gemma-7b-borpo-noisy-5e-5-norm / trainer_state.json
silviasapora's picture
Model save
1b4d2ba 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": 875.1458129882812,
"learning_rate": 7.8125e-06,
"log_odds_chosen": 4.998103618621826,
"log_odds_ratio": -9.803796768188477,
"logits/chosen": 138.401611328125,
"logits/rejected": 153.0216064453125,
"logps/chosen": -20.76495361328125,
"logps/rejected": -25.763286590576172,
"loss": 171.0682,
"nll_loss": 8.368230819702148,
"rewards/accuracies": 0.53125,
"rewards/chosen": -10.382476806640625,
"rewards/margins": 2.4991683959960938,
"rewards/rejected": -12.881643295288086,
"step": 5
},
{
"epoch": 0.09481481481481481,
"grad_norm": 1327.8896484375,
"learning_rate": 1.5625e-05,
"log_odds_chosen": 2.35148024559021,
"log_odds_ratio": -7.3204779624938965,
"logits/chosen": 132.05416870117188,
"logits/rejected": 156.443603515625,
"logps/chosen": -15.675480842590332,
"logps/rejected": -18.02592658996582,
"loss": 245.0754,
"nll_loss": 7.408373832702637,
"rewards/accuracies": 0.543749988079071,
"rewards/chosen": -7.837740421295166,
"rewards/margins": 1.175222635269165,
"rewards/rejected": -9.01296329498291,
"step": 10
},
{
"epoch": 0.14222222222222222,
"grad_norm": 697.5897827148438,
"learning_rate": 2.34375e-05,
"log_odds_chosen": 4.538907051086426,
"log_odds_ratio": -8.128061294555664,
"logits/chosen": 115.15523529052734,
"logits/rejected": 142.7751922607422,
"logps/chosen": -20.854244232177734,
"logps/rejected": -25.39171028137207,
"loss": 260.4323,
"nll_loss": 8.780919075012207,
"rewards/accuracies": 0.5687500238418579,
"rewards/chosen": -10.427122116088867,
"rewards/margins": 2.268732786178589,
"rewards/rejected": -12.695855140686035,
"step": 15
},
{
"epoch": 0.18962962962962962,
"grad_norm": 1123.942626953125,
"learning_rate": 3.125e-05,
"log_odds_chosen": -0.19573011994361877,
"log_odds_ratio": -7.7527055740356445,
"logits/chosen": 145.3936309814453,
"logits/rejected": 150.71353149414062,
"logps/chosen": -17.822494506835938,
"logps/rejected": -17.628124237060547,
"loss": 155.0807,
"nll_loss": 8.136808395385742,
"rewards/accuracies": 0.543749988079071,
"rewards/chosen": -8.911247253417969,
"rewards/margins": -0.09718628227710724,
"rewards/rejected": -8.814062118530273,
"step": 20
},
{
"epoch": 0.23703703703703705,
"grad_norm": 1145.396484375,
"learning_rate": 3.90625e-05,
"log_odds_chosen": 4.295346260070801,
"log_odds_ratio": -7.432453155517578,
"logits/chosen": 151.52081298828125,
"logits/rejected": 160.0590362548828,
"logps/chosen": -17.441072463989258,
"logps/rejected": -21.736679077148438,
"loss": 370.5377,
"nll_loss": 6.786558628082275,
"rewards/accuracies": 0.5249999761581421,
"rewards/chosen": -8.720536231994629,
"rewards/margins": 2.147803783416748,
"rewards/rejected": -10.868339538574219,
"step": 25
},
{
"epoch": 0.28444444444444444,
"grad_norm": 2596.38818359375,
"learning_rate": 4.6875e-05,
"log_odds_chosen": 0.009631728753447533,
"log_odds_ratio": -5.7961225509643555,
"logits/chosen": 152.17312622070312,
"logits/rejected": 188.4014129638672,
"logps/chosen": -13.26770305633545,
"logps/rejected": -13.267865180969238,
"loss": 140.9719,
"nll_loss": 5.588016510009766,
"rewards/accuracies": 0.574999988079071,
"rewards/chosen": -6.633851528167725,
"rewards/margins": 8.130073547363281e-05,
"rewards/rejected": -6.633932590484619,
"step": 30
},
{
"epoch": 0.33185185185185184,
"grad_norm": 2872.860107421875,
"learning_rate": 4.998613757348784e-05,
"log_odds_chosen": -0.7113814353942871,
"log_odds_ratio": -2.4185237884521484,
"logits/chosen": 196.71951293945312,
"logits/rejected": 181.12623596191406,
"logps/chosen": -5.303072929382324,
"logps/rejected": -4.604456901550293,
"loss": 104.958,
"nll_loss": 2.8738203048706055,
"rewards/accuracies": 0.5062500238418579,
"rewards/chosen": -2.651536464691162,
"rewards/margins": -0.3493083119392395,
"rewards/rejected": -2.3022284507751465,
"step": 35
},
{
"epoch": 0.37925925925925924,
"grad_norm": 25351.75,
"learning_rate": 4.990147841143462e-05,
"log_odds_chosen": 1.3610541820526123,
"log_odds_ratio": -2.1230006217956543,
"logits/chosen": 193.37913513183594,
"logits/rejected": 204.16732788085938,
"logps/chosen": -7.698246002197266,
"logps/rejected": -9.058138847351074,
"loss": 204.6441,
"nll_loss": 5.2039594650268555,
"rewards/accuracies": 0.53125,
"rewards/chosen": -3.849123001098633,
"rewards/margins": 0.6799466013908386,
