diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,6754 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.96,
+  "eval_steps": 500,
+  "global_step": 4800,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.001,
+      "grad_norm": 13.726217269897461,
+      "learning_rate": 4e-06,
+      "loss": 10.4133,
+      "step": 5
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 4.9693684577941895,
+      "learning_rate": 8.999999999999999e-06,
+      "loss": 9.8384,
+      "step": 10
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 2.879361152648926,
+      "learning_rate": 1.4e-05,
+      "loss": 9.5182,
+      "step": 15
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 3.023117780685425,
+      "learning_rate": 1.9e-05,
+      "loss": 9.3697,
+      "step": 20
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 3.9188971519470215,
+      "learning_rate": 2.3999999999999997e-05,
+      "loss": 9.1674,
+      "step": 25
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 4.881481647491455,
+      "learning_rate": 2.8999999999999997e-05,
+      "loss": 9.0057,
+      "step": 30
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 5.366204261779785,
+      "learning_rate": 3.399999999999999e-05,
+      "loss": 8.8219,
+      "step": 35
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 12.275991439819336,
+      "learning_rate": 3.9e-05,
+      "loss": 8.6271,
+      "step": 40
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 3.798144578933716,
+      "learning_rate": 4.4e-05,
+      "loss": 8.3997,
+      "step": 45
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.8132495880126953,
+      "learning_rate": 4.899999999999999e-05,
+      "loss": 8.211,
+      "step": 50
+    },
+    {
+      "epoch": 0.011,
+      "grad_norm": 3.747354507446289,
+      "learning_rate": 5.399999999999999e-05,
+      "loss": 8.0857,
+      "step": 55
+    },
+    {
+      "epoch": 0.012,
+      "grad_norm": 3.045440435409546,
+      "learning_rate": 5.899999999999999e-05,
+      "loss": 7.7861,
+      "step": 60
+    },
+    {
+      "epoch": 0.013,
+      "grad_norm": 2.5404093265533447,
+      "learning_rate": 6.4e-05,
+      "loss": 7.5872,
+      "step": 65
+    },
+    {
+      "epoch": 0.014,
+      "grad_norm": 2.668727159500122,
+      "learning_rate": 6.9e-05,
+      "loss": 7.2799,
+      "step": 70
+    },
+    {
+      "epoch": 0.015,
+      "grad_norm": 2.7760627269744873,
+      "learning_rate": 7.4e-05,
+      "loss": 7.1264,
+      "step": 75
+    },
+    {
+      "epoch": 0.016,
+      "grad_norm": 2.291050434112549,
+      "learning_rate": 7.899999999999998e-05,
+      "loss": 6.9572,
+      "step": 80
+    },
+    {
+      "epoch": 0.017,
+      "grad_norm": 2.175490140914917,
+      "learning_rate": 8.4e-05,
+      "loss": 6.7924,
+      "step": 85
+    },
+    {
+      "epoch": 0.018,
+      "grad_norm": 2.3527913093566895,
+      "learning_rate": 8.9e-05,
+      "loss": 6.7404,
+      "step": 90
+    },
+    {
+      "epoch": 0.019,
+      "grad_norm": 2.1863811016082764,
+      "learning_rate": 9.4e-05,
+      "loss": 6.4791,
+      "step": 95
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.8337147235870361,
+      "learning_rate": 9.9e-05,
+      "loss": 6.3777,
+      "step": 100
+    },
+    {
+      "epoch": 0.021,
+      "grad_norm": 2.6010560989379883,
+      "learning_rate": 0.000104,
+      "loss": 6.2828,
+      "step": 105
+    },
+    {
+      "epoch": 0.022,
+      "grad_norm": 2.021232843399048,
+      "learning_rate": 0.00010899999999999999,
+      "loss": 6.1714,
+      "step": 110
+    },
+    {
+      "epoch": 0.023,
+      "grad_norm": 2.872624397277832,
+      "learning_rate": 0.00011399999999999999,
+      "loss": 6.055,
+      "step": 115
+    },
+    {
+      "epoch": 0.024,
+      "grad_norm": 1.730695366859436,
+      "learning_rate": 0.00011899999999999999,
+      "loss": 5.9749,
+      "step": 120
+    },
+    {
+      "epoch": 0.025,
+      "grad_norm": 1.8848776817321777,
+      "learning_rate": 0.00012399999999999998,
+      "loss": 5.8248,
+      "step": 125
+    },
+    {
+      "epoch": 0.026,
+      "grad_norm": 1.6250139474868774,
+      "learning_rate": 0.000129,
+      "loss": 5.8104,
+      "step": 130
+    },
+    {
+      "epoch": 0.027,
+      "grad_norm": 1.776007056236267,
+      "learning_rate": 0.00013399999999999998,
+      "loss": 5.6435,
+      "step": 135
+    },
+    {
+      "epoch": 0.028,
+      "grad_norm": 1.5446497201919556,
+      "learning_rate": 0.000139,
+      "loss": 5.6116,
+      "step": 140
+    },
+    {
+      "epoch": 0.029,
+      "grad_norm": 1.7168349027633667,
+      "learning_rate": 0.00014399999999999998,
+      "loss": 5.5175,
+      "step": 145
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.5292563438415527,
+      "learning_rate": 0.000149,
+      "loss": 5.4173,
+      "step": 150
+    },
+    {
+      "epoch": 0.031,
+      "grad_norm": 1.3585489988327026,
+      "learning_rate": 0.00015399999999999998,
+      "loss": 5.363,
+      "step": 155
+    },
+    {
+      "epoch": 0.032,
+      "grad_norm": 1.428066372871399,
+      "learning_rate": 0.000159,
+      "loss": 5.3579,
+      "step": 160
+    },
+    {
+      "epoch": 0.033,
+      "grad_norm": 1.2797551155090332,
+      "learning_rate": 0.00016399999999999997,
+      "loss": 5.2454,
+      "step": 165
+    },
+    {
+      "epoch": 0.034,
+      "grad_norm": 1.2886139154434204,
+      "learning_rate": 0.000169,
+      "loss": 5.2456,
+      "step": 170
+    },
+    {
+      "epoch": 0.035,
+      "grad_norm": 1.220340609550476,
+      "learning_rate": 0.00017399999999999997,
+      "loss": 5.2171,
+      "step": 175
+    },
+    {
+      "epoch": 0.036,
+      "grad_norm": 1.3801219463348389,
+      "learning_rate": 0.000179,
+      "loss": 5.1411,
+      "step": 180
+    },
+    {
+      "epoch": 0.037,
+      "grad_norm": 1.2507349252700806,
+      "learning_rate": 0.00018399999999999997,
+      "loss": 5.0654,
+      "step": 185
+    },
+    {
+      "epoch": 0.038,
+      "grad_norm": 1.1602240800857544,
+      "learning_rate": 0.00018899999999999999,
+      "loss": 4.9534,
+      "step": 190
+    },
+    {
+      "epoch": 0.039,
+      "grad_norm": 1.1360636949539185,
+      "learning_rate": 0.00019399999999999997,
+      "loss": 5.003,
+      "step": 195
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.3035187721252441,
+      "learning_rate": 0.00019899999999999999,
+      "loss": 5.0598,
+      "step": 200
+    },
+    {
+      "epoch": 0.041,
+      "grad_norm": 1.2107629776000977,
+      "learning_rate": 0.000204,
+      "loss": 4.862,
+      "step": 205
+    },
+    {
+      "epoch": 0.042,
+      "grad_norm": 1.2512259483337402,
+      "learning_rate": 0.00020899999999999998,
+      "loss": 4.8316,
+      "step": 210
+    },
+    {
+      "epoch": 0.043,
+      "grad_norm": 1.1587598323822021,
+      "learning_rate": 0.000214,
+      "loss": 4.8094,
+      "step": 215
+    },
+    {
+      "epoch": 0.044,
+      "grad_norm": 1.0695734024047852,
+      "learning_rate": 0.00021899999999999998,
+      "loss": 4.7331,
+      "step": 220
+    },
+    {
+      "epoch": 0.045,
+      "grad_norm": 1.2294334173202515,
+      "learning_rate": 0.000224,
+      "loss": 4.7008,
+      "step": 225
+    },
+    {
+      "epoch": 0.046,
+      "grad_norm": 0.9978997707366943,
+      "learning_rate": 0.00022899999999999998,
+      "loss": 4.6802,
+      "step": 230
+    },
+    {
+      "epoch": 0.047,
+      "grad_norm": 0.9799301028251648,
+      "learning_rate": 0.000234,
+      "loss": 4.6729,
+      "step": 235
+    },
+    {
+      "epoch": 0.048,
+      "grad_norm": 0.9935582280158997,
+      "learning_rate": 0.00023899999999999998,
+      "loss": 4.6054,
+      "step": 240
+    },
+    {
+      "epoch": 0.049,
+      "grad_norm": 1.0166147947311401,
+      "learning_rate": 0.000244,
+      "loss": 4.5731,
+      "step": 245
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.9937739372253418,
+      "learning_rate": 0.000249,
+      "loss": 4.4931,
+      "step": 250
+    },
+    {
+      "epoch": 0.051,
+      "grad_norm": 1.3121594190597534,
+      "learning_rate": 0.000254,
+      "loss": 4.5523,
+      "step": 255
+    },
+    {
+      "epoch": 0.052,
+      "grad_norm": 0.8952345252037048,
+      "learning_rate": 0.00025899999999999995,
+      "loss": 4.458,
+      "step": 260
+    },
+    {
+      "epoch": 0.053,
+      "grad_norm": 0.9221830368041992,
+      "learning_rate": 0.00026399999999999997,
+      "loss": 4.47,
+      "step": 265
+    },
+    {
+      "epoch": 0.054,
+      "grad_norm": 0.9459202289581299,
+      "learning_rate": 0.000269,
+      "loss": 4.3436,
+      "step": 270
+    },
+    {
+      "epoch": 0.055,
+      "grad_norm": 0.907535970211029,
+      "learning_rate": 0.000274,
+      "loss": 4.399,
+      "step": 275
+    },
+    {
+      "epoch": 0.056,
+      "grad_norm": 0.9379016757011414,
+      "learning_rate": 0.000279,
+      "loss": 4.3173,
+      "step": 280
+    },
+    {
+      "epoch": 0.057,
+      "grad_norm": 0.8357326984405518,
+      "learning_rate": 0.00028399999999999996,
+      "loss": 4.4404,
+      "step": 285
+    },
+    {
+      "epoch": 0.058,
+      "grad_norm": 0.8483668565750122,
+      "learning_rate": 0.000289,
+      "loss": 4.3564,
+      "step": 290
+    },
+    {
+      "epoch": 0.059,
+      "grad_norm": 0.8923192024230957,
+      "learning_rate": 0.000294,
+      "loss": 4.2908,
+      "step": 295
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.8518162369728088,
+      "learning_rate": 0.000299,
+      "loss": 4.3578,
+      "step": 300
+    },
+    {
+      "epoch": 0.061,
+      "grad_norm": 0.7717773914337158,
+      "learning_rate": 0.00029999946385159685,
+      "loss": 4.274,
+      "step": 305
+    },
+    {
+      "epoch": 0.062,
+      "grad_norm": 0.8026501536369324,
+      "learning_rate": 0.00029999728575527777,
+      "loss": 4.2535,
+      "step": 310
+    },
+    {
+      "epoch": 0.063,
+      "grad_norm": 0.7490338087081909,
+      "learning_rate": 0.0002999934322260777,
+      "loss": 4.1746,
+      "step": 315
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 0.7846081256866455,
+      "learning_rate": 0.0002999879033070396,
+      "loss": 4.191,
+      "step": 320
+    },
+    {
+      "epoch": 0.065,
+      "grad_norm": 0.9395541548728943,
+      "learning_rate": 0.0002999806990599202,
+      "loss": 4.1992,
+      "step": 325
+    },
+    {
+      "epoch": 0.066,
+      "grad_norm": 0.7602784633636475,
+      "learning_rate": 0.00029997181956518905,
+      "loss": 4.0873,
+      "step": 330
+    },
+    {
+      "epoch": 0.067,
+      "grad_norm": 0.769402027130127,
+      "learning_rate": 0.000299961264922028,
+      "loss": 4.0827,
+      "step": 335
+    },
+    {
+      "epoch": 0.068,
+      "grad_norm": 0.7498162984848022,
+      "learning_rate": 0.0002999490352483299,
+      "loss": 4.1329,
+      "step": 340
+    },
+    {
+      "epoch": 0.069,
+      "grad_norm": 0.7378515005111694,
+      "learning_rate": 0.00029993513068069704,
+      "loss": 4.0957,
+      "step": 345
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.7770788669586182,
+      "learning_rate": 0.00029991955137444014,
+      "loss": 3.9895,
+      "step": 350
+    },
+    {
+      "epoch": 0.071,
+      "grad_norm": 0.6819906830787659,
+      "learning_rate": 0.0002999022975035762,
+      "loss": 3.9696,
+      "step": 355
+    },
+    {
+      "epoch": 0.072,
+      "grad_norm": 0.7527605295181274,
+      "learning_rate": 0.0002998833692608267,
+      "loss": 4.0049,
+      "step": 360
+    },
+    {
+      "epoch": 0.073,
+      "grad_norm": 0.8272974491119385,
+      "learning_rate": 0.0002998627668576155,
+      "loss": 3.9313,
+      "step": 365
+    },
+    {
+      "epoch": 0.074,
+      "grad_norm": 0.6840147376060486,
+      "learning_rate": 0.00029984049052406646,
+      "loss": 3.9424,
+      "step": 370
+    },
+    {
+      "epoch": 0.075,
+      "grad_norm": 0.6854280233383179,
+      "learning_rate": 0.00029981654050900057,
+      "loss": 3.9876,
+      "step": 375
+    },
+    {
+      "epoch": 0.076,
+      "grad_norm": 0.6815478205680847,
+      "learning_rate": 0.00029979091707993383,
+      "loss": 3.8654,
+      "step": 380
+    },
+    {
+      "epoch": 0.077,
+      "grad_norm": 0.7019010186195374,
+      "learning_rate": 0.00029976362052307355,
+      "loss": 3.9083,
+      "step": 385
+    },
+    {
+      "epoch": 0.078,
+      "grad_norm": 0.7006312608718872,
+      "learning_rate": 0.00029973465114331555,
+      "loss": 3.8683,
+      "step": 390
+    },
+    {
+      "epoch": 0.079,
+      "grad_norm": 0.6919441223144531,
+      "learning_rate": 0.0002997040092642407,
+      "loss": 3.8363,
+      "step": 395
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.6442705392837524,
+      "learning_rate": 0.00029967169522811135,
+      "loss": 3.7315,
+      "step": 400
+    },
+    {
+      "epoch": 0.081,
+      "grad_norm": 0.6530693769454956,
+      "learning_rate": 0.0002996377093958673,
+      "loss": 3.8249,
+      "step": 405
+    },
+    {
+      "epoch": 0.082,
+      "grad_norm": 0.7034955024719238,
+      "learning_rate": 0.0002996020521471219,
+      "loss": 3.7612,
+      "step": 410
+    },
+    {
+      "epoch": 0.083,
+      "grad_norm": 0.6602813601493835,
+      "learning_rate": 0.000299564723880158,
+      "loss": 3.7912,
+      "step": 415
+    },
+    {
+      "epoch": 0.084,
+      "grad_norm": 0.6340587139129639,
+      "learning_rate": 0.00029952572501192314,
+      "loss": 3.7946,
+      "step": 420
+    },
+    {
+      "epoch": 0.085,
+      "grad_norm": 0.6614922285079956,
+      "learning_rate": 0.000299485055978025,
+      "loss": 3.7346,
+      "step": 425
+    },
+    {
+      "epoch": 0.086,
+      "grad_norm": 0.6904277205467224,
+      "learning_rate": 0.00029944271723272685,
+      "loss": 3.608,
+      "step": 430
+    },
+    {
+      "epoch": 0.087,
+      "grad_norm": 1.0566864013671875,
+      "learning_rate": 0.00029939870924894206,
+      "loss": 3.7429,
+      "step": 435
+    },
+    {
+      "epoch": 0.088,
+      "grad_norm": 0.6728774905204773,
+      "learning_rate": 0.000299353032518229,
+      "loss": 3.7887,
+      "step": 440
+    },
+    {
+      "epoch": 0.089,
+      "grad_norm": 0.6397653818130493,
+      "learning_rate": 0.00029930568755078544,
+      "loss": 3.7278,
+      "step": 445
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.649704098701477,
+      "learning_rate": 0.0002992566748754432,
+      "loss": 3.6609,
+      "step": 450
+    },
+    {
+      "epoch": 0.091,
+      "grad_norm": 0.6457221508026123,
+      "learning_rate": 0.0002992059950396618,
+      "loss": 3.6433,
+      "step": 455
+    },
+    {
+      "epoch": 0.092,
+      "grad_norm": 0.6439397931098938,
+      "learning_rate": 0.00029915364860952256,
+      "loss": 3.5866,
+      "step": 460
+    },
+    {
+      "epoch": 0.093,
+      "grad_norm": 0.6577675342559814,
+      "learning_rate": 0.00029909963616972225,
+      "loss": 3.6028,
+      "step": 465
+    },
+    {
+      "epoch": 0.094,
+      "grad_norm": 0.6369614601135254,
+      "learning_rate": 0.00029904395832356653,
+      "loss": 3.6547,
+      "step": 470
+    },
+    {
+      "epoch": 0.095,
+      "grad_norm": 0.6148671507835388,
+      "learning_rate": 0.00029898661569296336,
+      "loss": 3.5645,
+      "step": 475
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 0.6119360327720642,
+      "learning_rate": 0.00029892760891841585,
+      "loss": 3.5271,
+      "step": 480
+    },
+    {
+      "epoch": 0.097,
+      "grad_norm": 0.6091258525848389,
+      "learning_rate": 0.0002988669386590151,
+      "loss": 3.5216,
+      "step": 485
+    },
+    {
+      "epoch": 0.098,
+      "grad_norm": 0.6266127228736877,
+      "learning_rate": 0.00029880460559243314,
+      "loss": 3.5137,
+      "step": 490
+    },
+    {
+      "epoch": 0.099,
+      "grad_norm": 0.6600483059883118,
+      "learning_rate": 0.00029874061041491493,
+      "loss": 3.6216,
+      "step": 495
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.5879938006401062,
+      "learning_rate": 0.000298674953841271,
+      "loss": 3.4859,
+      "step": 500
+    },
+    {
+      "epoch": 0.101,
+      "grad_norm": 0.5879422426223755,
+      "learning_rate": 0.00029860763660486913,
+      "loss": 3.5691,
+      "step": 505
+    },
+    {
+      "epoch": 0.102,
+      "grad_norm": 0.6330602169036865,
+      "learning_rate": 0.0002985386594576263,
+      "loss": 3.4669,
+      "step": 510
+    },
+    {
+      "epoch": 0.103,
+      "grad_norm": 0.6023596525192261,
+      "learning_rate": 0.0002984680231700004,
+      "loss": 3.5629,
+      "step": 515
+    },
+    {
+      "epoch": 0.104,
+      "grad_norm": 0.6126276850700378,
+      "learning_rate": 0.00029839572853098133,
+      "loss": 3.4725,
+      "step": 520
+    },
+    {
+      "epoch": 0.105,
+      "grad_norm": 0.5834094882011414,
+      "learning_rate": 0.00029832177634808255,
+      "loss": 3.4099,
+      "step": 525
+    },
+    {
+      "epoch": 0.106,
+      "grad_norm": 0.6275287866592407,
+      "learning_rate": 0.0002982461674473317,
+      "loss": 3.4758,
+      "step": 530
+    },
+    {
+      "epoch": 0.107,
+      "grad_norm": 0.5847580432891846,
+      "learning_rate": 0.00029816890267326184,
+      "loss": 3.319,
+      "step": 535
+    },
+    {
+      "epoch": 0.108,
+      "grad_norm": 0.6252226829528809,
+      "learning_rate": 0.0002980899828889013,
+      "loss": 3.4195,
+      "step": 540
+    },
+    {
+      "epoch": 0.109,
+      "grad_norm": 0.6148211359977722,
+      "learning_rate": 0.0002980094089757649,
+      "loss": 3.3954,
+      "step": 545
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.6043537855148315,
+      "learning_rate": 0.0002979271818338434,
+      "loss": 3.3332,
+      "step": 550
+    },
+    {
+      "epoch": 0.111,
+      "grad_norm": 0.6230714321136475,
+      "learning_rate": 0.0002978433023815939,
+      "loss": 3.3496,
+      "step": 555
+    },
+    {
+      "epoch": 0.112,
+      "grad_norm": 0.6967390179634094,
+      "learning_rate": 0.00029775777155592907,
+      "loss": 3.3825,
+      "step": 560
+    },
+    {
+      "epoch": 0.113,
+      "grad_norm": 0.6039656400680542,
+      "learning_rate": 0.00029767059031220747,
+      "loss": 3.3775,
+      "step": 565
+    },
+    {
+      "epoch": 0.114,
+      "grad_norm": 0.6448938846588135,
+      "learning_rate": 0.00029758175962422214,
+      "loss": 3.3608,
+      "step": 570
+    },
+    {
+      "epoch": 0.115,
+      "grad_norm": 0.5955671072006226,
+      "learning_rate": 0.00029749128048419006,
+      "loss": 3.2822,
+      "step": 575
+    },
+    {
+      "epoch": 0.116,
+      "grad_norm": 1.0059808492660522,
+      "learning_rate": 0.00029739915390274103,
+      "loss": 3.2733,
+      "step": 580
+    },
+    {
+      "epoch": 0.117,
+      "grad_norm": 0.589187502861023,
+      "learning_rate": 0.0002973053809089063,
+      "loss": 3.3818,
+      "step": 585
+    },
+    {
+      "epoch": 0.118,
+      "grad_norm": 0.5997525453567505,
+      "learning_rate": 0.0002972099625501073,
+      "loss": 3.3073,
+      "step": 590
+    },
+    {
+      "epoch": 0.119,
+      "grad_norm": 0.595435380935669,
+      "learning_rate": 0.0002971128998921436,
+      "loss": 3.2607,
+      "step": 595
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.6213010549545288,
+      "learning_rate": 0.0002970141940191814,
+      "loss": 3.2259,
+      "step": 600
+    },
+    {
+      "epoch": 0.121,
+      "grad_norm": 0.5790759325027466,
+      "learning_rate": 0.00029691384603374094,
+      "loss": 3.2976,
+      "step": 605
+    },
+    {
+      "epoch": 0.122,
+      "grad_norm": 0.5658262372016907,
+      "learning_rate": 0.0002968118570566847,
+      "loss": 3.2429,
+      "step": 610
+    },
+    {
+      "epoch": 0.123,
+      "grad_norm": 0.5907502174377441,
+      "learning_rate": 0.0002967082282272045,
