{
  "best_global_step": null,
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 1.0,
  "eval_steps": 500,
  "global_step": 3000,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.0003333333333333333,
      "grad_norm": 193.30125427246094,
      "learning_rate": 1.9833333333333332e-05,
      "loss": 154.99522705078124,
      "original_ce_loss": 9.6865,
      "step": 10
    },
    {
      "epoch": 0.0006666666666666666,
      "grad_norm": 207.8564453125,
      "learning_rate": 4.816666666666666e-05,
      "loss": 154.4616943359375,
      "original_ce_loss": 9.6532,
      "step": 20
    },
    {
      "epoch": 0.001,
      "grad_norm": 280.3179931640625,
      "learning_rate": 7.649999999999999e-05,
      "loss": 152.58868408203125,
      "original_ce_loss": 9.5361,
      "step": 30
    },
    {
      "epoch": 0.0013333333333333333,
      "grad_norm": 359.151611328125,
      "learning_rate": 9.916666666666666e-05,
      "loss": 145.5450927734375,
      "original_ce_loss": 9.0959,
      "step": 40
    },
    {
      "epoch": 0.0016666666666666668,
      "grad_norm": 104.84541320800781,
      "learning_rate": 0.00012749999999999998,
      "loss": 129.752197265625,
      "original_ce_loss": 8.1088,
      "step": 50
    },
    {
      "epoch": 0.002,
      "grad_norm": 42.41929244995117,
      "learning_rate": 0.00015583333333333332,
      "loss": 123.40166015625,
      "original_ce_loss": 7.7119,
      "step": 60
    },
    {
      "epoch": 0.0023333333333333335,
      "grad_norm": 43.07599639892578,
      "learning_rate": 0.00018416666666666665,
      "loss": 116.051904296875,
      "original_ce_loss": 7.2526,
      "step": 70
    },
    {
      "epoch": 0.0026666666666666666,
      "grad_norm": 15.110007286071777,
      "learning_rate": 0.0002125,
      "loss": 107.8390625,
      "original_ce_loss": 6.7393,
      "step": 80
    },
    {
      "epoch": 0.003,
      "grad_norm": 19.066268920898438,
      "learning_rate": 0.00024083333333333332,
      "loss": 104.07568359375,
      "original_ce_loss": 6.504,
      "step": 90
    },
    {
      "epoch": 0.0033333333333333335,
      "grad_norm": 15.286686897277832,
      "learning_rate": 0.00026916666666666663,
      "loss": 102.61002197265626,
      "original_ce_loss": 6.4124,
      "step": 100
    },
    {
      "epoch": 0.0036666666666666666,
      "grad_norm": 15.667778015136719,
      "learning_rate": 0.00029749999999999997,
      "loss": 101.6726806640625,
      "original_ce_loss": 6.3539,
      "step": 110
    },
    {
      "epoch": 0.004,
      "grad_norm": 20.178363800048828,
      "learning_rate": 0.00032583333333333336,
      "loss": 101.5388427734375,
      "original_ce_loss": 6.3455,
      "step": 120
    },
    {
      "epoch": 0.004333333333333333,
      "grad_norm": 127.03045654296875,
      "learning_rate": 0.00035416666666666664,
      "loss": 101.36513061523438,
      "original_ce_loss": 6.3346,
      "step": 130
    },
    {
      "epoch": 0.004666666666666667,
      "grad_norm": 21.85209846496582,
      "learning_rate": 0.0003825,
      "loss": 100.66131591796875,
      "original_ce_loss": 6.2907,
      "step": 140
    },
    {
      "epoch": 0.005,
      "grad_norm": 28.871206283569336,
      "learning_rate": 0.00040799999999999994,
      "loss": 100.21246337890625,
      "original_ce_loss": 6.2626,
      "step": 150
    },
    {
      "epoch": 0.005333333333333333,
      "grad_norm": 8.257862091064453,
      "learning_rate": 0.0004249999811695807,
      "loss": 99.4580810546875,
      "original_ce_loss": 6.2155,
      "step": 160
    },
    {
      "epoch": 0.005666666666666667,
      "grad_norm": 16.51659393310547,
      "learning_rate": 0.0004249997693274025,
      "loss": 98.74091796875,
      "original_ce_loss": 6.1707,
      "step": 170
    },
    {
      "epoch": 0.006,
      "grad_norm": NaN,
      "learning_rate": 0.0004249993774195584,
      "loss": 98.07858276367188,
      "original_ce_loss": 6.1293,
      "step": 180
    },
    {
      "epoch": 0.006333333333333333,
      "grad_norm": 29.44127082824707,
      "learning_rate": 0.0004249987948542895,
      "loss": 97.93662719726562,
      "original_ce_loss": 6.1204,
      "step": 190
    },
    {
      "epoch": 0.006666666666666667,
      "grad_norm": 14.971574783325195,
      "learning_rate": 0.0004249979239496277,
      "loss": 97.34913940429688,
      "original_ce_loss": 6.0837,
      "step": 200
    },
    {
      "epoch": 0.007,
      "grad_norm": 12.689688682556152,
      "learning_rate": 0.0004249968176670431,
      "loss": 96.178271484375,
      "original_ce_loss": 6.0105,
      "step": 210
    },
    {
      "epoch": 0.007333333333333333,
      "grad_norm": 20.04000473022461,
      "learning_rate": 0.0004249954760077612,
      "loss": 94.8289306640625,
      "original_ce_loss": 5.9262,
      "step": 220
    },
    {
      "epoch": 0.007666666666666666,
      "grad_norm": 11.919169425964355,
      "learning_rate": 0.000424993898973268,
      "loss": 93.77874755859375,
      "original_ce_loss": 5.8605,
      "step": 230
    },
    {
      "epoch": 0.008,
      "grad_norm": 21.34636688232422,
      "learning_rate": 0.00042499208656531043,
      "loss": 92.63297729492187,
      "original_ce_loss": 5.7889,
      "step": 240
    },
    {
      "epoch": 0.008333333333333333,
      "grad_norm": 14.039680480957031,
      "learning_rate": 0.000424990038785896,
      "loss": 91.2647705078125,
      "original_ce_loss": 5.7034,
      "step": 250
    },
    {
      "epoch": 0.008666666666666666,
      "grad_norm": 16.5980224609375,
      "learning_rate": 0.00042498775563729296,
      "loss": 89.99480590820312,
      "original_ce_loss": 5.6241,
      "step": 260
    },
    {
      "epoch": 0.009,
      "grad_norm": 35.983943939208984,
      "learning_rate": 0.0004249852371220303,
      "loss": 88.72568359375,
      "original_ce_loss": 5.5447,
      "step": 270
    },
    {
      "epoch": 0.009333333333333334,
      "grad_norm": 10.545525550842285,
      "learning_rate": 0.00042498248324289776,
      "loss": 87.27744750976562,
      "original_ce_loss": 5.4542,
      "step": 280
    },
    {
      "epoch": 0.009666666666666667,
      "grad_norm": 14.494406700134277,
      "learning_rate": 0.0004249794940029456,
      "loss": 85.4427490234375,
      "original_ce_loss": 5.3396,
      "step": 290
    },
    {
      "epoch": 0.01,
      "grad_norm": 46.0245361328125,
      "learning_rate": 0.00042497626940548515,
      "loss": 84.02860107421876,
      "original_ce_loss": 5.2512,
      "step": 300
    },
    {
      "epoch": 0.010333333333333333,
      "grad_norm": 14.768983840942383,
      "learning_rate": 0.000424972809454088,
      "loss": 82.96371459960938,
      "original_ce_loss": 5.1846,
      "step": 310
    },
    {
      "epoch": 0.010666666666666666,
      "grad_norm": 16.24750328063965,
      "learning_rate": 0.00042496911415258666,
      "loss": 82.49393310546876,
      "original_ce_loss": 5.1553,
      "step": 320
    },
    {
      "epoch": 0.011,
      "grad_norm": 14.52138900756836,
      "learning_rate": 0.0004249655871602741,
      "loss": 80.85582275390625,
      "original_ce_loss": 5.0529,
      "step": 330
    },
    {
      "epoch": 0.011333333333333334,
      "grad_norm": 19.26217269897461,
      "learning_rate": 0.00042496144470506705,
      "loss": 79.52462158203124,
      "original_ce_loss": 4.9697,
      "step": 340
    },
    {
      "epoch": 0.011666666666666667,
      "grad_norm": 20.684001922607422,
      "learning_rate": 0.00042495706691234427,
      "loss": 77.98438110351563,
      "original_ce_loss": 4.8734,
      "step": 350
    },
    {
      "epoch": 0.012,
      "grad_norm": 23.948087692260742,
      "learning_rate": 0.000424952453786955,
      "loss": 76.44223022460938,
      "original_ce_loss": 4.777,
      "step": 360
    },
    {
      "epoch": 0.012333333333333333,
      "grad_norm": 36.791988372802734,
      "learning_rate": 0.0004249476053340089,
      "loss": 75.355859375,
      "original_ce_loss": 4.7092,
      "step": 370
    },
    {
      "epoch": 0.012666666666666666,
      "grad_norm": 21.555706024169922,
