math-phi-1-5-FULL-Arithmetic-4k / trainer_state.json
UltimoUno's picture
Uploaded checkpoint-4000
90435d5 verified
raw
history blame contribute delete
No virus
59.3 kB
{
"best_metric": null,
"best_model_checkpoint": null,
"epoch": 0.2,
"eval_steps": 2000,
"global_step": 4000,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 0.0,
"grad_norm": 20.375,
"learning_rate": 3.0000000000000004e-07,
"loss": 1.7815,
"step": 10
},
{
"epoch": 0.0,
"grad_norm": 19.125,
"learning_rate": 6.000000000000001e-07,
"loss": 1.84,
"step": 20
},
{
"epoch": 0.0,
"grad_norm": 17.0,
"learning_rate": 9e-07,
"loss": 2.098,
"step": 30
},
{
"epoch": 0.0,
"grad_norm": 20.125,
"learning_rate": 1.2000000000000002e-06,
"loss": 1.67,
"step": 40
},
{
"epoch": 0.0,
"grad_norm": 18.0,
"learning_rate": 1.5e-06,
"loss": 1.8452,
"step": 50
},
{
"epoch": 0.0,
"grad_norm": 19.25,
"learning_rate": 1.8e-06,
"loss": 2.1664,
"step": 60
},
{
"epoch": 0.0,
"grad_norm": 17.25,
"learning_rate": 2.1000000000000002e-06,
"loss": 1.6483,
"step": 70
},
{
"epoch": 0.0,
"grad_norm": 18.625,
"learning_rate": 2.4000000000000003e-06,
"loss": 1.9371,
"step": 80
},
{
"epoch": 0.0,
"grad_norm": 16.75,
"learning_rate": 2.7e-06,
"loss": 1.8967,
"step": 90
},
{
"epoch": 0.01,
"grad_norm": 14.3125,
"learning_rate": 3e-06,
"loss": 1.8326,
"step": 100
},
{
"epoch": 0.01,
"grad_norm": 15.9375,
"learning_rate": 3.3e-06,
"loss": 1.9072,
"step": 110
},
{
"epoch": 0.01,
"grad_norm": 21.5,
"learning_rate": 3.6e-06,
"loss": 1.9562,
"step": 120
},
{
"epoch": 0.01,
"grad_norm": 21.75,
"learning_rate": 3.9e-06,
"loss": 1.6976,
"step": 130
},
{
"epoch": 0.01,
"grad_norm": 16.0,
"learning_rate": 4.2000000000000004e-06,
"loss": 2.2788,
"step": 140
},
{
"epoch": 0.01,
"grad_norm": 17.625,
"learning_rate": 4.5e-06,
"loss": 1.8362,
"step": 150
},
{
"epoch": 0.01,
"grad_norm": 22.5,
"learning_rate": 4.800000000000001e-06,
"loss": 1.8372,
"step": 160
},
{
"epoch": 0.01,
"grad_norm": 16.75,
"learning_rate": 5.1e-06,
"loss": 1.9134,
"step": 170
},
{
"epoch": 0.01,
"grad_norm": 21.875,
"learning_rate": 5.4e-06,
"loss": 1.8973,
"step": 180
},
{
"epoch": 0.01,
"grad_norm": 15.8125,
"learning_rate": 5.7000000000000005e-06,
"loss": 1.6868,
"step": 190
},
{
"epoch": 0.01,
"grad_norm": 19.75,
"learning_rate": 6e-06,
"loss": 1.6079,
"step": 200
},
{
"epoch": 0.01,
"grad_norm": 19.5,
"learning_rate": 6.3e-06,
"loss": 1.4638,
"step": 210
},
{
"epoch": 0.01,
"grad_norm": 18.625,
"learning_rate": 6.6e-06,
"loss": 1.8714,
"step": 220
},
{
"epoch": 0.01,
"grad_norm": 24.25,
"learning_rate": 6.900000000000001e-06,
"loss": 1.9379,
"step": 230
},
{
"epoch": 0.01,
"grad_norm": 13.0,
"learning_rate": 7.2e-06,
"loss": 1.364,
"step": 240
},
{
"epoch": 0.01,
"grad_norm": 17.75,
"learning_rate": 7.5e-06,
"loss": 1.8867,
"step": 250
},
{
"epoch": 0.01,
"grad_norm": 14.0,
"learning_rate": 7.8e-06,
"loss": 1.9215,
"step": 260
},
{
"epoch": 0.01,
"grad_norm": 23.5,
"learning_rate": 8.1e-06,
"loss": 1.8065,
"step": 270
},
{
"epoch": 0.01,
"grad_norm": 17.25,
"learning_rate": 8.400000000000001e-06,
"loss": 1.6864,
"step": 280
},
{
"epoch": 0.01,
"grad_norm": 25.25,
"learning_rate": 8.7e-06,
"loss": 1.6924,
"step": 290
},
{
"epoch": 0.01,
"grad_norm": 25.0,
"learning_rate": 9e-06,
"loss": 1.7671,
"step": 300
},
{
"epoch": 0.02,
"grad_norm": 18.375,
"learning_rate": 9.3e-06,
"loss": 1.8781,
"step": 310
},
{
"epoch": 0.02,
"grad_norm": 20.5,
"learning_rate": 9.600000000000001e-06,
"loss": 1.454,
"step": 320
},
{
"epoch": 0.02,
"grad_norm": 14.875,
"learning_rate": 9.9e-06,
"loss": 1.2016,
"step": 330
},
{
"epoch": 0.02,
"grad_norm": 23.625,
"learning_rate": 1.02e-05,
"loss": 1.6703,
"step": 340
},
{
"epoch": 0.02,
"grad_norm": 26.125,
"learning_rate": 1.05e-05,
"loss": 1.3712,
"step": 350
},
{
"epoch": 0.02,
"grad_norm": 18.375,
"learning_rate": 1.08e-05,
"loss": 1.3751,
"step": 360
},
{
"epoch": 0.02,
"grad_norm": 17.625,
"learning_rate": 1.11e-05,
"loss": 1.7888,
"step": 370
},
{
"epoch": 0.02,
"grad_norm": 34.25,
"learning_rate": 1.1400000000000001e-05,
"loss": 1.5242,
"step": 380
},
{
"epoch": 0.02,
"grad_norm": 22.5,
"learning_rate": 1.1700000000000001e-05,
"loss": 1.5371,
"step": 390
},
{
"epoch": 0.02,
"grad_norm": 31.375,
"learning_rate": 1.2e-05,
"loss": 1.3818,
"step": 400
},
{
"epoch": 0.02,
"grad_norm": 13.375,
"learning_rate": 1.2299999999999999e-05,
"loss": 1.9301,
"step": 410
},
{
"epoch": 0.02,
"grad_norm": 17.25,
"learning_rate": 1.26e-05,
"loss": 1.7813,
"step": 420
},
{
"epoch": 0.02,
"grad_norm": 22.125,
"learning_rate": 1.29e-05,
"loss": 1.6075,
"step": 430
},
{
"epoch": 0.02,
"grad_norm": 25.125,
"learning_rate": 1.32e-05,
"loss": 1.9157,
"step": 440
},
{
"epoch": 0.02,
"grad_norm": 16.375,
"learning_rate": 1.3500000000000001e-05,
"loss": 1.6057,
"step": 450
},
{
"epoch": 0.02,
"grad_norm": 19.0,
"learning_rate": 1.3800000000000002e-05,
"loss": 1.6342,
"step": 460
},
{
"epoch": 0.02,
"grad_norm": 13.625,
"learning_rate": 1.4099999999999999e-05,
"loss": 1.4012,
"step": 470
},
{
"epoch": 0.02,
"grad_norm": 14.0625,
"learning_rate": 1.44e-05,
"loss": 1.853,
"step": 480
},
{
"epoch": 0.02,
"grad_norm": 21.375,
"learning_rate": 1.47e-05,
"loss": 1.6546,
"step": 490
},
{
"epoch": 0.03,
"grad_norm": 20.0,
"learning_rate": 1.5e-05,
"loss": 1.3404,
"step": 500
},
{
"epoch": 0.03,
"grad_norm": 18.25,
"learning_rate": 1.4984210526315789e-05,
"loss": 1.5801,
"step": 510
},
{
"epoch": 0.03,
"grad_norm": 27.5,
"learning_rate": 1.496842105263158e-05,
"loss": 1.5567,
"step": 520
},
{
"epoch": 0.03,
"grad_norm": 28.25,
