MeedoSam's picture
Uploaded checkpoint-4000
354f907 verified
{
"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": 6.03125,
"learning_rate": 3.0000000000000004e-08,
"loss": 0.5688,
"step": 10
},
{
"epoch": 0.0,
"grad_norm": 5.5,
"learning_rate": 6.000000000000001e-08,
"loss": 0.4811,
"step": 20
},
{
"epoch": 0.0,
"grad_norm": 4.71875,
"learning_rate": 9e-08,
"loss": 0.6172,
"step": 30
},
{
"epoch": 0.0,
"grad_norm": 6.59375,
"learning_rate": 1.2000000000000002e-07,
"loss": 0.5229,
"step": 40
},
{
"epoch": 0.0,
"grad_norm": 7.125,
"learning_rate": 1.5000000000000002e-07,
"loss": 0.5498,
"step": 50
},
{
"epoch": 0.0,
"grad_norm": 4.875,
"learning_rate": 1.8e-07,
"loss": 0.6114,
"step": 60
},
{
"epoch": 0.0,
"grad_norm": 5.40625,
"learning_rate": 2.1000000000000003e-07,
"loss": 0.5094,
"step": 70
},
{
"epoch": 0.0,
"grad_norm": 6.46875,
"learning_rate": 2.4000000000000003e-07,
"loss": 0.5955,
"step": 80
},
{
"epoch": 0.0,
"grad_norm": 5.0625,
"learning_rate": 2.7e-07,
"loss": 0.5804,
"step": 90
},
{
"epoch": 0.01,
"grad_norm": 6.0625,
"learning_rate": 3.0000000000000004e-07,
"loss": 0.5598,
"step": 100
},
{
"epoch": 0.01,
"grad_norm": 5.40625,
"learning_rate": 3.3e-07,
"loss": 0.5773,
"step": 110
},
{
"epoch": 0.01,
"grad_norm": 6.6875,
"learning_rate": 3.6e-07,
"loss": 0.6036,
"step": 120
},
{
"epoch": 0.01,
"grad_norm": 8.25,
"learning_rate": 3.9e-07,
"loss": 0.5187,
"step": 130
},
{
"epoch": 0.01,
"grad_norm": 6.09375,
"learning_rate": 4.2000000000000006e-07,
"loss": 0.6931,
"step": 140
},
{
"epoch": 0.01,
"grad_norm": 4.40625,
"learning_rate": 4.5e-07,
"loss": 0.5209,
"step": 150
},
{
"epoch": 0.01,
"grad_norm": 4.375,
"learning_rate": 4.800000000000001e-07,
"loss": 0.6007,
"step": 160
},
{
"epoch": 0.01,
"grad_norm": 7.375,
"learning_rate": 5.100000000000001e-07,
"loss": 0.5626,
"step": 170
},
{
"epoch": 0.01,
"grad_norm": 5.9375,
"learning_rate": 5.4e-07,
"loss": 0.5869,
"step": 180
},
{
"epoch": 0.01,
"grad_norm": 5.53125,
"learning_rate": 5.7e-07,
"loss": 0.5639,
"step": 190
},
{
"epoch": 0.01,
"grad_norm": 4.25,
"learning_rate": 6.000000000000001e-07,
"loss": 0.5354,
"step": 200
},
{
"epoch": 0.01,
"grad_norm": 7.09375,
"learning_rate": 6.3e-07,
"loss": 0.492,
"step": 210
},
{
"epoch": 0.01,
"grad_norm": 4.9375,
"learning_rate": 6.6e-07,
"loss": 0.5569,
"step": 220
},
{
"epoch": 0.01,
"grad_norm": 6.71875,
"learning_rate": 6.900000000000001e-07,
"loss": 0.5664,
"step": 230
},
{
"epoch": 0.01,
"grad_norm": 6.375,
"learning_rate": 7.2e-07,
"loss": 0.4673,
"step": 240
},
{
"epoch": 0.01,
"grad_norm": 5.09375,
"learning_rate": 7.5e-07,
"loss": 0.6147,
"step": 250
},
{
"epoch": 0.01,
"grad_norm": 5.65625,
"learning_rate": 7.8e-07,
"loss": 0.6587,
"step": 260
},
{
"epoch": 0.01,
"grad_norm": 5.53125,
"learning_rate": 8.100000000000001e-07,
"loss": 0.5524,
"step": 270
},
{
"epoch": 0.01,
"grad_norm": 5.6875,
"learning_rate": 8.400000000000001e-07,
"loss": 0.5877,
"step": 280
},
{
"epoch": 0.01,
"grad_norm": 6.125,
"learning_rate": 8.699999999999999e-07,
"loss": 0.5573,
"step": 290
},
{
"epoch": 0.01,
"grad_norm": 5.71875,
"learning_rate": 9e-07,
"loss": 0.5665,
"step": 300
},
{
"epoch": 0.02,
"grad_norm": 4.65625,
"learning_rate": 9.3e-07,
"loss": 0.5659,
"step": 310
},
{
"epoch": 0.02,
"grad_norm": 5.5625,
"learning_rate": 9.600000000000001e-07,
"loss": 0.4814,
"step": 320
},
{
"epoch": 0.02,
"grad_norm": 4.875,
"learning_rate": 9.9e-07,
"loss": 0.5048,
"step": 330
},
{
"epoch": 0.02,
"grad_norm": 8.5625,
"learning_rate": 1.0200000000000002e-06,
"loss": 0.5021,
"step": 340
},
{
"epoch": 0.02,
"grad_norm": 6.5,
"learning_rate": 1.05e-06,
"loss": 0.5175,
"step": 350
},
{
"epoch": 0.02,
"grad_norm": 5.0625,
"learning_rate": 1.08e-06,
"loss": 0.55,
"step": 360
},
{
"epoch": 0.02,
"grad_norm": 5.6875,
"learning_rate": 1.11e-06,
"loss": 0.5565,
"step": 370
},
{
"epoch": 0.02,
"grad_norm": 6.90625,
"learning_rate": 1.14e-06,
"loss": 0.4719,
"step": 380
},
{
"epoch": 0.02,
"grad_norm": 5.59375,
"learning_rate": 1.17e-06,
"loss": 0.5117,
"step": 390
},
{
"epoch": 0.02,
"grad_norm": 5.09375,
"learning_rate": 1.2000000000000002e-06,
"loss": 0.4718,
"step": 400
},
{
"epoch": 0.02,
"grad_norm": 3.921875,
"learning_rate": 1.2299999999999999e-06,
"loss": 0.5335,
"step": 410
},
{
"epoch": 0.02,
"grad_norm": 6.21875,
"learning_rate": 1.26e-06,
"loss": 0.539,
"step": 420
},
{
"epoch": 0.02,
"grad_norm": 5.96875,
"learning_rate": 1.29e-06,
"loss": 0.4928,
"step": 430
},
{
"epoch": 0.02,
"grad_norm": 7.0625,
"learning_rate": 1.32e-06,
"loss": 0.6026,
"step": 440
},
{
"epoch": 0.02,
"grad_norm": 5.3125,
"learning_rate": 1.35e-06,
"loss": 0.5829,
"step": 450
},
{
"epoch": 0.02,
"grad_norm": 5.21875,
"learning_rate": 1.3800000000000001e-06,
"loss": 0.5167,
"step": 460
},
{
"epoch": 0.02,
"grad_norm": 4.8125,
"learning_rate": 1.41e-06,
"loss": 0.475,
"step": 470
},
{
"epoch": 0.02,
"grad_norm": 5.34375,
"learning_rate": 1.44e-06,
"loss": 0.5965,
"step": 480
},
{
"epoch": 0.02,
"grad_norm": 7.6875,
"learning_rate": 1.4700000000000001e-06,
"loss": 0.5396,
"step": 490
},
{
"epoch": 0.03,
"grad_norm": 5.53125,
"learning_rate": 1.5e-06,
"loss": 0.4778,
"step": 500
},
{
"epoch": 0.03,
"grad_norm": 4.78125,
"learning_rate": 1.498421052631579e-06,
"loss": 0.4872,
"step": 510
},
{
"epoch": 0.03,
"grad_norm": 6.375,
"learning_rate": 1.4968421052631579e-06,
"loss": 0.5117,
"step": 520
},
{
"epoch": 0.03,
"grad_norm": 5.375,
"learning_rate": 1.4952631578947368e-06,
"loss": 0.5354,
"step": 530
},
{
"epoch": 0.03,
"grad_norm": 4.875,
"learning_rate": 1.4936842105263159e-06,
