gemma-7b-borpo-low-quality-v3 / trainer_state.json
c-alfano's picture
Model save
4a5e5f6 verified
{
"best_metric": null,
"best_model_checkpoint": null,
"epoch": 3.0,
"eval_steps": 500,
"global_step": 504,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 0.02976190476190476,
"grad_norm": 1643.52392578125,
"learning_rate": 5.000000000000001e-07,
"log_odds_chosen": -0.34639421105384827,
"log_odds_ratio": -1.0579421520233154,
"logits/chosen": 125.67509460449219,
"logits/rejected": 180.79092407226562,
"logps/chosen": -15.30119514465332,
"logps/rejected": -14.954809188842773,
"loss": 15.2275,
"nll_loss": 14.854708671569824,
"rewards/accuracies": 0.25,
"rewards/chosen": -7.65059757232666,
"rewards/margins": -0.1731930673122406,
"rewards/rejected": -7.477404594421387,
"step": 5
},
{
"epoch": 0.05952380952380952,
"grad_norm": 859.1522827148438,
"learning_rate": 1.0000000000000002e-06,
"log_odds_chosen": -0.04612647369503975,
"log_odds_ratio": -0.8707455396652222,
"logits/chosen": 226.11355590820312,
"logits/rejected": 262.4227600097656,
"logps/chosen": -10.240728378295898,
"logps/rejected": -10.194613456726074,
"loss": 10.6297,
"nll_loss": 10.136636734008789,
"rewards/accuracies": 0.4000000059604645,
"rewards/chosen": -5.120364189147949,
"rewards/margins": -0.023057078942656517,
"rewards/rejected": -5.097306728363037,
"step": 10
},
{
"epoch": 0.08928571428571429,
"grad_norm": 192.85568237304688,
"learning_rate": 1.5e-06,
"log_odds_chosen": -0.1222303956747055,
"log_odds_ratio": -0.8104267120361328,
"logits/chosen": 293.97882080078125,
"logits/rejected": 282.1849670410156,
"logps/chosen": -5.908555507659912,
"logps/rejected": -5.787174224853516,
"loss": 6.2671,
"nll_loss": 5.8525919914245605,
"rewards/accuracies": 0.4000000059604645,
"rewards/chosen": -2.954277753829956,
"rewards/margins": -0.06069115549325943,
"rewards/rejected": -2.893587112426758,
"step": 15
},
{
"epoch": 0.11904761904761904,
"grad_norm": 181.04627990722656,
"learning_rate": 2.0000000000000003e-06,
"log_odds_chosen": 0.1651725322008133,
"log_odds_ratio": -0.8192933797836304,
"logits/chosen": 280.179931640625,
"logits/rejected": 263.22296142578125,
"logps/chosen": -3.191753625869751,
"logps/rejected": -3.352200984954834,
"loss": 3.8066,
"nll_loss": 3.3215103149414062,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -1.5958768129348755,
"rewards/margins": 0.08022388815879822,
"rewards/rejected": -1.676100492477417,
"step": 20
},
{
"epoch": 0.1488095238095238,
"grad_norm": 124.88711547851562,
"learning_rate": 2.5e-06,
"log_odds_chosen": 0.21610824763774872,
"log_odds_ratio": -0.6467095613479614,
"logits/chosen": 337.0777587890625,
"logits/rejected": 361.3624267578125,
"logps/chosen": -2.110491991043091,
"logps/rejected": -2.2703230381011963,
"loss": 2.9167,
"nll_loss": 2.676313638687134,
"rewards/accuracies": 0.550000011920929,
"rewards/chosen": -1.0552459955215454,
"rewards/margins": 0.07991557568311691,
"rewards/rejected": -1.1351615190505981,
"step": 25
},
{
"epoch": 0.17857142857142858,
"grad_norm": 120.593994140625,
"learning_rate": 3e-06,
"log_odds_chosen": 0.2517230808734894,
"log_odds_ratio": -0.639795184135437,
"logits/chosen": 322.35894775390625,
"logits/rejected": 409.290771484375,
"logps/chosen": -2.053133964538574,
"logps/rejected": -2.262042999267578,
"loss": 2.4846,
"nll_loss": 2.1683788299560547,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -1.026566982269287,
"rewards/margins": 0.10445437580347061,
"rewards/rejected": -1.131021499633789,
"step": 30
},
{
"epoch": 0.20833333333333334,
"grad_norm": 45.12126541137695,
"learning_rate": 3.5e-06,
"log_odds_chosen": 0.5470780730247498,
"log_odds_ratio": -0.5346588492393494,
"logits/chosen": 374.33734130859375,
"logits/rejected": 398.3028259277344,
"logps/chosen": -1.4850542545318604,
"logps/rejected": -1.9500999450683594,
"loss": 2.3067,
"nll_loss": 1.7573463916778564,
"rewards/accuracies": 0.699999988079071,
"rewards/chosen": -0.7425271272659302,
"rewards/margins": 0.23252280056476593,
"rewards/rejected": -0.9750499725341797,
"step": 35
},
{
"epoch": 0.23809523809523808,
"grad_norm": 34.55175018310547,
"learning_rate": 4.000000000000001e-06,
"log_odds_chosen": 0.48626986145973206,
"log_odds_ratio": -0.5875475406646729,
"logits/chosen": 343.8048400878906,
"logits/rejected": 384.3130187988281,
"logps/chosen": -1.6461362838745117,
"logps/rejected": -2.0677056312561035,
"loss": 2.2439,
"nll_loss": 1.988368272781372,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -0.8230681419372559,
"rewards/margins": 0.2107846736907959,
"rewards/rejected": -1.0338528156280518,
"step": 40
},
{
"epoch": 0.26785714285714285,
"grad_norm": 45.629512786865234,
"learning_rate": 4.5e-06,
"log_odds_chosen": -0.03384453058242798,
"log_odds_ratio": -0.7663249969482422,
"logits/chosen": 391.72265625,
"logits/rejected": 376.98858642578125,
"logps/chosen": -1.8141120672225952,
"logps/rejected": -1.77732253074646,
"loss": 2.1803,
"nll_loss": 1.894079566001892,
"rewards/accuracies": 0.44999998807907104,
"rewards/chosen": -0.9070560336112976,
"rewards/margins": -0.018394792452454567,
"rewards/rejected": -0.88866126537323,
"step": 45
},
{
"epoch": 0.2976190476190476,
"grad_norm": 47.596744537353516,
"learning_rate": 5e-06,
"log_odds_chosen": 0.6274509429931641,
"log_odds_ratio": -0.5308854579925537,
"logits/chosen": 436.2086486816406,
"logits/rejected": 442.98974609375,
"logps/chosen": -1.5697168111801147,
"logps/rejected": -2.105163097381592,
"loss": 2.1885,
"nll_loss": 2.0384533405303955,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.7848584055900574,
"rewards/margins": 0.26772308349609375,
"rewards/rejected": -1.052581548690796,
"step": 50
},
{
"epoch": 0.3273809523809524,
"grad_norm": 35.20968246459961,
"learning_rate": 5.500000000000001e-06,
"log_odds_chosen": 0.3415950834751129,
"log_odds_ratio": -0.6474028825759888,
"logits/chosen": 370.7908020019531,
"logits/rejected": 369.24395751953125,
"logps/chosen": -1.539645791053772,
"logps/rejected": -1.8215446472167969,
"loss": 2.2186,
"nll_loss": 1.9263942241668701,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -0.769822895526886,
"rewards/margins": 0.14094945788383484,
"rewards/rejected": -0.9107723236083984,
"step": 55
},
{
"epoch": 0.35714285714285715,
"grad_norm": 56.7140998840332,
"learning_rate": 6e-06,
"log_odds_chosen": 0.23950794339179993,
"log_odds_ratio": -0.634242057800293,
"logits/chosen": 393.73089599609375,
"logits/rejected": 364.10833740234375,
"logps/chosen": -1.4558594226837158,
"logps/rejected": -1.6270105838775635,
"loss": 2.0389,
"nll_loss": 2.0331270694732666,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.7279297113418579,
"rewards/margins": 0.08557556569576263,
"rewards/rejected": -0.8135052919387817,
"step": 60
},
{
"epoch": 0.3869047619047619,
"grad_norm": 35.615726470947266,
"learning_rate": 6.5000000000000004e-06,
"log_odds_chosen": 0.3183743357658386,
"log_odds_ratio": -0.6254990696907043,
"logits/chosen": 390.70111083984375,
"logits/rejected": 394.5395812988281,
"logps/chosen": -1.4384119510650635,
"logps/rejected": -1.6543052196502686,
"loss": 1.9284,
"nll_loss": 1.6583305597305298,
"rewards/accuracies": 0.699999988079071,
