gemma-7b-orpo-noisy-5e-6-norm / trainer_state.json
silviasapora's picture
Model save
de3df65 verified
{
"best_metric": null,
"best_model_checkpoint": null,
"epoch": 2.9765925925925925,
"eval_steps": 500,
"global_step": 315,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 0.047407407407407405,
"grad_norm": 2865.851318359375,
"learning_rate": 7.8125e-07,
"log_odds_chosen": 5.047066688537598,
"log_odds_ratio": -9.840336799621582,
"logits/chosen": 138.34872436523438,
"logits/rejected": 152.904296875,
"logps/chosen": -20.9003849029541,
"logps/rejected": -25.947729110717773,
"loss": 459.6987,
"nll_loss": 8.465580940246582,
"rewards/accuracies": 0.53125,
"rewards/chosen": -10.45019245147705,
"rewards/margins": 2.523672580718994,
"rewards/rejected": -12.973864555358887,
"step": 5
},
{
"epoch": 0.09481481481481481,
"grad_norm": 2037.79638671875,
"learning_rate": 1.5625e-06,
"log_odds_chosen": 2.4296860694885254,
"log_odds_ratio": -7.415003299713135,
"logits/chosen": 131.40785217285156,
"logits/rejected": 155.49392700195312,
"logps/chosen": -16.34769058227539,
"logps/rejected": -18.776687622070312,
"loss": 405.2566,
"nll_loss": 7.926710605621338,
"rewards/accuracies": 0.543749988079071,
"rewards/chosen": -8.173845291137695,
"rewards/margins": 1.21449875831604,
"rewards/rejected": -9.388343811035156,
"step": 10
},
{
"epoch": 0.14222222222222222,
"grad_norm": 716.259521484375,
"learning_rate": 2.3437500000000002e-06,
"log_odds_chosen": 4.6704511642456055,
"log_odds_ratio": -8.288759231567383,
"logits/chosen": 103.96134948730469,
"logits/rejected": 130.3046417236328,
"logps/chosen": -21.693960189819336,
"logps/rejected": -26.36324691772461,
"loss": 396.9101,
"nll_loss": 9.263435363769531,
"rewards/accuracies": 0.5687500238418579,
"rewards/chosen": -10.846980094909668,
"rewards/margins": 2.334644079208374,
"rewards/rejected": -13.181623458862305,
"step": 15
},
{
"epoch": 0.18962962962962962,
"grad_norm": 773.0562133789062,
"learning_rate": 3.125e-06,
"log_odds_chosen": -0.22766943275928497,
"log_odds_ratio": -8.08221435546875,
"logits/chosen": 127.30647277832031,
"logits/rejected": 133.12646484375,
"logps/chosen": -18.663068771362305,
"logps/rejected": -18.436843872070312,
"loss": 433.5117,
"nll_loss": 8.696657180786133,
"rewards/accuracies": 0.53125,
"rewards/chosen": -9.331534385681152,
"rewards/margins": -0.11311338096857071,
"rewards/rejected": -9.218421936035156,
"step": 20
},
{
"epoch": 0.23703703703703705,
"grad_norm": 461.5646667480469,
"learning_rate": 3.90625e-06,
"log_odds_chosen": 4.552544593811035,
"log_odds_ratio": -8.04813003540039,
"logits/chosen": 129.98939514160156,
"logits/rejected": 138.5162811279297,
"logps/chosen": -19.01546287536621,
"logps/rejected": -23.568456649780273,
"loss": 361.2036,
"nll_loss": 7.481656074523926,
"rewards/accuracies": 0.53125,
"rewards/chosen": -9.507731437683105,
"rewards/margins": 2.276496648788452,
"rewards/rejected": -11.784228324890137,
"step": 25
},
{
"epoch": 0.28444444444444444,
"grad_norm": 364.19580078125,
"learning_rate": 4.6875000000000004e-06,
"log_odds_chosen": 1.745486855506897,
"log_odds_ratio": -7.753143310546875,
"logits/chosen": 116.81257629394531,
"logits/rejected": 154.61524963378906,
"logps/chosen": -19.596527099609375,
"logps/rejected": -21.339160919189453,
"loss": 394.3339,
"nll_loss": 8.240059852600098,
"rewards/accuracies": 0.574999988079071,
"rewards/chosen": -9.798263549804688,
"rewards/margins": 0.8713172674179077,
"rewards/rejected": -10.669580459594727,
"step": 30
},
{
"epoch": 0.33185185185185184,
"grad_norm": 526.892333984375,
"learning_rate": 4.998613757348784e-06,
"log_odds_chosen": -2.7288818359375,
"log_odds_ratio": -8.869112968444824,
"logits/chosen": 149.9911651611328,
"logits/rejected": 133.53860473632812,
"logps/chosen": -19.254487991333008,
"logps/rejected": -16.524951934814453,
"loss": 371.9777,
"nll_loss": 7.2948455810546875,
"rewards/accuracies": 0.4937500059604645,
"rewards/chosen": -9.627243995666504,
"rewards/margins": -1.3647680282592773,
"rewards/rejected": -8.262475967407227,
"step": 35
},
{
"epoch": 0.37925925925925924,
"grad_norm": 638.0831909179688,
"learning_rate": 4.990147841143462e-06,
"log_odds_chosen": 2.0575106143951416,
"log_odds_ratio": -4.877079963684082,
"logits/chosen": 153.21884155273438,
"logits/rejected": 167.37942504882812,
"logps/chosen": -14.817087173461914,
"logps/rejected": -16.873952865600586,
"loss": 333.9902,
"nll_loss": 7.247918128967285,
