entropy": -49.64140319824219, "objective/kl": 31.527286529541016, "objective/non_score_reward": -1.5763643980026245, "objective/rlhf_reward": 11.714372634887695, "objective/scores": 13.290736198425293, "policy/approxkl_avg": 0.0627245381474495, "policy/clipfrac_avg": 0.189453125, "policy/entropy_avg": 0.21471458673477173, "step": 106, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3010, "val/ratio": 0.9998877048492432, "val/ratio_var": 6.754234505024215e-07 }, { "episode": 55296, "epoch": 0.2368745716244003, "eps": 7, "loss/policy_avg": 0.05563385412096977, "lr": 1.74609375e-06, "objective/entropy": -48.93874740600586, "objective/kl": 31.15831756591797, "objective/non_score_reward": -1.5579159259796143, "objective/rlhf_reward": 12.166226387023926, "objective/scores": 13.724142074584961, "policy/approxkl_avg": 0.04459884762763977, "policy/clipfrac_avg": 0.181640625, "policy/entropy_avg": 0.21589021384716034, "step": 107, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3050, "val/ratio": 1.0001426935195923, "val/ratio_var": 8.754975056035619e-07 }, { "episode": 55808, "epoch": 0.23906785469499658, "eps": 7, "loss/policy_avg": 0.07894216477870941, "lr": 1.734375e-06, "objective/entropy": -48.76978302001953, "objective/kl": 31.012630462646484, "objective/non_score_reward": -1.5506315231323242, "objective/rlhf_reward": 11.895812034606934, "objective/scores": 13.446443557739258, "policy/approxkl_avg": 0.0571858212351799, "policy/clipfrac_avg": 0.224609375, "policy/entropy_avg": 0.21034139394760132, "step": 108, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 2946, "val/ratio": 0.9999710321426392, "val/ratio_var": 1.362170110041916e-06 }, { "episode": 56320, "epoch": 0.24126113776559288, "eps": 7, "loss/policy_avg": 0.05730992928147316, "lr": 1.72265625e-06, "objective/entropy": -50.21818923950195, "objective/kl": 31.104856491088867, "objective/non_score_reward": -1.555242896080017, "objective/rlhf_reward": 11.923521041870117, "objective/scores": 13.478763580322266, "policy/approxkl_avg": 0.06301895529031754, "policy/clipfrac_avg": 0.22265625, "policy/entropy_avg": 0.20434075593948364, "step": 109, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3144, "val/ratio": 0.9999529123306274, "val/ratio_var": 8.132472544275515e-07 }, { "episode": 56832, "epoch": 0.24345442083618918, "eps": 7, "loss/policy_avg": 0.05458497628569603, "lr": 1.7109375e-06, "objective/entropy": -49.27418899536133, "objective/kl": 30.324630737304688, "objective/non_score_reward": -1.5162315368652344, "objective/rlhf_reward": 12.01490592956543, "objective/scores": 13.531137466430664, "policy/approxkl_avg": 0.052419357001781464, "policy/clipfrac_avg": 0.201171875, "policy/entropy_avg": 0.21094170212745667, "step": 110, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3055, "val/ratio": 1.0001752376556396, "val/ratio_var": 1.080083507076779e-06 }, { "episode": 57344, "epoch": 0.24564770390678548, "eps": 7, "loss/policy_avg": 0.038599204272031784, "lr": 1.69921875e-06, "objective/entropy": -46.80632781982422, "objective/kl": 31.30860137939453, "objective/non_score_reward": -1.5654300451278687, "objective/rlhf_reward": 11.990469932556152, "objective/scores": 13.555900573730469, "policy/approxkl_avg": 0.05481341481208801, "policy/clipfrac_avg": 0.1953125, "policy/entropy_avg": 0.21038272976875305, "step": 111, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3174, "val/ratio": 1.00007164478302, "val/ratio_var": 8.40473433072475e-07 }, { "episode": 57856, "epoch": 0.24784098697738177, "eps": 7, "loss/policy_avg": 0.06401612609624863, "lr": 1.6875e-06, "objective/entropy": -49.942535400390625, "objective/kl": 31.591415405273438, "objective/non_score_reward": -1.5795707702636719, "objective/rlhf_reward": 11.994038581848145, "objective/scores": 13.573609352111816, "policy/approxkl_avg": 0.05404981970787048, "policy/clipfrac_avg": 0.189453125, "policy/entropy_avg": 0.2190382182598114, "step": 112, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3124, "val/ratio": 1.0002111196517944, "val/ratio_var": 7.723161843387061e-07 }, { "episode": 58368, "epoch": 0.25003427004797807, "eps": 7, "loss/policy_avg": 0.03556077182292938, "lr": 1.67578125e-06, "objective/entropy": -50.223976135253906, "objective/kl": 31.2110538482666, "objective/non_score_reward": -1.560552716255188, "objective/rlhf_reward": 11.979619979858398, "objective/scores": 13.540172576904297, "policy/approxkl_avg": 0.05359281599521637, "policy/clipfrac_avg": 0.216796875, "policy/entropy_avg": 0.20700502395629883, "step": 113, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3152, "val/ratio": 1.000028371810913, "val/ratio_var": 8.244367677434639e-07 }, { "episode": 58880, "epoch": 0.25222755311857437, "eps": 7, "loss/policy_avg": 0.010434262454509735, "lr": 1.6640625e-06, "objective/entropy": -46.467777252197266, "objective/kl": 31.24257469177246, "objective/non_score_reward": -1.5621287822723389, "objective/rlhf_reward": 11.877924919128418, "objective/scores": 13.440053939819336, "policy/approxkl_avg": 0.06819692254066467, "policy/clipfrac_avg": 0.234375, "policy/entropy_avg": 0.2004886418581009, "step": 114, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3142, "val/ratio": 0.9997619390487671, "val/ratio_var": 7.186379207269056e-07 }, { "episode": 59392, "epoch": 0.25442083618917066, "eps": 7, "loss/policy_avg": 0.036439333111047745, "lr": 1.6523437500000001e-06, "objective/entropy": -49.938045501708984, "objective/kl": 30.839561462402344, "objective/non_score_reward": -1.5419780015945435, "objective/rlhf_reward": 11.90538501739502, "objective/scores": 13.447362899780273, "policy/approxkl_avg": 0.05128248780965805, "policy/clipfrac_avg": 0.201171875, "policy/entropy_avg": 0.20895111560821533, "step": 115, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3327, "val/ratio": 1.0000596046447754, "val/ratio_var": 8.727883482606558e-07 }, { "episode": 59904, "epoch": 0.25661411925976696, "eps": 7, "loss/policy_avg": 0.03798438236117363, "lr": 1.640625e-06, "objective/entropy": -49.26465606689453, "objective/kl": 29.985055923461914, "objective/non_score_reward": -1.4992527961730957, "objective/rlhf_reward": 12.096814155578613, "objective/scores": 13.59606647491455, "policy/approxkl_avg": 0.058065369725227356, "policy/clipfrac_avg": 0.203125, "policy/entropy_avg": 0.20883342623710632, "step": 116, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3162, "val/ratio": 0.9998736381530762, "val/ratio_var": 7.116667575246538e-07 }, { "episode": 60416, "epoch": 0.25880740233036326, "eps": 7, "loss/policy_avg": 0.07099050283432007, "lr": 1.62890625e-06, "objective/entropy": -47.13051223754883, "objective/kl": 31.923290252685547, "objective/non_score_reward": -1.5961644649505615, "objective/rlhf_reward": 11.913225173950195, "objective/scores": 13.509389877319336, "policy/approxkl_avg": 0.04841821268200874, "policy/clipfrac_avg": 0.181640625, "policy/entropy_avg": 0.202653706073761, "step": 117, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3143, "val/ratio": 1.0000693798065186, "val/ratio_var": 5.713557698072691e-07 }, { "episode": 60928, "epoch": 0.26100068540095955, "eps": 7, "loss/policy_avg": 0.029071252793073654, "lr": 1.6171875000000001e-06, "objective/entropy": -52.13547134399414, "objective/kl": 29.580509185791016, "objective/non_score_reward": -1.4790253639221191, "objective/rlhf_reward": 12.157207489013672, "objective/scores": 13.636232376098633, "policy/approxkl_avg": 0.04795447736978531, "policy/clipfrac_avg": 0.16796875, "policy/entropy_avg": 0.1991284042596817, "step": 118, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3306, "val/ratio": 0.9998592138290405, "val/ratio_var": 5.243380769570649e-07 }, { "episode": 61440, "epoch": 0.26319396847155585, "eps": 7, "loss/policy_avg": 0.029355045408010483, "lr": 1.60546875e-06, "objective/entropy": -51.54493713378906, "objective/kl": 31.228269577026367, "objective/non_score_reward": -1.5614135265350342, "objective/rlhf_reward": 11.880817413330078, "objective/scores": 13.442231178283691, "policy/approxkl_avg": 0.05728745460510254, "policy/clipfrac_avg": 0.216796875, "policy/entropy_avg": 0.21894299983978271, "step": 119, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3491, "val/ratio": 1.0003288984298706, "val/ratio_var": 4.954263204126619e-06 }, { "episode": 61952, "epoch": 0.26538725154215215, "eps": 7, "loss/policy_avg": 0.017333194613456726, "lr": 1.59375e-06, "objective/entropy": -51.9672966003418, "objective/kl": 30.101791381835938, "objective/non_score_reward": -1.5050896406173706, "objective/rlhf_reward": 12.025577545166016, "objective/scores": 13.530667304992676, "policy/approxkl_avg": 0.0607779435813427, "policy/clipfrac_avg": 0.181640625, "policy/entropy_avg": 0.2030687928199768, "step": 120, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3547, "val/ratio": 0.9999595880508423, "val/ratio_var": 6.275835175983957e-07 }, { "episode": 62464, "epoch": 0.26758053461274844, "eps": 7, "loss/policy_avg": 0.04087621718645096, "lr": 1.5820312500000001e-06, "objective/entropy": -51.47197723388672, "objective/kl": 30.397109985351562, "objective/non_score_reward": -1.5198554992675781, "objective/rlhf_reward": 12.032506942749023, "objective/scores": 13.552363395690918, "policy/approxkl_avg": 0.041269607841968536, "policy/clipfrac_avg": 0.1875, "policy/entropy_avg": 0.20843926072120667, "step": 121, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3534, "val/ratio": 1.0000314712524414, "val/ratio_var": 6.925745310581988e-07 }, { "episode": 62976, "epoch": 0.26977381768334474, "eps": 