diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5199 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9999279383151978,
+  "eval_steps": 100,
+  "global_step": 3469,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0014412336960438136,
+      "grad_norm": 15.083786010742188,
+      "learning_rate": 2.8735632183908047e-07,
+      "loss": 2.483,
+      "step": 5
+    },
+    {
+      "epoch": 0.0028824673920876272,
+      "grad_norm": 14.648148536682129,
+      "learning_rate": 5.747126436781609e-07,
+      "loss": 2.217,
+      "step": 10
+    },
+    {
+      "epoch": 0.004323701088131441,
+      "grad_norm": 15.199370384216309,
+      "learning_rate": 8.620689655172415e-07,
+      "loss": 2.2391,
+      "step": 15
+    },
+    {
+      "epoch": 0.0057649347841752544,
+      "grad_norm": 15.044349670410156,
+      "learning_rate": 1.1494252873563219e-06,
+      "loss": 2.4945,
+      "step": 20
+    },
+    {
+      "epoch": 0.007206168480219067,
+      "grad_norm": 14.505260467529297,
+      "learning_rate": 1.4367816091954023e-06,
+      "loss": 2.2728,
+      "step": 25
+    },
+    {
+      "epoch": 0.008647402176262881,
+      "grad_norm": 14.381186485290527,
+      "learning_rate": 1.724137931034483e-06,
+      "loss": 2.1199,
+      "step": 30
+    },
+    {
+      "epoch": 0.010088635872306694,
+      "grad_norm": 12.466582298278809,
+      "learning_rate": 2.0114942528735633e-06,
+      "loss": 1.8471,
+      "step": 35
+    },
+    {
+      "epoch": 0.011529869568350509,
+      "grad_norm": 13.661079406738281,
+      "learning_rate": 2.2988505747126437e-06,
+      "loss": 1.7115,
+      "step": 40
+    },
+    {
+      "epoch": 0.012971103264394322,
+      "grad_norm": 13.910940170288086,
+      "learning_rate": 2.5862068965517246e-06,
+      "loss": 1.7219,
+      "step": 45
+    },
+    {
+      "epoch": 0.014412336960438135,
+      "grad_norm": 14.404576301574707,
+      "learning_rate": 2.8735632183908046e-06,
+      "loss": 1.6988,
+      "step": 50
+    },
+    {
+      "epoch": 0.01585357065648195,
+      "grad_norm": 13.622361183166504,
+      "learning_rate": 3.1609195402298854e-06,
+      "loss": 1.1913,
+      "step": 55
+    },
+    {
+      "epoch": 0.017294804352525762,
+      "grad_norm": 4.847095966339111,
+      "learning_rate": 3.448275862068966e-06,
+      "loss": 0.9125,
+      "step": 60
+    },
+    {
+      "epoch": 0.018736038048569575,
+      "grad_norm": 5.12880277633667,
+      "learning_rate": 3.7356321839080462e-06,
+      "loss": 0.8373,
+      "step": 65
+    },
+    {
+      "epoch": 0.02017727174461339,
+      "grad_norm": 1.7816091775894165,
+      "learning_rate": 4.022988505747127e-06,
+      "loss": 0.6926,
+      "step": 70
+    },
+    {
+      "epoch": 0.0216185054406572,
+      "grad_norm": 2.607332706451416,
+      "learning_rate": 4.310344827586207e-06,
+      "loss": 0.6349,
+      "step": 75
+    },
+    {
+      "epoch": 0.023059739136701018,
+      "grad_norm": 1.3011763095855713,
+      "learning_rate": 4.5977011494252875e-06,
+      "loss": 0.557,
+      "step": 80
+    },
+    {
+      "epoch": 0.02450097283274483,
+      "grad_norm": 1.5208930969238281,
+      "learning_rate": 4.885057471264369e-06,
+      "loss": 0.5506,
+      "step": 85
+    },
+    {
+      "epoch": 0.025942206528788644,
+      "grad_norm": 2.021416425704956,
+      "learning_rate": 5.172413793103449e-06,
+      "loss": 0.6639,
+      "step": 90
+    },
+    {
+      "epoch": 0.027383440224832457,
+      "grad_norm": 1.7699145078659058,
+      "learning_rate": 5.45977011494253e-06,
+      "loss": 0.4614,
+      "step": 95
+    },
+    {
+      "epoch": 0.02882467392087627,
+      "grad_norm": 4.601930141448975,
+      "learning_rate": 5.747126436781609e-06,
+      "loss": 0.5361,
+      "step": 100
+    },
+    {
+      "epoch": 0.02882467392087627,
+      "eval_loss": 0.49337437748908997,
+      "eval_mse": 0.49337437438964843,
+      "eval_runtime": 3.4708,
+      "eval_samples_per_second": 288.122,
+      "eval_steps_per_second": 18.152,
+      "step": 100
+    },
+    {
+      "epoch": 0.030265907616920083,
+      "grad_norm": 4.816018581390381,
+      "learning_rate": 6.03448275862069e-06,
+      "loss": 0.5876,
+      "step": 105
+    },
+    {
+      "epoch": 0.0317071413129639,
+      "grad_norm": 4.016970157623291,
+      "learning_rate": 6.321839080459771e-06,
+      "loss": 0.4752,
+      "step": 110
+    },
+    {
+      "epoch": 0.03314837500900771,
+      "grad_norm": 2.397547483444214,
+      "learning_rate": 6.609195402298851e-06,
+      "loss": 0.5025,
+      "step": 115
+    },
+    {
+      "epoch": 0.034589608705051525,
+      "grad_norm": 2.304940938949585,
+      "learning_rate": 6.896551724137932e-06,
+      "loss": 0.4742,
+      "step": 120
+    },
+    {
+      "epoch": 0.03603084240109534,
+      "grad_norm": 4.651687145233154,
+      "learning_rate": 7.183908045977011e-06,
+      "loss": 0.4382,
+      "step": 125
+    },
+    {
+      "epoch": 0.03747207609713915,
+      "grad_norm": 2.2445216178894043,
+      "learning_rate": 7.4712643678160925e-06,
+      "loss": 0.3867,
+      "step": 130
+    },
+    {
+      "epoch": 0.038913309793182964,
+      "grad_norm": 2.322697877883911,
+      "learning_rate": 7.758620689655173e-06,
+      "loss": 0.4629,
+      "step": 135
+    },
+    {
+      "epoch": 0.04035454348922678,
+      "grad_norm": 2.4827287197113037,
+      "learning_rate": 8.045977011494253e-06,
+      "loss": 0.3633,
+      "step": 140
+    },
+    {
+      "epoch": 0.04179577718527059,
+      "grad_norm": 3.959317445755005,
+      "learning_rate": 8.333333333333334e-06,
+      "loss": 0.3743,
+      "step": 145
+    },
+    {
+      "epoch": 0.0432370108813144,
+      "grad_norm": 2.0469164848327637,
+      "learning_rate": 8.620689655172414e-06,
+      "loss": 0.3836,
+      "step": 150
+    },
+    {
+      "epoch": 0.044678244577358216,
+      "grad_norm": 2.5716373920440674,
+      "learning_rate": 8.908045977011495e-06,
+      "loss": 0.3875,
+      "step": 155
+    },
+    {
+      "epoch": 0.046119478273402036,
+      "grad_norm": 6.04033088684082,
+      "learning_rate": 9.195402298850575e-06,
+      "loss": 0.3584,
+      "step": 160
+    },
+    {
+      "epoch": 0.04756071196944585,
+      "grad_norm": 10.341621398925781,
+      "learning_rate": 9.482758620689655e-06,
+      "loss": 0.3824,
+      "step": 165
+    },
+    {
+      "epoch": 0.04900194566548966,
+      "grad_norm": 3.144866704940796,
+      "learning_rate": 9.770114942528738e-06,
+      "loss": 0.3281,
+      "step": 170
+    },
+    {
+      "epoch": 0.050443179361533474,
+      "grad_norm": 5.703673362731934,
+      "learning_rate": 9.996965098634295e-06,
+      "loss": 0.354,
+      "step": 175
+    },
+    {
+      "epoch": 0.05188441305757729,
+      "grad_norm": 2.0394303798675537,
+      "learning_rate": 9.981790591805767e-06,
+      "loss": 0.3734,
+      "step": 180
+    },
+    {
+      "epoch": 0.0533256467536211,
+      "grad_norm": 3.995746374130249,
+      "learning_rate": 9.966616084977238e-06,
+      "loss": 0.3225,
+      "step": 185
+    },
+    {
+      "epoch": 0.05476688044966491,
+      "grad_norm": 8.836956977844238,
+      "learning_rate": 9.951441578148711e-06,
+      "loss": 0.3439,
+      "step": 190
+    },
+    {
+      "epoch": 0.056208114145708726,
+      "grad_norm": 2.608346939086914,
+      "learning_rate": 9.936267071320182e-06,
+      "loss": 0.371,
+      "step": 195
+    },
+    {
+      "epoch": 0.05764934784175254,
+      "grad_norm": 4.031014442443848,
+      "learning_rate": 9.921092564491654e-06,
+      "loss": 0.3483,
+      "step": 200
+    },
+    {
+      "epoch": 0.05764934784175254,
+      "eval_loss": 0.352454274892807,
+      "eval_mse": 0.3524542865753174,
+      "eval_runtime": 3.6486,
+      "eval_samples_per_second": 274.078,
+      "eval_steps_per_second": 17.267,
+      "step": 200
+    },
+    {
+      "epoch": 0.05909058153779635,
+      "grad_norm": 4.824347972869873,
+      "learning_rate": 9.905918057663127e-06,
+      "loss": 0.3258,
+      "step": 205
+    },
+    {
+      "epoch": 0.060531815233840165,
+      "grad_norm": 2.2714250087738037,
+      "learning_rate": 9.890743550834598e-06,
+      "loss": 0.3321,
+      "step": 210
+    },
+    {
+      "epoch": 0.06197304892988398,
+      "grad_norm": 5.385800361633301,
+      "learning_rate": 9.87556904400607e-06,
+      "loss": 0.2992,
+      "step": 215
+    },
+    {
+      "epoch": 0.0634142826259278,
+      "grad_norm": 2.349539279937744,
+      "learning_rate": 9.860394537177543e-06,
+      "loss": 0.3111,
+      "step": 220
+    },
+    {
+      "epoch": 0.0648555163219716,
+      "grad_norm": 3.6904749870300293,
+      "learning_rate": 9.845220030349014e-06,
+      "loss": 0.3812,
+      "step": 225
+    },
+    {
+      "epoch": 0.06629675001801542,
+      "grad_norm": 3.031271457672119,
+      "learning_rate": 9.830045523520486e-06,
+      "loss": 0.2955,
+      "step": 230
+    },
+    {
+      "epoch": 0.06773798371405923,
+      "grad_norm": 5.256127834320068,
+      "learning_rate": 9.814871016691959e-06,
+      "loss": 0.3384,
+      "step": 235
+    },
+    {
+      "epoch": 0.06917921741010305,
+      "grad_norm": 3.7872514724731445,
+      "learning_rate": 9.79969650986343e-06,
+      "loss": 0.3169,
+      "step": 240
+    },
+    {
+      "epoch": 0.07062045110614686,
+      "grad_norm": 2.257871150970459,
+      "learning_rate": 9.784522003034902e-06,
+      "loss": 0.3045,
+      "step": 245
+    },
+    {
+      "epoch": 0.07206168480219068,
+      "grad_norm": 3.1732468605041504,
+      "learning_rate": 9.769347496206375e-06,
+      "loss": 0.2859,
+      "step": 250
+    },
+    {
+      "epoch": 0.07350291849823448,
+      "grad_norm": 1.8080546855926514,
+      "learning_rate": 9.754172989377846e-06,
+      "loss": 0.2752,
+      "step": 255
+    },
+    {
+      "epoch": 0.0749441521942783,
+      "grad_norm": 3.3401401042938232,
+      "learning_rate": 9.738998482549317e-06,
+      "loss": 0.3189,
+      "step": 260
+    },
+    {
+      "epoch": 0.07638538589032212,
+      "grad_norm": 4.685635089874268,
+      "learning_rate": 9.72382397572079e-06,
+      "loss": 0.3365,
+      "step": 265
+    },
+    {
+      "epoch": 0.07782661958636593,
+      "grad_norm": 3.807539224624634,
+      "learning_rate": 9.708649468892262e-06,
+      "loss": 0.3306,
+      "step": 270
+    },
+    {
+      "epoch": 0.07926785328240975,
+      "grad_norm": 2.541351795196533,
+      "learning_rate": 9.693474962063733e-06,
+      "loss": 0.3665,
+      "step": 275
+    },
+    {
+      "epoch": 0.08070908697845355,
+      "grad_norm": 2.3847010135650635,
+      "learning_rate": 9.678300455235205e-06,
+      "loss": 0.3359,
+      "step": 280
+    },
+    {
+      "epoch": 0.08215032067449737,
+      "grad_norm": 9.623117446899414,
+      "learning_rate": 9.663125948406678e-06,
+      "loss": 0.3699,
+      "step": 285
+    },
+    {
+      "epoch": 0.08359155437054118,
+      "grad_norm": 9.328727722167969,
+      "learning_rate": 9.64795144157815e-06,
+      "loss": 0.3608,
+      "step": 290
+    },
+    {
+      "epoch": 0.085032788066585,
+      "grad_norm": 6.9004435539245605,
+      "learning_rate": 9.63277693474962e-06,
+      "loss": 0.338,
+      "step": 295
+    },
+    {
+      "epoch": 0.0864740217626288,
+      "grad_norm": 5.4706830978393555,
+      "learning_rate": 9.617602427921094e-06,
+      "loss": 0.3238,
+      "step": 300
+    },
+    {
+      "epoch": 0.0864740217626288,
+      "eval_loss": 0.29314419627189636,
+      "eval_mse": 0.2931441988945007,
+      "eval_runtime": 3.5222,
+      "eval_samples_per_second": 283.916,
+      "eval_steps_per_second": 17.887,
+      "step": 300
+    },
+    {
+      "epoch": 0.08791525545867263,
+      "grad_norm": 2.093715190887451,
+      "learning_rate": 9.602427921092565e-06,
+      "loss": 0.3098,
+      "step": 305
+    },
+    {
+      "epoch": 0.08935648915471643,
+      "grad_norm": 2.114649534225464,
+      "learning_rate": 9.587253414264037e-06,
+      "loss": 0.2742,
+      "step": 310
+    },
+    {
+      "epoch": 0.09079772285076025,
+      "grad_norm": 7.1367387771606445,
+      "learning_rate": 9.57207890743551e-06,
+      "loss": 0.3443,
+      "step": 315
+    },
+    {
+      "epoch": 0.09223895654680407,
+      "grad_norm": 3.2078042030334473,
+      "learning_rate": 9.556904400606981e-06,
+      "loss": 0.2965,
+      "step": 320
+    },
+    {
+      "epoch": 0.09368019024284788,
+      "grad_norm": 4.233352184295654,
+      "learning_rate": 9.541729893778452e-06,
+      "loss": 0.2553,
+      "step": 325
+    },
+    {
+      "epoch": 0.0951214239388917,
+      "grad_norm": 8.637604713439941,
+      "learning_rate": 9.526555386949926e-06,
+      "loss": 0.3027,
+      "step": 330
+    },
+    {
+      "epoch": 0.0965626576349355,
+      "grad_norm": 2.668332576751709,
+      "learning_rate": 9.511380880121397e-06,
+      "loss": 0.3195,
+      "step": 335
+    },
+    {
+      "epoch": 0.09800389133097932,
+      "grad_norm": 5.594767093658447,
+      "learning_rate": 9.496206373292868e-06,
+      "loss": 0.3111,
+      "step": 340
+    },
+    {
+      "epoch": 0.09944512502702313,
+      "grad_norm": 1.80435049533844,
+      "learning_rate": 9.481031866464341e-06,
+      "loss": 0.2833,
+      "step": 345
+    },
+    {
+      "epoch": 0.10088635872306695,
+      "grad_norm": 6.785181045532227,
+      "learning_rate": 9.465857359635813e-06,
+      "loss": 0.3184,
+      "step": 350
+    },
+    {
+      "epoch": 0.10232759241911075,
+      "grad_norm": 7.8451008796691895,
+      "learning_rate": 9.450682852807284e-06,
+      "loss": 0.3031,
+      "step": 355
+    },
+    {
+      "epoch": 0.10376882611515457,
+      "grad_norm": 2.5645573139190674,
+      "learning_rate": 9.435508345978757e-06,
+      "loss": 0.2877,
+      "step": 360
+    },
+    {
+      "epoch": 0.10521005981119838,
+      "grad_norm": 2.994230031967163,
+      "learning_rate": 9.420333839150229e-06,
+      "loss": 0.3102,
+      "step": 365
+    },
+    {
+      "epoch": 0.1066512935072422,
+      "grad_norm": 5.640050411224365,
+      "learning_rate": 9.4051593323217e-06,
+      "loss": 0.293,
+      "step": 370
+    },
+    {
+      "epoch": 0.108092527203286,
+      "grad_norm": 7.134743690490723,
+      "learning_rate": 9.389984825493173e-06,
+      "loss": 0.3103,
+      "step": 375
+    },
+    {
+      "epoch": 0.10953376089932983,
+      "grad_norm": 3.617605209350586,
+      "learning_rate": 9.374810318664645e-06,
+      "loss": 0.2883,
+      "step": 380
+    },
+    {
+      "epoch": 0.11097499459537365,
+      "grad_norm": 1.7953038215637207,
+      "learning_rate": 9.359635811836116e-06,
+      "loss": 0.2946,
+      "step": 385
+    },
+    {
+      "epoch": 0.11241622829141745,
+      "grad_norm": 4.107087135314941,
+      "learning_rate": 9.344461305007587e-06,
+      "loss": 0.3091,
+      "step": 390
+    },
+    {
+      "epoch": 0.11385746198746127,
+      "grad_norm": 2.5875136852264404,
+      "learning_rate": 9.32928679817906e-06,
+      "loss": 0.2776,
+      "step": 395
+    },
+    {
+      "epoch": 0.11529869568350508,
+      "grad_norm": 2.0523602962493896,
+      "learning_rate": 9.314112291350532e-06,
+      "loss": 0.2734,
+      "step": 400
+    },
+    {
+      "epoch": 0.11529869568350508,
+      "eval_loss": 0.31296026706695557,
+      "eval_mse": 0.31296028327941894,
+      "eval_runtime": 3.4119,
+      "eval_samples_per_second": 293.091,
+      "eval_steps_per_second": 18.465,
+      "step": 400
+    },
+    {
+      "epoch": 0.1167399293795489,
+      "grad_norm": 2.2445826530456543,
+      "learning_rate": 9.298937784522003e-06,
+      "loss": 0.3148,
+      "step": 405
+    },
+    {
+      "epoch": 0.1181811630755927,
+      "grad_norm": 3.3342812061309814,
+      "learning_rate": 9.283763277693477e-06,
+      "loss": 0.2743,
+      "step": 410
+    },
+    {
+      "epoch": 0.11962239677163652,
+      "grad_norm": 2.5152697563171387,
+      "learning_rate": 9.268588770864948e-06,
+      "loss": 0.3092,
+      "step": 415
+    },
+    {
+      "epoch": 0.12106363046768033,
+      "grad_norm": 2.3605024814605713,
+      "learning_rate": 9.25341426403642e-06,
+      "loss": 0.3348,
+      "step": 420
+    },
+    {
+      "epoch": 0.12250486416372415,
+      "grad_norm": 4.979254245758057,
+      "learning_rate": 9.238239757207892e-06,
+      "loss": 0.3268,
+      "step": 425
+    },
+    {
+      "epoch": 0.12394609785976796,
+      "grad_norm": 2.3477885723114014,
+      "learning_rate": 9.223065250379364e-06,
+      "loss": 0.309,
+      "step": 430
+    },
+    {
+      "epoch": 0.12538733155581178,
+      "grad_norm": 2.372631549835205,
