diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,6697 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.989373007438894,
+  "eval_steps": 500,
+  "global_step": 4700,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0021253985122210413,
+      "grad_norm": 512.0,
+      "learning_rate": 4.2553191489361704e-07,
+      "loss": 23.7791,
+      "step": 1
+    },
+    {
+      "epoch": 0.010626992561105207,
+      "grad_norm": 488.0,
+      "learning_rate": 2.1276595744680853e-06,
+      "loss": 23.5159,
+      "step": 5
+    },
+    {
+      "epoch": 0.021253985122210415,
+      "grad_norm": 244.0,
+      "learning_rate": 4.255319148936171e-06,
+      "loss": 20.5987,
+      "step": 10
+    },
+    {
+      "epoch": 0.031880977683315624,
+      "grad_norm": 83.5,
+      "learning_rate": 6.3829787234042555e-06,
+      "loss": 17.7875,
+      "step": 15
+    },
+    {
+      "epoch": 0.04250797024442083,
+      "grad_norm": 45.25,
+      "learning_rate": 8.510638297872341e-06,
+      "loss": 15.0704,
+      "step": 20
+    },
+    {
+      "epoch": 0.053134962805526036,
+      "grad_norm": 19.125,
+      "learning_rate": 1.0638297872340426e-05,
+      "loss": 14.1038,
+      "step": 25
+    },
+    {
+      "epoch": 0.06376195536663125,
+      "grad_norm": 9.5,
+      "learning_rate": 1.2765957446808511e-05,
+      "loss": 13.5893,
+      "step": 30
+    },
+    {
+      "epoch": 0.07438894792773645,
+      "grad_norm": 5.96875,
+      "learning_rate": 1.4893617021276596e-05,
+      "loss": 12.8509,
+      "step": 35
+    },
+    {
+      "epoch": 0.08501594048884166,
+      "grad_norm": 6.34375,
+      "learning_rate": 1.7021276595744682e-05,
+      "loss": 12.4395,
+      "step": 40
+    },
+    {
+      "epoch": 0.09564293304994687,
+      "grad_norm": 8.875,
+      "learning_rate": 1.9148936170212766e-05,
+      "loss": 12.1862,
+      "step": 45
+    },
+    {
+      "epoch": 0.10626992561105207,
+      "grad_norm": 14.5625,
+      "learning_rate": 2.1276595744680852e-05,
+      "loss": 11.1895,
+      "step": 50
+    },
+    {
+      "epoch": 0.11689691817215728,
+      "grad_norm": 29.5,
+      "learning_rate": 2.340425531914894e-05,
+      "loss": 9.9832,
+      "step": 55
+    },
+    {
+      "epoch": 0.1275239107332625,
+      "grad_norm": 46.75,
+      "learning_rate": 2.5531914893617022e-05,
+      "loss": 6.8574,
+      "step": 60
+    },
+    {
+      "epoch": 0.1381509032943677,
+      "grad_norm": 9.8125,
+      "learning_rate": 2.765957446808511e-05,
+      "loss": 2.6513,
+      "step": 65
+    },
+    {
+      "epoch": 0.1487778958554729,
+      "grad_norm": 2.9375,
+      "learning_rate": 2.9787234042553192e-05,
+      "loss": 1.4605,
+      "step": 70
+    },
+    {
+      "epoch": 0.1594048884165781,
+      "grad_norm": 3.453125,
+      "learning_rate": 3.191489361702128e-05,
+      "loss": 1.1919,
+      "step": 75
+    },
+    {
+      "epoch": 0.17003188097768332,
+      "grad_norm": 2.015625,
+      "learning_rate": 3.4042553191489365e-05,
+      "loss": 1.0938,
+      "step": 80
+    },
+    {
+      "epoch": 0.1806588735387885,
+      "grad_norm": 2.265625,
+      "learning_rate": 3.617021276595745e-05,
+      "loss": 0.9763,
+      "step": 85
+    },
+    {
+      "epoch": 0.19128586609989373,
+      "grad_norm": 4.96875,
+      "learning_rate": 3.829787234042553e-05,
+      "loss": 0.9307,
+      "step": 90
+    },
+    {
+      "epoch": 0.20191285866099895,
+      "grad_norm": 1.5390625,
+      "learning_rate": 4.0425531914893614e-05,
+      "loss": 0.866,
+      "step": 95
+    },
+    {
+      "epoch": 0.21253985122210414,
+      "grad_norm": 4.40625,
+      "learning_rate": 4.2553191489361704e-05,
+      "loss": 0.8456,
+      "step": 100
+    },
+    {
+      "epoch": 0.22316684378320936,
+      "grad_norm": 2.6875,
+      "learning_rate": 4.468085106382979e-05,
+      "loss": 0.821,
+      "step": 105
+    },
+    {
+      "epoch": 0.23379383634431455,
+      "grad_norm": 2.75,
+      "learning_rate": 4.680851063829788e-05,
+      "loss": 0.7823,
+      "step": 110
+    },
+    {
+      "epoch": 0.24442082890541977,
+      "grad_norm": 2.953125,
+      "learning_rate": 4.893617021276596e-05,
+      "loss": 0.7831,
+      "step": 115
+    },
+    {
+      "epoch": 0.255047821466525,
+      "grad_norm": 3.90625,
+      "learning_rate": 5.1063829787234044e-05,
+      "loss": 0.7535,
+      "step": 120
+    },
+    {
+      "epoch": 0.26567481402763016,
+      "grad_norm": 4.3125,
+      "learning_rate": 5.319148936170213e-05,
+      "loss": 0.714,
+      "step": 125
+    },
+    {
+      "epoch": 0.2763018065887354,
+      "grad_norm": 7.09375,
+      "learning_rate": 5.531914893617022e-05,
+      "loss": 0.7084,
+      "step": 130
+    },
+    {
+      "epoch": 0.2869287991498406,
+      "grad_norm": 4.15625,
+      "learning_rate": 5.744680851063831e-05,
+      "loss": 0.7074,
+      "step": 135
+    },
+    {
+      "epoch": 0.2975557917109458,
+      "grad_norm": 3.203125,
+      "learning_rate": 5.9574468085106384e-05,
+      "loss": 0.6913,
+      "step": 140
+    },
+    {
+      "epoch": 0.30818278427205104,
+      "grad_norm": 1.9296875,
+      "learning_rate": 6.170212765957447e-05,
+      "loss": 0.6714,
+      "step": 145
+    },
+    {
+      "epoch": 0.3188097768331562,
+      "grad_norm": 29.25,
+      "learning_rate": 6.382978723404256e-05,
+      "loss": 0.6761,
+      "step": 150
+    },
+    {
+      "epoch": 0.3294367693942614,
+      "grad_norm": 2.921875,
+      "learning_rate": 6.595744680851063e-05,
+      "loss": 0.6739,
+      "step": 155
+    },
+    {
+      "epoch": 0.34006376195536664,
+      "grad_norm": 3.921875,
+      "learning_rate": 6.808510638297873e-05,
+      "loss": 0.6572,
+      "step": 160
+    },
+    {
+      "epoch": 0.35069075451647186,
+      "grad_norm": 8.25,
+      "learning_rate": 7.021276595744681e-05,
+      "loss": 0.644,
+      "step": 165
+    },
+    {
+      "epoch": 0.361317747077577,
+      "grad_norm": 11.125,
+      "learning_rate": 7.23404255319149e-05,
+      "loss": 0.6699,
+      "step": 170
+    },
+    {
+      "epoch": 0.37194473963868224,
+      "grad_norm": 2.828125,
+      "learning_rate": 7.446808510638298e-05,
+      "loss": 0.6613,
+      "step": 175
+    },
+    {
+      "epoch": 0.38257173219978746,
+      "grad_norm": 4.84375,
+      "learning_rate": 7.659574468085106e-05,
+      "loss": 0.6376,
+      "step": 180
+    },
+    {
+      "epoch": 0.3931987247608927,
+      "grad_norm": 3.796875,
+      "learning_rate": 7.872340425531916e-05,
+      "loss": 0.6444,
+      "step": 185
+    },
+    {
+      "epoch": 0.4038257173219979,
+      "grad_norm": 2.078125,
+      "learning_rate": 8.085106382978723e-05,
+      "loss": 0.6349,
+      "step": 190
+    },
+    {
+      "epoch": 0.41445270988310307,
+      "grad_norm": 3.53125,
+      "learning_rate": 8.297872340425533e-05,
+      "loss": 0.644,
+      "step": 195
+    },
+    {
+      "epoch": 0.4250797024442083,
+      "grad_norm": 1.9921875,
+      "learning_rate": 8.510638297872341e-05,
+      "loss": 0.6224,
+      "step": 200
+    },
+    {
+      "epoch": 0.4357066950053135,
+      "grad_norm": 3.1875,
+      "learning_rate": 8.723404255319149e-05,
+      "loss": 0.6213,
+      "step": 205
+    },
+    {
+      "epoch": 0.4463336875664187,
+      "grad_norm": 6.375,
+      "learning_rate": 8.936170212765958e-05,
+      "loss": 0.6095,
+      "step": 210
+    },
+    {
+      "epoch": 0.4569606801275239,
+      "grad_norm": 4.28125,
+      "learning_rate": 9.148936170212766e-05,
+      "loss": 0.6226,
+      "step": 215
+    },
+    {
+      "epoch": 0.4675876726886291,
+      "grad_norm": 2.671875,
+      "learning_rate": 9.361702127659576e-05,
+      "loss": 0.6042,
+      "step": 220
+    },
+    {
+      "epoch": 0.4782146652497343,
+      "grad_norm": 0.88671875,
+      "learning_rate": 9.574468085106384e-05,
+      "loss": 0.6009,
+      "step": 225
+    },
+    {
+      "epoch": 0.48884165781083955,
+      "grad_norm": 1.2265625,
+      "learning_rate": 9.787234042553192e-05,
+      "loss": 0.5855,
+      "step": 230
+    },
+    {
+      "epoch": 0.49946865037194477,
+      "grad_norm": 1.8671875,
+      "learning_rate": 0.0001,
+      "loss": 0.5729,
+      "step": 235
+    },
+    {
+      "epoch": 0.51009564293305,
+      "grad_norm": 2.171875,
+      "learning_rate": 0.00010212765957446809,
+      "loss": 0.5751,
+      "step": 240
+    },
+    {
+      "epoch": 0.5207226354941552,
+      "grad_norm": 13.5625,
+      "learning_rate": 0.00010425531914893618,
+      "loss": 0.5832,
+      "step": 245
+    },
+    {
+      "epoch": 0.5313496280552603,
+      "grad_norm": 1.0859375,
+      "learning_rate": 0.00010638297872340425,
+      "loss": 0.5858,
+      "step": 250
+    },
+    {
+      "epoch": 0.5419766206163655,
+      "grad_norm": 14.5,
+      "learning_rate": 0.00010851063829787234,
+      "loss": 0.5908,
+      "step": 255
+    },
+    {
+      "epoch": 0.5526036131774708,
+      "grad_norm": 1.5546875,
+      "learning_rate": 0.00011063829787234043,
+      "loss": 0.5924,
+      "step": 260
+    },
+    {
+      "epoch": 0.563230605738576,
+      "grad_norm": 1.7578125,
+      "learning_rate": 0.00011276595744680852,
+      "loss": 0.5704,
+      "step": 265
+    },
+    {
+      "epoch": 0.5738575982996812,
+      "grad_norm": 0.7734375,
+      "learning_rate": 0.00011489361702127661,
+      "loss": 0.5695,
+      "step": 270
+    },
+    {
+      "epoch": 0.5844845908607864,
+      "grad_norm": 2.25,
+      "learning_rate": 0.00011702127659574468,
+      "loss": 0.5654,
+      "step": 275
+    },
+    {
+      "epoch": 0.5951115834218916,
+      "grad_norm": 2.703125,
+      "learning_rate": 0.00011914893617021277,
+      "loss": 0.5816,
+      "step": 280
+    },
+    {
+      "epoch": 0.6057385759829969,
+      "grad_norm": 0.95703125,
+      "learning_rate": 0.00012127659574468086,
+      "loss": 0.5627,
+      "step": 285
+    },
+    {
+      "epoch": 0.6163655685441021,
+      "grad_norm": 2.703125,
+      "learning_rate": 0.00012340425531914893,
+      "loss": 0.5557,
+      "step": 290
+    },
+    {
+      "epoch": 0.6269925611052072,
+      "grad_norm": 1.8046875,
+      "learning_rate": 0.00012553191489361702,
+      "loss": 0.5571,
+      "step": 295
+    },
+    {
+      "epoch": 0.6376195536663124,
+      "grad_norm": 4.46875,
+      "learning_rate": 0.00012765957446808513,
+      "loss": 0.544,
+      "step": 300
+    },
+    {
+      "epoch": 0.6482465462274176,
+      "grad_norm": 3.421875,
+      "learning_rate": 0.00012978723404255318,
+      "loss": 0.5593,
+      "step": 305
+    },
+    {
+      "epoch": 0.6588735387885228,
+      "grad_norm": 3.5625,
+      "learning_rate": 0.00013191489361702127,
+      "loss": 0.5513,
+      "step": 310
+    },
+    {
+      "epoch": 0.6695005313496281,
+      "grad_norm": 9.0,
+      "learning_rate": 0.00013404255319148938,
+      "loss": 0.5681,
+      "step": 315
+    },
+    {
+      "epoch": 0.6801275239107333,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.00013617021276595746,
+      "loss": 0.554,
+      "step": 320
+    },
+    {
+      "epoch": 0.6907545164718385,
+      "grad_norm": 1.484375,
+      "learning_rate": 0.00013829787234042554,
+      "loss": 0.5383,
+      "step": 325
+    },
+    {
+      "epoch": 0.7013815090329437,
+      "grad_norm": 1.5234375,
+      "learning_rate": 0.00014042553191489363,
+      "loss": 0.5381,
+      "step": 330
+    },
+    {
+      "epoch": 0.7120085015940489,
+      "grad_norm": 1.125,
+      "learning_rate": 0.0001425531914893617,
+      "loss": 0.5544,
+      "step": 335
+    },
+    {
+      "epoch": 0.722635494155154,
+      "grad_norm": 0.96484375,
+      "learning_rate": 0.0001446808510638298,
+      "loss": 0.539,
+      "step": 340
+    },
+    {
+      "epoch": 0.7332624867162593,
+      "grad_norm": 2.890625,
+      "learning_rate": 0.00014680851063829788,
+      "loss": 0.542,
+      "step": 345
+    },
+    {
+      "epoch": 0.7438894792773645,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.00014893617021276596,
+      "loss": 0.531,
+      "step": 350
+    },
+    {
+      "epoch": 0.7545164718384697,
+      "grad_norm": 3.015625,
+      "learning_rate": 0.00015106382978723407,
+      "loss": 0.5358,
+      "step": 355
+    },
+    {
+      "epoch": 0.7651434643995749,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.00015319148936170213,
+      "loss": 0.5366,
+      "step": 360
+    },
+    {
+      "epoch": 0.7757704569606801,
+      "grad_norm": 2.71875,
+      "learning_rate": 0.0001553191489361702,
+      "loss": 0.5359,
+      "step": 365
+    },
+    {
+      "epoch": 0.7863974495217854,
+      "grad_norm": 1.8984375,
+      "learning_rate": 0.00015744680851063832,
+      "loss": 0.5506,
+      "step": 370
+    },
+    {
+      "epoch": 0.7970244420828906,
+      "grad_norm": 2.484375,
+      "learning_rate": 0.00015957446808510637,
+      "loss": 0.5286,
+      "step": 375
+    },
+    {
+      "epoch": 0.8076514346439958,
+      "grad_norm": 0.72265625,
+      "learning_rate": 0.00016170212765957446,
+      "loss": 0.5319,
+      "step": 380
+    },
+    {
+      "epoch": 0.8182784272051009,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.00016382978723404257,
+      "loss": 0.5312,
+      "step": 385
+    },
+    {
+      "epoch": 0.8289054197662061,
+      "grad_norm": 1.234375,
+      "learning_rate": 0.00016595744680851065,
+      "loss": 0.5284,
+      "step": 390
+    },
+    {
+      "epoch": 0.8395324123273114,
+      "grad_norm": 2.859375,
+      "learning_rate": 0.00016808510638297873,
+      "loss": 0.5448,
+      "step": 395
+    },
+    {
+      "epoch": 0.8501594048884166,
+      "grad_norm": 1.90625,
+      "learning_rate": 0.00017021276595744682,
+      "loss": 0.5292,
+      "step": 400
+    },
+    {
+      "epoch": 0.8607863974495218,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.0001723404255319149,
+      "loss": 0.5246,
+      "step": 405
+    },
+    {
+      "epoch": 0.871413390010627,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.00017446808510638298,
+      "loss": 0.5353,
+      "step": 410
+    },
+    {
+      "epoch": 0.8820403825717322,
+      "grad_norm": 21.0,
+      "learning_rate": 0.00017659574468085107,
+      "loss": 0.5399,
+      "step": 415
+    },
+    {
+      "epoch": 0.8926673751328374,
+      "grad_norm": 1.296875,
+      "learning_rate": 0.00017872340425531915,
+      "loss": 0.5405,
+      "step": 420
+    },
+    {
+      "epoch": 0.9032943676939427,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.00018085106382978726,
+      "loss": 0.5233,
+      "step": 425
+    },
+    {
+      "epoch": 0.9139213602550478,
+      "grad_norm": 2.21875,
+      "learning_rate": 0.00018297872340425532,
+      "loss": 0.5225,
+      "step": 430
+    },
+    {
+      "epoch": 0.924548352816153,
+      "grad_norm": 0.6328125,
+      "learning_rate": 0.0001851063829787234,
+      "loss": 0.532,
+      "step": 435
+    },
+    {
+      "epoch": 0.9351753453772582,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.0001872340425531915,
+      "loss": 0.5353,
+      "step": 440
+    },
+    {
+      "epoch": 0.9458023379383634,
+      "grad_norm": 1.8671875,
+      "learning_rate": 0.00018936170212765957,
+      "loss": 0.5365,
+      "step": 445
+    },
+    {
+      "epoch": 0.9564293304994687,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.00019148936170212768,
+      "loss": 0.5395,
+      "step": 450
+    },
+    {
+      "epoch": 0.9670563230605739,
+      "grad_norm": 2.421875,
+      "learning_rate": 0.00019361702127659576,
+      "loss": 0.5309,
+      "step": 455
+    },
+    {
+      "epoch": 0.9776833156216791,
+      "grad_norm": 4.96875,
+      "learning_rate": 0.00019574468085106384,
+      "loss": 0.5281,
+      "step": 460
+    },
+    {
+      "epoch": 0.9883103081827843,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.00019787234042553193,
+      "loss": 0.5182,
+      "step": 465
+    },
+    {
+      "epoch": 0.9989373007438895,
+      "grad_norm": 1.7109375,
+      "learning_rate": 0.0002,
+      "loss": 0.5262,
+      "step": 470
+    },
+    {
+      "epoch": 0.9989373007438895,
+      "eval_loss": 1.3223822116851807,
+      "eval_runtime": 0.8107,
+      "eval_samples_per_second": 6.167,
+      "eval_steps_per_second": 1.233,
+      "step": 470
+    },
+    {
+      "epoch": 1.0095642933049946,
+      "grad_norm": 1.484375,
+      "learning_rate": 0.00019999931050912237,
+      "loss": 0.5078,
+      "step": 475
+    },
+    {
+      "epoch": 1.0201912858661,
+      "grad_norm": 1.125,
+      "learning_rate": 0.00019999724204599747,
+      "loss": 0.4809,
+      "step": 480
+    },
+    {
+      "epoch": 1.030818278427205,
+      "grad_norm": 2.703125,
+      "learning_rate": 0.00019999379463914898,
+      "loss": 0.4983,
+      "step": 485
+    },
+    {
+      "epoch": 1.0414452709883104,
+      "grad_norm": 1.90625,
+      "learning_rate": 0.00019998896833611603,
+      "loss": 0.4957,
+      "step": 490
+    },
+    {
+      "epoch": 1.0520722635494155,
+      "grad_norm": 0.6640625,
+      "learning_rate": 0.00019998276320345247,
+      "loss": 0.4901,
+      "step": 495
+    },
+    {
+      "epoch": 1.0626992561105206,
+      "grad_norm": 2.171875,
+      "learning_rate": 0.0001999751793267259,
+      "loss": 0.4931,
+      "step": 500
+    },
+    {
+      "epoch": 1.073326248671626,
+      "grad_norm": 1.40625,
+      "learning_rate": 0.00019996621681051669,
+      "loss": 0.4916,
+      "step": 505
+    },
+    {
+      "epoch": 1.083953241232731,
+      "grad_norm": 2.75,
+      "learning_rate": 0.0001999558757784162,
+      "loss": 0.501,
+      "step": 510
+    },
+    {
+      "epoch": 1.0945802337938364,
+      "grad_norm": 1.390625,
+      "learning_rate": 0.00019994415637302547,
+      "loss": 0.5011,
+      "step": 515
+    },
+    {
+      "epoch": 1.1052072263549415,
+      "grad_norm": 2.625,
+      "learning_rate": 0.0001999310587559529,
+      "loss": 0.4966,
+      "step": 520
+    },
+    {
+      "epoch": 1.1158342189160468,
+      "grad_norm": 2.703125,
+      "learning_rate": 0.00019991658310781224,
+      "loss": 0.5037,
+      "step": 525
+    },
+    {
+      "epoch": 1.126461211477152,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.00019990072962822007,
+      "loss": 0.5134,
+      "step": 530
+    },
+    {
+      "epoch": 1.1370882040382573,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.00019988349853579295,
+      "loss": 0.4997,
+      "step": 535
+    },
+    {
+      "epoch": 1.1477151965993624,
+      "grad_norm": 1.0390625,
+      "learning_rate": 0.00019986489006814452,
+      "loss": 0.4965,
+      "step": 540
+    },
+    {
+      "epoch": 1.1583421891604675,
+      "grad_norm": 7.1875,
+      "learning_rate": 0.00019984490448188218,
+      "loss": 0.4843,
+      "step": 545
+    },
+    {
+      "epoch": 1.1689691817215728,
+      "grad_norm": 2.546875,
+      "learning_rate": 0.00019982354205260347,
+      "loss": 0.5033,
+      "step": 550
+    },
+    {
+      "epoch": 1.179596174282678,
+      "grad_norm": 0.443359375,
+      "learning_rate": 0.0001998008030748924,
+      "loss": 0.4976,
+      "step": 555
+    },
+    {
+      "epoch": 1.1902231668437833,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.00019977668786231534,
+      "loss": 0.497,
+      "step": 560
+    },
+    {
+      "epoch": 1.2008501594048884,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.00019975119674741664,
+      "loss": 0.4973,
+      "step": 565
+    },
+    {
+      "epoch": 1.2114771519659937,
+      "grad_norm": 1.171875,
+      "learning_rate": 0.00019972433008171416,
+      "loss": 0.5,
+      "step": 570
+    },
+    {
+      "epoch": 1.2221041445270988,
+      "grad_norm": 1.609375,
+      "learning_rate": 0.00019969608823569433,
+      "loss": 0.4925,
+      "step": 575
+    },
+    {
+      "epoch": 1.2327311370882041,
+      "grad_norm": 1.6875,
+      "learning_rate": 0.00019966647159880703,
+      "loss": 0.4848,
+      "step": 580
+    },
+    {
+      "epoch": 1.2433581296493093,
+      "grad_norm": 2.734375,
+      "learning_rate": 0.00019963548057946024,
+      "loss": 0.5033,
+      "step": 585
+    },
+    {
+      "epoch": 1.2539851222104144,
