diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -10,12578 +10,12578 @@
   "log_history": [
     {
       "epoch": 0.0,
-      "grad_norm": 14.375,
+      "grad_norm": 468.0,
       "learning_rate": 0.001,
-      "loss": 10.66,
+      "loss": 11.4202,
       "step": 1
     },
     {
       "epoch": 0.0,
-      "grad_norm": 2.15625,
+      "grad_norm": 30.75,
       "learning_rate": 0.001,
-      "loss": 9.0861,
+      "loss": 19.6347,
       "step": 5
     },
     {
       "epoch": 0.0,
-      "grad_norm": 0.89453125,
+      "grad_norm": 14.0,
       "learning_rate": 0.001,
-      "loss": 7.5929,
+      "loss": 10.8315,
       "step": 10
     },
     {
       "epoch": 0.0,
-      "grad_norm": 0.80078125,
+      "grad_norm": 6.6875,
       "learning_rate": 0.001,
-      "loss": 7.2385,
+      "loss": 9.9845,
       "step": 15
     },
     {
       "epoch": 0.0,
-      "grad_norm": 0.890625,
+      "grad_norm": 7.65625,
       "learning_rate": 0.001,
-      "loss": 7.1247,
+      "loss": 9.1775,
       "step": 20
     },
     {
       "epoch": 0.0,
-      "grad_norm": 0.55078125,
+      "grad_norm": 4.84375,
       "learning_rate": 0.001,
-      "loss": 6.9193,
+      "loss": 8.5303,
       "step": 25
     },
     {
       "epoch": 0.0,
-      "grad_norm": 0.78515625,
+      "grad_norm": 3.671875,
       "learning_rate": 0.001,
-      "loss": 6.8173,
+      "loss": 8.0888,
       "step": 30
     },
     {
       "epoch": 0.0,
-      "grad_norm": 0.68359375,
+      "grad_norm": 3.328125,
       "learning_rate": 0.001,
-      "loss": 6.7162,
+      "loss": 7.8222,
       "step": 35
     },
     {
       "epoch": 0.0,
-      "grad_norm": 0.67578125,
+      "grad_norm": 3.265625,
       "learning_rate": 0.001,
-      "loss": 6.6121,
+      "loss": 7.3621,
       "step": 40
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.7265625,
+      "grad_norm": 2.859375,
       "learning_rate": 0.001,
-      "loss": 6.4806,
+      "loss": 7.1087,
       "step": 45
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.63671875,
+      "grad_norm": 3.390625,
       "learning_rate": 0.001,
-      "loss": 6.4596,
+      "loss": 6.9641,
       "step": 50
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.875,
+      "grad_norm": 3.65625,
       "learning_rate": 0.001,
-      "loss": 6.3081,
+      "loss": 6.7323,
       "step": 55
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.8125,
+      "grad_norm": 2.34375,
       "learning_rate": 0.001,
-      "loss": 6.2485,
+      "loss": 6.5935,
       "step": 60
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.765625,
+      "grad_norm": 3.328125,
       "learning_rate": 0.001,
-      "loss": 6.1595,
+      "loss": 6.7137,
       "step": 65
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.80859375,
+      "grad_norm": 3.015625,
       "learning_rate": 0.001,
-      "loss": 6.113,
+      "loss": 6.4158,
       "step": 70
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.89453125,
+      "grad_norm": 3.09375,
       "learning_rate": 0.001,
-      "loss": 6.0143,
+      "loss": 6.3833,
       "step": 75
     },
     {
       "epoch": 0.01,
-      "grad_norm": 1.203125,
+      "grad_norm": 2.265625,
       "learning_rate": 0.001,
-      "loss": 5.9454,
+      "loss": 6.4733,
       "step": 80
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.78515625,
+      "grad_norm": 2.90625,
       "learning_rate": 0.001,
-      "loss": 5.8563,
+      "loss": 6.2427,
       "step": 85
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.89453125,
+      "grad_norm": 2.078125,
       "learning_rate": 0.001,
-      "loss": 5.8343,
+      "loss": 6.2661,
       "step": 90
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.6015625,
+      "grad_norm": 2.140625,
       "learning_rate": 0.001,
-      "loss": 5.8936,
+      "loss": 6.1774,
       "step": 95
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.765625,
+      "grad_norm": 2.234375,
       "learning_rate": 0.001,
-      "loss": 5.6957,
+      "loss": 6.1615,
       "step": 100
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.5,
+      "grad_norm": 2.171875,
       "learning_rate": 0.001,
-      "loss": 5.6288,
+      "loss": 6.1184,
       "step": 105
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.73828125,
+      "grad_norm": 2.328125,
       "learning_rate": 0.001,
-      "loss": 5.601,
+      "loss": 5.9304,
       "step": 110
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.64453125,
+      "grad_norm": 2.21875,
       "learning_rate": 0.001,
-      "loss": 5.6385,
+      "loss": 5.9465,
       "step": 115
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.5625,
+      "grad_norm": 2.234375,
       "learning_rate": 0.001,
-      "loss": 5.5606,
+      "loss": 5.8792,
       "step": 120
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.625,
+      "grad_norm": 2.03125,
       "learning_rate": 0.001,
-      "loss": 5.53,
+      "loss": 5.768,
       "step": 125
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.62109375,
+      "grad_norm": 2.09375,
       "learning_rate": 0.001,
-      "loss": 5.512,
+      "loss": 5.7801,
       "step": 130
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.91015625,
+      "grad_norm": 2.640625,
       "learning_rate": 0.001,
-      "loss": 5.3661,
+      "loss": 5.7555,
       "step": 135
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.5546875,
+      "grad_norm": 2.796875,
       "learning_rate": 0.001,
-      "loss": 5.4498,
+      "loss": 5.5612,
       "step": 140
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.54296875,
+      "grad_norm": 1.8359375,
       "learning_rate": 0.001,
-      "loss": 5.3858,
+      "loss": 5.6745,
       "step": 145
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.59375,
+      "grad_norm": 1.703125,
       "learning_rate": 0.001,
-      "loss": 5.3242,
+      "loss": 5.6213,
       "step": 150
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.9140625,
+      "grad_norm": 2.046875,
       "learning_rate": 0.001,
-      "loss": 5.3414,
+      "loss": 5.5163,
       "step": 155
     },
     {
       "epoch": 0.02,
-      "grad_norm": 1.15625,
+      "grad_norm": 2.234375,
       "learning_rate": 0.001,
-      "loss": 5.3306,
+      "loss": 5.4944,
       "step": 160
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.79296875,
+      "grad_norm": 1.8515625,
       "learning_rate": 0.001,
-      "loss": 5.2121,
+      "loss": 5.5869,
       "step": 165
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.578125,
+      "grad_norm": 2.515625,
       "learning_rate": 0.001,
-      "loss": 5.1801,
+      "loss": 5.4195,
       "step": 170
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.71484375,
+      "grad_norm": 1.96875,
       "learning_rate": 0.001,
-      "loss": 5.1879,
+      "loss": 5.3747,
       "step": 175
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.76953125,
+      "grad_norm": 2.03125,
       "learning_rate": 0.001,
-      "loss": 5.0849,
+      "loss": 5.3658,
       "step": 180
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.75,
+      "grad_norm": 1.953125,
       "learning_rate": 0.001,
-      "loss": 5.1201,
+      "loss": 5.2998,
       "step": 185
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.77734375,
+      "grad_norm": 1.9375,
       "learning_rate": 0.001,
-      "loss": 5.0651,
+      "loss": 5.4196,
       "step": 190
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.7109375,
+      "grad_norm": 2.125,
       "learning_rate": 0.001,
-      "loss": 5.0007,
+      "loss": 5.3192,
       "step": 195
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.60546875,
+      "grad_norm": 1.9453125,
       "learning_rate": 0.001,
-      "loss": 5.0496,
+      "loss": 5.2397,
       "step": 200
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.5390625,
+      "grad_norm": 2.28125,
       "learning_rate": 0.001,
-      "loss": 4.9546,
+      "loss": 5.2464,
       "step": 205
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.73828125,
+      "grad_norm": 1.84375,
       "learning_rate": 0.001,
-      "loss": 4.9174,
+      "loss": 5.2192,
       "step": 210
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.609375,
+      "grad_norm": 2.203125,
       "learning_rate": 0.001,
-      "loss": 4.8822,
+      "loss": 5.1029,
       "step": 215
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.6171875,
+      "grad_norm": 1.9296875,
       "learning_rate": 0.001,
-      "loss": 4.9179,
+      "loss": 5.1451,
       "step": 220
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.55859375,
+      "grad_norm": 1.9375,
       "learning_rate": 0.001,
-      "loss": 4.8214,
+      "loss": 5.1545,
       "step": 225
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.625,
+      "grad_norm": 1.828125,
       "learning_rate": 0.001,
-      "loss": 4.8761,
+      "loss": 5.0767,
       "step": 230
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.55078125,
+      "grad_norm": 1.7890625,
       "learning_rate": 0.001,
-      "loss": 4.9586,
+      "loss": 5.0355,
       "step": 235
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.6484375,
+      "grad_norm": 2.328125,
       "learning_rate": 0.001,
-      "loss": 4.8608,
+      "loss": 5.0593,
       "step": 240
     },
     {
       "epoch": 0.03,
-      "grad_norm": 1.0703125,
+      "grad_norm": 1.984375,
       "learning_rate": 0.001,
-      "loss": 4.8165,
+      "loss": 5.0839,
       "step": 245
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.71484375,
+      "grad_norm": 1.5859375,
       "learning_rate": 0.001,
-      "loss": 4.8616,
+      "loss": 4.9712,
       "step": 250
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.671875,
+      "grad_norm": 1.84375,
       "learning_rate": 0.001,
-      "loss": 4.7908,
+      "loss": 4.968,
       "step": 255
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.52734375,
+      "grad_norm": 2.0625,
       "learning_rate": 0.001,
-      "loss": 4.7085,
+      "loss": 5.0095,
       "step": 260
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.6484375,
+      "grad_norm": 2.203125,
       "learning_rate": 0.001,
-      "loss": 4.6968,
+      "loss": 4.9378,
       "step": 265
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.67578125,
+      "grad_norm": 1.515625,
       "learning_rate": 0.001,
-      "loss": 4.648,
+      "loss": 4.9852,
       "step": 270
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.62109375,
+      "grad_norm": 1.75,
       "learning_rate": 0.001,
-      "loss": 4.7703,
+      "loss": 4.9289,
       "step": 275
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.71484375,
+      "grad_norm": 2.09375,
       "learning_rate": 0.001,
-      "loss": 4.6552,
+      "loss": 4.9237,
       "step": 280
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.51953125,
+      "grad_norm": 2.171875,
       "learning_rate": 0.001,
-      "loss": 4.6484,
+      "loss": 4.7961,
       "step": 285
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.546875,
+      "grad_norm": 1.6640625,
       "learning_rate": 0.001,
-      "loss": 4.6077,
+      "loss": 4.8586,
       "step": 290
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.69140625,
+      "grad_norm": 1.6796875,
       "learning_rate": 0.001,
-      "loss": 4.5301,
+      "loss": 4.7868,
       "step": 295
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.50390625,
+      "grad_norm": 1.59375,
       "learning_rate": 0.001,
-      "loss": 4.5485,
+      "loss": 4.7914,
       "step": 300
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.73046875,
+      "grad_norm": 1.734375,
       "learning_rate": 0.001,
-      "loss": 4.6582,
+      "loss": 4.8173,
       "step": 305
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.546875,
+      "grad_norm": 1.609375,
       "learning_rate": 0.001,
-      "loss": 4.6396,
+      "loss": 4.8026,
       "step": 310
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.5625,
+      "grad_norm": 1.5703125,
       "learning_rate": 0.001,
-      "loss": 4.4431,
+      "loss": 4.7621,
       "step": 315
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.5859375,
+      "grad_norm": 1.7734375,
       "learning_rate": 0.001,
-      "loss": 4.4276,
+      "loss": 4.7661,
       "step": 320
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.58203125,
+      "grad_norm": 1.8125,
       "learning_rate": 0.001,
-      "loss": 4.5243,
+      "loss": 4.7696,
       "step": 325
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.6796875,
+      "grad_norm": 1.546875,
       "learning_rate": 0.001,
-      "loss": 4.4766,
+      "loss": 4.657,
       "step": 330
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.4765625,
+      "grad_norm": 1.8984375,
       "learning_rate": 0.001,
-      "loss": 4.5035,
+      "loss": 4.7683,
       "step": 335
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.5,
+      "grad_norm": 2.71875,
       "learning_rate": 0.001,
-      "loss": 4.4449,
+      "loss": 4.6882,
       "step": 340
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.5625,
+      "grad_norm": 2.171875,
       "learning_rate": 0.001,
-      "loss": 4.3884,
+      "loss": 4.7274,
       "step": 345
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.58203125,
+      "grad_norm": 1.6953125,
       "learning_rate": 0.001,
-      "loss": 4.4832,
+      "loss": 4.7224,
       "step": 350
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.546875,
+      "grad_norm": 1.921875,
       "learning_rate": 0.001,
-      "loss": 4.3619,
+      "loss": 4.753,
       "step": 355
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.64453125,
+      "grad_norm": 1.4921875,
       "learning_rate": 0.001,
-      "loss": 4.4318,
+      "loss": 4.5646,
       "step": 360
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.59375,
+      "grad_norm": 1.828125,
       "learning_rate": 0.001,
-      "loss": 4.3091,
+      "loss": 4.6671,
       "step": 365
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.609375,
+      "grad_norm": 2.0625,
       "learning_rate": 0.001,
-      "loss": 4.3793,
+      "loss": 4.6817,
       "step": 370
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.60546875,
+      "grad_norm": 1.734375,
       "learning_rate": 0.001,
-      "loss": 4.3673,
+      "loss": 4.5382,
       "step": 375
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.7734375,
+      "grad_norm": 1.7734375,
       "learning_rate": 0.001,
-      "loss": 4.2584,
+      "loss": 4.5105,
       "step": 380
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.5078125,
+      "grad_norm": 1.515625,
       "learning_rate": 0.001,
-      "loss": 4.432,
+      "loss": 4.6714,
       "step": 385
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.53515625,
+      "grad_norm": 1.5234375,
       "learning_rate": 0.001,
-      "loss": 4.2351,
+      "loss": 4.5989,
       "step": 390
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.462890625,
+      "grad_norm": 1.6171875,
       "learning_rate": 0.001,
-      "loss": 4.3768,
+      "loss": 4.6162,
       "step": 395
     },
     {
       "epoch": 0.04,
-      "grad_norm": 0.5,
+      "grad_norm": 1.546875,
       "learning_rate": 0.001,
-      "loss": 4.2398,
+      "loss": 4.5086,
       "step": 400
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.484375,
+      "grad_norm": 1.734375,
       "learning_rate": 0.001,
-      "loss": 4.2591,
+      "loss": 4.5757,
       "step": 405
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.51953125,
+      "grad_norm": 2.171875,
       "learning_rate": 0.001,
-      "loss": 4.2083,
+      "loss": 4.5496,
       "step": 410
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.52734375,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 4.2354,
+      "loss": 4.5085,
       "step": 415
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.640625,
+      "grad_norm": 1.5,
       "learning_rate": 0.001,
-      "loss": 4.2133,
+      "loss": 4.4707,
       "step": 420
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.58203125,
+      "grad_norm": 1.4609375,
       "learning_rate": 0.001,
-      "loss": 4.2055,
+      "loss": 4.4716,
       "step": 425
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.56640625,
+      "grad_norm": 1.7109375,
       "learning_rate": 0.001,
-      "loss": 4.1682,
+      "loss": 4.5159,
       "step": 430
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.53515625,
+      "grad_norm": 1.6953125,
       "learning_rate": 0.001,
-      "loss": 4.2379,
+      "loss": 4.5706,
       "step": 435
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.578125,
+      "grad_norm": 1.6640625,
       "learning_rate": 0.001,
-      "loss": 4.1974,
+      "loss": 4.4737,
       "step": 440
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.49609375,
+      "grad_norm": 1.640625,
       "learning_rate": 0.001,
-      "loss": 4.2177,
+      "loss": 4.4327,
       "step": 445
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.59375,
+      "grad_norm": 1.828125,
       "learning_rate": 0.001,
-      "loss": 4.1886,
+      "loss": 4.4394,
       "step": 450
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.58203125,
+      "grad_norm": 1.71875,
       "learning_rate": 0.001,
-      "loss": 4.1948,
+      "loss": 4.495,
       "step": 455
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.5390625,
+      "grad_norm": 1.6015625,
       "learning_rate": 0.001,
-      "loss": 4.2911,
+      "loss": 4.3558,
       "step": 460
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.515625,
+      "grad_norm": 1.390625,
       "learning_rate": 0.001,
-      "loss": 4.2212,
+      "loss": 4.3452,
       "step": 465
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.49609375,
+      "grad_norm": 1.6484375,
       "learning_rate": 0.001,
-      "loss": 4.1132,
+      "loss": 4.3795,
       "step": 470
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.6796875,
+      "grad_norm": 1.4140625,
       "learning_rate": 0.001,
-      "loss": 4.1323,
+      "loss": 4.3633,
       "step": 475
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.5546875,
+      "grad_norm": 1.828125,
       "learning_rate": 0.001,
-      "loss": 4.0926,
+      "loss": 4.3212,
       "step": 480
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.53125,
+      "grad_norm": 1.3828125,
       "learning_rate": 0.001,
-      "loss": 4.0398,
+      "loss": 4.4133,
       "step": 485
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.484375,
+      "grad_norm": 1.40625,
       "learning_rate": 0.001,
-      "loss": 4.1016,
+      "loss": 4.3349,
       "step": 490
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.578125,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 3.9908,
+      "loss": 4.355,
       "step": 495
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.5,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 4.1144,
+      "loss": 4.2691,
       "step": 500
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.59375,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 4.0445,
+      "loss": 4.2382,
       "step": 505
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.52734375,
+      "grad_norm": 1.6328125,
       "learning_rate": 0.001,
-      "loss": 4.0426,
+      "loss": 4.3767,
       "step": 510
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.5234375,
+      "grad_norm": 1.4375,
       "learning_rate": 0.001,
-      "loss": 4.0339,
+      "loss": 4.323,
       "step": 515
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.55859375,
+      "grad_norm": 1.5625,
       "learning_rate": 0.001,
-      "loss": 4.0517,
+      "loss": 4.3758,
       "step": 520
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.5625,
+      "grad_norm": 1.6875,
       "learning_rate": 0.001,
-      "loss": 3.9779,
+      "loss": 4.3027,
       "step": 525
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.482421875,
+      "grad_norm": 1.5546875,
       "learning_rate": 0.001,
-      "loss": 3.9787,
+      "loss": 4.3068,
       "step": 530
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.494140625,
+      "grad_norm": 1.4375,
       "learning_rate": 0.001,
-      "loss": 3.9433,
+      "loss": 4.3135,
       "step": 535
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.58984375,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 3.9909,
+      "loss": 4.2856,
       "step": 540
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.498046875,
+      "grad_norm": 1.5078125,
       "learning_rate": 0.001,
-      "loss": 3.9582,
+      "loss": 4.291,
       "step": 545
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.453125,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 3.996,
+      "loss": 4.262,
       "step": 550
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.546875,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 4.0016,
+      "loss": 4.3267,
       "step": 555
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.64453125,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 3.9915,
+      "loss": 4.268,
       "step": 560
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.640625,
+      "grad_norm": 1.90625,
       "learning_rate": 0.001,
-      "loss": 3.8457,
+      "loss": 4.3043,
       "step": 565
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.5,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 3.8556,
+      "loss": 4.083,
       "step": 570
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.578125,
+      "grad_norm": 1.5234375,
       "learning_rate": 0.001,
-      "loss": 3.9233,
+      "loss": 4.1305,
       "step": 575
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.62109375,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 3.9413,
+      "loss": 4.1327,
       "step": 580
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.474609375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 3.968,
+      "loss": 4.2689,
       "step": 585
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.486328125,
+      "grad_norm": 1.515625,
       "learning_rate": 0.001,
-      "loss": 3.971,
+      "loss": 4.1684,
       "step": 590
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.4765625,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 4.0015,
+      "loss": 4.1342,
       "step": 595
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.60546875,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 3.8761,
+      "loss": 4.1759,
       "step": 600
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.48828125,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 3.8149,
+      "loss": 4.1624,
       "step": 605
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.5,
+      "grad_norm": 1.484375,
       "learning_rate": 0.001,
-      "loss": 3.8305,
+      "loss": 4.201,
       "step": 610
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.498046875,
+      "grad_norm": 1.5703125,
       "learning_rate": 0.001,
-      "loss": 3.7626,
+      "loss": 4.2148,
       "step": 615
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.46875,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 3.8127,
+      "loss": 4.1464,
       "step": 620
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.52734375,
+      "grad_norm": 1.6328125,
       "learning_rate": 0.001,
-      "loss": 3.8408,
+      "loss": 4.1184,
       "step": 625
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.4765625,
+      "grad_norm": 1.8359375,
       "learning_rate": 0.001,
-      "loss": 3.9199,
+      "loss": 4.027,
       "step": 630
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.60546875,
+      "grad_norm": 1.453125,
       "learning_rate": 0.001,
-      "loss": 3.8861,
+      "loss": 4.0665,
       "step": 635
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.451171875,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 3.8037,
+      "loss": 4.1904,
       "step": 640
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.50390625,
+      "grad_norm": 1.640625,
       "learning_rate": 0.001,
-      "loss": 3.8155,
+      "loss": 4.0893,
       "step": 645
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.5859375,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 3.8044,
+      "loss": 4.0258,
       "step": 650
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.4921875,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 3.7604,
+      "loss": 4.0683,
       "step": 655
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.5390625,
+      "grad_norm": 1.5,
       "learning_rate": 0.001,
-      "loss": 3.7374,
+      "loss": 4.0854,
       "step": 660
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.54296875,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 3.7128,
+      "loss": 3.9921,
       "step": 665
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.4453125,
+      "grad_norm": 1.8515625,
       "learning_rate": 0.001,
-      "loss": 3.6821,
+      "loss": 4.1034,
       "step": 670
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.466796875,
+      "grad_norm": 1.5,
       "learning_rate": 0.001,
-      "loss": 3.689,
+      "loss": 4.119,
       "step": 675
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.53125,
+      "grad_norm": 1.515625,
       "learning_rate": 0.001,
-      "loss": 3.7706,
+      "loss": 4.1259,
       "step": 680
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.53125,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 3.7512,
+      "loss": 4.0325,
       "step": 685
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.53515625,
+      "grad_norm": 1.390625,
       "learning_rate": 0.001,
-      "loss": 3.687,
+      "loss": 4.002,
       "step": 690
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.57421875,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 3.6497,
+      "loss": 4.1466,
       "step": 695
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.55859375,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 3.7325,
+      "loss": 3.9527,
       "step": 700
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.59375,
+      "grad_norm": 1.5390625,
       "learning_rate": 0.001,
-      "loss": 3.6482,
+      "loss": 4.1504,
       "step": 705
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.48046875,
+      "grad_norm": 1.453125,
       "learning_rate": 0.001,
-      "loss": 3.7139,
+      "loss": 4.0782,
       "step": 710
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.58984375,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 3.6265,
+      "loss": 3.9826,
       "step": 715
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.478515625,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 3.6359,
+      "loss": 4.0219,
       "step": 720
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.4609375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 3.689,
+      "loss": 4.0031,
       "step": 725
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.4921875,
+      "grad_norm": 1.5390625,
       "learning_rate": 0.001,
-      "loss": 3.664,
+      "loss": 3.9652,
       "step": 730
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.5234375,
+      "grad_norm": 1.3828125,
       "learning_rate": 0.001,
-      "loss": 3.5956,
+      "loss": 3.9176,
       "step": 735
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.458984375,
+      "grad_norm": 1.4609375,
       "learning_rate": 0.001,
-      "loss": 3.6903,
+      "loss": 3.9693,
       "step": 740
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.47265625,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 3.7775,
+      "loss": 4.1027,
       "step": 745
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.462890625,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 3.604,
+      "loss": 3.9645,
       "step": 750
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.5390625,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 3.5746,
+      "loss": 3.8947,
       "step": 755
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.455078125,
+      "grad_norm": 1.453125,
       "learning_rate": 0.001,
-      "loss": 3.5907,
+      "loss": 4.015,
       "step": 760
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.41796875,
+      "grad_norm": 1.578125,
       "learning_rate": 0.001,
-      "loss": 3.6306,
+      "loss": 3.9853,
       "step": 765
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.498046875,
+      "grad_norm": 1.390625,
       "learning_rate": 0.001,
-      "loss": 3.7143,
+      "loss": 3.9032,
       "step": 770
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.478515625,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 3.6161,
+      "loss": 3.9793,
       "step": 775
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.462890625,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 3.6404,
+      "loss": 3.8555,
       "step": 780
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.40625,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 3.6052,
+      "loss": 3.853,
       "step": 785
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.55078125,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 3.5776,
+      "loss": 3.7918,
       "step": 790
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.484375,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 3.56,
+      "loss": 3.8864,
       "step": 795
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.484375,
+      "grad_norm": 1.6640625,
       "learning_rate": 0.001,
-      "loss": 3.5469,
+      "loss": 3.9658,
       "step": 800
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.5625,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 3.637,
+      "loss": 3.9346,
       "step": 805
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.50390625,
+      "grad_norm": 1.5234375,
       "learning_rate": 0.001,
-      "loss": 3.582,
+      "loss": 3.9032,
       "step": 810
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.515625,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 3.6892,
+      "loss": 3.8612,
       "step": 815
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.494140625,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 3.5234,
+      "loss": 3.8483,
       "step": 820
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.43359375,
+      "grad_norm": 1.65625,
       "learning_rate": 0.001,
-      "loss": 3.5884,
+      "loss": 3.8366,
       "step": 825
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.515625,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 3.541,
+      "loss": 3.8611,
       "step": 830
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.474609375,
+      "grad_norm": 1.4453125,
       "learning_rate": 0.001,
-      "loss": 3.4975,
+      "loss": 3.8658,
       "step": 835
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.462890625,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 3.5908,
+      "loss": 3.7546,
       "step": 840
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.4765625,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 3.5859,
+      "loss": 3.8955,
       "step": 845
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.41796875,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 3.4903,
+      "loss": 3.8909,
       "step": 850
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.49609375,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 3.4214,
+      "loss": 3.7381,
       "step": 855
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.455078125,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 3.4849,
+      "loss": 3.9318,
       "step": 860
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.451171875,
+      "grad_norm": 1.4296875,
       "learning_rate": 0.001,
-      "loss": 3.4287,
+      "loss": 3.8855,
       "step": 865
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.50390625,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 3.3918,
+      "loss": 3.9868,
       "step": 870
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.4921875,
+      "grad_norm": 1.421875,
       "learning_rate": 0.001,
-      "loss": 3.4713,
+      "loss": 3.837,
       "step": 875
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.51171875,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 3.4109,
+      "loss": 3.7942,
       "step": 880