+      "learning_rate": 9.207890743550835e-06,
+      "loss": 0.2893,
+      "step": 435
+    },
+    {
+      "epoch": 0.1268285652518556,
+      "grad_norm": 4.252530097961426,
+      "learning_rate": 9.192716236722308e-06,
+      "loss": 0.2758,
+      "step": 440
+    },
+    {
+      "epoch": 0.1282697989478994,
+      "grad_norm": 5.626593112945557,
+      "learning_rate": 9.17754172989378e-06,
+      "loss": 0.2904,
+      "step": 445
+    },
+    {
+      "epoch": 0.1297110326439432,
+      "grad_norm": 6.885512351989746,
+      "learning_rate": 9.162367223065251e-06,
+      "loss": 0.2694,
+      "step": 450
+    },
+    {
+      "epoch": 0.13115226633998703,
+      "grad_norm": 3.3293423652648926,
+      "learning_rate": 9.147192716236724e-06,
+      "loss": 0.2826,
+      "step": 455
+    },
+    {
+      "epoch": 0.13259350003603085,
+      "grad_norm": 2.934630870819092,
+      "learning_rate": 9.132018209408196e-06,
+      "loss": 0.3319,
+      "step": 460
+    },
+    {
+      "epoch": 0.13403473373207467,
+      "grad_norm": 4.578772068023682,
+      "learning_rate": 9.116843702579667e-06,
+      "loss": 0.34,
+      "step": 465
+    },
+    {
+      "epoch": 0.13547596742811846,
+      "grad_norm": 2.9086310863494873,
+      "learning_rate": 9.10166919575114e-06,
+      "loss": 0.2912,
+      "step": 470
+    },
+    {
+      "epoch": 0.13691720112416228,
+      "grad_norm": 2.20721697807312,
+      "learning_rate": 9.08649468892261e-06,
+      "loss": 0.2737,
+      "step": 475
+    },
+    {
+      "epoch": 0.1383584348202061,
+      "grad_norm": 2.334871530532837,
+      "learning_rate": 9.071320182094083e-06,
+      "loss": 0.3058,
+      "step": 480
+    },
+    {
+      "epoch": 0.13979966851624992,
+      "grad_norm": 4.9962592124938965,
+      "learning_rate": 9.056145675265554e-06,
+      "loss": 0.2907,
+      "step": 485
+    },
+    {
+      "epoch": 0.1412409022122937,
+      "grad_norm": 5.542920112609863,
+      "learning_rate": 9.040971168437026e-06,
+      "loss": 0.3144,
+      "step": 490
+    },
+    {
+      "epoch": 0.14268213590833753,
+      "grad_norm": 2.817955255508423,
+      "learning_rate": 9.025796661608497e-06,
+      "loss": 0.2554,
+      "step": 495
+    },
+    {
+      "epoch": 0.14412336960438135,
+      "grad_norm": 3.7314751148223877,
+      "learning_rate": 9.01062215477997e-06,
+      "loss": 0.2891,
+      "step": 500
+    },
+    {
+      "epoch": 0.14412336960438135,
+      "eval_loss": 0.3298364281654358,
+      "eval_mse": 0.32983644971996545,
+      "eval_runtime": 3.402,
+      "eval_samples_per_second": 293.943,
+      "eval_steps_per_second": 18.518,
+      "step": 500
+    },
+    {
+      "epoch": 0.14556460330042517,
+      "grad_norm": 4.2389445304870605,
+      "learning_rate": 8.995447647951442e-06,
+      "loss": 0.2957,
+      "step": 505
+    },
+    {
+      "epoch": 0.14700583699646896,
+      "grad_norm": 4.599143981933594,
+      "learning_rate": 8.980273141122913e-06,
+      "loss": 0.3008,
+      "step": 510
+    },
+    {
+      "epoch": 0.14844707069251278,
+      "grad_norm": 7.0564751625061035,
+      "learning_rate": 8.965098634294386e-06,
+      "loss": 0.2908,
+      "step": 515
+    },
+    {
+      "epoch": 0.1498883043885566,
+      "grad_norm": 1.786997675895691,
+      "learning_rate": 8.949924127465858e-06,
+      "loss": 0.276,
+      "step": 520
+    },
+    {
+      "epoch": 0.15132953808460042,
+      "grad_norm": 3.3596365451812744,
+      "learning_rate": 8.934749620637329e-06,
+      "loss": 0.3179,
+      "step": 525
+    },
+    {
+      "epoch": 0.15277077178064424,
+      "grad_norm": 4.094123363494873,
+      "learning_rate": 8.919575113808802e-06,
+      "loss": 0.2689,
+      "step": 530
+    },
+    {
+      "epoch": 0.15421200547668804,
+      "grad_norm": 1.8090394735336304,
+      "learning_rate": 8.904400606980273e-06,
+      "loss": 0.266,
+      "step": 535
+    },
+    {
+      "epoch": 0.15565323917273186,
+      "grad_norm": 2.1146230697631836,
+      "learning_rate": 8.889226100151745e-06,
+      "loss": 0.2918,
+      "step": 540
+    },
+    {
+      "epoch": 0.15709447286877568,
+      "grad_norm": 2.7003800868988037,
+      "learning_rate": 8.874051593323218e-06,
+      "loss": 0.3014,
+      "step": 545
+    },
+    {
+      "epoch": 0.1585357065648195,
+      "grad_norm": 2.025775909423828,
+      "learning_rate": 8.85887708649469e-06,
+      "loss": 0.2933,
+      "step": 550
+    },
+    {
+      "epoch": 0.1599769402608633,
+      "grad_norm": 3.2832581996917725,
+      "learning_rate": 8.84370257966616e-06,
+      "loss": 0.2697,
+      "step": 555
+    },
+    {
+      "epoch": 0.1614181739569071,
+      "grad_norm": 4.607889175415039,
+      "learning_rate": 8.828528072837634e-06,
+      "loss": 0.32,
+      "step": 560
+    },
+    {
+      "epoch": 0.16285940765295093,
+      "grad_norm": 7.916262149810791,
+      "learning_rate": 8.813353566009105e-06,
+      "loss": 0.3381,
+      "step": 565
+    },
+    {
+      "epoch": 0.16430064134899475,
+      "grad_norm": 1.8572163581848145,
+      "learning_rate": 8.798179059180577e-06,
+      "loss": 0.2642,
+      "step": 570
+    },
+    {
+      "epoch": 0.16574187504503854,
+      "grad_norm": 1.9104336500167847,
+      "learning_rate": 8.78300455235205e-06,
+      "loss": 0.3111,
+      "step": 575
+    },
+    {
+      "epoch": 0.16718310874108236,
+      "grad_norm": 5.321012020111084,
+      "learning_rate": 8.767830045523521e-06,
+      "loss": 0.2776,
+      "step": 580
+    },
+    {
+      "epoch": 0.16862434243712618,
+      "grad_norm": 6.43941593170166,
+      "learning_rate": 8.752655538694993e-06,
+      "loss": 0.3395,
+      "step": 585
+    },
+    {
+      "epoch": 0.17006557613317,
+      "grad_norm": 4.0543646812438965,
+      "learning_rate": 8.737481031866466e-06,
+      "loss": 0.2403,
+      "step": 590
+    },
+    {
+      "epoch": 0.17150680982921382,
+      "grad_norm": 3.5150725841522217,
+      "learning_rate": 8.722306525037937e-06,
+      "loss": 0.2943,
+      "step": 595
+    },
+    {
+      "epoch": 0.1729480435252576,
+      "grad_norm": 4.989763259887695,
+      "learning_rate": 8.707132018209408e-06,
+      "loss": 0.2807,
+      "step": 600
+    },
+    {
+      "epoch": 0.1729480435252576,
+      "eval_loss": 0.26590582728385925,
+      "eval_mse": 0.2659058196991682,
+      "eval_runtime": 3.4384,
+      "eval_samples_per_second": 290.832,
+      "eval_steps_per_second": 18.322,
+      "step": 600
+    },
+    {
+      "epoch": 0.17438927722130143,
+      "grad_norm": 2.0770037174224854,
+      "learning_rate": 8.69195751138088e-06,
+      "loss": 0.2716,
+      "step": 605
+    },
+    {
+      "epoch": 0.17583051091734525,
+      "grad_norm": 3.540919065475464,
+      "learning_rate": 8.676783004552353e-06,
+      "loss": 0.2697,
+      "step": 610
+    },
+    {
+      "epoch": 0.17727174461338907,
+      "grad_norm": 4.88574743270874,
+      "learning_rate": 8.661608497723824e-06,
+      "loss": 0.2922,
+      "step": 615
+    },
+    {
+      "epoch": 0.17871297830943286,
+      "grad_norm": 2.6014974117279053,
+      "learning_rate": 8.646433990895296e-06,
+      "loss": 0.2942,
+      "step": 620
+    },
+    {
+      "epoch": 0.18015421200547668,
+      "grad_norm": 4.289388179779053,
+      "learning_rate": 8.631259484066769e-06,
+      "loss": 0.2707,
+      "step": 625
+    },
+    {
+      "epoch": 0.1815954457015205,
+      "grad_norm": 2.1737794876098633,
+      "learning_rate": 8.61608497723824e-06,
+      "loss": 0.2716,
+      "step": 630
+    },
+    {
+      "epoch": 0.18303667939756432,
+      "grad_norm": 3.533886432647705,
+      "learning_rate": 8.600910470409712e-06,
+      "loss": 0.2821,
+      "step": 635
+    },
+    {
+      "epoch": 0.18447791309360814,
+      "grad_norm": 3.032208204269409,
+      "learning_rate": 8.585735963581185e-06,
+      "loss": 0.2641,
+      "step": 640
+    },
+    {
+      "epoch": 0.18591914678965193,
+      "grad_norm": 1.685655951499939,
+      "learning_rate": 8.570561456752656e-06,
+      "loss": 0.3315,
+      "step": 645
+    },
+    {
+      "epoch": 0.18736038048569575,
+      "grad_norm": 1.627114176750183,
+      "learning_rate": 8.555386949924128e-06,
+      "loss": 0.2782,
+      "step": 650
+    },
+    {
+      "epoch": 0.18880161418173957,
+      "grad_norm": 3.437077760696411,
+      "learning_rate": 8.5402124430956e-06,
+      "loss": 0.2733,
+      "step": 655
+    },
+    {
+      "epoch": 0.1902428478777834,
+      "grad_norm": 3.599497079849243,
+      "learning_rate": 8.525037936267072e-06,
+      "loss": 0.2509,
+      "step": 660
+    },
+    {
+      "epoch": 0.1916840815738272,
+      "grad_norm": 2.423408269882202,
+      "learning_rate": 8.509863429438544e-06,
+      "loss": 0.2672,
+      "step": 665
+    },
+    {
+      "epoch": 0.193125315269871,
+      "grad_norm": 3.0533182621002197,
+      "learning_rate": 8.494688922610017e-06,
+      "loss": 0.2876,
+      "step": 670
+    },
+    {
+      "epoch": 0.19456654896591483,
+      "grad_norm": 2.392457962036133,
+      "learning_rate": 8.479514415781488e-06,
+      "loss": 0.254,
+      "step": 675
+    },
+    {
+      "epoch": 0.19600778266195865,
+      "grad_norm": 1.6407463550567627,
+      "learning_rate": 8.46433990895296e-06,
+      "loss": 0.2659,
+      "step": 680
+    },
+    {
+      "epoch": 0.19744901635800244,
+      "grad_norm": 2.979224443435669,
+      "learning_rate": 8.449165402124433e-06,
+      "loss": 0.2974,
+      "step": 685
+    },
+    {
+      "epoch": 0.19889025005404626,
+      "grad_norm": 5.143660545349121,
+      "learning_rate": 8.433990895295904e-06,
+      "loss": 0.2529,
+      "step": 690
+    },
+    {
+      "epoch": 0.20033148375009008,
+      "grad_norm": 2.1899147033691406,
+      "learning_rate": 8.418816388467375e-06,
+      "loss": 0.2532,
+      "step": 695
+    },
+    {
+      "epoch": 0.2017727174461339,
+      "grad_norm": 3.272409677505493,
+      "learning_rate": 8.403641881638848e-06,
+      "loss": 0.2727,
+      "step": 700
+    },
+    {
+      "epoch": 0.2017727174461339,
+      "eval_loss": 0.26900768280029297,
+      "eval_mse": 0.2690076799674425,
+      "eval_runtime": 3.5072,
+      "eval_samples_per_second": 285.129,
+      "eval_steps_per_second": 17.963,
+      "step": 700
+    },
+    {
+      "epoch": 0.20321395114217772,
+      "grad_norm": 3.7777695655822754,
+      "learning_rate": 8.38846737481032e-06,
+      "loss": 0.2701,
+      "step": 705
+    },
+    {
+      "epoch": 0.2046551848382215,
+      "grad_norm": 2.7959799766540527,
+      "learning_rate": 8.373292867981791e-06,
+      "loss": 0.2697,
+      "step": 710
+    },
+    {
+      "epoch": 0.20609641853426533,
+      "grad_norm": 2.1683778762817383,
+      "learning_rate": 8.358118361153263e-06,
+      "loss": 0.2919,
+      "step": 715
+    },
+    {
+      "epoch": 0.20753765223030915,
+      "grad_norm": 4.523341655731201,
+      "learning_rate": 8.342943854324736e-06,
+      "loss": 0.291,
+      "step": 720
+    },
+    {
+      "epoch": 0.20897888592635297,
+      "grad_norm": 3.7949328422546387,
+      "learning_rate": 8.327769347496207e-06,
+      "loss": 0.2673,
+      "step": 725
+    },
+    {
+      "epoch": 0.21042011962239676,
+      "grad_norm": 2.369877815246582,
+      "learning_rate": 8.312594840667679e-06,
+      "loss": 0.3002,
+      "step": 730
+    },
+    {
+      "epoch": 0.21186135331844058,
+      "grad_norm": 3.621488094329834,
+      "learning_rate": 8.297420333839152e-06,
+      "loss": 0.2715,
+      "step": 735
+    },
+    {
+      "epoch": 0.2133025870144844,
+      "grad_norm": 6.6333394050598145,
+      "learning_rate": 8.282245827010623e-06,
+      "loss": 0.3091,
+      "step": 740
+    },
+    {
+      "epoch": 0.21474382071052822,
+      "grad_norm": 4.557513236999512,
+      "learning_rate": 8.267071320182094e-06,
+      "loss": 0.3085,
+      "step": 745
+    },
+    {
+      "epoch": 0.216185054406572,
+      "grad_norm": 1.754773497581482,
+      "learning_rate": 8.251896813353568e-06,
+      "loss": 0.2687,
+      "step": 750
+    },
+    {
+      "epoch": 0.21762628810261583,
+      "grad_norm": 1.8777259588241577,
+      "learning_rate": 8.236722306525039e-06,
+      "loss": 0.261,
+      "step": 755
+    },
+    {
+      "epoch": 0.21906752179865965,
+      "grad_norm": 1.8965113162994385,
+      "learning_rate": 8.22154779969651e-06,
+      "loss": 0.2664,
+      "step": 760
+    },
+    {
+      "epoch": 0.22050875549470347,
+      "grad_norm": 1.9722626209259033,
+      "learning_rate": 8.206373292867983e-06,
+      "loss": 0.2866,
+      "step": 765
+    },
+    {
+      "epoch": 0.2219499891907473,
+      "grad_norm": 2.329744815826416,
+      "learning_rate": 8.191198786039455e-06,
+      "loss": 0.2818,
+      "step": 770
+    },
+    {
+      "epoch": 0.22339122288679109,
+      "grad_norm": 5.612697601318359,
+      "learning_rate": 8.176024279210926e-06,
+      "loss": 0.2852,
+      "step": 775
+    },
+    {
+      "epoch": 0.2248324565828349,
+      "grad_norm": 1.9772837162017822,
+      "learning_rate": 8.1608497723824e-06,
+      "loss": 0.2627,
+      "step": 780
+    },
+    {
+      "epoch": 0.22627369027887873,
+      "grad_norm": 1.7957868576049805,
+      "learning_rate": 8.14567526555387e-06,
+      "loss": 0.2903,
+      "step": 785
+    },
+    {
+      "epoch": 0.22771492397492255,
+      "grad_norm": 1.7506916522979736,
+      "learning_rate": 8.130500758725342e-06,
+      "loss": 0.2651,
+      "step": 790
+    },
+    {
+      "epoch": 0.22915615767096634,
+      "grad_norm": 4.4279046058654785,
+      "learning_rate": 8.115326251896815e-06,
+      "loss": 0.3138,
+      "step": 795
+    },
+    {
+      "epoch": 0.23059739136701016,
+      "grad_norm": 2.2973923683166504,
+      "learning_rate": 8.100151745068287e-06,
+      "loss": 0.2701,
+      "step": 800
+    },
+    {
+      "epoch": 0.23059739136701016,
+      "eval_loss": 0.25545039772987366,
+      "eval_mse": 0.2554504075245932,
+      "eval_runtime": 3.4804,
+      "eval_samples_per_second": 287.327,
+      "eval_steps_per_second": 18.102,
+      "step": 800
+    },
+    {
+      "epoch": 0.23203862506305398,
+      "grad_norm": 5.442203044891357,
+      "learning_rate": 8.084977238239758e-06,
+      "loss": 0.2784,
+      "step": 805
+    },
+    {
+      "epoch": 0.2334798587590978,
+      "grad_norm": 4.991244316101074,
+      "learning_rate": 8.06980273141123e-06,
+      "loss": 0.2731,
+      "step": 810
+    },
+    {
+      "epoch": 0.2349210924551416,
+      "grad_norm": 3.7908804416656494,
+      "learning_rate": 8.054628224582701e-06,
+      "loss": 0.253,
+      "step": 815
+    },
+    {
+      "epoch": 0.2363623261511854,
+      "grad_norm": 3.7784805297851562,
+      "learning_rate": 8.039453717754174e-06,
+      "loss": 0.292,
+      "step": 820
+    },
+    {
+      "epoch": 0.23780355984722923,
+      "grad_norm": 1.8876696825027466,
+      "learning_rate": 8.024279210925645e-06,
+      "loss": 0.2637,
+      "step": 825
+    },
+    {
+      "epoch": 0.23924479354327305,
+      "grad_norm": 2.832737922668457,
+      "learning_rate": 8.009104704097117e-06,
+      "loss": 0.2975,
+      "step": 830
+    },
+    {
+      "epoch": 0.24068602723931687,
+      "grad_norm": 3.5835187435150146,
+      "learning_rate": 7.993930197268588e-06,
+      "loss": 0.2746,
+      "step": 835
+    },
+    {
+      "epoch": 0.24212726093536066,
+      "grad_norm": 2.2903802394866943,
+      "learning_rate": 7.978755690440061e-06,
+      "loss": 0.2832,
+      "step": 840
+    },
+    {
+      "epoch": 0.24356849463140448,
+      "grad_norm": 2.3398866653442383,
+      "learning_rate": 7.963581183611533e-06,
+      "loss": 0.2816,
+      "step": 845
+    },
+    {
+      "epoch": 0.2450097283274483,
+      "grad_norm": 1.899884819984436,
+      "learning_rate": 7.948406676783004e-06,
+      "loss": 0.2942,
+      "step": 850
+    },
+    {
+      "epoch": 0.24645096202349212,
+      "grad_norm": 1.6949434280395508,
+      "learning_rate": 7.933232169954477e-06,
+      "loss": 0.2617,
+      "step": 855
+    },
+    {
+      "epoch": 0.2478921957195359,
+      "grad_norm": 2.029087543487549,
+      "learning_rate": 7.918057663125949e-06,
+      "loss": 0.2493,
+      "step": 860
+    },
+    {
+      "epoch": 0.24933342941557973,
+      "grad_norm": 2.05265474319458,
+      "learning_rate": 7.90288315629742e-06,
+      "loss": 0.2965,
+      "step": 865
+    },
+    {
+      "epoch": 0.25077466311162355,
+      "grad_norm": 4.606462001800537,
+      "learning_rate": 7.887708649468893e-06,
+      "loss": 0.2815,
+      "step": 870
+    },
+    {
+      "epoch": 0.2522158968076674,
