diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5030 @@
+{
+  "best_metric": 0.9616935483870968,
+  "best_model_checkpoint": "YAHIA/vivit-b-16x2-mixed-dataset\\checkpoint-7044",
+  "epoch": 5.166666666666667,
+  "eval_steps": 500,
+  "global_step": 7044,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 17.826011657714844,
+      "learning_rate": 7.092198581560284e-07,
+      "loss": 1.5047,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 21.9292049407959,
+      "learning_rate": 1.4184397163120568e-06,
+      "loss": 1.8599,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 27.537826538085938,
+      "learning_rate": 2.1276595744680853e-06,
+      "loss": 1.7293,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 21.80863380432129,
+      "learning_rate": 2.8368794326241136e-06,
+      "loss": 1.7558,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 21.152454376220703,
+      "learning_rate": 3.5460992907801423e-06,
+      "loss": 1.7329,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 20.616695404052734,
+      "learning_rate": 4.255319148936171e-06,
+      "loss": 1.8497,
+      "step": 60
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 22.662025451660156,
+      "learning_rate": 4.964539007092199e-06,
+      "loss": 1.6038,
+      "step": 70
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 21.61168098449707,
+      "learning_rate": 5.673758865248227e-06,
+      "loss": 1.6562,
+      "step": 80
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 16.91636848449707,
+      "learning_rate": 6.3829787234042555e-06,
+      "loss": 1.566,
+      "step": 90
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 22.01127052307129,
+      "learning_rate": 7.092198581560285e-06,
+      "loss": 1.5511,
+      "step": 100
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 21.30156898498535,
+      "learning_rate": 7.801418439716313e-06,
+      "loss": 1.5119,
+      "step": 110
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 22.595388412475586,
+      "learning_rate": 8.510638297872341e-06,
+      "loss": 1.6976,
+      "step": 120
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 16.5057315826416,
+      "learning_rate": 9.219858156028368e-06,
+      "loss": 1.6186,
+      "step": 130
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 23.195261001586914,
+      "learning_rate": 9.929078014184398e-06,
+      "loss": 1.7527,
+      "step": 140
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 17.528949737548828,
+      "learning_rate": 1.0638297872340426e-05,
+      "loss": 1.4839,
+      "step": 150
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 15.22777271270752,
+      "learning_rate": 1.1347517730496454e-05,
+      "loss": 1.4998,
+      "step": 160
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 15.931229591369629,
+      "learning_rate": 1.2056737588652483e-05,
+      "loss": 1.5211,
+      "step": 170
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 23.619646072387695,
+      "learning_rate": 1.2765957446808511e-05,
+      "loss": 1.4134,
+      "step": 180
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 13.800497055053711,
+      "learning_rate": 1.347517730496454e-05,
+      "loss": 1.322,
+      "step": 190
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 17.691619873046875,
+      "learning_rate": 1.418439716312057e-05,
+      "loss": 1.3451,
+      "step": 200
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 23.416200637817383,
+      "learning_rate": 1.4893617021276596e-05,
+      "loss": 1.3731,
+      "step": 210
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 28.64614486694336,
+      "learning_rate": 1.5602836879432626e-05,
+      "loss": 1.286,
+      "step": 220
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 17.008581161499023,
+      "learning_rate": 1.6312056737588656e-05,
+      "loss": 0.971,
+      "step": 230
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 18.605478286743164,
+      "learning_rate": 1.7021276595744682e-05,
+      "loss": 0.995,
+      "step": 240
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 13.493790626525879,
+      "learning_rate": 1.773049645390071e-05,
+      "loss": 1.0098,
+      "step": 250
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 8.223065376281738,
+      "learning_rate": 1.8439716312056736e-05,
+      "loss": 0.8352,
+      "step": 260
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 23.46603775024414,
+      "learning_rate": 1.9148936170212766e-05,
+      "loss": 0.934,
+      "step": 270
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 10.902445793151855,
+      "learning_rate": 1.9858156028368796e-05,
+      "loss": 1.1034,
+      "step": 280
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 31.208297729492188,
+      "learning_rate": 2.0567375886524822e-05,
+      "loss": 0.9828,
+      "step": 290
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 18.870407104492188,
+      "learning_rate": 2.1276595744680852e-05,
+      "loss": 0.9068,
+      "step": 300
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 24.936309814453125,
+      "learning_rate": 2.1985815602836882e-05,
+      "loss": 1.2959,
+      "step": 310
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 17.00774574279785,
+      "learning_rate": 2.269503546099291e-05,
+      "loss": 0.8883,
+      "step": 320
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 18.303964614868164,
+      "learning_rate": 2.340425531914894e-05,
+      "loss": 0.8418,
+      "step": 330
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 14.832561492919922,
+      "learning_rate": 2.4113475177304965e-05,
+      "loss": 0.6663,
+      "step": 340
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 8.988161087036133,
+      "learning_rate": 2.4822695035460995e-05,
+      "loss": 0.7949,
+      "step": 350
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 18.644752502441406,
+      "learning_rate": 2.5531914893617022e-05,
+      "loss": 1.2958,
+      "step": 360
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 22.70672607421875,
+      "learning_rate": 2.624113475177305e-05,
+      "loss": 0.9979,
+      "step": 370
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 26.379438400268555,
+      "learning_rate": 2.695035460992908e-05,
+      "loss": 1.0713,
+      "step": 380
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 21.667213439941406,
+      "learning_rate": 2.765957446808511e-05,
+      "loss": 0.8801,
+      "step": 390
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 28.279443740844727,
+      "learning_rate": 2.836879432624114e-05,
+      "loss": 1.1175,
+      "step": 400
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 7.201148986816406,
+      "learning_rate": 2.9078014184397162e-05,
+      "loss": 1.147,
+      "step": 410
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 13.392162322998047,
+      "learning_rate": 2.9787234042553192e-05,
+      "loss": 1.0524,
+      "step": 420
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 23.281951904296875,
+      "learning_rate": 3.0496453900709222e-05,
+      "loss": 0.7554,
+      "step": 430
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 33.47028732299805,
+      "learning_rate": 3.120567375886525e-05,
+      "loss": 1.2212,
+      "step": 440
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 16.247968673706055,
+      "learning_rate": 3.191489361702128e-05,
+      "loss": 1.0956,
+      "step": 450
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 7.920285701751709,
+      "learning_rate": 3.262411347517731e-05,
+      "loss": 0.7296,
+      "step": 460
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 28.956188201904297,
+      "learning_rate": 3.3333333333333335e-05,
+      "loss": 1.2698,
+      "step": 470
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 15.583264350891113,
+      "learning_rate": 3.4042553191489365e-05,
+      "loss": 0.7685,
+      "step": 480
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 33.40299606323242,
+      "learning_rate": 3.4751773049645395e-05,
+      "loss": 0.9291,
+      "step": 490
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 7.630702495574951,
+      "learning_rate": 3.546099290780142e-05,
+      "loss": 0.6593,
+      "step": 500
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 21.33745002746582,
+      "learning_rate": 3.617021276595745e-05,
+      "loss": 0.8438,
+      "step": 510
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 10.764616966247559,
+      "learning_rate": 3.687943262411347e-05,
+      "loss": 0.7591,
+      "step": 520
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 15.835653305053711,
+      "learning_rate": 3.75886524822695e-05,
+      "loss": 0.4482,
+      "step": 530
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 10.10210132598877,
+      "learning_rate": 3.829787234042553e-05,
+      "loss": 0.6634,
+      "step": 540
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 18.194259643554688,
+      "learning_rate": 3.900709219858156e-05,
+      "loss": 0.8918,
+      "step": 550
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.5565059781074524,
+      "learning_rate": 3.971631205673759e-05,
+      "loss": 0.6003,
+      "step": 560
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 32.85580825805664,
+      "learning_rate": 4.0425531914893614e-05,
+      "loss": 1.0914,
+      "step": 570
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 25.506961822509766,
+      "learning_rate": 4.1134751773049644e-05,
+      "loss": 0.9405,
+      "step": 580
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 20.16202735900879,
+      "learning_rate": 4.1843971631205674e-05,
+      "loss": 0.7998,
+      "step": 590
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.6806721687316895,
+      "learning_rate": 4.2553191489361704e-05,
+      "loss": 1.0418,
+      "step": 600
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 21.450342178344727,
+      "learning_rate": 4.3262411347517734e-05,
+      "loss": 0.9131,
+      "step": 610
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.693479061126709,
+      "learning_rate": 4.3971631205673764e-05,
+      "loss": 0.6994,
+      "step": 620
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 29.120101928710938,
+      "learning_rate": 4.468085106382979e-05,
+      "loss": 0.8137,
+      "step": 630
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 7.881028175354004,
+      "learning_rate": 4.539007092198582e-05,
+      "loss": 0.5136,
+      "step": 640
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 13.252199172973633,
+      "learning_rate": 4.609929078014185e-05,
+      "loss": 1.004,
+      "step": 650
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 27.444059371948242,
+      "learning_rate": 4.680851063829788e-05,
+      "loss": 0.8117,
+      "step": 660
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.8997368812561035,
+      "learning_rate": 4.751773049645391e-05,
+      "loss": 0.378,
+      "step": 670
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 27.491910934448242,
+      "learning_rate": 4.822695035460993e-05,
+      "loss": 0.526,
+      "step": 680
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.568950653076172,
+      "learning_rate": 4.893617021276596e-05,
+      "loss": 0.7545,
+      "step": 690
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 42.4417724609375,
+      "learning_rate": 4.964539007092199e-05,
+      "loss": 0.7227,
+      "step": 700
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 27.481216430664062,
+      "learning_rate": 4.9960561602776466e-05,
+      "loss": 0.9159,
+      "step": 710
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.950145959854126,
+      "learning_rate": 4.988168480832939e-05,
+      "loss": 0.3451,
+      "step": 720
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 21.8098201751709,
+      "learning_rate": 4.980280801388232e-05,
+      "loss": 0.5815,
+      "step": 730
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 23.565208435058594,
+      "learning_rate": 4.972393121943525e-05,
+      "loss": 0.8437,
+      "step": 740
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.0152504444122314,
+      "learning_rate": 4.964505442498817e-05,
+      "loss": 0.4328,
+      "step": 750
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.38031005859375,
+      "learning_rate": 4.95661776305411e-05,
+      "loss": 0.833,
+      "step": 760
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.787426948547363,
+      "learning_rate": 4.948730083609403e-05,
+      "loss": 1.2013,
+      "step": 770
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 19.72786521911621,
+      "learning_rate": 4.940842404164695e-05,
+      "loss": 0.982,
+      "step": 780
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 14.003949165344238,
+      "learning_rate": 4.9329547247199875e-05,
+      "loss": 0.5557,
+      "step": 790
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 38.31557083129883,
+      "learning_rate": 4.92506704527528e-05,
+      "loss": 0.7053,
+      "step": 800
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 35.58180618286133,
+      "learning_rate": 4.917179365830573e-05,
+      "loss": 1.1468,
+      "step": 810
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.2903843820095062,
+      "learning_rate": 4.9092916863858656e-05,
+      "loss": 0.6123,
+      "step": 820
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 24.000940322875977,
+      "learning_rate": 4.901404006941158e-05,
+      "loss": 0.7748,
+      "step": 830
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.914603590965271,
+      "learning_rate": 4.893516327496451e-05,
+      "loss": 0.2499,
+      "step": 840
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 37.609840393066406,
+      "learning_rate": 4.885628648051743e-05,
+      "loss": 0.7046,
+      "step": 850
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 28.55645751953125,
+      "learning_rate": 4.877740968607036e-05,
+      "loss": 0.7313,
+      "step": 860
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.9080527424812317,
+      "learning_rate": 4.869853289162329e-05,
+      "loss": 0.6403,
+      "step": 870
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 32.73764419555664,
+      "learning_rate": 4.861965609717621e-05,
+      "loss": 1.1874,
+      "step": 880
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 19.233123779296875,
+      "learning_rate": 4.854077930272914e-05,
+      "loss": 0.539,
+      "step": 890
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.6417540311813354,
+      "learning_rate": 4.8461902508282066e-05,
+      "loss": 1.0256,
+      "step": 900
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 23.07563018798828,
+      "learning_rate": 4.838302571383499e-05,
