diff --git "a/checkpoint-6014/trainer_state.json" "b/checkpoint-6014/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-6014/trainer_state.json"
@@ -0,0 +1,8443 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 6014,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.08855848014354706,
+      "learning_rate": 0.0004999991472481453,
+      "loss": 2.6557,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.07915773242712021,
+      "learning_rate": 0.0004999965889983988,
+      "loss": 2.4772,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.12245654314756393,
+      "learning_rate": 0.0004999923252682129,
+      "loss": 2.4706,
+      "step": 15
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.10732001066207886,
+      "learning_rate": 0.0004999863560866746,
+      "loss": 2.205,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.13363118469715118,
+      "learning_rate": 0.0004999786814945059,
+      "loss": 2.3767,
+      "step": 25
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.09244401752948761,
+      "learning_rate": 0.0004999693015440632,
+      "loss": 2.3077,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.0993221327662468,
+      "learning_rate": 0.0004999582162993363,
+      "loss": 2.3961,
+      "step": 35
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.09801439195871353,
+      "learning_rate": 0.0004999454258359492,
+      "loss": 2.303,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.1130354180932045,
+      "learning_rate": 0.0004999309302411583,
+      "loss": 2.3285,
+      "step": 45
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.15056972205638885,
+      "learning_rate": 0.0004999147296138531,
+      "loss": 2.4721,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.13582268357276917,
+      "learning_rate": 0.0004998968240645544,
+      "loss": 2.4008,
+      "step": 55
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.13640083372592926,
+      "learning_rate": 0.0004998772137154141,
+      "loss": 2.3995,
+      "step": 60
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.12711608409881592,
+      "learning_rate": 0.0004998558987002143,
+      "loss": 2.356,
+      "step": 65
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.19234400987625122,
+      "learning_rate": 0.0004998328791643663,
+      "loss": 2.2295,
+      "step": 70
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.1442757099866867,
+      "learning_rate": 0.0004998081552649098,
+      "loss": 2.4544,
+      "step": 75
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.18835151195526123,
+      "learning_rate": 0.0004997817271705116,
+      "loss": 2.3927,
+      "step": 80
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.13745005428791046,
+      "learning_rate": 0.0004997535950614643,
+      "loss": 2.3474,
+      "step": 85
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.1309432089328766,
+      "learning_rate": 0.0004997237591296861,
+      "loss": 2.3056,
+      "step": 90
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.1625588983297348,
+      "learning_rate": 0.0004996922195787177,
+      "loss": 2.4774,
+      "step": 95
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.1700213998556137,
+      "learning_rate": 0.0004996589766237225,
+      "loss": 2.3015,
+      "step": 100
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.15182705223560333,
+      "learning_rate": 0.0004996240304914846,
+      "loss": 2.4763,
+      "step": 105
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.15341205894947052,
+      "learning_rate": 0.0004995873814204069,
+      "loss": 2.3662,
+      "step": 110
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.14351683855056763,
+      "learning_rate": 0.0004995490296605098,
+      "loss": 1.967,
+      "step": 115
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.19977103173732758,
+      "learning_rate": 0.0004995089754734297,
+      "loss": 2.4405,
+      "step": 120
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.16674907505512238,
+      "learning_rate": 0.0004994672191324169,
+      "loss": 2.3259,
+      "step": 125
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.1730051189661026,
+      "learning_rate": 0.0004994237609223337,
+      "loss": 2.457,
+      "step": 130
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.22182777523994446,
+      "learning_rate": 0.0004993786011396528,
+      "loss": 2.4724,
+      "step": 135
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.1570582240819931,
+      "learning_rate": 0.0004993317400924547,
+      "loss": 2.3075,
+      "step": 140
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.14569781720638275,
+      "learning_rate": 0.0004992831781004262,
+      "loss": 2.1955,
+      "step": 145
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.19999592006206512,
+      "learning_rate": 0.000499232915494858,
+      "loss": 2.4308,
+      "step": 150
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.20882728695869446,
+      "learning_rate": 0.0004991809526186424,
+      "loss": 2.2685,
+      "step": 155
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.15529118478298187,
+      "learning_rate": 0.0004991272898262708,
+      "loss": 2.4197,
+      "step": 160
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.20917291939258575,
+      "learning_rate": 0.0004990719274838315,
+      "loss": 2.111,
+      "step": 165
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.16249670088291168,
+      "learning_rate": 0.0004990148659690073,
+      "loss": 2.1891,
+      "step": 170
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.18368925154209137,
+      "learning_rate": 0.0004989561056710729,
+      "loss": 2.445,
+      "step": 175
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.21063920855522156,
+      "learning_rate": 0.0004988956469908916,
+      "loss": 2.2818,
+      "step": 180
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.16046880185604095,
+      "learning_rate": 0.0004988334903409137,
+      "loss": 2.3213,
+      "step": 185
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.1568533182144165,
+      "learning_rate": 0.0004987696361451725,
+      "loss": 2.2785,
+      "step": 190
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.2682538330554962,
+      "learning_rate": 0.0004987040848392824,
+      "loss": 2.3157,
+      "step": 195
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.2094998061656952,
+      "learning_rate": 0.0004986368368704355,
+      "loss": 2.4451,
+      "step": 200
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.17797984182834625,
+      "learning_rate": 0.0004985678926973982,
+      "loss": 2.4541,
+      "step": 205
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.1685200035572052,
+      "learning_rate": 0.000498497252790509,
+      "loss": 2.5848,
+      "step": 210
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.18856477737426758,
+      "learning_rate": 0.0004984249176316741,
+      "loss": 2.2977,
+      "step": 215
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.1670752465724945,
+      "learning_rate": 0.0004983508877143651,
+      "loss": 2.2476,
+      "step": 220
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.20539650321006775,
+      "learning_rate": 0.0004982751635436152,
+      "loss": 2.4784,
+      "step": 225
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.19674716889858246,
+      "learning_rate": 0.000498197745636016,
+      "loss": 2.3826,
+      "step": 230
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.1808764785528183,
+      "learning_rate": 0.0004981186345197133,
+      "loss": 2.1915,
+      "step": 235
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.1649690568447113,
+      "learning_rate": 0.0004980378307344044,
+      "loss": 2.3101,
+      "step": 240
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.19910891354084015,
+      "learning_rate": 0.0004979553348313341,
+      "loss": 2.3597,
+      "step": 245
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.17429497838020325,
+      "learning_rate": 0.0004978711473732906,
+      "loss": 2.4553,
+      "step": 250
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.17228329181671143,
+      "learning_rate": 0.000497785268934602,
+      "loss": 2.4496,
+      "step": 255
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.18831633031368256,
+      "learning_rate": 0.0004976977001011321,
+      "loss": 2.2841,
+      "step": 260
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.1668764352798462,
+      "learning_rate": 0.000497608441470277,
+      "loss": 2.5694,
+      "step": 265
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.17521929740905762,
+      "learning_rate": 0.0004975174936509602,
+      "loss": 2.3644,
+      "step": 270
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.1602141261100769,
+      "learning_rate": 0.0004974248572636292,
+      "loss": 2.2046,
+      "step": 275
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.2694706618785858,
+      "learning_rate": 0.0004973305329402509,
+      "loss": 2.3489,
+      "step": 280
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.19102543592453003,
+      "learning_rate": 0.0004972345213243071,
+      "loss": 2.4592,
+      "step": 285
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.15223553776741028,
+      "learning_rate": 0.0004971368230707905,
+      "loss": 2.4645,
+      "step": 290
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.21530242264270782,
+      "learning_rate": 0.0004970374388462,
+      "loss": 2.2222,
+      "step": 295
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.15933960676193237,
+      "learning_rate": 0.0004969363693285363,
+      "loss": 2.2653,
+      "step": 300
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.2127295583486557,
+      "learning_rate": 0.0004968336152072971,
+      "loss": 2.3073,
+      "step": 305
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.18392804265022278,
+      "learning_rate": 0.0004967291771834727,
+      "loss": 2.1329,
+      "step": 310
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.15626537799835205,
+      "learning_rate": 0.0004966230559695406,
+      "loss": 2.2159,
+      "step": 315
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.19132393598556519,
+      "learning_rate": 0.0004965152522894615,
+      "loss": 2.476,
+      "step": 320
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.20226113498210907,
+      "learning_rate": 0.0004964057668786736,
+      "loss": 2.3081,
+      "step": 325
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.1876431554555893,
+      "learning_rate": 0.000496294600484088,
+      "loss": 2.3597,
+      "step": 330
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.16817514598369598,
+      "learning_rate": 0.0004961817538640836,
+      "loss": 2.2757,
+      "step": 335
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.18912410736083984,
+      "learning_rate": 0.0004960672277885016,
+      "loss": 2.3976,
+      "step": 340
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.2327001988887787,
+      "learning_rate": 0.0004959510230386406,
+      "loss": 2.3661,
+      "step": 345
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.2154412865638733,
+      "learning_rate": 0.0004958331404072511,
+      "loss": 2.3553,
+      "step": 350
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.18963538110256195,
+      "learning_rate": 0.0004957135806985303,
+      "loss": 2.166,
+      "step": 355
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.19671855866909027,
+      "learning_rate": 0.0004955923447281162,
+      "loss": 2.4749,
+      "step": 360
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.20475703477859497,
+      "learning_rate": 0.0004954694333230824,
+      "loss": 2.2721,
+      "step": 365
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.17598754167556763,
+      "learning_rate": 0.0004953448473219323,
+      "loss": 2.3505,
+      "step": 370
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.21133027970790863,
+      "learning_rate": 0.0004952185875745934,
+      "loss": 2.3477,
+      "step": 375
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.2046145498752594,
+      "learning_rate": 0.0004950906549424119,
+      "loss": 2.3582,
+      "step": 380
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.19380740821361542,
+      "learning_rate": 0.0004949610502981457,
+      "loss": 2.4032,
+      "step": 385
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.17790696024894714,
+      "learning_rate": 0.0004948297745259598,
+      "loss": 2.5021,
+      "step": 390
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.15176743268966675,
+      "learning_rate": 0.0004946968285214194,
+      "loss": 2.2358,
+      "step": 395
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.20856505632400513,
+      "learning_rate": 0.0004945622131914843,
+      "loss": 2.2877,
+      "step": 400
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.2029004991054535,
+      "learning_rate": 0.0004944259294545019,
+      "loss": 2.4357,
+      "step": 405
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.15777291357517242,
+      "learning_rate": 0.0004942879782402022,
+      "loss": 2.34,
+      "step": 410
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.18712708353996277,
+      "learning_rate": 0.0004941483604896904,
+      "loss": 2.3396,
+      "step": 415
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.20823413133621216,
+      "learning_rate": 0.0004940070771554408,
+      "loss": 2.3517,
+      "step": 420
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.3073880970478058,
+      "learning_rate": 0.0004938641292012904,
+      "loss": 2.1421,
+      "step": 425
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.2047368884086609,
+      "learning_rate": 0.0004937195176024323,
+      "loss": 2.3541,
+      "step": 430
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.1642436683177948,
+      "learning_rate": 0.0004935732433454088,
+      "loss": 2.3114,
+      "step": 435
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.24149803817272186,
+      "learning_rate": 0.0004934253074281054,
+      "loss": 2.0895,
+      "step": 440
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.18440380692481995,
+      "learning_rate": 0.0004932757108597428,
+      "loss": 2.3473,
+      "step": 445
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.18851065635681152,
+      "learning_rate": 0.0004931244546608711,
+      "loss": 2.7386,
+      "step": 450
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.2015257179737091,
+      "learning_rate": 0.0004929715398633624,
+      "loss": 2.5105,
+      "step": 455
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.2113213837146759,
+      "learning_rate": 0.0004928169675104037,
+      "loss": 2.2766,
+      "step": 460
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.16831834614276886,
+      "learning_rate": 0.0004926607386564898,
+      "loss": 2.4742,
+      "step": 465
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.22336164116859436,
+      "learning_rate": 0.0004925028543674164,
+      "loss": 2.5234,
+      "step": 470
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.19430772960186005,
+      "learning_rate": 0.0004923433157202723,
+      "loss": 2.1902,
+      "step": 475
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.1814448982477188,
+      "learning_rate": 0.0004921821238034326,
+      "loss": 2.3091,
+      "step": 480
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.2538805305957794,
+      "learning_rate": 0.000492019279716551,
+      "loss": 2.3677,
+      "step": 485
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.22240865230560303,
+      "learning_rate": 0.0004918547845705523,
+      "loss": 1.6151,
+      "step": 490
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.21643097698688507,
+      "learning_rate": 0.0004916886394876247,
+      "loss": 2.3328,
+      "step": 495
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.1595735102891922,
+      "learning_rate": 0.0004915208456012125,
+      "loss": 2.2963,
+      "step": 500
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.1994553655385971,
+      "learning_rate": 0.0004913514040560081,
+      "loss": 2.4716,
+      "step": 505
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.19502036273479462,
+      "learning_rate": 0.000491180316007944,
+      "loss": 2.2055,
+      "step": 510
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.33217477798461914,
+      "learning_rate": 0.0004910075826241858,
+      "loss": 2.3901,
+      "step": 515
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.17329910397529602,
+      "learning_rate": 0.0004908332050831229,
+      "loss": 2.4412,
+      "step": 520
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.19649653136730194,
+      "learning_rate": 0.0004906571845743616,
+      "loss": 2.5036,
+      "step": 525
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.20772477984428406,
+      "learning_rate": 0.0004904795222987164,
+      "loss": 2.2438,
+      "step": 530
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.17279602587223053,
+      "learning_rate": 0.000490300219468202,
+      "loss": 2.2795,
+      "step": 535
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.1734314262866974,
+      "learning_rate": 0.0004901192773060249,
+      "loss": 2.286,
+      "step": 540
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.20874248445034027,
+      "learning_rate": 0.0004899366970465753,
+      "loss": 2.335,
+      "step": 545
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.17237204313278198,
+      "learning_rate": 0.0004897524799354184,
+      "loss": 2.5809,
+      "step": 550
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.19388693571090698,
+      "learning_rate": 0.0004895666272292862,
+      "loss": 2.4632,
+      "step": 555
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.17896869778633118,
+      "learning_rate": 0.0004893791401960683,
+      "loss": 1.9323,
+      "step": 560
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.2121048867702484,
+      "learning_rate": 0.0004891900201148043,
+      "loss": 2.1472,
+      "step": 565
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.20368768274784088,
+      "learning_rate": 0.000488999268275674,
+      "loss": 2.3966,
+      "step": 570
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.1953553706407547,
+      "learning_rate": 0.0004888068859799895,
+      "loss": 2.1608,
+      "step": 575
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.168823704123497,
+      "learning_rate": 0.0004886128745401855,
+      "loss": 2.572,
+      "step": 580
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.213327556848526,
+      "learning_rate": 0.000488417235279811,
+      "loss": 2.4827,
+      "step": 585
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.19623854756355286,
+      "learning_rate": 0.00048821996953352,
+      "loss": 2.3122,
+      "step": 590
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.15413519740104675,
+      "learning_rate": 0.00048802107864706217,
+      "loss": 2.3699,
+      "step": 595
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.1929747462272644,
+      "learning_rate": 0.00048782056397727425,
+      "loss": 2.462,
+      "step": 600
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.19839173555374146,
+      "learning_rate": 0.0004876184268920702,
+      "loss": 2.2265,
+      "step": 605
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.1684265285730362,
+      "learning_rate": 0.0004874146687704323,
+      "loss": 2.0534,
+      "step": 610
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.19233009219169617,
+      "learning_rate": 0.0004872092910024014,
+      "loss": 2.2648,
+      "step": 615
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.1821518987417221,
+      "learning_rate": 0.0004870022949890676,
+      "loss": 2.1656,
+      "step": 620
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.17291715741157532,
+      "learning_rate": 0.000486793682142561,
+      "loss": 2.2383,
+      "step": 625
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.18019358813762665,
+      "learning_rate": 0.00048658345388604134,
+      "loss": 2.4353,
+      "step": 630
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.2849899232387543,
+      "learning_rate": 0.00048637161165368905,
+      "loss": 2.331,
+      "step": 635
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.3656805753707886,
+      "learning_rate": 0.00048615815689069487,
+      "loss": 2.3611,
+      "step": 640
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.18143434822559357,
+      "learning_rate": 0.0004859430910532504,
+      "loss": 2.4043,
+      "step": 645
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.1801663637161255,
+      "learning_rate": 0.00048572641560853805,
+      "loss": 2.2429,
+      "step": 650
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.16966284811496735,
+      "learning_rate": 0.0004855081320347207,
+      "loss": 2.3253,
+      "step": 655
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.25287890434265137,
+      "learning_rate": 0.0004852882418209323,
+      "loss": 2.2341,
+      "step": 660
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.1766708791255951,
+      "learning_rate": 0.0004850667464672672,
+      "loss": 2.6625,
+      "step": 665
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.23353531956672668,
+      "learning_rate": 0.00048484364748476995,
+      "loss": 1.9082,
+      "step": 670
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.19202041625976562,
+      "learning_rate": 0.00048461894639542505,
+      "loss": 2.5067,
+      "step": 675
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.75723135471344,
+      "learning_rate": 0.0004843926447321466,
+      "loss": 2.1948,
+      "step": 680
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.18087565898895264,
+      "learning_rate": 0.0004841647440387681,
+      "loss": 2.273,
+      "step": 685
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.20649802684783936,
+      "learning_rate": 0.0004839352458700314,
+      "loss": 2.3464,
+      "step": 690
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.19096232950687408,
+      "learning_rate": 0.0004837041517915762,
+      "loss": 2.1913,
+      "step": 695
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.1938013732433319,
+      "learning_rate": 0.0004834714633799301,
+      "loss": 2.1721,
+      "step": 700
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.3532910943031311,
+      "learning_rate": 0.00048323718222249655,
+      "loss": 2.0894,
+      "step": 705
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.1876286268234253,
+      "learning_rate": 0.00048300130991754534,
+      "loss": 2.4283,
+      "step": 710
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.1776292324066162,
+      "learning_rate": 0.0004827638480742007,
+      "loss": 2.2125,
+      "step": 715
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.22727273404598236,
+      "learning_rate": 0.0004825247983124309,
+      "loss": 2.2988,
+      "step": 720
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.16823545098304749,
+      "learning_rate": 0.00048228416226303697,
+      "loss": 2.1763,
+      "step": 725
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.21675339341163635,
+      "learning_rate": 0.00048204194156764146,
+      "loss": 2.4963,
+      "step": 730
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.1901399940252304,
+      "learning_rate": 0.0004817981378786778,
+      "loss": 2.2221,
+      "step": 735
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.20318692922592163,
+      "learning_rate": 0.00048155275285937814,
+      "loss": 2.601,
+      "step": 740
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.17421816289424896,
+      "learning_rate": 0.0004813057881837629,
+      "loss": 2.3034,
+      "step": 745
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.2204848974943161,
+      "learning_rate": 0.00048105724553662864,
+      "loss": 2.2784,
+      "step": 750
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.21678051352500916,
+      "learning_rate": 0.00048080712661353707,
+      "loss": 2.1565,
+      "step": 755