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.5625,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 3.4379,
+      "loss": 3.7073,
       "step": 885
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.4921875,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 3.4546,
+      "loss": 3.7289,
       "step": 890
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.482421875,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 3.3923,
+      "loss": 3.7217,
       "step": 895
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.47265625,
+      "grad_norm": 1.5,
       "learning_rate": 0.001,
-      "loss": 3.4302,
+      "loss": 3.8807,
       "step": 900
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.494140625,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 3.4475,
+      "loss": 3.6495,
       "step": 905
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.51953125,
+      "grad_norm": 1.609375,
       "learning_rate": 0.001,
-      "loss": 3.3023,
+      "loss": 3.7232,
       "step": 910
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.439453125,
+      "grad_norm": 1.421875,
       "learning_rate": 0.001,
-      "loss": 3.3485,
+      "loss": 3.7907,
       "step": 915
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.4375,
+      "grad_norm": 1.4296875,
       "learning_rate": 0.001,
-      "loss": 3.3677,
+      "loss": 3.7654,
       "step": 920
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.52734375,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 3.5076,
+      "loss": 3.6906,
       "step": 925
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.53515625,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 3.421,
+      "loss": 3.7171,
       "step": 930
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.474609375,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 3.4751,
+      "loss": 3.7887,
       "step": 935
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.5234375,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 3.2864,
+      "loss": 3.841,
       "step": 940
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.4453125,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 3.3964,
+      "loss": 3.7338,
       "step": 945
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.51171875,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 3.367,
+      "loss": 3.7955,
       "step": 950
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.45703125,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 3.3819,
+      "loss": 3.7954,
       "step": 955
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.435546875,
+      "grad_norm": 1.421875,
       "learning_rate": 0.001,
-      "loss": 3.2415,
+      "loss": 3.6753,
       "step": 960
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.48828125,
+      "grad_norm": 1.6484375,
       "learning_rate": 0.001,
-      "loss": 3.2919,
+      "loss": 3.731,
       "step": 965
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.47265625,
+      "grad_norm": 1.5703125,
       "learning_rate": 0.001,
-      "loss": 3.4433,
+      "loss": 3.7734,
       "step": 970
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.470703125,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 3.3476,
+      "loss": 3.7521,
       "step": 975
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.451171875,
+      "grad_norm": 1.3984375,
       "learning_rate": 0.001,
-      "loss": 3.2323,
+      "loss": 3.786,
       "step": 980
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.5390625,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 3.3249,
+      "loss": 3.7505,
       "step": 985
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.51171875,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 3.215,
+      "loss": 3.6746,
       "step": 990
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.435546875,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 3.2893,
+      "loss": 3.6631,
       "step": 995
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.48828125,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 3.3386,
+      "loss": 3.6193,
       "step": 1000
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.4921875,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 3.2875,
+      "loss": 3.6175,
       "step": 1005
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.5234375,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 3.3014,
+      "loss": 3.8167,
       "step": 1010
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.458984375,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 3.176,
+      "loss": 3.7462,
       "step": 1015
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.46484375,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 3.2795,
+      "loss": 3.7337,
       "step": 1020
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.494140625,
+      "grad_norm": 1.4296875,
       "learning_rate": 0.001,
-      "loss": 3.2852,
+      "loss": 3.8234,
       "step": 1025
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.5703125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 3.278,
+      "loss": 3.7221,
       "step": 1030
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.54296875,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 3.2953,
+      "loss": 3.61,
       "step": 1035
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.482421875,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 3.217,
+      "loss": 3.696,
       "step": 1040
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.48046875,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 3.1406,
+      "loss": 3.6931,
       "step": 1045
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.439453125,
+      "grad_norm": 1.390625,
       "learning_rate": 0.001,
-      "loss": 3.2346,
+      "loss": 3.8107,
       "step": 1050
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.451171875,
+      "grad_norm": 1.796875,
       "learning_rate": 0.001,
-      "loss": 3.1984,
+      "loss": 3.7183,
       "step": 1055
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.49609375,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 3.2343,
+      "loss": 3.703,
       "step": 1060
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.455078125,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 3.1503,
+      "loss": 3.7271,
       "step": 1065
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.427734375,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 3.224,
+      "loss": 3.805,
       "step": 1070
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.43359375,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 3.1678,
+      "loss": 3.6972,
       "step": 1075
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.451171875,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 3.1891,
+      "loss": 3.6404,
       "step": 1080
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.515625,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 3.197,
+      "loss": 3.6821,
       "step": 1085
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.42578125,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 3.1608,
+      "loss": 3.718,
       "step": 1090
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.46484375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 3.2191,
+      "loss": 3.6238,
       "step": 1095
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.50390625,
+      "grad_norm": 1.53125,
       "learning_rate": 0.001,
-      "loss": 3.1926,
+      "loss": 3.727,
       "step": 1100
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.474609375,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 3.2302,
+      "loss": 3.6449,
       "step": 1105
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.466796875,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 3.1557,
+      "loss": 3.7375,
       "step": 1110
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.4453125,
+      "grad_norm": 1.4296875,
       "learning_rate": 0.001,
-      "loss": 3.1696,
+      "loss": 3.7099,
       "step": 1115
     },
     {
       "epoch": 0.12,
-      "grad_norm": 0.4921875,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 3.2021,
+      "loss": 3.7355,
       "step": 1120
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.4609375,
+      "grad_norm": 1.4375,
       "learning_rate": 0.001,
-      "loss": 3.1396,
+      "loss": 3.6202,
       "step": 1125
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.42578125,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 3.126,
+      "loss": 3.6891,
       "step": 1130
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.44921875,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 3.1153,
+      "loss": 3.6321,
       "step": 1135
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.46484375,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 3.2244,
+      "loss": 3.6889,
       "step": 1140
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.44140625,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 3.1862,
+      "loss": 3.6289,
       "step": 1145
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.458984375,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 3.1397,
+      "loss": 3.6676,
       "step": 1150
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.44921875,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 3.1137,
+      "loss": 3.5734,
       "step": 1155
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.3984375,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 3.0842,
+      "loss": 3.536,
       "step": 1160
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.4765625,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 3.1646,
+      "loss": 3.5922,
       "step": 1165
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.455078125,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 3.1085,
+      "loss": 3.5445,
       "step": 1170
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.431640625,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 3.1618,
+      "loss": 3.6754,
       "step": 1175
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.421875,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 3.1494,
+      "loss": 3.5503,
       "step": 1180
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.455078125,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 3.1187,
+      "loss": 3.6787,
       "step": 1185
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.4921875,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 3.1668,
+      "loss": 3.6661,
       "step": 1190
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.47265625,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 3.0082,
+      "loss": 3.629,
       "step": 1195
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.5234375,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 3.0856,
+      "loss": 3.6414,
       "step": 1200
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.43359375,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 3.0244,
+      "loss": 3.6187,
       "step": 1205
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.51953125,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 3.1119,
+      "loss": 3.6122,
       "step": 1210
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.423828125,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 3.1893,
+      "loss": 3.6192,
       "step": 1215
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.46484375,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 3.0317,
+      "loss": 3.6335,
       "step": 1220
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.498046875,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 3.0754,
+      "loss": 3.5721,
       "step": 1225
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.4921875,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 3.0712,
+      "loss": 3.5466,
       "step": 1230
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.443359375,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 3.0459,
+      "loss": 3.5322,
       "step": 1235
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.451171875,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 3.1248,
+      "loss": 3.5597,
       "step": 1240
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.515625,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 3.0001,
+      "loss": 3.583,
       "step": 1245
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.451171875,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 3.0746,
+      "loss": 3.6154,
       "step": 1250
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.4921875,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 3.0217,
+      "loss": 3.5325,
       "step": 1255
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.65234375,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 3.0922,
+      "loss": 3.5572,
       "step": 1260
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.478515625,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 3.0769,
+      "loss": 3.5091,
       "step": 1265
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.408203125,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 3.0214,
+      "loss": 3.6152,
       "step": 1270
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.42578125,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 3.0603,
+      "loss": 3.4866,
       "step": 1275
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.50390625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 3.0081,
+      "loss": 3.5305,
       "step": 1280
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.419921875,
+      "grad_norm": 1.6484375,
       "learning_rate": 0.001,
-      "loss": 3.1073,
+      "loss": 3.5229,
       "step": 1285
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.400390625,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 3.0161,
+      "loss": 3.4918,
       "step": 1290
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.4453125,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 3.0399,
+      "loss": 3.5985,
       "step": 1295
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.427734375,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 3.0176,
+      "loss": 3.467,
       "step": 1300
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.466796875,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 3.0438,
+      "loss": 3.5325,
       "step": 1305
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.45703125,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.9605,
+      "loss": 3.591,
       "step": 1310
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.490234375,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 3.1799,
+      "loss": 3.4826,
       "step": 1315
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.4453125,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 3.0367,
+      "loss": 3.4778,
       "step": 1320
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.48046875,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 3.0772,
+      "loss": 3.5079,
       "step": 1325
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.453125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.9758,
+      "loss": 3.5206,
       "step": 1330
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.435546875,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.9411,
+      "loss": 3.5359,
       "step": 1335
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.419921875,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 3.0461,
+      "loss": 3.5054,
       "step": 1340
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.466796875,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 3.0639,
+      "loss": 3.3876,
       "step": 1345
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.51171875,
+      "grad_norm": 1.5078125,
       "learning_rate": 0.001,
-      "loss": 3.0514,
+      "loss": 3.5013,
       "step": 1350
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.396484375,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.9288,
+      "loss": 3.4053,
       "step": 1355
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.44921875,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.9909,
+      "loss": 3.4865,
       "step": 1360
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.416015625,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 3.0404,
+      "loss": 3.556,
       "step": 1365
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.4296875,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.9636,
+      "loss": 3.4782,
       "step": 1370
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.44921875,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.9691,
+      "loss": 3.4029,
       "step": 1375
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.41015625,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 3.0421,
+      "loss": 3.5246,
       "step": 1380
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.453125,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 3.0106,
+      "loss": 3.4893,
       "step": 1385
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.408203125,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.9206,
+      "loss": 3.4819,
       "step": 1390
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.412109375,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.9851,
+      "loss": 3.5376,
       "step": 1395
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.427734375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.8973,
+      "loss": 3.4777,
       "step": 1400
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.4453125,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.9926,
+      "loss": 3.438,
       "step": 1405
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.40625,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.9713,
+      "loss": 3.4363,
       "step": 1410
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.5,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.9641,
+      "loss": 3.5214,
       "step": 1415
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.431640625,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.9165,
+      "loss": 3.5338,
       "step": 1420
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.42578125,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.9043,
+      "loss": 3.3541,
       "step": 1425
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.4140625,
+      "grad_norm": 1.4453125,
       "learning_rate": 0.001,
-      "loss": 3.0078,
+      "loss": 3.4375,
       "step": 1430
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.40625,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.9917,
+      "loss": 3.4896,
       "step": 1435
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.51953125,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.9517,
+      "loss": 3.5238,
       "step": 1440
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.431640625,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.9723,
+      "loss": 3.3979,
       "step": 1445
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.4140625,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.8912,
+      "loss": 3.3255,
       "step": 1450
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.439453125,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.997,
+      "loss": 3.4087,
       "step": 1455
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.419921875,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.8812,
+      "loss": 3.3471,
       "step": 1460
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.396484375,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.9449,
+      "loss": 3.3955,
       "step": 1465
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.400390625,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.8727,
+      "loss": 3.4651,
       "step": 1470
     },
     {
       "epoch": 0.16,
-      "grad_norm": 0.4296875,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.9621,
+      "loss": 3.4859,
       "step": 1475
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.3984375,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.8718,
+      "loss": 3.3995,
       "step": 1480
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.412109375,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.924,
+      "loss": 3.5291,
       "step": 1485
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.431640625,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.975,
+      "loss": 3.5589,
       "step": 1490
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.373046875,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.8352,
+      "loss": 3.3669,
       "step": 1495
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.4375,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.8661,
+      "loss": 3.373,
       "step": 1500
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.41796875,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.8946,
+      "loss": 3.4424,
       "step": 1505
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.42578125,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.8299,
+      "loss": 3.3791,
       "step": 1510
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.439453125,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.8354,
+      "loss": 3.4511,
       "step": 1515
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.455078125,
+      "grad_norm": 1.4375,
       "learning_rate": 0.001,
-      "loss": 2.8615,
+      "loss": 3.3719,
       "step": 1520
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.3984375,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.858,
+      "loss": 3.3859,
       "step": 1525
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.43359375,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.9946,
+      "loss": 3.414,
       "step": 1530
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.470703125,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.9449,
+      "loss": 3.4589,
       "step": 1535
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.447265625,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.8862,
+      "loss": 3.3673,
       "step": 1540
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.404296875,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.8575,
+      "loss": 3.399,
       "step": 1545
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.8596,
+      "loss": 3.3631,
       "step": 1550
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.419921875,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.9418,
+      "loss": 3.3376,
       "step": 1555
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.4140625,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.8957,
+      "loss": 3.2622,
       "step": 1560
     },
     {
       "epoch": 0.17,
-      "grad_norm": 0.4296875,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.8291,
+      "loss": 3.3692,
       "step": 1565
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.455078125,
+      "grad_norm": 1.3828125,
       "learning_rate": 0.001,
-      "loss": 2.8985,
+      "loss": 3.401,
       "step": 1570
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.439453125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.9283,
+      "loss": 3.3968,
       "step": 1575
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.416015625,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.9447,
+      "loss": 3.4257,
       "step": 1580
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.42578125,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.9154,
+      "loss": 3.3389,
       "step": 1585
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.470703125,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.9456,
+      "loss": 3.3529,
       "step": 1590
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.439453125,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.8641,
+      "loss": 3.4625,
       "step": 1595
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.41796875,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.996,
+      "loss": 3.3096,
       "step": 1600
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.431640625,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.9102,
+      "loss": 3.3768,
       "step": 1605
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.416015625,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.8548,
+      "loss": 3.2796,
       "step": 1610
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.423828125,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.8116,
+      "loss": 3.2988,
       "step": 1615
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.9482,
+      "loss": 3.4475,
       "step": 1620
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.4921875,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.8992,
+      "loss": 3.3677,
       "step": 1625
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.3984375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.8682,
+      "loss": 3.2559,
       "step": 1630
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.412109375,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.8866,
+      "loss": 3.2786,
       "step": 1635
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.421875,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.8674,
+      "loss": 3.3831,
       "step": 1640
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.451171875,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.8261,
+      "loss": 3.3864,
       "step": 1645
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.3984375,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 3.0015,
+      "loss": 3.3959,
       "step": 1650
     },
     {
       "epoch": 0.18,
-      "grad_norm": 0.4453125,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.859,
+      "loss": 3.2918,
       "step": 1655
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.451171875,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.8403,
+      "loss": 3.2775,
       "step": 1660
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.408203125,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.8866,
+      "loss": 3.3034,
       "step": 1665
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.8863,
+      "loss": 3.4116,
       "step": 1670
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.435546875,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.8138,
+      "loss": 3.3383,
       "step": 1675
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.390625,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.9055,
+      "loss": 3.3282,
       "step": 1680
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.494140625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.8598,
+      "loss": 3.2493,
       "step": 1685
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.8609,
+      "loss": 3.2698,
       "step": 1690
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.4765625,
       "learning_rate": 0.001,
-      "loss": 2.7287,
+      "loss": 3.2708,
       "step": 1695
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.43359375,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.8634,
+      "loss": 3.2881,
       "step": 1700
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.419921875,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.8631,
+      "loss": 3.3755,
       "step": 1705
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.43359375,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.8012,
+      "loss": 3.3396,
       "step": 1710
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.435546875,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.8285,
+      "loss": 3.3008,
       "step": 1715
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.439453125,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.8212,
+      "loss": 3.2302,
       "step": 1720
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.4140625,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.8448,
+      "loss": 3.3688,
       "step": 1725
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.453125,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.8296,
+      "loss": 3.2136,
       "step": 1730
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.412109375,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.8172,
+      "loss": 3.3421,
       "step": 1735
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.416015625,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.8287,
+      "loss": 3.2805,
       "step": 1740
     },
     {
       "epoch": 0.19,
-      "grad_norm": 0.396484375,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.8966,
+      "loss": 3.2745,
       "step": 1745
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.462890625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.7409,
+      "loss": 3.2147,
       "step": 1750
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.421875,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.8212,
+      "loss": 3.1843,
       "step": 1755
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.7865,
+      "loss": 3.2455,
       "step": 1760
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.423828125,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.8383,
+      "loss": 3.2143,
       "step": 1765
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.421875,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.8843,
+      "loss": 3.1009,
       "step": 1770
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.439453125,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.7954,
+      "loss": 3.2751,
       "step": 1775
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.404296875,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.7362,
+      "loss": 3.3078,
       "step": 1780
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.41796875,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.7698,
+      "loss": 3.2281,
       "step": 1785
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.4921875,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.9273,
+      "loss": 3.1855,
       "step": 1790
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.400390625,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.774,
+      "loss": 3.3936,
       "step": 1795
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.842,
+      "loss": 3.1809,
       "step": 1800
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.443359375,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.7766,
+      "loss": 3.2644,
       "step": 1805
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.462890625,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.7165,
+      "loss": 3.4201,
       "step": 1810
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.427734375,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.8396,
+      "loss": 3.2126,
       "step": 1815
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.4140625,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.7133,
+      "loss": 3.3431,
       "step": 1820
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.3984375,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.7195,
+      "loss": 3.3002,
       "step": 1825
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.47265625,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.7769,
+      "loss": 3.3268,
       "step": 1830
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.416015625,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.8127,
+      "loss": 3.2517,
       "step": 1835
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.435546875,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.7074,
+      "loss": 3.3158,
       "step": 1840
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.380859375,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.8092,
+      "loss": 3.1521,
       "step": 1845
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.4375,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.8355,
+      "loss": 3.1654,
       "step": 1850
