diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -3,4120 +3,4078 @@
   "best_model_checkpoint": null,
   "epoch": 5.0,
   "eval_steps": 500,
-  "global_step": 2890,
+  "global_step": 2860,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.0,
-      "grad_norm": 4.816294846566506,
-      "learning_rate": 1.7301038062283738e-08,
-      "loss": 0.8531,
+      "grad_norm": 5.587369652925466,
+      "learning_rate": 1.7482517482517483e-08,
+      "loss": 0.8755,
       "step": 1
     },
     {
       "epoch": 0.01,
-      "grad_norm": 4.872596535971759,
-      "learning_rate": 8.65051903114187e-08,
-      "loss": 0.8583,
+      "grad_norm": 5.500934350042366,
+      "learning_rate": 8.741258741258742e-08,
+      "loss": 0.8733,
       "step": 5
     },
     {
       "epoch": 0.02,
-      "grad_norm": 3.6310650985999184,
-      "learning_rate": 1.730103806228374e-07,
-      "loss": 0.8626,
+      "grad_norm": 4.778154128284402,
+      "learning_rate": 1.7482517482517484e-07,
+      "loss": 0.8812,
       "step": 10
     },
     {
       "epoch": 0.03,
-      "grad_norm": 3.744415952772262,
-      "learning_rate": 2.5951557093425607e-07,
-      "loss": 0.8594,
+      "grad_norm": 4.157465345945996,
+      "learning_rate": 2.622377622377623e-07,
+      "loss": 0.8595,
       "step": 15
     },
     {
       "epoch": 0.03,
-      "grad_norm": 4.189549391166134,
-      "learning_rate": 3.460207612456748e-07,
-      "loss": 0.8542,
+      "grad_norm": 4.01463883737159,
+      "learning_rate": 3.496503496503497e-07,
+      "loss": 0.8627,
       "step": 20
     },
     {
       "epoch": 0.04,
-      "grad_norm": 2.388097158161717,
-      "learning_rate": 4.325259515570934e-07,
-      "loss": 0.8489,
+      "grad_norm": 2.545403355372151,
+      "learning_rate": 4.3706293706293707e-07,
+      "loss": 0.8508,
       "step": 25
     },
     {
       "epoch": 0.05,
-      "grad_norm": 2.131273138920553,
-      "learning_rate": 5.190311418685121e-07,
-      "loss": 0.8319,
+      "grad_norm": 2.2986226033618844,
+      "learning_rate": 5.244755244755246e-07,
+      "loss": 0.8633,
       "step": 30
     },
     {
       "epoch": 0.06,
-      "grad_norm": 1.9741009575137782,
-      "learning_rate": 6.055363321799308e-07,
-      "loss": 0.8334,
+      "grad_norm": 2.0882287948426668,
+      "learning_rate": 6.118881118881119e-07,
+      "loss": 0.822,
       "step": 35
     },
     {
       "epoch": 0.07,
-      "grad_norm": 1.9300494756762427,
-      "learning_rate": 6.920415224913496e-07,
-      "loss": 0.8206,
+      "grad_norm": 1.9905683806362928,
+      "learning_rate": 6.993006993006994e-07,
+      "loss": 0.8195,
       "step": 40
     },
     {
       "epoch": 0.08,
-      "grad_norm": 1.7903640743849025,
-      "learning_rate": 7.785467128027683e-07,
-      "loss": 0.7872,
+      "grad_norm": 1.8546442781858452,
+      "learning_rate": 7.867132867132868e-07,
+      "loss": 0.8423,
       "step": 45
     },
     {
       "epoch": 0.09,
-      "grad_norm": 1.9336160245419374,
-      "learning_rate": 8.650519031141868e-07,
-      "loss": 0.8006,
+      "grad_norm": 1.9539088930354693,
+      "learning_rate": 8.741258741258741e-07,
+      "loss": 0.7999,
       "step": 50
     },
     {
       "epoch": 0.1,
-      "grad_norm": 1.8895155835527617,
-      "learning_rate": 9.515570934256056e-07,
-      "loss": 0.8079,
+      "grad_norm": 1.8944469761122955,
+      "learning_rate": 9.615384615384617e-07,
+      "loss": 0.8234,
       "step": 55
     },
     {
       "epoch": 0.1,
-      "grad_norm": 1.802498228874264,
-      "learning_rate": 1.0380622837370243e-06,
-      "loss": 0.803,
+      "grad_norm": 1.976723504907432,
+      "learning_rate": 1.0489510489510491e-06,
+      "loss": 0.8047,
       "step": 60
     },
     {
       "epoch": 0.11,
-      "grad_norm": 1.777164779599201,
-      "learning_rate": 1.124567474048443e-06,
-      "loss": 0.8124,
+      "grad_norm": 1.8383713435930058,
+      "learning_rate": 1.1363636363636364e-06,
+      "loss": 0.8046,
       "step": 65
     },
     {
       "epoch": 0.12,
-      "grad_norm": 1.8054173138128906,
-      "learning_rate": 1.2110726643598616e-06,
-      "loss": 0.8183,
+      "grad_norm": 1.9093043777289103,
+      "learning_rate": 1.2237762237762238e-06,
+      "loss": 0.7935,
       "step": 70
     },
     {
       "epoch": 0.13,
-      "grad_norm": 1.9350493231536463,
-      "learning_rate": 1.2975778546712802e-06,
-      "loss": 0.8029,
+      "grad_norm": 1.8651426939128903,
+      "learning_rate": 1.3111888111888113e-06,
+      "loss": 0.8036,
       "step": 75
     },
     {
       "epoch": 0.14,
-      "grad_norm": 1.8950537177996267,
-      "learning_rate": 1.3840830449826992e-06,
-      "loss": 0.8138,
+      "grad_norm": 1.8471862984018954,
+      "learning_rate": 1.3986013986013987e-06,
+      "loss": 0.8034,
       "step": 80
     },
     {
       "epoch": 0.15,
-      "grad_norm": 1.8019972698305375,
-      "learning_rate": 1.4705882352941177e-06,
-      "loss": 0.8036,
+      "grad_norm": 1.9262668779136634,
+      "learning_rate": 1.486013986013986e-06,
+      "loss": 0.8008,
       "step": 85
     },
     {
       "epoch": 0.16,
-      "grad_norm": 1.9102546088528127,
-      "learning_rate": 1.5570934256055365e-06,
-      "loss": 0.7893,
+      "grad_norm": 1.9222277764812408,
+      "learning_rate": 1.5734265734265736e-06,
+      "loss": 0.7835,
       "step": 90
     },
     {
-      "epoch": 0.16,
-      "grad_norm": 1.8718974603674414,
-      "learning_rate": 1.643598615916955e-06,
-      "loss": 0.7821,
+      "epoch": 0.17,
+      "grad_norm": 1.8688469770089136,
+      "learning_rate": 1.660839160839161e-06,
+      "loss": 0.7967,
       "step": 95
     },
     {
       "epoch": 0.17,
-      "grad_norm": 1.8905357930834985,
-      "learning_rate": 1.7301038062283736e-06,
-      "loss": 0.7811,
+      "grad_norm": 1.8814770181802394,
+      "learning_rate": 1.7482517482517483e-06,
+      "loss": 0.7998,
       "step": 100
     },
     {
       "epoch": 0.18,
-      "grad_norm": 1.8446962030261018,
-      "learning_rate": 1.8166089965397926e-06,
-      "loss": 0.7687,
+      "grad_norm": 2.037977585299295,
+      "learning_rate": 1.8356643356643357e-06,
+      "loss": 0.8056,
       "step": 105
     },
     {
       "epoch": 0.19,
-      "grad_norm": 1.9762667024263982,
-      "learning_rate": 1.9031141868512112e-06,
-      "loss": 0.7876,
+      "grad_norm": 2.001270020691376,
+      "learning_rate": 1.9230769230769234e-06,
+      "loss": 0.7756,
       "step": 110
     },
     {
       "epoch": 0.2,
-      "grad_norm": 1.9200753281543292,
-      "learning_rate": 1.9896193771626298e-06,
-      "loss": 0.7856,
+      "grad_norm": 1.9295825867916314,
+      "learning_rate": 2.0104895104895104e-06,
+      "loss": 0.788,
       "step": 115
     },
     {
       "epoch": 0.21,
-      "grad_norm": 1.9558456094506125,
-      "learning_rate": 2.0761245674740485e-06,
-      "loss": 0.7532,
+      "grad_norm": 2.0467456581571577,
+      "learning_rate": 2.0979020979020983e-06,
+      "loss": 0.7852,
       "step": 120
     },
     {
       "epoch": 0.22,
-      "grad_norm": 1.9761173872923619,
-      "learning_rate": 2.1626297577854673e-06,
-      "loss": 0.7732,
+      "grad_norm": 2.140303916653523,
+      "learning_rate": 2.1853146853146857e-06,
+      "loss": 0.7749,
       "step": 125
     },
     {
-      "epoch": 0.22,
-      "grad_norm": 1.9910728274837823,
-      "learning_rate": 2.249134948096886e-06,
-      "loss": 0.7694,
+      "epoch": 0.23,
+      "grad_norm": 2.093503142861556,
+      "learning_rate": 2.2727272727272728e-06,
+      "loss": 0.7905,
       "step": 130
     },
     {
-      "epoch": 0.23,
-      "grad_norm": 2.1510352769592007,
-      "learning_rate": 2.335640138408305e-06,
-      "loss": 0.7612,
+      "epoch": 0.24,
+      "grad_norm": 2.064301983027922,
+      "learning_rate": 2.36013986013986e-06,
+      "loss": 0.7659,
       "step": 135
     },
     {
       "epoch": 0.24,
-      "grad_norm": 2.0884221130022373,
-      "learning_rate": 2.4221453287197232e-06,
-      "loss": 0.7667,
+      "grad_norm": 1.997358747946483,
+      "learning_rate": 2.4475524475524477e-06,
+      "loss": 0.7831,
       "step": 140
     },
     {
       "epoch": 0.25,
-      "grad_norm": 2.0730526163013345,
-      "learning_rate": 2.508650519031142e-06,
-      "loss": 0.7734,
+      "grad_norm": 2.1027121897036274,
+      "learning_rate": 2.534965034965035e-06,
+      "loss": 0.7702,
       "step": 145
     },
     {
       "epoch": 0.26,
-      "grad_norm": 2.071966255411125,
-      "learning_rate": 2.5951557093425604e-06,
-      "loss": 0.7558,
+      "grad_norm": 2.1291996409910015,
+      "learning_rate": 2.6223776223776225e-06,
+      "loss": 0.7658,
       "step": 150
     },
     {
       "epoch": 0.27,
-      "grad_norm": 2.140324284318863,
-      "learning_rate": 2.6816608996539796e-06,
-      "loss": 0.7453,
+      "grad_norm": 41.9980343074638,
+      "learning_rate": 2.70979020979021e-06,
+      "loss": 0.7668,
       "step": 155
     },
     {
       "epoch": 0.28,
-      "grad_norm": 2.0794979181710866,
-      "learning_rate": 2.7681660899653983e-06,
-      "loss": 0.7413,
+      "grad_norm": 2.198061904985496,
+      "learning_rate": 2.7972027972027974e-06,
+      "loss": 0.7742,
       "step": 160
     },
     {
       "epoch": 0.29,
-      "grad_norm": 2.079407078921554,
-      "learning_rate": 2.8546712802768167e-06,
-      "loss": 0.7331,
+      "grad_norm": 2.129322746365122,
+      "learning_rate": 2.8846153846153845e-06,
+      "loss": 0.7458,
       "step": 165
     },
     {
-      "epoch": 0.29,
-      "grad_norm": 2.0051529944042445,
-      "learning_rate": 2.9411764705882355e-06,
-      "loss": 0.751,
+      "epoch": 0.3,
+      "grad_norm": 2.0235052008880867,
+      "learning_rate": 2.972027972027972e-06,
+      "loss": 0.7434,
       "step": 170
     },
     {
-      "epoch": 0.3,
-      "grad_norm": 2.075289040593278,
-      "learning_rate": 3.027681660899654e-06,
-      "loss": 0.7369,
+      "epoch": 0.31,
+      "grad_norm": 2.0254135513369156,
+      "learning_rate": 3.0594405594405598e-06,
+      "loss": 0.7338,
       "step": 175
     },
     {
       "epoch": 0.31,
-      "grad_norm": 2.009911706722013,
-      "learning_rate": 3.114186851211073e-06,
-      "loss": 0.7239,
+      "grad_norm": 1.985119509495075,
+      "learning_rate": 3.1468531468531472e-06,
+      "loss": 0.7341,
       "step": 180
     },
     {
       "epoch": 0.32,
-      "grad_norm": 2.114683663985779,
-      "learning_rate": 3.200692041522492e-06,
-      "loss": 0.7219,
+      "grad_norm": 2.250959168282524,
+      "learning_rate": 3.2342657342657347e-06,
+      "loss": 0.7307,
       "step": 185
     },
     {
       "epoch": 0.33,
-      "grad_norm": 2.1266461925060343,
-      "learning_rate": 3.28719723183391e-06,
-      "loss": 0.7163,
+      "grad_norm": 1.9968379378052319,
+      "learning_rate": 3.321678321678322e-06,
+      "loss": 0.7385,
       "step": 190
     },
     {
       "epoch": 0.34,
-      "grad_norm": 2.0699467756030048,
-      "learning_rate": 3.373702422145329e-06,
-      "loss": 0.7246,
+      "grad_norm": 2.630990506587568,
+      "learning_rate": 3.409090909090909e-06,
+      "loss": 0.7328,
       "step": 195
     },
     {
       "epoch": 0.35,
-      "grad_norm": 1.9178177043552207,
-      "learning_rate": 3.4602076124567473e-06,
-      "loss": 0.715,
+      "grad_norm": 3.57779109043985,
+      "learning_rate": 3.4965034965034966e-06,
+      "loss": 0.7429,
       "step": 200
     },
     {
-      "epoch": 0.35,
-      "grad_norm": 1.9401101711848872,
-      "learning_rate": 3.5467128027681665e-06,
-      "loss": 0.6888,
+      "epoch": 0.36,
+      "grad_norm": 26.759360492511966,
+      "learning_rate": 3.583916083916084e-06,
+      "loss": 0.7541,
       "step": 205
     },
     {
-      "epoch": 0.36,
-      "grad_norm": 1.7637994756169328,
-      "learning_rate": 3.6332179930795853e-06,
-      "loss": 0.7095,
+      "epoch": 0.37,
+      "grad_norm": 2.779819046550619,
+      "learning_rate": 3.6713286713286715e-06,
+      "loss": 0.7457,
       "step": 210
     },
     {
-      "epoch": 0.37,
-      "grad_norm": 1.8879618486444918,
-      "learning_rate": 3.7197231833910036e-06,
-      "loss": 0.7159,
+      "epoch": 0.38,
+      "grad_norm": 1.9676288431898625,
+      "learning_rate": 3.7587412587412593e-06,
+      "loss": 0.7347,
       "step": 215
     },
     {
       "epoch": 0.38,
-      "grad_norm": 1.7360031299470728,
-      "learning_rate": 3.8062283737024224e-06,
-      "loss": 0.7111,
+      "grad_norm": 1.7250524315438622,
+      "learning_rate": 3.846153846153847e-06,
+      "loss": 0.7383,
       "step": 220
     },
     {
       "epoch": 0.39,
-      "grad_norm": 1.6156910333813757,
-      "learning_rate": 3.892733564013842e-06,
-      "loss": 0.7136,
+      "grad_norm": 1.8839750025590591,
+      "learning_rate": 3.933566433566433e-06,
+      "loss": 0.7318,
       "step": 225
     },
     {
       "epoch": 0.4,
-      "grad_norm": 1.5128961988226293,
-      "learning_rate": 3.9792387543252595e-06,
-      "loss": 0.7199,
+      "grad_norm": 2.6586996717057443,
+      "learning_rate": 4.020979020979021e-06,
+      "loss": 0.7227,
       "step": 230
     },
     {
       "epoch": 0.41,
-      "grad_norm": 1.5314262584578413,
-      "learning_rate": 4.065743944636678e-06,
+      "grad_norm": 3.585620352276218,
+      "learning_rate": 4.108391608391608e-06,
       "loss": 0.7051,
       "step": 235
     },
     {
       "epoch": 0.42,
-      "grad_norm": 1.7503445297432856,
-      "learning_rate": 4.152249134948097e-06,
-      "loss": 0.6951,
+      "grad_norm": 1.676940387665252,
+      "learning_rate": 4.195804195804197e-06,
+      "loss": 0.7296,
       "step": 240
     },
     {
-      "epoch": 0.42,
-      "grad_norm": 1.727205989077596,
-      "learning_rate": 4.238754325259516e-06,
-      "loss": 0.6967,
+      "epoch": 0.43,
+      "grad_norm": 1.4297941750638832,
+      "learning_rate": 4.283216783216784e-06,
+      "loss": 0.7034,
       "step": 245
     },
     {
-      "epoch": 0.43,
-      "grad_norm": 1.512173137191571,
-      "learning_rate": 4.325259515570935e-06,
-      "loss": 0.6951,
+      "epoch": 0.44,
+      "grad_norm": 1.5752492466565178,
+      "learning_rate": 4.3706293706293715e-06,
+      "loss": 0.7114,
       "step": 250
     },
     {
-      "epoch": 0.44,
-      "grad_norm": 1.5906205947939107,
-      "learning_rate": 4.411764705882353e-06,
-      "loss": 0.7177,
+      "epoch": 0.45,
+      "grad_norm": 23.603921846562006,
+      "learning_rate": 4.458041958041958e-06,
+      "loss": 0.7189,
       "step": 255
     },
     {
       "epoch": 0.45,
-      "grad_norm": 1.4011742158056826,
-      "learning_rate": 4.498269896193772e-06,
-      "loss": 0.6955,
+      "grad_norm": 1.4838961986379053,
+      "learning_rate": 4.5454545454545455e-06,
+      "loss": 0.7134,
       "step": 260
     },
     {
       "epoch": 0.46,
-      "grad_norm": 1.3418261511629705,
-      "learning_rate": 4.58477508650519e-06,
-      "loss": 0.6845,
+      "grad_norm": 1.553567707360373,
+      "learning_rate": 4.632867132867133e-06,
+      "loss": 0.7244,
       "step": 265
     },
     {
       "epoch": 0.47,
-      "grad_norm": 1.5187095864301963,
-      "learning_rate": 4.67128027681661e-06,
-      "loss": 0.7068,
+      "grad_norm": 1.6872621755065427,
+      "learning_rate": 4.72027972027972e-06,
+      "loss": 0.7221,
       "step": 270
     },
     {
       "epoch": 0.48,
-      "grad_norm": 1.62902883041799,
-      "learning_rate": 4.7577854671280285e-06,
-      "loss": 0.6937,
+      "grad_norm": 26.978554548474214,
+      "learning_rate": 4.807692307692308e-06,
+      "loss": 0.6954,
       "step": 275
     },
     {
-      "epoch": 0.48,
-      "grad_norm": 1.3634212608530258,
-      "learning_rate": 4.8442906574394464e-06,
-      "loss": 0.6858,
+      "epoch": 0.49,
+      "grad_norm": 1.5442233324126737,
+      "learning_rate": 4.895104895104895e-06,
+      "loss": 0.7098,
       "step": 280
     },
     {
-      "epoch": 0.49,
-      "grad_norm": 1.5038430309774804,
-      "learning_rate": 4.930795847750865e-06,
-      "loss": 0.6988,
+      "epoch": 0.5,
+      "grad_norm": 1.5826467174296213,
+      "learning_rate": 4.982517482517483e-06,
+      "loss": 0.6956,
       "step": 285
     },
     {
-      "epoch": 0.5,
-      "grad_norm": 1.3823728989860273,
-      "learning_rate": 4.999998176402445e-06,
-      "loss": 0.6895,
+      "epoch": 0.51,
+      "grad_norm": 1.529332769657097,
+      "learning_rate": 4.999970207167611e-06,
+      "loss": 0.7054,
       "step": 290
     },
     {
-      "epoch": 0.51,
-      "grad_norm": 1.7851921485205167,
-      "learning_rate": 4.999934350767369e-06,
-      "loss": 0.6972,
+      "epoch": 0.52,
+      "grad_norm": 1.4134102929361827,
+      "learning_rate": 4.999849175003032e-06,
+      "loss": 0.6925,
       "step": 295
     },
     {
       "epoch": 0.52,
-      "grad_norm": 1.5012843187747695,
-      "learning_rate": 4.99977934791494e-06,
-      "loss": 0.6909,
+      "grad_norm": 1.5561380803552274,
+      "learning_rate": 4.999635045958129e-06,
+      "loss": 0.6777,
       "step": 300
     },
     {
       "epoch": 0.53,
-      "grad_norm": 1.4846464217080884,
-      "learning_rate": 4.9995331734983985e-06,
-      "loss": 0.6784,
+      "grad_norm": 1.7122419800735684,
+      "learning_rate": 4.999327828007281e-06,
+      "loss": 0.6782,
       "step": 305
     },
     {
       "epoch": 0.54,
-      "grad_norm": 1.4416671840566515,
-      "learning_rate": 4.999195836496179e-06,
-      "loss": 0.6885,
+      "grad_norm": 3.884234946585116,
+      "learning_rate": 4.998927532591592e-06,
+      "loss": 0.714,
       "step": 310
     },
     {
-      "epoch": 0.54,
-      "grad_norm": 1.3968335334596063,
-      "learning_rate": 4.998767349211587e-06,
-      "loss": 0.6632,
+      "epoch": 0.55,
+      "grad_norm": 1.5140404948967683,
+      "learning_rate": 4.998434174618464e-06,
+      "loss": 0.6969,
       "step": 315
     },
     {
-      "epoch": 0.55,
-      "grad_norm": 1.4727540225441376,
-      "learning_rate": 4.9982477272723404e-06,
-      "loss": 0.6997,
+      "epoch": 0.56,
+      "grad_norm": 1.5157313273979758,
+      "learning_rate": 4.997847772461038e-06,
+      "loss": 0.6937,
       "step": 320
     },
     {
-      "epoch": 0.56,
-      "grad_norm": 2.0150510241724295,
-      "learning_rate": 4.997636989630012e-06,
-      "loss": 0.681,
+      "epoch": 0.57,
+      "grad_norm": 1.4006324095309106,
+      "learning_rate": 4.997168347957521e-06,
+      "loss": 0.6707,
       "step": 325
     },
     {
-      "epoch": 0.57,
-      "grad_norm": 1.5786095690969104,
-      "learning_rate": 4.996935158559329e-06,
-      "loss": 0.7052,
+      "epoch": 0.58,
+      "grad_norm": 1.4555169305723101,
+      "learning_rate": 4.996395926410354e-06,
+      "loss": 0.6901,
       "step": 330
     },
     {
-      "epoch": 0.58,
-      "grad_norm": 1.4930115902628418,
-      "learning_rate": 4.9961422596573674e-06,
-      "loss": 0.683,
+      "epoch": 0.59,
+      "grad_norm": 1.4699402702938433,
+      "learning_rate": 4.995530536585293e-06,
+      "loss": 0.6954,
       "step": 335
     },
     {
       "epoch": 0.59,
-      "grad_norm": 1.417488194165555,
-      "learning_rate": 4.995258321842611e-06,
-      "loss": 0.6724,
+      "grad_norm": 1.423966140797764,
+      "learning_rate": 4.994572210710315e-06,
+      "loss": 0.6834,
       "step": 340
     },
     {
       "epoch": 0.6,
-      "grad_norm": 1.5814802052424026,
-      "learning_rate": 4.994283377353902e-06,
-      "loss": 0.708,
+      "grad_norm": 1.4114672539255717,
+      "learning_rate": 4.993520984474435e-06,
+      "loss": 0.7091,
       "step": 345
     },
     {
       "epoch": 0.61,
-      "grad_norm": 1.626327112384558,
-      "learning_rate": 4.993217461749266e-06,
-      "loss": 0.6769,
+      "grad_norm": 1.4495020382615775,
+      "learning_rate": 4.9923768970263675e-06,
+      "loss": 0.6909,
       "step": 350
     },
     {
-      "epoch": 0.61,
-      "grad_norm": 1.4561720499145596,
-      "learning_rate": 4.992060613904611e-06,
-      "loss": 0.6827,
+      "epoch": 0.62,
+      "grad_norm": 1.5759646703195747,
+      "learning_rate": 4.991139990973071e-06,
+      "loss": 0.6943,
       "step": 355
     },
     {
-      "epoch": 0.62,
-      "grad_norm": 1.4209579264748098,
-      "learning_rate": 4.990812876012311e-06,
-      "loss": 0.6843,
+      "epoch": 0.63,
+      "grad_norm": 1.5472655442648218,
+      "learning_rate": 4.989810312378165e-06,
+      "loss": 0.699,
       "step": 360
     },
     {
-      "epoch": 0.63,
-      "grad_norm": 1.3963649914734755,
-      "learning_rate": 4.989474293579669e-06,
-      "loss": 0.678,
+      "epoch": 0.64,
+      "grad_norm": 1.4257086908354295,
+      "learning_rate": 4.988387910760206e-06,
+      "loss": 0.688,
       "step": 365
     },
     {
-      "epoch": 0.64,
-      "grad_norm": 1.5340840894447103,
-      "learning_rate": 4.988044915427257e-06,
-      "loss": 0.6753,
+      "epoch": 0.65,
+      "grad_norm": 1.4575544094857862,
+      "learning_rate": 4.986872839090853e-06,
+      "loss": 0.6715,
       "step": 370
     },
     {
-      "epoch": 0.65,
-      "grad_norm": 1.4322002446736155,
-      "learning_rate": 4.986524793687131e-06,
-      "loss": 0.6913,
+      "epoch": 0.66,
+      "grad_norm": 1.5103716324462753,
+      "learning_rate": 4.985265153792887e-06,
+      "loss": 0.6818,
       "step": 375
     },
     {
