Files
gkd-qwen-2.5-0.5b-base_v4_f…/checkpoint-2625/trainer_state.json

1869 lines
42 KiB
JSON
Raw Normal View History

{
"best_global_step": null,
"best_metric": null,
"best_model_checkpoint": null,
"epoch": 3.0,
"eval_steps": 500,
"global_step": 2625,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 0.011428571428571429,
"grad_norm": 8.8125,
"learning_rate": 1.9931428571428572e-05,
"loss": 2.0244,
"step": 10
},
{
"epoch": 0.022857142857142857,
"grad_norm": 12.4375,
"learning_rate": 1.9855238095238097e-05,
"loss": 2.0268,
"step": 20
},
{
"epoch": 0.03428571428571429,
"grad_norm": 5.6875,
"learning_rate": 1.977904761904762e-05,
"loss": 2.0396,
"step": 30
},
{
"epoch": 0.045714285714285714,
"grad_norm": 6.03125,
"learning_rate": 1.9702857142857144e-05,
"loss": 1.8937,
"step": 40
},
{
"epoch": 0.05714285714285714,
"grad_norm": 10.0,
"learning_rate": 1.9626666666666666e-05,
"loss": 1.8841,
"step": 50
},
{
"epoch": 0.06857142857142857,
"grad_norm": 8.625,
"learning_rate": 1.955047619047619e-05,
"loss": 1.8955,
"step": 60
},
{
"epoch": 0.08,
"grad_norm": 8.1875,
"learning_rate": 1.9474285714285717e-05,
"loss": 1.8829,
"step": 70
},
{
"epoch": 0.09142857142857143,
"grad_norm": 5.5625,
"learning_rate": 1.9398095238095242e-05,
"loss": 1.9389,
"step": 80
},
{
"epoch": 0.10285714285714286,
"grad_norm": 5.625,
"learning_rate": 1.9321904761904764e-05,
"loss": 1.9335,
"step": 90
},
{
"epoch": 0.11428571428571428,
"grad_norm": 11.4375,
"learning_rate": 1.924571428571429e-05,
"loss": 1.8842,
"step": 100
},
{
"epoch": 0.12571428571428572,
"grad_norm": 6.03125,
"learning_rate": 1.916952380952381e-05,
"loss": 1.9089,
"step": 110
},
{
"epoch": 0.13714285714285715,
"grad_norm": 9.625,
"learning_rate": 1.9093333333333336e-05,
"loss": 1.7602,
"step": 120
},
{
"epoch": 0.14857142857142858,
"grad_norm": 9.4375,
"learning_rate": 1.9017142857142858e-05,
"loss": 1.721,
"step": 130
},
{
"epoch": 0.16,
"grad_norm": 10.375,
"learning_rate": 1.8940952380952383e-05,
"loss": 1.8059,
"step": 140
},
{
"epoch": 0.17142857142857143,
"grad_norm": 9.0625,
"learning_rate": 1.8864761904761905e-05,
"loss": 1.7363,
"step": 150
},
{
"epoch": 0.18285714285714286,
"grad_norm": 11.0625,
"learning_rate": 1.878857142857143e-05,
"loss": 1.7634,
"step": 160
},
{
"epoch": 0.19428571428571428,
"grad_norm": 5.46875,
"learning_rate": 1.871238095238095e-05,
"loss": 1.7298,
"step": 170
},
{
"epoch": 0.2057142857142857,
"grad_norm": 5.375,
"learning_rate": 1.8636190476190477e-05,
"loss": 1.7439,
"step": 180
},
{
"epoch": 0.21714285714285714,
"grad_norm": 8.125,
"learning_rate": 1.8560000000000002e-05,
"loss": 1.8381,
"step": 190
},
{
"epoch": 0.22857142857142856,
"grad_norm": 7.65625,
"learning_rate": 1.8483809523809527e-05,
"loss": 1.6778,
"step": 200
},
{
"epoch": 0.24,
"grad_norm": 9.25,
"learning_rate": 1.840761904761905e-05,
"loss": 1.6378,
"step": 210
},
{
"epoch": 0.25142857142857145,
"grad_norm": 8.25,
"learning_rate": 1.8331428571428574e-05,
"loss": 1.6447,
"step": 220
},
{
"epoch": 0.26285714285714284,
"grad_norm": 10.9375,
"learning_rate": 1.8255238095238096e-05,
"loss": 1.7685,
"step": 230
},
{
"epoch": 0.2742857142857143,
"grad_norm": 11.0,
"learning_rate": 1.817904761904762e-05,
"loss": 1.6609,
"step": 240
},
{
"epoch": 0.2857142857142857,
"grad_norm": 4.75,
"learning_rate": 1.8102857142857143e-05,
"loss": 1.6822,
"step": 250
},
{
"epoch": 0.29714285714285715,
"grad_norm": 7.28125,
"learning_rate": 1.8026666666666668e-05,
"loss": 1.6153,
"step": 260
},
{
"epoch": 0.30857142857142855,
"grad_norm": 7.9375,
"learning_rate": 1.7950476190476193e-05,
"loss": 1.7153,
"step": 270
},
{
"epoch": 0.32,
"grad_norm": 9.0625,
"learning_rate": 1.7874285714285715e-05,
"loss": 1.7104,
"step": 280
},
{
"epoch": 0.3314285714285714,
"grad_norm": 7.90625,
"learning_rate": 1.779809523809524e-05,
"loss": 1.6752,
"step": 290
},
{
"epoch": 0.34285714285714286,
"grad_norm": 7.03125,
"learning_rate": 1.7721904761904762e-05,
"loss": 1.6701,
"step": 300
},
{
"epoch": 0.35428571428571426,
"grad_norm": 7.5625,
"learning_rate": 1.7645714285714287e-05,
"loss": 1.6393,
