diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4281 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.9628224582701064,
+  "eval_steps": 247,
+  "global_step": 2964,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0010116337885685382,
+      "eval_loss": 8.720296859741211,
+      "eval_runtime": 269.4019,
+      "eval_samples_per_second": 6.462,
+      "eval_steps_per_second": 6.462,
+      "step": 1
+    },
+    {
+      "epoch": 0.005058168942842691,
+      "grad_norm": 356.0,
+      "learning_rate": 2.5e-06,
+      "loss": 8.6533,
+      "step": 5
+    },
+    {
+      "epoch": 0.010116337885685382,
+      "grad_norm": 254.0,
+      "learning_rate": 5e-06,
+      "loss": 7.3554,
+      "step": 10
+    },
+    {
+      "epoch": 0.015174506828528073,
+      "grad_norm": 732.0,
+      "learning_rate": 7.5e-06,
+      "loss": 6.8742,
+      "step": 15
+    },
+    {
+      "epoch": 0.020232675771370764,
+      "grad_norm": 229.0,
+      "learning_rate": 1e-05,
+      "loss": 6.7736,
+      "step": 20
+    },
+    {
+      "epoch": 0.025290844714213456,
+      "grad_norm": 114.5,
+      "learning_rate": 1.25e-05,
+      "loss": 6.462,
+      "step": 25
+    },
+    {
+      "epoch": 0.030349013657056147,
+      "grad_norm": 112.5,
+      "learning_rate": 1.5e-05,
+      "loss": 5.7131,
+      "step": 30
+    },
+    {
+      "epoch": 0.035407182599898834,
+      "grad_norm": 141.0,
+      "learning_rate": 1.75e-05,
+      "loss": 5.4139,
+      "step": 35
+    },
+    {
+      "epoch": 0.04046535154274153,
+      "grad_norm": 122.5,
+      "learning_rate": 2e-05,
+      "loss": 5.3972,
+      "step": 40
+    },
+    {
+      "epoch": 0.04552352048558422,
+      "grad_norm": 84.5,
+      "learning_rate": 2.25e-05,
+      "loss": 5.2905,
+      "step": 45
+    },
+    {
+      "epoch": 0.05058168942842691,
+      "grad_norm": 73.5,
+      "learning_rate": 2.5e-05,
+      "loss": 5.3254,
+      "step": 50
+    },
+    {
+      "epoch": 0.0556398583712696,
+      "grad_norm": 88.5,
+      "learning_rate": 2.7500000000000004e-05,
+      "loss": 4.9147,
+      "step": 55
+    },
+    {
+      "epoch": 0.06069802731411229,
+      "grad_norm": 68.0,
+      "learning_rate": 3e-05,
+      "loss": 5.1287,
+      "step": 60
+    },
+    {
+      "epoch": 0.06575619625695499,
+      "grad_norm": 60.75,
+      "learning_rate": 3.2500000000000004e-05,
+      "loss": 4.998,
+      "step": 65
+    },
+    {
+      "epoch": 0.07081436519979767,
+      "grad_norm": 63.75,
+      "learning_rate": 3.5e-05,
+      "loss": 4.5915,
+      "step": 70
+    },
+    {
+      "epoch": 0.07587253414264036,
+      "grad_norm": 58.25,
+      "learning_rate": 3.7500000000000003e-05,
+      "loss": 4.2293,
+      "step": 75
+    },
+    {
+      "epoch": 0.08093070308548306,
+      "grad_norm": 47.0,
+      "learning_rate": 4e-05,
+      "loss": 4.2928,
+      "step": 80
+    },
+    {
+      "epoch": 0.08598887202832575,
+      "grad_norm": 49.25,
+      "learning_rate": 4.25e-05,
+      "loss": 4.3066,
+      "step": 85
+    },
+    {
+      "epoch": 0.09104704097116843,
+      "grad_norm": 45.25,
+      "learning_rate": 4.5e-05,
+      "loss": 4.0678,
+      "step": 90
+    },
+    {
+      "epoch": 0.09610520991401113,
+      "grad_norm": 42.75,
+      "learning_rate": 4.75e-05,
+      "loss": 3.8672,
+      "step": 95
+    },
+    {
+      "epoch": 0.10116337885685382,
+      "grad_norm": 40.75,
+      "learning_rate": 5e-05,
+      "loss": 3.7882,
+      "step": 100
+    },
+    {
+      "epoch": 0.1062215477996965,
+      "grad_norm": 46.5,
+      "learning_rate": 4.999962398718099e-05,
+      "loss": 3.5627,
+      "step": 105
+    },
+    {
+      "epoch": 0.1112797167425392,
+      "grad_norm": 39.75,
+      "learning_rate": 4.999849596003482e-05,
+      "loss": 3.5157,
+      "step": 110
+    },
+    {
+      "epoch": 0.11633788568538189,
+      "grad_norm": 37.5,
+      "learning_rate": 4.999661595249369e-05,
+      "loss": 3.4912,
+      "step": 115
+    },
+    {
+      "epoch": 0.12139605462822459,
+      "grad_norm": 33.75,
+      "learning_rate": 4.9993984021110154e-05,
+      "loss": 3.1001,
+      "step": 120
+    },
+    {
+      "epoch": 0.12645422357106728,
+      "grad_norm": 32.75,
+      "learning_rate": 4.9990600245055416e-05,
+      "loss": 3.0767,
+      "step": 125
+    },
+    {
+      "epoch": 0.13151239251390998,
+      "grad_norm": 29.375,
+      "learning_rate": 4.998646472611693e-05,
+      "loss": 3.0194,
+      "step": 130
+    },
+    {
+      "epoch": 0.13657056145675264,
+      "grad_norm": 33.0,
+      "learning_rate": 4.998157758869534e-05,
+      "loss": 3.0627,
+      "step": 135
+    },
+    {
+      "epoch": 0.14162873039959534,
+      "grad_norm": 32.75,
+      "learning_rate": 4.9975938979800755e-05,
+      "loss": 3.0349,
+      "step": 140
+    },
+    {
+      "epoch": 0.14668689934243803,
+      "grad_norm": 31.5,
+      "learning_rate": 4.9969549069048314e-05,
+      "loss": 2.969,
+      "step": 145
+    },
+    {
+      "epoch": 0.15174506828528073,
+      "grad_norm": 27.125,
+      "learning_rate": 4.996240804865309e-05,
+      "loss": 2.8384,
+      "step": 150
+    },
+    {
+      "epoch": 0.15680323722812342,
+      "grad_norm": 28.0,
+      "learning_rate": 4.995451613342429e-05,
+      "loss": 2.8856,
+      "step": 155
+    },
+    {
+      "epoch": 0.16186140617096612,
+      "grad_norm": 23.625,
+      "learning_rate": 4.994587356075883e-05,
+      "loss": 2.7377,
+      "step": 160
+    },
+    {
+      "epoch": 0.1669195751138088,
+      "grad_norm": 25.875,
+      "learning_rate": 4.993648059063414e-05,
+      "loss": 2.883,
+      "step": 165
+    },
+    {
+      "epoch": 0.1719777440566515,
+      "grad_norm": 29.5,
+      "learning_rate": 4.992633750560041e-05,
+      "loss": 2.768,
+      "step": 170
+    },
+    {
+      "epoch": 0.17703591299949417,
+      "grad_norm": 26.625,
+      "learning_rate": 4.991544461077203e-05,
+      "loss": 2.6471,
+      "step": 175
+    },
+    {
+      "epoch": 0.18209408194233687,
+      "grad_norm": 26.125,
+      "learning_rate": 4.990380223381846e-05,
+      "loss": 2.5026,
+      "step": 180
+    },
+    {
+      "epoch": 0.18715225088517956,
+      "grad_norm": 23.0,
+      "learning_rate": 4.9891410724954326e-05,
+      "loss": 2.6369,
+      "step": 185
+    },
+    {
+      "epoch": 0.19221041982802226,
+      "grad_norm": 25.125,
+      "learning_rate": 4.9878270456928926e-05,
+      "loss": 2.5479,
+      "step": 190
+    },
+    {
+      "epoch": 0.19726858877086495,
+      "grad_norm": 24.125,
+      "learning_rate": 4.9864381825015e-05,
+      "loss": 2.5968,
+      "step": 195
+    },
+    {
+      "epoch": 0.20232675771370764,
+      "grad_norm": 23.0,
+      "learning_rate": 4.9849745246996835e-05,
+      "loss": 2.595,
+      "step": 200
+    },
+    {
+      "epoch": 0.20738492665655034,
+      "grad_norm": 24.625,
+      "learning_rate": 4.9834361163157714e-05,
+      "loss": 2.6123,
+      "step": 205
+    },
+    {
+      "epoch": 0.212443095599393,
+      "grad_norm": 21.75,
+      "learning_rate": 4.9818230036266656e-05,
+      "loss": 2.522,
+      "step": 210
+    },
+    {
+      "epoch": 0.2175012645422357,
+      "grad_norm": 26.625,
+      "learning_rate": 4.980135235156449e-05,
+      "loss": 2.4647,
+      "step": 215
+    },
+    {
+      "epoch": 0.2225594334850784,
+      "grad_norm": 21.125,
+      "learning_rate": 4.9783728616749286e-05,
+      "loss": 2.2818,
+      "step": 220
+    },
+    {
+      "epoch": 0.2276176024279211,
+      "grad_norm": 28.125,
+      "learning_rate": 4.976535936196106e-05,
+      "loss": 2.4421,
+      "step": 225
+    },
+    {
+      "epoch": 0.23267577137076378,
+      "grad_norm": 23.75,
+      "learning_rate": 4.974624513976585e-05,
+      "loss": 2.5393,
+      "step": 230
+    },
+    {
+      "epoch": 0.23773394031360648,
+      "grad_norm": 20.625,
+      "learning_rate": 4.9726386525139035e-05,
+      "loss": 2.51,
+      "step": 235
+    },
+    {
+      "epoch": 0.24279210925644917,
+      "grad_norm": 23.125,
+      "learning_rate": 4.9705784115448115e-05,
+      "loss": 2.4146,
+      "step": 240
+    },
+    {
+      "epoch": 0.24785027819929187,
+      "grad_norm": 25.875,
+      "learning_rate": 4.968443853043472e-05,
+      "loss": 2.5361,
+      "step": 245
+    },
+    {
+      "epoch": 0.24987354577642892,
+      "eval_loss": 2.346703290939331,
+      "eval_runtime": 257.0086,
+      "eval_samples_per_second": 6.774,
+      "eval_steps_per_second": 6.774,
+      "step": 247
+    },
+    {
+      "epoch": 0.25290844714213456,
+      "grad_norm": 23.375,
+      "learning_rate": 4.966235041219592e-05,
+      "loss": 2.4028,
+      "step": 250
+    },
+    {
+      "epoch": 0.25796661608497723,
+      "grad_norm": 24.875,
+      "learning_rate": 4.9639520425164975e-05,
+      "loss": 2.2069,
+      "step": 255
+    },
+    {
+      "epoch": 0.26302478502781995,
+      "grad_norm": 22.125,
+      "learning_rate": 4.961594925609129e-05,
+      "loss": 2.3506,
+      "step": 260
+    },
+    {
+      "epoch": 0.2680829539706626,
+      "grad_norm": 20.375,
+      "learning_rate": 4.9591637614019824e-05,
+      "loss": 2.424,
+      "step": 265
+    },
+    {
+      "epoch": 0.2731411229135053,
+      "grad_norm": 19.5,
+      "learning_rate": 4.9566586230269684e-05,
+      "loss": 2.2442,
+      "step": 270
+    },
+    {
+      "epoch": 0.278199291856348,
+      "grad_norm": 21.375,
+      "learning_rate": 4.95407958584122e-05,
+      "loss": 2.0972,
+      "step": 275
+    },
+    {
+      "epoch": 0.2832574607991907,
+      "grad_norm": 23.25,
+      "learning_rate": 4.9514267274248205e-05,
+      "loss": 2.2188,
+      "step": 280
+    },
+    {
+      "epoch": 0.2883156297420334,
+      "grad_norm": 24.0,
+      "learning_rate": 4.94870012757847e-05,
+      "loss": 2.2483,
+      "step": 285
+    },
+    {
+      "epoch": 0.29337379868487606,
+      "grad_norm": 19.375,
+      "learning_rate": 4.94589986832109e-05,
+      "loss": 2.2795,
+      "step": 290
+    },
+    {
+      "epoch": 0.2984319676277188,
+      "grad_norm": 18.5,
+      "learning_rate": 4.9430260338873485e-05,
+      "loss": 2.3224,
+      "step": 295
+    },
+    {
+      "epoch": 0.30349013657056145,
+      "grad_norm": 19.625,
+      "learning_rate": 4.9400787107251347e-05,
+      "loss": 2.3763,
+      "step": 300
+    },
+    {
+      "epoch": 0.3085483055134042,
+      "grad_norm": 18.375,
+      "learning_rate": 4.9370579874929504e-05,
+      "loss": 2.2439,
+      "step": 305
+    },
+    {
+      "epoch": 0.31360647445624684,
+      "grad_norm": 20.875,
+      "learning_rate": 4.933963955057249e-05,
+      "loss": 2.2622,
+      "step": 310
+    },
+    {
+      "epoch": 0.3186646433990895,
+      "grad_norm": 23.875,
+      "learning_rate": 4.930796706489699e-05,
+      "loss": 2.2202,
+      "step": 315
+    },
+    {
+      "epoch": 0.32372281234193223,
+      "grad_norm": 21.125,
+      "learning_rate": 4.927556337064385e-05,
+      "loss": 2.3505,
+      "step": 320
+    },
+    {
+      "epoch": 0.3287809812847749,
+      "grad_norm": 18.875,
+      "learning_rate": 4.924242944254943e-05,
+      "loss": 2.169,
+      "step": 325
+    },
+    {
+      "epoch": 0.3338391502276176,
+      "grad_norm": 22.125,
+      "learning_rate": 4.920856627731626e-05,
+      "loss": 2.327,
+      "step": 330
+    },
+    {
+      "epoch": 0.3388973191704603,
+      "grad_norm": 24.0,
+      "learning_rate": 4.917397489358308e-05,
+      "loss": 2.1776,
+      "step": 335
+    },
+    {
+      "epoch": 0.343955488113303,
+      "grad_norm": 24.125,
+      "learning_rate": 4.913865633189419e-05,
+      "loss": 2.3587,
+      "step": 340
+    },
+    {
+      "epoch": 0.3490136570561457,
+      "grad_norm": 20.375,
+      "learning_rate": 4.910261165466814e-05,
+      "loss": 2.0693,
+      "step": 345
+    },
+    {
+      "epoch": 0.35407182599898834,
+      "grad_norm": 18.875,
+      "learning_rate": 4.906584194616579e-05,
+      "loss": 2.2131,
+      "step": 350
+    },
+    {
+      "epoch": 0.35912999494183107,
+      "grad_norm": 20.5,
+      "learning_rate": 4.9028348312457685e-05,
+      "loss": 2.2065,
+      "step": 355
+    },
+    {
+      "epoch": 0.36418816388467373,
+      "grad_norm": 20.875,
+      "learning_rate": 4.899013188139077e-05,
+      "loss": 2.2963,
+      "step": 360
+    },
+    {
+      "epoch": 0.36924633282751645,
+      "grad_norm": 25.875,
+      "learning_rate": 4.8951193802554475e-05,
+      "loss": 2.2212,
+      "step": 365
+    },
+    {
+      "epoch": 0.3743045017703591,
+      "grad_norm": 17.25,
+      "learning_rate": 4.891153524724616e-05,
+      "loss": 2.1413,
+      "step": 370
+    },
+    {