+      "grad_norm": 0.5390625,
+      "learning_rate": 0.00019960311560501454,
+      "loss": 0.4925,
+      "step": 590
+    },
+    {
+      "epoch": 1.2646121147715197,
+      "grad_norm": 1.3125,
+      "learning_rate": 0.000199569377121777,
+      "loss": 0.487,
+      "step": 595
+    },
+    {
+      "epoch": 1.2752391073326248,
+      "grad_norm": 0.44921875,
+      "learning_rate": 0.0001995342655949951,
+      "loss": 0.497,
+      "step": 600
+    },
+    {
+      "epoch": 1.2858660998937301,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.00019949778150885042,
+      "loss": 0.4877,
+      "step": 605
+    },
+    {
+      "epoch": 1.2964930924548352,
+      "grad_norm": 0.56640625,
+      "learning_rate": 0.00019945992536645187,
+      "loss": 0.4942,
+      "step": 610
+    },
+    {
+      "epoch": 1.3071200850159406,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.00019942069768982872,
+      "loss": 0.4881,
+      "step": 615
+    },
+    {
+      "epoch": 1.3177470775770457,
+      "grad_norm": 0.5390625,
+      "learning_rate": 0.0001993800990199235,
+      "loss": 0.4959,
+      "step": 620
+    },
+    {
+      "epoch": 1.328374070138151,
+      "grad_norm": 0.61328125,
+      "learning_rate": 0.0001993381299165844,
+      "loss": 0.4883,
+      "step": 625
+    },
+    {
+      "epoch": 1.3390010626992561,
+      "grad_norm": 0.546875,
+      "learning_rate": 0.0001992947909585578,
+      "loss": 0.4861,
+      "step": 630
+    },
+    {
+      "epoch": 1.3496280552603612,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.00019925008274347995,
+      "loss": 0.4868,
+      "step": 635
+    },
+    {
+      "epoch": 1.3602550478214666,
+      "grad_norm": 0.66796875,
+      "learning_rate": 0.000199204005887869,
+      "loss": 0.4979,
+      "step": 640
+    },
+    {
+      "epoch": 1.3708820403825717,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.00019915656102711634,
+      "loss": 0.4885,
+      "step": 645
+    },
+    {
+      "epoch": 1.381509032943677,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.000199107748815478,
+      "loss": 0.4928,
+      "step": 650
+    },
+    {
+      "epoch": 1.392136025504782,
+      "grad_norm": 1.46875,
+      "learning_rate": 0.00019905756992606548,
+      "loss": 0.4879,
+      "step": 655
+    },
+    {
+      "epoch": 1.4027630180658874,
+      "grad_norm": 0.70703125,
+      "learning_rate": 0.00019900602505083648,
+      "loss": 0.4854,
+      "step": 660
+    },
+    {
+      "epoch": 1.4133900106269925,
+      "grad_norm": 5.0,
+      "learning_rate": 0.00019895311490058542,
+      "loss": 0.4932,
+      "step": 665
+    },
+    {
+      "epoch": 1.4240170031880979,
+      "grad_norm": 0.5390625,
+      "learning_rate": 0.0001988988402049336,
+      "loss": 0.4911,
+      "step": 670
+    },
+    {
+      "epoch": 1.434643995749203,
+      "grad_norm": 0.490234375,
+      "learning_rate": 0.00019884320171231925,
+      "loss": 0.4967,
+      "step": 675
+    },
+    {
+      "epoch": 1.445270988310308,
+      "grad_norm": 0.859375,
+      "learning_rate": 0.00019878620018998696,
+      "loss": 0.4893,
+      "step": 680
+    },
+    {
+      "epoch": 1.4558979808714134,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.00019872783642397733,
+      "loss": 0.4849,
+      "step": 685
+    },
+    {
+      "epoch": 1.4665249734325185,
+      "grad_norm": 0.431640625,
+      "learning_rate": 0.00019866811121911607,
+      "loss": 0.485,
+      "step": 690
+    },
+    {
+      "epoch": 1.4771519659936239,
+      "grad_norm": 0.984375,
+      "learning_rate": 0.00019860702539900287,
+      "loss": 0.4993,
+      "step": 695
+    },
+    {
+      "epoch": 1.487778958554729,
+      "grad_norm": 3.546875,
+      "learning_rate": 0.000198544579806,
+      "loss": 0.4858,
+      "step": 700
+    },
+    {
+      "epoch": 1.4984059511158343,
+      "grad_norm": 0.8359375,
+      "learning_rate": 0.00019848077530122083,
+      "loss": 0.4781,
+      "step": 705
+    },
+    {
+      "epoch": 1.5090329436769394,
+      "grad_norm": 1.109375,
+      "learning_rate": 0.0001984156127645178,
+      "loss": 0.4771,
+      "step": 710
+    },
+    {
+      "epoch": 1.5196599362380447,
+      "grad_norm": 1.734375,
+      "learning_rate": 0.00019834909309447045,
+      "loss": 0.4913,
+      "step": 715
+    },
+    {
+      "epoch": 1.5302869287991498,
+      "grad_norm": 0.62890625,
+      "learning_rate": 0.00019828121720837286,
+      "loss": 0.4957,
+      "step": 720
+    },
+    {
+      "epoch": 1.540913921360255,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.00019821198604222113,
+      "loss": 0.4829,
+      "step": 725
+    },
+    {
+      "epoch": 1.5515409139213603,
+      "grad_norm": 1.9140625,
+      "learning_rate": 0.00019814140055070042,
+      "loss": 0.4901,
+      "step": 730
+    },
+    {
+      "epoch": 1.5621679064824656,
+      "grad_norm": 0.6640625,
+      "learning_rate": 0.00019806946170717175,
+      "loss": 0.4813,
+      "step": 735
+    },
+    {
+      "epoch": 1.5727948990435707,
+      "grad_norm": 0.408203125,
+      "learning_rate": 0.0001979961705036587,
+      "loss": 0.472,
+      "step": 740
+    },
+    {
+      "epoch": 1.5834218916046758,
+      "grad_norm": 0.484375,
+      "learning_rate": 0.00019792152795083351,
+      "loss": 0.4895,
+      "step": 745
+    },
+    {
+      "epoch": 1.594048884165781,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.00019784553507800349,
+      "loss": 0.4911,
+      "step": 750
+    },
+    {
+      "epoch": 1.6046758767268863,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.00019776819293309633,
+      "loss": 0.4801,
+      "step": 755
+    },
+    {
+      "epoch": 1.6153028692879916,
+      "grad_norm": 0.98046875,
+      "learning_rate": 0.00019768950258264623,
+      "loss": 0.488,
+      "step": 760
+    },
+    {
+      "epoch": 1.6259298618490967,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.00019760946511177872,
+      "loss": 0.4795,
+      "step": 765
+    },
+    {
+      "epoch": 1.6365568544102018,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.0001975280816241959,
+      "loss": 0.4797,
+      "step": 770
+    },
+    {
+      "epoch": 1.6471838469713072,
+      "grad_norm": 5.125,
+      "learning_rate": 0.00019744535324216127,
+      "loss": 0.4913,
+      "step": 775
+    },
+    {
+      "epoch": 1.6578108395324125,
+      "grad_norm": 0.7109375,
+      "learning_rate": 0.00019736128110648407,
+      "loss": 0.4751,
+      "step": 780
+    },
+    {
+      "epoch": 1.6684378320935176,
+      "grad_norm": 6.5,
+      "learning_rate": 0.00019727586637650373,
+      "loss": 0.4956,
+      "step": 785
+    },
+    {
+      "epoch": 1.6790648246546227,
+      "grad_norm": 1.1640625,
+      "learning_rate": 0.0001971891102300738,
+      "loss": 0.4918,
+      "step": 790
+    },
+    {
+      "epoch": 1.6896918172157278,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.0001971010138635457,
+      "loss": 0.4869,
+      "step": 795
+    },
+    {
+      "epoch": 1.7003188097768331,
+      "grad_norm": 0.451171875,
+      "learning_rate": 0.00019701157849175228,
+      "loss": 0.4823,
+      "step": 800
+    },
+    {
+      "epoch": 1.7109458023379385,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.00019692080534799096,
+      "loss": 0.4922,
+      "step": 805
+    },
+    {
+      "epoch": 1.7215727948990436,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.00019682869568400684,
+      "loss": 0.4918,
+      "step": 810
+    },
+    {
+      "epoch": 1.7321997874601487,
+      "grad_norm": 0.5,
+      "learning_rate": 0.0001967352507699754,
+      "loss": 0.4834,
+      "step": 815
+    },
+    {
+      "epoch": 1.742826780021254,
+      "grad_norm": 6.21875,
+      "learning_rate": 0.00019664047189448493,
+      "loss": 0.4807,
+      "step": 820
+    },
+    {
+      "epoch": 1.7534537725823593,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.0001965443603645189,
+      "loss": 0.486,
+      "step": 825
+    },
+    {
+      "epoch": 1.7640807651434645,
+      "grad_norm": 1.3046875,
+      "learning_rate": 0.00019644691750543767,
+      "loss": 0.5035,
+      "step": 830
+    },
+    {
+      "epoch": 1.7747077577045696,
+      "grad_norm": 1.40625,
+      "learning_rate": 0.00019634814466096056,
+      "loss": 0.4976,
+      "step": 835
+    },
+    {
+      "epoch": 1.7853347502656747,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.00019624804319314705,
+      "loss": 0.5016,
+      "step": 840
+    },
+    {
+      "epoch": 1.79596174282678,
+      "grad_norm": 1.125,
+      "learning_rate": 0.0001961466144823781,
+      "loss": 0.4952,
+      "step": 845
+    },
+    {
+      "epoch": 1.8065887353878853,
+      "grad_norm": 0.5546875,
+      "learning_rate": 0.00019604385992733715,
+      "loss": 0.4849,
+      "step": 850
+    },
+    {
+      "epoch": 1.8172157279489904,
+      "grad_norm": 0.61328125,
+      "learning_rate": 0.00019593978094499076,
+      "loss": 0.4792,
+      "step": 855
+    },
+    {
+      "epoch": 1.8278427205100956,
+      "grad_norm": 0.494140625,
+      "learning_rate": 0.00019583437897056915,
+      "loss": 0.4858,
+      "step": 860
+    },
+    {
+      "epoch": 1.8384697130712009,
+      "grad_norm": 0.85546875,
+      "learning_rate": 0.00019572765545754626,
+      "loss": 0.4905,
+      "step": 865
+    },
+    {
+      "epoch": 1.8490967056323062,
+      "grad_norm": 1.3359375,
+      "learning_rate": 0.00019561961187761985,
+      "loss": 0.4786,
+      "step": 870
+    },
+    {
+      "epoch": 1.8597236981934113,
+      "grad_norm": 0.5078125,
+      "learning_rate": 0.00019551024972069126,
+      "loss": 0.4694,
+      "step": 875
+    },
+    {
+      "epoch": 1.8703506907545164,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.00019539957049484458,
+      "loss": 0.474,
+      "step": 880
+    },
+    {
+      "epoch": 1.8809776833156215,
+      "grad_norm": 1.5,
+      "learning_rate": 0.00019528757572632622,
+      "loss": 0.4803,
+      "step": 885
+    },
+    {
+      "epoch": 1.8916046758767269,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.00019517426695952358,
+      "loss": 0.4798,
+      "step": 890
+    },
+    {
+      "epoch": 1.9022316684378322,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.00019505964575694385,
+      "loss": 0.4779,
+      "step": 895
+    },
+    {
+      "epoch": 1.9128586609989373,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.0001949437136991925,
+      "loss": 0.4796,
+      "step": 900
+    },
+    {
+      "epoch": 1.9234856535600424,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.00019482647238495152,
+      "loss": 0.4808,
+      "step": 905
+    },
+    {
+      "epoch": 1.9341126461211477,
+      "grad_norm": 0.69140625,
+      "learning_rate": 0.00019470792343095718,
+      "loss": 0.4837,
+      "step": 910
+    },
+    {
+      "epoch": 1.944739638682253,
+      "grad_norm": 0.4921875,
+      "learning_rate": 0.000194588068471978,
+      "loss": 0.4861,
+      "step": 915
+    },
+    {
+      "epoch": 1.9553666312433582,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.0001944669091607919,
+      "loss": 0.4632,
+      "step": 920
+    },
+    {
+      "epoch": 1.9659936238044633,
+      "grad_norm": 0.76171875,
+      "learning_rate": 0.00019434444716816374,
+      "loss": 0.4767,
+      "step": 925
+    },
+    {
+      "epoch": 1.9766206163655684,
+      "grad_norm": 0.47265625,
+      "learning_rate": 0.00019422068418282202,
+      "loss": 0.465,
+      "step": 930
+    },
+    {
+      "epoch": 1.9872476089266737,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.00019409562191143577,
+      "loss": 0.4613,
+      "step": 935
+    },
+    {
+      "epoch": 1.997874601487779,
+      "grad_norm": 0.484375,
+      "learning_rate": 0.00019396926207859084,
+      "loss": 0.4826,
+      "step": 940
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 1.3435465097427368,
+      "eval_runtime": 0.7971,
+      "eval_samples_per_second": 6.273,
+      "eval_steps_per_second": 1.255,
+      "step": 941
+    },
+    {
+      "epoch": 2.008501594048884,
+      "grad_norm": 0.734375,
+      "learning_rate": 0.00019384160642676635,
+      "loss": 0.4356,
+      "step": 945
+    },
+    {
+      "epoch": 2.0191285866099893,
+      "grad_norm": 0.5625,
+      "learning_rate": 0.00019371265671631037,
+      "loss": 0.4104,
+      "step": 950
+    },
+    {
+      "epoch": 2.0297555791710944,
+      "grad_norm": 0.484375,
+      "learning_rate": 0.0001935824147254159,
+      "loss": 0.4108,
+      "step": 955
+    },
+    {
+      "epoch": 2.0403825717322,
+      "grad_norm": 0.59765625,
+      "learning_rate": 0.00019345088225009626,
+      "loss": 0.4187,
+      "step": 960
+    },
+    {
+      "epoch": 2.051009564293305,
+      "grad_norm": 0.470703125,
+      "learning_rate": 0.00019331806110416027,
+      "loss": 0.4235,
+      "step": 965
+    },
+    {
+      "epoch": 2.06163655685441,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.0001931839531191873,
+      "loss": 0.4266,
+      "step": 970
+    },
+    {
+      "epoch": 2.0722635494155153,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.000193048560144502,
+      "loss": 0.4157,
+      "step": 975
+    },
+    {
+      "epoch": 2.082890541976621,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.00019291188404714878,
+      "loss": 0.4255,
+      "step": 980
+    },
+    {
+      "epoch": 2.093517534537726,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.00019277392671186608,
+      "loss": 0.4309,
+      "step": 985
+    },
+    {
+      "epoch": 2.104144527098831,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.0001926346900410604,
+      "loss": 0.4288,
+      "step": 990
+    },
+    {
+      "epoch": 2.114771519659936,
+      "grad_norm": 1.8359375,
+      "learning_rate": 0.00019249417595478002,
+      "loss": 0.4208,
+      "step": 995
+    },
+    {
+      "epoch": 2.1253985122210413,
+      "grad_norm": 0.427734375,
+      "learning_rate": 0.00019235238639068856,
+      "loss": 0.4249,
+      "step": 1000
+    },
+    {
+      "epoch": 2.136025504782147,
+      "grad_norm": 0.61328125,
+      "learning_rate": 0.00019220932330403823,
+      "loss": 0.427,
+      "step": 1005
+    },
+    {
+      "epoch": 2.146652497343252,
+      "grad_norm": 0.6875,
+      "learning_rate": 0.00019206498866764288,
+      "loss": 0.4276,
+      "step": 1010
+    },
+    {
+      "epoch": 2.157279489904357,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.00019191938447185084,
+      "loss": 0.4201,
+      "step": 1015
+    },
+    {
+      "epoch": 2.167906482465462,
+      "grad_norm": 1.6328125,
+      "learning_rate": 0.0001917725127245174,
+      "loss": 0.4299,
+      "step": 1020
+    },
+    {
+      "epoch": 2.1785334750265677,
+      "grad_norm": 0.5078125,
+      "learning_rate": 0.00019162437545097719,
+      "loss": 0.4268,
+      "step": 1025
+    },
+    {
+      "epoch": 2.189160467587673,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.0001914749746940161,
+      "loss": 0.4241,
+      "step": 1030
+    },
+    {
+      "epoch": 2.199787460148778,
+      "grad_norm": 0.69140625,
+      "learning_rate": 0.00019132431251384335,
+      "loss": 0.4297,
+      "step": 1035
+    },
+    {
+      "epoch": 2.210414452709883,
+      "grad_norm": 0.423828125,
+      "learning_rate": 0.00019117239098806295,
+      "loss": 0.4275,
+      "step": 1040
+    },
+    {
+      "epoch": 2.221041445270988,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.000191019212211645,
+      "loss": 0.4293,
+      "step": 1045
+    },
+    {
+      "epoch": 2.2316684378320937,
+      "grad_norm": 0.427734375,
+      "learning_rate": 0.00019086477829689685,
+      "loss": 0.421,
+      "step": 1050
+    },
+    {
+      "epoch": 2.242295430393199,
+      "grad_norm": 0.419921875,
+      "learning_rate": 0.00019070909137343408,
+      "loss": 0.4312,
+      "step": 1055
+    },
+    {
+      "epoch": 2.252922422954304,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.0001905521535881509,
+      "loss": 0.4299,
+      "step": 1060
+    },
+    {
+      "epoch": 2.263549415515409,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.00019039396710519076,
+      "loss": 0.4291,
+      "step": 1065
+    },
+    {
+      "epoch": 2.2741764080765146,
+      "grad_norm": 0.404296875,
+      "learning_rate": 0.00019023453410591635,
+      "loss": 0.4236,
+      "step": 1070
+    },
+    {
+      "epoch": 2.2848034006376197,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.00019007385678887975,
+      "loss": 0.4201,
+      "step": 1075
+    },
+    {
+      "epoch": 2.2954303931987248,
+      "grad_norm": 0.47265625,
+      "learning_rate": 0.00018991193736979175,
+      "loss": 0.4257,
+      "step": 1080
+    },
+    {
+      "epoch": 2.30605738575983,
+      "grad_norm": 0.439453125,
+      "learning_rate": 0.0001897487780814916,
+      "loss": 0.4219,
+      "step": 1085
+    },
+    {
+      "epoch": 2.316684378320935,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.00018958438117391618,
+      "loss": 0.4212,
+      "step": 1090
+    },
+    {
+      "epoch": 2.3273113708820405,
+      "grad_norm": 0.4765625,
+      "learning_rate": 0.00018941874891406882,
+      "loss": 0.4351,
+      "step": 1095
+    },
+    {
+      "epoch": 2.3379383634431457,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.00018925188358598813,
+      "loss": 0.4141,
+      "step": 1100
+    },
+    {
+      "epoch": 2.3485653560042508,
+      "grad_norm": 1.8515625,
+      "learning_rate": 0.0001890837874907166,
+      "loss": 0.4243,
+      "step": 1105
+    },
+    {
+      "epoch": 2.359192348565356,
+      "grad_norm": 0.66015625,
+      "learning_rate": 0.00018891446294626866,
+      "loss": 0.4408,
+      "step": 1110
+    },
+    {
+      "epoch": 2.369819341126461,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.00018874391228759893,
+      "loss": 0.4277,
+      "step": 1115
+    },
+    {
+      "epoch": 2.3804463336875665,
+      "grad_norm": 0.546875,
+      "learning_rate": 0.00018857213786656985,
+      "loss": 0.4279,
+      "step": 1120
+    },
+    {
+      "epoch": 2.3910733262486716,
+      "grad_norm": 0.77734375,
+      "learning_rate": 0.00018839914205191936,
+      "loss": 0.4259,
+      "step": 1125
+    },
+    {
+      "epoch": 2.4017003188097767,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.0001882249272292282,
+      "loss": 0.4287,
+      "step": 1130
+    },
+    {
+      "epoch": 2.412327311370882,
+      "grad_norm": 0.76953125,
+      "learning_rate": 0.00018804949580088692,
+      "loss": 0.4394,
+      "step": 1135
+    },
+    {
+      "epoch": 2.4229543039319874,
+      "grad_norm": 0.435546875,
+      "learning_rate": 0.00018787285018606297,
+      "loss": 0.4268,
+      "step": 1140
+    },
+    {
+      "epoch": 2.4335812964930925,
+      "grad_norm": 0.462890625,
+      "learning_rate": 0.00018769499282066717,
+      "loss": 0.4271,
+      "step": 1145
+    },
+    {
+      "epoch": 2.4442082890541976,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.00018751592615732005,
+      "loss": 0.4338,
+      "step": 1150
+    },
+    {
+      "epoch": 2.4548352816153027,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.0001873356526653183,
+      "loss": 0.4306,
+      "step": 1155
+    },
+    {
+      "epoch": 2.4654622741764083,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.0001871541748306005,
+      "loss": 0.4374,
+      "step": 1160
+    },
+    {
+      "epoch": 2.4760892667375134,
+      "grad_norm": 0.44921875,
+      "learning_rate": 0.00018697149515571284,
+      "loss": 0.429,
+      "step": 1165
+    },
+    {
+      "epoch": 2.4867162592986185,
+      "grad_norm": 1.3203125,
+      "learning_rate": 0.00018678761615977468,
+      "loss": 0.4351,
+      "step": 1170
+    },
+    {
+      "epoch": 2.4973432518597236,
+      "grad_norm": 0.47265625,
+      "learning_rate": 0.00018660254037844388,
+      "loss": 0.4264,
+      "step": 1175
+    },
+    {
+      "epoch": 2.5079702444208287,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.00018641627036388169,
+      "loss": 0.431,
+      "step": 1180
+    },
+    {
