diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5617 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 3982,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.71875,
+      "learning_rate": 5.0125313283208025e-08,
+      "loss": 2.0023,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.578125,
+      "learning_rate": 2.506265664160401e-07,
+      "loss": 1.9652,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.484375,
+      "learning_rate": 5.012531328320802e-07,
+      "loss": 1.9643,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 3.40625,
+      "learning_rate": 7.518796992481203e-07,
+      "loss": 1.9965,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.40625,
+      "learning_rate": 1.0025062656641603e-06,
+      "loss": 1.9539,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.578125,
+      "learning_rate": 1.2531328320802005e-06,
+      "loss": 1.9449,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.4375,
+      "learning_rate": 1.5037593984962406e-06,
+      "loss": 1.9467,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.546875,
+      "learning_rate": 1.7543859649122807e-06,
+      "loss": 1.9068,
+      "step": 35
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.234375,
+      "learning_rate": 2.0050125313283207e-06,
+      "loss": 2.0021,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.15625,
+      "learning_rate": 2.255639097744361e-06,
+      "loss": 1.9756,
+      "step": 45
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.171875,
+      "learning_rate": 2.506265664160401e-06,
+      "loss": 1.9301,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.96875,
+      "learning_rate": 2.7568922305764413e-06,
+      "loss": 1.8914,
+      "step": 55
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.921875,
+      "learning_rate": 3.007518796992481e-06,
+      "loss": 1.9242,
+      "step": 60
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.828125,
+      "learning_rate": 3.2581453634085216e-06,
+      "loss": 1.8973,
+      "step": 65
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.59375,
+      "learning_rate": 3.5087719298245615e-06,
+      "loss": 1.9371,
+      "step": 70
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.59375,
+      "learning_rate": 3.7593984962406014e-06,
+      "loss": 1.8551,
+      "step": 75
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.53125,
+      "learning_rate": 4.010025062656641e-06,
+      "loss": 1.8906,
+      "step": 80
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.265625,
+      "learning_rate": 4.260651629072682e-06,
+      "loss": 1.8873,
+      "step": 85
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.359375,
+      "learning_rate": 4.511278195488722e-06,
+      "loss": 1.8961,
+      "step": 90
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.328125,
+      "learning_rate": 4.761904761904762e-06,
+      "loss": 1.7845,
+      "step": 95
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.125,
+      "learning_rate": 5.012531328320802e-06,
+      "loss": 1.8556,
+      "step": 100
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.109375,
+      "learning_rate": 5.263157894736842e-06,
+      "loss": 1.8533,
+      "step": 105
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.078125,
+      "learning_rate": 5.5137844611528826e-06,
+      "loss": 1.8227,
+      "step": 110
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.0625,
+      "learning_rate": 5.764411027568922e-06,
+      "loss": 1.7946,
+      "step": 115
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.96875,
+      "learning_rate": 6.015037593984962e-06,
+      "loss": 1.8398,
+      "step": 120
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.9609375,
+      "learning_rate": 6.265664160401003e-06,
+      "loss": 1.7794,
+      "step": 125
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.9921875,
+      "learning_rate": 6.516290726817043e-06,
+      "loss": 1.8295,
+      "step": 130
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.9765625,
+      "learning_rate": 6.766917293233083e-06,
+      "loss": 1.7832,
+      "step": 135
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.9140625,
+      "learning_rate": 7.017543859649123e-06,
+      "loss": 1.7952,
+      "step": 140
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.9140625,
+      "learning_rate": 7.268170426065163e-06,
+      "loss": 1.7996,
+      "step": 145
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.8984375,
+      "learning_rate": 7.518796992481203e-06,
+      "loss": 1.7931,
+      "step": 150
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.921875,
+      "learning_rate": 7.769423558897243e-06,
+      "loss": 1.7922,
+      "step": 155
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.90625,
+      "learning_rate": 8.020050125313283e-06,
+      "loss": 1.7766,
+      "step": 160
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.125,
+      "learning_rate": 8.270676691729324e-06,
+      "loss": 1.7967,
+      "step": 165
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.8828125,
+      "learning_rate": 8.521303258145363e-06,
+      "loss": 1.7879,
+      "step": 170
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.875,
+      "learning_rate": 8.771929824561405e-06,
+      "loss": 1.7487,
+      "step": 175
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.8125,
+      "learning_rate": 9.022556390977444e-06,
+      "loss": 1.8029,
+      "step": 180
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.984375,
+      "learning_rate": 9.273182957393484e-06,
+      "loss": 1.7654,
+      "step": 185
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.015625,
+      "learning_rate": 9.523809523809525e-06,
+      "loss": 1.7712,
+      "step": 190
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.859375,
+      "learning_rate": 9.774436090225564e-06,
+      "loss": 1.7906,
+      "step": 195
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.8046875,
+      "learning_rate": 1.0025062656641604e-05,
+      "loss": 1.7939,
+      "step": 200
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.7421875,
+      "learning_rate": 1.0275689223057645e-05,
+      "loss": 1.8455,
+      "step": 205
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.890625,
+      "learning_rate": 1.0526315789473684e-05,
+      "loss": 1.7572,
+      "step": 210
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.890625,
+      "learning_rate": 1.0776942355889726e-05,
+      "loss": 1.7663,
+      "step": 215
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.8984375,
+      "learning_rate": 1.1027568922305765e-05,
+      "loss": 1.7555,
+      "step": 220
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.8828125,
+      "learning_rate": 1.1278195488721806e-05,
+      "loss": 1.742,
+      "step": 225
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.875,
+      "learning_rate": 1.1528822055137844e-05,
+      "loss": 1.7503,
+      "step": 230
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.859375,
+      "learning_rate": 1.1779448621553885e-05,
+      "loss": 1.7498,
+      "step": 235
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.84375,
+      "learning_rate": 1.2030075187969925e-05,
+      "loss": 1.7433,
+      "step": 240
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.984375,
+      "learning_rate": 1.2280701754385966e-05,
+      "loss": 1.7346,
+      "step": 245
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.7890625,
+      "learning_rate": 1.2531328320802006e-05,
+      "loss": 1.7372,
+      "step": 250
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.84375,
+      "learning_rate": 1.2781954887218047e-05,
+      "loss": 1.7331,
+      "step": 255
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.875,
+      "learning_rate": 1.3032581453634086e-05,
+      "loss": 1.7573,
+      "step": 260
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.9296875,
+      "learning_rate": 1.3283208020050127e-05,
+      "loss": 1.7163,
+      "step": 265
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.046875,
+      "learning_rate": 1.3533834586466165e-05,
+      "loss": 1.6598,
+      "step": 270
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.9296875,
+      "learning_rate": 1.3784461152882206e-05,
+      "loss": 1.7192,
+      "step": 275
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.90625,
+      "learning_rate": 1.4035087719298246e-05,
+      "loss": 1.7514,
+      "step": 280
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.953125,
+      "learning_rate": 1.4285714285714287e-05,
+      "loss": 1.7656,
+      "step": 285
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.7890625,
+      "learning_rate": 1.4536340852130327e-05,
+      "loss": 1.7618,
+      "step": 290
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.953125,
+      "learning_rate": 1.4786967418546368e-05,
+      "loss": 1.7497,
+      "step": 295
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.7734375,
+      "learning_rate": 1.5037593984962406e-05,
+      "loss": 1.737,
+      "step": 300
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.734375,
+      "learning_rate": 1.5288220551378447e-05,
+      "loss": 1.7304,
+      "step": 305
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.9140625,
+      "learning_rate": 1.5538847117794486e-05,
+      "loss": 1.7251,
+      "step": 310
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.90625,
+      "learning_rate": 1.578947368421053e-05,
+      "loss": 1.7399,
+      "step": 315
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.765625,
+      "learning_rate": 1.6040100250626565e-05,
+      "loss": 1.7211,
+      "step": 320
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.890625,
+      "learning_rate": 1.6290726817042608e-05,
+      "loss": 1.676,
+      "step": 325
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.875,
+      "learning_rate": 1.6541353383458648e-05,
+      "loss": 1.7265,
+      "step": 330
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.828125,
+      "learning_rate": 1.6791979949874687e-05,
+      "loss": 1.7135,
+      "step": 335
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.8359375,
+      "learning_rate": 1.7042606516290727e-05,
+      "loss": 1.7441,
+      "step": 340
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.6953125,
+      "learning_rate": 1.729323308270677e-05,
+      "loss": 1.7039,
+      "step": 345
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.7734375,
+      "learning_rate": 1.754385964912281e-05,
+      "loss": 1.6405,
+      "step": 350
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.7109375,
+      "learning_rate": 1.779448621553885e-05,
+      "loss": 1.7578,
+      "step": 355
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.9453125,
+      "learning_rate": 1.8045112781954888e-05,
+      "loss": 1.7595,
+      "step": 360
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.640625,
+      "learning_rate": 1.8295739348370928e-05,
+      "loss": 1.7058,
+      "step": 365
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.875,
+      "learning_rate": 1.8546365914786967e-05,
+      "loss": 1.6768,
+      "step": 370
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.8671875,
+      "learning_rate": 1.879699248120301e-05,
+      "loss": 1.7218,
+      "step": 375
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.6796875,
+      "learning_rate": 1.904761904761905e-05,
+      "loss": 1.6976,
+      "step": 380
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.7890625,
+      "learning_rate": 1.929824561403509e-05,
+      "loss": 1.7352,
+      "step": 385
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.71875,
+      "learning_rate": 1.954887218045113e-05,
+      "loss": 1.6723,
+      "step": 390
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.6796875,
+      "learning_rate": 1.9799498746867168e-05,
+      "loss": 1.719,
+      "step": 395
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.8828125,
+      "learning_rate": 1.9999996156064378e-05,
+      "loss": 1.6525,
+      "step": 400
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.78125,
+      "learning_rate": 1.9999861618627885e-05,
+      "loss": 1.6935,
+      "step": 405
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.859375,
+      "learning_rate": 1.999953488736543e-05,
+      "loss": 1.6571,
+      "step": 410
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.734375,
+      "learning_rate": 1.999901596855667e-05,
+      "loss": 1.7119,
+      "step": 415
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.6796875,
+      "learning_rate": 1.9998304872175046e-05,
+      "loss": 1.7224,
+      "step": 420
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.8046875,
+      "learning_rate": 1.9997401611887574e-05,
+      "loss": 1.7085,
+      "step": 425
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.7109375,
+      "learning_rate": 1.9996306205054605e-05,
+      "loss": 1.7252,
+      "step": 430
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.8125,
+      "learning_rate": 1.999501867272947e-05,
+      "loss": 1.7133,
+      "step": 435
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.828125,
+      "learning_rate": 1.999353903965809e-05,
+      "loss": 1.6672,
+      "step": 440
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.6015625,
+      "learning_rate": 1.9991867334278497e-05,
+      "loss": 1.6256,
+      "step": 445
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.7421875,
+      "learning_rate": 1.999000358872027e-05,
+      "loss": 1.6789,
+      "step": 450
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.703125,
+      "learning_rate": 1.9987947838803955e-05,
+      "loss": 1.68,
+      "step": 455
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.625,
