diff --git "a/checkpoint-1695/trainer_state.json" "b/checkpoint-1695/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-1695/trainer_state.json"
@@ -0,0 +1,11982 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.9986731534719153,
+  "eval_steps": 142,
+  "global_step": 1695,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 4.705155372619629,
+      "learning_rate": 0.0001,
+      "loss": 3.3182,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "eval_loss": 3.3362529277801514,
+      "eval_runtime": 14.4366,
+      "eval_samples_per_second": 33.041,
+      "eval_steps_per_second": 8.312,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 4.644563674926758,
+      "learning_rate": 0.0002,
+      "loss": 3.2788,
+      "step": 2
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.3825764656066895,
+      "learning_rate": 0.0003,
+      "loss": 2.9231,
+      "step": 3
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.904296636581421,
+      "learning_rate": 0.0004,
+      "loss": 1.1824,
+      "step": 4
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.6988284587860107,
+      "learning_rate": 0.0005,
+      "loss": 0.3472,
+      "step": 5
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.5742101073265076,
+      "learning_rate": 0.0006,
+      "loss": 0.1478,
+      "step": 6
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.6511944532394409,
+      "learning_rate": 0.0007,
+      "loss": 0.1532,
+      "step": 7
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.305083751678467,
+      "learning_rate": 0.0008,
+      "loss": 0.2397,
+      "step": 8
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.7435078620910645,
+      "learning_rate": 0.0009000000000000001,
+      "loss": 0.434,
+      "step": 9
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.6199779510498047,
+      "learning_rate": 0.001,
+      "loss": 0.1743,
+      "step": 10
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.14406554400920868,
+      "learning_rate": 0.0009999991309598973,
+      "loss": 0.1404,
+      "step": 11
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.1965201050043106,
+      "learning_rate": 0.0009999965238426103,
+      "loss": 0.1418,
+      "step": 12
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 26.520109176635742,
+      "learning_rate": 0.0009999921786572016,
+      "loss": 0.2689,
+      "step": 13
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.0870603695511818,
+      "learning_rate": 0.0009999860954187755,
+      "loss": 0.1338,
+      "step": 14
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.05200817808508873,
+      "learning_rate": 0.0009999782741484788,
+      "loss": 0.1308,
+      "step": 15
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.2145700752735138,
+      "learning_rate": 0.0009999687148734995,
+      "loss": 0.1375,
+      "step": 16
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.19921083748340607,
+      "learning_rate": 0.0009999574176270667,
+      "loss": 0.1388,
+      "step": 17
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.193419337272644,
+      "learning_rate": 0.0009999443824484518,
+      "loss": 0.1978,
+      "step": 18
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.4399484395980835,
+      "learning_rate": 0.0009999296093829671,
+      "loss": 0.1518,
+      "step": 19
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 44.88853073120117,
+      "learning_rate": 0.0009999130984819661,
+      "loss": 0.9033,
+      "step": 20
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.3220385015010834,
+      "learning_rate": 0.0009998948498028434,
+      "loss": 0.1234,
+      "step": 21
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.5420748591423035,
+      "learning_rate": 0.0009998748634090344,
+      "loss": 0.1602,
+      "step": 22
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.5249865651130676,
+      "learning_rate": 0.0009998531393700149,
+      "loss": 0.1538,
+      "step": 23
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.056158341467380524,
+      "learning_rate": 0.000999829677761301,
+      "loss": 0.1374,
+      "step": 24
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.19818872213363647,
+      "learning_rate": 0.0009998044786644492,
+      "loss": 0.1413,
+      "step": 25
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.27901849150657654,
+      "learning_rate": 0.0009997775421670557,
+      "loss": 0.1395,
+      "step": 26
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.22768354415893555,
+      "learning_rate": 0.0009997488683627558,
+      "loss": 0.1241,
+      "step": 27
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.14878959953784943,
+      "learning_rate": 0.0009997184573512245,
+      "loss": 0.1243,
+      "step": 28
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.0589066743850708,
+      "learning_rate": 0.000999686309238175,
+      "loss": 0.2499,
+      "step": 29
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.11455405503511429,
+      "learning_rate": 0.00099965242413536,
+      "loss": 0.1254,
+      "step": 30
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.16566088795661926,
+      "learning_rate": 0.000999616802160569,
+      "loss": 0.1416,
+      "step": 31
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.3691716194152832,
+      "learning_rate": 0.0009995794434376297,
+      "loss": 0.1465,
+      "step": 32
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.09674070030450821,
+      "learning_rate": 0.000999540348096407,
+      "loss": 0.1373,
+      "step": 33
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.5034632086753845,
+      "learning_rate": 0.000999499516272803,
+      "loss": 0.1471,
+      "step": 34
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.26572930812835693,
+      "learning_rate": 0.0009994569481087553,
+      "loss": 0.1424,
+      "step": 35
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.20631802082061768,
+      "learning_rate": 0.0009994126437522376,
+      "loss": 0.1449,
+      "step": 36
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.11268749833106995,
+      "learning_rate": 0.0009993666033572591,
+      "loss": 0.1403,
+      "step": 37
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.6610996723175049,
+      "learning_rate": 0.0009993188270838635,
+      "loss": 0.1424,
+      "step": 38
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 98.93838500976562,
+      "learning_rate": 0.0009992693150981291,
+      "loss": 2.775,
+      "step": 39
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 32.53168869018555,
+      "learning_rate": 0.0009992180675721671,
+      "loss": 0.6932,
+      "step": 40
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 54.8778076171875,
+      "learning_rate": 0.0009991650846841226,
+      "loss": 5.7008,
+      "step": 41
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.0524775981903076,
+      "learning_rate": 0.000999110366618172,
+      "loss": 0.1623,
+      "step": 42
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.404278427362442,
+      "learning_rate": 0.0009990539135645246,
+      "loss": 0.1427,
+      "step": 43
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.7963409423828125,
+      "learning_rate": 0.0009989957257194198,
+      "loss": 0.174,
+      "step": 44
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.11620022356510162,
+      "learning_rate": 0.0009989358032851284,
+      "loss": 0.1339,
+      "step": 45
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.5025681853294373,
+      "learning_rate": 0.00099887414646995,
+      "loss": 0.1558,
+      "step": 46
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 78.1165771484375,
+      "learning_rate": 0.0009988107554882138,
+      "loss": 2.2938,
+      "step": 47
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.08389786630868912,
+      "learning_rate": 0.0009987456305602768,
+      "loss": 0.1409,
+      "step": 48
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 7.123101711273193,
+      "learning_rate": 0.0009986787719125242,
+      "loss": 0.1524,
+      "step": 49
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.5341290235519409,
+      "learning_rate": 0.0009986101797773666,
+      "loss": 0.1598,
+      "step": 50
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.05239284038543701,
+      "learning_rate": 0.000998539854393242,
+      "loss": 0.1386,
+      "step": 51
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.0722254291176796,
+      "learning_rate": 0.0009984677960046123,
+      "loss": 0.1385,
+      "step": 52
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.11535236239433289,
+      "learning_rate": 0.000998394004861964,
+      "loss": 0.1369,
+      "step": 53
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.7584894299507141,
+      "learning_rate": 0.0009983184812218072,
+      "loss": 0.108,
+      "step": 54
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.8361538052558899,
+      "learning_rate": 0.000998241225346674,
+      "loss": 0.1703,
+      "step": 55
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.37683162093162537,
+      "learning_rate": 0.0009981622375051184,
+      "loss": 0.1368,
+      "step": 56
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.5335961580276489,
+      "learning_rate": 0.0009980815179717144,
+      "loss": 0.1559,
+      "step": 57
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.2806299328804016,
+      "learning_rate": 0.0009979990670270565,
+      "loss": 0.1397,
+      "step": 58
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.4967437982559204,
+      "learning_rate": 0.0009979148849577574,
+      "loss": 0.1543,
+      "step": 59
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.10350017994642258,
+      "learning_rate": 0.0009978289720564471,
+      "loss": 0.1367,
+      "step": 60
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 78.35698699951172,
+      "learning_rate": 0.0009977413286217727,
+      "loss": 2.2474,
+      "step": 61
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.15264186263084412,
+      "learning_rate": 0.0009976519549583973,
+      "loss": 0.1311,
+      "step": 62
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.31865784525871277,
+      "learning_rate": 0.0009975608513769975,
+      "loss": 0.1407,
+      "step": 63
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.32891547679901123,
+      "learning_rate": 0.0009974680181942645,
+      "loss": 0.1423,
+      "step": 64
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.15653717517852783,
+      "learning_rate": 0.0009973734557329008,
+      "loss": 0.1365,
+      "step": 65
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.3237778842449188,
+      "learning_rate": 0.0009972771643216212,
+      "loss": 0.1407,
+      "step": 66
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.13634416460990906,
+      "learning_rate": 0.0009971791442951496,
+      "loss": 0.1378,
+      "step": 67
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.3488883376121521,
+      "learning_rate": 0.0009970793959942197,
+      "loss": 0.1429,
+      "step": 68
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.5150622129440308,
+      "learning_rate": 0.0009969779197655725,
+      "loss": 0.1492,
+      "step": 69
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.3482552468776703,
+      "learning_rate": 0.0009968747159619555,
+      "loss": 0.1415,
+      "step": 70
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.22551549971103668,
+      "learning_rate": 0.000996769784942122,
+      "loss": 0.1418,
+      "step": 71
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.20759086310863495,
+      "learning_rate": 0.0009966631270708287,
+      "loss": 0.1366,
+      "step": 72
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 13.050313949584961,
+      "learning_rate": 0.0009965547427188356,
+      "loss": 0.1375,
+      "step": 73
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.18372055888175964,
+      "learning_rate": 0.0009964446322629043,
+      "loss": 0.1285,
+      "step": 74
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.4404817819595337,
+      "learning_rate": 0.000996332796085796,
+      "loss": 0.1501,
+      "step": 75
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.269240379333496,
+      "learning_rate": 0.0009962192345762716,
+      "loss": 0.1346,
+      "step": 76
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 32.20164108276367,
+      "learning_rate": 0.0009961039481290888,
+      "loss": 0.3348,
+      "step": 77
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 82.92976379394531,
+      "learning_rate": 0.0009959869371450021,
+      "loss": 5.8309,
+      "step": 78
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.3416314721107483,
+      "learning_rate": 0.0009958682020307602,
+      "loss": 0.1418,
+      "step": 79
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 31.961870193481445,
+      "learning_rate": 0.0009957477431991053,
+      "loss": 0.1899,
+      "step": 80
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 38.58375930786133,
+      "learning_rate": 0.000995625561068772,
+      "loss": 0.5641,
+      "step": 81
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.32622194290161133,
+      "learning_rate": 0.0009955016560644846,
+      "loss": 0.1144,
+      "step": 82
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.264970779418945,
+      "learning_rate": 0.0009953760286169572,
+      "loss": 0.4788,
+      "step": 83
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.07168668508529663,
+      "learning_rate": 0.0009952486791628904,
+      "loss": 0.1326,
+      "step": 84
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 35.18340301513672,
+      "learning_rate": 0.000995119608144972,
+      "loss": 0.3884,
+      "step": 85
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.03896519914269447,
+      "learning_rate": 0.000994988816011873,
+      "loss": 0.1249,
+      "step": 86
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 14.499520301818848,
+      "learning_rate": 0.000994856303218248,
+      "loss": 0.3756,
+      "step": 87
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.3134947419166565,
+      "learning_rate": 0.000994722070224733,
+      "loss": 0.1539,
+      "step": 88
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 117.39696502685547,
+      "learning_rate": 0.000994586117497943,
+      "loss": 0.5885,
+      "step": 89
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 37.93465805053711,
+      "learning_rate": 0.0009944484455104716,
+      "loss": 0.7709,
+      "step": 90
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 236.63330078125,
+      "learning_rate": 0.0009943090547408888,
+      "loss": 6.0182,
+      "step": 91
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.1088515520095825,
+      "learning_rate": 0.0009941679456737394,
+      "loss": 0.1931,
+      "step": 92
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.11310256272554398,
+      "learning_rate": 0.0009940251187995411,
+      "loss": 0.1293,
+      "step": 93
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.6143047213554382,
+      "learning_rate": 0.0009938805746147828,
+      "loss": 0.2364,
+      "step": 94
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.2461577206850052,
+      "learning_rate": 0.0009937343136219232,
+      "loss": 0.1504,
+      "step": 95
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 97.17162322998047,
+      "learning_rate": 0.0009935863363293895,
+      "loss": 5.764,
+      "step": 96
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.5417380928993225,
+      "learning_rate": 0.000993436643251574,
+      "loss": 0.1576,
+      "step": 97
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.2737255096435547,
+      "learning_rate": 0.0009932852349088341,
+      "loss": 0.1437,
+      "step": 98
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 138.00778198242188,
+      "learning_rate": 0.0009931321118274896,
+      "loss": 4.1331,
+      "step": 99
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 46.4688606262207,
+      "learning_rate": 0.0009929772745398205,
+      "loss": 0.6178,
+      "step": 100
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.49907386302948,
+      "learning_rate": 0.0009928207235840663,
+      "loss": 0.1445,
+      "step": 101
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.33814460039138794,
+      "learning_rate": 0.0009926624595044233,
+      "loss": 0.139,
+      "step": 102
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.3241071403026581,
+      "learning_rate": 0.0009925024828510427,
+      "loss": 0.1404,
+      "step": 103
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 78.4036865234375,
+      "learning_rate": 0.000992340794180029,
+      "loss": 1.2663,
+      "step": 104
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.113776445388794,
+      "learning_rate": 0.000992177394053438,
+      "loss": 0.162,
+      "step": 105
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 16.512048721313477,
+      "learning_rate": 0.0009920122830392748,
+      "loss": 3.3373,
+      "step": 106
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 111.53176879882812,
+      "learning_rate": 0.0009918454617114918,
+      "loss": 2.3969,
+      "step": 107
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 14.91741943359375,
+      "learning_rate": 0.0009916769306499865,
+      "loss": 1.8837,
+      "step": 108
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 61.30055618286133,
+      "learning_rate": 0.0009915066904406,
+      "loss": 10.4922,
+      "step": 109
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.6948704123497009,
+      "learning_rate": 0.0009913347416751147,
+      "loss": 0.1536,
+      "step": 110
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.7721084356307983,
+      "learning_rate": 0.000991161084951252,
+      "loss": 0.1356,
+      "step": 111
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.24614596366882324,
+      "learning_rate": 0.0009909857208726704,
+      "loss": 0.1339,
+      "step": 112
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 7.189969062805176,
+      "learning_rate": 0.0009908086500489638,
+      "loss": 0.2551,
+      "step": 113
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.8675662279129028,
+      "learning_rate": 0.0009906298730956585,
+      "loss": 0.1668,
+      "step": 114
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.605249285697937,
+      "learning_rate": 0.0009904493906342123,
+      "loss": 0.1478,
+      "step": 115
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.8765722513198853,
+      "learning_rate": 0.0009902672032920106,
+      "loss": 0.1598,
+      "step": 116
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.6021157503128052,
+      "learning_rate": 0.0009900833117023665,
+      "loss": 0.1506,
+      "step": 117
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.28180792927742004,
+      "learning_rate": 0.000989897716504516,
+      "loss": 0.1389,
+      "step": 118
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.21730898320674896,
+      "learning_rate": 0.0009897104183436184,
+      "loss": 0.1377,
+      "step": 119
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.977118730545044,
+      "learning_rate": 0.0009895214178707516,
+      "loss": 0.1698,
+      "step": 120
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 2.674729585647583,
+      "learning_rate": 0.0009893307157429118,
+      "loss": 0.1559,
+      "step": 121
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.9852035045623779,
+      "learning_rate": 0.0009891383126230102,
+      "loss": 0.2027,
+      "step": 122
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.36689773201942444,
+      "learning_rate": 0.0009889442091798712,
+      "loss": 0.1498,
+      "step": 123
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.104621522128582,
+      "learning_rate": 0.000988748406088229,
+      "loss": 0.1379,
+      "step": 124
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 74.17496490478516,
+      "learning_rate": 0.0009885509040287268,
+      "loss": 0.7724,
+      "step": 125
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.2943025827407837,
+      "learning_rate": 0.0009883517036879132,
+      "loss": 0.2643,
+      "step": 126
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.828774094581604,
+      "learning_rate": 0.000988150805758241,
+      "loss": 0.1852,
+      "step": 127
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.13165877759456635,
+      "learning_rate": 0.0009879482109380632,
+      "loss": 0.1429,
+      "step": 128
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.662426769733429,