"step": 310
},
{
"epoch": 0.3657142857142857,
"grad_norm": 7.65625,
"learning_rate": 1.7569523809523812e-05,
"loss": 1.5169,
"step": 320
},
{
"epoch": 0.37714285714285717,
"grad_norm": 4.6875,
"learning_rate": 1.7493333333333334e-05,
"loss": 1.5707,
"step": 330
},
{
"epoch": 0.38857142857142857,
"grad_norm": 4.78125,
"learning_rate": 1.741714285714286e-05,
"loss": 1.5962,
"step": 340
},
{
"epoch": 0.4,
"grad_norm": 7.59375,
"learning_rate": 1.734095238095238e-05,
"loss": 1.7404,
"step": 350
},
{
"epoch": 0.4114285714285714,
"grad_norm": 5.21875,
"learning_rate": 1.7264761904761906e-05,
"loss": 1.6232,
"step": 360
},
{
"epoch": 0.4228571428571429,
"grad_norm": 4.46875,
"learning_rate": 1.718857142857143e-05,
"loss": 1.7321,
"step": 370
},
{
"epoch": 0.4342857142857143,
"grad_norm": 4.40625,
"learning_rate": 1.7112380952380953e-05,
"loss": 1.5638,
"step": 380
},
{
"epoch": 0.44571428571428573,
"grad_norm": 4.625,
"learning_rate": 1.703619047619048e-05,
"loss": 1.6591,
"step": 390
},
{
"epoch": 0.45714285714285713,
"grad_norm": 9.0625,
"learning_rate": 1.696e-05,
"loss": 1.5868,
"step": 400
},
{
"epoch": 0.4685714285714286,
"grad_norm": 9.125,
"learning_rate": 1.6883809523809525e-05,
"loss": 1.6323,
"step": 410
},
{
"epoch": 0.48,
"grad_norm": 7.5,
"learning_rate": 1.6807619047619047e-05,
"loss": 1.6079,
"step": 420
},
{
"epoch": 0.49142857142857144,
"grad_norm": 9.9375,
"learning_rate": 1.6731428571428572e-05,
"loss": 1.6279,
"step": 430
},
{
"epoch": 0.5028571428571429,
"grad_norm": 4.90625,
"learning_rate": 1.6655238095238098e-05,
"loss": 1.6033,
"step": 440
},
{
"epoch": 0.5142857142857142,
"grad_norm": 9.375,
"learning_rate": 1.6579047619047623e-05,
"loss": 1.5813,
"step": 450
},
{
"epoch": 0.5257142857142857,
"grad_norm": 12.625,
"learning_rate": 1.6502857142857145e-05,
"loss": 1.7047,
"step": 460
},
{
"epoch": 0.5371428571428571,
"grad_norm": 7.75,
"learning_rate": 1.642666666666667e-05,
"loss": 1.6662,
"step": 470
},
{
"epoch": 0.5485714285714286,
"grad_norm": 7.03125,
"learning_rate": 1.635047619047619e-05,
"loss": 1.618,
"step": 480
},
{
"epoch": 0.56,
"grad_norm": 5.1875,
"learning_rate": 1.6274285714285717e-05,
"loss": 1.5469,
"step": 490
},
{
"epoch": 0.5714285714285714,
"grad_norm": 11.0625,
"learning_rate": 1.619809523809524e-05,
"loss": 1.5454,
"step": 500
},
{
"epoch": 0.5828571428571429,
"grad_norm": 7.84375,
"learning_rate": 1.6121904761904764e-05,
"loss": 1.4719,
"step": 510
},
{
"epoch": 0.5942857142857143,
"grad_norm": 4.53125,
"learning_rate": 1.6045714285714286e-05,
"loss": 1.4144,
"step": 520
},
{
"epoch": 0.6057142857142858,
"grad_norm": 8.125,
"learning_rate": 1.596952380952381e-05,
"loss": 1.5868,
"step": 530
},
{
"epoch": 0.6171428571428571,
"grad_norm": 6.9375,
"learning_rate": 1.5893333333333333e-05,
"loss": 1.5727,
"step": 540
},
{
"epoch": 0.6285714285714286,
"grad_norm": 9.3125,
"learning_rate": 1.5817142857142858e-05,
"loss": 1.6083,
"step": 550
},
{
"epoch": 0.64,
"grad_norm": 11.6875,
"learning_rate": 1.5740952380952383e-05,
"loss": 1.6169,
"step": 560
},
{
"epoch": 0.6514285714285715,
"grad_norm": 7.84375,
"learning_rate": 1.5664761904761908e-05,
"loss": 1.6161,
"step": 570
},
{
"epoch": 0.6628571428571428,
"grad_norm": 7.5,
"learning_rate": 1.558857142857143e-05,
"loss": 1.5572,
"step": 580
},
{
"epoch": 0.6742857142857143,
"grad_norm": 11.375,
"learning_rate": 1.5512380952380955e-05,
"loss": 1.4882,
"step": 590
},
{
"epoch": 0.6857142857142857,
"grad_norm": 7.59375,
"learning_rate": 1.5436190476190477e-05,
"loss": 1.6175,
"step": 600
},
{
"epoch": 0.6971428571428572,
"grad_norm": 4.59375,
"learning_rate": 1.5360000000000002e-05,
"loss": 1.5326,
"step": 610
},
{
"epoch": 0.7085714285714285,
"grad_norm": 4.3125,
"learning_rate": 1.5283809523809524e-05,
"loss": 1.5323,
"step": 620
},
{
"epoch": 0.72,
"grad_norm": 4.53125,
"learning_rate": 1.5207619047619049e-05,
"loss": 1.5665,
"step": 630
},
{
"epoch": 0.7314285714285714,
"grad_norm": 11.25,
"learning_rate": 1.5131428571428572e-05,
"loss": 1.4393,
"step": 640
},
{
"epoch": 0.7428571428571429,
"grad_norm": 7.03125,