"rewards/chosen": -0.7192059755325317,
"rewards/margins": 0.10794667154550552,
"rewards/rejected": -0.8271526098251343,
"step": 65
},
{
"epoch": 0.4166666666666667,
"grad_norm": 39.04438018798828,
"learning_rate": 7e-06,
"log_odds_chosen": 0.40005454421043396,
"log_odds_ratio": -0.5705705881118774,
"logits/chosen": 397.2193603515625,
"logits/rejected": 392.36126708984375,
"logps/chosen": -1.761604905128479,
"logps/rejected": -2.098273515701294,
"loss": 2.0169,
"nll_loss": 1.894805908203125,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.8808024525642395,
"rewards/margins": 0.16833437979221344,
"rewards/rejected": -1.049136757850647,
"step": 70
},
{
"epoch": 0.44642857142857145,
"grad_norm": 81.70245361328125,
"learning_rate": 7.500000000000001e-06,
"log_odds_chosen": 0.6616519689559937,
"log_odds_ratio": -0.5056720972061157,
"logits/chosen": 399.26202392578125,
"logits/rejected": 387.3080139160156,
"logps/chosen": -1.2724144458770752,
"logps/rejected": -1.7848689556121826,
"loss": 1.9258,
"nll_loss": 1.534188985824585,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.6362072229385376,
"rewards/margins": 0.2562272548675537,
"rewards/rejected": -0.8924344778060913,
"step": 75
},
{
"epoch": 0.47619047619047616,
"grad_norm": 50.36211395263672,
"learning_rate": 8.000000000000001e-06,
"log_odds_chosen": 0.6271190643310547,
"log_odds_ratio": -0.5201026797294617,
"logits/chosen": 406.4871520996094,
"logits/rejected": 419.87884521484375,
"logps/chosen": -1.585078239440918,
"logps/rejected": -2.038538694381714,
"loss": 2.0136,
"nll_loss": 1.7463384866714478,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.792539119720459,
"rewards/margins": 0.22673015296459198,
"rewards/rejected": -1.019269347190857,
"step": 80
},
{
"epoch": 0.5059523809523809,
"grad_norm": 63.27934265136719,
"learning_rate": 8.5e-06,
"log_odds_chosen": -0.1369583010673523,
"log_odds_ratio": -0.9432764053344727,
"logits/chosen": 416.71893310546875,
"logits/rejected": 396.90789794921875,
"logps/chosen": -2.0069127082824707,
"logps/rejected": -1.7957664728164673,
"loss": 2.0904,
"nll_loss": 2.148869514465332,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -1.0034563541412354,
"rewards/margins": -0.10557299852371216,
"rewards/rejected": -0.8978832364082336,
"step": 85
},
{
"epoch": 0.5357142857142857,
"grad_norm": 43.4862060546875,
"learning_rate": 9e-06,
"log_odds_chosen": 1.0741676092147827,
"log_odds_ratio": -0.4483945965766907,
"logits/chosen": 430.83160400390625,
"logits/rejected": 437.00006103515625,
"logps/chosen": -1.3649173974990845,
"logps/rejected": -2.2547924518585205,
"loss": 1.8406,
"nll_loss": 1.6345653533935547,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.6824586987495422,
"rewards/margins": 0.444937527179718,
"rewards/rejected": -1.1273962259292603,
"step": 90
},
{
"epoch": 0.5654761904761905,
"grad_norm": 158.6744384765625,
"learning_rate": 9.5e-06,
"log_odds_chosen": 0.3580256700515747,
"log_odds_ratio": -0.5983911752700806,
"logits/chosen": 365.0293273925781,
"logits/rejected": 395.13739013671875,
"logps/chosen": -1.2539499998092651,
"logps/rejected": -1.4669833183288574,
"loss": 1.9122,
"nll_loss": 1.5921684503555298,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -0.6269749999046326,
"rewards/margins": 0.10651664435863495,
"rewards/rejected": -0.7334916591644287,
"step": 95
},
{
"epoch": 0.5952380952380952,
"grad_norm": 60.67824172973633,
"learning_rate": 1e-05,
"log_odds_chosen": 0.5569584965705872,
"log_odds_ratio": -0.5973536372184753,
"logits/chosen": 427.83270263671875,
"logits/rejected": 444.78289794921875,
"logps/chosen": -1.552487850189209,
"logps/rejected": -2.052288770675659,
"loss": 1.9997,
"nll_loss": 1.694284439086914,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.7762439250946045,
"rewards/margins": 0.2499004304409027,
"rewards/rejected": -1.0261443853378296,
"step": 100
},
{
"epoch": 0.625,
"grad_norm": 66.58687591552734,
"learning_rate": 9.759000729485331e-06,
"log_odds_chosen": 0.6742503643035889,
"log_odds_ratio": -0.5492368340492249,
"logits/chosen": 421.02435302734375,
"logits/rejected": 404.02593994140625,
"logps/chosen": -1.4385154247283936,
"logps/rejected": -1.9352591037750244,
"loss": 1.9337,
"nll_loss": 1.8405609130859375,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.7192577123641968,
"rewards/margins": 0.24837179481983185,
"rewards/rejected": -0.9676295518875122,
"step": 105
},
{
"epoch": 0.6547619047619048,
"grad_norm": 378.0513916015625,
"learning_rate": 9.534625892455923e-06,
"log_odds_chosen": 0.7585327625274658,
"log_odds_ratio": -0.467681884765625,
"logits/chosen": 386.55181884765625,
"logits/rejected": 365.63861083984375,
"logps/chosen": -1.359151840209961,
"logps/rejected": -1.9564090967178345,
"loss": 2.0541,
"nll_loss": 1.9982373714447021,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.6795759201049805,
"rewards/margins": 0.29862865805625916,
"rewards/rejected": -0.9782045483589172,
"step": 110
},
{
"epoch": 0.6845238095238095,
"grad_norm": 68.57817077636719,
"learning_rate": 9.325048082403139e-06,
"log_odds_chosen": 0.9356454610824585,
"log_odds_ratio": -0.41365212202072144,
"logits/chosen": 406.5351867675781,
"logits/rejected": 452.47161865234375,
"logps/chosen": -1.2247552871704102,
"logps/rejected": -1.8875564336776733,
"loss": 2.0096,
"nll_loss": 1.6129186153411865,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.6123776435852051,
"rewards/margins": 0.33140069246292114,
"rewards/rejected": -0.9437782168388367,
"step": 115
},
{
"epoch": 0.7142857142857143,
"grad_norm": 70.14920806884766,
"learning_rate": 9.12870929175277e-06,
"log_odds_chosen": 0.6218617558479309,
"log_odds_ratio": -0.5581934452056885,
"logits/chosen": 400.44232177734375,
"logits/rejected": 403.1569519042969,
"logps/chosen": -1.2707955837249756,
"logps/rejected": -1.6316139698028564,
"loss": 2.0261,
"nll_loss": 1.7148948907852173,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -0.6353977918624878,
"rewards/margins": 0.18040914833545685,
"rewards/rejected": -0.8158069849014282,
"step": 120
},
{
"epoch": 0.7440476190476191,
"grad_norm": 28.353559494018555,
"learning_rate": 8.94427190999916e-06,
"log_odds_chosen": 0.06747283786535263,
"log_odds_ratio": -0.7747208476066589,
"logits/chosen": 394.8547668457031,
"logits/rejected": 402.99114990234375,
"logps/chosen": -1.4064565896987915,
"logps/rejected": -1.443865180015564,
"loss": 1.9533,
"nll_loss": 1.529036283493042,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.7032282948493958,
"rewards/margins": 0.018704283982515335,
"rewards/rejected": -0.721932590007782,
"step": 125
},
{
"epoch": 0.7738095238095238,
"grad_norm": 50.000038146972656,
"learning_rate": 8.770580193070294e-06,
"log_odds_chosen": 0.487846702337265,
"log_odds_ratio": -0.62076336145401,
"logits/chosen": 419.7767639160156,
"logits/rejected": 387.2704772949219,
"logps/chosen": -1.2864845991134644,
"logps/rejected": -1.5687696933746338,
"loss": 1.9848,
"nll_loss": 1.5705018043518066,
"rewards/accuracies": 0.5,
"rewards/chosen": -0.6432422995567322,
"rewards/margins": 0.14114244282245636,
"rewards/rejected": -0.7843848466873169,
"step": 130
},
{
