{
  "best_global_step": null,
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 4.0,
  "eval_steps": 500,
  "global_step": 968,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.041483017889551464,
      "grad_norm": 194.1380615234375,
      "learning_rate": 2.975e-05,
      "loss": 288.27666015625,
      "original_ce_loss": 9.008,
      "step": 10
    },
    {
      "epoch": 0.08296603577910293,
      "grad_norm": 201.3096160888672,
      "learning_rate": 7.225000000000001e-05,
      "loss": 286.1044189453125,
      "original_ce_loss": 8.9401,
      "step": 20
    },
    {
      "epoch": 0.1244490536686544,
      "grad_norm": 330.6499328613281,
      "learning_rate": 0.00011475,
      "loss": 275.3582275390625,
      "original_ce_loss": 8.6043,
      "step": 30
    },
    {
      "epoch": 0.16593207155820586,
      "grad_norm": 88.29187774658203,
      "learning_rate": 0.00015299999999999998,
      "loss": 240.4803466796875,
      "original_ce_loss": 7.5143,
      "step": 40
    },
    {
      "epoch": 0.20741508944775733,
      "grad_norm": 60.00925827026367,
      "learning_rate": 0.0001955,
      "loss": 226.5456298828125,
      "original_ce_loss": 7.0789,
      "step": 50
    },
    {
      "epoch": 0.2488981073373088,
      "grad_norm": 46.71499252319336,
      "learning_rate": 0.000238,
      "loss": 212.7708251953125,
      "original_ce_loss": 6.6484,
      "step": 60
    },
    {
      "epoch": 0.29038112522686027,
      "grad_norm": NaN,
      "learning_rate": 0.0002805,
      "loss": 202.953662109375,
      "original_ce_loss": 6.3416,
      "step": 70
    },
    {
      "epoch": 0.3318641431164117,
      "grad_norm": NaN,
      "learning_rate": 0.00029749999999999997,
      "loss": 209.846923828125,
      "original_ce_loss": 6.557,
      "step": 80
    },
    {
      "epoch": 0.37334716100596316,
      "grad_norm": 60.61952209472656,
      "learning_rate": 0.0003315,
      "loss": 203.94407958984374,
      "original_ce_loss": 6.3726,
      "step": 90
    },
    {
      "epoch": 0.41483017889551466,
      "grad_norm": 144.21949768066406,
      "learning_rate": 0.000374,
      "loss": 199.93363037109376,
      "original_ce_loss": 6.2473,
      "step": 100
    },
    {
      "epoch": 0.4563131967850661,
      "grad_norm": 18.162376403808594,
      "learning_rate": 0.0004165,
      "loss": 198.72691650390624,
      "original_ce_loss": 6.2096,
      "step": 110
    },
    {
      "epoch": 0.4977962146746176,
      "grad_norm": 32.539649963378906,
      "learning_rate": 0.0004216803067723292,
      "loss": 196.5206298828125,
      "original_ce_loss": 6.1406,
      "step": 120
    },
    {
      "epoch": 0.5392792325641691,
      "grad_norm": 30.620323181152344,
      "learning_rate": 0.0004083716200231366,
      "loss": 193.46846923828124,
      "original_ce_loss": 6.0452,
      "step": 130
    },
    {
      "epoch": 0.5807622504537205,
      "grad_norm": 18.298351287841797,
      "learning_rate": 0.00038551471342450957,
      "loss": 190.6060546875,
      "original_ce_loss": 5.9558,
      "step": 140
    },
    {
      "epoch": 0.622245268343272,
      "grad_norm": 25.024852752685547,
      "learning_rate": 0.00035422380030278826,
      "loss": 187.1262451171875,
      "original_ce_loss": 5.8471,
      "step": 150
    },
    {
      "epoch": 0.6637282862328234,
      "grad_norm": 27.06348419189453,
      "learning_rate": 0.0003201042547134993,
      "loss": 183.2828857421875,
      "original_ce_loss": 5.727,
      "step": 160
    },
    {