"rewards/rejected": -4.529069423675537,
"step": 40
},
{
"epoch": 0.4266666666666667,
"grad_norm": 7550.92822265625,
"learning_rate": 4.97401218720448e-05,
"log_odds_chosen": -0.7872304916381836,
"log_odds_ratio": -2.196615695953369,
"logits/chosen": 213.85107421875,
"logits/rejected": 207.1276092529297,
"logps/chosen": -5.1990180015563965,
"logps/rejected": -4.413318157196045,
"loss": 74.1465,
"nll_loss": 3.3599143028259277,
"rewards/accuracies": 0.4625000059604645,
"rewards/chosen": -2.5995090007781982,
"rewards/margins": -0.39284998178482056,
"rewards/rejected": -2.2066590785980225,
"step": 45
},
{
"epoch": 0.4740740740740741,
"grad_norm": 19450.84375,
"learning_rate": 4.9502564938797946e-05,
"log_odds_chosen": 0.3536006510257721,
"log_odds_ratio": -0.9575241208076477,
"logits/chosen": 228.44052124023438,
"logits/rejected": 228.40786743164062,
"logps/chosen": -2.896968126296997,
"logps/rejected": -3.246020555496216,
"loss": 153.1728,
"nll_loss": 2.854421615600586,
"rewards/accuracies": 0.5625,
"rewards/chosen": -1.4484840631484985,
"rewards/margins": 0.17452631890773773,
"rewards/rejected": -1.623010277748108,
"step": 50
},
{
"epoch": 0.5214814814814814,
"grad_norm": 1182.28564453125,
"learning_rate": 4.918953929490768e-05,
"log_odds_chosen": 0.05445149540901184,
"log_odds_ratio": -1.0194499492645264,
"logits/chosen": 261.8747863769531,
"logits/rejected": 269.243896484375,
"logps/chosen": -2.4033851623535156,
"logps/rejected": -2.4503068923950195,
"loss": 106.4384,
"nll_loss": 2.4143242835998535,
"rewards/accuracies": 0.5375000238418579,
"rewards/chosen": -1.2016925811767578,
"rewards/margins": 0.02346091903746128,
"rewards/rejected": -1.2251534461975098,
"step": 55
},
{
"epoch": 0.5688888888888889,
"grad_norm": 11634.9443359375,
"learning_rate": 4.88020090697132e-05,
"log_odds_chosen": 0.2782652974128723,
"log_odds_ratio": -0.8532482981681824,
"logits/chosen": 250.95071411132812,
"logits/rejected": 266.6951599121094,
"logps/chosen": -2.0876777172088623,
"logps/rejected": -2.3408493995666504,
"loss": 109.2007,
"nll_loss": 2.205104351043701,
"rewards/accuracies": 0.5625,
"rewards/chosen": -1.0438388586044312,
"rewards/margins": 0.1265856921672821,
"rewards/rejected": -1.1704246997833252,
"step": 60
},
{
"epoch": 0.6162962962962963,
"grad_norm": 2758.363525390625,
"learning_rate": 4.834116786912897e-05,
"log_odds_chosen": 0.15672044456005096,
"log_odds_ratio": -1.0671449899673462,
"logits/chosen": 279.3172607421875,
"logits/rejected": 281.0866394042969,
"logps/chosen": -2.3741352558135986,
"logps/rejected": -2.5186164379119873,
"loss": 131.6787,
"nll_loss": 2.280827760696411,
"rewards/accuracies": 0.4937500059604645,
"rewards/chosen": -1.1870676279067993,
"rewards/margins": 0.07224057614803314,
"rewards/rejected": -1.2593082189559937,
"step": 65
},
{
"epoch": 0.6637037037037037,
"grad_norm": 2126.38232421875,
"learning_rate": 4.7808435099299045e-05,
"log_odds_chosen": 0.7884730100631714,
"log_odds_ratio": -0.7412351369857788,
"logits/chosen": 259.4008483886719,
"logits/rejected": 279.88018798828125,
"logps/chosen": -2.333249092102051,
"logps/rejected": -3.116851329803467,
"loss": 105.9595,
"nll_loss": 2.3083605766296387,
"rewards/accuracies": 0.5375000238418579,
"rewards/chosen": -1.1666245460510254,
"rewards/margins": 0.39180102944374084,
"rewards/rejected": -1.5584256649017334,
"step": 70
},
{
"epoch": 0.7111111111111111,
"grad_norm": 1613.800048828125,
"learning_rate": 4.720545159477922e-05,
"log_odds_chosen": 0.15139682590961456,
"log_odds_ratio": -2.676135301589966,
"logits/chosen": 253.4937286376953,
"logits/rejected": 245.2483673095703,
"logps/chosen": -5.529114723205566,
"logps/rejected": -5.663898468017578,
"loss": 205.7061,
"nll_loss": 3.4207730293273926,
"rewards/accuracies": 0.612500011920929,
"rewards/chosen": -2.764557361602783,
"rewards/margins": 0.06739232689142227,
"rewards/rejected": -2.831949234008789,
"step": 75
},
{
"epoch": 0.7585185185185185,
"grad_norm": 1384.3203125,
"learning_rate": 4.653407456471222e-05,
"log_odds_chosen": 1.3567793369293213,
"log_odds_ratio": -1.5762678384780884,
"logits/chosen": 249.103759765625,
"logits/rejected": 261.83154296875,
"logps/chosen": -4.300660610198975,
"logps/rejected": -5.641573905944824,
"loss": 96.7149,
"nll_loss": 2.916186809539795,
"rewards/accuracies": 0.543749988079071,
"rewards/chosen": -2.1503303050994873,
"rewards/margins": 0.6704565286636353,