+      "epoch": 2.5185972369819343,
+      "grad_norm": 0.41015625,
+      "learning_rate": 0.00018622880868471756,
+      "loss": 0.4265,
+      "step": 1185
+    },
+    {
+      "epoch": 2.5292242295430394,
+      "grad_norm": 0.400390625,
+      "learning_rate": 0.00018604015792601396,
+      "loss": 0.4259,
+      "step": 1190
+    },
+    {
+      "epoch": 2.5398512221041445,
+      "grad_norm": 7.6875,
+      "learning_rate": 0.00018585032068923032,
+      "loss": 0.4309,
+      "step": 1195
+    },
+    {
+      "epoch": 2.5504782146652496,
+      "grad_norm": 0.4296875,
+      "learning_rate": 0.00018565929959218758,
+      "loss": 0.4254,
+      "step": 1200
+    },
+    {
+      "epoch": 2.5611052072263547,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.00018546709726903178,
+      "loss": 0.4253,
+      "step": 1205
+    },
+    {
+      "epoch": 2.5717321997874603,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.0001852737163701979,
+      "loss": 0.4403,
+      "step": 1210
+    },
+    {
+      "epoch": 2.5823591923485654,
+      "grad_norm": 0.640625,
+      "learning_rate": 0.00018507915956237326,
+      "loss": 0.4326,
+      "step": 1215
+    },
+    {
+      "epoch": 2.5929861849096705,
+      "grad_norm": 2.25,
+      "learning_rate": 0.00018488342952846073,
+      "loss": 0.437,
+      "step": 1220
+    },
+    {
+      "epoch": 2.603613177470776,
+      "grad_norm": 0.90625,
+      "learning_rate": 0.00018468652896754177,
+      "loss": 0.4344,
+      "step": 1225
+    },
+    {
+      "epoch": 2.614240170031881,
+      "grad_norm": 0.87890625,
+      "learning_rate": 0.0001844884605948392,
+      "loss": 0.4387,
+      "step": 1230
+    },
+    {
+      "epoch": 2.6248671625929862,
+      "grad_norm": 0.66796875,
+      "learning_rate": 0.0001842892271416797,
+      "loss": 0.4371,
+      "step": 1235
+    },
+    {
+      "epoch": 2.6354941551540914,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.00018408883135545632,
+      "loss": 0.4345,
+      "step": 1240
+    },
+    {
+      "epoch": 2.6461211477151965,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.00018388727599959033,
+      "loss": 0.4405,
+      "step": 1245
+    },
+    {
+      "epoch": 2.656748140276302,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.00018368456385349334,
+      "loss": 0.4369,
+      "step": 1250
+    },
+    {
+      "epoch": 2.667375132837407,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.0001834806977125288,
+      "loss": 0.4354,
+      "step": 1255
+    },
+    {
+      "epoch": 2.6780021253985122,
+      "grad_norm": 0.41796875,
+      "learning_rate": 0.0001832756803879737,
+      "loss": 0.4297,
+      "step": 1260
+    },
+    {
+      "epoch": 2.6886291179596173,
+      "grad_norm": 0.578125,
+      "learning_rate": 0.00018306951470697946,
+      "loss": 0.4268,
+      "step": 1265
+    },
+    {
+      "epoch": 2.6992561105207225,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.0001828622035125332,
+      "loss": 0.4299,
+      "step": 1270
+    },
+    {
+      "epoch": 2.709883103081828,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.0001826537496634186,
+      "loss": 0.4362,
+      "step": 1275
+    },
+    {
+      "epoch": 2.720510095642933,
+      "grad_norm": 1.484375,
+      "learning_rate": 0.00018244415603417603,
+      "loss": 0.4364,
+      "step": 1280
+    },
+    {
+      "epoch": 2.731137088204038,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.0001822334255150635,
+      "loss": 0.439,
+      "step": 1285
+    },
+    {
+      "epoch": 2.7417640807651433,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.00018202156101201645,
+      "loss": 0.4342,
+      "step": 1290
+    },
+    {
+      "epoch": 2.7523910733262484,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.0001818085654466076,
+      "loss": 0.4308,
+      "step": 1295
+    },
+    {
+      "epoch": 2.763018065887354,
+      "grad_norm": 0.494140625,
+      "learning_rate": 0.00018159444175600703,
+      "loss": 0.4306,
+      "step": 1300
+    },
+    {
+      "epoch": 2.773645058448459,
+      "grad_norm": 1.125,
+      "learning_rate": 0.00018137919289294135,
+      "loss": 0.4342,
+      "step": 1305
+    },
+    {
+      "epoch": 2.784272051009564,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.00018116282182565311,
+      "loss": 0.428,
+      "step": 1310
+    },
+    {
+      "epoch": 2.7948990435706698,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.00018094533153785984,
+      "loss": 0.4221,
+      "step": 1315
+    },
+    {
+      "epoch": 2.805526036131775,
+      "grad_norm": 1.5859375,
+      "learning_rate": 0.00018072672502871296,
+      "loss": 0.433,
+      "step": 1320
+    },
+    {
+      "epoch": 2.81615302869288,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.0001805070053127563,
+      "loss": 0.4306,
+      "step": 1325
+    },
+    {
+      "epoch": 2.826780021253985,
+      "grad_norm": 0.6171875,
+      "learning_rate": 0.00018028617541988472,
+      "loss": 0.4217,
+      "step": 1330
+    },
+    {
+      "epoch": 2.83740701381509,
+      "grad_norm": 0.49609375,
+      "learning_rate": 0.00018006423839530205,
+      "loss": 0.426,
+      "step": 1335
+    },
+    {
+      "epoch": 2.8480340063761957,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.00017984119729947944,
+      "loss": 0.4402,
+      "step": 1340
+    },
+    {
+      "epoch": 2.858660998937301,
+      "grad_norm": 0.4609375,
+      "learning_rate": 0.00017961705520811293,
+      "loss": 0.4306,
+      "step": 1345
+    },
+    {
+      "epoch": 2.869287991498406,
+      "grad_norm": 0.375,
+      "learning_rate": 0.000179391815212081,
+      "loss": 0.434,
+      "step": 1350
+    },
+    {
+      "epoch": 2.879914984059511,
+      "grad_norm": 0.70703125,
+      "learning_rate": 0.00017916548041740213,
+      "loss": 0.4295,
+      "step": 1355
+    },
+    {
+      "epoch": 2.890541976620616,
+      "grad_norm": 0.72265625,
+      "learning_rate": 0.0001789380539451919,
+      "loss": 0.4373,
+      "step": 1360
+    },
+    {
+      "epoch": 2.9011689691817217,
+      "grad_norm": 0.4296875,
+      "learning_rate": 0.00017870953893161975,
+      "loss": 0.4287,
+      "step": 1365
+    },
+    {
+      "epoch": 2.911795961742827,
+      "grad_norm": 0.5625,
+      "learning_rate": 0.0001784799385278661,
+      "loss": 0.4358,
+      "step": 1370
+    },
+    {
+      "epoch": 2.922422954303932,
+      "grad_norm": 0.419921875,
+      "learning_rate": 0.00017824925590007864,
+      "loss": 0.4304,
+      "step": 1375
+    },
+    {
+      "epoch": 2.933049946865037,
+      "grad_norm": 0.98046875,
+      "learning_rate": 0.0001780174942293287,
+      "loss": 0.4296,
+      "step": 1380
+    },
+    {
+      "epoch": 2.943676939426142,
+      "grad_norm": 0.6796875,
+      "learning_rate": 0.00017778465671156743,
+      "loss": 0.4326,
+      "step": 1385
+    },
+    {
+      "epoch": 2.9543039319872477,
+      "grad_norm": 0.486328125,
+      "learning_rate": 0.00017755074655758174,
+      "loss": 0.4348,
+      "step": 1390
+    },
+    {
+      "epoch": 2.964930924548353,
+      "grad_norm": 0.61328125,
+      "learning_rate": 0.0001773157669929499,
+      "loss": 0.4345,
+      "step": 1395
+    },
+    {
+      "epoch": 2.975557917109458,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.00017707972125799735,
+      "loss": 0.4259,
+      "step": 1400
+    },
+    {
+      "epoch": 2.9861849096705635,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.00017684261260775163,
+      "loss": 0.4306,
+      "step": 1405
+    },
+    {
+      "epoch": 2.9968119022316686,
+      "grad_norm": 0.60546875,
+      "learning_rate": 0.0001766044443118978,
+      "loss": 0.4369,
+      "step": 1410
+    },
+    {
+      "epoch": 2.9989373007438895,
+      "eval_loss": 1.4787013530731201,
+      "eval_runtime": 0.8123,
+      "eval_samples_per_second": 6.155,
+      "eval_steps_per_second": 1.231,
+      "step": 1411
+    },
+    {
+      "epoch": 3.0074388947927737,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.00017636521965473323,
+      "loss": 0.3835,
+      "step": 1415
+    },
+    {
+      "epoch": 3.018065887353879,
+      "grad_norm": 0.455078125,
+      "learning_rate": 0.0001761249419351222,
+      "loss": 0.3594,
+      "step": 1420
+    },
+    {
+      "epoch": 3.028692879914984,
+      "grad_norm": 0.478515625,
+      "learning_rate": 0.00017588361446645073,
+      "loss": 0.3596,
+      "step": 1425
+    },
+    {
+      "epoch": 3.0393198724760895,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.00017564124057658056,
+      "loss": 0.3504,
+      "step": 1430
+    },
+    {
+      "epoch": 3.0499468650371946,
+      "grad_norm": 0.4375,
+      "learning_rate": 0.00017539782360780334,
+      "loss": 0.3575,
+      "step": 1435
+    },
+    {
+      "epoch": 3.0605738575982997,
+      "grad_norm": 0.419921875,
+      "learning_rate": 0.00017515336691679477,
+      "loss": 0.3619,
+      "step": 1440
+    },
+    {
+      "epoch": 3.071200850159405,
+      "grad_norm": 0.41015625,
+      "learning_rate": 0.0001749078738745679,
+      "loss": 0.3678,
+      "step": 1445
+    },
+    {
+      "epoch": 3.08182784272051,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.0001746613478664271,
+      "loss": 0.3551,
+      "step": 1450
+    },
+    {
+      "epoch": 3.0924548352816155,
+      "grad_norm": 0.75,
+      "learning_rate": 0.00017441379229192098,
+      "loss": 0.3668,
+      "step": 1455
+    },
+    {
+      "epoch": 3.1030818278427206,
+      "grad_norm": 0.498046875,
+      "learning_rate": 0.00017416521056479577,
+      "loss": 0.3664,
+      "step": 1460
+    },
+    {
+      "epoch": 3.1137088204038257,
+      "grad_norm": 1.6640625,
+      "learning_rate": 0.0001739156061129481,
+      "loss": 0.3709,
+      "step": 1465
+    },
+    {
+      "epoch": 3.124335812964931,
+      "grad_norm": 0.8125,
+      "learning_rate": 0.0001736649823783779,
+      "loss": 0.3612,
+      "step": 1470
+    },
+    {
+      "epoch": 3.134962805526036,
+      "grad_norm": 0.64453125,
+      "learning_rate": 0.00017341334281714064,
+      "loss": 0.3619,
+      "step": 1475
+    },
+    {
+      "epoch": 3.1455897980871415,
+      "grad_norm": 1.0859375,
+      "learning_rate": 0.00017316069089930007,
+      "loss": 0.37,
+      "step": 1480
+    },
+    {
+      "epoch": 3.1562167906482466,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.00017290703010887994,
+      "loss": 0.3717,
+      "step": 1485
+    },
+    {
+      "epoch": 3.1668437832093517,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.00017265236394381633,
+      "loss": 0.3749,
+      "step": 1490
+    },
+    {
+      "epoch": 3.177470775770457,
+      "grad_norm": 0.5,
+      "learning_rate": 0.00017239669591590916,
+      "loss": 0.3754,
+      "step": 1495
+    },
+    {
+      "epoch": 3.1880977683315623,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.00017214002955077393,
+      "loss": 0.3726,
+      "step": 1500
+    },
+    {
+      "epoch": 3.1987247608926674,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.00017188236838779295,
+      "loss": 0.3726,
+      "step": 1505
+    },
+    {
+      "epoch": 3.2093517534537725,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.00017162371598006666,
+      "loss": 0.3798,
+      "step": 1510
+    },
+    {
+      "epoch": 3.2199787460148777,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.00017136407589436457,
+      "loss": 0.3767,
+      "step": 1515
+    },
+    {
+      "epoch": 3.230605738575983,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.0001711034517110761,
+      "loss": 0.3763,
+      "step": 1520
+    },
+    {
+      "epoch": 3.2412327311370883,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.00017084184702416115,
+      "loss": 0.3703,
+      "step": 1525
+    },
+    {
+      "epoch": 3.2518597236981934,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.0001705792654411007,
+      "loss": 0.378,
+      "step": 1530
+    },
+    {
+      "epoch": 3.2624867162592985,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.00017031571058284678,
+      "loss": 0.3818,
+      "step": 1535
+    },
+    {
+      "epoch": 3.2731137088204036,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.00017005118608377288,
+      "loss": 0.3709,
+      "step": 1540
+    },
+    {
+      "epoch": 3.283740701381509,
+      "grad_norm": 0.59375,
+      "learning_rate": 0.00016978569559162357,
+      "loss": 0.3761,
+      "step": 1545
+    },
+    {
+      "epoch": 3.2943676939426143,
+      "grad_norm": 0.44921875,
+      "learning_rate": 0.00016951924276746425,
+      "loss": 0.3786,
+      "step": 1550
+    },
+    {
+      "epoch": 3.3049946865037194,
+      "grad_norm": 0.451171875,
+      "learning_rate": 0.00016925183128563078,
+      "loss": 0.3661,
+      "step": 1555
+    },
+    {
+      "epoch": 3.3156216790648245,
+      "grad_norm": 0.490234375,
+      "learning_rate": 0.00016898346483367867,
+      "loss": 0.3721,
+      "step": 1560
+    },
+    {
+      "epoch": 3.3262486716259296,
+      "grad_norm": 0.4375,
+      "learning_rate": 0.0001687141471123324,
+      "loss": 0.3752,
+      "step": 1565
+    },
+    {
+      "epoch": 3.336875664187035,
+      "grad_norm": 0.4921875,
+      "learning_rate": 0.00016844388183543418,
+      "loss": 0.3784,
+      "step": 1570
+    },
+    {
+      "epoch": 3.3475026567481403,
+      "grad_norm": 0.44921875,
+      "learning_rate": 0.00016817267272989286,
+      "loss": 0.3807,
+      "step": 1575
+    },
+    {
+      "epoch": 3.3581296493092454,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.00016790052353563253,
+      "loss": 0.3887,
+      "step": 1580
+    },
+    {
+      "epoch": 3.3687566418703505,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.0001676274380055409,
+      "loss": 0.3831,
+      "step": 1585
+    },
+    {
+      "epoch": 3.379383634431456,
+      "grad_norm": 0.45703125,
+      "learning_rate": 0.00016735341990541764,
+      "loss": 0.3772,
+      "step": 1590
+    },
+    {
+      "epoch": 3.390010626992561,
+      "grad_norm": 0.56640625,
+      "learning_rate": 0.00016707847301392236,
+      "loss": 0.3799,
+      "step": 1595
+    },
+    {
+      "epoch": 3.4006376195536663,
+      "grad_norm": 0.431640625,
+      "learning_rate": 0.0001668026011225225,
+      "loss": 0.3854,
+      "step": 1600
+    },
+    {
+      "epoch": 3.4112646121147714,
+      "grad_norm": 0.439453125,
+      "learning_rate": 0.00016652580803544112,
+      "loss": 0.3725,
+      "step": 1605
+    },
+    {
+      "epoch": 3.421891604675877,
+      "grad_norm": 0.439453125,
+      "learning_rate": 0.00016624809756960444,
+      "loss": 0.3785,
+      "step": 1610
+    },
+    {
+      "epoch": 3.432518597236982,
+      "grad_norm": 0.490234375,
+      "learning_rate": 0.00016596947355458904,
+      "loss": 0.3803,
+      "step": 1615
+    },
+    {
+      "epoch": 3.443145589798087,
+      "grad_norm": 0.640625,
+      "learning_rate": 0.0001656899398325693,
+      "loss": 0.3893,
+      "step": 1620
+    },
+    {
+      "epoch": 3.4537725823591923,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.00016540950025826422,
+      "loss": 0.3776,
+      "step": 1625
+    },
+    {
+      "epoch": 3.4643995749202974,
+      "grad_norm": 0.494140625,
+      "learning_rate": 0.0001651281586988844,
+      "loss": 0.3802,
+      "step": 1630
+    },
+    {
+      "epoch": 3.475026567481403,
+      "grad_norm": 0.73046875,
+      "learning_rate": 0.00016484591903407857,
+      "loss": 0.3765,
+      "step": 1635
+    },
+    {
+      "epoch": 3.485653560042508,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.00016456278515588024,
+      "loss": 0.3843,
+      "step": 1640
+    },
+    {
+      "epoch": 3.496280552603613,
+      "grad_norm": 0.59375,
+      "learning_rate": 0.00016427876096865394,
+      "loss": 0.3748,
+      "step": 1645
+    },
+    {
+      "epoch": 3.5069075451647183,
+      "grad_norm": 0.4609375,
+      "learning_rate": 0.00016399385038904138,
+      "loss": 0.3785,
+      "step": 1650
+    },
+    {
+      "epoch": 3.5175345377258234,
+      "grad_norm": 0.466796875,
+      "learning_rate": 0.00016370805734590747,
+      "loss": 0.3879,
+      "step": 1655
+    },
+    {
+      "epoch": 3.528161530286929,
+      "grad_norm": 0.4609375,
+      "learning_rate": 0.00016342138578028613,
+      "loss": 0.382,
+      "step": 1660
+    },
+    {
+      "epoch": 3.538788522848034,
+      "grad_norm": 0.423828125,
+      "learning_rate": 0.00016313383964532596,
+      "loss": 0.3915,
+      "step": 1665
+    },
+    {
+      "epoch": 3.549415515409139,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.00016284542290623567,
+      "loss": 0.3767,
+      "step": 1670
+    },
+    {
+      "epoch": 3.5600425079702447,
+      "grad_norm": 0.443359375,
+      "learning_rate": 0.0001625561395402295,
+      "loss": 0.373,
+      "step": 1675
+    },
+    {
+      "epoch": 3.57066950053135,
+      "grad_norm": 0.4453125,
+      "learning_rate": 0.00016226599353647228,
+      "loss": 0.3811,
+      "step": 1680
+    },
+    {
+      "epoch": 3.581296493092455,
+      "grad_norm": 0.43359375,
+      "learning_rate": 0.00016197498889602448,
+      "loss": 0.3842,
+      "step": 1685
+    },
+    {
+      "epoch": 3.59192348565356,
+      "grad_norm": 0.59765625,
+      "learning_rate": 0.00016168312963178697,
+      "loss": 0.3864,
+      "step": 1690
+    },
+    {
+      "epoch": 3.602550478214665,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.00016139041976844583,
+      "loss": 0.3835,
+      "step": 1695
+    },
+    {
+      "epoch": 3.6131774707757707,
+      "grad_norm": 0.578125,
+      "learning_rate": 0.00016109686334241655,
+      "loss": 0.3801,
+      "step": 1700
+    },
+    {
+      "epoch": 3.623804463336876,
+      "grad_norm": 0.5078125,
+      "learning_rate": 0.00016080246440178874,
+      "loss": 0.382,
+      "step": 1705
+    },
+    {
+      "epoch": 3.634431455897981,
+      "grad_norm": 0.66796875,
+      "learning_rate": 0.00016050722700627012,
+      "loss": 0.3923,
+      "step": 1710
+    },
+    {
+      "epoch": 3.645058448459086,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.00016021115522713047,
+      "loss": 0.3836,
+      "step": 1715
+    },
+    {
+      "epoch": 3.655685441020191,
+      "grad_norm": 0.77734375,
+      "learning_rate": 0.0001599142531471456,
+      "loss": 0.3801,
+      "step": 1720
+    },
+    {
+      "epoch": 3.6663124335812967,
+      "grad_norm": 0.61328125,
+      "learning_rate": 0.00015961652486054103,
+      "loss": 0.3871,
+      "step": 1725
+    },
+    {
+      "epoch": 3.6769394261424018,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.00015931797447293552,
+      "loss": 0.3843,
+      "step": 1730
+    },
+    {
+      "epoch": 3.687566418703507,
+      "grad_norm": 0.4765625,
+      "learning_rate": 0.00015901860610128448,
+      "loss": 0.3846,
+      "step": 1735
+    },
+    {
+      "epoch": 3.698193411264612,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.00015871842387382305,
+      "loss": 0.3788,
+      "step": 1740
+    },
+    {
+      "epoch": 3.708820403825717,
+      "grad_norm": 0.66015625,
+      "learning_rate": 0.00015841743193000944,
+      "loss": 0.3899,
+      "step": 1745
+    },
+    {
+      "epoch": 3.7194473963868226,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.00015811563442046767,
+      "loss": 0.3863,
+      "step": 1750
+    },
+    {
+      "epoch": 3.7300743889479278,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.00015781303550693024,
+      "loss": 0.3822,
+      "step": 1755
+    },
+    {
+      "epoch": 3.740701381509033,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.00015750963936218105,
+      "loss": 0.3821,
+      "step": 1760
+    },
+    {
+      "epoch": 3.7513283740701384,
+      "grad_norm": 0.70703125,
+      "learning_rate": 0.00015720545016999752,
+      "loss": 0.3809,
+      "step": 1765
+    },
+    {
+      "epoch": 3.761955366631243,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.00015690047212509316,
+      "loss": 0.3809,
+      "step": 1770
+    },
+    {
+      "epoch": 3.7725823591923486,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.00015659470943305955,
