{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 12.0,
  "eval_steps": 500,
  "global_step": 5892,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.05091649694501019,
      "grad_norm": 2.806674003601074,
      "learning_rate": 0.0002,
      "loss": 1.8959,
      "step": 25
    },
    {
      "epoch": 0.10183299389002037,
      "grad_norm": 3.7694294452667236,
      "learning_rate": 0.0002,
      "loss": 1.5457,
      "step": 50
    },
    {
      "epoch": 0.15274949083503056,
      "grad_norm": 2.5580432415008545,
      "learning_rate": 0.0002,
      "loss": 1.2675,
      "step": 75
    },
    {
      "epoch": 0.20366598778004075,
      "grad_norm": 3.3819141387939453,
      "learning_rate": 0.0002,
      "loss": 1.1332,
      "step": 100
    },
    {
      "epoch": 0.2545824847250509,
      "grad_norm": 2.542970895767212,
      "learning_rate": 0.0002,
      "loss": 1.1962,
      "step": 125
    },
    {
      "epoch": 0.3054989816700611,
      "grad_norm": 3.139821767807007,
      "learning_rate": 0.0002,
      "loss": 1.0642,
      "step": 150
    },
    {
      "epoch": 0.3564154786150713,
      "grad_norm": 2.8020615577697754,
      "learning_rate": 0.0002,
      "loss": 1.1643,
      "step": 175
    },
    {
      "epoch": 0.4073319755600815,
      "grad_norm": 2.742553234100342,
      "learning_rate": 0.0002,
      "loss": 1.0901,
      "step": 200
    },
    {
      "epoch": 0.45824847250509165,
      "grad_norm": 2.8044979572296143,
      "learning_rate": 0.0002,
      "loss": 1.1531,
      "step": 225
    },
    {
      "epoch": 0.5091649694501018,
      "grad_norm": 2.295903444290161,
      "learning_rate": 0.0002,
      "loss": 0.981,
      "step": 250
    },
    {
      "epoch": 0.560081466395112,
      "grad_norm": 2.5899577140808105,
      "learning_rate": 0.0002,
      "loss": 1.1367,
      "step": 275
    },
    {
      "epoch": 0.6109979633401222,
      "grad_norm": 3.2374327182769775,
      "learning_rate": 0.0002,
      "loss": 1.0535,
      "step": 300
    },
    {
      "epoch": 0.6619144602851323,
      "grad_norm": 2.467207193374634,
      "learning_rate": 0.0002,
      "loss": 1.1175,
      "step": 325
    },
    {
      "epoch": 0.7128309572301426,
      "grad_norm": 2.5034849643707275,
      "learning_rate": 0.0002,
      "loss": 1.078,
      "step": 350
    },
    {
      "epoch": 0.7637474541751528,
      "grad_norm": 2.7688446044921875,
      "learning_rate": 0.0002,
      "loss": 1.0662,
      "step": 375
    },
    {
      "epoch": 0.814663951120163,
      "grad_norm": 2.9710495471954346,
      "learning_rate": 0.0002,
      "loss": 1.0392,
      "step": 400
    },
    {
      "epoch": 0.8655804480651731,
      "grad_norm": 2.105386734008789,
      "learning_rate": 0.0002,
      "loss": 1.0551,
      "step": 425
    },
    {
      "epoch": 0.9164969450101833,
      "grad_norm": 2.467886209487915,
      "learning_rate": 0.0002,
      "loss": 1.0188,
      "step": 450
    },
    {
      "epoch": 0.9674134419551935,
      "grad_norm": 2.3255183696746826,
      "learning_rate": 0.0002,
      "loss": 1.0985,
      "step": 475
    },
    {
      "epoch": 1.0183299389002036,
      "grad_norm": 1.9878566265106201,
      "learning_rate": 0.0002,
      "loss": 0.9582,
      "step": 500
    },
    {
      "epoch": 1.0692464358452138,
      "grad_norm": 2.2030067443847656,
      "learning_rate": 0.0002,
      "loss": 0.8562,
      "step": 525
    },
    {
      "epoch": 1.120162932790224,
      "grad_norm": 2.0550765991210938,
      "learning_rate": 0.0002,
      "loss": 0.8564,
      "step": 550
    },
    {
      "epoch": 1.1710794297352343,
      "grad_norm": 2.8419840335845947,
      "learning_rate": 0.0002,
      "loss": 0.8446,
      "step": 575
    },
    {
      "epoch": 1.2219959266802445,
      "grad_norm": 1.9293944835662842,
      "learning_rate": 0.0002,
      "loss": 0.8239,
      "step": 600
    },
    {
      "epoch": 1.2729124236252547,
      "grad_norm": 2.2307465076446533,
      "learning_rate": 0.0002,
      "loss": 0.7875,
      "step": 625
    },
    {
      "epoch": 1.3238289205702647,
      "grad_norm": 2.4380366802215576,
      "learning_rate": 0.0002,
      "loss": 0.8686,
      "step": 650
    },
    {
      "epoch": 1.374745417515275,
      "grad_norm": 2.5356967449188232,
      "learning_rate": 0.0002,
      "loss": 0.891,
      "step": 675
    },
    {
      "epoch": 1.4256619144602851,
      "grad_norm": 1.957305669784546,
      "learning_rate": 0.0002,
      "loss": 0.8923,
      "step": 700
    },
    {
      "epoch": 1.4765784114052953,
      "grad_norm": 3.196012020111084,
      "learning_rate": 0.0002,
      "loss": 0.8906,
      "step": 725
    },
    {