"rewards/rejected": -2.820786952972412,
"step": 80
},
{
"epoch": 0.8059259259259259,
"grad_norm": 6067.19921875,
"learning_rate": 4.579637187256222e-05,
"log_odds_chosen": 0.5072377920150757,
"log_odds_ratio": -1.1216537952423096,
"logits/chosen": 266.18145751953125,
"logits/rejected": 295.95086669921875,
"logps/chosen": -2.4558284282684326,
"logps/rejected": -2.9313106536865234,
"loss": 114.5914,
"nll_loss": 2.423276901245117,
"rewards/accuracies": 0.625,
"rewards/chosen": -1.2279142141342163,
"rewards/margins": 0.23774108290672302,
"rewards/rejected": -1.4656553268432617,
"step": 85
},
{
"epoch": 0.8533333333333334,
"grad_norm": 3012.88671875,
"learning_rate": 4.499461566702685e-05,
"log_odds_chosen": 0.59010249376297,
"log_odds_ratio": -1.1751501560211182,
"logits/chosen": 277.5395812988281,
"logits/rejected": 293.25543212890625,
"logps/chosen": -2.758129835128784,
"logps/rejected": -3.3149051666259766,
"loss": 100.0177,
"nll_loss": 2.5514206886291504,
"rewards/accuracies": 0.5874999761581421,
"rewards/chosen": -1.379064917564392,
"rewards/margins": 0.27838796377182007,
"rewards/rejected": -1.6574525833129883,
"step": 90
},
{
"epoch": 0.9007407407407407,
"grad_norm": 2200.108154296875,
"learning_rate": 4.413127538374411e-05,
"log_odds_chosen": 0.4700884222984314,
"log_odds_ratio": -0.7967466711997986,
"logits/chosen": 276.75836181640625,
"logits/rejected": 292.6000671386719,
"logps/chosen": -1.8207991123199463,
"logps/rejected": -2.2410740852355957,
"loss": 66.5128,
"nll_loss": 2.0072360038757324,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -0.9103995561599731,
"rewards/margins": 0.2101375162601471,
"rewards/rejected": -1.1205370426177979,
"step": 95
},
{
"epoch": 0.9481481481481482,
"grad_norm": 3112.162353515625,
"learning_rate": 4.320901013934887e-05,
"log_odds_chosen": 0.36486712098121643,
"log_odds_ratio": -0.9160555005073547,
"logits/chosen": 289.7991027832031,
"logits/rejected": 304.82513427734375,
"logps/chosen": -2.173243284225464,
"logps/rejected": -2.5329384803771973,
"loss": 62.2007,
"nll_loss": 2.255070209503174,
"rewards/accuracies": 0.5687500238418579,
"rewards/chosen": -1.086621642112732,
"rewards/margins": 0.17984741926193237,
"rewards/rejected": -1.2664692401885986,
"step": 100
},
{
"epoch": 0.9955555555555555,
"grad_norm": 1157.5224609375,
"learning_rate": 4.223066054130568e-05,
"log_odds_chosen": 0.3608945906162262,
"log_odds_ratio": -0.7305700182914734,
"logits/chosen": 281.0811462402344,
"logits/rejected": 300.54522705078125,
"logps/chosen": -1.7265421152114868,
"logps/rejected": -2.0366110801696777,
"loss": 50.5091,
"nll_loss": 1.927724838256836,
"rewards/accuracies": 0.6499999761581421,
"rewards/chosen": -0.8632710576057434,
"rewards/margins": 0.15503451228141785,
"rewards/rejected": -1.0183055400848389,
"step": 105
},
{
"epoch": 1.037925925925926,
"grad_norm": 1815.9822998046875,
"learning_rate": 4.1199239938743797e-05,
"log_odds_chosen": 0.05243406444787979,
"log_odds_ratio": -0.8580684065818787,
"logits/chosen": 291.1260070800781,
"logits/rejected": 301.667236328125,
"logps/chosen": -1.6703624725341797,
"logps/rejected": -1.733909010887146,
"loss": 55.2378,
"nll_loss": 1.869752049446106,
"rewards/accuracies": 0.5384615659713745,
"rewards/chosen": -0.8351812362670898,
"rewards/margins": 0.03177327290177345,
"rewards/rejected": -0.866954505443573,
"step": 110
},
{
"epoch": 1.0853333333333333,
"grad_norm": 473.7134094238281,
"learning_rate": 4.0117925141242174e-05,
"log_odds_chosen": 0.2678685784339905,
"log_odds_ratio": -0.8520981073379517,
"logits/chosen": 275.2568359375,
"logits/rejected": 300.43701171875,
"logps/chosen": -1.8224836587905884,
"logps/rejected": -2.0139191150665283,
"loss": 73.8236,
"nll_loss": 1.8877710103988647,
"rewards/accuracies": 0.637499988079071,
"rewards/chosen": -0.9112418293952942,
"rewards/margins": 0.09571774303913116,
"rewards/rejected": -1.0069595575332642,
"step": 115
},
{
"epoch": 1.1327407407407408,
"grad_norm": 652.1748046875,
"learning_rate": 3.899004663415084e-05,
"log_odds_chosen": 0.3419615924358368,
"log_odds_ratio": -0.8070743680000305,
"logits/chosen": 255.7921142578125,
"logits/rejected": 267.5791931152344,
"logps/chosen": -1.7789255380630493,
"logps/rejected": -2.0702507495880127,
"loss": 72.9068,
"nll_loss": 1.9794025421142578,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -0.8894627690315247,
