diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5465 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.9988358556461003,
+  "eval_steps": 500,
+  "global_step": 3864,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 1.796875,
+      "learning_rate": 8.620689655172414e-08,
+      "loss": 3.4377,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 1.7890625,
+      "learning_rate": 4.3103448275862073e-07,
+      "loss": 3.4043,
+      "step": 5
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.640625,
+      "learning_rate": 8.620689655172415e-07,
+      "loss": 3.4252,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.6640625,
+      "learning_rate": 1.2931034482758623e-06,
+      "loss": 3.3862,
+      "step": 15
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.8125,
+      "learning_rate": 1.724137931034483e-06,
+      "loss": 3.3863,
+      "step": 20
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2.1551724137931035e-06,
+      "loss": 3.3899,
+      "step": 25
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.546875,
+      "learning_rate": 2.5862068965517246e-06,
+      "loss": 3.3691,
+      "step": 30
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.640625,
+      "learning_rate": 3.017241379310345e-06,
+      "loss": 3.3793,
+      "step": 35
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.453125,
+      "learning_rate": 3.448275862068966e-06,
+      "loss": 3.3317,
+      "step": 40
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.390625,
+      "learning_rate": 3.8793103448275865e-06,
+      "loss": 3.3268,
+      "step": 45
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.1484375,
+      "learning_rate": 4.310344827586207e-06,
+      "loss": 3.291,
+      "step": 50
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.0546875,
+      "learning_rate": 4.741379310344828e-06,
+      "loss": 3.2941,
+      "step": 55
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.99609375,
+      "learning_rate": 5.172413793103449e-06,
+      "loss": 3.2778,
+      "step": 60
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.91796875,
+      "learning_rate": 5.603448275862069e-06,
+      "loss": 3.2182,
+      "step": 65
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.8203125,
+      "learning_rate": 6.03448275862069e-06,
+      "loss": 3.2205,
+      "step": 70
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.80078125,
+      "learning_rate": 6.465517241379311e-06,
+      "loss": 3.1764,
+      "step": 75
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.76953125,
+      "learning_rate": 6.896551724137932e-06,
+      "loss": 3.1814,
+      "step": 80
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.76953125,
+      "learning_rate": 7.327586206896552e-06,
+      "loss": 3.1258,
+      "step": 85
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.7265625,
+      "learning_rate": 7.758620689655173e-06,
+      "loss": 3.0397,
+      "step": 90
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.7421875,
+      "learning_rate": 8.189655172413794e-06,
+      "loss": 3.0763,
+      "step": 95
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.73828125,
+      "learning_rate": 8.620689655172414e-06,
+      "loss": 3.0838,
+      "step": 100
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.80078125,
+      "learning_rate": 9.051724137931036e-06,
+      "loss": 2.9988,
+      "step": 105
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.72265625,
+      "learning_rate": 9.482758620689655e-06,
+      "loss": 2.9653,
+      "step": 110
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.7265625,
+      "learning_rate": 9.913793103448277e-06,
+      "loss": 2.9536,
+      "step": 115
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.67578125,
+      "learning_rate": 9.999971896515836e-06,
+      "loss": 2.8924,
+      "step": 120
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.65234375,
+      "learning_rate": 9.999857726652867e-06,
+      "loss": 2.8778,
+      "step": 125
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.60546875,
+      "learning_rate": 9.999655735947143e-06,
+      "loss": 2.8204,
+      "step": 130
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.5390625,
+      "learning_rate": 9.999365927946568e-06,
+      "loss": 2.8143,
+      "step": 135
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.6015625,
+      "learning_rate": 9.998988307741521e-06,
+      "loss": 2.8075,
+      "step": 140
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.58203125,
+      "learning_rate": 9.998522881964777e-06,
+      "loss": 2.7778,
+      "step": 145
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.4921875,
+      "learning_rate": 9.997969658791384e-06,
+      "loss": 2.6992,
+      "step": 150
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.51953125,
+      "learning_rate": 9.997328647938524e-06,
+      "loss": 2.7048,
+      "step": 155
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.4609375,
+      "learning_rate": 9.996599860665342e-06,
+      "loss": 2.7525,
+      "step": 160
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.478515625,
+      "learning_rate": 9.995783309772743e-06,
+      "loss": 2.7152,
+      "step": 165
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.4921875,
+      "learning_rate": 9.99487900960317e-06,
+      "loss": 2.6601,
+      "step": 170
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.439453125,
+      "learning_rate": 9.99388697604036e-06,
+      "loss": 2.6125,
+      "step": 175
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.423828125,
+      "learning_rate": 9.99280722650905e-06,
+      "loss": 2.6056,
+      "step": 180
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.5234375,
+      "learning_rate": 9.991639779974688e-06,
+      "loss": 2.6514,
+      "step": 185
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.462890625,
+      "learning_rate": 9.99038465694308e-06,
+      "loss": 2.641,
+      "step": 190
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.462890625,
+      "learning_rate": 9.989041879460046e-06,
+      "loss": 2.6054,
+      "step": 195
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.474609375,
+      "learning_rate": 9.987611471111027e-06,
+      "loss": 2.596,
+      "step": 200
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.42578125,
+      "learning_rate": 9.986093457020676e-06,
+      "loss": 2.5794,
+      "step": 205
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.42578125,
+      "learning_rate": 9.984487863852401e-06,
+      "loss": 2.5771,
+      "step": 210
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.435546875,
+      "learning_rate": 9.982794719807916e-06,
+      "loss": 2.54,
+      "step": 215
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.421875,
+      "learning_rate": 9.981014054626737e-06,
+      "loss": 2.5216,
+      "step": 220
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.412109375,
+      "learning_rate": 9.979145899585653e-06,
+      "loss": 2.5178,
+      "step": 225
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.439453125,
+      "learning_rate": 9.977190287498191e-06,
+      "loss": 2.5356,
+      "step": 230
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.4140625,
+      "learning_rate": 9.975147252714025e-06,
+      "loss": 2.5361,
+      "step": 235
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.423828125,
+      "learning_rate": 9.973016831118389e-06,
+      "loss": 2.5285,
+      "step": 240
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.408203125,
+      "learning_rate": 9.970799060131428e-06,
+      "loss": 2.4848,
+      "step": 245
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.41015625,
+      "learning_rate": 9.968493978707555e-06,
+      "loss": 2.4364,
+      "step": 250
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.455078125,
+      "learning_rate": 9.966101627334758e-06,
+      "loss": 2.4305,
+      "step": 255
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.390625,
+      "learning_rate": 9.9636220480339e-06,
+      "loss": 2.4391,
+      "step": 260
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.396484375,
+      "learning_rate": 9.961055284357965e-06,
+      "loss": 2.4398,
+      "step": 265
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.423828125,
+      "learning_rate": 9.958401381391306e-06,
+      "loss": 2.5026,
+      "step": 270
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.3828125,
+      "learning_rate": 9.955660385748853e-06,
+      "loss": 2.4484,
+      "step": 275
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.41796875,
+      "learning_rate": 9.952832345575283e-06,
+      "loss": 2.4266,
+      "step": 280
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.37890625,
+      "learning_rate": 9.94991731054418e-06,
+      "loss": 2.4405,
+      "step": 285
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.3671875,
+      "learning_rate": 9.946915331857172e-06,
+      "loss": 2.4422,
+      "step": 290
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.396484375,
+      "learning_rate": 9.943826462243019e-06,
+      "loss": 2.4495,
+      "step": 295
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.390625,
+      "learning_rate": 9.940650755956686e-06,
+      "loss": 2.3972,
+      "step": 300
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.412109375,
+      "learning_rate": 9.937388268778409e-06,
+      "loss": 2.4367,
+      "step": 305
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.38671875,
+      "learning_rate": 9.934039058012685e-06,
+      "loss": 2.3607,
+      "step": 310
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.341796875,
+      "learning_rate": 9.930603182487297e-06,
+      "loss": 2.393,
+      "step": 315
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.431640625,
+      "learning_rate": 9.927080702552256e-06,
+      "loss": 2.3503,
+      "step": 320
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.396484375,
+      "learning_rate": 9.923471680078753e-06,
+      "loss": 2.392,
+      "step": 325
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.408203125,
+      "learning_rate": 9.919776178458072e-06,
+      "loss": 2.3518,
+      "step": 330
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.392578125,
+      "learning_rate": 9.915994262600475e-06,
+      "loss": 2.3383,
+      "step": 335
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.357421875,
+      "learning_rate": 9.912125998934055e-06,
+      "loss": 2.3516,
+      "step": 340
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.4140625,
+      "learning_rate": 9.908171455403586e-06,
+      "loss": 2.3174,
+      "step": 345
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.443359375,
+      "learning_rate": 9.90413070146931e-06,
+      "loss": 2.3188,
+      "step": 350
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.365234375,
+      "learning_rate": 9.900003808105729e-06,
+      "loss": 2.3135,
+      "step": 355
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.375,
+      "learning_rate": 9.895790847800361e-06,
+      "loss": 2.3136,
+      "step": 360
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.38671875,
+      "learning_rate": 9.891491894552452e-06,
+      "loss": 2.3092,
+      "step": 365
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.376953125,
+      "learning_rate": 9.887107023871692e-06,
+      "loss": 2.2987,
+      "step": 370
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.359375,
+      "learning_rate": 9.882636312776878e-06,
+      "loss": 2.3219,
+      "step": 375
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.3515625,
+      "learning_rate": 9.878079839794572e-06,
+      "loss": 2.2945,
+      "step": 380
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.359375,
+      "learning_rate": 9.873437684957705e-06,
+      "loss": 2.3071,
+      "step": 385
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.41015625,
+      "learning_rate": 9.868709929804193e-06,
+      "loss": 2.3112,
+      "step": 390
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.41796875,
+      "learning_rate": 9.863896657375485e-06,
+      "loss": 2.2789,
+      "step": 395
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.36328125,
+      "learning_rate": 9.858997952215112e-06,
+      "loss": 2.2764,
+      "step": 400
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.390625,
+      "learning_rate": 9.85401390036721e-06,
+      "loss": 2.3129,
+      "step": 405
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.3671875,
+      "learning_rate": 9.848944589374993e-06,
+      "loss": 2.3237,
+      "step": 410
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.37890625,
+      "learning_rate": 9.84379010827923e-06,
+      "loss": 2.2283,
+      "step": 415
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.375,
+      "learning_rate": 9.838550547616671e-06,
+      "loss": 2.2523,
+      "step": 420
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.37109375,
+      "learning_rate": 9.833225999418461e-06,
+      "loss": 2.2641,
+      "step": 425
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.396484375,
+      "learning_rate": 9.827816557208523e-06,
+      "loss": 2.2737,
+      "step": 430
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.396484375,
+      "learning_rate": 9.822322316001917e-06,
+      "loss": 2.2647,
+      "step": 435
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.390625,
+      "learning_rate": 9.816743372303166e-06,
+      "loss": 2.2409,
+      "step": 440
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.353515625,
+      "learning_rate": 9.811079824104566e-06,
+      "loss": 2.2018,
+      "step": 445
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.369140625,
+      "learning_rate": 9.805331770884462e-06,
+      "loss": 2.2343,
+      "step": 450
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.412109375,
+      "learning_rate": 9.799499313605506e-06,
+      "loss": 2.2398,
+      "step": 455
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.40234375,
+      "learning_rate": 9.793582554712873e-06,
+      "loss": 2.2474,
+      "step": 460
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.390625,
+      "learning_rate": 9.787581598132469e-06,
+      "loss": 2.2416,
+      "step": 465
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.365234375,
+      "learning_rate": 9.781496549269108e-06,
+      "loss": 2.2005,
+      "step": 470
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.41015625,
+      "learning_rate": 9.775327515004656e-06,
+      "loss": 2.2498,
+      "step": 475
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.359375,
+      "learning_rate": 9.769074603696153e-06,
+      "loss": 2.2456,
+      "step": 480