+      "epoch": 0.37936267071320184,
+      "grad_norm": 20.0,
+      "learning_rate": 4.887115740843583e-05,
+      "loss": 1.9985,
+      "step": 375
+    },
+    {
+      "epoch": 0.3844208396560445,
+      "grad_norm": 20.75,
+      "learning_rate": 4.883006150073028e-05,
+      "loss": 2.2051,
+      "step": 380
+    },
+    {
+      "epoch": 0.3894790085988872,
+      "grad_norm": 21.125,
+      "learning_rate": 4.878824876033658e-05,
+      "loss": 2.1988,
+      "step": 385
+    },
+    {
+      "epoch": 0.3945371775417299,
+      "grad_norm": 20.5,
+      "learning_rate": 4.874572044502481e-05,
+      "loss": 2.0261,
+      "step": 390
+    },
+    {
+      "epoch": 0.39959534648457257,
+      "grad_norm": 18.875,
+      "learning_rate": 4.870247783409033e-05,
+      "loss": 2.0358,
+      "step": 395
+    },
+    {
+      "epoch": 0.4046535154274153,
+      "grad_norm": 19.125,
+      "learning_rate": 4.8658522228315226e-05,
+      "loss": 2.172,
+      "step": 400
+    },
+    {
+      "epoch": 0.40971168437025796,
+      "grad_norm": 19.875,
+      "learning_rate": 4.8613854949929174e-05,
+      "loss": 2.1072,
+      "step": 405
+    },
+    {
+      "epoch": 0.4147698533131007,
+      "grad_norm": 20.5,
+      "learning_rate": 4.856847734256974e-05,
+      "loss": 2.0351,
+      "step": 410
+    },
+    {
+      "epoch": 0.41982802225594334,
+      "grad_norm": 19.5,
+      "learning_rate": 4.8522390771241877e-05,
+      "loss": 2.1653,
+      "step": 415
+    },
+    {
+      "epoch": 0.424886191198786,
+      "grad_norm": 22.5,
+      "learning_rate": 4.847559662227691e-05,
+      "loss": 2.1664,
+      "step": 420
+    },
+    {
+      "epoch": 0.42994436014162873,
+      "grad_norm": 19.625,
+      "learning_rate": 4.842809630329084e-05,
+      "loss": 1.9003,
+      "step": 425
+    },
+    {
+      "epoch": 0.4350025290844714,
+      "grad_norm": 16.375,
+      "learning_rate": 4.837989124314196e-05,
+      "loss": 2.0239,
+      "step": 430
+    },
+    {
+      "epoch": 0.4400606980273141,
+      "grad_norm": 19.875,
+      "learning_rate": 4.833098289188792e-05,
+      "loss": 2.0788,
+      "step": 435
+    },
+    {
+      "epoch": 0.4451188669701568,
+      "grad_norm": 19.75,
+      "learning_rate": 4.8281372720742095e-05,
+      "loss": 2.013,
+      "step": 440
+    },
+    {
+      "epoch": 0.4501770359129995,
+      "grad_norm": 18.875,
+      "learning_rate": 4.823106222202929e-05,
+      "loss": 2.1045,
+      "step": 445
+    },
+    {
+      "epoch": 0.4552352048558422,
+      "grad_norm": 19.5,
+      "learning_rate": 4.818005290914092e-05,
+      "loss": 2.0627,
+      "step": 450
+    },
+    {
+      "epoch": 0.4602933737986849,
+      "grad_norm": 23.125,
+      "learning_rate": 4.8128346316489406e-05,
+      "loss": 2.0449,
+      "step": 455
+    },
+    {
+      "epoch": 0.46535154274152757,
+      "grad_norm": 16.375,
+      "learning_rate": 4.8075943999462106e-05,
+      "loss": 1.9173,
+      "step": 460
+    },
+    {
+      "epoch": 0.47040971168437024,
+      "grad_norm": 21.625,
+      "learning_rate": 4.8022847534374424e-05,
+      "loss": 1.9477,
+      "step": 465
+    },
+    {
+      "epoch": 0.47546788062721296,
+      "grad_norm": 16.625,
+      "learning_rate": 4.796905851842252e-05,
+      "loss": 2.0886,
+      "step": 470
+    },
+    {
+      "epoch": 0.4805260495700556,
+      "grad_norm": 16.875,
+      "learning_rate": 4.7914578569635115e-05,
+      "loss": 2.0808,
+      "step": 475
+    },
+    {
+      "epoch": 0.48558421851289835,
+      "grad_norm": 16.75,
+      "learning_rate": 4.785940932682497e-05,
+      "loss": 2.1545,
+      "step": 480
+    },
+    {
+      "epoch": 0.490642387455741,
+      "grad_norm": 17.875,
+      "learning_rate": 4.780355244953947e-05,
+      "loss": 2.0563,
+      "step": 485
+    },
+    {
+      "epoch": 0.49570055639858374,
+      "grad_norm": 16.5,
+      "learning_rate": 4.7747009618010776e-05,
+      "loss": 2.0622,
+      "step": 490
+    },
+    {
+      "epoch": 0.49974709155285785,
+      "eval_loss": 1.9648030996322632,
+      "eval_runtime": 256.852,
+      "eval_samples_per_second": 6.778,
+      "eval_steps_per_second": 6.778,
+      "step": 494
+    },
+    {
+      "epoch": 0.5007587253414264,
+      "grad_norm": 19.5,
+      "learning_rate": 4.768978253310524e-05,
+      "loss": 2.1208,
+      "step": 495
+    },
+    {
+      "epoch": 0.5058168942842691,
+      "grad_norm": 18.375,
+      "learning_rate": 4.763187291627227e-05,
+      "loss": 1.8682,
+      "step": 500
+    },
+    {
+      "epoch": 0.5108750632271117,
+      "grad_norm": 24.0,
+      "learning_rate": 4.757328250949252e-05,
+      "loss": 2.0287,
+      "step": 505
+    },
+    {
+      "epoch": 0.5159332321699545,
+      "grad_norm": 22.375,
+      "learning_rate": 4.751401307522551e-05,
+      "loss": 1.9339,
+      "step": 510
+    },
+    {
+      "epoch": 0.5209914011127972,
+      "grad_norm": 15.75,
+      "learning_rate": 4.745406639635661e-05,
+      "loss": 1.9712,
+      "step": 515
+    },
+    {
+      "epoch": 0.5260495700556399,
+      "grad_norm": 17.125,
+      "learning_rate": 4.73934442761434e-05,
+      "loss": 1.9365,
+      "step": 520
+    },
+    {
+      "epoch": 0.5311077389984825,
+      "grad_norm": 20.625,
+      "learning_rate": 4.733214853816142e-05,
+      "loss": 1.8835,
+      "step": 525
+    },
+    {
+      "epoch": 0.5361659079413252,
+      "grad_norm": 21.125,
+      "learning_rate": 4.727018102624933e-05,
+      "loss": 1.9561,
+      "step": 530
+    },
+    {
+      "epoch": 0.541224076884168,
+      "grad_norm": 18.125,
+      "learning_rate": 4.7207543604453434e-05,
+      "loss": 1.7768,
+      "step": 535
+    },
+    {
+      "epoch": 0.5462822458270106,
+      "grad_norm": 18.5,
+      "learning_rate": 4.714423815697162e-05,
+      "loss": 2.0149,
+      "step": 540
+    },
+    {
+      "epoch": 0.5513404147698533,
+      "grad_norm": 16.625,
+      "learning_rate": 4.7080266588096674e-05,
+      "loss": 1.9243,
+      "step": 545
+    },
+    {
+      "epoch": 0.556398583712696,
+      "grad_norm": 19.5,
+      "learning_rate": 4.701563082215898e-05,
+      "loss": 1.9856,
+      "step": 550
+    },
+    {
+      "epoch": 0.5614567526555387,
+      "grad_norm": 16.375,
+      "learning_rate": 4.6950332803468665e-05,
+      "loss": 1.9061,
+      "step": 555
+    },
+    {
+      "epoch": 0.5665149215983813,
+      "grad_norm": 17.375,
+      "learning_rate": 4.688437449625711e-05,
+      "loss": 1.8953,
+      "step": 560
+    },
+    {
+      "epoch": 0.5715730905412241,
+      "grad_norm": 19.875,
+      "learning_rate": 4.681775788461782e-05,
+      "loss": 1.8439,
+      "step": 565
+    },
+    {
+      "epoch": 0.5766312594840668,
+      "grad_norm": 18.5,
+      "learning_rate": 4.67504849724468e-05,
+      "loss": 2.0273,
+      "step": 570
+    },
+    {
+      "epoch": 0.5816894284269094,
+      "grad_norm": 17.125,
+      "learning_rate": 4.668255778338223e-05,
+      "loss": 1.7345,
+      "step": 575
+    },
+    {
+      "epoch": 0.5867475973697521,
+      "grad_norm": 16.75,
+      "learning_rate": 4.6613978360743625e-05,
+      "loss": 1.7621,
+      "step": 580
+    },
+    {
+      "epoch": 0.5918057663125948,
+      "grad_norm": 16.375,
+      "learning_rate": 4.654474876747035e-05,
+      "loss": 1.8954,
+      "step": 585
+    },
+    {
+      "epoch": 0.5968639352554376,
+      "grad_norm": 17.375,
+      "learning_rate": 4.6474871086059556e-05,
+      "loss": 1.8496,
+      "step": 590
+    },
+    {
+      "epoch": 0.6019221041982802,
+      "grad_norm": 14.4375,
+      "learning_rate": 4.640434741850358e-05,
+      "loss": 1.9292,
+      "step": 595
+    },
+    {
+      "epoch": 0.6069802731411229,
+      "grad_norm": 19.125,
+      "learning_rate": 4.6333179886226634e-05,
+      "loss": 1.9507,
+      "step": 600
+    },
+    {
+      "epoch": 0.6120384420839656,
+      "grad_norm": 17.625,
+      "learning_rate": 4.62613706300211e-05,
+      "loss": 1.8745,
+      "step": 605
+    },
+    {
+      "epoch": 0.6170966110268084,
+      "grad_norm": 16.25,
+      "learning_rate": 4.618892180998304e-05,
+      "loss": 1.7686,
+      "step": 610
+    },
+    {
+      "epoch": 0.622154779969651,
+      "grad_norm": 16.625,
+      "learning_rate": 4.611583560544725e-05,
+      "loss": 1.8471,
+      "step": 615
+    },
+    {
+      "epoch": 0.6272129489124937,
+      "grad_norm": 18.25,
+      "learning_rate": 4.6042114214921724e-05,
+      "loss": 1.9049,
+      "step": 620
+    },
+    {
+      "epoch": 0.6322711178553364,
+      "grad_norm": 17.625,
+      "learning_rate": 4.596775985602148e-05,
+      "loss": 1.7335,
+      "step": 625
+    },
+    {
+      "epoch": 0.637329286798179,
+      "grad_norm": 15.1875,
+      "learning_rate": 4.5892774765401906e-05,
+      "loss": 1.8359,
+      "step": 630
+    },
+    {
+      "epoch": 0.6423874557410217,
+      "grad_norm": 24.25,
+      "learning_rate": 4.581716119869141e-05,
+      "loss": 1.8779,
+      "step": 635
+    },
+    {
+      "epoch": 0.6474456246838645,
+      "grad_norm": 15.9375,
+      "learning_rate": 4.5740921430423626e-05,
+      "loss": 1.928,
+      "step": 640
+    },
+    {
+      "epoch": 0.6525037936267072,
+      "grad_norm": 15.1875,
+      "learning_rate": 4.566405775396897e-05,
+      "loss": 1.7928,
+      "step": 645
+    },
+    {
+      "epoch": 0.6575619625695498,
+      "grad_norm": 14.3125,
+      "learning_rate": 4.558657248146565e-05,
+      "loss": 1.704,
+      "step": 650
+    },
+    {
+      "epoch": 0.6626201315123925,
+      "grad_norm": 23.125,
+      "learning_rate": 4.550846794375013e-05,
+      "loss": 1.7983,
+      "step": 655
+    },
+    {
+      "epoch": 0.6676783004552352,
+      "grad_norm": 20.125,
+      "learning_rate": 4.542974649028701e-05,
+      "loss": 1.9233,
+      "step": 660
+    },
+    {
+      "epoch": 0.6727364693980779,
+      "grad_norm": 21.375,
+      "learning_rate": 4.5350410489098324e-05,
+      "loss": 1.7701,
+      "step": 665
+    },
+    {
+      "epoch": 0.6777946383409206,
+      "grad_norm": 16.625,
+      "learning_rate": 4.5270462326692363e-05,
+      "loss": 1.8647,
+      "step": 670
+    },
+    {
+      "epoch": 0.6828528072837633,
+      "grad_norm": 17.75,
+      "learning_rate": 4.5189904407991834e-05,
+      "loss": 1.8829,
+      "step": 675
+    },
+    {
+      "epoch": 0.687910976226606,
+      "grad_norm": 20.875,
+      "learning_rate": 4.510873915626155e-05,
+      "loss": 1.8871,
+      "step": 680
+    },
+    {
+      "epoch": 0.6929691451694486,
+      "grad_norm": 15.0625,
+      "learning_rate": 4.50269690130355e-05,
+      "loss": 1.8591,
+      "step": 685
+    },
+    {
+      "epoch": 0.6980273141122914,
+      "grad_norm": 15.1875,
+      "learning_rate": 4.494459643804349e-05,
+      "loss": 1.823,
+      "step": 690
+    },
+    {
+      "epoch": 0.7030854830551341,
+      "grad_norm": 14.1875,
+      "learning_rate": 4.486162390913701e-05,
+      "loss": 1.8,
+      "step": 695
+    },
+    {
+      "epoch": 0.7081436519979767,
+      "grad_norm": 17.125,
+      "learning_rate": 4.4778053922214835e-05,
+      "loss": 1.8762,
+      "step": 700
+    },
+    {
+      "epoch": 0.7132018209408194,
+      "grad_norm": 14.75,
+      "learning_rate": 4.469388899114788e-05,
+      "loss": 1.7496,
+      "step": 705
+    },
+    {
+      "epoch": 0.7182599898836621,
+      "grad_norm": 19.875,
+      "learning_rate": 4.4609131647703575e-05,
+      "loss": 1.849,
+      "step": 710
+    },
+    {
+      "epoch": 0.7233181588265049,
+      "grad_norm": 15.625,
+      "learning_rate": 4.4523784441469735e-05,
+      "loss": 1.8202,
+      "step": 715
+    },
+    {
+      "epoch": 0.7283763277693475,
+      "grad_norm": 15.1875,
+      "learning_rate": 4.4437849939777855e-05,
+      "loss": 1.8944,
+      "step": 720
+    },
+    {
+      "epoch": 0.7334344967121902,
+      "grad_norm": 15.375,
+      "learning_rate": 4.4351330727625864e-05,
+      "loss": 1.8257,
+      "step": 725
+    },
+    {
+      "epoch": 0.7384926656550329,
+      "grad_norm": 17.125,
+      "learning_rate": 4.426422940760039e-05,
+      "loss": 1.8551,
+      "step": 730
+    },
+    {
+      "epoch": 0.7435508345978755,
+      "grad_norm": 15.0625,
+      "learning_rate": 4.4176548599798475e-05,
+      "loss": 1.6399,
+      "step": 735
+    },
+    {
+      "epoch": 0.7486090035407182,
+      "grad_norm": 14.875,
+      "learning_rate": 4.408829094174872e-05,
+      "loss": 1.7237,
+      "step": 740
+    },
+    {
+      "epoch": 0.7496206373292867,
+      "eval_loss": 1.755345106124878,
+      "eval_runtime": 255.916,
+      "eval_samples_per_second": 6.803,
+      "eval_steps_per_second": 6.803,
+      "step": 741
+    },