"rewards/margins": 0.14566253125667572,
"rewards/rejected": -1.0351253747940063,
"step": 120
},
{
"epoch": 1.1801481481481482,
"grad_norm": 1498.9239501953125,
"learning_rate": 3.781907832058587e-05,
"log_odds_chosen": 0.06549356132745743,
"log_odds_ratio": -0.8704937100410461,
"logits/chosen": 256.5135803222656,
"logits/rejected": 259.24957275390625,
"logps/chosen": -1.743499755859375,
"logps/rejected": -1.771620750427246,
"loss": 67.5531,
"nll_loss": 1.8592092990875244,
"rewards/accuracies": 0.606249988079071,
"rewards/chosen": -0.8717498779296875,
"rewards/margins": 0.014060517773032188,
"rewards/rejected": -0.885810375213623,
"step": 125
},
{
"epoch": 1.2275555555555555,
"grad_norm": 785.5113525390625,
"learning_rate": 3.660862682169282e-05,
"log_odds_chosen": 0.061852775514125824,
"log_odds_ratio": -0.8480948209762573,
"logits/chosen": 251.6976318359375,
"logits/rejected": 261.1220397949219,
"logps/chosen": -1.5321893692016602,
"logps/rejected": -1.576385498046875,
"loss": 61.6414,
"nll_loss": 1.7627137899398804,
"rewards/accuracies": 0.5375000238418579,
"rewards/chosen": -0.7660946846008301,
"rewards/margins": 0.022097986191511154,
"rewards/rejected": -0.7881927490234375,
"step": 130
},
{
"epoch": 1.274962962962963,
"grad_norm": 1566.5196533203125,
"learning_rate": 3.5362420368134356e-05,
"log_odds_chosen": -0.06569625437259674,
"log_odds_ratio": -1.0464322566986084,
"logits/chosen": 257.01080322265625,
"logits/rejected": 264.09246826171875,
"logps/chosen": -1.8015447854995728,
"logps/rejected": -1.727919578552246,
"loss": 49.7802,
"nll_loss": 1.8332347869873047,
"rewards/accuracies": 0.5375000238418579,
"rewards/chosen": -0.9007723927497864,
"rewards/margins": -0.036812592297792435,
"rewards/rejected": -0.863959789276123,
"step": 135
},
{
"epoch": 1.3223703703703704,
"grad_norm": 3892.252685546875,
"learning_rate": 3.408429731701635e-05,
"log_odds_chosen": 0.12928451597690582,
"log_odds_ratio": -1.1143265962600708,
"logits/chosen": 249.24832153320312,
"logits/rejected": 242.64840698242188,
"logps/chosen": -2.36354923248291,
"logps/rejected": -2.504887342453003,
"loss": 68.5205,
"nll_loss": 2.1543807983398438,
"rewards/accuracies": 0.5375000238418579,
"rewards/chosen": -1.181774616241455,
"rewards/margins": 0.07066915184259415,
"rewards/rejected": -1.2524436712265015,
"step": 140
},
{
"epoch": 1.3697777777777778,
"grad_norm": 1954.7860107421875,
"learning_rate": 3.2778194329621104e-05,
"log_odds_chosen": 0.6030800342559814,
"log_odds_ratio": -0.880083441734314,
"logits/chosen": 238.984619140625,
"logits/rejected": 257.8157653808594,
"logps/chosen": -2.005361318588257,
"logps/rejected": -2.592895984649658,
"loss": 106.2088,
"nll_loss": 1.994653344154358,
"rewards/accuracies": 0.612500011920929,
"rewards/chosen": -1.0026806592941284,
"rewards/margins": 0.29376715421676636,
"rewards/rejected": -1.296447992324829,
"step": 145
},
{
"epoch": 1.417185185185185,
"grad_norm": 4879.1201171875,
"learning_rate": 3.144813424636031e-05,
"log_odds_chosen": -0.14236605167388916,
"log_odds_ratio": -1.0077848434448242,
"logits/chosen": 254.04440307617188,
"logits/rejected": 244.98947143554688,
"logps/chosen": -1.827314019203186,
"logps/rejected": -1.7160383462905884,
"loss": 76.4689,
"nll_loss": 1.9274555444717407,
"rewards/accuracies": 0.48124998807907104,
"rewards/chosen": -0.913657009601593,
"rewards/margins": -0.055637799203395844,
"rewards/rejected": -0.8580191731452942,
"step": 150
},
{
"epoch": 1.4645925925925927,
"grad_norm": 3663.07275390625,
"learning_rate": 3.0098213696293542e-05,
"log_odds_chosen": 0.31877607107162476,
"log_odds_ratio": -0.8575546145439148,
"logits/chosen": 244.96865844726562,
"logits/rejected": 271.86456298828125,
"logps/chosen": -1.880878210067749,
"logps/rejected": -2.1720147132873535,
"loss": 81.7164,
"nll_loss": 2.0283596515655518,
"rewards/accuracies": 0.5625,
"rewards/chosen": -0.9404391050338745,
"rewards/margins": 0.14556822180747986,
"rewards/rejected": -1.0860073566436768,
"step": 155
},
{
"epoch": 1.512,
"grad_norm": 2128.646240234375,
"learning_rate": 2.8732590479375165e-05,
"log_odds_chosen": 0.2958715856075287,
"log_odds_ratio": -0.7991440892219543,
"logits/chosen": 261.80657958984375,
"logits/rejected": 264.6015930175781,
"logps/chosen": -1.747931718826294,
"logps/rejected": -2.023869037628174,
"loss": 64.2607,
"nll_loss": 1.854034423828125,
"rewards/accuracies": 0.5687500238418579,