"learning_rate": 1.5055238095238096e-05,
"loss": 1.528,
"step": 650
},
{
"epoch": 0.7542857142857143,
"grad_norm": 7.40625,
"learning_rate": 1.497904761904762e-05,
"loss": 1.5457,
"step": 660
},
{
"epoch": 0.7657142857142857,
"grad_norm": 9.4375,
"learning_rate": 1.4902857142857143e-05,
"loss": 1.5291,
"step": 670
},
{
"epoch": 0.7771428571428571,
"grad_norm": 9.375,
"learning_rate": 1.4826666666666666e-05,
"loss": 1.5229,
"step": 680
},
{
"epoch": 0.7885714285714286,
"grad_norm": 4.46875,
"learning_rate": 1.475047619047619e-05,
"loss": 1.4647,
"step": 690
},
{
"epoch": 0.8,
"grad_norm": 4.03125,
"learning_rate": 1.4674285714285717e-05,
"loss": 1.5565,
"step": 700
},
{
"epoch": 0.8114285714285714,
"grad_norm": 7.375,
"learning_rate": 1.459809523809524e-05,
"loss": 1.5298,
"step": 710
},
{
"epoch": 0.8228571428571428,
"grad_norm": 9.625,
"learning_rate": 1.4521904761904764e-05,
"loss": 1.5642,
"step": 720
},
{
"epoch": 0.8342857142857143,
"grad_norm": 7.4375,
"learning_rate": 1.4445714285714287e-05,
"loss": 1.5435,
"step": 730
},
{
"epoch": 0.8457142857142858,
"grad_norm": 12.375,
"learning_rate": 1.436952380952381e-05,
"loss": 1.4951,
"step": 740
},
{
"epoch": 0.8571428571428571,
"grad_norm": 7.9375,
"learning_rate": 1.4293333333333334e-05,
"loss": 1.5446,
"step": 750
},
{
"epoch": 0.8685714285714285,
"grad_norm": 7.21875,
"learning_rate": 1.4217142857142858e-05,
"loss": 1.5318,
"step": 760
},
{
"epoch": 0.88,
"grad_norm": 7.1875,
"learning_rate": 1.4140952380952381e-05,
"loss": 1.5279,
"step": 770
},
{
"epoch": 0.8914285714285715,
"grad_norm": 7.09375,
"learning_rate": 1.4064761904761905e-05,
"loss": 1.4712,
"step": 780
},
{
"epoch": 0.9028571428571428,
"grad_norm": 7.46875,
"learning_rate": 1.398857142857143e-05,
"loss": 1.566,
"step": 790
},
{
"epoch": 0.9142857142857143,
"grad_norm": 6.40625,
"learning_rate": 1.3912380952380953e-05,
"loss": 1.4607,
"step": 800
},
{
"epoch": 0.9257142857142857,
"grad_norm": 7.6875,
"learning_rate": 1.3836190476190477e-05,
"loss": 1.4935,
"step": 810
},
{
"epoch": 0.9371428571428572,
"grad_norm": 4.84375,
"learning_rate": 1.376e-05,
"loss": 1.4306,
"step": 820
},
{
"epoch": 0.9485714285714286,
"grad_norm": 9.75,
"learning_rate": 1.3683809523809526e-05,
"loss": 1.4454,
"step": 830
},
{
"epoch": 0.96,
"grad_norm": 7.5,
"learning_rate": 1.3607619047619049e-05,
"loss": 1.4285,
"step": 840
},
{
"epoch": 0.9714285714285714,
"grad_norm": 4.25,
"learning_rate": 1.3531428571428573e-05,
"loss": 1.5716,
"step": 850
},
{
"epoch": 0.9828571428571429,
"grad_norm": 10.8125,
"learning_rate": 1.3455238095238096e-05,
"loss": 1.5742,
"step": 860
},
{
"epoch": 0.9942857142857143,
"grad_norm": 4.375,
"learning_rate": 1.3379047619047621e-05,
"loss": 1.5958,
"step": 870
},
{
"epoch": 1.0057142857142858,
"grad_norm": 10.3125,
"learning_rate": 1.3302857142857145e-05,
"loss": 1.4893,
"step": 880
},
{
"epoch": 1.0171428571428571,
"grad_norm": 6.6875,
"learning_rate": 1.3226666666666668e-05,
"loss": 1.4789,
"step": 890
},
{
"epoch": 1.0285714285714285,
"grad_norm": 9.0625,
"learning_rate": 1.3150476190476192e-05,
"loss": 1.4813,
"step": 900
},
{
"epoch": 1.04,
"grad_norm": 7.21875,
"learning_rate": 1.3074285714285715e-05,
"loss": 1.4797,
"step": 910
},
{
"epoch": 1.0514285714285714,
"grad_norm": 7.21875,
"learning_rate": 1.2998095238095239e-05,
"loss": 1.5455,
"step": 920
},
{
"epoch": 1.062857142857143,
"grad_norm": 10.8125,
"learning_rate": 1.2921904761904762e-05,
"loss": 1.44,
"step": 930
},
{
"epoch": 1.0742857142857143,
"grad_norm": 4.21875,
"learning_rate": 1.2845714285714286e-05,
"loss": 1.53,
"step": 940
},
{
"epoch": 1.0857142857142856,
"grad_norm": 6.625,
"learning_rate": 1.2769523809523811e-05,
"loss": 1.4885,
"step": 950
},
{
"epoch": 1.0971428571428572,
"grad_norm": 9.8125,
"learning_rate": 1.2693333333333336e-05,
"loss": 1.4834,
"step": 960
},
{
"epoch": 1.1085714285714285,
"grad_norm": 7.59375,
"learning_rate": 1.261714285714286e-05,
"loss": 1.4658,
"step": 970
},
{
"epoch": 1.12,
"grad_norm": 9.9375,
"learning_rate": 1.2540952380952383e-05,
"loss": 1.5289,
"step": 980
},
{