      "learning_rate": 0.0004249425215588765,
      "loss": 73.73348999023438,
      "original_ce_loss": 4.6078,
      "step": 380
    },
    {
      "epoch": 0.013,
      "grad_norm": 33.586387634277344,
      "learning_rate": 0.0004249377449654369,
      "loss": 72.61731567382813,
      "original_ce_loss": 4.538,
      "step": 390
    },
    {
      "epoch": 0.013333333333333334,
      "grad_norm": 27.996273040771484,
      "learning_rate": 0.00042493221409387993,
      "loss": 71.43023681640625,
      "original_ce_loss": 4.4638,
      "step": 400
    },
    {
      "epoch": 0.013666666666666667,
      "grad_norm": 45.84962463378906,
      "learning_rate": 0.0004249264479171851,
      "loss": 70.55713500976563,
      "original_ce_loss": 4.4092,
      "step": 410
    },
    {
      "epoch": 0.014,
      "grad_norm": 48.37245178222656,
      "learning_rate": 0.00042492044644173955,
      "loss": 69.7059326171875,
      "original_ce_loss": 4.356,
      "step": 420
    },
    {
      "epoch": 0.014333333333333333,
      "grad_norm": 26.593931198120117,
      "learning_rate": 0.0004249142096741908,
      "loss": 68.96209716796875,
      "original_ce_loss": 4.3096,
      "step": 430
    },
    {
      "epoch": 0.014666666666666666,
      "grad_norm": 25.2656307220459,
      "learning_rate": 0.00042490773762144724,
      "loss": 67.66499633789063,
      "original_ce_loss": 4.2285,
      "step": 440
    },
    {
      "epoch": 0.015,
      "grad_norm": 87.45381927490234,
      "learning_rate": 0.0004249010302906778,
      "loss": 66.64647827148437,
      "original_ce_loss": 4.1648,
      "step": 450
    },
    {
      "epoch": 0.015333333333333332,
      "grad_norm": 26.007844924926758,
      "learning_rate": 0.0004248940876893119,
      "loss": 66.23974609375,
      "original_ce_loss": 4.1394,
      "step": 460
    },
    {
      "epoch": 0.015666666666666666,
      "grad_norm": 38.981483459472656,
      "learning_rate": 0.0004248869098250397,
      "loss": 65.3707763671875,
      "original_ce_loss": 4.0851,
      "step": 470
    },
    {
      "epoch": 0.016,
      "grad_norm": 30.139528274536133,
      "learning_rate": 0.00042487949670581203,
      "loss": 64.82785034179688,
      "original_ce_loss": 4.0512,
      "step": 480
    },
    {
      "epoch": 0.01633333333333333,
      "grad_norm": 22.18630599975586,
      "learning_rate": 0.00042487184833984,
      "loss": 63.92879028320313,
      "original_ce_loss": 3.995,
      "step": 490
    },
    {
      "epoch": 0.016666666666666666,
      "grad_norm": 20.91753578186035,
      "learning_rate": 0.00042486396473559565,
      "loss": 62.890411376953125,
      "original_ce_loss": 3.9301,
      "step": 500
    },
    {
      "epoch": 0.017,
      "grad_norm": 16.906227111816406,
      "learning_rate": 0.00042485584590181135,
      "loss": 61.85081176757812,
      "original_ce_loss": 3.8651,
      "step": 510
    },
    {
      "epoch": 0.017333333333333333,
      "grad_norm": 20.301300048828125,
      "learning_rate": 0.00042484749184748007,
      "loss": 60.9402099609375,
      "original_ce_loss": 3.8082,
      "step": 520
    },
    {
      "epoch": 0.017666666666666667,
      "grad_norm": 15.232793807983398,
      "learning_rate": 0.0004248389025818554,
      "loss": 60.15031127929687,
      "original_ce_loss": 3.7589,
      "step": 530
    },
    {
      "epoch": 0.018,
      "grad_norm": 25.332082748413086,
      "learning_rate": 0.0004248300781144515,
      "loss": 59.50396728515625,
      "original_ce_loss": 3.7185,
      "step": 540
    },
    {
      "epoch": 0.018333333333333333,
      "grad_norm": 18.957500457763672,
      "learning_rate": 0.0004248210184550428,
      "loss": 58.9510986328125,
      "original_ce_loss": 3.6839,
      "step": 550
    },
    {
      "epoch": 0.018666666666666668,
      "grad_norm": 17.083209991455078,
      "learning_rate": 0.00042481172361366463,
      "loss": 58.551019287109376,
      "original_ce_loss": 3.6589,
      "step": 560
    },
    {
      "epoch": 0.019,
      "grad_norm": 20.623079299926758,
      "learning_rate": 0.00042480219360061243,
      "loss": 57.89994506835937,
      "original_ce_loss": 3.6182,
      "step": 570
    },
    {
      "epoch": 0.019333333333333334,
      "grad_norm": 20.007261276245117,
      "learning_rate": 0.0004247924284264425,
      "loss": 57.207183837890625,
      "original_ce_loss": 3.5749,
      "step": 580
    },
    {
      "epoch": 0.019666666666666666,
      "grad_norm": 24.606914520263672,
      "learning_rate": 0.00042478242810197137,
      "loss": 56.6140380859375,
      "original_ce_loss": 3.5379,
      "step": 590
    },
    {
      "epoch": 0.02,
      "grad_norm": 23.800609588623047,
      "learning_rate": 0.00042477219263827606,
      "loss": 56.085272216796874,
      "original_ce_loss": 3.5048,
      "step": 600
    },
    {
      "epoch": 0.02033333333333333,
      "grad_norm": 18.29131507873535,
      "learning_rate": 0.0004247617220466942,
      "loss": 55.73380737304687,
      "original_ce_loss": 3.4828,
      "step": 610
    },
    {
      "epoch": 0.020666666666666667,
      "grad_norm": 20.64620018005371,
      "learning_rate": 0.00042475101633882375,
      "loss": 55.267041015625,
      "original_ce_loss": 3.4537,
      "step": 620
    },
    {
      "epoch": 0.021,
      "grad_norm": 16.878942489624023,
      "learning_rate": 0.0004247400755265231,
      "loss": 54.697265625,
      "original_ce_loss": 3.4181,
      "step": 630
    },
    {
      "epoch": 0.021333333333333333,
      "grad_norm": 24.191560745239258,
      "learning_rate": 0.00042472889962191107,
      "loss": 54.2244140625,
      "original_ce_loss": 3.3885,
      "step": 640
    },
    {
      "epoch": 0.021666666666666667,
      "grad_norm": 19.806337356567383,
      "learning_rate": 0.00042471748863736686,
      "loss": 53.91731567382813,
      "original_ce_loss": 3.3693,
      "step": 650
    },
    {
      "epoch": 0.022,
      "grad_norm": 19.804443359375,
      "learning_rate": 0.0004247058425855301,
      "loss": 53.446173095703124,
      "original_ce_loss": 3.3399,
      "step": 660
    },
    {
      "epoch": 0.022333333333333334,
      "grad_norm": 20.795433044433594,
      "learning_rate": 0.0004246939614793009,
      "loss": 53.202099609375,
      "original_ce_loss": 3.3246,
      "step": 670
    },
    {
      "epoch": 0.02266666666666667,
      "grad_norm": 18.797666549682617,
      "learning_rate": 0.0004246818453318394,
      "loss": 52.8059814453125,
      "original_ce_loss": 3.2999,
      "step": 680
    },
    {
      "epoch": 0.023,
      "grad_norm": 15.52590274810791,
      "learning_rate": 0.0004246694941565666,
      "loss": 52.476416015625,
      "original_ce_loss": 3.2793,
      "step": 690
    },
    {
      "epoch": 0.023333333333333334,
      "grad_norm": 20.646650314331055,
      "learning_rate": 0.0004246569079671633,
      "loss": 52.04948120117187,
      "original_ce_loss": 3.2526,
      "step": 700
    },
    {
      "epoch": 0.023666666666666666,
      "grad_norm": 20.110410690307617,
      "learning_rate": 0.00042464408677757103,
      "loss": 51.823065185546874,
      "original_ce_loss": 3.2384,
      "step": 710
    },
    {
      "epoch": 0.024,
      "grad_norm": 16.53462028503418,
      "learning_rate": 0.00042463103060199145,
      "loss": 51.42784423828125,
      "original_ce_loss": 3.2137,
      "step": 720
    },
    {
      "epoch": 0.024333333333333332,
      "grad_norm": 18.70224952697754,
      "learning_rate": 0.0004246177394548864,
      "loss": 51.2402587890625,
      "original_ce_loss": 3.202,
      "step": 730
    },
    {
      "epoch": 0.024666666666666667,
      "grad_norm": 17.058368682861328,
      "learning_rate": 0.0004246042133509782,
      "loss": 50.905398559570315,
      "original_ce_loss": 3.1811,
      "step": 740
    },
    {
      "epoch": 0.025,
      "grad_norm": 18.16584587097168,
      "learning_rate": 0.0004245904523052492,
      "loss": 50.543515014648435,
      "original_ce_loss": 3.1585,
      "step": 750
    },
    {
      "epoch": 0.025333333333333333,
