diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,6868 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 15.0,
+  "eval_steps": 500,
+  "global_step": 4785,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.003134796238244514,
+      "grad_norm": 1200.0,
+      "learning_rate": 4.175365344467641e-07,
+      "loss": 56.0196,
+      "step": 1
+    },
+    {
+      "epoch": 0.01567398119122257,
+      "grad_norm": 1104.0,
+      "learning_rate": 2.0876826722338207e-06,
+      "loss": 57.1065,
+      "step": 5
+    },
+    {
+      "epoch": 0.03134796238244514,
+      "grad_norm": 780.0,
+      "learning_rate": 4.175365344467641e-06,
+      "loss": 53.982,
+      "step": 10
+    },
+    {
+      "epoch": 0.047021943573667714,
+      "grad_norm": 378.0,
+      "learning_rate": 6.2630480167014616e-06,
+      "loss": 39.7621,
+      "step": 15
+    },
+    {
+      "epoch": 0.06269592476489028,
+      "grad_norm": 119.0,
+      "learning_rate": 8.350730688935283e-06,
+      "loss": 31.2881,
+      "step": 20
+    },
+    {
+      "epoch": 0.07836990595611286,
+      "grad_norm": 59.0,
+      "learning_rate": 1.0438413361169103e-05,
+      "loss": 29.3676,
+      "step": 25
+    },
+    {
+      "epoch": 0.09404388714733543,
+      "grad_norm": 27.75,
+      "learning_rate": 1.2526096033402923e-05,
+      "loss": 27.221,
+      "step": 30
+    },
+    {
+      "epoch": 0.109717868338558,
+      "grad_norm": 17.0,
+      "learning_rate": 1.4613778705636743e-05,
+      "loss": 25.3228,
+      "step": 35
+    },
+    {
+      "epoch": 0.12539184952978055,
+      "grad_norm": 11.375,
+      "learning_rate": 1.6701461377870565e-05,
+      "loss": 24.7964,
+      "step": 40
+    },
+    {
+      "epoch": 0.14106583072100312,
+      "grad_norm": 14.6875,
+      "learning_rate": 1.8789144050104384e-05,
+      "loss": 23.6672,
+      "step": 45
+    },
+    {
+      "epoch": 0.15673981191222572,
+      "grad_norm": 23.5,
+      "learning_rate": 2.0876826722338206e-05,
+      "loss": 22.938,
+      "step": 50
+    },
+    {
+      "epoch": 0.1724137931034483,
+      "grad_norm": 46.75,
+      "learning_rate": 2.2964509394572024e-05,
+      "loss": 20.5447,
+      "step": 55
+    },
+    {
+      "epoch": 0.18808777429467086,
+      "grad_norm": 102.5,
+      "learning_rate": 2.5052192066805846e-05,
+      "loss": 15.6089,
+      "step": 60
+    },
+    {
+      "epoch": 0.20376175548589343,
+      "grad_norm": 21.875,
+      "learning_rate": 2.7139874739039668e-05,
+      "loss": 6.57,
+      "step": 65
+    },
+    {
+      "epoch": 0.219435736677116,
+      "grad_norm": 7.15625,
+      "learning_rate": 2.9227557411273487e-05,
+      "loss": 2.9537,
+      "step": 70
+    },
+    {
+      "epoch": 0.23510971786833856,
+      "grad_norm": 3.953125,
+      "learning_rate": 3.131524008350731e-05,
+      "loss": 2.4063,
+      "step": 75
+    },
+    {
+      "epoch": 0.2507836990595611,
+      "grad_norm": 3.234375,
+      "learning_rate": 3.340292275574113e-05,
+      "loss": 2.1568,
+      "step": 80
+    },
+    {
+      "epoch": 0.2664576802507837,
+      "grad_norm": 4.75,
+      "learning_rate": 3.5490605427974946e-05,
+      "loss": 1.9549,
+      "step": 85
+    },
+    {
+      "epoch": 0.28213166144200624,
+      "grad_norm": 5.90625,
+      "learning_rate": 3.757828810020877e-05,
+      "loss": 1.8178,
+      "step": 90
+    },
+    {
+      "epoch": 0.29780564263322884,
+      "grad_norm": 16.625,
+      "learning_rate": 3.966597077244259e-05,
+      "loss": 1.7357,
+      "step": 95
+    },
+    {
+      "epoch": 0.31347962382445144,
+      "grad_norm": 10.625,
+      "learning_rate": 4.175365344467641e-05,
+      "loss": 1.6526,
+      "step": 100
+    },
+    {
+      "epoch": 0.329153605015674,
+      "grad_norm": 16.875,
+      "learning_rate": 4.3841336116910233e-05,
+      "loss": 1.6119,
+      "step": 105
+    },
+    {
+      "epoch": 0.3448275862068966,
+      "grad_norm": 15.9375,
+      "learning_rate": 4.592901878914405e-05,
+      "loss": 1.5619,
+      "step": 110
+    },
+    {
+      "epoch": 0.3605015673981191,
+      "grad_norm": 6.25,
+      "learning_rate": 4.801670146137787e-05,
+      "loss": 1.553,
+      "step": 115
+    },
+    {
+      "epoch": 0.3761755485893417,
+      "grad_norm": 16.875,
+      "learning_rate": 5.010438413361169e-05,
+      "loss": 1.5499,
+      "step": 120
+    },
+    {
+      "epoch": 0.39184952978056425,
+      "grad_norm": 18.875,
+      "learning_rate": 5.219206680584552e-05,
+      "loss": 1.5358,
+      "step": 125
+    },
+    {
+      "epoch": 0.40752351097178685,
+      "grad_norm": 10.5,
+      "learning_rate": 5.4279749478079336e-05,
+      "loss": 1.5022,
+      "step": 130
+    },
+    {
+      "epoch": 0.4231974921630094,
+      "grad_norm": 8.6875,
+      "learning_rate": 5.636743215031316e-05,
+      "loss": 1.4674,
+      "step": 135
+    },
+    {
+      "epoch": 0.438871473354232,
+      "grad_norm": 3.25,
+      "learning_rate": 5.8455114822546973e-05,
+      "loss": 1.4238,
+      "step": 140
+    },
+    {
+      "epoch": 0.45454545454545453,
+      "grad_norm": 9.0,
+      "learning_rate": 6.05427974947808e-05,
+      "loss": 1.3893,
+      "step": 145
+    },
+    {
+      "epoch": 0.4702194357366771,
+      "grad_norm": 25.875,
+      "learning_rate": 6.263048016701462e-05,
+      "loss": 1.3631,
+      "step": 150
+    },
+    {
+      "epoch": 0.48589341692789967,
+      "grad_norm": 5.21875,
+      "learning_rate": 6.471816283924845e-05,
+      "loss": 1.3578,
+      "step": 155
+    },
+    {
+      "epoch": 0.5015673981191222,
+      "grad_norm": 4.96875,
+      "learning_rate": 6.680584551148226e-05,
+      "loss": 1.2742,
+      "step": 160
+    },
+    {
+      "epoch": 0.5172413793103449,
+      "grad_norm": 7.21875,
+      "learning_rate": 6.889352818371608e-05,
+      "loss": 1.2685,
+      "step": 165
+    },
+    {
+      "epoch": 0.5329153605015674,
+      "grad_norm": 8.5,
+      "learning_rate": 7.098121085594989e-05,
+      "loss": 1.2819,
+      "step": 170
+    },
+    {
+      "epoch": 0.54858934169279,
+      "grad_norm": 9.1875,
+      "learning_rate": 7.306889352818372e-05,
+      "loss": 1.284,
+      "step": 175
+    },
+    {
+      "epoch": 0.5642633228840125,
+      "grad_norm": 6.40625,
+      "learning_rate": 7.515657620041754e-05,
+      "loss": 1.2541,
+      "step": 180
+    },
+    {
+      "epoch": 0.5799373040752351,
+      "grad_norm": 11.5625,
+      "learning_rate": 7.724425887265136e-05,
+      "loss": 1.25,
+      "step": 185
+    },
+    {
+      "epoch": 0.5956112852664577,
+      "grad_norm": 11.875,
+      "learning_rate": 7.933194154488518e-05,
+      "loss": 1.2324,
+      "step": 190
+    },
+    {
+      "epoch": 0.6112852664576802,
+      "grad_norm": 4.6875,
+      "learning_rate": 8.141962421711901e-05,
+      "loss": 1.2093,
+      "step": 195
+    },
+    {
+      "epoch": 0.6269592476489029,
+      "grad_norm": 10.0,
+      "learning_rate": 8.350730688935282e-05,
+      "loss": 1.2225,
+      "step": 200
+    },
+    {
+      "epoch": 0.6426332288401254,
+      "grad_norm": 2.03125,
+      "learning_rate": 8.559498956158665e-05,
+      "loss": 1.1948,
+      "step": 205
+    },
+    {
+      "epoch": 0.658307210031348,
+      "grad_norm": 6.125,
+      "learning_rate": 8.768267223382047e-05,
+      "loss": 1.1823,
+      "step": 210
+    },
+    {
+      "epoch": 0.6739811912225705,
+      "grad_norm": 4.34375,
+      "learning_rate": 8.977035490605428e-05,
+      "loss": 1.1812,
+      "step": 215
+    },
+    {
+      "epoch": 0.6896551724137931,
+      "grad_norm": 5.3125,
+      "learning_rate": 9.18580375782881e-05,
+      "loss": 1.1833,
+      "step": 220
+    },
+    {
+      "epoch": 0.7053291536050157,
+      "grad_norm": 9.875,
+      "learning_rate": 9.394572025052193e-05,
+      "loss": 1.1606,
+      "step": 225
+    },
+    {
+      "epoch": 0.7210031347962382,
+      "grad_norm": 2.609375,
+      "learning_rate": 9.603340292275574e-05,
+      "loss": 1.1662,
+      "step": 230
+    },
+    {
+      "epoch": 0.7366771159874608,
+      "grad_norm": 2.96875,
+      "learning_rate": 9.812108559498957e-05,
+      "loss": 1.1573,
+      "step": 235
+    },
+    {
+      "epoch": 0.7523510971786834,
+      "grad_norm": 2.0,
+      "learning_rate": 0.00010020876826722338,
+      "loss": 1.1386,
+      "step": 240
+    },
+    {
+      "epoch": 0.768025078369906,
+      "grad_norm": 5.3125,
+      "learning_rate": 0.00010229645093945721,
+      "loss": 1.1822,
+      "step": 245
+    },
+    {
+      "epoch": 0.7836990595611285,
+      "grad_norm": 8.125,
+      "learning_rate": 0.00010438413361169104,
+      "loss": 1.1415,
+      "step": 250
+    },
+    {
+      "epoch": 0.799373040752351,
+      "grad_norm": 4.65625,
+      "learning_rate": 0.00010647181628392484,
+      "loss": 1.1534,
+      "step": 255
+    },
+    {
+      "epoch": 0.8150470219435737,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.00010855949895615867,
+      "loss": 1.1338,
+      "step": 260
+    },
+    {
+      "epoch": 0.8307210031347962,
+      "grad_norm": 3.03125,
+      "learning_rate": 0.00011064718162839249,
+      "loss": 1.1435,
+      "step": 265
+    },
+    {
+      "epoch": 0.8463949843260188,
+      "grad_norm": 2.90625,
+      "learning_rate": 0.00011273486430062632,
+      "loss": 1.1213,
+      "step": 270
+    },
+    {
+      "epoch": 0.8620689655172413,
+      "grad_norm": 3.875,
+      "learning_rate": 0.00011482254697286012,
+      "loss": 1.123,
+      "step": 275
+    },
+    {
+      "epoch": 0.877742946708464,
+      "grad_norm": 9.8125,
+      "learning_rate": 0.00011691022964509395,
+      "loss": 1.154,
+      "step": 280
+    },
+    {
+      "epoch": 0.8934169278996865,
+      "grad_norm": 3.15625,
+      "learning_rate": 0.00011899791231732778,
+      "loss": 1.1346,
+      "step": 285
+    },
+    {
+      "epoch": 0.9090909090909091,
+      "grad_norm": 15.75,
+      "learning_rate": 0.0001210855949895616,
+      "loss": 1.0998,
+      "step": 290
+    },
+    {
+      "epoch": 0.9247648902821317,
+      "grad_norm": 4.375,
+      "learning_rate": 0.0001231732776617954,
+      "loss": 1.0742,
+      "step": 295
+    },
+    {
+      "epoch": 0.9404388714733543,
+      "grad_norm": 3.015625,
+      "learning_rate": 0.00012526096033402923,
+      "loss": 1.0956,
+      "step": 300
+    },
+    {
+      "epoch": 0.9561128526645768,
+      "grad_norm": 3.796875,
+      "learning_rate": 0.00012734864300626306,
+      "loss": 1.0698,
+      "step": 305
+    },
+    {
+      "epoch": 0.9717868338557993,
+      "grad_norm": 2.75,
+      "learning_rate": 0.0001294363256784969,
+      "loss": 1.0526,
+      "step": 310
+    },
+    {
+      "epoch": 0.987460815047022,
+      "grad_norm": 2.078125,
+      "learning_rate": 0.0001315240083507307,
+      "loss": 1.0589,
+      "step": 315
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 1.7604742050170898,
+      "eval_runtime": 0.8071,
+      "eval_samples_per_second": 2.478,
+      "eval_steps_per_second": 1.239,
+      "step": 319
+    },
+    {
+      "epoch": 1.0031347962382444,
+      "grad_norm": 2.359375,
+      "learning_rate": 0.00013361169102296452,
+      "loss": 1.06,
+      "step": 320
+    },
+    {
+      "epoch": 1.0188087774294672,
+      "grad_norm": 34.0,
+      "learning_rate": 0.00013569937369519835,
+      "loss": 1.0232,
+      "step": 325
+    },
+    {
+      "epoch": 1.0344827586206897,
+      "grad_norm": 8.6875,
+      "learning_rate": 0.00013778705636743215,
+      "loss": 1.0624,
+      "step": 330
+    },
+    {
+      "epoch": 1.0501567398119123,
+      "grad_norm": 10.8125,
+      "learning_rate": 0.00013987473903966598,
+      "loss": 1.0581,
+      "step": 335
+    },
+    {
+      "epoch": 1.0658307210031348,
+      "grad_norm": 3.109375,
+      "learning_rate": 0.00014196242171189978,
+      "loss": 1.0274,
+      "step": 340
+    },
+    {
+      "epoch": 1.0815047021943573,
+      "grad_norm": 3.828125,
+      "learning_rate": 0.0001440501043841336,
+      "loss": 1.0301,
+      "step": 345
+    },
+    {
+      "epoch": 1.09717868338558,
+      "grad_norm": 10.25,
+      "learning_rate": 0.00014613778705636744,
+      "loss": 1.0221,
+      "step": 350
+    },
+    {
+      "epoch": 1.1128526645768024,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.00014822546972860124,
+      "loss": 1.0185,
+      "step": 355
+    },
+    {
+      "epoch": 1.1285266457680252,
+      "grad_norm": 6.65625,
+      "learning_rate": 0.00015031315240083507,
+      "loss": 1.0146,
+      "step": 360
+    },
+    {
+      "epoch": 1.1442006269592477,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.0001524008350730689,
+      "loss": 1.0134,
+      "step": 365
+    },
+    {
+      "epoch": 1.1598746081504703,
+      "grad_norm": 5.375,
+      "learning_rate": 0.00015448851774530273,
+      "loss": 1.0372,
+      "step": 370
+    },
+    {
+      "epoch": 1.1755485893416928,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.00015657620041753653,
+      "loss": 1.0096,
+      "step": 375
+    },
+    {
+      "epoch": 1.1912225705329154,
+      "grad_norm": 3.734375,
+      "learning_rate": 0.00015866388308977036,
+      "loss": 1.0143,
+      "step": 380
+    },
+    {
+      "epoch": 1.206896551724138,
+      "grad_norm": 3.265625,
+      "learning_rate": 0.0001607515657620042,
+      "loss": 1.0159,
+      "step": 385
+    },
+    {
+      "epoch": 1.2225705329153604,
+      "grad_norm": 1.875,
+      "learning_rate": 0.00016283924843423802,
+      "loss": 1.0069,
+      "step": 390
+    },
+    {
+      "epoch": 1.238244514106583,
+      "grad_norm": 1.9609375,
+      "learning_rate": 0.00016492693110647182,
+      "loss": 1.0019,
+      "step": 395
+    },
+    {
+      "epoch": 1.2539184952978055,
+      "grad_norm": 5.0,
+      "learning_rate": 0.00016701461377870565,
+      "loss": 0.9979,
+      "step": 400
+    },
+    {
+      "epoch": 1.2695924764890283,
+      "grad_norm": 0.85546875,
+      "learning_rate": 0.00016910229645093947,
+      "loss": 1.0066,
+      "step": 405
+    },
+    {
+      "epoch": 1.2852664576802508,
+      "grad_norm": 3.25,
+      "learning_rate": 0.0001711899791231733,
+      "loss": 1.0002,
+      "step": 410
+    },
+    {
+      "epoch": 1.3009404388714734,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.0001732776617954071,
+      "loss": 1.0036,
+      "step": 415
+    },
+    {
+      "epoch": 1.316614420062696,
+      "grad_norm": 3.125,
+      "learning_rate": 0.00017536534446764093,
+      "loss": 1.003,
+      "step": 420
+    },
+    {
+      "epoch": 1.3322884012539185,
+      "grad_norm": 3.390625,
+      "learning_rate": 0.00017745302713987476,
+      "loss": 1.0207,
+      "step": 425
+    },
+    {
+      "epoch": 1.347962382445141,
+      "grad_norm": 9.3125,
+      "learning_rate": 0.00017954070981210856,
+      "loss": 1.0097,
+      "step": 430
+    },
+    {
+      "epoch": 1.3636363636363638,
+      "grad_norm": 2.140625,
+      "learning_rate": 0.0001816283924843424,
+      "loss": 0.9984,
+      "step": 435
+    },
+    {
+      "epoch": 1.3793103448275863,
+      "grad_norm": 0.93359375,
+      "learning_rate": 0.0001837160751565762,
+      "loss": 1.019,
+      "step": 440
+    },
+    {
+      "epoch": 1.3949843260188088,
+      "grad_norm": 1.4140625,
+      "learning_rate": 0.00018580375782881002,
+      "loss": 1.0093,
+      "step": 445
+    },
+    {
+      "epoch": 1.4106583072100314,
+      "grad_norm": 2.90625,
+      "learning_rate": 0.00018789144050104385,
+      "loss": 0.9906,
+      "step": 450
+    },
+    {
+      "epoch": 1.426332288401254,
+      "grad_norm": 1.8203125,
+      "learning_rate": 0.00018997912317327765,
+      "loss": 0.9956,
+      "step": 455
+    },
+    {
+      "epoch": 1.4420062695924765,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.00019206680584551148,
+      "loss": 1.0026,
+      "step": 460
+    },
+    {
+      "epoch": 1.457680250783699,
+      "grad_norm": 2.46875,
+      "learning_rate": 0.0001941544885177453,
+      "loss": 0.9779,
+      "step": 465
+    },
+    {
+      "epoch": 1.4733542319749215,
+      "grad_norm": 1.0859375,
+      "learning_rate": 0.00019624217118997914,
+      "loss": 0.9981,
+      "step": 470
+    },
+    {
+      "epoch": 1.489028213166144,
+      "grad_norm": 2.34375,
+      "learning_rate": 0.00019832985386221294,
+      "loss": 1.0087,
+      "step": 475
+    },
+    {
+      "epoch": 1.5047021943573666,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.0001999999733852936,
+      "loss": 1.0154,
+      "step": 480
+    },
+    {
+      "epoch": 1.5203761755485894,
+      "grad_norm": 2.125,
+      "learning_rate": 0.00019999904187205744,
+      "loss": 1.0196,
+      "step": 485
+    },
+    {
+      "epoch": 1.536050156739812,
+      "grad_norm": 5.1875,
+      "learning_rate": 0.00019999677963766844,
+      "loss": 1.0196,
+      "step": 490
+    },
+    {
+      "epoch": 1.5517241379310345,
+      "grad_norm": 3.015625,
+      "learning_rate": 0.00019999318671223102,
+      "loss": 1.003,
+      "step": 495
+    },
+    {
+      "epoch": 1.567398119122257,
+      "grad_norm": 2.453125,
+      "learning_rate": 0.0001999882631435574,
+      "loss": 1.0022,
+      "step": 500
+    },
+    {
+      "epoch": 1.5830721003134798,
+      "grad_norm": 5.78125,
+      "learning_rate": 0.00019998200899716724,
+      "loss": 0.9932,
+      "step": 505
+    },
+    {
+      "epoch": 1.5987460815047023,
+      "grad_norm": 3.625,
+      "learning_rate": 0.00019997442435628653,
+      "loss": 1.0083,
+      "step": 510
+    },
+    {
+      "epoch": 1.6144200626959249,
+      "grad_norm": 6.375,
+      "learning_rate": 0.00019996550932184666,
+      "loss": 1.0153,
+      "step": 515
+    },
+    {
+      "epoch": 1.6300940438871474,
+      "grad_norm": 1.59375,
+      "learning_rate": 0.00019995526401248302,
+      "loss": 1.0158,
+      "step": 520
+    },
+    {
+      "epoch": 1.64576802507837,
+      "grad_norm": 4.84375,
+      "learning_rate": 0.00019994368856453341,
+      "loss": 1.0308,
+      "step": 525
+    },
+    {
+      "epoch": 1.6614420062695925,
+      "grad_norm": 2.796875,
+      "learning_rate": 0.00019993078313203632,
+      "loss": 0.9834,
+      "step": 530
+    },
+    {
+      "epoch": 1.677115987460815,
+      "grad_norm": 1.28125,
+      "learning_rate": 0.0001999165478867286,
+      "loss": 0.9937,
+      "step": 535
+    },
+    {
+      "epoch": 1.6927899686520376,
+      "grad_norm": 1.9921875,
+      "learning_rate": 0.00019990098301804357,
+      "loss": 0.9884,
+      "step": 540
+    },
+    {
+      "epoch": 1.70846394984326,
+      "grad_norm": 1.671875,
+      "learning_rate": 0.00019988408873310815,
+      "loss": 0.9846,
+      "step": 545
+    },
+    {
+      "epoch": 1.7241379310344827,
+      "grad_norm": 0.9921875,
+      "learning_rate": 0.00019986586525674036,
+      "loss": 0.9711,
+      "step": 550
+    },
+    {
+      "epoch": 1.7398119122257052,
+      "grad_norm": 1.921875,
+      "learning_rate": 0.00019984631283144616,
+      "loss": 0.9789,
+      "step": 555
+    },
+    {
+      "epoch": 1.7554858934169277,
+      "grad_norm": 1.953125,
+      "learning_rate": 0.0001998254317174163,
+      "loss": 0.984,
+      "step": 560
+    },
+    {
+      "epoch": 1.7711598746081505,
+      "grad_norm": 3.046875,
+      "learning_rate": 0.00019980322219252284,
+      "loss": 0.9609,
+      "step": 565
+    },
+    {
+      "epoch": 1.786833855799373,
+      "grad_norm": 1.7578125,
+      "learning_rate": 0.0001997796845523155,
+      "loss": 0.9997,
+      "step": 570
+    },
+    {
+      "epoch": 1.8025078369905956,
+      "grad_norm": 2.9375,
+      "learning_rate": 0.00019975481911001762,
+      "loss": 0.9772,
+      "step": 575
+    },
+    {
+      "epoch": 1.8181818181818183,
+      "grad_norm": 1.5234375,
+      "learning_rate": 0.00019972862619652203,
+      "loss": 0.9939,
+      "step": 580
+    },
+    {
+      "epoch": 1.8338557993730409,
+      "grad_norm": 1.6953125,
+      "learning_rate": 0.00019970110616038673,
+      "loss": 0.9794,
+      "step": 585
+    },
+    {
+      "epoch": 1.8495297805642634,
+      "grad_norm": 3.109375,
+      "learning_rate": 0.0001996722593678302,
+      "loss": 0.9702,
+      "step": 590
+    },
+    {
+      "epoch": 1.865203761755486,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.00019964208620272647,
+      "loss": 0.9667,
+      "step": 595
+    },
+    {
+      "epoch": 1.8808777429467085,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.00019961058706660005,
+      "loss": 0.9632,
+      "step": 600
+    },
+    {
+      "epoch": 1.896551724137931,