"rewards/accuracies": 0.48750001192092896,
"rewards/chosen": -7.408543586730957,
"rewards/margins": 1.0284324884414673,
"rewards/rejected": -8.436976432800293,
"step": 40
},
{
"epoch": 0.4266666666666667,
"grad_norm": 895.48095703125,
"learning_rate": 4.97401218720448e-06,
"log_odds_chosen": 1.4577323198318481,
"log_odds_ratio": -8.850030899047852,
"logits/chosen": 176.9611053466797,
"logits/rejected": 164.4369354248047,
"logps/chosen": -18.82620620727539,
"logps/rejected": -20.282642364501953,
"loss": 335.1132,
"nll_loss": 6.7939629554748535,
"rewards/accuracies": 0.53125,
"rewards/chosen": -9.413103103637695,
"rewards/margins": 0.7282184362411499,
"rewards/rejected": -10.141321182250977,
"step": 45
},
{
"epoch": 0.4740740740740741,
"grad_norm": 1942.3775634765625,
"learning_rate": 4.950256493879795e-06,
"log_odds_chosen": 2.4757702350616455,
"log_odds_ratio": -5.5582990646362305,
"logits/chosen": 197.7135772705078,
"logits/rejected": 197.5384521484375,
"logps/chosen": -13.839834213256836,
"logps/rejected": -16.314041137695312,
"loss": 313.9627,
"nll_loss": 6.590758323669434,
"rewards/accuracies": 0.4937500059604645,
"rewards/chosen": -6.919917106628418,
"rewards/margins": 1.237102746963501,
"rewards/rejected": -8.157020568847656,
"step": 50
},
{
"epoch": 0.5214814814814814,
"grad_norm": 1690.2572021484375,
"learning_rate": 4.918953929490768e-06,
"log_odds_chosen": 0.49450284242630005,
"log_odds_ratio": -4.888993740081787,
"logits/chosen": 212.30068969726562,
"logits/rejected": 219.568115234375,
"logps/chosen": -10.383580207824707,
"logps/rejected": -10.86988353729248,
"loss": 223.3722,
"nll_loss": 5.05554723739624,
"rewards/accuracies": 0.543749988079071,
"rewards/chosen": -5.1917901039123535,
"rewards/margins": 0.24315185844898224,
"rewards/rejected": -5.43494176864624,
"step": 55
},
{
"epoch": 0.5688888888888889,
"grad_norm": 983.438720703125,
"learning_rate": 4.880200906971321e-06,
"log_odds_chosen": 0.5941994786262512,
"log_odds_ratio": -1.8958613872528076,
"logits/chosen": 207.96792602539062,
"logits/rejected": 225.57461547851562,
"logps/chosen": -5.315521717071533,
"logps/rejected": -5.895586967468262,
"loss": 129.5271,
"nll_loss": 3.1816208362579346,
"rewards/accuracies": 0.5625,
"rewards/chosen": -2.6577608585357666,
"rewards/margins": 0.2900325059890747,
"rewards/rejected": -2.947793483734131,
"step": 60
},
{
"epoch": 0.6162962962962963,
"grad_norm": 435.369384765625,
"learning_rate": 4.834116786912897e-06,
"log_odds_chosen": -0.35661205649375916,
"log_odds_ratio": -1.3505533933639526,
"logits/chosen": 233.6535186767578,
"logits/rejected": 232.8253936767578,
"logps/chosen": -2.915100574493408,
"logps/rejected": -2.5467069149017334,
"loss": 93.9753,
"nll_loss": 2.4082727432250977,
"rewards/accuracies": 0.5062500238418579,
"rewards/chosen": -1.457550287246704,
"rewards/margins": -0.18419687449932098,
"rewards/rejected": -1.2733534574508667,
"step": 65
},
{
"epoch": 0.6637037037037037,
"grad_norm": 266.7743835449219,
"learning_rate": 4.780843509929905e-06,
"log_odds_chosen": 0.5245118141174316,
"log_odds_ratio": -0.6836276650428772,
"logits/chosen": 223.84109497070312,
"logits/rejected": 249.83383178710938,
"logps/chosen": -1.803598165512085,
"logps/rejected": -2.3056464195251465,
"loss": 80.6457,
"nll_loss": 2.1085317134857178,
"rewards/accuracies": 0.5625,
"rewards/chosen": -0.9017990827560425,
"rewards/margins": 0.25102415680885315,
"rewards/rejected": -1.1528232097625732,
"step": 70
},
{
"epoch": 0.7111111111111111,
"grad_norm": 146.09365844726562,
"learning_rate": 4.720545159477921e-06,
"log_odds_chosen": 0.19110290706157684,
"log_odds_ratio": -0.896527886390686,
"logits/chosen": 229.1388702392578,
"logits/rejected": 218.0093536376953,
"logps/chosen": -1.8344182968139648,
"logps/rejected": -1.9851324558258057,
"loss": 76.4303,
"nll_loss": 2.0232486724853516,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.9172091484069824,
"rewards/margins": 0.0753571167588234,
"rewards/rejected": -0.9925662279129028,
"step": 75
},
{
"epoch": 0.7585185185185185,
"grad_norm": 188.69973754882812,
"learning_rate": 4.653407456471222e-06,
"log_odds_chosen": 0.3417088985443115,
"log_odds_ratio": -0.7933382391929626,
"logits/chosen": 206.59188842773438,
"logits/rejected": 223.7476348876953,
"logps/chosen": -1.6943140029907227,
"logps/rejected": -1.9918110370635986,
"loss": 73.3448,
"nll_loss": 1.9511661529541016,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -0.8471570014953613,
