diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5621 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.8313847752663029,
+  "eval_steps": 500,
+  "global_step": 4000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 24.954891204833984,
+      "learning_rate": 5e-06,
+      "loss": 9.5118,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 14.123294830322266,
+      "learning_rate": 1e-05,
+      "loss": 9.3162,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 11.853605270385742,
+      "learning_rate": 1.5e-05,
+      "loss": 8.9519,
+      "step": 15
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 8.794404983520508,
+      "learning_rate": 2e-05,
+      "loss": 8.511,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.945346355438232,
+      "learning_rate": 2.5e-05,
+      "loss": 8.1291,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.806889057159424,
+      "learning_rate": 3e-05,
+      "loss": 7.8939,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.950843572616577,
+      "learning_rate": 3.5e-05,
+      "loss": 7.808,
+      "step": 35
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.9586803913116455,
+      "learning_rate": 4e-05,
+      "loss": 7.7356,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.5460443496704102,
+      "learning_rate": 4.5e-05,
+      "loss": 7.6671,
+      "step": 45
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.6627991199493408,
+      "learning_rate": 5e-05,
+      "loss": 7.5727,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.6151450872421265,
+      "learning_rate": 5.500000000000001e-05,
+      "loss": 7.4721,
+      "step": 55
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.017178773880005,
+      "learning_rate": 6e-05,
+      "loss": 7.306,
+      "step": 60
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.126413345336914,
+      "learning_rate": 6.500000000000001e-05,
+      "loss": 7.1504,
+      "step": 65
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.792532444000244,
+      "learning_rate": 7e-05,
+      "loss": 6.9702,
+      "step": 70
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.8019553422927856,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 6.7947,
+      "step": 75
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.513380527496338,
+      "learning_rate": 8e-05,
+      "loss": 6.678,
+      "step": 80
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.021646022796631,
+      "learning_rate": 8.5e-05,
+      "loss": 6.5002,
+      "step": 85
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 5.583535194396973,
+      "learning_rate": 9e-05,
+      "loss": 6.4036,
+      "step": 90
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.9732987880706787,
+      "learning_rate": 9.5e-05,
+      "loss": 6.3155,
+      "step": 95
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.5459728240966797,
+      "learning_rate": 0.0001,
+      "loss": 6.1436,
+      "step": 100
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.8032758235931396,
+      "learning_rate": 9.999972205865686e-05,
+      "loss": 6.099,
+      "step": 105
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.432061195373535,
+      "learning_rate": 9.999888823771751e-05,
+      "loss": 5.9576,
+      "step": 110
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.0502891540527344,
+      "learning_rate": 9.999749854645204e-05,
+      "loss": 5.8158,
+      "step": 115
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.0243988037109375,
+      "learning_rate": 9.99955530003106e-05,
+      "loss": 5.7212,
+      "step": 120
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.176409959793091,
+      "learning_rate": 9.99930516209231e-05,
+      "loss": 5.6199,
+      "step": 125
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.457683563232422,
+      "learning_rate": 9.998999443609897e-05,
+      "loss": 5.5896,
+      "step": 130
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.6271471977233887,
+      "learning_rate": 9.998638147982696e-05,
+      "loss": 5.3765,
+      "step": 135
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.7946786880493164,
+      "learning_rate": 9.998221279227467e-05,
+      "loss": 5.3568,
+      "step": 140
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.4872028827667236,
+      "learning_rate": 9.997748841978812e-05,
+      "loss": 5.3366,
+      "step": 145
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.0820538997650146,
+      "learning_rate": 9.997220841489122e-05,
+      "loss": 5.2398,
+      "step": 150
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.47347354888916,
+      "learning_rate": 9.996637283628528e-05,
+      "loss": 5.1045,
+      "step": 155
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.871932029724121,
+      "learning_rate": 9.995998174884821e-05,
+      "loss": 5.09,
+      "step": 160
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.723886251449585,
+      "learning_rate": 9.995303522363394e-05,
+      "loss": 5.0264,
+      "step": 165
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.1860122680664062,
+      "learning_rate": 9.99455333378715e-05,
+      "loss": 4.9358,
+      "step": 170
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.0296273231506348,
+      "learning_rate": 9.993747617496428e-05,
+      "loss": 4.9147,
+      "step": 175
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.251325845718384,
+      "learning_rate": 9.9928863824489e-05,
+      "loss": 4.8911,
+      "step": 180
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.8915772438049316,
+      "learning_rate": 9.99196963821948e-05,
+      "loss": 4.8019,
+      "step": 185
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.635601758956909,
+      "learning_rate": 9.990997395000217e-05,
+      "loss": 4.7388,
+      "step": 190
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.413235664367676,
+      "learning_rate": 9.989969663600169e-05,
+      "loss": 4.6891,
+      "step": 195
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.7400410175323486,
+      "learning_rate": 9.9888864554453e-05,
+      "loss": 4.6709,
+      "step": 200
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.8295705318450928,
+      "learning_rate": 9.987747782578342e-05,
+      "loss": 4.5952,
+      "step": 205
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.3502957820892334,
+      "learning_rate": 9.986553657658668e-05,
+      "loss": 4.5546,
+      "step": 210
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.5216689109802246,
+      "learning_rate": 9.985304093962145e-05,
+      "loss": 4.5998,
+      "step": 215
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.301525115966797,
+      "learning_rate": 9.983999105380988e-05,
+      "loss": 4.5215,
+      "step": 220
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.87137508392334,
+      "learning_rate": 9.982638706423608e-05,
+      "loss": 4.5889,
+      "step": 225
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.8169140815734863,
+      "learning_rate": 9.98122291221445e-05,
+      "loss": 4.4965,
+      "step": 230
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.8885302543640137,
+      "learning_rate": 9.979751738493826e-05,
+      "loss": 4.4749,
+      "step": 235
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.511671304702759,
+      "learning_rate": 9.978225201617732e-05,
+      "loss": 4.4188,
+      "step": 240
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.7718353271484375,
+      "learning_rate": 9.976643318557678e-05,
+      "loss": 4.4041,
+      "step": 245
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.0946099758148193,
+      "learning_rate": 9.975006106900495e-05,
+      "loss": 4.4226,
+      "step": 250
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.0956268310546875,
+      "learning_rate": 9.973313584848132e-05,
+      "loss": 4.3768,
+      "step": 255
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.8771021366119385,
+      "learning_rate": 9.971565771217464e-05,
+      "loss": 4.3758,
+      "step": 260
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.6710236072540283,
+      "learning_rate": 9.969762685440076e-05,
+      "loss": 4.2965,
+      "step": 265
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.7884382009506226,
+      "learning_rate": 9.967904347562054e-05,
+      "loss": 4.2645,
+      "step": 270
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.9744821786880493,
+      "learning_rate": 9.965990778243755e-05,
+      "loss": 4.3013,
+      "step": 275
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.9575978517532349,
+      "learning_rate": 9.964021998759577e-05,
+      "loss": 4.2548,
+      "step": 280
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.7722605466842651,
+      "learning_rate": 9.961998030997733e-05,
+      "loss": 4.1621,
+      "step": 285
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.2758901119232178,
+      "learning_rate": 9.95991889745999e-05,
+      "loss": 4.2537,
+      "step": 290
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.0251128673553467,
+      "learning_rate": 9.957784621261441e-05,
+      "loss": 4.1987,
+      "step": 295
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.0095252990722656,
+      "learning_rate": 9.955595226130226e-05,
+      "loss": 4.2274,
+      "step": 300
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.6288301944732666,
+      "learning_rate": 9.953350736407282e-05,
+      "loss": 4.1804,
+      "step": 305
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.0467007160186768,
+      "learning_rate": 9.951051177046069e-05,
+      "loss": 4.1055,
+      "step": 310
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.819460391998291,
+      "learning_rate": 9.948696573612292e-05,
+      "loss": 4.1044,
+      "step": 315
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.1974942684173584,
+      "learning_rate": 9.946286952283618e-05,
+      "loss": 4.1271,
+      "step": 320
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.006293773651123,
+      "learning_rate": 9.943822339849381e-05,
+      "loss": 4.0771,
+      "step": 325
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.8218380212783813,
+      "learning_rate": 9.941302763710288e-05,
+      "loss": 3.9985,
+      "step": 330
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.2746903896331787,
+      "learning_rate": 9.938728251878116e-05,
+      "loss": 4.1088,
+      "step": 335
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.773939609527588,
+      "learning_rate": 9.936098832975393e-05,
+      "loss": 4.0133,
+      "step": 340
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.7421215772628784,
+      "learning_rate": 9.933414536235091e-05,
+      "loss": 3.9569,
+      "step": 345
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.555828332901001,
+      "learning_rate": 9.93067539150029e-05,
+      "loss": 3.9792,
+      "step": 350
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.7069153785705566,
+      "learning_rate": 9.927881429223853e-05,
+      "loss": 3.9499,
+      "step": 355
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.7734919786453247,
+      "learning_rate": 9.925032680468085e-05,
+      "loss": 3.9715,
+      "step": 360
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.539474606513977,
+      "learning_rate": 9.922129176904388e-05,
+      "loss": 3.9238,
+      "step": 365
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.5540482997894287,
+      "learning_rate": 9.919170950812911e-05,
+      "loss": 3.8796,
+      "step": 370
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.549526333808899,
+      "learning_rate": 9.916158035082184e-05,
+      "loss": 3.8584,
+      "step": 375
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.7199825048446655,
+      "learning_rate": 9.913090463208763e-05,
+      "loss": 3.8513,
+      "step": 380
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.535899043083191,
+      "learning_rate": 9.90996826929685e-05,
+      "loss": 3.873,
+      "step": 385
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.3458640575408936,
+      "learning_rate": 9.906791488057916e-05,
+      "loss": 3.8087,
+      "step": 390
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.6906614303588867,
+      "learning_rate": 9.903560154810313e-05,
+      "loss": 3.8095,
+      "step": 395
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.5479685068130493,
+      "learning_rate": 9.900274305478887e-05,
+      "loss": 3.8874,
+      "step": 400
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.3803184032440186,
+      "learning_rate": 9.896933976594572e-05,
+      "loss": 3.8076,
+      "step": 405
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.563531517982483,
+      "learning_rate": 9.893539205293989e-05,
+      "loss": 3.8089,
+      "step": 410
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.5819116830825806,
+      "learning_rate": 9.890090029319028e-05,
+      "loss": 3.7708,
+      "step": 415
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.441551923751831,
+      "learning_rate": 9.886586487016433e-05,
+      "loss": 3.641,
+      "step": 420
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.2000863552093506,
+      "learning_rate": 9.883028617337378e-05,
+      "loss": 3.8062,
+      "step": 425
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.5903886556625366,
+      "learning_rate": 9.879416459837022e-05,
+      "loss": 3.6873,
+      "step": 430
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3820626735687256,
+      "learning_rate": 9.875750054674082e-05,
+      "loss": 3.6398,
+      "step": 435
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3108272552490234,
+      "learning_rate": 9.872029442610382e-05,
+      "loss": 3.6804,
+      "step": 440
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.4747930765151978,
+      "learning_rate": 9.8682546650104e-05,
+      "loss": 3.693,
+      "step": 445
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3883565664291382,
+      "learning_rate": 9.864425763840802e-05,
+      "loss": 3.6498,
+      "step": 450
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3403358459472656,
+      "learning_rate": 9.860542781669988e-05,
+      "loss": 3.6682,
+      "step": 455
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.2842109203338623,
+      "learning_rate": 9.85660576166761e-05,
+      "loss": 3.649,
+      "step": 460
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.319282054901123,
+      "learning_rate": 9.852614747604093e-05,
+      "loss": 3.6335,
+      "step": 465
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.3801344633102417,
+      "learning_rate": 9.848569783850145e-05,
+      "loss": 3.6318,
+      "step": 470
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.342846393585205,
+      "learning_rate": 9.844470915376278e-05,
+      "loss": 3.604,
+      "step": 475
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.452356219291687,
+      "learning_rate": 9.840318187752292e-05,
+      "loss": 3.599,
+      "step": 480
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.323322057723999,
+      "learning_rate": 9.836111647146771e-05,
+      "loss": 3.5778,
+      "step": 485
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.1092493534088135,
+      "learning_rate": 9.831851340326577e-05,
+      "loss": 3.6146,
+      "step": 490
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.1545424461364746,
+      "learning_rate": 9.82753731465633e-05,
+      "loss": 3.6122,
+      "step": 495
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.1281070709228516,
+      "learning_rate": 9.823169618097871e-05,
+      "loss": 3.6009,
+      "step": 500
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.1365258693695068,
+      "learning_rate": 9.81874829920974e-05,
+      "loss": 3.5677,
+      "step": 505
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1423096656799316,
+      "learning_rate": 9.814273407146623e-05,
+      "loss": 3.5811,
+      "step": 510
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1574337482452393,
