diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,6743 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.999843448311851,
+  "eval_steps": 500,
+  "global_step": 4790,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.78125,
+      "learning_rate": 4.175365344467641e-07,
+      "loss": 1.6189,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.46875,
+      "learning_rate": 2.0876826722338207e-06,
+      "loss": 1.7423,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 9.9375,
+      "learning_rate": 4.175365344467641e-06,
+      "loss": 1.6192,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 13.375,
+      "learning_rate": 6.2630480167014616e-06,
+      "loss": 1.557,
+      "step": 15
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 9.4375,
+      "learning_rate": 8.350730688935283e-06,
+      "loss": 1.5582,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 10.625,
+      "learning_rate": 1.0438413361169103e-05,
+      "loss": 1.6962,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.2526096033402923e-05,
+      "loss": 1.424,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.75,
+      "learning_rate": 1.4613778705636743e-05,
+      "loss": 1.4941,
+      "step": 35
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 8.3125,
+      "learning_rate": 1.6701461377870565e-05,
+      "loss": 1.4198,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 40.75,
+      "learning_rate": 1.8789144050104384e-05,
+      "loss": 1.3663,
+      "step": 45
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2.0876826722338206e-05,
+      "loss": 1.2479,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.2964509394572024e-05,
+      "loss": 1.5104,
+      "step": 55
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2.5052192066805846e-05,
+      "loss": 1.1935,
+      "step": 60
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.46875,
+      "learning_rate": 2.7139874739039668e-05,
+      "loss": 1.304,
+      "step": 65
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2.9227557411273487e-05,
+      "loss": 1.3205,
+      "step": 70
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.4140625,
+      "learning_rate": 3.131524008350731e-05,
+      "loss": 1.1825,
+      "step": 75
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.359375,
+      "learning_rate": 3.340292275574113e-05,
+      "loss": 1.17,
+      "step": 80
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.828125,
+      "learning_rate": 3.5490605427974946e-05,
+      "loss": 1.1457,
+      "step": 85
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.9296875,
+      "learning_rate": 3.757828810020877e-05,
+      "loss": 1.1369,
+      "step": 90
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.171875,
+      "learning_rate": 3.966597077244259e-05,
+      "loss": 1.1408,
+      "step": 95
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.63671875,
+      "learning_rate": 4.175365344467641e-05,
+      "loss": 1.1776,
+      "step": 100
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.91796875,
+      "learning_rate": 4.3841336116910233e-05,
+      "loss": 1.0779,
+      "step": 105
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.6484375,
+      "learning_rate": 4.592901878914405e-05,
+      "loss": 1.1247,
+      "step": 110
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.0078125,
+      "learning_rate": 4.801670146137787e-05,
+      "loss": 1.2182,
+      "step": 115
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.3125,
+      "learning_rate": 5.010438413361169e-05,
+      "loss": 1.2062,
+      "step": 120
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.75,
+      "learning_rate": 5.219206680584552e-05,
+      "loss": 1.1983,
+      "step": 125
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.5859375,
+      "learning_rate": 5.4279749478079336e-05,
+      "loss": 1.2013,
+      "step": 130
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.390625,
+      "learning_rate": 5.636743215031316e-05,
+      "loss": 1.158,
+      "step": 135
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.6484375,
+      "learning_rate": 5.8455114822546973e-05,
+      "loss": 1.1559,
+      "step": 140
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.4765625,
+      "learning_rate": 6.05427974947808e-05,
+      "loss": 1.1545,
+      "step": 145
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.578125,
+      "learning_rate": 6.263048016701462e-05,
+      "loss": 1.1553,
+      "step": 150
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.484375,
+      "learning_rate": 6.471816283924845e-05,
+      "loss": 1.0636,
+      "step": 155
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.6015625,
+      "learning_rate": 6.680584551148226e-05,
+      "loss": 1.1026,
+      "step": 160
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.82421875,
+      "learning_rate": 6.889352818371608e-05,
+      "loss": 1.1697,
+      "step": 165
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.0078125,
+      "learning_rate": 7.098121085594989e-05,
+      "loss": 1.1397,
+      "step": 170
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.52734375,
+      "learning_rate": 7.306889352818372e-05,
+      "loss": 1.1051,
+      "step": 175
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.53515625,
+      "learning_rate": 7.515657620041754e-05,
+      "loss": 1.0573,
+      "step": 180
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.41015625,
+      "learning_rate": 7.724425887265136e-05,
+      "loss": 1.1934,
+      "step": 185
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.5625,
+      "learning_rate": 7.933194154488518e-05,
+      "loss": 1.1273,
+      "step": 190
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.50390625,
+      "learning_rate": 8.141962421711901e-05,
+      "loss": 1.1812,
+      "step": 195
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.482421875,
+      "learning_rate": 8.350730688935282e-05,
+      "loss": 1.1121,
+      "step": 200
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.453125,
+      "learning_rate": 8.559498956158665e-05,
+      "loss": 1.1803,
+      "step": 205
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.50390625,
+      "learning_rate": 8.768267223382047e-05,
+      "loss": 1.139,
+      "step": 210
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.3984375,
+      "learning_rate": 8.977035490605428e-05,
+      "loss": 1.1549,
+      "step": 215
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.58203125,
+      "learning_rate": 9.18580375782881e-05,
+      "loss": 1.1616,
+      "step": 220
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.8203125,
+      "learning_rate": 9.394572025052193e-05,
+      "loss": 1.1901,
+      "step": 225
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.53125,
+      "learning_rate": 9.603340292275574e-05,
+      "loss": 1.121,
+      "step": 230
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.51953125,
+      "learning_rate": 9.812108559498957e-05,
+      "loss": 1.0502,
+      "step": 235
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.37109375,
+      "learning_rate": 0.00010020876826722338,
+      "loss": 1.0768,
+      "step": 240
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.470703125,
+      "learning_rate": 0.00010229645093945721,
+      "loss": 1.2241,
+      "step": 245
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.00010438413361169104,
+      "loss": 1.1032,
+      "step": 250
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.00010647181628392484,
+      "loss": 1.0418,
+      "step": 255
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.427734375,
+      "learning_rate": 0.00010855949895615867,
+      "loss": 1.1505,
+      "step": 260
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.5546875,
+      "learning_rate": 0.00011064718162839249,
+      "loss": 1.0452,
+      "step": 265
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.380859375,
+      "learning_rate": 0.00011273486430062632,
+      "loss": 1.062,
+      "step": 270
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.375,
+      "learning_rate": 0.00011482254697286012,
+      "loss": 1.1468,
+      "step": 275
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.390625,
+      "learning_rate": 0.00011691022964509395,
+      "loss": 1.1486,
+      "step": 280
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.47265625,
+      "learning_rate": 0.00011899791231732778,
+      "loss": 1.1573,
+      "step": 285
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.171875,
+      "learning_rate": 0.0001210855949895616,
+      "loss": 1.0818,
+      "step": 290
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.416015625,
+      "learning_rate": 0.0001231732776617954,
+      "loss": 1.1409,
+      "step": 295
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.466796875,
+      "learning_rate": 0.00012526096033402923,
+      "loss": 1.2038,
+      "step": 300
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.419921875,
+      "learning_rate": 0.00012734864300626306,
+      "loss": 1.1181,
+      "step": 305
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.380859375,
+      "learning_rate": 0.0001294363256784969,
+      "loss": 1.1275,
+      "step": 310
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.0001315240083507307,
+      "loss": 1.1049,
+      "step": 315
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.41796875,
+      "learning_rate": 0.00013361169102296452,
+      "loss": 1.0732,
+      "step": 320
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.00013569937369519835,
+      "loss": 1.1349,
+      "step": 325
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.00013778705636743215,
+      "loss": 1.1851,
+      "step": 330
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.00013987473903966598,
+      "loss": 1.1427,
+      "step": 335
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.00014196242171189978,
+      "loss": 1.1225,
+      "step": 340
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.333984375,
+      "learning_rate": 0.0001440501043841336,
+      "loss": 1.0863,
+      "step": 345
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.416015625,
+      "learning_rate": 0.00014613778705636744,
+      "loss": 1.1447,
+      "step": 350
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.00014822546972860124,
+      "loss": 1.0539,
+      "step": 355
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.37109375,
+      "learning_rate": 0.00015031315240083507,
+      "loss": 1.1486,
+      "step": 360
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.3515625,
+      "learning_rate": 0.0001524008350730689,
+      "loss": 1.0433,
+      "step": 365
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.380859375,
+      "learning_rate": 0.00015448851774530273,
+      "loss": 1.0943,
+      "step": 370
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.4140625,
+      "learning_rate": 0.00015657620041753653,
+      "loss": 1.0619,
+      "step": 375
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.416015625,
+      "learning_rate": 0.00015866388308977036,
+      "loss": 1.0754,
+      "step": 380
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.453125,
+      "learning_rate": 0.0001607515657620042,
+      "loss": 1.1213,
+      "step": 385
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.00016283924843423802,
+      "loss": 1.2721,
+      "step": 390
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.44921875,
+      "learning_rate": 0.00016492693110647182,
+      "loss": 1.0433,
+      "step": 395
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.00016701461377870565,
+      "loss": 1.0319,
+      "step": 400
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.48828125,
+      "learning_rate": 0.00016910229645093947,
+      "loss": 1.1656,
+      "step": 405
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.0001711899791231733,
+      "loss": 1.0521,
+      "step": 410
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.0001732776617954071,
+      "loss": 1.19,
+      "step": 415
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.361328125,
+      "learning_rate": 0.00017536534446764093,
+      "loss": 1.1473,
+      "step": 420
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.462890625,
+      "learning_rate": 0.00017745302713987476,
+      "loss": 1.0851,
+      "step": 425
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.00017954070981210856,
+      "loss": 1.0106,
+      "step": 430
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.416015625,
+      "learning_rate": 0.0001816283924843424,
+      "loss": 1.0673,
+      "step": 435
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.546875,
+      "learning_rate": 0.0001837160751565762,
+      "loss": 1.0706,
+      "step": 440
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.484375,
+      "learning_rate": 0.00018580375782881002,
+      "loss": 1.1294,
+      "step": 445
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.00018789144050104385,
+      "loss": 1.2233,
+      "step": 450
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.439453125,
+      "learning_rate": 0.00018997912317327765,
+      "loss": 1.0567,
+      "step": 455
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00019206680584551148,
+      "loss": 1.1792,
+      "step": 460
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.392578125,
+      "learning_rate": 0.0001941544885177453,
+      "loss": 1.2399,
+      "step": 465
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.455078125,
+      "learning_rate": 0.00019624217118997914,
+      "loss": 1.0533,
+      "step": 470
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.6796875,
+      "learning_rate": 0.00019832985386221294,
+      "loss": 1.1026,
+      "step": 475
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.455078125,
+      "learning_rate": 0.00019999997344699456,
+      "loss": 1.3194,
+      "step": 480
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.451171875,
+      "learning_rate": 0.000199999044093284,
+      "loss": 1.2249,
+      "step": 485
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.0001999967871034016,
+      "loss": 1.0606,
+      "step": 490
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.00019999320250731225,
+      "loss": 1.1662,
+      "step": 495
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.38671875,
+      "learning_rate": 0.00019998829035260677,
+      "loss": 1.0885,
+      "step": 500
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.375,
+      "learning_rate": 0.0001999820507045013,
+      "loss": 1.2287,
+      "step": 505
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.859375,
+      "learning_rate": 0.00019997448364583652,
+      "loss": 1.2669,
+      "step": 510
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.00019996558927707635,
+      "loss": 1.0815,
+      "step": 515
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.0390625,
+      "learning_rate": 0.00019995536771630682,
+      "loss": 1.2549,
+      "step": 520
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.375,
+      "learning_rate": 0.00019994381909923433,
+      "loss": 1.0642,
+      "step": 525
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.43359375,
+      "learning_rate": 0.00019993094357918396,
+      "loss": 1.2543,
+      "step": 530
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.494140625,
+      "learning_rate": 0.00019991674132709743,
+      "loss": 1.1474,
+      "step": 535
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.00019990121253153075,
+      "loss": 1.0659,
+      "step": 540
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.00019988435739865182,
+      "loss": 1.1007,
+      "step": 545
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.00019986617615223767,
+      "loss": 1.1559,
+      "step": 550
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.546875,
+      "learning_rate": 0.0001998466690336713,
+      "loss": 0.9967,
+      "step": 555
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.49609375,
+      "learning_rate": 0.0001998258363019388,
+      "loss": 1.1432,
+      "step": 560
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.38671875,
+      "learning_rate": 0.00019980367823362573,
+      "loss": 1.0002,
+      "step": 565
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 8.625,
+      "learning_rate": 0.00019978019512291337,
+      "loss": 1.0953,
+      "step": 570
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.00019975538728157503,
+      "loss": 1.1463,
+      "step": 575
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.384765625,
+      "learning_rate": 0.0001997292550389717,
+      "loss": 1.0971,
+      "step": 580
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.0001997017987420478,
+      "loss": 1.0907,
+      "step": 585
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.00019967301875532657,
+      "loss": 1.1168,
+      "step": 590
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.0390625,
+      "learning_rate": 0.00019964291546090517,
+      "loss": 0.9757,
+      "step": 595
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.451171875,
+      "learning_rate": 0.0001996114892584496,
+      "loss": 1.067,
+      "step": 600
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.00019957874056518955,
+      "loss": 1.1446,
+      "step": 605
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.00019954466981591263,
+      "loss": 1.1734,
+      "step": 610