       "epoch": 0.66,
-      "grad_norm": 1.494759555714755,
-      "learning_rate": 4.984913983800936e-06,
-      "loss": 0.6601,
+      "grad_norm": 1.8027777365944309,
+      "learning_rate": 4.983564914738113e-06,
+      "loss": 0.6927,
       "step": 380
     },
     {
       "epoch": 0.67,
-      "grad_norm": 1.5286688918221005,
-      "learning_rate": 4.983212544517878e-06,
-      "loss": 0.6708,
+      "grad_norm": 1.4256278816465284,
+      "learning_rate": 4.981772185245135e-06,
+      "loss": 0.6886,
       "step": 385
     },
     {
-      "epoch": 0.67,
-      "grad_norm": 1.52282481083145,
-      "learning_rate": 4.9814205378925895e-06,
-      "loss": 0.6582,
+      "epoch": 0.68,
+      "grad_norm": 1.795771667278661,
+      "learning_rate": 4.9798870320769884e-06,
+      "loss": 0.6716,
       "step": 390
     },
     {
-      "epoch": 0.68,
-      "grad_norm": 1.6753513410463947,
-      "learning_rate": 4.979538029282856e-06,
-      "loss": 0.679,
+      "epoch": 0.69,
+      "grad_norm": 1.3370646701877693,
+      "learning_rate": 4.9779095254386605e-06,
+      "loss": 0.6939,
       "step": 395
     },
     {
-      "epoch": 0.69,
-      "grad_norm": 1.5212718767367754,
-      "learning_rate": 4.977565087347239e-06,
-      "loss": 0.667,
+      "epoch": 0.7,
+      "grad_norm": 1.4886996027080686,
+      "learning_rate": 4.975839738974473e-06,
+      "loss": 0.698,
       "step": 400
     },
     {
-      "epoch": 0.7,
-      "grad_norm": 1.4524081871963455,
-      "learning_rate": 4.975501784042572e-06,
-      "loss": 0.6707,
+      "epoch": 0.71,
+      "grad_norm": 1.4334461526468911,
+      "learning_rate": 4.9736777497653425e-06,
+      "loss": 0.6706,
       "step": 405
     },
     {
-      "epoch": 0.71,
-      "grad_norm": 1.7921858867602187,
-      "learning_rate": 4.973348194621329e-06,
-      "loss": 0.6741,
+      "epoch": 0.72,
+      "grad_norm": 1.500198901246391,
+      "learning_rate": 4.971423638325906e-06,
+      "loss": 0.6687,
       "step": 410
     },
     {
-      "epoch": 0.72,
-      "grad_norm": 1.5845517579748196,
-      "learning_rate": 4.9711043976288935e-06,
-      "loss": 0.6791,
+      "epoch": 0.73,
+      "grad_norm": 1.5396586819167384,
+      "learning_rate": 4.969077488601525e-06,
+      "loss": 0.6734,
       "step": 415
     },
     {
       "epoch": 0.73,
-      "grad_norm": 1.4343033765555364,
-      "learning_rate": 4.968770474900677e-06,
-      "loss": 0.6607,
+      "grad_norm": 1.4725153996617695,
+      "learning_rate": 4.966639387965158e-06,
+      "loss": 0.6999,
       "step": 420
     },
     {
       "epoch": 0.74,
-      "grad_norm": 1.5295122616615218,
-      "learning_rate": 4.966346511559149e-06,
-      "loss": 0.6624,
+      "grad_norm": 1.5885070209974053,
+      "learning_rate": 4.964109427214111e-06,
+      "loss": 0.6853,
       "step": 425
     },
     {
-      "epoch": 0.74,
-      "grad_norm": 1.4896437285857649,
-      "learning_rate": 4.963832596010723e-06,
-      "loss": 0.6677,
+      "epoch": 0.75,
+      "grad_norm": 1.7768817102056875,
+      "learning_rate": 4.961487700566646e-06,
+      "loss": 0.6758,
       "step": 430
     },
     {
-      "epoch": 0.75,
-      "grad_norm": 1.5044244825090314,
-      "learning_rate": 4.961228819942538e-06,
-      "loss": 0.6722,
+      "epoch": 0.76,
+      "grad_norm": 1.4156688514395073,
+      "learning_rate": 4.958774305658484e-06,
+      "loss": 0.682,
       "step": 435
     },
     {
-      "epoch": 0.76,
-      "grad_norm": 1.4992891216972632,
-      "learning_rate": 4.958535278319109e-06,
-      "loss": 0.6529,
+      "epoch": 0.77,
+      "grad_norm": 1.615821708366966,
+      "learning_rate": 4.955969343539162e-06,
+      "loss": 0.6623,
       "step": 440
     },
     {
-      "epoch": 0.77,
-      "grad_norm": 1.5676366345308723,
-      "learning_rate": 4.955752069378867e-06,
-      "loss": 0.6682,
+      "epoch": 0.78,
+      "grad_norm": 1.5778334799669915,
+      "learning_rate": 4.95307291866827e-06,
+      "loss": 0.7118,
       "step": 445
     },
     {
-      "epoch": 0.78,
-      "grad_norm": 1.566761422297736,
-      "learning_rate": 4.952879294630577e-06,
-      "loss": 0.6674,
+      "epoch": 0.79,
+      "grad_norm": 1.5459039639149776,
+      "learning_rate": 4.9500851389115645e-06,
+      "loss": 0.6822,
       "step": 450
     },
     {
-      "epoch": 0.79,
-      "grad_norm": 1.593607408287518,
-      "learning_rate": 4.949917058849635e-06,
-      "loss": 0.6685,
+      "epoch": 0.8,
+      "grad_norm": 1.950166551192635,
+      "learning_rate": 4.947006115536947e-06,
+      "loss": 0.6896,
       "step": 455
     },
     {
       "epoch": 0.8,
-      "grad_norm": 1.4655841734364505,
-      "learning_rate": 4.946865470074241e-06,
-      "loss": 0.6706,
+      "grad_norm": 1.4583416417638757,
+      "learning_rate": 4.943835963210324e-06,
+      "loss": 0.6737,
       "step": 460
     },
     {
-      "epoch": 0.8,
-      "grad_norm": 1.4680095746651978,
-      "learning_rate": 4.9437246396014695e-06,
-      "loss": 0.6503,
+      "epoch": 0.81,
+      "grad_norm": 1.3818410898501687,
+      "learning_rate": 4.9405747999913355e-06,
+      "loss": 0.6663,
       "step": 465
     },
     {
-      "epoch": 0.81,
-      "grad_norm": 1.6414158311618077,
-      "learning_rate": 4.940494681983201e-06,
-      "loss": 0.6483,
+      "epoch": 0.82,
+      "grad_norm": 1.642028714364656,
+      "learning_rate": 4.937222747328956e-06,
+      "loss": 0.6702,
       "step": 470
     },
     {
-      "epoch": 0.82,
-      "grad_norm": 1.647070417024019,
-      "learning_rate": 4.9371757150219475e-06,
-      "loss": 0.6708,
+      "epoch": 0.83,
+      "grad_norm": 1.4162244952189953,
+      "learning_rate": 4.933779930056975e-06,
+      "loss": 0.6709,
       "step": 475
     },
     {
-      "epoch": 0.83,
-      "grad_norm": 1.5786850858034276,
-      "learning_rate": 4.933767859766557e-06,
-      "loss": 0.6406,
+      "epoch": 0.84,
+      "grad_norm": 1.5391809516965524,
+      "learning_rate": 4.9302464763893474e-06,
+      "loss": 0.6581,
       "step": 480
     },
     {
-      "epoch": 0.84,
-      "grad_norm": 2.3967395390307664,
-      "learning_rate": 4.930271240507797e-06,
-      "loss": 0.6469,
+      "epoch": 0.85,
+      "grad_norm": 1.4577041041003882,
+      "learning_rate": 4.926622517915417e-06,
+      "loss": 0.6523,
       "step": 485
     },
     {
-      "epoch": 0.85,
-      "grad_norm": 1.863770908509767,
-      "learning_rate": 4.926685984773822e-06,
-      "loss": 0.6721,
+      "epoch": 0.86,
+      "grad_norm": 1.4232896131606396,
+      "learning_rate": 4.9229081895950185e-06,
+      "loss": 0.6828,
       "step": 490
     },
     {
-      "epoch": 0.86,
-      "grad_norm": 1.4091042690875146,
-      "learning_rate": 4.923012223325522e-06,
-      "loss": 0.6418,
+      "epoch": 0.87,
+      "grad_norm": 1.5308862157877259,
+      "learning_rate": 4.9191036297534455e-06,
+      "loss": 0.6671,
       "step": 495
     },
     {
       "epoch": 0.87,
-      "grad_norm": 1.4211964673604418,
-      "learning_rate": 4.919250090151754e-06,
-      "loss": 0.6384,
+      "grad_norm": 1.4998597450310402,
+      "learning_rate": 4.91520898007631e-06,
+      "loss": 0.6559,
       "step": 500
     },
     {
-      "epoch": 0.87,
-      "grad_norm": 1.5350233939025446,
-      "learning_rate": 4.915399722464455e-06,
-      "loss": 0.6365,
+      "epoch": 0.88,
+      "grad_norm": 1.5868254943762528,
+      "learning_rate": 4.911224385604255e-06,
+      "loss": 0.6731,
       "step": 505
     },
     {
-      "epoch": 0.88,
-      "grad_norm": 1.5608101804229912,
-      "learning_rate": 4.911461260693639e-06,
-      "loss": 0.6349,
+      "epoch": 0.89,
+      "grad_norm": 1.5018477168697573,
+      "learning_rate": 4.907149994727559e-06,
+      "loss": 0.6643,
       "step": 510
     },
     {
-      "epoch": 0.89,
-      "grad_norm": 1.471894754845902,
-      "learning_rate": 4.907434848482268e-06,
-      "loss": 0.6324,
+      "epoch": 0.9,
+      "grad_norm": 1.4716960928109946,
+      "learning_rate": 4.902985959180608e-06,
+      "loss": 0.6556,
       "step": 515
     },
     {
-      "epoch": 0.9,
-      "grad_norm": 1.5341990166130748,
-      "learning_rate": 4.903320632681026e-06,
-      "loss": 0.6434,
+      "epoch": 0.91,
+      "grad_norm": 1.4130064465598204,
+      "learning_rate": 4.8987324340362445e-06,
+      "loss": 0.6675,
       "step": 520
     },
     {
-      "epoch": 0.91,
-      "grad_norm": 1.4680803495559631,
-      "learning_rate": 4.899118763342952e-06,
-      "loss": 0.6466,
+      "epoch": 0.92,
+      "grad_norm": 1.4999119765126108,
+      "learning_rate": 4.894389577699994e-06,
+      "loss": 0.6484,
       "step": 525
     },
     {
-      "epoch": 0.92,
-      "grad_norm": 1.786833703301667,
-      "learning_rate": 4.89482939371797e-06,
-      "loss": 0.6599,
+      "epoch": 0.93,
+      "grad_norm": 1.4835440823652937,
+      "learning_rate": 4.889957551904164e-06,
+      "loss": 0.6566,
       "step": 530
     },
     {
-      "epoch": 0.93,
-      "grad_norm": 1.5749569225927729,
-      "learning_rate": 4.890452680247305e-06,
-      "loss": 0.641,
+      "epoch": 0.94,
+      "grad_norm": 1.5704867480691889,
+      "learning_rate": 4.885436521701824e-06,
+      "loss": 0.6589,
       "step": 535
     },
     {
-      "epoch": 0.93,
-      "grad_norm": 1.3756860207273784,
-      "learning_rate": 4.885988782557771e-06,
-      "loss": 0.6372,
+      "epoch": 0.94,
+      "grad_norm": 1.5154576884271134,
+      "learning_rate": 4.8808266554606535e-06,
+      "loss": 0.6476,
       "step": 540
     },
     {
-      "epoch": 0.94,
-      "grad_norm": 1.6093553203322744,
-      "learning_rate": 4.88143786345595e-06,
-      "loss": 0.6053,
+      "epoch": 0.95,
+      "grad_norm": 1.4656996812743353,
+      "learning_rate": 4.876128124856676e-06,
+      "loss": 0.6652,
       "step": 545
     },
     {
-      "epoch": 0.95,
-      "grad_norm": 1.417201052972172,
-      "learning_rate": 4.876800088922257e-06,
-      "loss": 0.6288,
+      "epoch": 0.96,
+      "grad_norm": 1.6670494857895617,
+      "learning_rate": 4.8713411048678635e-06,
+      "loss": 0.6711,
       "step": 550
     },
     {
-      "epoch": 0.96,
-      "grad_norm": 1.512172943891037,
-      "learning_rate": 4.872075628104884e-06,
-      "loss": 0.6352,
+      "epoch": 0.97,
+      "grad_norm": 12.070438425219901,
+      "learning_rate": 4.866465773767625e-06,
+      "loss": 0.6941,
       "step": 555
     },
     {
-      "epoch": 0.97,
-      "grad_norm": 1.6791678410000617,
-      "learning_rate": 4.867264653313633e-06,
-      "loss": 0.6473,
+      "epoch": 0.98,
+      "grad_norm": 1.9537677638893325,
+      "learning_rate": 4.861502313118157e-06,
+      "loss": 0.6606,
       "step": 560
     },
     {
-      "epoch": 0.98,
-      "grad_norm": 1.4591202016380542,
-      "learning_rate": 4.8623673400136295e-06,
-      "loss": 0.6364,
+      "epoch": 0.99,
+      "grad_norm": 2.0272499350973807,
+      "learning_rate": 4.856450907763693e-06,
+      "loss": 0.6506,
       "step": 565
     },
     {
-      "epoch": 0.99,
-      "grad_norm": 1.5298614351963449,
-      "learning_rate": 4.8573838668189225e-06,
-      "loss": 0.6295,
+      "epoch": 1.0,
+      "grad_norm": 1.7554436009464944,
+      "learning_rate": 4.851311745823616e-06,
+      "loss": 0.6695,
       "step": 570
     },
-    {
-      "epoch": 0.99,
-      "grad_norm": 1.5276546684651868,
-      "learning_rate": 4.852314415485973e-06,
-      "loss": 0.6331,
-      "step": 575
-    },
     {
       "epoch": 1.0,
-      "eval_loss": 0.622621476650238,
-      "eval_runtime": 330.1558,
-      "eval_samples_per_second": 22.78,
+      "eval_loss": 0.6609551310539246,
+      "eval_runtime": 330.32,
+      "eval_samples_per_second": 22.769,
       "eval_steps_per_second": 0.357,
-      "step": 578
+      "step": 572
     },
     {
-      "epoch": 1.0,
-      "grad_norm": 2.564141000897205,
-      "learning_rate": 4.847159170907022e-06,
-      "loss": 0.5882,
-      "step": 580
+      "epoch": 1.01,
+      "grad_norm": 2.145254776756945,
+      "learning_rate": 4.846085018685449e-06,
+      "loss": 0.6166,
+      "step": 575
     },
     {
       "epoch": 1.01,
-      "grad_norm": 1.702572539352643,
-      "learning_rate": 4.841918321103349e-06,
-      "loss": 0.5574,
-      "step": 585
+      "grad_norm": 1.7225148009689535,
+      "learning_rate": 4.84077092099773e-06,
+      "loss": 0.5855,
+      "step": 580
     },
     {
       "epoch": 1.02,
-      "grad_norm": 1.8542783824172295,
-      "learning_rate": 4.836592057218416e-06,
-      "loss": 0.5321,
-      "step": 590
+      "grad_norm": 1.5799570982565634,
+      "learning_rate": 4.835369650662767e-06,
+      "loss": 0.589,
+      "step": 585
     },
     {
       "epoch": 1.03,
-      "grad_norm": 1.6168277685945358,
-      "learning_rate": 4.83118057351089e-06,
-      "loss": 0.5301,
-      "step": 595
+      "grad_norm": 1.7146719749288577,
+      "learning_rate": 4.829881408829262e-06,
+      "loss": 0.5809,
+      "step": 590
     },
     {
       "epoch": 1.04,
-      "grad_norm": 1.7040979617839176,
-      "learning_rate": 4.825684067347565e-06,
-      "loss": 0.5369,
-      "step": 600
+      "grad_norm": 1.6210603375147141,
+      "learning_rate": 4.824306399884822e-06,
+      "loss": 0.5619,
+      "step": 595
     },
     {
       "epoch": 1.05,
-      "grad_norm": 1.6826187176546037,
-      "learning_rate": 4.82010273919616e-06,
-      "loss": 0.521,
-      "step": 605
+      "grad_norm": 1.8183201949889993,
+      "learning_rate": 4.81864483144835e-06,
+      "loss": 0.5821,
+      "step": 600
     },
     {
       "epoch": 1.06,
-      "grad_norm": 1.7537542015921341,
-      "learning_rate": 4.8144367926180084e-06,
-      "loss": 0.5362,
+      "grad_norm": 1.5245577439422557,
+      "learning_rate": 4.81289691436231e-06,
+      "loss": 0.5622,
+      "step": 605
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 1.6453822729343333,
+      "learning_rate": 4.807062862684874e-06,
+      "loss": 0.6047,
       "step": 610
     },
     {
-      "epoch": 1.06,
-      "grad_norm": 1.5596851176953166,
-      "learning_rate": 4.8086864342606345e-06,
-      "loss": 0.5421,
+      "epoch": 1.08,
+      "grad_norm": 1.7446190682981846,
+      "learning_rate": 4.801142893681955e-06,
+      "loss": 0.5696,
       "step": 615
     },
     {
-      "epoch": 1.07,
-      "grad_norm": 1.612414749994373,
-      "learning_rate": 4.802851873850214e-06,
-      "loss": 0.5542,
+      "epoch": 1.08,
+      "grad_norm": 1.6147794969895501,
+      "learning_rate": 4.795137227819113e-06,
+      "loss": 0.5523,
       "step": 620
     },
     {
-      "epoch": 1.08,
-      "grad_norm": 1.6043297699907004,
-      "learning_rate": 4.796933324183928e-06,
-      "loss": 0.526,
+      "epoch": 1.09,
+      "grad_norm": 1.6086737084266636,
+      "learning_rate": 4.7890460887533415e-06,
+      "loss": 0.5871,
       "step": 625
     },
     {
-      "epoch": 1.09,
-      "grad_norm": 1.5705483698698184,
-      "learning_rate": 4.790931001122202e-06,
-      "loss": 0.5267,
+      "epoch": 1.1,
+      "grad_norm": 1.5378514408370791,
+      "learning_rate": 4.782869703324746e-06,
+      "loss": 0.5687,
       "step": 630
     },
     {
-      "epoch": 1.1,
-      "grad_norm": 1.6281592397386129,
-      "learning_rate": 4.784845123580829e-06,
-      "loss": 0.5367,
+      "epoch": 1.11,
+      "grad_norm": 1.6893413050433177,
+      "learning_rate": 4.7766083015480876e-06,
+      "loss": 0.5748,
       "step": 635
     },
     {
-      "epoch": 1.11,
-      "grad_norm": 1.5966059820228211,
-      "learning_rate": 4.77867591352299e-06,
-      "loss": 0.535,
+      "epoch": 1.12,
+      "grad_norm": 1.6460419873518315,
+      "learning_rate": 4.770262116604224e-06,
+      "loss": 0.566,
       "step": 640
     },
     {
-      "epoch": 1.12,
-      "grad_norm": 1.8510817844041834,
-      "learning_rate": 4.7724235959511566e-06,
-      "loss": 0.5495,
+      "epoch": 1.13,
+      "grad_norm": 1.5227788895596694,
+      "learning_rate": 4.763831384831421e-06,
+      "loss": 0.542,
       "step": 645
     },
     {
-      "epoch": 1.12,
-      "grad_norm": 1.8997498768876209,
-      "learning_rate": 4.766088398898884e-06,
-      "loss": 0.5403,
+      "epoch": 1.14,
+      "grad_norm": 2.0335648439912455,
+      "learning_rate": 4.757316345716554e-06,
+      "loss": 0.5659,
       "step": 650
     },
     {
-      "epoch": 1.13,
-      "grad_norm": 1.7160489659947078,
-      "learning_rate": 4.759670553422494e-06,
-      "loss": 0.5347,
+      "epoch": 1.15,
+      "grad_norm": 1.5041239731809826,
+      "learning_rate": 4.750717241886186e-06,
+      "loss": 0.5529,
       "step": 655
     },
     {
-      "epoch": 1.14,
-      "grad_norm": 1.7971855635226874,
-      "learning_rate": 4.7531702935926505e-06,
-      "loss": 0.537,
+      "epoch": 1.15,
+      "grad_norm": 1.900645473248817,
+      "learning_rate": 4.744034319097536e-06,
+      "loss": 0.5726,
       "step": 660
     },
     {
-      "epoch": 1.15,
-      "grad_norm": 1.7064772479008743,
-      "learning_rate": 4.74658785648582e-06,
-      "loss": 0.5579,
+      "epoch": 1.16,
+      "grad_norm": 1.9943298775514051,
+      "learning_rate": 4.7372678262293235e-06,
+      "loss": 0.5709,
       "step": 665
     },
     {
-      "epoch": 1.16,
-      "grad_norm": 1.4680823102221994,
-      "learning_rate": 4.7399234821756285e-06,
-      "loss": 0.5276,
+      "epoch": 1.17,
+      "grad_norm": 1.6801680941350656,
+      "learning_rate": 4.7304180152725035e-06,
+      "loss": 0.5592,
       "step": 670
     },
     {
-      "epoch": 1.17,
-      "grad_norm": 2.0480120228699814,
-      "learning_rate": 4.733177413724098e-06,
-      "loss": 0.5417,
+      "epoch": 1.18,
+      "grad_norm": 1.6643180642477764,
+      "learning_rate": 4.723485141320877e-06,
+      "loss": 0.5755,
       "step": 675
     },
     {
-      "epoch": 1.18,
-      "grad_norm": 4.470377211022557,
-      "learning_rate": 4.726349897172791e-06,
-      "loss": 0.5292,
+      "epoch": 1.19,
+      "grad_norm": 1.7468633841451968,
+      "learning_rate": 4.716469462561595e-06,
+      "loss": 0.5737,
       "step": 680
     },
     {
-      "epoch": 1.19,
-      "grad_norm": 1.7455236825499205,
-      "learning_rate": 4.71944118153383e-06,
-      "loss": 0.5486,
+      "epoch": 1.2,
+      "grad_norm": 1.565224204964231,
+      "learning_rate": 4.709371240265543e-06,
+      "loss": 0.5713,
       "step": 685
     },
     {
-      "epoch": 1.19,
-      "grad_norm": 1.6229800058296617,
-      "learning_rate": 4.7124515187808186e-06,
-      "loss": 0.5403,
+      "epoch": 1.21,
+      "grad_norm": 1.704523210767029,
+      "learning_rate": 4.702190738777608e-06,
+      "loss": 0.5765,
       "step": 690
     },
     {
-      "epoch": 1.2,
-      "grad_norm": 1.6120491331538838,
-      "learning_rate": 4.705381163839652e-06,
-      "loss": 0.5212,
+      "epoch": 1.22,
+      "grad_norm": 1.643895772580424,
+      "learning_rate": 4.69492822550684e-06,
+      "loss": 0.5646,
       "step": 695
     },
     {
-      "epoch": 1.21,
-      "grad_norm": 1.5707477382476385,
-      "learning_rate": 4.698230374579216e-06,
-      "loss": 0.513,
+      "epoch": 1.22,
+      "grad_norm": 1.8094188227885684,
+      "learning_rate": 4.687583970916487e-06,
+      "loss": 0.5652,
       "step": 700
     },
     {
-      "epoch": 1.22,
-      "grad_norm": 1.8038956520870928,
-      "learning_rate": 4.690999411801987e-06,
-      "loss": 0.5505,
+      "epoch": 1.23,
+      "grad_norm": 1.7680146082507486,
+      "learning_rate": 4.680158248513924e-06,
+      "loss": 0.5789,
       "step": 705
     },
     {
-      "epoch": 1.23,
-      "grad_norm": 1.4977964665899992,
-      "learning_rate": 4.683688539234515e-06,
-      "loss": 0.5447,
+      "epoch": 1.24,
+      "grad_norm": 1.900227044096463,
+      "learning_rate": 4.6726513348404736e-06,
+      "loss": 0.5806,
       "step": 710
     },
     {
-      "epoch": 1.24,
-      "grad_norm": 1.64524294757822,
-      "learning_rate": 4.676298023517811e-06,
-      "loss": 0.5252,
+      "epoch": 1.25,
+      "grad_norm": 1.6673076686826593,
+      "learning_rate": 4.665063509461098e-06,
+      "loss": 0.5685,
       "step": 715
     },
     {
-      "epoch": 1.25,
-      "grad_norm": 1.8439638298190486,
-      "learning_rate": 4.668828134197615e-06,
-      "loss": 0.5288,
+      "epoch": 1.26,
+      "grad_norm": 1.8123226938487078,
+      "learning_rate": 4.657395054953992e-06,
+      "loss": 0.5886,
       "step": 720
     },
     {
-      "epoch": 1.25,
-      "grad_norm": 1.6551302109627803,
-      "learning_rate": 4.66127914371457e-06,
-      "loss": 0.5042,
+      "epoch": 1.27,
+      "grad_norm": 1.7238311806282647,
+      "learning_rate": 4.649646256900064e-06,
+      "loss": 0.5635,
       "step": 725
     },
     {
-      "epoch": 1.26,
-      "grad_norm": 1.8432213805384519,
-      "learning_rate": 4.653651327394284e-06,
-      "loss": 0.5075,
+      "epoch": 1.28,
+      "grad_norm": 1.6188026163610223,
+      "learning_rate": 4.641817403872293e-06,
+      "loss": 0.5731,
       "step": 730
     },
     {
-      "epoch": 1.27,
-      "grad_norm": 1.7991942501765485,
-      "learning_rate": 4.64594496343729e-06,
-      "loss": 0.5234,
+      "epoch": 1.28,
+      "grad_norm": 1.4913698739002144,
+      "learning_rate": 4.633908787424986e-06,
+      "loss": 0.5732,
       "step": 735
     },
     {
-      "epoch": 1.28,
-      "grad_norm": 1.560979934022783,