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.365234375,
+      "learning_rate": 9.762737925173914e-06,
+      "loss": 2.2294,
+      "step": 485
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.390625,
+      "learning_rate": 9.756317590739592e-06,
+      "loss": 2.1614,
+      "step": 490
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.40234375,
+      "learning_rate": 9.749813713164233e-06,
+      "loss": 2.1949,
+      "step": 495
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.400390625,
+      "learning_rate": 9.743226406686293e-06,
+      "loss": 2.2037,
+      "step": 500
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.345703125,
+      "learning_rate": 9.736555787009622e-06,
+      "loss": 2.2386,
+      "step": 505
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.38671875,
+      "learning_rate": 9.729801971301443e-06,
+      "loss": 2.1857,
+      "step": 510
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.390625,
+      "learning_rate": 9.72296507819029e-06,
+      "loss": 2.2382,
+      "step": 515
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.365234375,
+      "learning_rate": 9.716045227763923e-06,
+      "loss": 2.1812,
+      "step": 520
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.376953125,
+      "learning_rate": 9.70904254156722e-06,
+      "loss": 2.1713,
+      "step": 525
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.357421875,
+      "learning_rate": 9.70195714260004e-06,
+      "loss": 2.1758,
+      "step": 530
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.330078125,
+      "learning_rate": 9.694789155315067e-06,
+      "loss": 2.1967,
+      "step": 535
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.40234375,
+      "learning_rate": 9.68753870561562e-06,
+      "loss": 2.2178,
+      "step": 540
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.357421875,
+      "learning_rate": 9.680205920853442e-06,
+      "loss": 2.1908,
+      "step": 545
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.38671875,
+      "learning_rate": 9.67279092982647e-06,
+      "loss": 2.1777,
+      "step": 550
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.412109375,
+      "learning_rate": 9.665293862776557e-06,
+      "loss": 2.1835,
+      "step": 555
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.361328125,
+      "learning_rate": 9.657714851387204e-06,
+      "loss": 2.1737,
+      "step": 560
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.376953125,
+      "learning_rate": 9.650054028781226e-06,
+      "loss": 2.1535,
+      "step": 565
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.392578125,
+      "learning_rate": 9.642311529518438e-06,
+      "loss": 2.1533,
+      "step": 570
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.34375,
+      "learning_rate": 9.634487489593268e-06,
+      "loss": 2.1289,
+      "step": 575
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.380859375,
+      "learning_rate": 9.626582046432384e-06,
+      "loss": 2.1073,
+      "step": 580
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.369140625,
+      "learning_rate": 9.618595338892272e-06,
+      "loss": 2.1438,
+      "step": 585
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.33984375,
+      "learning_rate": 9.610527507256803e-06,
+      "loss": 2.1822,
+      "step": 590
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.419921875,
+      "learning_rate": 9.602378693234762e-06,
+      "loss": 2.1667,
+      "step": 595
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.34765625,
+      "learning_rate": 9.594149039957366e-06,
+      "loss": 2.1701,
+      "step": 600
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.380859375,
+      "learning_rate": 9.585838691975744e-06,
+      "loss": 2.1239,
+      "step": 605
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.388671875,
+      "learning_rate": 9.577447795258403e-06,
+      "loss": 2.19,
+      "step": 610
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.39453125,
+      "learning_rate": 9.56897649718866e-06,
+      "loss": 2.1609,
+      "step": 615
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.365234375,
+      "learning_rate": 9.56042494656206e-06,
+      "loss": 2.1666,
+      "step": 620
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.384765625,
+      "learning_rate": 9.551793293583749e-06,
+      "loss": 2.1561,
+      "step": 625
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.380859375,
+      "learning_rate": 9.54308168986585e-06,
+      "loss": 2.1633,
+      "step": 630
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.41796875,
+      "learning_rate": 9.534290288424795e-06,
+      "loss": 2.1065,
+      "step": 635
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.423828125,
+      "learning_rate": 9.525419243678633e-06,
+      "loss": 2.1084,
+      "step": 640
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.361328125,
+      "learning_rate": 9.516468711444323e-06,
+      "loss": 2.1653,
+      "step": 645
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.423828125,
+      "learning_rate": 9.507438848934996e-06,
+      "loss": 2.1056,
+      "step": 650
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.333984375,
+      "learning_rate": 9.49832981475719e-06,
+      "loss": 2.1215,
+      "step": 655
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.388671875,
+      "learning_rate": 9.48914176890807e-06,
+      "loss": 2.1386,
+      "step": 660
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.396484375,
+      "learning_rate": 9.479874872772618e-06,
+      "loss": 2.1141,
+      "step": 665
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.392578125,
+      "learning_rate": 9.470529289120786e-06,
+      "loss": 2.151,
+      "step": 670
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.3828125,
+      "learning_rate": 9.461105182104655e-06,
+      "loss": 2.0897,
+      "step": 675
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.375,
+      "learning_rate": 9.451602717255536e-06,
+      "loss": 2.1514,
+      "step": 680
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.37109375,
+      "learning_rate": 9.442022061481076e-06,
+      "loss": 2.1311,
+      "step": 685
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.37890625,
+      "learning_rate": 9.432363383062314e-06,
+      "loss": 2.1416,
+      "step": 690
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.439453125,
+      "learning_rate": 9.422626851650733e-06,
+      "loss": 2.1198,
+      "step": 695
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.361328125,
+      "learning_rate": 9.412812638265279e-06,
+      "loss": 2.1094,
+      "step": 700
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.42578125,
+      "learning_rate": 9.402920915289355e-06,
+      "loss": 2.1226,
+      "step": 705
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.369140625,
+      "learning_rate": 9.392951856467795e-06,
+      "loss": 2.1302,
+      "step": 710
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.447265625,
+      "learning_rate": 9.382905636903809e-06,
+      "loss": 2.0773,
+      "step": 715
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.412109375,
+      "learning_rate": 9.372782433055915e-06,
+      "loss": 2.1493,
+      "step": 720
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.380859375,
+      "learning_rate": 9.362582422734826e-06,
+      "loss": 2.1009,
+      "step": 725
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.4140625,
+      "learning_rate": 9.352305785100348e-06,
+      "loss": 2.1114,
+      "step": 730
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.431640625,
+      "learning_rate": 9.341952700658205e-06,
+      "loss": 2.1021,
+      "step": 735
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.357421875,
+      "learning_rate": 9.331523351256898e-06,
+      "loss": 2.0768,
+      "step": 740
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.373046875,
+      "learning_rate": 9.321017920084485e-06,
+      "loss": 2.0724,
+      "step": 745
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.40234375,
+      "learning_rate": 9.310436591665383e-06,
+      "loss": 2.085,
+      "step": 750
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.396484375,
+      "learning_rate": 9.299779551857118e-06,
+      "loss": 2.0719,
+      "step": 755
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.451171875,
+      "learning_rate": 9.289046987847058e-06,
+      "loss": 2.0916,
+      "step": 760
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.41015625,
+      "learning_rate": 9.278239088149128e-06,
+      "loss": 2.061,
+      "step": 765
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.396484375,
+      "learning_rate": 9.267356042600506e-06,
+      "loss": 2.0851,
+      "step": 770
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.38671875,
+      "learning_rate": 9.256398042358274e-06,
+      "loss": 2.1034,
+      "step": 775
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.423828125,
+      "learning_rate": 9.245365279896077e-06,
+      "loss": 2.0681,
+      "step": 780
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.412109375,
+      "learning_rate": 9.234257949000723e-06,
+      "loss": 2.072,
+      "step": 785
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.330078125,
+      "learning_rate": 9.223076244768797e-06,
+      "loss": 2.064,
+      "step": 790
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.392578125,
+      "learning_rate": 9.211820363603226e-06,
+      "loss": 2.0746,
+      "step": 795
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.4296875,
+      "learning_rate": 9.200490503209831e-06,
+      "loss": 2.1001,
+      "step": 800
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.416015625,
+      "learning_rate": 9.18908686259385e-06,
+      "loss": 2.1237,
+      "step": 805
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.369140625,
+      "learning_rate": 9.17760964205645e-06,
+      "loss": 2.0758,
+      "step": 810
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.40234375,
+      "learning_rate": 9.166059043191204e-06,
+      "loss": 2.0421,
+      "step": 815
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.3828125,
+      "learning_rate": 9.154435268880547e-06,
+      "loss": 2.0514,
+      "step": 820
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.427734375,
+      "learning_rate": 9.142738523292222e-06,
+      "loss": 2.081,
+      "step": 825
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.4140625,
+      "learning_rate": 9.130969011875686e-06,
+      "loss": 2.0762,
+      "step": 830
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.3828125,
+      "learning_rate": 9.119126941358502e-06,
+      "loss": 2.0405,
+      "step": 835
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.423828125,
+      "learning_rate": 9.107212519742714e-06,
+      "loss": 2.0439,
+      "step": 840
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.357421875,
+      "learning_rate": 9.095225956301186e-06,
+      "loss": 2.0508,
+      "step": 845
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.431640625,
+      "learning_rate": 9.083167461573926e-06,
+      "loss": 2.038,
+      "step": 850
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.48046875,
+      "learning_rate": 9.071037247364396e-06,
+      "loss": 2.0457,
+      "step": 855
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.416015625,
+      "learning_rate": 9.058835526735788e-06,
+      "loss": 2.0627,
+      "step": 860
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.40234375,
+      "learning_rate": 9.046562514007275e-06,
+      "loss": 2.0369,
+      "step": 865
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.4296875,
+      "learning_rate": 9.034218424750258e-06,
+      "loss": 2.0828,
+      "step": 870
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.421875,
+      "learning_rate": 9.02180347578457e-06,
+      "loss": 2.0686,
+      "step": 875
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.431640625,
+      "learning_rate": 9.009317885174672e-06,
+      "loss": 2.0532,
+      "step": 880
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.421875,
+      "learning_rate": 8.99676187222582e-06,
+      "loss": 2.0406,
+      "step": 885
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.435546875,
+      "learning_rate": 8.98413565748022e-06,
+      "loss": 2.0875,
+      "step": 890
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.4296875,
+      "learning_rate": 8.971439462713145e-06,
+      "loss": 2.0521,
+      "step": 895
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.498046875,
+      "learning_rate": 8.958673510929046e-06,
+      "loss": 2.0569,
+      "step": 900
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.42578125,
+      "learning_rate": 8.945838026357633e-06,
+      "loss": 2.0385,
+      "step": 905
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.45703125,
+      "learning_rate": 8.932933234449933e-06,
+      "loss": 2.0561,
+      "step": 910
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.3984375,
+      "learning_rate": 8.91995936187434e-06,
+      "loss": 2.0482,
+      "step": 915
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.38671875,
+      "learning_rate": 8.906916636512618e-06,
+      "loss": 2.0214,
+      "step": 920
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.4296875,
+      "learning_rate": 8.89380528745592e-06,
+      "loss": 2.0194,
+      "step": 925
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.349609375,
+      "learning_rate": 8.880625545000733e-06,
+      "loss": 2.0503,
+      "step": 930
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.44921875,
+      "learning_rate": 8.86737764064487e-06,
+      "loss": 2.0215,
+      "step": 935
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.486328125,
+      "learning_rate": 8.854061807083376e-06,
+      "loss": 2.0278,
+      "step": 940
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.474609375,
+      "learning_rate": 8.84067827820445e-06,
+      "loss": 2.0763,
+      "step": 945
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.416015625,
+      "learning_rate": 8.827227289085338e-06,
+      "loss": 2.0128,
+      "step": 950
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.44140625,
+      "learning_rate": 8.813709075988205e-06,
+      "loss": 2.0198,
+      "step": 955
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.400390625,
+      "learning_rate": 8.800123876355976e-06,
+      "loss": 2.0678,
+      "step": 960
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.4140625,
+      "learning_rate": 8.786471928808182e-06,
+      "loss": 1.9936,
+      "step": 965
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.42578125,