+      "learning_rate": 1.9985700124040336e-05,
+      "loss": 1.6815,
+      "step": 460
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.65625,
+      "learning_rate": 1.9983260487629698e-05,
+      "loss": 1.6655,
+      "step": 465
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.6875,
+      "learning_rate": 1.9980628976461004e-05,
+      "loss": 1.6622,
+      "step": 470
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.546875,
+      "learning_rate": 1.9977805641110968e-05,
+      "loss": 1.6893,
+      "step": 475
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.765625,
+      "learning_rate": 1.997479053584311e-05,
+      "loss": 1.7168,
+      "step": 480
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.7265625,
+      "learning_rate": 1.997158371860668e-05,
+      "loss": 1.6251,
+      "step": 485
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.6875,
+      "learning_rate": 1.9968185251035594e-05,
+      "loss": 1.7348,
+      "step": 490
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.7421875,
+      "learning_rate": 1.9964595198447194e-05,
+      "loss": 1.6948,
+      "step": 495
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.996081362984103e-05,
+      "loss": 1.6973,
+      "step": 500
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.995684061789753e-05,
+      "loss": 1.6703,
+      "step": 505
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.7578125,
+      "learning_rate": 1.9952676238976574e-05,
+      "loss": 1.6736,
+      "step": 510
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.6328125,
+      "learning_rate": 1.9948320573116073e-05,
+      "loss": 1.6812,
+      "step": 515
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.71875,
+      "learning_rate": 1.9943773704030387e-05,
+      "loss": 1.6962,
+      "step": 520
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.65625,
+      "learning_rate": 1.993903571910875e-05,
+      "loss": 1.6456,
+      "step": 525
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.9934106709413555e-05,
+      "loss": 1.6697,
+      "step": 530
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.6171875,
+      "learning_rate": 1.9928986769678643e-05,
+      "loss": 1.6567,
+      "step": 535
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.65625,
+      "learning_rate": 1.992367599830746e-05,
+      "loss": 1.685,
+      "step": 540
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.5546875,
+      "learning_rate": 1.9918174497371157e-05,
+      "loss": 1.6425,
+      "step": 545
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.6484375,
+      "learning_rate": 1.9912482372606656e-05,
+      "loss": 1.6892,
+      "step": 550
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.640625,
+      "learning_rate": 1.990659973341459e-05,
+      "loss": 1.7066,
+      "step": 555
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.578125,
+      "learning_rate": 1.990052669285722e-05,
+      "loss": 1.7015,
+      "step": 560
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.6640625,
+      "learning_rate": 1.989426336765625e-05,
+      "loss": 1.6728,
+      "step": 565
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.9887809878190584e-05,
+      "loss": 1.6693,
+      "step": 570
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.988116634849403e-05,
+      "loss": 1.6836,
+      "step": 575
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.578125,
+      "learning_rate": 1.9874332906252888e-05,
+      "loss": 1.6245,
+      "step": 580
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.640625,
+      "learning_rate": 1.9867309682803518e-05,
+      "loss": 1.6658,
+      "step": 585
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.986009681312981e-05,
+      "loss": 1.6716,
+      "step": 590
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.6015625,
+      "learning_rate": 1.985269443586058e-05,
+      "loss": 1.6758,
+      "step": 595
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.984510269326692e-05,
+      "loss": 1.633,
+      "step": 600
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.6328125,
+      "learning_rate": 1.9837321731259452e-05,
+      "loss": 1.6552,
+      "step": 605
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.982935169938553e-05,
+      "loss": 1.6836,
+      "step": 610
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.671875,
+      "learning_rate": 1.9821192750826373e-05,
+      "loss": 1.6429,
+      "step": 615
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.9812845042394095e-05,
+      "loss": 1.6634,
+      "step": 620
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.578125,
+      "learning_rate": 1.980430873452873e-05,
+      "loss": 1.6964,
+      "step": 625
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.9795583991295104e-05,
+      "loss": 1.7073,
+      "step": 630
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.9786670980379724e-05,
+      "loss": 1.6364,
+      "step": 635
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.640625,
+      "learning_rate": 1.977756987308752e-05,
+      "loss": 1.7014,
+      "step": 640
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.976828084433858e-05,
+      "loss": 1.6339,
+      "step": 645
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.5234375,
+      "learning_rate": 1.9758804072664773e-05,
+      "loss": 1.6574,
+      "step": 650
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.6015625,
+      "learning_rate": 1.9749139740206318e-05,
+      "loss": 1.6337,
+      "step": 655
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.973928803270829e-05,
+      "loss": 1.6727,
+      "step": 660
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.5546875,
+      "learning_rate": 1.9729249139517045e-05,
+      "loss": 1.6769,
+      "step": 665
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.971902325357658e-05,
+      "loss": 1.6548,
+      "step": 670
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.640625,
+      "learning_rate": 1.9708610571424825e-05,
+      "loss": 1.6741,
+      "step": 675
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.6328125,
+      "learning_rate": 1.9698011293189884e-05,
+      "loss": 1.6463,
+      "step": 680
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.6484375,
+      "learning_rate": 1.968722562258615e-05,
+      "loss": 1.6643,
+      "step": 685
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.53125,
+      "learning_rate": 1.967625376691042e-05,
+      "loss": 1.6968,
+      "step": 690
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.53125,
+      "learning_rate": 1.966509593703791e-05,
+      "loss": 1.6512,
+      "step": 695
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.7109375,
+      "learning_rate": 1.965375234741819e-05,
+      "loss": 1.648,
+      "step": 700
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.964222321607106e-05,
+      "loss": 1.6885,
+      "step": 705
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.5234375,
+      "learning_rate": 1.9630508764582374e-05,
+      "loss": 1.6735,
+      "step": 710
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.578125,
+      "learning_rate": 1.9618609218099773e-05,
+      "loss": 1.6813,
+      "step": 715
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.5546875,
+      "learning_rate": 1.960652480532836e-05,
+      "loss": 1.682,
+      "step": 720
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.59375,
+      "learning_rate": 1.9594255758526305e-05,
+      "loss": 1.6807,
+      "step": 725
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.6015625,
+      "learning_rate": 1.9581802313500367e-05,
+      "loss": 1.6265,
+      "step": 730
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.546875,
+      "learning_rate": 1.9569164709601388e-05,
+      "loss": 1.6634,
+      "step": 735
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.5,
+      "learning_rate": 1.9556343189719675e-05,
+      "loss": 1.6635,
+      "step": 740
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.954333800028033e-05,
+      "loss": 1.6683,
+      "step": 745
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.953014939123853e-05,
+      "loss": 1.6442,
+      "step": 750
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.578125,
+      "learning_rate": 1.95167776160747e-05,
+      "loss": 1.6648,
+      "step": 755
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.9503222931789656e-05,
+      "loss": 1.6437,
+      "step": 760
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.53125,
+      "learning_rate": 1.9489485598899673e-05,
+      "loss": 1.6726,
+      "step": 765
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.9475565881431458e-05,
+      "loss": 1.6561,
+      "step": 770
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.9461464046917092e-05,
+      "loss": 1.6713,
+      "step": 775
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.944718036638888e-05,
+      "loss": 1.6848,
+      "step": 780
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.546875,
+      "learning_rate": 1.9432715114374133e-05,
+      "loss": 1.6622,
+      "step": 785
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.6171875,
+      "learning_rate": 1.9418068568889918e-05,
+      "loss": 1.7046,
+      "step": 790
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.515625,
+      "learning_rate": 1.940324101143769e-05,
+      "loss": 1.6199,
+      "step": 795
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.938823272699789e-05,
+      "loss": 1.6755,
+      "step": 800
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.937304400402448e-05,
+      "loss": 1.6476,
+      "step": 805
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.546875,
+      "learning_rate": 1.9357675134439366e-05,
+      "loss": 1.6813,
+      "step": 810
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.9342126413626825e-05,
+      "loss": 1.6479,
+      "step": 815
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.9326398140427806e-05,
+      "loss": 1.668,
+      "step": 820
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.578125,
+      "learning_rate": 1.9310490617134196e-05,
+      "loss": 1.6647,
+      "step": 825
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.5546875,
+      "learning_rate": 1.9294404149482996e-05,
+      "loss": 1.637,
+      "step": 830
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.9278139046650463e-05,
+      "loss": 1.6636,
+      "step": 835
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.6171875,
+      "learning_rate": 1.9261695621246156e-05,
+      "loss": 1.6229,
+      "step": 840
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.5078125,
+      "learning_rate": 1.924507418930694e-05,
+      "loss": 1.6384,
+      "step": 845
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.5078125,
+      "learning_rate": 1.9228275070290885e-05,
+      "loss": 1.6909,
+      "step": 850
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.9211298587071165e-05,
+      "loss": 1.6706,
+      "step": 855
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.5859375,
+      "learning_rate": 1.919414506592981e-05,
+      "loss": 1.6457,
+      "step": 860
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.9176814836551478e-05,
+      "loss": 1.6747,
+      "step": 865
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.9159308232017076e-05,
+      "loss": 1.6199,
+      "step": 870
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.9141625588797404e-05,
+      "loss": 1.7087,
+      "step": 875
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.515625,
+      "learning_rate": 1.912376724674664e-05,
+      "loss": 1.7007,
+      "step": 880
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.910573354909585e-05,
+      "loss": 1.6024,
+      "step": 885
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.484375,
+      "learning_rate": 1.9087524842446357e-05,
+      "loss": 1.6815,
+      "step": 890
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.484375,
+      "learning_rate": 1.9069141476763108e-05,
+      "loss": 1.6368,
+      "step": 895
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.9050583805367932e-05,
+      "loss": 1.6118,
+      "step": 900
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.5078125,
+      "learning_rate": 1.903185218493275e-05,
+      "loss": 1.6702,
+      "step": 905
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.5078125,
+      "learning_rate": 1.901294697547272e-05,
+      "loss": 1.6367,
+      "step": 910
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.899386854033933e-05,
+      "loss": 1.6737,
+      "step": 915
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.8974617246213392e-05,
+      "loss": 1.6564,
+      "step": 920
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.8955193463098017e-05,
+      "loss": 1.6926,
+      "step": 925
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.8935597564311493e-05,
+      "loss": 1.6554,
+      "step": 930
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.8915829926480103e-05,
+      "loss": 1.6878,
+      "step": 935
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.515625,
+      "learning_rate": 1.8895890929530903e-05,
+      "loss": 1.658,
+      "step": 940
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.8875780956684403e-05,
+      "loss": 1.6548,
+      "step": 945
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.546875,
+      "learning_rate": 1.8855500394447214e-05,
+      "loss": 1.698,
+      "step": 950
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.53125,
+      "learning_rate": 1.8835049632604618e-05,
+      "loss": 1.6291,
+      "step": 955
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.881442906421307e-05,