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.466796875,
+      "learning_rate": 0.00019950927746295875,
+      "loss": 1.175,
+      "step": 615
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.482421875,
+      "learning_rate": 0.0001994725639762141,
+      "loss": 1.1536,
+      "step": 620
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.6171875,
+      "learning_rate": 0.0001994345298431049,
+      "loss": 1.1167,
+      "step": 625
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.462890625,
+      "learning_rate": 0.00019939517556859083,
+      "loss": 1.1631,
+      "step": 630
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.416015625,
+      "learning_rate": 0.00019935450167515848,
+      "loss": 1.2179,
+      "step": 635
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.458984375,
+      "learning_rate": 0.00019931250870281435,
+      "loss": 1.183,
+      "step": 640
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.5,
+      "learning_rate": 0.00019926919720907762,
+      "loss": 1.1522,
+      "step": 645
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.404296875,
+      "learning_rate": 0.00019922456776897294,
+      "loss": 1.23,
+      "step": 650
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.34765625,
+      "learning_rate": 0.00019917862097502246,
+      "loss": 1.1084,
+      "step": 655
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.0001991313574372383,
+      "loss": 1.0821,
+      "step": 660
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.462890625,
+      "learning_rate": 0.0001990827777831143,
+      "loss": 1.1474,
+      "step": 665
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.453125,
+      "learning_rate": 0.0001990328826576177,
+      "loss": 1.1337,
+      "step": 670
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.48828125,
+      "learning_rate": 0.00019898167272318053,
+      "loss": 1.0593,
+      "step": 675
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.41015625,
+      "learning_rate": 0.00019892914865969093,
+      "loss": 1.2849,
+      "step": 680
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.00019887531116448399,
+      "loss": 1.133,
+      "step": 685
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.00019882016095233262,
+      "loss": 1.1837,
+      "step": 690
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.41015625,
+      "learning_rate": 0.00019876369875543803,
+      "loss": 1.1552,
+      "step": 695
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.00019870592532341989,
+      "loss": 1.1478,
+      "step": 700
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.46484375,
+      "learning_rate": 0.0001986468414233065,
+      "loss": 1.2432,
+      "step": 705
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.3984375,
+      "learning_rate": 0.00019858644783952466,
+      "loss": 1.136,
+      "step": 710
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.15625,
+      "learning_rate": 0.000198524745373889,
+      "loss": 1.2197,
+      "step": 715
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.00019846173484559167,
+      "loss": 1.1506,
+      "step": 720
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.0001983974170911912,
+      "loss": 0.9831,
+      "step": 725
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.37890625,
+      "learning_rate": 0.00019833179296460156,
+      "loss": 1.0984,
+      "step": 730
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.0001982648633370807,
+      "loss": 1.0827,
+      "step": 735
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.2890625,
+      "learning_rate": 0.00019819662909721905,
+      "loss": 1.1374,
+      "step": 740
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.38671875,
+      "learning_rate": 0.00019812709115092774,
+      "loss": 1.0391,
+      "step": 745
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.35546875,
+      "learning_rate": 0.0001980562504214265,
+      "loss": 1.0482,
+      "step": 750
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.435546875,
+      "learning_rate": 0.0001979841078492315,
+      "loss": 1.0314,
+      "step": 755
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.0001979106643921427,
+      "loss": 0.9726,
+      "step": 760
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 28.875,
+      "learning_rate": 0.00019783592102523144,
+      "loss": 1.2156,
+      "step": 765
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.4375,
+      "learning_rate": 0.00019775987874082714,
+      "loss": 1.1393,
+      "step": 770
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.0001976825385485043,
+      "loss": 0.9954,
+      "step": 775
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.35546875,
+      "learning_rate": 0.00019760390147506907,
+      "loss": 1.1198,
+      "step": 780
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.44921875,
+      "learning_rate": 0.00019752396856454575,
+      "loss": 1.0508,
+      "step": 785
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00019744274087816262,
+      "loss": 1.0594,
+      "step": 790
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.00019736021949433825,
+      "loss": 1.0685,
+      "step": 795
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.7734375,
+      "learning_rate": 0.00019727640550866677,
+      "loss": 1.18,
+      "step": 800
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.392578125,
+      "learning_rate": 0.00019719130003390364,
+      "loss": 1.1706,
+      "step": 805
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.0001971049041999507,
+      "loss": 1.2276,
+      "step": 810
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.4140625,
+      "learning_rate": 0.00019701721915384125,
+      "loss": 1.2431,
+      "step": 815
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.345703125,
+      "learning_rate": 0.00019692824605972486,
+      "loss": 1.0623,
+      "step": 820
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.00019683798609885174,
+      "loss": 1.062,
+      "step": 825
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.00019674644046955732,
+      "loss": 1.1808,
+      "step": 830
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.36328125,
+      "learning_rate": 0.00019665361038724607,
+      "loss": 1.1256,
+      "step": 835
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00019655949708437544,
+      "loss": 1.0927,
+      "step": 840
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.40234375,
+      "learning_rate": 0.00019646410181043973,
+      "loss": 1.0819,
+      "step": 845
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.34765625,
+      "learning_rate": 0.00019636742583195317,
+      "loss": 1.1013,
+      "step": 850
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.4765625,
+      "learning_rate": 0.00019626947043243327,
+      "loss": 1.1833,
+      "step": 855
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.6015625,
+      "learning_rate": 0.00019617023691238382,
+      "loss": 1.1131,
+      "step": 860
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.388671875,
+      "learning_rate": 0.0001960697265892775,
+      "loss": 1.1258,
+      "step": 865
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.384765625,
+      "learning_rate": 0.00019596794079753847,
+      "loss": 1.2949,
+      "step": 870
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.384765625,
+      "learning_rate": 0.00019586488088852463,
+      "loss": 1.1795,
+      "step": 875
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.640625,
+      "learning_rate": 0.00019576054823050972,
+      "loss": 1.1488,
+      "step": 880
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.388671875,
+      "learning_rate": 0.00019565494420866502,
+      "loss": 1.1299,
+      "step": 885
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.5,
+      "learning_rate": 0.00019554807022504123,
+      "loss": 1.1733,
+      "step": 890
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.375,
+      "learning_rate": 0.0001954399276985495,
+      "loss": 1.0494,
+      "step": 895
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.38671875,
+      "learning_rate": 0.00019533051806494285,
+      "loss": 1.1806,
+      "step": 900
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.458984375,
+      "learning_rate": 0.00019521984277679708,
+      "loss": 1.2164,
+      "step": 905
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.400390625,
+      "learning_rate": 0.00019510790330349136,
+      "loss": 1.0622,
+      "step": 910
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.00019499470113118887,
+      "loss": 1.112,
+      "step": 915
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00019488023776281695,
+      "loss": 1.1172,
+      "step": 920
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.00019476451471804715,
+      "loss": 1.1562,
+      "step": 925
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.333984375,
+      "learning_rate": 0.00019464753353327525,
+      "loss": 1.1697,
+      "step": 930
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.00019452929576160057,
+      "loss": 1.0866,
+      "step": 935
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.00019440980297280556,
+      "loss": 1.1256,
+      "step": 940
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.00019428905675333486,
+      "loss": 1.0887,
+      "step": 945
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.333984375,
+      "learning_rate": 0.00019416705870627427,
+      "loss": 1.0343,
+      "step": 950
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.33203125,
+      "learning_rate": 0.0001940438104513295,
+      "loss": 1.1126,
+      "step": 955
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00019391931362480456,
+      "loss": 1.114,
+      "step": 960
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.419921875,
+      "learning_rate": 0.0001937935698795802,
+      "loss": 1.0697,
+      "step": 965
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.62109375,
+      "learning_rate": 0.00019366658088509184,
+      "loss": 1.1501,
+      "step": 970
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.427734375,
+      "learning_rate": 0.00019353834832730736,
+      "loss": 1.104,
+      "step": 975
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.40234375,
+      "learning_rate": 0.00019340887390870487,
+      "loss": 1.1771,
+      "step": 980
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.431640625,
+      "learning_rate": 0.00019327815934825005,
+      "loss": 1.0209,
+      "step": 985
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.00019314620638137325,
+      "loss": 1.0394,
+      "step": 990
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.390625,
+      "learning_rate": 0.00019301301675994653,
+      "loss": 1.1564,
+      "step": 995
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.00019287859225226042,
+      "loss": 1.165,
+      "step": 1000
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.00019274293464300032,
+      "loss": 1.0667,
+      "step": 1005
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.357421875,
+      "learning_rate": 0.000192606045733223,
+      "loss": 1.0807,
+      "step": 1010
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.375,
+      "learning_rate": 0.00019246792734033244,
+      "loss": 1.1203,
+      "step": 1015
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.00019232858129805593,
+      "loss": 1.0797,
+      "step": 1020
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.36328125,
+      "learning_rate": 0.00019218800945641963,
+      "loss": 1.1469,
+      "step": 1025
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.00019204621368172403,
+      "loss": 0.9757,
+      "step": 1030
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.00019190319585651907,
+      "loss": 1.056,
+      "step": 1035
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.34375,
+      "learning_rate": 0.00019175895787957933,
+      "loss": 1.1211,
+      "step": 1040
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.349609375,
+      "learning_rate": 0.0001916135016658786,
+      "loss": 1.1362,
+      "step": 1045
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.361328125,
+      "learning_rate": 0.00019146682914656472,
+      "loss": 1.0445,
+      "step": 1050
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.384765625,
+      "learning_rate": 0.0001913189422689337,
+      "loss": 1.1363,
+      "step": 1055
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.384765625,
+      "learning_rate": 0.00019116984299640395,
+      "loss": 1.0689,
+      "step": 1060
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.41015625,
+      "learning_rate": 0.00019101953330849031,
+      "loss": 1.1139,
+      "step": 1065
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.00019086801520077766,
+      "loss": 1.0314,
+      "step": 1070
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.00019071529068489436,
+      "loss": 1.0727,
+      "step": 1075
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.384765625,
+      "learning_rate": 0.0001905613617884858,
+      "loss": 1.1662,
+      "step": 1080
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.408203125,
+      "learning_rate": 0.00019040623055518712,
+      "loss": 1.2001,
+      "step": 1085
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.00019024989904459647,
+      "loss": 1.1899,
+      "step": 1090
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.361328125,
+      "learning_rate": 0.00019009236933224734,
+      "loss": 1.1244,
+      "step": 1095
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.412109375,
+      "learning_rate": 0.00018993364350958114,
+      "loss": 1.0512,
+      "step": 1100
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.34375,
+      "learning_rate": 0.00018977372368391947,
+      "loss": 1.1879,
+      "step": 1105
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.451171875,
+      "learning_rate": 0.0001896126119784361,
+      "loss": 1.1454,
+      "step": 1110
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.69921875,
+      "learning_rate": 0.0001894503105321287,
+      "loss": 1.1091,
+      "step": 1115
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.00018928682149979063,
+      "loss": 1.1875,
+      "step": 1120
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.00018912214705198212,
+      "loss": 1.1465,
+      "step": 1125
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.458984375,
+      "learning_rate": 0.00018895628937500166,
+      "loss": 1.0818,
+      "step": 1130
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.0001887892506708568,
+      "loss": 1.1909,
+      "step": 1135
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.4296875,
+      "learning_rate": 0.00018862103315723494,
+      "loss": 1.08,
+      "step": 1140
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.443359375,
+      "learning_rate": 0.00018845163906747406,
+      "loss": 1.0751,
+      "step": 1145
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.0001882810706505328,
+      "loss": 1.1537,
+      "step": 1150
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.349609375,
+      "learning_rate": 0.00018810933017096088,
+      "loss": 1.0934,
+      "step": 1155
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.00018793641990886876,
+      "loss": 1.029,
+      "step": 1160
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.00018776234215989762,
+      "loss": 1.0162,
+      "step": 1165
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.453125,
+      "learning_rate": 0.00018758709923518863,
+      "loss": 1.0271,
+      "step": 1170
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.0001874106934613526,
+      "loss": 1.1106,
+      "step": 1175
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.00018723312718043873,
+      "loss": 1.1233,
+      "step": 1180
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.00018705440274990371,
+      "loss": 1.0942,
+      "step": 1185
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.38671875,
+      "learning_rate": 0.00018687452254258045,
+      "loss": 1.1519,
+      "step": 1190
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.00018669348894664642,
+      "loss": 1.1205,
+      "step": 1195
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00018651130436559215,
+      "loss": 1.179,
+      "step": 1200
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.00018632797121818914,
+      "loss": 1.0658,
+      "step": 1205
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.455078125,
+      "learning_rate": 0.00018614349193845785,
+      "loss": 1.048,
+      "step": 1210
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.00018595786897563529,
+      "loss": 1.2007,
+      "step": 1215
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.0001857711047941427,
+      "loss": 1.2782,
+      "step": 1220
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.447265625,
+      "learning_rate": 0.00018558320187355253,
+      "loss": 1.0843,
+      "step": 1225
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.0001853941627085558,
+      "loss": 1.1152,
+      "step": 1230
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.00018520398980892888,
+      "loss": 1.1012,
+      "step": 1235
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00018501268569950003,
+      "loss": 1.1181,
+      "step": 1240