"rewards/chosen": -0.873965859413147,
"rewards/margins": 0.13796871900558472,
"rewards/rejected": -1.011934518814087,
"step": 160
},
{
"epoch": 1.5594074074074074,
"grad_norm": 1701.7451171875,
"learning_rate": 2.7355470760292956e-05,
"log_odds_chosen": 0.026621544733643532,
"log_odds_ratio": -0.913711428642273,
"logits/chosen": 275.054931640625,
"logits/rejected": 260.64874267578125,
"logps/chosen": -1.8813059329986572,
"logps/rejected": -1.8814128637313843,
"loss": 60.1293,
"nll_loss": 1.8933929204940796,
"rewards/accuracies": 0.550000011920929,
"rewards/chosen": -0.9406529664993286,
"rewards/margins": 5.351304935174994e-05,
"rewards/rejected": -0.9407064318656921,
"step": 165
},
{
"epoch": 1.6068148148148147,
"grad_norm": 892.0233154296875,
"learning_rate": 2.597109611334169e-05,
"log_odds_chosen": 0.10228142887353897,
"log_odds_ratio": -0.8465485572814941,
"logits/chosen": 270.9327087402344,
"logits/rejected": 275.8879699707031,
"logps/chosen": -1.8762061595916748,
"logps/rejected": -1.9606273174285889,
"loss": 59.1188,
"nll_loss": 1.9105665683746338,
"rewards/accuracies": 0.5375000238418579,
"rewards/chosen": -0.9381030797958374,
"rewards/margins": 0.042210519313812256,
"rewards/rejected": -0.9803136587142944,
"step": 170
},
{
"epoch": 1.6542222222222223,
"grad_norm": 363.839111328125,
"learning_rate": 2.458373045823404e-05,
"log_odds_chosen": 0.25977933406829834,
"log_odds_ratio": -0.8015204668045044,
"logits/chosen": 272.14178466796875,
"logits/rejected": 288.24725341796875,
"logps/chosen": -1.7361786365509033,
"logps/rejected": -1.966091513633728,
"loss": 63.3352,
"nll_loss": 1.8658736944198608,
"rewards/accuracies": 0.5874999761581421,
"rewards/chosen": -0.8680893182754517,
"rewards/margins": 0.11495651304721832,
"rewards/rejected": -0.983045756816864,
"step": 175
},
{
"epoch": 1.7016296296296296,
"grad_norm": 1070.2236328125,
"learning_rate": 2.3197646927086697e-05,
"log_odds_chosen": -0.2573428750038147,
"log_odds_ratio": -1.1848556995391846,
"logits/chosen": 278.9873352050781,
"logits/rejected": 272.8277587890625,
"logps/chosen": -2.0303432941436768,
"logps/rejected": -1.8005205392837524,
"loss": 60.8183,
"nll_loss": 1.9714701175689697,
"rewards/accuracies": 0.44999998807907104,
"rewards/chosen": -1.0151716470718384,
"rewards/margins": -0.11491117626428604,
"rewards/rejected": -0.9002602696418762,
"step": 180
},
{
"epoch": 1.749037037037037,
"grad_norm": 344.8516540527344,
"learning_rate": 2.1817114703032176e-05,
"log_odds_chosen": 0.2049332559108734,
"log_odds_ratio": -0.8525213003158569,
"logits/chosen": 267.901123046875,
"logits/rejected": 269.7948303222656,
"logps/chosen": -1.668211579322815,
"logps/rejected": -1.8583217859268188,
"loss": 65.7532,
"nll_loss": 1.8099273443222046,
"rewards/accuracies": 0.574999988079071,
"rewards/chosen": -0.8341057896614075,
"rewards/margins": 0.09505517780780792,
"rewards/rejected": -0.9291608929634094,
"step": 185
},
{
"epoch": 1.7964444444444445,
"grad_norm": 1921.921630859375,
"learning_rate": 2.0446385870993467e-05,
"log_odds_chosen": -0.010730123147368431,
"log_odds_ratio": -1.041703462600708,
"logits/chosen": 247.10665893554688,
"logits/rejected": 266.3430480957031,
"logps/chosen": -2.0603463649749756,
"logps/rejected": -2.0158324241638184,
"loss": 4.8149,
"nll_loss": 2.0877997875213623,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -1.0301731824874878,
"rewards/margins": -0.022256921976804733,
"rewards/rejected": -1.0079162120819092,
"step": 190
},
{
"epoch": 1.8438518518518519,
"grad_norm": 1207.0048828125,
"learning_rate": 1.9089682321121834e-05,
"log_odds_chosen": 0.04649388790130615,
"log_odds_ratio": -0.9928682446479797,
"logits/chosen": 263.4989929199219,
"logits/rejected": 280.4521789550781,
"logps/chosen": -1.9523179531097412,
"logps/rejected": -1.9815547466278076,
"loss": 37.5282,
"nll_loss": 2.034578800201416,
"rewards/accuracies": 0.5625,
"rewards/chosen": -0.9761589765548706,
"rewards/margins": 0.014618346467614174,
"rewards/rejected": -0.9907773733139038,
"step": 195
},
{
"epoch": 1.8912592592592592,
"grad_norm": 1364.85009765625,
"learning_rate": 1.775118274523545e-05,
"log_odds_chosen": -0.1880253702402115,
"log_odds_ratio": -1.0025999546051025,
"logits/chosen": 259.7791748046875,
"logits/rejected": 271.92303466796875,
"logps/chosen": -1.6487674713134766,
"logps/rejected": -1.4585245847702026,
"loss": 2.3249,
"nll_loss": 1.7344061136245728,
