diff --git "a/checkpoint-5500/trainer_state.json" "b/checkpoint-5500/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-5500/trainer_state.json"
@@ -0,0 +1,3972 @@
+{
+  "best_global_step": 5500,
+  "best_metric": 0.9099580645561218,
+  "best_model_checkpoint": "models/MNLP_M3_rag_model_test/checkpoint-5500",
+  "epoch": 2.8732210471340904,
+  "eval_steps": 500,
+  "global_step": 5500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.005222613918266093,
+      "grad_norm": 17.1588077545166,
+      "learning_rate": 3.757828810020877e-07,
+      "loss": 3.1712,
+      "step": 10
+    },
+    {
+      "epoch": 0.010445227836532185,
+      "grad_norm": 12.760807991027832,
+      "learning_rate": 7.933194154488519e-07,
+      "loss": 3.1084,
+      "step": 20
+    },
+    {
+      "epoch": 0.015667841754798278,
+      "grad_norm": 9.724287986755371,
+      "learning_rate": 1.210855949895616e-06,
+      "loss": 2.7273,
+      "step": 30
+    },
+    {
+      "epoch": 0.02089045567306437,
+      "grad_norm": 6.329061031341553,
+      "learning_rate": 1.62839248434238e-06,
+      "loss": 2.5386,
+      "step": 40
+    },
+    {
+      "epoch": 0.02611306959133046,
+      "grad_norm": 8.292409896850586,
+      "learning_rate": 2.0459290187891444e-06,
+      "loss": 2.4021,
+      "step": 50
+    },
+    {
+      "epoch": 0.031335683509596556,
+      "grad_norm": 7.1383161544799805,
+      "learning_rate": 2.4634655532359085e-06,
+      "loss": 2.3112,
+      "step": 60
+    },
+    {
+      "epoch": 0.036558297427862645,
+      "grad_norm": 6.580810546875,
+      "learning_rate": 2.8810020876826722e-06,
+      "loss": 2.3126,
+      "step": 70
+    },
+    {
+      "epoch": 0.04178091134612874,
+      "grad_norm": 7.729620456695557,
+      "learning_rate": 3.2985386221294363e-06,
+      "loss": 2.3074,
+      "step": 80
+    },
+    {
+      "epoch": 0.04700352526439483,
+      "grad_norm": 6.321948528289795,
+      "learning_rate": 3.716075156576201e-06,
+      "loss": 2.2807,
+      "step": 90
+    },
+    {
+      "epoch": 0.05222613918266092,
+      "grad_norm": 6.126489162445068,
+      "learning_rate": 4.1336116910229655e-06,
+      "loss": 2.2118,
+      "step": 100
+    },
+    {
+      "epoch": 0.057448753100927015,
+      "grad_norm": 6.092052936553955,
+      "learning_rate": 4.551148225469729e-06,
+      "loss": 2.2692,
+      "step": 110
+    },
+    {
+      "epoch": 0.06267136701919311,
+      "grad_norm": 6.273594856262207,
+      "learning_rate": 4.968684759916494e-06,
+      "loss": 2.2302,
+      "step": 120
+    },
+    {
+      "epoch": 0.0678939809374592,
+      "grad_norm": 6.156396389007568,
+      "learning_rate": 5.3862212943632574e-06,
+      "loss": 2.1471,
+      "step": 130
+    },
+    {
+      "epoch": 0.07311659485572529,
+      "grad_norm": 6.262785911560059,
+      "learning_rate": 5.803757828810022e-06,
+      "loss": 2.2413,
+      "step": 140
+    },
+    {
+      "epoch": 0.07833920877399138,
+      "grad_norm": 9.364627838134766,
+      "learning_rate": 6.221294363256786e-06,
+      "loss": 2.0653,
+      "step": 150
+    },
+    {
+      "epoch": 0.08356182269225748,
+      "grad_norm": 5.752789497375488,
+      "learning_rate": 6.638830897703549e-06,
+      "loss": 2.2157,
+      "step": 160
+    },
+    {
+      "epoch": 0.08878443661052357,
+      "grad_norm": 6.819472789764404,
+      "learning_rate": 7.056367432150314e-06,
+      "loss": 2.1066,
+      "step": 170
+    },
+    {
+      "epoch": 0.09400705052878966,
+      "grad_norm": 7.4179182052612305,
+      "learning_rate": 7.473903966597078e-06,
+      "loss": 2.1168,
+      "step": 180
+    },
+    {
+      "epoch": 0.09922966444705575,
+      "grad_norm": 7.955780029296875,
+      "learning_rate": 7.891440501043842e-06,
+      "loss": 2.0347,
+      "step": 190
+    },
+    {
+      "epoch": 0.10445227836532184,
+      "grad_norm": 6.121779441833496,
+      "learning_rate": 8.308977035490606e-06,
+      "loss": 2.0351,
+      "step": 200
+    },
+    {
+      "epoch": 0.10967489228358794,
+      "grad_norm": 7.683338642120361,
+      "learning_rate": 8.72651356993737e-06,
+      "loss": 2.0195,
+      "step": 210
+    },
+    {
+      "epoch": 0.11489750620185403,
+      "grad_norm": 6.982047080993652,
+      "learning_rate": 9.144050104384135e-06,
+      "loss": 1.9903,
+      "step": 220
+    },
+    {
+      "epoch": 0.12012012012012012,
+      "grad_norm": 6.351634502410889,
+      "learning_rate": 9.561586638830899e-06,
+      "loss": 1.7746,
+      "step": 230
+    },
+    {
+      "epoch": 0.12534273403838622,
+      "grad_norm": 6.279501438140869,
+      "learning_rate": 9.979123173277662e-06,
+      "loss": 2.1087,
+      "step": 240
+    },
+    {
+      "epoch": 0.1305653479566523,
+      "grad_norm": 6.24404239654541,
+      "learning_rate": 1.0396659707724426e-05,
+      "loss": 1.9282,
+      "step": 250
+    },
+    {
+      "epoch": 0.1357879618749184,
+      "grad_norm": 7.28134298324585,
+      "learning_rate": 1.0814196242171192e-05,
+      "loss": 1.8358,
+      "step": 260
+    },
+    {
+      "epoch": 0.1410105757931845,
+      "grad_norm": 7.017399787902832,
+      "learning_rate": 1.1231732776617954e-05,
+      "loss": 1.8979,
+      "step": 270
+    },
+    {
+      "epoch": 0.14623318971145058,
+      "grad_norm": 6.723785400390625,
+      "learning_rate": 1.1649269311064719e-05,
+      "loss": 1.8811,
+      "step": 280
+    },
+    {
+      "epoch": 0.15145580362971667,
+      "grad_norm": 6.5550360679626465,
+      "learning_rate": 1.2066805845511483e-05,
+      "loss": 1.7359,
+      "step": 290
+    },
+    {
+      "epoch": 0.15667841754798276,
+      "grad_norm": 6.1090240478515625,
+      "learning_rate": 1.2484342379958248e-05,
+      "loss": 1.8842,
+      "step": 300
+    },
+    {
+      "epoch": 0.16190103146624885,
+      "grad_norm": 7.078915596008301,
+      "learning_rate": 1.290187891440501e-05,
+      "loss": 1.9297,
+      "step": 310
+    },
+    {
+      "epoch": 0.16712364538451496,
+      "grad_norm": 7.479031085968018,
+      "learning_rate": 1.3319415448851776e-05,
+      "loss": 1.8578,
+      "step": 320
+    },
+    {
+      "epoch": 0.17234625930278105,
+      "grad_norm": 6.292983531951904,
+      "learning_rate": 1.373695198329854e-05,
+      "loss": 1.7184,
+      "step": 330
+    },
+    {
+      "epoch": 0.17756887322104714,
+      "grad_norm": 7.733270168304443,
+      "learning_rate": 1.4154488517745305e-05,
+      "loss": 1.8098,
+      "step": 340
+    },
+    {
+      "epoch": 0.18279148713931323,
+      "grad_norm": 8.41698169708252,
+      "learning_rate": 1.4572025052192067e-05,
+      "loss": 1.8588,
+      "step": 350
+    },
+    {
+      "epoch": 0.18801410105757932,
+      "grad_norm": 7.4168243408203125,
+      "learning_rate": 1.4989561586638832e-05,
+      "loss": 1.6814,
+      "step": 360
+    },
+    {
+      "epoch": 0.1932367149758454,
+      "grad_norm": 6.406525611877441,
+      "learning_rate": 1.5407098121085594e-05,
+      "loss": 1.805,
+      "step": 370
+    },
+    {
+      "epoch": 0.1984593288941115,
+      "grad_norm": 7.147383689880371,
+      "learning_rate": 1.582463465553236e-05,
+      "loss": 1.7771,
+      "step": 380
+    },
+    {
+      "epoch": 0.2036819428123776,
+      "grad_norm": 6.644751071929932,
+      "learning_rate": 1.6242171189979125e-05,
+      "loss": 1.7242,
+      "step": 390
+    },
+    {
+      "epoch": 0.20890455673064368,
+      "grad_norm": 6.507500648498535,
+      "learning_rate": 1.665970772442589e-05,
+      "loss": 1.8213,
+      "step": 400
+    },
+    {
+      "epoch": 0.2141271706489098,
+      "grad_norm": 6.171652317047119,
+      "learning_rate": 1.7077244258872652e-05,
+      "loss": 1.6563,
+      "step": 410
+    },
+    {
+      "epoch": 0.21934978456717588,
+      "grad_norm": 6.022638320922852,
+      "learning_rate": 1.745302713987474e-05,
+      "loss": 1.7445,
+      "step": 420
+    },
+    {
+      "epoch": 0.22457239848544197,
+      "grad_norm": 6.484378814697266,
+      "learning_rate": 1.7870563674321503e-05,
+      "loss": 1.3785,
+      "step": 430
+    },
+    {
+      "epoch": 0.22979501240370806,
+      "grad_norm": 6.780409336090088,
+      "learning_rate": 1.828810020876827e-05,
+      "loss": 1.6468,
+      "step": 440
+    },
+    {
+      "epoch": 0.23501762632197415,
+      "grad_norm": 6.643301963806152,
+      "learning_rate": 1.870563674321503e-05,
+      "loss": 1.6664,
+      "step": 450
+    },
+    {
+      "epoch": 0.24024024024024024,
+      "grad_norm": 6.810093879699707,
+      "learning_rate": 1.9123173277661798e-05,
+      "loss": 1.4001,
+      "step": 460
+    },
+    {
+      "epoch": 0.24546285415850633,
+      "grad_norm": 7.195308208465576,
+      "learning_rate": 1.954070981210856e-05,
+      "loss": 1.5031,
+      "step": 470
+    },
+    {
+      "epoch": 0.25068546807677244,
+      "grad_norm": 7.472316741943359,
+      "learning_rate": 1.9958246346555325e-05,
+      "loss": 1.5853,
+      "step": 480
+    },
+    {
+      "epoch": 0.2559080819950385,
+      "grad_norm": 6.748635768890381,
+      "learning_rate": 1.998020019799802e-05,
+      "loss": 1.5791,
+      "step": 490
+    },
+    {
+      "epoch": 0.2611306959133046,
+      "grad_norm": 6.5012431144714355,
+      "learning_rate": 1.9958200417995823e-05,
+      "loss": 1.4822,
+      "step": 500
+    },
+    {
+      "epoch": 0.2611306959133046,
+      "eval_loss": 1.5763118267059326,
+      "eval_runtime": 46.6118,
+      "eval_samples_per_second": 36.514,
+      "eval_steps_per_second": 4.57,
+      "step": 500
+    },
+    {
+      "epoch": 0.2663533098315707,
+      "grad_norm": 6.5022149085998535,
+      "learning_rate": 1.9936200637993622e-05,
+      "loss": 1.5332,
+      "step": 510
+    },
+    {
+      "epoch": 0.2715759237498368,
+      "grad_norm": 8.452847480773926,
+      "learning_rate": 1.9914200857991422e-05,
+      "loss": 1.547,
+      "step": 520
+    },
+    {
+      "epoch": 0.27679853766810286,
+      "grad_norm": 6.139389991760254,
+      "learning_rate": 1.9892201077989222e-05,
+      "loss": 1.6269,
+      "step": 530
+    },
+    {
+      "epoch": 0.282021151586369,
+      "grad_norm": 6.797521114349365,
+      "learning_rate": 1.9870201297987022e-05,
+      "loss": 1.5635,
+      "step": 540
+    },
+    {
+      "epoch": 0.2872437655046351,
+      "grad_norm": 7.221142292022705,
+      "learning_rate": 1.9848201517984822e-05,
+      "loss": 1.3621,
+      "step": 550
+    },
+    {
+      "epoch": 0.29246637942290116,
+      "grad_norm": 6.7414984703063965,
+      "learning_rate": 1.982620173798262e-05,
+      "loss": 1.4587,
+      "step": 560
+    },
+    {
+      "epoch": 0.2976889933411673,
+      "grad_norm": 6.582213401794434,
+      "learning_rate": 1.980420195798042e-05,
+      "loss": 1.5187,
+      "step": 570
+    },
+    {
+      "epoch": 0.30291160725943334,
+      "grad_norm": 6.121609210968018,
+      "learning_rate": 1.978220217797822e-05,
+      "loss": 1.7261,
+      "step": 580
+    },
+    {
+      "epoch": 0.30813422117769945,
+      "grad_norm": 7.791438579559326,
+      "learning_rate": 1.976020239797602e-05,
+      "loss": 1.6164,
+      "step": 590
+    },
+    {
+      "epoch": 0.3133568350959655,
+      "grad_norm": 8.201029777526855,
+      "learning_rate": 1.9738202617973824e-05,
+      "loss": 1.4859,
+      "step": 600
+    },
+    {
+      "epoch": 0.31857944901423163,
+      "grad_norm": 5.579436302185059,
+      "learning_rate": 1.971620283797162e-05,
+      "loss": 1.2939,
+      "step": 610
+    },
+    {
+      "epoch": 0.3238020629324977,
+      "grad_norm": 6.592574119567871,
+      "learning_rate": 1.969420305796942e-05,
+      "loss": 1.3201,
+      "step": 620
+    },
+    {
+      "epoch": 0.3290246768507638,
+      "grad_norm": 5.446815490722656,
+      "learning_rate": 1.967220327796722e-05,
+      "loss": 1.5061,
+      "step": 630
+    },
+    {
+      "epoch": 0.3342472907690299,
+      "grad_norm": 5.549929141998291,
+      "learning_rate": 1.965020349796502e-05,
+      "loss": 1.5843,
+      "step": 640
+    },
+    {
+      "epoch": 0.339469904687296,
+      "grad_norm": 5.987018585205078,
+      "learning_rate": 1.962820371796282e-05,
+      "loss": 1.3113,
+      "step": 650
+    },
+    {
+      "epoch": 0.3446925186055621,
+      "grad_norm": 5.247632026672363,
+      "learning_rate": 1.9606203937960623e-05,
+      "loss": 1.476,
+      "step": 660
+    },
+    {
+      "epoch": 0.34991513252382817,
+      "grad_norm": 6.408908367156982,
+      "learning_rate": 1.9584204157958423e-05,
+      "loss": 1.67,
+      "step": 670
+    },
+    {
+      "epoch": 0.3551377464420943,
+      "grad_norm": 7.674783229827881,