"learning_rate": 1.4952631578947368e-05,
"loss": 1.7417,
"step": 530
},
{
"epoch": 0.03,
"grad_norm": 27.0,
"learning_rate": 1.4936842105263158e-05,
"loss": 1.6283,
"step": 540
},
{
"epoch": 0.03,
"grad_norm": 25.5,
"learning_rate": 1.4921052631578947e-05,
"loss": 1.8534,
"step": 550
},
{
"epoch": 0.03,
"grad_norm": 11.625,
"learning_rate": 1.4905263157894737e-05,
"loss": 1.668,
"step": 560
},
{
"epoch": 0.03,
"grad_norm": 39.75,
"learning_rate": 1.4889473684210526e-05,
"loss": 1.4711,
"step": 570
},
{
"epoch": 0.03,
"grad_norm": 20.125,
"learning_rate": 1.4873684210526315e-05,
"loss": 1.4484,
"step": 580
},
{
"epoch": 0.03,
"grad_norm": 15.3125,
"learning_rate": 1.4857894736842107e-05,
"loss": 1.6434,
"step": 590
},
{
"epoch": 0.03,
"grad_norm": 21.375,
"learning_rate": 1.4842105263157895e-05,
"loss": 1.8766,
"step": 600
},
{
"epoch": 0.03,
"grad_norm": 22.125,
"learning_rate": 1.4826315789473686e-05,
"loss": 1.329,
"step": 610
},
{
"epoch": 0.03,
"grad_norm": 22.5,
"learning_rate": 1.4810526315789474e-05,
"loss": 1.5956,
"step": 620
},
{
"epoch": 0.03,
"grad_norm": 9.125,
"learning_rate": 1.4794736842105265e-05,
"loss": 1.5795,
"step": 630
},
{
"epoch": 0.03,
"grad_norm": 15.25,
"learning_rate": 1.4778947368421053e-05,
"loss": 1.7082,
"step": 640
},
{
"epoch": 0.03,
"grad_norm": 25.0,
"learning_rate": 1.4763157894736842e-05,
"loss": 1.7773,
"step": 650
},
{
"epoch": 0.03,
"grad_norm": 25.5,
"learning_rate": 1.4747368421052632e-05,
"loss": 1.3858,
"step": 660
},
{
"epoch": 0.03,
"grad_norm": 19.75,
"learning_rate": 1.4731578947368421e-05,
"loss": 1.6927,
"step": 670
},
{
"epoch": 0.03,
"grad_norm": 10.75,
"learning_rate": 1.4715789473684211e-05,
"loss": 1.5281,
"step": 680
},
{
"epoch": 0.03,
"grad_norm": 30.125,
"learning_rate": 1.47e-05,
"loss": 1.3842,
"step": 690
},
{
"epoch": 0.04,
"grad_norm": 41.5,
"learning_rate": 1.468421052631579e-05,
"loss": 1.7584,
"step": 700
},
{
"epoch": 0.04,
"grad_norm": 18.875,
"learning_rate": 1.4668421052631579e-05,
"loss": 1.5485,
"step": 710
},
{
"epoch": 0.04,
"grad_norm": 35.25,
"learning_rate": 1.4652631578947367e-05,
"loss": 1.61,
"step": 720
},
{
"epoch": 0.04,
"grad_norm": 19.25,
"learning_rate": 1.4636842105263158e-05,
"loss": 1.6709,
"step": 730
},
{
"epoch": 0.04,
"grad_norm": 17.5,
"learning_rate": 1.4621052631578946e-05,
"loss": 1.4464,
"step": 740
},
{
"epoch": 0.04,
"grad_norm": 18.625,
"learning_rate": 1.4605263157894737e-05,
"loss": 1.5036,
"step": 750
},
{
"epoch": 0.04,
"grad_norm": 22.25,
"learning_rate": 1.4589473684210527e-05,
"loss": 1.6983,
"step": 760
},
{
"epoch": 0.04,
"grad_norm": 32.5,
"learning_rate": 1.4573684210526317e-05,
"loss": 1.4551,
"step": 770
},
{
"epoch": 0.04,
"grad_norm": 18.625,
"learning_rate": 1.4557894736842106e-05,
"loss": 1.2903,
"step": 780
},
{
"epoch": 0.04,
"grad_norm": 25.875,
"learning_rate": 1.4542105263157895e-05,
"loss": 1.5937,
"step": 790
},
{
"epoch": 0.04,
"grad_norm": 24.125,
"learning_rate": 1.4526315789473685e-05,
"loss": 1.5994,
"step": 800
},
{
"epoch": 0.04,
"grad_norm": 26.375,
"learning_rate": 1.4510526315789474e-05,
"loss": 1.4018,
"step": 810
},
{
"epoch": 0.04,
"grad_norm": 18.75,
"learning_rate": 1.4494736842105264e-05,
"loss": 1.1078,
"step": 820
},
{
"epoch": 0.04,
"grad_norm": 19.25,
"learning_rate": 1.4478947368421053e-05,
"loss": 1.5669,
"step": 830
},
{
"epoch": 0.04,
"grad_norm": 16.75,
"learning_rate": 1.4463157894736843e-05,
"loss": 1.325,
"step": 840
},
{
"epoch": 0.04,
"grad_norm": 18.5,
"learning_rate": 1.4447368421052632e-05,
"loss": 1.3645,
"step": 850
},
{
"epoch": 0.04,
"grad_norm": 26.125,
"learning_rate": 1.443157894736842e-05,
"loss": 1.5002,
"step": 860
},
{
"epoch": 0.04,
"grad_norm": 22.875,
"learning_rate": 1.441578947368421e-05,
"loss": 1.3608,
"step": 870
},
{
"epoch": 0.04,
"grad_norm": 20.625,
"learning_rate": 1.44e-05,
"loss": 1.5648,
"step": 880
},
{
"epoch": 0.04,
"grad_norm": 32.0,
"learning_rate": 1.438421052631579e-05,
"loss": 1.1969,
"step": 890
},
{
"epoch": 0.04,
"grad_norm": 18.25,
"learning_rate": 1.4368421052631578e-05,
"loss": 1.5523,
"step": 900
},
{
"epoch": 0.05,
"grad_norm": 27.5,
"learning_rate": 1.4352631578947369e-05,
"loss": 1.5062,
"step": 910
},
{
"epoch": 0.05,
"grad_norm": 14.75,
"learning_rate": 1.4336842105263159e-05,
"loss": 1.5052,
"step": 920
},
{
"epoch": 0.05,
"grad_norm": 12.8125,
"learning_rate": 1.4321052631578948e-05,
"loss": 1.6696,
"step": 930
},
{
"epoch": 0.05,
"grad_norm": 18.0,
"learning_rate": 1.4305263157894738e-05,
"loss": 1.2593,
"step": 940
},
{
"epoch": 0.05,
"grad_norm": 22.0,
"learning_rate": 1.4289473684210527e-05,
"loss": 0.9601,
"step": 950
},
{
"epoch": 0.05,
"grad_norm": 17.5,
"learning_rate": 1.4273684210526317e-05,
"loss": 1.3436,
"step": 960
},
{
"epoch": 0.05,
"grad_norm": 29.25,
"learning_rate": 1.4257894736842106e-05,
"loss": 1.9493,
"step": 970
},
{
"epoch": 0.05,
"grad_norm": 14.375,
"learning_rate": 1.4242105263157896e-05,
"loss": 1.3317,
"step": 980
},
{
"epoch": 0.05,
"grad_norm": 12.625,
"learning_rate": 1.4226315789473685e-05,
"loss": 1.263,
"step": 990
},
{
"epoch": 0.05,
"grad_norm": 25.625,
"learning_rate": 1.4210526315789473e-05,
"loss": 1.4426,
"step": 1000
},
{
"epoch": 0.05,
"grad_norm": 27.75,
"learning_rate": 1.4194736842105264e-05,
"loss": 1.738,
"step": 1010
},
{
"epoch": 0.05,
"grad_norm": 20.875,
"learning_rate": 1.4178947368421052e-05,
"loss": 1.474,
"step": 1020
},
{
"epoch": 0.05,
"grad_norm": 23.25,
"learning_rate": 1.4163157894736843e-05,
"loss": 1.8161,
"step": 1030
},
{