"epoch": 0.8035714285714286,
"grad_norm": 32.915321350097656,
"learning_rate": 8.606629658238705e-06,
"log_odds_chosen": 0.5996901988983154,
"log_odds_ratio": -0.5842069387435913,
"logits/chosen": 392.77679443359375,
"logits/rejected": 423.60272216796875,
"logps/chosen": -1.3194632530212402,
"logps/rejected": -1.8708839416503906,
"loss": 1.9914,
"nll_loss": 1.6256046295166016,
"rewards/accuracies": 0.699999988079071,
"rewards/chosen": -0.6597316265106201,
"rewards/margins": 0.2757102847099304,
"rewards/rejected": -0.9354419708251953,
"step": 135
},
{
"epoch": 0.8333333333333334,
"grad_norm": 60.51993179321289,
"learning_rate": 8.451542547285167e-06,
"log_odds_chosen": 0.14726313948631287,
"log_odds_ratio": -0.6693505048751831,
"logits/chosen": 411.6986389160156,
"logits/rejected": 375.326416015625,
"logps/chosen": -1.321712851524353,
"logps/rejected": -1.4512436389923096,
"loss": 1.9797,
"nll_loss": 1.6686627864837646,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -0.6608564257621765,
"rewards/margins": 0.0647653192281723,
"rewards/rejected": -0.7256218194961548,
"step": 140
},
{
"epoch": 0.8630952380952381,
"grad_norm": 31.022499084472656,
"learning_rate": 8.304547985373997e-06,
"log_odds_chosen": 0.1282106339931488,
"log_odds_ratio": -0.6646836996078491,
"logits/chosen": 381.526611328125,
"logits/rejected": 395.548095703125,
"logps/chosen": -1.3501091003417969,
"logps/rejected": -1.4500309228897095,
"loss": 1.9768,
"nll_loss": 1.590570330619812,
"rewards/accuracies": 0.699999988079071,
"rewards/chosen": -0.6750545501708984,
"rewards/margins": 0.0499609112739563,
"rewards/rejected": -0.7250154614448547,
"step": 145
},
{
"epoch": 0.8928571428571429,
"grad_norm": 48.54278564453125,
"learning_rate": 8.164965809277262e-06,
"log_odds_chosen": 1.1463629007339478,
"log_odds_ratio": -0.46680259704589844,
"logits/chosen": 403.89031982421875,
"logits/rejected": 409.86151123046875,
"logps/chosen": -1.3419028520584106,
"logps/rejected": -2.324528217315674,
"loss": 1.901,
"nll_loss": 1.6430673599243164,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.6709514260292053,
"rewards/margins": 0.49131274223327637,
"rewards/rejected": -1.162264108657837,
"step": 150
},
{
"epoch": 0.9226190476190477,
"grad_norm": 44.35458755493164,
"learning_rate": 8.03219328902499e-06,
"log_odds_chosen": 0.22458314895629883,
"log_odds_ratio": -0.6567160487174988,
"logits/chosen": 399.331298828125,
"logits/rejected": 395.67767333984375,
"logps/chosen": -1.445773959159851,
"logps/rejected": -1.61488938331604,
"loss": 1.8732,
"nll_loss": 1.569331407546997,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -0.7228869795799255,
"rewards/margins": 0.08455771207809448,
"rewards/rejected": -0.80744469165802,
"step": 155
},
{
"epoch": 0.9523809523809523,
"grad_norm": 44.52214431762695,
"learning_rate": 7.905694150420949e-06,
"log_odds_chosen": 0.3395001292228699,
"log_odds_ratio": -0.6508662104606628,
"logits/chosen": 404.9471435546875,
"logits/rejected": 443.93865966796875,
"logps/chosen": -1.2228091955184937,
"logps/rejected": -1.4965049028396606,
"loss": 1.8769,
"nll_loss": 1.4934804439544678,
"rewards/accuracies": 0.44999998807907104,
"rewards/chosen": -0.6114045977592468,
"rewards/margins": 0.13684777915477753,
"rewards/rejected": -0.7482524514198303,
"step": 160
},
{
"epoch": 0.9821428571428571,
"grad_norm": 51.537784576416016,
"learning_rate": 7.78498944161523e-06,
"log_odds_chosen": 0.2777930200099945,
"log_odds_ratio": -0.6248766779899597,
"logits/chosen": 440.94366455078125,
"logits/rejected": 444.90240478515625,
"logps/chosen": -1.3909051418304443,
"logps/rejected": -1.5702444314956665,
"loss": 1.9721,
"nll_loss": 1.7042747735977173,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -0.6954525709152222,
"rewards/margins": 0.08966972678899765,
"rewards/rejected": -0.7851222157478333,
"step": 165
},
{
"epoch": 1.0,
"eval_log_odds_chosen": 0.26492369174957275,
"eval_log_odds_ratio": -0.6573201417922974,
"eval_logits/chosen": 336.2867126464844,
"eval_logits/rejected": 282.1214904785156,
"eval_logps/chosen": -1.2143747806549072,
"eval_logps/rejected": -1.4054285287857056,
"eval_loss": 1.9526195526123047,
"eval_nll_loss": 1.6514703035354614,
"eval_rewards/accuracies": 0.5571428537368774,
"eval_rewards/chosen": -0.6071873903274536,
"eval_rewards/margins": 0.09552692621946335,
"eval_rewards/rejected": -0.7027142643928528,
"eval_runtime": 201.1918,
"eval_samples_per_second": 2.749,
"eval_steps_per_second": 0.348,
"step": 168
},
{
"epoch": 1.0119047619047619,
"grad_norm": 31.192705154418945,
"learning_rate": 7.669649888473705e-06,
"log_odds_chosen": 0.9334543347358704,
"log_odds_ratio": -0.48719945549964905,
"logits/chosen": 377.3502502441406,
"logits/rejected": 415.46356201171875,
"logps/chosen": -1.0376964807510376,
"logps/rejected": -1.7006464004516602,
"loss": 1.7058,
"nll_loss": 1.2853295803070068,
"rewards/accuracies": 0.699999988079071,
"rewards/chosen": -0.5188482403755188,
"rewards/margins": 0.3314751386642456,
"rewards/rejected": -0.8503232002258301,
"step": 170
},
{
"epoch": 1.0416666666666667,
"grad_norm": 23.733516693115234,
"learning_rate": 7.559289460184545e-06,
"log_odds_chosen": 1.4257056713104248,
"log_odds_ratio": -0.36321204900741577,
"logits/chosen": 406.3605651855469,
"logits/rejected": 437.1375427246094,
"logps/chosen": -0.8629204034805298,
"logps/rejected": -1.8850151300430298,
"loss": 1.2977,
"nll_loss": 1.387385368347168,
"rewards/accuracies": 0.8500000238418579,
"rewards/chosen": -0.4314602017402649,
"rewards/margins": 0.5110472440719604,
"rewards/rejected": -0.9425075650215149,
"step": 175
},
{
"epoch": 1.0714285714285714,
"grad_norm": 24.075864791870117,
"learning_rate": 7.4535599249993e-06,
"log_odds_chosen": 1.4416855573654175,
"log_odds_ratio": -0.30401644110679626,
"logits/chosen": 357.9011535644531,
"logits/rejected": 352.5140380859375,
"logps/chosen": -0.899684727191925,
"logps/rejected": -1.9276078939437866,
"loss": 1.3288,
"nll_loss": 1.3482264280319214,
"rewards/accuracies": 0.8500000238418579,
"rewards/chosen": -0.4498423635959625,
"rewards/margins": 0.513961672782898,
"rewards/rejected": -0.9638039469718933,
"step": 180
},
{
"epoch": 1.1011904761904763,
"grad_norm": 39.60210418701172,
"learning_rate": 7.352146220938079e-06,
"log_odds_chosen": 2.3201422691345215,
"log_odds_ratio": -0.20258066058158875,
"logits/chosen": 449.66485595703125,
"logits/rejected": 380.39923095703125,
"logps/chosen": -0.8746849894523621,
"logps/rejected": -2.625277042388916,
"loss": 1.2651,
"nll_loss": 1.2770421504974365,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.43734249472618103,
"rewards/margins": 0.8752959370613098,
"rewards/rejected": -1.312638521194458,
"step": 185
},
{
"epoch": 1.130952380952381,
"grad_norm": 19.941560745239258,
"learning_rate": 7.254762501100117e-06,
"log_odds_chosen": 1.2973986864089966,
"log_odds_ratio": -0.36347365379333496,
"logits/chosen": 403.78106689453125,
"logits/rejected": 423.5147399902344,
"logps/chosen": -0.9356532096862793,
"logps/rejected": -1.7905277013778687,
"loss": 1.3584,
"nll_loss": 1.2576459646224976,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.46782660484313965,