+      "learning_rate": 9.809744991658829e-05,
+      "loss": 3.5232,
+      "step": 515
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.2431848049163818,
+      "learning_rate": 9.805163103091708e-05,
+      "loss": 3.5447,
+      "step": 520
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1095994710922241,
+      "learning_rate": 9.800527792385112e-05,
+      "loss": 3.4747,
+      "step": 525
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.259093165397644,
+      "learning_rate": 9.79583911107282e-05,
+      "loss": 3.4696,
+      "step": 530
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.3247113227844238,
+      "learning_rate": 9.791097111281968e-05,
+      "loss": 3.4693,
+      "step": 535
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1355781555175781,
+      "learning_rate": 9.786301845732467e-05,
+      "loss": 3.4687,
+      "step": 540
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.9854279160499573,
+      "learning_rate": 9.781453367736418e-05,
+      "loss": 3.4761,
+      "step": 545
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.998266339302063,
+      "learning_rate": 9.776551731197524e-05,
+      "loss": 3.4733,
+      "step": 550
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.0570634603500366,
+      "learning_rate": 9.771596990610478e-05,
+      "loss": 3.483,
+      "step": 555
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.182489037513733,
+      "learning_rate": 9.766589201060372e-05,
+      "loss": 3.4698,
+      "step": 560
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.718491554260254,
+      "learning_rate": 9.761528418222077e-05,
+      "loss": 3.4412,
+      "step": 565
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.1469898223876953,
+      "learning_rate": 9.756414698359624e-05,
+      "loss": 3.3577,
+      "step": 570
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.014533281326294,
+      "learning_rate": 9.75124809832558e-05,
+      "loss": 3.3677,
+      "step": 575
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.0099555253982544,
+      "learning_rate": 9.746028675560413e-05,
+      "loss": 3.4261,
+      "step": 580
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.081205129623413,
+      "learning_rate": 9.740756488091861e-05,
+      "loss": 3.3498,
+      "step": 585
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.2028361558914185,
+      "learning_rate": 9.735431594534277e-05,
+      "loss": 3.4458,
+      "step": 590
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.1756874322891235,
+      "learning_rate": 9.730054054087983e-05,
+      "loss": 3.3875,
+      "step": 595
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.140699028968811,
+      "learning_rate": 9.724623926538612e-05,
+      "loss": 3.3088,
+      "step": 600
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.9892724752426147,
+      "learning_rate": 9.719141272256443e-05,
+      "loss": 3.3534,
+      "step": 605
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.045665979385376,
+      "learning_rate": 9.713606152195726e-05,
+      "loss": 3.4133,
+      "step": 610
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.0393067598342896,
+      "learning_rate": 9.708018627894011e-05,
+      "loss": 3.4175,
+      "step": 615
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.099212169647217,
+      "learning_rate": 9.702378761471456e-05,
+      "loss": 3.4136,
+      "step": 620
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.9463175535202026,
+      "learning_rate": 9.696686615630146e-05,
+      "loss": 3.3175,
+      "step": 625
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.9630415439605713,
+      "learning_rate": 9.690942253653385e-05,
+      "loss": 3.343,
+      "step": 630
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.9103713035583496,
+      "learning_rate": 9.685145739405002e-05,
+      "loss": 3.3291,
+      "step": 635
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.01008939743042,
+      "learning_rate": 9.679297137328634e-05,
+      "loss": 3.3743,
+      "step": 640
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.0484918355941772,
+      "learning_rate": 9.673396512447013e-05,
+      "loss": 3.2935,
+      "step": 645
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.9054334759712219,
+      "learning_rate": 9.667443930361247e-05,
+      "loss": 3.2916,
+      "step": 650
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.2290186882019043,
+      "learning_rate": 9.661439457250076e-05,
+      "loss": 3.3105,
+      "step": 655
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.8617240786552429,
+      "learning_rate": 9.655383159869158e-05,
+      "loss": 3.2651,
+      "step": 660
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.8485407829284668,
+      "learning_rate": 9.649275105550309e-05,
+      "loss": 3.1712,
+      "step": 665
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.8689857125282288,
+      "learning_rate": 9.643115362200762e-05,
+      "loss": 3.2274,
+      "step": 670
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.8368767499923706,
+      "learning_rate": 9.636903998302409e-05,
+      "loss": 3.264,
+      "step": 675
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.9838525652885437,
+      "learning_rate": 9.630641082911045e-05,
+      "loss": 3.2652,
+      "step": 680
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.9460552334785461,
+      "learning_rate": 9.624326685655593e-05,
+      "loss": 3.3197,
+      "step": 685
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.9261606931686401,
+      "learning_rate": 9.617960876737337e-05,
+      "loss": 3.3094,
+      "step": 690
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.8362047672271729,
+      "learning_rate": 9.611543726929134e-05,
+      "loss": 3.2742,
+      "step": 695
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.8609175086021423,
+      "learning_rate": 9.605075307574635e-05,
+      "loss": 3.2256,
+      "step": 700
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.8735069036483765,
+      "learning_rate": 9.598555690587487e-05,
+      "loss": 3.3234,
+      "step": 705
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.8475310206413269,
+      "learning_rate": 9.591984948450532e-05,
+      "loss": 3.2328,
+      "step": 710
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.831333577632904,
+      "learning_rate": 9.585363154215008e-05,
+      "loss": 3.1681,
+      "step": 715
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.7871400713920593,
+      "learning_rate": 9.578690381499728e-05,
+      "loss": 3.2017,
+      "step": 720
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.8291134238243103,
+      "learning_rate": 9.571966704490271e-05,
+      "loss": 3.2111,
+      "step": 725
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.8258323669433594,
+      "learning_rate": 9.565192197938148e-05,
+      "loss": 3.213,
+      "step": 730
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.8596241474151611,
+      "learning_rate": 9.558366937159977e-05,
+      "loss": 3.1932,
+      "step": 735
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.8011757135391235,
+      "learning_rate": 9.551490998036646e-05,
+      "loss": 3.2353,
+      "step": 740
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.8874036073684692,
+      "learning_rate": 9.544564457012463e-05,
+      "loss": 3.1661,
+      "step": 745
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.9632325768470764,
+      "learning_rate": 9.537587391094314e-05,
+      "loss": 3.1521,
+      "step": 750
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.7759344577789307,
+      "learning_rate": 9.5305598778508e-05,
+      "loss": 3.209,
+      "step": 755
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.8687903881072998,
+      "learning_rate": 9.52348199541138e-05,
+      "loss": 3.1717,
+      "step": 760
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.7881929278373718,
+      "learning_rate": 9.516353822465504e-05,
+      "loss": 3.203,
+      "step": 765
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.8782910704612732,
+      "learning_rate": 9.509175438261726e-05,
+      "loss": 3.1721,
+      "step": 770
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.8087301850318909,
+      "learning_rate": 9.501946922606838e-05,
+      "loss": 3.1684,
+      "step": 775
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.737407922744751,
+      "learning_rate": 9.494668355864973e-05,
+      "loss": 3.0995,
+      "step": 780
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.8467393517494202,
+      "learning_rate": 9.487339818956716e-05,
+      "loss": 3.1303,
+      "step": 785
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.7596563100814819,
+      "learning_rate": 9.479961393358203e-05,
+      "loss": 3.1472,
+      "step": 790
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.8828564882278442,
+      "learning_rate": 9.472533161100215e-05,
+      "loss": 3.166,
+      "step": 795
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.787274956703186,
+      "learning_rate": 9.465055204767265e-05,
+      "loss": 3.2096,
+      "step": 800
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.7758961915969849,
+      "learning_rate": 9.457527607496685e-05,
+      "loss": 3.1402,
+      "step": 805
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.8374167680740356,
+      "learning_rate": 9.44995045297769e-05,
+      "loss": 3.13,
+      "step": 810
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.9025399088859558,
+      "learning_rate": 9.442323825450464e-05,
+      "loss": 3.0811,
+      "step": 815
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.7858222723007202,
+      "learning_rate": 9.43464780970521e-05,
+      "loss": 3.1224,
+      "step": 820
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.8241803646087646,
+      "learning_rate": 9.426922491081212e-05,
+      "loss": 3.1728,
+      "step": 825
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.8176282048225403,
+      "learning_rate": 9.419147955465888e-05,
+      "loss": 3.1275,
+      "step": 830
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.7991018295288086,
+      "learning_rate": 9.411324289293832e-05,
+      "loss": 3.0839,
+      "step": 835
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.8595659732818604,
+      "learning_rate": 9.403451579545859e-05,
+      "loss": 3.1161,
+      "step": 840
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.8479962944984436,
+      "learning_rate": 9.395529913748025e-05,
+      "loss": 3.0946,
+      "step": 845
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.775111198425293,
+      "learning_rate": 9.387559379970672e-05,
+      "loss": 3.0524,
+      "step": 850
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.7162839770317078,
+      "learning_rate": 9.379540066827431e-05,
+      "loss": 3.0587,
+      "step": 855
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.787935197353363,
+      "learning_rate": 9.371472063474248e-05,
+      "loss": 3.0199,
+      "step": 860
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.7565710544586182,
+      "learning_rate": 9.363355459608394e-05,
+      "loss": 3.0524,
+      "step": 865
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.7326767444610596,
+      "learning_rate": 9.355190345467457e-05,
+      "loss": 3.1168,
+      "step": 870
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.773398220539093,
+      "learning_rate": 9.346976811828352e-05,
+      "loss": 3.0634,
+      "step": 875
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.733391523361206,
+      "learning_rate": 9.338714950006297e-05,
+      "loss": 3.082,
+      "step": 880
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.767120897769928,
+      "learning_rate": 9.330404851853817e-05,
+      "loss": 3.0013,
+      "step": 885
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.738296627998352,
+      "learning_rate": 9.3220466097597e-05,
+      "loss": 3.1004,
+      "step": 890
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.7394375205039978,
+      "learning_rate": 9.313640316647991e-05,
+      "loss": 3.0697,
+      "step": 895
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.7176501154899597,
+      "learning_rate": 9.305186065976945e-05,
+      "loss": 3.1356,
+      "step": 900
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.7473928928375244,
+      "learning_rate": 9.296683951737993e-05,
+      "loss": 3.0234,
+      "step": 905
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.8314266800880432,
+      "learning_rate": 9.288134068454697e-05,
+      "loss": 3.0555,
+      "step": 910
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.6912246942520142,
+      "learning_rate": 9.2795365111817e-05,
+      "loss": 3.0148,
+      "step": 915
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.7679839730262756,
+      "learning_rate": 9.270891375503665e-05,
+      "loss": 3.0237,
+      "step": 920
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.7263458371162415,
+      "learning_rate": 9.262198757534218e-05,
+      "loss": 3.1093,
+      "step": 925
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.7467177510261536,
+      "learning_rate": 9.253458753914874e-05,
+      "loss": 3.0606,
+      "step": 930
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.7480751872062683,
+      "learning_rate": 9.244671461813969e-05,
+      "loss": 3.1225,
+      "step": 935
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.7728322148323059,
+      "learning_rate": 9.235836978925572e-05,
+      "loss": 3.0865,
+      "step": 940
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.7638208866119385,
+      "learning_rate": 9.226955403468406e-05,
+      "loss": 3.0544,
+      "step": 945
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.7154102921485901,
+      "learning_rate": 9.21802683418475e-05,
+      "loss": 2.9843,
+      "step": 950
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.6908529996871948,
+      "learning_rate": 9.209051370339347e-05,
+      "loss": 2.9612,
+      "step": 955
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.7261789441108704,
+      "learning_rate": 9.200029111718295e-05,
+      "loss": 3.0185,
+      "step": 960
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.6965435147285461,
+      "learning_rate": 9.190960158627941e-05,
+      "loss": 3.0435,
+      "step": 965
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.7368409037590027,
+      "learning_rate": 9.181844611893766e-05,
+      "loss": 2.9824,
+      "step": 970
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.8360817432403564,
+      "learning_rate": 9.172682572859261e-05,
+      "loss": 3.0772,
+      "step": 975
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.7241373658180237,
+      "learning_rate": 9.163474143384806e-05,
+      "loss": 3.015,
+      "step": 980
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.7527982592582703,
+      "learning_rate": 9.154219425846528e-05,
+      "loss": 3.0295,
+      "step": 985
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.7530322074890137,
+      "learning_rate": 9.144918523135175e-05,
+      "loss": 3.0634,
+      "step": 990
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.7819015979766846,
+      "learning_rate": 9.13557153865496e-05,
+      "loss": 3.057,
+      "step": 995
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.7606942653656006,
+      "learning_rate": 9.12617857632242e-05,
+      "loss": 2.9505,
+      "step": 1000
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.7142955660820007,
+      "learning_rate": 9.116739740565259e-05,
+      "loss": 2.9851,
+      "step": 1005
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.6874934434890747,
+      "learning_rate": 9.107255136321184e-05,
+      "loss": 2.9952,
+      "step": 1010
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.6932684779167175,
+      "learning_rate": 9.09772486903674e-05,
+      "loss": 2.9791,