+      "grad_norm": 12.5,
+      "learning_rate": 0.00019957776237862067,
+      "loss": 0.9554,
+      "step": 605
+    },
+    {
+      "epoch": 1.9122257053291536,
+      "grad_norm": 1.6328125,
+      "learning_rate": 0.00019954361257559756,
+      "loss": 0.9902,
+      "step": 610
+    },
+    {
+      "epoch": 1.9278996865203761,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.0001995081381119737,
+      "loss": 0.998,
+      "step": 615
+    },
+    {
+      "epoch": 1.9435736677115987,
+      "grad_norm": 2.28125,
+      "learning_rate": 0.00019947133945981987,
+      "loss": 0.9781,
+      "step": 620
+    },
+    {
+      "epoch": 1.9592476489028212,
+      "grad_norm": 2.796875,
+      "learning_rate": 0.00019943321710882815,
+      "loss": 0.9956,
+      "step": 625
+    },
+    {
+      "epoch": 1.9749216300940438,
+      "grad_norm": 3.046875,
+      "learning_rate": 0.0001993937715663056,
+      "loss": 0.9776,
+      "step": 630
+    },
+    {
+      "epoch": 1.9905956112852663,
+      "grad_norm": 1.9140625,
+      "learning_rate": 0.00019935300335716748,
+      "loss": 1.0,
+      "step": 635
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 1.7305909395217896,
+      "eval_runtime": 0.8036,
+      "eval_samples_per_second": 2.489,
+      "eval_steps_per_second": 1.244,
+      "step": 638
+    },
+    {
+      "epoch": 2.006269592476489,
+      "grad_norm": 3.921875,
+      "learning_rate": 0.00019931091302393008,
+      "loss": 0.9559,
+      "step": 640
+    },
+    {
+      "epoch": 2.0219435736677114,
+      "grad_norm": 1.6015625,
+      "learning_rate": 0.00019926750112670382,
+      "loss": 0.8412,
+      "step": 645
+    },
+    {
+      "epoch": 2.0376175548589344,
+      "grad_norm": 1.8203125,
+      "learning_rate": 0.00019922276824318547,
+      "loss": 0.8475,
+      "step": 650
+    },
+    {
+      "epoch": 2.053291536050157,
+      "grad_norm": 1.3359375,
+      "learning_rate": 0.0001991767149686507,
+      "loss": 0.8512,
+      "step": 655
+    },
+    {
+      "epoch": 2.0689655172413794,
+      "grad_norm": 1.0,
+      "learning_rate": 0.0001991293419159461,
+      "loss": 0.8301,
+      "step": 660
+    },
+    {
+      "epoch": 2.084639498432602,
+      "grad_norm": 2.390625,
+      "learning_rate": 0.00019908064971548085,
+      "loss": 0.8622,
+      "step": 665
+    },
+    {
+      "epoch": 2.1003134796238245,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.0001990306390152186,
+      "loss": 0.8375,
+      "step": 670
+    },
+    {
+      "epoch": 2.115987460815047,
+      "grad_norm": 3.390625,
+      "learning_rate": 0.00019897931048066877,
+      "loss": 0.8571,
+      "step": 675
+    },
+    {
+      "epoch": 2.1316614420062696,
+      "grad_norm": 1.8203125,
+      "learning_rate": 0.00019892666479487744,
+      "loss": 0.904,
+      "step": 680
+    },
+    {
+      "epoch": 2.147335423197492,
+      "grad_norm": 2.859375,
+      "learning_rate": 0.00019887270265841868,
+      "loss": 0.8602,
+      "step": 685
+    },
+    {
+      "epoch": 2.1630094043887147,
+      "grad_norm": 2.125,
+      "learning_rate": 0.00019881742478938496,
+      "loss": 0.8618,
+      "step": 690
+    },
+    {
+      "epoch": 2.1786833855799372,
+      "grad_norm": 3.09375,
+      "learning_rate": 0.00019876083192337757,
+      "loss": 0.8737,
+      "step": 695
+    },
+    {
+      "epoch": 2.19435736677116,
+      "grad_norm": 3.9375,
+      "learning_rate": 0.00019870292481349698,
+      "loss": 0.8688,
+      "step": 700
+    },
+    {
+      "epoch": 2.2100313479623823,
+      "grad_norm": 1.7578125,
+      "learning_rate": 0.00019864370423033274,
+      "loss": 0.8821,
+      "step": 705
+    },
+    {
+      "epoch": 2.225705329153605,
+      "grad_norm": 1.3203125,
+      "learning_rate": 0.00019858317096195323,
+      "loss": 0.8746,
+      "step": 710
+    },
+    {
+      "epoch": 2.2413793103448274,
+      "grad_norm": 2.796875,
+      "learning_rate": 0.00019852132581389513,
+      "loss": 0.8742,
+      "step": 715
+    },
+    {
+      "epoch": 2.2570532915360504,
+      "grad_norm": 2.421875,
+      "learning_rate": 0.00019845816960915286,
+      "loss": 0.8747,
+      "step": 720
+    },
+    {
+      "epoch": 2.2727272727272725,
+      "grad_norm": 2.046875,
+      "learning_rate": 0.0001983937031881674,
+      "loss": 0.843,
+      "step": 725
+    },
+    {
+      "epoch": 2.2884012539184955,
+      "grad_norm": 3.140625,
+      "learning_rate": 0.0001983279274088153,
+      "loss": 0.8682,
+      "step": 730
+    },
+    {
+      "epoch": 2.304075235109718,
+      "grad_norm": 2.9375,
+      "learning_rate": 0.00019826084314639714,
+      "loss": 0.856,
+      "step": 735
+    },
+    {
+      "epoch": 2.3197492163009406,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.00019819245129362595,
+      "loss": 0.8663,
+      "step": 740
+    },
+    {
+      "epoch": 2.335423197492163,
+      "grad_norm": 2.21875,
+      "learning_rate": 0.00019812275276061533,
+      "loss": 0.8483,
+      "step": 745
+    },
+    {
+      "epoch": 2.3510971786833856,
+      "grad_norm": 9.875,
+      "learning_rate": 0.00019805174847486721,
+      "loss": 0.8416,
+      "step": 750
+    },
+    {
+      "epoch": 2.366771159874608,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.00019797943938125977,
+      "loss": 0.8743,
+      "step": 755
+    },
+    {
+      "epoch": 2.3824451410658307,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.00019790582644203458,
+      "loss": 0.8529,
+      "step": 760
+    },
+    {
+      "epoch": 2.3981191222570533,
+      "grad_norm": 6.15625,
+      "learning_rate": 0.00019783091063678402,
+      "loss": 0.8628,
+      "step": 765
+    },
+    {
+      "epoch": 2.413793103448276,
+      "grad_norm": 2.65625,
+      "learning_rate": 0.00019775469296243807,
+      "loss": 0.8689,
+      "step": 770
+    },
+    {
+      "epoch": 2.4294670846394983,
+      "grad_norm": 1.59375,
+      "learning_rate": 0.0001976771744332512,
+      "loss": 0.8671,
+      "step": 775
+    },
+    {
+      "epoch": 2.445141065830721,
+      "grad_norm": 2.0,
+      "learning_rate": 0.00019759835608078877,
+      "loss": 0.8832,
+      "step": 780
+    },
+    {
+      "epoch": 2.4608150470219434,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.00019751823895391323,
+      "loss": 0.878,
+      "step": 785
+    },
+    {
+      "epoch": 2.476489028213166,
+      "grad_norm": 1.5,
+      "learning_rate": 0.00019743682411877046,
+      "loss": 0.8882,
+      "step": 790
+    },
+    {
+      "epoch": 2.492163009404389,
+      "grad_norm": 1.7265625,
+      "learning_rate": 0.00019735411265877522,
+      "loss": 0.8934,
+      "step": 795
+    },
+    {
+      "epoch": 2.507836990595611,
+      "grad_norm": 1.4140625,
+      "learning_rate": 0.00019727010567459696,
+      "loss": 0.8815,
+      "step": 800
+    },
+    {
+      "epoch": 2.523510971786834,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.00019718480428414505,
+      "loss": 0.8925,
+      "step": 805
+    },
+    {
+      "epoch": 2.5391849529780566,
+      "grad_norm": 1.4765625,
+      "learning_rate": 0.00019709820962255409,
+      "loss": 0.8956,
+      "step": 810
+    },
+    {
+      "epoch": 2.554858934169279,
+      "grad_norm": 3.546875,
+      "learning_rate": 0.00019701032284216857,
+      "loss": 0.8828,
+      "step": 815
+    },
+    {
+      "epoch": 2.5705329153605017,
+      "grad_norm": 1.125,
+      "learning_rate": 0.00019692114511252767,
+      "loss": 0.9194,
+      "step": 820
+    },
+    {
+      "epoch": 2.586206896551724,
+      "grad_norm": 1.734375,
+      "learning_rate": 0.00019683067762034967,
+      "loss": 0.8825,
+      "step": 825
+    },
+    {
+      "epoch": 2.6018808777429467,
+      "grad_norm": 4.1875,
+      "learning_rate": 0.00019673892156951613,
+      "loss": 0.8725,
+      "step": 830
+    },
+    {
+      "epoch": 2.6175548589341693,
+      "grad_norm": 1.296875,
+      "learning_rate": 0.00019664587818105596,
+      "loss": 0.877,
+      "step": 835
+    },
+    {
+      "epoch": 2.633228840125392,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.0001965515486931291,
+      "loss": 0.8764,
+      "step": 840
+    },
+    {
+      "epoch": 2.6489028213166144,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.00019645593436101,
+      "loss": 0.8571,
+      "step": 845
+    },
+    {
+      "epoch": 2.664576802507837,
+      "grad_norm": 2.40625,
+      "learning_rate": 0.00019635903645707096,
+      "loss": 0.887,
+      "step": 850
+    },
+    {
+      "epoch": 2.6802507836990594,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.00019626085627076528,
+      "loss": 0.8755,
+      "step": 855
+    },
+    {
+      "epoch": 2.695924764890282,
+      "grad_norm": 1.4140625,
+      "learning_rate": 0.00019616139510861,
+      "loss": 0.8664,
+      "step": 860
+    },
+    {
+      "epoch": 2.7115987460815045,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.00019606065429416848,
+      "loss": 0.8888,
+      "step": 865
+    },
+    {
+      "epoch": 2.7272727272727275,
+      "grad_norm": 1.4296875,
+      "learning_rate": 0.00019595863516803293,
+      "loss": 0.8772,
+      "step": 870
+    },
+    {
+      "epoch": 2.7429467084639496,
+      "grad_norm": 1.5859375,
+      "learning_rate": 0.0001958553390878064,
+      "loss": 0.8919,
+      "step": 875
+    },
+    {
+      "epoch": 2.7586206896551726,
+      "grad_norm": 1.609375,
+      "learning_rate": 0.00019575076742808488,
+      "loss": 0.8806,
+      "step": 880
+    },
+    {
+      "epoch": 2.774294670846395,
+      "grad_norm": 0.81640625,
+      "learning_rate": 0.00019564492158043891,
+      "loss": 0.8722,
+      "step": 885
+    },
+    {
+      "epoch": 2.7899686520376177,
+      "grad_norm": 1.359375,
+      "learning_rate": 0.0001955378029533951,
+      "loss": 0.8887,
+      "step": 890
+    },
+    {
+      "epoch": 2.80564263322884,
+      "grad_norm": 1.1015625,
+      "learning_rate": 0.00019542941297241722,
+      "loss": 0.9079,
+      "step": 895
+    },
+    {
+      "epoch": 2.8213166144200628,
+      "grad_norm": 2.0,
+      "learning_rate": 0.00019531975307988763,
+      "loss": 0.877,
+      "step": 900
+    },
+    {
+      "epoch": 2.8369905956112853,
+      "grad_norm": 1.4140625,
+      "learning_rate": 0.00019520882473508762,
+      "loss": 0.8953,
+      "step": 905
+    },
+    {
+      "epoch": 2.852664576802508,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.00019509662941417826,
+      "loss": 0.8886,
+      "step": 910
+    },
+    {
+      "epoch": 2.8683385579937304,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.00019498316861018086,
+      "loss": 0.9104,
+      "step": 915
+    },
+    {
+      "epoch": 2.884012539184953,
+      "grad_norm": 1.6015625,
+      "learning_rate": 0.0001948684438329566,
+      "loss": 0.8815,
+      "step": 920
+    },
+    {
+      "epoch": 2.8996865203761755,
+      "grad_norm": 1.8125,
+      "learning_rate": 0.00019475245660918717,
+      "loss": 0.8718,
+      "step": 925
+    },
+    {
+      "epoch": 2.915360501567398,
+      "grad_norm": 6.46875,
+      "learning_rate": 0.00019463520848235377,
+      "loss": 0.8774,
+      "step": 930
+    },
+    {
+      "epoch": 2.9310344827586206,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.000194516701012717,
+      "loss": 0.8853,
+      "step": 935
+    },
+    {
+      "epoch": 2.946708463949843,
+      "grad_norm": 1.5546875,
+      "learning_rate": 0.00019439693577729593,
+      "loss": 0.8833,
+      "step": 940
+    },
+    {
+      "epoch": 2.962382445141066,
+      "grad_norm": 1.796875,
+      "learning_rate": 0.0001942759143698472,
+      "loss": 0.8867,
+      "step": 945
+    },
+    {
+      "epoch": 2.978056426332288,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.0001941536384008437,
+      "loss": 0.8923,
+      "step": 950
+    },
+    {
+      "epoch": 2.993730407523511,
+      "grad_norm": 1.75,
+      "learning_rate": 0.0001940301094974531,
+      "loss": 0.8741,
+      "step": 955
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 1.8115239143371582,
+      "eval_runtime": 0.806,
+      "eval_samples_per_second": 2.481,
+      "eval_steps_per_second": 1.241,
+      "step": 957
+    },
+    {
+      "epoch": 3.0094043887147337,
+      "grad_norm": 0.7578125,
+      "learning_rate": 0.00019390532930351652,
+      "loss": 0.7936,
+      "step": 960
+    },
+    {
+      "epoch": 3.0250783699059562,
+      "grad_norm": 1.7734375,
+      "learning_rate": 0.00019377929947952626,
+      "loss": 0.7324,
+      "step": 965
+    },
+    {
+      "epoch": 3.040752351097179,
+      "grad_norm": 1.7109375,
+      "learning_rate": 0.00019365202170260393,
+      "loss": 0.726,
+      "step": 970
+    },
+    {
+      "epoch": 3.0564263322884013,
+      "grad_norm": 1.0859375,
+      "learning_rate": 0.000193523497666478,
+      "loss": 0.7355,
+      "step": 975
+    },
+    {
+      "epoch": 3.072100313479624,
+      "grad_norm": 2.578125,
+      "learning_rate": 0.00019339372908146147,
+      "loss": 0.7393,
+      "step": 980
+    },
+    {
+      "epoch": 3.0877742946708464,
+      "grad_norm": 1.0,
+      "learning_rate": 0.00019326271767442884,
+      "loss": 0.736,
+      "step": 985
+    },
+    {
+      "epoch": 3.103448275862069,
+      "grad_norm": 1.3203125,
+      "learning_rate": 0.00019313046518879337,
+      "loss": 0.7157,
+      "step": 990
+    },
+    {
+      "epoch": 3.1191222570532915,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.00019299697338448369,
+      "loss": 0.7231,
+      "step": 995
+    },
+    {
+      "epoch": 3.134796238244514,
+      "grad_norm": 1.2734375,
+      "learning_rate": 0.0001928622440379205,
+      "loss": 0.7221,
+      "step": 1000
+    },
+    {
+      "epoch": 3.1504702194357366,
+      "grad_norm": 0.9140625,
+      "learning_rate": 0.0001927262789419929,
+      "loss": 0.7352,
+      "step": 1005
+    },
+    {
+      "epoch": 3.166144200626959,
+      "grad_norm": 0.9375,
+      "learning_rate": 0.0001925890799060345,
+      "loss": 0.7196,
+      "step": 1010
+    },
+    {
+      "epoch": 3.1818181818181817,
+      "grad_norm": 0.90625,
+      "learning_rate": 0.00019245064875579942,
+      "loss": 0.7269,
+      "step": 1015
+    },
+    {
+      "epoch": 3.197492163009404,
+      "grad_norm": 0.98828125,
+      "learning_rate": 0.00019231098733343783,
+      "loss": 0.7225,
+      "step": 1020
+    },
+    {
+      "epoch": 3.2131661442006267,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.00019217009749747174,
+      "loss": 0.734,
+      "step": 1025
+    },
+    {
+      "epoch": 3.2288401253918497,
+      "grad_norm": 0.828125,
+      "learning_rate": 0.0001920279811227699,
+      "loss": 0.7387,
+      "step": 1030
+    },
+    {
+      "epoch": 3.2445141065830723,
+      "grad_norm": 1.5703125,
+      "learning_rate": 0.00019188464010052312,
+      "loss": 0.7303,
+      "step": 1035
+    },
+    {
+      "epoch": 3.260188087774295,
+      "grad_norm": 1.53125,
+      "learning_rate": 0.00019174007633821893,
+      "loss": 0.7565,
+      "step": 1040
+    },
+    {
+      "epoch": 3.2758620689655173,
+      "grad_norm": 2.921875,
+      "learning_rate": 0.00019159429175961634,
+      "loss": 0.7588,
+      "step": 1045
+    },
+    {
+      "epoch": 3.29153605015674,
+      "grad_norm": 1.0,
+      "learning_rate": 0.0001914472883047202,
+      "loss": 0.7452,
+      "step": 1050
+    },
+    {
+      "epoch": 3.3072100313479624,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.00019129906792975527,
+      "loss": 0.7395,
+      "step": 1055
+    },
+    {
+      "epoch": 3.322884012539185,
+      "grad_norm": 1.203125,
+      "learning_rate": 0.0001911496326071404,
+      "loss": 0.7429,
+      "step": 1060
+    },
+    {
+      "epoch": 3.3385579937304075,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.00019099898432546202,
+      "loss": 0.7643,
+      "step": 1065
+    },
+    {
+      "epoch": 3.35423197492163,
+      "grad_norm": 1.3203125,
+      "learning_rate": 0.00019084712508944793,
+      "loss": 0.755,
+      "step": 1070
+    },
+    {
+      "epoch": 3.3699059561128526,
+      "grad_norm": 1.8671875,
+      "learning_rate": 0.00019069405691994045,
+      "loss": 0.7381,
+      "step": 1075
+    },
+    {
+      "epoch": 3.385579937304075,
+      "grad_norm": 1.46875,
+      "learning_rate": 0.00019053978185386964,
+      "loss": 0.7546,
+      "step": 1080
+    },
+    {
+      "epoch": 3.4012539184952977,
+      "grad_norm": 2.609375,
+      "learning_rate": 0.00019038430194422606,
+      "loss": 0.7624,
+      "step": 1085
+    },
+    {
+      "epoch": 3.41692789968652,
+      "grad_norm": 1.3203125,
+      "learning_rate": 0.00019022761926003359,
+      "loss": 0.7657,
+      "step": 1090
+    },
+    {
+      "epoch": 3.4326018808777428,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.00019006973588632184,
+      "loss": 0.7433,
+      "step": 1095
+    },
+    {
+      "epoch": 3.4482758620689653,
+      "grad_norm": 1.6015625,
+      "learning_rate": 0.0001899106539240984,
+      "loss": 0.7767,
+      "step": 1100
+    },
+    {
+      "epoch": 3.4639498432601883,
+      "grad_norm": 33.75,
+      "learning_rate": 0.00018975037549032086,
+      "loss": 0.755,
+      "step": 1105
+    },
+    {
+      "epoch": 3.479623824451411,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.0001895889027178687,
+      "loss": 0.7631,
+      "step": 1110
+    },
+    {
+      "epoch": 3.4952978056426334,
+      "grad_norm": 1.1015625,
+      "learning_rate": 0.0001894262377555148,
+      "loss": 0.7545,
+      "step": 1115
+    },
+    {
+      "epoch": 3.510971786833856,
+      "grad_norm": 0.9765625,
+      "learning_rate": 0.00018926238276789704,
+      "loss": 0.7491,
+      "step": 1120
+    },
+    {
+      "epoch": 3.5266457680250785,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.0001890973399354892,
+      "loss": 0.7663,
+      "step": 1125
+    },
+    {
+      "epoch": 3.542319749216301,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.00018893111145457225,
+      "loss": 0.755,
+      "step": 1130
+    },
+    {
+      "epoch": 3.5579937304075235,
+      "grad_norm": 1.8046875,
+      "learning_rate": 0.00018876369953720496,
+      "loss": 0.7681,
+      "step": 1135
+    },
+    {
+      "epoch": 3.573667711598746,
+      "grad_norm": 5.125,
+      "learning_rate": 0.00018859510641119448,
+      "loss": 0.766,
+      "step": 1140
+    },
+    {
+      "epoch": 3.5893416927899686,
+      "grad_norm": 1.3828125,
+      "learning_rate": 0.00018842533432006662,
+      "loss": 0.7801,
+      "step": 1145
+    },
+    {
+      "epoch": 3.605015673981191,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.00018825438552303621,
+      "loss": 0.7647,
+      "step": 1150
+    },
+    {
+      "epoch": 3.6206896551724137,
+      "grad_norm": 1.109375,
+      "learning_rate": 0.00018808226229497684,
+      "loss": 0.7768,
+      "step": 1155
+    },
+    {
+      "epoch": 3.6363636363636362,
+      "grad_norm": 1.4921875,
+      "learning_rate": 0.00018790896692639068,
+      "loss": 0.7786,
+      "step": 1160
+    },
+    {
+      "epoch": 3.652037617554859,
+      "grad_norm": 1.2890625,
+      "learning_rate": 0.00018773450172337793,
+      "loss": 0.762,
+      "step": 1165
+    },
+    {
+      "epoch": 3.6677115987460818,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.00018755886900760619,
+      "loss": 0.7612,
+      "step": 1170
+    },
+    {
+      "epoch": 3.683385579937304,
+      "grad_norm": 0.81640625,
+      "learning_rate": 0.00018738207111627958,
+      "loss": 0.7718,
+      "step": 1175
+    },
+    {
+      "epoch": 3.699059561128527,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.00018720411040210752,
+      "loss": 0.7577,
+      "step": 1180
+    },
+    {
+      "epoch": 3.714733542319749,
+      "grad_norm": 0.8828125,
+      "learning_rate": 0.00018702498923327366,
+      "loss": 0.7429,
+      "step": 1185
+    },
+    {
+      "epoch": 3.730407523510972,
+      "grad_norm": 0.828125,
+      "learning_rate": 0.00018684470999340405,
+      "loss": 0.7552,
+      "step": 1190
+    },
+    {
+      "epoch": 3.7460815047021945,
+      "grad_norm": 1.46875,
+      "learning_rate": 0.00018666327508153567,
+      "loss": 0.7606,
+      "step": 1195
+    },
+    {
+      "epoch": 3.761755485893417,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.0001864806869120844,
+      "loss": 0.7678,
+      "step": 1200
+    },
+    {
+      "epoch": 3.7774294670846396,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.00018629694791481296,
+      "loss": 0.7985,