+      "loss": 0.3718,
+      "step": 910
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 36.97256851196289,
+      "learning_rate": 4.830414891938792e-05,
+      "loss": 0.5361,
+      "step": 920
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 25.37937355041504,
+      "learning_rate": 4.822527212494085e-05,
+      "loss": 0.8879,
+      "step": 930
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 29.83109474182129,
+      "learning_rate": 4.8146395330493774e-05,
+      "loss": 1.2576,
+      "step": 940
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.070791244506836,
+      "learning_rate": 4.8067518536046694e-05,
+      "loss": 0.3345,
+      "step": 950
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.04020560905337334,
+      "learning_rate": 4.798864174159962e-05,
+      "loss": 0.6677,
+      "step": 960
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 40.472267150878906,
+      "learning_rate": 4.7909764947152555e-05,
+      "loss": 0.5822,
+      "step": 970
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 14.475014686584473,
+      "learning_rate": 4.7830888152705475e-05,
+      "loss": 0.6471,
+      "step": 980
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.560985803604126,
+      "learning_rate": 4.77520113582584e-05,
+      "loss": 0.307,
+      "step": 990
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 24.90871238708496,
+      "learning_rate": 4.767313456381133e-05,
+      "loss": 1.3033,
+      "step": 1000
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.05692366883158684,
+      "learning_rate": 4.7594257769364256e-05,
+      "loss": 0.9878,
+      "step": 1010
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 29.222652435302734,
+      "learning_rate": 4.751538097491718e-05,
+      "loss": 0.3651,
+      "step": 1020
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 50.54731369018555,
+      "learning_rate": 4.743650418047011e-05,
+      "loss": 0.6615,
+      "step": 1030
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.4181761145591736,
+      "learning_rate": 4.735762738602304e-05,
+      "loss": 0.8965,
+      "step": 1040
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 20.379026412963867,
+      "learning_rate": 4.727875059157596e-05,
+      "loss": 0.4088,
+      "step": 1050
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.3473999798297882,
+      "learning_rate": 4.7199873797128884e-05,
+      "loss": 0.665,
+      "step": 1060
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 31.2557315826416,
+      "learning_rate": 4.712099700268182e-05,
+      "loss": 0.2745,
+      "step": 1070
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.036079246550798416,
+      "learning_rate": 4.704212020823474e-05,
+      "loss": 1.0468,
+      "step": 1080
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 32.4503288269043,
+      "learning_rate": 4.6963243413787665e-05,
+      "loss": 1.1,
+      "step": 1090
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 21.370407104492188,
+      "learning_rate": 4.688436661934059e-05,
+      "loss": 0.7678,
+      "step": 1100
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 17.9803409576416,
+      "learning_rate": 4.680548982489352e-05,
+      "loss": 0.7589,
+      "step": 1110
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.019895076751709,
+      "learning_rate": 4.672661303044644e-05,
+      "loss": 0.9177,
+      "step": 1120
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.14212968945503235,
+      "learning_rate": 4.6647736235999374e-05,
+      "loss": 0.2,
+      "step": 1130
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.2734150886535645,
+      "learning_rate": 4.65688594415523e-05,
+      "loss": 0.9089,
+      "step": 1140
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.033594999462366104,
+      "learning_rate": 4.648998264710522e-05,
+      "loss": 0.5866,
+      "step": 1150
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.04664573445916176,
+      "learning_rate": 4.641110585265815e-05,
+      "loss": 0.3159,
+      "step": 1160
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.028405319899320602,
+      "learning_rate": 4.6332229058211075e-05,
+      "loss": 0.2111,
+      "step": 1170
+    },
+    {
+      "epoch": 0.17,
+      "eval_accuracy": 0.7862903225806451,
+      "eval_loss": 0.8878931999206543,
+      "eval_runtime": 1585.7551,
+      "eval_samples_per_second": 0.313,
+      "eval_steps_per_second": 0.156,
+      "step": 1174
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.055226802825927734,
+      "learning_rate": 4.6253352263764e-05,
+      "loss": 0.5272,
+      "step": 1180
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 17.787506103515625,
+      "learning_rate": 4.617447546931693e-05,
+      "loss": 0.0974,
+      "step": 1190
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.16207976639270782,
+      "learning_rate": 4.6095598674869856e-05,
+      "loss": 0.0117,
+      "step": 1200
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.8650420904159546,
+      "learning_rate": 4.601672188042278e-05,
+      "loss": 0.0401,
+      "step": 1210
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 37.10027313232422,
+      "learning_rate": 4.59378450859757e-05,
+      "loss": 0.7591,
+      "step": 1220
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 36.38993835449219,
+      "learning_rate": 4.585896829152864e-05,
+      "loss": 0.9267,
+      "step": 1230
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.3329768478870392,
+      "learning_rate": 4.5780091497081564e-05,
+      "loss": 0.7411,
+      "step": 1240
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.028457719832658768,
+      "learning_rate": 4.5701214702634484e-05,
+      "loss": 0.9231,
+      "step": 1250
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 4.023280143737793,
+      "learning_rate": 4.562233790818741e-05,
+      "loss": 0.0775,
+      "step": 1260
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 40.81041717529297,
+      "learning_rate": 4.554346111374034e-05,
+      "loss": 0.7082,
+      "step": 1270
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.12308614701032639,
+      "learning_rate": 4.5464584319293265e-05,
+      "loss": 0.3318,
+      "step": 1280
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.19766148924827576,
+      "learning_rate": 4.538570752484619e-05,
+      "loss": 0.6948,
+      "step": 1290
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.09755626320838928,
+      "learning_rate": 4.530683073039912e-05,
+      "loss": 0.6711,
+      "step": 1300
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 1.6713837385177612,
+      "learning_rate": 4.5227953935952046e-05,
+      "loss": 0.2754,
+      "step": 1310
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 3.5718274116516113,
+      "learning_rate": 4.5149077141504967e-05,
+      "loss": 0.4493,
+      "step": 1320
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 33.49106216430664,
+      "learning_rate": 4.50702003470579e-05,
+      "loss": 0.3082,
+      "step": 1330
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 2.482903003692627,
+      "learning_rate": 4.499132355261083e-05,
+      "loss": 0.2746,
+      "step": 1340
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.7113821506500244,
+      "learning_rate": 4.491244675816375e-05,
+      "loss": 0.2867,
+      "step": 1350
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.017704015597701073,
+      "learning_rate": 4.4833569963716675e-05,
+      "loss": 0.9497,
+      "step": 1360
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 45.62956237792969,
+      "learning_rate": 4.47546931692696e-05,
+      "loss": 0.4519,
+      "step": 1370
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 5.461789131164551,
+      "learning_rate": 4.467581637482253e-05,
+      "loss": 0.4275,
+      "step": 1380
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 7.486189365386963,
+      "learning_rate": 4.4596939580375456e-05,
+      "loss": 0.6762,
+      "step": 1390
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.06275622546672821,
+      "learning_rate": 4.451806278592838e-05,
+      "loss": 0.8083,
+      "step": 1400
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 42.63979721069336,
+      "learning_rate": 4.443918599148131e-05,
+      "loss": 0.8268,
+      "step": 1410
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 43.551658630371094,
+      "learning_rate": 4.436030919703423e-05,
+      "loss": 0.5091,
+      "step": 1420
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 7.80751371383667,
+      "learning_rate": 4.428143240258716e-05,
+      "loss": 0.4187,
+      "step": 1430
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 51.5152587890625,
+      "learning_rate": 4.420255560814009e-05,
+      "loss": 0.7459,
+      "step": 1440
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.11915197223424911,
+      "learning_rate": 4.412367881369302e-05,
+      "loss": 1.2659,
+      "step": 1450
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 38.34433364868164,
+      "learning_rate": 4.404480201924594e-05,
+      "loss": 0.5682,
+      "step": 1460
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.5860255360603333,
+      "learning_rate": 4.3965925224798865e-05,
+      "loss": 0.2477,
+      "step": 1470
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 17.89107894897461,
+      "learning_rate": 4.388704843035179e-05,
+      "loss": 0.6877,
+      "step": 1480
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.05183021351695061,
+      "learning_rate": 4.380817163590472e-05,
+      "loss": 0.2217,
+      "step": 1490
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.036865703761577606,
+      "learning_rate": 4.3729294841457646e-05,
+      "loss": 0.7752,
+      "step": 1500
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.7843734622001648,
+      "learning_rate": 4.365041804701057e-05,
+      "loss": 0.12,
+      "step": 1510
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 4.3226237297058105,
+      "learning_rate": 4.35715412525635e-05,
+      "loss": 0.294,
+      "step": 1520
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 22.04405403137207,
+      "learning_rate": 4.349266445811642e-05,
+      "loss": 0.984,
+      "step": 1530
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.09485697746276855,
+      "learning_rate": 4.3413787663669354e-05,
+      "loss": 0.7627,
+      "step": 1540
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 5.8401689529418945,
+      "learning_rate": 4.333491086922228e-05,
+      "loss": 0.3515,
+      "step": 1550
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 58.50148391723633,
+      "learning_rate": 4.32560340747752e-05,
+      "loss": 0.6762,
+      "step": 1560
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.020496351644396782,
+      "learning_rate": 4.317715728032813e-05,
+      "loss": 0.2281,
+      "step": 1570
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 25.28636360168457,
+      "learning_rate": 4.3098280485881055e-05,
+      "loss": 0.8747,
+      "step": 1580
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.021903250366449356,
+      "learning_rate": 4.301940369143398e-05,
+      "loss": 0.4165,
+      "step": 1590
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 28.756746292114258,
+      "learning_rate": 4.294052689698691e-05,
+      "loss": 0.5279,
+      "step": 1600
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.8276819586753845,
+      "learning_rate": 4.2861650102539837e-05,
+      "loss": 0.3545,
+      "step": 1610
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.025062019005417824,
+      "learning_rate": 4.2782773308092764e-05,
+      "loss": 0.4562,
+      "step": 1620
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 3.913560628890991,
+      "learning_rate": 4.2703896513645684e-05,
+      "loss": 0.1353,
+      "step": 1630
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 1.3891974687576294,
+      "learning_rate": 4.262501971919861e-05,
+      "loss": 1.095,
+      "step": 1640
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.1321750432252884,
+      "learning_rate": 4.2546142924751545e-05,
+      "loss": 0.3972,
+      "step": 1650
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.6905998587608337,
+      "learning_rate": 4.2467266130304465e-05,
+      "loss": 0.6448,
+      "step": 1660
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.17736536264419556,
+      "learning_rate": 4.238838933585739e-05,
+      "loss": 0.0182,
+      "step": 1670
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 23.287622451782227,
+      "learning_rate": 4.230951254141032e-05,
+      "loss": 0.9857,
+      "step": 1680
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.051244575530290604,
+      "learning_rate": 4.2230635746963246e-05,
+      "loss": 0.0602,
+      "step": 1690
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.007085567340254784,
+      "learning_rate": 4.215175895251617e-05,
+      "loss": 0.2775,
+      "step": 1700
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.03806788846850395,
+      "learning_rate": 4.20728821580691e-05,
+      "loss": 1.0469,
+      "step": 1710
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.014668051153421402,
+      "learning_rate": 4.199400536362203e-05,
+      "loss": 0.5808,
+      "step": 1720
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 41.8901481628418,
+      "learning_rate": 4.191512856917495e-05,
+      "loss": 1.1275,
+      "step": 1730
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.03532877191901207,
+      "learning_rate": 4.1836251774727874e-05,
+      "loss": 0.3069,
+      "step": 1740
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.010979109443724155,
+      "learning_rate": 4.175737498028081e-05,
+      "loss": 0.0925,
+      "step": 1750
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.9177700281143188,
+      "learning_rate": 4.167849818583373e-05,
+      "loss": 0.0827,
+      "step": 1760
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.488022118806839,
+      "learning_rate": 4.1599621391386655e-05,
+      "loss": 0.3724,
+      "step": 1770
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 1.7911025285720825,
+      "learning_rate": 4.152074459693958e-05,
+      "loss": 0.2346,
+      "step": 1780
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 1.4934897422790527,
+      "learning_rate": 4.144186780249251e-05,
+      "loss": 0.2851,
+      "step": 1790
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 1.4438055753707886,
+      "learning_rate": 4.1362991008045436e-05,
+      "loss": 0.0501,
+      "step": 1800
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 23.35430335998535,