7, "loss/policy_avg": 0.07551665604114532, "lr": 1.5703125e-06, "objective/entropy": -51.15126037597656, "objective/kl": 30.321029663085938, "objective/non_score_reward": -1.5160515308380127, "objective/rlhf_reward": 11.992143630981445, "objective/scores": 13.508195877075195, "policy/approxkl_avg": 0.052853114902973175, "policy/clipfrac_avg": 0.171875, "policy/entropy_avg": 0.2155429720878601, "step": 122, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3662, "val/ratio": 1.0000662803649902, "val/ratio_var": 1.1502866072987672e-06 }, { "episode": 63488, "epoch": 0.27196710075394104, "eps": 7, "loss/policy_avg": 0.03800806403160095, "lr": 1.55859375e-06, "objective/entropy": -49.645782470703125, "objective/kl": 30.97983169555664, "objective/non_score_reward": -1.5489915609359741, "objective/rlhf_reward": 12.060916900634766, "objective/scores": 13.609909057617188, "policy/approxkl_avg": 0.04618716239929199, "policy/clipfrac_avg": 0.201171875, "policy/entropy_avg": 0.2078637331724167, "step": 123, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3616, "val/ratio": 0.9998440742492676, "val/ratio_var": 7.410563398480008e-07 }, { "episode": 64000, "epoch": 0.27416038382453733, "eps": 7, "loss/policy_avg": 0.032623402774333954, "lr": 1.5468750000000001e-06, "objective/entropy": -51.170257568359375, "objective/kl": 31.303388595581055, "objective/non_score_reward": -1.5651694536209106, "objective/rlhf_reward": 11.771775245666504, "objective/scores": 13.336943626403809, "policy/approxkl_avg": 0.05362207815051079, "policy/clipfrac_avg": 0.208984375, "policy/entropy_avg": 0.21997956931591034, "step": 124, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3862, "val/ratio": 1.000096321105957, "val/ratio_var": 7.600924618600402e-07 }, { "episode": 64512, "epoch": 0.27635366689513363, "eps": 7, "loss/policy_avg": 0.031085459515452385, "lr": 1.53515625e-06, "objective/entropy": -49.02000427246094, "objective/kl": 31.002422332763672, "objective/non_score_reward": -1.5501210689544678, "objective/rlhf_reward": 12.034141540527344, "objective/scores": 13.58426284790039, "policy/approxkl_avg": 0.05696885287761688, "policy/clipfrac_avg": 0.18359375, "policy/entropy_avg": 0.20467829704284668, "step": 125, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3761, "val/ratio": 1.0001591444015503, "val/ratio_var": 3.589534117054427e-06 }, { "episode": 65024, "epoch": 0.27854694996572993, "eps": 7, "loss/policy_avg": 0.07421186566352844, "lr": 1.5234375e-06, "objective/entropy": -47.81622314453125, "objective/kl": 31.5286922454834, "objective/non_score_reward": -1.5764347314834595, "objective/rlhf_reward": 11.903038024902344, "objective/scores": 13.479473114013672, "policy/approxkl_avg": 0.05290243774652481, "policy/clipfrac_avg": 0.205078125, "policy/entropy_avg": 0.2120390087366104, "step": 126, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3738, "val/ratio": 0.9998422265052795, "val/ratio_var": 8.801233661870356e-07 }, { "episode": 65536, "epoch": 0.2807402330363262, "eps": 7, "loss/policy_avg": 0.03595343977212906, "lr": 1.5117187500000001e-06, "objective/entropy": -48.716453552246094, "objective/kl": 30.322792053222656, "objective/non_score_reward": -1.5161396265029907, "objective/rlhf_reward": 12.159992218017578, "objective/scores": 13.676132202148438, "policy/approxkl_avg": 0.04783044382929802, "policy/clipfrac_avg": 0.16796875, "policy/entropy_avg": 0.19197297096252441, "step": 127, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3735, "val/ratio": 0.9998753666877747, "val/ratio_var": 7.83547932314832e-07 }, { "episode": 66048, "epoch": 0.2829335161069225, "eps": 7, "loss/policy_avg": 0.05838695913553238, "lr": 1.5e-06, "objective/entropy": -48.72858428955078, "objective/kl": 31.214242935180664, "objective/non_score_reward": -1.560712218284607, "objective/rlhf_reward": 12.080748558044434, "objective/scores": 13.641460418701172, "policy/approxkl_avg": 0.05561618506908417, "policy/clipfrac_avg": 0.185546875, "policy/entropy_avg": 0.1964370757341385, "step": 128, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3712, "val/ratio": 0.999783456325531, "val/ratio_var": 6.5260616111118e-07 }, { "episode": 66560, "epoch": 0.2851267991775189, "eps": 7, "loss/policy_avg": 0.01231492031365633, "lr": 1.48828125e-06, "objective/entropy": -46.436729431152344, "objective/kl": 31.581687927246094, "objective/non_score_reward": -1.5790843963623047, "objective/rlhf_reward": 12.098869323730469, "objective/scores": 13.67795467376709, "policy/approxkl_avg": 0.05094413086771965, "policy/clipfrac_avg": 0.169921875, "policy/entropy_avg": 0.19242434203624725, "step": 129, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3583, "val/ratio": 1.0000998973846436, "val/ratio_var": 8.682492875777825e-07 }, { "episode": 67072, "epoch": 0.28732008224811517, "eps": 7, "loss/policy_avg": 0.061362750828266144, "lr": 1.4765625e-06, "objective/entropy": -47.93026351928711, "objective/kl": 31.09593963623047, "objective/non_score_reward": -1.5547969341278076, "objective/rlhf_reward": 11.994403839111328, "objective/scores": 13.549201011657715, "policy/approxkl_avg": 0.08151525259017944, "policy/clipfrac_avg": 0.189453125, "policy/entropy_avg": 0.200167715549469, "step": 130, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3685, "val/ratio": 0.999895453453064, "val/ratio_var": 9.56301391852321e-07 }, { "episode": 67584, "epoch": 0.28951336531871147, "eps": 7, "loss/policy_avg": 0.04139713570475578, "lr": 1.46484375e-06, "objective/entropy": -46.84960174560547, "objective/kl": 31.893877029418945, "objective/non_score_reward": -1.5946937799453735, "objective/rlhf_reward": 12.212833404541016, "objective/scores": 13.807527542114258, "policy/approxkl_avg": 0.03567758947610855, "policy/clipfrac_avg": 0.158203125, "policy/entropy_avg": 0.20527958869934082, "step": 131, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3760, "val/ratio": 0.9999046325683594, "val/ratio_var": 5.736228558816947e-07 }, { "episode": 68096, "epoch": 0.29170664838930777, "eps": 6, "loss/policy_avg": 0.015449430793523788, "lr": 1.453125e-06, "objective/entropy": -46.26481246948242, "objective/kl": 31.340667724609375, "objective/non_score_reward": -1.5670334100723267, "objective/rlhf_reward": 12.008962631225586, "objective/scores": 13.575996398925781, "policy/approxkl_avg": 0.06165052950382233, "policy/clipfrac_avg": 0.220703125, "policy/entropy_avg": 0.19051343202590942, "step": 132, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3637, "val/ratio": 1.000138282775879, "val/ratio_var": 8.083251827883942e-07 }, { "episode": 68608, "epoch": 0.29389993145990406, "eps": 6, "loss/policy_avg": 0.08820777386426926, "lr": 1.44140625e-06, "objective/entropy": -45.314002990722656, "objective/kl": 31.91384506225586, "objective/non_score_reward": -1.5956922769546509, "objective/rlhf_reward": 11.910893440246582, "objective/scores": 13.506586074829102, "policy/approxkl_avg": 0.04902680218219757, "policy/clipfrac_avg": 0.173828125, "policy/entropy_avg": 0.2006945013999939, "step": 133, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3793, "val/ratio": 1.0002683401107788, "val/ratio_var": 8.941294709075009e-07 }, { "episode": 69120, "epoch": 0.29609321453050036, "eps": 6, "loss/policy_avg": 0.06826784461736679, "lr": 1.4296875e-06, "objective/entropy": -47.16330337524414, "objective/kl": 32.027706146240234, "objective/non_score_reward": -1.6013853549957275, "objective/rlhf_reward": 12.007841110229492, "objective/scores": 13.60922622680664, "policy/approxkl_avg": 0.057559214532375336, "policy/clipfrac_avg": 0.205078125, "policy/entropy_avg": 0.20745880901813507, "step": 134, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3756, "val/ratio": 1.000096082687378, "val/ratio_var": 6.857612220301235e-07 }, { "episode": 69632, "epoch": 0.29828649760109666, "eps": 6, "loss/policy_avg": 0.06609275937080383, "lr": 1.41796875e-06, "objective/entropy": -45.43036651611328, "objective/kl": 31.174110412597656, "objective/non_score_reward": -1.5587055683135986, "objective/rlhf_reward": 12.18134880065918, "objective/scores": 13.7400541305542, "policy/approxkl_avg": 0.04403318464756012, "policy/clipfrac_avg": 0.166015625, "policy/entropy_avg": 0.1890881061553955, "step": 135, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3700, "val/ratio": 1.0000509023666382, "val/ratio_var": 6.750472039129818e-07 }, { "episode": 70144, "epoch": 0.30047978067169295, "eps": 6, "loss/policy_avg": 0.04146716743707657, "lr": 1.40625e-06, "objective/entropy": -45.801666259765625, "objective/kl": 32.07624053955078, "objective/non_score_reward": -1.6038119792938232, "objective/rlhf_reward": 12.237869262695312, "objective/scores": 13.841680526733398, "policy/approxkl_avg": 0.052955564111471176, "policy/clipfrac_avg": 0.18359375, "policy/entropy_avg": 0.19059383869171143, "step": 136, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3521, "val/ratio": 1.0001215934753418, "val/ratio_var": 4.5439765017363243e-07 }, { "episode": 70656, "epoch": 0.30267306374228925, "eps": 6, "loss/policy_avg": 0.012542471289634705, "lr": 1.39453125e-06, "objective/entropy": -45.08157730102539, "objective/kl": 31.95647430419922, "objective/non_score_reward": -1.5978238582611084, "objective/rlhf_reward": 12.09853744506836, "objective/scores": 13.69636058807373, "policy/approxkl_avg": 0.05132647603750229, "policy/clipfrac_avg": 0.22265625, "policy/entropy_avg": 0.18995261192321777, "step": 137, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3716, "val/ratio": 1.0001007318496704, "val/ratio_var": 9.187869522975234e-07 }, { "episode": 71168, "epoch": 0.30486634681288555, "eps": 6, "loss/policy_avg": 0.06143996864557266, "lr": 1.3828125e-06, "objective/entropy": -47.724342346191406, "objective/kl": 31.745656967163086, "objective/non_score_reward": -1.5872828960418701, "objective/rlhf_reward": 12.249744415283203, "objective/scores": 13.837027549743652, "policy/approxkl_avg": 0.0399334654211998, "policy/clipfrac_avg": 0.171875, "policy/entropy_avg": 0.18892788887023926, "step": 138, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3768, "val/ratio": 1.0001490116119385, "val/ratio_var": 5.469966026794282e-07 }, { "episode": 71680, "epoch": 0.30705962988348184, "eps": 6, "loss/policy_avg": 0.05115976929664612, "lr": 1.37109375e-06, "objective/entropy": -47.63806915283203, "objective/kl": 31.581478118896484, "objective/non_score_reward": -1.5790740251541138, "objective/rlhf_reward": 12.079187393188477, "objective/scores": 13.6582612991333, "policy/approxkl_avg": 0.04344845563173294, "policy/clipfrac_avg": 0.189453125, "policy/entropy_avg": 0.19101229310035706, "step": 139, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3701, "val/ratio": 1.0000908374786377, "val/ratio_var": 6.132410135251121e-07 }, { "episode": 72192, "epoch": 0.30925291295407814, "eps": 6, "loss/policy_avg": 0.016339406371116638, "lr": 1.359375e-06, "objective/entropy": -46.384056091308594, "objective/kl": 31.883922576904297, "objective/non_score_reward": -1.5941962003707886, "objective/rlhf_reward": 11.978250503540039, "objective/scores": 13.572446823120117, "policy/approxkl_avg": 0.05104593187570572, "policy/clipfrac_avg": 0.1875, "policy/entropy_avg": 0.19467219710350037, "step": 140, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3737, "val/ratio": 1.0000483989715576, "val/ratio_var": 5.585275175690185e-07 }, { "episode": 72704, "epoch": 0.31144619602467444, "eps": 6, "loss/policy_avg": 0.0317990705370903, "lr": 1.34765625e-06, "objective/entropy": -45.89112091064453, "objective/kl": 31.302913665771484, "objective/non_score_reward": -1.56514573097229, "objective/rlhf_reward": 12.107915878295898, "objective/scores": 13.673062324523926, "policy/approxkl_avg": 0.049749329686164856, "policy/clipfrac_avg": 0.169921875, "policy/entropy_avg": 0.1856738030910492, "step": 141, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3801, "val/ratio": 0.9998741745948792, "val/ratio_var": 7.143764833017485e-07 }, { "episode": 73216, "epoch": 0.31363947909527073, "eps": 6, "loss/policy_avg": 0.08547420799732208, "lr": 1.3359375e-06, "objective/entropy": -46.3120231628418, "objective/kl": 32.43736267089844, "objective/non_score_reward": -1.6218681335449219, "objective/rlhf_reward": 12.044233322143555, "objective/scores": 13.66610050201416, "policy/approxkl_avg": 0.04088249430060387, "policy/clipfrac_avg": 0.134765625, "policy/entropy_avg": 0.17669671773910522, "step": 142, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3764, "val/ratio": 1.0001461505889893, "val/ratio_var": 7.019012855380424e-07 }, { "episode": 73728, "epoch": 0.31583276216586703, "eps": 6, "loss/policy_avg": 0.05418046936392784, "lr": 1.32421875e-06, "objective/entropy": -47.57442855834961, "objective/kl": 31.7020263671875, "objective/non_score_reward": -1.5851011276245117, "objective/rlhf_reward": 12.053739547729492, "objective/scores": 13.638840675354004, "policy/approxkl_avg": 0.05347238853573799, "policy/clipfrac_avg": 0.169921875, "policy/entropy_avg": 0.18706151843070984, "step": 143, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3753, "val/ratio": 1.0001330375671387, "val/ratio_var": 5.384726478041557e-07 }, { "episode": 74240, "epoch": 0.31802604523646333, "eps": 6, "loss/policy_avg": 0.03316599130630493, "lr": 1.3125000000000001e-06, "objective/entropy": -46.67085266113281, "objective/kl": 30.501739501953125, "objective/non_score_reward": -1.5250871181488037, "objective/rlhf_reward": 12.170660018920898, "objective/scores": 13.695746421813965, "policy/approxkl_avg": 0.04256698489189148, "policy/clipfrac_avg": 0.1640625, "policy/entropy_avg": 0.1765584498643875, "step": 144, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3793, "val/ratio": 0.9999911189079285, "val/ratio_var": 4.4938965970686695e-07 }, { "episode": 74752, "epoch": 0.3202193283070596, "eps": 6, "loss/policy_avg": 0.0750068873167038, "lr": 1.30078125e-06, "objective/entropy": -45.76214599609375, "objective/kl": 32.935794830322266, "objective/non_score_reward": -1.646789789199829, "objective/rlhf_reward": 11.971220970153809, "objective/scores": 13.618010520935059, "policy/approxkl_avg": 0.048812054097652435, "policy/clipfrac_avg": 0.162109375, "policy/entropy_avg": 0.1770566999912262, "step": 145, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3665, "val/ratio": 0.9999445676803589, "val/ratio_var": 6.499543587779044e-07 }, { "episode": 75264, "epoch": 0.3224126113776559, "eps": 6, "loss/policy_avg": 0.04462973773479462, "lr": 1.2890625e-06, "objective/entropy": -48.1773681640625, "objective/kl": 31.754302978515625, "objective/non_score_reward": -1.5877151489257812, "objective/rlhf_reward": 11.940765380859375, "objective/scores": 13.528480529785156, "policy/approxkl_avg": 0.04459899663925171, "policy/clipfrac_avg": 0.16015625, "policy/entropy_avg": 0.1767611801624298, "step": 146, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3754, "val/ratio": 0.9999408721923828, "val/ratio_var": 6.486638994829264e-07 }, { "episode": 75776, "epoch": 0.3246058944482522, "eps": 6, "loss/policy_avg": 0.045020490884780884, "lr": 1.2773437500000001e-06, "objective/entropy": -48.329246520996094, "objective/kl": 31.735971450805664, "objective/non_score_reward": -1.5867986679077148, "objective/rlhf_reward": 12.37378215789795, "objective/scores": 13.960580825805664, "policy/approxkl_avg": 0.049151502549648285, "policy/clipfrac_avg": 0.162109375, "policy/entropy_avg": 0.1765923798084259, "step": 147, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3932, "val/ratio": 1.000140905380249, "val/ratio_var": 8.023799864531611e-07 }, { "episode": 76288, "epoch": 0.3267991775188485, "eps": 6, "loss/policy_avg": 0.031616922467947006, "lr": 1.265625e-06, "objective/entropy": -46.86928176879883, "objective/kl": 32.57931900024414, "objective/non_score_reward": -1.6289660930633545, "objective/rlhf_reward": 11.952181816101074, "objective/scores": 13.581148147583008, "policy/approxkl_avg": 0.04584876447916031, "policy/clipfrac_avg": 0.1796875, "policy/entropy_avg": 0.18007265031337738, "step": 148, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3726, "val/ratio": 1.0000200271606445, "val/ratio_var": 6.545841984006984e-07 }, { "episode": 76800, "epoch": 0.3289924605894448, "eps": 6, "loss/policy_avg": 0.046732205897569656, "lr": 1.25390625e-06, "objective/entropy": -48.688018798828125, "objective/kl": 32.15502166748047, "objective/non_score_reward": -1.6077510118484497, "objective/rlhf_reward": 12.327468872070312, "objective/scores": 13.935219764709473, "policy/approxkl_avg": 0.04535512998700142, "policy/clipfrac_avg": 0.166015625, "policy/entropy_avg": 0.17792150378227234, "step": 149, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3964, "val/ratio": 1.0000827312469482, "val/ratio_var": 7.98607004526275e-07 }, { "episode": 77312, "epoch": 0.3311857436600411, "eps": 6, "loss/policy_avg": 0.055171530693769455, "lr": 1.2421875000000001e-06, "objective/entropy": -47.665096282958984, "objective/kl": 32.77416229248047, "objective/non_score_reward": -1.6387081146240234, "objective/rlhf_reward": 11.97776985168457, "objective/scores": 13.616477966308594, "policy/approxkl_avg": 0.04910632222890854, "policy/clipfrac_avg": 0.171875, "policy/entropy_avg": 0.18203596770763397, "step": 150, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3866, "val/ratio": 1.0001187324523926, "val/ratio_var": 7.911350508038595e-07 }, { "episode": 77824, "epoch": 0.3333790267306374, "eps": 6, "loss/policy_avg": 0.025989336892962456, "lr": 1.23046875e-06, "objective/entropy": -46.58367919921875, "objective/kl": 32.51264953613281, "objective/non_score_reward": -1.625632643699646, "objective/rlhf_reward": 12.152078628540039, "objective/scores": 13.777711868286133, "policy/approxkl_avg": 0.04476388543844223, "policy/clipfrac_avg": 0.197265625, "policy/entropy_avg": 0.17051802575588226, "step": 151, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3766, "val/ratio": 1.0001044273376465, "val/ratio_var": 1.0048654530692147e-06 }, { "episode": 78336, "epoch": 0.3355723098012337, "eps": 6, "loss/policy_avg": 0.04051756486296654, "lr": 1.21875e-06, "objective/entropy": -45.364601135253906, "objective/kl": 33.264549255371094, "objective/non_score_reward": -1.6632275581359863, "objective/rlhf_reward": 12.1052885055542, "objective/scores": 13.768516540527344, "policy/approxkl_avg": 0.0416666604578495, "policy/clipfrac_avg": 0.181640625, "policy/entropy_avg": 0.16822485625743866, "step": 152, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3805, "val/ratio": 0.9999629855155945, "val/ratio_var": 4.985236614629684e-07 }, { "episode": 78848, "epoch": 0.33776559287183, "eps": 6, "loss/policy_avg": 0.0353720486164093, "lr": 1.2070312500000001e-06, "objective/entropy": -46.887535095214844, "objective/kl": 33.40437316894531, "objective/non_score_reward": -1.670218586921692, "objective/rlhf_reward": 12.102897644042969, "objective/scores": 13.773115158081055, "policy/approxkl_avg": 0.042422834783792496, "policy/clipfrac_avg": 0.169921875, "policy/entropy_avg": 0.16671255230903625, "step": 153, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3630, "val/ratio": 