+      "learning_rate": 0.0009877439199316323,
+      "loss": 0.1643,
+      "step": 129
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.6256189942359924,
+      "learning_rate": 0.0009875379334490962,
+      "loss": 0.157,
+      "step": 130
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.5049256086349487,
+      "learning_rate": 0.0009873302522064972,
+      "loss": 0.1484,
+      "step": 131
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.4133671522140503,
+      "learning_rate": 0.0009871208769257685,
+      "loss": 0.1736,
+      "step": 132
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.7930824756622314,
+      "learning_rate": 0.0009869098083347323,
+      "loss": 0.1543,
+      "step": 133
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.5717449188232422,
+      "learning_rate": 0.0009866970471670965,
+      "loss": 0.1338,
+      "step": 134
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.582081139087677,
+      "learning_rate": 0.0009864825941624537,
+      "loss": 0.1692,
+      "step": 135
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 10.226588249206543,
+      "learning_rate": 0.0009862664500662763,
+      "loss": 0.2425,
+      "step": 136
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.1186953783035278,
+      "learning_rate": 0.0009860486156299164,
+      "loss": 0.2052,
+      "step": 137
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.2953661382198334,
+      "learning_rate": 0.000985829091610601,
+      "loss": 0.1408,
+      "step": 138
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.8647088408470154,
+      "learning_rate": 0.000985607878771431,
+      "loss": 0.1571,
+      "step": 139
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.41964420676231384,
+      "learning_rate": 0.0009853849778813776,
+      "loss": 0.1477,
+      "step": 140
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.25675931572914124,
+      "learning_rate": 0.0009851603897152803,
+      "loss": 0.1398,
+      "step": 141
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.2311631143093109,
+      "learning_rate": 0.0009849341150538434,
+      "loss": 0.1432,
+      "step": 142
+    },
+    {
+      "epoch": 0.25,
+      "eval_loss": 1.5366541147232056,
+      "eval_runtime": 14.6962,
+      "eval_samples_per_second": 32.457,
+      "eval_steps_per_second": 8.165,
+      "step": 142
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 41.83562469482422,
+      "learning_rate": 0.0009847061546836339,
+      "loss": 1.1525,
+      "step": 143
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.27440375089645386,
+      "learning_rate": 0.0009844765093970787,
+      "loss": 0.1452,
+      "step": 144
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.27643319964408875,
+      "learning_rate": 0.0009842451799924616,
+      "loss": 0.1069,
+      "step": 145
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.21519601345062256,
+      "learning_rate": 0.0009840121672739207,
+      "loss": 0.1358,
+      "step": 146
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.4073689877986908,
+      "learning_rate": 0.0009837774720514456,
+      "loss": 0.1545,
+      "step": 147
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.13685636222362518,
+      "learning_rate": 0.0009835410951408747,
+      "loss": 0.1259,
+      "step": 148
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.07474564015865326,
+      "learning_rate": 0.000983303037363892,
+      "loss": 0.1356,
+      "step": 149
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.45116662979125977,
+      "learning_rate": 0.0009830632995480241,
+      "loss": 0.1379,
+      "step": 150
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.1297813504934311,
+      "learning_rate": 0.0009828218825266388,
+      "loss": 0.1343,
+      "step": 151
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.5846492052078247,
+      "learning_rate": 0.00098257878713894,
+      "loss": 0.1563,
+      "step": 152
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.38457778096199036,
+      "learning_rate": 0.0009823340142299662,
+      "loss": 0.1477,
+      "step": 153
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.09184035658836365,
+      "learning_rate": 0.0009820875646505873,
+      "loss": 0.1376,
+      "step": 154
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.5166211128234863,
+      "learning_rate": 0.0009818394392575019,
+      "loss": 0.1498,
+      "step": 155
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.2788640260696411,
+      "learning_rate": 0.0009815896389132332,
+      "loss": 0.1434,
+      "step": 156
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.3762676417827606,
+      "learning_rate": 0.0009813381644861276,
+      "loss": 0.1482,
+      "step": 157
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.3615610897541046,
+      "learning_rate": 0.0009810850168503506,
+      "loss": 0.1312,
+      "step": 158
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.03483320027589798,
+      "learning_rate": 0.0009808301968858837,
+      "loss": 0.1239,
+      "step": 159
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.5616227984428406,
+      "learning_rate": 0.0009805737054785222,
+      "loss": 0.1881,
+      "step": 160
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.029542161151766777,
+      "learning_rate": 0.000980315543519871,
+      "loss": 0.1254,
+      "step": 161
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.142581045627594,
+      "learning_rate": 0.0009800557119073433,
+      "loss": 0.1258,
+      "step": 162
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.7289375066757202,
+      "learning_rate": 0.0009797942115441546,
+      "loss": 0.1526,
+      "step": 163
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.6975064873695374,
+      "learning_rate": 0.0009795310433393224,
+      "loss": 0.1487,
+      "step": 164
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.3072260618209839,
+      "learning_rate": 0.0009792662082076617,
+      "loss": 0.1712,
+      "step": 165
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.2993917465209961,
+      "learning_rate": 0.000978999707069782,
+      "loss": 0.1424,
+      "step": 166
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.3258236050605774,
+      "learning_rate": 0.0009787315408520839,
+      "loss": 0.135,
+      "step": 167
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.26566603779792786,
+      "learning_rate": 0.000978461710486756,
+      "loss": 0.1441,
+      "step": 168
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.1709599494934082,
+      "learning_rate": 0.0009781902169117718,
+      "loss": 0.2084,
+      "step": 169
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.6554279923439026,
+      "learning_rate": 0.000977917061070887,
+      "loss": 0.1634,
+      "step": 170
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.1635073721408844,
+      "learning_rate": 0.000977642243913635,
+      "loss": 0.1371,
+      "step": 171
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.4419834017753601,
+      "learning_rate": 0.0009773657663953242,
+      "loss": 0.1523,
+      "step": 172
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.839259147644043,
+      "learning_rate": 0.000977087629477035,
+      "loss": 0.1628,
+      "step": 173
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.1979222148656845,
+      "learning_rate": 0.0009768078341256155,
+      "loss": 0.1367,
+      "step": 174
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.2939910888671875,
+      "learning_rate": 0.0009765263813136795,
+      "loss": 0.1349,
+      "step": 175
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.19882674515247345,
+      "learning_rate": 0.0009762432720196024,
+      "loss": 0.1424,
+      "step": 176
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.07146954536437988,
+      "learning_rate": 0.000975958507227517,
+      "loss": 0.1237,
+      "step": 177
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.5031868815422058,
+      "learning_rate": 0.0009756720879273117,
+      "loss": 0.1592,
+      "step": 178
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.14860151708126068,
+      "learning_rate": 0.0009753840151146258,
+      "loss": 0.1396,
+      "step": 179
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.10280521959066391,
+      "learning_rate": 0.0009750942897908468,
+      "loss": 0.1333,
+      "step": 180
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.4652903974056244,
+      "learning_rate": 0.0009748029129631061,
+      "loss": 0.1421,
+      "step": 181
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.3985591530799866,
+      "learning_rate": 0.0009745098856442768,
+      "loss": 0.1459,
+      "step": 182
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.20321591198444366,
+      "learning_rate": 0.0009742152088529683,
+      "loss": 0.1381,
+      "step": 183
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.7694361805915833,
+      "learning_rate": 0.0009739188836135246,
+      "loss": 0.1676,
+      "step": 184
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.04469340294599533,
+      "learning_rate": 0.0009736209109560201,
+      "loss": 0.136,
+      "step": 185
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.08576061576604843,
+      "learning_rate": 0.0009733212919162549,
+      "loss": 0.1408,
+      "step": 186
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.042906519025564194,
+      "learning_rate": 0.0009730200275357535,
+      "loss": 0.1364,
+      "step": 187
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.30054494738578796,
+      "learning_rate": 0.0009727171188617588,
+      "loss": 0.1539,
+      "step": 188
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.05149005725979805,
+      "learning_rate": 0.0009724125669472299,
+      "loss": 0.1352,
+      "step": 189
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.1381620466709137,
+      "learning_rate": 0.0009721063728508383,
+      "loss": 0.1409,
+      "step": 190
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.37344205379486084,
+      "learning_rate": 0.0009717985376369639,
+      "loss": 0.1299,
+      "step": 191
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.1037706583738327,
+      "learning_rate": 0.0009714890623756912,
+      "loss": 0.1341,
+      "step": 192
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.14189712703227997,
+      "learning_rate": 0.0009711779481428056,
+      "loss": 0.1418,
+      "step": 193
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.15108801424503326,
+      "learning_rate": 0.0009708651960197903,
+      "loss": 0.142,
+      "step": 194
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.037045519798994064,
+      "learning_rate": 0.0009705508070938218,
+      "loss": 0.1315,
+      "step": 195
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.23301652073860168,
+      "learning_rate": 0.0009702347824577666,
+      "loss": 0.1396,
+      "step": 196
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.08476269990205765,
+      "learning_rate": 0.0009699171232101768,
+      "loss": 0.1392,
+      "step": 197
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.4222690463066101,
+      "learning_rate": 0.000969597830455287,
+      "loss": 0.1463,
+      "step": 198
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.3234136402606964,
+      "learning_rate": 0.0009692769053030099,
+      "loss": 0.1257,
+      "step": 199
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.04025443643331528,
+      "learning_rate": 0.0009689543488689332,
+      "loss": 0.1303,
+      "step": 200
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.07074520736932755,
+      "learning_rate": 0.0009686301622743144,
+      "loss": 0.1289,
+      "step": 201
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.0788850486278534,
+      "learning_rate": 0.0009683043466460782,
+      "loss": 0.1236,
+      "step": 202
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.525541365146637,
+      "learning_rate": 0.000967976903116812,
+      "loss": 0.1564,
+      "step": 203
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.6145509481430054,
+      "learning_rate": 0.0009676478328247623,
+      "loss": 0.156,
+      "step": 204
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.230132058262825,
+      "learning_rate": 0.0009673171369138296,
+      "loss": 0.1425,
+      "step": 205
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.03262978792190552,
+      "learning_rate": 0.0009669848165335666,
+      "loss": 0.1297,
+      "step": 206
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.0462469644844532,
+      "learning_rate": 0.0009666508728391718,
+      "loss": 0.1177,
+      "step": 207
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.06880385428667068,
+      "learning_rate": 0.0009663153069914874,
+      "loss": 0.1207,
+      "step": 208
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.4248260259628296,
+      "learning_rate": 0.000965978120156994,
+      "loss": 0.1571,
+      "step": 209
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.060492075979709625,
+      "learning_rate": 0.0009656393135078068,
+      "loss": 0.1219,
+      "step": 210
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.12135621905326843,
+      "learning_rate": 0.0009652988882216725,
+      "loss": 0.1323,
+      "step": 211
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.252119243144989,
+      "learning_rate": 0.0009649568454819637,
+      "loss": 0.1366,
+      "step": 212
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.5283567905426025,
+      "learning_rate": 0.0009646131864776761,
+      "loss": 0.1246,
+      "step": 213
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 2.224665880203247,
+      "learning_rate": 0.0009642679124034233,
+      "loss": 0.2582,
+      "step": 214
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.9277523756027222,
+      "learning_rate": 0.0009639210244594335,
+      "loss": 0.2131,
+      "step": 215
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.5668452978134155,
+      "learning_rate": 0.0009635725238515446,
+      "loss": 0.141,
+      "step": 216
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.13912492990493774,
+      "learning_rate": 0.000963222411791201,
+      "loss": 0.1418,
+      "step": 217
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.39307814836502075,
+      "learning_rate": 0.0009628706894954479,
+      "loss": 0.1477,
+      "step": 218
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.26248928904533386,
+      "learning_rate": 0.000962517358186929,
+      "loss": 0.1315,
+      "step": 219
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.2875257730484009,
+      "learning_rate": 0.0009621624190938803,
+      "loss": 0.1321,
+      "step": 220
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.6386964917182922,
+      "learning_rate": 0.0009618058734501269,
+      "loss": 0.1668,
+      "step": 221
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.16165001690387726,
+      "learning_rate": 0.0009614477224950789,
+      "loss": 0.1272,
+      "step": 222
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.6959558129310608,
+      "learning_rate": 0.0009610879674737262,
+      "loss": 0.1381,
+      "step": 223
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.1701437532901764,
+      "learning_rate": 0.0009607266096366352,
+      "loss": 0.1366,
+      "step": 224
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.2511409819126129,
+      "learning_rate": 0.0009603636502399437,
+      "loss": 0.126,
+      "step": 225
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.04554220288991928,
+      "learning_rate": 0.0009599990905453566,
+      "loss": 0.1321,
+      "step": 226
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.3964705765247345,
+      "learning_rate": 0.000959632931820142,
+      "loss": 0.1383,
+      "step": 227
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.10925984382629395,
+      "learning_rate": 0.0009592651753371264,
+      "loss": 0.1226,
+      "step": 228
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.19012318551540375,
+      "learning_rate": 0.0009588958223746903,
+      "loss": 0.1255,
+      "step": 229
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.23432157933712006,
+      "learning_rate": 0.0009585248742167639,
+      "loss": 0.1152,
+      "step": 230
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.1737753301858902,
+      "learning_rate": 0.0009581523321528223,
+      "loss": 0.1468,
+      "step": 231
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.2625434100627899,
+      "learning_rate": 0.0009577781974778817,
+      "loss": 0.1296,
+      "step": 232
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.3056884706020355,
+      "learning_rate": 0.000957402471492494,
+      "loss": 0.1574,
+      "step": 233
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.4111999273300171,
+      "learning_rate": 0.0009570251555027432,
+      "loss": 0.1434,
+      "step": 234
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.056673482060432434,
+      "learning_rate": 0.0009566462508202401,
+      "loss": 0.1337,
+      "step": 235
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.3861597180366516,
+      "learning_rate": 0.0009562657587621184,
+      "loss": 0.1609,
+      "step": 236
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.35893362760543823,
+      "learning_rate": 0.0009558836806510292,
+      "loss": 0.1189,
+      "step": 237
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.40538331866264343,
+      "learning_rate": 0.0009555000178151374,
+      "loss": 0.1504,
+      "step": 238
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 81.36141967773438,
+      "learning_rate": 0.0009551147715881167,
+      "loss": 4.7235,
+      "step": 239
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.21178042888641357,
+      "learning_rate": 0.0009547279433091446,
+      "loss": 0.1139,
+      "step": 240
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.27380529046058655,
+      "learning_rate": 0.0009543395343228983,
+      "loss": 0.1504,
+      "step": 241
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 41.42683410644531,
+      "learning_rate": 0.0009539495459795498,
+      "loss": 1.2477,
+      "step": 242
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.14853385090827942,
+      "learning_rate": 0.0009535579796347612,
+      "loss": 0.1343,
+      "step": 243
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.3484509289264679,
+      "learning_rate": 0.0009531648366496798,
+      "loss": 0.15,
+      "step": 244
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.20152732729911804,
+      "learning_rate": 0.0009527701183909336,
+      "loss": 0.1399,
+      "step": 245
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 80.84031677246094,
+      "learning_rate": 0.000952373826230627,
+      "loss": 3.1939,
+      "step": 246
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 15.475607872009277,
+      "learning_rate": 0.0009519759615463346,
+      "loss": 3.3935,
+      "step": 247
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 77.19477081298828,
+      "learning_rate": 0.0009515765257210979,
+      "loss": 6.5034,
+      "step": 248
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.1174071803689003,
+      "learning_rate": 0.0009511755201434205,
+      "loss": 0.1212,
+      "step": 249
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 16.503982543945312,
+      "learning_rate": 0.0009507729462072614,
+      "loss": 0.3753,
+      "step": 250
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 76.65412902832031,
+      "learning_rate": 0.0009503688053120326,
+      "loss": 0.9386,
+      "step": 251
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 94.82160186767578,
+      "learning_rate": 0.0009499630988625925,
+      "loss": 4.7449,
+      "step": 252
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.2721010148525238,
+      "learning_rate": 0.0009495558282692421,
+      "loss": 0.1358,
+      "step": 253
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.5150814056396484,
+      "learning_rate": 0.0009491469949477187,
+      "loss": 0.1622,
+      "step": 254
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 51.050167083740234,
+      "learning_rate": 0.0009487366003191931,
+      "loss": 0.7818,
+      "step": 255
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 11.698090553283691,
+      "learning_rate": 0.0009483246458102625,
+      "loss": 0.3862,
+      "step": 256
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.648543655872345,
+      "learning_rate": 0.0009479111328529472,
+      "loss": 0.1884,
