diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4087 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 100,
+  "global_step": 2730,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.001098901098901099,
+      "grad_norm": 2.5938866235339426,
+      "learning_rate": 7.326007326007327e-08,
+      "loss": 1.4331,
+      "step": 1
+    },
+    {
+      "epoch": 0.005494505494505495,
+      "grad_norm": 2.3305040971210316,
+      "learning_rate": 3.6630036630036635e-07,
+      "loss": 1.3861,
+      "step": 5
+    },
+    {
+      "epoch": 0.01098901098901099,
+      "grad_norm": 2.3032315871842037,
+      "learning_rate": 7.326007326007327e-07,
+      "loss": 1.3892,
+      "step": 10
+    },
+    {
+      "epoch": 0.016483516483516484,
+      "grad_norm": 2.0293111765459386,
+      "learning_rate": 1.098901098901099e-06,
+      "loss": 1.3496,
+      "step": 15
+    },
+    {
+      "epoch": 0.02197802197802198,
+      "grad_norm": 1.4027172304084468,
+      "learning_rate": 1.4652014652014654e-06,
+      "loss": 1.2629,
+      "step": 20
+    },
+    {
+      "epoch": 0.027472527472527472,
+      "grad_norm": 1.4093182507371345,
+      "learning_rate": 1.8315018315018316e-06,
+      "loss": 1.1094,
+      "step": 25
+    },
+    {
+      "epoch": 0.03296703296703297,
+      "grad_norm": 0.6639612973057083,
+      "learning_rate": 2.197802197802198e-06,
+      "loss": 1.0554,
+      "step": 30
+    },
+    {
+      "epoch": 0.038461538461538464,
+      "grad_norm": 0.6143428370522973,
+      "learning_rate": 2.564102564102564e-06,
+      "loss": 0.9925,
+      "step": 35
+    },
+    {
+      "epoch": 0.04395604395604396,
+      "grad_norm": 0.49724360049126803,
+      "learning_rate": 2.930402930402931e-06,
+      "loss": 0.964,
+      "step": 40
+    },
+    {
+      "epoch": 0.04945054945054945,
+      "grad_norm": 0.4775849339023127,
+      "learning_rate": 3.2967032967032968e-06,
+      "loss": 0.9479,
+      "step": 45
+    },
+    {
+      "epoch": 0.054945054945054944,
+      "grad_norm": 0.33493876785224186,
+      "learning_rate": 3.663003663003663e-06,
+      "loss": 0.9174,
+      "step": 50
+    },
+    {
+      "epoch": 0.06043956043956044,
+      "grad_norm": 0.32489963376779446,
+      "learning_rate": 4.0293040293040296e-06,
+      "loss": 0.8859,
+      "step": 55
+    },
+    {
+      "epoch": 0.06593406593406594,
+      "grad_norm": 0.492730880028892,
+      "learning_rate": 4.395604395604396e-06,
+      "loss": 0.9012,
+      "step": 60
+    },
+    {
+      "epoch": 0.07142857142857142,
+      "grad_norm": 0.3109753841977473,
+      "learning_rate": 4.761904761904762e-06,
+      "loss": 0.8766,
+      "step": 65
+    },
+    {
+      "epoch": 0.07692307692307693,
+      "grad_norm": 0.2867877510996371,
+      "learning_rate": 5.128205128205128e-06,
+      "loss": 0.8571,
+      "step": 70
+    },
+    {
+      "epoch": 0.08241758241758242,
+      "grad_norm": 0.27490143167772046,
+      "learning_rate": 5.494505494505495e-06,
+      "loss": 0.8589,
+      "step": 75
+    },
+    {
+      "epoch": 0.08791208791208792,
+      "grad_norm": 0.28545809911171605,
+      "learning_rate": 5.860805860805862e-06,
+      "loss": 0.868,
+      "step": 80
+    },
+    {
+      "epoch": 0.09340659340659341,
+      "grad_norm": 0.2757573337248957,
+      "learning_rate": 6.227106227106228e-06,
+      "loss": 0.8488,
+      "step": 85
+    },
+    {
+      "epoch": 0.0989010989010989,
+      "grad_norm": 0.28168284468857463,
+      "learning_rate": 6.5934065934065935e-06,
+      "loss": 0.8415,
+      "step": 90
+    },
+    {
+      "epoch": 0.1043956043956044,
+      "grad_norm": 0.2789025930294136,
+      "learning_rate": 6.95970695970696e-06,
+      "loss": 0.8478,
+      "step": 95
+    },
+    {
+      "epoch": 0.10989010989010989,
+      "grad_norm": 0.2793158898424537,
+      "learning_rate": 7.326007326007326e-06,
+      "loss": 0.8402,
+      "step": 100
+    },
+    {
+      "epoch": 0.10989010989010989,
+      "eval_loss": 0.8307385444641113,
+      "eval_runtime": 58.2285,
+      "eval_samples_per_second": 23.665,
+      "eval_steps_per_second": 0.103,
+      "step": 100
+    },
+    {
+      "epoch": 0.11538461538461539,
+      "grad_norm": 0.28714132101165046,
+      "learning_rate": 7.692307692307694e-06,
+      "loss": 0.8132,
+      "step": 105
+    },
+    {
+      "epoch": 0.12087912087912088,
+      "grad_norm": 0.25219489491882496,
+      "learning_rate": 8.058608058608059e-06,
+      "loss": 0.8212,
+      "step": 110
+    },
+    {
+      "epoch": 0.12637362637362637,
+      "grad_norm": 0.2621955241800005,
+      "learning_rate": 8.424908424908426e-06,
+      "loss": 0.8152,
+      "step": 115
+    },
+    {
+      "epoch": 0.13186813186813187,
+      "grad_norm": 0.2699161655817323,
+      "learning_rate": 8.791208791208792e-06,
+      "loss": 0.8227,
+      "step": 120
+    },
+    {
+      "epoch": 0.13736263736263737,
+      "grad_norm": 0.283439689013222,
+      "learning_rate": 9.157509157509158e-06,
+      "loss": 0.7977,
+      "step": 125
+    },
+    {
+      "epoch": 0.14285714285714285,
+      "grad_norm": 0.2513858523994278,
+      "learning_rate": 9.523809523809525e-06,
+      "loss": 0.7834,
+      "step": 130
+    },
+    {
+      "epoch": 0.14835164835164835,
+      "grad_norm": 0.2563724649349777,
+      "learning_rate": 9.890109890109892e-06,
+      "loss": 0.8214,
+      "step": 135
+    },
+    {
+      "epoch": 0.15384615384615385,
+      "grad_norm": 0.26746604704320437,
+      "learning_rate": 1.0256410256410256e-05,
+      "loss": 0.7852,
+      "step": 140
+    },
+    {
+      "epoch": 0.15934065934065933,
+      "grad_norm": 0.280443649102156,
+      "learning_rate": 1.0622710622710623e-05,
+      "loss": 0.7943,
+      "step": 145
+    },
+    {
+      "epoch": 0.16483516483516483,
+      "grad_norm": 0.2661703492350103,
+      "learning_rate": 1.098901098901099e-05,
+      "loss": 0.7846,
+      "step": 150
+    },
+    {
+      "epoch": 0.17032967032967034,
+      "grad_norm": 0.25295008227570925,
+      "learning_rate": 1.1355311355311356e-05,
+      "loss": 0.79,
+      "step": 155
+    },
+    {
+      "epoch": 0.17582417582417584,
+      "grad_norm": 0.2572179650619883,
+      "learning_rate": 1.1721611721611723e-05,
+      "loss": 0.7945,
+      "step": 160
+    },
+    {
+      "epoch": 0.1813186813186813,
+      "grad_norm": 0.28163509900788125,
+      "learning_rate": 1.2087912087912089e-05,
+      "loss": 0.7699,
+      "step": 165
+    },
+    {
+      "epoch": 0.18681318681318682,
+      "grad_norm": 0.2998101858443158,
+      "learning_rate": 1.2454212454212456e-05,
+      "loss": 0.789,
+      "step": 170
+    },
+    {
+      "epoch": 0.19230769230769232,
+      "grad_norm": 0.259731613700112,
+      "learning_rate": 1.2820512820512823e-05,
+      "loss": 0.7518,
+      "step": 175
+    },
+    {
+      "epoch": 0.1978021978021978,
+      "grad_norm": 0.26085689374421417,
+      "learning_rate": 1.3186813186813187e-05,
+      "loss": 0.7778,
+      "step": 180
+    },
+    {
+      "epoch": 0.2032967032967033,
+      "grad_norm": 0.25933950442397596,
+      "learning_rate": 1.3553113553113554e-05,
+      "loss": 0.7752,
+      "step": 185
+    },
+    {
+      "epoch": 0.2087912087912088,
+      "grad_norm": 0.2655972864098073,
+      "learning_rate": 1.391941391941392e-05,
+      "loss": 0.7754,
+      "step": 190
+    },
+    {
+      "epoch": 0.21428571428571427,
+      "grad_norm": 0.3027958618309835,
+      "learning_rate": 1.4285714285714287e-05,
+      "loss": 0.7738,
+      "step": 195
+    },
+    {
+      "epoch": 0.21978021978021978,
+      "grad_norm": 0.2636680908875259,
+      "learning_rate": 1.4652014652014653e-05,
+      "loss": 0.7611,
+      "step": 200
+    },
+    {
+      "epoch": 0.21978021978021978,
+      "eval_loss": 0.7792791724205017,
+      "eval_runtime": 60.1437,
+      "eval_samples_per_second": 22.912,
+      "eval_steps_per_second": 0.1,
+      "step": 200
+    },
+    {
+      "epoch": 0.22527472527472528,
+      "grad_norm": 0.28418095535412446,
+      "learning_rate": 1.501831501831502e-05,
+      "loss": 0.7773,
+      "step": 205
+    },
+    {
+      "epoch": 0.23076923076923078,
+      "grad_norm": 0.253579945341443,
+      "learning_rate": 1.5384615384615387e-05,
+      "loss": 0.7466,
+      "step": 210
+    },
+    {
+      "epoch": 0.23626373626373626,
+      "grad_norm": 0.2709322711520115,
+      "learning_rate": 1.575091575091575e-05,
+      "loss": 0.7486,
+      "step": 215
+    },
+    {
+      "epoch": 0.24175824175824176,
+      "grad_norm": 0.25784522395547,
+      "learning_rate": 1.6117216117216118e-05,
+      "loss": 0.7602,
+      "step": 220
+    },
+    {
+      "epoch": 0.24725274725274726,
+      "grad_norm": 0.25800289350422184,
+      "learning_rate": 1.6483516483516486e-05,
+      "loss": 0.7593,
+      "step": 225
+    },
+    {
+      "epoch": 0.25274725274725274,
+      "grad_norm": 0.4335652422653119,
+      "learning_rate": 1.6849816849816853e-05,
+      "loss": 0.7357,
+      "step": 230
+    },
+    {
+      "epoch": 0.25824175824175827,
+      "grad_norm": 0.3289531244189634,
+      "learning_rate": 1.721611721611722e-05,
+      "loss": 0.7456,
+      "step": 235
+    },
+    {
+      "epoch": 0.26373626373626374,
+      "grad_norm": 1.2095875802009959,
+      "learning_rate": 1.7582417582417584e-05,
+      "loss": 0.7226,
+      "step": 240
+    },
+    {
+      "epoch": 0.2692307692307692,
+      "grad_norm": 0.2550501846115449,
+      "learning_rate": 1.794871794871795e-05,
+      "loss": 0.7304,
+      "step": 245
+    },
+    {
+      "epoch": 0.27472527472527475,
+      "grad_norm": 0.32241036727466216,
+      "learning_rate": 1.8315018315018315e-05,
+      "loss": 0.7322,
+      "step": 250
+    },
+    {
+      "epoch": 0.2802197802197802,
+      "grad_norm": 0.23942812421802775,
+      "learning_rate": 1.8681318681318682e-05,
+      "loss": 0.7261,
+      "step": 255
+    },
+    {
+      "epoch": 0.2857142857142857,
+      "grad_norm": 0.2532890631657975,
+      "learning_rate": 1.904761904761905e-05,
+      "loss": 0.752,
+      "step": 260
+    },
+    {
+      "epoch": 0.29120879120879123,
+      "grad_norm": 0.24364925383721642,
+      "learning_rate": 1.9413919413919417e-05,
+      "loss": 0.755,
+      "step": 265
+    },
+    {
+      "epoch": 0.2967032967032967,
+      "grad_norm": 0.24181710554401747,
+      "learning_rate": 1.9780219780219784e-05,
+      "loss": 0.7554,
+      "step": 270
+    },
+    {
+      "epoch": 0.3021978021978022,
+      "grad_norm": 0.23828479627116422,
+      "learning_rate": 1.9999967302150437e-05,
+      "loss": 0.7246,
+      "step": 275
+    },
+    {
+      "epoch": 0.3076923076923077,
+      "grad_norm": 0.26360754943139847,
+      "learning_rate": 1.9999599453798523e-05,
+      "loss": 0.7293,
+      "step": 280
+    },
+    {
+      "epoch": 0.3131868131868132,
+      "grad_norm": 0.2624620835053123,
+      "learning_rate": 1.9998822899867633e-05,
+      "loss": 0.7282,
+      "step": 285
+    },
+    {
+      "epoch": 0.31868131868131866,
+      "grad_norm": 0.24888154268515242,
+      "learning_rate": 1.9997637672097222e-05,
+      "loss": 0.7343,
+      "step": 290
+    },
+    {
+      "epoch": 0.3241758241758242,
+      "grad_norm": 0.23324534028548077,
+      "learning_rate": 1.9996043818930153e-05,
+      "loss": 0.7158,
+      "step": 295
+    },
+    {
+      "epoch": 0.32967032967032966,
+      "grad_norm": 0.23821533213282017,
+      "learning_rate": 1.9994041405510705e-05,
+      "loss": 0.7361,
+      "step": 300
+    },
+    {
+      "epoch": 0.32967032967032966,
+      "eval_loss": 0.7524750828742981,
+      "eval_runtime": 58.5251,
+      "eval_samples_per_second": 23.545,
+      "eval_steps_per_second": 0.103,
+      "step": 300
+    },
+    {
+      "epoch": 0.33516483516483514,
+      "grad_norm": 0.2419083485626964,
+      "learning_rate": 1.999163051368191e-05,
+      "loss": 0.7392,
+      "step": 305
+    },
+    {
+      "epoch": 0.34065934065934067,
+      "grad_norm": 0.23969743620339276,
+      "learning_rate": 1.9988811241982206e-05,
+      "loss": 0.722,
+      "step": 310
+    },
+    {
+      "epoch": 0.34615384615384615,
+      "grad_norm": 0.24732719030591468,
+      "learning_rate": 1.9985583705641418e-05,
+      "loss": 0.7375,
+      "step": 315
+    },
+    {
+      "epoch": 0.3516483516483517,
+      "grad_norm": 0.2524078352421696,
+      "learning_rate": 1.9981948036576045e-05,
+      "loss": 0.7174,
+      "step": 320
+    },
+    {
+      "epoch": 0.35714285714285715,
+      "grad_norm": 0.2450128000419284,
+      "learning_rate": 1.997790438338385e-05,
+      "loss": 0.7193,
+      "step": 325
+    },
+    {
+      "epoch": 0.3626373626373626,
+      "grad_norm": 0.22547733974392728,
+      "learning_rate": 1.997345291133783e-05,
+      "loss": 0.7121,
+      "step": 330
+    },
+    {
+      "epoch": 0.36813186813186816,
+      "grad_norm": 0.23499045457962084,
+      "learning_rate": 1.9968593802379405e-05,
+      "loss": 0.7067,
+      "step": 335
+    },
+    {
+      "epoch": 0.37362637362637363,