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 3.109375,
+      "learning_rate": 0.00018482025292011612,
+      "loss": 1.1313,
+      "step": 1245
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.4140625,
+      "learning_rate": 0.0001846266940256088,
+      "loss": 1.0561,
+      "step": 1250
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.4453125,
+      "learning_rate": 0.00018443201158576048,
+      "loss": 1.1315,
+      "step": 1255
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.9453125,
+      "learning_rate": 0.00018423620818527035,
+      "loss": 1.0607,
+      "step": 1260
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.380859375,
+      "learning_rate": 0.00018403928642372005,
+      "loss": 1.1792,
+      "step": 1265
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.447265625,
+      "learning_rate": 0.0001838412489155391,
+      "loss": 0.975,
+      "step": 1270
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.4140625,
+      "learning_rate": 0.00018364209828997027,
+      "loss": 1.1655,
+      "step": 1275
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.41796875,
+      "learning_rate": 0.00018344183719103452,
+      "loss": 1.1981,
+      "step": 1280
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.609375,
+      "learning_rate": 0.00018324046827749606,
+      "loss": 1.1521,
+      "step": 1285
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.349609375,
+      "learning_rate": 0.000183037994222827,
+      "loss": 1.1028,
+      "step": 1290
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.00018283441771517182,
+      "loss": 1.1165,
+      "step": 1295
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.4765625,
+      "learning_rate": 0.00018262974145731172,
+      "loss": 1.0332,
+      "step": 1300
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.40234375,
+      "learning_rate": 0.0001824239681666287,
+      "loss": 1.1238,
+      "step": 1305
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.392578125,
+      "learning_rate": 0.00018221710057506947,
+      "loss": 1.077,
+      "step": 1310
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.419921875,
+      "learning_rate": 0.00018200914142910935,
+      "loss": 1.0885,
+      "step": 1315
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.00018180009348971548,
+      "loss": 1.1532,
+      "step": 1320
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.45703125,
+      "learning_rate": 0.00018158995953231055,
+      "loss": 1.159,
+      "step": 1325
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.384765625,
+      "learning_rate": 0.00018137874234673557,
+      "loss": 1.1303,
+      "step": 1330
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.36328125,
+      "learning_rate": 0.00018116644473721314,
+      "loss": 0.9563,
+      "step": 1335
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.408203125,
+      "learning_rate": 0.00018095306952231006,
+      "loss": 1.1481,
+      "step": 1340
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.48828125,
+      "learning_rate": 0.00018073861953489992,
+      "loss": 1.1995,
+      "step": 1345
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.40234375,
+      "learning_rate": 0.00018052309762212556,
+      "loss": 1.1704,
+      "step": 1350
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.3515625,
+      "learning_rate": 0.00018030650664536122,
+      "loss": 1.0302,
+      "step": 1355
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.0001800888494801745,
+      "loss": 1.199,
+      "step": 1360
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.384765625,
+      "learning_rate": 0.00017987012901628832,
+      "loss": 1.0257,
+      "step": 1365
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.0001796503481575424,
+      "loss": 1.2543,
+      "step": 1370
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.357421875,
+      "learning_rate": 0.00017942950982185484,
+      "loss": 1.0754,
+      "step": 1375
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.412109375,
+      "learning_rate": 0.00017920761694118327,
+      "loss": 1.1201,
+      "step": 1380
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.404296875,
+      "learning_rate": 0.000178984672461486,
+      "loss": 1.0574,
+      "step": 1385
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.3515625,
+      "learning_rate": 0.0001787606793426829,
+      "loss": 1.1474,
+      "step": 1390
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.96484375,
+      "learning_rate": 0.00017853564055861603,
+      "loss": 1.2119,
+      "step": 1395
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.416015625,
+      "learning_rate": 0.0001783095590970103,
+      "loss": 1.1593,
+      "step": 1400
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.0001780824379594336,
+      "loss": 1.1803,
+      "step": 1405
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.33984375,
+      "learning_rate": 0.00017785428016125717,
+      "loss": 1.0862,
+      "step": 1410
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.3984375,
+      "learning_rate": 0.00017762508873161542,
+      "loss": 1.1009,
+      "step": 1415
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.412109375,
+      "learning_rate": 0.00017739486671336575,
+      "loss": 1.1385,
+      "step": 1420
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.484375,
+      "learning_rate": 0.0001771636171630482,
+      "loss": 1.1652,
+      "step": 1425
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.00017693134315084475,
+      "loss": 1.1657,
+      "step": 1430
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.408203125,
+      "learning_rate": 0.0001766980477605387,
+      "loss": 1.0517,
+      "step": 1435
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.408203125,
+      "learning_rate": 0.00017646373408947364,
+      "loss": 1.0722,
+      "step": 1440
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.400390625,
+      "learning_rate": 0.00017622840524851235,
+      "loss": 1.1638,
+      "step": 1445
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.00017599206436199556,
+      "loss": 1.1001,
+      "step": 1450
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6171875,
+      "learning_rate": 0.0001757547145677003,
+      "loss": 1.0481,
+      "step": 1455
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.9453125,
+      "learning_rate": 0.00017551635901679849,
+      "loss": 1.1408,
+      "step": 1460
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.00017527700087381482,
+      "loss": 1.0477,
+      "step": 1465
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.5625,
+      "learning_rate": 0.00017503664331658498,
+      "loss": 1.0721,
+      "step": 1470
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.322265625,
+      "learning_rate": 0.00017479528953621336,
+      "loss": 1.1062,
+      "step": 1475
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00017455294273703066,
+      "loss": 1.1635,
+      "step": 1480
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.00017430960613655142,
+      "loss": 0.9949,
+      "step": 1485
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.0001740652829654312,
+      "loss": 1.156,
+      "step": 1490
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.00017381997646742387,
+      "loss": 1.1216,
+      "step": 1495
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.00017357368989933834,
+      "loss": 1.1143,
+      "step": 1500
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.00017332642653099543,
+      "loss": 1.1209,
+      "step": 1505
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.00017307818964518444,
+      "loss": 1.0504,
+      "step": 1510
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.392578125,
+      "learning_rate": 0.00017282898253761952,
+      "loss": 1.1089,
+      "step": 1515
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.5546875,
+      "learning_rate": 0.00017257880851689604,
+      "loss": 1.0741,
+      "step": 1520
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.357421875,
+      "learning_rate": 0.00017232767090444656,
+      "loss": 1.0457,
+      "step": 1525
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.45703125,
+      "learning_rate": 0.00017207557303449672,
+      "loss": 1.1921,
+      "step": 1530
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.34765625,
+      "learning_rate": 0.000171822518254021,
+      "loss": 1.1792,
+      "step": 1535
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.00017156850992269837,
+      "loss": 1.081,
+      "step": 1540
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.00017131355141286747,
+      "loss": 1.0944,
+      "step": 1545
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.380859375,
+      "learning_rate": 0.00017105764610948212,
+      "loss": 1.0123,
+      "step": 1550
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.484375,
+      "learning_rate": 0.00017080079741006618,
+      "loss": 0.9975,
+      "step": 1555
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00017054300872466842,
+      "loss": 1.0664,
+      "step": 1560
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.498046875,
+      "learning_rate": 0.00017028428347581746,
+      "loss": 1.108,
+      "step": 1565
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.4140625,
+      "learning_rate": 0.00017002462509847606,
+      "loss": 1.0253,
+      "step": 1570
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.00016976403703999578,
+      "loss": 1.0861,
+      "step": 1575
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.404296875,
+      "learning_rate": 0.00016950252276007102,
+      "loss": 1.065,
+      "step": 1580
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.4375,
+      "learning_rate": 0.00016924008573069305,
+      "loss": 1.0213,
+      "step": 1585
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.00016897672943610417,
+      "loss": 1.1514,
+      "step": 1590
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.322265625,
+      "learning_rate": 0.00016871245737275125,
+      "loss": 1.1526,
+      "step": 1595
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.416015625,
+      "learning_rate": 0.00016844727304923931,
+      "loss": 1.1442,
+      "step": 1600
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.392578125,
+      "learning_rate": 0.00016818117998628499,
+      "loss": 1.0245,
+      "step": 1605
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.37109375,
+      "learning_rate": 0.00016791418171666983,
+      "loss": 1.0738,
+      "step": 1610
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.00016764628178519335,
+      "loss": 1.0517,
+      "step": 1615
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.0001673774837486259,
+      "loss": 1.2543,
+      "step": 1620
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.404296875,
+      "learning_rate": 0.00016710779117566158,
+      "loss": 1.1085,
+      "step": 1625
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.00016683720764687076,
+      "loss": 1.1164,
+      "step": 1630
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.36328125,
+      "learning_rate": 0.00016656573675465257,
+      "loss": 1.1121,
+      "step": 1635
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00016629338210318723,
+      "loss": 1.1551,
+      "step": 1640
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.40234375,
+      "learning_rate": 0.00016602014730838818,
+      "loss": 1.149,
+      "step": 1645
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.390625,
+      "learning_rate": 0.00016574603599785412,
+      "loss": 1.0301,
+      "step": 1650
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.357421875,
+      "learning_rate": 0.00016547105181082062,
+      "loss": 1.0695,
+      "step": 1655
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.5546875,
+      "learning_rate": 0.00016519519839811224,
+      "loss": 1.0497,
+      "step": 1660
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.75,
+      "learning_rate": 0.0001649184794220936,
+      "loss": 1.1484,
+      "step": 1665
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.38671875,
+      "learning_rate": 0.00016464089855662111,
+      "loss": 1.056,
+      "step": 1670
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.390625,
+      "learning_rate": 0.0001643624594869939,
+      "loss": 1.1911,
+      "step": 1675
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.466796875,
+      "learning_rate": 0.00016408316590990513,
+      "loss": 1.0945,
+      "step": 1680
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.40234375,
+      "learning_rate": 0.00016380302153339282,
+      "loss": 1.1104,
+      "step": 1685
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.435546875,
+      "learning_rate": 0.0001635220300767906,
+      "loss": 1.0678,
+      "step": 1690
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.59765625,
+      "learning_rate": 0.0001632401952706783,
+      "loss": 1.0883,
+      "step": 1695
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.392578125,
+      "learning_rate": 0.00016295752085683258,
+      "loss": 1.221,
+      "step": 1700
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.388671875,
+      "learning_rate": 0.00016267401058817709,
+      "loss": 1.1249,
+      "step": 1705
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.0001623896682287326,
+      "loss": 1.0804,
+      "step": 1710
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.375,
+      "learning_rate": 0.0001621044975535673,
+      "loss": 1.0464,
+      "step": 1715
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.41796875,
+      "learning_rate": 0.00016181850234874636,
+      "loss": 1.1579,
+      "step": 1720
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.345703125,
+      "learning_rate": 0.00016153168641128192,
+      "loss": 1.0265,
+      "step": 1725
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.33984375,
+      "learning_rate": 0.00016124405354908246,
+      "loss": 0.9793,
+      "step": 1730
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.41015625,
+      "learning_rate": 0.00016095560758090245,
+      "loss": 1.1421,
+      "step": 1735
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.439453125,
+      "learning_rate": 0.00016066635233629145,
+      "loss": 1.0974,
+      "step": 1740
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.37109375,
+      "learning_rate": 0.0001603762916555435,
+      "loss": 1.0533,
+      "step": 1745
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.380859375,
+      "learning_rate": 0.00016008542938964588,
+      "loss": 1.1346,
+      "step": 1750
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.34375,
+      "learning_rate": 0.00015979376940022813,
+      "loss": 1.1334,
+      "step": 1755
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.3515625,
+      "learning_rate": 0.00015950131555951086,
+      "loss": 1.0466,
+      "step": 1760
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.357421875,
+      "learning_rate": 0.0001592080717502541,
+      "loss": 1.0547,
+      "step": 1765
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.36328125,
+      "learning_rate": 0.000158914041865706,
+      "loss": 1.1026,
+      "step": 1770
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.35546875,
+      "learning_rate": 0.00015861922980955093,
+      "loss": 0.9938,
+      "step": 1775
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.314453125,
+      "learning_rate": 0.0001583236394958578,
+      "loss": 1.0709,
+      "step": 1780
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.38671875,
+      "learning_rate": 0.0001580272748490281,
+      "loss": 1.2601,
+      "step": 1785
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.00015773013980374364,
+      "loss": 1.021,
+      "step": 1790
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.404296875,
+      "learning_rate": 0.0001574322383049145,
+      "loss": 0.9888,
+      "step": 1795
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.0001571335743076265,
+      "loss": 1.0506,
+      "step": 1800
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.388671875,
+      "learning_rate": 0.00015683415177708887,
+      "loss": 1.0916,
+      "step": 1805
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.0001565339746885814,
+      "loss": 1.068,
+      "step": 1810
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.59375,
+      "learning_rate": 0.00015623304702740178,
+      "loss": 1.1657,
+      "step": 1815
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.357421875,
+      "learning_rate": 0.00015593137278881273,
+      "loss": 0.9727,
+      "step": 1820
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.0001556289559779888,
+      "loss": 1.1467,
+      "step": 1825
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.37109375,
+      "learning_rate": 0.0001553258006099633,
+      "loss": 1.1254,
+      "step": 1830
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.435546875,
+      "learning_rate": 0.00015502191070957514,
+      "loss": 1.0301,
+      "step": 1835
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.00015471729031141503,
+      "loss": 1.0504,
+      "step": 1840
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.0001544119434597723,
+      "loss": 1.1046,
+      "step": 1845
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.00015410587420858088,
+      "loss": 1.065,
+      "step": 1850