+      "learning_rate": 8.772753473136751e-06,
+      "loss": 2.0451,
+      "step": 970
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.40625,
+      "learning_rate": 8.758968750301808e-06,
+      "loss": 1.9825,
+      "step": 975
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.36328125,
+      "learning_rate": 8.74511800242744e-06,
+      "loss": 2.0519,
+      "step": 980
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.4140625,
+      "learning_rate": 8.731201472797436e-06,
+      "loss": 2.0726,
+      "step": 985
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.462890625,
+      "learning_rate": 8.717219405851024e-06,
+      "loss": 2.0264,
+      "step": 990
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.439453125,
+      "learning_rate": 8.703172047178576e-06,
+      "loss": 2.0735,
+      "step": 995
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.4609375,
+      "learning_rate": 8.689059643517285e-06,
+      "loss": 2.0678,
+      "step": 1000
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.470703125,
+      "learning_rate": 8.674882442746844e-06,
+      "loss": 2.0188,
+      "step": 1005
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.4375,
+      "learning_rate": 8.66064069388508e-06,
+      "loss": 2.0203,
+      "step": 1010
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.451171875,
+      "learning_rate": 8.646334647083587e-06,
+      "loss": 2.0728,
+      "step": 1015
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.46484375,
+      "learning_rate": 8.631964553623336e-06,
+      "loss": 2.038,
+      "step": 1020
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.46875,
+      "learning_rate": 8.617530665910253e-06,
+      "loss": 2.0447,
+      "step": 1025
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.435546875,
+      "learning_rate": 8.603033237470783e-06,
+      "loss": 2.046,
+      "step": 1030
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.462890625,
+      "learning_rate": 8.588472522947456e-06,
+      "loss": 2.0006,
+      "step": 1035
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.396484375,
+      "learning_rate": 8.57384877809439e-06,
+      "loss": 2.0695,
+      "step": 1040
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.458984375,
+      "learning_rate": 8.559162259772811e-06,
+      "loss": 2.0498,
+      "step": 1045
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.5,
+      "learning_rate": 8.544413225946547e-06,
+      "loss": 2.0221,
+      "step": 1050
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.419921875,
+      "learning_rate": 8.529601935677481e-06,
+      "loss": 1.9977,
+      "step": 1055
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.388671875,
+      "learning_rate": 8.514728649121017e-06,
+      "loss": 2.0469,
+      "step": 1060
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.38671875,
+      "learning_rate": 8.499793627521501e-06,
+      "loss": 2.0298,
+      "step": 1065
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.41796875,
+      "learning_rate": 8.484797133207634e-06,
+      "loss": 2.0354,
+      "step": 1070
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.48046875,
+      "learning_rate": 8.469739429587865e-06,
+      "loss": 2.0555,
+      "step": 1075
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.43359375,
+      "learning_rate": 8.454620781145761e-06,
+      "loss": 2.0557,
+      "step": 1080
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.380859375,
+      "learning_rate": 8.439441453435371e-06,
+      "loss": 2.0416,
+      "step": 1085
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.486328125,
+      "learning_rate": 8.424201713076553e-06,
+      "loss": 2.0333,
+      "step": 1090
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.4921875,
+      "learning_rate": 8.408901827750289e-06,
+      "loss": 2.0203,
+      "step": 1095
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.45703125,
+      "learning_rate": 8.393542066193994e-06,
+      "loss": 2.0092,
+      "step": 1100
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.4375,
+      "learning_rate": 8.37812269819678e-06,
+      "loss": 2.0201,
+      "step": 1105
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.47265625,
+      "learning_rate": 8.36264399459474e-06,
+      "loss": 1.9957,
+      "step": 1110
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.48046875,
+      "learning_rate": 8.347106227266161e-06,
+      "loss": 2.02,
+      "step": 1115
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.447265625,
+      "learning_rate": 8.331509669126778e-06,
+      "loss": 2.0649,
+      "step": 1120
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.421875,
+      "learning_rate": 8.31585459412496e-06,
+      "loss": 2.0384,
+      "step": 1125
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.4453125,
+      "learning_rate": 8.30014127723691e-06,
+      "loss": 2.0024,
+      "step": 1130
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.53125,
+      "learning_rate": 8.284369994461825e-06,
+      "loss": 2.0536,
+      "step": 1135
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.4140625,
+      "learning_rate": 8.268541022817058e-06,
+      "loss": 2.0044,
+      "step": 1140
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.419921875,
+      "learning_rate": 8.252654640333246e-06,
+      "loss": 2.0584,
+      "step": 1145
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.470703125,
+      "learning_rate": 8.236711126049426e-06,
+      "loss": 2.0532,
+      "step": 1150
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.435546875,
+      "learning_rate": 8.220710760008143e-06,
+      "loss": 1.9918,
+      "step": 1155
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.470703125,
+      "learning_rate": 8.204653823250516e-06,
+      "loss": 2.0174,
+      "step": 1160
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.4921875,
+      "learning_rate": 8.188540597811316e-06,
+      "loss": 1.9979,
+      "step": 1165
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.51171875,
+      "learning_rate": 8.172371366714004e-06,
+      "loss": 2.0073,
+      "step": 1170
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.462890625,
+      "learning_rate": 8.156146413965763e-06,
+      "loss": 2.0156,
+      "step": 1175
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.396484375,
+      "learning_rate": 8.1398660245525e-06,
+      "loss": 1.985,
+      "step": 1180
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.443359375,
+      "learning_rate": 8.123530484433867e-06,
+      "loss": 2.0104,
+      "step": 1185
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.42578125,
+      "learning_rate": 8.107140080538197e-06,
+      "loss": 1.9925,
+      "step": 1190
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.40625,
+      "learning_rate": 8.090695100757505e-06,
+      "loss": 2.0538,
+      "step": 1195
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.447265625,
+      "learning_rate": 8.074195833942405e-06,
+      "loss": 2.0415,
+      "step": 1200
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.40234375,
+      "learning_rate": 8.057642569897049e-06,
+      "loss": 2.0044,
+      "step": 1205
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.44921875,
+      "learning_rate": 8.041035599374026e-06,
+      "loss": 2.0371,
+      "step": 1210
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.47265625,
+      "learning_rate": 8.024375214069269e-06,
+      "loss": 2.017,
+      "step": 1215
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.41796875,
+      "learning_rate": 8.007661706616919e-06,
+      "loss": 2.0202,
+      "step": 1220
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.453125,
+      "learning_rate": 7.99089537058419e-06,
+      "loss": 2.0547,
+      "step": 1225
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.46875,
+      "learning_rate": 7.974076500466215e-06,
+      "loss": 2.005,
+      "step": 1230
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.447265625,
+      "learning_rate": 7.957205391680872e-06,
+      "loss": 2.012,
+      "step": 1235
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.5,
+      "learning_rate": 7.940282340563586e-06,
+      "loss": 2.0097,
+      "step": 1240
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.431640625,
+      "learning_rate": 7.92330764436214e-06,
+      "loss": 2.0385,
+      "step": 1245
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.47265625,
+      "learning_rate": 7.906281601231449e-06,
+      "loss": 2.0071,
+      "step": 1250
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.43359375,
+      "learning_rate": 7.889204510228304e-06,
+      "loss": 2.0146,
+      "step": 1255
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.482421875,
+      "learning_rate": 7.87207667130615e-06,
+      "loss": 1.9922,
+      "step": 1260
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.48046875,
+      "learning_rate": 7.854898385309797e-06,
+      "loss": 1.9942,
+      "step": 1265
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.451171875,
+      "learning_rate": 7.83766995397014e-06,
+      "loss": 2.0493,
+      "step": 1270
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.400390625,
+      "learning_rate": 7.820391679898863e-06,
+      "loss": 2.021,
+      "step": 1275
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.39453125,
+      "learning_rate": 7.803063866583119e-06,
+      "loss": 2.0249,
+      "step": 1280
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.4765625,
+      "learning_rate": 7.7856868183802e-06,
+      "loss": 1.9945,
+      "step": 1285
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 1.9995732307434082,
+      "eval_runtime": 249.4314,
+      "eval_samples_per_second": 9.197,
+      "eval_steps_per_second": 4.598,
+      "step": 1288
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.482421875,
+      "learning_rate": 7.7682608405122e-06,
+      "loss": 2.0239,
+      "step": 1290
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.435546875,
+      "learning_rate": 7.750786239060635e-06,
+      "loss": 2.0123,
+      "step": 1295
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.458984375,
+      "learning_rate": 7.733263320961087e-06,
+      "loss": 1.977,
+      "step": 1300
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.4296875,
+      "learning_rate": 7.715692393997802e-06,
+      "loss": 2.0119,
+      "step": 1305
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.451171875,
+      "learning_rate": 7.698073766798281e-06,
+      "loss": 2.0071,
+      "step": 1310
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.48828125,
+      "learning_rate": 7.680407748827869e-06,
+      "loss": 1.9962,
+      "step": 1315
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.5546875,
+      "learning_rate": 7.662694650384315e-06,
+      "loss": 1.9844,
+      "step": 1320
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.37890625,
+      "learning_rate": 7.644934782592315e-06,
+      "loss": 2.0186,
+      "step": 1325
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.490234375,
+      "learning_rate": 7.627128457398061e-06,
+      "loss": 2.0253,
+      "step": 1330
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.4453125,
+      "learning_rate": 7.609275987563745e-06,
+      "loss": 2.0297,
+      "step": 1335
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.5546875,
+      "learning_rate": 7.591377686662081e-06,
+      "loss": 1.9942,
+      "step": 1340
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.41796875,
+      "learning_rate": 7.573433869070788e-06,
+      "loss": 1.9983,
+      "step": 1345
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.443359375,
+      "learning_rate": 7.555444849967073e-06,
+      "loss": 1.9873,
+      "step": 1350
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.4296875,
+      "learning_rate": 7.537410945322085e-06,
+      "loss": 2.045,
+      "step": 1355
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.453125,
+      "learning_rate": 7.519332471895384e-06,
+      "loss": 2.0481,
+      "step": 1360
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.4921875,
+      "learning_rate": 7.501209747229356e-06,
+      "loss": 1.9862,
+      "step": 1365
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.52734375,
+      "learning_rate": 7.483043089643654e-06,
+      "loss": 2.0116,
+      "step": 1370
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.458984375,
+      "learning_rate": 7.464832818229586e-06,
+      "loss": 2.0211,
+      "step": 1375
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.419921875,
+      "learning_rate": 7.446579252844536e-06,
+      "loss": 1.9864,
+      "step": 1380
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.42578125,
+      "learning_rate": 7.428282714106324e-06,
+      "loss": 2.0303,
+      "step": 1385
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.44140625,
+      "learning_rate": 7.409943523387586e-06,
+      "loss": 1.9974,
+      "step": 1390
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.46484375,
+      "learning_rate": 7.391562002810123e-06,
+      "loss": 1.9797,
+      "step": 1395
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.474609375,
+      "learning_rate": 7.37313847523925e-06,
+      "loss": 1.9493,
+      "step": 1400
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.546875,
+      "learning_rate": 7.354673264278115e-06,
+      "loss": 2.0172,
+      "step": 1405
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.498046875,
+      "learning_rate": 7.336166694262028e-06,
+      "loss": 2.0123,
+      "step": 1410
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.3828125,
+      "learning_rate": 7.3176190902527526e-06,
+      "loss": 1.9659,
+      "step": 1415
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.400390625,
+      "learning_rate": 7.299030778032799e-06,
+      "loss": 1.9849,
+      "step": 1420
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.51953125,
+      "learning_rate": 7.280402084099711e-06,
+      "loss": 1.9803,
+      "step": 1425
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.462890625,
+      "learning_rate": 7.261733335660317e-06,
+      "loss": 2.024,
+      "step": 1430
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.53515625,
+      "learning_rate": 7.243024860624991e-06,
+      "loss": 1.9692,
+      "step": 1435
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.48046875,
+      "learning_rate": 7.224276987601895e-06,
+      "loss": 2.0185,
+      "step": 1440
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.4140625,
+      "learning_rate": 7.205490045891198e-06,
+      "loss": 2.0265,
+      "step": 1445
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.470703125,