+    {
+      "epoch": 0.753667172483561,
+      "grad_norm": 17.125,
+      "learning_rate": 4.399945908833201e-05,
+      "loss": 1.8802,
+      "step": 745
+    },
+    {
+      "epoch": 0.7587253414264037,
+      "grad_norm": 15.9375,
+      "learning_rate": 4.3910055711701574e-05,
+      "loss": 1.8188,
+      "step": 750
+    },
+    {
+      "epoch": 0.7637835103692463,
+      "grad_norm": 18.875,
+      "learning_rate": 4.382008350120268e-05,
+      "loss": 1.7184,
+      "step": 755
+    },
+    {
+      "epoch": 0.768841679312089,
+      "grad_norm": 15.9375,
+      "learning_rate": 4.372954516329169e-05,
+      "loss": 1.8027,
+      "step": 760
+    },
+    {
+      "epoch": 0.7738998482549317,
+      "grad_norm": 16.875,
+      "learning_rate": 4.3638443421454636e-05,
+      "loss": 1.6897,
+      "step": 765
+    },
+    {
+      "epoch": 0.7789580171977744,
+      "grad_norm": 15.4375,
+      "learning_rate": 4.354678101612536e-05,
+      "loss": 1.7347,
+      "step": 770
+    },
+    {
+      "epoch": 0.7840161861406171,
+      "grad_norm": 15.3125,
+      "learning_rate": 4.345456070460301e-05,
+      "loss": 1.826,
+      "step": 775
+    },
+    {
+      "epoch": 0.7890743550834598,
+      "grad_norm": 15.6875,
+      "learning_rate": 4.336178526096913e-05,
+      "loss": 1.8267,
+      "step": 780
+    },
+    {
+      "epoch": 0.7941325240263025,
+      "grad_norm": 17.75,
+      "learning_rate": 4.326845747600421e-05,
+      "loss": 1.9181,
+      "step": 785
+    },
+    {
+      "epoch": 0.7991906929691451,
+      "grad_norm": 15.6875,
+      "learning_rate": 4.317458015710373e-05,
+      "loss": 2.0775,
+      "step": 790
+    },
+    {
+      "epoch": 0.8042488619119879,
+      "grad_norm": 15.5,
+      "learning_rate": 4.308015612819371e-05,
+      "loss": 1.7947,
+      "step": 795
+    },
+    {
+      "epoch": 0.8093070308548306,
+      "grad_norm": 15.875,
+      "learning_rate": 4.298518822964578e-05,
+      "loss": 1.6998,
+      "step": 800
+    },
+    {
+      "epoch": 0.8143651997976732,
+      "grad_norm": 13.75,
+      "learning_rate": 4.288967931819172e-05,
+      "loss": 1.6267,
+      "step": 805
+    },
+    {
+      "epoch": 0.8194233687405159,
+      "grad_norm": 19.25,
+      "learning_rate": 4.279363226683753e-05,
+      "loss": 1.7968,
+      "step": 810
+    },
+    {
+      "epoch": 0.8244815376833586,
+      "grad_norm": 16.125,
+      "learning_rate": 4.2697049964777014e-05,
+      "loss": 1.6982,
+      "step": 815
+    },
+    {
+      "epoch": 0.8295397066262014,
+      "grad_norm": 17.0,
+      "learning_rate": 4.2599935317304865e-05,
+      "loss": 1.7218,
+      "step": 820
+    },
+    {
+      "epoch": 0.834597875569044,
+      "grad_norm": 17.625,
+      "learning_rate": 4.2502291245729276e-05,
+      "loss": 1.6961,
+      "step": 825
+    },
+    {
+      "epoch": 0.8396560445118867,
+      "grad_norm": 16.625,
+      "learning_rate": 4.240412068728404e-05,
+      "loss": 1.6979,
+      "step": 830
+    },
+    {
+      "epoch": 0.8447142134547294,
+      "grad_norm": 17.5,
+      "learning_rate": 4.2305426595040253e-05,
+      "loss": 1.8622,
+      "step": 835
+    },
+    {
+      "epoch": 0.849772382397572,
+      "grad_norm": 16.75,
+      "learning_rate": 4.2206211937817416e-05,
+      "loss": 1.8359,
+      "step": 840
+    },
+    {
+      "epoch": 0.8548305513404147,
+      "grad_norm": 13.875,
+      "learning_rate": 4.210647970009416e-05,
+      "loss": 1.6391,
+      "step": 845
+    },
+    {
+      "epoch": 0.8598887202832575,
+      "grad_norm": 16.375,
+      "learning_rate": 4.200623288191847e-05,
+      "loss": 1.7364,
+      "step": 850
+    },
+    {
+      "epoch": 0.8649468892261002,
+      "grad_norm": 15.0625,
+      "learning_rate": 4.1905474498817455e-05,
+      "loss": 1.6429,
+      "step": 855
+    },
+    {
+      "epoch": 0.8700050581689428,
+      "grad_norm": 14.0625,
+      "learning_rate": 4.1804207581706586e-05,
+      "loss": 1.8574,
+      "step": 860
+    },
+    {
+      "epoch": 0.8750632271117855,
+      "grad_norm": 13.9375,
+      "learning_rate": 4.170243517679861e-05,
+      "loss": 1.6108,
+      "step": 865
+    },
+    {
+      "epoch": 0.8801213960546282,
+      "grad_norm": 15.375,
+      "learning_rate": 4.160016034551182e-05,
+      "loss": 1.7286,
+      "step": 870
+    },
+    {
+      "epoch": 0.885179564997471,
+      "grad_norm": 15.0,
+      "learning_rate": 4.1497386164378014e-05,
+      "loss": 1.8675,
+      "step": 875
+    },
+    {
+      "epoch": 0.8902377339403136,
+      "grad_norm": 14.625,
+      "learning_rate": 4.139411572494999e-05,
+      "loss": 1.6737,
+      "step": 880
+    },
+    {
+      "epoch": 0.8952959028831563,
+      "grad_norm": 17.375,
+      "learning_rate": 4.129035213370844e-05,
+      "loss": 1.7275,
+      "step": 885
+    },
+    {
+      "epoch": 0.900354071825999,
+      "grad_norm": 14.75,
+      "learning_rate": 4.118609851196862e-05,
+      "loss": 1.7576,
+      "step": 890
+    },
+    {
+      "epoch": 0.9054122407688416,
+      "grad_norm": 13.75,
+      "learning_rate": 4.108135799578637e-05,
+      "loss": 1.686,
+      "step": 895
+    },
+    {
+      "epoch": 0.9104704097116844,
+      "grad_norm": 15.25,
+      "learning_rate": 4.097613373586385e-05,
+      "loss": 1.612,
+      "step": 900
+    },
+    {
+      "epoch": 0.9155285786545271,
+      "grad_norm": 16.0,
+      "learning_rate": 4.08704288974547e-05,
+      "loss": 1.5407,
+      "step": 905
+    },
+    {
+      "epoch": 0.9205867475973698,
+      "grad_norm": 14.3125,
+      "learning_rate": 4.0764246660268855e-05,
+      "loss": 1.7227,
+      "step": 910
+    },
+    {
+      "epoch": 0.9256449165402124,
+      "grad_norm": 13.375,
+      "learning_rate": 4.065759021837691e-05,
+      "loss": 1.6943,
+      "step": 915
+    },
+    {
+      "epoch": 0.9307030854830551,
+      "grad_norm": 16.375,
+      "learning_rate": 4.055046278011401e-05,
+      "loss": 1.775,
+      "step": 920
+    },
+    {
+      "epoch": 0.9357612544258979,
+      "grad_norm": 15.5625,
+      "learning_rate": 4.0442867567983366e-05,
+      "loss": 1.6218,
+      "step": 925
+    },
+    {
+      "epoch": 0.9408194233687405,
+      "grad_norm": 16.625,
+      "learning_rate": 4.033480781855929e-05,
+      "loss": 1.7498,
+      "step": 930
+    },
+    {
+      "epoch": 0.9458775923115832,
+      "grad_norm": 15.875,
+      "learning_rate": 4.022628678238987e-05,
+      "loss": 1.6703,
+      "step": 935
+    },
+    {
+      "epoch": 0.9509357612544259,
+      "grad_norm": 18.5,
+      "learning_rate": 4.011730772389916e-05,
+      "loss": 1.6881,
+      "step": 940
+    },
+    {
+      "epoch": 0.9559939301972686,
+      "grad_norm": 15.0,
+      "learning_rate": 4.0007873921289005e-05,
+      "loss": 1.6605,
+      "step": 945
+    },
+    {
+      "epoch": 0.9610520991401112,
+      "grad_norm": 16.75,
+      "learning_rate": 3.989798866644041e-05,
+      "loss": 1.6931,
+      "step": 950
+    },
+    {
+      "epoch": 0.966110268082954,
+      "grad_norm": 13.1875,
+      "learning_rate": 3.978765526481453e-05,
+      "loss": 1.6725,
+      "step": 955
+    },
+    {
+      "epoch": 0.9711684370257967,
+      "grad_norm": 14.375,
+      "learning_rate": 3.967687703535324e-05,
+      "loss": 1.7586,
+      "step": 960
+    },
+    {
+      "epoch": 0.9762266059686393,
+      "grad_norm": 14.625,
+      "learning_rate": 3.956565731037928e-05,
+      "loss": 1.5827,
+      "step": 965
+    },
+    {
+      "epoch": 0.981284774911482,
+      "grad_norm": 14.375,
+      "learning_rate": 3.945399943549605e-05,
+      "loss": 1.6964,
+      "step": 970
+    },
+    {
+      "epoch": 0.9863429438543247,
+      "grad_norm": 16.625,
+      "learning_rate": 3.934190676948691e-05,
+      "loss": 1.8158,
+      "step": 975
+    },
+    {
+      "epoch": 0.9914011127971675,
+      "grad_norm": 13.8125,
+      "learning_rate": 3.922938268421423e-05,
+      "loss": 1.6117,
+      "step": 980
+    },
+    {
+      "epoch": 0.9964592817400101,
+      "grad_norm": 14.875,
+      "learning_rate": 3.911643056451788e-05,
+      "loss": 1.5232,
+      "step": 985
+    },
+    {
+      "epoch": 0.9994941831057157,
+      "eval_loss": 1.5983182191848755,
+      "eval_runtime": 256.1404,
+      "eval_samples_per_second": 6.797,
+      "eval_steps_per_second": 6.797,
+      "step": 988
+    },
+    {
+      "epoch": 1.0015174506828528,
+      "grad_norm": 14.25,
+      "learning_rate": 3.9003053808113454e-05,
+      "loss": 1.5804,
+      "step": 990
+    },
+    {
+      "epoch": 1.0065756196256954,
+      "grad_norm": 12.625,
+      "learning_rate": 3.888925582549006e-05,
+      "loss": 1.5701,
+      "step": 995
+    },
+    {
+      "epoch": 1.0116337885685383,
+      "grad_norm": 13.5625,
+      "learning_rate": 3.8775040039807706e-05,
+      "loss": 1.6593,
+      "step": 1000
+    },
+    {
+      "epoch": 1.0166919575113809,
+      "grad_norm": 17.5,
+      "learning_rate": 3.866040988679438e-05,
+      "loss": 1.8465,
+      "step": 1005
+    },
+    {
+      "epoch": 1.0040465351542742,
+      "grad_norm": 15.6875,
+      "learning_rate": 3.8545368814642615e-05,
+      "loss": 1.3083,
+      "step": 1010
+    },
+    {
+      "epoch": 1.0091047040971168,
+      "grad_norm": 15.75,
+      "learning_rate": 3.842992028390585e-05,
+      "loss": 1.242,
+      "step": 1015
+    },
+    {
+      "epoch": 1.0141628730399594,
+      "grad_norm": 13.875,
+      "learning_rate": 3.831406776739428e-05,
+      "loss": 1.1496,
+      "step": 1020
+    },
+    {
+      "epoch": 1.0192210419828023,
+      "grad_norm": 14.875,
+      "learning_rate": 3.819781475007042e-05,
+      "loss": 1.1189,
+      "step": 1025
+    },
+    {
+      "epoch": 1.024279210925645,
+      "grad_norm": 15.0625,
+      "learning_rate": 3.808116472894425e-05,
+      "loss": 1.2838,
+      "step": 1030
+    },
+    {
+      "epoch": 1.0293373798684875,
+      "grad_norm": 12.8125,
+      "learning_rate": 3.796412121296801e-05,
+      "loss": 1.0891,
+      "step": 1035
+    },
+    {
+      "epoch": 1.0343955488113303,
+      "grad_norm": 12.3125,
+      "learning_rate": 3.7846687722930724e-05,
+      "loss": 1.1669,
+      "step": 1040
+    },
+    {
+      "epoch": 1.039453717754173,
+      "grad_norm": 18.5,
+      "learning_rate": 3.772886779135217e-05,
+      "loss": 1.0916,
+      "step": 1045
+    },
+    {
+      "epoch": 1.0445118866970158,
+      "grad_norm": 19.25,
+      "learning_rate": 3.7610664962376755e-05,
+      "loss": 1.3949,
+      "step": 1050
+    },
+    {
+      "epoch": 1.0495700556398584,
+      "grad_norm": 9.875,
+      "learning_rate": 3.749208279166676e-05,
+      "loss": 1.1869,
+      "step": 1055
+    },
+    {
+      "epoch": 1.054628224582701,
+      "grad_norm": 13.75,
+      "learning_rate": 3.73731248462955e-05,
+      "loss": 0.9339,
+      "step": 1060
+    },
+    {
+      "epoch": 1.0596863935255438,
+      "grad_norm": 13.3125,
+      "learning_rate": 3.7253794704639976e-05,
+      "loss": 1.2436,
+      "step": 1065
+    },
+    {
+      "epoch": 1.0647445624683864,
+      "grad_norm": 13.625,
+      "learning_rate": 3.713409595627321e-05,
+      "loss": 1.224,
+      "step": 1070
+    },
+    {
+      "epoch": 1.069802731411229,
+      "grad_norm": 14.3125,
+      "learning_rate": 3.7014032201856316e-05,
+      "loss": 1.2101,
+      "step": 1075
+    },
+    {
+      "epoch": 1.074860900354072,
+      "grad_norm": 12.625,
+      "learning_rate": 3.6893607053030145e-05,
+      "loss": 1.2266,
+      "step": 1080
+    },
+    {
+      "epoch": 1.0799190692969145,
+      "grad_norm": 11.5625,
+      "learning_rate": 3.677282413230668e-05,
+      "loss": 1.2284,
+      "step": 1085
+    },
+    {
+      "epoch": 1.0849772382397571,
+      "grad_norm": 13.5,
+      "learning_rate": 3.665168707296004e-05,
+      "loss": 1.1808,
+      "step": 1090
+    },
+    {
+      "epoch": 1.0900354071826,
+      "grad_norm": 16.0,
+      "learning_rate": 3.653019951891719e-05,
+      "loss": 1.2844,
+      "step": 1095
+    },
+    {
+      "epoch": 1.0950935761254426,
+      "grad_norm": 14.3125,
+      "learning_rate": 3.640836512464837e-05,
+      "loss": 1.2463,
+      "step": 1100
+    },
+    {
+      "epoch": 1.1001517450682852,
+      "grad_norm": 12.8125,
+      "learning_rate": 3.6286187555057075e-05,
+      "loss": 1.2075,
+      "step": 1105
+    },
+    {
+      "epoch": 1.105209914011128,
+      "grad_norm": 12.75,
+      "learning_rate": 3.61636704853699e-05,
+      "loss": 1.1847,
+      "step": 1110
+    },
+    {
+      "epoch": 1.1102680829539706,
+      "grad_norm": 13.125,
+      "learning_rate": 3.604081760102596e-05,
+      "loss": 1.1479,
+      "step": 1115
+    },
+    {
