{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 20.0,
  "eval_steps": 500,
  "global_step": 500,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.2,
      "grad_norm": 1.559026837348938,
      "learning_rate": 4.998766400914329e-05,
      "loss": 4.2591,
      "step": 5
    },
    {
      "epoch": 0.4,
      "grad_norm": 2.0623748302459717,
      "learning_rate": 4.995066821070679e-05,
      "loss": 4.084,
      "step": 10
    },
    {
      "epoch": 0.6,
      "grad_norm": 1.7944402694702148,
      "learning_rate": 4.9889049115077005e-05,
      "loss": 4.0724,
      "step": 15
    },
    {
      "epoch": 0.8,
      "grad_norm": 2.1178815364837646,
      "learning_rate": 4.980286753286195e-05,
      "loss": 3.9116,
      "step": 20
    },
    {
      "epoch": 1.0,
      "grad_norm": 2.5003485679626465,
      "learning_rate": 4.9692208514878444e-05,
      "loss": 3.7709,
      "step": 25
    },
    {
      "epoch": 1.2,
      "grad_norm": 1.663772463798523,
      "learning_rate": 4.9557181268217227e-05,
      "loss": 3.5777,
      "step": 30
    },
    {
      "epoch": 1.4,
      "grad_norm": 1.7687804698944092,
      "learning_rate": 4.939791904846869e-05,
      "loss": 3.2731,
      "step": 35
    },
    {
      "epoch": 1.6,
      "grad_norm": 1.3904050588607788,
      "learning_rate": 4.9214579028215776e-05,
      "loss": 3.3322,
      "step": 40
    },
    {
      "epoch": 1.8,
      "grad_norm": 1.6052061319351196,
      "learning_rate": 4.900734214192358e-05,
      "loss": 3.1392,
      "step": 45
    },
    {
      "epoch": 2.0,
      "grad_norm": 1.6881974935531616,
      "learning_rate": 4.877641290737884e-05,
      "loss": 3.0443,
      "step": 50
    },
    {
      "epoch": 2.2,
      "grad_norm": 1.8101818561553955,
      "learning_rate": 4.852201922385564e-05,
      "loss": 3.1381,
      "step": 55
    },
    {
      "epoch": 2.4,
      "grad_norm": 1.9677544832229614,
      "learning_rate": 4.8244412147206284e-05,
      "loss": 2.907,
      "step": 60
    },
    {
      "epoch": 2.6,
      "grad_norm": 1.909866213798523,
      "learning_rate": 4.794386564209953e-05,
      "loss": 2.9507,
      "step": 65
    },
    {
      "epoch": 2.8,
      "grad_norm": 1.7826056480407715,
      "learning_rate": 4.762067631165049e-05,
      "loss": 2.9334,
      "step": 70
    },
    {
      "epoch": 3.0,
      "grad_norm": 2.4618849754333496,
      "learning_rate": 4.72751631047092e-05,
      "loss": 2.7628,
      "step": 75
    },
    {
      "epoch": 3.2,
      "grad_norm": 1.7693885564804077,
      "learning_rate": 4.690766700109659e-05,
      "loss": 2.7438,
      "step": 80
    },
    {
      "epoch": 3.4,
      "grad_norm": 2.086184024810791,
      "learning_rate": 4.65185506750986e-05,
      "loss": 2.7949,
      "step": 85
    },
    {
      "epoch": 3.6,
      "grad_norm": 2.5929203033447266,
      "learning_rate": 4.610819813755038e-05,
      "loss": 2.7056,
      "step": 90
    },
    {
      "epoch": 3.8,
      "grad_norm": 2.221397638320923,
      "learning_rate": 4.567701435686404e-05,
      "loss": 2.5453,
      "step": 95
    },
    {
      "epoch": 4.0,
      "grad_norm": 2.4999215602874756,
      "learning_rate": 4.522542485937369e-05,
      "loss": 2.7453,
      "step": 100
    },
