diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,961 +1,4705 @@
 {
-  "best_metric": 0.8342792281498297,
-  "best_model_checkpoint": "resnet-50-bottomCleanedData/checkpoint-1410",
-  "epoch": 9.959636730575177,
-  "global_step": 1410,
+  "best_metric": 0.9761634506242906,
+  "best_model_checkpoint": "resnet-50-bottomCleanedData/checkpoint-5946",
+  "epoch": 49.798183652875885,
+  "global_step": 7050,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.07,
-      "learning_rate": 3.3333333333333335e-05,
-      "loss": 1.3868,
+      "learning_rate": 7.042253521126762e-06,
+      "loss": 1.3922,
       "step": 10
     },
     {
       "epoch": 0.14,
-      "learning_rate": 4.982078853046595e-05,
-      "loss": 1.3831,
+      "learning_rate": 1.4084507042253523e-05,
+      "loss": 1.3925,
       "step": 20
     },
     {
       "epoch": 0.21,
-      "learning_rate": 4.9462365591397855e-05,
-      "loss": 1.3793,
+      "learning_rate": 2.112676056338028e-05,
+      "loss": 1.3875,
       "step": 30
     },
     {
       "epoch": 0.28,
-      "learning_rate": 4.910394265232976e-05,
-      "loss": 1.3799,
+      "learning_rate": 2.8169014084507046e-05,
+      "loss": 1.3867,
       "step": 40
     },
     {
       "epoch": 0.35,
-      "learning_rate": 4.874551971326165e-05,
-      "loss": 1.3732,
+      "learning_rate": 3.5211267605633805e-05,
+      "loss": 1.3887,
       "step": 50
     },
     {
       "epoch": 0.42,
-      "learning_rate": 4.8387096774193554e-05,
-      "loss": 1.3681,
+      "learning_rate": 4.225352112676056e-05,
+      "loss": 1.3804,
       "step": 60
     },
     {
       "epoch": 0.49,
-      "learning_rate": 4.802867383512545e-05,
-      "loss": 1.3693,
+      "learning_rate": 4.929577464788733e-05,
+      "loss": 1.3804,
       "step": 70
     },
     {
       "epoch": 0.57,
-      "learning_rate": 4.767025089605735e-05,
-      "loss": 1.3644,
+      "learning_rate": 4.993552084825906e-05,
+      "loss": 1.3758,
       "step": 80
     },
     {
       "epoch": 0.64,
-      "learning_rate": 4.731182795698925e-05,
-      "loss": 1.3626,
+      "learning_rate": 4.9863877346324687e-05,
+      "loss": 1.3704,
       "step": 90
     },
     {
       "epoch": 0.71,
-      "learning_rate": 4.695340501792115e-05,
-      "loss": 1.3576,
+      "learning_rate": 4.979223384439032e-05,
+      "loss": 1.3653,
       "step": 100
     },
     {
       "epoch": 0.78,
-      "learning_rate": 4.659498207885305e-05,
-      "loss": 1.348,
+      "learning_rate": 4.972059034245594e-05,
+      "loss": 1.3534,
       "step": 110
     },
     {
       "epoch": 0.85,
-      "learning_rate": 4.6236559139784944e-05,
-      "loss": 1.34,
+      "learning_rate": 4.9648946840521565e-05,
+      "loss": 1.3509,
       "step": 120
     },
     {
       "epoch": 0.92,
-      "learning_rate": 4.5878136200716846e-05,
-      "loss": 1.3357,
+      "learning_rate": 4.9577303338587196e-05,
+      "loss": 1.3496,
       "step": 130
     },
     {
       "epoch": 0.99,
-      "learning_rate": 4.551971326164875e-05,
-      "loss": 1.3235,
+      "learning_rate": 4.950565983665282e-05,
+      "loss": 1.3323,
       "step": 140
     },
     {
       "epoch": 1.0,
-      "eval_accuracy": 0.5096481271282634,
-      "eval_loss": 1.3266013860702515,
-      "eval_runtime": 648.045,
-      "eval_samples_per_second": 1.359,
-      "eval_steps_per_second": 0.171,
+      "eval_accuracy": 0.5187287173666288,
+      "eval_loss": 1.3318836688995361,
+      "eval_runtime": 6.1537,
+      "eval_samples_per_second": 143.166,
+      "eval_steps_per_second": 18.038,
       "step": 141
     },
     {
       "epoch": 1.06,
-      "learning_rate": 4.516129032258064e-05,
-      "loss": 1.3128,
+      "learning_rate": 4.9434016334718444e-05,
+      "loss": 1.3211,
       "step": 150
     },
     {
       "epoch": 1.13,
-      "learning_rate": 4.4802867383512545e-05,
-      "loss": 1.311,
+      "learning_rate": 4.936237283278407e-05,
+      "loss": 1.3129,
       "step": 160
     },
     {
       "epoch": 1.2,
-      "learning_rate": 4.4444444444444447e-05,
-      "loss": 1.2957,
+      "learning_rate": 4.929072933084969e-05,
+      "loss": 1.2989,
       "step": 170
     },
     {
       "epoch": 1.27,
-      "learning_rate": 4.408602150537635e-05,
-      "loss": 1.2956,
+      "learning_rate": 4.9219085828915316e-05,
+      "loss": 1.2976,
       "step": 180
     },
     {
       "epoch": 1.34,
-      "learning_rate": 4.372759856630825e-05,
-      "loss": 1.2798,
+      "learning_rate": 4.9147442326980947e-05,
+      "loss": 1.2906,
       "step": 190
     },
     {
       "epoch": 1.41,
-      "learning_rate": 4.3369175627240145e-05,
-      "loss": 1.2693,
+      "learning_rate": 4.907579882504657e-05,
+      "loss": 1.2608,
       "step": 200
     },
     {
       "epoch": 1.48,
-      "learning_rate": 4.301075268817205e-05,
-      "loss": 1.265,
+      "learning_rate": 4.9004155323112194e-05,
+      "loss": 1.2644,
       "step": 210
     },
     {
       "epoch": 1.55,
-      "learning_rate": 4.265232974910394e-05,
-      "loss": 1.2554,
+      "learning_rate": 4.8932511821177825e-05,
+      "loss": 1.2446,
       "step": 220
     },
     {
       "epoch": 1.62,
-      "learning_rate": 4.2293906810035844e-05,
-      "loss": 1.2435,
+      "learning_rate": 4.886086831924345e-05,
+      "loss": 1.2314,
       "step": 230
     },
     {
       "epoch": 1.7,
-      "learning_rate": 4.1935483870967746e-05,
-      "loss": 1.2251,
+      "learning_rate": 4.878922481730907e-05,
+      "loss": 1.204,
       "step": 240
     },
     {
       "epoch": 1.77,
-      "learning_rate": 4.157706093189964e-05,
-      "loss": 1.2136,
+      "learning_rate": 4.87175813153747e-05,
+      "loss": 1.1636,
       "step": 250
     },
     {
       "epoch": 1.84,
-      "learning_rate": 4.121863799283154e-05,
-      "loss": 1.1784,
+      "learning_rate": 4.864593781344032e-05,
+      "loss": 1.1508,
       "step": 260
     },
     {
       "epoch": 1.91,
-      "learning_rate": 4.0860215053763444e-05,
-      "loss": 1.1691,
+      "learning_rate": 4.8574294311505945e-05,
+      "loss": 1.117,
       "step": 270
     },
     {
       "epoch": 1.98,
-      "learning_rate": 4.050179211469534e-05,
-      "loss": 1.1546,
+      "learning_rate": 4.8502650809571576e-05,
+      "loss": 1.1302,
       "step": 280
     },
     {
       "epoch": 2.0,
-      "eval_accuracy": 0.5153234960272418,
-      "eval_loss": 1.1380490064620972,
-      "eval_runtime": 7.1826,
-      "eval_samples_per_second": 122.658,
-      "eval_steps_per_second": 15.454,
+      "eval_accuracy": 0.5334846765039728,
+      "eval_loss": 1.1058534383773804,
+      "eval_runtime": 6.0856,
+      "eval_samples_per_second": 144.767,
+      "eval_steps_per_second": 18.24,
       "step": 283
     },
     {
       "epoch": 2.05,
-      "learning_rate": 4.014336917562724e-05,
-      "loss": 1.1353,
+      "learning_rate": 4.84310073076372e-05,
+      "loss": 1.118,
       "step": 290
     },
     {
       "epoch": 2.12,
-      "learning_rate": 3.978494623655914e-05,
-      "loss": 1.1329,
+      "learning_rate": 4.8359363805702824e-05,
+      "loss": 1.0764,
       "step": 300
     },
     {
       "epoch": 2.19,
-      "learning_rate": 3.9426523297491045e-05,
-      "loss": 1.1246,
+      "learning_rate": 4.8287720303768454e-05,
+      "loss": 1.0558,
       "step": 310
     },
     {
       "epoch": 2.26,
-      "learning_rate": 3.906810035842295e-05,
-      "loss": 1.0971,
+      "learning_rate": 4.821607680183408e-05,
+      "loss": 1.0255,
       "step": 320
     },
     {
       "epoch": 2.33,
-      "learning_rate": 3.870967741935484e-05,
-      "loss": 1.0763,
+      "learning_rate": 4.81444332998997e-05,
+      "loss": 1.0148,
       "step": 330
     },
     {
       "epoch": 2.4,
-      "learning_rate": 3.8351254480286743e-05,
-      "loss": 1.0741,
+      "learning_rate": 4.8072789797965326e-05,
+      "loss": 0.9665,
       "step": 340
     },
     {
       "epoch": 2.47,
-      "learning_rate": 3.799283154121864e-05,
-      "loss": 1.0241,
+      "learning_rate": 4.800114629603095e-05,
+      "loss": 0.9797,
       "step": 350
     },
     {
       "epoch": 2.54,
-      "learning_rate": 3.763440860215054e-05,
-      "loss": 1.0103,
+      "learning_rate": 4.7929502794096574e-05,
+      "loss": 0.9602,
       "step": 360
     },
     {
       "epoch": 2.61,
-      "learning_rate": 3.727598566308244e-05,
-      "loss": 1.0049,
+      "learning_rate": 4.7857859292162205e-05,
+      "loss": 0.9274,
       "step": 370
     },
     {
       "epoch": 2.68,
-      "learning_rate": 3.691756272401434e-05,
-      "loss": 0.9883,
+      "learning_rate": 4.778621579022783e-05,
+      "loss": 0.917,
       "step": 380
     },
     {
       "epoch": 2.75,
-      "learning_rate": 3.655913978494624e-05,
-      "loss": 0.9864,
+      "learning_rate": 4.771457228829345e-05,
+      "loss": 0.9209,
       "step": 390
     },
     {
       "epoch": 2.83,
-      "learning_rate": 3.6200716845878134e-05,
-      "loss": 0.9871,
+      "learning_rate": 4.7642928786359084e-05,
+      "loss": 0.8891,
       "step": 400
     },
     {
       "epoch": 2.9,
-      "learning_rate": 3.5842293906810036e-05,
-      "loss": 0.9547,
+      "learning_rate": 4.757128528442471e-05,
+      "loss": 0.889,
       "step": 410
     },
     {
       "epoch": 2.97,
-      "learning_rate": 3.548387096774194e-05,
-      "loss": 0.9412,
+      "learning_rate": 4.749964178249033e-05,
+      "loss": 0.8793,
       "step": 420
     },
     {
       "epoch": 2.99,
-      "eval_accuracy": 0.6515323496027242,
-      "eval_loss": 0.8690479397773743,
-      "eval_runtime": 6.9421,
-      "eval_samples_per_second": 126.907,
-      "eval_steps_per_second": 15.989,
+      "eval_accuracy": 0.7094211123723042,
+      "eval_loss": 0.7847815155982971,
+      "eval_runtime": 6.1857,
+      "eval_samples_per_second": 142.425,
+      "eval_steps_per_second": 17.945,
       "step": 424
     },
     {
       "epoch": 3.04,
-      "learning_rate": 3.512544802867383e-05,
-      "loss": 0.9049,
+      "learning_rate": 4.7427998280555955e-05,
+      "loss": 0.886,
       "step": 430
     },
     {
       "epoch": 3.11,
-      "learning_rate": 3.4767025089605734e-05,
-      "loss": 0.9843,
+      "learning_rate": 4.735635477862158e-05,
+      "loss": 0.8801,
       "step": 440
     },
     {
       "epoch": 3.18,
-      "learning_rate": 3.4408602150537636e-05,
-      "loss": 0.9074,
+      "learning_rate": 4.72847112766872e-05,
+      "loss": 0.8465,
       "step": 450
     },
     {
       "epoch": 3.25,
-      "learning_rate": 3.405017921146954e-05,
-      "loss": 0.9251,
+      "learning_rate": 4.7213067774752834e-05,
+      "loss": 0.8312,
       "step": 460
     },
     {
       "epoch": 3.32,
-      "learning_rate": 3.369175627240144e-05,
-      "loss": 0.9401,
+      "learning_rate": 4.714142427281846e-05,
+      "loss": 0.8442,
       "step": 470
     },
     {
       "epoch": 3.39,
-      "learning_rate": 3.3333333333333335e-05,
-      "loss": 0.907,
+      "learning_rate": 4.706978077088408e-05,
+      "loss": 0.8356,
       "step": 480
     },
     {
       "epoch": 3.46,
-      "learning_rate": 3.297491039426524e-05,
-      "loss": 0.909,
+      "learning_rate": 4.699813726894971e-05,
+      "loss": 0.8252,
       "step": 490
     },
     {
       "epoch": 3.53,
-      "learning_rate": 3.261648745519714e-05,
-      "loss": 0.8676,
+      "learning_rate": 4.692649376701534e-05,
+      "loss": 0.8338,
       "step": 500
     },
     {
       "epoch": 3.6,
-      "learning_rate": 3.2258064516129034e-05,
-      "loss": 0.8435,
+      "learning_rate": 4.685485026508096e-05,
+      "loss": 0.8432,
       "step": 510
     },
     {
       "epoch": 3.67,
-      "learning_rate": 3.1899641577060935e-05,
-      "loss": 0.8634,
+      "learning_rate": 4.6783206763146585e-05,
+      "loss": 0.8556,
       "step": 520
     },
     {
       "epoch": 3.74,
-      "learning_rate": 3.154121863799283e-05,
-      "loss": 0.8802,
+      "learning_rate": 4.671156326121221e-05,
+      "loss": 0.7924,
       "step": 530
     },
     {
       "epoch": 3.81,
-      "learning_rate": 3.118279569892473e-05,
-      "loss": 0.8687,
+      "learning_rate": 4.663991975927783e-05,
+      "loss": 0.7948,
       "step": 540
     },
     {
       "epoch": 3.88,
-      "learning_rate": 3.0824372759856634e-05,
-      "loss": 0.8587,
+      "learning_rate": 4.656827625734346e-05,
+      "loss": 0.8202,
       "step": 550
     },
     {
       "epoch": 3.96,
-      "learning_rate": 3.046594982078853e-05,
-      "loss": 0.8539,
+      "learning_rate": 4.649663275540909e-05,
+      "loss": 0.7652,
       "step": 560
     },
     {
       "epoch": 4.0,
-      "eval_accuracy": 0.7593643586833144,
-      "eval_loss": 0.6672292351722717,
-      "eval_runtime": 7.0907,
-      "eval_samples_per_second": 124.247,
-      "eval_steps_per_second": 15.654,
+      "eval_accuracy": 0.7219069239500567,
+      "eval_loss": 0.7254674434661865,
+      "eval_runtime": 6.1097,
+      "eval_samples_per_second": 144.196,
+      "eval_steps_per_second": 18.168,
       "step": 566
     },
     {
       "epoch": 4.03,
-      "learning_rate": 3.010752688172043e-05,
-      "loss": 0.8467,
+      "learning_rate": 4.642498925347471e-05,
+      "loss": 0.7821,
       "step": 570
     },
     {
       "epoch": 4.1,
-      "learning_rate": 2.974910394265233e-05,
-      "loss": 0.8657,
+      "learning_rate": 4.635334575154034e-05,
+      "loss": 0.803,
       "step": 580
     },
     {
       "epoch": 4.17,
-      "learning_rate": 2.939068100358423e-05,
-      "loss": 0.8671,
+      "learning_rate": 4.6281702249605966e-05,
+      "loss": 0.7648,
       "step": 590
     },
     {
       "epoch": 4.24,
-      "learning_rate": 2.9032258064516133e-05,
-      "loss": 0.814,
+      "learning_rate": 4.621005874767158e-05,
+      "loss": 0.767,
       "step": 600
     },
     {
       "epoch": 4.31,
-      "learning_rate": 2.8673835125448028e-05,
-      "loss": 0.8287,
+      "learning_rate": 4.6138415245737214e-05,
+      "loss": 0.789,
       "step": 610
     },
     {
       "epoch": 4.38,
-      "learning_rate": 2.831541218637993e-05,
-      "loss": 0.8733,
+      "learning_rate": 4.606677174380284e-05,
+      "loss": 0.7957,
       "step": 620
     },
     {
       "epoch": 4.45,
-      "learning_rate": 2.7956989247311828e-05,
-      "loss": 0.8241,