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.18651427328586578,
+      "learning_rate": 0.0004805554331208032,
+      "loss": 2.2581,
+      "step": 760
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.2580089569091797,
+      "learning_rate": 0.00048030216677548367,
+      "loss": 2.0844,
+      "step": 765
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.20590724050998688,
+      "learning_rate": 0.00048004732930536525,
+      "loss": 2.1468,
+      "step": 770
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.21979926526546478,
+      "learning_rate": 0.00047979092244895306,
+      "loss": 2.1362,
+      "step": 775
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.20203465223312378,
+      "learning_rate": 0.00047953294795545845,
+      "loss": 2.2438,
+      "step": 780
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.19766996800899506,
+      "learning_rate": 0.0004792734075847871,
+      "loss": 2.2673,
+      "step": 785
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.18661414086818695,
+      "learning_rate": 0.0004790123031075275,
+      "loss": 2.1558,
+      "step": 790
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.20814205706119537,
+      "learning_rate": 0.00047874963630493804,
+      "loss": 2.2609,
+      "step": 795
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.20795443654060364,
+      "learning_rate": 0.00047848540896893577,
+      "loss": 2.0754,
+      "step": 800
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.20875173807144165,
+      "learning_rate": 0.00047821962290208326,
+      "loss": 2.1128,
+      "step": 805
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.18199613690376282,
+      "learning_rate": 0.00047795227991757715,
+      "loss": 2.3072,
+      "step": 810
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.18696178495883942,
+      "learning_rate": 0.00047768338183923527,
+      "loss": 2.1862,
+      "step": 815
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.1970338076353073,
+      "learning_rate": 0.0004774129305014842,
+      "loss": 2.1828,
+      "step": 820
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.19044700264930725,
+      "learning_rate": 0.00047714092774934705,
+      "loss": 2.2603,
+      "step": 825
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.20430251955986023,
+      "learning_rate": 0.00047686737543843063,
+      "loss": 2.1969,
+      "step": 830
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.31919530034065247,
+      "learning_rate": 0.0004765922754349128,
+      "loss": 2.023,
+      "step": 835
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.18964442610740662,
+      "learning_rate": 0.0004763156296155299,
+      "loss": 2.3088,
+      "step": 840
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.1777138113975525,
+      "learning_rate": 0.0004760374398675638,
+      "loss": 2.2829,
+      "step": 845
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.19124820828437805,
+      "learning_rate": 0.0004757577080888291,
+      "loss": 2.332,
+      "step": 850
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.20289361476898193,
+      "learning_rate": 0.0004754764361876602,
+      "loss": 2.3203,
+      "step": 855
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.2134077399969101,
+      "learning_rate": 0.0004751936260828982,
+      "loss": 2.3119,
+      "step": 860
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.2681266963481903,
+      "learning_rate": 0.00047490927970387766,
+      "loss": 2.1415,
+      "step": 865
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.2137921005487442,
+      "learning_rate": 0.000474623398990414,
+      "loss": 2.1536,
+      "step": 870
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.2895020842552185,
+      "learning_rate": 0.0004743359858927895,
+      "loss": 2.0187,
+      "step": 875
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.193098783493042,
+      "learning_rate": 0.0004740470423717407,
+      "loss": 2.0446,
+      "step": 880
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.1691964864730835,
+      "learning_rate": 0.00047375657039844455,
+      "loss": 2.2179,
+      "step": 885
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.18233279883861542,
+      "learning_rate": 0.0004734645719545051,
+      "loss": 2.2694,
+      "step": 890
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.2980597913265228,
+      "learning_rate": 0.0004731710490319403,
+      "loss": 2.0281,
+      "step": 895
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.21693329513072968,
+      "learning_rate": 0.0004728760036331676,
+      "loss": 2.2698,
+      "step": 900
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.20274919271469116,
+      "learning_rate": 0.0004725794377709912,
+      "loss": 2.4936,
+      "step": 905
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.2515089213848114,
+      "learning_rate": 0.0004722813534685879,
+      "loss": 2.1535,
+      "step": 910
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.20583480596542358,
+      "learning_rate": 0.00047198175275949315,
+      "loss": 2.1036,
+      "step": 915
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.18232561647891998,
+      "learning_rate": 0.0004716806376875874,
+      "loss": 2.3023,
+      "step": 920
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.21218007802963257,
+      "learning_rate": 0.00047137801030708217,
+      "loss": 2.3002,
+      "step": 925
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.2054254710674286,
+      "learning_rate": 0.0004710738726825059,
+      "loss": 2.3362,
+      "step": 930
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.21346160769462585,
+      "learning_rate": 0.00047076822688869006,
+      "loss": 2.0738,
+      "step": 935
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.18373315036296844,
+      "learning_rate": 0.00047046107501075475,
+      "loss": 2.2725,
+      "step": 940
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.1660277098417282,
+      "learning_rate": 0.0004701524191440947,
+      "loss": 2.2974,
+      "step": 945
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.20062261819839478,
+      "learning_rate": 0.0004698422613943647,
+      "loss": 2.4897,
+      "step": 950
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.20721417665481567,
+      "learning_rate": 0.00046953060387746557,
+      "loss": 2.2635,
+      "step": 955
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.22309726476669312,
+      "learning_rate": 0.00046921744871952954,
+      "loss": 2.2008,
+      "step": 960
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.19087284803390503,
+      "learning_rate": 0.00046890279805690565,
+      "loss": 2.1818,
+      "step": 965
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.1862783133983612,
+      "learning_rate": 0.00046858665403614556,
+      "loss": 2.2324,
+      "step": 970
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.21848712861537933,
+      "learning_rate": 0.0004682690188139883,
+      "loss": 2.4289,
+      "step": 975
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.20349426567554474,
+      "learning_rate": 0.00046794989455734617,
+      "loss": 2.3544,
+      "step": 980
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.22449104487895966,
+      "learning_rate": 0.00046762928344328957,
+      "loss": 2.33,
+      "step": 985
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.20320549607276917,
+      "learning_rate": 0.0004673071876590322,
+      "loss": 2.3229,
+      "step": 990
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.33305850625038147,
+      "learning_rate": 0.00046698360940191643,
+      "loss": 2.3729,
+      "step": 995
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.20021797716617584,
+      "learning_rate": 0.00046665855087939777,
+      "loss": 2.2189,
+      "step": 1000
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.18848876655101776,
+      "learning_rate": 0.0004663320143090304,
+      "loss": 2.1184,
+      "step": 1005
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.2134588062763214,
+      "learning_rate": 0.00046600400191845157,
+      "loss": 2.3574,
+      "step": 1010
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.20047910511493683,
+      "learning_rate": 0.0004656745159453667,
+      "loss": 2.4937,
+      "step": 1015
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.2156955897808075,
+      "learning_rate": 0.00046534355863753397,
+      "loss": 2.443,
+      "step": 1020
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.2098633348941803,
+      "learning_rate": 0.00046501113225274915,
+      "loss": 2.5053,
+      "step": 1025
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.20685449242591858,
+      "learning_rate": 0.00046467723905882984,
+      "loss": 2.3909,
+      "step": 1030
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.21232719719409943,
+      "learning_rate": 0.00046434188133360045,
+      "loss": 2.2625,
+      "step": 1035
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.1747647374868393,
+      "learning_rate": 0.00046400506136487626,
+      "loss": 2.0894,
+      "step": 1040
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.30385705828666687,
+      "learning_rate": 0.00046366678145044814,
+      "loss": 2.3737,
+      "step": 1045
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.21601806581020355,
+      "learning_rate": 0.0004633270438980668,
+      "loss": 2.2979,
+      "step": 1050
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.19758005440235138,
+      "learning_rate": 0.0004629858510254267,
+      "loss": 2.346,
+      "step": 1055
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.23565934598445892,
+      "learning_rate": 0.0004626432051601507,
+      "loss": 2.3861,
+      "step": 1060
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.20269787311553955,
+      "learning_rate": 0.0004622991086397741,
+      "loss": 2.2194,
+      "step": 1065
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.17142094671726227,
+      "learning_rate": 0.00046195356381172834,
+      "loss": 2.2701,
+      "step": 1070
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.1847238689661026,
+      "learning_rate": 0.0004616065730333254,
+      "loss": 2.0997,
+      "step": 1075
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.2138655036687851,
+      "learning_rate": 0.00046125813867174155,
+      "loss": 2.5526,
+      "step": 1080
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.23903928697109222,
+      "learning_rate": 0.00046090826310400115,
+      "loss": 2.4783,
+      "step": 1085
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.26335886120796204,
+      "learning_rate": 0.0004605569487169605,
+      "loss": 2.394,
+      "step": 1090
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.2013441026210785,
+      "learning_rate": 0.0004602041979072916,
+      "loss": 2.2272,
+      "step": 1095
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.17305096983909607,
+      "learning_rate": 0.0004598500130814657,
+      "loss": 2.4204,
+      "step": 1100
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.1825268715620041,
+      "learning_rate": 0.00045949439665573686,
+      "loss": 2.2598,
+      "step": 1105
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.21777556836605072,
+      "learning_rate": 0.00045913735105612577,
+      "loss": 2.1326,
+      "step": 1110
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.23001907765865326,
+      "learning_rate": 0.00045877887871840264,
+      "loss": 2.4948,
+      "step": 1115
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.20269040763378143,
+      "learning_rate": 0.0004584189820880712,
+      "loss": 2.127,
+      "step": 1120
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.1827128827571869,
+      "learning_rate": 0.0004580576636203515,
+      "loss": 2.2534,
+      "step": 1125
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.17216241359710693,
+      "learning_rate": 0.0004576949257801636,
+      "loss": 2.4425,
+      "step": 1130
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.19745908677577972,
+      "learning_rate": 0.0004573307710421103,
+      "loss": 2.3759,
+      "step": 1135
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.21618859469890594,
+      "learning_rate": 0.00045696520189046077,
+      "loss": 2.2339,
+      "step": 1140
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.18233755230903625,
+      "learning_rate": 0.00045659822081913306,
+      "loss": 2.2468,
+      "step": 1145
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.17987078428268433,
+      "learning_rate": 0.00045622983033167755,
+      "loss": 2.2023,
+      "step": 1150
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.20248466730117798,
+      "learning_rate": 0.0004558600329412596,
+      "loss": 2.1912,
+      "step": 1155
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.20506608486175537,
+      "learning_rate": 0.0004554888311706425,
+      "loss": 2.3273,
+      "step": 1160
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.22549524903297424,
+      "learning_rate": 0.0004551162275521702,
+      "loss": 2.1894,
+      "step": 1165
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.18408866226673126,
+      "learning_rate": 0.00045474222462775016,
+      "loss": 2.3541,
+      "step": 1170
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.2189367264509201,
+      "learning_rate": 0.00045436682494883585,
+      "loss": 2.119,
+      "step": 1175
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.1773710697889328,
+      "learning_rate": 0.00045399003107640944,
+      "loss": 2.3584,
+      "step": 1180
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.1946442872285843,
+      "learning_rate": 0.00045361184558096433,
+      "loss": 2.237,
+      "step": 1185
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.2449074685573578,
+      "learning_rate": 0.00045323227104248763,
+      "loss": 2.2493,
+      "step": 1190
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.1789473295211792,
+      "learning_rate": 0.00045285131005044245,
+      "loss": 2.1959,
+      "step": 1195
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.20725607872009277,
+      "learning_rate": 0.00045246896520375035,
+      "loss": 2.3345,
+      "step": 1200
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.21160945296287537,
+      "learning_rate": 0.0004520852391107734,
+      "loss": 2.2686,
+      "step": 1205
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.20865868031978607,
+      "learning_rate": 0.0004517001343892969,
+      "loss": 2.2881,
+      "step": 1210
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.23047909140586853,
+      "learning_rate": 0.00045131365366651085,
+      "loss": 2.1912,
+      "step": 1215
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.2568853795528412,
+      "learning_rate": 0.0004509257995789925,
+      "loss": 2.4391,
+      "step": 1220
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.17928802967071533,
+      "learning_rate": 0.0004505365747726882,
+      "loss": 2.3539,
+      "step": 1225
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.20399945974349976,
+      "learning_rate": 0.00045014598190289536,
+      "loss": 2.1629,
+      "step": 1230
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.20776867866516113,
+      "learning_rate": 0.0004497540236342443,
+      "loss": 2.3111,
+      "step": 1235
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.2226899415254593,
+      "learning_rate": 0.00044936070264068017,
+      "loss": 2.37,
+      "step": 1240
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.1808883249759674,
+      "learning_rate": 0.00044896602160544467,
+      "loss": 2.0685,
+      "step": 1245
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.21628347039222717,
+      "learning_rate": 0.00044856998322105764,
+      "loss": 2.2747,
+      "step": 1250
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.21809974312782288,
+      "learning_rate": 0.0004481725901892988,
+      "loss": 2.4779,
+      "step": 1255
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.2099200189113617,
+      "learning_rate": 0.00044777384522118926,
+      "loss": 2.1749,
+      "step": 1260
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.22972367703914642,
+      "learning_rate": 0.0004473737510369732,
+      "loss": 2.4255,
+      "step": 1265
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.3171684443950653,
+      "learning_rate": 0.0004469723103660991,
+      "loss": 2.2226,
+      "step": 1270
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.23672665655612946,
+      "learning_rate": 0.000446569525947201,
+      "loss": 2.3501,
+      "step": 1275
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.20475780963897705,
+      "learning_rate": 0.0004461654005280804,
+      "loss": 2.3695,
+      "step": 1280
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.1991017460823059,
+      "learning_rate": 0.00044575993686568675,
+      "loss": 2.2487,
+      "step": 1285
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.24624770879745483,
+      "learning_rate": 0.00044535313772609925,
+      "loss": 2.3519,
+      "step": 1290
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.17691905796527863,
+      "learning_rate": 0.0004449450058845077,
+      "loss": 2.3415,
+      "step": 1295
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.19319698214530945,
+      "learning_rate": 0.0004445355441251935,
+      "loss": 2.2697,
+      "step": 1300
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.20446337759494781,
+      "learning_rate": 0.0004441247552415107,
+      "loss": 2.2208,
+      "step": 1305
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.22464382648468018,
+      "learning_rate": 0.00044371264203586753,
+      "loss": 2.2629,
+      "step": 1310
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.22489066421985626,
+      "learning_rate": 0.0004432992073197062,
+      "loss": 2.3359,
+      "step": 1315
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.19456712901592255,
+      "learning_rate": 0.0004428844539134844,
+      "loss": 2.196,
+      "step": 1320
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.22781945765018463,
+      "learning_rate": 0.00044246838464665606,
+      "loss": 2.3108,
+      "step": 1325
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.20936964452266693,
+      "learning_rate": 0.00044205100235765194,
+      "loss": 2.1782,
+      "step": 1330
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.1924324333667755,
+      "learning_rate": 0.0004416323098938602,
+      "loss": 2.297,
+      "step": 1335
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.18653331696987152,
+      "learning_rate": 0.00044121231011160704,
+      "loss": 2.2042,
+      "step": 1340
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.24487438797950745,
+      "learning_rate": 0.0004407910058761372,
+      "loss": 2.2889,
+      "step": 1345
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.19584549963474274,
+      "learning_rate": 0.00044036840006159443,
+      "loss": 2.3466,
+      "step": 1350
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.21697990596294403,
+      "learning_rate": 0.00043994449555100193,
+      "loss": 2.2742,
+      "step": 1355
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.2454603612422943,
+      "learning_rate": 0.00043951929523624233,
+      "loss": 2.4419,
+      "step": 1360
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.24156726896762848,
+      "learning_rate": 0.0004390928020180388,
+      "loss": 2.1757,
+      "step": 1365
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.1978321224451065,
+      "learning_rate": 0.00043866501880593416,
+      "loss": 2.0642,
+      "step": 1370
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.18749172985553741,
+      "learning_rate": 0.000438235948518272,
+      "loss": 2.3689,
+      "step": 1375
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.25640445947647095,
+      "learning_rate": 0.000437805594082176,
+      "loss": 2.2029,
+      "step": 1380
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.21351024508476257,
+      "learning_rate": 0.0004373739584335308,
+      "loss": 2.4231,
+      "step": 1385
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.1982639878988266,
+      "learning_rate": 0.0004369410445169608,
+      "loss": 2.2831,
+      "step": 1390
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.31549686193466187,
+      "learning_rate": 0.00043650685528581155,
+      "loss": 2.3625,
+      "step": 1395
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.26613035798072815,
+      "learning_rate": 0.00043607139370212814,
+      "loss": 2.1986,
+      "step": 1400
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.20148108899593353,
+      "learning_rate": 0.0004356346627366361,
+      "loss": 2.2001,
+      "step": 1405
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.20938289165496826,
+      "learning_rate": 0.0004351966653687206,
+      "loss": 2.196,
+      "step": 1410
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.19232477247714996,
+      "learning_rate": 0.00043475740458640607,
+      "loss": 2.2149,
+      "step": 1415
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.23008297383785248,
+      "learning_rate": 0.0004343168833863361,
+      "loss": 2.2319,
+      "step": 1420
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.1995578110218048,
+      "learning_rate": 0.0004338751047737529,
+      "loss": 2.3673,
+      "step": 1425
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.17232060432434082,
+      "learning_rate": 0.0004334320717624767,
+      "loss": 2.1701,
+      "step": 1430
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.25711289048194885,
+      "learning_rate": 0.0004329877873748853,
+      "loss": 2.1933,
+      "step": 1435
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.2659030854701996,
+      "learning_rate": 0.00043254225464189335,
+      "loss": 1.9466,
+      "step": 1440
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.22041670978069305,
+      "learning_rate": 0.00043209547660293167,
+      "loss": 2.0898,
+      "step": 1445
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.18253591656684875,
+      "learning_rate": 0.00043164745630592686,
+      "loss": 2.2224,
+      "step": 1450
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.21765796840190887,
+      "learning_rate": 0.00043119819680728,
+      "loss": 2.2998,
+      "step": 1455
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.2172078639268875,
+      "learning_rate": 0.00043074770117184594,
+      "loss": 2.2644,
+      "step": 1460
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.19673798978328705,
+      "learning_rate": 0.0004302959724729127,
+      "loss": 2.1841,
+      "step": 1465
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.20035311579704285,
+      "learning_rate": 0.0004298430137921801,
+      "loss": 2.4498,
+      "step": 1470
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.18350964784622192,
+      "learning_rate": 0.000429388828219739,
+      "loss": 2.3168,
+      "step": 1475
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.18951267004013062,
+      "learning_rate": 0.00042893341885405007,
+      "loss": 2.2359,
+      "step": 1480
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.19564704596996307,
+      "learning_rate": 0.0004284767888019229,
+      "loss": 2.4583,
+      "step": 1485
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.1913597732782364,
+      "learning_rate": 0.00042801894117849425,
+      "loss": 2.2388,
+      "step": 1490
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.22768037021160126,
+      "learning_rate": 0.00042755987910720765,
+      "loss": 2.4552,
+      "step": 1495
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.20260490477085114,
+      "learning_rate": 0.00042709960571979123,
+      "loss": 2.4709,
+      "step": 1500
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.2456539422273636,
+      "learning_rate": 0.00042663812415623685,
+      "loss": 2.0556,
+      "step": 1505
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.20906373858451843,
+      "learning_rate": 0.00042617543756477867,
