diff --git "a/Qwen3.4B-Math-R1-CoT-SFT/trainer_state.json" "b/Qwen3.4B-Math-R1-CoT-SFT/trainer_state.json"
--- "a/Qwen3.4B-Math-R1-CoT-SFT/trainer_state.json"
+++ "b/Qwen3.4B-Math-R1-CoT-SFT/trainer_state.json"
@@ -2,7 +2,7 @@
   "best_global_step": null,
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 4.0,
+  "epoch": 3.9988184324537217,
   "eval_steps": 500,
   "global_step": 2539,
   "is_hyper_param_search": false,
@@ -11,3568 +11,3568 @@
   "log_history": [
     {
       "epoch": 0.007877116975187082,
-      "grad_norm": 3.437255382537842,
+      "grad_norm": 3399.454833984375,
       "learning_rate": 1.5748031496062992e-06,
-      "loss": 0.672,
+      "loss": 131.9922,
       "step": 5
     },
     {
       "epoch": 0.015754233950374164,
-      "grad_norm": 1.9117826223373413,
+      "grad_norm": 1452.986572265625,
       "learning_rate": 3.5433070866141735e-06,
-      "loss": 0.6371,
+      "loss": 88.8359,
       "step": 10
     },
     {
       "epoch": 0.023631350925561245,
-      "grad_norm": 1.74402916431427,
+      "grad_norm": 534.6153564453125,
       "learning_rate": 5.511811023622048e-06,
-      "loss": 0.6065,
+      "loss": 48.5337,
       "step": 15
     },
     {
       "epoch": 0.03150846790074833,
-      "grad_norm": 0.9878258109092712,
+      "grad_norm": 281.75048828125,
       "learning_rate": 7.4803149606299226e-06,
-      "loss": 0.5662,
+      "loss": 35.3826,
       "step": 20
     },
     {
       "epoch": 0.03938558487593541,
-      "grad_norm": 0.8288599252700806,
+      "grad_norm": 87.65968322753906,
       "learning_rate": 9.448818897637795e-06,
-      "loss": 0.5486,
+      "loss": 27.0295,
       "step": 25
     },
     {
       "epoch": 0.04726270185112249,
-      "grad_norm": 0.6100791096687317,
+      "grad_norm": 65.45500183105469,
       "learning_rate": 1.141732283464567e-05,
-      "loss": 0.5297,
+      "loss": 23.913,
       "step": 30
     },
     {
       "epoch": 0.05513981882630957,
-      "grad_norm": 0.4823300540447235,
+      "grad_norm": 65.12550354003906,
       "learning_rate": 1.3385826771653545e-05,
-      "loss": 0.5211,
+      "loss": 20.433,
       "step": 35
     },
     {
       "epoch": 0.06301693580149666,
-      "grad_norm": 0.4592387080192566,
+      "grad_norm": 20.998428344726562,
       "learning_rate": 1.535433070866142e-05,
-      "loss": 0.5176,
+      "loss": 18.3731,
       "step": 40
     },
     {
       "epoch": 0.07089405277668373,
-      "grad_norm": 0.3606854975223541,
+      "grad_norm": 29.879304885864258,
       "learning_rate": 1.732283464566929e-05,
-      "loss": 0.5071,
+      "loss": 16.9914,
       "step": 45
     },
     {
       "epoch": 0.07877116975187082,
-      "grad_norm": 0.4212338328361511,
+      "grad_norm": 22.706512451171875,
       "learning_rate": 1.9291338582677166e-05,
-      "loss": 0.4941,
+      "loss": 15.2466,
       "step": 50
     },
     {
       "epoch": 0.08664828672705789,
-      "grad_norm": 0.33868658542633057,
+      "grad_norm": 18.713308334350586,
       "learning_rate": 2.125984251968504e-05,
-      "loss": 0.4888,
+      "loss": 13.814,
       "step": 55
     },
     {
       "epoch": 0.09452540370224498,
-      "grad_norm": 0.3226923942565918,
+      "grad_norm": 29.742462158203125,
       "learning_rate": 2.3228346456692916e-05,
-      "loss": 0.4819,
+      "loss": 14.2114,
       "step": 60
     },
     {
       "epoch": 0.10240252067743207,
-      "grad_norm": 0.43986427783966064,
+      "grad_norm": 29.92841339111328,
       "learning_rate": 2.5196850393700788e-05,
-      "loss": 0.4881,
+      "loss": 14.0749,
       "step": 65
     },
     {
       "epoch": 0.11027963765261914,
-      "grad_norm": 0.40936681628227234,
+      "grad_norm": 17.866037368774414,
       "learning_rate": 2.7165354330708666e-05,
-      "loss": 0.4871,
+      "loss": 13.3448,
       "step": 70
     },
     {
       "epoch": 0.11815675462780623,
-      "grad_norm": 0.3969690799713135,
+      "grad_norm": 26.47714614868164,
       "learning_rate": 2.9133858267716534e-05,
-      "loss": 0.4721,
+      "loss": 13.2347,
       "step": 75
     },
     {
       "epoch": 0.1260338716029933,
-      "grad_norm": 0.4242527186870575,
+      "grad_norm": 22.94607162475586,
       "learning_rate": 3.110236220472441e-05,
-      "loss": 0.4841,
+      "loss": 12.4648,
       "step": 80
     },
     {
       "epoch": 0.13391098857818037,
-      "grad_norm": 0.48484691977500916,
+      "grad_norm": 14.325417518615723,
       "learning_rate": 3.3070866141732284e-05,
-      "loss": 0.4771,
+      "loss": 12.8074,
       "step": 85
     },
     {
       "epoch": 0.14178810555336746,
-      "grad_norm": 0.5061110854148865,
+      "grad_norm": 14.305035591125488,
       "learning_rate": 3.5039370078740156e-05,
-      "loss": 0.4718,
+      "loss": 12.1508,
       "step": 90
     },
     {
       "epoch": 0.14966522252855455,
-      "grad_norm": 0.4648071527481079,
+      "grad_norm": 14.955709457397461,
       "learning_rate": 3.7007874015748034e-05,
-      "loss": 0.474,
+      "loss": 12.5528,
       "step": 95
     },
     {
       "epoch": 0.15754233950374164,
-      "grad_norm": 0.5444895625114441,
+      "grad_norm": 14.336165428161621,
       "learning_rate": 3.8976377952755905e-05,
-      "loss": 0.4761,
+      "loss": 11.6424,
       "step": 100
     },
     {
       "epoch": 0.16541945647892872,
-      "grad_norm": 0.5805575251579285,
+      "grad_norm": 14.17673110961914,
       "learning_rate": 4.0944881889763784e-05,
-      "loss": 0.4668,
+      "loss": 12.1279,
       "step": 105
     },
     {
       "epoch": 0.17329657345411578,
-      "grad_norm": 0.40570878982543945,
+      "grad_norm": 12.128265380859375,
       "learning_rate": 4.2913385826771655e-05,
-      "loss": 0.467,
+      "loss": 11.7178,
       "step": 110
     },
     {
       "epoch": 0.18117369042930287,
-      "grad_norm": 0.44745564460754395,
+      "grad_norm": 13.574759483337402,
       "learning_rate": 4.488188976377953e-05,
-      "loss": 0.4801,
+      "loss": 11.9381,
       "step": 115
     },
     {
       "epoch": 0.18905080740448996,
-      "grad_norm": 0.5328453183174133,
+      "grad_norm": 11.064770698547363,
       "learning_rate": 4.6850393700787405e-05,
-      "loss": 0.4663,
+      "loss": 11.1822,
       "step": 120
     },
     {
       "epoch": 0.19692792437967704,
-      "grad_norm": 0.3628937602043152,
+      "grad_norm": 13.90650463104248,
       "learning_rate": 4.881889763779528e-05,
-      "loss": 0.4644,
+      "loss": 12.0308,
       "step": 125
     },
     {
       "epoch": 0.20480504135486413,
-      "grad_norm": 0.44744938611984253,
+      "grad_norm": 12.934874534606934,
       "learning_rate": 4.999991524704281e-05,
-      "loss": 0.4695,
+      "loss": 11.8371,
       "step": 130
     },
     {
       "epoch": 0.2126821583300512,
-      "grad_norm": 0.6168819069862366,
+      "grad_norm": 11.595288276672363,
       "learning_rate": 4.999896178287378e-05,
-      "loss": 0.4636,
+      "loss": 11.3976,
       "step": 135
     },
     {
       "epoch": 0.22055927530523828,
-      "grad_norm": 0.4544437527656555,
+      "grad_norm": 10.271947860717773,
       "learning_rate": 4.999694895387824e-05,
-      "loss": 0.4728,
+      "loss": 11.6096,
       "step": 140
     },
     {
       "epoch": 0.22843639228042537,
-      "grad_norm": 0.3462134003639221,
+      "grad_norm": 12.979105949401855,
       "learning_rate": 4.999387684535255e-05,
-      "loss": 0.4704,
+      "loss": 11.5108,
       "step": 145
     },
     {
       "epoch": 0.23631350925561245,
-      "grad_norm": 0.40045836567878723,
+      "grad_norm": 12.064156532287598,
       "learning_rate": 4.998974558748146e-05,
-      "loss": 0.4688,
+      "loss": 11.7928,
       "step": 150
     },
     {
       "epoch": 0.24419062623079954,
-      "grad_norm": 0.46955132484436035,
+      "grad_norm": 12.464822769165039,
       "learning_rate": 4.9984555355332604e-05,
-      "loss": 0.467,
+      "loss": 11.2361,
       "step": 155
     },
     {
       "epoch": 0.2520677432059866,
-      "grad_norm": 0.4244947135448456,
+      "grad_norm": 11.160205841064453,
       "learning_rate": 4.997830636884911e-05,
-      "loss": 0.4656,
+      "loss": 11.118,
       "step": 160
     },
     {
       "epoch": 0.2599448601811737,
-      "grad_norm": 0.41033992171287537,
+      "grad_norm": 10.261472702026367,
       "learning_rate": 4.997099889284021e-05,
-      "loss": 0.4679,
+      "loss": 11.6838,
       "step": 165
     },
     {
       "epoch": 0.26782197715636075,
-      "grad_norm": 0.39735573530197144,
+      "grad_norm": 9.86904525756836,
       "learning_rate": 4.996263323697011e-05,
-      "loss": 0.4686,
+      "loss": 11.4614,
       "step": 170
     },
     {
       "epoch": 0.27569909413154786,
-      "grad_norm": 0.3747316300868988,
+      "grad_norm": 10.730037689208984,
       "learning_rate": 4.9953209755744784e-05,
-      "loss": 0.4701,
+      "loss": 11.69,
       "step": 175
     },
     {
       "epoch": 0.2835762111067349,
-      "grad_norm": 0.39879927039146423,
+      "grad_norm": 10.000442504882812,
       "learning_rate": 4.9942728848496995e-05,
-      "loss": 0.4629,
+      "loss": 11.3298,
       "step": 180
     },
     {
       "epoch": 0.29145332808192204,
-      "grad_norm": 0.3395926356315613,
+      "grad_norm": 9.375894546508789,
       "learning_rate": 4.993119095936937e-05,
-      "loss": 0.4676,
+      "loss": 11.5506,
       "step": 185
     },
     {
       "epoch": 0.2993304450571091,
-      "grad_norm": 0.35948389768600464,
+      "grad_norm": 11.61759090423584,
       "learning_rate": 4.991859657729557e-05,
-      "loss": 0.4468,
+      "loss": 11.0027,
       "step": 190
     },
     {
       "epoch": 0.30720756203229616,
-      "grad_norm": 0.33689770102500916,
+      "grad_norm": 8.72737979888916,
       "learning_rate": 4.9904946235979574e-05,
-      "loss": 0.4624,
+      "loss": 11.1883,
       "step": 195
     },
     {
       "epoch": 0.31508467900748327,
-      "grad_norm": 0.38664934039115906,
+      "grad_norm": 8.50748348236084,
       "learning_rate": 4.9890240513873064e-05,
-      "loss": 0.4611,
+      "loss": 11.0641,
       "step": 200
     },
     {
       "epoch": 0.32296179598267033,
-      "grad_norm": 0.3196653723716736,
+      "grad_norm": 8.511228561401367,
       "learning_rate": 4.9874480034150915e-05,
-      "loss": 0.4604,
+      "loss": 11.1942,
       "step": 205
     },
     {
       "epoch": 0.33083891295785745,
-      "grad_norm": 0.38587889075279236,
+      "grad_norm": 9.39171028137207,
       "learning_rate": 4.985766546468477e-05,
-      "loss": 0.4597,
+      "loss": 10.6584,
       "step": 210
     },
     {
       "epoch": 0.3387160299330445,
-      "grad_norm": 0.3755643963813782,
+      "grad_norm": 7.405202865600586,
       "learning_rate": 4.983979751801475e-05,
-      "loss": 0.4638,
+      "loss": 11.1103,
       "step": 215
     },
     {
       "epoch": 0.34659314690823156,
-      "grad_norm": 0.3800589442253113,
+      "grad_norm": 7.648876667022705,
       "learning_rate": 4.982087695131928e-05,
-      "loss": 0.4579,
+      "loss": 11.0276,
       "step": 220
     },
     {
       "epoch": 0.3544702638834187,
-      "grad_norm": 0.3112618327140808,
+      "grad_norm": 7.407284259796143,
       "learning_rate": 4.9800904566382984e-05,
-      "loss": 0.445,
+      "loss": 9.9736,
       "step": 225
     },
     {
       "epoch": 0.36234738085860574,
-      "grad_norm": 0.33558526635169983,
+      "grad_norm": 7.161754608154297,
       "learning_rate": 4.9779881209562675e-05,
-      "loss": 0.4552,
+      "loss": 10.8461,
       "step": 230
     },
     {
       "epoch": 0.37022449783379285,
-      "grad_norm": 0.35156235098838806,
+      "grad_norm": 9.168039321899414,
       "learning_rate": 4.975780777175154e-05,
-      "loss": 0.4642,
+      "loss": 10.7467,
       "step": 235
     },
     {
       "epoch": 0.3781016148089799,
-      "grad_norm": 0.34751108288764954,
+      "grad_norm": 8.102660179138184,
       "learning_rate": 4.9734685188341376e-05,
-      "loss": 0.454,
+      "loss": 10.8515,
       "step": 240
     },
     {
       "epoch": 0.385978731784167,
-      "grad_norm": 0.3409808874130249,
+      "grad_norm": 8.496058464050293,
       "learning_rate": 4.971051443918294e-05,
-      "loss": 0.4605,
+      "loss": 10.1448,
       "step": 245
     },
     {
       "epoch": 0.3938558487593541,
-      "grad_norm": 0.3360814154148102,
+      "grad_norm": 8.0912446975708,
       "learning_rate": 4.9685296548544424e-05,
-      "loss": 0.4642,
+      "loss": 10.299,
       "step": 250
     },
     {
       "epoch": 0.40173296573454115,
-      "grad_norm": 0.3765697777271271,
+      "grad_norm": 8.432981491088867,
       "learning_rate": 4.965903258506806e-05,
-      "loss": 0.4516,
+      "loss": 10.3602,
       "step": 255
     },
     {
       "epoch": 0.40961008270972826,
-      "grad_norm": 0.3913894295692444,
+      "grad_norm": 7.998964786529541,
       "learning_rate": 4.9631723661724836e-05,
-      "loss": 0.4598,
+      "loss": 10.3991,
       "step": 260
     },
     {
       "epoch": 0.4174871996849153,
-      "grad_norm": 0.3312707841396332,
+      "grad_norm": 7.924742698669434,
       "learning_rate": 4.9603370935767324e-05,
-      "loss": 0.4554,
+      "loss": 10.7044,
       "step": 265
     },
     {
       "epoch": 0.4253643166601024,
-      "grad_norm": 0.2816666066646576,
+      "grad_norm": 7.88558292388916,
       "learning_rate": 4.957397560868064e-05,
-      "loss": 0.4565,
+      "loss": 10.2436,
       "step": 270
     },
     {
       "epoch": 0.4332414336352895,
-      "grad_norm": 0.3484485447406769,
+      "grad_norm": 7.450520038604736,
       "learning_rate": 4.954353892613154e-05,
-      "loss": 0.4589,
+      "loss": 10.2202,
       "step": 275
     },
     {
       "epoch": 0.44111855061047656,
-      "grad_norm": 0.3825587332248688,
+      "grad_norm": 8.56113052368164,
       "learning_rate": 4.951206217791564e-05,
-      "loss": 0.4598,
+      "loss": 10.4133,
       "step": 280
     },
     {
       "epoch": 0.44899566758566367,
-      "grad_norm": 0.3513055443763733,
+      "grad_norm": 6.7272491455078125,
       "learning_rate": 4.94795466979027e-05,
-      "loss": 0.4495,
+      "loss": 10.1886,
       "step": 285
     },
     {
       "epoch": 0.45687278456085073,
-      "grad_norm": 0.38703492283821106,
+      "grad_norm": 8.858597755432129,
       "learning_rate": 4.94459938639802e-05,
-      "loss": 0.442,
+      "loss": 10.2836,
       "step": 290
     },
     {
       "epoch": 0.4647499015360378,
-      "grad_norm": 0.28819456696510315,
+      "grad_norm": 7.176908493041992,
       "learning_rate": 4.941140509799487e-05,
-      "loss": 0.4462,
+      "loss": 10.657,
       "step": 295
     },
     {
       "epoch": 0.4726270185112249,
-      "grad_norm": 0.2926427125930786,
+      "grad_norm": 6.454217433929443,
       "learning_rate": 4.937578186569244e-05,
-      "loss": 0.4513,
+      "loss": 9.67,
       "step": 300
     },
     {
       "epoch": 0.48050413548641197,
-      "grad_norm": 0.28402814269065857,
+      "grad_norm": 8.159943580627441,
       "learning_rate": 4.933912567665561e-05,
-      "loss": 0.4553,
+      "loss": 10.7037,
       "step": 305
     },
     {
       "epoch": 0.4883812524615991,
-      "grad_norm": 0.31271660327911377,
+      "grad_norm": 7.20767879486084,
       "learning_rate": 4.930143808423995e-05,
-      "loss": 0.4389,
+      "loss": 9.7757,
       "step": 310
     },
     {
       "epoch": 0.49625836943678614,
-      "grad_norm": 0.39874470233917236,
+      "grad_norm": 8.055448532104492,
       "learning_rate": 4.926272068550819e-05,
-      "loss": 0.4583,
+      "loss": 10.9187,
       "step": 315
     },
     {
       "epoch": 0.5041354864119733,
-      "grad_norm": 0.3334454298019409,
+      "grad_norm": 7.440671443939209,
       "learning_rate": 4.922297512116248e-05,
-      "loss": 0.4603,
+      "loss": 10.756,
       "step": 320
     },
     {
       "epoch": 0.5120126033871603,
-      "grad_norm": 0.28501057624816895,
+      "grad_norm": 7.851231098175049,
       "learning_rate": 4.9182203075474884e-05,
-      "loss": 0.4539,
+      "loss": 9.9799,
       "step": 325
     },
     {
       "epoch": 0.5198897203623474,
-      "grad_norm": 0.3027922809123993,