+      "grad_norm": 2.490574359893799,
+      "learning_rate": 7.872534142640365e-06,
+      "loss": 0.2488,
+      "step": 875
+    },
+    {
+      "epoch": 0.2536571305037112,
+      "grad_norm": 1.7441041469573975,
+      "learning_rate": 7.857359635811836e-06,
+      "loss": 0.2621,
+      "step": 880
+    },
+    {
+      "epoch": 0.255098364199755,
+      "grad_norm": 2.0153064727783203,
+      "learning_rate": 7.842185128983309e-06,
+      "loss": 0.2612,
+      "step": 885
+    },
+    {
+      "epoch": 0.2565395978957988,
+      "grad_norm": 2.384290933609009,
+      "learning_rate": 7.82701062215478e-06,
+      "loss": 0.2721,
+      "step": 890
+    },
+    {
+      "epoch": 0.2579808315918426,
+      "grad_norm": 2.901371717453003,
+      "learning_rate": 7.811836115326252e-06,
+      "loss": 0.2611,
+      "step": 895
+    },
+    {
+      "epoch": 0.2594220652878864,
+      "grad_norm": 2.0064079761505127,
+      "learning_rate": 7.796661608497725e-06,
+      "loss": 0.2954,
+      "step": 900
+    },
+    {
+      "epoch": 0.2594220652878864,
+      "eval_loss": 0.25009381771087646,
+      "eval_mse": 0.2500938248075545,
+      "eval_runtime": 3.4662,
+      "eval_samples_per_second": 288.504,
+      "eval_steps_per_second": 18.176,
+      "step": 900
+    },
+    {
+      "epoch": 0.26086329898393024,
+      "grad_norm": 3.5631322860717773,
+      "learning_rate": 7.781487101669196e-06,
+      "loss": 0.2746,
+      "step": 905
+    },
+    {
+      "epoch": 0.26230453267997406,
+      "grad_norm": 2.9283640384674072,
+      "learning_rate": 7.766312594840668e-06,
+      "loss": 0.2791,
+      "step": 910
+    },
+    {
+      "epoch": 0.2637457663760179,
+      "grad_norm": 2.096801996231079,
+      "learning_rate": 7.75113808801214e-06,
+      "loss": 0.2549,
+      "step": 915
+    },
+    {
+      "epoch": 0.2651870000720617,
+      "grad_norm": 2.258312225341797,
+      "learning_rate": 7.735963581183612e-06,
+      "loss": 0.2512,
+      "step": 920
+    },
+    {
+      "epoch": 0.2666282337681055,
+      "grad_norm": 5.4184699058532715,
+      "learning_rate": 7.720789074355084e-06,
+      "loss": 0.2648,
+      "step": 925
+    },
+    {
+      "epoch": 0.26806946746414934,
+      "grad_norm": 2.8605494499206543,
+      "learning_rate": 7.705614567526557e-06,
+      "loss": 0.2619,
+      "step": 930
+    },
+    {
+      "epoch": 0.2695107011601931,
+      "grad_norm": 3.81048846244812,
+      "learning_rate": 7.690440060698028e-06,
+      "loss": 0.2873,
+      "step": 935
+    },
+    {
+      "epoch": 0.2709519348562369,
+      "grad_norm": 2.1442720890045166,
+      "learning_rate": 7.6752655538695e-06,
+      "loss": 0.2927,
+      "step": 940
+    },
+    {
+      "epoch": 0.27239316855228074,
+      "grad_norm": 2.3381333351135254,
+      "learning_rate": 7.660091047040971e-06,
+      "loss": 0.3048,
+      "step": 945
+    },
+    {
+      "epoch": 0.27383440224832456,
+      "grad_norm": 5.771536350250244,
+      "learning_rate": 7.644916540212444e-06,
+      "loss": 0.3202,
+      "step": 950
+    },
+    {
+      "epoch": 0.2752756359443684,
+      "grad_norm": 4.533117771148682,
+      "learning_rate": 7.629742033383915e-06,
+      "loss": 0.2824,
+      "step": 955
+    },
+    {
+      "epoch": 0.2767168696404122,
+      "grad_norm": 1.5523793697357178,
+      "learning_rate": 7.614567526555388e-06,
+      "loss": 0.2394,
+      "step": 960
+    },
+    {
+      "epoch": 0.278158103336456,
+      "grad_norm": 3.8669960498809814,
+      "learning_rate": 7.599393019726859e-06,
+      "loss": 0.2838,
+      "step": 965
+    },
+    {
+      "epoch": 0.27959933703249984,
+      "grad_norm": 6.376443862915039,
+      "learning_rate": 7.584218512898331e-06,
+      "loss": 0.3049,
+      "step": 970
+    },
+    {
+      "epoch": 0.28104057072854366,
+      "grad_norm": 2.06191349029541,
+      "learning_rate": 7.569044006069804e-06,
+      "loss": 0.2668,
+      "step": 975
+    },
+    {
+      "epoch": 0.2824818044245874,
+      "grad_norm": 8.73715877532959,
+      "learning_rate": 7.553869499241275e-06,
+      "loss": 0.3139,
+      "step": 980
+    },
+    {
+      "epoch": 0.28392303812063124,
+      "grad_norm": 2.4534220695495605,
+      "learning_rate": 7.538694992412747e-06,
+      "loss": 0.276,
+      "step": 985
+    },
+    {
+      "epoch": 0.28536427181667506,
+      "grad_norm": 1.751397728919983,
+      "learning_rate": 7.5235204855842195e-06,
+      "loss": 0.2545,
+      "step": 990
+    },
+    {
+      "epoch": 0.2868055055127189,
+      "grad_norm": 1.4238481521606445,
+      "learning_rate": 7.508345978755691e-06,
+      "loss": 0.2841,
+      "step": 995
+    },
+    {
+      "epoch": 0.2882467392087627,
+      "grad_norm": 1.4616000652313232,
+      "learning_rate": 7.493171471927163e-06,
+      "loss": 0.2618,
+      "step": 1000
+    },
+    {
+      "epoch": 0.2882467392087627,
+      "eval_loss": 0.248306006193161,
+      "eval_mse": 0.24830602062121035,
+      "eval_runtime": 3.5569,
+      "eval_samples_per_second": 281.142,
+      "eval_steps_per_second": 17.712,
+      "step": 1000
+    },
+    {
+      "epoch": 0.2896879729048065,
+      "grad_norm": 2.0549042224884033,
+      "learning_rate": 7.477996965098635e-06,
+      "loss": 0.2566,
+      "step": 1005
+    },
+    {
+      "epoch": 0.29112920660085034,
+      "grad_norm": 4.6840715408325195,
+      "learning_rate": 7.462822458270107e-06,
+      "loss": 0.3632,
+      "step": 1010
+    },
+    {
+      "epoch": 0.29257044029689416,
+      "grad_norm": 1.8822166919708252,
+      "learning_rate": 7.447647951441579e-06,
+      "loss": 0.2745,
+      "step": 1015
+    },
+    {
+      "epoch": 0.2940116739929379,
+      "grad_norm": 4.04414176940918,
+      "learning_rate": 7.4324734446130505e-06,
+      "loss": 0.2286,
+      "step": 1020
+    },
+    {
+      "epoch": 0.29545290768898175,
+      "grad_norm": 1.8269232511520386,
+      "learning_rate": 7.417298937784523e-06,
+      "loss": 0.2764,
+      "step": 1025
+    },
+    {
+      "epoch": 0.29689414138502557,
+      "grad_norm": 2.176198720932007,
+      "learning_rate": 7.402124430955995e-06,
+      "loss": 0.271,
+      "step": 1030
+    },
+    {
+      "epoch": 0.2983353750810694,
+      "grad_norm": 2.1249961853027344,
+      "learning_rate": 7.386949924127466e-06,
+      "loss": 0.2552,
+      "step": 1035
+    },
+    {
+      "epoch": 0.2997766087771132,
+      "grad_norm": 2.1173455715179443,
+      "learning_rate": 7.371775417298939e-06,
+      "loss": 0.2931,
+      "step": 1040
+    },
+    {
+      "epoch": 0.301217842473157,
+      "grad_norm": 1.8627756834030151,
+      "learning_rate": 7.356600910470411e-06,
+      "loss": 0.2653,
+      "step": 1045
+    },
+    {
+      "epoch": 0.30265907616920085,
+      "grad_norm": 3.421990394592285,
+      "learning_rate": 7.341426403641882e-06,
+      "loss": 0.2465,
+      "step": 1050
+    },
+    {
+      "epoch": 0.30410030986524467,
+      "grad_norm": 5.541344165802002,
+      "learning_rate": 7.3262518968133545e-06,
+      "loss": 0.2628,
+      "step": 1055
+    },
+    {
+      "epoch": 0.3055415435612885,
+      "grad_norm": 4.115916728973389,
+      "learning_rate": 7.311077389984827e-06,
+      "loss": 0.2809,
+      "step": 1060
+    },
+    {
+      "epoch": 0.30698277725733225,
+      "grad_norm": 5.273794651031494,
+      "learning_rate": 7.295902883156298e-06,
+      "loss": 0.2583,
+      "step": 1065
+    },
+    {
+      "epoch": 0.30842401095337607,
+      "grad_norm": 2.4409689903259277,
+      "learning_rate": 7.2807283763277704e-06,
+      "loss": 0.2551,
+      "step": 1070
+    },
+    {
+      "epoch": 0.3098652446494199,
+      "grad_norm": 1.662716269493103,
+      "learning_rate": 7.265553869499242e-06,
+      "loss": 0.2529,
+      "step": 1075
+    },
+    {
+      "epoch": 0.3113064783454637,
+      "grad_norm": 1.8716989755630493,
+      "learning_rate": 7.250379362670714e-06,
+      "loss": 0.2837,
+      "step": 1080
+    },
+    {
+      "epoch": 0.31274771204150753,
+      "grad_norm": 1.728489637374878,
+      "learning_rate": 7.235204855842186e-06,
+      "loss": 0.2885,
+      "step": 1085
+    },
+    {
+      "epoch": 0.31418894573755135,
+      "grad_norm": 2.6448850631713867,
+      "learning_rate": 7.220030349013658e-06,
+      "loss": 0.2883,
+      "step": 1090
+    },
+    {
+      "epoch": 0.31563017943359517,
+      "grad_norm": 2.3963072299957275,
+      "learning_rate": 7.20485584218513e-06,
+      "loss": 0.2581,
+      "step": 1095
+    },
+    {
+      "epoch": 0.317071413129639,
+      "grad_norm": 3.5989487171173096,
+      "learning_rate": 7.189681335356602e-06,
+      "loss": 0.3081,
+      "step": 1100
+    },
+    {
+      "epoch": 0.317071413129639,
+      "eval_loss": 0.2456330806016922,
+      "eval_mse": 0.2456330911256373,
+      "eval_runtime": 3.4886,
+      "eval_samples_per_second": 286.649,
+      "eval_steps_per_second": 18.059,
+      "step": 1100
+    },
+    {
+      "epoch": 0.3185126468256828,
+      "grad_norm": 6.0055012702941895,
+      "learning_rate": 7.174506828528074e-06,
+      "loss": 0.2705,
+      "step": 1105
+    },
+    {
+      "epoch": 0.3199538805217266,
+      "grad_norm": 2.128803014755249,
+      "learning_rate": 7.159332321699546e-06,
+      "loss": 0.2451,
+      "step": 1110
+    },
+    {
+      "epoch": 0.3213951142177704,
+      "grad_norm": 4.768571376800537,
+      "learning_rate": 7.144157814871018e-06,
+      "loss": 0.3035,
+      "step": 1115
+    },
+    {
+      "epoch": 0.3228363479138142,
+      "grad_norm": 3.2803542613983154,
+      "learning_rate": 7.1289833080424896e-06,
+      "loss": 0.2706,
+      "step": 1120
+    },
+    {
+      "epoch": 0.32427758160985803,
+      "grad_norm": 3.9689948558807373,
+      "learning_rate": 7.113808801213962e-06,
+      "loss": 0.2767,
+      "step": 1125
+    },
+    {
+      "epoch": 0.32571881530590185,
+      "grad_norm": 3.02150821685791,
+      "learning_rate": 7.098634294385432e-06,
+      "loss": 0.2778,
+      "step": 1130
+    },
+    {
+      "epoch": 0.3271600490019457,
+      "grad_norm": 1.8594034910202026,
+      "learning_rate": 7.083459787556905e-06,
+      "loss": 0.2732,
+      "step": 1135
+    },
+    {
+      "epoch": 0.3286012826979895,
+      "grad_norm": 5.168829917907715,
+      "learning_rate": 7.068285280728376e-06,
+      "loss": 0.2669,
+      "step": 1140
+    },
+    {
+      "epoch": 0.3300425163940333,
+      "grad_norm": 2.214434862136841,
+      "learning_rate": 7.053110773899848e-06,
+      "loss": 0.2847,
+      "step": 1145
+    },
+    {
+      "epoch": 0.3314837500900771,
+      "grad_norm": 8.464557647705078,
+      "learning_rate": 7.0379362670713205e-06,
+      "loss": 0.2758,
+      "step": 1150
+    },
+    {
+      "epoch": 0.3329249837861209,
+      "grad_norm": 4.346470355987549,
+      "learning_rate": 7.022761760242792e-06,
+      "loss": 0.2456,
+      "step": 1155
+    },
+    {
+      "epoch": 0.3343662174821647,
+      "grad_norm": 2.3579695224761963,
+      "learning_rate": 7.007587253414264e-06,
+      "loss": 0.2654,
+      "step": 1160
+    },
+    {
+      "epoch": 0.33580745117820854,
+      "grad_norm": 2.9135961532592773,
+      "learning_rate": 6.9924127465857364e-06,
+      "loss": 0.2676,
+      "step": 1165
+    },
+    {
+      "epoch": 0.33724868487425236,
+      "grad_norm": 2.272632122039795,
+      "learning_rate": 6.977238239757208e-06,
+      "loss": 0.2579,
+      "step": 1170
+    },
+    {
+      "epoch": 0.3386899185702962,
+      "grad_norm": 4.541927814483643,
+      "learning_rate": 6.96206373292868e-06,
+      "loss": 0.2448,
+      "step": 1175
+    },
+    {
+      "epoch": 0.34013115226634,
+      "grad_norm": 1.557702898979187,
+      "learning_rate": 6.946889226100152e-06,
+      "loss": 0.2729,
+      "step": 1180
+    },
+    {
+      "epoch": 0.3415723859623838,
+      "grad_norm": 5.71626615524292,
+      "learning_rate": 6.931714719271624e-06,
+      "loss": 0.2691,
+      "step": 1185
+    },
+    {
+      "epoch": 0.34301361965842764,
+      "grad_norm": 1.9943206310272217,
+      "learning_rate": 6.916540212443096e-06,
+      "loss": 0.2357,
+      "step": 1190
+    },
+    {
+      "epoch": 0.3444548533544714,
+      "grad_norm": 1.7325447797775269,
+      "learning_rate": 6.901365705614567e-06,
+      "loss": 0.2543,
+      "step": 1195
+    },
+    {
+      "epoch": 0.3458960870505152,
+      "grad_norm": 2.789851427078247,
+      "learning_rate": 6.88619119878604e-06,
+      "loss": 0.2544,
+      "step": 1200
+    },
+    {
+      "epoch": 0.3458960870505152,
+      "eval_loss": 0.23698876798152924,
+      "eval_mse": 0.2369887590147555,
+      "eval_runtime": 3.6793,
+      "eval_samples_per_second": 271.794,
+      "eval_steps_per_second": 17.123,
+      "step": 1200
+    },
+    {
+      "epoch": 0.34733732074655904,
+      "grad_norm": 3.181692361831665,
+      "learning_rate": 6.871016691957512e-06,
+      "loss": 0.2553,
+      "step": 1205
+    },
+    {
+      "epoch": 0.34877855444260286,
+      "grad_norm": 1.9078682661056519,
+      "learning_rate": 6.855842185128983e-06,
+      "loss": 0.2621,
+      "step": 1210
+    },
+    {
+      "epoch": 0.3502197881386467,
+      "grad_norm": 8.280984878540039,
+      "learning_rate": 6.8406676783004556e-06,
+      "loss": 0.2373,
+      "step": 1215
+    },
+    {
+      "epoch": 0.3516610218346905,
+      "grad_norm": 2.165010929107666,
+      "learning_rate": 6.825493171471928e-06,
+      "loss": 0.2674,
+      "step": 1220
+    },
+    {
+      "epoch": 0.3531022555307343,
+      "grad_norm": 1.7144526243209839,
+      "learning_rate": 6.810318664643399e-06,
+      "loss": 0.2628,
+      "step": 1225
+    },
+    {
+      "epoch": 0.35454348922677814,
+      "grad_norm": 2.039090394973755,
+      "learning_rate": 6.7951441578148715e-06,
+      "loss": 0.257,
+      "step": 1230
+    },
+    {
+      "epoch": 0.35598472292282196,
+      "grad_norm": 2.4305779933929443,
+      "learning_rate": 6.779969650986343e-06,
+      "loss": 0.2385,
+      "step": 1235
+    },
+    {
+      "epoch": 0.3574259566188657,
+      "grad_norm": 2.1004676818847656,
+      "learning_rate": 6.764795144157815e-06,
+      "loss": 0.2667,
+      "step": 1240
+    },
+    {
+      "epoch": 0.35886719031490955,
+      "grad_norm": 3.492386817932129,
+      "learning_rate": 6.749620637329287e-06,
+      "loss": 0.2809,
+      "step": 1245
+    },
+    {
+      "epoch": 0.36030842401095337,
+      "grad_norm": 2.710477352142334,
+      "learning_rate": 6.734446130500759e-06,
+      "loss": 0.2571,
+      "step": 1250
+    },
+    {
+      "epoch": 0.3617496577069972,
+      "grad_norm": 2.1155197620391846,
+      "learning_rate": 6.719271623672231e-06,
+      "loss": 0.2362,
+      "step": 1255
+    },
+    {
+      "epoch": 0.363190891403041,
+      "grad_norm": 2.056640625,
+      "learning_rate": 6.704097116843703e-06,
+      "loss": 0.2744,
+      "step": 1260
+    },
+    {
+      "epoch": 0.3646321250990848,
+      "grad_norm": 3.4779868125915527,
+      "learning_rate": 6.688922610015175e-06,
+      "loss": 0.2635,
+      "step": 1265
+    },
+    {
+      "epoch": 0.36607335879512864,
+      "grad_norm": 2.7803256511688232,
+      "learning_rate": 6.673748103186647e-06,
+      "loss": 0.2594,
+      "step": 1270
+    },
+    {
+      "epoch": 0.36751459249117246,
+      "grad_norm": 3.466438055038452,
+      "learning_rate": 6.658573596358119e-06,
+      "loss": 0.2791,
+      "step": 1275
+    },
+    {
+      "epoch": 0.3689558261872163,
+      "grad_norm": 1.9647109508514404,
+      "learning_rate": 6.643399089529591e-06,
+      "loss": 0.2589,
+      "step": 1280
+    },
+    {
+      "epoch": 0.37039705988326005,
+      "grad_norm": 3.195117950439453,
+      "learning_rate": 6.628224582701063e-06,
+      "loss": 0.2722,
+      "step": 1285
+    },
+    {
+      "epoch": 0.37183829357930387,
+      "grad_norm": 2.193518877029419,
+      "learning_rate": 6.613050075872534e-06,
+      "loss": 0.2467,
+      "step": 1290
+    },
+    {
+      "epoch": 0.3732795272753477,
+      "grad_norm": 3.5873641967773438,
+      "learning_rate": 6.5978755690440065e-06,
+      "loss": 0.2837,
+      "step": 1295
+    },
+    {
+      "epoch": 0.3747207609713915,
+      "grad_norm": 1.9536793231964111,
+      "learning_rate": 6.582701062215479e-06,
+      "loss": 0.2593,
+      "step": 1300
+    },
+    {
+      "epoch": 0.3747207609713915,
+      "eval_loss": 0.23485003411769867,
+      "eval_mse": 0.23485004005674273,
+      "eval_runtime": 3.6259,
+      "eval_samples_per_second": 275.793,
+      "eval_steps_per_second": 17.375,
+      "step": 1300
+    },
+    {
+      "epoch": 0.37616199466743533,