+      "loss": 1.6031,
+      "step": 960
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.8793639085592645e-05,
+      "loss": 1.6021,
+      "step": 965
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.877268009631943e-05,
+      "loss": 1.5901,
+      "step": 970
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.546875,
+      "learning_rate": 1.8751552499217825e-05,
+      "loss": 1.6181,
+      "step": 975
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.873025670035283e-05,
+      "loss": 1.6659,
+      "step": 980
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.870879310902221e-05,
+      "loss": 1.6853,
+      "step": 985
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.8687162137748647e-05,
+      "loss": 1.6478,
+      "step": 990
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.866536420227181e-05,
+      "loss": 1.647,
+      "step": 995
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.6015625,
+      "learning_rate": 1.864339972154036e-05,
+      "loss": 1.651,
+      "step": 1000
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.609375,
+      "learning_rate": 1.8621269117703887e-05,
+      "loss": 1.6311,
+      "step": 1005
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.8598972816104833e-05,
+      "loss": 1.6365,
+      "step": 1010
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.5,
+      "learning_rate": 1.8576511245270266e-05,
+      "loss": 1.6289,
+      "step": 1015
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.515625,
+      "learning_rate": 1.8553884836903695e-05,
+      "loss": 1.6246,
+      "step": 1020
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.5234375,
+      "learning_rate": 1.853109402587673e-05,
+      "loss": 1.6639,
+      "step": 1025
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.8508139250220753e-05,
+      "loss": 1.6381,
+      "step": 1030
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.5390625,
+      "learning_rate": 1.8485020951118478e-05,
+      "loss": 1.6427,
+      "step": 1035
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.5,
+      "learning_rate": 1.8461739572895497e-05,
+      "loss": 1.6531,
+      "step": 1040
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.5,
+      "learning_rate": 1.8438295563011713e-05,
+      "loss": 1.633,
+      "step": 1045
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.484375,
+      "learning_rate": 1.841468937205276e-05,
+      "loss": 1.6537,
+      "step": 1050
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.578125,
+      "learning_rate": 1.839092145372133e-05,
+      "loss": 1.6309,
+      "step": 1055
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.5,
+      "learning_rate": 1.8366992264828463e-05,
+      "loss": 1.6552,
+      "step": 1060
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.53125,
+      "learning_rate": 1.8342902265284755e-05,
+      "loss": 1.6517,
+      "step": 1065
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.8318651918091535e-05,
+      "loss": 1.6566,
+      "step": 1070
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.829424168933195e-05,
+      "loss": 1.6815,
+      "step": 1075
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.826967204816202e-05,
+      "loss": 1.5931,
+      "step": 1080
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.546875,
+      "learning_rate": 1.8244943466801615e-05,
+      "loss": 1.6596,
+      "step": 1085
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.578125,
+      "learning_rate": 1.8220056420525378e-05,
+      "loss": 1.6963,
+      "step": 1090
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.5546875,
+      "learning_rate": 1.819501138765359e-05,
+      "loss": 1.6305,
+      "step": 1095
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.8169808849542986e-05,
+      "loss": 1.6534,
+      "step": 1100
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.814444929057748e-05,
+      "loss": 1.6411,
+      "step": 1105
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.53125,
+      "learning_rate": 1.8118933198158893e-05,
+      "loss": 1.6309,
+      "step": 1110
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.8093261062697553e-05,
+      "loss": 1.5924,
+      "step": 1115
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.65625,
+      "learning_rate": 1.806743337760288e-05,
+      "loss": 1.6534,
+      "step": 1120
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.8041450639273902e-05,
+      "loss": 1.6357,
+      "step": 1125
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.8015313347089723e-05,
+      "loss": 1.6311,
+      "step": 1130
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.798902200339991e-05,
+      "loss": 1.6626,
+      "step": 1135
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.5078125,
+      "learning_rate": 1.7962577113514846e-05,
+      "loss": 1.6675,
+      "step": 1140
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.7935979185696027e-05,
+      "loss": 1.6463,
+      "step": 1145
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.7909228731146278e-05,
+      "loss": 1.612,
+      "step": 1150
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.7882326263999927e-05,
+      "loss": 1.653,
+      "step": 1155
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.625,
+      "learning_rate": 1.785527230131295e-05,
+      "loss": 1.6389,
+      "step": 1160
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.5390625,
+      "learning_rate": 1.7828067363052984e-05,
+      "loss": 1.6481,
+      "step": 1165
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.5078125,
+      "learning_rate": 1.78007119720894e-05,
+      "loss": 1.6829,
+      "step": 1170
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.7773206654183187e-05,
+      "loss": 1.5923,
+      "step": 1175
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.7745551937976897e-05,
+      "loss": 1.6356,
+      "step": 1180
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.5,
+      "learning_rate": 1.771774835498445e-05,
+      "loss": 1.6434,
+      "step": 1185
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.484375,
+      "learning_rate": 1.7689796439580942e-05,
+      "loss": 1.6394,
+      "step": 1190
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.766169672899236e-05,
+      "loss": 1.6625,
+      "step": 1195
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.7633449763285275e-05,
+      "loss": 1.6714,
+      "step": 1200
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.760505608535643e-05,
+      "loss": 1.68,
+      "step": 1205
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.7576516240922344e-05,
+      "loss": 1.6307,
+      "step": 1210
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.75478307785088e-05,
+      "loss": 1.6423,
+      "step": 1215
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.7519000249440293e-05,
+      "loss": 1.6374,
+      "step": 1220
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.7490025207829466e-05,
+      "loss": 1.6324,
+      "step": 1225
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.7460906210566435e-05,
+      "loss": 1.6392,
+      "step": 1230
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.5,
+      "learning_rate": 1.7431643817308097e-05,
+      "loss": 1.6503,
+      "step": 1235
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.546875,
+      "learning_rate": 1.740223859046736e-05,
+      "loss": 1.6207,
+      "step": 1240
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.7372691095202355e-05,
+      "loss": 1.5971,
+      "step": 1245
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.7343001899405553e-05,
+      "loss": 1.6922,
+      "step": 1250
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.515625,
+      "learning_rate": 1.7313171573692867e-05,
+      "loss": 1.6626,
+      "step": 1255
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.7283200691392674e-05,
+      "loss": 1.6359,
+      "step": 1260
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.484375,
+      "learning_rate": 1.7253089828534794e-05,
+      "loss": 1.5985,
+      "step": 1265
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.7222839563839432e-05,
+      "loss": 1.6303,
+      "step": 1270
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.7192450478706045e-05,
+      "loss": 1.6414,
+      "step": 1275
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.7161923157202168e-05,
+      "loss": 1.6167,
+      "step": 1280
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.7131258186052195e-05,
+      "loss": 1.6023,
+      "step": 1285
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.710045615462609e-05,
+      "loss": 1.6223,
+      "step": 1290
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.7069517654928073e-05,
+      "loss": 1.6865,
+      "step": 1295
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.7038443281585238e-05,
+      "loss": 1.6107,
+      "step": 1300
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.5078125,
+      "learning_rate": 1.7007233631836115e-05,
+      "loss": 1.6117,
+      "step": 1305
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.697588930551921e-05,
+      "loss": 1.6448,
+      "step": 1310
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.6944410905061454e-05,
+      "loss": 1.6116,
+      "step": 1315
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.691279903546664e-05,
+      "loss": 1.6388,
+      "step": 1320
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.5234375,
+      "learning_rate": 1.68810543043038e-05,
+      "loss": 1.6123,
+      "step": 1325
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.6849177321695505e-05,
+      "loss": 1.6201,
+      "step": 1330
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.6817168700306154e-05,
+      "loss": 1.6428,
+      "step": 1335
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.678502905533021e-05,
+      "loss": 1.6362,
+      "step": 1340
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.515625,
+      "learning_rate": 1.675275900448035e-05,
+      "loss": 1.6151,
+      "step": 1345
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.672035916797561e-05,
+      "loss": 1.6055,
+      "step": 1350
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.5,
+      "learning_rate": 1.6687830168529458e-05,
+      "loss": 1.6513,
+      "step": 1355
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.515625,
+      "learning_rate": 1.6655172631337836e-05,
+      "loss": 1.5987,
+      "step": 1360
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.6622387184067134e-05,
+      "loss": 1.5954,
+      "step": 1365
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.6589474456842117e-05,
+      "loss": 1.6548,
+      "step": 1370
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.6556435082233843e-05,
+      "loss": 1.5937,
+      "step": 1375
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.6523269695247482e-05,
+      "loss": 1.6298,
+      "step": 1380
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.648997893331011e-05,
+      "loss": 1.6436,
+      "step": 1385
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.5078125,
+      "learning_rate": 1.6456563436258475e-05,
+      "loss": 1.647,
+      "step": 1390
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.6423023846326686e-05,
+      "loss": 1.6254,
+      "step": 1395
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.5078125,
+      "learning_rate": 1.6389360808133877e-05,
+      "loss": 1.6002,
+      "step": 1400
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.6355574968671807e-05,
+      "loss": 1.6898,
+      "step": 1405
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.5,
+      "learning_rate": 1.6321666977292442e-05,
+      "loss": 1.6357,
+      "step": 1410
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.5546875,
+      "learning_rate": 1.6287637485695452e-05,
+      "loss": 1.6306,
+      "step": 1415
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.6253487147915718e-05,
+      "loss": 1.6559,
+      "step": 1420
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.578125,
+      "learning_rate": 1.6219216620310717e-05,
+      "loss": 1.5824,
+      "step": 1425
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.484375,
+      "learning_rate": 1.6184826561547955e-05,
+      "loss": 1.6891,
+      "step": 1430
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.5,
+      "learning_rate": 1.6150317632592272e-05,
+      "loss": 1.604,
+      "step": 1435
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.6115690496693156e-05,
+      "loss": 1.6402,
+      "step": 1440
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.608094581937199e-05,
+      "loss": 1.5887,
+      "step": 1445
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.6046084268409255e-05,
+      "loss": 1.6338,
+      "step": 1450
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.601110651383172e-05,
+      "loss": 1.6046,
+      "step": 1455
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.5976013227899527e-05,
+      "loss": 1.6425,
+      "step": 1460
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.59408050850933e-05,
+      "loss": 1.6547,
+      "step": 1465
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.5905482762101173e-05,
+      "loss": 1.6704,
+      "step": 1470
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.587004693780578e-05,
+      "loss": 1.6419,
+      "step": 1475
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.546875,
+      "learning_rate": 1.5834498293271202e-05,
+      "loss": 1.6424,
+      "step": 1480
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.5798837511729907e-05,
+      "loss": 1.6314,
+      "step": 1485