    {
      "epoch": 4.2,
      "grad_norm": 2.4940710067749023,
      "learning_rate": 4.4753875309392266e-05,
      "loss": 2.5124,
      "step": 105
    },
    {
      "epoch": 4.4,
      "grad_norm": 3.088253974914551,
      "learning_rate": 4.426283106939474e-05,
      "loss": 2.5935,
      "step": 110
    },
    {
      "epoch": 4.6,
      "grad_norm": 2.6688263416290283,
      "learning_rate": 4.375277674076149e-05,
      "loss": 2.5463,
      "step": 115
    },
    {
      "epoch": 4.8,
      "grad_norm": 2.6959311962127686,
      "learning_rate": 4.3224215685535294e-05,
      "loss": 2.6062,
      "step": 120
    },
    {
      "epoch": 5.0,
      "grad_norm": 3.0782675743103027,
      "learning_rate": 4.267766952966369e-05,
      "loss": 2.4869,
      "step": 125
    },
    {
      "epoch": 5.2,
      "grad_norm": 3.190086603164673,
      "learning_rate": 4.211367764821722e-05,
      "loss": 2.457,
      "step": 130
    },
    {
      "epoch": 5.4,
      "grad_norm": 3.3880529403686523,
      "learning_rate": 4.1532796633091296e-05,
      "loss": 2.4223,
      "step": 135
    },
    {
      "epoch": 5.6,
      "grad_norm": 2.6318459510803223,
      "learning_rate": 4.093559974371725e-05,
      "loss": 2.385,
      "step": 140
    },
    {
      "epoch": 5.8,
      "grad_norm": 2.92958402633667,
      "learning_rate": 4.0322676341324415e-05,
      "loss": 2.2885,
      "step": 145
    },
    {
      "epoch": 6.0,
      "grad_norm": 3.336378574371338,
      "learning_rate": 3.969463130731183e-05,
      "loss": 2.4326,
      "step": 150
    },
    {
      "epoch": 6.2,
      "grad_norm": 3.508898973464966,
      "learning_rate": 3.905208444630327e-05,
      "loss": 2.1526,
      "step": 155
    },
    {
      "epoch": 6.4,
      "grad_norm": 4.225854396820068,
      "learning_rate": 3.8395669874474915e-05,
      "loss": 2.4194,
      "step": 160
    },
    {
      "epoch": 6.6,
      "grad_norm": 4.112431526184082,
      "learning_rate": 3.7726035393759285e-05,
      "loss": 2.2839,
      "step": 165
    },
    {
      "epoch": 6.8,
      "grad_norm": 3.599271774291992,
      "learning_rate": 3.704384185254288e-05,
      "loss": 2.378,
      "step": 170
    },
    {
      "epoch": 7.0,
      "grad_norm": 4.479362964630127,
      "learning_rate": 3.634976249348867e-05,
      "loss": 2.0319,
      "step": 175
    },
    {
      "epoch": 7.2,
      "grad_norm": 4.234251499176025,
      "learning_rate": 3.564448228912682e-05,
      "loss": 2.021,
      "step": 180
    },
    {
      "epoch": 7.4,
      "grad_norm": 4.8591461181640625,
      "learning_rate": 3.4928697265869515e-05,
      "loss": 1.9893,
      "step": 185
    },
    {
      "epoch": 7.6,
      "grad_norm": 6.527431488037109,
      "learning_rate": 3.4203113817116957e-05,
      "loss": 2.1713,
      "step": 190
    },
    {
      "epoch": 7.8,
      "grad_norm": 5.432151794433594,
      "learning_rate": 3.346844800613229e-05,
      "loss": 2.2043,
      "step": 195
    },
    {
      "epoch": 8.0,
      "grad_norm": 4.323482513427734,
      "learning_rate": 3.272542485937369e-05,
      "loss": 2.2359,
      "step": 200
    },
    {
      "epoch": 8.2,
      "grad_norm": 3.9754505157470703,
      "learning_rate": 3.1974777650980735e-05,
      "loss": 2.0855,