      "epoch": 0.7052113041223749,
      "grad_norm": 22.74102210998535,
      "learning_rate": 0.0002772712258541234,
      "loss": 179.60921630859374,
      "original_ce_loss": 5.6122,
      "step": 170
    },
    {
      "epoch": 0.7466943220119263,
      "grad_norm": 28.52726173400879,
      "learning_rate": 0.00023128077222693525,
      "loss": 175.43724365234374,
      "original_ce_loss": 5.4818,
      "step": 180
    },
    {
      "epoch": 0.7881773399014779,
      "grad_norm": 34.97484588623047,
      "learning_rate": 0.00018437480596885273,
      "loss": 171.78480224609376,
      "original_ce_loss": 5.3677,
      "step": 190
    },
    {
      "epoch": 0.8296603577910293,
      "grad_norm": 27.37944984436035,
      "learning_rate": 0.0001388398680194434,
      "loss": 168.92047119140625,
      "original_ce_loss": 5.2782,
      "step": 200
    },
    {
      "epoch": 0.8711433756805808,
      "grad_norm": 33.82849884033203,
      "learning_rate": 0.000418625,
      "loss": 166.3930908203125,
      "original_ce_loss": 5.1992,
      "step": 210
    },
    {
      "epoch": 0.9126263935701322,
      "grad_norm": 61.042945861816406,
      "learning_rate": 0.00042436324689416107,
      "loss": 161.55086669921874,
      "original_ce_loss": 5.0479,
      "step": 220
    },
    {
      "epoch": 0.9541094114596836,
      "grad_norm": 60.506507873535156,
      "learning_rate": 0.00042125363916220815,
      "loss": 158.47862548828124,
      "original_ce_loss": 4.9519,
      "step": 230
    },
    {
      "epoch": 0.9955924293492352,
      "grad_norm": 34.90744400024414,
      "learning_rate": 0.0004155921868767752,
      "loss": 153.83856201171875,
      "original_ce_loss": 4.8069,
      "step": 240
    },
    {
      "epoch": 1.0331864143116412,
      "grad_norm": 34.04861831665039,
      "learning_rate": 0.00040744809671726373,
      "loss": 136.09080810546874,
      "original_ce_loss": 4.6922,
      "step": 250
    },
    {
      "epoch": 1.0746694322011927,
      "grad_norm": 36.69175338745117,
      "learning_rate": 0.0003969209236049676,
      "loss": 146.020751953125,
      "original_ce_loss": 4.5626,
      "step": 260
    },
    {
      "epoch": 1.116152450090744,
      "grad_norm": 55.71281433105469,
      "learning_rate": 0.00038413935372460655,
      "loss": 141.931298828125,
      "original_ce_loss": 4.4348,
      "step": 270
    },
    {
      "epoch": 1.1576354679802956,
      "grad_norm": 93.65548706054688,
      "learning_rate": 0.00036925963143970405,
      "loss": 139.0336669921875,
      "original_ce_loss": 4.3442,
      "step": 280
    },
    {
      "epoch": 1.199118485869847,
      "grad_norm": 50.466190338134766,
      "learning_rate": 0.0003524636493314982,
      "loss": 136.1361083984375,
      "original_ce_loss": 4.2537,
      "step": 290
    },
    {
      "epoch": 1.2406015037593985,
      "grad_norm": 40.19679641723633,
      "learning_rate": 0.00033395672470911547,
      "loss": 132.7816650390625,
      "original_ce_loss": 4.1489,
      "step": 300
    },
    {
      "epoch": 1.28208452164895,
      "grad_norm": 39.51124954223633,
      "learning_rate": 0.0003139650897713756,
      "loss": 130.66419677734376,
      "original_ce_loss": 4.0827,
      "step": 310
    },
    {
      "epoch": 1.3235675395385014,
      "grad_norm": 36.439571380615234,
      "learning_rate": 0.00029273312610096993,
      "loss": 127.60654296875,
      "original_ce_loss": 3.9871,
      "step": 320
    },
    {
      "epoch": 1.365050557428053,
      "grad_norm": 29.422285079956055,
