diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,3948 @@
+{
+  "best_global_step": 5574,
+  "best_metric": 1.932178020477295,
+  "best_model_checkpoint": "./best_mcqa_model/checkpoint-5574",
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 5574,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0001794043774668102,
+      "grad_norm": Infinity,
+      "learning_rate": 5e-05,
+      "loss": 2.0537,
+      "step": 1
+    },
+    {
+      "epoch": 0.001794043774668102,
+      "grad_norm": 50.121803283691406,
+      "learning_rate": 4.9994617868676e-05,
+      "loss": 2.3674,
+      "step": 10
+    },
+    {
+      "epoch": 0.003588087549336204,
+      "grad_norm": 25.215618133544922,
+      "learning_rate": 4.997667743092932e-05,
+      "loss": 2.2324,
+      "step": 20
+    },
+    {
+      "epoch": 0.005382131324004306,
+      "grad_norm": 17.250703811645508,
+      "learning_rate": 4.995873699318264e-05,
+      "loss": 2.1936,
+      "step": 30
+    },
+    {
+      "epoch": 0.007176175098672408,
+      "grad_norm": 16.687049865722656,
+      "learning_rate": 4.994079655543596e-05,
+      "loss": 2.1778,
+      "step": 40
+    },
+    {
+      "epoch": 0.00897021887334051,
+      "grad_norm": 18.19718360900879,
+      "learning_rate": 4.9922856117689274e-05,
+      "loss": 1.9997,
+      "step": 50
+    },
+    {
+      "epoch": 0.010764262648008612,
+      "grad_norm": 18.301307678222656,
+      "learning_rate": 4.990491567994259e-05,
+      "loss": 2.1654,
+      "step": 60
+    },
+    {
+      "epoch": 0.012558306422676713,
+      "grad_norm": 18.938064575195312,
+      "learning_rate": 4.988697524219591e-05,
+      "loss": 2.1311,
+      "step": 70
+    },
+    {
+      "epoch": 0.014352350197344816,
+      "grad_norm": 24.058290481567383,
+      "learning_rate": 4.986903480444923e-05,
+      "loss": 2.1101,
+      "step": 80
+    },
+    {
+      "epoch": 0.016146393972012917,
+      "grad_norm": 20.780832290649414,
+      "learning_rate": 4.985109436670255e-05,
+      "loss": 2.043,
+      "step": 90
+    },
+    {
+      "epoch": 0.01794043774668102,
+      "grad_norm": 18.546377182006836,
+      "learning_rate": 4.983315392895587e-05,
+      "loss": 2.1232,
+      "step": 100
+    },
+    {
+      "epoch": 0.019734481521349122,
+      "grad_norm": 15.411616325378418,
+      "learning_rate": 4.981521349120919e-05,
+      "loss": 2.1469,
+      "step": 110
+    },
+    {
+      "epoch": 0.021528525296017224,
+      "grad_norm": 20.646089553833008,
+      "learning_rate": 4.979727305346251e-05,
+      "loss": 2.2586,
+      "step": 120
+    },
+    {
+      "epoch": 0.023322569070685324,
+      "grad_norm": 17.679092407226562,
+      "learning_rate": 4.9779332615715826e-05,
+      "loss": 2.0833,
+      "step": 130
+    },
+    {
+      "epoch": 0.025116612845353426,
+      "grad_norm": 16.85576629638672,
+      "learning_rate": 4.976139217796914e-05,
+      "loss": 2.0967,
+      "step": 140
+    },
+    {
+      "epoch": 0.02691065662002153,
+      "grad_norm": 13.570154190063477,
+      "learning_rate": 4.974345174022246e-05,
+      "loss": 2.1794,
+      "step": 150
+    },
+    {
+      "epoch": 0.02870470039468963,
+      "grad_norm": 20.535985946655273,
+      "learning_rate": 4.972551130247578e-05,
+      "loss": 2.1686,
+      "step": 160
+    },
+    {
+      "epoch": 0.030498744169357734,
+      "grad_norm": 16.294347763061523,
+      "learning_rate": 4.97075708647291e-05,
+      "loss": 2.0392,
+      "step": 170
+    },
+    {
+      "epoch": 0.03229278794402583,
+      "grad_norm": 12.76968765258789,
+      "learning_rate": 4.968963042698242e-05,
+      "loss": 2.1039,
+      "step": 180
+    },
+    {
+      "epoch": 0.034086831718693936,
+      "grad_norm": 14.646772384643555,
+      "learning_rate": 4.967168998923574e-05,
+      "loss": 2.0321,
+      "step": 190
+    },
+    {
+      "epoch": 0.03588087549336204,
+      "grad_norm": 13.305708885192871,
+      "learning_rate": 4.965374955148906e-05,
+      "loss": 2.1168,
+      "step": 200
+    },
+    {
+      "epoch": 0.03767491926803014,
+      "grad_norm": 14.862943649291992,
+      "learning_rate": 4.963580911374238e-05,
+      "loss": 2.0901,
+      "step": 210
+    },
+    {
+      "epoch": 0.039468963042698244,
+      "grad_norm": 15.443259239196777,
+      "learning_rate": 4.96178686759957e-05,
+      "loss": 2.0921,
+      "step": 220
+    },
+    {
+      "epoch": 0.041263006817366346,
+      "grad_norm": 15.883851051330566,
+      "learning_rate": 4.959992823824902e-05,
+      "loss": 2.2321,
+      "step": 230
+    },
+    {
+      "epoch": 0.04305705059203445,
+      "grad_norm": 12.791455268859863,
+      "learning_rate": 4.9581987800502336e-05,
+      "loss": 2.0071,
+      "step": 240
+    },
+    {
+      "epoch": 0.044851094366702544,
+      "grad_norm": 15.706707954406738,
+      "learning_rate": 4.956404736275565e-05,
+      "loss": 2.1264,
+      "step": 250
+    },
+    {
+      "epoch": 0.04664513814137065,
+      "grad_norm": 18.551183700561523,
+      "learning_rate": 4.954610692500897e-05,
+      "loss": 1.8886,
+      "step": 260
+    },
+    {
+      "epoch": 0.04843918191603875,
+      "grad_norm": 24.170883178710938,
+      "learning_rate": 4.952816648726229e-05,
+      "loss": 2.2594,
+      "step": 270
+    },
+    {
+      "epoch": 0.05023322569070685,
+      "grad_norm": 12.178893089294434,
+      "learning_rate": 4.951022604951561e-05,
+      "loss": 2.0534,
+      "step": 280
+    },
+    {
+      "epoch": 0.052027269465374955,
+      "grad_norm": 14.537866592407227,
+      "learning_rate": 4.949228561176893e-05,
+      "loss": 1.9462,
+      "step": 290
+    },
+    {
+      "epoch": 0.05382131324004306,
+      "grad_norm": 15.807729721069336,
+      "learning_rate": 4.9474345174022246e-05,
+      "loss": 2.0243,
+      "step": 300
+    },
+    {
+      "epoch": 0.05561535701471116,
+      "grad_norm": 14.98117446899414,
+      "learning_rate": 4.945640473627557e-05,
+      "loss": 2.0688,
+      "step": 310
+    },
+    {
+      "epoch": 0.05740940078937926,
+      "grad_norm": 17.541954040527344,
+      "learning_rate": 4.943846429852889e-05,
+      "loss": 1.9925,
+      "step": 320
+    },
+    {
+      "epoch": 0.059203444564047365,
+      "grad_norm": 9.95796012878418,
+      "learning_rate": 4.9420523860782205e-05,
+      "loss": 1.9715,
+      "step": 330
+    },
+    {
+      "epoch": 0.06099748833871547,
+      "grad_norm": 12.977546691894531,
+      "learning_rate": 4.940258342303553e-05,
+      "loss": 2.0491,
+      "step": 340
+    },
+    {
+      "epoch": 0.06279153211338356,
+      "grad_norm": 15.887458801269531,
+      "learning_rate": 4.938464298528884e-05,
+      "loss": 2.1484,
+      "step": 350
+    },
+    {
+      "epoch": 0.06458557588805167,
+      "grad_norm": 11.04765510559082,
+      "learning_rate": 4.9366702547542157e-05,
+      "loss": 2.1138,
+      "step": 360
+    },
+    {
+      "epoch": 0.06637961966271977,
+      "grad_norm": 13.182682991027832,
+      "learning_rate": 4.934876210979548e-05,
+      "loss": 2.1559,
+      "step": 370
+    },
+    {
+      "epoch": 0.06817366343738787,
+      "grad_norm": 11.596452713012695,
+      "learning_rate": 4.93308216720488e-05,
+      "loss": 2.0895,
+      "step": 380
+    },
+    {
+      "epoch": 0.06996770721205597,
+      "grad_norm": 13.062545776367188,
+      "learning_rate": 4.931288123430212e-05,
+      "loss": 2.1482,
+      "step": 390
+    },
+    {
+      "epoch": 0.07176175098672408,
+      "grad_norm": 12.839446067810059,
+      "learning_rate": 4.929494079655544e-05,
+      "loss": 2.0827,
+      "step": 400
+    },
+    {
+      "epoch": 0.07355579476139218,
+      "grad_norm": 11.77519702911377,
+      "learning_rate": 4.9277000358808756e-05,
+      "loss": 2.1411,
+      "step": 410
+    },
+    {
+      "epoch": 0.07534983853606028,
+      "grad_norm": 10.112687110900879,
+      "learning_rate": 4.925905992106208e-05,
+      "loss": 2.0311,
+      "step": 420
+    },
+    {
+      "epoch": 0.07714388231072838,
+      "grad_norm": 13.721075057983398,
+      "learning_rate": 4.92411194833154e-05,
+      "loss": 1.9533,
+      "step": 430
+    },
+    {
+      "epoch": 0.07893792608539649,
+      "grad_norm": 13.185443878173828,
+      "learning_rate": 4.9223179045568715e-05,
+      "loss": 2.1598,
+      "step": 440
+    },
+    {
+      "epoch": 0.08073196986006459,
+      "grad_norm": 11.687894821166992,
+      "learning_rate": 4.920523860782203e-05,
+      "loss": 2.2673,
+      "step": 450
+    },
+    {
+      "epoch": 0.08252601363473269,
+      "grad_norm": 17.916851043701172,
+      "learning_rate": 4.918729817007535e-05,
+      "loss": 2.1737,
+      "step": 460
+    },
+    {
+      "epoch": 0.0843200574094008,
+      "grad_norm": 11.431639671325684,
+      "learning_rate": 4.916935773232867e-05,
+      "loss": 2.1655,
+      "step": 470
+    },
+    {
+      "epoch": 0.0861141011840689,
+      "grad_norm": 14.9763822555542,
+      "learning_rate": 4.915141729458199e-05,
+      "loss": 2.0351,
+      "step": 480
+    },
+    {
+      "epoch": 0.087908144958737,
+      "grad_norm": 12.372869491577148,
+      "learning_rate": 4.913347685683531e-05,
+      "loss": 2.2227,
+      "step": 490
+    },
+    {
+      "epoch": 0.08970218873340509,
+      "grad_norm": 11.000882148742676,
+      "learning_rate": 4.9115536419088625e-05,
+      "loss": 2.0837,
+      "step": 500
+    },
+    {
+      "epoch": 0.09149623250807319,
+      "grad_norm": 22.3309326171875,
+      "learning_rate": 4.909759598134195e-05,
+      "loss": 2.1196,
+      "step": 510
+    },
+    {
+      "epoch": 0.0932902762827413,
+      "grad_norm": 15.499113082885742,
+      "learning_rate": 4.9079655543595267e-05,
+      "loss": 2.2164,
+      "step": 520
+    },
+    {
+      "epoch": 0.0950843200574094,
+      "grad_norm": 9.863581657409668,
+      "learning_rate": 4.9061715105848584e-05,
+      "loss": 2.1818,
+      "step": 530
+    },
+    {
+      "epoch": 0.0968783638320775,
+      "grad_norm": 14.396673202514648,
+      "learning_rate": 4.904377466810191e-05,
+      "loss": 2.0059,
+      "step": 540
+    },
+    {
+      "epoch": 0.0986724076067456,
+      "grad_norm": 10.207282066345215,
+      "learning_rate": 4.9025834230355225e-05,
+      "loss": 2.1265,
+      "step": 550
+    },
+    {
+      "epoch": 0.1004664513814137,
+      "grad_norm": 11.846447944641113,
+      "learning_rate": 4.900789379260854e-05,
+      "loss": 2.2348,
+      "step": 560
+    },
+    {
+      "epoch": 0.10226049515608181,
+      "grad_norm": 15.00061321258545,
+      "learning_rate": 4.898995335486186e-05,
+      "loss": 2.0822,
+      "step": 570
+    },
+    {
+      "epoch": 0.10405453893074991,
+      "grad_norm": 19.457279205322266,
+      "learning_rate": 4.897201291711518e-05,
+      "loss": 2.1838,
+      "step": 580
+    },
+    {
+      "epoch": 0.10584858270541801,
+      "grad_norm": 9.659820556640625,
+      "learning_rate": 4.89540724793685e-05,
+      "loss": 1.9626,
+      "step": 590
+    },
+    {
+      "epoch": 0.10764262648008611,
+      "grad_norm": 14.713663101196289,
+      "learning_rate": 4.893613204162182e-05,
+      "loss": 2.2248,
+      "step": 600
+    },
+    {
+      "epoch": 0.10943667025475422,
+      "grad_norm": 10.287516593933105,
+      "learning_rate": 4.8918191603875135e-05,
+      "loss": 2.1525,
+      "step": 610
+    },
+    {
+      "epoch": 0.11123071402942232,
+      "grad_norm": 10.284120559692383,
+      "learning_rate": 4.890025116612846e-05,
+      "loss": 2.1112,
+      "step": 620
+    },
+    {
+      "epoch": 0.11302475780409042,
+      "grad_norm": 9.741727828979492,
+      "learning_rate": 4.888231072838178e-05,
+      "loss": 2.0442,
+      "step": 630
+    },
+    {
+      "epoch": 0.11481880157875853,
+      "grad_norm": 8.978920936584473,
+      "learning_rate": 4.8864370290635094e-05,
+      "loss": 2.0471,
+      "step": 640
+    },
+    {
+      "epoch": 0.11661284535342663,
+      "grad_norm": 12.450641632080078,
+      "learning_rate": 4.884642985288841e-05,
+      "loss": 2.1083,
+      "step": 650
+    },
+    {
+      "epoch": 0.11840688912809473,
+      "grad_norm": 11.721051216125488,
+      "learning_rate": 4.882848941514173e-05,
+      "loss": 2.2182,
+      "step": 660
+    },
+    {
+      "epoch": 0.12020093290276283,
+      "grad_norm": 9.241230964660645,
+      "learning_rate": 4.8810548977395046e-05,
+      "loss": 2.1022,
+      "step": 670
+    },
+    {
+      "epoch": 0.12199497667743094,
+      "grad_norm": 9.87799072265625,
+      "learning_rate": 4.879260853964837e-05,