+      "grad_norm": 0.24009998902460783,
+      "learning_rate": 1.9963327255111033e-05,
+      "loss": 0.724,
+      "step": 340
+    },
+    {
+      "epoch": 0.3791208791208791,
+      "grad_norm": 0.27381888592320386,
+      "learning_rate": 1.9957653484788054e-05,
+      "loss": 0.7088,
+      "step": 345
+    },
+    {
+      "epoch": 0.38461538461538464,
+      "grad_norm": 0.23470929489474682,
+      "learning_rate": 1.9951572723309918e-05,
+      "loss": 0.7123,
+      "step": 350
+    },
+    {
+      "epoch": 0.3901098901098901,
+      "grad_norm": 0.234461070541495,
+      "learning_rate": 1.99450852192107e-05,
+      "loss": 0.7132,
+      "step": 355
+    },
+    {
+      "epoch": 0.3956043956043956,
+      "grad_norm": 0.25459804884418324,
+      "learning_rate": 1.9938191237648924e-05,
+      "loss": 0.7076,
+      "step": 360
+    },
+    {
+      "epoch": 0.4010989010989011,
+      "grad_norm": 0.245152397924957,
+      "learning_rate": 1.9930891060396757e-05,
+      "loss": 0.6918,
+      "step": 365
+    },
+    {
+      "epoch": 0.4065934065934066,
+      "grad_norm": 0.23860847532916626,
+      "learning_rate": 1.992318498582846e-05,
+      "loss": 0.7304,
+      "step": 370
+    },
+    {
+      "epoch": 0.41208791208791207,
+      "grad_norm": 0.2519260957646762,
+      "learning_rate": 1.9915073328908217e-05,
+      "loss": 0.709,
+      "step": 375
+    },
+    {
+      "epoch": 0.4175824175824176,
+      "grad_norm": 0.23360487704945615,
+      "learning_rate": 1.9906556421177256e-05,
+      "loss": 0.7258,
+      "step": 380
+    },
+    {
+      "epoch": 0.4230769230769231,
+      "grad_norm": 0.23254394099980408,
+      "learning_rate": 1.989763461074029e-05,
+      "loss": 0.707,
+      "step": 385
+    },
+    {
+      "epoch": 0.42857142857142855,
+      "grad_norm": 0.22834651014549015,
+      "learning_rate": 1.9888308262251286e-05,
+      "loss": 0.7085,
+      "step": 390
+    },
+    {
+      "epoch": 0.4340659340659341,
+      "grad_norm": 0.22350055020087028,
+      "learning_rate": 1.987857775689859e-05,
+      "loss": 0.7153,
+      "step": 395
+    },
+    {
+      "epoch": 0.43956043956043955,
+      "grad_norm": 0.2559347137216262,
+      "learning_rate": 1.9868443492389307e-05,
+      "loss": 0.6854,
+      "step": 400
+    },
+    {
+      "epoch": 0.43956043956043955,
+      "eval_loss": 0.7337242960929871,
+      "eval_runtime": 57.135,
+      "eval_samples_per_second": 24.118,
+      "eval_steps_per_second": 0.105,
+      "step": 400
+    },
+    {
+      "epoch": 0.44505494505494503,
+      "grad_norm": 0.27398370193799876,
+      "learning_rate": 1.985790588293308e-05,
+      "loss": 0.716,
+      "step": 405
+    },
+    {
+      "epoch": 0.45054945054945056,
+      "grad_norm": 0.49293536001562244,
+      "learning_rate": 1.9846965359225127e-05,
+      "loss": 0.7187,
+      "step": 410
+    },
+    {
+      "epoch": 0.45604395604395603,
+      "grad_norm": 0.23991864498334772,
+      "learning_rate": 1.9835622368428673e-05,
+      "loss": 0.6915,
+      "step": 415
+    },
+    {
+      "epoch": 0.46153846153846156,
+      "grad_norm": 0.2223500820145743,
+      "learning_rate": 1.9823877374156647e-05,
+      "loss": 0.7076,
+      "step": 420
+    },
+    {
+      "epoch": 0.46703296703296704,
+      "grad_norm": 0.2327646676318351,
+      "learning_rate": 1.9811730856452754e-05,
+      "loss": 0.6865,
+      "step": 425
+    },
+    {
+      "epoch": 0.4725274725274725,
+      "grad_norm": 0.23366623666692093,
+      "learning_rate": 1.9799183311771823e-05,
+      "loss": 0.697,
+      "step": 430
+    },
+    {
+      "epoch": 0.47802197802197804,
+      "grad_norm": 0.23211216480431496,
+      "learning_rate": 1.9786235252959555e-05,
+      "loss": 0.6949,
+      "step": 435
+    },
+    {
+      "epoch": 0.4835164835164835,
+      "grad_norm": 0.2335049326202,
+      "learning_rate": 1.977288720923153e-05,
+      "loss": 0.688,
+      "step": 440
+    },
+    {
+      "epoch": 0.489010989010989,
+      "grad_norm": 0.23612214303202572,
+      "learning_rate": 1.9759139726151597e-05,
+      "loss": 0.7006,
+      "step": 445
+    },
+    {
+      "epoch": 0.4945054945054945,
+      "grad_norm": 0.2296371958742712,
+      "learning_rate": 1.9744993365609563e-05,
+      "loss": 0.7096,
+      "step": 450
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.23792816575947545,
+      "learning_rate": 1.973044870579824e-05,
+      "loss": 0.6799,
+      "step": 455
+    },
+    {
+      "epoch": 0.5054945054945055,
+      "grad_norm": 0.23888207855755858,
+      "learning_rate": 1.9715506341189795e-05,
+      "loss": 0.7189,
+      "step": 460
+    },
+    {
+      "epoch": 0.510989010989011,
+      "grad_norm": 0.22873637589635504,
+      "learning_rate": 1.970016688251147e-05,
+      "loss": 0.7041,
+      "step": 465
+    },
+    {
+      "epoch": 0.5164835164835165,
+      "grad_norm": 0.22109885665030304,
+      "learning_rate": 1.9684430956720613e-05,
+      "loss": 0.6889,
+      "step": 470
+    },
+    {
+      "epoch": 0.521978021978022,
+      "grad_norm": 0.22943401975482253,
+      "learning_rate": 1.966829920697905e-05,
+      "loss": 0.701,
+      "step": 475
+    },
+    {
+      "epoch": 0.5274725274725275,
+      "grad_norm": 0.24368759924523126,
+      "learning_rate": 1.9651772292626804e-05,
+      "loss": 0.6798,
+      "step": 480
+    },
+    {
+      "epoch": 0.532967032967033,
+      "grad_norm": 0.22502323506511537,
+      "learning_rate": 1.963485088915514e-05,
+      "loss": 0.6732,
+      "step": 485
+    },
+    {
+      "epoch": 0.5384615384615384,
+      "grad_norm": 0.23406619299034903,
+      "learning_rate": 1.961753568817896e-05,
+      "loss": 0.6808,
+      "step": 490
+    },
+    {
+      "epoch": 0.5439560439560439,
+      "grad_norm": 0.24379416412543173,
+      "learning_rate": 1.959982739740854e-05,
+      "loss": 0.6732,
+      "step": 495
+    },
+    {
+      "epoch": 0.5494505494505495,
+      "grad_norm": 0.22395929004707257,
+      "learning_rate": 1.9581726740620585e-05,
+      "loss": 0.6926,
+      "step": 500
+    },
+    {
+      "epoch": 0.5494505494505495,
+      "eval_loss": 0.7197047472000122,
+      "eval_runtime": 57.3825,
+      "eval_samples_per_second": 24.014,
+      "eval_steps_per_second": 0.105,
+      "step": 500
+    },
+    {
+      "epoch": 0.554945054945055,
+      "grad_norm": 0.23635721578380114,
+      "learning_rate": 1.9563234457628678e-05,
+      "loss": 0.7019,
+      "step": 505
+    },
+    {
+      "epoch": 0.5604395604395604,
+      "grad_norm": 0.22361881412683712,
+      "learning_rate": 1.954435130425301e-05,
+      "loss": 0.6725,
+      "step": 510
+    },
+    {
+      "epoch": 0.5659340659340659,
+      "grad_norm": 0.27988489835986297,
+      "learning_rate": 1.952507805228951e-05,
+      "loss": 0.6806,
+      "step": 515
+    },
+    {
+      "epoch": 0.5714285714285714,
+      "grad_norm": 0.22714617789786667,
+      "learning_rate": 1.9505415489478293e-05,
+      "loss": 0.6642,
+      "step": 520
+    },
+    {
+      "epoch": 0.5769230769230769,
+      "grad_norm": 0.234738186404094,
+      "learning_rate": 1.9485364419471454e-05,
+      "loss": 0.6788,
+      "step": 525
+    },
+    {
+      "epoch": 0.5824175824175825,
+      "grad_norm": 0.22212354431865575,
+      "learning_rate": 1.9464925661800247e-05,
+      "loss": 0.6714,
+      "step": 530
+    },
+    {
+      "epoch": 0.5879120879120879,
+      "grad_norm": 0.2260832469755207,
+      "learning_rate": 1.9444100051841556e-05,
+      "loss": 0.6742,
+      "step": 535
+    },
+    {
+      "epoch": 0.5934065934065934,
+      "grad_norm": 0.22847007927311688,
+      "learning_rate": 1.9422888440783773e-05,
+      "loss": 0.6825,
+      "step": 540
+    },
+    {
+      "epoch": 0.5989010989010989,
+      "grad_norm": 0.24175296826977385,
+      "learning_rate": 1.9401291695592e-05,
+      "loss": 0.6748,
+      "step": 545
+    },
+    {
+      "epoch": 0.6043956043956044,
+      "grad_norm": 0.22595935576781812,
+      "learning_rate": 1.9379310698972618e-05,
+      "loss": 0.6922,
+      "step": 550
+    },
+    {
+      "epoch": 0.6098901098901099,
+      "grad_norm": 0.2357931340905214,
+      "learning_rate": 1.935694634933721e-05,
+      "loss": 0.6961,
+      "step": 555
+    },
+    {
+      "epoch": 0.6153846153846154,
+      "grad_norm": 0.23436108471733894,
+      "learning_rate": 1.933419956076584e-05,
+      "loss": 0.6539,
+      "step": 560
+    },
+    {
+      "epoch": 0.6208791208791209,
+      "grad_norm": 0.21960475435288737,
+      "learning_rate": 1.9311071262969675e-05,
+      "loss": 0.6815,
+      "step": 565
+    },
+    {
+      "epoch": 0.6263736263736264,
+      "grad_norm": 0.22404812837767762,
+      "learning_rate": 1.9287562401253023e-05,
+      "loss": 0.6654,
+      "step": 570
+    },
+    {
+      "epoch": 0.6318681318681318,
+      "grad_norm": 0.2283357687835303,
+      "learning_rate": 1.9263673936474662e-05,
+      "loss": 0.6791,
+      "step": 575
+    },
+    {
+      "epoch": 0.6373626373626373,
+      "grad_norm": 0.22007214714544407,
+      "learning_rate": 1.9239406845008583e-05,
+      "loss": 0.6707,
+      "step": 580
+    },
+    {
+      "epoch": 0.6428571428571429,
+      "grad_norm": 0.21680436366260772,
+      "learning_rate": 1.921476211870408e-05,
+      "loss": 0.6875,
+      "step": 585
+    },
+    {
+      "epoch": 0.6483516483516484,
+      "grad_norm": 0.2206990118778276,
+      "learning_rate": 1.918974076484521e-05,
+      "loss": 0.689,
+      "step": 590
+    },
+    {
+      "epoch": 0.6538461538461539,
+      "grad_norm": 0.23773830363891224,
+      "learning_rate": 1.916434380610963e-05,
+      "loss": 0.6904,
+      "step": 595
+    },
+    {
+      "epoch": 0.6593406593406593,
+      "grad_norm": 0.22280953580435744,
+      "learning_rate": 1.9138572280526795e-05,
+      "loss": 0.7125,
+      "step": 600
+    },
+    {
+      "epoch": 0.6593406593406593,
+      "eval_loss": 0.709740936756134,
+      "eval_runtime": 56.8842,
+      "eval_samples_per_second": 24.225,
+      "eval_steps_per_second": 0.105,
+      "step": 600
+    },
+    {
+      "epoch": 0.6648351648351648,
+      "grad_norm": 0.21474109145946474,
+      "learning_rate": 1.911242724143552e-05,
+      "loss": 0.6883,
+      "step": 605
+    },
+    {
+      "epoch": 0.6703296703296703,
+      "grad_norm": 0.22132757673296177,
+      "learning_rate": 1.908590975744094e-05,
+      "loss": 0.6664,
+      "step": 610
+    },
+    {
+      "epoch": 0.6758241758241759,
+      "grad_norm": 0.22862364892697207,
+      "learning_rate": 1.9059020912370836e-05,
+      "loss": 0.6628,
+      "step": 615
+    },
+    {
+      "epoch": 0.6813186813186813,
+      "grad_norm": 0.22871592729546456,
+      "learning_rate": 1.9031761805231322e-05,
+      "loss": 0.6672,
+      "step": 620
+    },
+    {
+      "epoch": 0.6868131868131868,
+      "grad_norm": 0.22248256090434504,
+      "learning_rate": 1.9004133550161953e-05,
+      "loss": 0.6601,
+      "step": 625
+    },
+    {
+      "epoch": 0.6923076923076923,
+      "grad_norm": 0.22365938486859313,
+      "learning_rate": 1.8976137276390145e-05,
+      "loss": 0.6826,
+      "step": 630
+    },
+    {
+      "epoch": 0.6978021978021978,
+      "grad_norm": 0.2199011359745548,
+      "learning_rate": 1.894777412818506e-05,
+      "loss": 0.6641,
+      "step": 635
+    },
+    {
+      "epoch": 0.7032967032967034,
+      "grad_norm": 0.22510686451837608,
+      "learning_rate": 1.891904526481083e-05,
+      "loss": 0.6859,
+      "step": 640
+    },
+    {
+      "epoch": 0.7087912087912088,
+      "grad_norm": 0.21671102889261787,
+      "learning_rate": 1.8889951860479165e-05,
+      "loss": 0.6714,
+      "step": 645
+    },
+    {
+      "epoch": 0.7142857142857143,
+      "grad_norm": 0.2281707606944894,
+      "learning_rate": 1.8860495104301346e-05,
+      "loss": 0.6606,
+      "step": 650
+    },
+    {
+      "epoch": 0.7197802197802198,
+      "grad_norm": 0.2307798038244349,
+      "learning_rate": 1.8830676200239666e-05,
+      "loss": 0.6743,
+      "step": 655
+    },
+    {
+      "epoch": 0.7252747252747253,
+      "grad_norm": 0.2231346316824852,
+      "learning_rate": 1.8800496367058187e-05,
+      "loss": 0.6779,
+      "step": 660
+    },
+    {
+      "epoch": 0.7307692307692307,
+      "grad_norm": 0.2163289792210269,
+      "learning_rate": 1.8769956838272937e-05,
+      "loss": 0.6589,
+      "step": 665
+    },
+    {
+      "epoch": 0.7362637362637363,
+      "grad_norm": 0.21367697388384943,
+      "learning_rate": 1.8739058862101487e-05,
+      "loss": 0.6663,
+      "step": 670
+    },
+    {
+      "epoch": 0.7417582417582418,
+      "grad_norm": 0.2129068920296721,
+      "learning_rate": 1.8707803701411946e-05,
+      "loss": 0.6476,
+      "step": 675
+    },
+    {
+      "epoch": 0.7472527472527473,
+      "grad_norm": 0.225469317489798,
+      "learning_rate": 1.8676192633671342e-05,
+      "loss": 0.6617,
+      "step": 680
+    },
+    {
+      "epoch": 0.7527472527472527,
+      "grad_norm": 0.22021116546638458,
+      "learning_rate": 1.8644226950893394e-05,