+      "grad_norm": 1.7588564157485962,
+      "learning_rate": 6.56752655538695e-06,
+      "loss": 0.2633,
+      "step": 1305
+    },
+    {
+      "epoch": 0.37760322836347915,
+      "grad_norm": 2.348902940750122,
+      "learning_rate": 6.552352048558422e-06,
+      "loss": 0.2499,
+      "step": 1310
+    },
+    {
+      "epoch": 0.37904446205952297,
+      "grad_norm": 3.767819881439209,
+      "learning_rate": 6.537177541729895e-06,
+      "loss": 0.2939,
+      "step": 1315
+    },
+    {
+      "epoch": 0.3804856957555668,
+      "grad_norm": 5.1513495445251465,
+      "learning_rate": 6.522003034901366e-06,
+      "loss": 0.2408,
+      "step": 1320
+    },
+    {
+      "epoch": 0.38192692945161055,
+      "grad_norm": 1.6535985469818115,
+      "learning_rate": 6.506828528072838e-06,
+      "loss": 0.2944,
+      "step": 1325
+    },
+    {
+      "epoch": 0.3833681631476544,
+      "grad_norm": 3.1404776573181152,
+      "learning_rate": 6.4916540212443106e-06,
+      "loss": 0.2764,
+      "step": 1330
+    },
+    {
+      "epoch": 0.3848093968436982,
+      "grad_norm": 5.461513996124268,
+      "learning_rate": 6.476479514415782e-06,
+      "loss": 0.2748,
+      "step": 1335
+    },
+    {
+      "epoch": 0.386250630539742,
+      "grad_norm": 3.2750980854034424,
+      "learning_rate": 6.461305007587254e-06,
+      "loss": 0.2512,
+      "step": 1340
+    },
+    {
+      "epoch": 0.38769186423578583,
+      "grad_norm": 1.7520034313201904,
+      "learning_rate": 6.446130500758726e-06,
+      "loss": 0.2625,
+      "step": 1345
+    },
+    {
+      "epoch": 0.38913309793182965,
+      "grad_norm": 2.008507251739502,
+      "learning_rate": 6.430955993930198e-06,
+      "loss": 0.2512,
+      "step": 1350
+    },
+    {
+      "epoch": 0.3905743316278735,
+      "grad_norm": 2.8682732582092285,
+      "learning_rate": 6.41578148710167e-06,
+      "loss": 0.2745,
+      "step": 1355
+    },
+    {
+      "epoch": 0.3920155653239173,
+      "grad_norm": 6.220619201660156,
+      "learning_rate": 6.4006069802731415e-06,
+      "loss": 0.2537,
+      "step": 1360
+    },
+    {
+      "epoch": 0.3934567990199611,
+      "grad_norm": 4.935564994812012,
+      "learning_rate": 6.385432473444614e-06,
+      "loss": 0.2733,
+      "step": 1365
+    },
+    {
+      "epoch": 0.3948980327160049,
+      "grad_norm": 3.9264633655548096,
+      "learning_rate": 6.370257966616086e-06,
+      "loss": 0.283,
+      "step": 1370
+    },
+    {
+      "epoch": 0.3963392664120487,
+      "grad_norm": 2.657594680786133,
+      "learning_rate": 6.3550834597875574e-06,
+      "loss": 0.3157,
+      "step": 1375
+    },
+    {
+      "epoch": 0.3977805001080925,
+      "grad_norm": 4.369290351867676,
+      "learning_rate": 6.33990895295903e-06,
+      "loss": 0.2876,
+      "step": 1380
+    },
+    {
+      "epoch": 0.39922173380413634,
+      "grad_norm": 1.5138938426971436,
+      "learning_rate": 6.324734446130502e-06,
+      "loss": 0.2504,
+      "step": 1385
+    },
+    {
+      "epoch": 0.40066296750018016,
+      "grad_norm": 2.0192885398864746,
+      "learning_rate": 6.309559939301973e-06,
+      "loss": 0.2467,
+      "step": 1390
+    },
+    {
+      "epoch": 0.402104201196224,
+      "grad_norm": 3.4528374671936035,
+      "learning_rate": 6.294385432473446e-06,
+      "loss": 0.2801,
+      "step": 1395
+    },
+    {
+      "epoch": 0.4035454348922678,
+      "grad_norm": 3.7361555099487305,
+      "learning_rate": 6.279210925644917e-06,
+      "loss": 0.2361,
+      "step": 1400
+    },
+    {
+      "epoch": 0.4035454348922678,
+      "eval_loss": 0.24057632684707642,
+      "eval_mse": 0.24057634546631015,
+      "eval_runtime": 3.4722,
+      "eval_samples_per_second": 288.002,
+      "eval_steps_per_second": 18.144,
+      "step": 1400
+    },
+    {
+      "epoch": 0.4049866685883116,
+      "grad_norm": 5.139772891998291,
+      "learning_rate": 6.264036418816389e-06,
+      "loss": 0.2905,
+      "step": 1405
+    },
+    {
+      "epoch": 0.40642790228435544,
+      "grad_norm": 1.7238795757293701,
+      "learning_rate": 6.2488619119878615e-06,
+      "loss": 0.2533,
+      "step": 1410
+    },
+    {
+      "epoch": 0.4078691359803992,
+      "grad_norm": 3.280794620513916,
+      "learning_rate": 6.233687405159333e-06,
+      "loss": 0.2654,
+      "step": 1415
+    },
+    {
+      "epoch": 0.409310369676443,
+      "grad_norm": 4.247371673583984,
+      "learning_rate": 6.218512898330805e-06,
+      "loss": 0.2631,
+      "step": 1420
+    },
+    {
+      "epoch": 0.41075160337248684,
+      "grad_norm": 2.0990066528320312,
+      "learning_rate": 6.203338391502277e-06,
+      "loss": 0.2299,
+      "step": 1425
+    },
+    {
+      "epoch": 0.41219283706853066,
+      "grad_norm": 2.9505622386932373,
+      "learning_rate": 6.188163884673749e-06,
+      "loss": 0.3077,
+      "step": 1430
+    },
+    {
+      "epoch": 0.4136340707645745,
+      "grad_norm": 5.29465913772583,
+      "learning_rate": 6.172989377845221e-06,
+      "loss": 0.2661,
+      "step": 1435
+    },
+    {
+      "epoch": 0.4150753044606183,
+      "grad_norm": 4.061058044433594,
+      "learning_rate": 6.157814871016693e-06,
+      "loss": 0.2862,
+      "step": 1440
+    },
+    {
+      "epoch": 0.4165165381566621,
+      "grad_norm": 2.681509017944336,
+      "learning_rate": 6.142640364188165e-06,
+      "loss": 0.2757,
+      "step": 1445
+    },
+    {
+      "epoch": 0.41795777185270594,
+      "grad_norm": 5.9882493019104,
+      "learning_rate": 6.127465857359637e-06,
+      "loss": 0.2501,
+      "step": 1450
+    },
+    {
+      "epoch": 0.4193990055487497,
+      "grad_norm": 1.7219948768615723,
+      "learning_rate": 6.112291350531108e-06,
+      "loss": 0.251,
+      "step": 1455
+    },
+    {
+      "epoch": 0.4208402392447935,
+      "grad_norm": 3.918973207473755,
+      "learning_rate": 6.09711684370258e-06,
+      "loss": 0.2806,
+      "step": 1460
+    },
+    {
+      "epoch": 0.42228147294083734,
+      "grad_norm": 4.979806900024414,
+      "learning_rate": 6.081942336874051e-06,
+      "loss": 0.2602,
+      "step": 1465
+    },
+    {
+      "epoch": 0.42372270663688116,
+      "grad_norm": 2.2970187664031982,
+      "learning_rate": 6.0667678300455234e-06,
+      "loss": 0.2471,
+      "step": 1470
+    },
+    {
+      "epoch": 0.425163940332925,
+      "grad_norm": 2.741425037384033,
+      "learning_rate": 6.051593323216996e-06,
+      "loss": 0.2476,
+      "step": 1475
+    },
+    {
+      "epoch": 0.4266051740289688,
+      "grad_norm": 2.258983612060547,
+      "learning_rate": 6.036418816388467e-06,
+      "loss": 0.2542,
+      "step": 1480
+    },
+    {
+      "epoch": 0.4280464077250126,
+      "grad_norm": 2.742643356323242,
+      "learning_rate": 6.021244309559939e-06,
+      "loss": 0.2677,
+      "step": 1485
+    },
+    {
+      "epoch": 0.42948764142105644,
+      "grad_norm": 2.7305283546447754,
+      "learning_rate": 6.006069802731412e-06,
+      "loss": 0.2645,
+      "step": 1490
+    },
+    {
+      "epoch": 0.43092887511710026,
+      "grad_norm": 4.1162519454956055,
+      "learning_rate": 5.990895295902883e-06,
+      "loss": 0.2201,
+      "step": 1495
+    },
+    {
+      "epoch": 0.432370108813144,
+      "grad_norm": 2.419027805328369,
+      "learning_rate": 5.975720789074355e-06,
+      "loss": 0.2536,
+      "step": 1500
+    },
+    {
+      "epoch": 0.432370108813144,
+      "eval_loss": 0.245252326130867,
+      "eval_mse": 0.24525232070405037,
+      "eval_runtime": 3.5207,
+      "eval_samples_per_second": 284.032,
+      "eval_steps_per_second": 17.894,
+      "step": 1500
+    },
+    {
+      "epoch": 0.43381134250918785,
+      "grad_norm": 1.9734622240066528,
+      "learning_rate": 5.9605462822458275e-06,
+      "loss": 0.2538,
+      "step": 1505
+    },
+    {
+      "epoch": 0.43525257620523167,
+      "grad_norm": 2.8142011165618896,
+      "learning_rate": 5.945371775417299e-06,
+      "loss": 0.286,
+      "step": 1510
+    },
+    {
+      "epoch": 0.4366938099012755,
+      "grad_norm": 3.7075448036193848,
+      "learning_rate": 5.930197268588771e-06,
+      "loss": 0.2504,
+      "step": 1515
+    },
+    {
+      "epoch": 0.4381350435973193,
+      "grad_norm": 3.2373270988464355,
+      "learning_rate": 5.9150227617602426e-06,
+      "loss": 0.2605,
+      "step": 1520
+    },
+    {
+      "epoch": 0.4395762772933631,
+      "grad_norm": 7.530263423919678,
+      "learning_rate": 5.899848254931715e-06,
+      "loss": 0.2825,
+      "step": 1525
+    },
+    {
+      "epoch": 0.44101751098940695,
+      "grad_norm": 4.141541481018066,
+      "learning_rate": 5.884673748103187e-06,
+      "loss": 0.2765,
+      "step": 1530
+    },
+    {
+      "epoch": 0.44245874468545077,
+      "grad_norm": 2.2406415939331055,
+      "learning_rate": 5.8694992412746585e-06,
+      "loss": 0.2392,
+      "step": 1535
+    },
+    {
+      "epoch": 0.4438999783814946,
+      "grad_norm": 5.379973411560059,
+      "learning_rate": 5.854324734446131e-06,
+      "loss": 0.2416,
+      "step": 1540
+    },
+    {
+      "epoch": 0.44534121207753835,
+      "grad_norm": 1.5827113389968872,
+      "learning_rate": 5.839150227617603e-06,
+      "loss": 0.2664,
+      "step": 1545
+    },
+    {
+      "epoch": 0.44678244577358217,
+      "grad_norm": 3.172206401824951,
+      "learning_rate": 5.823975720789074e-06,
+      "loss": 0.2655,
+      "step": 1550
+    },
+    {
+      "epoch": 0.448223679469626,
+      "grad_norm": 1.9343550205230713,
+      "learning_rate": 5.808801213960547e-06,
+      "loss": 0.2716,
+      "step": 1555
+    },
+    {
+      "epoch": 0.4496649131656698,
+      "grad_norm": 1.622921109199524,
+      "learning_rate": 5.793626707132019e-06,
+      "loss": 0.2295,
+      "step": 1560
+    },
+    {
+      "epoch": 0.45110614686171363,
+      "grad_norm": 2.6616415977478027,
+      "learning_rate": 5.77845220030349e-06,
+      "loss": 0.2832,
+      "step": 1565
+    },
+    {
+      "epoch": 0.45254738055775745,
+      "grad_norm": 3.7298784255981445,
+      "learning_rate": 5.7632776934749625e-06,
+      "loss": 0.2376,
+      "step": 1570
+    },
+    {
+      "epoch": 0.45398861425380127,
+      "grad_norm": 2.567563056945801,
+      "learning_rate": 5.748103186646434e-06,
+      "loss": 0.2639,
+      "step": 1575
+    },
+    {
+      "epoch": 0.4554298479498451,
+      "grad_norm": 3.1561615467071533,
+      "learning_rate": 5.732928679817906e-06,
+      "loss": 0.2606,
+      "step": 1580
+    },
+    {
+      "epoch": 0.45687108164588885,
+      "grad_norm": 2.5020530223846436,
+      "learning_rate": 5.7177541729893784e-06,
+      "loss": 0.2384,
+      "step": 1585
+    },
+    {
+      "epoch": 0.4583123153419327,
+      "grad_norm": 1.5679707527160645,
+      "learning_rate": 5.70257966616085e-06,
+      "loss": 0.2332,
+      "step": 1590
+    },
+    {
+      "epoch": 0.4597535490379765,
+      "grad_norm": 1.720269799232483,
+      "learning_rate": 5.687405159332322e-06,
+      "loss": 0.2548,
+      "step": 1595
+    },
+    {
+      "epoch": 0.4611947827340203,
+      "grad_norm": 6.216645240783691,
+      "learning_rate": 5.672230652503794e-06,
+      "loss": 0.26,
+      "step": 1600
+    },
+    {
+      "epoch": 0.4611947827340203,
+      "eval_loss": 0.2568361163139343,
+      "eval_mse": 0.256836162135005,
+      "eval_runtime": 3.4453,
+      "eval_samples_per_second": 290.255,
+      "eval_steps_per_second": 18.286,
+      "step": 1600
+    },
+    {
+      "epoch": 0.46263601643006413,
+      "grad_norm": 1.7799232006072998,
+      "learning_rate": 5.657056145675266e-06,
+      "loss": 0.2363,
+      "step": 1605
+    },
+    {
+      "epoch": 0.46407725012610795,
+      "grad_norm": 3.4367353916168213,
+      "learning_rate": 5.641881638846738e-06,
+      "loss": 0.2506,
+      "step": 1610
+    },
+    {
+      "epoch": 0.4655184838221518,
+      "grad_norm": 6.947693824768066,
+      "learning_rate": 5.62670713201821e-06,
+      "loss": 0.2813,
+      "step": 1615
+    },
+    {
+      "epoch": 0.4669597175181956,
+      "grad_norm": 2.6440799236297607,
+      "learning_rate": 5.611532625189682e-06,
+      "loss": 0.2371,
+      "step": 1620
+    },
+    {
+      "epoch": 0.4684009512142394,
+      "grad_norm": 2.7614519596099854,
+      "learning_rate": 5.596358118361154e-06,
+      "loss": 0.2631,
+      "step": 1625
+    },
+    {
+      "epoch": 0.4698421849102832,
+      "grad_norm": 3.076611042022705,
+      "learning_rate": 5.581183611532625e-06,
+      "loss": 0.2669,
+      "step": 1630
+    },
+    {
+      "epoch": 0.471283418606327,
+      "grad_norm": 2.3479678630828857,
+      "learning_rate": 5.5660091047040976e-06,
+      "loss": 0.2498,
+      "step": 1635
+    },
+    {
+      "epoch": 0.4727246523023708,
+      "grad_norm": 3.953763723373413,
+      "learning_rate": 5.55083459787557e-06,
+      "loss": 0.2795,
+      "step": 1640
+    },
+    {
+      "epoch": 0.47416588599841464,
+      "grad_norm": 2.7778728008270264,
+      "learning_rate": 5.535660091047041e-06,
+      "loss": 0.2549,
+      "step": 1645
+    },
+    {
+      "epoch": 0.47560711969445846,
+      "grad_norm": 2.6670873165130615,
+      "learning_rate": 5.5204855842185135e-06,
+      "loss": 0.2808,
+      "step": 1650
+    },
+    {
+      "epoch": 0.4770483533905023,
+      "grad_norm": 2.9756405353546143,
+      "learning_rate": 5.505311077389986e-06,
+      "loss": 0.2716,
+      "step": 1655
+    },
+    {
+      "epoch": 0.4784895870865461,
+      "grad_norm": 2.9402782917022705,
+      "learning_rate": 5.490136570561457e-06,
+      "loss": 0.2515,
+      "step": 1660
+    },
+    {
+      "epoch": 0.4799308207825899,
+      "grad_norm": 3.0504231452941895,
+      "learning_rate": 5.474962063732929e-06,
+      "loss": 0.2305,
+      "step": 1665
+    },
+    {
+      "epoch": 0.48137205447863374,
+      "grad_norm": 2.28179931640625,
+      "learning_rate": 5.459787556904402e-06,
+      "loss": 0.2637,
+      "step": 1670
+    },
+    {
+      "epoch": 0.4828132881746775,
+      "grad_norm": 1.8344733715057373,
+      "learning_rate": 5.444613050075873e-06,
+      "loss": 0.261,
+      "step": 1675
+    },
+    {
+      "epoch": 0.4842545218707213,
+      "grad_norm": 2.4778668880462646,
+      "learning_rate": 5.429438543247345e-06,
+      "loss": 0.2681,
+      "step": 1680
+    },
+    {
+      "epoch": 0.48569575556676514,
+      "grad_norm": 2.5058882236480713,
+      "learning_rate": 5.414264036418817e-06,
+      "loss": 0.2576,
+      "step": 1685
+    },
+    {
+      "epoch": 0.48713698926280896,
+      "grad_norm": 5.368293762207031,
+      "learning_rate": 5.399089529590289e-06,
+      "loss": 0.2598,
+      "step": 1690
+    },
+    {
+      "epoch": 0.4885782229588528,
+      "grad_norm": 2.2240071296691895,
+      "learning_rate": 5.383915022761761e-06,
+      "loss": 0.2631,
+      "step": 1695
+    },
+    {
+      "epoch": 0.4900194566548966,
+      "grad_norm": 2.5634841918945312,
+      "learning_rate": 5.368740515933233e-06,
+      "loss": 0.2897,
+      "step": 1700
+    },
+    {
+      "epoch": 0.4900194566548966,
+      "eval_loss": 0.2567506432533264,
+      "eval_mse": 0.2567506678728969,
+      "eval_runtime": 3.488,
+      "eval_samples_per_second": 286.701,
+      "eval_steps_per_second": 18.062,
+      "step": 1700
+    },
+    {
+      "epoch": 0.4914606903509404,
+      "grad_norm": 2.202760696411133,
+      "learning_rate": 5.353566009104705e-06,
+      "loss": 0.2736,
+      "step": 1705
+    },
+    {
+      "epoch": 0.49290192404698424,
+      "grad_norm": 2.0145926475524902,
+      "learning_rate": 5.338391502276177e-06,
+      "loss": 0.2326,
+      "step": 1710
+    },
+    {
+      "epoch": 0.494343157743028,
+      "grad_norm": 3.5262668132781982,
+      "learning_rate": 5.3232169954476485e-06,
+      "loss": 0.2561,
+      "step": 1715
+    },
+    {
+      "epoch": 0.4957843914390718,
+      "grad_norm": 2.2268214225769043,
+      "learning_rate": 5.308042488619121e-06,
+      "loss": 0.2755,
+      "step": 1720
+    },
+    {
+      "epoch": 0.49722562513511565,
+      "grad_norm": 3.3295412063598633,
+      "learning_rate": 5.292867981790593e-06,
+      "loss": 0.228,
+      "step": 1725
+    },
+    {
+      "epoch": 0.49866685883115947,
+      "grad_norm": 4.007626533508301,
+      "learning_rate": 5.277693474962064e-06,
+      "loss": 0.2713,
+      "step": 1730
+    },
+    {
+      "epoch": 0.5001080925272033,
+      "grad_norm": 2.5232491493225098,
+      "learning_rate": 5.262518968133537e-06,
+      "loss": 0.2558,
+      "step": 1735
+    },
+    {
+      "epoch": 0.5015493262232471,
+      "grad_norm": 1.9738813638687134,
+      "learning_rate": 5.247344461305008e-06,