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.412109375,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.7602,
+      "loss": 3.198,
       "step": 1855
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.7212,
+      "loss": 3.1798,
       "step": 1860
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.8304,
+      "loss": 3.192,
       "step": 1865
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.8164,
+      "loss": 3.2478,
       "step": 1870
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.40625,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.8169,
+      "loss": 3.165,
       "step": 1875
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.7487,
+      "loss": 3.2028,
       "step": 1880
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.7734,
+      "loss": 3.2514,
       "step": 1885
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.4453125,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.7757,
+      "loss": 3.2128,
       "step": 1890
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.8023,
+      "loss": 3.1707,
       "step": 1895
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.7523,
+      "loss": 3.2027,
       "step": 1900
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.390625,
+      "grad_norm": 1.640625,
       "learning_rate": 0.001,
-      "loss": 2.818,
+      "loss": 3.3155,
       "step": 1905
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.41015625,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.7938,
+      "loss": 3.1595,
       "step": 1910
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.41015625,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.7914,
+      "loss": 3.2946,
       "step": 1915
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.431640625,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.7482,
+      "loss": 3.2708,
       "step": 1920
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.474609375,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.7364,
+      "loss": 3.2351,
       "step": 1925
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.7383,
+      "loss": 3.2298,
       "step": 1930
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.416015625,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 2.7172,
+      "loss": 3.2229,
       "step": 1935
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.6953,
+      "loss": 3.2835,
       "step": 1940
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.41796875,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 2.6664,
+      "loss": 3.258,
       "step": 1945
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.41796875,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.6742,
+      "loss": 3.3006,
       "step": 1950
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.44140625,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.7493,
+      "loss": 3.1846,
       "step": 1955
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.384765625,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.7087,
+      "loss": 3.2121,
       "step": 1960
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.400390625,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.7275,
+      "loss": 3.0832,
       "step": 1965
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.4296875,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.7522,
+      "loss": 3.2398,
       "step": 1970
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.41796875,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.7288,
+      "loss": 3.1761,
       "step": 1975
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.7241,
+      "loss": 3.1958,
       "step": 1980
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.7156,
+      "loss": 3.2419,
       "step": 1985
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.435546875,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.7864,
+      "loss": 3.1544,
       "step": 1990
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.46875,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.7317,
+      "loss": 3.1974,
       "step": 1995
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.4453125,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.6977,
+      "loss": 3.0909,
       "step": 2000
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.392578125,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.7876,
+      "loss": 3.1775,
       "step": 2005
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.380859375,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.8518,
+      "loss": 3.0954,
       "step": 2010
     },
     {
       "epoch": 0.22,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.6466,
+      "loss": 3.1065,
       "step": 2015
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.384765625,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.8142,
+      "loss": 3.2287,
       "step": 2020
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.6646,
+      "loss": 3.1186,
       "step": 2025
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.421875,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.7533,
+      "loss": 3.2063,
       "step": 2030
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.384765625,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.7672,
+      "loss": 3.2042,
       "step": 2035
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.421875,
       "learning_rate": 0.001,
-      "loss": 2.6538,
+      "loss": 3.208,
       "step": 2040
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.5546875,
       "learning_rate": 0.001,
-      "loss": 2.7903,
+      "loss": 3.203,
       "step": 2045
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.3828125,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.6225,
+      "loss": 3.0422,
       "step": 2050
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.369140625,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.6761,
+      "loss": 3.3197,
       "step": 2055
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.7514,
+      "loss": 3.2007,
       "step": 2060
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.3984375,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.7695,
+      "loss": 3.1827,
       "step": 2065
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.41015625,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.6763,
+      "loss": 3.1037,
       "step": 2070
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.7686,
+      "loss": 3.0695,
       "step": 2075
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.6337,
+      "loss": 3.1399,
       "step": 2080
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.42578125,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.7801,
+      "loss": 3.1754,
       "step": 2085
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.7185,
+      "loss": 3.0863,
       "step": 2090
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.390625,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.7413,
+      "loss": 3.1698,
       "step": 2095
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.6984,
+      "loss": 3.2883,
       "step": 2100
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.7253,
+      "loss": 3.1698,
       "step": 2105
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.408203125,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.7485,
+      "loss": 3.2341,
       "step": 2110
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.373046875,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.6853,
+      "loss": 3.0373,
       "step": 2115
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.6931,
+      "loss": 3.1708,
       "step": 2120
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.6173,
+      "loss": 3.1345,
       "step": 2125
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.421875,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.624,
+      "loss": 3.0521,
       "step": 2130
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.3984375,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 2.6629,
+      "loss": 3.1135,
       "step": 2135
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.7097,
+      "loss": 3.2268,
       "step": 2140
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.7286,
+      "loss": 3.0748,
       "step": 2145
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.404296875,
+      "grad_norm": 1.6015625,
       "learning_rate": 0.001,
-      "loss": 2.7264,
+      "loss": 3.0793,
       "step": 2150
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.47265625,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 2.6504,
+      "loss": 3.0799,
       "step": 2155
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.392578125,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.7058,
+      "loss": 3.0898,
       "step": 2160
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.365234375,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.7127,
+      "loss": 3.1049,
       "step": 2165
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.6349,
+      "loss": 3.0061,
       "step": 2170
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.375,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.7502,
+      "loss": 3.1097,
       "step": 2175
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.7021,
+      "loss": 3.1246,
       "step": 2180
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.470703125,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.6564,
+      "loss": 3.1433,
       "step": 2185
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.3828125,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.6715,
+      "loss": 3.0436,
       "step": 2190
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.421875,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.7572,
+      "loss": 3.1603,
       "step": 2195
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.6954,
+      "loss": 3.0944,
       "step": 2200
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.41015625,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.7125,
+      "loss": 3.1839,
       "step": 2205
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.408203125,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.6772,
+      "loss": 3.0935,
       "step": 2210
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.8311,
+      "loss": 3.1054,
       "step": 2215
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.4140625,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.6858,
+      "loss": 3.0938,
       "step": 2220
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.44140625,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.6552,
+      "loss": 3.0771,
       "step": 2225
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.380859375,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.7589,
+      "loss": 3.0439,
       "step": 2230
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.6547,
+      "loss": 3.1046,
       "step": 2235
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.6727,
+      "loss": 3.1784,
       "step": 2240
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.7218,
+      "loss": 3.1278,
       "step": 2245
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.439453125,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.6825,
+      "loss": 3.0594,
       "step": 2250
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.7252,
+      "loss": 3.1073,
       "step": 2255
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.65625,
       "learning_rate": 0.001,
-      "loss": 2.7026,
+      "loss": 3.0362,
       "step": 2260
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3828125,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.7449,
+      "loss": 3.0847,
       "step": 2265
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.5876,
+      "loss": 2.9977,
       "step": 2270
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.373046875,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.6584,
+      "loss": 3.0647,
       "step": 2275
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.4140625,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.6238,
+      "loss": 3.1407,
       "step": 2280
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.41015625,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.7556,
+      "loss": 2.9501,
       "step": 2285
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.6804,
+      "loss": 2.9796,
       "step": 2290
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.6864,
+      "loss": 2.9868,
       "step": 2295
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.369140625,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.5995,
+      "loss": 3.1029,
       "step": 2300
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.41796875,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.6537,
+      "loss": 3.0329,
       "step": 2305
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.419921875,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.6645,
+      "loss": 3.0854,
       "step": 2310
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.7136,
+      "loss": 3.0628,
       "step": 2315
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.6908,
+      "loss": 3.1078,
       "step": 2320
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.7102,
+      "loss": 3.0313,
       "step": 2325
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.50390625,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.7089,
+      "loss": 3.0872,
       "step": 2330
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.4140625,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.6295,
+      "loss": 2.9973,
       "step": 2335
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.384765625,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.6105,
+      "loss": 2.9909,
       "step": 2340
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.41796875,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.6332,
+      "loss": 2.958,
       "step": 2345
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.41015625,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.6549,
+      "loss": 2.9788,
       "step": 2350
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.7258,
+      "loss": 3.0805,
       "step": 2355
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.4453125,
       "learning_rate": 0.001,
-      "loss": 2.6932,
+      "loss": 2.981,
       "step": 2360
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.6171875,
       "learning_rate": 0.001,
-      "loss": 2.5853,
+      "loss": 3.0405,
       "step": 2365
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.6075,
+      "loss": 3.0477,
       "step": 2370
     },
     {
       "epoch": 0.26,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.7695,
+      "loss": 2.9396,
       "step": 2375
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.675,
+      "loss": 3.0082,
       "step": 2380
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.6556,
+      "loss": 2.948,
       "step": 2385
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.390625,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.6259,
+      "loss": 3.0341,
       "step": 2390
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.627,
+      "loss": 2.9754,
       "step": 2395
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.4453125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.6976,
+      "loss": 3.0006,
       "step": 2400
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.6732,
+      "loss": 2.9661,
       "step": 2405
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.6387,
+      "loss": 2.9024,
       "step": 2410
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.613,
+      "loss": 2.9043,
       "step": 2415
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.6863,
+      "loss": 3.1154,
       "step": 2420
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.384765625,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.6637,
+      "loss": 3.0356,
       "step": 2425
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.5994,
+      "loss": 2.9301,
       "step": 2430
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.6521,
+      "loss": 2.932,
       "step": 2435
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.423828125,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.6928,
+      "loss": 2.9737,
       "step": 2440
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.408203125,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.6497,
+      "loss": 2.9865,
       "step": 2445
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.412109375,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.6095,
+      "loss": 2.9408,
       "step": 2450
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.637,
+      "loss": 2.8823,
       "step": 2455
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.400390625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.5598,
+      "loss": 2.9784,
       "step": 2460
     },
     {
       "epoch": 0.27,
-      "grad_norm": 0.419921875,
+      "grad_norm": 2.25,
       "learning_rate": 0.001,
-      "loss": 2.5683,
+      "loss": 2.9849,
       "step": 2465
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.7179,
+      "loss": 3.0205,
       "step": 2470
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.373046875,
+      "grad_norm": 1.53125,
       "learning_rate": 0.001,
-      "loss": 2.5267,
+      "loss": 2.9721,
       "step": 2475
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.390625,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.643,
+      "loss": 2.9403,
       "step": 2480
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.390625,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.5632,
+      "loss": 2.9458,
       "step": 2485
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.611,
+      "loss": 2.993,
       "step": 2490
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.572,
+      "loss": 2.9178,
       "step": 2495
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.3984375,
       "learning_rate": 0.001,
-      "loss": 2.6059,
+      "loss": 3.0433,
       "step": 2500
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.384765625,
+      "grad_norm": 1.453125,
       "learning_rate": 0.001,
-      "loss": 2.5368,
+      "loss": 2.8434,
       "step": 2505
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.447265625,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.6391,
+      "loss": 2.8175,
       "step": 2510
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.6533,
+      "loss": 2.9316,
       "step": 2515
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.40625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.6177,
+      "loss": 2.8653,
       "step": 2520
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.5711,
+      "loss": 2.8296,
       "step": 2525
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.392578125,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.6115,
+      "loss": 2.7935,
       "step": 2530
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.384765625,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.674,
+      "loss": 2.9531,
       "step": 2535
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.6826,
+      "loss": 2.8647,
       "step": 2540
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.4140625,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.5226,
+      "loss": 2.9229,
       "step": 2545
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.6192,
+      "loss": 2.8533,
       "step": 2550
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.6914,
+      "loss": 2.906,
       "step": 2555
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.6802,
+      "loss": 2.9979,
       "step": 2560
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.41015625,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.5906,
+      "loss": 2.9295,
       "step": 2565
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.365234375,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.5617,
+      "loss": 2.8835,
       "step": 2570
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.5559,
+      "loss": 2.8187,
       "step": 2575
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.484375,
       "learning_rate": 0.001,
-      "loss": 2.5432,
+      "loss": 2.8908,
       "step": 2580
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.3984375,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.6188,
+      "loss": 2.9334,
       "step": 2585
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.375,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.594,
+      "loss": 2.8633,
       "step": 2590
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.4375,
       "learning_rate": 0.001,
-      "loss": 2.6327,
+      "loss": 2.8475,
       "step": 2595
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.396484375,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.6117,
+      "loss": 2.9979,
       "step": 2600
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.6003,
+      "loss": 2.8673,
       "step": 2605
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.5405,
+      "loss": 2.8746,
       "step": 2610
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.365234375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.5711,
+      "loss": 2.9179,
       "step": 2615
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.6618,
+      "loss": 2.9361,
       "step": 2620
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.400390625,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.63,
+      "loss": 2.8913,
       "step": 2625
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.404296875,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.5909,
+      "loss": 2.8556,
       "step": 2630
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.5804,
+      "loss": 2.8247,
       "step": 2635
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.392578125,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.6332,
+      "loss": 2.9346,
       "step": 2640
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.6762,
+      "loss": 2.8897,
       "step": 2645
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.390625,
+      "grad_norm": 1.5703125,
       "learning_rate": 0.001,
-      "loss": 2.5429,
+      "loss": 2.9438,
       "step": 2650
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.6548,
+      "loss": 2.7655,
       "step": 2655
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.5452,
+      "loss": 2.8996,
       "step": 2660
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.400390625,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.6768,
+      "loss": 2.9045,
       "step": 2665
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.6133,
+      "loss": 2.8311,
       "step": 2670
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.365234375,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.6124,
+      "loss": 2.9492,
       "step": 2675
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.5989,
+      "loss": 2.8869,
       "step": 2680
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.5802,
+      "loss": 2.8418,
       "step": 2685
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.5502,
+      "loss": 2.9442,
       "step": 2690
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.416015625,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.5251,
+      "loss": 2.8688,
       "step": 2695
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.6167,
+      "loss": 2.9113,
       "step": 2700
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.5672,
+      "loss": 2.8833,
       "step": 2705
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.3828125,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.5343,
+      "loss": 2.9694,
       "step": 2710
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.5582,
+      "loss": 2.8323,
       "step": 2715
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.5223,
+      "loss": 2.8322,
       "step": 2720
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.5679,
+      "loss": 2.8532,
       "step": 2725
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.5854,
+      "loss": 2.8846,
       "step": 2730
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.375,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.5077,
+      "loss": 2.7953,
       "step": 2735
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.5738,
+      "loss": 2.7917,
       "step": 2740
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.5829,
+      "loss": 2.7999,
       "step": 2745
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.5914,
+      "loss": 2.7832,
       "step": 2750
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.5735,
+      "loss": 2.7757,
       "step": 2755
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.400390625,
+      "grad_norm": 1.6015625,
       "learning_rate": 0.001,
-      "loss": 2.5015,
+      "loss": 2.9216,
       "step": 2760
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.6339,
+      "loss": 2.7279,
       "step": 2765
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.396484375,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.6208,
+      "loss": 2.8547,
       "step": 2770
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.373046875,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.5872,
+      "loss": 2.7974,
       "step": 2775
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.5484,
+      "loss": 2.7325,
       "step": 2780
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.6061,
+      "loss": 2.8651,
       "step": 2785
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.3828125,
       "learning_rate": 0.001,
-      "loss": 2.7287,
+      "loss": 2.8027,
       "step": 2790
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.4569,
+      "loss": 2.8023,
       "step": 2795
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.365234375,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.5641,
+      "loss": 2.7645,
       "step": 2800
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.5877,
+      "loss": 2.8216,
       "step": 2805
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.40234375,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.5578,
+      "loss": 2.7154,
       "step": 2810
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.5764,
+      "loss": 2.7438,
       "step": 2815
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.365234375,
+      "grad_norm": 1.4921875,
       "learning_rate": 0.001,
-      "loss": 2.5233,
+      "loss": 2.7664,
       "step": 2820
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.5845,
+      "loss": 2.8217,
       "step": 2825
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.6143,
+      "loss": 2.7378,
       "step": 2830
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.5839,
+      "loss": 2.8457,
       "step": 2835
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.5518,
+      "loss": 2.7347,
       "step": 2840
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.5763,
+      "loss": 2.7388,
       "step": 2845
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.365234375,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.5789,
+      "loss": 2.8177,
       "step": 2850
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.3984375,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.6161,
+      "loss": 2.8578,
       "step": 2855
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.4979,
+      "loss": 2.7455,
       "step": 2860
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.462890625,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.5239,
+      "loss": 2.7682,
       "step": 2865
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.369140625,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.5278,
+      "loss": 2.8223,
       "step": 2870
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.5839,
+      "loss": 2.8354,
       "step": 2875
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.5451,
+      "loss": 2.8103,
       "step": 2880
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.369140625,
+      "grad_norm": 1.515625,
       "learning_rate": 0.001,
-      "loss": 2.5832,
+      "loss": 2.8094,
       "step": 2885
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.6994,
+      "loss": 2.7705,
       "step": 2890
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.5226,
+      "loss": 2.7731,
       "step": 2895
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.568,
+      "loss": 2.7048,
       "step": 2900
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.3984375,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.5504,
+      "loss": 2.7777,
       "step": 2905
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.3984375,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.5429,
+      "loss": 2.8811,
       "step": 2910
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.5131,
+      "loss": 2.7433,
       "step": 2915
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.5571,
+      "loss": 2.7328,
       "step": 2920
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.40625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.63,
+      "loss": 2.7672,
       "step": 2925
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.369140625,
+      "grad_norm": 1.3828125,
       "learning_rate": 0.001,
-      "loss": 2.5114,
+      "loss": 2.6969,
       "step": 2930
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.587,
+      "loss": 2.6923,
       "step": 2935
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.6734,
+      "loss": 2.7048,
       "step": 2940
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.6092,
+      "loss": 2.7554,
       "step": 2945
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.4977,
+      "loss": 2.7421,
       "step": 2950
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.5451,
+      "loss": 2.7542,
       "step": 2955
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.390625,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.6318,
+      "loss": 2.7626,
       "step": 2960
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.3828125,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.489,
+      "loss": 2.7957,
       "step": 2965
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.4592,
+      "loss": 2.7428,
       "step": 2970
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.392578125,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.5569,
+      "loss": 2.7071,
       "step": 2975
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.380859375,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.4976,
+      "loss": 2.7833,
       "step": 2980
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.5439,
+      "loss": 2.7262,
       "step": 2985
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.5567,
+      "loss": 2.706,
       "step": 2990
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.5471,
+      "loss": 2.7063,
       "step": 2995
     },
     {
       "epoch": 0.33,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.5934,
+      "loss": 2.7755,
       "step": 3000
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.527,
+      "loss": 2.7543,
       "step": 3005
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.5382,
+      "loss": 2.7099,
       "step": 3010
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.5225,
+      "loss": 2.6922,
       "step": 3015
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.373046875,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.571,
+      "loss": 2.7727,
       "step": 3020
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.5245,
+      "loss": 2.7885,
       "step": 3025
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.5326,
+      "loss": 2.8368,
       "step": 3030
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.5773,
+      "loss": 2.6819,
       "step": 3035
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.5171,
+      "loss": 2.6766,
       "step": 3040
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.5531,
+      "loss": 2.734,
       "step": 3045