1.0000197887420654, "val/ratio_var": 5.721728939533932e-07 }, { "episode": 79360, "epoch": 0.3399588759424263, "eps": 6, "loss/policy_avg": 0.03647676482796669, "lr": 1.1953125e-06, "objective/entropy": -45.58281707763672, "objective/kl": 33.2143669128418, "objective/non_score_reward": -1.6607184410095215, "objective/rlhf_reward": 11.99952220916748, "objective/scores": 13.660240173339844, "policy/approxkl_avg": 0.06722655892372131, "policy/clipfrac_avg": 0.19921875, "policy/entropy_avg": 0.17539045214653015, "step": 154, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3841, "val/ratio": 1.0002102851867676, "val/ratio_var": 1.4592494608223205e-06 }, { "episode": 79872, "epoch": 0.3421521590130226, "eps": 6, "loss/policy_avg": 0.07390489429235458, "lr": 1.18359375e-06, "objective/entropy": -48.43956756591797, "objective/kl": 32.89295196533203, "objective/non_score_reward": -1.6446475982666016, "objective/rlhf_reward": 12.093433380126953, "objective/scores": 13.738080978393555, "policy/approxkl_avg": 0.03513135761022568, "policy/clipfrac_avg": 0.171875, "policy/entropy_avg": 0.17300638556480408, "step": 155, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3858, "val/ratio": 1.0000462532043457, "val/ratio_var": 7.432458346556814e-07 }, { "episode": 80384, "epoch": 0.3443454420836189, "eps": 6, "loss/policy_avg": 0.062229886651039124, "lr": 1.1718750000000001e-06, "objective/entropy": -47.075225830078125, "objective/kl": 33.54029846191406, "objective/non_score_reward": -1.677014946937561, "objective/rlhf_reward": 12.176800727844238, "objective/scores": 13.853815078735352, "policy/approxkl_avg": 0.05141585320234299, "policy/clipfrac_avg": 0.1796875, "policy/entropy_avg": 0.16987097263336182, "step": 156, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3846, "val/ratio": 1.0001312494277954, "val/ratio_var": 6.718596523569431e-07 }, { "episode": 80896, "epoch": 0.34653872515421524, "eps": 6, "loss/policy_avg": 0.06592176854610443, "lr": 1.16015625e-06, "objective/entropy": -45.0014533996582, "objective/kl": 33.521873474121094, "objective/non_score_reward": -1.6760938167572021, "objective/rlhf_reward": 11.935114860534668, "objective/scores": 13.611207962036133, "policy/approxkl_avg": 0.045386653393507004, "policy/clipfrac_avg": 0.158203125, "policy/entropy_avg": 0.16817906498908997, "step": 157, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3799, "val/ratio": 1.000162124633789, "val/ratio_var": 6.280766342570132e-07 }, { "episode": 81408, "epoch": 0.34873200822481154, "eps": 6, "loss/policy_avg": 0.017087705433368683, "lr": 1.1484375e-06, "objective/entropy": -44.60365676879883, "objective/kl": 33.34742736816406, "objective/non_score_reward": -1.6673715114593506, "objective/rlhf_reward": 12.169751167297363, "objective/scores": 13.837122917175293, "policy/approxkl_avg": 0.042451776564121246, "policy/clipfrac_avg": 0.138671875, "policy/entropy_avg": 0.1639825701713562, "step": 158, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3827, "val/ratio": 0.9999348521232605, "val/ratio_var": 5.52525762032019e-07 }, { "episode": 81920, "epoch": 0.35092529129540784, "eps": 6, "loss/policy_avg": 0.030591148883104324, "lr": 1.13671875e-06, "objective/entropy": -45.3563232421875, "objective/kl": 32.8660888671875, "objective/non_score_reward": -1.6433045864105225, "objective/rlhf_reward": 12.179360389709473, "objective/scores": 13.82266616821289, "policy/approxkl_avg": 0.061483725905418396, "policy/clipfrac_avg": 0.185546875, "policy/entropy_avg": 0.1684301197528839, "step": 159, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3832, "val/ratio": 1.000175952911377, "val/ratio_var": 7.2921358196254e-07 }, { "episode": 82432, "epoch": 0.35311857436600413, "eps": 6, "loss/policy_avg": 0.04850190877914429, "lr": 1.125e-06, "objective/entropy": -45.8430290222168, "objective/kl": 32.72407531738281, "objective/non_score_reward": -1.6362038850784302, "objective/rlhf_reward": 11.770318984985352, "objective/scores": 13.406523704528809, "policy/approxkl_avg": 0.04402074217796326, "policy/clipfrac_avg": 0.166015625, "policy/entropy_avg": 0.16379833221435547, "step": 160, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3855, "val/ratio": 1.000004768371582, "val/ratio_var": 7.950519602673012e-07 }, { "episode": 82944, "epoch": 0.35531185743660043, "eps": 6, "loss/policy_avg": 0.04634499549865723, "lr": 1.11328125e-06, "objective/entropy": -44.28105926513672, "objective/kl": 32.60356903076172, "objective/non_score_reward": -1.630178451538086, "objective/rlhf_reward": 12.07042407989502, "objective/scores": 13.700602531433105, "policy/approxkl_avg": 0.04746092110872269, "policy/clipfrac_avg": 0.150390625, "policy/entropy_avg": 0.16100072860717773, "step": 161, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3809, "val/ratio": 0.9998874664306641, "val/ratio_var": 6.901511255819059e-07 }, { "episode": 83456, "epoch": 0.35750514050719673, "eps": 6, "loss/policy_avg": 0.053014010190963745, "lr": 1.1015625e-06, "objective/entropy": -44.34575271606445, "objective/kl": 33.12031555175781, "objective/non_score_reward": -1.6560157537460327, "objective/rlhf_reward": 12.135385513305664, "objective/scores": 13.791401863098145, "policy/approxkl_avg": 0.040675077587366104, "policy/clipfrac_avg": 0.162109375, "policy/entropy_avg": 0.16191574931144714, "step": 162, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3910, "val/ratio": 1.0000176429748535, "val/ratio_var": 7.805236919011804e-07 }, { "episode": 83968, "epoch": 0.359698423577793, "eps": 6, "loss/policy_avg": 0.04132367670536041, "lr": 1.08984375e-06, "objective/entropy": -45.620033264160156, "objective/kl": 32.308406829833984, "objective/non_score_reward": -1.6154203414916992, "objective/rlhf_reward": 12.091035842895508, "objective/scores": 13.706456184387207, "policy/approxkl_avg": 0.03995639830827713, "policy/clipfrac_avg": 0.15234375, "policy/entropy_avg": 0.16175366938114166, "step": 163, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 4079, "val/ratio": 1.0000017881393433, "val/ratio_var": 5.264938067739422e-07 }, { "episode": 84480, "epoch": 0.3618917066483893, "eps": 6, "loss/policy_avg": 0.02604106441140175, "lr": 1.078125e-06, "objective/entropy": -47.93902587890625, "objective/kl": 32.54497528076172, "objective/non_score_reward": -1.6272486448287964, "objective/rlhf_reward": 12.043754577636719, "objective/scores": 13.671003341674805, "policy/approxkl_avg": 0.03975088521838188, "policy/clipfrac_avg": 0.16015625, "policy/entropy_avg": 0.1707342565059662, "step": 164, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 4060, "val/ratio": 1.0001680850982666, "val/ratio_var": 9.182730309476028e-07 }, { "episode": 84992, "epoch": 0.3640849897189856, "eps": 6, "loss/policy_avg": 0.055100806057453156, "lr": 1.06640625e-06, "objective/entropy": -46.63789367675781, "objective/kl": 32.33283996582031, "objective/non_score_reward": -1.6166419982910156, "objective/rlhf_reward": 12.14315414428711, "objective/scores": 13.759796142578125, "policy/approxkl_avg": 0.051204077899456024, "policy/clipfrac_avg": 0.158203125, "policy/entropy_avg": 0.16239672899246216, "step": 165, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3850, "val/ratio": 1.0000569820404053, "val/ratio_var": 9.678119567979593e-07 }, { "episode": 85504, "epoch": 0.3662782727895819, "eps": 6, "loss/policy_avg": 0.06828349828720093, "lr": 1.0546875e-06, "objective/entropy": -48.122962951660156, "objective/kl": 32.6610221862793, "objective/non_score_reward": -1.6330510377883911, "objective/rlhf_reward": 12.024824142456055, "objective/scores": 13.657875061035156, "policy/approxkl_avg": 0.043141476809978485, "policy/clipfrac_avg": 0.119140625, "policy/entropy_avg": 0.17185088992118835, "step": 166, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3976, "val/ratio": 1.000027060508728, "val/ratio_var": 6.513710673061723e-07 }, { "episode": 86016, "epoch": 0.3684715558601782, "eps": 6, "loss/policy_avg": 0.0533636100590229, "lr": 1.04296875e-06, "objective/entropy": -47.8603515625, "objective/kl": 32.19922637939453, "objective/non_score_reward": -1.6099613904953003, "objective/rlhf_reward": 12.20279312133789, "objective/scores": 13.812755584716797, "policy/approxkl_avg": 0.06584059447050095, "policy/clipfrac_avg": 0.16796875, "policy/entropy_avg": 0.16326408088207245, "step": 167, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3902, "val/ratio": 0.999927818775177, "val/ratio_var": 7.594405815325445e-07 }, { "episode": 86528, "epoch": 0.3706648389307745, "eps": 6, "loss/policy_avg": 0.03568360209465027, "lr": 1.03125e-06, "objective/entropy": -46.847633361816406, "objective/kl": 32.593719482421875, "objective/non_score_reward": -1.6296861171722412, "objective/rlhf_reward": 12.053540229797363, "objective/scores": 13.683225631713867, "policy/approxkl_avg": 0.03915669023990631, "policy/clipfrac_avg": 0.1484375, "policy/entropy_avg": 0.16720905900001526, "step": 168, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 4019, "val/ratio": 0.9999830722808838, "val/ratio_var": 6.107963486101653e-07 }, { "episode": 87040, "epoch": 0.3728581220013708, "eps": 6, "loss/policy_avg": 0.02935672365128994, "lr": 1.01953125e-06, "objective/entropy": -47.342681884765625, "objective/kl": 32.99451446533203, "objective/non_score_reward": -1.6497256755828857, "objective/rlhf_reward": 12.179969787597656, "objective/scores": 13.829694747924805, "policy/approxkl_avg": 0.045646052807569504, "policy/clipfrac_avg": 0.1640625, "policy/entropy_avg": 0.17241185903549194, "step": 169, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 4012, "val/ratio": 0.9999792575836182, "val/ratio_var": 6.063951332180295e-07 }, { "episode": 87552, "epoch": 0.3750514050719671, "eps": 6, "loss/policy_avg": 0.04190219193696976, "lr": 1.0078125e-06, "objective/entropy": -47.80221176147461, "objective/kl": 33.12531280517578, "objective/non_score_reward": -1.6562654972076416, "objective/rlhf_reward": 12.086249351501465, "objective/scores": 13.742515563964844, "policy/approxkl_avg": 0.04789485037326813, "policy/clipfrac_avg": 0.177734375, "policy/entropy_avg": 0.16719411313533783, "step": 170, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3887, "val/ratio": 0.9998824596405029, "val/ratio_var": 6.53410666018317e-07 }, { "episode": 88064, "epoch": 0.3772446881425634, "eps": 6, "loss/policy_avg": 0.03752657026052475, "lr": 9.9609375e-07, "objective/entropy": -46.271549224853516, "objective/kl": 34.17296600341797, "objective/non_score_reward": -1.7086482048034668, "objective/rlhf_reward": 12.148984909057617, "objective/scores": 13.857633590698242, "policy/approxkl_avg": 0.04762439802289009, "policy/clipfrac_avg": 0.166015625, "policy/entropy_avg": 0.16880956292152405, "step": 171, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3887, "val/ratio": 0.9999531507492065, "val/ratio_var": 7.634708367731946e-07 }, { "episode": 88576, "epoch": 0.3794379712131597, "eps": 6, "loss/policy_avg": 0.03277211636304855, "lr": 9.84375e-07, "objective/entropy": -45.934234619140625, "objective/kl": 33.02567672729492, "objective/non_score_reward": -1.6512837409973145, "objective/rlhf_reward": 12.084634780883789, "objective/scores": 13.735918998718262, "policy/approxkl_avg": 0.039996273815631866, "policy/clipfrac_avg": 0.185546875, "policy/entropy_avg": 0.16980771720409393, "step": 172, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3856, "val/ratio": 1.0000903606414795, "val/ratio_var": 6.715124527545413e-07 }, { "episode": 89088, "epoch": 0.381631254283756, "eps": 6, "loss/policy_avg": 0.031661853194236755, "lr": 9.7265625e-07, "objective/entropy": -45.18244552612305, "objective/kl": 33.14325714111328, "objective/non_score_reward": -1.6571629047393799, "objective/rlhf_reward": 12.03643798828125, "objective/scores": 13.693601608276367, "policy/approxkl_avg": 0.044380709528923035, "policy/clipfrac_avg": 0.171875, "policy/entropy_avg": 0.1570526361465454, "step": 173, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3754, "val/ratio": 0.9999233484268188, "val/ratio_var": 4.877489914179023e-07 }, { "episode": 89600, "epoch": 0.3838245373543523, "eps": 6, "loss/policy_avg": 0.0002702837809920311, "lr": 9.609375e-07, "objective/entropy": -45.407840728759766, "objective/kl": 33.20573425292969, "objective/non_score_reward": -1.6602869033813477, "objective/rlhf_reward": 12.160284042358398, "objective/scores": 13.82056999206543, "policy/approxkl_avg": 0.03693336248397827, "policy/clipfrac_avg": 0.169921875, "policy/entropy_avg": 0.16722910106182098, "step": 174, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3814, "val/ratio": 0.9998776316642761, "val/ratio_var": 4.7349922738249006e-07 }, { "episode": 90112, "epoch": 0.3860178204249486, "eps": 6, "loss/policy_avg": 0.01162320002913475, "lr": 9.4921875e-07, "objective/entropy": -44.86945343017578, "objective/kl": 34.09245300292969, "objective/non_score_reward": -1.7046226263046265, "objective/rlhf_reward": 12.237493515014648, "objective/scores": 13.942116737365723, "policy/approxkl_avg": 0.04271651431918144, "policy/clipfrac_avg": 0.203125, "policy/entropy_avg": 0.16733479499816895, "step": 175, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3761, "val/ratio": 0.9998403787612915, "val/ratio_var": 4.857267299485102e-07 }, { "episode": 90624, "epoch": 0.3882111034955449, "eps": 6, "loss/policy_avg": 0.017360839992761612, "lr": 9.375e-07, "objective/entropy": -42.75486755371094, "objective/kl": 33.723506927490234, "objective/non_score_reward": -1.6861752271652222, "objective/rlhf_reward": 12.02360725402832, "objective/scores": 13.709782600402832, "policy/approxkl_avg": 0.04002365097403526, "policy/clipfrac_avg": 0.1484375, "policy/entropy_avg": 0.16249923408031464, "step": 176, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3749, "val/ratio": 1.000248670578003, "val/ratio_var": 6.447249347729667e-07 }, { "episode": 91136, "epoch": 0.3904043865661412, "eps": 6, "loss/policy_avg": 0.009092383086681366, "lr": 9.257812500000001e-07, "objective/entropy": -43.39929962158203, "objective/kl": 33.58355712890625, "objective/non_score_reward": -1.6791778802871704, "objective/rlhf_reward": 12.095142364501953, "objective/scores": 13.774320602416992, "policy/approxkl_avg": 0.04662185162305832, "policy/clipfrac_avg": 0.181640625, "policy/entropy_avg": 0.15745332837104797, "step": 177, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3739, "val/ratio": 1.000197172164917, "val/ratio_var": 8.259135597654677e-07 }, { "episode": 91648, "epoch": 0.3925976696367375, "eps": 6, "loss/policy_avg": 0.04981373995542526, "lr": 9.140625e-07, "objective/entropy": -43.12224197387695, "objective/kl": 33.532493591308594, "objective/non_score_reward": -1.6766246557235718, "objective/rlhf_reward": 12.181105613708496, "objective/scores": 13.857730865478516, "policy/approxkl_avg": 0.05293203517794609, "policy/clipfrac_avg": 0.19140625, "policy/entropy_avg": 0.16689547896385193, "step": 178, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3660, "val/ratio": 0.9998425245285034, "val/ratio_var": 6.505214287244598e-07 }, { "episode": 92160, "epoch": 0.3947909527073338, "eps": 6, "loss/policy_avg": 0.004799778573215008, "lr": 9.0234375e-07, "objective/entropy": -42.42271423339844, "objective/kl": 34.58287048339844, "objective/non_score_reward": -1.7291436195373535, "objective/rlhf_reward": 12.183492660522461, "objective/scores": 13.912635803222656, "policy/approxkl_avg": 0.04634490981698036, "policy/clipfrac_avg": 0.19140625, "policy/entropy_avg": 0.16785326600074768, "step": 179, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3563, "val/ratio": 1.0002268552780151, "val/ratio_var": 7.743488481537497e-07 }, { "episode": 92672, "epoch": 0.3969842357779301, "eps": 6, "loss/policy_avg": 0.05486229807138443, "lr": 8.90625e-07, "objective/entropy": -44.227134704589844, "objective/kl": 34.72454071044922, "objective/non_score_reward": -1.736227035522461, "objective/rlhf_reward": 12.078829765319824, "objective/scores": 13.815057754516602, "policy/approxkl_avg": 0.04358441010117531, "policy/clipfrac_avg": 0.193359375, "policy/entropy_avg": 0.1603843867778778, "step": 180, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3578, "val/ratio": 1.0001356601715088, "val/ratio_var": 6.527110372189782e-07 }, { "episode": 93184, "epoch": 0.39917751884852637, "eps": 6, "loss/policy_avg": 0.013856533914804459, "lr": 8.7890625e-07, "objective/entropy": -43.287506103515625, "objective/kl": 34.65958023071289, "objective/non_score_reward": -1.7329790592193604, "objective/rlhf_reward": 12.218202590942383, "objective/scores": 13.951181411743164, "policy/approxkl_avg": 0.05023915693163872, "policy/clipfrac_avg": 0.177734375, "policy/entropy_avg": 0.16447854042053223, "step": 181, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3627, "val/ratio": 1.0000338554382324, "val/ratio_var": 6.67180074742646e-07 }, { "episode": 93696, "epoch": 0.40137080191912267, "eps": 6, "loss/policy_avg": 0.013517491519451141, "lr": 8.671875e-07, "objective/entropy": -44.75653076171875, "objective/kl": 33.92699432373047, "objective/non_score_reward": -1.6963496208190918, "objective/rlhf_reward": 12.361766815185547, "objective/scores": 14.058116912841797, "policy/approxkl_avg": 0.04134466126561165, "policy/clipfrac_avg": 0.1484375, "policy/entropy_avg": 0.163140207529068, "step": 182, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3701, "val/ratio": 0.9998599290847778, "val/ratio_var": 5.315375801728806e-07 }, { "episode": 94208, "epoch": 0.40356408498971896, "eps": 6, "loss/policy_avg": 0.04708061367273331, "lr": 8.5546875e-07, "objective/entropy": -42.86664962768555, "objective/kl": 33.939781188964844, "objective/non_score_reward": -1.6969891786575317, "objective/rlhf_reward": 11.99999713897705, "objective/scores": 13.69698715209961, "policy/approxkl_avg": 0.03837224096059799, "policy/clipfrac_avg": 0.18359375, "policy/entropy_avg": 0.16601251065731049, "step": 183, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3579, "val/ratio": 1.0000877380371094, "val/ratio_var": 5.547088903767872e-07 }, { "episode": 94720, "epoch": 0.40575736806031526, "eps": 6, "loss/policy_avg": 0.02863866463303566, "lr": 8.4375e-07, "objective/entropy": -41.63829803466797, "objective/kl": 35.059444427490234, "objective/non_score_reward": -1.7529723644256592, "objective/rlhf_reward": 12.022567749023438, "objective/scores": 13.775540351867676, "policy/approxkl_avg": 0.05269748345017433, "policy/clipfrac_avg": 0.193359375, "policy/entropy_avg": 0.16794031858444214, "step": 184, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3701, "val/ratio": 0.9999144077301025, "val/ratio_var": 5.281810899759876e-07 }, { "episode": 95232, "epoch": 0.40795065113091156, "eps": 6, "loss/policy_avg": 0.04807991534471512, "lr": 8.3203125e-07, "objective/entropy": -42.897056579589844, "objective/kl": 35.038509368896484, "objective/non_score_reward": -1.7519254684448242, "objective/rlhf_reward": 11.840540885925293, "objective/scores": 13.592466354370117, "policy/approxkl_avg": 0.05131963640451431, "policy/clipfrac_avg": 0.1953125, "policy/entropy_avg": 