diff --git "a/last-checkpoint/trainer_state.json" "b/last-checkpoint/trainer_state.json"
--- "a/last-checkpoint/trainer_state.json"
+++ "b/last-checkpoint/trainer_state.json"
@@ -2,2112 +2,3372 @@
   "best_global_step": null,
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 1.8717628705148206,
+  "epoch": 2.994695787831513,
   "eval_steps": 500,
-  "global_step": 3000,
+  "global_step": 4800,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.0062402496099844,
-      "grad_norm": 18.16543197631836,
+      "grad_norm": 26.80913734436035,
       "learning_rate": 9e-05,
-      "loss": 2.5416,
+      "loss": 2.5572,
       "step": 10
     },
     {
       "epoch": 0.0124804992199688,
-      "grad_norm": 4.823841094970703,
+      "grad_norm": 5.056921482086182,
       "learning_rate": 0.00019,
-      "loss": 0.2597,
+      "loss": 0.274,
       "step": 20
     },
     {
       "epoch": 0.0187207488299532,
-      "grad_norm": 2.6858906745910645,
+      "grad_norm": 2.284580945968628,
       "learning_rate": 0.00019999825713466112,
-      "loss": 0.1064,
+      "loss": 0.117,
       "step": 30
     },
     {
       "epoch": 0.0249609984399376,
-      "grad_norm": 1.4995951652526855,
+      "grad_norm": 0.7552870512008667,
       "learning_rate": 0.00019999223249296698,
-      "loss": 0.074,
+      "loss": 0.0607,
       "step": 40
     },
     {
       "epoch": 0.031201248049921998,
-      "grad_norm": 4.664715766906738,
+      "grad_norm": 1.2389109134674072,
       "learning_rate": 0.00019998190481726233,
-      "loss": 0.0804,
+      "loss": 0.0508,
       "step": 50
     },
     {
       "epoch": 0.0374414976599064,
-      "grad_norm": 0.7267784476280212,
+      "grad_norm": 1.2077319622039795,
       "learning_rate": 0.00019996727455198513,
-      "loss": 0.0427,
+      "loss": 0.0437,
       "step": 60
     },
     {
       "epoch": 0.0436817472698908,
-      "grad_norm": 0.29998767375946045,
+      "grad_norm": 0.7111845016479492,
       "learning_rate": 0.00019994834232672947,
-      "loss": 0.0267,
+      "loss": 0.0365,
       "step": 70
     },
     {
       "epoch": 0.0499219968798752,
-      "grad_norm": 0.43115338683128357,
+      "grad_norm": 0.47715747356414795,
       "learning_rate": 0.00019992510895621885,
-      "loss": 0.0193,
+      "loss": 0.0262,
       "step": 80
     },
     {
       "epoch": 0.056162246489859596,
-      "grad_norm": 0.339687705039978,
+      "grad_norm": 3.2088232040405273,
       "learning_rate": 0.00019989757544027072,
-      "loss": 0.0324,
+      "loss": 0.0298,
       "step": 90
     },
     {
       "epoch": 0.062402496099843996,
-      "grad_norm": 0.648891806602478,
+      "grad_norm": 3.7542457580566406,
       "learning_rate": 0.00019986574296375364,
-      "loss": 0.0339,
+      "loss": 0.0347,
       "step": 100
     },
     {
       "epoch": 0.0686427457098284,
-      "grad_norm": 1.3054920434951782,
+      "grad_norm": 0.11264568567276001,
       "learning_rate": 0.0001998296128965363,
-      "loss": 0.0243,
+      "loss": 0.0155,
       "step": 110
     },
     {
       "epoch": 0.0748829953198128,
-      "grad_norm": 1.6219518184661865,
+      "grad_norm": 2.73014760017395,
       "learning_rate": 0.00019978918679342865,
-      "loss": 0.0357,
+      "loss": 0.0184,
       "step": 120
     },
     {
       "epoch": 0.0811232449297972,
-      "grad_norm": 0.8563578128814697,
+      "grad_norm": 1.8296796083450317,
       "learning_rate": 0.00019974446639411464,
-      "loss": 0.0224,
+      "loss": 0.0462,
       "step": 130
     },
     {
       "epoch": 0.0873634945397816,
-      "grad_norm": 0.16929180920124054,
+      "grad_norm": 1.1681585311889648,
       "learning_rate": 0.00019969545362307785,
-      "loss": 0.0115,
+      "loss": 0.0238,
       "step": 140
     },
     {
       "epoch": 0.093603744149766,
-      "grad_norm": 1.1288150548934937,
+      "grad_norm": 0.7835023403167725,
       "learning_rate": 0.0001996421505895183,
-      "loss": 0.0134,
+      "loss": 0.0191,
       "step": 150
     },
     {
       "epoch": 0.0998439937597504,
-      "grad_norm": 0.6310799717903137,
+      "grad_norm": 1.4719984531402588,
       "learning_rate": 0.0001995845595872618,
-      "loss": 0.0242,
+      "loss": 0.024,
       "step": 160
     },
     {
       "epoch": 0.1060842433697348,
-      "grad_norm": 1.2807273864746094,
+      "grad_norm": 0.526832103729248,
       "learning_rate": 0.0001995226830946613,
-      "loss": 0.0163,
+      "loss": 0.0168,
       "step": 170
     },
     {
       "epoch": 0.11232449297971919,
-      "grad_norm": 0.5181924700737,
+      "grad_norm": 0.3308654725551605,
       "learning_rate": 0.00019945652377449012,
-      "loss": 0.0314,
+      "loss": 0.0187,
       "step": 180
     },
     {
       "epoch": 0.11856474258970359,
-      "grad_norm": 2.7124929428100586,
+      "grad_norm": 0.7355976104736328,
       "learning_rate": 0.0001993860844738275,
-      "loss": 0.0241,
+      "loss": 0.0148,
       "step": 190
     },
     {
       "epoch": 0.12480499219968799,
-      "grad_norm": 0.7487115859985352,
+      "grad_norm": 2.0277516841888428,
       "learning_rate": 0.0001993113682239359,
-      "loss": 0.0206,
+      "loss": 0.0335,
       "step": 200
     },
     {
       "epoch": 0.1310452418096724,
-      "grad_norm": 0.9230497479438782,
+      "grad_norm": 1.4253898859024048,
       "learning_rate": 0.00019923237824013076,
-      "loss": 0.011,
+      "loss": 0.0267,
       "step": 210
     },
     {
       "epoch": 0.1372854914196568,
-      "grad_norm": 4.362138271331787,
+      "grad_norm": 2.561232328414917,
       "learning_rate": 0.000199149117921642,
-      "loss": 0.0182,
+      "loss": 0.0259,
       "step": 220
     },
     {
       "epoch": 0.1435257410296412,
-      "grad_norm": 2.675947904586792,
+      "grad_norm": 0.4792596399784088,
       "learning_rate": 0.00019906159085146768,
-      "loss": 0.0205,
+      "loss": 0.0177,
       "step": 230
     },
     {
       "epoch": 0.1497659906396256,
-      "grad_norm": 0.9578750729560852,
+      "grad_norm": 0.43000543117523193,
       "learning_rate": 0.00019896980079622,
-      "loss": 0.0142,
+      "loss": 0.0085,
       "step": 240
     },
     {
       "epoch": 0.15600624024961,
-      "grad_norm": 0.5006669759750366,
+      "grad_norm": 0.7919816374778748,
       "learning_rate": 0.00019887375170596312,
-      "loss": 0.0296,
+      "loss": 0.0154,
       "step": 250
     },
     {
       "epoch": 0.1622464898595944,
-      "grad_norm": 1.6358978748321533,
+      "grad_norm": 0.6787700057029724,
       "learning_rate": 0.00019877344771404307,
-      "loss": 0.0261,
+      "loss": 0.0171,
       "step": 260
     },
     {
       "epoch": 0.1684867394695788,
-      "grad_norm": 5.082851886749268,
+      "grad_norm": 0.015748390927910805,
       "learning_rate": 0.00019866889313691,
-      "loss": 0.0169,
+      "loss": 0.0074,
       "step": 270
     },
     {
       "epoch": 0.1747269890795632,
-      "grad_norm": 0.11479347199201584,
+      "grad_norm": 2.1442184448242188,
       "learning_rate": 0.0001985600924739325,
-      "loss": 0.0229,
+      "loss": 0.0198,
       "step": 280
     },
     {
       "epoch": 0.1809672386895476,
-      "grad_norm": 0.9521530866622925,
+      "grad_norm": 0.4856598973274231,
       "learning_rate": 0.0001984470504072037,
-      "loss": 0.0193,
+      "loss": 0.0148,
       "step": 290
     },
     {
       "epoch": 0.187207488299532,
-      "grad_norm": 2.097628593444824,
+      "grad_norm": 0.5050826072692871,
       "learning_rate": 0.0001983297718013401,
-      "loss": 0.0217,
+      "loss": 0.0158,
       "step": 300
     },
     {
       "epoch": 0.1934477379095164,
-      "grad_norm": 0.75714111328125,
+      "grad_norm": 0.29614710807800293,
       "learning_rate": 0.00019820826170327194,
-      "loss": 0.0176,
+      "loss": 0.0129,
       "step": 310
     },
     {
       "epoch": 0.1996879875195008,
-      "grad_norm": 1.0679302215576172,
+      "grad_norm": 0.34757232666015625,
       "learning_rate": 0.0001980825253420263,
-      "loss": 0.0145,
+      "loss": 0.0138,
       "step": 320
     },
     {
       "epoch": 0.2059282371294852,
-      "grad_norm": 0.7506147623062134,
+      "grad_norm": 0.24157005548477173,
       "learning_rate": 0.0001979525681285018,
-      "loss": 0.0132,
+      "loss": 0.0114,
       "step": 330
     },
     {
       "epoch": 0.2121684867394696,
-      "grad_norm": 0.606416642665863,
+      "grad_norm": 5.074306488037109,
       "learning_rate": 0.000197818395655236,
-      "loss": 0.0231,
+      "loss": 0.028,
       "step": 340
     },
     {
       "epoch": 0.21840873634945399,
-      "grad_norm": 0.4706982672214508,
+      "grad_norm": 0.11908160150051117,
       "learning_rate": 0.0001976800136961645,
-      "loss": 0.0304,
+      "loss": 0.016,
       "step": 350
     },
     {
       "epoch": 0.22464898595943839,
-      "grad_norm": 0.04006557539105415,
+      "grad_norm": 0.06858685612678528,
       "learning_rate": 0.0001975374282063727,
-      "loss": 0.0249,
+      "loss": 0.0189,
       "step": 360
     },
     {
       "epoch": 0.23088923556942278,
-      "grad_norm": 0.1513279676437378,
+      "grad_norm": 0.1790592521429062,
       "learning_rate": 0.0001973906453218392,
-      "loss": 0.0173,
+      "loss": 0.0145,
       "step": 370
     },
     {
       "epoch": 0.23712948517940718,
-      "grad_norm": 0.34040647745132446,
+      "grad_norm": 0.31875643134117126,
       "learning_rate": 0.00019723967135917223,
-      "loss": 0.0121,
+      "loss": 0.0149,
       "step": 380
     },
     {
       "epoch": 0.24336973478939158,
-      "grad_norm": 0.6798901557922363,
+      "grad_norm": 0.09313604235649109,
       "learning_rate": 0.0001970845128153373,
-      "loss": 0.0265,
+      "loss": 0.0288,
       "step": 390
     },
     {
       "epoch": 0.24960998439937598,
-      "grad_norm": 0.1554585099220276,
+      "grad_norm": 0.5442845225334167,
       "learning_rate": 0.000196925176367378,
-      "loss": 0.0105,
+      "loss": 0.0073,
       "step": 400
     },
     {
       "epoch": 0.25585023400936036,
-      "grad_norm": 0.2256622165441513,
+      "grad_norm": 0.025376079604029655,
       "learning_rate": 0.0001967616688721285,
-      "loss": 0.0128,
+      "loss": 0.0158,
       "step": 410
     },
     {
       "epoch": 0.2620904836193448,
-      "grad_norm": 0.2520933151245117,
+      "grad_norm": 1.513388991355896,
       "learning_rate": 0.00019659399736591843,
-      "loss": 0.017,
+      "loss": 0.0199,
       "step": 420
     },
     {
       "epoch": 0.26833073322932915,
-      "grad_norm": 0.6182873249053955,
+      "grad_norm": 0.5486435294151306,
       "learning_rate": 0.00019642216906427026,
-      "loss": 0.0128,
+      "loss": 0.0135,
       "step": 430
     },
     {
       "epoch": 0.2745709828393136,
-      "grad_norm": 0.028039440512657166,
+      "grad_norm": 0.5338599681854248,
       "learning_rate": 0.00019624619136158854,
-      "loss": 0.0155,
+      "loss": 0.0145,
       "step": 440
     },
     {
       "epoch": 0.28081123244929795,
-      "grad_norm": 0.018251167610287666,
+      "grad_norm": 0.9042736887931824,
       "learning_rate": 0.00019606607183084192,
-      "loss": 0.0107,
+      "loss": 0.0089,
       "step": 450
     },
     {
       "epoch": 0.2870514820592824,
-      "grad_norm": 0.1820351779460907,
+      "grad_norm": 1.2041606903076172,
       "learning_rate": 0.00019588181822323717,
-      "loss": 0.0063,
+      "loss": 0.0166,
       "step": 460
     },
     {
       "epoch": 0.29329173166926675,
-      "grad_norm": 0.4565224051475525,
+      "grad_norm": 0.25932419300079346,
       "learning_rate": 0.00019569343846788557,
-      "loss": 0.012,
+      "loss": 0.0178,
       "step": 470
     },
     {
       "epoch": 0.2995319812792512,
-      "grad_norm": 0.01261977944523096,
+      "grad_norm": 0.10857085883617401,
       "learning_rate": 0.00019550094067146174,
-      "loss": 0.0076,
+      "loss": 0.013,
       "step": 480
     },
     {
       "epoch": 0.30577223088923555,
-      "grad_norm": 0.038973432034254074,
+      "grad_norm": 0.9906659126281738,
       "learning_rate": 0.0001953043331178548,
-      "loss": 0.0048,
+      "loss": 0.0108,
       "step": 490
     },
     {
       "epoch": 0.31201248049922,
-      "grad_norm": 0.2936696410179138,
+      "grad_norm": 1.4083600044250488,
       "learning_rate": 0.0001951036242678118,
-      "loss": 0.0062,
+      "loss": 0.0247,
       "step": 500
     },
     {
       "epoch": 0.31825273010920435,
-      "grad_norm": 0.6721827387809753,
+      "grad_norm": 0.9176926016807556,
       "learning_rate": 0.00019489882275857377,
-      "loss": 0.0152,
+      "loss": 0.0138,
       "step": 510
     },
     {
       "epoch": 0.3244929797191888,
-      "grad_norm": 0.0886274203658104,
+      "grad_norm": 0.5357791781425476,
       "learning_rate": 0.00019468993740350387,
-      "loss": 0.0073,
+      "loss": 0.0118,
       "step": 520
     },
     {
       "epoch": 0.33073322932917315,
-      "grad_norm": 0.03001420944929123,
+      "grad_norm": 0.8709370493888855,
       "learning_rate": 0.00019447697719170815,
-      "loss": 0.0091,
+      "loss": 0.0296,
       "step": 530
     },
     {
       "epoch": 0.3369734789391576,
-      "grad_norm": 0.3404795229434967,