      "grad_norm": 18.601837158203125,
      "learning_rate": 0.0004245764563329423,
      "loss": 50.39247131347656,
      "original_ce_loss": 3.149,
      "step": 760
    },
    {
      "epoch": 0.025666666666666667,
      "grad_norm": 20.248645782470703,
      "learning_rate": 0.00042456222544956035,
      "loss": 50.04454040527344,
      "original_ce_loss": 3.1273,
      "step": 770
    },
    {
      "epoch": 0.026,
      "grad_norm": 19.146419525146484,
      "learning_rate": 0.0004245477596708665,
      "loss": 49.74447937011719,
      "original_ce_loss": 3.1086,
      "step": 780
    },
    {
      "epoch": 0.026333333333333334,
      "grad_norm": 17.926576614379883,
      "learning_rate": 0.0004245330590128841,
      "loss": 49.5018310546875,
      "original_ce_loss": 3.0934,
      "step": 790
    },
    {
      "epoch": 0.02666666666666667,
      "grad_norm": 15.49186897277832,
      "learning_rate": 0.0004245181234918967,
      "loss": 49.32894287109375,
      "original_ce_loss": 3.0826,
      "step": 800
    },
    {
      "epoch": 0.027,
      "grad_norm": 16.708791732788086,
      "learning_rate": 0.0004245029531244478,
      "loss": 49.1858642578125,
      "original_ce_loss": 3.0736,
      "step": 810
    },
    {
      "epoch": 0.027333333333333334,
      "grad_norm": 16.17323875427246,
      "learning_rate": 0.00042448754792734133,
      "loss": 48.92313842773437,
      "original_ce_loss": 3.0572,
      "step": 820
    },
    {
      "epoch": 0.027666666666666666,
      "grad_norm": 14.645228385925293,
      "learning_rate": 0.00042447190791764113,
      "loss": 48.708493041992185,
      "original_ce_loss": 3.0438,
      "step": 830
    },
    {
      "epoch": 0.028,
      "grad_norm": 16.37765121459961,
      "learning_rate": 0.0004244560331126712,
      "loss": 48.501834106445315,
      "original_ce_loss": 3.0309,
      "step": 840
    },
    {
      "epoch": 0.028333333333333332,
      "grad_norm": 20.280181884765625,
      "learning_rate": 0.0004244399235300156,
      "loss": 48.502603149414064,
      "original_ce_loss": 3.0309,
      "step": 850
    },
    {
      "epoch": 0.028666666666666667,
      "grad_norm": 18.538236618041992,
      "learning_rate": 0.00042442357918751853,
      "loss": 48.18049926757813,
      "original_ce_loss": 3.0108,
      "step": 860
    },
    {
      "epoch": 0.029,
      "grad_norm": 19.755647659301758,
      "learning_rate": 0.00042440700010328404,
      "loss": 48.11741333007812,
      "original_ce_loss": 3.0069,
      "step": 870
    },
    {
      "epoch": 0.029333333333333333,
      "grad_norm": 17.37371253967285,
      "learning_rate": 0.0004243901862956764,
      "loss": 47.78751220703125,
      "original_ce_loss": 2.9863,
      "step": 880
    },
    {
      "epoch": 0.029666666666666668,
      "grad_norm": 20.684240341186523,
      "learning_rate": 0.0004243731377833197,
      "loss": 47.599853515625,
      "original_ce_loss": 2.9745,
      "step": 890
    },
    {
      "epoch": 0.03,
      "grad_norm": 20.71816062927246,
      "learning_rate": 0.00042435585458509823,
      "loss": 47.446356201171874,
      "original_ce_loss": 2.9649,
      "step": 900
    },
    {
      "epoch": 0.030333333333333334,
      "grad_norm": 19.639667510986328,
      "learning_rate": 0.00042433833672015606,
      "loss": 47.269107055664065,
      "original_ce_loss": 2.9539,
      "step": 910
    },
    {
      "epoch": 0.030666666666666665,
      "grad_norm": 13.578326225280762,
      "learning_rate": 0.0004243205842078971,
      "loss": 46.974429321289065,
      "original_ce_loss": 2.9355,
      "step": 920
    },
    {
      "epoch": 0.031,
      "grad_norm": 23.431640625,
      "learning_rate": 0.00042430259706798543,
      "loss": 46.828652954101564,
      "original_ce_loss": 2.9263,
      "step": 930
    },
    {
      "epoch": 0.03133333333333333,
      "grad_norm": 15.45804214477539,
      "learning_rate": 0.0004242843753203448,
      "loss": 46.75359191894531,
      "original_ce_loss": 2.9217,
      "step": 940
    },
    {
      "epoch": 0.03166666666666667,
      "grad_norm": 21.198925018310547,
      "learning_rate": 0.0004242659189851591,
      "loss": 46.800146484375,
      "original_ce_loss": 2.9246,
      "step": 950
    },
    {
      "epoch": 0.032,
      "grad_norm": 16.371259689331055,
      "learning_rate": 0.0004242472280828717,
      "loss": 46.56090393066406,
      "original_ce_loss": 2.9096,
      "step": 960
    },
    {
      "epoch": 0.03233333333333333,
      "grad_norm": 16.652050018310547,
      "learning_rate": 0.00042422830263418607,
      "loss": 46.34129638671875,
      "original_ce_loss": 2.8959,
      "step": 970
    },
    {
      "epoch": 0.03266666666666666,
      "grad_norm": 15.668427467346191,
      "learning_rate": 0.00042420914266006534,
      "loss": 46.112457275390625,
      "original_ce_loss": 2.8816,
      "step": 980
    },
    {
      "epoch": 0.033,
      "grad_norm": 19.35259437561035,
      "learning_rate": 0.0004241897481817326,
      "loss": 45.83723754882813,
      "original_ce_loss": 2.8644,
      "step": 990
    },
    {
      "epoch": 0.03333333333333333,
      "grad_norm": 19.012178421020508,
      "learning_rate": 0.00042417011922067036,
      "loss": 45.88760986328125,
      "original_ce_loss": 2.8675,
      "step": 1000
    },
    {
      "epoch": 0.33666666666666667,
      "grad_norm": 12.40669059753418,
      "learning_rate": 0.0003383499911913083,
      "loss": 45.56795654296875,
      "original_ce_loss": 2.8476,
      "step": 1010
    },
    {
      "epoch": 0.34,
      "grad_norm": 11.26506519317627,
      "learning_rate": 0.0003364549467776501,
      "loss": 45.49089050292969,
      "original_ce_loss": 2.8427,
      "step": 1020
    },
    {
      "epoch": 0.3433333333333333,
      "grad_norm": 16.84862518310547,
      "learning_rate": 0.0003345448408201255,
      "loss": 45.356015014648435,
      "original_ce_loss": 2.8343,
      "step": 1030
    },
    {
      "epoch": 0.3466666666666667,
      "grad_norm": 13.147734642028809,
      "learning_rate": 0.00033261990541229276,
      "loss": 45.2113037109375,
      "original_ce_loss": 2.8253,
      "step": 1040
    },
    {
      "epoch": 0.35,
      "grad_norm": 15.466135025024414,
      "learning_rate": 0.00033068037444961007,
      "loss": 44.94834594726562,
      "original_ce_loss": 2.8088,
      "step": 1050
    },
    {
      "epoch": 0.35333333333333333,
      "grad_norm": 14.858617782592773,
      "learning_rate": 0.00032872648360101573,
      "loss": 44.842855834960936,
      "original_ce_loss": 2.8022,
      "step": 1060
    },
    {
      "epoch": 0.3566666666666667,
      "grad_norm": 12.897237777709961,
      "learning_rate": 0.0003267584702802916,
      "loss": 44.86846008300781,
      "original_ce_loss": 2.8039,
      "step": 1070
    },
    {
      "epoch": 0.36,
      "grad_norm": 14.901575088500977,
      "learning_rate": 0.00032477657361721626,
      "loss": 44.76924438476563,
      "original_ce_loss": 2.7977,
      "step": 1080
    },
    {
      "epoch": 0.36333333333333334,
      "grad_norm": 11.741254806518555,
      "learning_rate": 0.00032278103442850823,
      "loss": 44.71358642578125,
      "original_ce_loss": 2.7942,
      "step": 1090
    },
    {
      "epoch": 0.36666666666666664,
      "grad_norm": 15.501832008361816,
      "learning_rate": 0.00032077209518856486,
      "loss": 44.47730407714844,
      "original_ce_loss": 2.7794,
      "step": 1100
    },
    {
      "epoch": 0.37,
      "grad_norm": 11.20630931854248,
      "learning_rate": 0.00031874999999999997,
      "loss": 44.381414794921874,
      "original_ce_loss": 2.7734,
      "step": 1110
    },
    {
      "epoch": 0.37333333333333335,
      "grad_norm": 12.469181060791016,
      "learning_rate": 0.000316714994563983,
      "loss": 44.42494201660156,
      "original_ce_loss": 2.7761,
      "step": 1120
    },
    {
      "epoch": 0.37666666666666665,
      "grad_norm": 11.791640281677246,
      "learning_rate": 0.00031466732615038434,
      "loss": 44.274844360351565,
      "original_ce_loss": 2.7668,
      "step": 1130