+      "step": 257
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.745293140411377,
+      "learning_rate": 0.0009474960628846843,
+      "loss": 0.1562,
+      "step": 258
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.17890043556690216,
+      "learning_rate": 0.0009470794373483235,
+      "loss": 0.1425,
+      "step": 259
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.5058090686798096,
+      "learning_rate": 0.0009466612576921223,
+      "loss": 0.17,
+      "step": 260
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.3177820444107056,
+      "learning_rate": 0.00094624152536974,
+      "loss": 0.15,
+      "step": 261
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.49652573466300964,
+      "learning_rate": 0.0009458202418402337,
+      "loss": 0.145,
+      "step": 262
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 11.423394203186035,
+      "learning_rate": 0.0009453974085680526,
+      "loss": 0.349,
+      "step": 263
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.5422337055206299,
+      "learning_rate": 0.0009449730270230326,
+      "loss": 0.211,
+      "step": 264
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 103.68435668945312,
+      "learning_rate": 0.0009445470986803921,
+      "loss": 17.4069,
+      "step": 265
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 54.51758575439453,
+      "learning_rate": 0.0009441196250207267,
+      "loss": 15.685,
+      "step": 266
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 14.596623420715332,
+      "learning_rate": 0.0009436906075300032,
+      "loss": 0.791,
+      "step": 267
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 3.3164780139923096,
+      "learning_rate": 0.000943260047699555,
+      "loss": 0.3611,
+      "step": 268
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.3087855577468872,
+      "learning_rate": 0.0009428279470260776,
+      "loss": 0.1332,
+      "step": 269
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.1544523239135742,
+      "learning_rate": 0.0009423943070116219,
+      "loss": 0.2405,
+      "step": 270
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.27010253071784973,
+      "learning_rate": 0.00094195912916359,
+      "loss": 0.1241,
+      "step": 271
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.2287709265947342,
+      "learning_rate": 0.0009415224149947306,
+      "loss": 0.1366,
+      "step": 272
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.5216432809829712,
+      "learning_rate": 0.0009410841660231316,
+      "loss": 0.1641,
+      "step": 273
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.3091949224472046,
+      "learning_rate": 0.0009406443837722167,
+      "loss": 0.2524,
+      "step": 274
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.11813609302043915,
+      "learning_rate": 0.0009402030697707398,
+      "loss": 0.1353,
+      "step": 275
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.3709551095962524,
+      "learning_rate": 0.000939760225552779,
+      "loss": 0.2714,
+      "step": 276
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 8.527563095092773,
+      "learning_rate": 0.0009393158526577322,
+      "loss": 0.1955,
+      "step": 277
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 21.874027252197266,
+      "learning_rate": 0.0009388699526303105,
+      "loss": 0.2398,
+      "step": 278
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 51.793731689453125,
+      "learning_rate": 0.0009384225270205339,
+      "loss": 1.3069,
+      "step": 279
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.6711062788963318,
+      "learning_rate": 0.0009379735773837259,
+      "loss": 0.1664,
+      "step": 280
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 5.93789005279541,
+      "learning_rate": 0.0009375231052805072,
+      "loss": 0.2455,
+      "step": 281
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 62.527198791503906,
+      "learning_rate": 0.0009370711122767912,
+      "loss": 6.6447,
+      "step": 282
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 22.35348129272461,
+      "learning_rate": 0.000936617599943778,
+      "loss": 2.5015,
+      "step": 283
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.7277780175209045,
+      "learning_rate": 0.0009361625698579493,
+      "loss": 0.1667,
+      "step": 284
+    },
+    {
+      "epoch": 0.5,
+      "eval_loss": 0.14179374277591705,
+      "eval_runtime": 14.7139,
+      "eval_samples_per_second": 32.418,
+      "eval_steps_per_second": 8.156,
+      "step": 284
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.26271358132362366,
+      "learning_rate": 0.0009357060236010625,
+      "loss": 0.1429,
+      "step": 285
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 21.24464988708496,
+      "learning_rate": 0.0009352479627601457,
+      "loss": 2.0706,
+      "step": 286
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.5764265060424805,
+      "learning_rate": 0.0009347883889274922,
+      "loss": 0.3337,
+      "step": 287
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.6868380904197693,
+      "learning_rate": 0.0009343273037006539,
+      "loss": 0.1994,
+      "step": 288
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.9018234610557556,
+      "learning_rate": 0.0009338647086824372,
+      "loss": 0.1908,
+      "step": 289
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.7751502990722656,
+      "learning_rate": 0.0009334006054808966,
+      "loss": 0.2028,
+      "step": 290
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.5386408567428589,
+      "learning_rate": 0.0009329349957093293,
+      "loss": 0.1853,
+      "step": 291
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.4171103239059448,
+      "learning_rate": 0.0009324678809862695,
+      "loss": 0.3597,
+      "step": 292
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.4105970561504364,
+      "learning_rate": 0.0009319992629354827,
+      "loss": 0.1344,
+      "step": 293
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.26628127694129944,
+      "learning_rate": 0.000931529143185961,
+      "loss": 0.1453,
+      "step": 294
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 14.981964111328125,
+      "learning_rate": 0.0009310575233719154,
+      "loss": 0.2563,
+      "step": 295
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.6945788264274597,
+      "learning_rate": 0.0009305844051327725,
+      "loss": 0.1229,
+      "step": 296
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 31.034496307373047,
+      "learning_rate": 0.000930109790113167,
+      "loss": 1.2974,
+      "step": 297
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.5794603824615479,
+      "learning_rate": 0.0009296336799629368,
+      "loss": 0.22,
+      "step": 298
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.33219394087791443,
+      "learning_rate": 0.0009291560763371172,
+      "loss": 0.1262,
+      "step": 299
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 2.597118377685547,
+      "learning_rate": 0.000928676980895935,
+      "loss": 0.4026,
+      "step": 300
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 13.547090530395508,
+      "learning_rate": 0.0009281963953048029,
+      "loss": 1.3086,
+      "step": 301
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.289302945137024,
+      "learning_rate": 0.0009277143212343134,
+      "loss": 0.2215,
+      "step": 302
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.2176313400268555,
+      "learning_rate": 0.0009272307603602334,
+      "loss": 0.15,
+      "step": 303
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 4.436944007873535,
+      "learning_rate": 0.0009267457143634979,
+      "loss": 0.514,
+      "step": 304
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 29.960241317749023,
+      "learning_rate": 0.0009262591849302047,
+      "loss": 3.5389,
+      "step": 305
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 5.514049530029297,
+      "learning_rate": 0.0009257711737516082,
+      "loss": 0.2902,
+      "step": 306
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 2.331019401550293,
+      "learning_rate": 0.0009252816825241135,
+      "loss": 0.2775,
+      "step": 307
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.5708584189414978,
+      "learning_rate": 0.0009247907129492707,
+      "loss": 0.1438,
+      "step": 308
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 2.16607928276062,
+      "learning_rate": 0.0009242982667337685,
+      "loss": 0.2383,
+      "step": 309
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.5346423387527466,
+      "learning_rate": 0.0009238043455894293,
+      "loss": 0.1793,
+      "step": 310
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.6200052499771118,
+      "learning_rate": 0.000923308951233202,
+      "loss": 0.1473,
+      "step": 311
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 64.87612915039062,
+      "learning_rate": 0.0009228120853871572,
+      "loss": 0.8875,
+      "step": 312
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.077471137046814,
+      "learning_rate": 0.0009223137497784797,
+      "loss": 0.2114,
+      "step": 313
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 3.5934722423553467,
+      "learning_rate": 0.0009218139461394644,
+      "loss": 0.2852,
+      "step": 314
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.10276800394058228,
+      "learning_rate": 0.0009213126762075088,
+      "loss": 0.1365,
+      "step": 315
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 3.9422831535339355,
+      "learning_rate": 0.0009208099417251077,
+      "loss": 0.2949,
+      "step": 316
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.7574914693832397,
+      "learning_rate": 0.0009203057444398468,
+      "loss": 0.2621,
+      "step": 317
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.29479530453681946,
+      "learning_rate": 0.0009198000861043967,
+      "loss": 0.1341,
+      "step": 318
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.5362827181816101,
+      "learning_rate": 0.0009192929684765068,
+      "loss": 0.1398,
+      "step": 319
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.8159481287002563,
+      "learning_rate": 0.0009187843933189994,
+      "loss": 0.1863,
+      "step": 320
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.9413295388221741,
+      "learning_rate": 0.0009182743623997634,
+      "loss": 0.2104,
+      "step": 321
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.5306220650672913,
+      "learning_rate": 0.0009177628774917479,
+      "loss": 0.1537,
+      "step": 322
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.8887706398963928,
+      "learning_rate": 0.0009172499403729567,
+      "loss": 0.1963,
+      "step": 323
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.8467744588851929,
+      "learning_rate": 0.0009167355528264414,
+      "loss": 0.204,
+      "step": 324
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.19867151975631714,
+      "learning_rate": 0.0009162197166402956,
+      "loss": 0.1407,
+      "step": 325
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.13638383150100708,
+      "learning_rate": 0.0009157024336076487,
+      "loss": 0.1408,
+      "step": 326
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.2027496099472046,
+      "learning_rate": 0.0009151837055266594,
+      "loss": 0.1444,
+      "step": 327
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.370151549577713,
+      "learning_rate": 0.0009146635342005098,
+      "loss": 0.158,
+      "step": 328
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.3114052414894104,
+      "learning_rate": 0.000914141921437399,
+      "loss": 0.1464,
+      "step": 329
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.15394961833953857,
+      "learning_rate": 0.0009136188690505362,
+      "loss": 0.1341,
+      "step": 330
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.46498528122901917,
+      "learning_rate": 0.0009130943788581359,
+      "loss": 0.1426,
+      "step": 331
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.28067877888679504,
+      "learning_rate": 0.00091256845268341,
+      "loss": 0.1409,
+      "step": 332
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.061186857521533966,
+      "learning_rate": 0.0009120410923545619,
+      "loss": 0.1401,
+      "step": 333
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.26736098527908325,
+      "learning_rate": 0.0009115122997047811,
+      "loss": 0.1467,
+      "step": 334
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.5139696598052979,
+      "learning_rate": 0.0009109820765722356,
+      "loss": 0.1585,
+      "step": 335
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.40007275342941284,
+      "learning_rate": 0.000910450424800066,
+      "loss": 0.1473,
+      "step": 336
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.66825270652771,
+      "learning_rate": 0.0009099173462363792,
+      "loss": 0.1572,
+      "step": 337
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.5313024520874023,
+      "learning_rate": 0.0009093828427342418,
+      "loss": 0.1555,
+      "step": 338
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.4224655330181122,
+      "learning_rate": 0.0009088469161516735,
+      "loss": 0.1429,
+      "step": 339
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.03462248668074608,
+      "learning_rate": 0.0009083095683516414,
+      "loss": 0.1325,
+      "step": 340
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.542322039604187,
+      "learning_rate": 0.0009077708012020524,
+      "loss": 0.1755,
+      "step": 341
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.2164747267961502,
+      "learning_rate": 0.0009072306165757476,
+      "loss": 0.1458,
+      "step": 342
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.27414461970329285,
+      "learning_rate": 0.0009066890163504955,
+      "loss": 0.1512,
+      "step": 343
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.1911482959985733,
+      "learning_rate": 0.0009061460024089853,
+      "loss": 0.1185,
+      "step": 344
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.1287711262702942,
+      "learning_rate": 0.0009056015766388205,
+      "loss": 0.1372,
+      "step": 345
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.18598809838294983,
+      "learning_rate": 0.0009050557409325125,
+      "loss": 0.1341,
+      "step": 346
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.18694853782653809,
+      "learning_rate": 0.0009045084971874737,
+      "loss": 0.141,
+      "step": 347
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.06479912996292114,
+      "learning_rate": 0.0009039598473060113,
+      "loss": 0.1368,
+      "step": 348
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.17768733203411102,
+      "learning_rate": 0.0009034097931953201,
+      "loss": 0.1381,
+      "step": 349
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.28938984870910645,
+      "learning_rate": 0.0009028583367674765,
+      "loss": 0.1365,
+      "step": 350
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.2924034893512726,
+      "learning_rate": 0.0009023054799394316,
+      "loss": 0.1282,
+      "step": 351
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.28439652919769287,
+      "learning_rate": 0.0009017512246330042,
+      "loss": 0.151,
+      "step": 352
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.14329224824905396,
+      "learning_rate": 0.0009011955727748749,
+      "loss": 0.1419,
+      "step": 353
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.15245947241783142,
+      "learning_rate": 0.0009006385262965785,
+      "loss": 0.1163,
+      "step": 354
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.052399642765522,
+      "learning_rate": 0.000900080087134498,
+      "loss": 0.1241,
+      "step": 355
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.030301153659820557,
+      "learning_rate": 0.0008995202572298575,
+      "loss": 0.1232,
+      "step": 356
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.41738417744636536,
+      "learning_rate": 0.0008989590385287155,
+      "loss": 0.1675,
+      "step": 357
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.19307875633239746,
+      "learning_rate": 0.0008983964329819583,
+      "loss": 0.1328,
+      "step": 358
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.05682377517223358,
+      "learning_rate": 0.000897832442545293,
+      "loss": 0.1322,
+      "step": 359
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.15418089926242828,
+      "learning_rate": 0.0008972670691792409,
+      "loss": 0.1414,
+      "step": 360
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.07167459279298782,
+      "learning_rate": 0.0008967003148491304,
+      "loss": 0.1414,
+      "step": 361
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.2866109609603882,
+      "learning_rate": 0.0008961321815250904,
+      "loss": 0.1381,
+      "step": 362
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.281264990568161,
+      "learning_rate": 0.0008955626711820438,
+      "loss": 0.1365,
+      "step": 363
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.19263768196105957,
+      "learning_rate": 0.0008949917857996997,
+      "loss": 0.1394,
+      "step": 364
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.30531641840934753,
+      "learning_rate": 0.0008944195273625471,
+      "loss": 0.1478,
+      "step": 365
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.16229306161403656,
+      "learning_rate": 0.0008938458978598483,
+      "loss": 0.1412,
+      "step": 366
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.09315463900566101,
+      "learning_rate": 0.0008932708992856315,
+      "loss": 0.1397,
+      "step": 367
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.04228806868195534,
+      "learning_rate": 0.0008926945336386838,
+      "loss": 0.1383,
+      "step": 368
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.2209407389163971,
+      "learning_rate": 0.0008921168029225448,
+      "loss": 0.1434,
+      "step": 369
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.04254443198442459,
+      "learning_rate": 0.0008915377091454992,
+      "loss": 0.1326,
+      "step": 370
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.09651175886392593,
+      "learning_rate": 0.0008909572543205698,
+      "loss": 0.134,
+      "step": 371
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.2821654975414276,
+      "learning_rate": 0.0008903754404655105,
+      "loss": 0.1498,
+      "step": 372
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.43042680621147156,
+      "learning_rate": 0.0008897922696027998,
+      "loss": 0.1571,
+      "step": 373
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.06591568142175674,
+      "learning_rate": 0.0008892077437596332,
+      "loss": 0.1391,
+      "step": 374
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.08771979063749313,
+      "learning_rate": 0.0008886218649679161,
+      "loss": 0.1375,
+      "step": 375
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.03339942544698715,
+      "learning_rate": 0.0008880346352642574,
+      "loss": 0.1368,
+      "step": 376
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.15352453291416168,
+      "learning_rate": 0.0008874460566899616,
+      "loss": 0.1447,
+      "step": 377
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.1778584122657776,
+      "learning_rate": 0.0008868561312910222,
+      "loss": 0.1189,
+      "step": 378
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.11893154680728912,
+      "learning_rate": 0.0008862648611181144,
+      "loss": 0.1167,
+      "step": 379
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.4323861598968506,
+      "learning_rate": 0.0008856722482265886,
+      "loss": 0.1691,
+      "step": 380
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.28813356161117554,
+      "learning_rate": 0.0008850782946764618,
+      "loss": 0.1505,
+      "step": 381
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.5008757710456848,
+      "learning_rate": 0.0008844830025324122,
+      "loss": 0.1671,
+      "step": 382
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.12061876803636551,
+      "learning_rate": 0.0008838863738637705,
+      "loss": 0.1375,
+      "step": 383
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.6747052073478699,
+      "learning_rate": 0.0008832884107445138,
+      "loss": 0.1663,
+      "step": 384
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.18846777081489563,
+      "learning_rate": 0.0008826891152532579,
+      "loss": 0.1148,
+      "step": 385
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.0950111448764801,