+      "step": 1205
+    },
+    {
+      "epoch": 3.793103448275862,
+      "grad_norm": 5.125,
+      "learning_rate": 0.00018611206053479842,
+      "loss": 0.7712,
+      "step": 1210
+    },
+    {
+      "epoch": 3.8087774294670846,
+      "grad_norm": 1.78125,
+      "learning_rate": 0.00018592602723239984,
+      "loss": 0.7745,
+      "step": 1215
+    },
+    {
+      "epoch": 3.824451410658307,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.00018573885048322547,
+      "loss": 0.7684,
+      "step": 1220
+    },
+    {
+      "epoch": 3.8401253918495297,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.00018555053277809975,
+      "loss": 0.7811,
+      "step": 1225
+    },
+    {
+      "epoch": 3.8557993730407523,
+      "grad_norm": 1.1484375,
+      "learning_rate": 0.00018536107662303026,
+      "loss": 0.7732,
+      "step": 1230
+    },
+    {
+      "epoch": 3.871473354231975,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.00018517048453917424,
+      "loss": 0.7668,
+      "step": 1235
+    },
+    {
+      "epoch": 3.8871473354231973,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.00018497875906280515,
+      "loss": 0.759,
+      "step": 1240
+    },
+    {
+      "epoch": 3.9028213166144203,
+      "grad_norm": 1.46875,
+      "learning_rate": 0.00018478590274527898,
+      "loss": 0.7763,
+      "step": 1245
+    },
+    {
+      "epoch": 3.9184952978056424,
+      "grad_norm": 1.6171875,
+      "learning_rate": 0.0001845919181530001,
+      "loss": 0.7633,
+      "step": 1250
+    },
+    {
+      "epoch": 3.9341692789968654,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.00018439680786738722,
+      "loss": 0.7853,
+      "step": 1255
+    },
+    {
+      "epoch": 3.9498432601880875,
+      "grad_norm": 1.46875,
+      "learning_rate": 0.00018420057448483905,
+      "loss": 0.7856,
+      "step": 1260
+    },
+    {
+      "epoch": 3.9655172413793105,
+      "grad_norm": 1.359375,
+      "learning_rate": 0.00018400322061669982,
+      "loss": 0.7831,
+      "step": 1265
+    },
+    {
+      "epoch": 3.981191222570533,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.00018380474888922426,
+      "loss": 0.7952,
+      "step": 1270
+    },
+    {
+      "epoch": 3.9968652037617556,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.000183605161943543,
+      "loss": 0.7735,
+      "step": 1275
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 1.923946738243103,
+      "eval_runtime": 0.8,
+      "eval_samples_per_second": 2.5,
+      "eval_steps_per_second": 1.25,
+      "step": 1276
+    },
+    {
+      "epoch": 4.012539184952978,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.0001834044624356272,
+      "loss": 0.6611,
+      "step": 1280
+    },
+    {
+      "epoch": 4.028213166144201,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.0001832026530362532,
+      "loss": 0.5993,
+      "step": 1285
+    },
+    {
+      "epoch": 4.043887147335423,
+      "grad_norm": 1.3125,
+      "learning_rate": 0.00018299973643096714,
+      "loss": 0.6197,
+      "step": 1290
+    },
+    {
+      "epoch": 4.059561128526646,
+      "grad_norm": 1.1640625,
+      "learning_rate": 0.00018279571532004907,
+      "loss": 0.6147,
+      "step": 1295
+    },
+    {
+      "epoch": 4.075235109717869,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.00018259059241847707,
+      "loss": 0.6295,
+      "step": 1300
+    },
+    {
+      "epoch": 4.090909090909091,
+      "grad_norm": 0.97265625,
+      "learning_rate": 0.00018238437045589115,
+      "loss": 0.6219,
+      "step": 1305
+    },
+    {
+      "epoch": 4.106583072100314,
+      "grad_norm": 0.98828125,
+      "learning_rate": 0.00018217705217655689,
+      "loss": 0.6033,
+      "step": 1310
+    },
+    {
+      "epoch": 4.122257053291536,
+      "grad_norm": 0.875,
+      "learning_rate": 0.0001819686403393289,
+      "loss": 0.622,
+      "step": 1315
+    },
+    {
+      "epoch": 4.137931034482759,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.00018175913771761417,
+      "loss": 0.6166,
+      "step": 1320
+    },
+    {
+      "epoch": 4.153605015673981,
+      "grad_norm": 1.234375,
+      "learning_rate": 0.0001815485470993351,
+      "loss": 0.6335,
+      "step": 1325
+    },
+    {
+      "epoch": 4.169278996865204,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.00018133687128689242,
+      "loss": 0.6204,
+      "step": 1330
+    },
+    {
+      "epoch": 4.184952978056426,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.000181124113097128,
+      "loss": 0.635,
+      "step": 1335
+    },
+    {
+      "epoch": 4.200626959247649,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.00018091027536128716,
+      "loss": 0.6224,
+      "step": 1340
+    },
+    {
+      "epoch": 4.216300940438871,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.00018069536092498112,
+      "loss": 0.6314,
+      "step": 1345
+    },
+    {
+      "epoch": 4.231974921630094,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.00018047937264814917,
+      "loss": 0.6421,
+      "step": 1350
+    },
+    {
+      "epoch": 4.247648902821316,
+      "grad_norm": 2.25,
+      "learning_rate": 0.00018026231340502057,
+      "loss": 0.6335,
+      "step": 1355
+    },
+    {
+      "epoch": 4.263322884012539,
+      "grad_norm": 1.0,
+      "learning_rate": 0.00018004418608407626,
+      "loss": 0.6365,
+      "step": 1360
+    },
+    {
+      "epoch": 4.278996865203762,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.00017982499358801037,
+      "loss": 0.6289,
+      "step": 1365
+    },
+    {
+      "epoch": 4.294670846394984,
+      "grad_norm": 0.95703125,
+      "learning_rate": 0.00017960473883369186,
+      "loss": 0.6297,
+      "step": 1370
+    },
+    {
+      "epoch": 4.310344827586207,
+      "grad_norm": 1.484375,
+      "learning_rate": 0.00017938342475212532,
+      "loss": 0.6496,
+      "step": 1375
+    },
+    {
+      "epoch": 4.326018808777429,
+      "grad_norm": 0.9140625,
+      "learning_rate": 0.00017916105428841234,
+      "loss": 0.6454,
+      "step": 1380
+    },
+    {
+      "epoch": 4.341692789968652,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.00017893763040171203,
+      "loss": 0.6452,
+      "step": 1385
+    },
+    {
+      "epoch": 4.3573667711598745,
+      "grad_norm": 2.25,
+      "learning_rate": 0.00017871315606520183,
+      "loss": 0.6542,
+      "step": 1390
+    },
+    {
+      "epoch": 4.3730407523510975,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.0001784876342660378,
+      "loss": 0.6448,
+      "step": 1395
+    },
+    {
+      "epoch": 4.38871473354232,
+      "grad_norm": 0.95703125,
+      "learning_rate": 0.00017826106800531498,
+      "loss": 0.63,
+      "step": 1400
+    },
+    {
+      "epoch": 4.4043887147335425,
+      "grad_norm": 1.828125,
+      "learning_rate": 0.0001780334602980275,
+      "loss": 0.6317,
+      "step": 1405
+    },
+    {
+      "epoch": 4.420062695924765,
+      "grad_norm": 1.75,
+      "learning_rate": 0.0001778048141730282,
+      "loss": 0.6375,
+      "step": 1410
+    },
+    {
+      "epoch": 4.435736677115988,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.00017757513267298856,
+      "loss": 0.645,
+      "step": 1415
+    },
+    {
+      "epoch": 4.45141065830721,
+      "grad_norm": 1.6328125,
+      "learning_rate": 0.00017734441885435828,
+      "loss": 0.6431,
+      "step": 1420
+    },
+    {
+      "epoch": 4.467084639498433,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.00017711267578732423,
+      "loss": 0.6494,
+      "step": 1425
+    },
+    {
+      "epoch": 4.482758620689655,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.00017687990655577008,
+      "loss": 0.6464,
+      "step": 1430
+    },
+    {
+      "epoch": 4.498432601880878,
+      "grad_norm": 33.75,
+      "learning_rate": 0.00017664611425723486,
+      "loss": 0.6598,
+      "step": 1435
+    },
+    {
+      "epoch": 4.514106583072101,
+      "grad_norm": 1.5234375,
+      "learning_rate": 0.00017641130200287197,
+      "loss": 0.6602,
+      "step": 1440
+    },
+    {
+      "epoch": 4.529780564263323,
+      "grad_norm": 1.34375,
+      "learning_rate": 0.00017617547291740767,
+      "loss": 0.6443,
+      "step": 1445
+    },
+    {
+      "epoch": 4.545454545454545,
+      "grad_norm": 1.2890625,
+      "learning_rate": 0.00017593863013909956,
+      "loss": 0.6441,
+      "step": 1450
+    },
+    {
+      "epoch": 4.561128526645768,
+      "grad_norm": 1.375,
+      "learning_rate": 0.00017570077681969474,
+      "loss": 0.6405,
+      "step": 1455
+    },
+    {
+      "epoch": 4.576802507836991,
+      "grad_norm": 2.15625,
+      "learning_rate": 0.00017546191612438804,
+      "loss": 0.6605,
+      "step": 1460
+    },
+    {
+      "epoch": 4.592476489028213,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.0001752220512317797,
+      "loss": 0.6572,
+      "step": 1465
+    },
+    {
+      "epoch": 4.608150470219436,
+      "grad_norm": 1.96875,
+      "learning_rate": 0.00017498118533383316,
+      "loss": 0.6396,
+      "step": 1470
+    },
+    {
+      "epoch": 4.623824451410658,
+      "grad_norm": 1.2265625,
+      "learning_rate": 0.0001747393216358326,
+      "loss": 0.656,
+      "step": 1475
+    },
+    {
+      "epoch": 4.639498432601881,
+      "grad_norm": 1.0390625,
+      "learning_rate": 0.00017449646335634017,
+      "loss": 0.6602,
+      "step": 1480
+    },
+    {
+      "epoch": 4.655172413793103,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.00017425261372715345,
+      "loss": 0.6519,
+      "step": 1485
+    },
+    {
+      "epoch": 4.670846394984326,
+      "grad_norm": 0.875,
+      "learning_rate": 0.00017400777599326203,
+      "loss": 0.6475,
+      "step": 1490
+    },
+    {
+      "epoch": 4.686520376175548,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.00017376195341280468,
+      "loss": 0.6656,
+      "step": 1495
+    },
+    {
+      "epoch": 4.702194357366771,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.00017351514925702583,
+      "loss": 0.6655,
+      "step": 1500
+    },
+    {
+      "epoch": 4.717868338557993,
+      "grad_norm": 1.4296875,
+      "learning_rate": 0.00017326736681023204,
+      "loss": 0.672,
+      "step": 1505
+    },
+    {
+      "epoch": 4.733542319749216,
+      "grad_norm": 0.98828125,
+      "learning_rate": 0.00017301860936974834,
+      "loss": 0.6635,
+      "step": 1510
+    },
+    {
+      "epoch": 4.749216300940439,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.00017276888024587433,
+      "loss": 0.6666,
+      "step": 1515
+    },
+    {
+      "epoch": 4.764890282131661,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.00017251818276184012,
+      "loss": 0.6692,
+      "step": 1520
+    },
+    {
+      "epoch": 4.7805642633228835,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.0001722665202537621,
+      "loss": 0.6578,
+      "step": 1525
+    },
+    {
+      "epoch": 4.7962382445141065,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.00017201389607059863,
+      "loss": 0.6607,
+      "step": 1530
+    },
+    {
+      "epoch": 4.8119122257053295,
+      "grad_norm": 2.140625,
+      "learning_rate": 0.00017176031357410537,
+      "loss": 0.6538,
+      "step": 1535
+    },
+    {
+      "epoch": 4.827586206896552,
+      "grad_norm": 1.828125,
+      "learning_rate": 0.0001715057761387905,
+      "loss": 0.6703,
+      "step": 1540
+    },
+    {
+      "epoch": 4.843260188087775,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.00017125028715187,
+      "loss": 0.6761,
+      "step": 1545
+    },
+    {
+      "epoch": 4.858934169278997,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.0001709938500132225,
+      "loss": 0.6616,
+      "step": 1550
+    },
+    {
+      "epoch": 4.87460815047022,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.00017073646813534388,
+      "loss": 0.6597,
+      "step": 1555
+    },
+    {
+      "epoch": 4.890282131661442,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.00017047814494330207,
+      "loss": 0.6733,
+      "step": 1560
+    },
+    {
+      "epoch": 4.905956112852665,
+      "grad_norm": 1.859375,
+      "learning_rate": 0.00017021888387469135,
+      "loss": 0.6737,
+      "step": 1565
+    },
+    {
+      "epoch": 4.921630094043887,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.00016995868837958665,
+      "loss": 0.6736,
+      "step": 1570
+    },
+    {
+      "epoch": 4.93730407523511,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.0001696975619204977,
+      "loss": 0.68,
+      "step": 1575
+    },
+    {
+      "epoch": 4.952978056426332,
+      "grad_norm": 2.0,
+      "learning_rate": 0.0001694355079723227,
+      "loss": 0.6755,
+      "step": 1580
+    },
+    {
+      "epoch": 4.968652037617555,
+      "grad_norm": 0.78125,
+      "learning_rate": 0.0001691725300223025,
+      "loss": 0.6827,
+      "step": 1585
+    },
+    {
+      "epoch": 4.984326018808778,
+      "grad_norm": 0.859375,
+      "learning_rate": 0.0001689086315699738,
+      "loss": 0.6681,
+      "step": 1590
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 0.91015625,
+      "learning_rate": 0.00016864381612712276,
+      "loss": 0.6599,
+      "step": 1595
+    },
+    {
+      "epoch": 5.0,
+      "eval_loss": 2.0616867542266846,
+      "eval_runtime": 0.7937,
+      "eval_samples_per_second": 2.52,
+      "eval_steps_per_second": 1.26,
+      "step": 1595
+    },
+    {
+      "epoch": 5.015673981191223,
+      "grad_norm": 1.9296875,
+      "learning_rate": 0.00016837808721773827,
+      "loss": 0.5239,
+      "step": 1600
+    },
+    {
+      "epoch": 5.031347962382445,
+      "grad_norm": 1.125,
+      "learning_rate": 0.000168111448377965,
+      "loss": 0.509,
+      "step": 1605
+    },
+    {
+      "epoch": 5.047021943573668,
+      "grad_norm": 2.5625,
+      "learning_rate": 0.0001678439031560564,
+      "loss": 0.5041,
+      "step": 1610
+    },
+    {
+      "epoch": 5.06269592476489,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.00016757545511232746,
+      "loss": 0.5195,
+      "step": 1615
+    },
+    {
+      "epoch": 5.078369905956113,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.00016730610781910728,
+      "loss": 0.4963,
+      "step": 1620
+    },
+    {
+      "epoch": 5.094043887147335,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.00016703586486069164,
+      "loss": 0.5159,
+      "step": 1625
+    },
+    {
+      "epoch": 5.109717868338558,
+      "grad_norm": 1.0859375,
+      "learning_rate": 0.0001667647298332952,
+      "loss": 0.5111,
+      "step": 1630
+    },
+    {
+      "epoch": 5.12539184952978,
+      "grad_norm": 0.9765625,
+      "learning_rate": 0.00016649270634500366,
+      "loss": 0.5309,
+      "step": 1635
+    },
+    {
+      "epoch": 5.141065830721003,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.00016621979801572585,
+      "loss": 0.5254,
+      "step": 1640
+    },
+    {
+      "epoch": 5.156739811912225,
+      "grad_norm": 0.8359375,
+      "learning_rate": 0.00016594600847714538,
+      "loss": 0.5274,
+      "step": 1645
+    },
+    {
+      "epoch": 5.172413793103448,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.0001656713413726725,
+      "loss": 0.5239,
+      "step": 1650
+    },
+    {
+      "epoch": 5.1880877742946705,
+      "grad_norm": 0.984375,
+      "learning_rate": 0.00016539580035739547,
+      "loss": 0.5246,
+      "step": 1655
+    },
+    {
+      "epoch": 5.2037617554858935,
+      "grad_norm": 1.6953125,
+      "learning_rate": 0.00016511938909803204,
+      "loss": 0.5367,
+      "step": 1660
+    },
+    {
+      "epoch": 5.219435736677116,
+      "grad_norm": 0.9453125,
+      "learning_rate": 0.00016484211127288048,
+      "loss": 0.5356,
+      "step": 1665
+    },
+    {
+      "epoch": 5.235109717868339,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.00016456397057177085,
+      "loss": 0.5367,
+      "step": 1670
+    },
+    {
+      "epoch": 5.250783699059561,
+      "grad_norm": 1.25,
+      "learning_rate": 0.00016428497069601578,
+      "loss": 0.534,
+      "step": 1675
+    },
+    {
+      "epoch": 5.266457680250784,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.00016400511535836118,
+      "loss": 0.5476,
+      "step": 1680
+    },
+    {
+      "epoch": 5.282131661442007,
+      "grad_norm": 1.1015625,
+      "learning_rate": 0.0001637244082829369,
+      "loss": 0.544,
+      "step": 1685
+    },
+    {
+      "epoch": 5.297805642633229,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.00016344285320520717,
+      "loss": 0.5414,
+      "step": 1690
+    },
+    {
+      "epoch": 5.313479623824452,
+      "grad_norm": 1.0,
+      "learning_rate": 0.00016316045387192087,
+      "loss": 0.5435,
+      "step": 1695
+    },
+    {
+      "epoch": 5.329153605015674,
+      "grad_norm": 0.85546875,
+      "learning_rate": 0.00016287721404106167,
+      "loss": 0.5412,
+      "step": 1700
+    },
+    {
+      "epoch": 5.344827586206897,
+      "grad_norm": 1.0,
+      "learning_rate": 0.00016259313748179802,
+      "loss": 0.5448,
+      "step": 1705
+    },
+    {
+      "epoch": 5.360501567398119,
+      "grad_norm": 0.76171875,
+      "learning_rate": 0.000162308227974433,
+      "loss": 0.5523,
+      "step": 1710
+    },
+    {
+      "epoch": 5.376175548589342,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.00016202248931035404,
+      "loss": 0.5382,
+      "step": 1715
+    },
+    {
+      "epoch": 5.391849529780564,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.0001617359252919824,
+      "loss": 0.5427,
+      "step": 1720
+    },
+    {
+      "epoch": 5.407523510971787,
+      "grad_norm": 1.203125,
+      "learning_rate": 0.00016144853973272262,
+      "loss": 0.5426,
+      "step": 1725
+    },
+    {
+      "epoch": 5.423197492163009,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.00016116033645691174,
+      "loss": 0.5514,
+      "step": 1730
+    },
+    {
+      "epoch": 5.438871473354232,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.00016087131929976852,
+      "loss": 0.5471,
+      "step": 1735
+    },
+    {
+      "epoch": 5.454545454545454,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.00016058149210734223,
+      "loss": 0.5617,
+      "step": 1740
+    },
+    {
+      "epoch": 5.470219435736677,
+      "grad_norm": 0.90234375,
+      "learning_rate": 0.0001602908587364616,
+      "loss": 0.5598,
+      "step": 1745
+    },
+    {
+      "epoch": 5.485893416927899,
+      "grad_norm": 0.875,
+      "learning_rate": 0.00015999942305468338,
+      "loss": 0.5457,
+      "step": 1750
+    },
+    {
+      "epoch": 5.501567398119122,
+      "grad_norm": 0.99609375,
+      "learning_rate": 0.000159707188940241,
+      "loss": 0.5667,
+      "step": 1755
+    },
+    {
+      "epoch": 5.517241379310345,
+      "grad_norm": 2.90625,
+      "learning_rate": 0.00015941416028199298,
+      "loss": 0.552,
+      "step": 1760
+    },
+    {
+      "epoch": 5.532915360501567,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.00015912034097937094,
+      "loss": 0.5408,
+      "step": 1765
+    },
+    {
+      "epoch": 5.54858934169279,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.00015882573494232797,
+      "loss": 0.5516,
+      "step": 1770
+    },
+    {
+      "epoch": 5.564263322884012,
+      "grad_norm": 0.94921875,
+      "learning_rate": 0.00015853034609128648,
+      "loss": 0.5649,
+      "step": 1775
+    },
+    {
+      "epoch": 5.579937304075235,
+      "grad_norm": 1.25,
+      "learning_rate": 0.00015823417835708606,
+      "loss": 0.5666,
+      "step": 1780
+    },
+    {
+      "epoch": 5.5956112852664575,
+      "grad_norm": 1.515625,
+      "learning_rate": 0.00015793723568093118,
+      "loss": 0.5653,
+      "step": 1785
+    },
+    {
+      "epoch": 5.61128526645768,
+      "grad_norm": 1.4296875,
+      "learning_rate": 0.00015763952201433866,
+      "loss": 0.5663,
+      "step": 1790
+    },
+    {
+      "epoch": 5.6269592476489025,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.00015734104131908522,
+      "loss": 0.5582,
+      "step": 1795
+    },
+    {
+      "epoch": 5.6426332288401255,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.00015704179756715467,
+      "loss": 0.5569,
+      "step": 1800
+    },
+    {
+      "epoch": 5.658307210031348,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.00015674179474068508,
+      "loss": 0.5618,
+      "step": 1805
+    },
+    {
+      "epoch": 5.673981191222571,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.00015644103683191575,
+      "loss": 0.5636,