"epoch": 1.1314285714285715,
"grad_norm": 4.5625,
"learning_rate": 1.2464761904761907e-05,
"loss": 1.5231,
"step": 990
},
{
"epoch": 1.1428571428571428,
"grad_norm": 4.34375,
"learning_rate": 1.238857142857143e-05,
"loss": 1.5437,
"step": 1000
},
{
"epoch": 1.1542857142857144,
"grad_norm": 8.125,
"learning_rate": 1.2312380952380953e-05,
"loss": 1.4484,
"step": 1010
},
{
"epoch": 1.1657142857142857,
"grad_norm": 7.5,
"learning_rate": 1.2236190476190477e-05,
"loss": 1.5001,
"step": 1020
},
{
"epoch": 1.177142857142857,
"grad_norm": 7.84375,
"learning_rate": 1.216e-05,
"loss": 1.5136,
"step": 1030
},
{
"epoch": 1.1885714285714286,
"grad_norm": 10.1875,
"learning_rate": 1.2083809523809524e-05,
"loss": 1.5088,
"step": 1040
},
{
"epoch": 1.2,
"grad_norm": 6.6875,
"learning_rate": 1.2007619047619047e-05,
"loss": 1.4638,
"step": 1050
},
{
"epoch": 1.2114285714285715,
"grad_norm": 4.15625,
"learning_rate": 1.1931428571428571e-05,
"loss": 1.535,
"step": 1060
},
{
"epoch": 1.2228571428571429,
"grad_norm": 4.625,
"learning_rate": 1.1855238095238094e-05,
"loss": 1.5093,
"step": 1070
},
{
"epoch": 1.2342857142857142,
"grad_norm": 9.5,
"learning_rate": 1.1779047619047621e-05,
"loss": 1.48,
"step": 1080
},
{
"epoch": 1.2457142857142858,
"grad_norm": 6.84375,
"learning_rate": 1.1702857142857145e-05,
"loss": 1.4696,
"step": 1090
},
{
"epoch": 1.2571428571428571,
"grad_norm": 7.5,
"learning_rate": 1.1626666666666668e-05,
"loss": 1.4727,
"step": 1100
},
{
"epoch": 1.2685714285714287,
"grad_norm": 6.59375,
"learning_rate": 1.1550476190476192e-05,
"loss": 1.5753,
"step": 1110
},
{
"epoch": 1.28,
"grad_norm": 8.125,
"learning_rate": 1.1474285714285715e-05,
"loss": 1.4507,
"step": 1120
},
{
"epoch": 1.2914285714285714,
"grad_norm": 3.859375,
"learning_rate": 1.1398095238095239e-05,
"loss": 1.3666,
"step": 1130
},
{
"epoch": 1.302857142857143,
"grad_norm": 6.25,
"learning_rate": 1.1321904761904762e-05,
"loss": 1.4768,
"step": 1140
},
{
"epoch": 1.3142857142857143,
"grad_norm": 4.5625,
"learning_rate": 1.1245714285714286e-05,
"loss": 1.5304,
"step": 1150
},
{
"epoch": 1.3257142857142856,
"grad_norm": 8.1875,
"learning_rate": 1.1169523809523811e-05,
"loss": 1.5348,
"step": 1160
},
{
"epoch": 1.3371428571428572,
"grad_norm": 13.375,
"learning_rate": 1.1093333333333334e-05,
"loss": 1.4743,
"step": 1170
},
{
"epoch": 1.3485714285714285,
"grad_norm": 6.6875,
"learning_rate": 1.1017142857142858e-05,
"loss": 1.3598,
"step": 1180
},
{
"epoch": 1.3599999999999999,
"grad_norm": 7.875,
"learning_rate": 1.0940952380952381e-05,
"loss": 1.4255,
"step": 1190
},
{
"epoch": 1.3714285714285714,
"grad_norm": 9.6875,
"learning_rate": 1.0864761904761905e-05,
"loss": 1.6216,
"step": 1200
},
{
"epoch": 1.3828571428571428,
"grad_norm": 11.1875,
"learning_rate": 1.078857142857143e-05,
"loss": 1.4601,
"step": 1210
},
{
"epoch": 1.3942857142857144,
"grad_norm": 9.25,
"learning_rate": 1.0712380952380954e-05,
"loss": 1.4096,
"step": 1220
},
{
"epoch": 1.4057142857142857,
"grad_norm": 7.8125,
"learning_rate": 1.0636190476190477e-05,
"loss": 1.52,
"step": 1230
},
{
"epoch": 1.4171428571428573,
"grad_norm": 5.0,
"learning_rate": 1.056e-05,
"loss": 1.3965,
"step": 1240
},
{
"epoch": 1.4285714285714286,
"grad_norm": 8.875,
"learning_rate": 1.0483809523809526e-05,
"loss": 1.4751,
"step": 1250
},
{
"epoch": 1.44,
"grad_norm": 10.0625,
"learning_rate": 1.040761904761905e-05,
"loss": 1.451,
"step": 1260
},
{
"epoch": 1.4514285714285715,
"grad_norm": 7.5,
"learning_rate": 1.0331428571428573e-05,
"loss": 1.4057,
"step": 1270
},
{
"epoch": 1.4628571428571429,
"grad_norm": 9.0,
"learning_rate": 1.0255238095238096e-05,
"loss": 1.4995,
"step": 1280
},
{
"epoch": 1.4742857142857142,
"grad_norm": 6.3125,
"learning_rate": 1.017904761904762e-05,
"loss": 1.3841,
"step": 1290
},
{
"epoch": 1.4857142857142858,
"grad_norm": 8.3125,
"learning_rate": 1.0102857142857143e-05,
"loss": 1.5022,
"step": 1300
},
{
"epoch": 1.497142857142857,
"grad_norm": 7.1875,
"learning_rate": 1.0026666666666667e-05,
"loss": 1.416,
"step": 1310
},
{
"epoch": 1.5085714285714285,
"grad_norm": 4.1875,