+      "loss": 3.339,
+      "step": 615
+    },
+    {
+      "epoch": 0.124,
+      "grad_norm": 0.5808840990066528,
+      "learning_rate": 0.00029660296070280895,
+      "loss": 3.2019,
+      "step": 620
+    },
+    {
+      "epoch": 0.125,
+      "grad_norm": 0.5826807022094727,
+      "learning_rate": 0.00029649605565931043,
+      "loss": 3.2223,
+      "step": 625
+    },
+    {
+      "epoch": 0.126,
+      "grad_norm": 0.5372958779335022,
+      "learning_rate": 0.0002963875142908121,
+      "loss": 3.2082,
+      "step": 630
+    },
+    {
+      "epoch": 0.127,
+      "grad_norm": 0.5273796916007996,
+      "learning_rate": 0.0002962773378096944,
+      "loss": 3.2612,
+      "step": 635
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 0.5920493602752686,
+      "learning_rate": 0.0002961655274466015,
+      "loss": 3.1669,
+      "step": 640
+    },
+    {
+      "epoch": 0.129,
+      "grad_norm": 0.575118362903595,
+      "learning_rate": 0.0002960520844504279,
+      "loss": 3.1589,
+      "step": 645
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.5747135877609253,
+      "learning_rate": 0.0002959370100883039,
+      "loss": 3.1427,
+      "step": 650
+    },
+    {
+      "epoch": 0.131,
+      "grad_norm": 0.5302339792251587,
+      "learning_rate": 0.000295820305645582,
+      "loss": 3.139,
+      "step": 655
+    },
+    {
+      "epoch": 0.132,
+      "grad_norm": 0.5856277346611023,
+      "learning_rate": 0.00029570197242582204,
+      "loss": 3.1229,
+      "step": 660
+    },
+    {
+      "epoch": 0.133,
+      "grad_norm": 0.5387154817581177,
+      "learning_rate": 0.00029558201175077723,
+      "loss": 3.1793,
+      "step": 665
+    },
+    {
+      "epoch": 0.134,
+      "grad_norm": 0.519659161567688,
+      "learning_rate": 0.0002954604249603787,
+      "loss": 3.019,
+      "step": 670
+    },
+    {
+      "epoch": 0.135,
+      "grad_norm": 0.5343477725982666,
+      "learning_rate": 0.0002953372134127212,
+      "loss": 3.1136,
+      "step": 675
+    },
+    {
+      "epoch": 0.136,
+      "grad_norm": 0.5857408046722412,
+      "learning_rate": 0.0002952123784840475,
+      "loss": 3.1447,
+      "step": 680
+    },
+    {
+      "epoch": 0.137,
+      "grad_norm": 0.5493313670158386,
+      "learning_rate": 0.000295085921568733,
+      "loss": 3.159,
+      "step": 685
+    },
+    {
+      "epoch": 0.138,
+      "grad_norm": 0.5469555854797363,
+      "learning_rate": 0.0002949578440792705,
+      "loss": 3.0828,
+      "step": 690
+    },
+    {
+      "epoch": 0.139,
+      "grad_norm": 0.5394929647445679,
+      "learning_rate": 0.0002948281474462541,
+      "loss": 3.0478,
+      "step": 695
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.528149425983429,
+      "learning_rate": 0.0002946968331183634,
+      "loss": 3.1151,
+      "step": 700
+    },
+    {
+      "epoch": 0.141,
+      "grad_norm": 0.5426072478294373,
+      "learning_rate": 0.0002945639025623473,
+      "loss": 3.1065,
+      "step": 705
+    },
+    {
+      "epoch": 0.142,
+      "grad_norm": 0.5489263534545898,
+      "learning_rate": 0.00029442935726300734,
+      "loss": 3.0476,
+      "step": 710
+    },
+    {
+      "epoch": 0.143,
+      "grad_norm": 0.5520033836364746,
+      "learning_rate": 0.0002942931987231816,
+      "loss": 3.0378,
+      "step": 715
+    },
+    {
+      "epoch": 0.144,
+      "grad_norm": 0.5429459810256958,
+      "learning_rate": 0.0002941554284637276,
+      "loss": 3.0479,
+      "step": 720
+    },
+    {
+      "epoch": 0.145,
+      "grad_norm": 0.5990952253341675,
+      "learning_rate": 0.00029401604802350523,
+      "loss": 3.0617,
+      "step": 725
+    },
+    {
+      "epoch": 0.146,
+      "grad_norm": 0.5248637199401855,
+      "learning_rate": 0.0002938750589593599,
+      "loss": 3.0854,
+      "step": 730
+    },
+    {
+      "epoch": 0.147,
+      "grad_norm": 0.59062260389328,
+      "learning_rate": 0.0002937324628461048,
+      "loss": 3.0252,
+      "step": 735
+    },
+    {
+      "epoch": 0.148,
+      "grad_norm": 0.51813805103302,
+      "learning_rate": 0.0002935882612765035,
+      "loss": 3.0378,
+      "step": 740
+    },
+    {
+      "epoch": 0.149,
+      "grad_norm": 0.5538212656974792,
+      "learning_rate": 0.00029344245586125227,
+      "loss": 2.9998,
+      "step": 745
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.5211405158042908,
+      "learning_rate": 0.0002932950482289618,
+      "loss": 3.0235,
+      "step": 750
+    },
+    {
+      "epoch": 0.151,
+      "grad_norm": 0.5439658761024475,
+      "learning_rate": 0.00029314604002613913,
+      "loss": 3.0212,
+      "step": 755
+    },
+    {
+      "epoch": 0.152,
+      "grad_norm": 0.5444567799568176,
+      "learning_rate": 0.0002929954329171693,
+      "loss": 3.0763,
+      "step": 760
+    },
+    {
+      "epoch": 0.153,
+      "grad_norm": 0.5256481170654297,
+      "learning_rate": 0.0002928432285842969,
+      "loss": 3.0378,
+      "step": 765
+    },
+    {
+      "epoch": 0.154,
+      "grad_norm": 0.5187762379646301,
+      "learning_rate": 0.0002926894287276068,
+      "loss": 3.0012,
+      "step": 770
+    },
+    {
+      "epoch": 0.155,
+      "grad_norm": 0.5280289649963379,
+      "learning_rate": 0.0002925340350650058,
+      "loss": 2.9581,
+      "step": 775
+    },
+    {
+      "epoch": 0.156,
+      "grad_norm": 0.5345909595489502,
+      "learning_rate": 0.0002923770493322029,
+      "loss": 2.9386,
+      "step": 780
+    },
+    {
+      "epoch": 0.157,
+      "grad_norm": 0.5289936661720276,
+      "learning_rate": 0.00029221847328269034,
+      "loss": 2.9282,
+      "step": 785
+    },
+    {
+      "epoch": 0.158,
+      "grad_norm": 0.5417344570159912,
+      "learning_rate": 0.00029205830868772364,
+      "loss": 3.0096,
+      "step": 790
+    },
+    {
+      "epoch": 0.159,
+      "grad_norm": 0.5449649095535278,
+      "learning_rate": 0.00029189655733630193,
+      "loss": 2.9267,
+      "step": 795
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.5082869529724121,
+      "learning_rate": 0.0002917332210351482,
+      "loss": 2.9358,
+      "step": 800
+    },
+    {
+      "epoch": 0.161,
+      "grad_norm": 0.5396576523780823,
+      "learning_rate": 0.0002915683016086889,
+      "loss": 3.0027,
+      "step": 805
+    },
+    {
+      "epoch": 0.162,
+      "grad_norm": 0.5313181281089783,
+      "learning_rate": 0.0002914018008990335,
+      "loss": 2.9455,
+      "step": 810
+    },
+    {
+      "epoch": 0.163,
+      "grad_norm": 0.49431952834129333,
+      "learning_rate": 0.0002912337207659541,
+      "loss": 3.0227,
+      "step": 815
+    },
+    {
+      "epoch": 0.164,
+      "grad_norm": 0.535514235496521,
+      "learning_rate": 0.0002910640630868645,
+      "loss": 2.8925,
+      "step": 820
+    },
+    {
+      "epoch": 0.165,
+      "grad_norm": 0.5321778059005737,
+      "learning_rate": 0.00029089282975679946,
+      "loss": 2.898,
+      "step": 825
+    },
+    {
+      "epoch": 0.166,
+      "grad_norm": 0.5481770038604736,
+      "learning_rate": 0.00029072002268839316,
+      "loss": 3.0421,
+      "step": 830
+    },
+    {
+      "epoch": 0.167,
+      "grad_norm": 0.5205419659614563,
+      "learning_rate": 0.0002905456438118582,
+      "loss": 2.9817,
+      "step": 835
+    },
+    {
+      "epoch": 0.168,
+      "grad_norm": 0.519214391708374,
+      "learning_rate": 0.00029036969507496385,
+      "loss": 2.8693,
+      "step": 840
+    },
+    {
+      "epoch": 0.169,
+      "grad_norm": 0.5142780542373657,
+      "learning_rate": 0.00029019217844301437,
+      "loss": 2.9589,
+      "step": 845
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.5100776553153992,
+      "learning_rate": 0.0002900130958988269,
+      "loss": 2.9288,
+      "step": 850
+    },
+    {
+      "epoch": 0.171,
+      "grad_norm": 0.49259114265441895,
+      "learning_rate": 0.00028983244944270957,
+      "loss": 2.8171,
+      "step": 855
+    },
+    {
+      "epoch": 0.172,
+      "grad_norm": 0.5288705825805664,
+      "learning_rate": 0.000289650241092439,
+      "loss": 2.8789,
+      "step": 860
+    },
+    {
+      "epoch": 0.173,
+      "grad_norm": 0.5391842126846313,
+      "learning_rate": 0.00028946647288323766,
+      "loss": 2.915,
+      "step": 865
+    },
+    {
+      "epoch": 0.174,
+      "grad_norm": 0.5194993615150452,
+      "learning_rate": 0.00028928114686775153,
+      "loss": 2.8859,
+      "step": 870
+    },
+    {
+      "epoch": 0.175,
+      "grad_norm": 0.503380298614502,
+      "learning_rate": 0.00028909426511602657,
+      "loss": 2.9212,
+      "step": 875
+    },
+    {
+      "epoch": 0.176,
+      "grad_norm": 0.5249530076980591,
+      "learning_rate": 0.0002889058297154862,
+      "loss": 2.8832,
+      "step": 880
+    },
+    {
+      "epoch": 0.177,
+      "grad_norm": 0.4950558543205261,
+      "learning_rate": 0.0002887158427709075,
+      "loss": 2.8848,
+      "step": 885
+    },
+    {
+      "epoch": 0.178,
+      "grad_norm": 0.5167747735977173,
+      "learning_rate": 0.0002885243064043982,
+      "loss": 2.8871,
+      "step": 890
+    },
+    {
+      "epoch": 0.179,
+      "grad_norm": 0.49764567613601685,
+      "learning_rate": 0.0002883312227553724,
+      "loss": 2.8416,
+      "step": 895
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.5111367702484131,
+      "learning_rate": 0.00028813659398052707,
+      "loss": 2.8673,
+      "step": 900
+    },
+    {
+      "epoch": 0.181,
+      "grad_norm": 0.5087841749191284,
+      "learning_rate": 0.00028794042225381794,
+      "loss": 2.8911,
+      "step": 905
+    },
+    {
+      "epoch": 0.182,
+      "grad_norm": 0.5020551085472107,
+      "learning_rate": 0.00028774270976643503,
+      "loss": 2.8075,
+      "step": 910
+    },
+    {
+      "epoch": 0.183,
+      "grad_norm": 0.47799405455589294,
+      "learning_rate": 0.00028754345872677837,
+      "loss": 2.8803,
+      "step": 915
+    },
+    {
+      "epoch": 0.184,
+      "grad_norm": 0.4816206693649292,
+      "learning_rate": 0.0002873426713604331,
+      "loss": 2.7745,
+      "step": 920
+    },
+    {
+      "epoch": 0.185,
+      "grad_norm": 0.5181537866592407,
+      "learning_rate": 0.00028714034991014493,
+      "loss": 2.787,
+      "step": 925
+    },
+    {
+      "epoch": 0.186,
+      "grad_norm": 0.49121904373168945,
+      "learning_rate": 0.00028693649663579483,
+      "loss": 2.8193,
+      "step": 930
+    },
+    {
+      "epoch": 0.187,
+      "grad_norm": 0.49521416425704956,
+      "learning_rate": 0.0002867311138143737,
+      "loss": 2.767,
+      "step": 935
+    },
+    {
+      "epoch": 0.188,
+      "grad_norm": 0.47037652134895325,
+      "learning_rate": 0.00028652420373995744,
+      "loss": 2.8282,
+      "step": 940
+    },
+    {
+      "epoch": 0.189,
+      "grad_norm": 0.4956018328666687,
+      "learning_rate": 0.0002863157687236808,
+      "loss": 2.7311,
+      "step": 945
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.521686851978302,
+      "learning_rate": 0.0002861058110937116,
+      "loss": 2.7677,
+      "step": 950
+    },
+    {
+      "epoch": 0.191,
+      "grad_norm": 0.5066460371017456,
+      "learning_rate": 0.0002858943331952253,
+      "loss": 2.793,
+      "step": 955
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 0.4976024031639099,
+      "learning_rate": 0.00028568133739037793,
+      "loss": 2.776,
+      "step": 960
+    },
+    {
+      "epoch": 0.193,
+      "grad_norm": 0.4874838590621948,
+      "learning_rate": 0.00028546682605828054,
+      "loss": 2.8554,
+      "step": 965
+    },
+    {
+      "epoch": 0.194,
+      "grad_norm": 0.4944135546684265,
+      "learning_rate": 0.000285250801594972,
+      "loss": 2.8121,
+      "step": 970
+    },
+    {
+      "epoch": 0.195,
+      "grad_norm": 0.48091182112693787,
+      "learning_rate": 0.0002850332664133925,
+      "loss": 2.7394,
+      "step": 975
+    },
+    {
+      "epoch": 0.196,
+      "grad_norm": 0.46700772643089294,
+      "learning_rate": 0.0002848142229433568,
+      "loss": 2.6952,
+      "step": 980
+    },
+    {
+      "epoch": 0.197,
+      "grad_norm": 0.5188003182411194,
+      "learning_rate": 0.0002845936736315267,
+      "loss": 2.8399,
+      "step": 985
+    },
+    {
+      "epoch": 0.198,
+      "grad_norm": 0.5063126683235168,
+      "learning_rate": 0.00028437162094138366,
+      "loss": 2.7841,
+      "step": 990
+    },
+    {
+      "epoch": 0.199,
+      "grad_norm": 0.49878448247909546,
+      "learning_rate": 0.0002841480673532021,
+      "loss": 2.7651,
+      "step": 995
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.5015583634376526,
+      "learning_rate": 0.0002839230153640205,
+      "loss": 2.813,
+      "step": 1000
+    },
+    {
+      "epoch": 0.201,
+      "grad_norm": 0.5300015211105347,
+      "learning_rate": 0.00028369646748761443,
+      "loss": 2.7488,
+      "step": 1005
+    },
+    {
+      "epoch": 0.202,
+      "grad_norm": 0.4953711926937103,
+      "learning_rate": 0.0002834684262544682,
+      "loss": 2.7762,
+      "step": 1010
+    },
+    {
+      "epoch": 0.203,
+      "grad_norm": 0.49907276034355164,
+      "learning_rate": 0.0002832388942117462,
+      "loss": 2.7878,
+      "step": 1015
+    },
+    {
+      "epoch": 0.204,
+      "grad_norm": 0.4764520525932312,
+      "learning_rate": 0.0002830078739232653,
+      "loss": 2.8229,
+      "step": 1020
+    },
+    {
+      "epoch": 0.205,
+      "grad_norm": 0.46354764699935913,
+      "learning_rate": 0.00028277536796946526,
+      "loss": 2.7619,
+      "step": 1025
+    },
+    {
+      "epoch": 0.206,
+      "grad_norm": 0.463844895362854,
+      "learning_rate": 0.0002825413789473806,
+      "loss": 2.6742,
+      "step": 1030
+    },
+    {
+      "epoch": 0.207,
+      "grad_norm": 0.5160843729972839,
+      "learning_rate": 0.0002823059094706113,
+      "loss": 2.7154,
+      "step": 1035
+    },
+    {
+      "epoch": 0.208,
+      "grad_norm": 0.47835180163383484,
+      "learning_rate": 0.00028206896216929363,
+      "loss": 2.733,
+      "step": 1040
+    },
+    {
+      "epoch": 0.209,
+      "grad_norm": 0.5362768769264221,
+      "learning_rate": 0.0002818305396900708,
+      "loss": 2.7082,
+      "step": 1045
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.4910009503364563,
+      "learning_rate": 0.0002815906446960635,
+      "loss": 2.693,
+      "step": 1050
+    },
+    {
+      "epoch": 0.211,
+      "grad_norm": 0.5005692839622498,
+      "learning_rate": 0.0002813492798668398,
+      "loss": 2.793,
+      "step": 1055
+    },
+    {
+      "epoch": 0.212,
+      "grad_norm": 0.47488823533058167,
+      "learning_rate": 0.0002811064478983857,
+      "loss": 2.7746,
+      "step": 1060
+    },
+    {
+      "epoch": 0.213,
+      "grad_norm": 0.4562411606311798,
+      "learning_rate": 0.00028086215150307473,
+      "loss": 2.6761,
+      "step": 1065
+    },
+    {
+      "epoch": 0.214,
+      "grad_norm": 0.47498100996017456,
+      "learning_rate": 0.00028061639340963763,
+      "loss": 2.6934,
+      "step": 1070
+    },
+    {
+      "epoch": 0.215,
+      "grad_norm": 0.4689179062843323,
+      "learning_rate": 0.0002803691763631321,
+      "loss": 2.6655,
+      "step": 1075
+    },
+    {
+      "epoch": 0.216,
+      "grad_norm": 0.46847668290138245,
+      "learning_rate": 0.0002801205031249118,
+      "loss": 2.7459,
+      "step": 1080
+    },
+    {
+      "epoch": 0.217,
+      "grad_norm": 0.4786347448825836,
+      "learning_rate": 0.00027987037647259587,
+      "loss": 2.7196,
+      "step": 1085
+    },
+    {
+      "epoch": 0.218,
+      "grad_norm": 0.48441487550735474,
+      "learning_rate": 0.00027961879920003767,
+      "loss": 2.7219,
+      "step": 1090
+    },
+    {
+      "epoch": 0.219,
+      "grad_norm": 0.5033813714981079,
+      "learning_rate": 0.0002793657741172935,
+      "loss": 2.7095,
+      "step": 1095
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.4371601641178131,
+      "learning_rate": 0.0002791113040505915,
+      "loss": 2.6165,
+      "step": 1100
+    },
+    {
+      "epoch": 0.221,
+      "grad_norm": 0.47073352336883545,
+      "learning_rate": 0.0002788553918422999,
+      "loss": 2.6581,
+      "step": 1105
+    },
+    {
+      "epoch": 0.222,
+      "grad_norm": 0.5119361281394958,
+      "learning_rate": 0.00027859804035089533,
+      "loss": 2.7874,
+      "step": 1110
+    },
+    {
+      "epoch": 0.223,
+      "grad_norm": 0.45070821046829224,
+      "learning_rate": 0.0002783392524509307,
+      "loss": 2.6972,
+      "step": 1115
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 0.45409971475601196,
+      "learning_rate": 0.0002780790310330035,
+      "loss": 2.6349,
+      "step": 1120
+    },
+    {
+      "epoch": 0.225,
+      "grad_norm": 0.46642905473709106,
+      "learning_rate": 0.00027781737900372296,
+      "loss": 2.6704,
+      "step": 1125
+    },
+    {
+      "epoch": 0.226,
+      "grad_norm": 0.46489498019218445,
+      "learning_rate": 0.0002775542992856781,
+      "loss": 2.7102,
+      "step": 1130
+    },
+    {
+      "epoch": 0.227,
+      "grad_norm": 0.9031617641448975,
+      "learning_rate": 0.000277289794817405,
+      "loss": 2.6664,
+      "step": 1135
+    },
+    {
+      "epoch": 0.228,
+      "grad_norm": 0.4736745357513428,
+      "learning_rate": 0.0002770238685533534,
+      "loss": 2.7807,
+      "step": 1140
+    },
+    {
+      "epoch": 0.229,
+      "grad_norm": 0.4746663272380829,
+      "learning_rate": 0.0002767565234638546,
+      "loss": 2.7733,
+      "step": 1145
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.4613768756389618,
+      "learning_rate": 0.0002764877625350876,
+      "loss": 2.6382,
+      "step": 1150
+    },
+    {
+      "epoch": 0.231,
+      "grad_norm": 0.45333442091941833,
+      "learning_rate": 0.0002762175887690461,
+      "loss": 2.6333,
+      "step": 1155
+    },
+    {
+      "epoch": 0.232,
+      "grad_norm": 0.48350557684898376,
+      "learning_rate": 0.0002759460051835048,
+      "loss": 2.7099,
+      "step": 1160
+    },
+    {
+      "epoch": 0.233,
+      "grad_norm": 0.46377626061439514,
+      "learning_rate": 0.0002756730148119855,
+      "loss": 2.673,
+      "step": 1165
+    },
+    {
+      "epoch": 0.234,
+      "grad_norm": 0.4856124818325043,
+      "learning_rate": 0.00027539862070372393,
+      "loss": 2.7564,
+      "step": 1170
+    },
+    {
+      "epoch": 0.235,
+      "grad_norm": 0.48508885502815247,
+      "learning_rate": 0.0002751228259236348,
+      "loss": 2.6935,
+      "step": 1175
+    },
+    {
+      "epoch": 0.236,
+      "grad_norm": 0.4603351652622223,
+      "learning_rate": 0.00027484563355227826,
+      "loss": 2.6816,
+      "step": 1180
+    },
+    {
+      "epoch": 0.237,
+      "grad_norm": 0.4790993928909302,
+      "learning_rate": 0.000274567046685825,
+      "loss": 2.7326,
+      "step": 1185
+    },
+    {
+      "epoch": 0.238,
+      "grad_norm": 0.4396674335002899,
+      "learning_rate": 0.000274287068436022,
+      "loss": 2.6504,
+      "step": 1190
+    },
+    {
+      "epoch": 0.239,
+      "grad_norm": 0.4911787211894989,
+      "learning_rate": 0.0002740057019301576,
+      "loss": 2.7139,
+      "step": 1195
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.45822376012802124,
+      "learning_rate": 0.0002737229503110267,
+      "loss": 2.6546,
+      "step": 1200
+    },
+    {
+      "epoch": 0.241,
+      "grad_norm": 0.44974634051322937,
+      "learning_rate": 0.0002734388167368954,
+      "loss": 2.6673,
+      "step": 1205
+    },
+    {
+      "epoch": 0.242,
+      "grad_norm": 0.45818230509757996,
+      "learning_rate": 0.0002731533043814661,
+      "loss": 2.7511,
+      "step": 1210
+    },
+    {
+      "epoch": 0.243,
+      "grad_norm": 0.45645296573638916,
+      "learning_rate": 0.0002728664164338418,
+      "loss": 2.6447,
+      "step": 1215
+    },
+    {