+      "loss": 2.2327,
+      "step": 1510
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.30633488297462463,
+      "learning_rate": 0.0004257115491018714,
+      "loss": 2.3643,
+      "step": 1515
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.2021452635526657,
+      "learning_rate": 0.00042524646193216906,
+      "loss": 2.2815,
+      "step": 1520
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.20578056573867798,
+      "learning_rate": 0.0004247801792285032,
+      "loss": 2.1693,
+      "step": 1525
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.24151162803173065,
+      "learning_rate": 0.0004243127041718614,
+      "loss": 2.4737,
+      "step": 1530
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.2197379320859909,
+      "learning_rate": 0.00042384403995136534,
+      "loss": 2.4527,
+      "step": 1535
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.24394835531711578,
+      "learning_rate": 0.00042337418976424936,
+      "loss": 2.3551,
+      "step": 1540
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.2169724851846695,
+      "learning_rate": 0.0004229031568158383,
+      "loss": 2.2993,
+      "step": 1545
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.193971648812294,
+      "learning_rate": 0.0004224309443195261,
+      "loss": 2.3481,
+      "step": 1550
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.22565196454524994,
+      "learning_rate": 0.00042195755549675324,
+      "loss": 2.4907,
+      "step": 1555
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.24645136296749115,
+      "learning_rate": 0.0004214829935769854,
+      "loss": 2.2397,
+      "step": 1560
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.18104226887226105,
+      "learning_rate": 0.000421007261797691,
+      "loss": 2.2246,
+      "step": 1565
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.17285558581352234,
+      "learning_rate": 0.0004205303634043194,
+      "loss": 2.1925,
+      "step": 1570
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.17167840898036957,
+      "learning_rate": 0.0004200523016502783,
+      "loss": 2.2859,
+      "step": 1575
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.2422027587890625,
+      "learning_rate": 0.00041957307979691226,
+      "loss": 2.1167,
+      "step": 1580
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.22238920629024506,
+      "learning_rate": 0.00041909270111347986,
+      "loss": 2.1963,
+      "step": 1585
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.31004032492637634,
+      "learning_rate": 0.0004186111688771315,
+      "loss": 2.0878,
+      "step": 1590
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.20740267634391785,
+      "learning_rate": 0.00041812848637288736,
+      "loss": 2.2541,
+      "step": 1595
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.220609650015831,
+      "learning_rate": 0.00041764465689361454,
+      "loss": 2.3849,
+      "step": 1600
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.24316507577896118,
+      "learning_rate": 0.000417159683740005,
+      "loss": 2.2295,
+      "step": 1605
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2829028069972992,
+      "learning_rate": 0.0004166735702205527,
+      "loss": 2.2568,
+      "step": 1610
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.20122820138931274,
+      "learning_rate": 0.0004161863196515314,
+      "loss": 2.2641,
+      "step": 1615
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.23964501917362213,
+      "learning_rate": 0.00041569793535697165,
+      "loss": 2.3745,
+      "step": 1620
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.20462176203727722,
+      "learning_rate": 0.0004152084206686384,
+      "loss": 2.3948,
+      "step": 1625
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.21822762489318848,
+      "learning_rate": 0.00041471777892600805,
+      "loss": 2.4378,
+      "step": 1630
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.18578322231769562,
+      "learning_rate": 0.0004142260134762459,
+      "loss": 2.2125,
+      "step": 1635
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.1855502426624298,
+      "learning_rate": 0.0004137331276741831,
+      "loss": 2.1584,
+      "step": 1640
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2354944944381714,
+      "learning_rate": 0.00041323912488229394,
+      "loss": 2.4169,
+      "step": 1645
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2282811552286148,
+      "learning_rate": 0.0004127440084706727,
+      "loss": 2.1559,
+      "step": 1650
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.2994524836540222,
+      "learning_rate": 0.0004122477818170111,
+      "loss": 1.9802,
+      "step": 1655
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.17210961878299713,
+      "learning_rate": 0.00041175044830657454,
+      "loss": 2.1983,
+      "step": 1660
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.21259650588035583,
+      "learning_rate": 0.0004112520113321797,
+      "loss": 2.184,
+      "step": 1665
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.1957891881465912,
+      "learning_rate": 0.00041075247429417095,
+      "loss": 2.3489,
+      "step": 1670
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.202979177236557,
+      "learning_rate": 0.00041025184060039743,
+      "loss": 2.3709,
+      "step": 1675
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.1887211799621582,
+      "learning_rate": 0.0004097501136661896,
+      "loss": 2.4586,
+      "step": 1680
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.1808994561433792,
+      "learning_rate": 0.00040924729691433606,
+      "loss": 2.3032,
+      "step": 1685
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.20630981028079987,
+      "learning_rate": 0.0004087433937750601,
+      "loss": 2.3186,
+      "step": 1690
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.2361355423927307,
+      "learning_rate": 0.00040823840768599656,
+      "loss": 2.1686,
+      "step": 1695
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.23437006771564484,
+      "learning_rate": 0.0004077323420921679,
+      "loss": 2.2247,
+      "step": 1700
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.20080877840518951,
+      "learning_rate": 0.00040722520044596114,
+      "loss": 2.453,
+      "step": 1705
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.22303621470928192,
+      "learning_rate": 0.0004067169862071041,
+      "loss": 2.095,
+      "step": 1710
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.1971607655286789,
+      "learning_rate": 0.0004062077028426419,
+      "loss": 2.2782,
+      "step": 1715
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.22378380596637726,
+      "learning_rate": 0.0004056973538269132,
+      "loss": 2.2542,
+      "step": 1720
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.1935906559228897,
+      "learning_rate": 0.00040518594264152654,
+      "loss": 2.1513,
+      "step": 1725
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.288876473903656,
+      "learning_rate": 0.00040467347277533655,
+      "loss": 2.2316,
+      "step": 1730
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.2500283122062683,
+      "learning_rate": 0.0004041599477244204,
+      "loss": 2.3276,
+      "step": 1735
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.2323485165834427,
+      "learning_rate": 0.0004036453709920535,
+      "loss": 2.3284,
+      "step": 1740
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.22176632285118103,
+      "learning_rate": 0.000403129746088686,
+      "loss": 2.245,
+      "step": 1745
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.21016566455364227,
+      "learning_rate": 0.0004026130765319187,
+      "loss": 2.2102,
+      "step": 1750
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.22619667649269104,
+      "learning_rate": 0.00040209536584647875,
+      "loss": 2.4151,
+      "step": 1755
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.2051662653684616,
+      "learning_rate": 0.00040157661756419646,
+      "loss": 2.2177,
+      "step": 1760
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.20449289679527283,
+      "learning_rate": 0.00040105683522398004,
+      "loss": 2.219,
+      "step": 1765
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.20390522480010986,
+      "learning_rate": 0.00040053602237179246,
+      "loss": 2.1071,
+      "step": 1770
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.20242749154567719,
+      "learning_rate": 0.00040001418256062674,
+      "loss": 2.0979,
+      "step": 1775
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.22173428535461426,
+      "learning_rate": 0.0003994913193504817,
+      "loss": 2.3197,
+      "step": 1780
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.22555288672447205,
+      "learning_rate": 0.000398967436308338,
+      "loss": 2.4671,
+      "step": 1785
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.2446102499961853,
+      "learning_rate": 0.0003984425370081335,
+      "loss": 2.3669,
+      "step": 1790
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.20918509364128113,
+      "learning_rate": 0.000397916625030739,
+      "loss": 2.4581,
+      "step": 1795
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.2591955363750458,
+      "learning_rate": 0.00039738970396393386,
+      "loss": 2.2427,
+      "step": 1800
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.21759961545467377,
+      "learning_rate": 0.0003968617774023814,
+      "loss": 2.3986,
+      "step": 1805
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.29073455929756165,
+      "learning_rate": 0.0003963328489476045,
+      "loss": 2.4886,
+      "step": 1810
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.1800486147403717,
+      "learning_rate": 0.00039580292220796077,
+      "loss": 2.4326,
+      "step": 1815
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.23151837289333344,
+      "learning_rate": 0.0003952720007986185,
+      "loss": 2.3614,
+      "step": 1820
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.24493078887462616,
+      "learning_rate": 0.0003947400883415313,
+      "loss": 2.4108,
+      "step": 1825
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.18288132548332214,
+      "learning_rate": 0.00039420718846541384,
+      "loss": 2.3391,
+      "step": 1830
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.22715741395950317,
+      "learning_rate": 0.000393673304805717,
+      "loss": 2.1612,
+      "step": 1835
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.19885744154453278,
+      "learning_rate": 0.0003931384410046031,
+      "loss": 2.2497,
+      "step": 1840
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.20505961775779724,
+      "learning_rate": 0.0003926026007109207,
+      "loss": 2.368,
+      "step": 1845
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.20146214962005615,
+      "learning_rate": 0.00039206578758018047,
+      "loss": 2.457,
+      "step": 1850
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2406214475631714,
+      "learning_rate": 0.0003915280052745295,
+      "loss": 2.2507,
+      "step": 1855
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.219930961728096,
+      "learning_rate": 0.0003909892574627266,
+      "loss": 2.4327,
+      "step": 1860
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.24521400034427643,
+      "learning_rate": 0.0003904495478201174,
+      "loss": 1.8198,
+      "step": 1865
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.20486341416835785,
+      "learning_rate": 0.00038990888002860913,
+      "loss": 2.1002,
+      "step": 1870
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.21982859075069427,
+      "learning_rate": 0.0003893672577766453,
+      "loss": 2.2058,
+      "step": 1875
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.21171775460243225,
+      "learning_rate": 0.0003888246847591811,
+      "loss": 2.3034,
+      "step": 1880
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.23152461647987366,
+      "learning_rate": 0.0003882811646776577,
+      "loss": 2.2811,
+      "step": 1885
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2419670820236206,
+      "learning_rate": 0.0003877367012399772,
+      "loss": 2.3064,
+      "step": 1890
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2318524569272995,
+      "learning_rate": 0.00038719129816047705,
+      "loss": 2.3313,
+      "step": 1895
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.26805251836776733,
+      "learning_rate": 0.0003866449591599053,
+      "loss": 2.3949,
+      "step": 1900
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.20099863409996033,
+      "learning_rate": 0.00038609768796539474,
+      "loss": 2.4194,
+      "step": 1905
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2393646538257599,
+      "learning_rate": 0.00038554948831043755,
+      "loss": 1.9689,
+      "step": 1910
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2404375821352005,
+      "learning_rate": 0.0003850003639348598,
+      "loss": 2.1439,
+      "step": 1915
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.22678737342357635,
+      "learning_rate": 0.0003844503185847964,
+      "loss": 2.2073,
+      "step": 1920
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2190311998128891,
+      "learning_rate": 0.00038389935601266454,
+      "loss": 2.3895,
+      "step": 1925
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.36514168977737427,
+      "learning_rate": 0.0003833474799771394,
+      "loss": 2.1925,
+      "step": 1930
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.19080831110477448,
+      "learning_rate": 0.00038279469424312723,
+      "loss": 2.2825,
+      "step": 1935
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.19559559226036072,
+      "learning_rate": 0.0003822410025817406,
+      "loss": 1.9923,
+      "step": 1940
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.204594224691391,
+      "learning_rate": 0.0003816864087702723,
+      "loss": 2.2341,
+      "step": 1945
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.2163582295179367,
+      "learning_rate": 0.0003811309165921695,
+      "loss": 2.3263,
+      "step": 1950
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.22616557776927948,
+      "learning_rate": 0.00038057452983700793,
+      "loss": 2.4578,
+      "step": 1955
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.21181915700435638,
+      "learning_rate": 0.00038001725230046645,
+      "loss": 2.2881,
+      "step": 1960
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.21244682371616364,
+      "learning_rate": 0.00037945908778430073,
+      "loss": 2.3482,
+      "step": 1965
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.22712396085262299,
+      "learning_rate": 0.00037890004009631727,
+      "loss": 2.467,
+      "step": 1970
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.20208491384983063,
+      "learning_rate": 0.00037834011305034774,
+      "loss": 2.4241,
+      "step": 1975
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.23634742200374603,
+      "learning_rate": 0.00037777931046622273,
+      "loss": 2.1896,
+      "step": 1980
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.20562510192394257,
+      "learning_rate": 0.0003772176361697458,
+      "loss": 2.3983,
+      "step": 1985
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.19427534937858582,
+      "learning_rate": 0.0003766550939926674,
+      "loss": 2.0804,
+      "step": 1990
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.22697383165359497,
+      "learning_rate": 0.00037609168777265843,
+      "loss": 2.3416,
+      "step": 1995
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.2205856442451477,
+      "learning_rate": 0.0003755274213532847,
+      "loss": 2.2662,
+      "step": 2000
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.24093294143676758,
+      "learning_rate": 0.0003749622985839799,
+      "loss": 2.1266,
+      "step": 2005
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.2476285696029663,
+      "learning_rate": 0.0003743963233200201,
+      "loss": 2.2735,
+      "step": 2010
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.23264004290103912,
+      "learning_rate": 0.00037382949942249696,
+      "loss": 2.2642,
+      "step": 2015
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.20969323813915253,
+      "learning_rate": 0.00037326183075829126,
+      "loss": 2.3931,
+      "step": 2020
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.23997239768505096,
+      "learning_rate": 0.0003726933212000474,
+      "loss": 2.4217,
+      "step": 2025
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.25848066806793213,
+      "learning_rate": 0.0003721239746261458,
+      "loss": 2.3721,
+      "step": 2030
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.1921965628862381,
+      "learning_rate": 0.0003715537949206773,
+      "loss": 2.2568,
+      "step": 2035
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.19738563895225525,
+      "learning_rate": 0.0003709827859734163,
+      "loss": 2.245,
+      "step": 2040
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.23472926020622253,
+      "learning_rate": 0.0003704109516797943,
+      "loss": 2.2329,
+      "step": 2045
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.19815848767757416,
+      "learning_rate": 0.00036983829594087336,
+      "loss": 2.5266,
+      "step": 2050
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.22582224011421204,
+      "learning_rate": 0.00036926482266331947,
+      "loss": 2.2198,
+      "step": 2055
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.21971648931503296,
+      "learning_rate": 0.0003686905357593758,
+      "loss": 2.2141,
+      "step": 2060
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.20423489809036255,
+      "learning_rate": 0.00036811543914683605,
+      "loss": 2.3125,
+      "step": 2065
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.19768038392066956,
+      "learning_rate": 0.0003675395367490179,
+      "loss": 2.364,
+      "step": 2070
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.2276119887828827,
+      "learning_rate": 0.00036696283249473604,
+      "loss": 2.2745,
+      "step": 2075
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.2061648964881897,
+      "learning_rate": 0.0003663853303182756,
+      "loss": 2.3182,
+      "step": 2080
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.22242146730422974,
+      "learning_rate": 0.00036580703415936474,
+      "loss": 2.2696,
+      "step": 2085
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.2070762813091278,
+      "learning_rate": 0.0003652279479631486,
+      "loss": 2.157,
+      "step": 2090
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.250929594039917,
+      "learning_rate": 0.00036464807568016176,
+      "loss": 2.1939,
+      "step": 2095
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.26398998498916626,
+      "learning_rate": 0.00036406742126630166,
+      "loss": 2.6459,
+      "step": 2100
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.22033217549324036,
+      "learning_rate": 0.00036348598868280114,
+      "loss": 2.5577,
+      "step": 2105
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.25079113245010376,
+      "learning_rate": 0.00036290378189620195,
+      "loss": 2.227,
+      "step": 2110
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.20717765390872955,
+      "learning_rate": 0.00036232080487832747,
+      "loss": 2.1095,
+      "step": 2115
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.19811852276325226,
+      "learning_rate": 0.0003617370616062555,
+      "loss": 2.1457,
+      "step": 2120
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.20729579031467438,
+      "learning_rate": 0.00036115255606229154,
+      "loss": 2.3231,
+      "step": 2125
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.20803453028202057,
+      "learning_rate": 0.00036056729223394083,
+      "loss": 2.3809,
+      "step": 2130
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.387298583984375,
+      "learning_rate": 0.00035998127411388187,
+      "loss": 2.2312,
+      "step": 2135
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.2171965390443802,
+      "learning_rate": 0.0003593945056999391,
+      "loss": 2.2823,
+      "step": 2140
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.19715160131454468,
+      "learning_rate": 0.00035880699099505536,
+      "loss": 2.5001,
+      "step": 2145
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.2282709926366806,
+      "learning_rate": 0.00035821873400726463,
+      "loss": 2.4297,
+      "step": 2150
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.20915524661540985,
+      "learning_rate": 0.0003576297387496648,
+      "loss": 2.445,
+      "step": 2155
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.28219184279441833,
+      "learning_rate": 0.0003570400092403903,
+      "loss": 2.2412,
+      "step": 2160
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.21221624314785004,
+      "learning_rate": 0.00035644954950258444,
+      "loss": 2.2826,
+      "step": 2165
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.22089391946792603,
+      "learning_rate": 0.00035585836356437264,
+      "loss": 2.1981,
+      "step": 2170
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.2844483256340027,
+      "learning_rate": 0.0003552664554588338,
+      "loss": 2.3541,
+      "step": 2175
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.2894970774650574,
+      "learning_rate": 0.00035467382922397395,
+      "loss": 1.8961,
+      "step": 2180
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.24491670727729797,
+      "learning_rate": 0.00035408048890269805,
+      "loss": 2.2435,
+      "step": 2185
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.2114434689283371,
+      "learning_rate": 0.00035348643854278257,
+      "loss": 2.2917,
+      "step": 2190
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.24633482098579407,
+      "learning_rate": 0.0003528916821968479,
+      "loss": 2.363,
+      "step": 2195
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.19707679748535156,
+      "learning_rate": 0.0003522962239223306,
+      "loss": 2.268,
+      "step": 2200
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.1920863687992096,
+      "learning_rate": 0.00035170006778145583,
+      "loss": 2.3193,
+      "step": 2205
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.19514207541942596,
+      "learning_rate": 0.0003511032178412098,
+      "loss": 2.2747,
+      "step": 2210
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.23632396757602692,
+      "learning_rate": 0.00035050567817331154,
+      "loss": 2.4162,
+      "step": 2215
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.20080457627773285,
+      "learning_rate": 0.0003499074528541855,
+      "loss": 2.2029,
+      "step": 2220
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.20242765545845032,
+      "learning_rate": 0.00034930854596493374,
+      "loss": 2.286,
+      "step": 2225
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.24630364775657654,
+      "learning_rate": 0.00034870896159130786,
+      "loss": 2.3279,
+      "step": 2230
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.32333359122276306,
+      "learning_rate": 0.0003481087038236815,
+      "loss": 1.89,
+      "step": 2235
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.22153715789318085,
+      "learning_rate": 0.0003475077767570219,
+      "loss": 2.2267,
+      "step": 2240
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.20011873543262482,
+      "learning_rate": 0.0003469061844908626,
+      "loss": 2.5615,
+      "step": 2245
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.20656999945640564,
+      "learning_rate": 0.00034630393112927477,
+      "loss": 2.1442,
+      "step": 2250
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.2093532532453537,
+      "learning_rate": 0.00034570102078083985,
+      "loss": 2.2969,
+      "step": 2255
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.2096518725156784,
+      "learning_rate": 0.0003450974575586213,
+      "loss": 2.2671,
+      "step": 2260
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.28272372484207153,