+      "step": 1015
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.7075040936470032,
+      "learning_rate": 9.08814904466614e-05,
+      "loss": 3.0384,
+      "step": 1020
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.7663232684135437,
+      "learning_rate": 9.078527769670085e-05,
+      "loss": 2.979,
+      "step": 1025
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.7197521924972534,
+      "learning_rate": 9.068861151014575e-05,
+      "loss": 2.9968,
+      "step": 1030
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.7028963565826416,
+      "learning_rate": 9.05914929616973e-05,
+      "loss": 3.0382,
+      "step": 1035
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.5042543411254883,
+      "learning_rate": 9.04939231310859e-05,
+      "loss": 3.0451,
+      "step": 1040
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.7034931182861328,
+      "learning_rate": 9.039590310305914e-05,
+      "loss": 3.0157,
+      "step": 1045
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.6842995882034302,
+      "learning_rate": 9.029743396736974e-05,
+      "loss": 2.9701,
+      "step": 1050
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.6938946843147278,
+      "learning_rate": 9.019851681876348e-05,
+      "loss": 3.0551,
+      "step": 1055
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.7267816662788391,
+      "learning_rate": 9.009915275696693e-05,
+      "loss": 2.9805,
+      "step": 1060
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.1045072078704834,
+      "learning_rate": 8.999934288667534e-05,
+      "loss": 3.0012,
+      "step": 1065
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.7654233574867249,
+      "learning_rate": 8.989908831754028e-05,
+      "loss": 2.8833,
+      "step": 1070
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.7957326769828796,
+      "learning_rate": 8.979839016415735e-05,
+      "loss": 2.8604,
+      "step": 1075
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.7361999154090881,
+      "learning_rate": 8.969724954605373e-05,
+      "loss": 3.0241,
+      "step": 1080
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.6803563833236694,
+      "learning_rate": 8.959566758767581e-05,
+      "loss": 3.0452,
+      "step": 1085
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.733856737613678,
+      "learning_rate": 8.949364541837661e-05,
+      "loss": 3.0064,
+      "step": 1090
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.7043154239654541,
+      "learning_rate": 8.939118417240329e-05,
+      "loss": 2.9622,
+      "step": 1095
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.7002702355384827,
+      "learning_rate": 8.92882849888845e-05,
+      "loss": 2.9593,
+      "step": 1100
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.6819352507591248,
+      "learning_rate": 8.918494901181773e-05,
+      "loss": 2.921,
+      "step": 1105
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.6527635455131531,
+      "learning_rate": 8.908117739005659e-05,
+      "loss": 2.9374,
+      "step": 1110
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.6832794547080994,
+      "learning_rate": 8.897697127729805e-05,
+      "loss": 2.9854,
+      "step": 1115
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.6709647178649902,
+      "learning_rate": 8.887233183206957e-05,
+      "loss": 2.9673,
+      "step": 1120
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.6724316477775574,
+      "learning_rate": 8.876726021771627e-05,
+      "loss": 2.9038,
+      "step": 1125
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.6602643132209778,
+      "learning_rate": 8.866175760238798e-05,
+      "loss": 2.8952,
+      "step": 1130
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.7111104130744934,
+      "learning_rate": 8.855582515902625e-05,
+      "loss": 2.9609,
+      "step": 1135
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.7529614567756653,
+      "learning_rate": 8.844946406535131e-05,
+      "loss": 2.9714,
+      "step": 1140
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.6821660399436951,
+      "learning_rate": 8.834267550384893e-05,
+      "loss": 2.925,
+      "step": 1145
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.6634496450424194,
+      "learning_rate": 8.823546066175741e-05,
+      "loss": 2.8966,
+      "step": 1150
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.7059546709060669,
+      "learning_rate": 8.81278207310542e-05,
+      "loss": 2.9919,
+      "step": 1155
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.7027754187583923,
+      "learning_rate": 8.801975690844278e-05,
+      "loss": 2.9371,
+      "step": 1160
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.7034463286399841,
+      "learning_rate": 8.791127039533934e-05,
+      "loss": 2.9084,
+      "step": 1165
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.6744077801704407,
+      "learning_rate": 8.780236239785935e-05,
+      "loss": 2.9022,
+      "step": 1170
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.6845873594284058,
+      "learning_rate": 8.76930341268042e-05,
+      "loss": 2.9669,
+      "step": 1175
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.7137369513511658,
+      "learning_rate": 8.758328679764776e-05,
+      "loss": 2.9028,
+      "step": 1180
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.7238683104515076,
+      "learning_rate": 8.747312163052284e-05,
+      "loss": 2.8834,
+      "step": 1185
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.6741464734077454,
+      "learning_rate": 8.736253985020761e-05,
+      "loss": 2.8644,
+      "step": 1190
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.657223641872406,
+      "learning_rate": 8.725154268611203e-05,
+      "loss": 2.8719,
+      "step": 1195
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.6820774674415588,
+      "learning_rate": 8.714013137226411e-05,
+      "loss": 3.0099,
+      "step": 1200
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.6592907905578613,
+      "learning_rate": 8.702830714729628e-05,
+      "loss": 2.9659,
+      "step": 1205
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.6504175066947937,
+      "learning_rate": 8.691607125443153e-05,
+      "loss": 2.9834,
+      "step": 1210
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.655095636844635,
+      "learning_rate": 8.680342494146967e-05,
+      "loss": 2.9123,
+      "step": 1215
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.64875328540802,
+      "learning_rate": 8.66903694607734e-05,
+      "loss": 2.891,
+      "step": 1220
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.6668553352355957,
+      "learning_rate": 8.65769060692544e-05,
+      "loss": 3.0055,
+      "step": 1225
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.6586368680000305,
+      "learning_rate": 8.646303602835936e-05,
+      "loss": 2.9121,
+      "step": 1230
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.649715006351471,
+      "learning_rate": 8.634876060405597e-05,
+      "loss": 3.0043,
+      "step": 1235
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.6875714659690857,
+      "learning_rate": 8.623408106681884e-05,
+      "loss": 2.8886,
+      "step": 1240
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.6406840682029724,
+      "learning_rate": 8.611899869161535e-05,
+      "loss": 2.9263,
+      "step": 1245
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.6714261770248413,
+      "learning_rate": 8.600351475789147e-05,
+      "loss": 2.9036,
+      "step": 1250
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.6701438426971436,
+      "learning_rate": 8.588763054955764e-05,
+      "loss": 2.9502,
+      "step": 1255
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.6833099722862244,
+      "learning_rate": 8.57713473549743e-05,
+      "loss": 2.8397,
+      "step": 1260
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.652505099773407,
+      "learning_rate": 8.565466646693778e-05,
+      "loss": 2.9765,
+      "step": 1265
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.6483748555183411,
+      "learning_rate": 8.553758918266578e-05,
+      "loss": 2.8748,
+      "step": 1270
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.6926413774490356,
+      "learning_rate": 8.5420116803783e-05,
+      "loss": 2.952,
+      "step": 1275
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.661097526550293,
+      "learning_rate": 8.530225063630668e-05,
+      "loss": 2.9308,
+      "step": 1280
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.7419840693473816,
+      "learning_rate": 8.518399199063205e-05,
+      "loss": 2.9051,
+      "step": 1285
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.707297682762146,
+      "learning_rate": 8.50653421815178e-05,
+      "loss": 2.9121,
+      "step": 1290
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.66552734375,
+      "learning_rate": 8.494630252807138e-05,
+      "loss": 2.88,
+      "step": 1295
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 3.651979923248291,
+      "learning_rate": 8.482687435373449e-05,
+      "loss": 2.9026,
+      "step": 1300
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.6851710081100464,
+      "learning_rate": 8.470705898626817e-05,
+      "loss": 2.8257,
+      "step": 1305
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.6279444694519043,
+      "learning_rate": 8.458685775773822e-05,
+      "loss": 2.8641,
+      "step": 1310
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.6904920339584351,
+      "learning_rate": 8.446627200450025e-05,
+      "loss": 2.9092,
+      "step": 1315
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.6325361728668213,
+      "learning_rate": 8.434530306718493e-05,
+      "loss": 2.8976,
+      "step": 1320
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.6655163168907166,
+      "learning_rate": 8.4223952290683e-05,
+      "loss": 2.8918,
+      "step": 1325
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.6457725763320923,
+      "learning_rate": 8.41022210241304e-05,
+      "loss": 2.8592,
+      "step": 1330
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.6553127765655518,
+      "learning_rate": 8.398011062089316e-05,
+      "loss": 2.9329,
+      "step": 1335
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.6595431566238403,
+      "learning_rate": 8.385762243855249e-05,
+      "loss": 2.8627,
+      "step": 1340
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.6419198513031006,
+      "learning_rate": 8.373475783888958e-05,
+      "loss": 2.8726,
+      "step": 1345
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.6688870787620544,
+      "learning_rate": 8.36115181878705e-05,
+      "loss": 2.8129,
+      "step": 1350
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.6275585889816284,
+      "learning_rate": 8.348790485563101e-05,
+      "loss": 2.8333,
+      "step": 1355
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.658042848110199,
+      "learning_rate": 8.336391921646134e-05,
+      "loss": 2.7352,
+      "step": 1360
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.6524467468261719,
+      "learning_rate": 8.323956264879089e-05,
+      "loss": 2.9171,
+      "step": 1365
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.6397958993911743,
+      "learning_rate": 8.311483653517294e-05,
+      "loss": 2.8176,
+      "step": 1370
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.6147693395614624,
+      "learning_rate": 8.298974226226919e-05,
+      "loss": 2.8676,
+      "step": 1375
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.6508784294128418,
+      "learning_rate": 8.28642812208345e-05,
+      "loss": 2.842,
+      "step": 1380
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.6087005734443665,
+      "learning_rate": 8.273845480570123e-05,
+      "loss": 2.8126,
+      "step": 1385
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.7282595634460449,
+      "learning_rate": 8.26122644157639e-05,
+      "loss": 2.9029,
+      "step": 1390
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.619777500629425,
+      "learning_rate": 8.248571145396362e-05,
+      "loss": 2.8496,
+      "step": 1395
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.6248059868812561,
+      "learning_rate": 8.235879732727236e-05,
+      "loss": 2.8447,
+      "step": 1400
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.6536071300506592,
+      "learning_rate": 8.223152344667745e-05,
+      "loss": 2.831,
+      "step": 1405
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.6239855289459229,
+      "learning_rate": 8.21038912271658e-05,
+      "loss": 2.8611,
+      "step": 1410
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.6541547775268555,
+      "learning_rate": 8.197590208770824e-05,
+      "loss": 2.8558,
+      "step": 1415
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6467083692550659,
+      "learning_rate": 8.184755745124371e-05,
+      "loss": 2.8409,
+      "step": 1420
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6722072958946228,
+      "learning_rate": 8.171885874466342e-05,
+      "loss": 2.8381,
+      "step": 1425
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6703723669052124,
+      "learning_rate": 8.158980739879507e-05,
+      "loss": 2.8365,
+      "step": 1430
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6237878799438477,
+      "learning_rate": 8.146040484838677e-05,
+      "loss": 2.8091,
+      "step": 1435
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6343963146209717,
+      "learning_rate": 8.133065253209132e-05,
+      "loss": 2.8389,
+      "step": 1440
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6207173466682434,
+      "learning_rate": 8.120055189245e-05,
+      "loss": 2.8748,
+      "step": 1445
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6682604551315308,
+      "learning_rate": 8.10701043758767e-05,
+      "loss": 2.8611,
+      "step": 1450
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6844489574432373,
+      "learning_rate": 8.093931143264174e-05,
+      "loss": 2.8423,
+      "step": 1455
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6315253376960754,
+      "learning_rate": 8.080817451685576e-05,
+      "loss": 2.7389,
+      "step": 1460
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6357865929603577,
+      "learning_rate": 8.067669508645356e-05,
+      "loss": 2.8418,
+      "step": 1465
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.6491623520851135,
+      "learning_rate": 8.054487460317797e-05,
+      "loss": 2.8975,
+      "step": 1470
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.6408995985984802,
+      "learning_rate": 8.041271453256345e-05,
+      "loss": 2.782,
+      "step": 1475
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.6269375085830688,
+      "learning_rate": 8.02802163439199e-05,
+      "loss": 2.8292,
+      "step": 1480
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.7113586068153381,
+      "learning_rate": 8.01473815103163e-05,
+      "loss": 2.8488,
+      "step": 1485
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.6262251138687134,
+      "learning_rate": 8.001421150856434e-05,
+      "loss": 2.8008,
+      "step": 1490
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.6527858972549438,
+      "learning_rate": 7.988070781920197e-05,
+      "loss": 2.8618,
+      "step": 1495
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.6287201642990112,
+      "learning_rate": 7.9746871926477e-05,
+      "loss": 2.8064,
+      "step": 1500
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.6469153165817261,
+      "learning_rate": 7.961270531833052e-05,
+      "loss": 2.8638,
+      "step": 1505
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.6609588265419006,
+      "learning_rate": 7.947820948638045e-05,
+      "loss": 2.8039,
+      "step": 1510
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.7042279243469238,
+      "learning_rate": 7.934338592590486e-05,
+      "loss": 2.9047,
+      "step": 1515