"rewards/accuracies": 0.53125,
"rewards/chosen": -0.8243837356567383,
"rewards/margins": -0.09512142091989517,
"rewards/rejected": -0.7292622923851013,
"step": 200
},
{
"epoch": 1.9386666666666668,
"grad_norm": 423.42108154296875,
"learning_rate": 1.643500976631037e-05,
"log_odds_chosen": 0.18183811008930206,
"log_odds_ratio": -1.011243462562561,
"logits/chosen": 255.23471069335938,
"logits/rejected": 276.84356689453125,
"logps/chosen": -1.7864488363265991,
"logps/rejected": -1.945235252380371,
"loss": 53.8824,
"nll_loss": 1.8585844039916992,
"rewards/accuracies": 0.5375000238418579,
"rewards/chosen": -0.8932244181632996,
"rewards/margins": 0.07939319312572479,
"rewards/rejected": -0.9726176261901855,
"step": 205
},
{
"epoch": 1.986074074074074,
"grad_norm": 217.96726989746094,
"learning_rate": 1.514521724066537e-05,
"log_odds_chosen": -0.03663633018732071,
"log_odds_ratio": -0.8743799924850464,
"logits/chosen": 272.13055419921875,
"logits/rejected": 262.0260925292969,
"logps/chosen": -1.5016670227050781,
"logps/rejected": -1.47659170627594,
"loss": 53.2491,
"nll_loss": 1.7205402851104736,
"rewards/accuracies": 0.4937500059604645,
"rewards/chosen": -0.7508335113525391,
"rewards/margins": -0.0125376982614398,
"rewards/rejected": -0.73829585313797,
"step": 210
},
{
"epoch": 2.0284444444444443,
"grad_norm": 1592.707275390625,
"learning_rate": 1.3885777771950348e-05,
"log_odds_chosen": -0.2431742548942566,
"log_odds_ratio": -1.065582275390625,
"logits/chosen": 264.0666809082031,
"logits/rejected": 270.9798278808594,
"logps/chosen": -1.8248804807662964,
"logps/rejected": -1.5742560625076294,
"loss": 23.2293,
"nll_loss": 1.8352609872817993,
"rewards/accuracies": 0.5524475574493408,
"rewards/chosen": -0.9124402403831482,
"rewards/margins": -0.1253122240304947,
"rewards/rejected": -0.7871280312538147,
"step": 215
},
{
"epoch": 2.075851851851852,
"grad_norm": 2064.20458984375,
"learning_rate": 1.2660570475395683e-05,
"log_odds_chosen": 0.07270346581935883,
"log_odds_ratio": -0.8794776201248169,
"logits/chosen": 257.7349548339844,
"logits/rejected": 261.220703125,
"logps/chosen": -1.5441248416900635,
"logps/rejected": -1.5986078977584839,
"loss": 46.8863,
"nll_loss": 1.7352149486541748,
"rewards/accuracies": 0.5687500238418579,
"rewards/chosen": -0.7720624208450317,
"rewards/margins": 0.027241546660661697,
"rewards/rejected": -0.7993039488792419,
"step": 220
},
{
"epoch": 2.1232592592592594,
"grad_norm": 718.3240966796875,
"learning_rate": 1.1473369030008974e-05,
"log_odds_chosen": -0.3160693943500519,
"log_odds_ratio": -1.118138074874878,
"logits/chosen": 255.883056640625,
"logits/rejected": 249.1800537109375,
"logps/chosen": -1.7974827289581299,
"logps/rejected": -1.5016992092132568,
"loss": -6.6175,
"nll_loss": 1.8773807287216187,
"rewards/accuracies": 0.4937500059604645,
"rewards/chosen": -0.8987413644790649,
"rewards/margins": -0.14789175987243652,
"rewards/rejected": -0.7508496046066284,
"step": 225
},
{
"epoch": 2.1706666666666665,
"grad_norm": 2849.17138671875,
"learning_rate": 1.0327830055518842e-05,
"log_odds_chosen": 0.05444896221160889,
"log_odds_ratio": -1.0051008462905884,
"logits/chosen": 249.77755737304688,
"logits/rejected": 262.1796569824219,
"logps/chosen": -1.957166314125061,
"logps/rejected": -1.9977023601531982,
"loss": 32.8461,
"nll_loss": 1.9164683818817139,
"rewards/accuracies": 0.5625,
"rewards/chosen": -0.9785831570625305,
"rewards/margins": 0.020268145948648453,
"rewards/rejected": -0.9988511800765991,
"step": 230
},
{
"epoch": 2.218074074074074,
"grad_norm": 772.2219848632812,
"learning_rate": 9.227481849865235e-06,
"log_odds_chosen": -0.5832428932189941,
"log_odds_ratio": -1.5142714977264404,
"logits/chosen": 238.1352996826172,
"logits/rejected": 243.27197265625,
"logps/chosen": -2.4328596591949463,
"logps/rejected": -1.8463695049285889,
"loss": -10.1247,
"nll_loss": 2.1351406574249268,
"rewards/accuracies": 0.518750011920929,
"rewards/chosen": -1.2164298295974731,
"rewards/margins": -0.2932450473308563,
"rewards/rejected": -0.9231847524642944,
"step": 235
},
{
"epoch": 2.2654814814814817,
"grad_norm": 206.4999237060547,
"learning_rate": 8.175713521924978e-06,
"log_odds_chosen": -0.37373510003089905,
"log_odds_ratio": -1.2141059637069702,
"logits/chosen": 247.3380126953125,
"logits/rejected": 246.8715057373047,
"logps/chosen": -2.0405170917510986,
"logps/rejected": -1.6850570440292358,