"epoch": 0.05,
"grad_norm": 18.25,
"learning_rate": 1.4147368421052631e-05,
"loss": 1.2116,
"step": 1040
},
{
"epoch": 0.05,
"grad_norm": 25.25,
"learning_rate": 1.4131578947368422e-05,
"loss": 1.7062,
"step": 1050
},
{
"epoch": 0.05,
"grad_norm": 21.125,
"learning_rate": 1.411578947368421e-05,
"loss": 1.5641,
"step": 1060
},
{
"epoch": 0.05,
"grad_norm": 15.625,
"learning_rate": 1.4099999999999999e-05,
"loss": 1.5193,
"step": 1070
},
{
"epoch": 0.05,
"grad_norm": 20.75,
"learning_rate": 1.408421052631579e-05,
"loss": 1.4775,
"step": 1080
},
{
"epoch": 0.05,
"grad_norm": 16.375,
"learning_rate": 1.406842105263158e-05,
"loss": 1.4353,
"step": 1090
},
{
"epoch": 0.06,
"grad_norm": 29.375,
"learning_rate": 1.405263157894737e-05,
"loss": 1.6741,
"step": 1100
},
{
"epoch": 0.06,
"grad_norm": 36.5,
"learning_rate": 1.4036842105263158e-05,
"loss": 1.4158,
"step": 1110
},
{
"epoch": 0.06,
"grad_norm": 20.5,
"learning_rate": 1.4021052631578949e-05,
"loss": 1.2708,
"step": 1120
},
{
"epoch": 0.06,
"grad_norm": 19.25,
"learning_rate": 1.4005263157894737e-05,
"loss": 1.1376,
"step": 1130
},
{
"epoch": 0.06,
"grad_norm": 18.5,
"learning_rate": 1.3989473684210526e-05,
"loss": 1.3171,
"step": 1140
},
{
"epoch": 0.06,
"grad_norm": 21.375,
"learning_rate": 1.3973684210526316e-05,
"loss": 1.5621,
"step": 1150
},
{
"epoch": 0.06,
"grad_norm": 19.5,
"learning_rate": 1.3957894736842105e-05,
"loss": 1.8743,
"step": 1160
},
{
"epoch": 0.06,
"grad_norm": 13.4375,
"learning_rate": 1.3942105263157895e-05,
"loss": 1.5132,
"step": 1170
},
{
"epoch": 0.06,
"grad_norm": 22.5,
"learning_rate": 1.3926315789473684e-05,
"loss": 1.6475,
"step": 1180
},
{
"epoch": 0.06,
"grad_norm": 13.25,
"learning_rate": 1.3910526315789474e-05,
"loss": 1.4806,
"step": 1190
},
{
"epoch": 0.06,
"grad_norm": 19.0,
"learning_rate": 1.3894736842105263e-05,
"loss": 1.5761,
"step": 1200
},
{
"epoch": 0.06,
"grad_norm": 17.125,
"learning_rate": 1.3878947368421052e-05,
"loss": 1.88,
"step": 1210
},
{
"epoch": 0.06,
"grad_norm": 19.0,
"learning_rate": 1.3863157894736842e-05,
"loss": 1.3399,
"step": 1220
},
{
"epoch": 0.06,
"grad_norm": 21.25,
"learning_rate": 1.384736842105263e-05,
"loss": 1.6421,
"step": 1230
},
{
"epoch": 0.06,
"grad_norm": 19.625,
"learning_rate": 1.3831578947368421e-05,
"loss": 1.7212,
"step": 1240
},
{
"epoch": 0.06,
"grad_norm": 17.75,
"learning_rate": 1.3815789473684211e-05,
"loss": 1.6537,
"step": 1250
},
{
"epoch": 0.06,
"grad_norm": 13.5625,
"learning_rate": 1.3800000000000002e-05,
"loss": 1.6197,
"step": 1260
},
{
"epoch": 0.06,
"grad_norm": 18.875,
"learning_rate": 1.378421052631579e-05,
"loss": 1.5681,
"step": 1270
},
{
"epoch": 0.06,
"grad_norm": 40.75,
"learning_rate": 1.3768421052631579e-05,
"loss": 1.3511,
"step": 1280
},
{
"epoch": 0.06,
"grad_norm": 13.0,
"learning_rate": 1.375263157894737e-05,
"loss": 1.3007,
"step": 1290
},
{
"epoch": 0.07,
"grad_norm": 10.9375,
"learning_rate": 1.3736842105263158e-05,
"loss": 1.18,
"step": 1300
},
{
"epoch": 0.07,
"grad_norm": 19.125,
"learning_rate": 1.3721052631578948e-05,
"loss": 1.6356,
"step": 1310
},
{
"epoch": 0.07,
"grad_norm": 16.375,
"learning_rate": 1.3705263157894737e-05,
"loss": 1.4107,
"step": 1320
},
{
"epoch": 0.07,
"grad_norm": 26.0,
"learning_rate": 1.3689473684210527e-05,
"loss": 1.3306,
"step": 1330
},
{
"epoch": 0.07,
"grad_norm": 22.5,
"learning_rate": 1.3673684210526316e-05,
"loss": 1.5359,
"step": 1340
},
{
"epoch": 0.07,
"grad_norm": 20.0,
"learning_rate": 1.3657894736842106e-05,
"loss": 1.404,
"step": 1350
},
{
"epoch": 0.07,
"grad_norm": 19.125,
"learning_rate": 1.3642105263157895e-05,
"loss": 1.2873,
"step": 1360
},
{
"epoch": 0.07,
"grad_norm": 11.1875,
"learning_rate": 1.3626315789473684e-05,
"loss": 1.3952,
"step": 1370
},
{
"epoch": 0.07,
"grad_norm": 22.0,
"learning_rate": 1.3610526315789474e-05,
"loss": 1.4608,
"step": 1380
},
{
"epoch": 0.07,
"grad_norm": 20.875,
"learning_rate": 1.3594736842105263e-05,
"loss": 1.5216,
"step": 1390
},
{
"epoch": 0.07,
"grad_norm": 18.5,
"learning_rate": 1.3578947368421053e-05,
"loss": 1.4667,
"step": 1400
},
{
"epoch": 0.07,
"grad_norm": 31.75,
"learning_rate": 1.3563157894736842e-05,
"loss": 1.6675,
"step": 1410
},
{
"epoch": 0.07,
"grad_norm": 22.125,
"learning_rate": 1.3547368421052634e-05,
"loss": 1.4568,
"step": 1420
},
{
"epoch": 0.07,
"grad_norm": 15.75,
"learning_rate": 1.3531578947368422e-05,
"loss": 1.4352,
"step": 1430
},
{
"epoch": 0.07,
"grad_norm": 19.75,
"learning_rate": 1.3515789473684211e-05,
"loss": 1.4362,
"step": 1440
},
{
"epoch": 0.07,
"grad_norm": 29.5,
"learning_rate": 1.3500000000000001e-05,
"loss": 1.2622,
"step": 1450
},
{
"epoch": 0.07,
"grad_norm": 13.75,
"learning_rate": 1.348421052631579e-05,
"loss": 1.4709,
"step": 1460
},
{
"epoch": 0.07,
"grad_norm": 18.25,
"learning_rate": 1.346842105263158e-05,
"loss": 1.3638,
"step": 1470
},
{
"epoch": 0.07,
"grad_norm": 22.375,
"learning_rate": 1.3452631578947369e-05,
"loss": 1.7496,
"step": 1480
},
{
"epoch": 0.07,
"grad_norm": 24.625,
"learning_rate": 1.343684210526316e-05,
"loss": 1.7741,
"step": 1490
},
{
"epoch": 0.07,
"grad_norm": 19.0,
"learning_rate": 1.3421052631578948e-05,
"loss": 1.3499,
"step": 1500
},
{
"epoch": 0.08,
"grad_norm": 31.625,
"learning_rate": 1.3405263157894736e-05,
"loss": 1.3528,
"step": 1510
},
{
"epoch": 0.08,
"grad_norm": 25.75,
"learning_rate": 1.3389473684210527e-05,
"loss": 1.4893,
"step": 1520
},
{
"epoch": 0.08,
"grad_norm": 12.3125,
"learning_rate": 1.3373684210526315e-05,
"loss": 1.4222,
"step": 1530
},
{
"epoch": 0.08,