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.5763065278569572e-05,
+      "loss": 1.6517,
+      "step": 1490
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.572718228131996e-05,
+      "loss": 1.5977,
+      "step": 1495
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.569118920963965e-05,
+      "loss": 1.6342,
+      "step": 1500
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.565508675530284e-05,
+      "loss": 1.6403,
+      "step": 1505
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.561887561218602e-05,
+      "loss": 1.5749,
+      "step": 1510
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.5582556476254627e-05,
+      "loss": 1.6235,
+      "step": 1515
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.5546130045549687e-05,
+      "loss": 1.635,
+      "step": 1520
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.5509597020174404e-05,
+      "loss": 1.6273,
+      "step": 1525
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.5472958102280683e-05,
+      "loss": 1.6525,
+      "step": 1530
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.515625,
+      "learning_rate": 1.543621399605565e-05,
+      "loss": 1.6509,
+      "step": 1535
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.5399365407708113e-05,
+      "loss": 1.6284,
+      "step": 1540
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.484375,
+      "learning_rate": 1.536241304545499e-05,
+      "loss": 1.594,
+      "step": 1545
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.53253576195077e-05,
+      "loss": 1.6252,
+      "step": 1550
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.5,
+      "learning_rate": 1.5288199842058497e-05,
+      "loss": 1.6424,
+      "step": 1555
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.484375,
+      "learning_rate": 1.5250940427266818e-05,
+      "loss": 1.6434,
+      "step": 1560
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.5213580091245513e-05,
+      "loss": 1.6102,
+      "step": 1565
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.517611955204711e-05,
+      "loss": 1.6414,
+      "step": 1570
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.375,
+      "learning_rate": 1.5138559529650014e-05,
+      "loss": 1.6163,
+      "step": 1575
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.5100900745944649e-05,
+      "loss": 1.6522,
+      "step": 1580
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.5063143924719603e-05,
+      "loss": 1.621,
+      "step": 1585
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.5025289791647714e-05,
+      "loss": 1.6983,
+      "step": 1590
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.4987339074272116e-05,
+      "loss": 1.6299,
+      "step": 1595
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.494929250199226e-05,
+      "loss": 1.605,
+      "step": 1600
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.4911150806049887e-05,
+      "loss": 1.6289,
+      "step": 1605
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.4872914719514998e-05,
+      "loss": 1.6703,
+      "step": 1610
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.4834584977271735e-05,
+      "loss": 1.6258,
+      "step": 1615
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.479616231600427e-05,
+      "loss": 1.615,
+      "step": 1620
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.4757647474182657e-05,
+      "loss": 1.6475,
+      "step": 1625
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.4719041192048613e-05,
+      "loss": 1.6328,
+      "step": 1630
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.4680344211601313e-05,
+      "loss": 1.6485,
+      "step": 1635
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.4641557276583126e-05,
+      "loss": 1.6417,
+      "step": 1640
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.460268113246531e-05,
+      "loss": 1.6082,
+      "step": 1645
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.456371652643369e-05,
+      "loss": 1.6131,
+      "step": 1650
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.3671875,
+      "learning_rate": 1.4524664207374306e-05,
+      "loss": 1.632,
+      "step": 1655
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.4485524925858998e-05,
+      "loss": 1.6555,
+      "step": 1660
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.4446299434131011e-05,
+      "loss": 1.6406,
+      "step": 1665
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.44069884860905e-05,
+      "loss": 1.6423,
+      "step": 1670
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.4367592837280077e-05,
+      "loss": 1.6825,
+      "step": 1675
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.4328113244870267e-05,
+      "loss": 1.6249,
+      "step": 1680
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.59375,
+      "learning_rate": 1.4288550467644961e-05,
+      "loss": 1.659,
+      "step": 1685
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.4248905265986835e-05,
+      "loss": 1.6215,
+      "step": 1690
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.4209178401862724e-05,
+      "loss": 1.6452,
+      "step": 1695
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.375,
+      "learning_rate": 1.4169370638809004e-05,
+      "loss": 1.5952,
+      "step": 1700
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.4129482741916889e-05,
+      "loss": 1.5953,
+      "step": 1705
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.4089515477817737e-05,
+      "loss": 1.5673,
+      "step": 1710
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.4049469614668318e-05,
+      "loss": 1.644,
+      "step": 1715
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.4009345922136048e-05,
+      "loss": 1.6232,
+      "step": 1720
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.3969145171384197e-05,
+      "loss": 1.6291,
+      "step": 1725
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.392886813505707e-05,
+      "loss": 1.652,
+      "step": 1730
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.3888515587265144e-05,
+      "loss": 1.628,
+      "step": 1735
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.3848088303570209e-05,
+      "loss": 1.628,
+      "step": 1740
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.3807587060970449e-05,
+      "loss": 1.6372,
+      "step": 1745
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.3767012637885513e-05,
+      "loss": 1.5894,
+      "step": 1750
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.3726365814141554e-05,
+      "loss": 1.631,
+      "step": 1755
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.3685647370956241e-05,
+      "loss": 1.5791,
+      "step": 1760
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.3644858090923741e-05,
+      "loss": 1.6218,
+      "step": 1765
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.375,
+      "learning_rate": 1.3603998757999676e-05,
+      "loss": 1.5844,
+      "step": 1770
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.3563070157486076e-05,
+      "loss": 1.6515,
+      "step": 1775
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.3522073076016247e-05,
+      "loss": 1.6189,
+      "step": 1780
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.3481008301539698e-05,
+      "loss": 1.6147,
+      "step": 1785
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.3439876623306956e-05,
+      "loss": 1.6187,
+      "step": 1790
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.3398678831854417e-05,
+      "loss": 1.6313,
+      "step": 1795
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.3357415718989163e-05,
+      "loss": 1.5335,
+      "step": 1800
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.3316088077773722e-05,
+      "loss": 1.621,
+      "step": 1805
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.3274696702510825e-05,
+      "loss": 1.6076,
+      "step": 1810
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.3233242388728167e-05,
+      "loss": 1.6324,
+      "step": 1815
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.3191725933163087e-05,
+      "loss": 1.6292,
+      "step": 1820
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.3150148133747269e-05,
+      "loss": 1.577,
+      "step": 1825
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.3108509789591407e-05,
+      "loss": 1.6413,
+      "step": 1830
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.3066811700969844e-05,
+      "loss": 1.6291,
+      "step": 1835
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.5,
+      "learning_rate": 1.3025054669305187e-05,
+      "loss": 1.6541,
+      "step": 1840
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.2983239497152905e-05,
+      "loss": 1.6797,
+      "step": 1845
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.2941366988185919e-05,
+      "loss": 1.5779,
+      "step": 1850
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.546875,
+      "learning_rate": 1.2899437947179135e-05,
+      "loss": 1.6196,
+      "step": 1855
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.285745317999398e-05,
+      "loss": 1.6481,
+      "step": 1860
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.2815413493562927e-05,
+      "loss": 1.6075,
+      "step": 1865
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.2773319695873977e-05,
+      "loss": 1.6253,
+      "step": 1870
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.3671875,
+      "learning_rate": 1.2731172595955125e-05,
+      "loss": 1.6332,
+      "step": 1875
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.2688973003858824e-05,
+      "loss": 1.6095,
+      "step": 1880
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.2646721730646398e-05,
+      "loss": 1.6107,
+      "step": 1885
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.260441958837247e-05,
+      "loss": 1.611,
+      "step": 1890
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.2562067390069353e-05,
+      "loss": 1.617,
+      "step": 1895
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.2519665949731405e-05,
+      "loss": 1.6319,
+      "step": 1900
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.2477216082299419e-05,
+      "loss": 1.6473,
+      "step": 1905
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.243471860364492e-05,
+      "loss": 1.6115,
+      "step": 1910
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.515625,
+      "learning_rate": 1.2392174330554515e-05,
+      "loss": 1.6295,
+      "step": 1915
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.234958408071418e-05,
+      "loss": 1.6129,
+      "step": 1920
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.2306948672693555e-05,
+      "loss": 1.6285,
+      "step": 1925
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.226426892593019e-05,
+      "loss": 1.6052,
+      "step": 1930
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.2221545660713825e-05,
+      "loss": 1.6283,
+      "step": 1935
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.515625,
+      "learning_rate": 1.2178779698170597e-05,
+      "loss": 1.6261,
+      "step": 1940
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.53125,
+      "learning_rate": 1.2135971860247282e-05,
+      "loss": 1.6627,
+      "step": 1945
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.2093122969695476e-05,
+      "loss": 1.6299,
+      "step": 1950
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.484375,
+      "learning_rate": 1.2050233850055798e-05,
+      "loss": 1.6575,
+      "step": 1955
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.2007305325642055e-05,
+      "loss": 1.6312,
+      "step": 1960
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.1964338221525401e-05,
+      "loss": 1.6781,
+      "step": 1965
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.1921333363518474e-05,
+      "loss": 1.6375,
+      "step": 1970
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.1878291578159536e-05,
+      "loss": 1.6452,
+      "step": 1975
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.183521369269657e-05,
+      "loss": 1.6708,
+      "step": 1980
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.1792100535071396e-05,
+      "loss": 1.6155,
+      "step": 1985
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.1748952933903754e-05,
+      "loss": 1.6023,
+      "step": 1990
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.1705771718475372e-05,
+      "loss": 1.6147,
+      "step": 1995
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.1662557718714039e-05,
+      "loss": 1.6652,
+      "step": 2000
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.1619311765177639e-05,
+      "loss": 1.6058,
+      "step": 2005
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.1576034689038201e-05,
+      "loss": 1.5889,
+      "step": 2010
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.1532727322065919e-05,
+      "loss": 1.626,
+      "step": 2015
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.1489390496613174e-05,
+      "loss": 1.5881,
+      "step": 2020
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.5234375,
+      "learning_rate": 1.144602504559851e-05,
+      "loss": 1.6013,
+      "step": 2025
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.1402631802490662e-05,
+      "loss": 1.6822,
+      "step": 2030
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.1359211601292507e-05,
+      "loss": 1.6418,
+      "step": 2035
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.5390625,
+      "learning_rate": 1.1315765276525061e-05,
+      "loss": 1.6366,
+      "step": 2040
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.1272293663211412e-05,
+      "loss": 1.6232,
+      "step": 2045