+      "learning_rate": 4.128411421359836e-05,
+      "loss": 0.714,
+      "step": 1810
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 1.5962296724319458,
+      "learning_rate": 4.120523741915129e-05,
+      "loss": 1.3421,
+      "step": 1820
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.019027821719646454,
+      "learning_rate": 4.112636062470421e-05,
+      "loss": 0.7925,
+      "step": 1830
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.40448516607284546,
+      "learning_rate": 4.104748383025714e-05,
+      "loss": 0.0505,
+      "step": 1840
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 52.31330871582031,
+      "learning_rate": 4.096860703581007e-05,
+      "loss": 0.4829,
+      "step": 1850
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 39.16224670410156,
+      "learning_rate": 4.088973024136299e-05,
+      "loss": 0.2865,
+      "step": 1860
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 28.03236198425293,
+      "learning_rate": 4.081085344691592e-05,
+      "loss": 0.4357,
+      "step": 1870
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.016243135556578636,
+      "learning_rate": 4.0731976652468846e-05,
+      "loss": 0.6881,
+      "step": 1880
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.16010265052318573,
+      "learning_rate": 4.065309985802177e-05,
+      "loss": 0.7461,
+      "step": 1890
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 2.2465898990631104,
+      "learning_rate": 4.057422306357469e-05,
+      "loss": 0.2457,
+      "step": 1900
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 50.493865966796875,
+      "learning_rate": 4.049534626912763e-05,
+      "loss": 0.4798,
+      "step": 1910
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 1.1454970836639404,
+      "learning_rate": 4.0416469474680554e-05,
+      "loss": 0.3395,
+      "step": 1920
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.009125631302595139,
+      "learning_rate": 4.0337592680233474e-05,
+      "loss": 0.4666,
+      "step": 1930
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.10130707174539566,
+      "learning_rate": 4.02587158857864e-05,
+      "loss": 0.0233,
+      "step": 1940
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 1.5332127809524536,
+      "learning_rate": 4.017983909133933e-05,
+      "loss": 0.2434,
+      "step": 1950
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.023429008200764656,
+      "learning_rate": 4.0100962296892255e-05,
+      "loss": 0.6019,
+      "step": 1960
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.028758693486452103,
+      "learning_rate": 4.002208550244518e-05,
+      "loss": 0.3796,
+      "step": 1970
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 1.5673832893371582,
+      "learning_rate": 3.994320870799811e-05,
+      "loss": 0.4689,
+      "step": 1980
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.010430915281176567,
+      "learning_rate": 3.9864331913551036e-05,
+      "loss": 0.3058,
+      "step": 1990
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.08046963810920715,
+      "learning_rate": 3.9785455119103956e-05,
+      "loss": 0.005,
+      "step": 2000
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.037964820861816406,
+      "learning_rate": 3.970657832465689e-05,
+      "loss": 0.5847,
+      "step": 2010
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 18.253530502319336,
+      "learning_rate": 3.962770153020982e-05,
+      "loss": 0.2832,
+      "step": 2020
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.012116353027522564,
+      "learning_rate": 3.954882473576274e-05,
+      "loss": 0.7095,
+      "step": 2030
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.005259596277028322,
+      "learning_rate": 3.9469947941315664e-05,
+      "loss": 0.7079,
+      "step": 2040
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.13410820066928864,
+      "learning_rate": 3.939107114686859e-05,
+      "loss": 0.5924,
+      "step": 2050
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.045772209763526917,
+      "learning_rate": 3.9312194352421525e-05,
+      "loss": 0.1816,
+      "step": 2060
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 25.16946792602539,
+      "learning_rate": 3.9233317557974445e-05,
+      "loss": 0.2418,
+      "step": 2070
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.010729226283729076,
+      "learning_rate": 3.915444076352737e-05,
+      "loss": 0.0113,
+      "step": 2080
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 15.555303573608398,
+      "learning_rate": 3.90755639690803e-05,
+      "loss": 0.0163,
+      "step": 2090
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.07517635077238083,
+      "learning_rate": 3.899668717463322e-05,
+      "loss": 0.2035,
+      "step": 2100
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 58.81978988647461,
+      "learning_rate": 3.8917810380186154e-05,
+      "loss": 0.3969,
+      "step": 2110
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 1.662779450416565,
+      "learning_rate": 3.883893358573908e-05,
+      "loss": 1.0464,
+      "step": 2120
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.010347407311201096,
+      "learning_rate": 3.876005679129201e-05,
+      "loss": 0.3169,
+      "step": 2130
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 92.21438598632812,
+      "learning_rate": 3.868117999684493e-05,
+      "loss": 0.6992,
+      "step": 2140
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.013454782776534557,
+      "learning_rate": 3.8602303202397855e-05,
+      "loss": 0.0588,
+      "step": 2150
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.07685262709856033,
+      "learning_rate": 3.852342640795078e-05,
+      "loss": 0.405,
+      "step": 2160
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 32.91915512084961,
+      "learning_rate": 3.844454961350371e-05,
+      "loss": 0.5075,
+      "step": 2170
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.007491769269108772,
+      "learning_rate": 3.8365672819056636e-05,
+      "loss": 0.0138,
+      "step": 2180
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 47.2808952331543,
+      "learning_rate": 3.828679602460956e-05,
+      "loss": 1.1644,
+      "step": 2190
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.38593998551368713,
+      "learning_rate": 3.820791923016249e-05,
+      "loss": 0.102,
+      "step": 2200
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.03193847835063934,
+      "learning_rate": 3.812904243571541e-05,
+      "loss": 0.0032,
+      "step": 2210
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 3.3530170917510986,
+      "learning_rate": 3.8050165641268344e-05,
+      "loss": 0.4006,
+      "step": 2220
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 32.89242172241211,
+      "learning_rate": 3.797128884682127e-05,
+      "loss": 0.3695,
+      "step": 2230
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.007152862381190062,
+      "learning_rate": 3.789241205237419e-05,
+      "loss": 0.3245,
+      "step": 2240
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.06642629951238632,
+      "learning_rate": 3.781353525792712e-05,
+      "loss": 0.0625,
+      "step": 2250
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.010737092234194279,
+      "learning_rate": 3.7734658463480045e-05,
+      "loss": 0.0037,
+      "step": 2260
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 50.68439865112305,
+      "learning_rate": 3.765578166903297e-05,
+      "loss": 0.3923,
+      "step": 2270
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.006863988004624844,
+      "learning_rate": 3.75769048745859e-05,
+      "loss": 0.2766,
+      "step": 2280
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 4.621837139129639,
+      "learning_rate": 3.7498028080138826e-05,
+      "loss": 0.3579,
+      "step": 2290
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.07001708447933197,
+      "learning_rate": 3.741915128569175e-05,
+      "loss": 0.7549,
+      "step": 2300
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.018876327201724052,
+      "learning_rate": 3.7340274491244674e-05,
+      "loss": 0.8258,
+      "step": 2310
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 52.4869384765625,
+      "learning_rate": 3.726139769679761e-05,
+      "loss": 0.6847,
+      "step": 2320
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.027168385684490204,
+      "learning_rate": 3.7182520902350534e-05,
+      "loss": 0.0682,
+      "step": 2330
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 34.695613861083984,
+      "learning_rate": 3.7103644107903455e-05,
+      "loss": 0.0911,
+      "step": 2340
+    },
+    {
+      "epoch": 1.17,
+      "eval_accuracy": 0.8830645161290323,
+      "eval_loss": 0.5390264987945557,
+      "eval_runtime": 1625.0663,
+      "eval_samples_per_second": 0.305,
+      "eval_steps_per_second": 0.153,
+      "step": 2348
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 83.50694274902344,
+      "learning_rate": 3.702476731345638e-05,
+      "loss": 0.4141,
+      "step": 2350
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.07181308418512344,
+      "learning_rate": 3.694589051900931e-05,
+      "loss": 0.0345,
+      "step": 2360
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 7.896732330322266,
+      "learning_rate": 3.6867013724562236e-05,
+      "loss": 0.1293,
+      "step": 2370
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 5.4950127601623535,
+      "learning_rate": 3.678813693011516e-05,
+      "loss": 0.2118,
+      "step": 2380
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.1615094095468521,
+      "learning_rate": 3.670926013566809e-05,
+      "loss": 0.2933,
+      "step": 2390
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.07830177247524261,
+      "learning_rate": 3.663038334122102e-05,
+      "loss": 0.3135,
+      "step": 2400
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.004315768834203482,
+      "learning_rate": 3.655150654677394e-05,
+      "loss": 0.3462,
+      "step": 2410
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.01801230013370514,
+      "learning_rate": 3.6472629752326864e-05,
+      "loss": 0.0197,
+      "step": 2420
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.01667596399784088,
+      "learning_rate": 3.63937529578798e-05,
+      "loss": 0.1788,
+      "step": 2430
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.9513064622879028,
+      "learning_rate": 3.631487616343272e-05,
+      "loss": 0.0015,
+      "step": 2440
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.005059552378952503,
+      "learning_rate": 3.6235999368985645e-05,
+      "loss": 0.0011,
+      "step": 2450
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.9061123728752136,
+      "learning_rate": 3.615712257453857e-05,
+      "loss": 0.0886,
+      "step": 2460
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.009590793401002884,
+      "learning_rate": 3.60782457800915e-05,
+      "loss": 0.0043,
+      "step": 2470
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 32.69272994995117,
+      "learning_rate": 3.5999368985644426e-05,
+      "loss": 0.4264,
+      "step": 2480
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.02591569721698761,
+      "learning_rate": 3.592049219119735e-05,
+      "loss": 0.1742,
+      "step": 2490
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.19918805360794067,
+      "learning_rate": 3.584161539675028e-05,
+      "loss": 0.0196,
+      "step": 2500
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.05109681561589241,
+      "learning_rate": 3.57627386023032e-05,
+      "loss": 0.6586,
+      "step": 2510
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.0862836092710495,
+      "learning_rate": 3.568386180785613e-05,
+      "loss": 0.4987,
+      "step": 2520
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.004215422552078962,
+      "learning_rate": 3.560498501340906e-05,
+      "loss": 0.5482,
+      "step": 2530
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.006058989092707634,
+      "learning_rate": 3.552610821896198e-05,
+      "loss": 0.0132,
+      "step": 2540
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.006959158461540937,
+      "learning_rate": 3.544723142451491e-05,
+      "loss": 0.1954,
+      "step": 2550
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.7044851779937744,
+      "learning_rate": 3.5368354630067835e-05,
+      "loss": 0.0451,
+      "step": 2560
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 57.56528091430664,
+      "learning_rate": 3.528947783562076e-05,
+      "loss": 0.0514,
+      "step": 2570
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 59.18427658081055,
+      "learning_rate": 3.521060104117369e-05,
+      "loss": 0.2163,
+      "step": 2580
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 27.565649032592773,
+      "learning_rate": 3.5131724246726617e-05,
+      "loss": 0.4024,
+      "step": 2590
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.049286678433418274,
+      "learning_rate": 3.5052847452279544e-05,
+      "loss": 0.3371,
+      "step": 2600
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.02281220071017742,
+      "learning_rate": 3.4973970657832464e-05,
+      "loss": 0.0695,
+      "step": 2610
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.004120226483792067,
+      "learning_rate": 3.489509386338539e-05,
+      "loss": 0.2743,
+      "step": 2620
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.005320386495441198,
+      "learning_rate": 3.4816217068938325e-05,
+      "loss": 0.3163,
+      "step": 2630
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 62.42177200317383,
+      "learning_rate": 3.4737340274491245e-05,
+      "loss": 0.6724,
+      "step": 2640
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.5263716578483582,
+      "learning_rate": 3.465846348004417e-05,
+      "loss": 0.4357,
+      "step": 2650
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 1.5683003664016724,
+      "learning_rate": 3.45795866855971e-05,
+      "loss": 0.0465,
+      "step": 2660
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.002447244944050908,
+      "learning_rate": 3.4500709891150026e-05,
+      "loss": 0.6059,
+      "step": 2670
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.0026684117037802935,
+      "learning_rate": 3.4421833096702946e-05,
+      "loss": 0.4851,
+      "step": 2680
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.009127967990934849,
+      "learning_rate": 3.434295630225588e-05,
+      "loss": 0.0971,
+      "step": 2690
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 42.93687438964844,
+      "learning_rate": 3.426407950780881e-05,