+      "grad_norm": 7.782782077789307,
       "learning_rate": 4.914040627621599e-05,
-      "loss": 0.4573,
+      "loss": 10.0362,
       "step": 330
     },
     {
       "epoch": 0.5277668373375345,
-      "grad_norm": 0.31392157077789307,
+      "grad_norm": 7.82265043258667,
       "learning_rate": 4.909758649458172e-05,
-      "loss": 0.4547,
+      "loss": 10.2922,
       "step": 335
     },
     {
       "epoch": 0.5356439543127215,
-      "grad_norm": 0.30944597721099854,
+      "grad_norm": 7.081122875213623,
       "learning_rate": 4.905374554511825e-05,
-      "loss": 0.443,
+      "loss": 9.93,
       "step": 340
     },
     {
       "epoch": 0.5435210712879086,
-      "grad_norm": 0.2820224463939667,
+      "grad_norm": 7.180002212524414,
       "learning_rate": 4.9008885285645134e-05,
-      "loss": 0.4535,
+      "loss": 10.1384,
       "step": 345
     },
     {
       "epoch": 0.5513981882630957,
-      "grad_norm": 0.2703908681869507,
+      "grad_norm": 8.38278579711914,
       "learning_rate": 4.8963007617176535e-05,
-      "loss": 0.4456,
+      "loss": 10.1262,
       "step": 350
     },
     {
       "epoch": 0.5592753052382828,
-      "grad_norm": 0.2866289019584656,
+      "grad_norm": 7.965811729431152,
       "learning_rate": 4.8916114483840746e-05,
-      "loss": 0.4468,
+      "loss": 10.2287,
       "step": 355
     },
     {
       "epoch": 0.5671524222134698,
-      "grad_norm": 0.30911338329315186,
+      "grad_norm": 6.411648750305176,
       "learning_rate": 4.886820787279773e-05,
-      "loss": 0.4419,
+      "loss": 9.9416,
       "step": 360
     },
     {
       "epoch": 0.575029539188657,
-      "grad_norm": 0.2732813060283661,
+      "grad_norm": 6.27194356918335,
       "learning_rate": 4.8819289814154935e-05,
-      "loss": 0.4333,
+      "loss": 9.9728,
       "step": 365
     },
     {
       "epoch": 0.5829066561638441,
-      "grad_norm": 0.32516050338745117,
+      "grad_norm": 5.867300987243652,
       "learning_rate": 4.876936238088127e-05,
-      "loss": 0.4466,
+      "loss": 10.0274,
       "step": 370
     },
     {
       "epoch": 0.5907837731390311,
-      "grad_norm": 0.2863224446773529,
+      "grad_norm": 6.166661739349365,
       "learning_rate": 4.871842768871928e-05,
-      "loss": 0.4408,
+      "loss": 10.2888,
       "step": 375
     },
     {
       "epoch": 0.5986608901142182,
-      "grad_norm": 0.27347731590270996,
+      "grad_norm": 6.584941387176514,
       "learning_rate": 4.866648789609545e-05,
-      "loss": 0.4422,
+      "loss": 9.8767,
       "step": 380
     },
     {
       "epoch": 0.6065380070894053,
-      "grad_norm": 0.27762892842292786,
+      "grad_norm": 6.197508811950684,
       "learning_rate": 4.8613545204028764e-05,
-      "loss": 0.4425,
+      "loss": 10.2098,
       "step": 385
     },
     {
       "epoch": 0.6144151240645923,
-      "grad_norm": 0.3145371675491333,
+      "grad_norm": 7.481927394866943,
       "learning_rate": 4.855960185603742e-05,
-      "loss": 0.4451,
+      "loss": 9.8902,
       "step": 390
     },
     {
       "epoch": 0.6222922410397794,
-      "grad_norm": 0.2932167649269104,
+      "grad_norm": 7.165607452392578,
       "learning_rate": 4.850466013804377e-05,
-      "loss": 0.4434,
+      "loss": 10.2372,
       "step": 395
     },
     {
       "epoch": 0.6301693580149665,
-      "grad_norm": 0.2880720794200897,
+      "grad_norm": 6.94336462020874,
       "learning_rate": 4.844872237827745e-05,
-      "loss": 0.4475,
+      "loss": 9.9813,
       "step": 400
     },
     {
       "epoch": 0.6380464749901537,
-      "grad_norm": 0.3783859610557556,
+      "grad_norm": 6.348086833953857,
       "learning_rate": 4.8391790947176696e-05,
-      "loss": 0.446,
+      "loss": 10.0216,
       "step": 405
     },
     {
       "epoch": 0.6459235919653407,
-      "grad_norm": 0.31301021575927734,
+      "grad_norm": 6.404892444610596,
       "learning_rate": 4.833386825728794e-05,
-      "loss": 0.4462,
+      "loss": 10.108,
       "step": 410
     },
     {
       "epoch": 0.6538007089405278,
-      "grad_norm": 0.28316691517829895,
+      "grad_norm": 7.904120445251465,
       "learning_rate": 4.8274956763163506e-05,
-      "loss": 0.4485,
+      "loss": 10.0952,
       "step": 415
     },
     {
       "epoch": 0.6616778259157149,
-      "grad_norm": 0.3179584741592407,
+      "grad_norm": 6.545925617218018,
       "learning_rate": 4.821505896125769e-05,
-      "loss": 0.4372,
+      "loss": 9.8825,
       "step": 420
     },
     {
       "epoch": 0.6695549428909019,
-      "grad_norm": 0.3088538646697998,
+      "grad_norm": 6.014336109161377,
       "learning_rate": 4.815417738982085e-05,
-      "loss": 0.4377,
+      "loss": 10.0608,
       "step": 425
     },
     {
       "epoch": 0.677432059866089,
-      "grad_norm": 0.31807076930999756,
+      "grad_norm": 7.037391185760498,
       "learning_rate": 4.809231462879196e-05,
-      "loss": 0.4373,
+      "loss": 9.6694,
       "step": 430
     },
     {
       "epoch": 0.6853091768412761,
-      "grad_norm": 0.2823260724544525,
+      "grad_norm": 6.266141414642334,
       "learning_rate": 4.802947329968921e-05,
-      "loss": 0.4375,
+      "loss": 9.5179,
       "step": 435
     },
     {
       "epoch": 0.6931862938164631,
-      "grad_norm": 0.30851051211357117,
+      "grad_norm": 6.4194111824035645,
       "learning_rate": 4.7965656065498944e-05,
-      "loss": 0.4498,
+      "loss": 9.9608,
       "step": 440
     },
     {
       "epoch": 0.7010634107916502,
-      "grad_norm": 0.28332722187042236,
+      "grad_norm": 6.709506511688232,
       "learning_rate": 4.7900865630562786e-05,
-      "loss": 0.4342,
+      "loss": 9.8622,
       "step": 445
     },
     {
       "epoch": 0.7089405277668374,
-      "grad_norm": 0.26585566997528076,
+      "grad_norm": 5.767233848571777,
       "learning_rate": 4.783510474046308e-05,
-      "loss": 0.4402,
+      "loss": 9.3161,
       "step": 450
     },
     {
       "epoch": 0.7168176447420245,
-      "grad_norm": 0.26203709840774536,
+      "grad_norm": 5.416118621826172,
       "learning_rate": 4.77683761819065e-05,
-      "loss": 0.4393,
+      "loss": 9.719,
       "step": 455
     },
     {
       "epoch": 0.7246947617172115,
-      "grad_norm": 0.27073633670806885,
+      "grad_norm": 6.180813789367676,
       "learning_rate": 4.7700682782606015e-05,
-      "loss": 0.4378,
+      "loss": 9.9941,
       "step": 460
     },
     {
       "epoch": 0.7325718786923986,
-      "grad_norm": 0.2871417999267578,
+      "grad_norm": 5.783968448638916,
       "learning_rate": 4.763202741116098e-05,
-      "loss": 0.4362,
+      "loss": 10.4704,
       "step": 465
     },
     {
       "epoch": 0.7404489956675857,
-      "grad_norm": 0.27669113874435425,
+      "grad_norm": 5.99013614654541,
       "learning_rate": 4.756241297693566e-05,
-      "loss": 0.4269,
+      "loss": 9.3854,
       "step": 470
     },
     {
       "epoch": 0.7483261126427727,
-      "grad_norm": 0.26247602701187134,
+      "grad_norm": 6.445564270019531,
       "learning_rate": 4.749184242993587e-05,
-      "loss": 0.442,
+      "loss": 9.835,
       "step": 475
     },
     {
       "epoch": 0.7562032296179598,
-      "grad_norm": 0.2640887498855591,
+      "grad_norm": 5.977458477020264,
       "learning_rate": 4.742031876068402e-05,
-      "loss": 0.4435,
+      "loss": 10.2198,
       "step": 480
     },
     {
       "epoch": 0.7640803465931469,
-      "grad_norm": 0.2536977231502533,
+      "grad_norm": 7.140058517456055,
       "learning_rate": 4.734784500009237e-05,
-      "loss": 0.4317,
+      "loss": 9.456,
       "step": 485
     },
     {
       "epoch": 0.771957463568334,
-      "grad_norm": 0.2515846788883209,
+      "grad_norm": 6.452182769775391,
       "learning_rate": 4.727442421933456e-05,
-      "loss": 0.4326,
+      "loss": 9.904,
       "step": 490
     },
     {
       "epoch": 0.7798345805435211,
-      "grad_norm": 0.2645900547504425,
+      "grad_norm": 5.870535373687744,
       "learning_rate": 4.720005952971551e-05,
-      "loss": 0.4336,
+      "loss": 9.9197,
       "step": 495
     },
     {
       "epoch": 0.7877116975187082,
-      "grad_norm": 0.2777593433856964,
+      "grad_norm": 6.265655040740967,
       "learning_rate": 4.7124754082539546e-05,
-      "loss": 0.4402,
+      "loss": 9.8827,
       "step": 500
     },
     {
       "epoch": 0.7955888144938952,
-      "grad_norm": 0.29428908228874207,
+      "grad_norm": 6.607460975646973,
       "learning_rate": 4.704851106897689e-05,
-      "loss": 0.4292,
+      "loss": 9.6845,
       "step": 505
     },
     {
       "epoch": 0.8034659314690823,
-      "grad_norm": 0.28475651144981384,
+      "grad_norm": 7.221487045288086,
       "learning_rate": 4.6971333719928354e-05,
-      "loss": 0.4307,
+      "loss": 9.9776,
       "step": 510
     },
     {
       "epoch": 0.8113430484442694,
-      "grad_norm": 0.2999843955039978,
+      "grad_norm": 6.316247463226318,
       "learning_rate": 4.6893225305888556e-05,
-      "loss": 0.4408,
+      "loss": 9.2224,
       "step": 515
     },
     {
       "epoch": 0.8192201654194565,
-      "grad_norm": 0.28412893414497375,
+      "grad_norm": 5.804502010345459,
       "learning_rate": 4.68141891368072e-05,
-      "loss": 0.4375,
+      "loss": 9.8942,
       "step": 520
     },
     {
       "epoch": 0.8270972823946435,
-      "grad_norm": 0.24423280358314514,
+      "grad_norm": 6.048508644104004,
       "learning_rate": 4.6734228561948864e-05,
-      "loss": 0.4413,
+      "loss": 9.9273,
       "step": 525
     },
     {
       "epoch": 0.8349743993698306,
-      "grad_norm": 0.3219039738178253,
+      "grad_norm": 5.967385768890381,
       "learning_rate": 4.6653346969751096e-05,
-      "loss": 0.4402,
+      "loss": 9.8039,
       "step": 530
     },
     {
       "epoch": 0.8428515163450178,
-      "grad_norm": 0.2778218984603882,
+      "grad_norm": 5.828855514526367,
       "learning_rate": 4.65715477876808e-05,
-      "loss": 0.4299,
+      "loss": 9.5876,
       "step": 535
     },
     {
       "epoch": 0.8507286333202048,
-      "grad_norm": 0.2895450294017792,
+      "grad_norm": 6.0214314460754395,
       "learning_rate": 4.6488834482088965e-05,
-      "loss": 0.4351,
+      "loss": 9.478,
       "step": 540
     },
     {
       "epoch": 0.8586057502953919,
-      "grad_norm": 0.28678175806999207,
+      "grad_norm": 6.222189903259277,
       "learning_rate": 4.640521055806383e-05,
-      "loss": 0.4467,
+      "loss": 9.6082,
       "step": 545
     },
     {
       "epoch": 0.866482867270579,
-      "grad_norm": 0.28237128257751465,
+      "grad_norm": 6.3668437004089355,
       "learning_rate": 4.632067955928231e-05,
-      "loss": 0.4256,
+      "loss": 9.515,
       "step": 550
     },
     {
       "epoch": 0.874359984245766,
-      "grad_norm": 0.2734551727771759,
+      "grad_norm": 6.134618282318115,
       "learning_rate": 4.6235245067859825e-05,
-      "loss": 0.4346,
+      "loss": 9.7517,
       "step": 555
     },
     {
       "epoch": 0.8822371012209531,
-      "grad_norm": 0.25271713733673096,
+      "grad_norm": 6.289186000823975,
       "learning_rate": 4.614891070419855e-05,
-      "loss": 0.43,
+      "loss": 9.7513,
       "step": 560
     },
     {
       "epoch": 0.8901142181961402,
-      "grad_norm": 0.2673031687736511,
+      "grad_norm": 6.570099830627441,
       "learning_rate": 4.606168012683394e-05,
-      "loss": 0.4334,
+      "loss": 9.5023,
       "step": 565
     },
     {
       "epoch": 0.8979913351713273,
-      "grad_norm": 0.24039021134376526,
+      "grad_norm": 6.501761436462402,
       "learning_rate": 4.5973557032279726e-05,
-      "loss": 0.4354,
+      "loss": 9.7911,
       "step": 570
     },
     {
       "epoch": 0.9058684521465143,
-      "grad_norm": 0.2770460546016693,
+      "grad_norm": 6.731903553009033,
       "learning_rate": 4.588454515487125e-05,
-      "loss": 0.4277,
+      "loss": 9.2721,
       "step": 575
     },
     {
       "epoch": 0.9137455691217015,
-      "grad_norm": 0.2645154893398285,
+      "grad_norm": 6.2406086921691895,
       "learning_rate": 4.5794648266607254e-05,
-      "loss": 0.4242,
+      "loss": 9.2431,
       "step": 580
     },
     {
       "epoch": 0.9216226860968886,
-      "grad_norm": 0.2790268659591675,
+      "grad_norm": 8.897172927856445,
       "learning_rate": 4.570387017699e-05,
-      "loss": 0.4276,
+      "loss": 9.4321,
       "step": 585
     },
     {
       "epoch": 0.9294998030720756,
-      "grad_norm": 0.2759014070034027,
+      "grad_norm": 6.036581039428711,
       "learning_rate": 4.561221473286384e-05,
-      "loss": 0.4365,
+      "loss": 10.3099,
       "step": 590
     },
     {
       "epoch": 0.9373769200472627,
-      "grad_norm": 0.27472007274627686,
+      "grad_norm": 5.789550304412842,
       "learning_rate": 4.551968581825222e-05,
-      "loss": 0.4318,
+      "loss": 10.217,
       "step": 595
     },
     {
       "epoch": 0.9452540370224498,
-      "grad_norm": 0.27848538756370544,
+      "grad_norm": 6.0984110832214355,
       "learning_rate": 4.5426287354193095e-05,
-      "loss": 0.4302,
+      "loss": 9.667,
       "step": 600
     },
     {
       "epoch": 0.9531311539976368,
-      "grad_norm": 0.25084635615348816,
+      "grad_norm": 6.172824382781982,
       "learning_rate": 4.533202329857271e-05,
-      "loss": 0.4398,
+      "loss": 9.7751,
       "step": 605
     },
     {
       "epoch": 0.9610082709728239,
-      "grad_norm": 0.2645019590854645,
+      "grad_norm": 6.124137878417969,
       "learning_rate": 4.523689764595797e-05,
-      "loss": 0.4352,
+      "loss": 9.4172,
       "step": 610
     },
     {
       "epoch": 0.968885387948011,
-      "grad_norm": 0.2712864875793457,
+      "grad_norm": 5.909232139587402,
       "learning_rate": 4.5140914427427094e-05,
-      "loss": 0.4266,
+      "loss": 9.2955,
       "step": 615
     },
     {
       "epoch": 0.9767625049231982,
-      "grad_norm": 0.2347964495420456,
+      "grad_norm": 5.878952503204346,
       "learning_rate": 4.504407771039883e-05,
-      "loss": 0.4249,
+      "loss": 9.4788,
       "step": 620
     },
     {
       "epoch": 0.9846396218983852,
-      "grad_norm": 0.23019365966320038,
+      "grad_norm": 7.975442886352539,
       "learning_rate": 4.494639159846006e-05,
-      "loss": 0.4292,
+      "loss": 9.15,
       "step": 625
     },
     {
       "epoch": 0.9925167388735723,
-      "grad_norm": 0.22928465902805328,
+      "grad_norm": 6.191237449645996,
       "learning_rate": 4.484786023119194e-05,
-      "loss": 0.4309,
+      "loss": 10.031,
       "step": 630
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.253717303276062,
+      "grad_norm": 5.575296401977539,
       "learning_rate": 4.474848778399446e-05,
-      "loss": 0.4261,
+      "loss": 9.1086,
       "step": 635
     },
     {
       "epoch": 1.007877116975187,
-      "grad_norm": 0.35631492733955383,
+      "grad_norm": 5.996377944946289,
       "learning_rate": 4.464827846790952e-05,
-      "loss": 0.388,
+      "loss": 9.3717,
       "step": 640
     },
     {
       "epoch": 1.0157542339503742,
-      "grad_norm": 0.3446408212184906,
+      "grad_norm": 5.924043655395508,
       "learning_rate": 4.454723652944246e-05,
-      "loss": 0.3889,
+      "loss": 9.4512,
       "step": 645
     },
     {
       "epoch": 1.0236313509255612,
-      "grad_norm": 0.29462265968322754,
+      "grad_norm": 6.695152282714844,
       "learning_rate": 4.4445366250382106e-05,
-      "loss": 0.3927,
+      "loss": 8.4621,
       "step": 650
     },
     {
       "epoch": 1.0315084679007482,
-      "grad_norm": 0.26057687401771545,
+      "grad_norm": 7.838866233825684,
       "learning_rate": 4.434267194761936e-05,
-      "loss": 0.3883,
+      "loss": 9.3423,
       "step": 655
     },
     {
       "epoch": 1.0393855848759355,
-      "grad_norm": 0.2740859091281891,
+      "grad_norm": 5.655851364135742,
       "learning_rate": 4.423915797296425e-05,
-      "loss": 0.394,
+      "loss": 9.2772,
       "step": 660
     },
     {
       "epoch": 1.0472627018511225,
-      "grad_norm": 0.2503489553928375,
+      "grad_norm": 6.232006072998047,
       "learning_rate": 4.413482871296148e-05,
-      "loss": 0.3769,
+      "loss": 9.1979,
       "step": 665
     },
     {
       "epoch": 1.0551398188263095,