+      "grad_norm": 0.21523749828338623,
       "learning_rate": 0.00019425995128764887,
-      "loss": 0.0155,
+      "loss": 0.0191,
       "step": 540
     },
     {
       "epoch": 0.34321372854914195,
-      "grad_norm": 0.07943115383386612,
+      "grad_norm": 0.09816073626279831,
       "learning_rate": 0.00019403886903074987,
-      "loss": 0.0076,
+      "loss": 0.0068,
       "step": 550
     },
     {
       "epoch": 0.3494539781591264,
-      "grad_norm": 0.018820637837052345,
+      "grad_norm": 0.08138906955718994,
       "learning_rate": 0.00019381373993499493,
-      "loss": 0.0112,
+      "loss": 0.0101,
       "step": 560
     },
     {
       "epoch": 0.35569422776911075,
-      "grad_norm": 0.49909457564353943,
+      "grad_norm": 0.08689755201339722,
       "learning_rate": 0.0001935845736885181,
-      "loss": 0.018,
+      "loss": 0.009,
       "step": 570
     },
     {
       "epoch": 0.3619344773790952,
-      "grad_norm": 0.30171993374824524,
+      "grad_norm": 0.6227536797523499,
       "learning_rate": 0.0001933513801531869,
-      "loss": 0.0076,
+      "loss": 0.008,
       "step": 580
     },
     {
       "epoch": 0.36817472698907955,
-      "grad_norm": 1.3735274076461792,
+      "grad_norm": 0.41413530707359314,
       "learning_rate": 0.000193114169364178,
-      "loss": 0.0112,
+      "loss": 0.0119,
       "step": 590
     },
     {
       "epoch": 0.374414976599064,
-      "grad_norm": 0.22438661754131317,
+      "grad_norm": 0.2208208590745926,
       "learning_rate": 0.0001928729515295452,
-      "loss": 0.0117,
+      "loss": 0.0101,
       "step": 600
     },
     {
       "epoch": 0.38065522620904835,
-      "grad_norm": 0.005140058230608702,
+      "grad_norm": 0.26638615131378174,
       "learning_rate": 0.00019262773702978026,
-      "loss": 0.0085,
+      "loss": 0.0105,
       "step": 610
     },
     {
       "epoch": 0.3868954758190328,
-      "grad_norm": 0.33452799916267395,
+      "grad_norm": 0.2926352024078369,
       "learning_rate": 0.0001923785364173662,
-      "loss": 0.0295,
+      "loss": 0.0139,
       "step": 620
     },
     {
       "epoch": 0.39313572542901715,
-      "grad_norm": 0.2755588889122009,
+      "grad_norm": 0.20466898381710052,
       "learning_rate": 0.0001921253604163231,
-      "loss": 0.0106,
+      "loss": 0.0221,
       "step": 630
     },
     {
       "epoch": 0.3993759750390016,
-      "grad_norm": 0.36959072947502136,
+      "grad_norm": 0.13136933743953705,
       "learning_rate": 0.00019186821992174665,
-      "loss": 0.0085,
+      "loss": 0.014,
       "step": 640
     },
     {
       "epoch": 0.40561622464898595,
-      "grad_norm": 0.2535085082054138,
+      "grad_norm": 0.42323336005210876,
       "learning_rate": 0.00019160712599933938,
-      "loss": 0.017,
+      "loss": 0.0135,
       "step": 650
     },
     {
       "epoch": 0.4118564742589704,
-      "grad_norm": 0.06113250553607941,
+      "grad_norm": 0.05264485627412796,
       "learning_rate": 0.00019134208988493424,
-      "loss": 0.0089,
+      "loss": 0.026,
       "step": 660
     },
     {
       "epoch": 0.41809672386895474,
-      "grad_norm": 0.12086120992898941,
+      "grad_norm": 0.36449646949768066,
       "learning_rate": 0.00019107312298401137,
-      "loss": 0.0037,
+      "loss": 0.0217,
       "step": 670
     },
     {
       "epoch": 0.4243369734789392,
-      "grad_norm": 0.2163705676794052,
+      "grad_norm": 4.810301780700684,
       "learning_rate": 0.00019080023687120698,
-      "loss": 0.0125,
+      "loss": 0.0154,
       "step": 680
     },
     {
       "epoch": 0.43057722308892354,
-      "grad_norm": 0.904460608959198,
+      "grad_norm": 1.3598051071166992,
       "learning_rate": 0.00019052344328981553,
-      "loss": 0.0155,
+      "loss": 0.0145,
       "step": 690
     },
     {
       "epoch": 0.43681747269890797,
-      "grad_norm": 0.5320896506309509,
+      "grad_norm": 0.04880363494157791,
       "learning_rate": 0.00019024275415128412,
-      "loss": 0.0175,
+      "loss": 0.0187,
       "step": 700
     },
     {
       "epoch": 0.44305772230889234,
-      "grad_norm": 0.31751132011413574,
+      "grad_norm": 0.06181839480996132,
       "learning_rate": 0.00018995818153470017,
-      "loss": 0.0447,
+      "loss": 0.0095,
       "step": 710
     },
     {
       "epoch": 0.44929797191887677,
-      "grad_norm": 0.023313090205192566,
+      "grad_norm": 0.830059826374054,
       "learning_rate": 0.00018966973768627134,
-      "loss": 0.0147,
+      "loss": 0.0115,
       "step": 720
     },
     {
       "epoch": 0.45553822152886114,
-      "grad_norm": 0.3516482710838318,
+      "grad_norm": 0.05761188268661499,
       "learning_rate": 0.0001893774350187987,
-      "loss": 0.0117,
+      "loss": 0.0106,
       "step": 730
     },
     {
       "epoch": 0.46177847113884557,
-      "grad_norm": 0.8473337292671204,
+      "grad_norm": 0.2695203125476837,
       "learning_rate": 0.0001890812861111425,
-      "loss": 0.0154,
+      "loss": 0.0143,
       "step": 740
     },
     {
       "epoch": 0.46801872074882994,
-      "grad_norm": 0.50248122215271,
+      "grad_norm": 0.20724090933799744,
       "learning_rate": 0.0001887813037076809,
-      "loss": 0.0093,
+      "loss": 0.0077,
       "step": 750
     },
     {
       "epoch": 0.47425897035881437,
-      "grad_norm": 1.4992247819900513,
+      "grad_norm": 1.1357392072677612,
       "learning_rate": 0.00018847750071776143,
-      "loss": 0.0264,
+      "loss": 0.0104,
       "step": 760
     },
     {
       "epoch": 0.48049921996879874,
-      "grad_norm": 1.2603468894958496,
+      "grad_norm": 0.15949776768684387,
       "learning_rate": 0.00018816989021514568,
-      "loss": 0.0111,
+      "loss": 0.0087,
       "step": 770
     },
     {
       "epoch": 0.48673946957878317,
-      "grad_norm": 0.49606853723526,
+      "grad_norm": 1.279613733291626,
       "learning_rate": 0.00018785848543744632,
-      "loss": 0.0125,
+      "loss": 0.0085,
       "step": 780
     },
     {
       "epoch": 0.49297971918876754,
-      "grad_norm": 0.9916170835494995,
+      "grad_norm": 0.5072678327560425,
       "learning_rate": 0.00018754329978555782,
-      "loss": 0.01,
+      "loss": 0.013,
       "step": 790
     },
     {
       "epoch": 0.49921996879875197,
-      "grad_norm": 1.2155410051345825,
+      "grad_norm": 0.025297529995441437,
       "learning_rate": 0.0001872243468230795,
-      "loss": 0.0126,
+      "loss": 0.0061,
       "step": 800
     },
     {
       "epoch": 0.5054602184087363,
-      "grad_norm": 2.0329480171203613,
+      "grad_norm": 4.629367351531982,
       "learning_rate": 0.000186901640275732,
-      "loss": 0.0165,
+      "loss": 0.0137,
       "step": 810
     },
     {
       "epoch": 0.5117004680187207,
-      "grad_norm": 0.0864962488412857,
+      "grad_norm": 0.5096446871757507,
       "learning_rate": 0.00018657519403076644,
-      "loss": 0.0161,
+      "loss": 0.0117,
       "step": 820
     },
     {
       "epoch": 0.5179407176287052,
-      "grad_norm": 0.1404924988746643,
+      "grad_norm": 0.8323133587837219,
       "learning_rate": 0.00018624502213636702,
-      "loss": 0.0216,
+      "loss": 0.0147,
       "step": 830
     },
     {
       "epoch": 0.5241809672386896,
-      "grad_norm": 1.004807472229004,
+      "grad_norm": 0.33263689279556274,
       "learning_rate": 0.00018591113880104626,
-      "loss": 0.0135,
+      "loss": 0.0132,
       "step": 840
     },
     {
       "epoch": 0.5304212168486739,
-      "grad_norm": 0.3545820116996765,
+      "grad_norm": 2.3099992275238037,
       "learning_rate": 0.0001855735583930337,
-      "loss": 0.0124,
+      "loss": 0.0148,
       "step": 850
     },
     {
       "epoch": 0.5366614664586583,
-      "grad_norm": 1.5320320129394531,
+      "grad_norm": 0.25143998861312866,
       "learning_rate": 0.0001852322954396575,
-      "loss": 0.0096,
+      "loss": 0.0108,
       "step": 860
     },
     {
       "epoch": 0.5429017160686428,
-      "grad_norm": 0.41443562507629395,
+      "grad_norm": 0.8478009700775146,
       "learning_rate": 0.0001848873646267194,
-      "loss": 0.0164,
+      "loss": 0.0177,
       "step": 870
     },
     {
       "epoch": 0.5491419656786272,
-      "grad_norm": 0.3798370063304901,
+      "grad_norm": 0.481947124004364,
       "learning_rate": 0.0001845387807978625,
-      "loss": 0.008,
+      "loss": 0.0065,
       "step": 880
     },
     {
       "epoch": 0.5553822152886115,
-      "grad_norm": 0.16353817284107208,
+      "grad_norm": 0.16595329344272614,
       "learning_rate": 0.0001841865589539327,
-      "loss": 0.0094,
+      "loss": 0.0118,
       "step": 890
     },
     {
       "epoch": 0.5616224648985959,
-      "grad_norm": 0.03544636815786362,
+      "grad_norm": 0.01159042026847601,
       "learning_rate": 0.00018383071425233315,
-      "loss": 0.0054,
+      "loss": 0.0049,
       "step": 900
     },
     {
       "epoch": 0.5678627145085804,
-      "grad_norm": 0.6518317461013794,
+      "grad_norm": 0.22808197140693665,
       "learning_rate": 0.0001834712620063718,
       "loss": 0.0089,
       "step": 910
     },
     {
       "epoch": 0.5741029641185648,
-      "grad_norm": 2.557168960571289,
+      "grad_norm": 0.635155439376831,
       "learning_rate": 0.00018310821768460257,
-      "loss": 0.0142,
+      "loss": 0.0164,
       "step": 920
     },
     {
       "epoch": 0.5803432137285491,
-      "grad_norm": 0.38063621520996094,
+      "grad_norm": 0.4370354413986206,
       "learning_rate": 0.00018274159691015966,
-      "loss": 0.0087,
+      "loss": 0.0111,
       "step": 930
     },
     {
       "epoch": 0.5865834633385335,
-      "grad_norm": 0.10065897554159164,
+      "grad_norm": 0.21154633164405823,
       "learning_rate": 0.0001823714154600851,
-      "loss": 0.0072,
+      "loss": 0.0076,
       "step": 940
     },
     {
       "epoch": 0.592823712948518,
-      "grad_norm": 0.2000608593225479,
+      "grad_norm": 0.0030444941949099302,
       "learning_rate": 0.00018199768926465003,
-      "loss": 0.0041,
+      "loss": 0.0035,
       "step": 950
     },
     {
       "epoch": 0.5990639625585024,
-      "grad_norm": 0.5174477696418762,
+      "grad_norm": 0.5866973400115967,
       "learning_rate": 0.000181620434406669,
-      "loss": 0.0117,
+      "loss": 0.0189,
       "step": 960
     },
     {
       "epoch": 0.6053042121684867,
-      "grad_norm": 0.14495687186717987,
+      "grad_norm": 1.6045774221420288,
       "learning_rate": 0.00018123966712080782,
-      "loss": 0.0052,
+      "loss": 0.0101,
       "step": 970
     },
     {
       "epoch": 0.6115444617784711,
-      "grad_norm": 0.00905101839452982,
+      "grad_norm": 0.379751980304718,
       "learning_rate": 0.0001808554037928851,
-      "loss": 0.0053,
+      "loss": 0.0122,
       "step": 980
     },
     {
       "epoch": 0.6177847113884556,
-      "grad_norm": 0.7881221175193787,
+      "grad_norm": 0.12018060684204102,
       "learning_rate": 0.00018046766095916697,
-      "loss": 0.0051,
+      "loss": 0.0061,
       "step": 990
     },
     {
       "epoch": 0.62402496099844,
-      "grad_norm": 0.008211757056415081,
+      "grad_norm": 0.16439475119113922,
       "learning_rate": 0.00018007645530565555,
-      "loss": 0.0054,
+      "loss": 0.0181,
       "step": 1000
     },
     {
       "epoch": 0.6302652106084243,
-      "grad_norm": 0.19711071252822876,
+      "grad_norm": 0.21596841514110565,
       "learning_rate": 0.00017968180366737083,
-      "loss": 0.0108,
+      "loss": 0.0092,
       "step": 1010
     },
     {
       "epoch": 0.6365054602184087,
-      "grad_norm": 0.4079477787017822,
+      "grad_norm": 0.03322902321815491,
       "learning_rate": 0.00017928372302762616,
-      "loss": 0.0091,
+      "loss": 0.0058,
       "step": 1020
     },
     {
       "epoch": 0.6427457098283932,
-      "grad_norm": 0.38131341338157654,
+      "grad_norm": 0.0635453462600708,
       "learning_rate": 0.00017888223051729755,
-      "loss": 0.0197,
+      "loss": 0.0115,
       "step": 1030
     },
     {
       "epoch": 0.6489859594383776,
-      "grad_norm": 0.2548066973686218,
+      "grad_norm": 0.8183733820915222,
       "learning_rate": 0.00017847734341408628,
-      "loss": 0.0221,
+      "loss": 0.0128,
       "step": 1040
     },
     {
       "epoch": 0.6552262090483619,
-      "grad_norm": 0.016785508021712303,
+      "grad_norm": 0.03648073971271515,
       "learning_rate": 0.00017806907914177554,
-      "loss": 0.0082,
+      "loss": 0.0058,
       "step": 1050
     },
     {
       "epoch": 0.6614664586583463,
-      "grad_norm": 0.17338435351848602,
+      "grad_norm": 0.14891788363456726,
       "learning_rate": 0.00017765745526948046,
-      "loss": 0.0141,
+      "loss": 0.0109,
       "step": 1060
     },
     {
       "epoch": 0.6677067082683308,
-      "grad_norm": 1.0973267555236816,
+      "grad_norm": 0.23602107167243958,
       "learning_rate": 0.0001772424895108922,
-      "loss": 0.0133,
+      "loss": 0.0099,
       "step": 1070
     },
     {
       "epoch": 0.6739469578783152,
-      "grad_norm": 0.015839748084545135,
+      "grad_norm": 0.016226977109909058,
       "learning_rate": 0.00017682419972351554,
-      "loss": 0.0083,
+      "loss": 0.0061,
       "step": 1080