+      "learning_rate": 7.186664365479301e-06,
+      "loss": 2.0226,
+      "step": 1450
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.43359375,
+      "learning_rate": 7.167800277033039e-06,
+      "loss": 2.0106,
+      "step": 1455
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.43359375,
+      "learning_rate": 7.148898111893867e-06,
+      "loss": 2.0205,
+      "step": 1460
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.4921875,
+      "learning_rate": 7.129958202072047e-06,
+      "loss": 1.9522,
+      "step": 1465
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.4921875,
+      "learning_rate": 7.110980880240814e-06,
+      "loss": 1.9759,
+      "step": 1470
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.443359375,
+      "learning_rate": 7.091966479730532e-06,
+      "loss": 2.02,
+      "step": 1475
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.45703125,
+      "learning_rate": 7.07291533452284e-06,
+      "loss": 2.0119,
+      "step": 1480
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.5,
+      "learning_rate": 7.053827779244784e-06,
+      "loss": 2.0199,
+      "step": 1485
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.453125,
+      "learning_rate": 7.034704149162944e-06,
+      "loss": 2.0191,
+      "step": 1490
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.5,
+      "learning_rate": 7.015544780177539e-06,
+      "loss": 1.9632,
+      "step": 1495
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.44921875,
+      "learning_rate": 6.996350008816532e-06,
+      "loss": 2.0271,
+      "step": 1500
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.4375,
+      "learning_rate": 6.977120172229713e-06,
+      "loss": 1.9492,
+      "step": 1505
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.44140625,
+      "learning_rate": 6.957855608182788e-06,
+      "loss": 1.9785,
+      "step": 1510
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.443359375,
+      "learning_rate": 6.938556655051432e-06,
+      "loss": 1.991,
+      "step": 1515
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.484375,
+      "learning_rate": 6.9192236518153566e-06,
+      "loss": 1.9875,
+      "step": 1520
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.416015625,
+      "learning_rate": 6.8998569380523535e-06,
+      "loss": 1.9848,
+      "step": 1525
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.458984375,
+      "learning_rate": 6.880456853932327e-06,
+      "loss": 2.0381,
+      "step": 1530
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.44921875,
+      "learning_rate": 6.861023740211318e-06,
+      "loss": 1.9968,
+      "step": 1535
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.458984375,
+      "learning_rate": 6.841557938225527e-06,
+      "loss": 1.9984,
+      "step": 1540
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.4765625,
+      "learning_rate": 6.822059789885308e-06,
+      "loss": 1.9702,
+      "step": 1545
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.44140625,
+      "learning_rate": 6.802529637669171e-06,
+      "loss": 1.9838,
+      "step": 1550
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.48828125,
+      "learning_rate": 6.782967824617762e-06,
+      "loss": 1.9962,
+      "step": 1555
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.48828125,
+      "learning_rate": 6.76337469432784e-06,
+      "loss": 2.0012,
+      "step": 1560
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.51953125,
+      "learning_rate": 6.743750590946239e-06,
+      "loss": 1.9384,
+      "step": 1565
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.47265625,
+      "learning_rate": 6.72409585916383e-06,
+      "loss": 1.9947,
+      "step": 1570
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.42578125,
+      "learning_rate": 6.704410844209453e-06,
+      "loss": 2.0145,
+      "step": 1575
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.451171875,
+      "learning_rate": 6.684695891843871e-06,
+      "loss": 2.0012,
+      "step": 1580
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.43359375,
+      "learning_rate": 6.664951348353682e-06,
+      "loss": 1.9855,
+      "step": 1585
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.498046875,
+      "learning_rate": 6.645177560545245e-06,
+      "loss": 2.0137,
+      "step": 1590
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.515625,
+      "learning_rate": 6.625374875738585e-06,
+      "loss": 2.0052,
+      "step": 1595
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.5390625,
+      "learning_rate": 6.605543641761293e-06,
+      "loss": 1.9763,
+      "step": 1600
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.46875,
+      "learning_rate": 6.585684206942416e-06,
+      "loss": 2.0089,
+      "step": 1605
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.490234375,
+      "learning_rate": 6.56579692010634e-06,
+      "loss": 1.9824,
+      "step": 1610
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.56640625,
+      "learning_rate": 6.545882130566663e-06,
+      "loss": 1.9818,
+      "step": 1615
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.451171875,
+      "learning_rate": 6.525940188120059e-06,
+      "loss": 2.0417,
+      "step": 1620
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.40625,
+      "learning_rate": 6.505971443040132e-06,
+      "loss": 1.9811,
+      "step": 1625
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.46875,
+      "learning_rate": 6.485976246071268e-06,
+      "loss": 2.0328,
+      "step": 1630
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.5625,
+      "learning_rate": 6.4659549484224704e-06,
+      "loss": 1.9734,
+      "step": 1635
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.47265625,
+      "learning_rate": 6.445907901761189e-06,
+      "loss": 2.0472,
+      "step": 1640
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.50390625,
+      "learning_rate": 6.425835458207155e-06,
+      "loss": 2.0085,
+      "step": 1645
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.482421875,
+      "learning_rate": 6.405737970326179e-06,
+      "loss": 2.022,
+      "step": 1650
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.51953125,
+      "learning_rate": 6.3856157911239714e-06,
+      "loss": 2.0213,
+      "step": 1655
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.48046875,
+      "learning_rate": 6.365469274039936e-06,
+      "loss": 1.9837,
+      "step": 1660
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.462890625,
+      "learning_rate": 6.345298772940962e-06,
+      "loss": 1.999,
+      "step": 1665
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.412109375,
+      "learning_rate": 6.325104642115214e-06,
+      "loss": 1.9988,
+      "step": 1670
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.486328125,
+      "learning_rate": 6.304887236265902e-06,
+      "loss": 1.9888,
+      "step": 1675
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.52734375,
+      "learning_rate": 6.2846469105050545e-06,
+      "loss": 1.9944,
+      "step": 1680
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.50390625,
+      "learning_rate": 6.264384020347281e-06,
+      "loss": 1.975,
+      "step": 1685
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.421875,
+      "learning_rate": 6.244098921703524e-06,
+      "loss": 2.0254,
+      "step": 1690
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.474609375,
+      "learning_rate": 6.223791970874817e-06,
+      "loss": 1.9662,
+      "step": 1695
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.52734375,
+      "learning_rate": 6.203463524546017e-06,
+      "loss": 1.9826,
+      "step": 1700
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.474609375,
+      "learning_rate": 6.183113939779539e-06,
+      "loss": 2.0226,
+      "step": 1705
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.52734375,
+      "learning_rate": 6.162743574009094e-06,
+      "loss": 1.9639,
+      "step": 1710
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.51171875,
+      "learning_rate": 6.142352785033395e-06,
+      "loss": 2.0038,
+      "step": 1715
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.50390625,
+      "learning_rate": 6.121941931009894e-06,
+      "loss": 1.9956,
+      "step": 1720
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.50390625,
+      "learning_rate": 6.101511370448468e-06,
+      "loss": 2.025,
+      "step": 1725
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.53515625,
+      "learning_rate": 6.08106146220514e-06,
+      "loss": 1.9768,
+      "step": 1730
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.5,
+      "learning_rate": 6.060592565475765e-06,
+      "loss": 2.0067,
+      "step": 1735
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.5234375,
+      "learning_rate": 6.040105039789726e-06,
+      "loss": 2.0106,
+      "step": 1740
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.48046875,
+      "learning_rate": 6.019599245003616e-06,
+      "loss": 1.9991,
+      "step": 1745
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.482421875,
+      "learning_rate": 5.999075541294921e-06,
+      "loss": 2.0292,
+      "step": 1750
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.47265625,
+      "learning_rate": 5.978534289155689e-06,
+      "loss": 2.0486,
+      "step": 1755
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.5078125,
+      "learning_rate": 5.957975849386202e-06,
+      "loss": 2.0203,
+      "step": 1760
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.53125,
+      "learning_rate": 5.937400583088636e-06,
+      "loss": 2.0101,
+      "step": 1765
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.5078125,
+      "learning_rate": 5.916808851660718e-06,
+      "loss": 2.0079,
+      "step": 1770
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.494140625,
+      "learning_rate": 5.896201016789385e-06,
+      "loss": 1.9835,
+      "step": 1775
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.5234375,
+      "learning_rate": 5.8755774404444175e-06,
+      "loss": 1.9988,
+      "step": 1780
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.53515625,
+      "learning_rate": 5.8549384848720965e-06,
+      "loss": 2.0291,
+      "step": 1785
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.46484375,
+      "learning_rate": 5.834284512588832e-06,
+      "loss": 1.9633,
+      "step": 1790
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.478515625,
+      "learning_rate": 5.813615886374795e-06,
+      "loss": 1.9922,
+      "step": 1795
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.47265625,
+      "learning_rate": 5.792932969267553e-06,
+      "loss": 1.9726,
+      "step": 1800
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.48828125,
+      "learning_rate": 5.772236124555684e-06,
+      "loss": 2.0007,
+      "step": 1805
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.4609375,
+      "learning_rate": 5.751525715772401e-06,
+      "loss": 1.978,
+      "step": 1810
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.458984375,
+      "learning_rate": 5.730802106689168e-06,
+      "loss": 2.0138,
+      "step": 1815
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.58984375,
+      "learning_rate": 5.7100656613093005e-06,
+      "loss": 1.9792,
+      "step": 1820
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.498046875,
+      "learning_rate": 5.689316743861589e-06,
+      "loss": 2.0005,
+      "step": 1825
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.515625,
+      "learning_rate": 5.668555718793884e-06,
+      "loss": 2.0156,
+      "step": 1830
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.44921875,
+      "learning_rate": 5.647782950766709e-06,
+      "loss": 1.9823,
+      "step": 1835
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.455078125,
+      "learning_rate": 5.626998804646842e-06,
+      "loss": 1.9878,
+      "step": 1840
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.47265625,
+      "learning_rate": 5.606203645500916e-06,
+      "loss": 2.0163,
+      "step": 1845
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.435546875,
+      "learning_rate": 5.5853978385890054e-06,
+      "loss": 2.0128,
+      "step": 1850
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.5234375,
+      "learning_rate": 5.564581749358209e-06,
+      "loss": 2.0038,
+      "step": 1855
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.427734375,
+      "learning_rate": 5.543755743436231e-06,
+      "loss": 2.0005,
+      "step": 1860
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.43359375,
+      "learning_rate": 5.522920186624958e-06,
+      "loss": 1.9923,
+      "step": 1865
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.5234375,
+      "learning_rate": 5.502075444894035e-06,
+      "loss": 2.0108,
+      "step": 1870
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.392578125,
+      "learning_rate": 5.481221884374435e-06,
+      "loss": 1.9527,
+      "step": 1875
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.439453125,
+      "learning_rate": 5.4603598713520354e-06,
+      "loss": 1.9828,
+      "step": 1880
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.484375,
+      "learning_rate": 5.439489772261176e-06,
+      "loss": 1.9947,
+      "step": 1885
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.455078125,
+      "learning_rate": 5.418611953678224e-06,
+      "loss": 2.0235,
+      "step": 1890
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.490234375,
+      "learning_rate": 5.39772678231514e-06,
+      "loss": 1.9974,
+      "step": 1895
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.451171875,
+      "learning_rate": 5.376834625013031e-06,
+      "loss": 2.012,
+      "step": 1900
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.45703125,
+      "learning_rate": 5.355935848735712e-06,
+      "loss": 1.9656,
+      "step": 1905
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.486328125,
+      "learning_rate": 5.335030820563258e-06,
+      "loss": 1.9831,
+      "step": 1910
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.5,
+      "learning_rate": 5.3141199076855546e-06,
+      "loss": 2.0297,
+      "step": 1915
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.478515625,
+      "learning_rate": 5.293203477395851e-06,
+      "loss": 1.9985,
+      "step": 1920
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.51171875,
+      "learning_rate": 5.272281897084307e-06,
+      "loss": 2.0078,
+      "step": 1925
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.51953125,
+      "learning_rate": 5.251355534231546e-06,
+      "loss": 1.9816,
+      "step": 1930
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.54296875,