+      "loss": 0.2377,
+      "step": 1740
+    },
+    {
+      "epoch": 0.5029905599192909,
+      "grad_norm": 1.8534965515136719,
+      "learning_rate": 5.23216995447648e-06,
+      "loss": 0.2598,
+      "step": 1745
+    },
+    {
+      "epoch": 0.5044317936153347,
+      "grad_norm": 2.304119110107422,
+      "learning_rate": 5.2169954476479526e-06,
+      "loss": 0.2546,
+      "step": 1750
+    },
+    {
+      "epoch": 0.5058730273113785,
+      "grad_norm": 3.076138734817505,
+      "learning_rate": 5.201820940819424e-06,
+      "loss": 0.2757,
+      "step": 1755
+    },
+    {
+      "epoch": 0.5073142610074224,
+      "grad_norm": 3.0180857181549072,
+      "learning_rate": 5.186646433990896e-06,
+      "loss": 0.2987,
+      "step": 1760
+    },
+    {
+      "epoch": 0.5087554947034661,
+      "grad_norm": 1.7307599782943726,
+      "learning_rate": 5.1714719271623685e-06,
+      "loss": 0.2563,
+      "step": 1765
+    },
+    {
+      "epoch": 0.51019672839951,
+      "grad_norm": 3.3566253185272217,
+      "learning_rate": 5.15629742033384e-06,
+      "loss": 0.2677,
+      "step": 1770
+    },
+    {
+      "epoch": 0.5116379620955538,
+      "grad_norm": 5.262298107147217,
+      "learning_rate": 5.141122913505312e-06,
+      "loss": 0.2835,
+      "step": 1775
+    },
+    {
+      "epoch": 0.5130791957915976,
+      "grad_norm": 2.29787540435791,
+      "learning_rate": 5.125948406676784e-06,
+      "loss": 0.278,
+      "step": 1780
+    },
+    {
+      "epoch": 0.5145204294876414,
+      "grad_norm": 2.0509817600250244,
+      "learning_rate": 5.110773899848256e-06,
+      "loss": 0.2633,
+      "step": 1785
+    },
+    {
+      "epoch": 0.5159616631836852,
+      "grad_norm": 3.3420064449310303,
+      "learning_rate": 5.095599393019726e-06,
+      "loss": 0.2354,
+      "step": 1790
+    },
+    {
+      "epoch": 0.5174028968797291,
+      "grad_norm": 4.251697063446045,
+      "learning_rate": 5.080424886191199e-06,
+      "loss": 0.2567,
+      "step": 1795
+    },
+    {
+      "epoch": 0.5188441305757728,
+      "grad_norm": 1.7263171672821045,
+      "learning_rate": 5.065250379362671e-06,
+      "loss": 0.2597,
+      "step": 1800
+    },
+    {
+      "epoch": 0.5188441305757728,
+      "eval_loss": 0.23586300015449524,
+      "eval_mse": 0.23586300941184163,
+      "eval_runtime": 3.5548,
+      "eval_samples_per_second": 281.313,
+      "eval_steps_per_second": 17.723,
+      "step": 1800
+    },
+    {
+      "epoch": 0.5202853642718167,
+      "grad_norm": 2.9523870944976807,
+      "learning_rate": 5.050075872534142e-06,
+      "loss": 0.2565,
+      "step": 1805
+    },
+    {
+      "epoch": 0.5217265979678605,
+      "grad_norm": 2.9274137020111084,
+      "learning_rate": 5.0349013657056145e-06,
+      "loss": 0.2499,
+      "step": 1810
+    },
+    {
+      "epoch": 0.5231678316639043,
+      "grad_norm": 2.106623411178589,
+      "learning_rate": 5.019726858877087e-06,
+      "loss": 0.2448,
+      "step": 1815
+    },
+    {
+      "epoch": 0.5246090653599481,
+      "grad_norm": 2.8787355422973633,
+      "learning_rate": 5.004552352048558e-06,
+      "loss": 0.2668,
+      "step": 1820
+    },
+    {
+      "epoch": 0.5260502990559919,
+      "grad_norm": 2.016101837158203,
+      "learning_rate": 4.989377845220031e-06,
+      "loss": 0.2573,
+      "step": 1825
+    },
+    {
+      "epoch": 0.5274915327520358,
+      "grad_norm": 3.1823604106903076,
+      "learning_rate": 4.974203338391503e-06,
+      "loss": 0.2628,
+      "step": 1830
+    },
+    {
+      "epoch": 0.5289327664480795,
+      "grad_norm": 2.1019415855407715,
+      "learning_rate": 4.959028831562975e-06,
+      "loss": 0.2825,
+      "step": 1835
+    },
+    {
+      "epoch": 0.5303740001441234,
+      "grad_norm": 2.71608829498291,
+      "learning_rate": 4.943854324734446e-06,
+      "loss": 0.2495,
+      "step": 1840
+    },
+    {
+      "epoch": 0.5318152338401672,
+      "grad_norm": 1.8275377750396729,
+      "learning_rate": 4.9286798179059185e-06,
+      "loss": 0.2535,
+      "step": 1845
+    },
+    {
+      "epoch": 0.533256467536211,
+      "grad_norm": 3.9251434803009033,
+      "learning_rate": 4.913505311077391e-06,
+      "loss": 0.2881,
+      "step": 1850
+    },
+    {
+      "epoch": 0.5346977012322548,
+      "grad_norm": 2.2690351009368896,
+      "learning_rate": 4.898330804248862e-06,
+      "loss": 0.2495,
+      "step": 1855
+    },
+    {
+      "epoch": 0.5361389349282987,
+      "grad_norm": 1.780943751335144,
+      "learning_rate": 4.8831562974203345e-06,
+      "loss": 0.2658,
+      "step": 1860
+    },
+    {
+      "epoch": 0.5375801686243424,
+      "grad_norm": 2.1613879203796387,
+      "learning_rate": 4.867981790591807e-06,
+      "loss": 0.2688,
+      "step": 1865
+    },
+    {
+      "epoch": 0.5390214023203862,
+      "grad_norm": 4.357624530792236,
+      "learning_rate": 4.852807283763278e-06,
+      "loss": 0.2557,
+      "step": 1870
+    },
+    {
+      "epoch": 0.5404626360164301,
+      "grad_norm": 3.1877598762512207,
+      "learning_rate": 4.8376327769347495e-06,
+      "loss": 0.2637,
+      "step": 1875
+    },
+    {
+      "epoch": 0.5419038697124738,
+      "grad_norm": 1.587506651878357,
+      "learning_rate": 4.822458270106222e-06,
+      "loss": 0.243,
+      "step": 1880
+    },
+    {
+      "epoch": 0.5433451034085177,
+      "grad_norm": 1.8893779516220093,
+      "learning_rate": 4.807283763277694e-06,
+      "loss": 0.2513,
+      "step": 1885
+    },
+    {
+      "epoch": 0.5447863371045615,
+      "grad_norm": 5.860842704772949,
+      "learning_rate": 4.792109256449165e-06,
+      "loss": 0.27,
+      "step": 1890
+    },
+    {
+      "epoch": 0.5462275708006054,
+      "grad_norm": 1.7181977033615112,
+      "learning_rate": 4.776934749620638e-06,
+      "loss": 0.2545,
+      "step": 1895
+    },
+    {
+      "epoch": 0.5476688044966491,
+      "grad_norm": 2.9068872928619385,
+      "learning_rate": 4.761760242792109e-06,
+      "loss": 0.2489,
+      "step": 1900
+    },
+    {
+      "epoch": 0.5476688044966491,
+      "eval_loss": 0.24126099050045013,
+      "eval_mse": 0.24126099393393452,
+      "eval_runtime": 3.4541,
+      "eval_samples_per_second": 289.514,
+      "eval_steps_per_second": 18.239,
+      "step": 1900
+    },
+    {
+      "epoch": 0.549110038192693,
+      "grad_norm": 2.1639840602874756,
+      "learning_rate": 4.746585735963581e-06,
+      "loss": 0.2583,
+      "step": 1905
+    },
+    {
+      "epoch": 0.5505512718887368,
+      "grad_norm": 3.045797824859619,
+      "learning_rate": 4.731411229135054e-06,
+      "loss": 0.2373,
+      "step": 1910
+    },
+    {
+      "epoch": 0.5519925055847805,
+      "grad_norm": 2.1132616996765137,
+      "learning_rate": 4.716236722306525e-06,
+      "loss": 0.2742,
+      "step": 1915
+    },
+    {
+      "epoch": 0.5534337392808244,
+      "grad_norm": 4.257943153381348,
+      "learning_rate": 4.701062215477997e-06,
+      "loss": 0.2401,
+      "step": 1920
+    },
+    {
+      "epoch": 0.5548749729768682,
+      "grad_norm": 1.8180277347564697,
+      "learning_rate": 4.6858877086494695e-06,
+      "loss": 0.2211,
+      "step": 1925
+    },
+    {
+      "epoch": 0.556316206672912,
+      "grad_norm": 1.9830940961837769,
+      "learning_rate": 4.670713201820941e-06,
+      "loss": 0.2611,
+      "step": 1930
+    },
+    {
+      "epoch": 0.5577574403689558,
+      "grad_norm": 1.979367971420288,
+      "learning_rate": 4.655538694992413e-06,
+      "loss": 0.2499,
+      "step": 1935
+    },
+    {
+      "epoch": 0.5591986740649997,
+      "grad_norm": 3.538367748260498,
+      "learning_rate": 4.640364188163885e-06,
+      "loss": 0.2495,
+      "step": 1940
+    },
+    {
+      "epoch": 0.5606399077610434,
+      "grad_norm": 2.1496660709381104,
+      "learning_rate": 4.625189681335357e-06,
+      "loss": 0.2233,
+      "step": 1945
+    },
+    {
+      "epoch": 0.5620811414570873,
+      "grad_norm": 2.313380002975464,
+      "learning_rate": 4.610015174506829e-06,
+      "loss": 0.2713,
+      "step": 1950
+    },
+    {
+      "epoch": 0.5635223751531311,
+      "grad_norm": 1.9493372440338135,
+      "learning_rate": 4.5948406676783005e-06,
+      "loss": 0.2745,
+      "step": 1955
+    },
+    {
+      "epoch": 0.5649636088491748,
+      "grad_norm": 1.78380286693573,
+      "learning_rate": 4.579666160849773e-06,
+      "loss": 0.2705,
+      "step": 1960
+    },
+    {
+      "epoch": 0.5664048425452187,
+      "grad_norm": 5.174985885620117,
+      "learning_rate": 4.564491654021245e-06,
+      "loss": 0.2626,
+      "step": 1965
+    },
+    {
+      "epoch": 0.5678460762412625,
+      "grad_norm": 5.110122203826904,
+      "learning_rate": 4.549317147192716e-06,
+      "loss": 0.2482,
+      "step": 1970
+    },
+    {
+      "epoch": 0.5692873099373064,
+      "grad_norm": 2.324326753616333,
+      "learning_rate": 4.534142640364189e-06,
+      "loss": 0.2478,
+      "step": 1975
+    },
+    {
+      "epoch": 0.5707285436333501,
+      "grad_norm": 2.391932249069214,
+      "learning_rate": 4.518968133535661e-06,
+      "loss": 0.2335,
+      "step": 1980
+    },
+    {
+      "epoch": 0.572169777329394,
+      "grad_norm": 2.217822313308716,
+      "learning_rate": 4.503793626707132e-06,
+      "loss": 0.2382,
+      "step": 1985
+    },
+    {
+      "epoch": 0.5736110110254378,
+      "grad_norm": 3.003229856491089,
+      "learning_rate": 4.4886191198786045e-06,
+      "loss": 0.2585,
+      "step": 1990
+    },
+    {
+      "epoch": 0.5750522447214816,
+      "grad_norm": 3.210510492324829,
+      "learning_rate": 4.473444613050077e-06,
+      "loss": 0.2145,
+      "step": 1995
+    },
+    {
+      "epoch": 0.5764934784175254,
+      "grad_norm": 2.712481737136841,
+      "learning_rate": 4.458270106221548e-06,
+      "loss": 0.2376,
+      "step": 2000
+    },
+    {
+      "epoch": 0.5764934784175254,
+      "eval_loss": 0.24156416952610016,
+      "eval_mse": 0.2415641685180599,
+      "eval_runtime": 3.4679,
+      "eval_samples_per_second": 288.359,
+      "eval_steps_per_second": 18.167,
+      "step": 2000
+    },
+    {
+      "epoch": 0.5779347121135692,
+      "grad_norm": 4.5920209884643555,
+      "learning_rate": 4.44309559939302e-06,
+      "loss": 0.2551,
+      "step": 2005
+    },
+    {
+      "epoch": 0.579375945809613,
+      "grad_norm": 2.582946538925171,
+      "learning_rate": 4.427921092564492e-06,
+      "loss": 0.2492,
+      "step": 2010
+    },
+    {
+      "epoch": 0.5808171795056568,
+      "grad_norm": 3.868809700012207,
+      "learning_rate": 4.412746585735964e-06,
+      "loss": 0.271,
+      "step": 2015
+    },
+    {
+      "epoch": 0.5822584132017007,
+      "grad_norm": 2.1130621433258057,
+      "learning_rate": 4.397572078907436e-06,
+      "loss": 0.2543,
+      "step": 2020
+    },
+    {
+      "epoch": 0.5836996468977445,
+      "grad_norm": 3.2280352115631104,
+      "learning_rate": 4.382397572078908e-06,
+      "loss": 0.264,
+      "step": 2025
+    },
+    {
+      "epoch": 0.5851408805937883,
+      "grad_norm": 1.9445719718933105,
+      "learning_rate": 4.36722306525038e-06,
+      "loss": 0.2465,
+      "step": 2030
+    },
+    {
+      "epoch": 0.5865821142898321,
+      "grad_norm": 2.186732292175293,
+      "learning_rate": 4.352048558421852e-06,
+      "loss": 0.2687,
+      "step": 2035
+    },
+    {
+      "epoch": 0.5880233479858759,
+      "grad_norm": 6.876237392425537,
+      "learning_rate": 4.336874051593324e-06,
+      "loss": 0.2861,
+      "step": 2040
+    },
+    {
+      "epoch": 0.5894645816819197,
+      "grad_norm": 4.177600860595703,
+      "learning_rate": 4.321699544764795e-06,
+      "loss": 0.2553,
+      "step": 2045
+    },
+    {
+      "epoch": 0.5909058153779635,
+      "grad_norm": 1.9591648578643799,
+      "learning_rate": 4.306525037936267e-06,
+      "loss": 0.2375,
+      "step": 2050
+    },
+    {
+      "epoch": 0.5923470490740074,
+      "grad_norm": 4.677102565765381,
+      "learning_rate": 4.2913505311077395e-06,
+      "loss": 0.264,
+      "step": 2055
+    },
+    {
+      "epoch": 0.5937882827700511,
+      "grad_norm": 2.2224631309509277,
+      "learning_rate": 4.276176024279211e-06,
+      "loss": 0.2485,
+      "step": 2060
+    },
+    {
+      "epoch": 0.595229516466095,
+      "grad_norm": 2.693962574005127,
+      "learning_rate": 4.261001517450683e-06,
+      "loss": 0.2388,
+      "step": 2065
+    },
+    {
+      "epoch": 0.5966707501621388,
+      "grad_norm": 2.4235777854919434,
+      "learning_rate": 4.245827010622155e-06,
+      "loss": 0.2399,
+      "step": 2070
+    },
+    {
+      "epoch": 0.5981119838581826,
+      "grad_norm": 1.874377965927124,
+      "learning_rate": 4.230652503793627e-06,
+      "loss": 0.2326,
+      "step": 2075
+    },
+    {
+      "epoch": 0.5995532175542264,
+      "grad_norm": 3.6788363456726074,
+      "learning_rate": 4.215477996965099e-06,
+      "loss": 0.2338,
+      "step": 2080
+    },
+    {
+      "epoch": 0.6009944512502702,
+      "grad_norm": 2.06693172454834,
+      "learning_rate": 4.2003034901365705e-06,
+      "loss": 0.2347,
+      "step": 2085
+    },
+    {
+      "epoch": 0.602435684946314,
+      "grad_norm": 4.380660533905029,
+      "learning_rate": 4.185128983308043e-06,
+      "loss": 0.2569,
+      "step": 2090
+    },
+    {
+      "epoch": 0.6038769186423578,
+      "grad_norm": 2.032864809036255,
+      "learning_rate": 4.169954476479515e-06,
+      "loss": 0.2471,
+      "step": 2095
+    },
+    {
+      "epoch": 0.6053181523384017,
+      "grad_norm": 4.096874237060547,
+      "learning_rate": 4.154779969650986e-06,
+      "loss": 0.2424,
+      "step": 2100
+    },
+    {
+      "epoch": 0.6053181523384017,
+      "eval_loss": 0.24177902936935425,
+      "eval_mse": 0.24177903914719717,
+      "eval_runtime": 3.4534,
+      "eval_samples_per_second": 289.57,
+      "eval_steps_per_second": 18.243,
+      "step": 2100
+    },
+    {
+      "epoch": 0.6067593860344455,
+      "grad_norm": 2.5046980381011963,
+      "learning_rate": 4.139605462822459e-06,
+      "loss": 0.234,
+      "step": 2105
+    },
+    {
+      "epoch": 0.6082006197304893,
+      "grad_norm": 4.800332069396973,
+      "learning_rate": 4.124430955993931e-06,
+      "loss": 0.2742,
+      "step": 2110
+    },
+    {
+      "epoch": 0.6096418534265331,
+      "grad_norm": 2.0443146228790283,
+      "learning_rate": 4.109256449165402e-06,
+      "loss": 0.2408,
+      "step": 2115
+    },
+    {
+      "epoch": 0.611083087122577,
+      "grad_norm": 5.306743144989014,
+      "learning_rate": 4.0940819423368746e-06,
+      "loss": 0.2447,
+      "step": 2120
+    },
+    {
+      "epoch": 0.6125243208186207,
+      "grad_norm": 2.0920376777648926,
+      "learning_rate": 4.078907435508346e-06,
+      "loss": 0.2598,
+      "step": 2125
+    },
+    {
+      "epoch": 0.6139655545146645,
+      "grad_norm": 2.3251876831054688,
+      "learning_rate": 4.063732928679818e-06,
+      "loss": 0.2343,
+      "step": 2130
+    },
+    {
+      "epoch": 0.6154067882107084,
+      "grad_norm": 1.6690804958343506,
+      "learning_rate": 4.0485584218512905e-06,
+      "loss": 0.2488,
+      "step": 2135
+    },
+    {
+      "epoch": 0.6168480219067521,
+      "grad_norm": 4.557926177978516,
+      "learning_rate": 4.033383915022762e-06,
+      "loss": 0.2371,
+      "step": 2140
+    },
+    {
+      "epoch": 0.618289255602796,
+      "grad_norm": 1.826122760772705,
+      "learning_rate": 4.018209408194234e-06,
+      "loss": 0.2562,
+      "step": 2145
+    },
+    {
+      "epoch": 0.6197304892988398,
+      "grad_norm": 2.284559488296509,
+      "learning_rate": 4.003034901365706e-06,
+      "loss": 0.2647,
+      "step": 2150
+    },
+    {
+      "epoch": 0.6211717229948837,
+      "grad_norm": 1.968127727508545,
+      "learning_rate": 3.987860394537178e-06,
+      "loss": 0.24,
+      "step": 2155
+    },
+    {
+      "epoch": 0.6226129566909274,
+      "grad_norm": 1.8883932828903198,
+      "learning_rate": 3.97268588770865e-06,
+      "loss": 0.2508,
+      "step": 2160
+    },
+    {
+      "epoch": 0.6240541903869713,
+      "grad_norm": 4.677165985107422,
+      "learning_rate": 3.957511380880122e-06,
+      "loss": 0.2698,
+      "step": 2165
+    },
+    {
+      "epoch": 0.6254954240830151,
+      "grad_norm": 3.8323287963867188,
+      "learning_rate": 3.942336874051594e-06,
+      "loss": 0.2505,
+      "step": 2170
+    },
+    {
+      "epoch": 0.6269366577790588,
+      "grad_norm": 2.6321306228637695,
+      "learning_rate": 3.927162367223066e-06,
+      "loss": 0.2578,
+      "step": 2175
+    },
+    {
+      "epoch": 0.6283778914751027,