     },
     {
       "epoch": 0.6801872074882995,
-      "grad_norm": 0.014509073458611965,
+      "grad_norm": 0.002975013805553317,
       "learning_rate": 0.00017640260390790053,
-      "loss": 0.0085,
+      "loss": 0.007,
       "step": 1090
     },
     {
       "epoch": 0.6864274570982839,
-      "grad_norm": 0.03248611092567444,
+      "grad_norm": 0.3831285834312439,
       "learning_rate": 0.00017597772020686768,
-      "loss": 0.0037,
+      "loss": 0.0021,
       "step": 1100
     },
     {
       "epoch": 0.6926677067082684,
-      "grad_norm": 0.06335289031267166,
+      "grad_norm": 0.005971380975097418,
       "learning_rate": 0.00017554956690472748,
-      "loss": 0.0071,
+      "loss": 0.0062,
       "step": 1110
     },
     {
       "epoch": 0.6989079563182528,
-      "grad_norm": 0.04545082896947861,
+      "grad_norm": 0.02962116152048111,
       "learning_rate": 0.00017511816242649322,
-      "loss": 0.0125,
+      "loss": 0.0065,
       "step": 1120
     },
     {
       "epoch": 0.7051482059282371,
-      "grad_norm": 1.186499834060669,
+      "grad_norm": 0.24174004793167114,
       "learning_rate": 0.00017468352533708836,
-      "loss": 0.0142,
+      "loss": 0.0059,
       "step": 1130
     },
     {
       "epoch": 0.7113884555382215,
-      "grad_norm": 0.7652639746665955,
+      "grad_norm": 0.025318020954728127,
       "learning_rate": 0.00017424567434054758,
-      "loss": 0.0061,
+      "loss": 0.0056,
       "step": 1140
     },
     {
       "epoch": 0.717628705148206,
-      "grad_norm": 0.2659410238265991,
+      "grad_norm": 0.0799591913819313,
       "learning_rate": 0.0001738046282792117,
-      "loss": 0.0097,
+      "loss": 0.0095,
       "step": 1150
     },
     {
       "epoch": 0.7238689547581904,
-      "grad_norm": 0.14636874198913574,
+      "grad_norm": 1.6064969301223755,
       "learning_rate": 0.000173360406132917,
-      "loss": 0.0056,
+      "loss": 0.0089,
       "step": 1160
     },
     {
       "epoch": 0.7301092043681747,
-      "grad_norm": 0.26678094267845154,
+      "grad_norm": 0.11897823959589005,
       "learning_rate": 0.00017291302701817843,
-      "loss": 0.01,
+      "loss": 0.0076,
       "step": 1170
     },
     {
       "epoch": 0.7363494539781591,
-      "grad_norm": 0.11483141034841537,
+      "grad_norm": 0.136907160282135,
       "learning_rate": 0.00017246251018736687,
-      "loss": 0.0114,
+      "loss": 0.0136,
       "step": 1180
     },
     {
       "epoch": 0.7425897035881436,
-      "grad_norm": 0.010227694176137447,
+      "grad_norm": 0.08185099810361862,
       "learning_rate": 0.0001720088750278807,
-      "loss": 0.0073,
+      "loss": 0.0096,
       "step": 1190
     },
     {
       "epoch": 0.748829953198128,
-      "grad_norm": 0.15639427304267883,
+      "grad_norm": 0.21249833703041077,
       "learning_rate": 0.0001715521410613114,
-      "loss": 0.0132,
+      "loss": 0.0093,
       "step": 1200
     },
     {
       "epoch": 0.7550702028081123,
-      "grad_norm": 0.1970348358154297,
+      "grad_norm": 0.12882816791534424,
       "learning_rate": 0.00017109232794260362,
-      "loss": 0.0062,
+      "loss": 0.0081,
       "step": 1210
     },
     {
       "epoch": 0.7613104524180967,
-      "grad_norm": 0.13912296295166016,
+      "grad_norm": 0.11902020871639252,
       "learning_rate": 0.00017062945545920927,
-      "loss": 0.0112,
+      "loss": 0.0097,
       "step": 1220
     },
     {
       "epoch": 0.7675507020280812,
-      "grad_norm": 0.0700615793466568,
+      "grad_norm": 0.2858002483844757,
       "learning_rate": 0.00017016354353023593,
-      "loss": 0.0086,
+      "loss": 0.0087,
       "step": 1230
     },
     {
       "epoch": 0.7737909516380655,
-      "grad_norm": 0.08227328956127167,
+      "grad_norm": 0.11392389237880707,
       "learning_rate": 0.00016969461220558976,
-      "loss": 0.005,
+      "loss": 0.0109,
       "step": 1240
     },
     {
       "epoch": 0.7800312012480499,
-      "grad_norm": 0.5434132814407349,
+      "grad_norm": 0.005757195875048637,
       "learning_rate": 0.00016922268166511255,
-      "loss": 0.0078,
+      "loss": 0.0083,
       "step": 1250
     },
     {
       "epoch": 0.7862714508580343,
-      "grad_norm": 0.14495456218719482,
+      "grad_norm": 0.024745512753725052,
       "learning_rate": 0.00016874777221771354,
-      "loss": 0.0053,
+      "loss": 0.0081,
       "step": 1260
     },
     {
       "epoch": 0.7925117004680188,
-      "grad_norm": 0.6543499827384949,
+      "grad_norm": 0.08866913616657257,
       "learning_rate": 0.00016826990430049513,
-      "loss": 0.0021,
+      "loss": 0.0127,
       "step": 1270
     },
     {
       "epoch": 0.7987519500780031,
-      "grad_norm": 1.2169020175933838,
+      "grad_norm": 0.25643572211265564,
       "learning_rate": 0.00016778909847787378,
-      "loss": 0.0109,
+      "loss": 0.0116,
       "step": 1280
     },
     {
       "epoch": 0.8049921996879875,
-      "grad_norm": 0.059403009712696075,
+      "grad_norm": 0.185395285487175,
       "learning_rate": 0.00016730537544069464,
-      "loss": 0.0083,
+      "loss": 0.0087,
       "step": 1290
     },
     {
       "epoch": 0.8112324492979719,
-      "grad_norm": 1.1702568531036377,
+      "grad_norm": 0.09554590284824371,
       "learning_rate": 0.0001668187560053415,
-      "loss": 0.012,
+      "loss": 0.0059,
       "step": 1300
     },
     {
       "epoch": 0.8174726989079563,
-      "grad_norm": 0.35837844014167786,
+      "grad_norm": 0.10387805849313736,
       "learning_rate": 0.00016632926111284066,
-      "loss": 0.0059,
+      "loss": 0.0026,
       "step": 1310
     },
     {
       "epoch": 0.8237129485179407,
-      "grad_norm": 0.26777470111846924,
+      "grad_norm": 0.11453939974308014,
       "learning_rate": 0.00016583691182796013,
-      "loss": 0.0079,
+      "loss": 0.0073,
       "step": 1320
     },
     {
       "epoch": 0.8299531981279251,
-      "grad_norm": 0.6792253255844116,
+      "grad_norm": 0.41253453493118286,
       "learning_rate": 0.00016534172933830275,
-      "loss": 0.0116,
+      "loss": 0.0087,
       "step": 1330
     },
     {
       "epoch": 0.8361934477379095,
-      "grad_norm": 0.23119303584098816,
+      "grad_norm": 0.017546722665429115,
       "learning_rate": 0.00016484373495339477,
-      "loss": 0.0048,
+      "loss": 0.004,
       "step": 1340
     },
     {
       "epoch": 0.8424336973478939,
-      "grad_norm": 0.2668294608592987,
+      "grad_norm": 0.2553124725818634,
       "learning_rate": 0.0001643429501037685,
-      "loss": 0.0106,
+      "loss": 0.0101,
       "step": 1350
     },
     {
       "epoch": 0.8486739469578783,
-      "grad_norm": 0.18860620260238647,
+      "grad_norm": 0.2235170602798462,
       "learning_rate": 0.00016383939634004034,
-      "loss": 0.0111,
+      "loss": 0.0102,
       "step": 1360
     },
     {
       "epoch": 0.8549141965678627,
-      "grad_norm": 0.2770467698574066,
+      "grad_norm": 0.030884789302945137,
       "learning_rate": 0.00016333309533198316,
-      "loss": 0.0094,
+      "loss": 0.0075,
       "step": 1370
     },
     {
       "epoch": 0.8611544461778471,
-      "grad_norm": 0.14996962249279022,
+      "grad_norm": 0.13557137548923492,
       "learning_rate": 0.00016282406886759392,
-      "loss": 0.006,
+      "loss": 0.0094,
       "step": 1380
     },
     {
       "epoch": 0.8673946957878315,
-      "grad_norm": 0.0023365733213722706,
+      "grad_norm": 0.3623427450656891,
       "learning_rate": 0.000162312338852156,
-      "loss": 0.0062,
+      "loss": 0.0252,
       "step": 1390
     },
     {
       "epoch": 0.8736349453978159,
-      "grad_norm": 0.06916924566030502,
+      "grad_norm": 0.06469976902008057,
       "learning_rate": 0.00016179792730729652,
-      "loss": 0.0013,
+      "loss": 0.0154,
       "step": 1400
     },
     {
       "epoch": 0.8798751950078003,
-      "grad_norm": 0.5850982666015625,
+      "grad_norm": 0.1621229499578476,
       "learning_rate": 0.0001612808563700388,
-      "loss": 0.011,
+      "loss": 0.0118,
       "step": 1410
     },
     {
       "epoch": 0.8861154446177847,
-      "grad_norm": 0.3427271246910095,
+      "grad_norm": 0.1326974332332611,
       "learning_rate": 0.00016076114829184938,
-      "loss": 0.0122,
+      "loss": 0.0152,
       "step": 1420
     },
     {
       "epoch": 0.8923556942277691,
-      "grad_norm": 0.300658643245697,
+      "grad_norm": 0.6861521005630493,
       "learning_rate": 0.00016023882543768093,
       "loss": 0.0129,
       "step": 1430
     },
     {
       "epoch": 0.8985959438377535,
-      "grad_norm": 0.20494520664215088,
+      "grad_norm": 0.007267525419592857,
       "learning_rate": 0.00015971391028500934,
-      "loss": 0.0098,
+      "loss": 0.0072,
       "step": 1440
     },
     {
       "epoch": 0.9048361934477379,
-      "grad_norm": 0.2109510898590088,
+      "grad_norm": 0.022586530074477196,
       "learning_rate": 0.00015918642542286687,
-      "loss": 0.0222,
+      "loss": 0.0132,
       "step": 1450
     },
     {
       "epoch": 0.9110764430577223,
-      "grad_norm": 0.05611221864819527,
+      "grad_norm": 1.0569708347320557,
       "learning_rate": 0.00015865639355086967,
-      "loss": 0.0145,
+      "loss": 0.0133,
       "step": 1460
     },
     {
       "epoch": 0.9173166926677067,
-      "grad_norm": 0.3887752592563629,
+      "grad_norm": 1.3910530805587769,
       "learning_rate": 0.00015812383747824116,
       "loss": 0.0111,
       "step": 1470
     },
     {
       "epoch": 0.9235569422776911,
-      "grad_norm": 1.0526059865951538,
+      "grad_norm": 0.2169407606124878,
       "learning_rate": 0.00015758878012283044,
-      "loss": 0.0126,
+      "loss": 0.0146,
       "step": 1480
     },
     {
       "epoch": 0.9297971918876755,
-      "grad_norm": 0.2805418074131012,
+      "grad_norm": 0.03829345107078552,
       "learning_rate": 0.00015705124451012602,
-      "loss": 0.0076,
+      "loss": 0.0058,
       "step": 1490
     },
     {
       "epoch": 0.9360374414976599,
-      "grad_norm": 0.014235242269933224,
+      "grad_norm": 0.05906001478433609,
       "learning_rate": 0.0001565112537722649,
-      "loss": 0.01,
+      "loss": 0.0043,
       "step": 1500
     },
     {
       "epoch": 0.9422776911076443,
-      "grad_norm": 0.3192802965641022,
+      "grad_norm": 0.4823719263076782,
       "learning_rate": 0.00015596883114703714,
-      "loss": 0.015,
+      "loss": 0.0089,
       "step": 1510
     },
     {
       "epoch": 0.9485179407176287,
-      "grad_norm": 0.06423596292734146,
+      "grad_norm": 0.10682803392410278,
       "learning_rate": 0.00015542399997688598,
-      "loss": 0.0145,
+      "loss": 0.0111,
       "step": 1520
     },
     {
       "epoch": 0.9547581903276131,
-      "grad_norm": 0.06878869235515594,
+      "grad_norm": 0.20365221798419952,
       "learning_rate": 0.00015487678370790307,
-      "loss": 0.0129,
+      "loss": 0.0108,
       "step": 1530
     },
     {
       "epoch": 0.9609984399375975,
-      "grad_norm": 0.3116801381111145,
+      "grad_norm": 0.19743436574935913,
       "learning_rate": 0.0001543272058888198,
-      "loss": 0.0071,
+      "loss": 0.0074,
       "step": 1540
     },
     {
       "epoch": 0.9672386895475819,
-      "grad_norm": 0.40218546986579895,
+      "grad_norm": 0.1283200979232788,
       "learning_rate": 0.00015377529016999365,
-      "loss": 0.0117,
+      "loss": 0.0088,
       "step": 1550
     },
     {
       "epoch": 0.9734789391575663,
-      "grad_norm": 0.9296953678131104,
+      "grad_norm": 0.7578624486923218,
       "learning_rate": 0.00015322106030239054,
-      "loss": 0.0125,
+      "loss": 0.0056,
       "step": 1560
     },
     {
       "epoch": 0.9797191887675507,
-      "grad_norm": 0.10756753385066986,
+      "grad_norm": 0.02659941278398037,
       "learning_rate": 0.00015266454013656278,
-      "loss": 0.0119,
+      "loss": 0.0053,
       "step": 1570
     },
     {
       "epoch": 0.9859594383775351,
-      "grad_norm": 0.09447533637285233,
+      "grad_norm": 0.009654139168560505,
       "learning_rate": 0.0001521057536216227,
-      "loss": 0.0112,
+      "loss": 0.0041,
       "step": 1580
     },
     {
       "epoch": 0.9921996879875195,
-      "grad_norm": 0.5510380268096924,
+      "grad_norm": 0.06220339611172676,
       "learning_rate": 0.0001515447248042118,
-      "loss": 0.0126,
+      "loss": 0.0077,
       "step": 1590
     },
     {
       "epoch": 0.9984399375975039,
-      "grad_norm": 0.004704979248344898,
+      "grad_norm": 0.08257798105478287,
       "learning_rate": 0.00015098147782746633,
-      "loss": 0.0057,
+      "loss": 0.0022,
       "step": 1600
     },
     {
       "epoch": 1.004368174726989,
-      "grad_norm": 0.24134884774684906,
+      "grad_norm": 0.001243369304575026,
       "learning_rate": 0.00015041603692997788,
-      "loss": 0.006,
+      "loss": 0.0022,
       "step": 1610
     },
     {
       "epoch": 1.0106084243369735,
-      "grad_norm": 0.23953481018543243,
+      "grad_norm": 0.006356839090585709,
       "learning_rate": 0.00014984842644475075,
-      "loss": 0.0048,
+      "loss": 0.0037,
       "step": 1620
     },
     {
       "epoch": 1.0168486739469578,
-      "grad_norm": 0.008176998235285282,
+      "grad_norm": 0.02987987920641899,