    },
    {
      "epoch": 0.38,
      "grad_norm": 13.211021423339844,
      "learning_rate": 0.0003126072435677305,
      "loss": 44.124505615234376,
      "original_ce_loss": 2.7574,
      "step": 1140
    },
    {
      "epoch": 0.38333333333333336,
      "grad_norm": 14.31765079498291,
      "learning_rate": 0.00031053499713297114,
      "loss": 43.99041748046875,
      "original_ce_loss": 2.749,
      "step": 1150
    },
    {
      "epoch": 0.38666666666666666,
      "grad_norm": 15.122718811035156,
      "learning_rate": 0.00030845083864106397,
      "loss": 43.92890930175781,
      "original_ce_loss": 2.7451,
      "step": 1160
    },
    {
      "epoch": 0.39,
      "grad_norm": 13.994558334350586,
      "learning_rate": 0.0003063550213343795,
      "loss": 43.74768981933594,
      "original_ce_loss": 2.7338,
      "step": 1170
    },
    {
      "epoch": 0.3933333333333333,
      "grad_norm": 11.93791389465332,
      "learning_rate": 0.0003042477998719297,
      "loss": 43.84468994140625,
      "original_ce_loss": 2.7399,
      "step": 1180
    },
    {
      "epoch": 0.39666666666666667,
      "grad_norm": 12.416786193847656,
      "learning_rate": 0.0003021294302984256,
      "loss": 43.70210266113281,
      "original_ce_loss": 2.731,
      "step": 1190
    },
    {
      "epoch": 0.4,
      "grad_norm": 11.777953147888184,
      "learning_rate": 0.0003000001700131648,
      "loss": 43.617486572265626,
      "original_ce_loss": 2.7257,
      "step": 1200
    },
    {
      "epoch": 0.4033333333333333,
      "grad_norm": 12.980205535888672,
      "learning_rate": 0.00029786027773875604,
      "loss": 43.557089233398436,
      "original_ce_loss": 2.7219,
      "step": 1210
    },
    {
      "epoch": 0.4066666666666667,
      "grad_norm": 12.5612211227417,
      "learning_rate": 0.00029571001348968195,
      "loss": 43.4889892578125,
      "original_ce_loss": 2.7177,
      "step": 1220
    },
    {
      "epoch": 0.41,
      "grad_norm": 16.189455032348633,
      "learning_rate": 0.0002935496385407053,
      "loss": 43.246743774414064,
      "original_ce_loss": 2.7025,
      "step": 1230
    },
    {
      "epoch": 0.41333333333333333,
      "grad_norm": 11.432961463928223,
      "learning_rate": 0.00029137941539512187,
      "loss": 43.24491882324219,
      "original_ce_loss": 2.7024,
      "step": 1240
    },
    {
      "epoch": 0.4166666666666667,
      "grad_norm": 12.180628776550293,
      "learning_rate": 0.0002891996077528644,
      "loss": 43.221884155273436,
      "original_ce_loss": 2.701,
      "step": 1250
    },
    {
      "epoch": 0.42,
      "grad_norm": 12.448768615722656,
      "learning_rate": 0.00028701048047846033,
      "loss": 43.094137573242186,
      "original_ce_loss": 2.693,
      "step": 1260
    },
    {
      "epoch": 0.42333333333333334,
      "grad_norm": 11.567203521728516,
      "learning_rate": 0.0002848122995688491,
      "loss": 43.016519165039064,
      "original_ce_loss": 2.6881,
      "step": 1270
    },
    {
      "epoch": 0.4266666666666667,
      "grad_norm": 13.275468826293945,
      "learning_rate": 0.00028260533212106135,
      "loss": 42.92235412597656,
      "original_ce_loss": 2.6822,
      "step": 1280
    },
    {
      "epoch": 0.43,
      "grad_norm": 10.33851146697998,
      "learning_rate": 0.00028038984629976397,
      "loss": 42.876287841796874,
      "original_ce_loss": 2.6794,
      "step": 1290
    },
    {
      "epoch": 0.43333333333333335,
      "grad_norm": 14.586567878723145,
      "learning_rate": 0.0002781661113046763,
      "loss": 42.8263671875,
      "original_ce_loss": 2.6762,
      "step": 1300
    },
    {
      "epoch": 0.43666666666666665,
      "grad_norm": 12.394359588623047,
      "learning_rate": 0.0002759343973378601,
      "loss": 42.7820068359375,
      "original_ce_loss": 2.6735,
      "step": 1310
    },
    {
      "epoch": 0.44,
      "grad_norm": 10.222810745239258,
      "learning_rate": 0.00027369497557088753,
      "loss": 42.73019714355469,
      "original_ce_loss": 2.6702,
      "step": 1320
    },
    {
      "epoch": 0.44333333333333336,
      "grad_norm": 11.414692878723145,
      "learning_rate": 0.00027144811811189194,
      "loss": 42.600067138671875,
      "original_ce_loss": 2.6621,
      "step": 1330
    },
    {
      "epoch": 0.44666666666666666,
      "grad_norm": 12.681991577148438,
      "learning_rate": 0.0002691940979725043,
      "loss": 42.56636352539063,
      "original_ce_loss": 2.66,
      "step": 1340
    },
    {
      "epoch": 0.45,
      "grad_norm": 11.895796775817871,
      "learning_rate": 0.0002669331890346799,
      "loss": 42.446875,
      "original_ce_loss": 2.6525,
      "step": 1350
    },
    {
      "epoch": 0.4533333333333333,
      "grad_norm": 12.874837875366211,
      "learning_rate": 0.0002646656660174198,
      "loss": 42.47828979492188,
      "original_ce_loss": 2.6545,
      "step": 1360
    },
    {
      "epoch": 0.45666666666666667,
      "grad_norm": 13.898730278015137,
      "learning_rate": 0.00026239180444338994,
      "loss": 42.523861694335935,
      "original_ce_loss": 2.6573,
      "step": 1370
    },
    {
      "epoch": 0.46,
      "grad_norm": 11.172188758850098,
      "learning_rate": 0.0002601118806054429,
      "loss": 42.32704467773438,
      "original_ce_loss": 2.645,
      "step": 1380
    },
    {
      "epoch": 0.4633333333333333,
      "grad_norm": 10.018525123596191,
      "learning_rate": 0.0002578261715330462,
      "loss": 42.28916931152344,
      "original_ce_loss": 2.6427,
      "step": 1390
    },
    {
      "epoch": 0.4666666666666667,
      "grad_norm": 11.347637176513672,
      "learning_rate": 0.00025553495495862083,
      "loss": 42.21034545898438,
      "original_ce_loss": 2.6378,
      "step": 1400
    },
    {
      "epoch": 0.47,
      "grad_norm": 11.695727348327637,
      "learning_rate": 0.0002532385092837944,
      "loss": 42.206451416015625,
      "original_ce_loss": 2.6375,
      "step": 1410
    },
    {
      "epoch": 0.47333333333333333,
      "grad_norm": 12.01218032836914,
      "learning_rate": 0.00025093711354557335,
      "loss": 42.03828430175781,
      "original_ce_loss": 2.627,
      "step": 1420
    },
    {
      "epoch": 0.4766666666666667,
      "grad_norm": 13.935858726501465,
      "learning_rate": 0.00024863104738243725,
      "loss": 41.985330200195314,
      "original_ce_loss": 2.6237,
      "step": 1430
    },
    {
      "epoch": 0.48,
      "grad_norm": 11.189215660095215,
      "learning_rate": 0.00024632059100036086,
      "loss": 41.950033569335936,
      "original_ce_loss": 2.6215,
      "step": 1440
    },
    {
      "epoch": 0.48333333333333334,
      "grad_norm": 10.566532135009766,
      "learning_rate": 0.0002440060251387664,
      "loss": 41.87472839355469,
      "original_ce_loss": 2.6168,
      "step": 1450
    },
    {
      "epoch": 0.4866666666666667,
      "grad_norm": 11.969350814819336,
      "learning_rate": 0.00024168763103641171,
      "loss": 41.94146728515625,
      "original_ce_loss": 2.6209,
      "step": 1460
    },
    {
      "epoch": 0.49,
      "grad_norm": 10.838825225830078,
      "learning_rate": 0.00023936569039721732,
      "loss": 41.90030517578125,
      "original_ce_loss": 2.6184,
      "step": 1470
    },
    {
      "epoch": 0.49333333333333335,
      "grad_norm": 8.70041275024414,
      "learning_rate": 0.0002370404853560371,
      "loss": 41.783303833007814,
      "original_ce_loss": 2.6111,
      "step": 1480
    },
    {
      "epoch": 0.49666666666666665,
      "grad_norm": 11.917825698852539,
      "learning_rate": 0.00023471229844437637,
      "loss": 41.73698120117187,
      "original_ce_loss": 2.6082,
      "step": 1490
    },
    {
      "epoch": 0.5,
      "grad_norm": 14.329609870910645,
      "learning_rate": 0.00023238141255606254,
      "loss": 41.66865844726563,
      "original_ce_loss": 2.6039,
      "step": 1500
    },
    {
      "epoch": 0.5033333333333333,
      "grad_norm": 12.779980659484863,
      "learning_rate": 0.00023004811091287063,
      "loss": 41.554812622070315,