      "learning_rate": 0.0002705203772970877,
      "loss": 124.6323486328125,
      "original_ce_loss": 3.8942,
      "step": 330
    },
    {
      "epoch": 1.4065335753176043,
      "grad_norm": 40.57709503173828,
      "learning_rate": 0.00024759837626464246,
      "loss": 122.33271484375,
      "original_ce_loss": 3.8223,
      "step": 340
    },
    {
      "epoch": 1.4480165932071558,
      "grad_norm": 50.97930908203125,
      "learning_rate": 0.00022424732594392442,
      "loss": 119.6706298828125,
      "original_ce_loss": 3.7392,
      "step": 350
    },
    {
      "epoch": 1.4894996110967074,
      "grad_norm": 39.640533447265625,
      "learning_rate": 0.00020075267405607558,
      "loss": 118.19874267578125,
      "original_ce_loss": 3.6932,
      "step": 360
    },
    {
      "epoch": 1.5309826289862587,
      "grad_norm": 42.26484298706055,
      "learning_rate": 0.00017740162373535744,
      "loss": 116.33406982421874,
      "original_ce_loss": 3.6349,
      "step": 370
    },
    {
      "epoch": 1.5724656468758103,
      "grad_norm": 24.80182456970215,
      "learning_rate": 0.00015447962270291233,
      "loss": 114.62059326171875,
      "original_ce_loss": 3.5813,
      "step": 380
    },
    {
      "epoch": 1.6139486647653616,
      "grad_norm": 26.771968841552734,
      "learning_rate": 0.0001322668738990301,
      "loss": 113.4220947265625,
      "original_ce_loss": 3.5439,
      "step": 390
    },
    {
      "epoch": 1.6554316826549131,
      "grad_norm": 31.066268920898438,
      "learning_rate": 0.00011103491022862441,
      "loss": 112.212158203125,
      "original_ce_loss": 3.5061,
      "step": 400
    },
    {
      "epoch": 1.6969147005444647,
      "grad_norm": 22.266321182250977,
      "learning_rate": 9.104327529088453e-05,
      "loss": 111.1876953125,
      "original_ce_loss": 3.4741,
      "step": 410
    },
    {
      "epoch": 1.738397718434016,
      "grad_norm": 21.909481048583984,
      "learning_rate": 7.253635066850176e-05,
      "loss": 110.20638427734374,
      "original_ce_loss": 3.4434,
      "step": 420
    },
    {
      "epoch": 1.7798807363235676,
      "grad_norm": Infinity,
      "learning_rate": 5.57403685602959e-05,
      "loss": 109.29989013671874,
      "original_ce_loss": 3.4151,
      "step": 430
    },
    {
      "epoch": 1.821363754213119,
      "grad_norm": 16.10231590270996,
      "learning_rate": 4.225700869359102e-05,
      "loss": 108.821923828125,
      "original_ce_loss": 3.4001,
      "step": 440
    },
    {
      "epoch": 1.8628467721026705,
      "grad_norm": 13.391860961914062,
      "learning_rate": 2.925815333949408e-05,
      "loss": 108.2069580078125,
      "original_ce_loss": 3.3809,
      "step": 450
    },
    {
      "epoch": 1.904329789992222,
      "grad_norm": 12.280927658081055,
      "learning_rate": 1.849928149057992e-05,
      "loss": 107.8180419921875,
      "original_ce_loss": 3.3688,
      "step": 460
    },
    {
      "epoch": 1.9458128078817734,
      "grad_norm": 11.25769329071045,
      "learning_rate": 1.0111911661397095e-05,
      "loss": 107.88870849609376,
      "original_ce_loss": 3.371,
      "step": 470
    },
    {
      "epoch": 1.987295825771325,
      "grad_norm": 11.092090606689453,
      "learning_rate": 4.1985726706168925e-06,
      "loss": 107.57252197265625,
      "original_ce_loss": 3.3611,
      "step": 480
    },
    {
      "epoch": 2.0248898107337308,
      "grad_norm": 146.8853759765625,
      "learning_rate": 0.00019601723888601857,