0.17112389206886292, "step": 185, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3506, "val/ratio": 1.0001037120819092, "val/ratio_var": 1.1791954648288083e-06 }, { "episode": 95744, "epoch": 0.4101439342015079, "eps": 6, "loss/policy_avg": 0.021313758566975594, "lr": 8.203125e-07, "objective/entropy": -41.886253356933594, "objective/kl": 33.41462707519531, "objective/non_score_reward": -1.670731544494629, "objective/rlhf_reward": 12.376485824584961, "objective/scores": 14.047216415405273, "policy/approxkl_avg": 0.04311320185661316, "policy/clipfrac_avg": 0.189453125, "policy/entropy_avg": 0.1567443311214447, "step": 186, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3687, "val/ratio": 0.9999748468399048, "val/ratio_var": 5.931812552262272e-07 }, { "episode": 96256, "epoch": 0.4123372172721042, "eps": 6, "loss/policy_avg": 0.0319538488984108, "lr": 8.085937500000001e-07, "objective/entropy": -46.733802795410156, "objective/kl": 33.61097717285156, "objective/non_score_reward": -1.680548906326294, "objective/rlhf_reward": 11.839179039001465, "objective/scores": 13.51972770690918, "policy/approxkl_avg": 0.0666745975613594, "policy/clipfrac_avg": 0.201171875, "policy/entropy_avg": 0.17065854370594025, "step": 187, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3545, "val/ratio": 0.9998564720153809, "val/ratio_var": 4.6152018740031053e-07 }, { "episode": 96768, "epoch": 0.4145305003427005, "eps": 6, "loss/policy_avg": 0.04184500873088837, "lr": 7.96875e-07, "objective/entropy": -43.91098403930664, "objective/kl": 33.17008590698242, "objective/non_score_reward": -1.6585042476654053, "objective/rlhf_reward": 12.247800827026367, "objective/scores": 13.906305313110352, "policy/approxkl_avg": 0.048506151884794235, "policy/clipfrac_avg": 0.14453125, "policy/entropy_avg": 0.16719208657741547, "step": 188, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3754, "val/ratio": 0.999748945236206, "val/ratio_var": 4.2723382875919924e-07 }, { "episode": 97280, "epoch": 0.4167237834132968, "eps": 6, "loss/policy_avg": 0.04508175328373909, "lr": 7.8515625e-07, "objective/entropy": -44.22284698486328, "objective/kl": 33.55169677734375, "objective/non_score_reward": -1.6775848865509033, "objective/rlhf_reward": 12.015632629394531, "objective/scores": 13.693218231201172, "policy/approxkl_avg": 0.053490396589040756, "policy/clipfrac_avg": 0.1484375, "policy/entropy_avg": 0.17472797632217407, "step": 189, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3656, "val/ratio": 0.999805212020874, "val/ratio_var": 7.422577823490428e-07 }, { "episode": 97792, "epoch": 0.4189170664838931, "eps": 6, "loss/policy_avg": 0.05971541628241539, "lr": 7.734375000000001e-07, "objective/entropy": -45.186431884765625, "objective/kl": 32.71662139892578, "objective/non_score_reward": -1.6358308792114258, "objective/rlhf_reward": 12.227357864379883, "objective/scores": 13.863187789916992, "policy/approxkl_avg": 0.048808276653289795, "policy/clipfrac_avg": 0.146484375, "policy/entropy_avg": 0.1684289276599884, "step": 190, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3526, "val/ratio": 1.0000661611557007, "val/ratio_var": 8.512056410836522e-07 }, { "episode": 98304, "epoch": 0.4211103495544894, "eps": 6, "loss/policy_avg": 0.05087277665734291, "lr": 7.6171875e-07, "objective/entropy": -45.842369079589844, "objective/kl": 32.24187469482422, "objective/non_score_reward": -1.6120936870574951, "objective/rlhf_reward": 12.114307403564453, "objective/scores": 13.726400375366211, "policy/approxkl_avg": 0.038973987102508545, "policy/clipfrac_avg": 0.18359375, "policy/entropy_avg": 0.16362987458705902, "step": 191, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3493, "val/ratio": 0.9998780488967896, "val/ratio_var": 6.553620437443897e-07 }, { "episode": 98816, "epoch": 0.4233036326250857, "eps": 6, "loss/policy_avg": 0.020076967775821686, "lr": 7.5e-07, "objective/entropy": -47.215694427490234, "objective/kl": 32.64807891845703, "objective/non_score_reward": -1.632404088973999, "objective/rlhf_reward": 12.276786804199219, "objective/scores": 13.909191131591797, "policy/approxkl_avg": 0.04490412771701813, "policy/clipfrac_avg": 0.15625, "policy/entropy_avg": 0.17147311568260193, "step": 192, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3601, "val/ratio": 1.000018835067749, "val/ratio_var": 7.1356157604896e-07 }, { "episode": 99328, "epoch": 0.425496915695682, "eps": 6, "loss/policy_avg": 0.0046792663633823395, "lr": 7.3828125e-07, "objective/entropy": -46.06641387939453, "objective/kl": 32.497032165527344, "objective/non_score_reward": -1.6248518228530884, "objective/rlhf_reward": 12.361309051513672, "objective/scores": 13.986160278320312, "policy/approxkl_avg": 0.036182649433612823, "policy/clipfrac_avg": 0.134765625, "policy/entropy_avg": 0.1672893613576889, "step": 193, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3515, "val/ratio": 1.0000345706939697, "val/ratio_var": 7.649219355698733e-07 }, { "episode": 99840, "epoch": 0.4276901987662783, "eps": 6, "loss/policy_avg": 0.029073424637317657, "lr": 7.265625e-07, "objective/entropy": -46.39115905761719, "objective/kl": 32.59736251831055, "objective/non_score_reward": -1.6298680305480957, "objective/rlhf_reward": 12.244403839111328, "objective/scores": 13.874271392822266, "policy/approxkl_avg": 0.038327522575855255, "policy/clipfrac_avg": 0.1328125, "policy/entropy_avg": 0.16699525713920593, "step": 194, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3552, "val/ratio": 0.9999607801437378, "val/ratio_var": 5.253468771115877e-07 }, { "episode": 100352, "epoch": 0.4298834818368746, "eps": 6, "loss/policy_avg": 0.03488843888044357, "lr": 7.1484375e-07, "objective/entropy": -44.52246856689453, "objective/kl": 33.070701599121094, "objective/non_score_reward": -1.65353524684906, "objective/rlhf_reward": 12.094472885131836, "objective/scores": 13.748006820678711, "policy/approxkl_avg": 0.05242970585823059, "policy/clipfrac_avg": 0.177734375, "policy/entropy_avg": 0.17484037578105927, "step": 195, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3767, "val/ratio": 0.9999789595603943, "val/ratio_var": 8.076226549746934e-07 }, { "episode": 100864, "epoch": 0.4320767649074709, "eps": 6, "loss/policy_avg": 0.03715651482343674, "lr": 7.03125e-07, "objective/entropy": -46.404510498046875, "objective/kl": 32.77263641357422, "objective/non_score_reward": -1.638631820678711, "objective/rlhf_reward": 11.959657669067383, "objective/scores": 13.598289489746094, "policy/approxkl_avg": 0.08408388495445251, "policy/clipfrac_avg": 0.185546875, "policy/entropy_avg": 0.16849398612976074, "step": 196, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3600, "val/ratio": 1.0001060962677002, "val/ratio_var": 8.36843753404537e-07 }, { "episode": 101376, "epoch": 0.4342700479780672, "eps": 6, "loss/policy_avg": 0.009729292243719101, "lr": 6.9140625e-07, "objective/entropy": -46.91974639892578, "objective/kl": 32.46935272216797, "objective/non_score_reward": -1.6234676837921143, "objective/rlhf_reward": 12.182170867919922, "objective/scores": 13.805639266967773, "policy/approxkl_avg": 0.038409337401390076, "policy/clipfrac_avg": 0.169921875, "policy/entropy_avg": 0.17712263762950897, "step": 197, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3690, "val/ratio": 0.9999710917472839, "val/ratio_var": 4.84107999909611e-07 }, { "episode": 101888, "epoch": 0.43646333104866347, "eps": 6, "loss/policy_avg": 0.041521988809108734, "lr": 6.796875e-07, "objective/entropy": -47.788963317871094, "objective/kl": 32.61534118652344, "objective/non_score_reward": -1.6307669878005981, "objective/rlhf_reward": 12.14478874206543, "objective/scores": 13.775555610656738, "policy/approxkl_avg": 0.04595543444156647, "policy/clipfrac_avg": 0.177734375, "policy/entropy_avg": 0.1775820106267929, "step": 198, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3967, "val/ratio": 0.9999727010726929, "val/ratio_var": 5.872706196896615e-07 }, { "episode": 102400, "epoch": 0.43865661411925977, "eps": 6, "loss/policy_avg": 0.013786925002932549, "lr": 6.6796875e-07, "objective/entropy": -47.44401168823242, "objective/kl": 31.362531661987305, "objective/non_score_reward": -1.5681265592575073, "objective/rlhf_reward": 12.159399032592773, "objective/scores": 13.727526664733887, "policy/approxkl_avg": 0.04408939927816391, "policy/clipfrac_avg": 0.1875, "policy/entropy_avg": 0.1726619303226471, "step": 199, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3723, "val/ratio": 1.0000035762786865, "val/ratio_var": 8.266886197816348e-07 }, { "episode": 102912, "epoch": 0.44084989718985607, "eps": 6, "loss/policy_avg": 0.02519533410668373, "lr": 6.562500000000001e-07, "objective/entropy": -48.725502014160156, "objective/kl": 31.51070213317871, "objective/non_score_reward": -1.5755351781845093, "objective/rlhf_reward": 12.353096008300781, "objective/scores": 13.928630828857422, "policy/approxkl_avg": 0.035048454999923706, "policy/clipfrac_avg": 0.158203125, "policy/entropy_avg": 0.17824092507362366, "step": 200, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3734, "val/ratio": 1.0000141859054565, "val/ratio_var": 5.391907507146243e-07 }, { "episode": 103424, "epoch": 0.44304318026045236, "eps": 6, "loss/policy_avg": 0.02977118268609047, "lr": 6.4453125e-07, "objective/entropy": -49.671939849853516, "objective/kl": 31.477449417114258, "objective/non_score_reward": -1.5738725662231445, "objective/rlhf_reward": 12.222738265991211, "objective/scores": 13.796609878540039, "policy/approxkl_avg": 