"rewards/margins": 0.4274372160434723,
"rewards/rejected": -0.8952638506889343,
"step": 190
},
{
"epoch": 1.1607142857142858,
"grad_norm": 22.545778274536133,
"learning_rate": 7.1611487403943295e-06,
"log_odds_chosen": 1.494046688079834,
"log_odds_ratio": -0.28673815727233887,
"logits/chosen": 383.939697265625,
"logits/rejected": 414.34100341796875,
"logps/chosen": -1.043168067932129,
"logps/rejected": -2.1456902027130127,
"loss": 1.3101,
"nll_loss": 1.2422401905059814,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.5215840339660645,
"rewards/margins": 0.5512610673904419,
"rewards/rejected": -1.0728451013565063,
"step": 195
},
{
"epoch": 1.1904761904761905,
"grad_norm": 29.15802574157715,
"learning_rate": 7.0710678118654756e-06,
"log_odds_chosen": 2.355477809906006,
"log_odds_ratio": -0.23161384463310242,
"logits/chosen": 443.63848876953125,
"logits/rejected": 404.1456298828125,
"logps/chosen": -0.6128617525100708,
"logps/rejected": -2.293651580810547,
"loss": 1.2885,
"nll_loss": 1.0000107288360596,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.3064308762550354,
"rewards/margins": 0.8403949737548828,
"rewards/rejected": -1.1468257904052734,
"step": 200
},
{
"epoch": 1.2202380952380953,
"grad_norm": 25.433277130126953,
"learning_rate": 6.984302957695783e-06,
"log_odds_chosen": 2.206387758255005,
"log_odds_ratio": -0.20685645937919617,
"logits/chosen": 345.31134033203125,
"logits/rejected": 396.1018371582031,
"logps/chosen": -0.8278995752334595,
"logps/rejected": -2.5073094367980957,
"loss": 1.2908,
"nll_loss": 1.2252819538116455,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.41394978761672974,
"rewards/margins": 0.8397049903869629,
"rewards/rejected": -1.2536547183990479,
"step": 205
},
{
"epoch": 1.25,
"grad_norm": 20.637460708618164,
"learning_rate": 6.900655593423542e-06,
"log_odds_chosen": 1.841051697731018,
"log_odds_ratio": -0.25199171900749207,
"logits/chosen": 375.1767272949219,
"logits/rejected": 373.2859191894531,
"logps/chosen": -0.8490890264511108,
"logps/rejected": -2.084282159805298,
"loss": 1.1767,
"nll_loss": 1.2052141427993774,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.4245445132255554,
"rewards/margins": 0.617596447467804,
"rewards/rejected": -1.042141079902649,
"step": 210
},
{
"epoch": 1.2797619047619047,
"grad_norm": 44.84423065185547,
"learning_rate": 6.819943394704736e-06,
"log_odds_chosen": 2.0608296394348145,
"log_odds_ratio": -0.26352304220199585,
"logits/chosen": 417.9048767089844,
"logits/rejected": 423.32275390625,
"logps/chosen": -0.8630622029304504,
"logps/rejected": -2.449791669845581,
"loss": 1.2804,
"nll_loss": 1.2531594038009644,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.4315311014652252,
"rewards/margins": 0.7933648824691772,
"rewards/rejected": -1.2248958349227905,
"step": 215
},
{
"epoch": 1.3095238095238095,
"grad_norm": 27.052371978759766,
"learning_rate": 6.741998624632421e-06,
"log_odds_chosen": 1.9642305374145508,
"log_odds_ratio": -0.2985631823539734,
"logits/chosen": 365.51226806640625,
"logits/rejected": 383.2918395996094,
"logps/chosen": -0.7986623048782349,
"logps/rejected": -2.1426806449890137,
"loss": 1.2905,
"nll_loss": 1.1290483474731445,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.39933115243911743,
"rewards/margins": 0.6720091700553894,
"rewards/rejected": -1.0713403224945068,
"step": 220
},
{
"epoch": 1.3392857142857144,
"grad_norm": 33.49964141845703,
"learning_rate": 6.666666666666667e-06,
"log_odds_chosen": 1.2250487804412842,
"log_odds_ratio": -0.34452176094055176,
"logits/chosen": 399.16046142578125,
"logits/rejected": 376.3241882324219,
"logps/chosen": -1.1176211833953857,
"logps/rejected": -2.0609307289123535,
"loss": 1.2819,
"nll_loss": 1.3445219993591309,
"rewards/accuracies": 0.8500000238418579,
"rewards/chosen": -0.5588105916976929,
"rewards/margins": 0.47165459394454956,
"rewards/rejected": -1.0304653644561768,
"step": 225
},
{
"epoch": 1.369047619047619,
"grad_norm": 20.881301879882812,
"learning_rate": 6.593804733957872e-06,
"log_odds_chosen": 2.1563384532928467,
"log_odds_ratio": -0.24819080531597137,
"logits/chosen": 367.4872131347656,
"logits/rejected": 376.99822998046875,
"logps/chosen": -0.9552156329154968,
"logps/rejected": -2.5535261631011963,
"loss": 1.2569,
"nll_loss": 0.9765374064445496,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.4776078164577484,
"rewards/margins": 0.7991552948951721,
"rewards/rejected": -1.2767630815505981,
"step": 230
},
{
"epoch": 1.3988095238095237,
"grad_norm": 20.453393936157227,
"learning_rate": 6.523280730534423e-06,
"log_odds_chosen": 2.148529529571533,
"log_odds_ratio": -0.2209801971912384,
"logits/chosen": 396.610595703125,
"logits/rejected": 402.3414306640625,
"logps/chosen": -0.6876020431518555,
"logps/rejected": -2.1799914836883545,
"loss": 1.2833,
"nll_loss": 1.0352197885513306,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.34380102157592773,
"rewards/margins": 0.7461946606636047,
"rewards/rejected": -1.0899957418441772,
"step": 235
},
{
"epoch": 1.4285714285714286,
"grad_norm": 23.43726348876953,
"learning_rate": 6.4549722436790284e-06,
"log_odds_chosen": 1.4042203426361084,
"log_odds_ratio": -0.30760836601257324,
"logits/chosen": 426.6971740722656,
"logits/rejected": 434.68292236328125,
"logps/chosen": -0.9825772047042847,
"logps/rejected": -2.0324692726135254,
"loss": 1.258,
"nll_loss": 1.1937551498413086,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.49128860235214233,
"rewards/margins": 0.5249461531639099,
"rewards/rejected": -1.0162346363067627,
"step": 240
},
{
"epoch": 1.4583333333333333,
"grad_norm": 20.512117385864258,
"learning_rate": 6.3887656499994e-06,
"log_odds_chosen": 1.7208242416381836,
"log_odds_ratio": -0.2485727071762085,
"logits/chosen": 406.27337646484375,
"logits/rejected": 381.36297607421875,
"logps/chosen": -0.7839127779006958,
"logps/rejected": -2.025153398513794,
"loss": 1.3187,
"nll_loss": 1.2436693906784058,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.3919563889503479,
"rewards/margins": 0.6206203699111938,
"rewards/rejected": -1.012576699256897,
"step": 245
},
{
"epoch": 1.4880952380952381,
"grad_norm": 31.212268829345703,
"learning_rate": 6.324555320336759e-06,
"log_odds_chosen": 2.171504497528076,
"log_odds_ratio": -0.2626289427280426,
"logits/chosen": 385.3293151855469,
"logits/rejected": 403.4792785644531,
"logps/chosen": -0.912436842918396,
"logps/rejected": -2.581498146057129,
"loss": 1.2409,
"nll_loss": 0.9795435070991516,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.456218421459198,
"rewards/margins": 0.8345306515693665,
"rewards/rejected": -1.2907490730285645,
"step": 250
},
{
"epoch": 1.5178571428571428,
"grad_norm": 15.0794095993042,
"learning_rate": 6.262242910851496e-06,
"log_odds_chosen": 1.983006238937378,
"log_odds_ratio": -0.19495443999767303,
"logits/chosen": 411.68487548828125,
"logits/rejected": 416.602783203125,
"logps/chosen": -0.9126818776130676,
"logps/rejected": -2.4313483238220215,
"loss": 1.2696,
"nll_loss": 1.102388620376587,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.4563409388065338,
"rewards/margins": 0.7593332529067993,
"rewards/rejected": -1.2156741619110107,