+      "learning_rate": 1.9562204377956223e-05,
+      "loss": 1.4345,
+      "step": 680
+    },
+    {
+      "epoch": 0.36036036036036034,
+      "grad_norm": 5.847611427307129,
+      "learning_rate": 1.9540204597954023e-05,
+      "loss": 1.4084,
+      "step": 690
+    },
+    {
+      "epoch": 0.36558297427862646,
+      "grad_norm": 7.193668365478516,
+      "learning_rate": 1.9518204817951823e-05,
+      "loss": 1.4818,
+      "step": 700
+    },
+    {
+      "epoch": 0.3708055881968925,
+      "grad_norm": 5.585901260375977,
+      "learning_rate": 1.9496205037949622e-05,
+      "loss": 1.3425,
+      "step": 710
+    },
+    {
+      "epoch": 0.37602820211515864,
+      "grad_norm": 5.371877193450928,
+      "learning_rate": 1.9474205257947422e-05,
+      "loss": 1.4872,
+      "step": 720
+    },
+    {
+      "epoch": 0.38125081603342476,
+      "grad_norm": 7.625179767608643,
+      "learning_rate": 1.9452205477945222e-05,
+      "loss": 1.4589,
+      "step": 730
+    },
+    {
+      "epoch": 0.3864734299516908,
+      "grad_norm": 6.194158554077148,
+      "learning_rate": 1.9430205697943022e-05,
+      "loss": 1.3672,
+      "step": 740
+    },
+    {
+      "epoch": 0.39169604386995693,
+      "grad_norm": 5.667837619781494,
+      "learning_rate": 1.940820591794082e-05,
+      "loss": 1.2324,
+      "step": 750
+    },
+    {
+      "epoch": 0.396918657788223,
+      "grad_norm": 5.823015213012695,
+      "learning_rate": 1.938620613793862e-05,
+      "loss": 1.5731,
+      "step": 760
+    },
+    {
+      "epoch": 0.4021412717064891,
+      "grad_norm": 5.3874945640563965,
+      "learning_rate": 1.936420635793642e-05,
+      "loss": 1.456,
+      "step": 770
+    },
+    {
+      "epoch": 0.4073638856247552,
+      "grad_norm": 9.306315422058105,
+      "learning_rate": 1.934220657793422e-05,
+      "loss": 1.3355,
+      "step": 780
+    },
+    {
+      "epoch": 0.4125864995430213,
+      "grad_norm": 7.826662540435791,
+      "learning_rate": 1.932020679793202e-05,
+      "loss": 1.2842,
+      "step": 790
+    },
+    {
+      "epoch": 0.41780911346128735,
+      "grad_norm": 5.670297145843506,
+      "learning_rate": 1.929820701792982e-05,
+      "loss": 1.3126,
+      "step": 800
+    },
+    {
+      "epoch": 0.42303172737955347,
+      "grad_norm": 5.819605827331543,
+      "learning_rate": 1.927620723792762e-05,
+      "loss": 1.4907,
+      "step": 810
+    },
+    {
+      "epoch": 0.4282543412978196,
+      "grad_norm": 5.6237335205078125,
+      "learning_rate": 1.9254207457925424e-05,
+      "loss": 1.3278,
+      "step": 820
+    },
+    {
+      "epoch": 0.43347695521608565,
+      "grad_norm": 8.079651832580566,
+      "learning_rate": 1.9232207677923224e-05,
+      "loss": 1.468,
+      "step": 830
+    },
+    {
+      "epoch": 0.43869956913435176,
+      "grad_norm": 5.166318416595459,
+      "learning_rate": 1.9210207897921023e-05,
+      "loss": 1.3633,
+      "step": 840
+    },
+    {
+      "epoch": 0.4439221830526178,
+      "grad_norm": 5.308135509490967,
+      "learning_rate": 1.9188208117918823e-05,
+      "loss": 1.3076,
+      "step": 850
+    },
+    {
+      "epoch": 0.44914479697088394,
+      "grad_norm": 5.224324703216553,
+      "learning_rate": 1.9166208337916623e-05,
+      "loss": 1.3712,
+      "step": 860
+    },
+    {
+      "epoch": 0.45436741088915,
+      "grad_norm": 5.71431827545166,
+      "learning_rate": 1.9144208557914423e-05,
+      "loss": 1.4854,
+      "step": 870
+    },
+    {
+      "epoch": 0.4595900248074161,
+      "grad_norm": 8.291099548339844,
+      "learning_rate": 1.9122208777912223e-05,
+      "loss": 1.3062,
+      "step": 880
+    },
+    {
+      "epoch": 0.4648126387256822,
+      "grad_norm": 6.186370849609375,
+      "learning_rate": 1.9100208997910023e-05,
+      "loss": 1.3916,
+      "step": 890
+    },
+    {
+      "epoch": 0.4700352526439483,
+      "grad_norm": 5.299021244049072,
+      "learning_rate": 1.9078209217907822e-05,
+      "loss": 1.3207,
+      "step": 900
+    },
+    {
+      "epoch": 0.4752578665622144,
+      "grad_norm": 5.656353950500488,
+      "learning_rate": 1.9056209437905622e-05,
+      "loss": 1.2343,
+      "step": 910
+    },
+    {
+      "epoch": 0.4804804804804805,
+      "grad_norm": 5.802785396575928,
+      "learning_rate": 1.9034209657903422e-05,
+      "loss": 1.4532,
+      "step": 920
+    },
+    {
+      "epoch": 0.4857030943987466,
+      "grad_norm": 6.065880298614502,
+      "learning_rate": 1.9012209877901222e-05,
+      "loss": 1.3908,
+      "step": 930
+    },
+    {
+      "epoch": 0.49092570831701265,
+      "grad_norm": 5.135191440582275,
+      "learning_rate": 1.8990210097899022e-05,
+      "loss": 1.3211,
+      "step": 940
+    },
+    {
+      "epoch": 0.49614832223527877,
+      "grad_norm": 5.89790153503418,
+      "learning_rate": 1.896821031789682e-05,
+      "loss": 1.4183,
+      "step": 950
+    },
+    {
+      "epoch": 0.5013709361535449,
+      "grad_norm": 5.3774566650390625,
+      "learning_rate": 1.894621053789462e-05,
+      "loss": 1.3267,
+      "step": 960
+    },
+    {
+      "epoch": 0.506593550071811,
+      "grad_norm": 6.830160617828369,
+      "learning_rate": 1.892421075789242e-05,
+      "loss": 1.3054,
+      "step": 970
+    },
+    {
+      "epoch": 0.511816163990077,
+      "grad_norm": 5.20271110534668,
+      "learning_rate": 1.8902210977890224e-05,
+      "loss": 1.3637,
+      "step": 980
+    },
+    {
+      "epoch": 0.5170387779083431,
+      "grad_norm": 7.338613986968994,
+      "learning_rate": 1.8880211197888024e-05,
+      "loss": 1.2787,
+      "step": 990
+    },
+    {
+      "epoch": 0.5222613918266092,
+      "grad_norm": 5.433679580688477,
+      "learning_rate": 1.8858211417885824e-05,
+      "loss": 1.2041,
+      "step": 1000
+    },
+    {
+      "epoch": 0.5222613918266092,
+      "eval_loss": 1.2922958135604858,
+      "eval_runtime": 46.4161,
+      "eval_samples_per_second": 36.668,
+      "eval_steps_per_second": 4.589,
+      "step": 1000
+    },
+    {
+      "epoch": 0.5274840057448753,
+      "grad_norm": 6.289940357208252,
+      "learning_rate": 1.8836211637883624e-05,
+      "loss": 1.4183,
+      "step": 1010
+    },
+    {
+      "epoch": 0.5327066196631414,
+      "grad_norm": 4.168149471282959,
+      "learning_rate": 1.8814211857881424e-05,
+      "loss": 1.209,
+      "step": 1020
+    },
+    {
+      "epoch": 0.5379292335814075,
+      "grad_norm": 7.0136189460754395,
+      "learning_rate": 1.879221207787922e-05,
+      "loss": 1.38,
+      "step": 1030
+    },
+    {
+      "epoch": 0.5431518474996736,
+      "grad_norm": 4.89089822769165,
+      "learning_rate": 1.8770212297877023e-05,
+      "loss": 1.5683,
+      "step": 1040
+    },
+    {
+      "epoch": 0.5483744614179397,
+      "grad_norm": 6.398843288421631,
+      "learning_rate": 1.8748212517874823e-05,
+      "loss": 1.3035,
+      "step": 1050
+    },
+    {
+      "epoch": 0.5535970753362057,
+      "grad_norm": 5.6711859703063965,
+      "learning_rate": 1.8726212737872623e-05,
+      "loss": 1.2262,
+      "step": 1060
+    },
+    {
+      "epoch": 0.5588196892544719,
+      "grad_norm": 5.641086578369141,
+      "learning_rate": 1.8704212957870423e-05,
+      "loss": 1.442,
+      "step": 1070
+    },
+    {
+      "epoch": 0.564042303172738,
+      "grad_norm": 5.436042308807373,
+      "learning_rate": 1.8682213177868223e-05,
+      "loss": 1.2896,
+      "step": 1080
+    },
+    {
+      "epoch": 0.569264917091004,
+      "grad_norm": 5.110105514526367,
+      "learning_rate": 1.8660213397866022e-05,
+      "loss": 1.2142,
+      "step": 1090
+    },
+    {
+      "epoch": 0.5744875310092702,
+      "grad_norm": 4.662496089935303,
+      "learning_rate": 1.8638213617863822e-05,
+      "loss": 1.4738,
+      "step": 1100
+    },
+    {
+      "epoch": 0.5797101449275363,
+      "grad_norm": 6.685956954956055,
+      "learning_rate": 1.8616213837861622e-05,
+      "loss": 1.1842,
+      "step": 1110
+    },
+    {
+      "epoch": 0.5849327588458023,
+      "grad_norm": 4.521024227142334,
+      "learning_rate": 1.8594214057859422e-05,
+      "loss": 1.2215,
+      "step": 1120
+    },
+    {
+      "epoch": 0.5901553727640684,
+      "grad_norm": 5.817269802093506,
+      "learning_rate": 1.8572214277857222e-05,
+      "loss": 1.4296,
+      "step": 1130
+    },
+    {
+      "epoch": 0.5953779866823345,
+      "grad_norm": 6.144318103790283,
+      "learning_rate": 1.8550214497855025e-05,
+      "loss": 1.3538,
+      "step": 1140
+    },
+    {
+      "epoch": 0.6006006006006006,
+      "grad_norm": 5.195668697357178,
+      "learning_rate": 1.8528214717852825e-05,
+      "loss": 1.3292,
+      "step": 1150
+    },
+    {
+      "epoch": 0.6058232145188667,
+      "grad_norm": 5.556060314178467,
+      "learning_rate": 1.8506214937850625e-05,
+      "loss": 1.3371,
+      "step": 1160
+    },
+    {
+      "epoch": 0.6110458284371327,
+      "grad_norm": 5.5624613761901855,
+      "learning_rate": 1.8484215157848424e-05,
+      "loss": 1.3969,
+      "step": 1170
+    },
+    {
+      "epoch": 0.6162684423553989,
+      "grad_norm": 4.879144668579102,
+      "learning_rate": 1.846221537784622e-05,
+      "loss": 1.3443,
+      "step": 1180
+    },
+    {
+      "epoch": 0.621491056273665,
+      "grad_norm": 5.067091941833496,
+      "learning_rate": 1.844021559784402e-05,
+      "loss": 1.2455,
+      "step": 1190
+    },
+    {
+      "epoch": 0.626713670191931,
+      "grad_norm": 5.5850443840026855,
+      "learning_rate": 1.8418215817841824e-05,
+      "loss": 1.0956,
+      "step": 1200
+    },
+    {
+      "epoch": 0.6319362841101972,
+      "grad_norm": 4.342463493347168,
+      "learning_rate": 1.8396216037839624e-05,
+      "loss": 1.2106,
+      "step": 1210
+    },
+    {
+      "epoch": 0.6371588980284633,
+      "grad_norm": 5.82832670211792,
+      "learning_rate": 1.8374216257837424e-05,
+      "loss": 1.3382,
+      "step": 1220
+    },
+    {
+      "epoch": 0.6423815119467293,
+      "grad_norm": 5.469226360321045,
+      "learning_rate": 1.8352216477835223e-05,
+      "loss": 1.4335,
+      "step": 1230
+    },
+    {
+      "epoch": 0.6476041258649954,
+      "grad_norm": 5.302371978759766,
+      "learning_rate": 1.8330216697833023e-05,
+      "loss": 1.312,
+      "step": 1240
+    },
+    {
+      "epoch": 0.6528267397832616,
+      "grad_norm": 5.672764778137207,
+      "learning_rate": 1.8308216917830823e-05,
+      "loss": 1.1967,
+      "step": 1250
+    },
+    {
+      "epoch": 0.6580493537015276,
+      "grad_norm": 4.915512561798096,
+      "learning_rate": 1.8286217137828623e-05,
+      "loss": 1.4107,
+      "step": 1260
+    },
+    {
+      "epoch": 0.6632719676197937,
+      "grad_norm": 5.713113784790039,
+      "learning_rate": 1.8264217357826423e-05,
+      "loss": 1.2576,
+      "step": 1270
+    },
+    {
+      "epoch": 0.6684945815380599,
+      "grad_norm": 6.491368293762207,
+      "learning_rate": 1.8242217577824222e-05,
+      "loss": 1.3966,
+      "step": 1280
+    },
+    {
+      "epoch": 0.6737171954563259,
+      "grad_norm": 5.070196628570557,
+      "learning_rate": 1.8220217797822026e-05,
+      "loss": 1.3004,
+      "step": 1290
+    },
+    {
+      "epoch": 0.678939809374592,
+      "grad_norm": 5.9302287101745605,
+      "learning_rate": 1.8198218017819826e-05,
+      "loss": 1.2584,
+      "step": 1300
+    },
+    {
+      "epoch": 0.684162423292858,
+      "grad_norm": 4.848388671875,
+      "learning_rate": 1.8176218237817625e-05,
+      "loss": 1.2861,
+      "step": 1310
+    },
+    {
+      "epoch": 0.6893850372111242,
+      "grad_norm": 5.589436054229736,
+      "learning_rate": 1.8154218457815422e-05,
+      "loss": 1.1153,
+      "step": 1320
+    },
+    {
+      "epoch": 0.6946076511293903,
+      "grad_norm": 4.449565887451172,
+      "learning_rate": 1.813221867781322e-05,
+      "loss": 1.2184,
+      "step": 1330
+    },
+    {
+      "epoch": 0.6998302650476563,
+      "grad_norm": 5.761178970336914,
+      "learning_rate": 1.811021889781102e-05,
+      "loss": 1.0727,
+      "step": 1340
+    },
+    {
+      "epoch": 0.7050528789659224,
+      "grad_norm": 6.625110149383545,
+      "learning_rate": 1.808821911780882e-05,
+      "loss": 1.2343,
+      "step": 1350
+    },
+    {
+      "epoch": 0.7102754928841886,
+      "grad_norm": 6.265412330627441,
+      "learning_rate": 1.8066219337806624e-05,
+      "loss": 1.2446,
+      "step": 1360
+    },
+    {
+      "epoch": 0.7154981068024546,
+      "grad_norm": 6.213257312774658,
+      "learning_rate": 1.8044219557804424e-05,
+      "loss": 1.2587,
+      "step": 1370
+    },
+    {
+      "epoch": 0.7207207207207207,