      "epoch": 1.5274949083503055,
      "grad_norm": 2.050201654434204,
      "learning_rate": 0.0002,
      "loss": 0.9081,
      "step": 750
    },
    {
      "epoch": 1.5784114052953155,
      "grad_norm": 2.0173072814941406,
      "learning_rate": 0.0002,
      "loss": 0.9078,
      "step": 775
    },
    {
      "epoch": 1.629327902240326,
      "grad_norm": 2.157409906387329,
      "learning_rate": 0.0002,
      "loss": 0.8954,
      "step": 800
    },
    {
      "epoch": 1.680244399185336,
      "grad_norm": 3.089580535888672,
      "learning_rate": 0.0002,
      "loss": 0.8553,
      "step": 825
    },
    {
      "epoch": 1.7311608961303462,
      "grad_norm": 2.2709248065948486,
      "learning_rate": 0.0002,
      "loss": 0.9085,
      "step": 850
    },
    {
      "epoch": 1.7820773930753564,
      "grad_norm": 2.866403102874756,
      "learning_rate": 0.0002,
      "loss": 0.8565,
      "step": 875
    },
    {
      "epoch": 1.8329938900203666,
      "grad_norm": 2.305607795715332,
      "learning_rate": 0.0002,
      "loss": 0.8674,
      "step": 900
    },
    {
      "epoch": 1.8839103869653768,
      "grad_norm": 2.287306070327759,
      "learning_rate": 0.0002,
      "loss": 0.9006,
      "step": 925
    },
    {
      "epoch": 1.9348268839103868,
      "grad_norm": 2.0112550258636475,
      "learning_rate": 0.0002,
      "loss": 0.8583,
      "step": 950
    },
    {
      "epoch": 1.9857433808553973,
      "grad_norm": 2.2255215644836426,
      "learning_rate": 0.0002,
      "loss": 0.8373,
      "step": 975
    },
    {
      "epoch": 2.0366598778004072,
      "grad_norm": 2.1194534301757812,
      "learning_rate": 0.0002,
      "loss": 0.7602,
      "step": 1000
    },
    {
      "epoch": 2.0875763747454177,
      "grad_norm": 1.874557614326477,
      "learning_rate": 0.0002,
      "loss": 0.6402,
      "step": 1025
    },
    {
      "epoch": 2.1384928716904277,
      "grad_norm": 2.008828639984131,
      "learning_rate": 0.0002,
      "loss": 0.6897,
      "step": 1050
    },
    {
      "epoch": 2.189409368635438,
      "grad_norm": 1.8525766134262085,
      "learning_rate": 0.0002,
      "loss": 0.6611,
      "step": 1075
    },
    {
      "epoch": 2.240325865580448,
      "grad_norm": 2.184070348739624,
      "learning_rate": 0.0002,
      "loss": 0.7043,
      "step": 1100
    },
    {
      "epoch": 2.291242362525458,
      "grad_norm": 1.9643105268478394,
      "learning_rate": 0.0002,
      "loss": 0.6444,
      "step": 1125
    },
    {
      "epoch": 2.3421588594704685,
      "grad_norm": 2.2633492946624756,
      "learning_rate": 0.0002,
      "loss": 0.6915,
      "step": 1150
    },
    {
      "epoch": 2.3930753564154785,
      "grad_norm": 2.3236186504364014,
      "learning_rate": 0.0002,
      "loss": 0.6539,
      "step": 1175
    },
    {
      "epoch": 2.443991853360489,
      "grad_norm": 2.5817580223083496,
      "learning_rate": 0.0002,
      "loss": 0.6922,
      "step": 1200
    },
    {
      "epoch": 2.494908350305499,
      "grad_norm": 2.4054062366485596,
      "learning_rate": 0.0002,
      "loss": 0.6485,
      "step": 1225
    },
    {
      "epoch": 2.5458248472505094,
      "grad_norm": 2.730226516723633,
      "learning_rate": 0.0002,
      "loss": 0.7163,
      "step": 1250
    },
    {
      "epoch": 2.5967413441955194,
      "grad_norm": 2.436521530151367,
      "learning_rate": 0.0002,
      "loss": 0.6899,
      "step": 1275
    },
    {
      "epoch": 2.6476578411405294,
      "grad_norm": 2.4914846420288086,
      "learning_rate": 0.0002,
      "loss": 0.7301,
      "step": 1300
    },
    {
      "epoch": 2.69857433808554,
      "grad_norm": 2.550816774368286,
      "learning_rate": 0.0002,
      "loss": 0.6853,
      "step": 1325
    },
    {
      "epoch": 2.74949083503055,
      "grad_norm": 2.2780368328094482,
      "learning_rate": 0.0002,
      "loss": 0.7165,
      "step": 1350
    },
    {
      "epoch": 2.8004073319755602,
      "grad_norm": 2.6065120697021484,
      "learning_rate": 0.0002,
      "loss": 0.6841,
      "step": 1375
    },
    {
      "epoch": 2.8513238289205702,
      "grad_norm": 2.3689310550689697,
      "learning_rate": 0.0002,
      "loss": 0.7647,
      "step": 1400
    },
    {
      "epoch": 2.9022403258655807,
      "grad_norm": 3.0321147441864014,
      "learning_rate": 0.0002,
      "loss": 0.6813,
      "step": 1425
    },
    {
      "epoch": 2.9531568228105907,
      "grad_norm": 2.069146156311035,
      "learning_rate": 0.0002,
      "loss": 0.7123,
      "step": 1450
    },
    {
      "epoch": 3.0040733197556007,
      "grad_norm": 2.1185224056243896,
      "learning_rate": 0.0002,
      "loss": 0.7189,
      "step": 1475