-      "grad_norm": 0.2788833677768707,
+      "grad_norm": 5.975604057312012,
       "learning_rate": 4.402968858870459e-05,
-      "loss": 0.3866,
+      "loss": 9.4245,
       "step": 670
     },
     {
       "epoch": 1.0630169358014967,
-      "grad_norm": 0.24859720468521118,
+      "grad_norm": 8.961718559265137,
       "learning_rate": 4.39237420556486e-05,
-      "loss": 0.3856,
+      "loss": 9.2747,
       "step": 675
     },
     {
       "epoch": 1.0708940527766837,
-      "grad_norm": 0.26013121008872986,
+      "grad_norm": 6.198498249053955,
       "learning_rate": 4.381699360342116e-05,
-      "loss": 0.3929,
+      "loss": 9.1819,
       "step": 680
     },
     {
       "epoch": 1.078771169751871,
-      "grad_norm": 0.2571970522403717,
+      "grad_norm": 5.1888861656188965,
       "learning_rate": 4.370944775563239e-05,
-      "loss": 0.3664,
+      "loss": 8.2554,
       "step": 685
     },
     {
       "epoch": 1.086648286727058,
-      "grad_norm": 0.2991076111793518,
+      "grad_norm": 5.546014308929443,
       "learning_rate": 4.360110906968305e-05,
-      "loss": 0.379,
+      "loss": 9.2945,
       "step": 690
     },
     {
       "epoch": 1.094525403702245,
-      "grad_norm": 0.7817800045013428,
+      "grad_norm": 5.926428318023682,
       "learning_rate": 4.3491982136571554e-05,
-      "loss": 0.3871,
+      "loss": 9.384,
       "step": 695
     },
     {
       "epoch": 1.1024025206774322,
-      "grad_norm": 0.31027108430862427,
+      "grad_norm": 6.317972660064697,
       "learning_rate": 4.3382071580699314e-05,
-      "loss": 0.39,
+      "loss": 8.7085,
       "step": 700
     },
     {
       "epoch": 1.1102796376526192,
-      "grad_norm": 0.27426421642303467,
+      "grad_norm": 5.735546588897705,
       "learning_rate": 4.327138205967484e-05,
-      "loss": 0.3832,
+      "loss": 9.1479,
       "step": 705
     },
     {
       "epoch": 1.1181567546278062,
-      "grad_norm": 0.27434617280960083,
+      "grad_norm": 5.673744201660156,
       "learning_rate": 4.3159918264116334e-05,
-      "loss": 0.3937,
+      "loss": 9.5043,
       "step": 710
     },
     {
       "epoch": 1.1260338716029934,
-      "grad_norm": 0.2739649713039398,
+      "grad_norm": 5.20525598526001,
       "learning_rate": 4.3047684917452936e-05,
-      "loss": 0.3971,
+      "loss": 9.1163,
       "step": 715
     },
     {
       "epoch": 1.1339109885781804,
-      "grad_norm": 0.2901259958744049,
+      "grad_norm": 6.533771514892578,
       "learning_rate": 4.293468677572454e-05,
-      "loss": 0.387,
+      "loss": 8.9555,
       "step": 720
     },
     {
       "epoch": 1.1417881055533674,
-      "grad_norm": 0.30935901403427124,
+      "grad_norm": 5.279290676116943,
       "learning_rate": 4.282092862738029e-05,
-      "loss": 0.3787,
+      "loss": 8.6573,
       "step": 725
     },
     {
       "epoch": 1.1496652225285546,
-      "grad_norm": 0.27035894989967346,
+      "grad_norm": 5.301630020141602,
       "learning_rate": 4.270641529307562e-05,
-      "loss": 0.3727,
+      "loss": 8.9388,
       "step": 730
     },
     {
       "epoch": 1.1575423395037416,
-      "grad_norm": 0.262545645236969,
+      "grad_norm": 7.306978225708008,
       "learning_rate": 4.2591151625467986e-05,
-      "loss": 0.3866,
+      "loss": 9.1701,
       "step": 735
     },
     {
       "epoch": 1.1654194564789286,
-      "grad_norm": 0.2450433075428009,
+      "grad_norm": 5.707004070281982,
       "learning_rate": 4.2475142509011254e-05,
-      "loss": 0.3836,
+      "loss": 9.3798,
       "step": 740
     },
     {
       "epoch": 1.1732965734541159,
-      "grad_norm": 0.24818231165409088,
+      "grad_norm": 7.129739284515381,
       "learning_rate": 4.235839285974868e-05,
-      "loss": 0.3854,
+      "loss": 9.1854,
       "step": 745
     },
     {
       "epoch": 1.1811736904293029,
-      "grad_norm": 0.241587832570076,
+      "grad_norm": 9.110360145568848,
       "learning_rate": 4.2240907625104606e-05,
-      "loss": 0.3862,
+      "loss": 8.9261,
       "step": 750
     },
     {
       "epoch": 1.1890508074044899,
-      "grad_norm": 0.24221687018871307,
+      "grad_norm": 6.3835554122924805,
       "learning_rate": 4.2122691783674786e-05,
-      "loss": 0.382,
+      "loss": 8.9869,
       "step": 755
     },
     {
       "epoch": 1.196927924379677,
-      "grad_norm": 0.24578852951526642,
+      "grad_norm": 5.634960651397705,
       "learning_rate": 4.200375034501543e-05,
-      "loss": 0.3812,
+      "loss": 8.6859,
       "step": 760
     },
     {
       "epoch": 1.204805041354864,
-      "grad_norm": 0.27418452501296997,
+      "grad_norm": 5.98690938949585,
       "learning_rate": 4.188408834943093e-05,
-      "loss": 0.3894,
+      "loss": 8.9667,
       "step": 765
     },
     {
       "epoch": 1.212682158330051,
-      "grad_norm": 0.28048843145370483,
+      "grad_norm": 5.745028972625732,
       "learning_rate": 4.176371086776023e-05,
-      "loss": 0.3695,
+      "loss": 8.3641,
       "step": 770
     },
     {
       "epoch": 1.2205592753052383,
-      "grad_norm": 0.2632172703742981,
+      "grad_norm": 5.545426845550537,
       "learning_rate": 4.1642623001161976e-05,
-      "loss": 0.3846,
+      "loss": 9.2566,
       "step": 775
     },
     {
       "epoch": 1.2284363922804253,
-      "grad_norm": 0.25057244300842285,
+      "grad_norm": 6.151079177856445,
       "learning_rate": 4.152082988089834e-05,
-      "loss": 0.3892,
+      "loss": 9.1436,
       "step": 780
     },
     {
       "epoch": 1.2363135092556123,
-      "grad_norm": 0.28263312578201294,
+      "grad_norm": 5.903685092926025,
       "learning_rate": 4.139833666811756e-05,
-      "loss": 0.3832,
+      "loss": 9.0897,
       "step": 785
     },
     {
       "epoch": 1.2441906262307996,
-      "grad_norm": 0.26317495107650757,
+      "grad_norm": 5.456060886383057,
       "learning_rate": 4.127514855363524e-05,
-      "loss": 0.3831,
+      "loss": 9.2121,
       "step": 790
     },
     {
       "epoch": 1.2520677432059866,
-      "grad_norm": 0.28825777769088745,
+      "grad_norm": 5.680197238922119,
       "learning_rate": 4.1151270757714405e-05,
-      "loss": 0.3902,
+      "loss": 9.1453,
       "step": 795
     },
     {
       "epoch": 1.2599448601811738,
-      "grad_norm": 0.23847556114196777,
+      "grad_norm": 6.367788791656494,
       "learning_rate": 4.102670852984425e-05,
-      "loss": 0.3871,
+      "loss": 9.0336,
       "step": 800
     },
     {
       "epoch": 1.2678219771563608,
-      "grad_norm": 0.27880316972732544,
+      "grad_norm": 5.573480606079102,
       "learning_rate": 4.090146714851769e-05,
-      "loss": 0.3822,
+      "loss": 8.6367,
       "step": 805
     },
     {
       "epoch": 1.2756990941315478,
-      "grad_norm": 0.25444021821022034,
+      "grad_norm": 5.5287861824035645,
       "learning_rate": 4.077555192100772e-05,
-      "loss": 0.382,
+      "loss": 8.8872,
       "step": 810
     },
     {
       "epoch": 1.2835762111067348,
-      "grad_norm": 0.2789304554462433,
+      "grad_norm": 5.406901836395264,
       "learning_rate": 4.0648968183142424e-05,
-      "loss": 0.3783,
+      "loss": 8.9955,
       "step": 815
     },
     {
       "epoch": 1.291453328081922,
-      "grad_norm": 0.2782425880432129,
+      "grad_norm": 5.834075927734375,
       "learning_rate": 4.052172129907897e-05,
-      "loss": 0.3965,
+      "loss": 8.9981,
       "step": 820
     },
     {
       "epoch": 1.299330445057109,
-      "grad_norm": 0.25539177656173706,
+      "grad_norm": 6.222182273864746,
       "learning_rate": 4.039381666107621e-05,
-      "loss": 0.3897,
+      "loss": 9.2987,
       "step": 825
     },
     {
       "epoch": 1.3072075620322963,
-      "grad_norm": 0.26664361357688904,
+      "grad_norm": 5.359075546264648,
       "learning_rate": 4.026525968926624e-05,
-      "loss": 0.3816,
+      "loss": 8.97,
       "step": 830
     },
     {
       "epoch": 1.3150846790074833,
-      "grad_norm": 0.2982277274131775,
+      "grad_norm": 7.37460994720459,
       "learning_rate": 4.013605583142464e-05,
-      "loss": 0.3967,
+      "loss": 8.8287,
       "step": 835
     },
     {
       "epoch": 1.3229617959826703,
-      "grad_norm": 0.24720807373523712,
+      "grad_norm": 7.033757209777832,
       "learning_rate": 4.000621056273971e-05,
-      "loss": 0.3859,
+      "loss": 8.9608,
       "step": 840
     },
     {
       "epoch": 1.3308389129578575,
-      "grad_norm": 0.2673453986644745,
+      "grad_norm": 5.98358154296875,
       "learning_rate": 3.987572938558035e-05,
-      "loss": 0.3862,
+      "loss": 8.7479,
       "step": 845
     },
     {
       "epoch": 1.3387160299330445,
-      "grad_norm": 0.242339089512825,
+      "grad_norm": 8.184402465820312,
       "learning_rate": 3.974461782926299e-05,
-      "loss": 0.3903,
+      "loss": 8.6626,
       "step": 850
     },
     {
       "epoch": 1.3465931469082315,
-      "grad_norm": 0.24661767482757568,
+      "grad_norm": 9.955259323120117,
       "learning_rate": 3.96128814498172e-05,
-      "loss": 0.3831,
+      "loss": 8.926,
       "step": 855
     },
     {
       "epoch": 1.3544702638834187,
-      "grad_norm": 0.25107643008232117,
+      "grad_norm": 6.086458683013916,
       "learning_rate": 3.9480525829750295e-05,
-      "loss": 0.3815,
+      "loss": 8.6291,
       "step": 860
     },
     {
       "epoch": 1.3623473808586057,
-      "grad_norm": 0.25703874230384827,
+      "grad_norm": 6.604151725769043,
       "learning_rate": 3.9347556577810724e-05,
-      "loss": 0.3784,
+      "loss": 8.8916,
       "step": 865
     },
     {
       "epoch": 1.3702244978337927,
-      "grad_norm": 0.24667410552501678,
+      "grad_norm": 6.385220050811768,
       "learning_rate": 3.921397932875044e-05,
-      "loss": 0.3832,
+      "loss": 8.6173,
       "step": 870
     },
     {
       "epoch": 1.37810161480898,
-      "grad_norm": 0.24698962271213531,
+      "grad_norm": 5.808775424957275,
       "learning_rate": 3.9079799743086076e-05,
-      "loss": 0.3807,
+      "loss": 8.8478,
       "step": 875
     },
     {
       "epoch": 1.385978731784167,
-      "grad_norm": 0.2628852128982544,
+      "grad_norm": 5.057977676391602,
       "learning_rate": 3.8945023506859125e-05,
-      "loss": 0.3833,
+      "loss": 8.8043,
       "step": 880
     },
     {
       "epoch": 1.3938558487593542,
-      "grad_norm": 0.24828363955020905,
+      "grad_norm": 5.664560794830322,
       "learning_rate": 3.880965633139493e-05,
-      "loss": 0.3893,
+      "loss": 8.517,
       "step": 885
     },
     {
       "epoch": 1.4017329657345412,
-      "grad_norm": 0.25798603892326355,
+      "grad_norm": 6.176719665527344,
       "learning_rate": 3.867370395306068e-05,
-      "loss": 0.3908,
+      "loss": 9.4023,
       "step": 890
     },
     {
       "epoch": 1.4096100827097282,
-      "grad_norm": 0.24343079328536987,
+      "grad_norm": 5.604133605957031,
       "learning_rate": 3.8537172133022334e-05,
-      "loss": 0.3838,
+      "loss": 8.819,
       "step": 895
     },
     {
       "epoch": 1.4174871996849152,
-      "grad_norm": 0.25076451897621155,
+      "grad_norm": 6.113736152648926,
       "learning_rate": 3.840006665700049e-05,
-      "loss": 0.3865,
+      "loss": 8.5058,
       "step": 900
     },
     {
       "epoch": 1.4253643166601024,
-      "grad_norm": 0.24691860377788544,
+      "grad_norm": 5.689216136932373,
       "learning_rate": 3.8262393335025175e-05,
-      "loss": 0.3785,
+      "loss": 9.0603,
       "step": 905
     },
     {
       "epoch": 1.4332414336352894,
-      "grad_norm": 0.2401050329208374,
+      "grad_norm": 5.811161994934082,
       "learning_rate": 3.812415800118968e-05,
-      "loss": 0.3907,
+      "loss": 8.7633,
       "step": 910
     },
     {
       "epoch": 1.4411185506104767,
-      "grad_norm": 0.24668364226818085,
+      "grad_norm": 6.930866241455078,
       "learning_rate": 3.7985366513403275e-05,
-      "loss": 0.3744,
+      "loss": 8.8962,
       "step": 915
     },
     {
       "epoch": 1.4489956675856637,
-      "grad_norm": 0.24690192937850952,
+      "grad_norm": 4.998627662658691,
       "learning_rate": 3.784602475314303e-05,
-      "loss": 0.3841,
+      "loss": 8.8133,
       "step": 920
     },
     {
       "epoch": 1.4568727845608507,
-      "grad_norm": 0.25426268577575684,
+      "grad_norm": 5.640780925750732,
       "learning_rate": 3.7706138625204565e-05,
-      "loss": 0.3867,
+      "loss": 8.5275,
       "step": 925
     },
     {
       "epoch": 1.4647499015360377,
-      "grad_norm": 0.23657885193824768,
+      "grad_norm": 6.085617542266846,
       "learning_rate": 3.756571405745178e-05,
-      "loss": 0.3793,
+      "loss": 8.2861,
       "step": 930
     },
     {
       "epoch": 1.472627018511225,
-      "grad_norm": 0.2615351974964142,
+      "grad_norm": 6.3180108070373535,
       "learning_rate": 3.7424757000565734e-05,
-      "loss": 0.3904,
+      "loss": 8.8205,
       "step": 935
     },
     {
       "epoch": 1.480504135486412,
-      "grad_norm": 0.30141061544418335,
+      "grad_norm": 7.119158744812012,
       "learning_rate": 3.7283273427792407e-05,
-      "loss": 0.3871,
+      "loss": 8.8586,
       "step": 940
     },
     {
       "epoch": 1.4883812524615991,
-      "grad_norm": 0.23688402771949768,
+      "grad_norm": 7.279252529144287,
       "learning_rate": 3.714126933468959e-05,
-      "loss": 0.3846,
+      "loss": 8.2839,
       "step": 945
     },
     {
       "epoch": 1.4962583694367861,
-      "grad_norm": 0.22304469347000122,
+      "grad_norm": 5.607680320739746,
       "learning_rate": 3.6998750738872876e-05,
-      "loss": 0.3819,
+      "loss": 9.2545,
       "step": 950
     },
     {
       "epoch": 1.5041354864119731,
-      "grad_norm": 0.25306451320648193,
+      "grad_norm": 5.1762590408325195,
       "learning_rate": 3.6855723679760545e-05,
-      "loss": 0.3884,
+      "loss": 9.3281,
       "step": 955
     },
     {
       "epoch": 1.5120126033871601,
-      "grad_norm": 0.24044828116893768,
+      "grad_norm": 5.414069652557373,
       "learning_rate": 3.671219421831775e-05,
-      "loss": 0.3819,
+      "loss": 8.5181,
       "step": 960
     },
     {
       "epoch": 1.5198897203623474,
-      "grad_norm": 0.2118435502052307,
+      "grad_norm": 6.324887275695801,
       "learning_rate": 3.656816843679959e-05,
-      "loss": 0.3794,
+      "loss": 9.1248,
       "step": 965
     },
     {
       "epoch": 1.5277668373375346,
-      "grad_norm": 0.2519252598285675,
+      "grad_norm": 6.40478515625,
       "learning_rate": 3.64236524384934e-05,
-      "loss": 0.3896,
+      "loss": 9.0107,
       "step": 970
     },
     {
       "epoch": 1.5356439543127216,
-      "grad_norm": 0.24061793088912964,
+      "grad_norm": 6.342474937438965,
       "learning_rate": 3.627865234746014e-05,
-      "loss": 0.3874,
+      "loss": 8.8478,
       "step": 975
     },
     {
       "epoch": 1.5435210712879086,
-      "grad_norm": 0.237691268324852,
+      "grad_norm": 5.396797180175781,
       "learning_rate": 3.6133174308274826e-05,
-      "loss": 0.3955,
+      "loss": 8.8895,
       "step": 980
     },
     {
       "epoch": 1.5513981882630956,
-      "grad_norm": 0.2733243703842163,
+      "grad_norm": 5.2031378746032715,
       "learning_rate": 3.598722448576619e-05,
-      "loss": 0.3935,
+      "loss": 8.6624,
       "step": 985
     },
     {
       "epoch": 1.5592753052382828,
-      "grad_norm": 0.2568168044090271,
+      "grad_norm": 6.283775329589844,
       "learning_rate": 3.584080906475542e-05,
-      "loss": 0.3848,
+      "loss": 8.7416,
       "step": 990
     },
     {
       "epoch": 1.5671524222134698,
-      "grad_norm": 0.2655527889728546,
+      "grad_norm": 5.148292064666748,
       "learning_rate": 3.569393424979405e-05,
-      "loss": 0.3901,
+      "loss": 8.5302,
       "step": 995
     },
     {
       "epoch": 1.575029539188657,
-      "grad_norm": 0.25115740299224854,
+      "grad_norm": 5.293840408325195,
       "learning_rate": 3.554660626490109e-05,
-      "loss": 0.3814,
+      "loss": 8.4594,
       "step": 1000
     },
     {
       "epoch": 1.582906656163844,
-      "grad_norm": 0.23420318961143494,
+      "grad_norm": 4.885067462921143,
       "learning_rate": 3.53988313532992e-05,
-      "loss": 0.386,
+      "loss": 8.5849,
       "step": 1005
     },
     {