+      "loss": 0.3815,
+      "step": 1775
+    },
+    {
+      "epoch": 3.7832093517534537,
+      "grad_norm": 0.443359375,
+      "learning_rate": 0.00015628816631030836,
+      "loss": 0.3808,
+      "step": 1780
+    },
+    {
+      "epoch": 3.793836344314559,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.00015598084698401342,
+      "loss": 0.38,
+      "step": 1785
+    },
+    {
+      "epoch": 3.8044633368756644,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.00015567275569205218,
+      "loss": 0.3851,
+      "step": 1790
+    },
+    {
+      "epoch": 3.8150903294367695,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.00015536389668294724,
+      "loss": 0.3917,
+      "step": 1795
+    },
+    {
+      "epoch": 3.8257173219978746,
+      "grad_norm": 0.40625,
+      "learning_rate": 0.00015505427421580808,
+      "loss": 0.3799,
+      "step": 1800
+    },
+    {
+      "epoch": 3.8363443145589797,
+      "grad_norm": 0.45703125,
+      "learning_rate": 0.000154743892560272,
+      "loss": 0.3846,
+      "step": 1805
+    },
+    {
+      "epoch": 3.846971307120085,
+      "grad_norm": 0.46484375,
+      "learning_rate": 0.00015443275599644538,
+      "loss": 0.3865,
+      "step": 1810
+    },
+    {
+      "epoch": 3.8575982996811904,
+      "grad_norm": 0.486328125,
+      "learning_rate": 0.0001541208688148447,
+      "loss": 0.3802,
+      "step": 1815
+    },
+    {
+      "epoch": 3.8682252922422955,
+      "grad_norm": 0.462890625,
+      "learning_rate": 0.00015380823531633729,
+      "loss": 0.3823,
+      "step": 1820
+    },
+    {
+      "epoch": 3.8788522848034006,
+      "grad_norm": 0.61328125,
+      "learning_rate": 0.00015349485981208202,
+      "loss": 0.3866,
+      "step": 1825
+    },
+    {
+      "epoch": 3.8894792773645057,
+      "grad_norm": 0.490234375,
+      "learning_rate": 0.00015318074662346994,
+      "loss": 0.3758,
+      "step": 1830
+    },
+    {
+      "epoch": 3.900106269925611,
+      "grad_norm": 0.5625,
+      "learning_rate": 0.00015286590008206465,
+      "loss": 0.3858,
+      "step": 1835
+    },
+    {
+      "epoch": 3.9107332624867164,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.00015255032452954245,
+      "loss": 0.3867,
+      "step": 1840
+    },
+    {
+      "epoch": 3.9213602550478215,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.00015223402431763269,
+      "loss": 0.3893,
+      "step": 1845
+    },
+    {
+      "epoch": 3.9319872476089266,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.00015191700380805752,
+      "loss": 0.3791,
+      "step": 1850
+    },
+    {
+      "epoch": 3.942614240170032,
+      "grad_norm": 0.51953125,
+      "learning_rate": 0.00015159926737247202,
+      "loss": 0.3932,
+      "step": 1855
+    },
+    {
+      "epoch": 3.953241232731137,
+      "grad_norm": 0.69921875,
+      "learning_rate": 0.00015128081939240357,
+      "loss": 0.3794,
+      "step": 1860
+    },
+    {
+      "epoch": 3.9638682252922424,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.00015096166425919175,
+      "loss": 0.3857,
+      "step": 1865
+    },
+    {
+      "epoch": 3.9744952178533475,
+      "grad_norm": 0.412109375,
+      "learning_rate": 0.00015064180637392764,
+      "loss": 0.3887,
+      "step": 1870
+    },
+    {
+      "epoch": 3.9851222104144526,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.0001503212501473931,
+      "loss": 0.3886,
+      "step": 1875
+    },
+    {
+      "epoch": 3.995749202975558,
+      "grad_norm": 0.5078125,
+      "learning_rate": 0.00015000000000000001,
+      "loss": 0.3819,
+      "step": 1880
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 1.7431671619415283,
+      "eval_runtime": 0.7941,
+      "eval_samples_per_second": 6.296,
+      "eval_steps_per_second": 1.259,
+      "step": 1882
+    },
+    {
+      "epoch": 4.006376195536663,
+      "grad_norm": 0.47265625,
+      "learning_rate": 0.0001496780603617293,
+      "loss": 0.3464,
+      "step": 1885
+    },
+    {
+      "epoch": 4.017003188097768,
+      "grad_norm": 0.46484375,
+      "learning_rate": 0.00014935543567206984,
+      "loss": 0.3125,
+      "step": 1890
+    },
+    {
+      "epoch": 4.027630180658874,
+      "grad_norm": 0.5546875,
+      "learning_rate": 0.00014903213037995724,
+      "loss": 0.3159,
+      "step": 1895
+    },
+    {
+      "epoch": 4.038257173219979,
+      "grad_norm": 0.478515625,
+      "learning_rate": 0.00014870814894371245,
+      "loss": 0.3227,
+      "step": 1900
+    },
+    {
+      "epoch": 4.048884165781084,
+      "grad_norm": 0.470703125,
+      "learning_rate": 0.00014838349583098045,
+      "loss": 0.3122,
+      "step": 1905
+    },
+    {
+      "epoch": 4.059511158342189,
+      "grad_norm": 0.43359375,
+      "learning_rate": 0.00014805817551866838,
+      "loss": 0.3194,
+      "step": 1910
+    },
+    {
+      "epoch": 4.070138150903294,
+      "grad_norm": 0.458984375,
+      "learning_rate": 0.00014773219249288402,
+      "loss": 0.3123,
+      "step": 1915
+    },
+    {
+      "epoch": 4.0807651434644,
+      "grad_norm": 0.4765625,
+      "learning_rate": 0.00014740555124887375,
+      "loss": 0.3254,
+      "step": 1920
+    },
+    {
+      "epoch": 4.0913921360255046,
+      "grad_norm": 0.478515625,
+      "learning_rate": 0.00014707825629096084,
+      "loss": 0.3178,
+      "step": 1925
+    },
+    {
+      "epoch": 4.10201912858661,
+      "grad_norm": 0.609375,
+      "learning_rate": 0.00014675031213248296,
+      "loss": 0.3214,
+      "step": 1930
+    },
+    {
+      "epoch": 4.112646121147715,
+      "grad_norm": 0.470703125,
+      "learning_rate": 0.00014642172329573026,
+      "loss": 0.321,
+      "step": 1935
+    },
+    {
+      "epoch": 4.12327311370882,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.00014609249431188278,
+      "loss": 0.3227,
+      "step": 1940
+    },
+    {
+      "epoch": 4.133900106269926,
+      "grad_norm": 0.470703125,
+      "learning_rate": 0.00014576262972094828,
+      "loss": 0.3277,
+      "step": 1945
+    },
+    {
+      "epoch": 4.1445270988310305,
+      "grad_norm": 0.578125,
+      "learning_rate": 0.0001454321340716992,
+      "loss": 0.3216,
+      "step": 1950
+    },
+    {
+      "epoch": 4.155154091392136,
+      "grad_norm": 0.4453125,
+      "learning_rate": 0.00014510101192161018,
+      "loss": 0.3196,
+      "step": 1955
+    },
+    {
+      "epoch": 4.165781083953242,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.00014476926783679538,
+      "loss": 0.3284,
+      "step": 1960
+    },
+    {
+      "epoch": 4.176408076514346,
+      "grad_norm": 0.46875,
+      "learning_rate": 0.00014443690639194515,
+      "loss": 0.3244,
+      "step": 1965
+    },
+    {
+      "epoch": 4.187035069075452,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.00014410393217026318,
+      "loss": 0.3205,
+      "step": 1970
+    },
+    {
+      "epoch": 4.1976620616365565,
+      "grad_norm": 0.71875,
+      "learning_rate": 0.0001437703497634032,
+      "loss": 0.3216,
+      "step": 1975
+    },
+    {
+      "epoch": 4.208289054197662,
+      "grad_norm": 0.76171875,
+      "learning_rate": 0.00014343616377140582,
+      "loss": 0.3178,
+      "step": 1980
+    },
+    {
+      "epoch": 4.218916046758768,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.00014310137880263482,
+      "loss": 0.3219,
+      "step": 1985
+    },
+    {
+      "epoch": 4.229543039319872,
+      "grad_norm": 0.5546875,
+      "learning_rate": 0.00014276599947371388,
+      "loss": 0.3262,
+      "step": 1990
+    },
+    {
+      "epoch": 4.240170031880978,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.00014243003040946274,
+      "loss": 0.3287,
+      "step": 1995
+    },
+    {
+      "epoch": 4.2507970244420825,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.0001420934762428335,
+      "loss": 0.328,
+      "step": 2000
+    },
+    {
+      "epoch": 4.261424017003188,
+      "grad_norm": 0.64453125,
+      "learning_rate": 0.00014175634161484675,
+      "loss": 0.3334,
+      "step": 2005
+    },
+    {
+      "epoch": 4.272051009564294,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.00014141863117452745,
+      "loss": 0.3255,
+      "step": 2010
+    },
+    {
+      "epoch": 4.282678002125398,
+      "grad_norm": 0.419921875,
+      "learning_rate": 0.00014108034957884094,
+      "loss": 0.3226,
+      "step": 2015
+    },
+    {
+      "epoch": 4.293304994686504,
+      "grad_norm": 0.48046875,
+      "learning_rate": 0.0001407415014926288,
+      "loss": 0.3314,
+      "step": 2020
+    },
+    {
+      "epoch": 4.3039319872476085,
+      "grad_norm": 0.67578125,
+      "learning_rate": 0.00014040209158854426,
+      "loss": 0.3334,
+      "step": 2025
+    },
+    {
+      "epoch": 4.314558979808714,
+      "grad_norm": 0.48828125,
+      "learning_rate": 0.00014006212454698797,
+      "loss": 0.3349,
+      "step": 2030
+    },
+    {
+      "epoch": 4.32518597236982,
+      "grad_norm": 0.455078125,
+      "learning_rate": 0.00013972160505604342,
+      "loss": 0.3301,
+      "step": 2035
+    },
+    {
+      "epoch": 4.335812964930924,
+      "grad_norm": 0.6171875,
+      "learning_rate": 0.00013938053781141222,
+      "loss": 0.325,
+      "step": 2040
+    },
+    {
+      "epoch": 4.34643995749203,
+      "grad_norm": 0.5,
+      "learning_rate": 0.00013903892751634947,
+      "loss": 0.3293,
+      "step": 2045
+    },
+    {
+      "epoch": 4.357066950053135,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.00013869677888159887,
+      "loss": 0.3285,
+      "step": 2050
+    },
+    {
+      "epoch": 4.36769394261424,
+      "grad_norm": 0.71875,
+      "learning_rate": 0.00013835409662532762,
+      "loss": 0.3271,
+      "step": 2055
+    },
+    {
+      "epoch": 4.378320935175346,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.00013801088547306148,
+      "loss": 0.3296,
+      "step": 2060
+    },
+    {
+      "epoch": 4.38894792773645,
+      "grad_norm": 0.6015625,
+      "learning_rate": 0.0001376671501576197,
+      "loss": 0.3292,
+      "step": 2065
+    },
+    {
+      "epoch": 4.399574920297556,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.00013732289541904948,
+      "loss": 0.3324,
+      "step": 2070
+    },
+    {
+      "epoch": 4.410201912858661,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.00013697812600456093,
+      "loss": 0.3326,
+      "step": 2075
+    },
+    {
+      "epoch": 4.420828905419766,
+      "grad_norm": 0.462890625,
+      "learning_rate": 0.00013663284666846134,
+      "loss": 0.3329,
+      "step": 2080
+    },
+    {
+      "epoch": 4.431455897980872,
+      "grad_norm": 0.65234375,
+      "learning_rate": 0.00013628706217208976,
+      "loss": 0.3305,
+      "step": 2085
+    },
+    {
+      "epoch": 4.442082890541976,
+      "grad_norm": 0.4765625,
+      "learning_rate": 0.00013594077728375128,
+      "loss": 0.3305,
+      "step": 2090
+    },
+    {
+      "epoch": 4.452709883103082,
+      "grad_norm": 0.451171875,
+      "learning_rate": 0.0001355939967786514,
+      "loss": 0.3327,
+      "step": 2095
+    },
+    {
+      "epoch": 4.463336875664187,
+      "grad_norm": 0.427734375,
+      "learning_rate": 0.00013524672543882996,
+      "loss": 0.3329,
+      "step": 2100
+    },
+    {
+      "epoch": 4.473963868225292,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.00013489896805309542,
+      "loss": 0.3376,
+      "step": 2105
+    },
+    {
+      "epoch": 4.484590860786398,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.00013455072941695863,
+      "loss": 0.329,
+      "step": 2110
+    },
+    {
+      "epoch": 4.495217853347503,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.00013420201433256689,
+      "loss": 0.333,
+      "step": 2115
+    },
+    {
+      "epoch": 4.505844845908608,
+      "grad_norm": 0.470703125,
+      "learning_rate": 0.00013385282760863758,
+      "loss": 0.328,
+      "step": 2120
+    },
+    {
+      "epoch": 4.516471838469713,
+      "grad_norm": 0.439453125,
+      "learning_rate": 0.00013350317406039187,
+      "loss": 0.3321,
+      "step": 2125
+    },
+    {
+      "epoch": 4.527098831030818,
+      "grad_norm": 0.48046875,
+      "learning_rate": 0.00013315305850948846,
+      "loss": 0.3325,
+      "step": 2130
+    },
+    {
+      "epoch": 4.537725823591924,
+      "grad_norm": 0.48046875,
+      "learning_rate": 0.0001328024857839569,
+      "loss": 0.3324,
+      "step": 2135
+    },
+    {
+      "epoch": 4.548352816153029,
+      "grad_norm": 0.447265625,
+      "learning_rate": 0.00013245146071813114,
+      "loss": 0.3351,
+      "step": 2140
+    },
+    {
+      "epoch": 4.558979808714134,
+      "grad_norm": 0.41796875,
+      "learning_rate": 0.00013209998815258273,
+      "loss": 0.3264,
+      "step": 2145
+    },
+    {
+      "epoch": 4.569606801275239,
+      "grad_norm": 0.6640625,
+      "learning_rate": 0.00013174807293405428,
+      "loss": 0.3303,
+      "step": 2150
+    },
+    {
+      "epoch": 4.580233793836344,
+      "grad_norm": 0.5625,
+      "learning_rate": 0.00013139571991539238,
+      "loss": 0.3308,
+      "step": 2155
+    },
+    {
+      "epoch": 4.5908607863974495,
+      "grad_norm": 0.494140625,
+      "learning_rate": 0.00013104293395548098,
+      "loss": 0.3314,
+      "step": 2160
+    },
+    {
+      "epoch": 4.601487778958555,
+      "grad_norm": 0.45703125,
+      "learning_rate": 0.000130689719919174,
+      "loss": 0.3328,
+      "step": 2165
+    },
+    {
+      "epoch": 4.61211477151966,
+      "grad_norm": 0.453125,
+      "learning_rate": 0.00013033608267722858,
+      "loss": 0.3337,
+      "step": 2170
+    },
+    {
+      "epoch": 4.622741764080765,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.0001299820271062378,
+      "loss": 0.3343,
+      "step": 2175
+    },
+    {
+      "epoch": 4.63336875664187,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.00012962755808856342,
+      "loss": 0.3317,
+      "step": 2180
+    },
+    {
+      "epoch": 4.6439957492029755,
+      "grad_norm": 0.478515625,
+      "learning_rate": 0.00012927268051226844,
+      "loss": 0.3311,
+      "step": 2185
+    },
+    {
+      "epoch": 4.654622741764081,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.0001289173992710499,
+      "loss": 0.3347,
+      "step": 2190
+    },
+    {
+      "epoch": 4.665249734325186,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.00012856171926417133,
+      "loss": 0.3353,
+      "step": 2195
+    },
+    {
+      "epoch": 4.675876726886291,
+      "grad_norm": 0.59375,
+      "learning_rate": 0.00012820564539639512,
+      "loss": 0.3393,
+      "step": 2200
+    },
+    {
+      "epoch": 4.686503719447396,
+      "grad_norm": 0.73046875,
+      "learning_rate": 0.00012784918257791495,
+      "loss": 0.3373,
+      "step": 2205
+    },
+    {
+      "epoch": 4.6971307120085015,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.00012749233572428804,
+      "loss": 0.3271,
+      "step": 2210
+    },
+    {
+      "epoch": 4.707757704569607,
+      "grad_norm": 0.77734375,
+      "learning_rate": 0.00012713510975636741,
+      "loss": 0.3363,
+      "step": 2215
+    },
+    {
+      "epoch": 4.718384697130712,
+      "grad_norm": 0.67578125,
+      "learning_rate": 0.00012677750960023396,
+      "loss": 0.3355,
+      "step": 2220
+    },
+    {
+      "epoch": 4.729011689691817,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.00012641954018712863,
+      "loss": 0.3378,
+      "step": 2225
+    },
+    {
+      "epoch": 4.739638682252922,
+      "grad_norm": 0.484375,
+      "learning_rate": 0.0001260612064533843,
+      "loss": 0.3354,
+      "step": 2230
+    },
+    {
+      "epoch": 4.7502656748140275,
+      "grad_norm": 0.5,
+      "learning_rate": 0.0001257025133403577,
+      "loss": 0.3363,
+      "step": 2235
+    },
+    {
+      "epoch": 4.760892667375133,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.0001253434657943616,
+      "loss": 0.3296,
+      "step": 2240
+    },
+    {
+      "epoch": 4.771519659936238,
+      "grad_norm": 0.5546875,
+      "learning_rate": 0.00012498406876659598,
+      "loss": 0.3381,
+      "step": 2245
+    },
+    {
+      "epoch": 4.782146652497343,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.0001246243272130804,
+      "loss": 0.3222,
+      "step": 2250
+    },
+    {
+      "epoch": 4.792773645058449,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.00012426424609458518,
+      "loss": 0.3348,
+      "step": 2255
+    },
+    {
+      "epoch": 4.8034006376195535,
+      "grad_norm": 0.431640625,
+      "learning_rate": 0.00012390383037656327,
+      "loss": 0.3321,
+      "step": 2260
+    },
+    {
+      "epoch": 4.814027630180659,
+      "grad_norm": 0.43359375,
+      "learning_rate": 0.00012354308502908164,
+      "loss": 0.3349,
+      "step": 2265
+    },
+    {
+      "epoch": 4.824654622741764,
+      "grad_norm": 0.453125,
+      "learning_rate": 0.00012318201502675285,
+      "loss": 0.3269,
+      "step": 2270
+    },
+    {
+      "epoch": 4.835281615302869,
+      "grad_norm": 0.4453125,
+      "learning_rate": 0.00012282062534866632,
+      "loss": 0.3379,
+      "step": 2275
+    },
+    {
+      "epoch": 4.845908607863975,
+      "grad_norm": 0.423828125,
+      "learning_rate": 0.00012245892097831982,
+      "loss": 0.3316,
+      "step": 2280
+    },
+    {
+      "epoch": 4.8565356004250795,
+      "grad_norm": 0.4609375,
+      "learning_rate": 0.0001220969069035506,
+      "loss": 0.3411,
+      "step": 2285
+    },
+    {
+      "epoch": 4.867162592986185,
+      "grad_norm": 0.5546875,
+      "learning_rate": 0.0001217345881164667,
+      "loss": 0.3362,
+      "step": 2290
+    },
+    {
+      "epoch": 4.877789585547291,
+      "grad_norm": 0.46484375,
+      "learning_rate": 0.00012137196961337811,
+      "loss": 0.3316,
+      "step": 2295
+    },
+    {
+      "epoch": 4.888416578108395,
+      "grad_norm": 0.546875,
+      "learning_rate": 0.00012100905639472779,
+      "loss": 0.3309,
+      "step": 2300
+    },
+    {
+      "epoch": 4.899043570669501,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.00012064585346502286,
+      "loss": 0.3341,
+      "step": 2305
+    },
+    {
+      "epoch": 4.9096705632306055,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.00012028236583276542,
+      "loss": 0.3362,
+      "step": 2310
+    },
+    {
+      "epoch": 4.920297555791711,
+      "grad_norm": 0.734375,
+      "learning_rate": 0.0001199185985103836,
+      "loss": 0.3397,
+      "step": 2315
+    },
+    {
+      "epoch": 4.930924548352817,
+      "grad_norm": 0.49609375,
+      "learning_rate": 0.00011955455651416246,
+      "loss": 0.3363,
+      "step": 2320
+    },
+    {
+      "epoch": 4.941551540913921,
+      "grad_norm": 0.48828125,
+      "learning_rate": 0.00011919024486417463,
+      "loss": 0.3335,
+      "step": 2325
+    },
+    {
+      "epoch": 4.952178533475027,
+      "grad_norm": 0.5078125,
+      "learning_rate": 0.00011882566858421135,
+      "loss": 0.3311,
+      "step": 2330
+    },
+    {
+      "epoch": 4.9628055260361315,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.000118460832701713,
+      "loss": 0.3346,
+      "step": 2335
+    },
+    {
+      "epoch": 4.973432518597237,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.00011809574224769981,
+      "loss": 0.3334,
+      "step": 2340
+    },
+    {
+      "epoch": 4.984059511158343,
+      "grad_norm": 0.462890625,
+      "learning_rate": 0.00011773040225670256,
+      "loss": 0.3309,
+      "step": 2345
+    },
+    {
+      "epoch": 4.994686503719447,
+      "grad_norm": 0.490234375,
+      "learning_rate": 0.00011736481776669306,
+      "loss": 0.3345,
+      "step": 2350
+    },
+    {
+      "epoch": 4.99893730074389,
+      "eval_loss": 2.1233744621276855,
+      "eval_runtime": 0.8188,
+      "eval_samples_per_second": 6.106,
+      "eval_steps_per_second": 1.221,
+      "step": 2352
+    },
+    {
+      "epoch": 5.005313496280553,
+      "grad_norm": 0.435546875,
+      "learning_rate": 0.0001169989938190147,
+      "loss": 0.3151,
+      "step": 2355
+    },
+    {