+      "epoch": 1.1153262518968134,
+      "grad_norm": 13.9375,
+      "learning_rate": 3.591763259756598e-05,
+      "loss": 1.2051,
+      "step": 1120
+    },
+    {
+      "epoch": 1.120384420839656,
+      "grad_norm": 12.75,
+      "learning_rate": 3.579411918052121e-05,
+      "loss": 1.1309,
+      "step": 1125
+    },
+    {
+      "epoch": 1.1254425897824987,
+      "grad_norm": 13.75,
+      "learning_rate": 3.56702810653019e-05,
+      "loss": 1.2491,
+      "step": 1130
+    },
+    {
+      "epoch": 1.1305007587253415,
+      "grad_norm": 13.4375,
+      "learning_rate": 3.554612197708555e-05,
+      "loss": 1.0678,
+      "step": 1135
+    },
+    {
+      "epoch": 1.1355589276681841,
+      "grad_norm": 12.375,
+      "learning_rate": 3.542164565070486e-05,
+      "loss": 1.071,
+      "step": 1140
+    },
+    {
+      "epoch": 1.1406170966110267,
+      "grad_norm": 13.125,
+      "learning_rate": 3.529685583053538e-05,
+      "loss": 1.1301,
+      "step": 1145
+    },
+    {
+      "epoch": 1.1456752655538696,
+      "grad_norm": 12.4375,
+      "learning_rate": 3.517175627038289e-05,
+      "loss": 1.2263,
+      "step": 1150
+    },
+    {
+      "epoch": 1.1507334344967122,
+      "grad_norm": 12.5,
+      "learning_rate": 3.504635073337044e-05,
+      "loss": 1.2053,
+      "step": 1155
+    },
+    {
+      "epoch": 1.155791603439555,
+      "grad_norm": 11.3125,
+      "learning_rate": 3.492064299182517e-05,
+      "loss": 1.258,
+      "step": 1160
+    },
+    {
+      "epoch": 1.1608497723823976,
+      "grad_norm": 13.3125,
+      "learning_rate": 3.4794636827164894e-05,
+      "loss": 1.1801,
+      "step": 1165
+    },
+    {
+      "epoch": 1.1659079413252402,
+      "grad_norm": 12.25,
+      "learning_rate": 3.466833602978425e-05,
+      "loss": 1.2318,
+      "step": 1170
+    },
+    {
+      "epoch": 1.170966110268083,
+      "grad_norm": 12.5625,
+      "learning_rate": 3.4541744398940757e-05,
+      "loss": 1.1059,
+      "step": 1175
+    },
+    {
+      "epoch": 1.1760242792109257,
+      "grad_norm": 12.5625,
+      "learning_rate": 3.441486574264048e-05,
+      "loss": 1.2274,
+      "step": 1180
+    },
+    {
+      "epoch": 1.1810824481537683,
+      "grad_norm": 14.0625,
+      "learning_rate": 3.4287703877523526e-05,
+      "loss": 1.2024,
+      "step": 1185
+    },
+    {
+      "epoch": 1.1861406170966111,
+      "grad_norm": 13.75,
+      "learning_rate": 3.4160262628749204e-05,
+      "loss": 1.2123,
+      "step": 1190
+    },
+    {
+      "epoch": 1.1911987860394537,
+      "grad_norm": 11.4375,
+      "learning_rate": 3.403254582988096e-05,
+      "loss": 1.1231,
+      "step": 1195
+    },
+    {
+      "epoch": 1.1962569549822963,
+      "grad_norm": 12.8125,
+      "learning_rate": 3.39045573227711e-05,
+      "loss": 1.0944,
+      "step": 1200
+    },
+    {
+      "epoch": 1.2013151239251392,
+      "grad_norm": 12.6875,
+      "learning_rate": 3.377630095744516e-05,
+      "loss": 1.3242,
+      "step": 1205
+    },
+    {
+      "epoch": 1.2063732928679818,
+      "grad_norm": 12.5625,
+      "learning_rate": 3.364778059198614e-05,
+      "loss": 1.1179,
+      "step": 1210
+    },
+    {
+      "epoch": 1.2114314618108244,
+      "grad_norm": 12.3125,
+      "learning_rate": 3.351900009241844e-05,
+      "loss": 1.2562,
+      "step": 1215
+    },
+    {
+      "epoch": 1.2164896307536672,
+      "grad_norm": 11.125,
+      "learning_rate": 3.338996333259155e-05,
+      "loss": 1.1211,
+      "step": 1220
+    },
+    {
+      "epoch": 1.2215477996965098,
+      "grad_norm": 11.8125,
+      "learning_rate": 3.326067419406352e-05,
+      "loss": 1.0939,
+      "step": 1225
+    },
+    {
+      "epoch": 1.2266059686393525,
+      "grad_norm": 11.8125,
+      "learning_rate": 3.3131136565984255e-05,
+      "loss": 1.2619,
+      "step": 1230
+    },
+    {
+      "epoch": 1.2316641375821953,
+      "grad_norm": 13.625,
+      "learning_rate": 3.300135434497843e-05,
+      "loss": 1.1344,
+      "step": 1235
+    },
+    {
+      "epoch": 1.2316641375821953,
+      "eval_loss": 1.553524374961853,
+      "eval_runtime": 256.0029,
+      "eval_samples_per_second": 6.801,
+      "eval_steps_per_second": 6.801,
+      "step": 1235
+    },
+    {
+      "epoch": 1.236722306525038,
+      "grad_norm": 10.25,
+      "learning_rate": 3.287133143502835e-05,
+      "loss": 1.0952,
+      "step": 1240
+    },
+    {
+      "epoch": 1.2417804754678805,
+      "grad_norm": 14.375,
+      "learning_rate": 3.274107174735649e-05,
+      "loss": 1.2879,
+      "step": 1245
+    },
+    {
+      "epoch": 1.2468386444107233,
+      "grad_norm": 16.75,
+      "learning_rate": 3.261057920030784e-05,
+      "loss": 1.132,
+      "step": 1250
+    },
+    {
+      "epoch": 1.251896813353566,
+      "grad_norm": 11.0,
+      "learning_rate": 3.247985771923205e-05,
+      "loss": 1.1082,
+      "step": 1255
+    },
+    {
+      "epoch": 1.2569549822964086,
+      "grad_norm": 12.1875,
+      "learning_rate": 3.23489112363653e-05,
+      "loss": 1.014,
+      "step": 1260
+    },
+    {
+      "epoch": 1.2620131512392514,
+      "grad_norm": 13.625,
+      "learning_rate": 3.22177436907121e-05,
+      "loss": 1.2431,
+      "step": 1265
+    },
+    {
+      "epoch": 1.267071320182094,
+      "grad_norm": 13.0,
+      "learning_rate": 3.2086359027926743e-05,
+      "loss": 1.0752,
+      "step": 1270
+    },
+    {
+      "epoch": 1.2721294891249368,
+      "grad_norm": 15.0,
+      "learning_rate": 3.195476120019462e-05,
+      "loss": 1.2593,
+      "step": 1275
+    },
+    {
+      "epoch": 1.2771876580677795,
+      "grad_norm": 13.625,
+      "learning_rate": 3.182295416611334e-05,
+      "loss": 1.272,
+      "step": 1280
+    },
+    {
+      "epoch": 1.2822458270106223,
+      "grad_norm": 12.0625,
+      "learning_rate": 3.1690941890573663e-05,
+      "loss": 1.1347,
+      "step": 1285
+    },
+    {
+      "epoch": 1.287303995953465,
+      "grad_norm": 12.1875,
+      "learning_rate": 3.155872834464021e-05,
+      "loss": 1.1248,
+      "step": 1290
+    },
+    {
+      "epoch": 1.2923621648963075,
+      "grad_norm": 11.875,
+      "learning_rate": 3.142631750543205e-05,
+      "loss": 1.1644,
+      "step": 1295
+    },
+    {
+      "epoch": 1.2974203338391503,
+      "grad_norm": 10.0,
+      "learning_rate": 3.1293713356003005e-05,
+      "loss": 0.9635,
+      "step": 1300
+    },
+    {
+      "epoch": 1.302478502781993,
+      "grad_norm": 12.3125,
+      "learning_rate": 3.116091988522186e-05,
+      "loss": 1.2456,
+      "step": 1305
+    },
+    {
+      "epoch": 1.3075366717248356,
+      "grad_norm": 16.125,
+      "learning_rate": 3.102794108765244e-05,
+      "loss": 1.1195,
+      "step": 1310
+    },
+    {
+      "epoch": 1.3125948406676784,
+      "grad_norm": 12.5,
+      "learning_rate": 3.0894780963433306e-05,
+      "loss": 1.1988,
+      "step": 1315
+    },
+    {
+      "epoch": 1.317653009610521,
+      "grad_norm": 12.3125,
+      "learning_rate": 3.0761443518157577e-05,
+      "loss": 1.148,
+      "step": 1320
+    },
+    {
+      "epoch": 1.3227111785533636,
+      "grad_norm": 12.875,
+      "learning_rate": 3.062793276275235e-05,
+      "loss": 1.1715,
+      "step": 1325
+    },
+    {
+      "epoch": 1.3277693474962065,
+      "grad_norm": 13.75,
+      "learning_rate": 3.0494252713358052e-05,
+      "loss": 1.213,
+      "step": 1330
+    },
+    {
+      "epoch": 1.332827516439049,
+      "grad_norm": 12.1875,
+      "learning_rate": 3.0360407391207663e-05,
+      "loss": 1.0854,
+      "step": 1335
+    },
+    {
+      "epoch": 1.3378856853818917,
+      "grad_norm": 13.1875,
+      "learning_rate": 3.0226400822505735e-05,
+      "loss": 1.1538,
+      "step": 1340
+    },
+    {
+      "epoch": 1.3429438543247345,
+      "grad_norm": 12.5,
+      "learning_rate": 3.0092237038307293e-05,
+      "loss": 1.1367,
+      "step": 1345
+    },
+    {
+      "epoch": 1.3480020232675771,
+      "grad_norm": 11.875,
+      "learning_rate": 2.9957920074396535e-05,
+      "loss": 1.1167,
+      "step": 1350
+    },
+    {
+      "epoch": 1.3530601922104197,
+      "grad_norm": 11.875,
+      "learning_rate": 2.9823453971165495e-05,
+      "loss": 1.1201,
+      "step": 1355
+    },
+    {
+      "epoch": 1.3581183611532626,
+      "grad_norm": 11.0,
+      "learning_rate": 2.9688842773492443e-05,
+      "loss": 1.0583,
+      "step": 1360
+    },
+    {
+      "epoch": 1.3631765300961052,
+      "grad_norm": 13.25,
+      "learning_rate": 2.9554090530620264e-05,
+      "loss": 1.1696,
+      "step": 1365
+    },
+    {
+      "epoch": 1.3682346990389478,
+      "grad_norm": 13.4375,
+      "learning_rate": 2.941920129603461e-05,
+      "loss": 1.1046,
+      "step": 1370
+    },
+    {
+      "epoch": 1.3732928679817906,
+      "grad_norm": 12.375,
+      "learning_rate": 2.9284179127341987e-05,
+      "loss": 1.1178,
+      "step": 1375
+    },
+    {
+      "epoch": 1.3783510369246332,
+      "grad_norm": 13.4375,
+      "learning_rate": 2.91490280861477e-05,
+      "loss": 1.1733,
+      "step": 1380
+    },
+    {
+      "epoch": 1.3834092058674758,
+      "grad_norm": 12.1875,
+      "learning_rate": 2.9013752237933668e-05,
+      "loss": 1.2754,
+      "step": 1385
+    },
+    {
+      "epoch": 1.3884673748103187,
+      "grad_norm": 11.875,
+      "learning_rate": 2.8878355651936128e-05,
+      "loss": 1.1508,
+      "step": 1390
+    },
+    {
+      "epoch": 1.3935255437531613,
+      "grad_norm": 10.1875,
+      "learning_rate": 2.8742842401023246e-05,
+      "loss": 1.0746,
+      "step": 1395
+    },
+    {
+      "epoch": 1.3985837126960041,
+      "grad_norm": 13.0,
+      "learning_rate": 2.8607216561572585e-05,
+      "loss": 1.2091,
+      "step": 1400
+    },
+    {
+      "epoch": 1.4036418816388467,
+      "grad_norm": 12.125,
+      "learning_rate": 2.8471482213348467e-05,
+      "loss": 1.0808,
+      "step": 1405
+    },
+    {
+      "epoch": 1.4087000505816896,
+      "grad_norm": 16.375,
+      "learning_rate": 2.83356434393793e-05,
+      "loss": 1.1411,
+      "step": 1410
+    },
+    {
+      "epoch": 1.4137582195245322,
+      "grad_norm": 11.25,
+      "learning_rate": 2.8199704325834704e-05,
+      "loss": 0.9701,
+      "step": 1415
+    },
+    {
+      "epoch": 1.4188163884673748,
+      "grad_norm": 12.4375,
+      "learning_rate": 2.8063668961902622e-05,
+      "loss": 1.1979,
+      "step": 1420
+    },
+    {
+      "epoch": 1.4238745574102176,
+      "grad_norm": 12.25,
+      "learning_rate": 2.792754143966631e-05,
+      "loss": 1.1325,
+      "step": 1425
+    },
+    {
+      "epoch": 1.4289327263530602,
+      "grad_norm": 11.125,
+      "learning_rate": 2.779132585398123e-05,
+      "loss": 1.1739,
+      "step": 1430
+    },
+    {
+      "epoch": 1.4339908952959028,
+      "grad_norm": 12.25,
+      "learning_rate": 2.7655026302351917e-05,
+      "loss": 1.1228,
+      "step": 1435
+    },
+    {
+      "epoch": 1.4390490642387457,
+      "grad_norm": 10.5,
+      "learning_rate": 2.7518646884808635e-05,
+      "loss": 1.0809,
+      "step": 1440
+    },
+    {
+      "epoch": 1.4441072331815883,
+      "grad_norm": 15.5,
+      "learning_rate": 2.7382191703784133e-05,
+      "loss": 1.0879,
+      "step": 1445
+    },
+    {
+      "epoch": 1.449165402124431,
+      "grad_norm": 12.3125,
+      "learning_rate": 2.7245664863990188e-05,
+      "loss": 1.0487,
+      "step": 1450
+    },
+    {
+      "epoch": 1.4542235710672737,
+      "grad_norm": 11.875,
+      "learning_rate": 2.7109070472294173e-05,
+      "loss": 1.0813,
+      "step": 1455
+    },
+    {
+      "epoch": 1.4592817400101163,
+      "grad_norm": 12.0,
+      "learning_rate": 2.6972412637595444e-05,
+      "loss": 1.1697,
+      "step": 1460
+    },
+    {
+      "epoch": 1.464339908952959,
+      "grad_norm": 10.625,
+      "learning_rate": 2.6835695470701834e-05,
+      "loss": 1.2109,
+      "step": 1465
+    },
+    {
+      "epoch": 1.4693980778958018,
+      "grad_norm": 10.8125,
+      "learning_rate": 2.669892308420592e-05,
+      "loss": 1.1037,
+      "step": 1470
+    },
+    {
+      "epoch": 1.4744562468386444,
+      "grad_norm": 11.625,
+      "learning_rate": 2.656209959236135e-05,
+      "loss": 1.0944,
+      "step": 1475
+    },
+    {
+      "epoch": 1.479514415781487,
+      "grad_norm": 10.6875,
+      "learning_rate": 2.6425229110959078e-05,
+      "loss": 1.0946,
+      "step": 1480
+    },
+    {
+      "epoch": 1.4815376833586242,
+      "eval_loss": 1.4312002658843994,
+      "eval_runtime": 256.0588,
+      "eval_samples_per_second": 6.799,
+      "eval_steps_per_second": 6.799,
+      "step": 1482
+    },
+    {
+      "epoch": 1.4845725847243298,
+      "grad_norm": 10.125,