+      "epoch": 0.244,
+      "grad_norm": 0.46149763464927673,
+      "learning_rate": 0.0002725781560984904,
+      "loss": 2.6297,
+      "step": 1220
+    },
+    {
+      "epoch": 0.245,
+      "grad_norm": 0.47193628549575806,
+      "learning_rate": 0.00027228852659520915,
+      "loss": 2.6723,
+      "step": 1225
+    },
+    {
+      "epoch": 0.246,
+      "grad_norm": 0.47553935647010803,
+      "learning_rate": 0.0002719975311590885,
+      "loss": 2.6032,
+      "step": 1230
+    },
+    {
+      "epoch": 0.247,
+      "grad_norm": 0.4629911482334137,
+      "learning_rate": 0.0002717051730404762,
+      "loss": 2.6566,
+      "step": 1235
+    },
+    {
+      "epoch": 0.248,
+      "grad_norm": 0.46325111389160156,
+      "learning_rate": 0.00027141145550494056,
+      "loss": 2.5453,
+      "step": 1240
+    },
+    {
+      "epoch": 0.249,
+      "grad_norm": 0.46191033720970154,
+      "learning_rate": 0.00027111638183323453,
+      "loss": 2.5978,
+      "step": 1245
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.45227476954460144,
+      "learning_rate": 0.00027081995532125857,
+      "loss": 2.6556,
+      "step": 1250
+    },
+    {
+      "epoch": 0.251,
+      "grad_norm": 0.4406815767288208,
+      "learning_rate": 0.000270522179280024,
+      "loss": 2.6529,
+      "step": 1255
+    },
+    {
+      "epoch": 0.252,
+      "grad_norm": 0.4414892792701721,
+      "learning_rate": 0.0002702230570356163,
+      "loss": 2.6808,
+      "step": 1260
+    },
+    {
+      "epoch": 0.253,
+      "grad_norm": 0.45952120423316956,
+      "learning_rate": 0.00026992259192915746,
+      "loss": 2.5599,
+      "step": 1265
+    },
+    {
+      "epoch": 0.254,
+      "grad_norm": 0.45652300119400024,
+      "learning_rate": 0.000269620787316769,
+      "loss": 2.61,
+      "step": 1270
+    },
+    {
+      "epoch": 0.255,
+      "grad_norm": 0.4613508880138397,
+      "learning_rate": 0.00026931764656953444,
+      "loss": 2.7847,
+      "step": 1275
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 0.4642449915409088,
+      "learning_rate": 0.0002690131730734615,
+      "loss": 2.6241,
+      "step": 1280
+    },
+    {
+      "epoch": 0.257,
+      "grad_norm": 0.6733850240707397,
+      "learning_rate": 0.0002687073702294445,
+      "loss": 2.605,
+      "step": 1285
+    },
+    {
+      "epoch": 0.258,
+      "grad_norm": 0.4731089174747467,
+      "learning_rate": 0.0002684002414532261,
+      "loss": 2.6458,
+      "step": 1290
+    },
+    {
+      "epoch": 0.259,
+      "grad_norm": 0.5482076406478882,
+      "learning_rate": 0.0002680917901753594,
+      "loss": 2.5853,
+      "step": 1295
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.4559101164340973,
+      "learning_rate": 0.0002677820198411696,
+      "loss": 2.5809,
+      "step": 1300
+    },
+    {
+      "epoch": 0.261,
+      "grad_norm": 0.480686753988266,
+      "learning_rate": 0.0002674709339107152,
+      "loss": 2.6749,
+      "step": 1305
+    },
+    {
+      "epoch": 0.262,
+      "grad_norm": 0.42668047547340393,
+      "learning_rate": 0.00026715853585874977,
+      "loss": 2.6271,
+      "step": 1310
+    },
+    {
+      "epoch": 0.263,
+      "grad_norm": 0.4340451657772064,
+      "learning_rate": 0.00026684482917468293,
+      "loss": 2.6019,
+      "step": 1315
+    },
+    {
+      "epoch": 0.264,
+      "grad_norm": 0.4473606050014496,
+      "learning_rate": 0.00026652981736254125,
+      "loss": 2.5944,
+      "step": 1320
+    },
+    {
+      "epoch": 0.265,
+      "grad_norm": 0.4584132134914398,
+      "learning_rate": 0.0002662135039409296,
+      "loss": 2.6534,
+      "step": 1325
+    },
+    {
+      "epoch": 0.266,
+      "grad_norm": 0.43533068895339966,
+      "learning_rate": 0.00026589589244299113,
+      "loss": 2.6027,
+      "step": 1330
+    },
+    {
+      "epoch": 0.267,
+      "grad_norm": 0.456048846244812,
+      "learning_rate": 0.00026557698641636835,
+      "loss": 2.5276,
+      "step": 1335
+    },
+    {
+      "epoch": 0.268,
+      "grad_norm": 0.441222608089447,
+      "learning_rate": 0.0002652567894231634,
+      "loss": 2.5891,
+      "step": 1340
+    },
+    {
+      "epoch": 0.269,
+      "grad_norm": 0.4482981860637665,
+      "learning_rate": 0.0002649353050398982,
+      "loss": 2.7258,
+      "step": 1345
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.4718676805496216,
+      "learning_rate": 0.0002646125368574743,
+      "loss": 2.6594,
+      "step": 1350
+    },
+    {
+      "epoch": 0.271,
+      "grad_norm": 0.44714292883872986,
+      "learning_rate": 0.00026428848848113315,
+      "loss": 2.5474,
+      "step": 1355
+    },
+    {
+      "epoch": 0.272,
+      "grad_norm": 0.42453283071517944,
+      "learning_rate": 0.00026396316353041564,
+      "loss": 2.5412,
+      "step": 1360
+    },
+    {
+      "epoch": 0.273,
+      "grad_norm": 0.42612630128860474,
+      "learning_rate": 0.0002636365656391216,
+      "loss": 2.6426,
+      "step": 1365
+    },
+    {
+      "epoch": 0.274,
+      "grad_norm": 0.45025870203971863,
+      "learning_rate": 0.00026330869845526944,
+      "loss": 2.6238,
+      "step": 1370
+    },
+    {
+      "epoch": 0.275,
+      "grad_norm": 0.46621114015579224,
+      "learning_rate": 0.0002629795656410551,
+      "loss": 2.6543,
+      "step": 1375
+    },
+    {
+      "epoch": 0.276,
+      "grad_norm": 0.4684096872806549,
+      "learning_rate": 0.00026264917087281134,
+      "loss": 2.5498,
+      "step": 1380
+    },
+    {
+      "epoch": 0.277,
+      "grad_norm": 0.444062739610672,
+      "learning_rate": 0.0002623175178409667,
+      "loss": 2.5517,
+      "step": 1385
+    },
+    {
+      "epoch": 0.278,
+      "grad_norm": 0.46531593799591064,
+      "learning_rate": 0.00026198461025000417,
+      "loss": 2.591,
+      "step": 1390
+    },
+    {
+      "epoch": 0.279,
+      "grad_norm": 0.43859636783599854,
+      "learning_rate": 0.0002616504518184199,
+      "loss": 2.6114,
+      "step": 1395
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.4471001923084259,
+      "learning_rate": 0.0002613150462786815,
+      "loss": 2.5127,
+      "step": 1400
+    },
+    {
+      "epoch": 0.281,
+      "grad_norm": 0.4279544949531555,
+      "learning_rate": 0.00026097839737718664,
+      "loss": 2.6325,
+      "step": 1405
+    },
+    {
+      "epoch": 0.282,
+      "grad_norm": 0.45175451040267944,
+      "learning_rate": 0.0002606405088742209,
+      "loss": 2.5789,
+      "step": 1410
+    },
+    {
+      "epoch": 0.283,
+      "grad_norm": 0.4278489351272583,
+      "learning_rate": 0.00026030138454391597,
+      "loss": 2.4853,
+      "step": 1415
+    },
+    {
+      "epoch": 0.284,
+      "grad_norm": 0.48170074820518494,
+      "learning_rate": 0.00025996102817420745,
+      "loss": 2.5067,
+      "step": 1420
+    },
+    {
+      "epoch": 0.285,
+      "grad_norm": 0.4388919770717621,
+      "learning_rate": 0.0002596194435667925,
+      "loss": 2.4981,
+      "step": 1425
+    },
+    {
+      "epoch": 0.286,
+      "grad_norm": 0.4510611593723297,
+      "learning_rate": 0.00025927663453708733,
+      "loss": 2.6489,
+      "step": 1430
+    },
+    {
+      "epoch": 0.287,
+      "grad_norm": 0.43777748942375183,
+      "learning_rate": 0.0002589326049141847,
+      "loss": 2.5824,
+      "step": 1435
+    },
+    {
+      "epoch": 0.288,
+      "grad_norm": 0.43350744247436523,
+      "learning_rate": 0.0002585873585408111,
+      "loss": 2.6057,
+      "step": 1440
+    },
+    {
+      "epoch": 0.289,
+      "grad_norm": 0.42946937680244446,
+      "learning_rate": 0.00025824089927328384,
+      "loss": 2.5705,
+      "step": 1445
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.4467051029205322,
+      "learning_rate": 0.00025789323098146787,
+      "loss": 2.6092,
+      "step": 1450
+    },
+    {
+      "epoch": 0.291,
+      "grad_norm": 0.4496210515499115,
+      "learning_rate": 0.0002575443575487328,
+      "loss": 2.4654,
+      "step": 1455
+    },
+    {
+      "epoch": 0.292,
+      "grad_norm": 0.43781545758247375,
+      "learning_rate": 0.0002571942828719092,
+      "loss": 2.4598,
+      "step": 1460
+    },
+    {
+      "epoch": 0.293,
+      "grad_norm": 0.430301308631897,
+      "learning_rate": 0.0002568430108612454,
+      "loss": 2.5123,
+      "step": 1465
+    },
+    {
+      "epoch": 0.294,
+      "grad_norm": 0.42806944251060486,
+      "learning_rate": 0.00025649054544036356,
+      "loss": 2.5518,
+      "step": 1470
+    },
+    {
+      "epoch": 0.295,
+      "grad_norm": 0.4417232573032379,
+      "learning_rate": 0.0002561368905462159,
+      "loss": 2.5158,
+      "step": 1475
+    },
+    {
+      "epoch": 0.296,
+      "grad_norm": 0.4251876473426819,
+      "learning_rate": 0.00025578205012904086,
+      "loss": 2.5811,
+      "step": 1480
+    },
+    {
+      "epoch": 0.297,
+      "grad_norm": 0.45378217101097107,
+      "learning_rate": 0.0002554260281523188,
+      "loss": 2.4669,
+      "step": 1485
+    },
+    {
+      "epoch": 0.298,
+      "grad_norm": 0.42655450105667114,
+      "learning_rate": 0.00025506882859272796,
+      "loss": 2.4976,
+      "step": 1490
+    },
+    {
+      "epoch": 0.299,
+      "grad_norm": 0.44121378660202026,
+      "learning_rate": 0.00025471045544009965,
+      "loss": 2.5124,
+      "step": 1495
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.4258913993835449,
+      "learning_rate": 0.00025435091269737414,
+      "loss": 2.5127,
+      "step": 1500
+    },
+    {
+      "epoch": 0.301,
+      "grad_norm": 0.46033087372779846,
+      "learning_rate": 0.0002539902043805556,
+      "loss": 2.4782,
+      "step": 1505
+    },
+    {
+      "epoch": 0.302,
+      "grad_norm": 0.4258130192756653,
+      "learning_rate": 0.00025362833451866753,
+      "loss": 2.4894,
+      "step": 1510
+    },
+    {
+      "epoch": 0.303,
+      "grad_norm": 0.44221916794776917,
+      "learning_rate": 0.00025326530715370744,
+      "loss": 2.4741,
+      "step": 1515
+    },
+    {
+      "epoch": 0.304,
+      "grad_norm": 0.42666810750961304,
+      "learning_rate": 0.00025290112634060186,
+      "loss": 2.5112,
+      "step": 1520
+    },
+    {
+      "epoch": 0.305,
+      "grad_norm": 0.4374025762081146,
+      "learning_rate": 0.0002525357961471613,
+      "loss": 2.5195,
+      "step": 1525
+    },
+    {
+      "epoch": 0.306,
+      "grad_norm": 0.4391957223415375,
+      "learning_rate": 0.00025216932065403425,
+      "loss": 2.4956,
+      "step": 1530
+    },
+    {
+      "epoch": 0.307,
+      "grad_norm": 0.43732672929763794,
+      "learning_rate": 0.00025180170395466204,
+      "loss": 2.4465,
+      "step": 1535
+    },
+    {
+      "epoch": 0.308,
+      "grad_norm": 0.45370855927467346,
+      "learning_rate": 0.0002514329501552331,
+      "loss": 2.5087,
+      "step": 1540
+    },
+    {
+      "epoch": 0.309,
+      "grad_norm": 0.4213542342185974,
+      "learning_rate": 0.00025106306337463686,
+      "loss": 2.4634,
+      "step": 1545
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.4372135102748871,
+      "learning_rate": 0.0002506920477444179,
+      "loss": 2.5841,
+      "step": 1550
+    },
+    {
+      "epoch": 0.311,
+      "grad_norm": 0.4450632631778717,
+      "learning_rate": 0.0002503199074087298,
+      "loss": 2.4903,
+      "step": 1555
+    },
+    {
+      "epoch": 0.312,
+      "grad_norm": 0.4359963536262512,
+      "learning_rate": 0.00024994664652428877,
+      "loss": 2.4622,
+      "step": 1560
+    },
+    {
+      "epoch": 0.313,
+      "grad_norm": 0.4272458255290985,
+      "learning_rate": 0.0002495722692603274,
+      "loss": 2.5966,
+      "step": 1565
+    },
+    {
+      "epoch": 0.314,
+      "grad_norm": 0.430254191160202,
+      "learning_rate": 0.00024919677979854776,
+      "loss": 2.4504,
+      "step": 1570
+    },
+    {
+      "epoch": 0.315,
+      "grad_norm": 0.41149428486824036,
+      "learning_rate": 0.0002488201823330751,
+      "loss": 2.5111,
+      "step": 1575
+    },
+    {
+      "epoch": 0.316,
+      "grad_norm": 0.4507811665534973,
+      "learning_rate": 0.0002484424810704107,
+      "loss": 2.4763,
+      "step": 1580
+    },
+    {
+      "epoch": 0.317,
+      "grad_norm": 0.43168288469314575,
+      "learning_rate": 0.00024806368022938495,
+      "loss": 2.5124,
+      "step": 1585
+    },
+    {
+      "epoch": 0.318,
+      "grad_norm": 0.44648832082748413,
+      "learning_rate": 0.0002476837840411103,
+      "loss": 2.46,
+      "step": 1590
+    },
+    {
+      "epoch": 0.319,
+      "grad_norm": 0.4257158041000366,
+      "learning_rate": 0.0002473027967489341,
+      "loss": 2.5119,
+      "step": 1595
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.4564734101295471,
+      "learning_rate": 0.0002469207226083908,
+      "loss": 2.5977,
+      "step": 1600
+    },
+    {
+      "epoch": 0.321,
+      "grad_norm": 0.46141913533210754,
+      "learning_rate": 0.00024653756588715486,
+      "loss": 2.611,
+      "step": 1605
+    },
+    {
+      "epoch": 0.322,
+      "grad_norm": 0.44771572947502136,
+      "learning_rate": 0.0002461533308649929,
+      "loss": 2.4669,
+      "step": 1610
+    },
+    {
+      "epoch": 0.323,
+      "grad_norm": 0.43828845024108887,
+      "learning_rate": 0.0002457680218337157,
+      "loss": 2.433,
+      "step": 1615
+    },
+    {
+      "epoch": 0.324,
+      "grad_norm": 0.4548298120498657,
+      "learning_rate": 0.0002453816430971307,
+      "loss": 2.4799,
+      "step": 1620
+    },
+    {
+      "epoch": 0.325,
+      "grad_norm": 0.4636451303958893,
+      "learning_rate": 0.0002449941989709936,
+      "loss": 2.5127,
+      "step": 1625
+    },
+    {
+      "epoch": 0.326,
+      "grad_norm": 0.422651082277298,
+      "learning_rate": 0.0002446056937829603,
+      "loss": 2.539,
+      "step": 1630
+    },
+    {
+      "epoch": 0.327,
+      "grad_norm": 0.45776715874671936,
+      "learning_rate": 0.00024421613187253823,
+      "loss": 2.5262,
+      "step": 1635
+    },
+    {
+      "epoch": 0.328,
+      "grad_norm": 0.43375980854034424,
+      "learning_rate": 0.00024382551759103853,
+      "loss": 2.533,
+      "step": 1640
+    },
+    {
+      "epoch": 0.329,
+      "grad_norm": 0.4225177764892578,
+      "learning_rate": 0.00024343385530152683,
+      "loss": 2.5418,
+      "step": 1645
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.42413005232810974,
+      "learning_rate": 0.00024304114937877486,
+      "loss": 2.3906,
+      "step": 1650
+    },
+    {
+      "epoch": 0.331,
+      "grad_norm": 0.4245309829711914,
+      "learning_rate": 0.00024264740420921132,
+      "loss": 2.4826,
+      "step": 1655
+    },
+    {
+      "epoch": 0.332,
+      "grad_norm": 0.45272067189216614,
+      "learning_rate": 0.00024225262419087323,
+      "loss": 2.4663,
+      "step": 1660
+    },
+    {
+      "epoch": 0.333,
+      "grad_norm": 0.43966102600097656,
+      "learning_rate": 0.00024185681373335656,
+      "loss": 2.3933,
+      "step": 1665
+    },
+    {
+      "epoch": 0.334,
+      "grad_norm": 0.42622506618499756,
+      "learning_rate": 0.00024145997725776697,
+      "loss": 2.4733,
+      "step": 1670
+    },
+    {
+      "epoch": 0.335,
+      "grad_norm": 0.44191116094589233,
+      "learning_rate": 0.0002410621191966705,
+      "loss": 2.5164,
+      "step": 1675
+    },
+    {
+      "epoch": 0.336,
+      "grad_norm": 0.40515631437301636,
+      "learning_rate": 0.00024066324399404416,
+      "loss": 2.5272,
+      "step": 1680
+    },
+    {
+      "epoch": 0.337,
+      "grad_norm": 0.42186275124549866,
+      "learning_rate": 0.00024026335610522604,
+      "loss": 2.4699,
+      "step": 1685
+    },
+    {
+      "epoch": 0.338,
+      "grad_norm": 0.4451305568218231,
+      "learning_rate": 0.0002398624599968658,
+      "loss": 2.4666,
+      "step": 1690
+    },
+    {
+      "epoch": 0.339,
+      "grad_norm": 0.4346270263195038,
+      "learning_rate": 0.00023946056014687454,
+      "loss": 2.5605,
+      "step": 1695
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.42730483412742615,
+      "learning_rate": 0.00023905766104437504,
+      "loss": 2.4154,
+      "step": 1700
+    },
+    {
+      "epoch": 0.341,
+      "grad_norm": 0.41405901312828064,
+      "learning_rate": 0.0002386537671896514,
+      "loss": 2.3792,
+      "step": 1705
+    },
+    {
+      "epoch": 0.342,
+      "grad_norm": 0.42070284485816956,
+      "learning_rate": 0.0002382488830940989,
+      "loss": 2.4676,
+      "step": 1710
+    },
+    {
+      "epoch": 0.343,
+      "grad_norm": 0.4284003973007202,
+      "learning_rate": 0.00023784301328017354,
+      "loss": 2.433,
+      "step": 1715
+    },
+    {
+      "epoch": 0.344,
+      "grad_norm": 0.4249774217605591,
+      "learning_rate": 0.00023743616228134154,
+      "loss": 2.4119,
+      "step": 1720
+    },
+    {
+      "epoch": 0.345,
+      "grad_norm": 0.4497978389263153,
+      "learning_rate": 0.00023702833464202882,
+      "loss": 2.4146,
+      "step": 1725
+    },
+    {
+      "epoch": 0.346,
+      "grad_norm": 0.421589732170105,
+      "learning_rate": 0.00023661953491756996,
+      "loss": 2.4051,
+      "step": 1730
+    },
+    {
+      "epoch": 0.347,
+      "grad_norm": 0.44604167342185974,
+      "learning_rate": 0.00023620976767415764,
+      "loss": 2.519,
+      "step": 1735
+    },
+    {
+      "epoch": 0.348,
+      "grad_norm": 0.414231538772583,
+      "learning_rate": 0.00023579903748879145,
+      "loss": 2.4193,
+      "step": 1740
+    },
+    {
+      "epoch": 0.349,
+      "grad_norm": 0.4184156060218811,
+      "learning_rate": 0.00023538734894922673,
+      "loss": 2.4478,
+      "step": 1745
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.43089163303375244,
+      "learning_rate": 0.0002349747066539235,
+      "loss": 2.3942,
+      "step": 1750
+    },
+    {
+      "epoch": 0.351,
+      "grad_norm": 0.4269482493400574,
+      "learning_rate": 0.00023456111521199494,
+      "loss": 2.5309,
+      "step": 1755
+    },
+    {
+      "epoch": 0.352,
+      "grad_norm": 0.4043903350830078,
+      "learning_rate": 0.00023414657924315598,
+      "loss": 2.4896,
+      "step": 1760
+    },
+    {
+      "epoch": 0.353,
+      "grad_norm": 0.4351564347743988,
+      "learning_rate": 0.00023373110337767175,
+      "loss": 2.3659,
+      "step": 1765
+    },
+    {
+      "epoch": 0.354,
+      "grad_norm": 0.41638249158859253,
+      "learning_rate": 0.00023331469225630567,
+      "loss": 2.4415,
+      "step": 1770
+    },
+    {
+      "epoch": 0.355,
+      "grad_norm": 0.41405797004699707,
+      "learning_rate": 0.00023289735053026785,
+      "loss": 2.4004,
+      "step": 1775
+    },
+    {
+      "epoch": 0.356,
+      "grad_norm": 0.43309521675109863,
+      "learning_rate": 0.00023247908286116287,
+      "loss": 2.4222,
+      "step": 1780
+    },
+    {
+      "epoch": 0.357,
+      "grad_norm": 0.43367326259613037,
+      "learning_rate": 0.00023205989392093812,
+      "loss": 2.4417,
+      "step": 1785
+    },
+    {
+      "epoch": 0.358,
+      "grad_norm": 0.4391366243362427,
+      "learning_rate": 0.00023163978839183113,
+      "loss": 2.4042,
+      "step": 1790
+    },
+    {
+      "epoch": 0.359,
+      "grad_norm": 0.4290434718132019,
+      "learning_rate": 0.0002312187709663176,
+      "loss": 2.4001,
+      "step": 1795
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.4561871588230133,
+      "learning_rate": 0.0002307968463470589,
+      "loss": 2.4138,
+      "step": 1800
+    },
+    {
+      "epoch": 0.361,
+      "grad_norm": 0.4181992709636688,
+      "learning_rate": 0.00023037401924684946,
+      "loss": 2.3321,
+      "step": 1805
+    },
+    {
+      "epoch": 0.362,
+      "grad_norm": 0.4237087666988373,
+      "learning_rate": 0.00022995029438856437,
+      "loss": 2.4286,
+      "step": 1810
+    },
+    {
+      "epoch": 0.363,
+      "grad_norm": 0.4123045802116394,
+      "learning_rate": 0.00022952567650510617,