-      "learning_rate": 4.638160332908896e-06,
-      "loss": 0.5076,
+      "epoch": 1.29,
+      "grad_norm": 1.9372955869658495,
+      "learning_rate": 4.625920702082918e-06,
+      "loss": 0.5571,
       "step": 740
     },
     {
-      "epoch": 1.29,
-      "grad_norm": 1.5466509824875485,
-      "learning_rate": 4.630297719728939e-06,
-      "loss": 0.5204,
+      "epoch": 1.3,
+      "grad_norm": 1.7995058667015358,
+      "learning_rate": 4.617853445330367e-06,
+      "loss": 0.5607,
       "step": 745
     },
     {
-      "epoch": 1.3,
-      "grad_norm": 1.7134330421496313,
-      "learning_rate": 4.622357410661425e-06,
-      "loss": 0.5136,
+      "epoch": 1.31,
+      "grad_norm": 1.6274220841761087,
+      "learning_rate": 4.6097073176000325e-06,
+      "loss": 0.559,
       "step": 750
     },
     {
-      "epoch": 1.31,
-      "grad_norm": 1.9274332369383436,
-      "learning_rate": 4.6143396953040735e-06,
-      "loss": 0.5121,
+      "epoch": 1.32,
+      "grad_norm": 1.6385978867845423,
+      "learning_rate": 4.601482622261848e-06,
+      "loss": 0.5737,
       "step": 755
     },
     {
-      "epoch": 1.31,
-      "grad_norm": 1.4937551003793812,
-      "learning_rate": 4.606244866077752e-06,
-      "loss": 0.527,
+      "epoch": 1.33,
+      "grad_norm": 1.9049655839724005,
+      "learning_rate": 4.593179665611685e-06,
+      "loss": 0.5578,
       "step": 760
     },
     {
-      "epoch": 1.32,
-      "grad_norm": 2.658551374926485,
-      "learning_rate": 4.598073218215817e-06,
-      "loss": 0.5458,
+      "epoch": 1.34,
+      "grad_norm": 1.8343333848444217,
+      "learning_rate": 4.584798756859941e-06,
+      "loss": 0.5666,
       "step": 765
     },
     {
-      "epoch": 1.33,
-      "grad_norm": 2.224040870079962,
-      "learning_rate": 4.5898250497533395e-06,
-      "loss": 0.5436,
+      "epoch": 1.35,
+      "grad_norm": 1.6845673541242205,
+      "learning_rate": 4.5763402081200295e-06,
+      "loss": 0.5718,
       "step": 770
     },
     {
-      "epoch": 1.34,
-      "grad_norm": 2.7319048521917635,
-      "learning_rate": 4.581500661516237e-06,
-      "loss": 0.5313,
+      "epoch": 1.35,
+      "grad_norm": 1.772559396539105,
+      "learning_rate": 4.567804334396756e-06,
+      "loss": 0.5716,
       "step": 775
     },
     {
-      "epoch": 1.35,
-      "grad_norm": 1.6998409225129496,
-      "learning_rate": 4.573100357110306e-06,
-      "loss": 0.5173,
+      "epoch": 1.36,
+      "grad_norm": 1.748801886170466,
+      "learning_rate": 4.559191453574582e-06,
+      "loss": 0.5426,
       "step": 780
     },
     {
-      "epoch": 1.36,
-      "grad_norm": 1.5581752222534802,
-      "learning_rate": 4.564624442910144e-06,
-      "loss": 0.5301,
+      "epoch": 1.37,
+      "grad_norm": 1.5929925572389767,
+      "learning_rate": 4.550501886405795e-06,
+      "loss": 0.5748,
       "step": 785
     },
     {
-      "epoch": 1.37,
-      "grad_norm": 1.6962356595619175,
-      "learning_rate": 4.556073228047977e-06,
-      "loss": 0.5313,
+      "epoch": 1.38,
+      "grad_norm": 1.9926799279560756,
+      "learning_rate": 4.541735956498555e-06,
+      "loss": 0.5704,
       "step": 790
     },
     {
-      "epoch": 1.38,
-      "grad_norm": 1.8219089406617814,
-      "learning_rate": 4.547447024402385e-06,
-      "loss": 0.5335,
+      "epoch": 1.39,
+      "grad_norm": 1.6789209872224784,
+      "learning_rate": 4.532893990304848e-06,
+      "loss": 0.5581,
       "step": 795
     },
     {
-      "epoch": 1.38,
-      "grad_norm": 1.62352160937354,
-      "learning_rate": 4.5387461465869275e-06,
-      "loss": 0.5459,
+      "epoch": 1.4,
+      "grad_norm": 1.7322217739168355,
+      "learning_rate": 4.523976317108326e-06,
+      "loss": 0.5608,
       "step": 800
     },
     {
-      "epoch": 1.39,
-      "grad_norm": 1.5257648113589735,
-      "learning_rate": 4.529970911938669e-06,
-      "loss": 0.5111,
+      "epoch": 1.41,
+      "grad_norm": 29.58409303047747,
+      "learning_rate": 4.514983269012048e-06,
+      "loss": 0.5806,
       "step": 805
     },
     {
-      "epoch": 1.4,
-      "grad_norm": 2.5973864011536416,
-      "learning_rate": 4.521121640506605e-06,
-      "loss": 0.5352,
+      "epoch": 1.42,
+      "grad_norm": 2.307561877023173,
+      "learning_rate": 4.5059151809261085e-06,
+      "loss": 0.5521,
       "step": 810
     },
     {
-      "epoch": 1.41,
-      "grad_norm": 2.1347665405267384,
-      "learning_rate": 4.512198655039988e-06,
-      "loss": 0.5208,
+      "epoch": 1.42,
+      "grad_norm": 1.8846862840887817,
+      "learning_rate": 4.496772390555164e-06,
+      "loss": 0.5516,
       "step": 815
     },
     {
-      "epoch": 1.42,
-      "grad_norm": 1.8537528730656059,
-      "learning_rate": 4.503202280976557e-06,
-      "loss": 0.5064,
+      "epoch": 1.43,
+      "grad_norm": 2.8880758982566186,
+      "learning_rate": 4.487555238385862e-06,
+      "loss": 0.5643,
       "step": 820
     },
     {
-      "epoch": 1.43,
-      "grad_norm": 2.1504491301566784,
-      "learning_rate": 4.494132846430672e-06,
-      "loss": 0.5186,
+      "epoch": 1.44,
+      "grad_norm": 1.7726833097207872,
+      "learning_rate": 4.478264067674155e-06,
+      "loss": 0.5603,
       "step": 825
     },
     {
-      "epoch": 1.44,
-      "grad_norm": 14.695605966854037,
-      "learning_rate": 4.4849906821813406e-06,
-      "loss": 0.5349,
+      "epoch": 1.45,
+      "grad_norm": 1.9926460899793028,
+      "learning_rate": 4.4688992244325215e-06,
+      "loss": 0.5648,
       "step": 830
     },
     {
-      "epoch": 1.44,
-      "grad_norm": 2.063212121669853,
-      "learning_rate": 4.475776121660156e-06,
-      "loss": 0.531,
+      "epoch": 1.46,
+      "grad_norm": 1.959915507617199,
+      "learning_rate": 4.459461057417078e-06,
+      "loss": 0.5408,
       "step": 835
     },
     {
-      "epoch": 1.45,
-      "grad_norm": 1.834832161110646,
-      "learning_rate": 4.466489500939138e-06,
-      "loss": 0.5143,
+      "epoch": 1.47,
+      "grad_norm": 1.8617496123839707,
+      "learning_rate": 4.449949918114593e-06,
+      "loss": 0.5528,
       "step": 840
     },
     {
-      "epoch": 1.46,
-      "grad_norm": 1.7633276487477196,
-      "learning_rate": 4.457131158718478e-06,
-      "loss": 0.5331,
+      "epoch": 1.48,
+      "grad_norm": 1.9538793006795439,
+      "learning_rate": 4.440366160729393e-06,
+      "loss": 0.5674,
       "step": 845
     },
     {
-      "epoch": 1.47,
-      "grad_norm": 1.809450693070927,
-      "learning_rate": 4.447701436314176e-06,
-      "loss": 0.5283,
+      "epoch": 1.49,
+      "grad_norm": 1.8693300905967214,
+      "learning_rate": 4.430710142170176e-06,
+      "loss": 0.5578,
       "step": 850
     },
     {
-      "epoch": 1.48,
-      "grad_norm": 1.6601361692100707,
-      "learning_rate": 4.438200677645607e-06,
-      "loss": 0.5371,
+      "epoch": 1.49,
+      "grad_norm": 1.9612803924260176,
+      "learning_rate": 4.420982222036719e-06,
+      "loss": 0.5435,
       "step": 855
     },
     {
-      "epoch": 1.49,
-      "grad_norm": 1.6494075536487005,
-      "learning_rate": 4.428629229222962e-06,
-      "loss": 0.5334,
+      "epoch": 1.5,
+      "grad_norm": 1.995791143771985,
+      "learning_rate": 4.411182762606484e-06,
+      "loss": 0.5613,
       "step": 860
     },
     {
-      "epoch": 1.5,
-      "grad_norm": 1.5995037205122475,
-      "learning_rate": 4.418987440134624e-06,
-      "loss": 0.5171,
+      "epoch": 1.51,
+      "grad_norm": 5.458759085906611,
+      "learning_rate": 4.401312128821131e-06,
+      "loss": 0.5535,
       "step": 865
     },
     {
-      "epoch": 1.51,
-      "grad_norm": 1.6352796020571934,
-      "learning_rate": 4.4092756620344246e-06,
-      "loss": 0.5148,
+      "epoch": 1.52,
+      "grad_norm": 3.017681632112604,
+      "learning_rate": 4.391370688272919e-06,
+      "loss": 0.5761,
       "step": 870
     },
     {
-      "epoch": 1.51,
-      "grad_norm": 7.045247439882991,
-      "learning_rate": 4.3994942491288275e-06,
-      "loss": 1.7556,
+      "epoch": 1.53,
+      "grad_norm": 1.6117486364142628,
+      "learning_rate": 4.381358811191025e-06,
+      "loss": 0.5424,
       "step": 875
     },
     {
-      "epoch": 1.52,
-      "grad_norm": 6.128928116038949,
-      "learning_rate": 4.389643558164005e-06,
-      "loss": 0.5763,
+      "epoch": 1.54,
+      "grad_norm": 1.6889717368205845,
+      "learning_rate": 4.3712768704277535e-06,
+      "loss": 0.5578,
       "step": 880
     },
     {
-      "epoch": 1.53,
-      "grad_norm": 36.49282493029366,
-      "learning_rate": 4.379723948412828e-06,
-      "loss": 0.6,
+      "epoch": 1.55,
+      "grad_norm": 1.6263535777974591,
+      "learning_rate": 4.361125241444647e-06,
+      "loss": 0.5517,
       "step": 885
     },
     {
-      "epoch": 1.54,
-      "grad_norm": 9.579002986726392,
-      "learning_rate": 4.369735781661763e-06,
-      "loss": 0.6819,
+      "epoch": 1.56,
+      "grad_norm": 1.5391004007544808,
+      "learning_rate": 4.350904302298511e-06,
+      "loss": 0.5475,
       "step": 890
     },
     {
-      "epoch": 1.55,
-      "grad_norm": 12.266502746290532,
-      "learning_rate": 4.3596794221976755e-06,
-      "loss": 0.6053,
+      "epoch": 1.56,
+      "grad_norm": 3.919504264594871,
+      "learning_rate": 4.3406144336273284e-06,
+      "loss": 0.5488,
       "step": 895
     },
     {
-      "epoch": 1.56,
-      "grad_norm": 3.9630821498228874,
-      "learning_rate": 4.349555236794546e-06,
-      "loss": 0.6229,
+      "epoch": 1.57,
+      "grad_norm": 1.7006307109389633,
+      "learning_rate": 4.330256018636086e-06,
+      "loss": 0.5385,
       "step": 900
     },
     {
-      "epoch": 1.57,
-      "grad_norm": 3.4771323398284566,
-      "learning_rate": 4.339363594700093e-06,
-      "loss": 0.585,
+      "epoch": 1.58,
+      "grad_norm": 1.6821530119030959,
+      "learning_rate": 4.319829443082506e-06,
+      "loss": 0.5392,
       "step": 905
     },
     {
-      "epoch": 1.57,
-      "grad_norm": 2.201457058274829,
-      "learning_rate": 4.329104867622302e-06,
-      "loss": 0.5611,
+      "epoch": 1.59,
+      "grad_norm": 1.81327230754709,
+      "learning_rate": 4.309335095262675e-06,
+      "loss": 0.5463,
       "step": 910
     },
     {
-      "epoch": 1.58,
-      "grad_norm": 1.9654853072206555,
-      "learning_rate": 4.3187794297158765e-06,
-      "loss": 0.5648,
+      "epoch": 1.6,
+      "grad_norm": 1.7582986070443545,
+      "learning_rate": 4.298773365996591e-06,
+      "loss": 0.5698,
       "step": 915
     },
     {
-      "epoch": 1.59,
-      "grad_norm": 4.227758915564248,
-      "learning_rate": 4.308387657568582e-06,
-      "loss": 0.5481,
+      "epoch": 1.61,
+      "grad_norm": 1.6167984983083563,
+      "learning_rate": 4.288144648613601e-06,
+      "loss": 0.5451,
       "step": 920
     },
     {
-      "epoch": 1.6,
-      "grad_norm": 1.7148546445696602,
-      "learning_rate": 4.297929930187519e-06,
-      "loss": 0.548,
+      "epoch": 1.62,
+      "grad_norm": 3.2311390071409125,
+      "learning_rate": 4.277449338937754e-06,
+      "loss": 0.5516,
       "step": 925
     },
     {
-      "epoch": 1.61,
-      "grad_norm": 1.7282534502834466,
-      "learning_rate": 4.287406628985295e-06,
-      "loss": 0.5669,
+      "epoch": 1.63,
+      "grad_norm": 1.6218961380455157,
+      "learning_rate": 4.266687835273071e-06,
+      "loss": 0.538,
       "step": 930
     },
     {
-      "epoch": 1.62,
-      "grad_norm": 1.5604162135037978,
-      "learning_rate": 4.276818137766118e-06,
-      "loss": 0.5357,
+      "epoch": 1.63,
+      "grad_norm": 1.6454054434252245,
+      "learning_rate": 4.255860538388694e-06,
+      "loss": 0.5678,
       "step": 935
     },
     {
-      "epoch": 1.63,
-      "grad_norm": 1.7385091646846929,
-      "learning_rate": 4.266164842711797e-06,
-      "loss": 0.5371,
+      "epoch": 1.64,
+      "grad_norm": 1.9639702122144873,
+      "learning_rate": 4.244967851503975e-06,
+      "loss": 0.5446,
       "step": 940
     },
     {
-      "epoch": 1.63,
-      "grad_norm": 1.8757343048736053,
-      "learning_rate": 4.255447132367651e-06,
-      "loss": 0.5356,
+      "epoch": 1.65,
+      "grad_norm": 1.7028448251386163,
+      "learning_rate": 4.234010180273455e-06,
+      "loss": 0.5297,
       "step": 945
     },
     {
-      "epoch": 1.64,
-      "grad_norm": 1.5821779705858339,
-      "learning_rate": 4.2446653976283485e-06,
-      "loss": 0.534,
+      "epoch": 1.66,
+      "grad_norm": 1.689519412982808,
+      "learning_rate": 4.2229879327717545e-06,
+      "loss": 0.5432,
       "step": 950
     },
     {
-      "epoch": 1.65,
-      "grad_norm": 1.7181301166792315,
-      "learning_rate": 4.233820031723644e-06,
-      "loss": 0.5192,
+      "epoch": 1.67,
+      "grad_norm": 1.64362516189522,
+      "learning_rate": 4.211901519478382e-06,
+      "loss": 0.5414,
       "step": 955
     },
     {
-      "epoch": 1.66,
-      "grad_norm": 1.7620628413002903,
-      "learning_rate": 4.22291143020404e-06,
-      "loss": 0.527,
+      "epoch": 1.68,
+      "grad_norm": 1.9767208825933924,
+      "learning_rate": 4.200751353262442e-06,
+      "loss": 0.5578,
       "step": 960
     },
     {
-      "epoch": 1.67,
-      "grad_norm": 1.588909360094358,
-      "learning_rate": 4.211939990926355e-06,
-      "loss": 0.5205,
+      "epoch": 1.69,
+      "grad_norm": 1.7757696899561655,
+      "learning_rate": 4.1895378493672615e-06,
+      "loss": 0.5447,
       "step": 965
     },
     {
-      "epoch": 1.68,
-      "grad_norm": 1.6838208492860955,
-      "learning_rate": 4.200906114039219e-06,
-      "loss": 0.5298,
+      "epoch": 1.7,
+      "grad_norm": 1.7488808503021864,
+      "learning_rate": 4.178261425394926e-06,
+      "loss": 0.5362,
       "step": 970
     },
     {
-      "epoch": 1.69,
-      "grad_norm": 1.9081461502343906,
-      "learning_rate": 4.1898102019684765e-06,
-      "loss": 0.502,
+      "epoch": 1.7,
+      "grad_norm": 1.720469955142088,
+      "learning_rate": 4.16692250129073e-06,
+      "loss": 0.5339,
       "step": 975
     },
     {
-      "epoch": 1.7,
-      "grad_norm": 1.4961424837742658,
-      "learning_rate": 4.1786526594025065e-06,
-      "loss": 0.5372,
+      "epoch": 1.71,
+      "grad_norm": 1.656259449924422,
+      "learning_rate": 4.15552149932753e-06,
+      "loss": 0.5324,
       "step": 980
     },
     {
-      "epoch": 1.7,
-      "grad_norm": 1.6252683024976837,
-      "learning_rate": 4.167433893277473e-06,
-      "loss": 0.5269,
+      "epoch": 1.72,
+      "grad_norm": 1.605065327847084,
+      "learning_rate": 4.144058844090032e-06,
+      "loss": 0.5421,
       "step": 985
     },
     {
-      "epoch": 1.71,
-      "grad_norm": 1.5490581530552676,
-      "learning_rate": 4.156154312762467e-06,
-      "loss": 0.5285,
+      "epoch": 1.73,
+      "grad_norm": 2.400014331232814,
+      "learning_rate": 4.1325349624589625e-06,
+      "loss": 0.5345,
       "step": 990
     },
     {
-      "epoch": 1.72,
-      "grad_norm": 1.8061470924061633,
-      "learning_rate": 4.1448143292446e-06,
-      "loss": 0.5503,
+      "epoch": 1.74,
+      "grad_norm": 1.8217771753224867,
+      "learning_rate": 4.120950283595188e-06,
+      "loss": 0.5645,
       "step": 995
     },
     {
-      "epoch": 1.73,
-      "grad_norm": 1.5936238911397005,
-      "learning_rate": 4.133414356313988e-06,
-      "loss": 0.535,
+      "epoch": 1.75,
+      "grad_norm": 1.7606143199953959,
+      "learning_rate": 4.109305238923718e-06,
+      "loss": 0.5385,
       "step": 1000
     },
     {
-      "epoch": 1.74,
-      "grad_norm": 1.5524395421750514,
-      "learning_rate": 4.121954809748675e-06,
-      "loss": 0.5278,
+      "epoch": 1.76,
+      "grad_norm": 1.854649950712452,
+      "learning_rate": 4.09760026211765e-06,
+      "loss": 0.537,
       "step": 1005
     },
     {
-      "epoch": 1.75,
-      "grad_norm": 1.5740976269214233,
-      "learning_rate": 4.110436107499462e-06,
-      "loss": 0.5197,
+      "epoch": 1.77,
+      "grad_norm": 1.834481039824983,
+      "learning_rate": 4.0858357890820115e-06,
+      "loss": 0.5256,
       "step": 1010
     },
     {
-      "epoch": 1.76,
-      "grad_norm": 1.5214013327972284,
-      "learning_rate": 4.098858669674667e-06,
-      "loss": 0.5202,
+      "epoch": 1.77,
+      "grad_norm": 1.7881039817362976,
+      "learning_rate": 4.0740122579375284e-06,
+      "loss": 0.5492,
       "step": 1015
     },
     {
-      "epoch": 1.76,
-      "grad_norm": 1.5237741216662357,
-      "learning_rate": 4.087222918524807e-06,
-      "loss": 0.5088,
+      "epoch": 1.78,
+      "grad_norm": 1.99136744573053,
+      "learning_rate": 4.062130109004313e-06,
+      "loss": 0.5423,
       "step": 1020
     },
     {
-      "epoch": 1.77,
-      "grad_norm": 1.6738078012815194,
-      "learning_rate": 4.07552927842719e-06,
-      "loss": 0.5166,
+      "epoch": 1.79,
+      "grad_norm": 2.067114672482736,
+      "learning_rate": 4.0501897847854596e-06,
+      "loss": 0.5248,
       "step": 1025
     },
     {
-      "epoch": 1.78,
-      "grad_norm": 1.5517613536326162,
-      "learning_rate": 4.06377817587044e-06,
-      "loss": 0.4918,
+      "epoch": 1.8,
+      "grad_norm": 2.3619737058273698,
+      "learning_rate": 4.038191729950569e-06,
+      "loss": 0.5548,
       "step": 1030
     },
     {
-      "epoch": 1.79,
-      "grad_norm": 1.431747331311476,
-      "learning_rate": 4.051970039438945e-06,
-      "loss": 0.5133,
+      "epoch": 1.81,
+      "grad_norm": 1.6422923327155017,
+      "learning_rate": 4.026136391319187e-06,
+      "loss": 0.5661,
       "step": 1035
     },
     {
-      "epoch": 1.8,
-      "grad_norm": 1.4767269458087147,
-      "learning_rate": 4.0401052997972235e-06,
-      "loss": 0.5033,
+      "epoch": 1.82,
+      "grad_norm": 1.7084270040887584,
+      "learning_rate": 4.014024217844167e-06,
+      "loss": 0.5413,
       "step": 1040
     },
     {
-      "epoch": 1.81,
-      "grad_norm": 1.5180384694427076,
-      "learning_rate": 4.028184389674216e-06,
-      "loss": 0.5058,
+      "epoch": 1.83,
+      "grad_norm": 2.4216667822493387,
+      "learning_rate": 4.001855660594948e-06,
+      "loss": 0.5251,
       "step": 1045
     },
     {
-      "epoch": 1.82,
-      "grad_norm": 1.5889385254960862,
-      "learning_rate": 4.016207743847505e-06,
-      "loss": 0.515,
+      "epoch": 1.84,
+      "grad_norm": 1.6894496135090933,
+      "learning_rate": 3.989631172740756e-06,
+      "loss": 0.5282,
       "step": 1050
     },
     {
-      "epoch": 1.83,
-      "grad_norm": 1.5532809579850755,
-      "learning_rate": 4.004175799127458e-06,
-      "loss": 0.5252,
+      "epoch": 1.84,
+      "grad_norm": 2.3374959162830224,
+      "learning_rate": 3.97735120953373e-06,
+      "loss": 0.5375,
       "step": 1055
     },
     {
-      "epoch": 1.83,
-      "grad_norm": 1.5669745733483815,
-      "learning_rate": 3.992088994341292e-06,
-      "loss": 0.5237,
+      "epoch": 1.85,
+      "grad_norm": 1.6036838276324374,
+      "learning_rate": 3.965016228291966e-06,
+      "loss": 0.5482,
       "step": 1060
     },
     {
-      "epoch": 1.84,
-      "grad_norm": 1.7971179594367128,
-      "learning_rate": 3.979947770317077e-06,
-      "loss": 0.5401,
+      "epoch": 1.86,
+      "grad_norm": 1.5838188110377205,
+      "learning_rate": 3.9526266883824865e-06,
+      "loss": 0.5406,
       "step": 1065
     },
     {
-      "epoch": 1.85,
-      "grad_norm": 1.6564666981215765,
-      "learning_rate": 3.967752569867645e-06,
-      "loss": 0.5315,
+      "epoch": 1.87,
+      "grad_norm": 2.925850402777526,
+      "learning_rate": 3.940183051204133e-06,
+      "loss": 0.5514,
       "step": 1070
     },
     {
-      "epoch": 1.86,
-      "grad_norm": 1.596740218503028,
-      "learning_rate": 3.955503837774458e-06,
-      "loss": 0.5124,
+      "epoch": 1.88,
+      "grad_norm": 1.8345221327535834,
+      "learning_rate": 3.927685780170385e-06,
+      "loss": 0.5351,
       "step": 1075
     },
     {
-      "epoch": 1.87,
-      "grad_norm": 1.5508137957824288,
-      "learning_rate": 3.943202020771367e-06,
-      "loss": 0.5219,
+      "epoch": 1.89,
+      "grad_norm": 1.7672217360317701,
+      "learning_rate": 3.915135340692098e-06,
+      "loss": 0.5431,
       "step": 1080
     },
     {
-      "epoch": 1.88,
-      "grad_norm": 1.7093498845192874,
-      "learning_rate": 3.9308475675283326e-06,
-      "loss": 0.5265,
+      "epoch": 1.9,
+      "grad_norm": 1.7690557869414958,
+      "learning_rate": 3.902532200160174e-06,
+      "loss": 0.5332,
       "step": 1085
     },
     {
-      "epoch": 1.89,
-      "grad_norm": 1.5364211220041548,
-      "learning_rate": 3.9184409286350535e-06,
-      "loss": 0.5249,
+      "epoch": 1.91,
+      "grad_norm": 1.8016714829119445,
+      "learning_rate": 3.889876827928156e-06,
+      "loss": 0.5547,
       "step": 1090
     },
     {
-      "epoch": 1.89,
-      "grad_norm": 1.588131611670702,
-      "learning_rate": 3.905982556584537e-06,
-      "loss": 0.5132,
+      "epoch": 1.91,
+      "grad_norm": 1.8101317872578178,
+      "learning_rate": 3.877169695294749e-06,
+      "loss": 0.5152,
       "step": 1095
     },
     {
-      "epoch": 1.9,
-      "grad_norm": 1.4898401463528494,
-      "learning_rate": 3.893472905756593e-06,
-      "loss": 0.5241,
+      "epoch": 1.92,
+      "grad_norm": 1.863558573257381,
+      "learning_rate": 3.8644112754862614e-06,
+      "loss": 0.5496,
       "step": 1100
     },
     {
-      "epoch": 1.91,