"loss": 0.5136,
"step": 540
},
{
"epoch": 0.03,
"grad_norm": 6.0625,
"learning_rate": 1.4921052631578948e-06,
"loss": 0.6154,
"step": 550
},
{
"epoch": 0.03,
"grad_norm": 5.09375,
"learning_rate": 1.4905263157894737e-06,
"loss": 0.5474,
"step": 560
},
{
"epoch": 0.03,
"grad_norm": 5.625,
"learning_rate": 1.4889473684210526e-06,
"loss": 0.4899,
"step": 570
},
{
"epoch": 0.03,
"grad_norm": 5.125,
"learning_rate": 1.4873684210526315e-06,
"loss": 0.5327,
"step": 580
},
{
"epoch": 0.03,
"grad_norm": 5.625,
"learning_rate": 1.4857894736842106e-06,
"loss": 0.5615,
"step": 590
},
{
"epoch": 0.03,
"grad_norm": 4.28125,
"learning_rate": 1.4842105263157895e-06,
"loss": 0.5382,
"step": 600
},
{
"epoch": 0.03,
"grad_norm": 6.125,
"learning_rate": 1.4826315789473684e-06,
"loss": 0.4741,
"step": 610
},
{
"epoch": 0.03,
"grad_norm": 5.15625,
"learning_rate": 1.4810526315789473e-06,
"loss": 0.4843,
"step": 620
},
{
"epoch": 0.03,
"grad_norm": 5.09375,
"learning_rate": 1.4794736842105265e-06,
"loss": 0.5417,
"step": 630
},
{
"epoch": 0.03,
"grad_norm": 4.8125,
"learning_rate": 1.4778947368421054e-06,
"loss": 0.5187,
"step": 640
},
{
"epoch": 0.03,
"grad_norm": 6.21875,
"learning_rate": 1.4763157894736843e-06,
"loss": 0.5706,
"step": 650
},
{
"epoch": 0.03,
"grad_norm": 6.21875,
"learning_rate": 1.4747368421052632e-06,
"loss": 0.4604,
"step": 660
},
{
"epoch": 0.03,
"grad_norm": 6.25,
"learning_rate": 1.473157894736842e-06,
"loss": 0.5064,
"step": 670
},
{
"epoch": 0.03,
"grad_norm": 4.09375,
"learning_rate": 1.4715789473684212e-06,
"loss": 0.54,
"step": 680
},
{
"epoch": 0.03,
"grad_norm": 5.3125,
"learning_rate": 1.4700000000000001e-06,
"loss": 0.5333,
"step": 690
},
{
"epoch": 0.04,
"grad_norm": 6.09375,
"learning_rate": 1.468421052631579e-06,
"loss": 0.5975,
"step": 700
},
{
"epoch": 0.04,
"grad_norm": 6.1875,
"learning_rate": 1.466842105263158e-06,
"loss": 0.5563,
"step": 710
},
{
"epoch": 0.04,
"grad_norm": 5.1875,
"learning_rate": 1.4652631578947368e-06,
"loss": 0.4694,
"step": 720
},
{
"epoch": 0.04,
"grad_norm": 5.09375,
"learning_rate": 1.463684210526316e-06,
"loss": 0.6063,
"step": 730
},
{
"epoch": 0.04,
"grad_norm": 6.125,
"learning_rate": 1.4621052631578946e-06,
"loss": 0.4839,
"step": 740
},
{
"epoch": 0.04,
"grad_norm": 4.625,
"learning_rate": 1.4605263157894738e-06,
"loss": 0.4172,
"step": 750
},
{
"epoch": 0.04,
"grad_norm": 6.34375,
"learning_rate": 1.4589473684210527e-06,
"loss": 0.4914,
"step": 760
},
{
"epoch": 0.04,
"grad_norm": 7.4375,
"learning_rate": 1.4573684210526316e-06,
"loss": 0.4758,
"step": 770
},
{
"epoch": 0.04,
"grad_norm": 4.625,
"learning_rate": 1.4557894736842105e-06,
"loss": 0.4667,
"step": 780
},
{
"epoch": 0.04,
"grad_norm": 5.21875,
"learning_rate": 1.4542105263157894e-06,
"loss": 0.5244,
"step": 790
},
{
"epoch": 0.04,
"grad_norm": 5.625,
"learning_rate": 1.4526315789473685e-06,
"loss": 0.4719,
"step": 800
},
{
"epoch": 0.04,
"grad_norm": 6.21875,
"learning_rate": 1.4510526315789474e-06,
"loss": 0.4166,
"step": 810
},
{
"epoch": 0.04,
"grad_norm": 3.78125,
"learning_rate": 1.4494736842105263e-06,
"loss": 0.4209,
"step": 820
},
{
"epoch": 0.04,
"grad_norm": 6.46875,
"learning_rate": 1.4478947368421052e-06,
"loss": 0.4946,
"step": 830
},
{
"epoch": 0.04,
"grad_norm": 5.0625,
"learning_rate": 1.4463157894736844e-06,
"loss": 0.4399,
"step": 840
},
{
"epoch": 0.04,
"grad_norm": 4.65625,
"learning_rate": 1.4447368421052633e-06,
"loss": 0.4578,
"step": 850
},
{
"epoch": 0.04,
"grad_norm": 5.25,
"learning_rate": 1.4431578947368422e-06,
"loss": 0.5727,
"step": 860
},
{
"epoch": 0.04,
"grad_norm": 5.75,
"learning_rate": 1.441578947368421e-06,
"loss": 0.4572,
"step": 870
},
{
"epoch": 0.04,
"grad_norm": 5.03125,
"learning_rate": 1.44e-06,
"loss": 0.5346,
"step": 880
},
{
"epoch": 0.04,
"grad_norm": 5.6875,
"learning_rate": 1.438421052631579e-06,
"loss": 0.4093,
"step": 890
},
{
"epoch": 0.04,
"grad_norm": 4.875,
"learning_rate": 1.436842105263158e-06,
"loss": 0.5276,
"step": 900
},
{
"epoch": 0.05,
"grad_norm": 6.21875,
"learning_rate": 1.435263157894737e-06,
"loss": 0.4809,
"step": 910
},
{
"epoch": 0.05,
"grad_norm": 5.3125,
"learning_rate": 1.4336842105263158e-06,
"loss": 0.5037,
"step": 920
},
{
"epoch": 0.05,
"grad_norm": 3.984375,
"learning_rate": 1.4321052631578947e-06,
"loss": 0.5046,
"step": 930
},
{
"epoch": 0.05,
"grad_norm": 3.890625,
"learning_rate": 1.4305263157894738e-06,
"loss": 0.4633,
"step": 940
},
{
"epoch": 0.05,
"grad_norm": 5.0625,
"learning_rate": 1.4289473684210525e-06,
"loss": 0.3924,
"step": 950
},
{
"epoch": 0.05,
"grad_norm": 5.0,
"learning_rate": 1.4273684210526317e-06,
"loss": 0.5488,
"step": 960
},
{
"epoch": 0.05,
"grad_norm": 6.0,
"learning_rate": 1.4257894736842106e-06,
"loss": 0.5566,
"step": 970
},
{
"epoch": 0.05,
"grad_norm": 4.71875,
"learning_rate": 1.4242105263157895e-06,
"loss": 0.4756,
"step": 980
},
{
"epoch": 0.05,
"grad_norm": 4.1875,
"learning_rate": 1.4226315789473684e-06,
"loss": 0.3859,
"step": 990
},
{
"epoch": 0.05,
"grad_norm": 4.8125,
"learning_rate": 1.4210526315789473e-06,
"loss": 0.4438,
"step": 1000
},
{
"epoch": 0.05,
"grad_norm": 5.875,
"learning_rate": 1.4194736842105264e-06,
"loss": 0.5533,
"step": 1010
},
{
"epoch": 0.05,
"grad_norm": 6.09375,
"learning_rate": 1.4178947368421053e-06,
"loss": 0.5438,
"step": 1020
},
{
"epoch": 0.05,
"grad_norm": 4.59375,
"learning_rate": 1.4163157894736842e-06,
"loss": 0.574,
"step": 1030
},
{
"epoch": 0.05,
"grad_norm": 4.375,
"learning_rate": 1.4147368421052631e-06,
"loss": 0.4077,
"step": 1040