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.1228797596860686e-05,
+      "loss": 1.6611,
+      "step": 2050
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.1185277913451997e-05,
+      "loss": 1.6388,
+      "step": 2055
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.1141735449418368e-05,
+      "loss": 1.5679,
+      "step": 2060
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.109817104163065e-05,
+      "loss": 1.6797,
+      "step": 2065
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.1054585527381453e-05,
+      "loss": 1.6294,
+      "step": 2070
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.101097974436904e-05,
+      "loss": 1.6137,
+      "step": 2075
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.0967354530681243e-05,
+      "loss": 1.6225,
+      "step": 2080
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.0923710724779331e-05,
+      "loss": 1.6253,
+      "step": 2085
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.0880049165481916e-05,
+      "loss": 1.6296,
+      "step": 2090
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.0836370691948826e-05,
+      "loss": 1.6128,
+      "step": 2095
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.079267614366497e-05,
+      "loss": 1.651,
+      "step": 2100
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.0748966360424212e-05,
+      "loss": 1.6071,
+      "step": 2105
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.0705242182313222e-05,
+      "loss": 1.6355,
+      "step": 2110
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.3671875,
+      "learning_rate": 1.0661504449695337e-05,
+      "loss": 1.6353,
+      "step": 2115
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.0617754003194406e-05,
+      "loss": 1.6562,
+      "step": 2120
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.484375,
+      "learning_rate": 1.057399168367864e-05,
+      "loss": 1.599,
+      "step": 2125
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.0530218332244435e-05,
+      "loss": 1.6245,
+      "step": 2130
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.0486434790200226e-05,
+      "loss": 1.6382,
+      "step": 2135
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.04426418990503e-05,
+      "loss": 1.6523,
+      "step": 2140
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.0398840500478635e-05,
+      "loss": 1.6372,
+      "step": 2145
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.0355031436332724e-05,
+      "loss": 1.6227,
+      "step": 2150
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.5,
+      "learning_rate": 1.0311215548607373e-05,
+      "loss": 1.6283,
+      "step": 2155
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.375,
+      "learning_rate": 1.0267393679428554e-05,
+      "loss": 1.6355,
+      "step": 2160
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.022356667103719e-05,
+      "loss": 1.5956,
+      "step": 2165
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.0179735365772978e-05,
+      "loss": 1.6589,
+      "step": 2170
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.01359006060582e-05,
+      "loss": 1.6513,
+      "step": 2175
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.0092063234381536e-05,
+      "loss": 1.668,
+      "step": 2180
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.0048224093281862e-05,
+      "loss": 1.6202,
+      "step": 2185
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.0004384025332063e-05,
+      "loss": 1.6059,
+      "step": 2190
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.5234375,
+      "learning_rate": 9.960543873122836e-06,
+      "loss": 1.6166,
+      "step": 2195
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.421875,
+      "learning_rate": 9.916704479246505e-06,
+      "loss": 1.6034,
+      "step": 2200
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.4296875,
+      "learning_rate": 9.872866686280809e-06,
+      "loss": 1.629,
+      "step": 2205
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.4453125,
+      "learning_rate": 9.829031336772719e-06,
+      "loss": 1.6198,
+      "step": 2210
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.421875,
+      "learning_rate": 9.785199273222254e-06,
+      "loss": 1.6402,
+      "step": 2215
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.40625,
+      "learning_rate": 9.74137133806627e-06,
+      "loss": 1.6183,
+      "step": 2220
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.4296875,
+      "learning_rate": 9.697548373662272e-06,
+      "loss": 1.6332,
+      "step": 2225
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.3984375,
+      "learning_rate": 9.653731222272242e-06,
+      "loss": 1.6507,
+      "step": 2230
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.4609375,
+      "learning_rate": 9.609920726046424e-06,
+      "loss": 1.6473,
+      "step": 2235
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.4140625,
+      "learning_rate": 9.566117727007165e-06,
+      "loss": 1.5653,
+      "step": 2240
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.4609375,
+      "learning_rate": 9.52232306703271e-06,
+      "loss": 1.6348,
+      "step": 2245
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.5,
+      "learning_rate": 9.478537587841028e-06,
+      "loss": 1.6096,
+      "step": 2250
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.4375,
+      "learning_rate": 9.434762130973646e-06,
+      "loss": 1.6153,
+      "step": 2255
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.359375,
+      "learning_rate": 9.390997537779462e-06,
+      "loss": 1.6025,
+      "step": 2260
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.4609375,
+      "learning_rate": 9.347244649398573e-06,
+      "loss": 1.64,
+      "step": 2265
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.3984375,
+      "learning_rate": 9.30350430674612e-06,
+      "loss": 1.621,
+      "step": 2270
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.453125,
+      "learning_rate": 9.259777350496116e-06,
+      "loss": 1.5929,
+      "step": 2275
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.4140625,
+      "learning_rate": 9.21606462106529e-06,
+      "loss": 1.5973,
+      "step": 2280
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.421875,
+      "learning_rate": 9.172366958596943e-06,
+      "loss": 1.6234,
+      "step": 2285
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.390625,
+      "learning_rate": 9.128685202944786e-06,
+      "loss": 1.5659,
+      "step": 2290
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.421875,
+      "learning_rate": 9.085020193656817e-06,
+      "loss": 1.6311,
+      "step": 2295
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.4609375,
+      "learning_rate": 9.041372769959166e-06,
+      "loss": 1.6372,
+      "step": 2300
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.4375,
+      "learning_rate": 8.997743770739974e-06,
+      "loss": 1.6044,
+      "step": 2305
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.4296875,
+      "learning_rate": 8.95413403453328e-06,
+      "loss": 1.6755,
+      "step": 2310
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.390625,
+      "learning_rate": 8.91054439950288e-06,
+      "loss": 1.64,
+      "step": 2315
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.453125,
+      "learning_rate": 8.866975703426244e-06,
+      "loss": 1.6352,
+      "step": 2320
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.4375,
+      "learning_rate": 8.823428783678402e-06,
+      "loss": 1.6373,
+      "step": 2325
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.421875,
+      "learning_rate": 8.77990447721584e-06,
+      "loss": 1.6642,
+      "step": 2330
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.390625,
+      "learning_rate": 8.73640362056044e-06,
+      "loss": 1.6322,
+      "step": 2335
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.4453125,
+      "learning_rate": 8.692927049783372e-06,
+      "loss": 1.6259,
+      "step": 2340
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.4140625,
+      "learning_rate": 8.649475600489052e-06,
+      "loss": 1.5984,
+      "step": 2345
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.4375,
+      "learning_rate": 8.606050107799063e-06,
+      "loss": 1.6185,
+      "step": 2350
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.328125,
+      "learning_rate": 8.56265140633611e-06,
+      "loss": 1.6013,
+      "step": 2355
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.4609375,
+      "learning_rate": 8.51928033020799e-06,
+      "loss": 1.6586,
+      "step": 2360
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.4453125,
+      "learning_rate": 8.475937712991534e-06,
+      "loss": 1.6162,
+      "step": 2365
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.4296875,
+      "learning_rate": 8.432624387716625e-06,
+      "loss": 1.6052,
+      "step": 2370
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.421875,
+      "learning_rate": 8.389341186850144e-06,
+      "loss": 1.6685,
+      "step": 2375
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.3828125,
+      "learning_rate": 8.346088942280003e-06,
+      "loss": 1.5978,
+      "step": 2380
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.46875,
+      "learning_rate": 8.30286848529914e-06,
+      "loss": 1.6138,
+      "step": 2385
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.4609375,
+      "learning_rate": 8.259680646589553e-06,
+      "loss": 1.645,
+      "step": 2390
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.375,
+      "learning_rate": 8.216526256206322e-06,
+      "loss": 1.6061,
+      "step": 2395
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.5078125,
+      "learning_rate": 8.173406143561661e-06,
+      "loss": 1.7134,
+      "step": 2400
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.53125,
+      "learning_rate": 8.130321137408985e-06,
+      "loss": 1.625,
+      "step": 2405
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.3671875,
+      "learning_rate": 8.087272065826967e-06,
+      "loss": 1.6675,
+      "step": 2410
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.4296875,
+      "learning_rate": 8.044259756203639e-06,
+      "loss": 1.6193,
+      "step": 2415
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.3671875,
+      "learning_rate": 8.001285035220472e-06,
+      "loss": 1.577,
+      "step": 2420
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.4140625,
+      "learning_rate": 7.958348728836499e-06,
+      "loss": 1.6533,
+      "step": 2425
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.3984375,
+      "learning_rate": 7.915451662272442e-06,
+      "loss": 1.6036,
+      "step": 2430
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.390625,
+      "learning_rate": 7.872594659994842e-06,
+      "loss": 1.5943,
+      "step": 2435
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.4453125,
+      "learning_rate": 7.829778545700225e-06,
+      "loss": 1.6511,
+      "step": 2440
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.390625,
+      "learning_rate": 7.787004142299258e-06,
+      "loss": 1.6339,
+      "step": 2445
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.4296875,
+      "learning_rate": 7.744272271900939e-06,
+      "loss": 1.6608,
+      "step": 2450
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.53125,
+      "learning_rate": 7.701583755796804e-06,
+      "loss": 1.6451,
+      "step": 2455
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.4375,
+      "learning_rate": 7.658939414445127e-06,
+      "loss": 1.6154,
+      "step": 2460
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.421875,
+      "learning_rate": 7.616340067455165e-06,
+      "loss": 1.6226,
+      "step": 2465
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.4453125,
+      "learning_rate": 7.57378653357139e-06,
+      "loss": 1.6419,
+      "step": 2470
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.390625,
+      "learning_rate": 7.531279630657776e-06,
+      "loss": 1.6246,
+      "step": 2475
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.4140625,
+      "learning_rate": 7.488820175682054e-06,
+      "loss": 1.6596,
+      "step": 2480
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.40625,
+      "learning_rate": 7.446408984700029e-06,
+      "loss": 1.6198,
+      "step": 2485
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.3984375,
+      "learning_rate": 7.404046872839886e-06,
+      "loss": 1.6242,
+      "step": 2490
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.4296875,
+      "learning_rate": 7.361734654286524e-06,
+      "loss": 1.6338,
+      "step": 2495
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.421875,
+      "learning_rate": 7.3194731422659185e-06,
+      "loss": 1.5629,
+      "step": 2500
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.453125,
+      "learning_rate": 7.277263149029472e-06,
+      "loss": 1.5845,
+      "step": 2505
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.4765625,
+      "learning_rate": 7.235105485838427e-06,
+      "loss": 1.6297,
+      "step": 2510
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.4375,
+      "learning_rate": 7.193000962948252e-06,
+      "loss": 1.6451,
+      "step": 2515
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.40625,
+      "learning_rate": 7.1509503895930765e-06,
+      "loss": 1.6632,
+      "step": 2520
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.4140625,
+      "learning_rate": 7.108954573970146e-06,
+      "loss": 1.612,
+      "step": 2525
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.421875,
+      "learning_rate": 7.067014323224273e-06,
+      "loss": 1.6612,
+      "step": 2530