      "loss": 110.22643025716145,
      "original_ce_loss": 3.444,
      "step": 490
    },
    {
      "epoch": 2.0663728286232823,
      "grad_norm": 51.29483413696289,
      "learning_rate": 0.00018340059260297552,
      "loss": 109.75345458984376,
      "original_ce_loss": 3.4293,
      "step": 500
    },
    {
      "epoch": 2.107855846512834,
      "grad_norm": 71664.6640625,
      "learning_rate": 0.00017088771906293497,
      "loss": 106.969921875,
      "original_ce_loss": 3.3423,
      "step": 510
    },
    {
      "epoch": 2.1493388644023854,
      "grad_norm": 40.492515563964844,
      "learning_rate": 0.00015852324100313608,
      "loss": 106.52847900390626,
      "original_ce_loss": 3.3285,
      "step": 520
    },
    {
      "epoch": 2.190821882291937,
      "grad_norm": 43.579524993896484,
      "learning_rate": 0.0001463512519608304,
      "loss": 104.58763427734375,
      "original_ce_loss": 3.2678,
      "step": 530
    },
    {
      "epoch": 2.232304900181488,
      "grad_norm": 33.966487884521484,
      "learning_rate": 0.00013441515902927877,
      "loss": 104.3513916015625,
      "original_ce_loss": 3.2604,
      "step": 540
    },
    {
      "epoch": 2.2737879180710396,
      "grad_norm": 102.76095581054688,
      "learning_rate": 0.00012275752806170748,
      "loss": 103.5652587890625,
      "original_ce_loss": 3.2359,
      "step": 550
    },
    {
      "epoch": 2.315270935960591,
      "grad_norm": 33.42119216918945,
      "learning_rate": 0.00011141993187525056,
      "loss": 103.26700439453126,
      "original_ce_loss": 3.2266,
      "step": 560
    },
    {
      "epoch": 2.3567539538501427,
      "grad_norm": 28.796308517456055,
      "learning_rate": 0.00010044280199620206,
      "loss": 102.09954223632812,
      "original_ce_loss": 3.1901,
      "step": 570
    },
    {
      "epoch": 2.398236971739694,
      "grad_norm": 21.81608772277832,
      "learning_rate": 8.986528447528069e-05,
      "loss": 101.95626220703124,
      "original_ce_loss": 3.1856,
      "step": 580
    },
    {
      "epoch": 2.4397199896292454,
      "grad_norm": 33.46965789794922,
      "learning_rate": 7.972510028708652e-05,
      "loss": 100.96932983398438,
      "original_ce_loss": 3.1548,
      "step": 590
    },
    {
      "epoch": 2.481203007518797,
      "grad_norm": 24.240360260009766,
      "learning_rate": 7.005841081158877e-05,
      "loss": 100.52698364257813,
      "original_ce_loss": 3.1409,
      "step": 600
    },
    {
      "epoch": 2.5226860254083485,
      "grad_norm": 95.81848907470703,
      "learning_rate": 0.00020207311920542368,
      "loss": 101.27193603515624,
      "original_ce_loss": 3.1642,
      "step": 610
    },
    {
      "epoch": 2.5641690432979,
      "grad_norm": 50.80052947998047,
      "learning_rate": 0.00019340217558391255,
      "loss": 100.44633178710937,
      "original_ce_loss": 3.1384,
      "step": 620
    },
    {
      "epoch": 2.6056520611874516,
      "grad_norm": 67.34901428222656,
      "learning_rate": 0.00018476318415323908,
      "loss": 99.65368041992187,
      "original_ce_loss": 3.1137,
      "step": 630
    },
    {
      "epoch": 2.6471350790770027,
      "grad_norm": 44.9225959777832,
      "learning_rate": 0.00017617059863843596,
      "loss": 98.57659912109375,
      "original_ce_loss": 3.08,
      "step": 640
    },
    {
      "epoch": 2.6886180969665543,
      "grad_norm": 40.193748474121094,
      "learning_rate": 0.0001676387951237199,
      "loss": 98.15628662109376,