+      "loss": 1.9777,
+      "step": 680
+    },
+    {
+      "epoch": 0.12378902045209902,
+      "grad_norm": 13.317924499511719,
+      "learning_rate": 4.877466810190169e-05,
+      "loss": 2.0508,
+      "step": 690
+    },
+    {
+      "epoch": 0.12558306422676713,
+      "grad_norm": 11.3270845413208,
+      "learning_rate": 4.8756727664155004e-05,
+      "loss": 2.0391,
+      "step": 700
+    },
+    {
+      "epoch": 0.12737710800143523,
+      "grad_norm": 10.925955772399902,
+      "learning_rate": 4.873878722640833e-05,
+      "loss": 2.1051,
+      "step": 710
+    },
+    {
+      "epoch": 0.12917115177610333,
+      "grad_norm": 11.26830768585205,
+      "learning_rate": 4.8720846788661646e-05,
+      "loss": 1.994,
+      "step": 720
+    },
+    {
+      "epoch": 0.13096519555077143,
+      "grad_norm": 16.53739356994629,
+      "learning_rate": 4.870290635091496e-05,
+      "loss": 2.1491,
+      "step": 730
+    },
+    {
+      "epoch": 0.13275923932543954,
+      "grad_norm": 11.774067878723145,
+      "learning_rate": 4.868496591316829e-05,
+      "loss": 2.1163,
+      "step": 740
+    },
+    {
+      "epoch": 0.13455328310010764,
+      "grad_norm": 11.250940322875977,
+      "learning_rate": 4.8667025475421604e-05,
+      "loss": 2.1339,
+      "step": 750
+    },
+    {
+      "epoch": 0.13634732687477574,
+      "grad_norm": 12.242491722106934,
+      "learning_rate": 4.864908503767492e-05,
+      "loss": 1.8935,
+      "step": 760
+    },
+    {
+      "epoch": 0.13814137064944385,
+      "grad_norm": 9.951498985290527,
+      "learning_rate": 4.863114459992824e-05,
+      "loss": 1.9869,
+      "step": 770
+    },
+    {
+      "epoch": 0.13993541442411195,
+      "grad_norm": 8.47032356262207,
+      "learning_rate": 4.8613204162181556e-05,
+      "loss": 2.008,
+      "step": 780
+    },
+    {
+      "epoch": 0.14172945819878005,
+      "grad_norm": 10.949905395507812,
+      "learning_rate": 4.859526372443488e-05,
+      "loss": 2.1019,
+      "step": 790
+    },
+    {
+      "epoch": 0.14352350197344815,
+      "grad_norm": 11.062277793884277,
+      "learning_rate": 4.85773232866882e-05,
+      "loss": 2.1903,
+      "step": 800
+    },
+    {
+      "epoch": 0.14531754574811626,
+      "grad_norm": 16.415149688720703,
+      "learning_rate": 4.8559382848941514e-05,
+      "loss": 2.1214,
+      "step": 810
+    },
+    {
+      "epoch": 0.14711158952278436,
+      "grad_norm": 10.11135196685791,
+      "learning_rate": 4.854144241119484e-05,
+      "loss": 1.8267,
+      "step": 820
+    },
+    {
+      "epoch": 0.14890563329745246,
+      "grad_norm": 10.645978927612305,
+      "learning_rate": 4.8523501973448156e-05,
+      "loss": 2.0422,
+      "step": 830
+    },
+    {
+      "epoch": 0.15069967707212056,
+      "grad_norm": 8.619297981262207,
+      "learning_rate": 4.850556153570147e-05,
+      "loss": 2.2584,
+      "step": 840
+    },
+    {
+      "epoch": 0.15249372084678867,
+      "grad_norm": 9.050743103027344,
+      "learning_rate": 4.84876210979548e-05,
+      "loss": 2.0256,
+      "step": 850
+    },
+    {
+      "epoch": 0.15428776462145677,
+      "grad_norm": 10.283220291137695,
+      "learning_rate": 4.8469680660208114e-05,
+      "loss": 2.0351,
+      "step": 860
+    },
+    {
+      "epoch": 0.15608180839612487,
+      "grad_norm": 10.877766609191895,
+      "learning_rate": 4.8451740222461425e-05,
+      "loss": 2.0931,
+      "step": 870
+    },
+    {
+      "epoch": 0.15787585217079297,
+      "grad_norm": 13.12762451171875,
+      "learning_rate": 4.843379978471475e-05,
+      "loss": 2.1018,
+      "step": 880
+    },
+    {
+      "epoch": 0.15966989594546108,
+      "grad_norm": 9.463711738586426,
+      "learning_rate": 4.8415859346968066e-05,
+      "loss": 2.0868,
+      "step": 890
+    },
+    {
+      "epoch": 0.16146393972012918,
+      "grad_norm": 7.113370418548584,
+      "learning_rate": 4.839791890922138e-05,
+      "loss": 2.0611,
+      "step": 900
+    },
+    {
+      "epoch": 0.16325798349479728,
+      "grad_norm": 11.066067695617676,
+      "learning_rate": 4.837997847147471e-05,
+      "loss": 2.0621,
+      "step": 910
+    },
+    {
+      "epoch": 0.16505202726946538,
+      "grad_norm": 12.794750213623047,
+      "learning_rate": 4.8362038033728025e-05,
+      "loss": 2.0361,
+      "step": 920
+    },
+    {
+      "epoch": 0.1668460710441335,
+      "grad_norm": 10.758418083190918,
+      "learning_rate": 4.834409759598135e-05,
+      "loss": 1.9764,
+      "step": 930
+    },
+    {
+      "epoch": 0.1686401148188016,
+      "grad_norm": 13.04681396484375,
+      "learning_rate": 4.8326157158234666e-05,
+      "loss": 2.0678,
+      "step": 940
+    },
+    {
+      "epoch": 0.1704341585934697,
+      "grad_norm": 12.170126914978027,
+      "learning_rate": 4.830821672048798e-05,
+      "loss": 2.1812,
+      "step": 950
+    },
+    {
+      "epoch": 0.1722282023681378,
+      "grad_norm": 9.722840309143066,
+      "learning_rate": 4.82902762827413e-05,
+      "loss": 2.0603,
+      "step": 960
+    },
+    {
+      "epoch": 0.1740222461428059,
+      "grad_norm": 11.84780216217041,
+      "learning_rate": 4.827233584499462e-05,
+      "loss": 1.9741,
+      "step": 970
+    },
+    {
+      "epoch": 0.175816289917474,
+      "grad_norm": 11.813640594482422,
+      "learning_rate": 4.8254395407247935e-05,
+      "loss": 2.0533,
+      "step": 980
+    },
+    {
+      "epoch": 0.17761033369214208,
+      "grad_norm": 14.080793380737305,
+      "learning_rate": 4.823645496950126e-05,
+      "loss": 2.0094,
+      "step": 990
+    },
+    {
+      "epoch": 0.17940437746681018,
+      "grad_norm": 10.01248550415039,
+      "learning_rate": 4.8218514531754576e-05,
+      "loss": 2.1019,
+      "step": 1000
+    },
+    {
+      "epoch": 0.18119842124147828,
+      "grad_norm": 10.648508071899414,
+      "learning_rate": 4.8200574094007894e-05,
+      "loss": 2.018,
+      "step": 1010
+    },
+    {
+      "epoch": 0.18299246501614638,
+      "grad_norm": 12.019718170166016,
+      "learning_rate": 4.818263365626122e-05,
+      "loss": 1.9094,
+      "step": 1020
+    },
+    {
+      "epoch": 0.18478650879081449,
+      "grad_norm": 8.853078842163086,
+      "learning_rate": 4.8164693218514535e-05,
+      "loss": 2.1722,
+      "step": 1030
+    },
+    {
+      "epoch": 0.1865805525654826,
+      "grad_norm": 8.063078880310059,
+      "learning_rate": 4.814675278076785e-05,
+      "loss": 2.0377,
+      "step": 1040
+    },
+    {
+      "epoch": 0.1883745963401507,
+      "grad_norm": 9.952075958251953,
+      "learning_rate": 4.8128812343021176e-05,
+      "loss": 2.1291,
+      "step": 1050
+    },
+    {
+      "epoch": 0.1901686401148188,
+      "grad_norm": 10.393296241760254,
+      "learning_rate": 4.811087190527449e-05,
+      "loss": 2.0707,
+      "step": 1060
+    },
+    {
+      "epoch": 0.1919626838894869,
+      "grad_norm": 10.250103950500488,
+      "learning_rate": 4.809293146752781e-05,
+      "loss": 2.0411,
+      "step": 1070
+    },
+    {
+      "epoch": 0.193756727664155,
+      "grad_norm": 10.035394668579102,
+      "learning_rate": 4.807499102978113e-05,
+      "loss": 1.9696,
+      "step": 1080
+    },
+    {
+      "epoch": 0.1955507714388231,
+      "grad_norm": 12.22006893157959,
+      "learning_rate": 4.8057050592034445e-05,
+      "loss": 2.0681,
+      "step": 1090
+    },
+    {
+      "epoch": 0.1973448152134912,
+      "grad_norm": 9.505834579467773,
+      "learning_rate": 4.803911015428776e-05,
+      "loss": 2.0469,
+      "step": 1100
+    },
+    {
+      "epoch": 0.1991388589881593,
+      "grad_norm": 8.87895393371582,
+      "learning_rate": 4.8021169716541086e-05,
+      "loss": 2.1125,
+      "step": 1110
+    },
+    {
+      "epoch": 0.2009329027628274,
+      "grad_norm": 8.545978546142578,
+      "learning_rate": 4.8003229278794404e-05,
+      "loss": 2.008,
+      "step": 1120
+    },
+    {
+      "epoch": 0.2027269465374955,
+      "grad_norm": 9.056685447692871,
+      "learning_rate": 4.798528884104773e-05,
+      "loss": 1.9602,
+      "step": 1130
+    },
+    {
+      "epoch": 0.20452099031216361,
+      "grad_norm": 15.332698822021484,
+      "learning_rate": 4.7967348403301045e-05,
+      "loss": 1.8802,
+      "step": 1140
+    },
+    {
+      "epoch": 0.20631503408683172,
+      "grad_norm": 12.204120635986328,
+      "learning_rate": 4.794940796555436e-05,
+      "loss": 1.9994,
+      "step": 1150
+    },
+    {
+      "epoch": 0.20810907786149982,
+      "grad_norm": 10.399602890014648,
+      "learning_rate": 4.7931467527807686e-05,
+      "loss": 2.1549,
+      "step": 1160
+    },
+    {
+      "epoch": 0.20990312163616792,
+      "grad_norm": 9.073091506958008,
+      "learning_rate": 4.7913527090061e-05,
+      "loss": 2.1483,
+      "step": 1170
+    },
+    {
+      "epoch": 0.21169716541083602,
+      "grad_norm": 11.733457565307617,
+      "learning_rate": 4.7895586652314314e-05,
+      "loss": 2.1155,
+      "step": 1180
+    },
+    {
+      "epoch": 0.21349120918550413,
+      "grad_norm": 7.083014011383057,
+      "learning_rate": 4.787764621456764e-05,
+      "loss": 2.0738,
+      "step": 1190
+    },
+    {
+      "epoch": 0.21528525296017223,
+      "grad_norm": 11.031323432922363,
+      "learning_rate": 4.7859705776820955e-05,
+      "loss": 2.1221,
+      "step": 1200
+    },
+    {
+      "epoch": 0.21707929673484033,
+      "grad_norm": 9.98110580444336,
+      "learning_rate": 4.784176533907427e-05,
+      "loss": 2.0356,
+      "step": 1210
+    },
+    {
+      "epoch": 0.21887334050950844,
+      "grad_norm": 13.873873710632324,
+      "learning_rate": 4.78238249013276e-05,
+      "loss": 2.0523,
+      "step": 1220
+    },
+    {
+      "epoch": 0.22066738428417654,
+      "grad_norm": 10.365226745605469,
+      "learning_rate": 4.7805884463580914e-05,
+      "loss": 2.0989,
+      "step": 1230
+    },
+    {
+      "epoch": 0.22246142805884464,
+      "grad_norm": 9.975032806396484,
+      "learning_rate": 4.778794402583423e-05,
+      "loss": 2.0278,
+      "step": 1240
+    },
+    {
+      "epoch": 0.22425547183351274,
+      "grad_norm": 10.711677551269531,
+      "learning_rate": 4.7770003588087555e-05,
+      "loss": 2.1347,
+      "step": 1250
+    },
+    {
+      "epoch": 0.22604951560818085,
+      "grad_norm": 8.741484642028809,
+      "learning_rate": 4.775206315034087e-05,
+      "loss": 2.0043,
+      "step": 1260
+    },
+    {
+      "epoch": 0.22784355938284895,
+      "grad_norm": 11.853199005126953,
+      "learning_rate": 4.773412271259419e-05,
+      "loss": 2.1624,
+      "step": 1270
+    },
+    {
+      "epoch": 0.22963760315751705,
+      "grad_norm": 8.458366394042969,
+      "learning_rate": 4.771618227484751e-05,
+      "loss": 2.0635,
+      "step": 1280
+    },
+    {
+      "epoch": 0.23143164693218515,
+      "grad_norm": 9.640998840332031,
+      "learning_rate": 4.7698241837100824e-05,
+      "loss": 1.9351,
+      "step": 1290
+    },
+    {
+      "epoch": 0.23322569070685326,
+      "grad_norm": 11.35208511352539,
+      "learning_rate": 4.768030139935415e-05,
+      "loss": 2.0016,
+      "step": 1300
+    },
+    {
+      "epoch": 0.23501973448152136,
+      "grad_norm": 12.047490119934082,
+      "learning_rate": 4.7662360961607465e-05,
+      "loss": 1.9982,
+      "step": 1310
+    },
+    {
+      "epoch": 0.23681377825618946,
+      "grad_norm": 7.340638637542725,
+      "learning_rate": 4.764442052386078e-05,
+      "loss": 1.9317,
+      "step": 1320
+    },
+    {
+      "epoch": 0.23860782203085756,
+      "grad_norm": 9.259571075439453,
+      "learning_rate": 4.762648008611411e-05,
+      "loss": 1.867,
+      "step": 1330
+    },
+    {
+      "epoch": 0.24040186580552567,
+      "grad_norm": 10.415928840637207,
+      "learning_rate": 4.7608539648367424e-05,
+      "loss": 1.8926,
+      "step": 1340
+    },
+    {
+      "epoch": 0.24219590958019377,
+      "grad_norm": 10.523111343383789,
+      "learning_rate": 4.759059921062074e-05,
+      "loss": 1.9432,
+      "step": 1350
+    },
+    {
+      "epoch": 0.24398995335486187,
+      "grad_norm": 10.31205940246582,
+      "learning_rate": 4.7572658772874065e-05,
+      "loss": 2.0679,
+      "step": 1360
+    },
+    {
+      "epoch": 0.24578399712952997,
+      "grad_norm": 11.592795372009277,
+      "learning_rate": 4.755471833512738e-05,
+      "loss": 1.8786,
+      "step": 1370
+    },
+    {
+      "epoch": 0.24757804090419805,
+      "grad_norm": 13.131718635559082,
+      "learning_rate": 4.75367778973807e-05,