    },
    {
      "epoch": 3.054989816700611,
      "grad_norm": 2.558671236038208,
      "learning_rate": 0.0002,
      "loss": 0.5302,
      "step": 1500
    },
    {
      "epoch": 3.105906313645621,
      "grad_norm": 2.6581151485443115,
      "learning_rate": 0.0002,
      "loss": 0.4756,
      "step": 1525
    },
    {
      "epoch": 3.1568228105906315,
      "grad_norm": 2.602369785308838,
      "learning_rate": 0.0002,
      "loss": 0.527,
      "step": 1550
    },
    {
      "epoch": 3.2077393075356415,
      "grad_norm": 2.8149938583374023,
      "learning_rate": 0.0002,
      "loss": 0.5173,
      "step": 1575
    },
    {
      "epoch": 3.258655804480652,
      "grad_norm": 2.065443515777588,
      "learning_rate": 0.0002,
      "loss": 0.5756,
      "step": 1600
    },
    {
      "epoch": 3.309572301425662,
      "grad_norm": 2.308039903640747,
      "learning_rate": 0.0002,
      "loss": 0.5096,
      "step": 1625
    },
    {
      "epoch": 3.360488798370672,
      "grad_norm": 2.4268381595611572,
      "learning_rate": 0.0002,
      "loss": 0.5538,
      "step": 1650
    },
    {
      "epoch": 3.4114052953156824,
      "grad_norm": 2.6498641967773438,
      "learning_rate": 0.0002,
      "loss": 0.5189,
      "step": 1675
    },
    {
      "epoch": 3.4623217922606924,
      "grad_norm": 2.2553043365478516,
      "learning_rate": 0.0002,
      "loss": 0.5786,
      "step": 1700
    },
    {
      "epoch": 3.513238289205703,
      "grad_norm": 2.8816471099853516,
      "learning_rate": 0.0002,
      "loss": 0.5232,
      "step": 1725
    },
    {
      "epoch": 3.564154786150713,
      "grad_norm": 1.7681572437286377,
      "learning_rate": 0.0002,
      "loss": 0.569,
      "step": 1750
    },
    {
      "epoch": 3.6150712830957232,
      "grad_norm": 2.772834062576294,
      "learning_rate": 0.0002,
      "loss": 0.5357,
      "step": 1775
    },
    {
      "epoch": 3.6659877800407332,
      "grad_norm": 2.4505577087402344,
      "learning_rate": 0.0002,
      "loss": 0.5848,
      "step": 1800
    },
    {
      "epoch": 3.716904276985743,
      "grad_norm": 3.0437214374542236,
      "learning_rate": 0.0002,
      "loss": 0.5571,
      "step": 1825
    },
    {
      "epoch": 3.7678207739307537,
      "grad_norm": 3.2281458377838135,
      "learning_rate": 0.0002,
      "loss": 0.5956,
      "step": 1850
    },
    {
      "epoch": 3.8187372708757636,
      "grad_norm": 2.437544345855713,
      "learning_rate": 0.0002,
      "loss": 0.5542,
      "step": 1875
    },
    {
      "epoch": 3.869653767820774,
      "grad_norm": 2.759650230407715,
      "learning_rate": 0.0002,
      "loss": 0.5823,
      "step": 1900
    },
    {
      "epoch": 3.920570264765784,
      "grad_norm": 2.3260252475738525,
      "learning_rate": 0.0002,
      "loss": 0.5724,
      "step": 1925
    },
    {
      "epoch": 3.9714867617107945,
      "grad_norm": 2.2468202114105225,
      "learning_rate": 0.0002,
      "loss": 0.5884,
      "step": 1950
    },
    {
      "epoch": 4.022403258655804,
      "grad_norm": 2.221639394760132,
      "learning_rate": 0.0002,
      "loss": 0.5266,
      "step": 1975
    },
    {
      "epoch": 4.0733197556008145,
      "grad_norm": 1.9102641344070435,
      "learning_rate": 0.0002,
      "loss": 0.3866,
      "step": 2000
    },
    {
      "epoch": 4.124236252545825,
      "grad_norm": 2.9819774627685547,
      "learning_rate": 0.0002,
      "loss": 0.4212,
      "step": 2025
    },
    {
      "epoch": 4.175152749490835,
      "grad_norm": 2.39497709274292,
      "learning_rate": 0.0002,
      "loss": 0.4073,
      "step": 2050
    },
    {
      "epoch": 4.226069246435845,
      "grad_norm": 2.4051284790039062,
      "learning_rate": 0.0002,
      "loss": 0.4328,
      "step": 2075
    },
    {
      "epoch": 4.276985743380855,
      "grad_norm": 2.677963972091675,
      "learning_rate": 0.0002,
      "loss": 0.4294,
      "step": 2100
    },
    {
      "epoch": 4.327902240325866,
      "grad_norm": 2.484499216079712,
      "learning_rate": 0.0002,
      "loss": 0.4386,
      "step": 2125
    },
    {
      "epoch": 4.378818737270876,
      "grad_norm": 3.8193187713623047,
      "learning_rate": 0.0002,
      "loss": 0.4317,
      "step": 2150
    },
    {
      "epoch": 4.429735234215886,
      "grad_norm": 2.5229299068450928,
      "learning_rate": 0.0002,
      "loss": 0.4597,
      "step": 2175
    },
    {
      "epoch": 4.480651731160896,
      "grad_norm": 2.6942062377929688,
      "learning_rate": 0.0002,
      "loss": 0.4462,
      "step": 2200
    },
    {
      "epoch": 4.531568228105907,
      "grad_norm": 2.4558463096618652,
      "learning_rate": 0.0002,