+      "learning_rate": 0.0003444932455801362,
+      "loss": 2.2979,
+      "step": 2265
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.19493138790130615,
+      "learning_rate": 0.0003438883889673278,
+      "loss": 2.3548,
+      "step": 2270
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.21109020709991455,
+      "learning_rate": 0.0003432828918465367,
+      "loss": 2.3726,
+      "step": 2275
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.22046278417110443,
+      "learning_rate": 0.00034267675834847346,
+      "loss": 2.176,
+      "step": 2280
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.2979690432548523,
+      "learning_rate": 0.0003420699926081897,
+      "loss": 2.0627,
+      "step": 2285
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.2563689649105072,
+      "learning_rate": 0.00034146259876505013,
+      "loss": 2.3314,
+      "step": 2290
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.19333158433437347,
+      "learning_rate": 0.00034085458096270484,
+      "loss": 2.3283,
+      "step": 2295
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.2114039808511734,
+      "learning_rate": 0.00034024594334906014,
+      "loss": 2.2269,
+      "step": 2300
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.22176076471805573,
+      "learning_rate": 0.00033963669007625085,
+      "loss": 2.225,
+      "step": 2305
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.2213573306798935,
+      "learning_rate": 0.0003390268253006119,
+      "loss": 2.2757,
+      "step": 2310
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.2310647964477539,
+      "learning_rate": 0.00033841635318264987,
+      "loss": 2.2503,
+      "step": 2315
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.22906078398227692,
+      "learning_rate": 0.00033780527788701445,
+      "loss": 2.4459,
+      "step": 2320
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.21090683341026306,
+      "learning_rate": 0.0003371936035824705,
+      "loss": 2.3689,
+      "step": 2325
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.28067609667778015,
+      "learning_rate": 0.00033658133444186935,
+      "loss": 2.2164,
+      "step": 2330
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.24392075836658478,
+      "learning_rate": 0.0003359684746421199,
+      "loss": 2.1782,
+      "step": 2335
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.22776876389980316,
+      "learning_rate": 0.0003353550283641608,
+      "loss": 2.4948,
+      "step": 2340
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.22832733392715454,
+      "learning_rate": 0.0003347409997929318,
+      "loss": 2.3384,
+      "step": 2345
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.23752398788928986,
+      "learning_rate": 0.0003341263931173449,
+      "loss": 2.0114,
+      "step": 2350
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.20741015672683716,
+      "learning_rate": 0.00033351121253025597,
+      "loss": 2.3307,
+      "step": 2355
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.23571203649044037,
+      "learning_rate": 0.000332895462228436,
+      "loss": 2.433,
+      "step": 2360
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.21684762835502625,
+      "learning_rate": 0.00033227914641254273,
+      "loss": 2.6377,
+      "step": 2365
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.3023325502872467,
+      "learning_rate": 0.00033166226928709185,
+      "loss": 2.2254,
+      "step": 2370
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.24143949151039124,
+      "learning_rate": 0.0003310448350604283,
+      "loss": 2.3094,
+      "step": 2375
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.20807060599327087,
+      "learning_rate": 0.0003304268479446974,
+      "loss": 2.4342,
+      "step": 2380
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.21055589616298676,
+      "learning_rate": 0.0003298083121558165,
+      "loss": 2.4467,
+      "step": 2385
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.22990992665290833,
+      "learning_rate": 0.000329189231913446,
+      "loss": 2.512,
+      "step": 2390
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.20713160932064056,
+      "learning_rate": 0.0003285696114409603,
+      "loss": 2.2196,
+      "step": 2395
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.21360722184181213,
+      "learning_rate": 0.0003279494549654197,
+      "loss": 2.1737,
+      "step": 2400
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.19622068107128143,
+      "learning_rate": 0.0003273287667175407,
+      "loss": 2.3779,
+      "step": 2405
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.2596871554851532,
+      "learning_rate": 0.0003267075509316678,
+      "loss": 2.2926,
+      "step": 2410
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.21403126418590546,
+      "learning_rate": 0.00032608581184574427,
+      "loss": 2.2394,
+      "step": 2415
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.1995462030172348,
+      "learning_rate": 0.0003254635537012834,
+      "loss": 2.3674,
+      "step": 2420
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.19879235327243805,
+      "learning_rate": 0.00032484078074333957,
+      "loss": 2.2684,
+      "step": 2425
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.2155233919620514,
+      "learning_rate": 0.00032421749722047894,
+      "loss": 2.1418,
+      "step": 2430
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.2114148586988449,
+      "learning_rate": 0.00032359370738475115,
+      "loss": 2.3007,
+      "step": 2435
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.24906907975673676,
+      "learning_rate": 0.00032296941549165957,
+      "loss": 2.07,
+      "step": 2440
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.21433790028095245,
+      "learning_rate": 0.00032234462580013303,
+      "loss": 2.3135,
+      "step": 2445
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.21331354975700378,
+      "learning_rate": 0.0003217193425724957,
+      "loss": 2.2919,
+      "step": 2450
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.21881960332393646,
+      "learning_rate": 0.00032109357007443926,
+      "loss": 2.2984,
+      "step": 2455
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.24045515060424805,
+      "learning_rate": 0.0003204673125749929,
+      "loss": 2.315,
+      "step": 2460
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.19170185923576355,
+      "learning_rate": 0.0003198405743464946,
+      "loss": 2.1179,
+      "step": 2465
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.21934069693088531,
+      "learning_rate": 0.0003192133596645619,
+      "loss": 2.347,
+      "step": 2470
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.20186646282672882,
+      "learning_rate": 0.0003185856728080626,
+      "loss": 2.3665,
+      "step": 2475
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.20092110335826874,
+      "learning_rate": 0.00031795751805908576,
+      "loss": 2.2575,
+      "step": 2480
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.19863560795783997,
+      "learning_rate": 0.0003173288997029124,
+      "loss": 2.4422,
+      "step": 2485
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.2436000406742096,
+      "learning_rate": 0.00031669982202798625,
+      "loss": 2.4637,
+      "step": 2490
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.23264776170253754,
+      "learning_rate": 0.0003160702893258846,
+      "loss": 2.509,
+      "step": 2495
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.23909597098827362,
+      "learning_rate": 0.00031544030589128886,
+      "loss": 2.254,
+      "step": 2500
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.21964912116527557,
+      "learning_rate": 0.0003148098760219553,
+      "loss": 2.3039,
+      "step": 2505
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.1836385279893875,
+      "learning_rate": 0.00031417900401868603,
+      "loss": 2.0602,
+      "step": 2510
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.22736455500125885,
+      "learning_rate": 0.00031354769418529905,
+      "loss": 2.4169,
+      "step": 2515
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.18662111461162567,
+      "learning_rate": 0.0003129159508285995,
+      "loss": 2.1888,
+      "step": 2520
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.20640943944454193,
+      "learning_rate": 0.00031228377825834975,
+      "loss": 2.2545,
+      "step": 2525
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.1896309107542038,
+      "learning_rate": 0.0003116511807872406,
+      "loss": 2.2505,
+      "step": 2530
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.20263446867465973,
+      "learning_rate": 0.0003110181627308615,
+      "loss": 2.1097,
+      "step": 2535
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.20662613213062286,
+      "learning_rate": 0.00031038472840767085,
+      "loss": 2.3434,
+      "step": 2540
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.24093309044837952,
+      "learning_rate": 0.000309750882138967,
+      "loss": 2.2969,
+      "step": 2545
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.1911349594593048,
+      "learning_rate": 0.00030911662824885866,
+      "loss": 2.4546,
+      "step": 2550
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.2091955691576004,
+      "learning_rate": 0.00030848197106423525,
+      "loss": 2.2451,
+      "step": 2555
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.20242448151111603,
+      "learning_rate": 0.0003078469149147376,
+      "loss": 2.1631,
+      "step": 2560
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.21048593521118164,
+      "learning_rate": 0.00030721146413272807,
+      "loss": 1.9984,
+      "step": 2565
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.18805581331253052,
+      "learning_rate": 0.00030657562305326134,
+      "loss": 2.4625,
+      "step": 2570
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.19612343609333038,
+      "learning_rate": 0.0003059393960140547,
+      "loss": 2.1259,
+      "step": 2575
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.3192237913608551,
+      "learning_rate": 0.00030530278735545847,
+      "loss": 2.3196,
+      "step": 2580
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.20382805168628693,
+      "learning_rate": 0.00030466580142042636,
+      "loss": 2.3005,
+      "step": 2585
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.20354492962360382,
+      "learning_rate": 0.00030402844255448587,
+      "loss": 1.9433,
+      "step": 2590
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.18796123564243317,
+      "learning_rate": 0.0003033907151057086,
+      "loss": 2.1679,
+      "step": 2595
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.22691605985164642,
+      "learning_rate": 0.0003027526234246807,
+      "loss": 2.184,
+      "step": 2600
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.19387076795101166,
+      "learning_rate": 0.00030211417186447304,
+      "loss": 2.3213,
+      "step": 2605
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.23006758093833923,
+      "learning_rate": 0.0003014753647806117,
+      "loss": 2.2843,
+      "step": 2610
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.2078760713338852,
+      "learning_rate": 0.0003008362065310481,
+      "loss": 2.0728,
+      "step": 2615
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2124081254005432,
+      "learning_rate": 0.0003001967014761292,
+      "loss": 2.1527,
+      "step": 2620
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.21508193016052246,
+      "learning_rate": 0.0002995568539785681,
+      "loss": 2.3249,
+      "step": 2625
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.22573944926261902,
+      "learning_rate": 0.0002989166684034139,
+      "loss": 2.5555,
+      "step": 2630
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.27325788140296936,
+      "learning_rate": 0.000298276149118022,
+      "loss": 2.1355,
+      "step": 2635
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2094535082578659,
+      "learning_rate": 0.0002976353004920245,
+      "loss": 2.2193,
+      "step": 2640
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.1797902137041092,
+      "learning_rate": 0.0002969941268973003,
+      "loss": 1.8739,
+      "step": 2645
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2198173552751541,
+      "learning_rate": 0.0002963526327079451,
+      "loss": 2.4586,
+      "step": 2650
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.23765875399112701,
+      "learning_rate": 0.0002957108223002419,
+      "loss": 2.2355,
+      "step": 2655
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2252362072467804,
+      "learning_rate": 0.0002950687000526307,
+      "loss": 2.3512,
+      "step": 2660
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.2460881769657135,
+      "learning_rate": 0.0002944262703456789,
+      "loss": 2.5086,
+      "step": 2665
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.18150685727596283,
+      "learning_rate": 0.0002937835375620518,
+      "loss": 2.1693,
+      "step": 2670
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.21276025474071503,
+      "learning_rate": 0.00029314050608648175,
+      "loss": 2.2578,
+      "step": 2675
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.21988217532634735,
+      "learning_rate": 0.00029249718030573905,
+      "loss": 2.2901,
+      "step": 2680
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.22700655460357666,
+      "learning_rate": 0.00029185356460860185,
+      "loss": 2.4618,
+      "step": 2685
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.24527020752429962,
+      "learning_rate": 0.00029120966338582583,
+      "loss": 2.2635,
+      "step": 2690
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.21412739157676697,
+      "learning_rate": 0.00029056548103011473,
+      "loss": 2.026,
+      "step": 2695
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.22026726603507996,
+      "learning_rate": 0.0002899210219360902,
+      "loss": 2.2242,
+      "step": 2700
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.23526617884635925,
+      "learning_rate": 0.00028927629050026164,
+      "loss": 2.466,
+      "step": 2705
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.1999528706073761,
+      "learning_rate": 0.00028863129112099663,
+      "loss": 2.1298,
+      "step": 2710
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.19018436968326569,
+      "learning_rate": 0.0002879860281984903,
+      "loss": 2.1991,
+      "step": 2715
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.23844487965106964,
+      "learning_rate": 0.000287340506134736,
+      "loss": 2.4711,
+      "step": 2720
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.22726939618587494,
+      "learning_rate": 0.00028669472933349485,
+      "loss": 2.3945,
+      "step": 2725
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.24586647748947144,
+      "learning_rate": 0.00028604870220026566,
+      "loss": 2.41,
+      "step": 2730
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.2375824749469757,
+      "learning_rate": 0.0002854024291422553,
+      "loss": 2.0514,
+      "step": 2735
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2294244021177292,
+      "learning_rate": 0.00028475591456834795,
+      "loss": 2.2207,
+      "step": 2740
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.252311646938324,
+      "learning_rate": 0.00028410916288907587,
+      "loss": 2.3079,
+      "step": 2745
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2053811103105545,
+      "learning_rate": 0.0002834621785165883,
+      "loss": 2.2646,
+      "step": 2750
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.20166157186031342,
+      "learning_rate": 0.0002828149658646225,
+      "loss": 2.348,
+      "step": 2755
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.20390185713768005,
+      "learning_rate": 0.0002821675293484726,
+      "loss": 2.1899,
+      "step": 2760
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.23896300792694092,
+      "learning_rate": 0.0002815198733849602,
+      "loss": 2.2631,
+      "step": 2765
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2838144302368164,
+      "learning_rate": 0.00028087200239240403,
+      "loss": 2.0516,
+      "step": 2770
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2110898345708847,
+      "learning_rate": 0.0002802239207905894,
+      "loss": 2.3501,
+      "step": 2775
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.18373216688632965,
+      "learning_rate": 0.00027957563300073864,
+      "loss": 2.0434,
+      "step": 2780
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.2768537998199463,
+      "learning_rate": 0.0002789271434454807,
+      "loss": 2.312,
+      "step": 2785
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.21091219782829285,
+      "learning_rate": 0.0002782784565488211,
+      "loss": 2.2531,
+      "step": 2790
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.19119109213352203,
+      "learning_rate": 0.0002776295767361113,
+      "loss": 2.3189,
+      "step": 2795
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.2107253223657608,
+      "learning_rate": 0.00027698050843401903,
+      "loss": 2.3991,
+      "step": 2800
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.20837560296058655,
+      "learning_rate": 0.00027633125607049786,
+      "loss": 2.2885,
+      "step": 2805
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.2551201283931732,
+      "learning_rate": 0.0002756818240747571,
+      "loss": 2.3301,
+      "step": 2810
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.24259725213050842,
+      "learning_rate": 0.0002750322168772316,
+      "loss": 2.1009,
+      "step": 2815
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.21995201706886292,
+      "learning_rate": 0.000274382438909551,
+      "loss": 2.2959,
+      "step": 2820
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.1952590048313141,
+      "learning_rate": 0.0002737324946045104,
+      "loss": 2.2697,
+      "step": 2825
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.2953265309333801,
+      "learning_rate": 0.0002730823883960395,
+      "loss": 2.1697,
+      "step": 2830
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.22461754083633423,
+      "learning_rate": 0.00027243212471917246,
+      "loss": 2.2691,
+      "step": 2835
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.21608883142471313,
+      "learning_rate": 0.0002717817080100177,
+      "loss": 2.3351,
+      "step": 2840
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.19717416167259216,
+      "learning_rate": 0.0002711311427057278,
+      "loss": 2.2855,
+      "step": 2845
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.212899848818779,
+      "learning_rate": 0.0002704804332444688,
+      "loss": 2.2025,
+      "step": 2850
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.2357543259859085,
+      "learning_rate": 0.0002698295840653903,
+      "loss": 2.3831,
+      "step": 2855
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.2233133614063263,
+      "learning_rate": 0.0002691785996085952,
+      "loss": 2.3989,
+      "step": 2860
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.22534696757793427,
+      "learning_rate": 0.0002685274843151089,
+      "loss": 2.3525,
+      "step": 2865
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.21673683822155,
+      "learning_rate": 0.00026787624262684977,
+      "loss": 2.3969,
+      "step": 2870
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.24868395924568176,
+      "learning_rate": 0.0002672248789865981,
+      "loss": 2.369,
+      "step": 2875
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.2035851776599884,
+      "learning_rate": 0.00026657339783796655,
+      "loss": 2.2208,
+      "step": 2880
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.2544965147972107,
+      "learning_rate": 0.000265921803625369,
+      "loss": 2.5607,
+      "step": 2885
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.17845949530601501,
+      "learning_rate": 0.00026527010079399084,
+      "loss": 2.3066,
+      "step": 2890
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.22786450386047363,
+      "learning_rate": 0.00026461829378975846,
+      "loss": 2.309,
+      "step": 2895
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.2404799461364746,
+      "learning_rate": 0.00026396638705930895,
+      "loss": 2.5578,
+      "step": 2900
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.19382694363594055,
+      "learning_rate": 0.0002633143850499598,
+      "loss": 2.3082,
+      "step": 2905
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.19829370081424713,
+      "learning_rate": 0.0002626622922096782,
+      "loss": 2.4811,
+      "step": 2910
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.23021139204502106,
+      "learning_rate": 0.0002620101129870513,
+      "loss": 2.2642,
+      "step": 2915
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.21615907549858093,
+      "learning_rate": 0.00026135785183125537,
+      "loss": 2.2167,
+      "step": 2920
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.23788948357105255,
+      "learning_rate": 0.00026070551319202573,
+      "loss": 2.3856,
+      "step": 2925
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.1953219771385193,
+      "learning_rate": 0.00026005310151962636,
+      "loss": 2.2164,
+      "step": 2930
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.2500264644622803,
+      "learning_rate": 0.0002594006212648192,
+      "loss": 2.4068,
+      "step": 2935
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.203148752450943,
+      "learning_rate": 0.00025874807687883424,
+      "loss": 2.3142,
+      "step": 2940
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.21646182239055634,
+      "learning_rate": 0.000258095472813339,
+      "loss": 2.0725,
+      "step": 2945
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.19882526993751526,
+      "learning_rate": 0.00025744281352040814,
+      "loss": 2.2717,
+      "step": 2950
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.22494642436504364,
+      "learning_rate": 0.00025679010345249306,
+      "loss": 2.1359,
+      "step": 2955
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.22282272577285767,
+      "learning_rate": 0.0002561373470623914,
+      "loss": 2.2144,
+      "step": 2960
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.20449025928974152,
+      "learning_rate": 0.000255484548803217,
+      "loss": 2.1777,
+      "step": 2965
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.23310497403144836,
+      "learning_rate": 0.0002548317131283694,
+      "loss": 2.1825,
+      "step": 2970
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.20909251272678375,
+      "learning_rate": 0.0002541788444915031,
+      "loss": 2.22,
+      "step": 2975
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.22772087156772614,
+      "learning_rate": 0.0002535259473464977,
+      "loss": 2.392,
+      "step": 2980
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.26885297894477844,
+      "learning_rate": 0.00025287302614742714,
+      "loss": 1.9405,
+      "step": 2985
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.2358844131231308,
+      "learning_rate": 0.00025222008534852956,
+      "loss": 2.2033,
+      "step": 2990
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.20965200662612915,
+      "learning_rate": 0.00025156712940417684,
+      "loss": 2.3244,
+      "step": 2995
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.20102304220199585,
+      "learning_rate": 0.0002509141627688441,
+      "loss": 2.258,
+      "step": 3000
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.23354652523994446,
+      "learning_rate": 0.00025026118989707934,
+      "loss": 2.2546,
+      "step": 3005
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.21666738390922546,
+      "learning_rate": 0.0002496082152434732,
+      "loss": 1.8389,
+      "step": 3010
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.2279592752456665,
+      "learning_rate": 0.00024895524326262855,