+      "learning_rate": 2.6288315757203536e-05,
+      "loss": 1.0402,
+      "step": 1485
+    },
+    {
+      "epoch": 1.4896307536671725,
+      "grad_norm": 11.3125,
+      "learning_rate": 2.615136364958884e-05,
+      "loss": 1.1341,
+      "step": 1490
+    },
+    {
+      "epoch": 1.494688922610015,
+      "grad_norm": 11.0,
+      "learning_rate": 2.6014376907774802e-05,
+      "loss": 1.0339,
+      "step": 1495
+    },
+    {
+      "epoch": 1.499747091552858,
+      "grad_norm": 10.0,
+      "learning_rate": 2.5877359652463125e-05,
+      "loss": 1.0749,
+      "step": 1500
+    },
+    {
+      "epoch": 1.5048052604957005,
+      "grad_norm": 11.9375,
+      "learning_rate": 2.574031600527334e-05,
+      "loss": 0.9995,
+      "step": 1505
+    },
+    {
+      "epoch": 1.5098634294385431,
+      "grad_norm": 12.75,
+      "learning_rate": 2.5603250088618917e-05,
+      "loss": 1.068,
+      "step": 1510
+    },
+    {
+      "epoch": 1.514921598381386,
+      "grad_norm": 10.125,
+      "learning_rate": 2.5466166025583184e-05,
+      "loss": 1.1409,
+      "step": 1515
+    },
+    {
+      "epoch": 1.5199797673242288,
+      "grad_norm": 12.5,
+      "learning_rate": 2.5329067939795324e-05,
+      "loss": 1.2126,
+      "step": 1520
+    },
+    {
+      "epoch": 1.5250379362670712,
+      "grad_norm": 10.25,
+      "learning_rate": 2.5191959955306387e-05,
+      "loss": 1.0113,
+      "step": 1525
+    },
+    {
+      "epoch": 1.530096105209914,
+      "grad_norm": 11.5625,
+      "learning_rate": 2.505484619646512e-05,
+      "loss": 1.1573,
+      "step": 1530
+    },
+    {
+      "epoch": 1.5351542741527568,
+      "grad_norm": 11.6875,
+      "learning_rate": 2.4917730787794023e-05,
+      "loss": 1.007,
+      "step": 1535
+    },
+    {
+      "epoch": 1.5402124430955992,
+      "grad_norm": 11.4375,
+      "learning_rate": 2.4780617853865197e-05,
+      "loss": 1.0884,
+      "step": 1540
+    },
+    {
+      "epoch": 1.545270612038442,
+      "grad_norm": 12.625,
+      "learning_rate": 2.464351151917631e-05,
+      "loss": 1.0363,
+      "step": 1545
+    },
+    {
+      "epoch": 1.550328780981285,
+      "grad_norm": 13.9375,
+      "learning_rate": 2.450641590802652e-05,
+      "loss": 1.1238,
+      "step": 1550
+    },
+    {
+      "epoch": 1.5553869499241275,
+      "grad_norm": 11.3125,
+      "learning_rate": 2.436933514439239e-05,
+      "loss": 1.0095,
+      "step": 1555
+    },
+    {
+      "epoch": 1.5604451188669701,
+      "grad_norm": 13.4375,
+      "learning_rate": 2.423227335180387e-05,
+      "loss": 1.0518,
+      "step": 1560
+    },
+    {
+      "epoch": 1.565503287809813,
+      "grad_norm": 11.5,
+      "learning_rate": 2.4095234653220254e-05,
+      "loss": 1.1204,
+      "step": 1565
+    },
+    {
+      "epoch": 1.5705614567526556,
+      "grad_norm": 10.6875,
+      "learning_rate": 2.3958223170906126e-05,
+      "loss": 1.0028,
+      "step": 1570
+    },
+    {
+      "epoch": 1.5756196256954982,
+      "grad_norm": 9.8125,
+      "learning_rate": 2.382124302630738e-05,
+      "loss": 0.9897,
+      "step": 1575
+    },
+    {
+      "epoch": 1.580677794638341,
+      "grad_norm": 10.875,
+      "learning_rate": 2.368429833992724e-05,
+      "loss": 1.078,
+      "step": 1580
+    },
+    {
+      "epoch": 1.5857359635811836,
+      "grad_norm": 12.125,
+      "learning_rate": 2.354739323120231e-05,
+      "loss": 1.1009,
+      "step": 1585
+    },
+    {
+      "epoch": 1.5907941325240262,
+      "grad_norm": 10.5,
+      "learning_rate": 2.3410531818378684e-05,
+      "loss": 1.0235,
+      "step": 1590
+    },
+    {
+      "epoch": 1.595852301466869,
+      "grad_norm": 11.25,
+      "learning_rate": 2.3273718218387974e-05,
+      "loss": 1.0469,
+      "step": 1595
+    },
+    {
+      "epoch": 1.6009104704097117,
+      "grad_norm": 12.0,
+      "learning_rate": 2.31369565467236e-05,
+      "loss": 1.0201,
+      "step": 1600
+    },
+    {
+      "epoch": 1.6059686393525543,
+      "grad_norm": 11.1875,
+      "learning_rate": 2.3000250917316886e-05,
+      "loss": 1.0471,
+      "step": 1605
+    },
+    {
+      "epoch": 1.6110268082953971,
+      "grad_norm": 12.1875,
+      "learning_rate": 2.2863605442413365e-05,
+      "loss": 1.08,
+      "step": 1610
+    },
+    {
+      "epoch": 1.6160849772382397,
+      "grad_norm": 12.875,
+      "learning_rate": 2.2727024232449056e-05,
+      "loss": 1.0742,
+      "step": 1615
+    },
+    {
+      "epoch": 1.6211431461810824,
+      "grad_norm": 12.8125,
+      "learning_rate": 2.259051139592681e-05,
+      "loss": 1.1045,
+      "step": 1620
+    },
+    {
+      "epoch": 1.6262013151239252,
+      "grad_norm": 13.0625,
+      "learning_rate": 2.2454071039292777e-05,
+      "loss": 1.0407,
+      "step": 1625
+    },
+    {
+      "epoch": 1.6312594840667678,
+      "grad_norm": 11.8125,
+      "learning_rate": 2.2317707266812767e-05,
+      "loss": 1.0816,
+      "step": 1630
+    },
+    {
+      "epoch": 1.6363176530096104,
+      "grad_norm": 9.125,
+      "learning_rate": 2.218142418044892e-05,
+      "loss": 0.9233,
+      "step": 1635
+    },
+    {
+      "epoch": 1.6413758219524532,
+      "grad_norm": 12.625,
+      "learning_rate": 2.204522587973623e-05,
+      "loss": 1.0843,
+      "step": 1640
+    },
+    {
+      "epoch": 1.6464339908952959,
+      "grad_norm": 10.375,
+      "learning_rate": 2.1909116461659266e-05,
+      "loss": 1.0033,
+      "step": 1645
+    },
+    {
+      "epoch": 1.6514921598381385,
+      "grad_norm": 11.625,
+      "learning_rate": 2.17731000205289e-05,
+      "loss": 1.009,
+      "step": 1650
+    },
+    {
+      "epoch": 1.6565503287809813,
+      "grad_norm": 11.5,
+      "learning_rate": 2.1637180647859162e-05,
+      "loss": 1.1145,
+      "step": 1655
+    },
+    {
+      "epoch": 1.6616084977238241,
+      "grad_norm": 11.5625,
+      "learning_rate": 2.1501362432244193e-05,
+      "loss": 0.9659,
+      "step": 1660
+    },
+    {
+      "epoch": 1.6666666666666665,
+      "grad_norm": 12.625,
+      "learning_rate": 2.136564945923518e-05,
+      "loss": 1.0536,
+      "step": 1665
+    },
+    {
+      "epoch": 1.6717248356095094,
+      "grad_norm": 10.125,
+      "learning_rate": 2.1230045811217533e-05,
+      "loss": 0.8739,
+      "step": 1670
+    },
+    {
+      "epoch": 1.6767830045523522,
+      "grad_norm": 12.4375,
+      "learning_rate": 2.1094555567288048e-05,
+      "loss": 1.0167,
+      "step": 1675
+    },
+    {
+      "epoch": 1.6818411734951946,
+      "grad_norm": 11.1875,
+      "learning_rate": 2.0959182803132217e-05,
+      "loss": 1.0289,
+      "step": 1680
+    },
+    {
+      "epoch": 1.6868993424380374,
+      "grad_norm": 11.0,
+      "learning_rate": 2.0823931590901606e-05,
+      "loss": 1.0191,
+      "step": 1685
+    },
+    {
+      "epoch": 1.6919575113808802,
+      "grad_norm": 9.875,
+      "learning_rate": 2.068880599909139e-05,
+      "loss": 0.9128,
+      "step": 1690
+    },
+    {
+      "epoch": 1.6970156803237229,
+      "grad_norm": 10.6875,
+      "learning_rate": 2.0553810092417942e-05,
+      "loss": 1.0658,
+      "step": 1695
+    },
+    {
+      "epoch": 1.7020738492665655,
+      "grad_norm": 10.1875,
+      "learning_rate": 2.0418947931696574e-05,
+      "loss": 0.9534,
+      "step": 1700
+    },
+    {
+      "epoch": 1.7071320182094083,
+      "grad_norm": 11.75,
+      "learning_rate": 2.028422357371938e-05,
+      "loss": 1.0896,
+      "step": 1705
+    },
+    {
+      "epoch": 1.712190187152251,
+      "grad_norm": 15.0,
+      "learning_rate": 2.0149641071133217e-05,
+      "loss": 1.027,
+      "step": 1710
+    },
+    {
+      "epoch": 1.7172483560950935,
+      "grad_norm": 11.9375,
+      "learning_rate": 2.001520447231778e-05,
+      "loss": 1.0639,
+      "step": 1715
+    },
+    {
+      "epoch": 1.7223065250379364,
+      "grad_norm": 13.4375,
+      "learning_rate": 1.988091782126382e-05,
+      "loss": 1.0887,
+      "step": 1720
+    },
+    {
+      "epoch": 1.727364693980779,
+      "grad_norm": 11.4375,
+      "learning_rate": 1.974678515745153e-05,
+      "loss": 0.9878,
+      "step": 1725
+    },
+    {
+      "epoch": 1.7314112291350532,
+      "eval_loss": 1.327903389930725,
+      "eval_runtime": 256.0013,
+      "eval_samples_per_second": 6.801,
+      "eval_steps_per_second": 6.801,
+      "step": 1729
+    },
+    {
+      "epoch": 1.7324228629236216,
+      "grad_norm": 11.375,
+      "learning_rate": 1.9612810515728987e-05,
+      "loss": 1.0591,
+      "step": 1730
+    },
+    {
+      "epoch": 1.7374810318664644,
+      "grad_norm": 14.625,
+      "learning_rate": 1.94789979261908e-05,
+      "loss": 1.2031,
+      "step": 1735
+    },
+    {
+      "epoch": 1.742539200809307,
+      "grad_norm": 11.0,
+      "learning_rate": 1.9345351414056893e-05,
+      "loss": 1.0341,
+      "step": 1740
+    },
+    {
+      "epoch": 1.7475973697521496,
+      "grad_norm": 10.25,
+      "learning_rate": 1.9211874999551414e-05,
+      "loss": 0.971,
+      "step": 1745
+    },
+    {
+      "epoch": 1.7526555386949925,
+      "grad_norm": 10.5625,
+      "learning_rate": 1.9078572697781792e-05,
+      "loss": 1.079,
+      "step": 1750
+    },
+    {
+      "epoch": 1.757713707637835,
+      "grad_norm": 10.1875,
+      "learning_rate": 1.8945448518617963e-05,
+      "loss": 1.008,
+      "step": 1755
+    },
+    {
+      "epoch": 1.7627718765806777,
+      "grad_norm": 11.3125,
+      "learning_rate": 1.8812506466571768e-05,
+      "loss": 0.9894,
+      "step": 1760
+    },
+    {
+      "epoch": 1.7678300455235205,
+      "grad_norm": 10.5,
+      "learning_rate": 1.8679750540676462e-05,
+      "loss": 0.987,
+      "step": 1765
+    },
+    {
+      "epoch": 1.7728882144663631,
+      "grad_norm": 10.1875,
+      "learning_rate": 1.854718473436644e-05,
+      "loss": 1.0557,
+      "step": 1770
+    },
+    {
+      "epoch": 1.7779463834092057,
+      "grad_norm": 10.3125,
+      "learning_rate": 1.8414813035357094e-05,
+      "loss": 1.4272,
+      "step": 1775
+    },
+    {
+      "epoch": 1.7830045523520486,
+      "grad_norm": 11.0625,
+      "learning_rate": 1.82826394255249e-05,
+      "loss": 0.9094,
+      "step": 1780
+    },
+    {
+      "epoch": 1.7880627212948914,
+      "grad_norm": 10.25,
+      "learning_rate": 1.815066788078757e-05,
+      "loss": 0.9773,
+      "step": 1785
+    },
+    {
+      "epoch": 1.7931208902377338,
+      "grad_norm": 14.625,
+      "learning_rate": 1.8018902370984524e-05,
+      "loss": 1.0779,
+      "step": 1790
+    },
+    {
+      "epoch": 1.7981790591805766,
+      "grad_norm": 11.875,
+      "learning_rate": 1.788734685975742e-05,
+      "loss": 1.0728,
+      "step": 1795
+    },
+    {
+      "epoch": 1.8032372281234195,
+      "grad_norm": 11.5,
+      "learning_rate": 1.7756005304430944e-05,
+      "loss": 0.9823,
+      "step": 1800
+    },
+    {
+      "epoch": 1.8082953970662619,
+      "grad_norm": 11.0625,
+      "learning_rate": 1.762488165589379e-05,
+      "loss": 1.0346,
+      "step": 1805
+    },
+    {
+      "epoch": 1.8133535660091047,
+      "grad_norm": 9.6875,
+      "learning_rate": 1.7493979858479755e-05,
+      "loss": 0.9136,
+      "step": 1810
+    },
+    {
+      "epoch": 1.8184117349519475,
+      "grad_norm": 10.8125,
+      "learning_rate": 1.736330384984916e-05,
+      "loss": 0.8479,
+      "step": 1815
+    },
+    {
+      "epoch": 1.8234699038947901,
+      "grad_norm": 11.6875,
+      "learning_rate": 1.723285756087034e-05,
+      "loss": 1.0573,
+      "step": 1820
+    },
+    {
+      "epoch": 1.8285280728376327,
+      "grad_norm": 9.375,
+      "learning_rate": 1.7102644915501468e-05,
+      "loss": 1.0019,
+      "step": 1825
+    },
+    {
+      "epoch": 1.8335862417804756,
+      "grad_norm": 10.6875,
+      "learning_rate": 1.6972669830672434e-05,
+      "loss": 0.9915,
+      "step": 1830
+    },
+    {
+      "epoch": 1.8386444107233182,
+      "grad_norm": 12.4375,
+      "learning_rate": 1.6842936216167087e-05,
+      "loss": 0.9397,
+      "step": 1835
+    },
+    {
+      "epoch": 1.8437025796661608,
+      "grad_norm": 11.8125,
+      "learning_rate": 1.6713447974505602e-05,
+      "loss": 1.0044,
+      "step": 1840
+    },
+    {
+      "epoch": 1.8487607486090036,
+      "grad_norm": 11.375,
+      "learning_rate": 1.658420900082707e-05,
+      "loss": 1.0673,
+      "step": 1845
+    },
+    {
+      "epoch": 1.8538189175518462,
+      "grad_norm": 12.25,
+      "learning_rate": 1.6455223182772367e-05,
+      "loss": 0.9595,
+      "step": 1850
+    },
+    {
+      "epoch": 1.8588770864946889,
+      "grad_norm": 10.875,
+      "learning_rate": 1.632649440036716e-05,
+      "loss": 1.0325,
+      "step": 1855
+    },
+    {