"loss": -7.9574,
"nll_loss": 1.9059865474700928,
"rewards/accuracies": 0.512499988079071,
"rewards/chosen": -1.0202585458755493,
"rewards/margins": -0.17772991955280304,
"rewards/rejected": -0.8425285220146179,
"step": 240
},
{
"epoch": 2.3128888888888888,
"grad_norm": 5002.2255859375,
"learning_rate": 7.1757645529443665e-06,
"log_odds_chosen": -0.3410179018974304,
"log_odds_ratio": -1.1818878650665283,
"logits/chosen": 239.45654296875,
"logits/rejected": 262.42694091796875,
"logps/chosen": -2.027069568634033,
"logps/rejected": -1.6753078699111938,
"loss": -1.4536,
"nll_loss": 2.0037341117858887,
"rewards/accuracies": 0.543749988079071,
"rewards/chosen": -1.0135347843170166,
"rewards/margins": -0.17588095366954803,
"rewards/rejected": -0.8376539349555969,
"step": 245
},
{
"epoch": 2.3602962962962963,
"grad_norm": 3950.781982421875,
"learning_rate": 6.230714818829733e-06,
"log_odds_chosen": 0.018829286098480225,
"log_odds_ratio": -0.8809181451797485,
"logits/chosen": 248.4897918701172,
"logits/rejected": 255.39193725585938,
"logps/chosen": -1.5735387802124023,
"logps/rejected": -1.5727895498275757,
"loss": 29.9793,
"nll_loss": 1.7044947147369385,
"rewards/accuracies": 0.5625,
"rewards/chosen": -0.7867693901062012,
"rewards/margins": -0.00037460625753737986,
"rewards/rejected": -0.7863947749137878,
"step": 250
},
{
"epoch": 2.407703703703704,
"grad_norm": 896.8823852539062,
"learning_rate": 5.343475104027743e-06,
"log_odds_chosen": -0.20623116195201874,
"log_odds_ratio": -1.1634149551391602,
"logits/chosen": 247.5481414794922,
"logits/rejected": 250.3152618408203,
"logps/chosen": -1.9348275661468506,
"logps/rejected": -1.7173500061035156,
"loss": 10.0473,
"nll_loss": 1.8512252569198608,
"rewards/accuracies": 0.6000000238418579,
"rewards/chosen": -0.9674137830734253,
"rewards/margins": -0.10873879492282867,
"rewards/rejected": -0.8586750030517578,
"step": 255
},
{
"epoch": 2.455111111111111,
"grad_norm": 3805.9501953125,
"learning_rate": 4.516778136213037e-06,
"log_odds_chosen": -0.09736748039722443,
"log_odds_ratio": -1.0045772790908813,
"logits/chosen": 240.8005828857422,
"logits/rejected": 259.3901062011719,
"logps/chosen": -1.9838569164276123,
"logps/rejected": -1.8580623865127563,
"loss": 28.1764,
"nll_loss": 1.8320090770721436,
"rewards/accuracies": 0.5687500238418579,
"rewards/chosen": -0.9919284582138062,
"rewards/margins": -0.0628972202539444,
"rewards/rejected": -0.9290311932563782,
"step": 260
},
{
"epoch": 2.5025185185185186,
"grad_norm": 839.794677734375,
"learning_rate": 3.7531701693965554e-06,
"log_odds_chosen": -0.46628624200820923,
"log_odds_ratio": -1.336814522743225,
"logits/chosen": 253.2608184814453,
"logits/rejected": 247.44625854492188,
"logps/chosen": -1.995568037033081,
"logps/rejected": -1.50832998752594,
"loss": -22.3793,
"nll_loss": 1.8523412942886353,
"rewards/accuracies": 0.550000011920929,
"rewards/chosen": -0.9977840185165405,
"rewards/margins": -0.24361905455589294,
"rewards/rejected": -0.75416499376297,
"step": 265
},
{
"epoch": 2.549925925925926,
"grad_norm": 415.7859802246094,
"learning_rate": 3.055003141378948e-06,
"log_odds_chosen": 0.025583285838365555,
"log_odds_ratio": -1.0144554376602173,
"logits/chosen": 242.5679168701172,
"logits/rejected": 247.4667205810547,
"logps/chosen": -2.184016466140747,
"logps/rejected": -2.171065092086792,
"loss": 31.7258,
"nll_loss": 1.8770112991333008,
"rewards/accuracies": 0.625,
"rewards/chosen": -1.0920082330703735,
"rewards/margins": -0.006475582718849182,
"rewards/rejected": -1.085532546043396,
"step": 270
},
{
"epoch": 2.5973333333333333,
"grad_norm": 871.19482421875,
"learning_rate": 2.424427429704365e-06,
"log_odds_chosen": -0.06566087901592255,
"log_odds_ratio": -1.05461847782135,
"logits/chosen": 243.1831512451172,
"logits/rejected": 265.54376220703125,
"logps/chosen": -1.7891054153442383,
"logps/rejected": -1.672284483909607,
"loss": 12.4619,
"nll_loss": 1.8154491186141968,
"rewards/accuracies": 0.6187499761581421,
"rewards/chosen": -0.8945527076721191,
"rewards/margins": -0.05841045826673508,
"rewards/rejected": -0.8361422419548035,
"step": 275
},
{
"epoch": 2.644740740740741,
"grad_norm": 3279.919921875,
"learning_rate": 1.8633852284264508e-06,
"log_odds_chosen": -0.36494511365890503,
"log_odds_ratio": -1.200988531112671,
"logits/chosen": 246.86233520507812,
"logits/rejected": 247.6204071044922,