+      "loss": 2.2966,
+      "step": 3015
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.2545997202396393,
+      "learning_rate": 0.0002483022784091298,
+      "loss": 2.3153,
+      "step": 3020
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.22384227812290192,
+      "learning_rate": 0.000247649325137513,
+      "loss": 2.1932,
+      "step": 3025
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.22192451357841492,
+      "learning_rate": 0.0002469963879022349,
+      "loss": 2.3922,
+      "step": 3030
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.22841066122055054,
+      "learning_rate": 0.00024634347115764317,
+      "loss": 2.3009,
+      "step": 3035
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.19970761239528656,
+      "learning_rate": 0.00024569057935794537,
+      "loss": 2.0438,
+      "step": 3040
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.21427589654922485,
+      "learning_rate": 0.0002450377169571792,
+      "loss": 1.9967,
+      "step": 3045
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.23419593274593353,
+      "learning_rate": 0.0002443848884091816,
+      "loss": 2.3208,
+      "step": 3050
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.1909918338060379,
+      "learning_rate": 0.0002437320981675585,
+      "loss": 2.3367,
+      "step": 3055
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.20571142435073853,
+      "learning_rate": 0.0002430793506856548,
+      "loss": 2.1166,
+      "step": 3060
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.22468572854995728,
+      "learning_rate": 0.00024242665041652333,
+      "loss": 2.1951,
+      "step": 3065
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.2436290830373764,
+      "learning_rate": 0.00024177400181289518,
+      "loss": 2.2756,
+      "step": 3070
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.24295543134212494,
+      "learning_rate": 0.0002411214093271486,
+      "loss": 2.161,
+      "step": 3075
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.2191152274608612,
+      "learning_rate": 0.00024046887741127943,
+      "loss": 2.4584,
+      "step": 3080
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.23705042898654938,
+      "learning_rate": 0.00023981641051686986,
+      "loss": 2.2818,
+      "step": 3085
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.20371532440185547,
+      "learning_rate": 0.00023916401309505893,
+      "loss": 2.4664,
+      "step": 3090
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.2301088273525238,
+      "learning_rate": 0.00023851168959651147,
+      "loss": 2.3175,
+      "step": 3095
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.2044006735086441,
+      "learning_rate": 0.000237859444471388,
+      "loss": 2.2875,
+      "step": 3100
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.24112215638160706,
+      "learning_rate": 0.0002372072821693144,
+      "loss": 2.4344,
+      "step": 3105
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.23517994582653046,
+      "learning_rate": 0.00023655520713935176,
+      "loss": 2.2492,
+      "step": 3110
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.2575826346874237,
+      "learning_rate": 0.0002359032238299655,
+      "loss": 2.1888,
+      "step": 3115
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.22532737255096436,
+      "learning_rate": 0.00023525133668899536,
+      "loss": 2.3007,
+      "step": 3120
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.23337632417678833,
+      "learning_rate": 0.0002345995501636252,
+      "loss": 2.2881,
+      "step": 3125
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.21395531296730042,
+      "learning_rate": 0.0002339478687003523,
+      "loss": 2.2067,
+      "step": 3130
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.2543134093284607,
+      "learning_rate": 0.00023329629674495735,
+      "loss": 2.4062,
+      "step": 3135
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.22206565737724304,
+      "learning_rate": 0.00023264483874247383,
+      "loss": 2.1208,
+      "step": 3140
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.20878835022449493,
+      "learning_rate": 0.00023199349913715796,
+      "loss": 2.2897,
+      "step": 3145
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.2773876190185547,
+      "learning_rate": 0.000231342282372458,
+      "loss": 2.1967,
+      "step": 3150
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.22530554234981537,
+      "learning_rate": 0.00023069119289098467,
+      "loss": 2.0708,
+      "step": 3155
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.20695586502552032,
+      "learning_rate": 0.00023004023513447988,
+      "loss": 2.2647,
+      "step": 3160
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.2549115717411041,
+      "learning_rate": 0.0002293894135437871,
+      "loss": 2.3193,
+      "step": 3165
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.2138691544532776,
+      "learning_rate": 0.00022873873255882104,
+      "loss": 2.4966,
+      "step": 3170
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.24796932935714722,
+      "learning_rate": 0.00022808819661853682,
+      "loss": 2.3338,
+      "step": 3175
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.27691391110420227,
+      "learning_rate": 0.00022743781016090048,
+      "loss": 2.3752,
+      "step": 3180
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.22704118490219116,
+      "learning_rate": 0.00022678757762285793,
+      "loss": 2.4301,
+      "step": 3185
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.21799629926681519,
+      "learning_rate": 0.00022613750344030532,
+      "loss": 2.5869,
+      "step": 3190
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.20468707382678986,
+      "learning_rate": 0.00022548759204805817,
+      "loss": 2.3938,
+      "step": 3195
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.23200130462646484,
+      "learning_rate": 0.00022483784787982188,
+      "loss": 2.0689,
+      "step": 3200
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.22721296548843384,
+      "learning_rate": 0.00022418827536816068,
+      "loss": 2.2533,
+      "step": 3205
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.3599529564380646,
+      "learning_rate": 0.0002235388789444679,
+      "loss": 2.5117,
+      "step": 3210
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.22146733105182648,
+      "learning_rate": 0.00022288966303893548,
+      "loss": 2.115,
+      "step": 3215
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.20554299652576447,
+      "learning_rate": 0.00022224063208052408,
+      "loss": 2.2785,
+      "step": 3220
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.2130589336156845,
+      "learning_rate": 0.0002215917904969326,
+      "loss": 2.1105,
+      "step": 3225
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.21884912252426147,
+      "learning_rate": 0.00022094314271456779,
+      "loss": 2.397,
+      "step": 3230
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.2159346342086792,
+      "learning_rate": 0.00022029469315851458,
+      "loss": 2.1086,
+      "step": 3235
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.21349260210990906,
+      "learning_rate": 0.00021964644625250526,
+      "loss": 2.2467,
+      "step": 3240
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.19596756994724274,
+      "learning_rate": 0.0002189984064188901,
+      "loss": 2.199,
+      "step": 3245
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.2176314741373062,
+      "learning_rate": 0.0002183505780786063,
+      "loss": 2.2719,
+      "step": 3250
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.20054878294467926,
+      "learning_rate": 0.00021770296565114846,
+      "loss": 2.2418,
+      "step": 3255
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.21248690783977509,
+      "learning_rate": 0.00021705557355453808,
+      "loss": 2.1406,
+      "step": 3260
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.23126669228076935,
+      "learning_rate": 0.0002164084062052938,
+      "loss": 2.3417,
+      "step": 3265
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.24822832643985748,
+      "learning_rate": 0.00021576146801840072,
+      "loss": 2.2398,
+      "step": 3270
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.19908185303211212,
+      "learning_rate": 0.0002151147634072809,
+      "loss": 2.4964,
+      "step": 3275
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2246454805135727,
+      "learning_rate": 0.00021446829678376273,
+      "loss": 2.2703,
+      "step": 3280
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.23698629438877106,
+      "learning_rate": 0.00021382207255805097,
+      "loss": 2.2822,
+      "step": 3285
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2192452996969223,
+      "learning_rate": 0.00021317609513869717,
+      "loss": 2.1605,
+      "step": 3290
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.22096799314022064,
+      "learning_rate": 0.00021253036893256863,
+      "loss": 2.4929,
+      "step": 3295
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.23533667623996735,
+      "learning_rate": 0.00021188489834481926,
+      "loss": 2.5708,
+      "step": 3300
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2246759682893753,
+      "learning_rate": 0.00021123968777885877,
+      "loss": 2.3319,
+      "step": 3305
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2108616828918457,
+      "learning_rate": 0.0002105947416363235,
+      "loss": 2.306,
+      "step": 3310
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2039867490530014,
+      "learning_rate": 0.00020995006431704532,
+      "loss": 2.18,
+      "step": 3315
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2507101893424988,
+      "learning_rate": 0.00020930566021902275,
+      "loss": 2.4027,
+      "step": 3320
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.20139947533607483,
+      "learning_rate": 0.0002086615337383899,
+      "loss": 2.2227,
+      "step": 3325
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2274298071861267,
+      "learning_rate": 0.00020801768926938726,
+      "loss": 2.3289,
+      "step": 3330
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.2205355018377304,
+      "learning_rate": 0.00020737413120433128,
+      "loss": 2.1413,
+      "step": 3335
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.21190853416919708,
+      "learning_rate": 0.00020673086393358471,
+      "loss": 2.0141,
+      "step": 3340
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.22356361150741577,
+      "learning_rate": 0.00020608789184552642,
+      "loss": 2.0832,
+      "step": 3345
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.18460409343242645,
+      "learning_rate": 0.00020544521932652144,
+      "loss": 2.2919,
+      "step": 3350
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.2091461569070816,
+      "learning_rate": 0.00020480285076089136,
+      "loss": 1.8256,
+      "step": 3355
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.23582054674625397,
+      "learning_rate": 0.00020416079053088391,
+      "loss": 2.3861,
+      "step": 3360
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.24472753703594208,
+      "learning_rate": 0.0002035190430166437,
+      "loss": 2.4747,
+      "step": 3365
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.2311311960220337,
+      "learning_rate": 0.00020287761259618164,
+      "loss": 2.2835,
+      "step": 3370
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.25772756338119507,
+      "learning_rate": 0.00020223650364534568,
+      "loss": 2.2397,
+      "step": 3375
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.36200013756752014,
+      "learning_rate": 0.00020159572053779043,
+      "loss": 1.8731,
+      "step": 3380
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.21725964546203613,
+      "learning_rate": 0.00020095526764494796,
+      "loss": 2.3772,
+      "step": 3385
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.21761581301689148,
+      "learning_rate": 0.00020031514933599727,
+      "loss": 2.4379,
+      "step": 3390
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.22678567469120026,
+      "learning_rate": 0.00019967536997783495,
+      "loss": 2.1252,
+      "step": 3395
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.19343462586402893,
+      "learning_rate": 0.00019903593393504542,
+      "loss": 2.3515,
+      "step": 3400
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.24662891030311584,
+      "learning_rate": 0.00019839684556987076,
+      "loss": 2.1898,
+      "step": 3405
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.19625911116600037,
+      "learning_rate": 0.00019775810924218125,
+      "loss": 2.2743,
+      "step": 3410
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.25908294320106506,
+      "learning_rate": 0.0001971197293094456,
+      "loss": 2.4319,
+      "step": 3415
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.22597509622573853,
+      "learning_rate": 0.0001964817101267013,
+      "loss": 2.1773,
+      "step": 3420
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.23188504576683044,
+      "learning_rate": 0.00019584405604652444,
+      "loss": 2.3682,
+      "step": 3425
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.23394878208637238,
+      "learning_rate": 0.00019520677141900093,
+      "loss": 2.0693,
+      "step": 3430
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.20653656125068665,
+      "learning_rate": 0.00019456986059169568,
+      "loss": 2.2845,
+      "step": 3435
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.23298628628253937,
+      "learning_rate": 0.00019393332790962402,
+      "loss": 2.454,
+      "step": 3440
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.19534069299697876,
+      "learning_rate": 0.00019329717771522108,
+      "loss": 2.4001,
+      "step": 3445
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.19667889177799225,
+      "learning_rate": 0.00019266141434831326,
+      "loss": 2.3783,
+      "step": 3450
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.22638565301895142,
+      "learning_rate": 0.00019202604214608744,
+      "loss": 2.2708,
+      "step": 3455
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.24946092069149017,
+      "learning_rate": 0.0001913910654430622,
+      "loss": 2.4107,
+      "step": 3460
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.23316776752471924,
+      "learning_rate": 0.00019075648857105825,
+      "loss": 2.3116,
+      "step": 3465
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.20107421278953552,
+      "learning_rate": 0.00019012231585916817,
+      "loss": 2.2918,
+      "step": 3470
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.21125708520412445,
+      "learning_rate": 0.00018948855163372782,
+      "loss": 2.1628,
+      "step": 3475
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.23127441108226776,
+      "learning_rate": 0.00018885520021828602,
+      "loss": 2.1008,
+      "step": 3480
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.2128669172525406,
+      "learning_rate": 0.00018822226593357563,
+      "loss": 2.3034,
+      "step": 3485
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.20843705534934998,
+      "learning_rate": 0.00018758975309748354,
+      "loss": 2.4882,
+      "step": 3490
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.20661096274852753,
+      "learning_rate": 0.00018695766602502196,
+      "loss": 2.3373,
+      "step": 3495
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.23054824769496918,
+      "learning_rate": 0.00018632600902829808,
+      "loss": 2.542,
+      "step": 3500
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.2079750895500183,
+      "learning_rate": 0.00018569478641648535,
+      "loss": 1.9235,
+      "step": 3505
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.22011853754520416,
+      "learning_rate": 0.00018506400249579383,
+      "loss": 2.4797,
+      "step": 3510
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.19704569876194,
+      "learning_rate": 0.00018443366156944068,
+      "loss": 2.2134,
+      "step": 3515
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.20672884583473206,
+      "learning_rate": 0.0001838037679376213,
+      "loss": 2.2853,
+      "step": 3520
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.21415551006793976,
+      "learning_rate": 0.00018317432589747915,
+      "loss": 2.316,
+      "step": 3525
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.23471008241176605,
+      "learning_rate": 0.0001825453397430773,
+      "loss": 2.213,
+      "step": 3530
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.22792291641235352,
+      "learning_rate": 0.00018191681376536844,
+      "loss": 2.2338,
+      "step": 3535
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.20559385418891907,
+      "learning_rate": 0.0001812887522521664,
+      "loss": 2.4132,
+      "step": 3540
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.19658410549163818,
+      "learning_rate": 0.00018066115948811595,
+      "loss": 2.3406,
+      "step": 3545
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.27269044518470764,
+      "learning_rate": 0.0001800340397546643,
+      "loss": 2.0845,
+      "step": 3550
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.19628912210464478,
+      "learning_rate": 0.00017940739733003148,
+      "loss": 2.0894,
+      "step": 3555
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.20123037695884705,
+      "learning_rate": 0.0001787812364891816,
+      "loss": 2.2939,
+      "step": 3560
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.2504325211048126,
+      "learning_rate": 0.00017815556150379297,
+      "loss": 2.3354,
+      "step": 3565
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.211093932390213,
+      "learning_rate": 0.0001775303766422298,
+      "loss": 2.2469,
+      "step": 3570
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.22017726302146912,
+      "learning_rate": 0.00017690568616951247,
+      "loss": 2.0667,
+      "step": 3575
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.19698256254196167,
+      "learning_rate": 0.00017628149434728858,
+      "loss": 2.1873,
+      "step": 3580
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.1986425369977951,
+      "learning_rate": 0.0001756578054338041,
+      "loss": 2.3461,
+      "step": 3585
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.23417946696281433,
+      "learning_rate": 0.00017503462368387396,
+      "loss": 2.1092,
+      "step": 3590
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.2174525111913681,
+      "learning_rate": 0.00017441195334885341,
+      "loss": 2.3174,
+      "step": 3595
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.1901862919330597,
+      "learning_rate": 0.00017378979867660848,
+      "loss": 2.341,
+      "step": 3600
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.22507810592651367,
+      "learning_rate": 0.00017316816391148792,
+      "loss": 2.0203,
+      "step": 3605
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.23076196014881134,
+      "learning_rate": 0.00017254705329429302,
+      "loss": 2.2094,
+      "step": 3610
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.2597287893295288,
+      "learning_rate": 0.00017192647106224984,
+      "loss": 2.2167,
+      "step": 3615
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.18957456946372986,
+      "learning_rate": 0.00017130642144897942,
+      "loss": 2.2965,
+      "step": 3620
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.208304300904274,
+      "learning_rate": 0.00017068690868446957,
+      "loss": 2.4638,
+      "step": 3625
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.23491379618644714,
+      "learning_rate": 0.00017006793699504537,
+      "loss": 2.607,
+      "step": 3630
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.22672338783740997,
+      "learning_rate": 0.000169449510603341,
+      "loss": 2.3012,
+      "step": 3635
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.23909202218055725,
+      "learning_rate": 0.00016883163372827053,
+      "loss": 2.2488,
+      "step": 3640
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.21005770564079285,
+      "learning_rate": 0.00016821431058499896,
+      "loss": 2.2493,
+      "step": 3645
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.212179496884346,
+      "learning_rate": 0.00016759754538491422,
+      "loss": 1.8298,
+      "step": 3650
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.23341526091098785,
+      "learning_rate": 0.00016698134233559736,
+      "loss": 2.0907,
+      "step": 3655
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.2126818150281906,
+      "learning_rate": 0.000166365705640795,
+      "loss": 2.229,
+      "step": 3660
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.22661077976226807,
+      "learning_rate": 0.00016575063950038962,
+      "loss": 2.3294,
+      "step": 3665
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.21582266688346863,
+      "learning_rate": 0.00016513614811037168,
+      "loss": 2.2743,
+      "step": 3670
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.21785694360733032,
+      "learning_rate": 0.00016452223566281035,
+      "loss": 2.2205,
+      "step": 3675
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.21953895688056946,
+      "learning_rate": 0.00016390890634582572,
+      "loss": 2.386,
+      "step": 3680
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.23978057503700256,
+      "learning_rate": 0.00016329616434355933,
+      "loss": 2.3754,
+      "step": 3685
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.22289539873600006,
+      "learning_rate": 0.0001626840138361462,
+      "loss": 2.3556,
+      "step": 3690
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.2133941650390625,
+      "learning_rate": 0.00016207245899968633,
+      "loss": 2.2085,
+      "step": 3695
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.22433440387248993,
+      "learning_rate": 0.00016146150400621565,
+      "loss": 2.3706,
+      "step": 3700
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.22882284224033356,
+      "learning_rate": 0.00016085115302367844,
+      "loss": 2.175,
+      "step": 3705
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.21862849593162537,
+      "learning_rate": 0.000160241410215898,
+      "loss": 2.405,
+      "step": 3710
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.2291046679019928,
+      "learning_rate": 0.00015963227974254891,
+      "loss": 2.3178,
+      "step": 3715
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.25386229157447815,
+      "learning_rate": 0.00015902376575912814,
+      "loss": 2.2498,
+      "step": 3720
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.23087728023529053,
+      "learning_rate": 0.0001584158724169273,
+      "loss": 2.0111,
+      "step": 3725
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.35551658272743225,
+      "learning_rate": 0.00015780860386300366,
+      "loss": 2.0455,
+      "step": 3730
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.2827921509742737,
+      "learning_rate": 0.00015720196424015237,
+      "loss": 2.0207,
+      "step": 3735
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.21254152059555054,
+      "learning_rate": 0.00015659595768687787,
+      "loss": 2.3726,
+      "step": 3740
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.24501194059848785,
+      "learning_rate": 0.0001559905883373659,
+      "loss": 2.0826,
+      "step": 3745
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.25266799330711365,
+      "learning_rate": 0.00015538586032145523,
+      "loss": 2.4135,
+      "step": 3750
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.21061569452285767,
+      "learning_rate": 0.00015478177776460922,
+      "loss": 2.2499,
+      "step": 3755
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.2409357875585556,
+      "learning_rate": 0.00015417834478788817,
+      "loss": 2.1908,
+      "step": 3760
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.2231035977602005,
+      "learning_rate": 0.00015357556550792064,
+      "loss": 2.3146,