+      "epoch": 1.8639352554375317,
+      "grad_norm": 11.1875,
+      "learning_rate": 1.6198026525905258e-05,
+      "loss": 0.9897,
+      "step": 1860
+    },
+    {
+      "epoch": 1.8689934243803743,
+      "grad_norm": 9.625,
+      "learning_rate": 1.6069823423832057e-05,
+      "loss": 0.9835,
+      "step": 1865
+    },
+    {
+      "epoch": 1.874051593323217,
+      "grad_norm": 12.125,
+      "learning_rate": 1.5941888950628358e-05,
+      "loss": 1.0567,
+      "step": 1870
+    },
+    {
+      "epoch": 1.8791097622660597,
+      "grad_norm": 10.25,
+      "learning_rate": 1.5814226954694307e-05,
+      "loss": 1.0093,
+      "step": 1875
+    },
+    {
+      "epoch": 1.8841679312089024,
+      "grad_norm": 9.625,
+      "learning_rate": 1.5686841276233655e-05,
+      "loss": 0.8856,
+      "step": 1880
+    },
+    {
+      "epoch": 1.889226100151745,
+      "grad_norm": 10.5,
+      "learning_rate": 1.5559735747138247e-05,
+      "loss": 0.9369,
+      "step": 1885
+    },
+    {
+      "epoch": 1.8942842690945878,
+      "grad_norm": 10.6875,
+      "learning_rate": 1.5432914190872757e-05,
+      "loss": 0.9683,
+      "step": 1890
+    },
+    {
+      "epoch": 1.8993424380374304,
+      "grad_norm": 11.875,
+      "learning_rate": 1.5306380422359652e-05,
+      "loss": 1.0251,
+      "step": 1895
+    },
+    {
+      "epoch": 1.904400606980273,
+      "grad_norm": 8.75,
+      "learning_rate": 1.5180138247864445e-05,
+      "loss": 0.8458,
+      "step": 1900
+    },
+    {
+      "epoch": 1.9094587759231159,
+      "grad_norm": 10.75,
+      "learning_rate": 1.5054191464881219e-05,
+      "loss": 1.0583,
+      "step": 1905
+    },
+    {
+      "epoch": 1.9145169448659585,
+      "grad_norm": 10.625,
+      "learning_rate": 1.4928543862018368e-05,
+      "loss": 0.9317,
+      "step": 1910
+    },
+    {
+      "epoch": 1.919575113808801,
+      "grad_norm": 9.8125,
+      "learning_rate": 1.4803199218884628e-05,
+      "loss": 0.9129,
+      "step": 1915
+    },
+    {
+      "epoch": 1.924633282751644,
+      "grad_norm": 9.75,
+      "learning_rate": 1.4678161305975414e-05,
+      "loss": 1.0195,
+      "step": 1920
+    },
+    {
+      "epoch": 1.9296914516944867,
+      "grad_norm": 11.625,
+      "learning_rate": 1.4553433884559375e-05,
+      "loss": 0.9456,
+      "step": 1925
+    },
+    {
+      "epoch": 1.9347496206373291,
+      "grad_norm": 12.375,
+      "learning_rate": 1.4429020706565257e-05,
+      "loss": 0.9962,
+      "step": 1930
+    },
+    {
+      "epoch": 1.939807789580172,
+      "grad_norm": 10.1875,
+      "learning_rate": 1.4304925514469056e-05,
+      "loss": 0.8925,
+      "step": 1935
+    },
+    {
+      "epoch": 1.9448659585230148,
+      "grad_norm": 11.125,
+      "learning_rate": 1.418115204118139e-05,
+      "loss": 0.9802,
+      "step": 1940
+    },
+    {
+      "epoch": 1.9499241274658572,
+      "grad_norm": 10.0625,
+      "learning_rate": 1.4057704009935288e-05,
+      "loss": 0.8725,
+      "step": 1945
+    },
+    {
+      "epoch": 1.9549822964087,
+      "grad_norm": 10.375,
+      "learning_rate": 1.3934585134174111e-05,
+      "loss": 0.9031,
+      "step": 1950
+    },
+    {
+      "epoch": 1.9600404653515429,
+      "grad_norm": 9.4375,
+      "learning_rate": 1.3811799117439916e-05,
+      "loss": 0.9882,
+      "step": 1955
+    },
+    {
+      "epoch": 1.9650986342943855,
+      "grad_norm": 10.875,
+      "learning_rate": 1.3689349653261996e-05,
+      "loss": 0.9635,
+      "step": 1960
+    },
+    {
+      "epoch": 1.970156803237228,
+      "grad_norm": 11.0625,
+      "learning_rate": 1.3567240425045807e-05,
+      "loss": 1.0934,
+      "step": 1965
+    },
+    {
+      "epoch": 1.975214972180071,
+      "grad_norm": 11.4375,
+      "learning_rate": 1.3445475105962172e-05,
+      "loss": 0.9353,
+      "step": 1970
+    },
+    {
+      "epoch": 1.9802731411229135,
+      "grad_norm": 10.125,
+      "learning_rate": 1.332405735883676e-05,
+      "loss": 0.9503,
+      "step": 1975
+    },
+    {
+      "epoch": 1.9812847749114821,
+      "eval_loss": 1.235831618309021,
+      "eval_runtime": 256.0612,
+      "eval_samples_per_second": 6.799,
+      "eval_steps_per_second": 6.799,
+      "step": 1976
+    },
+    {
+      "epoch": 1.9853313100657561,
+      "grad_norm": 10.125,
+      "learning_rate": 1.320299083603992e-05,
+      "loss": 0.9683,
+      "step": 1980
+    },
+    {
+      "epoch": 1.990389479008599,
+      "grad_norm": 10.5625,
+      "learning_rate": 1.3082279179376793e-05,
+      "loss": 0.8768,
+      "step": 1985
+    },
+    {
+      "epoch": 1.9954476479514416,
+      "grad_norm": 9.9375,
+      "learning_rate": 1.2961926019977828e-05,
+      "loss": 1.0119,
+      "step": 1990
+    },
+    {
+      "epoch": 2.000505816894284,
+      "grad_norm": 10.0,
+      "learning_rate": 1.2841934978189482e-05,
+      "loss": 0.9268,
+      "step": 1995
+    },
+    {
+      "epoch": 2.005563985837127,
+      "grad_norm": 10.5625,
+      "learning_rate": 1.2722309663465332e-05,
+      "loss": 0.9064,
+      "step": 2000
+    },
+    {
+      "epoch": 2.01062215477997,
+      "grad_norm": 10.0625,
+      "learning_rate": 1.2603053674257531e-05,
+      "loss": 0.9505,
+      "step": 2005
+    },
+    {
+      "epoch": 2.0156803237228123,
+      "grad_norm": 8.375,
+      "learning_rate": 1.2484170597908546e-05,
+      "loss": 0.9074,
+      "step": 2010
+    },
+    {
+      "epoch": 2.0027819929185635,
+      "grad_norm": 8.75,
+      "learning_rate": 1.2365664010543206e-05,
+      "loss": 0.7564,
+      "step": 2015
+    },
+    {
+      "epoch": 2.0078401618614063,
+      "grad_norm": 8.125,
+      "learning_rate": 1.2247537476961212e-05,
+      "loss": 0.6274,
+      "step": 2020
+    },
+    {
+      "epoch": 2.0128983308042487,
+      "grad_norm": 9.3125,
+      "learning_rate": 1.2129794550529822e-05,
+      "loss": 0.721,
+      "step": 2025
+    },
+    {
+      "epoch": 2.0179564997470916,
+      "grad_norm": 8.875,
+      "learning_rate": 1.201243877307702e-05,
+      "loss": 0.6828,
+      "step": 2030
+    },
+    {
+      "epoch": 2.0230146686899344,
+      "grad_norm": 8.875,
+      "learning_rate": 1.189547367478493e-05,
+      "loss": 0.7442,
+      "step": 2035
+    },
+    {
+      "epoch": 2.028072837632777,
+      "grad_norm": 8.875,
+      "learning_rate": 1.1778902774083663e-05,
+      "loss": 0.6343,
+      "step": 2040
+    },
+    {
+      "epoch": 2.0331310065756196,
+      "grad_norm": 9.875,
+      "learning_rate": 1.1662729577545475e-05,
+      "loss": 0.6979,
+      "step": 2045
+    },
+    {
+      "epoch": 2.0381891755184625,
+      "grad_norm": 10.5,
+      "learning_rate": 1.154695757977923e-05,
+      "loss": 0.6676,
+      "step": 2050
+    },
+    {
+      "epoch": 2.043247344461305,
+      "grad_norm": 9.3125,
+      "learning_rate": 1.1431590263325372e-05,
+      "loss": 0.6416,
+      "step": 2055
+    },
+    {
+      "epoch": 2.0483055134041477,
+      "grad_norm": 8.0625,
+      "learning_rate": 1.1316631098551066e-05,
+      "loss": 0.6696,
+      "step": 2060
+    },
+    {
+      "epoch": 2.0533636823469905,
+      "grad_norm": 8.9375,
+      "learning_rate": 1.1202083543545907e-05,
+      "loss": 0.6854,
+      "step": 2065
+    },
+    {
+      "epoch": 2.058421851289833,
+      "grad_norm": 20.625,
+      "learning_rate": 1.1087951044017799e-05,
+      "loss": 0.6977,
+      "step": 2070
+    },
+    {
+      "epoch": 2.0634800202326757,
+      "grad_norm": 8.875,
+      "learning_rate": 1.097423703318938e-05,
+      "loss": 0.677,
+      "step": 2075
+    },
+    {
+      "epoch": 2.0685381891755186,
+      "grad_norm": 9.75,
+      "learning_rate": 1.0860944931694726e-05,
+      "loss": 0.664,
+      "step": 2080
+    },
+    {
+      "epoch": 2.073596358118361,
+      "grad_norm": 9.1875,
+      "learning_rate": 1.074807814747641e-05,
+      "loss": 0.7,
+      "step": 2085
+    },
+    {
+      "epoch": 2.078654527061204,
+      "grad_norm": 9.625,
+      "learning_rate": 1.0635640075683068e-05,
+      "loss": 0.6898,
+      "step": 2090
+    },
+    {
+      "epoch": 2.0837126960040466,
+      "grad_norm": 10.625,
+      "learning_rate": 1.0523634098567193e-05,
+      "loss": 0.6989,
+      "step": 2095
+    },
+    {
+      "epoch": 2.088770864946889,
+      "grad_norm": 8.625,
+      "learning_rate": 1.0412063585383452e-05,
+      "loss": 0.6244,
+      "step": 2100
+    },
+    {
+      "epoch": 2.093829033889732,
+      "grad_norm": 10.125,
+      "learning_rate": 1.0300931892287288e-05,
+      "loss": 0.6122,
+      "step": 2105
+    },
+    {
+      "epoch": 2.0988872028325747,
+      "grad_norm": 9.3125,
+      "learning_rate": 1.0190242362234002e-05,
+      "loss": 0.7064,
+      "step": 2110
+    },
+    {
+      "epoch": 2.103945371775417,
+      "grad_norm": 8.6875,
+      "learning_rate": 1.0079998324878184e-05,
+      "loss": 0.7044,
+      "step": 2115
+    },
+    {
+      "epoch": 2.10900354071826,
+      "grad_norm": 8.5625,
+      "learning_rate": 9.970203096473515e-06,
+      "loss": 0.5713,
+      "step": 2120
+    },
+    {
+      "epoch": 2.1140617096611027,
+      "grad_norm": 9.8125,
+      "learning_rate": 9.860859979773085e-06,
+      "loss": 0.677,
+      "step": 2125
+    },
+    {
+      "epoch": 2.1191198786039456,
+      "grad_norm": 9.375,
+      "learning_rate": 9.751972263929953e-06,
+      "loss": 0.5211,
+      "step": 2130
+    },
+    {
+      "epoch": 2.124178047546788,
+      "grad_norm": 8.25,
+      "learning_rate": 9.643543224398306e-06,
+      "loss": 0.5727,
+      "step": 2135
+    },
+    {
+      "epoch": 2.129236216489631,
+      "grad_norm": 8.0,
+      "learning_rate": 9.535576122834819e-06,
+      "loss": 0.6739,
+      "step": 2140
+    },
+    {
+      "epoch": 2.1342943854324736,
+      "grad_norm": 9.875,
+      "learning_rate": 9.428074207000649e-06,
+      "loss": 0.6485,
+      "step": 2145
+    },
+    {
+      "epoch": 2.139352554375316,
+      "grad_norm": 8.6875,
+      "learning_rate": 9.321040710663664e-06,
+      "loss": 0.5869,
+      "step": 2150
+    },
+    {
+      "epoch": 2.144410723318159,
+      "grad_norm": 8.5625,
+      "learning_rate": 9.21447885350121e-06,
+      "loss": 0.7034,
+      "step": 2155
+    },
+    {
+      "epoch": 2.1494688922610017,
+      "grad_norm": 8.3125,
+      "learning_rate": 9.108391841003222e-06,
+      "loss": 0.7334,
+      "step": 2160
+    },
+    {
+      "epoch": 2.154527061203844,
+      "grad_norm": 9.4375,
+      "learning_rate": 9.002782864375822e-06,
+      "loss": 0.7109,
+      "step": 2165
+    },
+    {
+      "epoch": 2.159585230146687,
+      "grad_norm": 9.4375,
+      "learning_rate": 8.897655100445343e-06,
+      "loss": 0.7259,
+      "step": 2170
+    },
+    {
+      "epoch": 2.1646433990895297,
+      "grad_norm": 9.125,
+      "learning_rate": 8.793011711562737e-06,
+      "loss": 0.5756,
+      "step": 2175
+    },
+    {
+      "epoch": 2.169701568032372,
+      "grad_norm": 7.8125,
+      "learning_rate": 8.688855845508443e-06,
+      "loss": 0.6581,
+      "step": 2180
+    },
+    {
+      "epoch": 2.174759736975215,
+      "grad_norm": 9.875,
+      "learning_rate": 8.585190635397733e-06,
+      "loss": 0.6387,
+      "step": 2185
+    },
+    {
+      "epoch": 2.179817905918058,
+      "grad_norm": 9.25,
+      "learning_rate": 8.482019199586446e-06,
+      "loss": 0.5967,
+      "step": 2190
+    },
+    {
+      "epoch": 2.1848760748609,
+      "grad_norm": 9.6875,
+      "learning_rate": 8.379344641577169e-06,
+      "loss": 0.6694,
+      "step": 2195
+    },
+    {
+      "epoch": 2.189934243803743,
+      "grad_norm": 9.75,
+      "learning_rate": 8.27717004992589e-06,
+      "loss": 0.6873,
+      "step": 2200
+    },
+    {
+      "epoch": 2.194992412746586,
+      "grad_norm": 9.1875,
+      "learning_rate": 8.17549849814912e-06,
+      "loss": 0.7475,
+      "step": 2205
+    },
+    {
+      "epoch": 2.2000505816894282,
+      "grad_norm": 8.8125,
+      "learning_rate": 8.07433304463141e-06,
+      "loss": 0.59,
+      "step": 2210
+    },
+    {
+      "epoch": 2.205108750632271,
+      "grad_norm": 9.4375,
+      "learning_rate": 7.973676732533333e-06,
+      "loss": 0.6716,
+      "step": 2215
+    },
+    {
+      "epoch": 2.210166919575114,
+      "grad_norm": 8.4375,
+      "learning_rate": 7.873532589699988e-06,
+      "loss": 0.6639,
+      "step": 2220
+    },
+    {
+      "epoch": 2.2132018209408195,
+      "eval_loss": 1.2539501190185547,
+      "eval_runtime": 255.9656,
+      "eval_samples_per_second": 6.802,
+      "eval_steps_per_second": 6.802,
+      "step": 2223
+    },
+    {
+      "epoch": 2.2152250885179563,
+      "grad_norm": 10.0625,
+      "learning_rate": 7.773903628569903e-06,