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.5284,
+      "loss": 2.8116,
       "step": 3050
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.369140625,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.5586,
+      "loss": 2.6718,
       "step": 3055
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.5146,
+      "loss": 2.6894,
       "step": 3060
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.5981,
+      "loss": 2.6376,
       "step": 3065
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.5523,
+      "loss": 2.8011,
       "step": 3070
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.5508,
+      "loss": 2.7382,
       "step": 3075
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.5361,
+      "loss": 2.7586,
       "step": 3080
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.396484375,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.6033,
+      "loss": 2.8034,
       "step": 3085
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.408203125,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.6071,
+      "loss": 2.7853,
       "step": 3090
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.5747,
+      "loss": 2.6979,
       "step": 3095
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.4875,
+      "loss": 2.6383,
       "step": 3100
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.5199,
+      "loss": 2.7292,
       "step": 3105
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.5829,
+      "loss": 2.6234,
       "step": 3110
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.4627,
+      "loss": 2.7024,
       "step": 3115
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.375,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.5592,
+      "loss": 2.7145,
       "step": 3120
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.6057,
+      "loss": 2.6933,
       "step": 3125
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.380859375,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.538,
+      "loss": 2.6224,
       "step": 3130
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.384765625,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.5633,
+      "loss": 2.7334,
       "step": 3135
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.3828125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.5373,
+      "loss": 2.6673,
       "step": 3140
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.4773,
+      "loss": 2.7149,
       "step": 3145
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.41015625,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.5214,
+      "loss": 2.5578,
       "step": 3150
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.6875,
       "learning_rate": 0.001,
-      "loss": 2.498,
+      "loss": 2.6527,
       "step": 3155
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.4634,
+      "loss": 2.6242,
       "step": 3160
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.4578,
+      "loss": 2.5558,
       "step": 3165
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.546875,
       "learning_rate": 0.001,
-      "loss": 2.5839,
+      "loss": 2.7399,
       "step": 3170
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.40625,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.5751,
+      "loss": 2.8314,
       "step": 3175
     },
     {
       "epoch": 0.35,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.4664,
+      "loss": 2.7,
       "step": 3180
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.4624,
+      "loss": 2.7248,
       "step": 3185
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.42578125,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.5735,
+      "loss": 2.6502,
       "step": 3190
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.5191,
+      "loss": 2.692,
       "step": 3195
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.408203125,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.4841,
+      "loss": 2.7619,
       "step": 3200
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.5304,
+      "loss": 2.7757,
       "step": 3205
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.4461,
+      "loss": 2.6067,
       "step": 3210
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.384765625,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.5183,
+      "loss": 2.6181,
       "step": 3215
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.369140625,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.481,
+      "loss": 2.5756,
       "step": 3220
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.6264,
+      "loss": 2.6381,
       "step": 3225
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.544,
+      "loss": 2.7008,
       "step": 3230
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.5118,
+      "loss": 2.7283,
       "step": 3235
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.365234375,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.5537,
+      "loss": 2.599,
       "step": 3240
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.5551,
+      "loss": 2.6984,
       "step": 3245
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.4869,
+      "loss": 2.6424,
       "step": 3250
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.5788,
+      "loss": 2.6042,
       "step": 3255
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.5389,
+      "loss": 2.7193,
       "step": 3260
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.4296875,
       "learning_rate": 0.001,
-      "loss": 2.5014,
+      "loss": 2.6789,
       "step": 3265
     },
     {
       "epoch": 0.36,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.5224,
+      "loss": 2.6459,
       "step": 3270
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.4611,
+      "loss": 2.6203,
       "step": 3275
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.5234375,
       "learning_rate": 0.001,
-      "loss": 2.5188,
+      "loss": 2.6347,
       "step": 3280
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.5296,
+      "loss": 2.7011,
       "step": 3285
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.5364,
+      "loss": 2.6418,
       "step": 3290
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.5207,
+      "loss": 2.6523,
       "step": 3295
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.5104,
+      "loss": 2.567,
       "step": 3300
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.513,
+      "loss": 2.5959,
       "step": 3305
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.523,
+      "loss": 2.6507,
       "step": 3310
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.4677,
+      "loss": 2.5627,
       "step": 3315
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.528,
+      "loss": 2.6199,
       "step": 3320
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.373046875,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.5352,
+      "loss": 2.6042,
       "step": 3325
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.4736,
+      "loss": 2.6501,
       "step": 3330
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.5234,
+      "loss": 2.6254,
       "step": 3335
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.4768,
+      "loss": 2.611,
       "step": 3340
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.5709,
+      "loss": 2.5959,
       "step": 3345
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.4996,
+      "loss": 2.5988,
       "step": 3350
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.4553,
+      "loss": 2.5914,
       "step": 3355
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.5061,
+      "loss": 2.5807,
       "step": 3360
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.4686,
+      "loss": 2.5726,
       "step": 3365
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.7890625,
       "learning_rate": 0.001,
-      "loss": 2.5281,
+      "loss": 2.6163,
       "step": 3370
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.5412,
+      "loss": 2.7316,
       "step": 3375
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.4708,
+      "loss": 2.5764,
       "step": 3380
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.9609375,
       "learning_rate": 0.001,
-      "loss": 2.4491,
+      "loss": 2.619,
       "step": 3385
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.453125,
       "learning_rate": 0.001,
-      "loss": 2.4432,
+      "loss": 2.7615,
       "step": 3390
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.5601,
+      "loss": 2.6547,
       "step": 3395
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.4834,
+      "loss": 2.5934,
       "step": 3400
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.5377,
+      "loss": 2.5967,
       "step": 3405
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.90625,
       "learning_rate": 0.001,
-      "loss": 2.4399,
+      "loss": 2.599,
       "step": 3410
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.4765625,
       "learning_rate": 0.001,
-      "loss": 2.4336,
+      "loss": 2.6839,
       "step": 3415
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.5348,
+      "loss": 2.5772,
       "step": 3420
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 2.4751,
+      "loss": 2.6305,
       "step": 3425
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.39453125,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.4556,
+      "loss": 2.5354,
       "step": 3430
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.426,
+      "loss": 2.5099,
       "step": 3435
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.4736,
+      "loss": 2.575,
       "step": 3440
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.4077,
+      "loss": 2.5827,
       "step": 3445
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.5495,
+      "loss": 2.6243,
       "step": 3450
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.380859375,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.4385,
+      "loss": 2.6192,
       "step": 3455
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.369140625,
+      "grad_norm": 1.484375,
       "learning_rate": 0.001,
-      "loss": 2.5411,
+      "loss": 2.5388,
       "step": 3460
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.4391,
+      "loss": 2.587,
       "step": 3465
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.8671875,
       "learning_rate": 0.001,
-      "loss": 2.5512,
+      "loss": 2.5446,
       "step": 3470
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.5625,
       "learning_rate": 0.001,
-      "loss": 2.5375,
+      "loss": 2.6233,
       "step": 3475
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.8359375,
       "learning_rate": 0.001,
-      "loss": 2.4875,
+      "loss": 2.6057,
       "step": 3480
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.5244,
+      "loss": 2.5322,
       "step": 3485
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.5694,
+      "loss": 2.6265,
       "step": 3490
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.5128,
+      "loss": 2.6747,
       "step": 3495
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.578125,
       "learning_rate": 0.001,
-      "loss": 2.4621,
+      "loss": 2.5641,
       "step": 3500
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.353515625,
+      "grad_norm": 2.34375,
       "learning_rate": 0.001,
-      "loss": 2.3882,
+      "loss": 2.643,
       "step": 3505
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.6484375,
       "learning_rate": 0.001,
-      "loss": 2.4457,
+      "loss": 2.5575,
       "step": 3510
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.6640625,
       "learning_rate": 0.001,
-      "loss": 2.6129,
+      "loss": 2.6074,
       "step": 3515
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.4673,
+      "loss": 2.5276,
       "step": 3520
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.40625,
       "learning_rate": 0.001,
-      "loss": 2.4322,
+      "loss": 2.5163,
       "step": 3525
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.5703125,
       "learning_rate": 0.001,
-      "loss": 2.4767,
+      "loss": 2.5769,
       "step": 3530
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.5015,
+      "loss": 2.5597,
       "step": 3535
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.4496,
+      "loss": 2.5999,
       "step": 3540
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.4416,
+      "loss": 2.5962,
       "step": 3545
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.4924,
+      "loss": 2.5758,
       "step": 3550
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.5,
       "learning_rate": 0.001,
-      "loss": 2.444,
+      "loss": 2.6153,
       "step": 3555
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.5554,
+      "loss": 2.5332,
       "step": 3560
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.5222,
+      "loss": 2.6169,
       "step": 3565
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.4904,
+      "loss": 2.5641,
       "step": 3570
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.369140625,
+      "grad_norm": 1.6484375,
       "learning_rate": 0.001,
-      "loss": 2.51,
+      "loss": 2.5202,
       "step": 3575
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.491,
+      "loss": 2.5156,
       "step": 3580
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.5569,
+      "loss": 2.485,
       "step": 3585
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.4702,
+      "loss": 2.557,
       "step": 3590
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.421875,
       "learning_rate": 0.001,
-      "loss": 2.4147,
+      "loss": 2.6137,
       "step": 3595
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.5733,
+      "loss": 2.5028,
       "step": 3600
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.4938,
+      "loss": 2.5713,
       "step": 3605
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.4076,
+      "loss": 2.5385,
       "step": 3610
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.75,
       "learning_rate": 0.001,
-      "loss": 2.4868,
+      "loss": 2.568,
       "step": 3615
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.4921875,
       "learning_rate": 0.001,
-      "loss": 2.4626,
+      "loss": 2.5535,
       "step": 3620
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.5284,
+      "loss": 2.4861,
       "step": 3625
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.4889,
+      "loss": 2.5154,
       "step": 3630
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.3828125,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 2.4571,
+      "loss": 2.5672,
       "step": 3635
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.4296875,
       "learning_rate": 0.001,
-      "loss": 2.5282,
+      "loss": 2.5576,
       "step": 3640
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.4765625,
       "learning_rate": 0.001,
-      "loss": 2.5391,
+      "loss": 2.5735,
       "step": 3645
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.5301,
+      "loss": 2.5436,
       "step": 3650
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.369140625,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.4556,
+      "loss": 2.512,
       "step": 3655
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.4623,
+      "loss": 2.4986,
       "step": 3660
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.421875,
       "learning_rate": 0.001,
-      "loss": 2.5289,
+      "loss": 2.5719,
       "step": 3665
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.373046875,
+      "grad_norm": 1.4140625,
       "learning_rate": 0.001,
-      "loss": 2.4913,
+      "loss": 2.558,
       "step": 3670
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.4986,
+      "loss": 2.6109,
       "step": 3675
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.4453125,
       "learning_rate": 0.001,
-      "loss": 2.5109,
+      "loss": 2.4702,
       "step": 3680
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 2.5675,
+      "loss": 2.5414,
       "step": 3685
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.404296875,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.5055,
+      "loss": 2.4311,
       "step": 3690
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.40625,
       "learning_rate": 0.001,
-      "loss": 2.4677,
+      "loss": 2.5505,
       "step": 3695
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.5347,
+      "loss": 2.5599,
       "step": 3700
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.4140625,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 2.4168,
+      "loss": 2.5399,
       "step": 3705
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.5,
       "learning_rate": 0.001,
-      "loss": 2.4156,
+      "loss": 2.5745,
       "step": 3710
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.7109375,
       "learning_rate": 0.001,
-      "loss": 2.5117,
+      "loss": 2.5591,
       "step": 3715
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.765625,
       "learning_rate": 0.001,
-      "loss": 2.5008,
+      "loss": 2.5645,
       "step": 3720
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.5078125,
       "learning_rate": 0.001,
-      "loss": 2.5145,
+      "loss": 2.5349,
       "step": 3725
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.4302,
+      "loss": 2.64,
       "step": 3730
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.326171875,
+      "grad_norm": 2.546875,
       "learning_rate": 0.001,
-      "loss": 2.5553,
+      "loss": 2.5241,
       "step": 3735
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.3984375,
       "learning_rate": 0.001,
-      "loss": 2.4049,
+      "loss": 2.5465,
       "step": 3740
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.5078125,
       "learning_rate": 0.001,
-      "loss": 2.4319,
+      "loss": 2.548,
       "step": 3745
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.5425,
+      "loss": 2.5006,
       "step": 3750
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.5042,
+      "loss": 2.4411,
       "step": 3755
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.4445,
+      "loss": 2.6081,
       "step": 3760
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.4256,
+      "loss": 2.5661,
       "step": 3765
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.6953125,
       "learning_rate": 0.001,
-      "loss": 2.4933,
+      "loss": 2.4901,
       "step": 3770
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.37109375,
+      "grad_norm": 1.4921875,
       "learning_rate": 0.001,
-      "loss": 2.4679,
+      "loss": 2.5783,
       "step": 3775
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.4482,
+      "loss": 2.5633,
       "step": 3780
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.4219,
+      "loss": 2.4383,
       "step": 3785
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.3828125,
       "learning_rate": 0.001,
-      "loss": 2.5463,
+      "loss": 2.5552,
       "step": 3790
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.4565,
+      "loss": 2.4974,
       "step": 3795
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 2.4338,
+      "loss": 2.4294,
       "step": 3800
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.3828125,
       "learning_rate": 0.001,
-      "loss": 2.4545,
+      "loss": 2.5585,
       "step": 3805
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.53125,
       "learning_rate": 0.001,
-      "loss": 2.4431,
+      "loss": 2.4823,
       "step": 3810
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.5105,
+      "loss": 2.4983,
       "step": 3815
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.390625,
       "learning_rate": 0.001,
-      "loss": 2.4994,
+      "loss": 2.4712,
       "step": 3820
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.373046875,
+      "grad_norm": 2.6875,
       "learning_rate": 0.001,
-      "loss": 2.4036,
+      "loss": 2.5071,
       "step": 3825
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.365234375,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.3943,
+      "loss": 2.378,
       "step": 3830
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.4016,
+      "loss": 2.5846,
       "step": 3835
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.3733,
+      "loss": 2.4814,
       "step": 3840
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.373046875,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.4746,
+      "loss": 2.5205,
       "step": 3845
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.453125,
       "learning_rate": 0.001,
-      "loss": 2.4641,
+      "loss": 2.6069,
       "step": 3850
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.4296875,
       "learning_rate": 0.001,
-      "loss": 2.4163,
+      "loss": 2.6136,
       "step": 3855
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.5641,
+      "loss": 2.5675,
       "step": 3860
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.365234375,
+      "grad_norm": 2.34375,
       "learning_rate": 0.001,
-      "loss": 2.4459,
+      "loss": 2.4644,
       "step": 3865
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.48,
+      "loss": 2.5886,
       "step": 3870
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.384765625,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.3711,
+      "loss": 2.6005,
       "step": 3875
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.484375,
       "learning_rate": 0.001,
-      "loss": 2.4004,
+      "loss": 2.5263,
       "step": 3880
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.3915,
+      "loss": 2.6189,
       "step": 3885
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.3828125,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.4762,
+      "loss": 2.4413,
       "step": 3890
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.5390625,
       "learning_rate": 0.001,
-      "loss": 2.4994,
+      "loss": 2.517,
       "step": 3895
     },
     {
       "epoch": 0.43,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.4486,
+      "loss": 2.5185,
       "step": 3900
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.484375,
       "learning_rate": 0.001,
-      "loss": 2.4192,
+      "loss": 2.5422,
       "step": 3905
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.4359,
+      "loss": 2.4782,
       "step": 3910
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.4987,
+      "loss": 2.5178,
       "step": 3915
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.390625,
       "learning_rate": 0.001,
-      "loss": 2.494,
+      "loss": 2.5289,
       "step": 3920
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.6953125,
       "learning_rate": 0.001,
-      "loss": 2.4081,
+      "loss": 2.5569,
       "step": 3925
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.5546875,
       "learning_rate": 0.001,
-      "loss": 2.4449,
+      "loss": 2.5402,
       "step": 3930
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.625,
       "learning_rate": 0.001,
-      "loss": 2.4008,
+      "loss": 2.4138,
       "step": 3935
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.515625,
       "learning_rate": 0.001,
-      "loss": 2.4596,
+      "loss": 2.4729,
       "step": 3940
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.333984375,
+      "grad_norm": 2.125,
       "learning_rate": 0.001,
-      "loss": 2.4376,
+      "loss": 2.4512,
       "step": 3945
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.4765,
+      "loss": 2.3337,
       "step": 3950
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.345703125,
+      "grad_norm": 2.96875,
       "learning_rate": 0.001,
-      "loss": 2.4726,
+      "loss": 2.4999,
       "step": 3955
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.5,
       "learning_rate": 0.001,
-      "loss": 2.4462,
+      "loss": 2.4464,
       "step": 3960
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.4140625,
       "learning_rate": 0.001,
-      "loss": 2.3776,
+      "loss": 2.4477,
       "step": 3965
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.3864,
+      "loss": 2.4136,
       "step": 3970
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.8046875,
       "learning_rate": 0.001,
-      "loss": 2.4396,
+      "loss": 2.5954,
       "step": 3975
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.4140625,
       "learning_rate": 0.001,
-      "loss": 2.5275,
+      "loss": 2.582,
       "step": 3980
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 2.5021,
+      "loss": 2.4916,
       "step": 3985
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.5053,
+      "loss": 2.4851,
       "step": 3990
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.4765625,
       "learning_rate": 0.001,
-      "loss": 2.4067,
+      "loss": 2.5241,
       "step": 3995
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.3718,
+      "loss": 2.4568,
       "step": 4000
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.396,
+      "loss": 2.4604,
       "step": 4005
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.4531,
+      "loss": 2.4243,
       "step": 4010
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.4581,
+      "loss": 2.4554,
       "step": 4015
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.4616,
+      "loss": 2.4236,
       "step": 4020
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.369140625,
+      "grad_norm": 2.125,
       "learning_rate": 0.001,
-      "loss": 2.4446,
+      "loss": 2.4336,
       "step": 4025
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.3241,
+      "loss": 2.4906,
       "step": 4030
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.5,
       "learning_rate": 0.001,
-      "loss": 2.5278,
+      "loss": 2.4008,
       "step": 4035
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.4402,
+      "loss": 2.4638,
       "step": 4040
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.375,
+      "grad_norm": 1.5703125,
       "learning_rate": 0.001,
-      "loss": 2.526,
+      "loss": 2.4525,
       "step": 4045
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.4471,
+      "loss": 2.4088,
       "step": 4050
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.4765625,
       "learning_rate": 0.001,
-      "loss": 2.5027,
+      "loss": 2.5329,
       "step": 4055
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.5,
       "learning_rate": 0.001,
-      "loss": 2.4513,
+      "loss": 2.5568,
       "step": 4060
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.40625,
       "learning_rate": 0.001,
-      "loss": 2.4194,
+      "loss": 2.4684,
       "step": 4065
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.4691,
+      "loss": 2.4425,
       "step": 4070
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.4777,
+      "loss": 2.4634,
       "step": 4075
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.3957,
+      "loss": 2.5238,
       "step": 4080
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.4746,
+      "loss": 2.4865,
       "step": 4085
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.3968,
+      "loss": 2.551,
       "step": 4090
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.4296875,
       "learning_rate": 0.001,
-      "loss": 2.4137,
+      "loss": 2.4611,
       "step": 4095
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.5123,
+      "loss": 2.5396,
       "step": 4100
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.4194,
+      "loss": 2.4379,
       "step": 4105
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.7734375,
       "learning_rate": 0.001,
-      "loss": 2.3577,
+      "loss": 2.4577,
       "step": 4110
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.6796875,
       "learning_rate": 0.001,
-      "loss": 2.4128,
+      "loss": 2.5041,
       "step": 4115
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.40625,
       "learning_rate": 0.001,
-      "loss": 2.4063,
+      "loss": 2.4483,
       "step": 4120
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.37890625,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.4535,
+      "loss": 2.5035,
       "step": 4125
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.3966,
+      "loss": 2.5613,
       "step": 4130
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.390625,
       "learning_rate": 0.001,
-      "loss": 2.4123,
+      "loss": 2.4445,
       "step": 4135
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.45703125,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.4634,
+      "loss": 2.6129,
       "step": 4140
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.8984375,
       "learning_rate": 0.001,
-      "loss": 2.3939,
+      "loss": 2.3973,
       "step": 4145
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.3828125,
       "learning_rate": 0.001,
-      "loss": 2.3884,
+      "loss": 2.3411,
       "step": 4150
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.4213,
+      "loss": 2.5052,
       "step": 4155
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.8359375,
       "learning_rate": 0.001,
-      "loss": 2.3715,
+      "loss": 2.4608,
       "step": 4160
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.5625,
       "learning_rate": 0.001,
-      "loss": 2.4166,
+      "loss": 2.5404,
       "step": 4165
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.5234375,
       "learning_rate": 0.001,
-      "loss": 2.4946,
+      "loss": 2.4894,
       "step": 4170
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.4918,
+      "loss": 2.4287,
       "step": 4175
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.6875,
       "learning_rate": 0.001,
-      "loss": 2.381,
+      "loss": 2.4712,
       "step": 4180
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.421875,
       "learning_rate": 0.001,
-      "loss": 2.3991,
+      "loss": 2.411,
       "step": 4185
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.4139,
+      "loss": 2.449,
       "step": 4190
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.59375,
       "learning_rate": 0.001,
-      "loss": 2.5107,
+      "loss": 2.5091,
       "step": 4195
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.875,
       "learning_rate": 0.001,
-      "loss": 2.387,
+      "loss": 2.4494,
       "step": 4200
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.447,
+      "loss": 2.4689,
       "step": 4205
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.3473,
+      "loss": 2.5923,
       "step": 4210
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.3907,
+      "loss": 2.4636,
       "step": 4215
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.3656,
+      "loss": 2.4478,
       "step": 4220
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.5121,
+      "loss": 2.4567,
       "step": 4225
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.4212,
+      "loss": 2.3894,
       "step": 4230
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.3811,
+      "loss": 2.4179,
       "step": 4235
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.368,
+      "loss": 2.5042,
       "step": 4240
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.4631,
+      "loss": 2.4784,
       "step": 4245
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.4267,
+      "loss": 2.4666,
       "step": 4250
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.390625,
       "learning_rate": 0.001,
-      "loss": 2.4453,
+      "loss": 2.3777,
       "step": 4255
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.3846,
+      "loss": 2.4888,
       "step": 4260
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.3578,
+      "loss": 2.4483,
       "step": 4265
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.4811,
+      "loss": 2.3399,
       "step": 4270
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.4921875,
       "learning_rate": 0.001,
-      "loss": 2.4718,
+      "loss": 2.4195,
       "step": 4275
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.453125,
       "learning_rate": 0.001,
-      "loss": 2.4631,
+      "loss": 2.4564,
       "step": 4280
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.428,
+      "loss": 2.4967,
       "step": 4285
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.3984375,