"learning_rate": 9.950476190476192e-06,
"loss": 1.5316,
"step": 1320
},
{
"epoch": 1.52,
"grad_norm": 10.375,
"learning_rate": 9.874285714285715e-06,
"loss": 1.5099,
"step": 1330
},
{
"epoch": 1.5314285714285716,
"grad_norm": 9.9375,
"learning_rate": 9.798095238095239e-06,
"loss": 1.463,
"step": 1340
},
{
"epoch": 1.5428571428571427,
"grad_norm": 4.3125,
"learning_rate": 9.721904761904762e-06,
"loss": 1.4075,
"step": 1350
},
{
"epoch": 1.5542857142857143,
"grad_norm": 6.78125,
"learning_rate": 9.645714285714286e-06,
"loss": 1.437,
"step": 1360
},
{
"epoch": 1.5657142857142858,
"grad_norm": 9.6875,
"learning_rate": 9.569523809523811e-06,
"loss": 1.4965,
"step": 1370
},
{
"epoch": 1.5771428571428572,
"grad_norm": 9.875,
"learning_rate": 9.493333333333334e-06,
"loss": 1.3657,
"step": 1380
},
{
"epoch": 1.5885714285714285,
"grad_norm": 6.875,
"learning_rate": 9.417142857142858e-06,
"loss": 1.5096,
"step": 1390
},
{
"epoch": 1.6,
"grad_norm": 7.03125,
"learning_rate": 9.340952380952381e-06,
"loss": 1.4665,
"step": 1400
},
{
"epoch": 1.6114285714285714,
"grad_norm": 4.21875,
"learning_rate": 9.264761904761905e-06,
"loss": 1.4288,
"step": 1410
},
{
"epoch": 1.6228571428571428,
"grad_norm": 6.0625,
"learning_rate": 9.188571428571428e-06,
"loss": 1.3966,
"step": 1420
},
{
"epoch": 1.6342857142857143,
"grad_norm": 7.03125,
"learning_rate": 9.112380952380954e-06,
"loss": 1.376,
"step": 1430
},
{
"epoch": 1.6457142857142857,
"grad_norm": 8.625,
"learning_rate": 9.036190476190477e-06,
"loss": 1.4886,
"step": 1440
},
{
"epoch": 1.657142857142857,
"grad_norm": 7.03125,
"learning_rate": 8.96e-06,
"loss": 1.449,
"step": 1450
},
{
"epoch": 1.6685714285714286,
"grad_norm": 4.46875,
"learning_rate": 8.883809523809524e-06,
"loss": 1.5675,
"step": 1460
},
{
"epoch": 1.6800000000000002,
"grad_norm": 6.65625,
"learning_rate": 8.807619047619048e-06,
"loss": 1.4398,
"step": 1470
},
{
"epoch": 1.6914285714285713,
"grad_norm": 7.875,
"learning_rate": 8.731428571428571e-06,
"loss": 1.4698,
"step": 1480
},
{
"epoch": 1.7028571428571428,
"grad_norm": 6.5625,
"learning_rate": 8.655238095238096e-06,
"loss": 1.4593,
"step": 1490
},
{
"epoch": 1.7142857142857144,
"grad_norm": 4.3125,
"learning_rate": 8.57904761904762e-06,
"loss": 1.4622,
"step": 1500
},
{
"epoch": 1.7257142857142858,
"grad_norm": 4.34375,
"learning_rate": 8.502857142857143e-06,
"loss": 1.5299,
"step": 1510
},
{
"epoch": 1.737142857142857,
"grad_norm": 4.34375,
"learning_rate": 8.426666666666667e-06,
"loss": 1.4833,
"step": 1520
},
{
"epoch": 1.7485714285714287,
"grad_norm": 4.03125,
"learning_rate": 8.35047619047619e-06,
"loss": 1.3867,
"step": 1530
},
{
"epoch": 1.76,
"grad_norm": 4.1875,
"learning_rate": 8.274285714285715e-06,
"loss": 1.434,
"step": 1540
},
{
"epoch": 1.7714285714285714,
"grad_norm": 10.1875,
"learning_rate": 8.198095238095239e-06,
"loss": 1.4841,
"step": 1550
},
{
"epoch": 1.782857142857143,
"grad_norm": 11.125,
"learning_rate": 8.121904761904762e-06,
"loss": 1.4636,
"step": 1560
},
{
"epoch": 1.7942857142857143,
"grad_norm": 10.4375,
"learning_rate": 8.045714285714286e-06,
"loss": 1.4761,
"step": 1570
},
{
"epoch": 1.8057142857142856,
"grad_norm": 10.1875,
"learning_rate": 7.969523809523811e-06,
"loss": 1.5066,
"step": 1580
},
{
"epoch": 1.8171428571428572,
"grad_norm": 7.125,
"learning_rate": 7.893333333333335e-06,
"loss": 1.4319,
"step": 1590
},
{
"epoch": 1.8285714285714287,
"grad_norm": 8.75,
"learning_rate": 7.817142857142858e-06,
"loss": 1.4641,
"step": 1600
},
{
"epoch": 1.8399999999999999,
"grad_norm": 4.34375,
"learning_rate": 7.740952380952382e-06,
"loss": 1.4205,
"step": 1610
},
{
"epoch": 1.8514285714285714,
"grad_norm": 9.9375,
"learning_rate": 7.664761904761905e-06,
"loss": 1.5164,
"step": 1620
},
{
"epoch": 1.862857142857143,
"grad_norm": 6.875,
"learning_rate": 7.588571428571429e-06,
"loss": 1.5392,
"step": 1630
},
{
"epoch": 1.8742857142857143,
"grad_norm": 4.34375,
"learning_rate": 7.512380952380953e-06,
"loss": 1.4799,
"step": 1640
},
{
"epoch": 1.8857142857142857,
"grad_norm": 7.0,
"learning_rate": 7.436190476190477e-06,
"loss": 1.5299,