"grad_norm": 23.0,
"learning_rate": 1.3357894736842106e-05,
"loss": 1.7875,
"step": 1540
},
{
"epoch": 0.08,
"grad_norm": 30.375,
"learning_rate": 1.3342105263157894e-05,
"loss": 1.426,
"step": 1550
},
{
"epoch": 0.08,
"grad_norm": 14.25,
"learning_rate": 1.3326315789473685e-05,
"loss": 1.8679,
"step": 1560
},
{
"epoch": 0.08,
"grad_norm": 19.875,
"learning_rate": 1.3310526315789473e-05,
"loss": 1.4155,
"step": 1570
},
{
"epoch": 0.08,
"grad_norm": 29.125,
"learning_rate": 1.3294736842105262e-05,
"loss": 1.7515,
"step": 1580
},
{
"epoch": 0.08,
"grad_norm": 18.125,
"learning_rate": 1.3278947368421054e-05,
"loss": 1.4494,
"step": 1590
},
{
"epoch": 0.08,
"grad_norm": 18.75,
"learning_rate": 1.3263157894736843e-05,
"loss": 1.4159,
"step": 1600
},
{
"epoch": 0.08,
"grad_norm": 24.75,
"learning_rate": 1.3247368421052633e-05,
"loss": 1.5582,
"step": 1610
},
{
"epoch": 0.08,
"grad_norm": 24.25,
"learning_rate": 1.3231578947368422e-05,
"loss": 1.5761,
"step": 1620
},
{
"epoch": 0.08,
"grad_norm": 30.375,
"learning_rate": 1.3215789473684212e-05,
"loss": 1.5033,
"step": 1630
},
{
"epoch": 0.08,
"grad_norm": 16.75,
"learning_rate": 1.32e-05,
"loss": 1.4209,
"step": 1640
},
{
"epoch": 0.08,
"grad_norm": 22.25,
"learning_rate": 1.318421052631579e-05,
"loss": 1.5761,
"step": 1650
},
{
"epoch": 0.08,
"grad_norm": 25.0,
"learning_rate": 1.316842105263158e-05,
"loss": 1.4146,
"step": 1660
},
{
"epoch": 0.08,
"grad_norm": 20.375,
"learning_rate": 1.3152631578947368e-05,
"loss": 1.2064,
"step": 1670
},
{
"epoch": 0.08,
"grad_norm": 25.75,
"learning_rate": 1.3136842105263159e-05,
"loss": 1.1254,
"step": 1680
},
{
"epoch": 0.08,
"grad_norm": 20.0,
"learning_rate": 1.3121052631578947e-05,
"loss": 1.5665,
"step": 1690
},
{
"epoch": 0.09,
"grad_norm": 23.625,
"learning_rate": 1.3105263157894738e-05,
"loss": 1.5582,
"step": 1700
},
{
"epoch": 0.09,
"grad_norm": 22.875,
"learning_rate": 1.3089473684210526e-05,
"loss": 1.2198,
"step": 1710
},
{
"epoch": 0.09,
"grad_norm": 15.875,
"learning_rate": 1.3073684210526315e-05,
"loss": 1.4875,
"step": 1720
},
{
"epoch": 0.09,
"grad_norm": 22.25,
"learning_rate": 1.3057894736842105e-05,
"loss": 1.3077,
"step": 1730
},
{
"epoch": 0.09,
"grad_norm": 13.125,
"learning_rate": 1.3042105263157894e-05,
"loss": 1.571,
"step": 1740
},
{
"epoch": 0.09,
"grad_norm": 16.875,
"learning_rate": 1.3026315789473684e-05,
"loss": 1.2261,
"step": 1750
},
{
"epoch": 0.09,
"grad_norm": 12.0,
"learning_rate": 1.3010526315789475e-05,
"loss": 1.1795,
"step": 1760
},
{
"epoch": 0.09,
"grad_norm": 23.0,
"learning_rate": 1.2994736842105265e-05,
"loss": 1.2311,
"step": 1770
},
{
"epoch": 0.09,
"grad_norm": 10.4375,
"learning_rate": 1.2978947368421054e-05,
"loss": 1.5966,
"step": 1780
},
{
"epoch": 0.09,
"grad_norm": 18.75,
"learning_rate": 1.2963157894736842e-05,
"loss": 1.5122,
"step": 1790
},
{
"epoch": 0.09,
"grad_norm": 10.5,
"learning_rate": 1.2947368421052633e-05,
"loss": 1.2532,
"step": 1800
},
{
"epoch": 0.09,
"grad_norm": 9.6875,
"learning_rate": 1.2931578947368421e-05,
"loss": 1.3394,
"step": 1810
},
{
"epoch": 0.09,
"grad_norm": 19.375,
"learning_rate": 1.2915789473684212e-05,
"loss": 1.6067,
"step": 1820
},
{
"epoch": 0.09,
"grad_norm": 23.0,
"learning_rate": 1.29e-05,
"loss": 1.4977,
"step": 1830
},
{
"epoch": 0.09,
"grad_norm": 16.375,
"learning_rate": 1.288421052631579e-05,
"loss": 1.4339,
"step": 1840
},
{
"epoch": 0.09,
"grad_norm": 19.625,
"learning_rate": 1.2868421052631579e-05,
"loss": 1.4789,
"step": 1850
},
{
"epoch": 0.09,
"grad_norm": 11.125,
"learning_rate": 1.2852631578947368e-05,
"loss": 1.3857,
"step": 1860
},
{
"epoch": 0.09,
"grad_norm": 9.3125,
"learning_rate": 1.2836842105263158e-05,
"loss": 1.4344,
"step": 1870
},
{
"epoch": 0.09,
"grad_norm": 21.5,
"learning_rate": 1.2821052631578947e-05,
"loss": 1.416,
"step": 1880
},
{
"epoch": 0.09,
"grad_norm": 23.875,
"learning_rate": 1.2805263157894737e-05,
"loss": 1.4628,
"step": 1890
},
{
"epoch": 0.1,
"grad_norm": 16.5,
"learning_rate": 1.2789473684210526e-05,
"loss": 1.3098,
"step": 1900
},
{
"epoch": 0.1,
"grad_norm": 21.375,
"learning_rate": 1.2773684210526316e-05,
"loss": 1.4423,
"step": 1910
},
{
"epoch": 0.1,
"grad_norm": 18.625,
"learning_rate": 1.2757894736842106e-05,
"loss": 1.1756,
"step": 1920
},
{
"epoch": 0.1,
"grad_norm": 24.625,
"learning_rate": 1.2742105263157895e-05,
"loss": 1.3788,
"step": 1930
},
{
"epoch": 0.1,
"grad_norm": 14.1875,
"learning_rate": 1.2726315789473685e-05,
"loss": 1.2287,
"step": 1940
},
{
"epoch": 0.1,
"grad_norm": 24.25,
"learning_rate": 1.2710526315789474e-05,
"loss": 1.1927,
"step": 1950
},
{
"epoch": 0.1,
"grad_norm": 14.625,
"learning_rate": 1.2694736842105264e-05,
"loss": 1.5478,
"step": 1960
},
{
"epoch": 0.1,
"grad_norm": 11.4375,
"learning_rate": 1.2678947368421053e-05,
"loss": 1.2426,
"step": 1970
},
{
"epoch": 0.1,
"grad_norm": 30.25,
"learning_rate": 1.2663157894736843e-05,
"loss": 1.1653,
"step": 1980
},
{
"epoch": 0.1,
"grad_norm": 21.875,
"learning_rate": 1.2647368421052632e-05,
"loss": 1.3693,
"step": 1990
},
{
"epoch": 0.1,
"grad_norm": 14.3125,
"learning_rate": 1.263157894736842e-05,
"loss": 1.7071,
"step": 2000
},
{
"epoch": 0.1,
"eval_loss": 1.4587359428405762,
"eval_runtime": 31.2193,
"eval_samples_per_second": 32.032,
"eval_steps_per_second": 32.032,
"step": 2000
},
{
"epoch": 0.1,
"grad_norm": 12.25,
"learning_rate": 1.2615789473684211e-05,
"loss": 1.1877,
"step": 2010
},
{
"epoch": 0.1,
"grad_norm": 33.25,
"learning_rate": 1.26e-05,
"loss": 1.6576,
"step": 2020
},
{
"epoch": 0.1,