"logps/chosen": -1.8717445135116577,
"logps/rejected": -1.5146552324295044,
"loss": 14.1496,
"nll_loss": 1.7873508930206299,
"rewards/accuracies": 0.518750011920929,
"rewards/chosen": -0.9358722567558289,
"rewards/margins": -0.17854467034339905,
"rewards/rejected": -0.7573276162147522,
"step": 280
},
{
"epoch": 2.6921481481481484,
"grad_norm": 591.1192626953125,
"learning_rate": 1.3736045660864034e-06,
"log_odds_chosen": -0.10209669917821884,
"log_odds_ratio": -1.0389295816421509,
"logits/chosen": 244.7912139892578,
"logits/rejected": 257.46722412109375,
"logps/chosen": -1.8741945028305054,
"logps/rejected": -1.7793128490447998,
"loss": 13.954,
"nll_loss": 1.839137315750122,
"rewards/accuracies": 0.512499988079071,
"rewards/chosen": -0.9370972514152527,
"rewards/margins": -0.04744075611233711,
"rewards/rejected": -0.8896564245223999,
"step": 285
},
{
"epoch": 2.7395555555555555,
"grad_norm": 3014.538818359375,
"learning_rate": 9.565939833279192e-07,
"log_odds_chosen": -0.32551804184913635,
"log_odds_ratio": -1.3097738027572632,
"logits/chosen": 253.1048126220703,
"logits/rejected": 256.01220703125,
"logps/chosen": -2.233328104019165,
"logps/rejected": -1.8837789297103882,
"loss": -51.2977,
"nll_loss": 2.0668323040008545,
"rewards/accuracies": 0.581250011920929,
"rewards/chosen": -1.1166640520095825,
"rewards/margins": -0.1747743934392929,
"rewards/rejected": -0.9418894648551941,
"step": 290
},
{
"epoch": 2.786962962962963,
"grad_norm": 1583.771728515625,
"learning_rate": 6.136378865420872e-07,
"log_odds_chosen": -0.180761456489563,
"log_odds_ratio": -1.1848331689834595,
"logits/chosen": 246.4217071533203,
"logits/rejected": 238.6134796142578,
"logps/chosen": -2.0827651023864746,
"logps/rejected": -1.8982934951782227,
"loss": -4.1222,
"nll_loss": 1.8400661945343018,
"rewards/accuracies": 0.5249999761581421,
"rewards/chosen": -1.0413825511932373,
"rewards/margins": -0.09223591536283493,
"rewards/rejected": -0.9491467475891113,
"step": 295
},
{
"epoch": 2.83437037037037,
"grad_norm": 370.90655517578125,
"learning_rate": 3.45792591853214e-07,
"log_odds_chosen": 0.15687787532806396,
"log_odds_ratio": -0.8642757534980774,
"logits/chosen": 231.45712280273438,
"logits/rejected": 254.4613494873047,
"logps/chosen": -1.6515134572982788,
"logps/rejected": -1.7699718475341797,
"loss": 36.0028,
"nll_loss": 1.6890325546264648,
"rewards/accuracies": 0.5874999761581421,
"rewards/chosen": -0.8257567286491394,
"rewards/margins": 0.05922911688685417,
"rewards/rejected": -0.8849859237670898,
"step": 300
},
{
"epoch": 2.8817777777777778,
"grad_norm": 1475.66357421875,
"learning_rate": 1.538830716302092e-07,
"log_odds_chosen": 0.0742514431476593,
"log_odds_ratio": -0.9600369334220886,
"logits/chosen": 239.23818969726562,
"logits/rejected": 255.12649536132812,
"logps/chosen": -1.6762956380844116,
"logps/rejected": -1.7098366022109985,
"loss": 16.3917,
"nll_loss": 1.6855227947235107,
"rewards/accuracies": 0.625,
"rewards/chosen": -0.8381478190422058,
"rewards/margins": 0.01677042618393898,
"rewards/rejected": -0.8549183011054993,
"step": 305
},
{
"epoch": 2.9291851851851853,
"grad_norm": 4500.62158203125,
"learning_rate": 3.8500413544415025e-08,
"log_odds_chosen": 0.004546451382339001,
"log_odds_ratio": -1.0055034160614014,
"logits/chosen": 246.24240112304688,
"logits/rejected": 267.38336181640625,
"logps/chosen": -1.752664566040039,
"logps/rejected": -1.7212101221084595,
"loss": 12.0693,
"nll_loss": 1.76467764377594,
"rewards/accuracies": 0.5874999761581421,
"rewards/chosen": -0.8763322830200195,
"rewards/margins": -0.015727439895272255,
"rewards/rejected": -0.8606050610542297,
"step": 310
},
{
"epoch": 2.9765925925925925,
"grad_norm": 349.34295654296875,
"learning_rate": 0.0,
"log_odds_chosen": -0.17005635797977448,
"log_odds_ratio": -1.134798526763916,
"logits/chosen": 242.64279174804688,
"logits/rejected": 257.0933532714844,
"logps/chosen": -1.8541597127914429,
"logps/rejected": -1.6562083959579468,
"loss": -6.7425,
"nll_loss": 1.7939285039901733,
"rewards/accuracies": 0.574999988079071,
"rewards/chosen": -0.9270798563957214,
"rewards/margins": -0.09897565841674805,
"rewards/rejected": -0.8281041979789734,
"step": 315
},
{
"epoch": 2.9765925925925925,
"step": 315,
"total_flos": 0.0,
"train_loss": 70.88754168313648,
"train_runtime": 8139.5297,
"train_samples_per_second": 2.488,
"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
}