+      "epoch": 5.015940488841657,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.00011663293545831302,
+      "loss": 0.2706,
+      "step": 2360
+    },
+    {
+      "epoch": 5.026567481402763,
+      "grad_norm": 0.5390625,
+      "learning_rate": 0.000116266647732466,
+      "loss": 0.276,
+      "step": 2365
+    },
+    {
+      "epoch": 5.0371944739638685,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.00011590013569251457,
+      "loss": 0.2787,
+      "step": 2370
+    },
+    {
+      "epoch": 5.047821466524973,
+      "grad_norm": 0.56640625,
+      "learning_rate": 0.00011553340439259286,
+      "loss": 0.2753,
+      "step": 2375
+    },
+    {
+      "epoch": 5.058448459086079,
+      "grad_norm": 0.466796875,
+      "learning_rate": 0.0001151664588898586,
+      "loss": 0.2707,
+      "step": 2380
+    },
+    {
+      "epoch": 5.069075451647183,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.00011479930424442335,
+      "loss": 0.2703,
+      "step": 2385
+    },
+    {
+      "epoch": 5.079702444208289,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.00011443194551928266,
+      "loss": 0.2715,
+      "step": 2390
+    },
+    {
+      "epoch": 5.0903294367693945,
+      "grad_norm": 0.6328125,
+      "learning_rate": 0.00011406438778024635,
+      "loss": 0.2739,
+      "step": 2395
+    },
+    {
+      "epoch": 5.100956429330499,
+      "grad_norm": 0.5625,
+      "learning_rate": 0.00011369663609586854,
+      "loss": 0.278,
+      "step": 2400
+    },
+    {
+      "epoch": 5.111583421891605,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.0001133286955373779,
+      "loss": 0.2809,
+      "step": 2405
+    },
+    {
+      "epoch": 5.12221041445271,
+      "grad_norm": 0.57421875,
+      "learning_rate": 0.00011296057117860759,
+      "loss": 0.2706,
+      "step": 2410
+    },
+    {
+      "epoch": 5.132837407013815,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.00011259226809592534,
+      "loss": 0.2765,
+      "step": 2415
+    },
+    {
+      "epoch": 5.1434643995749205,
+      "grad_norm": 0.46484375,
+      "learning_rate": 0.00011222379136816345,
+      "loss": 0.2717,
+      "step": 2420
+    },
+    {
+      "epoch": 5.154091392136025,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.00011185514607654881,
+      "loss": 0.2796,
+      "step": 2425
+    },
+    {
+      "epoch": 5.164718384697131,
+      "grad_norm": 0.51953125,
+      "learning_rate": 0.00011148633730463273,
+      "loss": 0.2801,
+      "step": 2430
+    },
+    {
+      "epoch": 5.175345377258236,
+      "grad_norm": 0.5078125,
+      "learning_rate": 0.00011111737013822088,
+      "loss": 0.2745,
+      "step": 2435
+    },
+    {
+      "epoch": 5.185972369819341,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.00011074824966530312,
+      "loss": 0.2786,
+      "step": 2440
+    },
+    {
+      "epoch": 5.1965993623804465,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.00011037898097598352,
+      "loss": 0.2728,
+      "step": 2445
+    },
+    {
+      "epoch": 5.207226354941551,
+      "grad_norm": 0.5625,
+      "learning_rate": 0.00011000956916240985,
+      "loss": 0.2827,
+      "step": 2450
+    },
+    {
+      "epoch": 5.217853347502657,
+      "grad_norm": 0.59765625,
+      "learning_rate": 0.00010964001931870365,
+      "loss": 0.2783,
+      "step": 2455
+    },
+    {
+      "epoch": 5.228480340063762,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.00010927033654088983,
+      "loss": 0.2847,
+      "step": 2460
+    },
+    {
+      "epoch": 5.239107332624867,
+      "grad_norm": 0.78125,
+      "learning_rate": 0.0001089005259268265,
+      "loss": 0.2833,
+      "step": 2465
+    },
+    {
+      "epoch": 5.2497343251859725,
+      "grad_norm": 0.6953125,
+      "learning_rate": 0.00010853059257613448,
+      "loss": 0.2819,
+      "step": 2470
+    },
+    {
+      "epoch": 5.260361317747078,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.00010816054159012723,
+      "loss": 0.2766,
+      "step": 2475
+    },
+    {
+      "epoch": 5.270988310308183,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.00010779037807174033,
+      "loss": 0.2767,
+      "step": 2480
+    },
+    {
+      "epoch": 5.281615302869288,
+      "grad_norm": 0.48828125,
+      "learning_rate": 0.00010742010712546116,
+      "loss": 0.2737,
+      "step": 2485
+    },
+    {
+      "epoch": 5.292242295430393,
+      "grad_norm": 0.47265625,
+      "learning_rate": 0.00010704973385725851,
+      "loss": 0.2799,
+      "step": 2490
+    },
+    {
+      "epoch": 5.3028692879914985,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.00010667926337451217,
+      "loss": 0.2815,
+      "step": 2495
+    },
+    {
+      "epoch": 5.313496280552604,
+      "grad_norm": 0.48046875,
+      "learning_rate": 0.00010630870078594249,
+      "loss": 0.2796,
+      "step": 2500
+    },
+    {
+      "epoch": 5.324123273113709,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.00010593805120154001,
+      "loss": 0.2764,
+      "step": 2505
+    },
+    {
+      "epoch": 5.334750265674814,
+      "grad_norm": 0.48828125,
+      "learning_rate": 0.00010556731973249485,
+      "loss": 0.279,
+      "step": 2510
+    },
+    {
+      "epoch": 5.345377258235919,
+      "grad_norm": 0.5625,
+      "learning_rate": 0.00010519651149112631,
+      "loss": 0.2817,
+      "step": 2515
+    },
+    {
+      "epoch": 5.3560042507970245,
+      "grad_norm": 0.578125,
+      "learning_rate": 0.00010482563159081238,
+      "loss": 0.283,
+      "step": 2520
+    },
+    {
+      "epoch": 5.36663124335813,
+      "grad_norm": 0.5390625,
+      "learning_rate": 0.00010445468514591925,
+      "loss": 0.2815,
+      "step": 2525
+    },
+    {
+      "epoch": 5.377258235919235,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.00010408367727173067,
+      "loss": 0.2793,
+      "step": 2530
+    },
+    {
+      "epoch": 5.38788522848034,
+      "grad_norm": 0.5078125,
+      "learning_rate": 0.0001037126130843776,
+      "loss": 0.2839,
+      "step": 2535
+    },
+    {
+      "epoch": 5.398512221041445,
+      "grad_norm": 0.46875,
+      "learning_rate": 0.00010334149770076747,
+      "loss": 0.2843,
+      "step": 2540
+    },
+    {
+      "epoch": 5.4091392136025505,
+      "grad_norm": 0.494140625,
+      "learning_rate": 0.00010297033623851369,
+      "loss": 0.2864,
+      "step": 2545
+    },
+    {
+      "epoch": 5.419766206163656,
+      "grad_norm": 0.609375,
+      "learning_rate": 0.0001025991338158651,
+      "loss": 0.2883,
+      "step": 2550
+    },
+    {
+      "epoch": 5.430393198724761,
+      "grad_norm": 0.609375,
+      "learning_rate": 0.0001022278955516354,
+      "loss": 0.2919,
+      "step": 2555
+    },
+    {
+      "epoch": 5.441020191285866,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.00010185662656513251,
+      "loss": 0.28,
+      "step": 2560
+    },
+    {
+      "epoch": 5.451647183846971,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.00010148533197608803,
+      "loss": 0.2815,
+      "step": 2565
+    },
+    {
+      "epoch": 5.462274176408076,
+      "grad_norm": 0.5546875,
+      "learning_rate": 0.00010111401690458654,
+      "loss": 0.2858,
+      "step": 2570
+    },
+    {
+      "epoch": 5.472901168969182,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.00010074268647099525,
+      "loss": 0.2824,
+      "step": 2575
+    },
+    {
+      "epoch": 5.483528161530287,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.00010037134579589302,
+      "loss": 0.2863,
+      "step": 2580
+    },
+    {
+      "epoch": 5.494155154091392,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.0001,
+      "loss": 0.2755,
+      "step": 2585
+    },
+    {
+      "epoch": 5.504782146652497,
+      "grad_norm": 0.59375,
+      "learning_rate": 9.962865420410701e-05,
+      "loss": 0.2812,
+      "step": 2590
+    },
+    {
+      "epoch": 5.515409139213602,
+      "grad_norm": 0.72265625,
+      "learning_rate": 9.925731352900478e-05,
+      "loss": 0.2781,
+      "step": 2595
+    },
+    {
+      "epoch": 5.526036131774708,
+      "grad_norm": 0.73828125,
+      "learning_rate": 9.888598309541347e-05,
+      "loss": 0.2864,
+      "step": 2600
+    },
+    {
+      "epoch": 5.536663124335813,
+      "grad_norm": 0.5234375,
+      "learning_rate": 9.851466802391201e-05,
+      "loss": 0.2813,
+      "step": 2605
+    },
+    {
+      "epoch": 5.547290116896918,
+      "grad_norm": 0.60546875,
+      "learning_rate": 9.814337343486754e-05,
+      "loss": 0.2801,
+      "step": 2610
+    },
+    {
+      "epoch": 5.557917109458024,
+      "grad_norm": 0.546875,
+      "learning_rate": 9.777210444836463e-05,
+      "loss": 0.281,
+      "step": 2615
+    },
+    {
+      "epoch": 5.568544102019128,
+      "grad_norm": 0.51953125,
+      "learning_rate": 9.740086618413495e-05,
+      "loss": 0.2817,
+      "step": 2620
+    },
+    {
+      "epoch": 5.579171094580234,
+      "grad_norm": 0.578125,
+      "learning_rate": 9.702966376148635e-05,
+      "loss": 0.2859,
+      "step": 2625
+    },
+    {
+      "epoch": 5.589798087141339,
+      "grad_norm": 0.52734375,
+      "learning_rate": 9.665850229923258e-05,
+      "loss": 0.2814,
+      "step": 2630
+    },
+    {
+      "epoch": 5.600425079702444,
+      "grad_norm": 0.609375,
+      "learning_rate": 9.62873869156224e-05,
+      "loss": 0.2827,
+      "step": 2635
+    },
+    {
+      "epoch": 5.61105207226355,
+      "grad_norm": 0.48046875,
+      "learning_rate": 9.591632272826934e-05,
+      "loss": 0.2831,
+      "step": 2640
+    },
+    {
+      "epoch": 5.621679064824654,
+      "grad_norm": 0.609375,
+      "learning_rate": 9.554531485408078e-05,
+      "loss": 0.2791,
+      "step": 2645
+    },
+    {
+      "epoch": 5.63230605738576,
+      "grad_norm": 0.51953125,
+      "learning_rate": 9.517436840918766e-05,
+      "loss": 0.2829,
+      "step": 2650
+    },
+    {
+      "epoch": 5.6429330499468655,
+      "grad_norm": 0.5390625,
+      "learning_rate": 9.480348850887373e-05,
+      "loss": 0.2826,
+      "step": 2655
+    },
+    {
+      "epoch": 5.65356004250797,
+      "grad_norm": 0.490234375,
+      "learning_rate": 9.44326802675052e-05,
+      "loss": 0.2827,
+      "step": 2660
+    },
+    {
+      "epoch": 5.664187035069076,
+      "grad_norm": 0.46875,
+      "learning_rate": 9.406194879846e-05,
+      "loss": 0.2857,
+      "step": 2665
+    },
+    {
+      "epoch": 5.67481402763018,
+      "grad_norm": 0.484375,
+      "learning_rate": 9.369129921405754e-05,
+      "loss": 0.2818,
+      "step": 2670
+    },
+    {
+      "epoch": 5.685441020191286,
+      "grad_norm": 0.45703125,
+      "learning_rate": 9.332073662548784e-05,
+      "loss": 0.278,
+      "step": 2675
+    },
+    {
+      "epoch": 5.6960680127523915,
+      "grad_norm": 0.5234375,
+      "learning_rate": 9.295026614274152e-05,
+      "loss": 0.2815,
+      "step": 2680
+    },
+    {
+      "epoch": 5.706695005313496,
+      "grad_norm": 0.5546875,
+      "learning_rate": 9.257989287453883e-05,
+      "loss": 0.2832,
+      "step": 2685
+    },
+    {
+      "epoch": 5.717321997874602,
+      "grad_norm": 0.59375,
+      "learning_rate": 9.220962192825968e-05,
+      "loss": 0.285,
+      "step": 2690
+    },
+    {
+      "epoch": 5.727948990435706,
+      "grad_norm": 0.45703125,
+      "learning_rate": 9.183945840987276e-05,
+      "loss": 0.283,
+      "step": 2695
+    },
+    {
+      "epoch": 5.738575982996812,
+      "grad_norm": 0.4921875,
+      "learning_rate": 9.146940742386553e-05,
+      "loss": 0.2869,
+      "step": 2700
+    },
+    {
+      "epoch": 5.7492029755579175,
+      "grad_norm": 0.466796875,
+      "learning_rate": 9.109947407317352e-05,
+      "loss": 0.2876,
+      "step": 2705
+    },
+    {
+      "epoch": 5.759829968119022,
+      "grad_norm": 0.51953125,
+      "learning_rate": 9.072966345911019e-05,
+      "loss": 0.2879,
+      "step": 2710
+    },
+    {
+      "epoch": 5.770456960680128,
+      "grad_norm": 0.5,
+      "learning_rate": 9.035998068129636e-05,
+      "loss": 0.2773,
+      "step": 2715
+    },
+    {
+      "epoch": 5.781083953241232,
+      "grad_norm": 0.5390625,
+      "learning_rate": 8.999043083759017e-05,
+      "loss": 0.2863,
+      "step": 2720
+    },
+    {
+      "epoch": 5.791710945802338,
+      "grad_norm": 0.47265625,
+      "learning_rate": 8.962101902401648e-05,
+      "loss": 0.2813,
+      "step": 2725
+    },
+    {
+      "epoch": 5.8023379383634435,
+      "grad_norm": 0.54296875,
+      "learning_rate": 8.925175033469688e-05,
+      "loss": 0.2772,
+      "step": 2730
+    },
+    {
+      "epoch": 5.812964930924548,
+      "grad_norm": 0.5234375,
+      "learning_rate": 8.888262986177913e-05,
+      "loss": 0.2842,
+      "step": 2735
+    },
+    {
+      "epoch": 5.823591923485654,
+      "grad_norm": 0.53125,
+      "learning_rate": 8.851366269536729e-05,
+      "loss": 0.2806,
+      "step": 2740
+    },
+    {
+      "epoch": 5.834218916046758,
+      "grad_norm": 0.51953125,
+      "learning_rate": 8.814485392345118e-05,
+      "loss": 0.2754,
+      "step": 2745
+    },
+    {
+      "epoch": 5.844845908607864,
+      "grad_norm": 0.482421875,
+      "learning_rate": 8.777620863183657e-05,
+      "loss": 0.2852,
+      "step": 2750
+    },
+    {
+      "epoch": 5.8554729011689695,
+      "grad_norm": 0.50390625,
+      "learning_rate": 8.740773190407471e-05,
+      "loss": 0.2782,
+      "step": 2755
+    },
+    {
+      "epoch": 5.866099893730074,
+      "grad_norm": 0.486328125,
+      "learning_rate": 8.703942882139245e-05,
+      "loss": 0.2887,
+      "step": 2760
+    },
+    {
+      "epoch": 5.87672688629118,
+      "grad_norm": 0.46875,
+      "learning_rate": 8.667130446262214e-05,
+      "loss": 0.2812,
+      "step": 2765
+    },
+    {
+      "epoch": 5.887353878852284,
+      "grad_norm": 0.53515625,
+      "learning_rate": 8.630336390413147e-05,
+      "loss": 0.2809,
+      "step": 2770
+    },
+    {
+      "epoch": 5.89798087141339,
+      "grad_norm": 0.498046875,
+      "learning_rate": 8.59356122197537e-05,
+      "loss": 0.2816,
+      "step": 2775
+    },
+    {
+      "epoch": 5.9086078639744954,
+      "grad_norm": 0.59375,
+      "learning_rate": 8.556805448071735e-05,
+      "loss": 0.2788,
+      "step": 2780
+    },
+    {
+      "epoch": 5.9192348565356,
+      "grad_norm": 0.515625,
+      "learning_rate": 8.520069575557667e-05,
+      "loss": 0.2773,
+      "step": 2785
+    },
+    {
+      "epoch": 5.929861849096706,
+      "grad_norm": 0.53125,
+      "learning_rate": 8.483354111014141e-05,
+      "loss": 0.2882,
+      "step": 2790
+    },
+    {
+      "epoch": 5.940488841657811,
+      "grad_norm": 0.50390625,
+      "learning_rate": 8.446659560740717e-05,
+      "loss": 0.285,
+      "step": 2795
+    },
+    {
+      "epoch": 5.951115834218916,
+      "grad_norm": 0.515625,
+      "learning_rate": 8.409986430748545e-05,
+      "loss": 0.2788,
+      "step": 2800
+    },
+    {
+      "epoch": 5.961742826780021,
+      "grad_norm": 0.50390625,
+      "learning_rate": 8.373335226753404e-05,
+      "loss": 0.2799,
+      "step": 2805
+    },
+    {
+      "epoch": 5.972369819341126,
+      "grad_norm": 0.478515625,
+      "learning_rate": 8.336706454168701e-05,
+      "loss": 0.2835,
+      "step": 2810
+    },
+    {
+      "epoch": 5.982996811902232,
+      "grad_norm": 0.5234375,
+      "learning_rate": 8.300100618098534e-05,
+      "loss": 0.2865,
+      "step": 2815
+    },
+    {
+      "epoch": 5.993623804463337,
+      "grad_norm": 0.53515625,
+      "learning_rate": 8.263518223330697e-05,
+      "loss": 0.2875,
+      "step": 2820
+    },
+    {
+      "epoch": 6.0,
+      "eval_loss": 2.5845518112182617,
+      "eval_runtime": 0.7933,
+      "eval_samples_per_second": 6.303,
+      "eval_steps_per_second": 1.261,
+      "step": 2823
+    },
+    {
+      "epoch": 6.004250797024442,
+      "grad_norm": 0.3984375,
+      "learning_rate": 8.226959774329747e-05,
+      "loss": 0.2603,
+      "step": 2825
+    },
+    {
+      "epoch": 6.014877789585547,
+      "grad_norm": 0.578125,
+      "learning_rate": 8.190425775230021e-05,
+      "loss": 0.2316,
+      "step": 2830
+    },
+    {
+      "epoch": 6.025504782146652,
+      "grad_norm": 0.470703125,
+      "learning_rate": 8.153916729828703e-05,
+      "loss": 0.2244,
+      "step": 2835
+    },
+    {
+      "epoch": 6.036131774707758,
+      "grad_norm": 0.470703125,
+      "learning_rate": 8.117433141578866e-05,
+      "loss": 0.2276,
+      "step": 2840
+    },
+    {
+      "epoch": 6.046758767268863,
+      "grad_norm": 0.484375,
+      "learning_rate": 8.080975513582539e-05,
+      "loss": 0.2338,
+      "step": 2845
+    },
+    {
+      "epoch": 6.057385759829968,
+      "grad_norm": 0.51171875,
+      "learning_rate": 8.044544348583755e-05,
+      "loss": 0.2334,
+      "step": 2850
+    },
+    {
+      "epoch": 6.068012752391073,
+      "grad_norm": 0.51953125,
+      "learning_rate": 8.008140148961641e-05,
+      "loss": 0.2319,
+      "step": 2855
+    },
+    {
+      "epoch": 6.078639744952179,
+      "grad_norm": 0.53125,
+      "learning_rate": 7.971763416723459e-05,
+      "loss": 0.2326,
+      "step": 2860
+    },
+    {
+      "epoch": 6.089266737513284,
+      "grad_norm": 0.494140625,
+      "learning_rate": 7.935414653497715e-05,
+      "loss": 0.2305,
+      "step": 2865
+    },
+    {
+      "epoch": 6.099893730074389,
+      "grad_norm": 0.46484375,
+      "learning_rate": 7.89909436052722e-05,
+      "loss": 0.2297,
+      "step": 2870
+    },
+    {
+      "epoch": 6.110520722635494,
+      "grad_norm": 0.482421875,
+      "learning_rate": 7.862803038662191e-05,
+      "loss": 0.2251,
+      "step": 2875
+    },
+    {
+      "epoch": 6.121147715196599,
+      "grad_norm": 0.515625,
+      "learning_rate": 7.826541188353329e-05,
+      "loss": 0.2321,
+      "step": 2880
+    },
+    {
+      "epoch": 6.131774707757705,
+      "grad_norm": 0.52734375,
+      "learning_rate": 7.790309309644942e-05,
+      "loss": 0.2324,
+      "step": 2885
+    },
+    {
+      "epoch": 6.14240170031881,
+      "grad_norm": 0.5078125,
+      "learning_rate": 7.754107902168019e-05,
+      "loss": 0.2305,
+      "step": 2890
+    },
+    {
+      "epoch": 6.153028692879915,
+      "grad_norm": 0.52734375,
+      "learning_rate": 7.717937465133371e-05,
+      "loss": 0.2258,
+      "step": 2895
+    },
+    {
+      "epoch": 6.16365568544102,
+      "grad_norm": 0.5078125,
+      "learning_rate": 7.681798497324716e-05,
+      "loss": 0.2375,
+      "step": 2900
+    },
+    {
+      "epoch": 6.174282678002125,
+      "grad_norm": 0.50390625,
+      "learning_rate": 7.645691497091838e-05,
+      "loss": 0.2351,
+      "step": 2905
+    },
+    {
+      "epoch": 6.184909670563231,
+      "grad_norm": 0.52734375,
+      "learning_rate": 7.609616962343675e-05,
+      "loss": 0.2359,
+      "step": 2910
+    },
+    {
+      "epoch": 6.195536663124336,
+      "grad_norm": 0.5234375,
+      "learning_rate": 7.573575390541485e-05,
+      "loss": 0.2396,
+      "step": 2915
+    },
+    {
+      "epoch": 6.206163655685441,
+      "grad_norm": 0.5390625,
+      "learning_rate": 7.537567278691964e-05,
+      "loss": 0.2316,
+      "step": 2920
+    },
+    {
+      "epoch": 6.216790648246546,
+      "grad_norm": 0.486328125,
+      "learning_rate": 7.501593123340403e-05,
+      "loss": 0.2356,
+      "step": 2925
+    },
+    {
+      "epoch": 6.227417640807651,
+      "grad_norm": 0.52734375,
+      "learning_rate": 7.465653420563845e-05,
+      "loss": 0.2373,
+      "step": 2930
+    },
+    {
+      "epoch": 6.238044633368757,
+      "grad_norm": 0.53515625,
+      "learning_rate": 7.42974866596423e-05,
+      "loss": 0.2349,
+      "step": 2935
+    },
+    {
+      "epoch": 6.248671625929862,
+      "grad_norm": 0.5625,
+      "learning_rate": 7.393879354661577e-05,
+      "loss": 0.2326,
+      "step": 2940
+    },
+    {
+      "epoch": 6.259298618490967,
+      "grad_norm": 0.60546875,
+      "learning_rate": 7.358045981287141e-05,
+      "loss": 0.2366,
+      "step": 2945
+    },
+    {
+      "epoch": 6.269925611052072,
+      "grad_norm": 0.5390625,
+      "learning_rate": 7.322249039976608e-05,
+      "loss": 0.2372,