"rewards/margins": 0.1487484872341156,
"rewards/rejected": -0.9959055185317993,
"step": 80
},
{
"epoch": 0.8059259259259259,
"grad_norm": 137.05148315429688,
"learning_rate": 4.579637187256222e-06,
"log_odds_chosen": 0.39623409509658813,
"log_odds_ratio": -0.7564057111740112,
"logits/chosen": 211.7617950439453,
"logits/rejected": 254.6263885498047,
"logps/chosen": -1.5494643449783325,
"logps/rejected": -1.891304612159729,
"loss": 71.0347,
"nll_loss": 1.7963597774505615,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.7747321724891663,
"rewards/margins": 0.17092011868953705,
"rewards/rejected": -0.9456523060798645,
"step": 85
},
{
"epoch": 0.8533333333333334,
"grad_norm": 158.83348083496094,
"learning_rate": 4.499461566702685e-06,
"log_odds_chosen": 0.28660133481025696,
"log_odds_ratio": -0.7941745519638062,
"logits/chosen": 222.36099243164062,
"logits/rejected": 244.03665161132812,
"logps/chosen": -1.6755253076553345,
"logps/rejected": -1.9040807485580444,
"loss": 73.0658,
"nll_loss": 1.8526337146759033,
"rewards/accuracies": 0.612500011920929,
"rewards/chosen": -0.8377626538276672,
"rewards/margins": 0.11427769809961319,
"rewards/rejected": -0.9520403742790222,
"step": 90
},
{
"epoch": 0.9007407407407407,
"grad_norm": 170.69134521484375,
"learning_rate": 4.413127538374411e-06,
"log_odds_chosen": 0.4414753317832947,
"log_odds_ratio": -0.6728295087814331,
"logits/chosen": 227.6732940673828,
"logits/rejected": 249.4246063232422,
"logps/chosen": -1.4834215641021729,
"logps/rejected": -1.850646734237671,
"loss": 70.3617,
"nll_loss": 1.7307571172714233,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -0.7417107820510864,
"rewards/margins": 0.18361257016658783,
"rewards/rejected": -0.9253233671188354,
"step": 95
},
{
"epoch": 0.9481481481481482,
"grad_norm": 195.4022216796875,
"learning_rate": 4.320901013934887e-06,
"log_odds_chosen": 0.26656374335289,
"log_odds_ratio": -0.7893905639648438,
"logits/chosen": 230.3835906982422,
"logits/rejected": 250.1712188720703,
"logps/chosen": -1.6341091394424438,
"logps/rejected": -1.8852962255477905,
"loss": 70.7829,
"nll_loss": 1.838405966758728,
"rewards/accuracies": 0.5375000238418579,
"rewards/chosen": -0.8170545697212219,
"rewards/margins": 0.12559358775615692,
"rewards/rejected": -0.9426481127738953,
"step": 100
},
{
"epoch": 0.9955555555555555,
"grad_norm": 120.9432144165039,
"learning_rate": 4.223066054130568e-06,
"log_odds_chosen": 0.40909630060195923,
"log_odds_ratio": -0.6454111337661743,
"logits/chosen": 230.5096435546875,
"logits/rejected": 254.70449829101562,
"logps/chosen": -1.4842649698257446,
"logps/rejected": -1.8271305561065674,
"loss": 65.2115,
"nll_loss": 1.6993058919906616,
"rewards/accuracies": 0.6312500238418579,
"rewards/chosen": -0.7421324849128723,
"rewards/margins": 0.17143282294273376,
"rewards/rejected": -0.9135652780532837,
"step": 105
},
{
"epoch": 1.037925925925926,
"grad_norm": 135.30653381347656,
"learning_rate": 4.11992399387438e-06,
"log_odds_chosen": 0.27806416153907776,
"log_odds_ratio": -0.739980936050415,
"logits/chosen": 248.7511749267578,
"logits/rejected": 262.7025146484375,
"logps/chosen": -1.4985274076461792,
"logps/rejected": -1.7603181600570679,
"loss": 58.8669,
"nll_loss": 1.6877117156982422,
"rewards/accuracies": 0.5874125957489014,
"rewards/chosen": -0.7492637038230896,
"rewards/margins": 0.1308954954147339,
"rewards/rejected": -0.8801590800285339,
"step": 110
},
{
"epoch": 1.0853333333333333,
"grad_norm": 111.60261535644531,
"learning_rate": 4.011792514124217e-06,
"log_odds_chosen": 0.49480900168418884,
"log_odds_ratio": -0.6541386842727661,
"logits/chosen": 237.9220733642578,
"logits/rejected": 269.5302429199219,
"logps/chosen": -1.4647657871246338,
"logps/rejected": -1.8502311706542969,
"loss": 64.0187,
"nll_loss": 1.6229032278060913,
"rewards/accuracies": 0.6812499761581421,
"rewards/chosen": -0.7323828935623169,
"rewards/margins": 0.19273266196250916,
"rewards/rejected": -0.9251155853271484,
"step": 115
},
{
"epoch": 1.1327407407407408,
"grad_norm": 136.4354248046875,
"learning_rate": 3.899004663415083e-06,
"log_odds_chosen": 0.5176212191581726,
"log_odds_ratio": -0.6403207778930664,
"logits/chosen": 233.19796752929688,
"logits/rejected": 246.4919891357422,
"logps/chosen": -1.4554225206375122,
"logps/rejected": -1.8822691440582275,
"loss": 66.2991,
"nll_loss": 1.7435401678085327,
"rewards/accuracies": 0.6312500238418579,
"rewards/chosen": -0.7277112603187561,