+      "step": 1810
+    },
+    {
+      "epoch": 5.689655172413794,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.00015613952784313418,
+      "loss": 0.5562,
+      "step": 1815
+    },
+    {
+      "epoch": 5.705329153605016,
+      "grad_norm": 1.7421875,
+      "learning_rate": 0.00015583727178662262,
+      "loss": 0.5566,
+      "step": 1820
+    },
+    {
+      "epoch": 5.721003134796238,
+      "grad_norm": 1.4375,
+      "learning_rate": 0.00015553427268460496,
+      "loss": 0.5591,
+      "step": 1825
+    },
+    {
+      "epoch": 5.736677115987461,
+      "grad_norm": 0.98046875,
+      "learning_rate": 0.00015523053456919294,
+      "loss": 0.5666,
+      "step": 1830
+    },
+    {
+      "epoch": 5.752351097178684,
+      "grad_norm": 0.98828125,
+      "learning_rate": 0.00015492606148233265,
+      "loss": 0.5664,
+      "step": 1835
+    },
+    {
+      "epoch": 5.768025078369906,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.00015462085747575068,
+      "loss": 0.5624,
+      "step": 1840
+    },
+    {
+      "epoch": 5.783699059561129,
+      "grad_norm": 0.85546875,
+      "learning_rate": 0.00015431492661090022,
+      "loss": 0.5587,
+      "step": 1845
+    },
+    {
+      "epoch": 5.799373040752351,
+      "grad_norm": 0.99609375,
+      "learning_rate": 0.00015400827295890702,
+      "loss": 0.5645,
+      "step": 1850
+    },
+    {
+      "epoch": 5.815047021943574,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.0001537009006005152,
+      "loss": 0.5631,
+      "step": 1855
+    },
+    {
+      "epoch": 5.830721003134796,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.0001533928136260329,
+      "loss": 0.5712,
+      "step": 1860
+    },
+    {
+      "epoch": 5.846394984326019,
+      "grad_norm": 1.140625,
+      "learning_rate": 0.00015308401613527796,
+      "loss": 0.5676,
+      "step": 1865
+    },
+    {
+      "epoch": 5.862068965517241,
+      "grad_norm": 0.9375,
+      "learning_rate": 0.00015277451223752326,
+      "loss": 0.5761,
+      "step": 1870
+    },
+    {
+      "epoch": 5.877742946708464,
+      "grad_norm": 0.98046875,
+      "learning_rate": 0.00015246430605144216,
+      "loss": 0.5685,
+      "step": 1875
+    },
+    {
+      "epoch": 5.893416927899686,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.00015215340170505348,
+      "loss": 0.5678,
+      "step": 1880
+    },
+    {
+      "epoch": 5.909090909090909,
+      "grad_norm": 1.6484375,
+      "learning_rate": 0.0001518418033356668,
+      "loss": 0.569,
+      "step": 1885
+    },
+    {
+      "epoch": 5.924764890282132,
+      "grad_norm": 0.9765625,
+      "learning_rate": 0.00015152951508982726,
+      "loss": 0.5669,
+      "step": 1890
+    },
+    {
+      "epoch": 5.940438871473354,
+      "grad_norm": 0.91015625,
+      "learning_rate": 0.0001512165411232604,
+      "loss": 0.5759,
+      "step": 1895
+    },
+    {
+      "epoch": 5.956112852664576,
+      "grad_norm": 0.87890625,
+      "learning_rate": 0.00015090288560081692,
+      "loss": 0.5641,
+      "step": 1900
+    },
+    {
+      "epoch": 5.971786833855799,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.0001505885526964172,
+      "loss": 0.574,
+      "step": 1905
+    },
+    {
+      "epoch": 5.987460815047022,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.00015027354659299578,
+      "loss": 0.5764,
+      "step": 1910
+    },
+    {
+      "epoch": 6.0,
+      "eval_loss": 2.323450803756714,
+      "eval_runtime": 0.8121,
+      "eval_samples_per_second": 2.463,
+      "eval_steps_per_second": 1.231,
+      "step": 1914
+    },
+    {
+      "epoch": 6.003134796238244,
+      "grad_norm": 0.96484375,
+      "learning_rate": 0.00014995787148244563,
+      "loss": 0.5514,
+      "step": 1915
+    },
+    {
+      "epoch": 6.018808777429467,
+      "grad_norm": 1.1484375,
+      "learning_rate": 0.00014964153156556245,
+      "loss": 0.4392,
+      "step": 1920
+    },
+    {
+      "epoch": 6.0344827586206895,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.00014932453105198884,
+      "loss": 0.4499,
+      "step": 1925
+    },
+    {
+      "epoch": 6.0501567398119125,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.0001490068741601581,
+      "loss": 0.4379,
+      "step": 1930
+    },
+    {
+      "epoch": 6.065830721003135,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.00014868856511723814,
+      "loss": 0.431,
+      "step": 1935
+    },
+    {
+      "epoch": 6.081504702194358,
+      "grad_norm": 0.98046875,
+      "learning_rate": 0.00014836960815907532,
+      "loss": 0.4428,
+      "step": 1940
+    },
+    {
+      "epoch": 6.09717868338558,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.000148050007530138,
+      "loss": 0.4482,
+      "step": 1945
+    },
+    {
+      "epoch": 6.112852664576803,
+      "grad_norm": 1.453125,
+      "learning_rate": 0.00014772976748346015,
+      "loss": 0.4478,
+      "step": 1950
+    },
+    {
+      "epoch": 6.128526645768025,
+      "grad_norm": 2.453125,
+      "learning_rate": 0.00014740889228058462,
+      "loss": 0.4414,
+      "step": 1955
+    },
+    {
+      "epoch": 6.144200626959248,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.0001470873861915065,
+      "loss": 0.4466,
+      "step": 1960
+    },
+    {
+      "epoch": 6.15987460815047,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.00014676525349461637,
+      "loss": 0.4356,
+      "step": 1965
+    },
+    {
+      "epoch": 6.175548589341693,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.00014644249847664317,
+      "loss": 0.4526,
+      "step": 1970
+    },
+    {
+      "epoch": 6.191222570532915,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.00014611912543259742,
+      "loss": 0.442,
+      "step": 1975
+    },
+    {
+      "epoch": 6.206896551724138,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.00014579513866571378,
+      "loss": 0.4463,
+      "step": 1980
+    },
+    {
+      "epoch": 6.222570532915361,
+      "grad_norm": 0.875,
+      "learning_rate": 0.00014547054248739404,
+      "loss": 0.444,
+      "step": 1985
+    },
+    {
+      "epoch": 6.238244514106583,
+      "grad_norm": 0.96484375,
+      "learning_rate": 0.0001451453412171496,
+      "loss": 0.4513,
+      "step": 1990
+    },
+    {
+      "epoch": 6.253918495297806,
+      "grad_norm": 0.9140625,
+      "learning_rate": 0.000144819539182544,
+      "loss": 0.4494,
+      "step": 1995
+    },
+    {
+      "epoch": 6.269592476489028,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.00014449314071913533,
+      "loss": 0.4493,
+      "step": 2000
+    },
+    {
+      "epoch": 6.285266457680251,
+      "grad_norm": 0.95703125,
+      "learning_rate": 0.00014416615017041868,
+      "loss": 0.4605,
+      "step": 2005
+    },
+    {
+      "epoch": 6.300940438871473,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.00014383857188776807,
+      "loss": 0.4642,
+      "step": 2010
+    },
+    {
+      "epoch": 6.316614420062696,
+      "grad_norm": 0.91015625,
+      "learning_rate": 0.00014351041023037884,
+      "loss": 0.4526,
+      "step": 2015
+    },
+    {
+      "epoch": 6.332288401253918,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.00014318166956520936,
+      "loss": 0.4634,
+      "step": 2020
+    },
+    {
+      "epoch": 6.347962382445141,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.00014285235426692315,
+      "loss": 0.4608,
+      "step": 2025
+    },
+    {
+      "epoch": 6.363636363636363,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.00014252246871783051,
+      "loss": 0.4588,
+      "step": 2030
+    },
+    {
+      "epoch": 6.379310344827586,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.00014219201730783024,
+      "loss": 0.467,
+      "step": 2035
+    },
+    {
+      "epoch": 6.394984326018808,
+      "grad_norm": 0.97265625,
+      "learning_rate": 0.0001418610044343514,
+      "loss": 0.4641,
+      "step": 2040
+    },
+    {
+      "epoch": 6.410658307210031,
+      "grad_norm": 0.9609375,
+      "learning_rate": 0.00014152943450229443,
+      "loss": 0.4647,
+      "step": 2045
+    },
+    {
+      "epoch": 6.4263322884012535,
+      "grad_norm": 0.875,
+      "learning_rate": 0.00014119731192397284,
+      "loss": 0.4627,
+      "step": 2050
+    },
+    {
+      "epoch": 6.4420062695924765,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.0001408646411190544,
+      "loss": 0.4664,
+      "step": 2055
+    },
+    {
+      "epoch": 6.4576802507836994,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.0001405314265145023,
+      "loss": 0.4697,
+      "step": 2060
+    },
+    {
+      "epoch": 6.4733542319749215,
+      "grad_norm": 0.95703125,
+      "learning_rate": 0.0001401976725445162,
+      "loss": 0.4664,
+      "step": 2065
+    },
+    {
+      "epoch": 6.4890282131661445,
+      "grad_norm": 0.9375,
+      "learning_rate": 0.0001398633836504734,
+      "loss": 0.4692,
+      "step": 2070
+    },
+    {
+      "epoch": 6.504702194357367,
+      "grad_norm": 0.8828125,
+      "learning_rate": 0.00013952856428086952,
+      "loss": 0.4658,
+      "step": 2075
+    },
+    {
+      "epoch": 6.52037617554859,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.00013919321889125941,
+      "loss": 0.4711,
+      "step": 2080
+    },
+    {
+      "epoch": 6.536050156739812,
+      "grad_norm": 0.859375,
+      "learning_rate": 0.0001388573519441979,
+      "loss": 0.4675,
+      "step": 2085
+    },
+    {
+      "epoch": 6.551724137931035,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.00013852096790918026,
+      "loss": 0.4677,
+      "step": 2090
+    },
+    {
+      "epoch": 6.567398119122257,
+      "grad_norm": 0.9140625,
+      "learning_rate": 0.00013818407126258293,
+      "loss": 0.4728,
+      "step": 2095
+    },
+    {
+      "epoch": 6.58307210031348,
+      "grad_norm": 2.75,
+      "learning_rate": 0.0001378466664876038,
+      "loss": 0.4769,
+      "step": 2100
+    },
+    {
+      "epoch": 6.598746081504702,
+      "grad_norm": 2.203125,
+      "learning_rate": 0.00013750875807420259,
+      "loss": 0.463,
+      "step": 2105
+    },
+    {
+      "epoch": 6.614420062695925,
+      "grad_norm": 1.1015625,
+      "learning_rate": 0.00013717035051904114,
+      "loss": 0.4663,
+      "step": 2110
+    },
+    {
+      "epoch": 6.630094043887147,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.00013683144832542352,
+      "loss": 0.4699,
+      "step": 2115
+    },
+    {
+      "epoch": 6.64576802507837,
+      "grad_norm": 0.96875,
+      "learning_rate": 0.00013649205600323609,
+      "loss": 0.4703,
+      "step": 2120
+    },
+    {
+      "epoch": 6.661442006269592,
+      "grad_norm": 0.91015625,
+      "learning_rate": 0.00013615217806888755,
+      "loss": 0.4643,
+      "step": 2125
+    },
+    {
+      "epoch": 6.677115987460815,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.0001358118190452488,
+      "loss": 0.4669,
+      "step": 2130
+    },
+    {
+      "epoch": 6.692789968652038,
+      "grad_norm": 0.9375,
+      "learning_rate": 0.0001354709834615928,
+      "loss": 0.4703,
+      "step": 2135
+    },
+    {
+      "epoch": 6.70846394984326,
+      "grad_norm": 0.875,
+      "learning_rate": 0.00013512967585353413,
+      "loss": 0.4714,
+      "step": 2140
+    },
+    {
+      "epoch": 6.724137931034483,
+      "grad_norm": 0.95703125,
+      "learning_rate": 0.00013478790076296892,
+      "loss": 0.4658,
+      "step": 2145
+    },
+    {
+      "epoch": 6.739811912225705,
+      "grad_norm": 0.97265625,
+      "learning_rate": 0.00013444566273801414,
+      "loss": 0.4649,
+      "step": 2150
+    },
+    {
+      "epoch": 6.755485893416928,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.00013410296633294727,
+      "loss": 0.4783,
+      "step": 2155
+    },
+    {
+      "epoch": 6.77115987460815,
+      "grad_norm": 1.171875,
+      "learning_rate": 0.00013375981610814545,
+      "loss": 0.4741,
+      "step": 2160
+    },
+    {
+      "epoch": 6.786833855799373,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.00013341621663002514,
+      "loss": 0.4651,
+      "step": 2165
+    },
+    {
+      "epoch": 6.802507836990595,
+      "grad_norm": 0.9765625,
+      "learning_rate": 0.0001330721724709811,
+      "loss": 0.4682,
+      "step": 2170
+    },
+    {
+      "epoch": 6.818181818181818,
+      "grad_norm": 0.8671875,
+      "learning_rate": 0.00013272768820932554,
+      "loss": 0.4761,
+      "step": 2175
+    },
+    {
+      "epoch": 6.83385579937304,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.0001323827684292273,
+      "loss": 0.467,
+      "step": 2180
+    },
+    {
+      "epoch": 6.849529780564263,
+      "grad_norm": 0.9453125,
+      "learning_rate": 0.0001320374177206509,
+      "loss": 0.4719,
+      "step": 2185
+    },
+    {
+      "epoch": 6.8652037617554855,
+      "grad_norm": 0.87890625,
+      "learning_rate": 0.00013169164067929526,
+      "loss": 0.4829,
+      "step": 2190
+    },
+    {
+      "epoch": 6.8808777429467085,
+      "grad_norm": 0.9140625,
+      "learning_rate": 0.00013134544190653274,
+      "loss": 0.4743,
+      "step": 2195
+    },
+    {
+      "epoch": 6.896551724137931,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.00013099882600934773,
+      "loss": 0.4701,
+      "step": 2200
+    },
+    {
+      "epoch": 6.912225705329154,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.00013065179760027556,
+      "loss": 0.4703,
+      "step": 2205
+    },
+    {
+      "epoch": 6.927899686520377,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.00013030436129734082,
+      "loss": 0.4802,
+      "step": 2210
+    },
+    {
+      "epoch": 6.943573667711599,
+      "grad_norm": 1.2265625,
+      "learning_rate": 0.00012995652172399623,
+      "loss": 0.4781,
+      "step": 2215
+    },
+    {
+      "epoch": 6.959247648902822,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.00012960828350906095,
+      "loss": 0.4838,
+      "step": 2220
+    },
+    {
+      "epoch": 6.974921630094044,
+      "grad_norm": 1.546875,
+      "learning_rate": 0.00012925965128665897,
+      "loss": 0.4751,
+      "step": 2225
+    },
+    {
+      "epoch": 6.990595611285267,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.0001289106296961574,
+      "loss": 0.4817,
+      "step": 2230
+    },
+    {
+      "epoch": 7.0,
+      "eval_loss": 2.6759016513824463,
+      "eval_runtime": 0.7898,
+      "eval_samples_per_second": 2.532,
+      "eval_steps_per_second": 1.266,
+      "step": 2233
+    },
+    {
+      "epoch": 7.006269592476489,
+      "grad_norm": 0.78125,
+      "learning_rate": 0.00012856122338210493,
+      "loss": 0.4256,
+      "step": 2235
+    },
+    {
+      "epoch": 7.021943573667712,
+      "grad_norm": 0.8828125,
+      "learning_rate": 0.00012821143699416984,
+      "loss": 0.3775,
+      "step": 2240
+    },
+    {
+      "epoch": 7.037617554858934,
+      "grad_norm": 0.73828125,
+      "learning_rate": 0.00012786127518707818,
+      "loss": 0.3705,
+      "step": 2245
+    },
+    {
+      "epoch": 7.053291536050157,
+      "grad_norm": 0.859375,
+      "learning_rate": 0.00012751074262055178,
+      "loss": 0.3732,
+      "step": 2250
+    },
+    {
+      "epoch": 7.068965517241379,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.00012715984395924643,
+      "loss": 0.3737,
+      "step": 2255
+    },
+    {
+      "epoch": 7.084639498432602,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.00012680858387268952,
+      "loss": 0.3788,
+      "step": 2260
+    },
+    {
+      "epoch": 7.100313479623824,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.00012645696703521818,
+      "loss": 0.3711,
+      "step": 2265
+    },
+    {
+      "epoch": 7.115987460815047,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.00012610499812591673,
+      "loss": 0.3725,
+      "step": 2270
+    },
+    {
+      "epoch": 7.131661442006269,
+      "grad_norm": 0.90234375,
+      "learning_rate": 0.0001257526818285549,
+      "loss": 0.371,
+      "step": 2275
+    },
+    {
+      "epoch": 7.147335423197492,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.0001254000228315251,
+      "loss": 0.3751,
+      "step": 2280
+    },
+    {
+      "epoch": 7.163009404388715,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.00012504702582778008,
+      "loss": 0.3798,
+      "step": 2285
+    },
+    {
+      "epoch": 7.178683385579937,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.00012469369551477074,
+      "loss": 0.3826,
+      "step": 2290
+    },
+    {
+      "epoch": 7.19435736677116,
+      "grad_norm": 0.8671875,
+      "learning_rate": 0.0001243400365943833,
+      "loss": 0.3846,
+      "step": 2295
+    },
+    {
+      "epoch": 7.210031347962382,
+      "grad_norm": 0.87890625,
+      "learning_rate": 0.00012398605377287694,
+      "loss": 0.383,
+      "step": 2300
+    },
+    {
+      "epoch": 7.225705329153605,
+      "grad_norm": 0.78125,
+      "learning_rate": 0.000123631751760821,
+      "loss": 0.387,
+      "step": 2305
+    },
+    {
+      "epoch": 7.241379310344827,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.00012327713527303255,
+      "loss": 0.3752,
+      "step": 2310
+    },
+    {
+      "epoch": 7.25705329153605,
+      "grad_norm": 0.9453125,
+      "learning_rate": 0.0001229222090285134,
+      "loss": 0.3832,
+      "step": 2315
+    },
+    {
+      "epoch": 7.2727272727272725,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.00012256697775038741,
+      "loss": 0.39,
+      "step": 2320
+    },
+    {
+      "epoch": 7.2884012539184955,
+      "grad_norm": 0.828125,
+      "learning_rate": 0.00012221144616583765,
+      "loss": 0.3902,
+      "step": 2325
+    },
+    {
+      "epoch": 7.304075235109718,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.00012185561900604341,
+      "loss": 0.376,
+      "step": 2330
+    },
+    {
+      "epoch": 7.3197492163009406,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.00012149950100611738,
+      "loss": 0.3873,
+      "step": 2335
+    },
+    {
+      "epoch": 7.335423197492163,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.00012114309690504249,
+      "loss": 0.388,
+      "step": 2340
+    },
+    {
+      "epoch": 7.351097178683386,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.00012078641144560898,
+      "loss": 0.3889,
+      "step": 2345
+    },
+    {
+      "epoch": 7.366771159874608,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.00012042944937435116,
+      "loss": 0.395,
+      "step": 2350
+    },
+    {
+      "epoch": 7.382445141065831,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.00012007221544148435,
+      "loss": 0.3957,
+      "step": 2355
+    },
+    {
+      "epoch": 7.398119122257054,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.00011971471440084157,
+      "loss": 0.393,
+      "step": 2360
+    },
+    {
+      "epoch": 7.413793103448276,
+      "grad_norm": 0.91015625,
+      "learning_rate": 0.00011935695100981041,
+      "loss": 0.3884,
+      "step": 2365
+    },
+    {
+      "epoch": 7.429467084639499,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.00011899893002926958,
+      "loss": 0.3907,
+      "step": 2370
+    },
+    {
+      "epoch": 7.445141065830721,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.00011864065622352568,
+      "loss": 0.3865,
+      "step": 2375
+    },
+    {
+      "epoch": 7.460815047021944,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.00011828213436024968,
+      "loss": 0.3866,
+      "step": 2380
+    },
+    {
+      "epoch": 7.476489028213166,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.00011792336921041359,
+      "loss": 0.3878,
+      "step": 2385
+    },
+    {
+      "epoch": 7.492163009404389,
+      "grad_norm": 0.9140625,
+      "learning_rate": 0.00011756436554822685,
+      "loss": 0.3933,
+      "step": 2390
+    },
+    {
+      "epoch": 7.507836990595611,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.00011720512815107292,
+      "loss": 0.3825,
+      "step": 2395
+    },
+    {
+      "epoch": 7.523510971786834,
+      "grad_norm": 0.82421875,
+      "learning_rate": 0.00011684566179944567,
+      "loss": 0.3892,
+      "step": 2400
+    },
+    {
+      "epoch": 7.539184952978056,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.00011648597127688567,
+      "loss": 0.3918,
+      "step": 2405
+    },
+    {
+      "epoch": 7.554858934169279,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.00011612606136991665,
+      "loss": 0.3952,
+      "step": 2410
+    },
+    {
+      "epoch": 7.570532915360501,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.00011576593686798181,