+      "step": 2950
+    },
+    {
+      "epoch": 6.280552603613177,
+      "grad_norm": 0.4765625,
+      "learning_rate": 7.286489024363261e-05,
+      "loss": 0.2319,
+      "step": 2955
+    },
+    {
+      "epoch": 6.291179596174283,
+      "grad_norm": 0.546875,
+      "learning_rate": 7.2507664275712e-05,
+      "loss": 0.2341,
+      "step": 2960
+    },
+    {
+      "epoch": 6.301806588735388,
+      "grad_norm": 0.5703125,
+      "learning_rate": 7.215081742208508e-05,
+      "loss": 0.2334,
+      "step": 2965
+    },
+    {
+      "epoch": 6.312433581296493,
+      "grad_norm": 0.515625,
+      "learning_rate": 7.179435460360491e-05,
+      "loss": 0.2305,
+      "step": 2970
+    },
+    {
+      "epoch": 6.323060573857599,
+      "grad_norm": 0.61328125,
+      "learning_rate": 7.14382807358287e-05,
+      "loss": 0.2351,
+      "step": 2975
+    },
+    {
+      "epoch": 6.333687566418703,
+      "grad_norm": 0.51953125,
+      "learning_rate": 7.108260072895013e-05,
+      "loss": 0.2312,
+      "step": 2980
+    },
+    {
+      "epoch": 6.344314558979809,
+      "grad_norm": 0.52734375,
+      "learning_rate": 7.072731948773159e-05,
+      "loss": 0.2309,
+      "step": 2985
+    },
+    {
+      "epoch": 6.354941551540914,
+      "grad_norm": 0.54296875,
+      "learning_rate": 7.037244191143661e-05,
+      "loss": 0.2392,
+      "step": 2990
+    },
+    {
+      "epoch": 6.365568544102019,
+      "grad_norm": 0.56640625,
+      "learning_rate": 7.001797289376218e-05,
+      "loss": 0.2363,
+      "step": 2995
+    },
+    {
+      "epoch": 6.376195536663125,
+      "grad_norm": 0.55859375,
+      "learning_rate": 6.966391732277143e-05,
+      "loss": 0.2342,
+      "step": 3000
+    },
+    {
+      "epoch": 6.386822529224229,
+      "grad_norm": 0.55859375,
+      "learning_rate": 6.931028008082602e-05,
+      "loss": 0.2329,
+      "step": 3005
+    },
+    {
+      "epoch": 6.397449521785335,
+      "grad_norm": 0.55078125,
+      "learning_rate": 6.895706604451905e-05,
+      "loss": 0.2377,
+      "step": 3010
+    },
+    {
+      "epoch": 6.4080765143464395,
+      "grad_norm": 0.53515625,
+      "learning_rate": 6.860428008460762e-05,
+      "loss": 0.2402,
+      "step": 3015
+    },
+    {
+      "epoch": 6.418703506907545,
+      "grad_norm": 0.5703125,
+      "learning_rate": 6.825192706594575e-05,
+      "loss": 0.2366,
+      "step": 3020
+    },
+    {
+      "epoch": 6.429330499468651,
+      "grad_norm": 0.6015625,
+      "learning_rate": 6.790001184741728e-05,
+      "loss": 0.2331,
+      "step": 3025
+    },
+    {
+      "epoch": 6.439957492029755,
+      "grad_norm": 0.56640625,
+      "learning_rate": 6.75485392818689e-05,
+      "loss": 0.2337,
+      "step": 3030
+    },
+    {
+      "epoch": 6.450584484590861,
+      "grad_norm": 0.53515625,
+      "learning_rate": 6.719751421604309e-05,
+      "loss": 0.2357,
+      "step": 3035
+    },
+    {
+      "epoch": 6.461211477151966,
+      "grad_norm": 0.5,
+      "learning_rate": 6.684694149051156e-05,
+      "loss": 0.2395,
+      "step": 3040
+    },
+    {
+      "epoch": 6.471838469713071,
+      "grad_norm": 0.54296875,
+      "learning_rate": 6.649682593960814e-05,
+      "loss": 0.2258,
+      "step": 3045
+    },
+    {
+      "epoch": 6.482465462274177,
+      "grad_norm": 0.46484375,
+      "learning_rate": 6.614717239136246e-05,
+      "loss": 0.234,
+      "step": 3050
+    },
+    {
+      "epoch": 6.493092454835281,
+      "grad_norm": 0.5234375,
+      "learning_rate": 6.579798566743314e-05,
+      "loss": 0.2417,
+      "step": 3055
+    },
+    {
+      "epoch": 6.503719447396387,
+      "grad_norm": 0.66796875,
+      "learning_rate": 6.54492705830414e-05,
+      "loss": 0.2319,
+      "step": 3060
+    },
+    {
+      "epoch": 6.514346439957492,
+      "grad_norm": 0.58203125,
+      "learning_rate": 6.51010319469046e-05,
+      "loss": 0.2325,
+      "step": 3065
+    },
+    {
+      "epoch": 6.524973432518597,
+      "grad_norm": 0.578125,
+      "learning_rate": 6.475327456117005e-05,
+      "loss": 0.2353,
+      "step": 3070
+    },
+    {
+      "epoch": 6.535600425079703,
+      "grad_norm": 0.490234375,
+      "learning_rate": 6.440600322134864e-05,
+      "loss": 0.2345,
+      "step": 3075
+    },
+    {
+      "epoch": 6.546227417640807,
+      "grad_norm": 0.52734375,
+      "learning_rate": 6.405922271624874e-05,
+      "loss": 0.2327,
+      "step": 3080
+    },
+    {
+      "epoch": 6.556854410201913,
+      "grad_norm": 0.54296875,
+      "learning_rate": 6.371293782791028e-05,
+      "loss": 0.2356,
+      "step": 3085
+    },
+    {
+      "epoch": 6.567481402763018,
+      "grad_norm": 0.49609375,
+      "learning_rate": 6.336715333153869e-05,
+      "loss": 0.2336,
+      "step": 3090
+    },
+    {
+      "epoch": 6.578108395324123,
+      "grad_norm": 0.51953125,
+      "learning_rate": 6.302187399543911e-05,
+      "loss": 0.2365,
+      "step": 3095
+    },
+    {
+      "epoch": 6.588735387885229,
+      "grad_norm": 0.46875,
+      "learning_rate": 6.267710458095053e-05,
+      "loss": 0.2377,
+      "step": 3100
+    },
+    {
+      "epoch": 6.599362380446333,
+      "grad_norm": 0.53515625,
+      "learning_rate": 6.233284984238035e-05,
+      "loss": 0.2297,
+      "step": 3105
+    },
+    {
+      "epoch": 6.609989373007439,
+      "grad_norm": 0.53515625,
+      "learning_rate": 6.198911452693853e-05,
+      "loss": 0.2347,
+      "step": 3110
+    },
+    {
+      "epoch": 6.620616365568544,
+      "grad_norm": 0.51953125,
+      "learning_rate": 6.164590337467243e-05,
+      "loss": 0.2363,
+      "step": 3115
+    },
+    {
+      "epoch": 6.631243358129649,
+      "grad_norm": 0.5078125,
+      "learning_rate": 6.130322111840114e-05,
+      "loss": 0.2339,
+      "step": 3120
+    },
+    {
+      "epoch": 6.641870350690755,
+      "grad_norm": 0.57421875,
+      "learning_rate": 6.0961072483650526e-05,
+      "loss": 0.2328,
+      "step": 3125
+    },
+    {
+      "epoch": 6.652497343251859,
+      "grad_norm": 0.578125,
+      "learning_rate": 6.0619462188587793e-05,
+      "loss": 0.234,
+      "step": 3130
+    },
+    {
+      "epoch": 6.663124335812965,
+      "grad_norm": 0.5390625,
+      "learning_rate": 6.027839494395664e-05,
+      "loss": 0.2397,
+      "step": 3135
+    },
+    {
+      "epoch": 6.67375132837407,
+      "grad_norm": 0.50390625,
+      "learning_rate": 5.993787545301204e-05,
+      "loss": 0.2379,
+      "step": 3140
+    },
+    {
+      "epoch": 6.684378320935175,
+      "grad_norm": 0.50390625,
+      "learning_rate": 5.959790841145577e-05,
+      "loss": 0.236,
+      "step": 3145
+    },
+    {
+      "epoch": 6.695005313496281,
+      "grad_norm": 0.53125,
+      "learning_rate": 5.9258498507371194e-05,
+      "loss": 0.2393,
+      "step": 3150
+    },
+    {
+      "epoch": 6.705632306057386,
+      "grad_norm": 0.50390625,
+      "learning_rate": 5.891965042115907e-05,
+      "loss": 0.2399,
+      "step": 3155
+    },
+    {
+      "epoch": 6.716259298618491,
+      "grad_norm": 0.59765625,
+      "learning_rate": 5.8581368825472585e-05,
+      "loss": 0.2261,
+      "step": 3160
+    },
+    {
+      "epoch": 6.726886291179596,
+      "grad_norm": 0.51953125,
+      "learning_rate": 5.8243658385153285e-05,
+      "loss": 0.2379,
+      "step": 3165
+    },
+    {
+      "epoch": 6.737513283740701,
+      "grad_norm": 0.55859375,
+      "learning_rate": 5.790652375716652e-05,
+      "loss": 0.2327,
+      "step": 3170
+    },
+    {
+      "epoch": 6.748140276301807,
+      "grad_norm": 0.58984375,
+      "learning_rate": 5.7569969590537284e-05,
+      "loss": 0.2299,
+      "step": 3175
+    },
+    {
+      "epoch": 6.758767268862912,
+      "grad_norm": 0.490234375,
+      "learning_rate": 5.7234000526286156e-05,
+      "loss": 0.2373,
+      "step": 3180
+    },
+    {
+      "epoch": 6.769394261424017,
+      "grad_norm": 0.55859375,
+      "learning_rate": 5.689862119736522e-05,
+      "loss": 0.2406,
+      "step": 3185
+    },
+    {
+      "epoch": 6.780021253985122,
+      "grad_norm": 0.55078125,
+      "learning_rate": 5.656383622859418e-05,
+      "loss": 0.2394,
+      "step": 3190
+    },
+    {
+      "epoch": 6.790648246546228,
+      "grad_norm": 0.5625,
+      "learning_rate": 5.622965023659683e-05,
+      "loss": 0.2378,
+      "step": 3195
+    },
+    {
+      "epoch": 6.801275239107333,
+      "grad_norm": 0.5703125,
+      "learning_rate": 5.589606782973683e-05,
+      "loss": 0.2342,
+      "step": 3200
+    },
+    {
+      "epoch": 6.811902231668438,
+      "grad_norm": 0.51171875,
+      "learning_rate": 5.55630936080549e-05,
+      "loss": 0.2344,
+      "step": 3205
+    },
+    {
+      "epoch": 6.822529224229543,
+      "grad_norm": 0.53125,
+      "learning_rate": 5.5230732163204615e-05,
+      "loss": 0.2344,
+      "step": 3210
+    },
+    {
+      "epoch": 6.833156216790648,
+      "grad_norm": 0.5390625,
+      "learning_rate": 5.48989880783898e-05,
+      "loss": 0.2352,
+      "step": 3215
+    },
+    {
+      "epoch": 6.843783209351754,
+      "grad_norm": 0.51953125,
+      "learning_rate": 5.456786592830083e-05,
+      "loss": 0.2337,
+      "step": 3220
+    },
+    {
+      "epoch": 6.8544102019128585,
+      "grad_norm": 0.50390625,
+      "learning_rate": 5.423737027905173e-05,
+      "loss": 0.2326,
+      "step": 3225
+    },
+    {
+      "epoch": 6.865037194473964,
+      "grad_norm": 0.50390625,
+      "learning_rate": 5.39075056881172e-05,
+      "loss": 0.2348,
+      "step": 3230
+    },
+    {
+      "epoch": 6.875664187035069,
+      "grad_norm": 0.54296875,
+      "learning_rate": 5.357827670426977e-05,
+      "loss": 0.2361,
+      "step": 3235
+    },
+    {
+      "epoch": 6.886291179596174,
+      "grad_norm": 0.4921875,
+      "learning_rate": 5.3249687867517095e-05,
+      "loss": 0.2298,
+      "step": 3240
+    },
+    {
+      "epoch": 6.89691817215728,
+      "grad_norm": 0.53515625,
+      "learning_rate": 5.292174370903919e-05,
+      "loss": 0.2344,
+      "step": 3245
+    },
+    {
+      "epoch": 6.9075451647183845,
+      "grad_norm": 0.51953125,
+      "learning_rate": 5.259444875112624e-05,
+      "loss": 0.2366,
+      "step": 3250
+    },
+    {
+      "epoch": 6.91817215727949,
+      "grad_norm": 0.8203125,
+      "learning_rate": 5.226780750711602e-05,
+      "loss": 0.2343,
+      "step": 3255
+    },
+    {
+      "epoch": 6.928799149840595,
+      "grad_norm": 0.53125,
+      "learning_rate": 5.1941824481331626e-05,
+      "loss": 0.236,
+      "step": 3260
+    },
+    {
+      "epoch": 6.9394261424017,
+      "grad_norm": 0.498046875,
+      "learning_rate": 5.1616504169019564e-05,
+      "loss": 0.2271,
+      "step": 3265
+    },
+    {
+      "epoch": 6.950053134962806,
+      "grad_norm": 0.50390625,
+      "learning_rate": 5.129185105628756e-05,
+      "loss": 0.2338,
+      "step": 3270
+    },
+    {
+      "epoch": 6.9606801275239105,
+      "grad_norm": 0.515625,
+      "learning_rate": 5.0967869620042794e-05,
+      "loss": 0.2311,
+      "step": 3275
+    },
+    {
+      "epoch": 6.971307120085016,
+      "grad_norm": 0.51171875,
+      "learning_rate": 5.064456432793019e-05,
+      "loss": 0.23,
+      "step": 3280
+    },
+    {
+      "epoch": 6.981934112646121,
+      "grad_norm": 0.59765625,
+      "learning_rate": 5.032193963827073e-05,
+      "loss": 0.2347,
+      "step": 3285
+    },
+    {
+      "epoch": 6.992561105207226,
+      "grad_norm": 0.56640625,
+      "learning_rate": 5.000000000000002e-05,
+      "loss": 0.2319,
+      "step": 3290
+    },
+    {
+      "epoch": 6.99893730074389,
+      "eval_loss": 3.1057090759277344,
+      "eval_runtime": 0.816,
+      "eval_samples_per_second": 6.127,
+      "eval_steps_per_second": 1.225,
+      "step": 3293
+    },
+    {
+      "epoch": 7.003188097768332,
+      "grad_norm": 0.40234375,
+      "learning_rate": 4.96787498526069e-05,
+      "loss": 0.2177,
+      "step": 3295
+    },
+    {
+      "epoch": 7.0138150903294365,
+      "grad_norm": 0.458984375,
+      "learning_rate": 4.93581936260724e-05,
+      "loss": 0.1956,
+      "step": 3300
+    },
+    {
+      "epoch": 7.024442082890542,
+      "grad_norm": 0.56640625,
+      "learning_rate": 4.903833574080825e-05,
+      "loss": 0.1954,
+      "step": 3305
+    },
+    {
+      "epoch": 7.035069075451648,
+      "grad_norm": 0.494140625,
+      "learning_rate": 4.8719180607596484e-05,
+      "loss": 0.1939,
+      "step": 3310
+    },
+    {
+      "epoch": 7.045696068012752,
+      "grad_norm": 0.4453125,
+      "learning_rate": 4.8400732627528e-05,
+      "loss": 0.1996,
+      "step": 3315
+    },
+    {
+      "epoch": 7.056323060573858,
+      "grad_norm": 0.55859375,
+      "learning_rate": 4.808299619194251e-05,
+      "loss": 0.1946,
+      "step": 3320
+    },
+    {
+      "epoch": 7.0669500531349625,
+      "grad_norm": 0.54296875,
+      "learning_rate": 4.776597568236731e-05,
+      "loss": 0.1914,
+      "step": 3325
+    },
+    {
+      "epoch": 7.077577045696068,
+      "grad_norm": 0.4609375,
+      "learning_rate": 4.744967547045754e-05,
+      "loss": 0.1963,
+      "step": 3330
+    },
+    {
+      "epoch": 7.088204038257174,
+      "grad_norm": 0.4921875,
+      "learning_rate": 4.713409991793536e-05,
+      "loss": 0.2024,
+      "step": 3335
+    },
+    {
+      "epoch": 7.098831030818278,
+      "grad_norm": 0.5625,
+      "learning_rate": 4.681925337653006e-05,
+      "loss": 0.1974,
+      "step": 3340
+    },
+    {
+      "epoch": 7.109458023379384,
+      "grad_norm": 0.53515625,
+      "learning_rate": 4.650514018791799e-05,
+      "loss": 0.1954,
+      "step": 3345
+    },
+    {
+      "epoch": 7.1200850159404885,
+      "grad_norm": 0.53125,
+      "learning_rate": 4.6191764683662744e-05,
+      "loss": 0.2007,
+      "step": 3350
+    },
+    {
+      "epoch": 7.130712008501594,
+      "grad_norm": 0.478515625,
+      "learning_rate": 4.587913118515532e-05,
+      "loss": 0.1959,
+      "step": 3355
+    },
+    {
+      "epoch": 7.1413390010627,
+      "grad_norm": 0.5078125,
+      "learning_rate": 4.5567244003554645e-05,
+      "loss": 0.1994,
+      "step": 3360
+    },
+    {
+      "epoch": 7.151965993623804,
+      "grad_norm": 0.51953125,
+      "learning_rate": 4.5256107439728035e-05,
+      "loss": 0.1979,
+      "step": 3365
+    },
+    {
+      "epoch": 7.16259298618491,
+      "grad_norm": 0.53125,
+      "learning_rate": 4.494572578419194e-05,
+      "loss": 0.1951,
+      "step": 3370
+    },
+    {
+      "epoch": 7.1732199787460145,
+      "grad_norm": 0.474609375,
+      "learning_rate": 4.463610331705273e-05,
+      "loss": 0.1996,
+      "step": 3375
+    },
+    {
+      "epoch": 7.18384697130712,
+      "grad_norm": 0.474609375,
+      "learning_rate": 4.432724430794786e-05,
+      "loss": 0.1989,
+      "step": 3380
+    },
+    {
+      "epoch": 7.194473963868226,
+      "grad_norm": 0.51953125,
+      "learning_rate": 4.4019153015986546e-05,
+      "loss": 0.2009,
+      "step": 3385
+    },
+    {
+      "epoch": 7.20510095642933,
+      "grad_norm": 0.53125,
+      "learning_rate": 4.371183368969165e-05,
+      "loss": 0.1926,
+      "step": 3390
+    },
+    {
+      "epoch": 7.215727948990436,
+      "grad_norm": 0.51171875,
+      "learning_rate": 4.340529056694047e-05,
+      "loss": 0.1997,
+      "step": 3395
+    },
+    {
+      "epoch": 7.226354941551541,
+      "grad_norm": 0.4765625,
+      "learning_rate": 4.309952787490689e-05,
+      "loss": 0.1969,
+      "step": 3400
+    },
+    {
+      "epoch": 7.236981934112646,
+      "grad_norm": 0.48046875,
+      "learning_rate": 4.279454983000251e-05,
+      "loss": 0.1974,
+      "step": 3405
+    },
+    {
+      "epoch": 7.247608926673752,
+      "grad_norm": 0.63671875,
+      "learning_rate": 4.249036063781896e-05,
+      "loss": 0.1969,
+      "step": 3410
+    },
+    {
+      "epoch": 7.258235919234856,
+      "grad_norm": 0.5234375,
+      "learning_rate": 4.2186964493069794e-05,
+      "loss": 0.1953,
+      "step": 3415
+    },
+    {
+      "epoch": 7.268862911795962,
+      "grad_norm": 0.54296875,
+      "learning_rate": 4.1884365579532346e-05,
+      "loss": 0.1973,
+      "step": 3420
+    },
+    {
+      "epoch": 7.279489904357067,
+      "grad_norm": 0.62890625,
+      "learning_rate": 4.158256806999059e-05,
+      "loss": 0.1938,
+      "step": 3425
+    },
+    {
+      "epoch": 7.290116896918172,
+      "grad_norm": 0.51953125,
+      "learning_rate": 4.128157612617696e-05,
+      "loss": 0.1964,
+      "step": 3430
+    },
+    {
+      "epoch": 7.3007438894792775,
+      "grad_norm": 0.53125,
+      "learning_rate": 4.0981393898715545e-05,
+      "loss": 0.202,
+      "step": 3435
+    },
+    {
+      "epoch": 7.311370882040382,
+      "grad_norm": 0.494140625,
+      "learning_rate": 4.0682025527064486e-05,
+      "loss": 0.1964,
+      "step": 3440
+    },
+    {
+      "epoch": 7.321997874601488,
+      "grad_norm": 0.53515625,
+      "learning_rate": 4.038347513945898e-05,
+      "loss": 0.1999,
+      "step": 3445
+    },
+    {
+      "epoch": 7.332624867162593,
+      "grad_norm": 0.4921875,
+      "learning_rate": 4.008574685285442e-05,
+      "loss": 0.1938,
+      "step": 3450
+    },
+    {
+      "epoch": 7.343251859723698,
+      "grad_norm": 0.94921875,
+      "learning_rate": 3.978884477286956e-05,
+      "loss": 0.1941,
+      "step": 3455
+    },
+    {
+      "epoch": 7.3538788522848035,
+      "grad_norm": 0.5859375,
+      "learning_rate": 3.94927729937299e-05,
+      "loss": 0.1996,
+      "step": 3460
+    },
+    {
+      "epoch": 7.364505844845908,
+      "grad_norm": 0.578125,
+      "learning_rate": 3.9197535598211274e-05,
+      "loss": 0.2,
+      "step": 3465
+    },
+    {
+      "epoch": 7.375132837407014,
+      "grad_norm": 0.50390625,
+      "learning_rate": 3.890313665758348e-05,
+      "loss": 0.1937,
+      "step": 3470
+    },
+    {
+      "epoch": 7.385759829968119,
+      "grad_norm": 0.5390625,
+      "learning_rate": 3.860958023155422e-05,
+      "loss": 0.1991,
+      "step": 3475
+    },
+    {
+      "epoch": 7.396386822529224,
+      "grad_norm": 0.5078125,
+      "learning_rate": 3.8316870368213e-05,
+      "loss": 0.1979,
+      "step": 3480
+    },
+    {
+      "epoch": 7.4070138150903295,
+      "grad_norm": 0.546875,
+      "learning_rate": 3.802501110397553e-05,
+      "loss": 0.2021,
+      "step": 3485
+    },
+    {
+      "epoch": 7.417640807651434,
+      "grad_norm": 0.54296875,
+      "learning_rate": 3.773400646352769e-05,
+      "loss": 0.2034,
+      "step": 3490
+    },
+    {
+      "epoch": 7.42826780021254,
+      "grad_norm": 0.50390625,
+      "learning_rate": 3.744386045977052e-05,
+      "loss": 0.2033,
+      "step": 3495
+    },
+    {
+      "epoch": 7.438894792773645,
+      "grad_norm": 0.5,
+      "learning_rate": 3.7154577093764334e-05,
+      "loss": 0.2012,
+      "step": 3500
+    },
+    {
+      "epoch": 7.44952178533475,
+      "grad_norm": 0.53515625,
+      "learning_rate": 3.686616035467408e-05,
+      "loss": 0.1952,
+      "step": 3505
+    },
+    {
+      "epoch": 7.4601487778958555,
+      "grad_norm": 0.57421875,
+      "learning_rate": 3.657861421971388e-05,
+      "loss": 0.2011,
+      "step": 3510
+    },
+    {
+      "epoch": 7.470775770456961,
+      "grad_norm": 0.490234375,
+      "learning_rate": 3.629194265409255e-05,
+      "loss": 0.2033,
+      "step": 3515
+    },
+    {
+      "epoch": 7.481402763018066,
+      "grad_norm": 0.546875,
+      "learning_rate": 3.6006149610958625e-05,
+      "loss": 0.2003,
+      "step": 3520
+    },
+    {
+      "epoch": 7.492029755579171,
+      "grad_norm": 0.56640625,
+      "learning_rate": 3.5721239031346066e-05,
+      "loss": 0.1988,
+      "step": 3525
+    },
+    {
+      "epoch": 7.502656748140276,
+      "grad_norm": 0.484375,
+      "learning_rate": 3.543721484411976e-05,
+      "loss": 0.1903,
+      "step": 3530
+    },
+    {
+      "epoch": 7.5132837407013815,
+      "grad_norm": 0.53125,
+      "learning_rate": 3.515408096592144e-05,
+      "loss": 0.1991,
+      "step": 3535
+    },
+    {
+      "epoch": 7.523910733262487,
+      "grad_norm": 0.49609375,
+      "learning_rate": 3.487184130111562e-05,