      "loss": 0.4645,
      "step": 2225
    },
    {
      "epoch": 4.582484725050916,
      "grad_norm": 2.276397466659546,
      "learning_rate": 0.0002,
      "loss": 0.4721,
      "step": 2250
    },
    {
      "epoch": 4.633401221995927,
      "grad_norm": 2.844794750213623,
      "learning_rate": 0.0002,
      "loss": 0.477,
      "step": 2275
    },
    {
      "epoch": 4.684317718940937,
      "grad_norm": 2.6256089210510254,
      "learning_rate": 0.0002,
      "loss": 0.4553,
      "step": 2300
    },
    {
      "epoch": 4.7352342158859475,
      "grad_norm": 2.599666118621826,
      "learning_rate": 0.0002,
      "loss": 0.4753,
      "step": 2325
    },
    {
      "epoch": 4.786150712830957,
      "grad_norm": 2.470028877258301,
      "learning_rate": 0.0002,
      "loss": 0.4433,
      "step": 2350
    },
    {
      "epoch": 4.8370672097759675,
      "grad_norm": 2.4316930770874023,
      "learning_rate": 0.0002,
      "loss": 0.493,
      "step": 2375
    },
    {
      "epoch": 4.887983706720978,
      "grad_norm": 2.4588210582733154,
      "learning_rate": 0.0002,
      "loss": 0.465,
      "step": 2400
    },
    {
      "epoch": 4.9389002036659875,
      "grad_norm": 2.5438883304595947,
      "learning_rate": 0.0002,
      "loss": 0.4802,
      "step": 2425
    },
    {
      "epoch": 4.989816700610998,
      "grad_norm": 2.764341354370117,
      "learning_rate": 0.0002,
      "loss": 0.4788,
      "step": 2450
    },
    {
      "epoch": 5.040733197556008,
      "grad_norm": 2.4877161979675293,
      "learning_rate": 0.0002,
      "loss": 0.3577,
      "step": 2475
    },
    {
      "epoch": 5.091649694501018,
      "grad_norm": 2.5782573223114014,
      "learning_rate": 0.0002,
      "loss": 0.3434,
      "step": 2500
    },
    {
      "epoch": 5.142566191446028,
      "grad_norm": 2.363449811935425,
      "learning_rate": 0.0002,
      "loss": 0.3595,
      "step": 2525
    },
    {
      "epoch": 5.193482688391039,
      "grad_norm": 2.2839596271514893,
      "learning_rate": 0.0002,
      "loss": 0.3566,
      "step": 2550
    },
    {
      "epoch": 5.244399185336049,
      "grad_norm": 2.3375349044799805,
      "learning_rate": 0.0002,
      "loss": 0.3704,
      "step": 2575
    },
    {
      "epoch": 5.295315682281059,
      "grad_norm": 2.2514779567718506,
      "learning_rate": 0.0002,
      "loss": 0.3586,
      "step": 2600
    },
    {
      "epoch": 5.346232179226069,
      "grad_norm": 2.687700033187866,
      "learning_rate": 0.0002,
      "loss": 0.3797,
      "step": 2625
    },
    {
      "epoch": 5.39714867617108,
      "grad_norm": 2.498506784439087,
      "learning_rate": 0.0002,
      "loss": 0.3761,
      "step": 2650
    },
    {
      "epoch": 5.44806517311609,
      "grad_norm": 2.632052183151245,
      "learning_rate": 0.0002,
      "loss": 0.4052,
      "step": 2675
    },
    {
      "epoch": 5.4989816700611,
      "grad_norm": 2.7227554321289062,
      "learning_rate": 0.0002,
      "loss": 0.3939,
      "step": 2700
    },
    {
      "epoch": 5.54989816700611,
      "grad_norm": 2.041354179382324,
      "learning_rate": 0.0002,
      "loss": 0.3995,
      "step": 2725
    },
    {
      "epoch": 5.6008146639511205,
      "grad_norm": 4.209399223327637,
      "learning_rate": 0.0002,
      "loss": 0.3882,
      "step": 2750
    },
    {
      "epoch": 5.65173116089613,
      "grad_norm": 1.9786779880523682,
      "learning_rate": 0.0002,
      "loss": 0.401,
      "step": 2775
    },
    {
      "epoch": 5.7026476578411405,
      "grad_norm": 3.278057098388672,
      "learning_rate": 0.0002,
      "loss": 0.391,
      "step": 2800
    },
    {
      "epoch": 5.753564154786151,
      "grad_norm": 2.4366769790649414,
      "learning_rate": 0.0002,
      "loss": 0.4022,
      "step": 2825
    },
    {
      "epoch": 5.804480651731161,
      "grad_norm": 2.606545925140381,
      "learning_rate": 0.0002,
      "loss": 0.3864,
      "step": 2850
    },
    {
      "epoch": 5.855397148676171,
      "grad_norm": 2.517683982849121,
      "learning_rate": 0.0002,
      "loss": 0.4029,
      "step": 2875
    },
    {
      "epoch": 5.906313645621181,
      "grad_norm": 2.00010085105896,
      "learning_rate": 0.0002,
      "loss": 0.3909,
      "step": 2900
    },
    {
      "epoch": 5.957230142566192,
      "grad_norm": 2.561739683151245,
      "learning_rate": 0.0002,
      "loss": 0.3977,
      "step": 2925
    },
    {
      "epoch": 6.008146639511201,
      "grad_norm": 2.0152885913848877,
      "learning_rate": 0.0002,
      "loss": 0.3885,
      "step": 2950
    },
    {
      "epoch": 6.059063136456212,
      "grad_norm": 2.1359171867370605,
      "learning_rate": 0.0002,