+      "loss": 0.6709,
+      "step": 685
+    },
+    {
+      "epoch": 0.7582417582417582,
+      "grad_norm": 0.21699187730804526,
+      "learning_rate": 1.861190795958573e-05,
+      "loss": 0.6758,
+      "step": 690
+    },
+    {
+      "epoch": 0.7637362637362637,
+      "grad_norm": 0.22500716309661084,
+      "learning_rate": 1.857923698069646e-05,
+      "loss": 0.6623,
+      "step": 695
+    },
+    {
+      "epoch": 0.7692307692307693,
+      "grad_norm": 0.22095635026785526,
+      "learning_rate": 1.8546215349560204e-05,
+      "loss": 0.6662,
+      "step": 700
+    },
+    {
+      "epoch": 0.7692307692307693,
+      "eval_loss": 0.7014729976654053,
+      "eval_runtime": 58.7488,
+      "eval_samples_per_second": 23.456,
+      "eval_steps_per_second": 0.102,
+      "step": 700
+    },
+    {
+      "epoch": 0.7747252747252747,
+      "grad_norm": 0.2309523218504365,
+      "learning_rate": 1.8512844415843514e-05,
+      "loss": 0.6638,
+      "step": 705
+    },
+    {
+      "epoch": 0.7802197802197802,
+      "grad_norm": 0.2603467352522321,
+      "learning_rate": 1.8479125543489694e-05,
+      "loss": 0.6585,
+      "step": 710
+    },
+    {
+      "epoch": 0.7857142857142857,
+      "grad_norm": 0.25516737652528354,
+      "learning_rate": 1.844506011066308e-05,
+      "loss": 0.6731,
+      "step": 715
+    },
+    {
+      "epoch": 0.7912087912087912,
+      "grad_norm": 0.20905095426392303,
+      "learning_rate": 1.841064950969268e-05,
+      "loss": 0.6516,
+      "step": 720
+    },
+    {
+      "epoch": 0.7967032967032966,
+      "grad_norm": 0.21250771141180422,
+      "learning_rate": 1.8375895147015285e-05,
+      "loss": 0.6597,
+      "step": 725
+    },
+    {
+      "epoch": 0.8021978021978022,
+      "grad_norm": 0.21331846730755302,
+      "learning_rate": 1.8340798443117992e-05,
+      "loss": 0.6551,
+      "step": 730
+    },
+    {
+      "epoch": 0.8076923076923077,
+      "grad_norm": 0.21452172763665409,
+      "learning_rate": 1.8305360832480118e-05,
+      "loss": 0.6284,
+      "step": 735
+    },
+    {
+      "epoch": 0.8131868131868132,
+      "grad_norm": 0.22124518551235928,
+      "learning_rate": 1.8269583763514603e-05,
+      "loss": 0.6629,
+      "step": 740
+    },
+    {
+      "epoch": 0.8186813186813187,
+      "grad_norm": 0.20994044274257415,
+      "learning_rate": 1.8233468698508786e-05,
+      "loss": 0.6379,
+      "step": 745
+    },
+    {
+      "epoch": 0.8241758241758241,
+      "grad_norm": 0.21519633358820012,
+      "learning_rate": 1.819701711356464e-05,
+      "loss": 0.657,
+      "step": 750
+    },
+    {
+      "epoch": 0.8296703296703297,
+      "grad_norm": 0.2131452043844892,
+      "learning_rate": 1.8160230498538464e-05,
+      "loss": 0.6607,
+      "step": 755
+    },
+    {
+      "epoch": 0.8351648351648352,
+      "grad_norm": 0.21693178791987028,
+      "learning_rate": 1.8123110356979955e-05,
+      "loss": 0.6691,
+      "step": 760
+    },
+    {
+      "epoch": 0.8406593406593407,
+      "grad_norm": 0.2122186850793786,
+      "learning_rate": 1.808565820607078e-05,
+      "loss": 0.6605,
+      "step": 765
+    },
+    {
+      "epoch": 0.8461538461538461,
+      "grad_norm": 0.21668819567741526,
+      "learning_rate": 1.8047875576562556e-05,
+      "loss": 0.6518,
+      "step": 770
+    },
+    {
+      "epoch": 0.8516483516483516,
+      "grad_norm": 0.21597653641956766,
+      "learning_rate": 1.8009764012714283e-05,
+      "loss": 0.6552,
+      "step": 775
+    },
+    {
+      "epoch": 0.8571428571428571,
+      "grad_norm": 0.21403410821941848,
+      "learning_rate": 1.7971325072229227e-05,
+      "loss": 0.6524,
+      "step": 780
+    },
+    {
+      "epoch": 0.8626373626373627,
+      "grad_norm": 0.21858357009497015,
+      "learning_rate": 1.7932560326191265e-05,
+      "loss": 0.6432,
+      "step": 785
+    },
+    {
+      "epoch": 0.8681318681318682,
+      "grad_norm": 0.20835225444574582,
+      "learning_rate": 1.789347135900066e-05,
+      "loss": 0.6521,
+      "step": 790
+    },
+    {
+      "epoch": 0.8736263736263736,
+      "grad_norm": 0.2147631378246422,
+      "learning_rate": 1.7854059768309292e-05,
+      "loss": 0.6635,
+      "step": 795
+    },
+    {
+      "epoch": 0.8791208791208791,
+      "grad_norm": 0.21742833943392462,
+      "learning_rate": 1.7814327164955388e-05,
+      "loss": 0.6517,
+      "step": 800
+    },
+    {
+      "epoch": 0.8791208791208791,
+      "eval_loss": 0.6937416791915894,
+      "eval_runtime": 58.6272,
+      "eval_samples_per_second": 23.504,
+      "eval_steps_per_second": 0.102,
+      "step": 800
+    },
+    {
+      "epoch": 0.8846153846153846,
+      "grad_norm": 0.21815615754521292,
+      "learning_rate": 1.777427517289766e-05,
+      "loss": 0.6513,
+      "step": 805
+    },
+    {
+      "epoch": 0.8901098901098901,
+      "grad_norm": 0.21132299920641998,
+      "learning_rate": 1.773390542914894e-05,
+      "loss": 0.6451,
+      "step": 810
+    },
+    {
+      "epoch": 0.8956043956043956,
+      "grad_norm": 0.2153981254584012,
+      "learning_rate": 1.7693219583709266e-05,
+      "loss": 0.6472,
+      "step": 815
+    },
+    {
+      "epoch": 0.9010989010989011,
+      "grad_norm": 0.21516183847901768,
+      "learning_rate": 1.765221929949845e-05,
+      "loss": 0.6455,
+      "step": 820
+    },
+    {
+      "epoch": 0.9065934065934066,
+      "grad_norm": 0.21374150031251749,
+      "learning_rate": 1.7610906252288097e-05,
+      "loss": 0.6599,
+      "step": 825
+    },
+    {
+      "epoch": 0.9120879120879121,
+      "grad_norm": 0.22521849898029836,
+      "learning_rate": 1.7569282130633137e-05,
+      "loss": 0.6618,
+      "step": 830
+    },
+    {
+      "epoch": 0.9175824175824175,
+      "grad_norm": 0.20871378565296117,
+      "learning_rate": 1.752734863580278e-05,
+      "loss": 0.6435,
+      "step": 835
+    },
+    {
+      "epoch": 0.9230769230769231,
+      "grad_norm": 0.20579905659289663,
+      "learning_rate": 1.7485107481711014e-05,
+      "loss": 0.6384,
+      "step": 840
+    },
+    {
+      "epoch": 0.9285714285714286,
+      "grad_norm": 0.24382815508466385,
+      "learning_rate": 1.7442560394846518e-05,
+      "loss": 0.6433,
+      "step": 845
+    },
+    {
+      "epoch": 0.9340659340659341,
+      "grad_norm": 0.20272371923265584,
+      "learning_rate": 1.739970911420213e-05,
+      "loss": 0.6647,
+      "step": 850
+    },
+    {
+      "epoch": 0.9395604395604396,
+      "grad_norm": 0.21061337303398175,
+      "learning_rate": 1.7356555391203745e-05,
+      "loss": 0.6627,
+      "step": 855
+    },
+    {
+      "epoch": 0.945054945054945,
+      "grad_norm": 0.22526349877300406,
+      "learning_rate": 1.7313100989638745e-05,
+      "loss": 0.6528,
+      "step": 860
+    },
+    {
+      "epoch": 0.9505494505494505,
+      "grad_norm": 0.2405922721275717,
+      "learning_rate": 1.7269347685583913e-05,
+      "loss": 0.641,
+      "step": 865
+    },
+    {
+      "epoch": 0.9560439560439561,
+      "grad_norm": 0.22058952892730063,
+      "learning_rate": 1.7225297267332815e-05,
+      "loss": 0.6632,
+      "step": 870
+    },
+    {
+      "epoch": 0.9615384615384616,
+      "grad_norm": 0.21098073518114888,
+      "learning_rate": 1.7180951535322742e-05,
+      "loss": 0.651,
+      "step": 875
+    },
+    {
+      "epoch": 0.967032967032967,
+      "grad_norm": 0.21900984210996413,
+      "learning_rate": 1.7136312302061097e-05,
+      "loss": 0.6651,
+      "step": 880
+    },
+    {
+      "epoch": 0.9725274725274725,
+      "grad_norm": 0.20847336917155376,
+      "learning_rate": 1.7091381392051333e-05,
+      "loss": 0.6364,
+      "step": 885
+    },
+    {
+      "epoch": 0.978021978021978,
+      "grad_norm": 0.22051072508635897,
+      "learning_rate": 1.704616064171836e-05,
+      "loss": 0.6577,
+      "step": 890
+    },
+    {
+      "epoch": 0.9835164835164835,
+      "grad_norm": 0.21269562851079712,
+      "learning_rate": 1.7000651899333512e-05,
+      "loss": 0.6457,
+      "step": 895
+    },
+    {
+      "epoch": 0.989010989010989,
+      "grad_norm": 0.20529825370314203,
+      "learning_rate": 1.6954857024938976e-05,
+      "loss": 0.6234,
+      "step": 900
+    },
+    {
+      "epoch": 0.989010989010989,
+      "eval_loss": 0.6869306564331055,
+      "eval_runtime": 57.713,
+      "eval_samples_per_second": 23.877,
+      "eval_steps_per_second": 0.104,
+      "step": 900
+    },
+    {
+      "epoch": 0.9945054945054945,
+      "grad_norm": 0.2434572328244998,
+      "learning_rate": 1.6908777890271794e-05,
+      "loss": 0.6504,
+      "step": 905
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.21741173594960767,
+      "learning_rate": 1.686241637868734e-05,
+      "loss": 0.6335,
+      "step": 910
+    },
+    {
+      "epoch": 1.0054945054945055,
+      "grad_norm": 0.2130051055875561,
+      "learning_rate": 1.6815774385082355e-05,
+      "loss": 0.5875,
+      "step": 915
+    },
+    {
+      "epoch": 1.010989010989011,
+      "grad_norm": 0.21037185236973288,
+      "learning_rate": 1.6768853815817506e-05,
+      "loss": 0.5925,
+      "step": 920
+    },
+    {
+      "epoch": 1.0164835164835164,
+      "grad_norm": 0.22203340887583678,
+      "learning_rate": 1.6721656588639444e-05,
+      "loss": 0.6027,
+      "step": 925
+    },
+    {
+      "epoch": 1.021978021978022,
+      "grad_norm": 0.2835076660446732,
+      "learning_rate": 1.6674184632602447e-05,
+      "loss": 0.5949,
+      "step": 930
+    },
+    {
+      "epoch": 1.0274725274725274,
+      "grad_norm": 0.21059794691991446,
+      "learning_rate": 1.6626439887989552e-05,
+      "loss": 0.5952,
+      "step": 935
+    },
+    {
+      "epoch": 1.032967032967033,
+      "grad_norm": 0.21091351684344237,
+      "learning_rate": 1.6578424306233282e-05,
+      "loss": 0.5912,
+      "step": 940
+    },
+    {
+      "epoch": 1.0384615384615385,
+      "grad_norm": 0.21087167061209808,
+      "learning_rate": 1.653013984983585e-05,
+      "loss": 0.5862,
+      "step": 945
+    },
+    {
+      "epoch": 1.043956043956044,
+      "grad_norm": 0.21380595384587936,
+      "learning_rate": 1.6481588492288985e-05,
+      "loss": 0.599,
+      "step": 950
+    },
+    {
+      "epoch": 1.0494505494505495,
+      "grad_norm": 0.20714451035024295,
+      "learning_rate": 1.643277221799323e-05,
+      "loss": 0.5841,
+      "step": 955
+    },
+    {
+      "epoch": 1.054945054945055,
+      "grad_norm": 0.2159347638582695,
+      "learning_rate": 1.638369302217687e-05,
+      "loss": 0.5919,
+      "step": 960
+    },
+    {
+      "epoch": 1.0604395604395604,
+      "grad_norm": 0.20846702806767928,
+      "learning_rate": 1.633435291081437e-05,
+      "loss": 0.5943,
+      "step": 965
+    },
+    {
+      "epoch": 1.065934065934066,
+      "grad_norm": 0.21732557354687126,
+      "learning_rate": 1.6284753900544384e-05,
+      "loss": 0.6074,
+      "step": 970
+    },
+    {
+      "epoch": 1.0714285714285714,
+      "grad_norm": 0.21491296524750328,
+      "learning_rate": 1.6234898018587336e-05,
+      "loss": 0.6033,
+      "step": 975
+    },
+    {
+      "epoch": 1.0769230769230769,
+      "grad_norm": 0.2420317609022985,
+      "learning_rate": 1.618478730266255e-05,
+      "loss": 0.6081,
+      "step": 980
+    },
+    {
+      "epoch": 1.0824175824175823,
+      "grad_norm": 0.21022308551794014,
+      "learning_rate": 1.6134423800904985e-05,
+      "loss": 0.5755,
+      "step": 985
+    },
+    {
+      "epoch": 1.0879120879120878,
+      "grad_norm": 0.20675117493646727,
+      "learning_rate": 1.6083809571781498e-05,
+      "loss": 0.5977,
+      "step": 990
+    },
+    {
+      "epoch": 1.0934065934065935,
+      "grad_norm": 0.21348585093868772,
+      "learning_rate": 1.6032946684006745e-05,
+      "loss": 0.5859,
+      "step": 995
+    },
+    {
+      "epoch": 1.098901098901099,
+      "grad_norm": 0.2840550425577152,
+      "learning_rate": 1.598183721645858e-05,
+      "loss": 0.5925,
+      "step": 1000
+    },
+    {
+      "epoch": 1.098901098901099,
+      "eval_loss": 0.6866068243980408,
+      "eval_runtime": 56.3212,
+      "eval_samples_per_second": 24.467,
+      "eval_steps_per_second": 0.107,
+      "step": 1000
+    },
+    {
+      "epoch": 1.1043956043956045,
+      "grad_norm": 0.21980107752404765,
+      "learning_rate": 1.5930483258093144e-05,
+      "loss": 0.5916,
+      "step": 1005
+    },
+    {
+      "epoch": 1.10989010989011,
+      "grad_norm": 0.22069775219844529,
+      "learning_rate": 1.5878886907859423e-05,
+      "loss": 0.6005,
+      "step": 1010
+    },
+    {
+      "epoch": 1.1153846153846154,
+      "grad_norm": 0.22181894571988076,
+      "learning_rate": 1.5827050274613512e-05,
+      "loss": 0.5862,
+      "step": 1015
+    },
+    {
+      "epoch": 1.120879120879121,
+      "grad_norm": 0.20906581970205632,
+      "learning_rate": 1.57749754770324e-05,
+      "loss": 0.5745,
+      "step": 1020
+    },
+    {
+      "epoch": 1.1263736263736264,
+      "grad_norm": 0.21467487180581052,
+      "learning_rate": 1.5722664643527362e-05,
+      "loss": 0.5796,