+      "loss": 2.1331,
+      "step": 1380
+    },
+    {
+      "epoch": 0.24937208467886615,
+      "grad_norm": 8.448001861572266,
+      "learning_rate": 4.751883745963402e-05,
+      "loss": 2.2519,
+      "step": 1390
+    },
+    {
+      "epoch": 0.25116612845353425,
+      "grad_norm": 9.400849342346191,
+      "learning_rate": 4.7500897021887334e-05,
+      "loss": 2.162,
+      "step": 1400
+    },
+    {
+      "epoch": 0.2529601722282024,
+      "grad_norm": 9.269533157348633,
+      "learning_rate": 4.748295658414065e-05,
+      "loss": 2.0009,
+      "step": 1410
+    },
+    {
+      "epoch": 0.25475421600287046,
+      "grad_norm": 12.824959754943848,
+      "learning_rate": 4.7465016146393976e-05,
+      "loss": 2.1757,
+      "step": 1420
+    },
+    {
+      "epoch": 0.2565482597775386,
+      "grad_norm": 6.765061378479004,
+      "learning_rate": 4.744707570864729e-05,
+      "loss": 1.9529,
+      "step": 1430
+    },
+    {
+      "epoch": 0.25834230355220666,
+      "grad_norm": 8.876534461975098,
+      "learning_rate": 4.742913527090061e-05,
+      "loss": 1.9964,
+      "step": 1440
+    },
+    {
+      "epoch": 0.2601363473268748,
+      "grad_norm": 13.438302040100098,
+      "learning_rate": 4.7411194833153934e-05,
+      "loss": 2.0535,
+      "step": 1450
+    },
+    {
+      "epoch": 0.26193039110154287,
+      "grad_norm": 11.55459213256836,
+      "learning_rate": 4.739325439540725e-05,
+      "loss": 2.0917,
+      "step": 1460
+    },
+    {
+      "epoch": 0.263724434876211,
+      "grad_norm": 13.218855857849121,
+      "learning_rate": 4.737531395766057e-05,
+      "loss": 1.9388,
+      "step": 1470
+    },
+    {
+      "epoch": 0.2655184786508791,
+      "grad_norm": 8.17874526977539,
+      "learning_rate": 4.7357373519913886e-05,
+      "loss": 1.9581,
+      "step": 1480
+    },
+    {
+      "epoch": 0.2673125224255472,
+      "grad_norm": 9.68444538116455,
+      "learning_rate": 4.73394330821672e-05,
+      "loss": 2.0179,
+      "step": 1490
+    },
+    {
+      "epoch": 0.2691065662002153,
+      "grad_norm": 10.616148948669434,
+      "learning_rate": 4.732149264442053e-05,
+      "loss": 2.024,
+      "step": 1500
+    },
+    {
+      "epoch": 0.2709006099748834,
+      "grad_norm": 11.970083236694336,
+      "learning_rate": 4.7303552206673845e-05,
+      "loss": 2.1263,
+      "step": 1510
+    },
+    {
+      "epoch": 0.2726946537495515,
+      "grad_norm": 9.173564910888672,
+      "learning_rate": 4.728561176892716e-05,
+      "loss": 1.9028,
+      "step": 1520
+    },
+    {
+      "epoch": 0.2744886975242196,
+      "grad_norm": 12.469501495361328,
+      "learning_rate": 4.7267671331180486e-05,
+      "loss": 2.1234,
+      "step": 1530
+    },
+    {
+      "epoch": 0.2762827412988877,
+      "grad_norm": 7.576513290405273,
+      "learning_rate": 4.72497308934338e-05,
+      "loss": 1.9832,
+      "step": 1540
+    },
+    {
+      "epoch": 0.2780767850735558,
+      "grad_norm": 7.813937187194824,
+      "learning_rate": 4.723179045568712e-05,
+      "loss": 2.0337,
+      "step": 1550
+    },
+    {
+      "epoch": 0.2798708288482239,
+      "grad_norm": 8.702999114990234,
+      "learning_rate": 4.7213850017940444e-05,
+      "loss": 1.8968,
+      "step": 1560
+    },
+    {
+      "epoch": 0.281664872622892,
+      "grad_norm": 9.221305847167969,
+      "learning_rate": 4.719590958019376e-05,
+      "loss": 1.8972,
+      "step": 1570
+    },
+    {
+      "epoch": 0.2834589163975601,
+      "grad_norm": 7.7957892417907715,
+      "learning_rate": 4.717796914244708e-05,
+      "loss": 1.985,
+      "step": 1580
+    },
+    {
+      "epoch": 0.2852529601722282,
+      "grad_norm": 15.21849250793457,
+      "learning_rate": 4.7160028704700396e-05,
+      "loss": 1.8525,
+      "step": 1590
+    },
+    {
+      "epoch": 0.2870470039468963,
+      "grad_norm": 11.758828163146973,
+      "learning_rate": 4.7142088266953713e-05,
+      "loss": 2.039,
+      "step": 1600
+    },
+    {
+      "epoch": 0.2888410477215644,
+      "grad_norm": 9.403512001037598,
+      "learning_rate": 4.712414782920703e-05,
+      "loss": 2.0726,
+      "step": 1610
+    },
+    {
+      "epoch": 0.2906350914962325,
+      "grad_norm": 10.458137512207031,
+      "learning_rate": 4.7106207391460355e-05,
+      "loss": 2.1689,
+      "step": 1620
+    },
+    {
+      "epoch": 0.2924291352709006,
+      "grad_norm": 7.665112495422363,
+      "learning_rate": 4.708826695371367e-05,
+      "loss": 1.9077,
+      "step": 1630
+    },
+    {
+      "epoch": 0.2942231790455687,
+      "grad_norm": 11.904746055603027,
+      "learning_rate": 4.707032651596699e-05,
+      "loss": 2.0425,
+      "step": 1640
+    },
+    {
+      "epoch": 0.2960172228202368,
+      "grad_norm": 8.795477867126465,
+      "learning_rate": 4.705238607822031e-05,
+      "loss": 1.9992,
+      "step": 1650
+    },
+    {
+      "epoch": 0.2978112665949049,
+      "grad_norm": 7.0219550132751465,
+      "learning_rate": 4.703444564047363e-05,
+      "loss": 1.9143,
+      "step": 1660
+    },
+    {
+      "epoch": 0.299605310369573,
+      "grad_norm": 7.745930194854736,
+      "learning_rate": 4.7016505202726955e-05,
+      "loss": 1.8667,
+      "step": 1670
+    },
+    {
+      "epoch": 0.3013993541442411,
+      "grad_norm": 7.579682350158691,
+      "learning_rate": 4.699856476498027e-05,
+      "loss": 2.0123,
+      "step": 1680
+    },
+    {
+      "epoch": 0.3031933979189092,
+      "grad_norm": 7.5498199462890625,
+      "learning_rate": 4.698062432723358e-05,
+      "loss": 2.1165,
+      "step": 1690
+    },
+    {
+      "epoch": 0.30498744169357733,
+      "grad_norm": 8.588129043579102,
+      "learning_rate": 4.6962683889486906e-05,
+      "loss": 2.0662,
+      "step": 1700
+    },
+    {
+      "epoch": 0.3067814854682454,
+      "grad_norm": 6.789324760437012,
+      "learning_rate": 4.6944743451740224e-05,
+      "loss": 1.9604,
+      "step": 1710
+    },
+    {
+      "epoch": 0.30857552924291354,
+      "grad_norm": 8.46413803100586,
+      "learning_rate": 4.692680301399354e-05,
+      "loss": 2.0078,
+      "step": 1720
+    },
+    {
+      "epoch": 0.3103695730175816,
+      "grad_norm": 9.2481689453125,
+      "learning_rate": 4.6908862576246865e-05,
+      "loss": 2.0001,
+      "step": 1730
+    },
+    {
+      "epoch": 0.31216361679224974,
+      "grad_norm": 9.034040451049805,
+      "learning_rate": 4.689092213850018e-05,
+      "loss": 2.1038,
+      "step": 1740
+    },
+    {
+      "epoch": 0.3139576605669178,
+      "grad_norm": 9.660133361816406,
+      "learning_rate": 4.68729817007535e-05,
+      "loss": 2.0491,
+      "step": 1750
+    },
+    {
+      "epoch": 0.31575170434158595,
+      "grad_norm": 8.319841384887695,
+      "learning_rate": 4.6855041263006823e-05,
+      "loss": 1.9269,
+      "step": 1760
+    },
+    {
+      "epoch": 0.317545748116254,
+      "grad_norm": 12.393325805664062,
+      "learning_rate": 4.683710082526014e-05,
+      "loss": 2.0983,
+      "step": 1770
+    },
+    {
+      "epoch": 0.31933979189092215,
+      "grad_norm": 9.26393985748291,
+      "learning_rate": 4.681916038751346e-05,
+      "loss": 1.8658,
+      "step": 1780
+    },
+    {
+      "epoch": 0.32113383566559023,
+      "grad_norm": 10.711555480957031,
+      "learning_rate": 4.6801219949766775e-05,
+      "loss": 2.0091,
+      "step": 1790
+    },
+    {
+      "epoch": 0.32292787944025836,
+      "grad_norm": 9.118010520935059,
+      "learning_rate": 4.678327951202009e-05,
+      "loss": 2.119,
+      "step": 1800
+    },
+    {
+      "epoch": 0.32472192321492643,
+      "grad_norm": 10.101414680480957,
+      "learning_rate": 4.676533907427341e-05,
+      "loss": 2.1448,
+      "step": 1810
+    },
+    {
+      "epoch": 0.32651596698959456,
+      "grad_norm": 11.349935531616211,
+      "learning_rate": 4.6747398636526734e-05,
+      "loss": 2.0042,
+      "step": 1820
+    },
+    {
+      "epoch": 0.32831001076426264,
+      "grad_norm": 8.576199531555176,
+      "learning_rate": 4.672945819878005e-05,
+      "loss": 2.0786,
+      "step": 1830
+    },
+    {
+      "epoch": 0.33010405453893077,
+      "grad_norm": 8.958368301391602,
+      "learning_rate": 4.671151776103337e-05,
+      "loss": 1.9364,
+      "step": 1840
+    },
+    {
+      "epoch": 0.33189809831359884,
+      "grad_norm": 10.209322929382324,
+      "learning_rate": 4.669357732328669e-05,
+      "loss": 2.1192,
+      "step": 1850
+    },
+    {
+      "epoch": 0.333692142088267,
+      "grad_norm": 10.517390251159668,
+      "learning_rate": 4.667563688554001e-05,
+      "loss": 1.8833,
+      "step": 1860
+    },
+    {
+      "epoch": 0.33548618586293505,
+      "grad_norm": 11.347796440124512,
+      "learning_rate": 4.6657696447793334e-05,
+      "loss": 2.0498,
+      "step": 1870
+    },
+    {
+      "epoch": 0.3372802296376032,
+      "grad_norm": 9.35456371307373,
+      "learning_rate": 4.663975601004665e-05,
+      "loss": 2.0158,
+      "step": 1880
+    },
+    {
+      "epoch": 0.33907427341227125,
+      "grad_norm": 7.018476486206055,
+      "learning_rate": 4.662181557229997e-05,
+      "loss": 1.9527,
+      "step": 1890
+    },
+    {
+      "epoch": 0.3408683171869394,
+      "grad_norm": 13.440537452697754,
+      "learning_rate": 4.6603875134553285e-05,
+      "loss": 1.9492,
+      "step": 1900
+    },
+    {
+      "epoch": 0.34266236096160746,
+      "grad_norm": 7.040363788604736,
+      "learning_rate": 4.65859346968066e-05,
+      "loss": 2.085,
+      "step": 1910
+    },
+    {
+      "epoch": 0.3444564047362756,
+      "grad_norm": 8.57967472076416,
+      "learning_rate": 4.656799425905992e-05,
+      "loss": 1.9639,
+      "step": 1920
+    },
+    {
+      "epoch": 0.34625044851094366,
+      "grad_norm": 7.877899169921875,
+      "learning_rate": 4.6550053821313244e-05,
+      "loss": 1.9315,
+      "step": 1930
+    },
+    {
+      "epoch": 0.3480444922856118,
+      "grad_norm": 10.133997917175293,
+      "learning_rate": 4.653211338356656e-05,
+      "loss": 1.9489,
+      "step": 1940
+    },
+    {
+      "epoch": 0.34983853606027987,
+      "grad_norm": 10.46199893951416,
+      "learning_rate": 4.651417294581988e-05,
+      "loss": 2.1294,
+      "step": 1950
+    },
+    {
+      "epoch": 0.351632579834948,
+      "grad_norm": 10.541289329528809,
+      "learning_rate": 4.64962325080732e-05,
+      "loss": 2.0334,
+      "step": 1960
+    },
+    {
+      "epoch": 0.3534266236096161,
+      "grad_norm": 10.026679039001465,
+      "learning_rate": 4.647829207032652e-05,
+      "loss": 1.9669,
+      "step": 1970
+    },
+    {
+      "epoch": 0.35522066738428415,
+      "grad_norm": 9.2806978225708,
+      "learning_rate": 4.646035163257984e-05,
+      "loss": 2.0065,
+      "step": 1980
+    },
+    {
+      "epoch": 0.3570147111589523,
+      "grad_norm": 11.332128524780273,
+      "learning_rate": 4.6442411194833154e-05,
+      "loss": 2.1402,
+      "step": 1990
+    },
+    {
+      "epoch": 0.35880875493362036,
+      "grad_norm": 9.25808048248291,
+      "learning_rate": 4.642447075708647e-05,
+      "loss": 2.1242,
+      "step": 2000
+    },
+    {
+      "epoch": 0.3606027987082885,
+      "grad_norm": 9.891722679138184,
+      "learning_rate": 4.640653031933979e-05,
+      "loss": 2.0954,
+      "step": 2010
+    },
+    {
+      "epoch": 0.36239684248295656,
+      "grad_norm": 8.027229309082031,
+      "learning_rate": 4.638858988159311e-05,
+      "loss": 1.94,
+      "step": 2020
+    },
+    {
+      "epoch": 0.3641908862576247,
+      "grad_norm": 7.155013084411621,
+      "learning_rate": 4.637064944384643e-05,
+      "loss": 2.0099,
+      "step": 2030
+    },
+    {
+      "epoch": 0.36598493003229277,
+      "grad_norm": 7.851609230041504,
+      "learning_rate": 4.6352709006099754e-05,
+      "loss": 2.0819,
+      "step": 2040
+    },
+    {
+      "epoch": 0.3677789738069609,
+      "grad_norm": 9.718694686889648,
+      "learning_rate": 4.633476856835307e-05,
+      "loss": 2.0265,
+      "step": 2050
+    },
+    {
+      "epoch": 0.36957301758162897,
+      "grad_norm": 7.794171333312988,
+      "learning_rate": 4.631682813060639e-05,
+      "loss": 1.9589,
+      "step": 2060
+    },
+    {
+      "epoch": 0.3713670613562971,
+      "grad_norm": 9.896645545959473,
+      "learning_rate": 4.629888769285971e-05,
+      "loss": 1.8671,
+      "step": 2070
+    },
+    {
+      "epoch": 0.3731611051309652,
+      "grad_norm": 8.257375717163086,
+      "learning_rate": 4.628094725511303e-05,