      "original_ce_loss": 2.5968,
      "step": 1510
    },
    {
      "epoch": 0.5066666666666667,
      "grad_norm": 13.428671836853027,
      "learning_rate": 0.00022771267703010974,
      "loss": 41.5691650390625,
      "original_ce_loss": 2.5977,
      "step": 1520
    },
    {
      "epoch": 0.51,
      "grad_norm": 11.918360710144043,
      "learning_rate": 0.00022537539468217365,
      "loss": 41.67054138183594,
      "original_ce_loss": 2.604,
      "step": 1530
    },
    {
      "epoch": 0.5133333333333333,
      "grad_norm": 10.47590160369873,
      "learning_rate": 0.00022303654786805973,
      "loss": 41.43465576171875,
      "original_ce_loss": 2.5893,
      "step": 1540
    },
    {
      "epoch": 0.5166666666666667,
      "grad_norm": 11.461536407470703,
      "learning_rate": 0.00022069642077686093,
      "loss": 41.446713256835935,
      "original_ce_loss": 2.59,
      "step": 1550
    },
    {
      "epoch": 0.52,
      "grad_norm": 9.24561595916748,
      "learning_rate": 0.00021835529775323444,
      "loss": 41.314047241210936,
      "original_ce_loss": 2.5817,
      "step": 1560
    },
    {
      "epoch": 0.5233333333333333,
      "grad_norm": 11.406410217285156,
      "learning_rate": 0.00021601346326285118,
      "loss": 41.320367431640626,
      "original_ce_loss": 2.5821,
      "step": 1570
    },
    {
      "epoch": 0.5266666666666666,
      "grad_norm": 12.614730834960938,
      "learning_rate": 0.00021367120185783136,
      "loss": 41.229287719726564,
      "original_ce_loss": 2.5764,
      "step": 1580
    },
    {
      "epoch": 0.53,
      "grad_norm": 10.465003967285156,
      "learning_rate": 0.00021132879814216864,
      "loss": 41.22291870117188,
      "original_ce_loss": 2.576,
      "step": 1590
    },
    {
      "epoch": 0.5333333333333333,
      "grad_norm": 10.220480918884277,
      "learning_rate": 0.00020898653673714882,
      "loss": 41.216635131835936,
      "original_ce_loss": 2.5757,
      "step": 1600
    },
    {
      "epoch": 0.5366666666666666,
      "grad_norm": 11.79033088684082,
      "learning_rate": 0.0002066447022467656,
      "loss": 41.119195556640626,
      "original_ce_loss": 2.5696,
      "step": 1610
    },
    {
      "epoch": 0.54,
      "grad_norm": 12.365918159484863,
      "learning_rate": 0.000204303579223139,
      "loss": 41.012420654296875,
      "original_ce_loss": 2.5629,
      "step": 1620
    },
    {
      "epoch": 0.5433333333333333,
      "grad_norm": 9.81035327911377,
      "learning_rate": 0.00020196345213194022,
      "loss": 40.986734008789064,
      "original_ce_loss": 2.5613,
      "step": 1630
    },
    {
      "epoch": 0.5466666666666666,
      "grad_norm": 10.371267318725586,
      "learning_rate": 0.00019962460531782638,
      "loss": 41.023834228515625,
      "original_ce_loss": 2.5636,
      "step": 1640
    },
    {
      "epoch": 0.55,
      "grad_norm": 9.000577926635742,
      "learning_rate": 0.00019728732296989032,
      "loss": 40.93628845214844,
      "original_ce_loss": 2.5581,
      "step": 1650
    },
    {
      "epoch": 0.5533333333333333,
      "grad_norm": 10.659809112548828,
      "learning_rate": 0.0001949518890871294,
      "loss": 40.941802978515625,
      "original_ce_loss": 2.5585,
      "step": 1660
    },
    {
      "epoch": 0.5566666666666666,
      "grad_norm": 9.251291275024414,
      "learning_rate": 0.00019261858744393747,
      "loss": 40.82298889160156,
      "original_ce_loss": 2.5511,
      "step": 1670
    },
    {
      "epoch": 0.56,
      "grad_norm": 10.543051719665527,
      "learning_rate": 0.00019028770155562366,
      "loss": 40.922186279296874,
      "original_ce_loss": 2.5573,
      "step": 1680
    },
    {
      "epoch": 0.5633333333333334,
      "grad_norm": 10.208642959594727,
      "learning_rate": 0.00018795951464396294,
      "loss": 40.73258972167969,
      "original_ce_loss": 2.5454,
      "step": 1690
    },
    {
      "epoch": 0.5666666666666667,
      "grad_norm": 14.32315731048584,
      "learning_rate": 0.00018563430960278265,
      "loss": 40.719366455078124,
      "original_ce_loss": 2.5446,
      "step": 1700
    },
    {
      "epoch": 0.57,
      "grad_norm": 13.02042007446289,
      "learning_rate": 0.00018331236896358826,
      "loss": 40.71995239257812,
      "original_ce_loss": 2.5446,
      "step": 1710
    },
    {
      "epoch": 0.5733333333333334,
      "grad_norm": 11.142119407653809,
      "learning_rate": 0.00018099397486123355,
      "loss": 40.743798828125,
      "original_ce_loss": 2.5461,
      "step": 1720
    },
    {
      "epoch": 0.5766666666666667,
      "grad_norm": 9.898249626159668,
      "learning_rate": 0.0001786794089996391,
      "loss": 40.69387512207031,
      "original_ce_loss": 2.543,
      "step": 1730
    },
    {
      "epoch": 0.58,
      "grad_norm": 30.856658935546875,
      "learning_rate": 0.00017636895261756276,
      "loss": 40.60739440917969,
      "original_ce_loss": 2.5376,
      "step": 1740
    },
    {
      "epoch": 0.5833333333333334,
      "grad_norm": 10.264257431030273,
      "learning_rate": 0.0001740628864544267,
      "loss": 40.628298950195315,
      "original_ce_loss": 2.5389,
      "step": 1750
    },
    {
      "epoch": 0.5866666666666667,
      "grad_norm": 11.025504112243652,
      "learning_rate": 0.00017176149071620564,
      "loss": 40.60066528320313,
      "original_ce_loss": 2.5372,
      "step": 1760
    },
    {
      "epoch": 0.59,
      "grad_norm": 12.942453384399414,
      "learning_rate": 0.0001694650450413792,
      "loss": 40.47138671875,
      "original_ce_loss": 2.5291,
      "step": 1770
    },
    {
      "epoch": 0.5933333333333334,
      "grad_norm": 11.431750297546387,
      "learning_rate": 0.00016717382846695382,
      "loss": 40.45237426757812,
      "original_ce_loss": 2.5279,
      "step": 1780
    },
    {
      "epoch": 0.5966666666666667,
      "grad_norm": 9.413339614868164,
      "learning_rate": 0.00016488811939455712,
      "loss": 40.36336669921875,
      "original_ce_loss": 2.5223,
      "step": 1790
    },
    {
      "epoch": 0.6,
      "grad_norm": 8.911304473876953,
      "learning_rate": 0.00016260819555661006,
      "loss": 40.387020874023435,
      "original_ce_loss": 2.5238,
      "step": 1800
    },
    {
      "epoch": 0.6033333333333334,
      "grad_norm": 8.465132713317871,
      "learning_rate": 0.00016033433398258019,
      "loss": 40.344757080078125,
      "original_ce_loss": 2.5212,
      "step": 1810
    },
    {
      "epoch": 0.6066666666666667,
      "grad_norm": 10.04110336303711,
      "learning_rate": 0.00015806681096532002,
      "loss": 40.456539916992185,
      "original_ce_loss": 2.5282,
      "step": 1820
    },
    {
      "epoch": 0.61,
      "grad_norm": 9.178019523620605,
      "learning_rate": 0.00015580590202749568,
      "loss": 40.34690551757812,
      "original_ce_loss": 2.5213,
      "step": 1830
    },
    {
      "epoch": 0.6133333333333333,
      "grad_norm": 7.764488697052002,
      "learning_rate": 0.00015355188188810807,
      "loss": 40.279901123046876,
      "original_ce_loss": 2.5171,
      "step": 1840
    },
    {
      "epoch": 0.6166666666666667,
      "grad_norm": 8.292529106140137,
      "learning_rate": 0.00015130502442911248,
      "loss": 40.32528076171875,
      "original_ce_loss": 2.52,
      "step": 1850
    },
    {
      "epoch": 0.62,
      "grad_norm": 8.208005905151367,
      "learning_rate": 0.00014906560266213996,
      "loss": 40.189996337890626,
      "original_ce_loss": 2.5115,
      "step": 1860
    },
    {
      "epoch": 0.6233333333333333,
      "grad_norm": 9.50411605834961,
      "learning_rate": 0.00014683388869532368,
      "loss": 40.22713928222656,
      "original_ce_loss": 2.5138,
      "step": 1870
    },
    {
      "epoch": 0.6266666666666667,
      "grad_norm": 10.043655395507812,
      "learning_rate": 0.00014461015370023604,
      "loss": 40.089666748046874,
      "original_ce_loss": 2.5052,
      "step": 1880
    },
    {
      "epoch": 0.63,
      "grad_norm": 8.17435073852539,
      "learning_rate": 0.00014239466787893866,