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.6446994543075562,
+      "learning_rate": 7.92082361358254e-05,
+      "loss": 2.8422,
+      "step": 1520
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.9173874258995056,
+      "learning_rate": 7.907276161869065e-05,
+      "loss": 2.8219,
+      "step": 1525
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.6591891646385193,
+      "learning_rate": 7.893696388065936e-05,
+      "loss": 2.7361,
+      "step": 1530
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.6696944832801819,
+      "learning_rate": 7.88008444314838e-05,
+      "loss": 2.8942,
+      "step": 1535
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.6576460003852844,
+      "learning_rate": 7.866440478449283e-05,
+      "loss": 2.8264,
+      "step": 1540
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.6158410906791687,
+      "learning_rate": 7.852764645657522e-05,
+      "loss": 2.7558,
+      "step": 1545
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.6890105605125427,
+      "learning_rate": 7.839057096816271e-05,
+      "loss": 2.8061,
+      "step": 1550
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.1392295360565186,
+      "learning_rate": 7.82531798432131e-05,
+      "loss": 2.8412,
+      "step": 1555
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.6717019081115723,
+      "learning_rate": 7.811547460919333e-05,
+      "loss": 2.781,
+      "step": 1560
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.6458261609077454,
+      "learning_rate": 7.797745679706254e-05,
+      "loss": 2.8394,
+      "step": 1565
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.6580215096473694,
+      "learning_rate": 7.783912794125496e-05,
+      "loss": 2.8779,
+      "step": 1570
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.6356989741325378,
+      "learning_rate": 7.770048957966291e-05,
+      "loss": 2.828,
+      "step": 1575
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.6654439568519592,
+      "learning_rate": 7.756154325361967e-05,
+      "loss": 2.8222,
+      "step": 1580
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.6044449806213379,
+      "learning_rate": 7.74222905078824e-05,
+      "loss": 2.8407,
+      "step": 1585
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.6402572393417358,
+      "learning_rate": 7.728273289061489e-05,
+      "loss": 2.7892,
+      "step": 1590
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.9658334255218506,
+      "learning_rate": 7.714287195337044e-05,
+      "loss": 2.8384,
+      "step": 1595
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.6388287544250488,
+      "learning_rate": 7.700270925107448e-05,
+      "loss": 2.8379,
+      "step": 1600
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.6642169952392578,
+      "learning_rate": 7.686224634200742e-05,
+      "loss": 2.8276,
+      "step": 1605
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.6531822681427002,
+      "learning_rate": 7.672148478778722e-05,
+      "loss": 2.8161,
+      "step": 1610
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.6064309477806091,
+      "learning_rate": 7.658042615335212e-05,
+      "loss": 2.736,
+      "step": 1615
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.6005924344062805,
+      "learning_rate": 7.643907200694318e-05,
+      "loss": 2.8041,
+      "step": 1620
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.596092939376831,
+      "learning_rate": 7.629742392008684e-05,
+      "loss": 2.7951,
+      "step": 1625
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.6009310483932495,
+      "learning_rate": 7.615548346757749e-05,
+      "loss": 2.8565,
+      "step": 1630
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.6022703051567078,
+      "learning_rate": 7.60132522274599e-05,
+      "loss": 2.8197,
+      "step": 1635
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.5911956429481506,
+      "learning_rate": 7.587073178101178e-05,
+      "loss": 2.7829,
+      "step": 1640
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.6114310026168823,
+      "learning_rate": 7.572792371272609e-05,
+      "loss": 2.7378,
+      "step": 1645
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.6166604161262512,
+      "learning_rate": 7.55848296102935e-05,
+      "loss": 2.8319,
+      "step": 1650
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.596598744392395,
+      "learning_rate": 7.544145106458465e-05,
+      "loss": 2.7407,
+      "step": 1655
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.6231933236122131,
+      "learning_rate": 7.529778966963259e-05,
+      "loss": 2.7703,
+      "step": 1660
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.5864799618721008,
+      "learning_rate": 7.515384702261496e-05,
+      "loss": 2.792,
+      "step": 1665
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.629953145980835,
+      "learning_rate": 7.500962472383627e-05,
+      "loss": 2.7735,
+      "step": 1670
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.6393805742263794,
+      "learning_rate": 7.486512437671011e-05,
+      "loss": 2.8102,
+      "step": 1675
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.6531060934066772,
+      "learning_rate": 7.472034758774128e-05,
+      "loss": 2.7946,
+      "step": 1680
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.6248769760131836,
+      "learning_rate": 7.457529596650797e-05,
+      "loss": 2.815,
+      "step": 1685
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.6464139223098755,
+      "learning_rate": 7.442997112564392e-05,
+      "loss": 2.7105,
+      "step": 1690
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.6017792224884033,
+      "learning_rate": 7.428437468082037e-05,
+      "loss": 2.7475,
+      "step": 1695
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.6099601984024048,
+      "learning_rate": 7.413850825072817e-05,
+      "loss": 2.7917,
+      "step": 1700
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.5857229828834534,
+      "learning_rate": 7.39923734570598e-05,
+      "loss": 2.7858,
+      "step": 1705
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 2.5941388607025146,
+      "learning_rate": 7.384597192449126e-05,
+      "loss": 2.8045,
+      "step": 1710
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.6371538639068604,
+      "learning_rate": 7.369930528066412e-05,
+      "loss": 2.8803,
+      "step": 1715
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.6334919333457947,
+      "learning_rate": 7.355237515616732e-05,
+      "loss": 2.8386,
+      "step": 1720
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.6036331057548523,
+      "learning_rate": 7.340518318451914e-05,
+      "loss": 2.7614,
+      "step": 1725
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.5964059233665466,
+      "learning_rate": 7.325773100214893e-05,
+      "loss": 2.7165,
+      "step": 1730
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.6223915815353394,
+      "learning_rate": 7.311002024837899e-05,
+      "loss": 2.7807,
+      "step": 1735
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.5919489860534668,
+      "learning_rate": 7.296205256540633e-05,
+      "loss": 2.8301,
+      "step": 1740
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.6090957522392273,
+      "learning_rate": 7.281382959828443e-05,
+      "loss": 2.8146,
+      "step": 1745
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.5976056456565857,
+      "learning_rate": 7.26653529949049e-05,
+      "loss": 2.7906,
+      "step": 1750
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.6359770894050598,
+      "learning_rate": 7.25166244059792e-05,
+      "loss": 2.7609,
+      "step": 1755
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.6062914133071899,
+      "learning_rate": 7.236764548502029e-05,
+      "loss": 2.7974,
+      "step": 1760
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.6405194401741028,
+      "learning_rate": 7.221841788832421e-05,
+      "loss": 2.8288,
+      "step": 1765
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.6273607015609741,
+      "learning_rate": 7.206894327495173e-05,
+      "loss": 2.8076,
+      "step": 1770
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.5930814743041992,
+      "learning_rate": 7.191922330670982e-05,
+      "loss": 2.8182,
+      "step": 1775
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.5818787217140198,
+      "learning_rate": 7.176925964813326e-05,
+      "loss": 2.709,
+      "step": 1780
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.6052417755126953,
+      "learning_rate": 7.161905396646607e-05,
+      "loss": 2.8197,
+      "step": 1785
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.6246456503868103,
+      "learning_rate": 7.146860793164299e-05,
+      "loss": 2.7605,
+      "step": 1790
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.6177834868431091,
+      "learning_rate": 7.131792321627098e-05,
+      "loss": 2.8023,
+      "step": 1795
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.6127104163169861,
+      "learning_rate": 7.116700149561048e-05,
+      "loss": 2.8207,
+      "step": 1800
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.6975888013839722,
+      "learning_rate": 7.101584444755696e-05,
+      "loss": 2.7521,
+      "step": 1805
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.6271196603775024,
+      "learning_rate": 7.086445375262212e-05,
+      "loss": 2.771,
+      "step": 1810
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.6058980822563171,
+      "learning_rate": 7.071283109391528e-05,
+      "loss": 2.7284,
+      "step": 1815
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.6291877031326294,
+      "learning_rate": 7.056097815712466e-05,
+      "loss": 2.8608,
+      "step": 1820
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.6075469851493835,
+      "learning_rate": 7.040889663049862e-05,
+      "loss": 2.7491,
+      "step": 1825
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.6630153656005859,
+      "learning_rate": 7.025658820482693e-05,
+      "loss": 2.8098,
+      "step": 1830
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.6005200743675232,
+      "learning_rate": 7.010405457342192e-05,
+      "loss": 2.79,
+      "step": 1835
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.6074357032775879,
+      "learning_rate": 6.995129743209967e-05,
+      "loss": 2.7717,
+      "step": 1840
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.6157153248786926,
+      "learning_rate": 6.97983184791612e-05,
+      "loss": 2.7526,
+      "step": 1845
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.5990545749664307,
+      "learning_rate": 6.964511941537355e-05,
+      "loss": 2.7891,
+      "step": 1850
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.5907087326049805,
+      "learning_rate": 6.949170194395083e-05,
+      "loss": 2.728,
+      "step": 1855
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.5958890318870544,
+      "learning_rate": 6.933806777053536e-05,
+      "loss": 2.7638,
+      "step": 1860
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.6144511103630066,
+      "learning_rate": 6.918421860317872e-05,
+      "loss": 2.785,
+      "step": 1865
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.595562756061554,
+      "learning_rate": 6.903015615232263e-05,
+      "loss": 2.7708,
+      "step": 1870
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.6055405139923096,
+      "learning_rate": 6.887588213078012e-05,
+      "loss": 2.7603,
+      "step": 1875
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.6407721638679504,
+      "learning_rate": 6.87213982537163e-05,
+      "loss": 2.7656,
+      "step": 1880
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.5973005294799805,
+      "learning_rate": 6.856670623862943e-05,
+      "loss": 2.74,
+      "step": 1885
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.6174443960189819,
+      "learning_rate": 6.841180780533179e-05,
+      "loss": 2.737,
+      "step": 1890
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.6259763836860657,
+      "learning_rate": 6.82567046759305e-05,
+      "loss": 2.7718,
+      "step": 1895
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.619278609752655,
+      "learning_rate": 6.810139857480844e-05,
+      "loss": 2.7495,
+      "step": 1900
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.6113199591636658,
+      "learning_rate": 6.794589122860509e-05,
+      "loss": 2.7093,
+      "step": 1905
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.6377850770950317,
+      "learning_rate": 6.779018436619725e-05,
+      "loss": 2.7287,
+      "step": 1910
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.5981298089027405,
+      "learning_rate": 6.763427971867992e-05,
+      "loss": 2.7516,
+      "step": 1915
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.6440865993499756,
+      "learning_rate": 6.747817901934699e-05,
+      "loss": 2.7265,
+      "step": 1920
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.6392199993133545,
+      "learning_rate": 6.732188400367197e-05,
+      "loss": 2.7461,
+      "step": 1925
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.5952336192131042,
+      "learning_rate": 6.716539640928871e-05,
+      "loss": 2.7049,
+      "step": 1930
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.5947147011756897,
+      "learning_rate": 6.70087179759721e-05,
+      "loss": 2.6609,
+      "step": 1935
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.5876834392547607,
+      "learning_rate": 6.685185044561874e-05,
+      "loss": 2.7298,
+      "step": 1940
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.6168666481971741,
+      "learning_rate": 6.669479556222747e-05,
+      "loss": 2.7995,
+      "step": 1945
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.5867111682891846,
+      "learning_rate": 6.653755507188013e-05,
+      "loss": 2.7873,
+      "step": 1950
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.6168385148048401,
+      "learning_rate": 6.638013072272205e-05,
+      "loss": 2.7616,
+      "step": 1955
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.6352885365486145,
+      "learning_rate": 6.622252426494259e-05,
+      "loss": 2.7235,
+      "step": 1960
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.6029419302940369,
+      "learning_rate": 6.606473745075581e-05,
+      "loss": 2.755,
+      "step": 1965
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.6367367506027222,
+      "learning_rate": 6.590677203438084e-05,
+      "loss": 2.7381,
+      "step": 1970
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.5971164107322693,
+      "learning_rate": 6.574862977202252e-05,
+      "loss": 2.7641,
+      "step": 1975
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.6724832653999329,
+      "learning_rate": 6.559031242185174e-05,
+      "loss": 2.7543,
+      "step": 1980
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.6365452408790588,
+      "learning_rate": 6.543182174398597e-05,
+      "loss": 2.7156,
+      "step": 1985
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.5962063670158386,
+      "learning_rate": 6.52731595004697e-05,
+      "loss": 2.6921,
+      "step": 1990
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.6351775527000427,
+      "learning_rate": 6.51143274552548e-05,
+      "loss": 2.6404,
+      "step": 1995
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.6487362384796143,
+      "learning_rate": 6.495532737418098e-05,
+      "loss": 2.6959,
+      "step": 2000
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.6513057351112366,
+      "learning_rate": 6.479616102495605e-05,
+      "loss": 2.7508,
+      "step": 2005
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.5903010368347168,
+      "learning_rate": 6.463683017713638e-05,
+      "loss": 2.7474,
+      "step": 2010
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.6075739860534668,
+      "learning_rate": 6.447733660210715e-05,
+      "loss": 2.7462,