+      "grad_norm": 6.336606979370117,
+      "learning_rate": 1.8022219777802224e-05,
+      "loss": 1.2753,
+      "step": 1380
+    },
+    {
+      "epoch": 0.7259433346389869,
+      "grad_norm": 5.962548732757568,
+      "learning_rate": 1.8000219997800024e-05,
+      "loss": 1.2126,
+      "step": 1390
+    },
+    {
+      "epoch": 0.7311659485572529,
+      "grad_norm": 4.752743721008301,
+      "learning_rate": 1.7978220217797824e-05,
+      "loss": 1.1085,
+      "step": 1400
+    },
+    {
+      "epoch": 0.736388562475519,
+      "grad_norm": 4.578768253326416,
+      "learning_rate": 1.7956220437795624e-05,
+      "loss": 0.9734,
+      "step": 1410
+    },
+    {
+      "epoch": 0.741611176393785,
+      "grad_norm": 5.62348747253418,
+      "learning_rate": 1.7934220657793423e-05,
+      "loss": 1.1462,
+      "step": 1420
+    },
+    {
+      "epoch": 0.7468337903120512,
+      "grad_norm": 5.4975738525390625,
+      "learning_rate": 1.7912220877791223e-05,
+      "loss": 1.0505,
+      "step": 1430
+    },
+    {
+      "epoch": 0.7520564042303173,
+      "grad_norm": 5.553978443145752,
+      "learning_rate": 1.7890221097789023e-05,
+      "loss": 1.1606,
+      "step": 1440
+    },
+    {
+      "epoch": 0.7572790181485833,
+      "grad_norm": 5.423792362213135,
+      "learning_rate": 1.7868221317786826e-05,
+      "loss": 1.2759,
+      "step": 1450
+    },
+    {
+      "epoch": 0.7625016320668495,
+      "grad_norm": 3.9736289978027344,
+      "learning_rate": 1.7846221537784623e-05,
+      "loss": 1.1232,
+      "step": 1460
+    },
+    {
+      "epoch": 0.7677242459851156,
+      "grad_norm": 4.78857421875,
+      "learning_rate": 1.7826421735782644e-05,
+      "loss": 1.2644,
+      "step": 1470
+    },
+    {
+      "epoch": 0.7729468599033816,
+      "grad_norm": 4.863739967346191,
+      "learning_rate": 1.7804421955780444e-05,
+      "loss": 1.1593,
+      "step": 1480
+    },
+    {
+      "epoch": 0.7781694738216477,
+      "grad_norm": 5.5411882400512695,
+      "learning_rate": 1.7782422175778244e-05,
+      "loss": 1.1533,
+      "step": 1490
+    },
+    {
+      "epoch": 0.7833920877399139,
+      "grad_norm": 4.7969584465026855,
+      "learning_rate": 1.7760422395776043e-05,
+      "loss": 1.2054,
+      "step": 1500
+    },
+    {
+      "epoch": 0.7833920877399139,
+      "eval_loss": 1.152809739112854,
+      "eval_runtime": 46.4044,
+      "eval_samples_per_second": 36.678,
+      "eval_steps_per_second": 4.59,
+      "step": 1500
+    },
+    {
+      "epoch": 0.7886147016581799,
+      "grad_norm": 5.786462306976318,
+      "learning_rate": 1.7738422615773843e-05,
+      "loss": 1.1893,
+      "step": 1510
+    },
+    {
+      "epoch": 0.793837315576446,
+      "grad_norm": 5.743650913238525,
+      "learning_rate": 1.7716422835771643e-05,
+      "loss": 1.1873,
+      "step": 1520
+    },
+    {
+      "epoch": 0.799059929494712,
+      "grad_norm": 5.861995220184326,
+      "learning_rate": 1.7694423055769443e-05,
+      "loss": 1.2693,
+      "step": 1530
+    },
+    {
+      "epoch": 0.8042825434129782,
+      "grad_norm": 5.11861515045166,
+      "learning_rate": 1.7672423275767243e-05,
+      "loss": 1.1807,
+      "step": 1540
+    },
+    {
+      "epoch": 0.8095051573312443,
+      "grad_norm": 5.063669204711914,
+      "learning_rate": 1.7650423495765043e-05,
+      "loss": 1.1939,
+      "step": 1550
+    },
+    {
+      "epoch": 0.8147277712495103,
+      "grad_norm": 6.106082439422607,
+      "learning_rate": 1.7628423715762842e-05,
+      "loss": 1.0904,
+      "step": 1560
+    },
+    {
+      "epoch": 0.8199503851677765,
+      "grad_norm": 5.699225902557373,
+      "learning_rate": 1.7606423935760642e-05,
+      "loss": 1.0069,
+      "step": 1570
+    },
+    {
+      "epoch": 0.8251729990860426,
+      "grad_norm": 5.60951566696167,
+      "learning_rate": 1.7584424155758445e-05,
+      "loss": 1.196,
+      "step": 1580
+    },
+    {
+      "epoch": 0.8303956130043086,
+      "grad_norm": 6.9652099609375,
+      "learning_rate": 1.7562424375756245e-05,
+      "loss": 1.2167,
+      "step": 1590
+    },
+    {
+      "epoch": 0.8356182269225747,
+      "grad_norm": 6.079868316650391,
+      "learning_rate": 1.7540424595754045e-05,
+      "loss": 1.296,
+      "step": 1600
+    },
+    {
+      "epoch": 0.8408408408408409,
+      "grad_norm": 5.60435152053833,
+      "learning_rate": 1.7518424815751845e-05,
+      "loss": 1.2494,
+      "step": 1610
+    },
+    {
+      "epoch": 0.8460634547591069,
+      "grad_norm": 6.481119632720947,
+      "learning_rate": 1.7496425035749645e-05,
+      "loss": 1.2123,
+      "step": 1620
+    },
+    {
+      "epoch": 0.851286068677373,
+      "grad_norm": 4.173206806182861,
+      "learning_rate": 1.7474425255747444e-05,
+      "loss": 1.1377,
+      "step": 1630
+    },
+    {
+      "epoch": 0.8565086825956392,
+      "grad_norm": 4.763253688812256,
+      "learning_rate": 1.7452425475745244e-05,
+      "loss": 1.0706,
+      "step": 1640
+    },
+    {
+      "epoch": 0.8617312965139052,
+      "grad_norm": 5.0582709312438965,
+      "learning_rate": 1.7430425695743044e-05,
+      "loss": 1.0366,
+      "step": 1650
+    },
+    {
+      "epoch": 0.8669539104321713,
+      "grad_norm": 4.170965194702148,
+      "learning_rate": 1.7408425915740844e-05,
+      "loss": 1.1789,
+      "step": 1660
+    },
+    {
+      "epoch": 0.8721765243504374,
+      "grad_norm": 3.9636096954345703,
+      "learning_rate": 1.7386426135738644e-05,
+      "loss": 0.9923,
+      "step": 1670
+    },
+    {
+      "epoch": 0.8773991382687035,
+      "grad_norm": 4.601221561431885,
+      "learning_rate": 1.7364426355736444e-05,
+      "loss": 1.2057,
+      "step": 1680
+    },
+    {
+      "epoch": 0.8826217521869696,
+      "grad_norm": 5.6236114501953125,
+      "learning_rate": 1.7342426575734243e-05,
+      "loss": 1.0889,
+      "step": 1690
+    },
+    {
+      "epoch": 0.8878443661052356,
+      "grad_norm": 5.330966472625732,
+      "learning_rate": 1.7320426795732043e-05,
+      "loss": 1.1021,
+      "step": 1700
+    },
+    {
+      "epoch": 0.8930669800235017,
+      "grad_norm": 5.453662872314453,
+      "learning_rate": 1.7298427015729843e-05,
+      "loss": 1.2257,
+      "step": 1710
+    },
+    {
+      "epoch": 0.8982895939417679,
+      "grad_norm": 5.54590368270874,
+      "learning_rate": 1.7276427235727643e-05,
+      "loss": 1.2001,
+      "step": 1720
+    },
+    {
+      "epoch": 0.903512207860034,
+      "grad_norm": 5.4851508140563965,
+      "learning_rate": 1.7254427455725443e-05,
+      "loss": 1.2087,
+      "step": 1730
+    },
+    {
+      "epoch": 0.9087348217783,
+      "grad_norm": 5.15001106262207,
+      "learning_rate": 1.7232427675723246e-05,
+      "loss": 1.178,
+      "step": 1740
+    },
+    {
+      "epoch": 0.9139574356965662,
+      "grad_norm": 5.020768165588379,
+      "learning_rate": 1.7210427895721046e-05,
+      "loss": 1.1614,
+      "step": 1750
+    },
+    {
+      "epoch": 0.9191800496148322,
+      "grad_norm": 4.394384384155273,
+      "learning_rate": 1.7188428115718846e-05,
+      "loss": 1.1276,
+      "step": 1760
+    },
+    {
+      "epoch": 0.9244026635330983,
+      "grad_norm": 4.4511237144470215,
+      "learning_rate": 1.7166428335716645e-05,
+      "loss": 1.0134,
+      "step": 1770
+    },
+    {
+      "epoch": 0.9296252774513644,
+      "grad_norm": 5.19103479385376,
+      "learning_rate": 1.7144428555714445e-05,
+      "loss": 1.1161,
+      "step": 1780
+    },
+    {
+      "epoch": 0.9348478913696305,
+      "grad_norm": 5.5525126457214355,
+      "learning_rate": 1.7122428775712245e-05,
+      "loss": 1.0225,
+      "step": 1790
+    },
+    {
+      "epoch": 0.9400705052878966,
+      "grad_norm": 6.351354598999023,
+      "learning_rate": 1.7100428995710045e-05,
+      "loss": 1.0203,
+      "step": 1800
+    },
+    {
+      "epoch": 0.9452931192061627,
+      "grad_norm": 4.5916666984558105,
+      "learning_rate": 1.7078429215707845e-05,
+      "loss": 1.1395,
+      "step": 1810
+    },
+    {
+      "epoch": 0.9505157331244288,
+      "grad_norm": 5.519266128540039,
+      "learning_rate": 1.7056429435705645e-05,
+      "loss": 0.9899,
+      "step": 1820
+    },
+    {
+      "epoch": 0.9557383470426949,
+      "grad_norm": 5.209677696228027,
+      "learning_rate": 1.7034429655703444e-05,
+      "loss": 1.3285,
+      "step": 1830
+    },
+    {
+      "epoch": 0.960960960960961,
+      "grad_norm": 5.857298374176025,
+      "learning_rate": 1.7012429875701244e-05,
+      "loss": 1.2475,
+      "step": 1840
+    },
+    {
+      "epoch": 0.966183574879227,
+      "grad_norm": 6.121822834014893,
+      "learning_rate": 1.6990430095699044e-05,
+      "loss": 1.0665,
+      "step": 1850
+    },
+    {
+      "epoch": 0.9714061887974932,
+      "grad_norm": 4.8332295417785645,
+      "learning_rate": 1.6968430315696844e-05,
+      "loss": 1.179,
+      "step": 1860
+    },
+    {
+      "epoch": 0.9766288027157592,
+      "grad_norm": 5.4480695724487305,
+      "learning_rate": 1.6946430535694644e-05,
+      "loss": 0.9849,
+      "step": 1870
+    },
+    {
+      "epoch": 0.9818514166340253,
+      "grad_norm": 5.037874698638916,
+      "learning_rate": 1.6924430755692443e-05,
+      "loss": 1.0214,
+      "step": 1880
+    },
+    {
+      "epoch": 0.9870740305522914,
+      "grad_norm": 4.6786112785339355,
+      "learning_rate": 1.6902430975690243e-05,
+      "loss": 1.1205,
+      "step": 1890
+    },
+    {
+      "epoch": 0.9922966444705575,
+      "grad_norm": 5.271369457244873,
+      "learning_rate": 1.6880431195688047e-05,
+      "loss": 1.1756,
+      "step": 1900
+    },
+    {
+      "epoch": 0.9975192583888236,
+      "grad_norm": 5.059353828430176,
+      "learning_rate": 1.6858431415685846e-05,
+      "loss": 1.1025,
+      "step": 1910
+    },
+    {
+      "epoch": 1.0031335683509597,
+      "grad_norm": 4.742678165435791,
+      "learning_rate": 1.6836431635683646e-05,
+      "loss": 0.9805,
+      "step": 1920
+    },
+    {
+      "epoch": 1.0083561822692257,
+      "grad_norm": 4.218075275421143,
+      "learning_rate": 1.6814431855681446e-05,
+      "loss": 0.8862,
+      "step": 1930
+    },
+    {
+      "epoch": 1.0135787961874918,
+      "grad_norm": 7.259308338165283,
+      "learning_rate": 1.6792432075679242e-05,
+      "loss": 0.7837,
+      "step": 1940
+    },
+    {
+      "epoch": 1.0188014101057579,
+      "grad_norm": 3.9883408546447754,
+      "learning_rate": 1.6770432295677042e-05,
+      "loss": 0.7887,
+      "step": 1950
+    },
+    {
+      "epoch": 1.024024024024024,
+      "grad_norm": 5.002141952514648,
+      "learning_rate": 1.6748432515674845e-05,
+      "loss": 0.8512,
+      "step": 1960
+    },
+    {
+      "epoch": 1.0292466379422902,
+      "grad_norm": 4.857590198516846,
+      "learning_rate": 1.6726432735672645e-05,
+      "loss": 0.8973,
+      "step": 1970
+    },
+    {
+      "epoch": 1.0344692518605563,
+      "grad_norm": 5.385519981384277,
+      "learning_rate": 1.6704432955670445e-05,
+      "loss": 0.8722,
+      "step": 1980
+    },
+    {
+      "epoch": 1.0396918657788223,
+      "grad_norm": 4.655357360839844,
+      "learning_rate": 1.6682433175668245e-05,
+      "loss": 0.9631,
+      "step": 1990
+    },
+    {
+      "epoch": 1.0449144796970884,
+      "grad_norm": 4.768300533294678,
+      "learning_rate": 1.6660433395666045e-05,
+      "loss": 0.9284,
+      "step": 2000
+    },
+    {
+      "epoch": 1.0449144796970884,
+      "eval_loss": 1.062979817390442,
+      "eval_runtime": 46.384,
+      "eval_samples_per_second": 36.694,
+      "eval_steps_per_second": 4.592,
+      "step": 2000
+    },
+    {
+      "epoch": 1.0501370936153545,
+      "grad_norm": 5.307561874389648,
+      "learning_rate": 1.6638433615663845e-05,
+      "loss": 1.1394,
+      "step": 2010
+    },
+    {
+      "epoch": 1.0553597075336205,
+      "grad_norm": 6.255516529083252,
+      "learning_rate": 1.6616433835661644e-05,
+      "loss": 0.933,
+      "step": 2020
+    },
+    {
+      "epoch": 1.0605823214518866,
+      "grad_norm": 4.4736433029174805,
+      "learning_rate": 1.6594434055659444e-05,
+      "loss": 0.8371,
+      "step": 2030
+    },
+    {
+      "epoch": 1.0658049353701529,
+      "grad_norm": 3.445369005203247,
+      "learning_rate": 1.6572434275657244e-05,
+      "loss": 0.8323,
+      "step": 2040
+    },
+    {
+      "epoch": 1.071027549288419,
+      "grad_norm": 3.695296287536621,
+      "learning_rate": 1.6550434495655044e-05,
+      "loss": 0.8505,
+      "step": 2050
+    },
+    {
+      "epoch": 1.076250163206685,
+      "grad_norm": 5.231265544891357,