"step": 1650
},
{
"epoch": 1.8971428571428572,
"grad_norm": 7.96875,
"learning_rate": 7.360000000000001e-06,
"loss": 1.4444,
"step": 1660
},
{
"epoch": 1.9085714285714286,
"grad_norm": 4.34375,
"learning_rate": 7.283809523809524e-06,
"loss": 1.5101,
"step": 1670
},
{
"epoch": 1.92,
"grad_norm": 7.15625,
"learning_rate": 7.207619047619048e-06,
"loss": 1.4646,
"step": 1680
},
{
"epoch": 1.9314285714285715,
"grad_norm": 4.625,
"learning_rate": 7.131428571428573e-06,
"loss": 1.5244,
"step": 1690
},
{
"epoch": 1.9428571428571428,
"grad_norm": 10.0625,
"learning_rate": 7.055238095238096e-06,
"loss": 1.4385,
"step": 1700
},
{
"epoch": 1.9542857142857142,
"grad_norm": 7.78125,
"learning_rate": 6.97904761904762e-06,
"loss": 1.4275,
"step": 1710
},
{
"epoch": 1.9657142857142857,
"grad_norm": 8.6875,
"learning_rate": 6.902857142857143e-06,
"loss": 1.449,
"step": 1720
},
{
"epoch": 1.977142857142857,
"grad_norm": 4.4375,
"learning_rate": 6.826666666666667e-06,
"loss": 1.5125,
"step": 1730
},
{
"epoch": 1.9885714285714284,
"grad_norm": 6.28125,
"learning_rate": 6.75047619047619e-06,
"loss": 1.3555,
"step": 1740
},
{
"epoch": 2.0,
"grad_norm": 4.0625,
"learning_rate": 6.6742857142857155e-06,
"loss": 1.404,
"step": 1750
},
{
"epoch": 2.0114285714285716,
"grad_norm": 6.6875,
"learning_rate": 6.598095238095239e-06,
"loss": 1.3851,
"step": 1760
},
{
"epoch": 2.0228571428571427,
"grad_norm": 6.84375,
"learning_rate": 6.5219047619047625e-06,
"loss": 1.4035,
"step": 1770
},
{
"epoch": 2.0342857142857143,
"grad_norm": 8.0625,
"learning_rate": 6.445714285714286e-06,
"loss": 1.4249,
"step": 1780
},
{
"epoch": 2.045714285714286,
"grad_norm": 3.875,
"learning_rate": 6.36952380952381e-06,
"loss": 1.5085,
"step": 1790
},
{
"epoch": 2.057142857142857,
"grad_norm": 8.6875,
"learning_rate": 6.293333333333334e-06,
"loss": 1.4788,
"step": 1800
},
{
"epoch": 2.0685714285714285,
"grad_norm": 10.3125,
"learning_rate": 6.217142857142857e-06,
"loss": 1.5104,
"step": 1810
},
{
"epoch": 2.08,
"grad_norm": 7.0625,
"learning_rate": 6.140952380952382e-06,
"loss": 1.397,
"step": 1820
},
{
"epoch": 2.0914285714285716,
"grad_norm": 10.9375,
"learning_rate": 6.064761904761905e-06,
"loss": 1.5341,
"step": 1830
},
{
"epoch": 2.1028571428571428,
"grad_norm": 7.625,
"learning_rate": 5.9885714285714294e-06,
"loss": 1.5131,
"step": 1840
},
{
"epoch": 2.1142857142857143,
"grad_norm": 8.125,
"learning_rate": 5.912380952380953e-06,
"loss": 1.4619,
"step": 1850
},
{
"epoch": 2.125714285714286,
"grad_norm": 8.875,
"learning_rate": 5.836190476190476e-06,
"loss": 1.4592,
"step": 1860
},
{
"epoch": 2.137142857142857,
"grad_norm": 7.71875,
"learning_rate": 5.76e-06,
"loss": 1.4432,
"step": 1870
},
{
"epoch": 2.1485714285714286,
"grad_norm": 9.25,
"learning_rate": 5.683809523809525e-06,
"loss": 1.4692,
"step": 1880
},
{
"epoch": 2.16,
"grad_norm": 4.90625,
"learning_rate": 5.607619047619049e-06,
"loss": 1.3839,
"step": 1890
},
{
"epoch": 2.1714285714285713,
"grad_norm": 10.0,
"learning_rate": 5.531428571428572e-06,
"loss": 1.5186,
"step": 1900
},
{
"epoch": 2.182857142857143,
"grad_norm": 10.1875,
"learning_rate": 5.4552380952380956e-06,
"loss": 1.479,
"step": 1910
},
{
"epoch": 2.1942857142857144,
"grad_norm": 7.15625,
"learning_rate": 5.379047619047619e-06,
"loss": 1.3692,
"step": 1920
},
{
"epoch": 2.2057142857142855,
"grad_norm": 9.3125,
"learning_rate": 5.3028571428571425e-06,
"loss": 1.4544,
"step": 1930
},
{
"epoch": 2.217142857142857,
"grad_norm": 6.25,
"learning_rate": 5.226666666666667e-06,
"loss": 1.4225,
"step": 1940
},
{
"epoch": 2.2285714285714286,
"grad_norm": 9.125,
"learning_rate": 5.150476190476191e-06,
"loss": 1.4868,
"step": 1950
},
{
"epoch": 2.24,
"grad_norm": 7.125,
"learning_rate": 5.074285714285715e-06,
"loss": 1.4007,
"step": 1960
},
{
"epoch": 2.2514285714285713,
"grad_norm": 7.84375,
"learning_rate": 4.998095238095238e-06,
"loss": 1.5,
"step": 1970
},
{
"epoch": 2.262857142857143,
"grad_norm": 6.5,
"learning_rate": 4.9219047619047625e-06,
"loss": 1.3791,
"step": 1980
},
{
"epoch": 2.2742857142857145,
"grad_norm": 7.34375,