       "learning_rate": 0.001,
-      "loss": 2.4018,
+      "loss": 2.44,
       "step": 4290
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.7421875,
       "learning_rate": 0.001,
-      "loss": 2.3948,
+      "loss": 2.4448,
       "step": 4295
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.3828125,
       "learning_rate": 0.001,
-      "loss": 2.3939,
+      "loss": 2.4057,
       "step": 4300
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.3984375,
       "learning_rate": 0.001,
-      "loss": 2.3406,
+      "loss": 2.443,
       "step": 4305
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.384765625,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.427,
+      "loss": 2.4622,
       "step": 4310
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.84375,
       "learning_rate": 0.001,
-      "loss": 2.4806,
+      "loss": 2.3848,
       "step": 4315
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.3565,
+      "loss": 2.509,
       "step": 4320
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.341796875,
+      "grad_norm": 4.0,
       "learning_rate": 0.001,
-      "loss": 2.4658,
+      "loss": 2.5265,
       "step": 4325
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.4375,
       "learning_rate": 0.001,
-      "loss": 2.3383,
+      "loss": 2.5115,
       "step": 4330
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.4433,
+      "loss": 2.4607,
       "step": 4335
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.4428,
+      "loss": 2.5261,
       "step": 4340
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.3914,
+      "loss": 2.419,
       "step": 4345
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.4283,
+      "loss": 2.4465,
       "step": 4350
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.4238,
+      "loss": 2.4207,
       "step": 4355
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.3536,
+      "loss": 2.4445,
       "step": 4360
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.3306,
+      "loss": 2.5083,
       "step": 4365
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.369140625,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.4716,
+      "loss": 2.4347,
       "step": 4370
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.3698,
+      "loss": 2.4745,
       "step": 4375
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.4516,
+      "loss": 2.515,
       "step": 4380
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.3544,
+      "loss": 2.5217,
       "step": 4385
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.4579,
+      "loss": 2.4455,
       "step": 4390
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.4471,
+      "loss": 2.4412,
       "step": 4395
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.34765625,
+      "grad_norm": 4.4375,
       "learning_rate": 0.001,
-      "loss": 2.4688,
+      "loss": 2.4647,
       "step": 4400
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.4272,
+      "loss": 2.4476,
       "step": 4405
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.3917,
+      "loss": 2.4614,
       "step": 4410
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.4167,
+      "loss": 2.5262,
       "step": 4415
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.4342,
+      "loss": 2.4971,
       "step": 4420
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.5076,
+      "loss": 2.4713,
       "step": 4425
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.6015625,
       "learning_rate": 0.001,
-      "loss": 2.4897,
+      "loss": 2.3779,
       "step": 4430
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.349609375,
+      "grad_norm": 2.03125,
       "learning_rate": 0.001,
-      "loss": 2.4231,
+      "loss": 2.4566,
       "step": 4435
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.358,
+      "loss": 2.5437,
       "step": 4440
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.59375,
       "learning_rate": 0.001,
-      "loss": 2.3515,
+      "loss": 2.4165,
       "step": 4445
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.4139,
+      "loss": 2.4083,
       "step": 4450
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.5031,
+      "loss": 2.3645,
       "step": 4455
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.3742,
+      "loss": 2.4619,
       "step": 4460
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.467,
+      "loss": 2.4517,
       "step": 4465
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.3063,
+      "loss": 2.4964,
       "step": 4470
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.2742,
+      "loss": 2.3903,
       "step": 4475
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.390625,
+      "grad_norm": 1.6328125,
       "learning_rate": 0.001,
-      "loss": 2.3993,
+      "loss": 2.4048,
       "step": 4480
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.6328125,
       "learning_rate": 0.001,
-      "loss": 2.3711,
+      "loss": 2.5493,
       "step": 4485
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.3366,
+      "loss": 2.4204,
       "step": 4490
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.359375,
+      "grad_norm": 5.0625,
       "learning_rate": 0.001,
-      "loss": 2.3711,
+      "loss": 2.4851,
       "step": 4495
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.4112,
+      "loss": 2.4582,
       "step": 4500
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.5269,
+      "loss": 2.3493,
       "step": 4505
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.4256,
+      "loss": 2.4725,
       "step": 4510
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3359375,
+      "grad_norm": 2.078125,
       "learning_rate": 0.001,
-      "loss": 2.4714,
+      "loss": 2.4007,
       "step": 4515
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.65625,
       "learning_rate": 0.001,
-      "loss": 2.4565,
+      "loss": 2.4246,
       "step": 4520
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.4071,
+      "loss": 2.4925,
       "step": 4525
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.9375,
       "learning_rate": 0.001,
-      "loss": 2.4281,
+      "loss": 2.3759,
       "step": 4530
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.4609375,
       "learning_rate": 0.001,
-      "loss": 2.4384,
+      "loss": 2.431,
       "step": 4535
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.4587,
+      "loss": 2.387,
       "step": 4540
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.4184,
+      "loss": 2.5083,
       "step": 4545
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.3867,
+      "loss": 2.3847,
       "step": 4550
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.326171875,
+      "grad_norm": 2.125,
       "learning_rate": 0.001,
-      "loss": 2.3673,
+      "loss": 2.4038,
       "step": 4555
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.310546875,
+      "grad_norm": 4.1875,
       "learning_rate": 0.001,
-      "loss": 2.4,
+      "loss": 2.3835,
       "step": 4560
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.4027,
+      "loss": 2.4496,
       "step": 4565
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.4282,
+      "loss": 2.4231,
       "step": 4570
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.4324,
+      "loss": 2.3516,
       "step": 4575
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.4159,
+      "loss": 2.3787,
       "step": 4580
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.3382,
+      "loss": 2.331,
       "step": 4585
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.3968,
+      "loss": 2.4552,
       "step": 4590
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.3331,
+      "loss": 2.3459,
       "step": 4595
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.4229,
+      "loss": 2.4763,
       "step": 4600
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.37109375,
+      "grad_norm": 2.265625,
       "learning_rate": 0.001,
-      "loss": 2.3908,
+      "loss": 2.3367,
       "step": 4605
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.4211,
+      "loss": 2.4216,
       "step": 4610
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.4101,
+      "loss": 2.4127,
       "step": 4615
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.3875,
+      "loss": 2.5296,
       "step": 4620
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.3807,
+      "loss": 2.4159,
       "step": 4625
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.53125,
       "learning_rate": 0.001,
-      "loss": 2.3716,
+      "loss": 2.4337,
       "step": 4630
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.4305,
+      "loss": 2.3075,
       "step": 4635
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.4326,
+      "loss": 2.3824,
       "step": 4640
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.4278,
+      "loss": 2.4537,
       "step": 4645
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.6484375,
       "learning_rate": 0.001,
-      "loss": 2.3428,
+      "loss": 2.4326,
       "step": 4650
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.3823,
+      "loss": 2.4517,
       "step": 4655
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.4197,
+      "loss": 2.4543,
       "step": 4660
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.4177,
+      "loss": 2.5044,
       "step": 4665
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.2897,
+      "loss": 2.3716,
       "step": 4670
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.3867,
+      "loss": 2.4272,
       "step": 4675
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.2943,
+      "loss": 2.4517,
       "step": 4680
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.3359375,
+      "grad_norm": 0.99609375,
       "learning_rate": 0.001,
-      "loss": 2.4413,
+      "loss": 2.3564,
       "step": 4685
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.3828,
+      "loss": 2.4594,
       "step": 4690
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.4453125,
       "learning_rate": 0.001,
-      "loss": 2.451,
+      "loss": 2.4573,
       "step": 4695
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.3077,
+      "loss": 2.4016,
       "step": 4700
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.5546875,
       "learning_rate": 0.001,
-      "loss": 2.4278,
+      "loss": 2.4329,
       "step": 4705
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.371,
+      "loss": 2.4285,
       "step": 4710
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.429,
+      "loss": 2.4041,
       "step": 4715
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.3383,
+      "loss": 2.4431,
       "step": 4720
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.8359375,
       "learning_rate": 0.001,
-      "loss": 2.3408,
+      "loss": 2.4893,
       "step": 4725
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.388671875,
+      "grad_norm": 1.8671875,
       "learning_rate": 0.001,
-      "loss": 2.4862,
+      "loss": 2.4273,
       "step": 4730
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.5457,
+      "loss": 2.3858,
       "step": 4735
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.4219,
+      "loss": 2.4082,
       "step": 4740
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.277,
+      "loss": 2.3506,
       "step": 4745
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.671875,
       "learning_rate": 0.001,
-      "loss": 2.3562,
+      "loss": 2.3004,
       "step": 4750
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.4333,
+      "loss": 2.4124,
       "step": 4755
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.3555,
+      "loss": 2.3558,
       "step": 4760
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.365234375,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.4844,
+      "loss": 2.394,
       "step": 4765
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.435546875,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.4931,
+      "loss": 2.3589,
       "step": 4770
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.3959,
+      "loss": 2.3722,
       "step": 4775
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.3608,
+      "loss": 2.3893,
       "step": 4780
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.3661,
+      "loss": 2.386,
       "step": 4785
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.4261,
+      "loss": 2.4663,
       "step": 4790
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.3362,
+      "loss": 2.4395,
       "step": 4795
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.4043,
+      "loss": 2.4552,
       "step": 4800
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.3257,
+      "loss": 2.3771,
       "step": 4805
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.4065,
+      "loss": 2.4676,
       "step": 4810
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.3205,
+      "loss": 2.4427,
       "step": 4815
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.4849,
+      "loss": 2.3732,
       "step": 4820
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.373046875,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.4492,
+      "loss": 2.3805,
       "step": 4825
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.3474,
+      "loss": 2.3883,
       "step": 4830
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.3716,
+      "loss": 2.4687,
       "step": 4835
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.4609375,
       "learning_rate": 0.001,
-      "loss": 2.3561,
+      "loss": 2.4263,
       "step": 4840
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.3908,
+      "loss": 2.3938,
       "step": 4845
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.4253,
+      "loss": 2.371,
       "step": 4850
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.484375,
       "learning_rate": 0.001,
-      "loss": 2.3876,
+      "loss": 2.4461,
       "step": 4855
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.3442,
+      "loss": 2.4477,
       "step": 4860
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.3125,
+      "grad_norm": 2.078125,
       "learning_rate": 0.001,
-      "loss": 2.3322,
+      "loss": 2.2894,
       "step": 4865
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.3931,
+      "loss": 2.3719,
       "step": 4870
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.4097,
+      "loss": 2.4017,
       "step": 4875
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.3871,
+      "loss": 2.3104,
       "step": 4880
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.383,
+      "loss": 2.4409,
       "step": 4885
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.375,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 2.3096,
+      "loss": 2.3999,
       "step": 4890
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.890625,
       "learning_rate": 0.001,
-      "loss": 2.3468,
+      "loss": 2.3703,
       "step": 4895
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.339,
+      "loss": 2.3998,
       "step": 4900
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.365234375,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.3719,
+      "loss": 2.3721,
       "step": 4905
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.3645,
+      "loss": 2.3626,
       "step": 4910
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.5703125,
       "learning_rate": 0.001,
-      "loss": 2.3295,
+      "loss": 2.4011,
       "step": 4915
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 2.4452,
+      "loss": 2.3683,
       "step": 4920
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.3908,
+      "loss": 2.3509,
       "step": 4925
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.3639,
+      "loss": 2.2919,
       "step": 4930
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.32421875,
+      "grad_norm": 2.140625,
       "learning_rate": 0.001,
-      "loss": 2.3453,
+      "loss": 2.304,
       "step": 4935
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.318359375,
+      "grad_norm": 0.91015625,
       "learning_rate": 0.001,
-      "loss": 2.4583,
+      "loss": 2.3685,
       "step": 4940
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.5078125,
       "learning_rate": 0.001,
-      "loss": 2.3326,
+      "loss": 2.3315,
       "step": 4945
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 2.3368,
+      "loss": 2.3417,
       "step": 4950
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.330078125,
+      "grad_norm": 2.09375,
       "learning_rate": 0.001,
-      "loss": 2.3238,
+      "loss": 2.5319,
       "step": 4955
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.4453125,
       "learning_rate": 0.001,
-      "loss": 2.3857,
+      "loss": 2.3706,
       "step": 4960
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 2.4269,
+      "loss": 2.3488,
       "step": 4965
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.4069,
+      "loss": 2.343,
       "step": 4970
     },
     {
       "epoch": 0.55,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.7734375,
       "learning_rate": 0.001,
-      "loss": 2.4184,
+      "loss": 2.339,
       "step": 4975
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.3375,
+      "loss": 2.3289,
       "step": 4980
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.326171875,
+      "grad_norm": 2.453125,
       "learning_rate": 0.001,
-      "loss": 2.3276,
+      "loss": 2.3635,
       "step": 4985
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.4921875,
       "learning_rate": 0.001,
-      "loss": 2.387,
+      "loss": 2.3701,
       "step": 4990
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.341796875,
+      "grad_norm": 0.99609375,
       "learning_rate": 0.001,
-      "loss": 2.4152,
+      "loss": 2.3865,
       "step": 4995
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.0,
       "learning_rate": 0.001,
-      "loss": 2.4344,
+      "loss": 2.3312,
       "step": 5000
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.40625,
       "learning_rate": 0.001,
-      "loss": 2.3746,
+      "loss": 2.3723,
       "step": 5005
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.0078125,
       "learning_rate": 0.001,
-      "loss": 2.3692,
+      "loss": 2.3693,
       "step": 5010
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.3626,
+      "loss": 2.4211,
       "step": 5015
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.4002,
+      "loss": 2.3302,
       "step": 5020
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.30859375,
+      "grad_norm": 4.03125,
       "learning_rate": 0.001,
-      "loss": 2.4398,
+      "loss": 2.4063,
       "step": 5025
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.2824,
+      "loss": 2.4358,
       "step": 5030
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.5703125,
       "learning_rate": 0.001,
-      "loss": 2.3933,
+      "loss": 2.327,
       "step": 5035
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.3513,
+      "loss": 2.2614,
       "step": 5040
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.3421,
+      "loss": 2.2946,
       "step": 5045
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.328125,
+      "grad_norm": 2.328125,
       "learning_rate": 0.001,
-      "loss": 2.3887,
+      "loss": 2.4384,
       "step": 5050
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.3859,
+      "loss": 2.3465,
       "step": 5055
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.3026,
+      "loss": 2.355,
       "step": 5060
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.326,
+      "loss": 2.3374,
       "step": 5065
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.3422,
+      "loss": 2.3319,
       "step": 5070
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.3424,
+      "loss": 2.4058,
       "step": 5075
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.339,
+      "loss": 2.3335,
       "step": 5080
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.4453125,
       "learning_rate": 0.001,
-      "loss": 2.3462,
+      "loss": 2.4275,
       "step": 5085
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.3583,
+      "loss": 2.2337,
       "step": 5090
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.7109375,
       "learning_rate": 0.001,
-      "loss": 2.4476,
+      "loss": 2.472,
       "step": 5095
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.3964,
+      "loss": 2.3425,
       "step": 5100
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.3881,
+      "loss": 2.3226,
       "step": 5105
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.3989,
+      "loss": 2.3202,
       "step": 5110
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.3513,
+      "loss": 2.4078,
       "step": 5115
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.4226,
+      "loss": 2.3514,
       "step": 5120
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.384765625,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.318,
+      "loss": 2.3719,
       "step": 5125
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.55078125,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.3471,
+      "loss": 2.3483,
       "step": 5130
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.3504,
+      "loss": 2.3145,
       "step": 5135
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.3703,
+      "loss": 2.3698,
       "step": 5140
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.356,
+      "loss": 2.4254,
       "step": 5145
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.3881,
+      "loss": 2.3652,
       "step": 5150
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.3372,
+      "loss": 2.3388,
       "step": 5155
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.2871,
+      "loss": 2.3005,
       "step": 5160
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.4192,
+      "loss": 2.343,
       "step": 5165
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.3266,
+      "loss": 2.4356,
       "step": 5170
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.3311,
+      "loss": 2.3843,
       "step": 5175
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.8984375,
       "learning_rate": 0.001,
-      "loss": 2.3426,
+      "loss": 2.3718,
       "step": 5180
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.314453125,
+      "grad_norm": 2.078125,
       "learning_rate": 0.001,
-      "loss": 2.3464,
+      "loss": 2.2947,
       "step": 5185
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.6796875,
       "learning_rate": 0.001,
-      "loss": 2.4051,
+      "loss": 2.3222,
       "step": 5190
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.3669,
+      "loss": 2.4169,
       "step": 5195
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.4088,
+      "loss": 2.3647,
       "step": 5200
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.3046875,
+      "grad_norm": 3.78125,
       "learning_rate": 0.001,
-      "loss": 2.3947,
+      "loss": 2.3263,
       "step": 5205
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.5859375,
       "learning_rate": 0.001,
-      "loss": 2.3573,
+      "loss": 2.363,
       "step": 5210
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.4141,
+      "loss": 2.2993,
       "step": 5215
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.3164,
+      "loss": 2.339,
       "step": 5220
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.3358,
+      "loss": 2.331,
       "step": 5225
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.324,
+      "loss": 2.3842,
       "step": 5230
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.4203,
+      "loss": 2.4073,
       "step": 5235
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.4609375,
       "learning_rate": 0.001,
-      "loss": 2.3093,
+      "loss": 2.3942,
       "step": 5240
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.3203125,
+      "grad_norm": 2.25,
       "learning_rate": 0.001,
-      "loss": 2.3323,
+      "loss": 2.4361,
       "step": 5245
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.484375,
       "learning_rate": 0.001,
-      "loss": 2.3769,
+      "loss": 2.2972,
       "step": 5250
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.421875,
       "learning_rate": 0.001,
-      "loss": 2.3383,
+      "loss": 2.2457,
       "step": 5255
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.298828125,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.2656,
+      "loss": 2.3928,
       "step": 5260
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.9375,
       "learning_rate": 0.001,
-      "loss": 2.2729,
+      "loss": 2.3553,
       "step": 5265
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.3279,
+      "loss": 2.308,
       "step": 5270
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.8125,
       "learning_rate": 0.001,
-      "loss": 2.4237,
+      "loss": 2.4344,
       "step": 5275
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.3641,
+      "loss": 2.345,
       "step": 5280
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.3602,
+      "loss": 2.3255,
       "step": 5285
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.385,
+      "loss": 2.3048,
       "step": 5290
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.3023,
+      "loss": 2.3411,
       "step": 5295
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.361328125,
+      "grad_norm": 0.98046875,
       "learning_rate": 0.001,
-      "loss": 2.3724,
+      "loss": 2.208,
       "step": 5300
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.3638,
+      "loss": 2.2552,
       "step": 5305
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.609375,
       "learning_rate": 0.001,
-      "loss": 2.3663,
+      "loss": 2.2885,
       "step": 5310
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.365234375,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.4862,
+      "loss": 2.3439,
       "step": 5315
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.3922,
+      "loss": 2.3595,
       "step": 5320
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.3677,
+      "loss": 2.2758,
       "step": 5325
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.5078125,
       "learning_rate": 0.001,
-      "loss": 2.3606,
+      "loss": 2.3548,
       "step": 5330
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.3088,
+      "loss": 2.4227,
       "step": 5335
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.4073,
+      "loss": 2.394,
       "step": 5340
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.3641,
+      "loss": 2.3507,
       "step": 5345
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.3953,
+      "loss": 2.382,
       "step": 5350
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.9375,
       "learning_rate": 0.001,
-      "loss": 2.3133,
+      "loss": 2.3912,
       "step": 5355
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.3236,
+      "loss": 2.3112,
       "step": 5360
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.3984,
+      "loss": 2.408,
       "step": 5365
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.3896,
+      "loss": 2.2344,
       "step": 5370
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.4453125,
       "learning_rate": 0.001,
-      "loss": 2.3215,
+      "loss": 2.4095,
       "step": 5375
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.4614,
+      "loss": 2.3205,
       "step": 5380
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.3136,
+      "loss": 2.3861,
       "step": 5385
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.3548,
+      "loss": 2.3335,
       "step": 5390
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.7265625,
       "learning_rate": 0.001,
-      "loss": 2.3609,
+      "loss": 2.3337,
       "step": 5395
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.84375,
       "learning_rate": 0.001,
-      "loss": 2.3043,
+      "loss": 2.3939,
       "step": 5400
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.3837,
+      "loss": 2.3017,
       "step": 5405
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.4236,
+      "loss": 2.2953,
       "step": 5410
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.3492,
+      "loss": 2.3509,
       "step": 5415
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.4296875,
       "learning_rate": 0.001,
-      "loss": 2.3594,
+      "loss": 2.3585,
       "step": 5420
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.4765625,
       "learning_rate": 0.001,
-      "loss": 2.4392,
+      "loss": 2.2923,
       "step": 5425
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.640625,
       "learning_rate": 0.001,
-      "loss": 2.3757,
+      "loss": 2.3309,
       "step": 5430
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.5,
       "learning_rate": 0.001,
-      "loss": 2.3445,
+      "loss": 2.3466,
       "step": 5435
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.4049,
+      "loss": 2.2919,
       "step": 5440
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.546875,
       "learning_rate": 0.001,
-      "loss": 2.4535,
+      "loss": 2.3037,
       "step": 5445
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.6640625,
       "learning_rate": 0.001,
-      "loss": 2.3116,
+      "loss": 2.3836,
       "step": 5450
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.640625,
       "learning_rate": 0.001,
-      "loss": 2.3645,
+      "loss": 2.3335,
       "step": 5455
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.3229,
+      "loss": 2.3305,
       "step": 5460
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.3648,
+      "loss": 2.2483,
       "step": 5465
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.4006,
+      "loss": 2.31,
       "step": 5470
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.357421875,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.3397,
+      "loss": 2.3389,
       "step": 5475
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.3543,
+      "loss": 2.3941,
       "step": 5480
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.3028,
+      "loss": 2.3019,
       "step": 5485
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.3388,
+      "loss": 2.2751,
       "step": 5490
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.3664,
+      "loss": 2.3538,
       "step": 5495
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.3046875,
+      "grad_norm": 2.53125,
       "learning_rate": 0.001,
-      "loss": 2.2125,
+      "loss": 2.3456,
       "step": 5500
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.4234,
+      "loss": 2.3096,
       "step": 5505
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.4765625,
       "learning_rate": 0.001,
-      "loss": 2.3316,
+      "loss": 2.3386,
       "step": 5510
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.367,
+      "loss": 2.328,
       "step": 5515
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.5,
       "learning_rate": 0.001,
-      "loss": 2.4614,
+      "loss": 2.2578,
       "step": 5520
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.2681,
+      "loss": 2.2215,
       "step": 5525
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.3821,
+      "loss": 2.3991,
       "step": 5530
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.3998,
+      "loss": 2.2228,
       "step": 5535
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.3163,
+      "loss": 2.259,
       "step": 5540
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.3876,
+      "loss": 2.2526,
       "step": 5545
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.3965,
+      "loss": 2.3117,
       "step": 5550
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.382,
+      "loss": 2.3628,
       "step": 5555
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.2978,
+      "loss": 2.3557,
       "step": 5560
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.3291,
+      "loss": 2.2213,
       "step": 5565
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.3339,
+      "loss": 2.3814,
       "step": 5570
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.3565,
+      "loss": 2.3071,
       "step": 5575