+      "learning_rate": 1.6528434715652847e-05,
+      "loss": 0.8781,
+      "step": 2060
+    },
+    {
+      "epoch": 1.081472777124951,
+      "grad_norm": 4.590741157531738,
+      "learning_rate": 1.6506434935650647e-05,
+      "loss": 0.8584,
+      "step": 2070
+    },
+    {
+      "epoch": 1.086695391043217,
+      "grad_norm": 5.293918132781982,
+      "learning_rate": 1.6484435155648443e-05,
+      "loss": 0.7935,
+      "step": 2080
+    },
+    {
+      "epoch": 1.0919180049614832,
+      "grad_norm": 5.582162380218506,
+      "learning_rate": 1.6462435375646243e-05,
+      "loss": 0.8391,
+      "step": 2090
+    },
+    {
+      "epoch": 1.0971406188797492,
+      "grad_norm": 4.409743785858154,
+      "learning_rate": 1.6440435595644043e-05,
+      "loss": 0.8929,
+      "step": 2100
+    },
+    {
+      "epoch": 1.1023632327980155,
+      "grad_norm": 6.849720001220703,
+      "learning_rate": 1.6418435815641843e-05,
+      "loss": 1.0301,
+      "step": 2110
+    },
+    {
+      "epoch": 1.1075858467162816,
+      "grad_norm": 4.769338607788086,
+      "learning_rate": 1.6396436035639646e-05,
+      "loss": 0.8172,
+      "step": 2120
+    },
+    {
+      "epoch": 1.1128084606345476,
+      "grad_norm": 5.646458625793457,
+      "learning_rate": 1.6374436255637446e-05,
+      "loss": 0.8616,
+      "step": 2130
+    },
+    {
+      "epoch": 1.1180310745528137,
+      "grad_norm": 5.374930381774902,
+      "learning_rate": 1.6352436475635246e-05,
+      "loss": 0.8015,
+      "step": 2140
+    },
+    {
+      "epoch": 1.1232536884710798,
+      "grad_norm": 3.9189627170562744,
+      "learning_rate": 1.6330436695633046e-05,
+      "loss": 0.8825,
+      "step": 2150
+    },
+    {
+      "epoch": 1.1284763023893458,
+      "grad_norm": 4.851017951965332,
+      "learning_rate": 1.6308436915630845e-05,
+      "loss": 0.7428,
+      "step": 2160
+    },
+    {
+      "epoch": 1.1336989163076119,
+      "grad_norm": 5.55581521987915,
+      "learning_rate": 1.6286437135628645e-05,
+      "loss": 0.8404,
+      "step": 2170
+    },
+    {
+      "epoch": 1.1389215302258782,
+      "grad_norm": 6.765121936798096,
+      "learning_rate": 1.6264437355626445e-05,
+      "loss": 0.9473,
+      "step": 2180
+    },
+    {
+      "epoch": 1.1441441441441442,
+      "grad_norm": 4.86257266998291,
+      "learning_rate": 1.6242437575624245e-05,
+      "loss": 0.8989,
+      "step": 2190
+    },
+    {
+      "epoch": 1.1493667580624103,
+      "grad_norm": 5.461528301239014,
+      "learning_rate": 1.6220437795622045e-05,
+      "loss": 0.7735,
+      "step": 2200
+    },
+    {
+      "epoch": 1.1545893719806763,
+      "grad_norm": 5.794458389282227,
+      "learning_rate": 1.6198438015619848e-05,
+      "loss": 0.8712,
+      "step": 2210
+    },
+    {
+      "epoch": 1.1598119858989424,
+      "grad_norm": 5.1557841300964355,
+      "learning_rate": 1.6176438235617644e-05,
+      "loss": 0.9436,
+      "step": 2220
+    },
+    {
+      "epoch": 1.1650345998172085,
+      "grad_norm": 5.397255897521973,
+      "learning_rate": 1.6154438455615444e-05,
+      "loss": 0.9447,
+      "step": 2230
+    },
+    {
+      "epoch": 1.1702572137354745,
+      "grad_norm": 4.012696743011475,
+      "learning_rate": 1.6132438675613244e-05,
+      "loss": 0.7914,
+      "step": 2240
+    },
+    {
+      "epoch": 1.1754798276537408,
+      "grad_norm": 4.5587286949157715,
+      "learning_rate": 1.6110438895611044e-05,
+      "loss": 0.849,
+      "step": 2250
+    },
+    {
+      "epoch": 1.1807024415720069,
+      "grad_norm": 5.750669956207275,
+      "learning_rate": 1.6088439115608844e-05,
+      "loss": 0.9119,
+      "step": 2260
+    },
+    {
+      "epoch": 1.185925055490273,
+      "grad_norm": 4.828431606292725,
+      "learning_rate": 1.6066439335606643e-05,
+      "loss": 0.8488,
+      "step": 2270
+    },
+    {
+      "epoch": 1.191147669408539,
+      "grad_norm": 4.56440544128418,
+      "learning_rate": 1.6044439555604447e-05,
+      "loss": 0.8229,
+      "step": 2280
+    },
+    {
+      "epoch": 1.196370283326805,
+      "grad_norm": 4.551214218139648,
+      "learning_rate": 1.6022439775602246e-05,
+      "loss": 0.8505,
+      "step": 2290
+    },
+    {
+      "epoch": 1.2015928972450711,
+      "grad_norm": 5.286618232727051,
+      "learning_rate": 1.6000439995600046e-05,
+      "loss": 0.8631,
+      "step": 2300
+    },
+    {
+      "epoch": 1.2068155111633372,
+      "grad_norm": 5.9425482749938965,
+      "learning_rate": 1.5978440215597846e-05,
+      "loss": 0.6281,
+      "step": 2310
+    },
+    {
+      "epoch": 1.2120381250816032,
+      "grad_norm": 3.729724884033203,
+      "learning_rate": 1.5956440435595646e-05,
+      "loss": 0.8309,
+      "step": 2320
+    },
+    {
+      "epoch": 1.2172607389998695,
+      "grad_norm": 8.449934959411621,
+      "learning_rate": 1.5934440655593446e-05,
+      "loss": 0.9191,
+      "step": 2330
+    },
+    {
+      "epoch": 1.2224833529181356,
+      "grad_norm": 5.1881842613220215,
+      "learning_rate": 1.5912440875591246e-05,
+      "loss": 0.8585,
+      "step": 2340
+    },
+    {
+      "epoch": 1.2277059668364017,
+      "grad_norm": 6.055355548858643,
+      "learning_rate": 1.5890441095589045e-05,
+      "loss": 0.8272,
+      "step": 2350
+    },
+    {
+      "epoch": 1.2329285807546677,
+      "grad_norm": 4.473712921142578,
+      "learning_rate": 1.5868441315586845e-05,
+      "loss": 0.871,
+      "step": 2360
+    },
+    {
+      "epoch": 1.2381511946729338,
+      "grad_norm": 3.1991331577301025,
+      "learning_rate": 1.5846441535584645e-05,
+      "loss": 0.8593,
+      "step": 2370
+    },
+    {
+      "epoch": 1.2433738085911998,
+      "grad_norm": 5.2965288162231445,
+      "learning_rate": 1.5824441755582445e-05,
+      "loss": 0.8573,
+      "step": 2380
+    },
+    {
+      "epoch": 1.248596422509466,
+      "grad_norm": 5.760184288024902,
+      "learning_rate": 1.5802441975580245e-05,
+      "loss": 0.9083,
+      "step": 2390
+    },
+    {
+      "epoch": 1.253819036427732,
+      "grad_norm": 4.14992618560791,
+      "learning_rate": 1.5780442195578045e-05,
+      "loss": 0.8069,
+      "step": 2400
+    },
+    {
+      "epoch": 1.2590416503459982,
+      "grad_norm": 5.517070770263672,
+      "learning_rate": 1.5758442415575844e-05,
+      "loss": 0.8448,
+      "step": 2410
+    },
+    {
+      "epoch": 1.2642642642642643,
+      "grad_norm": 4.883249282836914,
+      "learning_rate": 1.5736442635573644e-05,
+      "loss": 0.9297,
+      "step": 2420
+    },
+    {
+      "epoch": 1.2694868781825304,
+      "grad_norm": 4.5844950675964355,
+      "learning_rate": 1.5714442855571444e-05,
+      "loss": 0.7879,
+      "step": 2430
+    },
+    {
+      "epoch": 1.2747094921007964,
+      "grad_norm": 6.477102279663086,
+      "learning_rate": 1.5692443075569247e-05,
+      "loss": 0.9081,
+      "step": 2440
+    },
+    {
+      "epoch": 1.2799321060190625,
+      "grad_norm": 4.495989799499512,
+      "learning_rate": 1.5670443295567047e-05,
+      "loss": 0.9186,
+      "step": 2450
+    },
+    {
+      "epoch": 1.2851547199373288,
+      "grad_norm": 5.04586124420166,
+      "learning_rate": 1.5648443515564847e-05,
+      "loss": 0.8438,
+      "step": 2460
+    },
+    {
+      "epoch": 1.2903773338555946,
+      "grad_norm": 4.635850429534912,
+      "learning_rate": 1.5626443735562647e-05,
+      "loss": 0.9261,
+      "step": 2470
+    },
+    {
+      "epoch": 1.295599947773861,
+      "grad_norm": 6.3529253005981445,
+      "learning_rate": 1.5604443955560446e-05,
+      "loss": 0.9077,
+      "step": 2480
+    },
+    {
+      "epoch": 1.300822561692127,
+      "grad_norm": 5.649443626403809,
+      "learning_rate": 1.5582444175558246e-05,
+      "loss": 0.8207,
+      "step": 2490
+    },
+    {
+      "epoch": 1.306045175610393,
+      "grad_norm": 4.3899760246276855,
+      "learning_rate": 1.5560444395556046e-05,
+      "loss": 0.8617,
+      "step": 2500
+    },
+    {
+      "epoch": 1.306045175610393,
+      "eval_loss": 1.0080269575119019,
+      "eval_runtime": 46.3858,
+      "eval_samples_per_second": 36.692,
+      "eval_steps_per_second": 4.592,
+      "step": 2500
+    },
+    {
+      "epoch": 1.311267789528659,
+      "grad_norm": 4.011806011199951,
+      "learning_rate": 1.5538444615553846e-05,
+      "loss": 0.7535,
+      "step": 2510
+    },
+    {
+      "epoch": 1.3164904034469251,
+      "grad_norm": 6.140223026275635,
+      "learning_rate": 1.5516444835551646e-05,
+      "loss": 0.7793,
+      "step": 2520
+    },
+    {
+      "epoch": 1.3217130173651912,
+      "grad_norm": 3.5991857051849365,
+      "learning_rate": 1.5494445055549446e-05,
+      "loss": 0.8774,
+      "step": 2530
+    },
+    {
+      "epoch": 1.3269356312834573,
+      "grad_norm": 4.690261363983154,
+      "learning_rate": 1.5472445275547245e-05,
+      "loss": 0.8758,
+      "step": 2540
+    },
+    {
+      "epoch": 1.3321582452017235,
+      "grad_norm": 4.750207424163818,
+      "learning_rate": 1.5450445495545045e-05,
+      "loss": 0.9743,
+      "step": 2550
+    },
+    {
+      "epoch": 1.3373808591199896,
+      "grad_norm": 6.83930778503418,
+      "learning_rate": 1.5428445715542845e-05,
+      "loss": 0.8209,
+      "step": 2560
+    },
+    {
+      "epoch": 1.3426034730382557,
+      "grad_norm": 5.2934370040893555,
+      "learning_rate": 1.5406445935540645e-05,
+      "loss": 0.8453,
+      "step": 2570
+    },
+    {
+      "epoch": 1.3478260869565217,
+      "grad_norm": 5.174552917480469,
+      "learning_rate": 1.5384446155538445e-05,
+      "loss": 0.7369,
+      "step": 2580
+    },
+    {
+      "epoch": 1.3530487008747878,
+      "grad_norm": 4.970681667327881,
+      "learning_rate": 1.5362446375536248e-05,
+      "loss": 0.7854,
+      "step": 2590
+    },
+    {
+      "epoch": 1.3582713147930539,
+      "grad_norm": 4.746755599975586,
+      "learning_rate": 1.5340446595534048e-05,
+      "loss": 0.7734,
+      "step": 2600
+    },
+    {
+      "epoch": 1.36349392871132,
+      "grad_norm": 2.6767237186431885,
+      "learning_rate": 1.5318446815531848e-05,
+      "loss": 0.6543,
+      "step": 2610
+    },
+    {
+      "epoch": 1.3687165426295862,
+      "grad_norm": 5.292436122894287,
+      "learning_rate": 1.5296447035529647e-05,
+      "loss": 0.8073,
+      "step": 2620
+    },
+    {
+      "epoch": 1.3739391565478523,
+      "grad_norm": 5.2635626792907715,
+      "learning_rate": 1.5274447255527447e-05,
+      "loss": 0.7409,
+      "step": 2630
+    },
+    {
+      "epoch": 1.3791617704661183,
+      "grad_norm": 5.708250522613525,
+      "learning_rate": 1.5252447475525245e-05,
+      "loss": 0.9656,
+      "step": 2640
+    },
+    {
+      "epoch": 1.3843843843843844,
+      "grad_norm": 5.420493125915527,
+      "learning_rate": 1.5230447695523045e-05,
+      "loss": 0.8856,
+      "step": 2650
+    },
+    {
+      "epoch": 1.3896069983026504,
+      "grad_norm": 4.980160713195801,
+      "learning_rate": 1.5208447915520845e-05,
+      "loss": 0.8956,
+      "step": 2660
+    },
+    {
+      "epoch": 1.3948296122209165,
+      "grad_norm": 6.127252101898193,
+      "learning_rate": 1.5186448135518645e-05,
+      "loss": 0.8999,
+      "step": 2670
+    },
+    {
+      "epoch": 1.4000522261391826,
+      "grad_norm": 5.1474080085754395,
+      "learning_rate": 1.5164448355516446e-05,
+      "loss": 0.8222,
+      "step": 2680
+    },
+    {
+      "epoch": 1.4052748400574488,
+      "grad_norm": 4.6120476722717285,
+      "learning_rate": 1.5142448575514246e-05,
+      "loss": 0.8835,
+      "step": 2690
+    },
+    {
+      "epoch": 1.410497453975715,
+      "grad_norm": 4.124380111694336,
+      "learning_rate": 1.5120448795512046e-05,
+      "loss": 0.7693,
+      "step": 2700
+    },
+    {
+      "epoch": 1.415720067893981,
+      "grad_norm": 5.343372821807861,
+      "learning_rate": 1.5098449015509846e-05,
+      "loss": 0.7564,
+      "step": 2710
+    },
+    {
+      "epoch": 1.420942681812247,
+      "grad_norm": 4.66859245300293,
+      "learning_rate": 1.5076449235507646e-05,
+      "loss": 0.9316,
+      "step": 2720
+    },
+    {
+      "epoch": 1.426165295730513,
+      "grad_norm": 5.207456111907959,
+      "learning_rate": 1.5054449455505447e-05,
+      "loss": 0.8185,
+      "step": 2730
+    },
+    {
+      "epoch": 1.4313879096487792,
+      "grad_norm": 5.222368240356445,
+      "learning_rate": 1.5032449675503247e-05,
+      "loss": 0.763,
+      "step": 2740
+    },
+    {
+      "epoch": 1.4366105235670452,
+      "grad_norm": 4.925426959991455,
+      "learning_rate": 1.5010449895501047e-05,
+      "loss": 0.843,