       "epoch": 1.590783773139031,
-      "grad_norm": 0.2673567235469818,
+      "grad_norm": 5.12977933883667,
       "learning_rate": 3.52506157771502e-05,
-      "loss": 0.3894,
+      "loss": 8.2942,
       "step": 1010
     },
     {
       "epoch": 1.598660890114218,
-      "grad_norm": 0.2313881516456604,
+      "grad_norm": 6.016343593597412,
       "learning_rate": 3.510196581728963e-05,
-      "loss": 0.3861,
+      "loss": 8.5103,
       "step": 1015
     },
     {
       "epoch": 1.6065380070894053,
-      "grad_norm": 0.23093390464782715,
+      "grad_norm": 6.179079055786133,
       "learning_rate": 3.495288777296067e-05,
-      "loss": 0.3882,
+      "loss": 8.3985,
       "step": 1020
     },
     {
       "epoch": 1.6144151240645923,
-      "grad_norm": 0.2772485017776489,
+      "grad_norm": 6.393824100494385,
       "learning_rate": 3.480338796154712e-05,
-      "loss": 0.3776,
+      "loss": 8.5286,
       "step": 1025
     },
     {
       "epoch": 1.6222922410397795,
-      "grad_norm": 0.2581574320793152,
+      "grad_norm": 6.632625102996826,
       "learning_rate": 3.4653472718305745e-05,
-      "loss": 0.3887,
+      "loss": 8.713,
       "step": 1030
     },
     {
       "epoch": 1.6301693580149665,
-      "grad_norm": 0.2368578016757965,
+      "grad_norm": 5.532236576080322,
       "learning_rate": 3.450314839609781e-05,
-      "loss": 0.3848,
+      "loss": 8.6,
       "step": 1035
     },
     {
       "epoch": 1.6380464749901535,
-      "grad_norm": 0.2310653030872345,
+      "grad_norm": 5.10947322845459,
       "learning_rate": 3.435242136511984e-05,
-      "loss": 0.372,
+      "loss": 8.7229,
       "step": 1040
     },
     {
       "epoch": 1.6459235919653405,
-      "grad_norm": 0.23389846086502075,
+      "grad_norm": 6.765851020812988,
       "learning_rate": 3.420129801263367e-05,
-      "loss": 0.3958,
+      "loss": 8.7038,
       "step": 1045
     },
     {
       "epoch": 1.6538007089405278,
-      "grad_norm": 0.23078757524490356,
+      "grad_norm": 6.051114082336426,
       "learning_rate": 3.404978474269583e-05,
-      "loss": 0.3807,
+      "loss": 8.9874,
       "step": 1050
     },
     {
       "epoch": 1.661677825915715,
-      "grad_norm": 0.23647628724575043,
+      "grad_norm": 5.545468807220459,
       "learning_rate": 3.389788797588611e-05,
-      "loss": 0.3702,
+      "loss": 8.7562,
       "step": 1055
     },
     {
       "epoch": 1.669554942890902,
-      "grad_norm": 0.23526796698570251,
+      "grad_norm": 5.190369129180908,
       "learning_rate": 3.374561414903552e-05,
-      "loss": 0.3795,
+      "loss": 8.7141,
       "step": 1060
     },
     {
       "epoch": 1.677432059866089,
-      "grad_norm": 0.23299497365951538,
+      "grad_norm": 5.89268684387207,
       "learning_rate": 3.3592969714953464e-05,
-      "loss": 0.3829,
+      "loss": 8.5692,
       "step": 1065
     },
     {
       "epoch": 1.685309176841276,
-      "grad_norm": 0.4668526351451874,
+      "grad_norm": 5.843109130859375,
       "learning_rate": 3.343996114215436e-05,
-      "loss": 0.3844,
+      "loss": 8.7083,
       "step": 1070
     },
     {
       "epoch": 1.693186293816463,
-      "grad_norm": 0.23494206368923187,
+      "grad_norm": 7.01223087310791,
       "learning_rate": 3.328659491458348e-05,
-      "loss": 0.3851,
+      "loss": 8.9503,
       "step": 1075
     },
     {
       "epoch": 1.7010634107916502,
-      "grad_norm": 0.250499427318573,
+      "grad_norm": 5.1653666496276855,
       "learning_rate": 3.31328775313422e-05,
-      "loss": 0.3811,
+      "loss": 8.6678,
       "step": 1080
     },
     {
       "epoch": 1.7089405277668375,
-      "grad_norm": 0.2643508315086365,
+      "grad_norm": 4.834114074707031,
       "learning_rate": 3.297881550641261e-05,
-      "loss": 0.3848,
+      "loss": 9.2416,
       "step": 1085
     },
     {
       "epoch": 1.7168176447420245,
-      "grad_norm": 0.2502589225769043,
+      "grad_norm": 5.249499320983887,
       "learning_rate": 3.282441536838144e-05,
-      "loss": 0.389,
+      "loss": 8.4502,
       "step": 1090
     },
     {
       "epoch": 1.7246947617172115,
-      "grad_norm": 0.2423262745141983,
+      "grad_norm": 5.539346694946289,
       "learning_rate": 3.266968366016342e-05,
-      "loss": 0.3784,
+      "loss": 8.7632,
       "step": 1095
     },
     {
       "epoch": 1.7325718786923985,
-      "grad_norm": 0.2834143340587616,
+      "grad_norm": 5.473989963531494,
       "learning_rate": 3.251462693872403e-05,
-      "loss": 0.3775,
+      "loss": 8.6672,
       "step": 1100
     },
     {
       "epoch": 1.7404489956675857,
-      "grad_norm": 0.24142897129058838,
+      "grad_norm": 5.088748455047607,
       "learning_rate": 3.2359251774801616e-05,
-      "loss": 0.3756,
+      "loss": 8.3875,
       "step": 1105
     },
     {
       "epoch": 1.7483261126427727,
-      "grad_norm": 0.24487896263599396,
+      "grad_norm": 5.558929920196533,
       "learning_rate": 3.2203564752628956e-05,
-      "loss": 0.3851,
+      "loss": 8.9813,
       "step": 1110
     },
     {
       "epoch": 1.75620322961796,
-      "grad_norm": 0.22995340824127197,
+      "grad_norm": 5.3655548095703125,
       "learning_rate": 3.204757246965424e-05,
-      "loss": 0.3888,
+      "loss": 8.9908,
       "step": 1115
     },
     {
       "epoch": 1.764080346593147,
-      "grad_norm": 0.21457414329051971,
+      "grad_norm": 6.439647197723389,
       "learning_rate": 3.18912815362615e-05,
-      "loss": 0.3858,
+      "loss": 8.1854,
       "step": 1120
     },
     {
       "epoch": 1.771957463568334,
-      "grad_norm": 0.21458269655704498,
+      "grad_norm": 5.290548324584961,
       "learning_rate": 3.173469857549048e-05,
-      "loss": 0.3812,
+      "loss": 8.9399,
       "step": 1125
     },
     {
       "epoch": 1.779834580543521,
-      "grad_norm": 0.23233187198638916,
+      "grad_norm": 5.242992401123047,
       "learning_rate": 3.157783022275599e-05,
-      "loss": 0.3849,
+      "loss": 8.6978,
       "step": 1130
     },
     {
       "epoch": 1.7877116975187082,
-      "grad_norm": 0.23942947387695312,
+      "grad_norm": 5.243824481964111,
       "learning_rate": 3.14206831255667e-05,
-      "loss": 0.3863,
+      "loss": 8.5899,
       "step": 1135
     },
     {
       "epoch": 1.7955888144938952,
-      "grad_norm": 0.22999627888202667,
+      "grad_norm": 6.161407947540283,
       "learning_rate": 3.126326394324346e-05,
-      "loss": 0.39,
+      "loss": 8.4425,
       "step": 1140
     },
     {
       "epoch": 1.8034659314690824,
-      "grad_norm": 0.22376017272472382,
+      "grad_norm": 5.396100997924805,
       "learning_rate": 3.110557934663708e-05,
-      "loss": 0.3776,
+      "loss": 8.4716,
       "step": 1145
     },
     {
       "epoch": 1.8113430484442694,
-      "grad_norm": 0.20644953846931458,
+      "grad_norm": 5.102015972137451,
       "learning_rate": 3.09476360178457e-05,
-      "loss": 0.3772,
+      "loss": 8.396,
       "step": 1150
     },
     {
       "epoch": 1.8192201654194564,
-      "grad_norm": 0.23101069033145905,
+      "grad_norm": 5.385177135467529,
       "learning_rate": 3.078944064993152e-05,
-      "loss": 0.3796,
+      "loss": 9.0414,
       "step": 1155
     },
     {
       "epoch": 1.8270972823946434,
-      "grad_norm": 0.23300951719284058,
+      "grad_norm": 5.5736165046691895,
       "learning_rate": 3.063099994663731e-05,
-      "loss": 0.3766,
+      "loss": 8.5122,
       "step": 1160
     },
     {
       "epoch": 1.8349743993698306,
-      "grad_norm": 0.4477635324001312,
+      "grad_norm": 5.4414286613464355,
       "learning_rate": 3.0472320622102207e-05,
-      "loss": 0.3864,
+      "loss": 8.3838,
       "step": 1165
     },
     {
       "epoch": 1.8428515163450179,
-      "grad_norm": 0.25588539242744446,
+      "grad_norm": 5.483443737030029,
       "learning_rate": 3.0313409400577264e-05,
-      "loss": 0.3766,
+      "loss": 8.0776,
       "step": 1170
     },
     {
       "epoch": 1.8507286333202049,
-      "grad_norm": 0.24095283448696136,
+      "grad_norm": 5.481230735778809,
       "learning_rate": 3.0154273016140504e-05,
-      "loss": 0.386,
+      "loss": 8.5493,
       "step": 1175
     },
     {
       "epoch": 1.8586057502953919,
-      "grad_norm": 0.2307787984609604,
+      "grad_norm": 5.8650312423706055,
       "learning_rate": 2.999491821241148e-05,
-      "loss": 0.3817,
+      "loss": 8.9861,
       "step": 1180
     },
     {
       "epoch": 1.8664828672705789,
-      "grad_norm": 0.24299356341362,
+      "grad_norm": 5.712615013122559,
       "learning_rate": 2.9835351742265622e-05,
-      "loss": 0.3888,
+      "loss": 8.5673,
       "step": 1185
     },
     {
       "epoch": 1.8743599842457659,
-      "grad_norm": 0.22085171937942505,
+      "grad_norm": 5.285681247711182,
       "learning_rate": 2.967558036754796e-05,
-      "loss": 0.376,
+      "loss": 8.4617,
       "step": 1190
     },
     {
       "epoch": 1.8822371012209531,
-      "grad_norm": 0.22518432140350342,
+      "grad_norm": 5.035665035247803,
       "learning_rate": 2.9515610858786657e-05,
-      "loss": 0.3841,
+      "loss": 8.1492,
       "step": 1195
     },
     {
       "epoch": 1.8901142181961403,
-      "grad_norm": 0.21670100092887878,
+      "grad_norm": 5.165551662445068,
       "learning_rate": 2.935544999490607e-05,
-      "loss": 0.3876,
+      "loss": 8.599,
       "step": 1200
     },
     {
       "epoch": 1.8979913351713273,
-      "grad_norm": 0.22420799732208252,
+      "grad_norm": 5.6296610832214355,
       "learning_rate": 2.9195104562939474e-05,
-      "loss": 0.39,
+      "loss": 8.6841,
       "step": 1205
     },
     {
       "epoch": 1.9058684521465143,
-      "grad_norm": 0.22058328986167908,
+      "grad_norm": 5.5025634765625,
       "learning_rate": 2.90345813577415e-05,
-      "loss": 0.3775,
+      "loss": 8.4796,
       "step": 1210
     },
     {
       "epoch": 1.9137455691217014,
-      "grad_norm": 0.22855107486248016,
+      "grad_norm": 5.420698642730713,
       "learning_rate": 2.8873887181700127e-05,
-      "loss": 0.3821,
+      "loss": 8.134,
       "step": 1215
     },
     {
       "epoch": 1.9216226860968886,
-      "grad_norm": 0.2261258214712143,
+      "grad_norm": 5.495443344116211,
       "learning_rate": 2.8713028844448464e-05,
-      "loss": 0.3881,
+      "loss": 8.4425,
       "step": 1220
     },
     {
       "epoch": 1.9294998030720756,
-      "grad_norm": 0.23165945708751678,
+      "grad_norm": 5.478509902954102,
       "learning_rate": 2.8552013162576186e-05,
-      "loss": 0.3813,
+      "loss": 8.2366,
       "step": 1225
     },
     {
       "epoch": 1.9373769200472628,
-      "grad_norm": 0.2159801721572876,
+      "grad_norm": 5.962779521942139,
       "learning_rate": 2.8390846959340638e-05,
-      "loss": 0.3827,
+      "loss": 8.4159,
       "step": 1230
     },
     {
       "epoch": 1.9452540370224498,
-      "grad_norm": 0.2017299383878708,
+      "grad_norm": 7.839330673217773,
       "learning_rate": 2.8229537064377735e-05,
-      "loss": 0.3784,
+      "loss": 8.2683,
       "step": 1235
     },
     {
       "epoch": 1.9531311539976368,
-      "grad_norm": 0.21241925656795502,
+      "grad_norm": 5.109673023223877,
       "learning_rate": 2.806809031341251e-05,
-      "loss": 0.3763,
+      "loss": 8.6595,
       "step": 1240
     },
     {
       "epoch": 1.9610082709728238,
-      "grad_norm": 0.22526709735393524,
+      "grad_norm": 5.204682350158691,
       "learning_rate": 2.790651354796945e-05,
-      "loss": 0.3762,
+      "loss": 8.6676,
       "step": 1245
     },
     {
       "epoch": 1.968885387948011,
-      "grad_norm": 0.21094514429569244,
+      "grad_norm": 7.151166915893555,
       "learning_rate": 2.7744813615082588e-05,
-      "loss": 0.3792,
+      "loss": 8.2718,
       "step": 1250
     },
     {
       "epoch": 1.9767625049231983,
-      "grad_norm": 0.2524140775203705,
+      "grad_norm": 5.639292240142822,
       "learning_rate": 2.7582997367005337e-05,
-      "loss": 0.3873,
+      "loss": 8.4982,
       "step": 1255
     },
     {
       "epoch": 1.9846396218983853,
-      "grad_norm": 0.23768506944179535,
+      "grad_norm": 7.254332065582275,
       "learning_rate": 2.742107166092015e-05,
-      "loss": 0.3696,
+      "loss": 8.3822,
       "step": 1260
     },
     {
       "epoch": 1.9925167388735723,
-      "grad_norm": 0.21980994939804077,
+      "grad_norm": 5.9872307777404785,
       "learning_rate": 2.7259043358647863e-05,
-      "loss": 0.3852,
+      "loss": 8.5006,
       "step": 1265
     },
     {
-      "epoch": 2.0,
-      "grad_norm": 0.244730606675148,
+      "epoch": 2.0015754233950376,
+      "grad_norm": 4.944819450378418,
       "learning_rate": 2.709691932635701e-05,
-      "loss": 0.3773,
+      "loss": 9.5234,
       "step": 1270
     },
     {
-      "epoch": 2.007877116975187,
-      "grad_norm": 0.2792533338069916,
+      "epoch": 2.0094525403702246,
+      "grad_norm": 4.903031349182129,
       "learning_rate": 2.6934706434272794e-05,
-      "loss": 0.3618,
+      "loss": 8.1337,
       "step": 1275
     },
     {
-      "epoch": 2.015754233950374,
-      "grad_norm": 0.2545369565486908,
+      "epoch": 2.0173296573454116,
+      "grad_norm": 5.022507667541504,
       "learning_rate": 2.6772411556385966e-05,
-      "loss": 0.3541,
+      "loss": 8.5159,
       "step": 1280
     },
     {
-      "epoch": 2.0236313509255615,
-      "grad_norm": 0.2422221451997757,
+      "epoch": 2.0252067743205986,
+      "grad_norm": 5.7532525062561035,
       "learning_rate": 2.6610041570161563e-05,
-      "loss": 0.3592,
+      "loss": 7.8905,
       "step": 1285
     },
     {
-      "epoch": 2.0315084679007485,
-      "grad_norm": 0.2332969456911087,
+      "epoch": 2.0330838912957856,
+      "grad_norm": 5.3462653160095215,
       "learning_rate": 2.6447603356247396e-05,
-      "loss": 0.3516,
+      "loss": 8.5373,
       "step": 1290
     },
     {
-      "epoch": 2.0393855848759355,
-      "grad_norm": 0.27313655614852905,
+      "epoch": 2.0409610082709726,
+      "grad_norm": 5.146194934844971,
       "learning_rate": 2.628510379818256e-05,
-      "loss": 0.354,
+      "loss": 8.375,
       "step": 1295
     },
     {
-      "epoch": 2.0472627018511225,
-      "grad_norm": 0.22538913786411285,
+      "epoch": 2.04883812524616,
+      "grad_norm": 4.806943893432617,
       "learning_rate": 2.6122549782105666e-05,
-      "loss": 0.3487,
+      "loss": 8.4479,
       "step": 1300
     },
     {
-      "epoch": 2.0551398188263095,
-      "grad_norm": 0.24524974822998047,
+      "epoch": 2.056715242221347,
+      "grad_norm": 6.281928062438965,
       "learning_rate": 2.5959948196463058e-05,
-      "loss": 0.3497,
+      "loss": 8.1086,
       "step": 1305
     },
     {
-      "epoch": 2.0630169358014965,
-      "grad_norm": 0.2330634593963623,
+      "epoch": 2.064592359196534,
+      "grad_norm": 5.9106550216674805,
       "learning_rate": 2.5797305931716924e-05,
-      "loss": 0.3539,
+      "loss": 8.3172,
       "step": 1310
     },
     {
-      "epoch": 2.070894052776684,
-      "grad_norm": 0.247636616230011,
+      "epoch": 2.072469476171721,
+      "grad_norm": 5.928641319274902,
       "learning_rate": 2.5634629880053256e-05,
-      "loss": 0.3492,
+      "loss": 8.3275,
       "step": 1315
     },
     {
-      "epoch": 2.078771169751871,
-      "grad_norm": 0.21172009408473969,
+      "epoch": 2.080346593146908,
+      "grad_norm": 5.428690433502197,
       "learning_rate": 2.5471926935089835e-05,
-      "loss": 0.343,
+      "loss": 8.5239,
       "step": 1320
     },
     {
-      "epoch": 2.086648286727058,
-      "grad_norm": 0.25462400913238525,
+      "epoch": 2.0882237101220955,
+      "grad_norm": 5.30228853225708,
       "learning_rate": 2.5309203991584073e-05,
-      "loss": 0.3417,
+      "loss": 8.3246,
       "step": 1325
     },
     {
-      "epoch": 2.094525403702245,
-      "grad_norm": 0.24423840641975403,
+      "epoch": 2.0961008270972825,
+      "grad_norm": 5.126616477966309,
       "learning_rate": 2.514646794514085e-05,
-      "loss": 0.3383,
+      "loss": 7.8035,
       "step": 1330
     },
     {
-      "epoch": 2.102402520677432,
-      "grad_norm": 0.23146118223667145,
+      "epoch": 2.1039779440724695,
+      "grad_norm": 5.645951747894287,
       "learning_rate": 2.4983725691920294e-05,