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.3237,
+      "loss": 2.2928,
       "step": 5580
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.4315,
+      "loss": 2.2477,
       "step": 5585
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.231,
+      "loss": 2.4124,
       "step": 5590
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.3903,
+      "loss": 2.2786,
       "step": 5595
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.6015625,
       "learning_rate": 0.001,
-      "loss": 2.2526,
+      "loss": 2.3296,
       "step": 5600
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.3466,
+      "loss": 2.3383,
       "step": 5605
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.3012,
+      "loss": 2.3517,
       "step": 5610
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.4253,
+      "loss": 2.2402,
       "step": 5615
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.2762,
+      "loss": 2.315,
       "step": 5620
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.2793,
+      "loss": 2.3727,
       "step": 5625
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.421875,
       "learning_rate": 0.001,
-      "loss": 2.3468,
+      "loss": 2.3322,
       "step": 5630
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.4084,
+      "loss": 2.3768,
       "step": 5635
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.347,
+      "loss": 2.3349,
       "step": 5640
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.8359375,
       "learning_rate": 0.001,
-      "loss": 2.3839,
+      "loss": 2.3919,
       "step": 5645
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.3001,
+      "loss": 2.4263,
       "step": 5650
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.2755,
+      "loss": 2.3149,
       "step": 5655
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.2986,
+      "loss": 2.383,
       "step": 5660
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.3897,
+      "loss": 2.4073,
       "step": 5665
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.3523,
+      "loss": 2.277,
       "step": 5670
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.3523,
+      "loss": 2.3267,
       "step": 5675
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.3354,
+      "loss": 2.283,
       "step": 5680
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.3529,
+      "loss": 2.3412,
       "step": 5685
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.546875,
       "learning_rate": 0.001,
-      "loss": 2.357,
+      "loss": 2.3126,
       "step": 5690
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.333984375,
+      "grad_norm": 2.046875,
       "learning_rate": 0.001,
-      "loss": 2.368,
+      "loss": 2.2937,
       "step": 5695
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.32421875,
+      "grad_norm": 0.9296875,
       "learning_rate": 0.001,
-      "loss": 2.3112,
+      "loss": 2.2517,
       "step": 5700
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.5859375,
       "learning_rate": 0.001,
-      "loss": 2.3848,
+      "loss": 2.3529,
       "step": 5705
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.3793,
+      "loss": 2.3374,
       "step": 5710
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.3272,
+      "loss": 2.3249,
       "step": 5715
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.4029,
+      "loss": 2.2751,
       "step": 5720
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.3996,
+      "loss": 2.2705,
       "step": 5725
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.329,
+      "loss": 2.2959,
       "step": 5730
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.3392,
+      "loss": 2.3365,
       "step": 5735
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.0078125,
       "learning_rate": 0.001,
-      "loss": 2.3056,
+      "loss": 2.3377,
       "step": 5740
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.3413,
+      "loss": 2.387,
       "step": 5745
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.3949,
+      "loss": 2.3485,
       "step": 5750
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.438,
+      "loss": 2.3841,
       "step": 5755
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.4453,
+      "loss": 2.2726,
       "step": 5760
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.7265625,
       "learning_rate": 0.001,
-      "loss": 2.3045,
+      "loss": 2.2472,
       "step": 5765
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.2906,
+      "loss": 2.3245,
       "step": 5770
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.294921875,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.2219,
+      "loss": 2.3735,
       "step": 5775
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.32421875,
+      "grad_norm": 0.96484375,
       "learning_rate": 0.001,
-      "loss": 2.3613,
+      "loss": 2.3062,
       "step": 5780
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.3199,
+      "loss": 2.3034,
       "step": 5785
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.3443,
+      "loss": 2.3342,
       "step": 5790
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.2965,
+      "loss": 2.2772,
       "step": 5795
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.4453125,
       "learning_rate": 0.001,
-      "loss": 2.2862,
+      "loss": 2.3663,
       "step": 5800
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.40625,
       "learning_rate": 0.001,
-      "loss": 2.3733,
+      "loss": 2.298,
       "step": 5805
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.2991,
+      "loss": 2.3009,
       "step": 5810
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.3591,
+      "loss": 2.2918,
       "step": 5815
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.3919,
+      "loss": 2.2476,
       "step": 5820
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.0078125,
       "learning_rate": 0.001,
-      "loss": 2.2326,
+      "loss": 2.2734,
       "step": 5825
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.341796875,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.3035,
+      "loss": 2.2312,
       "step": 5830
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.3228,
+      "loss": 2.3424,
       "step": 5835
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.3327,
+      "loss": 2.2396,
       "step": 5840
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.265,
+      "loss": 2.3923,
       "step": 5845
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.3723,
+      "loss": 2.4244,
       "step": 5850
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.2181,
+      "loss": 2.3257,
       "step": 5855
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.3984375,
       "learning_rate": 0.001,
-      "loss": 2.314,
+      "loss": 2.2825,
       "step": 5860
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.2494,
+      "loss": 2.3113,
       "step": 5865
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.38671875,
+      "grad_norm": 1.8125,
       "learning_rate": 0.001,
-      "loss": 2.2576,
+      "loss": 2.3198,
       "step": 5870
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.515625,
       "learning_rate": 0.001,
-      "loss": 2.3208,
+      "loss": 2.3237,
       "step": 5875
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.453125,
       "learning_rate": 0.001,
-      "loss": 2.3072,
+      "loss": 2.2635,
       "step": 5880
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.3374,
+      "loss": 2.3256,
       "step": 5885
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.322265625,
+      "grad_norm": 0.97265625,
       "learning_rate": 0.001,
-      "loss": 2.3156,
+      "loss": 2.2391,
       "step": 5890
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.3659,
+      "loss": 2.3338,
       "step": 5895
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.3847,
+      "loss": 2.2332,
       "step": 5900
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.29,
+      "loss": 2.2905,
       "step": 5905
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.282,
+      "loss": 2.3601,
       "step": 5910
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.3172,
+      "loss": 2.299,
       "step": 5915
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.2918,
+      "loss": 2.3948,
       "step": 5920
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.3004,
+      "loss": 2.4341,
       "step": 5925
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.30859375,
+      "grad_norm": 2.578125,
       "learning_rate": 0.001,
-      "loss": 2.3571,
+      "loss": 2.3247,
       "step": 5930
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.328125,
+      "grad_norm": 5.21875,
       "learning_rate": 0.001,
-      "loss": 2.3381,
+      "loss": 2.4271,
       "step": 5935
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.2733,
+      "loss": 2.2825,
       "step": 5940
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 2.3277,
+      "loss": 2.3141,
       "step": 5945
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.4188,
+      "loss": 2.364,
       "step": 5950
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.5859375,
       "learning_rate": 0.001,
-      "loss": 2.2578,
+      "loss": 2.3063,
       "step": 5955
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.392578125,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.2164,
+      "loss": 2.2324,
       "step": 5960
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.361328125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.3124,
+      "loss": 2.3069,
       "step": 5965
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2956,
+      "loss": 2.2703,
       "step": 5970
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.2811,
+      "loss": 2.2642,
       "step": 5975
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.3955,
+      "loss": 2.3167,
       "step": 5980
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.3116,
+      "loss": 2.3079,
       "step": 5985
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.3184,
+      "loss": 2.3698,
       "step": 5990
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.3773,
+      "loss": 2.2639,
       "step": 5995
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.3402,
+      "loss": 2.2953,
       "step": 6000
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.3409,
+      "loss": 2.2728,
       "step": 6005
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.3015,
+      "loss": 2.2379,
       "step": 6010
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.2731,
+      "loss": 2.2972,
       "step": 6015
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.4375,
       "learning_rate": 0.001,
-      "loss": 2.352,
+      "loss": 2.2669,
       "step": 6020
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.3109,
+      "loss": 2.3334,
       "step": 6025
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.3099,
+      "loss": 2.2345,
       "step": 6030
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.2662,
+      "loss": 2.2557,
       "step": 6035
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.3723,
+      "loss": 2.3104,
       "step": 6040
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.3657,
+      "loss": 2.2174,
       "step": 6045
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.3337,
+      "loss": 2.2894,
       "step": 6050
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.2631,
+      "loss": 2.3237,
       "step": 6055
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.546875,
       "learning_rate": 0.001,
-      "loss": 2.3135,
+      "loss": 2.2313,
       "step": 6060
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.3438,
+      "loss": 2.386,
       "step": 6065
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.3203125,
+      "grad_norm": 2.6875,
       "learning_rate": 0.001,
-      "loss": 2.2664,
+      "loss": 2.3001,
       "step": 6070
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.3946,
+      "loss": 2.1863,
       "step": 6075
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.296875,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.2588,
+      "loss": 2.3525,
       "step": 6080
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.2975,
+      "loss": 2.2239,
       "step": 6085
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.7265625,
       "learning_rate": 0.001,
-      "loss": 2.3101,
+      "loss": 2.2718,
       "step": 6090
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.3642,
+      "loss": 2.2479,
       "step": 6095
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.3494,
+      "loss": 2.2787,
       "step": 6100
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.298828125,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2839,
+      "loss": 2.2199,
       "step": 6105
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.3428,
+      "loss": 2.3662,
       "step": 6110
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.2826,
+      "loss": 2.284,
       "step": 6115
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.3481,
+      "loss": 2.2711,
       "step": 6120
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.3043,
+      "loss": 2.2889,
       "step": 6125
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.2965,
+      "loss": 2.256,
       "step": 6130
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.2599,
+      "loss": 2.2337,
       "step": 6135
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.75,
       "learning_rate": 0.001,
-      "loss": 2.2636,
+      "loss": 2.2716,
       "step": 6140
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.5546875,
       "learning_rate": 0.001,
-      "loss": 2.3423,
+      "loss": 2.2392,
       "step": 6145
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.2692,
+      "loss": 2.2354,
       "step": 6150
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.2963,
+      "loss": 2.1689,
       "step": 6155
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.453125,
       "learning_rate": 0.001,
-      "loss": 2.2212,
+      "loss": 2.2155,
       "step": 6160
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.296875,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.3136,
+      "loss": 2.3151,
       "step": 6165
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.34765625,
+      "grad_norm": 0.94140625,
       "learning_rate": 0.001,
-      "loss": 2.3659,
+      "loss": 2.2376,
       "step": 6170
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.2902,
+      "loss": 2.3778,
       "step": 6175
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.294921875,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.3175,
+      "loss": 2.2291,
       "step": 6180
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.3527,
+      "loss": 2.2969,
       "step": 6185
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.3352,
+      "loss": 2.3347,
       "step": 6190
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.458984375,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.3264,
+      "loss": 2.2838,
       "step": 6195
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.3021,
+      "loss": 2.2417,
       "step": 6200
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.314453125,
+      "grad_norm": 0.96875,
       "learning_rate": 0.001,
-      "loss": 2.3128,
+      "loss": 2.227,
       "step": 6205
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.3982,
+      "loss": 2.2919,
       "step": 6210
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.3052,
+      "loss": 2.1453,
       "step": 6215
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.269,
+      "loss": 2.4227,
       "step": 6220
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.4296875,
       "learning_rate": 0.001,
-      "loss": 2.357,
+      "loss": 2.2441,
       "step": 6225
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.99609375,
       "learning_rate": 0.001,
-      "loss": 2.3996,
+      "loss": 2.3082,
       "step": 6230
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.3695,
+      "loss": 2.2227,
       "step": 6235
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.6484375,
       "learning_rate": 0.001,
-      "loss": 2.3206,
+      "loss": 2.3381,
       "step": 6240
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.2325,
+      "loss": 2.2592,
       "step": 6245
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.334,
+      "loss": 2.3264,
       "step": 6250
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.2761,
+      "loss": 2.3429,
       "step": 6255
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.3984375,
       "learning_rate": 0.001,
-      "loss": 2.3011,
+      "loss": 2.258,
       "step": 6260
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.3958,
+      "loss": 2.3763,
       "step": 6265
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.3143,
+      "loss": 2.3234,
       "step": 6270
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.3211,
+      "loss": 2.2315,
       "step": 6275
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.6171875,
       "learning_rate": 0.001,
-      "loss": 2.2919,
+      "loss": 2.2926,
       "step": 6280
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.3387,
+      "loss": 2.2738,
       "step": 6285
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.3049,
+      "loss": 2.273,
       "step": 6290
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 2.3557,
+      "loss": 2.2739,
       "step": 6295
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.59375,
       "learning_rate": 0.001,
-      "loss": 2.3384,
+      "loss": 2.2088,
       "step": 6300
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.3372,
+      "loss": 2.3184,
       "step": 6305
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.2197,
+      "loss": 2.2584,
       "step": 6310
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.2533,
+      "loss": 2.2646,
       "step": 6315
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.2072,
+      "loss": 2.2761,
       "step": 6320
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.376953125,
+      "grad_norm": 0.91015625,
       "learning_rate": 0.001,
-      "loss": 2.2987,
+      "loss": 2.2677,
       "step": 6325
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.3984375,
       "learning_rate": 0.001,
-      "loss": 2.2634,
+      "loss": 2.2255,
       "step": 6330
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.353515625,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.3597,
+      "loss": 2.2457,
       "step": 6335
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.2872,
+      "loss": 2.3369,
       "step": 6340
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.3196,
+      "loss": 2.1955,
       "step": 6345
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.32421875,
+      "grad_norm": 3.765625,
       "learning_rate": 0.001,
-      "loss": 2.2671,
+      "loss": 2.2271,
       "step": 6350
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.453125,
       "learning_rate": 0.001,
-      "loss": 2.2474,
+      "loss": 2.1785,
       "step": 6355
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.318359375,
+      "grad_norm": 3.328125,
       "learning_rate": 0.001,
-      "loss": 2.334,
+      "loss": 2.2299,
       "step": 6360
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.314453125,
+      "grad_norm": 0.9765625,
       "learning_rate": 0.001,
-      "loss": 2.3749,
+      "loss": 2.1963,
       "step": 6365
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.2718,
+      "loss": 2.1968,
       "step": 6370
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.3395,
+      "loss": 2.2366,
       "step": 6375
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.3203125,
+      "grad_norm": 2.703125,
       "learning_rate": 0.001,
-      "loss": 2.3243,
+      "loss": 2.2328,
       "step": 6380
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.3901,
+      "loss": 2.2762,
       "step": 6385
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.3203125,
+      "grad_norm": 0.98046875,
       "learning_rate": 0.001,
-      "loss": 2.2454,
+      "loss": 2.1722,
       "step": 6390
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.330078125,
+      "grad_norm": 2.515625,
       "learning_rate": 0.001,
-      "loss": 2.287,
+      "loss": 2.2829,
       "step": 6395
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.283,
+      "loss": 2.2764,
       "step": 6400
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.298828125,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.3251,
+      "loss": 2.3164,
       "step": 6405
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.3293,
+      "loss": 2.2618,
       "step": 6410
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.353515625,
+      "grad_norm": 0.9453125,
       "learning_rate": 0.001,
-      "loss": 2.2905,
+      "loss": 2.2912,
       "step": 6415
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.345703125,
+      "grad_norm": 2.0,
       "learning_rate": 0.001,
-      "loss": 2.2483,
+      "loss": 2.2168,
       "step": 6420
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.4921875,
       "learning_rate": 0.001,
-      "loss": 2.3314,
+      "loss": 2.2411,
       "step": 6425
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.328125,
+      "grad_norm": 0.94921875,
       "learning_rate": 0.001,
-      "loss": 2.3804,
+      "loss": 2.2966,
       "step": 6430
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.345703125,
+      "grad_norm": 3.078125,
       "learning_rate": 0.001,
-      "loss": 2.3254,
+      "loss": 2.2664,
       "step": 6435
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.3487,
+      "loss": 2.2472,
       "step": 6440
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.609375,
       "learning_rate": 0.001,
-      "loss": 2.3102,
+      "loss": 2.2971,
       "step": 6445
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.2659,
+      "loss": 2.2707,
       "step": 6450
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.3119,
+      "loss": 2.2137,
       "step": 6455
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.3058,
+      "loss": 2.2746,
       "step": 6460
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.2881,
+      "loss": 2.3172,
       "step": 6465
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.40625,
       "learning_rate": 0.001,
-      "loss": 2.2166,
+      "loss": 2.215,
       "step": 6470
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.3242,
+      "loss": 2.3012,
       "step": 6475
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.65625,
       "learning_rate": 0.001,
-      "loss": 2.3057,
+      "loss": 2.2543,
       "step": 6480
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.2805,
+      "loss": 2.2402,
       "step": 6485
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.2795,
+      "loss": 2.2796,
       "step": 6490
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.2733,
+      "loss": 2.3052,
       "step": 6495
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.318359375,
+      "grad_norm": 0.89453125,
       "learning_rate": 0.001,
-      "loss": 2.2482,
+      "loss": 2.2445,
       "step": 6500
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.306640625,
+      "grad_norm": 2.5,
       "learning_rate": 0.001,
-      "loss": 2.2718,
+      "loss": 2.2294,
       "step": 6505
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.31640625,
+      "grad_norm": 2.484375,
       "learning_rate": 0.001,
-      "loss": 2.2996,
+      "loss": 2.2323,
       "step": 6510
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.32421875,
+      "grad_norm": 0.94140625,
       "learning_rate": 0.001,
-      "loss": 2.2118,
+      "loss": 2.2513,
       "step": 6515
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.28515625,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.3748,
+      "loss": 2.2792,
       "step": 6520
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.3675,
+      "loss": 2.2732,
       "step": 6525
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.5234375,
       "learning_rate": 0.001,
-      "loss": 2.2712,
+      "loss": 2.2925,
       "step": 6530
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2544,
+      "loss": 2.2928,
       "step": 6535
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.3075,
+      "loss": 2.2555,
       "step": 6540
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.299,
+      "loss": 2.2544,
       "step": 6545
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.327,
+      "loss": 2.3104,
       "step": 6550
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.3941,
+      "loss": 2.2852,
       "step": 6555
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.2842,
+      "loss": 2.238,
       "step": 6560
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.3262,
+      "loss": 2.1404,
       "step": 6565
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.359375,
+      "grad_norm": 1.5859375,
       "learning_rate": 0.001,
-      "loss": 2.3961,
+      "loss": 2.167,
       "step": 6570
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.3889,
+      "loss": 2.1843,
       "step": 6575
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.3397,
+      "loss": 2.3116,
       "step": 6580
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.0,
       "learning_rate": 0.001,
-      "loss": 2.2399,
+      "loss": 2.2582,
       "step": 6585
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 2.3182,
+      "loss": 2.1951,
       "step": 6590
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.3264,
+      "loss": 2.3272,
       "step": 6595
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.322265625,
+      "grad_norm": 0.94921875,
       "learning_rate": 0.001,
-      "loss": 2.3071,
+      "loss": 2.2032,
       "step": 6600
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.4024,
+      "loss": 2.2803,
       "step": 6605
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.2123,
+      "loss": 2.1923,
       "step": 6610
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.2309,
+      "loss": 2.278,
       "step": 6615
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.1942,
+      "loss": 2.208,
       "step": 6620
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.298828125,
+      "grad_norm": 0.9921875,
       "learning_rate": 0.001,
-      "loss": 2.2871,
+      "loss": 2.3667,
       "step": 6625
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.30859375,
+      "grad_norm": 2.0,
       "learning_rate": 0.001,
-      "loss": 2.3449,
+      "loss": 2.23,
       "step": 6630
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.310546875,
+      "grad_norm": 0.98828125,
       "learning_rate": 0.001,
-      "loss": 2.3352,
+      "loss": 2.3159,
       "step": 6635
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.30859375,
+      "grad_norm": 0.87109375,
       "learning_rate": 0.001,
-      "loss": 2.3952,
+      "loss": 2.1728,
       "step": 6640
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.2695,
+      "loss": 2.3895,
       "step": 6645
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.3078,
+      "loss": 2.2639,
       "step": 6650
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.3105,
+      "loss": 2.2976,
       "step": 6655
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.71875,
       "learning_rate": 0.001,
-      "loss": 2.3317,
+      "loss": 2.2421,
       "step": 6660
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.3116,
+      "loss": 2.3828,
       "step": 6665
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.2339,
+      "loss": 2.3166,
       "step": 6670
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.3095,
+      "loss": 2.2423,
       "step": 6675
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.3515625,
       "learning_rate": 0.001,
-      "loss": 2.239,
+      "loss": 2.1668,
       "step": 6680
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.4118,
+      "loss": 2.2309,
       "step": 6685
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.53125,
       "learning_rate": 0.001,
-      "loss": 2.2235,
+      "loss": 2.2477,
       "step": 6690
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.333984375,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.3407,
+      "loss": 2.1868,
       "step": 6695
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.2263,
+      "loss": 2.2333,
       "step": 6700
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.2005,
+      "loss": 2.232,
       "step": 6705
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.3457,
+      "loss": 2.2175,
       "step": 6710
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.1686,
+      "loss": 2.2547,
       "step": 6715
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.2701,
+      "loss": 2.2624,
       "step": 6720
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.2268,
+      "loss": 2.2089,
       "step": 6725
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.369140625,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.232,
+      "loss": 2.1886,
       "step": 6730
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.7265625,
       "learning_rate": 0.001,
-      "loss": 2.2332,
+      "loss": 2.2837,
       "step": 6735
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.3156,
+      "loss": 2.2128,
       "step": 6740
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.2292,
+      "loss": 2.2201,
       "step": 6745
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.267,
+      "loss": 2.1656,
       "step": 6750
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.3458,
+      "loss": 2.229,
       "step": 6755
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.3359375,
+      "grad_norm": 2.140625,
       "learning_rate": 0.001,
-      "loss": 2.2835,
+      "loss": 2.3482,
       "step": 6760
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.296875,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.2314,
+      "loss": 2.2294,
       "step": 6765
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.2215,
+      "loss": 2.2277,
       "step": 6770
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.3393,
+      "loss": 2.2251,
       "step": 6775
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.5390625,
       "learning_rate": 0.001,
-      "loss": 2.2511,
+      "loss": 2.2425,
       "step": 6780
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.2021,
+      "loss": 2.1912,
       "step": 6785
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.306640625,
+      "grad_norm": 0.94921875,
       "learning_rate": 0.001,
-      "loss": 2.2558,
+      "loss": 2.1729,
       "step": 6790
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.2729,
+      "loss": 2.2105,
       "step": 6795
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.291015625,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.2082,
+      "loss": 2.2288,
       "step": 6800
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.30859375,
+      "grad_norm": 0.98828125,
       "learning_rate": 0.001,
-      "loss": 2.2801,
+      "loss": 2.1926,
       "step": 6805
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.312,
+      "loss": 2.2566,
       "step": 6810
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.314453125,
+      "grad_norm": 2.46875,
       "learning_rate": 0.001,
-      "loss": 2.2518,
+      "loss": 2.1782,
       "step": 6815
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.2459,
+      "loss": 2.2487,
       "step": 6820
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.1997,
+      "loss": 2.2902,
       "step": 6825
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.3033,
+      "loss": 2.1668,
       "step": 6830
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.2142,
+      "loss": 2.2062,
       "step": 6835
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.8125,
       "learning_rate": 0.001,
-      "loss": 2.3164,
+      "loss": 2.2033,
       "step": 6840
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.2225,
+      "loss": 2.3064,
       "step": 6845