+      "learning_rate": 4.599512824186846e-05,
+      "loss": 0.7661,
       "step": 630
     },
     {
       "epoch": 4.52,
-      "learning_rate": 2.759856630824373e-05,
-      "loss": 0.8157,
+      "learning_rate": 4.592348473993409e-05,
+      "loss": 0.7733,
       "step": 640
     },
     {
       "epoch": 4.59,
-      "learning_rate": 2.7240143369175632e-05,
-      "loss": 0.8597,
+      "learning_rate": 4.5851841237999716e-05,
+      "loss": 0.7391,
       "step": 650
     },
     {
       "epoch": 4.66,
-      "learning_rate": 2.6881720430107527e-05,
-      "loss": 0.8599,
+      "learning_rate": 4.578019773606534e-05,
+      "loss": 0.7193,
       "step": 660
     },
     {
       "epoch": 4.73,
-      "learning_rate": 2.652329749103943e-05,
-      "loss": 0.7635,
+      "learning_rate": 4.570855423413097e-05,
+      "loss": 0.7415,
       "step": 670
     },
     {
       "epoch": 4.8,
-      "learning_rate": 2.616487455197133e-05,
-      "loss": 0.8585,
+      "learning_rate": 4.563691073219659e-05,
+      "loss": 0.7346,
       "step": 680
     },
     {
       "epoch": 4.87,
-      "learning_rate": 2.5806451612903226e-05,
-      "loss": 0.8159,
+      "learning_rate": 4.556526723026221e-05,
+      "loss": 0.6951,
       "step": 690
     },
     {
       "epoch": 4.94,
-      "learning_rate": 2.5448028673835127e-05,
-      "loss": 0.7967,
+      "learning_rate": 4.549362372832784e-05,
+      "loss": 0.7708,
       "step": 700
     },
     {
       "epoch": 4.99,
-      "eval_accuracy": 0.7502837684449489,
-      "eval_loss": 0.625583827495575,
-      "eval_runtime": 7.1065,
-      "eval_samples_per_second": 123.971,
-      "eval_steps_per_second": 15.62,
+      "eval_accuracy": 0.8172531214528944,
+      "eval_loss": 0.5280147790908813,
+      "eval_runtime": 6.2297,
+      "eval_samples_per_second": 141.42,
+      "eval_steps_per_second": 17.818,
       "step": 707
     },
     {
       "epoch": 5.02,
-      "learning_rate": 2.5089605734767026e-05,
-      "loss": 0.807,
+      "learning_rate": 4.542198022639347e-05,
+      "loss": 0.7218,
       "step": 710
     },
     {
       "epoch": 5.09,
-      "learning_rate": 2.4731182795698928e-05,
-      "loss": 0.8029,
+      "learning_rate": 4.535033672445909e-05,
+      "loss": 0.7357,
       "step": 720
     },
     {
       "epoch": 5.16,
-      "learning_rate": 2.4372759856630826e-05,
-      "loss": 0.7848,
+      "learning_rate": 4.527869322252472e-05,
+      "loss": 0.6776,
       "step": 730
     },
     {
       "epoch": 5.23,
-      "learning_rate": 2.4014336917562724e-05,
-      "loss": 0.7913,
+      "learning_rate": 4.5207049720590346e-05,
+      "loss": 0.7348,
       "step": 740
     },
     {
       "epoch": 5.3,
-      "learning_rate": 2.3655913978494626e-05,
-      "loss": 0.8001,
+      "learning_rate": 4.513540621865597e-05,
+      "loss": 0.7038,
       "step": 750
     },
     {
       "epoch": 5.37,
-      "learning_rate": 2.3297491039426525e-05,
-      "loss": 0.7831,
+      "learning_rate": 4.50637627167216e-05,
+      "loss": 0.7089,
       "step": 760
     },
     {
       "epoch": 5.44,
-      "learning_rate": 2.2939068100358423e-05,
-      "loss": 0.8103,
+      "learning_rate": 4.499211921478722e-05,
+      "loss": 0.7122,
       "step": 770
     },
     {
       "epoch": 5.51,
-      "learning_rate": 2.258064516129032e-05,
-      "loss": 0.7803,
+      "learning_rate": 4.492047571285284e-05,
+      "loss": 0.6919,
       "step": 780
     },
     {
       "epoch": 5.58,
-      "learning_rate": 2.2222222222222223e-05,
-      "loss": 0.7639,
+      "learning_rate": 4.484883221091847e-05,
+      "loss": 0.6947,
       "step": 790
     },
     {
       "epoch": 5.65,
-      "learning_rate": 2.1863799283154125e-05,
-      "loss": 0.7671,
+      "learning_rate": 4.4777188708984096e-05,
+      "loss": 0.722,
       "step": 800
     },
     {
       "epoch": 5.72,
-      "learning_rate": 2.1505376344086024e-05,
-      "loss": 0.7457,
+      "learning_rate": 4.470554520704972e-05,
+      "loss": 0.67,
       "step": 810
     },
     {
       "epoch": 5.79,
-      "learning_rate": 2.1146953405017922e-05,
-      "loss": 0.8118,
+      "learning_rate": 4.463390170511535e-05,
+      "loss": 0.6491,
       "step": 820
     },
     {
       "epoch": 5.86,
-      "learning_rate": 2.078853046594982e-05,
-      "loss": 0.7935,
+      "learning_rate": 4.4562258203180975e-05,
+      "loss": 0.6226,
       "step": 830
     },
     {
       "epoch": 5.93,
-      "learning_rate": 2.0430107526881722e-05,
-      "loss": 0.7679,
+      "learning_rate": 4.44906147012466e-05,
+      "loss": 0.6153,
       "step": 840
     },
     {
       "epoch": 6.0,
-      "eval_accuracy": 0.8229284903518729,
-      "eval_loss": 0.5356738567352295,
-      "eval_runtime": 6.9588,
-      "eval_samples_per_second": 126.603,
-      "eval_steps_per_second": 15.951,
+      "eval_accuracy": 0.8490351872871736,
+      "eval_loss": 0.4221389889717102,
+      "eval_runtime": 6.1027,
+      "eval_samples_per_second": 144.361,
+      "eval_steps_per_second": 18.189,
       "step": 849
     },
     {
       "epoch": 6.0,
-      "learning_rate": 2.007168458781362e-05,
-      "loss": 0.804,
+      "learning_rate": 4.441897119931222e-05,
+      "loss": 0.6918,
       "step": 850
     },
     {
       "epoch": 6.07,
-      "learning_rate": 1.9713261648745522e-05,
-      "loss": 0.7701,
+      "learning_rate": 4.434732769737785e-05,
+      "loss": 0.623,
       "step": 860
     },
     {
       "epoch": 6.15,
-      "learning_rate": 1.935483870967742e-05,
-      "loss": 0.7442,
+      "learning_rate": 4.427568419544347e-05,
+      "loss": 0.6691,
       "step": 870
     },
     {
       "epoch": 6.22,
-      "learning_rate": 1.899641577060932e-05,
-      "loss": 0.7167,
+      "learning_rate": 4.42040406935091e-05,
+      "loss": 0.692,
       "step": 880
     },
     {
       "epoch": 6.29,
-      "learning_rate": 1.863799283154122e-05,
-      "loss": 0.7309,
+      "learning_rate": 4.4132397191574725e-05,
+      "loss": 0.6622,
       "step": 890
     },
     {
       "epoch": 6.36,
-      "learning_rate": 1.827956989247312e-05,
-      "loss": 0.7587,
+      "learning_rate": 4.406075368964035e-05,
+      "loss": 0.6513,
       "step": 900
     },
     {
       "epoch": 6.43,
-      "learning_rate": 1.7921146953405018e-05,
-      "loss": 0.781,
+      "learning_rate": 4.398911018770598e-05,
+      "loss": 0.6774,
       "step": 910
     },
     {
       "epoch": 6.5,
-      "learning_rate": 1.7562724014336916e-05,
-      "loss": 0.785,
+      "learning_rate": 4.3917466685771604e-05,
+      "loss": 0.6543,
       "step": 920
     },
     {
       "epoch": 6.57,
-      "learning_rate": 1.7204301075268818e-05,
-      "loss": 0.7486,
+      "learning_rate": 4.384582318383723e-05,
+      "loss": 0.6502,
       "step": 930
     },
     {
       "epoch": 6.64,
-      "learning_rate": 1.684587813620072e-05,
-      "loss": 0.7597,
+      "learning_rate": 4.377417968190285e-05,
+      "loss": 0.6189,
       "step": 940
     },
     {
       "epoch": 6.71,
-      "learning_rate": 1.648745519713262e-05,
-      "loss": 0.6813,
+      "learning_rate": 4.3702536179968476e-05,
+      "loss": 0.6758,
       "step": 950
     },
     {
       "epoch": 6.78,
-      "learning_rate": 1.6129032258064517e-05,
-      "loss": 0.75,
+      "learning_rate": 4.36308926780341e-05,
+      "loss": 0.5804,
       "step": 960
     },
     {
       "epoch": 6.85,
-      "learning_rate": 1.5770609318996415e-05,
-      "loss": 0.7543,
+      "learning_rate": 4.355924917609973e-05,
+      "loss": 0.6634,
       "step": 970
     },
     {
       "epoch": 6.92,
-      "learning_rate": 1.5412186379928317e-05,
-      "loss": 0.7112,
+      "learning_rate": 4.3487605674165355e-05,
+      "loss": 0.6442,
       "step": 980
     },
     {
       "epoch": 6.99,
-      "learning_rate": 1.5053763440860215e-05,
-      "loss": 0.7265,
+      "learning_rate": 4.341596217223098e-05,
+      "loss": 0.5895,
       "step": 990
     },
     {
       "epoch": 7.0,
-      "eval_accuracy": 0.7832009080590239,
-      "eval_loss": 0.5698212385177612,
-      "eval_runtime": 7.061,
-      "eval_samples_per_second": 124.77,
-      "eval_steps_per_second": 15.72,
+      "eval_accuracy": 0.8569807037457434,
+      "eval_loss": 0.4015471041202545,
+      "eval_runtime": 6.2885,
+      "eval_samples_per_second": 140.098,
+      "eval_steps_per_second": 17.651,
       "step": 991
     },
     {
       "epoch": 7.06,
-      "learning_rate": 1.4695340501792116e-05,
-      "loss": 0.7152,
+      "learning_rate": 4.334431867029661e-05,
+      "loss": 0.6638,
       "step": 1000
     },
     {
       "epoch": 7.13,
-      "learning_rate": 1.4336917562724014e-05,
-      "loss": 0.742,
+      "learning_rate": 4.327267516836223e-05,
+      "loss": 0.618,
       "step": 1010
     },
     {
       "epoch": 7.2,
-      "learning_rate": 1.3978494623655914e-05,
-      "loss": 0.742,
+      "learning_rate": 4.320103166642786e-05,
+      "loss": 0.6145,
       "step": 1020
     },
     {
       "epoch": 7.28,
-      "learning_rate": 1.3620071684587816e-05,
-      "loss": 0.7282,
+      "learning_rate": 4.312938816449348e-05,
+      "loss": 0.5624,
       "step": 1030
     },
     {
       "epoch": 7.35,
-      "learning_rate": 1.3261648745519714e-05,
-      "loss": 0.7105,
+      "learning_rate": 4.3057744662559105e-05,
+      "loss": 0.6239,
       "step": 1040
     },
     {
       "epoch": 7.42,
-      "learning_rate": 1.2903225806451613e-05,
-      "loss": 0.7683,
+      "learning_rate": 4.298610116062473e-05,
+      "loss": 0.6381,
       "step": 1050
     },
     {
       "epoch": 7.49,
-      "learning_rate": 1.2544802867383513e-05,
-      "loss": 0.7343,
+      "learning_rate": 4.291445765869036e-05,
+      "loss": 0.6004,
       "step": 1060
     },
     {
       "epoch": 7.56,
-      "learning_rate": 1.2186379928315413e-05,
-      "loss": 0.6884,
+      "learning_rate": 4.2842814156755984e-05,
+      "loss": 0.606,
       "step": 1070
     },
     {
       "epoch": 7.63,
-      "learning_rate": 1.1827956989247313e-05,
-      "loss": 0.7065,
+      "learning_rate": 4.277117065482161e-05,
+      "loss": 0.5705,
       "step": 1080
     },
     {
       "epoch": 7.7,
-      "learning_rate": 1.1469534050179212e-05,
-      "loss": 0.6961,
+      "learning_rate": 4.269952715288724e-05,
+      "loss": 0.5445,
       "step": 1090
     },
     {
       "epoch": 7.77,
-      "learning_rate": 1.1111111111111112e-05,
-      "loss": 0.7079,
+      "learning_rate": 4.262788365095286e-05,
+      "loss": 0.5553,
       "step": 1100
     },
     {
       "epoch": 7.84,
-      "learning_rate": 1.0752688172043012e-05,
-      "loss": 0.7012,
+      "learning_rate": 4.2556240149018486e-05,
+      "loss": 0.6153,
       "step": 1110
     },
     {
       "epoch": 7.91,
-      "learning_rate": 1.039426523297491e-05,
-      "loss": 0.633,
+      "learning_rate": 4.248459664708411e-05,
+      "loss": 0.5137,
       "step": 1120
     },
     {
       "epoch": 7.98,
-      "learning_rate": 1.003584229390681e-05,
-      "loss": 0.7395,
+      "learning_rate": 4.2412953145149734e-05,
+      "loss": 0.5617,
       "step": 1130
     },
     {
       "epoch": 8.0,
-      "eval_accuracy": 0.8161180476730987,
-      "eval_loss": 0.5125163197517395,
-      "eval_runtime": 6.6771,
-      "eval_samples_per_second": 131.944,
-      "eval_steps_per_second": 16.624,
+      "eval_accuracy": 0.9001135073779796,
+      "eval_loss": 0.29984819889068604,
+      "eval_runtime": 6.164,
+      "eval_samples_per_second": 142.926,
+      "eval_steps_per_second": 18.008,
       "step": 1132
     },
     {
       "epoch": 8.05,
-      "learning_rate": 9.67741935483871e-06,
-      "loss": 0.7309,
+      "learning_rate": 4.234130964321536e-05,
+      "loss": 0.534,
       "step": 1140
     },
     {
       "epoch": 8.12,
-      "learning_rate": 9.31899641577061e-06,
-      "loss": 0.6818,
+      "learning_rate": 4.226966614128099e-05,
+      "loss": 0.5892,
       "step": 1150
     },
     {
       "epoch": 8.19,
-      "learning_rate": 8.960573476702509e-06,
-      "loss": 0.7509,
+      "learning_rate": 4.219802263934661e-05,
+      "loss": 0.5432,
       "step": 1160
     },
     {
       "epoch": 8.26,
-      "learning_rate": 8.602150537634409e-06,
-      "loss": 0.679,
+      "learning_rate": 4.212637913741224e-05,
+      "loss": 0.5921,
       "step": 1170
     },
     {
       "epoch": 8.34,
-      "learning_rate": 8.24372759856631e-06,
-      "loss": 0.7765,
+      "learning_rate": 4.205473563547787e-05,
+      "loss": 0.5802,
       "step": 1180
     },
     {
       "epoch": 8.41,
-      "learning_rate": 7.885304659498208e-06,
-      "loss": 0.7271,
+      "learning_rate": 4.198309213354349e-05,
+      "loss": 0.5153,
       "step": 1190
     },
     {
       "epoch": 8.48,
-      "learning_rate": 7.526881720430108e-06,
-      "loss": 0.6982,
+      "learning_rate": 4.1911448631609116e-05,
+      "loss": 0.5237,
       "step": 1200
     },
     {
       "epoch": 8.55,
-      "learning_rate": 7.168458781362007e-06,
-      "loss": 0.6922,
+      "learning_rate": 4.183980512967474e-05,
+      "loss": 0.5552,
       "step": 1210
     },
     {
       "epoch": 8.62,
-      "learning_rate": 6.810035842293908e-06,
-      "loss": 0.7429,
+      "learning_rate": 4.1768161627740364e-05,
+      "loss": 0.5363,
       "step": 1220
     },
     {
       "epoch": 8.69,
-      "learning_rate": 6.451612903225806e-06,
-      "loss": 0.7055,
+      "learning_rate": 4.169651812580599e-05,
+      "loss": 0.5509,
       "step": 1230
     },
     {
       "epoch": 8.76,
-      "learning_rate": 6.0931899641577065e-06,
-      "loss": 0.7341,
+      "learning_rate": 4.162487462387162e-05,
+      "loss": 0.5058,
       "step": 1240
     },
     {
       "epoch": 8.83,
-      "learning_rate": 5.734767025089606e-06,
-      "loss": 0.7545,
+      "learning_rate": 4.155323112193724e-05,
+      "loss": 0.5619,
       "step": 1250
     },
     {
       "epoch": 8.9,
-      "learning_rate": 5.376344086021506e-06,
-      "loss": 0.7029,
+      "learning_rate": 4.1481587620002866e-05,
+      "loss": 0.5433,
       "step": 1260
     },
     {
       "epoch": 8.97,
-      "learning_rate": 5.017921146953405e-06,
-      "loss": 0.7029,
+      "learning_rate": 4.14099441180685e-05,
+      "loss": 0.517,
       "step": 1270
     },