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.4921875,
+      "learning_rate": 7.0251304434323445e-06,
+      "loss": 1.6252,
+      "step": 2535
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.4296875,
+      "learning_rate": 6.983303739587806e-06,
+      "loss": 1.6268,
+      "step": 2540
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.375,
+      "learning_rate": 6.941535015585203e-06,
+      "loss": 1.566,
+      "step": 2545
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.46875,
+      "learning_rate": 6.899825074204739e-06,
+      "loss": 1.6449,
+      "step": 2550
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.453125,
+      "learning_rate": 6.858174717096821e-06,
+      "loss": 1.5949,
+      "step": 2555
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.40625,
+      "learning_rate": 6.816584744766683e-06,
+      "loss": 1.623,
+      "step": 2560
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.421875,
+      "learning_rate": 6.775055956558971e-06,
+      "loss": 1.6131,
+      "step": 2565
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.421875,
+      "learning_rate": 6.7335891506424e-06,
+      "loss": 1.6714,
+      "step": 2570
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.4453125,
+      "learning_rate": 6.69218512399441e-06,
+      "loss": 1.6585,
+      "step": 2575
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.40625,
+      "learning_rate": 6.6508446723858355e-06,
+      "loss": 1.6033,
+      "step": 2580
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.40625,
+      "learning_rate": 6.609568590365636e-06,
+      "loss": 1.6234,
+      "step": 2585
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.4296875,
+      "learning_rate": 6.568357671245598e-06,
+      "loss": 1.6487,
+      "step": 2590
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.3984375,
+      "learning_rate": 6.5272127070851035e-06,
+      "loss": 1.6592,
+      "step": 2595
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.4140625,
+      "learning_rate": 6.486134488675904e-06,
+      "loss": 1.6651,
+      "step": 2600
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.4140625,
+      "learning_rate": 6.445123805526922e-06,
+      "loss": 1.6068,
+      "step": 2605
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.4296875,
+      "learning_rate": 6.404181445849073e-06,
+      "loss": 1.6203,
+      "step": 2610
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.453125,
+      "learning_rate": 6.363308196540121e-06,
+      "loss": 1.6255,
+      "step": 2615
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.390625,
+      "learning_rate": 6.3225048431695564e-06,
+      "loss": 1.6608,
+      "step": 2620
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.421875,
+      "learning_rate": 6.281772169963488e-06,
+      "loss": 1.5833,
+      "step": 2625
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.40625,
+      "learning_rate": 6.241110959789583e-06,
+      "loss": 1.6541,
+      "step": 2630
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.4375,
+      "learning_rate": 6.200521994142006e-06,
+      "loss": 1.6032,
+      "step": 2635
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.375,
+      "learning_rate": 6.160006053126415e-06,
+      "loss": 1.6192,
+      "step": 2640
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.4296875,
+      "learning_rate": 6.1195639154449594e-06,
+      "loss": 1.5958,
+      "step": 2645
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.40625,
+      "learning_rate": 6.079196358381307e-06,
+      "loss": 1.6035,
+      "step": 2650
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.421875,
+      "learning_rate": 6.038904157785725e-06,
+      "loss": 1.5928,
+      "step": 2655
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.3671875,
+      "learning_rate": 5.998688088060144e-06,
+      "loss": 1.6468,
+      "step": 2660
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.484375,
+      "learning_rate": 5.958548922143289e-06,
+      "loss": 1.6246,
+      "step": 2665
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.4609375,
+      "learning_rate": 5.918487431495828e-06,
+      "loss": 1.6489,
+      "step": 2670
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.390625,
+      "learning_rate": 5.878504386085526e-06,
+      "loss": 1.648,
+      "step": 2675
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.40625,
+      "learning_rate": 5.838600554372469e-06,
+      "loss": 1.5877,
+      "step": 2680
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.4609375,
+      "learning_rate": 5.798776703294273e-06,
+      "loss": 1.6025,
+      "step": 2685
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.421875,
+      "learning_rate": 5.759033598251367e-06,
+      "loss": 1.6379,
+      "step": 2690
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.4375,
+      "learning_rate": 5.7193720030922605e-06,
+      "loss": 1.6229,
+      "step": 2695
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.421875,
+      "learning_rate": 5.679792680098869e-06,
+      "loss": 1.6149,
+      "step": 2700
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.4296875,
+      "learning_rate": 5.6402963899718775e-06,
+      "loss": 1.6051,
+      "step": 2705
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.4765625,
+      "learning_rate": 5.6008838918161e-06,
+      "loss": 1.5936,
+      "step": 2710
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.421875,
+      "learning_rate": 5.561555943125898e-06,
+      "loss": 1.5987,
+      "step": 2715
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.484375,
+      "learning_rate": 5.522313299770624e-06,
+      "loss": 1.6739,
+      "step": 2720
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.375,
+      "learning_rate": 5.483156715980089e-06,
+      "loss": 1.5944,
+      "step": 2725
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.40625,
+      "learning_rate": 5.4440869443300804e-06,
+      "loss": 1.5757,
+      "step": 2730
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.4453125,
+      "learning_rate": 5.405104735727873e-06,
+      "loss": 1.5867,
+      "step": 2735
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.3671875,
+      "learning_rate": 5.366210839397816e-06,
+      "loss": 1.5878,
+      "step": 2740
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.4609375,
+      "learning_rate": 5.327406002866932e-06,
+      "loss": 1.6447,
+      "step": 2745
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.3671875,
+      "learning_rate": 5.288690971950536e-06,
+      "loss": 1.5847,
+      "step": 2750
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.46875,
+      "learning_rate": 5.2500664907379174e-06,
+      "loss": 1.5775,
+      "step": 2755
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.359375,
+      "learning_rate": 5.211533301578036e-06,
+      "loss": 1.6567,
+      "step": 2760
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.34375,
+      "learning_rate": 5.1730921450652404e-06,
+      "loss": 1.6168,
+      "step": 2765
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.4921875,
+      "learning_rate": 5.1347437600250515e-06,
+      "loss": 1.6286,
+      "step": 2770
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.375,
+      "learning_rate": 5.0964888834999526e-06,
+      "loss": 1.6167,
+      "step": 2775
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.4609375,
+      "learning_rate": 5.0583282507352205e-06,
+      "loss": 1.6098,
+      "step": 2780
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.421875,
+      "learning_rate": 5.020262595164814e-06,
+      "loss": 1.5851,
+      "step": 2785
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.4453125,
+      "learning_rate": 4.982292648397252e-06,
+      "loss": 1.6563,
+      "step": 2790
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.3984375,
+      "learning_rate": 4.944419140201564e-06,
+      "loss": 1.6066,
+      "step": 2795
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.484375,
+      "learning_rate": 4.906642798493275e-06,
+      "loss": 1.6134,
+      "step": 2800
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.4609375,
+      "learning_rate": 4.868964349320391e-06,
+      "loss": 1.5946,
+      "step": 2805
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.421875,
+      "learning_rate": 4.831384516849461e-06,
+      "loss": 1.6384,
+      "step": 2810
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.4296875,
+      "learning_rate": 4.793904023351665e-06,
+      "loss": 1.6232,
+      "step": 2815
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.375,
+      "learning_rate": 4.756523589188913e-06,
+      "loss": 1.5959,
+      "step": 2820
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.3984375,
+      "learning_rate": 4.719243932800012e-06,
+      "loss": 1.6354,
+      "step": 2825
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.4765625,
+      "learning_rate": 4.682065770686855e-06,
+      "loss": 1.6081,
+      "step": 2830
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.4296875,
+      "learning_rate": 4.644989817400651e-06,
+      "loss": 1.6048,
+      "step": 2835
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.46875,
+      "learning_rate": 4.608016785528198e-06,
+      "loss": 1.6491,
+      "step": 2840
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.3828125,
+      "learning_rate": 4.5711473856781706e-06,
+      "loss": 1.5842,
+      "step": 2845
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.4453125,
+      "learning_rate": 4.534382326467477e-06,
+      "loss": 1.6355,
+      "step": 2850
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.5078125,
+      "learning_rate": 4.497722314507636e-06,
+      "loss": 1.6665,
+      "step": 2855
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.4375,
+      "learning_rate": 4.461168054391187e-06,
+      "loss": 1.6266,
+      "step": 2860
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.40625,
+      "learning_rate": 4.4247202486781735e-06,
+      "loss": 1.5948,
+      "step": 2865
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.375,
+      "learning_rate": 4.388379597882606e-06,
+      "loss": 1.5851,
+      "step": 2870
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.4296875,
+      "learning_rate": 4.352146800459023e-06,
+      "loss": 1.5952,
+      "step": 2875
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.375,
+      "learning_rate": 4.3160225527890556e-06,
+      "loss": 1.6181,
+      "step": 2880
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.3984375,
+      "learning_rate": 4.280007549168046e-06,
+      "loss": 1.64,
+      "step": 2885
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.453125,
+      "learning_rate": 4.244102481791713e-06,
+      "loss": 1.6083,
+      "step": 2890
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.46875,
+      "learning_rate": 4.208308040742831e-06,
+      "loss": 1.6524,
+      "step": 2895
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.421875,
+      "learning_rate": 4.172624913977973e-06,
+      "loss": 1.6263,
+      "step": 2900
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.3359375,
+      "learning_rate": 4.1370537873143025e-06,
+      "loss": 1.6471,
+      "step": 2905
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.4765625,
+      "learning_rate": 4.101595344416368e-06,
+      "loss": 1.6364,
+      "step": 2910
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.4140625,
+      "learning_rate": 4.066250266782979e-06,
+      "loss": 1.6406,
+      "step": 2915
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.4609375,
+      "learning_rate": 4.0310192337341106e-06,
+      "loss": 1.6156,
+      "step": 2920
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.46875,
+      "learning_rate": 3.995902922397834e-06,
+      "loss": 1.65,
+      "step": 2925
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.421875,
+      "learning_rate": 3.960902007697314e-06,
+      "loss": 1.6574,
+      "step": 2930
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.4140625,
+      "learning_rate": 3.926017162337826e-06,
+      "loss": 1.6318,
+      "step": 2935
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.3984375,
+      "learning_rate": 3.891249056793839e-06,
+      "loss": 1.625,
+      "step": 2940
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.4296875,
+      "learning_rate": 3.856598359296126e-06,
+      "loss": 1.6285,
+      "step": 2945
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.421875,
+      "learning_rate": 3.822065735818914e-06,
+      "loss": 1.6158,
+      "step": 2950
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.3984375,
+      "learning_rate": 3.78765185006709e-06,
+      "loss": 1.6165,
+      "step": 2955
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.4140625,
+      "learning_rate": 3.753357363463442e-06,
+      "loss": 1.584,
+      "step": 2960
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.4375,
+      "learning_rate": 3.7191829351359487e-06,
+      "loss": 1.6093,
+      "step": 2965
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.685129221905117e-06,
+      "loss": 1.6096,
+      "step": 2970
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.4296875,
+      "learning_rate": 3.6511968782713468e-06,
+      "loss": 1.6229,
+      "step": 2975
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.375,
+      "learning_rate": 3.6173865564023557e-06,
+      "loss": 1.6342,
+      "step": 2980
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.4296875,
+      "learning_rate": 3.5836989061206473e-06,
+      "loss": 1.6376,
+      "step": 2985
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.421875,
+      "learning_rate": 3.5501345748910223e-06,