+      "loss": 0.1967,
+      "step": 3540
+    },
+    {
+      "epoch": 7.534537725823592,
+      "grad_norm": 0.54296875,
+      "learning_rate": 3.459049974173579e-05,
+      "loss": 0.1983,
+      "step": 3545
+    },
+    {
+      "epoch": 7.545164718384697,
+      "grad_norm": 0.53125,
+      "learning_rate": 3.4310060167430725e-05,
+      "loss": 0.1998,
+      "step": 3550
+    },
+    {
+      "epoch": 7.555791710945803,
+      "grad_norm": 0.5390625,
+      "learning_rate": 3.4030526445410984e-05,
+      "loss": 0.198,
+      "step": 3555
+    },
+    {
+      "epoch": 7.5664187035069075,
+      "grad_norm": 0.49609375,
+      "learning_rate": 3.375190243039556e-05,
+      "loss": 0.1961,
+      "step": 3560
+    },
+    {
+      "epoch": 7.577045696068013,
+      "grad_norm": 0.478515625,
+      "learning_rate": 3.3474191964558885e-05,
+      "loss": 0.1974,
+      "step": 3565
+    },
+    {
+      "epoch": 7.587672688629118,
+      "grad_norm": 0.5234375,
+      "learning_rate": 3.319739887747752e-05,
+      "loss": 0.195,
+      "step": 3570
+    },
+    {
+      "epoch": 7.598299681190223,
+      "grad_norm": 0.51171875,
+      "learning_rate": 3.292152698607768e-05,
+      "loss": 0.1992,
+      "step": 3575
+    },
+    {
+      "epoch": 7.608926673751329,
+      "grad_norm": 0.50390625,
+      "learning_rate": 3.264658009458239e-05,
+      "loss": 0.1996,
+      "step": 3580
+    },
+    {
+      "epoch": 7.6195536663124335,
+      "grad_norm": 0.5,
+      "learning_rate": 3.2372561994459136e-05,
+      "loss": 0.1966,
+      "step": 3585
+    },
+    {
+      "epoch": 7.630180658873539,
+      "grad_norm": 0.5078125,
+      "learning_rate": 3.209947646436752e-05,
+      "loss": 0.1941,
+      "step": 3590
+    },
+    {
+      "epoch": 7.640807651434644,
+      "grad_norm": 0.54296875,
+      "learning_rate": 3.182732727010715e-05,
+      "loss": 0.1959,
+      "step": 3595
+    },
+    {
+      "epoch": 7.651434643995749,
+      "grad_norm": 0.53515625,
+      "learning_rate": 3.155611816456586e-05,
+      "loss": 0.2002,
+      "step": 3600
+    },
+    {
+      "epoch": 7.662061636556855,
+      "grad_norm": 0.51171875,
+      "learning_rate": 3.12858528876676e-05,
+      "loss": 0.196,
+      "step": 3605
+    },
+    {
+      "epoch": 7.6726886291179595,
+      "grad_norm": 0.53125,
+      "learning_rate": 3.1016535166321356e-05,
+      "loss": 0.1988,
+      "step": 3610
+    },
+    {
+      "epoch": 7.683315621679065,
+      "grad_norm": 0.51171875,
+      "learning_rate": 3.074816871436924e-05,
+      "loss": 0.2016,
+      "step": 3615
+    },
+    {
+      "epoch": 7.69394261424017,
+      "grad_norm": 0.54296875,
+      "learning_rate": 3.0480757232535772e-05,
+      "loss": 0.1973,
+      "step": 3620
+    },
+    {
+      "epoch": 7.704569606801275,
+      "grad_norm": 0.5234375,
+      "learning_rate": 3.021430440837646e-05,
+      "loss": 0.1945,
+      "step": 3625
+    },
+    {
+      "epoch": 7.715196599362381,
+      "grad_norm": 0.5234375,
+      "learning_rate": 2.9948813916227115e-05,
+      "loss": 0.1985,
+      "step": 3630
+    },
+    {
+      "epoch": 7.7258235919234854,
+      "grad_norm": 0.5859375,
+      "learning_rate": 2.968428941715321e-05,
+      "loss": 0.1963,
+      "step": 3635
+    },
+    {
+      "epoch": 7.736450584484591,
+      "grad_norm": 0.50390625,
+      "learning_rate": 2.9420734558899322e-05,
+      "loss": 0.2008,
+      "step": 3640
+    },
+    {
+      "epoch": 7.747077577045696,
+      "grad_norm": 0.5234375,
+      "learning_rate": 2.915815297583886e-05,
+      "loss": 0.1983,
+      "step": 3645
+    },
+    {
+      "epoch": 7.757704569606801,
+      "grad_norm": 0.51953125,
+      "learning_rate": 2.889654828892393e-05,
+      "loss": 0.2,
+      "step": 3650
+    },
+    {
+      "epoch": 7.768331562167907,
+      "grad_norm": 0.515625,
+      "learning_rate": 2.8635924105635458e-05,
+      "loss": 0.1978,
+      "step": 3655
+    },
+    {
+      "epoch": 7.778958554729011,
+      "grad_norm": 0.546875,
+      "learning_rate": 2.8376284019933373e-05,
+      "loss": 0.1944,
+      "step": 3660
+    },
+    {
+      "epoch": 7.789585547290117,
+      "grad_norm": 0.474609375,
+      "learning_rate": 2.8117631612207084e-05,
+      "loss": 0.1963,
+      "step": 3665
+    },
+    {
+      "epoch": 7.800212539851222,
+      "grad_norm": 0.53515625,
+      "learning_rate": 2.7859970449226104e-05,
+      "loss": 0.1961,
+      "step": 3670
+    },
+    {
+      "epoch": 7.810839532412327,
+      "grad_norm": 0.54296875,
+      "learning_rate": 2.760330408409083e-05,
+      "loss": 0.1974,
+      "step": 3675
+    },
+    {
+      "epoch": 7.821466524973433,
+      "grad_norm": 0.51171875,
+      "learning_rate": 2.73476360561837e-05,
+      "loss": 0.1977,
+      "step": 3680
+    },
+    {
+      "epoch": 7.832093517534537,
+      "grad_norm": 0.55078125,
+      "learning_rate": 2.7092969891120058e-05,
+      "loss": 0.1956,
+      "step": 3685
+    },
+    {
+      "epoch": 7.842720510095643,
+      "grad_norm": 0.51171875,
+      "learning_rate": 2.6839309100699973e-05,
+      "loss": 0.1905,
+      "step": 3690
+    },
+    {
+      "epoch": 7.8533475026567485,
+      "grad_norm": 0.5234375,
+      "learning_rate": 2.6586657182859343e-05,
+      "loss": 0.1984,
+      "step": 3695
+    },
+    {
+      "epoch": 7.863974495217853,
+      "grad_norm": 0.5625,
+      "learning_rate": 2.6335017621622116e-05,
+      "loss": 0.1949,
+      "step": 3700
+    },
+    {
+      "epoch": 7.874601487778959,
+      "grad_norm": 0.5703125,
+      "learning_rate": 2.6084393887051884e-05,
+      "loss": 0.196,
+      "step": 3705
+    },
+    {
+      "epoch": 7.885228480340063,
+      "grad_norm": 0.5234375,
+      "learning_rate": 2.5834789435204243e-05,
+      "loss": 0.1922,
+      "step": 3710
+    },
+    {
+      "epoch": 7.895855472901169,
+      "grad_norm": 0.478515625,
+      "learning_rate": 2.5586207708079034e-05,
+      "loss": 0.1921,
+      "step": 3715
+    },
+    {
+      "epoch": 7.9064824654622745,
+      "grad_norm": 0.55078125,
+      "learning_rate": 2.5338652133572915e-05,
+      "loss": 0.1969,
+      "step": 3720
+    },
+    {
+      "epoch": 7.917109458023379,
+      "grad_norm": 0.54296875,
+      "learning_rate": 2.5092126125432136e-05,
+      "loss": 0.2008,
+      "step": 3725
+    },
+    {
+      "epoch": 7.927736450584485,
+      "grad_norm": 0.54296875,
+      "learning_rate": 2.4846633083205263e-05,
+      "loss": 0.1963,
+      "step": 3730
+    },
+    {
+      "epoch": 7.93836344314559,
+      "grad_norm": 0.49609375,
+      "learning_rate": 2.4602176392196662e-05,
+      "loss": 0.1944,
+      "step": 3735
+    },
+    {
+      "epoch": 7.948990435706695,
+      "grad_norm": 0.51953125,
+      "learning_rate": 2.4358759423419474e-05,
+      "loss": 0.1954,
+      "step": 3740
+    },
+    {
+      "epoch": 7.9596174282678005,
+      "grad_norm": 0.5,
+      "learning_rate": 2.411638553354928e-05,
+      "loss": 0.1979,
+      "step": 3745
+    },
+    {
+      "epoch": 7.970244420828905,
+      "grad_norm": 0.50390625,
+      "learning_rate": 2.3875058064877807e-05,
+      "loss": 0.195,
+      "step": 3750
+    },
+    {
+      "epoch": 7.980871413390011,
+      "grad_norm": 0.515625,
+      "learning_rate": 2.3634780345266806e-05,
+      "loss": 0.1936,
+      "step": 3755
+    },
+    {
+      "epoch": 7.991498405951116,
+      "grad_norm": 0.5,
+      "learning_rate": 2.339555568810221e-05,
+      "loss": 0.1968,
+      "step": 3760
+    },
+    {
+      "epoch": 8.0,
+      "eval_loss": 3.6609108448028564,
+      "eval_runtime": 0.7997,
+      "eval_samples_per_second": 6.252,
+      "eval_steps_per_second": 1.25,
+      "step": 3764
+    },
+    {
+      "epoch": 8.002125398512222,
+      "grad_norm": 0.431640625,
+      "learning_rate": 2.3157387392248385e-05,
+      "loss": 0.1921,
+      "step": 3765
+    },
+    {
+      "epoch": 8.012752391073326,
+      "grad_norm": 0.41015625,
+      "learning_rate": 2.2920278742002676e-05,
+      "loss": 0.1828,
+      "step": 3770
+    },
+    {
+      "epoch": 8.023379383634431,
+      "grad_norm": 0.453125,
+      "learning_rate": 2.2684233007050115e-05,
+      "loss": 0.1811,
+      "step": 3775
+    },
+    {
+      "epoch": 8.034006376195537,
+      "grad_norm": 0.515625,
+      "learning_rate": 2.244925344241828e-05,
+      "loss": 0.1809,
+      "step": 3780
+    },
+    {
+      "epoch": 8.044633368756642,
+      "grad_norm": 0.4765625,
+      "learning_rate": 2.221534328843261e-05,
+      "loss": 0.1758,
+      "step": 3785
+    },
+    {
+      "epoch": 8.055260361317748,
+      "grad_norm": 0.51171875,
+      "learning_rate": 2.1982505770671303e-05,
+      "loss": 0.1764,
+      "step": 3790
+    },
+    {
+      "epoch": 8.065887353878852,
+      "grad_norm": 0.51953125,
+      "learning_rate": 2.1750744099921396e-05,
+      "loss": 0.1779,
+      "step": 3795
+    },
+    {
+      "epoch": 8.076514346439957,
+      "grad_norm": 0.478515625,
+      "learning_rate": 2.1520061472133902e-05,
+      "loss": 0.1798,
+      "step": 3800
+    },
+    {
+      "epoch": 8.087141339001063,
+      "grad_norm": 0.51953125,
+      "learning_rate": 2.1290461068380298e-05,
+      "loss": 0.1756,
+      "step": 3805
+    },
+    {
+      "epoch": 8.097768331562168,
+      "grad_norm": 0.4921875,
+      "learning_rate": 2.1061946054808146e-05,
+      "loss": 0.1804,
+      "step": 3810
+    },
+    {
+      "epoch": 8.108395324123274,
+      "grad_norm": 0.51953125,
+      "learning_rate": 2.0834519582597878e-05,
+      "loss": 0.1782,
+      "step": 3815
+    },
+    {
+      "epoch": 8.119022316684378,
+      "grad_norm": 0.490234375,
+      "learning_rate": 2.0608184787919026e-05,
+      "loss": 0.1804,
+      "step": 3820
+    },
+    {
+      "epoch": 8.129649309245483,
+      "grad_norm": 0.51171875,
+      "learning_rate": 2.0382944791887106e-05,
+      "loss": 0.1795,
+      "step": 3825
+    },
+    {
+      "epoch": 8.140276301806589,
+      "grad_norm": 0.490234375,
+      "learning_rate": 2.0158802700520574e-05,
+      "loss": 0.1815,
+      "step": 3830
+    },
+    {
+      "epoch": 8.150903294367694,
+      "grad_norm": 0.4921875,
+      "learning_rate": 1.9935761604697976e-05,
+      "loss": 0.1793,
+      "step": 3835
+    },
+    {
+      "epoch": 8.1615302869288,
+      "grad_norm": 0.4765625,
+      "learning_rate": 1.9713824580115335e-05,
+      "loss": 0.1785,
+      "step": 3840
+    },
+    {
+      "epoch": 8.172157279489904,
+      "grad_norm": 0.486328125,
+      "learning_rate": 1.9492994687243714e-05,
+      "loss": 0.1784,
+      "step": 3845
+    },
+    {
+      "epoch": 8.182784272051009,
+      "grad_norm": 0.48046875,
+      "learning_rate": 1.927327497128706e-05,
+      "loss": 0.177,
+      "step": 3850
+    },
+    {
+      "epoch": 8.193411264612115,
+      "grad_norm": 0.4921875,
+      "learning_rate": 1.9054668462140168e-05,
+      "loss": 0.1804,
+      "step": 3855
+    },
+    {
+      "epoch": 8.20403825717322,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.883717817434688e-05,
+      "loss": 0.1753,
+      "step": 3860
+    },
+    {
+      "epoch": 8.214665249734326,
+      "grad_norm": 0.484375,
+      "learning_rate": 1.8620807107058668e-05,
+      "loss": 0.1785,
+      "step": 3865
+    },
+    {
+      "epoch": 8.22529224229543,
+      "grad_norm": 0.46875,
+      "learning_rate": 1.840555824399296e-05,
+      "loss": 0.1775,
+      "step": 3870
+    },
+    {
+      "epoch": 8.235919234856535,
+      "grad_norm": 0.578125,
+      "learning_rate": 1.8191434553392428e-05,
+      "loss": 0.18,
+      "step": 3875
+    },
+    {
+      "epoch": 8.24654622741764,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.797843898798358e-05,
+      "loss": 0.1806,
+      "step": 3880
+    },
+    {
+      "epoch": 8.257173219978746,
+      "grad_norm": 0.51171875,
+      "learning_rate": 1.7766574484936482e-05,
+      "loss": 0.1837,
+      "step": 3885
+    },
+    {
+      "epoch": 8.267800212539852,
+      "grad_norm": 0.494140625,
+      "learning_rate": 1.7555843965823992e-05,
+      "loss": 0.1817,
+      "step": 3890
+    },
+    {
+      "epoch": 8.278427205100957,
+      "grad_norm": 0.5390625,
+      "learning_rate": 1.7346250336581427e-05,
+      "loss": 0.1806,
+      "step": 3895
+    },
+    {
+      "epoch": 8.289054197662061,
+      "grad_norm": 0.486328125,
+      "learning_rate": 1.7137796487466797e-05,
+      "loss": 0.1791,
+      "step": 3900
+    },
+    {
+      "epoch": 8.299681190223167,
+      "grad_norm": 0.498046875,
+      "learning_rate": 1.693048529302056e-05,
+      "loss": 0.1788,
+      "step": 3905
+    },
+    {
+      "epoch": 8.310308182784272,
+      "grad_norm": 0.46484375,
+      "learning_rate": 1.672431961202635e-05,
+      "loss": 0.1809,
+      "step": 3910
+    },
+    {
+      "epoch": 8.320935175345378,
+      "grad_norm": 0.484375,
+      "learning_rate": 1.6519302287471207e-05,
+      "loss": 0.1774,
+      "step": 3915
+    },
+    {
+      "epoch": 8.331562167906483,
+      "grad_norm": 0.5,
+      "learning_rate": 1.6315436146506703e-05,
+      "loss": 0.1737,
+      "step": 3920
+    },
+    {
+      "epoch": 8.342189160467587,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.6112724000409694e-05,
+      "loss": 0.1813,
+      "step": 3925
+    },
+    {
+      "epoch": 8.352816153028693,
+      "grad_norm": 0.4921875,
+      "learning_rate": 1.5911168644543707e-05,
+      "loss": 0.175,
+      "step": 3930
+    },
+    {
+      "epoch": 8.363443145589798,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.57107728583203e-05,
+      "loss": 0.1838,
+      "step": 3935
+    },
+    {
+      "epoch": 8.374070138150904,
+      "grad_norm": 0.51953125,
+      "learning_rate": 1.5511539405160825e-05,
+      "loss": 0.1783,
+      "step": 3940
+    },
+    {
+      "epoch": 8.38469713071201,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.5313471032458247e-05,
+      "loss": 0.1763,
+      "step": 3945
+    },
+    {
+      "epoch": 8.395324123273113,
+      "grad_norm": 0.51953125,
+      "learning_rate": 1.5116570471539293e-05,
+      "loss": 0.1758,
+      "step": 3950
+    },
+    {
+      "epoch": 8.405951115834219,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.4920840437626759e-05,
+      "loss": 0.1829,
+      "step": 3955
+    },
+    {
+      "epoch": 8.416578108395324,
+      "grad_norm": 0.53125,
+      "learning_rate": 1.4726283629802107e-05,
+      "loss": 0.1779,
+      "step": 3960
+    },
+    {
+      "epoch": 8.42720510095643,
+      "grad_norm": 0.53125,
+      "learning_rate": 1.4532902730968212e-05,
+      "loss": 0.1792,
+      "step": 3965
+    },
+    {
+      "epoch": 8.437832093517535,
+      "grad_norm": 0.48828125,
+      "learning_rate": 1.4340700407812435e-05,
+      "loss": 0.1813,
+      "step": 3970
+    },
+    {
+      "epoch": 8.448459086078639,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.414967931076967e-05,
+      "loss": 0.1785,
+      "step": 3975
+    },
+    {
+      "epoch": 8.459086078639745,
+      "grad_norm": 0.5625,
+      "learning_rate": 1.3959842073986085e-05,
+      "loss": 0.1746,
+      "step": 3980
+    },
+    {
+      "epoch": 8.46971307120085,
+      "grad_norm": 0.62109375,
+      "learning_rate": 1.3771191315282427e-05,
+      "loss": 0.1832,
+      "step": 3985
+    },
+    {
+      "epoch": 8.480340063761956,
+      "grad_norm": 0.490234375,
+      "learning_rate": 1.3583729636118358e-05,
+      "loss": 0.1843,
+      "step": 3990
+    },
+    {
+      "epoch": 8.490967056323061,
+      "grad_norm": 0.515625,
+      "learning_rate": 1.339745962155613e-05,
+      "loss": 0.1801,
+      "step": 3995
+    },
+    {
+      "epoch": 8.501594048884165,
+      "grad_norm": 0.54296875,
+      "learning_rate": 1.3212383840225329e-05,
+      "loss": 0.18,
+      "step": 4000
+    },
+    {
+      "epoch": 8.51222104144527,
+      "grad_norm": 0.55078125,
+      "learning_rate": 1.3028504844287182e-05,
+      "loss": 0.18,
+      "step": 4005
+    },
+    {
+      "epoch": 8.522848034006376,
+      "grad_norm": 0.51171875,
+      "learning_rate": 1.2845825169399507e-05,
+      "loss": 0.1778,
+      "step": 4010
+    },
+    {
+      "epoch": 8.533475026567482,
+      "grad_norm": 0.46875,
+      "learning_rate": 1.2664347334681681e-05,
+      "loss": 0.1849,
+      "step": 4015
+    },
+    {
+      "epoch": 8.544102019128587,
+      "grad_norm": 0.46875,
+      "learning_rate": 1.2484073842679944e-05,
+      "loss": 0.1799,
+      "step": 4020
+    },
+    {
+      "epoch": 8.554729011689691,
+      "grad_norm": 0.478515625,
+      "learning_rate": 1.230500717933285e-05,
+      "loss": 0.1802,
+      "step": 4025
+    },
+    {
+      "epoch": 8.565356004250797,
+      "grad_norm": 0.51953125,
+      "learning_rate": 1.2127149813937022e-05,
+      "loss": 0.1776,
+      "step": 4030
+    },
+    {
+      "epoch": 8.575982996811902,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.1950504199113088e-05,
+      "loss": 0.1794,
+      "step": 4035
+    },
+    {
+      "epoch": 8.586609989373008,
+      "grad_norm": 0.478515625,
+      "learning_rate": 1.1775072770771834e-05,
+      "loss": 0.1757,
+      "step": 4040
+    },
+    {
+      "epoch": 8.597236981934113,
+      "grad_norm": 0.5078125,
+      "learning_rate": 1.1600857948080624e-05,
+      "loss": 0.1786,
+      "step": 4045
+    },
+    {
+      "epoch": 8.607863974495217,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.1427862133430156e-05,
+      "loss": 0.1838,
+      "step": 4050
+    },
+    {
+      "epoch": 8.618490967056323,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.1256087712401087e-05,
+      "loss": 0.1755,
+      "step": 4055
+    },
+    {
+      "epoch": 8.629117959617428,
+      "grad_norm": 0.478515625,
+      "learning_rate": 1.1085537053731354e-05,
+      "loss": 0.1777,
+      "step": 4060
+    },
+    {
+      "epoch": 8.639744952178534,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.091621250928343e-05,
+      "loss": 0.1799,
+      "step": 4065
+    },
+    {
+      "epoch": 8.65037194473964,
+      "grad_norm": 0.51953125,
+      "learning_rate": 1.0748116414011888e-05,
+      "loss": 0.1774,
+      "step": 4070
+    },
+    {
+      "epoch": 8.660998937300743,
+      "grad_norm": 0.5,
+      "learning_rate": 1.058125108593122e-05,
+      "loss": 0.1844,
+      "step": 4075
+    },
+    {
+      "epoch": 8.671625929861849,
+      "grad_norm": 0.50390625,
+      "learning_rate": 1.0415618826083828e-05,
+      "loss": 0.1803,
+      "step": 4080
+    },
+    {
+      "epoch": 8.682252922422954,
+      "grad_norm": 0.546875,
+      "learning_rate": 1.0251221918508425e-05,
+      "loss": 0.1787,
+      "step": 4085
+    },
+    {
+      "epoch": 8.69287991498406,
+      "grad_norm": 0.5546875,
+      "learning_rate": 1.0088062630208273e-05,
+      "loss": 0.1804,
+      "step": 4090
+    },
+    {
+      "epoch": 8.703506907545165,
+      "grad_norm": 0.50390625,
+      "learning_rate": 9.926143211120286e-06,
+      "loss": 0.1868,
+      "step": 4095
+    },
+    {
+      "epoch": 8.71413390010627,
+      "grad_norm": 0.5,
+      "learning_rate": 9.765465894083636e-06,
+      "loss": 0.1831,
+      "step": 4100
+    },
+    {
+      "epoch": 8.724760892667375,
+      "grad_norm": 0.458984375,
+      "learning_rate": 9.606032894809258e-06,
+      "loss": 0.1774,
+      "step": 4105
+    },
+    {
+      "epoch": 8.73538788522848,
+      "grad_norm": 0.515625,
+      "learning_rate": 9.447846411849115e-06,
+      "loss": 0.1775,
+      "step": 4110
+    },
+    {
+      "epoch": 8.746014877789586,
+      "grad_norm": 0.51953125,
+      "learning_rate": 9.29090862656593e-06,
+      "loss": 0.1796,
+      "step": 4115
+    },
+    {
+      "epoch": 8.756641870350691,
+      "grad_norm": 0.5,
+      "learning_rate": 9.135221703103136e-06,
+      "loss": 0.1798,
+      "step": 4120
+    },
+    {
+      "epoch": 8.767268862911797,
+      "grad_norm": 0.5,
+      "learning_rate": 8.980787788355016e-06,
+      "loss": 0.179,
+      "step": 4125
+    },
+    {
+      "epoch": 8.7778958554729,
+      "grad_norm": 0.46484375,
+      "learning_rate": 8.827609011937066e-06,
+      "loss": 0.1802,
+      "step": 4130