-      "loss": 0.344,
+      "loss": 8.3698,
       "step": 1335
     },
     {
-      "epoch": 2.110279637652619,
-      "grad_norm": 0.229843407869339,
+      "epoch": 2.1118550610476565,
+      "grad_norm": 5.491515636444092,
       "learning_rate": 2.482098412834556e-05,
-      "loss": 0.3429,
+      "loss": 7.826,
       "step": 1340
     },
     {
-      "epoch": 2.1181567546278064,
-      "grad_norm": 0.24268518388271332,
+      "epoch": 2.1197321780228435,
+      "grad_norm": 5.176724910736084,
       "learning_rate": 2.4658250150810578e-05,
-      "loss": 0.3337,
+      "loss": 8.0034,
       "step": 1345
     },
     {
-      "epoch": 2.1260338716029934,
-      "grad_norm": 0.2576902210712433,
+      "epoch": 2.1276092949980305,
+      "grad_norm": 4.87959098815918,
       "learning_rate": 2.449553065538781e-05,
-      "loss": 0.344,
+      "loss": 8.2813,
       "step": 1350
     },
     {
-      "epoch": 2.1339109885781804,
-      "grad_norm": 0.2369920015335083,
+      "epoch": 2.135486411973218,
+      "grad_norm": 5.171248435974121,
       "learning_rate": 2.4332832537536e-05,
-      "loss": 0.3362,
+      "loss": 8.4077,
       "step": 1355
     },
     {
-      "epoch": 2.1417881055533674,
-      "grad_norm": 0.2687665522098541,
+      "epoch": 2.143363528948405,
+      "grad_norm": 5.680236339569092,
       "learning_rate": 2.4170162691808034e-05,
-      "loss": 0.3318,
+      "loss": 8.2604,
       "step": 1360
     },
     {
-      "epoch": 2.1496652225285544,
-      "grad_norm": 0.23115476965904236,
+      "epoch": 2.151240645923592,
+      "grad_norm": 5.374014854431152,
       "learning_rate": 2.4007528011558683e-05,
-      "loss": 0.3338,
+      "loss": 8.0889,
       "step": 1365
     },
     {
-      "epoch": 2.157542339503742,
-      "grad_norm": 0.3254106342792511,
+      "epoch": 2.159117762898779,
+      "grad_norm": 5.249903678894043,
       "learning_rate": 2.3844935388652545e-05,
-      "loss": 0.3337,
+      "loss": 8.3831,
       "step": 1370
     },
     {
-      "epoch": 2.165419456478929,
-      "grad_norm": 0.2389201521873474,
+      "epoch": 2.166994879873966,
+      "grad_norm": 5.103092670440674,
       "learning_rate": 2.3682391713171998e-05,
-      "loss": 0.3269,
+      "loss": 8.314,
       "step": 1375
     },
     {
-      "epoch": 2.173296573454116,
-      "grad_norm": 0.23049896955490112,
+      "epoch": 2.174871996849153,
+      "grad_norm": 5.576798915863037,
       "learning_rate": 2.351990387312519e-05,
-      "loss": 0.3255,
+      "loss": 7.9968,
       "step": 1380
     },
     {
-      "epoch": 2.181173690429303,
-      "grad_norm": 0.24909381568431854,
+      "epoch": 2.1827491138243404,
+      "grad_norm": 5.228267192840576,
       "learning_rate": 2.3357478754154173e-05,
-      "loss": 0.3362,
+      "loss": 8.3819,
       "step": 1385
     },
     {
-      "epoch": 2.18905080740449,
-      "grad_norm": 0.25661030411720276,
+      "epoch": 2.1906262307995275,
+      "grad_norm": 4.981462478637695,
       "learning_rate": 2.3195123239243098e-05,
-      "loss": 0.3246,
+      "loss": 8.5833,
       "step": 1390
     },
     {
-      "epoch": 2.196927924379677,
-      "grad_norm": 0.225275456905365,
+      "epoch": 2.1985033477747145,
+      "grad_norm": 5.161903381347656,
       "learning_rate": 2.303284420842658e-05,
-      "loss": 0.3246,
+      "loss": 8.0129,
       "step": 1395
     },
     {
-      "epoch": 2.2048050413548643,
-      "grad_norm": 0.2919777035713196,
+      "epoch": 2.2063804647499015,
+      "grad_norm": 5.56730842590332,
       "learning_rate": 2.2870648538498085e-05,
-      "loss": 0.3277,
+      "loss": 8.2146,
       "step": 1400
     },
     {
-      "epoch": 2.2126821583300513,
-      "grad_norm": 0.25202763080596924,
+      "epoch": 2.2142575817250885,
+      "grad_norm": 5.310222625732422,
       "learning_rate": 2.2708543102718542e-05,
-      "loss": 0.3249,
+      "loss": 8.3771,
       "step": 1405
     },
     {
-      "epoch": 2.2205592753052383,
-      "grad_norm": 0.292521595954895,
+      "epoch": 2.222134698700276,
+      "grad_norm": 5.077228546142578,
       "learning_rate": 2.254653477052514e-05,
-      "loss": 0.3302,
+      "loss": 8.0806,
       "step": 1410
     },
     {
-      "epoch": 2.2284363922804253,
-      "grad_norm": 0.2488042116165161,
+      "epoch": 2.230011815675463,
+      "grad_norm": 5.488475322723389,
       "learning_rate": 2.238463040724011e-05,
-      "loss": 0.3302,
+      "loss": 7.9597,
       "step": 1415
     },
     {
-      "epoch": 2.2363135092556123,
-      "grad_norm": 0.24722595512866974,
+      "epoch": 2.23788893265065,
+      "grad_norm": 4.879157543182373,
       "learning_rate": 2.2222836873779888e-05,
-      "loss": 0.3289,
+      "loss": 8.1727,
       "step": 1420
     },
     {
-      "epoch": 2.2441906262307993,
-      "grad_norm": 0.25138890743255615,
+      "epoch": 2.245766049625837,
+      "grad_norm": 5.107478618621826,
       "learning_rate": 2.2061161026364353e-05,
-      "loss": 0.3286,
+      "loss": 8.0123,
       "step": 1425
     },
     {
-      "epoch": 2.252067743205987,
-      "grad_norm": 0.23644718527793884,
+      "epoch": 2.253643166601024,
+      "grad_norm": 5.371547222137451,
       "learning_rate": 2.1899609716226263e-05,
-      "loss": 0.3278,
+      "loss": 8.2489,
       "step": 1430
     },
     {
-      "epoch": 2.259944860181174,
-      "grad_norm": 0.2490551620721817,
+      "epoch": 2.261520283576211,
+      "grad_norm": 5.334740161895752,
       "learning_rate": 2.173818978932095e-05,
-      "loss": 0.3275,
+      "loss": 8.536,
       "step": 1435
     },
     {
-      "epoch": 2.267821977156361,
-      "grad_norm": 0.22873945534229279,
+      "epoch": 2.2693974005513984,
+      "grad_norm": 5.4585394859313965,
       "learning_rate": 2.1576908086036195e-05,
-      "loss": 0.3315,
+      "loss": 8.2132,
       "step": 1440
     },
     {
-      "epoch": 2.275699094131548,
-      "grad_norm": 0.24433307349681854,
+      "epoch": 2.2772745175265854,
+      "grad_norm": 6.0415449142456055,
       "learning_rate": 2.1415771440902384e-05,
-      "loss": 0.3308,
+      "loss": 7.9417,
       "step": 1445
     },
     {
-      "epoch": 2.283576211106735,
-      "grad_norm": 0.24108295142650604,
+      "epoch": 2.2851516345017724,
+      "grad_norm": 5.975248336791992,
       "learning_rate": 2.1254786682302847e-05,
-      "loss": 0.325,
+      "loss": 8.2411,
       "step": 1450
     },
     {
-      "epoch": 2.2914533280819223,
-      "grad_norm": 0.275078147649765,
+      "epoch": 2.2930287514769594,
+      "grad_norm": 5.171903133392334,
       "learning_rate": 2.1093960632184513e-05,
-      "loss": 0.3317,
+      "loss": 8.4717,
       "step": 1455
     },
     {
-      "epoch": 2.2993304450571093,
-      "grad_norm": 0.24685098230838776,
+      "epoch": 2.3009058684521464,
+      "grad_norm": 5.419390678405762,
       "learning_rate": 2.0933300105768867e-05,
-      "loss": 0.3175,
+      "loss": 7.9774,
       "step": 1460
     },
     {
-      "epoch": 2.3072075620322963,
-      "grad_norm": 0.22858254611492157,
+      "epoch": 2.3087829854273334,
+      "grad_norm": 5.009170055389404,
       "learning_rate": 2.0772811911263048e-05,
-      "loss": 0.3268,
+      "loss": 7.9232,
       "step": 1465
     },
     {
-      "epoch": 2.3150846790074833,
-      "grad_norm": 0.25581493973731995,
+      "epoch": 2.316660102402521,
+      "grad_norm": 5.920315265655518,
       "learning_rate": 2.0612502849571425e-05,
-      "loss": 0.3259,
+      "loss": 8.3842,
       "step": 1470
     },
     {
-      "epoch": 2.3229617959826703,
-      "grad_norm": 0.2207001894712448,
+      "epoch": 2.324537219377708,
+      "grad_norm": 5.717631816864014,
       "learning_rate": 2.0452379714007375e-05,
-      "loss": 0.3268,
+      "loss": 8.3991,
       "step": 1475
     },
     {
-      "epoch": 2.3308389129578573,
-      "grad_norm": 0.2586366832256317,
+      "epoch": 2.332414336352895,
+      "grad_norm": 6.118537425994873,
       "learning_rate": 2.0292449290005395e-05,
-      "loss": 0.3264,
+      "loss": 8.1187,
       "step": 1480
     },
     {
-      "epoch": 2.3387160299330443,
-      "grad_norm": 0.2496010810136795,
+      "epoch": 2.340291453328082,
+      "grad_norm": 4.826847076416016,
       "learning_rate": 2.013271835483357e-05,
-      "loss": 0.3275,
+      "loss": 7.8728,
       "step": 1485
     },
     {
-      "epoch": 2.3465931469082317,
-      "grad_norm": 0.21064487099647522,
+      "epoch": 2.348168570303269,
+      "grad_norm": 5.45656681060791,
       "learning_rate": 1.9973193677306377e-05,
-      "loss": 0.324,
+      "loss": 8.1391,
       "step": 1490
     },
     {
-      "epoch": 2.3544702638834187,
-      "grad_norm": 0.22213426232337952,
+      "epoch": 2.3560456872784563,
+      "grad_norm": 5.478461265563965,
       "learning_rate": 1.9813882017497848e-05,
-      "loss": 0.3172,
+      "loss": 8.1173,
       "step": 1495
     },
     {
-      "epoch": 2.3623473808586057,
-      "grad_norm": 0.29100117087364197,
+      "epoch": 2.3639228042536433,
+      "grad_norm": 6.264642715454102,
       "learning_rate": 1.9654790126455103e-05,
-      "loss": 0.3235,
+      "loss": 8.1301,
       "step": 1500
     },
     {
-      "epoch": 2.3702244978337927,
-      "grad_norm": 0.24606060981750488,
+      "epoch": 2.3717999212288303,
+      "grad_norm": 5.903857231140137,
       "learning_rate": 1.949592474591224e-05,
-      "loss": 0.3332,
+      "loss": 7.9252,
       "step": 1505
     },
     {
-      "epoch": 2.3781016148089797,
-      "grad_norm": 0.23023582994937897,
+      "epoch": 2.3796770382040173,
+      "grad_norm": 4.8199896812438965,
       "learning_rate": 1.9337292608004715e-05,
-      "loss": 0.3238,
+      "loss": 8.3755,
       "step": 1510
     },
     {
-      "epoch": 2.385978731784167,
-      "grad_norm": 0.2493922859430313,
+      "epoch": 2.3875541551792043,
+      "grad_norm": 4.588783264160156,
       "learning_rate": 1.917890043498397e-05,
-      "loss": 0.3284,
+      "loss": 8.1026,
       "step": 1515
     },
     {
-      "epoch": 2.393855848759354,
-      "grad_norm": 0.2237468957901001,
+      "epoch": 2.3954312721543913,
+      "grad_norm": 5.213608741760254,
       "learning_rate": 1.9020754938932624e-05,
-      "loss": 0.3307,
+      "loss": 8.1211,
       "step": 1520
     },
     {
-      "epoch": 2.401732965734541,
-      "grad_norm": 0.24473683536052704,
+      "epoch": 2.4033083891295783,
+      "grad_norm": 5.581782817840576,
       "learning_rate": 1.8862862821480025e-05,
-      "loss": 0.3219,
+      "loss": 7.7893,
       "step": 1525
     },
     {
-      "epoch": 2.409610082709728,
-      "grad_norm": 0.2150821089744568,
+      "epoch": 2.411185506104766,
+      "grad_norm": 5.054122447967529,
       "learning_rate": 1.8705230773518267e-05,
-      "loss": 0.328,
+      "loss": 8.1924,
       "step": 1530
     },
     {
-      "epoch": 2.417487199684915,
-      "grad_norm": 0.24754087626934052,
+      "epoch": 2.419062623079953,
+      "grad_norm": 4.978759288787842,
       "learning_rate": 1.854786547491864e-05,
-      "loss": 0.3251,
+      "loss": 8.1678,
       "step": 1535
     },
     {
-      "epoch": 2.425364316660102,
-      "grad_norm": 0.23762992024421692,
+      "epoch": 2.42693974005514,
+      "grad_norm": 5.0260329246521,
       "learning_rate": 1.8390773594248568e-05,
-      "loss": 0.3266,
+      "loss": 8.1418,
       "step": 1540
     },
     {
-      "epoch": 2.4332414336352897,
-      "grad_norm": 0.23614896833896637,
+      "epoch": 2.434816857030327,
+      "grad_norm": 5.8550262451171875,
       "learning_rate": 1.823396178848903e-05,
-      "loss": 0.3283,
+      "loss": 8.2088,
       "step": 1545
     },
     {
-      "epoch": 2.4411185506104767,
-      "grad_norm": 0.2320897877216339,
+      "epoch": 2.442693974005514,
+      "grad_norm": 6.323260307312012,
       "learning_rate": 1.8077436702752448e-05,
-      "loss": 0.3286,
+      "loss": 7.9444,
       "step": 1550
     },
     {
-      "epoch": 2.4489956675856637,
-      "grad_norm": 0.22540679574012756,
+      "epoch": 2.4505710909807012,
+      "grad_norm": 5.396198272705078,
       "learning_rate": 1.7921204970001065e-05,
-      "loss": 0.3215,
+      "loss": 7.8484,
       "step": 1555
     },
     {
-      "epoch": 2.4568727845608507,
-      "grad_norm": 0.24557819962501526,
+      "epoch": 2.4584482079558883,
+      "grad_norm": 5.220623970031738,
       "learning_rate": 1.776527321076596e-05,
-      "loss": 0.3183,
+      "loss": 8.1014,
       "step": 1560
     },
     {
-      "epoch": 2.4647499015360377,
-      "grad_norm": 0.23252086341381073,
+      "epoch": 2.4663253249310753,
+      "grad_norm": 16.552457809448242,
       "learning_rate": 1.7609648032866362e-05,
-      "loss": 0.3209,
+      "loss": 8.2018,
       "step": 1565
     },
     {
-      "epoch": 2.4726270185112247,
-      "grad_norm": 0.27775880694389343,
+      "epoch": 2.4742024419062623,
+      "grad_norm": 5.957991600036621,
       "learning_rate": 1.7454336031129743e-05,
-      "loss": 0.324,
+      "loss": 7.8216,
       "step": 1570
     },
     {
-      "epoch": 2.480504135486412,
-      "grad_norm": 0.225099116563797,
+      "epoch": 2.4820795588814493,
+      "grad_norm": 5.393924713134766,
       "learning_rate": 1.729934378711228e-05,
-      "loss": 0.3263,
+      "loss": 8.3398,
       "step": 1575
     },
     {
-      "epoch": 2.488381252461599,
-      "grad_norm": 0.2224494367837906,
+      "epoch": 2.4899566758566367,
+      "grad_norm": 4.859684944152832,
       "learning_rate": 1.7144677868820026e-05,
-      "loss": 0.3128,
+      "loss": 8.6844,
       "step": 1580
     },
     {
-      "epoch": 2.496258369436786,
-      "grad_norm": 0.21441712975502014,
+      "epoch": 2.4978337928318237,
+      "grad_norm": 5.323124885559082,
       "learning_rate": 1.6990344830430498e-05,
-      "loss": 0.3294,
+      "loss": 8.0736,
       "step": 1585
     },
     {
-      "epoch": 2.504135486411973,
-      "grad_norm": 0.2346782237291336,
+      "epoch": 2.5057109098070107,
+      "grad_norm": 8.831464767456055,
       "learning_rate": 1.6836351212014996e-05,
-      "loss": 0.3302,
+      "loss": 8.1686,
       "step": 1590
     },
     {
       "epoch": 2.51201260338716,
-      "grad_norm": 0.23030629754066467,
+      "grad_norm": 5.187417507171631,
       "learning_rate": 1.6682703539261442e-05,
-      "loss": 0.3268,
+      "loss": 7.6429,
       "step": 1595
     },
     {
       "epoch": 2.5198897203623476,
-      "grad_norm": 0.2294086217880249,
+      "grad_norm": 6.1180500984191895,
       "learning_rate": 1.652940832319785e-05,
-      "loss": 0.3288,
+      "loss": 7.9255,
       "step": 1600
     },
     {
       "epoch": 2.5277668373375346,
-      "grad_norm": 0.32409849762916565,
+      "grad_norm": 8.949240684509277,
       "learning_rate": 1.6376472059916354e-05,
-      "loss": 0.3262,
+      "loss": 8.0755,
       "step": 1605
     },
     {
       "epoch": 2.5356439543127216,
-      "grad_norm": 0.22377949953079224,
+      "grad_norm": 4.905629634857178,
       "learning_rate": 1.6223901230298062e-05,
-      "loss": 0.3171,
+      "loss": 7.8017,
       "step": 1610
     },
     {
       "epoch": 2.5435210712879086,
-      "grad_norm": 0.2417742758989334,
+      "grad_norm": 5.435753345489502,
       "learning_rate": 1.6071702299738268e-05,
-      "loss": 0.3261,
+      "loss": 7.7271,
       "step": 1615
     },
     {
       "epoch": 2.5513981882630956,
-      "grad_norm": 0.22446629405021667,
+      "grad_norm": 5.005056858062744,
       "learning_rate": 1.5919881717872564e-05,
-      "loss": 0.3199,
+      "loss": 8.3681,
       "step": 1620
     },
     {
       "epoch": 2.559275305238283,
-      "grad_norm": 0.22529666125774384,
+      "grad_norm": 5.524580001831055,
       "learning_rate": 1.5768445918303505e-05,
-      "loss": 0.321,
+      "loss": 7.6268,
       "step": 1625
     },
     {
       "epoch": 2.5671524222134696,
-      "grad_norm": 0.22095133364200592,
+      "grad_norm": 5.758496284484863,
       "learning_rate": 1.5617401318327985e-05,
-      "loss": 0.3185,
+      "loss": 7.8986,
       "step": 1630
     },
     {
       "epoch": 2.575029539188657,
-      "grad_norm": 0.24521619081497192,
+      "grad_norm": 5.624476909637451,
       "learning_rate": 1.546675431866526e-05,