+      "loss": 1.6727,
+      "step": 2990
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.390625,
+      "learning_rate": 3.5166942078081266e-06,
+      "loss": 1.6045,
+      "step": 2995
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.4296875,
+      "learning_rate": 3.4833784475840705e-06,
+      "loss": 1.6253,
+      "step": 3000
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.3984375,
+      "learning_rate": 3.4501879345360534e-06,
+      "loss": 1.6315,
+      "step": 3005
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.3828125,
+      "learning_rate": 3.4171233065740684e-06,
+      "loss": 1.6292,
+      "step": 3010
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.384185199188652e-06,
+      "loss": 1.6187,
+      "step": 3015
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.3671875,
+      "learning_rate": 3.351374245438649e-06,
+      "loss": 1.6295,
+      "step": 3020
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.3671875,
+      "learning_rate": 3.3186910759390546e-06,
+      "loss": 1.5787,
+      "step": 3025
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.5,
+      "learning_rate": 3.2861363188489083e-06,
+      "loss": 1.684,
+      "step": 3030
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.453125,
+      "learning_rate": 3.2537105998591966e-06,
+      "loss": 1.6018,
+      "step": 3035
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.4375,
+      "learning_rate": 3.221414542180844e-06,
+      "loss": 1.6337,
+      "step": 3040
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.1892487665327265e-06,
+      "loss": 1.5943,
+      "step": 3045
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.3984375,
+      "learning_rate": 3.157213891129747e-06,
+      "loss": 1.601,
+      "step": 3050
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.4296875,
+      "learning_rate": 3.1253105316709574e-06,
+      "loss": 1.6267,
+      "step": 3055
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.421875,
+      "learning_rate": 3.0935393013277127e-06,
+      "loss": 1.6177,
+      "step": 3060
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.4375,
+      "learning_rate": 3.0619008107318924e-06,
+      "loss": 1.6532,
+      "step": 3065
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.390625,
+      "learning_rate": 3.0303956679641677e-06,
+      "loss": 1.6221,
+      "step": 3070
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2.9990244785423074e-06,
+      "loss": 1.6616,
+      "step": 3075
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.359375,
+      "learning_rate": 2.967787845409553e-06,
+      "loss": 1.6482,
+      "step": 3080
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2.936686368923013e-06,
+      "loss": 1.6737,
+      "step": 3085
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.375,
+      "learning_rate": 2.905720646842135e-06,
+      "loss": 1.6354,
+      "step": 3090
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.40625,
+      "learning_rate": 2.8748912743172152e-06,
+      "loss": 1.6529,
+      "step": 3095
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2.8441988438779543e-06,
+      "loss": 1.6664,
+      "step": 3100
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.8136439454220864e-06,
+      "loss": 1.627,
+      "step": 3105
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.7832271662040176e-06,
+      "loss": 1.6276,
+      "step": 3110
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.453125,
+      "learning_rate": 2.7529490908235523e-06,
+      "loss": 1.626,
+      "step": 3115
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.4375,
+      "learning_rate": 2.7228103012146657e-06,
+      "loss": 1.5861,
+      "step": 3120
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2.692811376634297e-06,
+      "loss": 1.6465,
+      "step": 3125
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.375,
+      "learning_rate": 2.6629528936512306e-06,
+      "loss": 1.5498,
+      "step": 3130
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.6332354261350247e-06,
+      "loss": 1.6333,
+      "step": 3135
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.4375,
+      "learning_rate": 2.6036595452449554e-06,
+      "loss": 1.6199,
+      "step": 3140
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.5742258194190607e-06,
+      "loss": 1.6126,
+      "step": 3145
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2.544934814363206e-06,
+      "loss": 1.6537,
+      "step": 3150
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2.515787093040212e-06,
+      "loss": 1.6169,
+      "step": 3155
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.34375,
+      "learning_rate": 2.4867832156590444e-06,
+      "loss": 1.5844,
+      "step": 3160
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.4375,
+      "learning_rate": 2.457923739664031e-06,
+      "loss": 1.6596,
+      "step": 3165
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.4292092197241577e-06,
+      "loss": 1.629,
+      "step": 3170
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.40625,
+      "learning_rate": 2.4006402077224067e-06,
+      "loss": 1.6467,
+      "step": 3175
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.372217252745145e-06,
+      "loss": 1.6157,
+      "step": 3180
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.40625,
+      "learning_rate": 2.343940901071584e-06,
+      "loss": 1.5702,
+      "step": 3185
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2.3158116961632626e-06,
+      "loss": 1.7161,
+      "step": 3190
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.421875,
+      "learning_rate": 2.2878301786536137e-06,
+      "loss": 1.6483,
+      "step": 3195
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2.259996886337571e-06,
+      "loss": 1.669,
+      "step": 3200
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.232312354161226e-06,
+      "loss": 1.6202,
+      "step": 3205
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.204777114211565e-06,
+      "loss": 1.6383,
+      "step": 3210
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.177391695706219e-06,
+      "loss": 1.6728,
+      "step": 3215
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2.1501566249833082e-06,
+      "loss": 1.5633,
+      "step": 3220
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.1230724254913184e-06,
+      "loss": 1.5665,
+      "step": 3225
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2.0961396177790494e-06,
+      "loss": 1.6195,
+      "step": 3230
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.453125,
+      "learning_rate": 2.0693587194855967e-06,
+      "loss": 1.6128,
+      "step": 3235
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2.0427302453304143e-06,
+      "loss": 1.6486,
+      "step": 3240
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.0162547071034187e-06,
+      "loss": 1.5957,
+      "step": 3245
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.98993261365515e-06,
+      "loss": 1.6555,
+      "step": 3250
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.963764470886992e-06,
+      "loss": 1.6126,
+      "step": 3255
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.9377507817414532e-06,
+      "loss": 1.5952,
+      "step": 3260
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.9118920461924938e-06,
+      "loss": 1.6249,
+      "step": 3265
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.8861887612359286e-06,
+      "loss": 1.6511,
+      "step": 3270
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.8606414208798596e-06,
+      "loss": 1.6364,
+      "step": 3275
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.8352505161351874e-06,
+      "loss": 1.6138,
+      "step": 3280
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.8100165350061816e-06,
+      "loss": 1.6323,
+      "step": 3285
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.7849399624810848e-06,
+      "loss": 1.6198,
+      "step": 3290
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.7600212805228156e-06,
+      "loss": 1.6198,
+      "step": 3295
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.735260968059681e-06,
+      "loss": 1.5782,
+      "step": 3300
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.710659500976185e-06,
+      "loss": 1.6216,
+      "step": 3305
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.6862173521038783e-06,
+      "loss": 1.6247,
+      "step": 3310
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.6619349912122695e-06,
+      "loss": 1.6394,
+      "step": 3315
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.6378128849998065e-06,
+      "loss": 1.5938,
+      "step": 3320
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.4609375,
+      "learning_rate": 1.6138514970848918e-06,
+      "loss": 1.588,
+      "step": 3325
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.5900512879969765e-06,
+      "loss": 1.6692,
+      "step": 3330
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.56641271516772e-06,
+      "loss": 1.6308,
+      "step": 3335
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.3671875,
+      "learning_rate": 1.542936232922182e-06,
+      "loss": 1.6051,
+      "step": 3340
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.5196222924700988e-06,
+      "loss": 1.6803,
+      "step": 3345
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.4964713418972166e-06,
+      "loss": 1.5859,
+      "step": 3350
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.4734838261566676e-06,
+      "loss": 1.6232,
+      "step": 3355
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.4506601870604253e-06,
+      "loss": 1.6776,
+      "step": 3360
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.428000863270812e-06,
+      "loss": 1.5972,
+      "step": 3365
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.5,
+      "learning_rate": 1.4055062902920657e-06,
+      "loss": 1.598,
+      "step": 3370
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.3831769004619766e-06,
+      "loss": 1.64,
+      "step": 3375
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.3610131229435696e-06,
+      "loss": 1.6315,
+      "step": 3380
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.3390153837168573e-06,
+      "loss": 1.6181,
+      "step": 3385
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.4453125,
+      "learning_rate": 1.3171841055706591e-06,
+      "loss": 1.5947,
+      "step": 3390
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.2955197080944648e-06,
+      "loss": 1.6375,
+      "step": 3395
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.2740226076703844e-06,
+      "loss": 1.6416,
+      "step": 3400
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.2526932174651307e-06,
+      "loss": 1.6436,
+      "step": 3405
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.375,
+      "learning_rate": 1.23153194742209e-06,
+      "loss": 1.6326,
+      "step": 3410
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.2105392042534325e-06,
+      "loss": 1.5622,
+      "step": 3415
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.1897153914323034e-06,
+      "loss": 1.6367,
+      "step": 3420
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.1690609091850713e-06,
+      "loss": 1.636,
+      "step": 3425
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.1485761544836238e-06,
+      "loss": 1.5795,
+      "step": 3430
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.1282615210377456e-06,
+      "loss": 1.6616,
+      "step": 3435
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.108117399287555e-06,
+      "loss": 1.6382,
+      "step": 3440
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.088144176395992e-06,
+      "loss": 1.6355,
+      "step": 3445
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.0683422362413797e-06,
+      "loss": 1.6833,
+      "step": 3450
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.048711959410048e-06,
+      "loss": 1.6637,
+      "step": 3455
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.029253723189022e-06,
+      "loss": 1.5878,
+      "step": 3460
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.375,
+      "learning_rate": 1.0099679015587616e-06,
+      "loss": 1.614,
+      "step": 3465
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.4453125,
+      "learning_rate": 9.908548651859807e-07,
+      "loss": 1.5783,
+      "step": 3470
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.4453125,
+      "learning_rate": 9.719149814165219e-07,
+      "loss": 1.6576,
+      "step": 3475
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.421875,
+      "learning_rate": 9.531486142682922e-07,
+      "loss": 1.6059,
+      "step": 3480
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.375,
+      "learning_rate": 9.345561244242773e-07,
+      "loss": 1.6049,
+      "step": 3485
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.453125,
+      "learning_rate": 9.161378692255963e-07,
+      "loss": 1.6458,
+      "step": 3490
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.46875,
+      "learning_rate": 8.978942026646409e-07,
+      "loss": 1.5813,
+      "step": 3495
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.3984375,
+      "learning_rate": 8.798254753782698e-07,
+      "loss": 1.6261,
+      "step": 3500
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.3671875,
+      "learning_rate": 8.61932034641072e-07,
+      "loss": 1.6255,
+      "step": 3505
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.40625,
+      "learning_rate": 8.442142243586904e-07,
+      "loss": 1.5771,
+      "step": 3510
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.421875,
+      "learning_rate": 8.266723850612124e-07,