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.3232,
+      "loss": 2.2519,
       "step": 6850
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.2207,
+      "loss": 2.2718,
       "step": 6855
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.34375,
+      "grad_norm": 0.9375,
       "learning_rate": 0.001,
-      "loss": 2.3717,
+      "loss": 2.2755,
       "step": 6860
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.2952,
+      "loss": 2.2878,
       "step": 6865
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.2302,
+      "loss": 2.1555,
       "step": 6870
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.3487,
+      "loss": 2.2228,
       "step": 6875
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.2465,
+      "loss": 2.3159,
       "step": 6880
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 2.3268,
+      "loss": 2.1421,
       "step": 6885
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2749,
+      "loss": 2.218,
       "step": 6890
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.3893,
+      "loss": 2.249,
       "step": 6895
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.298828125,
+      "grad_norm": 1.0078125,
       "learning_rate": 0.001,
-      "loss": 2.3373,
+      "loss": 2.257,
       "step": 6900
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.2797,
+      "loss": 2.2311,
       "step": 6905
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.3495,
+      "loss": 2.3024,
       "step": 6910
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.3984375,
       "learning_rate": 0.001,
-      "loss": 2.297,
+      "loss": 2.3048,
       "step": 6915
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.2722,
+      "loss": 2.2174,
       "step": 6920
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.2851,
+      "loss": 2.2053,
       "step": 6925
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.5859375,
       "learning_rate": 0.001,
-      "loss": 2.2873,
+      "loss": 2.2763,
       "step": 6930
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.609375,
       "learning_rate": 0.001,
-      "loss": 2.3203,
+      "loss": 2.2094,
       "step": 6935
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.3418,
+      "loss": 2.2562,
       "step": 6940
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.2656,
+      "loss": 2.1941,
       "step": 6945
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.6796875,
       "learning_rate": 0.001,
-      "loss": 2.2917,
+      "loss": 2.2871,
       "step": 6950
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.0078125,
       "learning_rate": 0.001,
-      "loss": 2.3157,
+      "loss": 2.2773,
       "step": 6955
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.3485,
+      "loss": 2.2026,
       "step": 6960
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.3457,
+      "loss": 2.1996,
       "step": 6965
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.3021,
+      "loss": 2.2896,
       "step": 6970
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.3554,
+      "loss": 2.2455,
       "step": 6975
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.302734375,
+      "grad_norm": 2.03125,
       "learning_rate": 0.001,
-      "loss": 2.214,
+      "loss": 2.2318,
       "step": 6980
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.2834,
+      "loss": 2.3176,
       "step": 6985
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.32421875,
+      "grad_norm": 2.34375,
       "learning_rate": 0.001,
-      "loss": 2.2508,
+      "loss": 2.302,
       "step": 6990
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.5625,
       "learning_rate": 0.001,
-      "loss": 2.3555,
+      "loss": 2.2314,
       "step": 6995
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.2949,
+      "loss": 2.1735,
       "step": 7000
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.3022,
+      "loss": 2.2345,
       "step": 7005
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.2868,
+      "loss": 2.1579,
       "step": 7010
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.3297,
+      "loss": 2.2407,
       "step": 7015
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.34765625,
+      "grad_norm": 1.5390625,
       "learning_rate": 0.001,
-      "loss": 2.257,
+      "loss": 2.2276,
       "step": 7020
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.328125,
+      "grad_norm": 8.0,
       "learning_rate": 0.001,
-      "loss": 2.2424,
+      "loss": 2.2251,
       "step": 7025
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.65625,
       "learning_rate": 0.001,
-      "loss": 2.2686,
+      "loss": 2.2293,
       "step": 7030
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.294921875,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.2936,
+      "loss": 2.164,
       "step": 7035
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.3123,
+      "loss": 2.1919,
       "step": 7040
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.2571,
+      "loss": 2.2568,
       "step": 7045
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.2647,
+      "loss": 2.1385,
       "step": 7050
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.304,
+      "loss": 2.1111,
       "step": 7055
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.2541,
+      "loss": 2.2902,
       "step": 7060
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.354,
+      "loss": 2.2044,
       "step": 7065
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.3359375,
       "learning_rate": 0.001,
-      "loss": 2.336,
+      "loss": 2.212,
       "step": 7070
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.484375,
       "learning_rate": 0.001,
-      "loss": 2.3124,
+      "loss": 2.1734,
       "step": 7075
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.5859375,
       "learning_rate": 0.001,
-      "loss": 2.3281,
+      "loss": 2.2047,
       "step": 7080
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 2.3212,
+      "loss": 2.1626,
       "step": 7085
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.3671875,
+      "grad_norm": 1.40625,
       "learning_rate": 0.001,
-      "loss": 2.3689,
+      "loss": 2.188,
       "step": 7090
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.2572,
+      "loss": 2.1195,
       "step": 7095
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.7734375,
       "learning_rate": 0.001,
-      "loss": 2.335,
+      "loss": 2.2254,
       "step": 7100
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.298828125,
+      "grad_norm": 1.8125,
       "learning_rate": 0.001,
-      "loss": 2.2974,
+      "loss": 2.1368,
       "step": 7105
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.3526,
+      "loss": 2.1321,
       "step": 7110
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.2890625,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.3439,
+      "loss": 2.1857,
       "step": 7115
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.5859375,
       "learning_rate": 0.001,
-      "loss": 2.3026,
+      "loss": 2.2296,
       "step": 7120
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.3288,
+      "loss": 2.2473,
       "step": 7125
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.29296875,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.2952,
+      "loss": 2.3584,
       "step": 7130
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.40625,
       "learning_rate": 0.001,
-      "loss": 2.2942,
+      "loss": 2.2099,
       "step": 7135
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.2384,
+      "loss": 2.1661,
       "step": 7140
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.5859375,
       "learning_rate": 0.001,
-      "loss": 2.2436,
+      "loss": 2.3145,
       "step": 7145
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.3166,
+      "loss": 2.2612,
       "step": 7150
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.2552,
+      "loss": 2.2163,
       "step": 7155
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.294921875,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2395,
+      "loss": 2.1993,
       "step": 7160
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.2334,
+      "loss": 2.1804,
       "step": 7165
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.314453125,
+      "grad_norm": 0.99609375,
       "learning_rate": 0.001,
-      "loss": 2.3261,
+      "loss": 2.2222,
       "step": 7170
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2233,
+      "loss": 2.2598,
       "step": 7175
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.3239,
+      "loss": 2.222,
       "step": 7180
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.208,
+      "loss": 2.1146,
       "step": 7185
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.2938,
+      "loss": 2.2832,
       "step": 7190
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.3055,
+      "loss": 2.2666,
       "step": 7195
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.318359375,
+      "grad_norm": 7.0625,
       "learning_rate": 0.001,
-      "loss": 2.2853,
+      "loss": 2.2489,
       "step": 7200
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.5,
       "learning_rate": 0.001,
-      "loss": 2.2827,
+      "loss": 2.2843,
       "step": 7205
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.390625,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.234,
+      "loss": 2.2427,
       "step": 7210
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.2621,
+      "loss": 2.1722,
       "step": 7215
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.625,
       "learning_rate": 0.001,
-      "loss": 2.2557,
+      "loss": 2.2615,
       "step": 7220
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.302734375,
+      "grad_norm": 2.484375,
       "learning_rate": 0.001,
-      "loss": 2.1625,
+      "loss": 2.2178,
       "step": 7225
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2853,
+      "loss": 2.2657,
       "step": 7230
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.1697,
+      "loss": 2.2596,
       "step": 7235
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.3439,
+      "loss": 2.3205,
       "step": 7240
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.2568,
+      "loss": 2.1141,
       "step": 7245
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.9765625,
       "learning_rate": 0.001,
-      "loss": 2.2698,
+      "loss": 2.2048,
       "step": 7250
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.4765625,
       "learning_rate": 0.001,
-      "loss": 2.3522,
+      "loss": 2.1992,
       "step": 7255
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.2786,
+      "loss": 2.2168,
       "step": 7260
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2758,
+      "loss": 2.1951,
       "step": 7265
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.2267,
+      "loss": 2.1946,
       "step": 7270
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.2619,
+      "loss": 2.2296,
       "step": 7275
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.296875,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.235,
+      "loss": 2.1909,
       "step": 7280
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.1946,
+      "loss": 2.2634,
       "step": 7285
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.3306,
+      "loss": 2.1766,
       "step": 7290
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.3515625,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.3197,
+      "loss": 2.2185,
       "step": 7295
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.390625,
       "learning_rate": 0.001,
-      "loss": 2.3484,
+      "loss": 2.3009,
       "step": 7300
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.2644,
+      "loss": 2.234,
       "step": 7305
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.3429,
+      "loss": 2.2108,
       "step": 7310
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.248,
+      "loss": 2.1009,
       "step": 7315
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.2096,
+      "loss": 2.1628,
       "step": 7320
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.376953125,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.1851,
+      "loss": 2.2285,
       "step": 7325
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2935,
+      "loss": 2.1172,
       "step": 7330
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.3234,
+      "loss": 2.2362,
       "step": 7335
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.298828125,
+      "grad_norm": 2.296875,
       "learning_rate": 0.001,
-      "loss": 2.2555,
+      "loss": 2.1721,
       "step": 7340
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.318359375,
+      "grad_norm": 0.9921875,
       "learning_rate": 0.001,
-      "loss": 2.2168,
+      "loss": 2.0911,
       "step": 7345
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.3627,
+      "loss": 2.1961,
       "step": 7350
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.2858,
+      "loss": 2.2337,
       "step": 7355
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.33984375,
+      "grad_norm": 3.53125,
       "learning_rate": 0.001,
-      "loss": 2.3002,
+      "loss": 2.2182,
       "step": 7360
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.345703125,
+      "grad_norm": 2.859375,
       "learning_rate": 0.001,
-      "loss": 2.2817,
+      "loss": 2.1035,
       "step": 7365
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.291015625,
+      "grad_norm": 0.83984375,
       "learning_rate": 0.001,
-      "loss": 2.2995,
+      "loss": 2.1426,
       "step": 7370
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.2896,
+      "loss": 2.2156,
       "step": 7375
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.2691,
+      "loss": 2.2241,
       "step": 7380
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.298828125,
+      "grad_norm": 0.98828125,
       "learning_rate": 0.001,
-      "loss": 2.3571,
+      "loss": 2.1933,
       "step": 7385
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.2753,
+      "loss": 2.2293,
       "step": 7390
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.1712,
+      "loss": 2.2288,
       "step": 7395
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.31640625,
+      "grad_norm": 0.9375,
       "learning_rate": 0.001,
-      "loss": 2.3137,
+      "loss": 2.2141,
       "step": 7400
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.318359375,
+      "grad_norm": 0.91796875,
       "learning_rate": 0.001,
-      "loss": 2.3335,
+      "loss": 2.1613,
       "step": 7405
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.1866,
+      "loss": 2.2172,
       "step": 7410
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.219,
+      "loss": 2.2277,
       "step": 7415
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.3086,
+      "loss": 2.1964,
       "step": 7420
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.2989,
+      "loss": 2.1474,
       "step": 7425
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.515625,
       "learning_rate": 0.001,
-      "loss": 2.2209,
+      "loss": 2.1842,
       "step": 7430
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.2487,
+      "loss": 2.2594,
       "step": 7435
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.294921875,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.2269,
+      "loss": 2.1938,
       "step": 7440
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.287109375,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.2807,
+      "loss": 2.2454,
       "step": 7445
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.3287,
+      "loss": 2.1547,
       "step": 7450
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.2568,
+      "loss": 2.2064,
       "step": 7455
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.0078125,
       "learning_rate": 0.001,
-      "loss": 2.2662,
+      "loss": 2.2425,
       "step": 7460
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.2783,
+      "loss": 2.2289,
       "step": 7465
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.239,
+      "loss": 2.1156,
       "step": 7470
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.3046875,
+      "grad_norm": 2.09375,
       "learning_rate": 0.001,
-      "loss": 2.2978,
+      "loss": 2.2579,
       "step": 7475
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.828125,
       "learning_rate": 0.001,
-      "loss": 2.3329,
+      "loss": 2.2268,
       "step": 7480
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.1576,
+      "loss": 2.1241,
       "step": 7485
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.2521,
+      "loss": 2.1078,
       "step": 7490
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.298828125,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.2536,
+      "loss": 2.2325,
       "step": 7495
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.2902,
+      "loss": 2.1536,
       "step": 7500
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.4453125,
       "learning_rate": 0.001,
-      "loss": 2.2477,
+      "loss": 2.2238,
       "step": 7505
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.3125,
+      "grad_norm": 0.96875,
       "learning_rate": 0.001,
-      "loss": 2.2292,
+      "loss": 2.223,
       "step": 7510
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.2579,
+      "loss": 2.2105,
       "step": 7515
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.2531,
+      "loss": 2.2338,
       "step": 7520
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.3611,
+      "loss": 2.182,
       "step": 7525
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.2527,
+      "loss": 2.1806,
       "step": 7530
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.0,
       "learning_rate": 0.001,
-      "loss": 2.3296,
+      "loss": 2.1253,
       "step": 7535
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.2596,
+      "loss": 2.065,
       "step": 7540
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.5390625,
       "learning_rate": 0.001,
-      "loss": 2.2396,
+      "loss": 2.1058,
       "step": 7545
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.2573,
+      "loss": 2.1723,
       "step": 7550
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.3203125,
       "learning_rate": 0.001,
-      "loss": 2.3043,
+      "loss": 2.2771,
       "step": 7555
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.2121,
+      "loss": 2.178,
       "step": 7560
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.3462,
+      "loss": 2.1832,
       "step": 7565
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.287109375,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.3078,
+      "loss": 2.106,
       "step": 7570
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.316,
+      "loss": 2.1701,
       "step": 7575
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.292,
+      "loss": 2.2016,
       "step": 7580
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.341796875,
+      "grad_norm": 0.98828125,
       "learning_rate": 0.001,
-      "loss": 2.3084,
+      "loss": 2.1198,
       "step": 7585
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.235,
+      "loss": 2.222,
       "step": 7590
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2572,
+      "loss": 2.1769,
       "step": 7595
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.2612,
+      "loss": 2.1416,
       "step": 7600
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.2995,
+      "loss": 2.1255,
       "step": 7605
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.2694,
+      "loss": 2.1879,
       "step": 7610
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.2282,
+      "loss": 2.1509,
       "step": 7615
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.296875,
+      "grad_norm": 1.296875,
       "learning_rate": 0.001,
-      "loss": 2.2096,
+      "loss": 2.1773,
       "step": 7620
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.30078125,
+      "grad_norm": 0.94140625,
       "learning_rate": 0.001,
-      "loss": 2.2345,
+      "loss": 2.1697,
       "step": 7625
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.890625,
       "learning_rate": 0.001,
-      "loss": 2.2535,
+      "loss": 2.1141,
       "step": 7630
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.2774,
+      "loss": 2.2385,
       "step": 7635
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.2084,
+      "loss": 2.1444,
       "step": 7640
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.306640625,
+      "grad_norm": 2.90625,
       "learning_rate": 0.001,
-      "loss": 2.2479,
+      "loss": 2.1821,
       "step": 7645
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.306640625,
+      "grad_norm": 2.5625,
       "learning_rate": 0.001,
-      "loss": 2.2153,
+      "loss": 2.2419,
       "step": 7650
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 2.2901,
+      "loss": 2.1277,
       "step": 7655
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.328125,
+      "grad_norm": 2.859375,
       "learning_rate": 0.001,
-      "loss": 2.2653,
+      "loss": 2.1443,
       "step": 7660
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2731,
+      "loss": 2.1499,
       "step": 7665
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.294921875,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.1943,
+      "loss": 2.1843,
       "step": 7670
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.203125,
       "learning_rate": 0.001,
-      "loss": 2.1499,
+      "loss": 2.1554,
       "step": 7675
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.2693,
+      "loss": 2.1742,
       "step": 7680
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.7890625,
       "learning_rate": 0.001,
-      "loss": 2.2261,
+      "loss": 2.253,
       "step": 7685
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.1544,
+      "loss": 2.2189,
       "step": 7690
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.2456,
+      "loss": 2.2244,
       "step": 7695
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.2612,
+      "loss": 2.2498,
       "step": 7700
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.2848,
+      "loss": 2.1689,
       "step": 7705
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.2257,
+      "loss": 2.1435,
       "step": 7710
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.2658,
+      "loss": 2.0968,
       "step": 7715
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.306640625,
+      "grad_norm": 2.171875,
       "learning_rate": 0.001,
-      "loss": 2.2202,
+      "loss": 2.1769,
       "step": 7720
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.390625,
       "learning_rate": 0.001,
-      "loss": 2.2599,
+      "loss": 2.2411,
       "step": 7725
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.369140625,
+      "grad_norm": 2.671875,
       "learning_rate": 0.001,
-      "loss": 2.3019,
+      "loss": 2.2172,
       "step": 7730
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.3055,
+      "loss": 2.1888,
       "step": 7735
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.296875,
+      "grad_norm": 1.7265625,
       "learning_rate": 0.001,
-      "loss": 2.1914,
+      "loss": 2.2317,
       "step": 7740
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.30859375,
+      "grad_norm": 0.93359375,
       "learning_rate": 0.001,
-      "loss": 2.2936,
+      "loss": 2.1126,
       "step": 7745
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.2759,
+      "loss": 2.1826,
       "step": 7750
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.34375,
       "learning_rate": 0.001,
-      "loss": 2.2421,
+      "loss": 2.1961,
       "step": 7755
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.3079,
+      "loss": 2.0871,
       "step": 7760
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.7578125,
       "learning_rate": 0.001,
-      "loss": 2.2364,
+      "loss": 2.2262,
       "step": 7765
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.2714,
+      "loss": 2.1916,
       "step": 7770
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.546875,
       "learning_rate": 0.001,
-      "loss": 2.2086,
+      "loss": 2.1953,
       "step": 7775
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.2275,
+      "loss": 2.2019,
       "step": 7780
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.3203125,
+      "grad_norm": 0.9453125,
       "learning_rate": 0.001,
-      "loss": 2.3385,
+      "loss": 2.2187,
       "step": 7785
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.291015625,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.2574,
+      "loss": 2.2037,
       "step": 7790
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.318359375,
+      "grad_norm": 0.94140625,
       "learning_rate": 0.001,
-      "loss": 2.2743,
+      "loss": 2.0825,
       "step": 7795
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.294921875,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.2457,
+      "loss": 2.1189,
       "step": 7800
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.298828125,
+      "grad_norm": 0.99609375,
       "learning_rate": 0.001,
-      "loss": 2.3624,
+      "loss": 2.2302,
       "step": 7805
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.291015625,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.2603,
+      "loss": 2.1359,
       "step": 7810
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.09375,
       "learning_rate": 0.001,
-      "loss": 2.2317,
+      "loss": 2.2286,
       "step": 7815
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.2466,
+      "loss": 2.2267,
       "step": 7820
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.3946,
+      "loss": 2.2747,
       "step": 7825
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.2734375,
       "learning_rate": 0.001,
-      "loss": 2.2815,
+      "loss": 2.1479,
       "step": 7830
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.2996,
+      "loss": 2.2157,
       "step": 7835
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.1796875,
       "learning_rate": 0.001,
-      "loss": 2.2776,
+      "loss": 2.2047,
       "step": 7840
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.2265625,
       "learning_rate": 0.001,
-      "loss": 2.184,
+      "loss": 2.1771,
       "step": 7845
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.703125,
       "learning_rate": 0.001,
-      "loss": 2.2714,
+      "loss": 2.2158,
       "step": 7850
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.2512,
+      "loss": 2.1596,
       "step": 7855
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.291015625,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.2536,
+      "loss": 2.226,
       "step": 7860
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.35546875,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.1873,
+      "loss": 2.138,
       "step": 7865
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.337890625,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.2324,
+      "loss": 2.166,
       "step": 7870
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.390625,
       "learning_rate": 0.001,
-      "loss": 2.3213,
+      "loss": 2.1684,
       "step": 7875
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.2541,
+      "loss": 2.1095,
       "step": 7880
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.515625,
       "learning_rate": 0.001,
-      "loss": 2.2901,
+      "loss": 2.1289,
       "step": 7885
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.310546875,
+      "grad_norm": 0.94921875,
       "learning_rate": 0.001,
-      "loss": 2.3655,
+      "loss": 2.2142,
       "step": 7890
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.291015625,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.1354,
+      "loss": 2.0919,
       "step": 7895
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.2691,
+      "loss": 2.2185,
       "step": 7900
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.159,
+      "loss": 2.1294,
       "step": 7905
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.291015625,
+      "grad_norm": 1.1875,
       "learning_rate": 0.001,
-      "loss": 2.1469,
+      "loss": 2.186,
       "step": 7910
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.609375,
       "learning_rate": 0.001,
-      "loss": 2.2344,
+      "loss": 2.1617,
       "step": 7915
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.2182,
+      "loss": 2.1857,
       "step": 7920
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.3306,
+      "loss": 2.2289,
       "step": 7925
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.5390625,
       "learning_rate": 0.001,
-      "loss": 2.1549,
+      "loss": 2.2337,
       "step": 7930
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.3984375,
       "learning_rate": 0.001,
-      "loss": 2.3377,
+      "loss": 2.2364,
       "step": 7935
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2853,
+      "loss": 2.2024,
       "step": 7940
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.2024,
+      "loss": 2.1375,
       "step": 7945
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.1640625,
       "learning_rate": 0.001,
-      "loss": 2.2813,
+      "loss": 2.1636,
       "step": 7950
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.298828125,
+      "grad_norm": 1.140625,
       "learning_rate": 0.001,
-      "loss": 2.3001,
+      "loss": 2.2233,
       "step": 7955
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.671875,
       "learning_rate": 0.001,
-      "loss": 2.2503,
+      "loss": 2.1445,
       "step": 7960
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.306640625,
+      "grad_norm": 0.953125,
       "learning_rate": 0.001,
-      "loss": 2.1995,
+      "loss": 2.145,
       "step": 7965
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.5703125,
       "learning_rate": 0.001,
-      "loss": 2.2804,
+      "loss": 2.2116,
       "step": 7970
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.2591,
+      "loss": 2.1848,
       "step": 7975
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.310546875,
+      "grad_norm": 0.96484375,
       "learning_rate": 0.001,
-      "loss": 2.2143,
+      "loss": 2.1689,
       "step": 7980
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.298828125,
+      "grad_norm": 1.3828125,
       "learning_rate": 0.001,
-      "loss": 2.2275,
+      "loss": 2.2128,
       "step": 7985
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.8828125,
       "learning_rate": 0.001,
-      "loss": 2.3919,
+      "loss": 2.2035,
       "step": 7990
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.309,
+      "loss": 2.1877,
       "step": 7995
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.345703125,
+      "grad_norm": 1.0078125,
       "learning_rate": 0.001,
-      "loss": 2.2363,
+      "loss": 2.2685,
       "step": 8000
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.298828125,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.3152,
+      "loss": 2.1212,
       "step": 8005
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.2732,
+      "loss": 2.1883,
       "step": 8010
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.2561,
+      "loss": 2.1043,
       "step": 8015
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.349609375,
+      "grad_norm": 0.83984375,
       "learning_rate": 0.001,
-      "loss": 2.2615,
+      "loss": 2.27,
       "step": 8020
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.30859375,
+      "grad_norm": 2.578125,
       "learning_rate": 0.001,
-      "loss": 2.2131,
+      "loss": 2.1822,
       "step": 8025
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.2894,
+      "loss": 2.2655,
       "step": 8030
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.306640625,
+      "grad_norm": 0.94140625,
       "learning_rate": 0.001,
-      "loss": 2.3007,
+      "loss": 2.2037,
       "step": 8035
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.306640625,
+      "grad_norm": 0.90625,
       "learning_rate": 0.001,
-      "loss": 2.2376,
+      "loss": 2.1545,