+      "loss": 2.4236,
+      "step": 1815
+    },
+    {
+      "epoch": 0.364,
+      "grad_norm": 0.41121742129325867,
+      "learning_rate": 0.00022910017033935258,
+      "loss": 2.4475,
+      "step": 1820
+    },
+    {
+      "epoch": 0.365,
+      "grad_norm": 0.4109913408756256,
+      "learning_rate": 0.00022867378064410303,
+      "loss": 2.4107,
+      "step": 1825
+    },
+    {
+      "epoch": 0.366,
+      "grad_norm": 0.4502372145652771,
+      "learning_rate": 0.00022824651218202578,
+      "loss": 2.4689,
+      "step": 1830
+    },
+    {
+      "epoch": 0.367,
+      "grad_norm": 0.4249553084373474,
+      "learning_rate": 0.00022781836972560473,
+      "loss": 2.4354,
+      "step": 1835
+    },
+    {
+      "epoch": 0.368,
+      "grad_norm": 0.4587641954421997,
+      "learning_rate": 0.0002273893580570861,
+      "loss": 2.4022,
+      "step": 1840
+    },
+    {
+      "epoch": 0.369,
+      "grad_norm": 0.4180518388748169,
+      "learning_rate": 0.000226959481968425,
+      "loss": 2.4303,
+      "step": 1845
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.4448012113571167,
+      "learning_rate": 0.0002265287462612318,
+      "loss": 2.3981,
+      "step": 1850
+    },
+    {
+      "epoch": 0.371,
+      "grad_norm": 0.4289359450340271,
+      "learning_rate": 0.0002260971557467187,
+      "loss": 2.3435,
+      "step": 1855
+    },
+    {
+      "epoch": 0.372,
+      "grad_norm": 0.42983385920524597,
+      "learning_rate": 0.0002256647152456459,
+      "loss": 2.349,
+      "step": 1860
+    },
+    {
+      "epoch": 0.373,
+      "grad_norm": 0.4299187660217285,
+      "learning_rate": 0.00022523142958826763,
+      "loss": 2.396,
+      "step": 1865
+    },
+    {
+      "epoch": 0.374,
+      "grad_norm": 0.42625340819358826,
+      "learning_rate": 0.00022479730361427844,
+      "loss": 2.2904,
+      "step": 1870
+    },
+    {
+      "epoch": 0.375,
+      "grad_norm": 0.4453998804092407,
+      "learning_rate": 0.00022436234217275888,
+      "loss": 2.5129,
+      "step": 1875
+    },
+    {
+      "epoch": 0.376,
+      "grad_norm": 0.4248093366622925,
+      "learning_rate": 0.00022392655012212163,
+      "loss": 2.3976,
+      "step": 1880
+    },
+    {
+      "epoch": 0.377,
+      "grad_norm": 0.42273494601249695,
+      "learning_rate": 0.0002234899323300569,
+      "loss": 2.4843,
+      "step": 1885
+    },
+    {
+      "epoch": 0.378,
+      "grad_norm": 0.40302085876464844,
+      "learning_rate": 0.00022305249367347835,
+      "loss": 2.4268,
+      "step": 1890
+    },
+    {
+      "epoch": 0.379,
+      "grad_norm": 0.4263085126876831,
+      "learning_rate": 0.00022261423903846846,
+      "loss": 2.4438,
+      "step": 1895
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.4482550621032715,
+      "learning_rate": 0.00022217517332022386,
+      "loss": 2.4004,
+      "step": 1900
+    },
+    {
+      "epoch": 0.381,
+      "grad_norm": 0.42222732305526733,
+      "learning_rate": 0.0002217353014230011,
+      "loss": 2.4649,
+      "step": 1905
+    },
+    {
+      "epoch": 0.382,
+      "grad_norm": 0.44026628136634827,
+      "learning_rate": 0.00022129462826006116,
+      "loss": 2.4186,
+      "step": 1910
+    },
+    {
+      "epoch": 0.383,
+      "grad_norm": 0.4032437801361084,
+      "learning_rate": 0.0002208531587536153,
+      "loss": 2.388,
+      "step": 1915
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 0.42311862111091614,
+      "learning_rate": 0.00022041089783476954,
+      "loss": 2.4084,
+      "step": 1920
+    },
+    {
+      "epoch": 0.385,
+      "grad_norm": 0.4101983904838562,
+      "learning_rate": 0.00021996785044346983,
+      "loss": 2.4068,
+      "step": 1925
+    },
+    {
+      "epoch": 0.386,
+      "grad_norm": 0.42865946888923645,
+      "learning_rate": 0.0002195240215284468,
+      "loss": 2.3719,
+      "step": 1930
+    },
+    {
+      "epoch": 0.387,
+      "grad_norm": 0.4461745023727417,
+      "learning_rate": 0.00021907941604716057,
+      "loss": 2.382,
+      "step": 1935
+    },
+    {
+      "epoch": 0.388,
+      "grad_norm": 0.42378079891204834,
+      "learning_rate": 0.00021863403896574534,
+      "loss": 2.3732,
+      "step": 1940
+    },
+    {
+      "epoch": 0.389,
+      "grad_norm": 0.4562043845653534,
+      "learning_rate": 0.00021818789525895375,
+      "loss": 2.3665,
+      "step": 1945
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.41517820954322815,
+      "learning_rate": 0.0002177409899101016,
+      "loss": 2.4582,
+      "step": 1950
+    },
+    {
+      "epoch": 0.391,
+      "grad_norm": 0.4133061170578003,
+      "learning_rate": 0.00021729332791101203,
+      "loss": 2.3971,
+      "step": 1955
+    },
+    {
+      "epoch": 0.392,
+      "grad_norm": 0.4128275513648987,
+      "learning_rate": 0.00021684491426195957,
+      "loss": 2.4198,
+      "step": 1960
+    },
+    {
+      "epoch": 0.393,
+      "grad_norm": 0.4075275957584381,
+      "learning_rate": 0.0002163957539716149,
+      "loss": 2.3646,
+      "step": 1965
+    },
+    {
+      "epoch": 0.394,
+      "grad_norm": 0.4190007448196411,
+      "learning_rate": 0.0002159458520569881,
+      "loss": 2.458,
+      "step": 1970
+    },
+    {
+      "epoch": 0.395,
+      "grad_norm": 0.4138437509536743,
+      "learning_rate": 0.0002154952135433732,
+      "loss": 2.3209,
+      "step": 1975
+    },
+    {
+      "epoch": 0.396,
+      "grad_norm": 0.44516539573669434,
+      "learning_rate": 0.00021504384346429187,
+      "loss": 2.4555,
+      "step": 1980
+    },
+    {
+      "epoch": 0.397,
+      "grad_norm": 0.4246795177459717,
+      "learning_rate": 0.00021459174686143718,
+      "loss": 2.3423,
+      "step": 1985
+    },
+    {
+      "epoch": 0.398,
+      "grad_norm": 0.42210015654563904,
+      "learning_rate": 0.00021413892878461717,
+      "loss": 2.4987,
+      "step": 1990
+    },
+    {
+      "epoch": 0.399,
+      "grad_norm": 0.43938738107681274,
+      "learning_rate": 0.0002136853942916987,
+      "loss": 2.4339,
+      "step": 1995
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.4321936368942261,
+      "learning_rate": 0.00021323114844855087,
+      "loss": 2.4217,
+      "step": 2000
+    },
+    {
+      "epoch": 0.401,
+      "grad_norm": 0.436798632144928,
+      "learning_rate": 0.00021277619632898817,
+      "loss": 2.2219,
+      "step": 2005
+    },
+    {
+      "epoch": 0.402,
+      "grad_norm": 0.4363677203655243,
+      "learning_rate": 0.00021232054301471428,
+      "loss": 2.3598,
+      "step": 2010
+    },
+    {
+      "epoch": 0.403,
+      "grad_norm": 0.43887215852737427,
+      "learning_rate": 0.0002118641935952648,
+      "loss": 2.4391,
+      "step": 2015
+    },
+    {
+      "epoch": 0.404,
+      "grad_norm": 0.41993457078933716,
+      "learning_rate": 0.0002114071531679509,
+      "loss": 2.428,
+      "step": 2020
+    },
+    {
+      "epoch": 0.405,
+      "grad_norm": 0.4155243933200836,
+      "learning_rate": 0.00021094942683780194,
+      "loss": 2.4014,
+      "step": 2025
+    },
+    {
+      "epoch": 0.406,
+      "grad_norm": 0.4476686716079712,
+      "learning_rate": 0.00021049101971750875,
+      "loss": 2.4425,
+      "step": 2030
+    },
+    {
+      "epoch": 0.407,
+      "grad_norm": 0.42122960090637207,
+      "learning_rate": 0.00021003193692736643,
+      "loss": 2.3467,
+      "step": 2035
+    },
+    {
+      "epoch": 0.408,
+      "grad_norm": 0.40306955575942993,
+      "learning_rate": 0.00020957218359521706,
+      "loss": 2.3764,
+      "step": 2040
+    },
+    {
+      "epoch": 0.409,
+      "grad_norm": 0.47617068886756897,
+      "learning_rate": 0.00020911176485639263,
+      "loss": 2.3937,
+      "step": 2045
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.42115652561187744,
+      "learning_rate": 0.00020865068585365745,
+      "loss": 2.3776,
+      "step": 2050
+    },
+    {
+      "epoch": 0.411,
+      "grad_norm": 0.4249623417854309,
+      "learning_rate": 0.00020818895173715083,
+      "loss": 2.3161,
+      "step": 2055
+    },
+    {
+      "epoch": 0.412,
+      "grad_norm": 0.40884312987327576,
+      "learning_rate": 0.00020772656766432961,
+      "loss": 2.3615,
+      "step": 2060
+    },
+    {
+      "epoch": 0.413,
+      "grad_norm": 0.4013335108757019,
+      "learning_rate": 0.0002072635387999104,
+      "loss": 2.3916,
+      "step": 2065
+    },
+    {
+      "epoch": 0.414,
+      "grad_norm": 0.40139567852020264,
+      "learning_rate": 0.00020679987031581206,
+      "loss": 2.3001,
+      "step": 2070
+    },
+    {
+      "epoch": 0.415,
+      "grad_norm": 0.40394365787506104,
+      "learning_rate": 0.00020633556739109782,
+      "loss": 2.3268,
+      "step": 2075
+    },
+    {
+      "epoch": 0.416,
+      "grad_norm": 0.4449470341205597,
+      "learning_rate": 0.00020587063521191736,
+      "loss": 2.4028,
+      "step": 2080
+    },
+    {
+      "epoch": 0.417,
+      "grad_norm": 0.42907705903053284,
+      "learning_rate": 0.0002054050789714491,
+      "loss": 2.3472,
+      "step": 2085
+    },
+    {
+      "epoch": 0.418,
+      "grad_norm": 0.4474325478076935,
+      "learning_rate": 0.00020493890386984198,
+      "loss": 2.331,
+      "step": 2090
+    },
+    {
+      "epoch": 0.419,
+      "grad_norm": 0.4234548807144165,
+      "learning_rate": 0.0002044721151141576,
+      "loss": 2.307,
+      "step": 2095
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.45159173011779785,
+      "learning_rate": 0.00020400471791831172,
+      "loss": 2.4009,
+      "step": 2100
+    },
+    {
+      "epoch": 0.421,
+      "grad_norm": 0.42569705843925476,
+      "learning_rate": 0.00020353671750301648,
+      "loss": 2.2994,
+      "step": 2105
+    },
+    {
+      "epoch": 0.422,
+      "grad_norm": 0.41479888558387756,
+      "learning_rate": 0.0002030681190957216,
+      "loss": 2.3603,
+      "step": 2110
+    },
+    {
+      "epoch": 0.423,
+      "grad_norm": 0.4099830389022827,
+      "learning_rate": 0.00020259892793055633,
+      "loss": 2.4141,
+      "step": 2115
+    },
+    {
+      "epoch": 0.424,
+      "grad_norm": 0.44319307804107666,
+      "learning_rate": 0.00020212914924827092,
+      "loss": 2.3434,
+      "step": 2120
+    },
+    {
+      "epoch": 0.425,
+      "grad_norm": 0.4080663323402405,
+      "learning_rate": 0.00020165878829617794,
+      "loss": 2.3563,
+      "step": 2125
+    },
+    {
+      "epoch": 0.426,
+      "grad_norm": 0.4360628128051758,
+      "learning_rate": 0.00020118785032809385,
+      "loss": 2.3404,
+      "step": 2130
+    },
+    {
+      "epoch": 0.427,
+      "grad_norm": 0.4054611027240753,
+      "learning_rate": 0.00020071634060428022,
+      "loss": 2.3795,
+      "step": 2135
+    },
+    {
+      "epoch": 0.428,
+      "grad_norm": 0.4493691623210907,
+      "learning_rate": 0.00020024426439138495,
+      "loss": 2.3898,
+      "step": 2140
+    },
+    {
+      "epoch": 0.429,
+      "grad_norm": 0.42616891860961914,
+      "learning_rate": 0.00019977162696238358,
+      "loss": 2.4323,
+      "step": 2145
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.43381640315055847,
+      "learning_rate": 0.00019929843359652017,
+      "loss": 2.3555,
+      "step": 2150
+    },
+    {
+      "epoch": 0.431,
+      "grad_norm": 0.39783987402915955,
+      "learning_rate": 0.00019882468957924855,
+      "loss": 2.3587,
+      "step": 2155
+    },
+    {
+      "epoch": 0.432,
+      "grad_norm": 0.4238894283771515,
+      "learning_rate": 0.00019835040020217316,
+      "loss": 2.3681,
+      "step": 2160
+    },
+    {
+      "epoch": 0.433,
+      "grad_norm": 0.4072917103767395,
+      "learning_rate": 0.00019787557076298998,
+      "loss": 2.3289,
+      "step": 2165
+    },
+    {
+      "epoch": 0.434,
+      "grad_norm": 0.40059733390808105,
+      "learning_rate": 0.00019740020656542733,
+      "loss": 2.4105,
+      "step": 2170
+    },
+    {
+      "epoch": 0.435,
+      "grad_norm": 0.44404855370521545,
+      "learning_rate": 0.00019692431291918667,
+      "loss": 2.2413,
+      "step": 2175
+    },
+    {
+      "epoch": 0.436,
+      "grad_norm": 0.41822120547294617,
+      "learning_rate": 0.00019644789513988337,
+      "loss": 2.3406,
+      "step": 2180
+    },
+    {
+      "epoch": 0.437,
+      "grad_norm": 0.4292788803577423,
+      "learning_rate": 0.00019597095854898697,
+      "loss": 2.3422,
+      "step": 2185
+    },
+    {
+      "epoch": 0.438,
+      "grad_norm": 0.41014593839645386,
+      "learning_rate": 0.0001954935084737623,
+      "loss": 2.3293,
+      "step": 2190
+    },
+    {
+      "epoch": 0.439,
+      "grad_norm": 0.4195305109024048,
+      "learning_rate": 0.00019501555024720944,
+      "loss": 2.3776,
+      "step": 2195
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.39849036931991577,
+      "learning_rate": 0.00019453708920800475,
+      "loss": 2.3109,
+      "step": 2200
+    },
+    {
+      "epoch": 0.441,
+      "grad_norm": 0.4177517592906952,
+      "learning_rate": 0.0001940581307004404,
+      "loss": 2.4408,
+      "step": 2205
+    },
+    {
+      "epoch": 0.442,
+      "grad_norm": 0.4089096486568451,
+      "learning_rate": 0.00019357868007436552,
+      "loss": 2.3956,
+      "step": 2210
+    },
+    {
+      "epoch": 0.443,
+      "grad_norm": 0.3905986547470093,
+      "learning_rate": 0.00019309874268512596,
+      "loss": 2.3823,
+      "step": 2215
+    },
+    {
+      "epoch": 0.444,
+      "grad_norm": 0.40985196828842163,
+      "learning_rate": 0.00019261832389350463,
+      "loss": 2.3608,
+      "step": 2220
+    },
+    {
+      "epoch": 0.445,
+      "grad_norm": 0.40191152691841125,
+      "learning_rate": 0.00019213742906566152,
+      "loss": 2.2658,
+      "step": 2225
+    },
+    {
+      "epoch": 0.446,
+      "grad_norm": 0.4191170632839203,
+      "learning_rate": 0.0001916560635730739,
+      "loss": 2.3718,
+      "step": 2230
+    },
+    {
+      "epoch": 0.447,
+      "grad_norm": 0.44027984142303467,
+      "learning_rate": 0.00019117423279247628,
+      "loss": 2.3377,
+      "step": 2235
+    },
+    {
+      "epoch": 0.448,
+      "grad_norm": 0.4240022897720337,
+      "learning_rate": 0.00019069194210580019,
+      "loss": 2.3589,
+      "step": 2240
+    },
+    {
+      "epoch": 0.449,
+      "grad_norm": 0.40984782576560974,
+      "learning_rate": 0.00019020919690011424,
+      "loss": 2.3659,
+      "step": 2245
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.40956535935401917,
+      "learning_rate": 0.000189726002567564,
+      "loss": 2.3578,
+      "step": 2250
+    },
+    {
+      "epoch": 0.451,
+      "grad_norm": 0.4261001944541931,
+      "learning_rate": 0.0001892423645053116,
+      "loss": 2.3671,
+      "step": 2255
+    },
+    {
+      "epoch": 0.452,
+      "grad_norm": 0.42046308517456055,
+      "learning_rate": 0.00018875828811547557,
+      "loss": 2.3699,
+      "step": 2260
+    },
+    {
+      "epoch": 0.453,
+      "grad_norm": 0.423777312040329,
+      "learning_rate": 0.00018827377880507035,
+      "loss": 2.3395,
+      "step": 2265
+    },
+    {
+      "epoch": 0.454,
+      "grad_norm": 0.41518712043762207,
+      "learning_rate": 0.00018778884198594615,
+      "loss": 2.2932,
+      "step": 2270
+    },
+    {
+      "epoch": 0.455,
+      "grad_norm": 0.43722566962242126,
+      "learning_rate": 0.00018730348307472824,
+      "loss": 2.2692,
+      "step": 2275
+    },
+    {
+      "epoch": 0.456,
+      "grad_norm": 0.4038710594177246,
+      "learning_rate": 0.00018681770749275647,
+      "loss": 2.283,
+      "step": 2280
+    },
+    {
+      "epoch": 0.457,
+      "grad_norm": 0.42532870173454285,
+      "learning_rate": 0.00018633152066602508,
+      "loss": 2.2813,
+      "step": 2285
+    },
+    {
+      "epoch": 0.458,
+      "grad_norm": 0.4140380024909973,
+      "learning_rate": 0.00018584492802512152,
+      "loss": 2.308,
+      "step": 2290
+    },
+    {
+      "epoch": 0.459,
+      "grad_norm": 0.4389931559562683,
+      "learning_rate": 0.00018535793500516626,
+      "loss": 2.3299,
+      "step": 2295
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.42715537548065186,
+      "learning_rate": 0.00018487054704575183,
+      "loss": 2.3266,
+      "step": 2300
+    },
+    {
+      "epoch": 0.461,
+      "grad_norm": 0.42225635051727295,
+      "learning_rate": 0.00018438276959088218,
+      "loss": 2.3171,
+      "step": 2305
+    },
+    {
+      "epoch": 0.462,
+      "grad_norm": 0.4294082820415497,
+      "learning_rate": 0.00018389460808891176,
+      "loss": 2.3159,
+      "step": 2310
+    },
+    {
+      "epoch": 0.463,
+      "grad_norm": 0.44747427105903625,
+      "learning_rate": 0.00018340606799248485,
+      "loss": 2.2908,
+      "step": 2315
+    },
+    {
+      "epoch": 0.464,
+      "grad_norm": 0.4157891571521759,
+      "learning_rate": 0.0001829171547584744,
+      "loss": 2.4158,
+      "step": 2320
+    },
+    {
+      "epoch": 0.465,
+      "grad_norm": 0.4599103331565857,
+      "learning_rate": 0.00018242787384792136,
+      "loss": 2.384,
+      "step": 2325
+    },
+    {
+      "epoch": 0.466,
+      "grad_norm": 0.40598711371421814,
+      "learning_rate": 0.00018193823072597338,
+      "loss": 2.3425,
+      "step": 2330
+    },
+    {
+      "epoch": 0.467,
+      "grad_norm": 0.43444859981536865,
+      "learning_rate": 0.00018144823086182406,
+      "loss": 2.3395,
+      "step": 2335
+    },
+    {
+      "epoch": 0.468,
+      "grad_norm": 0.45927369594573975,
+      "learning_rate": 0.00018095787972865162,
+      "loss": 2.5384,
+      "step": 2340
+    },
+    {
+      "epoch": 0.469,
+      "grad_norm": 0.45538032054901123,
+      "learning_rate": 0.0001804671828035579,
+      "loss": 2.3153,
+      "step": 2345
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.3852723240852356,
+      "learning_rate": 0.00017997614556750723,
+      "loss": 2.3461,
+      "step": 2350
+    },
+    {
+      "epoch": 0.471,
+      "grad_norm": 0.4270138144493103,
+      "learning_rate": 0.00017948477350526504,
+      "loss": 2.2924,
+      "step": 2355
+    },
+    {
+      "epoch": 0.472,
+      "grad_norm": 0.41092297434806824,
+      "learning_rate": 0.00017899307210533663,
+      "loss": 2.3212,
+      "step": 2360
+    },
+    {
+      "epoch": 0.473,
+      "grad_norm": 0.4263060986995697,
+      "learning_rate": 0.00017850104685990614,
+      "loss": 2.3173,
+      "step": 2365
+    },
+    {
+      "epoch": 0.474,
+      "grad_norm": 0.4177812933921814,
+      "learning_rate": 0.0001780087032647748,
+      "loss": 2.3849,
+      "step": 2370
+    },
+    {
+      "epoch": 0.475,
+      "grad_norm": 0.4262104630470276,
+      "learning_rate": 0.00017751604681929967,
+      "loss": 2.3068,
+      "step": 2375
+    },
+    {
+      "epoch": 0.476,
+      "grad_norm": 0.40943190455436707,
+      "learning_rate": 0.00017702308302633253,
+      "loss": 2.2976,
+      "step": 2380
+    },
+    {
+      "epoch": 0.477,
+      "grad_norm": 0.40264710783958435,
+      "learning_rate": 0.00017652981739215784,
+      "loss": 2.2258,
+      "step": 2385
+    },
+    {
+      "epoch": 0.478,
+      "grad_norm": 0.4030868411064148,
+      "learning_rate": 0.0001760362554264319,
+      "loss": 2.3503,
+      "step": 2390
+    },
+    {
+      "epoch": 0.479,
+      "grad_norm": 0.40495970845222473,
+      "learning_rate": 0.00017554240264212062,
+      "loss": 2.2957,
+      "step": 2395
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.413086473941803,
+      "learning_rate": 0.0001750482645554385,
+      "loss": 2.3362,
+      "step": 2400
+    },
+    {
+      "epoch": 0.481,
+      "grad_norm": 0.42103326320648193,
+      "learning_rate": 0.00017455384668578676,
+      "loss": 2.4136,
+      "step": 2405
+    },
+    {
+      "epoch": 0.482,
+      "grad_norm": 0.4094327688217163,
+      "learning_rate": 0.00017405915455569163,
+      "loss": 2.3414,
+      "step": 2410
+    },
+    {
+      "epoch": 0.483,