+      "step": 1025
+    },
+    {
+      "epoch": 1.1318681318681318,
+      "grad_norm": 0.2126168321569666,
+      "learning_rate": 1.567011991215699e-05,
+      "loss": 0.5842,
+      "step": 1030
+    },
+    {
+      "epoch": 1.1373626373626373,
+      "grad_norm": 0.21522455237738478,
+      "learning_rate": 1.561734343053979e-05,
+      "loss": 0.5979,
+      "step": 1035
+    },
+    {
+      "epoch": 1.1428571428571428,
+      "grad_norm": 0.22633801737454678,
+      "learning_rate": 1.5564337355766412e-05,
+      "loss": 0.5759,
+      "step": 1040
+    },
+    {
+      "epoch": 1.1483516483516483,
+      "grad_norm": 0.21358958977145795,
+      "learning_rate": 1.551110385431148e-05,
+      "loss": 0.5939,
+      "step": 1045
+    },
+    {
+      "epoch": 1.1538461538461537,
+      "grad_norm": 0.23012592316508443,
+      "learning_rate": 1.5457645101945046e-05,
+      "loss": 0.5844,
+      "step": 1050
+    },
+    {
+      "epoch": 1.1593406593406592,
+      "grad_norm": 0.22381921198752516,
+      "learning_rate": 1.540396328364367e-05,
+      "loss": 0.5952,
+      "step": 1055
+    },
+    {
+      "epoch": 1.164835164835165,
+      "grad_norm": 0.21581592323525847,
+      "learning_rate": 1.5350060593501086e-05,
+      "loss": 0.5964,
+      "step": 1060
+    },
+    {
+      "epoch": 1.1703296703296704,
+      "grad_norm": 0.21213106121775074,
+      "learning_rate": 1.5295939234638566e-05,
+      "loss": 0.5948,
+      "step": 1065
+    },
+    {
+      "epoch": 1.1758241758241759,
+      "grad_norm": 0.2148049280696334,
+      "learning_rate": 1.5241601419114842e-05,
+      "loss": 0.5764,
+      "step": 1070
+    },
+    {
+      "epoch": 1.1813186813186813,
+      "grad_norm": 0.22960383958999012,
+      "learning_rate": 1.5187049367835709e-05,
+      "loss": 0.6003,
+      "step": 1075
+    },
+    {
+      "epoch": 1.1868131868131868,
+      "grad_norm": 0.21916919776904648,
+      "learning_rate": 1.5132285310463243e-05,
+      "loss": 0.5821,
+      "step": 1080
+    },
+    {
+      "epoch": 1.1923076923076923,
+      "grad_norm": 0.20455516360752496,
+      "learning_rate": 1.507731148532468e-05,
+      "loss": 0.5725,
+      "step": 1085
+    },
+    {
+      "epoch": 1.1978021978021978,
+      "grad_norm": 0.21171630637088715,
+      "learning_rate": 1.5022130139320916e-05,
+      "loss": 0.6011,
+      "step": 1090
+    },
+    {
+      "epoch": 1.2032967032967032,
+      "grad_norm": 0.21663323232960344,
+      "learning_rate": 1.4966743527834691e-05,
+      "loss": 0.5824,
+      "step": 1095
+    },
+    {
+      "epoch": 1.2087912087912087,
+      "grad_norm": 0.21254281294030117,
+      "learning_rate": 1.4911153914638388e-05,
+      "loss": 0.585,
+      "step": 1100
+    },
+    {
+      "epoch": 1.2087912087912087,
+      "eval_loss": 0.6832450032234192,
+      "eval_runtime": 57.3698,
+      "eval_samples_per_second": 24.02,
+      "eval_steps_per_second": 0.105,
+      "step": 1100
+    },
+    {
+      "epoch": 1.2142857142857142,
+      "grad_norm": 0.21064353654883214,
+      "learning_rate": 1.4855363571801523e-05,
+      "loss": 0.5825,
+      "step": 1105
+    },
+    {
+      "epoch": 1.2197802197802199,
+      "grad_norm": 0.21183460637710103,
+      "learning_rate": 1.4799374779597866e-05,
+      "loss": 0.5845,
+      "step": 1110
+    },
+    {
+      "epoch": 1.2252747252747254,
+      "grad_norm": 0.21260336129469928,
+      "learning_rate": 1.474318982641225e-05,
+      "loss": 0.5677,
+      "step": 1115
+    },
+    {
+      "epoch": 1.2307692307692308,
+      "grad_norm": 0.2126511780407026,
+      "learning_rate": 1.4686811008647037e-05,
+      "loss": 0.5804,
+      "step": 1120
+    },
+    {
+      "epoch": 1.2362637362637363,
+      "grad_norm": 0.21779018291407423,
+      "learning_rate": 1.463024063062827e-05,
+      "loss": 0.5863,
+      "step": 1125
+    },
+    {
+      "epoch": 1.2417582417582418,
+      "grad_norm": 0.20654603994279122,
+      "learning_rate": 1.457348100451146e-05,
+      "loss": 0.594,
+      "step": 1130
+    },
+    {
+      "epoch": 1.2472527472527473,
+      "grad_norm": 0.2085523679075223,
+      "learning_rate": 1.4516534450187126e-05,
+      "loss": 0.5755,
+      "step": 1135
+    },
+    {
+      "epoch": 1.2527472527472527,
+      "grad_norm": 0.20261083379143977,
+      "learning_rate": 1.4459403295185933e-05,
+      "loss": 0.5898,
+      "step": 1140
+    },
+    {
+      "epoch": 1.2582417582417582,
+      "grad_norm": 0.21173700024568526,
+      "learning_rate": 1.4402089874583594e-05,
+      "loss": 0.5757,
+      "step": 1145
+    },
+    {
+      "epoch": 1.2637362637362637,
+      "grad_norm": 0.20789003864449884,
+      "learning_rate": 1.4344596530905412e-05,
+      "loss": 0.5956,
+      "step": 1150
+    },
+    {
+      "epoch": 1.2692307692307692,
+      "grad_norm": 0.21703411260373892,
+      "learning_rate": 1.4286925614030542e-05,
+      "loss": 0.5783,
+      "step": 1155
+    },
+    {
+      "epoch": 1.2747252747252746,
+      "grad_norm": 0.20821140503325464,
+      "learning_rate": 1.4229079481095949e-05,
+      "loss": 0.5698,
+      "step": 1160
+    },
+    {
+      "epoch": 1.2802197802197801,
+      "grad_norm": 0.20754015036739895,
+      "learning_rate": 1.4171060496400055e-05,
+      "loss": 0.5683,
+      "step": 1165
+    },
+    {
+      "epoch": 1.2857142857142856,
+      "grad_norm": 0.21632631037889694,
+      "learning_rate": 1.4112871031306118e-05,
+      "loss": 0.5722,
+      "step": 1170
+    },
+    {
+      "epoch": 1.2912087912087913,
+      "grad_norm": 0.21139448126540564,
+      "learning_rate": 1.4054513464145303e-05,
+      "loss": 0.5941,
+      "step": 1175
+    },
+    {
+      "epoch": 1.2967032967032968,
+      "grad_norm": 0.2005961523532221,
+      "learning_rate": 1.3995990180119478e-05,
+      "loss": 0.5783,
+      "step": 1180
+    },
+    {
+      "epoch": 1.3021978021978022,
+      "grad_norm": 0.22506737587889664,
+      "learning_rate": 1.3937303571203718e-05,
+      "loss": 0.5815,
+      "step": 1185
+    },
+    {
+      "epoch": 1.3076923076923077,
+      "grad_norm": 0.20776135745507532,
+      "learning_rate": 1.387845603604855e-05,
+      "loss": 0.5754,
+      "step": 1190
+    },
+    {
+      "epoch": 1.3131868131868132,
+      "grad_norm": 0.20550133841197799,
+      "learning_rate": 1.3819449979881907e-05,
+      "loss": 0.5784,
+      "step": 1195
+    },
+    {
+      "epoch": 1.3186813186813187,
+      "grad_norm": 0.20805118712132745,
+      "learning_rate": 1.3760287814410822e-05,
+      "loss": 0.5857,
+      "step": 1200
+    },
+    {
+      "epoch": 1.3186813186813187,
+      "eval_loss": 0.6798372864723206,
+      "eval_runtime": 56.9599,
+      "eval_samples_per_second": 24.192,
+      "eval_steps_per_second": 0.105,
+      "step": 1200
+    },
+    {
+      "epoch": 1.3241758241758241,
+      "grad_norm": 0.2139723251065959,
+      "learning_rate": 1.3700971957722861e-05,
+      "loss": 0.5755,
+      "step": 1205
+    },
+    {
+      "epoch": 1.3296703296703296,
+      "grad_norm": 0.20154696968263214,
+      "learning_rate": 1.3641504834187288e-05,
+      "loss": 0.5905,
+      "step": 1210
+    },
+    {
+      "epoch": 1.335164835164835,
+      "grad_norm": 0.21481447035842982,
+      "learning_rate": 1.3581888874355969e-05,
+      "loss": 0.5968,
+      "step": 1215
+    },
+    {
+      "epoch": 1.3406593406593408,
+      "grad_norm": 0.23044920349242504,
+      "learning_rate": 1.3522126514864047e-05,
+      "loss": 0.5788,
+      "step": 1220
+    },
+    {
+      "epoch": 1.3461538461538463,
+      "grad_norm": 0.19789912123306916,
+      "learning_rate": 1.346222019833033e-05,
+      "loss": 0.5808,
+      "step": 1225
+    },
+    {
+      "epoch": 1.3516483516483517,
+      "grad_norm": 0.21068837876070498,
+      "learning_rate": 1.3402172373257466e-05,
+      "loss": 0.5777,
+      "step": 1230
+    },
+    {
+      "epoch": 1.3571428571428572,
+      "grad_norm": 0.20881025023596286,
+      "learning_rate": 1.3341985493931877e-05,
+      "loss": 0.5918,
+      "step": 1235
+    },
+    {
+      "epoch": 1.3626373626373627,
+      "grad_norm": 0.21040201908308628,
+      "learning_rate": 1.3281662020323434e-05,
+      "loss": 0.5678,
+      "step": 1240
+    },
+    {
+      "epoch": 1.3681318681318682,
+      "grad_norm": 0.21648376396617017,
+      "learning_rate": 1.3221204417984907e-05,
+      "loss": 0.5816,
+      "step": 1245
+    },
+    {
+      "epoch": 1.3736263736263736,
+      "grad_norm": 0.23069884686246664,
+      "learning_rate": 1.3160615157951218e-05,
+      "loss": 0.5825,
+      "step": 1250
+    },
+    {
+      "epoch": 1.379120879120879,
+      "grad_norm": 0.21219224260123462,
+      "learning_rate": 1.3099896716638414e-05,
+      "loss": 0.5862,
+      "step": 1255
+    },
+    {
+      "epoch": 1.3846153846153846,
+      "grad_norm": 0.20937858254743488,
+      "learning_rate": 1.303905157574247e-05,
+      "loss": 0.5754,
+      "step": 1260
+    },
+    {
+      "epoch": 1.39010989010989,
+      "grad_norm": 0.287029581105537,
+      "learning_rate": 1.297808222213785e-05,
+      "loss": 0.5723,
+      "step": 1265
+    },
+    {
+      "epoch": 1.3956043956043955,
+      "grad_norm": 0.2115500774984686,
+      "learning_rate": 1.2916991147775867e-05,
+      "loss": 0.5903,
+      "step": 1270
+    },
+    {
+      "epoch": 1.401098901098901,
+      "grad_norm": 0.2360214862879323,
+      "learning_rate": 1.2855780849582828e-05,
+      "loss": 0.5995,
+      "step": 1275
+    },
+    {
+      "epoch": 1.4065934065934065,
+      "grad_norm": 0.20079252362217995,
+      "learning_rate": 1.2794453829357974e-05,
+      "loss": 0.5845,
+      "step": 1280
+    },
+    {
+      "epoch": 1.412087912087912,
+      "grad_norm": 0.21328822844172537,
+      "learning_rate": 1.2733012593671235e-05,
+      "loss": 0.5764,
+      "step": 1285
+    },
+    {
+      "epoch": 1.4175824175824177,
+      "grad_norm": 0.20595464274671632,
+      "learning_rate": 1.2671459653760781e-05,
+      "loss": 0.5722,
+      "step": 1290
+    },
+    {
+      "epoch": 1.4230769230769231,
+      "grad_norm": 0.21114157296745673,
+      "learning_rate": 1.2609797525430374e-05,
+      "loss": 0.5851,
+      "step": 1295
+    },
+    {
+      "epoch": 1.4285714285714286,
+      "grad_norm": 0.28130303512847965,
+      "learning_rate": 1.2548028728946548e-05,
+      "loss": 0.5736,
+      "step": 1300
+    },
+    {
+      "epoch": 1.4285714285714286,
+      "eval_loss": 0.6746197938919067,
+      "eval_runtime": 56.6881,
+      "eval_samples_per_second": 24.308,
+      "eval_steps_per_second": 0.106,
+      "step": 1300
+    },
+    {
+      "epoch": 1.434065934065934,
+      "grad_norm": 0.2112732742944607,
+      "learning_rate": 1.2486155788935599e-05,
+      "loss": 0.5856,
+      "step": 1305
+    },
+    {
+      "epoch": 1.4395604395604396,
+      "grad_norm": 0.21296297985278367,
+      "learning_rate": 1.24241812342804e-05,
+      "loss": 0.5741,
+      "step": 1310
+    },
+    {
+      "epoch": 1.445054945054945,
+      "grad_norm": 0.20477018149128742,
+      "learning_rate": 1.2362107598017037e-05,
+      "loss": 0.5747,
+      "step": 1315
+    },
+    {
+      "epoch": 1.4505494505494505,
+      "grad_norm": 0.20359961146660085,
+      "learning_rate": 1.2299937417231269e-05,
+      "loss": 0.5909,
+      "step": 1320
+    },
+    {
+      "epoch": 1.456043956043956,
+      "grad_norm": 0.19983003339561645,
+      "learning_rate": 1.2237673232954854e-05,
+      "loss": 0.5819,
+      "step": 1325
+    },
+    {
+      "epoch": 1.4615384615384617,
+      "grad_norm": 0.21585126952570063,
+      "learning_rate": 1.2175317590061676e-05,
+      "loss": 0.5744,
+      "step": 1330
+    },
+    {
+      "epoch": 1.4670329670329672,
+      "grad_norm": 0.2164749075408294,
+      "learning_rate": 1.2112873037163728e-05,
+      "loss": 0.5921,
+      "step": 1335
+    },
+    {
+      "epoch": 1.4725274725274726,
+      "grad_norm": 0.21214564202757996,
+      "learning_rate": 1.2050342126506958e-05,
+      "loss": 0.6007,
+      "step": 1340
+    },
+    {
+      "epoch": 1.478021978021978,
+      "grad_norm": 0.20961989493961736,
+      "learning_rate": 1.1987727413866936e-05,
+      "loss": 0.5922,
+      "step": 1345
+    },
+    {
+      "epoch": 1.4835164835164836,
+      "grad_norm": 0.2070781399263859,
+      "learning_rate": 1.1925031458444416e-05,
+      "loss": 0.5579,
+      "step": 1350
+    },
+    {
+      "epoch": 1.489010989010989,
+      "grad_norm": 0.21230956255901673,
+      "learning_rate": 1.1862256822760704e-05,
+      "loss": 0.583,
+      "step": 1355
+    },
+    {
+      "epoch": 1.4945054945054945,
+      "grad_norm": 0.20764551222516942,
+      "learning_rate": 1.1799406072552963e-05,
+      "loss": 0.5881,
+      "step": 1360
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.2216686537947161,
+      "learning_rate": 1.1736481776669307e-05,
+      "loss": 0.5884,
+      "step": 1365
+    },
+    {
+      "epoch": 1.5054945054945055,
+      "grad_norm": 0.20026575192100984,
+      "learning_rate": 1.1673486506963824e-05,
+      "loss": 0.5789,