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.00015379908662136574,
+      "loss": 1.0126,
+      "step": 1855
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.37109375,
+      "learning_rate": 0.00015349158477118884,
+      "loss": 1.0121,
+      "step": 1860
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.419921875,
+      "learning_rate": 0.00015318337274059491,
+      "loss": 1.1957,
+      "step": 1865
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.0001528744546215575,
+      "loss": 1.0178,
+      "step": 1870
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.0001525648345154245,
+      "loss": 1.1288,
+      "step": 1875
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.48046875,
+      "learning_rate": 0.00015225451653286376,
+      "loss": 1.2488,
+      "step": 1880
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.357421875,
+      "learning_rate": 0.00015194350479380843,
+      "loss": 1.0424,
+      "step": 1885
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.00015163180342740232,
+      "loss": 1.1758,
+      "step": 1890
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.36328125,
+      "learning_rate": 0.00015131941657194506,
+      "loss": 1.0522,
+      "step": 1895
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.00015100634837483728,
+      "loss": 1.0631,
+      "step": 1900
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 2.515625,
+      "learning_rate": 0.0001506926029925252,
+      "loss": 1.0577,
+      "step": 1905
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.400390625,
+      "learning_rate": 0.00015037818459044588,
+      "loss": 1.0577,
+      "step": 1910
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.384765625,
+      "learning_rate": 0.0001500630973429717,
+      "loss": 1.185,
+      "step": 1915
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.349609375,
+      "learning_rate": 0.00014974734543335487,
+      "loss": 1.0734,
+      "step": 1920
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.00014943093305367204,
+      "loss": 1.0677,
+      "step": 1925
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.458984375,
+      "learning_rate": 0.00014911386440476858,
+      "loss": 1.1153,
+      "step": 1930
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.0001487961436962028,
+      "loss": 1.115,
+      "step": 1935
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.423828125,
+      "learning_rate": 0.00014847777514619007,
+      "loss": 1.181,
+      "step": 1940
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.00014815876298154683,
+      "loss": 1.0703,
+      "step": 1945
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.380859375,
+      "learning_rate": 0.00014783911143763443,
+      "loss": 1.0116,
+      "step": 1950
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.357421875,
+      "learning_rate": 0.00014751882475830301,
+      "loss": 1.0731,
+      "step": 1955
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.00014719790719583498,
+      "loss": 0.9598,
+      "step": 1960
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.431640625,
+      "learning_rate": 0.00014687636301088873,
+      "loss": 1.0058,
+      "step": 1965
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.390625,
+      "learning_rate": 0.00014655419647244196,
+      "loss": 1.1702,
+      "step": 1970
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.392578125,
+      "learning_rate": 0.00014623141185773507,
+      "loss": 1.1904,
+      "step": 1975
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.0001459080134522143,
+      "loss": 1.0873,
+      "step": 1980
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.00014558400554947495,
+      "loss": 1.1216,
+      "step": 1985
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.349609375,
+      "learning_rate": 0.00014525939245120422,
+      "loss": 0.9945,
+      "step": 1990
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.00014493417846712424,
+      "loss": 1.1011,
+      "step": 1995
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.00014460836791493477,
+      "loss": 1.0888,
+      "step": 2000
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.390625,
+      "learning_rate": 0.00014428196512025586,
+      "loss": 1.0757,
+      "step": 2005
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.412109375,
+      "learning_rate": 0.0001439549744165706,
+      "loss": 1.069,
+      "step": 2010
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.00014362740014516723,
+      "loss": 1.1268,
+      "step": 2015
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.375,
+      "learning_rate": 0.0001432992466550819,
+      "loss": 1.0224,
+      "step": 2020
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.0001429705183030407,
+      "loss": 1.0092,
+      "step": 2025
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.00014264121945340187,
+      "loss": 1.102,
+      "step": 2030
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.00014231135447809778,
+      "loss": 1.0176,
+      "step": 2035
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.380859375,
+      "learning_rate": 0.00014198092775657704,
+      "loss": 1.1124,
+      "step": 2040
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.578125,
+      "learning_rate": 0.0001416499436757463,
+      "loss": 1.2026,
+      "step": 2045
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.00014131840662991196,
+      "loss": 1.1109,
+      "step": 2050
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.421875,
+      "learning_rate": 0.0001409863210207218,
+      "loss": 1.1017,
+      "step": 2055
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.4140625,
+      "learning_rate": 0.0001406536912571066,
+      "loss": 0.9984,
+      "step": 2060
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.4375,
+      "learning_rate": 0.0001403205217552217,
+      "loss": 1.0476,
+      "step": 2065
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.404296875,
+      "learning_rate": 0.00013998681693838818,
+      "loss": 1.17,
+      "step": 2070
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.328125,
+      "learning_rate": 0.0001396525812370342,
+      "loss": 1.0703,
+      "step": 2075
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00013931781908863635,
+      "loss": 1.0546,
+      "step": 2080
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.00013898253493766042,
+      "loss": 1.0501,
+      "step": 2085
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.00013864673323550264,
+      "loss": 0.9337,
+      "step": 2090
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.00013831041844043047,
+      "loss": 0.9993,
+      "step": 2095
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.00013797359501752348,
+      "loss": 1.0086,
+      "step": 2100
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.380859375,
+      "learning_rate": 0.00013763626743861404,
+      "loss": 1.0966,
+      "step": 2105
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.36328125,
+      "learning_rate": 0.00013729844018222788,
+      "loss": 1.1107,
+      "step": 2110
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.00013696011773352474,
+      "loss": 1.0831,
+      "step": 2115
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.00013662130458423876,
+      "loss": 1.0856,
+      "step": 2120
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.3984375,
+      "learning_rate": 0.00013628200523261888,
+      "loss": 1.0221,
+      "step": 2125
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.40234375,
+      "learning_rate": 0.00013594222418336906,
+      "loss": 1.118,
+      "step": 2130
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.81640625,
+      "learning_rate": 0.00013560196594758855,
+      "loss": 1.1019,
+      "step": 2135
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.408203125,
+      "learning_rate": 0.00013526123504271196,
+      "loss": 1.1237,
+      "step": 2140
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.00013492003599244922,
+      "loss": 1.005,
+      "step": 2145
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.4375,
+      "learning_rate": 0.0001345783733267257,
+      "loss": 1.0955,
+      "step": 2150
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.33203125,
+      "learning_rate": 0.0001342362515816219,
+      "loss": 1.1209,
+      "step": 2155
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.00013389367529931324,
+      "loss": 0.9948,
+      "step": 2160
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.361328125,
+      "learning_rate": 0.00013355064902800988,
+      "loss": 1.0733,
+      "step": 2165
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00013320717732189614,
+      "loss": 1.119,
+      "step": 2170
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.35546875,
+      "learning_rate": 0.0001328632647410703,
+      "loss": 0.9886,
+      "step": 2175
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.37890625,
+      "learning_rate": 0.0001325189158514838,
+      "loss": 1.1871,
+      "step": 2180
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.0001321741352248807,
+      "loss": 1.0246,
+      "step": 2185
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.392578125,
+      "learning_rate": 0.00013182892743873715,
+      "loss": 1.1359,
+      "step": 2190
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.361328125,
+      "learning_rate": 0.00013148329707620042,
+      "loss": 1.0841,
+      "step": 2195
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.41015625,
+      "learning_rate": 0.00013113724872602798,
+      "loss": 1.0725,
+      "step": 2200
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.37890625,
+      "learning_rate": 0.00013079078698252687,
+      "loss": 1.0312,
+      "step": 2205
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.00013044391644549252,
+      "loss": 1.0211,
+      "step": 2210
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.60546875,
+      "learning_rate": 0.00013009664172014765,
+      "loss": 1.0574,
+      "step": 2215
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.37890625,
+      "learning_rate": 0.00012974896741708115,
+      "loss": 1.0335,
+      "step": 2220
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.439453125,
+      "learning_rate": 0.00012940089815218704,
+      "loss": 1.1947,
+      "step": 2225
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.640625,
+      "learning_rate": 0.00012905243854660286,
+      "loss": 1.0073,
+      "step": 2230
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.333984375,
+      "learning_rate": 0.00012870359322664872,
+      "loss": 1.0853,
+      "step": 2235
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.34375,
+      "learning_rate": 0.00012835436682376544,
+      "loss": 1.0972,
+      "step": 2240
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.423828125,
+      "learning_rate": 0.0001280047639744534,
+      "loss": 1.0882,
+      "step": 2245
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.40234375,
+      "learning_rate": 0.00012765478932021094,
+      "loss": 1.0716,
+      "step": 2250
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.00012730444750747247,
+      "loss": 1.1108,
+      "step": 2255
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.4453125,
+      "learning_rate": 0.00012695374318754716,
+      "loss": 1.0067,
+      "step": 2260
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.37109375,
+      "learning_rate": 0.00012660268101655686,
+      "loss": 1.0883,
+      "step": 2265
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.00012625126565537456,
+      "loss": 1.1754,
+      "step": 2270
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.3515625,
+      "learning_rate": 0.00012589950176956227,
+      "loss": 1.1861,
+      "step": 2275
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.60546875,
+      "learning_rate": 0.00012554739402930923,
+      "loss": 1.0978,
+      "step": 2280
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.00012519494710936982,
+      "loss": 0.9778,
+      "step": 2285
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.46875,
+      "learning_rate": 0.00012484216568900162,
+      "loss": 1.0441,
+      "step": 2290
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.0001244890544519031,
+      "loss": 1.0973,
+      "step": 2295
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.00012413561808615159,
+      "loss": 1.1415,
+      "step": 2300
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.000123781861284141,
+      "loss": 1.1161,
+      "step": 2305
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.36328125,
+      "learning_rate": 0.0001234277887425195,
+      "loss": 1.0397,
+      "step": 2310
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.00012307340516212712,
+      "loss": 1.1378,
+      "step": 2315
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.00012271871524793348,
+      "loss": 1.0375,
+      "step": 2320
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.0001223637237089752,
+      "loss": 1.1031,
+      "step": 2325
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.0001220084352582933,
+      "loss": 1.0545,
+      "step": 2330
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.578125,
+      "learning_rate": 0.000121652854612871,
+      "loss": 1.0811,
+      "step": 2335
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.482421875,
+      "learning_rate": 0.0001212969864935706,
+      "loss": 0.9833,
+      "step": 2340
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.00012094083562507119,
+      "loss": 1.0778,
+      "step": 2345
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.37890625,
+      "learning_rate": 0.00012058440673580572,
+      "loss": 0.9276,
+      "step": 2350
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.408203125,
+      "learning_rate": 0.00012022770455789824,
+      "loss": 1.0733,
+      "step": 2355
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.00011987073382710123,
+      "loss": 1.0961,
+      "step": 2360
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.388671875,
+      "learning_rate": 0.00011951349928273253,
+      "loss": 1.094,
+      "step": 2365
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.00011915600566761243,
+      "loss": 1.1453,
+      "step": 2370
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.00011879825772800088,
+      "loss": 1.1801,
+      "step": 2375
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.48046875,
+      "learning_rate": 0.0001184402602135344,
+      "loss": 1.024,
+      "step": 2380
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.00011808201787716284,
+      "loss": 1.1035,
+      "step": 2385
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.37890625,
+      "learning_rate": 0.00011772353547508656,
+      "loss": 1.0209,
+      "step": 2390
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.87890625,
+      "learning_rate": 0.00011736481776669306,
+      "loss": 0.9905,
+      "step": 2395
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.390625,
+      "learning_rate": 0.00011700586951449393,
+      "loss": 1.1018,
+      "step": 2400
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.00011664669548406153,
+      "loss": 1.0384,
+      "step": 2405
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.00011628730044396583,
+      "loss": 1.0272,
+      "step": 2410
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.36328125,
+      "learning_rate": 0.00011592768916571095,
+      "loss": 0.9559,
+      "step": 2415
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.00011556786642367201,
+      "loss": 1.0277,
+      "step": 2420
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.00011520783699503148,
+      "loss": 1.0187,
+      "step": 2425
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.37890625,
+      "learning_rate": 0.00011484760565971602,
+      "loss": 0.995,
+      "step": 2430
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.333984375,
+      "learning_rate": 0.00011448717720033283,
+      "loss": 1.0452,
+      "step": 2435
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.0001141265564021063,
+      "loss": 1.0964,
+      "step": 2440
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.380859375,
+      "learning_rate": 0.0001137657480528143,
+      "loss": 1.2053,
+      "step": 2445
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.6875,
+      "learning_rate": 0.00011340475694272478,
+      "loss": 1.0627,
+      "step": 2450
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.455078125,
+      "learning_rate": 0.00011304358786453218,
+      "loss": 1.0991,
+      "step": 2455
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.3515625,
+      "learning_rate": 0.00011268224561329354,
+      "loss": 1.1173,
+      "step": 2460
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.00011232073498636526,
+      "loss": 1.1316,
+      "step": 2465
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.000111959060783339,
+      "loss": 1.0399,
+      "step": 2470
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.00011159722780597826,
+      "loss": 1.1867,
+      "step": 2475
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.0001112352408581544,
+      "loss": 1.039,
+      "step": 2480
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.00011087310474578305,
+      "loss": 1.0116,
+      "step": 2485
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.400390625,