"step": 255
},
{
"epoch": 1.5476190476190477,
"grad_norm": 27.49896812438965,
"learning_rate": 6.2017367294604225e-06,
"log_odds_chosen": 1.9481725692749023,
"log_odds_ratio": -0.2753888964653015,
"logits/chosen": 386.07427978515625,
"logits/rejected": 435.198974609375,
"logps/chosen": -0.995970606803894,
"logps/rejected": -2.5512473583221436,
"loss": 1.2264,
"nll_loss": 1.1618086099624634,
"rewards/accuracies": 0.8500000238418579,
"rewards/chosen": -0.497985303401947,
"rewards/margins": 0.7776384353637695,
"rewards/rejected": -1.2756236791610718,
"step": 260
},
{
"epoch": 1.5773809523809523,
"grad_norm": 23.22437286376953,
"learning_rate": 6.142951168339513e-06,
"log_odds_chosen": 1.3944361209869385,
"log_odds_ratio": -0.4270511567592621,
"logits/chosen": 435.14678955078125,
"logits/rejected": 427.51226806640625,
"logps/chosen": -1.104048252105713,
"logps/rejected": -2.160891056060791,
"loss": 1.4157,
"nll_loss": 1.3079249858856201,
"rewards/accuracies": 0.75,
"rewards/chosen": -0.5520241260528564,
"rewards/margins": 0.5284214019775391,
"rewards/rejected": -1.0804455280303955,
"step": 265
},
{
"epoch": 1.6071428571428572,
"grad_norm": 16.965179443359375,
"learning_rate": 6.0858061945018455e-06,
"log_odds_chosen": 1.3110793828964233,
"log_odds_ratio": -0.33642929792404175,
"logits/chosen": 370.2662658691406,
"logits/rejected": 376.83587646484375,
"logps/chosen": -1.1280148029327393,
"logps/rejected": -2.0509979724884033,
"loss": 1.3064,
"nll_loss": 1.3204824924468994,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.5640074014663696,
"rewards/margins": 0.46149152517318726,
"rewards/rejected": -1.0254989862442017,
"step": 270
},
{
"epoch": 1.6369047619047619,
"grad_norm": 16.231523513793945,
"learning_rate": 6.030226891555273e-06,
"log_odds_chosen": 1.2828176021575928,
"log_odds_ratio": -0.31571871042251587,
"logits/chosen": 336.78143310546875,
"logits/rejected": 352.0708312988281,
"logps/chosen": -0.8390641212463379,
"logps/rejected": -1.6753209829330444,
"loss": 1.2993,
"nll_loss": 1.0436512231826782,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.41953206062316895,
"rewards/margins": 0.4181283414363861,
"rewards/rejected": -0.8376604914665222,
"step": 275
},
{
"epoch": 1.6666666666666665,
"grad_norm": 19.623210906982422,
"learning_rate": 5.976143046671968e-06,
"log_odds_chosen": 1.4294779300689697,
"log_odds_ratio": -0.3383074104785919,
"logits/chosen": 406.8611145019531,
"logits/rejected": 369.0583190917969,
"logps/chosen": -0.7519195675849915,
"logps/rejected": -1.7197290658950806,
"loss": 1.3075,
"nll_loss": 1.0520720481872559,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.3759597837924957,
"rewards/margins": 0.4839046597480774,
"rewards/rejected": -0.8598645329475403,
"step": 280
},
{
"epoch": 1.6964285714285714,
"grad_norm": 27.15911293029785,
"learning_rate": 5.923488777590924e-06,
"log_odds_chosen": 1.6469166278839111,
"log_odds_ratio": -0.31998661160469055,
"logits/chosen": 351.8115234375,
"logits/rejected": 378.2767333984375,
"logps/chosen": -0.8496532440185547,
"logps/rejected": -2.034947156906128,
"loss": 1.2238,
"nll_loss": 1.0820589065551758,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.42482662200927734,
"rewards/margins": 0.5926468968391418,
"rewards/rejected": -1.017473578453064,
"step": 285
},
{
"epoch": 1.7261904761904763,
"grad_norm": 20.430383682250977,
"learning_rate": 5.8722021951470355e-06,
"log_odds_chosen": 2.2868614196777344,
"log_odds_ratio": -0.23290471732616425,
"logits/chosen": 376.9328918457031,
"logits/rejected": 444.3408203125,
"logps/chosen": -0.8962064981460571,
"logps/rejected": -2.7214858531951904,
"loss": 1.2798,
"nll_loss": 1.1513203382492065,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.44810324907302856,
"rewards/margins": 0.9126396179199219,
"rewards/rejected": -1.3607429265975952,
"step": 290
},
{
"epoch": 1.755952380952381,
"grad_norm": 23.614900588989258,
"learning_rate": 5.822225097395821e-06,
"log_odds_chosen": 2.100496768951416,
"log_odds_ratio": -0.2871156930923462,
"logits/chosen": 391.25726318359375,
"logits/rejected": 399.76214599609375,
"logps/chosen": -1.0318877696990967,
"logps/rejected": -2.5897607803344727,
"loss": 1.318,
"nll_loss": 1.339220404624939,
"rewards/accuracies": 0.8500000238418579,
"rewards/chosen": -0.5159438848495483,
"rewards/margins": 0.778936505317688,
"rewards/rejected": -1.2948803901672363,
"step": 295
},
{
"epoch": 1.7857142857142856,
"grad_norm": 24.594526290893555,
"learning_rate": 5.773502691896259e-06,
"log_odds_chosen": 1.4617587327957153,
"log_odds_ratio": -0.268736869096756,
"logits/chosen": 392.9894714355469,
"logits/rejected": 392.593994140625,
"logps/chosen": -0.716581404209137,
"logps/rejected": -1.6443252563476562,
"loss": 1.2342,
"nll_loss": 1.0071418285369873,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.3582907021045685,
"rewards/margins": 0.46387186646461487,
"rewards/rejected": -0.8221626281738281,
"step": 300
},
{
"epoch": 1.8154761904761905,
"grad_norm": 18.39592742919922,
"learning_rate": 5.725983343138682e-06,
"log_odds_chosen": 1.3188327550888062,
"log_odds_ratio": -0.39420002698898315,
"logits/chosen": 443.49505615234375,
"logits/rejected": 423.3663635253906,
"logps/chosen": -1.0477676391601562,
"logps/rejected": -2.009697914123535,
"loss": 1.2633,
"nll_loss": 1.2299957275390625,
"rewards/accuracies": 0.8500000238418579,
"rewards/chosen": -0.5238838195800781,
"rewards/margins": 0.4809652864933014,
"rewards/rejected": -1.0048489570617676,
"step": 305
},
{
"epoch": 1.8452380952380953,
"grad_norm": 16.925289154052734,
"learning_rate": 5.679618342470648e-06,
"log_odds_chosen": 1.8564685583114624,
"log_odds_ratio": -0.2928754985332489,
"logits/chosen": 358.899169921875,
"logits/rejected": 364.58160400390625,
"logps/chosen": -0.94548100233078,
"logps/rejected": -2.3730039596557617,
"loss": 1.2961,
"nll_loss": 1.2056444883346558,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.47274050116539,
"rewards/margins": 0.7137616276741028,
"rewards/rejected": -1.1865019798278809,
"step": 310
},
{
"epoch": 1.875,
"grad_norm": 27.0323543548584,
"learning_rate": 5.63436169819011e-06,
"log_odds_chosen": 1.889154076576233,
"log_odds_ratio": -0.3951663076877594,
"logits/chosen": 385.2926025390625,
"logits/rejected": 391.605712890625,
"logps/chosen": -1.166025161743164,
"logps/rejected": -2.7089405059814453,
"loss": 1.3311,
"nll_loss": 1.3940035104751587,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.583012580871582,
"rewards/margins": 0.7714576125144958,
"rewards/rejected": -1.3544702529907227,
"step": 315
},
{
"epoch": 1.9047619047619047,
"grad_norm": 18.699766159057617,
"learning_rate": 5.590169943749475e-06,
"log_odds_chosen": 1.6885192394256592,
"log_odds_ratio": -0.2244713306427002,
"logits/chosen": 402.7138366699219,
"logits/rejected": 435.62646484375,
"logps/chosen": -0.8159521222114563,
"logps/rejected": -2.033271551132202,
"loss": 1.2117,
"nll_loss": 0.9898951649665833,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.40797606110572815,
"rewards/margins": 0.6086598634719849,
"rewards/rejected": -1.016635775566101,
"step": 320
},
{
"epoch": 1.9345238095238095,