+      "learning_rate": 0.0008820884894732497,
+      "loss": 0.1138,
+      "step": 386
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.42371127009391785,
+      "learning_rate": 0.0008814865354923613,
+      "loss": 0.142,
+      "step": 387
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.17662374675273895,
+      "learning_rate": 0.0008808832554030808,
+      "loss": 0.1255,
+      "step": 388
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.7766286134719849,
+      "learning_rate": 0.0008802786513025068,
+      "loss": 0.1613,
+      "step": 389
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.49581214785575867,
+      "learning_rate": 0.0008796727252923403,
+      "loss": 0.1346,
+      "step": 390
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.6148929595947266,
+      "learning_rate": 0.0008790654794788768,
+      "loss": 0.1426,
+      "step": 391
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.15860037505626678,
+      "learning_rate": 0.0008784569159730007,
+      "loss": 0.1382,
+      "step": 392
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.6793199777603149,
+      "learning_rate": 0.0008778470368901761,
+      "loss": 0.1398,
+      "step": 393
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.40314817428588867,
+      "learning_rate": 0.0008772358443504404,
+      "loss": 0.1428,
+      "step": 394
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.6403933167457581,
+      "learning_rate": 0.0008766233404783974,
+      "loss": 0.1556,
+      "step": 395
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.33554157614707947,
+      "learning_rate": 0.0008760095274032083,
+      "loss": 0.1439,
+      "step": 396
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.45690324902534485,
+      "learning_rate": 0.000875394407258586,
+      "loss": 0.1374,
+      "step": 397
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.0541120283305645,
+      "learning_rate": 0.0008747779821827868,
+      "loss": 0.1314,
+      "step": 398
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.6533159613609314,
+      "learning_rate": 0.0008741602543186031,
+      "loss": 0.169,
+      "step": 399
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.4919282793998718,
+      "learning_rate": 0.0008735412258133561,
+      "loss": 0.1569,
+      "step": 400
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.30325594544410706,
+      "learning_rate": 0.0008729208988188881,
+      "loss": 0.1471,
+      "step": 401
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.3497300148010254,
+      "learning_rate": 0.0008722992754915554,
+      "loss": 0.1457,
+      "step": 402
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.22892774641513824,
+      "learning_rate": 0.0008716763579922203,
+      "loss": 0.1334,
+      "step": 403
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.20050272345542908,
+      "learning_rate": 0.0008710521484862439,
+      "loss": 0.1446,
+      "step": 404
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.5029633641242981,
+      "learning_rate": 0.0008704266491434787,
+      "loss": 0.171,
+      "step": 405
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.2720576226711273,
+      "learning_rate": 0.0008697998621382607,
+      "loss": 0.144,
+      "step": 406
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.10961242765188217,
+      "learning_rate": 0.000869171789649402,
+      "loss": 0.1349,
+      "step": 407
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.13584192097187042,
+      "learning_rate": 0.0008685424338601833,
+      "loss": 0.1385,
+      "step": 408
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.6586437821388245,
+      "learning_rate": 0.0008679117969583464,
+      "loss": 0.1459,
+      "step": 409
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.24006032943725586,
+      "learning_rate": 0.0008672798811360864,
+      "loss": 0.1344,
+      "step": 410
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.1859387755393982,
+      "learning_rate": 0.0008666466885900438,
+      "loss": 0.1358,
+      "step": 411
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.5095134973526001,
+      "learning_rate": 0.0008660122215212977,
+      "loss": 0.1387,
+      "step": 412
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.1827729493379593,
+      "learning_rate": 0.0008653764821353573,
+      "loss": 0.1377,
+      "step": 413
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.14332665503025055,
+      "learning_rate": 0.0008647394726421547,
+      "loss": 0.131,
+      "step": 414
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.383101224899292,
+      "learning_rate": 0.0008641011952560371,
+      "loss": 0.146,
+      "step": 415
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.19079791009426117,
+      "learning_rate": 0.000863461652195759,
+      "loss": 0.1255,
+      "step": 416
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.49537310004234314,
+      "learning_rate": 0.0008628208456844747,
+      "loss": 0.1602,
+      "step": 417
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.5658069849014282,
+      "learning_rate": 0.0008621787779497306,
+      "loss": 0.1518,
+      "step": 418
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.2572256326675415,
+      "learning_rate": 0.0008615354512234569,
+      "loss": 0.1369,
+      "step": 419
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.1088945865631104,
+      "learning_rate": 0.0008608908677419605,
+      "loss": 0.1773,
+      "step": 420
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.35405099391937256,
+      "learning_rate": 0.0008602450297459173,
+      "loss": 0.1441,
+      "step": 421
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.39150556921958923,
+      "learning_rate": 0.0008595979394803633,
+      "loss": 0.147,
+      "step": 422
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.07459918409585953,
+      "learning_rate": 0.0008589495991946885,
+      "loss": 0.1338,
+      "step": 423
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.2999761402606964,
+      "learning_rate": 0.0008583000111426276,
+      "loss": 0.1357,
+      "step": 424
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.28417065739631653,
+      "learning_rate": 0.0008576491775822525,
+      "loss": 0.1411,
+      "step": 425
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.32605019211769104,
+      "learning_rate": 0.0008569971007759657,
+      "loss": 0.1329,
+      "step": 426
+    },
+    {
+      "epoch": 0.75,
+      "eval_loss": 0.13750587403774261,
+      "eval_runtime": 15.1749,
+      "eval_samples_per_second": 31.433,
+      "eval_steps_per_second": 7.908,
+      "step": 426
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.047430120408535004,
+      "learning_rate": 0.0008563437829904903,
+      "loss": 0.1373,
+      "step": 427
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.4616542160511017,
+      "learning_rate": 0.0008556892264968639,
+      "loss": 0.1534,
+      "step": 428
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.12317585945129395,
+      "learning_rate": 0.0008550334335704297,
+      "loss": 0.1338,
+      "step": 429
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.39604276418685913,
+      "learning_rate": 0.0008543764064908295,
+      "loss": 0.1434,
+      "step": 430
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.3490678369998932,
+      "learning_rate": 0.0008537181475419944,
+      "loss": 0.1365,
+      "step": 431
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.15001270174980164,
+      "learning_rate": 0.0008530586590121383,
+      "loss": 0.1358,
+      "step": 432
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.33340635895729065,
+      "learning_rate": 0.0008523979431937492,
+      "loss": 0.1367,
+      "step": 433
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.06029750779271126,
+      "learning_rate": 0.0008517360023835809,
+      "loss": 0.1366,
+      "step": 434
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.07978738099336624,
+      "learning_rate": 0.0008510728388826463,
+      "loss": 0.1345,
+      "step": 435
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.27599036693573,
+      "learning_rate": 0.0008504084549962079,
+      "loss": 0.1447,
+      "step": 436
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.13302059471607208,
+      "learning_rate": 0.0008497428530337706,
+      "loss": 0.1407,
+      "step": 437
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.20869582891464233,
+      "learning_rate": 0.0008490760353090737,
+      "loss": 0.1374,
+      "step": 438
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.10881117731332779,
+      "learning_rate": 0.0008484080041400825,
+      "loss": 0.1429,
+      "step": 439
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.20344361662864685,
+      "learning_rate": 0.0008477387618489807,
+      "loss": 0.139,
+      "step": 440
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.07153432071208954,
+      "learning_rate": 0.0008470683107621615,
+      "loss": 0.1315,
+      "step": 441
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.08688751608133316,
+      "learning_rate": 0.0008463966532102207,
+      "loss": 0.1346,
+      "step": 442
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.06495650112628937,
+      "learning_rate": 0.0008457237915279476,
+      "loss": 0.1307,
+      "step": 443
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.1892390102148056,
+      "learning_rate": 0.0008450497280543173,
+      "loss": 0.12,
+      "step": 444
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.2579623758792877,
+      "learning_rate": 0.0008443744651324827,
+      "loss": 0.1531,
+      "step": 445
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.149379700422287,
+      "learning_rate": 0.000843698005109766,
+      "loss": 0.1385,
+      "step": 446
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.19281132519245148,
+      "learning_rate": 0.0008430203503376506,
+      "loss": 0.1033,
+      "step": 447
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.33208444714546204,
+      "learning_rate": 0.0008423415031717733,
+      "loss": 0.1525,
+      "step": 448
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.15149784088134766,
+      "learning_rate": 0.0008416614659719157,
+      "loss": 0.1282,
+      "step": 449
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.24646438658237457,
+      "learning_rate": 0.0008409802411019962,
+      "loss": 0.1393,
+      "step": 450
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.2505553662776947,
+      "learning_rate": 0.000840297830930062,
+      "loss": 0.1453,
+      "step": 451
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.1632508784532547,
+      "learning_rate": 0.0008396142378282799,
+      "loss": 0.1274,
+      "step": 452
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.12370573729276657,
+      "learning_rate": 0.0008389294641729292,
+      "loss": 0.1201,
+      "step": 453
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.08046772330999374,
+      "learning_rate": 0.0008382435123443934,
+      "loss": 0.1263,
+      "step": 454
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.19015488028526306,
+      "learning_rate": 0.0008375563847271506,
+      "loss": 0.1318,
+      "step": 455
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.3562954366207123,
+      "learning_rate": 0.0008368680837097669,
+      "loss": 0.132,
+      "step": 456
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.06315189599990845,
+      "learning_rate": 0.000836178611684887,
+      "loss": 0.1113,
+      "step": 457
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.43667125701904297,
+      "learning_rate": 0.0008354879710492264,
+      "loss": 0.1908,
+      "step": 458
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.0708879753947258,
+      "learning_rate": 0.0008347961642035624,
+      "loss": 0.1399,
+      "step": 459
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.04855835437774658,
+      "learning_rate": 0.0008341031935527267,
+      "loss": 0.1258,
+      "step": 460
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.1364990919828415,
+      "learning_rate": 0.0008334090615055965,
+      "loss": 0.1344,
+      "step": 461
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.08166524022817612,
+      "learning_rate": 0.0008327137704750862,
+      "loss": 0.134,
+      "step": 462
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.09308458864688873,
+      "learning_rate": 0.0008320173228781389,
+      "loss": 0.1507,
+      "step": 463
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.07796576619148254,
+      "learning_rate": 0.000831319721135718,
+      "loss": 0.1284,
+      "step": 464
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.12168626487255096,
+      "learning_rate": 0.0008306209676727993,
+      "loss": 0.148,
+      "step": 465
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.18862847983837128,
+      "learning_rate": 0.000829921064918362,
+      "loss": 0.1229,
+      "step": 466
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.23615515232086182,
+      "learning_rate": 0.00082922001530538,
+      "loss": 0.1322,
+      "step": 467
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.34108766913414,
+      "learning_rate": 0.0008285178212708142,
+      "loss": 0.1338,
+      "step": 468
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.39579400420188904,
+      "learning_rate": 0.0008278144852556042,
+      "loss": 0.1341,
+      "step": 469
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.2620592713356018,
+      "learning_rate": 0.0008271100097046585,
+      "loss": 0.1395,
+      "step": 470
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.08778171986341476,
+      "learning_rate": 0.0008264043970668469,
+      "loss": 0.1328,
+      "step": 471
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.6086364388465881,
+      "learning_rate": 0.0008256976497949924,
+      "loss": 0.1271,
+      "step": 472
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.08982394635677338,
+      "learning_rate": 0.0008249897703458619,
+      "loss": 0.1346,
+      "step": 473
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.054080091416835785,
+      "learning_rate": 0.0008242807611801578,
+      "loss": 0.1218,
+      "step": 474
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.5981457829475403,
+      "learning_rate": 0.0008235706247625098,
+      "loss": 0.1715,
+      "step": 475
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.9139420986175537,
+      "learning_rate": 0.0008228593635614659,
+      "loss": 0.1983,
+      "step": 476
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.05938498303294182,
+      "learning_rate": 0.0008221469800494841,
+      "loss": 0.1308,
+      "step": 477
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.11526026576757431,
+      "learning_rate": 0.0008214334767029239,
+      "loss": 0.1422,
+      "step": 478
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.3049907386302948,
+      "learning_rate": 0.0008207188560020373,
+      "loss": 0.1419,
+      "step": 479
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.04782035946846008,
+      "learning_rate": 0.0008200031204309604,
+      "loss": 0.138,
+      "step": 480
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.12950918078422546,
+      "learning_rate": 0.000819286272477705,
+      "loss": 0.1315,
+      "step": 481
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.0429329015314579,
+      "learning_rate": 0.0008185683146341496,
+      "loss": 0.1354,
+      "step": 482
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.4792588949203491,
+      "learning_rate": 0.0008178492493960308,
+      "loss": 0.1476,
+      "step": 483
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.19784927368164062,
+      "learning_rate": 0.0008171290792629346,
+      "loss": 0.1394,
+      "step": 484
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.1172945499420166,
+      "learning_rate": 0.000816407806738288,
+      "loss": 0.1302,
+      "step": 485
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.3732689917087555,
+      "learning_rate": 0.0008156854343293501,
+      "loss": 0.1416,
+      "step": 486
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.5152392983436584,
+      "learning_rate": 0.0008149619645472031,
+      "loss": 0.1403,
+      "step": 487
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.15429601073265076,
+      "learning_rate": 0.000814237399906744,
+      "loss": 0.1322,
+      "step": 488
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.0002127885818481,
+      "learning_rate": 0.0008135117429266756,
+      "loss": 0.1303,
+      "step": 489
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.7232715487480164,
+      "learning_rate": 0.0008127849961294984,
+      "loss": 0.143,
+      "step": 490
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.13510456681251526,
+      "learning_rate": 0.0008120571620415006,
+      "loss": 0.1536,
+      "step": 491
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.5168789625167847,
+      "learning_rate": 0.0008113282431927503,
+      "loss": 0.1312,
+      "step": 492
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.7039850950241089,
+      "learning_rate": 0.000810598242117086,
+      "loss": 0.118,
+      "step": 493
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.5126641988754272,
+      "learning_rate": 0.0008098671613521089,
+      "loss": 0.2343,
+      "step": 494
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.6958308815956116,
+      "learning_rate": 0.0008091350034391731,
+      "loss": 0.1648,
+      "step": 495
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.979303359985352,
+      "learning_rate": 0.0008084017709233766,
+      "loss": 0.2261,
+      "step": 496
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.3389752507209778,
+      "learning_rate": 0.0008076674663535537,
+      "loss": 0.146,
+      "step": 497
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.19990071654319763,
+      "learning_rate": 0.0008069320922822643,
+      "loss": 0.1429,
+      "step": 498
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.33689868450164795,
+      "learning_rate": 0.0008061956512657871,
+      "loss": 0.147,
+      "step": 499
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.09925112873315811,
+      "learning_rate": 0.000805458145864109,
+      "loss": 0.1342,
+      "step": 500
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.961702585220337,
+      "learning_rate": 0.0008047195786409172,
+      "loss": 0.1361,
+      "step": 501
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.4342229962348938,
+      "learning_rate": 0.0008039799521635895,
+      "loss": 0.1485,
+      "step": 502
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.1798858642578125,
+      "learning_rate": 0.0008032392690031867,
+      "loss": 0.1314,
+      "step": 503
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.3653756380081177,
+      "learning_rate": 0.0008024975317344421,
+      "loss": 0.1388,
+      "step": 504
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 9.677605628967285,
+      "learning_rate": 0.0008017547429357531,
+      "loss": 0.4186,
+      "step": 505
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 8.348475456237793,
+      "learning_rate": 0.0008010109051891731,
+      "loss": 0.3806,
+      "step": 506
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 35.19770050048828,
+      "learning_rate": 0.0008002660210804011,
+      "loss": 3.6145,
+      "step": 507
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 9.18663501739502,
+      "learning_rate": 0.0007995200931987743,
+      "loss": 0.6162,
+      "step": 508
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.05997322499752045,
+      "learning_rate": 0.0007987731241372571,
+      "loss": 0.1129,
+      "step": 509
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.41408172249794006,
+      "learning_rate": 0.000798025116492434,
+      "loss": 0.1512,
+      "step": 510
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.4445393979549408,
+      "learning_rate": 0.0007972760728644996,
+      "loss": 0.1463,
+      "step": 511
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.19678063690662384,
+      "learning_rate": 0.0007965259958572495,
+      "loss": 0.1386,
+      "step": 512
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.45497119426727295,
+      "learning_rate": 0.0007957748880780721,
+      "loss": 0.1373,
+      "step": 513
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.6455509066581726,
+      "learning_rate": 0.0007950227521379381,