+      "learning_rate": 0.00011051082427676017,
+      "loss": 1.0806,
+      "step": 2490
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.00011014840426089829,
+      "loss": 1.0751,
+      "step": 2495
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.408203125,
+      "learning_rate": 0.00010978584950986264,
+      "loss": 1.1224,
+      "step": 2500
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.00010942316483710722,
+      "loss": 1.07,
+      "step": 2505
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.390625,
+      "learning_rate": 0.00010906035505781101,
+      "loss": 1.059,
+      "step": 2510
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.00010869742498881388,
+      "loss": 1.2009,
+      "step": 2515
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.00010833437944855278,
+      "loss": 1.0943,
+      "step": 2520
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.369140625,
+      "learning_rate": 0.00010797122325699769,
+      "loss": 1.059,
+      "step": 2525
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.0001076079612355877,
+      "loss": 1.1079,
+      "step": 2530
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.00010724459820716683,
+      "loss": 0.9847,
+      "step": 2535
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.3515625,
+      "learning_rate": 0.00010688113899592025,
+      "loss": 1.0511,
+      "step": 2540
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.00010651758842731003,
+      "loss": 1.1623,
+      "step": 2545
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.00010615395132801118,
+      "loss": 1.026,
+      "step": 2550
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.00010579023252584748,
+      "loss": 1.0156,
+      "step": 2555
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.314453125,
+      "learning_rate": 0.00010542643684972748,
+      "loss": 1.112,
+      "step": 2560
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.00010506256912958037,
+      "loss": 1.0403,
+      "step": 2565
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.357421875,
+      "learning_rate": 0.00010469863419629176,
+      "loss": 1.0756,
+      "step": 2570
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.00010433463688163969,
+      "loss": 1.0707,
+      "step": 2575
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.0001039705820182303,
+      "loss": 1.0884,
+      "step": 2580
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.00010360647443943392,
+      "loss": 1.1053,
+      "step": 2585
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.37890625,
+      "learning_rate": 0.00010324231897932068,
+      "loss": 1.034,
+      "step": 2590
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.451171875,
+      "learning_rate": 0.00010287812047259632,
+      "loss": 1.1258,
+      "step": 2595
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.4921875,
+      "learning_rate": 0.00010251388375453827,
+      "loss": 1.0579,
+      "step": 2600
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.00010214961366093114,
+      "loss": 1.1272,
+      "step": 2605
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.00010178531502800266,
+      "loss": 1.0785,
+      "step": 2610
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.00010142099269235952,
+      "loss": 1.0422,
+      "step": 2615
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00010105665149092305,
+      "loss": 1.1162,
+      "step": 2620
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.349609375,
+      "learning_rate": 0.00010069229626086509,
+      "loss": 0.9534,
+      "step": 2625
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.3359375,
+      "learning_rate": 0.0001003279318395437,
+      "loss": 1.0155,
+      "step": 2630
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.56640625,
+      "learning_rate": 9.996356306443895e-05,
+      "loss": 1.1337,
+      "step": 2635
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.34765625,
+      "learning_rate": 9.959919477308878e-05,
+      "loss": 1.0822,
+      "step": 2640
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.5234375,
+      "learning_rate": 9.923483180302463e-05,
+      "loss": 1.1467,
+      "step": 2645
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.392578125,
+      "learning_rate": 9.887047899170735e-05,
+      "loss": 1.0377,
+      "step": 2650
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.349609375,
+      "learning_rate": 9.850614117646289e-05,
+      "loss": 1.0582,
+      "step": 2655
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.37890625,
+      "learning_rate": 9.814182319441814e-05,
+      "loss": 1.1596,
+      "step": 2660
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.3828125,
+      "learning_rate": 9.777752988243659e-05,
+      "loss": 1.1117,
+      "step": 2665
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.490234375,
+      "learning_rate": 9.74132660770543e-05,
+      "loss": 1.0621,
+      "step": 2670
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.39453125,
+      "learning_rate": 9.704903661441553e-05,
+      "loss": 1.0325,
+      "step": 2675
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.37890625,
+      "learning_rate": 9.668484633020859e-05,
+      "loss": 1.1046,
+      "step": 2680
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.484375,
+      "learning_rate": 9.632070005960164e-05,
+      "loss": 1.1527,
+      "step": 2685
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.369140625,
+      "learning_rate": 9.595660263717852e-05,
+      "loss": 1.0448,
+      "step": 2690
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.384765625,
+      "learning_rate": 9.559255889687454e-05,
+      "loss": 1.0799,
+      "step": 2695
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.392578125,
+      "learning_rate": 9.522857367191221e-05,
+      "loss": 1.0383,
+      "step": 2700
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.322265625,
+      "learning_rate": 9.486465179473729e-05,
+      "loss": 0.9467,
+      "step": 2705
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.3671875,
+      "learning_rate": 9.450079809695447e-05,
+      "loss": 1.1495,
+      "step": 2710
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.37890625,
+      "learning_rate": 9.413701740926313e-05,
+      "loss": 1.0475,
+      "step": 2715
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.58984375,
+      "learning_rate": 9.377331456139352e-05,
+      "loss": 0.9938,
+      "step": 2720
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.34765625,
+      "learning_rate": 9.340969438204232e-05,
+      "loss": 1.0686,
+      "step": 2725
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.36328125,
+      "learning_rate": 9.304616169880871e-05,
+      "loss": 1.0011,
+      "step": 2730
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.268272133813025e-05,
+      "loss": 1.1045,
+      "step": 2735
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.3671875,
+      "learning_rate": 9.231937812521875e-05,
+      "loss": 1.0832,
+      "step": 2740
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.37109375,
+      "learning_rate": 9.195613688399623e-05,
+      "loss": 1.0394,
+      "step": 2745
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.392578125,
+      "learning_rate": 9.159300243703091e-05,
+      "loss": 1.0736,
+      "step": 2750
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.4140625,
+      "learning_rate": 9.122997960547316e-05,
+      "loss": 1.1489,
+      "step": 2755
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.38671875,
+      "learning_rate": 9.086707320899144e-05,
+      "loss": 1.1488,
+      "step": 2760
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.373046875,
+      "learning_rate": 9.050428806570841e-05,
+      "loss": 0.9796,
+      "step": 2765
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.349609375,
+      "learning_rate": 9.014162899213689e-05,
+      "loss": 1.0584,
+      "step": 2770
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.333984375,
+      "learning_rate": 8.977910080311601e-05,
+      "loss": 1.0405,
+      "step": 2775
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.431640625,
+      "learning_rate": 8.941670831174706e-05,
+      "loss": 1.1109,
+      "step": 2780
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.40234375,
+      "learning_rate": 8.905445632932989e-05,
+      "loss": 1.1511,
+      "step": 2785
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.3359375,
+      "learning_rate": 8.869234966529883e-05,
+      "loss": 0.9418,
+      "step": 2790
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.365234375,
+      "learning_rate": 8.833039312715884e-05,
+      "loss": 1.0695,
+      "step": 2795
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.357421875,
+      "learning_rate": 8.79685915204218e-05,
+      "loss": 1.0157,
+      "step": 2800
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.38671875,
+      "learning_rate": 8.760694964854265e-05,
+      "loss": 1.1373,
+      "step": 2805
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.3359375,
+      "learning_rate": 8.724547231285558e-05,
+      "loss": 0.9823,
+      "step": 2810
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.349609375,
+      "learning_rate": 8.688416431251032e-05,
+      "loss": 1.0418,
+      "step": 2815
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.37890625,
+      "learning_rate": 8.652303044440841e-05,
+      "loss": 0.9956,
+      "step": 2820
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.365234375,
+      "learning_rate": 8.61620755031396e-05,
+      "loss": 1.1068,
+      "step": 2825
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.361328125,
+      "learning_rate": 8.580130428091802e-05,
+      "loss": 1.0246,
+      "step": 2830
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.310546875,
+      "learning_rate": 8.54407215675187e-05,
+      "loss": 1.0299,
+      "step": 2835
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.4453125,
+      "learning_rate": 8.508033215021395e-05,
+      "loss": 1.1331,
+      "step": 2840
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.375,
+      "learning_rate": 8.472014081370977e-05,
+      "loss": 0.9622,
+      "step": 2845
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.353515625,
+      "learning_rate": 8.436015234008233e-05,
+      "loss": 1.0842,
+      "step": 2850
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.388671875,
+      "learning_rate": 8.400037150871452e-05,
+      "loss": 1.0176,
+      "step": 2855
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.384765625,
+      "learning_rate": 8.364080309623244e-05,
+      "loss": 1.04,
+      "step": 2860
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.412109375,
+      "learning_rate": 8.328145187644206e-05,
+      "loss": 1.052,
+      "step": 2865
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.365234375,
+      "learning_rate": 8.292232262026574e-05,
+      "loss": 1.1369,
+      "step": 2870
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.54296875,
+      "learning_rate": 8.256342009567896e-05,
+      "loss": 1.0515,
+      "step": 2875
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.359375,
+      "learning_rate": 8.220474906764704e-05,
+      "loss": 1.0096,
+      "step": 2880
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 2.859375,
+      "learning_rate": 8.184631429806175e-05,
+      "loss": 1.0463,
+      "step": 2885
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.314453125,
+      "learning_rate": 8.148812054567834e-05,
+      "loss": 1.0897,
+      "step": 2890
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.37890625,
+      "learning_rate": 8.113017256605196e-05,
+      "loss": 1.1268,
+      "step": 2895
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.7578125,
+      "learning_rate": 8.077247511147495e-05,
+      "loss": 1.1213,
+      "step": 2900
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.375,
+      "learning_rate": 8.04150329309135e-05,
+      "loss": 1.006,
+      "step": 2905
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.353515625,
+      "learning_rate": 8.005785076994466e-05,
+      "loss": 1.0393,
+      "step": 2910
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.375,
+      "learning_rate": 7.97009333706933e-05,
+      "loss": 1.0877,
+      "step": 2915
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.369140625,
+      "learning_rate": 7.934428547176925e-05,
+      "loss": 0.9819,
+      "step": 2920
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.375,
+      "learning_rate": 7.898791180820428e-05,
+      "loss": 1.149,
+      "step": 2925
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.34765625,
+      "learning_rate": 7.863181711138925e-05,
+      "loss": 1.2709,
+      "step": 2930
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.345703125,
+      "learning_rate": 7.827600610901138e-05,
+      "loss": 1.0535,
+      "step": 2935
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.396484375,
+      "learning_rate": 7.792048352499142e-05,
+      "loss": 1.0346,
+      "step": 2940
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.421875,
+      "learning_rate": 7.756525407942085e-05,
+      "loss": 1.0115,
+      "step": 2945
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.390625,
+      "learning_rate": 7.72103224884994e-05,
+      "loss": 1.0973,
+      "step": 2950
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.357421875,
+      "learning_rate": 7.685569346447232e-05,
+      "loss": 1.0076,
+      "step": 2955
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.423828125,
+      "learning_rate": 7.650137171556781e-05,
+      "loss": 1.084,
+      "step": 2960
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.359375,
+      "learning_rate": 7.614736194593453e-05,
+      "loss": 1.0667,
+      "step": 2965
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.41015625,
+      "learning_rate": 7.579366885557914e-05,
+      "loss": 1.1245,
+      "step": 2970
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.369140625,
+      "learning_rate": 7.544029714030394e-05,
+      "loss": 0.9752,
+      "step": 2975
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.455078125,
+      "learning_rate": 7.508725149164455e-05,
+      "loss": 1.0817,
+      "step": 2980
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.359375,
+      "learning_rate": 7.473453659680743e-05,
+      "loss": 0.9972,
+      "step": 2985
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.361328125,
+      "learning_rate": 7.43821571386079e-05,
+      "loss": 0.9939,
+      "step": 2990
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.36328125,
+      "learning_rate": 7.403011779540785e-05,
+      "loss": 1.1253,
+      "step": 2995
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.7109375,
+      "learning_rate": 7.367842324105358e-05,
+      "loss": 1.1767,
+      "step": 3000
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.330078125,
+      "learning_rate": 7.332707814481387e-05,
+      "loss": 0.96,
+      "step": 3005
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.39453125,
+      "learning_rate": 7.297608717131788e-05,
+      "loss": 0.9991,
+      "step": 3010
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.341796875,
+      "learning_rate": 7.262545498049332e-05,
+      "loss": 1.1189,
+      "step": 3015
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.3359375,
+      "learning_rate": 7.227518622750443e-05,
+      "loss": 1.081,
+      "step": 3020
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.333984375,
+      "learning_rate": 7.192528556269034e-05,
+      "loss": 1.1145,
+      "step": 3025
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.4140625,
+      "learning_rate": 7.157575763150327e-05,
+      "loss": 1.1179,
+      "step": 3030
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.390625,
+      "learning_rate": 7.12266070744468e-05,
+      "loss": 1.0421,
+      "step": 3035
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.3828125,
+      "learning_rate": 7.087783852701432e-05,
+      "loss": 1.2192,
+      "step": 3040
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.35546875,
+      "learning_rate": 7.052945661962753e-05,
+      "loss": 1.0846,
+      "step": 3045
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.34375,
+      "learning_rate": 7.018146597757484e-05,
+      "loss": 1.0074,
+      "step": 3050
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.373046875,
+      "learning_rate": 6.983387122095004e-05,
+      "loss": 1.1082,
+      "step": 3055
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.37109375,
+      "learning_rate": 6.948667696459097e-05,
+      "loss": 1.1076,
+      "step": 3060
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.380859375,
+      "learning_rate": 6.913988781801826e-05,
+      "loss": 1.0275,
+      "step": 3065
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.55859375,
+      "learning_rate": 6.879350838537403e-05,
+      "loss": 1.204,
+      "step": 3070
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.36328125,
+      "learning_rate": 6.84475432653609e-05,
+      "loss": 1.1089,
+      "step": 3075
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.373046875,
+      "learning_rate": 6.810199705118081e-05,
+      "loss": 1.0234,
+      "step": 3080
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.373046875,