+      "loss": 2.0167,
+      "step": 2080
+    },
+    {
+      "epoch": 0.3749551489056333,
+      "grad_norm": 9.882346153259277,
+      "learning_rate": 4.626300681736635e-05,
+      "loss": 2.079,
+      "step": 2090
+    },
+    {
+      "epoch": 0.3767491926803014,
+      "grad_norm": 8.422935485839844,
+      "learning_rate": 4.6245066379619664e-05,
+      "loss": 1.998,
+      "step": 2100
+    },
+    {
+      "epoch": 0.3785432364549695,
+      "grad_norm": 9.090377807617188,
+      "learning_rate": 4.622712594187298e-05,
+      "loss": 1.8588,
+      "step": 2110
+    },
+    {
+      "epoch": 0.3803372802296376,
+      "grad_norm": 9.11992359161377,
+      "learning_rate": 4.62091855041263e-05,
+      "loss": 1.9268,
+      "step": 2120
+    },
+    {
+      "epoch": 0.3821313240043057,
+      "grad_norm": 7.935147285461426,
+      "learning_rate": 4.619124506637962e-05,
+      "loss": 2.1426,
+      "step": 2130
+    },
+    {
+      "epoch": 0.3839253677789738,
+      "grad_norm": 6.065644264221191,
+      "learning_rate": 4.617330462863294e-05,
+      "loss": 1.9426,
+      "step": 2140
+    },
+    {
+      "epoch": 0.3857194115536419,
+      "grad_norm": 7.789048194885254,
+      "learning_rate": 4.615536419088626e-05,
+      "loss": 1.9607,
+      "step": 2150
+    },
+    {
+      "epoch": 0.38751345532831,
+      "grad_norm": 10.809876441955566,
+      "learning_rate": 4.613742375313958e-05,
+      "loss": 2.0248,
+      "step": 2160
+    },
+    {
+      "epoch": 0.3893074991029781,
+      "grad_norm": 7.871123790740967,
+      "learning_rate": 4.61194833153929e-05,
+      "loss": 2.0491,
+      "step": 2170
+    },
+    {
+      "epoch": 0.3911015428776462,
+      "grad_norm": 10.08164119720459,
+      "learning_rate": 4.6101542877646216e-05,
+      "loss": 1.8583,
+      "step": 2180
+    },
+    {
+      "epoch": 0.39289558665231433,
+      "grad_norm": 8.326801300048828,
+      "learning_rate": 4.608360243989954e-05,
+      "loss": 2.0879,
+      "step": 2190
+    },
+    {
+      "epoch": 0.3946896304269824,
+      "grad_norm": 9.958331108093262,
+      "learning_rate": 4.606566200215286e-05,
+      "loss": 1.9132,
+      "step": 2200
+    },
+    {
+      "epoch": 0.39648367420165054,
+      "grad_norm": 8.36109733581543,
+      "learning_rate": 4.604772156440617e-05,
+      "loss": 1.9548,
+      "step": 2210
+    },
+    {
+      "epoch": 0.3982777179763186,
+      "grad_norm": 8.014139175415039,
+      "learning_rate": 4.602978112665949e-05,
+      "loss": 1.8601,
+      "step": 2220
+    },
+    {
+      "epoch": 0.40007176175098674,
+      "grad_norm": 8.810429573059082,
+      "learning_rate": 4.601184068891281e-05,
+      "loss": 2.0101,
+      "step": 2230
+    },
+    {
+      "epoch": 0.4018658055256548,
+      "grad_norm": 9.849298477172852,
+      "learning_rate": 4.599390025116613e-05,
+      "loss": 1.9178,
+      "step": 2240
+    },
+    {
+      "epoch": 0.40365984930032295,
+      "grad_norm": 7.431755065917969,
+      "learning_rate": 4.597595981341945e-05,
+      "loss": 1.9154,
+      "step": 2250
+    },
+    {
+      "epoch": 0.405453893074991,
+      "grad_norm": 10.968128204345703,
+      "learning_rate": 4.595801937567277e-05,
+      "loss": 2.0084,
+      "step": 2260
+    },
+    {
+      "epoch": 0.40724793684965915,
+      "grad_norm": 8.786876678466797,
+      "learning_rate": 4.594007893792609e-05,
+      "loss": 2.1216,
+      "step": 2270
+    },
+    {
+      "epoch": 0.40904198062432723,
+      "grad_norm": 8.992511749267578,
+      "learning_rate": 4.592213850017941e-05,
+      "loss": 1.9462,
+      "step": 2280
+    },
+    {
+      "epoch": 0.41083602439899536,
+      "grad_norm": 8.601082801818848,
+      "learning_rate": 4.5904198062432726e-05,
+      "loss": 1.988,
+      "step": 2290
+    },
+    {
+      "epoch": 0.41263006817366343,
+      "grad_norm": 9.053206443786621,
+      "learning_rate": 4.5886257624686044e-05,
+      "loss": 1.8687,
+      "step": 2300
+    },
+    {
+      "epoch": 0.41442411194833156,
+      "grad_norm": 10.55015754699707,
+      "learning_rate": 4.586831718693936e-05,
+      "loss": 1.9853,
+      "step": 2310
+    },
+    {
+      "epoch": 0.41621815572299964,
+      "grad_norm": 7.392729759216309,
+      "learning_rate": 4.585037674919268e-05,
+      "loss": 2.0305,
+      "step": 2320
+    },
+    {
+      "epoch": 0.41801219949766777,
+      "grad_norm": 8.890578269958496,
+      "learning_rate": 4.5832436311446e-05,
+      "loss": 2.1245,
+      "step": 2330
+    },
+    {
+      "epoch": 0.41980624327233584,
+      "grad_norm": 8.476337432861328,
+      "learning_rate": 4.581449587369932e-05,
+      "loss": 1.8404,
+      "step": 2340
+    },
+    {
+      "epoch": 0.421600287047004,
+      "grad_norm": 11.671123504638672,
+      "learning_rate": 4.5796555435952637e-05,
+      "loss": 2.0296,
+      "step": 2350
+    },
+    {
+      "epoch": 0.42339433082167205,
+      "grad_norm": 9.152827262878418,
+      "learning_rate": 4.577861499820596e-05,
+      "loss": 1.9146,
+      "step": 2360
+    },
+    {
+      "epoch": 0.4251883745963401,
+      "grad_norm": 7.549522876739502,
+      "learning_rate": 4.576067456045928e-05,
+      "loss": 2.0178,
+      "step": 2370
+    },
+    {
+      "epoch": 0.42698241837100825,
+      "grad_norm": 11.978499412536621,
+      "learning_rate": 4.5742734122712595e-05,
+      "loss": 1.7915,
+      "step": 2380
+    },
+    {
+      "epoch": 0.42877646214567633,
+      "grad_norm": 10.303351402282715,
+      "learning_rate": 4.572479368496592e-05,
+      "loss": 1.8197,
+      "step": 2390
+    },
+    {
+      "epoch": 0.43057050592034446,
+      "grad_norm": 9.132424354553223,
+      "learning_rate": 4.5706853247219236e-05,
+      "loss": 2.1076,
+      "step": 2400
+    },
+    {
+      "epoch": 0.43236454969501253,
+      "grad_norm": 9.768096923828125,
+      "learning_rate": 4.5688912809472554e-05,
+      "loss": 1.8818,
+      "step": 2410
+    },
+    {
+      "epoch": 0.43415859346968066,
+      "grad_norm": 7.977086067199707,
+      "learning_rate": 4.567097237172587e-05,
+      "loss": 1.9721,
+      "step": 2420
+    },
+    {
+      "epoch": 0.43595263724434874,
+      "grad_norm": 10.111530303955078,
+      "learning_rate": 4.565303193397919e-05,
+      "loss": 1.9108,
+      "step": 2430
+    },
+    {
+      "epoch": 0.43774668101901687,
+      "grad_norm": 8.572367668151855,
+      "learning_rate": 4.563509149623251e-05,
+      "loss": 1.8878,
+      "step": 2440
+    },
+    {
+      "epoch": 0.43954072479368494,
+      "grad_norm": 15.499300956726074,
+      "learning_rate": 4.561715105848583e-05,
+      "loss": 2.0646,
+      "step": 2450
+    },
+    {
+      "epoch": 0.4413347685683531,
+      "grad_norm": 8.507719993591309,
+      "learning_rate": 4.559921062073915e-05,
+      "loss": 2.0785,
+      "step": 2460
+    },
+    {
+      "epoch": 0.44312881234302115,
+      "grad_norm": 8.560482025146484,
+      "learning_rate": 4.558127018299247e-05,
+      "loss": 1.919,
+      "step": 2470
+    },
+    {
+      "epoch": 0.4449228561176893,
+      "grad_norm": 8.793256759643555,
+      "learning_rate": 4.556332974524579e-05,
+      "loss": 2.0275,
+      "step": 2480
+    },
+    {
+      "epoch": 0.44671689989235736,
+      "grad_norm": 6.947265625,
+      "learning_rate": 4.5545389307499105e-05,
+      "loss": 1.8708,
+      "step": 2490
+    },
+    {
+      "epoch": 0.4485109436670255,
+      "grad_norm": 8.579648971557617,
+      "learning_rate": 4.552744886975242e-05,
+      "loss": 2.0002,
+      "step": 2500
+    },
+    {
+      "epoch": 0.45030498744169356,
+      "grad_norm": 7.623202323913574,
+      "learning_rate": 4.550950843200574e-05,
+      "loss": 1.9444,
+      "step": 2510
+    },
+    {
+      "epoch": 0.4520990312163617,
+      "grad_norm": 8.441667556762695,
+      "learning_rate": 4.549156799425906e-05,
+      "loss": 2.0275,
+      "step": 2520
+    },
+    {
+      "epoch": 0.45389307499102977,
+      "grad_norm": 9.18626594543457,
+      "learning_rate": 4.547362755651238e-05,
+      "loss": 1.9512,
+      "step": 2530
+    },
+    {
+      "epoch": 0.4556871187656979,
+      "grad_norm": 8.11828899383545,
+      "learning_rate": 4.54556871187657e-05,
+      "loss": 1.9609,
+      "step": 2540
+    },
+    {
+      "epoch": 0.45748116254036597,
+      "grad_norm": 7.3665852546691895,
+      "learning_rate": 4.5437746681019016e-05,
+      "loss": 1.9721,
+      "step": 2550
+    },
+    {
+      "epoch": 0.4592752063150341,
+      "grad_norm": 8.686524391174316,
+      "learning_rate": 4.541980624327234e-05,
+      "loss": 2.0039,
+      "step": 2560
+    },
+    {
+      "epoch": 0.4610692500897022,
+      "grad_norm": 8.56877326965332,
+      "learning_rate": 4.540186580552566e-05,
+      "loss": 2.0762,
+      "step": 2570
+    },
+    {
+      "epoch": 0.4628632938643703,
+      "grad_norm": 9.964433670043945,
+      "learning_rate": 4.5383925367778974e-05,
+      "loss": 1.9773,
+      "step": 2580
+    },
+    {
+      "epoch": 0.4646573376390384,
+      "grad_norm": 8.93453311920166,
+      "learning_rate": 4.53659849300323e-05,
+      "loss": 1.9763,
+      "step": 2590
+    },
+    {
+      "epoch": 0.4664513814137065,
+      "grad_norm": 6.911281108856201,
+      "learning_rate": 4.5348044492285615e-05,
+      "loss": 1.8979,
+      "step": 2600
+    },
+    {
+      "epoch": 0.4682454251883746,
+      "grad_norm": 6.9323410987854,
+      "learning_rate": 4.533010405453893e-05,
+      "loss": 1.9935,
+      "step": 2610
+    },
+    {
+      "epoch": 0.4700394689630427,
+      "grad_norm": 7.738839626312256,
+      "learning_rate": 4.531216361679225e-05,
+      "loss": 1.9254,
+      "step": 2620
+    },
+    {
+      "epoch": 0.4718335127377108,
+      "grad_norm": 6.664251804351807,
+      "learning_rate": 4.529422317904557e-05,
+      "loss": 1.9573,
+      "step": 2630
+    },
+    {
+      "epoch": 0.4736275565123789,
+      "grad_norm": 8.574616432189941,
+      "learning_rate": 4.527628274129889e-05,
+      "loss": 1.9251,
+      "step": 2640
+    },
+    {
+      "epoch": 0.475421600287047,
+      "grad_norm": 7.2672834396362305,
+      "learning_rate": 4.525834230355221e-05,
+      "loss": 1.8782,
+      "step": 2650
+    },
+    {
+      "epoch": 0.4772156440617151,
+      "grad_norm": 6.810856342315674,
+      "learning_rate": 4.5240401865805526e-05,
+      "loss": 1.9968,
+      "step": 2660
+    },
+    {
+      "epoch": 0.4790096878363832,
+      "grad_norm": 16.551050186157227,
+      "learning_rate": 4.522246142805885e-05,
+      "loss": 1.9016,
+      "step": 2670
+    },
+    {
+      "epoch": 0.48080373161105133,
+      "grad_norm": 9.01229190826416,
+      "learning_rate": 4.520452099031217e-05,
+      "loss": 1.9715,
+      "step": 2680
+    },
+    {
+      "epoch": 0.4825977753857194,
+      "grad_norm": 9.2802152633667,
+      "learning_rate": 4.5186580552565484e-05,
+      "loss": 2.0513,
+      "step": 2690
+    },
+    {
+      "epoch": 0.48439181916038754,
+      "grad_norm": 7.945450782775879,
+      "learning_rate": 4.516864011481881e-05,
+      "loss": 1.849,
+      "step": 2700
+    },
+    {
+      "epoch": 0.4861858629350556,
+      "grad_norm": 9.4662446975708,
+      "learning_rate": 4.5150699677072126e-05,
+      "loss": 1.9349,
+      "step": 2710
+    },
+    {
+      "epoch": 0.48797990670972374,
+      "grad_norm": 7.22721004486084,
+      "learning_rate": 4.5132759239325436e-05,
+      "loss": 2.0256,
+      "step": 2720
+    },
+    {
+      "epoch": 0.4897739504843918,
+      "grad_norm": 8.505002975463867,
+      "learning_rate": 4.511481880157876e-05,
+      "loss": 2.0187,
+      "step": 2730
+    },
+    {
+      "epoch": 0.49156799425905995,
+      "grad_norm": 8.256681442260742,
+      "learning_rate": 4.509687836383208e-05,
+      "loss": 1.8934,
+      "step": 2740
+    },
+    {
+      "epoch": 0.493362038033728,
+      "grad_norm": 13.187761306762695,
+      "learning_rate": 4.5078937926085395e-05,
+      "loss": 1.9564,
+      "step": 2750
+    },
+    {
+      "epoch": 0.4951560818083961,
+      "grad_norm": 9.577531814575195,
+      "learning_rate": 4.506099748833872e-05,
+      "loss": 1.9369,
+      "step": 2760
+    },
+    {
+      "epoch": 0.49695012558306423,
+      "grad_norm": 8.296804428100586,
+      "learning_rate": 4.5043057050592036e-05,
+      "loss": 2.043,
+      "step": 2770
+    },
+    {
+      "epoch": 0.4987441693577323,
+      "grad_norm": 7.145248889923096,
+      "learning_rate": 4.502511661284536e-05,
+      "loss": 1.7847,
+      "step": 2780
+    },