"grad_norm": 17.0,
"learning_rate": 1.258421052631579e-05,
"loss": 1.3037,
"step": 2030
},
{
"epoch": 0.1,
"grad_norm": 17.625,
"learning_rate": 1.2568421052631579e-05,
"loss": 1.2634,
"step": 2040
},
{
"epoch": 0.1,
"grad_norm": 18.5,
"learning_rate": 1.2552631578947369e-05,
"loss": 1.5936,
"step": 2050
},
{
"epoch": 0.1,
"grad_norm": 23.625,
"learning_rate": 1.2536842105263158e-05,
"loss": 1.1655,
"step": 2060
},
{
"epoch": 0.1,
"grad_norm": 12.625,
"learning_rate": 1.2521052631578946e-05,
"loss": 1.3446,
"step": 2070
},
{
"epoch": 0.1,
"grad_norm": 7.34375,
"learning_rate": 1.2505263157894737e-05,
"loss": 1.5295,
"step": 2080
},
{
"epoch": 0.1,
"grad_norm": 12.4375,
"learning_rate": 1.2489473684210527e-05,
"loss": 1.5361,
"step": 2090
},
{
"epoch": 0.1,
"grad_norm": 15.9375,
"learning_rate": 1.2473684210526317e-05,
"loss": 1.3638,
"step": 2100
},
{
"epoch": 0.11,
"grad_norm": 15.4375,
"learning_rate": 1.2457894736842106e-05,
"loss": 1.472,
"step": 2110
},
{
"epoch": 0.11,
"grad_norm": 22.0,
"learning_rate": 1.2442105263157896e-05,
"loss": 1.0827,
"step": 2120
},
{
"epoch": 0.11,
"grad_norm": 12.0625,
"learning_rate": 1.2426315789473685e-05,
"loss": 1.2359,
"step": 2130
},
{
"epoch": 0.11,
"grad_norm": 12.75,
"learning_rate": 1.2410526315789474e-05,
"loss": 1.4345,
"step": 2140
},
{
"epoch": 0.11,
"grad_norm": 19.5,
"learning_rate": 1.2394736842105264e-05,
"loss": 1.5083,
"step": 2150
},
{
"epoch": 0.11,
"grad_norm": 29.125,
"learning_rate": 1.2378947368421053e-05,
"loss": 1.2703,
"step": 2160
},
{
"epoch": 0.11,
"grad_norm": 19.0,
"learning_rate": 1.2363157894736843e-05,
"loss": 1.2485,
"step": 2170
},
{
"epoch": 0.11,
"grad_norm": 30.0,
"learning_rate": 1.2347368421052631e-05,
"loss": 0.992,
"step": 2180
},
{
"epoch": 0.11,
"grad_norm": 27.625,
"learning_rate": 1.2331578947368422e-05,
"loss": 1.5069,
"step": 2190
},
{
"epoch": 0.11,
"grad_norm": 23.375,
"learning_rate": 1.231578947368421e-05,
"loss": 1.5471,
"step": 2200
},
{
"epoch": 0.11,
"grad_norm": 18.5,
"learning_rate": 1.2299999999999999e-05,
"loss": 1.3682,
"step": 2210
},
{
"epoch": 0.11,
"grad_norm": 17.125,
"learning_rate": 1.228421052631579e-05,
"loss": 1.5041,
"step": 2220
},
{
"epoch": 0.11,
"grad_norm": 15.875,
"learning_rate": 1.2268421052631578e-05,
"loss": 1.5487,
"step": 2230
},
{
"epoch": 0.11,
"grad_norm": 29.75,
"learning_rate": 1.2252631578947368e-05,
"loss": 1.2019,
"step": 2240
},
{
"epoch": 0.11,
"grad_norm": 12.625,
"learning_rate": 1.2236842105263159e-05,
"loss": 1.4014,
"step": 2250
},
{
"epoch": 0.11,
"grad_norm": 16.875,
"learning_rate": 1.2221052631578949e-05,
"loss": 1.3086,
"step": 2260
},
{
"epoch": 0.11,
"grad_norm": 15.5,
"learning_rate": 1.2205263157894738e-05,
"loss": 1.6796,
"step": 2270
},
{
"epoch": 0.11,
"grad_norm": 18.75,
"learning_rate": 1.2189473684210526e-05,
"loss": 1.2801,
"step": 2280
},
{
"epoch": 0.11,
"grad_norm": 9.75,
"learning_rate": 1.2173684210526317e-05,
"loss": 1.1822,
"step": 2290
},
{
"epoch": 0.12,
"grad_norm": 27.75,
"learning_rate": 1.2157894736842105e-05,
"loss": 1.428,
"step": 2300
},
{
"epoch": 0.12,
"grad_norm": 21.125,
"learning_rate": 1.2142105263157896e-05,
"loss": 1.4445,
"step": 2310
},
{
"epoch": 0.12,
"grad_norm": 20.5,
"learning_rate": 1.2126315789473684e-05,
"loss": 1.745,
"step": 2320
},
{
"epoch": 0.12,
"grad_norm": 14.3125,
"learning_rate": 1.2110526315789475e-05,
"loss": 1.8124,
"step": 2330
},
{
"epoch": 0.12,
"grad_norm": 31.0,
"learning_rate": 1.2094736842105263e-05,
"loss": 1.3533,
"step": 2340
},
{
"epoch": 0.12,
"grad_norm": 11.5,
"learning_rate": 1.2078947368421052e-05,
"loss": 1.4255,
"step": 2350
},
{
"epoch": 0.12,
"grad_norm": 17.5,
"learning_rate": 1.2063157894736842e-05,
"loss": 1.3926,
"step": 2360
},
{
"epoch": 0.12,
"grad_norm": 13.75,
"learning_rate": 1.2047368421052631e-05,
"loss": 1.295,
"step": 2370
},
{
"epoch": 0.12,
"grad_norm": 16.375,
"learning_rate": 1.2031578947368421e-05,
"loss": 1.0575,
"step": 2380
},
{
"epoch": 0.12,
"grad_norm": 30.5,
"learning_rate": 1.201578947368421e-05,
"loss": 1.4961,
"step": 2390
},
{
"epoch": 0.12,
"grad_norm": 17.5,
"learning_rate": 1.2e-05,
"loss": 1.3838,
"step": 2400
},
{
"epoch": 0.12,
"grad_norm": 16.75,
"learning_rate": 1.1984210526315789e-05,
"loss": 1.5833,
"step": 2410
},
{
"epoch": 0.12,
"grad_norm": 22.125,
"learning_rate": 1.196842105263158e-05,
"loss": 1.2771,
"step": 2420
},
{
"epoch": 0.12,
"grad_norm": 19.125,
"learning_rate": 1.195263157894737e-05,
"loss": 1.6264,
"step": 2430
},
{
"epoch": 0.12,
"grad_norm": 19.5,
"learning_rate": 1.1936842105263158e-05,
"loss": 1.4304,
"step": 2440
},
{
"epoch": 0.12,
"grad_norm": 11.1875,
"learning_rate": 1.1921052631578949e-05,
"loss": 1.1868,
"step": 2450
},
{
"epoch": 0.12,
"grad_norm": 39.25,
"learning_rate": 1.1905263157894737e-05,
"loss": 1.4066,
"step": 2460
},
{
"epoch": 0.12,
"grad_norm": 24.5,
"learning_rate": 1.1889473684210528e-05,
"loss": 1.3723,
"step": 2470
},
{
"epoch": 0.12,
"grad_norm": 21.75,
"learning_rate": 1.1873684210526316e-05,
"loss": 1.507,
"step": 2480
},
{
"epoch": 0.12,
"grad_norm": 22.625,
"learning_rate": 1.1857894736842105e-05,
"loss": 1.4822,
"step": 2490
},
{
"epoch": 0.12,
"grad_norm": 23.25,
"learning_rate": 1.1842105263157895e-05,
"loss": 1.3431,
"step": 2500
},
{
"epoch": 0.13,
"grad_norm": 14.25,
"learning_rate": 1.1826315789473684e-05,
"loss": 1.438,
"step": 2510
},
{
"epoch": 0.13,
"grad_norm": 38.0,
"learning_rate": 1.1810526315789474e-05,
"loss": 1.3837,
"step": 2520
},
{
"epoch": 0.13,
"grad_norm": 17.625,