       "learning_rate": 0.0001492786707981544,
-      "loss": 0.006,
+      "loss": 0.004,
       "step": 1630
     },
     {
       "epoch": 1.0230889235569423,
-      "grad_norm": 0.046635717153549194,
+      "grad_norm": 0.030943598598241806,
       "learning_rate": 0.00014870679450887261,
-      "loss": 0.0087,
+      "loss": 0.0091,
       "step": 1640
     },
     {
       "epoch": 1.0293291731669267,
-      "grad_norm": 0.03800414875149727,
+      "grad_norm": 0.10699480026960373,
       "learning_rate": 0.00014813282218684826,
-      "loss": 0.0037,
+      "loss": 0.0028,
       "step": 1650
     },
     {
       "epoch": 1.035569422776911,
-      "grad_norm": 0.10656531155109406,
+      "grad_norm": 0.0018822617130354047,
       "learning_rate": 0.000147556778532224,
-      "loss": 0.0058,
+      "loss": 0.006,
       "step": 1660
     },
     {
       "epoch": 1.0418096723868955,
-      "grad_norm": 0.28034937381744385,
+      "grad_norm": 0.05771805718541145,
       "learning_rate": 0.00014697868833427985,
-      "loss": 0.0052,
+      "loss": 0.0047,
       "step": 1670
     },
     {
       "epoch": 1.04804992199688,
-      "grad_norm": 0.09066982567310333,
+      "grad_norm": 0.027922987937927246,
       "learning_rate": 0.0001463985764703659,
-      "loss": 0.0058,
+      "loss": 0.0049,
       "step": 1680
     },
     {
       "epoch": 1.0542901716068642,
-      "grad_norm": 0.024250924587249756,
+      "grad_norm": 0.010962706990540028,
       "learning_rate": 0.00014581646790483218,
-      "loss": 0.0098,
+      "loss": 0.006,
       "step": 1690
     },
     {
       "epoch": 1.0605304212168487,
-      "grad_norm": 0.20950108766555786,
+      "grad_norm": 0.223402202129364,
       "learning_rate": 0.00014523238768795386,
-      "loss": 0.0085,
+      "loss": 0.0061,
       "step": 1700
     },
     {
       "epoch": 1.0667706708268332,
-      "grad_norm": 0.26374754309654236,
+      "grad_norm": 0.09500240534543991,
       "learning_rate": 0.00014464636095485385,
-      "loss": 0.0096,
+      "loss": 0.0062,
       "step": 1710
     },
     {
       "epoch": 1.0730109204368175,
-      "grad_norm": 0.004857131280004978,
+      "grad_norm": 0.003751072334125638,
       "learning_rate": 0.00014405841292442054,
-      "loss": 0.0089,
+      "loss": 0.0057,
       "step": 1720
     },
     {
       "epoch": 1.079251170046802,
-      "grad_norm": 0.1881384700536728,
+      "grad_norm": 0.0011663122568279505,
       "learning_rate": 0.00014346856889822305,
-      "loss": 0.0097,
+      "loss": 0.0041,
       "step": 1730
     },
     {
       "epoch": 1.0854914196567862,
-      "grad_norm": 0.3700352609157562,
+      "grad_norm": 0.10746866464614868,
       "learning_rate": 0.000142876854259422,
-      "loss": 0.0098,
+      "loss": 0.0052,
       "step": 1740
     },
     {
       "epoch": 1.0917316692667707,
-      "grad_norm": 0.029609568417072296,
+      "grad_norm": 0.008093472570180893,
       "learning_rate": 0.00014228329447167742,
-      "loss": 0.0068,
+      "loss": 0.0027,
       "step": 1750
     },
     {
       "epoch": 1.0979719188767552,
-      "grad_norm": 0.6126700043678284,
+      "grad_norm": 0.009198025800287724,
       "learning_rate": 0.000141687915078053,
-      "loss": 0.0201,
+      "loss": 0.0102,
       "step": 1760
     },
     {
       "epoch": 1.1042121684867394,
-      "grad_norm": 0.16146229207515717,
+      "grad_norm": 0.1056290939450264,
       "learning_rate": 0.00014109074169991662,
-      "loss": 0.014,
+      "loss": 0.0038,
       "step": 1770
     },
     {
       "epoch": 1.110452418096724,
-      "grad_norm": 0.13777479529380798,
+      "grad_norm": 0.07998266071081161,
       "learning_rate": 0.000140491800035838,
-      "loss": 0.0115,
+      "loss": 0.0063,
       "step": 1780
     },
     {
       "epoch": 1.1166926677067082,
-      "grad_norm": 0.006505697965621948,
+      "grad_norm": 0.001640097238123417,
       "learning_rate": 0.00013989111586048265,
-      "loss": 0.0044,
+      "loss": 0.0025,
       "step": 1790
     },
     {
       "epoch": 1.1229329173166926,
-      "grad_norm": 0.02420825883746147,
+      "grad_norm": 0.014512299560010433,
       "learning_rate": 0.00013928871502350286,
-      "loss": 0.0021,
+      "loss": 0.0011,
       "step": 1800
     },
     {
       "epoch": 1.1291731669266771,
-      "grad_norm": 1.3016235828399658,
+      "grad_norm": 0.001313922111876309,
       "learning_rate": 0.00013868462344842513,
-      "loss": 0.0082,
+      "loss": 0.0055,
       "step": 1810
     },
     {
       "epoch": 1.1354134165366614,
-      "grad_norm": 0.35244765877723694,
+      "grad_norm": 0.11844850331544876,
       "learning_rate": 0.00013807886713153463,
-      "loss": 0.0148,
+      "loss": 0.0132,
       "step": 1820
     },
     {
       "epoch": 1.1416536661466459,
-      "grad_norm": 0.011646454222500324,
+      "grad_norm": 0.02499815635383129,
       "learning_rate": 0.00013747147214075652,
-      "loss": 0.0072,
+      "loss": 0.0054,
       "step": 1830
     },
     {
       "epoch": 1.1478939157566304,
-      "grad_norm": 0.3862789571285248,
+      "grad_norm": 0.03466424345970154,
       "learning_rate": 0.00013686246461453417,
-      "loss": 0.0171,
+      "loss": 0.0092,
       "step": 1840
     },
     {
       "epoch": 1.1541341653666146,
-      "grad_norm": 0.8616077899932861,
+      "grad_norm": 0.03496575355529785,
       "learning_rate": 0.00013625187076070423,
-      "loss": 0.0092,
+      "loss": 0.0073,
       "step": 1850
     },
     {
       "epoch": 1.160374414976599,
-      "grad_norm": 0.04771306738257408,
+      "grad_norm": 0.2750947177410126,
       "learning_rate": 0.00013563971685536894,
-      "loss": 0.0062,
+      "loss": 0.0082,
       "step": 1860
     },
     {
       "epoch": 1.1666146645865836,
-      "grad_norm": 0.26229849457740784,
+      "grad_norm": 0.2440357208251953,
       "learning_rate": 0.00013502602924176516,
-      "loss": 0.0059,
+      "loss": 0.0079,
       "step": 1870
     },
     {
       "epoch": 1.1728549141965678,
-      "grad_norm": 0.017036737874150276,
+      "grad_norm": 0.0070082019083201885,
       "learning_rate": 0.00013441083432913107,
-      "loss": 0.0059,
+      "loss": 0.0088,
       "step": 1880
     },
     {
       "epoch": 1.1790951638065523,
-      "grad_norm": 0.10029294341802597,
+      "grad_norm": 0.009735733270645142,
       "learning_rate": 0.0001337941585915693,
-      "loss": 0.0061,
+      "loss": 0.008,
       "step": 1890
     },
     {
       "epoch": 1.1853354134165366,
-      "grad_norm": 0.15829603374004364,
+      "grad_norm": 0.03347931057214737,
       "learning_rate": 0.00013317602856690789,
-      "loss": 0.0045,
+      "loss": 0.004,
       "step": 1900
     },
     {
       "epoch": 1.191575663026521,
-      "grad_norm": 0.12337626516819,
+      "grad_norm": 0.059263862669467926,
       "learning_rate": 0.0001325564708555583,
-      "loss": 0.0042,
+      "loss": 0.0079,
       "step": 1910
     },
     {
       "epoch": 1.1978159126365056,
-      "grad_norm": 0.1467108130455017,
+      "grad_norm": 0.2200971245765686,
       "learning_rate": 0.00013193551211937042,
-      "loss": 0.0072,
+      "loss": 0.0079,
       "step": 1920
     },
     {
       "epoch": 1.2040561622464898,
-      "grad_norm": 0.12583035230636597,
+      "grad_norm": 11.123027801513672,
       "learning_rate": 0.0001313131790804856,
-      "loss": 0.0037,
+      "loss": 0.005,
       "step": 1930
     },
     {
       "epoch": 1.2102964118564743,
-      "grad_norm": 0.07843368500471115,
+      "grad_norm": 0.3263792395591736,
       "learning_rate": 0.00013068949852018634,
-      "loss": 0.0077,
+      "loss": 0.0129,
       "step": 1940
     },
     {
       "epoch": 1.2165366614664586,
-      "grad_norm": 0.0029602251015603542,
+      "grad_norm": 0.01347371470183134,
       "learning_rate": 0.00013006449727774401,
-      "loss": 0.0058,
+      "loss": 0.0091,
       "step": 1950
     },
     {
       "epoch": 1.222776911076443,
-      "grad_norm": 0.049546901136636734,
+      "grad_norm": 0.003924458287656307,
       "learning_rate": 0.0001294382022492638,
-      "loss": 0.0061,
+      "loss": 0.0062,
       "step": 1960
     },
     {
       "epoch": 1.2290171606864275,
-      "grad_norm": 0.16197310388088226,
+      "grad_norm": 0.1540151685476303,
       "learning_rate": 0.00012881064038652734,
-      "loss": 0.0058,
+      "loss": 0.0085,
       "step": 1970
     },
     {
       "epoch": 1.2352574102964118,
-      "grad_norm": 0.11641169339418411,
+      "grad_norm": 0.08336371183395386,
       "learning_rate": 0.00012818183869583264,
-      "loss": 0.0027,
+      "loss": 0.0123,
       "step": 1980
     },
     {
       "epoch": 1.2414976599063963,
-      "grad_norm": 0.001966765383258462,
+      "grad_norm": 0.4115087687969208,
       "learning_rate": 0.00012755182423683224,
-      "loss": 0.0014,
+      "loss": 0.0066,
       "step": 1990
     },
     {
       "epoch": 1.2477379095163807,
-      "grad_norm": 0.0025216981302946806,
+      "grad_norm": 0.06939776241779327,
       "learning_rate": 0.00012692062412136842,
-      "loss": 0.0037,
+      "loss": 0.0055,
       "step": 2000
     },
     {
       "epoch": 1.253978159126365,
-      "grad_norm": 0.23342859745025635,
+      "grad_norm": 0.2686580717563629,
       "learning_rate": 0.00012628826551230676,
-      "loss": 0.0043,
+      "loss": 0.0052,
       "step": 2010
     },
     {
       "epoch": 1.2602184087363495,
-      "grad_norm": 0.0014470007736235857,
+      "grad_norm": 0.03792508319020271,
       "learning_rate": 0.00012565477562236706,
-      "loss": 0.0024,
+      "loss": 0.0033,
       "step": 2020
     },
     {
       "epoch": 1.266458658346334,
-      "grad_norm": 0.09222376346588135,
+      "grad_norm": 0.07231654971837997,
       "learning_rate": 0.00012502018171295216,
-      "loss": 0.0027,
+      "loss": 0.0064,
       "step": 2030
     },
     {
       "epoch": 1.2726989079563182,
-      "grad_norm": 0.08314360678195953,
+      "grad_norm": 0.10225731134414673,
       "learning_rate": 0.0001243845110929751,
-      "loss": 0.0054,
+      "loss": 0.0147,
       "step": 2040
     },
     {
       "epoch": 1.2789391575663027,
-      "grad_norm": 0.03709840402007103,
+      "grad_norm": 0.12953227758407593,
       "learning_rate": 0.0001237477911176837,
-      "loss": 0.0064,
+      "loss": 0.0088,
       "step": 2050
     },
     {
       "epoch": 1.285179407176287,
-      "grad_norm": 0.1064552441239357,
+      "grad_norm": 0.08767446130514145,
       "learning_rate": 0.00012311004918748333,
-      "loss": 0.0044,
+      "loss": 0.0058,
       "step": 2060
     },
     {
       "epoch": 1.2914196567862715,
-      "grad_norm": 0.028765583410859108,
+      "grad_norm": 0.0033362265676259995,
       "learning_rate": 0.00012247131274675788,
-      "loss": 0.0098,
+      "loss": 0.0048,
       "step": 2070
     },
     {
       "epoch": 1.2976599063962557,
-      "grad_norm": 0.14255695044994354,
+      "grad_norm": 4.914682865142822,
       "learning_rate": 0.00012183160928268875,
-      "loss": 0.0063,
+      "loss": 0.0073,
       "step": 2080
     },
     {
       "epoch": 1.3039001560062402,
-      "grad_norm": 0.352822482585907,
+      "grad_norm": 0.32156863808631897,
       "learning_rate": 0.00012119096632407185,
-      "loss": 0.0116,
+      "loss": 0.0102,
       "step": 2090
     },
     {
       "epoch": 1.3101404056162247,
-      "grad_norm": 0.08606994897127151,
+      "grad_norm": 0.45002901554107666,
       "learning_rate": 0.00012054941144013298,
-      "loss": 0.0102,
+      "loss": 0.0115,
       "step": 2100
     },
     {
       "epoch": 1.316380655226209,
-      "grad_norm": 0.1337183713912964,
+      "grad_norm": 0.12745647132396698,
       "learning_rate": 0.00011990697223934151,
-      "loss": 0.0088,
+      "loss": 0.009,
       "step": 2110
     },
     {
       "epoch": 1.3226209048361934,
-      "grad_norm": 0.0092279938980937,
+      "grad_norm": 0.17609676718711853,
       "learning_rate": 0.00011926367636822223,
-      "loss": 0.0049,
+      "loss": 0.0066,
       "step": 2120
     },
     {
       "epoch": 1.328861154446178,
-      "grad_norm": 0.003310105064883828,
+      "grad_norm": 0.01337698195129633,
       "learning_rate": 0.00011861955151016548,
-      "loss": 0.0059,
+      "loss": 0.0056,
       "step": 2130
     },
     {
       "epoch": 1.3351014040561622,
-      "grad_norm": 0.06122155115008354,
+      "grad_norm": 2.005794048309326,
       "learning_rate": 0.00011797462538423614,
-      "loss": 0.0074,
+      "loss": 0.0115,
       "step": 2140
     },
     {
       "epoch": 1.3413416536661467,
-      "grad_norm": 0.16545473039150238,
+      "grad_norm": 0.22314704954624176,
       "learning_rate": 0.00011732892574398047,
-      "loss": 0.0092,
+      "loss": 0.013,
       "step": 2150
     },
     {
       "epoch": 1.3475819032761311,
-      "grad_norm": 0.004430962726473808,
+      "grad_norm": 0.13446640968322754,
       "learning_rate": 0.00011668248037623194,
-      "loss": 0.0037,
+      "loss": 0.0065,
       "step": 2160
     },
     {
       "epoch": 1.3538221528861154,
-      "grad_norm": 0.009139200672507286,
+      "grad_norm": 0.016804970800876617,
       "learning_rate": 0.00011603531709991544,