+      "step": 2750
+    },
+    {
+      "epoch": 1.4418331374853115,
+      "grad_norm": 4.546454429626465,
+      "learning_rate": 1.4988450115498847e-05,
+      "loss": 0.8244,
+      "step": 2760
+    },
+    {
+      "epoch": 1.4470557514035776,
+      "grad_norm": 5.662976264953613,
+      "learning_rate": 1.4966450335496646e-05,
+      "loss": 0.7424,
+      "step": 2770
+    },
+    {
+      "epoch": 1.4522783653218436,
+      "grad_norm": 6.152126789093018,
+      "learning_rate": 1.4944450555494445e-05,
+      "loss": 0.685,
+      "step": 2780
+    },
+    {
+      "epoch": 1.4575009792401097,
+      "grad_norm": 4.776817321777344,
+      "learning_rate": 1.4922450775492246e-05,
+      "loss": 0.8083,
+      "step": 2790
+    },
+    {
+      "epoch": 1.4627235931583757,
+      "grad_norm": 3.0789973735809326,
+      "learning_rate": 1.4900450995490046e-05,
+      "loss": 0.7356,
+      "step": 2800
+    },
+    {
+      "epoch": 1.4679462070766418,
+      "grad_norm": 4.335427284240723,
+      "learning_rate": 1.4878451215487846e-05,
+      "loss": 0.8127,
+      "step": 2810
+    },
+    {
+      "epoch": 1.4731688209949079,
+      "grad_norm": 6.07483434677124,
+      "learning_rate": 1.4856451435485646e-05,
+      "loss": 0.7791,
+      "step": 2820
+    },
+    {
+      "epoch": 1.4783914349131742,
+      "grad_norm": 5.220136642456055,
+      "learning_rate": 1.4834451655483445e-05,
+      "loss": 0.715,
+      "step": 2830
+    },
+    {
+      "epoch": 1.4836140488314402,
+      "grad_norm": 5.2860426902771,
+      "learning_rate": 1.4812451875481247e-05,
+      "loss": 0.8689,
+      "step": 2840
+    },
+    {
+      "epoch": 1.4888366627497063,
+      "grad_norm": 4.845248222351074,
+      "learning_rate": 1.4790452095479047e-05,
+      "loss": 0.8165,
+      "step": 2850
+    },
+    {
+      "epoch": 1.4940592766679723,
+      "grad_norm": 5.356618404388428,
+      "learning_rate": 1.4768452315476847e-05,
+      "loss": 0.779,
+      "step": 2860
+    },
+    {
+      "epoch": 1.4992818905862384,
+      "grad_norm": 5.140944957733154,
+      "learning_rate": 1.4746452535474646e-05,
+      "loss": 0.7765,
+      "step": 2870
+    },
+    {
+      "epoch": 1.5045045045045045,
+      "grad_norm": 4.167046070098877,
+      "learning_rate": 1.4724452755472446e-05,
+      "loss": 0.7353,
+      "step": 2880
+    },
+    {
+      "epoch": 1.5097271184227705,
+      "grad_norm": 6.490098476409912,
+      "learning_rate": 1.4702452975470248e-05,
+      "loss": 0.8059,
+      "step": 2890
+    },
+    {
+      "epoch": 1.5149497323410368,
+      "grad_norm": 5.053187847137451,
+      "learning_rate": 1.4680453195468048e-05,
+      "loss": 0.8564,
+      "step": 2900
+    },
+    {
+      "epoch": 1.5201723462593026,
+      "grad_norm": 4.322595119476318,
+      "learning_rate": 1.4658453415465847e-05,
+      "loss": 0.8182,
+      "step": 2910
+    },
+    {
+      "epoch": 1.525394960177569,
+      "grad_norm": 3.603776454925537,
+      "learning_rate": 1.4636453635463646e-05,
+      "loss": 0.9616,
+      "step": 2920
+    },
+    {
+      "epoch": 1.530617574095835,
+      "grad_norm": 4.849047660827637,
+      "learning_rate": 1.4614453855461445e-05,
+      "loss": 0.7226,
+      "step": 2930
+    },
+    {
+      "epoch": 1.535840188014101,
+      "grad_norm": 6.261702537536621,
+      "learning_rate": 1.4592454075459245e-05,
+      "loss": 0.7724,
+      "step": 2940
+    },
+    {
+      "epoch": 1.541062801932367,
+      "grad_norm": 4.070964336395264,
+      "learning_rate": 1.4570454295457047e-05,
+      "loss": 0.6157,
+      "step": 2950
+    },
+    {
+      "epoch": 1.5462854158506332,
+      "grad_norm": 5.896931171417236,
+      "learning_rate": 1.4548454515454847e-05,
+      "loss": 0.7687,
+      "step": 2960
+    },
+    {
+      "epoch": 1.5515080297688995,
+      "grad_norm": 4.215420722961426,
+      "learning_rate": 1.4526454735452646e-05,
+      "loss": 0.7984,
+      "step": 2970
+    },
+    {
+      "epoch": 1.5567306436871653,
+      "grad_norm": 5.551188945770264,
+      "learning_rate": 1.4504454955450446e-05,
+      "loss": 0.7284,
+      "step": 2980
+    },
+    {
+      "epoch": 1.5619532576054316,
+      "grad_norm": 7.92472505569458,
+      "learning_rate": 1.4482455175448246e-05,
+      "loss": 0.7007,
+      "step": 2990
+    },
+    {
+      "epoch": 1.5671758715236976,
+      "grad_norm": 4.330050468444824,
+      "learning_rate": 1.4460455395446048e-05,
+      "loss": 0.948,
+      "step": 3000
+    },
+    {
+      "epoch": 1.5671758715236976,
+      "eval_loss": 0.9684945344924927,
+      "eval_runtime": 46.3792,
+      "eval_samples_per_second": 36.697,
+      "eval_steps_per_second": 4.593,
+      "step": 3000
+    },
+    {
+      "epoch": 1.5723984854419637,
+      "grad_norm": 4.803574085235596,
+      "learning_rate": 1.4438455615443847e-05,
+      "loss": 0.7733,
+      "step": 3010
+    },
+    {
+      "epoch": 1.5776210993602298,
+      "grad_norm": 3.641155481338501,
+      "learning_rate": 1.4416455835441647e-05,
+      "loss": 0.7893,
+      "step": 3020
+    },
+    {
+      "epoch": 1.5828437132784958,
+      "grad_norm": 4.811826705932617,
+      "learning_rate": 1.4394456055439447e-05,
+      "loss": 0.9068,
+      "step": 3030
+    },
+    {
+      "epoch": 1.588066327196762,
+      "grad_norm": 4.911914348602295,
+      "learning_rate": 1.4372456275437247e-05,
+      "loss": 0.8413,
+      "step": 3040
+    },
+    {
+      "epoch": 1.593288941115028,
+      "grad_norm": 6.21580171585083,
+      "learning_rate": 1.4350456495435048e-05,
+      "loss": 0.7968,
+      "step": 3050
+    },
+    {
+      "epoch": 1.5985115550332942,
+      "grad_norm": 3.3916568756103516,
+      "learning_rate": 1.4328456715432846e-05,
+      "loss": 0.7357,
+      "step": 3060
+    },
+    {
+      "epoch": 1.6037341689515603,
+      "grad_norm": 4.651008605957031,
+      "learning_rate": 1.4306456935430646e-05,
+      "loss": 0.823,
+      "step": 3070
+    },
+    {
+      "epoch": 1.6089567828698264,
+      "grad_norm": 3.533646583557129,
+      "learning_rate": 1.4284457155428446e-05,
+      "loss": 0.8333,
+      "step": 3080
+    },
+    {
+      "epoch": 1.6141793967880924,
+      "grad_norm": 5.906960487365723,
+      "learning_rate": 1.4262457375426246e-05,
+      "loss": 0.9569,
+      "step": 3090
+    },
+    {
+      "epoch": 1.6194020107063585,
+      "grad_norm": 5.581897258758545,
+      "learning_rate": 1.4240457595424046e-05,
+      "loss": 0.7276,
+      "step": 3100
+    },
+    {
+      "epoch": 1.6246246246246248,
+      "grad_norm": 5.3108439445495605,
+      "learning_rate": 1.4218457815421847e-05,
+      "loss": 0.7249,
+      "step": 3110
+    },
+    {
+      "epoch": 1.6298472385428906,
+      "grad_norm": 9.554184913635254,
+      "learning_rate": 1.4196458035419647e-05,
+      "loss": 0.8382,
+      "step": 3120
+    },
+    {
+      "epoch": 1.6350698524611569,
+      "grad_norm": 5.487770080566406,
+      "learning_rate": 1.4174458255417447e-05,
+      "loss": 0.8928,
+      "step": 3130
+    },
+    {
+      "epoch": 1.640292466379423,
+      "grad_norm": 4.557746887207031,
+      "learning_rate": 1.4152458475415247e-05,
+      "loss": 0.7369,
+      "step": 3140
+    },
+    {
+      "epoch": 1.645515080297689,
+      "grad_norm": 4.392518043518066,
+      "learning_rate": 1.4130458695413047e-05,
+      "loss": 0.8775,
+      "step": 3150
+    },
+    {
+      "epoch": 1.650737694215955,
+      "grad_norm": 3.4565000534057617,
+      "learning_rate": 1.4108458915410848e-05,
+      "loss": 0.7641,
+      "step": 3160
+    },
+    {
+      "epoch": 1.6559603081342211,
+      "grad_norm": 3.457012891769409,
+      "learning_rate": 1.4086459135408648e-05,
+      "loss": 0.8069,
+      "step": 3170
+    },
+    {
+      "epoch": 1.6611829220524874,
+      "grad_norm": 4.115066051483154,
+      "learning_rate": 1.4064459355406448e-05,
+      "loss": 0.8344,
+      "step": 3180
+    },
+    {
+      "epoch": 1.6664055359707532,
+      "grad_norm": 4.821453094482422,
+      "learning_rate": 1.4042459575404248e-05,
+      "loss": 0.6791,
+      "step": 3190
+    },
+    {
+      "epoch": 1.6716281498890195,
+      "grad_norm": 4.313604354858398,
+      "learning_rate": 1.4020459795402046e-05,
+      "loss": 0.7312,
+      "step": 3200
+    },
+    {
+      "epoch": 1.6768507638072856,
+      "grad_norm": 5.048585891723633,
+      "learning_rate": 1.3998460015399846e-05,
+      "loss": 0.8069,
+      "step": 3210
+    },
+    {
+      "epoch": 1.6820733777255517,
+      "grad_norm": 4.065299987792969,
+      "learning_rate": 1.3976460235397647e-05,
+      "loss": 0.8265,
+      "step": 3220
+    },
+    {
+      "epoch": 1.6872959916438177,
+      "grad_norm": 3.214991569519043,
+      "learning_rate": 1.3954460455395447e-05,
+      "loss": 0.7092,
+      "step": 3230
+    },
+    {
+      "epoch": 1.6925186055620838,
+      "grad_norm": 7.845371723175049,
+      "learning_rate": 1.3932460675393247e-05,
+      "loss": 0.7499,
+      "step": 3240
+    },
+    {
+      "epoch": 1.69774121948035,
+      "grad_norm": 5.4474921226501465,
+      "learning_rate": 1.3910460895391046e-05,
+      "loss": 0.781,
+      "step": 3250
+    },
+    {
+      "epoch": 1.702963833398616,
+      "grad_norm": 7.43993616104126,
+      "learning_rate": 1.3888461115388846e-05,
+      "loss": 0.8104,
+      "step": 3260
+    },
+    {
+      "epoch": 1.7081864473168822,
+      "grad_norm": 5.410302639007568,
+      "learning_rate": 1.3866461335386648e-05,
+      "loss": 0.909,
+      "step": 3270
+    },
+    {
+      "epoch": 1.7134090612351482,
+      "grad_norm": 3.9956214427948,
+      "learning_rate": 1.3844461555384448e-05,
+      "loss": 0.7744,
+      "step": 3280
+    },
+    {
+      "epoch": 1.7186316751534143,
+      "grad_norm": 5.384256839752197,
+      "learning_rate": 1.3822461775382247e-05,
+      "loss": 0.7898,
+      "step": 3290
+    },
+    {
+      "epoch": 1.7238542890716804,
+      "grad_norm": 4.181092739105225,
+      "learning_rate": 1.3800461995380047e-05,
+      "loss": 0.7584,
+      "step": 3300
+    },
+    {
+      "epoch": 1.7290769029899464,
+      "grad_norm": 4.823882102966309,
+      "learning_rate": 1.3778462215377849e-05,
+      "loss": 0.9165,
+      "step": 3310
+    },
+    {
+      "epoch": 1.7342995169082127,
+      "grad_norm": 5.339610576629639,
+      "learning_rate": 1.3756462435375649e-05,
+      "loss": 0.6867,
+      "step": 3320
+    },
+    {
+      "epoch": 1.7395221308264786,
+      "grad_norm": 3.4958078861236572,
+      "learning_rate": 1.3734462655373448e-05,
+      "loss": 0.9282,
+      "step": 3330
+    },
+    {
+      "epoch": 1.7447447447447448,
+      "grad_norm": 4.378716945648193,
+      "learning_rate": 1.3712462875371247e-05,
+      "loss": 0.5787,
+      "step": 3340
+    },
+    {
+      "epoch": 1.7499673586630107,
+      "grad_norm": 9.347067832946777,
+      "learning_rate": 1.3690463095369046e-05,
+      "loss": 0.7691,
+      "step": 3350
+    },
+    {
+      "epoch": 1.755189972581277,
+      "grad_norm": 3.417569398880005,
+      "learning_rate": 1.3668463315366846e-05,
+      "loss": 0.7411,
+      "step": 3360
+    },
+    {
+      "epoch": 1.760412586499543,
+      "grad_norm": 4.640798568725586,
+      "learning_rate": 1.3646463535364646e-05,
+      "loss": 0.7407,
+      "step": 3370
+    },
+    {
+      "epoch": 1.765635200417809,
+      "grad_norm": 5.5894341468811035,
+      "learning_rate": 1.3624463755362448e-05,
+      "loss": 0.9005,
+      "step": 3380
+    },
+    {
+      "epoch": 1.7708578143360754,
+      "grad_norm": 5.964491844177246,
+      "learning_rate": 1.3602463975360247e-05,
+      "loss": 0.8281,
+      "step": 3390
+    },
+    {
+      "epoch": 1.7760804282543412,
+      "grad_norm": 5.019880294799805,
+      "learning_rate": 1.3580464195358047e-05,
+      "loss": 0.7772,
+      "step": 3400
+    },
+    {
+      "epoch": 1.7813030421726075,
+      "grad_norm": 7.011587619781494,
+      "learning_rate": 1.3558464415355847e-05,
+      "loss": 0.6852,
+      "step": 3410
+    },
+    {
+      "epoch": 1.7865256560908733,
+      "grad_norm": 5.2039055824279785,
+      "learning_rate": 1.3536464635353647e-05,
+      "loss": 0.8141,
+      "step": 3420
+    },
+    {
+      "epoch": 1.7917482700091396,
+      "grad_norm": 5.673884868621826,
+      "learning_rate": 1.3514464855351448e-05,
+      "loss": 0.7204,
+      "step": 3430
+    },
+    {
+      "epoch": 1.7969708839274057,
+      "grad_norm": 4.41068172454834,
+      "learning_rate": 1.3492465075349248e-05,
+      "loss": 0.8504,
+      "step": 3440
+    },
+    {
+      "epoch": 1.8021934978456717,
+      "grad_norm": 4.449737071990967,