+      "loss": 0.1584,
+      "step": 514
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.3793765604496002,
+      "learning_rate": 0.0007942695906513929,
+      "loss": 0.1236,
+      "step": 515
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.20562775433063507,
+      "learning_rate": 0.0007935154062365467,
+      "loss": 0.1364,
+      "step": 516
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.3131325244903564,
+      "learning_rate": 0.0007927602015150655,
+      "loss": 0.1556,
+      "step": 517
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.1705670803785324,
+      "learning_rate": 0.0007920039791121617,
+      "loss": 0.1372,
+      "step": 518
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 6.6207499504089355,
+      "learning_rate": 0.0007912467416565861,
+      "loss": 0.22,
+      "step": 519
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.34343230724334717,
+      "learning_rate": 0.0007904884917806173,
+      "loss": 0.1453,
+      "step": 520
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.4290754497051239,
+      "learning_rate": 0.0007897292321200537,
+      "loss": 0.1177,
+      "step": 521
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.24469922482967377,
+      "learning_rate": 0.0007889689653142036,
+      "loss": 0.1369,
+      "step": 522
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.5307168960571289,
+      "learning_rate": 0.0007882076940058763,
+      "loss": 0.1542,
+      "step": 523
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.13802866637706757,
+      "learning_rate": 0.000787445420841373,
+      "loss": 0.1372,
+      "step": 524
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.36055922508239746,
+      "learning_rate": 0.0007866821484704776,
+      "loss": 0.1413,
+      "step": 525
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.36655113101005554,
+      "learning_rate": 0.0007859178795464472,
+      "loss": 0.1438,
+      "step": 526
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.6237390637397766,
+      "learning_rate": 0.0007851526167260034,
+      "loss": 0.1382,
+      "step": 527
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.42217007279396057,
+      "learning_rate": 0.0007843863626693221,
+      "loss": 0.1408,
+      "step": 528
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 24.023250579833984,
+      "learning_rate": 0.0007836191200400256,
+      "loss": 0.1517,
+      "step": 529
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.31599146127700806,
+      "learning_rate": 0.0007828508915051723,
+      "loss": 0.1353,
+      "step": 530
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.6795622706413269,
+      "learning_rate": 0.0007820816797352479,
+      "loss": 0.1515,
+      "step": 531
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.37493640184402466,
+      "learning_rate": 0.0007813114874041557,
+      "loss": 0.141,
+      "step": 532
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.7365546226501465,
+      "learning_rate": 0.0007805403171892079,
+      "loss": 0.1347,
+      "step": 533
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 18.393390655517578,
+      "learning_rate": 0.000779768171771116,
+      "loss": 0.1753,
+      "step": 534
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 2.2978413105010986,
+      "learning_rate": 0.0007789950538339812,
+      "loss": 0.1418,
+      "step": 535
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.495151162147522,
+      "learning_rate": 0.0007782209660652854,
+      "loss": 0.146,
+      "step": 536
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 7.705572605133057,
+      "learning_rate": 0.0007774459111558821,
+      "loss": 0.2042,
+      "step": 537
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.6036086678504944,
+      "learning_rate": 0.0007766698917999862,
+      "loss": 0.1695,
+      "step": 538
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 127.21215057373047,
+      "learning_rate": 0.0007758929106951656,
+      "loss": 18.5136,
+      "step": 539
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 40.58448791503906,
+      "learning_rate": 0.0007751149705423312,
+      "loss": 0.5973,
+      "step": 540
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.6296218633651733,
+      "learning_rate": 0.0007743360740457278,
+      "loss": 0.1849,
+      "step": 541
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.4533160924911499,
+      "learning_rate": 0.0007735562239129247,
+      "loss": 0.1464,
+      "step": 542
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.2379036247730255,
+      "learning_rate": 0.0007727754228548058,
+      "loss": 0.1267,
+      "step": 543
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.8904889225959778,
+      "learning_rate": 0.000771993673585561,
+      "loss": 0.2181,
+      "step": 544
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.8934443593025208,
+      "learning_rate": 0.0007712109788226762,
+      "loss": 0.2158,
+      "step": 545
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.3368353545665741,
+      "learning_rate": 0.0007704273412869238,
+      "loss": 0.1489,
+      "step": 546
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.2570180594921112,
+      "learning_rate": 0.0007696427637023537,
+      "loss": 0.144,
+      "step": 547
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 2.865034580230713,
+      "learning_rate": 0.0007688572487962834,
+      "loss": 0.1664,
+      "step": 548
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.4369525611400604,
+      "learning_rate": 0.0007680707992992888,
+      "loss": 0.1777,
+      "step": 549
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.2545509934425354,
+      "learning_rate": 0.0007672834179451942,
+      "loss": 0.1536,
+      "step": 550
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.14455465972423553,
+      "learning_rate": 0.0007664951074710638,
+      "loss": 0.1256,
+      "step": 551
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.16001886129379272,
+      "learning_rate": 0.0007657058706171911,
+      "loss": 0.1356,
+      "step": 552
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.2537885308265686,
+      "learning_rate": 0.0007649157101270903,
+      "loss": 0.1393,
+      "step": 553
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.33060047030448914,
+      "learning_rate": 0.0007641246287474854,
+      "loss": 0.148,
+      "step": 554
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.691941499710083,
+      "learning_rate": 0.0007633326292283028,
+      "loss": 0.1764,
+      "step": 555
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.20472805202007294,
+      "learning_rate": 0.0007625397143226595,
+      "loss": 0.1424,
+      "step": 556
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.7124485969543457,
+      "learning_rate": 0.0007617458867868553,
+      "loss": 0.1482,
+      "step": 557
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.09631184488534927,
+      "learning_rate": 0.0007609511493803615,
+      "loss": 0.1392,
+      "step": 558
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.20814809203147888,
+      "learning_rate": 0.0007601555048658133,
+      "loss": 0.1384,
+      "step": 559
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.19566737115383148,
+      "learning_rate": 0.0007593589560089984,
+      "loss": 0.1394,
+      "step": 560
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.13406091928482056,
+      "learning_rate": 0.0007585615055788484,
+      "loss": 0.1389,
+      "step": 561
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.07635807991027832,
+      "learning_rate": 0.0007577631563474291,
+      "loss": 0.1376,
+      "step": 562
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.11265091598033905,
+      "learning_rate": 0.0007569639110899302,
+      "loss": 0.1395,
+      "step": 563
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.31152746081352234,
+      "learning_rate": 0.0007561637725846567,
+      "loss": 0.1407,
+      "step": 564
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.13474373519420624,
+      "learning_rate": 0.0007553627436130183,
+      "loss": 0.1386,
+      "step": 565
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.23706336319446564,
+      "learning_rate": 0.0007545608269595201,
+      "loss": 0.1417,
+      "step": 566
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.30558836460113525,
+      "learning_rate": 0.0007537580254117531,
+      "loss": 0.1452,
+      "step": 567
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.106146439909935,
+      "learning_rate": 0.0007529543417603843,
+      "loss": 0.1372,
+      "step": 568
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.13968442380428314,
+      "eval_runtime": 15.1558,
+      "eval_samples_per_second": 31.473,
+      "eval_steps_per_second": 7.918,
+      "step": 568
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.3243511915206909,
+      "learning_rate": 0.0007521497787991472,
+      "loss": 0.1424,
+      "step": 569
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.19688986241817474,
+      "learning_rate": 0.0007513443393248312,
+      "loss": 0.1403,
+      "step": 570
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.1128445565700531,
+      "learning_rate": 0.0007505380261372734,
+      "loss": 0.1397,
+      "step": 571
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.11025507003068924,
+      "learning_rate": 0.0007497308420393477,
+      "loss": 0.1391,
+      "step": 572
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 0.19862700998783112,
+      "learning_rate": 0.0007489227898369558,
+      "loss": 0.1345,
+      "step": 573
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.11129032075405121,
+      "learning_rate": 0.0007481138723390164,
+      "loss": 0.1342,
+      "step": 574
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.21451863646507263,
+      "learning_rate": 0.0007473040923574567,
+      "loss": 0.132,
+      "step": 575
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.6781334280967712,
+      "learning_rate": 0.0007464934527072016,
+      "loss": 0.1688,
+      "step": 576
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.3881673812866211,
+      "learning_rate": 0.0007456819562061648,
+      "loss": 0.15,
+      "step": 577
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.0530267171561718,
+      "learning_rate": 0.0007448696056752383,
+      "loss": 0.139,
+      "step": 578
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 0.2782767415046692,
+      "learning_rate": 0.0007440564039382827,
+      "loss": 0.1334,
+      "step": 579
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.693821370601654,
+      "learning_rate": 0.0007432423538221178,
+      "loss": 0.1673,
+      "step": 580
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.27020275592803955,
+      "learning_rate": 0.0007424274581565122,
+      "loss": 0.1464,
+      "step": 581
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.42129820585250854,
+      "learning_rate": 0.0007416117197741742,
+      "loss": 0.1507,
+      "step": 582
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.21161474287509918,
+      "learning_rate": 0.0007407951415107412,
+      "loss": 0.1398,
+      "step": 583
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.13954728841781616,
+      "learning_rate": 0.00073997772620477,
+      "loss": 0.1387,
+      "step": 584
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 0.08771730959415436,
+      "learning_rate": 0.0007391594766977276,
+      "loss": 0.1419,
+      "step": 585
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.2630119025707245,
+      "learning_rate": 0.0007383403958339806,
+      "loss": 0.1483,
+      "step": 586
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.07496945559978485,
+      "learning_rate": 0.0007375204864607851,
+      "loss": 0.1339,
+      "step": 587
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.25115033984184265,
+      "learning_rate": 0.0007366997514282782,
+      "loss": 0.129,
+      "step": 588
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.24599210917949677,
+      "learning_rate": 0.0007358781935894659,
+      "loss": 0.1491,
+      "step": 589
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.14762777090072632,
+      "learning_rate": 0.0007350558158002153,
+      "loss": 0.1287,
+      "step": 590
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.03114377148449421,
+      "learning_rate": 0.0007342326209192435,
+      "loss": 0.1379,
+      "step": 591
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.5076407194137573,
+      "learning_rate": 0.000733408611808108,
+      "loss": 0.1122,
+      "step": 592
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.10492309182882309,
+      "learning_rate": 0.0007325837913311966,
+      "loss": 0.1284,
+      "step": 593
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.1740669459104538,
+      "learning_rate": 0.0007317581623557177,
+      "loss": 0.1458,
+      "step": 594
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.20419681072235107,
+      "learning_rate": 0.00073093172775169,
+      "loss": 0.1511,
+      "step": 595
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 0.1906755119562149,
+      "learning_rate": 0.0007301044903919325,
+      "loss": 0.1052,
+      "step": 596
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.10201478004455566,
+      "learning_rate": 0.0007292764531520552,
+      "loss": 0.1147,
+      "step": 597
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.4594266712665558,
+      "learning_rate": 0.0007284476189104485,
+      "loss": 0.1739,
+      "step": 598
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.1697234809398651,
+      "learning_rate": 0.0007276179905482729,
+      "loss": 0.1049,
+      "step": 599
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.09107261896133423,
+      "learning_rate": 0.0007267875709494499,
+      "loss": 0.1319,
+      "step": 600
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 0.07888934016227722,
+      "learning_rate": 0.0007259563630006512,
+      "loss": 0.1323,
+      "step": 601
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.06149132549762726,
+      "learning_rate": 0.0007251243695912886,
+      "loss": 0.1239,
+      "step": 602
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.3240460157394409,
+      "learning_rate": 0.0007242915936135052,
+      "loss": 0.1614,
+      "step": 603
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.04239710047841072,
+      "learning_rate": 0.0007234580379621635,
+      "loss": 0.1336,
+      "step": 604
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.04415787383913994,
+      "learning_rate": 0.000722623705534837,
+      "loss": 0.1336,
+      "step": 605
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.13500471413135529,
+      "learning_rate": 0.0007217885992317985,
+      "loss": 0.1429,
+      "step": 606
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.09405327588319778,
+      "learning_rate": 0.0007209527219560119,
+      "loss": 0.1399,
+      "step": 607
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.16369308531284332,
+      "learning_rate": 0.0007201160766131207,
+      "loss": 0.1389,
+      "step": 608
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.24509336054325104,
+      "learning_rate": 0.0007192786661114383,
+      "loss": 0.1376,
+      "step": 609
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.29961100220680237,
+      "learning_rate": 0.0007184404933619377,
+      "loss": 0.133,
+      "step": 610
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.4641360640525818,
+      "learning_rate": 0.0007176015612782421,
+      "loss": 0.1491,
+      "step": 611
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.059663355350494385,
+      "learning_rate": 0.0007167618727766138,
+      "loss": 0.1365,
+      "step": 612
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 0.16221192479133606,
+      "learning_rate": 0.0007159214307759448,
+      "loss": 0.1395,
+      "step": 613
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.04930780455470085,
+      "learning_rate": 0.0007150802381977463,
+      "loss": 0.1368,
+      "step": 614
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.6152715086936951,
+      "learning_rate": 0.0007142382979661386,
+      "loss": 0.1291,
+      "step": 615
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.15247471630573273,
+      "learning_rate": 0.0007133956130078411,
+      "loss": 0.1404,
+      "step": 616
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.7167736887931824,
+      "learning_rate": 0.000712552186252162,
+      "loss": 0.1642,
+      "step": 617
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 0.2419363558292389,
+      "learning_rate": 0.0007117080206309878,
+      "loss": 0.1317,
+      "step": 618
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.5636677742004395,
+      "learning_rate": 0.0007108631190787735,
+      "loss": 0.147,
+      "step": 619
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.26012521982192993,
+      "learning_rate": 0.0007100174845325327,
+      "loss": 0.1344,
+      "step": 620
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.14845141768455505,
+      "learning_rate": 0.0007091711199318265,
+      "loss": 0.1299,
+      "step": 621
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.1344316601753235,
+      "learning_rate": 0.0007083240282187542,
+      "loss": 0.1401,
+      "step": 622
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.08974921703338623,
+      "learning_rate": 0.0007074762123379423,
+      "loss": 0.1257,
+      "step": 623
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 0.3263636529445648,
+      "learning_rate": 0.0007066276752365351,
+      "loss": 0.1587,
+      "step": 624
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.12418147176504135,
+      "learning_rate": 0.0007057784198641835,
+      "loss": 0.1361,
+      "step": 625
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.3086402714252472,
+      "learning_rate": 0.0007049284491730353,
+      "loss": 0.1496,
+      "step": 626
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.3529713749885559,
+      "learning_rate": 0.000704077766117725,
+      "loss": 0.1523,
+      "step": 627
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.11717434972524643,
+      "learning_rate": 0.0007032263736553634,
+      "loss": 0.1402,
+      "step": 628
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.19153334200382233,
+      "learning_rate": 0.0007023742747455275,
+      "loss": 0.1407,
+      "step": 629
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.22798961400985718,
+      "learning_rate": 0.0007015214723502495,
+      "loss": 0.1262,
+      "step": 630
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.4415830373764038,
+      "learning_rate": 0.0007006679694340073,
+      "loss": 0.1494,
+      "step": 631
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.554614782333374,
+      "learning_rate": 0.0006998137689637142,
+      "loss": 0.1324,
+      "step": 632
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.32604262232780457,
+      "learning_rate": 0.0006989588739087078,
+      "loss": 0.1433,
+      "step": 633
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.5145484209060669,
+      "learning_rate": 0.0006981032872407406,
+      "loss": 0.152,
+      "step": 634
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.3538295030593872,
+      "learning_rate": 0.0006972470119339691,
+      "loss": 0.137,
+      "step": 635
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.2451559156179428,
+      "learning_rate": 0.0006963900509649435,
+      "loss": 0.1418,
+      "step": 636
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.2240092009305954,
+      "learning_rate": 0.0006955324073125978,
+      "loss": 0.1406,
+      "step": 637
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.5672935843467712,
+      "learning_rate": 0.0006946740839582387,
+      "loss": 0.1682,
+      "step": 638
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.3396548628807068,
+      "learning_rate": 0.000693815083885536,
+      "loss": 0.1538,
+      "step": 639
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.2464788407087326,
+      "learning_rate": 0.0006929554100805117,
+      "loss": 0.145,
+      "step": 640
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 0.08380208164453506,
+      "learning_rate": 0.0006920950655315297,
+      "loss": 0.1333,
+      "step": 641
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.04563472419977188,
+      "learning_rate": 0.000691234053229286,
+      "loss": 0.1371,
+      "step": 642