+      "step": 2015
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.6202053427696228,
+      "learning_rate": 6.431768207306272e-05,
+      "loss": 2.8086,
+      "step": 2020
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.6007124185562134,
+      "learning_rate": 6.415786836498684e-05,
+      "loss": 2.7325,
+      "step": 2025
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.6256109476089478,
+      "learning_rate": 6.399789725463298e-05,
+      "loss": 2.7049,
+      "step": 2030
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.6217005252838135,
+      "learning_rate": 6.383777052050458e-05,
+      "loss": 2.7153,
+      "step": 2035
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.6184681057929993,
+      "learning_rate": 6.367748994283518e-05,
+      "loss": 2.694,
+      "step": 2040
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.5928269624710083,
+      "learning_rate": 6.351705730356877e-05,
+      "loss": 2.7265,
+      "step": 2045
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.6197706460952759,
+      "learning_rate": 6.335647438633987e-05,
+      "loss": 2.7359,
+      "step": 2050
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.6162838339805603,
+      "learning_rate": 6.319574297645374e-05,
+      "loss": 2.7239,
+      "step": 2055
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.5909050107002258,
+      "learning_rate": 6.303486486086654e-05,
+      "loss": 2.751,
+      "step": 2060
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.5814822912216187,
+      "learning_rate": 6.287384182816546e-05,
+      "loss": 2.7583,
+      "step": 2065
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.6004829406738281,
+      "learning_rate": 6.271267566854883e-05,
+      "loss": 2.7153,
+      "step": 2070
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.5877543687820435,
+      "learning_rate": 6.255136817380618e-05,
+      "loss": 2.6557,
+      "step": 2075
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.5787215828895569,
+      "learning_rate": 6.23899211372984e-05,
+      "loss": 2.7686,
+      "step": 2080
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.5978643298149109,
+      "learning_rate": 6.222833635393772e-05,
+      "loss": 2.7421,
+      "step": 2085
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.5708004832267761,
+      "learning_rate": 6.206661562016782e-05,
+      "loss": 2.7306,
+      "step": 2090
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.6145750284194946,
+      "learning_rate": 6.190476073394382e-05,
+      "loss": 2.7589,
+      "step": 2095
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.6016147136688232,
+      "learning_rate": 6.17427734947123e-05,
+      "loss": 2.6672,
+      "step": 2100
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.5984389185905457,
+      "learning_rate": 6.158065570339127e-05,
+      "loss": 2.7528,
+      "step": 2105
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.609862208366394,
+      "learning_rate": 6.141840916235021e-05,
+      "loss": 2.8206,
+      "step": 2110
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.5888879299163818,
+      "learning_rate": 6.125603567539001e-05,
+      "loss": 2.6867,
+      "step": 2115
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.5528971552848816,
+      "learning_rate": 6.109353704772284e-05,
+      "loss": 2.711,
+      "step": 2120
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.580680787563324,
+      "learning_rate": 6.0930915085952164e-05,
+      "loss": 2.6565,
+      "step": 2125
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.5929208993911743,
+      "learning_rate": 6.076817159805267e-05,
+      "loss": 2.681,
+      "step": 2130
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.5872501134872437,
+      "learning_rate": 6.06053083933501e-05,
+      "loss": 2.7475,
+      "step": 2135
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.5946284532546997,
+      "learning_rate": 6.044232728250116e-05,
+      "loss": 2.6727,
+      "step": 2140
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.5818427205085754,
+      "learning_rate": 6.027923007747339e-05,
+      "loss": 2.639,
+      "step": 2145
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.6036926507949829,
+      "learning_rate": 6.011601859152506e-05,
+      "loss": 2.7188,
+      "step": 2150
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.5915615558624268,
+      "learning_rate": 5.995269463918495e-05,
+      "loss": 2.6782,
+      "step": 2155
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.589364767074585,
+      "learning_rate": 5.97892600362322e-05,
+      "loss": 2.6975,
+      "step": 2160
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.570540189743042,
+      "learning_rate": 5.962571659967614e-05,
+      "loss": 2.7128,
+      "step": 2165
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.6136625409126282,
+      "learning_rate": 5.946206614773606e-05,
+      "loss": 2.6649,
+      "step": 2170
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.5848395228385925,
+      "learning_rate": 5.929831049982103e-05,
+      "loss": 2.7049,
+      "step": 2175
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.589984118938446,
+      "learning_rate": 5.9134451476509633e-05,
+      "loss": 2.7139,
+      "step": 2180
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.59041827917099,
+      "learning_rate": 5.897049089952974e-05,
+      "loss": 2.7045,
+      "step": 2185
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.5757389068603516,
+      "learning_rate": 5.880643059173826e-05,
+      "loss": 2.6226,
+      "step": 2190
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.592380702495575,
+      "learning_rate": 5.864227237710093e-05,
+      "loss": 2.7305,
+      "step": 2195
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.6019942760467529,
+      "learning_rate": 5.847801808067189e-05,
+      "loss": 2.6937,
+      "step": 2200
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.6186248660087585,
+      "learning_rate": 5.831366952857357e-05,
+      "loss": 2.7107,
+      "step": 2205
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.5778764486312866,
+      "learning_rate": 5.814922854797622e-05,
+      "loss": 2.7006,
+      "step": 2210
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.5956502556800842,
+      "learning_rate": 5.798469696707775e-05,
+      "loss": 2.7031,
+      "step": 2215
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.6054814457893372,
+      "learning_rate": 5.782007661508331e-05,
+      "loss": 2.7445,
+      "step": 2220
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.5853110551834106,
+      "learning_rate": 5.765536932218495e-05,
+      "loss": 2.6657,
+      "step": 2225
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.6011916399002075,
+      "learning_rate": 5.7490576919541315e-05,
+      "loss": 2.6646,
+      "step": 2230
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.5917201638221741,
+      "learning_rate": 5.732570123925729e-05,
+      "loss": 2.6769,
+      "step": 2235
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.6263756155967712,
+      "learning_rate": 5.7160744114363593e-05,
+      "loss": 2.6539,
+      "step": 2240
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.5738804340362549,
+      "learning_rate": 5.699570737879641e-05,
+      "loss": 2.6446,
+      "step": 2245
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.5955730080604553,
+      "learning_rate": 5.683059286737702e-05,
+      "loss": 2.7087,
+      "step": 2250
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.5945442914962769,
+      "learning_rate": 5.666540241579139e-05,
+      "loss": 2.6942,
+      "step": 2255
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.5907345414161682,
+      "learning_rate": 5.6500137860569766e-05,
+      "loss": 2.7556,
+      "step": 2260
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.5916737914085388,
+      "learning_rate": 5.633480103906624e-05,
+      "loss": 2.6653,
+      "step": 2265
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.5896185636520386,
+      "learning_rate": 5.616939378943834e-05,
+      "loss": 2.6891,
+      "step": 2270
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.6207943558692932,
+      "learning_rate": 5.6003917950626595e-05,
+      "loss": 2.7149,
+      "step": 2275
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.6069183945655823,
+      "learning_rate": 5.583837536233407e-05,
+      "loss": 2.7647,
+      "step": 2280
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.5955156087875366,
+      "learning_rate": 5.567276786500596e-05,
+      "loss": 2.6789,
+      "step": 2285
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.5711756944656372,
+      "learning_rate": 5.5507097299809054e-05,
+      "loss": 2.7099,
+      "step": 2290
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.5811170339584351,
+      "learning_rate": 5.534136550861133e-05,
+      "loss": 2.7018,
+      "step": 2295
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.6151323318481445,
+      "learning_rate": 5.5175574333961465e-05,
+      "loss": 2.6619,
+      "step": 2300
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.6350491642951965,
+      "learning_rate": 5.500972561906832e-05,
+      "loss": 2.6722,
+      "step": 2305
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.6088038682937622,
+      "learning_rate": 5.484382120778048e-05,
+      "loss": 2.6838,
+      "step": 2310
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.5795965790748596,
+      "learning_rate": 5.467786294456575e-05,
+      "loss": 2.7021,
+      "step": 2315
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.6044985055923462,
+      "learning_rate": 5.451185267449061e-05,
+      "loss": 2.648,
+      "step": 2320
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.5968245267868042,
+      "learning_rate": 5.43457922431998e-05,
+      "loss": 2.7042,
+      "step": 2325
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.5967344045639038,
+      "learning_rate": 5.417968349689566e-05,
+      "loss": 2.7427,
+      "step": 2330
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.6051455736160278,
+      "learning_rate": 5.401352828231772e-05,
+      "loss": 2.6792,
+      "step": 2335
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.6021824479103088,
+      "learning_rate": 5.384732844672211e-05,
+      "loss": 2.7174,
+      "step": 2340
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.5897824764251709,
+      "learning_rate": 5.368108583786107e-05,
+      "loss": 2.7425,
+      "step": 2345
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.5941648483276367,
+      "learning_rate": 5.3514802303962344e-05,
+      "loss": 2.7718,
+      "step": 2350
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.6310714483261108,
+      "learning_rate": 5.334847969370868e-05,
+      "loss": 2.7228,
+      "step": 2355
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.5781346559524536,
+      "learning_rate": 5.3182119856217284e-05,
+      "loss": 2.6547,
+      "step": 2360
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.5885051488876343,
+      "learning_rate": 5.3015724641019214e-05,
+      "loss": 2.6163,
+      "step": 2365
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.5877886414527893,
+      "learning_rate": 5.284929589803884e-05,
+      "loss": 2.6974,
+      "step": 2370
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.5883300304412842,
+      "learning_rate": 5.2682835477573336e-05,
+      "loss": 2.7209,
+      "step": 2375
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.6138569712638855,
+      "learning_rate": 5.2516345230271965e-05,
+      "loss": 2.6766,
+      "step": 2380
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.5742478966712952,
+      "learning_rate": 5.234982700711569e-05,
+      "loss": 2.6928,
+      "step": 2385
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.5935817360877991,
+      "learning_rate": 5.218328265939643e-05,
+      "loss": 2.6688,
+      "step": 2390
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.579456090927124,
+      "learning_rate": 5.201671403869657e-05,
+      "loss": 2.7296,
+      "step": 2395
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.6063280701637268,
+      "learning_rate": 5.1850122996868366e-05,
+      "loss": 2.6807,
+      "step": 2400
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.5990563035011292,
+      "learning_rate": 5.168351138601334e-05,
+      "loss": 2.6674,
+      "step": 2405
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.5913801193237305,
+      "learning_rate": 5.1516881058461675e-05,
+      "loss": 2.6167,
+      "step": 2410
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.5955457091331482,
+      "learning_rate": 5.135023386675166e-05,
+      "loss": 2.6794,
+      "step": 2415
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.5914067625999451,
+      "learning_rate": 5.118357166360906e-05,
+      "loss": 2.6578,
+      "step": 2420
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.6297724843025208,
+      "learning_rate": 5.101689630192655e-05,
+      "loss": 2.6634,
+      "step": 2425
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.6354207992553711,
+      "learning_rate": 5.085020963474307e-05,
+      "loss": 2.6888,
+      "step": 2430
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.6159840226173401,
+      "learning_rate": 5.068351351522329e-05,
+      "loss": 2.6681,
+      "step": 2435
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.5876420736312866,
+      "learning_rate": 5.0516809796636935e-05,
+      "loss": 2.6068,
+      "step": 2440
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.5940358638763428,
+      "learning_rate": 5.035010033233821e-05,
+      "loss": 2.5744,
+      "step": 2445
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.5878875851631165,
+      "learning_rate": 5.018338697574523e-05,
+      "loss": 2.6835,
+      "step": 2450
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.5701265335083008,
+      "learning_rate": 5.0016671580319354e-05,
+      "loss": 2.6463,
+      "step": 2455
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.6023485660552979,
+      "learning_rate": 4.984995599954461e-05,
+      "loss": 2.6397,
+      "step": 2460
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.6002631187438965,
+      "learning_rate": 4.968324208690712e-05,
+      "loss": 2.5806,
+      "step": 2465
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.5934431552886963,
+      "learning_rate": 4.951653169587441e-05,
+      "loss": 2.7039,
+      "step": 2470
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.5772780179977417,
+      "learning_rate": 4.93498266798749e-05,
+      "loss": 2.5367,
+      "step": 2475
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 2.262930393218994,
+      "learning_rate": 4.918312889227722e-05,
+      "loss": 2.6517,
+      "step": 2480
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.6189587712287903,
+      "learning_rate": 4.901644018636966e-05,
+      "loss": 2.6813,
+      "step": 2485
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.5724838376045227,
+      "learning_rate": 4.8849762415339526e-05,
+      "loss": 2.6501,
+      "step": 2490
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.5820484161376953,
+      "learning_rate": 4.868309743225256e-05,
+      "loss": 2.7631,
+      "step": 2495
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.5822151303291321,
+      "learning_rate": 4.851644709003233e-05,
+      "loss": 2.6944,
+      "step": 2500
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.5823991894721985,
+      "learning_rate": 4.834981324143964e-05,
+      "loss": 2.6122,
+      "step": 2505
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.5901163220405579,
+      "learning_rate": 4.818319773905191e-05,
+      "loss": 2.6254,
+      "step": 2510
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.5650957822799683,
+      "learning_rate": 4.801660243524261e-05,