+      "loss": 0.3941,
+      "step": 2415
+    },
+    {
+      "epoch": 7.586206896551724,
+      "grad_norm": 1.1015625,
+      "learning_rate": 0.00011540560256337995,
+      "loss": 0.3996,
+      "step": 2420
+    },
+    {
+      "epoch": 7.601880877742946,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.00011504506325120184,
+      "loss": 0.3973,
+      "step": 2425
+    },
+    {
+      "epoch": 7.617554858934169,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.0001146843237292663,
+      "loss": 0.3944,
+      "step": 2430
+    },
+    {
+      "epoch": 7.633228840125392,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.0001143233887980565,
+      "loss": 0.4008,
+      "step": 2435
+    },
+    {
+      "epoch": 7.648902821316614,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.00011396226326065593,
+      "loss": 0.3967,
+      "step": 2440
+    },
+    {
+      "epoch": 7.664576802507837,
+      "grad_norm": 0.87890625,
+      "learning_rate": 0.00011360095192268454,
+      "loss": 0.3979,
+      "step": 2445
+    },
+    {
+      "epoch": 7.6802507836990594,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.00011323945959223477,
+      "loss": 0.3957,
+      "step": 2450
+    },
+    {
+      "epoch": 7.695924764890282,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.00011287779107980766,
+      "loss": 0.398,
+      "step": 2455
+    },
+    {
+      "epoch": 7.7115987460815045,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.0001125159511982487,
+      "loss": 0.3976,
+      "step": 2460
+    },
+    {
+      "epoch": 7.7272727272727275,
+      "grad_norm": 0.8671875,
+      "learning_rate": 0.00011215394476268387,
+      "loss": 0.3941,
+      "step": 2465
+    },
+    {
+      "epoch": 7.74294670846395,
+      "grad_norm": 1.0390625,
+      "learning_rate": 0.00011179177659045554,
+      "loss": 0.4061,
+      "step": 2470
+    },
+    {
+      "epoch": 7.758620689655173,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.00011142945150105839,
+      "loss": 0.3991,
+      "step": 2475
+    },
+    {
+      "epoch": 7.774294670846395,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.00011106697431607518,
+      "loss": 0.3863,
+      "step": 2480
+    },
+    {
+      "epoch": 7.789968652037618,
+      "grad_norm": 0.90625,
+      "learning_rate": 0.00011070434985911271,
+      "loss": 0.3923,
+      "step": 2485
+    },
+    {
+      "epoch": 7.80564263322884,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.0001103415829557376,
+      "loss": 0.3988,
+      "step": 2490
+    },
+    {
+      "epoch": 7.821316614420063,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.00010997867843341198,
+      "loss": 0.3868,
+      "step": 2495
+    },
+    {
+      "epoch": 7.836990595611285,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.0001096156411214294,
+      "loss": 0.3949,
+      "step": 2500
+    },
+    {
+      "epoch": 7.852664576802508,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.00010925247585085044,
+      "loss": 0.4005,
+      "step": 2505
+    },
+    {
+      "epoch": 7.868338557993731,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.00010888918745443845,
+      "loss": 0.3974,
+      "step": 2510
+    },
+    {
+      "epoch": 7.884012539184953,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.00010852578076659535,
+      "loss": 0.4004,
+      "step": 2515
+    },
+    {
+      "epoch": 7.899686520376176,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.00010816226062329706,
+      "loss": 0.3997,
+      "step": 2520
+    },
+    {
+      "epoch": 7.915360501567398,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.00010779863186202943,
+      "loss": 0.4002,
+      "step": 2525
+    },
+    {
+      "epoch": 7.931034482758621,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.00010743489932172366,
+      "loss": 0.3973,
+      "step": 2530
+    },
+    {
+      "epoch": 7.946708463949843,
+      "grad_norm": 0.9453125,
+      "learning_rate": 0.00010707106784269196,
+      "loss": 0.3968,
+      "step": 2535
+    },
+    {
+      "epoch": 7.962382445141066,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.00010670714226656315,
+      "loss": 0.397,
+      "step": 2540
+    },
+    {
+      "epoch": 7.978056426332288,
+      "grad_norm": 0.90625,
+      "learning_rate": 0.00010634312743621832,
+      "loss": 0.4065,
+      "step": 2545
+    },
+    {
+      "epoch": 7.993730407523511,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.00010597902819572619,
+      "loss": 0.4016,
+      "step": 2550
+    },
+    {
+      "epoch": 8.0,
+      "eval_loss": 3.0132577419281006,
+      "eval_runtime": 0.7738,
+      "eval_samples_per_second": 2.585,
+      "eval_steps_per_second": 1.292,
+      "step": 2552
+    },
+    {
+      "epoch": 8.009404388714733,
+      "grad_norm": 0.76953125,
+      "learning_rate": 0.00010561484939027877,
+      "loss": 0.361,
+      "step": 2555
+    },
+    {
+      "epoch": 8.025078369905955,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.00010525059586612693,
+      "loss": 0.3253,
+      "step": 2560
+    },
+    {
+      "epoch": 8.04075235109718,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.0001048862724705158,
+      "loss": 0.3205,
+      "step": 2565
+    },
+    {
+      "epoch": 8.056426332288401,
+      "grad_norm": 0.72265625,
+      "learning_rate": 0.00010452188405162033,
+      "loss": 0.3241,
+      "step": 2570
+    },
+    {
+      "epoch": 8.072100313479623,
+      "grad_norm": 0.828125,
+      "learning_rate": 0.00010415743545848072,
+      "loss": 0.3193,
+      "step": 2575
+    },
+    {
+      "epoch": 8.087774294670846,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.00010379293154093796,
+      "loss": 0.3212,
+      "step": 2580
+    },
+    {
+      "epoch": 8.10344827586207,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.00010342837714956928,
+      "loss": 0.3161,
+      "step": 2585
+    },
+    {
+      "epoch": 8.119122257053291,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.00010306377713562354,
+      "loss": 0.321,
+      "step": 2590
+    },
+    {
+      "epoch": 8.134796238244514,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.00010269913635095676,
+      "loss": 0.3245,
+      "step": 2595
+    },
+    {
+      "epoch": 8.150470219435737,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.00010233445964796749,
+      "loss": 0.3279,
+      "step": 2600
+    },
+    {
+      "epoch": 8.16614420062696,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.00010196975187953221,
+      "loss": 0.3233,
+      "step": 2605
+    },
+    {
+      "epoch": 8.181818181818182,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.00010160501789894086,
+      "loss": 0.3207,
+      "step": 2610
+    },
+    {
+      "epoch": 8.197492163009404,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.0001012402625598322,
+      "loss": 0.3217,
+      "step": 2615
+    },
+    {
+      "epoch": 8.213166144200628,
+      "grad_norm": 0.76953125,
+      "learning_rate": 0.00010087549071612919,
+      "loss": 0.3255,
+      "step": 2620
+    },
+    {
+      "epoch": 8.22884012539185,
+      "grad_norm": 0.82421875,
+      "learning_rate": 0.00010051070722197438,
+      "loss": 0.3256,
+      "step": 2625
+    },
+    {
+      "epoch": 8.244514106583072,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.0001001459169316654,
+      "loss": 0.3233,
+      "step": 2630
+    },
+    {
+      "epoch": 8.260188087774294,
+      "grad_norm": 0.8046875,
+      "learning_rate": 9.978112469959033e-05,
+      "loss": 0.3232,
+      "step": 2635
+    },
+    {
+      "epoch": 8.275862068965518,
+      "grad_norm": 0.80859375,
+      "learning_rate": 9.941633538016315e-05,
+      "loss": 0.3237,
+      "step": 2640
+    },
+    {
+      "epoch": 8.29153605015674,
+      "grad_norm": 0.79296875,
+      "learning_rate": 9.90515538277589e-05,
+      "loss": 0.3307,
+      "step": 2645
+    },
+    {
+      "epoch": 8.307210031347962,
+      "grad_norm": 0.765625,
+      "learning_rate": 9.868678489664945e-05,
+      "loss": 0.3228,
+      "step": 2650
+    },
+    {
+      "epoch": 8.322884012539184,
+      "grad_norm": 0.75390625,
+      "learning_rate": 9.832203344093855e-05,
+      "loss": 0.3238,
+      "step": 2655
+    },
+    {
+      "epoch": 8.338557993730408,
+      "grad_norm": 0.78125,
+      "learning_rate": 9.795730431449759e-05,
+      "loss": 0.3302,
+      "step": 2660
+    },
+    {
+      "epoch": 8.35423197492163,
+      "grad_norm": 0.734375,
+      "learning_rate": 9.759260237090058e-05,
+      "loss": 0.3243,
+      "step": 2665
+    },
+    {
+      "epoch": 8.369905956112852,
+      "grad_norm": 0.84765625,
+      "learning_rate": 9.722793246336006e-05,
+      "loss": 0.3255,
+      "step": 2670
+    },
+    {
+      "epoch": 8.385579937304076,
+      "grad_norm": 0.7734375,
+      "learning_rate": 9.686329944466203e-05,
+      "loss": 0.323,
+      "step": 2675
+    },
+    {
+      "epoch": 8.401253918495298,
+      "grad_norm": 0.8125,
+      "learning_rate": 9.649870816710172e-05,
+      "loss": 0.3278,
+      "step": 2680
+    },
+    {
+      "epoch": 8.41692789968652,
+      "grad_norm": 0.875,
+      "learning_rate": 9.613416348241887e-05,
+      "loss": 0.3282,
+      "step": 2685
+    },
+    {
+      "epoch": 8.432601880877742,
+      "grad_norm": 0.90234375,
+      "learning_rate": 9.576967024173323e-05,
+      "loss": 0.3226,
+      "step": 2690
+    },
+    {
+      "epoch": 8.448275862068966,
+      "grad_norm": 0.80078125,
+      "learning_rate": 9.540523329547984e-05,
+      "loss": 0.3361,
+      "step": 2695
+    },
+    {
+      "epoch": 8.463949843260188,
+      "grad_norm": 0.8671875,
+      "learning_rate": 9.504085749334479e-05,
+      "loss": 0.3309,
+      "step": 2700
+    },
+    {
+      "epoch": 8.47962382445141,
+      "grad_norm": 0.74609375,
+      "learning_rate": 9.467654768420032e-05,
+      "loss": 0.325,
+      "step": 2705
+    },
+    {
+      "epoch": 8.495297805642632,
+      "grad_norm": 0.78515625,
+      "learning_rate": 9.431230871604067e-05,
+      "loss": 0.3265,
+      "step": 2710
+    },
+    {
+      "epoch": 8.510971786833856,
+      "grad_norm": 0.8203125,
+      "learning_rate": 9.394814543591719e-05,
+      "loss": 0.3302,
+      "step": 2715
+    },
+    {
+      "epoch": 8.526645768025078,
+      "grad_norm": 0.77734375,
+      "learning_rate": 9.358406268987417e-05,
+      "loss": 0.3299,
+      "step": 2720
+    },
+    {
+      "epoch": 8.5423197492163,
+      "grad_norm": 0.80859375,
+      "learning_rate": 9.322006532288411e-05,
+      "loss": 0.3303,
+      "step": 2725
+    },
+    {
+      "epoch": 8.557993730407524,
+      "grad_norm": 0.87109375,
+      "learning_rate": 9.285615817878342e-05,
+      "loss": 0.3246,
+      "step": 2730
+    },
+    {
+      "epoch": 8.573667711598747,
+      "grad_norm": 0.7578125,
+      "learning_rate": 9.249234610020779e-05,
+      "loss": 0.3256,
+      "step": 2735
+    },
+    {
+      "epoch": 8.589341692789969,
+      "grad_norm": 0.80859375,
+      "learning_rate": 9.212863392852793e-05,
+      "loss": 0.3286,
+      "step": 2740
+    },
+    {
+      "epoch": 8.60501567398119,
+      "grad_norm": 0.8359375,
+      "learning_rate": 9.176502650378499e-05,
+      "loss": 0.3301,
+      "step": 2745
+    },
+    {
+      "epoch": 8.620689655172415,
+      "grad_norm": 0.73828125,
+      "learning_rate": 9.140152866462629e-05,
+      "loss": 0.3345,
+      "step": 2750
+    },
+    {
+      "epoch": 8.636363636363637,
+      "grad_norm": 0.76171875,
+      "learning_rate": 9.103814524824073e-05,
+      "loss": 0.3335,
+      "step": 2755
+    },
+    {
+      "epoch": 8.652037617554859,
+      "grad_norm": 0.73828125,
+      "learning_rate": 9.067488109029474e-05,
+      "loss": 0.3287,
+      "step": 2760
+    },
+    {
+      "epoch": 8.66771159874608,
+      "grad_norm": 0.77734375,
+      "learning_rate": 9.031174102486752e-05,
+      "loss": 0.3286,
+      "step": 2765
+    },
+    {
+      "epoch": 8.683385579937305,
+      "grad_norm": 0.79296875,
+      "learning_rate": 8.994872988438711e-05,
+      "loss": 0.3283,
+      "step": 2770
+    },
+    {
+      "epoch": 8.699059561128527,
+      "grad_norm": 0.80078125,
+      "learning_rate": 8.958585249956578e-05,
+      "loss": 0.3308,
+      "step": 2775
+    },
+    {
+      "epoch": 8.714733542319749,
+      "grad_norm": 0.859375,
+      "learning_rate": 8.922311369933598e-05,
+      "loss": 0.3308,
+      "step": 2780
+    },
+    {
+      "epoch": 8.730407523510971,
+      "grad_norm": 0.859375,
+      "learning_rate": 8.886051831078582e-05,
+      "loss": 0.3314,
+      "step": 2785
+    },
+    {
+      "epoch": 8.746081504702195,
+      "grad_norm": 0.8046875,
+      "learning_rate": 8.849807115909513e-05,
+      "loss": 0.33,
+      "step": 2790
+    },
+    {
+      "epoch": 8.761755485893417,
+      "grad_norm": 0.86328125,
+      "learning_rate": 8.8135777067471e-05,
+      "loss": 0.329,
+      "step": 2795
+    },
+    {
+      "epoch": 8.77742946708464,
+      "grad_norm": 0.84375,
+      "learning_rate": 8.777364085708378e-05,
+      "loss": 0.3356,
+      "step": 2800
+    },
+    {
+      "epoch": 8.793103448275861,
+      "grad_norm": 0.83984375,
+      "learning_rate": 8.741166734700273e-05,
+      "loss": 0.3384,
+      "step": 2805
+    },
+    {
+      "epoch": 8.808777429467085,
+      "grad_norm": 0.80859375,
+      "learning_rate": 8.704986135413212e-05,
+      "loss": 0.3324,
+      "step": 2810
+    },
+    {
+      "epoch": 8.824451410658307,
+      "grad_norm": 0.78125,
+      "learning_rate": 8.668822769314691e-05,
+      "loss": 0.3291,
+      "step": 2815
+    },
+    {
+      "epoch": 8.84012539184953,
+      "grad_norm": 0.78515625,
+      "learning_rate": 8.632677117642892e-05,
+      "loss": 0.3358,
+      "step": 2820
+    },
+    {
+      "epoch": 8.855799373040753,
+      "grad_norm": 0.90625,
+      "learning_rate": 8.596549661400248e-05,
+      "loss": 0.3331,
+      "step": 2825
+    },
+    {
+      "epoch": 8.871473354231975,
+      "grad_norm": 0.7890625,
+      "learning_rate": 8.560440881347071e-05,
+      "loss": 0.3306,
+      "step": 2830
+    },
+    {
+      "epoch": 8.887147335423197,
+      "grad_norm": 0.796875,
+      "learning_rate": 8.524351257995135e-05,
+      "loss": 0.3322,
+      "step": 2835
+    },
+    {
+      "epoch": 8.90282131661442,
+      "grad_norm": 0.78515625,
+      "learning_rate": 8.488281271601302e-05,
+      "loss": 0.3246,
+      "step": 2840
+    },
+    {
+      "epoch": 8.918495297805643,
+      "grad_norm": 0.81640625,
+      "learning_rate": 8.452231402161099e-05,
+      "loss": 0.3339,
+      "step": 2845
+    },
+    {
+      "epoch": 8.934169278996865,
+      "grad_norm": 0.74609375,
+      "learning_rate": 8.416202129402371e-05,
+      "loss": 0.3291,
+      "step": 2850
+    },
+    {
+      "epoch": 8.949843260188088,
+      "grad_norm": 0.765625,
+      "learning_rate": 8.380193932778857e-05,
+      "loss": 0.3268,
+      "step": 2855
+    },
+    {
+      "epoch": 8.96551724137931,
+      "grad_norm": 0.84375,
+      "learning_rate": 8.344207291463843e-05,
+      "loss": 0.3258,
+      "step": 2860
+    },
+    {
+      "epoch": 8.981191222570533,
+      "grad_norm": 0.77734375,
+      "learning_rate": 8.30824268434376e-05,
+      "loss": 0.3345,
+      "step": 2865
+    },
+    {
+      "epoch": 8.996865203761756,
+      "grad_norm": 0.796875,
+      "learning_rate": 8.27230059001184e-05,
+      "loss": 0.327,
+      "step": 2870
+    },
+    {
+      "epoch": 9.0,
+      "eval_loss": 3.4537088871002197,
+      "eval_runtime": 0.7983,
+      "eval_samples_per_second": 2.505,
+      "eval_steps_per_second": 1.253,
+      "step": 2871
+    },
+    {
+      "epoch": 9.012539184952978,
+      "grad_norm": 0.5625,
+      "learning_rate": 8.2363814867617e-05,
+      "loss": 0.2932,
+      "step": 2875
+    },
+    {
+      "epoch": 9.0282131661442,
+      "grad_norm": 0.7734375,
+      "learning_rate": 8.200485852581036e-05,
+      "loss": 0.2798,
+      "step": 2880
+    },
+    {
+      "epoch": 9.043887147335424,
+      "grad_norm": 0.78125,
+      "learning_rate": 8.16461416514522e-05,
+      "loss": 0.2751,
+      "step": 2885
+    },
+    {
+      "epoch": 9.059561128526646,
+      "grad_norm": 0.71484375,
+      "learning_rate": 8.12876690181096e-05,
+      "loss": 0.2771,
+      "step": 2890
+    },
+    {
+      "epoch": 9.075235109717868,
+      "grad_norm": 0.66796875,
+      "learning_rate": 8.092944539609937e-05,
+      "loss": 0.281,
+      "step": 2895
+    },
+    {
+      "epoch": 9.090909090909092,
+      "grad_norm": 0.7109375,
+      "learning_rate": 8.057147555242473e-05,
+      "loss": 0.2709,
+      "step": 2900
+    },
+    {
+      "epoch": 9.106583072100314,
+      "grad_norm": 0.6875,
+      "learning_rate": 8.021376425071175e-05,
+      "loss": 0.2787,
+      "step": 2905
+    },
+    {
+      "epoch": 9.122257053291536,
+      "grad_norm": 0.64453125,
+      "learning_rate": 7.985631625114603e-05,
+      "loss": 0.2817,
+      "step": 2910
+    },
+    {
+      "epoch": 9.137931034482758,
+      "grad_norm": 0.68359375,
+      "learning_rate": 7.94991363104092e-05,
+      "loss": 0.2798,
+      "step": 2915
+    },
+    {
+      "epoch": 9.153605015673982,
+      "grad_norm": 0.72265625,
+      "learning_rate": 7.914222918161589e-05,
+      "loss": 0.2804,
+      "step": 2920
+    },
+    {
+      "epoch": 9.169278996865204,
+      "grad_norm": 0.71484375,
+      "learning_rate": 7.878559961425025e-05,
+      "loss": 0.2778,
+      "step": 2925
+    },
+    {
+      "epoch": 9.184952978056426,
+      "grad_norm": 0.67578125,
+      "learning_rate": 7.842925235410288e-05,
+      "loss": 0.2832,
+      "step": 2930
+    },
+    {
+      "epoch": 9.200626959247648,
+      "grad_norm": 0.703125,
+      "learning_rate": 7.807319214320747e-05,
+      "loss": 0.2781,
+      "step": 2935
+    },
+    {
+      "epoch": 9.216300940438872,
+      "grad_norm": 0.67578125,
+      "learning_rate": 7.771742371977811e-05,
+      "loss": 0.277,
+      "step": 2940
+    },
+    {
+      "epoch": 9.231974921630094,
+      "grad_norm": 0.69921875,
+      "learning_rate": 7.73619518181457e-05,
+      "loss": 0.2701,
+      "step": 2945
+    },
+    {
+      "epoch": 9.247648902821316,
+      "grad_norm": 0.73046875,
+      "learning_rate": 7.700678116869543e-05,
+      "loss": 0.2771,
+      "step": 2950
+    },
+    {
+      "epoch": 9.263322884012538,
+      "grad_norm": 0.7265625,
+      "learning_rate": 7.66519164978035e-05,
+      "loss": 0.2738,
+      "step": 2955
+    },
+    {
+      "epoch": 9.278996865203762,
+      "grad_norm": 0.77734375,
+      "learning_rate": 7.629736252777445e-05,
+      "loss": 0.2735,
+      "step": 2960
+    },
+    {
+      "epoch": 9.294670846394984,
+      "grad_norm": 0.6875,
+      "learning_rate": 7.594312397677809e-05,
+      "loss": 0.2773,
+      "step": 2965
+    },
+    {
+      "epoch": 9.310344827586206,
+      "grad_norm": 0.68359375,
+      "learning_rate": 7.558920555878696e-05,
+      "loss": 0.2764,
+      "step": 2970
+    },
+    {
+      "epoch": 9.32601880877743,
+      "grad_norm": 0.83984375,
+      "learning_rate": 7.523561198351342e-05,
+      "loss": 0.2828,
+      "step": 2975
+    },
+    {
+      "epoch": 9.341692789968652,
+      "grad_norm": 0.74609375,
+      "learning_rate": 7.48823479563471e-05,
+      "loss": 0.2834,
+      "step": 2980
+    },
+    {
+      "epoch": 9.357366771159874,
+      "grad_norm": 0.87109375,
+      "learning_rate": 7.452941817829212e-05,
+      "loss": 0.2848,
+      "step": 2985
+    },
+    {
+      "epoch": 9.373040752351097,
+      "grad_norm": 0.66015625,
+      "learning_rate": 7.417682734590469e-05,
+      "loss": 0.2801,
+      "step": 2990
+    },
+    {
+      "epoch": 9.38871473354232,
+      "grad_norm": 0.71875,
+      "learning_rate": 7.382458015123057e-05,
+      "loss": 0.2822,
+      "step": 2995
+    },
+    {
+      "epoch": 9.404388714733543,
+      "grad_norm": 0.69921875,
+      "learning_rate": 7.347268128174265e-05,
+      "loss": 0.2736,
+      "step": 3000
+    },
+    {
+      "epoch": 9.420062695924765,
+      "grad_norm": 0.703125,
+      "learning_rate": 7.31211354202784e-05,
+      "loss": 0.2801,
+      "step": 3005
+    },
+    {
+      "epoch": 9.435736677115987,
+      "grad_norm": 0.703125,
+      "learning_rate": 7.276994724497787e-05,
+      "loss": 0.2799,
+      "step": 3010
+    },
+    {
+      "epoch": 9.45141065830721,
+      "grad_norm": 0.6953125,
+      "learning_rate": 7.241912142922109e-05,