"rewards/margins": 0.21342334151268005,
"rewards/rejected": -0.9411345720291138,
"step": 120
},
{
"epoch": 1.1801481481481482,
"grad_norm": 115.97614288330078,
"learning_rate": 3.7819078320585865e-06,
"log_odds_chosen": 0.27887487411499023,
"log_odds_ratio": -0.7117995023727417,
"logits/chosen": 243.6487274169922,
"logits/rejected": 249.3553924560547,
"logps/chosen": -1.5128828287124634,
"logps/rejected": -1.7247974872589111,
"loss": 63.7844,
"nll_loss": 1.6903963088989258,
"rewards/accuracies": 0.6312500238418579,
"rewards/chosen": -0.7564414143562317,
"rewards/margins": 0.10595734417438507,
"rewards/rejected": -0.8623987436294556,
"step": 125
},
{
"epoch": 1.2275555555555555,
"grad_norm": 105.05059051513672,
"learning_rate": 3.660862682169283e-06,
"log_odds_chosen": 0.2080799639225006,
"log_odds_ratio": -0.7350869178771973,
"logits/chosen": 235.3159942626953,
"logits/rejected": 248.73782348632812,
"logps/chosen": -1.3867459297180176,
"logps/rejected": -1.5404752492904663,
"loss": 61.9593,
"nll_loss": 1.6034740209579468,
"rewards/accuracies": 0.5687500238418579,
"rewards/chosen": -0.6933729648590088,
"rewards/margins": 0.07686467468738556,
"rewards/rejected": -0.7702376246452332,
"step": 130
},
{
"epoch": 1.274962962962963,
"grad_norm": 129.73594665527344,
"learning_rate": 3.536242036813436e-06,
"log_odds_chosen": 0.2643691599369049,
"log_odds_ratio": -0.7503995895385742,
"logits/chosen": 247.5218048095703,
"logits/rejected": 255.9250030517578,
"logps/chosen": -1.4334195852279663,
"logps/rejected": -1.6370153427124023,
"loss": 62.8912,
"nll_loss": 1.6255853176116943,
"rewards/accuracies": 0.5874999761581421,
"rewards/chosen": -0.7167097926139832,
"rewards/margins": 0.10179783403873444,
"rewards/rejected": -0.8185076713562012,
"step": 135
},
{
"epoch": 1.3223703703703704,
"grad_norm": 156.6639404296875,
"learning_rate": 3.4084297317016353e-06,
"log_odds_chosen": 0.2968347668647766,
"log_odds_ratio": -0.7262933254241943,
"logits/chosen": 256.20166015625,
"logits/rejected": 252.2810516357422,
"logps/chosen": -1.4891420602798462,
"logps/rejected": -1.745237946510315,
"loss": 65.243,
"nll_loss": 1.665055274963379,
"rewards/accuracies": 0.6187499761581421,
"rewards/chosen": -0.7445710301399231,
"rewards/margins": 0.12804797291755676,
"rewards/rejected": -0.8726189732551575,
"step": 140
},
{
"epoch": 1.3697777777777778,
"grad_norm": 133.37393188476562,
"learning_rate": 3.2778194329621104e-06,
"log_odds_chosen": 0.48826122283935547,
"log_odds_ratio": -0.6439377665519714,
"logits/chosen": 242.45767211914062,
"logits/rejected": 267.6114501953125,
"logps/chosen": -1.3750956058502197,
"logps/rejected": -1.8073225021362305,
"loss": 62.7163,
"nll_loss": 1.6385784149169922,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -0.6875478029251099,
"rewards/margins": 0.21611353754997253,
"rewards/rejected": -0.9036612510681152,
"step": 145
},
{
"epoch": 1.417185185185185,
"grad_norm": 145.82296752929688,
"learning_rate": 3.144813424636031e-06,
"log_odds_chosen": 0.10933913290500641,
"log_odds_ratio": -0.7888091802597046,
"logits/chosen": 251.4317169189453,
"logits/rejected": 244.14462280273438,
"logps/chosen": -1.416845679283142,
"logps/rejected": -1.5029141902923584,
"loss": 62.2472,
"nll_loss": 1.6043399572372437,
"rewards/accuracies": 0.550000011920929,
"rewards/chosen": -0.708422839641571,
"rewards/margins": 0.043034255504608154,
"rewards/rejected": -0.7514570951461792,
"step": 150
},
{
"epoch": 1.4645925925925927,
"grad_norm": 124.13430786132812,
"learning_rate": 3.0098213696293542e-06,
"log_odds_chosen": 0.3432820439338684,
"log_odds_ratio": -0.6613715887069702,
"logits/chosen": 228.41439819335938,
"logits/rejected": 260.245849609375,
"logps/chosen": -1.31131911277771,
"logps/rejected": -1.5758918523788452,
"loss": 61.9618,
"nll_loss": 1.592642903327942,
"rewards/accuracies": 0.6187499761581421,
"rewards/chosen": -0.655659556388855,
"rewards/margins": 0.13228638470172882,
"rewards/rejected": -0.7879459261894226,
"step": 155
},
{
"epoch": 1.512,
"grad_norm": 169.77215576171875,
"learning_rate": 2.8732590479375167e-06,
"log_odds_chosen": 0.3919784128665924,
"log_odds_ratio": -0.6399356126785278,
"logits/chosen": 238.72976684570312,
"logits/rejected": 242.7633514404297,
"logps/chosen": -1.339658498764038,
"logps/rejected": -1.6707757711410522,
"loss": 62.1449,
"nll_loss": 1.5825494527816772,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -0.669829249382019,
"rewards/margins": 0.1655586212873459,