+    {
+      "epoch": 0.5005382131324004,
+      "grad_norm": 10.56977367401123,
+      "learning_rate": 4.500717617509868e-05,
+      "loss": 1.8364,
+      "step": 2790
+    },
+    {
+      "epoch": 0.5023322569070685,
+      "grad_norm": 9.34067440032959,
+      "learning_rate": 4.4989235737351995e-05,
+      "loss": 1.9133,
+      "step": 2800
+    },
+    {
+      "epoch": 0.5041263006817366,
+      "grad_norm": 8.067873001098633,
+      "learning_rate": 4.497129529960531e-05,
+      "loss": 1.8836,
+      "step": 2810
+    },
+    {
+      "epoch": 0.5059203444564048,
+      "grad_norm": 8.651354789733887,
+      "learning_rate": 4.495335486185863e-05,
+      "loss": 1.8979,
+      "step": 2820
+    },
+    {
+      "epoch": 0.5077143882310728,
+      "grad_norm": 6.594571590423584,
+      "learning_rate": 4.4935414424111946e-05,
+      "loss": 1.9502,
+      "step": 2830
+    },
+    {
+      "epoch": 0.5095084320057409,
+      "grad_norm": 10.328874588012695,
+      "learning_rate": 4.491747398636527e-05,
+      "loss": 2.0826,
+      "step": 2840
+    },
+    {
+      "epoch": 0.511302475780409,
+      "grad_norm": 6.956106185913086,
+      "learning_rate": 4.489953354861859e-05,
+      "loss": 1.9671,
+      "step": 2850
+    },
+    {
+      "epoch": 0.5130965195550772,
+      "grad_norm": 7.3767313957214355,
+      "learning_rate": 4.4881593110871905e-05,
+      "loss": 1.902,
+      "step": 2860
+    },
+    {
+      "epoch": 0.5148905633297453,
+      "grad_norm": 11.275064468383789,
+      "learning_rate": 4.486365267312523e-05,
+      "loss": 2.0932,
+      "step": 2870
+    },
+    {
+      "epoch": 0.5166846071044133,
+      "grad_norm": 7.849559783935547,
+      "learning_rate": 4.4845712235378546e-05,
+      "loss": 1.9519,
+      "step": 2880
+    },
+    {
+      "epoch": 0.5184786508790814,
+      "grad_norm": 7.759171962738037,
+      "learning_rate": 4.4827771797631863e-05,
+      "loss": 1.853,
+      "step": 2890
+    },
+    {
+      "epoch": 0.5202726946537496,
+      "grad_norm": 12.3310546875,
+      "learning_rate": 4.480983135988519e-05,
+      "loss": 2.017,
+      "step": 2900
+    },
+    {
+      "epoch": 0.5220667384284177,
+      "grad_norm": 7.477339267730713,
+      "learning_rate": 4.4791890922138505e-05,
+      "loss": 1.8495,
+      "step": 2910
+    },
+    {
+      "epoch": 0.5238607822030857,
+      "grad_norm": 8.375692367553711,
+      "learning_rate": 4.477395048439182e-05,
+      "loss": 1.9654,
+      "step": 2920
+    },
+    {
+      "epoch": 0.5256548259777538,
+      "grad_norm": 7.368261337280273,
+      "learning_rate": 4.475601004664514e-05,
+      "loss": 1.8585,
+      "step": 2930
+    },
+    {
+      "epoch": 0.527448869752422,
+      "grad_norm": 7.570740699768066,
+      "learning_rate": 4.4738069608898456e-05,
+      "loss": 1.965,
+      "step": 2940
+    },
+    {
+      "epoch": 0.5292429135270901,
+      "grad_norm": 7.81984281539917,
+      "learning_rate": 4.4720129171151774e-05,
+      "loss": 2.0262,
+      "step": 2950
+    },
+    {
+      "epoch": 0.5310369573017582,
+      "grad_norm": 7.456051349639893,
+      "learning_rate": 4.47021887334051e-05,
+      "loss": 1.8934,
+      "step": 2960
+    },
+    {
+      "epoch": 0.5328310010764262,
+      "grad_norm": 8.479120254516602,
+      "learning_rate": 4.4684248295658415e-05,
+      "loss": 1.9117,
+      "step": 2970
+    },
+    {
+      "epoch": 0.5346250448510944,
+      "grad_norm": 8.73019027709961,
+      "learning_rate": 4.466630785791174e-05,
+      "loss": 2.0238,
+      "step": 2980
+    },
+    {
+      "epoch": 0.5364190886257625,
+      "grad_norm": 8.31013298034668,
+      "learning_rate": 4.4648367420165056e-05,
+      "loss": 1.9404,
+      "step": 2990
+    },
+    {
+      "epoch": 0.5382131324004306,
+      "grad_norm": 6.616645812988281,
+      "learning_rate": 4.4630426982418374e-05,
+      "loss": 1.9167,
+      "step": 3000
+    },
+    {
+      "epoch": 0.5400071761750986,
+      "grad_norm": 7.4681172370910645,
+      "learning_rate": 4.46124865446717e-05,
+      "loss": 2.0548,
+      "step": 3010
+    },
+    {
+      "epoch": 0.5418012199497668,
+      "grad_norm": 11.563379287719727,
+      "learning_rate": 4.459454610692501e-05,
+      "loss": 2.0929,
+      "step": 3020
+    },
+    {
+      "epoch": 0.5435952637244349,
+      "grad_norm": 7.389353275299072,
+      "learning_rate": 4.4576605669178325e-05,
+      "loss": 1.7729,
+      "step": 3030
+    },
+    {
+      "epoch": 0.545389307499103,
+      "grad_norm": 8.560032844543457,
+      "learning_rate": 4.455866523143165e-05,
+      "loss": 1.8803,
+      "step": 3040
+    },
+    {
+      "epoch": 0.547183351273771,
+      "grad_norm": 8.208505630493164,
+      "learning_rate": 4.454072479368497e-05,
+      "loss": 1.8715,
+      "step": 3050
+    },
+    {
+      "epoch": 0.5489773950484392,
+      "grad_norm": 8.2820463180542,
+      "learning_rate": 4.4522784355938284e-05,
+      "loss": 2.0147,
+      "step": 3060
+    },
+    {
+      "epoch": 0.5507714388231073,
+      "grad_norm": 7.4810285568237305,
+      "learning_rate": 4.450484391819161e-05,
+      "loss": 1.9301,
+      "step": 3070
+    },
+    {
+      "epoch": 0.5525654825977754,
+      "grad_norm": 7.442780017852783,
+      "learning_rate": 4.4486903480444925e-05,
+      "loss": 2.0132,
+      "step": 3080
+    },
+    {
+      "epoch": 0.5543595263724435,
+      "grad_norm": 7.857019424438477,
+      "learning_rate": 4.446896304269824e-05,
+      "loss": 2.1327,
+      "step": 3090
+    },
+    {
+      "epoch": 0.5561535701471116,
+      "grad_norm": 8.195714950561523,
+      "learning_rate": 4.4451022604951567e-05,
+      "loss": 1.9133,
+      "step": 3100
+    },
+    {
+      "epoch": 0.5579476139217797,
+      "grad_norm": 9.424678802490234,
+      "learning_rate": 4.4433082167204884e-05,
+      "loss": 1.8079,
+      "step": 3110
+    },
+    {
+      "epoch": 0.5597416576964478,
+      "grad_norm": 7.7518768310546875,
+      "learning_rate": 4.44151417294582e-05,
+      "loss": 2.0345,
+      "step": 3120
+    },
+    {
+      "epoch": 0.5615357014711159,
+      "grad_norm": 8.170982360839844,
+      "learning_rate": 4.439720129171152e-05,
+      "loss": 2.089,
+      "step": 3130
+    },
+    {
+      "epoch": 0.563329745245784,
+      "grad_norm": 7.443562030792236,
+      "learning_rate": 4.4379260853964836e-05,
+      "loss": 1.805,
+      "step": 3140
+    },
+    {
+      "epoch": 0.5651237890204521,
+      "grad_norm": 8.59354019165039,
+      "learning_rate": 4.436132041621816e-05,
+      "loss": 1.9759,
+      "step": 3150
+    },
+    {
+      "epoch": 0.5669178327951202,
+      "grad_norm": 13.471951484680176,
+      "learning_rate": 4.434337997847148e-05,
+      "loss": 1.7935,
+      "step": 3160
+    },
+    {
+      "epoch": 0.5687118765697883,
+      "grad_norm": 9.660313606262207,
+      "learning_rate": 4.4325439540724794e-05,
+      "loss": 1.9807,
+      "step": 3170
+    },
+    {
+      "epoch": 0.5705059203444564,
+      "grad_norm": 6.917825222015381,
+      "learning_rate": 4.430749910297812e-05,
+      "loss": 2.0491,
+      "step": 3180
+    },
+    {
+      "epoch": 0.5722999641191245,
+      "grad_norm": 9.734037399291992,
+      "learning_rate": 4.4289558665231435e-05,
+      "loss": 1.8836,
+      "step": 3190
+    },
+    {
+      "epoch": 0.5740940078937926,
+      "grad_norm": 7.515590190887451,
+      "learning_rate": 4.427161822748475e-05,
+      "loss": 1.9391,
+      "step": 3200
+    },
+    {
+      "epoch": 0.5758880516684607,
+      "grad_norm": 7.2530670166015625,
+      "learning_rate": 4.425367778973808e-05,
+      "loss": 1.8615,
+      "step": 3210
+    },
+    {
+      "epoch": 0.5776820954431288,
+      "grad_norm": 8.29240894317627,
+      "learning_rate": 4.4235737351991394e-05,
+      "loss": 2.0954,
+      "step": 3220
+    },
+    {
+      "epoch": 0.579476139217797,
+      "grad_norm": 8.519623756408691,
+      "learning_rate": 4.421779691424471e-05,
+      "loss": 1.983,
+      "step": 3230
+    },
+    {
+      "epoch": 0.581270182992465,
+      "grad_norm": 7.667181015014648,
+      "learning_rate": 4.419985647649803e-05,
+      "loss": 1.9309,
+      "step": 3240
+    },
+    {
+      "epoch": 0.5830642267671331,
+      "grad_norm": 9.281998634338379,
+      "learning_rate": 4.4181916038751346e-05,
+      "loss": 1.9057,
+      "step": 3250
+    },
+    {
+      "epoch": 0.5848582705418012,
+      "grad_norm": 7.8026123046875,
+      "learning_rate": 4.416397560100466e-05,
+      "loss": 1.8834,
+      "step": 3260
+    },
+    {
+      "epoch": 0.5866523143164694,
+      "grad_norm": 7.448451042175293,
+      "learning_rate": 4.414603516325799e-05,
+      "loss": 1.8942,
+      "step": 3270
+    },
+    {
+      "epoch": 0.5884463580911374,
+      "grad_norm": 8.204652786254883,
+      "learning_rate": 4.4128094725511304e-05,
+      "loss": 2.0565,
+      "step": 3280
+    },
+    {
+      "epoch": 0.5902404018658055,
+      "grad_norm": 8.598773002624512,
+      "learning_rate": 4.411015428776462e-05,
+      "loss": 1.9708,
+      "step": 3290
+    },
+    {
+      "epoch": 0.5920344456404736,
+      "grad_norm": 7.167200088500977,
+      "learning_rate": 4.4092213850017946e-05,
+      "loss": 1.8534,
+      "step": 3300
+    },
+    {
+      "epoch": 0.5938284894151418,
+      "grad_norm": 11.437542915344238,
+      "learning_rate": 4.407427341227126e-05,
+      "loss": 2.0349,
+      "step": 3310
+    },
+    {
+      "epoch": 0.5956225331898098,
+      "grad_norm": 8.82004451751709,
+      "learning_rate": 4.405633297452458e-05,
+      "loss": 1.9427,
+      "step": 3320
+    },
+    {
+      "epoch": 0.5974165769644779,
+      "grad_norm": 6.192866802215576,
+      "learning_rate": 4.40383925367779e-05,
+      "loss": 1.9727,
+      "step": 3330
+    },
+    {
+      "epoch": 0.599210620739146,
+      "grad_norm": 7.089540481567383,
+      "learning_rate": 4.4020452099031215e-05,
+      "loss": 1.9497,
+      "step": 3340
+    },
+    {
+      "epoch": 0.6010046645138142,
+      "grad_norm": 8.135286331176758,
+      "learning_rate": 4.400251166128454e-05,
+      "loss": 1.8405,
+      "step": 3350
+    },
+    {
+      "epoch": 0.6027987082884823,
+      "grad_norm": 8.762147903442383,
+      "learning_rate": 4.3984571223537856e-05,
+      "loss": 1.9011,
+      "step": 3360
+    },
+    {
+      "epoch": 0.6045927520631503,
+      "grad_norm": 9.658001899719238,
+      "learning_rate": 4.396663078579117e-05,
+      "loss": 1.8995,
+      "step": 3370
+    },
+    {
+      "epoch": 0.6063867958378184,
+      "grad_norm": 7.004101276397705,
+      "learning_rate": 4.39486903480445e-05,
+      "loss": 1.8548,
+      "step": 3380
+    },
+    {
+      "epoch": 0.6081808396124866,
+      "grad_norm": 7.552975177764893,
+      "learning_rate": 4.3930749910297814e-05,
+      "loss": 1.9911,
+      "step": 3390
+    },
+    {
+      "epoch": 0.6099748833871547,
+      "grad_norm": 7.0711774826049805,
+      "learning_rate": 4.391280947255113e-05,
+      "loss": 1.8611,
+      "step": 3400
+    },
+    {
+      "epoch": 0.6117689271618227,
+      "grad_norm": 9.78061580657959,
+      "learning_rate": 4.3894869034804456e-05,
+      "loss": 1.9661,
+      "step": 3410
+    },
+    {
+      "epoch": 0.6135629709364908,
+      "grad_norm": 8.065385818481445,
+      "learning_rate": 4.387692859705777e-05,
+      "loss": 1.975,
+      "step": 3420
+    },
+    {
+      "epoch": 0.615357014711159,
+      "grad_norm": 9.28231430053711,
+      "learning_rate": 4.385898815931109e-05,
+      "loss": 1.9083,
+      "step": 3430
+    },
+    {
+      "epoch": 0.6171510584858271,
+      "grad_norm": 8.710234642028809,
+      "learning_rate": 4.384104772156441e-05,
+      "loss": 2.0402,
+      "step": 3440
+    },
+    {
+      "epoch": 0.6189451022604952,
+      "grad_norm": 8.679749488830566,
+      "learning_rate": 4.3823107283817725e-05,
+      "loss": 1.9162,
+      "step": 3450
+    },
+    {
+      "epoch": 0.6207391460351632,
+      "grad_norm": 7.5028228759765625,
+      "learning_rate": 4.380516684607104e-05,
+      "loss": 2.0918,
+      "step": 3460
+    },
+    {
+      "epoch": 0.6225331898098314,
+      "grad_norm": 8.817152976989746,
+      "learning_rate": 4.3787226408324366e-05,
+      "loss": 1.9089,
+      "step": 3470
+    },
+    {
+      "epoch": 0.6243272335844995,
+      "grad_norm": 7.720176696777344,
+      "learning_rate": 4.376928597057768e-05,
+      "loss": 1.8395,
+      "step": 3480
+    },
+    {
+      "epoch": 0.6261212773591676,
+      "grad_norm": 8.804288864135742,