+      "step": 3765
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.21707406640052795,
+      "learning_rate": 0.00015297344403687594,
+      "loss": 2.0881,
+      "step": 3770
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.25137150287628174,
+      "learning_rate": 0.00015237198448243565,
+      "loss": 2.3653,
+      "step": 3775
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.21213415265083313,
+      "learning_rate": 0.0001517711909477658,
+      "loss": 2.0724,
+      "step": 3780
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.22044523060321808,
+      "learning_rate": 0.00015117106753148864,
+      "loss": 2.3951,
+      "step": 3785
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.21106037497520447,
+      "learning_rate": 0.0001505716183276553,
+      "loss": 2.3063,
+      "step": 3790
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.19864067435264587,
+      "learning_rate": 0.00014997284742571687,
+      "loss": 2.3512,
+      "step": 3795
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.2682659327983856,
+      "learning_rate": 0.00014937475891049754,
+      "loss": 2.1263,
+      "step": 3800
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.22299188375473022,
+      "learning_rate": 0.0001487773568621659,
+      "loss": 2.1646,
+      "step": 3805
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.1971612125635147,
+      "learning_rate": 0.00014818064535620755,
+      "loss": 2.2026,
+      "step": 3810
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.230462908744812,
+      "learning_rate": 0.00014758462846339748,
+      "loss": 2.0656,
+      "step": 3815
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.21233803033828735,
+      "learning_rate": 0.00014698931024977152,
+      "loss": 2.0036,
+      "step": 3820
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.22545789182186127,
+      "learning_rate": 0.0001463946947765995,
+      "loss": 2.3219,
+      "step": 3825
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.25745728611946106,
+      "learning_rate": 0.0001458007861003568,
+      "loss": 2.1843,
+      "step": 3830
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.22463242709636688,
+      "learning_rate": 0.00014520758827269762,
+      "loss": 2.253,
+      "step": 3835
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.221448615193367,
+      "learning_rate": 0.00014461510534042593,
+      "loss": 2.2945,
+      "step": 3840
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.2240522801876068,
+      "learning_rate": 0.0001440233413454693,
+      "loss": 2.495,
+      "step": 3845
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.2830246090888977,
+      "learning_rate": 0.0001434323003248504,
+      "loss": 2.1455,
+      "step": 3850
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.22345268726348877,
+      "learning_rate": 0.0001428419863106599,
+      "loss": 2.1584,
+      "step": 3855
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.2145606279373169,
+      "learning_rate": 0.0001422524033300287,
+      "loss": 2.2282,
+      "step": 3860
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.20531050860881805,
+      "learning_rate": 0.00014166355540510072,
+      "loss": 2.3778,
+      "step": 3865
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.20582695305347443,
+      "learning_rate": 0.00014107544655300519,
+      "loss": 2.364,
+      "step": 3870
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.2178577184677124,
+      "learning_rate": 0.00014048808078582942,
+      "loss": 2.533,
+      "step": 3875
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.2206876426935196,
+      "learning_rate": 0.0001399014621105914,
+      "loss": 2.5413,
+      "step": 3880
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.24441972374916077,
+      "learning_rate": 0.00013931559452921247,
+      "loss": 2.0963,
+      "step": 3885
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.1869293451309204,
+      "learning_rate": 0.00013873048203849,
+      "loss": 2.0351,
+      "step": 3890
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.21759897470474243,
+      "learning_rate": 0.00013814612863006985,
+      "loss": 2.1202,
+      "step": 3895
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.2100670337677002,
+      "learning_rate": 0.00013756253829041998,
+      "loss": 2.3812,
+      "step": 3900
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.203419029712677,
+      "learning_rate": 0.00013697971500080207,
+      "loss": 2.4032,
+      "step": 3905
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.24871662259101868,
+      "learning_rate": 0.00013639766273724542,
+      "loss": 2.2387,
+      "step": 3910
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.24824783205986023,
+      "learning_rate": 0.0001358163854705191,
+      "loss": 2.2322,
+      "step": 3915
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.21718545258045197,
+      "learning_rate": 0.0001352358871661052,
+      "loss": 2.3572,
+      "step": 3920
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.22530969977378845,
+      "learning_rate": 0.00013465617178417202,
+      "loss": 2.2706,
+      "step": 3925
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.3539257049560547,
+      "learning_rate": 0.0001340772432795463,
+      "loss": 2.2439,
+      "step": 3930
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.22788652777671814,
+      "learning_rate": 0.00013349910560168688,
+      "loss": 2.2209,
+      "step": 3935
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.24959754943847656,
+      "learning_rate": 0.0001329217626946576,
+      "loss": 2.1656,
+      "step": 3940
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.2394050508737564,
+      "learning_rate": 0.0001323452184971004,
+      "loss": 2.2512,
+      "step": 3945
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.207530215382576,
+      "learning_rate": 0.00013176947694220828,
+      "loss": 2.3349,
+      "step": 3950
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.2458074539899826,
+      "learning_rate": 0.0001311945419576988,
+      "loss": 2.3544,
+      "step": 3955
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.23957344889640808,
+      "learning_rate": 0.00013062041746578668,
+      "loss": 2.1734,
+      "step": 3960
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.23613759875297546,
+      "learning_rate": 0.00013004710738315805,
+      "loss": 2.1414,
+      "step": 3965
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.20104865729808807,
+      "learning_rate": 0.0001294746156209425,
+      "loss": 2.2497,
+      "step": 3970
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.2077418565750122,
+      "learning_rate": 0.00012890294608468764,
+      "loss": 2.0288,
+      "step": 3975
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.190739706158638,
+      "learning_rate": 0.00012833210267433127,
+      "loss": 2.1426,
+      "step": 3980
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.21029919385910034,
+      "learning_rate": 0.00012776208928417558,
+      "loss": 2.3764,
+      "step": 3985
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.22516466677188873,
+      "learning_rate": 0.0001271929098028607,
+      "loss": 2.3506,
+      "step": 3990
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.25254198908805847,
+      "learning_rate": 0.00012662456811333718,
+      "loss": 2.3135,
+      "step": 3995
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.23021924495697021,
+      "learning_rate": 0.00012605706809284056,
+      "loss": 2.0895,
+      "step": 4000
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.2016134113073349,
+      "learning_rate": 0.00012549041361286438,
+      "loss": 2.3918,
+      "step": 4005
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.22848759591579437,
+      "learning_rate": 0.000124924608539134,
+      "loss": 2.0874,
+      "step": 4010
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.24023045599460602,
+      "learning_rate": 0.00012435965673157992,
+      "loss": 2.2176,
+      "step": 4015
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.2871743440628052,
+      "learning_rate": 0.00012379556204431184,
+      "loss": 2.0991,
+      "step": 4020
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.24772612750530243,
+      "learning_rate": 0.000123232328325592,
+      "loss": 2.3852,
+      "step": 4025
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.27144259214401245,
+      "learning_rate": 0.00012266995941780933,
+      "loss": 2.2463,
+      "step": 4030
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.22941002249717712,
+      "learning_rate": 0.0001221084591574526,
+      "loss": 1.9907,
+      "step": 4035
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.19429859519004822,
+      "learning_rate": 0.00012154783137508513,
+      "loss": 2.3679,
+      "step": 4040
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.2515440881252289,
+      "learning_rate": 0.00012098807989531804,
+      "loss": 2.2745,
+      "step": 4045
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.22819548845291138,
+      "learning_rate": 0.00012042920853678393,
+      "loss": 2.3092,
+      "step": 4050
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.23505914211273193,
+      "learning_rate": 0.00011987122111211185,
+      "loss": 2.237,
+      "step": 4055
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.208960622549057,
+      "learning_rate": 0.00011931412142790005,
+      "loss": 2.1689,
+      "step": 4060
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.24931026995182037,
+      "learning_rate": 0.00011875791328469088,
+      "loss": 2.2817,
+      "step": 4065
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.2663825452327728,
+      "learning_rate": 0.0001182026004769446,
+      "loss": 2.4229,
+      "step": 4070
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.32719677686691284,
+      "learning_rate": 0.0001176481867930134,
+      "loss": 2.2895,
+      "step": 4075
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.19801324605941772,
+      "learning_rate": 0.00011709467601511565,
+      "loss": 2.2313,
+      "step": 4080
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.2740652859210968,
+      "learning_rate": 0.0001165420719193101,
+      "loss": 2.4916,
+      "step": 4085
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.228300541639328,
+      "learning_rate": 0.0001159903782754701,
+      "loss": 1.8949,
+      "step": 4090
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.22289882600307465,
+      "learning_rate": 0.00011543959884725793,
+      "loss": 2.32,
+      "step": 4095
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.2218303680419922,
+      "learning_rate": 0.00011488973739209879,
+      "loss": 2.279,
+      "step": 4100
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.2361016869544983,
+      "learning_rate": 0.00011434079766115594,
+      "loss": 2.4161,
+      "step": 4105
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.2459881603717804,
+      "learning_rate": 0.0001137927833993043,
+      "loss": 2.2137,
+      "step": 4110
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.205433651804924,
+      "learning_rate": 0.00011324569834510523,
+      "loss": 2.4675,
+      "step": 4115
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.20708201825618744,
+      "learning_rate": 0.00011269954623078127,
+      "loss": 2.4206,
+      "step": 4120
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.20594294369220734,
+      "learning_rate": 0.0001121543307821899,
+      "loss": 2.1661,
+      "step": 4125
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.21572817862033844,
+      "learning_rate": 0.0001116100557187994,
+      "loss": 2.3033,
+      "step": 4130
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.27235686779022217,
+      "learning_rate": 0.00011106672475366203,
+      "loss": 2.4498,
+      "step": 4135
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.208398699760437,
+      "learning_rate": 0.0001105243415933898,
+      "loss": 2.3131,
+      "step": 4140
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2954976260662079,
+      "learning_rate": 0.0001099829099381287,
+      "loss": 2.0314,
+      "step": 4145
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2156749665737152,
+      "learning_rate": 0.00010944243348153352,
+      "loss": 2.2011,
+      "step": 4150
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.21214473247528076,
+      "learning_rate": 0.00010890291591074262,
+      "loss": 2.4135,
+      "step": 4155
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.18999165296554565,
+      "learning_rate": 0.00010836436090635288,
+      "loss": 2.3648,
+      "step": 4160
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.2245589941740036,
+      "learning_rate": 0.00010782677214239456,
+      "loss": 2.2032,
+      "step": 4165
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.21589384973049164,
+      "learning_rate": 0.00010729015328630617,
+      "loss": 2.2427,
+      "step": 4170
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.23020033538341522,
+      "learning_rate": 0.00010675450799890951,
+      "loss": 2.4896,
+      "step": 4175
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.21616196632385254,
+      "learning_rate": 0.00010621983993438467,
+      "loss": 2.1164,
+      "step": 4180
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.214460089802742,
+      "learning_rate": 0.0001056861527402452,
+      "loss": 2.3076,
+      "step": 4185
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.217409148812294,
+      "learning_rate": 0.00010515345005731283,
+      "loss": 2.3753,
+      "step": 4190
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.2724614143371582,
+      "learning_rate": 0.00010462173551969348,
+      "loss": 2.409,
+      "step": 4195
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.21833041310310364,
+      "learning_rate": 0.00010409101275475139,
+      "loss": 2.4145,
+      "step": 4200
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.22455398738384247,
+      "learning_rate": 0.00010356128538308519,
+      "loss": 2.4656,
+      "step": 4205
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.24207107722759247,
+      "learning_rate": 0.00010303255701850287,
+      "loss": 2.5042,
+      "step": 4210
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.21409133076667786,
+      "learning_rate": 0.00010250483126799711,
+      "loss": 2.4075,
+      "step": 4215
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.27256593108177185,
+      "learning_rate": 0.00010197811173172106,
+      "loss": 2.1344,
+      "step": 4220
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.2253335416316986,
+      "learning_rate": 0.00010145240200296296,
+      "loss": 2.095,
+      "step": 4225
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.22114749252796173,
+      "learning_rate": 0.00010092770566812245,
+      "loss": 2.3584,
+      "step": 4230
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.20449167490005493,
+      "learning_rate": 0.00010040402630668575,
+      "loss": 2.0765,
+      "step": 4235
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.2127346694469452,
+      "learning_rate": 9.988136749120117e-05,
+      "loss": 2.4296,
+      "step": 4240
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.24239586293697357,
+      "learning_rate": 9.935973278725497e-05,
+      "loss": 2.2133,
+      "step": 4245
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.20855852961540222,
+      "learning_rate": 9.883912575344689e-05,
+      "loss": 2.1825,
+      "step": 4250
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.22541500627994537,
+      "learning_rate": 9.831954994136558e-05,
+      "loss": 2.4416,
+      "step": 4255
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.21852771937847137,
+      "learning_rate": 9.780100889556528e-05,
+      "loss": 2.1338,
+      "step": 4260
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.2524181306362152,
+      "learning_rate": 9.728350615354037e-05,
+      "loss": 2.2715,
+      "step": 4265
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.223457932472229,
+      "learning_rate": 9.676704524570257e-05,
+      "loss": 2.1475,
+      "step": 4270
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.20966559648513794,
+      "learning_rate": 9.625162969535562e-05,
+      "loss": 2.1721,
+      "step": 4275
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.257662296295166,
+      "learning_rate": 9.5737263018672e-05,
+      "loss": 2.5223,
+      "step": 4280
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.21245898306369781,
+      "learning_rate": 9.522394872466913e-05,
+      "loss": 2.5489,
+      "step": 4285
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.2315073013305664,
+      "learning_rate": 9.471169031518445e-05,
+      "loss": 2.2782,
+      "step": 4290
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.21125444769859314,
+      "learning_rate": 9.420049128485253e-05,
+      "loss": 2.2031,
+      "step": 4295
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.21013091504573822,
+      "learning_rate": 9.36903551210807e-05,
+      "loss": 1.8751,
+      "step": 4300
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.25472453236579895,
+      "learning_rate": 9.318128530402548e-05,
+      "loss": 2.1909,
+      "step": 4305
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.20788568258285522,
+      "learning_rate": 9.26732853065687e-05,
+      "loss": 2.1907,
+      "step": 4310
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.22828350961208344,
+      "learning_rate": 9.216635859429389e-05,
+      "loss": 2.173,
+      "step": 4315
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.2420109063386917,
+      "learning_rate": 9.16605086254626e-05,
+      "loss": 2.2958,
+      "step": 4320
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.39677414298057556,
+      "learning_rate": 9.11557388509909e-05,
+      "loss": 2.0855,
+      "step": 4325
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.29712197184562683,
+      "learning_rate": 9.065205271442542e-05,
+      "loss": 2.4342,
+      "step": 4330
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.22807803750038147,
+      "learning_rate": 9.014945365192074e-05,
+      "loss": 2.1303,
+      "step": 4335
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.22611750662326813,
+      "learning_rate": 8.964794509221508e-05,
+      "loss": 2.3752,
+      "step": 4340
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.1972074657678604,
+      "learning_rate": 8.914753045660706e-05,
+      "loss": 2.2315,
+      "step": 4345
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.23331503570079803,
+      "learning_rate": 8.864821315893307e-05,
+      "loss": 2.0982,
+      "step": 4350
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.20077547430992126,
+      "learning_rate": 8.814999660554279e-05,
+      "loss": 2.3864,
+      "step": 4355
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.21390070021152496,
+      "learning_rate": 8.765288419527725e-05,
+      "loss": 2.4906,
+      "step": 4360
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.24563802778720856,
+      "learning_rate": 8.715687931944449e-05,
+      "loss": 2.3063,
+      "step": 4365
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.18615995347499847,
+      "learning_rate": 8.666198536179718e-05,
+      "loss": 2.4023,
+      "step": 4370
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2110363095998764,
+      "learning_rate": 8.616820569850922e-05,
+      "loss": 2.445,
+      "step": 4375
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.21250484883785248,
+      "learning_rate": 8.567554369815286e-05,
+      "loss": 2.2042,
+      "step": 4380
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.21482475101947784,
+      "learning_rate": 8.518400272167554e-05,
+      "loss": 2.0429,
+      "step": 4385
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2096213549375534,
+      "learning_rate": 8.46935861223771e-05,
+      "loss": 2.3407,
+      "step": 4390
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.2219087928533554,
+      "learning_rate": 8.420429724588682e-05,
+      "loss": 2.2471,
+      "step": 4395
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.22424234449863434,
+      "learning_rate": 8.371613943014076e-05,
+      "loss": 2.443,
+      "step": 4400
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.23273713886737823,
+      "learning_rate": 8.32291160053587e-05,
+      "loss": 2.0754,
+      "step": 4405
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.20582637190818787,
+      "learning_rate": 8.274323029402173e-05,
+      "loss": 2.3168,
+      "step": 4410
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.24268639087677002,
+      "learning_rate": 8.225848561084942e-05,
+      "loss": 2.3469,
+      "step": 4415
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.27532529830932617,
+      "learning_rate": 8.177488526277696e-05,
+      "loss": 2.1499,
+      "step": 4420
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.19710592925548553,
+      "learning_rate": 8.129243254893343e-05,
+      "loss": 2.3333,
+      "step": 4425
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.22231920063495636,
+      "learning_rate": 8.081113076061816e-05,
+      "loss": 2.617,
+      "step": 4430
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.24639876186847687,
+      "learning_rate": 8.033098318127919e-05,
+      "loss": 2.3569,
+      "step": 4435
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2539924383163452,
+      "learning_rate": 7.985199308649038e-05,
+      "loss": 2.3275,
+      "step": 4440
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2054421603679657,
+      "learning_rate": 7.937416374392931e-05,
+      "loss": 2.2305,
+      "step": 4445
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.21331799030303955,
+      "learning_rate": 7.88974984133548e-05,
+      "loss": 2.4914,
+      "step": 4450
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2482718825340271,
+      "learning_rate": 7.842200034658486e-05,
+      "loss": 2.4796,
+      "step": 4455
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2408265769481659,
+      "learning_rate": 7.794767278747434e-05,
+      "loss": 2.246,
+      "step": 4460
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2019105851650238,
+      "learning_rate": 7.747451897189287e-05,
+      "loss": 2.4315,
+      "step": 4465
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.23757004737854004,
+      "learning_rate": 7.70025421277028e-05,
+      "loss": 2.3275,
+      "step": 4470
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.21571137011051178,
+      "learning_rate": 7.653174547473717e-05,
+      "loss": 2.1763,
+      "step": 4475
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.24419759213924408,
+      "learning_rate": 7.606213222477784e-05,
+      "loss": 2.1517,
+      "step": 4480
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.23553812503814697,
+      "learning_rate": 7.559370558153308e-05,
+      "loss": 2.2369,
+      "step": 4485
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.25342613458633423,
+      "learning_rate": 7.512646874061677e-05,
+      "loss": 2.1151,
+      "step": 4490
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.21691171824932098,
+      "learning_rate": 7.466042488952521e-05,
+      "loss": 2.121,
+      "step": 4495
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2163485735654831,
+      "learning_rate": 7.419557720761678e-05,
+      "loss": 2.3613,
+      "step": 4500
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.23227451741695404,
+      "learning_rate": 7.373192886608907e-05,
+      "loss": 2.351,
+      "step": 4505
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2794537842273712,
+      "learning_rate": 7.326948302795792e-05,
+      "loss": 2.2958,
+      "step": 4510
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.25960350036621094,
+      "learning_rate": 7.280824284803595e-05,
+      "loss": 2.3849,
+      "step": 4515
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2537117898464203,