+      "step": 1370
+    },
+    {
+      "epoch": 1.510989010989011,
+      "grad_norm": 0.20025791445692404,
+      "learning_rate": 1.1610422838191473e-05,
+      "loss": 0.5757,
+      "step": 1375
+    },
+    {
+      "epoch": 1.5164835164835164,
+      "grad_norm": 0.21762293248064593,
+      "learning_rate": 1.1547293347902813e-05,
+      "loss": 0.595,
+      "step": 1380
+    },
+    {
+      "epoch": 1.521978021978022,
+      "grad_norm": 0.20522965991028572,
+      "learning_rate": 1.148410061633869e-05,
+      "loss": 0.5825,
+      "step": 1385
+    },
+    {
+      "epoch": 1.5274725274725274,
+      "grad_norm": 0.2007877147929741,
+      "learning_rate": 1.1420847226324746e-05,
+      "loss": 0.5699,
+      "step": 1390
+    },
+    {
+      "epoch": 1.5329670329670328,
+      "grad_norm": 0.2117099052435157,
+      "learning_rate": 1.135753576316588e-05,
+      "loss": 0.5814,
+      "step": 1395
+    },
+    {
+      "epoch": 1.5384615384615383,
+      "grad_norm": 0.26509052806023586,
+      "learning_rate": 1.1294168814540554e-05,
+      "loss": 0.5906,
+      "step": 1400
+    },
+    {
+      "epoch": 1.5384615384615383,
+      "eval_loss": 0.6723337769508362,
+      "eval_runtime": 55.2482,
+      "eval_samples_per_second": 24.942,
+      "eval_steps_per_second": 0.109,
+      "step": 1400
+    },
+    {
+      "epoch": 1.5439560439560438,
+      "grad_norm": 0.21156955434544839,
+      "learning_rate": 1.1230748970395056e-05,
+      "loss": 0.577,
+      "step": 1405
+    },
+    {
+      "epoch": 1.5494505494505495,
+      "grad_norm": 0.20751359958793425,
+      "learning_rate": 1.1167278822837621e-05,
+      "loss": 0.5655,
+      "step": 1410
+    },
+    {
+      "epoch": 1.554945054945055,
+      "grad_norm": 0.21010912643848495,
+      "learning_rate": 1.1103760966032497e-05,
+      "loss": 0.5657,
+      "step": 1415
+    },
+    {
+      "epoch": 1.5604395604395604,
+      "grad_norm": 0.20804119817991984,
+      "learning_rate": 1.1040197996093915e-05,
+      "loss": 0.5829,
+      "step": 1420
+    },
+    {
+      "epoch": 1.565934065934066,
+      "grad_norm": 0.200793527641334,
+      "learning_rate": 1.0976592510979982e-05,
+      "loss": 0.5695,
+      "step": 1425
+    },
+    {
+      "epoch": 1.5714285714285714,
+      "grad_norm": 0.21571689584942777,
+      "learning_rate": 1.0912947110386484e-05,
+      "loss": 0.5687,
+      "step": 1430
+    },
+    {
+      "epoch": 1.5769230769230769,
+      "grad_norm": 0.20881262462287056,
+      "learning_rate": 1.084926439564065e-05,
+      "loss": 0.5813,
+      "step": 1435
+    },
+    {
+      "epoch": 1.5824175824175826,
+      "grad_norm": 0.19724870907211803,
+      "learning_rate": 1.0785546969594813e-05,
+      "loss": 0.5695,
+      "step": 1440
+    },
+    {
+      "epoch": 1.587912087912088,
+      "grad_norm": 0.19796826932631745,
+      "learning_rate": 1.0721797436520044e-05,
+      "loss": 0.5745,
+      "step": 1445
+    },
+    {
+      "epoch": 1.5934065934065935,
+      "grad_norm": 0.20623299794109134,
+      "learning_rate": 1.0658018401999681e-05,
+      "loss": 0.5916,
+      "step": 1450
+    },
+    {
+      "epoch": 1.598901098901099,
+      "grad_norm": 0.22551529540554374,
+      "learning_rate": 1.0594212472822865e-05,
+      "loss": 0.5844,
+      "step": 1455
+    },
+    {
+      "epoch": 1.6043956043956045,
+      "grad_norm": 0.20013753224483868,
+      "learning_rate": 1.053038225687798e-05,
+      "loss": 0.5767,
+      "step": 1460
+    },
+    {
+      "epoch": 1.60989010989011,
+      "grad_norm": 0.20025644857315295,
+      "learning_rate": 1.0466530363046057e-05,
+      "loss": 0.57,
+      "step": 1465
+    },
+    {
+      "epoch": 1.6153846153846154,
+      "grad_norm": 0.22069359110995249,
+      "learning_rate": 1.0402659401094154e-05,
+      "loss": 0.5803,
+      "step": 1470
+    },
+    {
+      "epoch": 1.620879120879121,
+      "grad_norm": 0.2016457322498759,
+      "learning_rate": 1.033877198156868e-05,
+      "loss": 0.5711,
+      "step": 1475
+    },
+    {
+      "epoch": 1.6263736263736264,
+      "grad_norm": 0.20948058749001328,
+      "learning_rate": 1.0274870715688713e-05,
+      "loss": 0.5614,
+      "step": 1480
+    },
+    {
+      "epoch": 1.6318681318681318,
+      "grad_norm": 0.20613326525368722,
+      "learning_rate": 1.0210958215239249e-05,
+      "loss": 0.5617,
+      "step": 1485
+    },
+    {
+      "epoch": 1.6373626373626373,
+      "grad_norm": 0.20057743312747597,
+      "learning_rate": 1.0147037092464469e-05,
+      "loss": 0.5749,
+      "step": 1490
+    },
+    {
+      "epoch": 1.6428571428571428,
+      "grad_norm": 0.21757021973274904,
+      "learning_rate": 1.0083109959960974e-05,
+      "loss": 0.5719,
+      "step": 1495
+    },
+    {
+      "epoch": 1.6483516483516483,
+      "grad_norm": 0.1988420530692502,
+      "learning_rate": 1.0019179430570984e-05,
+      "loss": 0.569,
+      "step": 1500
+    },
+    {
+      "epoch": 1.6483516483516483,
+      "eval_loss": 0.6686215996742249,
+      "eval_runtime": 57.8791,
+      "eval_samples_per_second": 23.808,
+      "eval_steps_per_second": 0.104,
+      "step": 1500
+    },
+    {
+      "epoch": 1.6538461538461537,
+      "grad_norm": 0.21660113082980018,
+      "learning_rate": 9.955248117275566e-06,
+      "loss": 0.5584,
+      "step": 1505
+    },
+    {
+      "epoch": 1.6593406593406592,
+      "grad_norm": 0.23375649427094025,
+      "learning_rate": 9.891318633087831e-06,
+      "loss": 0.5515,
+      "step": 1510
+    },
+    {
+      "epoch": 1.6648351648351647,
+      "grad_norm": 0.20598295650184129,
+      "learning_rate": 9.827393590946116e-06,
+      "loss": 0.5591,
+      "step": 1515
+    },
+    {
+      "epoch": 1.6703296703296702,
+      "grad_norm": 0.19812798829418632,
+      "learning_rate": 9.763475603607215e-06,
+      "loss": 0.5589,
+      "step": 1520
+    },
+    {
+      "epoch": 1.6758241758241759,
+      "grad_norm": 0.20951557993767064,
+      "learning_rate": 9.699567283539567e-06,
+      "loss": 0.5777,
+      "step": 1525
+    },
+    {
+      "epoch": 1.6813186813186813,
+      "grad_norm": 0.20861568891956944,
+      "learning_rate": 9.635671242816503e-06,
+      "loss": 0.5793,
+      "step": 1530
+    },
+    {
+      "epoch": 1.6868131868131868,
+      "grad_norm": 0.2045724615591832,
+      "learning_rate": 9.571790093009445e-06,
+      "loss": 0.5507,
+      "step": 1535
+    },
+    {
+      "epoch": 1.6923076923076923,
+      "grad_norm": 0.2062400013339212,
+      "learning_rate": 9.50792644508122e-06,
+      "loss": 0.5859,
+      "step": 1540
+    },
+    {
+      "epoch": 1.6978021978021978,
+      "grad_norm": 0.20539716997080384,
+      "learning_rate": 9.44408290927929e-06,
+      "loss": 0.5883,
+      "step": 1545
+    },
+    {
+      "epoch": 1.7032967032967035,
+      "grad_norm": 0.2007069765651346,
+      "learning_rate": 9.380262095029113e-06,
+      "loss": 0.5433,
+      "step": 1550
+    },
+    {
+      "epoch": 1.708791208791209,
+      "grad_norm": 0.19988561870754676,
+      "learning_rate": 9.316466610827446e-06,
+      "loss": 0.5681,
+      "step": 1555
+    },
+    {
+      "epoch": 1.7142857142857144,
+      "grad_norm": 0.20674989840235833,
+      "learning_rate": 9.252699064135759e-06,
+      "loss": 0.56,
+      "step": 1560
+    },
+    {
+      "epoch": 1.7197802197802199,
+      "grad_norm": 0.19685538726018847,
+      "learning_rate": 9.188962061273664e-06,
+      "loss": 0.5762,
+      "step": 1565
+    },
+    {
+      "epoch": 1.7252747252747254,
+      "grad_norm": 0.20541519207163753,
+      "learning_rate": 9.125258207312365e-06,
+      "loss": 0.5637,
+      "step": 1570
+    },
+    {
+      "epoch": 1.7307692307692308,
+      "grad_norm": 0.1975839480015212,
+      "learning_rate": 9.061590105968208e-06,
+      "loss": 0.5529,
+      "step": 1575
+    },
+    {
+      "epoch": 1.7362637362637363,
+      "grad_norm": 0.22500565969016117,
+      "learning_rate": 8.997960359496248e-06,
+      "loss": 0.5736,
+      "step": 1580
+    },
+    {
+      "epoch": 1.7417582417582418,
+      "grad_norm": 0.19091706107610168,
+      "learning_rate": 8.934371568583893e-06,
+      "loss": 0.5683,
+      "step": 1585
+    },
+    {
+      "epoch": 1.7472527472527473,
+      "grad_norm": 0.2105931306233096,
+      "learning_rate": 8.8708263322446e-06,
+      "loss": 0.5751,
+      "step": 1590
+    },
+    {
+      "epoch": 1.7527472527472527,
+      "grad_norm": 0.20458297854035193,
+      "learning_rate": 8.807327247711667e-06,
+      "loss": 0.5873,
+      "step": 1595
+    },
+    {
+      "epoch": 1.7582417582417582,
+      "grad_norm": 0.19955511509412213,
+      "learning_rate": 8.743876910332057e-06,
+      "loss": 0.5756,
+      "step": 1600
+    },
+    {
+      "epoch": 1.7582417582417582,
+      "eval_loss": 0.6655026078224182,
+      "eval_runtime": 58.1146,
+      "eval_samples_per_second": 23.712,
+      "eval_steps_per_second": 0.103,
+      "step": 1600
+    },
+    {
+      "epoch": 1.7637362637362637,
+      "grad_norm": 0.20167066864184663,
+      "learning_rate": 8.680477913460339e-06,
+      "loss": 0.5625,
+      "step": 1605
+    },
+    {
+      "epoch": 1.7692307692307692,
+      "grad_norm": 0.20048559966418272,
+      "learning_rate": 8.617132848352672e-06,
+      "loss": 0.589,
+      "step": 1610
+    },
+    {
+      "epoch": 1.7747252747252746,
+      "grad_norm": 0.20227923458206726,
+      "learning_rate": 8.553844304060908e-06,
+      "loss": 0.5601,
+      "step": 1615
+    },
+    {
+      "epoch": 1.7802197802197801,
+      "grad_norm": 0.19698401864090306,
+      "learning_rate": 8.490614867326775e-06,
+      "loss": 0.5653,
+      "step": 1620
+    },
+    {
+      "epoch": 1.7857142857142856,
+      "grad_norm": 0.20175985199043966,
+      "learning_rate": 8.427447122476148e-06,
+      "loss": 0.576,
+      "step": 1625
+    },
+    {
+      "epoch": 1.791208791208791,
+      "grad_norm": 0.20649958059932677,
+      "learning_rate": 8.364343651313406e-06,
+      "loss": 0.5789,
+      "step": 1630
+    },
+    {
+      "epoch": 1.7967032967032965,
+      "grad_norm": 0.20362236096913583,
+      "learning_rate": 8.301307033015928e-06,
+      "loss": 0.5743,
+      "step": 1635
+    },
+    {
+      "epoch": 1.8021978021978022,
+      "grad_norm": 0.1986543134675767,
+      "learning_rate": 8.23833984402868e-06,
+      "loss": 0.5643,
+      "step": 1640
+    },
+    {
+      "epoch": 1.8076923076923077,
+      "grad_norm": 0.198913330885529,
+      "learning_rate": 8.175444657958875e-06,
+      "loss": 0.5735,
+      "step": 1645
+    },
+    {
+      "epoch": 1.8131868131868132,
+      "grad_norm": 0.19864368535876992,
+      "learning_rate": 8.112624045470834e-06,
+      "loss": 0.5717,
+      "step": 1650
+    },
+    {
+      "epoch": 1.8186813186813187,
+      "grad_norm": 0.19883717015134883,
+      "learning_rate": 8.04988057418088e-06,
+      "loss": 0.5557,
+      "step": 1655
+    },
+    {
+      "epoch": 1.8241758241758241,
+      "grad_norm": 0.20332693450451117,
+      "learning_rate": 7.987216808552409e-06,
+      "loss": 0.5736,
+      "step": 1660
+    },
+    {
+      "epoch": 1.8296703296703298,
+      "grad_norm": 0.2022305746327796,
+      "learning_rate": 7.924635309791065e-06,
+      "loss": 0.5605,
+      "step": 1665
+    },
+    {
+      "epoch": 1.8351648351648353,
+      "grad_norm": 0.20043133551344314,
+      "learning_rate": 7.862138635740078e-06,
+      "loss": 0.5509,
+      "step": 1670
+    },
+    {
+      "epoch": 1.8406593406593408,
+      "grad_norm": 0.22455898506641436,
+      "learning_rate": 7.799729340775688e-06,
+      "loss": 0.5869,
+      "step": 1675
+    },
+    {
+      "epoch": 1.8461538461538463,
+      "grad_norm": 0.19030565857505835,
+      "learning_rate": 7.73740997570278e-06,
+      "loss": 0.5657,
+      "step": 1680
+    },
+    {
+      "epoch": 1.8516483516483517,
+      "grad_norm": 0.20815080209673342,
+      "learning_rate": 7.675183087650592e-06,
+      "loss": 0.566,
+      "step": 1685
+    },
+    {
+      "epoch": 1.8571428571428572,
+      "grad_norm": 0.20412451434740683,
+      "learning_rate": 7.613051219968624e-06,
+      "loss": 0.5571,
+      "step": 1690
+    },
+    {
+      "epoch": 1.8626373626373627,
+      "grad_norm": 0.20467604912515247,
+      "learning_rate": 7.551016912122692e-06,
+      "loss": 0.5677,
+      "step": 1695
+    },
+    {
+      "epoch": 1.8681318681318682,
+      "grad_norm": 0.2049947541178511,
+      "learning_rate": 7.489082699591128e-06,
+      "loss": 0.545,
+      "step": 1700
+    },
+    {
+      "epoch": 1.8681318681318682,
+      "eval_loss": 0.6621970534324646,
+      "eval_runtime": 53.6631,
+      "eval_samples_per_second": 25.679,
+      "eval_steps_per_second": 0.112,
+      "step": 1700
+    },
+    {
+      "epoch": 1.8736263736263736,
+      "grad_norm": 0.20749354283076393,
+      "learning_rate": 7.4272511137611405e-06,
+      "loss": 0.5875,
+      "step": 1705
+    },
+    {
+      "epoch": 1.879120879120879,
+      "grad_norm": 0.19337165187460637,
+      "learning_rate": 7.3655246818253626e-06,