+      "learning_rate": 5.2304247564021886e-06,
+      "loss": 2.0208,
+      "step": 1935
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.46875,
+      "learning_rate": 5.209489931238405e-06,
+      "loss": 2.0368,
+      "step": 1940
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.5078125,
+      "learning_rate": 5.188551426453458e-06,
+      "loss": 1.9596,
+      "step": 1945
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.52734375,
+      "learning_rate": 5.167609609825238e-06,
+      "loss": 2.038,
+      "step": 1950
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.494140625,
+      "learning_rate": 5.14666484918981e-06,
+      "loss": 1.9991,
+      "step": 1955
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.51953125,
+      "learning_rate": 5.125717512434947e-06,
+      "loss": 2.0086,
+      "step": 1960
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.4921875,
+      "learning_rate": 5.104767967493675e-06,
+      "loss": 1.9864,
+      "step": 1965
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.435546875,
+      "learning_rate": 5.083816582337799e-06,
+      "loss": 1.9859,
+      "step": 1970
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.44921875,
+      "learning_rate": 5.062863724971453e-06,
+      "loss": 1.9661,
+      "step": 1975
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.5546875,
+      "learning_rate": 5.041909763424625e-06,
+      "loss": 2.0011,
+      "step": 1980
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.447265625,
+      "learning_rate": 5.020955065746702e-06,
+      "loss": 2.0104,
+      "step": 1985
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.494140625,
+      "learning_rate": 5e-06,
+      "loss": 1.9524,
+      "step": 1990
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.49609375,
+      "learning_rate": 4.979044934253299e-06,
+      "loss": 1.9903,
+      "step": 1995
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.451171875,
+      "learning_rate": 4.9580902365753765e-06,
+      "loss": 1.9945,
+      "step": 2000
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.482421875,
+      "learning_rate": 4.937136275028549e-06,
+      "loss": 1.9889,
+      "step": 2005
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.46875,
+      "learning_rate": 4.916183417662203e-06,
+      "loss": 1.9972,
+      "step": 2010
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.5234375,
+      "learning_rate": 4.895232032506326e-06,
+      "loss": 2.0227,
+      "step": 2015
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.466796875,
+      "learning_rate": 4.874282487565053e-06,
+      "loss": 2.0109,
+      "step": 2020
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.4453125,
+      "learning_rate": 4.853335150810191e-06,
+      "loss": 1.9937,
+      "step": 2025
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.41015625,
+      "learning_rate": 4.832390390174763e-06,
+      "loss": 2.0196,
+      "step": 2030
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.4921875,
+      "learning_rate": 4.811448573546543e-06,
+      "loss": 2.0225,
+      "step": 2035
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.46875,
+      "learning_rate": 4.790510068761596e-06,
+      "loss": 2.0174,
+      "step": 2040
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.498046875,
+      "learning_rate": 4.769575243597815e-06,
+      "loss": 2.0172,
+      "step": 2045
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.45703125,
+      "learning_rate": 4.7486444657684575e-06,
+      "loss": 2.0014,
+      "step": 2050
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.48046875,
+      "learning_rate": 4.727718102915694e-06,
+      "loss": 1.9513,
+      "step": 2055
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.478515625,
+      "learning_rate": 4.706796522604152e-06,
+      "loss": 1.987,
+      "step": 2060
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.5,
+      "learning_rate": 4.685880092314448e-06,
+      "loss": 1.9774,
+      "step": 2065
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.482421875,
+      "learning_rate": 4.664969179436744e-06,
+      "loss": 1.9865,
+      "step": 2070
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.49609375,
+      "learning_rate": 4.644064151264289e-06,
+      "loss": 1.9413,
+      "step": 2075
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.4453125,
+      "learning_rate": 4.623165374986971e-06,
+      "loss": 2.0014,
+      "step": 2080
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.4765625,
+      "learning_rate": 4.602273217684861e-06,
+      "loss": 2.0078,
+      "step": 2085
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.4609375,
+      "learning_rate": 4.581388046321777e-06,
+      "loss": 1.9833,
+      "step": 2090
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.5390625,
+      "learning_rate": 4.560510227738825e-06,
+      "loss": 1.9535,
+      "step": 2095
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.494140625,
+      "learning_rate": 4.539640128647965e-06,
+      "loss": 2.0031,
+      "step": 2100
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.427734375,
+      "learning_rate": 4.518778115625566e-06,
+      "loss": 2.023,
+      "step": 2105
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.51171875,
+      "learning_rate": 4.497924555105966e-06,
+      "loss": 1.9782,
+      "step": 2110
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.4921875,
+      "learning_rate": 4.477079813375043e-06,
+      "loss": 1.9828,
+      "step": 2115
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.46875,
+      "learning_rate": 4.456244256563769e-06,
+      "loss": 1.9677,
+      "step": 2120
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.4921875,
+      "learning_rate": 4.435418250641791e-06,
+      "loss": 1.9976,
+      "step": 2125
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.53125,
+      "learning_rate": 4.414602161410995e-06,
+      "loss": 2.0058,
+      "step": 2130
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.51171875,
+      "learning_rate": 4.393796354499086e-06,
+      "loss": 1.9918,
+      "step": 2135
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.48828125,
+      "learning_rate": 4.373001195353159e-06,
+      "loss": 1.9882,
+      "step": 2140
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.462890625,
+      "learning_rate": 4.352217049233294e-06,
+      "loss": 1.9824,
+      "step": 2145
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.46484375,
+      "learning_rate": 4.331444281206118e-06,
+      "loss": 2.0098,
+      "step": 2150
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.51171875,
+      "learning_rate": 4.310683256138414e-06,
+      "loss": 2.0122,
+      "step": 2155
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.45703125,
+      "learning_rate": 4.2899343386907e-06,
+      "loss": 2.0138,
+      "step": 2160
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.458984375,
+      "learning_rate": 4.269197893310834e-06,
+      "loss": 1.9788,
+      "step": 2165
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.48046875,
+      "learning_rate": 4.2484742842276e-06,
+      "loss": 2.0198,
+      "step": 2170
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.49609375,
+      "learning_rate": 4.227763875444317e-06,
+      "loss": 1.9988,
+      "step": 2175
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.5625,
+      "learning_rate": 4.207067030732449e-06,
+      "loss": 1.983,
+      "step": 2180
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.5390625,
+      "learning_rate": 4.186384113625207e-06,
+      "loss": 1.9983,
+      "step": 2185
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.53125,
+      "learning_rate": 4.165715487411169e-06,
+      "loss": 1.9702,
+      "step": 2190
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.5703125,
+      "learning_rate": 4.145061515127905e-06,
+      "loss": 1.9954,
+      "step": 2195
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.484375,
+      "learning_rate": 4.124422559555584e-06,
+      "loss": 2.0175,
+      "step": 2200
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.5,
+      "learning_rate": 4.103798983210617e-06,
+      "loss": 1.9919,
+      "step": 2205
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.50390625,
+      "learning_rate": 4.083191148339283e-06,
+      "loss": 1.9898,
+      "step": 2210
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.455078125,
+      "learning_rate": 4.0625994169113645e-06,
+      "loss": 2.017,
+      "step": 2215
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.53515625,
+      "learning_rate": 4.042024150613798e-06,
+      "loss": 1.9761,
+      "step": 2220
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.451171875,
+      "learning_rate": 4.021465710844311e-06,
+      "loss": 1.972,
+      "step": 2225
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.48828125,
+      "learning_rate": 4.000924458705079e-06,
+      "loss": 1.9848,
+      "step": 2230
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.546875,
+      "learning_rate": 3.980400754996384e-06,
+      "loss": 2.0117,
+      "step": 2235
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.478515625,
+      "learning_rate": 3.959894960210275e-06,
+      "loss": 2.0132,
+      "step": 2240
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.423828125,
+      "learning_rate": 3.939407434524238e-06,
+      "loss": 2.0503,
+      "step": 2245
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.515625,
+      "learning_rate": 3.918938537794863e-06,
+      "loss": 2.0558,
+      "step": 2250
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.4609375,
+      "learning_rate": 3.898488629551535e-06,
+      "loss": 1.9943,
+      "step": 2255
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.43359375,
+      "learning_rate": 3.87805806899011e-06,
+      "loss": 1.9821,
+      "step": 2260
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.52734375,
+      "learning_rate": 3.857647214966606e-06,
+      "loss": 1.9503,
+      "step": 2265
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.4375,
+      "learning_rate": 3.837256425990909e-06,
+      "loss": 1.9906,
+      "step": 2270
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.52734375,
+      "learning_rate": 3.816886060220462e-06,
+      "loss": 1.9814,
+      "step": 2275
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.51171875,
+      "learning_rate": 3.7965364754539845e-06,
+      "loss": 2.0166,
+      "step": 2280
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.47265625,
+      "learning_rate": 3.7762080291251836e-06,
+      "loss": 1.9964,
+      "step": 2285
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.48046875,
+      "learning_rate": 3.755901078296478e-06,
+      "loss": 2.0175,
+      "step": 2290
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.439453125,
+      "learning_rate": 3.7356159796527214e-06,
+      "loss": 2.0271,
+      "step": 2295
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.50390625,
+      "learning_rate": 3.7153530894949476e-06,
+      "loss": 2.0338,
+      "step": 2300
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.49609375,
+      "learning_rate": 3.6951127637340995e-06,
+      "loss": 1.9582,
+      "step": 2305
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.58203125,
+      "learning_rate": 3.6748953578847875e-06,
+      "loss": 1.9524,
+      "step": 2310
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.498046875,
+      "learning_rate": 3.6547012270590397e-06,
+      "loss": 1.9826,
+      "step": 2315
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.50390625,
+      "learning_rate": 3.6345307259600657e-06,
+      "loss": 1.9812,
+      "step": 2320
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.5078125,
+      "learning_rate": 3.6143842088760294e-06,
+      "loss": 1.9708,
+      "step": 2325
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.44140625,
+      "learning_rate": 3.5942620296738218e-06,
+      "loss": 2.01,
+      "step": 2330
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.416015625,
+      "learning_rate": 3.5741645417928457e-06,
+      "loss": 1.9714,
+      "step": 2335
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.5546875,
+      "learning_rate": 3.554092098238811e-06,
+      "loss": 1.9949,
+      "step": 2340
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.42578125,
+      "learning_rate": 3.5340450515775316e-06,
+      "loss": 1.9974,
+      "step": 2345
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.478515625,
+      "learning_rate": 3.514023753928734e-06,
+      "loss": 2.0229,
+      "step": 2350
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.474609375,
+      "learning_rate": 3.49402855695987e-06,
+      "loss": 1.9732,
+      "step": 2355
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.3984375,
+      "learning_rate": 3.474059811879944e-06,
+      "loss": 2.0332,
+      "step": 2360
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.494140625,
+      "learning_rate": 3.45411786943334e-06,
+      "loss": 1.9856,
+      "step": 2365
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.48828125,
+      "learning_rate": 3.434203079893662e-06,
+      "loss": 1.9775,
+      "step": 2370
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.494140625,
+      "learning_rate": 3.414315793057586e-06,
+      "loss": 1.9561,
+      "step": 2375
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.5703125,
+      "learning_rate": 3.3944563582387084e-06,
+      "loss": 1.9733,
+      "step": 2380
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.55078125,
+      "learning_rate": 3.374625124261416e-06,
+      "loss": 2.0123,
+      "step": 2385
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.490234375,
+      "learning_rate": 3.3548224394547557e-06,
+      "loss": 2.0205,
+      "step": 2390
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.3671875,
+      "learning_rate": 3.3350486516463194e-06,
+      "loss": 1.9915,
+      "step": 2395
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.478515625,
+      "learning_rate": 3.31530410815613e-06,
+      "loss": 2.0599,
+      "step": 2400
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.50390625,
+      "learning_rate": 3.2955891557905477e-06,
+      "loss": 1.9845,
+      "step": 2405
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.4765625,
+      "learning_rate": 3.275904140836172e-06,
+      "loss": 1.9529,
+      "step": 2410
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.46875,
+      "learning_rate": 3.256249409053761e-06,
+      "loss": 1.9569,
+      "step": 2415
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.4921875,