+      "grad_norm": 0.44162988662719727,
+      "learning_rate": 0.00017356419369074288,
+      "loss": 2.2554,
+      "step": 2415
+    },
+    {
+      "epoch": 0.484,
+      "grad_norm": 0.41292014718055725,
+      "learning_rate": 0.00017306896961953193,
+      "loss": 2.2463,
+      "step": 2420
+    },
+    {
+      "epoch": 0.485,
+      "grad_norm": 0.43508976697921753,
+      "learning_rate": 0.00017257348787359022,
+      "loss": 2.2801,
+      "step": 2425
+    },
+    {
+      "epoch": 0.486,
+      "grad_norm": 0.4286428689956665,
+      "learning_rate": 0.00017207775398732716,
+      "loss": 2.2726,
+      "step": 2430
+    },
+    {
+      "epoch": 0.487,
+      "grad_norm": 0.4207102060317993,
+      "learning_rate": 0.0001715817734979687,
+      "loss": 2.3964,
+      "step": 2435
+    },
+    {
+      "epoch": 0.488,
+      "grad_norm": 0.4012376368045807,
+      "learning_rate": 0.00017108555194549524,
+      "loss": 2.3938,
+      "step": 2440
+    },
+    {
+      "epoch": 0.489,
+      "grad_norm": 0.42363470792770386,
+      "learning_rate": 0.00017058909487257972,
+      "loss": 2.3354,
+      "step": 2445
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.4191186726093292,
+      "learning_rate": 0.0001700924078245259,
+      "loss": 2.3304,
+      "step": 2450
+    },
+    {
+      "epoch": 0.491,
+      "grad_norm": 0.4244017004966736,
+      "learning_rate": 0.00016959549634920623,
+      "loss": 2.2543,
+      "step": 2455
+    },
+    {
+      "epoch": 0.492,
+      "grad_norm": 0.418206125497818,
+      "learning_rate": 0.00016909836599699997,
+      "loss": 2.322,
+      "step": 2460
+    },
+    {
+      "epoch": 0.493,
+      "grad_norm": 0.4423177242279053,
+      "learning_rate": 0.0001686010223207312,
+      "loss": 2.1921,
+      "step": 2465
+    },
+    {
+      "epoch": 0.494,
+      "grad_norm": 0.4443405270576477,
+      "learning_rate": 0.00016810347087560672,
+      "loss": 2.3374,
+      "step": 2470
+    },
+    {
+      "epoch": 0.495,
+      "grad_norm": 0.4554826617240906,
+      "learning_rate": 0.0001676057172191542,
+      "loss": 2.3143,
+      "step": 2475
+    },
+    {
+      "epoch": 0.496,
+      "grad_norm": 0.41316652297973633,
+      "learning_rate": 0.00016710776691115968,
+      "loss": 2.3136,
+      "step": 2480
+    },
+    {
+      "epoch": 0.497,
+      "grad_norm": 0.4269903302192688,
+      "learning_rate": 0.00016660962551360613,
+      "loss": 2.3707,
+      "step": 2485
+    },
+    {
+      "epoch": 0.498,
+      "grad_norm": 0.4215530753135681,
+      "learning_rate": 0.00016611129859061066,
+      "loss": 2.3857,
+      "step": 2490
+    },
+    {
+      "epoch": 0.499,
+      "grad_norm": 0.42953386902809143,
+      "learning_rate": 0.00016561279170836267,
+      "loss": 2.2289,
+      "step": 2495
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.41060370206832886,
+      "learning_rate": 0.00016511411043506184,
+      "loss": 2.2564,
+      "step": 2500
+    },
+    {
+      "epoch": 0.501,
+      "grad_norm": 0.425135999917984,
+      "learning_rate": 0.00016461526034085563,
+      "loss": 2.4423,
+      "step": 2505
+    },
+    {
+      "epoch": 0.502,
+      "grad_norm": 0.4176621735095978,
+      "learning_rate": 0.00016411624699777717,
+      "loss": 2.3313,
+      "step": 2510
+    },
+    {
+      "epoch": 0.503,
+      "grad_norm": 0.4177638590335846,
+      "learning_rate": 0.000163617075979683,
+      "loss": 2.2815,
+      "step": 2515
+    },
+    {
+      "epoch": 0.504,
+      "grad_norm": 0.42569682002067566,
+      "learning_rate": 0.0001631177528621911,
+      "loss": 2.299,
+      "step": 2520
+    },
+    {
+      "epoch": 0.505,
+      "grad_norm": 0.41489818692207336,
+      "learning_rate": 0.00016261828322261803,
+      "loss": 2.2341,
+      "step": 2525
+    },
+    {
+      "epoch": 0.506,
+      "grad_norm": 0.4402053654193878,
+      "learning_rate": 0.0001621186726399172,
+      "loss": 2.3399,
+      "step": 2530
+    },
+    {
+      "epoch": 0.507,
+      "grad_norm": 0.4353001117706299,
+      "learning_rate": 0.0001616189266946162,
+      "loss": 2.2805,
+      "step": 2535
+    },
+    {
+      "epoch": 0.508,
+      "grad_norm": 0.4333752393722534,
+      "learning_rate": 0.00016111905096875468,
+      "loss": 2.2669,
+      "step": 2540
+    },
+    {
+      "epoch": 0.509,
+      "grad_norm": 0.4180094599723816,
+      "learning_rate": 0.0001606190510458218,
+      "loss": 2.3054,
+      "step": 2545
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.4101157784461975,
+      "learning_rate": 0.00016011893251069404,
+      "loss": 2.3033,
+      "step": 2550
+    },
+    {
+      "epoch": 0.511,
+      "grad_norm": 0.41150620579719543,
+      "learning_rate": 0.0001596187009495728,
+      "loss": 2.3246,
+      "step": 2555
+    },
+    {
+      "epoch": 0.512,
+      "grad_norm": 0.4234907925128937,
+      "learning_rate": 0.00015911836194992183,
+      "loss": 2.3146,
+      "step": 2560
+    },
+    {
+      "epoch": 0.513,
+      "grad_norm": 0.4023984968662262,
+      "learning_rate": 0.00015861792110040496,
+      "loss": 2.2686,
+      "step": 2565
+    },
+    {
+      "epoch": 0.514,
+      "grad_norm": 0.41848230361938477,
+      "learning_rate": 0.0001581173839908238,
+      "loss": 2.249,
+      "step": 2570
+    },
+    {
+      "epoch": 0.515,
+      "grad_norm": 0.3979667127132416,
+      "learning_rate": 0.000157616756212055,
+      "loss": 2.2636,
+      "step": 2575
+    },
+    {
+      "epoch": 0.516,
+      "grad_norm": 0.4226064085960388,
+      "learning_rate": 0.00015711604335598808,
+      "loss": 2.2615,
+      "step": 2580
+    },
+    {
+      "epoch": 0.517,
+      "grad_norm": 0.4342791736125946,
+      "learning_rate": 0.00015661525101546268,
+      "loss": 2.374,
+      "step": 2585
+    },
+    {
+      "epoch": 0.518,
+      "grad_norm": 0.4076896011829376,
+      "learning_rate": 0.00015611438478420648,
+      "loss": 2.2511,
+      "step": 2590
+    },
+    {
+      "epoch": 0.519,
+      "grad_norm": 0.42497551441192627,
+      "learning_rate": 0.00015561345025677235,
+      "loss": 2.306,
+      "step": 2595
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.40860408544540405,
+      "learning_rate": 0.00015511245302847606,
+      "loss": 2.2657,
+      "step": 2600
+    },
+    {
+      "epoch": 0.521,
+      "grad_norm": 0.43411552906036377,
+      "learning_rate": 0.00015461139869533383,
+      "loss": 2.3071,
+      "step": 2605
+    },
+    {
+      "epoch": 0.522,
+      "grad_norm": 0.44043874740600586,
+      "learning_rate": 0.00015411029285399946,
+      "loss": 2.316,
+      "step": 2610
+    },
+    {
+      "epoch": 0.523,
+      "grad_norm": 0.4515673518180847,
+      "learning_rate": 0.0001536091411017024,
+      "loss": 2.2077,
+      "step": 2615
+    },
+    {
+      "epoch": 0.524,
+      "grad_norm": 0.42663195729255676,
+      "learning_rate": 0.0001531079490361847,
+      "loss": 2.2715,
+      "step": 2620
+    },
+    {
+      "epoch": 0.525,
+      "grad_norm": 0.4274803102016449,
+      "learning_rate": 0.00015260672225563877,
+      "loss": 2.2968,
+      "step": 2625
+    },
+    {
+      "epoch": 0.526,
+      "grad_norm": 0.4248811900615692,
+      "learning_rate": 0.0001521054663586448,
+      "loss": 2.3694,
+      "step": 2630
+    },
+    {
+      "epoch": 0.527,
+      "grad_norm": 0.4073241651058197,
+      "learning_rate": 0.00015160418694410815,
+      "loss": 2.2967,
+      "step": 2635
+    },
+    {
+      "epoch": 0.528,
+      "grad_norm": 0.42025795578956604,
+      "learning_rate": 0.00015110288961119693,
+      "loss": 2.2745,
+      "step": 2640
+    },
+    {
+      "epoch": 0.529,
+      "grad_norm": 0.42329052090644836,
+      "learning_rate": 0.00015060157995927932,
+      "loss": 2.3771,
+      "step": 2645
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.38789117336273193,
+      "learning_rate": 0.0001501002635878612,
+      "loss": 2.2525,
+      "step": 2650
+    },
+    {
+      "epoch": 0.531,
+      "grad_norm": 0.43104180693626404,
+      "learning_rate": 0.00014959894609652334,
+      "loss": 2.29,
+      "step": 2655
+    },
+    {
+      "epoch": 0.532,
+      "grad_norm": 0.42107024788856506,
+      "learning_rate": 0.00014909763308485925,
+      "loss": 2.3289,
+      "step": 2660
+    },
+    {
+      "epoch": 0.533,
+      "grad_norm": 0.419460654258728,
+      "learning_rate": 0.0001485963301524122,
+      "loss": 2.24,
+      "step": 2665
+    },
+    {
+      "epoch": 0.534,
+      "grad_norm": 0.45686954259872437,
+      "learning_rate": 0.0001480950428986129,
+      "loss": 2.2585,
+      "step": 2670
+    },
+    {
+      "epoch": 0.535,
+      "grad_norm": 0.4176424741744995,
+      "learning_rate": 0.0001475937769227171,
+      "loss": 2.3315,
+      "step": 2675
+    },
+    {
+      "epoch": 0.536,
+      "grad_norm": 0.4299408197402954,
+      "learning_rate": 0.00014709253782374274,
+      "loss": 2.2319,
+      "step": 2680
+    },
+    {
+      "epoch": 0.537,
+      "grad_norm": 0.42534300684928894,
+      "learning_rate": 0.00014659133120040755,
+      "loss": 2.2121,
+      "step": 2685
+    },
+    {
+      "epoch": 0.538,
+      "grad_norm": 0.42528364062309265,
+      "learning_rate": 0.00014609016265106656,
+      "loss": 2.2479,
+      "step": 2690
+    },
+    {
+      "epoch": 0.539,
+      "grad_norm": 0.4252580404281616,
+      "learning_rate": 0.0001455890377736495,
+      "loss": 2.2287,
+      "step": 2695
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.4283309876918793,
+      "learning_rate": 0.00014508796216559832,
+      "loss": 2.2557,
+      "step": 2700
+    },
+    {
+      "epoch": 0.541,
+      "grad_norm": 0.434817373752594,
+      "learning_rate": 0.00014458694142380454,
+      "loss": 2.2359,
+      "step": 2705
+    },
+    {
+      "epoch": 0.542,
+      "grad_norm": 0.41941994428634644,
+      "learning_rate": 0.00014408598114454705,
+      "loss": 2.2425,
+      "step": 2710
+    },
+    {
+      "epoch": 0.543,
+      "grad_norm": 0.42644003033638,
+      "learning_rate": 0.00014358508692342926,
+      "loss": 2.3102,
+      "step": 2715
+    },
+    {
+      "epoch": 0.544,
+      "grad_norm": 0.45429447293281555,
+      "learning_rate": 0.0001430842643553166,
+      "loss": 2.2705,
+      "step": 2720
+    },
+    {
+      "epoch": 0.545,
+      "grad_norm": 0.38917839527130127,
+      "learning_rate": 0.00014258351903427438,
+      "loss": 2.2375,
+      "step": 2725
+    },
+    {
+      "epoch": 0.546,
+      "grad_norm": 0.42386049032211304,
+      "learning_rate": 0.00014208285655350496,
+      "loss": 2.2354,
+      "step": 2730
+    },
+    {
+      "epoch": 0.547,
+      "grad_norm": 0.43970856070518494,
+      "learning_rate": 0.00014158228250528546,
+      "loss": 2.3308,
+      "step": 2735
+    },
+    {
+      "epoch": 0.548,
+      "grad_norm": 0.4227691888809204,
+      "learning_rate": 0.00014108180248090506,
+      "loss": 2.3061,
+      "step": 2740
+    },
+    {
+      "epoch": 0.549,
+      "grad_norm": 0.39985859394073486,
+      "learning_rate": 0.000140581422070603,
+      "loss": 2.2377,
+      "step": 2745
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.3900357484817505,
+      "learning_rate": 0.00014008114686350556,
+      "loss": 2.2854,
+      "step": 2750
+    },
+    {
+      "epoch": 0.551,
+      "grad_norm": 0.4252527356147766,
+      "learning_rate": 0.00013958098244756418,
+      "loss": 2.2305,
+      "step": 2755
+    },
+    {
+      "epoch": 0.552,
+      "grad_norm": 0.3899967074394226,
+      "learning_rate": 0.0001390809344094926,
+      "loss": 2.3445,
+      "step": 2760
+    },
+    {
+      "epoch": 0.553,
+      "grad_norm": 0.4172196090221405,
+      "learning_rate": 0.0001385810083347047,
+      "loss": 2.2162,
+      "step": 2765
+    },
+    {
+      "epoch": 0.554,
+      "grad_norm": 0.43587014079093933,
+      "learning_rate": 0.0001380812098072522,
+      "loss": 2.2852,
+      "step": 2770
+    },
+    {
+      "epoch": 0.555,
+      "grad_norm": 0.41883230209350586,
+      "learning_rate": 0.0001375815444097619,
+      "loss": 2.2875,
+      "step": 2775
+    },
+    {
+      "epoch": 0.556,
+      "grad_norm": 0.41043713688850403,
+      "learning_rate": 0.0001370820177233738,
+      "loss": 2.3192,
+      "step": 2780
+    },
+    {
+      "epoch": 0.557,
+      "grad_norm": 0.42568516731262207,
+      "learning_rate": 0.00013658263532767835,
+      "loss": 2.2224,
+      "step": 2785
+    },
+    {
+      "epoch": 0.558,
+      "grad_norm": 0.44088488817214966,
+      "learning_rate": 0.00013608340280065446,
+      "loss": 2.2605,
+      "step": 2790
+    },
+    {
+      "epoch": 0.559,
+      "grad_norm": 0.39374229311943054,
+      "learning_rate": 0.0001355843257186069,
+      "loss": 2.2003,
+      "step": 2795
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.398308128118515,
+      "learning_rate": 0.00013508540965610424,
+      "loss": 2.27,
+      "step": 2800
+    },
+    {
+      "epoch": 0.561,
+      "grad_norm": 0.3940664231777191,
+      "learning_rate": 0.00013458666018591656,
+      "loss": 2.1863,
+      "step": 2805
+    },
+    {
+      "epoch": 0.562,
+      "grad_norm": 0.4169365167617798,
+      "learning_rate": 0.00013408808287895306,
+      "loss": 2.1897,
+      "step": 2810
+    },
+    {
+      "epoch": 0.563,
+      "grad_norm": 0.4238418638706207,
+      "learning_rate": 0.0001335896833041999,
+      "loss": 2.258,
+      "step": 2815
+    },
+    {
+      "epoch": 0.564,
+      "grad_norm": 0.423495352268219,
+      "learning_rate": 0.00013309146702865805,
+      "loss": 2.2247,
+      "step": 2820
+    },
+    {
+      "epoch": 0.565,
+      "grad_norm": 0.4261130690574646,
+      "learning_rate": 0.00013259343961728108,
+      "loss": 2.2895,
+      "step": 2825
+    },
+    {
+      "epoch": 0.566,
+      "grad_norm": 0.42107266187667847,
+      "learning_rate": 0.00013209560663291303,
+      "loss": 2.217,
+      "step": 2830
+    },
+    {
+      "epoch": 0.567,
+      "grad_norm": 0.4057311713695526,
+      "learning_rate": 0.0001315979736362261,
+      "loss": 2.2517,
+      "step": 2835
+    },
+    {
+      "epoch": 0.568,
+      "grad_norm": 0.41574499011039734,
+      "learning_rate": 0.00013110054618565886,
+      "loss": 2.2166,
+      "step": 2840
+    },
+    {
+      "epoch": 0.569,
+      "grad_norm": 0.3923725485801697,
+      "learning_rate": 0.0001306033298373539,
+      "loss": 2.2217,
+      "step": 2845
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.4277208149433136,
+      "learning_rate": 0.0001301063301450958,
+      "loss": 2.257,
+      "step": 2850
+    },
+    {
+      "epoch": 0.571,
+      "grad_norm": 0.4455857574939728,
+      "learning_rate": 0.00012960955266024917,
+      "loss": 2.2104,
+      "step": 2855
+    },
+    {
+      "epoch": 0.572,
+      "grad_norm": 0.43064215779304504,
+      "learning_rate": 0.00012911300293169664,
+      "loss": 2.2939,
+      "step": 2860
+    },
+    {
+      "epoch": 0.573,
+      "grad_norm": 0.4337853789329529,
+      "learning_rate": 0.00012861668650577688,
+      "loss": 2.3123,
+      "step": 2865
+    },
+    {
+      "epoch": 0.574,
+      "grad_norm": 0.41334277391433716,
+      "learning_rate": 0.00012812060892622264,
+      "loss": 2.2551,
+      "step": 2870
+    },
+    {
+      "epoch": 0.575,
+      "grad_norm": 0.4144851863384247,
+      "learning_rate": 0.00012762477573409868,
+      "loss": 2.241,
+      "step": 2875
+    },
+    {
+      "epoch": 0.576,
+      "grad_norm": 0.40655460953712463,
+      "learning_rate": 0.00012712919246774017,
+      "loss": 2.1983,
+      "step": 2880
+    },
+    {
+      "epoch": 0.577,
+      "grad_norm": 0.4161902666091919,
+      "learning_rate": 0.0001266338646626906,
+      "loss": 2.1633,
+      "step": 2885
+    },
+    {
+      "epoch": 0.578,
+      "grad_norm": 0.45048007369041443,
+      "learning_rate": 0.00012613879785163998,
+      "loss": 2.3365,
+      "step": 2890
+    },
+    {
+      "epoch": 0.579,
+      "grad_norm": 0.40051379799842834,
+      "learning_rate": 0.00012564399756436314,
+      "loss": 2.1469,
+      "step": 2895
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.4100329875946045,
+      "learning_rate": 0.00012514946932765794,
+      "loss": 2.1957,
+      "step": 2900
+    },
+    {
+      "epoch": 0.581,
+      "grad_norm": 0.40546661615371704,
+      "learning_rate": 0.00012465521866528346,
+      "loss": 2.182,
+      "step": 2905
+    },
+    {
+      "epoch": 0.582,
+      "grad_norm": 0.4651910662651062,
+      "learning_rate": 0.00012416125109789827,
+      "loss": 2.2542,
+      "step": 2910
+    },
+    {
+      "epoch": 0.583,
+      "grad_norm": 0.4111328721046448,
+      "learning_rate": 0.00012366757214299896,
+      "loss": 2.2598,
+      "step": 2915
+    },
+    {
+      "epoch": 0.584,
+      "grad_norm": 0.4306609630584717,
+      "learning_rate": 0.00012317418731485832,
+      "loss": 2.2524,
+      "step": 2920
+    },
+    {
+      "epoch": 0.585,
+      "grad_norm": 0.4010145664215088,
+      "learning_rate": 0.00012268110212446386,
+      "loss": 2.2163,
+      "step": 2925
+    },
+    {
+      "epoch": 0.586,
+      "grad_norm": 0.44942528009414673,
+      "learning_rate": 0.00012218832207945609,
+      "loss": 2.3242,
+      "step": 2930
+    },
+    {
+      "epoch": 0.587,
+      "grad_norm": 0.42698466777801514,
+      "learning_rate": 0.00012169585268406728,
+      "loss": 2.2714,
+      "step": 2935
+    },
+    {
+      "epoch": 0.588,
+      "grad_norm": 0.420461505651474,
+      "learning_rate": 0.00012120369943905976,
+      "loss": 2.2011,
+      "step": 2940
+    },
+    {
+      "epoch": 0.589,
+      "grad_norm": 0.4152168929576874,
+      "learning_rate": 0.00012071186784166456,
+      "loss": 2.2704,
+      "step": 2945
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.39992696046829224,
+      "learning_rate": 0.00012022036338551985,
+      "loss": 2.2062,
+      "step": 2950
+    },
+    {
+      "epoch": 0.591,
+      "grad_norm": 0.39946138858795166,
+      "learning_rate": 0.00011972919156060985,
+      "loss": 2.2606,
+      "step": 2955
+    },
+    {
+      "epoch": 0.592,
+      "grad_norm": 0.4226006865501404,
+      "learning_rate": 0.00011923835785320349,
+      "loss": 2.252,
+      "step": 2960
+    },
+    {
+      "epoch": 0.593,
+      "grad_norm": 0.416586697101593,
+      "learning_rate": 0.00011874786774579264,
+      "loss": 2.1411,
+      "step": 2965
+    },
+    {
+      "epoch": 0.594,
+      "grad_norm": 0.44908973574638367,
+      "learning_rate": 0.00011825772671703155,
+      "loss": 2.2244,
+      "step": 2970
+    },
+    {
+      "epoch": 0.595,
+      "grad_norm": 0.42042890191078186,
+      "learning_rate": 0.0001177679402416753,
+      "loss": 2.1188,
+      "step": 2975
+    },
+    {
+      "epoch": 0.596,
+      "grad_norm": 0.4261787235736847,
+      "learning_rate": 0.00011727851379051865,
+      "loss": 2.2926,
+      "step": 2980
+    },
+    {
+      "epoch": 0.597,
+      "grad_norm": 0.4221886992454529,
+      "learning_rate": 0.00011678945283033488,
+      "loss": 2.2446,
+      "step": 2985
+    },
+    {
+      "epoch": 0.598,
+      "grad_norm": 0.41435757279396057,
+      "learning_rate": 0.000116300762823815,
+      "loss": 2.2215,
+      "step": 2990
+    },
+    {
+      "epoch": 0.599,
+      "grad_norm": 0.4205993115901947,
+      "learning_rate": 0.00011581244922950652,
+      "loss": 2.2375,
+      "step": 2995
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.41477659344673157,
+      "learning_rate": 0.00011532451750175245,
+      "loss": 2.2337,
+      "step": 3000
+    },
+    {
+      "epoch": 0.601,
+      "grad_norm": 0.4197056293487549,
+      "learning_rate": 0.00011483697309063048,
+      "loss": 2.241,
+      "step": 3005
+    },
+    {
+      "epoch": 0.602,
+      "grad_norm": 0.4107452929019928,
+      "learning_rate": 0.00011434982144189203,