"rewards/rejected": -0.8353878855705261,
"step": 160
},
{
"epoch": 1.5594074074074074,
"grad_norm": 136.43710327148438,
"learning_rate": 2.735547076029296e-06,
"log_odds_chosen": 0.28347623348236084,
"log_odds_ratio": -0.6901515126228333,
"logits/chosen": 249.5279541015625,
"logits/rejected": 233.10983276367188,
"logps/chosen": -1.348504900932312,
"logps/rejected": -1.563481092453003,
"loss": 62.4916,
"nll_loss": 1.5731277465820312,
"rewards/accuracies": 0.637499988079071,
"rewards/chosen": -0.674252450466156,
"rewards/margins": 0.10748811066150665,
"rewards/rejected": -0.7817405462265015,
"step": 165
},
{
"epoch": 1.6068148148148147,
"grad_norm": 162.6702880859375,
"learning_rate": 2.5971096113341692e-06,
"log_odds_chosen": 0.3690604269504547,
"log_odds_ratio": -0.6491698622703552,
"logits/chosen": 239.92678833007812,
"logits/rejected": 247.5673828125,
"logps/chosen": -1.3685719966888428,
"logps/rejected": -1.6624857187271118,
"loss": 61.4999,
"nll_loss": 1.5894715785980225,
"rewards/accuracies": 0.6187499761581421,
"rewards/chosen": -0.6842859983444214,
"rewards/margins": 0.14695683121681213,
"rewards/rejected": -0.8312428593635559,
"step": 170
},
{
"epoch": 1.6542222222222223,
"grad_norm": 121.98951721191406,
"learning_rate": 2.458373045823404e-06,
"log_odds_chosen": 0.37640315294265747,
"log_odds_ratio": -0.6424993872642517,
"logits/chosen": 243.83737182617188,
"logits/rejected": 263.46844482421875,
"logps/chosen": -1.3080052137374878,
"logps/rejected": -1.6126611232757568,
"loss": 61.3657,
"nll_loss": 1.595855712890625,
"rewards/accuracies": 0.6312500238418579,
"rewards/chosen": -0.6540026068687439,
"rewards/margins": 0.15232795476913452,
"rewards/rejected": -0.8063305616378784,
"step": 175
},
{
"epoch": 1.7016296296296296,
"grad_norm": 126.68740844726562,
"learning_rate": 2.3197646927086697e-06,
"log_odds_chosen": 0.05865911766886711,
"log_odds_ratio": -0.7959011793136597,
"logits/chosen": 255.73562622070312,
"logits/rejected": 250.8922882080078,
"logps/chosen": -1.4331071376800537,
"logps/rejected": -1.4851996898651123,
"loss": 62.2336,
"nll_loss": 1.6543314456939697,
"rewards/accuracies": 0.5375000238418579,
"rewards/chosen": -0.7165535688400269,
"rewards/margins": 0.026046359911561012,
"rewards/rejected": -0.7425998449325562,
"step": 180
},
{
"epoch": 1.749037037037037,
"grad_norm": 126.31131744384766,
"learning_rate": 2.1817114703032176e-06,
"log_odds_chosen": 0.3499384820461273,
"log_odds_ratio": -0.6687244176864624,
"logits/chosen": 250.3800048828125,
"logits/rejected": 253.47647094726562,
"logps/chosen": -1.3136674165725708,
"logps/rejected": -1.603899598121643,
"loss": 61.6932,
"nll_loss": 1.5699495077133179,
"rewards/accuracies": 0.6187499761581421,
"rewards/chosen": -0.6568337082862854,
"rewards/margins": 0.14511603116989136,
"rewards/rejected": -0.8019497990608215,
"step": 185
},
{
"epoch": 1.7964444444444445,
"grad_norm": 117.38888549804688,
"learning_rate": 2.044638587099347e-06,
"log_odds_chosen": 0.3149377703666687,
"log_odds_ratio": -0.661319375038147,
"logits/chosen": 235.9986572265625,
"logits/rejected": 258.65423583984375,
"logps/chosen": -1.3008835315704346,
"logps/rejected": -1.5506656169891357,
"loss": 61.813,
"nll_loss": 1.5823218822479248,
"rewards/accuracies": 0.643750011920929,
"rewards/chosen": -0.6504417657852173,
"rewards/margins": 0.12489104270935059,
"rewards/rejected": -0.7753328084945679,
"step": 190
},
{
"epoch": 1.8438518518518519,
"grad_norm": 142.8717498779297,
"learning_rate": 1.9089682321121834e-06,
"log_odds_chosen": 0.4244133532047272,
"log_odds_ratio": -0.6273762583732605,
"logits/chosen": 248.58889770507812,
"logits/rejected": 268.8494567871094,
"logps/chosen": -1.3011976480484009,
"logps/rejected": -1.6522626876831055,
"loss": 60.3822,
"nll_loss": 1.586854100227356,
"rewards/accuracies": 0.637499988079071,
"rewards/chosen": -0.6505988240242004,
"rewards/margins": 0.17553254961967468,
"rewards/rejected": -0.8261313438415527,
"step": 195
},
{
"epoch": 1.8912592592592592,
"grad_norm": 140.85545349121094,
"learning_rate": 1.775118274523545e-06,
"log_odds_chosen": 0.1862161010503769,
"log_odds_ratio": -0.7570394277572632,
"logits/chosen": 239.0512237548828,
"logits/rejected": 254.9342041015625,
"logps/chosen": -1.307045578956604,
"logps/rejected": -1.4443647861480713,
"loss": 61.6621,
"nll_loss": 1.504206895828247,
"rewards/accuracies": 0.5687500238418579,
"rewards/chosen": -0.653522789478302,
"rewards/margins": 0.06865964829921722,