      "step": 205
    },
    {
      "epoch": 8.4,
      "grad_norm": 4.956503391265869,
      "learning_rate": 3.121724717912138e-05,
      "loss": 1.9709,
      "step": 210
    },
    {
      "epoch": 8.6,
      "grad_norm": 5.2445969581604,
      "learning_rate": 3.045358103491357e-05,
      "loss": 2.062,
      "step": 215
    },
    {
      "epoch": 8.8,
      "grad_norm": 4.901844024658203,
      "learning_rate": 2.9684532864643122e-05,
      "loss": 2.0368,
      "step": 220
    },
    {
      "epoch": 9.0,
      "grad_norm": 4.711243629455566,
      "learning_rate": 2.8910861626005776e-05,
      "loss": 1.8624,
      "step": 225
    },
    {
      "epoch": 9.2,
      "grad_norm": 4.800491809844971,
      "learning_rate": 2.8133330839107608e-05,
      "loss": 2.0423,
      "step": 230
    },
    {
      "epoch": 9.4,
      "grad_norm": 5.560423851013184,
      "learning_rate": 2.7352707832962865e-05,
      "loss": 1.905,
      "step": 235
    },
    {
      "epoch": 9.6,
      "grad_norm": 7.123173713684082,
      "learning_rate": 2.656976298823284e-05,
      "loss": 1.6245,
      "step": 240
    },
    {
      "epoch": 9.8,
      "grad_norm": 6.563107013702393,
      "learning_rate": 2.578526897695321e-05,
      "loss": 1.8757,
      "step": 245
    },
    {
      "epoch": 10.0,
      "grad_norm": 5.408273220062256,
      "learning_rate": 2.5e-05,
      "loss": 1.9541,
      "step": 250
    },
    {
      "epoch": 10.2,
      "grad_norm": 5.59449577331543,
      "learning_rate": 2.4214731023046793e-05,
      "loss": 1.822,
      "step": 255
    },
    {
      "epoch": 10.4,
      "grad_norm": 7.534347057342529,
      "learning_rate": 2.3430237011767167e-05,
      "loss": 1.8813,
      "step": 260
    },
    {
      "epoch": 10.6,
      "grad_norm": 7.906167984008789,
      "learning_rate": 2.2647292167037144e-05,
      "loss": 1.8534,
      "step": 265
    },
    {
      "epoch": 10.8,
      "grad_norm": 5.6607184410095215,
      "learning_rate": 2.186666916089239e-05,
      "loss": 1.7331,
      "step": 270
    },
    {
      "epoch": 11.0,
      "grad_norm": 6.6628217697143555,
      "learning_rate": 2.1089138373994223e-05,
      "loss": 1.6813,
      "step": 275
    },
    {
      "epoch": 11.2,
      "grad_norm": 6.011382579803467,
      "learning_rate": 2.031546713535688e-05,
      "loss": 1.5243,
      "step": 280
    },
    {
      "epoch": 11.4,
      "grad_norm": 6.615753173828125,
      "learning_rate": 1.9546418965086442e-05,
      "loss": 1.7315,
      "step": 285
    },
    {
      "epoch": 11.6,
      "grad_norm": 6.512957572937012,
      "learning_rate": 1.8782752820878634e-05,
      "loss": 1.797,
      "step": 290
    },
    {
      "epoch": 11.8,
      "grad_norm": 6.8899664878845215,
      "learning_rate": 1.802522234901927e-05,
      "loss": 1.7544,
      "step": 295
    },
    {
      "epoch": 12.0,
      "grad_norm": 6.717526435852051,
      "learning_rate": 1.7274575140626318e-05,
      "loss": 1.7081,
      "step": 300
    },
    {
      "epoch": 12.2,
      "grad_norm": 5.812675952911377,
      "learning_rate": 1.6531551993867717e-05,
      "loss": 1.6276,
      "step": 305
    },
    {
      "epoch": 12.4,
      "grad_norm": 6.911491870880127,
      "learning_rate": 1.5796886182883053e-05,