-      "loss": 0.3128,
+      "loss": 7.437,
       "step": 1635
     },
     {
       "epoch": 2.582906656163844,
-      "grad_norm": 0.23159317672252655,
+      "grad_norm": 4.7457594871521,
       "learning_rate": 1.5316511303185755e-05,
-      "loss": 0.3217,
+      "loss": 8.1143,
       "step": 1640
     },
     {
       "epoch": 2.590783773139031,
-      "grad_norm": 0.22384974360466003,
+      "grad_norm": 5.465321063995361,
       "learning_rate": 1.5166678638640521e-05,
-      "loss": 0.3173,
+      "loss": 7.4633,
       "step": 1645
     },
     {
       "epoch": 2.598660890114218,
-      "grad_norm": 0.22395044565200806,
+      "grad_norm": 5.189293384552002,
       "learning_rate": 1.5017262674391428e-05,
-      "loss": 0.3188,
+      "loss": 7.7685,
       "step": 1650
     },
     {
       "epoch": 2.606538007089405,
-      "grad_norm": 0.24889998137950897,
+      "grad_norm": 5.231342792510986,
       "learning_rate": 1.486826974214209e-05,
-      "loss": 0.3193,
+      "loss": 7.6534,
       "step": 1655
     },
     {
       "epoch": 2.6144151240645925,
-      "grad_norm": 0.25244370102882385,
+      "grad_norm": 5.773874282836914,
       "learning_rate": 1.4719706155669587e-05,
-      "loss": 0.3219,
+      "loss": 7.7808,
       "step": 1660
     },
     {
       "epoch": 2.6222922410397795,
-      "grad_norm": 0.22579914331436157,
+      "grad_norm": 5.353034019470215,
       "learning_rate": 1.4571578210556896e-05,
-      "loss": 0.3209,
+      "loss": 7.7763,
       "step": 1665
     },
     {
       "epoch": 2.6301693580149665,
-      "grad_norm": 0.21842360496520996,
+      "grad_norm": 5.128270149230957,
       "learning_rate": 1.4423892183926092e-05,
-      "loss": 0.3221,
+      "loss": 7.8019,
       "step": 1670
     },
     {
       "epoch": 2.6380464749901535,
-      "grad_norm": 0.2209043651819229,
+      "grad_norm": 5.596746444702148,
       "learning_rate": 1.4276654334172332e-05,
-      "loss": 0.3235,
+      "loss": 7.3829,
       "step": 1675
     },
     {
       "epoch": 2.6459235919653405,
-      "grad_norm": 0.24010902643203735,
+      "grad_norm": 4.946563720703125,
       "learning_rate": 1.4129870900698739e-05,
-      "loss": 0.3207,
+      "loss": 8.053,
       "step": 1680
     },
     {
       "epoch": 2.653800708940528,
-      "grad_norm": 0.2469053864479065,
+      "grad_norm": 5.08951473236084,
       "learning_rate": 1.3983548103651861e-05,
-      "loss": 0.3238,
+      "loss": 7.9546,
       "step": 1685
     },
     {
       "epoch": 2.661677825915715,
-      "grad_norm": 0.21121683716773987,
+      "grad_norm": 6.104579925537109,
       "learning_rate": 1.3837692143658182e-05,
-      "loss": 0.317,
+      "loss": 7.8367,
       "step": 1690
     },
     {
       "epoch": 2.669554942890902,
-      "grad_norm": 0.22918404638767242,
+      "grad_norm": 5.49988317489624,
       "learning_rate": 1.369230920156134e-05,
-      "loss": 0.3162,
+      "loss": 7.8629,
       "step": 1695
     },
     {
       "epoch": 2.677432059866089,
-      "grad_norm": 0.22199533879756927,
+      "grad_norm": 5.227181434631348,
       "learning_rate": 1.3547405438160205e-05,
-      "loss": 0.3173,
+      "loss": 8.1967,
       "step": 1700
     },
     {
       "epoch": 2.685309176841276,
-      "grad_norm": 0.22123722732067108,
+      "grad_norm": 6.716334819793701,
       "learning_rate": 1.340298699394777e-05,
-      "loss": 0.3156,
+      "loss": 7.4787,
       "step": 1705
     },
     {
       "epoch": 2.693186293816463,
-      "grad_norm": 0.2306554913520813,
+      "grad_norm": 5.50681209564209,
       "learning_rate": 1.3259059988851e-05,
-      "loss": 0.3236,
+      "loss": 7.2509,
       "step": 1710
     },
     {
       "epoch": 2.70106341079165,
-      "grad_norm": 0.21142704784870148,
+      "grad_norm": 4.969735145568848,
       "learning_rate": 1.3115630521971468e-05,
-      "loss": 0.3144,
+      "loss": 7.8345,
       "step": 1715
     },
     {
       "epoch": 2.7089405277668375,
-      "grad_norm": 0.215803861618042,
+      "grad_norm": 5.385042190551758,
       "learning_rate": 1.2972704671326863e-05,
-      "loss": 0.3196,
+      "loss": 7.6157,
       "step": 1720
     },
     {
       "epoch": 2.7168176447420245,
-      "grad_norm": 0.2311914712190628,
+      "grad_norm": 5.288397789001465,
       "learning_rate": 1.2830288493593451e-05,
-      "loss": 0.3179,
+      "loss": 7.6661,
       "step": 1725
     },
     {
       "epoch": 2.7246947617172115,
-      "grad_norm": 0.23759211599826813,
+      "grad_norm": 5.0901994705200195,
       "learning_rate": 1.2688388023849482e-05,
-      "loss": 0.3173,
+      "loss": 7.8334,
       "step": 1730
     },
     {
       "epoch": 2.7325718786923985,
-      "grad_norm": 0.22561126947402954,
+      "grad_norm": 5.180675983428955,
       "learning_rate": 1.2547009275319313e-05,
-      "loss": 0.319,
+      "loss": 7.3121,
       "step": 1735
     },
     {
       "epoch": 2.7404489956675855,
-      "grad_norm": 0.21002857387065887,
+      "grad_norm": 7.612630367279053,
       "learning_rate": 1.2406158239118677e-05,
-      "loss": 0.3104,
+      "loss": 7.5115,
       "step": 1740
     },
     {
       "epoch": 2.748326112642773,
-      "grad_norm": 0.218175008893013,
+      "grad_norm": 9.157015800476074,
       "learning_rate": 1.2265840884000792e-05,
-      "loss": 0.3209,
+      "loss": 7.7749,
       "step": 1745
     },
     {
       "epoch": 2.75620322961796,
-      "grad_norm": 0.27158647775650024,
+      "grad_norm": 5.494700908660889,
       "learning_rate": 1.2126063156103437e-05,
-      "loss": 0.3209,
+      "loss": 7.8249,
       "step": 1750
     },
     {
       "epoch": 2.764080346593147,
-      "grad_norm": 0.2084042727947235,
+      "grad_norm": 4.8552985191345215,
       "learning_rate": 1.1986830978696908e-05,
-      "loss": 0.3138,
+      "loss": 7.3518,
       "step": 1755
     },
     {
       "epoch": 2.771957463568334,
-      "grad_norm": 0.20862416923046112,
+      "grad_norm": 5.416152477264404,
       "learning_rate": 1.1848150251933102e-05,
-      "loss": 0.3134,
+      "loss": 7.7477,
       "step": 1760
     },
     {
       "epoch": 2.779834580543521,
-      "grad_norm": 0.19537927210330963,
+      "grad_norm": 4.609814643859863,
       "learning_rate": 1.1710026852595446e-05,
-      "loss": 0.3154,
+      "loss": 7.7182,
       "step": 1765
     },
     {
       "epoch": 2.7877116975187084,
-      "grad_norm": 0.2035813182592392,
+      "grad_norm": 4.770631790161133,
       "learning_rate": 1.157246663384984e-05,
-      "loss": 0.3204,
+      "loss": 7.5309,
       "step": 1770
     },
     {
       "epoch": 2.795588814493895,
-      "grad_norm": 0.20835839211940765,
+      "grad_norm": 5.455694675445557,
       "learning_rate": 1.1435475424996644e-05,
-      "loss": 0.3109,
+      "loss": 7.4731,
       "step": 1775
     },
     {
       "epoch": 2.8034659314690824,
-      "grad_norm": 0.21280139684677124,
+      "grad_norm": 5.270656585693359,
       "learning_rate": 1.1299059031223706e-05,
-      "loss": 0.3127,
+      "loss": 7.6206,
       "step": 1780
     },
     {
       "epoch": 2.8113430484442694,
-      "grad_norm": 0.21605828404426575,
+      "grad_norm": 4.653499126434326,
       "learning_rate": 1.1163223233360243e-05,
-      "loss": 0.3206,
+      "loss": 7.9029,
       "step": 1785
     },
     {
       "epoch": 2.8192201654194564,
-      "grad_norm": 0.20913977921009064,
+      "grad_norm": 5.981932163238525,
       "learning_rate": 1.1027973787631956e-05,
-      "loss": 0.3187,
+      "loss": 7.6331,
       "step": 1790
     },
     {
       "epoch": 2.8270972823946434,
-      "grad_norm": 0.20801085233688354,
+      "grad_norm": 6.355517864227295,
       "learning_rate": 1.0893316425417097e-05,
-      "loss": 0.3181,
+      "loss": 7.6632,
       "step": 1795
     },
     {
       "epoch": 2.8349743993698304,
-      "grad_norm": 0.21318671107292175,
+      "grad_norm": 7.381013870239258,
       "learning_rate": 1.0759256853003578e-05,
-      "loss": 0.3181,
+      "loss": 7.5561,
       "step": 1800
     },
     {
       "epoch": 2.842851516345018,
-      "grad_norm": 0.20348812639713287,
+      "grad_norm": 6.400808811187744,
       "learning_rate": 1.062580075134712e-05,
-      "loss": 0.3131,
+      "loss": 7.1496,
       "step": 1805
     },
     {
       "epoch": 2.850728633320205,
-      "grad_norm": 0.22590206563472748,
+      "grad_norm": 6.144219875335693,
       "learning_rate": 1.0492953775830603e-05,
-      "loss": 0.3182,
+      "loss": 7.6644,
       "step": 1810
     },
     {
       "epoch": 2.858605750295392,
-      "grad_norm": 0.2275339812040329,
+      "grad_norm": 4.925227642059326,
       "learning_rate": 1.0360721556024348e-05,
-      "loss": 0.3245,
+      "loss": 7.9462,
       "step": 1815
     },
     {
       "epoch": 2.866482867270579,
-      "grad_norm": 0.20729248225688934,
+      "grad_norm": 6.179694175720215,
       "learning_rate": 1.0229109695447553e-05,
-      "loss": 0.3113,
+      "loss": 7.6671,
       "step": 1820
     },
     {
       "epoch": 2.874359984245766,
-      "grad_norm": 0.21620720624923706,
+      "grad_norm": 5.008157730102539,
       "learning_rate": 1.0098123771330853e-05,
-      "loss": 0.3142,
+      "loss": 7.9779,
       "step": 1825
     },
     {
       "epoch": 2.8822371012209533,
-      "grad_norm": 0.20153531432151794,
+      "grad_norm": 5.26678466796875,
       "learning_rate": 9.967769334380029e-06,
-      "loss": 0.3136,
+      "loss": 7.4016,
       "step": 1830
     },
     {
       "epoch": 2.8901142181961403,
-      "grad_norm": 0.20845447480678558,
+      "grad_norm": 5.3563151359558105,
       "learning_rate": 9.83805190854068e-06,
-      "loss": 0.3134,
+      "loss": 7.6818,
       "step": 1835
     },
     {
       "epoch": 2.8979913351713273,
-      "grad_norm": 0.20220224559307098,
+      "grad_norm": 4.609752178192139,
       "learning_rate": 9.708976990764212e-06,
-      "loss": 0.317,
+      "loss": 7.826,
       "step": 1840
     },
     {
       "epoch": 2.9058684521465143,
-      "grad_norm": 0.22257234156131744,
+      "grad_norm": 4.820992946624756,
       "learning_rate": 9.580550050774906e-06,
-      "loss": 0.3121,
+      "loss": 7.7669,
       "step": 1845
     },
     {
       "epoch": 2.9137455691217014,
-      "grad_norm": 0.2114904224872589,
+      "grad_norm": 5.011669635772705,
       "learning_rate": 9.452776530838096e-06,
-      "loss": 0.3092,
+      "loss": 7.8166,
       "step": 1850
     },
     {
       "epoch": 2.921622686096889,
-      "grad_norm": 0.20534107089042664,
+      "grad_norm": 5.413022041320801,
       "learning_rate": 9.325661845529533e-06,
-      "loss": 0.3112,
+      "loss": 7.3276,
       "step": 1855
     },
     {
       "epoch": 2.9294998030720754,
-      "grad_norm": 0.21934354305267334,
+      "grad_norm": 5.638709545135498,
       "learning_rate": 9.199211381505993e-06,
-      "loss": 0.3195,
+      "loss": 7.6476,
       "step": 1860
     },
     {
       "epoch": 2.937376920047263,
-      "grad_norm": 0.24163486063480377,
+      "grad_norm": 6.355219841003418,
       "learning_rate": 9.073430497276969e-06,
-      "loss": 0.3135,
+      "loss": 7.5497,
       "step": 1865
     },
     {
       "epoch": 2.94525403702245,
-      "grad_norm": 0.22086267173290253,
+      "grad_norm": 4.824697494506836,
       "learning_rate": 8.94832452297759e-06,
-      "loss": 0.3132,
+      "loss": 7.4419,
       "step": 1870
     },
     {
       "epoch": 2.953131153997637,
-      "grad_norm": 0.20960792899131775,
+      "grad_norm": 10.201951026916504,
       "learning_rate": 8.823898760142759e-06,
-      "loss": 0.3218,
+      "loss": 7.8972,
       "step": 1875
     },
     {
       "epoch": 2.961008270972824,
-      "grad_norm": 0.2152036726474762,
+      "grad_norm": 5.421176433563232,
       "learning_rate": 8.700158481482543e-06,
-      "loss": 0.3179,
+      "loss": 7.5122,
       "step": 1880
     },
     {
       "epoch": 2.968885387948011,
-      "grad_norm": 0.2101101130247116,
+      "grad_norm": 4.769680500030518,
       "learning_rate": 8.577108930658636e-06,
-      "loss": 0.3123,
+      "loss": 7.4867,
       "step": 1885
     },
     {
       "epoch": 2.9767625049231983,
-      "grad_norm": 0.21300119161605835,
+      "grad_norm": 4.7165727615356445,
       "learning_rate": 8.454755322062228e-06,
-      "loss": 0.3119,
+      "loss": 7.9242,
       "step": 1890
     },
     {
       "epoch": 2.9846396218983853,
-      "grad_norm": 0.20476487278938293,
+      "grad_norm": 6.179513931274414,
       "learning_rate": 8.333102840593015e-06,
-      "loss": 0.3139,
+      "loss": 7.585,
       "step": 1895
     },
     {
       "epoch": 2.9925167388735723,
-      "grad_norm": 0.20163965225219727,
+      "grad_norm": 5.120509624481201,
       "learning_rate": 8.212156641439495e-06,
-      "loss": 0.3159,
+      "loss": 7.6742,
       "step": 1900
     },
     {
       "epoch": 3.0015754233950376,
-      "grad_norm": 0.3048245906829834,
+      "grad_norm": 5.389956951141357,
       "learning_rate": 8.091921849860466e-06,
-      "loss": 0.3674,
+      "loss": 8.6246,
       "step": 1905
     },
     {
-      "epoch": 3.0094525403702246,
-      "grad_norm": 0.26477697491645813,
+      "epoch": 3.007877116975187,
+      "grad_norm": 5.342672348022461,
       "learning_rate": 7.972403560967895e-06,
-      "loss": 0.2759,
+      "loss": 8.6445,
       "step": 1910
     },
     {
-      "epoch": 3.0173296573454116,
-      "grad_norm": 0.25195425748825073,
+      "epoch": 3.015754233950374,
+      "grad_norm": 5.166170597076416,
       "learning_rate": 7.853606839510976e-06,
-      "loss": 0.2867,
+      "loss": 8.1686,
       "step": 1915
     },
     {
-      "epoch": 3.0252067743205986,
-      "grad_norm": 0.2241075187921524,
+      "epoch": 3.0236313509255615,
+      "grad_norm": 4.708956241607666,
       "learning_rate": 7.735536719661495e-06,
-      "loss": 0.2845,
+      "loss": 8.1996,
       "step": 1920
     },
     {
-      "epoch": 3.0330838912957856,
-      "grad_norm": 0.2114437222480774,
+      "epoch": 3.0315084679007485,
+      "grad_norm": 5.225474834442139,
       "learning_rate": 7.618198204800509e-06,
-      "loss": 0.2803,
+      "loss": 8.5347,
       "step": 1925
     },
     {
-      "epoch": 3.0409610082709726,
-      "grad_norm": 0.21161282062530518,
+      "epoch": 3.0393855848759355,
+      "grad_norm": 4.633377552032471,
       "learning_rate": 7.5015962673063375e-06,
-      "loss": 0.2791,
+      "loss": 8.4879,
       "step": 1930
     },
     {
-      "epoch": 3.04883812524616,
-      "grad_norm": 0.21086104214191437,
+      "epoch": 3.0472627018511225,
+      "grad_norm": 6.010617256164551,
       "learning_rate": 7.3857358483438385e-06,
-      "loss": 0.2817,
+      "loss": 8.3143,
       "step": 1935
     },
     {
-      "epoch": 3.056715242221347,
-      "grad_norm": 0.21510441601276398,
+      "epoch": 3.0551398188263095,
+      "grad_norm": 5.735445022583008,
       "learning_rate": 7.270621857655002e-06,
-      "loss": 0.285,
+      "loss": 8.2262,
       "step": 1940
     },
     {
-      "epoch": 3.064592359196534,
-      "grad_norm": 0.20668771862983704,
+      "epoch": 3.0630169358014965,
+      "grad_norm": 5.311731815338135,
       "learning_rate": 7.156259173350926e-06,
-      "loss": 0.2789,
+      "loss": 8.3875,
       "step": 1945
     },
     {
-      "epoch": 3.072469476171721,
-      "grad_norm": 0.21624241769313812,
+      "epoch": 3.070894052776684,
+      "grad_norm": 4.930365562438965,
       "learning_rate": 7.042652641705083e-06,
-      "loss": 0.2795,
+      "loss": 8.2483,
       "step": 1950
     },
     {
-      "epoch": 3.080346593146908,
-      "grad_norm": 0.22468064725399017,
+      "epoch": 3.078771169751871,
+      "grad_norm": 4.509944915771484,
       "learning_rate": 6.9298070769479476e-06,
-      "loss": 0.2822,
+      "loss": 8.1581,
       "step": 1955
     },
     {
-      "epoch": 3.0882237101220955,
-      "grad_norm": 0.20948195457458496,
+      "epoch": 3.086648286727058,
+      "grad_norm": 6.027124404907227,
       "learning_rate": 6.817727261062973e-06,
-      "loss": 0.2847,
+      "loss": 8.4293,
       "step": 1960
     },
     {
-      "epoch": 3.0961008270972825,
-      "grad_norm": 0.21020089089870453,
+      "epoch": 3.094525403702245,