+      "learning_rate": 4.3751345532831e-05,
+      "loss": 2.0384,
+      "step": 3490
+    },
+    {
+      "epoch": 0.6279153211338356,
+      "grad_norm": 6.383409023284912,
+      "learning_rate": 4.3733405095084325e-05,
+      "loss": 1.917,
+      "step": 3500
+    },
+    {
+      "epoch": 0.6297093649085038,
+      "grad_norm": 7.6271233558654785,
+      "learning_rate": 4.371546465733764e-05,
+      "loss": 1.884,
+      "step": 3510
+    },
+    {
+      "epoch": 0.6315034086831719,
+      "grad_norm": 7.319206237792969,
+      "learning_rate": 4.3697524219590966e-05,
+      "loss": 2.0081,
+      "step": 3520
+    },
+    {
+      "epoch": 0.63329745245784,
+      "grad_norm": 7.624337196350098,
+      "learning_rate": 4.367958378184428e-05,
+      "loss": 1.9573,
+      "step": 3530
+    },
+    {
+      "epoch": 0.635091496232508,
+      "grad_norm": 8.105642318725586,
+      "learning_rate": 4.3661643344097594e-05,
+      "loss": 1.9264,
+      "step": 3540
+    },
+    {
+      "epoch": 0.6368855400071761,
+      "grad_norm": 6.624615669250488,
+      "learning_rate": 4.364370290635092e-05,
+      "loss": 1.9273,
+      "step": 3550
+    },
+    {
+      "epoch": 0.6386795837818443,
+      "grad_norm": 8.527731895446777,
+      "learning_rate": 4.3625762468604235e-05,
+      "loss": 1.8837,
+      "step": 3560
+    },
+    {
+      "epoch": 0.6404736275565124,
+      "grad_norm": 8.374712944030762,
+      "learning_rate": 4.360782203085755e-05,
+      "loss": 1.9182,
+      "step": 3570
+    },
+    {
+      "epoch": 0.6422676713311805,
+      "grad_norm": 6.726866245269775,
+      "learning_rate": 4.3589881593110876e-05,
+      "loss": 1.7699,
+      "step": 3580
+    },
+    {
+      "epoch": 0.6440617151058485,
+      "grad_norm": 10.059771537780762,
+      "learning_rate": 4.3571941155364194e-05,
+      "loss": 1.8803,
+      "step": 3590
+    },
+    {
+      "epoch": 0.6458557588805167,
+      "grad_norm": 6.3898024559021,
+      "learning_rate": 4.355400071761751e-05,
+      "loss": 1.9215,
+      "step": 3600
+    },
+    {
+      "epoch": 0.6476498026551848,
+      "grad_norm": 9.93798542022705,
+      "learning_rate": 4.3536060279870835e-05,
+      "loss": 1.9015,
+      "step": 3610
+    },
+    {
+      "epoch": 0.6494438464298529,
+      "grad_norm": 8.230923652648926,
+      "learning_rate": 4.351811984212415e-05,
+      "loss": 1.8842,
+      "step": 3620
+    },
+    {
+      "epoch": 0.6512378902045209,
+      "grad_norm": 8.391541481018066,
+      "learning_rate": 4.350017940437747e-05,
+      "loss": 2.0254,
+      "step": 3630
+    },
+    {
+      "epoch": 0.6530319339791891,
+      "grad_norm": 8.703720092773438,
+      "learning_rate": 4.3482238966630787e-05,
+      "loss": 1.9441,
+      "step": 3640
+    },
+    {
+      "epoch": 0.6548259777538572,
+      "grad_norm": 8.434385299682617,
+      "learning_rate": 4.3464298528884104e-05,
+      "loss": 2.1388,
+      "step": 3650
+    },
+    {
+      "epoch": 0.6566200215285253,
+      "grad_norm": 9.658421516418457,
+      "learning_rate": 4.344635809113742e-05,
+      "loss": 2.1284,
+      "step": 3660
+    },
+    {
+      "epoch": 0.6584140653031934,
+      "grad_norm": 9.099438667297363,
+      "learning_rate": 4.3428417653390745e-05,
+      "loss": 1.9191,
+      "step": 3670
+    },
+    {
+      "epoch": 0.6602081090778615,
+      "grad_norm": 5.330417633056641,
+      "learning_rate": 4.341047721564406e-05,
+      "loss": 1.7558,
+      "step": 3680
+    },
+    {
+      "epoch": 0.6620021528525296,
+      "grad_norm": 8.256141662597656,
+      "learning_rate": 4.339253677789738e-05,
+      "loss": 1.904,
+      "step": 3690
+    },
+    {
+      "epoch": 0.6637961966271977,
+      "grad_norm": 7.978524208068848,
+      "learning_rate": 4.3374596340150704e-05,
+      "loss": 1.9706,
+      "step": 3700
+    },
+    {
+      "epoch": 0.6655902404018658,
+      "grad_norm": 7.254574298858643,
+      "learning_rate": 4.335665590240402e-05,
+      "loss": 1.8038,
+      "step": 3710
+    },
+    {
+      "epoch": 0.667384284176534,
+      "grad_norm": 6.119344711303711,
+      "learning_rate": 4.3338715464657345e-05,
+      "loss": 2.0935,
+      "step": 3720
+    },
+    {
+      "epoch": 0.669178327951202,
+      "grad_norm": 10.237481117248535,
+      "learning_rate": 4.332077502691066e-05,
+      "loss": 1.9197,
+      "step": 3730
+    },
+    {
+      "epoch": 0.6709723717258701,
+      "grad_norm": 8.238425254821777,
+      "learning_rate": 4.330283458916398e-05,
+      "loss": 1.8335,
+      "step": 3740
+    },
+    {
+      "epoch": 0.6727664155005382,
+      "grad_norm": 10.77346420288086,
+      "learning_rate": 4.32848941514173e-05,
+      "loss": 1.9849,
+      "step": 3750
+    },
+    {
+      "epoch": 0.6745604592752064,
+      "grad_norm": 5.878365993499756,
+      "learning_rate": 4.3266953713670614e-05,
+      "loss": 1.8184,
+      "step": 3760
+    },
+    {
+      "epoch": 0.6763545030498744,
+      "grad_norm": 5.962438583374023,
+      "learning_rate": 4.324901327592393e-05,
+      "loss": 1.942,
+      "step": 3770
+    },
+    {
+      "epoch": 0.6781485468245425,
+      "grad_norm": 7.118070602416992,
+      "learning_rate": 4.3231072838177255e-05,
+      "loss": 1.8888,
+      "step": 3780
+    },
+    {
+      "epoch": 0.6799425905992106,
+      "grad_norm": 7.19279146194458,
+      "learning_rate": 4.321313240043057e-05,
+      "loss": 1.9875,
+      "step": 3790
+    },
+    {
+      "epoch": 0.6817366343738788,
+      "grad_norm": 9.004045486450195,
+      "learning_rate": 4.319519196268389e-05,
+      "loss": 1.8713,
+      "step": 3800
+    },
+    {
+      "epoch": 0.6835306781485468,
+      "grad_norm": 6.644644737243652,
+      "learning_rate": 4.3177251524937214e-05,
+      "loss": 1.8883,
+      "step": 3810
+    },
+    {
+      "epoch": 0.6853247219232149,
+      "grad_norm": 7.53091287612915,
+      "learning_rate": 4.315931108719053e-05,
+      "loss": 1.9133,
+      "step": 3820
+    },
+    {
+      "epoch": 0.687118765697883,
+      "grad_norm": 7.283283710479736,
+      "learning_rate": 4.314137064944385e-05,
+      "loss": 1.8786,
+      "step": 3830
+    },
+    {
+      "epoch": 0.6889128094725512,
+      "grad_norm": 8.955952644348145,
+      "learning_rate": 4.3123430211697166e-05,
+      "loss": 1.9249,
+      "step": 3840
+    },
+    {
+      "epoch": 0.6907068532472193,
+      "grad_norm": 6.595334053039551,
+      "learning_rate": 4.310548977395048e-05,
+      "loss": 1.8995,
+      "step": 3850
+    },
+    {
+      "epoch": 0.6925008970218873,
+      "grad_norm": 6.214903831481934,
+      "learning_rate": 4.30875493362038e-05,
+      "loss": 1.8988,
+      "step": 3860
+    },
+    {
+      "epoch": 0.6942949407965554,
+      "grad_norm": 5.9941182136535645,
+      "learning_rate": 4.3069608898457124e-05,
+      "loss": 1.8008,
+      "step": 3870
+    },
+    {
+      "epoch": 0.6960889845712236,
+      "grad_norm": 9.003954887390137,
+      "learning_rate": 4.305166846071044e-05,
+      "loss": 2.0354,
+      "step": 3880
+    },
+    {
+      "epoch": 0.6978830283458917,
+      "grad_norm": 7.329159259796143,
+      "learning_rate": 4.303372802296376e-05,
+      "loss": 2.0086,
+      "step": 3890
+    },
+    {
+      "epoch": 0.6996770721205597,
+      "grad_norm": 7.872637748718262,
+      "learning_rate": 4.301578758521708e-05,
+      "loss": 1.9595,
+      "step": 3900
+    },
+    {
+      "epoch": 0.7014711158952278,
+      "grad_norm": 10.439992904663086,
+      "learning_rate": 4.29978471474704e-05,
+      "loss": 1.9187,
+      "step": 3910
+    },
+    {
+      "epoch": 0.703265159669896,
+      "grad_norm": 10.078546524047852,
+      "learning_rate": 4.2979906709723724e-05,
+      "loss": 1.9979,
+      "step": 3920
+    },
+    {
+      "epoch": 0.7050592034445641,
+      "grad_norm": 8.9290771484375,
+      "learning_rate": 4.296196627197704e-05,
+      "loss": 1.9665,
+      "step": 3930
+    },
+    {
+      "epoch": 0.7068532472192322,
+      "grad_norm": 8.043295860290527,
+      "learning_rate": 4.294402583423036e-05,
+      "loss": 2.0803,
+      "step": 3940
+    },
+    {
+      "epoch": 0.7086472909939002,
+      "grad_norm": 7.653200149536133,
+      "learning_rate": 4.2926085396483676e-05,
+      "loss": 2.1244,
+      "step": 3950
+    },
+    {
+      "epoch": 0.7104413347685683,
+      "grad_norm": 12.027050971984863,
+      "learning_rate": 4.290814495873699e-05,
+      "loss": 2.0046,
+      "step": 3960
+    },
+    {
+      "epoch": 0.7122353785432365,
+      "grad_norm": 6.169637680053711,
+      "learning_rate": 4.289020452099031e-05,
+      "loss": 1.9957,
+      "step": 3970
+    },
+    {
+      "epoch": 0.7140294223179046,
+      "grad_norm": 8.492105484008789,
+      "learning_rate": 4.2872264083243634e-05,
+      "loss": 1.9959,
+      "step": 3980
+    },
+    {
+      "epoch": 0.7158234660925726,
+      "grad_norm": 7.080846309661865,
+      "learning_rate": 4.285432364549695e-05,
+      "loss": 1.9098,
+      "step": 3990
+    },
+    {
+      "epoch": 0.7176175098672407,
+      "grad_norm": 7.8670454025268555,
+      "learning_rate": 4.283638320775027e-05,
+      "loss": 1.8557,
+      "step": 4000
+    },
+    {
+      "epoch": 0.7194115536419089,
+      "grad_norm": 8.910687446594238,
+      "learning_rate": 4.281844277000359e-05,
+      "loss": 2.0108,
+      "step": 4010
+    },
+    {
+      "epoch": 0.721205597416577,
+      "grad_norm": 7.866197109222412,
+      "learning_rate": 4.280050233225691e-05,
+      "loss": 1.9711,
+      "step": 4020
+    },
+    {
+      "epoch": 0.722999641191245,
+      "grad_norm": 7.0763044357299805,
+      "learning_rate": 4.278256189451023e-05,
+      "loss": 1.9565,
+      "step": 4030
+    },
+    {
+      "epoch": 0.7247936849659131,
+      "grad_norm": 10.510825157165527,
+      "learning_rate": 4.276462145676355e-05,
+      "loss": 1.9513,
+      "step": 4040
+    },
+    {
+      "epoch": 0.7265877287405813,
+      "grad_norm": 6.813848495483398,
+      "learning_rate": 4.274668101901687e-05,
+      "loss": 1.9841,
+      "step": 4050
+    },
+    {
+      "epoch": 0.7283817725152494,
+      "grad_norm": 8.661791801452637,
+      "learning_rate": 4.272874058127018e-05,
+      "loss": 1.9382,
+      "step": 4060
+    },
+    {
+      "epoch": 0.7301758162899175,
+      "grad_norm": 8.576088905334473,
+      "learning_rate": 4.27108001435235e-05,
+      "loss": 1.7463,
+      "step": 4070
+    },
+    {
+      "epoch": 0.7319698600645855,
+      "grad_norm": 9.310657501220703,
+      "learning_rate": 4.269285970577682e-05,
+      "loss": 1.9832,
+      "step": 4080
+    },
+    {
+      "epoch": 0.7337639038392537,
+      "grad_norm": 8.533422470092773,
+      "learning_rate": 4.2674919268030145e-05,
+      "loss": 2.0184,
+      "step": 4090
+    },
+    {
+      "epoch": 0.7355579476139218,
+      "grad_norm": 6.781728267669678,
+      "learning_rate": 4.265697883028346e-05,
+      "loss": 1.9384,
+      "step": 4100
+    },
+    {
+      "epoch": 0.7373519913885899,
+      "grad_norm": 7.908256530761719,
+      "learning_rate": 4.263903839253678e-05,
+      "loss": 1.9602,
+      "step": 4110
+    },
+    {
+      "epoch": 0.7391460351632579,
+      "grad_norm": 6.28724479675293,
+      "learning_rate": 4.26210979547901e-05,
+      "loss": 2.0066,
+      "step": 4120
+    },
+    {
+      "epoch": 0.7409400789379261,
+      "grad_norm": 6.198331832885742,
+      "learning_rate": 4.260315751704342e-05,
+      "loss": 2.132,
+      "step": 4130
+    },
+    {
+      "epoch": 0.7427341227125942,
+      "grad_norm": 7.816977024078369,
+      "learning_rate": 4.258521707929674e-05,
+      "loss": 1.9536,
+      "step": 4140
+    },
+    {
+      "epoch": 0.7445281664872623,
+      "grad_norm": 7.433613300323486,
+      "learning_rate": 4.2567276641550055e-05,
+      "loss": 1.885,
+      "step": 4150
+    },
+    {
+      "epoch": 0.7463222102619304,
+      "grad_norm": 6.046330451965332,
+      "learning_rate": 4.254933620380337e-05,
+      "loss": 1.7975,
+      "step": 4160
+    },
+    {
+      "epoch": 0.7481162540365985,
+      "grad_norm": 8.465211868286133,
+      "learning_rate": 4.253139576605669e-05,
+      "loss": 2.0454,
+      "step": 4170
+    },
+    {
+      "epoch": 0.7499102978112666,
+      "grad_norm": 5.859003067016602,
+      "learning_rate": 4.2513455328310013e-05,
+      "loss": 2.0586,
+      "step": 4180
+    },
+    {
+      "epoch": 0.7517043415859347,
+      "grad_norm": 8.045632362365723,
+      "learning_rate": 4.249551489056333e-05,