      "loss": 40.06712951660156,
      "original_ce_loss": 2.5038,
      "step": 1890
    },
    {
      "epoch": 0.6333333333333333,
      "grad_norm": 8.872373580932617,
      "learning_rate": 0.00014018770043115083,
      "loss": 40.13961486816406,
      "original_ce_loss": 2.5084,
      "step": 1900
    },
    {
      "epoch": 0.6366666666666667,
      "grad_norm": 9.705024719238281,
      "learning_rate": 0.0001379895195215397,
      "loss": 40.03070068359375,
      "original_ce_loss": 2.5015,
      "step": 1910
    },
    {
      "epoch": 0.64,
      "grad_norm": 8.770875930786133,
      "learning_rate": 0.00013580039224713565,
      "loss": 40.00947570800781,
      "original_ce_loss": 2.5002,
      "step": 1920
    },
    {
      "epoch": 0.6433333333333333,
      "grad_norm": 8.644168853759766,
      "learning_rate": 0.00013362058460487805,
      "loss": 40.012994384765626,
      "original_ce_loss": 2.5004,
      "step": 1930
    },
    {
      "epoch": 0.6466666666666666,
      "grad_norm": 7.846497058868408,
      "learning_rate": 0.0001314503614592947,
      "loss": 40.023431396484376,
      "original_ce_loss": 2.5011,
      "step": 1940
    },
    {
      "epoch": 0.65,
      "grad_norm": 8.981230735778809,
      "learning_rate": 0.0001292899865103181,
      "loss": 39.95662231445313,
      "original_ce_loss": 2.4969,
      "step": 1950
    },
    {
      "epoch": 0.6533333333333333,
      "grad_norm": 9.10230541229248,
      "learning_rate": 0.00012713972226124402,
      "loss": 40.002069091796876,
      "original_ce_loss": 2.4998,
      "step": 1960
    },
    {
      "epoch": 0.6566666666666666,
      "grad_norm": 8.652543067932129,
      "learning_rate": 0.00012499982998683521,
      "loss": 39.91124572753906,
      "original_ce_loss": 2.4941,
      "step": 1970
    },
    {
      "epoch": 0.66,
      "grad_norm": 8.03935432434082,
      "learning_rate": 0.00012287056970157446,
      "loss": 39.8054443359375,
      "original_ce_loss": 2.4875,
      "step": 1980
    },
    {
      "epoch": 0.6633333333333333,
      "grad_norm": 9.71660041809082,
      "learning_rate": 0.00012075220012807027,
      "loss": 39.81572570800781,
      "original_ce_loss": 2.4881,
      "step": 1990
    },
    {
      "epoch": 0.6666666666666666,
      "grad_norm": 8.720709800720215,
      "learning_rate": 0.00011864497866562052,
      "loss": 39.940530395507814,
      "original_ce_loss": 2.4959,
      "step": 2000
    },
    {
      "epoch": 0.67,
      "grad_norm": 8.757256507873535,
      "learning_rate": 0.00011654916135893604,
      "loss": 39.902182006835936,
      "original_ce_loss": 2.4935,
      "step": 2010
    },
    {
      "epoch": 0.6733333333333333,
      "grad_norm": 9.392760276794434,
      "learning_rate": 0.00011446500286702886,
      "loss": 39.86787109375,
      "original_ce_loss": 2.4914,
      "step": 2020
    },
    {
      "epoch": 0.6766666666666666,
      "grad_norm": 9.689042091369629,
      "learning_rate": 0.00011239275643226945,
      "loss": 39.822994995117185,
      "original_ce_loss": 2.4886,
      "step": 2030
    },
    {
      "epoch": 0.68,
      "grad_norm": 9.171027183532715,
      "learning_rate": 0.00011033267384961554,
      "loss": 39.860556030273436,
      "original_ce_loss": 2.4909,
      "step": 2040
    },
    {
      "epoch": 0.6833333333333333,
      "grad_norm": 7.839738368988037,
      "learning_rate": 0.00010828500543601709,
      "loss": 39.73720092773438,
      "original_ce_loss": 2.4832,
      "step": 2050
    },
    {
      "epoch": 0.6866666666666666,
      "grad_norm": 8.569633483886719,
      "learning_rate": 0.00010625000000000003,
      "loss": 39.617706298828125,
      "original_ce_loss": 2.4757,
      "step": 2060
    },
    {
      "epoch": 0.69,
      "grad_norm": 8.060338973999023,
      "learning_rate": 0.0001042279048114351,
      "loss": 39.69352416992187,
      "original_ce_loss": 2.4805,
      "step": 2070
    },
    {
      "epoch": 0.6933333333333334,
      "grad_norm": 8.348586082458496,
      "learning_rate": 0.00010221896557149182,
      "loss": 39.686151123046876,
      "original_ce_loss": 2.48,
      "step": 2080
    },
    {
      "epoch": 0.6966666666666667,
      "grad_norm": 8.033924102783203,
      "learning_rate": 0.00010022342638278375,
      "loss": 39.68394165039062,
      "original_ce_loss": 2.4799,
      "step": 2090
    },
    {
      "epoch": 0.7,
      "grad_norm": 8.12784194946289,
      "learning_rate": 9.824152971970839e-05,
      "loss": 39.73549194335938,
      "original_ce_loss": 2.4831,
      "step": 2100
    },
    {
      "epoch": 0.7033333333333334,
      "grad_norm": 9.852293968200684,
      "learning_rate": 9.627351639898432e-05,
      "loss": 39.60615539550781,
      "original_ce_loss": 2.475,
      "step": 2110
    },
    {
      "epoch": 0.7066666666666667,
      "grad_norm": 8.127728462219238,
      "learning_rate": 9.431962555038987e-05,
      "loss": 39.702090454101565,
      "original_ce_loss": 2.481,
      "step": 2120
    },
    {
      "epoch": 0.71,
      "grad_norm": 7.37929630279541,
      "learning_rate": 9.238009458770722e-05,
      "loss": 39.58073425292969,
      "original_ce_loss": 2.4734,
      "step": 2130
    },
    {
      "epoch": 0.7133333333333334,
      "grad_norm": 8.563313484191895,
      "learning_rate": 9.045515917987448e-05,
      "loss": 39.58157043457031,
      "original_ce_loss": 2.4735,
      "step": 2140
    },
    {
      "epoch": 0.7166666666666667,
      "grad_norm": 7.394179821014404,
      "learning_rate": 8.854505322234993e-05,
      "loss": 39.54279479980469,
      "original_ce_loss": 2.4711,
      "step": 2150
    },
    {
      "epoch": 0.72,
      "grad_norm": 7.087237358093262,
      "learning_rate": 8.665000880869176e-05,
      "loss": 39.50851745605469,
      "original_ce_loss": 2.4689,
      "step": 2160
    },
    {
      "epoch": 0.7233333333333334,
      "grad_norm": 7.820181846618652,
      "learning_rate": 8.477025620235717e-05,
      "loss": 39.505514526367186,
      "original_ce_loss": 2.4687,
      "step": 2170
    },
    {
      "epoch": 0.7266666666666667,
      "grad_norm": 7.245079040527344,
      "learning_rate": 8.290602380872308e-05,
      "loss": 39.5278564453125,
      "original_ce_loss": 2.4701,
      "step": 2180
    },
    {
      "epoch": 0.73,
      "grad_norm": 8.576043128967285,
      "learning_rate": 8.105753814733303e-05,
      "loss": 39.471343994140625,
      "original_ce_loss": 2.4666,
      "step": 2190
    },
    {
      "epoch": 0.7333333333333333,
      "grad_norm": 8.827922821044922,
      "learning_rate": 7.922502382437337e-05,
      "loss": 39.40574645996094,
      "original_ce_loss": 2.4625,
      "step": 2200
    },
    {
      "epoch": 0.7366666666666667,
      "grad_norm": 7.869954586029053,
      "learning_rate": 7.740870350538184e-05,
      "loss": 39.3803955078125,
      "original_ce_loss": 2.4609,
      "step": 2210
    },
    {
      "epoch": 0.74,
      "grad_norm": 7.113214492797852,
      "learning_rate": 7.560879788819137e-05,
      "loss": 39.3686279296875,
      "original_ce_loss": 2.4602,
      "step": 2220
    },
    {
      "epoch": 0.7433333333333333,
      "grad_norm": 7.480007648468018,
      "learning_rate": 7.382552567611415e-05,
      "loss": 39.250030517578125,
      "original_ce_loss": 2.4528,
      "step": 2230
    },
    {
      "epoch": 0.7466666666666667,
      "grad_norm": 8.339986801147461,
      "learning_rate": 7.205910355136702e-05,
      "loss": 39.36780700683594,
      "original_ce_loss": 2.4601,
      "step": 2240
    },
    {
      "epoch": 0.75,
      "grad_norm": 7.813916206359863,
      "learning_rate": 7.03097461487427e-05,
      "loss": 39.40064392089844,
      "original_ce_loss": 2.4622,
      "step": 2250
    },
    {
      "epoch": 0.7533333333333333,
      "grad_norm": 8.291626930236816,
      "learning_rate": 6.857766602953006e-05,
      "loss": 39.31941528320313,
      "original_ce_loss": 2.4571,
      "step": 2260
    },
    {
      "epoch": 0.7566666666666667,
      "grad_norm": 7.5233869552612305,
      "learning_rate": 6.686307365568631e-05,