+      "learning_rate": 6.775687433047417e-05,
+      "loss": 1.17,
+      "step": 3085
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.353515625,
+      "learning_rate": 6.74121796852589e-05,
+      "loss": 1.1946,
+      "step": 3090
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.400390625,
+      "learning_rate": 6.706791769186946e-05,
+      "loss": 0.9443,
+      "step": 3095
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.400390625,
+      "learning_rate": 6.672409292089635e-05,
+      "loss": 1.0418,
+      "step": 3100
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.419921875,
+      "learning_rate": 6.638070993712521e-05,
+      "loss": 1.1374,
+      "step": 3105
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.359375,
+      "learning_rate": 6.603777329947635e-05,
+      "loss": 1.1197,
+      "step": 3110
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.365234375,
+      "learning_rate": 6.569528756094409e-05,
+      "loss": 1.01,
+      "step": 3115
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.35546875,
+      "learning_rate": 6.535325726853647e-05,
+      "loss": 1.0671,
+      "step": 3120
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.35546875,
+      "learning_rate": 6.501168696321475e-05,
+      "loss": 0.9577,
+      "step": 3125
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.3671875,
+      "learning_rate": 6.467058117983329e-05,
+      "loss": 1.1782,
+      "step": 3130
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.34765625,
+      "learning_rate": 6.432994444707906e-05,
+      "loss": 1.0968,
+      "step": 3135
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.341796875,
+      "learning_rate": 6.398978128741176e-05,
+      "loss": 0.987,
+      "step": 3140
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.359375,
+      "learning_rate": 6.36500962170038e-05,
+      "loss": 1.0588,
+      "step": 3145
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.365234375,
+      "learning_rate": 6.331089374568007e-05,
+      "loss": 1.0049,
+      "step": 3150
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.353515625,
+      "learning_rate": 6.297217837685833e-05,
+      "loss": 1.0989,
+      "step": 3155
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.353515625,
+      "learning_rate": 6.263395460748938e-05,
+      "loss": 1.1308,
+      "step": 3160
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.361328125,
+      "learning_rate": 6.229622692799724e-05,
+      "loss": 0.9604,
+      "step": 3165
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.3515625,
+      "learning_rate": 6.195899982221953e-05,
+      "loss": 1.0006,
+      "step": 3170
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.3359375,
+      "learning_rate": 6.162227776734817e-05,
+      "loss": 0.9775,
+      "step": 3175
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.357421875,
+      "learning_rate": 6.128606523386967e-05,
+      "loss": 0.978,
+      "step": 3180
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.3828125,
+      "learning_rate": 6.0950366685505864e-05,
+      "loss": 0.992,
+      "step": 3185
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.349609375,
+      "learning_rate": 6.0615186579154756e-05,
+      "loss": 1.103,
+      "step": 3190
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.384765625,
+      "learning_rate": 6.028052936483122e-05,
+      "loss": 1.0557,
+      "step": 3195
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.396484375,
+      "learning_rate": 5.994639948560794e-05,
+      "loss": 1.0671,
+      "step": 3200
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.357421875,
+      "learning_rate": 5.961280137755646e-05,
+      "loss": 0.9779,
+      "step": 3205
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.36328125,
+      "learning_rate": 5.927973946968827e-05,
+      "loss": 1.1016,
+      "step": 3210
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.453125,
+      "learning_rate": 5.8947218183896016e-05,
+      "loss": 1.0919,
+      "step": 3215
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.345703125,
+      "learning_rate": 5.861524193489474e-05,
+      "loss": 1.0643,
+      "step": 3220
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.353515625,
+      "learning_rate": 5.828381513016336e-05,
+      "loss": 1.1629,
+      "step": 3225
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.37890625,
+      "learning_rate": 5.795294216988606e-05,
+      "loss": 1.0624,
+      "step": 3230
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.32421875,
+      "learning_rate": 5.762262744689395e-05,
+      "loss": 0.9258,
+      "step": 3235
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.3203125,
+      "learning_rate": 5.72928753466067e-05,
+      "loss": 1.0168,
+      "step": 3240
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.357421875,
+      "learning_rate": 5.6963690246974354e-05,
+      "loss": 1.1368,
+      "step": 3245
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.357421875,
+      "learning_rate": 5.663507651841906e-05,
+      "loss": 1.0114,
+      "step": 3250
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.37109375,
+      "learning_rate": 5.630703852377732e-05,
+      "loss": 1.117,
+      "step": 3255
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.3359375,
+      "learning_rate": 5.5979580618241786e-05,
+      "loss": 1.0523,
+      "step": 3260
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.376953125,
+      "learning_rate": 5.565270714930364e-05,
+      "loss": 1.1847,
+      "step": 3265
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.67578125,
+      "learning_rate": 5.5326422456694774e-05,
+      "loss": 1.0423,
+      "step": 3270
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.33984375,
+      "learning_rate": 5.500073087233021e-05,
+      "loss": 0.9565,
+      "step": 3275
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.353515625,
+      "learning_rate": 5.4675636720250554e-05,
+      "loss": 0.9838,
+      "step": 3280
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.34375,
+      "learning_rate": 5.435114431656466e-05,
+      "loss": 0.9491,
+      "step": 3285
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.396484375,
+      "learning_rate": 5.402725796939216e-05,
+      "loss": 1.1402,
+      "step": 3290
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.375,
+      "learning_rate": 5.370398197880651e-05,
+      "loss": 0.9926,
+      "step": 3295
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.60546875,
+      "learning_rate": 5.33813206367777e-05,
+      "loss": 1.0699,
+      "step": 3300
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.390625,
+      "learning_rate": 5.3059278227115386e-05,
+      "loss": 1.0185,
+      "step": 3305
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.388671875,
+      "learning_rate": 5.2737859025411954e-05,
+      "loss": 1.1353,
+      "step": 3310
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.365234375,
+      "learning_rate": 5.241706729898579e-05,
+      "loss": 1.0359,
+      "step": 3315
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.34765625,
+      "learning_rate": 5.209690730682457e-05,
+      "loss": 0.9474,
+      "step": 3320
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.37109375,
+      "learning_rate": 5.177738329952888e-05,
+      "loss": 1.0324,
+      "step": 3325
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.369140625,
+      "learning_rate": 5.145849951925544e-05,
+      "loss": 1.1495,
+      "step": 3330
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.443359375,
+      "learning_rate": 5.11402601996612e-05,
+      "loss": 1.0776,
+      "step": 3335
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.38671875,
+      "learning_rate": 5.082266956584687e-05,
+      "loss": 1.0373,
+      "step": 3340
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.3828125,
+      "learning_rate": 5.0505731834300875e-05,
+      "loss": 1.1072,
+      "step": 3345
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.369140625,
+      "learning_rate": 5.018945121284342e-05,
+      "loss": 1.0765,
+      "step": 3350
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.43359375,
+      "learning_rate": 4.9873831900570554e-05,
+      "loss": 1.1338,
+      "step": 3355
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.54296875,
+      "learning_rate": 4.9558878087798554e-05,
+      "loss": 1.2131,
+      "step": 3360
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.412109375,
+      "learning_rate": 4.9244593956008046e-05,
+      "loss": 1.0869,
+      "step": 3365
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.35546875,
+      "learning_rate": 4.893098367778877e-05,
+      "loss": 1.072,
+      "step": 3370
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.353515625,
+      "learning_rate": 4.8618051416784006e-05,
+      "loss": 1.1649,
+      "step": 3375
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.341796875,
+      "learning_rate": 4.8305801327635383e-05,
+      "loss": 0.9534,
+      "step": 3380
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.439453125,
+      "learning_rate": 4.799423755592765e-05,
+      "loss": 1.0432,
+      "step": 3385
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.3984375,
+      "learning_rate": 4.768336423813368e-05,
+      "loss": 0.9562,
+      "step": 3390
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.32421875,
+      "learning_rate": 4.737318550155957e-05,
+      "loss": 0.8911,
+      "step": 3395
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.357421875,
+      "learning_rate": 4.706370546428972e-05,
+      "loss": 0.9981,
+      "step": 3400
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.349609375,
+      "learning_rate": 4.675492823513237e-05,
+      "loss": 0.9333,
+      "step": 3405
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.3515625,
+      "learning_rate": 4.644685791356489e-05,
+      "loss": 1.065,
+      "step": 3410
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.40234375,
+      "learning_rate": 4.613949858967938e-05,
+      "loss": 0.9751,
+      "step": 3415
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.369140625,
+      "learning_rate": 4.583285434412845e-05,
+      "loss": 1.2406,
+      "step": 3420
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.455078125,
+      "learning_rate": 4.5526929248070904e-05,
+      "loss": 1.0822,
+      "step": 3425
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.3984375,
+      "learning_rate": 4.522172736311786e-05,
+      "loss": 0.9694,
+      "step": 3430
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.349609375,
+      "learning_rate": 4.491725274127858e-05,
+      "loss": 1.0378,
+      "step": 3435
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.369140625,
+      "learning_rate": 4.461350942490701e-05,
+      "loss": 1.0835,
+      "step": 3440
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.353515625,
+      "learning_rate": 4.431050144664774e-05,
+      "loss": 1.0179,
+      "step": 3445
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.373046875,
+      "learning_rate": 4.4008232829382824e-05,
+      "loss": 1.1077,
+      "step": 3450
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.369140625,
+      "learning_rate": 4.370670758617808e-05,
+      "loss": 1.0818,
+      "step": 3455
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.349609375,
+      "learning_rate": 4.340592972023e-05,
+      "loss": 1.0678,
+      "step": 3460
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.36328125,
+      "learning_rate": 4.310590322481248e-05,
+      "loss": 1.0692,
+      "step": 3465
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.349609375,
+      "learning_rate": 4.2806632083223886e-05,
+      "loss": 0.9918,
+      "step": 3470
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.353515625,
+      "learning_rate": 4.2508120268734174e-05,
+      "loss": 1.1388,
+      "step": 3475
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.388671875,
+      "learning_rate": 4.2210371744531965e-05,
+      "loss": 1.0109,
+      "step": 3480
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.390625,
+      "learning_rate": 4.191339046367219e-05,
+      "loss": 1.0696,
+      "step": 3485
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.34765625,
+      "learning_rate": 4.161718036902345e-05,
+      "loss": 1.0385,
+      "step": 3490
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.34375,
+      "learning_rate": 4.132174539321572e-05,
+      "loss": 0.9972,
+      "step": 3495
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.40234375,
+      "learning_rate": 4.1027089458588065e-05,
+      "loss": 1.1009,
+      "step": 3500
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.41015625,
+      "learning_rate": 4.073321647713667e-05,
+      "loss": 0.9775,
+      "step": 3505
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.55078125,
+      "learning_rate": 4.0440130350462865e-05,
+      "loss": 1.0561,
+      "step": 3510
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.3515625,
+      "learning_rate": 4.014783496972121e-05,
+      "loss": 1.0012,
+      "step": 3515
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.380859375,
+      "learning_rate": 3.9856334215568035e-05,
+      "loss": 1.0508,
+      "step": 3520
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.345703125,
+      "learning_rate": 3.956563195810978e-05,
+      "loss": 1.0059,
+      "step": 3525
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.48828125,
+      "learning_rate": 3.927573205685167e-05,
+      "loss": 1.1234,
+      "step": 3530
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.353515625,
+      "learning_rate": 3.8986638360646443e-05,
+      "loss": 1.0895,
+      "step": 3535
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.353515625,
+      "learning_rate": 3.8698354707643284e-05,
+      "loss": 1.0191,
+      "step": 3540
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.34765625,
+      "learning_rate": 3.841088492523685e-05,
+      "loss": 1.0382,
+      "step": 3545
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.48828125,
+      "learning_rate": 3.812423283001637e-05,
+      "loss": 1.1604,
+      "step": 3550
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.40625,
+      "learning_rate": 3.783840222771518e-05,
+      "loss": 1.0974,
+      "step": 3555
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.35546875,
+      "learning_rate": 3.755339691316e-05,
+      "loss": 1.0917,
+      "step": 3560
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.39453125,
+      "learning_rate": 3.726922067022066e-05,
+      "loss": 0.8749,
+      "step": 3565
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.3515625,
+      "learning_rate": 3.698587727175982e-05,
+      "loss": 1.0033,
+      "step": 3570
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.345703125,
+      "learning_rate": 3.6703370479582875e-05,
+      "loss": 1.0835,
+      "step": 3575
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.357421875,
+      "learning_rate": 3.642170404438809e-05,
+      "loss": 1.156,
+      "step": 3580
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.94921875,
+      "learning_rate": 3.614088170571661e-05,
+      "loss": 1.0586,
+      "step": 3585
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.380859375,
+      "learning_rate": 3.586090719190306e-05,
+      "loss": 1.0405,
+      "step": 3590
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.357421875,
+      "learning_rate": 3.558178422002589e-05,
+      "loss": 1.027,
+      "step": 3595
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.357421875,
+      "learning_rate": 3.5303516495858057e-05,
+      "loss": 1.1363,
+      "step": 3600
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.314453125,
+      "learning_rate": 3.502610771381783e-05,
+      "loss": 1.106,
+      "step": 3605
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.380859375,
+      "learning_rate": 3.4749561556919775e-05,
+      "loss": 0.9968,
+      "step": 3610
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.337890625,
+      "learning_rate": 3.4473881696725816e-05,
+      "loss": 1.0708,
+      "step": 3615
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.3515625,
+      "learning_rate": 3.419907179329641e-05,
+      "loss": 1.1052,
+      "step": 3620
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.3984375,
+      "learning_rate": 3.392513549514219e-05,
+      "loss": 1.1304,
+      "step": 3625
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.37890625,
+      "learning_rate": 3.3652076439175237e-05,
+      "loss": 1.1509,
+      "step": 3630
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.412109375,
+      "learning_rate": 3.3379898250661034e-05,
+      "loss": 0.9853,
+      "step": 3635
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.384765625,
+      "learning_rate": 3.3108604543170206e-05,
+      "loss": 1.0669,
+      "step": 3640
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.38671875,
+      "learning_rate": 3.283819891853057e-05,
+      "loss": 1.1067,
+      "step": 3645
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.33984375,
+      "learning_rate": 3.256868496677935e-05,
+      "loss": 0.9935,
+      "step": 3650
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.361328125,
+      "learning_rate": 3.230006626611544e-05,
+      "loss": 0.9701,
+      "step": 3655