+      "grad_norm": 4.761337757110596,
       "learning_rate": 6.706417943584023e-06,
-      "loss": 0.2803,
+      "loss": 8.0434,
       "step": 1965
     },
     {
-      "epoch": 3.1039779440724695,
-      "grad_norm": 0.20508207380771637,
+      "epoch": 3.102402520677432,
+      "grad_norm": 4.544243335723877,
       "learning_rate": 6.5958838413940016e-06,
-      "loss": 0.2902,
+      "loss": 8.3465,
       "step": 1970
     },
     {
-      "epoch": 3.1118550610476565,
-      "grad_norm": 0.21120624244213104,
+      "epoch": 3.110279637652619,
+      "grad_norm": 4.9786882400512695,
       "learning_rate": 6.486129638525021e-06,
-      "loss": 0.2858,
+      "loss": 8.0994,
       "step": 1975
     },
     {
-      "epoch": 3.1197321780228435,
-      "grad_norm": 0.2145949751138687,
+      "epoch": 3.1181567546278064,
+      "grad_norm": 5.5463972091674805,
       "learning_rate": 6.377159985959921e-06,
-      "loss": 0.2805,
+      "loss": 7.9211,
       "step": 1980
     },
     {
-      "epoch": 3.1276092949980305,
-      "grad_norm": 0.20431272685527802,
+      "epoch": 3.1260338716029934,
+      "grad_norm": 4.427308082580566,
       "learning_rate": 6.2689795014351595e-06,
-      "loss": 0.2752,
+      "loss": 8.2482,
       "step": 1985
     },
     {
-      "epoch": 3.135486411973218,
-      "grad_norm": 0.20915834605693817,
+      "epoch": 3.1339109885781804,
+      "grad_norm": 4.5052642822265625,
       "learning_rate": 6.161592769245114e-06,
-      "loss": 0.2774,
+      "loss": 8.1749,
       "step": 1990
     },
     {
-      "epoch": 3.143363528948405,
-      "grad_norm": 0.20422811806201935,
+      "epoch": 3.1417881055533674,
+      "grad_norm": 5.253861904144287,
       "learning_rate": 6.055004340047849e-06,
-      "loss": 0.28,
+      "loss": 8.2287,
       "step": 1995
     },
     {
-      "epoch": 3.151240645923592,
-      "grad_norm": 0.20162755250930786,
+      "epoch": 3.1496652225285544,
+      "grad_norm": 8.412028312683105,
       "learning_rate": 5.949218730672259e-06,
-      "loss": 0.2846,
+      "loss": 7.489,
       "step": 2000
     },
     {
-      "epoch": 3.159117762898779,
-      "grad_norm": 0.2039792239665985,
+      "epoch": 3.157542339503742,
+      "grad_norm": 6.002096652984619,
       "learning_rate": 5.844240423926659e-06,
-      "loss": 0.2826,
+      "loss": 7.7578,
       "step": 2005
     },
     {
-      "epoch": 3.166994879873966,
-      "grad_norm": 0.211518332362175,
+      "epoch": 3.165419456478929,
+      "grad_norm": 5.330885410308838,
       "learning_rate": 5.740073868408799e-06,
-      "loss": 0.2796,
+      "loss": 7.6986,
       "step": 2010
     },
     {
-      "epoch": 3.174871996849153,
-      "grad_norm": 0.21034759283065796,
+      "epoch": 3.173296573454116,
+      "grad_norm": 6.712745666503906,
       "learning_rate": 5.636723478317429e-06,
-      "loss": 0.2747,
+      "loss": 7.765,
       "step": 2015
     },
     {
-      "epoch": 3.1827491138243404,
-      "grad_norm": 0.19976024329662323,
+      "epoch": 3.181173690429303,
+      "grad_norm": 5.10537052154541,
       "learning_rate": 5.5341936332651336e-06,
-      "loss": 0.2794,
+      "loss": 7.9934,
       "step": 2020
     },
     {
-      "epoch": 3.1906262307995275,
-      "grad_norm": 0.19816116988658905,
+      "epoch": 3.18905080740449,
+      "grad_norm": 5.30924129486084,
       "learning_rate": 5.432488678092807e-06,
-      "loss": 0.2743,
+      "loss": 7.8984,
       "step": 2025
     },
     {
-      "epoch": 3.1985033477747145,
-      "grad_norm": 0.20253697037696838,
+      "epoch": 3.196927924379677,
+      "grad_norm": 6.369449138641357,
       "learning_rate": 5.331612922685522e-06,
-      "loss": 0.2787,
+      "loss": 7.7312,
       "step": 2030
     },
     {
-      "epoch": 3.2063804647499015,
-      "grad_norm": 0.2093832790851593,
+      "epoch": 3.2048050413548643,
+      "grad_norm": 4.61285924911499,
       "learning_rate": 5.231570641789899e-06,
-      "loss": 0.2787,
+      "loss": 7.9155,
       "step": 2035
     },
     {
-      "epoch": 3.2142575817250885,
-      "grad_norm": 0.2073364406824112,
+      "epoch": 3.2126821583300513,
+      "grad_norm": 4.817512512207031,
       "learning_rate": 5.13236607483292e-06,
-      "loss": 0.2806,
+      "loss": 7.6799,
       "step": 2040
     },
     {
-      "epoch": 3.222134698700276,
-      "grad_norm": 0.19635286927223206,
+      "epoch": 3.2205592753052383,
+      "grad_norm": 5.9151082038879395,
       "learning_rate": 5.034003425742334e-06,
-      "loss": 0.2788,
+      "loss": 8.2068,
       "step": 2045
     },
     {
-      "epoch": 3.230011815675463,
-      "grad_norm": 0.20712953805923462,
+      "epoch": 3.2284363922804253,
+      "grad_norm": 4.645730972290039,
       "learning_rate": 4.9364868627684775e-06,
-      "loss": 0.2887,
+      "loss": 7.9157,
       "step": 2050
     },
     {
-      "epoch": 3.23788893265065,
-      "grad_norm": 0.2041015326976776,
+      "epoch": 3.2363135092556123,
+      "grad_norm": 4.885422706604004,
       "learning_rate": 4.839820518307628e-06,
-      "loss": 0.2821,
+      "loss": 7.7623,
       "step": 2055
     },
     {
-      "epoch": 3.245766049625837,
-      "grad_norm": 0.19976255297660828,
+      "epoch": 3.2441906262307993,
+      "grad_norm": 5.870239734649658,
       "learning_rate": 4.7440084887269035e-06,
-      "loss": 0.2823,
+      "loss": 8.0045,
       "step": 2060
     },
     {
-      "epoch": 3.253643166601024,
-      "grad_norm": 0.20047256350517273,
+      "epoch": 3.252067743205987,
+      "grad_norm": 4.6275153160095215,
       "learning_rate": 4.649054834190686e-06,
-      "loss": 0.2799,
+      "loss": 8.0402,
       "step": 2065
     },
     {
-      "epoch": 3.261520283576211,
-      "grad_norm": 0.20123493671417236,
+      "epoch": 3.259944860181174,
+      "grad_norm": 5.3590779304504395,
       "learning_rate": 4.554963578488561e-06,
-      "loss": 0.2825,
+      "loss": 7.8615,
       "step": 2070
     },
     {
-      "epoch": 3.2693974005513984,
-      "grad_norm": 0.19577762484550476,
+      "epoch": 3.267821977156361,
+      "grad_norm": 4.911188125610352,
       "learning_rate": 4.461738708864768e-06,
-      "loss": 0.2813,
+      "loss": 7.8886,
       "step": 2075
     },
     {
-      "epoch": 3.2772745175265854,
-      "grad_norm": 0.20611698925495148,
+      "epoch": 3.275699094131548,
+      "grad_norm": 5.0650954246521,
       "learning_rate": 4.369384175849287e-06,
-      "loss": 0.2825,
+      "loss": 8.1302,
       "step": 2080
     },
     {
-      "epoch": 3.2851516345017724,
-      "grad_norm": 0.20026405155658722,
+      "epoch": 3.283576211106735,
+      "grad_norm": 5.238993167877197,
       "learning_rate": 4.277903893090407e-06,
-      "loss": 0.2766,
+      "loss": 7.5243,
       "step": 2085
     },
     {
-      "epoch": 3.2930287514769594,
-      "grad_norm": 0.19801442325115204,
+      "epoch": 3.2914533280819223,
+      "grad_norm": 4.651856899261475,
       "learning_rate": 4.187301737188859e-06,
-      "loss": 0.2756,
+      "loss": 7.3692,
       "step": 2090
     },
     {
-      "epoch": 3.3009058684521464,
-      "grad_norm": 0.19530606269836426,
+      "epoch": 3.2993304450571093,
+      "grad_norm": 4.900179386138916,
       "learning_rate": 4.0975815475335855e-06,
-      "loss": 0.2767,
+      "loss": 7.0708,
       "step": 2095
     },
     {
-      "epoch": 3.3087829854273334,
-      "grad_norm": 0.20650427043437958,
+      "epoch": 3.3072075620322963,
+      "grad_norm": 7.008111953735352,
       "learning_rate": 4.0087471261390094e-06,
-      "loss": 0.2752,
+      "loss": 7.5399,
       "step": 2100
     },
     {
-      "epoch": 3.316660102402521,
-      "grad_norm": 0.20899270474910736,
+      "epoch": 3.3150846790074833,
+      "grad_norm": 4.802496910095215,
       "learning_rate": 3.920802237483912e-06,
-      "loss": 0.2755,
+      "loss": 7.7337,
       "step": 2105
     },
     {
-      "epoch": 3.324537219377708,
-      "grad_norm": 0.21143369376659393,
+      "epoch": 3.3229617959826703,
+      "grad_norm": 5.664666175842285,
       "learning_rate": 3.833750608351924e-06,
-      "loss": 0.2805,
+      "loss": 7.3805,
       "step": 2110
     },
     {
-      "epoch": 3.332414336352895,
-      "grad_norm": 0.20321451127529144,
+      "epoch": 3.3308389129578573,
+      "grad_norm": 5.779867649078369,
       "learning_rate": 3.747595927673611e-06,
-      "loss": 0.2788,
+      "loss": 7.4321,
       "step": 2115
     },
     {
-      "epoch": 3.340291453328082,
-      "grad_norm": 0.1997046321630478,
+      "epoch": 3.3387160299330443,
+      "grad_norm": 4.908834457397461,
       "learning_rate": 3.6623418463701293e-06,
-      "loss": 0.2825,
+      "loss": 7.8338,
       "step": 2120
     },
     {
-      "epoch": 3.348168570303269,
-      "grad_norm": 0.4038223326206207,
+      "epoch": 3.3465931469082317,
+      "grad_norm": 5.189285755157471,
       "learning_rate": 3.5779919771985025e-06,
-      "loss": 0.2767,
+      "loss": 7.2317,
       "step": 2125
     },
     {
-      "epoch": 3.3560456872784563,
-      "grad_norm": 0.20923250913619995,
+      "epoch": 3.3544702638834187,
+      "grad_norm": 4.881292819976807,
       "learning_rate": 3.494549894598556e-06,
-      "loss": 0.28,
+      "loss": 7.1418,
       "step": 2130
     },
     {
-      "epoch": 3.3639228042536433,
-      "grad_norm": 0.1994774043560028,
+      "epoch": 3.3623473808586057,
+      "grad_norm": 4.643113613128662,
       "learning_rate": 3.4120191345414415e-06,
-      "loss": 0.2842,
+      "loss": 7.9551,
       "step": 2135
     },
     {
-      "epoch": 3.3717999212288303,
-      "grad_norm": 0.20118853449821472,
+      "epoch": 3.3702244978337927,
+      "grad_norm": 6.662186622619629,
       "learning_rate": 3.3304031943797527e-06,
-      "loss": 0.281,
+      "loss": 7.9614,
       "step": 2140
     },
     {
-      "epoch": 3.3796770382040173,
-      "grad_norm": 0.21133875846862793,
+      "epoch": 3.3781016148089797,
+      "grad_norm": 4.886229038238525,
       "learning_rate": 3.249705532699379e-06,
-      "loss": 0.2821,
+      "loss": 7.7298,
       "step": 2145
     },
     {
-      "epoch": 3.3875541551792043,
-      "grad_norm": 0.19587914645671844,
+      "epoch": 3.385978731784167,
+      "grad_norm": 6.290960788726807,
       "learning_rate": 3.169929569172919e-06,
-      "loss": 0.2768,
+      "loss": 7.4452,
       "step": 2150
     },
     {
-      "epoch": 3.3954312721543913,
-      "grad_norm": 0.20176377892494202,
+      "epoch": 3.393855848759354,
+      "grad_norm": 5.830533981323242,
       "learning_rate": 3.0910786844147473e-06,
-      "loss": 0.2765,
+      "loss": 7.6276,
       "step": 2155
     },
     {
-      "epoch": 3.4033083891295783,
-      "grad_norm": 0.19805395603179932,
+      "epoch": 3.401732965734541,
+      "grad_norm": 4.647792339324951,
       "learning_rate": 3.013156219837776e-06,
-      "loss": 0.277,
+      "loss": 7.7677,
       "step": 2160
     },
     {
-      "epoch": 3.411185506104766,
-      "grad_norm": 0.2078244537115097,
+      "epoch": 3.409610082709728,
+      "grad_norm": 4.803062915802002,
       "learning_rate": 2.936165477511868e-06,
-      "loss": 0.2784,
+      "loss": 7.9931,
       "step": 2165
     },
     {
-      "epoch": 3.419062623079953,
-      "grad_norm": 0.199508935213089,
+      "epoch": 3.417487199684915,
+      "grad_norm": 5.041831970214844,
       "learning_rate": 2.8601097200238987e-06,
-      "loss": 0.2798,
+      "loss": 7.7018,
       "step": 2170
     },
     {
-      "epoch": 3.42693974005514,
-      "grad_norm": 0.2023046910762787,
+      "epoch": 3.425364316660102,
+      "grad_norm": 5.187193393707275,
       "learning_rate": 2.7849921703394732e-06,
-      "loss": 0.2833,
+      "loss": 7.4835,
       "step": 2175
     },
     {
-      "epoch": 3.434816857030327,
-      "grad_norm": 0.20601923763751984,
+      "epoch": 3.4332414336352897,
+      "grad_norm": 5.092511177062988,
       "learning_rate": 2.7108160116663893e-06,
-      "loss": 0.2811,
+      "loss": 7.4102,
       "step": 2180
     },
     {
-      "epoch": 3.442693974005514,
-      "grad_norm": 0.20845656096935272,
+      "epoch": 3.4411185506104767,
+      "grad_norm": 4.9641594886779785,
       "learning_rate": 2.637584387319733e-06,
-      "loss": 0.2844,
+      "loss": 8.2661,
       "step": 2185
     },
     {
-      "epoch": 3.4505710909807012,
-      "grad_norm": 0.19674457609653473,
+      "epoch": 3.4489956675856637,
+      "grad_norm": 4.644840240478516,
       "learning_rate": 2.565300400588647e-06,
-      "loss": 0.2846,
+      "loss": 7.5782,
       "step": 2190
     },
     {
-      "epoch": 3.4584482079558883,
-      "grad_norm": 0.197424054145813,
+      "epoch": 3.4568727845608507,
+      "grad_norm": 4.720969200134277,
       "learning_rate": 2.4939671146048727e-06,
-      "loss": 0.2792,
+      "loss": 7.7751,
       "step": 2195
     },
     {
-      "epoch": 3.4663253249310753,
-      "grad_norm": 0.19904725253582,
+      "epoch": 3.4647499015360377,
+      "grad_norm": 4.5188889503479,
       "learning_rate": 2.4235875522128997e-06,
-      "loss": 0.2903,
+      "loss": 7.6633,
       "step": 2200
     },
     {
-      "epoch": 3.4742024419062623,
-      "grad_norm": 0.19490273296833038,
+      "epoch": 3.4726270185112247,
+      "grad_norm": 4.929895877838135,
       "learning_rate": 2.3541646958419094e-06,
-      "loss": 0.2799,
+      "loss": 8.0803,
       "step": 2205
     },
     {
-      "epoch": 3.4820795588814493,
-      "grad_norm": 0.19701102375984192,
+      "epoch": 3.480504135486412,
+      "grad_norm": 4.896505832672119,
       "learning_rate": 2.2857014873793577e-06,
-      "loss": 0.2815,
+      "loss": 7.508,
       "step": 2210
     },
     {
-      "epoch": 3.4899566758566367,
-      "grad_norm": 0.1960253268480301,
+      "epoch": 3.488381252461599,
+      "grad_norm": 5.505984783172607,
       "learning_rate": 2.218200828046324e-06,
-      "loss": 0.2767,
+      "loss": 7.1282,
       "step": 2215
     },
     {
-      "epoch": 3.4978337928318237,
-      "grad_norm": 0.1989275962114334,
+      "epoch": 3.496258369436786,
+      "grad_norm": 5.277995586395264,
       "learning_rate": 2.151665578274581e-06,
-      "loss": 0.2795,
+      "loss": 7.9669,
       "step": 2220
     },
     {
-      "epoch": 3.5057109098070107,
-      "grad_norm": 0.19489678740501404,
+      "epoch": 3.504135486411973,
+      "grad_norm": 7.832403182983398,
       "learning_rate": 2.0860985575853366e-06,
-      "loss": 0.2816,
+      "loss": 8.0894,
       "step": 2225
     },
     {
-      "epoch": 3.5135880267821977,
-      "grad_norm": 0.20096373558044434,
+      "epoch": 3.51201260338716,
+      "grad_norm": 5.209780216217041,
       "learning_rate": 2.021502544469808e-06,
-      "loss": 0.2879,
+      "loss": 7.4866,
       "step": 2230
     },
     {
-      "epoch": 3.5214651437573847,
-      "grad_norm": 0.19433267414569855,
+      "epoch": 3.5198897203623476,
+      "grad_norm": 5.2757697105407715,
       "learning_rate": 1.9578802762714427e-06,
-      "loss": 0.2814,
+      "loss": 7.7545,
       "step": 2235
     },
     {
-      "epoch": 3.5293422607325717,
-      "grad_norm": 0.20079940557479858,
+      "epoch": 3.5277668373375346,
+      "grad_norm": 4.831712245941162,
       "learning_rate": 1.8952344490699243e-06,
-      "loss": 0.2805,
+      "loss": 7.9045,
       "step": 2240
     },
     {
-      "epoch": 3.5372193777077587,
-      "grad_norm": 0.19769088923931122,
+      "epoch": 3.5356439543127216,
+      "grad_norm": 4.441675662994385,
       "learning_rate": 1.833567717566939e-06,
-      "loss": 0.2833,
+      "loss": 7.379,
       "step": 2245
     },
     {
-      "epoch": 3.545096494682946,
-      "grad_norm": 0.20594307780265808,
+      "epoch": 3.5435210712879086,
+      "grad_norm": 5.417379856109619,
       "learning_rate": 1.7728826949736598e-06,
-      "loss": 0.2882,
+      "loss": 7.6097,
       "step": 2250
     },
     {
-      "epoch": 3.552973611658133,
-      "grad_norm": 0.19729258120059967,
+      "epoch": 3.5513981882630956,
+      "grad_norm": 4.504559516906738,
       "learning_rate": 1.7131819529000226e-06,
-      "loss": 0.2829,
+      "loss": 7.2856,
       "step": 2255
     },
     {
-      "epoch": 3.56085072863332,
-      "grad_norm": 0.19826674461364746,
+      "epoch": 3.559275305238283,