+      "learning_rate": 1.3470465295347048e-05,
+      "loss": 0.7652,
+      "step": 3450
+    },
+    {
+      "epoch": 1.807416111763938,
+      "grad_norm": 5.095055103302002,
+      "learning_rate": 1.3448465515344848e-05,
+      "loss": 0.8654,
+      "step": 3460
+    },
+    {
+      "epoch": 1.8126387256822039,
+      "grad_norm": 4.90816068649292,
+      "learning_rate": 1.342646573534265e-05,
+      "loss": 0.6933,
+      "step": 3470
+    },
+    {
+      "epoch": 1.8178613396004701,
+      "grad_norm": 3.835984468460083,
+      "learning_rate": 1.3406665933340667e-05,
+      "loss": 0.6106,
+      "step": 3480
+    },
+    {
+      "epoch": 1.823083953518736,
+      "grad_norm": 4.385377407073975,
+      "learning_rate": 1.3384666153338469e-05,
+      "loss": 0.797,
+      "step": 3490
+    },
+    {
+      "epoch": 1.8283065674370023,
+      "grad_norm": 5.261483192443848,
+      "learning_rate": 1.3362666373336269e-05,
+      "loss": 0.7343,
+      "step": 3500
+    },
+    {
+      "epoch": 1.8283065674370023,
+      "eval_loss": 0.9347423911094666,
+      "eval_runtime": 46.3797,
+      "eval_samples_per_second": 36.697,
+      "eval_steps_per_second": 4.593,
+      "step": 3500
+    },
+    {
+      "epoch": 1.8335291813552683,
+      "grad_norm": 5.269721508026123,
+      "learning_rate": 1.3340666593334068e-05,
+      "loss": 0.7558,
+      "step": 3510
+    },
+    {
+      "epoch": 1.8387517952735344,
+      "grad_norm": 4.808032512664795,
+      "learning_rate": 1.3318666813331868e-05,
+      "loss": 0.7597,
+      "step": 3520
+    },
+    {
+      "epoch": 1.8439744091918004,
+      "grad_norm": 5.560021877288818,
+      "learning_rate": 1.3296667033329668e-05,
+      "loss": 0.6428,
+      "step": 3530
+    },
+    {
+      "epoch": 1.8491970231100665,
+      "grad_norm": 4.8384013175964355,
+      "learning_rate": 1.3274667253327466e-05,
+      "loss": 0.8231,
+      "step": 3540
+    },
+    {
+      "epoch": 1.8544196370283328,
+      "grad_norm": 5.98041296005249,
+      "learning_rate": 1.3252667473325268e-05,
+      "loss": 0.8127,
+      "step": 3550
+    },
+    {
+      "epoch": 1.8596422509465986,
+      "grad_norm": 5.745590686798096,
+      "learning_rate": 1.3230667693323068e-05,
+      "loss": 0.7488,
+      "step": 3560
+    },
+    {
+      "epoch": 1.864864864864865,
+      "grad_norm": 6.292532920837402,
+      "learning_rate": 1.3208667913320867e-05,
+      "loss": 0.8194,
+      "step": 3570
+    },
+    {
+      "epoch": 1.870087478783131,
+      "grad_norm": 4.489058971405029,
+      "learning_rate": 1.3186668133318667e-05,
+      "loss": 0.8918,
+      "step": 3580
+    },
+    {
+      "epoch": 1.875310092701397,
+      "grad_norm": 7.152320861816406,
+      "learning_rate": 1.3164668353316467e-05,
+      "loss": 0.8746,
+      "step": 3590
+    },
+    {
+      "epoch": 1.880532706619663,
+      "grad_norm": 5.863579273223877,
+      "learning_rate": 1.3142668573314268e-05,
+      "loss": 0.6692,
+      "step": 3600
+    },
+    {
+      "epoch": 1.8857553205379292,
+      "grad_norm": 5.481932163238525,
+      "learning_rate": 1.3120668793312068e-05,
+      "loss": 0.9529,
+      "step": 3610
+    },
+    {
+      "epoch": 1.8909779344561954,
+      "grad_norm": 5.673171043395996,
+      "learning_rate": 1.3098669013309868e-05,
+      "loss": 0.7145,
+      "step": 3620
+    },
+    {
+      "epoch": 1.8962005483744613,
+      "grad_norm": 3.1669821739196777,
+      "learning_rate": 1.3076669233307668e-05,
+      "loss": 0.6498,
+      "step": 3630
+    },
+    {
+      "epoch": 1.9014231622927276,
+      "grad_norm": 5.213048934936523,
+      "learning_rate": 1.3054669453305468e-05,
+      "loss": 0.8128,
+      "step": 3640
+    },
+    {
+      "epoch": 1.9066457762109936,
+      "grad_norm": 3.5980887413024902,
+      "learning_rate": 1.303266967330327e-05,
+      "loss": 0.692,
+      "step": 3650
+    },
+    {
+      "epoch": 1.9118683901292597,
+      "grad_norm": 3.2975611686706543,
+      "learning_rate": 1.3010669893301069e-05,
+      "loss": 0.6698,
+      "step": 3660
+    },
+    {
+      "epoch": 1.9170910040475257,
+      "grad_norm": 4.254265308380127,
+      "learning_rate": 1.2988670113298869e-05,
+      "loss": 0.7827,
+      "step": 3670
+    },
+    {
+      "epoch": 1.9223136179657918,
+      "grad_norm": 4.849771022796631,
+      "learning_rate": 1.2966670333296667e-05,
+      "loss": 0.6954,
+      "step": 3680
+    },
+    {
+      "epoch": 1.927536231884058,
+      "grad_norm": 4.976161956787109,
+      "learning_rate": 1.2944670553294467e-05,
+      "loss": 0.6382,
+      "step": 3690
+    },
+    {
+      "epoch": 1.932758845802324,
+      "grad_norm": 5.311154365539551,
+      "learning_rate": 1.2922670773292267e-05,
+      "loss": 0.6891,
+      "step": 3700
+    },
+    {
+      "epoch": 1.9379814597205902,
+      "grad_norm": 6.911156177520752,
+      "learning_rate": 1.2900670993290068e-05,
+      "loss": 0.861,
+      "step": 3710
+    },
+    {
+      "epoch": 1.9432040736388563,
+      "grad_norm": 4.014489650726318,
+      "learning_rate": 1.2878671213287868e-05,
+      "loss": 0.8185,
+      "step": 3720
+    },
+    {
+      "epoch": 1.9484266875571223,
+      "grad_norm": 5.282459259033203,
+      "learning_rate": 1.2856671433285668e-05,
+      "loss": 0.7983,
+      "step": 3730
+    },
+    {
+      "epoch": 1.9536493014753884,
+      "grad_norm": 4.00164794921875,
+      "learning_rate": 1.2834671653283468e-05,
+      "loss": 0.7628,
+      "step": 3740
+    },
+    {
+      "epoch": 1.9588719153936545,
+      "grad_norm": 5.199289798736572,
+      "learning_rate": 1.2812671873281268e-05,
+      "loss": 0.8203,
+      "step": 3750
+    },
+    {
+      "epoch": 1.9640945293119207,
+      "grad_norm": 4.595690727233887,
+      "learning_rate": 1.2790672093279069e-05,
+      "loss": 0.6234,
+      "step": 3760
+    },
+    {
+      "epoch": 1.9693171432301866,
+      "grad_norm": 4.965597152709961,
+      "learning_rate": 1.2768672313276869e-05,
+      "loss": 0.7896,
+      "step": 3770
+    },
+    {
+      "epoch": 1.9745397571484529,
+      "grad_norm": 6.135955810546875,
+      "learning_rate": 1.2746672533274669e-05,
+      "loss": 0.7313,
+      "step": 3780
+    },
+    {
+      "epoch": 1.979762371066719,
+      "grad_norm": 3.6871345043182373,
+      "learning_rate": 1.2724672753272469e-05,
+      "loss": 0.9234,
+      "step": 3790
+    },
+    {
+      "epoch": 1.984984984984985,
+      "grad_norm": 4.930665493011475,
+      "learning_rate": 1.2702672973270268e-05,
+      "loss": 0.9072,
+      "step": 3800
+    },
+    {
+      "epoch": 1.990207598903251,
+      "grad_norm": 4.443655014038086,
+      "learning_rate": 1.268067319326807e-05,
+      "loss": 0.7923,
+      "step": 3810
+    },
+    {
+      "epoch": 1.9954302128215171,
+      "grad_norm": 4.998082637786865,
+      "learning_rate": 1.2658673413265868e-05,
+      "loss": 0.7933,
+      "step": 3820
+    },
+    {
+      "epoch": 2.001044522783653,
+      "grad_norm": 2.840280294418335,
+      "learning_rate": 1.2636673633263668e-05,
+      "loss": 0.7225,
+      "step": 3830
+    },
+    {
+      "epoch": 2.0062671367019194,
+      "grad_norm": 5.853653907775879,
+      "learning_rate": 1.2614673853261468e-05,
+      "loss": 0.548,
+      "step": 3840
+    },
+    {
+      "epoch": 2.011489750620185,
+      "grad_norm": 3.085155725479126,
+      "learning_rate": 1.2592674073259267e-05,
+      "loss": 0.5502,
+      "step": 3850
+    },
+    {
+      "epoch": 2.0167123645384515,
+      "grad_norm": 5.955913543701172,
+      "learning_rate": 1.2570674293257067e-05,
+      "loss": 0.5229,
+      "step": 3860
+    },
+    {
+      "epoch": 2.0219349784567178,
+      "grad_norm": 5.1260905265808105,
+      "learning_rate": 1.2548674513254869e-05,
+      "loss": 0.6059,
+      "step": 3870
+    },
+    {
+      "epoch": 2.0271575923749836,
+      "grad_norm": 2.684438943862915,
+      "learning_rate": 1.2526674733252669e-05,
+      "loss": 0.5465,
+      "step": 3880
+    },
+    {
+      "epoch": 2.03238020629325,
+      "grad_norm": 4.3153533935546875,
+      "learning_rate": 1.2504674953250468e-05,
+      "loss": 0.5319,
+      "step": 3890
+    },
+    {
+      "epoch": 2.0376028202115157,
+      "grad_norm": 5.260719299316406,
+      "learning_rate": 1.2482675173248268e-05,
+      "loss": 0.4983,
+      "step": 3900
+    },
+    {
+      "epoch": 2.042825434129782,
+      "grad_norm": 4.053082466125488,
+      "learning_rate": 1.2460675393246068e-05,
+      "loss": 0.6385,
+      "step": 3910
+    },
+    {
+      "epoch": 2.048048048048048,
+      "grad_norm": 4.619148254394531,
+      "learning_rate": 1.243867561324387e-05,
+      "loss": 0.7165,
+      "step": 3920
+    },
+    {
+      "epoch": 2.053270661966314,
+      "grad_norm": 6.127100944519043,
+      "learning_rate": 1.241667583324167e-05,
+      "loss": 0.5853,
+      "step": 3930
+    },
+    {
+      "epoch": 2.0584932758845804,
+      "grad_norm": 6.0645432472229,
+      "learning_rate": 1.239467605323947e-05,
+      "loss": 0.6654,
+      "step": 3940
+    },
+    {
+      "epoch": 2.0637158898028463,
+      "grad_norm": 4.367906093597412,
+      "learning_rate": 1.2372676273237269e-05,
+      "loss": 0.5563,
+      "step": 3950
+    },
+    {
+      "epoch": 2.0689385037211125,
+      "grad_norm": 6.469724178314209,
+      "learning_rate": 1.2350676493235067e-05,
+      "loss": 0.5929,
+      "step": 3960
+    },
+    {
+      "epoch": 2.0741611176393784,
+      "grad_norm": 3.372661828994751,
+      "learning_rate": 1.2328676713232867e-05,
+      "loss": 0.5706,
+      "step": 3970
+    },
+    {
+      "epoch": 2.0793837315576447,
+      "grad_norm": 6.373135566711426,
+      "learning_rate": 1.2306676933230669e-05,
+      "loss": 0.5192,
+      "step": 3980
+    },
+    {
+      "epoch": 2.0846063454759105,
+      "grad_norm": 5.228106498718262,
+      "learning_rate": 1.2284677153228468e-05,
+      "loss": 0.6097,
+      "step": 3990
+    },
+    {
+      "epoch": 2.089828959394177,
+      "grad_norm": 6.330024719238281,
+      "learning_rate": 1.2262677373226268e-05,
+      "loss": 0.5799,
+      "step": 4000
+    },
+    {
+      "epoch": 2.089828959394177,
+      "eval_loss": 0.9453912377357483,
+      "eval_runtime": 46.3882,
+      "eval_samples_per_second": 36.69,
+      "eval_steps_per_second": 4.592,
+      "step": 4000
+    },
+    {
+      "epoch": 2.095051573312443,
+      "grad_norm": 3.039299964904785,
+      "learning_rate": 1.2240677593224068e-05,
+      "loss": 0.6099,
+      "step": 4010
+    },
+    {
+      "epoch": 2.100274187230709,
+      "grad_norm": 5.28336763381958,
+      "learning_rate": 1.2218677813221868e-05,
+      "loss": 0.5661,
+      "step": 4020
+    },
+    {
+      "epoch": 2.105496801148975,
+      "grad_norm": 6.230477333068848,
+      "learning_rate": 1.219667803321967e-05,
+      "loss": 0.5823,
+      "step": 4030
+    },
+    {
+      "epoch": 2.110719415067241,
+      "grad_norm": 6.373923301696777,
+      "learning_rate": 1.217467825321747e-05,
+      "loss": 0.5674,
+      "step": 4040
+    },
+    {
+      "epoch": 2.1159420289855073,
+      "grad_norm": 4.496073246002197,
+      "learning_rate": 1.2152678473215269e-05,
+      "loss": 0.5449,
+      "step": 4050
+    },
+    {
+      "epoch": 2.121164642903773,
+      "grad_norm": 5.086572170257568,
+      "learning_rate": 1.2130678693213069e-05,
+      "loss": 0.478,
+      "step": 4060
+    },
+    {
+      "epoch": 2.1263872568220394,
+      "grad_norm": 3.9062867164611816,
+      "learning_rate": 1.2108678913210869e-05,
+      "loss": 0.4974,
+      "step": 4070
+    },
+    {
+      "epoch": 2.1316098707403057,
+      "grad_norm": 5.478116989135742,
+      "learning_rate": 1.208667913320867e-05,
+      "loss": 0.6237,
+      "step": 4080
+    },
+    {
+      "epoch": 2.1368324846585716,
+      "grad_norm": 4.788442611694336,
+      "learning_rate": 1.206467935320647e-05,
+      "loss": 0.4777,
+      "step": 4090
+    },
+    {
+      "epoch": 2.142055098576838,
+      "grad_norm": 5.6343889236450195,
+      "learning_rate": 1.2042679573204268e-05,
+      "loss": 0.5733,
+      "step": 4100
+    },
+    {
+      "epoch": 2.1472777124951037,
+      "grad_norm": 3.3089094161987305,
+      "learning_rate": 1.2020679793202068e-05,
+      "loss": 0.4911,
+      "step": 4110
+    },
+    {
+      "epoch": 2.15250032641337,
+      "grad_norm": 4.506272315979004,
+      "learning_rate": 1.1998680013199868e-05,
+      "loss": 0.516,
+      "step": 4120
+    },
+    {
+      "epoch": 2.157722940331636,
+      "grad_norm": 5.48958158493042,
+      "learning_rate": 1.1976680233197668e-05,
+      "loss": 0.5535,