+      "learning_rate": 3.2366253056721607e-06,
+      "loss": 2.0147,
+      "step": 2420
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.5,
+      "learning_rate": 3.2170321753822374e-06,
+      "loss": 2.0061,
+      "step": 2425
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.50390625,
+      "learning_rate": 3.1974703623308288e-06,
+      "loss": 2.0166,
+      "step": 2430
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.455078125,
+      "learning_rate": 3.1779402101146917e-06,
+      "loss": 2.0089,
+      "step": 2435
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.515625,
+      "learning_rate": 3.1584420617744737e-06,
+      "loss": 2.0264,
+      "step": 2440
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.58984375,
+      "learning_rate": 3.138976259788682e-06,
+      "loss": 1.9822,
+      "step": 2445
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.51953125,
+      "learning_rate": 3.119543146067675e-06,
+      "loss": 1.9841,
+      "step": 2450
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.443359375,
+      "learning_rate": 3.100143061947648e-06,
+      "loss": 2.0002,
+      "step": 2455
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.421875,
+      "learning_rate": 3.0807763481846455e-06,
+      "loss": 1.9902,
+      "step": 2460
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.53125,
+      "learning_rate": 3.0614433449485716e-06,
+      "loss": 1.967,
+      "step": 2465
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.55078125,
+      "learning_rate": 3.0421443918172155e-06,
+      "loss": 1.8929,
+      "step": 2470
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.466796875,
+      "learning_rate": 3.022879827770288e-06,
+      "loss": 2.0185,
+      "step": 2475
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.470703125,
+      "learning_rate": 3.00364999118347e-06,
+      "loss": 2.0332,
+      "step": 2480
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.48828125,
+      "learning_rate": 2.984455219822462e-06,
+      "loss": 1.9836,
+      "step": 2485
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.53125,
+      "learning_rate": 2.965295850837057e-06,
+      "loss": 1.9414,
+      "step": 2490
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.50390625,
+      "learning_rate": 2.946172220755217e-06,
+      "loss": 1.9639,
+      "step": 2495
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.56640625,
+      "learning_rate": 2.927084665477162e-06,
+      "loss": 1.9881,
+      "step": 2500
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.4921875,
+      "learning_rate": 2.90803352026947e-06,
+      "loss": 2.0282,
+      "step": 2505
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.55078125,
+      "learning_rate": 2.889019119759187e-06,
+      "loss": 1.9554,
+      "step": 2510
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.458984375,
+      "learning_rate": 2.8700417979279555e-06,
+      "loss": 2.0012,
+      "step": 2515
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.50390625,
+      "learning_rate": 2.8511018881061347e-06,
+      "loss": 1.9745,
+      "step": 2520
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.546875,
+      "learning_rate": 2.8321997229669616e-06,
+      "loss": 2.0432,
+      "step": 2525
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.5,
+      "learning_rate": 2.8133356345206996e-06,
+      "loss": 1.9733,
+      "step": 2530
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.5390625,
+      "learning_rate": 2.7945099541088026e-06,
+      "loss": 1.9746,
+      "step": 2535
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.46484375,
+      "learning_rate": 2.775723012398107e-06,
+      "loss": 2.0081,
+      "step": 2540
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.49609375,
+      "learning_rate": 2.7569751393750088e-06,
+      "loss": 2.0195,
+      "step": 2545
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.515625,
+      "learning_rate": 2.7382666643396825e-06,
+      "loss": 1.9804,
+      "step": 2550
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.498046875,
+      "learning_rate": 2.7195979159002918e-06,
+      "loss": 1.971,
+      "step": 2555
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.498046875,
+      "learning_rate": 2.7009692219672025e-06,
+      "loss": 2.0013,
+      "step": 2560
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.494140625,
+      "learning_rate": 2.6823809097472495e-06,
+      "loss": 1.9895,
+      "step": 2565
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.4609375,
+      "learning_rate": 2.6638333057379743e-06,
+      "loss": 2.0207,
+      "step": 2570
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.54296875,
+      "learning_rate": 2.645326735721886e-06,
+      "loss": 2.0215,
+      "step": 2575
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 1.9850621223449707,
+      "eval_runtime": 247.3914,
+      "eval_samples_per_second": 9.273,
+      "eval_steps_per_second": 4.636,
+      "step": 2577
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.443359375,
+      "learning_rate": 2.6268615247607533e-06,
+      "loss": 1.9937,
+      "step": 2580
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.52734375,
+      "learning_rate": 2.608437997189878e-06,
+      "loss": 2.0004,
+      "step": 2585
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.4921875,
+      "learning_rate": 2.5900564766124163e-06,
+      "loss": 1.9831,
+      "step": 2590
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.46875,
+      "learning_rate": 2.5717172858936767e-06,
+      "loss": 1.9916,
+      "step": 2595
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.50390625,
+      "learning_rate": 2.5534207471554644e-06,
+      "loss": 1.9886,
+      "step": 2600
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.4921875,
+      "learning_rate": 2.5351671817704156e-06,
+      "loss": 2.0097,
+      "step": 2605
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.54296875,
+      "learning_rate": 2.5169569103563484e-06,
+      "loss": 2.0168,
+      "step": 2610
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.546875,
+      "learning_rate": 2.498790252770645e-06,
+      "loss": 1.9968,
+      "step": 2615
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.5,
+      "learning_rate": 2.480667528104617e-06,
+      "loss": 1.9686,
+      "step": 2620
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.443359375,
+      "learning_rate": 2.462589054677914e-06,
+      "loss": 2.0257,
+      "step": 2625
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.45703125,
+      "learning_rate": 2.444555150032929e-06,
+      "loss": 1.9721,
+      "step": 2630
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.5390625,
+      "learning_rate": 2.4265661309292116e-06,
+      "loss": 1.9975,
+      "step": 2635
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.478515625,
+      "learning_rate": 2.40862231333792e-06,
+      "loss": 2.0092,
+      "step": 2640
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.5625,
+      "learning_rate": 2.390724012436255e-06,
+      "loss": 2.0166,
+      "step": 2645
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.4296875,
+      "learning_rate": 2.372871542601939e-06,
+      "loss": 2.0051,
+      "step": 2650
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.462890625,
+      "learning_rate": 2.3550652174076867e-06,
+      "loss": 1.9946,
+      "step": 2655
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.52734375,
+      "learning_rate": 2.3373053496156865e-06,
+      "loss": 1.9957,
+      "step": 2660
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.51171875,
+      "learning_rate": 2.319592251172133e-06,
+      "loss": 1.9535,
+      "step": 2665
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.458984375,
+      "learning_rate": 2.301926233201721e-06,
+      "loss": 1.9934,
+      "step": 2670
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.412109375,
+      "learning_rate": 2.2843076060022003e-06,
+      "loss": 2.0027,
+      "step": 2675
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.52734375,
+      "learning_rate": 2.2667366790389152e-06,
+      "loss": 1.9822,
+      "step": 2680
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.52734375,
+      "learning_rate": 2.2492137609393663e-06,
+      "loss": 1.9879,
+      "step": 2685
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.43359375,
+      "learning_rate": 2.231739159487803e-06,
+      "loss": 1.9868,
+      "step": 2690
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.46875,
+      "learning_rate": 2.2143131816198007e-06,
+      "loss": 2.0242,
+      "step": 2695
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.46875,
+      "learning_rate": 2.196936133416882e-06,
+      "loss": 1.9868,
+      "step": 2700
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.48046875,
+      "learning_rate": 2.1796083201011387e-06,
+      "loss": 1.9965,
+      "step": 2705
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.3984375,
+      "learning_rate": 2.16233004602986e-06,
+      "loss": 1.9783,
+      "step": 2710
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.5625,
+      "learning_rate": 2.145101614690205e-06,
+      "loss": 1.9721,
+      "step": 2715
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.4765625,
+      "learning_rate": 2.1279233286938503e-06,
+      "loss": 2.0304,
+      "step": 2720
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.5078125,
+      "learning_rate": 2.1107954897716976e-06,
+      "loss": 1.964,
+      "step": 2725
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.462890625,
+      "learning_rate": 2.093718398768553e-06,
+      "loss": 1.9907,
+      "step": 2730
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.5078125,
+      "learning_rate": 2.0766923556378576e-06,
+      "loss": 2.0063,
+      "step": 2735
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.50390625,
+      "learning_rate": 2.059717659436415e-06,
+      "loss": 1.9807,
+      "step": 2740
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.5546875,
+      "learning_rate": 2.0427946083191295e-06,
+      "loss": 1.9811,
+      "step": 2745
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.494140625,
+      "learning_rate": 2.0259234995337864e-06,
+      "loss": 1.9759,
+      "step": 2750
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.50390625,
+      "learning_rate": 2.009104629415813e-06,
+      "loss": 2.0222,
+      "step": 2755
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.51171875,
+      "learning_rate": 1.9923382933830836e-06,
+      "loss": 2.0119,
+      "step": 2760
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.4375,
+      "learning_rate": 1.9756247859307336e-06,
+      "loss": 1.9132,
+      "step": 2765
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.9589644006259754e-06,
+      "loss": 1.9956,
+      "step": 2770
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.478515625,
+      "learning_rate": 1.9423574301029524e-06,
+      "loss": 1.9716,
+      "step": 2775
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.5,
+      "learning_rate": 1.925804166057596e-06,
+      "loss": 2.0107,
+      "step": 2780
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.51171875,
+      "learning_rate": 1.9093048992424964e-06,
+      "loss": 1.9849,
+      "step": 2785
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.55078125,
+      "learning_rate": 1.8928599194618052e-06,
+      "loss": 2.0232,
+      "step": 2790
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.876469515566136e-06,
+      "loss": 1.976,
+      "step": 2795
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.8601339754475007e-06,
+      "loss": 2.0444,
+      "step": 2800
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.8438535860342398e-06,
+      "loss": 1.9898,
+      "step": 2805
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.482421875,
+      "learning_rate": 1.827628633285996e-06,
+      "loss": 1.9559,
+      "step": 2810
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.4375,
+      "learning_rate": 1.8114594021886845e-06,
+      "loss": 1.9971,
+      "step": 2815
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.53515625,
+      "learning_rate": 1.795346176749484e-06,
+      "loss": 1.9682,
+      "step": 2820
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.466796875,
+      "learning_rate": 1.7792892399918587e-06,
+      "loss": 2.0249,
+      "step": 2825
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.53125,
+      "learning_rate": 1.7632888739505744e-06,
+      "loss": 1.9901,
+      "step": 2830
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.46875,
+      "learning_rate": 1.7473453596667544e-06,
+      "loss": 1.9835,
+      "step": 2835
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.5078125,
+      "learning_rate": 1.7314589771829426e-06,
+      "loss": 1.9695,
+      "step": 2840
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.470703125,
+      "learning_rate": 1.715630005538174e-06,
+      "loss": 1.9969,
+      "step": 2845
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.6998587227630903e-06,
+      "loss": 2.0026,
+      "step": 2850
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.47265625,
+      "learning_rate": 1.684145405875039e-06,
+      "loss": 2.0031,
+      "step": 2855
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.48828125,
+      "learning_rate": 1.668490330873223e-06,
+      "loss": 2.0011,
+      "step": 2860
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.6528937727338411e-06,
+      "loss": 1.9748,
+      "step": 2865
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.470703125,
+      "learning_rate": 1.6373560054052629e-06,
+      "loss": 1.9881,
+      "step": 2870
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.484375,
+      "learning_rate": 1.6218773018032212e-06,
+      "loss": 1.9932,
+      "step": 2875
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.48828125,
+      "learning_rate": 1.6064579338060088e-06,
+      "loss": 2.0219,
+      "step": 2880
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.490234375,
+      "learning_rate": 1.5910981722497115e-06,
+      "loss": 2.0025,
+      "step": 2885
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.5625,
+      "learning_rate": 1.575798286923449e-06,
+      "loss": 2.0118,
+      "step": 2890
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.5078125,
+      "learning_rate": 1.5605585465646295e-06,
+      "loss": 1.9888,
+      "step": 2895
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.46875,
+      "learning_rate": 1.545379218854241e-06,
+      "loss": 1.9833,
+      "step": 2900
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.4609375,