       "step": 8040
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.265625,
       "learning_rate": 0.001,
-      "loss": 2.3124,
+      "loss": 2.1945,
       "step": 8045
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.1713,
+      "loss": 2.2214,
       "step": 8050
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.59375,
       "learning_rate": 0.001,
-      "loss": 2.3618,
+      "loss": 2.1683,
       "step": 8055
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.7421875,
       "learning_rate": 0.001,
-      "loss": 2.3386,
+      "loss": 2.1066,
       "step": 8060
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.1484375,
       "learning_rate": 0.001,
-      "loss": 2.2925,
+      "loss": 2.2586,
       "step": 8065
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.2421875,
       "learning_rate": 0.001,
-      "loss": 2.2123,
+      "loss": 2.1627,
       "step": 8070
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.484375,
       "learning_rate": 0.001,
-      "loss": 2.2016,
+      "loss": 2.2267,
       "step": 8075
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.2361,
+      "loss": 2.1868,
       "step": 8080
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.2677,
+      "loss": 2.1543,
       "step": 8085
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.3125,
+      "grad_norm": 2.296875,
       "learning_rate": 0.001,
-      "loss": 2.2364,
+      "loss": 2.191,
       "step": 8090
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.314453125,
+      "grad_norm": 0.8515625,
       "learning_rate": 0.001,
-      "loss": 2.2845,
+      "loss": 2.1399,
       "step": 8095
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.291015625,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.2399,
+      "loss": 2.1241,
       "step": 8100
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.5546875,
       "learning_rate": 0.001,
-      "loss": 2.2638,
+      "loss": 2.0788,
       "step": 8105
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.2932,
+      "loss": 2.154,
       "step": 8110
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.2401,
+      "loss": 2.1215,
       "step": 8115
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.2824,
+      "loss": 2.1879,
       "step": 8120
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.35546875,
+      "grad_norm": 2.171875,
       "learning_rate": 0.001,
-      "loss": 2.2077,
+      "loss": 2.1597,
       "step": 8125
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.2046,
+      "loss": 2.2198,
       "step": 8130
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.375,
       "learning_rate": 0.001,
-      "loss": 2.1893,
+      "loss": 2.2092,
       "step": 8135
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.306640625,
+      "grad_norm": 0.98046875,
       "learning_rate": 0.001,
-      "loss": 2.2769,
+      "loss": 2.177,
       "step": 8140
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.217,
+      "loss": 2.1563,
       "step": 8145
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.330078125,
+      "grad_norm": 0.90625,
       "learning_rate": 0.001,
-      "loss": 2.2389,
+      "loss": 2.1434,
       "step": 8150
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.291015625,
+      "grad_norm": 1.0,
       "learning_rate": 0.001,
-      "loss": 2.266,
+      "loss": 2.1589,
       "step": 8155
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.5625,
       "learning_rate": 0.001,
-      "loss": 2.2453,
+      "loss": 2.1215,
       "step": 8160
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.314453125,
+      "grad_norm": 0.91015625,
       "learning_rate": 0.001,
-      "loss": 2.281,
+      "loss": 2.2276,
       "step": 8165
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.2578125,
       "learning_rate": 0.001,
-      "loss": 2.2076,
+      "loss": 2.2408,
       "step": 8170
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.298828125,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.1804,
+      "loss": 2.1113,
       "step": 8175
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.0703125,
       "learning_rate": 0.001,
-      "loss": 2.2309,
+      "loss": 2.1163,
       "step": 8180
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.4296875,
       "learning_rate": 0.001,
-      "loss": 2.2703,
+      "loss": 2.1909,
       "step": 8185
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.5859375,
       "learning_rate": 0.001,
-      "loss": 2.3598,
+      "loss": 2.1932,
       "step": 8190
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.2038,
+      "loss": 2.2223,
       "step": 8195
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.3203125,
+      "grad_norm": 0.921875,
       "learning_rate": 0.001,
-      "loss": 2.2529,
+      "loss": 2.1728,
       "step": 8200
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.30078125,
+      "grad_norm": 0.93359375,
       "learning_rate": 0.001,
-      "loss": 2.2816,
+      "loss": 2.2604,
       "step": 8205
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.238,
+      "loss": 2.1194,
       "step": 8210
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.0078125,
       "learning_rate": 0.001,
-      "loss": 2.2579,
+      "loss": 2.0889,
       "step": 8215
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.294921875,
+      "grad_norm": 1.7421875,
       "learning_rate": 0.001,
-      "loss": 2.1992,
+      "loss": 2.1805,
       "step": 8220
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.0,
       "learning_rate": 0.001,
-      "loss": 2.1818,
+      "loss": 2.1755,
       "step": 8225
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.1506,
+      "loss": 2.2337,
       "step": 8230
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.298828125,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.2157,
+      "loss": 2.2678,
       "step": 8235
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.296875,
+      "grad_norm": 0.87890625,
       "learning_rate": 0.001,
-      "loss": 2.2456,
+      "loss": 2.1309,
       "step": 8240
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.078125,
       "learning_rate": 0.001,
-      "loss": 2.2597,
+      "loss": 2.1571,
       "step": 8245
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.0390625,
       "learning_rate": 0.001,
-      "loss": 2.2447,
+      "loss": 2.1756,
       "step": 8250
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.1631,
+      "loss": 2.2012,
       "step": 8255
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.390625,
       "learning_rate": 0.001,
-      "loss": 2.2394,
+      "loss": 2.1615,
       "step": 8260
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.046875,
       "learning_rate": 0.001,
-      "loss": 2.2786,
+      "loss": 2.19,
       "step": 8265
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.001,
-      "loss": 2.2592,
+      "loss": 2.1919,
       "step": 8270
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.2017,
+      "loss": 2.1195,
       "step": 8275
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.3046875,
       "learning_rate": 0.001,
-      "loss": 2.2654,
+      "loss": 2.1805,
       "step": 8280
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.95703125,
       "learning_rate": 0.001,
-      "loss": 2.2704,
+      "loss": 2.0899,
       "step": 8285
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.34375,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.3055,
+      "loss": 2.2288,
       "step": 8290
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.298828125,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.1985,
+      "loss": 2.1627,
       "step": 8295
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.0,
       "learning_rate": 0.001,
-      "loss": 2.259,
+      "loss": 2.0923,
       "step": 8300
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.276,
+      "loss": 2.1777,
       "step": 8305
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.1859,
+      "loss": 2.0752,
       "step": 8310
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.2696,
+      "loss": 2.1106,
       "step": 8315
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.001,
-      "loss": 2.2577,
+      "loss": 2.1276,
       "step": 8320
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.2533,
+      "loss": 2.1765,
       "step": 8325
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.306640625,
+      "grad_norm": 0.953125,
       "learning_rate": 0.001,
-      "loss": 2.3143,
+      "loss": 2.1726,
       "step": 8330
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.298828125,
+      "grad_norm": 0.95703125,
       "learning_rate": 0.001,
-      "loss": 2.2516,
+      "loss": 2.2546,
       "step": 8335
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.2624,
+      "loss": 2.1831,
       "step": 8340
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.2890625,
+      "grad_norm": 0.86328125,
       "learning_rate": 0.001,
-      "loss": 2.1764,
+      "loss": 2.1938,
       "step": 8345
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.98828125,
       "learning_rate": 0.001,
-      "loss": 2.2136,
+      "loss": 2.0966,
       "step": 8350
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.4765625,
       "learning_rate": 0.001,
-      "loss": 2.2297,
+      "loss": 2.0815,
       "step": 8355
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.2867,
+      "loss": 2.2203,
       "step": 8360
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.2896,
+      "loss": 2.0747,
       "step": 8365
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.302734375,
+      "grad_norm": 0.95703125,
       "learning_rate": 0.001,
-      "loss": 2.2175,
+      "loss": 2.159,
       "step": 8370
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.3125,
+      "grad_norm": 0.890625,
       "learning_rate": 0.001,
-      "loss": 2.19,
+      "loss": 2.1264,
       "step": 8375
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.134,
+      "loss": 2.1399,
       "step": 8380
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.318359375,
+      "grad_norm": 1.6171875,
       "learning_rate": 0.001,
-      "loss": 2.2358,
+      "loss": 2.1996,
       "step": 8385
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.287109375,
+      "grad_norm": 0.95703125,
       "learning_rate": 0.001,
-      "loss": 2.1986,
+      "loss": 2.2055,
       "step": 8390
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.203,
+      "loss": 2.2027,
       "step": 8395
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.287109375,
+      "grad_norm": 1.1328125,
       "learning_rate": 0.001,
-      "loss": 2.2181,
+      "loss": 2.1516,
       "step": 8400
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.294921875,
+      "grad_norm": 1.5546875,
       "learning_rate": 0.001,
-      "loss": 2.2655,
+      "loss": 2.1234,
       "step": 8405
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.0,
       "learning_rate": 0.001,
-      "loss": 2.2862,
+      "loss": 2.1325,
       "step": 8410
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.310546875,
+      "grad_norm": 0.9765625,
       "learning_rate": 0.001,
-      "loss": 2.3058,
+      "loss": 2.1561,
       "step": 8415
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.0859375,
       "learning_rate": 0.001,
-      "loss": 2.2694,
+      "loss": 2.2077,
       "step": 8420
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.0625,
       "learning_rate": 0.001,
-      "loss": 2.2549,
+      "loss": 2.0967,
       "step": 8425
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.34765625,
+      "grad_norm": 0.96484375,
       "learning_rate": 0.001,
-      "loss": 2.2976,
+      "loss": 2.1205,
       "step": 8430
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.31640625,
+      "grad_norm": 0.8203125,
       "learning_rate": 0.001,
-      "loss": 2.2406,
+      "loss": 2.1933,
       "step": 8435
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.306640625,
+      "grad_norm": 0.9375,
       "learning_rate": 0.001,
-      "loss": 2.2485,
+      "loss": 2.2021,
       "step": 8440
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.30859375,
+      "grad_norm": 0.7734375,
       "learning_rate": 0.001,
-      "loss": 2.1618,
+      "loss": 2.1096,
       "step": 8445
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.3203125,
+      "grad_norm": 0.93359375,
       "learning_rate": 0.001,
-      "loss": 2.2759,
+      "loss": 2.1006,
       "step": 8450
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.2618,
+      "loss": 2.0802,
       "step": 8455
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.314453125,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.203,
+      "loss": 2.1566,
       "step": 8460
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.1972,
+      "loss": 2.2041,
       "step": 8465
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.4921875,
       "learning_rate": 0.001,
-      "loss": 2.1704,
+      "loss": 2.2042,
       "step": 8470
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.2045,
+      "loss": 2.0913,
       "step": 8475
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.30859375,
+      "grad_norm": 0.94921875,
       "learning_rate": 0.001,
-      "loss": 2.188,
+      "loss": 2.1835,
       "step": 8480
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.94140625,
       "learning_rate": 0.001,
-      "loss": 2.2592,
+      "loss": 2.1648,
       "step": 8485
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.9609375,
       "learning_rate": 0.001,
-      "loss": 2.2913,
+      "loss": 2.1875,
       "step": 8490
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.294921875,
+      "grad_norm": 1.359375,
       "learning_rate": 0.001,
-      "loss": 2.2592,
+      "loss": 2.194,
       "step": 8495
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.310546875,
+      "grad_norm": 0.96484375,
       "learning_rate": 0.001,
-      "loss": 2.2068,
+      "loss": 2.098,
       "step": 8500
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.30078125,
+      "grad_norm": 0.9765625,
       "learning_rate": 0.001,
-      "loss": 2.2432,
+      "loss": 2.1823,
       "step": 8505
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.3125,
+      "grad_norm": 1.109375,
       "learning_rate": 0.001,
-      "loss": 2.1883,
+      "loss": 2.1069,
       "step": 8510
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.291015625,
+      "grad_norm": 0.94140625,
       "learning_rate": 0.001,
-      "loss": 2.2438,
+      "loss": 2.2228,
       "step": 8515
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.30078125,
+      "grad_norm": 0.9765625,
       "learning_rate": 0.001,
-      "loss": 2.1702,
+      "loss": 2.1541,
       "step": 8520
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.296875,
+      "grad_norm": 2.015625,
       "learning_rate": 0.001,
-      "loss": 2.2469,
+      "loss": 2.1089,
       "step": 8525
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.3203125,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.1156,
+      "loss": 2.1029,
       "step": 8530
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.322265625,
+      "grad_norm": 1.46875,
       "learning_rate": 0.001,
-      "loss": 2.1934,
+      "loss": 2.2125,
       "step": 8535
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.0078125,
       "learning_rate": 0.001,
-      "loss": 2.3777,
+      "loss": 2.1722,
       "step": 8540
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.306640625,
+      "grad_norm": 1.15625,
       "learning_rate": 0.001,
-      "loss": 2.2576,
+      "loss": 2.2255,
       "step": 8545
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.328125,
+      "grad_norm": 0.9453125,
       "learning_rate": 0.001,
-      "loss": 2.2462,
+      "loss": 2.1747,
       "step": 8550
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.33203125,
+      "grad_norm": 0.9453125,
       "learning_rate": 0.001,
-      "loss": 2.2936,
+      "loss": 2.1575,
       "step": 8555
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.890625,
       "learning_rate": 0.001,
-      "loss": 2.3116,
+      "loss": 2.24,
       "step": 8560
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.310546875,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.2689,
+      "loss": 2.2122,
       "step": 8565
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.36328125,
+      "grad_norm": 1.4375,
       "learning_rate": 0.001,
-      "loss": 2.3067,
+      "loss": 2.0925,
       "step": 8570
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.32421875,
+      "grad_norm": 0.89453125,
       "learning_rate": 0.001,
-      "loss": 2.3342,
+      "loss": 2.2339,
       "step": 8575
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.30078125,
+      "grad_norm": 0.94140625,
       "learning_rate": 0.001,
-      "loss": 2.2079,
+      "loss": 2.1486,
       "step": 8580
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.33984375,
+      "grad_norm": 1.2109375,
       "learning_rate": 0.001,
-      "loss": 2.243,
+      "loss": 2.208,
       "step": 8585
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.314453125,
+      "grad_norm": 0.828125,
       "learning_rate": 0.001,
-      "loss": 2.2,
+      "loss": 2.1253,
       "step": 8590
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.2890625,
+      "grad_norm": 1.171875,
       "learning_rate": 0.001,
-      "loss": 2.2437,
+      "loss": 2.1997,
       "step": 8595
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.298828125,
+      "grad_norm": 0.89453125,
       "learning_rate": 0.001,
-      "loss": 2.1937,
+      "loss": 2.0718,
       "step": 8600
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.314453125,
+      "grad_norm": 0.87890625,
       "learning_rate": 0.001,
-      "loss": 2.2336,
+      "loss": 2.1493,
       "step": 8605
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.314453125,
+      "grad_norm": 0.94921875,
       "learning_rate": 0.001,
-      "loss": 2.2758,
+      "loss": 2.1384,
       "step": 8610
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.345703125,
+      "grad_norm": 0.8984375,
       "learning_rate": 0.001,
-      "loss": 2.2071,
+      "loss": 2.1194,
       "step": 8615
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.3125,
+      "grad_norm": 0.95703125,
       "learning_rate": 0.001,
-      "loss": 2.149,
+      "loss": 2.1114,
       "step": 8620
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.318359375,
+      "grad_norm": 0.96875,
       "learning_rate": 0.001,
-      "loss": 2.257,
+      "loss": 2.0944,
       "step": 8625
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.296875,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.001,
-      "loss": 2.207,
+      "loss": 2.1073,
       "step": 8630
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.310546875,
+      "grad_norm": 0.875,
       "learning_rate": 0.001,
-      "loss": 2.2858,
+      "loss": 2.0895,
       "step": 8635
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.328125,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.2318,
+      "loss": 2.2211,
       "step": 8640
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.0,
       "learning_rate": 0.001,
-      "loss": 2.295,
+      "loss": 2.1039,
       "step": 8645
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.310546875,
+      "grad_norm": 0.94921875,
       "learning_rate": 0.001,
-      "loss": 2.1429,
+      "loss": 2.1231,
       "step": 8650
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.1171875,
       "learning_rate": 0.001,
-      "loss": 2.1635,
+      "loss": 2.128,
       "step": 8655
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.302734375,
+      "grad_norm": 0.8984375,
       "learning_rate": 0.001,
-      "loss": 2.3268,
+      "loss": 2.1983,
       "step": 8660
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.306640625,
+      "grad_norm": 0.95703125,
       "learning_rate": 0.001,
-      "loss": 2.2371,
+      "loss": 2.1355,
       "step": 8665
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.298828125,
+      "grad_norm": 0.9375,
       "learning_rate": 0.001,
-      "loss": 2.2604,
+      "loss": 2.175,
       "step": 8670
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.30859375,
+      "grad_norm": 0.8515625,
       "learning_rate": 0.001,
-      "loss": 2.2545,
+      "loss": 2.1366,
       "step": 8675
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.984375,
       "learning_rate": 0.001,
-      "loss": 2.1592,
+      "loss": 2.1074,
       "step": 8680
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.3125,
+      "grad_norm": 0.83203125,
       "learning_rate": 0.001,
-      "loss": 2.2776,
+      "loss": 2.0969,
       "step": 8685
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.302734375,
+      "grad_norm": 0.9296875,
       "learning_rate": 0.001,
-      "loss": 2.213,
+      "loss": 2.1339,
       "step": 8690
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.287109375,
+      "grad_norm": 0.89453125,
       "learning_rate": 0.001,
-      "loss": 2.1203,
+      "loss": 2.1782,
       "step": 8695
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.31640625,
+      "grad_norm": 0.96875,
       "learning_rate": 0.001,
-      "loss": 2.3063,
+      "loss": 2.141,
       "step": 8700
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.296875,
+      "grad_norm": 1.0078125,
       "learning_rate": 0.001,
-      "loss": 2.2479,
+      "loss": 2.1213,
       "step": 8705
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.306640625,
+      "grad_norm": 0.92578125,
       "learning_rate": 0.001,
-      "loss": 2.2708,
+      "loss": 2.1564,
       "step": 8710
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.326171875,
+      "grad_norm": 1.28125,
       "learning_rate": 0.001,
-      "loss": 2.308,
+      "loss": 2.2358,
       "step": 8715
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.03125,
       "learning_rate": 0.001,
-      "loss": 2.2732,
+      "loss": 2.1675,
       "step": 8720
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.296875,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.2556,
+      "loss": 2.1651,
       "step": 8725
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.3359375,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.2082,
+      "loss": 2.1643,
       "step": 8730
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.8359375,
       "learning_rate": 0.001,
-      "loss": 2.2652,
+      "loss": 2.1753,
       "step": 8735
     },
     {
       "epoch": 0.97,
-      "grad_norm": 0.296875,
+      "grad_norm": 1.125,
       "learning_rate": 0.001,
-      "loss": 2.3049,
+      "loss": 2.2613,
       "step": 8740
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.234375,
       "learning_rate": 0.001,
-      "loss": 2.2438,
+      "loss": 2.1604,
       "step": 8745
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.1015625,
       "learning_rate": 0.001,
-      "loss": 2.1741,
+      "loss": 2.1436,
       "step": 8750
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.734375,
       "learning_rate": 0.001,
-      "loss": 2.2807,
+      "loss": 2.1084,
       "step": 8755
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.322265625,
+      "grad_norm": 0.9296875,
       "learning_rate": 0.001,
-      "loss": 2.2431,
+      "loss": 2.1185,
       "step": 8760
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.291015625,
+      "grad_norm": 1.3125,
       "learning_rate": 0.001,
-      "loss": 2.2556,
+      "loss": 2.1427,
       "step": 8765
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.302734375,
+      "grad_norm": 0.86328125,
       "learning_rate": 0.001,
-      "loss": 2.2498,
+      "loss": 2.1018,
       "step": 8770
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.322265625,
+      "grad_norm": 0.96484375,
       "learning_rate": 0.001,
-      "loss": 2.2788,
+      "loss": 2.2008,
       "step": 8775
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.29296875,
+      "grad_norm": 0.97265625,
       "learning_rate": 0.001,
-      "loss": 2.2636,
+      "loss": 2.0736,
       "step": 8780
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.330078125,
+      "grad_norm": 1.40625,
       "learning_rate": 0.001,
-      "loss": 2.2968,
+      "loss": 2.1633,
       "step": 8785
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.306640625,
+      "grad_norm": 0.78125,
       "learning_rate": 0.001,
-      "loss": 2.2984,
+      "loss": 2.2193,
       "step": 8790
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.302734375,
+      "grad_norm": 0.7890625,
       "learning_rate": 0.001,
-      "loss": 2.2495,
+      "loss": 2.209,
       "step": 8795
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.322265625,
+      "grad_norm": 0.921875,
       "learning_rate": 0.001,
-      "loss": 2.2131,
+      "loss": 2.1654,
       "step": 8800
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.33203125,
+      "grad_norm": 1.40625,
       "learning_rate": 0.001,
-      "loss": 2.176,
+      "loss": 2.183,
       "step": 8805
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.890625,
       "learning_rate": 0.001,
-      "loss": 2.3043,
+      "loss": 2.1428,
       "step": 8810
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.30078125,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.2409,
+      "loss": 2.1481,
       "step": 8815
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.92578125,
       "learning_rate": 0.001,
-      "loss": 2.1861,
+      "loss": 2.0974,
       "step": 8820
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.328125,
       "learning_rate": 0.001,
-      "loss": 2.2472,
+      "loss": 2.04,
       "step": 8825
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.306640625,
+      "grad_norm": 0.9140625,
       "learning_rate": 0.001,
-      "loss": 2.2569,
+      "loss": 2.1917,
       "step": 8830
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.32421875,
+      "grad_norm": 1.5234375,
       "learning_rate": 0.001,
-      "loss": 2.2432,
+      "loss": 2.1512,
       "step": 8835
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.298828125,
+      "grad_norm": 0.95703125,
       "learning_rate": 0.001,
-      "loss": 2.2889,
+      "loss": 2.1508,
       "step": 8840
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.30859375,
+      "grad_norm": 0.890625,
       "learning_rate": 0.001,
-      "loss": 2.1974,
+      "loss": 2.1739,
       "step": 8845
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.3046875,
+      "grad_norm": 0.91796875,
       "learning_rate": 0.001,
-      "loss": 2.3096,
+      "loss": 2.1486,
       "step": 8850
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.30078125,
+      "grad_norm": 0.828125,
       "learning_rate": 0.001,
-      "loss": 2.1649,
+      "loss": 2.13,
       "step": 8855
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.298828125,
+      "grad_norm": 0.94140625,
       "learning_rate": 0.001,
-      "loss": 2.2163,
+      "loss": 2.1971,
       "step": 8860
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.302734375,
+      "grad_norm": 1.1953125,
       "learning_rate": 0.001,
-      "loss": 2.2611,
+      "loss": 2.1131,
       "step": 8865
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.30859375,
+      "grad_norm": 0.9296875,
       "learning_rate": 0.001,
-      "loss": 2.245,
+      "loss": 2.1483,
       "step": 8870
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.30078125,
+      "grad_norm": 0.8515625,
       "learning_rate": 0.001,
-      "loss": 2.1904,
+      "loss": 2.1244,
       "step": 8875
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.310546875,
+      "grad_norm": 0.82421875,
       "learning_rate": 0.001,
-      "loss": 2.2698,
+      "loss": 2.1369,
       "step": 8880
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.3046875,
+      "grad_norm": 1.25,
       "learning_rate": 0.001,
-      "loss": 2.2447,
+      "loss": 2.1743,
       "step": 8885
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.349609375,
+      "grad_norm": 1.015625,
       "learning_rate": 0.001,
-      "loss": 2.2229,
+      "loss": 2.0747,
       "step": 8890
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.294921875,
+      "grad_norm": 0.890625,
       "learning_rate": 0.001,
-      "loss": 2.1953,
+      "loss": 2.1792,
       "step": 8895
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.291015625,
+      "grad_norm": 0.8203125,
       "learning_rate": 0.001,
-      "loss": 2.2675,
+      "loss": 2.145,
       "step": 8900
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.322265625,
+      "grad_norm": 0.95703125,
       "learning_rate": 0.001,
-      "loss": 2.3104,
+      "loss": 2.1802,
       "step": 8905
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.31640625,
+      "grad_norm": 1.21875,
       "learning_rate": 0.001,
-      "loss": 2.3058,
+      "loss": 2.067,
       "step": 8910
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.294921875,
+      "grad_norm": 0.76171875,
       "learning_rate": 0.001,
-      "loss": 2.1981,
+      "loss": 2.1613,
       "step": 8915
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.30859375,
+      "grad_norm": 0.8046875,
       "learning_rate": 0.001,
-      "loss": 2.1727,
+      "loss": 2.142,
       "step": 8920
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.3125,
+      "grad_norm": 0.90625,
       "learning_rate": 0.001,
-      "loss": 2.1917,
+      "loss": 2.1794,
       "step": 8925
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.291015625,
+      "grad_norm": 0.90234375,
       "learning_rate": 0.001,
-      "loss": 2.1705,
+      "loss": 2.071,
       "step": 8930
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.3125,
+      "grad_norm": 0.95703125,
       "learning_rate": 0.001,
-      "loss": 2.2193,
+      "loss": 2.1635,
       "step": 8935
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.30859375,
+      "grad_norm": 0.875,
       "learning_rate": 0.001,
-      "loss": 2.2683,
+      "loss": 2.2223,
       "step": 8940
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.30859375,
+      "grad_norm": 0.875,
       "learning_rate": 0.001,
-      "loss": 2.1673,
+      "loss": 2.0432,
       "step": 8945
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.2890625,
+      "grad_norm": 0.91796875,
       "learning_rate": 0.001,
-      "loss": 2.2151,
+      "loss": 2.0315,
       "step": 8950
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.341796875,
+      "grad_norm": 0.87890625,
       "learning_rate": 0.001,
-      "loss": 2.2106,
+      "loss": 2.1065,
       "step": 8955
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.318359375,
+      "grad_norm": 0.9375,
       "learning_rate": 0.001,
-      "loss": 2.2439,
+      "loss": 2.1791,
       "step": 8960
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.30859375,
+      "grad_norm": 1.0234375,
       "learning_rate": 0.001,
-      "loss": 2.2279,
+      "loss": 2.2013,
       "step": 8965
     },
     {
       "epoch": 1.0,
-      "eval_loss": 2.2348995208740234,
-      "eval_runtime": 2639.8137,
-      "eval_samples_per_second": 6.016,
-      "eval_steps_per_second": 0.752,
+      "eval_loss": 2.164351224899292,
+      "eval_runtime": 1701.8515,
+      "eval_samples_per_second": 9.332,
+      "eval_steps_per_second": 1.167,
       "step": 8969
     },
     {
       "epoch": 1.0,
       "step": 8969,
-      "total_flos": 1.50079627163861e+17,
-      "train_loss": 2.666482654486044,
-      "train_runtime": 30048.5081,
-      "train_samples_per_second": 4.775,
-      "train_steps_per_second": 0.298
+      "total_flos": 1.8254804498880922e+18,
+      "train_loss": 2.789915239123786,
+      "train_runtime": 47169.5182,
+      "train_samples_per_second": 3.042,
+      "train_steps_per_second": 0.19
     }
   ],
   "logging_steps": 5,
@@ -12589,7 +12589,7 @@
   "num_input_tokens_seen": 0,
   "num_train_epochs": 1,
   "save_steps": 1000,
-  "total_flos": 1.50079627163861e+17,
+  "total_flos": 1.8254804498880922e+18,
   "train_batch_size": 16,
   "trial_name": null,
   "trial_params": null