+    },
+    {
+      "epoch": 8.788522848034006,
+      "grad_norm": 0.51171875,
+      "learning_rate": 8.675687486156659e-06,
+      "loss": 0.184,
+      "step": 4135
+    },
+    {
+      "epoch": 8.799149840595112,
+      "grad_norm": 0.60546875,
+      "learning_rate": 8.525025305983936e-06,
+      "loss": 0.173,
+      "step": 4140
+    },
+    {
+      "epoch": 8.809776833156217,
+      "grad_norm": 0.53125,
+      "learning_rate": 8.375624549022854e-06,
+      "loss": 0.179,
+      "step": 4145
+    },
+    {
+      "epoch": 8.820403825717323,
+      "grad_norm": 0.56640625,
+      "learning_rate": 8.227487275482592e-06,
+      "loss": 0.1812,
+      "step": 4150
+    },
+    {
+      "epoch": 8.831030818278427,
+      "grad_norm": 0.494140625,
+      "learning_rate": 8.08061552814917e-06,
+      "loss": 0.183,
+      "step": 4155
+    },
+    {
+      "epoch": 8.841657810839532,
+      "grad_norm": 0.51953125,
+      "learning_rate": 7.935011332357112e-06,
+      "loss": 0.1786,
+      "step": 4160
+    },
+    {
+      "epoch": 8.852284803400638,
+      "grad_norm": 0.474609375,
+      "learning_rate": 7.790676695961796e-06,
+      "loss": 0.178,
+      "step": 4165
+    },
+    {
+      "epoch": 8.862911795961743,
+      "grad_norm": 0.5,
+      "learning_rate": 7.647613609311455e-06,
+      "loss": 0.1809,
+      "step": 4170
+    },
+    {
+      "epoch": 8.873538788522849,
+      "grad_norm": 0.55078125,
+      "learning_rate": 7.505824045220011e-06,
+      "loss": 0.1756,
+      "step": 4175
+    },
+    {
+      "epoch": 8.884165781083952,
+      "grad_norm": 0.5390625,
+      "learning_rate": 7.365309958939615e-06,
+      "loss": 0.175,
+      "step": 4180
+    },
+    {
+      "epoch": 8.894792773645058,
+      "grad_norm": 0.5078125,
+      "learning_rate": 7.226073288133939e-06,
+      "loss": 0.1792,
+      "step": 4185
+    },
+    {
+      "epoch": 8.905419766206164,
+      "grad_norm": 0.486328125,
+      "learning_rate": 7.088115952851238e-06,
+      "loss": 0.1733,
+      "step": 4190
+    },
+    {
+      "epoch": 8.91604675876727,
+      "grad_norm": 0.49609375,
+      "learning_rate": 6.951439855498021e-06,
+      "loss": 0.1792,
+      "step": 4195
+    },
+    {
+      "epoch": 8.926673751328375,
+      "grad_norm": 0.498046875,
+      "learning_rate": 6.81604688081271e-06,
+      "loss": 0.1803,
+      "step": 4200
+    },
+    {
+      "epoch": 8.937300743889478,
+      "grad_norm": 0.45703125,
+      "learning_rate": 6.681938895839746e-06,
+      "loss": 0.177,
+      "step": 4205
+    },
+    {
+      "epoch": 8.947927736450584,
+      "grad_norm": 0.486328125,
+      "learning_rate": 6.549117749903755e-06,
+      "loss": 0.1791,
+      "step": 4210
+    },
+    {
+      "epoch": 8.95855472901169,
+      "grad_norm": 0.498046875,
+      "learning_rate": 6.417585274584115e-06,
+      "loss": 0.1778,
+      "step": 4215
+    },
+    {
+      "epoch": 8.969181721572795,
+      "grad_norm": 0.494140625,
+      "learning_rate": 6.287343283689661e-06,
+      "loss": 0.176,
+      "step": 4220
+    },
+    {
+      "epoch": 8.9798087141339,
+      "grad_norm": 0.5234375,
+      "learning_rate": 6.158393573233678e-06,
+      "loss": 0.1759,
+      "step": 4225
+    },
+    {
+      "epoch": 8.990435706695006,
+      "grad_norm": 0.5234375,
+      "learning_rate": 6.030737921409169e-06,
+      "loss": 0.1809,
+      "step": 4230
+    },
+    {
+      "epoch": 8.99893730074389,
+      "eval_loss": 3.939990520477295,
+      "eval_runtime": 0.8191,
+      "eval_samples_per_second": 6.105,
+      "eval_steps_per_second": 1.221,
+      "step": 4234
+    },
+    {
+      "epoch": 9.00106269925611,
+      "grad_norm": 0.47265625,
+      "learning_rate": 5.904378088564255e-06,
+      "loss": 0.1819,
+      "step": 4235
+    },
+    {
+      "epoch": 9.011689691817216,
+      "grad_norm": 0.5,
+      "learning_rate": 5.779315817178e-06,
+      "loss": 0.1763,
+      "step": 4240
+    },
+    {
+      "epoch": 9.022316684378321,
+      "grad_norm": 0.4921875,
+      "learning_rate": 5.655552831836297e-06,
+      "loss": 0.1767,
+      "step": 4245
+    },
+    {
+      "epoch": 9.032943676939427,
+      "grad_norm": 0.46875,
+      "learning_rate": 5.533090839208133e-06,
+      "loss": 0.1744,
+      "step": 4250
+    },
+    {
+      "epoch": 9.043570669500532,
+      "grad_norm": 0.490234375,
+      "learning_rate": 5.4119315280220405e-06,
+      "loss": 0.1777,
+      "step": 4255
+    },
+    {
+      "epoch": 9.054197662061636,
+      "grad_norm": 0.44921875,
+      "learning_rate": 5.292076569042826e-06,
+      "loss": 0.1744,
+      "step": 4260
+    },
+    {
+      "epoch": 9.064824654622742,
+      "grad_norm": 0.451171875,
+      "learning_rate": 5.1735276150484905e-06,
+      "loss": 0.1792,
+      "step": 4265
+    },
+    {
+      "epoch": 9.075451647183847,
+      "grad_norm": 0.50390625,
+      "learning_rate": 5.056286300807511e-06,
+      "loss": 0.1751,
+      "step": 4270
+    },
+    {
+      "epoch": 9.086078639744953,
+      "grad_norm": 0.47265625,
+      "learning_rate": 4.940354243056178e-06,
+      "loss": 0.174,
+      "step": 4275
+    },
+    {
+      "epoch": 9.096705632306058,
+      "grad_norm": 0.458984375,
+      "learning_rate": 4.825733040476465e-06,
+      "loss": 0.1732,
+      "step": 4280
+    },
+    {
+      "epoch": 9.107332624867162,
+      "grad_norm": 0.51953125,
+      "learning_rate": 4.712424273673788e-06,
+      "loss": 0.1767,
+      "step": 4285
+    },
+    {
+      "epoch": 9.117959617428268,
+      "grad_norm": 0.44921875,
+      "learning_rate": 4.600429505155424e-06,
+      "loss": 0.1726,
+      "step": 4290
+    },
+    {
+      "epoch": 9.128586609989373,
+      "grad_norm": 0.4765625,
+      "learning_rate": 4.489750279308757e-06,
+      "loss": 0.174,
+      "step": 4295
+    },
+    {
+      "epoch": 9.139213602550479,
+      "grad_norm": 0.484375,
+      "learning_rate": 4.380388122380141e-06,
+      "loss": 0.1787,
+      "step": 4300
+    },
+    {
+      "epoch": 9.149840595111584,
+      "grad_norm": 0.515625,
+      "learning_rate": 4.2723445424537544e-06,
+      "loss": 0.1802,
+      "step": 4305
+    },
+    {
+      "epoch": 9.160467587672688,
+      "grad_norm": 0.5,
+      "learning_rate": 4.165621029430855e-06,
+      "loss": 0.1786,
+      "step": 4310
+    },
+    {
+      "epoch": 9.171094580233794,
+      "grad_norm": 0.462890625,
+      "learning_rate": 4.060219055009229e-06,
+      "loss": 0.1752,
+      "step": 4315
+    },
+    {
+      "epoch": 9.181721572794899,
+      "grad_norm": 0.48828125,
+      "learning_rate": 3.9561400726628505e-06,
+      "loss": 0.1793,
+      "step": 4320
+    },
+    {
+      "epoch": 9.192348565356005,
+      "grad_norm": 0.56640625,
+      "learning_rate": 3.85338551762191e-06,
+      "loss": 0.175,
+      "step": 4325
+    },
+    {
+      "epoch": 9.20297555791711,
+      "grad_norm": 0.462890625,
+      "learning_rate": 3.7519568068529855e-06,
+      "loss": 0.1728,
+      "step": 4330
+    },
+    {
+      "epoch": 9.213602550478214,
+      "grad_norm": 0.478515625,
+      "learning_rate": 3.651855339039456e-06,
+      "loss": 0.1698,
+      "step": 4335
+    },
+    {
+      "epoch": 9.22422954303932,
+      "grad_norm": 0.69921875,
+      "learning_rate": 3.5530824945623542e-06,
+      "loss": 0.1765,
+      "step": 4340
+    },
+    {
+      "epoch": 9.234856535600425,
+      "grad_norm": 0.4765625,
+      "learning_rate": 3.4556396354811226e-06,
+      "loss": 0.1762,
+      "step": 4345
+    },
+    {
+      "epoch": 9.24548352816153,
+      "grad_norm": 0.490234375,
+      "learning_rate": 3.359528105515064e-06,
+      "loss": 0.1781,
+      "step": 4350
+    },
+    {
+      "epoch": 9.256110520722636,
+      "grad_norm": 0.49609375,
+      "learning_rate": 3.2647492300245994e-06,
+      "loss": 0.1754,
+      "step": 4355
+    },
+    {
+      "epoch": 9.26673751328374,
+      "grad_norm": 0.50390625,
+      "learning_rate": 3.1713043159931734e-06,
+      "loss": 0.1767,
+      "step": 4360
+    },
+    {
+      "epoch": 9.277364505844846,
+      "grad_norm": 0.515625,
+      "learning_rate": 3.079194652009054e-06,
+      "loss": 0.1708,
+      "step": 4365
+    },
+    {
+      "epoch": 9.287991498405951,
+      "grad_norm": 0.48046875,
+      "learning_rate": 2.9884215082477408e-06,
+      "loss": 0.1723,
+      "step": 4370
+    },
+    {
+      "epoch": 9.298618490967057,
+      "grad_norm": 0.51171875,
+      "learning_rate": 2.898986136454318e-06,
+      "loss": 0.1801,
+      "step": 4375
+    },
+    {
+      "epoch": 9.309245483528162,
+      "grad_norm": 0.494140625,
+      "learning_rate": 2.810889769926217e-06,
+      "loss": 0.1774,
+      "step": 4380
+    },
+    {
+      "epoch": 9.319872476089266,
+      "grad_norm": 0.447265625,
+      "learning_rate": 2.7241336234962944e-06,
+      "loss": 0.1722,
+      "step": 4385
+    },
+    {
+      "epoch": 9.330499468650371,
+      "grad_norm": 0.515625,
+      "learning_rate": 2.6387188935159456e-06,
+      "loss": 0.1779,
+      "step": 4390
+    },
+    {
+      "epoch": 9.341126461211477,
+      "grad_norm": 0.46875,
+      "learning_rate": 2.5546467578387635e-06,
+      "loss": 0.1743,
+      "step": 4395
+    },
+    {
+      "epoch": 9.351753453772583,
+      "grad_norm": 0.4921875,
+      "learning_rate": 2.471918375804105e-06,
+      "loss": 0.1733,
+      "step": 4400
+    },
+    {
+      "epoch": 9.362380446333688,
+      "grad_norm": 0.4921875,
+      "learning_rate": 2.3905348882212964e-06,
+      "loss": 0.179,
+      "step": 4405
+    },
+    {
+      "epoch": 9.373007438894792,
+      "grad_norm": 0.49609375,
+      "learning_rate": 2.3104974173537743e-06,
+      "loss": 0.1742,
+      "step": 4410
+    },
+    {
+      "epoch": 9.383634431455897,
+      "grad_norm": 0.484375,
+      "learning_rate": 2.2318070669036685e-06,
+      "loss": 0.1743,
+      "step": 4415
+    },
+    {
+      "epoch": 9.394261424017003,
+      "grad_norm": 0.4921875,
+      "learning_rate": 2.1544649219965575e-06,
+      "loss": 0.1734,
+      "step": 4420
+    },
+    {
+      "epoch": 9.404888416578109,
+      "grad_norm": 0.46875,
+      "learning_rate": 2.07847204916648e-06,
+      "loss": 0.1729,
+      "step": 4425
+    },
+    {
+      "epoch": 9.415515409139214,
+      "grad_norm": 0.546875,
+      "learning_rate": 2.003829496341325e-06,
+      "loss": 0.1762,
+      "step": 4430
+    },
+    {
+      "epoch": 9.426142401700318,
+      "grad_norm": 0.51953125,
+      "learning_rate": 1.9305382928282546e-06,
+      "loss": 0.1798,
+      "step": 4435
+    },
+    {
+      "epoch": 9.436769394261423,
+      "grad_norm": 0.51953125,
+      "learning_rate": 1.8585994492995916e-06,
+      "loss": 0.1757,
+      "step": 4440
+    },
+    {
+      "epoch": 9.447396386822529,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.788013957778878e-06,
+      "loss": 0.1765,
+      "step": 4445
+    },
+    {
+      "epoch": 9.458023379383635,
+      "grad_norm": 0.5,
+      "learning_rate": 1.7187827916271382e-06,
+      "loss": 0.1739,
+      "step": 4450
+    },
+    {
+      "epoch": 9.46865037194474,
+      "grad_norm": 0.45703125,
+      "learning_rate": 1.65090690552957e-06,
+      "loss": 0.1732,
+      "step": 4455
+    },
+    {
+      "epoch": 9.479277364505846,
+      "grad_norm": 0.498046875,
+      "learning_rate": 1.5843872354822097e-06,
+      "loss": 0.173,
+      "step": 4460
+    },
+    {
+      "epoch": 9.48990435706695,
+      "grad_norm": 0.5078125,
+      "learning_rate": 1.5192246987791981e-06,
+      "loss": 0.177,
+      "step": 4465
+    },
+    {
+      "epoch": 9.500531349628055,
+      "grad_norm": 0.49609375,
+      "learning_rate": 1.4554201940000123e-06,
+      "loss": 0.177,
+      "step": 4470
+    },
+    {
+      "epoch": 9.51115834218916,
+      "grad_norm": 0.5,
+      "learning_rate": 1.3929746009971433e-06,
+      "loss": 0.179,
+      "step": 4475
+    },
+    {
+      "epoch": 9.521785334750266,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.3318887808839274e-06,
+      "loss": 0.1714,
+      "step": 4480
+    },
+    {
+      "epoch": 9.532412327311372,
+      "grad_norm": 0.484375,
+      "learning_rate": 1.272163576022667e-06,
+      "loss": 0.1729,
+      "step": 4485
+    },
+    {
+      "epoch": 9.543039319872475,
+      "grad_norm": 0.46484375,
+      "learning_rate": 1.21379981001305e-06,
+      "loss": 0.1733,
+      "step": 4490
+    },
+    {
+      "epoch": 9.553666312433581,
+      "grad_norm": 0.5078125,
+      "learning_rate": 1.1567982876807603e-06,
+      "loss": 0.175,
+      "step": 4495
+    },
+    {
+      "epoch": 9.564293304994687,
+      "grad_norm": 0.474609375,
+      "learning_rate": 1.1011597950663865e-06,
+      "loss": 0.1816,
+      "step": 4500
+    },
+    {
+      "epoch": 9.574920297555792,
+      "grad_norm": 0.490234375,
+      "learning_rate": 1.0468850994145963e-06,
+      "loss": 0.1777,
+      "step": 4505
+    },
+    {
+      "epoch": 9.585547290116898,
+      "grad_norm": 0.5,
+      "learning_rate": 9.939749491635341e-07,
+      "loss": 0.1776,
+      "step": 4510
+    },
+    {
+      "epoch": 9.596174282678001,
+      "grad_norm": 0.5,
+      "learning_rate": 9.424300739345304e-07,
+      "loss": 0.1727,
+      "step": 4515
+    },
+    {
+      "epoch": 9.606801275239107,
+      "grad_norm": 0.478515625,
+      "learning_rate": 8.922511845219971e-07,
+      "loss": 0.1734,
+      "step": 4520
+    },
+    {
+      "epoch": 9.617428267800213,
+      "grad_norm": 0.49609375,
+      "learning_rate": 8.434389728836589e-07,
+      "loss": 0.1806,
+      "step": 4525
+    },
+    {
+      "epoch": 9.628055260361318,
+      "grad_norm": 0.51953125,
+      "learning_rate": 7.959941121310266e-07,
+      "loss": 0.1729,
+      "step": 4530
+    },
+    {
+      "epoch": 9.638682252922424,
+      "grad_norm": 0.486328125,
+      "learning_rate": 7.499172565200718e-07,
+      "loss": 0.1729,
+      "step": 4535
+    },
+    {
+      "epoch": 9.649309245483527,
+      "grad_norm": 0.455078125,
+      "learning_rate": 7.052090414422119e-07,
+      "loss": 0.1821,
+      "step": 4540
+    },
+    {
+      "epoch": 9.659936238044633,
+      "grad_norm": 0.47265625,
+      "learning_rate": 6.618700834155945e-07,
+      "loss": 0.1825,
+      "step": 4545
+    },
+    {
+      "epoch": 9.670563230605739,
+      "grad_norm": 0.51171875,
+      "learning_rate": 6.199009800765265e-07,
+      "loss": 0.1709,
+      "step": 4550
+    },
+    {
+      "epoch": 9.681190223166844,
+      "grad_norm": 0.5,
+      "learning_rate": 5.793023101712924e-07,
+      "loss": 0.1771,
+      "step": 4555
+    },
+    {
+      "epoch": 9.69181721572795,
+      "grad_norm": 0.5390625,
+      "learning_rate": 5.400746335481488e-07,
+      "loss": 0.1771,
+      "step": 4560
+    },
+    {
+      "epoch": 9.702444208289053,
+      "grad_norm": 0.52734375,
+      "learning_rate": 5.022184911495864e-07,
+      "loss": 0.1807,
+      "step": 4565
+    },
+    {
+      "epoch": 9.713071200850159,
+      "grad_norm": 0.474609375,
+      "learning_rate": 4.6573440500492504e-07,
+      "loss": 0.1677,
+      "step": 4570
+    },
+    {
+      "epoch": 9.723698193411265,
+      "grad_norm": 0.5,
+      "learning_rate": 4.306228782230304e-07,
+      "loss": 0.1755,
+      "step": 4575
+    },
+    {
+      "epoch": 9.73432518597237,
+      "grad_norm": 0.494140625,
+      "learning_rate": 3.96884394985475e-07,
+      "loss": 0.1741,
+      "step": 4580
+    },
+    {
+      "epoch": 9.744952178533476,
+      "grad_norm": 0.466796875,
+      "learning_rate": 3.6451942053975507e-07,
+      "loss": 0.1708,
+      "step": 4585
+    },
+    {
+      "epoch": 9.755579171094581,
+      "grad_norm": 0.490234375,
+      "learning_rate": 3.335284011929951e-07,
+      "loss": 0.1732,
+      "step": 4590
+    },
+    {
+      "epoch": 9.766206163655685,
+      "grad_norm": 0.48828125,
+      "learning_rate": 3.0391176430567547e-07,
+      "loss": 0.1757,
+      "step": 4595
+    },
+    {
+      "epoch": 9.77683315621679,
+      "grad_norm": 0.50390625,
+      "learning_rate": 2.756699182858369e-07,
+      "loss": 0.1786,
+      "step": 4600
+    },
+    {
+      "epoch": 9.787460148777896,
+      "grad_norm": 0.48828125,
+      "learning_rate": 2.488032525833628e-07,
+      "loss": 0.1746,
+      "step": 4605
+    },
+    {
+      "epoch": 9.798087141339002,
+      "grad_norm": 0.494140625,
+      "learning_rate": 2.2331213768468363e-07,
+      "loss": 0.1767,
+      "step": 4610
+    },
+    {
+      "epoch": 9.808714133900107,
+      "grad_norm": 0.484375,
+      "learning_rate": 1.9919692510762533e-07,
+      "loss": 0.1795,
+      "step": 4615
+    },
+    {
+      "epoch": 9.819341126461211,
+      "grad_norm": 0.5078125,
+      "learning_rate": 1.7645794739654665e-07,
+      "loss": 0.1764,
+      "step": 4620
+    },
+    {
+      "epoch": 9.829968119022316,
+      "grad_norm": 0.5234375,
+      "learning_rate": 1.5509551811783152e-07,
+      "loss": 0.1781,
+      "step": 4625
+    },
+    {
+      "epoch": 9.840595111583422,
+      "grad_norm": 0.52734375,
+      "learning_rate": 1.351099318554705e-07,
+      "loss": 0.1755,
+      "step": 4630
+    },
+    {
+      "epoch": 9.851222104144528,
+      "grad_norm": 0.498046875,
+      "learning_rate": 1.1650146420704166e-07,
+      "loss": 0.1758,
+      "step": 4635
+    },
+    {
+      "epoch": 9.861849096705633,
+      "grad_norm": 0.515625,
+      "learning_rate": 9.927037177993592e-08,
+      "loss": 0.1774,
+      "step": 4640
+    },
+    {
+      "epoch": 9.872476089266737,
+      "grad_norm": 0.5234375,
+      "learning_rate": 8.341689218775984e-08,
+      "loss": 0.1827,
+      "step": 4645
+    },
+    {
+      "epoch": 9.883103081827842,
+      "grad_norm": 0.49609375,
+      "learning_rate": 6.894124404711599e-08,
+      "loss": 0.1717,
+      "step": 4650
+    },
+    {
+      "epoch": 9.893730074388948,
+      "grad_norm": 0.4609375,
+      "learning_rate": 5.584362697453882e-08,
+      "loss": 0.1773,
+      "step": 4655
+    },
+    {
+      "epoch": 9.904357066950054,
+      "grad_norm": 0.494140625,
+      "learning_rate": 4.4124221583785595e-08,
+      "loss": 0.1738,
+      "step": 4660
+    },
+    {
+      "epoch": 9.91498405951116,
+      "grad_norm": 0.49609375,
+      "learning_rate": 3.378318948332737e-08,
+      "loss": 0.1792,
+      "step": 4665
+    },
+    {
+      "epoch": 9.925611052072263,
+      "grad_norm": 0.5859375,
+      "learning_rate": 2.482067327409521e-08,
+      "loss": 0.173,
+      "step": 4670
+    },
+    {
+      "epoch": 9.936238044633368,
+      "grad_norm": 0.455078125,
+      "learning_rate": 1.7236796547559497e-08,
+      "loss": 0.1719,
+      "step": 4675
+    },
+    {
+      "epoch": 9.946865037194474,
+      "grad_norm": 0.451171875,
+      "learning_rate": 1.103166388398691e-08,
+      "loss": 0.1741,
+      "step": 4680
+    },
+    {
+      "epoch": 9.95749202975558,
+      "grad_norm": 0.486328125,
+      "learning_rate": 6.205360851041508e-09,
+      "loss": 0.1773,
+      "step": 4685
+    },
+    {
+      "epoch": 9.968119022316685,
+      "grad_norm": 0.490234375,
+      "learning_rate": 2.7579540025524097e-09,
+      "loss": 0.1762,
+      "step": 4690
+    },
+    {
+      "epoch": 9.978746014877789,
+      "grad_norm": 0.44140625,
+      "learning_rate": 6.894908776255982e-10,
+      "loss": 0.1745,
+      "step": 4695
+    },
+    {
+      "epoch": 9.989373007438894,
+      "grad_norm": 0.51953125,
+      "learning_rate": 0.0,
+      "loss": 0.1757,
+      "step": 4700
+    },
+    {
+      "epoch": 9.989373007438894,
+      "eval_loss": 3.9657981395721436,
+      "eval_runtime": 0.7855,
+      "eval_samples_per_second": 6.366,
+      "eval_steps_per_second": 1.273,
+      "step": 4700
+    },
+    {
+      "epoch": 9.989373007438894,
+      "step": 4700,
+      "total_flos": 7.255013733871649e+18,
+      "train_loss": 0.5076825852977469,
+      "train_runtime": 24662.4426,
+      "train_samples_per_second": 6.102,
+      "train_steps_per_second": 0.191
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 4700,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 100,
+  "total_flos": 7.255013733871649e+18,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}