+      "loss": 0.027,
+      "step": 2700
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.00475093862041831,
+      "learning_rate": 3.418520271336173e-05,
+      "loss": 0.3864,
+      "step": 2710
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.00734454533085227,
+      "learning_rate": 3.4106325918914654e-05,
+      "loss": 0.3156,
+      "step": 2720
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.013374255038797855,
+      "learning_rate": 3.402744912446758e-05,
+      "loss": 0.3001,
+      "step": 2730
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.0032714835833758116,
+      "learning_rate": 3.394857233002051e-05,
+      "loss": 0.0005,
+      "step": 2740
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.050922248512506485,
+      "learning_rate": 3.3869695535573435e-05,
+      "loss": 0.4824,
+      "step": 2750
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.24358013272285461,
+      "learning_rate": 3.379081874112636e-05,
+      "loss": 0.0034,
+      "step": 2760
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.009304082952439785,
+      "learning_rate": 3.371194194667929e-05,
+      "loss": 0.4735,
+      "step": 2770
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 83.10357666015625,
+      "learning_rate": 3.363306515223221e-05,
+      "loss": 0.1614,
+      "step": 2780
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.010971483774483204,
+      "learning_rate": 3.355418835778514e-05,
+      "loss": 0.0006,
+      "step": 2790
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.021259542554616928,
+      "learning_rate": 3.347531156333807e-05,
+      "loss": 0.1619,
+      "step": 2800
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.03375878557562828,
+      "learning_rate": 3.339643476889099e-05,
+      "loss": 0.3193,
+      "step": 2810
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 4.055739879608154,
+      "learning_rate": 3.331755797444392e-05,
+      "loss": 0.3975,
+      "step": 2820
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 1.8192673921585083,
+      "learning_rate": 3.3238681179996845e-05,
+      "loss": 0.4128,
+      "step": 2830
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 3.6572928428649902,
+      "learning_rate": 3.315980438554978e-05,
+      "loss": 0.0194,
+      "step": 2840
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.013262529857456684,
+      "learning_rate": 3.30809275911027e-05,
+      "loss": 0.2099,
+      "step": 2850
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.003644093405455351,
+      "learning_rate": 3.3002050796655626e-05,
+      "loss": 0.0363,
+      "step": 2860
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.005795649252831936,
+      "learning_rate": 3.292317400220855e-05,
+      "loss": 0.0058,
+      "step": 2870
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.42943742871284485,
+      "learning_rate": 3.284429720776147e-05,
+      "loss": 0.4412,
+      "step": 2880
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 37.513389587402344,
+      "learning_rate": 3.276542041331441e-05,
+      "loss": 0.6395,
+      "step": 2890
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.003234813455492258,
+      "learning_rate": 3.2686543618867334e-05,
+      "loss": 0.0894,
+      "step": 2900
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.0032623664010316133,
+      "learning_rate": 3.260766682442026e-05,
+      "loss": 0.1503,
+      "step": 2910
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.012005943804979324,
+      "learning_rate": 3.252879002997318e-05,
+      "loss": 0.0045,
+      "step": 2920
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.006498160306364298,
+      "learning_rate": 3.244991323552611e-05,
+      "loss": 0.1571,
+      "step": 2930
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.005553847178816795,
+      "learning_rate": 3.2371036441079035e-05,
+      "loss": 0.2607,
+      "step": 2940
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.006091673392802477,
+      "learning_rate": 3.229215964663196e-05,
+      "loss": 0.0167,
+      "step": 2950
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.011702551506459713,
+      "learning_rate": 3.221328285218489e-05,
+      "loss": 0.0002,
+      "step": 2960
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.0019528231350705028,
+      "learning_rate": 3.2134406057737816e-05,
+      "loss": 0.0026,
+      "step": 2970
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.004342043772339821,
+      "learning_rate": 3.205552926329074e-05,
+      "loss": 0.33,
+      "step": 2980
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.0032574606593698263,
+      "learning_rate": 3.197665246884366e-05,
+      "loss": 0.3547,
+      "step": 2990
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.01980379782617092,
+      "learning_rate": 3.18977756743966e-05,
+      "loss": 0.601,
+      "step": 3000
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.15134751796722412,
+      "learning_rate": 3.1818898879949524e-05,
+      "loss": 0.3661,
+      "step": 3010
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.6195966005325317,
+      "learning_rate": 3.1740022085502444e-05,
+      "loss": 0.0714,
+      "step": 3020
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 53.342018127441406,
+      "learning_rate": 3.166114529105537e-05,
+      "loss": 0.3813,
+      "step": 3030
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.0044724950566887856,
+      "learning_rate": 3.15822684966083e-05,
+      "loss": 0.0166,
+      "step": 3040
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.023660195991396904,
+      "learning_rate": 3.1503391702161225e-05,
+      "loss": 0.0268,
+      "step": 3050
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.012484594248235226,
+      "learning_rate": 3.142451490771415e-05,
+      "loss": 0.347,
+      "step": 3060
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.005233893636614084,
+      "learning_rate": 3.134563811326708e-05,
+      "loss": 0.0135,
+      "step": 3070
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.0070632160641252995,
+      "learning_rate": 3.1266761318820007e-05,
+      "loss": 0.1943,
+      "step": 3080
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.008478705771267414,
+      "learning_rate": 3.118788452437293e-05,
+      "loss": 0.1199,
+      "step": 3090
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.7838729023933411,
+      "learning_rate": 3.110900772992586e-05,
+      "loss": 0.0169,
+      "step": 3100
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 3.1755118370056152,
+      "learning_rate": 3.103013093547879e-05,
+      "loss": 0.1134,
+      "step": 3110
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.002370638307183981,
+      "learning_rate": 3.095125414103171e-05,
+      "loss": 0.0516,
+      "step": 3120
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.0009597245953045785,
+      "learning_rate": 3.0872377346584635e-05,
+      "loss": 0.6948,
+      "step": 3130
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.007442851085215807,
+      "learning_rate": 3.079350055213756e-05,
+      "loss": 0.0012,
+      "step": 3140
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.0020674436818808317,
+      "learning_rate": 3.071462375769049e-05,
+      "loss": 0.0051,
+      "step": 3150
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.00474890461191535,
+      "learning_rate": 3.0635746963243416e-05,
+      "loss": 0.0013,
+      "step": 3160
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.010151269845664501,
+      "learning_rate": 3.055687016879634e-05,
+      "loss": 0.0005,
+      "step": 3170
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 50.9908561706543,
+      "learning_rate": 3.0477993374349267e-05,
+      "loss": 0.454,
+      "step": 3180
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.0189617108553648,
+      "learning_rate": 3.0399116579902194e-05,
+      "loss": 0.2066,
+      "step": 3190
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.00790422223508358,
+      "learning_rate": 3.0320239785455117e-05,
+      "loss": 0.0967,
+      "step": 3200
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.007165226619690657,
+      "learning_rate": 3.0241362991008048e-05,
+      "loss": 0.2621,
+      "step": 3210
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 89.41899871826172,
+      "learning_rate": 3.0162486196560975e-05,
+      "loss": 0.8914,
+      "step": 3220
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.08559765666723251,
+      "learning_rate": 3.0083609402113898e-05,
+      "loss": 0.2084,
+      "step": 3230
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.011442103423178196,
+      "learning_rate": 3.0004732607666825e-05,
+      "loss": 0.0171,
+      "step": 3240
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.003758377628400922,
+      "learning_rate": 2.992585581321975e-05,
+      "loss": 0.1336,
+      "step": 3250
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.0027232924476265907,
+      "learning_rate": 2.984697901877268e-05,
+      "loss": 0.0027,
+      "step": 3260
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.017452992498874664,
+      "learning_rate": 2.9768102224325606e-05,
+      "loss": 0.0007,
+      "step": 3270
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.3878909945487976,
+      "learning_rate": 2.968922542987853e-05,
+      "loss": 0.0232,
+      "step": 3280
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.002640586346387863,
+      "learning_rate": 2.9610348635431457e-05,
+      "loss": 0.0005,
+      "step": 3290
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.004480639472603798,
+      "learning_rate": 2.953147184098438e-05,
+      "loss": 0.0029,
+      "step": 3300
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.03782272711396217,
+      "learning_rate": 2.945259504653731e-05,
+      "loss": 0.0002,
+      "step": 3310
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.001542330370284617,
+      "learning_rate": 2.9373718252090238e-05,
+      "loss": 0.0631,
+      "step": 3320
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.003070350969210267,
+      "learning_rate": 2.929484145764316e-05,
+      "loss": 0.0376,
+      "step": 3330
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.0022313990630209446,
+      "learning_rate": 2.921596466319609e-05,
+      "loss": 0.1432,
+      "step": 3340
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 21.07185935974121,
+      "learning_rate": 2.9137087868749012e-05,
+      "loss": 0.0969,
+      "step": 3350
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.06916557252407074,
+      "learning_rate": 2.9058211074301943e-05,
+      "loss": 0.0002,
+      "step": 3360
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.009072001092135906,
+      "learning_rate": 2.897933427985487e-05,
+      "loss": 0.267,
+      "step": 3370
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.007433378137648106,
+      "learning_rate": 2.8900457485407793e-05,
+      "loss": 0.0246,
+      "step": 3380
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.004601635970175266,
+      "learning_rate": 2.882158069096072e-05,
+      "loss": 0.6894,
+      "step": 3390
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.10854679346084595,
+      "learning_rate": 2.8742703896513644e-05,
+      "loss": 0.2008,
+      "step": 3400
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.0007173072663135827,
+      "learning_rate": 2.8663827102066574e-05,
+      "loss": 0.2165,
+      "step": 3410
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.0021528834477066994,
+      "learning_rate": 2.85849503076195e-05,
+      "loss": 0.0393,
+      "step": 3420
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.0069760591723024845,
+      "learning_rate": 2.8506073513172425e-05,
+      "loss": 0.0085,
+      "step": 3430
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.009175879880785942,
+      "learning_rate": 2.8427196718725352e-05,
+      "loss": 0.0389,
+      "step": 3440
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 1.5355865955352783,
+      "learning_rate": 2.8348319924278276e-05,
+      "loss": 0.1762,
+      "step": 3450
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.0020140057895332575,
+      "learning_rate": 2.8269443129831203e-05,
+      "loss": 0.0039,
+      "step": 3460
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.001294136280193925,
+      "learning_rate": 2.8190566335384133e-05,
+      "loss": 0.0115,
+      "step": 3470
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 64.58512878417969,
+      "learning_rate": 2.8111689540937057e-05,
+      "loss": 0.2543,
+      "step": 3480
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.0034382932353764772,
+      "learning_rate": 2.8032812746489984e-05,
+      "loss": 0.0001,
+      "step": 3490
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 44.1510009765625,
+      "learning_rate": 2.7953935952042907e-05,
+      "loss": 1.4329,
+      "step": 3500
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.14346307516098022,
+      "learning_rate": 2.7875059157595834e-05,
+      "loss": 0.0152,
+      "step": 3510
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.6694988012313843,
+      "learning_rate": 2.7796182363148765e-05,
+      "loss": 0.0021,
+      "step": 3520
+    },
+    {
+      "epoch": 2.17,
+      "eval_accuracy": 0.9354838709677419,
+      "eval_loss": 0.2325863540172577,
+      "eval_runtime": 1633.604,
+      "eval_samples_per_second": 0.304,
+      "eval_steps_per_second": 0.152,
+      "step": 3522
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.9486463069915771,
+      "learning_rate": 2.7717305568701692e-05,
+      "loss": 0.3196,
+      "step": 3530
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.00858683418482542,
+      "learning_rate": 2.7638428774254615e-05,
+      "loss": 0.0004,
+      "step": 3540
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 82.04684448242188,
+      "learning_rate": 2.755955197980754e-05,
+      "loss": 0.3248,
+      "step": 3550
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 0.004381874576210976,
+      "learning_rate": 2.7480675185360466e-05,
+      "loss": 0.3727,
+      "step": 3560
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 0.014454508200287819,
+      "learning_rate": 2.7401798390913397e-05,
+      "loss": 0.1753,
+      "step": 3570
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 0.013255137018859386,
+      "learning_rate": 2.7322921596466324e-05,
+      "loss": 0.0315,
+      "step": 3580
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 0.007677259389311075,
+      "learning_rate": 2.7244044802019247e-05,