      "loss": 39.260711669921875,
      "original_ce_loss": 2.4534,
      "step": 2270
    },
    {
      "epoch": 0.76,
      "grad_norm": 7.651540756225586,
      "learning_rate": 6.516617736426403e-05,
      "loss": 39.325738525390626,
      "original_ce_loss": 2.4575,
      "step": 2280
    },
    {
      "epoch": 0.7633333333333333,
      "grad_norm": 6.938567161560059,
      "learning_rate": 6.348718334209617e-05,
      "loss": 39.248715209960935,
      "original_ce_loss": 2.4527,
      "step": 2290
    },
    {
      "epoch": 0.7666666666666667,
      "grad_norm": 6.557342052459717,
      "learning_rate": 6.182629560074332e-05,
      "loss": 39.28128662109375,
      "original_ce_loss": 2.4547,
      "step": 2300
    },
    {
      "epoch": 0.77,
      "grad_norm": 7.514851093292236,
      "learning_rate": 6.018371595170414e-05,
      "loss": 39.292431640625,
      "original_ce_loss": 2.4554,
      "step": 2310
    },
    {
      "epoch": 0.7733333333333333,
      "grad_norm": 7.411563873291016,
      "learning_rate": 5.8559643981893595e-05,
      "loss": 39.299197387695315,
      "original_ce_loss": 2.4558,
      "step": 2320
    },
    {
      "epoch": 0.7766666666666666,
      "grad_norm": 6.666036605834961,
      "learning_rate": 5.695427702939189e-05,
      "loss": 39.24348754882813,
      "original_ce_loss": 2.4524,
      "step": 2330
    },
    {
      "epoch": 0.78,
      "grad_norm": 7.235419750213623,
      "learning_rate": 5.536781015946593e-05,
      "loss": 39.20620422363281,
      "original_ce_loss": 2.45,
      "step": 2340
    },
    {
      "epoch": 0.7833333333333333,
      "grad_norm": 7.544335842132568,
      "learning_rate": 5.3800436140867365e-05,
      "loss": 39.27761535644531,
      "original_ce_loss": 2.4545,
      "step": 2350
    },
    {
      "epoch": 0.7866666666666666,
      "grad_norm": 6.542551517486572,
      "learning_rate": 5.225234542240957e-05,
      "loss": 39.25896301269531,
      "original_ce_loss": 2.4533,
      "step": 2360
    },
    {
      "epoch": 0.79,
      "grad_norm": 7.0305047035217285,
      "learning_rate": 5.0723726109826774e-05,
      "loss": 39.202761840820315,
      "original_ce_loss": 2.4498,
      "step": 2370
    },
    {
      "epoch": 0.7933333333333333,
      "grad_norm": 6.597095012664795,
      "learning_rate": 4.9214763942917504e-05,
      "loss": 39.177285766601564,
      "original_ce_loss": 2.4482,
      "step": 2380
    },
    {
      "epoch": 0.7966666666666666,
      "grad_norm": 7.227161884307861,
      "learning_rate": 4.772564227297543e-05,
      "loss": 39.163336181640624,
      "original_ce_loss": 2.4473,
      "step": 2390
    },
    {
      "epoch": 0.8,
      "grad_norm": 6.851435661315918,
      "learning_rate": 4.625654204051122e-05,
      "loss": 39.124533081054686,
      "original_ce_loss": 2.4449,
      "step": 2400
    },
    {
      "epoch": 0.8033333333333333,
      "grad_norm": 7.683741569519043,
      "learning_rate": 4.4807641753266386e-05,
      "loss": 39.1417724609375,
      "original_ce_loss": 2.446,
      "step": 2410
    },
    {
      "epoch": 0.8066666666666666,
      "grad_norm": 6.8996477127075195,
      "learning_rate": 4.337911746452307e-05,
      "loss": 39.13313293457031,
      "original_ce_loss": 2.4455,
      "step": 2420
    },
    {
      "epoch": 0.81,
      "grad_norm": 7.35669469833374,
      "learning_rate": 4.1971142751712525e-05,
      "loss": 39.178994750976564,
      "original_ce_loss": 2.4483,
      "step": 2430
    },
    {
      "epoch": 0.8133333333333334,
      "grad_norm": 6.518770694732666,
      "learning_rate": 4.0583888695323685e-05,
      "loss": 39.073080444335936,
      "original_ce_loss": 2.4417,
      "step": 2440
    },
    {
      "epoch": 0.8166666666666667,
      "grad_norm": 6.807189464569092,
      "learning_rate": 3.921752385811548e-05,
      "loss": 39.06864929199219,
      "original_ce_loss": 2.4414,
      "step": 2450
    },
    {
      "epoch": 0.82,
      "grad_norm": 7.491281509399414,
      "learning_rate": 3.787221426463542e-05,
      "loss": 39.04958801269531,
      "original_ce_loss": 2.4402,
      "step": 2460
    },
    {
      "epoch": 0.8233333333333334,
      "grad_norm": 6.779686450958252,
      "learning_rate": 3.654812338104574e-05,
      "loss": 38.97378234863281,
      "original_ce_loss": 2.4355,
      "step": 2470
    },
    {
      "epoch": 0.8266666666666667,
      "grad_norm": 6.429449081420898,
      "learning_rate": 3.524541209526144e-05,
      "loss": 39.070925903320315,
      "original_ce_loss": 2.4416,
      "step": 2480
    },
    {
      "epoch": 0.83,
      "grad_norm": 7.245244979858398,
      "learning_rate": 3.3964238697400586e-05,
      "loss": 39.09231872558594,
      "original_ce_loss": 2.4429,
      "step": 2490
    },
    {
      "epoch": 0.8333333333333334,
      "grad_norm": 6.734501838684082,
      "learning_rate": 3.27047588605512e-05,
      "loss": 38.96620483398438,
      "original_ce_loss": 2.435,
      "step": 2500
    },
    {
      "epoch": 0.8366666666666667,
      "grad_norm": 6.67666482925415,
      "learning_rate": 3.1467125621855524e-05,
      "loss": 39.005157470703125,
      "original_ce_loss": 2.4375,
      "step": 2510
    },
    {
      "epoch": 0.84,
      "grad_norm": 6.6135077476501465,
      "learning_rate": 3.0251489363914502e-05,
      "loss": 39.00813598632813,
      "original_ce_loss": 2.4376,
      "step": 2520
    },
    {
      "epoch": 0.8433333333333334,
      "grad_norm": 6.659499168395996,
      "learning_rate": 2.9057997796515544e-05,
      "loss": 39.01441650390625,
      "original_ce_loss": 2.438,
      "step": 2530
    },
    {
      "epoch": 0.8466666666666667,
      "grad_norm": 6.484838962554932,
      "learning_rate": 2.788679593868422e-05,
      "loss": 38.977987670898436,
      "original_ce_loss": 2.4358,
      "step": 2540
    },
    {
      "epoch": 0.85,
      "grad_norm": 6.826574325561523,
      "learning_rate": 2.673802610106332e-05,
      "loss": 39.01756896972656,
      "original_ce_loss": 2.4382,
      "step": 2550
    },
    {
      "epoch": 0.8533333333333334,
      "grad_norm": 6.7142157554626465,
      "learning_rate": 2.5611827868621095e-05,
      "loss": 38.96328430175781,
      "original_ce_loss": 2.4348,
      "step": 2560
    },
    {
      "epoch": 0.8566666666666667,
      "grad_norm": 7.200511932373047,
      "learning_rate": 2.4508338083690248e-05,
      "loss": 38.98872680664063,
      "original_ce_loss": 2.4364,
      "step": 2570
    },
    {
      "epoch": 0.86,
      "grad_norm": 6.844103813171387,
      "learning_rate": 2.342769082934085e-05,
      "loss": 38.981326293945315,
      "original_ce_loss": 2.436,
      "step": 2580
    },
    {
      "epoch": 0.8633333333333333,
      "grad_norm": 5.781346797943115,
      "learning_rate": 2.237001741308776e-05,
      "loss": 39.01437072753906,
      "original_ce_loss": 2.438,
      "step": 2590
    },
    {
      "epoch": 0.8666666666666667,
      "grad_norm": 5.936890125274658,
      "learning_rate": 2.1335446350935984e-05,
      "loss": 39.051177978515625,
      "original_ce_loss": 2.4403,
      "step": 2600
    },
    {
      "epoch": 0.87,
      "grad_norm": 6.188159465789795,
      "learning_rate": 2.0324103351764793e-05,
      "loss": 38.984521484375,
      "original_ce_loss": 2.4362,
      "step": 2610
    },
    {
      "epoch": 0.8733333333333333,
      "grad_norm": 6.051525115966797,
      "learning_rate": 1.9336111302052965e-05,
      "loss": 39.01229248046875,
      "original_ce_loss": 2.4379,
      "step": 2620
    },
    {
      "epoch": 0.8766666666666667,
      "grad_norm": 6.345283031463623,
      "learning_rate": 1.837159025094729e-05,
      "loss": 39.01556396484375,
      "original_ce_loss": 2.4381,
      "step": 2630
    },
    {
      "epoch": 0.88,
      "grad_norm": 6.247714519500732,
      "learning_rate": 1.7430657395675505e-05,
      "loss": 38.99752197265625,
      "original_ce_loss": 2.437,
      "step": 2640
    },
    {
      "epoch": 0.8833333333333333,
      "grad_norm": 6.06412935256958,
      "learning_rate": 1.651342706730574e-05,