     {
       "epoch": 9.0,
-      "eval_accuracy": 0.8149829738933031,
-      "eval_loss": 0.4992757737636566,
-      "eval_runtime": 6.2439,
-      "eval_samples_per_second": 141.097,
-      "eval_steps_per_second": 17.777,
+      "eval_accuracy": 0.9160045402951191,
+      "eval_loss": 0.2737162113189697,
+      "eval_runtime": 6.1693,
+      "eval_samples_per_second": 142.804,
+      "eval_steps_per_second": 17.992,
       "step": 1274
     },
     {
       "epoch": 9.04,
-      "learning_rate": 4.659498207885305e-06,
-      "loss": 0.7542,
+      "learning_rate": 4.133830061613412e-05,
+      "loss": 0.5417,
       "step": 1280
     },
     {
       "epoch": 9.11,
-      "learning_rate": 4.3010752688172045e-06,
-      "loss": 0.7201,
+      "learning_rate": 4.1266657114199745e-05,
+      "loss": 0.5581,
       "step": 1290
     },
     {
       "epoch": 9.18,
-      "learning_rate": 3.942652329749104e-06,
-      "loss": 0.6967,
+      "learning_rate": 4.119501361226537e-05,
+      "loss": 0.4931,
       "step": 1300
     },
     {
       "epoch": 9.25,
-      "learning_rate": 3.5842293906810035e-06,
-      "loss": 0.7151,
+      "learning_rate": 4.112337011033099e-05,
+      "loss": 0.4994,
       "step": 1310
     },
     {
       "epoch": 9.32,
-      "learning_rate": 3.225806451612903e-06,
-      "loss": 0.7127,
+      "learning_rate": 4.105172660839662e-05,
+      "loss": 0.518,
       "step": 1320
     },
     {
       "epoch": 9.39,
-      "learning_rate": 2.867383512544803e-06,
-      "loss": 0.6665,
+      "learning_rate": 4.098008310646225e-05,
+      "loss": 0.5148,
       "step": 1330
     },
     {
       "epoch": 9.47,
-      "learning_rate": 2.5089605734767026e-06,
-      "loss": 0.7662,
+      "learning_rate": 4.090843960452787e-05,
+      "loss": 0.4692,
       "step": 1340
     },
     {
       "epoch": 9.54,
-      "learning_rate": 2.1505376344086023e-06,
-      "loss": 0.7602,
+      "learning_rate": 4.0836796102593495e-05,
+      "loss": 0.5111,
       "step": 1350
     },
     {
       "epoch": 9.61,
-      "learning_rate": 1.7921146953405017e-06,
-      "loss": 0.7054,
+      "learning_rate": 4.0765152600659126e-05,
+      "loss": 0.4556,
       "step": 1360
     },
     {
       "epoch": 9.68,
-      "learning_rate": 1.4336917562724014e-06,
-      "loss": 0.6655,
+      "learning_rate": 4.069350909872475e-05,
+      "loss": 0.5025,
       "step": 1370
     },
     {
       "epoch": 9.75,
-      "learning_rate": 1.0752688172043011e-06,
-      "loss": 0.6564,
+      "learning_rate": 4.062186559679037e-05,
+      "loss": 0.5809,
       "step": 1380
     },
     {
       "epoch": 9.82,
-      "learning_rate": 7.168458781362007e-07,
-      "loss": 0.7077,
+      "learning_rate": 4.0550222094856e-05,
+      "loss": 0.5321,
       "step": 1390
     },
     {
       "epoch": 9.89,
-      "learning_rate": 3.5842293906810036e-07,
-      "loss": 0.6575,
+      "learning_rate": 4.047857859292162e-05,
+      "loss": 0.5143,
       "step": 1400
     },
     {
       "epoch": 9.96,
-      "learning_rate": 0.0,
-      "loss": 0.7275,
+      "learning_rate": 4.0406935090987246e-05,
+      "loss": 0.5366,
       "step": 1410
     },
     {
-      "epoch": 9.96,
-      "eval_accuracy": 0.8342792281498297,
-      "eval_loss": 0.48094892501831055,
-      "eval_runtime": 6.3075,
-      "eval_samples_per_second": 139.674,
-      "eval_steps_per_second": 17.598,
-      "step": 1410
+      "epoch": 9.99,
+      "eval_accuracy": 0.923950056753689,
+      "eval_loss": 0.22288767993450165,
+      "eval_runtime": 6.0958,
+      "eval_samples_per_second": 144.525,
+      "eval_steps_per_second": 18.209,
+      "step": 1415
     },
     {
-      "epoch": 9.96,
-      "step": 1410,
-      "total_flos": 1.6760839403221402e+18,
-      "train_loss": 0.9043752511342367,
-      "train_runtime": 8036.8099,
-      "train_samples_per_second": 9.858,
-      "train_steps_per_second": 0.175
+      "epoch": 10.03,
+      "learning_rate": 4.033529158905288e-05,
+      "loss": 0.5326,
+      "step": 1420
+    },
+    {
+      "epoch": 10.1,
+      "learning_rate": 4.02636480871185e-05,
+      "loss": 0.5089,
+      "step": 1430
+    },
+    {
+      "epoch": 10.17,
+      "learning_rate": 4.0192004585184125e-05,
+      "loss": 0.5244,
+      "step": 1440
+    },
+    {
+      "epoch": 10.24,
+      "learning_rate": 4.0120361083249755e-05,
+      "loss": 0.5221,
+      "step": 1450
+    },
+    {
+      "epoch": 10.31,
+      "learning_rate": 4.004871758131538e-05,
+      "loss": 0.4276,
+      "step": 1460
+    },
+    {
+      "epoch": 10.38,
+      "learning_rate": 3.9977074079380996e-05,
+      "loss": 0.4801,
+      "step": 1470
+    },
+    {
+      "epoch": 10.45,
+      "learning_rate": 3.990543057744663e-05,
+      "loss": 0.4969,
+      "step": 1480
+    },
+    {
+      "epoch": 10.52,
+      "learning_rate": 3.983378707551225e-05,
+      "loss": 0.5258,
+      "step": 1490
+    },
+    {
+      "epoch": 10.6,
+      "learning_rate": 3.9762143573577875e-05,
+      "loss": 0.5242,
+      "step": 1500
+    },
+    {
+      "epoch": 10.67,
+      "learning_rate": 3.9690500071643506e-05,
+      "loss": 0.4626,
+      "step": 1510
+    },
+    {
+      "epoch": 10.74,
+      "learning_rate": 3.961885656970913e-05,
+      "loss": 0.5292,
+      "step": 1520
+    },
+    {
+      "epoch": 10.81,
+      "learning_rate": 3.9547213067774754e-05,
+      "loss": 0.4741,
+      "step": 1530
+    },
+    {
+      "epoch": 10.88,
+      "learning_rate": 3.9475569565840384e-05,
+      "loss": 0.4873,
+      "step": 1540
+    },
+    {
+      "epoch": 10.95,
+      "learning_rate": 3.9403926063906e-05,
+      "loss": 0.4645,
+      "step": 1550
+    },
+    {
+      "epoch": 11.0,
+      "eval_accuracy": 0.9330306469920545,
+      "eval_loss": 0.20377282798290253,
+      "eval_runtime": 6.092,
+      "eval_samples_per_second": 144.617,
+      "eval_steps_per_second": 18.221,
+      "step": 1557
+    },
+    {
+      "epoch": 11.02,
+      "learning_rate": 3.9332282561971626e-05,
+      "loss": 0.5078,
+      "step": 1560
+    },
+    {
+      "epoch": 11.09,
+      "learning_rate": 3.9260639060037256e-05,
+      "loss": 0.4665,
+      "step": 1570
+    },
+    {
+      "epoch": 11.16,
+      "learning_rate": 3.918899555810288e-05,
+      "loss": 0.4408,
+      "step": 1580
+    },
+    {
+      "epoch": 11.23,
+      "learning_rate": 3.9117352056168504e-05,
+      "loss": 0.4783,
+      "step": 1590
+    },
+    {
+      "epoch": 11.3,
+      "learning_rate": 3.9045708554234135e-05,
+      "loss": 0.5538,
+      "step": 1600
+    },
+    {
+      "epoch": 11.37,
+      "learning_rate": 3.897406505229976e-05,
+      "loss": 0.5001,
+      "step": 1610
+    },
+    {
+      "epoch": 11.44,
+      "learning_rate": 3.890242155036538e-05,
+      "loss": 0.4253,
+      "step": 1620
+    },
+    {
+      "epoch": 11.51,
+      "learning_rate": 3.883077804843101e-05,
+      "loss": 0.4619,
+      "step": 1630
+    },
+    {
+      "epoch": 11.58,
+      "learning_rate": 3.875913454649663e-05,
+      "loss": 0.4665,
+      "step": 1640
+    },
+    {
+      "epoch": 11.65,
+      "learning_rate": 3.8687491044562255e-05,
+      "loss": 0.4721,
+      "step": 1650
+    },
+    {
+      "epoch": 11.73,
+      "learning_rate": 3.8615847542627886e-05,
+      "loss": 0.4166,
+      "step": 1660
+    },
+    {
+      "epoch": 11.8,
+      "learning_rate": 3.854420404069351e-05,
+      "loss": 0.5097,
+      "step": 1670
+    },
+    {
+      "epoch": 11.87,
+      "learning_rate": 3.8472560538759133e-05,
+      "loss": 0.4988,
+      "step": 1680
+    },
+    {
+      "epoch": 11.94,
+      "learning_rate": 3.8400917036824764e-05,
+      "loss": 0.4114,
+      "step": 1690
+    },
+    {
+      "epoch": 11.99,
+      "eval_accuracy": 0.9375709421112373,
+      "eval_loss": 0.1851072907447815,
+      "eval_runtime": 6.2123,
+      "eval_samples_per_second": 141.815,
+      "eval_steps_per_second": 17.868,
+      "step": 1698
+    },
+    {
+      "epoch": 12.01,
+      "learning_rate": 3.832927353489039e-05,
+      "loss": 0.4899,
+      "step": 1700
+    },
+    {
+      "epoch": 12.08,
+      "learning_rate": 3.825763003295601e-05,
+      "loss": 0.5046,
+      "step": 1710
+    },
+    {
+      "epoch": 12.15,
+      "learning_rate": 3.8185986531021636e-05,
+      "loss": 0.4511,
+      "step": 1720
+    },
+    {
+      "epoch": 12.22,
+      "learning_rate": 3.811434302908726e-05,
+      "loss": 0.4511,
+      "step": 1730
+    },
+    {
+      "epoch": 12.29,
+      "learning_rate": 3.8042699527152884e-05,
+      "loss": 0.4498,
+      "step": 1740
+    },
+    {
+      "epoch": 12.36,
+      "learning_rate": 3.7971056025218515e-05,
+      "loss": 0.4773,
+      "step": 1750
+    },
+    {
+      "epoch": 12.43,
+      "learning_rate": 3.789941252328414e-05,
+      "loss": 0.494,
+      "step": 1760
+    },
+    {
+      "epoch": 12.5,
+      "learning_rate": 3.782776902134976e-05,
+      "loss": 0.4634,
+      "step": 1770
+    },
+    {
+      "epoch": 12.57,
+      "learning_rate": 3.7756125519415393e-05,
+      "loss": 0.4953,
+      "step": 1780
+    },
+    {
+      "epoch": 12.64,
+      "learning_rate": 3.768448201748102e-05,
+      "loss": 0.4063,
+      "step": 1790
+    },
+    {
+      "epoch": 12.71,
+      "learning_rate": 3.761283851554664e-05,
+      "loss": 0.4812,
+      "step": 1800
+    },
+    {
+      "epoch": 12.79,
+      "learning_rate": 3.7541195013612265e-05,
+      "loss": 0.4629,
+      "step": 1810
+    },
+    {
+      "epoch": 12.86,
+      "learning_rate": 3.746955151167789e-05,
+      "loss": 0.4229,
+      "step": 1820
+    },
+    {
+      "epoch": 12.93,
+      "learning_rate": 3.739790800974352e-05,
+      "loss": 0.3726,
+      "step": 1830
+    },
+    {
+      "epoch": 13.0,
+      "learning_rate": 3.7326264507809144e-05,
+      "loss": 0.4528,
+      "step": 1840
+    },
+    {
+      "epoch": 13.0,
+      "eval_accuracy": 0.9432463110102156,
+      "eval_loss": 0.17957377433776855,
+      "eval_runtime": 6.2071,
+      "eval_samples_per_second": 141.935,
+      "eval_steps_per_second": 17.883,
+      "step": 1840
+    },
+    {
+      "epoch": 13.07,
+      "learning_rate": 3.725462100587477e-05,
+      "loss": 0.4288,
+      "step": 1850
+    },
+    {
+      "epoch": 13.14,
+      "learning_rate": 3.71829775039404e-05,
+      "loss": 0.4264,
+      "step": 1860
+    },
+    {
+      "epoch": 13.21,
+      "learning_rate": 3.711133400200602e-05,
+      "loss": 0.4351,
+      "step": 1870
+    },
+    {
+      "epoch": 13.28,
+      "learning_rate": 3.7039690500071647e-05,
+      "loss": 0.4547,
+      "step": 1880
+    },
+    {
+      "epoch": 13.35,
+      "learning_rate": 3.696804699813727e-05,
+      "loss": 0.4149,
+      "step": 1890
+    },
+    {
+      "epoch": 13.42,
+      "learning_rate": 3.6896403496202895e-05,
+      "loss": 0.3878,
+      "step": 1900
+    },
+    {
+      "epoch": 13.49,
+      "learning_rate": 3.682475999426852e-05,
+      "loss": 0.4646,
+      "step": 1910
+    },
+    {
+      "epoch": 13.56,
+      "learning_rate": 3.675311649233415e-05,
+      "loss": 0.432,
+      "step": 1920
+    },
+    {
+      "epoch": 13.63,
+      "learning_rate": 3.668147299039977e-05,
+      "loss": 0.4186,
+      "step": 1930
+    },
+    {
+      "epoch": 13.7,
+      "learning_rate": 3.66098294884654e-05,
+      "loss": 0.4586,
+      "step": 1940
+    },
+    {
+      "epoch": 13.77,
+      "learning_rate": 3.653818598653103e-05,
+      "loss": 0.4329,
+      "step": 1950
+    },
+    {
+      "epoch": 13.84,
+      "learning_rate": 3.646654248459665e-05,
+      "loss": 0.471,
+      "step": 1960
+    },
+    {
+      "epoch": 13.92,
+      "learning_rate": 3.6394898982662276e-05,
+      "loss": 0.4297,
+      "step": 1970
+    },
+    {
+      "epoch": 13.99,
+      "learning_rate": 3.63232554807279e-05,
+      "loss": 0.4182,
+      "step": 1980
+    },
+    {
+      "epoch": 14.0,
+      "eval_accuracy": 0.9523269012485811,
+      "eval_loss": 0.15777143836021423,
+      "eval_runtime": 6.2837,
+      "eval_samples_per_second": 140.204,
+      "eval_steps_per_second": 17.665,
+      "step": 1982
+    },
+    {
+      "epoch": 14.06,
+      "learning_rate": 3.6251611978793524e-05,
+      "loss": 0.4281,
+      "step": 1990
+    },
+    {
+      "epoch": 14.13,
+      "learning_rate": 3.617996847685915e-05,
+      "loss": 0.4153,
+      "step": 2000
+    },
+    {
+      "epoch": 14.2,
+      "learning_rate": 3.610832497492478e-05,
+      "loss": 0.4115,
+      "step": 2010
+    },
+    {
+      "epoch": 14.27,
+      "learning_rate": 3.60366814729904e-05,
+      "loss": 0.5146,
+      "step": 2020
+    },
+    {
+      "epoch": 14.34,
+      "learning_rate": 3.5965037971056026e-05,
+      "loss": 0.444,
+      "step": 2030
+    },
+    {
+      "epoch": 14.41,
+      "learning_rate": 3.589339446912166e-05,
+      "loss": 0.4059,
+      "step": 2040
+    },
+    {
+      "epoch": 14.48,
+      "learning_rate": 3.582175096718728e-05,
+      "loss": 0.4057,
+      "step": 2050
+    },
+    {
+      "epoch": 14.55,
+      "learning_rate": 3.5750107465252905e-05,
+      "loss": 0.4201,
+      "step": 2060
+    },
+    {
+      "epoch": 14.62,
+      "learning_rate": 3.567846396331853e-05,
+      "loss": 0.4526,
+      "step": 2070
+    },
+    {
+      "epoch": 14.69,
+      "learning_rate": 3.560682046138415e-05,
+      "loss": 0.4452,
+      "step": 2080
+    },
+    {
+      "epoch": 14.76,
+      "learning_rate": 3.553517695944978e-05,
+      "loss": 0.4063,
+      "step": 2090
+    },
+    {
+      "epoch": 14.83,
+      "learning_rate": 3.546353345751541e-05,
+      "loss": 0.4277,
+      "step": 2100
+    },
+    {
+      "epoch": 14.9,
+      "learning_rate": 3.539188995558103e-05,
+      "loss": 0.4436,
+      "step": 2110
+    },
+    {
+      "epoch": 14.97,
+      "learning_rate": 3.5320246453646656e-05,
+      "loss": 0.432,
+      "step": 2120
+    },
+    {
+      "epoch": 15.0,
+      "eval_accuracy": 0.94211123723042,
+      "eval_loss": 0.16604571044445038,
+      "eval_runtime": 6.3147,
+      "eval_samples_per_second": 139.516,
+      "eval_steps_per_second": 17.578,
+      "step": 2123
+    },
+    {
+      "epoch": 15.05,
+      "learning_rate": 3.5248602951712286e-05,
+      "loss": 0.4551,
+      "step": 2130
+    },
+    {
+      "epoch": 15.12,