"learning_rate": 1.1794736842105263e-05,
"loss": 1.3997,
"step": 2530
},
{
"epoch": 0.13,
"grad_norm": 13.375,
"learning_rate": 1.1778947368421053e-05,
"loss": 1.5068,
"step": 2540
},
{
"epoch": 0.13,
"grad_norm": 26.75,
"learning_rate": 1.1763157894736842e-05,
"loss": 1.5514,
"step": 2550
},
{
"epoch": 0.13,
"grad_norm": 23.375,
"learning_rate": 1.174736842105263e-05,
"loss": 1.4679,
"step": 2560
},
{
"epoch": 0.13,
"grad_norm": 12.6875,
"learning_rate": 1.173157894736842e-05,
"loss": 1.3298,
"step": 2570
},
{
"epoch": 0.13,
"grad_norm": 16.875,
"learning_rate": 1.171578947368421e-05,
"loss": 1.2123,
"step": 2580
},
{
"epoch": 0.13,
"grad_norm": 21.625,
"learning_rate": 1.1700000000000001e-05,
"loss": 1.3493,
"step": 2590
},
{
"epoch": 0.13,
"grad_norm": 25.75,
"learning_rate": 1.168421052631579e-05,
"loss": 1.3984,
"step": 2600
},
{
"epoch": 0.13,
"grad_norm": 20.5,
"learning_rate": 1.166842105263158e-05,
"loss": 1.435,
"step": 2610
},
{
"epoch": 0.13,
"grad_norm": 23.0,
"learning_rate": 1.1652631578947369e-05,
"loss": 1.273,
"step": 2620
},
{
"epoch": 0.13,
"grad_norm": 15.75,
"learning_rate": 1.1636842105263158e-05,
"loss": 1.2547,
"step": 2630
},
{
"epoch": 0.13,
"grad_norm": 7.71875,
"learning_rate": 1.1621052631578948e-05,
"loss": 1.1227,
"step": 2640
},
{
"epoch": 0.13,
"grad_norm": 29.125,
"learning_rate": 1.1605263157894737e-05,
"loss": 1.3045,
"step": 2650
},
{
"epoch": 0.13,
"grad_norm": 20.25,
"learning_rate": 1.1589473684210527e-05,
"loss": 1.4858,
"step": 2660
},
{
"epoch": 0.13,
"grad_norm": 14.4375,
"learning_rate": 1.1573684210526316e-05,
"loss": 1.1192,
"step": 2670
},
{
"epoch": 0.13,
"grad_norm": 22.875,
"learning_rate": 1.1557894736842106e-05,
"loss": 1.4049,
"step": 2680
},
{
"epoch": 0.13,
"grad_norm": 14.25,
"learning_rate": 1.1542105263157895e-05,
"loss": 1.3724,
"step": 2690
},
{
"epoch": 0.14,
"grad_norm": 24.25,
"learning_rate": 1.1526315789473683e-05,
"loss": 1.3053,
"step": 2700
},
{
"epoch": 0.14,
"grad_norm": 23.0,
"learning_rate": 1.1510526315789474e-05,
"loss": 1.3986,
"step": 2710
},
{
"epoch": 0.14,
"grad_norm": 20.0,
"learning_rate": 1.1494736842105262e-05,
"loss": 1.5444,
"step": 2720
},
{
"epoch": 0.14,
"grad_norm": 16.25,
"learning_rate": 1.1478947368421053e-05,
"loss": 1.6835,
"step": 2730
},
{
"epoch": 0.14,
"grad_norm": 19.375,
"learning_rate": 1.1463157894736841e-05,
"loss": 1.2312,
"step": 2740
},
{
"epoch": 0.14,
"grad_norm": 13.0625,
"learning_rate": 1.1447368421052632e-05,
"loss": 1.0698,
"step": 2750
},
{
"epoch": 0.14,
"grad_norm": 19.875,
"learning_rate": 1.1431578947368422e-05,
"loss": 1.4871,
"step": 2760
},
{
"epoch": 0.14,
"grad_norm": 10.875,
"learning_rate": 1.141578947368421e-05,
"loss": 1.257,
"step": 2770
},
{
"epoch": 0.14,
"grad_norm": 11.8125,
"learning_rate": 1.1400000000000001e-05,
"loss": 1.5711,
"step": 2780
},
{
"epoch": 0.14,
"grad_norm": 22.125,
"learning_rate": 1.138421052631579e-05,
"loss": 1.592,
"step": 2790
},
{
"epoch": 0.14,
"grad_norm": 25.5,
"learning_rate": 1.136842105263158e-05,
"loss": 0.9352,
"step": 2800
},
{
"epoch": 0.14,
"grad_norm": 17.625,
"learning_rate": 1.1352631578947369e-05,
"loss": 1.3347,
"step": 2810
},
{
"epoch": 0.14,
"grad_norm": 26.75,
"learning_rate": 1.1336842105263159e-05,
"loss": 1.0488,
"step": 2820
},
{
"epoch": 0.14,
"grad_norm": 18.75,
"learning_rate": 1.1321052631578948e-05,
"loss": 1.7698,
"step": 2830
},
{
"epoch": 0.14,
"grad_norm": 26.0,
"learning_rate": 1.1305263157894736e-05,
"loss": 1.2135,
"step": 2840
},
{
"epoch": 0.14,
"grad_norm": 19.125,
"learning_rate": 1.1289473684210527e-05,
"loss": 1.5619,
"step": 2850
},
{
"epoch": 0.14,
"grad_norm": 20.125,
"learning_rate": 1.1273684210526315e-05,
"loss": 1.5739,
"step": 2860
},
{
"epoch": 0.14,
"grad_norm": 51.25,
"learning_rate": 1.1257894736842106e-05,
"loss": 1.3658,
"step": 2870
},
{
"epoch": 0.14,
"grad_norm": 16.75,
"learning_rate": 1.1242105263157894e-05,
"loss": 1.6634,
"step": 2880
},
{
"epoch": 0.14,
"grad_norm": 16.375,
"learning_rate": 1.1226315789473685e-05,
"loss": 1.1902,
"step": 2890
},
{
"epoch": 0.14,
"grad_norm": 20.375,
"learning_rate": 1.1210526315789473e-05,
"loss": 1.2131,
"step": 2900
},
{
"epoch": 0.15,
"grad_norm": 18.875,
"learning_rate": 1.1194736842105264e-05,
"loss": 1.4465,
"step": 2910
},
{
"epoch": 0.15,
"grad_norm": 30.75,
"learning_rate": 1.1178947368421054e-05,
"loss": 1.6064,
"step": 2920
},
{
"epoch": 0.15,
"grad_norm": 15.1875,
"learning_rate": 1.1163157894736842e-05,
"loss": 1.5864,
"step": 2930
},
{
"epoch": 0.15,
"grad_norm": 23.25,
"learning_rate": 1.1147368421052633e-05,
"loss": 1.2329,
"step": 2940
},
{
"epoch": 0.15,
"grad_norm": 22.375,
"learning_rate": 1.1131578947368421e-05,
"loss": 1.3795,
"step": 2950
},
{
"epoch": 0.15,
"grad_norm": 24.0,
"learning_rate": 1.1115789473684212e-05,
"loss": 1.4095,
"step": 2960
},
{
"epoch": 0.15,
"grad_norm": 19.0,
"learning_rate": 1.11e-05,
"loss": 1.1682,
"step": 2970
},
{
"epoch": 0.15,
"grad_norm": 26.0,
"learning_rate": 1.108421052631579e-05,
"loss": 1.1633,
"step": 2980
},
{
"epoch": 0.15,
"grad_norm": 18.75,
"learning_rate": 1.106842105263158e-05,
"loss": 1.4457,
"step": 2990
},
{
"epoch": 0.15,
"grad_norm": 25.75,
"learning_rate": 1.1052631578947368e-05,
"loss": 1.4336,
"step": 3000
},
{
"epoch": 0.15,
"grad_norm": 26.625,
"learning_rate": 1.1036842105263158e-05,
"loss": 1.1143,
"step": 3010
},
{
"epoch": 0.15,
"grad_norm": 16.5,
"learning_rate": 1.1021052631578947e-05,
"loss": 1.2105,
"step": 3020
},
{
"epoch": 0.15,
"grad_norm": 38.75,