-      "grad_norm": 1.655966866153354,
-      "learning_rate": 3.880912432401265e-06,
-      "loss": 0.5098,
+      "epoch": 1.93,
+      "grad_norm": 1.6834072929471855,
+      "learning_rate": 3.8516020436389945e-06,
+      "loss": 0.5403,
       "step": 1105
     },
     {
-      "epoch": 1.92,
-      "grad_norm": 1.4977265046553758,
-      "learning_rate": 3.868301594622185e-06,
-      "loss": 0.5189,
+      "epoch": 1.94,
+      "grad_norm": 1.8343601221708872,
+      "learning_rate": 3.838742476781535e-06,
+      "loss": 0.5449,
       "step": 1110
     },
     {
-      "epoch": 1.93,
-      "grad_norm": 1.45173341143097,
-      "learning_rate": 3.855640852359869e-06,
-      "loss": 0.4974,
+      "epoch": 1.95,
+      "grad_norm": 1.8111180639710172,
+      "learning_rate": 3.825833053816998e-06,
+      "loss": 0.5177,
       "step": 1115
     },
     {
-      "epoch": 1.94,
-      "grad_norm": 1.8226392308673234,
-      "learning_rate": 3.842930667374945e-06,
-      "loss": 0.5122,
+      "epoch": 1.96,
+      "grad_norm": 1.760616458847631,
+      "learning_rate": 3.812874255505191e-06,
+      "loss": 0.5282,
       "step": 1120
     },
     {
-      "epoch": 1.95,
-      "grad_norm": 1.479895371145679,
-      "learning_rate": 3.830171503231302e-06,
-      "loss": 0.5077,
+      "epoch": 1.97,
+      "grad_norm": 1.662863038395529,
+      "learning_rate": 3.7998665644447064e-06,
+      "loss": 0.5462,
       "step": 1125
     },
     {
-      "epoch": 1.96,
-      "grad_norm": 1.8170301741742978,
-      "learning_rate": 3.817363825279195e-06,
-      "loss": 0.5174,
+      "epoch": 1.98,
+      "grad_norm": 1.5940854942690412,
+      "learning_rate": 3.786810465054953e-06,
+      "loss": 0.5155,
       "step": 1130
     },
     {
-      "epoch": 1.96,
-      "grad_norm": 1.6039721848461628,
-      "learning_rate": 3.8045081006382668e-06,
-      "loss": 0.5145,
+      "epoch": 1.98,
+      "grad_norm": 1.7490423857630772,
+      "learning_rate": 3.773706443558112e-06,
+      "loss": 0.5333,
       "step": 1135
     },
     {
-      "epoch": 1.97,
-      "grad_norm": 2.198046435353487,
-      "learning_rate": 3.7916047981805058e-06,
-      "loss": 0.507,
+      "epoch": 1.99,
+      "grad_norm": 1.7805842103529974,
+      "learning_rate": 3.7605549879610346e-06,
+      "loss": 0.523,
       "step": 1140
     },
     {
-      "epoch": 1.98,
-      "grad_norm": 2.538669249099628,
-      "learning_rate": 3.778654388513159e-06,
-      "loss": 0.4926,
+      "epoch": 2.0,
+      "eval_loss": 0.48952990770339966,
+      "eval_runtime": 329.9798,
+      "eval_samples_per_second": 22.792,
+      "eval_steps_per_second": 0.358,
+      "step": 1144
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 3.530130888128286,
+      "learning_rate": 3.747356588037064e-06,
+      "loss": 0.4829,
       "step": 1145
     },
     {
-      "epoch": 1.99,
-      "grad_norm": 1.490519700011853,
-      "learning_rate": 3.765657343961553e-06,
-      "loss": 0.5091,
+      "epoch": 2.01,
+      "grad_norm": 2.3455796414814656,
+      "learning_rate": 3.7341117353077964e-06,
+      "loss": 0.4185,
       "step": 1150
     },
     {
-      "epoch": 2.0,
-      "grad_norm": 1.609088164054738,
-      "learning_rate": 3.7526141385518794e-06,
-      "loss": 0.51,
+      "epoch": 2.02,
+      "grad_norm": 11.689021838621713,
+      "learning_rate": 3.7208209230247785e-06,
+      "loss": 0.4431,
       "step": 1155
     },
     {
-      "epoch": 2.0,
-      "eval_loss": 0.5023414492607117,
-      "eval_runtime": 329.8071,
-      "eval_samples_per_second": 22.804,
-      "eval_steps_per_second": 0.358,
-      "step": 1156
-    },
-    {
-      "epoch": 2.01,
-      "grad_norm": 2.3682882817547433,
-      "learning_rate": 3.739525247993901e-06,
-      "loss": 0.4345,
+      "epoch": 2.03,
+      "grad_norm": 2.136634988191819,
+      "learning_rate": 3.7074846461511336e-06,
+      "loss": 0.4166,
       "step": 1160
     },
     {
-      "epoch": 2.02,
-      "grad_norm": 2.198186338081821,
-      "learning_rate": 3.7263911496636003e-06,
-      "loss": 0.4221,
+      "epoch": 2.04,
+      "grad_norm": 1.8007601624193452,
+      "learning_rate": 3.694103401343136e-06,
+      "loss": 0.4247,
       "step": 1165
     },
     {
-      "epoch": 2.02,
-      "grad_norm": 1.9622396068630974,
-      "learning_rate": 3.71321232258577e-06,
-      "loss": 0.4166,
+      "epoch": 2.05,
+      "grad_norm": 1.8280089181701968,
+      "learning_rate": 3.6806776869317074e-06,
+      "loss": 0.4184,
       "step": 1170
     },
     {
-      "epoch": 2.03,
-      "grad_norm": 1.9405841953564362,
-      "learning_rate": 3.6999892474165477e-06,
-      "loss": 0.4124,
+      "epoch": 2.05,
+      "grad_norm": 1.8498758253178664,
+      "learning_rate": 3.667208002903863e-06,
+      "loss": 0.4015,
       "step": 1175
     },
     {
-      "epoch": 2.04,
-      "grad_norm": 1.6911138242049408,
-      "learning_rate": 3.6867224064258754e-06,
-      "loss": 0.3999,
+      "epoch": 2.06,
+      "grad_norm": 1.7616805224163088,
+      "learning_rate": 3.6536948508840915e-06,
+      "loss": 0.4139,
       "step": 1180
     },
     {
-      "epoch": 2.05,
-      "grad_norm": 1.805794848826172,
-      "learning_rate": 3.673412283479916e-06,
-      "loss": 0.3923,
+      "epoch": 2.07,
+      "grad_norm": 1.7368842598902192,
+      "learning_rate": 3.6401387341156715e-06,
+      "loss": 0.3943,
       "step": 1185
     },
     {
-      "epoch": 2.06,
-      "grad_norm": 1.7227193868132855,
-      "learning_rate": 3.660059364023409e-06,
-      "loss": 0.3922,
+      "epoch": 2.08,
+      "grad_norm": 1.735642439390496,
+      "learning_rate": 3.6265401574419316e-06,
+      "loss": 0.4362,
       "step": 1190
     },
     {
-      "epoch": 2.07,
-      "grad_norm": 1.5633624388497611,
-      "learning_rate": 3.6466641350619564e-06,
-      "loss": 0.4081,
+      "epoch": 2.09,
+      "grad_norm": 1.9606530829602056,
+      "learning_rate": 3.6128996272874523e-06,
+      "loss": 0.4177,
       "step": 1195
     },
     {
-      "epoch": 2.08,
-      "grad_norm": 1.6049499642389522,
-      "learning_rate": 3.6332270851442697e-06,
-      "loss": 0.3988,
+      "epoch": 2.1,
+      "grad_norm": 2.045121741468278,
+      "learning_rate": 3.5992176516392007e-06,
+      "loss": 0.4191,
       "step": 1200
     },
     {
-      "epoch": 2.08,
-      "grad_norm": 1.6302545860912692,
-      "learning_rate": 3.619748704344346e-06,
-      "loss": 0.4153,
+      "epoch": 2.11,
+      "grad_norm": 1.8838455742670526,
+      "learning_rate": 3.5854947400276164e-06,
+      "loss": 0.414,
       "step": 1205
     },
     {
-      "epoch": 2.09,
-      "grad_norm": 1.893580997382411,
-      "learning_rate": 3.6062294842435958e-06,
-      "loss": 0.4118,
+      "epoch": 2.12,
+      "grad_norm": 2.0713934176853366,
+      "learning_rate": 3.5717314035076355e-06,
+      "loss": 0.423,
       "step": 1210
     },
     {
-      "epoch": 2.1,
-      "grad_norm": 1.591008467389523,
-      "learning_rate": 3.5926699179129156e-06,
-      "loss": 0.3967,
+      "epoch": 2.12,
+      "grad_norm": 1.9586979622942737,
+      "learning_rate": 3.5579281546396582e-06,
+      "loss": 0.4089,
       "step": 1215
     },
     {
-      "epoch": 2.11,
-      "grad_norm": 1.8812505523353902,
-      "learning_rate": 3.579070499894703e-06,
-      "loss": 0.4105,
+      "epoch": 2.13,
+      "grad_norm": 1.6793543323852747,
+      "learning_rate": 3.54408550747046e-06,
+      "loss": 0.4045,
       "step": 1220
     },
     {
-      "epoch": 2.12,
-      "grad_norm": 1.7312982189755863,
-      "learning_rate": 3.5654317261848182e-06,
-      "loss": 0.4088,
+      "epoch": 2.14,
+      "grad_norm": 1.8609322408923432,
+      "learning_rate": 3.530203977514049e-06,
+      "loss": 0.4216,
       "step": 1225
     },
     {
-      "epoch": 2.13,
-      "grad_norm": 1.715794291601741,
-      "learning_rate": 3.5517540942144985e-06,
-      "loss": 0.3837,
+      "epoch": 2.15,
+      "grad_norm": 1.9301155764363938,
+      "learning_rate": 3.516284081732466e-06,
+      "loss": 0.4141,
       "step": 1230
     },
     {
-      "epoch": 2.14,
-      "grad_norm": 1.571536156806173,
-      "learning_rate": 3.538038102832213e-06,
-      "loss": 0.4047,
+      "epoch": 2.16,
+      "grad_norm": 1.8203713381481375,
+      "learning_rate": 3.5023263385165346e-06,
+      "loss": 0.4193,
       "step": 1235
     },
     {
-      "epoch": 2.15,
-      "grad_norm": 1.6926403850601441,
-      "learning_rate": 3.5242842522854687e-06,
-      "loss": 0.4097,
+      "epoch": 2.17,
+      "grad_norm": 1.986035063455061,
+      "learning_rate": 3.4883312676665537e-06,
+      "loss": 0.4102,
       "step": 1240
     },
     {
-      "epoch": 2.15,
-      "grad_norm": 1.5865520678177705,
-      "learning_rate": 3.5104930442025665e-06,
-      "loss": 0.3988,
+      "epoch": 2.18,
+      "grad_norm": 1.932701768003649,
+      "learning_rate": 3.4742993903729423e-06,
+      "loss": 0.4096,
       "step": 1245
     },
     {
-      "epoch": 2.16,
-      "grad_norm": 1.6304875271885666,
-      "learning_rate": 3.4966649815743055e-06,
-      "loss": 0.4002,
+      "epoch": 2.19,
+      "grad_norm": 2.072592848243265,
+      "learning_rate": 3.460231229196826e-06,
+      "loss": 0.4356,
       "step": 1250
     },
     {
-      "epoch": 2.17,
-      "grad_norm": 1.5709135265237464,
-      "learning_rate": 3.4828005687356382e-06,
-      "loss": 0.4115,
+      "epoch": 2.19,
+      "grad_norm": 1.8608782853161558,
+      "learning_rate": 3.446127308050579e-06,
+      "loss": 0.422,
       "step": 1255
     },
     {
-      "epoch": 2.18,
-      "grad_norm": 1.6424713964605273,
-      "learning_rate": 3.4689003113472754e-06,
-      "loss": 0.3978,
+      "epoch": 2.2,
+      "grad_norm": 1.8587160248700436,
+      "learning_rate": 3.431988152178315e-06,
+      "loss": 0.4055,
       "step": 1260
     },
     {
-      "epoch": 2.19,
-      "grad_norm": 1.8859757363715237,
-      "learning_rate": 3.454964716377247e-06,
-      "loss": 0.4203,
+      "epoch": 2.21,
+      "grad_norm": 1.7275839544864644,
+      "learning_rate": 3.4178142881363192e-06,
+      "loss": 0.3994,
       "step": 1265
     },
     {
-      "epoch": 2.2,
-      "grad_norm": 1.637179971274298,
-      "learning_rate": 3.440994292082408e-06,
-      "loss": 0.3908,
+      "epoch": 2.22,
+      "grad_norm": 1.7408220116241242,
+      "learning_rate": 3.4036062437734484e-06,
+      "loss": 0.4247,
       "step": 1270
     },
     {
-      "epoch": 2.21,
-      "grad_norm": 25.023011080924313,
-      "learning_rate": 3.4269895479899023e-06,
-      "loss": 0.4038,
+      "epoch": 2.23,
+      "grad_norm": 1.8684812995394824,
+      "learning_rate": 3.3893645482114663e-06,
+      "loss": 0.4013,
       "step": 1275
     },
     {
-      "epoch": 2.21,
-      "grad_norm": 2.830979911281441,
-      "learning_rate": 3.412950994878582e-06,
-      "loss": 0.4047,
+      "epoch": 2.24,
+      "grad_norm": 1.9169356401008089,
+      "learning_rate": 3.3750897318253407e-06,
+      "loss": 0.4436,
       "step": 1280
     },
     {
-      "epoch": 2.22,
-      "grad_norm": 1.8263955076269602,
-      "learning_rate": 3.3988791447603763e-06,
-      "loss": 0.3963,
+      "epoch": 2.25,
+      "grad_norm": 1.9965429262903887,
+      "learning_rate": 3.3607823262234936e-06,
+      "loss": 0.4052,
       "step": 1285
     },
     {
-      "epoch": 2.23,
-      "grad_norm": 1.6584308792186484,
-      "learning_rate": 3.3847745108616157e-06,
-      "loss": 0.4049,
+      "epoch": 2.26,
+      "grad_norm": 1.676108249215827,
+      "learning_rate": 3.3464428642280004e-06,
+      "loss": 0.4043,
       "step": 1290
     },
     {
-      "epoch": 2.24,
-      "grad_norm": 1.868836216780196,
-      "learning_rate": 3.3706376076043183e-06,
-      "loss": 0.4123,
+      "epoch": 2.26,
+      "grad_norm": 1.759052520530038,
+      "learning_rate": 3.3320718798547503e-06,
+      "loss": 0.3891,
       "step": 1295
     },
     {
-      "epoch": 2.25,
-      "grad_norm": 1.704922866865318,
-      "learning_rate": 3.356468950587424e-06,
-      "loss": 0.3919,
+      "epoch": 2.27,
+      "grad_norm": 1.8158002528007304,
+      "learning_rate": 3.3176699082935546e-06,
+      "loss": 0.4048,
       "step": 1300
     },
     {
-      "epoch": 2.26,
-      "grad_norm": 1.8511972094459495,
-      "learning_rate": 3.3422690565679895e-06,
-      "loss": 0.4118,
+      "epoch": 2.28,
+      "grad_norm": 1.8540661043983968,
+      "learning_rate": 3.303237485888221e-06,
+      "loss": 0.413,
       "step": 1305
     },
     {
-      "epoch": 2.27,
-      "grad_norm": 1.5826425092029799,
-      "learning_rate": 3.3280384434423447e-06,
-      "loss": 0.4072,
+      "epoch": 2.29,
+      "grad_norm": 1.8368083328897664,
+      "learning_rate": 3.2887751501165755e-06,
+      "loss": 0.4134,
       "step": 1310
     },
     {
-      "epoch": 2.28,
-      "grad_norm": 2.8458291903299555,
-      "learning_rate": 3.3137776302272017e-06,
-      "loss": 0.4044,
+      "epoch": 2.3,
+      "grad_norm": 1.943248250811944,
+      "learning_rate": 3.2742834395704486e-06,
+      "loss": 0.4225,
       "step": 1315
     },
     {
-      "epoch": 2.28,
-      "grad_norm": 2.696905442165675,
-      "learning_rate": 3.2994871370407242e-06,
-      "loss": 0.405,
+      "epoch": 2.31,
+      "grad_norm": 2.023022399396665,
+      "learning_rate": 3.2597628939356174e-06,
+      "loss": 0.4045,
       "step": 1320
     },
     {
-      "epoch": 2.29,
-      "grad_norm": 1.8719365939252683,
-      "learning_rate": 3.2851674850835615e-06,
-      "loss": 0.4091,
+      "epoch": 2.32,
+      "grad_norm": 2.0127015131643575,
+      "learning_rate": 3.2452140539717047e-06,
+      "loss": 0.4157,
       "step": 1325
     },
     {
-      "epoch": 2.3,
-      "grad_norm": 3.602956750004364,
-      "learning_rate": 3.2708191966198344e-06,
-      "loss": 0.3845,
+      "epoch": 2.33,
+      "grad_norm": 1.9502787177069507,
+      "learning_rate": 3.2306374614920434e-06,
+      "loss": 0.4037,
       "step": 1330
     },
     {
-      "epoch": 2.31,
-      "grad_norm": 1.7107089177149095,
-      "learning_rate": 3.2564427949580907e-06,
-      "loss": 0.4128,
+      "epoch": 2.33,
+      "grad_norm": 1.938220761297596,
+      "learning_rate": 3.2160336593434977e-06,
+      "loss": 0.4065,
       "step": 1335
     },
     {
-      "epoch": 2.32,
-      "grad_norm": 1.782125003384775,
-      "learning_rate": 3.2420388044322204e-06,
-      "loss": 0.3957,
+      "epoch": 2.34,
+      "grad_norm": 1.8287659925527244,
+      "learning_rate": 3.201403191386247e-06,
+      "loss": 0.4249,
       "step": 1340
     },
     {
-      "epoch": 2.33,
-      "grad_norm": 1.7157313501617861,
-      "learning_rate": 3.2276077503823276e-06,
-      "loss": 0.4164,
+      "epoch": 2.35,
+      "grad_norm": 2.115173595697826,
+      "learning_rate": 3.1867466024735327e-06,
+      "loss": 0.4083,
       "step": 1345
     },
     {
-      "epoch": 2.34,
-      "grad_norm": 1.6189210159528598,
-      "learning_rate": 3.213150159135575e-06,
-      "loss": 0.4017,
+      "epoch": 2.36,
+      "grad_norm": 1.8653664947018882,
+      "learning_rate": 3.1720644384313647e-06,
+      "loss": 0.4311,
       "step": 1350
     },
     {
-      "epoch": 2.34,
-      "grad_norm": 1.7965605988243851,
-      "learning_rate": 3.198666557986984e-06,
-      "loss": 0.3995,
+      "epoch": 2.37,
+      "grad_norm": 2.0817854592111025,
+      "learning_rate": 3.1573572460381992e-06,
+      "loss": 0.4184,
       "step": 1355
     },
     {
-      "epoch": 2.35,
-      "grad_norm": 1.6750382286037944,
-      "learning_rate": 3.184157475180208e-06,
-      "loss": 0.4065,
+      "epoch": 2.38,
+      "grad_norm": 1.9348176314156256,
+      "learning_rate": 3.1426255730045703e-06,
+      "loss": 0.4231,
       "step": 1360
     },
     {
-      "epoch": 2.36,
-      "grad_norm": 1.7466143412529116,
-      "learning_rate": 3.169623439888259e-06,
-      "loss": 0.3913,
+      "epoch": 2.39,
+      "grad_norm": 1.9693517430562117,
+      "learning_rate": 3.127869967952698e-06,
+      "loss": 0.4206,
       "step": 1365
     },
     {
-      "epoch": 2.37,
-      "grad_norm": 1.7094098689820845,
-      "learning_rate": 3.1550649821942165e-06,
-      "loss": 0.4055,
+      "epoch": 2.4,
+      "grad_norm": 1.8456135208525017,
+      "learning_rate": 3.1130909803960533e-06,
+      "loss": 0.3978,
       "step": 1370
     },
     {
-      "epoch": 2.38,
-      "grad_norm": 1.6546779771359428,
-      "learning_rate": 3.1404826330718884e-06,
-      "loss": 0.4072,
+      "epoch": 2.4,
+      "grad_norm": 1.9626926891274619,
+      "learning_rate": 3.0982891607188948e-06,
+      "loss": 0.4064,
       "step": 1375
     },
     {
-      "epoch": 2.39,
-      "grad_norm": 1.5822130245289208,
-      "learning_rate": 3.1258769243664457e-06,
-      "loss": 0.396,
+      "epoch": 2.41,
+      "grad_norm": 2.026442414272127,
+      "learning_rate": 3.0834650601557724e-06,
+      "loss": 0.4149,
       "step": 1380
     },
     {
-      "epoch": 2.4,
-      "grad_norm": 1.6175930524108444,
-      "learning_rate": 3.1112483887750285e-06,
-      "loss": 0.4009,
+      "epoch": 2.42,
+      "grad_norm": 2.063023886136181,
+      "learning_rate": 3.068619230770999e-06,
+      "loss": 0.4118,
       "step": 1385
     },
     {
-      "epoch": 2.4,
-      "grad_norm": 1.6572189285469252,
-      "learning_rate": 3.0965975598273147e-06,
-      "loss": 0.4135,
+      "epoch": 2.43,
+      "grad_norm": 1.9784482842258986,
+      "learning_rate": 3.0537522254380902e-06,
+      "loss": 0.4026,
       "step": 1390
     },
     {
-      "epoch": 2.41,
-      "grad_norm": 1.7351755497284653,
-      "learning_rate": 3.0819249718660625e-06,
-      "loss": 0.4077,
+      "epoch": 2.44,
+      "grad_norm": 1.921786550245392,
+      "learning_rate": 3.0388645978191745e-06,
+      "loss": 0.4044,
       "step": 1395
     },
     {
-      "epoch": 2.42,
-      "grad_norm": 1.722331718347925,
-      "learning_rate": 3.0672311600276205e-06,
-      "loss": 0.3961,
+      "epoch": 2.45,
+      "grad_norm": 1.9377855933194186,
+      "learning_rate": 3.0239569023443756e-06,
+      "loss": 0.4294,
       "step": 1400
     },
     {
-      "epoch": 2.43,
-      "grad_norm": 2.8787903703341704,
-      "learning_rate": 3.052516660222412e-06,
-      "loss": 0.4013,
+      "epoch": 2.46,
+      "grad_norm": 1.8926358118848756,
+      "learning_rate": 3.0090296941911633e-06,
+      "loss": 0.4131,
       "step": 1405
     },
     {
-      "epoch": 2.44,
-      "grad_norm": 1.7609799996300464,
-      "learning_rate": 3.0377820091153894e-06,
-      "loss": 0.4182,
+      "epoch": 2.47,
+      "grad_norm": 1.8195567532258379,
+      "learning_rate": 2.9940835292636806e-06,
+      "loss": 0.416,
       "step": 1410
     },
     {
-      "epoch": 2.45,
-      "grad_norm": 1.6624761502383696,
-      "learning_rate": 3.0230277441064576e-06,
-      "loss": 0.4174,
+      "epoch": 2.47,
+      "grad_norm": 1.9219319571815525,
+      "learning_rate": 2.9791189641720385e-06,
+      "loss": 0.4023,
       "step": 1415
     },
     {
-      "epoch": 2.46,
-      "grad_norm": 1.6422107924968583,
-      "learning_rate": 3.0082544033108792e-06,
-      "loss": 0.3968,
+      "epoch": 2.48,
+      "grad_norm": 1.9615684580403125,
+      "learning_rate": 2.9641365562115886e-06,
+      "loss": 0.4256,
       "step": 1420
     },
     {
-      "epoch": 2.47,
-      "grad_norm": 1.6979969793176557,
-      "learning_rate": 2.993462525539646e-06,
-      "loss": 0.3893,
+      "epoch": 2.49,
+      "grad_norm": 2.2610537590860145,
+      "learning_rate": 2.949136863342169e-06,
+      "loss": 0.4101,
       "step": 1425
     },
     {
-      "epoch": 2.47,
-      "grad_norm": 2.0089020705022027,
-      "learning_rate": 2.9786526502798245e-06,
-      "loss": 0.401,
+      "epoch": 2.5,
+      "grad_norm": 1.9357969284665328,
+      "learning_rate": 2.9341204441673267e-06,
+      "loss": 0.4202,
       "step": 1430
     },
     {
-      "epoch": 2.48,
-      "grad_norm": 2.4328077084567035,
-      "learning_rate": 2.9638253176748868e-06,
-      "loss": 0.4041,
+      "epoch": 2.51,
+      "grad_norm": 1.8520492192305298,
+      "learning_rate": 2.9190878579135077e-06,
+      "loss": 0.4155,
       "step": 1435
     },
     {
-      "epoch": 2.49,
-      "grad_norm": 2.1658392118895096,
-      "learning_rate": 2.948981068505004e-06,
-      "loss": 0.4152,
+      "epoch": 2.52,
+      "grad_norm": 1.9246616441113338,
+      "learning_rate": 2.904039664409244e-06,
+      "loss": 0.4056,
       "step": 1440
     },
     {
-      "epoch": 2.5,
-      "grad_norm": 1.9075735219009673,
-      "learning_rate": 2.9341204441673267e-06,
-      "loss": 0.4019,
+      "epoch": 2.53,
+      "grad_norm": 2.030321498137203,
+      "learning_rate": 2.888976424064289e-06,
+      "loss": 0.4176,
       "step": 1445
     },
     {
-      "epoch": 2.51,
-      "grad_norm": 79.58152669185955,
-      "learning_rate": 2.9192439866562363e-06,
-      "loss": 0.3986,
+      "epoch": 2.53,
+      "grad_norm": 2.238812621763622,
+      "learning_rate": 2.8738986978487625e-06,
+      "loss": 0.4049,
       "step": 1450
     },
     {
-      "epoch": 2.52,
-      "grad_norm": 1.9783510669105098,
-      "learning_rate": 2.904352238543582e-06,
-      "loss": 0.3977,
+      "epoch": 2.54,
+      "grad_norm": 2.259109743694307,
+      "learning_rate": 2.8588070472722486e-06,
+      "loss": 0.3983,
       "step": 1455
     },
     {
-      "epoch": 2.53,
-      "grad_norm": 2.0408912753861452,