+      "learning_rate": 3.517695944977791e-05,
+      "loss": 0.3895,
+      "step": 2140
+    },
+    {
+      "epoch": 15.19,
+      "learning_rate": 3.5105315947843534e-05,
+      "loss": 0.4432,
+      "step": 2150
+    },
+    {
+      "epoch": 15.26,
+      "learning_rate": 3.503367244590916e-05,
+      "loss": 0.4599,
+      "step": 2160
+    },
+    {
+      "epoch": 15.33,
+      "learning_rate": 3.496202894397478e-05,
+      "loss": 0.3682,
+      "step": 2170
+    },
+    {
+      "epoch": 15.4,
+      "learning_rate": 3.4890385442040406e-05,
+      "loss": 0.4004,
+      "step": 2180
+    },
+    {
+      "epoch": 15.47,
+      "learning_rate": 3.481874194010604e-05,
+      "loss": 0.4339,
+      "step": 2190
+    },
+    {
+      "epoch": 15.54,
+      "learning_rate": 3.474709843817166e-05,
+      "loss": 0.4134,
+      "step": 2200
+    },
+    {
+      "epoch": 15.61,
+      "learning_rate": 3.4675454936237285e-05,
+      "loss": 0.4399,
+      "step": 2210
+    },
+    {
+      "epoch": 15.68,
+      "learning_rate": 3.4603811434302915e-05,
+      "loss": 0.4276,
+      "step": 2220
+    },
+    {
+      "epoch": 15.75,
+      "learning_rate": 3.453216793236854e-05,
+      "loss": 0.4094,
+      "step": 2230
+    },
+    {
+      "epoch": 15.82,
+      "learning_rate": 3.446052443043416e-05,
+      "loss": 0.4073,
+      "step": 2240
+    },
+    {
+      "epoch": 15.89,
+      "learning_rate": 3.438888092849979e-05,
+      "loss": 0.3707,
+      "step": 2250
+    },
+    {
+      "epoch": 15.96,
+      "learning_rate": 3.431723742656541e-05,
+      "loss": 0.4442,
+      "step": 2260
+    },
+    {
+      "epoch": 16.0,
+      "eval_accuracy": 0.9557321225879682,
+      "eval_loss": 0.14012937247753143,
+      "eval_runtime": 6.467,
+      "eval_samples_per_second": 136.23,
+      "eval_steps_per_second": 17.164,
+      "step": 2265
+    },
+    {
+      "epoch": 16.03,
+      "learning_rate": 3.4245593924631035e-05,
+      "loss": 0.4051,
+      "step": 2270
+    },
+    {
+      "epoch": 16.1,
+      "learning_rate": 3.4173950422696666e-05,
+      "loss": 0.4999,
+      "step": 2280
+    },
+    {
+      "epoch": 16.18,
+      "learning_rate": 3.410230692076229e-05,
+      "loss": 0.436,
+      "step": 2290
+    },
+    {
+      "epoch": 16.25,
+      "learning_rate": 3.4030663418827914e-05,
+      "loss": 0.4243,
+      "step": 2300
+    },
+    {
+      "epoch": 16.32,
+      "learning_rate": 3.3959019916893545e-05,
+      "loss": 0.3869,
+      "step": 2310
+    },
+    {
+      "epoch": 16.39,
+      "learning_rate": 3.388737641495917e-05,
+      "loss": 0.4124,
+      "step": 2320
+    },
+    {
+      "epoch": 16.46,
+      "learning_rate": 3.3815732913024786e-05,
+      "loss": 0.4078,
+      "step": 2330
+    },
+    {
+      "epoch": 16.53,
+      "learning_rate": 3.3744089411090417e-05,
+      "loss": 0.3953,
+      "step": 2340
+    },
+    {
+      "epoch": 16.6,
+      "learning_rate": 3.367244590915604e-05,
+      "loss": 0.4336,
+      "step": 2350
+    },
+    {
+      "epoch": 16.67,
+      "learning_rate": 3.3600802407221664e-05,
+      "loss": 0.3616,
+      "step": 2360
+    },
+    {
+      "epoch": 16.74,
+      "learning_rate": 3.3529158905287295e-05,
+      "loss": 0.3745,
+      "step": 2370
+    },
+    {
+      "epoch": 16.81,
+      "learning_rate": 3.345751540335292e-05,
+      "loss": 0.4494,
+      "step": 2380
+    },
+    {
+      "epoch": 16.88,
+      "learning_rate": 3.338587190141854e-05,
+      "loss": 0.4444,
+      "step": 2390
+    },
+    {
+      "epoch": 16.95,
+      "learning_rate": 3.3314228399484174e-05,
+      "loss": 0.4059,
+      "step": 2400
+    },
+    {
+      "epoch": 16.99,
+      "eval_accuracy": 0.9591373439273553,
+      "eval_loss": 0.13323572278022766,
+      "eval_runtime": 6.9956,
+      "eval_samples_per_second": 125.936,
+      "eval_steps_per_second": 15.867,
+      "step": 2406
+    },
+    {
+      "epoch": 17.02,
+      "learning_rate": 3.32425848975498e-05,
+      "loss": 0.4014,
+      "step": 2410
+    },
+    {
+      "epoch": 17.09,
+      "learning_rate": 3.3170941395615415e-05,
+      "loss": 0.4046,
+      "step": 2420
+    },
+    {
+      "epoch": 17.16,
+      "learning_rate": 3.3099297893681046e-05,
+      "loss": 0.4055,
+      "step": 2430
+    },
+    {
+      "epoch": 17.24,
+      "learning_rate": 3.302765439174667e-05,
+      "loss": 0.4265,
+      "step": 2440
+    },
+    {
+      "epoch": 17.31,
+      "learning_rate": 3.2956010889812294e-05,
+      "loss": 0.3723,
+      "step": 2450
+    },
+    {
+      "epoch": 17.38,
+      "learning_rate": 3.2884367387877924e-05,
+      "loss": 0.3662,
+      "step": 2460
+    },
+    {
+      "epoch": 17.45,
+      "learning_rate": 3.281272388594355e-05,
+      "loss": 0.4136,
+      "step": 2470
+    },
+    {
+      "epoch": 17.52,
+      "learning_rate": 3.274108038400917e-05,
+      "loss": 0.4106,
+      "step": 2480
+    },
+    {
+      "epoch": 17.59,
+      "learning_rate": 3.26694368820748e-05,
+      "loss": 0.4042,
+      "step": 2490
+    },
+    {
+      "epoch": 17.66,
+      "learning_rate": 3.259779338014042e-05,
+      "loss": 0.3787,
+      "step": 2500
+    },
+    {
+      "epoch": 17.73,
+      "learning_rate": 3.2526149878206044e-05,
+      "loss": 0.3918,
+      "step": 2510
+    },
+    {
+      "epoch": 17.8,
+      "learning_rate": 3.2454506376271675e-05,
+      "loss": 0.351,
+      "step": 2520
+    },
+    {
+      "epoch": 17.87,
+      "learning_rate": 3.23828628743373e-05,
+      "loss": 0.4006,
+      "step": 2530
+    },
+    {
+      "epoch": 17.94,
+      "learning_rate": 3.231121937240292e-05,
+      "loss": 0.3498,
+      "step": 2540
+    },
+    {
+      "epoch": 18.0,
+      "eval_accuracy": 0.9534619750283768,
+      "eval_loss": 0.14309445023536682,
+      "eval_runtime": 7.2518,
+      "eval_samples_per_second": 121.487,
+      "eval_steps_per_second": 15.307,
+      "step": 2548
+    },
+    {
+      "epoch": 18.01,
+      "learning_rate": 3.2239575870468554e-05,
+      "loss": 0.3746,
+      "step": 2550
+    },
+    {
+      "epoch": 18.08,
+      "learning_rate": 3.216793236853418e-05,
+      "loss": 0.4102,
+      "step": 2560
+    },
+    {
+      "epoch": 18.15,
+      "learning_rate": 3.20962888665998e-05,
+      "loss": 0.433,
+      "step": 2570
+    },
+    {
+      "epoch": 18.22,
+      "learning_rate": 3.2024645364665425e-05,
+      "loss": 0.3613,
+      "step": 2580
+    },
+    {
+      "epoch": 18.29,
+      "learning_rate": 3.195300186273105e-05,
+      "loss": 0.3981,
+      "step": 2590
+    },
+    {
+      "epoch": 18.37,
+      "learning_rate": 3.1881358360796673e-05,
+      "loss": 0.3661,
+      "step": 2600
+    },
+    {
+      "epoch": 18.44,
+      "learning_rate": 3.1809714858862304e-05,
+      "loss": 0.3646,
+      "step": 2610
+    },
+    {
+      "epoch": 18.51,
+      "learning_rate": 3.173807135692793e-05,
+      "loss": 0.3595,
+      "step": 2620
+    },
+    {
+      "epoch": 18.58,
+      "learning_rate": 3.166642785499355e-05,
+      "loss": 0.3984,
+      "step": 2630
+    },
+    {
+      "epoch": 18.65,
+      "learning_rate": 3.159478435305918e-05,
+      "loss": 0.3975,
+      "step": 2640
+    },
+    {
+      "epoch": 18.72,
+      "learning_rate": 3.152314085112481e-05,
+      "loss": 0.3846,
+      "step": 2650
+    },
+    {
+      "epoch": 18.79,
+      "learning_rate": 3.145149734919043e-05,
+      "loss": 0.3295,
+      "step": 2660
+    },
+    {
+      "epoch": 18.86,
+      "learning_rate": 3.1379853847256055e-05,
+      "loss": 0.4528,
+      "step": 2670
+    },
+    {
+      "epoch": 18.93,
+      "learning_rate": 3.130821034532168e-05,
+      "loss": 0.3869,
+      "step": 2680
+    },
+    {
+      "epoch": 18.99,
+      "eval_accuracy": 0.9511918274687855,
+      "eval_loss": 0.12365122884511948,
+      "eval_runtime": 7.1176,
+      "eval_samples_per_second": 123.778,
+      "eval_steps_per_second": 15.595,
+      "step": 2689
+    },
+    {
+      "epoch": 19.0,
+      "learning_rate": 3.12365668433873e-05,
+      "loss": 0.3786,
+      "step": 2690
+    },
+    {
+      "epoch": 19.07,
+      "learning_rate": 3.116492334145293e-05,
+      "loss": 0.4174,
+      "step": 2700
+    },
+    {
+      "epoch": 19.14,
+      "learning_rate": 3.109327983951856e-05,
+      "loss": 0.3869,
+      "step": 2710
+    },
+    {
+      "epoch": 19.21,
+      "learning_rate": 3.102163633758418e-05,
+      "loss": 0.3583,
+      "step": 2720
+    },
+    {
+      "epoch": 19.28,
+      "learning_rate": 3.094999283564981e-05,
+      "loss": 0.3536,
+      "step": 2730
+    },
+    {
+      "epoch": 19.35,
+      "learning_rate": 3.0878349333715436e-05,
+      "loss": 0.3724,
+      "step": 2740
+    },
+    {
+      "epoch": 19.42,
+      "learning_rate": 3.080670583178106e-05,
+      "loss": 0.3147,
+      "step": 2750
+    },
+    {
+      "epoch": 19.5,
+      "learning_rate": 3.0735062329846684e-05,
+      "loss": 0.3465,
+      "step": 2760
+    },
+    {
+      "epoch": 19.57,
+      "learning_rate": 3.066341882791231e-05,
+      "loss": 0.378,
+      "step": 2770
+    },
+    {
+      "epoch": 19.64,
+      "learning_rate": 3.059177532597793e-05,
+      "loss": 0.3429,
+      "step": 2780
+    },
+    {
+      "epoch": 19.71,
+      "learning_rate": 3.052013182404356e-05,
+      "loss": 0.4112,
+      "step": 2790
+    },
+    {
+      "epoch": 19.78,
+      "learning_rate": 3.0448488322109186e-05,
+      "loss": 0.3739,
+      "step": 2800
+    },
+    {
+      "epoch": 19.85,
+      "learning_rate": 3.037684482017481e-05,
+      "loss": 0.3698,
+      "step": 2810
+    },
+    {
+      "epoch": 19.92,
+      "learning_rate": 3.0305201318240438e-05,
+      "loss": 0.373,
+      "step": 2820
+    },
+    {
+      "epoch": 19.99,
+      "learning_rate": 3.0233557816306062e-05,
+      "loss": 0.3639,
+      "step": 2830
+    },
+    {
+      "epoch": 20.0,
+      "eval_accuracy": 0.960272417707151,
+      "eval_loss": 0.11927589029073715,
+      "eval_runtime": 6.849,
+      "eval_samples_per_second": 128.632,
+      "eval_steps_per_second": 16.207,
+      "step": 2831
+    },
+    {
+      "epoch": 20.06,
+      "learning_rate": 3.0161914314371686e-05,
+      "loss": 0.3422,
+      "step": 2840
+    },
+    {
+      "epoch": 20.13,
+      "learning_rate": 3.0090270812437316e-05,
+      "loss": 0.3691,
+      "step": 2850
+    },
+    {
+      "epoch": 20.2,
+      "learning_rate": 3.001862731050294e-05,
+      "loss": 0.3654,
+      "step": 2860
+    },
+    {
+      "epoch": 20.27,
+      "learning_rate": 2.9946983808568564e-05,
+      "loss": 0.368,
+      "step": 2870
+    },
+    {
+      "epoch": 20.34,
+      "learning_rate": 2.9875340306634192e-05,
+      "loss": 0.4058,
+      "step": 2880
+    },
+    {
+      "epoch": 20.41,
+      "learning_rate": 2.9803696804699816e-05,
+      "loss": 0.4099,
+      "step": 2890
+    },
+    {
+      "epoch": 20.48,
+      "learning_rate": 2.973205330276544e-05,
+      "loss": 0.3929,
+      "step": 2900
+    },
+    {
+      "epoch": 20.55,
+      "learning_rate": 2.9660409800831067e-05,
+      "loss": 0.34,
+      "step": 2910
+    },
+    {
+      "epoch": 20.63,
+      "learning_rate": 2.958876629889669e-05,
+      "loss": 0.3025,
+      "step": 2920
+    },
+    {
+      "epoch": 20.7,
+      "learning_rate": 2.9517122796962315e-05,
+      "loss": 0.4087,
+      "step": 2930
+    },
+    {
+      "epoch": 20.77,
+      "learning_rate": 2.9445479295027946e-05,
+      "loss": 0.3894,
+      "step": 2940
+    },
+    {
+      "epoch": 20.84,
+      "learning_rate": 2.937383579309357e-05,
+      "loss": 0.3472,
+      "step": 2950
+    },
+    {
+      "epoch": 20.91,
+      "learning_rate": 2.930219229115919e-05,
+      "loss": 0.2909,
+      "step": 2960
+    },
+    {
+      "epoch": 20.98,
+      "learning_rate": 2.923054878922482e-05,
+      "loss": 0.3819,
+      "step": 2970
+    },
+    {
+      "epoch": 21.0,
+      "eval_accuracy": 0.9557321225879682,
+      "eval_loss": 0.12340272963047028,
+      "eval_runtime": 6.5462,
+      "eval_samples_per_second": 134.581,
+      "eval_steps_per_second": 16.956,
+      "step": 2973
+    },
+    {
+      "epoch": 21.05,
+      "learning_rate": 2.9158905287290445e-05,
+      "loss": 0.3768,
+      "step": 2980
+    },
+    {
+      "epoch": 21.12,
+      "learning_rate": 2.908726178535607e-05,
+      "loss": 0.3817,
+      "step": 2990
+    },
+    {
+      "epoch": 21.19,
+      "learning_rate": 2.9015618283421696e-05,
+      "loss": 0.3311,
+      "step": 3000
+    },
+    {
+      "epoch": 21.26,
+      "learning_rate": 2.894397478148732e-05,
+      "loss": 0.3117,
+      "step": 3010
+    },
+    {
+      "epoch": 21.33,
+      "learning_rate": 2.8872331279552944e-05,
+      "loss": 0.3106,
+      "step": 3020
+    },
+    {
+      "epoch": 21.4,
+      "learning_rate": 2.8800687777618575e-05,
+      "loss": 0.3228,
+      "step": 3030
+    },
+    {
+      "epoch": 21.47,
+      "learning_rate": 2.8729044275684195e-05,
+      "loss": 0.3495,
+      "step": 3040
+    },
+    {
+      "epoch": 21.54,
+      "learning_rate": 2.865740077374982e-05,
+      "loss": 0.3456,
+      "step": 3050
+    },
+    {
+      "epoch": 21.61,
+      "learning_rate": 2.858575727181545e-05,
+      "loss": 0.3742,
+      "step": 3060
+    },
+    {
+      "epoch": 21.69,
+      "learning_rate": 2.8514113769881074e-05,
+      "loss": 0.4374,
+      "step": 3070
+    },
+    {
+      "epoch": 21.76,
+      "learning_rate": 2.8442470267946698e-05,
+      "loss": 0.4181,
+      "step": 3080
+    },
+    {
+      "epoch": 21.83,
+      "learning_rate": 2.8370826766012325e-05,
+      "loss": 0.3352,
+      "step": 3090
+    },
+    {
+      "epoch": 21.9,
+      "learning_rate": 2.829918326407795e-05,
+      "loss": 0.3284,
+      "step": 3100
+    },
+    {
+      "epoch": 21.97,
+      "learning_rate": 2.8227539762143573e-05,
+      "loss": 0.3491,
+      "step": 3110
+    },
+    {
+      "epoch": 22.0,
+      "eval_accuracy": 0.9568671963677639,
+      "eval_loss": 0.12074992805719376,
+      "eval_runtime": 6.176,
+      "eval_samples_per_second": 142.648,
+      "eval_steps_per_second": 17.973,
+      "step": 3114
+    },
+    {
+      "epoch": 22.04,
+      "learning_rate": 2.8155896260209204e-05,
+      "loss": 0.3429,
+      "step": 3120
+    },
+    {
+      "epoch": 22.11,
+      "learning_rate": 2.8084252758274825e-05,