      "original_ce_loss": 3.0669,
      "step": 650
    },
    {
      "epoch": 2.730101114856106,
      "grad_norm": 37.45283126831055,
      "learning_rate": 0.00015918204800014976,
      "loss": 97.65645141601563,
      "original_ce_loss": 3.0512,
      "step": 660
    },
    {
      "epoch": 2.7715841327456574,
      "grad_norm": 32.841346740722656,
      "learning_rate": 0.00015081450608342678,
      "loss": 97.17821655273437,
      "original_ce_loss": 3.0363,
      "step": 670
    },
    {
      "epoch": 2.8130671506352085,
      "grad_norm": 38.73606872558594,
      "learning_rate": 0.0001425501689417917,
      "loss": 96.53353881835938,
      "original_ce_loss": 3.0162,
      "step": 680
    },
    {
      "epoch": 2.85455016852476,
      "grad_norm": 39.93632507324219,
      "learning_rate": 0.00013440286347362552,
      "loss": 96.160009765625,
      "original_ce_loss": 3.0045,
      "step": 690
    },
    {
      "epoch": 2.8960331864143116,
      "grad_norm": 32.806663513183594,
      "learning_rate": 0.00012638622077393966,
      "loss": 95.5531005859375,
      "original_ce_loss": 2.9855,
      "step": 700
    },
    {
      "epoch": 2.937516204303863,
      "grad_norm": 36.611412048339844,
      "learning_rate": 0.00011851365332846226,
      "loss": 95.086572265625,
      "original_ce_loss": 2.9709,
      "step": 710
    },
    {
      "epoch": 2.9789992221934147,
      "grad_norm": 33.177825927734375,
      "learning_rate": 0.00011079833257347411,
      "loss": 94.63262939453125,
      "original_ce_loss": 2.9568,
      "step": 720
    },
    {
      "epoch": 3.016593207155821,
      "grad_norm": 36.173824310302734,
      "learning_rate": 0.0001032531668589404,
      "loss": 85.64888916015624,
      "original_ce_loss": 2.9529,
      "step": 730
    },
    {
      "epoch": 3.058076225045372,
      "grad_norm": 28.60649871826172,
      "learning_rate": 9.589077985180664e-05,
      "loss": 93.87921752929688,
      "original_ce_loss": 2.9332,
      "step": 740
    },
    {
      "epoch": 3.0995592429349235,
      "grad_norm": 28.834732055664062,
      "learning_rate": 8.872348941559206e-05,
      "loss": 93.64618530273438,
      "original_ce_loss": 2.9259,
      "step": 750
    },
    {
      "epoch": 3.141042260824475,
      "grad_norm": 26.154630661010742,
      "learning_rate": 8.176328700161682e-05,
      "loss": 93.3215087890625,
      "original_ce_loss": 2.9158,
      "step": 760
    },
    {
      "epoch": 3.1825252787140266,
      "grad_norm": 21.94171714782715,
      "learning_rate": 7.502181758634282e-05,
      "loss": 92.95552368164063,
      "original_ce_loss": 2.9043,
      "step": 770
    },
    {
      "epoch": 3.2240082966035777,
      "grad_norm": 21.436670303344727,
      "learning_rate": 6.85103601883949e-05,
      "loss": 92.64935302734375,
      "original_ce_loss": 2.8948,
      "step": 780
    },
    {
      "epoch": 3.2654913144931292,
      "grad_norm": 30.240144729614258,
      "learning_rate": 6.223980899785866e-05,
      "loss": 92.23059692382813,
      "original_ce_loss": 2.8817,
      "step": 790
    },
    {
      "epoch": 3.306974332382681,
      "grad_norm": 23.84362030029297,
      "learning_rate": 5.6220655149427304e-05,
      "loss": 92.1412109375,
      "original_ce_loss": 2.8789,
      "step": 800
    },
    {
      "epoch": 3.3484573502722323,
      "grad_norm": 21.698253631591797,
      "learning_rate": 5.046296916989253e-05,
      "loss": 91.54824829101562,
      "original_ce_loss": 2.8604,
      "step": 810