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.0336502380669117,
+      "learning_rate": 0.0006903723761667972,
+      "loss": 0.1383,
+      "step": 643
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.11504160612821579,
+      "learning_rate": 0.0006895100373393912,
+      "loss": 0.1366,
+      "step": 644
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.4302406311035156,
+      "learning_rate": 0.0006886470397446957,
+      "loss": 0.1464,
+      "step": 645
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.13670873641967773,
+      "learning_rate": 0.0006877833863826295,
+      "loss": 0.1399,
+      "step": 646
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 0.11441440135240555,
+      "learning_rate": 0.0006869190802553894,
+      "loss": 0.1389,
+      "step": 647
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.07245034724473953,
+      "learning_rate": 0.0006860541243674426,
+      "loss": 0.1376,
+      "step": 648
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.12628068029880524,
+      "learning_rate": 0.0006851885217255144,
+      "loss": 0.1314,
+      "step": 649
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.345865935087204,
+      "learning_rate": 0.0006843222753385784,
+      "loss": 0.1469,
+      "step": 650
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.18721798062324524,
+      "learning_rate": 0.0006834553882178463,
+      "loss": 0.129,
+      "step": 651
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.1566080003976822,
+      "learning_rate": 0.0006825878633767564,
+      "loss": 0.1296,
+      "step": 652
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.13990430533885956,
+      "learning_rate": 0.0006817197038309643,
+      "loss": 0.1245,
+      "step": 653
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.26073744893074036,
+      "learning_rate": 0.000680850912598332,
+      "loss": 0.1437,
+      "step": 654
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.05034814029932022,
+      "learning_rate": 0.0006799814926989171,
+      "loss": 0.1209,
+      "step": 655
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.29498428106307983,
+      "learning_rate": 0.0006791114471549626,
+      "loss": 0.1476,
+      "step": 656
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.24109311401844025,
+      "learning_rate": 0.0006782407789908863,
+      "loss": 0.1421,
+      "step": 657
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 0.2070060819387436,
+      "learning_rate": 0.0006773694912332707,
+      "loss": 0.1174,
+      "step": 658
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.05099210515618324,
+      "learning_rate": 0.0006764975869108514,
+      "loss": 0.1325,
+      "step": 659
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.03778371214866638,
+      "learning_rate": 0.0006756250690545078,
+      "loss": 0.1326,
+      "step": 660
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.23074184358119965,
+      "learning_rate": 0.0006747519406972524,
+      "loss": 0.1417,
+      "step": 661
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.162948340177536,
+      "learning_rate": 0.0006738782048742187,
+      "loss": 0.1422,
+      "step": 662
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.1257455050945282,
+      "learning_rate": 0.0006730038646226531,
+      "loss": 0.1352,
+      "step": 663
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 0.1732119917869568,
+      "learning_rate": 0.0006721289229819024,
+      "loss": 0.1313,
+      "step": 664
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.15348908305168152,
+      "learning_rate": 0.0006712533829934043,
+      "loss": 0.139,
+      "step": 665
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.06923094391822815,
+      "learning_rate": 0.0006703772477006757,
+      "loss": 0.1381,
+      "step": 666
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.307449609041214,
+      "learning_rate": 0.0006695005201493037,
+      "loss": 0.1365,
+      "step": 667
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.09788268059492111,
+      "learning_rate": 0.0006686232033869343,
+      "loss": 0.1358,
+      "step": 668
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 0.23847998678684235,
+      "learning_rate": 0.0006677453004632608,
+      "loss": 0.1399,
+      "step": 669
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.08175510168075562,
+      "learning_rate": 0.0006668668144300149,
+      "loss": 0.1372,
+      "step": 670
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.18189309537410736,
+      "learning_rate": 0.0006659877483409545,
+      "loss": 0.1401,
+      "step": 671
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.08665986359119415,
+      "learning_rate": 0.000665108105251855,
+      "loss": 0.1345,
+      "step": 672
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.40454381704330444,
+      "learning_rate": 0.0006642278882204963,
+      "loss": 0.1342,
+      "step": 673
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.27606263756752014,
+      "learning_rate": 0.0006633471003066543,
+      "loss": 0.1363,
+      "step": 674
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.06796804070472717,
+      "learning_rate": 0.000662465744572089,
+      "loss": 0.1353,
+      "step": 675
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.4458450376987457,
+      "learning_rate": 0.0006615838240805343,
+      "loss": 0.1521,
+      "step": 676
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.3369523286819458,
+      "learning_rate": 0.0006607013418976873,
+      "loss": 0.1489,
+      "step": 677
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.20170435309410095,
+      "learning_rate": 0.0006598183010911978,
+      "loss": 0.1263,
+      "step": 678
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.11186213046312332,
+      "learning_rate": 0.0006589347047306571,
+      "loss": 0.1344,
+      "step": 679
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.12327159941196442,
+      "learning_rate": 0.0006580505558875878,
+      "loss": 0.1354,
+      "step": 680
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.05389246717095375,
+      "learning_rate": 0.0006571658576354334,
+      "loss": 0.1333,
+      "step": 681
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.20890717208385468,
+      "learning_rate": 0.0006562806130495466,
+      "loss": 0.1428,
+      "step": 682
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.12948615849018097,
+      "learning_rate": 0.0006553948252071799,
+      "loss": 0.1372,
+      "step": 683
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.16449519991874695,
+      "learning_rate": 0.0006545084971874737,
+      "loss": 0.1418,
+      "step": 684
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.04887047037482262,
+      "learning_rate": 0.0006536216320714466,
+      "loss": 0.139,
+      "step": 685
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 0.1712215691804886,
+      "learning_rate": 0.0006527342329419836,
+      "loss": 0.1389,
+      "step": 686
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.14935021102428436,
+      "learning_rate": 0.000651846302883827,
+      "loss": 0.1369,
+      "step": 687
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.16822853684425354,
+      "learning_rate": 0.0006509578449835636,
+      "loss": 0.1393,
+      "step": 688
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.04274258390069008,
+      "learning_rate": 0.0006500688623296158,
+      "loss": 0.1339,
+      "step": 689
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.20485758781433105,
+      "learning_rate": 0.00064917935801223,
+      "loss": 0.1232,
+      "step": 690
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.16438162326812744,
+      "learning_rate": 0.0006482893351234658,
+      "loss": 0.1272,
+      "step": 691
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.0820753276348114,
+      "learning_rate": 0.0006473987967571855,
+      "loss": 0.1368,
+      "step": 692
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.5247365832328796,
+      "learning_rate": 0.000646507746009043,
+      "loss": 0.1702,
+      "step": 693
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.21259160339832306,
+      "learning_rate": 0.0006456161859764745,
+      "loss": 0.1384,
+      "step": 694
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.10756111145019531,
+      "learning_rate": 0.0006447241197586847,
+      "loss": 0.1316,
+      "step": 695
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.32431429624557495,
+      "learning_rate": 0.0006438315504566397,
+      "loss": 0.1505,
+      "step": 696
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.09354525059461594,
+      "learning_rate": 0.0006429384811730528,
+      "loss": 0.1338,
+      "step": 697
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 0.25492650270462036,
+      "learning_rate": 0.0006420449150123767,
+      "loss": 0.1391,
+      "step": 698
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.28658854961395264,
+      "learning_rate": 0.0006411508550807905,
+      "loss": 0.1336,
+      "step": 699
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.21230942010879517,
+      "learning_rate": 0.0006402563044861899,
+      "loss": 0.1369,
+      "step": 700
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.13693292438983917,
+      "learning_rate": 0.0006393612663381763,
+      "loss": 0.1347,
+      "step": 701
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.20328965783119202,
+      "learning_rate": 0.0006384657437480457,
+      "loss": 0.1349,
+      "step": 702
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 0.1463640034198761,
+      "learning_rate": 0.0006375697398287788,
+      "loss": 0.1316,
+      "step": 703
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.47083455324172974,
+      "learning_rate": 0.0006366732576950283,
+      "loss": 0.1538,
+      "step": 704
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.18148604035377502,
+      "learning_rate": 0.0006357763004631105,
+      "loss": 0.1264,
+      "step": 705
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.10440527647733688,
+      "learning_rate": 0.000634878871250992,
+      "loss": 0.1209,
+      "step": 706
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.42732179164886475,
+      "learning_rate": 0.000633980973178281,
+      "loss": 0.1581,
+      "step": 707
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.09864400327205658,
+      "learning_rate": 0.0006330826093662157,
+      "loss": 0.1398,
+      "step": 708
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.2839510142803192,
+      "learning_rate": 0.000632183782937652,
+      "loss": 0.1448,
+      "step": 709
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.18296539783477783,
+      "learning_rate": 0.0006312844970170551,
+      "loss": 0.1369,
+      "step": 710
+    },
+    {
+      "epoch": 1.26,
+      "eval_loss": 0.13823845982551575,
+      "eval_runtime": 15.0864,
+      "eval_samples_per_second": 31.618,
+      "eval_steps_per_second": 7.954,
+      "step": 710
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.2305176705121994,
+      "learning_rate": 0.0006303847547304872,
+      "loss": 0.1253,
+      "step": 711
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.2792215049266815,
+      "learning_rate": 0.0006294845592055967,
+      "loss": 0.1292,
+      "step": 712
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.3560260236263275,
+      "learning_rate": 0.0006285839135716078,
+      "loss": 0.142,
+      "step": 713
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.5769198536872864,
+      "learning_rate": 0.000627682820959309,
+      "loss": 0.1629,
+      "step": 714
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.24223147332668304,
+      "learning_rate": 0.000626781284501043,
+      "loss": 0.1436,
+      "step": 715
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.20025451481342316,
+      "learning_rate": 0.0006258793073306948,
+      "loss": 0.1271,
+      "step": 716
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.5235323905944824,
+      "learning_rate": 0.0006249768925836822,
+      "loss": 0.1362,
+      "step": 717
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.12457533925771713,
+      "learning_rate": 0.0006240740433969432,
+      "loss": 0.1267,
+      "step": 718
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.22851230204105377,
+      "learning_rate": 0.0006231707629089263,
+      "loss": 0.1368,
+      "step": 719
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 0.12162783741950989,
+      "learning_rate": 0.0006222670542595799,
+      "loss": 0.1348,
+      "step": 720
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.21811127662658691,
+      "learning_rate": 0.0006213629205903399,
+      "loss": 0.1302,
+      "step": 721
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.1042797714471817,
+      "learning_rate": 0.0006204583650441201,
+      "loss": 0.1227,
+      "step": 722
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.5917842388153076,
+      "learning_rate": 0.0006195533907653003,
+      "loss": 0.1218,
+      "step": 723
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.6369093656539917,
+      "learning_rate": 0.000618648000899717,
+      "loss": 0.1309,
+      "step": 724
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.298677921295166,
+      "learning_rate": 0.0006177421985946498,
+      "loss": 0.1329,
+      "step": 725
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.5087531208992004,
+      "learning_rate": 0.0006168359869988133,
+      "loss": 0.1619,
+      "step": 726
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.5805624723434448,
+      "learning_rate": 0.0006159293692623443,
+      "loss": 0.1388,
+      "step": 727
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.595432698726654,
+      "learning_rate": 0.0006150223485367914,
+      "loss": 0.1363,
+      "step": 728
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 2.0664656162261963,
+      "learning_rate": 0.0006141149279751042,
+      "loss": 0.1373,
+      "step": 729
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 1.3190929889678955,
+      "learning_rate": 0.0006132071107316221,
+      "loss": 0.1434,
+      "step": 730
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 0.19045250117778778,
+      "learning_rate": 0.0006122988999620634,
+      "loss": 0.1177,
+      "step": 731
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 1.3130842447280884,
+      "learning_rate": 0.0006113902988235145,
+      "loss": 0.1542,
+      "step": 732
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.5767085552215576,
+      "learning_rate": 0.0006104813104744187,
+      "loss": 0.1627,
+      "step": 733
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.45873621106147766,
+      "learning_rate": 0.0006095719380745653,
+      "loss": 0.1369,
+      "step": 734
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.4458267092704773,
+      "learning_rate": 0.0006086621847850788,
+      "loss": 0.1207,
+      "step": 735
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.13178426027297974,
+      "learning_rate": 0.0006077520537684072,
+      "loss": 0.1263,
+      "step": 736
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.19360630214214325,
+      "learning_rate": 0.0006068415481883121,
+      "loss": 0.1366,
+      "step": 737
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.12965673208236694,
+      "learning_rate": 0.0006059306712098571,
+      "loss": 0.1436,
+      "step": 738
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.13222691416740417,
+      "learning_rate": 0.0006050194259993966,
+      "loss": 0.1294,
+      "step": 739
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.14453792572021484,
+      "learning_rate": 0.0006041078157245648,
+      "loss": 0.1273,
+      "step": 740
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.27612432837486267,
+      "learning_rate": 0.0006031958435542659,
+      "loss": 0.1145,
+      "step": 741
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.6110266447067261,
+      "learning_rate": 0.0006022835126586609,
+      "loss": 0.1299,
+      "step": 742
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 0.29062649607658386,
+      "learning_rate": 0.0006013708262091586,
+      "loss": 0.132,
+      "step": 743
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.44451919198036194,
+      "learning_rate": 0.0006004577873784034,
+      "loss": 0.1235,
+      "step": 744
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.15329335629940033,
+      "learning_rate": 0.0005995443993402648,
+      "loss": 0.1462,
+      "step": 745
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.7718572616577148,
+      "learning_rate": 0.000598630665269826,
+      "loss": 0.1309,
+      "step": 746
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.427112340927124,
+      "learning_rate": 0.0005977165883433733,
+      "loss": 0.1565,
+      "step": 747
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 0.18447764217853546,
+      "learning_rate": 0.0005968021717383849,
+      "loss": 0.1431,
+      "step": 748
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.2808150053024292,
+      "learning_rate": 0.0005958874186335193,
+      "loss": 0.1429,
+      "step": 749
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.29161331057548523,
+      "learning_rate": 0.0005949723322086053,
+      "loss": 0.1238,
+      "step": 750
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.3760308027267456,
+      "learning_rate": 0.0005940569156446298,
+      "loss": 0.1416,
+      "step": 751
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.24679379165172577,
+      "learning_rate": 0.0005931411721237279,
+      "loss": 0.1366,
+      "step": 752
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.8587498664855957,
+      "learning_rate": 0.0005922251048291707,
+      "loss": 0.1525,
+      "step": 753
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.17934030294418335,
+      "learning_rate": 0.0005913087169453553,
+      "loss": 0.1287,
+      "step": 754
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.43589112162590027,
+      "learning_rate": 0.0005903920116577931,
+      "loss": 0.1472,
+      "step": 755
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.061591099947690964,
+      "learning_rate": 0.0005894749921530983,
+      "loss": 0.129,
+      "step": 756
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 2.9088215827941895,
+      "learning_rate": 0.0005885576616189781,
+      "loss": 0.1379,
+      "step": 757
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.5655280351638794,
+      "learning_rate": 0.0005876400232442205,
+      "loss": 0.138,
+      "step": 758
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.05325045809149742,
+      "learning_rate": 0.0005867220802186837,
+      "loss": 0.1269,
+      "step": 759
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 0.08003886044025421,
+      "learning_rate": 0.000585803835733285,
+      "loss": 0.1386,
+      "step": 760
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.09640295058488846,
+      "learning_rate": 0.0005848852929799894,
+      "loss": 0.1364,
+      "step": 761
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.5506378412246704,
+      "learning_rate": 0.0005839664551517988,
+      "loss": 0.1424,
+      "step": 762
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.06252831220626831,
+      "learning_rate": 0.000583047325442741,
+      "loss": 0.1384,
+      "step": 763
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.11976012587547302,
+      "learning_rate": 0.0005821279070478583,
+      "loss": 0.1422,
+      "step": 764
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 0.13022871315479279,
+      "learning_rate": 0.0005812082031631966,
+      "loss": 0.1357,
+      "step": 765
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.12146733701229095,
+      "learning_rate": 0.0005802882169857938,
+      "loss": 0.14,
+      "step": 766
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.17471055686473846,
+      "learning_rate": 0.00057936795171367,
+      "loss": 0.1432,
+      "step": 767
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.07833831012248993,
+      "learning_rate": 0.0005784474105458143,
+      "loss": 0.1425,
+      "step": 768
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.17084555327892303,
+      "learning_rate": 0.000577526596682176,
+      "loss": 0.1437,
+      "step": 769
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.112625353038311,
+      "learning_rate": 0.0005766055133236513,
+      "loss": 0.1429,
+      "step": 770
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 0.15991152822971344,