"grad_norm": 29.9803409576416,
"learning_rate": 5.547001962252292e-06,
"log_odds_chosen": 2.429278612136841,
"log_odds_ratio": -0.19841960072517395,
"logits/chosen": 411.4713439941406,
"logits/rejected": 433.3556213378906,
"logps/chosen": -0.8200858235359192,
"logps/rejected": -2.58778977394104,
"loss": 1.1848,
"nll_loss": 1.0703952312469482,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.4100429117679596,
"rewards/margins": 0.8838518857955933,
"rewards/rejected": -1.29389488697052,
"step": 325
},
{
"epoch": 1.9642857142857144,
"grad_norm": 15.92293643951416,
"learning_rate": 5.504818825631804e-06,
"log_odds_chosen": 1.846983551979065,
"log_odds_ratio": -0.29716789722442627,
"logits/chosen": 407.4903869628906,
"logits/rejected": 413.0613708496094,
"logps/chosen": -0.9037710428237915,
"logps/rejected": -2.2958996295928955,
"loss": 1.245,
"nll_loss": 0.9782114028930664,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.45188552141189575,
"rewards/margins": 0.6960643529891968,
"rewards/rejected": -1.1479498147964478,
"step": 330
},
{
"epoch": 1.994047619047619,
"grad_norm": 15.895071029663086,
"learning_rate": 5.4635836470815305e-06,
"log_odds_chosen": 1.2502014636993408,
"log_odds_ratio": -0.3804013133049011,
"logits/chosen": 390.8824157714844,
"logits/rejected": 404.8577575683594,
"logps/chosen": -1.2015631198883057,
"logps/rejected": -2.143710136413574,
"loss": 1.3299,
"nll_loss": 1.439396619796753,
"rewards/accuracies": 0.800000011920929,
"rewards/chosen": -0.6007815599441528,
"rewards/margins": 0.47107353806495667,
"rewards/rejected": -1.071855068206787,
"step": 335
},
{
"epoch": 2.0,
"eval_log_odds_chosen": 0.2173391729593277,
"eval_log_odds_ratio": -0.6792100071907043,
"eval_logits/chosen": 345.2332763671875,
"eval_logits/rejected": 293.2820129394531,
"eval_logps/chosen": -1.1971725225448608,
"eval_logps/rejected": -1.3610831499099731,
"eval_loss": 1.9015214443206787,
"eval_nll_loss": 1.5933443307876587,
"eval_rewards/accuracies": 0.5,
"eval_rewards/chosen": -0.5985862612724304,
"eval_rewards/margins": 0.08195527642965317,
"eval_rewards/rejected": -0.6805415749549866,
"eval_runtime": 201.3424,
"eval_samples_per_second": 2.747,
"eval_steps_per_second": 0.348,
"step": 336
},
{
"epoch": 2.0238095238095237,
"grad_norm": 18.064809799194336,
"learning_rate": 5.423261445466404e-06,
"log_odds_chosen": 3.1630072593688965,
"log_odds_ratio": -0.14342114329338074,
"logits/chosen": 407.103759765625,
"logits/rejected": 396.9122009277344,
"logps/chosen": -0.5497515201568604,
"logps/rejected": -2.797853946685791,
"loss": 0.7506,
"nll_loss": 0.7164371609687805,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.2748757600784302,
"rewards/margins": 1.1240513324737549,
"rewards/rejected": -1.3989269733428955,
"step": 340
},
{
"epoch": 2.0535714285714284,
"grad_norm": 12.874629974365234,
"learning_rate": 5.383819020581656e-06,
"log_odds_chosen": 4.300416946411133,
"log_odds_ratio": -0.10298861563205719,
"logits/chosen": 406.04144287109375,
"logits/rejected": 365.400390625,
"logps/chosen": -0.5357804894447327,
"logps/rejected": -3.574709415435791,
"loss": 0.6675,
"nll_loss": 0.8001909255981445,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.26789024472236633,
"rewards/margins": 1.5194646120071411,
"rewards/rejected": -1.7873547077178955,
"step": 345
},
{
"epoch": 2.0833333333333335,
"grad_norm": 13.677016258239746,
"learning_rate": 5.345224838248489e-06,
"log_odds_chosen": 4.409786224365234,
"log_odds_ratio": -0.03061295673251152,
"logits/chosen": 372.30560302734375,
"logits/rejected": 401.3008728027344,
"logps/chosen": -0.46024495363235474,
"logps/rejected": -3.737445116043091,
"loss": 0.5993,
"nll_loss": 0.6028808355331421,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.23012247681617737,
"rewards/margins": 1.6386003494262695,
"rewards/rejected": -1.8687225580215454,
"step": 350
},
{
"epoch": 2.113095238095238,
"grad_norm": 10.955292701721191,
"learning_rate": 5.307448924342753e-06,
"log_odds_chosen": 4.361363410949707,
"log_odds_ratio": -0.04152694344520569,
"logits/chosen": 373.2965393066406,
"logits/rejected": 396.0318603515625,
"logps/chosen": -0.4621841311454773,
"logps/rejected": -3.721182346343994,
"loss": 0.5932,
"nll_loss": 0.5964406132698059,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.23109206557273865,
"rewards/margins": 1.6294991970062256,
"rewards/rejected": -1.860591173171997,
"step": 355
},
{
"epoch": 2.142857142857143,
"grad_norm": 28.92525863647461,
"learning_rate": 5.270462766947299e-06,
"log_odds_chosen": 4.147698402404785,
"log_odds_ratio": -0.032929155975580215,
"logits/chosen": 380.5809631347656,
"logits/rejected": 416.3023986816406,
"logps/chosen": -0.381146639585495,
"logps/rejected": -3.1521759033203125,
"loss": 0.612,
"nll_loss": 0.6254645586013794,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.1905733197927475,
"rewards/margins": 1.3855146169662476,
"rewards/rejected": -1.5760879516601562,
"step": 360
},
{
"epoch": 2.1726190476190474,
"grad_norm": 12.443119049072266,
"learning_rate": 5.234239225902137e-06,
"log_odds_chosen": 4.118858337402344,
"log_odds_ratio": -0.03749427944421768,
"logits/chosen": 345.683349609375,
"logits/rejected": 312.6221923828125,
"logps/chosen": -0.32834386825561523,
"logps/rejected": -3.117326498031616,
"loss": 0.5807,
"nll_loss": 0.531657338142395,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.16417193412780762,
"rewards/margins": 1.3944913148880005,
"rewards/rejected": -1.558663249015808,
"step": 365
},
{
"epoch": 2.2023809523809526,
"grad_norm": 10.90052318572998,
"learning_rate": 5.198752449100364e-06,
"log_odds_chosen": 3.4755806922912598,
"log_odds_ratio": -0.04619182273745537,
"logits/chosen": 332.7835693359375,
"logits/rejected": 340.1016845703125,
"logps/chosen": -0.4380973279476166,
"logps/rejected": -2.8499865531921387,
"loss": 0.6158,
"nll_loss": 0.6762995719909668,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.2190486639738083,
"rewards/margins": 1.2059446573257446,
"rewards/rejected": -1.4249932765960693,
"step": 370
},
{
"epoch": 2.232142857142857,
"grad_norm": 14.661003112792969,
"learning_rate": 5.163977794943223e-06,
"log_odds_chosen": 3.5086026191711426,
"log_odds_ratio": -0.11584819853305817,
"logits/chosen": 396.9617614746094,
"logits/rejected": 412.2447814941406,
"logps/chosen": -0.4914408326148987,
"logps/rejected": -2.892894744873047,
"loss": 0.6302,
"nll_loss": 0.5993391275405884,
"rewards/accuracies": 0.8999999761581421,
"rewards/chosen": -0.24572041630744934,
"rewards/margins": 1.2007267475128174,
"rewards/rejected": -1.4464473724365234,
"step": 375
},
{
"epoch": 2.261904761904762,
"grad_norm": 13.687707901000977,
"learning_rate": 5.129891760425772e-06,
"log_odds_chosen": 3.4452381134033203,
"log_odds_ratio": -0.04614276438951492,
"logits/chosen": 386.617919921875,
"logits/rejected": 348.6525573730469,
"logps/chosen": -0.43144264817237854,
"logps/rejected": -2.679448366165161,
"loss": 0.5916,
"nll_loss": 0.61055588722229,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.21572132408618927,
"rewards/margins": 1.1240026950836182,
"rewards/rejected": -1.3397241830825806,
"step": 380
},
{