},
{
"epoch": 0.05,
"grad_norm": 5.21875,
"learning_rate": 1.4131578947368422e-06,
"loss": 0.5732,
"step": 1050
},
{
"epoch": 0.05,
"grad_norm": 5.40625,
"learning_rate": 1.4115789473684211e-06,
"loss": 0.4751,
"step": 1060
},
{
"epoch": 0.05,
"grad_norm": 4.78125,
"learning_rate": 1.41e-06,
"loss": 0.4842,
"step": 1070
},
{
"epoch": 0.05,
"grad_norm": 5.125,
"learning_rate": 1.408421052631579e-06,
"loss": 0.5282,
"step": 1080
},
{
"epoch": 0.05,
"grad_norm": 4.6875,
"learning_rate": 1.4068421052631579e-06,
"loss": 0.4528,
"step": 1090
},
{
"epoch": 0.06,
"grad_norm": 4.1875,
"learning_rate": 1.405263157894737e-06,
"loss": 0.5392,
"step": 1100
},
{
"epoch": 0.06,
"grad_norm": 3.5,
"learning_rate": 1.4036842105263159e-06,
"loss": 0.4842,
"step": 1110
},
{
"epoch": 0.06,
"grad_norm": 5.3125,
"learning_rate": 1.4021052631578948e-06,
"loss": 0.4341,
"step": 1120
},
{
"epoch": 0.06,
"grad_norm": 4.84375,
"learning_rate": 1.4005263157894737e-06,
"loss": 0.4241,
"step": 1130
},
{
"epoch": 0.06,
"grad_norm": 4.96875,
"learning_rate": 1.3989473684210526e-06,
"loss": 0.4254,
"step": 1140
},
{
"epoch": 0.06,
"grad_norm": 5.8125,
"learning_rate": 1.3973684210526317e-06,
"loss": 0.5204,
"step": 1150
},
{
"epoch": 0.06,
"grad_norm": 5.15625,
"learning_rate": 1.3957894736842104e-06,
"loss": 0.5232,
"step": 1160
},
{
"epoch": 0.06,
"grad_norm": 5.5625,
"learning_rate": 1.3942105263157895e-06,
"loss": 0.4818,
"step": 1170
},
{
"epoch": 0.06,
"grad_norm": 4.15625,
"learning_rate": 1.3926315789473685e-06,
"loss": 0.5224,
"step": 1180
},
{
"epoch": 0.06,
"grad_norm": 5.53125,
"learning_rate": 1.3910526315789476e-06,
"loss": 0.5449,
"step": 1190
},
{
"epoch": 0.06,
"grad_norm": 5.53125,
"learning_rate": 1.3894736842105263e-06,
"loss": 0.5367,
"step": 1200
},
{
"epoch": 0.06,
"grad_norm": 5.71875,
"learning_rate": 1.3878947368421052e-06,
"loss": 0.5962,
"step": 1210
},
{
"epoch": 0.06,
"grad_norm": 4.125,
"learning_rate": 1.3863157894736843e-06,
"loss": 0.3955,
"step": 1220
},
{
"epoch": 0.06,
"grad_norm": 4.96875,
"learning_rate": 1.3847368421052632e-06,
"loss": 0.5646,
"step": 1230
},
{
"epoch": 0.06,
"grad_norm": 4.84375,
"learning_rate": 1.383157894736842e-06,
"loss": 0.515,
"step": 1240
},
{
"epoch": 0.06,
"grad_norm": 4.65625,
"learning_rate": 1.381578947368421e-06,
"loss": 0.4916,
"step": 1250
},
{
"epoch": 0.06,
"grad_norm": 4.34375,
"learning_rate": 1.3800000000000001e-06,
"loss": 0.4971,
"step": 1260
},
{
"epoch": 0.06,
"grad_norm": 5.5625,
"learning_rate": 1.378421052631579e-06,
"loss": 0.5121,
"step": 1270
},
{
"epoch": 0.06,
"grad_norm": 6.84375,
"learning_rate": 1.376842105263158e-06,
"loss": 0.4628,
"step": 1280
},
{
"epoch": 0.06,
"grad_norm": 5.6875,
"learning_rate": 1.3752631578947368e-06,
"loss": 0.4309,
"step": 1290
},
{
"epoch": 0.07,
"grad_norm": 4.09375,
"learning_rate": 1.3736842105263158e-06,
"loss": 0.4871,
"step": 1300
},
{
"epoch": 0.07,
"grad_norm": 4.5,
"learning_rate": 1.3721052631578949e-06,
"loss": 0.5438,
"step": 1310
},
{
"epoch": 0.07,
"grad_norm": 4.03125,
"learning_rate": 1.3705263157894738e-06,
"loss": 0.4459,
"step": 1320
},
{
"epoch": 0.07,
"grad_norm": 4.78125,
"learning_rate": 1.3689473684210527e-06,
"loss": 0.4734,
"step": 1330
},
{
"epoch": 0.07,
"grad_norm": 6.125,
"learning_rate": 1.3673684210526316e-06,
"loss": 0.5321,
"step": 1340
},
{
"epoch": 0.07,
"grad_norm": 5.59375,
"learning_rate": 1.3657894736842107e-06,
"loss": 0.4859,
"step": 1350
},
{
"epoch": 0.07,
"grad_norm": 5.59375,
"learning_rate": 1.3642105263157896e-06,
"loss": 0.4804,
"step": 1360
},
{
"epoch": 0.07,
"grad_norm": 5.21875,
"learning_rate": 1.3626315789473683e-06,
"loss": 0.5041,
"step": 1370
},
{
"epoch": 0.07,
"grad_norm": 6.71875,
"learning_rate": 1.3610526315789474e-06,
"loss": 0.5103,
"step": 1380
},
{
"epoch": 0.07,
"grad_norm": 5.84375,
"learning_rate": 1.3594736842105263e-06,
"loss": 0.4737,
"step": 1390
},
{
"epoch": 0.07,
"grad_norm": 6.125,
"learning_rate": 1.3578947368421055e-06,
"loss": 0.4954,
"step": 1400
},
{
"epoch": 0.07,
"grad_norm": 5.78125,
"learning_rate": 1.3563157894736842e-06,
"loss": 0.5358,
"step": 1410
},
{
"epoch": 0.07,
"grad_norm": 5.625,
"learning_rate": 1.3547368421052633e-06,
"loss": 0.4667,
"step": 1420
},
{
"epoch": 0.07,
"grad_norm": 5.84375,
"learning_rate": 1.3531578947368422e-06,
"loss": 0.4511,
"step": 1430
},
{
"epoch": 0.07,
"grad_norm": 4.875,
"learning_rate": 1.351578947368421e-06,
"loss": 0.4853,
"step": 1440
},
{
"epoch": 0.07,
"grad_norm": 7.34375,
"learning_rate": 1.35e-06,
"loss": 0.4906,
"step": 1450
},
{
"epoch": 0.07,
"grad_norm": 3.515625,
"learning_rate": 1.348421052631579e-06,
"loss": 0.4553,
"step": 1460
},
{
"epoch": 0.07,
"grad_norm": 5.1875,
"learning_rate": 1.346842105263158e-06,
"loss": 0.4767,
"step": 1470
},
{
"epoch": 0.07,
"grad_norm": 4.84375,
"learning_rate": 1.345263157894737e-06,
"loss": 0.5855,
"step": 1480
},
{
"epoch": 0.07,
"grad_norm": 5.90625,
"learning_rate": 1.3436842105263158e-06,
"loss": 0.548,
"step": 1490
},
{
"epoch": 0.07,
"grad_norm": 5.0625,
"learning_rate": 1.3421052631578947e-06,
"loss": 0.4498,
"step": 1500
},
{
"epoch": 0.08,
"grad_norm": 5.09375,
"learning_rate": 1.3405263157894736e-06,
"loss": 0.453,
"step": 1510
},
{
"epoch": 0.08,
"grad_norm": 5.3125,
"learning_rate": 1.3389473684210528e-06,
"loss": 0.4706,
"step": 1520
},
{
"epoch": 0.08,
"grad_norm": 3.984375,
"learning_rate": 1.3373684210526317e-06,
"loss": 0.492,
"step": 1530
},
{
"epoch": 0.08,
"grad_norm": 4.8125,
"learning_rate": 1.3357894736842106e-06,
"loss": 0.5406,
"step": 1540