+      "loss": 0.647,
+      "step": 2225
+    },
+    {
+      "epoch": 2.220283257460799,
+      "grad_norm": 10.375,
+      "learning_rate": 7.67479284608438e-06,
+      "loss": 0.6833,
+      "step": 2230
+    },
+    {
+      "epoch": 2.225341426403642,
+      "grad_norm": 10.0625,
+      "learning_rate": 7.5762032235974165e-06,
+      "loss": 0.7367,
+      "step": 2235
+    },
+    {
+      "epoch": 2.2303995953464844,
+      "grad_norm": 9.1875,
+      "learning_rate": 7.478137726785942e-06,
+      "loss": 0.6923,
+      "step": 2240
+    },
+    {
+      "epoch": 2.235457764289327,
+      "grad_norm": 8.9375,
+      "learning_rate": 7.38059930556069e-06,
+      "loss": 0.7006,
+      "step": 2245
+    },
+    {
+      "epoch": 2.24051593323217,
+      "grad_norm": 8.1875,
+      "learning_rate": 7.283590893977379e-06,
+      "loss": 0.5964,
+      "step": 2250
+    },
+    {
+      "epoch": 2.245574102175013,
+      "grad_norm": 9.25,
+      "learning_rate": 7.187115410148523e-06,
+      "loss": 0.6169,
+      "step": 2255
+    },
+    {
+      "epoch": 2.2506322711178552,
+      "grad_norm": 9.5,
+      "learning_rate": 7.091175756155618e-06,
+      "loss": 0.6767,
+      "step": 2260
+    },
+    {
+      "epoch": 2.255690440060698,
+      "grad_norm": 9.4375,
+      "learning_rate": 6.995774817961837e-06,
+      "loss": 0.6116,
+      "step": 2265
+    },
+    {
+      "epoch": 2.260748609003541,
+      "grad_norm": 9.75,
+      "learning_rate": 6.900915465325242e-06,
+      "loss": 0.642,
+      "step": 2270
+    },
+    {
+      "epoch": 2.2658067779463833,
+      "grad_norm": 7.96875,
+      "learning_rate": 6.80660055171243e-06,
+      "loss": 0.6496,
+      "step": 2275
+    },
+    {
+      "epoch": 2.270864946889226,
+      "grad_norm": 9.25,
+      "learning_rate": 6.712832914212741e-06,
+      "loss": 0.7172,
+      "step": 2280
+    },
+    {
+      "epoch": 2.275923115832069,
+      "grad_norm": 9.1875,
+      "learning_rate": 6.619615373452853e-06,
+      "loss": 0.7012,
+      "step": 2285
+    },
+    {
+      "epoch": 2.2809812847749114,
+      "grad_norm": 9.3125,
+      "learning_rate": 6.526950733511997e-06,
+      "loss": 0.6704,
+      "step": 2290
+    },
+    {
+      "epoch": 2.286039453717754,
+      "grad_norm": 10.125,
+      "learning_rate": 6.434841781837572e-06,
+      "loss": 0.6776,
+      "step": 2295
+    },
+    {
+      "epoch": 2.291097622660597,
+      "grad_norm": 8.875,
+      "learning_rate": 6.343291289161316e-06,
+      "loss": 0.6665,
+      "step": 2300
+    },
+    {
+      "epoch": 2.2961557916034394,
+      "grad_norm": 8.875,
+      "learning_rate": 6.25230200941592e-06,
+      "loss": 0.6947,
+      "step": 2305
+    },
+    {
+      "epoch": 2.3012139605462822,
+      "grad_norm": 9.5625,
+      "learning_rate": 6.161876679652229e-06,
+      "loss": 0.6923,
+      "step": 2310
+    },
+    {
+      "epoch": 2.306272129489125,
+      "grad_norm": 7.71875,
+      "learning_rate": 6.0720180199569015e-06,
+      "loss": 0.601,
+      "step": 2315
+    },
+    {
+      "epoch": 2.3113302984319675,
+      "grad_norm": 9.625,
+      "learning_rate": 5.98272873337058e-06,
+      "loss": 0.6752,
+      "step": 2320
+    },
+    {
+      "epoch": 2.3163884673748103,
+      "grad_norm": 9.375,
+      "learning_rate": 5.894011505806557e-06,
+      "loss": 0.651,
+      "step": 2325
+    },
+    {
+      "epoch": 2.321446636317653,
+      "grad_norm": 8.625,
+      "learning_rate": 5.805869005970027e-06,
+      "loss": 0.7164,
+      "step": 2330
+    },
+    {
+      "epoch": 2.3265048052604955,
+      "grad_norm": 10.0625,
+      "learning_rate": 5.718303885277784e-06,
+      "loss": 0.7191,
+      "step": 2335
+    },
+    {
+      "epoch": 2.3315629742033384,
+      "grad_norm": 9.3125,
+      "learning_rate": 5.631318777778452e-06,
+      "loss": 0.695,
+      "step": 2340
+    },
+    {
+      "epoch": 2.336621143146181,
+      "grad_norm": 8.5625,
+      "learning_rate": 5.544916300073264e-06,
+      "loss": 0.6296,
+      "step": 2345
+    },
+    {
+      "epoch": 2.341679312089024,
+      "grad_norm": 9.1875,
+      "learning_rate": 5.459099051237362e-06,
+      "loss": 0.7312,
+      "step": 2350
+    },
+    {
+      "epoch": 2.3467374810318664,
+      "grad_norm": 8.5,
+      "learning_rate": 5.3738696127416026e-06,
+      "loss": 0.5695,
+      "step": 2355
+    },
+    {
+      "epoch": 2.3517956499747092,
+      "grad_norm": 7.96875,
+      "learning_rate": 5.289230548374893e-06,
+      "loss": 0.6656,
+      "step": 2360
+    },
+    {
+      "epoch": 2.3568538189175516,
+      "grad_norm": 9.75,
+      "learning_rate": 5.20518440416709e-06,
+      "loss": 0.6532,
+      "step": 2365
+    },
+    {
+      "epoch": 2.3619119878603945,
+      "grad_norm": 8.375,
+      "learning_rate": 5.12173370831241e-06,
+      "loss": 0.6479,
+      "step": 2370
+    },
+    {
+      "epoch": 2.3669701568032373,
+      "grad_norm": 9.25,
+      "learning_rate": 5.03888097109336e-06,
+      "loss": 0.6562,
+      "step": 2375
+    },
+    {
+      "epoch": 2.37202832574608,
+      "grad_norm": 8.625,
+      "learning_rate": 4.9566286848052365e-06,
+      "loss": 0.5889,
+      "step": 2380
+    },
+    {
+      "epoch": 2.3770864946889225,
+      "grad_norm": 9.3125,
+      "learning_rate": 4.874979323681164e-06,
+      "loss": 0.7583,
+      "step": 2385
+    },
+    {
+      "epoch": 2.3821446636317654,
+      "grad_norm": 9.0625,
+      "learning_rate": 4.793935343817671e-06,
+      "loss": 0.641,
+      "step": 2390
+    },
+    {
+      "epoch": 2.3872028325746077,
+      "grad_norm": 10.0,
+      "learning_rate": 4.71349918310077e-06,
+      "loss": 0.6704,
+      "step": 2395
+    },
+    {
+      "epoch": 2.3922610015174506,
+      "grad_norm": 8.625,
+      "learning_rate": 4.633673261132665e-06,
+      "loss": 0.6579,
+      "step": 2400
+    },
+    {
+      "epoch": 2.3973191704602934,
+      "grad_norm": 7.28125,
+      "learning_rate": 4.554459979158965e-06,
+      "loss": 0.6143,
+      "step": 2405
+    },
+    {
+      "epoch": 2.4023773394031362,
+      "grad_norm": 8.8125,
+      "learning_rate": 4.475861719996419e-06,
+      "loss": 0.654,
+      "step": 2410
+    },
+    {
+      "epoch": 2.4074355083459786,
+      "grad_norm": 9.0625,
+      "learning_rate": 4.397880847961256e-06,
+      "loss": 0.6973,
+      "step": 2415
+    },
+    {
+      "epoch": 2.4124936772888215,
+      "grad_norm": 9.25,
+      "learning_rate": 4.32051970879809e-06,
+      "loss": 0.6216,
+      "step": 2420
+    },
+    {
+      "epoch": 2.4175518462316643,
+      "grad_norm": 9.625,
+      "learning_rate": 4.243780629609326e-06,
+      "loss": 0.6474,
+      "step": 2425
+    },
+    {
+      "epoch": 2.4226100151745067,
+      "grad_norm": 9.9375,
+      "learning_rate": 4.167665918785149e-06,
+      "loss": 0.6677,
+      "step": 2430
+    },
+    {
+      "epoch": 2.4276681841173495,
+      "grad_norm": 10.3125,
+      "learning_rate": 4.09217786593413e-06,
+      "loss": 0.6855,
+      "step": 2435
+    },
+    {
+      "epoch": 2.4327263530601924,
+      "grad_norm": 9.0625,
+      "learning_rate": 4.017318741814305e-06,
+      "loss": 0.6987,
+      "step": 2440
+    },
+    {
+      "epoch": 2.4377845220030347,
+      "grad_norm": 8.875,
+      "learning_rate": 3.9430907982649175e-06,
+      "loss": 0.6322,
+      "step": 2445
+    },
+    {
+      "epoch": 2.4428426909458776,
+      "grad_norm": 8.125,
+      "learning_rate": 3.869496268138603e-06,
+      "loss": 0.6044,
+      "step": 2450
+    },
+    {
+      "epoch": 2.4479008598887204,
+      "grad_norm": 9.125,
+      "learning_rate": 3.796537365234312e-06,
+      "loss": 0.5787,
+      "step": 2455
+    },
+    {
+      "epoch": 2.452959028831563,
+      "grad_norm": 9.1875,
+      "learning_rate": 3.724216284230672e-06,
+      "loss": 0.6162,
+      "step": 2460
+    },
+    {
+      "epoch": 2.4580171977744056,
+      "grad_norm": 9.3125,
+      "learning_rate": 3.6525352006199552e-06,
+      "loss": 0.6025,
+      "step": 2465
+    },
+    {
+      "epoch": 2.4630753667172485,
+      "grad_norm": 9.125,
+      "learning_rate": 3.581496270642673e-06,
+      "loss": 0.6624,
+      "step": 2470
+    },
+    {
+      "epoch": 2.4630753667172485,
+      "eval_loss": 1.2369153499603271,
+      "eval_runtime": 256.1258,
+      "eval_samples_per_second": 6.797,
+      "eval_steps_per_second": 6.797,
+      "step": 2470
+    },
+    {
+      "epoch": 2.468133535660091,
+      "grad_norm": 8.25,
+      "learning_rate": 3.5111016312226907e-06,
+      "loss": 0.6113,
+      "step": 2475
+    },
+    {
+      "epoch": 2.4731917046029337,
+      "grad_norm": 13.4375,
+      "learning_rate": 3.4413533999029577e-06,
+      "loss": 0.7113,
+      "step": 2480
+    },
+    {
+      "epoch": 2.4782498735457765,
+      "grad_norm": 8.625,
+      "learning_rate": 3.3722536747817947e-06,
+      "loss": 0.6582,
+      "step": 2485
+    },
+    {
+      "epoch": 2.483308042488619,
+      "grad_norm": 9.375,
+      "learning_rate": 3.303804534449792e-06,
+      "loss": 0.6877,
+      "step": 2490
+    },
+    {
+      "epoch": 2.4883662114314617,
+      "grad_norm": 8.9375,
+      "learning_rate": 3.2360080379272934e-06,
+      "loss": 0.618,
+      "step": 2495
+    },
+    {
+      "epoch": 2.4934243803743046,
+      "grad_norm": 7.40625,
+      "learning_rate": 3.168866224602443e-06,
+      "loss": 0.5701,
+      "step": 2500
+    },
+    {
+      "epoch": 2.4984825493171474,
+      "grad_norm": 10.5625,
+      "learning_rate": 3.1023811141698335e-06,
+      "loss": 0.6588,
+      "step": 2505
+    },
+    {
+      "epoch": 2.50354071825999,
+      "grad_norm": 7.5625,
+      "learning_rate": 3.0365547065697746e-06,
+      "loss": 0.6959,
+      "step": 2510
+    },
+    {
+      "epoch": 2.5085988872028326,
+      "grad_norm": 8.875,
+      "learning_rate": 2.9713889819281137e-06,
+      "loss": 0.5452,
+      "step": 2515
+    },
+    {
+      "epoch": 2.513657056145675,
+      "grad_norm": 10.375,
+      "learning_rate": 2.9068859004966753e-06,
+      "loss": 0.7035,
+      "step": 2520
+    },
+    {
+      "epoch": 2.518715225088518,
+      "grad_norm": 8.9375,
+      "learning_rate": 2.8430474025942926e-06,
+      "loss": 0.6461,
+      "step": 2525
+    },
+    {
+      "epoch": 2.5237733940313607,
+      "grad_norm": 9.75,
+      "learning_rate": 2.7798754085484547e-06,
+      "loss": 0.6859,
+      "step": 2530
+    },
+    {
+      "epoch": 2.5288315629742035,
+      "grad_norm": 9.1875,
+      "learning_rate": 2.7173718186375317e-06,
+      "loss": 0.6637,
+      "step": 2535
+    },
+    {
+      "epoch": 2.533889731917046,
+      "grad_norm": 9.375,
+      "learning_rate": 2.655538513033598e-06,
+      "loss": 0.718,
+      "step": 2540
+    },
+    {
+      "epoch": 2.5389479008598888,
+      "grad_norm": 9.4375,
+      "learning_rate": 2.5943773517459025e-06,
+      "loss": 0.7801,
+      "step": 2545
+    },
+    {
+      "epoch": 2.544006069802731,
+      "grad_norm": 7.25,
+      "learning_rate": 2.5338901745649012e-06,
+      "loss": 0.6276,
+      "step": 2550
+    },
+    {
+      "epoch": 2.549064238745574,
+      "grad_norm": 9.75,
+      "learning_rate": 2.47407880100691e-06,
+      "loss": 0.6224,
+      "step": 2555
+    },
+    {
+      "epoch": 2.554122407688417,
+      "grad_norm": 8.0625,
+      "learning_rate": 2.4149450302593827e-06,
+      "loss": 0.5454,
+      "step": 2560
+    },
+    {
+      "epoch": 2.5591805766312596,
+      "grad_norm": 7.9375,
+      "learning_rate": 2.356490641126785e-06,
+      "loss": 0.6294,
+      "step": 2565
+    },
+    {
+      "epoch": 2.564238745574102,
+      "grad_norm": 8.1875,
+      "learning_rate": 2.298717391977098e-06,
+      "loss": 0.5638,
+      "step": 2570
+    },
+    {
+      "epoch": 2.569296914516945,
+      "grad_norm": 9.3125,
+      "learning_rate": 2.241627020688891e-06,
+      "loss": 0.5839,
+      "step": 2575
+    },
+    {
+      "epoch": 2.5743550834597877,
+      "grad_norm": 10.125,
+      "learning_rate": 2.1852212445990893e-06,
+      "loss": 0.7468,
+      "step": 2580
+    },
+    {
+      "epoch": 2.57941325240263,
+      "grad_norm": 9.6875,
+      "learning_rate": 2.1295017604512802e-06,
+      "loss": 0.6948,
+      "step": 2585
+    },
+    {
+      "epoch": 2.584471421345473,
+      "grad_norm": 8.6875,
+      "learning_rate": 2.0744702443446893e-06,
+      "loss": 0.5727,
+      "step": 2590
+    },
+    {
+      "epoch": 2.5895295902883158,
+      "grad_norm": 9.0,
+      "learning_rate": 2.0201283516837504e-06,
+      "loss": 0.71,
+      "step": 2595
+    },
+    {
+      "epoch": 2.5945877592311586,
+      "grad_norm": 9.0,
+      "learning_rate": 1.966477717128329e-06,
+      "loss": 0.623,