-      "loss": 0.0037,
+      "loss": 0.0027,
       "step": 2170
     },
     {
       "epoch": 1.3600624024960999,
-      "grad_norm": 0.010984491556882858,
+      "grad_norm": 0.1096780076622963,
       "learning_rate": 0.00011538746376485003,
-      "loss": 0.0087,
+      "loss": 0.0096,
       "step": 2180
     },
     {
       "epoch": 1.3663026521060844,
-      "grad_norm": 0.05745226517319679,
+      "grad_norm": 0.09347089380025864,
       "learning_rate": 0.00011473894825055071,
-      "loss": 0.0058,
+      "loss": 0.0093,
       "step": 2190
     },
     {
       "epoch": 1.3725429017160686,
-      "grad_norm": 0.047253601253032684,
+      "grad_norm": 0.0883338525891304,
       "learning_rate": 0.00011408979846502828,
-      "loss": 0.0068,
+      "loss": 0.0143,
       "step": 2200
     },
     {
       "epoch": 1.3787831513260531,
-      "grad_norm": 0.0017802199581637979,
+      "grad_norm": 0.04370776563882828,
       "learning_rate": 0.00011344004234358872,
-      "loss": 0.0045,
+      "loss": 0.0087,
       "step": 2210
     },
     {
       "epoch": 1.3850234009360374,
-      "grad_norm": 0.07970303297042847,
+      "grad_norm": 0.6470524072647095,
       "learning_rate": 0.00011278970784763079,
-      "loss": 0.004,
+      "loss": 0.0071,
       "step": 2220
     },
     {
       "epoch": 1.3912636505460219,
-      "grad_norm": 0.1246202290058136,
+      "grad_norm": 0.5653970241546631,
       "learning_rate": 0.00011213882296344289,
-      "loss": 0.0065,
+      "loss": 0.0145,
       "step": 2230
     },
     {
       "epoch": 1.3975039001560061,
-      "grad_norm": 0.01450931467115879,
+      "grad_norm": 0.6832361817359924,
       "learning_rate": 0.00011148741570099863,
-      "loss": 0.0046,
+      "loss": 0.0055,
       "step": 2240
     },
     {
       "epoch": 1.4037441497659906,
-      "grad_norm": 0.4136829078197479,
+      "grad_norm": 0.2704048752784729,
       "learning_rate": 0.0001108355140927515,
-      "loss": 0.0067,
+      "loss": 0.0051,
       "step": 2250
     },
     {
       "epoch": 1.409984399375975,
-      "grad_norm": 0.0039209057576954365,
+      "grad_norm": 0.3748556077480316,
       "learning_rate": 0.00011018314619242848,
-      "loss": 0.005,
+      "loss": 0.0075,
       "step": 2260
     },
     {
       "epoch": 1.4162246489859593,
-      "grad_norm": 0.0021903226152062416,
+      "grad_norm": 0.15553052723407745,
       "learning_rate": 0.00010953034007382281,
-      "loss": 0.0089,
+      "loss": 0.0079,
       "step": 2270
     },
     {
       "epoch": 1.4224648985959438,
-      "grad_norm": 0.0021560548339039087,
+      "grad_norm": 0.036521948873996735,
       "learning_rate": 0.0001088771238295859,
-      "loss": 0.0059,
+      "loss": 0.0105,
       "step": 2280
     },
     {
       "epoch": 1.4287051482059283,
-      "grad_norm": 0.16054560244083405,
+      "grad_norm": 0.37633374333381653,
       "learning_rate": 0.0001082235255700184,
-      "loss": 0.0037,
+      "loss": 0.0057,
       "step": 2290
     },
     {
       "epoch": 1.4349453978159126,
-      "grad_norm": 0.09998723119497299,
+      "grad_norm": 0.237517312169075,
       "learning_rate": 0.00010756957342186042,
-      "loss": 0.004,
+      "loss": 0.0065,
       "step": 2300
     },
     {
       "epoch": 1.441185647425897,
-      "grad_norm": 0.08871794492006302,
+      "grad_norm": 0.10721524804830551,
       "learning_rate": 0.00010691529552708128,
-      "loss": 0.0067,
+      "loss": 0.0051,
       "step": 2310
     },
     {
       "epoch": 1.4474258970358815,
-      "grad_norm": 0.005393527913838625,
+      "grad_norm": 0.15115052461624146,
       "learning_rate": 0.00010626072004166829,
-      "loss": 0.0061,
+      "loss": 0.0044,
       "step": 2320
     },
     {
       "epoch": 1.4536661466458658,
-      "grad_norm": 0.20176663994789124,
+      "grad_norm": 1.0222440958023071,
       "learning_rate": 0.00010560587513441524,
-      "loss": 0.0084,
+      "loss": 0.0117,
       "step": 2330
     },
     {
       "epoch": 1.4599063962558503,
-      "grad_norm": 0.25814759731292725,
+      "grad_norm": 0.1690666824579239,
       "learning_rate": 0.00010495078898571014,
-      "loss": 0.0046,
+      "loss": 0.0072,
       "step": 2340
     },
     {
       "epoch": 1.4661466458658348,
-      "grad_norm": 0.0187391247600317,
+      "grad_norm": 0.05320679396390915,
       "learning_rate": 0.00010429548978632246,
-      "loss": 0.0041,
+      "loss": 0.0063,
       "step": 2350
     },
     {
       "epoch": 1.472386895475819,
-      "grad_norm": 0.0017654044786468148,
+      "grad_norm": 0.1913842111825943,
       "learning_rate": 0.00010364000573619008,
-      "loss": 0.0025,
+      "loss": 0.0051,
       "step": 2360
     },
     {
       "epoch": 1.4786271450858035,
-      "grad_norm": 0.04109564423561096,
+      "grad_norm": 0.05571167916059494,
       "learning_rate": 0.00010298436504320561,
-      "loss": 0.0031,
+      "loss": 0.0048,
       "step": 2370
     },
     {
       "epoch": 1.4848673946957878,
-      "grad_norm": 0.0455956868827343,
+      "grad_norm": 0.06032731384038925,
       "learning_rate": 0.0001023285959220027,
-      "loss": 0.007,
+      "loss": 0.0099,
       "step": 2380
     },
     {
       "epoch": 1.4911076443057723,
-      "grad_norm": 0.0018099998123943806,
+      "grad_norm": 0.0039248778484761715,
       "learning_rate": 0.0001016727265927416,
-      "loss": 0.0026,
+      "loss": 0.0034,
       "step": 2390
     },
     {
       "epoch": 1.4973478939157565,
-      "grad_norm": 0.10770522803068161,
+      "grad_norm": 0.09997803717851639,
       "learning_rate": 0.000101016785279895,
-      "loss": 0.0062,
+      "loss": 0.0063,
       "step": 2400
     },
     {
       "epoch": 1.503588143525741,
-      "grad_norm": 0.04892813041806221,
+      "grad_norm": 0.060774434357881546,
       "learning_rate": 0.00010036080021103323,
-      "loss": 0.003,
+      "loss": 0.0039,
       "step": 2410
     },
     {
       "epoch": 1.5098283931357255,
-      "grad_norm": 0.02330232597887516,
+      "grad_norm": 0.5938266515731812,
       "learning_rate": 9.970479961560964e-05,
-      "loss": 0.0055,
+      "loss": 0.0057,
       "step": 2420
     },
     {
       "epoch": 1.5160686427457097,
-      "grad_norm": 0.03708835691213608,
+      "grad_norm": 0.0014718938618898392,
       "learning_rate": 9.904881172374576e-05,
-      "loss": 0.0029,
+      "loss": 0.0024,
       "step": 2430
     },
     {
       "epoch": 1.5223088923556942,
-      "grad_norm": 0.01838928461074829,
+      "grad_norm": 0.025481415912508965,
       "learning_rate": 9.839286476501635e-05,
-      "loss": 0.0038,
+      "loss": 0.0036,
       "step": 2440
     },
     {
       "epoch": 1.5285491419656787,
-      "grad_norm": 1.0069528818130493,
+      "grad_norm": 0.04258560761809349,
       "learning_rate": 9.773698696723481e-05,
-      "loss": 0.0063,
+      "loss": 0.0046,
       "step": 2450
     },
     {
       "epoch": 1.534789391575663,
-      "grad_norm": 1.22986900806427,
+      "grad_norm": 0.11182672530412674,
       "learning_rate": 9.70812065552382e-05,
-      "loss": 0.0069,
+      "loss": 0.0045,
       "step": 2460
     },
     {
       "epoch": 1.5410296411856474,
-      "grad_norm": 0.04500283673405647,
+      "grad_norm": 0.006588339805603027,
       "learning_rate": 9.642555174967268e-05,
-      "loss": 0.011,
+      "loss": 0.0041,
       "step": 2470
     },
     {
       "epoch": 1.547269890795632,
-      "grad_norm": 0.21551202237606049,
+      "grad_norm": 0.13866789638996124,
       "learning_rate": 9.577005076577915e-05,
-      "loss": 0.007,
+      "loss": 0.0056,
       "step": 2480
     },
     {
       "epoch": 1.5535101404056162,
-      "grad_norm": 0.2890082895755768,
+      "grad_norm": 0.3087863028049469,
       "learning_rate": 9.511473181217904e-05,
-      "loss": 0.0046,
+      "loss": 0.0045,
       "step": 2490
     },
     {
       "epoch": 1.5597503900156007,
-      "grad_norm": 0.014139365404844284,
+      "grad_norm": 0.07177337259054184,
       "learning_rate": 9.445962308966028e-05,
-      "loss": 0.0046,
+      "loss": 0.0041,
       "step": 2500
     },
     {
       "epoch": 1.5659906396255852,
-      "grad_norm": 0.0018333615735173225,
+      "grad_norm": 0.0008801668882369995,
       "learning_rate": 9.380475278996375e-05,
-      "loss": 0.0042,
+      "loss": 0.0034,
       "step": 2510
     },
     {
       "epoch": 1.5722308892355694,
-      "grad_norm": 0.01185754593461752,
+      "grad_norm": 0.0008780327043496072,
       "learning_rate": 9.315014909457023e-05,
-      "loss": 0.0032,
+      "loss": 0.0028,
       "step": 2520
     },
     {
       "epoch": 1.5784711388455537,
-      "grad_norm": 0.21326202154159546,
+      "grad_norm": 0.12729932367801666,
       "learning_rate": 9.249584017348748e-05,
-      "loss": 0.0055,
+      "loss": 0.0036,
       "step": 2530
     },
     {
       "epoch": 1.5847113884555384,
-      "grad_norm": 0.0019281799905002117,
+      "grad_norm": 0.00842339638620615,
       "learning_rate": 9.184185418403799e-05,
-      "loss": 0.0042,
+      "loss": 0.0043,
       "step": 2540
     },
     {
       "epoch": 1.5909516380655226,
-      "grad_norm": 0.0017161122523248196,
+      "grad_norm": 0.0009154107538051903,
       "learning_rate": 9.118821926964733e-05,
-      "loss": 0.0038,
+      "loss": 0.0036,
       "step": 2550
     },
     {
       "epoch": 1.597191887675507,
-      "grad_norm": 0.050887156277894974,
+      "grad_norm": 0.06446767598390579,
       "learning_rate": 9.053496355863311e-05,
       "loss": 0.0044,
       "step": 2560
     },
     {
       "epoch": 1.6034321372854914,
-      "grad_norm": 0.0016990028088912368,
+      "grad_norm": 0.0021474426612257957,
       "learning_rate": 8.988211516299434e-05,
-      "loss": 0.0024,
+      "loss": 0.0026,
       "step": 2570
     },
     {
       "epoch": 1.6096723868954759,
-      "grad_norm": 0.1756485253572464,
+      "grad_norm": 0.1125708743929863,
       "learning_rate": 8.92297021772017e-05,
       "loss": 0.005,
       "step": 2580
     },
     {
       "epoch": 1.6159126365054601,
-      "grad_norm": 0.0019257533131167293,
+      "grad_norm": 0.002130712615326047,
       "learning_rate": 8.857775267698869e-05,
-      "loss": 0.004,
+      "loss": 0.0039,
       "step": 2590
     },
     {
       "epoch": 1.6221528861154446,
-      "grad_norm": 0.047810349613428116,
+      "grad_norm": 0.023599926382303238,
       "learning_rate": 8.792629471814325e-05,
-      "loss": 0.0028,
+      "loss": 0.0027,
       "step": 2600
     },
     {
       "epoch": 1.628393135725429,
-      "grad_norm": 0.0032332607079297304,
+      "grad_norm": 0.0017673929687589407,
       "learning_rate": 8.727535633530057e-05,
-      "loss": 0.003,
+      "loss": 0.0028,
       "step": 2610
     },
     {
       "epoch": 1.6346333853354134,
-      "grad_norm": 0.0024718448985368013,
+      "grad_norm": 0.0008239516173489392,
       "learning_rate": 8.662496554073642e-05,
-      "loss": 0.0044,
+      "loss": 0.0049,
       "step": 2620
     },
     {
       "epoch": 1.6408736349453978,
-      "grad_norm": 0.0009530312963761389,
+      "grad_norm": 0.0014444030821323395,
       "learning_rate": 8.597515032316197e-05,
       "loss": 0.0022,
       "step": 2630
     },
     {
       "epoch": 1.6471138845553823,
-      "grad_norm": 0.05391823127865791,
+      "grad_norm": 0.05863706022500992,
       "learning_rate": 8.532593864651913e-05,
-      "loss": 0.0039,
+      "loss": 0.004,
       "step": 2640
     },
     {
       "epoch": 1.6533541341653666,
-      "grad_norm": 0.20347720384597778,
+      "grad_norm": 0.18055737018585205,
       "learning_rate": 8.467735844877717e-05,
-      "loss": 0.0048,
+      "loss": 0.0047,
       "step": 2650
     },
     {
       "epoch": 1.659594383775351,
-      "grad_norm": 0.215189591050148,
+      "grad_norm": 0.18026234209537506,
       "learning_rate": 8.402943764073064e-05,
-      "loss": 0.0059,
+      "loss": 0.0053,
       "step": 2660
     },
     {
       "epoch": 1.6658346333853356,
-      "grad_norm": 0.008820041082799435,
+      "grad_norm": 0.0015067998319864273,
       "learning_rate": 8.338220410479799e-05,
-      "loss": 0.0047,
+      "loss": 0.0043,
       "step": 2670
     },
     {
       "epoch": 1.6720748829953198,
-      "grad_norm": 0.048564255237579346,
+      "grad_norm": 0.03375815600156784,
       "learning_rate": 8.273568569382193e-05,
-      "loss": 0.0031,
+      "loss": 0.0029,
       "step": 2680
     },
     {
       "epoch": 1.678315132605304,
-      "grad_norm": 0.0026437535416334867,
+      "grad_norm": 0.0008590230136178434,
       "learning_rate": 8.208991022987057e-05,
-      "loss": 0.0042,
+      "loss": 0.0041,
       "step": 2690
     },
     {
       "epoch": 1.6845553822152888,
-      "grad_norm": 0.19647392630577087,
+      "grad_norm": 0.17761924862861633,
       "learning_rate": 8.144490550304038e-05,
-      "loss": 0.0024,
+      "loss": 0.0023,
       "step": 2700
     },
     {
       "epoch": 1.690795631825273,
-      "grad_norm": 0.0007143426337279379,
+      "grad_norm": 0.0009983059717342257,
       "learning_rate": 8.080069927026018e-05,
       "loss": 0.0038,
       "step": 2710
     },
     {
       "epoch": 1.6970358814352573,
-      "grad_norm": 0.1795455664396286,