-      "learning_rate": 2.889445742958886e-06,
-      "loss": 0.395,
+      "epoch": 2.55,
+      "grad_norm": 2.437226455840826,
+      "learning_rate": 2.8437020343628896e-06,
+      "loss": 0.3935,
       "step": 1460
     },
     {
-      "epoch": 2.53,
-      "grad_norm": 2.1287347922745563,
-      "learning_rate": 2.8745250435695433e-06,
-      "loss": 0.3944,
+      "epoch": 2.56,
+      "grad_norm": 2.346717301291877,
+      "learning_rate": 2.8285842216464544e-06,
+      "loss": 0.4093,
       "step": 1465
     },
     {
-      "epoch": 2.54,
-      "grad_norm": 1.9220277314654948,
-      "learning_rate": 2.859590684560982e-06,
-      "loss": 0.4066,
+      "epoch": 2.57,
+      "grad_norm": 2.4385081373620197,
+      "learning_rate": 2.813454172125389e-06,
+      "loss": 0.4007,
       "step": 1470
     },
     {
-      "epoch": 2.55,
-      "grad_norm": 1.7440202984118327,
-      "learning_rate": 2.844643210616824e-06,
-      "loss": 0.3933,
+      "epoch": 2.58,
+      "grad_norm": 2.0580498358246557,
+      "learning_rate": 2.79831244925785e-06,
+      "loss": 0.4012,
       "step": 1475
     },
     {
-      "epoch": 2.56,
-      "grad_norm": 4.515419055182269,
-      "learning_rate": 2.8296831668990195e-06,
-      "loss": 0.408,
+      "epoch": 2.59,
+      "grad_norm": 1.8904939031404155,
+      "learning_rate": 2.783159616936723e-06,
+      "loss": 0.4247,
       "step": 1480
     },
     {
-      "epoch": 2.57,
-      "grad_norm": 1.804786297673,
-      "learning_rate": 2.8147110990279576e-06,
-      "loss": 0.3895,
+      "epoch": 2.6,
+      "grad_norm": 1.939667430958579,
+      "learning_rate": 2.76799623946862e-06,
+      "loss": 0.3983,
       "step": 1485
     },
     {
-      "epoch": 2.58,
-      "grad_norm": 1.8010185425321945,
-      "learning_rate": 2.7997275530625715e-06,
-      "loss": 0.4007,
+      "epoch": 2.6,
+      "grad_norm": 1.9363488042991572,
+      "learning_rate": 2.7528228815528622e-06,
+      "loss": 0.4054,
       "step": 1490
     },
     {
-      "epoch": 2.59,
-      "grad_norm": 1.7762627094164511,
-      "learning_rate": 2.7847330754804257e-06,
-      "loss": 0.4047,
+      "epoch": 2.61,
+      "grad_norm": 1.9510664856850064,
+      "learning_rate": 2.7376401082604563e-06,
+      "loss": 0.4098,
       "step": 1495
     },
     {
-      "epoch": 2.6,
-      "grad_norm": 1.713415455379404,
-      "learning_rate": 2.7697282131577752e-06,
-      "loss": 0.3981,
+      "epoch": 2.62,
+      "grad_norm": 1.838136670411088,
+      "learning_rate": 2.722448485013046e-06,
+      "loss": 0.3997,
       "step": 1500
     },
     {
-      "epoch": 2.6,
-      "grad_norm": 1.8140754605201832,
-      "learning_rate": 2.754713513349627e-06,
-      "loss": 0.3976,
+      "epoch": 2.63,
+      "grad_norm": 1.792723761937039,
+      "learning_rate": 2.707248577561854e-06,
+      "loss": 0.4029,
       "step": 1505
     },
     {
-      "epoch": 2.61,
-      "grad_norm": 1.925171255479977,
-      "learning_rate": 2.7396895236697833e-06,
-      "loss": 0.3839,
+      "epoch": 2.64,
+      "grad_norm": 1.8102828207451078,
+      "learning_rate": 2.6920409519666173e-06,
+      "loss": 0.4249,
       "step": 1510
     },
     {
-      "epoch": 2.62,
-      "grad_norm": 1.6916338723474953,
-      "learning_rate": 2.724656792070861e-06,
-      "loss": 0.4009,
+      "epoch": 2.65,
+      "grad_norm": 2.1220720458624465,
+      "learning_rate": 2.6768261745745037e-06,
+      "loss": 0.4106,
       "step": 1515
     },
     {
-      "epoch": 2.63,
-      "grad_norm": 1.7468445915903061,
-      "learning_rate": 2.7096158668243105e-06,
-      "loss": 0.3954,
+      "epoch": 2.66,
+      "grad_norm": 1.812641713649887,
+      "learning_rate": 2.6616048119990214e-06,
+      "loss": 0.3944,
       "step": 1520
     },
     {
-      "epoch": 2.64,
-      "grad_norm": 1.7964976583564236,
-      "learning_rate": 2.6945672965004236e-06,
-      "loss": 0.3943,
+      "epoch": 2.67,
+      "grad_norm": 2.4648885003841046,
+      "learning_rate": 2.6463774310989154e-06,
+      "loss": 0.3971,
       "step": 1525
     },
     {
-      "epoch": 2.65,
-      "grad_norm": 1.6933144469510995,
-      "learning_rate": 2.679511629948319e-06,
-      "loss": 0.3958,
+      "epoch": 2.67,
+      "grad_norm": 1.85108667649988,
+      "learning_rate": 2.6311445989570633e-06,
+      "loss": 0.3975,
       "step": 1530
     },
     {
-      "epoch": 2.66,
-      "grad_norm": 1.7617290180955545,
-      "learning_rate": 2.664449416275928e-06,
-      "loss": 0.3926,
+      "epoch": 2.68,
+      "grad_norm": 1.7524914725271679,
+      "learning_rate": 2.615906882859349e-06,
+      "loss": 0.392,
       "step": 1535
     },
     {
-      "epoch": 2.66,
-      "grad_norm": 1.6597635128518158,
-      "learning_rate": 2.64938120482997e-06,
-      "loss": 0.4052,
+      "epoch": 2.69,
+      "grad_norm": 2.0229317958148867,
+      "learning_rate": 2.6006648502735384e-06,
+      "loss": 0.4123,
       "step": 1540
     },
     {
-      "epoch": 2.67,
-      "grad_norm": 2.1579541798227213,
-      "learning_rate": 2.6343075451759098e-06,
-      "loss": 0.3962,
+      "epoch": 2.7,
+      "grad_norm": 1.8570564954534334,
+      "learning_rate": 2.585419068828152e-06,
+      "loss": 0.4174,
       "step": 1545
     },
     {
-      "epoch": 2.68,
-      "grad_norm": 1.6871485325523872,
-      "learning_rate": 2.619228987077923e-06,
-      "loss": 0.3975,
+      "epoch": 2.71,
+      "grad_norm": 1.8937349012574363,
+      "learning_rate": 2.5701701062913194e-06,
+      "loss": 0.4265,
       "step": 1550
     },
     {
-      "epoch": 2.69,
-      "grad_norm": 2.0167255639712436,
-      "learning_rate": 2.604146080478839e-06,
-      "loss": 0.3794,
+      "epoch": 2.72,
+      "grad_norm": 1.7590069556724997,
+      "learning_rate": 2.554918530549637e-06,
+      "loss": 0.3863,
       "step": 1555
     },
     {
-      "epoch": 2.7,
-      "grad_norm": 1.8181334973476126,
-      "learning_rate": 2.5890593754800826e-06,
-      "loss": 0.4061,
+      "epoch": 2.73,
+      "grad_norm": 2.096408205769661,
+      "learning_rate": 2.53966490958702e-06,
+      "loss": 0.396,
       "step": 1560
     },
     {
-      "epoch": 2.71,
-      "grad_norm": 1.753744913485219,
-      "learning_rate": 2.573969422321617e-06,
-      "loss": 0.386,
+      "epoch": 2.74,
+      "grad_norm": 1.962464797958802,
+      "learning_rate": 2.5244098114635503e-06,
+      "loss": 0.4142,
       "step": 1565
     },
     {
-      "epoch": 2.72,
-      "grad_norm": 1.7053212636407111,
-      "learning_rate": 2.558876771361869e-06,
-      "loss": 0.3835,
+      "epoch": 2.74,
+      "grad_norm": 1.8458575978769718,
+      "learning_rate": 2.5091538042943183e-06,
+      "loss": 0.3946,
       "step": 1570
     },
     {
-      "epoch": 2.72,
-      "grad_norm": 2.0262024426335903,
-      "learning_rate": 2.5437819730576608e-06,
-      "loss": 0.3925,
+      "epoch": 2.75,
+      "grad_norm": 1.8269976161284471,
+      "learning_rate": 2.4938974562282708e-06,
+      "loss": 0.4127,
       "step": 1575
     },
     {
-      "epoch": 2.73,
-      "grad_norm": 1.7593848195979234,
-      "learning_rate": 2.528685577944131e-06,
-      "loss": 0.3856,
+      "epoch": 2.76,
+      "grad_norm": 2.01278268309966,
+      "learning_rate": 2.4786413354270494e-06,
+      "loss": 0.4032,
       "step": 1580
     },
     {
-      "epoch": 2.74,
-      "grad_norm": 1.7817581606319834,
-      "learning_rate": 2.5135881366146585e-06,
-      "loss": 0.379,
+      "epoch": 2.77,
+      "grad_norm": 1.7849542357371841,
+      "learning_rate": 2.4633860100438317e-06,
+      "loss": 0.3905,
       "step": 1585
     },
     {
-      "epoch": 2.75,
-      "grad_norm": 1.9582822708483771,
-      "learning_rate": 2.498490199700778e-06,
-      "loss": 0.4028,
+      "epoch": 2.78,
+      "grad_norm": 69.63657032212085,
+      "learning_rate": 2.4481320482021716e-06,
+      "loss": 0.4038,
       "step": 1590
     },
     {
-      "epoch": 2.76,
-      "grad_norm": 1.737025758915202,
-      "learning_rate": 2.4833923178520984e-06,
-      "loss": 0.3918,
+      "epoch": 2.79,
+      "grad_norm": 2.1742738648421023,
+      "learning_rate": 2.4328800179748475e-06,
+      "loss": 0.3888,
       "step": 1595
     },
     {
-      "epoch": 2.77,
-      "grad_norm": 1.8457649877934572,
-      "learning_rate": 2.468295041716223e-06,
-      "loss": 0.3869,
+      "epoch": 2.8,
+      "grad_norm": 2.8552125373844985,
+      "learning_rate": 2.4176304873626983e-06,
+      "loss": 0.4066,
       "step": 1600
     },
     {
-      "epoch": 2.78,
-      "grad_norm": 1.7058419645908771,
-      "learning_rate": 2.4531989219186626e-06,
-      "loss": 0.3969,
+      "epoch": 2.81,
+      "grad_norm": 2.1841551939924244,
+      "learning_rate": 2.4023840242734774e-06,
+      "loss": 0.3997,
       "step": 1605
     },
     {
-      "epoch": 2.79,
-      "grad_norm": 1.7424467938884878,
-      "learning_rate": 2.4381045090427525e-06,
-      "loss": 0.3944,
+      "epoch": 2.81,
+      "grad_norm": 2.194040510042996,
+      "learning_rate": 2.3871411965006985e-06,
+      "loss": 0.4102,
       "step": 1610
     },
     {
-      "epoch": 2.79,
-      "grad_norm": 1.7229435032812541,
-      "learning_rate": 2.4230123536095746e-06,
-      "loss": 0.392,
+      "epoch": 2.82,
+      "grad_norm": 2.3916077476597475,
+      "learning_rate": 2.3719025717024946e-06,
+      "loss": 0.4014,
       "step": 1615
     },
     {
-      "epoch": 2.8,
-      "grad_norm": 1.6940195802109876,
-      "learning_rate": 2.4079230060578787e-06,
-      "loss": 0.3921,
+      "epoch": 2.83,
+      "grad_norm": 1.9141926801542115,
+      "learning_rate": 2.3566687173804747e-06,
+      "loss": 0.3977,
       "step": 1620
     },
     {
-      "epoch": 2.81,
-      "grad_norm": 1.926203653659556,
-      "learning_rate": 2.3928370167240037e-06,
-      "loss": 0.3772,
+      "epoch": 2.84,
+      "grad_norm": 1.8144717702331585,
+      "learning_rate": 2.341440200858589e-06,
+      "loss": 0.3964,
       "step": 1625
     },
     {
-      "epoch": 2.82,
-      "grad_norm": 2.3907546286805443,
-      "learning_rate": 2.3777549358218105e-06,
-      "loss": 0.3829,
+      "epoch": 2.85,
+      "grad_norm": 1.851593659208308,
+      "learning_rate": 2.3262175892620064e-06,
+      "loss": 0.4085,
       "step": 1630
     },
     {
-      "epoch": 2.83,
-      "grad_norm": 1.7061727789752485,
-      "learning_rate": 2.3626773134226114e-06,
-      "loss": 0.3821,
+      "epoch": 2.86,
+      "grad_norm": 1.8686287238442707,
+      "learning_rate": 2.311001449495986e-06,
+      "loss": 0.4002,
       "step": 1635
     },
     {
-      "epoch": 2.84,
-      "grad_norm": 4.31283876155316,
-      "learning_rate": 2.3476046994351085e-06,
-      "loss": 0.3885,
+      "epoch": 2.87,
+      "grad_norm": 1.898471354316753,
+      "learning_rate": 2.2957923482247745e-06,
+      "loss": 0.399,
       "step": 1640
     },
     {
-      "epoch": 2.85,
-      "grad_norm": 1.815971909695708,
-      "learning_rate": 2.332537643585337e-06,
-      "loss": 0.4065,
+      "epoch": 2.88,
+      "grad_norm": 1.8645673290000866,
+      "learning_rate": 2.280590851850493e-06,
+      "loss": 0.4025,
       "step": 1645
     },
     {
-      "epoch": 2.85,
-      "grad_norm": 1.8273993097485424,
-      "learning_rate": 2.3174766953966178e-06,
-      "loss": 0.403,
+      "epoch": 2.88,
+      "grad_norm": 1.8195968879508377,
+      "learning_rate": 2.265397526492052e-06,
+      "loss": 0.3942,
       "step": 1650
     },
     {
-      "epoch": 2.86,
-      "grad_norm": 2.3245389527585347,
-      "learning_rate": 2.302422404169516e-06,
-      "loss": 0.3813,
+      "epoch": 2.89,
+      "grad_norm": 2.133179133462322,
+      "learning_rate": 2.2502129379640644e-06,
+      "loss": 0.4135,
       "step": 1655
     },
     {
-      "epoch": 2.87,
-      "grad_norm": 2.5008928168129305,
-      "learning_rate": 2.2873753189618016e-06,
-      "loss": 0.3958,
+      "epoch": 2.9,
+      "grad_norm": 1.790567048936156,
+      "learning_rate": 2.235037651755773e-06,
+      "loss": 0.3889,
       "step": 1660
     },
     {
-      "epoch": 2.88,
-      "grad_norm": 1.7982648722278374,
-      "learning_rate": 2.272335988568429e-06,
-      "loss": 0.3939,
+      "epoch": 2.91,
+      "grad_norm": 1.7690615361620154,
+      "learning_rate": 2.2198722330099964e-06,
+      "loss": 0.3971,
       "step": 1665
     },
     {
-      "epoch": 2.89,
-      "grad_norm": 1.7113353706744565,
-      "learning_rate": 2.257304961501521e-06,
-      "loss": 0.3831,
+      "epoch": 2.92,
+      "grad_norm": 2.0091067058646543,
+      "learning_rate": 2.2047172465020757e-06,
+      "loss": 0.3953,
       "step": 1670
     },
     {
-      "epoch": 2.9,
-      "grad_norm": 1.8571633665118874,
-      "learning_rate": 2.242282785970361e-06,
-      "loss": 0.4036,
+      "epoch": 2.93,
+      "grad_norm": 1.9832659283001421,
+      "learning_rate": 2.1895732566188475e-06,
+      "loss": 0.396,
       "step": 1675
     },
     {
-      "epoch": 2.91,
-      "grad_norm": 1.6973636101504117,
-      "learning_rate": 2.2272700098614022e-06,
-      "loss": 0.3897,
+      "epoch": 2.94,
+      "grad_norm": 1.7695360334830164,
+      "learning_rate": 2.1744408273376204e-06,
+      "loss": 0.3864,
       "step": 1680
     },
     {
-      "epoch": 2.92,
-      "grad_norm": 1.8121544488710108,
-      "learning_rate": 2.2122671807182847e-06,
-      "loss": 0.3997,
+      "epoch": 2.95,
+      "grad_norm": 1.9266047240521331,
+      "learning_rate": 2.159320522205179e-06,
+      "loss": 0.3902,
       "step": 1685
     },
     {
-      "epoch": 2.92,
-      "grad_norm": 1.7018502297212512,
-      "learning_rate": 2.1972748457218608e-06,
-      "loss": 0.3907,
+      "epoch": 2.95,
+      "grad_norm": 1.9067745089946206,
+      "learning_rate": 2.1442129043167877e-06,
+      "loss": 0.4048,
       "step": 1690
     },
     {
-      "epoch": 2.93,
-      "grad_norm": 1.5654175282292644,
-      "learning_rate": 2.182293551670245e-06,
-      "loss": 0.3849,
+      "epoch": 2.96,
+      "grad_norm": 1.8121353203087656,
+      "learning_rate": 2.1291185362952274e-06,
+      "loss": 0.3949,
       "step": 1695
     },
     {
-      "epoch": 2.94,
-      "grad_norm": 1.8131129781015132,
-      "learning_rate": 2.167323844958867e-06,
-      "loss": 0.3991,
+      "epoch": 2.97,
+      "grad_norm": 1.8495794192789976,
+      "learning_rate": 2.114037980269842e-06,
+      "loss": 0.4038,
       "step": 1700
     },
     {
-      "epoch": 2.95,
-      "grad_norm": 1.6556951661263792,
-      "learning_rate": 2.152366271560543e-06,
-      "loss": 0.3876,
+      "epoch": 2.98,
+      "grad_norm": 1.912104342675586,
+      "learning_rate": 2.0989717978555992e-06,
+      "loss": 0.3919,
       "step": 1705
     },
     {
-      "epoch": 2.96,
-      "grad_norm": 1.6927349202503557,
-      "learning_rate": 2.1374213770055692e-06,
-      "loss": 0.3942,
+      "epoch": 2.99,
+      "grad_norm": 1.8275208022684235,
+      "learning_rate": 2.0839205501321844e-06,
+      "loss": 0.4066,
       "step": 1710
     },
     {
-      "epoch": 2.97,
-      "grad_norm": 1.8331415491875973,
-      "learning_rate": 2.122489706361818e-06,
-      "loss": 0.3791,
+      "epoch": 3.0,
+      "grad_norm": 2.057130603594985,
+      "learning_rate": 2.0688847976230952e-06,
+      "loss": 0.3858,
       "step": 1715
     },
     {
-      "epoch": 2.98,
-      "grad_norm": 1.8546912165375895,
-      "learning_rate": 2.107571804214861e-06,
-      "loss": 0.3863,
-      "step": 1720
+      "epoch": 3.0,
+      "eval_loss": 0.377420037984848,
+      "eval_runtime": 330.5022,
+      "eval_samples_per_second": 22.756,
+      "eval_steps_per_second": 0.357,
+      "step": 1716
     },
     {
-      "epoch": 2.98,
-      "grad_norm": 1.8701558336845174,
-      "learning_rate": 2.0926682146481094e-06,
-      "loss": 0.3918,
-      "step": 1725
+      "epoch": 3.01,
+      "grad_norm": 3.3388932256530697,
+      "learning_rate": 2.0538651002747745e-06,
+      "loss": 0.2972,
+      "step": 1720
     },
     {
-      "epoch": 2.99,
-      "grad_norm": 1.6438729463581823,
-      "learning_rate": 2.077779481222967e-06,
-      "loss": 0.4058,
-      "step": 1730
+      "epoch": 3.02,
+      "grad_norm": 2.8981240612419423,
+      "learning_rate": 2.0388620174357542e-06,
+      "loss": 0.3002,
+      "step": 1725
     },
     {
-      "epoch": 3.0,
-      "eval_loss": 0.41721072793006897,
-      "eval_runtime": 329.8932,
-      "eval_samples_per_second": 22.798,
-      "eval_steps_per_second": 0.358,
-      "step": 1734
+      "epoch": 3.02,
+      "grad_norm": 2.205591361217027,
+      "learning_rate": 2.023876107835825e-06,
+      "loss": 0.3135,
+      "step": 1730
     },
     {
-      "epoch": 3.0,
-      "grad_norm": 3.362012279132523,
-      "learning_rate": 2.0629061469590046e-06,
-      "loss": 0.3616,
+      "epoch": 3.03,
+      "grad_norm": 2.1495938527408835,
+      "learning_rate": 2.008907929565231e-06,
+      "loss": 0.3126,
       "step": 1735
     },
     {
-      "epoch": 3.01,
-      "grad_norm": 2.405517474268548,
-      "learning_rate": 2.048048754314162e-06,
-      "loss": 0.3281,
+      "epoch": 3.04,
+      "grad_norm": 2.3520040686767167,
+      "learning_rate": 1.993958040053881e-06,
+      "loss": 0.3209,
       "step": 1740
     },
     {
-      "epoch": 3.02,
-      "grad_norm": 2.252602528540252,
-      "learning_rate": 2.0332078451649557e-06,
-      "loss": 0.3167,
+      "epoch": 3.05,
+      "grad_norm": 2.0650913293200848,
+      "learning_rate": 1.9790269960505947e-06,
+      "loss": 0.2944,
       "step": 1745
     },
     {
-      "epoch": 3.03,
-      "grad_norm": 1.8994332344961664,
-      "learning_rate": 2.0183839607867177e-06,
-      "loss": 0.3061,
+      "epoch": 3.06,
+      "grad_norm": 2.1220138588069704,
+      "learning_rate": 1.9641153536023646e-06,
+      "loss": 0.308,
       "step": 1750
     },
     {
-      "epoch": 3.04,
-      "grad_norm": 2.0025910547043138,
-      "learning_rate": 2.003577641833859e-06,
-      "loss": 0.3205,
+      "epoch": 3.07,
+      "grad_norm": 2.150988951055739,
+      "learning_rate": 1.9492236680336486e-06,
+      "loss": 0.2931,
       "step": 1755
     },
     {
-      "epoch": 3.04,
-      "grad_norm": 1.8748618784934366,
-      "learning_rate": 1.988789428320141e-06,
-      "loss": 0.3101,
+      "epoch": 3.08,
+      "grad_norm": 2.124000470558878,
+      "learning_rate": 1.934352493925695e-06,
+      "loss": 0.2995,
       "step": 1760
     },
     {
-      "epoch": 3.05,
-      "grad_norm": 1.8182888367049812,
-      "learning_rate": 1.9740198595989935e-06,
-      "loss": 0.3071,
+      "epoch": 3.09,
+      "grad_norm": 1.9934393020695829,
+      "learning_rate": 1.9195023850958812e-06,
+      "loss": 0.2926,
       "step": 1765
     },
     {
-      "epoch": 3.06,
-      "grad_norm": 1.9091898687921862,
-      "learning_rate": 1.959269474343832e-06,
-      "loss": 0.308,
+      "epoch": 3.09,
+      "grad_norm": 2.154505232085453,
+      "learning_rate": 1.9046738945770932e-06,
+      "loss": 0.3057,
       "step": 1770
     },
     {
-      "epoch": 3.07,
-      "grad_norm": 1.7973778535481801,
-      "learning_rate": 1.9445388105284157e-06,
-      "loss": 0.3031,
+      "epoch": 3.1,
+      "grad_norm": 2.1900958746909565,
+      "learning_rate": 1.889867574597129e-06,
+      "loss": 0.3012,
       "step": 1775
     },
     {
-      "epoch": 3.08,
-      "grad_norm": 1.7575417892395457,
-      "learning_rate": 1.9298284054072278e-06,
-      "loss": 0.3063,
+      "epoch": 3.11,
+      "grad_norm": 1.8955650762693546,
+      "learning_rate": 1.875083976558136e-06,
+      "loss": 0.2838,
       "step": 1780
     },
     {
-      "epoch": 3.09,
-      "grad_norm": 1.8133263463584137,
-      "learning_rate": 1.9151387954958792e-06,
-      "loss": 0.3008,
+      "epoch": 3.12,
+      "grad_norm": 2.0442738100465783,
+      "learning_rate": 1.860323651016072e-06,
+      "loss": 0.2977,
       "step": 1785
     },
     {
-      "epoch": 3.1,
-      "grad_norm": 1.892423366282607,
-      "learning_rate": 1.9004705165515403e-06,
-      "loss": 0.33,
+      "epoch": 3.13,
+      "grad_norm": 2.110043278615698,
+      "learning_rate": 1.8455871476602023e-06,
+      "loss": 0.2984,
       "step": 1790
     },
     {
-      "epoch": 3.11,
-      "grad_norm": 3.4152436955589596,
-      "learning_rate": 1.885824103553404e-06,
-      "loss": 0.2967,
+      "epoch": 3.14,
+      "grad_norm": 2.3136510034991282,
+      "learning_rate": 1.8308750152926338e-06,
+      "loss": 0.2927,
       "step": 1795
     },
     {
-      "epoch": 3.11,
-      "grad_norm": 1.766725631679472,
-      "learning_rate": 1.8712000906831684e-06,
-      "loss": 0.3134,
+      "epoch": 3.15,
+      "grad_norm": 2.0131063879803826,
+      "learning_rate": 1.8161878018078693e-06,
+      "loss": 0.2908,
       "step": 1800
     },
     {
-      "epoch": 3.12,
-      "grad_norm": 1.7559925086788295,
-      "learning_rate": 1.8565990113055615e-06,
-      "loss": 0.3117,
+      "epoch": 3.16,
+      "grad_norm": 2.1355586799767674,
+      "learning_rate": 1.8015260541724128e-06,
+      "loss": 0.2905,
       "step": 1805
     },
     {
-      "epoch": 3.13,
-      "grad_norm": 1.7744634591819772,
-      "learning_rate": 1.8420213979488805e-06,
-      "loss": 0.3188,
+      "epoch": 3.16,
+      "grad_norm": 2.0664019629242087,
+      "learning_rate": 1.7868903184043888e-06,
+      "loss": 0.2955,
       "step": 1810
     },
     {
-      "epoch": 3.14,