+      "learning_rate": 1.5302605704121375e-06,
+      "loss": 2.0166,
+      "step": 2905
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.490234375,
+      "learning_rate": 1.5152028667923669e-06,
+      "loss": 2.0144,
+      "step": 2910
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.53125,
+      "learning_rate": 1.5002063724785004e-06,
+      "loss": 1.9796,
+      "step": 2915
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.490234375,
+      "learning_rate": 1.4852713508789835e-06,
+      "loss": 2.0341,
+      "step": 2920
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.51953125,
+      "learning_rate": 1.4703980643225208e-06,
+      "loss": 2.0127,
+      "step": 2925
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.4555867740534552e-06,
+      "loss": 1.9815,
+      "step": 2930
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.466796875,
+      "learning_rate": 1.4408377402271907e-06,
+      "loss": 2.0077,
+      "step": 2935
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.5,
+      "learning_rate": 1.4261512219056118e-06,
+      "loss": 2.008,
+      "step": 2940
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.411527477052544e-06,
+      "loss": 1.9747,
+      "step": 2945
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.3969667625292165e-06,
+      "loss": 1.9639,
+      "step": 2950
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.48046875,
+      "learning_rate": 1.3824693340897483e-06,
+      "loss": 1.9885,
+      "step": 2955
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.53125,
+      "learning_rate": 1.3680354463766642e-06,
+      "loss": 1.9922,
+      "step": 2960
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.48046875,
+      "learning_rate": 1.3536653529164146e-06,
+      "loss": 2.0102,
+      "step": 2965
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.3393593061149223e-06,
+      "loss": 1.998,
+      "step": 2970
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.498046875,
+      "learning_rate": 1.3251175572531583e-06,
+      "loss": 1.9725,
+      "step": 2975
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.55078125,
+      "learning_rate": 1.3109403564827155e-06,
+      "loss": 1.9735,
+      "step": 2980
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.484375,
+      "learning_rate": 1.2968279528214246e-06,
+      "loss": 1.9715,
+      "step": 2985
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.2827805941489767e-06,
+      "loss": 1.9592,
+      "step": 2990
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.5,
+      "learning_rate": 1.2687985272025655e-06,
+      "loss": 1.9479,
+      "step": 2995
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.451171875,
+      "learning_rate": 1.2548819975725624e-06,
+      "loss": 2.0103,
+      "step": 3000
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.462890625,
+      "learning_rate": 1.2410312496981923e-06,
+      "loss": 1.972,
+      "step": 3005
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.455078125,
+      "learning_rate": 1.2272465268632512e-06,
+      "loss": 1.9831,
+      "step": 3010
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.3828125,
+      "learning_rate": 1.2135280711918197e-06,
+      "loss": 1.9965,
+      "step": 3015
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.474609375,
+      "learning_rate": 1.1998761236440248e-06,
+      "loss": 2.0397,
+      "step": 3020
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.53125,
+      "learning_rate": 1.1862909240117982e-06,
+      "loss": 1.9739,
+      "step": 3025
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.43359375,
+      "learning_rate": 1.1727727109146632e-06,
+      "loss": 1.9653,
+      "step": 3030
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.51953125,
+      "learning_rate": 1.1593217217955522e-06,
+      "loss": 2.0289,
+      "step": 3035
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.40234375,
+      "learning_rate": 1.1459381929166251e-06,
+      "loss": 1.9883,
+      "step": 3040
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.439453125,
+      "learning_rate": 1.1326223593551294e-06,
+      "loss": 1.9846,
+      "step": 3045
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.44921875,
+      "learning_rate": 1.1193744549992675e-06,
+      "loss": 2.0217,
+      "step": 3050
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.546875,
+      "learning_rate": 1.1061947125440818e-06,
+      "loss": 2.0168,
+      "step": 3055
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.427734375,
+      "learning_rate": 1.0930833634873811e-06,
+      "loss": 1.981,
+      "step": 3060
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.0800406381256616e-06,
+      "loss": 1.9489,
+      "step": 3065
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.41796875,
+      "learning_rate": 1.0670667655500665e-06,
+      "loss": 1.9975,
+      "step": 3070
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.51171875,
+      "learning_rate": 1.0541619736423686e-06,
+      "loss": 1.969,
+      "step": 3075
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.494140625,
+      "learning_rate": 1.0413264890709546e-06,
+      "loss": 2.017,
+      "step": 3080
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.498046875,
+      "learning_rate": 1.0285605372868567e-06,
+      "loss": 2.0058,
+      "step": 3085
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.453125,
+      "learning_rate": 1.0158643425197817e-06,
+      "loss": 2.0093,
+      "step": 3090
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.003238127774181e-06,
+      "loss": 1.9601,
+      "step": 3095
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.578125,
+      "learning_rate": 9.906821148253303e-07,
+      "loss": 1.997,
+      "step": 3100
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.5625,
+      "learning_rate": 9.781965242154307e-07,
+      "loss": 2.0081,
+      "step": 3105
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.515625,
+      "learning_rate": 9.65781575249743e-07,
+      "loss": 1.9958,
+      "step": 3110
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.5390625,
+      "learning_rate": 9.534374859927248e-07,
+      "loss": 1.9913,
+      "step": 3115
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.52734375,
+      "learning_rate": 9.411644732642122e-07,
+      "loss": 2.0226,
+      "step": 3120
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.5546875,
+      "learning_rate": 9.289627526356048e-07,
+      "loss": 1.9847,
+      "step": 3125
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.419921875,
+      "learning_rate": 9.168325384260751e-07,
+      "loss": 1.9623,
+      "step": 3130
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.46875,
+      "learning_rate": 9.047740436988168e-07,
+      "loss": 1.9876,
+      "step": 3135
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.5546875,
+      "learning_rate": 8.927874802572861e-07,
+      "loss": 1.9982,
+      "step": 3140
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.515625,
+      "learning_rate": 8.808730586414982e-07,
+      "loss": 1.9776,
+      "step": 3145
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.4296875,
+      "learning_rate": 8.690309881243147e-07,
+      "loss": 2.0388,
+      "step": 3150
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.453125,
+      "learning_rate": 8.57261476707778e-07,
+      "loss": 2.0119,
+      "step": 3155
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.48046875,
+      "learning_rate": 8.455647311194537e-07,
+      "loss": 2.0039,
+      "step": 3160
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.470703125,
+      "learning_rate": 8.339409568087981e-07,
+      "loss": 1.9728,
+      "step": 3165
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.435546875,
+      "learning_rate": 8.223903579435499e-07,
+      "loss": 1.9684,
+      "step": 3170
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.416015625,
+      "learning_rate": 8.109131374061508e-07,
+      "loss": 1.9851,
+      "step": 3175
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.5234375,
+      "learning_rate": 7.995094967901701e-07,
+      "loss": 1.9839,
+      "step": 3180
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.50390625,
+      "learning_rate": 7.881796363967753e-07,
+      "loss": 2.0393,
+      "step": 3185
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.44140625,
+      "learning_rate": 7.769237552312048e-07,
+      "loss": 1.9957,
+      "step": 3190
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.5234375,
+      "learning_rate": 7.657420509992785e-07,
+      "loss": 1.9819,
+      "step": 3195
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.470703125,
+      "learning_rate": 7.546347201039255e-07,
+      "loss": 2.0461,
+      "step": 3200
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.5546875,
+      "learning_rate": 7.436019576417258e-07,
+      "loss": 2.0057,
+      "step": 3205
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.45703125,
+      "learning_rate": 7.326439573994953e-07,
+      "loss": 1.9784,
+      "step": 3210
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.45703125,
+      "learning_rate": 7.217609118508722e-07,
+      "loss": 1.9894,
+      "step": 3215
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.478515625,
+      "learning_rate": 7.109530121529439e-07,
+      "loss": 2.0161,
+      "step": 3220
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.466796875,
+      "learning_rate": 7.002204481428826e-07,
+      "loss": 1.9951,
+      "step": 3225
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.478515625,
+      "learning_rate": 6.895634083346159e-07,
+      "loss": 1.9595,
+      "step": 3230
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.53515625,
+      "learning_rate": 6.789820799155155e-07,
+      "loss": 1.9779,
+      "step": 3235
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.53515625,
+      "learning_rate": 6.684766487431027e-07,
+      "loss": 1.9571,
+      "step": 3240
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.51171875,
+      "learning_rate": 6.580472993417952e-07,
+      "loss": 2.0053,
+      "step": 3245
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.51171875,
+      "learning_rate": 6.476942148996529e-07,
+      "loss": 1.9901,
+      "step": 3250
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.44140625,
+      "learning_rate": 6.374175772651725e-07,
+      "loss": 2.011,
+      "step": 3255
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.5234375,
+      "learning_rate": 6.272175669440861e-07,
+      "loss": 1.9691,
+      "step": 3260
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.484375,
+      "learning_rate": 6.170943630961918e-07,
+      "loss": 1.9929,
+      "step": 3265
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.5625,
+      "learning_rate": 6.070481435322062e-07,
+      "loss": 2.0098,
+      "step": 3270
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.48828125,
+      "learning_rate": 5.970790847106461e-07,
+      "loss": 1.9893,
+      "step": 3275
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.470703125,
+      "learning_rate": 5.871873617347218e-07,
+      "loss": 2.0239,
+      "step": 3280
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.498046875,
+      "learning_rate": 5.773731483492684e-07,
+      "loss": 1.9829,
+      "step": 3285
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.46484375,
+      "learning_rate": 5.676366169376873e-07,
+      "loss": 2.0305,
+      "step": 3290
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.46484375,
+      "learning_rate": 5.579779385189261e-07,
+      "loss": 2.0101,
+      "step": 3295
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.52734375,
+      "learning_rate": 5.483972827444645e-07,
+      "loss": 2.0039,
+      "step": 3300
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.443359375,
+      "learning_rate": 5.388948178953462e-07,
+      "loss": 1.9907,
+      "step": 3305
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.42578125,
+      "learning_rate": 5.294707108792146e-07,
+      "loss": 1.9854,
+      "step": 3310
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.498046875,
+      "learning_rate": 5.201251272273833e-07,
+      "loss": 1.9777,
+      "step": 3315
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.482421875,
+      "learning_rate": 5.108582310919302e-07,
+      "loss": 1.9771,
+      "step": 3320
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.4921875,
+      "learning_rate": 5.016701852428113e-07,
+      "loss": 1.9773,
+      "step": 3325
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.443359375,
+      "learning_rate": 4.925611510650058e-07,
+      "loss": 2.0213,
+      "step": 3330
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.46875,
+      "learning_rate": 4.835312885556781e-07,
+      "loss": 2.045,
+      "step": 3335
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.58984375,
+      "learning_rate": 4.745807563213678e-07,
+      "loss": 2.0107,
+      "step": 3340
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.49609375,
+      "learning_rate": 4.6570971157520574e-07,
+      "loss": 1.9895,
+      "step": 3345
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.458984375,
+      "learning_rate": 4.569183101341501e-07,
+      "loss": 2.0058,
+      "step": 3350
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.470703125,
+      "learning_rate": 4.482067064162521e-07,
+      "loss": 1.9324,
+      "step": 3355
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.470703125,
+      "learning_rate": 4.3957505343794115e-07,
+      "loss": 1.9871,
+      "step": 3360
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.470703125,
+      "learning_rate": 4.310235028113402e-07,
+      "loss": 2.0171,
+      "step": 3365
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.53125,
+      "learning_rate": 4.225522047415992e-07,
+      "loss": 1.978,
+      "step": 3370
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.447265625,
+      "learning_rate": 4.141613080242579e-07,
+      "loss": 2.018,
+      "step": 3375
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.44140625,
+      "learning_rate": 4.058509600426358e-07,
+      "loss": 1.9975,
+      "step": 3380
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.5,
+      "learning_rate": 3.976213067652396e-07,
+      "loss": 2.0103,
+      "step": 3385
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.46875,
+      "learning_rate": 3.894724927431981e-07,
+      "loss": 1.9982,
+      "step": 3390
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.423828125,
+      "learning_rate": 3.8140466110772875e-07,
+      "loss": 2.0159,