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.33984375,
+      "learning_rate": 3.2032346382851995e-05,
+      "loss": 0.9201,
+      "step": 3660
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.369140625,
+      "learning_rate": 3.1765528871368945e-05,
+      "loss": 1.1023,
+      "step": 3665
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.44921875,
+      "learning_rate": 3.1499617274065986e-05,
+      "loss": 1.0817,
+      "step": 3670
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.37890625,
+      "learning_rate": 3.12346151213154e-05,
+      "loss": 1.1742,
+      "step": 3675
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.388671875,
+      "learning_rate": 3.097052593141527e-05,
+      "loss": 1.0696,
+      "step": 3680
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.369140625,
+      "learning_rate": 3.070735321054271e-05,
+      "loss": 1.2406,
+      "step": 3685
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.57421875,
+      "learning_rate": 3.0445100452707375e-05,
+      "loss": 1.0931,
+      "step": 3690
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.40625,
+      "learning_rate": 3.018377113970503e-05,
+      "loss": 1.1022,
+      "step": 3695
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.33984375,
+      "learning_rate": 2.992336874107129e-05,
+      "loss": 0.9678,
+      "step": 3700
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.357421875,
+      "learning_rate": 2.9663896714035644e-05,
+      "loss": 1.009,
+      "step": 3705
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.349609375,
+      "learning_rate": 2.9405358503475533e-05,
+      "loss": 0.9928,
+      "step": 3710
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.3828125,
+      "learning_rate": 2.9147757541870546e-05,
+      "loss": 1.1043,
+      "step": 3715
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.9296875,
+      "learning_rate": 2.8891097249256948e-05,
+      "loss": 1.0473,
+      "step": 3720
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.357421875,
+      "learning_rate": 2.8635381033182175e-05,
+      "loss": 1.0229,
+      "step": 3725
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.369140625,
+      "learning_rate": 2.83806122886597e-05,
+      "loss": 1.014,
+      "step": 3730
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.35546875,
+      "learning_rate": 2.81267943981238e-05,
+      "loss": 1.0675,
+      "step": 3735
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.353515625,
+      "learning_rate": 2.7873930731384845e-05,
+      "loss": 1.1727,
+      "step": 3740
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.369140625,
+      "learning_rate": 2.7622024645584423e-05,
+      "loss": 1.0164,
+      "step": 3745
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.361328125,
+      "learning_rate": 2.737107948515083e-05,
+      "loss": 1.183,
+      "step": 3750
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.380859375,
+      "learning_rate": 2.7121098581754634e-05,
+      "loss": 1.0301,
+      "step": 3755
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.35546875,
+      "learning_rate": 2.6872085254264446e-05,
+      "loss": 1.0661,
+      "step": 3760
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.3671875,
+      "learning_rate": 2.662404280870292e-05,
+      "loss": 1.0385,
+      "step": 3765
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.35546875,
+      "learning_rate": 2.6376974538202692e-05,
+      "loss": 1.0785,
+      "step": 3770
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.392578125,
+      "learning_rate": 2.6130883722962886e-05,
+      "loss": 1.0484,
+      "step": 3775
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.396484375,
+      "learning_rate": 2.5885773630205413e-05,
+      "loss": 1.1069,
+      "step": 3780
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.3515625,
+      "learning_rate": 2.564164751413162e-05,
+      "loss": 0.9637,
+      "step": 3785
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.349609375,
+      "learning_rate": 2.5398508615879126e-05,
+      "loss": 1.1005,
+      "step": 3790
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2.5156360163478764e-05,
+      "loss": 1.0053,
+      "step": 3795
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.359375,
+      "learning_rate": 2.4915205371811667e-05,
+      "loss": 1.0631,
+      "step": 3800
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.388671875,
+      "learning_rate": 2.467504744256669e-05,
+      "loss": 1.106,
+      "step": 3805
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.37109375,
+      "learning_rate": 2.443588956419791e-05,
+      "loss": 1.0389,
+      "step": 3810
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.37109375,
+      "learning_rate": 2.4197734911882098e-05,
+      "loss": 1.133,
+      "step": 3815
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.3515625,
+      "learning_rate": 2.3960586647476824e-05,
+      "loss": 1.1337,
+      "step": 3820
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.3515625,
+      "learning_rate": 2.372444791947832e-05,
+      "loss": 0.9681,
+      "step": 3825
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.515625,
+      "learning_rate": 2.3489321862979728e-05,
+      "loss": 1.062,
+      "step": 3830
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.37109375,
+      "learning_rate": 2.3255211599629467e-05,
+      "loss": 1.0479,
+      "step": 3835
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.396484375,
+      "learning_rate": 2.3022120237589783e-05,
+      "loss": 1.0702,
+      "step": 3840
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.384765625,
+      "learning_rate": 2.2790050871495517e-05,
+      "loss": 1.1138,
+      "step": 3845
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.376953125,
+      "learning_rate": 2.255900658241291e-05,
+      "loss": 0.992,
+      "step": 3850
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.31640625,
+      "learning_rate": 2.232899043779889e-05,
+      "loss": 0.9199,
+      "step": 3855
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.3984375,
+      "learning_rate": 2.2100005491460163e-05,
+      "loss": 1.0894,
+      "step": 3860
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.5234375,
+      "learning_rate": 2.1872054783512775e-05,
+      "loss": 1.1446,
+      "step": 3865
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.34765625,
+      "learning_rate": 2.16451413403417e-05,
+      "loss": 1.0609,
+      "step": 3870
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.3515625,
+      "learning_rate": 2.1419268174560693e-05,
+      "loss": 0.9734,
+      "step": 3875
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.349609375,
+      "learning_rate": 2.1194438284972294e-05,
+      "loss": 0.9959,
+      "step": 3880
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.345703125,
+      "learning_rate": 2.0970654656527922e-05,
+      "loss": 0.941,
+      "step": 3885
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.33984375,
+      "learning_rate": 2.074792026028841e-05,
+      "loss": 0.9481,
+      "step": 3890
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.357421875,
+      "learning_rate": 2.052623805338444e-05,
+      "loss": 1.0345,
+      "step": 3895
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.341796875,
+      "learning_rate": 2.0305610978977284e-05,
+      "loss": 1.0368,
+      "step": 3900
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.353515625,
+      "learning_rate": 2.0086041966219827e-05,
+      "loss": 1.0414,
+      "step": 3905
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.609375,
+      "learning_rate": 1.986753393021753e-05,
+      "loss": 1.0255,
+      "step": 3910
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.376953125,
+      "learning_rate": 1.965008977198991e-05,
+      "loss": 1.0517,
+      "step": 3915
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.376953125,
+      "learning_rate": 1.943371237843179e-05,
+      "loss": 1.202,
+      "step": 3920
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.36328125,
+      "learning_rate": 1.9218404622275197e-05,
+      "loss": 1.0329,
+      "step": 3925
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.349609375,
+      "learning_rate": 1.9004169362051116e-05,
+      "loss": 0.8898,
+      "step": 3930
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.357421875,
+      "learning_rate": 1.8791009442051545e-05,
+      "loss": 1.0638,
+      "step": 3935
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.40234375,
+      "learning_rate": 1.8578927692291737e-05,
+      "loss": 1.0068,
+      "step": 3940
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.42578125,
+      "learning_rate": 1.8367926928472623e-05,
+      "loss": 1.0017,
+      "step": 3945
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.333984375,
+      "learning_rate": 1.815800995194348e-05,
+      "loss": 1.0378,
+      "step": 3950
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.33984375,
+      "learning_rate": 1.7949179549664606e-05,
+      "loss": 1.1059,
+      "step": 3955
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.349609375,
+      "learning_rate": 1.7741438494170494e-05,
+      "loss": 0.9109,
+      "step": 3960
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.330078125,
+      "learning_rate": 1.7534789543532894e-05,
+      "loss": 0.976,
+      "step": 3965
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.353515625,
+      "learning_rate": 1.7329235441324253e-05,
+      "loss": 0.922,
+      "step": 3970
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.33203125,
+      "learning_rate": 1.7124778916581297e-05,
+      "loss": 0.9421,
+      "step": 3975
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.34375,
+      "learning_rate": 1.6921422683768706e-05,
+      "loss": 1.0366,
+      "step": 3980
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.369140625,
+      "learning_rate": 1.671916944274321e-05,
+      "loss": 1.1087,
+      "step": 3985
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.3671875,
+      "learning_rate": 1.6518021878717672e-05,
+      "loss": 1.1213,
+      "step": 3990
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.34765625,
+      "learning_rate": 1.6317982662225462e-05,
+      "loss": 1.1077,
+      "step": 3995
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.390625,
+      "learning_rate": 1.6119054449084902e-05,
+      "loss": 1.073,
+      "step": 4000
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.34375,
+      "learning_rate": 1.592123988036419e-05,
+      "loss": 1.0458,
+      "step": 4005
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.333984375,
+      "learning_rate": 1.5724541582346197e-05,
+      "loss": 1.0022,
+      "step": 4010
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.3515625,
+      "learning_rate": 1.5528962166493642e-05,
+      "loss": 1.0358,
+      "step": 4015
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.396484375,
+      "learning_rate": 1.5334504229414403e-05,
+      "loss": 1.0106,
+      "step": 4020
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.4140625,
+      "learning_rate": 1.5141170352827095e-05,
+      "loss": 1.0975,
+      "step": 4025
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.36328125,
+      "learning_rate": 1.4948963103526726e-05,
+      "loss": 1.013,
+      "step": 4030
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.3515625,
+      "learning_rate": 1.4757885033350627e-05,
+      "loss": 1.006,
+      "step": 4035
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.328125,
+      "learning_rate": 1.4567938679144633e-05,
+      "loss": 0.9755,
+      "step": 4040
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.353515625,
+      "learning_rate": 1.4379126562729373e-05,
+      "loss": 1.0473,
+      "step": 4045
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.361328125,
+      "learning_rate": 1.4191451190866756e-05,
+      "loss": 1.0553,
+      "step": 4050
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.400491505522673e-05,
+      "loss": 1.0685,
+      "step": 4055
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.365234375,
+      "learning_rate": 1.3819520632354166e-05,
+      "loss": 1.0261,
+      "step": 4060
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.345703125,
+      "learning_rate": 1.3635270383636046e-05,
+      "loss": 1.0794,
+      "step": 4065
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.34765625,
+      "learning_rate": 1.3452166755268635e-05,
+      "loss": 0.9351,
+      "step": 4070
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.9375,
+      "learning_rate": 1.3270212178225227e-05,
+      "loss": 0.9637,
+      "step": 4075
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.390625,
+      "learning_rate": 1.308940906822368e-05,
+      "loss": 0.9613,
+      "step": 4080
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.34765625,
+      "learning_rate": 1.2909759825694411e-05,
+      "loss": 1.0514,
+      "step": 4085
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.30859375,
+      "learning_rate": 1.273126683574858e-05,
+      "loss": 1.0097,
+      "step": 4090
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.4765625,
+      "learning_rate": 1.2553932468146312e-05,
+      "loss": 1.0627,
+      "step": 4095
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.44140625,
+      "learning_rate": 1.2377759077265361e-05,
+      "loss": 1.0088,
+      "step": 4100
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.404296875,
+      "learning_rate": 1.2202749002069691e-05,
+      "loss": 1.1016,
+      "step": 4105
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.359375,
+      "learning_rate": 1.2028904566078602e-05,
+      "loss": 1.0947,
+      "step": 4110
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.4609375,
+      "learning_rate": 1.1856228077335773e-05,
+      "loss": 1.0247,
+      "step": 4115
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.345703125,
+      "learning_rate": 1.1684721828378642e-05,
+      "loss": 1.0542,
+      "step": 4120
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.330078125,
+      "learning_rate": 1.1514388096207972e-05,
+      "loss": 0.974,
+      "step": 4125
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.63671875,
+      "learning_rate": 1.1345229142257629e-05,
+      "loss": 1.1255,
+      "step": 4130
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.349609375,
+      "learning_rate": 1.1177247212364528e-05,
+      "loss": 1.0134,
+      "step": 4135
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.349609375,
+      "learning_rate": 1.1010444536738884e-05,
+      "loss": 1.0641,
+      "step": 4140
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.34765625,
+      "learning_rate": 1.0844823329934472e-05,
+      "loss": 1.0691,
+      "step": 4145
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.353515625,
+      "learning_rate": 1.0680385790819391e-05,
+      "loss": 1.0196,
+      "step": 4150
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.39453125,
+      "learning_rate": 1.0517134102546766e-05,
+      "loss": 1.0477,
+      "step": 4155
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.333984375,
+      "learning_rate": 1.0355070432525772e-05,
+      "loss": 1.0956,
+      "step": 4160
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.361328125,
+      "learning_rate": 1.0194196932392874e-05,
+      "loss": 1.0475,
+      "step": 4165
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.486328125,
+      "learning_rate": 1.0034515737983264e-05,
+      "loss": 1.083,
+      "step": 4170
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.40234375,
+      "learning_rate": 9.87602896930252e-06,
+      "loss": 1.1112,
+      "step": 4175
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.349609375,
+      "learning_rate": 9.718738730498422e-06,
+      "loss": 1.0079,
+      "step": 4180
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.34375,
+      "learning_rate": 9.562647109833023e-06,
+      "loss": 1.0678,
+      "step": 4185
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.357421875,
+      "learning_rate": 9.40775617965496e-06,
+      "loss": 0.9273,
+      "step": 4190
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.353515625,
+      "learning_rate": 9.254067996371896e-06,
+      "loss": 1.1094,
+      "step": 4195
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.53125,
+      "learning_rate": 9.101584600423263e-06,
+      "loss": 1.1018,
+      "step": 4200
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.412109375,
+      "learning_rate": 8.950308016253129e-06,
+      "loss": 1.1271,
+      "step": 4205
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.39453125,
+      "learning_rate": 8.800240252283343e-06,
+      "loss": 1.1965,
+      "step": 4210
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.36328125,
+      "learning_rate": 8.651383300886884e-06,
+      "loss": 0.9942,
+      "step": 4215
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.34765625,
+      "learning_rate": 8.503739138361344e-06,
+      "loss": 1.0004,
+      "step": 4220
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.3671875,
+      "learning_rate": 8.357309724902773e-06,
+      "loss": 1.1956,
+      "step": 4225
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.328125,
+      "learning_rate": 8.212097004579622e-06,
+      "loss": 1.0484,