+      "loss": 0.0001,
+      "step": 3590
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 20.19415283203125,
+      "learning_rate": 2.7165168007572174e-05,
+      "loss": 0.0889,
+      "step": 3600
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 0.0009551621042191982,
+      "learning_rate": 2.7086291213125098e-05,
+      "loss": 0.351,
+      "step": 3610
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 0.0011309119872748852,
+      "learning_rate": 2.7007414418678028e-05,
+      "loss": 0.1626,
+      "step": 3620
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 0.15643049776554108,
+      "learning_rate": 2.6928537624230955e-05,
+      "loss": 0.0005,
+      "step": 3630
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 0.011252015829086304,
+      "learning_rate": 2.684966082978388e-05,
+      "loss": 0.1608,
+      "step": 3640
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 26.875606536865234,
+      "learning_rate": 2.6770784035336806e-05,
+      "loss": 0.245,
+      "step": 3650
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 0.0020260706078261137,
+      "learning_rate": 2.669190724088973e-05,
+      "loss": 0.0001,
+      "step": 3660
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 0.020889783278107643,
+      "learning_rate": 2.661303044644266e-05,
+      "loss": 0.0004,
+      "step": 3670
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 0.004248456563800573,
+      "learning_rate": 2.6534153651995587e-05,
+      "loss": 0.0002,
+      "step": 3680
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 0.0021925840992480516,
+      "learning_rate": 2.645527685754851e-05,
+      "loss": 0.0005,
+      "step": 3690
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 0.0018957388820126653,
+      "learning_rate": 2.6376400063101438e-05,
+      "loss": 0.0071,
+      "step": 3700
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 0.0038514058105647564,
+      "learning_rate": 2.629752326865436e-05,
+      "loss": 0.0002,
+      "step": 3710
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 0.008002794347703457,
+      "learning_rate": 2.6218646474207288e-05,
+      "loss": 0.0002,
+      "step": 3720
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 0.006867717951536179,
+      "learning_rate": 2.613976967976022e-05,
+      "loss": 0.0004,
+      "step": 3730
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 0.0030483207665383816,
+      "learning_rate": 2.6060892885313142e-05,
+      "loss": 0.373,
+      "step": 3740
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 2.781649112701416,
+      "learning_rate": 2.598201609086607e-05,
+      "loss": 0.144,
+      "step": 3750
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 0.0029240138828754425,
+      "learning_rate": 2.5903139296418993e-05,
+      "loss": 0.0655,
+      "step": 3760
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.005044338293373585,
+      "learning_rate": 2.582426250197192e-05,
+      "loss": 0.096,
+      "step": 3770
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.0017533600330352783,
+      "learning_rate": 2.574538570752485e-05,
+      "loss": 0.0027,
+      "step": 3780
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.005865835584700108,
+      "learning_rate": 2.5666508913077774e-05,
+      "loss": 0.0859,
+      "step": 3790
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 35.962486267089844,
+      "learning_rate": 2.55876321186307e-05,
+      "loss": 0.8106,
+      "step": 3800
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.002111587440595031,
+      "learning_rate": 2.5508755324183625e-05,
+      "loss": 0.003,
+      "step": 3810
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 76.0323257446289,
+      "learning_rate": 2.542987852973655e-05,
+      "loss": 0.1045,
+      "step": 3820
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.4966025650501251,
+      "learning_rate": 2.5351001735289482e-05,
+      "loss": 0.0008,
+      "step": 3830
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 0.005008663050830364,
+      "learning_rate": 2.5272124940842406e-05,
+      "loss": 0.0005,
+      "step": 3840
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 0.002375594340264797,
+      "learning_rate": 2.5193248146395333e-05,
+      "loss": 0.0052,
+      "step": 3850
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 23.257604598999023,
+      "learning_rate": 2.5114371351948256e-05,
+      "loss": 0.7102,
+      "step": 3860
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 0.002120954217389226,
+      "learning_rate": 2.5035494557501183e-05,
+      "loss": 0.0013,
+      "step": 3870
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 0.0004507018893491477,
+      "learning_rate": 2.495661776305411e-05,
+      "loss": 0.0195,
+      "step": 3880
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 0.0011175195686519146,
+      "learning_rate": 2.4877740968607034e-05,
+      "loss": 0.0925,
+      "step": 3890
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 0.002447870559990406,
+      "learning_rate": 2.4798864174159964e-05,
+      "loss": 0.0001,
+      "step": 3900
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.0006674313917756081,
+      "learning_rate": 2.4719987379712888e-05,
+      "loss": 0.013,
+      "step": 3910
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.057717230170965195,
+      "learning_rate": 2.464111058526582e-05,
+      "loss": 0.0004,
+      "step": 3920
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.05578107759356499,
+      "learning_rate": 2.4562233790818742e-05,
+      "loss": 0.0004,
+      "step": 3930
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.14482277631759644,
+      "learning_rate": 2.448335699637167e-05,
+      "loss": 0.5712,
+      "step": 3940
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.04522058740258217,
+      "learning_rate": 2.4404480201924596e-05,
+      "loss": 0.0026,
+      "step": 3950
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.002966469619423151,
+      "learning_rate": 2.432560340747752e-05,
+      "loss": 0.0006,
+      "step": 3960
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 0.00053648796165362,
+      "learning_rate": 2.424672661303045e-05,
+      "loss": 0.0505,
+      "step": 3970
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.0006470835069194436,
+      "learning_rate": 2.4167849818583374e-05,
+      "loss": 0.0007,
+      "step": 3980
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.02763412706553936,
+      "learning_rate": 2.40889730241363e-05,
+      "loss": 0.0001,
+      "step": 3990
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.008323222398757935,
+      "learning_rate": 2.4010096229689228e-05,
+      "loss": 0.0841,
+      "step": 4000
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.0019473332213237882,
+      "learning_rate": 2.393121943524215e-05,
+      "loss": 0.1134,
+      "step": 4010
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 28.365497589111328,
+      "learning_rate": 2.385234264079508e-05,
+      "loss": 0.1109,
+      "step": 4020
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.527344822883606,
+      "learning_rate": 2.3773465846348005e-05,
+      "loss": 0.0035,
+      "step": 4030
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.7328306436538696,
+      "learning_rate": 2.3694589051900933e-05,
+      "loss": 0.0062,
+      "step": 4040
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 12.828133583068848,
+      "learning_rate": 2.361571225745386e-05,
+      "loss": 0.0102,
+      "step": 4050
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 2.4953901767730713,
+      "learning_rate": 2.3536835463006783e-05,
+      "loss": 0.0076,
+      "step": 4060
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 0.0233890563249588,
+      "learning_rate": 2.345795866855971e-05,
+      "loss": 0.036,
+      "step": 4070
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 0.003167585236951709,
+      "learning_rate": 2.3379081874112637e-05,
+      "loss": 0.0003,
+      "step": 4080
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 0.0012705402914434671,
+      "learning_rate": 2.3300205079665564e-05,
+      "loss": 0.4268,
+      "step": 4090
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 48.02186965942383,
+      "learning_rate": 2.322132828521849e-05,
+      "loss": 0.4373,
+      "step": 4100
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 0.006636523641645908,
+      "learning_rate": 2.3142451490771415e-05,
+      "loss": 0.0002,
+      "step": 4110
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 50.187374114990234,
+      "learning_rate": 2.3063574696324342e-05,
+      "loss": 0.0308,
+      "step": 4120
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 0.0010502920486032963,
+      "learning_rate": 2.298469790187727e-05,
+      "loss": 0.0001,
+      "step": 4130
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 0.007113128434866667,
+      "learning_rate": 2.2905821107430196e-05,
+      "loss": 0.0008,
+      "step": 4140
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 0.036530692130327225,
+      "learning_rate": 2.282694431298312e-05,
+      "loss": 0.0002,
+      "step": 4150
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 0.06909912079572678,
+      "learning_rate": 2.2748067518536047e-05,
+      "loss": 0.0003,
+      "step": 4160
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 0.0019159259973093867,
+      "learning_rate": 2.2669190724088974e-05,
+      "loss": 0.0083,
+      "step": 4170
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 0.0021332651376724243,
+      "learning_rate": 2.25903139296419e-05,
+      "loss": 0.0001,
+      "step": 4180
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 0.0026469645090401173,
+      "learning_rate": 2.2511437135194828e-05,
+      "loss": 0.0002,
+      "step": 4190
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 0.008601084351539612,
+      "learning_rate": 2.243256034074775e-05,
+      "loss": 0.2097,
+      "step": 4200
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 0.015036378987133503,
+      "learning_rate": 2.235368354630068e-05,
+      "loss": 0.0004,
+      "step": 4210
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 0.0036619622260332108,
+      "learning_rate": 2.2274806751853605e-05,
+      "loss": 0.0002,
+      "step": 4220
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 3.0393753051757812,
+      "learning_rate": 2.2195929957406532e-05,
+      "loss": 0.0047,
+      "step": 4230
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 0.003793679177761078,
+      "learning_rate": 2.211705316295946e-05,
+      "loss": 0.3331,
+      "step": 4240
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 0.2550521790981293,
+      "learning_rate": 2.2038176368512383e-05,
+      "loss": 0.0231,
+      "step": 4250
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 0.0018365568248555064,
+      "learning_rate": 2.1959299574065313e-05,
+      "loss": 0.0034,
+      "step": 4260
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 0.003998609725385904,
+      "learning_rate": 2.1880422779618237e-05,
+      "loss": 0.3161,
+      "step": 4270
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 0.007352112792432308,
+      "learning_rate": 2.1801545985171164e-05,
+      "loss": 0.0003,
+      "step": 4280
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 0.0015179958427324891,
+      "learning_rate": 2.172266919072409e-05,
+      "loss": 0.3138,
+      "step": 4290
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 0.002980669727548957,
+      "learning_rate": 2.1643792396277015e-05,
+      "loss": 0.0023,
+      "step": 4300
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 0.003277245443314314,
+      "learning_rate": 2.1564915601829945e-05,
+      "loss": 0.2865,
+      "step": 4310
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 0.007209760136902332,
+      "learning_rate": 2.148603880738287e-05,
+      "loss": 0.2303,
+      "step": 4320
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 0.004860408138483763,
+      "learning_rate": 2.1407162012935796e-05,
+      "loss": 0.0002,
+      "step": 4330
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 0.006379884667694569,
+      "learning_rate": 2.1328285218488723e-05,
+      "loss": 0.3253,
+      "step": 4340
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 0.0946652889251709,
+      "learning_rate": 2.1249408424041646e-05,
+      "loss": 0.0003,
+      "step": 4350
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 0.24752424657344818,
+      "learning_rate": 2.1170531629594577e-05,
+      "loss": 0.0668,
+      "step": 4360
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 0.0038672536611557007,
+      "learning_rate": 2.10916548351475e-05,
+      "loss": 0.0003,
+      "step": 4370
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 0.0020559318363666534,
+      "learning_rate": 2.1012778040700427e-05,
+      "loss": 0.1577,
+      "step": 4380
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 0.004125183913856745,
+      "learning_rate": 2.0933901246253354e-05,
+      "loss": 0.0208,
+      "step": 4390
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 0.004827685654163361,
+      "learning_rate": 2.0855024451806278e-05,
+      "loss": 0.0001,
+      "step": 4400
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.03129200264811516,
+      "learning_rate": 2.0776147657359205e-05,
+      "loss": 0.1293,
+      "step": 4410
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.0035766016226261854,
+      "learning_rate": 2.0697270862912132e-05,
+      "loss": 0.0657,
+      "step": 4420
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.012179574929177761,
+      "learning_rate": 2.061839406846506e-05,
+      "loss": 0.0003,
+      "step": 4430
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.011766198091208935,
+      "learning_rate": 2.0539517274017986e-05,
+      "loss": 0.0042,
+      "step": 4440
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.004765970166772604,
+      "learning_rate": 2.046064047957091e-05,
+      "loss": 0.0001,
+      "step": 4450
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.009594051167368889,
+      "learning_rate": 2.0381763685123837e-05,
+      "loss": 0.2083,
+      "step": 4460
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 108.7445068359375,
+      "learning_rate": 2.0302886890676764e-05,
+      "loss": 0.1901,
+      "step": 4470
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 0.006083535496145487,
+      "learning_rate": 2.022401009622969e-05,
+      "loss": 0.0003,
+      "step": 4480
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 0.08322729915380478,