+      "loss": 2.6358,
+      "step": 2515
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.5851700901985168,
+      "learning_rate": 4.7850029182160626e-05,
+      "loss": 2.6393,
+      "step": 2520
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.581257164478302,
+      "learning_rate": 4.768347983170973e-05,
+      "loss": 2.5892,
+      "step": 2525
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.5895279049873352,
+      "learning_rate": 4.7516956235527884e-05,
+      "loss": 2.6424,
+      "step": 2530
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.7180891036987305,
+      "learning_rate": 4.735046024496682e-05,
+      "loss": 2.6909,
+      "step": 2535
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.5922880172729492,
+      "learning_rate": 4.7183993711071286e-05,
+      "loss": 2.71,
+      "step": 2540
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.5666274428367615,
+      "learning_rate": 4.7017558484558554e-05,
+      "loss": 2.7223,
+      "step": 2545
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.561943531036377,
+      "learning_rate": 4.6851156415797844e-05,
+      "loss": 2.6269,
+      "step": 2550
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.5793047547340393,
+      "learning_rate": 4.6684789354789746e-05,
+      "loss": 2.6433,
+      "step": 2555
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.6017892956733704,
+      "learning_rate": 4.651845915114563e-05,
+      "loss": 2.6989,
+      "step": 2560
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.575771689414978,
+      "learning_rate": 4.6352167654067095e-05,
+      "loss": 2.6372,
+      "step": 2565
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.579129695892334,
+      "learning_rate": 4.618591671232544e-05,
+      "loss": 2.6722,
+      "step": 2570
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.5836328268051147,
+      "learning_rate": 4.601970817424106e-05,
+      "loss": 2.6702,
+      "step": 2575
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.598733127117157,
+      "learning_rate": 4.585354388766292e-05,
+      "loss": 2.697,
+      "step": 2580
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.5923112630844116,
+      "learning_rate": 4.568742569994802e-05,
+      "loss": 2.6181,
+      "step": 2585
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.5779056549072266,
+      "learning_rate": 4.552135545794086e-05,
+      "loss": 2.6333,
+      "step": 2590
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.5684216618537903,
+      "learning_rate": 4.535533500795288e-05,
+      "loss": 2.6568,
+      "step": 2595
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.6125906109809875,
+      "learning_rate": 4.5189366195741953e-05,
+      "loss": 2.7044,
+      "step": 2600
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.5921022295951843,
+      "learning_rate": 4.502345086649186e-05,
+      "loss": 2.565,
+      "step": 2605
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.6385989785194397,
+      "learning_rate": 4.485759086479179e-05,
+      "loss": 2.6301,
+      "step": 2610
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.5802872180938721,
+      "learning_rate": 4.469178803461579e-05,
+      "loss": 2.5987,
+      "step": 2615
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.5786648392677307,
+      "learning_rate": 4.4526044219302326e-05,
+      "loss": 2.6468,
+      "step": 2620
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.5831460356712341,
+      "learning_rate": 4.4360361261533745e-05,
+      "loss": 2.6621,
+      "step": 2625
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.5802498459815979,
+      "learning_rate": 4.419474100331579e-05,
+      "loss": 2.6547,
+      "step": 2630
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.5955533981323242,
+      "learning_rate": 4.402918528595715e-05,
+      "loss": 2.6076,
+      "step": 2635
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.5763300061225891,
+      "learning_rate": 4.386369595004896e-05,
+      "loss": 2.6596,
+      "step": 2640
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.6035252213478088,
+      "learning_rate": 4.3698274835444354e-05,
+      "loss": 2.6467,
+      "step": 2645
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.5766133069992065,
+      "learning_rate": 4.3532923781238e-05,
+      "loss": 2.6246,
+      "step": 2650
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.5641186833381653,
+      "learning_rate": 4.336764462574566e-05,
+      "loss": 2.6427,
+      "step": 2655
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.5621420741081238,
+      "learning_rate": 4.320243920648376e-05,
+      "loss": 2.7131,
+      "step": 2660
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.587009847164154,
+      "learning_rate": 4.303730936014894e-05,
+      "loss": 2.5924,
+      "step": 2665
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.5865349173545837,
+      "learning_rate": 4.287225692259765e-05,
+      "loss": 2.6606,
+      "step": 2670
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.5877777338027954,
+      "learning_rate": 4.270728372882575e-05,
+      "loss": 2.5986,
+      "step": 2675
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.6098527908325195,
+      "learning_rate": 4.254239161294804e-05,
+      "loss": 2.7315,
+      "step": 2680
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.5707640051841736,
+      "learning_rate": 4.237758240817802e-05,
+      "loss": 2.6216,
+      "step": 2685
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.7130870819091797,
+      "learning_rate": 4.2212857946807336e-05,
+      "loss": 2.6666,
+      "step": 2690
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.5846860408782959,
+      "learning_rate": 4.2048220060185516e-05,
+      "loss": 2.6045,
+      "step": 2695
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.6190792322158813,
+      "learning_rate": 4.188367057869957e-05,
+      "loss": 2.6988,
+      "step": 2700
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.5834519863128662,
+      "learning_rate": 4.171921133175365e-05,
+      "loss": 2.6965,
+      "step": 2705
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.5982712507247925,
+      "learning_rate": 4.155484414774872e-05,
+      "loss": 2.5919,
+      "step": 2710
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.5789735317230225,
+      "learning_rate": 4.139057085406221e-05,
+      "loss": 2.6139,
+      "step": 2715
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.5745024681091309,
+      "learning_rate": 4.1226393277027726e-05,
+      "loss": 2.6619,
+      "step": 2720
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.5738227963447571,
+      "learning_rate": 4.106231324191471e-05,
+      "loss": 2.6202,
+      "step": 2725
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.5885658860206604,
+      "learning_rate": 4.089833257290817e-05,
+      "loss": 2.7186,
+      "step": 2730
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.5927093029022217,
+      "learning_rate": 4.073445309308842e-05,
+      "loss": 2.6959,
+      "step": 2735
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.5532147288322449,
+      "learning_rate": 4.0570676624410756e-05,
+      "loss": 2.5826,
+      "step": 2740
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.571595311164856,
+      "learning_rate": 4.040700498768525e-05,
+      "loss": 2.5981,
+      "step": 2745
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.6024319529533386,
+      "learning_rate": 4.024344000255648e-05,
+      "loss": 2.714,
+      "step": 2750
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.5821074843406677,
+      "learning_rate": 4.0079983487483313e-05,
+      "loss": 2.5985,
+      "step": 2755
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.5916585922241211,
+      "learning_rate": 3.9916637259718683e-05,
+      "loss": 2.6529,
+      "step": 2760
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.5751479864120483,
+      "learning_rate": 3.9753403135289396e-05,
+      "loss": 2.5818,
+      "step": 2765
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.5622469782829285,
+      "learning_rate": 3.9590282928975914e-05,
+      "loss": 2.545,
+      "step": 2770
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.5675203800201416,
+      "learning_rate": 3.942727845429221e-05,
+      "loss": 2.5872,
+      "step": 2775
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.5867101550102234,
+      "learning_rate": 3.926439152346558e-05,
+      "loss": 2.5448,
+      "step": 2780
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.5699689984321594,
+      "learning_rate": 3.910162394741653e-05,
+      "loss": 2.5741,
+      "step": 2785
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.5863639712333679,
+      "learning_rate": 3.893897753573861e-05,
+      "loss": 2.6459,
+      "step": 2790
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.5991702675819397,
+      "learning_rate": 3.877645409667829e-05,
+      "loss": 2.6341,
+      "step": 2795
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.6305561065673828,
+      "learning_rate": 3.861405543711491e-05,
+      "loss": 2.6204,
+      "step": 2800
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.5843254923820496,
+      "learning_rate": 3.8451783362540507e-05,
+      "loss": 2.6726,
+      "step": 2805
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.5682271122932434,
+      "learning_rate": 3.828963967703983e-05,
+      "loss": 2.6089,
+      "step": 2810
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.5877279043197632,
+      "learning_rate": 3.8127626183270223e-05,
+      "loss": 2.6249,
+      "step": 2815
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.5976167321205139,
+      "learning_rate": 3.796574468244161e-05,
+      "loss": 2.6543,
+      "step": 2820
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.5947586297988892,
+      "learning_rate": 3.7803996974296444e-05,
+      "loss": 2.6448,
+      "step": 2825
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.5799654722213745,
+      "learning_rate": 3.7642384857089776e-05,
+      "loss": 2.6903,
+      "step": 2830
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.5786498785018921,
+      "learning_rate": 3.748091012756915e-05,
+      "loss": 2.6172,
+      "step": 2835
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.5935390591621399,
+      "learning_rate": 3.731957458095467e-05,
+      "loss": 2.6681,
+      "step": 2840
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.5944668054580688,
+      "learning_rate": 3.71583800109191e-05,
+      "loss": 2.6853,
+      "step": 2845
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.5829348564147949,
+      "learning_rate": 3.699732820956784e-05,
+      "loss": 2.6258,
+      "step": 2850
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.5773081183433533,
+      "learning_rate": 3.6836420967419057e-05,
+      "loss": 2.6407,
+      "step": 2855
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.5687635540962219,
+      "learning_rate": 3.6675660073383745e-05,
+      "loss": 2.6407,
+      "step": 2860
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.5697055459022522,
+      "learning_rate": 3.6515047314745856e-05,
+      "loss": 2.651,
+      "step": 2865
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.574481189250946,
+      "learning_rate": 3.6354584477142437e-05,
+      "loss": 2.603,
+      "step": 2870
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.5846320390701294,
+      "learning_rate": 3.6194273344543736e-05,
+      "loss": 2.6002,
+      "step": 2875
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.5774704813957214,
+      "learning_rate": 3.6034115699233425e-05,
+      "loss": 2.6297,
+      "step": 2880
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.5734532475471497,
+      "learning_rate": 3.5874113321788736e-05,
+      "loss": 2.6309,
+      "step": 2885
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.5819912552833557,
+      "learning_rate": 3.571426799106071e-05,
+      "loss": 2.6724,
+      "step": 2890
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.5800731182098389,
+      "learning_rate": 3.555458148415437e-05,
+      "loss": 2.6073,
+      "step": 2895
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.6030119061470032,
+      "learning_rate": 3.539505557640901e-05,
+      "loss": 2.6547,
+      "step": 2900
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.5793662071228027,
+      "learning_rate": 3.523569204137843e-05,
+      "loss": 2.6946,
+      "step": 2905
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.5904523730278015,
+      "learning_rate": 3.5076492650811246e-05,
+      "loss": 2.6513,
+      "step": 2910
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.5770894289016724,
+      "learning_rate": 3.491745917463113e-05,
+      "loss": 2.6559,
+      "step": 2915
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.5736362338066101,
+      "learning_rate": 3.475859338091721e-05,
+      "loss": 2.5941,
+      "step": 2920
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.585515022277832,
+      "learning_rate": 3.4599897035884374e-05,
+      "loss": 2.6083,
+      "step": 2925
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.5967217683792114,
+      "learning_rate": 3.444137190386363e-05,
+      "loss": 2.629,
+      "step": 2930
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.5730523467063904,
+      "learning_rate": 3.4283019747282514e-05,
+      "loss": 2.6124,
+      "step": 2935
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.5762849450111389,
+      "learning_rate": 3.412484232664545e-05,
+      "loss": 2.6663,
+      "step": 2940
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.5997823476791382,
+      "learning_rate": 3.396684140051424e-05,
+      "loss": 2.5798,
+      "step": 2945
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.591468334197998,
+      "learning_rate": 3.3809018725488466e-05,
+      "loss": 2.6309,
+      "step": 2950
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.576914370059967,
+      "learning_rate": 3.365137605618598e-05,
+      "loss": 2.628,
+      "step": 2955
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.5974778532981873,
+      "learning_rate": 3.3493915145223395e-05,
+      "loss": 2.652,
+      "step": 2960
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.5637148022651672,
+      "learning_rate": 3.3336637743196584e-05,
+      "loss": 2.5695,
+      "step": 2965
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.5815153121948242,
+      "learning_rate": 3.317954559866126e-05,
+      "loss": 2.578,
+      "step": 2970
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.5781580209732056,
+      "learning_rate": 3.302264045811344e-05,
+      "loss": 2.65,
+      "step": 2975
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.5772373080253601,
+      "learning_rate": 3.286592406597021e-05,
+      "loss": 2.647,
+      "step": 2980
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.5850008130073547,
+      "learning_rate": 3.270939816455012e-05,
+      "loss": 2.5886,
+      "step": 2985
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.5770871639251709,
+      "learning_rate": 3.255306449405395e-05,
+      "loss": 2.6366,
+      "step": 2990
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.6196701526641846,
+      "learning_rate": 3.2396924792545304e-05,
+      "loss": 2.6013,
+      "step": 2995
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.5672704577445984,
+      "learning_rate": 3.224098079593132e-05,
+      "loss": 2.5439,
+      "step": 3000
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.5793458223342896,
+      "learning_rate": 3.2085234237943354e-05,
+      "loss": 2.5838,
+      "step": 3005
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.5920185446739197,
+      "learning_rate": 3.19296868501177e-05,
+      "loss": 2.7513,
+      "step": 3010
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.5743845105171204,