+      "grad_norm": 4.790838241577148,
+      "learning_rate": 3.911987860394537e-06,
+      "loss": 0.2682,
+      "step": 2180
+    },
+    {
+      "epoch": 0.6298191251711465,
+      "grad_norm": 1.60104501247406,
+      "learning_rate": 3.89681335356601e-06,
+      "loss": 0.2172,
+      "step": 2185
+    },
+    {
+      "epoch": 0.6312603588671903,
+      "grad_norm": 2.6829323768615723,
+      "learning_rate": 3.881638846737482e-06,
+      "loss": 0.319,
+      "step": 2190
+    },
+    {
+      "epoch": 0.6327015925632341,
+      "grad_norm": 5.924467086791992,
+      "learning_rate": 3.866464339908953e-06,
+      "loss": 0.2753,
+      "step": 2195
+    },
+    {
+      "epoch": 0.634142826259278,
+      "grad_norm": 2.6062183380126953,
+      "learning_rate": 3.8512898330804255e-06,
+      "loss": 0.2798,
+      "step": 2200
+    },
+    {
+      "epoch": 0.634142826259278,
+      "eval_loss": 0.2461671233177185,
+      "eval_mse": 0.24616713661520043,
+      "eval_runtime": 3.5245,
+      "eval_samples_per_second": 283.725,
+      "eval_steps_per_second": 17.875,
+      "step": 2200
+    },
+    {
+      "epoch": 0.6355840599553217,
+      "grad_norm": 5.319560527801514,
+      "learning_rate": 3.836115326251897e-06,
+      "loss": 0.2797,
+      "step": 2205
+    },
+    {
+      "epoch": 0.6370252936513656,
+      "grad_norm": 2.300398826599121,
+      "learning_rate": 3.820940819423369e-06,
+      "loss": 0.2638,
+      "step": 2210
+    },
+    {
+      "epoch": 0.6384665273474094,
+      "grad_norm": 1.8542537689208984,
+      "learning_rate": 3.8057663125948406e-06,
+      "loss": 0.2269,
+      "step": 2215
+    },
+    {
+      "epoch": 0.6399077610434531,
+      "grad_norm": 2.6789774894714355,
+      "learning_rate": 3.790591805766313e-06,
+      "loss": 0.2631,
+      "step": 2220
+    },
+    {
+      "epoch": 0.641348994739497,
+      "grad_norm": 3.5065345764160156,
+      "learning_rate": 3.7754172989377846e-06,
+      "loss": 0.2393,
+      "step": 2225
+    },
+    {
+      "epoch": 0.6427902284355408,
+      "grad_norm": 5.016166687011719,
+      "learning_rate": 3.7602427921092565e-06,
+      "loss": 0.2488,
+      "step": 2230
+    },
+    {
+      "epoch": 0.6442314621315847,
+      "grad_norm": 2.594754457473755,
+      "learning_rate": 3.7450682852807287e-06,
+      "loss": 0.2344,
+      "step": 2235
+    },
+    {
+      "epoch": 0.6456726958276284,
+      "grad_norm": 2.8543953895568848,
+      "learning_rate": 3.7298937784522006e-06,
+      "loss": 0.2477,
+      "step": 2240
+    },
+    {
+      "epoch": 0.6471139295236723,
+      "grad_norm": 1.9123929738998413,
+      "learning_rate": 3.7147192716236724e-06,
+      "loss": 0.274,
+      "step": 2245
+    },
+    {
+      "epoch": 0.6485551632197161,
+      "grad_norm": 2.7275772094726562,
+      "learning_rate": 3.699544764795144e-06,
+      "loss": 0.253,
+      "step": 2250
+    },
+    {
+      "epoch": 0.6499963969157599,
+      "grad_norm": 2.763127088546753,
+      "learning_rate": 3.6843702579666165e-06,
+      "loss": 0.2365,
+      "step": 2255
+    },
+    {
+      "epoch": 0.6514376306118037,
+      "grad_norm": 2.082650899887085,
+      "learning_rate": 3.6691957511380883e-06,
+      "loss": 0.2015,
+      "step": 2260
+    },
+    {
+      "epoch": 0.6528788643078475,
+      "grad_norm": 2.4569077491760254,
+      "learning_rate": 3.65402124430956e-06,
+      "loss": 0.2796,
+      "step": 2265
+    },
+    {
+      "epoch": 0.6543200980038913,
+      "grad_norm": 1.6337858438491821,
+      "learning_rate": 3.638846737481032e-06,
+      "loss": 0.2276,
+      "step": 2270
+    },
+    {
+      "epoch": 0.6557613316999351,
+      "grad_norm": 3.221034526824951,
+      "learning_rate": 3.623672230652504e-06,
+      "loss": 0.2193,
+      "step": 2275
+    },
+    {
+      "epoch": 0.657202565395979,
+      "grad_norm": 1.661058783531189,
+      "learning_rate": 3.608497723823976e-06,
+      "loss": 0.2468,
+      "step": 2280
+    },
+    {
+      "epoch": 0.6586437990920228,
+      "grad_norm": 3.846362590789795,
+      "learning_rate": 3.593323216995448e-06,
+      "loss": 0.2891,
+      "step": 2285
+    },
+    {
+      "epoch": 0.6600850327880666,
+      "grad_norm": 2.7255539894104004,
+      "learning_rate": 3.57814871016692e-06,
+      "loss": 0.2961,
+      "step": 2290
+    },
+    {
+      "epoch": 0.6615262664841104,
+      "grad_norm": 2.3028368949890137,
+      "learning_rate": 3.562974203338392e-06,
+      "loss": 0.2125,
+      "step": 2295
+    },
+    {
+      "epoch": 0.6629675001801542,
+      "grad_norm": 2.312697649002075,
+      "learning_rate": 3.5477996965098638e-06,
+      "loss": 0.2523,
+      "step": 2300
+    },
+    {
+      "epoch": 0.6629675001801542,
+      "eval_loss": 0.23224954307079315,
+      "eval_mse": 0.23224954986944796,
+      "eval_runtime": 3.5059,
+      "eval_samples_per_second": 285.236,
+      "eval_steps_per_second": 17.97,
+      "step": 2300
+    },
+    {
+      "epoch": 0.664408733876198,
+      "grad_norm": 1.403989315032959,
+      "learning_rate": 3.5326251896813356e-06,
+      "loss": 0.2127,
+      "step": 2305
+    },
+    {
+      "epoch": 0.6658499675722418,
+      "grad_norm": 2.4680073261260986,
+      "learning_rate": 3.517450682852808e-06,
+      "loss": 0.2631,
+      "step": 2310
+    },
+    {
+      "epoch": 0.6672912012682857,
+      "grad_norm": 2.680781841278076,
+      "learning_rate": 3.5022761760242797e-06,
+      "loss": 0.2498,
+      "step": 2315
+    },
+    {
+      "epoch": 0.6687324349643294,
+      "grad_norm": 2.0593039989471436,
+      "learning_rate": 3.4871016691957515e-06,
+      "loss": 0.2674,
+      "step": 2320
+    },
+    {
+      "epoch": 0.6701736686603733,
+      "grad_norm": 2.249380350112915,
+      "learning_rate": 3.4719271623672233e-06,
+      "loss": 0.2632,
+      "step": 2325
+    },
+    {
+      "epoch": 0.6716149023564171,
+      "grad_norm": 4.473768711090088,
+      "learning_rate": 3.4567526555386956e-06,
+      "loss": 0.274,
+      "step": 2330
+    },
+    {
+      "epoch": 0.673056136052461,
+      "grad_norm": 2.8323299884796143,
+      "learning_rate": 3.4415781487101674e-06,
+      "loss": 0.23,
+      "step": 2335
+    },
+    {
+      "epoch": 0.6744973697485047,
+      "grad_norm": 1.9655417203903198,
+      "learning_rate": 3.4264036418816392e-06,
+      "loss": 0.2741,
+      "step": 2340
+    },
+    {
+      "epoch": 0.6759386034445485,
+      "grad_norm": 2.6639366149902344,
+      "learning_rate": 3.4112291350531115e-06,
+      "loss": 0.2512,
+      "step": 2345
+    },
+    {
+      "epoch": 0.6773798371405924,
+      "grad_norm": 1.6781830787658691,
+      "learning_rate": 3.3960546282245833e-06,
+      "loss": 0.2326,
+      "step": 2350
+    },
+    {
+      "epoch": 0.6788210708366361,
+      "grad_norm": 2.3150107860565186,
+      "learning_rate": 3.380880121396055e-06,
+      "loss": 0.2317,
+      "step": 2355
+    },
+    {
+      "epoch": 0.68026230453268,
+      "grad_norm": 3.279113292694092,
+      "learning_rate": 3.365705614567527e-06,
+      "loss": 0.2542,
+      "step": 2360
+    },
+    {
+      "epoch": 0.6817035382287238,
+      "grad_norm": 1.7288917303085327,
+      "learning_rate": 3.350531107738999e-06,
+      "loss": 0.2458,
+      "step": 2365
+    },
+    {
+      "epoch": 0.6831447719247676,
+      "grad_norm": 1.6974340677261353,
+      "learning_rate": 3.3353566009104706e-06,
+      "loss": 0.2632,
+      "step": 2370
+    },
+    {
+      "epoch": 0.6845860056208114,
+      "grad_norm": 3.335235595703125,
+      "learning_rate": 3.3201820940819424e-06,
+      "loss": 0.2753,
+      "step": 2375
+    },
+    {
+      "epoch": 0.6860272393168553,
+      "grad_norm": 2.5615906715393066,
+      "learning_rate": 3.3050075872534143e-06,
+      "loss": 0.2435,
+      "step": 2380
+    },
+    {
+      "epoch": 0.687468473012899,
+      "grad_norm": 1.9486761093139648,
+      "learning_rate": 3.289833080424886e-06,
+      "loss": 0.2571,
+      "step": 2385
+    },
+    {
+      "epoch": 0.6889097067089428,
+      "grad_norm": 3.594895124435425,
+      "learning_rate": 3.2746585735963583e-06,
+      "loss": 0.2338,
+      "step": 2390
+    },
+    {
+      "epoch": 0.6903509404049867,
+      "grad_norm": 2.6904447078704834,
+      "learning_rate": 3.25948406676783e-06,
+      "loss": 0.2471,
+      "step": 2395
+    },
+    {
+      "epoch": 0.6917921741010304,
+      "grad_norm": 2.230098009109497,
+      "learning_rate": 3.244309559939302e-06,
+      "loss": 0.286,
+      "step": 2400
+    },
+    {
+      "epoch": 0.6917921741010304,
+      "eval_loss": 0.2431608885526657,
+      "eval_mse": 0.2431609066054225,
+      "eval_runtime": 3.4991,
+      "eval_samples_per_second": 285.789,
+      "eval_steps_per_second": 18.005,
+      "step": 2400
+    },
+    {
+      "epoch": 0.6932334077970743,
+      "grad_norm": 2.067410945892334,
+      "learning_rate": 3.229135053110774e-06,
+      "loss": 0.2474,
+      "step": 2405
+    },
+    {
+      "epoch": 0.6946746414931181,
+      "grad_norm": 2.398256540298462,
+      "learning_rate": 3.213960546282246e-06,
+      "loss": 0.2429,
+      "step": 2410
+    },
+    {
+      "epoch": 0.696115875189162,
+      "grad_norm": 3.5365965366363525,
+      "learning_rate": 3.198786039453718e-06,
+      "loss": 0.2478,
+      "step": 2415
+    },
+    {
+      "epoch": 0.6975571088852057,
+      "grad_norm": 2.981490135192871,
+      "learning_rate": 3.1836115326251897e-06,
+      "loss": 0.237,
+      "step": 2420
+    },
+    {
+      "epoch": 0.6989983425812496,
+      "grad_norm": 2.0100955963134766,
+      "learning_rate": 3.168437025796662e-06,
+      "loss": 0.2653,
+      "step": 2425
+    },
+    {
+      "epoch": 0.7004395762772934,
+      "grad_norm": 3.208611488342285,
+      "learning_rate": 3.153262518968134e-06,
+      "loss": 0.2621,
+      "step": 2430
+    },
+    {
+      "epoch": 0.7018808099733371,
+      "grad_norm": 3.3685500621795654,
+      "learning_rate": 3.1380880121396056e-06,
+      "loss": 0.2764,
+      "step": 2435
+    },
+    {
+      "epoch": 0.703322043669381,
+      "grad_norm": 1.9647629261016846,
+      "learning_rate": 3.1229135053110775e-06,
+      "loss": 0.2493,
+      "step": 2440
+    },
+    {
+      "epoch": 0.7047632773654248,
+      "grad_norm": 1.7301753759384155,
+      "learning_rate": 3.1077389984825497e-06,
+      "loss": 0.2773,
+      "step": 2445
+    },
+    {
+      "epoch": 0.7062045110614686,
+      "grad_norm": 2.260758638381958,
+      "learning_rate": 3.0925644916540215e-06,
+      "loss": 0.2487,
+      "step": 2450
+    },
+    {
+      "epoch": 0.7076457447575124,
+      "grad_norm": 1.7644176483154297,
+      "learning_rate": 3.0773899848254934e-06,
+      "loss": 0.27,
+      "step": 2455
+    },
+    {
+      "epoch": 0.7090869784535563,
+      "grad_norm": 2.1906681060791016,
+      "learning_rate": 3.062215477996965e-06,
+      "loss": 0.2653,
+      "step": 2460
+    },
+    {
+      "epoch": 0.7105282121496,
+      "grad_norm": 1.6636836528778076,
+      "learning_rate": 3.0470409711684375e-06,
+      "loss": 0.22,
+      "step": 2465
+    },
+    {
+      "epoch": 0.7119694458456439,
+      "grad_norm": 1.9646042585372925,
+      "learning_rate": 3.0318664643399093e-06,
+      "loss": 0.2279,
+      "step": 2470
+    },
+    {
+      "epoch": 0.7134106795416877,
+      "grad_norm": 3.8436663150787354,
+      "learning_rate": 3.016691957511381e-06,
+      "loss": 0.2675,
+      "step": 2475
+    },
+    {
+      "epoch": 0.7148519132377315,
+      "grad_norm": 1.7162784337997437,
+      "learning_rate": 3.0015174506828534e-06,
+      "loss": 0.2424,
+      "step": 2480
+    },
+    {
+      "epoch": 0.7162931469337753,
+      "grad_norm": 7.108310699462891,
+      "learning_rate": 2.986342943854325e-06,
+      "loss": 0.2585,
+      "step": 2485
+    },
+    {
+      "epoch": 0.7177343806298191,
+      "grad_norm": 1.4513427019119263,
+      "learning_rate": 2.971168437025797e-06,
+      "loss": 0.2531,
+      "step": 2490
+    },
+    {
+      "epoch": 0.719175614325863,
+      "grad_norm": 3.262995719909668,
+      "learning_rate": 2.955993930197269e-06,
+      "loss": 0.2407,
+      "step": 2495
+    },
+    {
+      "epoch": 0.7206168480219067,
+      "grad_norm": 2.73363995552063,
+      "learning_rate": 2.940819423368741e-06,
+      "loss": 0.247,
+      "step": 2500
+    },
+    {
+      "epoch": 0.7206168480219067,
+      "eval_loss": 0.23825927078723907,
+      "eval_mse": 0.23825928315892816,
+      "eval_runtime": 3.4721,
+      "eval_samples_per_second": 288.01,
+      "eval_steps_per_second": 18.145,
+      "step": 2500
+    },
+    {
+      "epoch": 0.7220580817179506,
+      "grad_norm": 2.238675594329834,
+      "learning_rate": 2.925644916540213e-06,
+      "loss": 0.2472,
+      "step": 2505
+    },
+    {
+      "epoch": 0.7234993154139944,
+      "grad_norm": 2.1989986896514893,
+      "learning_rate": 2.9104704097116847e-06,
+      "loss": 0.251,
+      "step": 2510
+    },
+    {
+      "epoch": 0.7249405491100382,
+      "grad_norm": 1.8326098918914795,
+      "learning_rate": 2.8952959028831566e-06,
+      "loss": 0.2885,
+      "step": 2515
+    },
+    {
+      "epoch": 0.726381782806082,
+      "grad_norm": 4.506540775299072,
+      "learning_rate": 2.880121396054629e-06,
+      "loss": 0.2509,
+      "step": 2520
+    },
+    {
+      "epoch": 0.7278230165021258,
+      "grad_norm": 2.287397623062134,
+      "learning_rate": 2.8649468892261007e-06,
+      "loss": 0.2502,
+      "step": 2525
+    },
+    {
+      "epoch": 0.7292642501981697,
+      "grad_norm": 1.5754716396331787,
+      "learning_rate": 2.849772382397572e-06,
+      "loss": 0.2345,
+      "step": 2530
+    },
+    {
+      "epoch": 0.7307054838942134,
+      "grad_norm": 2.309974193572998,
+      "learning_rate": 2.834597875569044e-06,
+      "loss": 0.2249,
+      "step": 2535
+    },
+    {
+      "epoch": 0.7321467175902573,
+      "grad_norm": 2.3784449100494385,
+      "learning_rate": 2.8194233687405157e-06,
+      "loss": 0.2505,
+      "step": 2540
+    },
+    {
+      "epoch": 0.733587951286301,
+      "grad_norm": 1.9537062644958496,
+      "learning_rate": 2.804248861911988e-06,
+      "loss": 0.2277,
+      "step": 2545
+    },
+    {
+      "epoch": 0.7350291849823449,
+      "grad_norm": 2.5582821369171143,
+      "learning_rate": 2.78907435508346e-06,
+      "loss": 0.2349,
+      "step": 2550
+    },
+    {
+      "epoch": 0.7364704186783887,
+      "grad_norm": 3.8102619647979736,
+      "learning_rate": 2.7738998482549316e-06,
+      "loss": 0.223,
+      "step": 2555
+    },
+    {
+      "epoch": 0.7379116523744326,
+      "grad_norm": 2.270439624786377,
+      "learning_rate": 2.758725341426404e-06,
+      "loss": 0.2737,
+      "step": 2560
+    },
+    {
+      "epoch": 0.7393528860704763,
+      "grad_norm": 1.8109246492385864,
+      "learning_rate": 2.7435508345978757e-06,
+      "loss": 0.2288,
+      "step": 2565
+    },
+    {
+      "epoch": 0.7407941197665201,
+      "grad_norm": 3.8969645500183105,
+      "learning_rate": 2.7283763277693475e-06,
+      "loss": 0.2474,
+      "step": 2570
+    },
+    {
+      "epoch": 0.742235353462564,
+      "grad_norm": 3.1700432300567627,
+      "learning_rate": 2.7132018209408194e-06,
+      "loss": 0.2739,
+      "step": 2575
+    },
+    {
+      "epoch": 0.7436765871586077,
+      "grad_norm": 1.7024991512298584,
+      "learning_rate": 2.6980273141122916e-06,
+      "loss": 0.2735,
+      "step": 2580
+    },
+    {
+      "epoch": 0.7451178208546516,
+      "grad_norm": 3.1429338455200195,
+      "learning_rate": 2.6828528072837634e-06,
+      "loss": 0.2599,
+      "step": 2585
+    },
+    {
+      "epoch": 0.7465590545506954,
+      "grad_norm": 2.875809907913208,
+      "learning_rate": 2.6676783004552353e-06,
+      "loss": 0.2446,
+      "step": 2590
+    },
+    {
+      "epoch": 0.7480002882467393,
+      "grad_norm": 2.0145938396453857,
+      "learning_rate": 2.652503793626707e-06,
+      "loss": 0.2458,
+      "step": 2595
+    },
+    {
+      "epoch": 0.749441521942783,
+      "grad_norm": 2.050328493118286,
+      "learning_rate": 2.6373292867981793e-06,
+      "loss": 0.2856,
+      "step": 2600
+    },
+    {
+      "epoch": 0.749441521942783,
+      "eval_loss": 0.23749777674674988,
+      "eval_mse": 0.2374977778196335,
+      "eval_runtime": 3.4587,
+      "eval_samples_per_second": 289.127,
+      "eval_steps_per_second": 18.215,
+      "step": 2600
+    },
+    {
+      "epoch": 0.7508827556388268,
+      "grad_norm": 2.5174169540405273,
+      "learning_rate": 2.622154779969651e-06,
+      "loss": 0.2519,
+      "step": 2605
+    },
+    {