+      "learning_rate": 7.234821147291023e-05,
+      "loss": 2.2611,
+      "step": 4520
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2555319666862488,
+      "learning_rate": 7.188939204092177e-05,
+      "loss": 2.2563,
+      "step": 4525
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2340226173400879,
+      "learning_rate": 7.143178768214348e-05,
+      "loss": 2.2398,
+      "step": 4530
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.18839353322982788,
+      "learning_rate": 7.097540151835913e-05,
+      "loss": 1.9488,
+      "step": 4535
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2070079892873764,
+      "learning_rate": 7.052023666304189e-05,
+      "loss": 2.3041,
+      "step": 4540
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.20171380043029785,
+      "learning_rate": 7.006629622133318e-05,
+      "loss": 2.2266,
+      "step": 4545
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.22852955758571625,
+      "learning_rate": 6.961358329002134e-05,
+      "loss": 2.2571,
+      "step": 4550
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.2285429835319519,
+      "learning_rate": 6.916210095752091e-05,
+      "loss": 2.0589,
+      "step": 4555
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.22552116215229034,
+      "learning_rate": 6.871185230385072e-05,
+      "loss": 2.2549,
+      "step": 4560
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.3575563132762909,
+      "learning_rate": 6.826284040061407e-05,
+      "loss": 2.1802,
+      "step": 4565
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.20591920614242554,
+      "learning_rate": 6.781506831097678e-05,
+      "loss": 2.4576,
+      "step": 4570
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.23244622349739075,
+      "learning_rate": 6.736853908964647e-05,
+      "loss": 2.4535,
+      "step": 4575
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.23467615246772766,
+      "learning_rate": 6.692325578285241e-05,
+      "loss": 2.3651,
+      "step": 4580
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.2762286365032196,
+      "learning_rate": 6.647922142832366e-05,
+      "loss": 2.321,
+      "step": 4585
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.23728686571121216,
+      "learning_rate": 6.603643905526947e-05,
+      "loss": 2.2093,
+      "step": 4590
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.24722512066364288,
+      "learning_rate": 6.559491168435749e-05,
+      "loss": 2.2426,
+      "step": 4595
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.1793704330921173,
+      "learning_rate": 6.51546423276941e-05,
+      "loss": 2.0517,
+      "step": 4600
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.2555958926677704,
+      "learning_rate": 6.47156339888034e-05,
+      "loss": 2.1139,
+      "step": 4605
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.23529459536075592,
+      "learning_rate": 6.427788966260675e-05,
+      "loss": 2.2764,
+      "step": 4610
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.24994830787181854,
+      "learning_rate": 6.384141233540248e-05,
+      "loss": 2.135,
+      "step": 4615
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.21067097783088684,
+      "learning_rate": 6.340620498484542e-05,
+      "loss": 2.2185,
+      "step": 4620
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.2730359435081482,
+      "learning_rate": 6.297227057992638e-05,
+      "loss": 1.9392,
+      "step": 4625
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.22600042819976807,
+      "learning_rate": 6.253961208095255e-05,
+      "loss": 2.2381,
+      "step": 4630
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.20506155490875244,
+      "learning_rate": 6.210823243952654e-05,
+      "loss": 2.2401,
+      "step": 4635
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.21911026537418365,
+      "learning_rate": 6.167813459852665e-05,
+      "loss": 2.349,
+      "step": 4640
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.3007972240447998,
+      "learning_rate": 6.124932149208681e-05,
+      "loss": 2.3174,
+      "step": 4645
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.2639479339122772,
+      "learning_rate": 6.082179604557617e-05,
+      "loss": 2.191,
+      "step": 4650
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.35899296402931213,
+      "learning_rate": 6.0395561175579984e-05,
+      "loss": 2.3853,
+      "step": 4655
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.22475650906562805,
+      "learning_rate": 5.997061978987861e-05,
+      "loss": 2.6994,
+      "step": 4660
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.22194553911685944,
+      "learning_rate": 5.95469747874286e-05,
+      "loss": 2.3443,
+      "step": 4665
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.2621094286441803,
+      "learning_rate": 5.912462905834243e-05,
+      "loss": 2.3218,
+      "step": 4670
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.22809450328350067,
+      "learning_rate": 5.8703585483868923e-05,
+      "loss": 2.0625,
+      "step": 4675
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.24936749041080475,
+      "learning_rate": 5.828384693637356e-05,
+      "loss": 2.2313,
+      "step": 4680
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.22074222564697266,
+      "learning_rate": 5.786541627931899e-05,
+      "loss": 2.4042,
+      "step": 4685
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.2226363569498062,
+      "learning_rate": 5.744829636724536e-05,
+      "loss": 2.2453,
+      "step": 4690
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.23048317432403564,
+      "learning_rate": 5.703249004575087e-05,
+      "loss": 2.2149,
+      "step": 4695
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.26077502965927124,
+      "learning_rate": 5.661800015147245e-05,
+      "loss": 2.4554,
+      "step": 4700
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.23060084879398346,
+      "learning_rate": 5.620482951206626e-05,
+      "loss": 2.0871,
+      "step": 4705
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.23240596055984497,
+      "learning_rate": 5.579298094618865e-05,
+      "loss": 2.068,
+      "step": 4710
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.25822100043296814,
+      "learning_rate": 5.5382457263476404e-05,
+      "loss": 2.3347,
+      "step": 4715
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.2152366042137146,
+      "learning_rate": 5.497326126452848e-05,
+      "loss": 2.2502,
+      "step": 4720
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.22713406383991241,
+      "learning_rate": 5.4565395740885784e-05,
+      "loss": 2.2203,
+      "step": 4725
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.19853945076465607,
+      "learning_rate": 5.4158863475013196e-05,
+      "loss": 2.3458,
+      "step": 4730
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.20661340653896332,
+      "learning_rate": 5.3753667240279716e-05,
+      "loss": 2.2679,
+      "step": 4735
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.2174801528453827,
+      "learning_rate": 5.334980980094004e-05,
+      "loss": 2.1825,
+      "step": 4740
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.24562963843345642,
+      "learning_rate": 5.2947293912115836e-05,
+      "loss": 2.5649,
+      "step": 4745
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.19424057006835938,
+      "learning_rate": 5.254612231977624e-05,
+      "loss": 2.1822,
+      "step": 4750
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.220481738448143,
+      "learning_rate": 5.214629776071986e-05,
+      "loss": 2.2884,
+      "step": 4755
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.2209591120481491,
+      "learning_rate": 5.174782296255581e-05,
+      "loss": 2.4984,
+      "step": 4760
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.23579350113868713,
+      "learning_rate": 5.135070064368508e-05,
+      "loss": 2.2501,
+      "step": 4765
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.2045588195323944,
+      "learning_rate": 5.095493351328201e-05,
+      "loss": 2.1386,
+      "step": 4770
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.23356984555721283,
+      "learning_rate": 5.0560524271275804e-05,
+      "loss": 2.087,
+      "step": 4775
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.24123229086399078,
+      "learning_rate": 5.016747560833221e-05,
+      "loss": 2.3423,
+      "step": 4780
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.23942683637142181,
+      "learning_rate": 4.977579020583509e-05,
+      "loss": 2.2332,
+      "step": 4785
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.22866415977478027,
+      "learning_rate": 4.938547073586783e-05,
+      "loss": 2.3039,
+      "step": 4790
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2132473886013031,
+      "learning_rate": 4.8996519861195964e-05,
+      "loss": 2.1985,
+      "step": 4795
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.24595317244529724,
+      "learning_rate": 4.860894023524792e-05,
+      "loss": 2.4184,
+      "step": 4800
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.21361759305000305,
+      "learning_rate": 4.822273450209766e-05,
+      "loss": 2.2322,
+      "step": 4805
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2093634456396103,
+      "learning_rate": 4.78379052964466e-05,
+      "loss": 2.283,
+      "step": 4810
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2636638283729553,
+      "learning_rate": 4.745445524360506e-05,
+      "loss": 2.1172,
+      "step": 4815
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.23667100071907043,
+      "learning_rate": 4.707238695947522e-05,
+      "loss": 2.2553,
+      "step": 4820
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.20943553745746613,
+      "learning_rate": 4.669170305053239e-05,
+      "loss": 2.2184,
+      "step": 4825
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.22824349999427795,
+      "learning_rate": 4.631240611380793e-05,
+      "loss": 2.2645,
+      "step": 4830
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.18925786018371582,
+      "learning_rate": 4.593449873687117e-05,
+      "loss": 2.3492,
+      "step": 4835
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.21233032643795013,
+      "learning_rate": 4.555798349781185e-05,
+      "loss": 2.2667,
+      "step": 4840
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.25917020440101624,
+      "learning_rate": 4.518286296522248e-05,
+      "loss": 2.3071,
+      "step": 4845
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.25229182839393616,
+      "learning_rate": 4.4809139698180985e-05,
+      "loss": 2.3561,
+      "step": 4850
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.2252894639968872,
+      "learning_rate": 4.4436816246232855e-05,
+      "loss": 2.5364,
+      "step": 4855
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.2240176647901535,
+      "learning_rate": 4.406589514937437e-05,
+      "loss": 2.2504,
+      "step": 4860
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.2048347145318985,
+      "learning_rate": 4.369637893803474e-05,
+      "loss": 2.243,
+      "step": 4865
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.21030034124851227,
+      "learning_rate": 4.3328270133058984e-05,
+      "loss": 2.2484,
+      "step": 4870
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.2221430093050003,
+      "learning_rate": 4.2961571245690945e-05,
+      "loss": 2.1924,
+      "step": 4875
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.22565896809101105,
+      "learning_rate": 4.2596284777555646e-05,
+      "loss": 2.2848,
+      "step": 4880
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.22470615804195404,
+      "learning_rate": 4.223241322064306e-05,
+      "loss": 2.2022,
+      "step": 4885
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.3345583975315094,
+      "learning_rate": 4.186995905729013e-05,
+      "loss": 2.195,
+      "step": 4890
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.24246099591255188,
+      "learning_rate": 4.1508924760164654e-05,
+      "loss": 2.2407,
+      "step": 4895
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.2674398124217987,
+      "learning_rate": 4.1149312792247956e-05,
+      "loss": 2.2382,
+      "step": 4900
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.23504556715488434,
+      "learning_rate": 4.079112560681816e-05,
+      "loss": 2.2465,
+      "step": 4905
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.26757216453552246,
+      "learning_rate": 4.043436564743361e-05,
+      "loss": 2.1608,
+      "step": 4910
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.2398817539215088,
+      "learning_rate": 4.007903534791604e-05,
+      "loss": 2.3138,
+      "step": 4915
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.21942365169525146,
+      "learning_rate": 3.9725137132334e-05,
+      "loss": 2.1774,
+      "step": 4920
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.33368757367134094,
+      "learning_rate": 3.937267341498641e-05,
+      "loss": 1.9173,
+      "step": 4925
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.2571778893470764,
+      "learning_rate": 3.90216466003859e-05,
+      "loss": 2.1053,
+      "step": 4930
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.21170833706855774,
+      "learning_rate": 3.86720590832427e-05,
+      "loss": 2.1862,
+      "step": 4935
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.2919519543647766,
+      "learning_rate": 3.8323913248448055e-05,
+      "loss": 2.0265,
+      "step": 4940
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.27031105756759644,
+      "learning_rate": 3.7977211471057814e-05,
+      "loss": 2.3355,
+      "step": 4945
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.22384324669837952,
+      "learning_rate": 3.763195611627693e-05,
+      "loss": 2.5411,
+      "step": 4950
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.24857838451862335,
+      "learning_rate": 3.728814953944229e-05,
+      "loss": 2.1088,
+      "step": 4955
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.2427876591682434,
+      "learning_rate": 3.69457940860077e-05,
+      "loss": 2.3843,
+      "step": 4960
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.24170845746994019,
+      "learning_rate": 3.660489209152701e-05,
+      "loss": 2.2145,
+      "step": 4965
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.23208285868167877,
+      "learning_rate": 3.6265445881638685e-05,
+      "loss": 1.9527,
+      "step": 4970
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.2421301156282425,
+      "learning_rate": 3.592745777204981e-05,
+      "loss": 2.3853,
+      "step": 4975
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.21723875403404236,
+      "learning_rate": 3.5590930068520324e-05,
+      "loss": 2.2972,
+      "step": 4980
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.23736509680747986,
+      "learning_rate": 3.525586506684719e-05,
+      "loss": 2.3117,
+      "step": 4985
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.229548841714859,
+      "learning_rate": 3.492226505284879e-05,
+      "loss": 2.3366,
+      "step": 4990
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.2537176311016083,
+      "learning_rate": 3.459013230234945e-05,
+      "loss": 2.3424,
+      "step": 4995
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.2257246971130371,
+      "learning_rate": 3.425946908116365e-05,
+      "loss": 2.3804,
+      "step": 5000
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.2141430377960205,
+      "learning_rate": 3.3930277645080804e-05,
+      "loss": 2.1892,
+      "step": 5005
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.26059871912002563,
+      "learning_rate": 3.3602560239849786e-05,
+      "loss": 2.1947,
+      "step": 5010
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.22793832421302795,
+      "learning_rate": 3.3276319101163636e-05,
+      "loss": 2.4303,
+      "step": 5015
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.2286398559808731,
+      "learning_rate": 3.295155645464412e-05,
+      "loss": 2.2965,
+      "step": 5020
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.26674941182136536,
+      "learning_rate": 3.262827451582701e-05,
+      "loss": 2.1779,
+      "step": 5025
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.2102755457162857,
+      "learning_rate": 3.230647549014634e-05,
+      "loss": 2.4149,
+      "step": 5030
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.23954099416732788,
+      "learning_rate": 3.1986161572919806e-05,
+      "loss": 2.3689,
+      "step": 5035
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.22200442850589752,
+      "learning_rate": 3.1667334949333935e-05,
+      "loss": 2.5024,
+      "step": 5040
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.23744402825832367,
+      "learning_rate": 3.13499977944284e-05,
+      "loss": 2.4955,
+      "step": 5045
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.26780271530151367,
+      "learning_rate": 3.1034152273082233e-05,
+      "loss": 1.8371,
+      "step": 5050
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.22308044135570526,
+      "learning_rate": 3.07198005399981e-05,
+      "loss": 2.42,
+      "step": 5055
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.22626711428165436,
+      "learning_rate": 3.0406944739688247e-05,
+      "loss": 1.995,
+      "step": 5060
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.2728048264980316,
+      "learning_rate": 3.0095587006459563e-05,
+      "loss": 2.1475,
+      "step": 5065
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.22108417749404907,
+      "learning_rate": 2.978572946439914e-05,
+      "loss": 2.263,
+      "step": 5070
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.21562892198562622,
+      "learning_rate": 2.9477374227359715e-05,
+      "loss": 2.2598,
+      "step": 5075
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.21353229880332947,
+      "learning_rate": 2.917052339894538e-05,
+      "loss": 2.2058,
+      "step": 5080
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.24220535159111023,
+      "learning_rate": 2.8865179072496834e-05,
+      "loss": 2.1816,
+      "step": 5085
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.21273867785930634,
+      "learning_rate": 2.8561343331077794e-05,
+      "loss": 2.4305,
+      "step": 5090
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.2410360872745514,
+      "learning_rate": 2.8259018247460104e-05,
+      "loss": 2.3677,
+      "step": 5095
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.23682256042957306,
+      "learning_rate": 2.7958205884110022e-05,
+      "loss": 2.2552,
+      "step": 5100
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.2274460345506668,
+      "learning_rate": 2.765890829317397e-05,
+      "loss": 2.1807,
+      "step": 5105
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.21735264360904694,
+      "learning_rate": 2.7361127516464413e-05,
+      "loss": 2.3206,
+      "step": 5110
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.2178596407175064,
+      "learning_rate": 2.7064865585446435e-05,
+      "loss": 2.4849,
+      "step": 5115
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.2024727314710617,
+      "learning_rate": 2.6770124521223156e-05,
+      "loss": 2.2573,
+      "step": 5120
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.23709452152252197,
+      "learning_rate": 2.6476906334522544e-05,
+      "loss": 2.4347,
+      "step": 5125
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.24053430557250977,
+      "learning_rate": 2.6185213025683442e-05,
+      "loss": 2.3239,
+      "step": 5130
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.24593400955200195,
+      "learning_rate": 2.5895046584641925e-05,
+      "loss": 1.9137,
+      "step": 5135
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.21003013849258423,
+      "learning_rate": 2.5606408990917734e-05,
+      "loss": 2.3909,
+      "step": 5140
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.22038382291793823,
+      "learning_rate": 2.531930221360082e-05,
+      "loss": 2.2642,
+      "step": 5145
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.21890124678611755,
+      "learning_rate": 2.50337282113379e-05,
+      "loss": 2.248,
+      "step": 5150
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.20427949726581573,
+      "learning_rate": 2.4749688932319046e-05,
+      "loss": 2.2361,
+      "step": 5155
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.18769937753677368,
+      "learning_rate": 2.4467186314264423e-05,
+      "loss": 2.3167,
+      "step": 5160
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.21898379921913147,
+      "learning_rate": 2.4186222284411115e-05,
+      "loss": 2.1296,
+      "step": 5165
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.2148793488740921,
+      "learning_rate": 2.3906798759499916e-05,
+      "loss": 2.2478,
+      "step": 5170
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.21558383107185364,
+      "learning_rate": 2.362891764576211e-05,
+      "loss": 2.2518,
+      "step": 5175
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.1823437213897705,
+      "learning_rate": 2.335258083890701e-05,
+      "loss": 1.9974,
+      "step": 5180
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.214274063706398,
+      "learning_rate": 2.307779022410819e-05,
+      "loss": 2.3863,
+      "step": 5185
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.2402074933052063,
+      "learning_rate": 2.2804547675991493e-05,
+      "loss": 2.2795,
+      "step": 5190
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.23359552025794983,
+      "learning_rate": 2.2532855058621486e-05,
+      "loss": 2.1839,
+      "step": 5195
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.1988850086927414,
+      "learning_rate": 2.226271422548931e-05,
+      "loss": 2.3525,
+      "step": 5200
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.23410113155841827,
+      "learning_rate": 2.199412701949974e-05,
+      "loss": 2.2221,
+      "step": 5205
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.28235799074172974,
+      "learning_rate": 2.1727095272958658e-05,
+      "loss": 2.3737,
+      "step": 5210
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.19444677233695984,
+      "learning_rate": 2.1461620807560588e-05,
+      "loss": 2.0633,
+      "step": 5215
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2312103509902954,
+      "learning_rate": 2.1197705434376323e-05,
+      "loss": 2.4128,
+      "step": 5220
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.24011710286140442,
+      "learning_rate": 2.0935350953840386e-05,
+      "loss": 2.0844,
+      "step": 5225
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.20998087525367737,
+      "learning_rate": 2.0674559155738986e-05,
+      "loss": 2.4681,
+      "step": 5230
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.23423248529434204,
+      "learning_rate": 2.0415331819197634e-05,
+      "loss": 2.3379,
+      "step": 5235
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.25770696997642517,
+      "learning_rate": 2.015767071266908e-05,
+      "loss": 2.329,
+      "step": 5240
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.22478801012039185,
+      "learning_rate": 1.9901577593921193e-05,
+      "loss": 2.2112,
+      "step": 5245
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.22718951106071472,
+      "learning_rate": 1.9647054210024968e-05,
+      "loss": 2.4007,
+      "step": 5250
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.2245934009552002,
+      "learning_rate": 1.9394102297342842e-05,
+      "loss": 1.9822,
+      "step": 5255
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.25233885645866394,
+      "learning_rate": 1.9142723581516418e-05,
+      "loss": 1.8663,
+      "step": 5260
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.22092613577842712,
+      "learning_rate": 1.8892919777455054e-05,
+      "loss": 2.24,
+      "step": 5265
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.2314857840538025,
+      "learning_rate": 1.864469258932397e-05,
+      "loss": 2.3645,
+      "step": 5270