+      "learning_rate": 3.177434036177636e-05,
+      "loss": 2.5961,
+      "step": 3015
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.5579830408096313,
+      "learning_rate": 3.1619196500007804e-05,
+      "loss": 2.5869,
+      "step": 3020
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.5578731894493103,
+      "learning_rate": 3.146425698964776e-05,
+      "loss": 2.6034,
+      "step": 3025
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.5633421540260315,
+      "learning_rate": 3.1309523553260046e-05,
+      "loss": 2.6938,
+      "step": 3030
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.5821592211723328,
+      "learning_rate": 3.115499791111743e-05,
+      "loss": 2.5988,
+      "step": 3035
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.568416953086853,
+      "learning_rate": 3.10006817811825e-05,
+      "loss": 2.6686,
+      "step": 3040
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.6258422136306763,
+      "learning_rate": 3.084657687908855e-05,
+      "loss": 2.6229,
+      "step": 3045
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.5760878324508667,
+      "learning_rate": 3.069268491812052e-05,
+      "loss": 2.5593,
+      "step": 3050
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.7546426057815552,
+      "learning_rate": 3.0539007609195934e-05,
+      "loss": 2.6005,
+      "step": 3055
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.5613300204277039,
+      "learning_rate": 3.0385546660845908e-05,
+      "loss": 2.58,
+      "step": 3060
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.5683727264404297,
+      "learning_rate": 3.0232303779196132e-05,
+      "loss": 2.6769,
+      "step": 3065
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.5826250910758972,
+      "learning_rate": 3.0079280667947885e-05,
+      "loss": 2.6511,
+      "step": 3070
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.6152758598327637,
+      "learning_rate": 2.9926479028359132e-05,
+      "loss": 2.6877,
+      "step": 3075
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.6035573482513428,
+      "learning_rate": 2.97739005592256e-05,
+      "loss": 2.6807,
+      "step": 3080
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.5841193795204163,
+      "learning_rate": 2.962154695686187e-05,
+      "loss": 2.6925,
+      "step": 3085
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.6009371280670166,
+      "learning_rate": 2.9469419915082536e-05,
+      "loss": 2.6164,
+      "step": 3090
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.566608190536499,
+      "learning_rate": 2.9317521125183368e-05,
+      "loss": 2.5822,
+      "step": 3095
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.5858410000801086,
+      "learning_rate": 2.9165852275922524e-05,
+      "loss": 2.6449,
+      "step": 3100
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.5713921189308167,
+      "learning_rate": 2.901441505350174e-05,
+      "loss": 2.5967,
+      "step": 3105
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.5789456367492676,
+      "learning_rate": 2.886321114154762e-05,
+      "loss": 2.6496,
+      "step": 3110
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.5742778182029724,
+      "learning_rate": 2.87122422210929e-05,
+      "loss": 2.6082,
+      "step": 3115
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.5781303644180298,
+      "learning_rate": 2.8561509970557736e-05,
+      "loss": 2.6587,
+      "step": 3120
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.5684245228767395,
+      "learning_rate": 2.8411016065731146e-05,
+      "loss": 2.6136,
+      "step": 3125
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.5729696750640869,
+      "learning_rate": 2.826076217975222e-05,
+      "loss": 2.6092,
+      "step": 3130
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.5810202360153198,
+      "learning_rate": 2.8110749983091632e-05,
+      "loss": 2.5651,
+      "step": 3135
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.5778184533119202,
+      "learning_rate": 2.7960981143533053e-05,
+      "loss": 2.654,
+      "step": 3140
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.5724374055862427,
+      "learning_rate": 2.781145732615457e-05,
+      "loss": 2.6412,
+      "step": 3145
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.5706700682640076,
+      "learning_rate": 2.7662180193310218e-05,
+      "loss": 2.6387,
+      "step": 3150
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.5863929986953735,
+      "learning_rate": 2.751315140461145e-05,
+      "loss": 2.5676,
+      "step": 3155
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.5613924860954285,
+      "learning_rate": 2.7364372616908744e-05,
+      "loss": 2.5772,
+      "step": 3160
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.6091950535774231,
+      "learning_rate": 2.7215845484273152e-05,
+      "loss": 2.7069,
+      "step": 3165
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.5914263129234314,
+      "learning_rate": 2.7067571657977893e-05,
+      "loss": 2.6698,
+      "step": 3170
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.5723788142204285,
+      "learning_rate": 2.691955278648003e-05,
+      "loss": 2.5516,
+      "step": 3175
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.5798773169517517,
+      "learning_rate": 2.6771790515402112e-05,
+      "loss": 2.605,
+      "step": 3180
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.5647338032722473,
+      "learning_rate": 2.6624286487513916e-05,
+      "loss": 2.6944,
+      "step": 3185
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.5808343291282654,
+      "learning_rate": 2.6477042342714137e-05,
+      "loss": 2.6245,
+      "step": 3190
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.6048632264137268,
+      "learning_rate": 2.633005971801219e-05,
+      "loss": 2.6853,
+      "step": 3195
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.566261351108551,
+      "learning_rate": 2.6183340247510013e-05,
+      "loss": 2.6689,
+      "step": 3200
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.5952037572860718,
+      "learning_rate": 2.6036885562383856e-05,
+      "loss": 2.6039,
+      "step": 3205
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.588478684425354,
+      "learning_rate": 2.5890697290866206e-05,
+      "loss": 2.6306,
+      "step": 3210
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.5835679769515991,
+      "learning_rate": 2.5744777058227642e-05,
+      "loss": 2.6297,
+      "step": 3215
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.5727607011795044,
+      "learning_rate": 2.5599126486758777e-05,
+      "loss": 2.5732,
+      "step": 3220
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.578033447265625,
+      "learning_rate": 2.5453747195752243e-05,
+      "loss": 2.6064,
+      "step": 3225
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.8947706818580627,
+      "learning_rate": 2.530864080148464e-05,
+      "loss": 2.614,
+      "step": 3230
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.5751860737800598,
+      "learning_rate": 2.5163808917198615e-05,
+      "loss": 2.6096,
+      "step": 3235
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.5922190546989441,
+      "learning_rate": 2.501925315308492e-05,
+      "loss": 2.5628,
+      "step": 3240
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.6162046194076538,
+      "learning_rate": 2.4874975116264477e-05,
+      "loss": 2.5608,
+      "step": 3245
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.5621728897094727,
+      "learning_rate": 2.4730976410770534e-05,
+      "loss": 2.5841,
+      "step": 3250
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.5650796294212341,
+      "learning_rate": 2.458725863753084e-05,
+      "loss": 2.5644,
+      "step": 3255
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.5697013139724731,
+      "learning_rate": 2.4443823394349834e-05,
+      "loss": 2.5701,
+      "step": 3260
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.5562264323234558,
+      "learning_rate": 2.430067227589088e-05,
+      "loss": 2.5848,
+      "step": 3265
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.5883675813674927,
+      "learning_rate": 2.4157806873658517e-05,
+      "loss": 2.6811,
+      "step": 3270
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.5807322859764099,
+      "learning_rate": 2.401522877598087e-05,
+      "loss": 2.5286,
+      "step": 3275
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.5690186023712158,
+      "learning_rate": 2.3872939567991827e-05,
+      "loss": 2.5263,
+      "step": 3280
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.5782911777496338,
+      "learning_rate": 2.373094083161353e-05,
+      "loss": 2.6415,
+      "step": 3285
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.566866934299469,
+      "learning_rate": 2.358923414553877e-05,
+      "loss": 2.5329,
+      "step": 3290
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.5811752080917358,
+      "learning_rate": 2.3447821085213405e-05,
+      "loss": 2.5956,
+      "step": 3295
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.6118722558021545,
+      "learning_rate": 2.3306703222818878e-05,
+      "loss": 2.597,
+      "step": 3300
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.5883946418762207,
+      "learning_rate": 2.3165882127254705e-05,
+      "loss": 2.6247,
+      "step": 3305
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.563856840133667,
+      "learning_rate": 2.302535936412108e-05,
+      "loss": 2.5334,
+      "step": 3310
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.7375384569168091,
+      "learning_rate": 2.2885136495701415e-05,
+      "loss": 2.7001,
+      "step": 3315
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.5989105105400085,
+      "learning_rate": 2.274521508094501e-05,
+      "loss": 2.6305,
+      "step": 3320
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.5700932145118713,
+      "learning_rate": 2.2605596675449698e-05,
+      "loss": 2.6133,
+      "step": 3325
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.5767858028411865,
+      "learning_rate": 2.246628283144457e-05,
+      "loss": 2.5287,
+      "step": 3330
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.5744432210922241,
+      "learning_rate": 2.232727509777269e-05,
+      "loss": 2.5014,
+      "step": 3335
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.5841879844665527,
+      "learning_rate": 2.2188575019873932e-05,
+      "loss": 2.643,
+      "step": 3340
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.5745596289634705,
+      "learning_rate": 2.2050184139767704e-05,
+      "loss": 2.5714,
+      "step": 3345
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.5794277191162109,
+      "learning_rate": 2.191210399603591e-05,
+      "loss": 2.5296,
+      "step": 3350
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.5828683376312256,
+      "learning_rate": 2.1774336123805772e-05,
+      "loss": 2.6062,
+      "step": 3355
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.5689579844474792,
+      "learning_rate": 2.1636882054732776e-05,
+      "loss": 2.571,
+      "step": 3360
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.5862215757369995,
+      "learning_rate": 2.1499743316983684e-05,
+      "loss": 2.6261,
+      "step": 3365
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.5827880501747131,
+      "learning_rate": 2.1362921435219473e-05,
+      "loss": 2.5678,
+      "step": 3370
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.573703944683075,
+      "learning_rate": 2.1226417930578464e-05,
+      "loss": 2.5523,
+      "step": 3375
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.5824171304702759,
+      "learning_rate": 2.109023432065935e-05,
+      "loss": 2.5961,
+      "step": 3380
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.5649211406707764,
+      "learning_rate": 2.095437211950434e-05,
+      "loss": 2.5863,
+      "step": 3385
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.5749054551124573,
+      "learning_rate": 2.0818832837582352e-05,
+      "loss": 2.6205,
+      "step": 3390
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.5825015902519226,
+      "learning_rate": 2.068361798177218e-05,
+      "loss": 2.6257,
+      "step": 3395
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.5777032971382141,
+      "learning_rate": 2.0548729055345778e-05,
+      "loss": 2.5669,
+      "step": 3400
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.5922849774360657,
+      "learning_rate": 2.0414167557951514e-05,
+      "loss": 2.6121,
+      "step": 3405
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.5713364481925964,
+      "learning_rate": 2.0279934985597527e-05,
+      "loss": 2.5375,
+      "step": 3410
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.5662713050842285,
+      "learning_rate": 2.0146032830635054e-05,
+      "loss": 2.5945,
+      "step": 3415
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.5747870206832886,
+      "learning_rate": 2.001246258174192e-05,
+      "loss": 2.5869,
+      "step": 3420
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.5959161520004272,
+      "learning_rate": 1.9879225723905886e-05,
+      "loss": 2.543,
+      "step": 3425
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.5774676203727722,
+      "learning_rate": 1.9746323738408203e-05,
+      "loss": 2.5901,
+      "step": 3430
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.5725018978118896,
+      "learning_rate": 1.9613758102807117e-05,
+      "loss": 2.584,
+      "step": 3435
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.5607089996337891,
+      "learning_rate": 1.9481530290921474e-05,
+      "loss": 2.5761,
+      "step": 3440
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.577150285243988,
+      "learning_rate": 1.934964177281428e-05,
+      "loss": 2.6497,
+      "step": 3445
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.5869901776313782,
+      "learning_rate": 1.9218094014776434e-05,
+      "loss": 2.5564,
+      "step": 3450
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.5867902636528015,
+      "learning_rate": 1.9086888479310333e-05,
+      "loss": 2.6265,
+      "step": 3455
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.5620519518852234,
+      "learning_rate": 1.895602662511371e-05,
+      "loss": 2.5874,
+      "step": 3460
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.5845962762832642,
+      "learning_rate": 1.8825509907063327e-05,
+      "loss": 2.5383,
+      "step": 3465
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.5725237131118774,
+      "learning_rate": 1.8695339776198872e-05,
+      "loss": 2.5967,
+      "step": 3470
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.58806312084198,
+      "learning_rate": 1.8565517679706783e-05,
+      "loss": 2.5904,
+      "step": 3475
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.5862602591514587,
+      "learning_rate": 1.8436045060904174e-05,
+      "loss": 2.5578,
+      "step": 3480
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.5779770612716675,
+      "learning_rate": 1.830692335922279e-05,
+      "loss": 2.6801,
+      "step": 3485
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.5615940690040588,
+      "learning_rate": 1.8178154010192994e-05,
+      "loss": 2.5917,
+      "step": 3490
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.5655120015144348,
+      "learning_rate": 1.8049738445427822e-05,
+      "loss": 2.541,
+      "step": 3495
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.5825996398925781,
+      "learning_rate": 1.7921678092607052e-05,
+      "loss": 2.6431,
+      "step": 3500
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.5582937002182007,
+      "learning_rate": 1.7793974375461352e-05,
+      "loss": 2.5971,
+      "step": 3505
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.5537931323051453,
+      "learning_rate": 1.7666628713756417e-05,