+      "epoch": 0.7523239893348707,
+      "grad_norm": 4.114897727966309,
+      "learning_rate": 2.606980273141123e-06,
+      "loss": 0.2592,
+      "step": 2610
+    },
+    {
+      "epoch": 0.7537652230309144,
+      "grad_norm": 1.5561412572860718,
+      "learning_rate": 2.5918057663125952e-06,
+      "loss": 0.2474,
+      "step": 2615
+    },
+    {
+      "epoch": 0.7552064567269583,
+      "grad_norm": 2.4157848358154297,
+      "learning_rate": 2.576631259484067e-06,
+      "loss": 0.2643,
+      "step": 2620
+    },
+    {
+      "epoch": 0.7566476904230021,
+      "grad_norm": 2.7704153060913086,
+      "learning_rate": 2.561456752655539e-06,
+      "loss": 0.2305,
+      "step": 2625
+    },
+    {
+      "epoch": 0.7580889241190459,
+      "grad_norm": 3.1055073738098145,
+      "learning_rate": 2.5462822458270107e-06,
+      "loss": 0.2451,
+      "step": 2630
+    },
+    {
+      "epoch": 0.7595301578150897,
+      "grad_norm": 1.9698563814163208,
+      "learning_rate": 2.531107738998483e-06,
+      "loss": 0.2292,
+      "step": 2635
+    },
+    {
+      "epoch": 0.7609713915111336,
+      "grad_norm": 1.9055688381195068,
+      "learning_rate": 2.515933232169955e-06,
+      "loss": 0.2681,
+      "step": 2640
+    },
+    {
+      "epoch": 0.7624126252071773,
+      "grad_norm": 3.620293617248535,
+      "learning_rate": 2.5007587253414266e-06,
+      "loss": 0.2469,
+      "step": 2645
+    },
+    {
+      "epoch": 0.7638538589032211,
+      "grad_norm": 2.351473569869995,
+      "learning_rate": 2.4855842185128985e-06,
+      "loss": 0.2612,
+      "step": 2650
+    },
+    {
+      "epoch": 0.765295092599265,
+      "grad_norm": 2.991499185562134,
+      "learning_rate": 2.4704097116843703e-06,
+      "loss": 0.2513,
+      "step": 2655
+    },
+    {
+      "epoch": 0.7667363262953087,
+      "grad_norm": 3.2980542182922363,
+      "learning_rate": 2.455235204855842e-06,
+      "loss": 0.2249,
+      "step": 2660
+    },
+    {
+      "epoch": 0.7681775599913526,
+      "grad_norm": 1.7889645099639893,
+      "learning_rate": 2.4400606980273144e-06,
+      "loss": 0.2566,
+      "step": 2665
+    },
+    {
+      "epoch": 0.7696187936873964,
+      "grad_norm": 2.3199574947357178,
+      "learning_rate": 2.424886191198786e-06,
+      "loss": 0.2419,
+      "step": 2670
+    },
+    {
+      "epoch": 0.7710600273834403,
+      "grad_norm": 1.5038098096847534,
+      "learning_rate": 2.409711684370258e-06,
+      "loss": 0.2308,
+      "step": 2675
+    },
+    {
+      "epoch": 0.772501261079484,
+      "grad_norm": 4.429576396942139,
+      "learning_rate": 2.39453717754173e-06,
+      "loss": 0.2702,
+      "step": 2680
+    },
+    {
+      "epoch": 0.7739424947755279,
+      "grad_norm": 3.6350767612457275,
+      "learning_rate": 2.379362670713202e-06,
+      "loss": 0.2343,
+      "step": 2685
+    },
+    {
+      "epoch": 0.7753837284715717,
+      "grad_norm": 2.1402158737182617,
+      "learning_rate": 2.364188163884674e-06,
+      "loss": 0.265,
+      "step": 2690
+    },
+    {
+      "epoch": 0.7768249621676154,
+      "grad_norm": 3.15134334564209,
+      "learning_rate": 2.3490136570561458e-06,
+      "loss": 0.2433,
+      "step": 2695
+    },
+    {
+      "epoch": 0.7782661958636593,
+      "grad_norm": 2.184305429458618,
+      "learning_rate": 2.333839150227618e-06,
+      "loss": 0.2216,
+      "step": 2700
+    },
+    {
+      "epoch": 0.7782661958636593,
+      "eval_loss": 0.2383294701576233,
+      "eval_mse": 0.23832946814969183,
+      "eval_runtime": 3.7811,
+      "eval_samples_per_second": 264.471,
+      "eval_steps_per_second": 16.662,
+      "step": 2700
+    },
+    {
+      "epoch": 0.7797074295597031,
+      "grad_norm": 3.330735206604004,
+      "learning_rate": 2.31866464339909e-06,
+      "loss": 0.2463,
+      "step": 2705
+    },
+    {
+      "epoch": 0.781148663255747,
+      "grad_norm": 3.6282222270965576,
+      "learning_rate": 2.3034901365705617e-06,
+      "loss": 0.2269,
+      "step": 2710
+    },
+    {
+      "epoch": 0.7825898969517907,
+      "grad_norm": 3.381441593170166,
+      "learning_rate": 2.2883156297420335e-06,
+      "loss": 0.2783,
+      "step": 2715
+    },
+    {
+      "epoch": 0.7840311306478346,
+      "grad_norm": 2.2831079959869385,
+      "learning_rate": 2.2731411229135057e-06,
+      "loss": 0.2633,
+      "step": 2720
+    },
+    {
+      "epoch": 0.7854723643438783,
+      "grad_norm": 3.28764271736145,
+      "learning_rate": 2.2579666160849776e-06,
+      "loss": 0.2424,
+      "step": 2725
+    },
+    {
+      "epoch": 0.7869135980399222,
+      "grad_norm": 5.361327648162842,
+      "learning_rate": 2.2427921092564494e-06,
+      "loss": 0.2379,
+      "step": 2730
+    },
+    {
+      "epoch": 0.788354831735966,
+      "grad_norm": 2.6280875205993652,
+      "learning_rate": 2.2276176024279212e-06,
+      "loss": 0.2405,
+      "step": 2735
+    },
+    {
+      "epoch": 0.7897960654320098,
+      "grad_norm": 2.9547958374023438,
+      "learning_rate": 2.212443095599393e-06,
+      "loss": 0.2531,
+      "step": 2740
+    },
+    {
+      "epoch": 0.7912372991280536,
+      "grad_norm": 2.026799201965332,
+      "learning_rate": 2.197268588770865e-06,
+      "loss": 0.2267,
+      "step": 2745
+    },
+    {
+      "epoch": 0.7926785328240974,
+      "grad_norm": 4.866410732269287,
+      "learning_rate": 2.182094081942337e-06,
+      "loss": 0.2465,
+      "step": 2750
+    },
+    {
+      "epoch": 0.7941197665201413,
+      "grad_norm": 1.9076712131500244,
+      "learning_rate": 2.166919575113809e-06,
+      "loss": 0.2622,
+      "step": 2755
+    },
+    {
+      "epoch": 0.795561000216185,
+      "grad_norm": 3.61765193939209,
+      "learning_rate": 2.1517450682852808e-06,
+      "loss": 0.2348,
+      "step": 2760
+    },
+    {
+      "epoch": 0.7970022339122289,
+      "grad_norm": 2.6565895080566406,
+      "learning_rate": 2.1365705614567526e-06,
+      "loss": 0.262,
+      "step": 2765
+    },
+    {
+      "epoch": 0.7984434676082727,
+      "grad_norm": 2.1399002075195312,
+      "learning_rate": 2.121396054628225e-06,
+      "loss": 0.2507,
+      "step": 2770
+    },
+    {
+      "epoch": 0.7998847013043165,
+      "grad_norm": 2.7440185546875,
+      "learning_rate": 2.1062215477996967e-06,
+      "loss": 0.2274,
+      "step": 2775
+    },
+    {
+      "epoch": 0.8013259350003603,
+      "grad_norm": 1.979787826538086,
+      "learning_rate": 2.0910470409711685e-06,
+      "loss": 0.2485,
+      "step": 2780
+    },
+    {
+      "epoch": 0.8027671686964041,
+      "grad_norm": 1.6057958602905273,
+      "learning_rate": 2.0758725341426408e-06,
+      "loss": 0.244,
+      "step": 2785
+    },
+    {
+      "epoch": 0.804208402392448,
+      "grad_norm": 1.937915563583374,
+      "learning_rate": 2.0606980273141126e-06,
+      "loss": 0.2574,
+      "step": 2790
+    },
+    {
+      "epoch": 0.8056496360884917,
+      "grad_norm": 2.035689115524292,
+      "learning_rate": 2.0455235204855844e-06,
+      "loss": 0.2744,
+      "step": 2795
+    },
+    {
+      "epoch": 0.8070908697845356,
+      "grad_norm": 2.3180646896362305,
+      "learning_rate": 2.0303490136570563e-06,
+      "loss": 0.255,
+      "step": 2800
+    },
+    {
+      "epoch": 0.8070908697845356,
+      "eval_loss": 0.2366907298564911,
+      "eval_mse": 0.23669074228499085,
+      "eval_runtime": 3.4073,
+      "eval_samples_per_second": 293.49,
+      "eval_steps_per_second": 18.49,
+      "step": 2800
+    },
+    {
+      "epoch": 0.8085321034805794,
+      "grad_norm": 1.9396241903305054,
+      "learning_rate": 2.0151745068285285e-06,
+      "loss": 0.2309,
+      "step": 2805
+    },
+    {
+      "epoch": 0.8099733371766232,
+      "grad_norm": 4.935720443725586,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 0.2523,
+      "step": 2810
+    },
+    {
+      "epoch": 0.811414570872667,
+      "grad_norm": 2.1301615238189697,
+      "learning_rate": 1.984825493171472e-06,
+      "loss": 0.2448,
+      "step": 2815
+    },
+    {
+      "epoch": 0.8128558045687109,
+      "grad_norm": 2.9700865745544434,
+      "learning_rate": 1.969650986342944e-06,
+      "loss": 0.239,
+      "step": 2820
+    },
+    {
+      "epoch": 0.8142970382647546,
+      "grad_norm": 2.1401546001434326,
+      "learning_rate": 1.954476479514416e-06,
+      "loss": 0.2349,
+      "step": 2825
+    },
+    {
+      "epoch": 0.8157382719607984,
+      "grad_norm": 3.954643726348877,
+      "learning_rate": 1.9393019726858876e-06,
+      "loss": 0.2609,
+      "step": 2830
+    },
+    {
+      "epoch": 0.8171795056568423,
+      "grad_norm": 1.8797838687896729,
+      "learning_rate": 1.92412746585736e-06,
+      "loss": 0.2492,
+      "step": 2835
+    },
+    {
+      "epoch": 0.818620739352886,
+      "grad_norm": 2.2840428352355957,
+      "learning_rate": 1.9089529590288317e-06,
+      "loss": 0.2382,
+      "step": 2840
+    },
+    {
+      "epoch": 0.8200619730489299,
+      "grad_norm": 1.9538135528564453,
+      "learning_rate": 1.8937784522003035e-06,
+      "loss": 0.2307,
+      "step": 2845
+    },
+    {
+      "epoch": 0.8215032067449737,
+      "grad_norm": 3.3001744747161865,
+      "learning_rate": 1.8786039453717756e-06,
+      "loss": 0.2311,
+      "step": 2850
+    },
+    {
+      "epoch": 0.8229444404410176,
+      "grad_norm": 2.6055421829223633,
+      "learning_rate": 1.8634294385432474e-06,
+      "loss": 0.2732,
+      "step": 2855
+    },
+    {
+      "epoch": 0.8243856741370613,
+      "grad_norm": 4.642452716827393,
+      "learning_rate": 1.8482549317147195e-06,
+      "loss": 0.2642,
+      "step": 2860
+    },
+    {
+      "epoch": 0.8258269078331051,
+      "grad_norm": 1.3938387632369995,
+      "learning_rate": 1.8330804248861913e-06,
+      "loss": 0.2478,
+      "step": 2865
+    },
+    {
+      "epoch": 0.827268141529149,
+      "grad_norm": 2.2278647422790527,
+      "learning_rate": 1.8179059180576633e-06,
+      "loss": 0.2533,
+      "step": 2870
+    },
+    {
+      "epoch": 0.8287093752251927,
+      "grad_norm": 4.208284378051758,
+      "learning_rate": 1.8027314112291352e-06,
+      "loss": 0.2449,
+      "step": 2875
+    },
+    {
+      "epoch": 0.8301506089212366,
+      "grad_norm": 1.990187644958496,
+      "learning_rate": 1.7875569044006072e-06,
+      "loss": 0.2511,
+      "step": 2880
+    },
+    {
+      "epoch": 0.8315918426172804,
+      "grad_norm": 1.8299331665039062,
+      "learning_rate": 1.7723823975720792e-06,
+      "loss": 0.2484,
+      "step": 2885
+    },
+    {
+      "epoch": 0.8330330763133242,
+      "grad_norm": 2.172791004180908,
+      "learning_rate": 1.757207890743551e-06,
+      "loss": 0.2365,
+      "step": 2890
+    },
+    {
+      "epoch": 0.834474310009368,
+      "grad_norm": 1.920158863067627,
+      "learning_rate": 1.742033383915023e-06,
+      "loss": 0.2335,
+      "step": 2895
+    },
+    {
+      "epoch": 0.8359155437054119,
+      "grad_norm": 1.844710111618042,
+      "learning_rate": 1.726858877086495e-06,
+      "loss": 0.2406,
+      "step": 2900
+    },
+    {
+      "epoch": 0.8359155437054119,
+      "eval_loss": 0.2344696968793869,
+      "eval_mse": 0.2344697018004954,
+      "eval_runtime": 3.4794,
+      "eval_samples_per_second": 287.41,
+      "eval_steps_per_second": 18.107,
+      "step": 2900
+    },
+    {
+      "epoch": 0.8373567774014556,
+      "grad_norm": 1.9122151136398315,
+      "learning_rate": 1.7116843702579665e-06,
+      "loss": 0.2462,
+      "step": 2905
+    },
+    {
+      "epoch": 0.8387980110974994,
+      "grad_norm": 1.5618327856063843,
+      "learning_rate": 1.6965098634294386e-06,
+      "loss": 0.2413,
+      "step": 2910
+    },
+    {
+      "epoch": 0.8402392447935433,
+      "grad_norm": 3.7296929359436035,
+      "learning_rate": 1.6813353566009106e-06,
+      "loss": 0.2405,
+      "step": 2915
+    },
+    {
+      "epoch": 0.841680478489587,
+      "grad_norm": 3.897550106048584,
+      "learning_rate": 1.6661608497723824e-06,
+      "loss": 0.2833,
+      "step": 2920
+    },
+    {
+      "epoch": 0.8431217121856309,
+      "grad_norm": 2.775019645690918,
+      "learning_rate": 1.6509863429438545e-06,
+      "loss": 0.2083,
+      "step": 2925
+    },
+    {
+      "epoch": 0.8445629458816747,
+      "grad_norm": 1.737260341644287,
+      "learning_rate": 1.6358118361153263e-06,
+      "loss": 0.2535,
+      "step": 2930
+    },
+    {
+      "epoch": 0.8460041795777186,
+      "grad_norm": 2.2079930305480957,
+      "learning_rate": 1.6206373292867984e-06,
+      "loss": 0.2649,
+      "step": 2935
+    },
+    {
+      "epoch": 0.8474454132737623,
+      "grad_norm": 2.6033072471618652,
+      "learning_rate": 1.6054628224582702e-06,
+      "loss": 0.2324,
+      "step": 2940
+    },
+    {
+      "epoch": 0.8488866469698062,
+      "grad_norm": 2.3140974044799805,
+      "learning_rate": 1.5902883156297422e-06,
+      "loss": 0.2314,
+      "step": 2945
+    },
+    {
+      "epoch": 0.85032788066585,
+      "grad_norm": 1.9832013845443726,
+      "learning_rate": 1.575113808801214e-06,
+      "loss": 0.2709,
+      "step": 2950
+    },
+    {
+      "epoch": 0.8517691143618937,
+      "grad_norm": 2.839756727218628,
+      "learning_rate": 1.559939301972686e-06,
+      "loss": 0.2349,
+      "step": 2955
+    },
+    {
+      "epoch": 0.8532103480579376,
+      "grad_norm": 2.0620322227478027,
+      "learning_rate": 1.544764795144158e-06,
+      "loss": 0.2697,
+      "step": 2960
+    },
+    {
+      "epoch": 0.8546515817539814,
+      "grad_norm": 2.1879138946533203,
+      "learning_rate": 1.52959028831563e-06,
+      "loss": 0.2415,
+      "step": 2965
+    },
+    {
+      "epoch": 0.8560928154500252,
+      "grad_norm": 2.201324224472046,
+      "learning_rate": 1.514415781487102e-06,
+      "loss": 0.2469,
+      "step": 2970
+    },
+    {
+      "epoch": 0.857534049146069,
+      "grad_norm": 1.4632495641708374,
+      "learning_rate": 1.4992412746585738e-06,
+      "loss": 0.2194,
+      "step": 2975
+    },
+    {
+      "epoch": 0.8589752828421129,
+      "grad_norm": 1.9901318550109863,
+      "learning_rate": 1.4840667678300459e-06,
+      "loss": 0.2597,
+      "step": 2980
+    },
+    {
+      "epoch": 0.8604165165381567,
+      "grad_norm": 1.8942216634750366,
+      "learning_rate": 1.4688922610015175e-06,
+      "loss": 0.2481,
+      "step": 2985
+    },
+    {
+      "epoch": 0.8618577502342005,
+      "grad_norm": 1.9384193420410156,
+      "learning_rate": 1.4537177541729893e-06,
+      "loss": 0.2399,
+      "step": 2990
+    },
+    {
+      "epoch": 0.8632989839302443,
+      "grad_norm": 1.9823272228240967,
+      "learning_rate": 1.4385432473444613e-06,
+      "loss": 0.2001,
+      "step": 2995
+    },
+    {
+      "epoch": 0.864740217626288,
+      "grad_norm": 4.13219690322876,
+      "learning_rate": 1.4233687405159332e-06,
+      "loss": 0.2388,
+      "step": 3000
+    },
+    {
+      "epoch": 0.864740217626288,
+      "eval_loss": 0.22815744578838348,
+      "eval_mse": 0.22815744203701616,
+      "eval_runtime": 3.6856,
+      "eval_samples_per_second": 271.329,
+      "eval_steps_per_second": 17.094,
+      "step": 3000
+    },
+    {
+      "epoch": 0.8661814513223319,
+      "grad_norm": 2.3792169094085693,
+      "learning_rate": 1.4081942336874052e-06,
+      "loss": 0.2651,
+      "step": 3005
+    },
+    {
+      "epoch": 0.8676226850183757,
+      "grad_norm": 1.8244715929031372,
+      "learning_rate": 1.3930197268588772e-06,
+      "loss": 0.2437,
+      "step": 3010
+    },
+    {
+      "epoch": 0.8690639187144196,
+      "grad_norm": 2.443810224533081,
+      "learning_rate": 1.377845220030349e-06,
+      "loss": 0.2527,
+      "step": 3015
+    },
+    {
+      "epoch": 0.8705051524104633,
+      "grad_norm": 2.1576616764068604,
+      "learning_rate": 1.3626707132018211e-06,
+      "loss": 0.2713,
+      "step": 3020
+    },
+    {
+      "epoch": 0.8719463861065072,
+      "grad_norm": 2.210902452468872,
+      "learning_rate": 1.347496206373293e-06,
+      "loss": 0.238,
+      "step": 3025
+    },
+    {
+      "epoch": 0.873387619802551,
+      "grad_norm": 2.990591049194336,
+      "learning_rate": 1.332321699544765e-06,
+      "loss": 0.2217,
+      "step": 3030
+    },
+    {
+      "epoch": 0.8748288534985948,
+      "grad_norm": 1.5626643896102905,
+      "learning_rate": 1.3171471927162368e-06,
+      "loss": 0.2387,
+      "step": 3035
+    },
+    {
+      "epoch": 0.8762700871946386,
+      "grad_norm": 1.5288246870040894,
+      "learning_rate": 1.3019726858877088e-06,
+      "loss": 0.2163,
+      "step": 3040
+    },
+    {
+      "epoch": 0.8777113208906824,
+      "grad_norm": 3.954557180404663,