+      "loss": 0.3946,
+      "step": 3130
+    },
+    {
+      "epoch": 22.18,
+      "learning_rate": 2.801260925634045e-05,
+      "loss": 0.3725,
+      "step": 3140
+    },
+    {
+      "epoch": 22.25,
+      "learning_rate": 2.794096575440608e-05,
+      "loss": 0.3322,
+      "step": 3150
+    },
+    {
+      "epoch": 22.32,
+      "learning_rate": 2.7869322252471703e-05,
+      "loss": 0.3044,
+      "step": 3160
+    },
+    {
+      "epoch": 22.39,
+      "learning_rate": 2.7797678750537327e-05,
+      "loss": 0.3284,
+      "step": 3170
+    },
+    {
+      "epoch": 22.46,
+      "learning_rate": 2.7726035248602955e-05,
+      "loss": 0.3336,
+      "step": 3180
+    },
+    {
+      "epoch": 22.53,
+      "learning_rate": 2.765439174666858e-05,
+      "loss": 0.372,
+      "step": 3190
+    },
+    {
+      "epoch": 22.6,
+      "learning_rate": 2.7582748244734203e-05,
+      "loss": 0.3659,
+      "step": 3200
+    },
+    {
+      "epoch": 22.67,
+      "learning_rate": 2.751110474279983e-05,
+      "loss": 0.3627,
+      "step": 3210
+    },
+    {
+      "epoch": 22.74,
+      "learning_rate": 2.7439461240865454e-05,
+      "loss": 0.347,
+      "step": 3220
+    },
+    {
+      "epoch": 22.82,
+      "learning_rate": 2.7367817738931078e-05,
+      "loss": 0.3686,
+      "step": 3230
+    },
+    {
+      "epoch": 22.89,
+      "learning_rate": 2.729617423699671e-05,
+      "loss": 0.3453,
+      "step": 3240
+    },
+    {
+      "epoch": 22.96,
+      "learning_rate": 2.7224530735062332e-05,
+      "loss": 0.3259,
+      "step": 3250
+    },
+    {
+      "epoch": 23.0,
+      "eval_accuracy": 0.9591373439273553,
+      "eval_loss": 0.12335461378097534,
+      "eval_runtime": 6.1713,
+      "eval_samples_per_second": 142.758,
+      "eval_steps_per_second": 17.987,
+      "step": 3256
+    },
+    {
+      "epoch": 23.03,
+      "learning_rate": 2.7152887233127956e-05,
+      "loss": 0.3471,
+      "step": 3260
+    },
+    {
+      "epoch": 23.1,
+      "learning_rate": 2.7081243731193584e-05,
+      "loss": 0.3599,
+      "step": 3270
+    },
+    {
+      "epoch": 23.17,
+      "learning_rate": 2.7009600229259208e-05,
+      "loss": 0.3938,
+      "step": 3280
+    },
+    {
+      "epoch": 23.24,
+      "learning_rate": 2.6937956727324832e-05,
+      "loss": 0.3609,
+      "step": 3290
+    },
+    {
+      "epoch": 23.31,
+      "learning_rate": 2.686631322539046e-05,
+      "loss": 0.4036,
+      "step": 3300
+    },
+    {
+      "epoch": 23.38,
+      "learning_rate": 2.6794669723456083e-05,
+      "loss": 0.3267,
+      "step": 3310
+    },
+    {
+      "epoch": 23.45,
+      "learning_rate": 2.6723026221521707e-05,
+      "loss": 0.324,
+      "step": 3320
+    },
+    {
+      "epoch": 23.52,
+      "learning_rate": 2.6651382719587338e-05,
+      "loss": 0.3424,
+      "step": 3330
+    },
+    {
+      "epoch": 23.59,
+      "learning_rate": 2.657973921765296e-05,
+      "loss": 0.3208,
+      "step": 3340
+    },
+    {
+      "epoch": 23.66,
+      "learning_rate": 2.6508095715718582e-05,
+      "loss": 0.2846,
+      "step": 3350
+    },
+    {
+      "epoch": 23.73,
+      "learning_rate": 2.6436452213784213e-05,
+      "loss": 0.3132,
+      "step": 3360
+    },
+    {
+      "epoch": 23.8,
+      "learning_rate": 2.6364808711849837e-05,
+      "loss": 0.3199,
+      "step": 3370
+    },
+    {
+      "epoch": 23.87,
+      "learning_rate": 2.629316520991546e-05,
+      "loss": 0.3802,
+      "step": 3380
+    },
+    {
+      "epoch": 23.95,
+      "learning_rate": 2.6221521707981088e-05,
+      "loss": 0.3199,
+      "step": 3390
+    },
+    {
+      "epoch": 23.99,
+      "eval_accuracy": 0.9659477866061293,
+      "eval_loss": 0.10280199348926544,
+      "eval_runtime": 6.2384,
+      "eval_samples_per_second": 141.222,
+      "eval_steps_per_second": 17.793,
+      "step": 3397
+    },
+    {
+      "epoch": 24.02,
+      "learning_rate": 2.6149878206046712e-05,
+      "loss": 0.3645,
+      "step": 3400
+    },
+    {
+      "epoch": 24.09,
+      "learning_rate": 2.6078234704112336e-05,
+      "loss": 0.3617,
+      "step": 3410
+    },
+    {
+      "epoch": 24.16,
+      "learning_rate": 2.6006591202177967e-05,
+      "loss": 0.333,
+      "step": 3420
+    },
+    {
+      "epoch": 24.23,
+      "learning_rate": 2.593494770024359e-05,
+      "loss": 0.3623,
+      "step": 3430
+    },
+    {
+      "epoch": 24.3,
+      "learning_rate": 2.586330419830921e-05,
+      "loss": 0.2935,
+      "step": 3440
+    },
+    {
+      "epoch": 24.37,
+      "learning_rate": 2.5791660696374842e-05,
+      "loss": 0.3686,
+      "step": 3450
+    },
+    {
+      "epoch": 24.44,
+      "learning_rate": 2.5720017194440466e-05,
+      "loss": 0.3415,
+      "step": 3460
+    },
+    {
+      "epoch": 24.51,
+      "learning_rate": 2.564837369250609e-05,
+      "loss": 0.3356,
+      "step": 3470
+    },
+    {
+      "epoch": 24.58,
+      "learning_rate": 2.5576730190571717e-05,
+      "loss": 0.3801,
+      "step": 3480
+    },
+    {
+      "epoch": 24.65,
+      "learning_rate": 2.550508668863734e-05,
+      "loss": 0.3473,
+      "step": 3490
+    },
+    {
+      "epoch": 24.72,
+      "learning_rate": 2.5433443186702965e-05,
+      "loss": 0.3293,
+      "step": 3500
+    },
+    {
+      "epoch": 24.79,
+      "learning_rate": 2.5361799684768596e-05,
+      "loss": 0.2884,
+      "step": 3510
+    },
+    {
+      "epoch": 24.86,
+      "learning_rate": 2.5290156182834217e-05,
+      "loss": 0.3605,
+      "step": 3520
+    },
+    {
+      "epoch": 24.93,
+      "learning_rate": 2.521851268089984e-05,
+      "loss": 0.3398,
+      "step": 3530
+    },
+    {
+      "epoch": 25.0,
+      "eval_accuracy": 0.960272417707151,
+      "eval_loss": 0.10104019939899445,
+      "eval_runtime": 6.188,
+      "eval_samples_per_second": 142.373,
+      "eval_steps_per_second": 17.938,
+      "step": 3539
+    },
+    {
+      "epoch": 25.01,
+      "learning_rate": 2.514686917896547e-05,
+      "loss": 0.3016,
+      "step": 3540
+    },
+    {
+      "epoch": 25.08,
+      "learning_rate": 2.5075225677031095e-05,
+      "loss": 0.3471,
+      "step": 3550
+    },
+    {
+      "epoch": 25.15,
+      "learning_rate": 2.500358217509672e-05,
+      "loss": 0.3086,
+      "step": 3560
+    },
+    {
+      "epoch": 25.22,
+      "learning_rate": 2.4931938673162343e-05,
+      "loss": 0.3697,
+      "step": 3570
+    },
+    {
+      "epoch": 25.29,
+      "learning_rate": 2.486029517122797e-05,
+      "loss": 0.3304,
+      "step": 3580
+    },
+    {
+      "epoch": 25.36,
+      "learning_rate": 2.4788651669293598e-05,
+      "loss": 0.3418,
+      "step": 3590
+    },
+    {
+      "epoch": 25.43,
+      "learning_rate": 2.4717008167359222e-05,
+      "loss": 0.3914,
+      "step": 3600
+    },
+    {
+      "epoch": 25.5,
+      "learning_rate": 2.4645364665424846e-05,
+      "loss": 0.3683,
+      "step": 3610
+    },
+    {
+      "epoch": 25.57,
+      "learning_rate": 2.4573721163490473e-05,
+      "loss": 0.3428,
+      "step": 3620
+    },
+    {
+      "epoch": 25.64,
+      "learning_rate": 2.4502077661556097e-05,
+      "loss": 0.2947,
+      "step": 3630
+    },
+    {
+      "epoch": 25.71,
+      "learning_rate": 2.4430434159621725e-05,
+      "loss": 0.3163,
+      "step": 3640
+    },
+    {
+      "epoch": 25.78,
+      "learning_rate": 2.435879065768735e-05,
+      "loss": 0.2936,
+      "step": 3650
+    },
+    {
+      "epoch": 25.85,
+      "learning_rate": 2.4287147155752972e-05,
+      "loss": 0.3367,
+      "step": 3660
+    },
+    {
+      "epoch": 25.92,
+      "learning_rate": 2.42155036538186e-05,
+      "loss": 0.315,
+      "step": 3670
+    },
+    {
+      "epoch": 25.99,
+      "learning_rate": 2.4143860151884227e-05,
+      "loss": 0.3108,
+      "step": 3680
+    },
+    {
+      "epoch": 25.99,
+      "eval_accuracy": 0.9670828603859251,
+      "eval_loss": 0.10149858891963959,
+      "eval_runtime": 6.1497,
+      "eval_samples_per_second": 143.26,
+      "eval_steps_per_second": 18.05,
+      "step": 3680
+    },
+    {
+      "epoch": 26.06,
+      "learning_rate": 2.407221664994985e-05,
+      "loss": 0.2994,
+      "step": 3690
+    },
+    {
+      "epoch": 26.14,
+      "learning_rate": 2.4000573148015475e-05,
+      "loss": 0.3851,
+      "step": 3700
+    },
+    {
+      "epoch": 26.21,
+      "learning_rate": 2.3928929646081102e-05,
+      "loss": 0.3489,
+      "step": 3710
+    },
+    {
+      "epoch": 26.28,
+      "learning_rate": 2.3857286144146726e-05,
+      "loss": 0.3918,
+      "step": 3720
+    },
+    {
+      "epoch": 26.35,
+      "learning_rate": 2.3785642642212354e-05,
+      "loss": 0.3422,
+      "step": 3730
+    },
+    {
+      "epoch": 26.42,
+      "learning_rate": 2.3713999140277978e-05,
+      "loss": 0.3474,
+      "step": 3740
+    },
+    {
+      "epoch": 26.49,
+      "learning_rate": 2.36423556383436e-05,
+      "loss": 0.3357,
+      "step": 3750
+    },
+    {
+      "epoch": 26.56,
+      "learning_rate": 2.357071213640923e-05,
+      "loss": 0.3424,
+      "step": 3760
+    },
+    {
+      "epoch": 26.63,
+      "learning_rate": 2.3499068634474856e-05,
+      "loss": 0.3289,
+      "step": 3770
+    },
+    {
+      "epoch": 26.7,
+      "learning_rate": 2.342742513254048e-05,
+      "loss": 0.3118,
+      "step": 3780
+    },
+    {
+      "epoch": 26.77,
+      "learning_rate": 2.3355781630606104e-05,
+      "loss": 0.3182,
+      "step": 3790
+    },
+    {
+      "epoch": 26.84,
+      "learning_rate": 2.328413812867173e-05,
+      "loss": 0.3191,
+      "step": 3800
+    },
+    {
+      "epoch": 26.91,
+      "learning_rate": 2.3212494626737356e-05,
+      "loss": 0.298,
+      "step": 3810
+    },
+    {
+      "epoch": 26.98,
+      "learning_rate": 2.3140851124802983e-05,
+      "loss": 0.3417,
+      "step": 3820
+    },
+    {
+      "epoch": 27.0,
+      "eval_accuracy": 0.9614074914869466,
+      "eval_loss": 0.10795789211988449,
+      "eval_runtime": 6.1816,
+      "eval_samples_per_second": 142.519,
+      "eval_steps_per_second": 17.956,
+      "step": 3822
+    },
+    {
+      "epoch": 27.05,
+      "learning_rate": 2.3069207622868607e-05,
+      "loss": 0.3285,
+      "step": 3830
+    },
+    {
+      "epoch": 27.12,
+      "learning_rate": 2.299756412093423e-05,
+      "loss": 0.3441,
+      "step": 3840
+    },
+    {
+      "epoch": 27.19,
+      "learning_rate": 2.2925920618999858e-05,
+      "loss": 0.324,
+      "step": 3850
+    },
+    {
+      "epoch": 27.27,
+      "learning_rate": 2.2854277117065486e-05,
+      "loss": 0.3254,
+      "step": 3860
+    },
+    {
+      "epoch": 27.34,
+      "learning_rate": 2.2782633615131106e-05,
+      "loss": 0.2996,
+      "step": 3870
+    },
+    {
+      "epoch": 27.41,
+      "learning_rate": 2.2710990113196733e-05,
+      "loss": 0.3197,
+      "step": 3880
+    },
+    {
+      "epoch": 27.48,
+      "learning_rate": 2.263934661126236e-05,
+      "loss": 0.3086,
+      "step": 3890
+    },
+    {
+      "epoch": 27.55,
+      "learning_rate": 2.2567703109327985e-05,
+      "loss": 0.3767,
+      "step": 3900
+    },
+    {
+      "epoch": 27.62,
+      "learning_rate": 2.249605960739361e-05,
+      "loss": 0.3127,
+      "step": 3910
+    },
+    {
+      "epoch": 27.69,
+      "learning_rate": 2.2424416105459236e-05,
+      "loss": 0.317,
+      "step": 3920
+    },
+    {
+      "epoch": 27.76,
+      "learning_rate": 2.235277260352486e-05,
+      "loss": 0.3517,
+      "step": 3930
+    },
+    {
+      "epoch": 27.83,
+      "learning_rate": 2.2281129101590487e-05,
+      "loss": 0.3339,
+      "step": 3940
+    },
+    {
+      "epoch": 27.9,
+      "learning_rate": 2.220948559965611e-05,
+      "loss": 0.3094,
+      "step": 3950
+    },
+    {
+      "epoch": 27.97,
+      "learning_rate": 2.2137842097721735e-05,
+      "loss": 0.3835,
+      "step": 3960
+    },
+    {
+      "epoch": 28.0,
+      "eval_accuracy": 0.9591373439273553,
+      "eval_loss": 0.10561925917863846,
+      "eval_runtime": 6.2527,
+      "eval_samples_per_second": 140.898,
+      "eval_steps_per_second": 17.752,
+      "step": 3964
+    },
+    {
+      "epoch": 28.04,
+      "learning_rate": 2.2066198595787363e-05,
+      "loss": 0.2687,
+      "step": 3970
+    },
+    {
+      "epoch": 28.11,
+      "learning_rate": 2.199455509385299e-05,
+      "loss": 0.2943,
+      "step": 3980
+    },
+    {
+      "epoch": 28.18,
+      "learning_rate": 2.1922911591918614e-05,
+      "loss": 0.3234,
+      "step": 3990
+    },
+    {
+      "epoch": 28.25,
+      "learning_rate": 2.1851268089984238e-05,
+      "loss": 0.3118,
+      "step": 4000
+    },
+    {
+      "epoch": 28.32,
+      "learning_rate": 2.1779624588049865e-05,
+      "loss": 0.3027,
+      "step": 4010
+    },
+    {
+      "epoch": 28.4,
+      "learning_rate": 2.170798108611549e-05,
+      "loss": 0.2919,
+      "step": 4020
+    },
+    {
+      "epoch": 28.47,
+      "learning_rate": 2.1636337584181117e-05,
+      "loss": 0.3546,
+      "step": 4030
+    },
+    {
+      "epoch": 28.54,
+      "learning_rate": 2.156469408224674e-05,
+      "loss": 0.2915,
+      "step": 4040
+    },
+    {
+      "epoch": 28.61,
+      "learning_rate": 2.1493050580312365e-05,
+      "loss": 0.3017,
+      "step": 4050
+    },
+    {
+      "epoch": 28.68,
+      "learning_rate": 2.1421407078377992e-05,
+      "loss": 0.336,
+      "step": 4060
+    },
+    {
+      "epoch": 28.75,
+      "learning_rate": 2.134976357644362e-05,
+      "loss": 0.3046,
+      "step": 4070
+    },
+    {
+      "epoch": 28.82,
+      "learning_rate": 2.1278120074509243e-05,
+      "loss": 0.3208,
+      "step": 4080
+    },
+    {
+      "epoch": 28.89,
+      "learning_rate": 2.1206476572574867e-05,
+      "loss": 0.3296,
+      "step": 4090
+    },
+    {
+      "epoch": 28.96,
+      "learning_rate": 2.1134833070640495e-05,
+      "loss": 0.3336,
+      "step": 4100
+    },
+    {
+      "epoch": 29.0,
+      "eval_accuracy": 0.963677639046538,
+      "eval_loss": 0.10110009461641312,
+      "eval_runtime": 6.2387,
+      "eval_samples_per_second": 141.214,
+      "eval_steps_per_second": 17.792,
+      "step": 4105
+    },
+    {
+      "epoch": 29.03,
+      "learning_rate": 2.106318956870612e-05,
+      "loss": 0.3282,
+      "step": 4110
+    },
+    {
+      "epoch": 29.1,