"epoch": 2.2916666666666665,
"grad_norm": 12.487543106079102,
"learning_rate": 5.096471914376255e-06,
"log_odds_chosen": 4.143012046813965,
"log_odds_ratio": -0.055028241127729416,
"logits/chosen": 369.7757873535156,
"logits/rejected": 385.3197326660156,
"logps/chosen": -0.42009004950523376,
"logps/rejected": -3.1088523864746094,
"loss": 0.5661,
"nll_loss": 0.5417618155479431,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.21004502475261688,
"rewards/margins": 1.3443810939788818,
"rewards/rejected": -1.5544261932373047,
"step": 385
},
{
"epoch": 2.3214285714285716,
"grad_norm": 14.711711883544922,
"learning_rate": 5.0636968354183334e-06,
"log_odds_chosen": 3.86775541305542,
"log_odds_ratio": -0.051612865179777145,
"logits/chosen": 386.55340576171875,
"logits/rejected": 420.259033203125,
"logps/chosen": -0.40563899278640747,
"logps/rejected": -3.13773775100708,
"loss": 0.6336,
"nll_loss": 0.5235159993171692,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.20281949639320374,
"rewards/margins": 1.3660494089126587,
"rewards/rejected": -1.56886887550354,
"step": 390
},
{
"epoch": 2.3511904761904763,
"grad_norm": 14.341813087463379,
"learning_rate": 5.031546054266276e-06,
"log_odds_chosen": 4.3614888191223145,
"log_odds_ratio": -0.02659059502184391,
"logits/chosen": 381.09478759765625,
"logits/rejected": 371.54730224609375,
"logps/chosen": -0.3334965705871582,
"logps/rejected": -3.2248425483703613,
"loss": 0.6006,
"nll_loss": 0.5441581010818481,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.1667482852935791,
"rewards/margins": 1.4456731081008911,
"rewards/rejected": -1.6124212741851807,
"step": 395
},
{
"epoch": 2.380952380952381,
"grad_norm": 10.141727447509766,
"learning_rate": 5e-06,
"log_odds_chosen": 5.030593395233154,
"log_odds_ratio": -0.060914844274520874,
"logits/chosen": 360.45758056640625,
"logits/rejected": 407.6477966308594,
"logps/chosen": -0.41616684198379517,
"logps/rejected": -4.148199558258057,
"loss": 0.5815,
"nll_loss": 0.6026356220245361,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.20808342099189758,
"rewards/margins": 1.8660163879394531,
"rewards/rejected": -2.0740997791290283,
"step": 400
},
{
"epoch": 2.4107142857142856,
"grad_norm": 12.312828063964844,
"learning_rate": 4.969039949999534e-06,
"log_odds_chosen": 3.7259132862091064,
"log_odds_ratio": -0.05210161954164505,
"logits/chosen": 441.5244140625,
"logits/rejected": 379.39215087890625,
"logps/chosen": -0.31666994094848633,
"logps/rejected": -2.7463951110839844,
"loss": 0.6506,
"nll_loss": 0.5460222959518433,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.15833497047424316,
"rewards/margins": 1.214862585067749,
"rewards/rejected": -1.3731975555419922,
"step": 405
},
{
"epoch": 2.4404761904761907,
"grad_norm": 10.084498405456543,
"learning_rate": 4.938647983247949e-06,
"log_odds_chosen": 4.13731575012207,
"log_odds_ratio": -0.07053720951080322,
"logits/chosen": 375.71875,
"logits/rejected": 404.51898193359375,
"logps/chosen": -0.4222637116909027,
"logps/rejected": -3.3887131214141846,
"loss": 0.5941,
"nll_loss": 0.5796958804130554,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.21113185584545135,
"rewards/margins": 1.483224630355835,
"rewards/rejected": -1.6943565607070923,
"step": 410
},
{
"epoch": 2.4702380952380953,
"grad_norm": 10.309317588806152,
"learning_rate": 4.9088069367381605e-06,
"log_odds_chosen": 4.0256781578063965,
"log_odds_ratio": -0.02946905419230461,
"logits/chosen": 410.3990783691406,
"logits/rejected": 388.10638427734375,
"logps/chosen": -0.34235674142837524,
"logps/rejected": -2.9506092071533203,
"loss": 0.6024,
"nll_loss": 0.5277774333953857,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.17117837071418762,
"rewards/margins": 1.3041261434555054,
"rewards/rejected": -1.4753046035766602,
"step": 415
},
{
"epoch": 2.5,
"grad_norm": 17.37739372253418,
"learning_rate": 4.8795003647426654e-06,
"log_odds_chosen": 4.524823188781738,
"log_odds_ratio": -0.054420001804828644,
"logits/chosen": 391.82403564453125,
"logits/rejected": 331.8238220214844,
"logps/chosen": -0.296555757522583,
"logps/rejected": -3.383528470993042,
"loss": 0.5781,
"nll_loss": 0.6599777340888977,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.1482778787612915,
"rewards/margins": 1.5434863567352295,
"rewards/rejected": -1.691764235496521,
"step": 420
},
{
"epoch": 2.5297619047619047,
"grad_norm": 14.520256042480469,
"learning_rate": 4.850712500726659e-06,
"log_odds_chosen": 4.5146284103393555,
"log_odds_ratio": -0.020355457440018654,
"logits/chosen": 364.44293212890625,
"logits/rejected": 381.6509094238281,
"logps/chosen": -0.4376618266105652,
"logps/rejected": -3.6549148559570312,
"loss": 0.6064,
"nll_loss": 0.7541596293449402,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.2188309133052826,
"rewards/margins": 1.6086266040802002,
"rewards/rejected": -1.8274574279785156,
"step": 425
},
{
"epoch": 2.5595238095238093,
"grad_norm": 18.008617401123047,
"learning_rate": 4.822428221704122e-06,
"log_odds_chosen": 4.2900166511535645,
"log_odds_ratio": -0.05251041799783707,
"logits/chosen": 418.76812744140625,
"logits/rejected": 422.1298828125,
"logps/chosen": -0.3535473942756653,
"logps/rejected": -3.3458237648010254,
"loss": 0.5961,
"nll_loss": 0.49905306100845337,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.17677369713783264,
"rewards/margins": 1.496138334274292,
"rewards/rejected": -1.6729118824005127,
"step": 430
},
{
"epoch": 2.5892857142857144,
"grad_norm": 14.501811027526855,
"learning_rate": 4.794633014853843e-06,
"log_odds_chosen": 3.7202048301696777,
"log_odds_ratio": -0.08040798455476761,
"logits/chosen": 378.04962158203125,
"logits/rejected": 374.47149658203125,
"logps/chosen": -0.4141341745853424,
"logps/rejected": -3.029280424118042,
"loss": 0.6443,
"nll_loss": 0.5974027514457703,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.2070670872926712,
"rewards/margins": 1.3075730800628662,
"rewards/rejected": -1.514640212059021,
"step": 435
},
{
"epoch": 2.619047619047619,
"grad_norm": 13.21033000946045,
"learning_rate": 4.767312946227961e-06,
"log_odds_chosen": 4.036019325256348,
"log_odds_ratio": -0.0349675677716732,
"logits/chosen": 374.5437927246094,
"logits/rejected": 379.2315368652344,
"logps/chosen": -0.357850581407547,
"logps/rejected": -3.0918896198272705,
"loss": 0.6481,
"nll_loss": 0.5479155778884888,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.1789252907037735,
"rewards/margins": 1.3670194149017334,
"rewards/rejected": -1.5459448099136353,
"step": 440
},
{
"epoch": 2.6488095238095237,
"grad_norm": 10.330572128295898,
"learning_rate": 4.740454631399773e-06,
"log_odds_chosen": 4.903168678283691,
"log_odds_ratio": -0.02322390116751194,
"logits/chosen": 356.780029296875,
"logits/rejected": 385.78631591796875,
"logps/chosen": -0.3364141881465912,
"logps/rejected": -3.9139695167541504,
"loss": 0.5982,
"nll_loss": 0.5599361658096313,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.1682070940732956,
"rewards/margins": 1.7887779474258423,
"rewards/rejected": -1.9569847583770752,
"step": 445
},
{
"epoch": 2.678571428571429,
"grad_norm": 11.032569885253906,
"learning_rate": 4.714045207910318e-06,