+      "loss": 1.6067,
+      "step": 3515
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.421875,
+      "learning_rate": 8.093068538966242e-07,
+      "loss": 1.6403,
+      "step": 3520
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.484375,
+      "learning_rate": 7.921179646243315e-07,
+      "loss": 1.6435,
+      "step": 3525
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.4765625,
+      "learning_rate": 7.751060476087435e-07,
+      "loss": 1.6065,
+      "step": 3530
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.4296875,
+      "learning_rate": 7.582714298129302e-07,
+      "loss": 1.6341,
+      "step": 3535
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.375,
+      "learning_rate": 7.416144347923271e-07,
+      "loss": 1.5997,
+      "step": 3540
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.390625,
+      "learning_rate": 7.251353826885232e-07,
+      "loss": 1.6263,
+      "step": 3545
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.46875,
+      "learning_rate": 7.08834590223112e-07,
+      "loss": 1.6525,
+      "step": 3550
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.3671875,
+      "learning_rate": 6.92712370691594e-07,
+      "loss": 1.6177,
+      "step": 3555
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.453125,
+      "learning_rate": 6.767690339573607e-07,
+      "loss": 1.6611,
+      "step": 3560
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.4296875,
+      "learning_rate": 6.610048864457452e-07,
+      "loss": 1.6394,
+      "step": 3565
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.3828125,
+      "learning_rate": 6.454202311381197e-07,
+      "loss": 1.6174,
+      "step": 3570
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.3828125,
+      "learning_rate": 6.300153675660836e-07,
+      "loss": 1.6583,
+      "step": 3575
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.4375,
+      "learning_rate": 6.147905918057007e-07,
+      "loss": 1.6458,
+      "step": 3580
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.40625,
+      "learning_rate": 5.997461964718076e-07,
+      "loss": 1.6303,
+      "step": 3585
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.390625,
+      "learning_rate": 5.84882470712399e-07,
+      "loss": 1.6162,
+      "step": 3590
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.4375,
+      "learning_rate": 5.701997002030602e-07,
+      "loss": 1.6281,
+      "step": 3595
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.4140625,
+      "learning_rate": 5.556981671414785e-07,
+      "loss": 1.6293,
+      "step": 3600
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.4296875,
+      "learning_rate": 5.413781502420234e-07,
+      "loss": 1.6084,
+      "step": 3605
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.421875,
+      "learning_rate": 5.272399247303861e-07,
+      "loss": 1.6464,
+      "step": 3610
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.453125,
+      "learning_rate": 5.132837623382947e-07,
+      "loss": 1.6099,
+      "step": 3615
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.40625,
+      "learning_rate": 4.995099312982843e-07,
+      "loss": 1.6073,
+      "step": 3620
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.421875,
+      "learning_rate": 4.859186963385465e-07,
+      "loss": 1.6105,
+      "step": 3625
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.46875,
+      "learning_rate": 4.725103186778424e-07,
+      "loss": 1.6115,
+      "step": 3630
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.421875,
+      "learning_rate": 4.5928505602047426e-07,
+      "loss": 1.6123,
+      "step": 3635
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.390625,
+      "learning_rate": 4.462431625513475e-07,
+      "loss": 1.6208,
+      "step": 3640
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.40625,
+      "learning_rate": 4.3338488893106876e-07,
+      "loss": 1.6154,
+      "step": 3645
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.375,
+      "learning_rate": 4.207104822911368e-07,
+      "loss": 1.6232,
+      "step": 3650
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.4375,
+      "learning_rate": 4.082201862291957e-07,
+      "loss": 1.6444,
+      "step": 3655
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.959142408043448e-07,
+      "loss": 1.6303,
+      "step": 3660
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.3828125,
+      "learning_rate": 3.837928825325299e-07,
+      "loss": 1.5837,
+      "step": 3665
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.3984375,
+      "learning_rate": 3.718563443819989e-07,
+      "loss": 1.622,
+      "step": 3670
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.4375,
+      "learning_rate": 3.60104855768818e-07,
+      "loss": 1.5705,
+      "step": 3675
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.4140625,
+      "learning_rate": 3.485386425524695e-07,
+      "loss": 1.6096,
+      "step": 3680
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.371579270315073e-07,
+      "loss": 1.6356,
+      "step": 3685
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.421875,
+      "learning_rate": 3.2596292793928174e-07,
+      "loss": 1.5869,
+      "step": 3690
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.390625,
+      "learning_rate": 3.1495386043974173e-07,
+      "loss": 1.6391,
+      "step": 3695
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.4765625,
+      "learning_rate": 3.0413093612329805e-07,
+      "loss": 1.6174,
+      "step": 3700
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2.934943630027509e-07,
+      "loss": 1.604,
+      "step": 3705
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2.830443455092957e-07,
+      "loss": 1.628,
+      "step": 3710
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.40625,
+      "learning_rate": 2.727810844885992e-07,
+      "loss": 1.6104,
+      "step": 3715
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.453125,
+      "learning_rate": 2.6270477719692823e-07,
+      "loss": 1.6315,
+      "step": 3720
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.4375,
+      "learning_rate": 2.528156172973706e-07,
+      "loss": 1.6311,
+      "step": 3725
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.40625,
+      "learning_rate": 2.431137948561024e-07,
+      "loss": 1.6617,
+      "step": 3730
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.3359949633874224e-07,
+      "loss": 1.5982,
+      "step": 3735
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.2427290460676378e-07,
+      "loss": 1.6156,
+      "step": 3740
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2.1513419891398103e-07,
+      "loss": 1.6412,
+      "step": 3745
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2.061835549031077e-07,
+      "loss": 1.5988,
+      "step": 3750
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.9742114460237439e-07,
+      "loss": 1.6496,
+      "step": 3755
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.8884713642222773e-07,
+      "loss": 1.5901,
+      "step": 3760
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.8046169515209432e-07,
+      "loss": 1.616,
+      "step": 3765
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.722649819572053e-07,
+      "loss": 1.6194,
+      "step": 3770
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.4140625,
+      "learning_rate": 1.6425715437550783e-07,
+      "loss": 1.5966,
+      "step": 3775
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.3671875,
+      "learning_rate": 1.5643836631463626e-07,
+      "loss": 1.5756,
+      "step": 3780
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.4880876804894807e-07,
+      "loss": 1.6281,
+      "step": 3785
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.4136850621663922e-07,
+      "loss": 1.6399,
+      "step": 3790
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.3411772381692888e-07,
+      "loss": 1.6618,
+      "step": 3795
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.2705656020730484e-07,
+      "loss": 1.5982,
+      "step": 3800
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.2018515110085227e-07,
+      "loss": 1.6163,
+      "step": 3805
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.1350362856363817e-07,
+      "loss": 1.6177,
+      "step": 3810
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.0701212101217772e-07,
+      "loss": 1.6262,
+      "step": 3815
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.0071075321096524e-07,
+      "loss": 1.6355,
+      "step": 3820
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.4140625,
+      "learning_rate": 9.459964627007378e-08,
+      "loss": 1.6219,
+      "step": 3825
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.390625,
+      "learning_rate": 8.867891764283154e-08,
+      "loss": 1.5891,
+      "step": 3830
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.4140625,
+      "learning_rate": 8.294868112356358e-08,
+      "loss": 1.5979,
+      "step": 3835
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.40625,
+      "learning_rate": 7.74090468454014e-08,
+      "loss": 1.6252,
+      "step": 3840
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.4140625,
+      "learning_rate": 7.206012127816686e-08,
+      "loss": 1.6028,
+      "step": 3845
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.421875,
+      "learning_rate": 6.690200722633156e-08,
+      "loss": 1.6106,
+      "step": 3850
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.4296875,
+      "learning_rate": 6.19348038270351e-08,
+      "loss": 1.6085,
+      "step": 3855
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.4375,
+      "learning_rate": 5.715860654817995e-08,
+      "loss": 1.6112,
+      "step": 3860
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.375,
+      "learning_rate": 5.2573507186599595e-08,
+      "loss": 1.639,
+      "step": 3865
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.4375,
+      "learning_rate": 4.817959386629212e-08,
+      "loss": 1.6482,
+      "step": 3870
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.390625,
+      "learning_rate": 4.397695103672828e-08,
+      "loss": 1.6347,
+      "step": 3875
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.996565947122499e-08,
+      "loss": 1.6203,
+      "step": 3880
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.4375,
+      "learning_rate": 3.6145796265397716e-08,
+      "loss": 1.6375,
+      "step": 3885
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.3984375,
+      "learning_rate": 3.2517434835674936e-08,
+      "loss": 1.6312,
+      "step": 3890
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.908064491788709e-08,
+      "loss": 1.6125,
+      "step": 3895
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2.5835492565928766e-08,
+      "loss": 1.652,
+      "step": 3900
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.278204015048746e-08,
+      "loss": 1.6038,
+      "step": 3905
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.9920346357846787e-08,
+      "loss": 1.6156,
+      "step": 3910
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.46875,
+      "learning_rate": 1.7250466188754035e-08,
+      "loss": 1.6265,
+      "step": 3915
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.4772450957368788e-08,
+      "loss": 1.6656,
+      "step": 3920
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.2486348290273731e-08,
+      "loss": 1.606,
+      "step": 3925
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.3984375,
+      "learning_rate": 1.0392202125556473e-08,
+      "loss": 1.6396,
+      "step": 3930
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.3984375,
+      "learning_rate": 8.490052711972452e-09,
+      "loss": 1.6434,
+      "step": 3935
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.4609375,
+      "learning_rate": 6.77993660816334e-09,
+      "loss": 1.6536,
+      "step": 3940
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.3515625,
+      "learning_rate": 5.261886681959816e-09,
+      "loss": 1.6093,
+      "step": 3945
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.453125,
+      "learning_rate": 3.935932109749852e-09,
+      "loss": 1.6293,
+      "step": 3950
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.8020983759124985e-09,
+      "loss": 1.6311,
+      "step": 3955
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.4375,
+      "learning_rate": 1.8604072723338306e-09,
+      "loss": 1.6457,
+      "step": 3960
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.3671875,
+      "learning_rate": 1.1108768979861685e-09,
+      "loss": 1.5786,
+      "step": 3965
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.4765625,
+      "learning_rate": 5.5352165857947e-10,
+      "loss": 1.5546,
+      "step": 3970
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.421875,
+      "learning_rate": 1.8835226628710623e-10,
+      "loss": 1.6215,
+      "step": 3975
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.5375739534917445e-11,
+      "loss": 1.5669,
+      "step": 3980
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 1.628558874130249,
+      "eval_runtime": 653.8126,
+      "eval_samples_per_second": 21.564,
+      "eval_steps_per_second": 1.349,
+      "step": 3982
+    },
+    {
+      "epoch": 1.0,
+      "step": 3982,
+      "total_flos": 2.7700535054303232e+17,
+      "train_loss": 1.6493343958478666,
+      "train_runtime": 16072.2247,
+      "train_samples_per_second": 7.928,
+      "train_steps_per_second": 0.248
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 3982,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 100,
+  "total_flos": 2.7700535054303232e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}