      "loss": 1.6319,
      "step": 310
    },
    {
      "epoch": 12.6,
      "grad_norm": 8.3098726272583,
      "learning_rate": 1.5071302734130489e-05,
      "loss": 1.5262,
      "step": 315
    },
    {
      "epoch": 12.8,
      "grad_norm": 6.8229756355285645,
      "learning_rate": 1.4355517710873184e-05,
      "loss": 1.6186,
      "step": 320
    },
    {
      "epoch": 13.0,
      "grad_norm": 6.8094258308410645,
      "learning_rate": 1.3650237506511331e-05,
      "loss": 1.6026,
      "step": 325
    },
    {
      "epoch": 13.2,
      "grad_norm": 6.559398651123047,
      "learning_rate": 1.2956158147457115e-05,
      "loss": 1.4957,
      "step": 330
    },
    {
      "epoch": 13.4,
      "grad_norm": 7.712869167327881,
      "learning_rate": 1.2273964606240718e-05,
      "loss": 1.5775,
      "step": 335
    },
    {
      "epoch": 13.6,
      "grad_norm": 8.607151985168457,
      "learning_rate": 1.1604330125525079e-05,
      "loss": 1.6038,
      "step": 340
    },
    {
      "epoch": 13.8,
      "grad_norm": 7.38192081451416,
      "learning_rate": 1.0947915553696742e-05,
      "loss": 1.4489,
      "step": 345
    },
    {
      "epoch": 14.0,
      "grad_norm": 9.552181243896484,
      "learning_rate": 1.0305368692688174e-05,
      "loss": 1.6264,
      "step": 350
    },
    {
      "epoch": 14.2,
      "grad_norm": 6.163412094116211,
      "learning_rate": 9.677323658675594e-06,
      "loss": 1.6046,
      "step": 355
    },
    {
      "epoch": 14.4,
      "grad_norm": 8.483848571777344,
      "learning_rate": 9.064400256282757e-06,
      "loss": 1.4736,
      "step": 360
    },
    {
      "epoch": 14.6,
      "grad_norm": 7.673094749450684,
      "learning_rate": 8.467203366908707e-06,
      "loss": 1.493,
      "step": 365
    },
    {
      "epoch": 14.8,
      "grad_norm": 7.977968692779541,
      "learning_rate": 7.886322351782783e-06,
      "loss": 1.5078,
      "step": 370
    },
    {
      "epoch": 15.0,
      "grad_norm": 7.63812255859375,
      "learning_rate": 7.3223304703363135e-06,
      "loss": 1.4198,
      "step": 375
    },
    {
      "epoch": 15.2,
      "grad_norm": 6.485208034515381,
      "learning_rate": 6.775784314464717e-06,
      "loss": 1.5416,
      "step": 380
    },
    {
      "epoch": 15.4,
      "grad_norm": 8.849021911621094,
      "learning_rate": 6.247223259238513e-06,
      "loss": 1.3864,
      "step": 385
    },
    {
      "epoch": 15.6,
      "grad_norm": 7.422196388244629,
      "learning_rate": 5.737168930605272e-06,
      "loss": 1.4536,
      "step": 390
    },
    {
      "epoch": 15.8,
      "grad_norm": 7.654792785644531,
      "learning_rate": 5.24612469060774e-06,
      "loss": 1.4819,
      "step": 395
    },
    {
      "epoch": 16.0,
      "grad_norm": 7.572329044342041,
      "learning_rate": 4.7745751406263165e-06,
      "loss": 1.4569,
      "step": 400
    },
    {
      "epoch": 16.2,
      "grad_norm": 7.488677024841309,
      "learning_rate": 4.322985643135952e-06,
      "loss": 1.4596,
      "step": 405
    },
    {
      "epoch": 16.4,
      "grad_norm": 7.173449993133545,
      "learning_rate": 3.891801862449629e-06,
      "loss": 1.2999,
      "step": 410
    },
    {
      "epoch": 16.6,