},
{
"epoch": 0.08,
"grad_norm": 5.3125,
"learning_rate": 1.3342105263157895e-06,
"loss": 0.4736,
"step": 1550
},
{
"epoch": 0.08,
"grad_norm": 5.5625,
"learning_rate": 1.3326315789473686e-06,
"loss": 0.6124,
"step": 1560
},
{
"epoch": 0.08,
"grad_norm": 4.96875,
"learning_rate": 1.3310526315789475e-06,
"loss": 0.4797,
"step": 1570
},
{
"epoch": 0.08,
"grad_norm": 7.65625,
"learning_rate": 1.3294736842105262e-06,
"loss": 0.5659,
"step": 1580
},
{
"epoch": 0.08,
"grad_norm": 4.53125,
"learning_rate": 1.3278947368421053e-06,
"loss": 0.4673,
"step": 1590
},
{
"epoch": 0.08,
"grad_norm": 5.625,
"learning_rate": 1.3263157894736842e-06,
"loss": 0.5162,
"step": 1600
},
{
"epoch": 0.08,
"grad_norm": 4.96875,
"learning_rate": 1.3247368421052633e-06,
"loss": 0.4925,
"step": 1610
},
{
"epoch": 0.08,
"grad_norm": 4.0625,
"learning_rate": 1.323157894736842e-06,
"loss": 0.4602,
"step": 1620
},
{
"epoch": 0.08,
"grad_norm": 4.03125,
"learning_rate": 1.3215789473684212e-06,
"loss": 0.4743,
"step": 1630
},
{
"epoch": 0.08,
"grad_norm": 5.5625,
"learning_rate": 1.32e-06,
"loss": 0.4944,
"step": 1640
},
{
"epoch": 0.08,
"grad_norm": 5.625,
"learning_rate": 1.318421052631579e-06,
"loss": 0.5525,
"step": 1650
},
{
"epoch": 0.08,
"grad_norm": 4.90625,
"learning_rate": 1.3168421052631579e-06,
"loss": 0.4512,
"step": 1660
},
{
"epoch": 0.08,
"grad_norm": 4.90625,
"learning_rate": 1.3152631578947368e-06,
"loss": 0.4107,
"step": 1670
},
{
"epoch": 0.08,
"grad_norm": 3.984375,
"learning_rate": 1.313684210526316e-06,
"loss": 0.3901,
"step": 1680
},
{
"epoch": 0.08,
"grad_norm": 5.59375,
"learning_rate": 1.3121052631578948e-06,
"loss": 0.4924,
"step": 1690
},
{
"epoch": 0.09,
"grad_norm": 5.84375,
"learning_rate": 1.3105263157894737e-06,
"loss": 0.6272,
"step": 1700
},
{
"epoch": 0.09,
"grad_norm": 6.4375,
"learning_rate": 1.3089473684210526e-06,
"loss": 0.4816,
"step": 1710
},
{
"epoch": 0.09,
"grad_norm": 5.09375,
"learning_rate": 1.3073684210526315e-06,
"loss": 0.4654,
"step": 1720
},
{
"epoch": 0.09,
"grad_norm": 4.96875,
"learning_rate": 1.3057894736842107e-06,
"loss": 0.3862,
"step": 1730
},
{
"epoch": 0.09,
"grad_norm": 4.65625,
"learning_rate": 1.3042105263157896e-06,
"loss": 0.491,
"step": 1740
},
{
"epoch": 0.09,
"grad_norm": 5.84375,
"learning_rate": 1.3026315789473685e-06,
"loss": 0.4092,
"step": 1750
},
{
"epoch": 0.09,
"grad_norm": 4.5625,
"learning_rate": 1.3010526315789474e-06,
"loss": 0.43,
"step": 1760
},
{
"epoch": 0.09,
"grad_norm": 5.28125,
"learning_rate": 1.2994736842105265e-06,
"loss": 0.4754,
"step": 1770
},
{
"epoch": 0.09,
"grad_norm": 4.71875,
"learning_rate": 1.2978947368421054e-06,
"loss": 0.5338,
"step": 1780
},
{
"epoch": 0.09,
"grad_norm": 6.3125,
"learning_rate": 1.296315789473684e-06,
"loss": 0.4727,
"step": 1790
},
{
"epoch": 0.09,
"grad_norm": 4.03125,
"learning_rate": 1.2947368421052632e-06,
"loss": 0.485,
"step": 1800
},
{
"epoch": 0.09,
"grad_norm": 4.3125,
"learning_rate": 1.2931578947368421e-06,
"loss": 0.4643,
"step": 1810
},
{
"epoch": 0.09,
"grad_norm": 4.28125,
"learning_rate": 1.2915789473684212e-06,
"loss": 0.5617,
"step": 1820
},
{
"epoch": 0.09,
"grad_norm": 4.875,
"learning_rate": 1.29e-06,
"loss": 0.4752,
"step": 1830
},
{
"epoch": 0.09,
"grad_norm": 4.6875,
"learning_rate": 1.288421052631579e-06,
"loss": 0.4927,
"step": 1840
},
{
"epoch": 0.09,
"grad_norm": 7.9375,
"learning_rate": 1.286842105263158e-06,
"loss": 0.5522,
"step": 1850
},
{
"epoch": 0.09,
"grad_norm": 4.53125,
"learning_rate": 1.2852631578947369e-06,
"loss": 0.5096,
"step": 1860
},
{
"epoch": 0.09,
"grad_norm": 3.390625,
"learning_rate": 1.2836842105263158e-06,
"loss": 0.5059,
"step": 1870
},
{
"epoch": 0.09,
"grad_norm": 5.0,
"learning_rate": 1.2821052631578947e-06,
"loss": 0.5205,
"step": 1880
},
{
"epoch": 0.09,
"grad_norm": 5.0625,
"learning_rate": 1.2805263157894738e-06,
"loss": 0.5575,
"step": 1890
},
{
"epoch": 0.1,
"grad_norm": 5.59375,
"learning_rate": 1.2789473684210527e-06,
"loss": 0.4836,
"step": 1900
},
{
"epoch": 0.1,
"grad_norm": 6.53125,
"learning_rate": 1.2773684210526316e-06,
"loss": 0.4563,
"step": 1910
},
{
"epoch": 0.1,
"grad_norm": 5.8125,
"learning_rate": 1.2757894736842105e-06,
"loss": 0.4431,
"step": 1920
},
{
"epoch": 0.1,
"grad_norm": 5.21875,
"learning_rate": 1.2742105263157894e-06,
"loss": 0.4686,
"step": 1930
},
{
"epoch": 0.1,
"grad_norm": 4.96875,
"learning_rate": 1.2726315789473685e-06,
"loss": 0.4486,
"step": 1940
},
{
"epoch": 0.1,
"grad_norm": 4.875,
"learning_rate": 1.2710526315789474e-06,
"loss": 0.4044,
"step": 1950
},
{
"epoch": 0.1,
"grad_norm": 5.03125,
"learning_rate": 1.2694736842105264e-06,
"loss": 0.5884,
"step": 1960
},
{
"epoch": 0.1,
"grad_norm": 4.5,
"learning_rate": 1.2678947368421053e-06,
"loss": 0.4116,
"step": 1970
},
{
"epoch": 0.1,
"grad_norm": 5.0625,
"learning_rate": 1.2663157894736844e-06,
"loss": 0.404,
"step": 1980
},
{
"epoch": 0.1,
"grad_norm": 4.03125,
"learning_rate": 1.2647368421052633e-06,
"loss": 0.4443,
"step": 1990
},
{
"epoch": 0.1,
"grad_norm": 5.0625,
"learning_rate": 1.263157894736842e-06,
"loss": 0.5572,
"step": 2000
},
{
"epoch": 0.1,
"eval_loss": 0.4979419410228729,
"eval_runtime": 64.9548,
"eval_samples_per_second": 15.395,
"eval_steps_per_second": 15.395,
"step": 2000
},
{
"epoch": 0.1,
"grad_norm": 5.03125,
"learning_rate": 1.261578947368421e-06,
"loss": 0.4184,
"step": 2010
},
{
"epoch": 0.1,
"grad_norm": 5.84375,
"learning_rate": 1.26e-06,
"loss": 0.4894,
"step": 2020
},
{
"epoch": 0.1,
"grad_norm": 4.03125,
"learning_rate": 1.2584210526315791e-06,