+      "learning_rate": 0.000575684163672074,
+      "loss": 0.1387,
+      "step": 771
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.18027663230895996,
+      "learning_rate": 0.0005747625509302033,
+      "loss": 0.1439,
+      "step": 772
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.4315040111541748,
+      "learning_rate": 0.0005738406783017127,
+      "loss": 0.1524,
+      "step": 773
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.3343091905117035,
+      "learning_rate": 0.0005729185489911797,
+      "loss": 0.1481,
+      "step": 774
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.16762638092041016,
+      "learning_rate": 0.0005719961662040733,
+      "loss": 0.1389,
+      "step": 775
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.11396286636590958,
+      "learning_rate": 0.0005710735331467444,
+      "loss": 0.1351,
+      "step": 776
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.11262958496809006,
+      "learning_rate": 0.0005701506530264132,
+      "loss": 0.1343,
+      "step": 777
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.06293229013681412,
+      "learning_rate": 0.0005692275290511592,
+      "loss": 0.1322,
+      "step": 778
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.037539321929216385,
+      "learning_rate": 0.0005683041644299093,
+      "loss": 0.13,
+      "step": 779
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.2424931526184082,
+      "learning_rate": 0.0005673805623724272,
+      "loss": 0.1333,
+      "step": 780
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.2825300991535187,
+      "learning_rate": 0.0005664567260893019,
+      "loss": 0.15,
+      "step": 781
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 0.12494263052940369,
+      "learning_rate": 0.000565532658791936,
+      "loss": 0.1317,
+      "step": 782
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.29031720757484436,
+      "learning_rate": 0.0005646083636925362,
+      "loss": 0.1593,
+      "step": 783
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.3038758933544159,
+      "learning_rate": 0.0005636838440041004,
+      "loss": 0.1551,
+      "step": 784
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.14983759820461273,
+      "learning_rate": 0.0005627591029404071,
+      "loss": 0.1402,
+      "step": 785
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.25971877574920654,
+      "learning_rate": 0.0005618341437160049,
+      "loss": 0.1389,
+      "step": 786
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.2425714135169983,
+      "learning_rate": 0.0005609089695462002,
+      "loss": 0.1399,
+      "step": 787
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 0.11737050861120224,
+      "learning_rate": 0.0005599835836470469,
+      "loss": 0.1237,
+      "step": 788
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.6737673878669739,
+      "learning_rate": 0.0005590579892353348,
+      "loss": 0.1837,
+      "step": 789
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.2361481785774231,
+      "learning_rate": 0.0005581321895285787,
+      "loss": 0.146,
+      "step": 790
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.47753140330314636,
+      "learning_rate": 0.0005572061877450068,
+      "loss": 0.1664,
+      "step": 791
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.2968634068965912,
+      "learning_rate": 0.0005562799871035495,
+      "loss": 0.1511,
+      "step": 792
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.20170801877975464,
+      "learning_rate": 0.0005553535908238294,
+      "loss": 0.1408,
+      "step": 793
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 0.11540532112121582,
+      "learning_rate": 0.0005544270021261482,
+      "loss": 0.1415,
+      "step": 794
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.10350099951028824,
+      "learning_rate": 0.0005535002242314772,
+      "loss": 0.1393,
+      "step": 795
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.06757602840662003,
+      "learning_rate": 0.0005525732603614444,
+      "loss": 0.1335,
+      "step": 796
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.11407013237476349,
+      "learning_rate": 0.0005516461137383254,
+      "loss": 0.1342,
+      "step": 797
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.40271708369255066,
+      "learning_rate": 0.0005507187875850305,
+      "loss": 0.1536,
+      "step": 798
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.35031354427337646,
+      "learning_rate": 0.000549791285125094,
+      "loss": 0.1489,
+      "step": 799
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.347901850938797,
+      "learning_rate": 0.0005488636095826636,
+      "loss": 0.1463,
+      "step": 800
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.17497143149375916,
+      "learning_rate": 0.0005479357641824877,
+      "loss": 0.1385,
+      "step": 801
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.42803797125816345,
+      "learning_rate": 0.0005470077521499062,
+      "loss": 0.1438,
+      "step": 802
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.77762371301651,
+      "learning_rate": 0.0005460795767108378,
+      "loss": 0.1616,
+      "step": 803
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.27612486481666565,
+      "learning_rate": 0.0005451512410917691,
+      "loss": 0.1424,
+      "step": 804
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 0.10936840623617172,
+      "learning_rate": 0.0005442227485197435,
+      "loss": 0.1379,
+      "step": 805
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.19322127103805542,
+      "learning_rate": 0.0005432941022223503,
+      "loss": 0.1279,
+      "step": 806
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.14601223170757294,
+      "learning_rate": 0.000542365305427713,
+      "loss": 0.1394,
+      "step": 807
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.05485713109374046,
+      "learning_rate": 0.0005414363613644781,
+      "loss": 0.1245,
+      "step": 808
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 3.4448142051696777,
+      "learning_rate": 0.0005405072732618043,
+      "loss": 0.3245,
+      "step": 809
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 0.3802805542945862,
+      "learning_rate": 0.0005395780443493508,
+      "loss": 0.1617,
+      "step": 810
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 2.19525146484375,
+      "learning_rate": 0.0005386486778572665,
+      "loss": 0.3246,
+      "step": 811
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.1237780973315239,
+      "learning_rate": 0.0005377191770161783,
+      "loss": 0.1348,
+      "step": 812
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.17096708714962006,
+      "learning_rate": 0.0005367895450571801,
+      "loss": 0.1417,
+      "step": 813
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.033038243651390076,
+      "learning_rate": 0.0005358597852118219,
+      "loss": 0.1308,
+      "step": 814
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.2681437134742737,
+      "learning_rate": 0.000534929900712098,
+      "loss": 0.1465,
+      "step": 815
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.11117050051689148,
+      "learning_rate": 0.0005339998947904363,
+      "loss": 0.1383,
+      "step": 816
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.25842952728271484,
+      "learning_rate": 0.0005330697706796861,
+      "loss": 0.1397,
+      "step": 817
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.08293187618255615,
+      "learning_rate": 0.0005321395316131083,
+      "loss": 0.1356,
+      "step": 818
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.14762446284294128,
+      "learning_rate": 0.0005312091808243631,
+      "loss": 0.1416,
+      "step": 819
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.4471145570278168,
+      "learning_rate": 0.0005302787215474991,
+      "loss": 0.1461,
+      "step": 820
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.2443719059228897,
+      "learning_rate": 0.0005293481570169421,
+      "loss": 0.1458,
+      "step": 821
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 0.22830860316753387,
+      "learning_rate": 0.0005284174904674835,
+      "loss": 0.139,
+      "step": 822
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.5184169411659241,
+      "learning_rate": 0.0005274867251342694,
+      "loss": 0.1417,
+      "step": 823
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.3812021017074585,
+      "learning_rate": 0.0005265558642527897,
+      "loss": 0.1346,
+      "step": 824
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.2922486662864685,
+      "learning_rate": 0.0005256249110588659,
+      "loss": 0.1294,
+      "step": 825
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.29819944500923157,
+      "learning_rate": 0.0005246938687886409,
+      "loss": 0.1401,
+      "step": 826
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.07050393521785736,
+      "learning_rate": 0.0005237627406785666,
+      "loss": 0.1307,
+      "step": 827
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 0.1988169550895691,
+      "learning_rate": 0.0005228315299653941,
+      "loss": 0.1359,
+      "step": 828
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.31983789801597595,
+      "learning_rate": 0.0005219002398861611,
+      "loss": 0.1459,
+      "step": 829
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.28960883617401123,
+      "learning_rate": 0.000520968873678181,
+      "loss": 0.1447,
+      "step": 830
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.36790764331817627,
+      "learning_rate": 0.0005200374345790325,
+      "loss": 0.1287,
+      "step": 831
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.055655404925346375,
+      "learning_rate": 0.0005191059258265471,
+      "loss": 0.1346,
+      "step": 832
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 0.42271995544433594,
+      "learning_rate": 0.0005181743506587989,
+      "loss": 0.1445,
+      "step": 833
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.4579026997089386,
+      "learning_rate": 0.0005172427123140923,
+      "loss": 0.1397,
+      "step": 834
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.30265891551971436,
+      "learning_rate": 0.0005163110140309518,
+      "loss": 0.1389,
+      "step": 835
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.23372715711593628,
+      "learning_rate": 0.0005153792590481101,
+      "loss": 0.1426,
+      "step": 836
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.22771018743515015,
+      "learning_rate": 0.0005144474506044969,
+      "loss": 0.1412,
+      "step": 837
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.32557952404022217,
+      "learning_rate": 0.000513515591939228,
+      "loss": 0.1382,
+      "step": 838
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.4409979283809662,
+      "learning_rate": 0.0005125836862915934,
+      "loss": 0.1382,
+      "step": 839
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 112.177978515625,
+      "learning_rate": 0.0005116517369010466,
+      "loss": 1.093,
+      "step": 840
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.13140363991260529,
+      "learning_rate": 0.0005107197470071933,
+      "loss": 0.1344,
+      "step": 841
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.0935206264257431,
+      "learning_rate": 0.00050978771984978,
+      "loss": 0.1315,
+      "step": 842
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.5304569602012634,
+      "learning_rate": 0.0005088556586686822,
+      "loss": 0.1549,
+      "step": 843
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.07438669353723526,
+      "learning_rate": 0.0005079235667038944,
+      "loss": 0.1311,
+      "step": 844
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 0.17763537168502808,
+      "learning_rate": 0.0005069914471955179,
+      "loss": 0.1342,
+      "step": 845
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.3256682753562927,
+      "learning_rate": 0.0005060593033837493,
+      "loss": 0.1435,
+      "step": 846
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.3771526515483856,
+      "learning_rate": 0.0005051271385088701,
+      "loss": 0.1434,
+      "step": 847
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.3716539740562439,
+      "learning_rate": 0.0005041949558112351,
+      "loss": 0.1329,
+      "step": 848
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.13685204088687897,
+      "learning_rate": 0.0005032627585312608,
+      "loss": 0.1415,
+      "step": 849
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 0.21241213381290436,
+      "learning_rate": 0.0005023305499094144,
+      "loss": 0.1384,
+      "step": 850
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.05780967324972153,
+      "learning_rate": 0.0005013983331862026,
+      "loss": 0.1366,
+      "step": 851
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.5117526650428772,
+      "learning_rate": 0.0005004661116021605,
+      "loss": 0.1537,
+      "step": 852
+    },
+    {
+      "epoch": 1.51,
+      "eval_loss": 0.14083661139011383,
+      "eval_runtime": 14.5613,
+      "eval_samples_per_second": 32.758,
+      "eval_steps_per_second": 8.241,
+      "step": 852
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.6577463150024414,
+      "learning_rate": 0.0004995338883978395,
+      "loss": 0.1461,
+      "step": 853
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.35039347410202026,
+      "learning_rate": 0.0004986016668137974,
+      "loss": 0.1345,
+      "step": 854
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.1379460096359253,
+      "learning_rate": 0.0004976694500905857,
+      "loss": 0.1425,
+      "step": 855
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 0.23959362506866455,
+      "learning_rate": 0.0004967372414687393,
+      "loss": 0.1535,
+      "step": 856
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.387977659702301,
+      "learning_rate": 0.000495805044188765,
+      "loss": 0.1709,
+      "step": 857
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.10591788589954376,
+      "learning_rate": 0.0004948728614911299,
+      "loss": 0.137,
+      "step": 858
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.1370454728603363,
+      "learning_rate": 0.0004939406966162507,
+      "loss": 0.1413,
+      "step": 859
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.10982546955347061,
+      "learning_rate": 0.0004930085528044823,
+      "loss": 0.1422,
+      "step": 860
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.1631615161895752,
+      "learning_rate": 0.0004920764332961055,
+      "loss": 0.1439,
+      "step": 861
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.4625565707683563,
+      "learning_rate": 0.0004911443413313179,
+      "loss": 0.13,
+      "step": 862
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.09019370377063751,
+      "learning_rate": 0.0004902122801502201,
+      "loss": 0.1367,
+      "step": 863
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.058873746544122696,
+      "learning_rate": 0.0004892802529928067,
+      "loss": 0.1388,
+      "step": 864
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.16651901602745056,
+      "learning_rate": 0.0004883482630989535,
+      "loss": 0.1383,
+      "step": 865
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.1222059577703476,
+      "learning_rate": 0.00048741631370840676,
+      "loss": 0.1391,
+      "step": 866
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 0.15417732298374176,
+      "learning_rate": 0.00048648440806077226,
+      "loss": 0.1368,
+      "step": 867
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.19719868898391724,
+      "learning_rate": 0.00048555254939550326,
+      "loss": 0.1423,
+      "step": 868
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.1811167150735855,
+      "learning_rate": 0.0004846207409518899,
+      "loss": 0.1382,
+      "step": 869
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.12746267020702362,
+      "learning_rate": 0.0004836889859690483,
+      "loss": 0.1375,
+      "step": 870
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.18294665217399597,
+      "learning_rate": 0.00048275728768590776,
+      "loss": 0.1376,
+      "step": 871
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.14346922934055328,
+      "learning_rate": 0.0004818256493412011,
+      "loss": 0.137,
+      "step": 872
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 0.07995035499334335,
+      "learning_rate": 0.00048089407417345296,
+      "loss": 0.1356,
+      "step": 873
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.14909203350543976,
+      "learning_rate": 0.0004799625654209675,
+      "loss": 0.1374,
+      "step": 874
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.06708569079637527,
+      "learning_rate": 0.00047903112632181904,
+      "loss": 0.1381,
+      "step": 875
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.22370800375938416,
+      "learning_rate": 0.00047809976011383906,
+      "loss": 0.1445,
+      "step": 876
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.05151727795600891,
+      "learning_rate": 0.0004771684700346059,
+      "loss": 0.1371,
+      "step": 877
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 0.12744151055812836,
+      "learning_rate": 0.0004762372593214335,
+      "loss": 0.1369,
+      "step": 878
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.13104400038719177,
+      "learning_rate": 0.0004753061312113592,
+      "loss": 0.1346,
+      "step": 879
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.11967893689870834,
+      "learning_rate": 0.00047437508894113416,
+      "loss": 0.1318,
+      "step": 880
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.035317592322826385,
+      "learning_rate": 0.00047344413574721046,
+      "loss": 0.1352,
+      "step": 881
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.15099988877773285,
+      "learning_rate": 0.0004725132748657307,
+      "loss": 0.1401,
+      "step": 882
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.24859024584293365,
+      "learning_rate": 0.0004715825095325168,
+      "loss": 0.1277,
+      "step": 883
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.11024681478738785,
+      "learning_rate": 0.00047065184298305797,
+      "loss": 0.1375,
+      "step": 884
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.031196558848023415,
+      "learning_rate": 0.00046972127845250084,
+      "loss": 0.133,
+      "step": 885
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.05172949284315109,
+      "learning_rate": 0.00046879081917563695,
+      "loss": 0.1324,
+      "step": 886
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.04595587030053139,
+      "learning_rate": 0.0004678604683868918,
+      "loss": 0.1361,
+      "step": 887
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.054625846445560455,
+      "learning_rate": 0.00046693022932031415,
+      "loss": 0.1334,
+      "step": 888
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.18956537544727325,
+      "learning_rate": 0.0004660001052095639,
+      "loss": 0.1419,
+      "step": 889
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 0.12293694168329239,
+      "learning_rate": 0.00046507009928790195,
+      "loss": 0.1234,
+      "step": 890
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.12140147387981415,
+      "learning_rate": 0.00046414021478817817,
+      "loss": 0.1282,
+      "step": 891
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.20622491836547852,
+      "learning_rate": 0.00046321045494282,
+      "loss": 0.1238,
+      "step": 892
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.1974942833185196,
+      "learning_rate": 0.00046228082298382196,
+      "loss": 0.1511,
+      "step": 893
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.13894042372703552,
+      "learning_rate": 0.0004613513221427337,
+      "loss": 0.1349,
+      "step": 894
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 0.023365622386336327,
+      "learning_rate": 0.00046042195565064914,
+      "loss": 0.1371,
+      "step": 895
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.16795076429843903,
+      "learning_rate": 0.0004594927267381958,
+      "loss": 0.1399,
+      "step": 896
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.06654185056686401,
+      "learning_rate": 0.00045856363863552195,
+      "loss": 0.1365,
+      "step": 897
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.21354056894779205,
+      "learning_rate": 0.00045763469457228695,
+      "loss": 0.1431,
+      "step": 898