+      "grad_norm": 0.19785909354686737,
       "learning_rate": 8.015731925409661e-05,
-      "loss": 0.0061,
+      "loss": 0.0062,
       "step": 2720
     },
     {
       "epoch": 1.7032761310452418,
-      "grad_norm": 0.0010726724285632372,
+      "grad_norm": 0.001470402698032558,
       "learning_rate": 7.951479314156113e-05,
-      "loss": 0.005,
+      "loss": 0.0051,
       "step": 2730
     },
     {
       "epoch": 1.7095163806552263,
-      "grad_norm": 0.1190568059682846,
+      "grad_norm": 0.08547449111938477,
       "learning_rate": 7.887314858291865e-05,
-      "loss": 0.0038,
+      "loss": 0.0039,
       "step": 2740
     },
     {
       "epoch": 1.7157566302652105,
-      "grad_norm": 0.07289066165685654,
+      "grad_norm": 0.07325121760368347,
       "learning_rate": 7.823241319049756e-05,
-      "loss": 0.0028,
+      "loss": 0.0029,
       "step": 2750
     },
     {
       "epoch": 1.721996879875195,
-      "grad_norm": 0.060338374227285385,
+      "grad_norm": 0.06296833604574203,
       "learning_rate": 7.759261453750146e-05,
       "loss": 0.0039,
       "step": 2760
     },
     {
       "epoch": 1.7282371294851795,
-      "grad_norm": 0.22540676593780518,
+      "grad_norm": 0.18311767280101776,
       "learning_rate": 7.695378015682256e-05,
       "loss": 0.007,
       "step": 2770
     },
     {
       "epoch": 1.7344773790951638,
-      "grad_norm": 0.11157722026109695,
+      "grad_norm": 0.09669390320777893,
       "learning_rate": 7.631593753985699e-05,
-      "loss": 0.0029,
+      "loss": 0.003,
       "step": 2780
     },
     {
       "epoch": 1.7407176287051482,
-      "grad_norm": 0.11412902176380157,
+      "grad_norm": 0.08832534402608871,
       "learning_rate": 7.567911413532156e-05,
-      "loss": 0.0038,
+      "loss": 0.004,
       "step": 2790
     },
     {
       "epoch": 1.7469578783151327,
-      "grad_norm": 0.03782040998339653,
+      "grad_norm": 0.008018124848604202,
       "learning_rate": 7.504333734807258e-05,
-      "loss": 0.0038,
+      "loss": 0.0036,
       "step": 2800
     },
     {
       "epoch": 1.753198127925117,
-      "grad_norm": 0.10296830534934998,
+      "grad_norm": 0.09240499883890152,
       "learning_rate": 7.440863453792665e-05,
-      "loss": 0.0045,
+      "loss": 0.0046,
       "step": 2810
     },
     {
       "epoch": 1.7594383775351012,
-      "grad_norm": 0.0008850370650179684,
+      "grad_norm": 0.000288833660306409,
       "learning_rate": 7.377503301848315e-05,
-      "loss": 0.0029,
+      "loss": 0.0028,
       "step": 2820
     },
     {
       "epoch": 1.765678627145086,
-      "grad_norm": 0.0012563092168420553,
+      "grad_norm": 0.001008062856271863,
       "learning_rate": 7.314256005594885e-05,
-      "loss": 0.0068,
+      "loss": 0.0067,
       "step": 2830
     },
     {
       "epoch": 1.7719188767550702,
-      "grad_norm": 0.08966930955648422,
+      "grad_norm": 0.08035475015640259,
       "learning_rate": 7.251124286796455e-05,
-      "loss": 0.0036,
+      "loss": 0.0034,
       "step": 2840
     },
     {
       "epoch": 1.7781591263650545,
-      "grad_norm": 0.006013158708810806,
+      "grad_norm": 0.003968788310885429,
       "learning_rate": 7.188110862243384e-05,
       "loss": 0.002,
       "step": 2850
     },
     {
       "epoch": 1.784399375975039,
-      "grad_norm": 0.07801369577646255,
+      "grad_norm": 0.06475794315338135,
       "learning_rate": 7.125218443635393e-05,
-      "loss": 0.0045,
+      "loss": 0.0046,
       "step": 2860
     },
     {
       "epoch": 1.7906396255850234,
-      "grad_norm": 0.04352996125817299,
+      "grad_norm": 0.04635635018348694,
       "learning_rate": 7.062449737464877e-05,
       "loss": 0.0047,
       "step": 2870
     },
     {
       "epoch": 1.7968798751950077,
-      "grad_norm": 0.011117925867438316,
+      "grad_norm": 0.008212746120989323,
       "learning_rate": 6.999807444900416e-05,
-      "loss": 0.0019,
+      "loss": 0.0021,
       "step": 2880
     },
     {
       "epoch": 1.8031201248049922,
-      "grad_norm": 0.0019138542702421546,
+      "grad_norm": 0.000851539196446538,
       "learning_rate": 6.937294261670558e-05,
-      "loss": 0.0021,
+      "loss": 0.002,
       "step": 2890
     },
     {
       "epoch": 1.8093603744149767,
-      "grad_norm": 0.0006595162558369339,
+      "grad_norm": 0.0004067673289682716,
       "learning_rate": 6.874912877947801e-05,
-      "loss": 0.0023,
+      "loss": 0.0025,
       "step": 2900
     },
     {
       "epoch": 1.815600624024961,
-      "grad_norm": 0.0925321877002716,
+      "grad_norm": 0.08847030997276306,
       "learning_rate": 6.812665978232814e-05,
       "loss": 0.0043,
       "step": 2910
     },
     {
       "epoch": 1.8218408736349454,
-      "grad_norm": 0.0005468401359394193,
+      "grad_norm": 0.0008237504516728222,
       "learning_rate": 6.750556241238938e-05,
-      "loss": 0.0033,
+      "loss": 0.0032,
       "step": 2920
     },
     {
       "epoch": 1.8280811232449299,
-      "grad_norm": 0.037381552159786224,
+      "grad_norm": 0.004662049934267998,
       "learning_rate": 6.688586339776886e-05,
-      "loss": 0.0058,
+      "loss": 0.0055,
       "step": 2930
     },
     {
       "epoch": 1.8343213728549141,
-      "grad_norm": 0.002739663003012538,
+      "grad_norm": 0.0011368844425305724,
       "learning_rate": 6.626758940639745e-05,
       "loss": 0.0042,
       "step": 2940
     },
     {
       "epoch": 1.8405616224648986,
-      "grad_norm": 0.11427415907382965,
+      "grad_norm": 0.05946766957640648,
       "learning_rate": 6.565076704488187e-05,
-      "loss": 0.004,
+      "loss": 0.0027,
       "step": 2950
     },
     {
       "epoch": 1.846801872074883,
-      "grad_norm": 0.03485134243965149,
+      "grad_norm": 0.03306344151496887,
       "learning_rate": 6.503542285735994e-05,
-      "loss": 0.0048,
+      "loss": 0.0043,
       "step": 2960
     },
     {
       "epoch": 1.8530421216848674,
-      "grad_norm": 0.009942654520273209,
+      "grad_norm": 0.0018961625173687935,
       "learning_rate": 6.442158332435826e-05,
-      "loss": 0.004,
+      "loss": 0.0033,
       "step": 2970
     },
     {
       "epoch": 1.8592823712948516,
-      "grad_norm": 0.06687472760677338,
+      "grad_norm": 0.012175627052783966,
       "learning_rate": 6.380927486165253e-05,
-      "loss": 0.0059,
+      "loss": 0.0056,
       "step": 2980
     },
     {
       "epoch": 1.8655226209048363,
-      "grad_norm": 0.002326112939044833,
+      "grad_norm": 0.00032657431438565254,
       "learning_rate": 6.319852381913087e-05,
-      "loss": 0.0041,
+      "loss": 0.0038,
       "step": 2990
     },
     {
       "epoch": 1.8717628705148206,
-      "grad_norm": 0.002811886603012681,
+      "grad_norm": 0.0009256178745999932,
       "learning_rate": 6.258935647965987e-05,
-      "loss": 0.0054,
+      "loss": 0.005,
       "step": 3000
+    },
+    {
+      "epoch": 1.8780031201248049,
+      "grad_norm": 0.0014349141856655478,
+      "learning_rate": 6.198179905795358e-05,
+      "loss": 0.0032,
+      "step": 3010
+    },
+    {
+      "epoch": 1.8842433697347893,
+      "grad_norm": 0.01866261102259159,
+      "learning_rate": 6.137587769944536e-05,
+      "loss": 0.0029,
+      "step": 3020
+    },
+    {
+      "epoch": 1.8904836193447738,
+      "grad_norm": 0.0008336464525200427,
+      "learning_rate": 6.0771618479162664e-05,
+      "loss": 0.0044,
+      "step": 3030
+    },
+    {
+      "epoch": 1.896723868954758,
+      "grad_norm": 0.0008074100478552282,
+      "learning_rate": 6.016904740060511e-05,
+      "loss": 0.0027,
+      "step": 3040
+    },
+    {
+      "epoch": 1.9029641185647426,
+      "grad_norm": 0.08711811900138855,
+      "learning_rate": 5.956819039462535e-05,
+      "loss": 0.0031,
+      "step": 3050
+    },
+    {
+      "epoch": 1.909204368174727,
+      "grad_norm": 0.0007693839725106955,
+      "learning_rate": 5.896907331831311e-05,
+      "loss": 0.0049,
+      "step": 3060
+    },
+    {
+      "epoch": 1.9154446177847113,
+      "grad_norm": 0.008787058293819427,
+      "learning_rate": 5.8371721953882575e-05,
+      "loss": 0.0032,
+      "step": 3070
+    },
+    {
+      "epoch": 1.9216848673946958,
+      "grad_norm": 0.0189735759049654,
+      "learning_rate": 5.7776162007562886e-05,
+      "loss": 0.0011,
+      "step": 3080
+    },
+    {
+      "epoch": 1.9279251170046803,
+      "grad_norm": 0.0007118944195099175,
+      "learning_rate": 5.718241910849187e-05,
+      "loss": 0.0038,
+      "step": 3090
+    },
+    {
+      "epoch": 1.9341653666146645,
+      "grad_norm": 0.0007943984237499535,
+      "learning_rate": 5.6590518807613026e-05,
+      "loss": 0.0024,
+      "step": 3100
+    },
+    {
+      "epoch": 1.940405616224649,
+      "grad_norm": 0.0037727924063801765,
+      "learning_rate": 5.6000486576576095e-05,
+      "loss": 0.0042,
+      "step": 3110
+    },
+    {
+      "epoch": 1.9466458658346335,
+      "grad_norm": 0.09510115534067154,
+      "learning_rate": 5.5412347806640955e-05,
+      "loss": 0.003,
+      "step": 3120
+    },
+    {
+      "epoch": 1.9528861154446178,
+      "grad_norm": 0.10276944190263748,
+      "learning_rate": 5.482612780758487e-05,
+      "loss": 0.0045,
+      "step": 3130
+    },
+    {
+      "epoch": 1.959126365054602,
+      "grad_norm": 0.0779290720820427,
+      "learning_rate": 5.42418518066133e-05,
+      "loss": 0.0047,
+      "step": 3140
+    },
+    {
+      "epoch": 1.9653666146645867,
+      "grad_norm": 0.0010850824182853103,
+      "learning_rate": 5.365954494727422e-05,
+      "loss": 0.0037,
+      "step": 3150
+    },
+    {
+      "epoch": 1.971606864274571,
+      "grad_norm": 0.0005516468081623316,
+      "learning_rate": 5.307923228837639e-05,
+      "loss": 0.0028,
+      "step": 3160
+    },
+    {
+      "epoch": 1.9778471138845553,
+      "grad_norm": 0.22346067428588867,
+      "learning_rate": 5.250093880291058e-05,
+      "loss": 0.0036,
+      "step": 3170
+    },
+    {
+      "epoch": 1.9840873634945397,
+      "grad_norm": 0.1451890766620636,
+      "learning_rate": 5.192468937697521e-05,
+      "loss": 0.0038,
+      "step": 3180
+    },
+    {
+      "epoch": 1.9903276131045242,
+      "grad_norm": 0.000913795200176537,
+      "learning_rate": 5.1350508808705314e-05,
+      "loss": 0.0032,
+      "step": 3190
+    },
+    {
+      "epoch": 1.9965678627145085,
+      "grad_norm": 0.04572800546884537,
+      "learning_rate": 5.077842180720527e-05,
+      "loss": 0.0042,
+      "step": 3200
+    },
+    {
+      "epoch": 2.002496099843994,
+      "grad_norm": 0.0017076911171898246,
+      "learning_rate": 5.020845299148555e-05,
+      "loss": 0.0026,
+      "step": 3210
+    },
+    {
+      "epoch": 2.008736349453978,
+      "grad_norm": 0.0012423003790900111,
+      "learning_rate": 4.96406268894034e-05,
+      "loss": 0.0019,
+      "step": 3220
+    },
+    {
+      "epoch": 2.0149765990639628,
+      "grad_norm": 0.09963550418615341,
+      "learning_rate": 4.907496793660717e-05,
+      "loss": 0.0063,
+      "step": 3230
+    },
+    {
+      "epoch": 2.021216848673947,
+      "grad_norm": 0.0007362994365394115,
+      "learning_rate": 4.8511500475484775e-05,
+      "loss": 0.0044,
+      "step": 3240
+    },
+    {
+      "epoch": 2.0274570982839313,
+      "grad_norm": 0.00047720750444568694,
+      "learning_rate": 4.795024875411612e-05,
+      "loss": 0.0025,
+      "step": 3250
+    },
+    {
+      "epoch": 2.0336973478939155,
+      "grad_norm": 0.0012651053257286549,
+      "learning_rate": 4.7391236925229755e-05,
+      "loss": 0.0033,
+      "step": 3260
+    },
+    {
+      "epoch": 2.0399375975039002,
+      "grad_norm": 0.09363336861133575,
+      "learning_rate": 4.68344890451634e-05,
+      "loss": 0.0052,
+      "step": 3270
+    },
+    {
+      "epoch": 2.0461778471138845,
+      "grad_norm": 0.04909437894821167,
+      "learning_rate": 4.628002907282871e-05,
+      "loss": 0.0032,
+      "step": 3280
+    },
+    {
+      "epoch": 2.0524180967238688,
+      "grad_norm": 0.06087338179349899,
+      "learning_rate": 4.572788086868021e-05,
+      "loss": 0.0047,
+      "step": 3290
+    },
+    {
+      "epoch": 2.0586583463338535,
+      "grad_norm": 0.002108914777636528,
+      "learning_rate": 4.51780681936885e-05,
+      "loss": 0.0031,
+      "step": 3300
+    },
+    {
+      "epoch": 2.0648985959438377,
+      "grad_norm": 0.001688466058112681,
+      "learning_rate": 4.4630614708317896e-05,
+      "loss": 0.0043,
+      "step": 3310
+    },
+    {
+      "epoch": 2.071138845553822,
+      "grad_norm": 0.19162900745868683,
+      "learning_rate": 4.408554397150792e-05,
+      "loss": 0.0052,
+      "step": 3320
+    },
+    {
+      "epoch": 2.0773790951638067,
+      "grad_norm": 0.09422574192285538,
+      "learning_rate": 4.3542879439659826e-05,
+      "loss": 0.0028,
+      "step": 3330
+    },
+    {
+      "epoch": 2.083619344773791,
+      "grad_norm": 0.000398104777559638,
+      "learning_rate": 4.300264446562682e-05,
+      "loss": 0.003,
+      "step": 3340
+    },
+    {
+      "epoch": 2.089859594383775,
+      "grad_norm": 0.16255362331867218,
+      "learning_rate": 4.2464862297709465e-05,
+      "loss": 0.0045,
+      "step": 3350
+    },
+    {