+      "loss": 0.5633,
+      "step": 1710
+    },
+    {
+      "epoch": 1.8846153846153846,
+      "grad_norm": 0.20317650293507142,
+      "learning_rate": 7.303905926678565e-06,
+      "loss": 0.5625,
+      "step": 1715
+    },
+    {
+      "epoch": 1.89010989010989,
+      "grad_norm": 0.20166481148644033,
+      "learning_rate": 7.242397366814516e-06,
+      "loss": 0.5444,
+      "step": 1720
+    },
+    {
+      "epoch": 1.8956043956043955,
+      "grad_norm": 0.20130249873133135,
+      "learning_rate": 7.181001516223074e-06,
+      "loss": 0.5501,
+      "step": 1725
+    },
+    {
+      "epoch": 1.901098901098901,
+      "grad_norm": 0.20104391428043816,
+      "learning_rate": 7.1197208842874175e-06,
+      "loss": 0.5568,
+      "step": 1730
+    },
+    {
+      "epoch": 1.9065934065934065,
+      "grad_norm": 0.19669831797494047,
+      "learning_rate": 7.058557975681488e-06,
+      "loss": 0.5691,
+      "step": 1735
+    },
+    {
+      "epoch": 1.912087912087912,
+      "grad_norm": 0.19994455544845724,
+      "learning_rate": 6.997515290267611e-06,
+      "loss": 0.5641,
+      "step": 1740
+    },
+    {
+      "epoch": 1.9175824175824174,
+      "grad_norm": 0.2022020053476063,
+      "learning_rate": 6.936595322994328e-06,
+      "loss": 0.5625,
+      "step": 1745
+    },
+    {
+      "epoch": 1.9230769230769231,
+      "grad_norm": 0.20426025496672773,
+      "learning_rate": 6.8758005637944245e-06,
+      "loss": 0.5662,
+      "step": 1750
+    },
+    {
+      "epoch": 1.9285714285714286,
+      "grad_norm": 0.19660653769715136,
+      "learning_rate": 6.815133497483157e-06,
+      "loss": 0.575,
+      "step": 1755
+    },
+    {
+      "epoch": 1.934065934065934,
+      "grad_norm": 0.2091108562276251,
+      "learning_rate": 6.754596603656687e-06,
+      "loss": 0.5668,
+      "step": 1760
+    },
+    {
+      "epoch": 1.9395604395604396,
+      "grad_norm": 0.1990872798241454,
+      "learning_rate": 6.694192356590743e-06,
+      "loss": 0.5677,
+      "step": 1765
+    },
+    {
+      "epoch": 1.945054945054945,
+      "grad_norm": 0.2035184385391121,
+      "learning_rate": 6.633923225139498e-06,
+      "loss": 0.5848,
+      "step": 1770
+    },
+    {
+      "epoch": 1.9505494505494505,
+      "grad_norm": 0.2087152884886787,
+      "learning_rate": 6.573791672634638e-06,
+      "loss": 0.5779,
+      "step": 1775
+    },
+    {
+      "epoch": 1.9560439560439562,
+      "grad_norm": 0.19903589160386567,
+      "learning_rate": 6.513800156784709e-06,
+      "loss": 0.5545,
+      "step": 1780
+    },
+    {
+      "epoch": 1.9615384615384617,
+      "grad_norm": 0.1988257810600432,
+      "learning_rate": 6.453951129574644e-06,
+      "loss": 0.5591,
+      "step": 1785
+    },
+    {
+      "epoch": 1.9670329670329672,
+      "grad_norm": 0.2440643065747747,
+      "learning_rate": 6.394247037165559e-06,
+      "loss": 0.5575,
+      "step": 1790
+    },
+    {
+      "epoch": 1.9725274725274726,
+      "grad_norm": 0.19915770600503882,
+      "learning_rate": 6.3346903197947564e-06,
+      "loss": 0.5713,
+      "step": 1795
+    },
+    {
+      "epoch": 1.978021978021978,
+      "grad_norm": 0.19824070469872943,
+      "learning_rate": 6.275283411676008e-06,
+      "loss": 0.5505,
+      "step": 1800
+    },
+    {
+      "epoch": 1.978021978021978,
+      "eval_loss": 0.6606090068817139,
+      "eval_runtime": 57.2329,
+      "eval_samples_per_second": 24.077,
+      "eval_steps_per_second": 0.105,
+      "step": 1800
+    },
+    {
+      "epoch": 1.9835164835164836,
+      "grad_norm": 0.20291412687117444,
+      "learning_rate": 6.216028740900042e-06,
+      "loss": 0.5715,
+      "step": 1805
+    },
+    {
+      "epoch": 1.989010989010989,
+      "grad_norm": 0.1992635144672437,
+      "learning_rate": 6.1569287293353274e-06,
+      "loss": 0.5648,
+      "step": 1810
+    },
+    {
+      "epoch": 1.9945054945054945,
+      "grad_norm": 0.20133042378281404,
+      "learning_rate": 6.097985792529055e-06,
+      "loss": 0.5718,
+      "step": 1815
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.18939749792824842,
+      "learning_rate": 6.039202339608432e-06,
+      "loss": 0.5509,
+      "step": 1820
+    },
+    {
+      "epoch": 2.0054945054945055,
+      "grad_norm": 0.21548294059114115,
+      "learning_rate": 5.980580773182214e-06,
+      "loss": 0.5084,
+      "step": 1825
+    },
+    {
+      "epoch": 2.010989010989011,
+      "grad_norm": 0.19938738432640987,
+      "learning_rate": 5.922123489242499e-06,
+      "loss": 0.5162,
+      "step": 1830
+    },
+    {
+      "epoch": 2.0164835164835164,
+      "grad_norm": 0.2027789624856937,
+      "learning_rate": 5.8638328770667905e-06,
+      "loss": 0.5119,
+      "step": 1835
+    },
+    {
+      "epoch": 2.021978021978022,
+      "grad_norm": 0.20197197450501087,
+      "learning_rate": 5.805711319120358e-06,
+      "loss": 0.5093,
+      "step": 1840
+    },
+    {
+      "epoch": 2.0274725274725274,
+      "grad_norm": 0.20124091857994694,
+      "learning_rate": 5.747761190958859e-06,
+      "loss": 0.5118,
+      "step": 1845
+    },
+    {
+      "epoch": 2.032967032967033,
+      "grad_norm": 0.2060214343274437,
+      "learning_rate": 5.689984861131221e-06,
+      "loss": 0.529,
+      "step": 1850
+    },
+    {
+      "epoch": 2.0384615384615383,
+      "grad_norm": 0.1949171402387555,
+      "learning_rate": 5.632384691082874e-06,
+      "loss": 0.5038,
+      "step": 1855
+    },
+    {
+      "epoch": 2.043956043956044,
+      "grad_norm": 0.2065371272676731,
+      "learning_rate": 5.5749630350592e-06,
+      "loss": 0.524,
+      "step": 1860
+    },
+    {
+      "epoch": 2.0494505494505493,
+      "grad_norm": 0.20211687410000778,
+      "learning_rate": 5.517722240009319e-06,
+      "loss": 0.5249,
+      "step": 1865
+    },
+    {
+      "epoch": 2.0549450549450547,
+      "grad_norm": 0.21155781039557575,
+      "learning_rate": 5.460664645490172e-06,
+      "loss": 0.5222,
+      "step": 1870
+    },
+    {
+      "epoch": 2.0604395604395602,
+      "grad_norm": 0.20557792586063273,
+      "learning_rate": 5.403792583570884e-06,
+      "loss": 0.5122,
+      "step": 1875
+    },
+    {
+      "epoch": 2.065934065934066,
+      "grad_norm": 0.20892091915093894,
+      "learning_rate": 5.347108378737469e-06,
+      "loss": 0.5209,
+      "step": 1880
+    },
+    {
+      "epoch": 2.0714285714285716,
+      "grad_norm": 0.21130418206749751,
+      "learning_rate": 5.290614347797802e-06,
+      "loss": 0.4999,
+      "step": 1885
+    },
+    {
+      "epoch": 2.076923076923077,
+      "grad_norm": 0.20265157501265516,
+      "learning_rate": 5.234312799786921e-06,
+      "loss": 0.5113,
+      "step": 1890
+    },
+    {
+      "epoch": 2.0824175824175826,
+      "grad_norm": 0.2380811857415137,
+      "learning_rate": 5.1782060358726885e-06,
+      "loss": 0.5137,
+      "step": 1895
+    },
+    {
+      "epoch": 2.087912087912088,
+      "grad_norm": 0.20021684752472454,
+      "learning_rate": 5.122296349261695e-06,
+      "loss": 0.5149,
+      "step": 1900
+    },
+    {
+      "epoch": 2.087912087912088,
+      "eval_loss": 0.6648159027099609,
+      "eval_runtime": 57.1476,
+      "eval_samples_per_second": 24.113,
+      "eval_steps_per_second": 0.105,
+      "step": 1900
+    },
+    {
+      "epoch": 2.0934065934065935,
+      "grad_norm": 0.20298609856841238,
+      "learning_rate": 5.066586025105558e-06,
+      "loss": 0.527,
+      "step": 1905
+    },
+    {
+      "epoch": 2.098901098901099,
+      "grad_norm": 0.2035387697050701,
+      "learning_rate": 5.011077340407509e-06,
+      "loss": 0.5055,
+      "step": 1910
+    },
+    {
+      "epoch": 2.1043956043956045,
+      "grad_norm": 0.19786708962281988,
+      "learning_rate": 4.955772563929334e-06,
+      "loss": 0.5151,
+      "step": 1915
+    },
+    {
+      "epoch": 2.10989010989011,
+      "grad_norm": 0.20387216209111395,
+      "learning_rate": 4.900673956098644e-06,
+      "loss": 0.5049,
+      "step": 1920
+    },
+    {
+      "epoch": 2.1153846153846154,
+      "grad_norm": 0.20554401944132458,
+      "learning_rate": 4.845783768916482e-06,
+      "loss": 0.519,
+      "step": 1925
+    },
+    {
+      "epoch": 2.120879120879121,
+      "grad_norm": 0.2111896896707393,
+      "learning_rate": 4.79110424586528e-06,
+      "loss": 0.5193,
+      "step": 1930
+    },
+    {
+      "epoch": 2.1263736263736264,
+      "grad_norm": 0.20409361712091278,
+      "learning_rate": 4.736637621817176e-06,
+      "loss": 0.5134,
+      "step": 1935
+    },
+    {
+      "epoch": 2.131868131868132,
+      "grad_norm": 0.20305198178835793,
+      "learning_rate": 4.682386122942649e-06,
+      "loss": 0.5338,
+      "step": 1940
+    },
+    {
+      "epoch": 2.1373626373626373,
+      "grad_norm": 0.20969960596506004,
+      "learning_rate": 4.628351966619531e-06,
+      "loss": 0.5285,
+      "step": 1945
+    },
+    {
+      "epoch": 2.142857142857143,
+      "grad_norm": 0.2016497737059224,
+      "learning_rate": 4.5745373613424075e-06,
+      "loss": 0.5196,
+      "step": 1950
+    },
+    {
+      "epoch": 2.1483516483516483,
+      "grad_norm": 0.20780959230200222,
+      "learning_rate": 4.520944506632314e-06,
+      "loss": 0.5102,
+      "step": 1955
+    },
+    {
+      "epoch": 2.1538461538461537,
+      "grad_norm": 0.20807222664357544,
+      "learning_rate": 4.467575592946865e-06,
+      "loss": 0.5325,
+      "step": 1960
+    },
+    {
+      "epoch": 2.159340659340659,
+      "grad_norm": 0.2167502541161845,
+      "learning_rate": 4.414432801590703e-06,
+      "loss": 0.5078,
+      "step": 1965
+    },
+    {
+      "epoch": 2.1648351648351647,
+      "grad_norm": 0.21490727997644427,
+      "learning_rate": 4.361518304626366e-06,
+      "loss": 0.5407,
+      "step": 1970
+    },
+    {
+      "epoch": 2.17032967032967,
+      "grad_norm": 0.21734918740860756,
+      "learning_rate": 4.308834264785483e-06,
+      "loss": 0.5141,
+      "step": 1975
+    },
+    {
+      "epoch": 2.1758241758241756,
+      "grad_norm": 0.1991788993982973,
+      "learning_rate": 4.256382835380421e-06,
+      "loss": 0.5069,
+      "step": 1980
+    },
+    {
+      "epoch": 2.181318681318681,
+      "grad_norm": 0.2143454079537833,
+      "learning_rate": 4.204166160216216e-06,
+      "loss": 0.5249,
+      "step": 1985
+    },
+    {
+      "epoch": 2.186813186813187,
+      "grad_norm": 0.19879558037154738,
+      "learning_rate": 4.1521863735030065e-06,
+      "loss": 0.4902,
+      "step": 1990
+    },
+    {
+      "epoch": 2.1923076923076925,
+      "grad_norm": 0.20002933623961325,
+      "learning_rate": 4.100445599768774e-06,
+      "loss": 0.5051,
+      "step": 1995
+    },
+    {
+      "epoch": 2.197802197802198,
+      "grad_norm": 0.20285409795219067,
+      "learning_rate": 4.048945953772504e-06,
+      "loss": 0.5234,
+      "step": 2000
+    },
+    {
+      "epoch": 2.197802197802198,
+      "eval_loss": 0.6638170480728149,
+      "eval_runtime": 57.5818,
+      "eval_samples_per_second": 23.931,
+      "eval_steps_per_second": 0.104,
+      "step": 2000
+    },
+    {
+      "epoch": 2.2032967032967035,
+      "grad_norm": 0.2018945000710767,
+      "learning_rate": 3.99768954041778e-06,
+      "loss": 0.5071,
+      "step": 2005
+    },
+    {
+      "epoch": 2.208791208791209,
+      "grad_norm": 0.20646923773133413,
+      "learning_rate": 3.946678454666719e-06,
+      "loss": 0.4988,
+      "step": 2010
+    },
+    {
+      "epoch": 2.2142857142857144,
+      "grad_norm": 0.20067691812172808,
+      "learning_rate": 3.89591478145437e-06,
+      "loss": 0.5182,
+      "step": 2015
+    },
+    {
+      "epoch": 2.21978021978022,
+      "grad_norm": 0.21205391255701184,
+      "learning_rate": 3.845400595603482e-06,
+      "loss": 0.5307,
+      "step": 2020
+    },
+    {
+      "epoch": 2.2252747252747254,
+      "grad_norm": 0.20460875351302696,
+      "learning_rate": 3.79513796173971e-06,
+      "loss": 0.5031,
+      "step": 2025
+    },
+    {
+      "epoch": 2.230769230769231,
+      "grad_norm": 0.1973070576811456,
+      "learning_rate": 3.745128934207225e-06,
+      "loss": 0.5284,
+      "step": 2030
+    },
+    {
+      "epoch": 2.2362637362637363,
+      "grad_norm": 0.21339279534660974,
+      "learning_rate": 3.695375556984764e-06,
+      "loss": 0.5252,
+      "step": 2035
+    },
+    {
+      "epoch": 2.241758241758242,
+      "grad_norm": 0.20247153184924946,
+      "learning_rate": 3.6458798636020477e-06,
+      "loss": 0.5128,
+      "step": 2040
+    },
+    {
+      "epoch": 2.2472527472527473,
+      "grad_norm": 0.19749668386741256,
+      "learning_rate": 3.59664387705672e-06,
+      "loss": 0.5062,
+      "step": 2045
+    },
+    {
+      "epoch": 2.2527472527472527,
+      "grad_norm": 0.21402795878492867,
+      "learning_rate": 3.5476696097316253e-06,
+      "loss": 0.5109,
+      "step": 2050
+    },
+    {
+      "epoch": 2.258241758241758,
+      "grad_norm": 0.2077870859543597,
+      "learning_rate": 3.4989590633125583e-06,
+      "loss": 0.5274,
+      "step": 2055
+    },
+    {