+      "step": 3395
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.474609375,
+      "learning_rate": 3.734179535676169e-07,
+      "loss": 1.9809,
+      "step": 3400
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.494140625,
+      "learning_rate": 3.655125104067325e-07,
+      "loss": 1.997,
+      "step": 3405
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.462890625,
+      "learning_rate": 3.576884704815631e-07,
+      "loss": 1.9811,
+      "step": 3410
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.458984375,
+      "learning_rate": 3.499459712187742e-07,
+      "loss": 2.0204,
+      "step": 3415
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.515625,
+      "learning_rate": 3.422851486127987e-07,
+      "loss": 1.9684,
+      "step": 3420
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.453125,
+      "learning_rate": 3.3470613722344335e-07,
+      "loss": 2.0207,
+      "step": 3425
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.6171875,
+      "learning_rate": 3.272090701735314e-07,
+      "loss": 1.9406,
+      "step": 3430
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.5390625,
+      "learning_rate": 3.1979407914655756e-07,
+      "loss": 2.0232,
+      "step": 3435
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.57421875,
+      "learning_rate": 3.1246129438438076e-07,
+      "loss": 1.9855,
+      "step": 3440
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.5078125,
+      "learning_rate": 3.0521084468493446e-07,
+      "loss": 2.0071,
+      "step": 3445
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.47265625,
+      "learning_rate": 2.980428573999605e-07,
+      "loss": 1.9886,
+      "step": 3450
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.435546875,
+      "learning_rate": 2.909574584327812e-07,
+      "loss": 2.0096,
+      "step": 3455
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.44140625,
+      "learning_rate": 2.839547722360769e-07,
+      "loss": 1.9977,
+      "step": 3460
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.53515625,
+      "learning_rate": 2.770349218097096e-07,
+      "loss": 1.956,
+      "step": 3465
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.4375,
+      "learning_rate": 2.7019802869855783e-07,
+      "loss": 2.0202,
+      "step": 3470
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.51953125,
+      "learning_rate": 2.6344421299037915e-07,
+      "loss": 2.0191,
+      "step": 3475
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.5234375,
+      "learning_rate": 2.5677359331370834e-07,
+      "loss": 2.0047,
+      "step": 3480
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.482421875,
+      "learning_rate": 2.5018628683576796e-07,
+      "loss": 1.9828,
+      "step": 3485
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.5,
+      "learning_rate": 2.436824092604101e-07,
+      "loss": 2.0087,
+      "step": 3490
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.48828125,
+      "learning_rate": 2.3726207482608842e-07,
+      "loss": 1.9827,
+      "step": 3495
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.455078125,
+      "learning_rate": 2.309253963038477e-07,
+      "loss": 1.9635,
+      "step": 3500
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.5,
+      "learning_rate": 2.246724849953452e-07,
+      "loss": 2.0159,
+      "step": 3505
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.462890625,
+      "learning_rate": 2.1850345073089253e-07,
+      "loss": 2.009,
+      "step": 3510
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.58203125,
+      "learning_rate": 2.1241840186753238e-07,
+      "loss": 1.9898,
+      "step": 3515
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.50390625,
+      "learning_rate": 2.0641744528712925e-07,
+      "loss": 2.0171,
+      "step": 3520
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.48046875,
+      "learning_rate": 2.0050068639449472e-07,
+      "loss": 2.006,
+      "step": 3525
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.46484375,
+      "learning_rate": 1.9466822911553772e-07,
+      "loss": 2.0039,
+      "step": 3530
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.458984375,
+      "learning_rate": 1.889201758954351e-07,
+      "loss": 2.0155,
+      "step": 3535
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.4921875,
+      "learning_rate": 1.832566276968345e-07,
+      "loss": 2.0047,
+      "step": 3540
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.4609375,
+      "learning_rate": 1.7767768399808372e-07,
+      "loss": 2.0088,
+      "step": 3545
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.458984375,
+      "learning_rate": 1.7218344279147702e-07,
+      "loss": 1.978,
+      "step": 3550
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.455078125,
+      "learning_rate": 1.6677400058153993e-07,
+      "loss": 1.9719,
+      "step": 3555
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.48828125,
+      "learning_rate": 1.6144945238332987e-07,
+      "loss": 1.9975,
+      "step": 3560
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.458984375,
+      "learning_rate": 1.5620989172077106e-07,
+      "loss": 1.9942,
+      "step": 3565
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.41015625,
+      "learning_rate": 1.5105541062500838e-07,
+      "loss": 1.9904,
+      "step": 3570
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.4598609963279164e-07,
+      "loss": 2.0009,
+      "step": 3575
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.5703125,
+      "learning_rate": 1.4100204778488947e-07,
+      "loss": 2.0399,
+      "step": 3580
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.4765625,
+      "learning_rate": 1.361033426245173e-07,
+      "loss": 1.9531,
+      "step": 3585
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.478515625,
+      "learning_rate": 1.3129007019580752e-07,
+      "loss": 1.9986,
+      "step": 3590
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.46875,
+      "learning_rate": 1.265623150422951e-07,
+      "loss": 1.9936,
+      "step": 3595
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.5078125,
+      "learning_rate": 1.2192016020542986e-07,
+      "loss": 2.03,
+      "step": 3600
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.486328125,
+      "learning_rate": 1.1736368722312263e-07,
+      "loss": 1.9686,
+      "step": 3605
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.45703125,
+      "learning_rate": 1.1289297612830984e-07,
+      "loss": 2.0041,
+      "step": 3610
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.0850810544754897e-07,
+      "loss": 2.0213,
+      "step": 3615
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.0420915219964023e-07,
+      "loss": 1.9643,
+      "step": 3620
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.435546875,
+      "learning_rate": 9.999619189427102e-08,
+      "loss": 2.0194,
+      "step": 3625
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.46484375,
+      "learning_rate": 9.586929853069139e-08,
+      "loss": 2.0342,
+      "step": 3630
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.5390625,
+      "learning_rate": 9.182854459641565e-08,
+      "loss": 1.9825,
+      "step": 3635
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.45703125,
+      "learning_rate": 8.787400106594568e-08,
+      "loss": 1.9844,
+      "step": 3640
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.51171875,
+      "learning_rate": 8.400573739952677e-08,
+      "loss": 2.0232,
+      "step": 3645
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.46875,
+      "learning_rate": 8.02238215419282e-08,
+      "loss": 2.0074,
+      "step": 3650
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.515625,
+      "learning_rate": 7.652831992124799e-08,
+      "loss": 1.9956,
+      "step": 3655
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.466796875,
+      "learning_rate": 7.291929744774495e-08,
+      "loss": 1.982,
+      "step": 3660
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.5234375,
+      "learning_rate": 6.939681751270356e-08,
+      "loss": 1.9714,
+      "step": 3665
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.46875,
+      "learning_rate": 6.596094198731473e-08,
+      "loss": 1.9725,
+      "step": 3670
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.55078125,
+      "learning_rate": 6.261173122159292e-08,
+      "loss": 1.9973,
+      "step": 3675
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.515625,
+      "learning_rate": 5.934924404331355e-08,
+      "loss": 1.9777,
+      "step": 3680
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.458984375,
+      "learning_rate": 5.6173537756982756e-08,
+      "loss": 1.9547,
+      "step": 3685
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.5546875,
+      "learning_rate": 5.30846681428282e-08,
+      "loss": 1.9781,
+      "step": 3690
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.46875,
+      "learning_rate": 5.00826894558204e-08,
+      "loss": 1.9667,
+      "step": 3695
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.5,
+      "learning_rate": 4.716765442471849e-08,
+      "loss": 2.0228,
+      "step": 3700
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.51953125,
+      "learning_rate": 4.433961425114708e-08,
+      "loss": 2.0139,
+      "step": 3705
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.48046875,
+      "learning_rate": 4.1598618608693074e-08,
+      "loss": 2.0082,
+      "step": 3710
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.5625,
+      "learning_rate": 3.894471564203639e-08,
+      "loss": 1.9803,
+      "step": 3715
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.494140625,
+      "learning_rate": 3.637795196610228e-08,
+      "loss": 1.9956,
+      "step": 3720
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.458984375,
+      "learning_rate": 3.3898372665243096e-08,
+      "loss": 2.0058,
+      "step": 3725
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.46484375,
+      "learning_rate": 3.1506021292447285e-08,
+      "loss": 1.9566,
+      "step": 3730
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.55078125,
+      "learning_rate": 2.9200939868573308e-08,
+      "loss": 2.0039,
+      "step": 3735
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.52734375,
+      "learning_rate": 2.6983168881611897e-08,
+      "loss": 1.9406,
+      "step": 3740
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.4453125,
+      "learning_rate": 2.485274728597442e-08,
+      "loss": 2.0118,
+      "step": 3745
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.482421875,
+      "learning_rate": 2.2809712501810632e-08,
+      "loss": 2.0342,
+      "step": 3750
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.546875,
+      "learning_rate": 2.08541004143481e-08,
+      "loss": 1.983,
+      "step": 3755
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.474609375,
+      "learning_rate": 1.898594537326437e-08,
+      "loss": 2.0319,
+      "step": 3760
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.4765625,
+      "learning_rate": 1.7205280192084117e-08,
+      "loss": 1.9244,
+      "step": 3765
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.5078125,
+      "learning_rate": 1.5512136147600167e-08,
+      "loss": 1.9726,
+      "step": 3770
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.466796875,
+      "learning_rate": 1.3906542979326143e-08,
+      "loss": 2.0218,
+      "step": 3775
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.53515625,
+      "learning_rate": 1.2388528888973017e-08,
+      "loss": 2.0047,
+      "step": 3780
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.412109375,
+      "learning_rate": 1.0958120539955596e-08,
+      "loss": 1.977,
+      "step": 3785
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.43359375,
+      "learning_rate": 9.615343056922355e-09,
+      "loss": 1.9978,
+      "step": 3790
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.55078125,
+      "learning_rate": 8.360220025313005e-09,
+      "loss": 1.9994,
+      "step": 3795
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.46484375,
+      "learning_rate": 7.1927734909488235e-09,
+      "loss": 2.0006,
+      "step": 3800
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.42578125,
+      "learning_rate": 6.113023959640751e-09,
+      "loss": 1.9753,
+      "step": 3805
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.4375,
+      "learning_rate": 5.120990396830228e-09,
+      "loss": 1.9885,
+      "step": 3810
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.48828125,
+      "learning_rate": 4.216690227258902e-09,
+      "loss": 2.0332,
+      "step": 3815
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.57421875,
+      "learning_rate": 3.400139334658881e-09,
+      "loss": 2.0213,
+      "step": 3820
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.54296875,
+      "learning_rate": 2.6713520614757295e-09,
+      "loss": 1.9891,
+      "step": 3825
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.48046875,
+      "learning_rate": 2.0303412086164487e-09,
+      "loss": 2.0187,
+      "step": 3830
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.412109375,
+      "learning_rate": 1.477118035223546e-09,
+      "loss": 1.9992,
+      "step": 3835
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.486328125,
+      "learning_rate": 1.01169225847908e-09,
+      "loss": 1.9885,
+      "step": 3840
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.4453125,
+      "learning_rate": 6.340720534325772e-10,
+      "loss": 1.9889,
+      "step": 3845
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.515625,
+      "learning_rate": 3.4426405285725715e-10,
+      "loss": 1.9909,
+      "step": 3850
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.439453125,
+      "learning_rate": 1.4227334713456943e-10,
+      "loss": 1.9585,
+      "step": 3855
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.447265625,
+      "learning_rate": 2.8103484164820894e-11,
+      "loss": 1.9766,
+      "step": 3860
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 1.9850057363510132,
+      "eval_runtime": 246.6106,
+      "eval_samples_per_second": 9.302,
+      "eval_steps_per_second": 4.651,
+      "step": 3864
+    },
+    {
+      "epoch": 3.0,
+      "step": 3864,
+      "total_flos": 8.050802347506401e+17,
+      "train_loss": 2.091761199090298,
+      "train_runtime": 21448.6469,
+      "train_samples_per_second": 2.883,
+      "train_steps_per_second": 0.18
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 3864,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 1000,
+  "total_flos": 8.050802347506401e+17,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}