+      "loss": 0.2768,
+      "step": 3015
+    },
+    {
+      "epoch": 9.467084639498433,
+      "grad_norm": 0.72265625,
+      "learning_rate": 7.206866264156623e-05,
+      "loss": 0.2851,
+      "step": 3020
+    },
+    {
+      "epoch": 9.482758620689655,
+      "grad_norm": 0.78125,
+      "learning_rate": 7.171857554568706e-05,
+      "loss": 0.284,
+      "step": 3025
+    },
+    {
+      "epoch": 9.498432601880877,
+      "grad_norm": 0.73046875,
+      "learning_rate": 7.136886480031138e-05,
+      "loss": 0.2826,
+      "step": 3030
+    },
+    {
+      "epoch": 9.5141065830721,
+      "grad_norm": 0.7265625,
+      "learning_rate": 7.101953505915857e-05,
+      "loss": 0.2823,
+      "step": 3035
+    },
+    {
+      "epoch": 9.529780564263323,
+      "grad_norm": 0.7109375,
+      "learning_rate": 7.067059097087796e-05,
+      "loss": 0.2767,
+      "step": 3040
+    },
+    {
+      "epoch": 9.545454545454545,
+      "grad_norm": 0.71484375,
+      "learning_rate": 7.03220371789868e-05,
+      "loss": 0.2808,
+      "step": 3045
+    },
+    {
+      "epoch": 9.561128526645769,
+      "grad_norm": 0.6953125,
+      "learning_rate": 6.997387832180864e-05,
+      "loss": 0.2847,
+      "step": 3050
+    },
+    {
+      "epoch": 9.576802507836991,
+      "grad_norm": 0.7109375,
+      "learning_rate": 6.962611903241132e-05,
+      "loss": 0.2829,
+      "step": 3055
+    },
+    {
+      "epoch": 9.592476489028213,
+      "grad_norm": 0.6875,
+      "learning_rate": 6.92787639385457e-05,
+      "loss": 0.2815,
+      "step": 3060
+    },
+    {
+      "epoch": 9.608150470219435,
+      "grad_norm": 0.94140625,
+      "learning_rate": 6.893181766258373e-05,
+      "loss": 0.281,
+      "step": 3065
+    },
+    {
+      "epoch": 9.623824451410659,
+      "grad_norm": 0.69140625,
+      "learning_rate": 6.858528482145716e-05,
+      "loss": 0.2807,
+      "step": 3070
+    },
+    {
+      "epoch": 9.639498432601881,
+      "grad_norm": 0.6953125,
+      "learning_rate": 6.823917002659596e-05,
+      "loss": 0.2884,
+      "step": 3075
+    },
+    {
+      "epoch": 9.655172413793103,
+      "grad_norm": 0.6484375,
+      "learning_rate": 6.789347788386706e-05,
+      "loss": 0.2789,
+      "step": 3080
+    },
+    {
+      "epoch": 9.670846394984325,
+      "grad_norm": 0.73828125,
+      "learning_rate": 6.754821299351299e-05,
+      "loss": 0.2733,
+      "step": 3085
+    },
+    {
+      "epoch": 9.68652037617555,
+      "grad_norm": 0.6953125,
+      "learning_rate": 6.720337995009076e-05,
+      "loss": 0.2876,
+      "step": 3090
+    },
+    {
+      "epoch": 9.702194357366771,
+      "grad_norm": 0.6875,
+      "learning_rate": 6.68589833424105e-05,
+      "loss": 0.2856,
+      "step": 3095
+    },
+    {
+      "epoch": 9.717868338557993,
+      "grad_norm": 0.77734375,
+      "learning_rate": 6.651502775347469e-05,
+      "loss": 0.2829,
+      "step": 3100
+    },
+    {
+      "epoch": 9.733542319749215,
+      "grad_norm": 0.71484375,
+      "learning_rate": 6.617151776041692e-05,
+      "loss": 0.2875,
+      "step": 3105
+    },
+    {
+      "epoch": 9.74921630094044,
+      "grad_norm": 0.69140625,
+      "learning_rate": 6.582845793444119e-05,
+      "loss": 0.2746,
+      "step": 3110
+    },
+    {
+      "epoch": 9.764890282131661,
+      "grad_norm": 0.703125,
+      "learning_rate": 6.548585284076084e-05,
+      "loss": 0.288,
+      "step": 3115
+    },
+    {
+      "epoch": 9.780564263322884,
+      "grad_norm": 0.7421875,
+      "learning_rate": 6.514370703853806e-05,
+      "loss": 0.2807,
+      "step": 3120
+    },
+    {
+      "epoch": 9.796238244514107,
+      "grad_norm": 0.7265625,
+      "learning_rate": 6.480202508082298e-05,
+      "loss": 0.2858,
+      "step": 3125
+    },
+    {
+      "epoch": 9.81191222570533,
+      "grad_norm": 0.69921875,
+      "learning_rate": 6.44608115144933e-05,
+      "loss": 0.2793,
+      "step": 3130
+    },
+    {
+      "epoch": 9.827586206896552,
+      "grad_norm": 0.7109375,
+      "learning_rate": 6.412007088019364e-05,
+      "loss": 0.2818,
+      "step": 3135
+    },
+    {
+      "epoch": 9.843260188087774,
+      "grad_norm": 0.75390625,
+      "learning_rate": 6.377980771227509e-05,
+      "loss": 0.2864,
+      "step": 3140
+    },
+    {
+      "epoch": 9.858934169278998,
+      "grad_norm": 0.73046875,
+      "learning_rate": 6.344002653873504e-05,
+      "loss": 0.2819,
+      "step": 3145
+    },
+    {
+      "epoch": 9.87460815047022,
+      "grad_norm": 0.75,
+      "learning_rate": 6.31007318811567e-05,
+      "loss": 0.2833,
+      "step": 3150
+    },
+    {
+      "epoch": 9.890282131661442,
+      "grad_norm": 0.7265625,
+      "learning_rate": 6.276192825464918e-05,
+      "loss": 0.2888,
+      "step": 3155
+    },
+    {
+      "epoch": 9.905956112852664,
+      "grad_norm": 0.66796875,
+      "learning_rate": 6.242362016778713e-05,
+      "loss": 0.2819,
+      "step": 3160
+    },
+    {
+      "epoch": 9.921630094043888,
+      "grad_norm": 0.703125,
+      "learning_rate": 6.208581212255104e-05,
+      "loss": 0.2858,
+      "step": 3165
+    },
+    {
+      "epoch": 9.93730407523511,
+      "grad_norm": 0.6875,
+      "learning_rate": 6.1748508614267e-05,
+      "loss": 0.2793,
+      "step": 3170
+    },
+    {
+      "epoch": 9.952978056426332,
+      "grad_norm": 0.73046875,
+      "learning_rate": 6.141171413154722e-05,
+      "loss": 0.2795,
+      "step": 3175
+    },
+    {
+      "epoch": 9.968652037617554,
+      "grad_norm": 0.6640625,
+      "learning_rate": 6.107543315623001e-05,
+      "loss": 0.2794,
+      "step": 3180
+    },
+    {
+      "epoch": 9.984326018808778,
+      "grad_norm": 0.7578125,
+      "learning_rate": 6.073967016332041e-05,
+      "loss": 0.2833,
+      "step": 3185
+    },
+    {
+      "epoch": 10.0,
+      "grad_norm": 0.6484375,
+      "learning_rate": 6.040442962093029e-05,
+      "loss": 0.2814,
+      "step": 3190
+    },
+    {
+      "epoch": 10.0,
+      "eval_loss": 3.8273370265960693,
+      "eval_runtime": 0.8006,
+      "eval_samples_per_second": 2.498,
+      "eval_steps_per_second": 1.249,
+      "step": 3190
+    },
+    {
+      "epoch": 10.015673981191222,
+      "grad_norm": 0.53515625,
+      "learning_rate": 6.006971599021928e-05,
+      "loss": 0.2526,
+      "step": 3195
+    },
+    {
+      "epoch": 10.031347962382446,
+      "grad_norm": 0.59375,
+      "learning_rate": 5.973553372533509e-05,
+      "loss": 0.2478,
+      "step": 3200
+    },
+    {
+      "epoch": 10.047021943573668,
+      "grad_norm": 0.6484375,
+      "learning_rate": 5.9401887273354475e-05,
+      "loss": 0.2487,
+      "step": 3205
+    },
+    {
+      "epoch": 10.06269592476489,
+      "grad_norm": 0.6640625,
+      "learning_rate": 5.9068781074223824e-05,
+      "loss": 0.2522,
+      "step": 3210
+    },
+    {
+      "epoch": 10.078369905956112,
+      "grad_norm": 0.6015625,
+      "learning_rate": 5.8736219560700324e-05,
+      "loss": 0.2503,
+      "step": 3215
+    },
+    {
+      "epoch": 10.094043887147336,
+      "grad_norm": 0.65625,
+      "learning_rate": 5.840420715829272e-05,
+      "loss": 0.247,
+      "step": 3220
+    },
+    {
+      "epoch": 10.109717868338558,
+      "grad_norm": 0.67578125,
+      "learning_rate": 5.807274828520266e-05,
+      "loss": 0.2521,
+      "step": 3225
+    },
+    {
+      "epoch": 10.12539184952978,
+      "grad_norm": 0.6640625,
+      "learning_rate": 5.774184735226571e-05,
+      "loss": 0.2484,
+      "step": 3230
+    },
+    {
+      "epoch": 10.141065830721002,
+      "grad_norm": 0.59375,
+      "learning_rate": 5.741150876289283e-05,
+      "loss": 0.2474,
+      "step": 3235
+    },
+    {
+      "epoch": 10.156739811912226,
+      "grad_norm": 0.58984375,
+      "learning_rate": 5.708173691301153e-05,
+      "loss": 0.2506,
+      "step": 3240
+    },
+    {
+      "epoch": 10.172413793103448,
+      "grad_norm": 0.61328125,
+      "learning_rate": 5.675253619100772e-05,
+      "loss": 0.2544,
+      "step": 3245
+    },
+    {
+      "epoch": 10.18808777429467,
+      "grad_norm": 0.66015625,
+      "learning_rate": 5.642391097766693e-05,
+      "loss": 0.249,
+      "step": 3250
+    },
+    {
+      "epoch": 10.203761755485893,
+      "grad_norm": 1.2578125,
+      "learning_rate": 5.609586564611631e-05,
+      "loss": 0.2535,
+      "step": 3255
+    },
+    {
+      "epoch": 10.219435736677116,
+      "grad_norm": 0.57421875,
+      "learning_rate": 5.576840456176631e-05,
+      "loss": 0.2504,
+      "step": 3260
+    },
+    {
+      "epoch": 10.235109717868339,
+      "grad_norm": 0.66015625,
+      "learning_rate": 5.544153208225265e-05,
+      "loss": 0.2524,
+      "step": 3265
+    },
+    {
+      "epoch": 10.25078369905956,
+      "grad_norm": 0.65234375,
+      "learning_rate": 5.511525255737815e-05,
+      "loss": 0.2549,
+      "step": 3270
+    },
+    {
+      "epoch": 10.266457680250785,
+      "grad_norm": 0.625,
+      "learning_rate": 5.478957032905514e-05,
+      "loss": 0.2521,
+      "step": 3275
+    },
+    {
+      "epoch": 10.282131661442007,
+      "grad_norm": 0.6171875,
+      "learning_rate": 5.446448973124736e-05,
+      "loss": 0.2531,
+      "step": 3280
+    },
+    {
+      "epoch": 10.297805642633229,
+      "grad_norm": 0.59375,
+      "learning_rate": 5.414001508991264e-05,
+      "loss": 0.2533,
+      "step": 3285
+    },
+    {
+      "epoch": 10.31347962382445,
+      "grad_norm": 0.63671875,
+      "learning_rate": 5.3816150722944916e-05,
+      "loss": 0.2544,
+      "step": 3290
+    },
+    {
+      "epoch": 10.329153605015675,
+      "grad_norm": 0.6328125,
+      "learning_rate": 5.3492900940117264e-05,
+      "loss": 0.2491,
+      "step": 3295
+    },
+    {
+      "epoch": 10.344827586206897,
+      "grad_norm": 0.63671875,
+      "learning_rate": 5.3170270043024015e-05,
+      "loss": 0.2522,
+      "step": 3300
+    },
+    {
+      "epoch": 10.360501567398119,
+      "grad_norm": 0.62890625,
+      "learning_rate": 5.284826232502399e-05,
+      "loss": 0.249,
+      "step": 3305
+    },
+    {
+      "epoch": 10.376175548589341,
+      "grad_norm": 0.77734375,
+      "learning_rate": 5.252688207118297e-05,
+      "loss": 0.2459,
+      "step": 3310
+    },
+    {
+      "epoch": 10.391849529780565,
+      "grad_norm": 0.625,
+      "learning_rate": 5.220613355821704e-05,
+      "loss": 0.2503,
+      "step": 3315
+    },
+    {
+      "epoch": 10.407523510971787,
+      "grad_norm": 0.6328125,
+      "learning_rate": 5.188602105443533e-05,
+      "loss": 0.2543,
+      "step": 3320
+    },
+    {
+      "epoch": 10.423197492163009,
+      "grad_norm": 0.65234375,
+      "learning_rate": 5.156654881968348e-05,
+      "loss": 0.2509,
+      "step": 3325
+    },
+    {
+      "epoch": 10.438871473354231,
+      "grad_norm": 0.65234375,
+      "learning_rate": 5.124772110528684e-05,
+      "loss": 0.2524,
+      "step": 3330
+    },
+    {
+      "epoch": 10.454545454545455,
+      "grad_norm": 0.6796875,
+      "learning_rate": 5.0929542153993926e-05,
+      "loss": 0.2525,
+      "step": 3335
+    },
+    {
+      "epoch": 10.470219435736677,
+      "grad_norm": 0.62890625,
+      "learning_rate": 5.061201619991984e-05,
+      "loss": 0.248,
+      "step": 3340
+    },
+    {
+      "epoch": 10.4858934169279,
+      "grad_norm": 0.72265625,
+      "learning_rate": 5.029514746849018e-05,
+      "loss": 0.2516,
+      "step": 3345
+    },
+    {
+      "epoch": 10.501567398119121,
+      "grad_norm": 0.75,
+      "learning_rate": 4.9978940176384514e-05,
+      "loss": 0.2509,
+      "step": 3350
+    },
+    {
+      "epoch": 10.517241379310345,
+      "grad_norm": 0.625,
+      "learning_rate": 4.966339853148055e-05,
+      "loss": 0.2506,
+      "step": 3355
+    },
+    {
+      "epoch": 10.532915360501567,
+      "grad_norm": 0.6171875,
+      "learning_rate": 4.934852673279787e-05,
+      "loss": 0.254,
+      "step": 3360
+    },
+    {
+      "epoch": 10.54858934169279,
+      "grad_norm": 0.63671875,
+      "learning_rate": 4.9034328970442275e-05,
+      "loss": 0.2519,
+      "step": 3365
+    },
+    {
+      "epoch": 10.564263322884013,
+      "grad_norm": 0.625,
+      "learning_rate": 4.8720809425549916e-05,
+      "loss": 0.2541,
+      "step": 3370
+    },
+    {
+      "epoch": 10.579937304075235,
+      "grad_norm": 0.6484375,
+      "learning_rate": 4.8407972270231704e-05,
+      "loss": 0.2563,
+      "step": 3375
+    },
+    {
+      "epoch": 10.595611285266457,
+      "grad_norm": 0.60546875,
+      "learning_rate": 4.809582166751765e-05,
+      "loss": 0.2529,
+      "step": 3380
+    },
+    {
+      "epoch": 10.61128526645768,
+      "grad_norm": 0.7265625,
+      "learning_rate": 4.778436177130173e-05,
+      "loss": 0.2546,
+      "step": 3385
+    },
+    {
+      "epoch": 10.626959247648903,
+      "grad_norm": 0.6328125,
+      "learning_rate": 4.747359672628631e-05,
+      "loss": 0.2545,
+      "step": 3390
+    },
+    {
+      "epoch": 10.642633228840126,
+      "grad_norm": 0.62890625,
+      "learning_rate": 4.7163530667927226e-05,
+      "loss": 0.2564,
+      "step": 3395
+    },
+    {
+      "epoch": 10.658307210031348,
+      "grad_norm": 0.66015625,
+      "learning_rate": 4.685416772237864e-05,
+      "loss": 0.2506,
+      "step": 3400
+    },
+    {
+      "epoch": 10.67398119122257,
+      "grad_norm": 0.6640625,
+      "learning_rate": 4.65455120064382e-05,
+      "loss": 0.2505,
+      "step": 3405
+    },
+    {
+      "epoch": 10.689655172413794,
+      "grad_norm": 0.68359375,
+      "learning_rate": 4.623756762749207e-05,
+      "loss": 0.2499,
+      "step": 3410
+    },
+    {
+      "epoch": 10.705329153605016,
+      "grad_norm": 0.6484375,
+      "learning_rate": 4.593033868346059e-05,
+      "loss": 0.2529,
+      "step": 3415
+    },
+    {
+      "epoch": 10.721003134796238,
+      "grad_norm": 0.63671875,
+      "learning_rate": 4.5623829262743414e-05,
+      "loss": 0.2549,
+      "step": 3420
+    },
+    {
+      "epoch": 10.736677115987462,
+      "grad_norm": 0.59375,
+      "learning_rate": 4.531804344416536e-05,
+      "loss": 0.2524,
+      "step": 3425
+    },
+    {
+      "epoch": 10.752351097178684,
+      "grad_norm": 0.6015625,
+      "learning_rate": 4.501298529692194e-05,
+      "loss": 0.2556,
+      "step": 3430
+    },
+    {
+      "epoch": 10.768025078369906,
+      "grad_norm": 0.64453125,
+      "learning_rate": 4.470865888052537e-05,
+      "loss": 0.2518,
+      "step": 3435
+    },
+    {
+      "epoch": 10.783699059561128,
+      "grad_norm": 0.640625,
+      "learning_rate": 4.4405068244750446e-05,
+      "loss": 0.2544,
+      "step": 3440
+    },
+    {
+      "epoch": 10.799373040752352,
+      "grad_norm": 0.61328125,
+      "learning_rate": 4.410221742958073e-05,
+      "loss": 0.2493,
+      "step": 3445
+    },
+    {
+      "epoch": 10.815047021943574,
+      "grad_norm": 0.76953125,
+      "learning_rate": 4.380011046515461e-05,
+      "loss": 0.2556,
+      "step": 3450
+    },
+    {
+      "epoch": 10.830721003134796,
+      "grad_norm": 0.6640625,
+      "learning_rate": 4.349875137171196e-05,
+      "loss": 0.2525,
+      "step": 3455
+    },
+    {
+      "epoch": 10.846394984326018,
+      "grad_norm": 0.63671875,
+      "learning_rate": 4.3198144159540346e-05,
+      "loss": 0.2532,
+      "step": 3460
+    },
+    {
+      "epoch": 10.862068965517242,
+      "grad_norm": 0.625,
+      "learning_rate": 4.289829282892188e-05,
+      "loss": 0.2494,
+      "step": 3465
+    },
+    {
+      "epoch": 10.877742946708464,
+      "grad_norm": 0.61328125,
+      "learning_rate": 4.2599201370079875e-05,
+      "loss": 0.2495,
+      "step": 3470
+    },
+    {
+      "epoch": 10.893416927899686,
+      "grad_norm": 0.6484375,
+      "learning_rate": 4.230087376312582e-05,
+      "loss": 0.2483,
+      "step": 3475
+    },
+    {
+      "epoch": 10.909090909090908,
+      "grad_norm": 0.71875,
+      "learning_rate": 4.2003313978006244e-05,
+      "loss": 0.2505,
+      "step": 3480
+    },
+    {
+      "epoch": 10.924764890282132,
+      "grad_norm": 0.609375,
+      "learning_rate": 4.170652597445016e-05,
+      "loss": 0.2499,
+      "step": 3485
+    },
+    {
+      "epoch": 10.940438871473354,
+      "grad_norm": 0.69921875,
+      "learning_rate": 4.1410513701916086e-05,
+      "loss": 0.2501,
+      "step": 3490
+    },
+    {
+      "epoch": 10.956112852664576,
+      "grad_norm": 0.66796875,
+      "learning_rate": 4.111528109953975e-05,
+      "loss": 0.2547,
+      "step": 3495
+    },
+    {
+      "epoch": 10.971786833855798,
+      "grad_norm": 0.60546875,
+      "learning_rate": 4.0820832096081415e-05,
+      "loss": 0.252,
+      "step": 3500
+    },
+    {
+      "epoch": 10.987460815047022,
+      "grad_norm": 0.625,
+      "learning_rate": 4.052717060987386e-05,
+      "loss": 0.2539,
+      "step": 3505
+    },
+    {
+      "epoch": 11.0,
+      "eval_loss": 4.135468006134033,
+      "eval_runtime": 0.805,
+      "eval_samples_per_second": 2.485,
+      "eval_steps_per_second": 1.242,
+      "step": 3509
+    },
+    {
+      "epoch": 11.003134796238244,
+      "grad_norm": 0.5078125,
+      "learning_rate": 4.023430054876999e-05,
+      "loss": 0.2514,
+      "step": 3510
+    },
+    {
+      "epoch": 11.018808777429467,
+      "grad_norm": 0.52734375,
+      "learning_rate": 3.994222581009107e-05,
+      "loss": 0.2418,
+      "step": 3515
+    },
+    {
+      "epoch": 11.03448275862069,
+      "grad_norm": 0.5390625,
+      "learning_rate": 3.965095028057461e-05,
+      "loss": 0.2369,
+      "step": 3520
+    },
+    {
+      "epoch": 11.050156739811912,
+      "grad_norm": 0.57421875,
+      "learning_rate": 3.936047783632286e-05,
+      "loss": 0.2349,
+      "step": 3525
+    },
+    {
+      "epoch": 11.065830721003135,
+      "grad_norm": 0.6015625,
+      "learning_rate": 3.907081234275109e-05,
+      "loss": 0.2392,
+      "step": 3530
+    },
+    {
+      "epoch": 11.081504702194357,
+      "grad_norm": 0.5859375,
+      "learning_rate": 3.878195765453626e-05,
+      "loss": 0.2316,
+      "step": 3535
+    },
+    {
+      "epoch": 11.09717868338558,
+      "grad_norm": 0.53515625,
+      "learning_rate": 3.849391761556559e-05,
+      "loss": 0.2321,
+      "step": 3540
+    },
+    {
+      "epoch": 11.112852664576803,
+      "grad_norm": 0.52734375,
+      "learning_rate": 3.820669605888556e-05,
+      "loss": 0.235,
+      "step": 3545
+    },
+    {
+      "epoch": 11.128526645768025,
+      "grad_norm": 0.55078125,
+      "learning_rate": 3.79202968066508e-05,
+      "loss": 0.2333,
+      "step": 3550
+    },
+    {
+      "epoch": 11.144200626959247,
+      "grad_norm": 0.55078125,
+      "learning_rate": 3.7634723670073294e-05,
+      "loss": 0.2376,
+      "step": 3555
+    },
+    {
+      "epoch": 11.15987460815047,
+      "grad_norm": 0.6171875,
+      "learning_rate": 3.7349980449371516e-05,
+      "loss": 0.2379,
+      "step": 3560
+    },
+    {
+      "epoch": 11.175548589341693,
+      "grad_norm": 0.5625,
+      "learning_rate": 3.706607093372012e-05,
+      "loss": 0.2344,
+      "step": 3565
+    },
+    {
+      "epoch": 11.191222570532915,
+      "grad_norm": 0.55078125,
+      "learning_rate": 3.67829989011992e-05,
+      "loss": 0.2355,
+      "step": 3570
+    },
+    {
+      "epoch": 11.206896551724139,
+      "grad_norm": 0.5703125,
+      "learning_rate": 3.65007681187443e-05,
+      "loss": 0.2332,
+      "step": 3575
+    },
+    {
+      "epoch": 11.22257053291536,
+      "grad_norm": 0.6015625,
+      "learning_rate": 3.621938234209613e-05,
+      "loss": 0.2346,
+      "step": 3580
+    },
+    {
+      "epoch": 11.238244514106583,
+      "grad_norm": 0.6015625,
+      "learning_rate": 3.5938845315750666e-05,
+      "loss": 0.2385,
+      "step": 3585
+    },
+    {
+      "epoch": 11.253918495297805,
+      "grad_norm": 0.5546875,
+      "learning_rate": 3.565916077290914e-05,
+      "loss": 0.2367,
+      "step": 3590
+    },
+    {
+      "epoch": 11.269592476489029,
+      "grad_norm": 0.66796875,
+      "learning_rate": 3.5380332435428655e-05,
+      "loss": 0.2383,
+      "step": 3595
+    },
+    {
+      "epoch": 11.285266457680251,
+      "grad_norm": 0.62890625,
+      "learning_rate": 3.510236401377236e-05,
+      "loss": 0.2373,
+      "step": 3600
+    },
+    {
+      "epoch": 11.300940438871473,
+      "grad_norm": 0.6015625,
+      "learning_rate": 3.482525920696036e-05,
+      "loss": 0.2352,
+      "step": 3605
+    },
+    {
+      "epoch": 11.316614420062695,
+      "grad_norm": 0.5234375,
+      "learning_rate": 3.454902170252019e-05,
+      "loss": 0.2343,