+      "loss": 2.2036,
+      "step": 3010
+    },
+    {
+      "epoch": 0.603,
+      "grad_norm": 0.4387679994106293,
+      "learning_rate": 0.00011386306799690153,
+      "loss": 2.1471,
+      "step": 3015
+    },
+    {
+      "epoch": 0.604,
+      "grad_norm": 0.4124211072921753,
+      "learning_rate": 0.0001133767181925756,
+      "loss": 2.1362,
+      "step": 3020
+    },
+    {
+      "epoch": 0.605,
+      "grad_norm": 0.4435650408267975,
+      "learning_rate": 0.00011289077746132207,
+      "loss": 2.2456,
+      "step": 3025
+    },
+    {
+      "epoch": 0.606,
+      "grad_norm": 0.4163254201412201,
+      "learning_rate": 0.00011240525123097982,
+      "loss": 2.1252,
+      "step": 3030
+    },
+    {
+      "epoch": 0.607,
+      "grad_norm": 0.44170472025871277,
+      "learning_rate": 0.00011192014492475771,
+      "loss": 2.2534,
+      "step": 3035
+    },
+    {
+      "epoch": 0.608,
+      "grad_norm": 0.41352444887161255,
+      "learning_rate": 0.00011143546396117416,
+      "loss": 2.2033,
+      "step": 3040
+    },
+    {
+      "epoch": 0.609,
+      "grad_norm": 0.44527488946914673,
+      "learning_rate": 0.00011095121375399656,
+      "loss": 2.1998,
+      "step": 3045
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.42093437910079956,
+      "learning_rate": 0.00011046739971218091,
+      "loss": 2.261,
+      "step": 3050
+    },
+    {
+      "epoch": 0.611,
+      "grad_norm": 0.44440940022468567,
+      "learning_rate": 0.00010998402723981153,
+      "loss": 2.182,
+      "step": 3055
+    },
+    {
+      "epoch": 0.612,
+      "grad_norm": 0.4243025779724121,
+      "learning_rate": 0.00010950110173604009,
+      "loss": 2.1064,
+      "step": 3060
+    },
+    {
+      "epoch": 0.613,
+      "grad_norm": 0.4191250801086426,
+      "learning_rate": 0.00010901862859502616,
+      "loss": 2.2331,
+      "step": 3065
+    },
+    {
+      "epoch": 0.614,
+      "grad_norm": 0.392995148897171,
+      "learning_rate": 0.00010853661320587631,
+      "loss": 2.2349,
+      "step": 3070
+    },
+    {
+      "epoch": 0.615,
+      "grad_norm": 0.41072773933410645,
+      "learning_rate": 0.00010805506095258419,
+      "loss": 2.2145,
+      "step": 3075
+    },
+    {
+      "epoch": 0.616,
+      "grad_norm": 0.42511963844299316,
+      "learning_rate": 0.00010757397721397026,
+      "loss": 2.3133,
+      "step": 3080
+    },
+    {
+      "epoch": 0.617,
+      "grad_norm": 0.4170040488243103,
+      "learning_rate": 0.00010709336736362188,
+      "loss": 2.2486,
+      "step": 3085
+    },
+    {
+      "epoch": 0.618,
+      "grad_norm": 0.44579964876174927,
+      "learning_rate": 0.0001066132367698332,
+      "loss": 2.2715,
+      "step": 3090
+    },
+    {
+      "epoch": 0.619,
+      "grad_norm": 0.4112447500228882,
+      "learning_rate": 0.00010613359079554517,
+      "loss": 2.205,
+      "step": 3095
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.4118829667568207,
+      "learning_rate": 0.00010565443479828558,
+      "loss": 2.2176,
+      "step": 3100
+    },
+    {
+      "epoch": 0.621,
+      "grad_norm": 0.4276077449321747,
+      "learning_rate": 0.00010517577413010938,
+      "loss": 2.2166,
+      "step": 3105
+    },
+    {
+      "epoch": 0.622,
+      "grad_norm": 0.41999852657318115,
+      "learning_rate": 0.00010469761413753881,
+      "loss": 2.1811,
+      "step": 3110
+    },
+    {
+      "epoch": 0.623,
+      "grad_norm": 0.43243396282196045,
+      "learning_rate": 0.0001042199601615037,
+      "loss": 2.2746,
+      "step": 3115
+    },
+    {
+      "epoch": 0.624,
+      "grad_norm": 0.3946484327316284,
+      "learning_rate": 0.00010374281753728167,
+      "loss": 2.1711,
+      "step": 3120
+    },
+    {
+      "epoch": 0.625,
+      "grad_norm": 0.40924206376075745,
+      "learning_rate": 0.00010326619159443889,
+      "loss": 2.2793,
+      "step": 3125
+    },
+    {
+      "epoch": 0.626,
+      "grad_norm": 0.44488149881362915,
+      "learning_rate": 0.00010279008765677017,
+      "loss": 2.1942,
+      "step": 3130
+    },
+    {
+      "epoch": 0.627,
+      "grad_norm": 0.43026819825172424,
+      "learning_rate": 0.00010231451104223966,
+      "loss": 2.184,
+      "step": 3135
+    },
+    {
+      "epoch": 0.628,
+      "grad_norm": 0.4194297194480896,
+      "learning_rate": 0.00010183946706292149,
+      "loss": 2.2779,
+      "step": 3140
+    },
+    {
+      "epoch": 0.629,
+      "grad_norm": 0.40315303206443787,
+      "learning_rate": 0.00010136496102494028,
+      "loss": 2.2575,
+      "step": 3145
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.42293116450309753,
+      "learning_rate": 0.00010089099822841223,
+      "loss": 2.1476,
+      "step": 3150
+    },
+    {
+      "epoch": 0.631,
+      "grad_norm": 0.4373874366283417,
+      "learning_rate": 0.00010041758396738528,
+      "loss": 2.1989,
+      "step": 3155
+    },
+    {
+      "epoch": 0.632,
+      "grad_norm": 0.4246419072151184,
+      "learning_rate": 9.994472352978073e-05,
+      "loss": 2.303,
+      "step": 3160
+    },
+    {
+      "epoch": 0.633,
+      "grad_norm": 0.41784340143203735,
+      "learning_rate": 9.94724221973336e-05,
+      "loss": 2.1698,
+      "step": 3165
+    },
+    {
+      "epoch": 0.634,
+      "grad_norm": 0.4105053246021271,
+      "learning_rate": 9.900068524553393e-05,
+      "loss": 2.1971,
+      "step": 3170
+    },
+    {
+      "epoch": 0.635,
+      "grad_norm": 0.4249609708786011,
+      "learning_rate": 9.852951794356768e-05,
+      "loss": 2.1984,
+      "step": 3175
+    },
+    {
+      "epoch": 0.636,
+      "grad_norm": 0.44941267371177673,
+      "learning_rate": 9.805892555425801e-05,
+      "loss": 2.1712,
+      "step": 3180
+    },
+    {
+      "epoch": 0.637,
+      "grad_norm": 0.43053391575813293,
+      "learning_rate": 9.758891333400652e-05,
+      "loss": 2.2137,
+      "step": 3185
+    },
+    {
+      "epoch": 0.638,
+      "grad_norm": 0.43545395135879517,
+      "learning_rate": 9.711948653273438e-05,
+      "loss": 2.2478,
+      "step": 3190
+    },
+    {
+      "epoch": 0.639,
+      "grad_norm": 0.420151025056839,
+      "learning_rate": 9.665065039382376e-05,
+      "loss": 2.2099,
+      "step": 3195
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.42247268557548523,
+      "learning_rate": 9.61824101540593e-05,
+      "loss": 2.1647,
+      "step": 3200
+    },
+    {
+      "epoch": 0.641,
+      "grad_norm": 0.4133879542350769,
+      "learning_rate": 9.571477104356962e-05,
+      "loss": 2.2214,
+      "step": 3205
+    },
+    {
+      "epoch": 0.642,
+      "grad_norm": 0.41226497292518616,
+      "learning_rate": 9.524773828576888e-05,
+      "loss": 2.2337,
+      "step": 3210
+    },
+    {
+      "epoch": 0.643,
+      "grad_norm": 0.4366952180862427,
+      "learning_rate": 9.47813170972983e-05,
+      "loss": 2.213,
+      "step": 3215
+    },
+    {
+      "epoch": 0.644,
+      "grad_norm": 0.4010978937149048,
+      "learning_rate": 9.43155126879682e-05,
+      "loss": 2.2454,
+      "step": 3220
+    },
+    {
+      "epoch": 0.645,
+      "grad_norm": 0.4122413992881775,
+      "learning_rate": 9.385033026069957e-05,
+      "loss": 2.2107,
+      "step": 3225
+    },
+    {
+      "epoch": 0.646,
+      "grad_norm": 0.4530417323112488,
+      "learning_rate": 9.338577501146599e-05,
+      "loss": 2.2506,
+      "step": 3230
+    },
+    {
+      "epoch": 0.647,
+      "grad_norm": 0.44792503118515015,
+      "learning_rate": 9.292185212923554e-05,
+      "loss": 2.267,
+      "step": 3235
+    },
+    {
+      "epoch": 0.648,
+      "grad_norm": 0.4253360331058502,
+      "learning_rate": 9.245856679591302e-05,
+      "loss": 2.1882,
+      "step": 3240
+    },
+    {
+      "epoch": 0.649,
+      "grad_norm": 0.42117589712142944,
+      "learning_rate": 9.199592418628206e-05,
+      "loss": 2.2687,
+      "step": 3245
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.43629190325737,
+      "learning_rate": 9.153392946794694e-05,
+      "loss": 2.2089,
+      "step": 3250
+    },
+    {
+      "epoch": 0.651,
+      "grad_norm": 0.419784814119339,
+      "learning_rate": 9.107258780127536e-05,
+      "loss": 2.157,
+      "step": 3255
+    },
+    {
+      "epoch": 0.652,
+      "grad_norm": 0.40995609760284424,
+      "learning_rate": 9.061190433934059e-05,
+      "loss": 2.233,
+      "step": 3260
+    },
+    {
+      "epoch": 0.653,
+      "grad_norm": 0.43481624126434326,
+      "learning_rate": 9.015188422786391e-05,
+      "loss": 2.1447,
+      "step": 3265
+    },
+    {
+      "epoch": 0.654,
+      "grad_norm": 0.40843117237091064,
+      "learning_rate": 8.9692532605157e-05,
+      "loss": 2.1295,
+      "step": 3270
+    },
+    {
+      "epoch": 0.655,
+      "grad_norm": 0.42146334052085876,
+      "learning_rate": 8.92338546020648e-05,
+      "loss": 2.1455,
+      "step": 3275
+    },
+    {
+      "epoch": 0.656,
+      "grad_norm": 0.3924829661846161,
+      "learning_rate": 8.877585534190814e-05,
+      "loss": 2.1831,
+      "step": 3280
+    },
+    {
+      "epoch": 0.657,
+      "grad_norm": 0.40246203541755676,
+      "learning_rate": 8.831853994042638e-05,
+      "loss": 2.2533,
+      "step": 3285
+    },
+    {
+      "epoch": 0.658,
+      "grad_norm": 0.4149978458881378,
+      "learning_rate": 8.786191350572032e-05,
+      "loss": 2.1951,
+      "step": 3290
+    },
+    {
+      "epoch": 0.659,
+      "grad_norm": 0.4230225384235382,
+      "learning_rate": 8.740598113819519e-05,
+      "loss": 2.2113,
+      "step": 3295
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.41584622859954834,
+      "learning_rate": 8.695074793050369e-05,
+      "loss": 2.2119,
+      "step": 3300
+    },
+    {
+      "epoch": 0.661,
+      "grad_norm": 0.4212537407875061,
+      "learning_rate": 8.649621896748913e-05,
+      "loss": 2.235,
+      "step": 3305
+    },
+    {
+      "epoch": 0.662,
+      "grad_norm": 0.39631447196006775,
+      "learning_rate": 8.604239932612844e-05,
+      "loss": 2.2289,
+      "step": 3310
+    },
+    {
+      "epoch": 0.663,
+      "grad_norm": 0.43470436334609985,
+      "learning_rate": 8.558929407547562e-05,
+      "loss": 2.1882,
+      "step": 3315
+    },
+    {
+      "epoch": 0.664,
+      "grad_norm": 0.4370731711387634,
+      "learning_rate": 8.513690827660527e-05,
+      "loss": 2.1841,
+      "step": 3320
+    },
+    {
+      "epoch": 0.665,
+      "grad_norm": 0.41377973556518555,
+      "learning_rate": 8.468524698255575e-05,
+      "loss": 2.1727,
+      "step": 3325
+    },
+    {
+      "epoch": 0.666,
+      "grad_norm": 0.3933526575565338,
+      "learning_rate": 8.42343152382728e-05,
+      "loss": 2.1645,
+      "step": 3330
+    },
+    {
+      "epoch": 0.667,
+      "grad_norm": 0.4057426154613495,
+      "learning_rate": 8.378411808055346e-05,
+      "loss": 2.1732,
+      "step": 3335
+    },
+    {
+      "epoch": 0.668,
+      "grad_norm": 0.45039045810699463,
+      "learning_rate": 8.333466053798954e-05,
+      "loss": 2.2185,
+      "step": 3340
+    },
+    {
+      "epoch": 0.669,
+      "grad_norm": 0.4107955992221832,
+      "learning_rate": 8.28859476309115e-05,
+      "loss": 2.1829,
+      "step": 3345
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.43991073966026306,
+      "learning_rate": 8.243798437133237e-05,
+      "loss": 2.1252,
+      "step": 3350
+    },
+    {
+      "epoch": 0.671,
+      "grad_norm": 0.4022817611694336,
+      "learning_rate": 8.199077576289188e-05,
+      "loss": 2.1649,
+      "step": 3355
+    },
+    {
+      "epoch": 0.672,
+      "grad_norm": 0.4041346311569214,
+      "learning_rate": 8.154432680080062e-05,
+      "loss": 2.1329,
+      "step": 3360
+    },
+    {
+      "epoch": 0.673,
+      "grad_norm": 0.4271688759326935,
+      "learning_rate": 8.109864247178372e-05,
+      "loss": 2.2025,
+      "step": 3365
+    },
+    {
+      "epoch": 0.674,
+      "grad_norm": 0.4121275842189789,
+      "learning_rate": 8.065372775402587e-05,
+      "loss": 2.2487,
+      "step": 3370
+    },
+    {
+      "epoch": 0.675,
+      "grad_norm": 0.42577147483825684,
+      "learning_rate": 8.020958761711541e-05,
+      "loss": 2.2157,
+      "step": 3375
+    },
+    {
+      "epoch": 0.676,
+      "grad_norm": 0.4191531240940094,
+      "learning_rate": 7.976622702198865e-05,
+      "loss": 2.1838,
+      "step": 3380
+    },
+    {
+      "epoch": 0.677,
+      "grad_norm": 0.4344123601913452,
+      "learning_rate": 7.932365092087456e-05,
+      "loss": 2.2468,
+      "step": 3385
+    },
+    {
+      "epoch": 0.678,
+      "grad_norm": 0.4339258670806885,
+      "learning_rate": 7.888186425723967e-05,
+      "loss": 2.2697,
+      "step": 3390
+    },
+    {
+      "epoch": 0.679,
+      "grad_norm": 0.4105294942855835,
+      "learning_rate": 7.844087196573273e-05,
+      "loss": 2.1848,
+      "step": 3395
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.4143899381160736,
+      "learning_rate": 7.800067897212935e-05,
+      "loss": 2.1466,
+      "step": 3400
+    },
+    {
+      "epoch": 0.681,
+      "grad_norm": 0.4071793854236603,
+      "learning_rate": 7.756129019327723e-05,
+      "loss": 2.1962,
+      "step": 3405
+    },
+    {
+      "epoch": 0.682,
+      "grad_norm": 0.410900354385376,
+      "learning_rate": 7.712271053704131e-05,
+      "loss": 2.1805,
+      "step": 3410
+    },
+    {
+      "epoch": 0.683,
+      "grad_norm": 0.39401447772979736,
+      "learning_rate": 7.668494490224872e-05,
+      "loss": 2.1064,
+      "step": 3415
+    },
+    {
+      "epoch": 0.684,
+      "grad_norm": 0.4203227162361145,
+      "learning_rate": 7.624799817863411e-05,
+      "loss": 2.2015,
+      "step": 3420
+    },
+    {
+      "epoch": 0.685,
+      "grad_norm": 0.4399988353252411,
+      "learning_rate": 7.581187524678525e-05,
+      "loss": 2.1792,
+      "step": 3425
+    },
+    {
+      "epoch": 0.686,
+      "grad_norm": 0.41790443658828735,
+      "learning_rate": 7.53765809780884e-05,
+      "loss": 2.2491,
+      "step": 3430
+    },
+    {
+      "epoch": 0.687,
+      "grad_norm": 0.40690094232559204,
+      "learning_rate": 7.494212023467359e-05,
+      "loss": 2.2041,
+      "step": 3435
+    },
+    {
+      "epoch": 0.688,
+      "grad_norm": 0.411528617143631,
+      "learning_rate": 7.450849786936075e-05,
+      "loss": 2.1019,
+      "step": 3440
+    },
+    {
+      "epoch": 0.689,
+      "grad_norm": 0.443026602268219,
+      "learning_rate": 7.407571872560541e-05,
+      "loss": 2.1692,
+      "step": 3445
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.4203071892261505,
+      "learning_rate": 7.364378763744429e-05,
+      "loss": 2.1624,
+      "step": 3450
+    },
+    {
+      "epoch": 0.691,
+      "grad_norm": 0.40647628903388977,
+      "learning_rate": 7.321270942944182e-05,
+      "loss": 2.2838,
+      "step": 3455
+    },
+    {
+      "epoch": 0.692,
+      "grad_norm": 0.4123462438583374,
+      "learning_rate": 7.278248891663567e-05,
+      "loss": 2.08,
+      "step": 3460
+    },
+    {
+      "epoch": 0.693,
+      "grad_norm": 0.3882538676261902,
+      "learning_rate": 7.235313090448357e-05,
+      "loss": 2.1054,
+      "step": 3465
+    },
+    {
+      "epoch": 0.694,
+      "grad_norm": 0.44436848163604736,
+      "learning_rate": 7.192464018880904e-05,
+      "loss": 2.1851,
+      "step": 3470
+    },
+    {
+      "epoch": 0.695,
+      "grad_norm": 0.45201411843299866,
+      "learning_rate": 7.149702155574834e-05,
+      "loss": 2.2164,
+      "step": 3475
+    },
+    {
+      "epoch": 0.696,
+      "grad_norm": 0.4433143734931946,
+      "learning_rate": 7.10702797816967e-05,
+      "loss": 2.1391,
+      "step": 3480
+    },
+    {
+      "epoch": 0.697,
+      "grad_norm": 0.4341137409210205,
+      "learning_rate": 7.064441963325487e-05,
+      "loss": 2.1427,
+      "step": 3485
+    },
+    {
+      "epoch": 0.698,
+      "grad_norm": 0.42594924569129944,
+      "learning_rate": 7.021944586717639e-05,
+      "loss": 2.2405,
+      "step": 3490
+    },
+    {
+      "epoch": 0.699,
+      "grad_norm": 0.4368554651737213,
+      "learning_rate": 6.979536323031396e-05,
+      "loss": 2.2518,
+      "step": 3495
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.3928126096725464,
+      "learning_rate": 6.937217645956655e-05,
+      "loss": 2.1556,
+      "step": 3500
+    },
+    {
+      "epoch": 0.701,
+      "grad_norm": 0.3956170082092285,
+      "learning_rate": 6.894989028182651e-05,
+      "loss": 2.1703,
+      "step": 3505
+    },
+    {
+      "epoch": 0.702,
+      "grad_norm": 0.39387911558151245,
+      "learning_rate": 6.852850941392695e-05,
+      "loss": 2.1718,
+      "step": 3510
+    },
+    {
+      "epoch": 0.703,
+      "grad_norm": 0.4357449412345886,
+      "learning_rate": 6.810803856258872e-05,
+      "loss": 2.223,
+      "step": 3515
+    },
+    {
+      "epoch": 0.704,
+      "grad_norm": 0.42475512623786926,
+      "learning_rate": 6.768848242436801e-05,
+      "loss": 2.2292,
+      "step": 3520
+    },
+    {
+      "epoch": 0.705,
+      "grad_norm": 0.44198477268218994,
+      "learning_rate": 6.7269845685604e-05,
+      "loss": 2.1521,
+      "step": 3525
+    },
+    {
+      "epoch": 0.706,
+      "grad_norm": 0.4247047007083893,
+      "learning_rate": 6.685213302236643e-05,
+      "loss": 2.1793,
+      "step": 3530
+    },
+    {
+      "epoch": 0.707,
+      "grad_norm": 0.4207407236099243,
+      "learning_rate": 6.643534910040319e-05,
+      "loss": 2.2442,
+      "step": 3535
+    },
+    {
+      "epoch": 0.708,
+      "grad_norm": 0.4190295934677124,
+      "learning_rate": 6.601949857508839e-05,
+      "loss": 2.234,
+      "step": 3540
+    },
+    {
+      "epoch": 0.709,
+      "grad_norm": 0.44241568446159363,
+      "learning_rate": 6.560458609137043e-05,
+      "loss": 2.1567,
+      "step": 3545
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.5026627779006958,
+      "learning_rate": 6.519061628372005e-05,
+      "loss": 2.1802,
+      "step": 3550
+    },
+    {
+      "epoch": 0.711,
+      "grad_norm": 0.4083501100540161,
+      "learning_rate": 6.477759377607821e-05,
+      "loss": 2.2652,
+      "step": 3555
+    },
+    {
+      "epoch": 0.712,
+      "grad_norm": 0.40966013073921204,
+      "learning_rate": 6.436552318180513e-05,
+      "loss": 2.1943,
+      "step": 3560
+    },
+    {
+      "epoch": 0.713,
+      "grad_norm": 0.4204246997833252,
+      "learning_rate": 6.395440910362828e-05,
+      "loss": 2.2324,
+      "step": 3565
+    },
+    {
+      "epoch": 0.714,
+      "grad_norm": 0.4089759290218353,
+      "learning_rate": 6.354425613359101e-05,
+      "loss": 2.2376,
+      "step": 3570
+    },
+    {
+      "epoch": 0.715,
+      "grad_norm": 0.40603703260421753,
+      "learning_rate": 6.31350688530013e-05,
+      "loss": 2.2388,
+      "step": 3575
+    },
+    {
+      "epoch": 0.716,
+      "grad_norm": 0.4146242141723633,
+      "learning_rate": 6.272685183238074e-05,
+      "loss": 2.1973,
+      "step": 3580
+    },
+    {
+      "epoch": 0.717,
+      "grad_norm": 0.4212912619113922,
+      "learning_rate": 6.231960963141334e-05,
+      "loss": 2.1458,
+      "step": 3585
+    },
+    {
+      "epoch": 0.718,
+      "grad_norm": 0.41953805088996887,
+      "learning_rate": 6.191334679889455e-05,
+      "loss": 2.176,
+      "step": 3590
+    },
+    {
+      "epoch": 0.719,
+      "grad_norm": 0.40272924304008484,
+      "learning_rate": 6.150806787268044e-05,
+      "loss": 2.1482,
+      "step": 3595
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.4300851821899414,
+      "learning_rate": 6.110377737963731e-05,
+      "loss": 2.204,
+      "step": 3600
+    },
+    {
+      "epoch": 0.721,
+      "grad_norm": 0.4585989713668823,
+      "learning_rate": 6.0700479835590733e-05,
+      "loss": 2.1059,
+      "step": 3605
+    },
+    {
+      "epoch": 0.722,
+      "grad_norm": 0.3827567994594574,
+      "learning_rate": 6.029817974527527e-05,