+      "step": 2600
+    },
+    {
+      "epoch": 2.599645928174001,
+      "grad_norm": 7.90625,
+      "learning_rate": 1.9135199545445377e-06,
+      "loss": 0.6839,
+      "step": 2605
+    },
+    {
+      "epoch": 2.604704097116844,
+      "grad_norm": 10.375,
+      "learning_rate": 1.861256656956173e-06,
+      "loss": 0.685,
+      "step": 2610
+    },
+    {
+      "epoch": 2.609762266059686,
+      "grad_norm": 8.9375,
+      "learning_rate": 1.8096893964968275e-06,
+      "loss": 0.6628,
+      "step": 2615
+    },
+    {
+      "epoch": 2.614820435002529,
+      "grad_norm": 9.0,
+      "learning_rate": 1.7588197243625815e-06,
+      "loss": 0.649,
+      "step": 2620
+    },
+    {
+      "epoch": 2.619878603945372,
+      "grad_norm": 9.3125,
+      "learning_rate": 1.7086491707653473e-06,
+      "loss": 0.6293,
+      "step": 2625
+    },
+    {
+      "epoch": 2.6249367728882147,
+      "grad_norm": 7.5625,
+      "learning_rate": 1.6591792448868132e-06,
+      "loss": 0.6218,
+      "step": 2630
+    },
+    {
+      "epoch": 2.629994941831057,
+      "grad_norm": 8.4375,
+      "learning_rate": 1.6104114348330896e-06,
+      "loss": 0.6793,
+      "step": 2635
+    },
+    {
+      "epoch": 2.6350531107739,
+      "grad_norm": 8.9375,
+      "learning_rate": 1.562347207589915e-06,
+      "loss": 0.7324,
+      "step": 2640
+    },
+    {
+      "epoch": 2.6401112797167423,
+      "grad_norm": 8.3125,
+      "learning_rate": 1.5149880089785408e-06,
+      "loss": 0.5812,
+      "step": 2645
+    },
+    {
+      "epoch": 2.645169448659585,
+      "grad_norm": 7.84375,
+      "learning_rate": 1.4683352636122194e-06,
+      "loss": 0.6439,
+      "step": 2650
+    },
+    {
+      "epoch": 2.650227617602428,
+      "grad_norm": 8.5,
+      "learning_rate": 1.4223903748533774e-06,
+      "loss": 0.6243,
+      "step": 2655
+    },
+    {
+      "epoch": 2.655285786545271,
+      "grad_norm": 9.1875,
+      "learning_rate": 1.3771547247713957e-06,
+      "loss": 0.6556,
+      "step": 2660
+    },
+    {
+      "epoch": 2.660343955488113,
+      "grad_norm": 10.125,
+      "learning_rate": 1.3326296741010074e-06,
+      "loss": 0.6481,
+      "step": 2665
+    },
+    {
+      "epoch": 2.665402124430956,
+      "grad_norm": 9.0625,
+      "learning_rate": 1.2888165622014003e-06,
+      "loss": 0.7203,
+      "step": 2670
+    },
+    {
+      "epoch": 2.6704602933737984,
+      "grad_norm": 8.8125,
+      "learning_rate": 1.2457167070159148e-06,
+      "loss": 0.6773,
+      "step": 2675
+    },
+    {
+      "epoch": 2.6755184623166413,
+      "grad_norm": 9.75,
+      "learning_rate": 1.203331405032393e-06,
+      "loss": 0.6175,
+      "step": 2680
+    },
+    {
+      "epoch": 2.680576631259484,
+      "grad_norm": 8.75,
+      "learning_rate": 1.1616619312441845e-06,
+      "loss": 0.6552,
+      "step": 2685
+    },
+    {
+      "epoch": 2.685634800202327,
+      "grad_norm": 8.125,
+      "learning_rate": 1.1207095391117945e-06,
+      "loss": 0.6226,
+      "step": 2690
+    },
+    {
+      "epoch": 2.6906929691451693,
+      "grad_norm": 8.375,
+      "learning_rate": 1.0804754605251793e-06,
+      "loss": 0.648,
+      "step": 2695
+    },
+    {
+      "epoch": 2.695751138088012,
+      "grad_norm": 8.0625,
+      "learning_rate": 1.040960905766683e-06,
+      "loss": 0.7131,
+      "step": 2700
+    },
+    {
+      "epoch": 2.700809307030855,
+      "grad_norm": 8.1875,
+      "learning_rate": 1.0021670634746266e-06,
+      "loss": 0.6678,
+      "step": 2705
+    },
+    {
+      "epoch": 2.7058674759736974,
+      "grad_norm": 7.9375,
+      "learning_rate": 9.640951006075816e-07,
+      "loss": 0.7006,
+      "step": 2710
+    },
+    {
+      "epoch": 2.71092564491654,
+      "grad_norm": 8.875,
+      "learning_rate": 9.267461624092283e-07,
+      "loss": 0.5855,
+      "step": 2715
+    },
+    {
+      "epoch": 2.7129489124936774,
+      "eval_loss": 1.229272484779358,
+      "eval_runtime": 255.986,
+      "eval_samples_per_second": 6.801,
+      "eval_steps_per_second": 6.801,
+      "step": 2717
+    },
+    {
+      "epoch": 2.715983813859383,
+      "grad_norm": 8.8125,
+      "learning_rate": 8.90121372373931e-07,
+      "loss": 0.7859,
+      "step": 2720
+    },
+    {
+      "epoch": 2.721041982802226,
+      "grad_norm": 8.1875,
+      "learning_rate": 8.542218322129341e-07,
+      "loss": 0.7668,
+      "step": 2725
+    },
+    {
+      "epoch": 2.7261001517450683,
+      "grad_norm": 8.625,
+      "learning_rate": 8.190486218212224e-07,
+      "loss": 0.6654,
+      "step": 2730
+    },
+    {
+      "epoch": 2.731158320687911,
+      "grad_norm": 7.84375,
+      "learning_rate": 7.846027992450378e-07,
+      "loss": 0.5874,
+      "step": 2735
+    },
+    {
+      "epoch": 2.7362164896307535,
+      "grad_norm": 8.5625,
+      "learning_rate": 7.508854006500393e-07,
+      "loss": 0.692,
+      "step": 2740
+    },
+    {
+      "epoch": 2.7412746585735963,
+      "grad_norm": 10.3125,
+      "learning_rate": 7.178974402901578e-07,
+      "loss": 0.6855,
+      "step": 2745
+    },
+    {
+      "epoch": 2.746332827516439,
+      "grad_norm": 7.8125,
+      "learning_rate": 6.856399104770789e-07,
+      "loss": 0.6668,
+      "step": 2750
+    },
+    {
+      "epoch": 2.751390996459282,
+      "grad_norm": 8.875,
+      "learning_rate": 6.541137815503695e-07,
+      "loss": 0.6741,
+      "step": 2755
+    },
+    {
+      "epoch": 2.7564491654021244,
+      "grad_norm": 8.625,
+      "learning_rate": 6.233200018483238e-07,
+      "loss": 0.6499,
+      "step": 2760
+    },
+    {
+      "epoch": 2.761507334344967,
+      "grad_norm": 7.46875,
+      "learning_rate": 5.93259497679416e-07,
+      "loss": 0.6286,
+      "step": 2765
+    },
+    {
+      "epoch": 2.7665655032878096,
+      "grad_norm": 7.8125,
+      "learning_rate": 5.639331732944397e-07,
+      "loss": 0.579,
+      "step": 2770
+    },
+    {
+      "epoch": 2.7716236722306524,
+      "grad_norm": 9.8125,
+      "learning_rate": 5.353419108592994e-07,
+      "loss": 0.695,
+      "step": 2775
+    },
+    {
+      "epoch": 2.7766818411734953,
+      "grad_norm": 10.375,
+      "learning_rate": 5.074865704284976e-07,
+      "loss": 0.6512,
+      "step": 2780
+    },
+    {
+      "epoch": 2.781740010116338,
+      "grad_norm": 9.9375,
+      "learning_rate": 4.803679899192392e-07,
+      "loss": 0.9856,
+      "step": 2785
+    },
+    {
+      "epoch": 2.7867981790591805,
+      "grad_norm": 7.28125,
+      "learning_rate": 4.539869850862355e-07,
+      "loss": 0.5678,
+      "step": 2790
+    },
+    {
+      "epoch": 2.7918563480020233,
+      "grad_norm": 10.75,
+      "learning_rate": 4.283443494971645e-07,
+      "loss": 0.6918,
+      "step": 2795
+    },
+    {
+      "epoch": 2.7969145169448657,
+      "grad_norm": 7.78125,
+      "learning_rate": 4.0344085450880755e-07,
+      "loss": 0.6091,
+      "step": 2800
+    },
+    {
+      "epoch": 2.8019726858877085,
+      "grad_norm": 10.1875,
+      "learning_rate": 3.7927724924382846e-07,
+      "loss": 0.7453,
+      "step": 2805
+    },
+    {
+      "epoch": 2.8070308548305514,
+      "grad_norm": 7.34375,
+      "learning_rate": 3.55854260568253e-07,
+      "loss": 0.6616,
+      "step": 2810
+    },
+    {
+      "epoch": 2.812089023773394,
+      "grad_norm": 7.875,
+      "learning_rate": 3.331725930696028e-07,
+      "loss": 0.6668,
+      "step": 2815
+    },
+    {
+      "epoch": 2.8171471927162366,
+      "grad_norm": 9.75,
+      "learning_rate": 3.1123292903569857e-07,
+      "loss": 0.6568,
+      "step": 2820
+    },
+    {
+      "epoch": 2.8222053616590794,
+      "grad_norm": 8.9375,
+      "learning_rate": 2.90035928434132e-07,
+      "loss": 0.6387,
+      "step": 2825
+    },
+    {
+      "epoch": 2.8272635306019223,
+      "grad_norm": 9.1875,
+      "learning_rate": 2.695822288924177e-07,
+      "loss": 0.6177,
+      "step": 2830
+    },
+    {
+      "epoch": 2.8323216995447646,
+      "grad_norm": 8.4375,
+      "learning_rate": 2.4987244567881705e-07,
+      "loss": 0.6505,
+      "step": 2835
+    },
+    {
+      "epoch": 2.8373798684876075,
+      "grad_norm": 8.25,
+      "learning_rate": 2.309071716838196e-07,
+      "loss": 0.6351,
+      "step": 2840
+    },
+    {
+      "epoch": 2.8424380374304503,
+      "grad_norm": 7.9375,
+      "learning_rate": 2.1268697740231826e-07,
+      "loss": 0.6083,
+      "step": 2845
+    },
+    {
+      "epoch": 2.8474962063732927,
+      "grad_norm": 9.6875,
+      "learning_rate": 1.9521241091643993e-07,
+      "loss": 0.6506,
+      "step": 2850
+    },
+    {
+      "epoch": 2.8525543753161355,
+      "grad_norm": 9.625,
+      "learning_rate": 1.7848399787906423e-07,
+      "loss": 0.6963,
+      "step": 2855
+    },
+    {
+      "epoch": 2.8576125442589784,
+      "grad_norm": 7.46875,
+      "learning_rate": 1.6250224149801652e-07,
+      "loss": 0.5868,
+      "step": 2860
+    },
+    {
+      "epoch": 2.8626707132018208,
+      "grad_norm": 9.1875,
+      "learning_rate": 1.4726762252091075e-07,
+      "loss": 0.6716,
+      "step": 2865
+    },
+    {
+      "epoch": 2.8677288821446636,
+      "grad_norm": 8.75,
+      "learning_rate": 1.3278059922071374e-07,
+      "loss": 0.6319,
+      "step": 2870
+    },
+    {
+      "epoch": 2.8727870510875064,
+      "grad_norm": 9.125,
+      "learning_rate": 1.1904160738194225e-07,
+      "loss": 0.646,
+      "step": 2875
+    },
+    {
+      "epoch": 2.8778452200303493,
+      "grad_norm": 9.5625,
+      "learning_rate": 1.060510602875625e-07,
+      "loss": 0.6335,
+      "step": 2880
+    },
+    {
+      "epoch": 2.8829033889731916,
+      "grad_norm": 9.0,
+      "learning_rate": 9.380934870654723e-08,
+      "loss": 0.6589,
+      "step": 2885
+    },
+    {
+      "epoch": 2.8879615579160345,
+      "grad_norm": 8.0625,
+      "learning_rate": 8.231684088214065e-08,
+      "loss": 0.5949,
+      "step": 2890
+    },
+    {
+      "epoch": 2.893019726858877,
+      "grad_norm": 8.8125,
+      "learning_rate": 7.157388252076181e-08,
+      "loss": 0.6551,
+      "step": 2895
+    },
+    {
+      "epoch": 2.8980778958017197,
+      "grad_norm": 7.71875,
+      "learning_rate": 6.158079678161565e-08,
+      "loss": 0.6209,
+      "step": 2900
+    },
+    {
+      "epoch": 2.9031360647445625,
+      "grad_norm": 8.1875,
+      "learning_rate": 5.2337884266970284e-08,
+      "loss": 0.5721,
+      "step": 2905
+    },
+    {
+      "epoch": 2.9081942336874054,
+      "grad_norm": 8.3125,
+      "learning_rate": 4.384542301311134e-08,
+      "loss": 0.6374,
+      "step": 2910
+    },
+    {
+      "epoch": 2.9132524026302478,
+      "grad_norm": 8.25,
+      "learning_rate": 3.61036684819821e-08,
+      "loss": 0.6139,
+      "step": 2915
+    },
+    {
+      "epoch": 2.9183105715730906,
+      "grad_norm": 9.0625,
+      "learning_rate": 2.9112853553497908e-08,
+      "loss": 0.674,
+      "step": 2920
+    },
+    {
+      "epoch": 2.923368740515933,
+      "grad_norm": 9.125,
+      "learning_rate": 2.2873188518543475e-08,
+      "loss": 0.6185,
+      "step": 2925
+    },
+    {
+      "epoch": 2.928426909458776,
+      "grad_norm": 8.1875,
+      "learning_rate": 1.7384861072641812e-08,
+      "loss": 0.5905,
+      "step": 2930
+    },
+    {
+      "epoch": 2.9334850784016187,
+      "grad_norm": 8.25,
+      "learning_rate": 1.2648036310308753e-08,
+      "loss": 0.5746,
+      "step": 2935
+    },
+    {
+      "epoch": 2.9385432473444615,
+      "grad_norm": 9.0625,
+      "learning_rate": 8.662856720090262e-09,
+      "loss": 0.6408,
+      "step": 2940
+    },
+    {
+      "epoch": 2.943601416287304,
+      "grad_norm": 8.5,
+      "learning_rate": 5.4294421802797426e-09,
+      "loss": 0.6383,
+      "step": 2945
+    },
+    {
+      "epoch": 2.9486595852301467,
+      "grad_norm": 7.8125,
+      "learning_rate": 2.9478899552987104e-09,
+      "loss": 0.631,
+      "step": 2950
+    },
+    {
+      "epoch": 2.9537177541729895,
+      "grad_norm": 10.1875,
+      "learning_rate": 1.2182746927824619e-09,
+      "loss": 0.7067,
+      "step": 2955
+    },
+    {
+      "epoch": 2.958775923115832,
+      "grad_norm": 8.0625,
+      "learning_rate": 2.406484213346438e-10,
+      "loss": 0.6411,
+      "step": 2960
+    },
+    {
+      "epoch": 2.9628224582701064,
+      "eval_loss": 1.229087233543396,
+      "eval_runtime": 256.1346,
+      "eval_samples_per_second": 6.797,
+      "eval_steps_per_second": 6.797,
+      "step": 2964
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 2964,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 494,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 8.385684104376484e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}