+      "step": 4230
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.3203125,
+      "learning_rate": 8.068102905306929e-06,
+      "loss": 0.9999,
+      "step": 4235
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.38671875,
+      "learning_rate": 7.925329338820709e-06,
+      "loss": 0.9657,
+      "step": 4240
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.396484375,
+      "learning_rate": 7.783778200652591e-06,
+      "loss": 1.1274,
+      "step": 4245
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.375,
+      "learning_rate": 7.643451370104648e-06,
+      "loss": 1.1039,
+      "step": 4250
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.7890625,
+      "learning_rate": 7.504350710224484e-06,
+      "loss": 1.1729,
+      "step": 4255
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.384765625,
+      "learning_rate": 7.366478067780369e-06,
+      "loss": 1.0566,
+      "step": 4260
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.3671875,
+      "learning_rate": 7.229835273236918e-06,
+      "loss": 1.0475,
+      "step": 4265
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.357421875,
+      "learning_rate": 7.094424140730615e-06,
+      "loss": 1.0622,
+      "step": 4270
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.376953125,
+      "learning_rate": 6.960246468045839e-06,
+      "loss": 0.9255,
+      "step": 4275
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.390625,
+      "learning_rate": 6.82730403659092e-06,
+      "loss": 1.0548,
+      "step": 4280
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.34765625,
+      "learning_rate": 6.695598611374554e-06,
+      "loss": 0.967,
+      "step": 4285
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.326171875,
+      "learning_rate": 6.565131940982316e-06,
+      "loss": 1.204,
+      "step": 4290
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.3203125,
+      "learning_rate": 6.4359057575534624e-06,
+      "loss": 1.0044,
+      "step": 4295
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.328125,
+      "learning_rate": 6.307921776757953e-06,
+      "loss": 1.0688,
+      "step": 4300
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.396484375,
+      "learning_rate": 6.181181697773664e-06,
+      "loss": 1.0628,
+      "step": 4305
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.55859375,
+      "learning_rate": 6.055687203263794e-06,
+      "loss": 0.8719,
+      "step": 4310
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.4765625,
+      "learning_rate": 5.9314399593545785e-06,
+      "loss": 0.9327,
+      "step": 4315
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.3515625,
+      "learning_rate": 5.808441615613147e-06,
+      "loss": 1.1326,
+      "step": 4320
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.365234375,
+      "learning_rate": 5.686693805025622e-06,
+      "loss": 1.0395,
+      "step": 4325
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.361328125,
+      "learning_rate": 5.566198143975398e-06,
+      "loss": 1.0369,
+      "step": 4330
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.322265625,
+      "learning_rate": 5.446956232221767e-06,
+      "loss": 1.1461,
+      "step": 4335
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.37109375,
+      "learning_rate": 5.328969652878624e-06,
+      "loss": 1.1625,
+      "step": 4340
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.33203125,
+      "learning_rate": 5.2122399723934066e-06,
+      "loss": 1.0224,
+      "step": 4345
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.349609375,
+      "learning_rate": 5.096768740526426e-06,
+      "loss": 1.1103,
+      "step": 4350
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.375,
+      "learning_rate": 4.982557490330175e-06,
+      "loss": 1.0779,
+      "step": 4355
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.34375,
+      "learning_rate": 4.869607738129011e-06,
+      "loss": 1.1403,
+      "step": 4360
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.365234375,
+      "learning_rate": 4.757920983499053e-06,
+      "loss": 1.1142,
+      "step": 4365
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.40625,
+      "learning_rate": 4.6474987092481996e-06,
+      "loss": 1.1299,
+      "step": 4370
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.330078125,
+      "learning_rate": 4.538342381396532e-06,
+      "loss": 0.9553,
+      "step": 4375
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.3671875,
+      "learning_rate": 4.430453449156802e-06,
+      "loss": 1.0479,
+      "step": 4380
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.35546875,
+      "learning_rate": 4.323833344915185e-06,
+      "loss": 1.0784,
+      "step": 4385
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.396484375,
+      "learning_rate": 4.218483484212277e-06,
+      "loss": 1.0951,
+      "step": 4390
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.37890625,
+      "learning_rate": 4.114405265724319e-06,
+      "loss": 1.067,
+      "step": 4395
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.330078125,
+      "learning_rate": 4.011600071244592e-06,
+      "loss": 1.0339,
+      "step": 4400
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.373046875,
+      "learning_rate": 3.910069265665106e-06,
+      "loss": 1.0851,
+      "step": 4405
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.3125,
+      "learning_rate": 3.8098141969584167e-06,
+      "loss": 1.0284,
+      "step": 4410
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.3515625,
+      "learning_rate": 3.710836196159806e-06,
+      "loss": 0.9502,
+      "step": 4415
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.36328125,
+      "learning_rate": 3.613136577349596e-06,
+      "loss": 0.9807,
+      "step": 4420
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.345703125,
+      "learning_rate": 3.516716637635664e-06,
+      "loss": 0.9914,
+      "step": 4425
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.392578125,
+      "learning_rate": 3.4215776571362213e-06,
+      "loss": 1.1339,
+      "step": 4430
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.326171875,
+      "learning_rate": 3.3277208989628826e-06,
+      "loss": 1.0262,
+      "step": 4435
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.3359375,
+      "learning_rate": 3.2351476092038367e-06,
+      "loss": 0.9285,
+      "step": 4440
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.345703125,
+      "learning_rate": 3.1438590169072914e-06,
+      "loss": 1.0673,
+      "step": 4445
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.37109375,
+      "learning_rate": 3.0538563340651973e-06,
+      "loss": 1.104,
+      "step": 4450
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.359375,
+      "learning_rate": 2.9651407555971734e-06,
+      "loss": 1.1608,
+      "step": 4455
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.35546875,
+      "learning_rate": 2.8777134593345854e-06,
+      "loss": 1.1182,
+      "step": 4460
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.36328125,
+      "learning_rate": 2.791575606004926e-06,
+      "loss": 1.0455,
+      "step": 4465
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.34765625,
+      "learning_rate": 2.7067283392164354e-06,
+      "loss": 1.029,
+      "step": 4470
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.3515625,
+      "learning_rate": 2.6231727854428843e-06,
+      "loss": 1.0893,
+      "step": 4475
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.40625,
+      "learning_rate": 2.5409100540086274e-06,
+      "loss": 1.0287,
+      "step": 4480
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.388671875,
+      "learning_rate": 2.4599412370738815e-06,
+      "loss": 0.948,
+      "step": 4485
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.34375,
+      "learning_rate": 2.3802674096202404e-06,
+      "loss": 0.9194,
+      "step": 4490
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.392578125,
+      "learning_rate": 2.301889629436349e-06,
+      "loss": 1.0878,
+      "step": 4495
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.369140625,
+      "learning_rate": 2.224808937103917e-06,
+      "loss": 1.0251,
+      "step": 4500
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.396484375,
+      "learning_rate": 2.149026355983896e-06,
+      "loss": 0.9898,
+      "step": 4505
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.375,
+      "learning_rate": 2.0745428922028576e-06,
+      "loss": 1.0231,
+      "step": 4510
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.376953125,
+      "learning_rate": 2.0013595346396462e-06,
+      "loss": 1.0693,
+      "step": 4515
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.37109375,
+      "learning_rate": 1.9294772549122932e-06,
+      "loss": 0.9087,
+      "step": 4520
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.3515625,
+      "learning_rate": 1.858897007365057e-06,
+      "loss": 1.015,
+      "step": 4525
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.341796875,
+      "learning_rate": 1.7896197290557914e-06,
+      "loss": 1.0226,
+      "step": 4530
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.365234375,
+      "learning_rate": 1.721646339743499e-06,
+      "loss": 1.0009,
+      "step": 4535
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.375,
+      "learning_rate": 1.654977741876107e-06,
+      "loss": 1.1028,
+      "step": 4540
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.337890625,
+      "learning_rate": 1.5896148205785e-06,
+      "loss": 1.0865,
+      "step": 4545
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.365234375,
+      "learning_rate": 1.5255584436407622e-06,
+      "loss": 0.9828,
+      "step": 4550
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.380859375,
+      "learning_rate": 1.4628094615066645e-06,
+      "loss": 0.9875,
+      "step": 4555
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.078125,
+      "learning_rate": 1.4013687072623294e-06,
+      "loss": 1.1048,
+      "step": 4560
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.333984375,
+      "learning_rate": 1.3412369966252392e-06,
+      "loss": 0.9696,
+      "step": 4565
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.337890625,
+      "learning_rate": 1.2824151279333673e-06,
+      "loss": 1.1658,
+      "step": 4570
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.396484375,
+      "learning_rate": 1.2249038821345982e-06,
+      "loss": 1.0483,
+      "step": 4575
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.3828125,
+      "learning_rate": 1.1687040227763124e-06,
+      "loss": 1.0442,
+      "step": 4580
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.388671875,
+      "learning_rate": 1.113816295995318e-06,
+      "loss": 1.1091,
+      "step": 4585
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.3515625,
+      "learning_rate": 1.0602414305079023e-06,
+      "loss": 1.089,
+      "step": 4590
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.34765625,
+      "learning_rate": 1.0079801376001398e-06,
+      "loss": 1.0591,
+      "step": 4595
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.333984375,
+      "learning_rate": 9.570331111184883e-07,
+      "loss": 1.102,
+      "step": 4600
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.3671875,
+      "learning_rate": 9.074010274605859e-07,
+      "loss": 1.075,
+      "step": 4605
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.419921875,
+      "learning_rate": 8.590845455662133e-07,
+      "loss": 1.1246,
+      "step": 4610
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.380859375,
+      "learning_rate": 8.120843069085893e-07,
+      "loss": 1.0597,
+      "step": 4615
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.365234375,
+      "learning_rate": 7.66400935485867e-07,
+      "loss": 1.1967,
+      "step": 4620
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.345703125,
+      "learning_rate": 7.220350378127961e-07,
+      "loss": 1.0108,
+      "step": 4625
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.375,
+      "learning_rate": 6.789872029127064e-07,
+      "loss": 1.0265,
+      "step": 4630
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.333984375,
+      "learning_rate": 6.37258002309693e-07,
+      "loss": 0.8866,
+      "step": 4635
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.337890625,
+      "learning_rate": 5.968479900210211e-07,
+      "loss": 0.9801,
+      "step": 4640
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.349609375,
+      "learning_rate": 5.577577025497438e-07,
+      "loss": 1.0485,
+      "step": 4645
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.365234375,
+      "learning_rate": 5.199876588776409e-07,
+      "loss": 0.9552,
+      "step": 4650
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.34765625,
+      "learning_rate": 4.835383604582467e-07,
+      "loss": 1.0244,
+      "step": 4655
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.322265625,
+      "learning_rate": 4.484102912102661e-07,
+      "loss": 0.9478,
+      "step": 4660
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.341796875,
+      "learning_rate": 4.1460391751110274e-07,
+      "loss": 1.0376,
+      "step": 4665
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.384765625,
+      "learning_rate": 3.821196881907074e-07,
+      "loss": 1.1337,
+      "step": 4670
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.4296875,
+      "learning_rate": 3.5095803452557254e-07,
+      "loss": 1.1147,
+      "step": 4675
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.462890625,
+      "learning_rate": 3.211193702330362e-07,
+      "loss": 0.9671,
+      "step": 4680
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.345703125,
+      "learning_rate": 2.9260409146578684e-07,
+      "loss": 1.0073,
+      "step": 4685
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.34765625,
+      "learning_rate": 2.654125768066118e-07,
+      "loss": 1.1421,
+      "step": 4690
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.376953125,
+      "learning_rate": 2.3954518726332365e-07,
+      "loss": 1.0143,
+      "step": 4695
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.3671875,
+      "learning_rate": 2.1500226626404163e-07,
+      "loss": 0.8934,
+      "step": 4700
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.35546875,
+      "learning_rate": 1.9178413965258435e-07,
+      "loss": 1.0531,
+      "step": 4705
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.333984375,
+      "learning_rate": 1.6989111568411764e-07,
+      "loss": 0.9631,
+      "step": 4710
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.35546875,
+      "learning_rate": 1.4932348502114669e-07,
+      "loss": 0.9886,
+      "step": 4715
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.3671875,
+      "learning_rate": 1.3008152072958579e-07,
+      "loss": 1.0686,
+      "step": 4720
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.1216547827515022e-07,
+      "loss": 1.0989,
+      "step": 4725
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.345703125,
+      "learning_rate": 9.557559551996998e-08,
+      "loss": 0.9859,
+      "step": 4730
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.373046875,
+      "learning_rate": 8.031209271944784e-08,
+      "loss": 0.98,
+      "step": 4735
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.376953125,
+      "learning_rate": 6.637517251928404e-08,
+      "loss": 1.0466,
+      "step": 4740
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.333984375,
+      "learning_rate": 5.37650199528672e-08,
+      "loss": 1.169,
+      "step": 4745
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.328125,
+      "learning_rate": 4.2481802438720794e-08,
+      "loss": 1.0257,
+      "step": 4750
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.37109375,
+      "learning_rate": 3.252566977837157e-08,
+      "loss": 1.0731,
+      "step": 4755
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.046875,
+      "learning_rate": 2.3896754154284497e-08,
+      "loss": 0.9857,
+      "step": 4760
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.3671875,
+      "learning_rate": 1.659517012817524e-08,
+      "loss": 1.0577,
+      "step": 4765
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 2.625,
+      "learning_rate": 1.062101463944476e-08,
+      "loss": 1.1085,
+      "step": 4770
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.3203125,
+      "learning_rate": 5.974367003880321e-09,
+      "loss": 1.1863,
+      "step": 4775
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.32421875,
+      "learning_rate": 2.6552889126563174e-09,
+      "loss": 1.0574,
+      "step": 4780
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.34375,
+      "learning_rate": 6.638244314793874e-10,
+      "loss": 0.9897,
+      "step": 4785
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.40234375,
+      "learning_rate": 0.0,
+      "loss": 1.1193,
+      "step": 4790
+    },
+    {
+      "epoch": 1.0,
+      "step": 4790,
+      "total_flos": 1.007099383382016e+18,
+      "train_loss": 1.0868773834192678,
+      "train_runtime": 33367.2438,
+      "train_samples_per_second": 1.149,
+      "train_steps_per_second": 0.144
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 4790,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 100,
+  "total_flos": 1.007099383382016e+18,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}