+      "learning_rate": 2.0991546066771746e-05,
+      "loss": 0.2584,
+      "step": 4120
+    },
+    {
+      "epoch": 29.17,
+      "learning_rate": 2.091990256483737e-05,
+      "loss": 0.3626,
+      "step": 4130
+    },
+    {
+      "epoch": 29.24,
+      "learning_rate": 2.0848259062902994e-05,
+      "loss": 0.2591,
+      "step": 4140
+    },
+    {
+      "epoch": 29.31,
+      "learning_rate": 2.077661556096862e-05,
+      "loss": 0.3293,
+      "step": 4150
+    },
+    {
+      "epoch": 29.38,
+      "learning_rate": 2.070497205903425e-05,
+      "loss": 0.29,
+      "step": 4160
+    },
+    {
+      "epoch": 29.46,
+      "learning_rate": 2.0633328557099872e-05,
+      "loss": 0.3254,
+      "step": 4170
+    },
+    {
+      "epoch": 29.53,
+      "learning_rate": 2.0561685055165496e-05,
+      "loss": 0.3111,
+      "step": 4180
+    },
+    {
+      "epoch": 29.6,
+      "learning_rate": 2.0490041553231124e-05,
+      "loss": 0.3249,
+      "step": 4190
+    },
+    {
+      "epoch": 29.67,
+      "learning_rate": 2.0418398051296748e-05,
+      "loss": 0.3228,
+      "step": 4200
+    },
+    {
+      "epoch": 29.74,
+      "learning_rate": 2.0346754549362375e-05,
+      "loss": 0.3304,
+      "step": 4210
+    },
+    {
+      "epoch": 29.81,
+      "learning_rate": 2.0275111047428e-05,
+      "loss": 0.3202,
+      "step": 4220
+    },
+    {
+      "epoch": 29.88,
+      "learning_rate": 2.0203467545493623e-05,
+      "loss": 0.3418,
+      "step": 4230
+    },
+    {
+      "epoch": 29.95,
+      "learning_rate": 2.013182404355925e-05,
+      "loss": 0.3035,
+      "step": 4240
+    },
+    {
+      "epoch": 30.0,
+      "eval_accuracy": 0.9614074914869466,
+      "eval_loss": 0.09720989316701889,
+      "eval_runtime": 6.1738,
+      "eval_samples_per_second": 142.701,
+      "eval_steps_per_second": 17.979,
+      "step": 4247
+    },
+    {
+      "epoch": 30.02,
+      "learning_rate": 2.0060180541624878e-05,
+      "loss": 0.3171,
+      "step": 4250
+    },
+    {
+      "epoch": 30.09,
+      "learning_rate": 1.9988537039690498e-05,
+      "loss": 0.3112,
+      "step": 4260
+    },
+    {
+      "epoch": 30.16,
+      "learning_rate": 1.9916893537756126e-05,
+      "loss": 0.2917,
+      "step": 4270
+    },
+    {
+      "epoch": 30.23,
+      "learning_rate": 1.9845250035821753e-05,
+      "loss": 0.3143,
+      "step": 4280
+    },
+    {
+      "epoch": 30.3,
+      "learning_rate": 1.9773606533887377e-05,
+      "loss": 0.3178,
+      "step": 4290
+    },
+    {
+      "epoch": 30.37,
+      "learning_rate": 1.9701963031953e-05,
+      "loss": 0.3074,
+      "step": 4300
+    },
+    {
+      "epoch": 30.44,
+      "learning_rate": 1.9630319530018628e-05,
+      "loss": 0.3104,
+      "step": 4310
+    },
+    {
+      "epoch": 30.51,
+      "learning_rate": 1.9558676028084252e-05,
+      "loss": 0.3005,
+      "step": 4320
+    },
+    {
+      "epoch": 30.59,
+      "learning_rate": 1.948703252614988e-05,
+      "loss": 0.3198,
+      "step": 4330
+    },
+    {
+      "epoch": 30.66,
+      "learning_rate": 1.9415389024215503e-05,
+      "loss": 0.3343,
+      "step": 4340
+    },
+    {
+      "epoch": 30.73,
+      "learning_rate": 1.9343745522281127e-05,
+      "loss": 0.3421,
+      "step": 4350
+    },
+    {
+      "epoch": 30.8,
+      "learning_rate": 1.9272102020346755e-05,
+      "loss": 0.3201,
+      "step": 4360
+    },
+    {
+      "epoch": 30.87,
+      "learning_rate": 1.9200458518412382e-05,
+      "loss": 0.3525,
+      "step": 4370
+    },
+    {
+      "epoch": 30.94,
+      "learning_rate": 1.9128815016478006e-05,
+      "loss": 0.2559,
+      "step": 4380
+    },
+    {
+      "epoch": 30.99,
+      "eval_accuracy": 0.9659477866061293,
+      "eval_loss": 0.09409745037555695,
+      "eval_runtime": 6.2004,
+      "eval_samples_per_second": 142.087,
+      "eval_steps_per_second": 17.902,
+      "step": 4388
+    },
+    {
+      "epoch": 31.01,
+      "learning_rate": 1.905717151454363e-05,
+      "loss": 0.3376,
+      "step": 4390
+    },
+    {
+      "epoch": 31.08,
+      "learning_rate": 1.8985528012609257e-05,
+      "loss": 0.2492,
+      "step": 4400
+    },
+    {
+      "epoch": 31.15,
+      "learning_rate": 1.891388451067488e-05,
+      "loss": 0.3028,
+      "step": 4410
+    },
+    {
+      "epoch": 31.22,
+      "learning_rate": 1.884224100874051e-05,
+      "loss": 0.277,
+      "step": 4420
+    },
+    {
+      "epoch": 31.29,
+      "learning_rate": 1.8770597506806133e-05,
+      "loss": 0.3199,
+      "step": 4430
+    },
+    {
+      "epoch": 31.36,
+      "learning_rate": 1.869895400487176e-05,
+      "loss": 0.3186,
+      "step": 4440
+    },
+    {
+      "epoch": 31.43,
+      "learning_rate": 1.8627310502937384e-05,
+      "loss": 0.3344,
+      "step": 4450
+    },
+    {
+      "epoch": 31.5,
+      "learning_rate": 1.855566700100301e-05,
+      "loss": 0.331,
+      "step": 4460
+    },
+    {
+      "epoch": 31.57,
+      "learning_rate": 1.8484023499068635e-05,
+      "loss": 0.3137,
+      "step": 4470
+    },
+    {
+      "epoch": 31.64,
+      "learning_rate": 1.841237999713426e-05,
+      "loss": 0.3383,
+      "step": 4480
+    },
+    {
+      "epoch": 31.72,
+      "learning_rate": 1.8340736495199887e-05,
+      "loss": 0.2932,
+      "step": 4490
+    },
+    {
+      "epoch": 31.79,
+      "learning_rate": 1.8269092993265514e-05,
+      "loss": 0.3344,
+      "step": 4500
+    },
+    {
+      "epoch": 31.86,
+      "learning_rate": 1.8197449491331138e-05,
+      "loss": 0.3648,
+      "step": 4510
+    },
+    {
+      "epoch": 31.93,
+      "learning_rate": 1.8125805989396762e-05,
+      "loss": 0.3042,
+      "step": 4520
+    },
+    {
+      "epoch": 32.0,
+      "learning_rate": 1.805416248746239e-05,
+      "loss": 0.378,
+      "step": 4530
+    },
+    {
+      "epoch": 32.0,
+      "eval_accuracy": 0.960272417707151,
+      "eval_loss": 0.09633893519639969,
+      "eval_runtime": 6.191,
+      "eval_samples_per_second": 142.303,
+      "eval_steps_per_second": 17.929,
+      "step": 4530
+    },
+    {
+      "epoch": 32.07,
+      "learning_rate": 1.7982518985528013e-05,
+      "loss": 0.3469,
+      "step": 4540
+    },
+    {
+      "epoch": 32.14,
+      "learning_rate": 1.791087548359364e-05,
+      "loss": 0.3094,
+      "step": 4550
+    },
+    {
+      "epoch": 32.21,
+      "learning_rate": 1.7839231981659264e-05,
+      "loss": 0.3114,
+      "step": 4560
+    },
+    {
+      "epoch": 32.28,
+      "learning_rate": 1.776758847972489e-05,
+      "loss": 0.285,
+      "step": 4570
+    },
+    {
+      "epoch": 32.35,
+      "learning_rate": 1.7695944977790516e-05,
+      "loss": 0.3464,
+      "step": 4580
+    },
+    {
+      "epoch": 32.42,
+      "learning_rate": 1.7624301475856143e-05,
+      "loss": 0.2776,
+      "step": 4590
+    },
+    {
+      "epoch": 32.49,
+      "learning_rate": 1.7552657973921767e-05,
+      "loss": 0.2839,
+      "step": 4600
+    },
+    {
+      "epoch": 32.56,
+      "learning_rate": 1.748101447198739e-05,
+      "loss": 0.3177,
+      "step": 4610
+    },
+    {
+      "epoch": 32.63,
+      "learning_rate": 1.740937097005302e-05,
+      "loss": 0.2979,
+      "step": 4620
+    },
+    {
+      "epoch": 32.7,
+      "learning_rate": 1.7337727468118642e-05,
+      "loss": 0.3049,
+      "step": 4630
+    },
+    {
+      "epoch": 32.77,
+      "learning_rate": 1.726608396618427e-05,
+      "loss": 0.2821,
+      "step": 4640
+    },
+    {
+      "epoch": 32.85,
+      "learning_rate": 1.7194440464249894e-05,
+      "loss": 0.3001,
+      "step": 4650
+    },
+    {
+      "epoch": 32.92,
+      "learning_rate": 1.7122796962315518e-05,
+      "loss": 0.2865,
+      "step": 4660
+    },
+    {
+      "epoch": 32.99,
+      "learning_rate": 1.7051153460381145e-05,
+      "loss": 0.2932,
+      "step": 4670
+    },
+    {
+      "epoch": 32.99,
+      "eval_accuracy": 0.9716231555051078,
+      "eval_loss": 0.09159873425960541,
+      "eval_runtime": 6.2721,
+      "eval_samples_per_second": 140.464,
+      "eval_steps_per_second": 17.698,
+      "step": 4671
+    },
+    {
+      "epoch": 33.06,
+      "learning_rate": 1.6979509958446772e-05,
+      "loss": 0.3428,
+      "step": 4680
+    },
+    {
+      "epoch": 33.13,
+      "learning_rate": 1.6907866456512393e-05,
+      "loss": 0.2876,
+      "step": 4690
+    },
+    {
+      "epoch": 33.2,
+      "learning_rate": 1.683622295457802e-05,
+      "loss": 0.2839,
+      "step": 4700
+    },
+    {
+      "epoch": 33.27,
+      "learning_rate": 1.6764579452643648e-05,
+      "loss": 0.2631,
+      "step": 4710
+    },
+    {
+      "epoch": 33.34,
+      "learning_rate": 1.669293595070927e-05,
+      "loss": 0.3336,
+      "step": 4720
+    },
+    {
+      "epoch": 33.41,
+      "learning_rate": 1.66212924487749e-05,
+      "loss": 0.3206,
+      "step": 4730
+    },
+    {
+      "epoch": 33.48,
+      "learning_rate": 1.6549648946840523e-05,
+      "loss": 0.2657,
+      "step": 4740
+    },
+    {
+      "epoch": 33.55,
+      "learning_rate": 1.6478005444906147e-05,
+      "loss": 0.3231,
+      "step": 4750
+    },
+    {
+      "epoch": 33.62,
+      "learning_rate": 1.6406361942971774e-05,
+      "loss": 0.2688,
+      "step": 4760
+    },
+    {
+      "epoch": 33.69,
+      "learning_rate": 1.63347184410374e-05,
+      "loss": 0.3003,
+      "step": 4770
+    },
+    {
+      "epoch": 33.76,
+      "learning_rate": 1.6263074939103022e-05,
+      "loss": 0.2996,
+      "step": 4780
+    },
+    {
+      "epoch": 33.83,
+      "learning_rate": 1.619143143716865e-05,
+      "loss": 0.2837,
+      "step": 4790
+    },
+    {
+      "epoch": 33.91,
+      "learning_rate": 1.6119787935234277e-05,
+      "loss": 0.2837,
+      "step": 4800
+    },
+    {
+      "epoch": 33.98,
+      "learning_rate": 1.60481444332999e-05,
+      "loss": 0.3072,
+      "step": 4810
+    },
+    {
+      "epoch": 34.0,
+      "eval_accuracy": 0.9670828603859251,
+      "eval_loss": 0.09172412008047104,
+      "eval_runtime": 6.2411,
+      "eval_samples_per_second": 141.16,
+      "eval_steps_per_second": 17.785,
+      "step": 4813
+    },
+    {
+      "epoch": 34.05,
+      "learning_rate": 1.5976500931365525e-05,
+      "loss": 0.3043,
+      "step": 4820
+    },
+    {
+      "epoch": 34.12,
+      "learning_rate": 1.5904857429431152e-05,
+      "loss": 0.2533,
+      "step": 4830
+    },
+    {
+      "epoch": 34.19,
+      "learning_rate": 1.5833213927496776e-05,
+      "loss": 0.2799,
+      "step": 4840
+    },
+    {
+      "epoch": 34.26,
+      "learning_rate": 1.5761570425562403e-05,
+      "loss": 0.3132,
+      "step": 4850
+    },
+    {
+      "epoch": 34.33,
+      "learning_rate": 1.5689926923628027e-05,
+      "loss": 0.2681,
+      "step": 4860
+    },
+    {
+      "epoch": 34.4,
+      "learning_rate": 1.561828342169365e-05,
+      "loss": 0.2843,
+      "step": 4870
+    },
+    {
+      "epoch": 34.47,
+      "learning_rate": 1.554663991975928e-05,
+      "loss": 0.2964,
+      "step": 4880
+    },
+    {
+      "epoch": 34.54,
+      "learning_rate": 1.5474996417824906e-05,
+      "loss": 0.3074,
+      "step": 4890
+    },
+    {
+      "epoch": 34.61,
+      "learning_rate": 1.540335291589053e-05,
+      "loss": 0.3897,
+      "step": 4900
+    },
+    {
+      "epoch": 34.68,
+      "learning_rate": 1.5331709413956154e-05,
+      "loss": 0.3322,
+      "step": 4910
+    },
+    {
+      "epoch": 34.75,
+      "learning_rate": 1.526006591202178e-05,
+      "loss": 0.296,
+      "step": 4920
+    },
+    {
+      "epoch": 34.82,
+      "learning_rate": 1.5188422410087405e-05,
+      "loss": 0.3511,
+      "step": 4930
+    },
+    {
+      "epoch": 34.89,
+      "learning_rate": 1.5116778908153031e-05,
+      "loss": 0.2865,
+      "step": 4940
+    },
+    {
+      "epoch": 34.96,
+      "learning_rate": 1.5045135406218658e-05,
+      "loss": 0.3081,
+      "step": 4950
+    },
+    {
+      "epoch": 35.0,
+      "eval_accuracy": 0.9625425652667423,
+      "eval_loss": 0.10247617214918137,
+      "eval_runtime": 6.223,
+      "eval_samples_per_second": 141.573,
+      "eval_steps_per_second": 17.837,
+      "step": 4955
+    },
+    {
+      "epoch": 35.04,
+      "learning_rate": 1.4973491904284282e-05,
+      "loss": 0.2365,
+      "step": 4960
+    },
+    {
+      "epoch": 35.11,
+      "learning_rate": 1.4901848402349908e-05,
+      "loss": 0.2982,
+      "step": 4970
+    },
+    {
+      "epoch": 35.18,
+      "learning_rate": 1.4830204900415534e-05,
+      "loss": 0.2889,
+      "step": 4980
+    },
+    {
+      "epoch": 35.25,
+      "learning_rate": 1.4758561398481157e-05,
+      "loss": 0.3014,
+      "step": 4990
+    },
+    {
+      "epoch": 35.32,
+      "learning_rate": 1.4686917896546785e-05,
+      "loss": 0.2814,
+      "step": 5000
+    },
+    {
+      "epoch": 35.39,
+      "learning_rate": 1.461527439461241e-05,
+      "loss": 0.3025,
+      "step": 5010
+    },
+    {
+      "epoch": 35.46,
+      "learning_rate": 1.4543630892678034e-05,
+      "loss": 0.3352,
+      "step": 5020
+    },
+    {
+      "epoch": 35.53,
+      "learning_rate": 1.447198739074366e-05,
+      "loss": 0.31,
+      "step": 5030
+    },
+    {
+      "epoch": 35.6,
+      "learning_rate": 1.4400343888809287e-05,
+      "loss": 0.2765,
+      "step": 5040
+    },
+    {
+      "epoch": 35.67,
+      "learning_rate": 1.432870038687491e-05,
+      "loss": 0.2585,
+      "step": 5050
+    },
+    {
+      "epoch": 35.74,
+      "learning_rate": 1.4257056884940537e-05,
+      "loss": 0.3092,
+      "step": 5060
+    },
+    {
+      "epoch": 35.81,
+      "learning_rate": 1.4185413383006163e-05,
+      "loss": 0.2737,
+      "step": 5070
+    },
+    {
+      "epoch": 35.88,
+      "learning_rate": 1.4113769881071787e-05,
+      "loss": 0.2914,
+      "step": 5080
+    },
+    {
+      "epoch": 35.95,
+      "learning_rate": 1.4042126379137412e-05,
+      "loss": 0.2724,
+      "step": 5090
+    },
+    {
+      "epoch": 36.0,
+      "eval_accuracy": 0.9670828603859251,
+      "eval_loss": 0.08742310851812363,
+      "eval_runtime": 6.1991,
+      "eval_samples_per_second": 142.117,
+      "eval_steps_per_second": 17.906,
+      "step": 5096
+    },
+    {
+      "epoch": 36.02,
+      "learning_rate": 1.397048287720304e-05,
+      "loss": 0.3187,
+      "step": 5100
+    },
+    {
+      "epoch": 36.09,