+      "loss": 1.9465,
+      "step": 4190
+    },
+    {
+      "epoch": 0.7534983853606028,
+      "grad_norm": 6.814916133880615,
+      "learning_rate": 4.247757445281665e-05,
+      "loss": 1.8779,
+      "step": 4200
+    },
+    {
+      "epoch": 0.755292429135271,
+      "grad_norm": 7.628875732421875,
+      "learning_rate": 4.245963401506997e-05,
+      "loss": 2.1836,
+      "step": 4210
+    },
+    {
+      "epoch": 0.757086472909939,
+      "grad_norm": 6.975657939910889,
+      "learning_rate": 4.244169357732329e-05,
+      "loss": 1.7787,
+      "step": 4220
+    },
+    {
+      "epoch": 0.7588805166846071,
+      "grad_norm": 7.706836700439453,
+      "learning_rate": 4.2423753139576606e-05,
+      "loss": 1.9749,
+      "step": 4230
+    },
+    {
+      "epoch": 0.7606745604592752,
+      "grad_norm": 8.721719741821289,
+      "learning_rate": 4.240581270182993e-05,
+      "loss": 1.9422,
+      "step": 4240
+    },
+    {
+      "epoch": 0.7624686042339434,
+      "grad_norm": 8.833395957946777,
+      "learning_rate": 4.238787226408325e-05,
+      "loss": 2.0886,
+      "step": 4250
+    },
+    {
+      "epoch": 0.7642626480086114,
+      "grad_norm": 8.308511734008789,
+      "learning_rate": 4.2369931826336565e-05,
+      "loss": 1.9319,
+      "step": 4260
+    },
+    {
+      "epoch": 0.7660566917832795,
+      "grad_norm": 7.8894476890563965,
+      "learning_rate": 4.235199138858988e-05,
+      "loss": 1.9378,
+      "step": 4270
+    },
+    {
+      "epoch": 0.7678507355579476,
+      "grad_norm": 6.138456344604492,
+      "learning_rate": 4.23340509508432e-05,
+      "loss": 1.9009,
+      "step": 4280
+    },
+    {
+      "epoch": 0.7696447793326158,
+      "grad_norm": 7.507815837860107,
+      "learning_rate": 4.2316110513096524e-05,
+      "loss": 1.9461,
+      "step": 4290
+    },
+    {
+      "epoch": 0.7714388231072838,
+      "grad_norm": 7.388694763183594,
+      "learning_rate": 4.229817007534984e-05,
+      "loss": 1.7823,
+      "step": 4300
+    },
+    {
+      "epoch": 0.7732328668819519,
+      "grad_norm": 7.483008861541748,
+      "learning_rate": 4.228022963760316e-05,
+      "loss": 2.0558,
+      "step": 4310
+    },
+    {
+      "epoch": 0.77502691065662,
+      "grad_norm": 8.476717948913574,
+      "learning_rate": 4.226228919985648e-05,
+      "loss": 1.97,
+      "step": 4320
+    },
+    {
+      "epoch": 0.7768209544312881,
+      "grad_norm": 8.137911796569824,
+      "learning_rate": 4.22443487621098e-05,
+      "loss": 2.0543,
+      "step": 4330
+    },
+    {
+      "epoch": 0.7786149982059563,
+      "grad_norm": 7.722476482391357,
+      "learning_rate": 4.222640832436312e-05,
+      "loss": 1.8836,
+      "step": 4340
+    },
+    {
+      "epoch": 0.7804090419806243,
+      "grad_norm": 7.449862003326416,
+      "learning_rate": 4.220846788661644e-05,
+      "loss": 1.9243,
+      "step": 4350
+    },
+    {
+      "epoch": 0.7822030857552924,
+      "grad_norm": 6.4095563888549805,
+      "learning_rate": 4.219052744886975e-05,
+      "loss": 1.8377,
+      "step": 4360
+    },
+    {
+      "epoch": 0.7839971295299605,
+      "grad_norm": 6.866125106811523,
+      "learning_rate": 4.217258701112307e-05,
+      "loss": 1.8246,
+      "step": 4370
+    },
+    {
+      "epoch": 0.7857911733046287,
+      "grad_norm": 9.195448875427246,
+      "learning_rate": 4.215464657337639e-05,
+      "loss": 1.8019,
+      "step": 4380
+    },
+    {
+      "epoch": 0.7875852170792967,
+      "grad_norm": 6.7557196617126465,
+      "learning_rate": 4.213670613562971e-05,
+      "loss": 1.8733,
+      "step": 4390
+    },
+    {
+      "epoch": 0.7893792608539648,
+      "grad_norm": 7.832233905792236,
+      "learning_rate": 4.211876569788303e-05,
+      "loss": 1.727,
+      "step": 4400
+    },
+    {
+      "epoch": 0.7911733046286329,
+      "grad_norm": 6.849626541137695,
+      "learning_rate": 4.210082526013635e-05,
+      "loss": 1.8051,
+      "step": 4410
+    },
+    {
+      "epoch": 0.7929673484033011,
+      "grad_norm": 7.668883323669434,
+      "learning_rate": 4.208288482238967e-05,
+      "loss": 1.9021,
+      "step": 4420
+    },
+    {
+      "epoch": 0.7947613921779692,
+      "grad_norm": 6.98665714263916,
+      "learning_rate": 4.2064944384642986e-05,
+      "loss": 1.9314,
+      "step": 4430
+    },
+    {
+      "epoch": 0.7965554359526372,
+      "grad_norm": 7.1887383460998535,
+      "learning_rate": 4.204700394689631e-05,
+      "loss": 1.8046,
+      "step": 4440
+    },
+    {
+      "epoch": 0.7983494797273053,
+      "grad_norm": 8.068037033081055,
+      "learning_rate": 4.202906350914963e-05,
+      "loss": 2.0018,
+      "step": 4450
+    },
+    {
+      "epoch": 0.8001435235019735,
+      "grad_norm": 8.662615776062012,
+      "learning_rate": 4.2011123071402944e-05,
+      "loss": 1.9545,
+      "step": 4460
+    },
+    {
+      "epoch": 0.8019375672766416,
+      "grad_norm": 6.354881286621094,
+      "learning_rate": 4.199318263365626e-05,
+      "loss": 1.699,
+      "step": 4470
+    },
+    {
+      "epoch": 0.8037316110513096,
+      "grad_norm": 7.5929059982299805,
+      "learning_rate": 4.197524219590958e-05,
+      "loss": 1.7685,
+      "step": 4480
+    },
+    {
+      "epoch": 0.8055256548259777,
+      "grad_norm": 7.677204608917236,
+      "learning_rate": 4.19573017581629e-05,
+      "loss": 1.876,
+      "step": 4490
+    },
+    {
+      "epoch": 0.8073196986006459,
+      "grad_norm": 6.630999565124512,
+      "learning_rate": 4.193936132041622e-05,
+      "loss": 1.6986,
+      "step": 4500
+    },
+    {
+      "epoch": 0.809113742375314,
+      "grad_norm": 6.430192947387695,
+      "learning_rate": 4.192142088266954e-05,
+      "loss": 1.9533,
+      "step": 4510
+    },
+    {
+      "epoch": 0.810907786149982,
+      "grad_norm": 8.051733016967773,
+      "learning_rate": 4.190348044492286e-05,
+      "loss": 1.8555,
+      "step": 4520
+    },
+    {
+      "epoch": 0.8127018299246501,
+      "grad_norm": 10.02955150604248,
+      "learning_rate": 4.188554000717618e-05,
+      "loss": 1.8035,
+      "step": 4530
+    },
+    {
+      "epoch": 0.8144958736993183,
+      "grad_norm": 9.238576889038086,
+      "learning_rate": 4.1867599569429496e-05,
+      "loss": 1.9818,
+      "step": 4540
+    },
+    {
+      "epoch": 0.8162899174739864,
+      "grad_norm": 6.678407192230225,
+      "learning_rate": 4.184965913168282e-05,
+      "loss": 1.808,
+      "step": 4550
+    },
+    {
+      "epoch": 0.8180839612486545,
+      "grad_norm": 8.583653450012207,
+      "learning_rate": 4.183171869393614e-05,
+      "loss": 1.7423,
+      "step": 4560
+    },
+    {
+      "epoch": 0.8198780050233225,
+      "grad_norm": 7.882835865020752,
+      "learning_rate": 4.1813778256189454e-05,
+      "loss": 1.9654,
+      "step": 4570
+    },
+    {
+      "epoch": 0.8216720487979907,
+      "grad_norm": 7.367639541625977,
+      "learning_rate": 4.179583781844277e-05,
+      "loss": 1.8759,
+      "step": 4580
+    },
+    {
+      "epoch": 0.8234660925726588,
+      "grad_norm": 5.808956146240234,
+      "learning_rate": 4.177789738069609e-05,
+      "loss": 1.8092,
+      "step": 4590
+    },
+    {
+      "epoch": 0.8252601363473269,
+      "grad_norm": 8.538715362548828,
+      "learning_rate": 4.1759956942949406e-05,
+      "loss": 1.9468,
+      "step": 4600
+    },
+    {
+      "epoch": 0.8270541801219949,
+      "grad_norm": 7.508995532989502,
+      "learning_rate": 4.174201650520273e-05,
+      "loss": 1.9887,
+      "step": 4610
+    },
+    {
+      "epoch": 0.8288482238966631,
+      "grad_norm": 7.2825446128845215,
+      "learning_rate": 4.172407606745605e-05,
+      "loss": 1.9767,
+      "step": 4620
+    },
+    {
+      "epoch": 0.8306422676713312,
+      "grad_norm": 7.998370170593262,
+      "learning_rate": 4.1706135629709365e-05,
+      "loss": 1.9921,
+      "step": 4630
+    },
+    {
+      "epoch": 0.8324363114459993,
+      "grad_norm": 6.514996528625488,
+      "learning_rate": 4.168819519196269e-05,
+      "loss": 1.8236,
+      "step": 4640
+    },
+    {
+      "epoch": 0.8342303552206674,
+      "grad_norm": 11.413960456848145,
+      "learning_rate": 4.1670254754216006e-05,
+      "loss": 1.9107,
+      "step": 4650
+    },
+    {
+      "epoch": 0.8360243989953355,
+      "grad_norm": 10.405426025390625,
+      "learning_rate": 4.165231431646932e-05,
+      "loss": 1.936,
+      "step": 4660
+    },
+    {
+      "epoch": 0.8378184427700036,
+      "grad_norm": 7.796232223510742,
+      "learning_rate": 4.163437387872264e-05,
+      "loss": 1.9083,
+      "step": 4670
+    },
+    {
+      "epoch": 0.8396124865446717,
+      "grad_norm": 7.413230895996094,
+      "learning_rate": 4.161643344097596e-05,
+      "loss": 1.9956,
+      "step": 4680
+    },
+    {
+      "epoch": 0.8414065303193398,
+      "grad_norm": 6.458949565887451,
+      "learning_rate": 4.159849300322928e-05,
+      "loss": 1.8525,
+      "step": 4690
+    },
+    {
+      "epoch": 0.843200574094008,
+      "grad_norm": 7.839207172393799,
+      "learning_rate": 4.15805525654826e-05,
+      "loss": 1.9213,
+      "step": 4700
+    },
+    {
+      "epoch": 0.844994617868676,
+      "grad_norm": 7.596676349639893,
+      "learning_rate": 4.1562612127735916e-05,
+      "loss": 1.7207,
+      "step": 4710
+    },
+    {
+      "epoch": 0.8467886616433441,
+      "grad_norm": 8.215608596801758,
+      "learning_rate": 4.154467168998924e-05,
+      "loss": 1.8861,
+      "step": 4720
+    },
+    {
+      "epoch": 0.8485827054180122,
+      "grad_norm": 8.735016822814941,
+      "learning_rate": 4.152673125224256e-05,
+      "loss": 1.7922,
+      "step": 4730
+    },
+    {
+      "epoch": 0.8503767491926802,
+      "grad_norm": 7.5839667320251465,
+      "learning_rate": 4.1508790814495875e-05,
+      "loss": 1.8327,
+      "step": 4740
+    },
+    {
+      "epoch": 0.8521707929673484,
+      "grad_norm": 8.134754180908203,
+      "learning_rate": 4.14908503767492e-05,
+      "loss": 2.0809,
+      "step": 4750
+    },
+    {
+      "epoch": 0.8539648367420165,
+      "grad_norm": 6.481571674346924,
+      "learning_rate": 4.1472909939002516e-05,
+      "loss": 1.8645,
+      "step": 4760
+    },
+    {
+      "epoch": 0.8557588805166846,
+      "grad_norm": 6.660789489746094,
+      "learning_rate": 4.145496950125583e-05,
+      "loss": 1.8249,
+      "step": 4770
+    },
+    {
+      "epoch": 0.8575529242913527,
+      "grad_norm": 9.439289093017578,
+      "learning_rate": 4.143702906350915e-05,
+      "loss": 1.9101,
+      "step": 4780
+    },
+    {
+      "epoch": 0.8593469680660208,
+      "grad_norm": 7.637185096740723,
+      "learning_rate": 4.141908862576247e-05,
+      "loss": 1.8607,
+      "step": 4790
+    },
+    {
+      "epoch": 0.8611410118406889,
+      "grad_norm": 6.203429698944092,
+      "learning_rate": 4.1401148188015785e-05,
+      "loss": 1.6808,
+      "step": 4800
+    },
+    {
+      "epoch": 0.862935055615357,
+      "grad_norm": 7.630656719207764,
+      "learning_rate": 4.138320775026911e-05,
+      "loss": 1.8628,
+      "step": 4810
+    },
+    {
+      "epoch": 0.8647290993900251,
+      "grad_norm": 7.22261905670166,
+      "learning_rate": 4.1365267312522426e-05,
+      "loss": 1.8176,
+      "step": 4820
+    },
+    {
+      "epoch": 0.8665231431646933,
+      "grad_norm": 7.170011520385742,
+      "learning_rate": 4.134732687477575e-05,
+      "loss": 1.9082,
+      "step": 4830
+    },
+    {
+      "epoch": 0.8683171869393613,
+      "grad_norm": 6.344324588775635,
+      "learning_rate": 4.132938643702907e-05,
+      "loss": 1.8772,
+      "step": 4840
+    },
+    {
+      "epoch": 0.8701112307140294,
+      "grad_norm": 8.296944618225098,
+      "learning_rate": 4.1311445999282385e-05,
+      "loss": 1.7995,
+      "step": 4850
+    },
+    {
+      "epoch": 0.8719052744886975,
+      "grad_norm": 10.2393159866333,
+      "learning_rate": 4.129350556153571e-05,
+      "loss": 1.7471,
+      "step": 4860
+    },
+    {
+      "epoch": 0.8736993182633657,
+      "grad_norm": 6.657433032989502,
+      "learning_rate": 4.127556512378902e-05,
+      "loss": 1.8175,
+      "step": 4870
+    },
+    {
+      "epoch": 0.8754933620380337,
+      "grad_norm": 9.101529121398926,
+      "learning_rate": 4.125762468604234e-05,
+      "loss": 1.8132,
+      "step": 4880
+    },
+    {
+      "epoch": 0.8772874058127018,
+      "grad_norm": 8.015061378479004,
+      "learning_rate": 4.123968424829566e-05,
+      "loss": 1.7798,
+      "step": 4890
+    },
+    {