+      "loss": 2.11,
+      "step": 3610
+    },
+    {
+      "epoch": 0.723,
+      "grad_norm": 0.4101276695728302,
+      "learning_rate": 5.9896881602284275e-05,
+      "loss": 2.2383,
+      "step": 3615
+    },
+    {
+      "epoch": 0.724,
+      "grad_norm": 0.42217323184013367,
+      "learning_rate": 5.949658988901959e-05,
+      "loss": 2.2011,
+      "step": 3620
+    },
+    {
+      "epoch": 0.725,
+      "grad_norm": 0.419291615486145,
+      "learning_rate": 5.9097309076641424e-05,
+      "loss": 2.1178,
+      "step": 3625
+    },
+    {
+      "epoch": 0.726,
+      "grad_norm": 0.4263475835323334,
+      "learning_rate": 5.8699043625018415e-05,
+      "loss": 2.1247,
+      "step": 3630
+    },
+    {
+      "epoch": 0.727,
+      "grad_norm": 0.4190998077392578,
+      "learning_rate": 5.830179798267806e-05,
+      "loss": 2.2097,
+      "step": 3635
+    },
+    {
+      "epoch": 0.728,
+      "grad_norm": 0.41809847950935364,
+      "learning_rate": 5.79055765867566e-05,
+      "loss": 2.2182,
+      "step": 3640
+    },
+    {
+      "epoch": 0.729,
+      "grad_norm": 0.43095046281814575,
+      "learning_rate": 5.7510383862949934e-05,
+      "loss": 2.1382,
+      "step": 3645
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.4270274341106415,
+      "learning_rate": 5.711622422546374e-05,
+      "loss": 2.1623,
+      "step": 3650
+    },
+    {
+      "epoch": 0.731,
+      "grad_norm": 0.4485456645488739,
+      "learning_rate": 5.672310207696454e-05,
+      "loss": 2.2079,
+      "step": 3655
+    },
+    {
+      "epoch": 0.732,
+      "grad_norm": 0.4084364175796509,
+      "learning_rate": 5.6331021808530195e-05,
+      "loss": 2.1675,
+      "step": 3660
+    },
+    {
+      "epoch": 0.733,
+      "grad_norm": 0.4245437681674957,
+      "learning_rate": 5.5939987799601204e-05,
+      "loss": 2.2246,
+      "step": 3665
+    },
+    {
+      "epoch": 0.734,
+      "grad_norm": 0.42533889412879944,
+      "learning_rate": 5.5550004417931477e-05,
+      "loss": 2.1298,
+      "step": 3670
+    },
+    {
+      "epoch": 0.735,
+      "grad_norm": 0.40783774852752686,
+      "learning_rate": 5.516107601953962e-05,
+      "loss": 2.1108,
+      "step": 3675
+    },
+    {
+      "epoch": 0.736,
+      "grad_norm": 0.43854960799217224,
+      "learning_rate": 5.4773206948660505e-05,
+      "loss": 2.1013,
+      "step": 3680
+    },
+    {
+      "epoch": 0.737,
+      "grad_norm": 0.4022478461265564,
+      "learning_rate": 5.4386401537696536e-05,
+      "loss": 2.1055,
+      "step": 3685
+    },
+    {
+      "epoch": 0.738,
+      "grad_norm": 0.4073084592819214,
+      "learning_rate": 5.40006641071692e-05,
+      "loss": 2.1329,
+      "step": 3690
+    },
+    {
+      "epoch": 0.739,
+      "grad_norm": 0.4003308117389679,
+      "learning_rate": 5.36159989656709e-05,
+      "loss": 2.2016,
+      "step": 3695
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.43242812156677246,
+      "learning_rate": 5.3232410409817006e-05,
+      "loss": 2.1525,
+      "step": 3700
+    },
+    {
+      "epoch": 0.741,
+      "grad_norm": 0.44237396121025085,
+      "learning_rate": 5.2849902724197526e-05,
+      "loss": 2.1776,
+      "step": 3705
+    },
+    {
+      "epoch": 0.742,
+      "grad_norm": 0.424785315990448,
+      "learning_rate": 5.2468480181329384e-05,
+      "loss": 2.1294,
+      "step": 3710
+    },
+    {
+      "epoch": 0.743,
+      "grad_norm": 0.4005042612552643,
+      "learning_rate": 5.208814704160888e-05,
+      "loss": 2.1107,
+      "step": 3715
+    },
+    {
+      "epoch": 0.744,
+      "grad_norm": 0.41858258843421936,
+      "learning_rate": 5.1708907553263925e-05,
+      "loss": 2.1498,
+      "step": 3720
+    },
+    {
+      "epoch": 0.745,
+      "grad_norm": 0.4518043100833893,
+      "learning_rate": 5.13307659523065e-05,
+      "loss": 2.1934,
+      "step": 3725
+    },
+    {
+      "epoch": 0.746,
+      "grad_norm": 0.42560485005378723,
+      "learning_rate": 5.095372646248547e-05,
+      "loss": 2.2194,
+      "step": 3730
+    },
+    {
+      "epoch": 0.747,
+      "grad_norm": 0.4324474632740021,
+      "learning_rate": 5.057779329523947e-05,
+      "loss": 2.1427,
+      "step": 3735
+    },
+    {
+      "epoch": 0.748,
+      "grad_norm": 0.4265004098415375,
+      "learning_rate": 5.020297064964985e-05,
+      "loss": 2.1932,
+      "step": 3740
+    },
+    {
+      "epoch": 0.749,
+      "grad_norm": 0.4324226975440979,
+      "learning_rate": 4.982926271239342e-05,
+      "loss": 2.0714,
+      "step": 3745
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.4050389230251312,
+      "learning_rate": 4.945667365769621e-05,
+      "loss": 2.1401,
+      "step": 3750
+    },
+    {
+      "epoch": 0.751,
+      "grad_norm": 0.4235606789588928,
+      "learning_rate": 4.908520764728665e-05,
+      "loss": 2.2192,
+      "step": 3755
+    },
+    {
+      "epoch": 0.752,
+      "grad_norm": 0.446185439825058,
+      "learning_rate": 4.8714868830348854e-05,
+      "loss": 2.1989,
+      "step": 3760
+    },
+    {
+      "epoch": 0.753,
+      "grad_norm": 0.39757952094078064,
+      "learning_rate": 4.83456613434765e-05,
+      "loss": 2.1534,
+      "step": 3765
+    },
+    {
+      "epoch": 0.754,
+      "grad_norm": 0.42775917053222656,
+      "learning_rate": 4.7977589310626704e-05,
+      "loss": 2.17,
+      "step": 3770
+    },
+    {
+      "epoch": 0.755,
+      "grad_norm": 0.4254078269004822,
+      "learning_rate": 4.761065684307383e-05,
+      "loss": 2.1444,
+      "step": 3775
+    },
+    {
+      "epoch": 0.756,
+      "grad_norm": 0.416073739528656,
+      "learning_rate": 4.724486803936337e-05,
+      "loss": 2.1706,
+      "step": 3780
+    },
+    {
+      "epoch": 0.757,
+      "grad_norm": 0.4071180820465088,
+      "learning_rate": 4.688022698526658e-05,
+      "loss": 2.1173,
+      "step": 3785
+    },
+    {
+      "epoch": 0.758,
+      "grad_norm": 0.41832828521728516,
+      "learning_rate": 4.651673775373463e-05,
+      "loss": 2.2064,
+      "step": 3790
+    },
+    {
+      "epoch": 0.759,
+      "grad_norm": 0.41309747099876404,
+      "learning_rate": 4.615440440485299e-05,
+      "loss": 2.1729,
+      "step": 3795
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.43315672874450684,
+      "learning_rate": 4.579323098579626e-05,
+      "loss": 2.2041,
+      "step": 3800
+    },
+    {
+      "epoch": 0.761,
+      "grad_norm": 0.43665811419487,
+      "learning_rate": 4.5433221530782945e-05,
+      "loss": 2.1855,
+      "step": 3805
+    },
+    {
+      "epoch": 0.762,
+      "grad_norm": 0.4120488166809082,
+      "learning_rate": 4.5074380061030366e-05,
+      "loss": 2.2696,
+      "step": 3810
+    },
+    {
+      "epoch": 0.763,
+      "grad_norm": 0.41258561611175537,
+      "learning_rate": 4.471671058470968e-05,
+      "loss": 2.2001,
+      "step": 3815
+    },
+    {
+      "epoch": 0.764,
+      "grad_norm": 0.4156684875488281,
+      "learning_rate": 4.436021709690113e-05,
+      "loss": 2.1246,
+      "step": 3820
+    },
+    {
+      "epoch": 0.765,
+      "grad_norm": 0.4078438878059387,
+      "learning_rate": 4.400490357954959e-05,
+      "loss": 2.1279,
+      "step": 3825
+    },
+    {
+      "epoch": 0.766,
+      "grad_norm": 0.41062772274017334,
+      "learning_rate": 4.365077400141981e-05,
+      "loss": 2.2157,
+      "step": 3830
+    },
+    {
+      "epoch": 0.767,
+      "grad_norm": 0.4282727539539337,
+      "learning_rate": 4.329783231805235e-05,
+      "loss": 2.3185,
+      "step": 3835
+    },
+    {
+      "epoch": 0.768,
+      "grad_norm": 0.4196500778198242,
+      "learning_rate": 4.294608247171917e-05,
+      "loss": 2.0916,
+      "step": 3840
+    },
+    {
+      "epoch": 0.769,
+      "grad_norm": 0.4065400958061218,
+      "learning_rate": 4.259552839137982e-05,
+      "loss": 2.1515,
+      "step": 3845
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.4283042550086975,
+      "learning_rate": 4.2246173992637276e-05,
+      "loss": 2.1246,
+      "step": 3850
+    },
+    {
+      "epoch": 0.771,
+      "grad_norm": 0.4189644753932953,
+      "learning_rate": 4.1898023177694563e-05,
+      "loss": 2.1942,
+      "step": 3855
+    },
+    {
+      "epoch": 0.772,
+      "grad_norm": 0.4234643876552582,
+      "learning_rate": 4.1551079835310816e-05,
+      "loss": 2.0987,
+      "step": 3860
+    },
+    {
+      "epoch": 0.773,
+      "grad_norm": 0.4286406934261322,
+      "learning_rate": 4.120534784075802e-05,
+      "loss": 2.109,
+      "step": 3865
+    },
+    {
+      "epoch": 0.774,
+      "grad_norm": 0.42778655886650085,
+      "learning_rate": 4.086083105577779e-05,
+      "loss": 2.1574,
+      "step": 3870
+    },
+    {
+      "epoch": 0.775,
+      "grad_norm": 0.40660133957862854,
+      "learning_rate": 4.051753332853803e-05,
+      "loss": 2.1345,
+      "step": 3875
+    },
+    {
+      "epoch": 0.776,
+      "grad_norm": 0.41279226541519165,
+      "learning_rate": 4.0175458493590195e-05,
+      "loss": 2.0764,
+      "step": 3880
+    },
+    {
+      "epoch": 0.777,
+      "grad_norm": 0.40562158823013306,
+      "learning_rate": 3.98346103718262e-05,
+      "loss": 2.1556,
+      "step": 3885
+    },
+    {
+      "epoch": 0.778,
+      "grad_norm": 0.42849069833755493,
+      "learning_rate": 3.9494992770436014e-05,
+      "loss": 2.1131,
+      "step": 3890
+    },
+    {
+      "epoch": 0.779,
+      "grad_norm": 0.4102763831615448,
+      "learning_rate": 3.915660948286486e-05,
+      "loss": 2.1785,
+      "step": 3895
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.44005340337753296,
+      "learning_rate": 3.8819464288771026e-05,
+      "loss": 2.159,
+      "step": 3900
+    },
+    {
+      "epoch": 0.781,
+      "grad_norm": 0.42726826667785645,
+      "learning_rate": 3.848356095398364e-05,
+      "loss": 2.1305,
+      "step": 3905
+    },
+    {
+      "epoch": 0.782,
+      "grad_norm": 0.4183489680290222,
+      "learning_rate": 3.81489032304606e-05,
+      "loss": 2.1276,
+      "step": 3910
+    },
+    {
+      "epoch": 0.783,
+      "grad_norm": 0.4101635217666626,
+      "learning_rate": 3.781549485624651e-05,
+      "loss": 2.1621,
+      "step": 3915
+    },
+    {
+      "epoch": 0.784,
+      "grad_norm": 0.4038468897342682,
+      "learning_rate": 3.7483339555431055e-05,
+      "loss": 2.1201,
+      "step": 3920
+    },
+    {
+      "epoch": 0.785,
+      "grad_norm": 0.4497060477733612,
+      "learning_rate": 3.715244103810755e-05,
+      "loss": 2.1344,
+      "step": 3925
+    },
+    {
+      "epoch": 0.786,
+      "grad_norm": 0.42213934659957886,
+      "learning_rate": 3.682280300033128e-05,
+      "loss": 2.1957,
+      "step": 3930
+    },
+    {
+      "epoch": 0.787,
+      "grad_norm": 0.4279301166534424,
+      "learning_rate": 3.649442912407811e-05,
+      "loss": 2.1426,
+      "step": 3935
+    },
+    {
+      "epoch": 0.788,
+      "grad_norm": 0.42436257004737854,
+      "learning_rate": 3.616732307720373e-05,
+      "loss": 2.1294,
+      "step": 3940
+    },
+    {
+      "epoch": 0.789,
+      "grad_norm": 0.4053291380405426,
+      "learning_rate": 3.584148851340252e-05,
+      "loss": 2.1124,
+      "step": 3945
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.38851654529571533,
+      "learning_rate": 3.551692907216652e-05,
+      "loss": 2.1285,
+      "step": 3950
+    },
+    {
+      "epoch": 0.791,
+      "grad_norm": 0.4097929894924164,
+      "learning_rate": 3.519364837874506e-05,
+      "loss": 2.1942,
+      "step": 3955
+    },
+    {
+      "epoch": 0.792,
+      "grad_norm": 0.42528805136680603,
+      "learning_rate": 3.48716500441042e-05,
+      "loss": 2.2122,
+      "step": 3960
+    },
+    {
+      "epoch": 0.793,
+      "grad_norm": 0.4303857386112213,
+      "learning_rate": 3.455093766488641e-05,
+      "loss": 2.1138,
+      "step": 3965
+    },
+    {
+      "epoch": 0.794,
+      "grad_norm": 0.40125489234924316,
+      "learning_rate": 3.423151482337013e-05,
+      "loss": 2.0879,
+      "step": 3970
+    },
+    {
+      "epoch": 0.795,
+      "grad_norm": 0.41210323572158813,
+      "learning_rate": 3.3913385087430204e-05,
+      "loss": 2.0709,
+      "step": 3975
+    },
+    {
+      "epoch": 0.796,
+      "grad_norm": 0.41345149278640747,
+      "learning_rate": 3.359655201049778e-05,
+      "loss": 2.1647,
+      "step": 3980
+    },
+    {
+      "epoch": 0.797,
+      "grad_norm": 0.41159018874168396,
+      "learning_rate": 3.3281019131520534e-05,
+      "loss": 2.216,
+      "step": 3985
+    },
+    {
+      "epoch": 0.798,
+      "grad_norm": 0.432202011346817,
+      "learning_rate": 3.2966789974923295e-05,
+      "loss": 2.122,
+      "step": 3990
+    },
+    {
+      "epoch": 0.799,
+      "grad_norm": 0.42457619309425354,
+      "learning_rate": 3.2653868050568695e-05,
+      "loss": 2.1532,
+      "step": 3995
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.4103912115097046,
+      "learning_rate": 3.2342256853717857e-05,
+      "loss": 2.0655,
+      "step": 4000
+    },
+    {
+      "epoch": 0.801,
+      "grad_norm": 0.4154665470123291,
+      "learning_rate": 3.203195986499138e-05,
+      "loss": 2.1186,
+      "step": 4005
+    },
+    {
+      "epoch": 0.802,
+      "grad_norm": 0.40893417596817017,
+      "learning_rate": 3.1722980550330465e-05,
+      "loss": 2.1407,
+      "step": 4010
+    },
+    {
+      "epoch": 0.803,
+      "grad_norm": 0.4399654269218445,
+      "learning_rate": 3.141532236095833e-05,
+      "loss": 2.0957,
+      "step": 4015
+    },
+    {
+      "epoch": 0.804,
+      "grad_norm": 0.4194296598434448,
+      "learning_rate": 3.110898873334137e-05,
+      "loss": 2.0873,
+      "step": 4020
+    },
+    {
+      "epoch": 0.805,
+      "grad_norm": 0.4045575261116028,
+      "learning_rate": 3.080398308915116e-05,
+      "loss": 2.2489,
+      "step": 4025
+    },
+    {
+      "epoch": 0.806,
+      "grad_norm": 0.42948484420776367,
+      "learning_rate": 3.0500308835225794e-05,
+      "loss": 2.1339,
+      "step": 4030
+    },
+    {
+      "epoch": 0.807,
+      "grad_norm": 0.40769994258880615,
+      "learning_rate": 3.0197969363532294e-05,
+      "loss": 2.123,
+      "step": 4035
+    },
+    {
+      "epoch": 0.808,
+      "grad_norm": 0.42292699217796326,
+      "learning_rate": 2.98969680511283e-05,
+      "loss": 2.1557,
+      "step": 4040
+    },
+    {
+      "epoch": 0.809,
+      "grad_norm": 0.4049204885959625,
+      "learning_rate": 2.9597308260124676e-05,
+      "loss": 2.2234,
+      "step": 4045
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.4200775623321533,
+      "learning_rate": 2.9298993337647776e-05,
+      "loss": 2.145,
+      "step": 4050
+    },
+    {
+      "epoch": 0.811,
+      "grad_norm": 0.4226503074169159,
+      "learning_rate": 2.900202661580201e-05,
+      "loss": 2.0963,
+      "step": 4055
+    },
+    {
+      "epoch": 0.812,
+      "grad_norm": 0.4314316213130951,
+      "learning_rate": 2.8706411411632886e-05,
+      "loss": 2.2098,
+      "step": 4060
+    },
+    {
+      "epoch": 0.813,
+      "grad_norm": 0.44541093707084656,
+      "learning_rate": 2.8412151027089663e-05,
+      "loss": 2.0709,
+      "step": 4065
+    },
+    {
+      "epoch": 0.814,
+      "grad_norm": 0.4003683030605316,
+      "learning_rate": 2.8119248748988682e-05,
+      "loss": 2.1063,
+      "step": 4070
+    },
+    {
+      "epoch": 0.815,
+      "grad_norm": 0.4231080412864685,
+      "learning_rate": 2.782770784897646e-05,
+      "loss": 2.1509,
+      "step": 4075
+    },
+    {
+      "epoch": 0.816,
+      "grad_norm": 0.41555121541023254,
+      "learning_rate": 2.7537531583493415e-05,
+      "loss": 2.1913,
+      "step": 4080
+    },
+    {
+      "epoch": 0.817,
+      "grad_norm": 0.42107847332954407,
+      "learning_rate": 2.7248723193737193e-05,
+      "loss": 2.1571,
+      "step": 4085
+    },
+    {
+      "epoch": 0.818,
+      "grad_norm": 0.43069687485694885,
+      "learning_rate": 2.6961285905626605e-05,
+      "loss": 2.1125,
+      "step": 4090
+    },
+    {
+      "epoch": 0.819,
+      "grad_norm": 0.4340486526489258,
+      "learning_rate": 2.667522292976569e-05,
+      "loss": 2.1622,
+      "step": 4095
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.41151127219200134,
+      "learning_rate": 2.639053746140772e-05,
+      "loss": 2.1475,
+      "step": 4100
+    },
+    {
+      "epoch": 0.821,
+      "grad_norm": 0.4273358881473541,
+      "learning_rate": 2.6107232680419514e-05,
+      "loss": 2.1178,
+      "step": 4105
+    },
+    {
+      "epoch": 0.822,
+      "grad_norm": 0.42515695095062256,
+      "learning_rate": 2.582531175124592e-05,
+      "loss": 2.0993,
+      "step": 4110
+    },
+    {
+      "epoch": 0.823,
+      "grad_norm": 0.40709659457206726,
+      "learning_rate": 2.5544777822874585e-05,
+      "loss": 2.1192,
+      "step": 4115
+    },
+    {
+      "epoch": 0.824,
+      "grad_norm": 0.402413010597229,
+      "learning_rate": 2.526563402880074e-05,
+      "loss": 2.1685,
+      "step": 4120
+    },
+    {
+      "epoch": 0.825,
+      "grad_norm": 0.42556092143058777,
+      "learning_rate": 2.498788348699194e-05,
+      "loss": 2.1758,
+      "step": 4125
+    },
+    {
+      "epoch": 0.826,
+      "grad_norm": 0.41573649644851685,
+      "learning_rate": 2.471152929985366e-05,
+      "loss": 2.117,
+      "step": 4130
+    },
+    {
+      "epoch": 0.827,
+      "grad_norm": 0.393812894821167,
+      "learning_rate": 2.44365745541944e-05,
+      "loss": 2.133,
+      "step": 4135
+    },
+    {
+      "epoch": 0.828,
+      "grad_norm": 0.39863842725753784,
+      "learning_rate": 2.4163022321191183e-05,
+      "loss": 2.1545,
+      "step": 4140
+    },
+    {
+      "epoch": 0.829,
+      "grad_norm": 0.40823498368263245,
+      "learning_rate": 2.389087565635529e-05,
+      "loss": 2.098,
+      "step": 4145
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.39885497093200684,
+      "learning_rate": 2.362013759949824e-05,
+      "loss": 2.0917,
+      "step": 4150
+    },
+    {
+      "epoch": 0.831,
+      "grad_norm": 0.4393814504146576,
+      "learning_rate": 2.335081117469777e-05,
+      "loss": 2.208,
+      "step": 4155
+    },
+    {
+      "epoch": 0.832,
+      "grad_norm": 0.3837684094905853,
+      "learning_rate": 2.3082899390263833e-05,
+      "loss": 2.1412,
+      "step": 4160
+    },
+    {
+      "epoch": 0.833,
+      "grad_norm": 0.4429077208042145,
+      "learning_rate": 2.28164052387054e-05,
+      "loss": 2.1168,
+      "step": 4165
+    },
+    {
+      "epoch": 0.834,
+      "grad_norm": 0.4012123644351959,
+      "learning_rate": 2.2551331696696797e-05,
+      "loss": 2.1188,
+      "step": 4170
+    },
+    {
+      "epoch": 0.835,
+      "grad_norm": 0.4135606288909912,
+      "learning_rate": 2.228768172504442e-05,
+      "loss": 2.2259,
+      "step": 4175
+    },
+    {
+      "epoch": 0.836,
+      "grad_norm": 0.441450834274292,
+      "learning_rate": 2.2025458268653727e-05,
+      "loss": 2.2108,
+      "step": 4180
+    },
+    {
+      "epoch": 0.837,
+      "grad_norm": 0.4205508828163147,
+      "learning_rate": 2.1764664256496433e-05,
+      "loss": 2.1244,
+      "step": 4185
+    },
+    {
+      "epoch": 0.838,
+      "grad_norm": 0.41583573818206787,
+      "learning_rate": 2.150530260157769e-05,
+      "loss": 2.1185,
+      "step": 4190
+    },
+    {
+      "epoch": 0.839,
+      "grad_norm": 0.425163209438324,
+      "learning_rate": 2.1247376200903534e-05,
+      "loss": 2.1598,
+      "step": 4195
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.4113369286060333,
+      "learning_rate": 2.0990887935448537e-05,
+      "loss": 2.0728,
+      "step": 4200
+    },
+    {
+      "epoch": 0.841,
+      "grad_norm": 0.41494420170783997,
+      "learning_rate": 2.073584067012376e-05,
+      "loss": 2.1892,
+      "step": 4205
+    },
+    {
+      "epoch": 0.842,
+      "grad_norm": 0.4220603108406067,