+      "step": 4130
+    },
+    {
+      "epoch": 2.162945554249902,
+      "grad_norm": 4.435242652893066,
+      "learning_rate": 1.195468045319547e-05,
+      "loss": 0.5861,
+      "step": 4140
+    },
+    {
+      "epoch": 2.1681681681681684,
+      "grad_norm": 4.332143306732178,
+      "learning_rate": 1.1932680673193269e-05,
+      "loss": 0.5009,
+      "step": 4150
+    },
+    {
+      "epoch": 2.173390782086434,
+      "grad_norm": 4.4287848472595215,
+      "learning_rate": 1.1910680893191069e-05,
+      "loss": 0.4957,
+      "step": 4160
+    },
+    {
+      "epoch": 2.1786133960047005,
+      "grad_norm": 5.4941229820251465,
+      "learning_rate": 1.1888681113188869e-05,
+      "loss": 0.5236,
+      "step": 4170
+    },
+    {
+      "epoch": 2.1838360099229663,
+      "grad_norm": 3.3228816986083984,
+      "learning_rate": 1.1866681333186668e-05,
+      "loss": 0.4355,
+      "step": 4180
+    },
+    {
+      "epoch": 2.1890586238412326,
+      "grad_norm": 6.6818952560424805,
+      "learning_rate": 1.184468155318447e-05,
+      "loss": 0.4244,
+      "step": 4190
+    },
+    {
+      "epoch": 2.1942812377594985,
+      "grad_norm": 4.551324844360352,
+      "learning_rate": 1.182268177318227e-05,
+      "loss": 0.5724,
+      "step": 4200
+    },
+    {
+      "epoch": 2.1995038516777647,
+      "grad_norm": 3.3153746128082275,
+      "learning_rate": 1.180068199318007e-05,
+      "loss": 0.4918,
+      "step": 4210
+    },
+    {
+      "epoch": 2.204726465596031,
+      "grad_norm": 4.966384410858154,
+      "learning_rate": 1.177868221317787e-05,
+      "loss": 0.4823,
+      "step": 4220
+    },
+    {
+      "epoch": 2.209949079514297,
+      "grad_norm": 4.28817081451416,
+      "learning_rate": 1.1756682433175671e-05,
+      "loss": 0.4717,
+      "step": 4230
+    },
+    {
+      "epoch": 2.215171693432563,
+      "grad_norm": 3.8175127506256104,
+      "learning_rate": 1.173468265317347e-05,
+      "loss": 0.4902,
+      "step": 4240
+    },
+    {
+      "epoch": 2.220394307350829,
+      "grad_norm": 3.9820878505706787,
+      "learning_rate": 1.1712682873171269e-05,
+      "loss": 0.5859,
+      "step": 4250
+    },
+    {
+      "epoch": 2.2256169212690953,
+      "grad_norm": 4.197390079498291,
+      "learning_rate": 1.1690683093169069e-05,
+      "loss": 0.5415,
+      "step": 4260
+    },
+    {
+      "epoch": 2.230839535187361,
+      "grad_norm": 6.945038795471191,
+      "learning_rate": 1.1668683313166869e-05,
+      "loss": 0.5892,
+      "step": 4270
+    },
+    {
+      "epoch": 2.2360621491056274,
+      "grad_norm": 4.9087982177734375,
+      "learning_rate": 1.1646683533164668e-05,
+      "loss": 0.5109,
+      "step": 4280
+    },
+    {
+      "epoch": 2.2412847630238932,
+      "grad_norm": 5.819643497467041,
+      "learning_rate": 1.1624683753162468e-05,
+      "loss": 0.6031,
+      "step": 4290
+    },
+    {
+      "epoch": 2.2465073769421595,
+      "grad_norm": 5.164392948150635,
+      "learning_rate": 1.160268397316027e-05,
+      "loss": 0.5399,
+      "step": 4300
+    },
+    {
+      "epoch": 2.251729990860426,
+      "grad_norm": 5.87556791305542,
+      "learning_rate": 1.158068419315807e-05,
+      "loss": 0.5744,
+      "step": 4310
+    },
+    {
+      "epoch": 2.2569526047786916,
+      "grad_norm": 2.9850735664367676,
+      "learning_rate": 1.155868441315587e-05,
+      "loss": 0.4386,
+      "step": 4320
+    },
+    {
+      "epoch": 2.262175218696958,
+      "grad_norm": 4.8038530349731445,
+      "learning_rate": 1.153668463315367e-05,
+      "loss": 0.5148,
+      "step": 4330
+    },
+    {
+      "epoch": 2.2673978326152238,
+      "grad_norm": 5.2686004638671875,
+      "learning_rate": 1.151468485315147e-05,
+      "loss": 0.5089,
+      "step": 4340
+    },
+    {
+      "epoch": 2.27262044653349,
+      "grad_norm": 6.421589374542236,
+      "learning_rate": 1.149268507314927e-05,
+      "loss": 0.6375,
+      "step": 4350
+    },
+    {
+      "epoch": 2.2778430604517563,
+      "grad_norm": 5.114859104156494,
+      "learning_rate": 1.147068529314707e-05,
+      "loss": 0.4898,
+      "step": 4360
+    },
+    {
+      "epoch": 2.283065674370022,
+      "grad_norm": 3.2084782123565674,
+      "learning_rate": 1.144868551314487e-05,
+      "loss": 0.5357,
+      "step": 4370
+    },
+    {
+      "epoch": 2.2882882882882885,
+      "grad_norm": 4.514894962310791,
+      "learning_rate": 1.142668573314267e-05,
+      "loss": 0.4976,
+      "step": 4380
+    },
+    {
+      "epoch": 2.2935109022065543,
+      "grad_norm": 4.541710376739502,
+      "learning_rate": 1.1404685953140468e-05,
+      "loss": 0.5235,
+      "step": 4390
+    },
+    {
+      "epoch": 2.2987335161248206,
+      "grad_norm": 3.9480032920837402,
+      "learning_rate": 1.1382686173138268e-05,
+      "loss": 0.571,
+      "step": 4400
+    },
+    {
+      "epoch": 2.3039561300430864,
+      "grad_norm": 5.118963718414307,
+      "learning_rate": 1.136068639313607e-05,
+      "loss": 0.5245,
+      "step": 4410
+    },
+    {
+      "epoch": 2.3091787439613527,
+      "grad_norm": 4.763105869293213,
+      "learning_rate": 1.133868661313387e-05,
+      "loss": 0.6002,
+      "step": 4420
+    },
+    {
+      "epoch": 2.3144013578796185,
+      "grad_norm": 3.849142074584961,
+      "learning_rate": 1.1316686833131669e-05,
+      "loss": 0.6067,
+      "step": 4430
+    },
+    {
+      "epoch": 2.319623971797885,
+      "grad_norm": 6.650665760040283,
+      "learning_rate": 1.1294687053129469e-05,
+      "loss": 0.5129,
+      "step": 4440
+    },
+    {
+      "epoch": 2.324846585716151,
+      "grad_norm": 4.083095073699951,
+      "learning_rate": 1.1272687273127269e-05,
+      "loss": 0.5995,
+      "step": 4450
+    },
+    {
+      "epoch": 2.330069199634417,
+      "grad_norm": 5.624079704284668,
+      "learning_rate": 1.125068749312507e-05,
+      "loss": 0.5652,
+      "step": 4460
+    },
+    {
+      "epoch": 2.3352918135526832,
+      "grad_norm": 6.2935028076171875,
+      "learning_rate": 1.122868771312287e-05,
+      "loss": 0.5226,
+      "step": 4470
+    },
+    {
+      "epoch": 2.340514427470949,
+      "grad_norm": 7.2632646560668945,
+      "learning_rate": 1.120668793312067e-05,
+      "loss": 0.5898,
+      "step": 4480
+    },
+    {
+      "epoch": 2.3457370413892153,
+      "grad_norm": 5.291088104248047,
+      "learning_rate": 1.118468815311847e-05,
+      "loss": 0.5239,
+      "step": 4490
+    },
+    {
+      "epoch": 2.3509596553074816,
+      "grad_norm": 4.0247907638549805,
+      "learning_rate": 1.1162688373116271e-05,
+      "loss": 0.5263,
+      "step": 4500
+    },
+    {
+      "epoch": 2.3509596553074816,
+      "eval_loss": 0.943185031414032,
+      "eval_runtime": 46.3699,
+      "eval_samples_per_second": 36.705,
+      "eval_steps_per_second": 4.593,
+      "step": 4500
+    },
+    {
+      "epoch": 2.3561822692257475,
+      "grad_norm": 6.087654113769531,
+      "learning_rate": 1.1140688593114071e-05,
+      "loss": 0.5452,
+      "step": 4510
+    },
+    {
+      "epoch": 2.3614048831440138,
+      "grad_norm": 5.286149024963379,
+      "learning_rate": 1.1118688813111871e-05,
+      "loss": 0.4648,
+      "step": 4520
+    },
+    {
+      "epoch": 2.3666274970622796,
+      "grad_norm": 5.043085098266602,
+      "learning_rate": 1.1096689033109669e-05,
+      "loss": 0.6365,
+      "step": 4530
+    },
+    {
+      "epoch": 2.371850110980546,
+      "grad_norm": 4.788007736206055,
+      "learning_rate": 1.1074689253107469e-05,
+      "loss": 0.4857,
+      "step": 4540
+    },
+    {
+      "epoch": 2.3770727248988117,
+      "grad_norm": 5.524052619934082,
+      "learning_rate": 1.1052689473105269e-05,
+      "loss": 0.6689,
+      "step": 4550
+    },
+    {
+      "epoch": 2.382295338817078,
+      "grad_norm": 4.40060567855835,
+      "learning_rate": 1.1030689693103069e-05,
+      "loss": 0.4604,
+      "step": 4560
+    },
+    {
+      "epoch": 2.387517952735344,
+      "grad_norm": 4.678870677947998,
+      "learning_rate": 1.100868991310087e-05,
+      "loss": 0.5077,
+      "step": 4570
+    },
+    {
+      "epoch": 2.39274056665361,
+      "grad_norm": 5.824218273162842,
+      "learning_rate": 1.098669013309867e-05,
+      "loss": 0.5265,
+      "step": 4580
+    },
+    {
+      "epoch": 2.3979631805718764,
+      "grad_norm": 6.174131393432617,
+      "learning_rate": 1.096469035309647e-05,
+      "loss": 0.4931,
+      "step": 4590
+    },
+    {
+      "epoch": 2.4031857944901422,
+      "grad_norm": 5.839408874511719,
+      "learning_rate": 1.094269057309427e-05,
+      "loss": 0.53,
+      "step": 4600
+    },
+    {
+      "epoch": 2.4084084084084085,
+      "grad_norm": 5.057196140289307,
+      "learning_rate": 1.0920690793092071e-05,
+      "loss": 0.5008,
+      "step": 4610
+    },
+    {
+      "epoch": 2.4136310223266744,
+      "grad_norm": 4.919235706329346,
+      "learning_rate": 1.0898691013089871e-05,
+      "loss": 0.5645,
+      "step": 4620
+    },
+    {
+      "epoch": 2.4188536362449407,
+      "grad_norm": 4.900165557861328,
+      "learning_rate": 1.087669123308767e-05,
+      "loss": 0.5037,
+      "step": 4630
+    },
+    {
+      "epoch": 2.4240762501632065,
+      "grad_norm": 5.373619079589844,
+      "learning_rate": 1.085469145308547e-05,
+      "loss": 0.5159,
+      "step": 4640
+    },
+    {
+      "epoch": 2.4292988640814728,
+      "grad_norm": 4.071446418762207,
+      "learning_rate": 1.083269167308327e-05,
+      "loss": 0.4935,
+      "step": 4650
+    },
+    {
+      "epoch": 2.434521477999739,
+      "grad_norm": 4.382746696472168,
+      "learning_rate": 1.0810691893081072e-05,
+      "loss": 0.5618,
+      "step": 4660
+    },
+    {
+      "epoch": 2.439744091918005,
+      "grad_norm": 5.00177001953125,
+      "learning_rate": 1.0788692113078868e-05,
+      "loss": 0.5216,
+      "step": 4670
+    },
+    {
+      "epoch": 2.444966705836271,
+      "grad_norm": 5.724264621734619,
+      "learning_rate": 1.076669233307667e-05,
+      "loss": 0.5516,
+      "step": 4680
+    },
+    {
+      "epoch": 2.450189319754537,
+      "grad_norm": 5.690515518188477,
+      "learning_rate": 1.074469255307447e-05,
+      "loss": 0.5124,
+      "step": 4690
+    },
+    {
+      "epoch": 2.4554119336728033,
+      "grad_norm": 5.372811794281006,
+      "learning_rate": 1.072269277307227e-05,
+      "loss": 0.514,
+      "step": 4700
+    },
+    {
+      "epoch": 2.460634547591069,
+      "grad_norm": 3.5407040119171143,
+      "learning_rate": 1.070069299307007e-05,
+      "loss": 0.5131,
+      "step": 4710
+    },
+    {
+      "epoch": 2.4658571615093354,
+      "grad_norm": 5.465915679931641,
+      "learning_rate": 1.067869321306787e-05,
+      "loss": 0.4962,
+      "step": 4720
+    },
+    {
+      "epoch": 2.4710797754276017,
+      "grad_norm": 3.357609272003174,
+      "learning_rate": 1.065669343306567e-05,
+      "loss": 0.495,
+      "step": 4730
+    },
+    {
+      "epoch": 2.4763023893458675,
+      "grad_norm": 6.003732204437256,
+      "learning_rate": 1.063469365306347e-05,
+      "loss": 0.47,
+      "step": 4740
+    },
+    {
+      "epoch": 2.481525003264134,
+      "grad_norm": 6.457789897918701,
+      "learning_rate": 1.061269387306127e-05,
+      "loss": 0.5406,
+      "step": 4750
+    },
+    {
+      "epoch": 2.4867476171823997,
+      "grad_norm": 4.927131652832031,
+      "learning_rate": 1.059069409305907e-05,
+      "loss": 0.5804,
+      "step": 4760
+    },
+    {
+      "epoch": 2.491970231100666,
+      "grad_norm": 4.805700302124023,
+      "learning_rate": 1.0568694313056872e-05,
+      "loss": 0.6091,
+      "step": 4770
+    },
+    {
+      "epoch": 2.497192845018932,
+      "grad_norm": 5.979767799377441,
+      "learning_rate": 1.0546694533054671e-05,
+      "loss": 0.5219,
+      "step": 4780
+    },
+    {
+      "epoch": 2.502415458937198,
+      "grad_norm": 3.6271190643310547,
+      "learning_rate": 1.0524694753052471e-05,
+      "loss": 0.5208,
+      "step": 4790
+    },
+    {
+      "epoch": 2.507638072855464,
+      "grad_norm": 4.7155442237854,
+      "learning_rate": 1.0502694973050271e-05,
+      "loss": 0.5011,
+      "step": 4800
+    },
+    {
+      "epoch": 2.51286068677373,
+      "grad_norm": 5.767155647277832,
+      "learning_rate": 1.048069519304807e-05,
+      "loss": 0.4557,
+      "step": 4810
+    },
+    {
+      "epoch": 2.5180833006919965,
+      "grad_norm": 3.412713050842285,
+      "learning_rate": 1.0458695413045869e-05,
+      "loss": 0.5196,
+      "step": 4820
+    },
+    {
+      "epoch": 2.5233059146102623,
+      "grad_norm": 6.184292793273926,
+      "learning_rate": 1.0436695633043669e-05,