+      "step": 3610
+    },
+    {
+      "epoch": 11.33228840125392,
+      "grad_norm": 0.64453125,
+      "learning_rate": 3.4273655176438014e-05,
+      "loss": 0.2317,
+      "step": 3615
+    },
+    {
+      "epoch": 11.347962382445141,
+      "grad_norm": 0.59765625,
+      "learning_rate": 3.3999163293109534e-05,
+      "loss": 0.2375,
+      "step": 3620
+    },
+    {
+      "epoch": 11.363636363636363,
+      "grad_norm": 0.625,
+      "learning_rate": 3.372554970529137e-05,
+      "loss": 0.2384,
+      "step": 3625
+    },
+    {
+      "epoch": 11.379310344827585,
+      "grad_norm": 0.63671875,
+      "learning_rate": 3.345281805405219e-05,
+      "loss": 0.2385,
+      "step": 3630
+    },
+    {
+      "epoch": 11.39498432601881,
+      "grad_norm": 0.63671875,
+      "learning_rate": 3.318097196872464e-05,
+      "loss": 0.2429,
+      "step": 3635
+    },
+    {
+      "epoch": 11.410658307210031,
+      "grad_norm": 0.58984375,
+      "learning_rate": 3.291001506685666e-05,
+      "loss": 0.2351,
+      "step": 3640
+    },
+    {
+      "epoch": 11.426332288401253,
+      "grad_norm": 0.5546875,
+      "learning_rate": 3.2639950954163644e-05,
+      "loss": 0.2377,
+      "step": 3645
+    },
+    {
+      "epoch": 11.442006269592476,
+      "grad_norm": 0.72265625,
+      "learning_rate": 3.23707832244803e-05,
+      "loss": 0.2353,
+      "step": 3650
+    },
+    {
+      "epoch": 11.4576802507837,
+      "grad_norm": 0.5390625,
+      "learning_rate": 3.2102515459712876e-05,
+      "loss": 0.2361,
+      "step": 3655
+    },
+    {
+      "epoch": 11.473354231974922,
+      "grad_norm": 0.61328125,
+      "learning_rate": 3.1835151229791435e-05,
+      "loss": 0.2357,
+      "step": 3660
+    },
+    {
+      "epoch": 11.489028213166144,
+      "grad_norm": 0.54296875,
+      "learning_rate": 3.1568694092622475e-05,
+      "loss": 0.2359,
+      "step": 3665
+    },
+    {
+      "epoch": 11.504702194357368,
+      "grad_norm": 0.5390625,
+      "learning_rate": 3.1303147594041394e-05,
+      "loss": 0.2333,
+      "step": 3670
+    },
+    {
+      "epoch": 11.52037617554859,
+      "grad_norm": 0.5859375,
+      "learning_rate": 3.1038515267765545e-05,
+      "loss": 0.236,
+      "step": 3675
+    },
+    {
+      "epoch": 11.536050156739812,
+      "grad_norm": 0.64453125,
+      "learning_rate": 3.0774800635346934e-05,
+      "loss": 0.2323,
+      "step": 3680
+    },
+    {
+      "epoch": 11.551724137931034,
+      "grad_norm": 0.578125,
+      "learning_rate": 3.0512007206125638e-05,
+      "loss": 0.2358,
+      "step": 3685
+    },
+    {
+      "epoch": 11.567398119122258,
+      "grad_norm": 0.6328125,
+      "learning_rate": 3.0250138477182886e-05,
+      "loss": 0.2379,
+      "step": 3690
+    },
+    {
+      "epoch": 11.58307210031348,
+      "grad_norm": 0.640625,
+      "learning_rate": 2.9989197933294687e-05,
+      "loss": 0.2353,
+      "step": 3695
+    },
+    {
+      "epoch": 11.598746081504702,
+      "grad_norm": 0.56640625,
+      "learning_rate": 2.9729189046885266e-05,
+      "loss": 0.2359,
+      "step": 3700
+    },
+    {
+      "epoch": 11.614420062695924,
+      "grad_norm": 0.5546875,
+      "learning_rate": 2.947011527798107e-05,
+      "loss": 0.2375,
+      "step": 3705
+    },
+    {
+      "epoch": 11.630094043887148,
+      "grad_norm": 0.609375,
+      "learning_rate": 2.9211980074164514e-05,
+      "loss": 0.2364,
+      "step": 3710
+    },
+    {
+      "epoch": 11.64576802507837,
+      "grad_norm": 0.62109375,
+      "learning_rate": 2.895478687052835e-05,
+      "loss": 0.2393,
+      "step": 3715
+    },
+    {
+      "epoch": 11.661442006269592,
+      "grad_norm": 0.59375,
+      "learning_rate": 2.8698539089629662e-05,
+      "loss": 0.235,
+      "step": 3720
+    },
+    {
+      "epoch": 11.677115987460816,
+      "grad_norm": 0.56640625,
+      "learning_rate": 2.844324014144457e-05,
+      "loss": 0.2341,
+      "step": 3725
+    },
+    {
+      "epoch": 11.692789968652038,
+      "grad_norm": 0.56640625,
+      "learning_rate": 2.818889342332275e-05,
+      "loss": 0.2299,
+      "step": 3730
+    },
+    {
+      "epoch": 11.70846394984326,
+      "grad_norm": 0.61328125,
+      "learning_rate": 2.793550231994222e-05,
+      "loss": 0.2388,
+      "step": 3735
+    },
+    {
+      "epoch": 11.724137931034482,
+      "grad_norm": 0.57421875,
+      "learning_rate": 2.768307020326425e-05,
+      "loss": 0.2341,
+      "step": 3740
+    },
+    {
+      "epoch": 11.739811912225706,
+      "grad_norm": 0.578125,
+      "learning_rate": 2.7431600432488657e-05,
+      "loss": 0.234,
+      "step": 3745
+    },
+    {
+      "epoch": 11.755485893416928,
+      "grad_norm": 0.60546875,
+      "learning_rate": 2.7181096354008884e-05,
+      "loss": 0.2364,
+      "step": 3750
+    },
+    {
+      "epoch": 11.77115987460815,
+      "grad_norm": 0.62890625,
+      "learning_rate": 2.6931561301367646e-05,
+      "loss": 0.2331,
+      "step": 3755
+    },
+    {
+      "epoch": 11.786833855799372,
+      "grad_norm": 0.60546875,
+      "learning_rate": 2.6682998595212505e-05,
+      "loss": 0.2357,
+      "step": 3760
+    },
+    {
+      "epoch": 11.802507836990596,
+      "grad_norm": 0.65625,
+      "learning_rate": 2.6435411543251677e-05,
+      "loss": 0.2389,
+      "step": 3765
+    },
+    {
+      "epoch": 11.818181818181818,
+      "grad_norm": 0.60546875,
+      "learning_rate": 2.6188803440209942e-05,
+      "loss": 0.241,
+      "step": 3770
+    },
+    {
+      "epoch": 11.83385579937304,
+      "grad_norm": 0.62109375,
+      "learning_rate": 2.5943177567785015e-05,
+      "loss": 0.2361,
+      "step": 3775
+    },
+    {
+      "epoch": 11.849529780564263,
+      "grad_norm": 0.578125,
+      "learning_rate": 2.5698537194603566e-05,
+      "loss": 0.2352,
+      "step": 3780
+    },
+    {
+      "epoch": 11.865203761755486,
+      "grad_norm": 0.58203125,
+      "learning_rate": 2.5454885576178067e-05,
+      "loss": 0.2389,
+      "step": 3785
+    },
+    {
+      "epoch": 11.880877742946709,
+      "grad_norm": 0.71875,
+      "learning_rate": 2.5212225954863132e-05,
+      "loss": 0.2367,
+      "step": 3790
+    },
+    {
+      "epoch": 11.89655172413793,
+      "grad_norm": 0.5703125,
+      "learning_rate": 2.4970561559812645e-05,
+      "loss": 0.2383,
+      "step": 3795
+    },
+    {
+      "epoch": 11.912225705329153,
+      "grad_norm": 0.61328125,
+      "learning_rate": 2.472989560693665e-05,
+      "loss": 0.2314,
+      "step": 3800
+    },
+    {
+      "epoch": 11.927899686520377,
+      "grad_norm": 0.66796875,
+      "learning_rate": 2.449023129885859e-05,
+      "loss": 0.2388,
+      "step": 3805
+    },
+    {
+      "epoch": 11.943573667711599,
+      "grad_norm": 0.56640625,
+      "learning_rate": 2.425157182487262e-05,
+      "loss": 0.2383,
+      "step": 3810
+    },
+    {
+      "epoch": 11.95924764890282,
+      "grad_norm": 0.67578125,
+      "learning_rate": 2.401392036090132e-05,
+      "loss": 0.2384,
+      "step": 3815
+    },
+    {
+      "epoch": 11.974921630094045,
+      "grad_norm": 0.62890625,
+      "learning_rate": 2.3777280069453245e-05,
+      "loss": 0.2358,
+      "step": 3820
+    },
+    {
+      "epoch": 11.990595611285267,
+      "grad_norm": 0.55859375,
+      "learning_rate": 2.3541654099581e-05,
+      "loss": 0.233,
+      "step": 3825
+    },
+    {
+      "epoch": 12.0,
+      "eval_loss": 4.354940891265869,
+      "eval_runtime": 0.795,
+      "eval_samples_per_second": 2.516,
+      "eval_steps_per_second": 1.258,
+      "step": 3828
+    },
+    {
+      "epoch": 12.006269592476489,
+      "grad_norm": 0.5546875,
+      "learning_rate": 2.330704558683926e-05,
+      "loss": 0.2349,
+      "step": 3830
+    },
+    {
+      "epoch": 12.021943573667711,
+      "grad_norm": 0.51171875,
+      "learning_rate": 2.307345765324306e-05,
+      "loss": 0.2297,
+      "step": 3835
+    },
+    {
+      "epoch": 12.037617554858935,
+      "grad_norm": 0.5390625,
+      "learning_rate": 2.284089340722618e-05,
+      "loss": 0.2337,
+      "step": 3840
+    },
+    {
+      "epoch": 12.053291536050157,
+      "grad_norm": 0.51953125,
+      "learning_rate": 2.2609355943599942e-05,
+      "loss": 0.2295,
+      "step": 3845
+    },
+    {
+      "epoch": 12.068965517241379,
+      "grad_norm": 0.578125,
+      "learning_rate": 2.2378848343511804e-05,
+      "loss": 0.2287,
+      "step": 3850
+    },
+    {
+      "epoch": 12.084639498432601,
+      "grad_norm": 0.5859375,
+      "learning_rate": 2.214937367440463e-05,
+      "loss": 0.2281,
+      "step": 3855
+    },
+    {
+      "epoch": 12.100313479623825,
+      "grad_norm": 0.56640625,
+      "learning_rate": 2.192093498997555e-05,
+      "loss": 0.2299,
+      "step": 3860
+    },
+    {
+      "epoch": 12.115987460815047,
+      "grad_norm": 0.58984375,
+      "learning_rate": 2.169353533013565e-05,
+      "loss": 0.2287,
+      "step": 3865
+    },
+    {
+      "epoch": 12.13166144200627,
+      "grad_norm": 0.5390625,
+      "learning_rate": 2.1467177720969268e-05,
+      "loss": 0.2281,
+      "step": 3870
+    },
+    {
+      "epoch": 12.147335423197493,
+      "grad_norm": 0.490234375,
+      "learning_rate": 2.12418651746939e-05,
+      "loss": 0.2238,
+      "step": 3875
+    },
+    {
+      "epoch": 12.163009404388715,
+      "grad_norm": 0.56640625,
+      "learning_rate": 2.101760068961992e-05,
+      "loss": 0.2331,
+      "step": 3880
+    },
+    {
+      "epoch": 12.178683385579937,
+      "grad_norm": 0.5234375,
+      "learning_rate": 2.0794387250110913e-05,
+      "loss": 0.2286,
+      "step": 3885
+    },
+    {
+      "epoch": 12.19435736677116,
+      "grad_norm": 0.53125,
+      "learning_rate": 2.0572227826543755e-05,
+      "loss": 0.2323,
+      "step": 3890
+    },
+    {
+      "epoch": 12.210031347962383,
+      "grad_norm": 0.5390625,
+      "learning_rate": 2.0351125375269264e-05,
+      "loss": 0.2269,
+      "step": 3895
+    },
+    {
+      "epoch": 12.225705329153605,
+      "grad_norm": 0.51171875,
+      "learning_rate": 2.0131082838572655e-05,
+      "loss": 0.2288,
+      "step": 3900
+    },
+    {
+      "epoch": 12.241379310344827,
+      "grad_norm": 0.546875,
+      "learning_rate": 1.991210314463461e-05,
+      "loss": 0.236,
+      "step": 3905
+    },
+    {
+      "epoch": 12.25705329153605,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.969418920749214e-05,
+      "loss": 0.229,
+      "step": 3910
+    },
+    {
+      "epoch": 12.272727272727273,
+      "grad_norm": 0.58203125,
+      "learning_rate": 1.9477343926999913e-05,
+      "loss": 0.2305,
+      "step": 3915
+    },
+    {
+      "epoch": 12.288401253918495,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.9261570188791555e-05,
+      "loss": 0.2287,
+      "step": 3920
+    },
+    {
+      "epoch": 12.304075235109718,
+      "grad_norm": 0.55859375,
+      "learning_rate": 1.9046870864241384e-05,
+      "loss": 0.2274,
+      "step": 3925
+    },
+    {
+      "epoch": 12.31974921630094,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.8833248810426073e-05,
+      "loss": 0.2282,
+      "step": 3930
+    },
+    {
+      "epoch": 12.335423197492164,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.8620706870086723e-05,
+      "loss": 0.2282,
+      "step": 3935
+    },
+    {
+      "epoch": 12.351097178683386,
+      "grad_norm": 0.578125,
+      "learning_rate": 1.8409247871591006e-05,
+      "loss": 0.2323,
+      "step": 3940
+    },
+    {
+      "epoch": 12.366771159874608,
+      "grad_norm": 0.57421875,
+      "learning_rate": 1.8198874628895524e-05,
+      "loss": 0.228,
+      "step": 3945
+    },
+    {
+      "epoch": 12.38244514106583,
+      "grad_norm": 0.55078125,
+      "learning_rate": 1.798958994150829e-05,
+      "loss": 0.2227,
+      "step": 3950
+    },
+    {
+      "epoch": 12.398119122257054,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.7781396594451637e-05,
+      "loss": 0.235,
+      "step": 3955
+    },
+    {
+      "epoch": 12.413793103448276,
+      "grad_norm": 0.55078125,
+      "learning_rate": 1.757429735822499e-05,
+      "loss": 0.2303,
+      "step": 3960
+    },
+    {
+      "epoch": 12.429467084639498,
+      "grad_norm": 0.6875,
+      "learning_rate": 1.7368294988768097e-05,
+      "loss": 0.2318,
+      "step": 3965
+    },
+    {
+      "epoch": 12.445141065830722,
+      "grad_norm": 0.59765625,
+      "learning_rate": 1.716339222742436e-05,
+      "loss": 0.2305,
+      "step": 3970
+    },
+    {
+      "epoch": 12.460815047021944,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.695959180090425e-05,
+      "loss": 0.2268,
+      "step": 3975
+    },
+    {
+      "epoch": 12.476489028213166,
+      "grad_norm": 0.65234375,
+      "learning_rate": 1.6756896421249168e-05,
+      "loss": 0.2279,
+      "step": 3980
+    },
+    {
+      "epoch": 12.492163009404388,
+      "grad_norm": 0.57421875,
+      "learning_rate": 1.6555308785795232e-05,
+      "loss": 0.2267,
+      "step": 3985
+    },
+    {
+      "epoch": 12.507836990595612,
+      "grad_norm": 0.59765625,
+      "learning_rate": 1.6354831577137485e-05,
+      "loss": 0.2287,
+      "step": 3990
+    },
+    {
+      "epoch": 12.523510971786834,
+      "grad_norm": 0.58984375,
+      "learning_rate": 1.6155467463094066e-05,
+      "loss": 0.2248,
+      "step": 3995
+    },
+    {
+      "epoch": 12.539184952978056,
+      "grad_norm": 0.5703125,
+      "learning_rate": 1.5957219096670883e-05,
+      "loss": 0.2299,
+      "step": 4000
+    },
+    {
+      "epoch": 12.554858934169278,
+      "grad_norm": 0.59765625,
+      "learning_rate": 1.576008911602609e-05,
+      "loss": 0.2288,
+      "step": 4005
+    },
+    {
+      "epoch": 12.570532915360502,
+      "grad_norm": 0.56640625,
+      "learning_rate": 1.5564080144435212e-05,
+      "loss": 0.2318,
+      "step": 4010
+    },
+    {
+      "epoch": 12.586206896551724,
+      "grad_norm": 0.5625,
+      "learning_rate": 1.536919479025609e-05,
+      "loss": 0.233,
+      "step": 4015
+    },
+    {
+      "epoch": 12.601880877742946,
+      "grad_norm": 0.55078125,
+      "learning_rate": 1.517543564689422e-05,
+      "loss": 0.2253,
+      "step": 4020
+    },
+    {
+      "epoch": 12.61755485893417,
+      "grad_norm": 0.60546875,
+      "learning_rate": 1.4982805292768165e-05,
+      "loss": 0.2266,
+      "step": 4025
+    },
+    {
+      "epoch": 12.633228840125392,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.4791306291275398e-05,
+      "loss": 0.2272,
+      "step": 4030
+    },
+    {
+      "epoch": 12.648902821316614,
+      "grad_norm": 0.55078125,
+      "learning_rate": 1.4600941190758022e-05,
+      "loss": 0.2304,
+      "step": 4035
+    },
+    {
+      "epoch": 12.664576802507836,
+      "grad_norm": 0.56640625,
+      "learning_rate": 1.4411712524469012e-05,
+      "loss": 0.2314,
+      "step": 4040
+    },
+    {
+      "epoch": 12.68025078369906,
+      "grad_norm": 0.59375,
+      "learning_rate": 1.4223622810538328e-05,
+      "loss": 0.2303,
+      "step": 4045
+    },
+    {
+      "epoch": 12.695924764890282,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.4036674551939599e-05,
+      "loss": 0.2323,
+      "step": 4050
+    },
+    {
+      "epoch": 12.711598746081505,
+      "grad_norm": 0.6015625,
+      "learning_rate": 1.385087023645667e-05,
+      "loss": 0.2307,
+      "step": 4055
+    },
+    {
+      "epoch": 12.727272727272727,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.3666212336650586e-05,
+      "loss": 0.2235,
+      "step": 4060
+    },
+    {
+      "epoch": 12.74294670846395,
+      "grad_norm": 0.51953125,
+      "learning_rate": 1.3482703309826584e-05,
+      "loss": 0.2285,
+      "step": 4065
+    },
+    {
+      "epoch": 12.758620689655173,
+      "grad_norm": 0.578125,
+      "learning_rate": 1.330034559800154e-05,
+      "loss": 0.2311,
+      "step": 4070
+    },
+    {
+      "epoch": 12.774294670846395,
+      "grad_norm": 0.578125,
+      "learning_rate": 1.31191416278713e-05,
+      "loss": 0.2286,
+      "step": 4075
+    },
+    {
+      "epoch": 12.789968652037617,
+      "grad_norm": 0.55859375,
+      "learning_rate": 1.293909381077858e-05,
+      "loss": 0.2294,
+      "step": 4080
+    },
+    {
+      "epoch": 12.80564263322884,
+      "grad_norm": 0.58203125,
+      "learning_rate": 1.2760204542680654e-05,
+      "loss": 0.2309,
+      "step": 4085
+    },
+    {
+      "epoch": 12.821316614420063,
+      "grad_norm": 0.57421875,
+      "learning_rate": 1.2582476204117755e-05,
+      "loss": 0.2294,
+      "step": 4090
+    },
+    {
+      "epoch": 12.836990595611285,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.2405911160181072e-05,
+      "loss": 0.2241,
+      "step": 4095
+    },
+    {
+      "epoch": 12.852664576802507,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.2230511760481533e-05,
+      "loss": 0.2253,
+      "step": 4100
+    },
+    {
+      "epoch": 12.86833855799373,
+      "grad_norm": 0.5703125,
+      "learning_rate": 1.2056280339118397e-05,
+      "loss": 0.2358,
+      "step": 4105
+    },
+    {
+      "epoch": 12.884012539184953,
+      "grad_norm": 0.5625,
+      "learning_rate": 1.188321921464829e-05,
+      "loss": 0.229,
+      "step": 4110
+    },
+    {
+      "epoch": 12.899686520376175,
+      "grad_norm": 0.578125,
+      "learning_rate": 1.1711330690054211e-05,
+      "loss": 0.2299,
+      "step": 4115
+    },
+    {
+      "epoch": 12.915360501567399,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.1540617052715074e-05,
+      "loss": 0.2283,
+      "step": 4120
+    },
+    {
+      "epoch": 12.931034482758621,
+      "grad_norm": 0.55859375,
+      "learning_rate": 1.1371080574375114e-05,
+      "loss": 0.2297,
+      "step": 4125
+    },
+    {
+      "epoch": 12.946708463949843,
+      "grad_norm": 0.5859375,
+      "learning_rate": 1.1202723511113766e-05,
+      "loss": 0.2338,
+      "step": 4130
+    },
+    {
+      "epoch": 12.962382445141065,
+      "grad_norm": 0.6171875,
+      "learning_rate": 1.1035548103315484e-05,
+      "loss": 0.2337,
+      "step": 4135
+    },
+    {
+      "epoch": 12.978056426332289,
+      "grad_norm": 0.5703125,
+      "learning_rate": 1.086955657564015e-05,
+      "loss": 0.2294,
+      "step": 4140
+    },
+    {
+      "epoch": 12.993730407523511,
+      "grad_norm": 0.6328125,
+      "learning_rate": 1.0704751136993251e-05,
+      "loss": 0.2281,
+      "step": 4145
+    },
+    {
+      "epoch": 13.0,
+      "eval_loss": 4.455935478210449,
+      "eval_runtime": 0.8034,
+      "eval_samples_per_second": 2.489,
+      "eval_steps_per_second": 1.245,
+      "step": 4147
+    },
+    {
+      "epoch": 13.009404388714733,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.0541133980496686e-05,
+      "loss": 0.2257,
+      "step": 4150
+    },
+    {
+      "epoch": 13.025078369905955,
+      "grad_norm": 0.57421875,
+      "learning_rate": 1.0378707283459376e-05,
+      "loss": 0.2262,
+      "step": 4155
+    },
+    {
+      "epoch": 13.04075235109718,
+      "grad_norm": 0.49609375,
+      "learning_rate": 1.0217473207348483e-05,
+      "loss": 0.2247,
+      "step": 4160
+    },
+    {
+      "epoch": 13.056426332288401,
+      "grad_norm": 0.5859375,
+      "learning_rate": 1.0057433897760493e-05,
+      "loss": 0.2277,
+      "step": 4165
+    },
+    {
+      "epoch": 13.072100313479623,
+      "grad_norm": 0.53515625,
+      "learning_rate": 9.898591484392793e-06,
+      "loss": 0.2256,
+      "step": 4170
+    },
+    {
+      "epoch": 13.087774294670846,
+      "grad_norm": 0.5625,
+      "learning_rate": 9.74094808101519e-06,
+      "loss": 0.23,
+      "step": 4175
+    },
+    {
+      "epoch": 13.10344827586207,
+      "grad_norm": 0.60546875,
+      "learning_rate": 9.584505785441932e-06,
+      "loss": 0.2266,
+      "step": 4180
+    },
+    {
+      "epoch": 13.119122257053291,
+      "grad_norm": 0.55078125,
+      "learning_rate": 9.429266679503657e-06,
+      "loss": 0.2283,
+      "step": 4185
+    },
+    {
+      "epoch": 13.134796238244514,
+      "grad_norm": 0.515625,
+      "learning_rate": 9.275232829019787e-06,
+      "loss": 0.2257,
+      "step": 4190
+    },
+    {
+      "epoch": 13.150470219435737,
+      "grad_norm": 0.51171875,
+      "learning_rate": 9.122406283771002e-06,
+      "loss": 0.2307,
+      "step": 4195
+    },
+    {
+      "epoch": 13.16614420062696,
+      "grad_norm": 0.4921875,
+      "learning_rate": 8.970789077471953e-06,
+      "loss": 0.2259,
+      "step": 4200
+    },
+    {
+      "epoch": 13.181818181818182,
+      "grad_norm": 0.54296875,
+      "learning_rate": 8.82038322774419e-06,