"loss": 0.4264,
"step": 2030
},
{
"epoch": 0.1,
"grad_norm": 4.96875,
"learning_rate": 1.2568421052631578e-06,
"loss": 0.4311,
"step": 2040
},
{
"epoch": 0.1,
"grad_norm": 5.59375,
"learning_rate": 1.255263157894737e-06,
"loss": 0.577,
"step": 2050
},
{
"epoch": 0.1,
"grad_norm": 4.0625,
"learning_rate": 1.2536842105263158e-06,
"loss": 0.3871,
"step": 2060
},
{
"epoch": 0.1,
"grad_norm": 4.78125,
"learning_rate": 1.2521052631578948e-06,
"loss": 0.4829,
"step": 2070
},
{
"epoch": 0.1,
"grad_norm": 5.25,
"learning_rate": 1.2505263157894737e-06,
"loss": 0.5132,
"step": 2080
},
{
"epoch": 0.1,
"grad_norm": 6.71875,
"learning_rate": 1.2489473684210526e-06,
"loss": 0.5001,
"step": 2090
},
{
"epoch": 0.1,
"grad_norm": 5.40625,
"learning_rate": 1.2473684210526317e-06,
"loss": 0.4924,
"step": 2100
},
{
"epoch": 0.11,
"grad_norm": 5.40625,
"learning_rate": 1.2457894736842106e-06,
"loss": 0.4732,
"step": 2110
},
{
"epoch": 0.11,
"grad_norm": 4.59375,
"learning_rate": 1.2442105263157895e-06,
"loss": 0.4098,
"step": 2120
},
{
"epoch": 0.11,
"grad_norm": 4.25,
"learning_rate": 1.2426315789473684e-06,
"loss": 0.4366,
"step": 2130
},
{
"epoch": 0.11,
"grad_norm": 4.875,
"learning_rate": 1.2410526315789473e-06,
"loss": 0.4782,
"step": 2140
},
{
"epoch": 0.11,
"grad_norm": 4.875,
"learning_rate": 1.2394736842105264e-06,
"loss": 0.4936,
"step": 2150
},
{
"epoch": 0.11,
"grad_norm": 4.4375,
"learning_rate": 1.2378947368421053e-06,
"loss": 0.4185,
"step": 2160
},
{
"epoch": 0.11,
"grad_norm": 4.53125,
"learning_rate": 1.2363157894736842e-06,
"loss": 0.4162,
"step": 2170
},
{
"epoch": 0.11,
"grad_norm": 4.75,
"learning_rate": 1.2347368421052631e-06,
"loss": 0.4217,
"step": 2180
},
{
"epoch": 0.11,
"grad_norm": 5.0625,
"learning_rate": 1.2331578947368423e-06,
"loss": 0.5339,
"step": 2190
},
{
"epoch": 0.11,
"grad_norm": 6.71875,
"learning_rate": 1.2315789473684212e-06,
"loss": 0.4711,
"step": 2200
},
{
"epoch": 0.11,
"grad_norm": 4.03125,
"learning_rate": 1.2299999999999999e-06,
"loss": 0.5137,
"step": 2210
},
{
"epoch": 0.11,
"grad_norm": 4.0625,
"learning_rate": 1.228421052631579e-06,
"loss": 0.4556,
"step": 2220
},
{
"epoch": 0.11,
"grad_norm": 3.15625,
"learning_rate": 1.226842105263158e-06,
"loss": 0.5223,
"step": 2230
},
{
"epoch": 0.11,
"grad_norm": 6.875,
"learning_rate": 1.225263157894737e-06,
"loss": 0.4735,
"step": 2240
},
{
"epoch": 0.11,
"grad_norm": 5.0625,
"learning_rate": 1.2236842105263157e-06,
"loss": 0.5419,
"step": 2250
},
{
"epoch": 0.11,
"grad_norm": 5.09375,
"learning_rate": 1.2221052631578948e-06,
"loss": 0.44,
"step": 2260
},
{
"epoch": 0.11,
"grad_norm": 4.03125,
"learning_rate": 1.2205263157894737e-06,
"loss": 0.4954,
"step": 2270
},
{
"epoch": 0.11,
"grad_norm": 6.53125,
"learning_rate": 1.2189473684210526e-06,
"loss": 0.4992,
"step": 2280
},
{
"epoch": 0.11,
"grad_norm": 4.0625,
"learning_rate": 1.2173684210526315e-06,
"loss": 0.4659,
"step": 2290
},
{
"epoch": 0.12,
"grad_norm": 4.9375,
"learning_rate": 1.2157894736842105e-06,
"loss": 0.4878,
"step": 2300
},
{
"epoch": 0.12,
"grad_norm": 5.03125,
"learning_rate": 1.2142105263157896e-06,
"loss": 0.4959,
"step": 2310
},
{
"epoch": 0.12,
"grad_norm": 5.03125,
"learning_rate": 1.2126315789473685e-06,
"loss": 0.5432,
"step": 2320
},
{
"epoch": 0.12,
"grad_norm": 4.3125,
"learning_rate": 1.2110526315789474e-06,
"loss": 0.5791,
"step": 2330
},
{
"epoch": 0.12,
"grad_norm": 6.40625,
"learning_rate": 1.2094736842105263e-06,
"loss": 0.4737,
"step": 2340
},
{
"epoch": 0.12,
"grad_norm": 3.53125,
"learning_rate": 1.2078947368421052e-06,
"loss": 0.457,
"step": 2350
},
{
"epoch": 0.12,
"grad_norm": 6.09375,
"learning_rate": 1.2063157894736843e-06,
"loss": 0.4794,
"step": 2360
},
{
"epoch": 0.12,
"grad_norm": 3.5,
"learning_rate": 1.2047368421052632e-06,
"loss": 0.4365,
"step": 2370
},
{
"epoch": 0.12,
"grad_norm": 4.9375,
"learning_rate": 1.2031578947368421e-06,
"loss": 0.3707,
"step": 2380
},
{
"epoch": 0.12,
"grad_norm": 5.90625,
"learning_rate": 1.201578947368421e-06,
"loss": 0.4804,
"step": 2390
},
{
"epoch": 0.12,
"grad_norm": 4.65625,
"learning_rate": 1.2000000000000002e-06,
"loss": 0.4923,
"step": 2400
},
{
"epoch": 0.12,
"grad_norm": 5.6875,
"learning_rate": 1.198421052631579e-06,
"loss": 0.5911,
"step": 2410
},
{
"epoch": 0.12,
"grad_norm": 5.40625,
"learning_rate": 1.1968421052631578e-06,
"loss": 0.4133,
"step": 2420
},
{
"epoch": 0.12,
"grad_norm": 5.71875,
"learning_rate": 1.1952631578947369e-06,
"loss": 0.5136,
"step": 2430
},
{
"epoch": 0.12,
"grad_norm": 5.5625,
"learning_rate": 1.1936842105263158e-06,
"loss": 0.4695,
"step": 2440
},
{
"epoch": 0.12,
"grad_norm": 5.125,
"learning_rate": 1.192105263157895e-06,
"loss": 0.4341,
"step": 2450
},
{
"epoch": 0.12,
"grad_norm": 5.0625,
"learning_rate": 1.1905263157894736e-06,
"loss": 0.5027,
"step": 2460
},
{
"epoch": 0.12,
"grad_norm": 5.0625,
"learning_rate": 1.1889473684210527e-06,
"loss": 0.5261,
"step": 2470
},
{
"epoch": 0.12,
"grad_norm": 6.125,
"learning_rate": 1.1873684210526316e-06,
"loss": 0.5187,
"step": 2480
},
{
"epoch": 0.12,
"grad_norm": 5.5625,
"learning_rate": 1.1857894736842105e-06,
"loss": 0.4773,
"step": 2490
},
{
"epoch": 0.12,
"grad_norm": 4.90625,
"learning_rate": 1.1842105263157894e-06,
"loss": 0.5043,
"step": 2500
},
{
"epoch": 0.13,
"grad_norm": 4.8125,
"learning_rate": 1.1826315789473683e-06,
"loss": 0.4606,
"step": 2510
},
{
"epoch": 0.13,
"grad_norm": 6.8125,
"learning_rate": 1.1810526315789475e-06,
"loss": 0.4696,
"step": 2520
},
{
"epoch": 0.13,
"grad_norm": 5.40625,
"learning_rate": 1.1794736842105264e-06,