-      "grad_norm": 1.7585191953453494,
-      "learning_rate": 1.8274677822855775e-06,
-      "loss": 0.3052,
+      "epoch": 3.17,
+      "grad_norm": 2.188481583652719,
+      "learning_rate": 1.772281139553218e-06,
+      "loss": 0.3071,
       "step": 1815
     },
     {
-      "epoch": 3.15,
-      "grad_norm": 1.8690590027859153,
-      "learning_rate": 1.81293869511286e-06,
-      "loss": 0.3113,
+      "epoch": 3.18,
+      "grad_norm": 2.0847120701500774,
+      "learning_rate": 1.7576990616793139e-06,
+      "loss": 0.2908,
       "step": 1820
     },
     {
-      "epoch": 3.16,
-      "grad_norm": 1.8668392799746427,
-      "learning_rate": 1.7984346663333415e-06,
-      "loss": 0.3115,
+      "epoch": 3.19,
+      "grad_norm": 2.018064853932535,
+      "learning_rate": 1.74314462783382e-06,
+      "loss": 0.2967,
       "step": 1825
     },
     {
-      "epoch": 3.17,
-      "grad_norm": 1.8555764810481752,
-      "learning_rate": 1.783956224935705e-06,
-      "loss": 0.3198,
+      "epoch": 3.2,
+      "grad_norm": 2.0242538886830843,
+      "learning_rate": 1.7286183800383937e-06,
+      "loss": 0.3017,
       "step": 1830
     },
     {
-      "epoch": 3.17,
-      "grad_norm": 1.7812617913871858,
-      "learning_rate": 1.7695038989754179e-06,
-      "loss": 0.3145,
+      "epoch": 3.21,
+      "grad_norm": 2.145977098033245,
+      "learning_rate": 1.714120859265011e-06,
+      "loss": 0.2872,
       "step": 1835
     },
     {
-      "epoch": 3.18,
-      "grad_norm": 1.7431068806527559,
-      "learning_rate": 1.7550782155554664e-06,
-      "loss": 0.2926,
+      "epoch": 3.22,
+      "grad_norm": 1.9029268616901622,
+      "learning_rate": 1.6996526054158283e-06,
+      "loss": 0.2888,
       "step": 1840
     },
     {
-      "epoch": 3.19,
-      "grad_norm": 1.8768744877192187,
-      "learning_rate": 1.7406797008071351e-06,
-      "loss": 0.3113,
+      "epoch": 3.23,
+      "grad_norm": 4.096519984140158,
+      "learning_rate": 1.685214157303069e-06,
+      "loss": 0.2966,
       "step": 1845
     },
     {
-      "epoch": 3.2,
-      "grad_norm": 1.8667603957089496,
-      "learning_rate": 1.7263088798708206e-06,
-      "loss": 0.3179,
+      "epoch": 3.23,
+      "grad_norm": 2.1188967978481013,
+      "learning_rate": 1.6708060526289648e-06,
+      "loss": 0.299,
       "step": 1850
     },
     {
-      "epoch": 3.21,
-      "grad_norm": 2.1299587342389072,
-      "learning_rate": 1.7119662768768694e-06,
-      "loss": 0.3171,
+      "epoch": 3.24,
+      "grad_norm": 2.7278823372957666,
+      "learning_rate": 1.6564288279657253e-06,
+      "loss": 0.3034,
       "step": 1855
     },
     {
-      "epoch": 3.22,
-      "grad_norm": 1.7486090043188862,
-      "learning_rate": 1.6976524149264707e-06,
-      "loss": 0.3364,
+      "epoch": 3.25,
+      "grad_norm": 2.0002017052238257,
+      "learning_rate": 1.6420830187355572e-06,
+      "loss": 0.2958,
       "step": 1860
     },
     {
-      "epoch": 3.23,
-      "grad_norm": 2.728922445411206,
-      "learning_rate": 1.6833678160725746e-06,
-      "loss": 0.3104,
+      "epoch": 3.26,
+      "grad_norm": 3.6893974280355097,
+      "learning_rate": 1.6277691591907272e-06,
+      "loss": 0.2954,
       "step": 1865
     },
     {
-      "epoch": 3.24,
-      "grad_norm": 1.662819449057055,
-      "learning_rate": 1.6691130013008514e-06,
-      "loss": 0.3128,
+      "epoch": 3.27,
+      "grad_norm": 1.8667517556916042,
+      "learning_rate": 1.613487782393661e-06,
+      "loss": 0.2933,
       "step": 1870
     },
     {
-      "epoch": 3.24,
-      "grad_norm": 1.944417319707403,
-      "learning_rate": 1.654888490510691e-06,
-      "loss": 0.3136,
+      "epoch": 3.28,
+      "grad_norm": 2.0838269309261364,
+      "learning_rate": 1.599239420197098e-06,
+      "loss": 0.297,
       "step": 1875
     },
     {
-      "epoch": 3.25,
-      "grad_norm": 2.543538356158386,
-      "learning_rate": 1.6406948024962438e-06,
-      "loss": 0.3139,
+      "epoch": 3.29,
+      "grad_norm": 1.9755359084479687,
+      "learning_rate": 1.5850246032242766e-06,
+      "loss": 0.2909,
       "step": 1880
     },
     {
-      "epoch": 3.26,
-      "grad_norm": 1.8386914789973714,
-      "learning_rate": 1.6265324549274924e-06,
-      "loss": 0.3164,
+      "epoch": 3.3,
+      "grad_norm": 2096.7504686136917,
+      "learning_rate": 1.5708438608491816e-06,
+      "loss": 0.2951,
       "step": 1885
     },
     {
-      "epoch": 3.27,
-      "grad_norm": 1.8542366087207915,
-      "learning_rate": 1.6124019643313783e-06,
-      "loss": 0.3086,
+      "epoch": 3.3,
+      "grad_norm": 2.918573785412608,
+      "learning_rate": 1.556697721176823e-06,
+      "loss": 0.3012,
       "step": 1890
     },
     {
-      "epoch": 3.28,
-      "grad_norm": 1.774546610521189,
-      "learning_rate": 1.59830384607296e-06,
-      "loss": 0.3128,
+      "epoch": 3.31,
+      "grad_norm": 4.274328804415084,
+      "learning_rate": 1.5425867110235717e-06,
+      "loss": 0.2876,
       "step": 1895
     },
     {
-      "epoch": 3.29,
-      "grad_norm": 2.1592993916723007,
-      "learning_rate": 1.5842386143366157e-06,
-      "loss": 0.3102,
+      "epoch": 3.32,
+      "grad_norm": 5.92446369343208,
+      "learning_rate": 1.5285113558975429e-06,
+      "loss": 0.3466,
       "step": 1900
     },
     {
-      "epoch": 3.3,
-      "grad_norm": 1.8046821102537511,
-      "learning_rate": 1.570206782107294e-06,
-      "loss": 0.3156,
+      "epoch": 3.33,
+      "grad_norm": 5.122609680763134,
+      "learning_rate": 1.5144721799790194e-06,
+      "loss": 0.3023,
       "step": 1905
     },
     {
-      "epoch": 3.3,
-      "grad_norm": 1.873775956948585,
-      "learning_rate": 1.5562088611518007e-06,
-      "loss": 0.3093,
+      "epoch": 3.34,
+      "grad_norm": 5.036779121717785,
+      "learning_rate": 1.5004697061009372e-06,
+      "loss": 0.3013,
       "step": 1910
     },
     {
-      "epoch": 3.31,
-      "grad_norm": 1.8924372370403861,
-      "learning_rate": 1.542245362000134e-06,
-      "loss": 0.3027,
+      "epoch": 3.35,
+      "grad_norm": 4.110935956575215,
+      "learning_rate": 1.486504455729408e-06,
+      "loss": 0.2915,
       "step": 1915
     },
     {
-      "epoch": 3.32,
-      "grad_norm": 1.879324682842581,
-      "learning_rate": 1.528316793926866e-06,
-      "loss": 0.3263,
+      "epoch": 3.36,
+      "grad_norm": 2.867011840176696,
+      "learning_rate": 1.4725769489443082e-06,
+      "loss": 0.2822,
       "step": 1920
     },
     {
-      "epoch": 3.33,
-      "grad_norm": 2.1352207068371465,
-      "learning_rate": 1.5144236649325692e-06,
-      "loss": 0.3163,
+      "epoch": 3.37,
+      "grad_norm": 2.9818315723679327,
+      "learning_rate": 1.4586877044199015e-06,
+      "loss": 0.286,
       "step": 1925
     },
     {
-      "epoch": 3.34,
-      "grad_norm": 2.164953693631762,
-      "learning_rate": 1.5005664817252844e-06,
-      "loss": 0.3104,
+      "epoch": 3.37,
+      "grad_norm": 3.284507097488489,
+      "learning_rate": 1.4448372394055249e-06,
+      "loss": 0.3031,
       "step": 1930
     },
     {
-      "epoch": 3.35,
-      "grad_norm": 1.6787075265475386,
-      "learning_rate": 1.486745749702047e-06,
-      "loss": 0.3004,
+      "epoch": 3.38,
+      "grad_norm": 2.6446231205609907,
+      "learning_rate": 1.431026069706335e-06,
+      "loss": 0.2951,
       "step": 1935
     },
     {
-      "epoch": 3.36,
-      "grad_norm": 2.3266338701503537,
-      "learning_rate": 1.4729619729304486e-06,
-      "loss": 0.3084,
+      "epoch": 3.39,
+      "grad_norm": 2.429736926817698,
+      "learning_rate": 1.4172547096640837e-06,
+      "loss": 0.2948,
       "step": 1940
     },
     {
-      "epoch": 3.37,
-      "grad_norm": 2.1291831085187827,
-      "learning_rate": 1.4592156541302542e-06,
-      "loss": 0.3212,
+      "epoch": 3.4,
+      "grad_norm": 2.4087174238529565,
+      "learning_rate": 1.4035236721379758e-06,
+      "loss": 0.3079,
       "step": 1945
     },
     {
-      "epoch": 3.37,
-      "grad_norm": 1.8834065519660157,
-      "learning_rate": 1.4455072946550688e-06,
-      "loss": 0.3018,
+      "epoch": 3.41,
+      "grad_norm": 2.346665896073256,
+      "learning_rate": 1.3898334684855647e-06,
+      "loss": 0.2884,
       "step": 1950
     },
     {
-      "epoch": 3.38,
-      "grad_norm": 1.7558438161458612,
-      "learning_rate": 1.4318373944740485e-06,
-      "loss": 0.3151,
+      "epoch": 3.42,
+      "grad_norm": 2.2112665369477806,
+      "learning_rate": 1.376184608543709e-06,
+      "loss": 0.2784,
       "step": 1955
     },
     {
-      "epoch": 3.39,
-      "grad_norm": 1.945827671485564,
-      "learning_rate": 1.4182064521536693e-06,
-      "loss": 0.3376,
+      "epoch": 3.43,
+      "grad_norm": 2.3555732075174705,
+      "learning_rate": 1.3625776006095882e-06,
+      "loss": 0.2951,
       "step": 1960
     },
     {
-      "epoch": 3.4,
-      "grad_norm": 1.807537024750888,
-      "learning_rate": 1.4046149648395451e-06,
-      "loss": 0.3275,
+      "epoch": 3.44,
+      "grad_norm": 2.453057863316602,
+      "learning_rate": 1.3490129514217665e-06,
+      "loss": 0.2912,
       "step": 1965
     },
     {
-      "epoch": 3.41,
-      "grad_norm": 2.0683028288702885,
-      "learning_rate": 1.391063428238288e-06,
-      "loss": 0.3141,
+      "epoch": 3.44,
+      "grad_norm": 2.349215074609662,
+      "learning_rate": 1.3354911661413305e-06,
+      "loss": 0.3175,
       "step": 1970
     },
     {
-      "epoch": 3.42,
-      "grad_norm": 2.0081489992818566,
-      "learning_rate": 1.3775523365994364e-06,
-      "loss": 0.3181,
+      "epoch": 3.45,
+      "grad_norm": 2.369663823495378,
+      "learning_rate": 1.3220127483330714e-06,
+      "loss": 0.3115,
       "step": 1975
     },
     {
-      "epoch": 3.43,
-      "grad_norm": 2.056354755418753,
-      "learning_rate": 1.3640821826974272e-06,
-      "loss": 0.3203,
+      "epoch": 3.46,
+      "grad_norm": 2.353086780550841,
+      "learning_rate": 1.3085781999467303e-06,
+      "loss": 0.2974,
       "step": 1980
     },
     {
-      "epoch": 3.43,
-      "grad_norm": 1.8074911366312743,
-      "learning_rate": 1.3506534578136184e-06,
-      "loss": 0.3056,
+      "epoch": 3.47,
+      "grad_norm": 2.283791495168621,
+      "learning_rate": 1.2951880212983106e-06,
+      "loss": 0.2963,
       "step": 1985
     },
     {
-      "epoch": 3.44,
-      "grad_norm": 1.9405362029094402,
-      "learning_rate": 1.3372666517183813e-06,
-      "loss": 0.3145,
+      "epoch": 3.48,
+      "grad_norm": 2.285838470923522,
+      "learning_rate": 1.2818427110514382e-06,
+      "loss": 0.2925,
       "step": 1990
     },
     {
-      "epoch": 3.45,
-      "grad_norm": 1.919499046536955,
-      "learning_rate": 1.323922252653228e-06,
-      "loss": 0.315,
+      "epoch": 3.49,
+      "grad_norm": 2.0947843282235863,
+      "learning_rate": 1.2685427661987975e-06,
+      "loss": 0.2944,
       "step": 1995
     },
     {
-      "epoch": 3.46,
-      "grad_norm": 1.9377990366808608,
-      "learning_rate": 1.3106207473130072e-06,
-      "loss": 0.3182,
+      "epoch": 3.5,
+      "grad_norm": 2.181297764037496,
+      "learning_rate": 1.2552886820436208e-06,
+      "loss": 0.2821,
       "step": 2000
     },
     {
-      "epoch": 3.47,
-      "grad_norm": 2.342962414340334,
-      "learning_rate": 1.2973626208281568e-06,
-      "loss": 0.3166,
+      "epoch": 3.51,
+      "grad_norm": 2.2572030477763847,
+      "learning_rate": 1.2420809521812406e-06,
+      "loss": 0.299,
       "step": 2005
     },
     {
-      "epoch": 3.48,
-      "grad_norm": 2.212340995208452,
-      "learning_rate": 1.284148356747007e-06,
-      "loss": 0.3141,
+      "epoch": 3.51,
+      "grad_norm": 2.2492785376097677,
+      "learning_rate": 1.2289200684807098e-06,
+      "loss": 0.3049,
       "step": 2010
     },
     {
-      "epoch": 3.49,
-      "grad_norm": 1.8649674715590283,
-      "learning_rate": 1.2709784370181424e-06,
-      "loss": 0.3077,
+      "epoch": 3.52,
+      "grad_norm": 2.334560206327675,
+      "learning_rate": 1.2158065210664848e-06,
+      "loss": 0.2877,
       "step": 2015
     },
     {
-      "epoch": 3.49,
-      "grad_norm": 1.8611701311779156,
-      "learning_rate": 1.257853341972833e-06,
-      "loss": 0.3052,
+      "epoch": 3.53,
+      "grad_norm": 2.0972126060506375,
+      "learning_rate": 1.2027407983001683e-06,
+      "loss": 0.2818,
       "step": 2020
     },
     {
-      "epoch": 3.5,
-      "grad_norm": 1.7753559046449237,
-      "learning_rate": 1.244773550307506e-06,
-      "loss": 0.3109,
+      "epoch": 3.54,
+      "grad_norm": 2.545800792955665,
+      "learning_rate": 1.189723386762328e-06,
+      "loss": 0.2927,
       "step": 2025
     },
     {
-      "epoch": 3.51,
-      "grad_norm": 2.016998664762509,
-      "learning_rate": 1.23173953906629e-06,
-      "loss": 0.3023,
+      "epoch": 3.55,
+      "grad_norm": 2.2889188636916624,
+      "learning_rate": 1.1767547712343722e-06,
+      "loss": 0.3015,
       "step": 2030
     },
     {
-      "epoch": 3.52,
-      "grad_norm": 1.8857492837767753,
-      "learning_rate": 1.21875178362362e-06,
-      "loss": 0.3247,
+      "epoch": 3.56,
+      "grad_norm": 2.073915350133538,
+      "learning_rate": 1.1638354346804974e-06,
+      "loss": 0.2784,
       "step": 2035
     },
     {
-      "epoch": 3.53,
-      "grad_norm": 1.868969694539886,
-      "learning_rate": 1.205810757666894e-06,
-      "loss": 0.3038,
+      "epoch": 3.57,
+      "grad_norm": 2.2964925609294826,
+      "learning_rate": 1.1509658582297025e-06,
+      "loss": 0.2906,
       "step": 2040
     },
     {
-      "epoch": 3.54,
-      "grad_norm": 1.8277877381221046,
-      "learning_rate": 1.1929169331792018e-06,
-      "loss": 0.3089,
+      "epoch": 3.58,
+      "grad_norm": 2.12295197546072,
+      "learning_rate": 1.1381465211578673e-06,
+      "loss": 0.2894,
       "step": 2045
     },
     {
-      "epoch": 3.55,
-      "grad_norm": 1.7809098288999228,
-      "learning_rate": 1.1800707804221068e-06,
-      "loss": 0.3127,
+      "epoch": 3.58,
+      "grad_norm": 21.13479225301652,
+      "learning_rate": 1.1253779008699131e-06,
+      "loss": 0.2995,
       "step": 2050
     },
     {
-      "epoch": 3.56,
-      "grad_norm": 1.9048483124043887,
-      "learning_rate": 1.1672727679184984e-06,
-      "loss": 0.3227,
+      "epoch": 3.59,
+      "grad_norm": 1.9921395090376566,
+      "learning_rate": 1.1126604728820102e-06,
+      "loss": 0.2766,
       "step": 2055
     },
     {
-      "epoch": 3.56,
-      "grad_norm": 1.7713895884274724,
-      "learning_rate": 1.154523362435499e-06,
-      "loss": 0.3149,
+      "epoch": 3.6,
+      "grad_norm": 2.1541306330776258,
+      "learning_rate": 1.0999947108038816e-06,
+      "loss": 0.3002,
       "step": 2060
     },
     {
-      "epoch": 3.57,
-      "grad_norm": 1.8725797518727274,
-      "learning_rate": 1.1418230289674456e-06,
-      "loss": 0.3021,
+      "epoch": 3.61,
+      "grad_norm": 2.3718321918672824,
+      "learning_rate": 1.0873810863211595e-06,
+      "loss": 0.2894,
       "step": 2065
     },
     {
-      "epoch": 3.58,
-      "grad_norm": 2.122666453139922,
-      "learning_rate": 1.1291722307189285e-06,
-      "loss": 0.3075,
+      "epoch": 3.62,
+      "grad_norm": 2.106333020947831,
+      "learning_rate": 1.074820069177816e-06,
+      "loss": 0.3026,
       "step": 2070
     },
     {
-      "epoch": 3.59,
-      "grad_norm": 1.9298160681079481,
-      "learning_rate": 1.1165714290878962e-06,
-      "loss": 0.3139,
+      "epoch": 3.63,
+      "grad_norm": 2.232973434321066,
+      "learning_rate": 1.0623121271586806e-06,
+      "loss": 0.287,
       "step": 2075
     },
     {
-      "epoch": 3.6,
-      "grad_norm": 1.8940463808973838,
-      "learning_rate": 1.1040210836488286e-06,
-      "loss": 0.3157,
+      "epoch": 3.64,
+      "grad_norm": 2.150526937883329,
+      "learning_rate": 1.049857726072005e-06,
+      "loss": 0.3079,
       "step": 2080
     },
     {
-      "epoch": 3.61,
-      "grad_norm": 2.061247706009593,
-      "learning_rate": 1.091521652135976e-06,
-      "loss": 0.3196,
+      "epoch": 3.65,
+      "grad_norm": 2.4299034348212034,
+      "learning_rate": 1.037457329732127e-06,
+      "loss": 0.2912,
       "step": 2085
     },
     {
-      "epoch": 3.62,
-      "grad_norm": 1.811438510725072,
-      "learning_rate": 1.0790735904266616e-06,
-      "loss": 0.3148,
+      "epoch": 3.65,
+      "grad_norm": 2.2110962609469644,
+      "learning_rate": 1.0251113999421936e-06,
+      "loss": 0.2983,
       "step": 2090
     },
     {
-      "epoch": 3.62,
-      "grad_norm": 1.805396357161127,
-      "learning_rate": 1.0666773525246593e-06,
-      "loss": 0.3036,
+      "epoch": 3.66,
+      "grad_norm": 2.2895025988377857,
+      "learning_rate": 1.0128203964769602e-06,
+      "loss": 0.2804,
       "step": 2095
     },
     {
-      "epoch": 3.63,
-      "grad_norm": 2.045401978025098,
-      "learning_rate": 1.0543333905436325e-06,
-      "loss": 0.3132,
+      "epoch": 3.67,
+      "grad_norm": 1.9663163475739167,
+      "learning_rate": 1.0005847770656757e-06,
+      "loss": 0.3099,
       "step": 2100
     },
     {
-      "epoch": 3.64,
-      "grad_norm": 1.9050403870259225,
-      "learning_rate": 1.0420421546906458e-06,
-      "loss": 0.3149,
+      "epoch": 3.68,
+      "grad_norm": 1.955773416240966,
+      "learning_rate": 9.884049973750268e-07,
+      "loss": 0.2843,
       "step": 2105
     },
     {
-      "epoch": 3.65,
-      "grad_norm": 1.8742997867603652,
-      "learning_rate": 1.0298040932497436e-06,
-      "loss": 0.3289,
+      "epoch": 3.69,
+      "grad_norm": 2.151841568064239,
+      "learning_rate": 9.762815109921762e-07,
+      "loss": 0.2898,
       "step": 2110
     },
     {
-      "epoch": 3.66,
-      "grad_norm": 1.877012236122311,
-      "learning_rate": 1.0176196525656032e-06,
-      "loss": 0.3183,
+      "epoch": 3.7,
+      "grad_norm": 2.2055014532745925,
+      "learning_rate": 9.642147694078664e-07,
+      "loss": 0.2922,
       "step": 2115
     },
     {
-      "epoch": 3.67,
-      "grad_norm": 1.873184037966506,
-      "learning_rate": 1.0054892770272506e-06,
-      "loss": 0.3092,
+      "epoch": 3.71,
+      "grad_norm": 2.0548163347288955,
+      "learning_rate": 9.522052219996072e-07,
+      "loss": 0.2875,
       "step": 2120
     },
     {
-      "epoch": 3.68,
-      "grad_norm": 2.0504833760417434,
-      "learning_rate": 9.934134090518593e-07,
-      "loss": 0.3277,
+      "epoch": 3.72,
+      "grad_norm": 2.0264167858867554,
+      "learning_rate": 9.402533160149415e-07,
+      "loss": 0.2916,
       "step": 2125
     },
     {
-      "epoch": 3.69,
-      "grad_norm": 1.9884415960242454,
-      "learning_rate": 9.813924890686107e-07,
-      "loss": 0.313,
+      "epoch": 3.72,
+      "grad_norm": 1.9310228298997407,
+      "learning_rate": 9.283594965547846e-07,
+      "loss": 0.288,
       "step": 2130
     },
     {
-      "epoch": 3.69,
-      "grad_norm": 1.83107837229764,
-      "learning_rate": 9.694269555026315e-07,
-      "loss": 0.3075,
+      "epoch": 3.73,
+      "grad_norm": 2.0775492416338994,
+      "learning_rate": 9.165242065568547e-07,
+      "loss": 0.3013,
       "step": 2135
     },
     {
-      "epoch": 3.7,
-      "grad_norm": 1.764340908962033,
-      "learning_rate": 9.575172447590032e-07,
-      "loss": 0.3193,
+      "epoch": 3.74,
+      "grad_norm": 4.273341280017886,
+      "learning_rate": 9.047478867791732e-07,
+      "loss": 0.2963,
       "step": 2140
     },
     {
-      "epoch": 3.71,
-      "grad_norm": 1.8477826895669585,
-      "learning_rate": 9.456637912068467e-07,
-      "loss": 0.3125,
+      "epoch": 3.75,
+      "grad_norm": 2.2046805609188,
+      "learning_rate": 8.930309757836517e-07,
+      "loss": 0.2884,
       "step": 2145
     },
     {
-      "epoch": 3.72,
-      "grad_norm": 1.6877441904666532,
-      "learning_rate": 9.338670271634778e-07,
-      "loss": 0.308,
+      "epoch": 3.76,
+      "grad_norm": 2.0943612862709022,
+      "learning_rate": 8.813739099197597e-07,
+      "loss": 0.288,
       "step": 2150
     },
     {
-      "epoch": 3.73,
-      "grad_norm": 1.729827667583889,
-      "learning_rate": 9.221273828786431e-07,
-      "loss": 0.3002,
+      "epoch": 3.77,
+      "grad_norm": 2.470082871683493,
+      "learning_rate": 8.697771233082744e-07,
+      "loss": 0.2958,
       "step": 2155
     },
     {
-      "epoch": 3.74,
-      "grad_norm": 1.8639866274721397,
-      "learning_rate": 9.104452865188254e-07,
-      "loss": 0.3168,
+      "epoch": 3.78,
+      "grad_norm": 2.514580950169932,
+      "learning_rate": 8.582410478251119e-07,
+      "loss": 0.2877,
       "step": 2160
     },
     {
-      "epoch": 3.75,
-      "grad_norm": 1.8584083780028213,
-      "learning_rate": 8.988211641516289e-07,
-      "loss": 0.3214,
+      "epoch": 3.78,
+      "grad_norm": 2.4225946060971326,
+      "learning_rate": 8.467661130852464e-07,
+      "loss": 0.2964,
       "step": 2165
     },
     {
-      "epoch": 3.75,
-      "grad_norm": 1.8211317436181427,
-      "learning_rate": 8.872554397302388e-07,
-      "loss": 0.3109,
+      "epoch": 3.79,
+      "grad_norm": 2.5213364012816752,
+      "learning_rate": 8.353527464267105e-07,
+      "loss": 0.3139,
       "step": 2170
     },
     {
-      "epoch": 3.76,
-      "grad_norm": 1.8219334782565464,
-      "learning_rate": 8.757485350779609e-07,
-      "loss": 0.301,
+      "epoch": 3.8,
+      "grad_norm": 2.1383450216449766,