    },
    {
      "epoch": 3.389940368161784,
      "grad_norm": 23.564599990844727,
      "learning_rate": 4.4976384129346174e-05,
      "loss": 91.6310791015625,
      "original_ce_loss": 2.863,
      "step": 820
    },
    {
      "epoch": 3.4314233860513355,
      "grad_norm": 24.447463989257812,
      "learning_rate": 3.977007952428178e-05,
      "loss": 91.59302368164063,
      "original_ce_loss": 2.8618,
      "step": 830
    },
    {
      "epoch": 3.4729064039408866,
      "grad_norm": 21.460962295532227,
      "learning_rate": 3.4852765919561326e-05,
      "loss": 91.24329223632813,
      "original_ce_loss": 2.8508,
      "step": 840
    },
    {
      "epoch": 3.514389421830438,
      "grad_norm": 30.3817081451416,
      "learning_rate": 3.0232670374942198e-05,
      "loss": 91.34154052734375,
      "original_ce_loss": 2.8539,
      "step": 850
    },
    {
      "epoch": 3.5558724397199897,
      "grad_norm": 21.52365493774414,
      "learning_rate": 2.5917522680546684e-05,
      "loss": 91.03560180664063,
      "original_ce_loss": 2.8444,
      "step": 860
    },
    {
      "epoch": 3.5973554576095412,
      "grad_norm": 17.532127380371094,
      "learning_rate": 2.1914542424303743e-05,
      "loss": 90.83184204101562,
      "original_ce_loss": 2.838,
      "step": 870
    },
    {
      "epoch": 3.6388384754990923,
      "grad_norm": 16.354034423828125,
      "learning_rate": 1.8230426912999726e-05,
      "loss": 90.87573852539063,
      "original_ce_loss": 2.8394,
      "step": 880
    },
    {
      "epoch": 3.680321493388644,
      "grad_norm": 16.013717651367188,
      "learning_rate": 1.4871339967147433e-05,
      "loss": 90.69482421875,
      "original_ce_loss": 2.8337,
      "step": 890
    },
    {
      "epoch": 3.7218045112781954,
      "grad_norm": 14.059792518615723,
      "learning_rate": 1.1842901608420243e-05,
      "loss": 90.68497924804687,
      "original_ce_loss": 2.8334,
      "step": 900
    },
    {
      "epoch": 3.763287529167747,
      "grad_norm": 16.36054229736328,
      "learning_rate": 9.150178656905624e-06,
      "loss": 90.73485107421875,
      "original_ce_loss": 2.835,
      "step": 910
    },
    {
      "epoch": 3.8047705470572986,
      "grad_norm": 13.24664306640625,
      "learning_rate": 6.797676253908929e-06,
      "loss": 90.66685180664062,
      "original_ce_loss": 2.8328,
      "step": 920
    },
    {
      "epoch": 3.84625356494685,
      "grad_norm": 14.047493934631348,
      "learning_rate": 4.789330324490995e-06,
      "loss": 90.6063720703125,
      "original_ce_loss": 2.8309,
      "step": 930
    },
    {
      "epoch": 3.8877365828364012,
      "grad_norm": 13.237540245056152,
      "learning_rate": 3.12850099234999e-06,
      "loss": 90.53909301757812,
      "original_ce_loss": 2.8288,
      "step": 940
    },
    {
      "epoch": 3.9292196007259528,
      "grad_norm": 13.06207275390625,
      "learning_rate": 1.8179669580652937e-06,
      "loss": 90.23057861328125,
      "original_ce_loss": 2.8192,
      "step": 950
    },
    {
      "epoch": 3.9707026186155043,
      "grad_norm": 13.120635032653809,
      "learning_rate": 8.599208501086114e-07,
      "loss": 90.44099731445313,
      "original_ce_loss": 2.8258,
      "step": 960
    }
  ],
  "logging_steps": 10,
  "max_steps": 968,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 4,
  "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": 7.84394310560809e+16,
  "train_batch_size": 64,
  "trial_name": null,
  "trial_params": null
}