"rewards/rejected": -0.7221823930740356,
"step": 200
},
{
"epoch": 1.9386666666666668,
"grad_norm": 124.06290435791016,
"learning_rate": 1.6435009766310372e-06,
"log_odds_chosen": 0.3265306353569031,
"log_odds_ratio": -0.6406243443489075,
"logits/chosen": 236.4462890625,
"logits/rejected": 262.8599548339844,
"logps/chosen": -1.22454035282135,
"logps/rejected": -1.4862773418426514,
"loss": 58.7041,
"nll_loss": 1.498868703842163,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -0.612270176410675,
"rewards/margins": 0.13086840510368347,
"rewards/rejected": -0.7431386709213257,
"step": 205
},
{
"epoch": 1.986074074074074,
"grad_norm": 139.6620330810547,
"learning_rate": 1.5145217240665373e-06,
"log_odds_chosen": 0.2167925387620926,
"log_odds_ratio": -0.6764565110206604,
"logits/chosen": 254.82162475585938,
"logits/rejected": 242.75906372070312,
"logps/chosen": -1.2110469341278076,
"logps/rejected": -1.3860079050064087,
"loss": 57.7464,
"nll_loss": 1.4964487552642822,
"rewards/accuracies": 0.5687500238418579,
"rewards/chosen": -0.6055234670639038,
"rewards/margins": 0.08748046308755875,
"rewards/rejected": -0.6930039525032043,
"step": 210
},
{
"epoch": 2.0284444444444443,
"grad_norm": 164.2100372314453,
"learning_rate": 1.388577777195035e-06,
"log_odds_chosen": 0.36885133385658264,
"log_odds_ratio": -0.6371248960494995,
"logits/chosen": 246.92779541015625,
"logits/rejected": 258.91119384765625,
"logps/chosen": -1.2711718082427979,
"logps/rejected": -1.5534406900405884,
"loss": 53.7807,
"nll_loss": 1.5334032773971558,
"rewards/accuracies": 0.6643356680870056,
"rewards/chosen": -0.6355859041213989,
"rewards/margins": 0.14113430678844452,
"rewards/rejected": -0.7767203450202942,
"step": 215
},
{
"epoch": 2.075851851851852,
"grad_norm": 153.154052734375,
"learning_rate": 1.2660570475395684e-06,
"log_odds_chosen": 0.4296096861362457,
"log_odds_ratio": -0.6230024695396423,
"logits/chosen": 241.6239013671875,
"logits/rejected": 247.5516815185547,
"logps/chosen": -1.179203748703003,
"logps/rejected": -1.5177547931671143,
"loss": 57.7354,
"nll_loss": 1.486169457435608,
"rewards/accuracies": 0.643750011920929,
"rewards/chosen": -0.5896018743515015,
"rewards/margins": 0.16927553713321686,
"rewards/rejected": -0.7588773965835571,
"step": 220
},
{
"epoch": 2.1232592592592594,
"grad_norm": 106.61084747314453,
"learning_rate": 1.1473369030008974e-06,
"log_odds_chosen": 0.32008111476898193,
"log_odds_ratio": -0.6721247434616089,
"logits/chosen": 247.1652069091797,
"logits/rejected": 241.549072265625,
"logps/chosen": -1.2260723114013672,
"logps/rejected": -1.4864741563796997,
"loss": 58.5367,
"nll_loss": 1.5331952571868896,
"rewards/accuracies": 0.6187499761581421,
"rewards/chosen": -0.6130361557006836,
"rewards/margins": 0.13020093739032745,
"rewards/rejected": -0.7432370781898499,
"step": 225
},
{
"epoch": 2.1706666666666665,
"grad_norm": 148.289794921875,
"learning_rate": 1.0327830055518843e-06,
"log_odds_chosen": 0.43876928091049194,
"log_odds_ratio": -0.6304255127906799,
"logits/chosen": 246.761962890625,
"logits/rejected": 262.81085205078125,
"logps/chosen": -1.3328922986984253,
"logps/rejected": -1.6859073638916016,
"loss": 58.6867,
"nll_loss": 1.554168462753296,
"rewards/accuracies": 0.6812499761581421,
"rewards/chosen": -0.6664461493492126,
"rewards/margins": 0.17650756239891052,
"rewards/rejected": -0.8429536819458008,
"step": 230
},
{
"epoch": 2.218074074074074,
"grad_norm": 181.2135467529297,
"learning_rate": 9.227481849865236e-07,
"log_odds_chosen": 0.3924606442451477,
"log_odds_ratio": -0.6264532208442688,
"logits/chosen": 233.6820831298828,
"logits/rejected": 240.0151824951172,
"logps/chosen": -1.2045910358428955,
"logps/rejected": -1.5118606090545654,
"loss": 58.7122,
"nll_loss": 1.4953967332839966,
"rewards/accuracies": 0.637499988079071,
"rewards/chosen": -0.6022955179214478,
"rewards/margins": 0.15363483130931854,
"rewards/rejected": -0.7559303045272827,
"step": 235
},
{
"epoch": 2.2654814814814817,
"grad_norm": 110.6124496459961,
"learning_rate": 8.175713521924977e-07,
"log_odds_chosen": 0.31769445538520813,
"log_odds_ratio": -0.6737982630729675,
"logits/chosen": 243.349853515625,
"logits/rejected": 242.6227264404297,
"logps/chosen": -1.2512071132659912,
"logps/rejected": -1.5016454458236694,
"loss": 58.2297,
"nll_loss": 1.538447618484497,
"rewards/accuracies": 0.606249988079071,
"rewards/chosen": -0.6256035566329956,
"rewards/margins": 0.12521910667419434,