+      "loss": 0.6146,
+      "step": 4830
+    },
+    {
+      "epoch": 2.5285285285285286,
+      "grad_norm": 6.836652755737305,
+      "learning_rate": 1.041469585304147e-05,
+      "loss": 0.5338,
+      "step": 4840
+    },
+    {
+      "epoch": 2.5337511424467944,
+      "grad_norm": 6.3216118812561035,
+      "learning_rate": 1.039269607303927e-05,
+      "loss": 0.5146,
+      "step": 4850
+    },
+    {
+      "epoch": 2.5389737563650607,
+      "grad_norm": 5.718419075012207,
+      "learning_rate": 1.037069629303707e-05,
+      "loss": 0.5192,
+      "step": 4860
+    },
+    {
+      "epoch": 2.544196370283327,
+      "grad_norm": 4.821221351623535,
+      "learning_rate": 1.034869651303487e-05,
+      "loss": 0.5044,
+      "step": 4870
+    },
+    {
+      "epoch": 2.549418984201593,
+      "grad_norm": 3.7728681564331055,
+      "learning_rate": 1.0326696733032671e-05,
+      "loss": 0.4557,
+      "step": 4880
+    },
+    {
+      "epoch": 2.554641598119859,
+      "grad_norm": 4.283219337463379,
+      "learning_rate": 1.0304696953030471e-05,
+      "loss": 0.6728,
+      "step": 4890
+    },
+    {
+      "epoch": 2.559864212038125,
+      "grad_norm": 3.8312339782714844,
+      "learning_rate": 1.0282697173028271e-05,
+      "loss": 0.4257,
+      "step": 4900
+    },
+    {
+      "epoch": 2.5650868259563913,
+      "grad_norm": 4.710253715515137,
+      "learning_rate": 1.0260697393026071e-05,
+      "loss": 0.4818,
+      "step": 4910
+    },
+    {
+      "epoch": 2.5703094398746575,
+      "grad_norm": 5.925652980804443,
+      "learning_rate": 1.023869761302387e-05,
+      "loss": 0.5167,
+      "step": 4920
+    },
+    {
+      "epoch": 2.5755320537929234,
+      "grad_norm": 4.76073694229126,
+      "learning_rate": 1.0216697833021672e-05,
+      "loss": 0.4363,
+      "step": 4930
+    },
+    {
+      "epoch": 2.580754667711189,
+      "grad_norm": 3.673877716064453,
+      "learning_rate": 1.0194698053019472e-05,
+      "loss": 0.5326,
+      "step": 4940
+    },
+    {
+      "epoch": 2.5859772816294555,
+      "grad_norm": 4.946410655975342,
+      "learning_rate": 1.017269827301727e-05,
+      "loss": 0.6238,
+      "step": 4950
+    },
+    {
+      "epoch": 2.591199895547722,
+      "grad_norm": 4.7166290283203125,
+      "learning_rate": 1.015069849301507e-05,
+      "loss": 0.479,
+      "step": 4960
+    },
+    {
+      "epoch": 2.5964225094659876,
+      "grad_norm": 4.268213748931885,
+      "learning_rate": 1.012869871301287e-05,
+      "loss": 0.5868,
+      "step": 4970
+    },
+    {
+      "epoch": 2.601645123384254,
+      "grad_norm": 5.25217866897583,
+      "learning_rate": 1.010669893301067e-05,
+      "loss": 0.5622,
+      "step": 4980
+    },
+    {
+      "epoch": 2.6068677373025197,
+      "grad_norm": 5.709757328033447,
+      "learning_rate": 1.0084699153008471e-05,
+      "loss": 0.4127,
+      "step": 4990
+    },
+    {
+      "epoch": 2.612090351220786,
+      "grad_norm": 5.0968732833862305,
+      "learning_rate": 1.0062699373006271e-05,
+      "loss": 0.5278,
+      "step": 5000
+    },
+    {
+      "epoch": 2.612090351220786,
+      "eval_loss": 0.9182845950126648,
+      "eval_runtime": 46.4198,
+      "eval_samples_per_second": 36.665,
+      "eval_steps_per_second": 4.589,
+      "step": 5000
+    },
+    {
+      "epoch": 2.6173129651390523,
+      "grad_norm": 4.902764320373535,
+      "learning_rate": 1.004069959300407e-05,
+      "loss": 0.5596,
+      "step": 5010
+    },
+    {
+      "epoch": 2.622535579057318,
+      "grad_norm": 5.16074800491333,
+      "learning_rate": 1.001869981300187e-05,
+      "loss": 0.496,
+      "step": 5020
+    },
+    {
+      "epoch": 2.6277581929755844,
+      "grad_norm": 5.447412490844727,
+      "learning_rate": 9.99670003299967e-06,
+      "loss": 0.5532,
+      "step": 5030
+    },
+    {
+      "epoch": 2.6329808068938503,
+      "grad_norm": 2.698256492614746,
+      "learning_rate": 9.974700252997472e-06,
+      "loss": 0.5541,
+      "step": 5040
+    },
+    {
+      "epoch": 2.6382034208121166,
+      "grad_norm": 5.540690898895264,
+      "learning_rate": 9.95270047299527e-06,
+      "loss": 0.5367,
+      "step": 5050
+    },
+    {
+      "epoch": 2.6434260347303824,
+      "grad_norm": 3.423121929168701,
+      "learning_rate": 9.93070069299307e-06,
+      "loss": 0.5361,
+      "step": 5060
+    },
+    {
+      "epoch": 2.6486486486486487,
+      "grad_norm": 5.744271278381348,
+      "learning_rate": 9.908700912990871e-06,
+      "loss": 0.5378,
+      "step": 5070
+    },
+    {
+      "epoch": 2.6538712625669145,
+      "grad_norm": 5.6931843757629395,
+      "learning_rate": 9.886701132988671e-06,
+      "loss": 0.5504,
+      "step": 5080
+    },
+    {
+      "epoch": 2.659093876485181,
+      "grad_norm": 5.743236064910889,
+      "learning_rate": 9.864701352986471e-06,
+      "loss": 0.5685,
+      "step": 5090
+    },
+    {
+      "epoch": 2.664316490403447,
+      "grad_norm": 5.774685382843018,
+      "learning_rate": 9.842701572984271e-06,
+      "loss": 0.6709,
+      "step": 5100
+    },
+    {
+      "epoch": 2.669539104321713,
+      "grad_norm": 3.9477055072784424,
+      "learning_rate": 9.82070179298207e-06,
+      "loss": 0.4714,
+      "step": 5110
+    },
+    {
+      "epoch": 2.674761718239979,
+      "grad_norm": 4.734736442565918,
+      "learning_rate": 9.79870201297987e-06,
+      "loss": 0.4373,
+      "step": 5120
+    },
+    {
+      "epoch": 2.679984332158245,
+      "grad_norm": 4.351528167724609,
+      "learning_rate": 9.77670223297767e-06,
+      "loss": 0.5032,
+      "step": 5130
+    },
+    {
+      "epoch": 2.6852069460765113,
+      "grad_norm": 5.193398475646973,
+      "learning_rate": 9.75470245297547e-06,
+      "loss": 0.4897,
+      "step": 5140
+    },
+    {
+      "epoch": 2.6904295599947776,
+      "grad_norm": 3.739581823348999,
+      "learning_rate": 9.732702672973272e-06,
+      "loss": 0.4485,
+      "step": 5150
+    },
+    {
+      "epoch": 2.6956521739130435,
+      "grad_norm": 4.9811787605285645,
+      "learning_rate": 9.710702892971072e-06,
+      "loss": 0.5092,
+      "step": 5160
+    },
+    {
+      "epoch": 2.7008747878313093,
+      "grad_norm": 5.68161678314209,
+      "learning_rate": 9.688703112968871e-06,
+      "loss": 0.5234,
+      "step": 5170
+    },
+    {
+      "epoch": 2.7060974017495756,
+      "grad_norm": 4.802379131317139,
+      "learning_rate": 9.666703332966671e-06,
+      "loss": 0.4929,
+      "step": 5180
+    },
+    {
+      "epoch": 2.711320015667842,
+      "grad_norm": 4.281787872314453,
+      "learning_rate": 9.644703552964471e-06,
+      "loss": 0.5053,
+      "step": 5190
+    },
+    {
+      "epoch": 2.7165426295861077,
+      "grad_norm": 5.739396572113037,
+      "learning_rate": 9.622703772962271e-06,
+      "loss": 0.572,
+      "step": 5200
+    },
+    {
+      "epoch": 2.721765243504374,
+      "grad_norm": 4.055601596832275,
+      "learning_rate": 9.60070399296007e-06,
+      "loss": 0.5056,
+      "step": 5210
+    },
+    {
+      "epoch": 2.72698785742264,
+      "grad_norm": 6.050775527954102,
+      "learning_rate": 9.57870421295787e-06,
+      "loss": 0.4743,
+      "step": 5220
+    },
+    {
+      "epoch": 2.732210471340906,
+      "grad_norm": 2.8432188034057617,
+      "learning_rate": 9.556704432955672e-06,
+      "loss": 0.4505,
+      "step": 5230
+    },
+    {
+      "epoch": 2.7374330852591724,
+      "grad_norm": 5.975009918212891,
+      "learning_rate": 9.534704652953472e-06,
+      "loss": 0.5604,
+      "step": 5240
+    },
+    {
+      "epoch": 2.7426556991774382,
+      "grad_norm": 4.10011100769043,
+      "learning_rate": 9.512704872951272e-06,
+      "loss": 0.4997,
+      "step": 5250
+    },
+    {
+      "epoch": 2.7478783130957045,
+      "grad_norm": 3.3179147243499756,
+      "learning_rate": 9.490705092949071e-06,
+      "loss": 0.4576,
+      "step": 5260
+    },
+    {
+      "epoch": 2.7531009270139704,
+      "grad_norm": 5.42781925201416,
+      "learning_rate": 9.468705312946871e-06,
+      "loss": 0.5374,
+      "step": 5270
+    },
+    {
+      "epoch": 2.7583235409322366,
+      "grad_norm": 5.195150852203369,
+      "learning_rate": 9.446705532944671e-06,
+      "loss": 0.4939,
+      "step": 5280
+    },
+    {
+      "epoch": 2.763546154850503,
+      "grad_norm": 3.490081548690796,
+      "learning_rate": 9.424705752942471e-06,
+      "loss": 0.4875,
+      "step": 5290
+    },
+    {
+      "epoch": 2.7687687687687688,
+      "grad_norm": 2.548433303833008,
+      "learning_rate": 9.40270597294027e-06,
+      "loss": 0.5088,
+      "step": 5300
+    },
+    {
+      "epoch": 2.7739913826870346,
+      "grad_norm": 4.056173801422119,
+      "learning_rate": 9.380706192938072e-06,
+      "loss": 0.5062,
+      "step": 5310
+    },
+    {
+      "epoch": 2.779213996605301,
+      "grad_norm": 3.9009056091308594,
+      "learning_rate": 9.358706412935872e-06,
+      "loss": 0.4859,
+      "step": 5320
+    },
+    {
+      "epoch": 2.784436610523567,
+      "grad_norm": 7.616847038269043,
+      "learning_rate": 9.336706632933672e-06,
+      "loss": 0.4976,
+      "step": 5330
+    },
+    {
+      "epoch": 2.789659224441833,
+      "grad_norm": 3.8724076747894287,
+      "learning_rate": 9.314706852931472e-06,
+      "loss": 0.4965,
+      "step": 5340
+    },
+    {
+      "epoch": 2.7948818383600993,
+      "grad_norm": 6.395397663116455,
+      "learning_rate": 9.292707072929272e-06,
+      "loss": 0.5324,
+      "step": 5350
+    },
+    {
+      "epoch": 2.800104452278365,
+      "grad_norm": 5.191283702850342,
+      "learning_rate": 9.270707292927071e-06,
+      "loss": 0.4321,
+      "step": 5360
+    },
+    {
+      "epoch": 2.8053270661966314,
+      "grad_norm": 5.025651454925537,
+      "learning_rate": 9.248707512924871e-06,
+      "loss": 0.5549,
+      "step": 5370
+    },
+    {
+      "epoch": 2.8105496801148977,
+      "grad_norm": 6.335343360900879,
+      "learning_rate": 9.226707732922671e-06,
+      "loss": 0.5857,
+      "step": 5380
+    },
+    {
+      "epoch": 2.8157722940331635,
+      "grad_norm": 4.999082565307617,
+      "learning_rate": 9.204707952920473e-06,
+      "loss": 0.4673,
+      "step": 5390
+    },
+    {
+      "epoch": 2.82099490795143,
+      "grad_norm": 5.395633697509766,
+      "learning_rate": 9.182708172918272e-06,
+      "loss": 0.5711,
+      "step": 5400
+    },
+    {
+      "epoch": 2.8262175218696957,
+      "grad_norm": 3.730288028717041,
+      "learning_rate": 9.16070839291607e-06,
+      "loss": 0.5602,
+      "step": 5410
+    },
+    {
+      "epoch": 2.831440135787962,
+      "grad_norm": 6.485812664031982,
+      "learning_rate": 9.138708612913872e-06,
+      "loss": 0.5606,
+      "step": 5420
+    },
+    {
+      "epoch": 2.8366627497062282,
+      "grad_norm": 6.046615123748779,
+      "learning_rate": 9.116708832911672e-06,
+      "loss": 0.5591,
+      "step": 5430
+    },
+    {
+      "epoch": 2.841885363624494,
+      "grad_norm": 6.039712429046631,
+      "learning_rate": 9.094709052909472e-06,
+      "loss": 0.7427,
+      "step": 5440
+    },
+    {
+      "epoch": 2.84710797754276,
+      "grad_norm": 5.540979385375977,
+      "learning_rate": 9.072709272907272e-06,
+      "loss": 0.5996,
+      "step": 5450
+    },
+    {
+      "epoch": 2.852330591461026,
+      "grad_norm": 4.887878894805908,
+      "learning_rate": 9.050709492905071e-06,
+      "loss": 0.5769,
+      "step": 5460
+    },
+    {
+      "epoch": 2.8575532053792925,
+      "grad_norm": 3.2837166786193848,
+      "learning_rate": 9.028709712902873e-06,
+      "loss": 0.4668,
+      "step": 5470
+    },
+    {
+      "epoch": 2.8627758192975583,
+      "grad_norm": 6.3544440269470215,
+      "learning_rate": 9.006709932900671e-06,
+      "loss": 0.4784,
+      "step": 5480
+    },
+    {
+      "epoch": 2.8679984332158246,
+      "grad_norm": 3.9592199325561523,
+      "learning_rate": 8.98471015289847e-06,
+      "loss": 0.521,
+      "step": 5490
+    },
+    {
+      "epoch": 2.8732210471340904,
+      "grad_norm": 3.39884614944458,
+      "learning_rate": 8.962710372896272e-06,
+      "loss": 0.5307,
+      "step": 5500
+    },
+    {
+      "epoch": 2.8732210471340904,
+      "eval_loss": 0.9099580645561218,
+      "eval_runtime": 46.3657,
+      "eval_samples_per_second": 36.708,
+      "eval_steps_per_second": 4.594,
+      "step": 5500
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 9570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 5,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5.954793002080666e+16,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}