+      "epoch": 2.2637362637362637,
+      "grad_norm": 0.2660838503313638,
+      "learning_rate": 3.450514228706482e-06,
+      "loss": 0.5059,
+      "step": 2060
+    },
+    {
+      "epoch": 2.269230769230769,
+      "grad_norm": 0.20089637585406683,
+      "learning_rate": 3.4023370859601192e-06,
+      "loss": 0.5044,
+      "step": 2065
+    },
+    {
+      "epoch": 2.2747252747252746,
+      "grad_norm": 0.19730583864844892,
+      "learning_rate": 3.3544296041790457e-06,
+      "loss": 0.5002,
+      "step": 2070
+    },
+    {
+      "epoch": 2.28021978021978,
+      "grad_norm": 0.202178274091821,
+      "learning_rate": 3.3067937414471986e-06,
+      "loss": 0.5244,
+      "step": 2075
+    },
+    {
+      "epoch": 2.2857142857142856,
+      "grad_norm": 0.19886623288328378,
+      "learning_rate": 3.2594314447468457e-06,
+      "loss": 0.5115,
+      "step": 2080
+    },
+    {
+      "epoch": 2.291208791208791,
+      "grad_norm": 0.20284655112337058,
+      "learning_rate": 3.2123446498790214e-06,
+      "loss": 0.5108,
+      "step": 2085
+    },
+    {
+      "epoch": 2.2967032967032965,
+      "grad_norm": 0.21307262525015022,
+      "learning_rate": 3.1655352813843886e-06,
+      "loss": 0.5205,
+      "step": 2090
+    },
+    {
+      "epoch": 2.302197802197802,
+      "grad_norm": 0.2016117227949298,
+      "learning_rate": 3.1190052524645752e-06,
+      "loss": 0.5036,
+      "step": 2095
+    },
+    {
+      "epoch": 2.3076923076923075,
+      "grad_norm": 0.20155504502839525,
+      "learning_rate": 3.0727564649040066e-06,
+      "loss": 0.5239,
+      "step": 2100
+    },
+    {
+      "epoch": 2.3076923076923075,
+      "eval_loss": 0.6632149815559387,
+      "eval_runtime": 55.6194,
+      "eval_samples_per_second": 24.776,
+      "eval_steps_per_second": 0.108,
+      "step": 2100
+    },
+    {
+      "epoch": 2.313186813186813,
+      "grad_norm": 0.21425188656541483,
+      "learning_rate": 3.0267908089921438e-06,
+      "loss": 0.5151,
+      "step": 2105
+    },
+    {
+      "epoch": 2.3186813186813184,
+      "grad_norm": 0.20081101739821433,
+      "learning_rate": 2.9811101634462414e-06,
+      "loss": 0.5147,
+      "step": 2110
+    },
+    {
+      "epoch": 2.3241758241758244,
+      "grad_norm": 0.20232270966888258,
+      "learning_rate": 2.93571639533455e-06,
+      "loss": 0.522,
+      "step": 2115
+    },
+    {
+      "epoch": 2.32967032967033,
+      "grad_norm": 0.20125405245316808,
+      "learning_rate": 2.8906113600000153e-06,
+      "loss": 0.5209,
+      "step": 2120
+    },
+    {
+      "epoch": 2.3351648351648353,
+      "grad_norm": 0.20454677450030168,
+      "learning_rate": 2.8457969009844354e-06,
+      "loss": 0.5131,
+      "step": 2125
+    },
+    {
+      "epoch": 2.340659340659341,
+      "grad_norm": 0.2068065615757479,
+      "learning_rate": 2.8012748499531195e-06,
+      "loss": 0.5268,
+      "step": 2130
+    },
+    {
+      "epoch": 2.3461538461538463,
+      "grad_norm": 0.20463770808980064,
+      "learning_rate": 2.7570470266200177e-06,
+      "loss": 0.5269,
+      "step": 2135
+    },
+    {
+      "epoch": 2.3516483516483517,
+      "grad_norm": 0.20997582101579929,
+      "learning_rate": 2.713115238673356e-06,
+      "loss": 0.5247,
+      "step": 2140
+    },
+    {
+      "epoch": 2.357142857142857,
+      "grad_norm": 0.20156173202574512,
+      "learning_rate": 2.669481281701739e-06,
+      "loss": 0.5039,
+      "step": 2145
+    },
+    {
+      "epoch": 2.3626373626373627,
+      "grad_norm": 0.19654077766547687,
+      "learning_rate": 2.626146939120757e-06,
+      "loss": 0.5185,
+      "step": 2150
+    },
+    {
+      "epoch": 2.368131868131868,
+      "grad_norm": 0.20382023726525478,
+      "learning_rate": 2.5831139821001184e-06,
+      "loss": 0.5259,
+      "step": 2155
+    },
+    {
+      "epoch": 2.3736263736263736,
+      "grad_norm": 0.20680053202116233,
+      "learning_rate": 2.5403841694912333e-06,
+      "loss": 0.513,
+      "step": 2160
+    },
+    {
+      "epoch": 2.379120879120879,
+      "grad_norm": 0.1941492774491302,
+      "learning_rate": 2.497959247755335e-06,
+      "loss": 0.5239,
+      "step": 2165
+    },
+    {
+      "epoch": 2.3846153846153846,
+      "grad_norm": 0.20405640437952138,
+      "learning_rate": 2.455840950892099e-06,
+      "loss": 0.5156,
+      "step": 2170
+    },
+    {
+      "epoch": 2.39010989010989,
+      "grad_norm": 0.2843212523020798,
+      "learning_rate": 2.414031000368767e-06,
+      "loss": 0.5034,
+      "step": 2175
+    },
+    {
+      "epoch": 2.3956043956043955,
+      "grad_norm": 0.20197161260199611,
+      "learning_rate": 2.372531105049789e-06,
+      "loss": 0.5003,
+      "step": 2180
+    },
+    {
+      "epoch": 2.401098901098901,
+      "grad_norm": 0.20727591825649075,
+      "learning_rate": 2.331342961126988e-06,
+      "loss": 0.5079,
+      "step": 2185
+    },
+    {
+      "epoch": 2.4065934065934065,
+      "grad_norm": 0.2054230166031501,
+      "learning_rate": 2.290468252050204e-06,
+      "loss": 0.5034,
+      "step": 2190
+    },
+    {
+      "epoch": 2.412087912087912,
+      "grad_norm": 0.2011557586955495,
+      "learning_rate": 2.2499086484585255e-06,
+      "loss": 0.5023,
+      "step": 2195
+    },
+    {
+      "epoch": 2.4175824175824174,
+      "grad_norm": 0.2043529990444942,
+      "learning_rate": 2.2096658081119793e-06,
+      "loss": 0.5142,
+      "step": 2200
+    },
+    {
+      "epoch": 2.4175824175824174,
+      "eval_loss": 0.662318229675293,
+      "eval_runtime": 56.3794,
+      "eval_samples_per_second": 24.442,
+      "eval_steps_per_second": 0.106,
+      "step": 2200
+    },
+    {
+      "epoch": 2.423076923076923,
+      "grad_norm": 0.2007104518256576,
+      "learning_rate": 2.1697413758237785e-06,
+      "loss": 0.5068,
+      "step": 2205
+    },
+    {
+      "epoch": 2.4285714285714284,
+      "grad_norm": 0.20519816271013885,
+      "learning_rate": 2.130136983393112e-06,
+      "loss": 0.5153,
+      "step": 2210
+    },
+    {
+      "epoch": 2.4340659340659343,
+      "grad_norm": 0.19385746620653593,
+      "learning_rate": 2.0908542495384276e-06,
+      "loss": 0.4915,
+      "step": 2215
+    },
+    {
+      "epoch": 2.4395604395604398,
+      "grad_norm": 0.20504945100422145,
+      "learning_rate": 2.051894779831286e-06,
+      "loss": 0.5117,
+      "step": 2220
+    },
+    {
+      "epoch": 2.4450549450549453,
+      "grad_norm": 0.20350739624315656,
+      "learning_rate": 2.0132601666307295e-06,
+      "loss": 0.5118,
+      "step": 2225
+    },
+    {
+      "epoch": 2.4505494505494507,
+      "grad_norm": 0.2038803889957168,
+      "learning_rate": 1.9749519890182035e-06,
+      "loss": 0.5144,
+      "step": 2230
+    },
+    {
+      "epoch": 2.456043956043956,
+      "grad_norm": 0.20292180332730542,
+      "learning_rate": 1.936971812733012e-06,
+      "loss": 0.513,
+      "step": 2235
+    },
+    {
+      "epoch": 2.4615384615384617,
+      "grad_norm": 0.19993755435515206,
+      "learning_rate": 1.8993211901083353e-06,
+      "loss": 0.5145,
+      "step": 2240
+    },
+    {
+      "epoch": 2.467032967032967,
+      "grad_norm": 0.20670573681391916,
+      "learning_rate": 1.8620016600077516e-06,
+      "loss": 0.5159,
+      "step": 2245
+    },
+    {
+      "epoch": 2.4725274725274726,
+      "grad_norm": 0.19479382540898788,
+      "learning_rate": 1.8250147477623836e-06,
+      "loss": 0.5068,
+      "step": 2250
+    },
+    {
+      "epoch": 2.478021978021978,
+      "grad_norm": 0.20005895519952888,
+      "learning_rate": 1.7883619651085194e-06,
+      "loss": 0.5165,
+      "step": 2255
+    },
+    {
+      "epoch": 2.4835164835164836,
+      "grad_norm": 0.2009142357322209,
+      "learning_rate": 1.7520448101258325e-06,
+      "loss": 0.4934,
+      "step": 2260
+    },
+    {
+      "epoch": 2.489010989010989,
+      "grad_norm": 0.20615350297111454,
+      "learning_rate": 1.716064767176172e-06,
+      "loss": 0.4966,
+      "step": 2265
+    },
+    {
+      "epoch": 2.4945054945054945,
+      "grad_norm": 0.19574790617423804,
+      "learning_rate": 1.6804233068428678e-06,
+      "loss": 0.4991,
+      "step": 2270
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.1968274274952375,
+      "learning_rate": 1.6451218858706374e-06,
+      "loss": 0.5132,
+      "step": 2275
+    },
+    {
+      "epoch": 2.5054945054945055,
+      "grad_norm": 0.19627672556133136,
+      "learning_rate": 1.6101619471060415e-06,
+      "loss": 0.4986,
+      "step": 2280
+    },
+    {
+      "epoch": 2.510989010989011,
+      "grad_norm": 0.232139426765921,
+      "learning_rate": 1.5755449194385164e-06,
+      "loss": 0.5119,
+      "step": 2285
+    },
+    {
+      "epoch": 2.5164835164835164,
+      "grad_norm": 0.20274183634685633,
+      "learning_rate": 1.5412722177419658e-06,
+      "loss": 0.5023,
+      "step": 2290
+    },
+    {
+      "epoch": 2.521978021978022,
+      "grad_norm": 0.1991951421133761,
+      "learning_rate": 1.5073452428169444e-06,
+      "loss": 0.4949,
+      "step": 2295
+    },
+    {
+      "epoch": 2.5274725274725274,
+      "grad_norm": 0.19379452216149648,
+      "learning_rate": 1.4737653813333774e-06,
+      "loss": 0.5086,
+      "step": 2300
+    },
+    {
+      "epoch": 2.5274725274725274,
+      "eval_loss": 0.6616050004959106,
+      "eval_runtime": 55.0037,
+      "eval_samples_per_second": 25.053,
+      "eval_steps_per_second": 0.109,
+      "step": 2300
+    },
+    {
+      "epoch": 2.532967032967033,
+      "grad_norm": 0.21494093339800838,
+      "learning_rate": 1.4405340057739203e-06,
+      "loss": 0.5233,
+      "step": 2305
+    },
+    {
+      "epoch": 2.5384615384615383,
+      "grad_norm": 0.20790725753107414,
+      "learning_rate": 1.407652474377832e-06,
+      "loss": 0.5262,
+      "step": 2310
+    },
+    {
+      "epoch": 2.543956043956044,
+      "grad_norm": 0.23490670194522875,
+      "learning_rate": 1.3751221310854778e-06,
+      "loss": 0.5064,
+      "step": 2315
+    },
+    {
+      "epoch": 2.5494505494505493,
+      "grad_norm": 0.2045871257636731,
+      "learning_rate": 1.3429443054833913e-06,
+      "loss": 0.5022,
+      "step": 2320
+    },
+    {
+      "epoch": 2.5549450549450547,
+      "grad_norm": 0.20260454811288686,
+      "learning_rate": 1.311120312749935e-06,
+      "loss": 0.5001,
+      "step": 2325
+    },
+    {
+      "epoch": 2.5604395604395602,
+      "grad_norm": 0.20417611729894805,
+      "learning_rate": 1.2796514536015492e-06,
+      "loss": 0.5051,
+      "step": 2330
+    },
+    {
+      "epoch": 2.5659340659340657,
+      "grad_norm": 0.20980337074664027,
+      "learning_rate": 1.2485390142395793e-06,
+      "loss": 0.5191,
+      "step": 2335
+    },
+    {
+      "epoch": 2.571428571428571,
+      "grad_norm": 0.20318999346422076,
+      "learning_rate": 1.2177842662977136e-06,
+      "loss": 0.5247,
+      "step": 2340
+    },
+    {
+      "epoch": 2.5769230769230766,
+      "grad_norm": 0.19629810633248806,
+      "learning_rate": 1.1873884667900125e-06,
+      "loss": 0.5117,
+      "step": 2345
+    },
+    {
+      "epoch": 2.5824175824175826,
+      "grad_norm": 0.20122953995800785,
+      "learning_rate": 1.1573528580595195e-06,
+      "loss": 0.5022,
+      "step": 2350
+    },
+    {
+      "epoch": 2.587912087912088,
+      "grad_norm": 0.19407168789538073,
+      "learning_rate": 1.1276786677274866e-06,
+      "loss": 0.5018,
+      "step": 2355
+    },
+    {
+      "epoch": 2.5934065934065935,
+      "grad_norm": 0.20866298600964597,
+      "learning_rate": 1.0983671086432146e-06,
+      "loss": 0.5131,
+      "step": 2360
+    },
+    {
+      "epoch": 2.598901098901099,
+      "grad_norm": 0.19859110363436613,
+      "learning_rate": 1.069419378834461e-06,
+      "loss": 0.5072,
+      "step": 2365
+    },
+    {
+      "epoch": 2.6043956043956045,
+      "grad_norm": 0.1972544497689168,
+      "learning_rate": 1.040836661458482e-06,
+      "loss": 0.5121,
+      "step": 2370
+    },
+    {
+      "epoch": 2.60989010989011,
+      "grad_norm": 0.2229095733014629,
+      "learning_rate": 1.0126201247536783e-06,
+      "loss": 0.5023,
+      "step": 2375
+    },
+    {
+      "epoch": 2.6153846153846154,
+      "grad_norm": 0.2078898580527157,
+      "learning_rate": 9.8477092199184e-07,
+      "loss": 0.4999,
+      "step": 2380
+    },
+    {
+      "epoch": 2.620879120879121,
+      "grad_norm": 0.20385043702217534,
+      "learning_rate": 9.57290191431013e-07,
+      "loss": 0.4973,
+      "step": 2385
+    },
+    {
+      "epoch": 2.6263736263736264,
+      "grad_norm": 0.20024077890781708,
+      "learning_rate": 9.301790562689794e-07,
+      "loss": 0.5115,
+      "step": 2390
+    },
+    {
+      "epoch": 2.631868131868132,
+      "grad_norm": 0.20229645144805108,
+      "learning_rate": 9.034386245973359e-07,
+      "loss": 0.518,
+      "step": 2395
+    },
+    {
+      "epoch": 2.6373626373626373,
+      "grad_norm": 0.20806472260190578,
+      "learning_rate": 8.770699893562273e-07,
+      "loss": 0.4998,
+      "step": 2400
+    },
+    {
+      "epoch": 2.6373626373626373,