"loss": 0.4454,
"step": 2530
},
{
"epoch": 0.13,
"grad_norm": 6.375,
"learning_rate": 1.1778947368421053e-06,
"loss": 0.4707,
"step": 2540
},
{
"epoch": 0.13,
"grad_norm": 5.0,
"learning_rate": 1.1763157894736842e-06,
"loss": 0.4945,
"step": 2550
},
{
"epoch": 0.13,
"grad_norm": 5.28125,
"learning_rate": 1.174736842105263e-06,
"loss": 0.5033,
"step": 2560
},
{
"epoch": 0.13,
"grad_norm": 5.75,
"learning_rate": 1.1731578947368422e-06,
"loss": 0.5153,
"step": 2570
},
{
"epoch": 0.13,
"grad_norm": 4.6875,
"learning_rate": 1.1715789473684211e-06,
"loss": 0.4832,
"step": 2580
},
{
"epoch": 0.13,
"grad_norm": 5.5625,
"learning_rate": 1.17e-06,
"loss": 0.4266,
"step": 2590
},
{
"epoch": 0.13,
"grad_norm": 5.4375,
"learning_rate": 1.168421052631579e-06,
"loss": 0.4664,
"step": 2600
},
{
"epoch": 0.13,
"grad_norm": 6.0,
"learning_rate": 1.166842105263158e-06,
"loss": 0.4107,
"step": 2610
},
{
"epoch": 0.13,
"grad_norm": 5.09375,
"learning_rate": 1.165263157894737e-06,
"loss": 0.5131,
"step": 2620
},
{
"epoch": 0.13,
"grad_norm": 5.75,
"learning_rate": 1.1636842105263156e-06,
"loss": 0.4338,
"step": 2630
},
{
"epoch": 0.13,
"grad_norm": 3.078125,
"learning_rate": 1.1621052631578948e-06,
"loss": 0.4013,
"step": 2640
},
{
"epoch": 0.13,
"grad_norm": 5.71875,
"learning_rate": 1.1605263157894737e-06,
"loss": 0.4419,
"step": 2650
},
{
"epoch": 0.13,
"grad_norm": 4.875,
"learning_rate": 1.1589473684210528e-06,
"loss": 0.5262,
"step": 2660
},
{
"epoch": 0.13,
"grad_norm": 4.0625,
"learning_rate": 1.1573684210526315e-06,
"loss": 0.4451,
"step": 2670
},
{
"epoch": 0.13,
"grad_norm": 4.5625,
"learning_rate": 1.1557894736842106e-06,
"loss": 0.536,
"step": 2680
},
{
"epoch": 0.13,
"grad_norm": 4.65625,
"learning_rate": 1.1542105263157895e-06,
"loss": 0.4086,
"step": 2690
},
{
"epoch": 0.14,
"grad_norm": 4.6875,
"learning_rate": 1.1526315789473684e-06,
"loss": 0.4409,
"step": 2700
},
{
"epoch": 0.14,
"grad_norm": 4.21875,
"learning_rate": 1.1510526315789473e-06,
"loss": 0.4708,
"step": 2710
},
{
"epoch": 0.14,
"grad_norm": 4.4375,
"learning_rate": 1.1494736842105262e-06,
"loss": 0.6023,
"step": 2720
},
{
"epoch": 0.14,
"grad_norm": 5.90625,
"learning_rate": 1.1478947368421054e-06,
"loss": 0.5468,
"step": 2730
},
{
"epoch": 0.14,
"grad_norm": 4.78125,
"learning_rate": 1.1463157894736843e-06,
"loss": 0.4219,
"step": 2740
},
{
"epoch": 0.14,
"grad_norm": 5.6875,
"learning_rate": 1.1447368421052632e-06,
"loss": 0.3978,
"step": 2750
},
{
"epoch": 0.14,
"grad_norm": 6.5625,
"learning_rate": 1.143157894736842e-06,
"loss": 0.4987,
"step": 2760
},
{
"epoch": 0.14,
"grad_norm": 4.75,
"learning_rate": 1.141578947368421e-06,
"loss": 0.5023,
"step": 2770
},
{
"epoch": 0.14,
"grad_norm": 6.25,
"learning_rate": 1.14e-06,
"loss": 0.5176,
"step": 2780
},
{
"epoch": 0.14,
"grad_norm": 4.59375,
"learning_rate": 1.138421052631579e-06,
"loss": 0.5293,
"step": 2790
},
{
"epoch": 0.14,
"grad_norm": 5.40625,
"learning_rate": 1.136842105263158e-06,
"loss": 0.404,
"step": 2800
},
{
"epoch": 0.14,
"grad_norm": 4.75,
"learning_rate": 1.1352631578947368e-06,
"loss": 0.4729,
"step": 2810
},
{
"epoch": 0.14,
"grad_norm": 6.59375,
"learning_rate": 1.133684210526316e-06,
"loss": 0.3948,
"step": 2820
},
{
"epoch": 0.14,
"grad_norm": 4.875,
"learning_rate": 1.1321052631578948e-06,
"loss": 0.5348,
"step": 2830
},
{
"epoch": 0.14,
"grad_norm": 5.875,
"learning_rate": 1.1305263157894735e-06,
"loss": 0.4634,
"step": 2840
},
{
"epoch": 0.14,
"grad_norm": 5.28125,
"learning_rate": 1.1289473684210527e-06,
"loss": 0.4997,
"step": 2850
},
{
"epoch": 0.14,
"grad_norm": 4.59375,
"learning_rate": 1.1273684210526316e-06,
"loss": 0.4732,
"step": 2860
},
{
"epoch": 0.14,
"grad_norm": 5.34375,
"learning_rate": 1.1257894736842107e-06,
"loss": 0.5044,
"step": 2870
},
{
"epoch": 0.14,
"grad_norm": 4.34375,
"learning_rate": 1.1242105263157894e-06,
"loss": 0.5032,
"step": 2880
},
{
"epoch": 0.14,
"grad_norm": 6.15625,
"learning_rate": 1.1226315789473685e-06,
"loss": 0.4681,
"step": 2890
},
{
"epoch": 0.14,
"grad_norm": 8.875,
"learning_rate": 1.1210526315789474e-06,
"loss": 0.4979,
"step": 2900
},
{
"epoch": 0.15,
"grad_norm": 5.09375,
"learning_rate": 1.1194736842105265e-06,
"loss": 0.54,
"step": 2910
},
{
"epoch": 0.15,
"grad_norm": 4.46875,
"learning_rate": 1.1178947368421052e-06,
"loss": 0.4824,
"step": 2920
},
{
"epoch": 0.15,
"grad_norm": 5.59375,
"learning_rate": 1.1163157894736841e-06,
"loss": 0.5232,
"step": 2930
},
{
"epoch": 0.15,
"grad_norm": 5.84375,
"learning_rate": 1.1147368421052632e-06,
"loss": 0.4755,
"step": 2940
},
{
"epoch": 0.15,
"grad_norm": 4.5,
"learning_rate": 1.1131578947368421e-06,
"loss": 0.4978,
"step": 2950
},
{
"epoch": 0.15,
"grad_norm": 4.09375,
"learning_rate": 1.111578947368421e-06,
"loss": 0.5024,
"step": 2960
},
{
"epoch": 0.15,
"grad_norm": 4.53125,
"learning_rate": 1.11e-06,
"loss": 0.4711,
"step": 2970
},
{
"epoch": 0.15,
"grad_norm": 5.5625,
"learning_rate": 1.108421052631579e-06,
"loss": 0.4807,
"step": 2980
},
{
"epoch": 0.15,
"grad_norm": 4.53125,
"learning_rate": 1.106842105263158e-06,
"loss": 0.4121,
"step": 2990
},
{
"epoch": 0.15,
"grad_norm": 3.90625,
"learning_rate": 1.1052631578947369e-06,
"loss": 0.4936,
"step": 3000
},
{
"epoch": 0.15,
"grad_norm": 5.46875,
"learning_rate": 1.1036842105263158e-06,
"loss": 0.3976,
"step": 3010
},
{
"epoch": 0.15,
"grad_norm": 4.34375,
"learning_rate": 1.1021052631578947e-06,
"loss": 0.4965,
"step": 3020
},
{
"epoch": 0.15,
"grad_norm": 6.46875,
"learning_rate": 1.1005263157894738e-06,
"loss": 0.5458,
"step": 3030