0.04726526886224747, "policy/clipfrac_avg": 0.14453125, "policy/entropy_avg": 0.17516601085662842, "step": 201, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3775, "val/ratio": 0.999923586845398, "val/ratio_var": 3.984151533131808e-07 }, { "episode": 103936, "epoch": 0.44523646333104866, "eps": 6, "loss/policy_avg": 0.024648478254675865, "lr": 6.328125e-07, "objective/entropy": -46.37347412109375, "objective/kl": 32.32550048828125, "objective/non_score_reward": -1.6162750720977783, "objective/rlhf_reward": 12.12025260925293, "objective/scores": 13.736526489257812, "policy/approxkl_avg": 0.04179628938436508, "policy/clipfrac_avg": 0.173828125, "policy/entropy_avg": 0.17158494889736176, "step": 202, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3651, "val/ratio": 0.9999801516532898, "val/ratio_var": 6.427766265915125e-07 }, { "episode": 104448, "epoch": 0.44742974640164496, "eps": 6, "loss/policy_avg": 0.01430027186870575, "lr": 6.210937500000001e-07, "objective/entropy": -46.783241271972656, "objective/kl": 31.894420623779297, "objective/non_score_reward": -1.5947210788726807, "objective/rlhf_reward": 12.257794380187988, "objective/scores": 13.85251522064209, "policy/approxkl_avg": 0.04405524581670761, "policy/clipfrac_avg": 0.166015625, "policy/entropy_avg": 0.1646786332130432, "step": 203, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3553, "val/ratio": 0.999962329864502, "val/ratio_var": 4.989540798305825e-07 }, { "episode": 104960, "epoch": 0.44962302947224125, "eps": 6, "loss/policy_avg": 0.017517562955617905, "lr": 6.09375e-07, "objective/entropy": -47.19154357910156, "objective/kl": 32.10340118408203, "objective/non_score_reward": -1.6051701307296753, "objective/rlhf_reward": 12.12353515625, "objective/scores": 13.728705406188965, "policy/approxkl_avg": 0.04066164046525955, "policy/clipfrac_avg": 0.158203125, "policy/entropy_avg": 0.17214670777320862, "step": 204, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3652, "val/ratio": 0.9999274015426636, "val/ratio_var": 4.265529014446656e-07 }, { "episode": 105472, "epoch": 0.45181631254283755, "eps": 6, "loss/policy_avg": 0.035035982728004456, "lr": 5.9765625e-07, "objective/entropy": -46.77722930908203, "objective/kl": 32.88157653808594, "objective/non_score_reward": -1.6440789699554443, "objective/rlhf_reward": 12.112733840942383, "objective/scores": 13.756813049316406, "policy/approxkl_avg": 0.04462597519159317, "policy/clipfrac_avg": 0.1640625, "policy/entropy_avg": 0.1622966229915619, "step": 205, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3642, "val/ratio": 1.0000214576721191, "val/ratio_var": 9.509098504167923e-07 }, { "episode": 105984, "epoch": 0.45400959561343385, "eps": 6, "loss/policy_avg": 0.02712424471974373, "lr": 5.859375000000001e-07, "objective/entropy": -47.40806198120117, "objective/kl": 32.84805679321289, "objective/non_score_reward": -1.6424027681350708, "objective/rlhf_reward": 12.075798988342285, "objective/scores": 13.718201637268066, "policy/approxkl_avg": 0.044767990708351135, "policy/clipfrac_avg": 0.1796875, "policy/entropy_avg": 0.17067429423332214, "step": 206, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3800, "val/ratio": 1.0000507831573486, "val/ratio_var": 5.416270028035797e-07 }, { "episode": 106496, "epoch": 0.45620287868403014, "eps": 6, "loss/policy_avg": 0.020231973379850388, "lr": 5.7421875e-07, "objective/entropy": -46.810264587402344, "objective/kl": 32.574867248535156, "objective/non_score_reward": -1.6287435293197632, "objective/rlhf_reward": 12.219034194946289, "objective/scores": 13.8477783203125, "policy/approxkl_avg": 0.04307110607624054, "policy/clipfrac_avg": 0.16796875, "policy/entropy_avg": 0.16806720197200775, "step": 207, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3683, "val/ratio": 1.0000402927398682, "val/ratio_var": 5.771587439085124e-07 }, { "episode": 107008, "epoch": 0.45839616175462644, "eps": 6, "loss/policy_avg": 0.023262523114681244, "lr": 5.625e-07, "objective/entropy": -47.67873001098633, "objective/kl": 32.05731964111328, "objective/non_score_reward": -1.6028659343719482, "objective/rlhf_reward": 12.151142120361328, "objective/scores": 13.754008293151855, "policy/approxkl_avg": 0.04319004714488983, "policy/clipfrac_avg": 0.181640625, "policy/entropy_avg": 0.17881324887275696, "step": 208, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3817, "val/ratio": 1.0000810623168945, "val/ratio_var": 5.869856636309123e-07 }, { "episode": 107520, "epoch": 0.46058944482522274, "eps": 6, "loss/policy_avg": 0.03269857168197632, "lr": 5.5078125e-07, "objective/entropy": -45.535865783691406, "objective/kl": 31.727317810058594, "objective/non_score_reward": -1.586365818977356, "objective/rlhf_reward": 12.213563919067383, "objective/scores": 13.799930572509766, "policy/approxkl_avg": 0.03817829117178917, "policy/clipfrac_avg": 0.185546875, "policy/entropy_avg": 0.16942265629768372, "step": 209, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3643, "val/ratio": 1.0001146793365479, "val/ratio_var": 5.890616989745467e-07 }, { "episode": 108032, "epoch": 0.46278272789581903, "eps": 6, "loss/policy_avg": 0.06721986085176468, "lr": 5.390625e-07, "objective/entropy": -45.77275085449219, "objective/kl": 32.46781921386719, "objective/non_score_reward": -1.623391032218933, "objective/rlhf_reward": 12.19780158996582, "objective/scores": 13.821191787719727, "policy/approxkl_avg": 0.04397875443100929, "policy/clipfrac_avg": 0.1328125, "policy/entropy_avg": 0.17651620507240295, "step": 210, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3783, "val/ratio": 1.0000160932540894, "val/ratio_var": 7.855404646761599e-07 }, { "episode": 108544, "epoch": 0.46497601096641533, "eps": 6, "loss/policy_avg": 0.026129083707928658, "lr": 5.2734375e-07, "objective/entropy": -46.74835968017578, "objective/kl": 32.184898376464844, "objective/non_score_reward": -1.6092448234558105, "objective/rlhf_reward": 12.210777282714844, "objective/scores": 13.820022583007812, "policy/approxkl_avg": 0.04312296211719513, "policy/clipfrac_avg": 0.18359375, "policy/entropy_avg": 0.16915397346019745, "step": 211, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3821, "val/ratio": 0.9999486804008484, "val/ratio_var": 5.002877969673136e-07 }, { "episode": 109056, "epoch": 0.46716929403701163, "eps": 6, "loss/policy_avg": 0.049704138189554214, "lr": 5.15625e-07, "objective/entropy": -45.145538330078125, "objective/kl": 32.364402770996094, "objective/non_score_reward": -1.6182202100753784, "objective/rlhf_reward": 12.053094863891602, "objective/scores": 13.67131519317627, "policy/approxkl_avg": 0.04747725650668144, "policy/clipfrac_avg": 0.189453125, "policy/entropy_avg": 0.17152246832847595, "step": 212, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3692, "val/ratio": 1.0000265836715698, "val/ratio_var": 7.351875979111355e-07 }, { "episode": 109568, "epoch": 0.4693625771076079, "eps": 6, "loss/policy_avg": 0.023406462743878365, "lr": 5.0390625e-07, "objective/entropy": -44.91341781616211, "objective/kl": 32.712379455566406, "objective/non_score_reward": -1.635619044303894, "objective/rlhf_reward": 12.297149658203125, "objective/scores": 13.932767868041992, "policy/approxkl_avg": 0.043134041130542755, "policy/clipfrac_avg": 0.169921875, "policy/entropy_avg": 0.170152947306633, "step": 213, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3830, "val/ratio": 1.0000108480453491, "val/ratio_var": 5.478732418850996e-07 }, { "episode": 110080, "epoch": 0.4715558601782042, "eps": 6, "loss/policy_avg": 0.025773677974939346, "lr": 4.921875e-07, "objective/entropy": -46.570350646972656, "objective/kl": 32.734310150146484, "objective/non_score_reward": -1.6367154121398926, "objective/rlhf_reward": 12.143056869506836, "objective/scores": 13.77977180480957, "policy/approxkl_avg": 0.06745807826519012, "policy/clipfrac_avg": 0.166015625, "policy/entropy_avg": 0.17417851090431213, "step": 214, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3843, "val/ratio": 1.0001730918884277, "val/ratio_var": 8.961763455772598e-07 }, { "episode": 110592, "epoch": 0.4737491432488006, "eps": 6, "loss/policy_avg": 0.019351951777935028, "lr": 4.8046875e-07, "objective/entropy": -45.34315490722656, "objective/kl": 32.31012725830078, "objective/non_score_reward": -1.615506649017334, "objective/rlhf_reward": 12.019478797912598, "objective/scores": 13.634984970092773, "policy/approxkl_avg": 0.03868725150823593, "policy/clipfrac_avg": 0.171875, "policy/entropy_avg": 0.16609841585159302, "step": 215, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3700, "val/ratio": 0.999989926815033, "val/ratio_var": 4.4976843582844594e-07 }, { "episode": 111104, "epoch": 0.47594242631939687, "eps": 6, "loss/policy_avg": 0.017778418958187103, "lr": 4.6875e-07, "objective/entropy": -44.903564453125, "objective/kl": 33.34663009643555, "objective/non_score_reward": -1.6673314571380615, "objective/rlhf_reward": 12.005241394042969, "objective/scores": 13.67257308959961, "policy/approxkl_avg": 0.04540884494781494, "policy/clipfrac_avg": 0.19140625, "policy/entropy_avg": 0.16867536306381226, "step": 216, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3720, "val/ratio": 0.9999477863311768, "val/ratio_var": 7.796459726705507e-07 }, { "episode": 111616, "epoch": 0.47813570938999317, "eps": 6, "loss/policy_avg": 0.028567474335432053, "lr": 4.5703125e-07, "objective/entropy": -41.529823303222656, "objective/kl": 32.95002746582031, "objective/non_score_reward": -1.6475014686584473, "objective/rlhf_reward": 12.306774139404297, "objective/scores": 