+      "learning_rate": 2.0145133301782618e-05,
+      "loss": 0.0018,
+      "step": 4490
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 0.0031785466708242893,
+      "learning_rate": 2.006625650733554e-05,
+      "loss": 0.0052,
+      "step": 4500
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 0.0017340343911200762,
+      "learning_rate": 1.998737971288847e-05,
+      "loss": 0.004,
+      "step": 4510
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 0.0038164989091455936,
+      "learning_rate": 1.9908502918441395e-05,
+      "loss": 0.0018,
+      "step": 4520
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 0.0010812565451487899,
+      "learning_rate": 1.9829626123994323e-05,
+      "loss": 0.0005,
+      "step": 4530
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 0.002185672055929899,
+      "learning_rate": 1.9750749329547246e-05,
+      "loss": 0.1151,
+      "step": 4540
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 0.0019652440678328276,
+      "learning_rate": 1.9671872535100173e-05,
+      "loss": 0.0001,
+      "step": 4550
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 0.0038963062688708305,
+      "learning_rate": 1.95929957406531e-05,
+      "loss": 0.0003,
+      "step": 4560
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 0.0006577034364454448,
+      "learning_rate": 1.9514118946206027e-05,
+      "loss": 0.0001,
+      "step": 4570
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 0.001732866745442152,
+      "learning_rate": 1.9435242151758954e-05,
+      "loss": 0.1175,
+      "step": 4580
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 0.0009296005009673536,
+      "learning_rate": 1.9356365357311878e-05,
+      "loss": 0.0473,
+      "step": 4590
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 0.0023930154275149107,
+      "learning_rate": 1.9277488562864808e-05,
+      "loss": 0.0001,
+      "step": 4600
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 0.0015132995322346687,
+      "learning_rate": 1.9198611768417732e-05,
+      "loss": 0.6946,
+      "step": 4610
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 0.0016138931969180703,
+      "learning_rate": 1.911973497397066e-05,
+      "loss": 0.0001,
+      "step": 4620
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 7.3594970703125,
+      "learning_rate": 1.9040858179523586e-05,
+      "loss": 0.3479,
+      "step": 4630
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 0.005374920088797808,
+      "learning_rate": 1.896198138507651e-05,
+      "loss": 0.0045,
+      "step": 4640
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 0.014135061763226986,
+      "learning_rate": 1.888310459062944e-05,
+      "loss": 0.0005,
+      "step": 4650
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 0.01300802268087864,
+      "learning_rate": 1.8804227796182364e-05,
+      "loss": 0.0001,
+      "step": 4660
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 0.004857921972870827,
+      "learning_rate": 1.872535100173529e-05,
+      "loss": 0.1488,
+      "step": 4670
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 0.004086265806108713,
+      "learning_rate": 1.8646474207288218e-05,
+      "loss": 0.5835,
+      "step": 4680
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 0.0017570651834830642,
+      "learning_rate": 1.856759741284114e-05,
+      "loss": 0.0028,
+      "step": 4690
+    },
+    {
+      "epoch": 3.17,
+      "eval_accuracy": 0.9395161290322581,
+      "eval_loss": 0.31246837973594666,
+      "eval_runtime": 1656.1068,
+      "eval_samples_per_second": 0.299,
+      "eval_steps_per_second": 0.15,
+      "step": 4696
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.009428204968571663,
+      "learning_rate": 1.848872061839407e-05,
+      "loss": 0.0002,
+      "step": 4700
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.18937572836875916,
+      "learning_rate": 1.8409843823946995e-05,
+      "loss": 0.0012,
+      "step": 4710
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.0004322623135522008,
+      "learning_rate": 1.8330967029499922e-05,
+      "loss": 0.0156,
+      "step": 4720
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.003526079934090376,
+      "learning_rate": 1.825209023505285e-05,
+      "loss": 0.0001,
+      "step": 4730
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 0.0006320773391053081,
+      "learning_rate": 1.8173213440605773e-05,
+      "loss": 0.0621,
+      "step": 4740
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 1.2660291194915771,
+      "learning_rate": 1.8094336646158703e-05,
+      "loss": 0.0007,
+      "step": 4750
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 0.0031638769432902336,
+      "learning_rate": 1.8015459851711627e-05,
+      "loss": 0.0001,
+      "step": 4760
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 0.013519646599888802,
+      "learning_rate": 1.7936583057264554e-05,
+      "loss": 0.0003,
+      "step": 4770
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 0.001212167670018971,
+      "learning_rate": 1.785770626281748e-05,
+      "loss": 0.222,
+      "step": 4780
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 0.0012346877483651042,
+      "learning_rate": 1.7778829468370405e-05,
+      "loss": 0.0004,
+      "step": 4790
+    },
+    {
+      "epoch": 4.01,
+      "grad_norm": 0.04241856932640076,
+      "learning_rate": 1.769995267392333e-05,
+      "loss": 0.0002,
+      "step": 4800
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 0.0030141491442918777,
+      "learning_rate": 1.762107587947626e-05,
+      "loss": 0.0002,
+      "step": 4810
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 0.013954492285847664,
+      "learning_rate": 1.7542199085029186e-05,
+      "loss": 0.0001,
+      "step": 4820
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 0.0023927518632262945,
+      "learning_rate": 1.7463322290582113e-05,
+      "loss": 0.0003,
+      "step": 4830
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 0.004521504510194063,
+      "learning_rate": 1.7384445496135036e-05,
+      "loss": 0.0003,
+      "step": 4840
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 0.004896995611488819,
+      "learning_rate": 1.7305568701687963e-05,
+      "loss": 0.0001,
+      "step": 4850
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 0.0036749409046024084,
+      "learning_rate": 1.722669190724089e-05,
+      "loss": 0.0256,
+      "step": 4860
+    },
+    {
+      "epoch": 4.02,
+      "grad_norm": 0.0008271436090581119,
+      "learning_rate": 1.7147815112793817e-05,
+      "loss": 0.0003,
+      "step": 4870
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 0.004261141177266836,
+      "learning_rate": 1.7068938318346744e-05,
+      "loss": 0.0001,
+      "step": 4880
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 0.0018433834193274379,
+      "learning_rate": 1.6990061523899668e-05,
+      "loss": 0.0001,
+      "step": 4890
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 0.015866931527853012,
+      "learning_rate": 1.6911184729452595e-05,
+      "loss": 0.0081,
+      "step": 4900
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 0.03334396332502365,
+      "learning_rate": 1.6832307935005522e-05,
+      "loss": 0.0004,
+      "step": 4910
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 0.056711308658123016,
+      "learning_rate": 1.675343114055845e-05,
+      "loss": 0.0039,
+      "step": 4920
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 35.81077194213867,
+      "learning_rate": 1.6674554346111373e-05,
+      "loss": 0.0307,
+      "step": 4930
+    },
+    {
+      "epoch": 4.03,
+      "grad_norm": 0.0010346799390390515,
+      "learning_rate": 1.6595677551664303e-05,
+      "loss": 0.0043,
+      "step": 4940
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 0.006793358828872442,
+      "learning_rate": 1.6516800757217227e-05,
+      "loss": 0.0012,
+      "step": 4950
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 6.214980602264404,
+      "learning_rate": 1.6437923962770154e-05,
+      "loss": 0.0029,
+      "step": 4960
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 0.0007869885885156691,
+      "learning_rate": 1.635904716832308e-05,
+      "loss": 0.0046,
+      "step": 4970
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 0.001759283128194511,
+      "learning_rate": 1.6280170373876004e-05,
+      "loss": 0.0023,
+      "step": 4980
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 0.004309698008000851,
+      "learning_rate": 1.6201293579428935e-05,
+      "loss": 0.0006,
+      "step": 4990
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 0.4593736231327057,
+      "learning_rate": 1.612241678498186e-05,
+      "loss": 0.66,
+      "step": 5000
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 0.001134719350375235,
+      "learning_rate": 1.6043539990534785e-05,
+      "loss": 0.0003,
+      "step": 5010
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 0.0006022979505360126,
+      "learning_rate": 1.5964663196087713e-05,
+      "loss": 0.0001,
+      "step": 5020
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 0.0009047788917087018,
+      "learning_rate": 1.5885786401640636e-05,
+      "loss": 0.0001,
+      "step": 5030
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 0.001102360780350864,
+      "learning_rate": 1.5806909607193567e-05,
+      "loss": 0.0023,
+      "step": 5040
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 0.0010434681316837668,
+      "learning_rate": 1.572803281274649e-05,
+      "loss": 0.0001,
+      "step": 5050
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 0.0007644995348528028,
+      "learning_rate": 1.5649156018299417e-05,
+      "loss": 0.0003,
+      "step": 5060
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 0.0008155221585184336,
+      "learning_rate": 1.5570279223852344e-05,
+      "loss": 0.4132,
+      "step": 5070
+    },
+    {
+      "epoch": 4.05,
+      "grad_norm": 0.0013060912024229765,
+      "learning_rate": 1.5491402429405268e-05,
+      "loss": 0.0001,
+      "step": 5080
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 21.223411560058594,
+      "learning_rate": 1.5412525634958198e-05,
+      "loss": 0.2948,
+      "step": 5090
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 0.0030962564051151276,
+      "learning_rate": 1.5333648840511122e-05,
+      "loss": 0.0001,
+      "step": 5100
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 0.0018799362005665898,
+      "learning_rate": 1.5254772046064047e-05,
+      "loss": 0.0001,
+      "step": 5110
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 0.006435804534703493,
+      "learning_rate": 1.5175895251616976e-05,
+      "loss": 0.0001,
+      "step": 5120
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 0.0019598493818193674,
+      "learning_rate": 1.5097018457169901e-05,
+      "loss": 0.0004,
+      "step": 5130
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 0.0014030305901542306,
+      "learning_rate": 1.5018141662722828e-05,
+      "loss": 0.0001,
+      "step": 5140
+    },
+    {
+      "epoch": 4.06,
+      "grad_norm": 0.003073945175856352,
+      "learning_rate": 1.4939264868275754e-05,
+      "loss": 0.0013,
+      "step": 5150
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 1.3647187948226929,
+      "learning_rate": 1.4860388073828679e-05,
+      "loss": 0.0028,
+      "step": 5160
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 0.002111067296937108,
+      "learning_rate": 1.4781511279381608e-05,
+      "loss": 0.0015,
+      "step": 5170
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 0.006969344802200794,
+      "learning_rate": 1.4702634484934533e-05,
+      "loss": 0.0001,
+      "step": 5180
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 0.00262584094889462,
+      "learning_rate": 1.4623757690487458e-05,
+      "loss": 0.0001,
+      "step": 5190
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 41.56041717529297,
+      "learning_rate": 1.4544880896040385e-05,
+      "loss": 0.3105,
+      "step": 5200
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 0.005535661708563566,
+      "learning_rate": 1.446600410159331e-05,
+      "loss": 0.0001,
+      "step": 5210
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 0.002079414902254939,
+      "learning_rate": 1.438712730714624e-05,
+      "loss": 0.0357,
+      "step": 5220
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 0.017037468031048775,
+      "learning_rate": 1.4308250512699165e-05,
+      "loss": 0.0006,
+      "step": 5230
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 0.0005684141651727259,
+      "learning_rate": 1.422937371825209e-05,
+      "loss": 0.0003,
+      "step": 5240
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 23.38873291015625,
+      "learning_rate": 1.4150496923805017e-05,
+      "loss": 0.443,
+      "step": 5250
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 0.006656976882368326,
+      "learning_rate": 1.4071620129357942e-05,
+      "loss": 0.0001,
+      "step": 5260
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 0.03132672235369682,
+      "learning_rate": 1.3992743334910871e-05,
+      "loss": 0.0001,
+      "step": 5270
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 0.0020499953534454107,
+      "learning_rate": 1.3913866540463796e-05,
+      "loss": 0.0285,
+      "step": 5280
+    },
+    {
+      "epoch": 4.08,
+      "grad_norm": 0.0015906492481008172,
+      "learning_rate": 1.3834989746016722e-05,
+      "loss": 0.0063,
+      "step": 5290
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 0.004495760891586542,
+      "learning_rate": 1.375611295156965e-05,
+      "loss": 0.0005,
+      "step": 5300
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 0.0008559366688132286,
+      "learning_rate": 1.3677236157122576e-05,
+      "loss": 0.0001,
+      "step": 5310
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 0.0009787013987079263,
+      "learning_rate": 1.35983593626755e-05,
+      "loss": 0.008,
+      "step": 5320
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 0.0012426511384546757,
+      "learning_rate": 1.3519482568228428e-05,
+      "loss": 0.0002,
+      "step": 5330
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 0.0016171043971553445,
+      "learning_rate": 1.3440605773781353e-05,
+      "loss": 0.0001,
+      "step": 5340
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 0.0006655818433500826,
+      "learning_rate": 1.3361728979334282e-05,
+      "loss": 0.0049,
+      "step": 5350
+    },
+    {
+      "epoch": 4.09,
+      "grad_norm": 0.0017938172677531838,
+      "learning_rate": 1.3282852184887207e-05,
+      "loss": 0.0021,
+      "step": 5360