+      "loss": 2.5717,
+      "step": 3510
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.5590839982032776,
+      "learning_rate": 1.7539642523277228e-05,
+      "loss": 2.6129,
+      "step": 3515
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.003821611404419,
+      "learning_rate": 1.7413017215812273e-05,
+      "loss": 2.5358,
+      "step": 3520
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.5575367212295532,
+      "learning_rate": 1.728675419913788e-05,
+      "loss": 2.5917,
+      "step": 3525
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.5800896286964417,
+      "learning_rate": 1.716085487700253e-05,
+      "loss": 2.5974,
+      "step": 3530
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.5730501413345337,
+      "learning_rate": 1.703532064911131e-05,
+      "loss": 2.6395,
+      "step": 3535
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.552326500415802,
+      "learning_rate": 1.6910152911110283e-05,
+      "loss": 2.5972,
+      "step": 3540
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.5704326033592224,
+      "learning_rate": 1.6785353054571024e-05,
+      "loss": 2.6062,
+      "step": 3545
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.58367919921875,
+      "learning_rate": 1.666092246697512e-05,
+      "loss": 2.5066,
+      "step": 3550
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.5703946352005005,
+      "learning_rate": 1.6536862531698766e-05,
+      "loss": 2.6073,
+      "step": 3555
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.5820434093475342,
+      "learning_rate": 1.6413174627997328e-05,
+      "loss": 2.6337,
+      "step": 3560
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.5775151252746582,
+      "learning_rate": 1.6289860130990147e-05,
+      "loss": 2.6206,
+      "step": 3565
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.5769268870353699,
+      "learning_rate": 1.6166920411645064e-05,
+      "loss": 2.5696,
+      "step": 3570
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.5777645111083984,
+      "learning_rate": 1.6044356836763315e-05,
+      "loss": 2.511,
+      "step": 3575
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.5798133611679077,
+      "learning_rate": 1.5922170768964285e-05,
+      "loss": 2.5312,
+      "step": 3580
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.5712454915046692,
+      "learning_rate": 1.5800363566670362e-05,
+      "loss": 2.5797,
+      "step": 3585
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.5668451189994812,
+      "learning_rate": 1.5678936584091852e-05,
+      "loss": 2.604,
+      "step": 3590
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.5877907872200012,
+      "learning_rate": 1.5557891171211892e-05,
+      "loss": 2.6074,
+      "step": 3595
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.5750396847724915,
+      "learning_rate": 1.5437228673771465e-05,
+      "loss": 2.6123,
+      "step": 3600
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.5754581689834595,
+      "learning_rate": 1.5316950433254445e-05,
+      "loss": 2.5287,
+      "step": 3605
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.567247211933136,
+      "learning_rate": 1.5197057786872649e-05,
+      "loss": 2.5492,
+      "step": 3610
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.5782046914100647,
+      "learning_rate": 1.5077552067551015e-05,
+      "loss": 2.5554,
+      "step": 3615
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.5800824165344238,
+      "learning_rate": 1.4958434603912747e-05,
+      "loss": 2.6436,
+      "step": 3620
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.5893679857254028,
+      "learning_rate": 1.4839706720264546e-05,
+      "loss": 2.5866,
+      "step": 3625
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.5560811161994934,
+      "learning_rate": 1.4721369736581924e-05,
+      "loss": 2.5058,
+      "step": 3630
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.6068143844604492,
+      "learning_rate": 1.4603424968494484e-05,
+      "loss": 2.6076,
+      "step": 3635
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.5767700672149658,
+      "learning_rate": 1.448587372727132e-05,
+      "loss": 2.5211,
+      "step": 3640
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.5597636103630066,
+      "learning_rate": 1.4368717319806419e-05,
+      "loss": 2.6155,
+      "step": 3645
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.5742618441581726,
+      "learning_rate": 1.4251957048604152e-05,
+      "loss": 2.5888,
+      "step": 3650
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.6016988158226013,
+      "learning_rate": 1.413559421176479e-05,
+      "loss": 2.5928,
+      "step": 3655
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.6035110354423523,
+      "learning_rate": 1.4019630102970056e-05,
+      "loss": 2.589,
+      "step": 3660
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.5645198822021484,
+      "learning_rate": 1.3904066011468753e-05,
+      "loss": 2.5914,
+      "step": 3665
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.5627663135528564,
+      "learning_rate": 1.3788903222062433e-05,
+      "loss": 2.5587,
+      "step": 3670
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.5593175292015076,
+      "learning_rate": 1.3674143015091118e-05,
+      "loss": 2.5776,
+      "step": 3675
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.5947297215461731,
+      "learning_rate": 1.355978666641905e-05,
+      "loss": 2.606,
+      "step": 3680
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.5737603306770325,
+      "learning_rate": 1.3445835447420507e-05,
+      "loss": 2.6054,
+      "step": 3685
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.5874439477920532,
+      "learning_rate": 1.3332290624965688e-05,
+      "loss": 2.5885,
+      "step": 3690
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.5544983744621277,
+      "learning_rate": 1.3219153461406609e-05,
+      "loss": 2.4788,
+      "step": 3695
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.5916979908943176,
+      "learning_rate": 1.3106425214563078e-05,
+      "loss": 2.5666,
+      "step": 3700
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.5403845310211182,
+      "learning_rate": 1.2994107137708716e-05,
+      "loss": 2.5592,
+      "step": 3705
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.5737205147743225,
+      "learning_rate": 1.2882200479556988e-05,
+      "loss": 2.5363,
+      "step": 3710
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.5680940747261047,
+      "learning_rate": 1.2770706484247397e-05,
+      "loss": 2.531,
+      "step": 3715
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.5880336165428162,
+      "learning_rate": 1.2659626391331564e-05,
+      "loss": 2.5294,
+      "step": 3720
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.5889505743980408,
+      "learning_rate": 1.2548961435759493e-05,
+      "loss": 2.5716,
+      "step": 3725
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.5553359389305115,
+      "learning_rate": 1.2438712847865846e-05,
+      "loss": 2.5521,
+      "step": 3730
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.5884903073310852,
+      "learning_rate": 1.2328881853356244e-05,
+      "loss": 2.5846,
+      "step": 3735
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.5808243155479431,
+      "learning_rate": 1.221946967329365e-05,
+      "loss": 2.6001,
+      "step": 3740
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.5739181637763977,
+      "learning_rate": 1.2110477524084796e-05,
+      "loss": 2.6288,
+      "step": 3745
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.5804181694984436,
+      "learning_rate": 1.2001906617466657e-05,
+      "loss": 2.5253,
+      "step": 3750
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.5736623406410217,
+      "learning_rate": 1.1893758160492978e-05,
+      "loss": 2.6306,
+      "step": 3755
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.5596362352371216,
+      "learning_rate": 1.1786033355520859e-05,
+      "loss": 2.5859,
+      "step": 3760
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.5757764577865601,
+      "learning_rate": 1.1678733400197373e-05,
+      "loss": 2.5413,
+      "step": 3765
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.5746675729751587,
+      "learning_rate": 1.1571859487446263e-05,
+      "loss": 2.5768,
+      "step": 3770
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.599814772605896,
+      "learning_rate": 1.1465412805454695e-05,
+      "loss": 2.6325,
+      "step": 3775
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.5795137882232666,
+      "learning_rate": 1.1359394537660011e-05,
+      "loss": 2.6282,
+      "step": 3780
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.5906032919883728,
+      "learning_rate": 1.125380586273661e-05,
+      "loss": 2.5769,
+      "step": 3785
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.5713207721710205,
+      "learning_rate": 1.1148647954582808e-05,
+      "loss": 2.5704,
+      "step": 3790
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.5629163980484009,
+      "learning_rate": 1.1043921982307819e-05,
+      "loss": 2.5626,
+      "step": 3795
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.583859384059906,
+      "learning_rate": 1.0939629110218735e-05,
+      "loss": 2.5515,
+      "step": 3800
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.5773959159851074,
+      "learning_rate": 1.0835770497807596e-05,
+      "loss": 2.5416,
+      "step": 3805
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.5668864250183105,
+      "learning_rate": 1.0732347299738493e-05,
+      "loss": 2.5416,
+      "step": 3810
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.5780789256095886,
+      "learning_rate": 1.0629360665834732e-05,
+      "loss": 2.5834,
+      "step": 3815
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.5882496237754822,
+      "learning_rate": 1.052681174106604e-05,
+      "loss": 2.623,
+      "step": 3820
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.5597360134124756,
+      "learning_rate": 1.0424701665535852e-05,
+      "loss": 2.5469,
+      "step": 3825
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.5949521660804749,
+      "learning_rate": 1.0323031574468638e-05,
+      "loss": 2.5835,
+      "step": 3830
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.5551010370254517,
+      "learning_rate": 1.0221802598197261e-05,
+      "loss": 2.605,
+      "step": 3835
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.5903000831604004,
+      "learning_rate": 1.0121015862150423e-05,
+      "loss": 2.5859,
+      "step": 3840
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.5738072395324707,
+      "learning_rate": 1.0020672486840154e-05,
+      "loss": 2.599,
+      "step": 3845
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.5860092639923096,
+      "learning_rate": 9.920773587849364e-06,
+      "loss": 2.5565,
+      "step": 3850
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.5723434090614319,
+      "learning_rate": 9.821320275819401e-06,
+      "loss": 2.5936,
+      "step": 3855
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.56388920545578,
+      "learning_rate": 9.72231365643777e-06,
+      "loss": 2.5033,
+      "step": 3860
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.5733988285064697,
+      "learning_rate": 9.623754830425779e-06,
+      "loss": 2.5936,
+      "step": 3865
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.5776930451393127,
+      "learning_rate": 9.52564489352632e-06,
+      "loss": 2.5885,
+      "step": 3870
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.5818698406219482,
+      "learning_rate": 9.427984936491702e-06,
+      "loss": 2.6229,
+      "step": 3875
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.5616958141326904,
+      "learning_rate": 9.330776045071509e-06,
+      "loss": 2.5224,
+      "step": 3880
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.5539332032203674,
+      "learning_rate": 9.23401930000054e-06,
+      "loss": 2.5469,
+      "step": 3885
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.5630046129226685,
+      "learning_rate": 9.137715776986772e-06,
+      "loss": 2.5892,
+      "step": 3890
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.5861016511917114,
+      "learning_rate": 9.041866546699434e-06,
+      "loss": 2.5578,
+      "step": 3895
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.5587556958198547,
+      "learning_rate": 8.946472674757078e-06,
+      "loss": 2.5814,
+      "step": 3900
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.560431957244873,
+      "learning_rate": 8.851535221715735e-06,
+      "loss": 2.5942,
+      "step": 3905
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.571217954158783,
+      "learning_rate": 8.757055243057132e-06,
+      "loss": 2.6199,
+      "step": 3910
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.6178560853004456,
+      "learning_rate": 8.663033789176967e-06,
+      "loss": 2.5234,
+      "step": 3915
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.5733814835548401,
+      "learning_rate": 8.5694719053732e-06,
+      "loss": 2.5297,
+      "step": 3920
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.5747708082199097,
+      "learning_rate": 8.476370631834458e-06,
+      "loss": 2.5832,
+      "step": 3925
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.5666419267654419,
+      "learning_rate": 8.383731003628452e-06,
+      "loss": 2.524,
+      "step": 3930
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.5650913119316101,
+      "learning_rate": 8.291554050690508e-06,
+      "loss": 2.5187,
+      "step": 3935
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.5808721780776978,
+      "learning_rate": 8.199840797812058e-06,
+      "loss": 2.5505,
+      "step": 3940
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.566299557685852,
+      "learning_rate": 8.108592264629295e-06,
+      "loss": 2.5946,
+      "step": 3945
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.5880241394042969,
+      "learning_rate": 8.017809465611803e-06,
+      "loss": 2.589,
+      "step": 3950
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.5790772438049316,
+      "learning_rate": 7.927493410051324e-06,
+      "loss": 2.5619,
+      "step": 3955
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.5659160614013672,
+      "learning_rate": 7.837645102050473e-06,
+      "loss": 2.562,
+      "step": 3960
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.5623981356620789,
+      "learning_rate": 7.748265540511635e-06,
+      "loss": 2.5778,
+      "step": 3965
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.5713520050048828,
+      "learning_rate": 7.65935571912582e-06,
+      "loss": 2.5206,
+      "step": 3970
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.5754117965698242,
+      "learning_rate": 7.5709166263616405e-06,
+      "loss": 2.5908,
+      "step": 3975
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.5733134746551514,
+      "learning_rate": 7.482949245454302e-06,
+      "loss": 2.5512,
+      "step": 3980
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.5658825635910034,
+      "learning_rate": 7.3954545543946876e-06,
+      "loss": 2.5486,
+      "step": 3985
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.5804485082626343,
+      "learning_rate": 7.308433525918468e-06,
+      "loss": 2.5343,
+      "step": 3990
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.559240460395813,
+      "learning_rate": 7.221887127495313e-06,
+      "loss": 2.5867,
+      "step": 3995
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.5767809152603149,
+      "learning_rate": 7.1358163213181114e-06,
+      "loss": 2.6351,
+      "step": 4000
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 4811,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 1000,
+  "total_flos": 1.0865380348133376e+19,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}