+      "epoch": 0.8790814495873699,
+      "grad_norm": 8.774176597595215,
+      "learning_rate": 4.122174381054898e-05,
+      "loss": 1.823,
+      "step": 4900
+    },
+    {
+      "epoch": 0.8808754933620381,
+      "grad_norm": 8.144107818603516,
+      "learning_rate": 4.1203803372802295e-05,
+      "loss": 1.8932,
+      "step": 4910
+    },
+    {
+      "epoch": 0.8826695371367062,
+      "grad_norm": 9.238556861877441,
+      "learning_rate": 4.118586293505562e-05,
+      "loss": 1.9033,
+      "step": 4920
+    },
+    {
+      "epoch": 0.8844635809113742,
+      "grad_norm": 7.814840316772461,
+      "learning_rate": 4.1167922497308937e-05,
+      "loss": 1.9437,
+      "step": 4930
+    },
+    {
+      "epoch": 0.8862576246860423,
+      "grad_norm": 8.720809936523438,
+      "learning_rate": 4.1149982059562254e-05,
+      "loss": 2.0195,
+      "step": 4940
+    },
+    {
+      "epoch": 0.8880516684607105,
+      "grad_norm": 8.207976341247559,
+      "learning_rate": 4.113204162181558e-05,
+      "loss": 1.6992,
+      "step": 4950
+    },
+    {
+      "epoch": 0.8898457122353786,
+      "grad_norm": 6.2852301597595215,
+      "learning_rate": 4.1114101184068895e-05,
+      "loss": 1.9237,
+      "step": 4960
+    },
+    {
+      "epoch": 0.8916397560100466,
+      "grad_norm": 7.93939208984375,
+      "learning_rate": 4.109616074632221e-05,
+      "loss": 1.8576,
+      "step": 4970
+    },
+    {
+      "epoch": 0.8934337997847147,
+      "grad_norm": 7.219597339630127,
+      "learning_rate": 4.107822030857553e-05,
+      "loss": 1.9087,
+      "step": 4980
+    },
+    {
+      "epoch": 0.8952278435593829,
+      "grad_norm": 6.388714790344238,
+      "learning_rate": 4.106027987082885e-05,
+      "loss": 1.8785,
+      "step": 4990
+    },
+    {
+      "epoch": 0.897021887334051,
+      "grad_norm": 7.048055648803711,
+      "learning_rate": 4.1042339433082164e-05,
+      "loss": 1.7598,
+      "step": 5000
+    },
+    {
+      "epoch": 0.898815931108719,
+      "grad_norm": 9.254620552062988,
+      "learning_rate": 4.102439899533549e-05,
+      "loss": 1.8903,
+      "step": 5010
+    },
+    {
+      "epoch": 0.9006099748833871,
+      "grad_norm": 7.499938488006592,
+      "learning_rate": 4.1006458557588805e-05,
+      "loss": 2.0262,
+      "step": 5020
+    },
+    {
+      "epoch": 0.9024040186580553,
+      "grad_norm": 7.229835033416748,
+      "learning_rate": 4.098851811984213e-05,
+      "loss": 1.7852,
+      "step": 5030
+    },
+    {
+      "epoch": 0.9041980624327234,
+      "grad_norm": 7.887816429138184,
+      "learning_rate": 4.097057768209545e-05,
+      "loss": 1.8154,
+      "step": 5040
+    },
+    {
+      "epoch": 0.9059921062073915,
+      "grad_norm": 11.909371376037598,
+      "learning_rate": 4.0952637244348764e-05,
+      "loss": 1.8699,
+      "step": 5050
+    },
+    {
+      "epoch": 0.9077861499820595,
+      "grad_norm": 9.439387321472168,
+      "learning_rate": 4.093469680660209e-05,
+      "loss": 2.0234,
+      "step": 5060
+    },
+    {
+      "epoch": 0.9095801937567277,
+      "grad_norm": 7.028390884399414,
+      "learning_rate": 4.0916756368855405e-05,
+      "loss": 1.8577,
+      "step": 5070
+    },
+    {
+      "epoch": 0.9113742375313958,
+      "grad_norm": 8.708728790283203,
+      "learning_rate": 4.089881593110872e-05,
+      "loss": 1.8517,
+      "step": 5080
+    },
+    {
+      "epoch": 0.9131682813060639,
+      "grad_norm": 7.153110027313232,
+      "learning_rate": 4.088087549336204e-05,
+      "loss": 1.8174,
+      "step": 5090
+    },
+    {
+      "epoch": 0.9149623250807319,
+      "grad_norm": 10.017348289489746,
+      "learning_rate": 4.086293505561536e-05,
+      "loss": 1.7079,
+      "step": 5100
+    },
+    {
+      "epoch": 0.9167563688554,
+      "grad_norm": 8.1173734664917,
+      "learning_rate": 4.0844994617868674e-05,
+      "loss": 1.914,
+      "step": 5110
+    },
+    {
+      "epoch": 0.9185504126300682,
+      "grad_norm": 7.267770290374756,
+      "learning_rate": 4.0827054180122e-05,
+      "loss": 1.8668,
+      "step": 5120
+    },
+    {
+      "epoch": 0.9203444564047363,
+      "grad_norm": 8.86645793914795,
+      "learning_rate": 4.0809113742375316e-05,
+      "loss": 1.8357,
+      "step": 5130
+    },
+    {
+      "epoch": 0.9221385001794044,
+      "grad_norm": 8.062966346740723,
+      "learning_rate": 4.079117330462863e-05,
+      "loss": 1.8604,
+      "step": 5140
+    },
+    {
+      "epoch": 0.9239325439540724,
+      "grad_norm": 7.72880220413208,
+      "learning_rate": 4.077323286688196e-05,
+      "loss": 1.9482,
+      "step": 5150
+    },
+    {
+      "epoch": 0.9257265877287406,
+      "grad_norm": 8.267526626586914,
+      "learning_rate": 4.0755292429135274e-05,
+      "loss": 1.6839,
+      "step": 5160
+    },
+    {
+      "epoch": 0.9275206315034087,
+      "grad_norm": 10.039454460144043,
+      "learning_rate": 4.073735199138859e-05,
+      "loss": 1.9365,
+      "step": 5170
+    },
+    {
+      "epoch": 0.9293146752780768,
+      "grad_norm": 9.875499725341797,
+      "learning_rate": 4.071941155364191e-05,
+      "loss": 1.8115,
+      "step": 5180
+    },
+    {
+      "epoch": 0.9311087190527448,
+      "grad_norm": 7.694703102111816,
+      "learning_rate": 4.0701471115895226e-05,
+      "loss": 1.8054,
+      "step": 5190
+    },
+    {
+      "epoch": 0.932902762827413,
+      "grad_norm": 6.072929382324219,
+      "learning_rate": 4.068353067814855e-05,
+      "loss": 1.9556,
+      "step": 5200
+    },
+    {
+      "epoch": 0.9346968066020811,
+      "grad_norm": 9.701952934265137,
+      "learning_rate": 4.066559024040187e-05,
+      "loss": 1.9576,
+      "step": 5210
+    },
+    {
+      "epoch": 0.9364908503767492,
+      "grad_norm": 8.353795051574707,
+      "learning_rate": 4.0647649802655185e-05,
+      "loss": 1.8579,
+      "step": 5220
+    },
+    {
+      "epoch": 0.9382848941514172,
+      "grad_norm": 7.89420747756958,
+      "learning_rate": 4.062970936490851e-05,
+      "loss": 1.8866,
+      "step": 5230
+    },
+    {
+      "epoch": 0.9400789379260854,
+      "grad_norm": 5.647555351257324,
+      "learning_rate": 4.0611768927161826e-05,
+      "loss": 1.742,
+      "step": 5240
+    },
+    {
+      "epoch": 0.9418729817007535,
+      "grad_norm": 8.360129356384277,
+      "learning_rate": 4.059382848941514e-05,
+      "loss": 1.9297,
+      "step": 5250
+    },
+    {
+      "epoch": 0.9436670254754216,
+      "grad_norm": 10.252833366394043,
+      "learning_rate": 4.057588805166847e-05,
+      "loss": 1.9036,
+      "step": 5260
+    },
+    {
+      "epoch": 0.9454610692500897,
+      "grad_norm": 8.29806900024414,
+      "learning_rate": 4.0557947613921784e-05,
+      "loss": 1.7497,
+      "step": 5270
+    },
+    {
+      "epoch": 0.9472551130247578,
+      "grad_norm": 6.64280891418457,
+      "learning_rate": 4.05400071761751e-05,
+      "loss": 1.9239,
+      "step": 5280
+    },
+    {
+      "epoch": 0.9490491567994259,
+      "grad_norm": 7.992321014404297,
+      "learning_rate": 4.052206673842842e-05,
+      "loss": 1.8058,
+      "step": 5290
+    },
+    {
+      "epoch": 0.950843200574094,
+      "grad_norm": 9.094325065612793,
+      "learning_rate": 4.0504126300681736e-05,
+      "loss": 1.9108,
+      "step": 5300
+    },
+    {
+      "epoch": 0.9526372443487621,
+      "grad_norm": 8.64013385772705,
+      "learning_rate": 4.048618586293505e-05,
+      "loss": 2.1198,
+      "step": 5310
+    },
+    {
+      "epoch": 0.9544312881234303,
+      "grad_norm": 6.910890579223633,
+      "learning_rate": 4.046824542518838e-05,
+      "loss": 1.9502,
+      "step": 5320
+    },
+    {
+      "epoch": 0.9562253318980983,
+      "grad_norm": 7.383171081542969,
+      "learning_rate": 4.0450304987441695e-05,
+      "loss": 1.7905,
+      "step": 5330
+    },
+    {
+      "epoch": 0.9580193756727664,
+      "grad_norm": 8.704651832580566,
+      "learning_rate": 4.043236454969501e-05,
+      "loss": 1.9428,
+      "step": 5340
+    },
+    {
+      "epoch": 0.9598134194474345,
+      "grad_norm": 8.296703338623047,
+      "learning_rate": 4.0414424111948336e-05,
+      "loss": 1.8346,
+      "step": 5350
+    },
+    {
+      "epoch": 0.9616074632221027,
+      "grad_norm": 7.613495349884033,
+      "learning_rate": 4.039648367420165e-05,
+      "loss": 1.9988,
+      "step": 5360
+    },
+    {
+      "epoch": 0.9634015069967707,
+      "grad_norm": 8.145048141479492,
+      "learning_rate": 4.037854323645497e-05,
+      "loss": 1.8056,
+      "step": 5370
+    },
+    {
+      "epoch": 0.9651955507714388,
+      "grad_norm": 9.955933570861816,
+      "learning_rate": 4.0360602798708295e-05,
+      "loss": 1.87,
+      "step": 5380
+    },
+    {
+      "epoch": 0.9669895945461069,
+      "grad_norm": 5.713760852813721,
+      "learning_rate": 4.0342662360961605e-05,
+      "loss": 1.8019,
+      "step": 5390
+    },
+    {
+      "epoch": 0.9687836383207751,
+      "grad_norm": 9.013385772705078,
+      "learning_rate": 4.032472192321493e-05,
+      "loss": 1.922,
+      "step": 5400
+    },
+    {
+      "epoch": 0.9705776820954432,
+      "grad_norm": 8.361518859863281,
+      "learning_rate": 4.0306781485468246e-05,
+      "loss": 1.8847,
+      "step": 5410
+    },
+    {
+      "epoch": 0.9723717258701112,
+      "grad_norm": 8.547319412231445,
+      "learning_rate": 4.0288841047721564e-05,
+      "loss": 1.9232,
+      "step": 5420
+    },
+    {
+      "epoch": 0.9741657696447793,
+      "grad_norm": 8.168745994567871,
+      "learning_rate": 4.027090060997489e-05,
+      "loss": 1.834,
+      "step": 5430
+    },
+    {
+      "epoch": 0.9759598134194475,
+      "grad_norm": 7.9649810791015625,
+      "learning_rate": 4.0252960172228205e-05,
+      "loss": 1.8909,
+      "step": 5440
+    },
+    {
+      "epoch": 0.9777538571941156,
+      "grad_norm": 7.361401557922363,
+      "learning_rate": 4.023501973448152e-05,
+      "loss": 1.8859,
+      "step": 5450
+    },
+    {
+      "epoch": 0.9795479009687836,
+      "grad_norm": 6.616429328918457,
+      "learning_rate": 4.0217079296734846e-05,
+      "loss": 1.8936,
+      "step": 5460
+    },
+    {
+      "epoch": 0.9813419447434517,
+      "grad_norm": 8.482680320739746,
+      "learning_rate": 4.0199138858988163e-05,
+      "loss": 1.8714,
+      "step": 5470
+    },
+    {
+      "epoch": 0.9831359885181199,
+      "grad_norm": 7.886808395385742,
+      "learning_rate": 4.018119842124148e-05,
+      "loss": 1.8772,
+      "step": 5480
+    },
+    {
+      "epoch": 0.984930032292788,
+      "grad_norm": 8.63215446472168,
+      "learning_rate": 4.01632579834948e-05,
+      "loss": 1.8233,
+      "step": 5490
+    },
+    {
+      "epoch": 0.986724076067456,
+      "grad_norm": 7.060615539550781,
+      "learning_rate": 4.0145317545748115e-05,
+      "loss": 1.7991,
+      "step": 5500
+    },
+    {
+      "epoch": 0.9885181198421241,
+      "grad_norm": 7.455167293548584,
+      "learning_rate": 4.012737710800143e-05,
+      "loss": 1.8871,
+      "step": 5510
+    },
+    {
+      "epoch": 0.9903121636167922,
+      "grad_norm": 8.120213508605957,
+      "learning_rate": 4.0109436670254756e-05,
+      "loss": 1.7909,
+      "step": 5520
+    },
+    {
+      "epoch": 0.9921062073914604,
+      "grad_norm": 8.847984313964844,
+      "learning_rate": 4.0091496232508074e-05,
+      "loss": 1.9595,
+      "step": 5530
+    },
+    {
+      "epoch": 0.9939002511661285,
+      "grad_norm": 7.193493366241455,
+      "learning_rate": 4.007355579476139e-05,
+      "loss": 1.7494,
+      "step": 5540
+    },
+    {
+      "epoch": 0.9956942949407965,
+      "grad_norm": 9.286087989807129,
+      "learning_rate": 4.0055615357014715e-05,
+      "loss": 1.7659,
+      "step": 5550
+    },
+    {
+      "epoch": 0.9974883387154646,
+      "grad_norm": 7.888460159301758,
+      "learning_rate": 4.003767491926803e-05,
+      "loss": 1.9127,
+      "step": 5560
+    },
+    {
+      "epoch": 0.9992823824901328,
+      "grad_norm": 9.469123840332031,
+      "learning_rate": 4.0019734481521356e-05,
+      "loss": 1.9231,
+      "step": 5570
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 1.932178020477295,
+      "eval_runtime": 189.8033,
+      "eval_samples_per_second": 13.05,
+      "eval_steps_per_second": 13.05,
+      "step": 5574
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 27870,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 5,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.0165015541579776e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}