"learning_rate": 4.845714285714286e-06,
"loss": 1.473,
"step": 1990
},
{
"epoch": 2.2857142857142856,
"grad_norm": 9.6875,
"learning_rate": 4.76952380952381e-06,
"loss": 1.4225,
"step": 2000
},
{
"epoch": 2.297142857142857,
"grad_norm": 8.5,
"learning_rate": 4.693333333333334e-06,
"loss": 1.4277,
"step": 2010
},
{
"epoch": 2.3085714285714287,
"grad_norm": 8.9375,
"learning_rate": 4.617142857142857e-06,
"loss": 1.4491,
"step": 2020
},
{
"epoch": 2.32,
"grad_norm": 10.5,
"learning_rate": 4.540952380952382e-06,
"loss": 1.5224,
"step": 2030
},
{
"epoch": 2.3314285714285714,
"grad_norm": 4.90625,
"learning_rate": 4.464761904761905e-06,
"loss": 1.4124,
"step": 2040
},
{
"epoch": 2.342857142857143,
"grad_norm": 4.34375,
"learning_rate": 4.388571428571429e-06,
"loss": 1.5009,
"step": 2050
},
{
"epoch": 2.354285714285714,
"grad_norm": 8.5,
"learning_rate": 4.312380952380953e-06,
"loss": 1.4571,
"step": 2060
},
{
"epoch": 2.3657142857142857,
"grad_norm": 6.84375,
"learning_rate": 4.2361904761904765e-06,
"loss": 1.51,
"step": 2070
},
{
"epoch": 2.3771428571428572,
"grad_norm": 8.5625,
"learning_rate": 4.16e-06,
"loss": 1.3797,
"step": 2080
},
{
"epoch": 2.388571428571429,
"grad_norm": 9.4375,
"learning_rate": 4.083809523809524e-06,
"loss": 1.4314,
"step": 2090
},
{
"epoch": 2.4,
"grad_norm": 7.96875,
"learning_rate": 4.007619047619048e-06,
"loss": 1.4811,
"step": 2100
},
{
"epoch": 2.4114285714285715,
"grad_norm": 4.125,
"learning_rate": 3.931428571428571e-06,
"loss": 1.4409,
"step": 2110
},
{
"epoch": 2.422857142857143,
"grad_norm": 4.0625,
"learning_rate": 3.855238095238096e-06,
"loss": 1.298,
"step": 2120
},
{
"epoch": 2.434285714285714,
"grad_norm": 8.75,
"learning_rate": 3.7790476190476196e-06,
"loss": 1.4243,
"step": 2130
},
{
"epoch": 2.4457142857142857,
"grad_norm": 6.9375,
"learning_rate": 3.702857142857143e-06,
"loss": 1.4088,
"step": 2140
},
{
"epoch": 2.4571428571428573,
"grad_norm": 4.09375,
"learning_rate": 3.6266666666666674e-06,
"loss": 1.3406,
"step": 2150
},
{
"epoch": 2.4685714285714284,
"grad_norm": 4.4375,
"learning_rate": 3.550476190476191e-06,
"loss": 1.4747,
"step": 2160
},
{
"epoch": 2.48,
"grad_norm": 8.75,
"learning_rate": 3.4742857142857144e-06,
"loss": 1.475,
"step": 2170
},
{
"epoch": 2.4914285714285715,
"grad_norm": 4.65625,
"learning_rate": 3.3980952380952383e-06,
"loss": 1.3833,
"step": 2180
},
{
"epoch": 2.5028571428571427,
"grad_norm": 8.0,
"learning_rate": 3.321904761904762e-06,
"loss": 1.455,
"step": 2190
},
{
"epoch": 2.5142857142857142,
"grad_norm": 7.21875,
"learning_rate": 3.245714285714286e-06,
"loss": 1.4182,
"step": 2200
},
{
"epoch": 2.525714285714286,
"grad_norm": 6.53125,
"learning_rate": 3.1695238095238096e-06,
"loss": 1.4157,
"step": 2210
},
{
"epoch": 2.5371428571428574,
"grad_norm": 8.125,
"learning_rate": 3.093333333333334e-06,
"loss": 1.4492,
"step": 2220
},
{
"epoch": 2.5485714285714285,
"grad_norm": 4.34375,
"learning_rate": 3.0171428571428574e-06,
"loss": 1.3572,
"step": 2230
},
{
"epoch": 2.56,
"grad_norm": 8.3125,
"learning_rate": 2.940952380952381e-06,
"loss": 1.4794,
"step": 2240
},
{
"epoch": 2.571428571428571,
"grad_norm": 9.0625,
"learning_rate": 2.8647619047619052e-06,
"loss": 1.2946,
"step": 2250
},
{
"epoch": 2.5828571428571427,
"grad_norm": 9.9375,
"learning_rate": 2.7885714285714287e-06,
"loss": 1.4978,
"step": 2260
},
{
"epoch": 2.5942857142857143,
"grad_norm": 7.9375,
"learning_rate": 2.7123809523809526e-06,
"loss": 1.4238,
"step": 2270
},
{
"epoch": 2.605714285714286,
"grad_norm": 6.53125,
"learning_rate": 2.6361904761904766e-06,
"loss": 1.4799,
"step": 2280
},
{
"epoch": 2.617142857142857,
"grad_norm": 4.375,
"learning_rate": 2.56e-06,
"loss": 1.359,
"step": 2290
},
{
"epoch": 2.6285714285714286,
"grad_norm": 4.09375,
"learning_rate": 2.483809523809524e-06,
"loss": 1.4785,
"step": 2300
},
{
"epoch": 2.64,
"grad_norm": 8.8125,
"learning_rate": 2.407619047619048e-06,
"loss": 1.4713,
"step": 2310
},
{
"epoch": 2.6514285714285712,
"grad_norm": 6.71875,
"learning_rate": 2.331428571428572e-06,
"loss": 1.5077,
"step": 2320
},
{
"epoch": 2.662857142857143,