+      "eval_loss": 0.6603860259056091,
+      "eval_runtime": 55.4261,
+      "eval_samples_per_second": 24.862,
+      "eval_steps_per_second": 0.108,
+      "step": 2400
+    },
+    {
+      "epoch": 2.642857142857143,
+      "grad_norm": 0.20212616235290484,
+      "learning_rate": 8.510742282896545e-07,
+      "loss": 0.5176,
+      "step": 2405
+    },
+    {
+      "epoch": 2.6483516483516483,
+      "grad_norm": 0.19926286905249757,
+      "learning_rate": 8.254524039014289e-07,
+      "loss": 0.504,
+      "step": 2410
+    },
+    {
+      "epoch": 2.6538461538461537,
+      "grad_norm": 0.20491396527113304,
+      "learning_rate": 8.002055634117578e-07,
+      "loss": 0.5127,
+      "step": 2415
+    },
+    {
+      "epoch": 2.659340659340659,
+      "grad_norm": 0.1984394044529964,
+      "learning_rate": 7.753347387144294e-07,
+      "loss": 0.4961,
+      "step": 2420
+    },
+    {
+      "epoch": 2.6648351648351647,
+      "grad_norm": 0.20457728804801084,
+      "learning_rate": 7.508409463346389e-07,
+      "loss": 0.5166,
+      "step": 2425
+    },
+    {
+      "epoch": 2.67032967032967,
+      "grad_norm": 0.2053681782347483,
+      "learning_rate": 7.26725187387446e-07,
+      "loss": 0.4985,
+      "step": 2430
+    },
+    {
+      "epoch": 2.675824175824176,
+      "grad_norm": 0.21034306160224336,
+      "learning_rate": 7.029884475368542e-07,
+      "loss": 0.5286,
+      "step": 2435
+    },
+    {
+      "epoch": 2.6813186813186816,
+      "grad_norm": 0.21208091059236112,
+      "learning_rate": 6.796316969555205e-07,
+      "loss": 0.5135,
+      "step": 2440
+    },
+    {
+      "epoch": 2.686813186813187,
+      "grad_norm": 0.20061617557170722,
+      "learning_rate": 6.566558902851161e-07,
+      "loss": 0.5248,
+      "step": 2445
+    },
+    {
+      "epoch": 2.6923076923076925,
+      "grad_norm": 0.2077389312882985,
+      "learning_rate": 6.340619665972847e-07,
+      "loss": 0.5265,
+      "step": 2450
+    },
+    {
+      "epoch": 2.697802197802198,
+      "grad_norm": 0.1907623497336921,
+      "learning_rate": 6.118508493552866e-07,
+      "loss": 0.5048,
+      "step": 2455
+    },
+    {
+      "epoch": 2.7032967032967035,
+      "grad_norm": 0.19886591589245772,
+      "learning_rate": 5.900234463762367e-07,
+      "loss": 0.506,
+      "step": 2460
+    },
+    {
+      "epoch": 2.708791208791209,
+      "grad_norm": 0.19691007260870758,
+      "learning_rate": 5.685806497940027e-07,
+      "loss": 0.511,
+      "step": 2465
+    },
+    {
+      "epoch": 2.7142857142857144,
+      "grad_norm": 0.19864992732710893,
+      "learning_rate": 5.475233360227516e-07,
+      "loss": 0.5093,
+      "step": 2470
+    },
+    {
+      "epoch": 2.71978021978022,
+      "grad_norm": 0.20129564143877238,
+      "learning_rate": 5.268523657211188e-07,
+      "loss": 0.4973,
+      "step": 2475
+    },
+    {
+      "epoch": 2.7252747252747254,
+      "grad_norm": 0.19814396411057444,
+      "learning_rate": 5.065685837570312e-07,
+      "loss": 0.4985,
+      "step": 2480
+    },
+    {
+      "epoch": 2.730769230769231,
+      "grad_norm": 0.19965550111609479,
+      "learning_rate": 4.866728191731829e-07,
+      "loss": 0.5111,
+      "step": 2485
+    },
+    {
+      "epoch": 2.7362637362637363,
+      "grad_norm": 0.19457379885149514,
+      "learning_rate": 4.671658851531424e-07,
+      "loss": 0.5057,
+      "step": 2490
+    },
+    {
+      "epoch": 2.741758241758242,
+      "grad_norm": 0.19940679212384896,
+      "learning_rate": 4.480485789881217e-07,
+      "loss": 0.5127,
+      "step": 2495
+    },
+    {
+      "epoch": 2.7472527472527473,
+      "grad_norm": 0.19251317932524864,
+      "learning_rate": 4.293216820443891e-07,
+      "loss": 0.5029,
+      "step": 2500
+    },
+    {
+      "epoch": 2.7472527472527473,
+      "eval_loss": 0.6601957678794861,
+      "eval_runtime": 54.7865,
+      "eval_samples_per_second": 25.152,
+      "eval_steps_per_second": 0.11,
+      "step": 2500
+    },
+    {
+      "epoch": 2.7527472527472527,
+      "grad_norm": 0.20141031688990624,
+      "learning_rate": 4.109859597313237e-07,
+      "loss": 0.5199,
+      "step": 2505
+    },
+    {
+      "epoch": 2.758241758241758,
+      "grad_norm": 0.2031083447799231,
+      "learning_rate": 3.9304216147014853e-07,
+      "loss": 0.5058,
+      "step": 2510
+    },
+    {
+      "epoch": 2.7637362637362637,
+      "grad_norm": 0.20105232675523538,
+      "learning_rate": 3.7549102066328226e-07,
+      "loss": 0.512,
+      "step": 2515
+    },
+    {
+      "epoch": 2.769230769230769,
+      "grad_norm": 0.39484901944232603,
+      "learning_rate": 3.5833325466437697e-07,
+      "loss": 0.5012,
+      "step": 2520
+    },
+    {
+      "epoch": 2.7747252747252746,
+      "grad_norm": 0.19709914420778502,
+      "learning_rate": 3.4156956474898805e-07,
+      "loss": 0.5045,
+      "step": 2525
+    },
+    {
+      "epoch": 2.78021978021978,
+      "grad_norm": 0.20621710050936062,
+      "learning_rate": 3.2520063608592165e-07,
+      "loss": 0.5121,
+      "step": 2530
+    },
+    {
+      "epoch": 2.7857142857142856,
+      "grad_norm": 0.1991105530357979,
+      "learning_rate": 3.0922713770922155e-07,
+      "loss": 0.5203,
+      "step": 2535
+    },
+    {
+      "epoch": 2.791208791208791,
+      "grad_norm": 0.20012961338063315,
+      "learning_rate": 2.9364972249082747e-07,
+      "loss": 0.5184,
+      "step": 2540
+    },
+    {
+      "epoch": 2.7967032967032965,
+      "grad_norm": 0.19896717513784648,
+      "learning_rate": 2.7846902711389236e-07,
+      "loss": 0.5108,
+      "step": 2545
+    },
+    {
+      "epoch": 2.802197802197802,
+      "grad_norm": 0.1977548907547512,
+      "learning_rate": 2.636856720467573e-07,
+      "loss": 0.5208,
+      "step": 2550
+    },
+    {
+      "epoch": 2.8076923076923075,
+      "grad_norm": 0.20270402323570907,
+      "learning_rate": 2.493002615175977e-07,
+      "loss": 0.5006,
+      "step": 2555
+    },
+    {
+      "epoch": 2.813186813186813,
+      "grad_norm": 0.1942837253435945,
+      "learning_rate": 2.3531338348971366e-07,
+      "loss": 0.5119,
+      "step": 2560
+    },
+    {
+      "epoch": 2.8186813186813184,
+      "grad_norm": 0.20474339797776248,
+      "learning_rate": 2.217256096375131e-07,
+      "loss": 0.5176,
+      "step": 2565
+    },
+    {
+      "epoch": 2.824175824175824,
+      "grad_norm": 0.2010552477495884,
+      "learning_rate": 2.0853749532314006e-07,
+      "loss": 0.5073,
+      "step": 2570
+    },
+    {
+      "epoch": 2.82967032967033,
+      "grad_norm": 0.2035067000005909,
+      "learning_rate": 1.9574957957377294e-07,
+      "loss": 0.5096,
+      "step": 2575
+    },
+    {
+      "epoch": 2.8351648351648353,
+      "grad_norm": 0.20125633333431672,
+      "learning_rate": 1.8336238505959892e-07,
+      "loss": 0.5121,
+      "step": 2580
+    },
+    {
+      "epoch": 2.840659340659341,
+      "grad_norm": 0.20332908603793057,
+      "learning_rate": 1.7137641807244754e-07,
+      "loss": 0.5125,
+      "step": 2585
+    },
+    {
+      "epoch": 2.8461538461538463,
+      "grad_norm": 0.19665455030518153,
+      "learning_rate": 1.5979216850509848e-07,
+      "loss": 0.4987,
+      "step": 2590
+    },
+    {
+      "epoch": 2.8516483516483517,
+      "grad_norm": 0.1966270172986848,
+      "learning_rate": 1.4861010983126202e-07,
+      "loss": 0.4874,
+      "step": 2595
+    },
+    {
+      "epoch": 2.857142857142857,
+      "grad_norm": 0.19834917582035766,
+      "learning_rate": 1.3783069908621772e-07,
+      "loss": 0.5146,
+      "step": 2600
+    },
+    {
+      "epoch": 2.857142857142857,
+      "eval_loss": 0.6599460244178772,
+      "eval_runtime": 56.5256,
+      "eval_samples_per_second": 24.378,
+      "eval_steps_per_second": 0.106,
+      "step": 2600
+    },
+    {
+      "epoch": 2.8626373626373627,
+      "grad_norm": 0.19885509465905268,
+      "learning_rate": 1.274543768481451e-07,
+      "loss": 0.5192,
+      "step": 2605
+    },
+    {
+      "epoch": 2.868131868131868,
+      "grad_norm": 0.19541257662283346,
+      "learning_rate": 1.1748156722011128e-07,
+      "loss": 0.4926,
+      "step": 2610
+    },
+    {
+      "epoch": 2.8736263736263736,
+      "grad_norm": 0.19388744876156347,
+      "learning_rate": 1.0791267781273263e-07,
+      "loss": 0.5094,
+      "step": 2615
+    },
+    {
+      "epoch": 2.879120879120879,
+      "grad_norm": 0.1976970112287548,
+      "learning_rate": 9.874809972752697e-08,
+      "loss": 0.5156,
+      "step": 2620
+    },
+    {
+      "epoch": 2.8846153846153846,
+      "grad_norm": 0.20191971179188523,
+      "learning_rate": 8.99882075409153e-08,
+      "loss": 0.5176,
+      "step": 2625
+    },
+    {
+      "epoch": 2.89010989010989,
+      "grad_norm": 0.20009589256600918,
+      "learning_rate": 8.16333592889207e-08,
+      "loss": 0.5135,
+      "step": 2630
+    },
+    {
+      "epoch": 2.8956043956043955,
+      "grad_norm": 0.19825056208420425,
+      "learning_rate": 7.368389645252772e-08,
+      "loss": 0.5048,
+      "step": 2635
+    },
+    {
+      "epoch": 2.901098901098901,
+      "grad_norm": 0.20207333625508234,
+      "learning_rate": 6.61401439437348e-08,
+      "loss": 0.4971,
+      "step": 2640
+    },
+    {
+      "epoch": 2.9065934065934065,
+      "grad_norm": 0.20352522502880846,
+      "learning_rate": 5.9002410092262593e-08,
+      "loss": 0.5106,
+      "step": 2645
+    },
+    {
+      "epoch": 2.912087912087912,
+      "grad_norm": 0.20403786605478308,
+      "learning_rate": 5.227098663296404e-08,
+      "loss": 0.5229,
+      "step": 2650
+    },
+    {
+      "epoch": 2.9175824175824174,
+      "grad_norm": 0.2041806025949136,
+      "learning_rate": 4.594614869388947e-08,
+      "loss": 0.5118,
+      "step": 2655
+    },
+    {
+      "epoch": 2.9230769230769234,
+      "grad_norm": 0.20136601507382682,
+      "learning_rate": 4.002815478505007e-08,
+      "loss": 0.4982,
+      "step": 2660
+    },
+    {
+      "epoch": 2.928571428571429,
+      "grad_norm": 0.20662702533640964,
+      "learning_rate": 3.451724678784518e-08,
+      "loss": 0.5103,
+      "step": 2665
+    },
+    {
+      "epoch": 2.9340659340659343,
+      "grad_norm": 0.21824571202036225,
+      "learning_rate": 2.9413649945182475e-08,
+      "loss": 0.5227,
+      "step": 2670
+    },
+    {
+      "epoch": 2.9395604395604398,
+      "grad_norm": 0.20023301939180432,
+      "learning_rate": 2.47175728522675e-08,
+      "loss": 0.5141,
+      "step": 2675
+    },
+    {
+      "epoch": 2.9450549450549453,
+      "grad_norm": 0.20718821800921122,
+      "learning_rate": 2.0429207448078302e-08,
+      "loss": 0.5065,
+      "step": 2680
+    },
+    {
+      "epoch": 2.9505494505494507,
+      "grad_norm": 0.20188941781097472,
+      "learning_rate": 1.654872900752169e-08,
+      "loss": 0.5202,
+      "step": 2685
+    },
+    {
+      "epoch": 2.956043956043956,
+      "grad_norm": 0.20379828773126438,
+      "learning_rate": 1.3076296134271194e-08,
+      "loss": 0.5097,
+      "step": 2690
+    },
+    {
+      "epoch": 2.9615384615384617,
+      "grad_norm": 0.2043103342841997,
+      "learning_rate": 1.0012050754277802e-08,
+      "loss": 0.5127,
+      "step": 2695
+    },
+    {
+      "epoch": 2.967032967032967,
+      "grad_norm": 0.19873220074389136,
+      "learning_rate": 7.356118109977939e-09,
+      "loss": 0.5293,
+      "step": 2700
+    },
+    {
+      "epoch": 2.967032967032967,
+      "eval_loss": 0.6600582599639893,
+      "eval_runtime": 57.0547,
+      "eval_samples_per_second": 24.152,
+      "eval_steps_per_second": 0.105,
+      "step": 2700
+    },
+    {
+      "epoch": 2.9725274725274726,
+      "grad_norm": 0.20314150689373117,
+      "learning_rate": 5.108606755168666e-09,
+      "loss": 0.5146,
+      "step": 2705
+    },
+    {
+      "epoch": 2.978021978021978,
+      "grad_norm": 0.19470505277621175,
+      "learning_rate": 3.269608550571235e-09,
+      "loss": 0.4919,
+      "step": 2710
+    },
+    {
+      "epoch": 2.9835164835164836,
+      "grad_norm": 0.19203738679178015,
+      "learning_rate": 1.839198660079644e-09,
+      "loss": 0.5043,
+      "step": 2715
+    },
+    {
+      "epoch": 2.989010989010989,
+      "grad_norm": 0.20387845762850706,
+      "learning_rate": 8.174355476864293e-10,
+      "loss": 0.5199,
+      "step": 2720
+    },
+    {
+      "epoch": 2.9945054945054945,
+      "grad_norm": 0.19811095343066756,
+      "learning_rate": 2.0436097509235475e-10,
+      "loss": 0.5033,
+      "step": 2725
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.1948663079714202,
+      "learning_rate": 0.0,
+      "loss": 0.4979,
+      "step": 2730
+    },
+    {
+      "epoch": 3.0,
+      "step": 2730,
+      "total_flos": 1.183665069490176e+16,
+      "train_loss": 0.5413940727492392,
+      "train_runtime": 63958.3461,
+      "train_samples_per_second": 5.459,
+      "train_steps_per_second": 0.043
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 2730,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.183665069490176e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}