      "loss": 0.3042,
      "step": 2975
    },
    {
      "epoch": 6.109979633401222,
      "grad_norm": 1.833449363708496,
      "learning_rate": 0.0002,
      "loss": 0.2968,
      "step": 3000
    },
    {
      "epoch": 6.160896130346233,
      "grad_norm": 1.7801095247268677,
      "learning_rate": 0.0002,
      "loss": 0.3064,
      "step": 3025
    },
    {
      "epoch": 6.211812627291242,
      "grad_norm": 2.25486159324646,
      "learning_rate": 0.0002,
      "loss": 0.3364,
      "step": 3050
    },
    {
      "epoch": 6.262729124236253,
      "grad_norm": 2.518338680267334,
      "learning_rate": 0.0002,
      "loss": 0.3269,
      "step": 3075
    },
    {
      "epoch": 6.313645621181263,
      "grad_norm": 1.4665368795394897,
      "learning_rate": 0.0002,
      "loss": 0.3359,
      "step": 3100
    },
    {
      "epoch": 6.364562118126273,
      "grad_norm": 2.3365631103515625,
      "learning_rate": 0.0002,
      "loss": 0.3233,
      "step": 3125
    },
    {
      "epoch": 6.415478615071283,
      "grad_norm": 2.406109094619751,
      "learning_rate": 0.0002,
      "loss": 0.3409,
      "step": 3150
    },
    {
      "epoch": 6.4663951120162935,
      "grad_norm": 3.327075958251953,
      "learning_rate": 0.0002,
      "loss": 0.3484,
      "step": 3175
    },
    {
      "epoch": 6.517311608961304,
      "grad_norm": 2.125439167022705,
      "learning_rate": 0.0002,
      "loss": 0.3423,
      "step": 3200
    },
    {
      "epoch": 6.5682281059063135,
      "grad_norm": 2.0253591537475586,
      "learning_rate": 0.0002,
      "loss": 0.3325,
      "step": 3225
    },
    {
      "epoch": 6.619144602851324,
      "grad_norm": 2.2506959438323975,
      "learning_rate": 0.0002,
      "loss": 0.3592,
      "step": 3250
    },
    {
      "epoch": 6.670061099796334,
      "grad_norm": 3.3501861095428467,
      "learning_rate": 0.0002,
      "loss": 0.3404,
      "step": 3275
    },
    {
      "epoch": 6.720977596741344,
      "grad_norm": 2.8377532958984375,
      "learning_rate": 0.0002,
      "loss": 0.3683,
      "step": 3300
    },
    {
      "epoch": 6.771894093686354,
      "grad_norm": 2.3918981552124023,
      "learning_rate": 0.0002,
      "loss": 0.3506,
      "step": 3325
    },
    {
      "epoch": 6.822810590631365,
      "grad_norm": 2.026571273803711,
      "learning_rate": 0.0002,
      "loss": 0.3714,
      "step": 3350
    },
    {
      "epoch": 6.873727087576375,
      "grad_norm": 1.9378567934036255,
      "learning_rate": 0.0002,
      "loss": 0.3529,
      "step": 3375
    },
    {
      "epoch": 6.924643584521385,
      "grad_norm": 2.5375149250030518,
      "learning_rate": 0.0002,
      "loss": 0.3614,
      "step": 3400
    },
    {
      "epoch": 6.975560081466395,
      "grad_norm": 2.6719155311584473,
      "learning_rate": 0.0002,
      "loss": 0.3543,
      "step": 3425
    },
    {
      "epoch": 7.026476578411406,
      "grad_norm": 1.9606503248214722,
      "learning_rate": 0.0002,
      "loss": 0.3204,
      "step": 3450
    },
    {
      "epoch": 7.077393075356415,
      "grad_norm": 2.2155184745788574,
      "learning_rate": 0.0002,
      "loss": 0.2657,
      "step": 3475
    },
    {
      "epoch": 7.128309572301426,
      "grad_norm": 2.116616725921631,
      "learning_rate": 0.0002,
      "loss": 0.29,
      "step": 3500
    },
    {
      "epoch": 7.179226069246436,
      "grad_norm": 1.7466342449188232,
      "learning_rate": 0.0002,
      "loss": 0.2942,
      "step": 3525
    },
    {
      "epoch": 7.2301425661914465,
      "grad_norm": 2.126159906387329,
      "learning_rate": 0.0002,
      "loss": 0.2951,
      "step": 3550
    },
    {
      "epoch": 7.281059063136456,
      "grad_norm": 1.983241081237793,
      "learning_rate": 0.0002,
      "loss": 0.3043,
      "step": 3575
    },
    {
      "epoch": 7.3319755600814664,
      "grad_norm": 2.3558452129364014,
      "learning_rate": 0.0002,
      "loss": 0.3011,
      "step": 3600
    },
    {
      "epoch": 7.382892057026477,
      "grad_norm": 2.587038516998291,
      "learning_rate": 0.0002,
      "loss": 0.3116,
      "step": 3625
    },
    {
      "epoch": 7.433808553971486,
      "grad_norm": 2.0329151153564453,
      "learning_rate": 0.0002,
      "loss": 0.3139,
      "step": 3650
    },
    {
      "epoch": 7.484725050916497,
      "grad_norm": 1.955492377281189,
      "learning_rate": 0.0002,
      "loss": 0.2996,
      "step": 3675
    },
    {
      "epoch": 7.535641547861507,
      "grad_norm": 1.8513798713684082,
      "learning_rate": 0.0002,
      "loss": 0.3288,
      "step": 3700
    },
    {
      "epoch": 7.586558044806518,
      "grad_norm": 1.7869365215301514,
      "learning_rate": 0.0002,