      "loss": 38.990499877929686,
      "original_ce_loss": 2.4365,
      "step": 2650
    },
    {
      "epoch": 0.8866666666666667,
      "grad_norm": 6.256269931793213,
      "learning_rate": 1.5620010716854626e-05,
      "loss": 38.9446044921875,
      "original_ce_loss": 2.4337,
      "step": 2660
    },
    {
      "epoch": 0.89,
      "grad_norm": 5.881938457489014,
      "learning_rate": 1.4750516901744854e-05,
      "loss": 38.98244018554688,
      "original_ce_loss": 2.436,
      "step": 2670
    },
    {
      "epoch": 0.8933333333333333,
      "grad_norm": 6.173046588897705,
      "learning_rate": 1.3905051272614592e-05,
      "loss": 38.942709350585936,
      "original_ce_loss": 2.4336,
      "step": 2680
    },
    {
      "epoch": 0.8966666666666666,
      "grad_norm": 6.209837913513184,
      "learning_rate": 1.3083716560480194e-05,
      "loss": 38.930197143554686,
      "original_ce_loss": 2.4328,
      "step": 2690
    },
    {
      "epoch": 0.9,
      "grad_norm": 5.919473171234131,
      "learning_rate": 1.228661256425346e-05,
      "loss": 38.91124572753906,
      "original_ce_loss": 2.4316,
      "step": 2700
    },
    {
      "epoch": 0.9033333333333333,
      "grad_norm": 5.999595642089844,
      "learning_rate": 1.151383613861514e-05,
      "loss": 38.893026733398436,
      "original_ce_loss": 2.4305,
      "step": 2710
    },
    {
      "epoch": 0.9066666666666666,
      "grad_norm": 5.834168434143066,
      "learning_rate": 1.076548118224649e-05,
      "loss": 38.92605285644531,
      "original_ce_loss": 2.4325,
      "step": 2720
    },
    {
      "epoch": 0.91,
      "grad_norm": 5.635613441467285,
      "learning_rate": 1.0041638626419734e-05,
      "loss": 38.94969482421875,
      "original_ce_loss": 2.434,
      "step": 2730
    },
    {
      "epoch": 0.9133333333333333,
      "grad_norm": 5.8296918869018555,
      "learning_rate": 9.342396423949e-06,
      "loss": 38.863006591796875,
      "original_ce_loss": 2.4286,
      "step": 2740
    },
    {
      "epoch": 0.9166666666666666,
      "grad_norm": 5.931970596313477,
      "learning_rate": 8.66783953850366e-06,
      "loss": 38.86027526855469,
      "original_ce_loss": 2.4284,
      "step": 2750
    },
    {
      "epoch": 0.92,
      "grad_norm": 5.655665874481201,
      "learning_rate": 8.018049934284286e-06,
      "loss": 38.78544311523437,
      "original_ce_loss": 2.4237,
      "step": 2760
    },
    {
      "epoch": 0.9233333333333333,
      "grad_norm": 5.667448043823242,
      "learning_rate": 7.39310656606343e-06,
      "loss": 38.823794555664065,
      "original_ce_loss": 2.4261,
      "step": 2770
    },
    {
      "epoch": 0.9266666666666666,
      "grad_norm": 5.631087779998779,
      "learning_rate": 6.79308536959188e-06,
      "loss": 38.836630249023436,
      "original_ce_loss": 2.4269,
      "step": 2780
    },
    {
      "epoch": 0.93,
      "grad_norm": 5.525851726531982,
      "learning_rate": 6.218059252372003e-06,
      "loss": 38.90266418457031,
      "original_ce_loss": 2.4311,
      "step": 2790
    },
    {
      "epoch": 0.9333333333333333,
      "grad_norm": 5.5109453201293945,
      "learning_rate": 5.668098084798779e-06,
      "loss": 38.83353271484375,
      "original_ce_loss": 2.4267,
      "step": 2800
    },
    {
      "epoch": 0.9366666666666666,
      "grad_norm": 5.430550575256348,
      "learning_rate": 5.143268691669931e-06,
      "loss": 38.83726806640625,
      "original_ce_loss": 2.427,
      "step": 2810
    },
    {
      "epoch": 0.94,
      "grad_norm": 7.3637590408325195,
      "learning_rate": 4.643634844066291e-06,
      "loss": 38.780398559570315,
      "original_ce_loss": 2.4234,
      "step": 2820
    },
    {
      "epoch": 0.9433333333333334,
      "grad_norm": 5.254461765289307,
      "learning_rate": 4.169257251603011e-06,
      "loss": 38.791683959960935,
      "original_ce_loss": 2.4241,
      "step": 2830
    },
    {
      "epoch": 0.9466666666666667,
      "grad_norm": 5.669284820556641,
      "learning_rate": 3.720193555052807e-06,
      "loss": 38.850555419921875,
      "original_ce_loss": 2.4278,
      "step": 2840
    },
    {
      "epoch": 0.95,
      "grad_norm": 5.320619106292725,
      "learning_rate": 3.2964983193422225e-06,
      "loss": 38.844931030273436,
      "original_ce_loss": 2.4274,
      "step": 2850
    },
    {
      "epoch": 0.9533333333333334,
      "grad_norm": 5.563847541809082,
      "learning_rate": 2.898223026921506e-06,
      "loss": 38.778683471679685,
      "original_ce_loss": 2.4233,
      "step": 2860
    },
    {
      "epoch": 0.9566666666666667,
      "grad_norm": 5.251751899719238,
      "learning_rate": 2.5254160715090346e-06,
      "loss": 38.865536499023435,
      "original_ce_loss": 2.4287,
      "step": 2870
    },
    {
      "epoch": 0.96,
      "grad_norm": 5.600255012512207,
      "learning_rate": 2.178122752211109e-06,
      "loss": 38.9463134765625,
      "original_ce_loss": 2.4338,
      "step": 2880
    },
    {
      "epoch": 0.9633333333333334,
      "grad_norm": 5.421494960784912,
      "learning_rate": 1.856385268017771e-06,
      "loss": 38.89384155273437,
      "original_ce_loss": 2.4305,
      "step": 2890
    },
    {
      "epoch": 0.9666666666666667,
      "grad_norm": 5.484427452087402,
      "learning_rate": 1.5602427126752638e-06,
      "loss": 38.77872009277344,
      "original_ce_loss": 2.4233,
      "step": 2900
    },
    {
      "epoch": 0.97,
      "grad_norm": 5.447332382202148,
      "learning_rate": 1.2897310699357345e-06,
      "loss": 38.856207275390624,
      "original_ce_loss": 2.4282,
      "step": 2910
    },
    {
      "epoch": 0.9733333333333334,
      "grad_norm": 5.485825061798096,
      "learning_rate": 1.044883209185074e-06,
      "loss": 38.89476013183594,
      "original_ce_loss": 2.4306,
      "step": 2920
    },
    {
      "epoch": 0.9766666666666667,
      "grad_norm": 5.441553115844727,
      "learning_rate": 8.257288814488686e-07,
      "loss": 38.93569641113281,
      "original_ce_loss": 2.4331,
      "step": 2930
    },
    {
      "epoch": 0.98,
      "grad_norm": 5.285153865814209,
      "learning_rate": 6.322947157774788e-07,
      "loss": 38.84034729003906,
      "original_ce_loss": 2.4272,
      "step": 2940
    },
    {
      "epoch": 0.9833333333333333,
      "grad_norm": 5.2883734703063965,
      "learning_rate": 4.646042160104113e-07,
      "loss": 38.926202392578126,
      "original_ce_loss": 2.4325,
      "step": 2950
    },
    {
      "epoch": 0.9866666666666667,
      "grad_norm": 5.316468715667725,
      "learning_rate": 3.22677757920313e-07,
      "loss": 38.85401611328125,
      "original_ce_loss": 2.428,
      "step": 2960
    },
    {
      "epoch": 0.99,
      "grad_norm": 5.307882785797119,
      "learning_rate": 2.065325867372486e-07,
      "loss": 38.951901245117185,
      "original_ce_loss": 2.4341,
      "step": 2970
    },
    {
      "epoch": 0.9933333333333333,
      "grad_norm": 5.168679714202881,
      "learning_rate": 1.1618281505321459e-07,
      "loss": 38.893191528320315,
      "original_ce_loss": 2.4305,
      "step": 2980
    },
    {
      "epoch": 0.9966666666666667,
      "grad_norm": 6.1123762130737305,
      "learning_rate": 5.163942110740782e-08,
      "loss": 38.831634521484375,
      "original_ce_loss": 2.4266,
      "step": 2990
    },
    {
      "epoch": 1.0,
      "grad_norm": 5.376128196716309,
      "learning_rate": 1.2910247452108114e-08,
      "loss": 38.82344360351563,
      "original_ce_loss": 2.4261,
      "step": 3000
    }
  ],
  "logging_steps": 10,
  "max_steps": 3000,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 9223372036854775807,
  "save_steps": 200,
  "stateful_callbacks": {
    "TrainerControl": {
      "args": {
        "should_epoch_stop": false,
        "should_evaluate": false,
        "should_log": false,
        "should_save": true,
        "should_training_stop": true
      },
      "attributes": {}
    }
  },
  "total_flos": 1.11674789462016e+17,
  "train_batch_size": 16,
  "trial_name": null,
  "trial_params": null
}