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.22606311738491058,
+      "learning_rate": 1.8398043710532748e-05,
+      "loss": 2.103,
+      "step": 5275
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.2500253915786743,
+      "learning_rate": 1.8152974823723772e-05,
+      "loss": 2.0572,
+      "step": 5280
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.21233054995536804,
+      "learning_rate": 1.7909487600760526e-05,
+      "loss": 2.4147,
+      "step": 5285
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.27136868238449097,
+      "learning_rate": 1.766758370271651e-05,
+      "loss": 2.4119,
+      "step": 5290
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.21505457162857056,
+      "learning_rate": 1.74272647798637e-05,
+      "loss": 2.0964,
+      "step": 5295
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.23759673535823822,
+      "learning_rate": 1.718853247166133e-05,
+      "loss": 2.4041,
+      "step": 5300
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.1982789784669876,
+      "learning_rate": 1.6951388406744732e-05,
+      "loss": 2.308,
+      "step": 5305
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.2191469967365265,
+      "learning_rate": 1.671583420291434e-05,
+      "loss": 2.1455,
+      "step": 5310
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.21145908534526825,
+      "learning_rate": 1.6481871467124267e-05,
+      "loss": 2.2348,
+      "step": 5315
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.2330002337694168,
+      "learning_rate": 1.6249501795471938e-05,
+      "loss": 2.1234,
+      "step": 5320
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.22962640225887299,
+      "learning_rate": 1.6018726773186525e-05,
+      "loss": 2.1627,
+      "step": 5325
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.24119548499584198,
+      "learning_rate": 1.578954797461879e-05,
+      "loss": 2.3492,
+      "step": 5330
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.21263697743415833,
+      "learning_rate": 1.5561966963229924e-05,
+      "loss": 2.3261,
+      "step": 5335
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.22236277163028717,
+      "learning_rate": 1.5335985291580785e-05,
+      "loss": 2.2343,
+      "step": 5340
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.21450620889663696,
+      "learning_rate": 1.5111604501321875e-05,
+      "loss": 2.2677,
+      "step": 5345
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.2557521462440491,
+      "learning_rate": 1.4888826123182153e-05,
+      "loss": 2.2499,
+      "step": 5350
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.22849878668785095,
+      "learning_rate": 1.4667651676959077e-05,
+      "loss": 2.1613,
+      "step": 5355
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.24565404653549194,
+      "learning_rate": 1.4448082671507989e-05,
+      "loss": 2.2011,
+      "step": 5360
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.26400044560432434,
+      "learning_rate": 1.4230120604731889e-05,
+      "loss": 2.5032,
+      "step": 5365
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.2253475785255432,
+      "learning_rate": 1.4013766963571262e-05,
+      "loss": 2.2235,
+      "step": 5370
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.2189287692308426,
+      "learning_rate": 1.3799023223993817e-05,
+      "loss": 2.3313,
+      "step": 5375
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.23772679269313812,
+      "learning_rate": 1.3585890850984579e-05,
+      "loss": 2.4549,
+      "step": 5380
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.23463352024555206,
+      "learning_rate": 1.337437129853572e-05,
+      "loss": 2.1687,
+      "step": 5385
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.22411875426769257,
+      "learning_rate": 1.3164466009636778e-05,
+      "loss": 2.3785,
+      "step": 5390
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.23704136908054352,
+      "learning_rate": 1.295617641626476e-05,
+      "loss": 2.3108,
+      "step": 5395
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.19436527788639069,
+      "learning_rate": 1.2749503939374379e-05,
+      "loss": 2.1299,
+      "step": 5400
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.19581307470798492,
+      "learning_rate": 1.2544449988888235e-05,
+      "loss": 2.2972,
+      "step": 5405
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.2575695216655731,
+      "learning_rate": 1.2341015963687557e-05,
+      "loss": 2.236,
+      "step": 5410
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.2203637808561325,
+      "learning_rate": 1.213920325160217e-05,
+      "loss": 2.0656,
+      "step": 5415
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.19933323562145233,
+      "learning_rate": 1.1939013229401524e-05,
+      "loss": 2.232,
+      "step": 5420
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.21780821681022644,
+      "learning_rate": 1.1740447262784781e-05,
+      "loss": 2.3308,
+      "step": 5425
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.2099289745092392,
+      "learning_rate": 1.1543506706371998e-05,
+      "loss": 2.3933,
+      "step": 5430
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.262759268283844,
+      "learning_rate": 1.1348192903694543e-05,
+      "loss": 2.1099,
+      "step": 5435
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.28049829602241516,
+      "learning_rate": 1.1154507187186052e-05,
+      "loss": 2.3418,
+      "step": 5440
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.21992290019989014,
+      "learning_rate": 1.0962450878173408e-05,
+      "loss": 2.1937,
+      "step": 5445
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.21217088401317596,
+      "learning_rate": 1.0772025286867577e-05,
+      "loss": 2.2756,
+      "step": 5450
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2052336037158966,
+      "learning_rate": 1.0583231712354756e-05,
+      "loss": 2.1314,
+      "step": 5455
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2902153432369232,
+      "learning_rate": 1.039607144258753e-05,
+      "loss": 2.0119,
+      "step": 5460
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.28160467743873596,
+      "learning_rate": 1.0210545754376021e-05,
+      "loss": 1.9997,
+      "step": 5465
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.2551237642765045,
+      "learning_rate": 1.0026655913379229e-05,
+      "loss": 2.5376,
+      "step": 5470
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.21193884313106537,
+      "learning_rate": 9.844403174096433e-06,
+      "loss": 2.1938,
+      "step": 5475
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.21645359694957733,
+      "learning_rate": 9.663788779858408e-06,
+      "loss": 2.1168,
+      "step": 5480
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.19699765741825104,
+      "learning_rate": 9.484813962819361e-06,
+      "loss": 2.1105,
+      "step": 5485
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.21276994049549103,
+      "learning_rate": 9.307479943948039e-06,
+      "loss": 2.2799,
+      "step": 5490
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.22967317700386047,
+      "learning_rate": 9.131787933019831e-06,
+      "loss": 2.5073,
+      "step": 5495
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.21648356318473816,
+      "learning_rate": 8.9577391286082e-06,
+      "loss": 2.3877,
+      "step": 5500
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.23259419202804565,
+      "learning_rate": 8.78533471807666e-06,
+      "loss": 2.4268,
+      "step": 5505
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.22746941447257996,
+      "learning_rate": 8.614575877570696e-06,
+      "loss": 2.212,
+      "step": 5510
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.3145082890987396,
+      "learning_rate": 8.445463772009638e-06,
+      "loss": 2.1488,
+      "step": 5515
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.23564931750297546,
+      "learning_rate": 8.27799955507877e-06,
+      "loss": 2.2061,
+      "step": 5520
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.22794529795646667,
+      "learning_rate": 8.112184369221482e-06,
+      "loss": 2.2778,
+      "step": 5525
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.21888861060142517,
+      "learning_rate": 7.948019345631385e-06,
+      "loss": 2.1319,
+      "step": 5530
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.2596055567264557,
+      "learning_rate": 7.78550560424479e-06,
+      "loss": 2.363,
+      "step": 5535
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.21776314079761505,
+      "learning_rate": 7.624644253732799e-06,
+      "loss": 2.1601,
+      "step": 5540
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.23371025919914246,
+      "learning_rate": 7.465436391493885e-06,
+      "loss": 2.2857,
+      "step": 5545
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.23128168284893036,
+      "learning_rate": 7.3078831036465785e-06,
+      "loss": 2.2666,
+      "step": 5550
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.20787528157234192,
+      "learning_rate": 7.151985465021604e-06,
+      "loss": 2.5238,
+      "step": 5555
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.19910071790218353,
+      "learning_rate": 6.997744539155026e-06,
+      "loss": 2.2942,
+      "step": 5560
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.19725759327411652,
+      "learning_rate": 6.845161378280784e-06,
+      "loss": 2.3778,
+      "step": 5565
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.20873206853866577,
+      "learning_rate": 6.694237023323335e-06,
+      "loss": 2.2222,
+      "step": 5570
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.2186024934053421,
+      "learning_rate": 6.544972503891023e-06,
+      "loss": 2.2635,
+      "step": 5575
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.21914984285831451,
+      "learning_rate": 6.3973688382684965e-06,
+      "loss": 1.8,
+      "step": 5580
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.2277553379535675,
+      "learning_rate": 6.251427033410195e-06,
+      "loss": 1.8708,
+      "step": 5585
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.2023899406194687,
+      "learning_rate": 6.1071480849332896e-06,
+      "loss": 2.3289,
+      "step": 5590
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.1917661875486374,
+      "learning_rate": 5.964532977110887e-06,
+      "loss": 2.175,
+      "step": 5595
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.3647346496582031,
+      "learning_rate": 5.82358268286537e-06,
+      "loss": 2.2159,
+      "step": 5600
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.2529740333557129,
+      "learning_rate": 5.684298163761759e-06,
+      "loss": 2.3442,
+      "step": 5605
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.3258639872074127,
+      "learning_rate": 5.546680370001028e-06,
+      "loss": 2.1641,
+      "step": 5610
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.21551235020160675,
+      "learning_rate": 5.410730240413942e-06,
+      "loss": 2.3734,
+      "step": 5615
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.2223397046327591,
+      "learning_rate": 5.276448702454195e-06,
+      "loss": 2.0221,
+      "step": 5620
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.2447679489850998,
+      "learning_rate": 5.143836672192481e-06,
+      "loss": 2.073,
+      "step": 5625
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.20842759311199188,
+      "learning_rate": 5.012895054310046e-06,
+      "loss": 2.3238,
+      "step": 5630
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.23142164945602417,
+      "learning_rate": 4.883624742092501e-06,
+      "loss": 2.2459,
+      "step": 5635
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.21301431953907013,
+      "learning_rate": 4.756026617423887e-06,
+      "loss": 2.3963,
+      "step": 5640
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.20507967472076416,
+      "learning_rate": 4.630101550780447e-06,
+      "loss": 2.3633,
+      "step": 5645
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.22849854826927185,
+      "learning_rate": 4.505850401224921e-06,
+      "loss": 2.4213,
+      "step": 5650
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.21087034046649933,
+      "learning_rate": 4.3832740164004295e-06,
+      "loss": 2.3358,
+      "step": 5655
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.2474902719259262,
+      "learning_rate": 4.2623732325249295e-06,
+      "loss": 1.9333,
+      "step": 5660
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.22979308664798737,
+      "learning_rate": 4.143148874385355e-06,
+      "loss": 2.4708,
+      "step": 5665
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.24336537718772888,
+      "learning_rate": 4.025601755332065e-06,
+      "loss": 2.2467,
+      "step": 5670
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.25388413667678833,
+      "learning_rate": 3.909732677273209e-06,
+      "loss": 2.0566,
+      "step": 5675
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.23465120792388916,
+      "learning_rate": 3.795542430669374e-06,
+      "loss": 2.2796,
+      "step": 5680
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.23578540980815887,
+      "learning_rate": 3.6830317945281133e-06,
+      "loss": 2.2443,
+      "step": 5685
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.23347768187522888,
+      "learning_rate": 3.5722015363986725e-06,
+      "loss": 2.1135,
+      "step": 5690
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.23163768649101257,
+      "learning_rate": 3.4630524123667184e-06,
+      "loss": 2.5128,
+      "step": 5695
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.20352327823638916,
+      "learning_rate": 3.3555851670491467e-06,
+      "loss": 2.2458,
+      "step": 5700
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.20329436659812927,
+      "learning_rate": 3.24980053358917e-06,
+      "loss": 2.2312,
+      "step": 5705
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.2172505259513855,
+      "learning_rate": 3.1456992336510726e-06,
+      "loss": 2.3672,
+      "step": 5710
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.23608310520648956,
+      "learning_rate": 3.0432819774154907e-06,
+      "loss": 2.2544,
+      "step": 5715
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.23203229904174805,
+      "learning_rate": 2.942549463574473e-06,
+      "loss": 2.4567,
+      "step": 5720
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.22394001483917236,
+      "learning_rate": 2.8435023793267068e-06,
+      "loss": 2.2673,
+      "step": 5725
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.2103736698627472,
+      "learning_rate": 2.7461414003728534e-06,
+      "loss": 2.347,
+      "step": 5730
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.22648335993289948,
+      "learning_rate": 2.650467190910999e-06,
+      "loss": 2.2428,
+      "step": 5735
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.24892614781856537,
+      "learning_rate": 2.556480403631989e-06,
+      "loss": 2.0718,
+      "step": 5740
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2146364003419876,
+      "learning_rate": 2.4641816797151272e-06,
+      "loss": 2.2748,
+      "step": 5745
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.24034225940704346,
+      "learning_rate": 2.373571648823597e-06,
+      "loss": 2.4486,
+      "step": 5750
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.19975844025611877,
+      "learning_rate": 2.284650929100435e-06,
+      "loss": 1.9455,
+      "step": 5755
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2771877646446228,
+      "learning_rate": 2.197420127164118e-06,
+      "loss": 2.3526,
+      "step": 5760
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.20685018599033356,
+      "learning_rate": 2.1118798381044023e-06,
+      "loss": 2.2497,
+      "step": 5765
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.25831371545791626,
+      "learning_rate": 2.02803064547849e-06,
+      "loss": 2.4169,
+      "step": 5770
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2329753190279007,
+      "learning_rate": 1.945873121306757e-06,
+      "loss": 2.1808,
+      "step": 5775
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.1924462765455246,
+      "learning_rate": 1.8654078260691153e-06,
+      "loss": 2.074,
+      "step": 5780
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.21585677564144135,
+      "learning_rate": 1.7866353087009345e-06,
+      "loss": 2.1959,
+      "step": 5785
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.23273953795433044,
+      "learning_rate": 1.709556106589516e-06,
+      "loss": 2.3726,
+      "step": 5790
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.23679770529270172,
+      "learning_rate": 1.6341707455703459e-06,
+      "loss": 2.0897,
+      "step": 5795
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2224605232477188,
+      "learning_rate": 1.5604797399234595e-06,
+      "loss": 2.4029,
+      "step": 5800
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.2761968672275543,
+      "learning_rate": 1.488483592369999e-06,
+      "loss": 2.1526,
+      "step": 5805
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.24104470014572144,
+      "learning_rate": 1.4181827940687453e-06,
+      "loss": 2.2732,
+      "step": 5810
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.19636249542236328,
+      "learning_rate": 1.3495778246127576e-06,
+      "loss": 2.1891,
+      "step": 5815
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.23419730365276337,
+      "learning_rate": 1.2826691520262114e-06,
+      "loss": 2.4011,
+      "step": 5820
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.21348519623279572,
+      "learning_rate": 1.2174572327610378e-06,
+      "loss": 2.3018,
+      "step": 5825
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.24424676597118378,
+      "learning_rate": 1.1539425116939006e-06,
+      "loss": 2.4398,
+      "step": 5830
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.3099956810474396,
+      "learning_rate": 1.092125422123197e-06,
+      "loss": 2.2096,
+      "step": 5835
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.20208124816417694,
+      "learning_rate": 1.0320063857660045e-06,
+      "loss": 2.235,
+      "step": 5840
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.22363179922103882,
+      "learning_rate": 9.735858127553066e-07,
+      "loss": 2.3201,
+      "step": 5845
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.22977089881896973,
+      "learning_rate": 9.168641016371049e-07,
+      "loss": 2.2843,
+      "step": 5850
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.22061312198638916,
+      "learning_rate": 8.618416393677552e-07,
+      "loss": 2.4087,
+      "step": 5855
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.22010397911071777,
+      "learning_rate": 8.085188013113032e-07,
+      "loss": 2.205,
+      "step": 5860
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.22382394969463348,
+      "learning_rate": 7.568959512369577e-07,
+      "loss": 2.2887,
+      "step": 5865
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.21971046924591064,
+      "learning_rate": 7.069734413165663e-07,
+      "loss": 2.4646,
+      "step": 5870
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.21887187659740448,
+      "learning_rate": 6.587516121221993e-07,
+      "loss": 2.3914,
+      "step": 5875
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.20574615895748138,
+      "learning_rate": 6.122307926239579e-07,
+      "loss": 2.4691,
+      "step": 5880
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.21829432249069214,
+      "learning_rate": 5.674113001875036e-07,
+      "loss": 2.328,
+      "step": 5885
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.20351313054561615,
+      "learning_rate": 5.242934405720879e-07,
+      "loss": 2.2867,
+      "step": 5890
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.22187699377536774,
+      "learning_rate": 4.828775079284142e-07,
+      "loss": 2.6343,
+      "step": 5895
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.25424492359161377,
+      "learning_rate": 4.431637847965575e-07,
+      "loss": 2.214,
+      "step": 5900
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2767295241355896,
+      "learning_rate": 4.0515254210415907e-07,
+      "loss": 2.1654,
+      "step": 5905
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2616841495037079,
+      "learning_rate": 3.6884403916445653e-07,
+      "loss": 2.1905,
+      "step": 5910
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.24903003871440887,
+      "learning_rate": 3.3423852367461817e-07,
+      "loss": 2.3208,
+      "step": 5915
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.21697789430618286,
+      "learning_rate": 3.0133623171396673e-07,
+      "loss": 2.1046,
+      "step": 5920
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.24143174290657043,
+      "learning_rate": 2.701373877424251e-07,
+      "loss": 2.2847,
+      "step": 5925
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.19240717589855194,
+      "learning_rate": 2.40642204598962e-07,
+      "loss": 2.371,
+      "step": 5930
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.22101587057113647,
+      "learning_rate": 2.1285088350017635e-07,
+      "loss": 2.1873,
+      "step": 5935
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.30621862411499023,
+      "learning_rate": 1.86763614038854e-07,
+      "loss": 2.4638,
+      "step": 5940
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.20043669641017914,
+      "learning_rate": 1.6238057418274666e-07,
+      "loss": 2.1642,
+      "step": 5945
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.2317458689212799,
+      "learning_rate": 1.3970193027332268e-07,
+      "loss": 2.4367,
+      "step": 5950
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.2099936455488205,
+      "learning_rate": 1.1872783702460144e-07,
+      "loss": 2.3994,
+      "step": 5955
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.197767436504364,
+      "learning_rate": 9.945843752218186e-08,
+      "loss": 2.1562,
+      "step": 5960
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.2169959545135498,
+      "learning_rate": 8.189386322218773e-08,
+      "loss": 2.2637,
+      "step": 5965
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.28220120072364807,
+      "learning_rate": 6.60342339504072e-08,
+      "loss": 2.2485,
+      "step": 5970
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.25559163093566895,
+      "learning_rate": 5.187965790143245e-08,
+      "loss": 2.3216,
+      "step": 5975
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.21520258486270905,
+      "learning_rate": 3.943023163804904e-08,
+      "loss": 2.3339,
+      "step": 5980
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.23260967433452606,
+      "learning_rate": 2.8686040090375453e-08,
+      "loss": 2.3379,
+      "step": 5985
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.24584035575389862,
+      "learning_rate": 1.9647156555474556e-08,
+      "loss": 2.3093,
+      "step": 5990
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.23227186501026154,
+      "learning_rate": 1.2313642696742955e-08,
+      "loss": 2.3064,
+      "step": 5995
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.24923665821552277,
+      "learning_rate": 6.685548543522435e-09,
+      "loss": 2.5481,
+      "step": 6000
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.2306099534034729,
+      "learning_rate": 2.7629124907391223e-09,
+      "loss": 2.3828,
+      "step": 6005
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.2319669872522354,
+      "learning_rate": 5.457612986814464e-10,
+      "loss": 2.4561,
+      "step": 6010
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 2.2096378803253174,
+      "eval_runtime": 372.0471,
+      "eval_samples_per_second": 3.22,
+      "eval_steps_per_second": 3.22,
+      "step": 6014
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 6014,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 500,
+  "total_flos": 6296538124910592.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}