+      "learning_rate": 2.0482237253744538e-05,
+      "loss": 2.0895,
+      "step": 4210
+    },
+    {
+      "epoch": 0.843,
+      "grad_norm": 0.43092748522758484,
+      "learning_rate": 2.023008051899884e-05,
+      "loss": 2.1202,
+      "step": 4215
+    },
+    {
+      "epoch": 0.844,
+      "grad_norm": 0.39946693181991577,
+      "learning_rate": 1.9979373282415482e-05,
+      "loss": 2.1241,
+      "step": 4220
+    },
+    {
+      "epoch": 0.845,
+      "grad_norm": 0.4280877113342285,
+      "learning_rate": 1.9730118344332874e-05,
+      "loss": 2.2293,
+      "step": 4225
+    },
+    {
+      "epoch": 0.846,
+      "grad_norm": 0.407251238822937,
+      "learning_rate": 1.9482318488867458e-05,
+      "loss": 2.1459,
+      "step": 4230
+    },
+    {
+      "epoch": 0.847,
+      "grad_norm": 0.43046170473098755,
+      "learning_rate": 1.92359764838828e-05,
+      "loss": 2.1587,
+      "step": 4235
+    },
+    {
+      "epoch": 0.848,
+      "grad_norm": 0.4038347601890564,
+      "learning_rate": 1.899109508095872e-05,
+      "loss": 2.021,
+      "step": 4240
+    },
+    {
+      "epoch": 0.849,
+      "grad_norm": 0.42142099142074585,
+      "learning_rate": 1.874767701536035e-05,
+      "loss": 2.1318,
+      "step": 4245
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.40510687232017517,
+      "learning_rate": 1.850572500600782e-05,
+      "loss": 2.2302,
+      "step": 4250
+    },
+    {
+      "epoch": 0.851,
+      "grad_norm": 0.4129900336265564,
+      "learning_rate": 1.8265241755445665e-05,
+      "loss": 2.2122,
+      "step": 4255
+    },
+    {
+      "epoch": 0.852,
+      "grad_norm": 0.4296528995037079,
+      "learning_rate": 1.8026229949812826e-05,
+      "loss": 2.235,
+      "step": 4260
+    },
+    {
+      "epoch": 0.853,
+      "grad_norm": 0.43158674240112305,
+      "learning_rate": 1.77886922588125e-05,
+      "loss": 2.1896,
+      "step": 4265
+    },
+    {
+      "epoch": 0.854,
+      "grad_norm": 0.3998924195766449,
+      "learning_rate": 1.755263133568246e-05,
+      "loss": 2.09,
+      "step": 4270
+    },
+    {
+      "epoch": 0.855,
+      "grad_norm": 0.41046082973480225,
+      "learning_rate": 1.7318049817165285e-05,
+      "loss": 2.2035,
+      "step": 4275
+    },
+    {
+      "epoch": 0.856,
+      "grad_norm": 0.39339983463287354,
+      "learning_rate": 1.7084950323478935e-05,
+      "loss": 2.1128,
+      "step": 4280
+    },
+    {
+      "epoch": 0.857,
+      "grad_norm": 0.41789743304252625,
+      "learning_rate": 1.685333545828761e-05,
+      "loss": 2.1941,
+      "step": 4285
+    },
+    {
+      "epoch": 0.858,
+      "grad_norm": 0.7934504747390747,
+      "learning_rate": 1.6623207808672556e-05,
+      "loss": 2.1591,
+      "step": 4290
+    },
+    {
+      "epoch": 0.859,
+      "grad_norm": 0.42379069328308105,
+      "learning_rate": 1.6394569945103116e-05,
+      "loss": 2.1738,
+      "step": 4295
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.4412902593612671,
+      "learning_rate": 1.6167424421408117e-05,
+      "loss": 2.0838,
+      "step": 4300
+    },
+    {
+      "epoch": 0.861,
+      "grad_norm": 0.42037296295166016,
+      "learning_rate": 1.594177377474736e-05,
+      "loss": 2.1533,
+      "step": 4305
+    },
+    {
+      "epoch": 0.862,
+      "grad_norm": 0.40737658739089966,
+      "learning_rate": 1.5717620525583264e-05,
+      "loss": 2.2523,
+      "step": 4310
+    },
+    {
+      "epoch": 0.863,
+      "grad_norm": 0.4244195818901062,
+      "learning_rate": 1.5494967177652485e-05,
+      "loss": 2.132,
+      "step": 4315
+    },
+    {
+      "epoch": 0.864,
+      "grad_norm": 0.3916720747947693,
+      "learning_rate": 1.527381621793839e-05,
+      "loss": 2.1012,
+      "step": 4320
+    },
+    {
+      "epoch": 0.865,
+      "grad_norm": 0.4201657772064209,
+      "learning_rate": 1.5054170116642956e-05,
+      "loss": 2.1597,
+      "step": 4325
+    },
+    {
+      "epoch": 0.866,
+      "grad_norm": 0.40828606486320496,
+      "learning_rate": 1.4836031327159203e-05,
+      "loss": 2.1062,
+      "step": 4330
+    },
+    {
+      "epoch": 0.867,
+      "grad_norm": 0.43217313289642334,
+      "learning_rate": 1.4619402286043875e-05,
+      "loss": 2.0693,
+      "step": 4335
+    },
+    {
+      "epoch": 0.868,
+      "grad_norm": 0.4222164750099182,
+      "learning_rate": 1.4404285412990247e-05,
+      "loss": 2.1159,
+      "step": 4340
+    },
+    {
+      "epoch": 0.869,
+      "grad_norm": 0.4298483729362488,
+      "learning_rate": 1.419068311080106e-05,
+      "loss": 2.2043,
+      "step": 4345
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.4016087055206299,
+      "learning_rate": 1.3978597765361504e-05,
+      "loss": 2.1043,
+      "step": 4350
+    },
+    {
+      "epoch": 0.871,
+      "grad_norm": 0.42064622044563293,
+      "learning_rate": 1.3768031745612923e-05,
+      "loss": 2.2281,
+      "step": 4355
+    },
+    {
+      "epoch": 0.872,
+      "grad_norm": 0.4412085711956024,
+      "learning_rate": 1.3558987403526118e-05,
+      "loss": 2.1408,
+      "step": 4360
+    },
+    {
+      "epoch": 0.873,
+      "grad_norm": 0.42196419835090637,
+      "learning_rate": 1.3351467074075068e-05,
+      "loss": 2.1301,
+      "step": 4365
+    },
+    {
+      "epoch": 0.874,
+      "grad_norm": 0.40128079056739807,
+      "learning_rate": 1.3145473075210921e-05,
+      "loss": 2.1288,
+      "step": 4370
+    },
+    {
+      "epoch": 0.875,
+      "grad_norm": 0.40371379256248474,
+      "learning_rate": 1.2941007707836148e-05,
+      "loss": 2.1145,
+      "step": 4375
+    },
+    {
+      "epoch": 0.876,
+      "grad_norm": 0.4210174083709717,
+      "learning_rate": 1.2738073255778741e-05,
+      "loss": 2.2467,
+      "step": 4380
+    },
+    {
+      "epoch": 0.877,
+      "grad_norm": 0.412621408700943,
+      "learning_rate": 1.2536671985766722e-05,
+      "loss": 2.12,
+      "step": 4385
+    },
+    {
+      "epoch": 0.878,
+      "grad_norm": 0.40745627880096436,
+      "learning_rate": 1.2336806147402828e-05,
+      "loss": 2.1071,
+      "step": 4390
+    },
+    {
+      "epoch": 0.879,
+      "grad_norm": 0.39555612206459045,
+      "learning_rate": 1.2138477973139515e-05,
+      "loss": 2.1897,
+      "step": 4395
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.45732036232948303,
+      "learning_rate": 1.1941689678253746e-05,
+      "loss": 2.1497,
+      "step": 4400
+    },
+    {
+      "epoch": 0.881,
+      "grad_norm": 0.44343504309654236,
+      "learning_rate": 1.1746443460822574e-05,
+      "loss": 2.1451,
+      "step": 4405
+    },
+    {
+      "epoch": 0.882,
+      "grad_norm": 0.4288816452026367,
+      "learning_rate": 1.1552741501698281e-05,
+      "loss": 2.1104,
+      "step": 4410
+    },
+    {
+      "epoch": 0.883,
+      "grad_norm": 0.41571128368377686,
+      "learning_rate": 1.136058596448428e-05,
+      "loss": 2.1129,
+      "step": 4415
+    },
+    {
+      "epoch": 0.884,
+      "grad_norm": 0.4085163474082947,
+      "learning_rate": 1.1169978995510753e-05,
+      "loss": 2.1633,
+      "step": 4420
+    },
+    {
+      "epoch": 0.885,
+      "grad_norm": 0.41918131709098816,
+      "learning_rate": 1.0980922723810765e-05,
+      "loss": 2.0841,
+      "step": 4425
+    },
+    {
+      "epoch": 0.886,
+      "grad_norm": 0.42264747619628906,
+      "learning_rate": 1.079341926109652e-05,
+      "loss": 2.083,
+      "step": 4430
+    },
+    {
+      "epoch": 0.887,
+      "grad_norm": 0.4294689893722534,
+      "learning_rate": 1.0607470701735632e-05,
+      "loss": 2.1732,
+      "step": 4435
+    },
+    {
+      "epoch": 0.888,
+      "grad_norm": 0.41121116280555725,
+      "learning_rate": 1.0423079122727974e-05,
+      "loss": 2.1585,
+      "step": 4440
+    },
+    {
+      "epoch": 0.889,
+      "grad_norm": 0.4106404185295105,
+      "learning_rate": 1.0240246583682177e-05,
+      "loss": 2.1152,
+      "step": 4445
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.4227830171585083,
+      "learning_rate": 1.0058975126792873e-05,
+      "loss": 2.1444,
+      "step": 4450
+    },
+    {
+      "epoch": 0.891,
+      "grad_norm": 0.4478638470172882,
+      "learning_rate": 9.879266776817757e-06,
+      "loss": 2.1113,
+      "step": 4455
+    },
+    {
+      "epoch": 0.892,
+      "grad_norm": 0.42579948902130127,
+      "learning_rate": 9.701123541055023e-06,
+      "loss": 2.189,
+      "step": 4460
+    },
+    {
+      "epoch": 0.893,
+      "grad_norm": 0.4126085937023163,
+      "learning_rate": 9.524547409320881e-06,
+      "loss": 2.1873,
+      "step": 4465
+    },
+    {
+      "epoch": 0.894,
+      "grad_norm": 0.3965235650539398,
+      "learning_rate": 9.349540353927415e-06,
+      "loss": 2.1206,
+      "step": 4470
+    },
+    {
+      "epoch": 0.895,
+      "grad_norm": 0.44063737988471985,
+      "learning_rate": 9.176104329660472e-06,
+      "loss": 2.0752,
+      "step": 4475
+    },
+    {
+      "epoch": 0.896,
+      "grad_norm": 0.41276708245277405,
+      "learning_rate": 9.004241273757957e-06,
+      "loss": 2.1658,
+      "step": 4480
+    },
+    {
+      "epoch": 0.897,
+      "grad_norm": 0.4311666190624237,
+      "learning_rate": 8.833953105887975e-06,
+      "loss": 2.1305,
+      "step": 4485
+    },
+    {
+      "epoch": 0.898,
+      "grad_norm": 0.4143565893173218,
+      "learning_rate": 8.665241728127592e-06,
+      "loss": 2.1187,
+      "step": 4490
+    },
+    {
+      "epoch": 0.899,
+      "grad_norm": 0.40346577763557434,
+      "learning_rate": 8.498109024941541e-06,
+      "loss": 2.2537,
+      "step": 4495
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.3936259150505066,
+      "learning_rate": 8.332556863161093e-06,
+      "loss": 2.162,
+      "step": 4500
+    },
+    {
+      "epoch": 0.901,
+      "grad_norm": 0.42583733797073364,
+      "learning_rate": 8.168587091963247e-06,
+      "loss": 2.1827,
+      "step": 4505
+    },
+    {
+      "epoch": 0.902,
+      "grad_norm": 0.4225499629974365,
+      "learning_rate": 8.006201542850132e-06,
+      "loss": 2.0864,
+      "step": 4510
+    },
+    {
+      "epoch": 0.903,
+      "grad_norm": 0.3910048305988312,
+      "learning_rate": 7.84540202962854e-06,
+      "loss": 2.1557,
+      "step": 4515
+    },
+    {
+      "epoch": 0.904,
+      "grad_norm": 0.42542725801467896,
+      "learning_rate": 7.686190348389542e-06,
+      "loss": 2.0824,
+      "step": 4520
+    },
+    {
+      "epoch": 0.905,
+      "grad_norm": 0.4061259627342224,
+      "learning_rate": 7.528568277488539e-06,
+      "loss": 2.1533,
+      "step": 4525
+    },
+    {
+      "epoch": 0.906,
+      "grad_norm": 0.4078681766986847,
+      "learning_rate": 7.372537577525395e-06,
+      "loss": 2.2207,
+      "step": 4530
+    },
+    {
+      "epoch": 0.907,
+      "grad_norm": 0.4450269341468811,
+      "learning_rate": 7.218099991324827e-06,
+      "loss": 2.1339,
+      "step": 4535
+    },
+    {
+      "epoch": 0.908,
+      "grad_norm": 0.42582967877388,
+      "learning_rate": 7.0652572439166726e-06,
+      "loss": 2.191,
+      "step": 4540
+    },
+    {
+      "epoch": 0.909,
+      "grad_norm": 0.4486072063446045,
+      "learning_rate": 6.914011042516965e-06,
+      "loss": 2.2439,
+      "step": 4545
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.43058037757873535,
+      "learning_rate": 6.7643630765086885e-06,
+      "loss": 2.1713,
+      "step": 4550
+    },
+    {
+      "epoch": 0.911,
+      "grad_norm": 0.4128859341144562,
+      "learning_rate": 6.616315017422891e-06,
+      "loss": 2.1686,
+      "step": 4555
+    },
+    {
+      "epoch": 0.912,
+      "grad_norm": 0.4264604151248932,
+      "learning_rate": 6.4698685189200625e-06,
+      "loss": 2.1281,
+      "step": 4560
+    },
+    {
+      "epoch": 0.913,
+      "grad_norm": 0.3913510739803314,
+      "learning_rate": 6.325025216771657e-06,
+      "loss": 2.1368,
+      "step": 4565
+    },
+    {
+      "epoch": 0.914,
+      "grad_norm": 0.40388935804367065,
+      "learning_rate": 6.181786728841847e-06,
+      "loss": 2.2191,
+      "step": 4570
+    },
+    {
+      "epoch": 0.915,
+      "grad_norm": 0.43146443367004395,
+      "learning_rate": 6.040154655069401e-06,
+      "loss": 2.131,
+      "step": 4575
+    },
+    {
+      "epoch": 0.916,
+      "grad_norm": 0.42876219749450684,
+      "learning_rate": 5.900130577449785e-06,
+      "loss": 2.1655,
+      "step": 4580
+    },
+    {
+      "epoch": 0.917,
+      "grad_norm": 0.4041782021522522,
+      "learning_rate": 5.761716060017652e-06,
+      "loss": 2.1558,
+      "step": 4585
+    },
+    {
+      "epoch": 0.918,
+      "grad_norm": 0.46385347843170166,
+      "learning_rate": 5.652144368961192e-06,
+      "loss": 2.2228,
+      "step": 4590
+    },
+    {
+      "epoch": 0.919,
+      "grad_norm": 0.4040263295173645,
+      "learning_rate": 5.516630943835609e-06,
+      "loss": 2.0635,
+      "step": 4595
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.4247514009475708,
+      "learning_rate": 5.382731362494269e-06,
+      "loss": 2.1747,
+      "step": 4600
+    },
+    {
+      "epoch": 0.921,
+      "grad_norm": 0.43409496545791626,
+      "learning_rate": 5.250447120562695e-06,
+      "loss": 2.136,
+      "step": 4605
+    },
+    {
+      "epoch": 0.922,
+      "grad_norm": 0.42186084389686584,
+      "learning_rate": 5.1197796956234106e-06,
+      "loss": 2.1409,
+      "step": 4610
+    },
+    {
+      "epoch": 0.923,
+      "grad_norm": 0.4255131483078003,
+      "learning_rate": 4.990730547199467e-06,
+      "loss": 2.1583,
+      "step": 4615
+    },
+    {
+      "epoch": 0.924,
+      "grad_norm": 0.41459619998931885,
+      "learning_rate": 4.863301116738211e-06,
+      "loss": 2.1897,
+      "step": 4620
+    },
+    {
+      "epoch": 0.925,
+      "grad_norm": 0.4171781539916992,
+      "learning_rate": 4.737492827595107e-06,
+      "loss": 2.161,
+      "step": 4625
+    },
+    {
+      "epoch": 0.926,
+      "grad_norm": 0.39910873770713806,
+      "learning_rate": 4.6133070850178924e-06,
+      "loss": 2.1911,
+      "step": 4630
+    },
+    {
+      "epoch": 0.927,
+      "grad_norm": 0.43180596828460693,
+      "learning_rate": 4.490745276130814e-06,
+      "loss": 2.1061,
+      "step": 4635
+    },
+    {
+      "epoch": 0.928,
+      "grad_norm": 0.43445250391960144,
+      "learning_rate": 4.369808769919214e-06,
+      "loss": 2.1358,
+      "step": 4640
+    },
+    {
+      "epoch": 0.929,
+      "grad_norm": 0.3996482789516449,
+      "learning_rate": 4.2504989172142405e-06,
+      "loss": 2.1734,
+      "step": 4645
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.42857295274734497,
+      "learning_rate": 4.132817050677639e-06,
+      "loss": 2.1879,
+      "step": 4650
+    },
+    {
+      "epoch": 0.931,
+      "grad_norm": 0.43241751194000244,
+      "learning_rate": 4.016764484786972e-06,
+      "loss": 2.1718,
+      "step": 4655
+    },
+    {
+      "epoch": 0.932,
+      "grad_norm": 0.41567280888557434,
+      "learning_rate": 3.9023425158209385e-06,
+      "loss": 2.1295,
+      "step": 4660
+    },
+    {
+      "epoch": 0.933,
+      "grad_norm": 0.41955962777137756,
+      "learning_rate": 3.78955242184491e-06,
+      "loss": 2.1306,
+      "step": 4665
+    },
+    {
+      "epoch": 0.934,
+      "grad_norm": 0.4202505946159363,
+      "learning_rate": 3.6783954626964873e-06,
+      "loss": 2.1312,
+      "step": 4670
+    },
+    {
+      "epoch": 0.935,
+      "grad_norm": 0.43055739998817444,
+      "learning_rate": 3.5688728799716626e-06,
+      "loss": 2.1468,
+      "step": 4675
+    },
+    {
+      "epoch": 0.936,
+      "grad_norm": 0.41296547651290894,
+      "learning_rate": 3.460985897010832e-06,
+      "loss": 2.1501,
+      "step": 4680
+    },
+    {
+      "epoch": 0.937,
+      "grad_norm": 0.4199470281600952,
+      "learning_rate": 3.354735718885121e-06,
+      "loss": 2.1343,
+      "step": 4685
+    },
+    {
+      "epoch": 0.938,
+      "grad_norm": 0.4009940028190613,
+      "learning_rate": 3.2501235323829146e-06,
+      "loss": 2.1858,
+      "step": 4690
+    },
+    {
+      "epoch": 0.939,
+      "grad_norm": 0.4397661089897156,
+      "learning_rate": 3.147150505996715e-06,
+      "loss": 2.219,
+      "step": 4695
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.4313606321811676,
+      "learning_rate": 3.0458177899099545e-06,
+      "loss": 2.1644,
+      "step": 4700
+    },
+    {
+      "epoch": 0.941,
+      "grad_norm": 0.40876680612564087,
+      "learning_rate": 2.9461265159841873e-06,
+      "loss": 2.2214,
+      "step": 4705
+    },
+    {
+      "epoch": 0.942,
+      "grad_norm": 0.44418561458587646,
+      "learning_rate": 2.848077797746501e-06,
+      "loss": 2.1538,
+      "step": 4710
+    },
+    {
+      "epoch": 0.943,
+      "grad_norm": 0.4055839478969574,
+      "learning_rate": 2.7516727303770248e-06,
+      "loss": 2.2004,
+      "step": 4715
+    },
+    {
+      "epoch": 0.944,
+      "grad_norm": 0.4120195209980011,
+      "learning_rate": 2.656912390696708e-06,
+      "loss": 2.1759,
+      "step": 4720
+    },
+    {
+      "epoch": 0.945,
+      "grad_norm": 0.41687390208244324,
+      "learning_rate": 2.5637978371552794e-06,
+      "loss": 2.1172,
+      "step": 4725
+    },
+    {
+      "epoch": 0.946,
+      "grad_norm": 0.42830410599708557,
+      "learning_rate": 2.4723301098194704e-06,
+      "loss": 2.2135,
+      "step": 4730
+    },
+    {
+      "epoch": 0.947,
+      "grad_norm": 0.4193302392959595,
+      "learning_rate": 2.3825102303613617e-06,
+      "loss": 2.141,
+      "step": 4735
+    },
+    {
+      "epoch": 0.948,
+      "grad_norm": 0.44555971026420593,
+      "learning_rate": 2.2943392020469742e-06,
+      "loss": 2.179,
+      "step": 4740
+    },
+    {
+      "epoch": 0.949,
+      "grad_norm": 0.3976489007472992,
+      "learning_rate": 2.2078180097250266e-06,
+      "loss": 2.1691,
+      "step": 4745
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.4367688000202179,
+      "learning_rate": 2.1229476198160456e-06,
+      "loss": 2.1502,
+      "step": 4750
+    },
+    {
+      "epoch": 0.951,
+      "grad_norm": 0.40627768635749817,
+      "learning_rate": 2.0397289803014584e-06,
+      "loss": 2.1399,
+      "step": 4755
+    },
+    {
+      "epoch": 0.952,
+      "grad_norm": 0.3884209096431732,
+      "learning_rate": 1.9581630207130326e-06,
+      "loss": 1.9814,
+      "step": 4760
+    },
+    {
+      "epoch": 0.953,
+      "grad_norm": 0.4176620841026306,
+      "learning_rate": 1.878250652122537e-06,
+      "loss": 2.1467,
+      "step": 4765
+    },
+    {
+      "epoch": 0.954,
+      "grad_norm": 0.41188085079193115,
+      "learning_rate": 1.7999927671315307e-06,
+      "loss": 2.1645,
+      "step": 4770
+    },
+    {
+      "epoch": 0.955,
+      "grad_norm": 0.41481921076774597,
+      "learning_rate": 1.7233902398613719e-06,
+      "loss": 2.1747,
+      "step": 4775
+    },
+    {
+      "epoch": 0.956,
+      "grad_norm": 0.40912356972694397,
+      "learning_rate": 1.6484439259435267e-06,
+      "loss": 2.1477,
+      "step": 4780
+    },
+    {
+      "epoch": 0.957,
+      "grad_norm": 0.44400301575660706,
+      "learning_rate": 1.575154662509892e-06,
+      "loss": 2.1439,
+      "step": 4785
+    },
+    {
+      "epoch": 0.958,
+      "grad_norm": 0.391618013381958,
+      "learning_rate": 1.5035232681835874e-06,
+      "loss": 2.0929,
+      "step": 4790
+    },
+    {
+      "epoch": 0.959,
+      "grad_norm": 0.4395817220211029,
+      "learning_rate": 1.4335505430696947e-06,
+      "loss": 2.2465,
+      "step": 4795
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.4161112606525421,
+      "learning_rate": 1.3652372687464164e-06,
+      "loss": 2.1842,
+      "step": 4800
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 5000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.93792398032896e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}