      "loss": 0.2988,
      "step": 3725
    },
    {
      "epoch": 7.637474541751527,
      "grad_norm": 2.4369406700134277,
      "learning_rate": 0.0002,
      "loss": 0.3168,
      "step": 3750
    },
    {
      "epoch": 7.688391038696538,
      "grad_norm": 2.287158727645874,
      "learning_rate": 0.0002,
      "loss": 0.3253,
      "step": 3775
    },
    {
      "epoch": 7.739307535641548,
      "grad_norm": 2.4751217365264893,
      "learning_rate": 0.0002,
      "loss": 0.3143,
      "step": 3800
    },
    {
      "epoch": 7.790224032586558,
      "grad_norm": 2.351529836654663,
      "learning_rate": 0.0002,
      "loss": 0.3135,
      "step": 3825
    },
    {
      "epoch": 7.841140529531568,
      "grad_norm": 1.7692896127700806,
      "learning_rate": 0.0002,
      "loss": 0.3226,
      "step": 3850
    },
    {
      "epoch": 7.892057026476579,
      "grad_norm": 1.483668327331543,
      "learning_rate": 0.0002,
      "loss": 0.3172,
      "step": 3875
    },
    {
      "epoch": 7.942973523421589,
      "grad_norm": 2.4964892864227295,
      "learning_rate": 0.0002,
      "loss": 0.3395,
      "step": 3900
    },
    {
      "epoch": 7.993890020366599,
      "grad_norm": 1.4458813667297363,
      "learning_rate": 0.0002,
      "loss": 0.3376,
      "step": 3925
    },
    {
      "epoch": 8.044806517311608,
      "grad_norm": 1.859740972518921,
      "learning_rate": 0.0002,
      "loss": 0.2504,
      "step": 3950
    },
    {
      "epoch": 8.095723014256619,
      "grad_norm": 1.3244179487228394,
      "learning_rate": 0.0002,
      "loss": 0.2643,
      "step": 3975
    },
    {
      "epoch": 8.146639511201629,
      "grad_norm": 1.5618747472763062,
      "learning_rate": 0.0002,
      "loss": 0.2579,
      "step": 4000
    },
    {
      "epoch": 8.19755600814664,
      "grad_norm": 1.806612491607666,
      "learning_rate": 0.0002,
      "loss": 0.2722,
      "step": 4025
    },
    {
      "epoch": 8.24847250509165,
      "grad_norm": 3.7666194438934326,
      "learning_rate": 0.0002,
      "loss": 0.2762,
      "step": 4050
    },
    {
      "epoch": 8.29938900203666,
      "grad_norm": 2.5826241970062256,
      "learning_rate": 0.0002,
      "loss": 0.2749,
      "step": 4075
    },
    {
      "epoch": 8.35030549898167,
      "grad_norm": 1.5352802276611328,
      "learning_rate": 0.0002,
      "loss": 0.2827,
      "step": 4100
    },
    {
      "epoch": 8.40122199592668,
      "grad_norm": 1.3692405223846436,
      "learning_rate": 0.0002,
      "loss": 0.2844,
      "step": 4125
    },
    {
      "epoch": 8.45213849287169,
      "grad_norm": 2.2680575847625732,
      "learning_rate": 0.0002,
      "loss": 0.2921,
      "step": 4150
    },
    {
      "epoch": 8.5030549898167,
      "grad_norm": 1.5913770198822021,
      "learning_rate": 0.0002,
      "loss": 0.294,
      "step": 4175
    },
    {
      "epoch": 8.55397148676171,
      "grad_norm": 2.9854307174682617,
      "learning_rate": 0.0002,
      "loss": 0.2971,
      "step": 4200
    },
    {
      "epoch": 8.604887983706721,
      "grad_norm": 2.6485278606414795,
      "learning_rate": 0.0002,
      "loss": 0.3029,
      "step": 4225
    },
    {
      "epoch": 8.655804480651732,
      "grad_norm": 1.9538838863372803,
      "learning_rate": 0.0002,
      "loss": 0.2878,
      "step": 4250
    },
    {
      "epoch": 8.706720977596742,
      "grad_norm": 1.6330325603485107,
      "learning_rate": 0.0002,
      "loss": 0.3144,
      "step": 4275
    },
    {
      "epoch": 8.757637474541752,
      "grad_norm": 1.4803214073181152,
      "learning_rate": 0.0002,
      "loss": 0.2976,
      "step": 4300
    },
    {
      "epoch": 8.808553971486761,
      "grad_norm": 3.5660393238067627,
      "learning_rate": 0.0002,
      "loss": 0.3142,
      "step": 4325
    },
    {
      "epoch": 8.859470468431772,
      "grad_norm": 2.3036141395568848,
      "learning_rate": 0.0002,
      "loss": 0.3012,
      "step": 4350
    },
    {
      "epoch": 8.910386965376782,
      "grad_norm": 2.528514862060547,
      "learning_rate": 0.0002,
      "loss": 0.3145,
      "step": 4375
    },
    {
      "epoch": 8.961303462321792,
      "grad_norm": 1.4351972341537476,
      "learning_rate": 0.0002,
      "loss": 0.3136,
      "step": 4400
    },
    {
      "epoch": 9.012219959266803,
      "grad_norm": 1.339784860610962,
      "learning_rate": 0.0002,
      "loss": 0.2982,
      "step": 4425
    },
    {
      "epoch": 9.063136456211813,
      "grad_norm": 1.6082321405410767,
      "learning_rate": 0.0002,
      "loss": 0.2447,
      "step": 4450
    },
    {
      "epoch": 9.114052953156822,
      "grad_norm": 2.039848566055298,