+      "epoch": 2.09609984399376,
+      "grad_norm": 0.0003849667846225202,
+      "learning_rate": 4.192955607865491e-05,
+      "loss": 0.0026,
+      "step": 3360
+    },
+    {
+      "epoch": 2.102340093603744,
+      "grad_norm": 0.16027703881263733,
+      "learning_rate": 4.139674884466117e-05,
+      "loss": 0.0043,
+      "step": 3370
+    },
+    {
+      "epoch": 2.1085803432137284,
+      "grad_norm": 0.09113962948322296,
+      "learning_rate": 4.086646352438577e-05,
+      "loss": 0.0041,
+      "step": 3380
+    },
+    {
+      "epoch": 2.114820592823713,
+      "grad_norm": 0.028218964114785194,
+      "learning_rate": 4.0338722937958906e-05,
+      "loss": 0.0049,
+      "step": 3390
+    },
+    {
+      "epoch": 2.1210608424336974,
+      "grad_norm": 0.13162770867347717,
+      "learning_rate": 3.9813549796001525e-05,
+      "loss": 0.0034,
+      "step": 3400
+    },
+    {
+      "epoch": 2.1273010920436817,
+      "grad_norm": 0.0004251877253409475,
+      "learning_rate": 3.929096669864802e-05,
+      "loss": 0.0027,
+      "step": 3410
+    },
+    {
+      "epoch": 2.1335413416536664,
+      "grad_norm": 0.0011246739886701107,
+      "learning_rate": 3.8770996134573675e-05,
+      "loss": 0.0035,
+      "step": 3420
+    },
+    {
+      "epoch": 2.1397815912636506,
+      "grad_norm": 0.20078203082084656,
+      "learning_rate": 3.825366048002673e-05,
+      "loss": 0.0046,
+      "step": 3430
+    },
+    {
+      "epoch": 2.146021840873635,
+      "grad_norm": 0.21198423206806183,
+      "learning_rate": 3.7738981997865595e-05,
+      "loss": 0.0027,
+      "step": 3440
+    },
+    {
+      "epoch": 2.152262090483619,
+      "grad_norm": 0.06202436983585358,
+      "learning_rate": 3.722698283660081e-05,
+      "loss": 0.0031,
+      "step": 3450
+    },
+    {
+      "epoch": 2.158502340093604,
+      "grad_norm": 0.07747945934534073,
+      "learning_rate": 3.6717685029441884e-05,
+      "loss": 0.0026,
+      "step": 3460
+    },
+    {
+      "epoch": 2.164742589703588,
+      "grad_norm": 0.12237684428691864,
+      "learning_rate": 3.621111049334905e-05,
+      "loss": 0.002,
+      "step": 3470
+    },
+    {
+      "epoch": 2.1709828393135724,
+      "grad_norm": 0.001747419242747128,
+      "learning_rate": 3.570728102809016e-05,
+      "loss": 0.0051,
+      "step": 3480
+    },
+    {
+      "epoch": 2.177223088923557,
+      "grad_norm": 0.05137491598725319,
+      "learning_rate": 3.5206218315302585e-05,
+      "loss": 0.0044,
+      "step": 3490
+    },
+    {
+      "epoch": 2.1834633385335414,
+      "grad_norm": 0.0013341855956241488,
+      "learning_rate": 3.47079439175602e-05,
+      "loss": 0.0011,
+      "step": 3500
+    },
+    {
+      "epoch": 2.1897035881435256,
+      "grad_norm": 0.11468301713466644,
+      "learning_rate": 3.4212479277445283e-05,
+      "loss": 0.0053,
+      "step": 3510
+    },
+    {
+      "epoch": 2.1959438377535103,
+      "grad_norm": 0.04506004974246025,
+      "learning_rate": 3.371984571662605e-05,
+      "loss": 0.0038,
+      "step": 3520
+    },
+    {
+      "epoch": 2.2021840873634946,
+      "grad_norm": 0.0011585105676203966,
+      "learning_rate": 3.323006443493879e-05,
+      "loss": 0.0011,
+      "step": 3530
+    },
+    {
+      "epoch": 2.208424336973479,
+      "grad_norm": 0.0005640931776724756,
+      "learning_rate": 3.274315650947587e-05,
+      "loss": 0.0032,
+      "step": 3540
+    },
+    {
+      "epoch": 2.2146645865834635,
+      "grad_norm": 0.0019072700524702668,
+      "learning_rate": 3.225914289367841e-05,
+      "loss": 0.0023,
+      "step": 3550
+    },
+    {
+      "epoch": 2.220904836193448,
+      "grad_norm": 0.016394929960370064,
+      "learning_rate": 3.177804441643486e-05,
+      "loss": 0.0029,
+      "step": 3560
+    },
+    {
+      "epoch": 2.227145085803432,
+      "grad_norm": 0.10553210973739624,
+      "learning_rate": 3.129988178118444e-05,
+      "loss": 0.0042,
+      "step": 3570
+    },
+    {
+      "epoch": 2.2333853354134163,
+      "grad_norm": 0.051088664680719376,
+      "learning_rate": 3.0824675565026265e-05,
+      "loss": 0.003,
+      "step": 3580
+    },
+    {
+      "epoch": 2.239625585023401,
+      "grad_norm": 0.0003461836895439774,
+      "learning_rate": 3.0352446217833896e-05,
+      "loss": 0.0015,
+      "step": 3590
+    },
+    {
+      "epoch": 2.2458658346333853,
+      "grad_norm": 0.0006262848619371653,
+      "learning_rate": 2.9883214061375276e-05,
+      "loss": 0.0024,
+      "step": 3600
+    },
+    {
+      "epoch": 2.2521060842433696,
+      "grad_norm": 0.08731874078512192,
+      "learning_rate": 2.941699928843816e-05,
+      "loss": 0.0043,
+      "step": 3610
+    },
+    {
+      "epoch": 2.2583463338533543,
+      "grad_norm": 0.003624439239501953,
+      "learning_rate": 2.895382196196117e-05,
+      "loss": 0.0036,
+      "step": 3620
+    },
+    {
+      "epoch": 2.2645865834633385,
+      "grad_norm": 0.06561044603586197,
+      "learning_rate": 2.8493702014170397e-05,
+      "loss": 0.0031,
+      "step": 3630
+    },
+    {
+      "epoch": 2.270826833073323,
+      "grad_norm": 0.0018900239374488592,
+      "learning_rate": 2.803665924572173e-05,
+      "loss": 0.0049,
+      "step": 3640
+    },
+    {
+      "epoch": 2.2770670826833075,
+      "grad_norm": 0.09807489067316055,
+      "learning_rate": 2.7582713324848687e-05,
+      "loss": 0.0038,
+      "step": 3650
+    },
+    {
+      "epoch": 2.2833073322932917,
+      "grad_norm": 0.000893085787538439,
+      "learning_rate": 2.7131883786515998e-05,
+      "loss": 0.0032,
+      "step": 3660
+    },
+    {
+      "epoch": 2.289547581903276,
+      "grad_norm": 0.07783768326044083,
+      "learning_rate": 2.668419003157896e-05,
+      "loss": 0.0027,
+      "step": 3670
+    },
+    {
+      "epoch": 2.2957878315132607,
+      "grad_norm": 0.0009885120671242476,
+      "learning_rate": 2.6239651325948634e-05,
+      "loss": 0.0045,
+      "step": 3680
+    },
+    {
+      "epoch": 2.302028081123245,
+      "grad_norm": 0.0022323562297970057,
+      "learning_rate": 2.5798286799762615e-05,
+      "loss": 0.0025,
+      "step": 3690
+    },
+    {
+      "epoch": 2.3082683307332292,
+      "grad_norm": 0.043047044426202774,
+      "learning_rate": 2.5360115446561905e-05,
+      "loss": 0.004,
+      "step": 3700
+    },
+    {
+      "epoch": 2.3145085803432135,
+      "grad_norm": 0.1055779680609703,
+      "learning_rate": 2.4925156122473547e-05,
+      "loss": 0.0037,
+      "step": 3710
+    },
+    {
+      "epoch": 2.320748829953198,
+      "grad_norm": 0.04654201865196228,
+      "learning_rate": 2.4493427545399072e-05,
+      "loss": 0.0037,
+      "step": 3720
+    },
+    {
+      "epoch": 2.3269890795631825,
+      "grad_norm": 0.0010084593668580055,
+      "learning_rate": 2.4064948294209076e-05,
+      "loss": 0.0038,
+      "step": 3730
+    },
+    {
+      "epoch": 2.333229329173167,
+      "grad_norm": 0.029136206954717636,
+      "learning_rate": 2.3639736807943747e-05,
+      "loss": 0.0041,
+      "step": 3740
+    },
+    {
+      "epoch": 2.3394695787831514,
+      "grad_norm": 0.0008335582679137588,
+      "learning_rate": 2.3217811385019307e-05,
+      "loss": 0.0045,
+      "step": 3750
+    },
+    {
+      "epoch": 2.3457098283931357,
+      "grad_norm": 0.13343648612499237,
+      "learning_rate": 2.2799190182440522e-05,
+      "loss": 0.0038,
+      "step": 3760
+    },
+    {
+      "epoch": 2.35195007800312,
+      "grad_norm": 0.0021896804682910442,
+      "learning_rate": 2.2383891215019405e-05,
+      "loss": 0.0026,
+      "step": 3770
+    },
+    {
+      "epoch": 2.3581903276131047,
+      "grad_norm": 0.026730315759778023,
+      "learning_rate": 2.1971932354599965e-05,
+      "loss": 0.0041,
+      "step": 3780
+    },
+    {
+      "epoch": 2.364430577223089,
+      "grad_norm": 0.0026145121082663536,
+      "learning_rate": 2.156333132928917e-05,
+      "loss": 0.0019,
+      "step": 3790
+    },
+    {
+      "epoch": 2.370670826833073,
+      "grad_norm": 0.059362031519412994,
+      "learning_rate": 2.1158105722693854e-05,
+      "loss": 0.0025,
+      "step": 3800
+    },
+    {
+      "epoch": 2.376911076443058,
+      "grad_norm": 0.001798074459657073,
+      "learning_rate": 2.0756272973164203e-05,
+      "loss": 0.0039,
+      "step": 3810
+    },
+    {
+      "epoch": 2.383151326053042,
+      "grad_norm": 0.054778262972831726,
+      "learning_rate": 2.035785037304331e-05,
+      "loss": 0.0043,
+      "step": 3820
+    },
+    {
+      "epoch": 2.3893915756630264,
+      "grad_norm": 0.0016084830276668072,
+      "learning_rate": 1.9962855067922957e-05,
+      "loss": 0.0029,
+      "step": 3830
+    },
+    {
+      "epoch": 2.395631825273011,
+      "grad_norm": 0.1474762260913849,
+      "learning_rate": 1.9571304055905758e-05,
+      "loss": 0.0061,
+      "step": 3840
+    },
+    {
+      "epoch": 2.4018720748829954,
+      "grad_norm": 0.10790376365184784,
+      "learning_rate": 1.9183214186873775e-05,
+      "loss": 0.0026,
+      "step": 3850
+    },
+    {
+      "epoch": 2.4081123244929796,
+      "grad_norm": 0.0012760792160406709,
+      "learning_rate": 1.879860216176329e-05,
+      "loss": 0.0038,
+      "step": 3860
+    },
+    {
+      "epoch": 2.4143525741029643,
+      "grad_norm": 0.06327309459447861,
+      "learning_rate": 1.8417484531846198e-05,
+      "loss": 0.0018,
+      "step": 3870
+    },
+    {
+      "epoch": 2.4205928237129486,
+      "grad_norm": 0.025750093162059784,
+      "learning_rate": 1.8039877698017636e-05,
+      "loss": 0.0022,
+      "step": 3880
+    },
+    {
+      "epoch": 2.426833073322933,
+      "grad_norm": 0.0012120454339310527,
+      "learning_rate": 1.7665797910090343e-05,
+      "loss": 0.0026,
+      "step": 3890
+    },
+    {
+      "epoch": 2.433073322932917,
+      "grad_norm": 0.002105908002704382,
+      "learning_rate": 1.7295261266095174e-05,
+      "loss": 0.0022,
+      "step": 3900
+    },
+    {
+      "epoch": 2.439313572542902,
+      "grad_norm": 0.07578789442777634,
+      "learning_rate": 1.692828371158859e-05,
+      "loss": 0.0041,
+      "step": 3910
+    },
+    {
+      "epoch": 2.445553822152886,
+      "grad_norm": 0.02772875316441059,
+      "learning_rate": 1.6564881038966175e-05,
+      "loss": 0.0021,
+      "step": 3920
+    },
+    {
+      "epoch": 2.4517940717628703,
+      "grad_norm": 0.12264281511306763,
+      "learning_rate": 1.6205068886783283e-05,
+      "loss": 0.0027,
+      "step": 3930
+    },
+    {
+      "epoch": 2.458034321372855,
+      "grad_norm": 0.0009135138243436813,
+      "learning_rate": 1.584886273908195e-05,
+      "loss": 0.0042,
+      "step": 3940
+    },
+    {
+      "epoch": 2.4642745709828393,
+      "grad_norm": 0.13282805681228638,
+      "learning_rate": 1.54962779247245e-05,
+      "loss": 0.0068,
+      "step": 3950
+    },
+    {
+      "epoch": 2.4705148205928236,
+      "grad_norm": 0.0021918078418821096,
+      "learning_rate": 1.5147329616733951e-05,
+      "loss": 0.0039,
+      "step": 3960
+    },
+    {
+      "epoch": 2.4767550702028083,
+      "grad_norm": 0.030358005315065384,
+      "learning_rate": 1.4802032831641121e-05,
+      "loss": 0.0022,
+      "step": 3970
+    },
+    {
+      "epoch": 2.4829953198127925,
+      "grad_norm": 0.10342071950435638,
+      "learning_rate": 1.4460402428838326e-05,
+      "loss": 0.0027,
+      "step": 3980
+    },
+    {
+      "epoch": 2.489235569422777,
+      "grad_norm": 0.11133008450269699,
+      "learning_rate": 1.4122453109939915e-05,
+      "loss": 0.0033,
+      "step": 3990
+    },
+    {
+      "epoch": 2.4954758190327615,
+      "grad_norm": 0.0008142650476656854,
+      "learning_rate": 1.3788199418149639e-05,
+      "loss": 0.0045,
+      "step": 4000
+    },
+    {
+      "epoch": 2.5017160686427458,
+      "grad_norm": 0.04364210367202759,
+      "learning_rate": 1.3457655737634866e-05,
+      "loss": 0.0031,
+      "step": 4010
+    },
+    {
+      "epoch": 2.50795631825273,
+      "grad_norm": 0.0011525236768648028,
+      "learning_rate": 1.3130836292907478e-05,
+      "loss": 0.0016,
+      "step": 4020
+    },
+    {
+      "epoch": 2.5141965678627143,
+      "grad_norm": 0.02383251115679741,
+      "learning_rate": 1.2807755148211741e-05,
+      "loss": 0.003,
+      "step": 4030
+    },
+    {
+      "epoch": 2.520436817472699,
+      "grad_norm": 0.0011006222339347005,
+      "learning_rate": 1.248842620691919e-05,
+      "loss": 0.0038,
+      "step": 4040
+    },
+    {
+      "epoch": 2.5266770670826832,
+      "grad_norm": 0.0005220650346018374,
+      "learning_rate": 1.2172863210930152e-05,
+      "loss": 0.0027,
+      "step": 4050
+    },
+    {
+      "epoch": 2.532917316692668,
+      "grad_norm": 0.11079895496368408,
+      "learning_rate": 1.1861079740082447e-05,
+      "loss": 0.0036,
+      "step": 4060
+    },
+    {
+      "epoch": 2.539157566302652,
+      "grad_norm": 0.08487123996019363,
+      "learning_rate": 1.1553089211567091e-05,
+      "loss": 0.0082,
+      "step": 4070
+    },
+    {
+      "epoch": 2.5453978159126365,
+      "grad_norm": 0.0006430468638427556,
+      "learning_rate": 1.1248904879350774e-05,
+      "loss": 0.0033,
+      "step": 4080
+    },
+    {
+      "epoch": 2.5516380655226207,