+      "learning_rate": 8.240013728946786e-07,
+      "loss": 0.3011,
       "step": 2175
     },
     {
-      "epoch": 3.77,
-      "grad_norm": 1.9771362675742574,
-      "learning_rate": 8.64300869872832e-07,
-      "loss": 0.3048,
+      "epoch": 3.81,
+      "grad_norm": 2.275623309460668,
+      "learning_rate": 8.127124152256408e-07,
+      "loss": 0.2877,
       "step": 2180
     },
     {
-      "epoch": 3.78,
-      "grad_norm": 1.875073162601576,
-      "learning_rate": 8.529128616323224e-07,
-      "loss": 0.2992,
+      "epoch": 3.82,
+      "grad_norm": 2.3231659654583945,
+      "learning_rate": 8.014862938316542e-07,
+      "loss": 0.2877,
       "step": 2185
     },
     {
-      "epoch": 3.79,
-      "grad_norm": 1.8832489255751728,
-      "learning_rate": 8.415849256980976e-07,
-      "loss": 0.3054,
+      "epoch": 3.83,
+      "grad_norm": 2.034432326023906,
+      "learning_rate": 7.903234267846965e-07,
+      "loss": 0.2891,
       "step": 2190
     },
     {
-      "epoch": 3.8,
-      "grad_norm": 10.069268574653966,
-      "learning_rate": 8.303174752208786e-07,
-      "loss": 0.3205,
+      "epoch": 3.84,
+      "grad_norm": 2.1427143907161024,
+      "learning_rate": 7.792242298010847e-07,
+      "loss": 0.2905,
       "step": 2195
     },
     {
-      "epoch": 3.81,
-      "grad_norm": 1.8175623122049562,
-      "learning_rate": 8.191109211453688e-07,
-      "loss": 0.3166,
+      "epoch": 3.85,
+      "grad_norm": 1.992125321368494,
+      "learning_rate": 7.681891162260016e-07,
+      "loss": 0.2904,
       "step": 2200
     },
     {
-      "epoch": 3.81,
-      "grad_norm": 1.884935738343672,
-      "learning_rate": 8.079656721952686e-07,
-      "loss": 0.3199,
+      "epoch": 3.85,
+      "grad_norm": 2.2912621172371073,
+      "learning_rate": 7.572184970181005e-07,
+      "loss": 0.2931,
       "step": 2205
     },
     {
-      "epoch": 3.82,
-      "grad_norm": 1.8325908870427998,
-      "learning_rate": 7.968821348583644e-07,
-      "loss": 0.3017,
+      "epoch": 3.86,
+      "grad_norm": 2.1841846462580303,
+      "learning_rate": 7.463127807341966e-07,
+      "loss": 0.2888,
       "step": 2210
     },
     {
-      "epoch": 3.83,
-      "grad_norm": 1.8664284061304677,
-      "learning_rate": 7.858607133717117e-07,
-      "loss": 0.3139,
+      "epoch": 3.87,
+      "grad_norm": 2.4186238871530947,
+      "learning_rate": 7.354723735140609e-07,
+      "loss": 0.2972,
       "step": 2215
     },
     {
-      "epoch": 3.84,
-      "grad_norm": 1.8379727282817748,
-      "learning_rate": 7.749018097068808e-07,
-      "loss": 0.3023,
+      "epoch": 3.88,
+      "grad_norm": 2.165303668523032,
+      "learning_rate": 7.246976790652843e-07,
+      "loss": 0.2894,
       "step": 2220
     },
     {
-      "epoch": 3.85,
-      "grad_norm": 1.7887589319537402,
-      "learning_rate": 7.640058235553044e-07,
-      "loss": 0.3019,
+      "epoch": 3.89,
+      "grad_norm": 1.971010064873144,
+      "learning_rate": 7.139890986482515e-07,
+      "loss": 0.2801,
       "step": 2225
     },
     {
-      "epoch": 3.86,
-      "grad_norm": 1.7307328404201077,
-      "learning_rate": 7.531731523136976e-07,
-      "loss": 0.3121,
+      "epoch": 3.9,
+      "grad_norm": 2.226759681716193,
+      "learning_rate": 7.033470310611945e-07,
+      "loss": 0.2936,
       "step": 2230
     },
     {
-      "epoch": 3.87,
-      "grad_norm": 1.8359406746662572,
-      "learning_rate": 7.424041910695631e-07,
-      "loss": 0.3223,
+      "epoch": 3.91,
+      "grad_norm": 2.128956719636366,
+      "learning_rate": 6.927718726253379e-07,
+      "loss": 0.2873,
       "step": 2235
     },
     {
-      "epoch": 3.88,
-      "grad_norm": 1.866144889812394,
-      "learning_rate": 7.316993325867827e-07,
-      "loss": 0.313,
+      "epoch": 3.92,
+      "grad_norm": 1.9680023419439248,
+      "learning_rate": 6.822640171701486e-07,
+      "loss": 0.2946,
       "step": 2240
     },
     {
-      "epoch": 3.88,
-      "grad_norm": 1.8657860951686938,
-      "learning_rate": 7.210589672912935e-07,
-      "loss": 0.314,
+      "epoch": 3.92,
+      "grad_norm": 2.2622423856131344,
+      "learning_rate": 6.718238560186572e-07,
+      "loss": 0.2825,
       "step": 2245
     },
     {
-      "epoch": 3.89,
-      "grad_norm": 1.9579670233637116,
-      "learning_rate": 7.104834832568438e-07,
-      "loss": 0.2997,
+      "epoch": 3.93,
+      "grad_norm": 1.947066365404126,
+      "learning_rate": 6.614517779728943e-07,
+      "loss": 0.2893,
       "step": 2250
     },
     {
-      "epoch": 3.9,
-      "grad_norm": 1.7626608239046955,
-      "learning_rate": 6.999732661908456e-07,
-      "loss": 0.2939,
+      "epoch": 3.94,
+      "grad_norm": 2.2897303224060064,
+      "learning_rate": 6.511481692994077e-07,
+      "loss": 0.2816,
       "step": 2255
     },
     {
-      "epoch": 3.91,
-      "grad_norm": 1.8904507508265878,
-      "learning_rate": 6.895286994203037e-07,
-      "loss": 0.3173,
+      "epoch": 3.95,
+      "grad_norm": 2.1661490360203937,
+      "learning_rate": 6.409134137148737e-07,
+      "loss": 0.2961,
       "step": 2260
     },
     {
-      "epoch": 3.92,
-      "grad_norm": 1.8375475778504446,
-      "learning_rate": 6.79150163877835e-07,
-      "loss": 0.2928,
+      "epoch": 3.96,
+      "grad_norm": 2.163171587598097,
+      "learning_rate": 6.307478923718171e-07,
+      "loss": 0.2858,
       "step": 2265
     },
     {
-      "epoch": 3.93,
-      "grad_norm": 2.071120288178449,
-      "learning_rate": 6.688380380877755e-07,
-      "loss": 0.3104,
+      "epoch": 3.97,
+      "grad_norm": 2.03844883078776,
+      "learning_rate": 6.206519838444044e-07,
+      "loss": 0.2802,
       "step": 2270
     },
     {
-      "epoch": 3.94,
-      "grad_norm": 1.8806639293010023,
-      "learning_rate": 6.585926981523761e-07,
-      "loss": 0.3113,
+      "epoch": 3.98,
+      "grad_norm": 1.950829082933861,
+      "learning_rate": 6.106260641143547e-07,
+      "loss": 0.3058,
       "step": 2275
     },
     {
-      "epoch": 3.94,
-      "grad_norm": 1.7866185026537371,
-      "learning_rate": 6.484145177380824e-07,
-      "loss": 0.3053,
+      "epoch": 3.99,
+      "grad_norm": 1.971999267799782,
+      "learning_rate": 6.006705065569329e-07,
+      "loss": 0.2809,
       "step": 2280
     },
     {
-      "epoch": 3.95,
-      "grad_norm": 1.8712955462594827,
-      "learning_rate": 6.3830386806191e-07,
-      "loss": 0.3054,
+      "epoch": 3.99,
+      "grad_norm": 1.9523923513832082,
+      "learning_rate": 5.907856819270471e-07,
+      "loss": 0.281,
       "step": 2285
     },
     {
-      "epoch": 3.96,
-      "grad_norm": 1.8278128484701799,
-      "learning_rate": 6.282611178779039e-07,
-      "loss": 0.3226,
+      "epoch": 4.0,
+      "eval_loss": 0.3295736014842987,
+      "eval_runtime": 330.2445,
+      "eval_samples_per_second": 22.774,
+      "eval_steps_per_second": 0.357,
+      "step": 2288
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 4.044140540928026,
+      "learning_rate": 5.809719583454415e-07,
+      "loss": 0.2618,
       "step": 2290
     },
     {
-      "epoch": 3.97,
-      "grad_norm": 1.7845284828008048,
-      "learning_rate": 6.182866334636889e-07,
-      "loss": 0.2995,
+      "epoch": 4.01,
+      "grad_norm": 2.5480857730573305,
+      "learning_rate": 5.712297012849826e-07,
+      "loss": 0.2349,
       "step": 2295
     },
     {
-      "epoch": 3.98,
-      "grad_norm": 1.908811432566092,
-      "learning_rate": 6.083807786071111e-07,
-      "loss": 0.3222,
+      "epoch": 4.02,
+      "grad_norm": 2.2011961310601276,
+      "learning_rate": 5.615592735570563e-07,
+      "loss": 0.2555,
       "step": 2300
     },
     {
-      "epoch": 3.99,
-      "grad_norm": 1.7539407257874886,
-      "learning_rate": 5.985439145929712e-07,
-      "loss": 0.3106,
+      "epoch": 4.03,
+      "grad_norm": 2.1457420228511497,
+      "learning_rate": 5.519610352980501e-07,
+      "loss": 0.2208,
       "step": 2305
     },
     {
-      "epoch": 4.0,
-      "grad_norm": 1.765533340506495,
-      "learning_rate": 5.887764001898441e-07,
-      "loss": 0.3003,
+      "epoch": 4.04,
+      "grad_norm": 1.989514783963616,
+      "learning_rate": 5.424353439559446e-07,
+      "loss": 0.24,
       "step": 2310
     },
     {
-      "epoch": 4.0,
-      "eval_loss": 0.37727871537208557,
-      "eval_runtime": 330.5667,
-      "eval_samples_per_second": 22.752,
-      "eval_steps_per_second": 0.357,
-      "step": 2312
-    },
-    {
-      "epoch": 4.01,
-      "grad_norm": 2.6249172724373158,
-      "learning_rate": 5.790785916369987e-07,
-      "loss": 0.2687,
+      "epoch": 4.05,
+      "grad_norm": 5.657828438864317,
+      "learning_rate": 5.32982554277e-07,
+      "loss": 0.2344,
       "step": 2315
     },
     {
-      "epoch": 4.01,
-      "grad_norm": 2.0234365981193108,
-      "learning_rate": 5.694508426314024e-07,
-      "loss": 0.2614,
+      "epoch": 4.06,
+      "grad_norm": 2.1124182295185783,
+      "learning_rate": 5.236030182925475e-07,
+      "loss": 0.2332,
       "step": 2320
     },
     {
-      "epoch": 4.02,
-      "grad_norm": 2.0671265707120643,
-      "learning_rate": 5.598935043148216e-07,
-      "loss": 0.2575,
+      "epoch": 4.06,
+      "grad_norm": 2.1054589415417797,
+      "learning_rate": 5.142970853058743e-07,
+      "loss": 0.2264,
       "step": 2325
     },
     {
-      "epoch": 4.03,
-      "grad_norm": 2.122780230514674,
-      "learning_rate": 5.504069252610148e-07,
-      "loss": 0.2634,
+      "epoch": 4.07,
+      "grad_norm": 2.1186956472831557,
+      "learning_rate": 5.050651018792213e-07,
+      "loss": 0.2331,
       "step": 2330
     },
     {
-      "epoch": 4.04,
-      "grad_norm": 1.943804613668437,
-      "learning_rate": 5.409914514630205e-07,
-      "loss": 0.2682,
+      "epoch": 4.08,
+      "grad_norm": 1.9513447351491608,
+      "learning_rate": 4.959074118208726e-07,
+      "loss": 0.2243,
       "step": 2335
     },
     {
-      "epoch": 4.05,
-      "grad_norm": 2.025428441019771,
-      "learning_rate": 5.316474263205352e-07,
-      "loss": 0.2499,
+      "epoch": 4.09,
+      "grad_norm": 2.020062184794575,
+      "learning_rate": 4.868243561723535e-07,
+      "loss": 0.2467,
       "step": 2340
     },
     {
-      "epoch": 4.06,
-      "grad_norm": 1.8116891241805655,
-      "learning_rate": 5.223751906273936e-07,
-      "loss": 0.2587,
+      "epoch": 4.1,
+      "grad_norm": 2.180369971274554,
+      "learning_rate": 4.7781627319573e-07,
+      "loss": 0.2396,
       "step": 2345
     },
     {
-      "epoch": 4.07,
-      "grad_norm": 1.9094711728143186,
-      "learning_rate": 5.131750825591353e-07,
-      "loss": 0.2624,
+      "epoch": 4.11,
+      "grad_norm": 1.9826937910162328,
+      "learning_rate": 4.688834983610083e-07,
+      "loss": 0.23,
       "step": 2350
     },
     {
-      "epoch": 4.07,
-      "grad_norm": 1.792419723948987,
-      "learning_rate": 5.04047437660673e-07,
-      "loss": 0.2596,
+      "epoch": 4.12,
+      "grad_norm": 1.9374964597970792,
+      "learning_rate": 4.6002636433364836e-07,
+      "loss": 0.233,
       "step": 2355
     },
     {
-      "epoch": 4.08,
-      "grad_norm": 1.8856130920393068,
-      "learning_rate": 4.94992588834054e-07,
-      "loss": 0.2696,
+      "epoch": 4.13,
+      "grad_norm": 2.2161679673685364,
+      "learning_rate": 4.512452009621665e-07,
+      "loss": 0.2355,
       "step": 2360
     },
     {
-      "epoch": 4.09,
-      "grad_norm": 4.886142600320586,
-      "learning_rate": 4.860108663263185e-07,
-      "loss": 0.279,
+      "epoch": 4.13,
+      "grad_norm": 2.0656182008910258,
+      "learning_rate": 4.4254033526585917e-07,
+      "loss": 0.2298,
       "step": 2365
     },
     {
-      "epoch": 4.1,
-      "grad_norm": 1.831554357960856,
-      "learning_rate": 4.77102597717454e-07,
-      "loss": 0.2627,
+      "epoch": 4.14,
+      "grad_norm": 1.9301077360385739,
+      "learning_rate": 4.3391209142261996e-07,
+      "loss": 0.2244,
       "step": 2370
     },
     {
-      "epoch": 4.11,
-      "grad_norm": 1.8619320299708513,
-      "learning_rate": 4.682681079084492e-07,
-      "loss": 0.2492,
+      "epoch": 4.15,
+      "grad_norm": 2.019772633601012,
+      "learning_rate": 4.2536079075686687e-07,
+      "loss": 0.239,
       "step": 2375
     },
     {
-      "epoch": 4.12,
-      "grad_norm": 2.026722052142745,
-      "learning_rate": 4.5950771910944603e-07,
-      "loss": 0.2795,
+      "epoch": 4.16,
+      "grad_norm": 1.9236814219014258,
+      "learning_rate": 4.168867517275807e-07,
+      "loss": 0.2294,
       "step": 2380
     },
     {
-      "epoch": 4.13,
-      "grad_norm": 1.8283283261223273,
-      "learning_rate": 4.5082175082798244e-07,
-      "loss": 0.2513,
+      "epoch": 4.17,
+      "grad_norm": 1.9496838165175443,
+      "learning_rate": 4.0849028991643726e-07,
+      "loss": 0.2351,
       "step": 2385
     },
     {
-      "epoch": 4.13,
-      "grad_norm": 1.959431295136227,
-      "learning_rate": 4.422105198573451e-07,
-      "loss": 0.2649,
+      "epoch": 4.18,
+      "grad_norm": 2.0751662501448687,
+      "learning_rate": 4.0017171801606245e-07,
+      "loss": 0.2328,
       "step": 2390
     },
     {
-      "epoch": 4.14,
-      "grad_norm": 1.7723428484005033,
-      "learning_rate": 4.3367434026501317e-07,
-      "loss": 0.2555,
+      "epoch": 4.19,
+      "grad_norm": 2.030778712243648,
+      "learning_rate": 3.9193134581838375e-07,
+      "loss": 0.2193,
       "step": 2395
     },
     {
-      "epoch": 4.15,
-      "grad_norm": 1.978790482643724,
-      "learning_rate": 4.2521352338120157e-07,
-      "loss": 0.2611,
+      "epoch": 4.2,
+      "grad_norm": 2.1108537998820696,
+      "learning_rate": 3.8376948020309083e-07,
+      "loss": 0.2259,
       "step": 2400
     },
     {
-      "epoch": 4.16,
-      "grad_norm": 1.8319604787059154,
-      "learning_rate": 4.1682837778750956e-07,
-      "loss": 0.2525,
+      "epoch": 4.2,
+      "grad_norm": 2.0388571280858003,
+      "learning_rate": 3.756864251262143e-07,
+      "loss": 0.2335,
       "step": 2405
     },
     {
-      "epoch": 4.17,
-      "grad_norm": 2.055092070648486,
-      "learning_rate": 4.085192093056667e-07,
-      "loss": 0.2683,
+      "epoch": 4.21,
+      "grad_norm": 2.4800994906904616,
+      "learning_rate": 3.6768248160879786e-07,
+      "loss": 0.2359,
       "step": 2410
     },
     {
-      "epoch": 4.18,
-      "grad_norm": 1.8816551471726852,
-      "learning_rate": 4.0028632098637335e-07,
-      "loss": 0.2599,
+      "epoch": 4.22,
+      "grad_norm": 2.0463977880911193,
+      "learning_rate": 3.597579477256932e-07,
+      "loss": 0.2296,
       "step": 2415
     },
     {
-      "epoch": 4.19,
-      "grad_norm": 1.8539364404128353,
-      "learning_rate": 3.9213001309825444e-07,
-      "loss": 0.2564,
+      "epoch": 4.23,
+      "grad_norm": 1.980879433572184,
+      "learning_rate": 3.51913118594458e-07,
+      "loss": 0.2244,
       "step": 2420
     },
     {
-      "epoch": 4.2,
-      "grad_norm": 1.802564889277574,
-      "learning_rate": 3.8405058311690473e-07,
-      "loss": 0.2604,
+      "epoch": 4.24,
+      "grad_norm": 2.201271813612607,
+      "learning_rate": 3.4414828636436525e-07,
+      "loss": 0.2355,
       "step": 2425
     },
     {
-      "epoch": 4.2,
-      "grad_norm": 1.921786615001957,
-      "learning_rate": 3.7604832571403823e-07,
-      "loss": 0.2577,
+      "epoch": 4.25,
+      "grad_norm": 1.9358796946633714,
+      "learning_rate": 3.364637402055235e-07,
+      "loss": 0.2235,
       "step": 2430
     },
     {
-      "epoch": 4.21,
-      "grad_norm": 1.8432967570763548,
-      "learning_rate": 3.681235327467453e-07,
-      "loss": 0.2482,
+      "epoch": 4.26,
+      "grad_norm": 1.9943772531112582,
+      "learning_rate": 3.288597662981069e-07,
+      "loss": 0.2363,
       "step": 2435
     },
     {
-      "epoch": 4.22,
-      "grad_norm": 1.8816609521253742,
-      "learning_rate": 3.6027649324684417e-07,
-      "loss": 0.2598,
+      "epoch": 4.27,
+      "grad_norm": 1.9701699768295131,
+      "learning_rate": 3.2133664782169947e-07,
+      "loss": 0.2211,
       "step": 2440
     },
     {
-      "epoch": 4.23,
-      "grad_norm": 1.8554972854976062,
-      "learning_rate": 3.5250749341034007e-07,
-      "loss": 0.2626,
+      "epoch": 4.27,
+      "grad_norm": 1.9846134376402154,
+      "learning_rate": 3.138946649447483e-07,
+      "loss": 0.2421,
       "step": 2445
     },
     {
-      "epoch": 4.24,
-      "grad_norm": 1.7689335842068374,
-      "learning_rate": 3.448168165869886e-07,
-      "loss": 0.2501,
+      "epoch": 4.28,
+      "grad_norm": 2.1267366719554084,
+      "learning_rate": 3.0653409481412906e-07,
+      "loss": 0.232,
       "step": 2450
     },
     {
-      "epoch": 4.25,
-      "grad_norm": 1.9815485309472562,
-      "learning_rate": 3.372047432699607e-07,
-      "loss": 0.2605,
+      "epoch": 4.29,
+      "grad_norm": 2.025895740497588,
+      "learning_rate": 2.9925521154482577e-07,
+      "loss": 0.2327,
       "step": 2455
     },
     {
-      "epoch": 4.26,
-      "grad_norm": 1.8170369399087385,
-      "learning_rate": 3.2967155108561076e-07,
-      "loss": 0.25,
+      "epoch": 4.3,
+      "grad_norm": 2.0801594057057304,
+      "learning_rate": 2.9205828620972267e-07,
+      "loss": 0.229,
       "step": 2460
     },
     {
-      "epoch": 4.26,
-      "grad_norm": 2.4030034838076886,
-      "learning_rate": 3.222175147833556e-07,
-      "loss": 0.2651,
+      "epoch": 4.31,
+      "grad_norm": 2.1905576705856413,
+      "learning_rate": 2.8494358682950725e-07,
+      "loss": 0.2343,
       "step": 2465
     },
     {
-      "epoch": 4.27,
-      "grad_norm": 1.9136750515911383,
-      "learning_rate": 3.148429062256497e-07,
-      "loss": 0.2621,
+      "epoch": 4.32,
+      "grad_norm": 2.104131043625037,
+      "learning_rate": 2.779113783626916e-07,
+      "loss": 0.2324,
       "step": 2470
     },
     {
-      "epoch": 4.28,
-      "grad_norm": 1.91289336281638,
-      "learning_rate": 3.075479943780693e-07,
-      "loss": 0.2608,
+      "epoch": 4.33,
+      "grad_norm": 1.9860664411869802,
+      "learning_rate": 2.70961922695743e-07,
+      "loss": 0.2238,
       "step": 2475
     },
     {
-      "epoch": 4.29,
-      "grad_norm": 1.916237195499518,
-      "learning_rate": 3.0033304529950733e-07,
-      "loss": 0.2591,
+      "epoch": 4.34,
+      "grad_norm": 2.0126712420413324,
+      "learning_rate": 2.6409547863333246e-07,
+      "loss": 0.2203,
       "step": 2480
     },
     {
-      "epoch": 4.3,
-      "grad_norm": 1.7917439220204743,
-      "learning_rate": 2.931983221324664e-07,
-      "loss": 0.2559,
+      "epoch": 4.34,
+      "grad_norm": 1.9772873060488734,
+      "learning_rate": 2.573123018886961e-07,
+      "loss": 0.231,
       "step": 2485
     },
     {
-      "epoch": 4.31,
-      "grad_norm": 1.8928174769958723,
-      "learning_rate": 2.861440850934599e-07,
-      "loss": 0.2554,
+      "epoch": 4.35,
+      "grad_norm": 1.8887328536766113,
+      "learning_rate": 2.5061264507411057e-07,
+      "loss": 0.2135,
       "step": 2490
     },
     {
-      "epoch": 4.32,
-      "grad_norm": 1.8697064920019573,
-      "learning_rate": 2.7917059146352703e-07,
-      "loss": 0.2529,
+      "epoch": 4.36,
+      "grad_norm": 2.1093653326387933,
+      "learning_rate": 2.4399675769148784e-07,
+      "loss": 0.2262,
       "step": 2495
     },
     {
-      "epoch": 4.33,
-      "grad_norm": 1.8149915602515383,
-      "learning_rate": 2.722780955788448e-07,
-      "loss": 0.2507,
+      "epoch": 4.37,
+      "grad_norm": 2.209645499404587,
+      "learning_rate": 2.37464886123083e-07,
+      "loss": 0.2307,
       "step": 2500
     },
     {
-      "epoch": 4.33,
-      "grad_norm": 1.8085679954098952,
-      "learning_rate": 2.6546684882145097e-07,
-      "loss": 0.2692,
+      "epoch": 4.38,
+      "grad_norm": 2.0867824783223208,
+      "learning_rate": 2.3101727362231762e-07,
+      "loss": 0.2363,
       "step": 2505
     },
     {
-      "epoch": 4.34,
-      "grad_norm": 1.8501785801054293,
-      "learning_rate": 2.587370996100802e-07,
-      "loss": 0.2635,
+      "epoch": 4.39,
+      "grad_norm": 2.1576991729700814,
+      "learning_rate": 2.2465416030472227e-07,
+      "loss": 0.2331,
       "step": 2510
     },
     {
-      "epoch": 4.35,
-      "grad_norm": 1.7841943537417788,
-      "learning_rate": 2.520890933911002e-07,
-      "loss": 0.2726,
+      "epoch": 4.4,
+      "grad_norm": 2.0751768584129495,
+      "learning_rate": 2.18375783138991e-07,
+      "loss": 0.2243,
       "step": 2515
     },
     {
-      "epoch": 4.36,
-      "grad_norm": 1.8236899418180625,
-      "learning_rate": 2.4552307262956154e-07,
-      "loss": 0.2629,
+      "epoch": 4.41,
+      "grad_norm": 1.8719170813482977,
+      "learning_rate": 2.1218237593816305e-07,
+      "loss": 0.2366,
       "step": 2520
     },
     {
-      "epoch": 4.37,
-      "grad_norm": 1.8527410868403758,
-      "learning_rate": 2.3903927680035357e-07,
-      "loss": 0.2601,
+      "epoch": 4.41,
+      "grad_norm": 1.9719981089033465,
+      "learning_rate": 2.0607416935090746e-07,
+      "loss": 0.2239,
       "step": 2525
     },
     {
-      "epoch": 4.38,
-      "grad_norm": 1.9912215998190843,
-      "learning_rate": 2.3263794237947084e-07,
-      "loss": 0.2541,
+      "epoch": 4.42,
+      "grad_norm": 2.1100472183339276,
+      "learning_rate": 2.0005139085293945e-07,
+      "loss": 0.2384,
       "step": 2530
     },
     {