+      "grad_norm": 4.984797954559326,
       "learning_rate": 1.6544680212457337e-06,
-      "loss": 0.2786,
+      "loss": 7.194,
       "step": 2260
     },
     {
-      "epoch": 3.568727845608507,
-      "grad_norm": 0.19805769622325897,
+      "epoch": 3.5671524222134696,
+      "grad_norm": 4.818814754486084,
       "learning_rate": 1.5967433880930889e-06,
-      "loss": 0.2808,
+      "loss": 7.051,
       "step": 2265
     },
     {
-      "epoch": 3.576604962583694,
-      "grad_norm": 0.20012123882770538,
+      "epoch": 3.575029539188657,
+      "grad_norm": 4.872919082641602,
       "learning_rate": 1.540010499601513e-06,
-      "loss": 0.2825,
+      "loss": 7.1664,
       "step": 2270
     },
     {
-      "epoch": 3.5844820795588817,
-      "grad_norm": 0.20687070488929749,
+      "epoch": 3.582906656163844,
+      "grad_norm": 6.4378862380981445,
       "learning_rate": 1.4842717599039047e-06,
-      "loss": 0.2848,
+      "loss": 7.3849,
       "step": 2275
     },
     {
-      "epoch": 3.5923591965340687,
-      "grad_norm": 0.19811828434467316,
+      "epoch": 3.590783773139031,
+      "grad_norm": 5.795922756195068,
       "learning_rate": 1.4295295310047668e-06,
-      "loss": 0.2785,
+      "loss": 7.4473,
       "step": 2280
     },
     {
-      "epoch": 3.6002363135092557,
-      "grad_norm": 0.19836102426052094,
+      "epoch": 3.598660890114218,
+      "grad_norm": 4.69597053527832,
       "learning_rate": 1.375786132680118e-06,
-      "loss": 0.2808,
+      "loss": 7.3833,
       "step": 2285
     },
     {
-      "epoch": 3.6081134304844427,
-      "grad_norm": 0.19285066425800323,
+      "epoch": 3.606538007089405,
+      "grad_norm": 5.1728901863098145,
       "learning_rate": 1.3230438423791625e-06,
-      "loss": 0.2818,
+      "loss": 7.0266,
       "step": 2290
     },
     {
-      "epoch": 3.6159905474596297,
-      "grad_norm": 0.19814734160900116,
+      "epoch": 3.6144151240645925,
+      "grad_norm": 4.9159440994262695,
       "learning_rate": 1.2713048951278144e-06,
-      "loss": 0.2817,
+      "loss": 7.3836,
       "step": 2295
     },
     {
-      "epoch": 3.623867664434817,
-      "grad_norm": 0.2252984642982483,
+      "epoch": 3.6222922410397795,
+      "grad_norm": 5.896880149841309,
       "learning_rate": 1.2205714834339521e-06,
-      "loss": 0.2822,
+      "loss": 7.517,
       "step": 2300
     },
     {
-      "epoch": 3.6317447814100037,
-      "grad_norm": 0.19117222726345062,
+      "epoch": 3.6301693580149665,
+      "grad_norm": 4.930107593536377,
       "learning_rate": 1.1708457571945386e-06,
-      "loss": 0.2753,
+      "loss": 7.6201,
       "step": 2305
     },
     {
-      "epoch": 3.639621898385191,
-      "grad_norm": 0.19470417499542236,
+      "epoch": 3.6380464749901535,
+      "grad_norm": 4.929466724395752,
       "learning_rate": 1.1221298236044953e-06,
-      "loss": 0.2822,
+      "loss": 7.285,
       "step": 2310
     },
     {
-      "epoch": 3.647499015360378,
-      "grad_norm": 0.2041274756193161,
+      "epoch": 3.6459235919653405,
+      "grad_norm": 5.38197135925293,
       "learning_rate": 1.0744257470674147e-06,
-      "loss": 0.2841,
+      "loss": 7.2539,
       "step": 2315
     },
     {
-      "epoch": 3.655376132335565,
-      "grad_norm": 0.19451037049293518,
+      "epoch": 3.653800708940528,
+      "grad_norm": 5.5960469245910645,
       "learning_rate": 1.0277355491080804e-06,
-      "loss": 0.2817,
+      "loss": 7.4543,
       "step": 2320
     },
     {
-      "epoch": 3.663253249310752,
-      "grad_norm": 0.1978326439857483,
+      "epoch": 3.661677825915715,
+      "grad_norm": 5.1494245529174805,
       "learning_rate": 9.820612082867969e-07,
-      "loss": 0.2797,
+      "loss": 7.4204,
       "step": 2325
     },
     {
-      "epoch": 3.671130366285939,
-      "grad_norm": 0.19272668659687042,
+      "epoch": 3.669554942890902,
+      "grad_norm": 5.0791015625,
       "learning_rate": 9.374046601155539e-07,
-      "loss": 0.2822,
+      "loss": 7.2228,
       "step": 2330
     },
     {
-      "epoch": 3.6790074832611266,
-      "grad_norm": 0.1939852088689804,
+      "epoch": 3.677432059866089,
+      "grad_norm": 4.948309421539307,
       "learning_rate": 8.937677969759861e-07,
-      "loss": 0.2792,
+      "loss": 7.5265,
       "step": 2335
     },
     {
-      "epoch": 3.6868846002363136,
-      "grad_norm": 0.1928287148475647,
+      "epoch": 3.685309176841276,
+      "grad_norm": 5.31011438369751,
       "learning_rate": 8.511524680392096e-07,
-      "loss": 0.2876,
+      "loss": 7.5065,
       "step": 2340
     },
     {
-      "epoch": 3.6947617172115006,
-      "grad_norm": 0.19896940886974335,
+      "epoch": 3.693186293816463,
+      "grad_norm": 4.728499889373779,
       "learning_rate": 8.09560479187449e-07,
-      "loss": 0.2744,
+      "loss": 7.6349,
       "step": 2345
     },
     {
-      "epoch": 3.7026388341866876,
-      "grad_norm": 0.19617894291877747,
+      "epoch": 3.70106341079165,
+      "grad_norm": 4.810965538024902,
       "learning_rate": 7.689935929374892e-07,
-      "loss": 0.2856,
+      "loss": 7.2619,
       "step": 2350
     },
     {
-      "epoch": 3.7105159511618746,
-      "grad_norm": 0.19769024848937988,
+      "epoch": 3.7089405277668375,
+      "grad_norm": 5.238432884216309,
       "learning_rate": 7.294535283660281e-07,
-      "loss": 0.279,
+      "loss": 7.1741,
       "step": 2355
     },
     {
-      "epoch": 3.718393068137062,
-      "grad_norm": 0.19515487551689148,
+      "epoch": 3.7168176447420245,
+      "grad_norm": 4.824338436126709,
       "learning_rate": 6.909419610367896e-07,
-      "loss": 0.2768,
+      "loss": 7.5995,
       "step": 2360
     },
     {
-      "epoch": 3.726270185112249,
-      "grad_norm": 0.19522498548030853,
+      "epoch": 3.7246947617172115,
+      "grad_norm": 5.660040855407715,
       "learning_rate": 6.534605229295415e-07,
-      "loss": 0.2839,
+      "loss": 7.513,
       "step": 2365
     },
     {
-      "epoch": 3.734147302087436,
-      "grad_norm": 0.18874140083789825,
+      "epoch": 3.7325718786923985,
+      "grad_norm": 5.301452159881592,
       "learning_rate": 6.170108023709348e-07,
-      "loss": 0.2818,
+      "loss": 7.3567,
       "step": 2370
     },
     {
-      "epoch": 3.742024419062623,
-      "grad_norm": 0.19244030117988586,
+      "epoch": 3.7404489956675855,
+      "grad_norm": 5.379169940948486,
       "learning_rate": 5.815943439671901e-07,
-      "loss": 0.2779,
+      "loss": 6.9589,
       "step": 2375
     },
     {
-      "epoch": 3.74990153603781,
-      "grad_norm": 0.190862774848938,
+      "epoch": 3.748326112642773,
+      "grad_norm": 5.965606212615967,
       "learning_rate": 5.472126485386475e-07,
-      "loss": 0.2784,
+      "loss": 7.6638,
       "step": 2380
     },
     {
-      "epoch": 3.7577786530129975,
-      "grad_norm": 0.197196364402771,
+      "epoch": 3.75620322961796,
+      "grad_norm": 4.746719837188721,
       "learning_rate": 5.138671730561623e-07,
-      "loss": 0.2803,
+      "loss": 7.3243,
       "step": 2385
     },
     {
-      "epoch": 3.765655769988184,
-      "grad_norm": 0.19586575031280518,
+      "epoch": 3.764080346593147,
+      "grad_norm": 4.829946994781494,
       "learning_rate": 4.815593305793703e-07,
-      "loss": 0.2776,
+      "loss": 7.6804,
       "step": 2390
     },
     {
-      "epoch": 3.7735328869633715,
-      "grad_norm": 0.19647184014320374,
+      "epoch": 3.771957463568334,
+      "grad_norm": 6.1915669441223145,
       "learning_rate": 4.502904901968141e-07,
-      "loss": 0.2824,
+      "loss": 7.2188,
       "step": 2395
     },
     {
-      "epoch": 3.7814100039385585,
-      "grad_norm": 0.20829348266124725,
+      "epoch": 3.779834580543521,
+      "grad_norm": 4.6367998123168945,
       "learning_rate": 4.200619769679004e-07,
-      "loss": 0.2821,
+      "loss": 6.9212,
       "step": 2400
     },
     {
-      "epoch": 3.7892871209137455,
-      "grad_norm": 0.19623522460460663,
+      "epoch": 3.7877116975187084,
+      "grad_norm": 5.0970611572265625,
       "learning_rate": 3.908750718667753e-07,
-      "loss": 0.2816,
+      "loss": 7.3765,
       "step": 2405
     },
     {
-      "epoch": 3.7971642378889325,
-      "grad_norm": 0.19415368139743805,
+      "epoch": 3.795588814493895,
+      "grad_norm": 5.42421817779541,
       "learning_rate": 3.6273101172802914e-07,
-      "loss": 0.2775,
+      "loss": 7.402,
       "step": 2410
     },
     {
-      "epoch": 3.8050413548641195,
-      "grad_norm": 0.1985977739095688,
+      "epoch": 3.8034659314690824,
+      "grad_norm": 5.175075054168701,
       "learning_rate": 3.3563098919428827e-07,
-      "loss": 0.2787,
+      "loss": 7.053,
       "step": 2415
     },
     {
-      "epoch": 3.812918471839307,
-      "grad_norm": 0.19479507207870483,
+      "epoch": 3.8113430484442694,
+      "grad_norm": 4.7831854820251465,
       "learning_rate": 3.095761526656693e-07,
-      "loss": 0.2778,
+      "loss": 7.9933,
       "step": 2420
     },
     {
-      "epoch": 3.820795588814494,
-      "grad_norm": 0.2967446446418762,
+      "epoch": 3.8192201654194564,
+      "grad_norm": 6.312967300415039,
       "learning_rate": 2.8456760625111823e-07,
-      "loss": 0.2887,
+      "loss": 7.4616,
       "step": 2425
     },
     {
-      "epoch": 3.828672705789681,
-      "grad_norm": 0.19328995048999786,
+      "epoch": 3.8270972823946434,
+      "grad_norm": 5.104128360748291,
       "learning_rate": 2.6060640972162543e-07,
-      "loss": 0.2822,
+      "loss": 7.5123,
       "step": 2430
     },
     {
-      "epoch": 3.836549822764868,
-      "grad_norm": 0.19140738248825073,
+      "epoch": 3.8349743993698304,
+      "grad_norm": 5.383700847625732,
       "learning_rate": 2.3769357846531171e-07,
-      "loss": 0.2771,
+      "loss": 6.8885,
       "step": 2435
     },
     {
-      "epoch": 3.844426939740055,
-      "grad_norm": 0.194104865193367,
+      "epoch": 3.842851516345018,
+      "grad_norm": 4.860473155975342,
       "learning_rate": 2.1583008344439603e-07,
-      "loss": 0.284,
+      "loss": 7.0768,
       "step": 2440
     },
     {
-      "epoch": 3.8523040567152425,
-      "grad_norm": 0.19096589088439941,
+      "epoch": 3.850728633320205,
+      "grad_norm": 5.095125675201416,
       "learning_rate": 1.9501685115406176e-07,
-      "loss": 0.2793,
+      "loss": 7.6519,
       "step": 2445
     },
     {
-      "epoch": 3.8601811736904295,
-      "grad_norm": 0.19132515788078308,
+      "epoch": 3.858605750295392,
+      "grad_norm": 5.066483497619629,
       "learning_rate": 1.7525476358318526e-07,
-      "loss": 0.2783,
+      "loss": 7.3912,
       "step": 2450
     },
     {
-      "epoch": 3.8680582906656165,
-      "grad_norm": 0.20566198229789734,
+      "epoch": 3.866482867270579,
+      "grad_norm": 6.104281425476074,
       "learning_rate": 1.5654465817696584e-07,
-      "loss": 0.2791,
+      "loss": 6.9311,
       "step": 2455
     },
     {
-      "epoch": 3.8759354076408035,
-      "grad_norm": 0.20506292581558228,
+      "epoch": 3.874359984245766,
+      "grad_norm": 4.735925674438477,
       "learning_rate": 1.3888732780142632e-07,
-      "loss": 0.2832,
+      "loss": 7.2688,
       "step": 2460
     },
     {
-      "epoch": 3.8838125246159905,
-      "grad_norm": 0.19119083881378174,
+      "epoch": 3.8822371012209533,
+      "grad_norm": 4.991415500640869,
       "learning_rate": 1.2228352070983719e-07,
-      "loss": 0.2808,
+      "loss": 6.8569,
       "step": 2465
     },
     {
-      "epoch": 3.8916896415911775,
-      "grad_norm": 0.19219276309013367,
+      "epoch": 3.8901142181961403,
+      "grad_norm": 5.310063362121582,
       "learning_rate": 1.0673394051098074e-07,
-      "loss": 0.2767,
+      "loss": 7.4524,
       "step": 2470
     },
     {
-      "epoch": 3.8995667585663645,
-      "grad_norm": 0.1935568004846573,
+      "epoch": 3.8979913351713273,
+      "grad_norm": 6.486708641052246,
       "learning_rate": 9.223924613935842e-08,
-      "loss": 0.2822,
+      "loss": 7.5994,
       "step": 2475
     },
     {
-      "epoch": 3.907443875541552,
-      "grad_norm": 0.19069112837314606,
+      "epoch": 3.9058684521465143,
+      "grad_norm": 4.825411319732666,
       "learning_rate": 7.880005182725469e-08,
-      "loss": 0.2862,
+      "loss": 7.0521,
       "step": 2480
     },
     {
-      "epoch": 3.915320992516739,
-      "grad_norm": 0.19846147298812866,
+      "epoch": 3.9137455691217014,
+      "grad_norm": 4.599216461181641,
       "learning_rate": 6.641692707870795e-08,
-      "loss": 0.2863,
+      "loss": 7.1754,
       "step": 2485
     },
     {
-      "epoch": 3.923198109491926,
-      "grad_norm": 0.19378872215747833,
+      "epoch": 3.921622686096889,
+      "grad_norm": 4.610249042510986,
       "learning_rate": 5.509039664538251e-08,
-      "loss": 0.2827,
+      "loss": 7.6455,
       "step": 2490
     },
     {
-      "epoch": 3.931075226467113,
-      "grad_norm": 0.19526971876621246,
+      "epoch": 3.9294998030720754,
+      "grad_norm": 5.008790016174316,
       "learning_rate": 4.482094050433372e-08,
-      "loss": 0.2843,
+      "loss": 7.7298,
       "step": 2495
     },
     {
-      "epoch": 3.9389523434423,
-      "grad_norm": 0.32608553767204285,
+      "epoch": 3.937376920047263,
+      "grad_norm": 5.213418006896973,
       "learning_rate": 3.5608993837651924e-08,
-      "loss": 0.2863,
+      "loss": 7.2896,
       "step": 2500
     },
     {
-      "epoch": 3.9468294604174874,
-      "grad_norm": 0.1961493343114853,
+      "epoch": 3.94525403702245,
+      "grad_norm": 4.882436275482178,
       "learning_rate": 2.745494701404394e-08,
-      "loss": 0.2796,
+      "loss": 7.3404,
       "step": 2505
     },
     {
-      "epoch": 3.9547065773926744,
-      "grad_norm": 0.19112332165241241,
+      "epoch": 3.953131153997637,
+      "grad_norm": 4.545849800109863,
       "learning_rate": 2.0359145572276804e-08,
-      "loss": 0.2821,
+      "loss": 7.5327,
       "step": 2510
     },
     {
-      "epoch": 3.9625836943678614,
-      "grad_norm": 0.19307857751846313,
+      "epoch": 3.961008270972824,
+      "grad_norm": 5.91042423248291,
       "learning_rate": 1.4321890206533961e-08,
-      "loss": 0.2815,
+      "loss": 7.468,
       "step": 2515
     },
     {
-      "epoch": 3.9704608113430484,
-      "grad_norm": 0.21154381334781647,
+      "epoch": 3.968885387948011,
+      "grad_norm": 5.8342766761779785,
       "learning_rate": 9.34343675368099e-09,
-      "loss": 0.2798,
+      "loss": 7.6057,
       "step": 2520
     },
     {
-      "epoch": 3.9783379283182354,
-      "grad_norm": 0.19480980932712555,
+      "epoch": 3.9767625049231983,
+      "grad_norm": 5.667447566986084,
       "learning_rate": 5.423996182415958e-09,
-      "loss": 0.2792,
+      "loss": 7.2114,
       "step": 2525
     },
     {
-      "epoch": 3.986215045293423,
-      "grad_norm": 0.1891375333070755,
+      "epoch": 3.9846396218983853,
+      "grad_norm": 5.622943878173828,
       "learning_rate": 2.5637345843348982e-09,
-      "loss": 0.2791,
+      "loss": 7.2359,
       "step": 2530
     },
     {
-      "epoch": 3.9940921622686094,
-      "grad_norm": 0.19509141147136688,
+      "epoch": 3.9925167388735723,
+      "grad_norm": 4.587869644165039,
       "learning_rate": 7.627731668985449e-10,
-      "loss": 0.2782,
+      "loss": 7.1989,
       "step": 2535
     },
     {
-      "epoch": 4.0,
+      "epoch": 3.9988184324537217,
       "step": 2539,
-      "total_flos": 4.3530238414695694e+19,
-      "train_loss": 0.15123671168837785,
-      "train_runtime": 82201.3053,
-      "train_samples_per_second": 1.977,
-      "train_steps_per_second": 0.031
+      "total_flos": 4.3565600076592054e+19,
+      "train_loss": 1.895302517482642,
+      "train_runtime": 61070.1581,
+      "train_samples_per_second": 2.661,
+      "train_steps_per_second": 0.042
     }
   ],
   "logging_steps": 5,
   "max_steps": 2540,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 4,
-  "save_steps": 318,
+  "save_steps": 159,
   "stateful_callbacks": {
     "TrainerControl": {
       "args": {
@@ -3585,7 +3585,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 4.3530238414695694e+19,
+  "total_flos": 4.3565600076592054e+19,
   "train_batch_size": 2,
   "trial_name": null,
   "trial_params": null