+      "grad_norm": 0.0645022988319397,
+      "learning_rate": 1.0948539833605553e-05,
+      "loss": 0.0045,
+      "step": 4090
+    },
+    {
+      "epoch": 2.5578783151326054,
+      "grad_norm": 0.001119063002988696,
+      "learning_rate": 1.065200700014547e-05,
+      "loss": 0.0031,
+      "step": 4100
+    },
+    {
+      "epoch": 2.5641185647425897,
+      "grad_norm": 0.027760177850723267,
+      "learning_rate": 1.0359319139870438e-05,
+      "loss": 0.0054,
+      "step": 4110
+    },
+    {
+      "epoch": 2.570358814352574,
+      "grad_norm": 0.04383605346083641,
+      "learning_rate": 1.0070488848216986e-05,
+      "loss": 0.0028,
+      "step": 4120
+    },
+    {
+      "epoch": 2.5765990639625587,
+      "grad_norm": 0.09240175783634186,
+      "learning_rate": 9.785528554616252e-06,
+      "loss": 0.0033,
+      "step": 4130
+    },
+    {
+      "epoch": 2.582839313572543,
+      "grad_norm": 0.07451074570417404,
+      "learning_rate": 9.504450521959107e-06,
+      "loss": 0.0039,
+      "step": 4140
+    },
+    {
+      "epoch": 2.589079563182527,
+      "grad_norm": 0.0004103380488231778,
+      "learning_rate": 9.227266846068416e-06,
+      "loss": 0.0038,
+      "step": 4150
+    },
+    {
+      "epoch": 2.5953198127925114,
+      "grad_norm": 0.08585964888334274,
+      "learning_rate": 8.953989455178558e-06,
+      "loss": 0.0034,
+      "step": 4160
+    },
+    {
+      "epoch": 2.601560062402496,
+      "grad_norm": 0.023947864770889282,
+      "learning_rate": 8.68463010942211e-06,
+      "loss": 0.0035,
+      "step": 4170
+    },
+    {
+      "epoch": 2.6078003120124804,
+      "grad_norm": 0.20187538862228394,
+      "learning_rate": 8.419200400323669e-06,
+      "loss": 0.0029,
+      "step": 4180
+    },
+    {
+      "epoch": 2.614040561622465,
+      "grad_norm": 0.0033060340210795403,
+      "learning_rate": 8.157711750301112e-06,
+      "loss": 0.0043,
+      "step": 4190
+    },
+    {
+      "epoch": 2.6202808112324494,
+      "grad_norm": 0.08134489506483078,
+      "learning_rate": 7.900175412174082e-06,
+      "loss": 0.0052,
+      "step": 4200
+    },
+    {
+      "epoch": 2.6265210608424336,
+      "grad_norm": 0.001747625763528049,
+      "learning_rate": 7.646602468679631e-06,
+      "loss": 0.0018,
+      "step": 4210
+    },
+    {
+      "epoch": 2.632761310452418,
+      "grad_norm": 0.0007245971937663853,
+      "learning_rate": 7.397003831995364e-06,
+      "loss": 0.0035,
+      "step": 4220
+    },
+    {
+      "epoch": 2.6390015600624026,
+      "grad_norm": 0.00047070434084162116,
+      "learning_rate": 7.151390243269851e-06,
+      "loss": 0.0011,
+      "step": 4230
+    },
+    {
+      "epoch": 2.645241809672387,
+      "grad_norm": 0.04296772554516792,
+      "learning_rate": 6.909772272160331e-06,
+      "loss": 0.0028,
+      "step": 4240
+    },
+    {
+      "epoch": 2.651482059282371,
+      "grad_norm": 0.040636662393808365,
+      "learning_rate": 6.672160316377896e-06,
+      "loss": 0.0035,
+      "step": 4250
+    },
+    {
+      "epoch": 2.657722308892356,
+      "grad_norm": 0.0022113157901912928,
+      "learning_rate": 6.4385646012400935e-06,
+      "loss": 0.0026,
+      "step": 4260
+    },
+    {
+      "epoch": 2.66396255850234,
+      "grad_norm": 0.0014167275512591004,
+      "learning_rate": 6.2089951792308255e-06,
+      "loss": 0.0047,
+      "step": 4270
+    },
+    {
+      "epoch": 2.6702028081123244,
+      "grad_norm": 0.0010946766706183553,
+      "learning_rate": 5.9834619295677686e-06,
+      "loss": 0.0019,
+      "step": 4280
+    },
+    {
+      "epoch": 2.6764430577223086,
+      "grad_norm": 0.07304741442203522,
+      "learning_rate": 5.761974557777216e-06,
+      "loss": 0.004,
+      "step": 4290
+    },
+    {
+      "epoch": 2.6826833073322933,
+      "grad_norm": 0.00039819118683226407,
+      "learning_rate": 5.5445425952764585e-06,
+      "loss": 0.003,
+      "step": 4300
+    },
+    {
+      "epoch": 2.6889235569422776,
+      "grad_norm": 0.056995488703250885,
+      "learning_rate": 5.3311753989636285e-06,
+      "loss": 0.0043,
+      "step": 4310
+    },
+    {
+      "epoch": 2.6951638065522623,
+      "grad_norm": 0.000514000013936311,
+      "learning_rate": 5.121882150814927e-06,
+      "loss": 0.0032,
+      "step": 4320
+    },
+    {
+      "epoch": 2.7014040561622465,
+      "grad_norm": 0.09580782800912857,
+      "learning_rate": 4.916671857489597e-06,
+      "loss": 0.0042,
+      "step": 4330
+    },
+    {
+      "epoch": 2.707644305772231,
+      "grad_norm": 0.22952914237976074,
+      "learning_rate": 4.715553349942303e-06,
+      "loss": 0.0038,
+      "step": 4340
+    },
+    {
+      "epoch": 2.713884555382215,
+      "grad_norm": 0.0012364761205390096,
+      "learning_rate": 4.5185352830431125e-06,
+      "loss": 0.0024,
+      "step": 4350
+    },
+    {
+      "epoch": 2.7201248049921998,
+      "grad_norm": 0.0007727920892648399,
+      "learning_rate": 4.3256261352049855e-06,
+      "loss": 0.0024,
+      "step": 4360
+    },
+    {
+      "epoch": 2.726365054602184,
+      "grad_norm": 0.0006329137831926346,
+      "learning_rate": 4.136834208019014e-06,
+      "loss": 0.0026,
+      "step": 4370
+    },
+    {
+      "epoch": 2.7326053042121687,
+      "grad_norm": 0.10149122774600983,
+      "learning_rate": 3.952167625897063e-06,
+      "loss": 0.006,
+      "step": 4380
+    },
+    {
+      "epoch": 2.738845553822153,
+      "grad_norm": 0.000572658667806536,
+      "learning_rate": 3.771634335722285e-06,
+      "loss": 0.0033,
+      "step": 4390
+    },
+    {
+      "epoch": 2.7450858034321373,
+      "grad_norm": 0.0005446093855425715,
+      "learning_rate": 3.5952421065069797e-06,
+      "loss": 0.0037,
+      "step": 4400
+    },
+    {
+      "epoch": 2.7513260530421215,
+      "grad_norm": 0.08697942644357681,
+      "learning_rate": 3.4229985290584077e-06,
+      "loss": 0.0027,
+      "step": 4410
+    },
+    {
+      "epoch": 2.7575663026521062,
+      "grad_norm": 0.0006207876140251756,
+      "learning_rate": 3.2549110156520068e-06,
+      "loss": 0.0018,
+      "step": 4420
+    },
+    {
+      "epoch": 2.7638065522620905,
+      "grad_norm": 0.0375247448682785,
+      "learning_rate": 3.090986799712525e-06,
+      "loss": 0.0026,
+      "step": 4430
+    },
+    {
+      "epoch": 2.7700468018720747,
+      "grad_norm": 0.04099486023187637,
+      "learning_rate": 2.931232935502648e-06,
+      "loss": 0.0041,
+      "step": 4440
+    },
+    {
+      "epoch": 2.7762870514820595,
+      "grad_norm": 0.05648237094283104,
+      "learning_rate": 2.7756562978195066e-06,
+      "loss": 0.0036,
+      "step": 4450
+    },
+    {
+      "epoch": 2.7825273010920437,
+      "grad_norm": 0.08683937788009644,
+      "learning_rate": 2.6242635816987606e-06,
+      "loss": 0.0021,
+      "step": 4460
+    },
+    {
+      "epoch": 2.788767550702028,
+      "grad_norm": 0.06880386173725128,
+      "learning_rate": 2.477061302126538e-06,
+      "loss": 0.0037,
+      "step": 4470
+    },
+    {
+      "epoch": 2.7950078003120122,
+      "grad_norm": 0.04714474081993103,
+      "learning_rate": 2.3340557937590045e-06,
+      "loss": 0.0029,
+      "step": 4480
+    },
+    {
+      "epoch": 2.801248049921997,
+      "grad_norm": 0.0878954529762268,
+      "learning_rate": 2.1952532106498613e-06,
+      "loss": 0.0041,
+      "step": 4490
+    },
+    {
+      "epoch": 2.807488299531981,
+      "grad_norm": 0.0005998958367854357,
+      "learning_rate": 2.0606595259854534e-06,
+      "loss": 0.002,
+      "step": 4500
+    },
+    {
+      "epoch": 2.813728549141966,
+      "grad_norm": 0.0010315885301679373,
+      "learning_rate": 1.930280531827677e-06,
+      "loss": 0.0034,
+      "step": 4510
+    },
+    {
+      "epoch": 2.81996879875195,
+      "grad_norm": 0.09765970706939697,
+      "learning_rate": 1.8041218388648029e-06,
+      "loss": 0.0029,
+      "step": 4520
+    },
+    {
+      "epoch": 2.8262090483619344,
+      "grad_norm": 0.0010021389462053776,
+      "learning_rate": 1.6821888761700212e-06,
+      "loss": 0.0032,
+      "step": 4530
+    },
+    {
+      "epoch": 2.8324492979719187,
+      "grad_norm": 0.0008393676835112274,
+      "learning_rate": 1.5644868909677535e-06,
+      "loss": 0.0013,
+      "step": 4540
+    },
+    {
+      "epoch": 2.8386895475819034,
+      "grad_norm": 0.1556025594472885,
+      "learning_rate": 1.4510209484078884e-06,
+      "loss": 0.0035,
+      "step": 4550
+    },
+    {
+      "epoch": 2.8449297971918877,
+      "grad_norm": 0.13432757556438446,
+      "learning_rate": 1.3417959313478112e-06,
+      "loss": 0.0026,
+      "step": 4560
+    },
+    {
+      "epoch": 2.851170046801872,
+      "grad_norm": 0.09762319177389145,
+      "learning_rate": 1.2368165401422493e-06,
+      "loss": 0.0034,
+      "step": 4570
+    },
+    {
+      "epoch": 2.8574102964118566,
+      "grad_norm": 0.0012418247060850263,
+      "learning_rate": 1.1360872924410127e-06,
+      "loss": 0.0014,
+      "step": 4580
+    },
+    {
+      "epoch": 2.863650546021841,
+      "grad_norm": 0.04660167917609215,
+      "learning_rate": 1.039612522994582e-06,
+      "loss": 0.004,
+      "step": 4590
+    },
+    {
+      "epoch": 2.869890795631825,
+      "grad_norm": 0.0005681141628883779,
+      "learning_rate": 9.473963834676136e-07,
+      "loss": 0.0043,
+      "step": 4600
+    },
+    {
+      "epoch": 2.8761310452418094,
+      "grad_norm": 0.0513504222035408,
+      "learning_rate": 8.594428422601608e-07,
+      "loss": 0.004,
+      "step": 4610
+    },
+    {
+      "epoch": 2.882371294851794,
+      "grad_norm": 0.0008744900114834309,
+      "learning_rate": 7.757556843370428e-07,
+      "loss": 0.002,
+      "step": 4620
+    },
+    {
+      "epoch": 2.8886115444617784,
+      "grad_norm": 0.0013541991356760263,
+      "learning_rate": 6.963385110648313e-07,
+      "loss": 0.0021,
+      "step": 4630
+    },
+    {
+      "epoch": 2.894851794071763,
+      "grad_norm": 0.04980359971523285,
+      "learning_rate": 6.211947400569962e-07,
+      "loss": 0.0025,
+      "step": 4640
+    },
+    {
+      "epoch": 2.9010920436817473,
+      "grad_norm": 0.003807028057053685,
+      "learning_rate": 5.503276050267458e-07,
+      "loss": 0.0026,
+      "step": 4650
+    },
+    {
+      "epoch": 2.9073322932917316,
+      "grad_norm": 0.10594616085290909,
+      "learning_rate": 4.837401556478493e-07,
+      "loss": 0.0055,
+      "step": 4660
+    },
+    {
+      "epoch": 2.913572542901716,
+      "grad_norm": 0.000727578008081764,
+      "learning_rate": 4.214352574235081e-07,
+      "loss": 0.0042,
+      "step": 4670
+    },
+    {
+      "epoch": 2.9198127925117006,
+      "grad_norm": 0.0013031769776716828,
+      "learning_rate": 3.634155915629323e-07,
+      "loss": 0.0037,
+      "step": 4680
+    },
+    {
+      "epoch": 2.926053042121685,
+      "grad_norm": 0.002906072186306119,
+      "learning_rate": 3.09683654866022e-07,
+      "loss": 0.0018,
+      "step": 4690
+    },
+    {
+      "epoch": 2.9322932917316695,
+      "grad_norm": 0.05608970671892166,
+      "learning_rate": 2.602417596158868e-07,
+      "loss": 0.0035,
+      "step": 4700
+    },
+    {
+      "epoch": 2.938533541341654,
+      "grad_norm": 0.0014681172324344516,
+      "learning_rate": 2.1509203347936934e-07,
+      "loss": 0.005,
+      "step": 4710
+    },
+    {
+      "epoch": 2.944773790951638,
+      "grad_norm": 0.1732673943042755,
+      "learning_rate": 1.7423641941546332e-07,
+      "loss": 0.003,
+      "step": 4720
+    },
+    {
+      "epoch": 2.9510140405616223,
+      "grad_norm": 0.0007420650799758732,
+      "learning_rate": 1.376766755916914e-07,
+      "loss": 0.0036,
+      "step": 4730
+    },
+    {
+      "epoch": 2.957254290171607,
+      "grad_norm": 0.0006529244128614664,
+      "learning_rate": 1.0541437530848796e-07,
+      "loss": 0.0035,
+      "step": 4740
+    },
+    {
+      "epoch": 2.9634945397815913,
+      "grad_norm": 0.0012812407221645117,
+      "learning_rate": 7.745090693143108e-08,
+      "loss": 0.0028,
+      "step": 4750
+    },
+    {
+      "epoch": 2.9697347893915755,
+      "grad_norm": 0.0007352688116952777,
+      "learning_rate": 5.378747383154581e-08,
+      "loss": 0.0024,
+      "step": 4760
+    },
+    {
+      "epoch": 2.9759750390015602,
+      "grad_norm": 0.0036806329153478146,
+      "learning_rate": 3.442509433352337e-08,
+      "loss": 0.0027,
+      "step": 4770
+    },
+    {
+      "epoch": 2.9822152886115445,
+      "grad_norm": 0.0006891420343890786,
+      "learning_rate": 1.9364601671845173e-08,
+      "loss": 0.0031,
+      "step": 4780
+    },
+    {
+      "epoch": 2.9884555382215288,
+      "grad_norm": 0.062437161803245544,
+      "learning_rate": 8.606643954978122e-09,
+      "loss": 0.0034,
+      "step": 4790
+    },
+    {
+      "epoch": 2.994695787831513,
+      "grad_norm": 0.12944412231445312,
+      "learning_rate": 2.1516841374635834e-09,
+      "loss": 0.0038,
+      "step": 4800
     }
   ],
   "logging_steps": 10,
@@ -2127,7 +3387,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 2.4257904823962624e+16,
+  "total_flos": 3.881044627922227e+16,
   "train_batch_size": 2,
   "trial_name": null,
   "trial_params": null