+      "learning_rate": 1.3898839375268664e-05,
+      "loss": 0.2791,
+      "step": 5110
+    },
+    {
+      "epoch": 36.17,
+      "learning_rate": 1.382719587333429e-05,
+      "loss": 0.3202,
+      "step": 5120
+    },
+    {
+      "epoch": 36.24,
+      "learning_rate": 1.3755552371399915e-05,
+      "loss": 0.3069,
+      "step": 5130
+    },
+    {
+      "epoch": 36.31,
+      "learning_rate": 1.3683908869465539e-05,
+      "loss": 0.2765,
+      "step": 5140
+    },
+    {
+      "epoch": 36.38,
+      "learning_rate": 1.3612265367531166e-05,
+      "loss": 0.3028,
+      "step": 5150
+    },
+    {
+      "epoch": 36.45,
+      "learning_rate": 1.3540621865596792e-05,
+      "loss": 0.2659,
+      "step": 5160
+    },
+    {
+      "epoch": 36.52,
+      "learning_rate": 1.3468978363662416e-05,
+      "loss": 0.2954,
+      "step": 5170
+    },
+    {
+      "epoch": 36.59,
+      "learning_rate": 1.3397334861728042e-05,
+      "loss": 0.2794,
+      "step": 5180
+    },
+    {
+      "epoch": 36.66,
+      "learning_rate": 1.3325691359793669e-05,
+      "loss": 0.2803,
+      "step": 5190
+    },
+    {
+      "epoch": 36.73,
+      "learning_rate": 1.3254047857859291e-05,
+      "loss": 0.2828,
+      "step": 5200
+    },
+    {
+      "epoch": 36.8,
+      "learning_rate": 1.3182404355924918e-05,
+      "loss": 0.2625,
+      "step": 5210
+    },
+    {
+      "epoch": 36.87,
+      "learning_rate": 1.3110760853990544e-05,
+      "loss": 0.3041,
+      "step": 5220
+    },
+    {
+      "epoch": 36.94,
+      "learning_rate": 1.3039117352056168e-05,
+      "loss": 0.2621,
+      "step": 5230
+    },
+    {
+      "epoch": 37.0,
+      "eval_accuracy": 0.9704880817253121,
+      "eval_loss": 0.08466104418039322,
+      "eval_runtime": 6.181,
+      "eval_samples_per_second": 142.533,
+      "eval_steps_per_second": 17.958,
+      "step": 5238
+    },
+    {
+      "epoch": 37.01,
+      "learning_rate": 1.2967473850121795e-05,
+      "loss": 0.2643,
+      "step": 5240
+    },
+    {
+      "epoch": 37.08,
+      "learning_rate": 1.2895830348187421e-05,
+      "loss": 0.2628,
+      "step": 5250
+    },
+    {
+      "epoch": 37.15,
+      "learning_rate": 1.2824186846253045e-05,
+      "loss": 0.2644,
+      "step": 5260
+    },
+    {
+      "epoch": 37.23,
+      "learning_rate": 1.275254334431867e-05,
+      "loss": 0.2571,
+      "step": 5270
+    },
+    {
+      "epoch": 37.3,
+      "learning_rate": 1.2680899842384298e-05,
+      "loss": 0.3253,
+      "step": 5280
+    },
+    {
+      "epoch": 37.37,
+      "learning_rate": 1.260925634044992e-05,
+      "loss": 0.3264,
+      "step": 5290
+    },
+    {
+      "epoch": 37.44,
+      "learning_rate": 1.2537612838515548e-05,
+      "loss": 0.2312,
+      "step": 5300
+    },
+    {
+      "epoch": 37.51,
+      "learning_rate": 1.2465969336581172e-05,
+      "loss": 0.2967,
+      "step": 5310
+    },
+    {
+      "epoch": 37.58,
+      "learning_rate": 1.2394325834646799e-05,
+      "loss": 0.3067,
+      "step": 5320
+    },
+    {
+      "epoch": 37.65,
+      "learning_rate": 1.2322682332712423e-05,
+      "loss": 0.3144,
+      "step": 5330
+    },
+    {
+      "epoch": 37.72,
+      "learning_rate": 1.2251038830778049e-05,
+      "loss": 0.2732,
+      "step": 5340
+    },
+    {
+      "epoch": 37.79,
+      "learning_rate": 1.2179395328843674e-05,
+      "loss": 0.3068,
+      "step": 5350
+    },
+    {
+      "epoch": 37.86,
+      "learning_rate": 1.21077518269093e-05,
+      "loss": 0.3079,
+      "step": 5360
+    },
+    {
+      "epoch": 37.93,
+      "learning_rate": 1.2036108324974926e-05,
+      "loss": 0.3521,
+      "step": 5370
+    },
+    {
+      "epoch": 37.99,
+      "eval_accuracy": 0.9727582292849035,
+      "eval_loss": 0.08290865272283554,
+      "eval_runtime": 6.1831,
+      "eval_samples_per_second": 142.485,
+      "eval_steps_per_second": 17.952,
+      "step": 5379
+    },
+    {
+      "epoch": 38.0,
+      "learning_rate": 1.1964464823040551e-05,
+      "loss": 0.3242,
+      "step": 5380
+    },
+    {
+      "epoch": 38.07,
+      "learning_rate": 1.1892821321106177e-05,
+      "loss": 0.2962,
+      "step": 5390
+    },
+    {
+      "epoch": 38.14,
+      "learning_rate": 1.18211778191718e-05,
+      "loss": 0.2441,
+      "step": 5400
+    },
+    {
+      "epoch": 38.21,
+      "learning_rate": 1.1749534317237428e-05,
+      "loss": 0.2776,
+      "step": 5410
+    },
+    {
+      "epoch": 38.28,
+      "learning_rate": 1.1677890815303052e-05,
+      "loss": 0.3171,
+      "step": 5420
+    },
+    {
+      "epoch": 38.36,
+      "learning_rate": 1.1606247313368678e-05,
+      "loss": 0.3235,
+      "step": 5430
+    },
+    {
+      "epoch": 38.43,
+      "learning_rate": 1.1534603811434303e-05,
+      "loss": 0.3009,
+      "step": 5440
+    },
+    {
+      "epoch": 38.5,
+      "learning_rate": 1.1462960309499929e-05,
+      "loss": 0.237,
+      "step": 5450
+    },
+    {
+      "epoch": 38.57,
+      "learning_rate": 1.1391316807565553e-05,
+      "loss": 0.3021,
+      "step": 5460
+    },
+    {
+      "epoch": 38.64,
+      "learning_rate": 1.131967330563118e-05,
+      "loss": 0.2664,
+      "step": 5470
+    },
+    {
+      "epoch": 38.71,
+      "learning_rate": 1.1248029803696804e-05,
+      "loss": 0.2985,
+      "step": 5480
+    },
+    {
+      "epoch": 38.78,
+      "learning_rate": 1.117638630176243e-05,
+      "loss": 0.2882,
+      "step": 5490
+    },
+    {
+      "epoch": 38.85,
+      "learning_rate": 1.1104742799828056e-05,
+      "loss": 0.2409,
+      "step": 5500
+    },
+    {
+      "epoch": 38.92,
+      "learning_rate": 1.1033099297893681e-05,
+      "loss": 0.3168,
+      "step": 5510
+    },
+    {
+      "epoch": 38.99,
+      "learning_rate": 1.0961455795959307e-05,
+      "loss": 0.2883,
+      "step": 5520
+    },
+    {
+      "epoch": 39.0,
+      "eval_accuracy": 0.9727582292849035,
+      "eval_loss": 0.08599155396223068,
+      "eval_runtime": 6.3582,
+      "eval_samples_per_second": 138.561,
+      "eval_steps_per_second": 17.458,
+      "step": 5521
+    },
+    {
+      "epoch": 39.06,
+      "learning_rate": 1.0889812294024933e-05,
+      "loss": 0.3102,
+      "step": 5530
+    },
+    {
+      "epoch": 39.13,
+      "learning_rate": 1.0818168792090558e-05,
+      "loss": 0.2801,
+      "step": 5540
+    },
+    {
+      "epoch": 39.2,
+      "learning_rate": 1.0746525290156182e-05,
+      "loss": 0.2933,
+      "step": 5550
+    },
+    {
+      "epoch": 39.27,
+      "learning_rate": 1.067488178822181e-05,
+      "loss": 0.3349,
+      "step": 5560
+    },
+    {
+      "epoch": 39.34,
+      "learning_rate": 1.0603238286287434e-05,
+      "loss": 0.2951,
+      "step": 5570
+    },
+    {
+      "epoch": 39.41,
+      "learning_rate": 1.053159478435306e-05,
+      "loss": 0.3278,
+      "step": 5580
+    },
+    {
+      "epoch": 39.49,
+      "learning_rate": 1.0459951282418685e-05,
+      "loss": 0.2953,
+      "step": 5590
+    },
+    {
+      "epoch": 39.56,
+      "learning_rate": 1.038830778048431e-05,
+      "loss": 0.2917,
+      "step": 5600
+    },
+    {
+      "epoch": 39.63,
+      "learning_rate": 1.0316664278549936e-05,
+      "loss": 0.2397,
+      "step": 5610
+    },
+    {
+      "epoch": 39.7,
+      "learning_rate": 1.0245020776615562e-05,
+      "loss": 0.2857,
+      "step": 5620
+    },
+    {
+      "epoch": 39.77,
+      "learning_rate": 1.0173377274681188e-05,
+      "loss": 0.3059,
+      "step": 5630
+    },
+    {
+      "epoch": 39.84,
+      "learning_rate": 1.0101733772746811e-05,
+      "loss": 0.3071,
+      "step": 5640
+    },
+    {
+      "epoch": 39.91,
+      "learning_rate": 1.0030090270812439e-05,
+      "loss": 0.264,
+      "step": 5650
+    },
+    {
+      "epoch": 39.98,
+      "learning_rate": 9.958446768878063e-06,
+      "loss": 0.2617,
+      "step": 5660
+    },
+    {
+      "epoch": 39.99,
+      "eval_accuracy": 0.9682179341657208,
+      "eval_loss": 0.08981039375066757,
+      "eval_runtime": 6.1751,
+      "eval_samples_per_second": 142.669,
+      "eval_steps_per_second": 17.975,
+      "step": 5662
+    },
+    {
+      "epoch": 40.05,
+      "learning_rate": 9.886803266943688e-06,
+      "loss": 0.2845,
+      "step": 5670
+    },
+    {
+      "epoch": 40.12,
+      "learning_rate": 9.815159765009314e-06,
+      "loss": 0.2579,
+      "step": 5680
+    },
+    {
+      "epoch": 40.19,
+      "learning_rate": 9.74351626307494e-06,
+      "loss": 0.2552,
+      "step": 5690
+    },
+    {
+      "epoch": 40.26,
+      "learning_rate": 9.671872761140564e-06,
+      "loss": 0.3055,
+      "step": 5700
+    },
+    {
+      "epoch": 40.33,
+      "learning_rate": 9.600229259206191e-06,
+      "loss": 0.2988,
+      "step": 5710
+    },
+    {
+      "epoch": 40.4,
+      "learning_rate": 9.528585757271815e-06,
+      "loss": 0.2743,
+      "step": 5720
+    },
+    {
+      "epoch": 40.47,
+      "learning_rate": 9.45694225533744e-06,
+      "loss": 0.2599,
+      "step": 5730
+    },
+    {
+      "epoch": 40.54,
+      "learning_rate": 9.385298753403066e-06,
+      "loss": 0.262,
+      "step": 5740
+    },
+    {
+      "epoch": 40.62,
+      "learning_rate": 9.313655251468692e-06,
+      "loss": 0.279,
+      "step": 5750
+    },
+    {
+      "epoch": 40.69,
+      "learning_rate": 9.242011749534318e-06,
+      "loss": 0.2444,
+      "step": 5760
+    },
+    {
+      "epoch": 40.76,
+      "learning_rate": 9.170368247599943e-06,
+      "loss": 0.2473,
+      "step": 5770
+    },
+    {
+      "epoch": 40.83,
+      "learning_rate": 9.098724745665569e-06,
+      "loss": 0.3152,
+      "step": 5780
+    },
+    {
+      "epoch": 40.9,
+      "learning_rate": 9.027081243731195e-06,
+      "loss": 0.3125,
+      "step": 5790
+    },
+    {
+      "epoch": 40.97,
+      "learning_rate": 8.95543774179682e-06,
+      "loss": 0.2893,
+      "step": 5800
+    },
+    {
+      "epoch": 41.0,
+      "eval_accuracy": 0.9670828603859251,
+      "eval_loss": 0.08769955486059189,
+      "eval_runtime": 6.1731,
+      "eval_samples_per_second": 142.715,
+      "eval_steps_per_second": 17.981,
+      "step": 5804
+    },
+    {
+      "epoch": 41.04,
+      "learning_rate": 8.883794239862444e-06,
+      "loss": 0.2922,
+      "step": 5810
+    },
+    {
+      "epoch": 41.11,
+      "learning_rate": 8.812150737928072e-06,
+      "loss": 0.2612,
+      "step": 5820
+    },
+    {
+      "epoch": 41.18,
+      "learning_rate": 8.740507235993696e-06,
+      "loss": 0.3029,
+      "step": 5830
+    },
+    {
+      "epoch": 41.25,
+      "learning_rate": 8.668863734059321e-06,
+      "loss": 0.3104,
+      "step": 5840
+    },
+    {
+      "epoch": 41.32,
+      "learning_rate": 8.597220232124947e-06,
+      "loss": 0.3205,
+      "step": 5850
+    },
+    {
+      "epoch": 41.39,
+      "learning_rate": 8.525576730190572e-06,
+      "loss": 0.2962,
+      "step": 5860
+    },
+    {
+      "epoch": 41.46,
+      "learning_rate": 8.453933228256196e-06,
+      "loss": 0.3315,
+      "step": 5870
+    },
+    {
+      "epoch": 41.53,
+      "learning_rate": 8.382289726321824e-06,
+      "loss": 0.3096,
+      "step": 5880
+    },
+    {
+      "epoch": 41.6,
+      "learning_rate": 8.31064622438745e-06,
+      "loss": 0.2607,
+      "step": 5890
+    },
+    {
+      "epoch": 41.68,
+      "learning_rate": 8.239002722453073e-06,
+      "loss": 0.3114,
+      "step": 5900
+    },
+    {
+      "epoch": 41.75,
+      "learning_rate": 8.1673592205187e-06,
+      "loss": 0.26,
+      "step": 5910
+    },
+    {
+      "epoch": 41.82,
+      "learning_rate": 8.095715718584325e-06,
+      "loss": 0.2806,
+      "step": 5920
+    },
+    {
+      "epoch": 41.89,
+      "learning_rate": 8.02407221664995e-06,
+      "loss": 0.2779,
+      "step": 5930
+    },
+    {
+      "epoch": 41.96,
+      "learning_rate": 7.952428714715576e-06,
+      "loss": 0.2994,
+      "step": 5940
+    },
+    {
+      "epoch": 42.0,
+      "eval_accuracy": 0.9761634506242906,
+      "eval_loss": 0.08218241482973099,
+      "eval_runtime": 6.2358,
+      "eval_samples_per_second": 141.28,
+      "eval_steps_per_second": 17.8,
+      "step": 5946
+    },
+    {
+      "epoch": 42.03,
+      "learning_rate": 7.880785212781202e-06,
+      "loss": 0.273,
+      "step": 5950
+    },
+    {
+      "epoch": 42.1,
+      "learning_rate": 7.809141710846826e-06,
+      "loss": 0.2522,
+      "step": 5960
+    },
+    {
+      "epoch": 42.17,
+      "learning_rate": 7.737498208912453e-06,
+      "loss": 0.2712,
+      "step": 5970
+    },
+    {
+      "epoch": 42.24,
+      "learning_rate": 7.665854706978077e-06,
+      "loss": 0.2691,
+      "step": 5980
+    },
+    {
+      "epoch": 42.31,
+      "learning_rate": 7.594211205043703e-06,
+      "loss": 0.2341,
+      "step": 5990
+    },
+    {
+      "epoch": 42.38,
+      "learning_rate": 7.522567703109329e-06,
+      "loss": 0.3191,
+      "step": 6000
+    },
+    {
+      "epoch": 42.45,
+      "learning_rate": 7.450924201174954e-06,
+      "loss": 0.3098,
+      "step": 6010
+    },
+    {
+      "epoch": 42.52,
+      "learning_rate": 7.379280699240579e-06,
+      "loss": 0.3204,
+      "step": 6020
+    },
+    {
+      "epoch": 42.59,
+      "learning_rate": 7.307637197306205e-06,
+      "loss": 0.2728,
+      "step": 6030
+    },
+    {
+      "epoch": 42.66,
+      "learning_rate": 7.23599369537183e-06,
+      "loss": 0.3128,
+      "step": 6040
+    },
+    {
+      "epoch": 42.73,
+      "learning_rate": 7.164350193437455e-06,
+      "loss": 0.2768,
+      "step": 6050
+    },
+    {
+      "epoch": 42.81,
+      "learning_rate": 7.092706691503081e-06,
+      "loss": 0.2934,
+      "step": 6060
+    },
+    {
+      "epoch": 42.88,
+      "learning_rate": 7.021063189568706e-06,
+      "loss": 0.281,
+      "step": 6070
+    },
+    {
+      "epoch": 42.95,
+      "learning_rate": 6.949419687634332e-06,
+      "loss": 0.2483,
+      "step": 6080
+    },
+    {
+      "epoch": 43.0,
+      "eval_accuracy": 0.9704880817253121,
+      "eval_loss": 0.08335914462804794,
+      "eval_runtime": 6.2487,
+      "eval_samples_per_second": 140.989,
+      "eval_steps_per_second": 17.764,
+      "step": 6087
+    },
+    {
+      "epoch": 43.02,
+      "learning_rate": 6.8777761856999575e-06,
+      "loss": 0.2943,
+      "step": 6090
+    },
+    {
+      "epoch": 43.09,
+      "learning_rate": 6.806132683765583e-06,