-      "epoch": 4.39,
-      "grad_norm": 2.06578491146987,
-      "learning_rate": 2.2631930283538745e-07,
-      "loss": 0.2501,
+      "epoch": 4.43,
+      "grad_norm": 2.194632892084896,
+      "learning_rate": 1.941142647385469e-07,
+      "loss": 0.2311,
       "step": 2535
     },
     {
-      "epoch": 4.39,
-      "grad_norm": 1.8706250122502321,
-      "learning_rate": 2.200835886205438e-07,
-      "loss": 0.2625,
+      "epoch": 4.44,
+      "grad_norm": 2.0340858963446475,
+      "learning_rate": 1.882630121122353e-07,
+      "loss": 0.2358,
       "step": 2540
     },
     {
-      "epoch": 4.4,
-      "grad_norm": 1.9781271983895208,
-      "learning_rate": 2.139310271629405e-07,
-      "loss": 0.2666,
+      "epoch": 4.45,
+      "grad_norm": 1.9791028474343204,
+      "learning_rate": 1.8249785088049894e-07,
+      "loss": 0.2389,
       "step": 2545
     },
     {
-      "epoch": 4.41,
-      "grad_norm": 1.7931386269528835,
-      "learning_rate": 2.07861842857843e-07,
-      "loss": 0.2693,
+      "epoch": 4.46,
+      "grad_norm": 2.2469036964697082,
+      "learning_rate": 1.7681899574369916e-07,
+      "loss": 0.2373,
       "step": 2550
     },
     {
-      "epoch": 4.42,
-      "grad_norm": 1.892205648139564,
-      "learning_rate": 2.0187625705959862e-07,
-      "loss": 0.2624,
+      "epoch": 4.47,
+      "grad_norm": 1.9763859240449058,
+      "learning_rate": 1.7122665818807478e-07,
+      "loss": 0.2242,
       "step": 2555
     },
     {
-      "epoch": 4.43,
-      "grad_norm": 1.8830690882464614,
-      "learning_rate": 1.9597448807356368e-07,
-      "loss": 0.2621,
+      "epoch": 4.48,
+      "grad_norm": 2.2911402959565224,
+      "learning_rate": 1.6572104647786247e-07,
+      "loss": 0.2411,
       "step": 2560
     },
     {
-      "epoch": 4.44,
-      "grad_norm": 2.792218645014677,
-      "learning_rate": 1.901567511481389e-07,
-      "loss": 0.2601,
+      "epoch": 4.48,
+      "grad_norm": 2.083914931198256,
+      "learning_rate": 1.6030236564754054e-07,
+      "loss": 0.2228,
       "step": 2565
     },
     {
-      "epoch": 4.45,
-      "grad_norm": 1.896041115050084,
-      "learning_rate": 1.8442325846692254e-07,
-      "loss": 0.2659,
+      "epoch": 4.49,
+      "grad_norm": 1.9520019226109886,
+      "learning_rate": 1.5497081749419745e-07,
+      "loss": 0.223,
       "step": 2570
     },
     {
-      "epoch": 4.46,
-      "grad_norm": 1.8177273159535672,
-      "learning_rate": 1.787742191409686e-07,
-      "loss": 0.2607,
+      "epoch": 4.5,
+      "grad_norm": 2.000958322889037,
+      "learning_rate": 1.497266005700107e-07,
+      "loss": 0.2366,
       "step": 2575
     },
     {
-      "epoch": 4.46,
-      "grad_norm": 1.8209339547730556,
-      "learning_rate": 1.7320983920116246e-07,
-      "loss": 0.2543,
+      "epoch": 4.51,
+      "grad_norm": 2.255516690447902,
+      "learning_rate": 1.4456991017485737e-07,
+      "loss": 0.2228,
       "step": 2580
     },
     {
-      "epoch": 4.47,
-      "grad_norm": 1.8208127681176174,
-      "learning_rate": 1.6773032159070474e-07,
-      "loss": 0.2559,
+      "epoch": 4.52,
+      "grad_norm": 1.9859817862123228,
+      "learning_rate": 1.3950093834903865e-07,
+      "loss": 0.2277,
       "step": 2585
     },
     {
-      "epoch": 4.48,
-      "grad_norm": 1.8887316977772421,
-      "learning_rate": 1.6233586615771097e-07,
-      "loss": 0.2675,
+      "epoch": 4.53,
+      "grad_norm": 1.9478379602583347,
+      "learning_rate": 1.3451987386612852e-07,
+      "loss": 0.2303,
       "step": 2590
     },
     {
-      "epoch": 4.49,
-      "grad_norm": 1.8402016996115687,
-      "learning_rate": 1.570266696479214e-07,
-      "loss": 0.2708,
+      "epoch": 4.54,
+      "grad_norm": 2.0551925040719135,
+      "learning_rate": 1.2962690222594444e-07,
+      "loss": 0.2213,
       "step": 2595
     },
     {
-      "epoch": 4.5,
-      "grad_norm": 1.8135826560339299,
-      "learning_rate": 1.5180292569752675e-07,
-      "loss": 0.2663,
+      "epoch": 4.55,
+      "grad_norm": 1.9946612862734567,
+      "learning_rate": 1.2482220564763669e-07,
+      "loss": 0.2332,
       "step": 2600
     },
     {
-      "epoch": 4.51,
-      "grad_norm": 1.8932652503035179,
-      "learning_rate": 1.4666482482610517e-07,
-      "loss": 0.2562,
+      "epoch": 4.55,
+      "grad_norm": 2.013120854912158,
+      "learning_rate": 1.2010596306290588e-07,
+      "loss": 0.2171,
       "step": 2605
     },
     {
-      "epoch": 4.52,
-      "grad_norm": 1.923411155855795,
-      "learning_rate": 1.4161255442967314e-07,
-      "loss": 0.2535,
+      "epoch": 4.56,
+      "grad_norm": 3.943001009546542,
+      "learning_rate": 1.154783501093365e-07,
+      "loss": 0.2351,
       "step": 2610
     },
     {
-      "epoch": 4.52,
-      "grad_norm": 1.8186664689085035,
-      "learning_rate": 1.3664629877385233e-07,
-      "loss": 0.2536,
+      "epoch": 4.57,
+      "grad_norm": 2.077111667283586,
+      "learning_rate": 1.1093953912385769e-07,
+      "loss": 0.2227,
       "step": 2615
     },
     {
-      "epoch": 4.53,
-      "grad_norm": 1.8324657945327985,
-      "learning_rate": 1.317662389871477e-07,
-      "loss": 0.2626,
+      "epoch": 4.58,
+      "grad_norm": 1.9319320014149188,
+      "learning_rate": 1.0648969913632401e-07,
+      "loss": 0.2349,
       "step": 2620
     },
     {
-      "epoch": 4.54,
-      "grad_norm": 1.825266320838444,
-      "learning_rate": 1.2697255305434087e-07,
-      "loss": 0.2669,
+      "epoch": 4.59,
+      "grad_norm": 2.0904879562505125,
+      "learning_rate": 1.0212899586322183e-07,
+      "loss": 0.2353,
       "step": 2625
     },
     {
-      "epoch": 4.55,
-      "grad_norm": 1.8907649399842439,
-      "learning_rate": 1.22265415810002e-07,
-      "loss": 0.2637,
+      "epoch": 4.6,
+      "grad_norm": 1.9353613819189108,
+      "learning_rate": 9.785759170149622e-08,
+      "loss": 0.2207,
       "step": 2630
     },
     {
-      "epoch": 4.56,
-      "grad_norm": 1.8689930525310243,
-      "learning_rate": 1.1764499893210879e-07,
-      "loss": 0.2625,
+      "epoch": 4.61,
+      "grad_norm": 2.070941108809218,
+      "learning_rate": 9.36756457225052e-08,
+      "loss": 0.2333,
       "step": 2635
     },
     {
-      "epoch": 4.57,
-      "grad_norm": 1.7851298479073345,
-      "learning_rate": 1.1311147093578778e-07,
-      "loss": 0.2515,
+      "epoch": 4.62,
+      "grad_norm": 1.962848104620626,
+      "learning_rate": 8.958331366609424e-08,
+      "loss": 0.2268,
       "step": 2640
     },
     {
-      "epoch": 4.58,
-      "grad_norm": 1.9022961798833327,
-      "learning_rate": 1.0866499716716855e-07,
-      "loss": 0.2683,
+      "epoch": 4.62,
+      "grad_norm": 2.0575315291559857,
+      "learning_rate": 8.55807479347967e-08,
+      "loss": 0.2253,
       "step": 2645
     },
     {
-      "epoch": 4.58,
-      "grad_norm": 1.79257289914677,
-      "learning_rate": 1.0430573979735154e-07,
-      "loss": 0.2654,
+      "epoch": 4.63,
+      "grad_norm": 2.107360332698912,
+      "learning_rate": 8.166809758815897e-08,
+      "loss": 0.224,
       "step": 2650
     },
     {
-      "epoch": 4.59,
-      "grad_norm": 1.9177553807521017,
-      "learning_rate": 1.000338578164936e-07,
-      "loss": 0.2607,
+      "epoch": 4.64,
+      "grad_norm": 2.144090062153445,
+      "learning_rate": 7.784550833718707e-08,
+      "loss": 0.2365,
       "step": 2655
     },
     {
-      "epoch": 4.6,
-      "grad_norm": 1.9212642014661736,
-      "learning_rate": 9.584950702801244e-08,
-      "loss": 0.2748,
+      "epoch": 4.65,
+      "grad_norm": 1.9482188832166278,
+      "learning_rate": 7.411312253892466e-08,
+      "loss": 0.2294,
       "step": 2660
     },
     {
-      "epoch": 4.61,
-      "grad_norm": 1.8444742116297173,
-      "learning_rate": 9.175284004289853e-08,
-      "loss": 0.2695,
+      "epoch": 4.66,
+      "grad_norm": 2.1524181670324616,
+      "learning_rate": 7.047107919114588e-08,
+      "loss": 0.2336,
       "step": 2665
     },
     {
-      "epoch": 4.62,
-      "grad_norm": 1.879187416864439,
-      "learning_rate": 8.774400627415498e-08,
-      "loss": 0.2648,
+      "epoch": 4.67,
+      "grad_norm": 2.152296301431861,
+      "learning_rate": 6.691951392718332e-08,
+      "loss": 0.2258,
       "step": 2670
     },
     {
-      "epoch": 4.63,
-      "grad_norm": 1.853575218259344,
-      "learning_rate": 8.382315193134372e-08,
-      "loss": 0.2531,
+      "epoch": 4.68,
+      "grad_norm": 1.9081910207653139,
+      "learning_rate": 6.345855901087522e-08,
+      "loss": 0.2338,
       "step": 2675
     },
     {
-      "epoch": 4.64,
-      "grad_norm": 1.8274985226049987,
-      "learning_rate": 7.999042001525598e-08,
-      "loss": 0.2608,
+      "epoch": 4.69,
+      "grad_norm": 1.938526307346203,
+      "learning_rate": 6.008834333163876e-08,
+      "loss": 0.2195,
       "step": 2680
     },
     {
-      "epoch": 4.65,
-      "grad_norm": 1.9822786870303424,
-      "learning_rate": 7.624595031269388e-08,
-      "loss": 0.2739,
+      "epoch": 4.69,
+      "grad_norm": 1.9617786271701327,
+      "learning_rate": 5.680899239967369e-08,
+      "loss": 0.2337,
       "step": 2685
     },
     {
-      "epoch": 4.65,
-      "grad_norm": 2.395868204461241,
-      "learning_rate": 7.258987939137596e-08,
-      "loss": 0.2608,
+      "epoch": 4.7,
+      "grad_norm": 2.070106598519499,
+      "learning_rate": 5.3620628341283234e-08,
+      "loss": 0.2339,
       "step": 2690
     },
     {
-      "epoch": 4.66,
-      "grad_norm": 1.9124430429661154,
-      "learning_rate": 6.902234059495222e-08,
-      "loss": 0.2621,
+      "epoch": 4.71,
+      "grad_norm": 2.063857678666163,
+      "learning_rate": 5.052336989433082e-08,
+      "loss": 0.2424,
       "step": 2695
     },
     {
-      "epoch": 4.67,
-      "grad_norm": 1.896035807781804,
-      "learning_rate": 6.554346403814387e-08,
-      "loss": 0.2604,
+      "epoch": 4.72,
+      "grad_norm": 1.9748870315279254,
+      "learning_rate": 4.75173324038139e-08,
+      "loss": 0.2284,
       "step": 2700
     },
     {
-      "epoch": 4.68,
-      "grad_norm": 1.881027284048797,
-      "learning_rate": 6.215337660199632e-08,
-      "loss": 0.2457,
+      "epoch": 4.73,
+      "grad_norm": 2.001809998998157,
+      "learning_rate": 4.4602627817571245e-08,
+      "loss": 0.2358,
       "step": 2705
     },
     {
-      "epoch": 4.69,
-      "grad_norm": 1.9007017472534657,
-      "learning_rate": 5.8852201929252005e-08,
-      "loss": 0.254,
+      "epoch": 4.74,
+      "grad_norm": 1.9729922140795935,
+      "learning_rate": 4.1779364682113796e-08,
+      "loss": 0.233,
       "step": 2710
     },
     {
-      "epoch": 4.7,
-      "grad_norm": 1.8930962106170162,
-      "learning_rate": 5.5640060419840655e-08,
-      "loss": 0.264,
+      "epoch": 4.75,
+      "grad_norm": 2.045685009725174,
+      "learning_rate": 3.904764813858014e-08,
+      "loss": 0.2323,
       "step": 2715
     },
     {
-      "epoch": 4.71,
-      "grad_norm": 1.7660321817481,
-      "learning_rate": 5.2517069226488694e-08,
-      "loss": 0.2845,
+      "epoch": 4.76,
+      "grad_norm": 1.9591449865359234,
+      "learning_rate": 3.640757991882349e-08,
+      "loss": 0.2349,
       "step": 2720
     },
     {
-      "epoch": 4.71,
-      "grad_norm": 1.794271544889709,
-      "learning_rate": 4.9483342250445934e-08,
-      "loss": 0.2657,
+      "epoch": 4.76,
+      "grad_norm": 1.9911019888259411,
+      "learning_rate": 3.385925834162113e-08,
+      "loss": 0.2257,
       "step": 2725
     },
     {
-      "epoch": 4.72,
-      "grad_norm": 1.8831928024041047,
-      "learning_rate": 4.653899013733115e-08,
-      "loss": 0.2622,
+      "epoch": 4.77,
+      "grad_norm": 2.050714072028577,
+      "learning_rate": 3.1402778309014284e-08,
+      "loss": 0.2299,
       "step": 2730
     },
     {
-      "epoch": 4.73,
-      "grad_norm": 1.8254137007243867,
-      "learning_rate": 4.368412027309754e-08,
-      "loss": 0.2613,
+      "epoch": 4.78,
+      "grad_norm": 2.0356703719044518,
+      "learning_rate": 2.903823130277289e-08,
+      "loss": 0.2277,
       "step": 2735
     },
     {
-      "epoch": 4.74,
-      "grad_norm": 1.921264151045264,
-      "learning_rate": 4.0918836780115265e-08,
-      "loss": 0.2479,
+      "epoch": 4.79,
+      "grad_norm": 1.9823356155432006,
+      "learning_rate": 2.676570538098944e-08,
+      "loss": 0.2431,
       "step": 2740
     },
     {
-      "epoch": 4.75,
-      "grad_norm": 1.8886495539506756,
-      "learning_rate": 3.8243240513374524e-08,
-      "loss": 0.2554,
+      "epoch": 4.8,
+      "grad_norm": 2.0568572715035063,
+      "learning_rate": 2.4585285174799357e-08,
+      "loss": 0.2331,
       "step": 2745
     },
     {
-      "epoch": 4.76,
-      "grad_norm": 1.842345668470861,
-      "learning_rate": 3.5657429056806523e-08,
-      "loss": 0.261,
+      "epoch": 4.81,
+      "grad_norm": 2.195948388536599,
+      "learning_rate": 2.2497051885228825e-08,
+      "loss": 0.2308,
       "step": 2750
     },
     {
-      "epoch": 4.77,
-      "grad_norm": 1.8470545043429003,
-      "learning_rate": 3.316149671972552e-08,
-      "loss": 0.2451,
+      "epoch": 4.82,
+      "grad_norm": 2.1647004516523523,
+      "learning_rate": 2.050108328017164e-08,
+      "loss": 0.2418,
       "step": 2755
     },
     {
-      "epoch": 4.78,
-      "grad_norm": 1.898197482257326,
-      "learning_rate": 3.0755534533387365e-08,
-      "loss": 0.2605,
+      "epoch": 4.83,
+      "grad_norm": 2.3975608426150656,
+      "learning_rate": 1.8597453691492628e-08,
+      "loss": 0.2318,
       "step": 2760
     },
-    {
-      "epoch": 4.78,
-      "grad_norm": 1.7715397317439983,
-      "learning_rate": 2.8439630247671934e-08,
-      "loss": 0.247,
-      "step": 2765
-    },
-    {
-      "epoch": 4.79,
-      "grad_norm": 1.8519562470145452,
-      "learning_rate": 2.6213868327880088e-08,
-      "loss": 0.2736,
-      "step": 2770
-    },
-    {
-      "epoch": 4.8,
-      "grad_norm": 1.9175537155386098,
-      "learning_rate": 2.4078329951655044e-08,
-      "loss": 0.2617,
-      "step": 2775
-    },
-    {
-      "epoch": 4.81,
-      "grad_norm": 1.8142907346054091,
-      "learning_rate": 2.2033093006021133e-08,
-      "loss": 0.2653,
-      "step": 2780
-    },
-    {
-      "epoch": 4.82,
-      "grad_norm": 1.7543346418920545,
-      "learning_rate": 2.007823208454218e-08,
-      "loss": 0.2478,
-      "step": 2785
-    },
     {
       "epoch": 4.83,
-      "grad_norm": 1.817163533821061,
-      "learning_rate": 1.821381848460202e-08,
-      "loss": 0.2513,
-      "step": 2790
-    },
-    {
-      "epoch": 4.84,
-      "grad_norm": 1.9039285159704047,
-      "learning_rate": 1.643992020480406e-08,
-      "loss": 0.257,
-      "step": 2795
+      "grad_norm": 2.0876506445205716,
+      "learning_rate": 1.678623401225876e-08,
+      "loss": 0.2227,
+      "step": 2765
     },
     {
       "epoch": 4.84,
-      "grad_norm": 1.8371177889007853,
-      "learning_rate": 1.4756601942490789e-08,
-      "loss": 0.2497,
-      "step": 2800
+      "grad_norm": 1.9235370393907136,
+      "learning_rate": 1.5067491694100156e-08,
+      "loss": 0.2317,
+      "step": 2770
     },
     {
       "epoch": 4.85,
-      "grad_norm": 1.8366707128631479,
-      "learning_rate": 1.3163925091384532e-08,
-      "loss": 0.2531,
-      "step": 2805
+      "grad_norm": 2.166013576809247,
+      "learning_rate": 1.3441290744697632e-08,
+      "loss": 0.2281,
+      "step": 2775
     },
     {
       "epoch": 4.86,
-      "grad_norm": 2.363794421622485,
-      "learning_rate": 1.1661947739347867e-08,
-      "loss": 0.2682,
-      "step": 2810
+      "grad_norm": 1.9685518937742286,
+      "learning_rate": 1.1907691725398795e-08,
+      "loss": 0.2238,
+      "step": 2780
     },
     {
       "epoch": 4.87,
-      "grad_norm": 1.820365380182807,
-      "learning_rate": 1.0250724666265588e-08,
-      "loss": 0.265,
-      "step": 2815
+      "grad_norm": 1.9483454828149802,
+      "learning_rate": 1.0466751748963443e-08,
+      "loss": 0.2304,
+      "step": 2785
     },
     {
       "epoch": 4.88,
-      "grad_norm": 1.8970735871610287,
-      "learning_rate": 8.930307342046307e-09,
-      "loss": 0.2625,
-      "step": 2820
+      "grad_norm": 1.9753685457102212,
+      "learning_rate": 9.118524477434999e-09,
+      "loss": 0.2365,
+      "step": 2790
     },
     {
       "epoch": 4.89,
-      "grad_norm": 1.9794621227492666,
-      "learning_rate": 7.700743924745902e-09,
-      "loss": 0.2743,
-      "step": 2825
+      "grad_norm": 1.8854947658053076,
+      "learning_rate": 7.863060120144316e-09,
+      "loss": 0.2142,
+      "step": 2795
     },
     {
       "epoch": 4.9,
-      "grad_norm": 2.1030207809550574,
-      "learning_rate": 6.5620792588097505e-09,
-      "loss": 0.2625,
-      "step": 2830
+      "grad_norm": 2.154842589588192,
+      "learning_rate": 6.700405431837587e-09,
+      "loss": 0.2315,
+      "step": 2800
     },
     {
       "epoch": 4.9,
-      "grad_norm": 1.8031485505651268,
-      "learning_rate": 5.51435487343932e-09,
-      "loss": 0.2589,
-      "step": 2835
+      "grad_norm": 2.0037090722639506,
+      "learning_rate": 5.6306037109371544e-09,
+      "loss": 0.2225,
+      "step": 2805
     },
     {
       "epoch": 4.91,
-      "grad_norm": 1.862632623834731,
-      "learning_rate": 4.557608981075323e-09,
-      "loss": 0.2651,
-      "step": 2840
+      "grad_norm": 1.9552444706301575,
+      "learning_rate": 4.653694797927544e-09,
+      "loss": 0.2331,
+      "step": 2810
     },
     {
       "epoch": 4.92,
-      "grad_norm": 1.796005099020042,
-      "learning_rate": 3.691876476005496e-09,
-      "loss": 0.2537,
-      "step": 2845
+      "grad_norm": 2.563219810008139,
+      "learning_rate": 3.769715073872749e-09,
+      "loss": 0.2407,
+      "step": 2815
     },
     {
       "epoch": 4.93,
-      "grad_norm": 1.7492013741327732,
-      "learning_rate": 2.917188933091175e-09,
-      "loss": 0.2461,
-      "step": 2850
+      "grad_norm": 1.9200483422222523,
+      "learning_rate": 2.978697459060098e-09,
+      "loss": 0.23,
+      "step": 2820
     },
     {
       "epoch": 4.94,
-      "grad_norm": 1.9157963617346956,
-      "learning_rate": 2.233574606616273e-09,
-      "loss": 0.2661,
-      "step": 2855
+      "grad_norm": 1.9640394162849764,
+      "learning_rate": 2.280671411776514e-09,
+      "loss": 0.2264,
+      "step": 2825
     },
     {
       "epoch": 4.95,
-      "grad_norm": 1.8994803862391993,
-      "learning_rate": 1.6410584292561592e-09,
-      "loss": 0.2665,
-      "step": 2860
+      "grad_norm": 2.1505121388781467,
+      "learning_rate": 1.6756629272085545e-09,
+      "loss": 0.2278,
+      "step": 2830
     },
     {
       "epoch": 4.96,
-      "grad_norm": 1.846063403180044,
-      "learning_rate": 1.139662011168663e-09,
-      "loss": 0.2503,
-      "step": 2865
+      "grad_norm": 1.9911030010645636,
+      "learning_rate": 1.1636945364768005e-09,
+      "loss": 0.2341,
+      "step": 2835
     },
     {
       "epoch": 4.97,
-      "grad_norm": 1.8871406230680454,
-      "learning_rate": 7.294036392066495e-10,
-      "loss": 0.2601,
-      "step": 2870
+      "grad_norm": 2.0018264068324187,
+      "learning_rate": 7.447853057954146e-10,
+      "loss": 0.2185,
+      "step": 2840
     },
     {
       "epoch": 4.97,
-      "grad_norm": 1.7693402402676852,
-      "learning_rate": 4.1029827624911033e-10,
-      "loss": 0.2494,
-      "step": 2875
+      "grad_norm": 1.9862221260079735,
+      "learning_rate": 4.1895083576271035e-10,
+      "loss": 0.221,
+      "step": 2845
     },
     {
       "epoch": 4.98,
-      "grad_norm": 1.8544595991462598,
-      "learning_rate": 1.8235756065770882e-10,
-      "loss": 0.2647,
-      "step": 2880
+      "grad_norm": 1.9962774038489774,
+      "learning_rate": 1.8620326077967155e-10,
+      "loss": 0.2316,
+      "step": 2850
     },
     {
       "epoch": 4.99,
-      "grad_norm": 1.9426687566855905,
-      "learning_rate": 4.5589805850454475e-11,
-      "loss": 0.2615,
-      "step": 2885
+      "grad_norm": 1.9534249748428105,
+      "learning_rate": 4.6551248598647457e-11,
+      "loss": 0.2333,
+      "step": 2855
     },
     {
       "epoch": 5.0,
-      "grad_norm": 1.8892335398265772,
+      "grad_norm": 2.0900193031037806,
       "learning_rate": 0.0,
-      "loss": 0.2508,
-      "step": 2890
+      "loss": 0.229,
+      "step": 2860
     },
     {
       "epoch": 5.0,
-      "eval_loss": 0.375461608171463,
-      "eval_runtime": 329.8865,
-      "eval_samples_per_second": 22.799,
+      "eval_loss": 0.3284313678741455,
+      "eval_runtime": 329.743,
+      "eval_samples_per_second": 22.809,
       "eval_steps_per_second": 0.358,
-      "step": 2890
+      "step": 2860
     },
     {
       "epoch": 5.0,
-      "step": 2890,
-      "total_flos": 2420428819660800.0,
-      "train_loss": 0.44474206024387714,
-      "train_runtime": 64422.034,
-      "train_samples_per_second": 5.737,
+      "step": 2860,
+      "total_flos": 2395303260979200.0,
+      "train_loss": 0.44255039914921446,
+      "train_runtime": 63778.4258,
+      "train_samples_per_second": 5.733,
       "train_steps_per_second": 0.045
     }
   ],
   "logging_steps": 5,
-  "max_steps": 2890,
+  "max_steps": 2860,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 5,
   "save_steps": 500,
-  "total_flos": 2420428819660800.0,
+  "total_flos": 2395303260979200.0,
   "train_batch_size": 16,
   "trial_name": null,
   "trial_params": null