"learning_rate": 1.1005263157894737e-05,
"loss": 1.6449,
"step": 3030
},
{
"epoch": 0.15,
"grad_norm": 23.375,
"learning_rate": 1.0989473684210526e-05,
"loss": 1.3151,
"step": 3040
},
{
"epoch": 0.15,
"grad_norm": 15.3125,
"learning_rate": 1.0973684210526316e-05,
"loss": 1.6362,
"step": 3050
},
{
"epoch": 0.15,
"grad_norm": 24.625,
"learning_rate": 1.0957894736842105e-05,
"loss": 1.6158,
"step": 3060
},
{
"epoch": 0.15,
"grad_norm": 22.125,
"learning_rate": 1.0942105263157894e-05,
"loss": 1.6301,
"step": 3070
},
{
"epoch": 0.15,
"grad_norm": 24.625,
"learning_rate": 1.0926315789473684e-05,
"loss": 1.4099,
"step": 3080
},
{
"epoch": 0.15,
"grad_norm": 28.625,
"learning_rate": 1.0910526315789474e-05,
"loss": 1.6574,
"step": 3090
},
{
"epoch": 0.15,
"grad_norm": 18.75,
"learning_rate": 1.0894736842105265e-05,
"loss": 1.4458,
"step": 3100
},
{
"epoch": 0.16,
"grad_norm": 16.25,
"learning_rate": 1.0878947368421053e-05,
"loss": 1.4384,
"step": 3110
},
{
"epoch": 0.16,
"grad_norm": 17.5,
"learning_rate": 1.0863157894736844e-05,
"loss": 1.4054,
"step": 3120
},
{
"epoch": 0.16,
"grad_norm": 13.5625,
"learning_rate": 1.0847368421052632e-05,
"loss": 1.1473,
"step": 3130
},
{
"epoch": 0.16,
"grad_norm": 20.5,
"learning_rate": 1.0831578947368421e-05,
"loss": 1.2524,
"step": 3140
},
{
"epoch": 0.16,
"grad_norm": 22.375,
"learning_rate": 1.0815789473684211e-05,
"loss": 1.1765,
"step": 3150
},
{
"epoch": 0.16,
"grad_norm": 17.375,
"learning_rate": 1.08e-05,
"loss": 1.315,
"step": 3160
},
{
"epoch": 0.16,
"grad_norm": 33.75,
"learning_rate": 1.078421052631579e-05,
"loss": 1.6893,
"step": 3170
},
{
"epoch": 0.16,
"grad_norm": 17.375,
"learning_rate": 1.0768421052631579e-05,
"loss": 1.3754,
"step": 3180
},
{
"epoch": 0.16,
"grad_norm": 24.125,
"learning_rate": 1.075263157894737e-05,
"loss": 1.0585,
"step": 3190
},
{
"epoch": 0.16,
"grad_norm": 26.75,
"learning_rate": 1.0736842105263158e-05,
"loss": 1.5381,
"step": 3200
},
{
"epoch": 0.16,
"grad_norm": 21.25,
"learning_rate": 1.0721052631578947e-05,
"loss": 1.2503,
"step": 3210
},
{
"epoch": 0.16,
"grad_norm": 21.625,
"learning_rate": 1.0705263157894737e-05,
"loss": 1.4186,
"step": 3220
},
{
"epoch": 0.16,
"grad_norm": 13.5625,
"learning_rate": 1.0689473684210526e-05,
"loss": 1.5066,
"step": 3230
},
{
"epoch": 0.16,
"grad_norm": 25.375,
"learning_rate": 1.0673684210526316e-05,
"loss": 1.2001,
"step": 3240
},
{
"epoch": 0.16,
"grad_norm": 21.5,
"learning_rate": 1.0657894736842106e-05,
"loss": 1.4222,
"step": 3250
},
{
"epoch": 0.16,
"grad_norm": 20.25,
"learning_rate": 1.0642105263157897e-05,
"loss": 1.556,
"step": 3260
},
{
"epoch": 0.16,
"grad_norm": 18.125,
"learning_rate": 1.0626315789473685e-05,
"loss": 1.1219,
"step": 3270
},
{
"epoch": 0.16,
"grad_norm": 20.125,
"learning_rate": 1.0610526315789474e-05,
"loss": 1.5552,
"step": 3280
},
{
"epoch": 0.16,
"grad_norm": 17.75,
"learning_rate": 1.0594736842105264e-05,
"loss": 1.4247,
"step": 3290
},
{
"epoch": 0.17,
"grad_norm": 17.0,
"learning_rate": 1.0578947368421053e-05,
"loss": 1.0846,
"step": 3300
},
{
"epoch": 0.17,
"grad_norm": 18.0,
"learning_rate": 1.0563157894736843e-05,
"loss": 1.4747,
"step": 3310
},
{
"epoch": 0.17,
"grad_norm": 13.5,
"learning_rate": 1.0547368421052632e-05,
"loss": 1.3696,
"step": 3320
},
{
"epoch": 0.17,
"grad_norm": 22.625,
"learning_rate": 1.0531578947368422e-05,
"loss": 1.2841,
"step": 3330
},
{
"epoch": 0.17,
"grad_norm": 19.125,
"learning_rate": 1.051578947368421e-05,
"loss": 1.4871,
"step": 3340
},
{
"epoch": 0.17,
"grad_norm": 16.875,
"learning_rate": 1.05e-05,
"loss": 1.0924,
"step": 3350
},
{
"epoch": 0.17,
"grad_norm": 21.875,
"learning_rate": 1.048421052631579e-05,
"loss": 1.5268,
"step": 3360
},
{
"epoch": 0.17,
"grad_norm": 23.5,
"learning_rate": 1.0468421052631578e-05,
"loss": 1.2527,
"step": 3370
},
{
"epoch": 0.17,
"grad_norm": 21.625,
"learning_rate": 1.0452631578947369e-05,
"loss": 1.0235,
"step": 3380
},
{
"epoch": 0.17,
"grad_norm": 20.5,
"learning_rate": 1.0436842105263157e-05,
"loss": 1.3699,
"step": 3390
},
{
"epoch": 0.17,
"grad_norm": 8.8125,
"learning_rate": 1.0421052631578948e-05,
"loss": 1.2931,
"step": 3400
},
{
"epoch": 0.17,
"grad_norm": 15.8125,
"learning_rate": 1.0405263157894736e-05,
"loss": 0.9101,
"step": 3410
},
{
"epoch": 0.17,
"grad_norm": 19.875,
"learning_rate": 1.0389473684210527e-05,
"loss": 1.0891,
"step": 3420
},
{
"epoch": 0.17,
"grad_norm": 17.5,
"learning_rate": 1.0373684210526317e-05,
"loss": 1.2518,
"step": 3430
},
{
"epoch": 0.17,
"grad_norm": 26.25,
"learning_rate": 1.0357894736842106e-05,
"loss": 1.4293,
"step": 3440
},
{
"epoch": 0.17,
"grad_norm": 13.25,
"learning_rate": 1.0342105263157896e-05,
"loss": 1.2779,
"step": 3450
},
{
"epoch": 0.17,
"grad_norm": 19.75,
"learning_rate": 1.0326315789473685e-05,
"loss": 1.6036,
"step": 3460
},
{
"epoch": 0.17,
"grad_norm": 10.5625,
"learning_rate": 1.0310526315789475e-05,
"loss": 1.1933,
"step": 3470
},
{
"epoch": 0.17,
"grad_norm": 24.375,
"learning_rate": 1.0294736842105264e-05,
"loss": 1.2207,
"step": 3480
},
{
"epoch": 0.17,
"grad_norm": 15.0625,
"learning_rate": 1.0278947368421052e-05,
"loss": 0.9875,
"step": 3490
},
{
"epoch": 0.17,
"grad_norm": 17.375,
"learning_rate": 1.0263157894736843e-05,
"loss": 1.6437,
"step": 3500
},
{
"epoch": 0.18,
"grad_norm": 21.125,
"learning_rate": 1.0247368421052631e-05,
"loss": 1.5689,
"step": 3510
},
{
"epoch": 0.18,
"grad_norm": 19.75,
"learning_rate": 1.0231578947368422e-05,
"loss": 1.4524,
"step": 3520
},
{
"epoch": 0.18,
"grad_norm": 28.125,
"learning_rate": 1.021578947368421e-05,