},
{
"epoch": 0.15,
"grad_norm": 4.84375,
"learning_rate": 1.0989473684210527e-06,
"loss": 0.4657,
"step": 3040
},
{
"epoch": 0.15,
"grad_norm": 6.3125,
"learning_rate": 1.0973684210526316e-06,
"loss": 0.4838,
"step": 3050
},
{
"epoch": 0.15,
"grad_norm": 4.96875,
"learning_rate": 1.0957894736842105e-06,
"loss": 0.5324,
"step": 3060
},
{
"epoch": 0.15,
"grad_norm": 4.53125,
"learning_rate": 1.0942105263157895e-06,
"loss": 0.4857,
"step": 3070
},
{
"epoch": 0.15,
"grad_norm": 5.3125,
"learning_rate": 1.0926315789473686e-06,
"loss": 0.4866,
"step": 3080
},
{
"epoch": 0.15,
"grad_norm": 6.15625,
"learning_rate": 1.0910526315789473e-06,
"loss": 0.5277,
"step": 3090
},
{
"epoch": 0.15,
"grad_norm": 4.9375,
"learning_rate": 1.0894736842105264e-06,
"loss": 0.5184,
"step": 3100
},
{
"epoch": 0.16,
"grad_norm": 3.890625,
"learning_rate": 1.0878947368421053e-06,
"loss": 0.4704,
"step": 3110
},
{
"epoch": 0.16,
"grad_norm": 4.5625,
"learning_rate": 1.0863157894736844e-06,
"loss": 0.4638,
"step": 3120
},
{
"epoch": 0.16,
"grad_norm": 4.46875,
"learning_rate": 1.084736842105263e-06,
"loss": 0.409,
"step": 3130
},
{
"epoch": 0.16,
"grad_norm": 4.78125,
"learning_rate": 1.083157894736842e-06,
"loss": 0.4403,
"step": 3140
},
{
"epoch": 0.16,
"grad_norm": 7.34375,
"learning_rate": 1.0815789473684211e-06,
"loss": 0.4339,
"step": 3150
},
{
"epoch": 0.16,
"grad_norm": 5.0625,
"learning_rate": 1.08e-06,
"loss": 0.447,
"step": 3160
},
{
"epoch": 0.16,
"grad_norm": 8.25,
"learning_rate": 1.078421052631579e-06,
"loss": 0.6107,
"step": 3170
},
{
"epoch": 0.16,
"grad_norm": 3.90625,
"learning_rate": 1.0768421052631578e-06,
"loss": 0.5194,
"step": 3180
},
{
"epoch": 0.16,
"grad_norm": 5.6875,
"learning_rate": 1.075263157894737e-06,
"loss": 0.4548,
"step": 3190
},
{
"epoch": 0.16,
"grad_norm": 4.25,
"learning_rate": 1.0736842105263159e-06,
"loss": 0.5024,
"step": 3200
},
{
"epoch": 0.16,
"grad_norm": 4.90625,
"learning_rate": 1.0721052631578948e-06,
"loss": 0.5341,
"step": 3210
},
{
"epoch": 0.16,
"grad_norm": 5.71875,
"learning_rate": 1.0705263157894737e-06,
"loss": 0.5375,
"step": 3220
},
{
"epoch": 0.16,
"grad_norm": 4.59375,
"learning_rate": 1.0689473684210526e-06,
"loss": 0.5262,
"step": 3230
},
{
"epoch": 0.16,
"grad_norm": 6.59375,
"learning_rate": 1.0673684210526317e-06,
"loss": 0.4006,
"step": 3240
},
{
"epoch": 0.16,
"grad_norm": 5.53125,
"learning_rate": 1.0657894736842106e-06,
"loss": 0.4687,
"step": 3250
},
{
"epoch": 0.16,
"grad_norm": 5.96875,
"learning_rate": 1.0642105263157895e-06,
"loss": 0.5541,
"step": 3260
},
{
"epoch": 0.16,
"grad_norm": 3.703125,
"learning_rate": 1.0626315789473684e-06,
"loss": 0.4609,
"step": 3270
},
{
"epoch": 0.16,
"grad_norm": 5.8125,
"learning_rate": 1.0610526315789473e-06,
"loss": 0.5103,
"step": 3280
},
{
"epoch": 0.16,
"grad_norm": 4.375,
"learning_rate": 1.0594736842105265e-06,
"loss": 0.4879,
"step": 3290
},
{
"epoch": 0.17,
"grad_norm": 4.34375,
"learning_rate": 1.0578947368421052e-06,
"loss": 0.3878,
"step": 3300
},
{
"epoch": 0.17,
"grad_norm": 4.71875,
"learning_rate": 1.0563157894736843e-06,
"loss": 0.5275,
"step": 3310
},
{
"epoch": 0.17,
"grad_norm": 3.65625,
"learning_rate": 1.0547368421052632e-06,
"loss": 0.5397,
"step": 3320
},
{
"epoch": 0.17,
"grad_norm": 5.53125,
"learning_rate": 1.0531578947368423e-06,
"loss": 0.5235,
"step": 3330
},
{
"epoch": 0.17,
"grad_norm": 5.125,
"learning_rate": 1.051578947368421e-06,
"loss": 0.5161,
"step": 3340
},
{
"epoch": 0.17,
"grad_norm": 4.125,
"learning_rate": 1.05e-06,
"loss": 0.4263,
"step": 3350
},
{
"epoch": 0.17,
"grad_norm": 4.28125,
"learning_rate": 1.048421052631579e-06,
"loss": 0.4468,
"step": 3360
},
{
"epoch": 0.17,
"grad_norm": 4.40625,
"learning_rate": 1.046842105263158e-06,
"loss": 0.4121,
"step": 3370
},
{
"epoch": 0.17,
"grad_norm": 4.78125,
"learning_rate": 1.0452631578947368e-06,
"loss": 0.3873,
"step": 3380
},
{
"epoch": 0.17,
"grad_norm": 5.0625,
"learning_rate": 1.0436842105263157e-06,
"loss": 0.4881,
"step": 3390
},
{
"epoch": 0.17,
"grad_norm": 3.5,
"learning_rate": 1.0421052631578949e-06,
"loss": 0.5261,
"step": 3400
},
{
"epoch": 0.17,
"grad_norm": 4.09375,
"learning_rate": 1.0405263157894738e-06,
"loss": 0.3614,
"step": 3410
},
{
"epoch": 0.17,
"grad_norm": 6.78125,
"learning_rate": 1.0389473684210527e-06,
"loss": 0.4622,
"step": 3420
},
{
"epoch": 0.17,
"grad_norm": 4.625,
"learning_rate": 1.0373684210526316e-06,
"loss": 0.3961,
"step": 3430
},
{
"epoch": 0.17,
"grad_norm": 4.5,
"learning_rate": 1.0357894736842105e-06,
"loss": 0.4891,
"step": 3440
},
{
"epoch": 0.17,
"grad_norm": 4.75,
"learning_rate": 1.0342105263157896e-06,
"loss": 0.4674,
"step": 3450
},
{
"epoch": 0.17,
"grad_norm": 3.71875,
"learning_rate": 1.0326315789473685e-06,
"loss": 0.4592,
"step": 3460
},
{
"epoch": 0.17,
"grad_norm": 6.65625,
"learning_rate": 1.0310526315789474e-06,
"loss": 0.4507,
"step": 3470
},
{
"epoch": 0.17,
"grad_norm": 4.15625,
"learning_rate": 1.0294736842105263e-06,
"loss": 0.425,
"step": 3480
},
{
"epoch": 0.17,
"grad_norm": 5.28125,
"learning_rate": 1.0278947368421052e-06,
"loss": 0.3396,
"step": 3490
},
{
"epoch": 0.17,
"grad_norm": 5.53125,
"learning_rate": 1.0263157894736843e-06,
"loss": 0.5464,
"step": 3500
},
{
"epoch": 0.18,
"grad_norm": 5.46875,
"learning_rate": 1.024736842105263e-06,
"loss": 0.5085,
"step": 3510
},
{
"epoch": 0.18,
"grad_norm": 4.875,
"learning_rate": 1.0231578947368422e-06,
"loss": 0.487,
"step": 3520
},
{
"epoch": 0.18,
"grad_norm": 6.5625,
"learning_rate": 1.021578947368421e-06,
"loss": 0.5404,
"step": 3530