"rewards/rejected": -0.7508227229118347,
"step": 240
},
{
"epoch": 2.3128888888888888,
"grad_norm": 142.4850616455078,
"learning_rate": 7.175764552944368e-07,
"log_odds_chosen": 0.4037380814552307,
"log_odds_ratio": -0.6004605293273926,
"logits/chosen": 235.74459838867188,
"logits/rejected": 261.41607666015625,
"logps/chosen": -1.1916182041168213,
"logps/rejected": -1.5017354488372803,
"loss": 56.862,
"nll_loss": 1.520082712173462,
"rewards/accuracies": 0.65625,
"rewards/chosen": -0.5958091020584106,
"rewards/margins": 0.1550586223602295,
"rewards/rejected": -0.7508677244186401,
"step": 245
},
{
"epoch": 2.3602962962962963,
"grad_norm": 135.36468505859375,
"learning_rate": 6.230714818829733e-07,
"log_odds_chosen": 0.402596652507782,
"log_odds_ratio": -0.6101157069206238,
"logits/chosen": 243.6725311279297,
"logits/rejected": 250.9963836669922,
"logps/chosen": -1.2181828022003174,
"logps/rejected": -1.5303680896759033,
"loss": 56.9306,
"nll_loss": 1.48748779296875,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -0.6090914011001587,
"rewards/margins": 0.1560925841331482,
"rewards/rejected": -0.7651840448379517,
"step": 250
},
{
"epoch": 2.407703703703704,
"grad_norm": 140.75233459472656,
"learning_rate": 5.343475104027743e-07,
"log_odds_chosen": 0.4917730391025543,
"log_odds_ratio": -0.6072665452957153,
"logits/chosen": 241.0843505859375,
"logits/rejected": 245.1985626220703,
"logps/chosen": -1.2005137205123901,
"logps/rejected": -1.5952293872833252,
"loss": 56.6815,
"nll_loss": 1.4808170795440674,
"rewards/accuracies": 0.675000011920929,
"rewards/chosen": -0.6002568602561951,
"rewards/margins": 0.1973578780889511,
"rewards/rejected": -0.7976146936416626,
"step": 255
},
{
"epoch": 2.455111111111111,
"grad_norm": 158.60487365722656,
"learning_rate": 4.5167781362130374e-07,
"log_odds_chosen": 0.5367686152458191,
"log_odds_ratio": -0.5729767084121704,
"logits/chosen": 228.5010528564453,
"logits/rejected": 252.38162231445312,
"logps/chosen": -1.2318106889724731,
"logps/rejected": -1.6468757390975952,
"loss": 57.0996,
"nll_loss": 1.498032808303833,
"rewards/accuracies": 0.71875,
"rewards/chosen": -0.6159053444862366,
"rewards/margins": 0.20753255486488342,
"rewards/rejected": -0.8234378695487976,
"step": 260
},
{
"epoch": 2.5025185185185186,
"grad_norm": 137.73834228515625,
"learning_rate": 3.7531701693965555e-07,
"log_odds_chosen": 0.32621318101882935,
"log_odds_ratio": -0.6674071550369263,
"logits/chosen": 242.7371063232422,
"logits/rejected": 238.8314666748047,
"logps/chosen": -1.201547622680664,
"logps/rejected": -1.4522160291671753,
"loss": 56.4953,
"nll_loss": 1.4682562351226807,
"rewards/accuracies": 0.643750011920929,
"rewards/chosen": -0.600773811340332,
"rewards/margins": 0.125334233045578,
"rewards/rejected": -0.7261080145835876,
"step": 265
},
{
"epoch": 2.549925925925926,
"grad_norm": 177.38958740234375,
"learning_rate": 3.0550031413789485e-07,
"log_odds_chosen": 0.5363808870315552,
"log_odds_ratio": -0.5923758745193481,
"logits/chosen": 230.21829223632812,
"logits/rejected": 238.24636840820312,
"logps/chosen": -1.2215114831924438,
"logps/rejected": -1.6396996974945068,
"loss": 55.9249,
"nll_loss": 1.4734989404678345,
"rewards/accuracies": 0.706250011920929,
"rewards/chosen": -0.6107557415962219,
"rewards/margins": 0.20909413695335388,
"rewards/rejected": -0.8198498487472534,
"step": 270
},
{
"epoch": 2.5973333333333333,
"grad_norm": 157.81935119628906,
"learning_rate": 2.4244274297043653e-07,
"log_odds_chosen": 0.5345597267150879,
"log_odds_ratio": -0.5606809258460999,
"logits/chosen": 236.24282836914062,
"logits/rejected": 264.61431884765625,
"logps/chosen": -1.1804494857788086,
"logps/rejected": -1.58183753490448,
"loss": 57.3731,
"nll_loss": 1.4926464557647705,
"rewards/accuracies": 0.71875,
"rewards/chosen": -0.5902247428894043,
"rewards/margins": 0.2006940096616745,
"rewards/rejected": -0.79091876745224,
"step": 275
},
{
"epoch": 2.644740740740741,
"grad_norm": 163.67861938476562,
"learning_rate": 1.8633852284264508e-07,
"log_odds_chosen": 0.37513643503189087,
"log_odds_ratio": -0.6290008425712585,
"logits/chosen": 242.14004516601562,
"logits/rejected": 243.6206817626953,
"logps/chosen": -1.1577939987182617,
"logps/rejected": -1.4585858583450317,
"loss": 56.5557,
"nll_loss": 1.4161715507507324,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -0.5788969993591309,
"rewards/margins": 0.15039584040641785,
"rewards/rejected": -0.7292929291725159,
"step": 280