+      "learning_rate": 1.2867981790591807e-06,
+      "loss": 0.252,
+      "step": 3045
+    },
+    {
+      "epoch": 0.8791525545867263,
+      "grad_norm": 5.214181900024414,
+      "learning_rate": 1.2716236722306527e-06,
+      "loss": 0.2403,
+      "step": 3050
+    },
+    {
+      "epoch": 0.88059378828277,
+      "grad_norm": 2.6781516075134277,
+      "learning_rate": 1.2564491654021245e-06,
+      "loss": 0.2493,
+      "step": 3055
+    },
+    {
+      "epoch": 0.8820350219788139,
+      "grad_norm": 4.418315410614014,
+      "learning_rate": 1.2412746585735964e-06,
+      "loss": 0.2694,
+      "step": 3060
+    },
+    {
+      "epoch": 0.8834762556748577,
+      "grad_norm": 2.1344754695892334,
+      "learning_rate": 1.2261001517450684e-06,
+      "loss": 0.2676,
+      "step": 3065
+    },
+    {
+      "epoch": 0.8849174893709015,
+      "grad_norm": 2.3457322120666504,
+      "learning_rate": 1.2109256449165402e-06,
+      "loss": 0.2516,
+      "step": 3070
+    },
+    {
+      "epoch": 0.8863587230669453,
+      "grad_norm": 1.9455342292785645,
+      "learning_rate": 1.1957511380880123e-06,
+      "loss": 0.2513,
+      "step": 3075
+    },
+    {
+      "epoch": 0.8877999567629892,
+      "grad_norm": 3.034874439239502,
+      "learning_rate": 1.1805766312594843e-06,
+      "loss": 0.2651,
+      "step": 3080
+    },
+    {
+      "epoch": 0.8892411904590329,
+      "grad_norm": 3.1415164470672607,
+      "learning_rate": 1.1654021244309561e-06,
+      "loss": 0.2494,
+      "step": 3085
+    },
+    {
+      "epoch": 0.8906824241550767,
+      "grad_norm": 2.136166572570801,
+      "learning_rate": 1.150227617602428e-06,
+      "loss": 0.2577,
+      "step": 3090
+    },
+    {
+      "epoch": 0.8921236578511206,
+      "grad_norm": 2.6537928581237793,
+      "learning_rate": 1.1350531107738998e-06,
+      "loss": 0.2432,
+      "step": 3095
+    },
+    {
+      "epoch": 0.8935648915471643,
+      "grad_norm": 1.8011763095855713,
+      "learning_rate": 1.1198786039453718e-06,
+      "loss": 0.2571,
+      "step": 3100
+    },
+    {
+      "epoch": 0.8935648915471643,
+      "eval_loss": 0.23310115933418274,
+      "eval_mse": 0.23310116421058774,
+      "eval_runtime": 3.4187,
+      "eval_samples_per_second": 292.511,
+      "eval_steps_per_second": 18.428,
+      "step": 3100
+    },
+    {
+      "epoch": 0.8950061252432082,
+      "grad_norm": 1.5741316080093384,
+      "learning_rate": 1.1047040971168439e-06,
+      "loss": 0.223,
+      "step": 3105
+    },
+    {
+      "epoch": 0.896447358939252,
+      "grad_norm": 1.5244059562683105,
+      "learning_rate": 1.0895295902883157e-06,
+      "loss": 0.2428,
+      "step": 3110
+    },
+    {
+      "epoch": 0.8978885926352959,
+      "grad_norm": 1.8226120471954346,
+      "learning_rate": 1.0743550834597877e-06,
+      "loss": 0.2548,
+      "step": 3115
+    },
+    {
+      "epoch": 0.8993298263313396,
+      "grad_norm": 2.6525444984436035,
+      "learning_rate": 1.0591805766312596e-06,
+      "loss": 0.2354,
+      "step": 3120
+    },
+    {
+      "epoch": 0.9007710600273834,
+      "grad_norm": 3.4096426963806152,
+      "learning_rate": 1.0440060698027316e-06,
+      "loss": 0.2535,
+      "step": 3125
+    },
+    {
+      "epoch": 0.9022122937234273,
+      "grad_norm": 1.6631042957305908,
+      "learning_rate": 1.0288315629742034e-06,
+      "loss": 0.2549,
+      "step": 3130
+    },
+    {
+      "epoch": 0.903653527419471,
+      "grad_norm": 4.044405460357666,
+      "learning_rate": 1.0136570561456753e-06,
+      "loss": 0.2438,
+      "step": 3135
+    },
+    {
+      "epoch": 0.9050947611155149,
+      "grad_norm": 2.1548964977264404,
+      "learning_rate": 9.984825493171473e-07,
+      "loss": 0.2171,
+      "step": 3140
+    },
+    {
+      "epoch": 0.9065359948115587,
+      "grad_norm": 2.5267910957336426,
+      "learning_rate": 9.833080424886191e-07,
+      "loss": 0.2341,
+      "step": 3145
+    },
+    {
+      "epoch": 0.9079772285076025,
+      "grad_norm": 1.7925020456314087,
+      "learning_rate": 9.681335356600912e-07,
+      "loss": 0.2653,
+      "step": 3150
+    },
+    {
+      "epoch": 0.9094184622036463,
+      "grad_norm": 2.5686533451080322,
+      "learning_rate": 9.529590288315631e-07,
+      "loss": 0.2671,
+      "step": 3155
+    },
+    {
+      "epoch": 0.9108596958996902,
+      "grad_norm": 1.508566975593567,
+      "learning_rate": 9.37784522003035e-07,
+      "loss": 0.2554,
+      "step": 3160
+    },
+    {
+      "epoch": 0.912300929595734,
+      "grad_norm": 2.414174795150757,
+      "learning_rate": 9.22610015174507e-07,
+      "loss": 0.242,
+      "step": 3165
+    },
+    {
+      "epoch": 0.9137421632917777,
+      "grad_norm": 2.007922887802124,
+      "learning_rate": 9.074355083459788e-07,
+      "loss": 0.2442,
+      "step": 3170
+    },
+    {
+      "epoch": 0.9151833969878216,
+      "grad_norm": 2.534485340118408,
+      "learning_rate": 8.922610015174507e-07,
+      "loss": 0.2478,
+      "step": 3175
+    },
+    {
+      "epoch": 0.9166246306838653,
+      "grad_norm": 2.9684510231018066,
+      "learning_rate": 8.770864946889227e-07,
+      "loss": 0.2607,
+      "step": 3180
+    },
+    {
+      "epoch": 0.9180658643799092,
+      "grad_norm": 2.6235086917877197,
+      "learning_rate": 8.619119878603946e-07,
+      "loss": 0.2442,
+      "step": 3185
+    },
+    {
+      "epoch": 0.919507098075953,
+      "grad_norm": 4.0555644035339355,
+      "learning_rate": 8.467374810318665e-07,
+      "loss": 0.2722,
+      "step": 3190
+    },
+    {
+      "epoch": 0.9209483317719969,
+      "grad_norm": 3.3704187870025635,
+      "learning_rate": 8.315629742033385e-07,
+      "loss": 0.2483,
+      "step": 3195
+    },
+    {
+      "epoch": 0.9223895654680406,
+      "grad_norm": 3.073582649230957,
+      "learning_rate": 8.163884673748104e-07,
+      "loss": 0.2672,
+      "step": 3200
+    },
+    {
+      "epoch": 0.9223895654680406,
+      "eval_loss": 0.2335718721151352,
+      "eval_mse": 0.23357188405096532,
+      "eval_runtime": 3.4739,
+      "eval_samples_per_second": 287.861,
+      "eval_steps_per_second": 18.135,
+      "step": 3200
+    },
+    {
+      "epoch": 0.9238307991640845,
+      "grad_norm": 2.310436248779297,
+      "learning_rate": 8.012139605462823e-07,
+      "loss": 0.2489,
+      "step": 3205
+    },
+    {
+      "epoch": 0.9252720328601283,
+      "grad_norm": 1.7982414960861206,
+      "learning_rate": 7.860394537177542e-07,
+      "loss": 0.2421,
+      "step": 3210
+    },
+    {
+      "epoch": 0.926713266556172,
+      "grad_norm": 1.9083003997802734,
+      "learning_rate": 7.708649468892261e-07,
+      "loss": 0.2398,
+      "step": 3215
+    },
+    {
+      "epoch": 0.9281545002522159,
+      "grad_norm": 3.1134302616119385,
+      "learning_rate": 7.55690440060698e-07,
+      "loss": 0.255,
+      "step": 3220
+    },
+    {
+      "epoch": 0.9295957339482597,
+      "grad_norm": 2.5554730892181396,
+      "learning_rate": 7.4051593323217e-07,
+      "loss": 0.255,
+      "step": 3225
+    },
+    {
+      "epoch": 0.9310369676443035,
+      "grad_norm": 2.0975911617279053,
+      "learning_rate": 7.253414264036419e-07,
+      "loss": 0.2405,
+      "step": 3230
+    },
+    {
+      "epoch": 0.9324782013403473,
+      "grad_norm": 2.0389764308929443,
+      "learning_rate": 7.101669195751138e-07,
+      "loss": 0.2298,
+      "step": 3235
+    },
+    {
+      "epoch": 0.9339194350363912,
+      "grad_norm": 2.1120331287384033,
+      "learning_rate": 6.949924127465859e-07,
+      "loss": 0.2455,
+      "step": 3240
+    },
+    {
+      "epoch": 0.935360668732435,
+      "grad_norm": 1.8716710805892944,
+      "learning_rate": 6.798179059180578e-07,
+      "loss": 0.2502,
+      "step": 3245
+    },
+    {
+      "epoch": 0.9368019024284788,
+      "grad_norm": 3.4095449447631836,
+      "learning_rate": 6.646433990895297e-07,
+      "loss": 0.2431,
+      "step": 3250
+    },
+    {
+      "epoch": 0.9382431361245226,
+      "grad_norm": 1.740820288658142,
+      "learning_rate": 6.494688922610016e-07,
+      "loss": 0.2493,
+      "step": 3255
+    },
+    {
+      "epoch": 0.9396843698205664,
+      "grad_norm": 1.8503363132476807,
+      "learning_rate": 6.342943854324735e-07,
+      "loss": 0.2537,
+      "step": 3260
+    },
+    {
+      "epoch": 0.9411256035166102,
+      "grad_norm": 2.1327052116394043,
+      "learning_rate": 6.191198786039454e-07,
+      "loss": 0.2346,
+      "step": 3265
+    },
+    {
+      "epoch": 0.942566837212654,
+      "grad_norm": 2.9109222888946533,
+      "learning_rate": 6.039453717754174e-07,
+      "loss": 0.2415,
+      "step": 3270
+    },
+    {
+      "epoch": 0.9440080709086979,
+      "grad_norm": 1.9122915267944336,
+      "learning_rate": 5.887708649468893e-07,
+      "loss": 0.2401,
+      "step": 3275
+    },
+    {
+      "epoch": 0.9454493046047416,
+      "grad_norm": 2.2561800479888916,
+      "learning_rate": 5.735963581183612e-07,
+      "loss": 0.2331,
+      "step": 3280
+    },
+    {
+      "epoch": 0.9468905383007855,
+      "grad_norm": 2.547837972640991,
+      "learning_rate": 5.584218512898331e-07,
+      "loss": 0.2373,
+      "step": 3285
+    },
+    {
+      "epoch": 0.9483317719968293,
+      "grad_norm": 2.524578332901001,
+      "learning_rate": 5.43247344461305e-07,
+      "loss": 0.2225,
+      "step": 3290
+    },
+    {
+      "epoch": 0.9497730056928732,
+      "grad_norm": 1.7476263046264648,
+      "learning_rate": 5.28072837632777e-07,
+      "loss": 0.2339,
+      "step": 3295
+    },
+    {
+      "epoch": 0.9512142393889169,
+      "grad_norm": 1.938265323638916,
+      "learning_rate": 5.12898330804249e-07,
+      "loss": 0.2375,
+      "step": 3300
+    },
+    {
+      "epoch": 0.9512142393889169,
+      "eval_loss": 0.23372192680835724,
+      "eval_mse": 0.23372193356230855,
+      "eval_runtime": 3.4568,
+      "eval_samples_per_second": 289.284,
+      "eval_steps_per_second": 18.225,
+      "step": 3300
+    },
+    {
+      "epoch": 0.9526554730849607,
+      "grad_norm": 2.301506519317627,
+      "learning_rate": 4.977238239757208e-07,
+      "loss": 0.2699,
+      "step": 3305
+    },
+    {
+      "epoch": 0.9540967067810046,
+      "grad_norm": 1.916329026222229,
+      "learning_rate": 4.825493171471927e-07,
+      "loss": 0.2101,
+      "step": 3310
+    },
+    {
+      "epoch": 0.9555379404770483,
+      "grad_norm": 2.0248494148254395,
+      "learning_rate": 4.673748103186647e-07,
+      "loss": 0.2202,
+      "step": 3315
+    },
+    {
+      "epoch": 0.9569791741730922,
+      "grad_norm": 3.3901896476745605,
+      "learning_rate": 4.5220030349013665e-07,
+      "loss": 0.2336,
+      "step": 3320
+    },
+    {
+      "epoch": 0.958420407869136,
+      "grad_norm": 1.915697455406189,
+      "learning_rate": 4.3702579666160853e-07,
+      "loss": 0.2444,
+      "step": 3325
+    },
+    {
+      "epoch": 0.9598616415651798,
+      "grad_norm": 3.5947072505950928,
+      "learning_rate": 4.2185128983308046e-07,
+      "loss": 0.2515,
+      "step": 3330
+    },
+    {
+      "epoch": 0.9613028752612236,
+      "grad_norm": 3.344456911087036,
+      "learning_rate": 4.066767830045524e-07,
+      "loss": 0.2347,
+      "step": 3335
+    },
+    {
+      "epoch": 0.9627441089572675,
+      "grad_norm": 2.0030887126922607,
+      "learning_rate": 3.9150227617602433e-07,
+      "loss": 0.2323,
+      "step": 3340
+    },
+    {
+      "epoch": 0.9641853426533112,
+      "grad_norm": 1.546491265296936,
+      "learning_rate": 3.763277693474962e-07,
+      "loss": 0.247,
+      "step": 3345
+    },
+    {
+      "epoch": 0.965626576349355,
+      "grad_norm": 1.7898391485214233,
+      "learning_rate": 3.6115326251896814e-07,
+      "loss": 0.2441,
+      "step": 3350
+    },
+    {
+      "epoch": 0.9670678100453989,
+      "grad_norm": 1.4800883531570435,
+      "learning_rate": 3.459787556904401e-07,
+      "loss": 0.2069,
+      "step": 3355
+    },
+    {
+      "epoch": 0.9685090437414426,
+      "grad_norm": 5.112166881561279,
+      "learning_rate": 3.3080424886191206e-07,
+      "loss": 0.2405,
+      "step": 3360
+    },
+    {
+      "epoch": 0.9699502774374865,
+      "grad_norm": 2.2341766357421875,
+      "learning_rate": 3.156297420333839e-07,
+      "loss": 0.2217,
+      "step": 3365
+    },
+    {
+      "epoch": 0.9713915111335303,
+      "grad_norm": 2.1282143592834473,
+      "learning_rate": 3.004552352048559e-07,
+      "loss": 0.2317,
+      "step": 3370
+    },
+    {
+      "epoch": 0.9728327448295742,
+      "grad_norm": 4.700996398925781,
+      "learning_rate": 2.852807283763278e-07,
+      "loss": 0.2458,
+      "step": 3375
+    },
+    {
+      "epoch": 0.9742739785256179,
+      "grad_norm": 2.893071413040161,
+      "learning_rate": 2.701062215477997e-07,
+      "loss": 0.2373,
+      "step": 3380
+    },
+    {
+      "epoch": 0.9757152122216617,
+      "grad_norm": 2.4132773876190186,
+      "learning_rate": 2.549317147192716e-07,
+      "loss": 0.2587,
+      "step": 3385
+    },
+    {
+      "epoch": 0.9771564459177056,
+      "grad_norm": 3.645531415939331,
+      "learning_rate": 2.3975720789074356e-07,
+      "loss": 0.2523,
+      "step": 3390
+    },
+    {
+      "epoch": 0.9785976796137493,
+      "grad_norm": 3.378278970718384,
+      "learning_rate": 2.245827010622155e-07,
+      "loss": 0.2336,
+      "step": 3395
+    },
+    {
+      "epoch": 0.9800389133097932,
+      "grad_norm": 2.213209867477417,
+      "learning_rate": 2.0940819423368745e-07,
+      "loss": 0.2423,
+      "step": 3400
+    },
+    {
+      "epoch": 0.9800389133097932,
+      "eval_loss": 0.23236581683158875,
+      "eval_mse": 0.2323658305592835,
+      "eval_runtime": 3.516,
+      "eval_samples_per_second": 284.418,
+      "eval_steps_per_second": 17.918,
+      "step": 3400
+    },
+    {
+      "epoch": 0.981480147005837,
+      "grad_norm": 1.9079331159591675,
+      "learning_rate": 1.9423368740515936e-07,
+      "loss": 0.2418,
+      "step": 3405
+    },
+    {
+      "epoch": 0.9829213807018808,
+      "grad_norm": 1.9052362442016602,
+      "learning_rate": 1.790591805766313e-07,
+      "loss": 0.2642,
+      "step": 3410
+    },
+    {
+      "epoch": 0.9843626143979246,
+      "grad_norm": 1.942834496498108,
+      "learning_rate": 1.638846737481032e-07,
+      "loss": 0.245,
+      "step": 3415
+    },
+    {
+      "epoch": 0.9858038480939685,
+      "grad_norm": 3.9796624183654785,
+      "learning_rate": 1.4871016691957513e-07,
+      "loss": 0.2705,
+      "step": 3420
+    },
+    {
+      "epoch": 0.9872450817900122,
+      "grad_norm": 1.7400703430175781,
+      "learning_rate": 1.3353566009104704e-07,
+      "loss": 0.2316,
+      "step": 3425
+    },
+    {
+      "epoch": 0.988686315486056,
+      "grad_norm": 2.3475584983825684,
+      "learning_rate": 1.1836115326251897e-07,
+      "loss": 0.2353,
+      "step": 3430
+    },
+    {
+      "epoch": 0.9901275491820999,
+      "grad_norm": 1.7061249017715454,
+      "learning_rate": 1.031866464339909e-07,
+      "loss": 0.2384,
+      "step": 3435
+    },
+    {
+      "epoch": 0.9915687828781437,
+      "grad_norm": 1.9734495878219604,
+      "learning_rate": 8.801213960546283e-08,
+      "loss": 0.2717,
+      "step": 3440
+    },
+    {
+      "epoch": 0.9930100165741875,
+      "grad_norm": 1.485177993774414,
+      "learning_rate": 7.283763277693476e-08,
+      "loss": 0.2242,
+      "step": 3445
+    },
+    {
+      "epoch": 0.9944512502702313,
+      "grad_norm": 2.0481748580932617,
+      "learning_rate": 5.7663125948406686e-08,
+      "loss": 0.2768,
+      "step": 3450
+    },
+    {
+      "epoch": 0.9958924839662752,
+      "grad_norm": 2.793379545211792,
+      "learning_rate": 4.248861911987861e-08,
+      "loss": 0.225,
+      "step": 3455
+    },
+    {
+      "epoch": 0.9973337176623189,
+      "grad_norm": 2.208509922027588,
+      "learning_rate": 2.7314112291350533e-08,
+      "loss": 0.2576,
+      "step": 3460
+    },
+    {
+      "epoch": 0.9987749513583628,
+      "grad_norm": 2.3073906898498535,
+      "learning_rate": 1.213960546282246e-08,
+      "loss": 0.2368,
+      "step": 3465
+    },
+    {
+      "epoch": 0.9999279383151978,
+      "step": 3469,
+      "total_flos": 5.881871499303322e+16,
+      "train_loss": 0.29927895559842593,
+      "train_runtime": 2787.8171,
+      "train_samples_per_second": 159.283,
+      "train_steps_per_second": 1.244
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 3469,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "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": 5.881871499303322e+16,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}