+      "loss": 0.2823,
+      "step": 6100
+    },
+    {
+      "epoch": 43.16,
+      "learning_rate": 6.734489181831208e-06,
+      "loss": 0.3293,
+      "step": 6110
+    },
+    {
+      "epoch": 43.23,
+      "learning_rate": 6.6628456798968344e-06,
+      "loss": 0.2691,
+      "step": 6120
+    },
+    {
+      "epoch": 43.3,
+      "learning_rate": 6.591202177962459e-06,
+      "loss": 0.2579,
+      "step": 6130
+    },
+    {
+      "epoch": 43.37,
+      "learning_rate": 6.519558676028084e-06,
+      "loss": 0.2751,
+      "step": 6140
+    },
+    {
+      "epoch": 43.44,
+      "learning_rate": 6.4479151740937106e-06,
+      "loss": 0.3124,
+      "step": 6150
+    },
+    {
+      "epoch": 43.51,
+      "learning_rate": 6.376271672159335e-06,
+      "loss": 0.2286,
+      "step": 6160
+    },
+    {
+      "epoch": 43.58,
+      "learning_rate": 6.30462817022496e-06,
+      "loss": 0.2765,
+      "step": 6170
+    },
+    {
+      "epoch": 43.65,
+      "learning_rate": 6.232984668290586e-06,
+      "loss": 0.2674,
+      "step": 6180
+    },
+    {
+      "epoch": 43.72,
+      "learning_rate": 6.1613411663562115e-06,
+      "loss": 0.2706,
+      "step": 6190
+    },
+    {
+      "epoch": 43.79,
+      "learning_rate": 6.089697664421837e-06,
+      "loss": 0.2729,
+      "step": 6200
+    },
+    {
+      "epoch": 43.86,
+      "learning_rate": 6.018054162487463e-06,
+      "loss": 0.2813,
+      "step": 6210
+    },
+    {
+      "epoch": 43.94,
+      "learning_rate": 5.9464106605530884e-06,
+      "loss": 0.301,
+      "step": 6220
+    },
+    {
+      "epoch": 44.0,
+      "eval_accuracy": 0.9693530079455165,
+      "eval_loss": 0.08827933669090271,
+      "eval_runtime": 6.3056,
+      "eval_samples_per_second": 139.718,
+      "eval_steps_per_second": 17.603,
+      "step": 6229
+    },
+    {
+      "epoch": 44.01,
+      "learning_rate": 5.874767158618714e-06,
+      "loss": 0.2597,
+      "step": 6230
+    },
+    {
+      "epoch": 44.08,
+      "learning_rate": 5.803123656684339e-06,
+      "loss": 0.2672,
+      "step": 6240
+    },
+    {
+      "epoch": 44.15,
+      "learning_rate": 5.7314801547499646e-06,
+      "loss": 0.3034,
+      "step": 6250
+    },
+    {
+      "epoch": 44.22,
+      "learning_rate": 5.65983665281559e-06,
+      "loss": 0.244,
+      "step": 6260
+    },
+    {
+      "epoch": 44.29,
+      "learning_rate": 5.588193150881215e-06,
+      "loss": 0.3157,
+      "step": 6270
+    },
+    {
+      "epoch": 44.36,
+      "learning_rate": 5.516549648946841e-06,
+      "loss": 0.2708,
+      "step": 6280
+    },
+    {
+      "epoch": 44.43,
+      "learning_rate": 5.444906147012466e-06,
+      "loss": 0.319,
+      "step": 6290
+    },
+    {
+      "epoch": 44.5,
+      "learning_rate": 5.373262645078091e-06,
+      "loss": 0.2733,
+      "step": 6300
+    },
+    {
+      "epoch": 44.57,
+      "learning_rate": 5.301619143143717e-06,
+      "loss": 0.2707,
+      "step": 6310
+    },
+    {
+      "epoch": 44.64,
+      "learning_rate": 5.2299756412093424e-06,
+      "loss": 0.266,
+      "step": 6320
+    },
+    {
+      "epoch": 44.71,
+      "learning_rate": 5.158332139274968e-06,
+      "loss": 0.287,
+      "step": 6330
+    },
+    {
+      "epoch": 44.78,
+      "learning_rate": 5.086688637340594e-06,
+      "loss": 0.2893,
+      "step": 6340
+    },
+    {
+      "epoch": 44.85,
+      "learning_rate": 5.015045135406219e-06,
+      "loss": 0.2802,
+      "step": 6350
+    },
+    {
+      "epoch": 44.92,
+      "learning_rate": 4.943401633471844e-06,
+      "loss": 0.284,
+      "step": 6360
+    },
+    {
+      "epoch": 44.99,
+      "learning_rate": 4.87175813153747e-06,
+      "loss": 0.2648,
+      "step": 6370
+    },
+    {
+      "epoch": 44.99,
+      "eval_accuracy": 0.9704880817253121,
+      "eval_loss": 0.08335667103528976,
+      "eval_runtime": 6.2077,
+      "eval_samples_per_second": 141.92,
+      "eval_steps_per_second": 17.881,
+      "step": 6370
+    },
+    {
+      "epoch": 45.07,
+      "learning_rate": 4.8001146296030955e-06,
+      "loss": 0.2652,
+      "step": 6380
+    },
+    {
+      "epoch": 45.14,
+      "learning_rate": 4.72847112766872e-06,
+      "loss": 0.3211,
+      "step": 6390
+    },
+    {
+      "epoch": 45.21,
+      "learning_rate": 4.656827625734346e-06,
+      "loss": 0.2702,
+      "step": 6400
+    },
+    {
+      "epoch": 45.28,
+      "learning_rate": 4.585184123799972e-06,
+      "loss": 0.2667,
+      "step": 6410
+    },
+    {
+      "epoch": 45.35,
+      "learning_rate": 4.513540621865597e-06,
+      "loss": 0.2929,
+      "step": 6420
+    },
+    {
+      "epoch": 45.42,
+      "learning_rate": 4.441897119931222e-06,
+      "loss": 0.2739,
+      "step": 6430
+    },
+    {
+      "epoch": 45.49,
+      "learning_rate": 4.370253617996848e-06,
+      "loss": 0.3129,
+      "step": 6440
+    },
+    {
+      "epoch": 45.56,
+      "learning_rate": 4.298610116062473e-06,
+      "loss": 0.2775,
+      "step": 6450
+    },
+    {
+      "epoch": 45.63,
+      "learning_rate": 4.226966614128098e-06,
+      "loss": 0.2688,
+      "step": 6460
+    },
+    {
+      "epoch": 45.7,
+      "learning_rate": 4.155323112193725e-06,
+      "loss": 0.2916,
+      "step": 6470
+    },
+    {
+      "epoch": 45.77,
+      "learning_rate": 4.08367961025935e-06,
+      "loss": 0.296,
+      "step": 6480
+    },
+    {
+      "epoch": 45.84,
+      "learning_rate": 4.012036108324975e-06,
+      "loss": 0.2611,
+      "step": 6490
+    },
+    {
+      "epoch": 45.91,
+      "learning_rate": 3.940392606390601e-06,
+      "loss": 0.3501,
+      "step": 6500
+    },
+    {
+      "epoch": 45.98,
+      "learning_rate": 3.8687491044562265e-06,
+      "loss": 0.2902,
+      "step": 6510
+    },
+    {
+      "epoch": 46.0,
+      "eval_accuracy": 0.9648127128263337,
+      "eval_loss": 0.08790680021047592,
+      "eval_runtime": 6.2553,
+      "eval_samples_per_second": 140.84,
+      "eval_steps_per_second": 17.745,
+      "step": 6512
+    },
+    {
+      "epoch": 46.05,
+      "learning_rate": 3.7971056025218513e-06,
+      "loss": 0.246,
+      "step": 6520
+    },
+    {
+      "epoch": 46.13,
+      "learning_rate": 3.725462100587477e-06,
+      "loss": 0.3185,
+      "step": 6530
+    },
+    {
+      "epoch": 46.2,
+      "learning_rate": 3.6538185986531026e-06,
+      "loss": 0.2685,
+      "step": 6540
+    },
+    {
+      "epoch": 46.27,
+      "learning_rate": 3.5821750967187274e-06,
+      "loss": 0.2837,
+      "step": 6550
+    },
+    {
+      "epoch": 46.34,
+      "learning_rate": 3.510531594784353e-06,
+      "loss": 0.3001,
+      "step": 6560
+    },
+    {
+      "epoch": 46.41,
+      "learning_rate": 3.4388880928499787e-06,
+      "loss": 0.3043,
+      "step": 6570
+    },
+    {
+      "epoch": 46.48,
+      "learning_rate": 3.367244590915604e-06,
+      "loss": 0.2775,
+      "step": 6580
+    },
+    {
+      "epoch": 46.55,
+      "learning_rate": 3.2956010889812296e-06,
+      "loss": 0.2847,
+      "step": 6590
+    },
+    {
+      "epoch": 46.62,
+      "learning_rate": 3.2239575870468553e-06,
+      "loss": 0.2708,
+      "step": 6600
+    },
+    {
+      "epoch": 46.69,
+      "learning_rate": 3.15231408511248e-06,
+      "loss": 0.2187,
+      "step": 6610
+    },
+    {
+      "epoch": 46.76,
+      "learning_rate": 3.0806705831781057e-06,
+      "loss": 0.2837,
+      "step": 6620
+    },
+    {
+      "epoch": 46.83,
+      "learning_rate": 3.0090270812437314e-06,
+      "loss": 0.2944,
+      "step": 6630
+    },
+    {
+      "epoch": 46.9,
+      "learning_rate": 2.937383579309357e-06,
+      "loss": 0.2549,
+      "step": 6640
+    },
+    {
+      "epoch": 46.97,
+      "learning_rate": 2.8657400773749823e-06,
+      "loss": 0.299,
+      "step": 6650
+    },
+    {
+      "epoch": 46.99,
+      "eval_accuracy": 0.9693530079455165,
+      "eval_loss": 0.08433925360441208,
+      "eval_runtime": 6.2093,
+      "eval_samples_per_second": 141.884,
+      "eval_steps_per_second": 17.876,
+      "step": 6653
+    },
+    {
+      "epoch": 47.04,
+      "learning_rate": 2.7940965754406075e-06,
+      "loss": 0.2713,
+      "step": 6660
+    },
+    {
+      "epoch": 47.11,
+      "learning_rate": 2.722453073506233e-06,
+      "loss": 0.3077,
+      "step": 6670
+    },
+    {
+      "epoch": 47.18,
+      "learning_rate": 2.6508095715718584e-06,
+      "loss": 0.3292,
+      "step": 6680
+    },
+    {
+      "epoch": 47.26,
+      "learning_rate": 2.579166069637484e-06,
+      "loss": 0.2728,
+      "step": 6690
+    },
+    {
+      "epoch": 47.33,
+      "learning_rate": 2.5075225677031097e-06,
+      "loss": 0.277,
+      "step": 6700
+    },
+    {
+      "epoch": 47.4,
+      "learning_rate": 2.435879065768735e-06,
+      "loss": 0.2725,
+      "step": 6710
+    },
+    {
+      "epoch": 47.47,
+      "learning_rate": 2.36423556383436e-06,
+      "loss": 0.3048,
+      "step": 6720
+    },
+    {
+      "epoch": 47.54,
+      "learning_rate": 2.292592061899986e-06,
+      "loss": 0.3192,
+      "step": 6730
+    },
+    {
+      "epoch": 47.61,
+      "learning_rate": 2.220948559965611e-06,
+      "loss": 0.2998,
+      "step": 6740
+    },
+    {
+      "epoch": 47.68,
+      "learning_rate": 2.1493050580312367e-06,
+      "loss": 0.2564,
+      "step": 6750
+    },
+    {
+      "epoch": 47.75,
+      "learning_rate": 2.0776615560968624e-06,
+      "loss": 0.2979,
+      "step": 6760
+    },
+    {
+      "epoch": 47.82,
+      "learning_rate": 2.0060180541624876e-06,
+      "loss": 0.2651,
+      "step": 6770
+    },
+    {
+      "epoch": 47.89,
+      "learning_rate": 1.9343745522281132e-06,
+      "loss": 0.2946,
+      "step": 6780
+    },
+    {
+      "epoch": 47.96,
+      "learning_rate": 1.8627310502937385e-06,
+      "loss": 0.2726,
+      "step": 6790
+    },
+    {
+      "epoch": 48.0,
+      "eval_accuracy": 0.9659477866061293,
+      "eval_loss": 0.0919908732175827,
+      "eval_runtime": 6.2684,
+      "eval_samples_per_second": 140.546,
+      "eval_steps_per_second": 17.708,
+      "step": 6795
+    },
+    {
+      "epoch": 48.03,
+      "learning_rate": 1.7910875483593637e-06,
+      "loss": 0.2345,
+      "step": 6800
+    },
+    {
+      "epoch": 48.1,
+      "learning_rate": 1.7194440464249894e-06,
+      "loss": 0.3091,
+      "step": 6810
+    },
+    {
+      "epoch": 48.17,
+      "learning_rate": 1.6478005444906148e-06,
+      "loss": 0.2844,
+      "step": 6820
+    },
+    {
+      "epoch": 48.24,
+      "learning_rate": 1.57615704255624e-06,
+      "loss": 0.2919,
+      "step": 6830
+    },
+    {
+      "epoch": 48.31,
+      "learning_rate": 1.5045135406218657e-06,
+      "loss": 0.2631,
+      "step": 6840
+    },
+    {
+      "epoch": 48.39,
+      "learning_rate": 1.4328700386874911e-06,
+      "loss": 0.2954,
+      "step": 6850
+    },
+    {
+      "epoch": 48.46,
+      "learning_rate": 1.3612265367531166e-06,
+      "loss": 0.2536,
+      "step": 6860
+    },
+    {
+      "epoch": 48.53,
+      "learning_rate": 1.289583034818742e-06,
+      "loss": 0.2613,
+      "step": 6870
+    },
+    {
+      "epoch": 48.6,
+      "learning_rate": 1.2179395328843675e-06,
+      "loss": 0.2801,
+      "step": 6880
+    },
+    {
+      "epoch": 48.67,
+      "learning_rate": 1.146296030949993e-06,
+      "loss": 0.2491,
+      "step": 6890
+    },
+    {
+      "epoch": 48.74,
+      "learning_rate": 1.0746525290156184e-06,
+      "loss": 0.2556,
+      "step": 6900
+    },
+    {
+      "epoch": 48.81,
+      "learning_rate": 1.0030090270812438e-06,
+      "loss": 0.2577,
+      "step": 6910
+    },
+    {
+      "epoch": 48.88,
+      "learning_rate": 9.313655251468692e-07,
+      "loss": 0.2781,
+      "step": 6920
+    },
+    {
+      "epoch": 48.95,
+      "learning_rate": 8.597220232124947e-07,
+      "loss": 0.3252,
+      "step": 6930
+    },
+    {
+      "epoch": 49.0,
+      "eval_accuracy": 0.9716231555051078,
+      "eval_loss": 0.0857309103012085,
+      "eval_runtime": 6.3021,
+      "eval_samples_per_second": 139.794,
+      "eval_steps_per_second": 17.613,
+      "step": 6937
+    },
+    {
+      "epoch": 49.02,
+      "learning_rate": 7.8807852127812e-07,
+      "loss": 0.2746,
+      "step": 6940
+    },
+    {
+      "epoch": 49.09,
+      "learning_rate": 7.164350193437456e-07,
+      "loss": 0.2916,
+      "step": 6950
+    },
+    {
+      "epoch": 49.16,
+      "learning_rate": 6.44791517409371e-07,
+      "loss": 0.3324,
+      "step": 6960
+    },
+    {
+      "epoch": 49.23,
+      "learning_rate": 5.731480154749965e-07,
+      "loss": 0.2405,
+      "step": 6970
+    },
+    {
+      "epoch": 49.3,
+      "learning_rate": 5.015045135406219e-07,
+      "loss": 0.2438,
+      "step": 6980
+    },
+    {
+      "epoch": 49.37,
+      "learning_rate": 4.2986101160624734e-07,
+      "loss": 0.2355,
+      "step": 6990
+    },
+    {
+      "epoch": 49.45,
+      "learning_rate": 3.582175096718728e-07,
+      "loss": 0.2183,
+      "step": 7000
+    },
+    {
+      "epoch": 49.52,
+      "learning_rate": 2.8657400773749823e-07,
+      "loss": 0.251,
+      "step": 7010
+    },
+    {
+      "epoch": 49.59,
+      "learning_rate": 2.1493050580312367e-07,
+      "loss": 0.3174,
+      "step": 7020
+    },
+    {
+      "epoch": 49.66,
+      "learning_rate": 1.4328700386874911e-07,
+      "loss": 0.237,
+      "step": 7030
+    },
+    {
+      "epoch": 49.73,
+      "learning_rate": 7.164350193437456e-08,
+      "loss": 0.2395,
+      "step": 7040
+    },
+    {
+      "epoch": 49.8,
+      "learning_rate": 0.0,
+      "loss": 0.274,
+      "step": 7050
+    },
+    {
+      "epoch": 49.8,
+      "eval_accuracy": 0.9761634506242906,
+      "eval_loss": 0.08129376918077469,
+      "eval_runtime": 7.191,
+      "eval_samples_per_second": 122.514,
+      "eval_steps_per_second": 15.436,
+      "step": 7050
+    },
+    {
+      "epoch": 49.8,
+      "step": 7050,
+      "total_flos": 8.379994919518495e+18,
+      "train_loss": 0.43237863398612814,
+      "train_runtime": 6732.481,
+      "train_samples_per_second": 58.842,
+      "train_steps_per_second": 1.047
     }
   ],
-  "max_steps": 1410,
-  "num_train_epochs": 10,
-  "total_flos": 1.6760839403221402e+18,
+  "max_steps": 7050,
+  "num_train_epochs": 50,
+  "total_flos": 8.379994919518495e+18,
   "trial_name": null,
   "trial_params": null
 }