13.954275131225586, "policy/approxkl_avg": 0.04487468674778938, "policy/clipfrac_avg": 0.189453125, "policy/entropy_avg": 0.1720014065504074, "step": 217, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3660, "val/ratio": 0.9999500513076782, "val/ratio_var": 8.629873491372564e-07 }, { "episode": 112128, "epoch": 0.48032899246058947, "eps": 6, "loss/policy_avg": 0.018710482865571976, "lr": 4.453125e-07, "objective/entropy": -44.33214569091797, "objective/kl": 33.13552474975586, "objective/non_score_reward": -1.6567763090133667, "objective/rlhf_reward": 12.17362117767334, "objective/scores": 13.830398559570312, "policy/approxkl_avg": 0.036987971514463425, "policy/clipfrac_avg": 0.15234375, "policy/entropy_avg": 0.16948255896568298, "step": 218, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3660, "val/ratio": 0.9999829530715942, "val/ratio_var": 6.785802497688564e-07 }, { "episode": 112640, "epoch": 0.48252227553118576, "eps": 6, "loss/policy_avg": 0.037629470229148865, "lr": 4.3359375e-07, "objective/entropy": -44.514896392822266, "objective/kl": 31.880420684814453, "objective/non_score_reward": -1.5940210819244385, "objective/rlhf_reward": 12.342854499816895, "objective/scores": 13.93687629699707, "policy/approxkl_avg": 0.041482266038656235, "policy/clipfrac_avg": 0.150390625, "policy/entropy_avg": 0.15429215133190155, "step": 219, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3657, "val/ratio": 1.0001139640808105, "val/ratio_var": 5.140203143128019e-07 }, { "episode": 113152, "epoch": 0.48471555860178206, "eps": 6, "loss/policy_avg": 0.042203765362501144, "lr": 4.21875e-07, "objective/entropy": -45.628173828125, "objective/kl": 32.86504364013672, "objective/non_score_reward": -1.6432521343231201, "objective/rlhf_reward": 12.352306365966797, "objective/scores": 13.995558738708496, "policy/approxkl_avg": 0.0412248820066452, "policy/clipfrac_avg": 0.13671875, "policy/entropy_avg": 0.1694689691066742, "step": 220, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3819, "val/ratio": 0.9999035596847534, "val/ratio_var": 5.895665822208684e-07 }, { "episode": 113664, "epoch": 0.48690884167237836, "eps": 6, "loss/policy_avg": 0.03167615458369255, "lr": 4.1015625e-07, "objective/entropy": -43.215972900390625, "objective/kl": 33.125850677490234, "objective/non_score_reward": -1.6562925577163696, "objective/rlhf_reward": 12.137786865234375, "objective/scores": 13.794078826904297, "policy/approxkl_avg": 0.0709940493106842, "policy/clipfrac_avg": 0.1796875, "policy/entropy_avg": 0.16062289476394653, "step": 221, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3614, "val/ratio": 0.9999601244926453, "val/ratio_var": 5.603689601230144e-07 }, { "episode": 114176, "epoch": 0.48910212474297465, "eps": 6, "loss/policy_avg": 0.025735924020409584, "lr": 3.984375e-07, "objective/entropy": -45.04023742675781, "objective/kl": 32.46979522705078, "objective/non_score_reward": -1.6234898567199707, "objective/rlhf_reward": 12.32956314086914, "objective/scores": 13.953052520751953, "policy/approxkl_avg": 0.06509754806756973, "policy/clipfrac_avg": 0.16796875, "policy/entropy_avg": 0.17296773195266724, "step": 222, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3792, "val/ratio": 0.9998451471328735, "val/ratio_var": 8.159282174347027e-07 }, { "episode": 114688, "epoch": 0.49129540781357095, "eps": 6, "loss/policy_avg": 0.041412077844142914, "lr": 3.8671875000000003e-07, "objective/entropy": -44.9599723815918, "objective/kl": 32.5138053894043, "objective/non_score_reward": -1.6256904602050781, "objective/rlhf_reward": 12.156733512878418, "objective/scores": 13.782423973083496, "policy/approxkl_avg": 0.0447232685983181, "policy/clipfrac_avg": 0.1640625, "policy/entropy_avg": 0.16872379183769226, "step": 223, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3821, "val/ratio": 1.0000337362289429, "val/ratio_var": 8.970671387942275e-07 }, { "episode": 115200, "epoch": 0.49348869088416725, "eps": 6, "loss/policy_avg": 0.07157529890537262, "lr": 3.75e-07, "objective/entropy": -44.725563049316406, "objective/kl": 33.00482940673828, "objective/non_score_reward": -1.6502416133880615, "objective/rlhf_reward": 12.270831108093262, "objective/scores": 13.921072006225586, "policy/approxkl_avg": 0.04013068601489067, "policy/clipfrac_avg": 0.1328125, "policy/entropy_avg": 0.16462165117263794, "step": 224, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3643, "val/ratio": 0.9998900890350342, "val/ratio_var": 5.518855346053897e-07 }, { "episode": 115712, "epoch": 0.49568197395476354, "eps": 6, "loss/policy_avg": 0.05166097357869148, "lr": 3.6328125e-07, "objective/entropy": -42.39487838745117, "objective/kl": 32.938636779785156, "objective/non_score_reward": -1.6469318866729736, "objective/rlhf_reward": 12.244211196899414, "objective/scores": 13.891142845153809, "policy/approxkl_avg": 0.03693312406539917, "policy/clipfrac_avg": 0.138671875, "policy/entropy_avg": 0.16154921054840088, "step": 225, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3702, "val/ratio": 0.9998941421508789, "val/ratio_var": 5.255814699012262e-07 }, { "episode": 116224, "epoch": 0.49787525702535984, "eps": 6, "loss/policy_avg": 0.02668042480945587, "lr": 3.515625e-07, "objective/entropy": -45.65013885498047, "objective/kl": 32.436126708984375, "objective/non_score_reward": -1.6218063831329346, "objective/rlhf_reward": 12.132859230041504, "objective/scores": 13.75466537475586, "policy/approxkl_avg": 0.045733705163002014, "policy/clipfrac_avg": 0.177734375, "policy/entropy_avg": 0.1665945202112198, "step": 226, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3756, "val/ratio": 1.0000441074371338, "val/ratio_var": 5.238281914898835e-07 }, { "episode": 116736, "epoch": 0.5000685400959561, "eps": 6, "loss/policy_avg": 0.023709766566753387, "lr": 3.3984375e-07, "objective/entropy": -43.99005889892578, "objective/kl": 33.31614685058594, "objective/non_score_reward": -1.6658073663711548, "objective/rlhf_reward": 12.003641128540039, "objective/scores": 13.669448852539062, "policy/approxkl_avg": 0.0620732381939888, "policy/clipfrac_avg": 0.1875, "policy/entropy_avg": 0.16674618422985077, "step": 227, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3670, "val/ratio": 0.9998377561569214, "val/ratio_var": 7.446521976817166e-07 }, { "episode": 117248, "epoch": 0.5022618231665524, "eps": 6, "loss/policy_avg": 0.04265175759792328, "lr": 3.2812500000000003e-07, "objective/entropy": -42.92007827758789, "objective/kl": 32.18259048461914, "objective/non_score_reward": -1.609129548072815, "objective/rlhf_reward": 12.223358154296875, "objective/scores": 13.832489013671875, "policy/approxkl_avg": 0.04245311766862869, "policy/clipfrac_avg": 0.15234375, "policy/entropy_avg": 0.16078408062458038, "step": 228, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3774, "val/ratio": 0.9999431371688843, "val/ratio_var": 3.725850206137693e-07 }, { "episode": 117760, "epoch": 0.5044551062371487, "eps": 6, "loss/policy_avg": 0.027984416112303734, "lr": 3.1640625e-07, "objective/entropy": -43.285343170166016, "objective/kl": 33.77119064331055, "objective/non_score_reward": -1.6885595321655273, "objective/rlhf_reward": 12.00825023651123, "objective/scores": 13.696809768676758, "policy/approxkl_avg": 0.037605203688144684, "policy/clipfrac_avg": 0.173828125, "policy/entropy_avg": 0.17539986968040466, "step": 229, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3831, "val/ratio": 1.0000276565551758, "val/ratio_var": 4.831003934668843e-07 }, { "episode": 118272, "epoch": 0.506648389307745, "eps": 6, "loss/policy_avg": 0.04504100978374481, "lr": 3.046875e-07, "objective/entropy": -45.07879638671875, "objective/kl": 32.74156188964844, "objective/non_score_reward": -1.637078046798706, "objective/rlhf_reward": 12.17453384399414, "objective/scores": 13.81161117553711, "policy/approxkl_avg": 0.04405646026134491, "policy/clipfrac_avg": 0.14453125, "policy/entropy_avg": 0.16691774129867554, "step": 230, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3676, "val/ratio": 0.999915361404419, "val/ratio_var": 6.945912787159614e-07 }, { "episode": 118784, "epoch": 0.5088416723783413, "eps": 6, "loss/policy_avg": 0.02043619565665722, "lr": 2.9296875000000003e-07, "objective/entropy": -45.20439529418945, "objective/kl": 32.45090103149414, "objective/non_score_reward": -1.6225450038909912, "objective/rlhf_reward": 12.122526168823242, "objective/scores": 13.745071411132812, "policy/approxkl_avg": 0.04984290897846222, "policy/clipfrac_avg": 0.1796875, "policy/entropy_avg": 0.16331960260868073, "step": 231, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3885, "val/ratio": 0.9999264478683472, "val/ratio_var": 8.994372819870478e-07 }, { "episode": 119296, "epoch": 0.5110349554489376, "eps": 6, "loss/policy_avg": 0.011941246688365936, "lr": 2.8125e-07, "objective/entropy": -44.99106979370117, "objective/kl": 33.3937873840332, "objective/non_score_reward": -1.669689416885376, "objective/rlhf_reward": 12.19953441619873, "objective/scores": 13.869223594665527, "policy/approxkl_avg": 0.05245044827461243, "policy/clipfrac_avg": 0.17578125, "policy/entropy_avg": 0.16594412922859192, "step": 232, "val/clipfrac_avg": 0.0, "val/num_eos_tokens": 3806, "val/ratio": 0.999961256980896, "val/ratio_var": 7.90271940331877e-07 } ], "logging_steps": 100, "max_steps": 128, "num_input_tokens_seen": 0, "num_train_epochs": 1.122960932145305, "save_steps": 26, "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, "train_batch_size": null, "trial_name": null, "trial_params": null }