      "learning_rate": 0.0002,
      "loss": 0.2687,
      "step": 4475
    },
    {
      "epoch": 9.164969450101832,
      "grad_norm": 1.8532267808914185,
      "learning_rate": 0.0002,
      "loss": 0.2656,
      "step": 4500
    },
    {
      "epoch": 9.215885947046843,
      "grad_norm": 2.1240642070770264,
      "learning_rate": 0.0002,
      "loss": 0.2685,
      "step": 4525
    },
    {
      "epoch": 9.266802443991853,
      "grad_norm": 1.2580517530441284,
      "learning_rate": 0.0002,
      "loss": 0.2563,
      "step": 4550
    },
    {
      "epoch": 9.317718940936864,
      "grad_norm": 1.3006818294525146,
      "learning_rate": 0.0002,
      "loss": 0.2726,
      "step": 4575
    },
    {
      "epoch": 9.368635437881874,
      "grad_norm": 1.2992304563522339,
      "learning_rate": 0.0002,
      "loss": 0.2611,
      "step": 4600
    },
    {
      "epoch": 9.419551934826885,
      "grad_norm": 2.6281898021698,
      "learning_rate": 0.0002,
      "loss": 0.2844,
      "step": 4625
    },
    {
      "epoch": 9.470468431771893,
      "grad_norm": 1.8424818515777588,
      "learning_rate": 0.0002,
      "loss": 0.2744,
      "step": 4650
    },
    {
      "epoch": 9.521384928716904,
      "grad_norm": 2.217327356338501,
      "learning_rate": 0.0002,
      "loss": 0.2943,
      "step": 4675
    },
    {
      "epoch": 9.572301425661914,
      "grad_norm": 1.463914394378662,
      "learning_rate": 0.0002,
      "loss": 0.2651,
      "step": 4700
    },
    {
      "epoch": 9.623217922606925,
      "grad_norm": 1.7850229740142822,
      "learning_rate": 0.0002,
      "loss": 0.2853,
      "step": 4725
    },
    {
      "epoch": 9.674134419551935,
      "grad_norm": 1.7324199676513672,
      "learning_rate": 0.0002,
      "loss": 0.2755,
      "step": 4750
    },
    {
      "epoch": 9.725050916496945,
      "grad_norm": 1.1688644886016846,
      "learning_rate": 0.0002,
      "loss": 0.2895,
      "step": 4775
    },
    {
      "epoch": 9.775967413441956,
      "grad_norm": 1.8048006296157837,
      "learning_rate": 0.0002,
      "loss": 0.2728,
      "step": 4800
    },
    {
      "epoch": 9.826883910386965,
      "grad_norm": 1.9764938354492188,
      "learning_rate": 0.0002,
      "loss": 0.2993,
      "step": 4825
    },
    {
      "epoch": 9.877800407331975,
      "grad_norm": 2.7224347591400146,
      "learning_rate": 0.0002,
      "loss": 0.2882,
      "step": 4850
    },
    {
      "epoch": 9.928716904276985,
      "grad_norm": 2.1471643447875977,
      "learning_rate": 0.0002,
      "loss": 0.292,
      "step": 4875
    },
    {
      "epoch": 9.979633401221996,
      "grad_norm": 2.122901678085327,
      "learning_rate": 0.0002,
      "loss": 0.2771,
      "step": 4900
    },
    {
      "epoch": 10.030549898167006,
      "grad_norm": 1.5285823345184326,
      "learning_rate": 0.0002,
      "loss": 0.2568,
      "step": 4925
    },
    {
      "epoch": 10.081466395112017,
      "grad_norm": 1.8790913820266724,
      "learning_rate": 0.0002,
      "loss": 0.2382,
      "step": 4950
    },
    {
      "epoch": 10.132382892057027,
      "grad_norm": 1.7328804731369019,
      "learning_rate": 0.0002,
      "loss": 0.2489,
      "step": 4975
    },
    {
      "epoch": 10.183299389002036,
      "grad_norm": 1.6446770429611206,
      "learning_rate": 0.0002,
      "loss": 0.2591,
      "step": 5000
    },
    {
      "epoch": 10.234215885947046,
      "grad_norm": 2.407752275466919,
      "learning_rate": 0.0002,
      "loss": 0.2636,
      "step": 5025
    },
    {
      "epoch": 10.285132382892057,
      "grad_norm": 1.7200428247451782,
      "learning_rate": 0.0002,
      "loss": 0.2528,
      "step": 5050
    },
    {
      "epoch": 10.336048879837067,
      "grad_norm": 4.040430545806885,
      "learning_rate": 0.0002,
      "loss": 0.2609,
      "step": 5075
    },
    {
      "epoch": 10.386965376782078,
      "grad_norm": 1.499848484992981,
      "learning_rate": 0.0002,
      "loss": 0.2699,
      "step": 5100
    },
    {
      "epoch": 10.437881873727088,
      "grad_norm": 1.3807271718978882,
      "learning_rate": 0.0002,
      "loss": 0.2698,
      "step": 5125
    },
    {
      "epoch": 10.488798370672098,
      "grad_norm": 2.1875016689300537,
      "learning_rate": 0.0002,
      "loss": 0.2718,
      "step": 5150
    },
    {
      "epoch": 10.539714867617107,
      "grad_norm": 1.8573893308639526,
      "learning_rate": 0.0002,
      "loss": 0.2658,
      "step": 5175
    },
    {
      "epoch": 10.590631364562118,
      "grad_norm": 1.7177698612213135,
      "learning_rate": 0.0002,
      "loss": 0.2666,
      "step": 5200
    },
    {
      "epoch": 10.641547861507128,