"grad_norm": 4.875,
"learning_rate": 2.2552380952380953e-06,
"loss": 1.4597,
"step": 2330
},
{
"epoch": 2.6742857142857144,
"grad_norm": 7.625,
"learning_rate": 2.179047619047619e-06,
"loss": 1.411,
"step": 2340
},
{
"epoch": 2.685714285714286,
"grad_norm": 13.1875,
"learning_rate": 2.102857142857143e-06,
"loss": 1.464,
"step": 2350
},
{
"epoch": 2.697142857142857,
"grad_norm": 6.90625,
"learning_rate": 2.0266666666666666e-06,
"loss": 1.4423,
"step": 2360
},
{
"epoch": 2.7085714285714286,
"grad_norm": 7.625,
"learning_rate": 1.9504761904761905e-06,
"loss": 1.3684,
"step": 2370
},
{
"epoch": 2.7199999999999998,
"grad_norm": 6.625,
"learning_rate": 1.8742857142857142e-06,
"loss": 1.4071,
"step": 2380
},
{
"epoch": 2.7314285714285713,
"grad_norm": 4.71875,
"learning_rate": 1.7980952380952381e-06,
"loss": 1.4424,
"step": 2390
},
{
"epoch": 2.742857142857143,
"grad_norm": 7.75,
"learning_rate": 1.721904761904762e-06,
"loss": 1.3702,
"step": 2400
},
{
"epoch": 2.7542857142857144,
"grad_norm": 11.3125,
"learning_rate": 1.6457142857142857e-06,
"loss": 1.4544,
"step": 2410
},
{
"epoch": 2.7657142857142856,
"grad_norm": 3.984375,
"learning_rate": 1.5695238095238097e-06,
"loss": 1.4182,
"step": 2420
},
{
"epoch": 2.777142857142857,
"grad_norm": 9.1875,
"learning_rate": 1.4933333333333336e-06,
"loss": 1.4721,
"step": 2430
},
{
"epoch": 2.7885714285714287,
"grad_norm": 6.5625,
"learning_rate": 1.4171428571428573e-06,
"loss": 1.4906,
"step": 2440
},
{
"epoch": 2.8,
"grad_norm": 4.21875,
"learning_rate": 1.3409523809523812e-06,
"loss": 1.5009,
"step": 2450
},
{
"epoch": 2.8114285714285714,
"grad_norm": 10.5,
"learning_rate": 1.2647619047619047e-06,
"loss": 1.4903,
"step": 2460
},
{
"epoch": 2.822857142857143,
"grad_norm": 6.96875,
"learning_rate": 1.1885714285714286e-06,
"loss": 1.4078,
"step": 2470
},
{
"epoch": 2.8342857142857145,
"grad_norm": 8.1875,
"learning_rate": 1.1123809523809525e-06,
"loss": 1.4798,
"step": 2480
},
{
"epoch": 2.8457142857142856,
"grad_norm": 4.28125,
"learning_rate": 1.0361904761904764e-06,
"loss": 1.3708,
"step": 2490
},
{
"epoch": 2.857142857142857,
"grad_norm": 7.46875,
"learning_rate": 9.600000000000001e-07,
"loss": 1.5206,
"step": 2500
},
{
"epoch": 2.8685714285714283,
"grad_norm": 10.625,
"learning_rate": 8.838095238095238e-07,
"loss": 1.4748,
"step": 2510
},
{
"epoch": 2.88,
"grad_norm": 10.625,
"learning_rate": 8.076190476190476e-07,
"loss": 1.3786,
"step": 2520
},
{
"epoch": 2.8914285714285715,
"grad_norm": 6.625,
"learning_rate": 7.314285714285715e-07,
"loss": 1.4883,
"step": 2530
},
{
"epoch": 2.902857142857143,
"grad_norm": 4.03125,
"learning_rate": 6.552380952380954e-07,
"loss": 1.4364,
"step": 2540
},
{
"epoch": 2.914285714285714,
"grad_norm": 6.84375,
"learning_rate": 5.790476190476191e-07,
"loss": 1.5024,
"step": 2550
},
{
"epoch": 2.9257142857142857,
"grad_norm": 4.09375,
"learning_rate": 5.028571428571429e-07,
"loss": 1.4519,
"step": 2560
},
{
"epoch": 2.9371428571428573,
"grad_norm": 6.625,
"learning_rate": 4.266666666666667e-07,
"loss": 1.4291,
"step": 2570
},
{
"epoch": 2.9485714285714284,
"grad_norm": 10.625,
"learning_rate": 3.5047619047619054e-07,
"loss": 1.4619,
"step": 2580
},
{
"epoch": 2.96,
"grad_norm": 4.34375,
"learning_rate": 2.742857142857143e-07,
"loss": 1.3314,
"step": 2590
},
{
"epoch": 2.9714285714285715,
"grad_norm": 8.8125,
"learning_rate": 1.980952380952381e-07,
"loss": 1.432,
"step": 2600
},
{
"epoch": 2.982857142857143,
"grad_norm": 8.875,
"learning_rate": 1.219047619047619e-07,
"loss": 1.4239,
"step": 2610
},
{
"epoch": 2.994285714285714,
"grad_norm": 7.21875,
"learning_rate": 4.571428571428572e-08,
"loss": 1.4812,
"step": 2620
}
],
"logging_steps": 10,
"max_steps": 2625,
"num_input_tokens_seen": 0,
"num_train_epochs": 3,
"save_steps": 500,
"stateful_callbacks": {
"TrainerControl": {
"args": {
"should_epoch_stop": false,
"should_evaluate": false,
"should_log": false,
"should_save": true,
"should_training_stop": true
},
"attributes": {}
}
},
"total_flos": 1.6196076078261862e+17,
"train_batch_size": 8,
"trial_name": null,
"trial_params": null
}