"loss": 1.7354,
"step": 3530
},
{
"epoch": 0.18,
"grad_norm": 27.75,
"learning_rate": 1.02e-05,
"loss": 1.5075,
"step": 3540
},
{
"epoch": 0.18,
"grad_norm": 22.0,
"learning_rate": 1.018421052631579e-05,
"loss": 1.5991,
"step": 3550
},
{
"epoch": 0.18,
"grad_norm": 31.5,
"learning_rate": 1.0168421052631578e-05,
"loss": 1.3886,
"step": 3560
},
{
"epoch": 0.18,
"grad_norm": 14.375,
"learning_rate": 1.0152631578947368e-05,
"loss": 1.2167,
"step": 3570
},
{
"epoch": 0.18,
"grad_norm": 15.75,
"learning_rate": 1.0136842105263157e-05,
"loss": 1.4397,
"step": 3580
},
{
"epoch": 0.18,
"grad_norm": 21.0,
"learning_rate": 1.0121052631578949e-05,
"loss": 1.8334,
"step": 3590
},
{
"epoch": 0.18,
"grad_norm": 15.75,
"learning_rate": 1.0105263157894738e-05,
"loss": 1.5306,
"step": 3600
},
{
"epoch": 0.18,
"grad_norm": 15.5625,
"learning_rate": 1.0089473684210528e-05,
"loss": 1.5956,
"step": 3610
},
{
"epoch": 0.18,
"grad_norm": 33.5,
"learning_rate": 1.0073684210526317e-05,
"loss": 1.6766,
"step": 3620
},
{
"epoch": 0.18,
"grad_norm": 24.625,
"learning_rate": 1.0057894736842105e-05,
"loss": 1.2643,
"step": 3630
},
{
"epoch": 0.18,
"grad_norm": 10.75,
"learning_rate": 1.0042105263157896e-05,
"loss": 1.4199,
"step": 3640
},
{
"epoch": 0.18,
"grad_norm": 10.4375,
"learning_rate": 1.0026315789473684e-05,
"loss": 1.3638,
"step": 3650
},
{
"epoch": 0.18,
"grad_norm": 19.625,
"learning_rate": 1.0010526315789474e-05,
"loss": 1.1781,
"step": 3660
},
{
"epoch": 0.18,
"grad_norm": 18.875,
"learning_rate": 9.994736842105263e-06,
"loss": 1.5178,
"step": 3670
},
{
"epoch": 0.18,
"grad_norm": 20.875,
"learning_rate": 9.978947368421053e-06,
"loss": 1.175,
"step": 3680
},
{
"epoch": 0.18,
"grad_norm": 14.5625,
"learning_rate": 9.963157894736842e-06,
"loss": 1.3539,
"step": 3690
},
{
"epoch": 0.18,
"grad_norm": 20.75,
"learning_rate": 9.94736842105263e-06,
"loss": 1.1516,
"step": 3700
},
{
"epoch": 0.19,
"grad_norm": 21.875,
"learning_rate": 9.931578947368421e-06,
"loss": 1.3705,
"step": 3710
},
{
"epoch": 0.19,
"grad_norm": 19.125,
"learning_rate": 9.91578947368421e-06,
"loss": 1.1967,
"step": 3720
},
{
"epoch": 0.19,
"grad_norm": 32.25,
"learning_rate": 9.9e-06,
"loss": 1.2262,
"step": 3730
},
{
"epoch": 0.19,
"grad_norm": 13.0625,
"learning_rate": 9.884210526315789e-06,
"loss": 1.0924,
"step": 3740
},
{
"epoch": 0.19,
"grad_norm": 19.5,
"learning_rate": 9.868421052631579e-06,
"loss": 1.4826,
"step": 3750
},
{
"epoch": 0.19,
"grad_norm": 18.5,
"learning_rate": 9.85263157894737e-06,
"loss": 1.7201,
"step": 3760
},
{
"epoch": 0.19,
"grad_norm": 22.75,
"learning_rate": 9.836842105263158e-06,
"loss": 1.8823,
"step": 3770
},
{
"epoch": 0.19,
"grad_norm": 16.75,
"learning_rate": 9.821052631578948e-06,
"loss": 1.2102,
"step": 3780
},
{
"epoch": 0.19,
"grad_norm": 19.0,
"learning_rate": 9.805263157894737e-06,
"loss": 0.9203,
"step": 3790
},
{
"epoch": 0.19,
"grad_norm": 27.125,
"learning_rate": 9.789473684210527e-06,
"loss": 1.3829,
"step": 3800
},
{
"epoch": 0.19,
"grad_norm": 16.0,
"learning_rate": 9.773684210526316e-06,
"loss": 1.4504,
"step": 3810
},
{
"epoch": 0.19,
"grad_norm": 19.0,
"learning_rate": 9.757894736842106e-06,
"loss": 1.6228,
"step": 3820
},
{
"epoch": 0.19,
"grad_norm": 15.0,
"learning_rate": 9.742105263157895e-06,
"loss": 1.1509,
"step": 3830
},
{
"epoch": 0.19,
"grad_norm": 13.3125,
"learning_rate": 9.726315789473684e-06,
"loss": 1.4963,
"step": 3840
},
{
"epoch": 0.19,
"grad_norm": 26.875,
"learning_rate": 9.710526315789474e-06,
"loss": 1.2939,
"step": 3850
},
{
"epoch": 0.19,
"grad_norm": 19.625,
"learning_rate": 9.694736842105263e-06,
"loss": 1.396,
"step": 3860
},
{
"epoch": 0.19,
"grad_norm": 17.25,
"learning_rate": 9.678947368421053e-06,
"loss": 1.5547,
"step": 3870
},
{
"epoch": 0.19,
"grad_norm": 23.0,
"learning_rate": 9.663157894736842e-06,
"loss": 1.3473,
"step": 3880
},
{
"epoch": 0.19,
"grad_norm": 17.5,
"learning_rate": 9.647368421052632e-06,
"loss": 1.3847,
"step": 3890
},
{
"epoch": 0.2,
"grad_norm": 25.5,
"learning_rate": 9.63157894736842e-06,
"loss": 1.0258,
"step": 3900
},
{
"epoch": 0.2,
"grad_norm": 20.25,
"learning_rate": 9.61578947368421e-06,
"loss": 1.4518,
"step": 3910
},
{
"epoch": 0.2,
"grad_norm": 21.5,
"learning_rate": 9.600000000000001e-06,
"loss": 1.5955,
"step": 3920
},
{
"epoch": 0.2,
"grad_norm": 16.875,
"learning_rate": 9.58421052631579e-06,
"loss": 1.3555,
"step": 3930
},
{
"epoch": 0.2,
"grad_norm": 29.25,
"learning_rate": 9.56842105263158e-06,
"loss": 1.306,
"step": 3940
},
{
"epoch": 0.2,
"grad_norm": 18.375,
"learning_rate": 9.552631578947369e-06,
"loss": 1.0536,
"step": 3950
},
{
"epoch": 0.2,
"grad_norm": 16.25,
"learning_rate": 9.53684210526316e-06,
"loss": 1.4253,
"step": 3960
},
{
"epoch": 0.2,
"grad_norm": 22.25,
"learning_rate": 9.521052631578948e-06,
"loss": 1.335,
"step": 3970
},
{
"epoch": 0.2,
"grad_norm": 27.75,
"learning_rate": 9.505263157894737e-06,
"loss": 1.2626,
"step": 3980
},
{
"epoch": 0.2,
"grad_norm": 32.0,
"learning_rate": 9.489473684210527e-06,
"loss": 1.3404,
"step": 3990
},
{
"epoch": 0.2,
"grad_norm": 10.25,
"learning_rate": 9.473684210526315e-06,
"loss": 1.5403,
"step": 4000
},
{
"epoch": 0.2,
"eval_loss": 1.389373540878296,
"eval_runtime": 30.7104,
"eval_samples_per_second": 32.562,
"eval_steps_per_second": 32.562,
"step": 4000
}
],
"logging_steps": 10,
"max_steps": 10000,
"num_input_tokens_seen": 0,
"num_train_epochs": 1,
"save_steps": 2000,
"total_flos": 3.227844083712e+16,
"train_batch_size": 1,
"trial_name": null,
"trial_params": null
}