      "grad_norm": 2.5458731651306152,
      "learning_rate": 0.0002,
      "loss": 0.2699,
      "step": 5225
    },
    {
      "epoch": 10.692464358452138,
      "grad_norm": 1.9686267375946045,
      "learning_rate": 0.0002,
      "loss": 0.2765,
      "step": 5250
    },
    {
      "epoch": 10.743380855397149,
      "grad_norm": 2.519333600997925,
      "learning_rate": 0.0002,
      "loss": 0.2746,
      "step": 5275
    },
    {
      "epoch": 10.79429735234216,
      "grad_norm": 1.456548810005188,
      "learning_rate": 0.0002,
      "loss": 0.2746,
      "step": 5300
    },
    {
      "epoch": 10.84521384928717,
      "grad_norm": 2.32214617729187,
      "learning_rate": 0.0002,
      "loss": 0.2921,
      "step": 5325
    },
    {
      "epoch": 10.89613034623218,
      "grad_norm": 1.5902856588363647,
      "learning_rate": 0.0002,
      "loss": 0.2849,
      "step": 5350
    },
    {
      "epoch": 10.947046843177189,
      "grad_norm": 2.1129226684570312,
      "learning_rate": 0.0002,
      "loss": 0.2725,
      "step": 5375
    },
    {
      "epoch": 10.9979633401222,
      "grad_norm": 2.040208578109741,
      "learning_rate": 0.0002,
      "loss": 0.2917,
      "step": 5400
    },
    {
      "epoch": 11.04887983706721,
      "grad_norm": 2.841794967651367,
      "learning_rate": 0.0002,
      "loss": 0.2304,
      "step": 5425
    },
    {
      "epoch": 11.09979633401222,
      "grad_norm": 1.1914277076721191,
      "learning_rate": 0.0002,
      "loss": 0.2458,
      "step": 5450
    },
    {
      "epoch": 11.15071283095723,
      "grad_norm": 1.49674391746521,
      "learning_rate": 0.0002,
      "loss": 0.2369,
      "step": 5475
    },
    {
      "epoch": 11.201629327902241,
      "grad_norm": 0.9921414256095886,
      "learning_rate": 0.0002,
      "loss": 0.2586,
      "step": 5500
    },
    {
      "epoch": 11.25254582484725,
      "grad_norm": 2.710859537124634,
      "learning_rate": 0.0002,
      "loss": 0.2436,
      "step": 5525
    },
    {
      "epoch": 11.30346232179226,
      "grad_norm": 1.3944896459579468,
      "learning_rate": 0.0002,
      "loss": 0.2654,
      "step": 5550
    },
    {
      "epoch": 11.35437881873727,
      "grad_norm": 1.8961577415466309,
      "learning_rate": 0.0002,
      "loss": 0.2411,
      "step": 5575
    },
    {
      "epoch": 11.405295315682281,
      "grad_norm": 3.050171136856079,
      "learning_rate": 0.0002,
      "loss": 0.2615,
      "step": 5600
    },
    {
      "epoch": 11.456211812627291,
      "grad_norm": 2.27951979637146,
      "learning_rate": 0.0002,
      "loss": 0.2534,
      "step": 5625
    },
    {
      "epoch": 11.507128309572302,
      "grad_norm": 1.4935775995254517,
      "learning_rate": 0.0002,
      "loss": 0.2687,
      "step": 5650
    },
    {
      "epoch": 11.558044806517312,
      "grad_norm": 2.3738980293273926,
      "learning_rate": 0.0002,
      "loss": 0.2485,
      "step": 5675
    },
    {
      "epoch": 11.608961303462323,
      "grad_norm": 2.942415237426758,
      "learning_rate": 0.0002,
      "loss": 0.2735,
      "step": 5700
    },
    {
      "epoch": 11.659877800407331,
      "grad_norm": 2.6978747844696045,
      "learning_rate": 0.0002,
      "loss": 0.2488,
      "step": 5725
    },
    {
      "epoch": 11.710794297352342,
      "grad_norm": 1.4899920225143433,
      "learning_rate": 0.0002,
      "loss": 0.267,
      "step": 5750
    },
    {
      "epoch": 11.761710794297352,
      "grad_norm": 2.248117208480835,
      "learning_rate": 0.0002,
      "loss": 0.253,
      "step": 5775
    },
    {
      "epoch": 11.812627291242363,
      "grad_norm": 1.3980762958526611,
      "learning_rate": 0.0002,
      "loss": 0.2742,
      "step": 5800
    },
    {
      "epoch": 11.863543788187373,
      "grad_norm": 3.564595937728882,
      "learning_rate": 0.0002,
      "loss": 0.2679,
      "step": 5825
    },
    {
      "epoch": 11.914460285132384,
      "grad_norm": 1.835825800895691,
      "learning_rate": 0.0002,
      "loss": 0.2799,
      "step": 5850
    },
    {
      "epoch": 11.965376782077392,
      "grad_norm": 1.2737057209014893,
      "learning_rate": 0.0002,
      "loss": 0.2761,
      "step": 5875
    }
  ],
  "logging_steps": 25,
  "max_steps": 5892,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 12,
  "save_steps": 25,
  "stateful_callbacks": {
    "TrainerControl": {
      "args": {
        "should_epoch_stop": false,
        "should_evaluate": false,
        "should_log": false,
        "should_save": true,
        "should_training_stop": true
      },
      "attributes": {}
    }
  },
  "total_flos": 5.5000869805056e+16,
  "train_batch_size": 4,
  "trial_name": null,
  "trial_params": null
}