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 0.0019568654242902994,
+      "learning_rate": 1.3203975390440133e-05,
+      "loss": 0.0001,
+      "step": 5370
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 0.032270222902297974,
+      "learning_rate": 1.312509859599306e-05,
+      "loss": 0.0001,
+      "step": 5380
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 0.0030949311330914497,
+      "learning_rate": 1.3046221801545985e-05,
+      "loss": 0.0001,
+      "step": 5390
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 0.0014502047561109066,
+      "learning_rate": 1.2967345007098914e-05,
+      "loss": 0.0,
+      "step": 5400
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 0.006577344611287117,
+      "learning_rate": 1.2888468212651839e-05,
+      "loss": 0.0001,
+      "step": 5410
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 0.0009933020919561386,
+      "learning_rate": 1.2809591418204764e-05,
+      "loss": 0.0001,
+      "step": 5420
+    },
+    {
+      "epoch": 4.1,
+      "grad_norm": 0.0014573619700968266,
+      "learning_rate": 1.2730714623757691e-05,
+      "loss": 0.0001,
+      "step": 5430
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 0.000841297791339457,
+      "learning_rate": 1.2651837829310617e-05,
+      "loss": 0.3696,
+      "step": 5440
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 0.000893387827090919,
+      "learning_rate": 1.2572961034863542e-05,
+      "loss": 0.003,
+      "step": 5450
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 0.0006943478365428746,
+      "learning_rate": 1.249408424041647e-05,
+      "loss": 0.0001,
+      "step": 5460
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 0.0015476369298994541,
+      "learning_rate": 1.2415207445969398e-05,
+      "loss": 0.0009,
+      "step": 5470
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 0.0017319549806416035,
+      "learning_rate": 1.2336330651522323e-05,
+      "loss": 0.0002,
+      "step": 5480
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 0.0007892731809988618,
+      "learning_rate": 1.2257453857075248e-05,
+      "loss": 0.0001,
+      "step": 5490
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 0.0014162284787744284,
+      "learning_rate": 1.2178577062628175e-05,
+      "loss": 0.0001,
+      "step": 5500
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 0.0019155018962919712,
+      "learning_rate": 1.2099700268181103e-05,
+      "loss": 0.1403,
+      "step": 5510
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 0.0012052778620272875,
+      "learning_rate": 1.2020823473734028e-05,
+      "loss": 0.0001,
+      "step": 5520
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 0.0014317920431494713,
+      "learning_rate": 1.1941946679286955e-05,
+      "loss": 0.0001,
+      "step": 5530
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 0.0008730898844078183,
+      "learning_rate": 1.186306988483988e-05,
+      "loss": 0.0006,
+      "step": 5540
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 0.0009261332452297211,
+      "learning_rate": 1.1784193090392807e-05,
+      "loss": 0.0002,
+      "step": 5550
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 0.003988543525338173,
+      "learning_rate": 1.1705316295945733e-05,
+      "loss": 0.3042,
+      "step": 5560
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 0.0012267963029444218,
+      "learning_rate": 1.162643950149866e-05,
+      "loss": 0.0002,
+      "step": 5570
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 20.188026428222656,
+      "learning_rate": 1.1547562707051587e-05,
+      "loss": 0.2263,
+      "step": 5580
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 0.012555709108710289,
+      "learning_rate": 1.1468685912604512e-05,
+      "loss": 0.0001,
+      "step": 5590
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 0.0016228832537308335,
+      "learning_rate": 1.1389809118157439e-05,
+      "loss": 0.0777,
+      "step": 5600
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 0.000622469000518322,
+      "learning_rate": 1.1310932323710364e-05,
+      "loss": 0.0002,
+      "step": 5610
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 0.0049230060540139675,
+      "learning_rate": 1.1232055529263291e-05,
+      "loss": 0.001,
+      "step": 5620
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 0.0005051741609349847,
+      "learning_rate": 1.1153178734816218e-05,
+      "loss": 0.0001,
+      "step": 5630
+    },
+    {
+      "epoch": 4.13,
+      "grad_norm": 20.605552673339844,
+      "learning_rate": 1.1074301940369145e-05,
+      "loss": 0.3086,
+      "step": 5640
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 0.00433269701898098,
+      "learning_rate": 1.0995425145922069e-05,
+      "loss": 0.0001,
+      "step": 5650
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 0.009856464341282845,
+      "learning_rate": 1.0916548351474996e-05,
+      "loss": 0.0121,
+      "step": 5660
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 0.0012663742527365685,
+      "learning_rate": 1.0837671557027923e-05,
+      "loss": 0.3383,
+      "step": 5670
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 0.006808194797486067,
+      "learning_rate": 1.075879476258085e-05,
+      "loss": 0.0001,
+      "step": 5680
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 0.0009030276560224593,
+      "learning_rate": 1.0679917968133775e-05,
+      "loss": 0.211,
+      "step": 5690
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 0.014718537218868732,
+      "learning_rate": 1.0601041173686702e-05,
+      "loss": 0.0036,
+      "step": 5700
+    },
+    {
+      "epoch": 4.14,
+      "grad_norm": 0.0008488602470606565,
+      "learning_rate": 1.0522164379239628e-05,
+      "loss": 0.0001,
+      "step": 5710
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 0.001407152391038835,
+      "learning_rate": 1.0443287584792555e-05,
+      "loss": 0.0001,
+      "step": 5720
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 0.0023314310237765312,
+      "learning_rate": 1.0364410790345482e-05,
+      "loss": 0.2512,
+      "step": 5730
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 0.028335148468613625,
+      "learning_rate": 1.0285533995898407e-05,
+      "loss": 0.0002,
+      "step": 5740
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 0.001055800705216825,
+      "learning_rate": 1.0206657201451334e-05,
+      "loss": 0.0004,
+      "step": 5750
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 0.0012961908942088485,
+      "learning_rate": 1.012778040700426e-05,
+      "loss": 0.0002,
+      "step": 5760
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 0.0017984593287110329,
+      "learning_rate": 1.0048903612557186e-05,
+      "loss": 0.0003,
+      "step": 5770
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 0.002949001733213663,
+      "learning_rate": 9.970026818110112e-06,
+      "loss": 0.0055,
+      "step": 5780
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 0.0291054118424654,
+      "learning_rate": 9.891150023663039e-06,
+      "loss": 0.1212,
+      "step": 5790
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 0.022276869043707848,
+      "learning_rate": 9.812273229215966e-06,
+      "loss": 0.0481,
+      "step": 5800
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 0.0016545113176107407,
+      "learning_rate": 9.733396434768893e-06,
+      "loss": 0.0001,
+      "step": 5810
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 31.92535972595215,
+      "learning_rate": 9.654519640321816e-06,
+      "loss": 0.1336,
+      "step": 5820
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 4.664388179779053,
+      "learning_rate": 9.575642845874743e-06,
+      "loss": 0.0096,
+      "step": 5830
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 0.006792199797928333,
+      "learning_rate": 9.49676605142767e-06,
+      "loss": 0.01,
+      "step": 5840
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 0.002514186082407832,
+      "learning_rate": 9.417889256980597e-06,
+      "loss": 0.0001,
+      "step": 5850
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 0.0008626214112155139,
+      "learning_rate": 9.339012462533524e-06,
+      "loss": 0.1226,
+      "step": 5860
+    },
+    {
+      "epoch": 4.17,
+      "grad_norm": 0.0056907725520431995,
+      "learning_rate": 9.26013566808645e-06,
+      "loss": 0.0008,
+      "step": 5870
+    },
+    {
+      "epoch": 4.17,
+      "eval_accuracy": 0.9475806451612904,
+      "eval_loss": 0.3077409267425537,
+      "eval_runtime": 1692.6192,
+      "eval_samples_per_second": 0.293,
+      "eval_steps_per_second": 0.147,
+      "step": 5870
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 0.0016933621373027563,
+      "learning_rate": 9.181258873639375e-06,
+      "loss": 0.0001,
+      "step": 5880
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 80.78791046142578,
+      "learning_rate": 9.102382079192302e-06,
+      "loss": 0.0298,
+      "step": 5890
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 0.0016686166636645794,
+      "learning_rate": 9.023505284745229e-06,
+      "loss": 0.0001,
+      "step": 5900
+    },
+    {
+      "epoch": 5.01,
+      "grad_norm": 0.01105748862028122,
+      "learning_rate": 8.944628490298154e-06,
+      "loss": 0.0009,
+      "step": 5910
+    },
+    {
+      "epoch": 5.01,
+      "grad_norm": 0.0029821807984262705,
+      "learning_rate": 8.865751695851081e-06,
+      "loss": 0.0001,
+      "step": 5920
+    },
+    {
+      "epoch": 5.01,
+      "grad_norm": 0.005396820139139891,
+      "learning_rate": 8.786874901404007e-06,
+      "loss": 0.0001,
+      "step": 5930
+    },
+    {
+      "epoch": 5.01,
+      "grad_norm": 0.006531829945743084,
+      "learning_rate": 8.707998106956934e-06,
+      "loss": 0.0001,
+      "step": 5940
+    },
+    {
+      "epoch": 5.01,
+      "grad_norm": 0.001007920247502625,
+      "learning_rate": 8.629121312509859e-06,
+      "loss": 0.0002,
+      "step": 5950
+    },
+    {
+      "epoch": 5.01,
+      "grad_norm": 0.05112405866384506,
+      "learning_rate": 8.550244518062786e-06,
+      "loss": 0.0001,
+      "step": 5960
+    },
+    {
+      "epoch": 5.01,
+      "grad_norm": 0.001093395403586328,
+      "learning_rate": 8.471367723615713e-06,
+      "loss": 0.0001,
+      "step": 5970
+    },
+    {
+      "epoch": 5.02,
+      "grad_norm": 0.0007405582000501454,
+      "learning_rate": 8.39249092916864e-06,
+      "loss": 0.0001,
+      "step": 5980
+    },
+    {
+      "epoch": 5.02,
+      "grad_norm": 0.0018120004096999764,
+      "learning_rate": 8.313614134721565e-06,
+      "loss": 0.0019,
+      "step": 5990
+    },
+    {
+      "epoch": 5.02,
+      "grad_norm": 0.1039557084441185,
+      "learning_rate": 8.23473734027449e-06,
+      "loss": 0.0005,
+      "step": 6000
+    },
+    {
+      "epoch": 5.02,
+      "grad_norm": 1.121138572692871,
+      "learning_rate": 8.155860545827418e-06,
+      "loss": 0.06,
+      "step": 6010
+    },
+    {
+      "epoch": 5.02,
+      "grad_norm": 0.0031882787588983774,
+      "learning_rate": 8.076983751380345e-06,
+      "loss": 0.0,
+      "step": 6020
+    },
+    {
+      "epoch": 5.02,
+      "grad_norm": 0.0027095868717879057,
+      "learning_rate": 7.998106956933272e-06,
+      "loss": 0.0001,
+      "step": 6030
+    },
+    {
+      "epoch": 5.02,
+      "grad_norm": 0.0024454870726913214,
+      "learning_rate": 7.919230162486197e-06,
+      "loss": 0.0004,
+      "step": 6040
+    },
+    {
+      "epoch": 5.03,
+      "grad_norm": 0.0022839908488094807,
+      "learning_rate": 7.840353368039123e-06,
+      "loss": 0.0001,
+      "step": 6050
+    },
+    {
+      "epoch": 5.03,
+      "grad_norm": 0.0016150075243785977,
+      "learning_rate": 7.76147657359205e-06,
+      "loss": 0.0001,
+      "step": 6060
+    },
+    {
+      "epoch": 5.03,
+      "grad_norm": 0.0014503969578072429,
+      "learning_rate": 7.682599779144977e-06,
+      "loss": 0.0001,
+      "step": 6070
+    },
+    {
+      "epoch": 5.03,
+      "grad_norm": 0.0006947741494514048,
+      "learning_rate": 7.603722984697902e-06,
+      "loss": 0.1056,
+      "step": 6080
+    },
+    {
+      "epoch": 5.03,
+      "grad_norm": 0.0009220929932780564,
+      "learning_rate": 7.524846190250828e-06,
+      "loss": 0.0001,
+      "step": 6090
+    },
+    {
+      "epoch": 5.03,
+      "grad_norm": 0.001653430750593543,
+      "learning_rate": 7.445969395803755e-06,
+      "loss": 0.0038,
+      "step": 6100
+    },
+    {
+      "epoch": 5.03,
+      "grad_norm": 0.001383556635119021,
+      "learning_rate": 7.367092601356681e-06,
+      "loss": 0.0004,
+      "step": 6110
+    },
+    {
+      "epoch": 5.04,
+      "grad_norm": 0.0017174333333969116,
+      "learning_rate": 7.288215806909608e-06,
+      "loss": 0.0571,
+      "step": 6120
+    },
+    {
+      "epoch": 5.04,
+      "grad_norm": 0.006595561280846596,
+      "learning_rate": 7.2093390124625336e-06,
+      "loss": 0.0006,
+      "step": 6130
+    },
+    {
+      "epoch": 5.04,
+      "grad_norm": 0.0009397428948432207,
+      "learning_rate": 7.13046221801546e-06,
+      "loss": 0.0013,
+      "step": 6140
+    },
+    {
+      "epoch": 5.04,
+      "grad_norm": 0.001213430892676115,
+      "learning_rate": 7.051585423568387e-06,
+      "loss": 0.0694,
+      "step": 6150
+    },
+    {
+      "epoch": 5.04,
+      "grad_norm": 0.0012336999643594027,
+      "learning_rate": 6.972708629121314e-06,
+      "loss": 0.0,
+      "step": 6160
+    },
+    {
+      "epoch": 5.04,
+      "grad_norm": 0.000609008886385709,
+      "learning_rate": 6.893831834674238e-06,
+      "loss": 0.0,
+      "step": 6170
+    },
+    {
+      "epoch": 5.04,
+      "grad_norm": 0.0030849073082208633,
+      "learning_rate": 6.814955040227165e-06,
+      "loss": 0.0017,
+      "step": 6180
+    },
+    {
+      "epoch": 5.05,
+      "grad_norm": 0.0020494009368121624,
+      "learning_rate": 6.736078245780092e-06,
+      "loss": 0.0001,
+      "step": 6190
+    },
+    {
+      "epoch": 5.05,
+      "grad_norm": 0.0008749588741920888,
+      "learning_rate": 6.6572014513330185e-06,
+      "loss": 0.0064,
+      "step": 6200
+    },
+    {
+      "epoch": 5.05,
+      "grad_norm": 0.0018605308141559362,
+      "learning_rate": 6.578324656885944e-06,
+      "loss": 0.0002,
+      "step": 6210
+    },
+    {
+      "epoch": 5.05,
+      "grad_norm": 0.0016896320739760995,
+      "learning_rate": 6.499447862438871e-06,
+      "loss": 0.0001,
+      "step": 6220
+    },
+    {
+      "epoch": 5.05,
+      "grad_norm": 0.0036882127169519663,
+      "learning_rate": 6.420571067991797e-06,
+      "loss": 0.0005,
+      "step": 6230
+    },
+    {
+      "epoch": 5.05,
+      "grad_norm": 0.0008862425456754863,