"log_odds_chosen": 4.497644901275635,
"log_odds_ratio": -0.030209308490157127,
"logits/chosen": 394.76336669921875,
"logits/rejected": 376.0134582519531,
"logps/chosen": -0.41811317205429077,
"logps/rejected": -3.552691698074341,
"loss": 0.6268,
"nll_loss": 0.5461404919624329,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.20905658602714539,
"rewards/margins": 1.567289113998413,
"rewards/rejected": -1.7763458490371704,
"step": 450
},
{
"epoch": 2.7083333333333335,
"grad_norm": 17.51761245727539,
"learning_rate": 4.688072309384955e-06,
"log_odds_chosen": 4.075136661529541,
"log_odds_ratio": -0.025673285126686096,
"logits/chosen": 384.58154296875,
"logits/rejected": 360.6435546875,
"logps/chosen": -0.29184406995773315,
"logps/rejected": -2.7901484966278076,
"loss": 0.6721,
"nll_loss": 0.5232545733451843,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.14592203497886658,
"rewards/margins": 1.249152421951294,
"rewards/rejected": -1.3950742483139038,
"step": 455
},
{
"epoch": 2.738095238095238,
"grad_norm": 13.356114387512207,
"learning_rate": 4.662524041201569e-06,
"log_odds_chosen": 3.7381584644317627,
"log_odds_ratio": -0.041400760412216187,
"logits/chosen": 424.15606689453125,
"logits/rejected": 436.59417724609375,
"logps/chosen": -0.3177054226398468,
"logps/rejected": -2.7719643115997314,
"loss": 0.6171,
"nll_loss": 0.5559448599815369,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.1588527113199234,
"rewards/margins": 1.2271292209625244,
"rewards/rejected": -1.3859821557998657,
"step": 460
},
{
"epoch": 2.767857142857143,
"grad_norm": 16.436243057250977,
"learning_rate": 4.6373889576016826e-06,
"log_odds_chosen": 3.9720139503479004,
"log_odds_ratio": -0.04150586202740669,
"logits/chosen": 419.3877868652344,
"logits/rejected": 399.4032287597656,
"logps/chosen": -0.3887273669242859,
"logps/rejected": -3.1958889961242676,
"loss": 0.6517,
"nll_loss": 0.7023509740829468,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.19436368346214294,
"rewards/margins": 1.403580665588379,
"rewards/rejected": -1.5979444980621338,
"step": 465
},
{
"epoch": 2.7976190476190474,
"grad_norm": 13.92647647857666,
"learning_rate": 4.6126560401444256e-06,
"log_odds_chosen": 3.6036312580108643,
"log_odds_ratio": -0.0841737613081932,
"logits/chosen": 376.7361145019531,
"logits/rejected": 346.24578857421875,
"logps/chosen": -0.38905754685401917,
"logps/rejected": -2.8610401153564453,
"loss": 0.5864,
"nll_loss": 0.5663691163063049,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.19452877342700958,
"rewards/margins": 1.235991358757019,
"rewards/rejected": -1.4305200576782227,
"step": 470
},
{
"epoch": 2.8273809523809526,
"grad_norm": 22.44719123840332,
"learning_rate": 4.588314677411235e-06,
"log_odds_chosen": 4.282876014709473,
"log_odds_ratio": -0.02305762842297554,
"logits/chosen": 351.80535888671875,
"logits/rejected": 435.5419006347656,
"logps/chosen": -0.376276433467865,
"logps/rejected": -3.4008407592773438,
"loss": 0.6159,
"nll_loss": 0.5037115216255188,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.1881382167339325,
"rewards/margins": 1.512282133102417,
"rewards/rejected": -1.7004203796386719,
"step": 475
},
{
"epoch": 2.857142857142857,
"grad_norm": 11.814839363098145,
"learning_rate": 4.564354645876385e-06,
"log_odds_chosen": 5.438357353210449,
"log_odds_ratio": -0.03407539427280426,
"logits/chosen": 393.2117614746094,
"logits/rejected": 400.9731140136719,
"logps/chosen": -0.38579824566841125,
"logps/rejected": -4.569952964782715,
"loss": 0.5976,
"nll_loss": 0.6172084212303162,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.19289912283420563,
"rewards/margins": 2.0920770168304443,
"rewards/rejected": -2.2849764823913574,
"step": 480
},
{
"epoch": 2.886904761904762,
"grad_norm": 12.605895042419434,
"learning_rate": 4.540766091864998e-06,
"log_odds_chosen": 3.7765631675720215,
"log_odds_ratio": -0.03285397216677666,
"logits/chosen": 376.1605529785156,
"logits/rejected": 400.92620849609375,
"logps/chosen": -0.4142071604728699,
"logps/rejected": -3.060633659362793,
"loss": 0.5919,
"nll_loss": 0.5258246660232544,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.20710358023643494,
"rewards/margins": 1.3232133388519287,
"rewards/rejected": -1.5303168296813965,
"step": 485
},
{
"epoch": 2.9166666666666665,
"grad_norm": 10.655911445617676,
"learning_rate": 4.517539514526257e-06,
"log_odds_chosen": 3.3025474548339844,
"log_odds_ratio": -0.07940138876438141,
"logits/chosen": 416.90509033203125,
"logits/rejected": 319.48687744140625,
"logps/chosen": -0.6181944012641907,
"logps/rejected": -2.8924221992492676,
"loss": 0.6671,
"nll_loss": 0.5718962550163269,
"rewards/accuracies": 0.949999988079071,
"rewards/chosen": -0.30909720063209534,
"rewards/margins": 1.1371139287948608,
"rewards/rejected": -1.4462110996246338,
"step": 490
},
{
"epoch": 2.946428571428571,
"grad_norm": 11.349733352661133,
"learning_rate": 4.4946657497549474e-06,
"log_odds_chosen": 4.556464672088623,
"log_odds_ratio": -0.033807143568992615,
"logits/chosen": 363.4445495605469,
"logits/rejected": 406.9335021972656,
"logps/chosen": -0.3999672532081604,
"logps/rejected": -3.7019195556640625,
"loss": 0.6043,
"nll_loss": 0.5209922790527344,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.1999836266040802,
"rewards/margins": 1.6509761810302734,
"rewards/rejected": -1.8509597778320312,
"step": 495
},
{
"epoch": 2.9761904761904763,
"grad_norm": 10.738035202026367,
"learning_rate": 4.47213595499958e-06,
"log_odds_chosen": 4.959429740905762,
"log_odds_ratio": -0.03518088907003403,
"logits/chosen": 417.14154052734375,
"logits/rejected": 379.5878601074219,
"logps/chosen": -0.2811127007007599,
"logps/rejected": -3.7553272247314453,
"loss": 0.6266,
"nll_loss": 0.42872363328933716,
"rewards/accuracies": 1.0,
"rewards/chosen": -0.14055635035037994,
"rewards/margins": 1.7371070384979248,
"rewards/rejected": -1.8776636123657227,
"step": 500
},
{
"epoch": 3.0,
"eval_log_odds_chosen": 0.44554343819618225,
"eval_log_odds_ratio": -0.6395100355148315,
"eval_logits/chosen": 319.3428649902344,
"eval_logits/rejected": 262.5517883300781,
"eval_logps/chosen": -1.3908888101577759,
"eval_logps/rejected": -1.6692452430725098,
"eval_loss": 2.1094672679901123,
"eval_nll_loss": 1.7835990190505981,
"eval_rewards/accuracies": 0.5571428537368774,
"eval_rewards/chosen": -0.6954444050788879,
"eval_rewards/margins": 0.13917820155620575,
"eval_rewards/rejected": -0.8346226215362549,
"eval_runtime": 201.4063,
"eval_samples_per_second": 2.746,
"eval_steps_per_second": 0.348,
"step": 504
},
{
"epoch": 3.0,
"step": 504,
"total_flos": 0.0,
"train_loss": 1.593297282854716,
"train_runtime": 15168.4679,
"train_samples_per_second": 1.061,
"train_steps_per_second": 0.033
}
],
"logging_steps": 5,
"max_steps": 504,
"num_input_tokens_seen": 0,
"num_train_epochs": 3,
"save_steps": 500,
"stateful_callbacks": {
"TrainerControl": {
"args": {
"should_epoch_stop": false,
"should_evaluate": false,
"should_log": false,
"should_save": false,
"should_training_stop": false
},
"attributes": {}
}
},
"total_flos": 0.0,
"train_batch_size": 2,
"trial_name": null,
"trial_params": null
}