},
{
"epoch": 2.6921481481481484,
"grad_norm": 131.90818786621094,
"learning_rate": 1.3736045660864034e-07,
"log_odds_chosen": 0.39001405239105225,
"log_odds_ratio": -0.6308606863021851,
"logits/chosen": 236.4365234375,
"logits/rejected": 251.0836181640625,
"logps/chosen": -1.2004508972167969,
"logps/rejected": -1.521728754043579,
"loss": 58.3533,
"nll_loss": 1.5017693042755127,
"rewards/accuracies": 0.612500011920929,
"rewards/chosen": -0.6002254486083984,
"rewards/margins": 0.1606389582157135,
"rewards/rejected": -0.7608643770217896,
"step": 285
},
{
"epoch": 2.7395555555555555,
"grad_norm": 123.62786865234375,
"learning_rate": 9.565939833279192e-08,
"log_odds_chosen": 0.4970114827156067,
"log_odds_ratio": -0.5855778455734253,
"logits/chosen": 245.94058227539062,
"logits/rejected": 250.78225708007812,
"logps/chosen": -1.196718454360962,
"logps/rejected": -1.5853480100631714,
"loss": 57.6325,
"nll_loss": 1.5269972085952759,
"rewards/accuracies": 0.6937500238418579,
"rewards/chosen": -0.598359227180481,
"rewards/margins": 0.19431476294994354,
"rewards/rejected": -0.7926740050315857,
"step": 290
},
{
"epoch": 2.786962962962963,
"grad_norm": 156.63650512695312,
"learning_rate": 6.136378865420872e-08,
"log_odds_chosen": 0.3856307566165924,
"log_odds_ratio": -0.6614529490470886,
"logits/chosen": 238.5367889404297,
"logits/rejected": 231.7664031982422,
"logps/chosen": -1.2782148122787476,
"logps/rejected": -1.5829877853393555,
"loss": 56.9746,
"nll_loss": 1.4864990711212158,
"rewards/accuracies": 0.65625,
"rewards/chosen": -0.6391074061393738,
"rewards/margins": 0.15238651633262634,
"rewards/rejected": -0.7914938926696777,
"step": 295
},
{
"epoch": 2.83437037037037,
"grad_norm": 150.2822723388672,
"learning_rate": 3.45792591853214e-08,
"log_odds_chosen": 0.5988565683364868,
"log_odds_ratio": -0.5730681419372559,
"logits/chosen": 223.4080810546875,
"logits/rejected": 250.70947265625,
"logps/chosen": -1.1857376098632812,
"logps/rejected": -1.6649852991104126,
"loss": 57.5247,
"nll_loss": 1.4728622436523438,
"rewards/accuracies": 0.6812499761581421,
"rewards/chosen": -0.5928688049316406,
"rewards/margins": 0.2396237850189209,
"rewards/rejected": -0.8324926495552063,
"step": 300
},
{
"epoch": 2.8817777777777778,
"grad_norm": 124.27201080322266,
"learning_rate": 1.538830716302092e-08,
"log_odds_chosen": 0.4612645208835602,
"log_odds_ratio": -0.6060934662818909,
"logits/chosen": 231.5652618408203,
"logits/rejected": 249.997802734375,
"logps/chosen": -1.2019062042236328,
"logps/rejected": -1.5464013814926147,
"loss": 55.1052,
"nll_loss": 1.423748254776001,
"rewards/accuracies": 0.675000011920929,
"rewards/chosen": -0.6009531021118164,
"rewards/margins": 0.17224761843681335,
"rewards/rejected": -0.7732006907463074,
"step": 305
},
{
"epoch": 2.9291851851851853,
"grad_norm": 174.55288696289062,
"learning_rate": 3.850041354441503e-09,
"log_odds_chosen": 0.38001030683517456,
"log_odds_ratio": -0.6417083144187927,
"logits/chosen": 238.8291473388672,
"logits/rejected": 261.79193115234375,
"logps/chosen": -1.2446626424789429,
"logps/rejected": -1.5207360982894897,
"loss": 57.5668,
"nll_loss": 1.507879376411438,
"rewards/accuracies": 0.65625,
"rewards/chosen": -0.6223313212394714,
"rewards/margins": 0.13803674280643463,
"rewards/rejected": -0.7603680491447449,
"step": 310
},
{
"epoch": 2.9765925925925925,
"grad_norm": 151.5247039794922,
"learning_rate": 0.0,
"log_odds_chosen": 0.4647287428379059,
"log_odds_ratio": -0.5936863422393799,
"logits/chosen": 231.6604461669922,
"logits/rejected": 250.37442016601562,
"logps/chosen": -1.172564148902893,
"logps/rejected": -1.534595251083374,
"loss": 55.8487,
"nll_loss": 1.4678246974945068,
"rewards/accuracies": 0.6625000238418579,
"rewards/chosen": -0.5862820744514465,
"rewards/margins": 0.18101553618907928,
"rewards/rejected": -0.767297625541687,
"step": 315
},
{
"epoch": 2.9765925925925925,
"step": 315,
"total_flos": 0.0,
"train_loss": 116.41342008076016,
"train_runtime": 8101.4659,
"train_samples_per_second": 2.5,
"train_steps_per_second": 0.039
}
],
"logging_steps": 5,
"max_steps": 315,
"num_input_tokens_seen": 0,
"num_train_epochs": 3,
"save_steps": 100000,
"stateful_callbacks": {
"TrainerControl": {
"args": {
"should_epoch_stop": false,
"should_evaluate": false,
"should_log": false,
"should_save": true,
"should_training_stop": true
},
"attributes": {}
}
},
"total_flos": 0.0,
"train_batch_size": 1,
"trial_name": null,
"trial_params": null
}