+      "learning_rate": 6.341694273544724e-06,
+      "loss": 0.0,
+      "step": 6240
+    },
+    {
+      "epoch": 5.05,
+      "grad_norm": 0.0007071091677062213,
+      "learning_rate": 6.26281747909765e-06,
+      "loss": 0.1574,
+      "step": 6250
+    },
+    {
+      "epoch": 5.06,
+      "grad_norm": 0.0014393229503184557,
+      "learning_rate": 6.183940684650576e-06,
+      "loss": 0.0006,
+      "step": 6260
+    },
+    {
+      "epoch": 5.06,
+      "grad_norm": 0.01002194732427597,
+      "learning_rate": 6.1050638902035025e-06,
+      "loss": 1.4686,
+      "step": 6270
+    },
+    {
+      "epoch": 5.06,
+      "grad_norm": 0.003706766525283456,
+      "learning_rate": 6.026187095756429e-06,
+      "loss": 0.0001,
+      "step": 6280
+    },
+    {
+      "epoch": 5.06,
+      "grad_norm": 0.0018802026752382517,
+      "learning_rate": 5.947310301309355e-06,
+      "loss": 0.0001,
+      "step": 6290
+    },
+    {
+      "epoch": 5.06,
+      "grad_norm": 0.004596124403178692,
+      "learning_rate": 5.868433506862282e-06,
+      "loss": 0.1912,
+      "step": 6300
+    },
+    {
+      "epoch": 5.06,
+      "grad_norm": 0.004123110324144363,
+      "learning_rate": 5.789556712415207e-06,
+      "loss": 0.0001,
+      "step": 6310
+    },
+    {
+      "epoch": 5.06,
+      "grad_norm": 0.001596176065504551,
+      "learning_rate": 5.710679917968134e-06,
+      "loss": 0.0001,
+      "step": 6320
+    },
+    {
+      "epoch": 5.07,
+      "grad_norm": 0.0048742955550551414,
+      "learning_rate": 5.63180312352106e-06,
+      "loss": 0.0015,
+      "step": 6330
+    },
+    {
+      "epoch": 5.07,
+      "grad_norm": 0.003633652813732624,
+      "learning_rate": 5.5529263290739866e-06,
+      "loss": 0.0001,
+      "step": 6340
+    },
+    {
+      "epoch": 5.07,
+      "grad_norm": 0.0006397954421117902,
+      "learning_rate": 5.474049534626913e-06,
+      "loss": 0.0002,
+      "step": 6350
+    },
+    {
+      "epoch": 5.07,
+      "grad_norm": 0.0008470194879919291,
+      "learning_rate": 5.39517274017984e-06,
+      "loss": 0.0001,
+      "step": 6360
+    },
+    {
+      "epoch": 5.07,
+      "grad_norm": 0.004509842023253441,
+      "learning_rate": 5.316295945732766e-06,
+      "loss": 0.0001,
+      "step": 6370
+    },
+    {
+      "epoch": 5.07,
+      "grad_norm": 0.0010576475178822875,
+      "learning_rate": 5.237419151285692e-06,
+      "loss": 0.0009,
+      "step": 6380
+    },
+    {
+      "epoch": 5.07,
+      "grad_norm": 0.0009775225771591067,
+      "learning_rate": 5.158542356838618e-06,
+      "loss": 0.0001,
+      "step": 6390
+    },
+    {
+      "epoch": 5.08,
+      "grad_norm": 0.0007809589733369648,
+      "learning_rate": 5.0796655623915444e-06,
+      "loss": 0.0001,
+      "step": 6400
+    },
+    {
+      "epoch": 5.08,
+      "grad_norm": 0.0018230247078463435,
+      "learning_rate": 5.0007887679444714e-06,
+      "loss": 0.0001,
+      "step": 6410
+    },
+    {
+      "epoch": 5.08,
+      "grad_norm": 0.0014742623316124082,
+      "learning_rate": 4.921911973497397e-06,
+      "loss": 0.0,
+      "step": 6420
+    },
+    {
+      "epoch": 5.08,
+      "grad_norm": 0.0010474827140569687,
+      "learning_rate": 4.843035179050324e-06,
+      "loss": 0.0017,
+      "step": 6430
+    },
+    {
+      "epoch": 5.08,
+      "grad_norm": 0.009958957321941853,
+      "learning_rate": 4.76415838460325e-06,
+      "loss": 0.0001,
+      "step": 6440
+    },
+    {
+      "epoch": 5.08,
+      "grad_norm": 0.0023771177511662245,
+      "learning_rate": 4.685281590156176e-06,
+      "loss": 0.0001,
+      "step": 6450
+    },
+    {
+      "epoch": 5.08,
+      "grad_norm": 0.0008558974368497729,
+      "learning_rate": 4.606404795709102e-06,
+      "loss": 0.0,
+      "step": 6460
+    },
+    {
+      "epoch": 5.09,
+      "grad_norm": 0.0010758903808891773,
+      "learning_rate": 4.527528001262029e-06,
+      "loss": 0.0017,
+      "step": 6470
+    },
+    {
+      "epoch": 5.09,
+      "grad_norm": 0.001098488923162222,
+      "learning_rate": 4.448651206814955e-06,
+      "loss": 0.0,
+      "step": 6480
+    },
+    {
+      "epoch": 5.09,
+      "grad_norm": 0.0007584207342006266,
+      "learning_rate": 4.369774412367882e-06,
+      "loss": 0.0001,
+      "step": 6490
+    },
+    {
+      "epoch": 5.09,
+      "grad_norm": 0.0009535447461530566,
+      "learning_rate": 4.290897617920808e-06,
+      "loss": 0.0001,
+      "step": 6500
+    },
+    {
+      "epoch": 5.09,
+      "grad_norm": 0.0011546735186129808,
+      "learning_rate": 4.212020823473734e-06,
+      "loss": 0.0005,
+      "step": 6510
+    },
+    {
+      "epoch": 5.09,
+      "grad_norm": 0.0004957587807439268,
+      "learning_rate": 4.133144029026661e-06,
+      "loss": 0.001,
+      "step": 6520
+    },
+    {
+      "epoch": 5.09,
+      "grad_norm": 0.0012522251345217228,
+      "learning_rate": 4.054267234579587e-06,
+      "loss": 0.0001,
+      "step": 6530
+    },
+    {
+      "epoch": 5.1,
+      "grad_norm": 0.0006259087822400033,
+      "learning_rate": 3.975390440132513e-06,
+      "loss": 0.0001,
+      "step": 6540
+    },
+    {
+      "epoch": 5.1,
+      "grad_norm": 0.004241630434989929,
+      "learning_rate": 3.8965136456854395e-06,
+      "loss": 0.0001,
+      "step": 6550
+    },
+    {
+      "epoch": 5.1,
+      "grad_norm": 0.000727738777641207,
+      "learning_rate": 3.817636851238366e-06,
+      "loss": 0.0,
+      "step": 6560
+    },
+    {
+      "epoch": 5.1,
+      "grad_norm": 0.0011699842289090157,
+      "learning_rate": 3.738760056791292e-06,
+      "loss": 0.0001,
+      "step": 6570
+    },
+    {
+      "epoch": 5.1,
+      "grad_norm": 0.0008908452000468969,
+      "learning_rate": 3.6598832623442185e-06,
+      "loss": 0.0,
+      "step": 6580
+    },
+    {
+      "epoch": 5.1,
+      "grad_norm": 0.0008112391224130988,
+      "learning_rate": 3.5810064678971446e-06,
+      "loss": 0.2038,
+      "step": 6590
+    },
+    {
+      "epoch": 5.1,
+      "grad_norm": 0.006017233710736036,
+      "learning_rate": 3.5021296734500712e-06,
+      "loss": 0.0026,
+      "step": 6600
+    },
+    {
+      "epoch": 5.11,
+      "grad_norm": 0.0013324839528650045,
+      "learning_rate": 3.423252879002997e-06,
+      "loss": 0.0,
+      "step": 6610
+    },
+    {
+      "epoch": 5.11,
+      "grad_norm": 0.0020767981186509132,
+      "learning_rate": 3.344376084555924e-06,
+      "loss": 0.0001,
+      "step": 6620
+    },
+    {
+      "epoch": 5.11,
+      "grad_norm": 21.080284118652344,
+      "learning_rate": 3.2654992901088506e-06,
+      "loss": 0.1978,
+      "step": 6630
+    },
+    {
+      "epoch": 5.11,
+      "grad_norm": 0.002964046085253358,
+      "learning_rate": 3.1866224956617763e-06,
+      "loss": 0.0001,
+      "step": 6640
+    },
+    {
+      "epoch": 5.11,
+      "grad_norm": 0.0009833782678470016,
+      "learning_rate": 3.107745701214703e-06,
+      "loss": 0.0,
+      "step": 6650
+    },
+    {
+      "epoch": 5.11,
+      "grad_norm": 1.337567687034607,
+      "learning_rate": 3.028868906767629e-06,
+      "loss": 0.0007,
+      "step": 6660
+    },
+    {
+      "epoch": 5.11,
+      "grad_norm": 0.000990193453617394,
+      "learning_rate": 2.9499921123205553e-06,
+      "loss": 0.0,
+      "step": 6670
+    },
+    {
+      "epoch": 5.11,
+      "grad_norm": 0.004708257038146257,
+      "learning_rate": 2.8711153178734814e-06,
+      "loss": 0.0001,
+      "step": 6680
+    },
+    {
+      "epoch": 5.12,
+      "grad_norm": 0.0014612562954425812,
+      "learning_rate": 2.792238523426408e-06,
+      "loss": 0.0001,
+      "step": 6690
+    },
+    {
+      "epoch": 5.12,
+      "grad_norm": 0.0018317296635359526,
+      "learning_rate": 2.7133617289793346e-06,
+      "loss": 0.0,
+      "step": 6700
+    },
+    {
+      "epoch": 5.12,
+      "grad_norm": 0.0012669299030676484,
+      "learning_rate": 2.634484934532261e-06,
+      "loss": 0.0,
+      "step": 6710
+    },
+    {
+      "epoch": 5.12,
+      "grad_norm": 0.018809571862220764,
+      "learning_rate": 2.555608140085187e-06,
+      "loss": 0.2779,
+      "step": 6720
+    },
+    {
+      "epoch": 5.12,
+      "grad_norm": 0.0017046189168468118,
+      "learning_rate": 2.4767313456381136e-06,
+      "loss": 0.0001,
+      "step": 6730
+    },
+    {
+      "epoch": 5.12,
+      "grad_norm": 1.444283366203308,
+      "learning_rate": 2.3978545511910397e-06,
+      "loss": 0.0009,
+      "step": 6740
+    },
+    {
+      "epoch": 5.12,
+      "grad_norm": 0.011415842920541763,
+      "learning_rate": 2.318977756743966e-06,
+      "loss": 0.0001,
+      "step": 6750
+    },
+    {
+      "epoch": 5.13,
+      "grad_norm": 0.0009098586742766201,
+      "learning_rate": 2.2401009622968925e-06,
+      "loss": 0.0001,
+      "step": 6760
+    },
+    {
+      "epoch": 5.13,
+      "grad_norm": 0.08870714157819748,
+      "learning_rate": 2.1612241678498187e-06,
+      "loss": 0.0003,
+      "step": 6770
+    },
+    {
+      "epoch": 5.13,
+      "grad_norm": 0.0012093151453882456,
+      "learning_rate": 2.082347373402745e-06,
+      "loss": 0.0001,
+      "step": 6780
+    },
+    {
+      "epoch": 5.13,
+      "grad_norm": 20.205219268798828,
+      "learning_rate": 2.0034705789556714e-06,
+      "loss": 0.1622,
+      "step": 6790
+    },
+    {
+      "epoch": 5.13,
+      "grad_norm": 0.0017173507949337363,
+      "learning_rate": 1.9245937845085976e-06,
+      "loss": 0.0003,
+      "step": 6800
+    },
+    {
+      "epoch": 5.13,
+      "grad_norm": 0.0005491850897669792,
+      "learning_rate": 1.8457169900615238e-06,
+      "loss": 0.0,
+      "step": 6810
+    },
+    {
+      "epoch": 5.13,
+      "grad_norm": 0.0018508133944123983,
+      "learning_rate": 1.7668401956144502e-06,
+      "loss": 0.271,
+      "step": 6820
+    },
+    {
+      "epoch": 5.14,
+      "grad_norm": 0.003221946069970727,
+      "learning_rate": 1.6879634011673768e-06,
+      "loss": 0.2781,
+      "step": 6830
+    },
+    {
+      "epoch": 5.14,
+      "grad_norm": 0.0021298727951943874,
+      "learning_rate": 1.6090866067203031e-06,
+      "loss": 0.0001,
+      "step": 6840
+    },
+    {
+      "epoch": 5.14,
+      "grad_norm": 0.0011486714938655496,
+      "learning_rate": 1.5302098122732293e-06,
+      "loss": 0.4311,
+      "step": 6850
+    },
+    {
+      "epoch": 5.14,
+      "grad_norm": 0.0011125507298856974,
+      "learning_rate": 1.4513330178261557e-06,
+      "loss": 0.0001,
+      "step": 6860
+    },
+    {
+      "epoch": 5.14,
+      "grad_norm": 26.37517547607422,
+      "learning_rate": 1.372456223379082e-06,
+      "loss": 0.26,
+      "step": 6870
+    },
+    {
+      "epoch": 5.14,
+      "grad_norm": 0.0004922157968394458,
+      "learning_rate": 1.2935794289320083e-06,
+      "loss": 0.0002,
+      "step": 6880
+    },
+    {
+      "epoch": 5.14,
+      "grad_norm": 0.0013155548367649317,
+      "learning_rate": 1.2147026344849346e-06,
+      "loss": 0.0,
+      "step": 6890
+    },
+    {
+      "epoch": 5.15,
+      "grad_norm": 0.0012173574650660157,
+      "learning_rate": 1.1358258400378608e-06,
+      "loss": 0.0718,
+      "step": 6900
+    },
+    {
+      "epoch": 5.15,
+      "grad_norm": 0.0024919668212532997,
+      "learning_rate": 1.0569490455907872e-06,
+      "loss": 0.0001,
+      "step": 6910
+    },
+    {
+      "epoch": 5.15,
+      "grad_norm": 0.002818644279614091,
+      "learning_rate": 9.780722511437136e-07,
+      "loss": 0.0001,
+      "step": 6920
+    },
+    {
+      "epoch": 5.15,
+      "grad_norm": 0.0013261339627206326,
+      "learning_rate": 8.9919545669664e-07,
+      "loss": 0.0001,
+      "step": 6930
+    },
+    {
+      "epoch": 5.15,
+      "grad_norm": 0.0017504541901871562,
+      "learning_rate": 8.203186622495662e-07,
+      "loss": 0.0001,
+      "step": 6940
+    },
+    {
+      "epoch": 5.15,
+      "grad_norm": 0.0027923909947276115,
+      "learning_rate": 7.414418678024925e-07,
+      "loss": 0.0005,
+      "step": 6950
+    },
+    {
+      "epoch": 5.15,
+      "grad_norm": 0.0008931371266953647,
+      "learning_rate": 6.625650733554189e-07,
+      "loss": 0.1283,
+      "step": 6960
+    },
+    {
+      "epoch": 5.16,
+      "grad_norm": 0.0017122598364949226,
+      "learning_rate": 5.836882789083453e-07,
+      "loss": 0.0,
+      "step": 6970
+    },
+    {
+      "epoch": 5.16,
+      "grad_norm": 0.0009121019975282252,
+      "learning_rate": 5.048114844612716e-07,
+      "loss": 0.0002,
+      "step": 6980
+    },
+    {
+      "epoch": 5.16,
+      "grad_norm": 0.0010635430226102471,
+      "learning_rate": 4.2593469001419783e-07,
+      "loss": 0.0,
+      "step": 6990
+    },
+    {
+      "epoch": 5.16,
+      "grad_norm": 0.0023219590075314045,
+      "learning_rate": 3.4705789556712416e-07,
+      "loss": 0.0018,
+      "step": 7000
+    },
+    {
+      "epoch": 5.16,
+      "grad_norm": 0.0009701193775981665,
+      "learning_rate": 2.681811011200505e-07,
+      "loss": 0.0004,
+      "step": 7010
+    },
+    {
+      "epoch": 5.16,
+      "grad_norm": 0.037384454160928726,
+      "learning_rate": 1.8930430667297682e-07,
+      "loss": 0.0858,
+      "step": 7020
+    },
+    {
+      "epoch": 5.16,
+      "grad_norm": 0.0022565952967852354,
+      "learning_rate": 1.1042751222590314e-07,
+      "loss": 0.0017,
+      "step": 7030
+    },
+    {
+      "epoch": 5.17,
+      "grad_norm": 0.0068207960575819016,
+      "learning_rate": 3.155071777882947e-08,
+      "loss": 0.125,
+      "step": 7040
+    },
+    {
+      "epoch": 5.17,
+      "eval_accuracy": 0.9616935483870968,
+      "eval_loss": 0.19818592071533203,
+      "eval_runtime": 1687.2131,
+      "eval_samples_per_second": 0.294,
+      "eval_steps_per_second": 0.147,
+      "step": 7044
+    },
+    {
+      "epoch": 5.17,
+      "step": 7044,
+      "total_flos": 3.6095062922884547e+19,
+      "train_loss": 0.29668668898395717,
+      "train_runtime": 69009.6757,
+      "train_samples_per_second": 0.204,
+      "train_steps_per_second": 0.102
+    },
+    {
+      "epoch": 5.17,
+      "eval_accuracy": 0.9739776951672863,
+      "eval_loss": 0.11183885484933853,
+      "eval_runtime": 927.8237,
+      "eval_samples_per_second": 0.29,
+      "eval_steps_per_second": 0.146,
+      "step": 7044
+    },
+    {
+      "epoch": 5.17,
+      "eval_accuracy": 0.9739776951672863,
+      "eval_loss": 0.11183883994817734,
+      "eval_runtime": 940.074,
+      "eval_samples_per_second": 0.286,
+      "eval_steps_per_second": 0.144,
+      "step": 7044
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 7044,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "save_steps": 500,
+  "total_flos": 3.6095062922884547e+19,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}