+      "loss": 0.2286,
+      "step": 4205
+    },
+    {
+      "epoch": 13.197492163009404,
+      "grad_norm": 0.5625,
+      "learning_rate": 8.671190736089373e-06,
+      "loss": 0.2277,
+      "step": 4210
+    },
+    {
+      "epoch": 13.213166144200628,
+      "grad_norm": 0.73046875,
+      "learning_rate": 8.523213587862533e-06,
+      "loss": 0.2287,
+      "step": 4215
+    },
+    {
+      "epoch": 13.22884012539185,
+      "grad_norm": 0.55859375,
+      "learning_rate": 8.376453752245795e-06,
+      "loss": 0.2266,
+      "step": 4220
+    },
+    {
+      "epoch": 13.244514106583072,
+      "grad_norm": 0.54296875,
+      "learning_rate": 8.230913182222e-06,
+      "loss": 0.2264,
+      "step": 4225
+    },
+    {
+      "epoch": 13.260188087774294,
+      "grad_norm": 0.5625,
+      "learning_rate": 8.086593814548882e-06,
+      "loss": 0.2258,
+      "step": 4230
+    },
+    {
+      "epoch": 13.275862068965518,
+      "grad_norm": 0.5703125,
+      "learning_rate": 7.943497569733183e-06,
+      "loss": 0.2295,
+      "step": 4235
+    },
+    {
+      "epoch": 13.29153605015674,
+      "grad_norm": 0.578125,
+      "learning_rate": 7.801626352005186e-06,
+      "loss": 0.2254,
+      "step": 4240
+    },
+    {
+      "epoch": 13.307210031347962,
+      "grad_norm": 0.51171875,
+      "learning_rate": 7.66098204929323e-06,
+      "loss": 0.2284,
+      "step": 4245
+    },
+    {
+      "epoch": 13.322884012539184,
+      "grad_norm": 0.578125,
+      "learning_rate": 7.521566533198765e-06,
+      "loss": 0.2263,
+      "step": 4250
+    },
+    {
+      "epoch": 13.338557993730408,
+      "grad_norm": 0.53125,
+      "learning_rate": 7.383381658971311e-06,
+      "loss": 0.2279,
+      "step": 4255
+    },
+    {
+      "epoch": 13.35423197492163,
+      "grad_norm": 0.4921875,
+      "learning_rate": 7.246429265483856e-06,
+      "loss": 0.2238,
+      "step": 4260
+    },
+    {
+      "epoch": 13.369905956112852,
+      "grad_norm": 0.5234375,
+      "learning_rate": 7.1107111752083175e-06,
+      "loss": 0.2273,
+      "step": 4265
+    },
+    {
+      "epoch": 13.385579937304076,
+      "grad_norm": 0.53125,
+      "learning_rate": 6.976229194191352e-06,
+      "loss": 0.2244,
+      "step": 4270
+    },
+    {
+      "epoch": 13.401253918495298,
+      "grad_norm": 0.53515625,
+      "learning_rate": 6.842985112030253e-06,
+      "loss": 0.2256,
+      "step": 4275
+    },
+    {
+      "epoch": 13.41692789968652,
+      "grad_norm": 0.53125,
+      "learning_rate": 6.710980701849223e-06,
+      "loss": 0.2246,
+      "step": 4280
+    },
+    {
+      "epoch": 13.432601880877742,
+      "grad_norm": 0.498046875,
+      "learning_rate": 6.580217720275661e-06,
+      "loss": 0.2275,
+      "step": 4285
+    },
+    {
+      "epoch": 13.448275862068966,
+      "grad_norm": 0.5234375,
+      "learning_rate": 6.450697907416936e-06,
+      "loss": 0.2269,
+      "step": 4290
+    },
+    {
+      "epoch": 13.463949843260188,
+      "grad_norm": 0.5234375,
+      "learning_rate": 6.3224229868370845e-06,
+      "loss": 0.2311,
+      "step": 4295
+    },
+    {
+      "epoch": 13.47962382445141,
+      "grad_norm": 0.5859375,
+      "learning_rate": 6.19539466553396e-06,
+      "loss": 0.2254,
+      "step": 4300
+    },
+    {
+      "epoch": 13.495297805642632,
+      "grad_norm": 0.58984375,
+      "learning_rate": 6.0696146339165095e-06,
+      "loss": 0.2289,
+      "step": 4305
+    },
+    {
+      "epoch": 13.510971786833856,
+      "grad_norm": 0.55078125,
+      "learning_rate": 5.945084565782277e-06,
+      "loss": 0.2319,
+      "step": 4310
+    },
+    {
+      "epoch": 13.526645768025078,
+      "grad_norm": 0.498046875,
+      "learning_rate": 5.82180611829507e-06,
+      "loss": 0.224,
+      "step": 4315
+    },
+    {
+      "epoch": 13.5423197492163,
+      "grad_norm": 0.5703125,
+      "learning_rate": 5.699780931963006e-06,
+      "loss": 0.2297,
+      "step": 4320
+    },
+    {
+      "epoch": 13.557993730407524,
+      "grad_norm": 0.54296875,
+      "learning_rate": 5.5790106306165766e-06,
+      "loss": 0.228,
+      "step": 4325
+    },
+    {
+      "epoch": 13.573667711598747,
+      "grad_norm": 0.6171875,
+      "learning_rate": 5.459496821387166e-06,
+      "loss": 0.2315,
+      "step": 4330
+    },
+    {
+      "epoch": 13.589341692789969,
+      "grad_norm": 0.53515625,
+      "learning_rate": 5.341241094685523e-06,
+      "loss": 0.2283,
+      "step": 4335
+    },
+    {
+      "epoch": 13.60501567398119,
+      "grad_norm": 0.6953125,
+      "learning_rate": 5.2242450241806964e-06,
+      "loss": 0.2277,
+      "step": 4340
+    },
+    {
+      "epoch": 13.620689655172415,
+      "grad_norm": 0.55859375,
+      "learning_rate": 5.108510166779068e-06,
+      "loss": 0.2264,
+      "step": 4345
+    },
+    {
+      "epoch": 13.636363636363637,
+      "grad_norm": 0.51953125,
+      "learning_rate": 4.994038062603645e-06,
+      "loss": 0.232,
+      "step": 4350
+    },
+    {
+      "epoch": 13.652037617554859,
+      "grad_norm": 0.546875,
+      "learning_rate": 4.880830234973499e-06,
+      "loss": 0.2254,
+      "step": 4355
+    },
+    {
+      "epoch": 13.66771159874608,
+      "grad_norm": 0.58984375,
+      "learning_rate": 4.7688881903835915e-06,
+      "loss": 0.2277,
+      "step": 4360
+    },
+    {
+      "epoch": 13.683385579937305,
+      "grad_norm": 0.515625,
+      "learning_rate": 4.658213418484636e-06,
+      "loss": 0.2292,
+      "step": 4365
+    },
+    {
+      "epoch": 13.699059561128527,
+      "grad_norm": 0.5,
+      "learning_rate": 4.548807392063359e-06,
+      "loss": 0.2246,
+      "step": 4370
+    },
+    {
+      "epoch": 13.714733542319749,
+      "grad_norm": 0.51171875,
+      "learning_rate": 4.4406715670228474e-06,
+      "loss": 0.226,
+      "step": 4375
+    },
+    {
+      "epoch": 13.730407523510971,
+      "grad_norm": 0.5546875,
+      "learning_rate": 4.333807382363197e-06,
+      "loss": 0.2292,
+      "step": 4380
+    },
+    {
+      "epoch": 13.746081504702195,
+      "grad_norm": 0.55078125,
+      "learning_rate": 4.22821626016231e-06,
+      "loss": 0.2289,
+      "step": 4385
+    },
+    {
+      "epoch": 13.761755485893417,
+      "grad_norm": 0.5234375,
+      "learning_rate": 4.123899605557091e-06,
+      "loss": 0.2247,
+      "step": 4390
+    },
+    {
+      "epoch": 13.77742946708464,
+      "grad_norm": 0.55078125,
+      "learning_rate": 4.020858806724592e-06,
+      "loss": 0.2221,
+      "step": 4395
+    },
+    {
+      "epoch": 13.793103448275861,
+      "grad_norm": 0.55078125,
+      "learning_rate": 3.91909523486369e-06,
+      "loss": 0.2277,
+      "step": 4400
+    },
+    {
+      "epoch": 13.808777429467085,
+      "grad_norm": 0.55859375,
+      "learning_rate": 3.818610244176702e-06,
+      "loss": 0.2298,
+      "step": 4405
+    },
+    {
+      "epoch": 13.824451410658307,
+      "grad_norm": 0.625,
+      "learning_rate": 3.719405171851487e-06,
+      "loss": 0.2223,
+      "step": 4410
+    },
+    {
+      "epoch": 13.84012539184953,
+      "grad_norm": 0.546875,
+      "learning_rate": 3.621481338043564e-06,
+      "loss": 0.2269,
+      "step": 4415
+    },
+    {
+      "epoch": 13.855799373040753,
+      "grad_norm": 0.57421875,
+      "learning_rate": 3.5248400458586127e-06,
+      "loss": 0.2266,
+      "step": 4420
+    },
+    {
+      "epoch": 13.871473354231975,
+      "grad_norm": 0.546875,
+      "learning_rate": 3.429482581335053e-06,
+      "loss": 0.2296,
+      "step": 4425
+    },
+    {
+      "epoch": 13.887147335423197,
+      "grad_norm": 0.54296875,
+      "learning_rate": 3.3354102134269927e-06,
+      "loss": 0.2281,
+      "step": 4430
+    },
+    {
+      "epoch": 13.90282131661442,
+      "grad_norm": 0.5625,
+      "learning_rate": 3.2426241939873313e-06,
+      "loss": 0.2288,
+      "step": 4435
+    },
+    {
+      "epoch": 13.918495297805643,
+      "grad_norm": 0.62109375,
+      "learning_rate": 3.151125757751083e-06,
+      "loss": 0.2299,
+      "step": 4440
+    },
+    {
+      "epoch": 13.934169278996865,
+      "grad_norm": 0.578125,
+      "learning_rate": 3.0609161223189575e-06,
+      "loss": 0.2238,
+      "step": 4445
+    },
+    {
+      "epoch": 13.949843260188088,
+      "grad_norm": 0.58203125,
+      "learning_rate": 2.9719964881411712e-06,
+      "loss": 0.2243,
+      "step": 4450
+    },
+    {
+      "epoch": 13.96551724137931,
+      "grad_norm": 0.890625,
+      "learning_rate": 2.8843680385014284e-06,
+      "loss": 0.2257,
+      "step": 4455
+    },
+    {
+      "epoch": 13.981191222570533,
+      "grad_norm": 0.609375,
+      "learning_rate": 2.798031939501222e-06,
+      "loss": 0.2286,
+      "step": 4460
+    },
+    {
+      "epoch": 13.996865203761756,
+      "grad_norm": 0.52734375,
+      "learning_rate": 2.7129893400442807e-06,
+      "loss": 0.2274,
+      "step": 4465
+    },
+    {
+      "epoch": 14.0,
+      "eval_loss": 4.4672441482543945,
+      "eval_runtime": 0.8024,
+      "eval_samples_per_second": 2.492,
+      "eval_steps_per_second": 1.246,
+      "step": 4466
+    },
+    {
+      "epoch": 14.012539184952978,
+      "grad_norm": 0.58984375,
+      "learning_rate": 2.629241371821334e-06,
+      "loss": 0.2259,
+      "step": 4470
+    },
+    {
+      "epoch": 14.0282131661442,
+      "grad_norm": 0.51953125,
+      "learning_rate": 2.546789149294959e-06,
+      "loss": 0.2248,
+      "step": 4475
+    },
+    {
+      "epoch": 14.043887147335424,
+      "grad_norm": 0.59765625,
+      "learning_rate": 2.4656337696848496e-06,
+      "loss": 0.226,
+      "step": 4480
+    },
+    {
+      "epoch": 14.059561128526646,
+      "grad_norm": 0.5625,
+      "learning_rate": 2.3857763129531473e-06,
+      "loss": 0.2239,
+      "step": 4485
+    },
+    {
+      "epoch": 14.075235109717868,
+      "grad_norm": 0.51953125,
+      "learning_rate": 2.3072178417901326e-06,
+      "loss": 0.2277,
+      "step": 4490
+    },
+    {
+      "epoch": 14.090909090909092,
+      "grad_norm": 0.57421875,
+      "learning_rate": 2.229959401599968e-06,
+      "loss": 0.2246,
+      "step": 4495
+    },
+    {
+      "epoch": 14.106583072100314,
+      "grad_norm": 0.5546875,
+      "learning_rate": 2.154002020486945e-06,
+      "loss": 0.2244,
+      "step": 4500
+    },
+    {
+      "epoch": 14.122257053291536,
+      "grad_norm": 0.5546875,
+      "learning_rate": 2.0793467092416696e-06,
+      "loss": 0.2226,
+      "step": 4505
+    },
+    {
+      "epoch": 14.137931034482758,
+      "grad_norm": 0.53515625,
+      "learning_rate": 2.005994461327698e-06,
+      "loss": 0.2299,
+      "step": 4510
+    },
+    {
+      "epoch": 14.153605015673982,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.933946252868224e-06,
+      "loss": 0.2287,
+      "step": 4515
+    },
+    {
+      "epoch": 14.169278996865204,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.8632030426332215e-06,
+      "loss": 0.2226,
+      "step": 4520
+    },
+    {
+      "epoch": 14.184952978056426,
+      "grad_norm": 0.59765625,
+      "learning_rate": 1.7937657720265454e-06,
+      "loss": 0.2262,
+      "step": 4525
+    },
+    {
+      "epoch": 14.200626959247648,
+      "grad_norm": 0.5703125,
+      "learning_rate": 1.7256353650735302e-06,
+      "loss": 0.2298,
+      "step": 4530
+    },
+    {
+      "epoch": 14.216300940438872,
+      "grad_norm": 0.5859375,
+      "learning_rate": 1.6588127284085652e-06,
+      "loss": 0.2286,
+      "step": 4535
+    },
+    {
+      "epoch": 14.231974921630094,
+      "grad_norm": 0.65234375,
+      "learning_rate": 1.5932987512631614e-06,
+      "loss": 0.2282,
+      "step": 4540
+    },
+    {
+      "epoch": 14.247648902821316,
+      "grad_norm": 0.58203125,
+      "learning_rate": 1.529094305453993e-06,
+      "loss": 0.222,
+      "step": 4545
+    },
+    {
+      "epoch": 14.263322884012538,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.4662002453714074e-06,
+      "loss": 0.2256,
+      "step": 4550
+    },
+    {
+      "epoch": 14.278996865203762,
+      "grad_norm": 0.59375,
+      "learning_rate": 1.4046174079679787e-06,
+      "loss": 0.2266,
+      "step": 4555
+    },
+    {
+      "epoch": 14.294670846394984,
+      "grad_norm": 0.59375,
+      "learning_rate": 1.3443466127474046e-06,
+      "loss": 0.2245,
+      "step": 4560
+    },
+    {
+      "epoch": 14.310344827586206,
+      "grad_norm": 0.546875,
+      "learning_rate": 1.285388661753595e-06,
+      "loss": 0.2332,
+      "step": 4565
+    },
+    {
+      "epoch": 14.32601880877743,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.2277443395599886e-06,
+      "loss": 0.2266,
+      "step": 4570
+    },
+    {
+      "epoch": 14.341692789968652,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.1714144132591199e-06,
+      "loss": 0.2281,
+      "step": 4575
+    },
+    {
+      "epoch": 14.357366771159874,
+      "grad_norm": 0.53125,
+      "learning_rate": 1.116399632452414e-06,
+      "loss": 0.2292,
+      "step": 4580
+    },
+    {
+      "epoch": 14.373040752351097,
+      "grad_norm": 0.51171875,
+      "learning_rate": 1.062700729240218e-06,
+      "loss": 0.2246,
+      "step": 4585
+    },
+    {
+      "epoch": 14.38871473354232,
+      "grad_norm": 0.57421875,
+      "learning_rate": 1.0103184182120418e-06,
+      "loss": 0.228,
+      "step": 4590
+    },
+    {
+      "epoch": 14.404388714733543,
+      "grad_norm": 0.54296875,
+      "learning_rate": 9.592533964370542e-07,
+      "loss": 0.2257,
+      "step": 4595
+    },
+    {
+      "epoch": 14.420062695924765,
+      "grad_norm": 0.51171875,
+      "learning_rate": 9.095063434548135e-07,
+      "loss": 0.229,
+      "step": 4600
+    },
+    {
+      "epoch": 14.435736677115987,
+      "grad_norm": 0.490234375,
+      "learning_rate": 8.61077921266229e-07,
+      "loss": 0.2273,
+      "step": 4605
+    },
+    {
+      "epoch": 14.45141065830721,
+      "grad_norm": 0.5546875,
+      "learning_rate": 8.139687743247138e-07,
+      "loss": 0.2256,
+      "step": 4610
+    },
+    {
+      "epoch": 14.467084639498433,
+      "grad_norm": 0.640625,
+      "learning_rate": 7.681795295276684e-07,
+      "loss": 0.229,
+      "step": 4615
+    },
+    {
+      "epoch": 14.482758620689655,
+      "grad_norm": 0.5234375,
+      "learning_rate": 7.237107962080991e-07,
+      "loss": 0.2248,
+      "step": 4620
+    },
+    {
+      "epoch": 14.498432601880877,
+      "grad_norm": 0.54296875,
+      "learning_rate": 6.805631661265133e-07,
+      "loss": 0.2292,
+      "step": 4625
+    },
+    {
+      "epoch": 14.5141065830721,
+      "grad_norm": 0.578125,
+      "learning_rate": 6.387372134630587e-07,
+      "loss": 0.2225,
+      "step": 4630
+    },
+    {
+      "epoch": 14.529780564263323,
+      "grad_norm": 0.5078125,
+      "learning_rate": 5.982334948098522e-07,
+      "loss": 0.2291,
+      "step": 4635
+    },
+    {
+      "epoch": 14.545454545454545,
+      "grad_norm": 0.5703125,
+      "learning_rate": 5.5905254916363e-07,
+      "loss": 0.2228,
+      "step": 4640
+    },
+    {
+      "epoch": 14.561128526645769,
+      "grad_norm": 0.5546875,
+      "learning_rate": 5.211948979184978e-07,
+      "loss": 0.2282,
+      "step": 4645
+    },
+    {
+      "epoch": 14.576802507836991,
+      "grad_norm": 0.53515625,
+      "learning_rate": 4.846610448590804e-07,
+      "loss": 0.2291,
+      "step": 4650
+    },
+    {
+      "epoch": 14.592476489028213,
+      "grad_norm": 0.5625,
+      "learning_rate": 4.4945147615372827e-07,
+      "loss": 0.229,
+      "step": 4655
+    },
+    {
+      "epoch": 14.608150470219435,
+      "grad_norm": 0.56640625,
+      "learning_rate": 4.1556666034811007e-07,
+      "loss": 0.2294,
+      "step": 4660
+    },
+    {
+      "epoch": 14.623824451410659,
+      "grad_norm": 0.546875,
+      "learning_rate": 3.8300704835896316e-07,
+      "loss": 0.2268,
+      "step": 4665
+    },
+    {
+      "epoch": 14.639498432601881,
+      "grad_norm": 0.5390625,
+      "learning_rate": 3.517730734680869e-07,
+      "loss": 0.2263,
+      "step": 4670
+    },
+    {
+      "epoch": 14.655172413793103,
+      "grad_norm": 0.58984375,
+      "learning_rate": 3.2186515131655823e-07,
+      "loss": 0.2267,
+      "step": 4675
+    },
+    {
+      "epoch": 14.670846394984325,
+      "grad_norm": 0.5078125,
+      "learning_rate": 2.932836798992589e-07,
+      "loss": 0.2281,
+      "step": 4680
+    },
+    {
+      "epoch": 14.68652037617555,
+      "grad_norm": 0.52734375,
+      "learning_rate": 2.660290395595011e-07,
+      "loss": 0.2286,
+      "step": 4685
+    },
+    {
+      "epoch": 14.702194357366771,
+      "grad_norm": 0.54296875,
+      "learning_rate": 2.401015929840322e-07,
+      "loss": 0.2275,
+      "step": 4690
+    },
+    {
+      "epoch": 14.717868338557993,
+      "grad_norm": 0.55859375,
+      "learning_rate": 2.155016851981717e-07,
+      "loss": 0.229,
+      "step": 4695
+    },
+    {
+      "epoch": 14.733542319749215,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.9222964356123696e-07,
+      "loss": 0.2284,
+      "step": 4700
+    },
+    {
+      "epoch": 14.74921630094044,
+      "grad_norm": 0.56640625,
+      "learning_rate": 1.7028577776216915e-07,
+      "loss": 0.2328,
+      "step": 4705
+    },
+    {
+      "epoch": 14.764890282131661,
+      "grad_norm": 0.59765625,
+      "learning_rate": 1.496703798154364e-07,
+      "loss": 0.2272,
+      "step": 4710
+    },
+    {
+      "epoch": 14.780564263322884,
+      "grad_norm": 0.482421875,
+      "learning_rate": 1.3038372405711487e-07,
+      "loss": 0.2307,
+      "step": 4715
+    },
+    {
+      "epoch": 14.796238244514107,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.1242606714129134e-07,
+      "loss": 0.2253,
+      "step": 4720
+    },
+    {
+      "epoch": 14.81191222570533,
+      "grad_norm": 0.55859375,
+      "learning_rate": 9.579764803658853e-08,
+      "loss": 0.2238,
+      "step": 4725
+    },
+    {
+      "epoch": 14.827586206896552,
+      "grad_norm": 0.52734375,
+      "learning_rate": 8.049868802301187e-08,
+      "loss": 0.2273,
+      "step": 4730
+    },
+    {
+      "epoch": 14.843260188087774,
+      "grad_norm": 0.5,
+      "learning_rate": 6.652939068899633e-08,
+      "loss": 0.2315,
+      "step": 4735
+    },
+    {
+      "epoch": 14.858934169278998,
+      "grad_norm": 0.52734375,
+      "learning_rate": 5.388994192875307e-08,
+      "loss": 0.2232,
+      "step": 4740
+    },
+    {
+      "epoch": 14.87460815047022,
+      "grad_norm": 0.515625,
+      "learning_rate": 4.258050993967144e-08,
+      "loss": 0.2258,
+      "step": 4745
+    },
+    {
+      "epoch": 14.890282131661442,
+      "grad_norm": 0.515625,
+      "learning_rate": 3.260124522023178e-08,
+      "loss": 0.2287,
+      "step": 4750
+    },
+    {
+      "epoch": 14.905956112852664,
+      "grad_norm": 0.484375,
+      "learning_rate": 2.3952280567873796e-08,
+      "loss": 0.2208,
+      "step": 4755
+    },
+    {
+      "epoch": 14.921630094043888,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.6633731077297933e-08,
+      "loss": 0.2233,
+      "step": 4760
+    },
+    {
+      "epoch": 14.93730407523511,
+      "grad_norm": 0.58203125,
+      "learning_rate": 1.0645694138933237e-08,
+      "loss": 0.2239,
+      "step": 4765
+    },
+    {
+      "epoch": 14.952978056426332,
+      "grad_norm": 0.50390625,
+      "learning_rate": 5.988249437627325e-09,
+      "loss": 0.2257,
+      "step": 4770
+    },
+    {
+      "epoch": 14.968652037617554,
+      "grad_norm": 0.609375,
+      "learning_rate": 2.6614589515583377e-09,
+      "loss": 0.2308,
+      "step": 4775
+    },
+    {
+      "epoch": 14.984326018808778,
+      "grad_norm": 0.55859375,
+      "learning_rate": 6.65366951457802e-10,
+      "loss": 0.2262,
+      "step": 4780
+    },
+    {
+      "epoch": 15.0,
+      "grad_norm": 0.5625,
+      "learning_rate": 0.0,
+      "loss": 0.2251,
+      "step": 4785
+    },
+    {
+      "epoch": 15.0,
+      "eval_loss": 4.466753005981445,
+      "eval_runtime": 0.7937,
+      "eval_samples_per_second": 2.52,
+      "eval_steps_per_second": 1.26,
+      "step": 4785
+    },
+    {
+      "epoch": 15.0,
+      "step": 4785,
+      "total_flos": 5.539666200113447e+18,
+      "train_loss": 0.8980661474674348,
+      "train_runtime": 27834.6576,
+      "train_samples_per_second": 4.124,
+      "train_steps_per_second": 0.172
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 4785,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 15,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5.539666200113447e+18,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}