      "grad_norm": 7.496935844421387,
      "learning_rate": 3.4814493249014116e-06,
      "loss": 1.4009,
      "step": 415
    },
    {
      "epoch": 16.8,
      "grad_norm": 7.768974304199219,
      "learning_rate": 3.0923329989034132e-06,
      "loss": 1.6062,
      "step": 420
    },
    {
      "epoch": 17.0,
      "grad_norm": 7.606301784515381,
      "learning_rate": 2.7248368952908053e-06,
      "loss": 1.3563,
      "step": 425
    },
    {
      "epoch": 17.2,
      "grad_norm": 9.161933898925781,
      "learning_rate": 2.379323688349516e-06,
      "loss": 1.2701,
      "step": 430
    },
    {
      "epoch": 17.4,
      "grad_norm": 9.012870788574219,
      "learning_rate": 2.0561343579004715e-06,
      "loss": 1.539,
      "step": 435
    },
    {
      "epoch": 17.6,
      "grad_norm": 7.092794418334961,
      "learning_rate": 1.7555878527937164e-06,
      "loss": 1.4352,
      "step": 440
    },
    {
      "epoch": 17.8,
      "grad_norm": 6.655755996704102,
      "learning_rate": 1.4779807761443636e-06,
      "loss": 1.2718,
      "step": 445
    },
    {
      "epoch": 18.0,
      "grad_norm": 6.042492866516113,
      "learning_rate": 1.2235870926211619e-06,
      "loss": 1.5636,
      "step": 450
    },
    {
      "epoch": 18.2,
      "grad_norm": 7.48867654800415,
      "learning_rate": 9.926578580764234e-07,
      "loss": 1.6322,
      "step": 455
    },
    {
      "epoch": 18.4,
      "grad_norm": 7.198723793029785,
      "learning_rate": 7.854209717842231e-07,
      "loss": 1.3335,
      "step": 460
    },
    {
      "epoch": 18.6,
      "grad_norm": 8.145171165466309,
      "learning_rate": 6.020809515313142e-07,
      "loss": 1.2486,
      "step": 465
    },
    {
      "epoch": 18.8,
      "grad_norm": 8.835026741027832,
      "learning_rate": 4.4281873178278475e-07,
      "loss": 1.41,
      "step": 470
    },
    {
      "epoch": 19.0,
      "grad_norm": 8.490466117858887,
      "learning_rate": 3.077914851215585e-07,
      "loss": 1.3855,
      "step": 475
    },
    {
      "epoch": 19.2,
      "grad_norm": 6.034891605377197,
      "learning_rate": 1.9713246713805588e-07,
      "loss": 1.4357,
      "step": 480
    },
    {
      "epoch": 19.4,
      "grad_norm": 7.8184919357299805,
      "learning_rate": 1.109508849230001e-07,
      "loss": 1.224,
      "step": 485
    },
    {
      "epoch": 19.6,
      "grad_norm": 8.390954971313477,
      "learning_rate": 4.9331789293211026e-08,
      "loss": 1.4121,
      "step": 490
    },
    {
      "epoch": 19.8,
      "grad_norm": 8.34197998046875,
      "learning_rate": 1.233599085671e-08,
      "loss": 1.4843,
      "step": 495
    },
    {
      "epoch": 20.0,
      "grad_norm": 7.5700273513793945,
      "learning_rate": 0.0,
      "loss": 1.52,
      "step": 500
    },
    {
      "epoch": 20.0,
      "step": 500,
      "total_flos": 1.3524244538720256e+16,
      "train_loss": 2.0708214292526246,
      "train_runtime": 754.3437,
      "train_samples_per_second": 10.605,
      "train_steps_per_second": 0.663
    }
  ],
  "logging_steps": 5,
  "max_steps": 500,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 20,
  "save_steps": 100,
  "total_flos": 1.3524244538720256e+16,
  "train_batch_size": 2,
  "trial_name": null,
  "trial_params": null
}