},
{
"epoch": 0.18,
"grad_norm": 5.40625,
"learning_rate": 1.0200000000000002e-06,
"loss": 0.4775,
"step": 3540
},
{
"epoch": 0.18,
"grad_norm": 5.6875,
"learning_rate": 1.0184210526315789e-06,
"loss": 0.4434,
"step": 3550
},
{
"epoch": 0.18,
"grad_norm": 4.9375,
"learning_rate": 1.0168421052631578e-06,
"loss": 0.4978,
"step": 3560
},
{
"epoch": 0.18,
"grad_norm": 5.28125,
"learning_rate": 1.015263157894737e-06,
"loss": 0.4715,
"step": 3570
},
{
"epoch": 0.18,
"grad_norm": 4.90625,
"learning_rate": 1.0136842105263158e-06,
"loss": 0.4897,
"step": 3580
},
{
"epoch": 0.18,
"grad_norm": 8.3125,
"learning_rate": 1.0121052631578947e-06,
"loss": 0.58,
"step": 3590
},
{
"epoch": 0.18,
"grad_norm": 4.21875,
"learning_rate": 1.0105263157894736e-06,
"loss": 0.5142,
"step": 3600
},
{
"epoch": 0.18,
"grad_norm": 5.28125,
"learning_rate": 1.0089473684210527e-06,
"loss": 0.4952,
"step": 3610
},
{
"epoch": 0.18,
"grad_norm": 6.1875,
"learning_rate": 1.0073684210526317e-06,
"loss": 0.592,
"step": 3620
},
{
"epoch": 0.18,
"grad_norm": 5.40625,
"learning_rate": 1.0057894736842106e-06,
"loss": 0.4829,
"step": 3630
},
{
"epoch": 0.18,
"grad_norm": 4.4375,
"learning_rate": 1.0042105263157895e-06,
"loss": 0.5296,
"step": 3640
},
{
"epoch": 0.18,
"grad_norm": 4.1875,
"learning_rate": 1.0026315789473684e-06,
"loss": 0.446,
"step": 3650
},
{
"epoch": 0.18,
"grad_norm": 4.1875,
"learning_rate": 1.0010526315789475e-06,
"loss": 0.4925,
"step": 3660
},
{
"epoch": 0.18,
"grad_norm": 5.78125,
"learning_rate": 9.994736842105264e-07,
"loss": 0.5038,
"step": 3670
},
{
"epoch": 0.18,
"grad_norm": 5.34375,
"learning_rate": 9.978947368421053e-07,
"loss": 0.4442,
"step": 3680
},
{
"epoch": 0.18,
"grad_norm": 4.0,
"learning_rate": 9.963157894736842e-07,
"loss": 0.4393,
"step": 3690
},
{
"epoch": 0.18,
"grad_norm": 3.75,
"learning_rate": 9.947368421052631e-07,
"loss": 0.4791,
"step": 3700
},
{
"epoch": 0.19,
"grad_norm": 4.53125,
"learning_rate": 9.931578947368422e-07,
"loss": 0.4682,
"step": 3710
},
{
"epoch": 0.19,
"grad_norm": 7.5625,
"learning_rate": 9.91578947368421e-07,
"loss": 0.4932,
"step": 3720
},
{
"epoch": 0.19,
"grad_norm": 3.96875,
"learning_rate": 9.9e-07,
"loss": 0.4941,
"step": 3730
},
{
"epoch": 0.19,
"grad_norm": 4.46875,
"learning_rate": 9.88421052631579e-07,
"loss": 0.4873,
"step": 3740
},
{
"epoch": 0.19,
"grad_norm": 3.859375,
"learning_rate": 9.86842105263158e-07,
"loss": 0.4924,
"step": 3750
},
{
"epoch": 0.19,
"grad_norm": 5.5625,
"learning_rate": 9.852631578947368e-07,
"loss": 0.5725,
"step": 3760
},
{
"epoch": 0.19,
"grad_norm": 5.96875,
"learning_rate": 9.836842105263157e-07,
"loss": 0.5883,
"step": 3770
},
{
"epoch": 0.19,
"grad_norm": 6.03125,
"learning_rate": 9.821052631578948e-07,
"loss": 0.5129,
"step": 3780
},
{
"epoch": 0.19,
"grad_norm": 3.96875,
"learning_rate": 9.805263157894737e-07,
"loss": 0.3405,
"step": 3790
},
{
"epoch": 0.19,
"grad_norm": 5.6875,
"learning_rate": 9.789473684210526e-07,
"loss": 0.4876,
"step": 3800
},
{
"epoch": 0.19,
"grad_norm": 3.90625,
"learning_rate": 9.773684210526315e-07,
"loss": 0.3936,
"step": 3810
},
{
"epoch": 0.19,
"grad_norm": 5.28125,
"learning_rate": 9.757894736842106e-07,
"loss": 0.5705,
"step": 3820
},
{
"epoch": 0.19,
"grad_norm": 5.5625,
"learning_rate": 9.742105263157895e-07,
"loss": 0.4709,
"step": 3830
},
{
"epoch": 0.19,
"grad_norm": 5.25,
"learning_rate": 9.726315789473684e-07,
"loss": 0.4787,
"step": 3840
},
{
"epoch": 0.19,
"grad_norm": 5.71875,
"learning_rate": 9.710526315789474e-07,
"loss": 0.4842,
"step": 3850
},
{
"epoch": 0.19,
"grad_norm": 4.4375,
"learning_rate": 9.694736842105263e-07,
"loss": 0.5223,
"step": 3860
},
{
"epoch": 0.19,
"grad_norm": 5.40625,
"learning_rate": 9.678947368421054e-07,
"loss": 0.4831,
"step": 3870
},
{
"epoch": 0.19,
"grad_norm": 5.5625,
"learning_rate": 9.663157894736843e-07,
"loss": 0.4824,
"step": 3880
},
{
"epoch": 0.19,
"grad_norm": 4.875,
"learning_rate": 9.647368421052632e-07,
"loss": 0.4704,
"step": 3890
},
{
"epoch": 0.2,
"grad_norm": 5.4375,
"learning_rate": 9.63157894736842e-07,
"loss": 0.4286,
"step": 3900
},
{
"epoch": 0.2,
"grad_norm": 4.75,
"learning_rate": 9.61578947368421e-07,
"loss": 0.4933,
"step": 3910
},
{
"epoch": 0.2,
"grad_norm": 5.53125,
"learning_rate": 9.600000000000001e-07,
"loss": 0.5162,
"step": 3920
},
{
"epoch": 0.2,
"grad_norm": 5.96875,
"learning_rate": 9.584210526315788e-07,
"loss": 0.4726,
"step": 3930
},
{
"epoch": 0.2,
"grad_norm": 5.25,
"learning_rate": 9.56842105263158e-07,
"loss": 0.4791,
"step": 3940
},
{
"epoch": 0.2,
"grad_norm": 5.03125,
"learning_rate": 9.552631578947368e-07,
"loss": 0.4233,
"step": 3950
},
{
"epoch": 0.2,
"grad_norm": 5.96875,
"learning_rate": 9.53684210526316e-07,
"loss": 0.5278,
"step": 3960
},
{
"epoch": 0.2,
"grad_norm": 4.75,
"learning_rate": 9.521052631578948e-07,
"loss": 0.5032,
"step": 3970
},
{
"epoch": 0.2,
"grad_norm": 5.09375,
"learning_rate": 9.505263157894737e-07,
"loss": 0.485,
"step": 3980
},
{
"epoch": 0.2,
"grad_norm": 5.0,
"learning_rate": 9.489473684210527e-07,
"loss": 0.502,
"step": 3990
},
{
"epoch": 0.2,
"grad_norm": 6.125,
"learning_rate": 9.473684210526316e-07,
"loss": 0.4658,
"step": 4000
},
{
"epoch": 0.2,
"eval_loss": 0.48478230834007263,
"eval_runtime": 64.7694,
"eval_samples_per_second": 15.439,
"eval_steps_per_second": 15.439,
"step": 4000
}
],
"logging_steps": 10,
"max_steps": 10000,
"num_input_tokens_seen": 0,
"num_train_epochs": 1,
"save_steps": 2000,
"total_flos": 6.455688167424e+16,
"train_batch_size": 1,
"trial_name": null,
"trial_params": null
}