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.1247892901301384,
+      "learning_rate": 0.0004567058977776498,
+      "loss": 0.1391,
+      "step": 899
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.1229679062962532,
+      "learning_rate": 0.00045577725148025647,
+      "loss": 0.1324,
+      "step": 900
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.0285334512591362,
+      "learning_rate": 0.000454848758908231,
+      "loss": 0.1417,
+      "step": 901
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.11522159725427628,
+      "learning_rate": 0.0004539204232891622,
+      "loss": 0.1349,
+      "step": 902
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.04999208077788353,
+      "learning_rate": 0.00045299224785009374,
+      "loss": 0.1395,
+      "step": 903
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.19387230277061462,
+      "learning_rate": 0.00045206423581751245,
+      "loss": 0.1367,
+      "step": 904
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.030587391927838326,
+      "learning_rate": 0.0004511363904173366,
+      "loss": 0.1392,
+      "step": 905
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.031090332195162773,
+      "learning_rate": 0.0004502087148749061,
+      "loss": 0.137,
+      "step": 906
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.07691047340631485,
+      "learning_rate": 0.0004492812124149696,
+      "loss": 0.144,
+      "step": 907
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.0668129101395607,
+      "learning_rate": 0.0004483538862616747,
+      "loss": 0.1337,
+      "step": 908
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.3449445962905884,
+      "learning_rate": 0.00044742673963855576,
+      "loss": 0.1526,
+      "step": 909
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.19746670126914978,
+      "learning_rate": 0.000446499775768523,
+      "loss": 0.1319,
+      "step": 910
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.11988267302513123,
+      "learning_rate": 0.0004455729978738517,
+      "loss": 0.1411,
+      "step": 911
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 0.17063240706920624,
+      "learning_rate": 0.00044464640917617063,
+      "loss": 0.1354,
+      "step": 912
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.26187554001808167,
+      "learning_rate": 0.00044372001289645044,
+      "loss": 0.136,
+      "step": 913
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.05965143442153931,
+      "learning_rate": 0.00044279381225499344,
+      "loss": 0.1398,
+      "step": 914
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.07176820188760757,
+      "learning_rate": 0.00044186781047142134,
+      "loss": 0.1388,
+      "step": 915
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.038787998259067535,
+      "learning_rate": 0.0004409420107646652,
+      "loss": 0.1383,
+      "step": 916
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.03987140208482742,
+      "learning_rate": 0.0004400164163529532,
+      "loss": 0.1366,
+      "step": 917
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 0.12179240584373474,
+      "learning_rate": 0.00043909103045379987,
+      "loss": 0.1306,
+      "step": 918
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.09804455190896988,
+      "learning_rate": 0.0004381658562839953,
+      "loss": 0.128,
+      "step": 919
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.08840085566043854,
+      "learning_rate": 0.00043724089705959304,
+      "loss": 0.1364,
+      "step": 920
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.18564368784427643,
+      "learning_rate": 0.00043631615599589964,
+      "loss": 0.1485,
+      "step": 921
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.3653159737586975,
+      "learning_rate": 0.00043539163630746384,
+      "loss": 0.1486,
+      "step": 922
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.08679798990488052,
+      "learning_rate": 0.000434467341208064,
+      "loss": 0.1291,
+      "step": 923
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.1024034321308136,
+      "learning_rate": 0.00043354327391069826,
+      "loss": 0.1275,
+      "step": 924
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.041372958570718765,
+      "learning_rate": 0.0004326194376275729,
+      "loss": 0.1328,
+      "step": 925
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.06509742885828018,
+      "learning_rate": 0.0004316958355700906,
+      "loss": 0.1324,
+      "step": 926
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.09408631920814514,
+      "learning_rate": 0.0004307724709488409,
+      "loss": 0.1405,
+      "step": 927
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.1963924914598465,
+      "learning_rate": 0.0004298493469735869,
+      "loss": 0.1436,
+      "step": 928
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 0.10209079831838608,
+      "learning_rate": 0.0004289264668532557,
+      "loss": 0.1277,
+      "step": 929
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.026920847594738007,
+      "learning_rate": 0.00042800383379592677,
+      "loss": 0.1295,
+      "step": 930
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.03551056608557701,
+      "learning_rate": 0.00042708145100882035,
+      "loss": 0.1281,
+      "step": 931
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.14194993674755096,
+      "learning_rate": 0.00042615932169828743,
+      "loss": 0.1398,
+      "step": 932
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.2725144326686859,
+      "learning_rate": 0.00042523744906979683,
+      "loss": 0.1217,
+      "step": 933
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.22893387079238892,
+      "learning_rate": 0.00042431583632792605,
+      "loss": 0.1517,
+      "step": 934
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 0.20985311269760132,
+      "learning_rate": 0.00042339448667634886,
+      "loss": 0.1433,
+      "step": 935
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.053967542946338654,
+      "learning_rate": 0.00042247340331782416,
+      "loss": 0.12,
+      "step": 936
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.22838272154331207,
+      "learning_rate": 0.0004215525894541856,
+      "loss": 0.1176,
+      "step": 937
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.3237338066101074,
+      "learning_rate": 0.0004206320482863301,
+      "loss": 0.1476,
+      "step": 938
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.09525377303361893,
+      "learning_rate": 0.0004197117830142062,
+      "loss": 0.1342,
+      "step": 939
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.05312574282288551,
+      "learning_rate": 0.0004187917968368036,
+      "loss": 0.1311,
+      "step": 940
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 0.11625714600086212,
+      "learning_rate": 0.00041787209295214177,
+      "loss": 0.1133,
+      "step": 941
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.04892723262310028,
+      "learning_rate": 0.000416952674557259,
+      "loss": 0.106,
+      "step": 942
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.44417399168014526,
+      "learning_rate": 0.00041603354484820134,
+      "loss": 0.1653,
+      "step": 943
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.07979090511798859,
+      "learning_rate": 0.00041511470702001074,
+      "loss": 0.1144,
+      "step": 944
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 3.163567304611206,
+      "learning_rate": 0.00041419616426671517,
+      "loss": 0.1576,
+      "step": 945
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.11929760128259659,
+      "learning_rate": 0.0004132779197813164,
+      "loss": 0.1329,
+      "step": 946
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.21021872758865356,
+      "learning_rate": 0.0004123599767557795,
+      "loss": 0.0956,
+      "step": 947
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.4803867042064667,
+      "learning_rate": 0.00041144233838102197,
+      "loss": 0.2027,
+      "step": 948
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.0795937329530716,
+      "learning_rate": 0.0004105250078469018,
+      "loss": 0.1226,
+      "step": 949
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.19914481043815613,
+      "learning_rate": 0.00040960798834220705,
+      "loss": 0.1457,
+      "step": 950
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 1.662695288658142,
+      "learning_rate": 0.00040869128305464475,
+      "loss": 0.1465,
+      "step": 951
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 0.1512700617313385,
+      "learning_rate": 0.00040777489517082924,
+      "loss": 0.1391,
+      "step": 952
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.3317195773124695,
+      "learning_rate": 0.00040685882787627227,
+      "loss": 0.1397,
+      "step": 953
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.2609153985977173,
+      "learning_rate": 0.00040594308435537026,
+      "loss": 0.1217,
+      "step": 954
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.40559151768684387,
+      "learning_rate": 0.00040502766779139485,
+      "loss": 0.1317,
+      "step": 955
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.21796320378780365,
+      "learning_rate": 0.0004041125813664808,
+      "loss": 0.1364,
+      "step": 956
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.6307505369186401,
+      "learning_rate": 0.0004031978282616151,
+      "loss": 0.1553,
+      "step": 957
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 0.28565332293510437,
+      "learning_rate": 0.00040228341165662683,
+      "loss": 0.1344,
+      "step": 958
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.4165158271789551,
+      "learning_rate": 0.0004013693347301741,
+      "loss": 0.1455,
+      "step": 959
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.821273922920227,
+      "learning_rate": 0.0004004556006597353,
+      "loss": 0.1287,
+      "step": 960
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.11364096403121948,
+      "learning_rate": 0.0003995422126215967,
+      "loss": 0.1177,
+      "step": 961
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.349627822637558,
+      "learning_rate": 0.0003986291737908414,
+      "loss": 0.1217,
+      "step": 962
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.10462171584367752,
+      "learning_rate": 0.0003977164873413391,
+      "loss": 0.1168,
+      "step": 963
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.11335984617471695,
+      "learning_rate": 0.0003968041564457342,
+      "loss": 0.1313,
+      "step": 964
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.37488850951194763,
+      "learning_rate": 0.0003958921842754351,
+      "loss": 0.133,
+      "step": 965
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.09337367117404938,
+      "learning_rate": 0.00039498057400060363,
+      "loss": 0.1464,
+      "step": 966
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.27405792474746704,
+      "learning_rate": 0.000394069328790143,
+      "loss": 0.1322,
+      "step": 967
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.5987095832824707,
+      "learning_rate": 0.00039315845181168784,
+      "loss": 0.1307,
+      "step": 968
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 0.3096538484096527,
+      "learning_rate": 0.00039224794623159294,
+      "loss": 0.1349,
+      "step": 969
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.5547122359275818,
+      "learning_rate": 0.0003913378152149214,
+      "loss": 0.1455,
+      "step": 970
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.4886229634284973,
+      "learning_rate": 0.0003904280619254348,
+      "loss": 0.1251,
+      "step": 971
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.18052807450294495,
+      "learning_rate": 0.0003895186895255814,
+      "loss": 0.1407,
+      "step": 972
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.09001462161540985,
+      "learning_rate": 0.0003886097011764855,
+      "loss": 0.1143,
+      "step": 973
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.3248112201690674,
+      "learning_rate": 0.0003877011000379367,
+      "loss": 0.1212,
+      "step": 974
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 0.11648620665073395,
+      "learning_rate": 0.000386792889268378,
+      "loss": 0.1167,
+      "step": 975
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 1.3816261291503906,
+      "learning_rate": 0.00038588507202489585,
+      "loss": 0.1518,
+      "step": 976
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.6389634013175964,
+      "learning_rate": 0.00038497765146320873,
+      "loss": 0.1372,
+      "step": 977
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.6133326888084412,
+      "learning_rate": 0.0003840706307376557,
+      "loss": 0.1252,
+      "step": 978
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.20734143257141113,
+      "learning_rate": 0.00038316401300118674,
+      "loss": 0.1115,
+      "step": 979
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 0.06368093192577362,
+      "learning_rate": 0.0003822578014053502,
+      "loss": 0.145,
+      "step": 980
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.07665737718343735,
+      "learning_rate": 0.0003813519991002831,
+      "loss": 0.1557,
+      "step": 981
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.09085717052221298,
+      "learning_rate": 0.00038044660923469963,
+      "loss": 0.1251,
+      "step": 982
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.08342912048101425,
+      "learning_rate": 0.00037954163495588,
+      "loss": 0.1256,
+      "step": 983
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.6068560481071472,
+      "learning_rate": 0.00037863707940966024,
+      "loss": 0.1506,
+      "step": 984
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.22720251977443695,
+      "learning_rate": 0.00037773294574042015,
+      "loss": 0.1151,
+      "step": 985
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.5528678297996521,
+      "learning_rate": 0.00037682923709107363,
+      "loss": 0.1436,
+      "step": 986
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.6791836619377136,
+      "learning_rate": 0.00037592595660305707,
+      "loss": 0.1405,
+      "step": 987
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.07115644961595535,
+      "learning_rate": 0.0003750231074163179,
+      "loss": 0.0997,
+      "step": 988
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.36401447653770447,
+      "learning_rate": 0.00037412069266930514,
+      "loss": 0.1471,
+      "step": 989
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.7831732630729675,
+      "learning_rate": 0.00037321871549895715,
+      "loss": 0.1314,
+      "step": 990
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.12779557704925537,
+      "learning_rate": 0.00037231717904069096,
+      "loss": 0.1446,
+      "step": 991
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 0.41478636860847473,
+      "learning_rate": 0.0003714160864283923,
+      "loss": 0.1429,
+      "step": 992
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.1117364913225174,
+      "learning_rate": 0.00037051544079440334,
+      "loss": 0.1148,
+      "step": 993
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.5920963287353516,
+      "learning_rate": 0.00036961524526951277,
+      "loss": 0.1204,
+      "step": 994
+    },
+    {
+      "epoch": 1.76,
+      "eval_loss": 0.13537168502807617,
+      "eval_runtime": 14.0251,
+      "eval_samples_per_second": 34.01,
+      "eval_steps_per_second": 8.556,
+      "step": 994
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.43262025713920593,
+      "learning_rate": 0.000368715502982945,
+      "loss": 0.1302,
+      "step": 995
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.10716990381479263,
+      "learning_rate": 0.00036781621706234816,
+      "loss": 0.133,
+      "step": 996
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.0918804183602333,
+      "learning_rate": 0.0003669173906337846,
+      "loss": 0.1354,
+      "step": 997
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.13421286642551422,
+      "learning_rate": 0.0003660190268217189,
+      "loss": 0.121,
+      "step": 998
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.15904036164283752,
+      "learning_rate": 0.00036512112874900797,
+      "loss": 0.1341,
+      "step": 999
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.3201177418231964,
+      "learning_rate": 0.00036422369953688973,
+      "loss": 0.1454,
+      "step": 1000
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.3361368477344513,
+      "learning_rate": 0.0003633267423049717,
+      "loss": 0.152,
+      "step": 1001
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.09073235839605331,
+      "learning_rate": 0.0003624302601712213,
+      "loss": 0.1331,
+      "step": 1002
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 0.2717398405075073,
+      "learning_rate": 0.0003615342562519542,
+      "loss": 0.1373,
+      "step": 1003
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.0666472539305687,
+      "learning_rate": 0.0003606387336618237,
+      "loss": 0.1496,
+      "step": 1004
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.15450868010520935,
+      "learning_rate": 0.0003597436955138102,
+      "loss": 0.1464,
+      "step": 1005
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.236429825425148,
+      "learning_rate": 0.0003588491449192096,
+      "loss": 0.1342,
+      "step": 1006
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.06421036273241043,
+      "learning_rate": 0.0003579550849876233,
+      "loss": 0.1352,
+      "step": 1007
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.07432877272367477,
+      "learning_rate": 0.00035706151882694727,
+      "loss": 0.131,
+      "step": 1008
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.10126742720603943,
+      "learning_rate": 0.00035616844954336046,
+      "loss": 0.1471,
+      "step": 1009
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.15761522948741913,
+      "learning_rate": 0.0003552758802413154,
+      "loss": 0.1358,
+      "step": 1010
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.02369426190853119,
+      "learning_rate": 0.0003543838140235257,
+      "loss": 0.1296,
+      "step": 1011
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.27005845308303833,
+      "learning_rate": 0.0003534922539909569,
+      "loss": 0.1412,
+      "step": 1012
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.0638512596487999,
+      "learning_rate": 0.00035260120324281474,
+      "loss": 0.1366,
+      "step": 1013
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 0.5134268403053284,
+      "learning_rate": 0.00035171066487653423,
+      "loss": 0.146,
+      "step": 1014
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.2569257915019989,
+      "learning_rate": 0.00035082064198776997,
+      "loss": 0.1507,
+      "step": 1015
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.16015255451202393,
+      "learning_rate": 0.0003499311376703842,
+      "loss": 0.1297,
+      "step": 1016
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.22499139606952667,
+      "learning_rate": 0.0003490421550164364,
+      "loss": 0.1357,
+      "step": 1017
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.5117542743682861,
+      "learning_rate": 0.0003481536971161732,
+      "loss": 0.1418,
+      "step": 1018
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.27242857217788696,
+      "learning_rate": 0.00034726576705801636,
+      "loss": 0.1358,
+      "step": 1019
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 0.3779907822608948,
+      "learning_rate": 0.0003463783679285535,
+      "loss": 0.1512,
+      "step": 1020
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.2148500680923462,
+      "learning_rate": 0.00034549150281252633,
+      "loss": 0.1445,
+      "step": 1021
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.08610748499631882,
+      "learning_rate": 0.0003446051747928202,
+      "loss": 0.1333,
+      "step": 1022
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.07877200841903687,
+      "learning_rate": 0.0003437193869504535,
+      "loss": 0.1333,
+      "step": 1023
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.1410919725894928,
+      "learning_rate": 0.0003428341423645668,
+      "loss": 0.1303,
+      "step": 1024
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.1181708350777626,
+      "learning_rate": 0.00034194944411241213,
+      "loss": 0.1234,
+      "step": 1025
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.13002073764801025,
+      "learning_rate": 0.00034106529526934303,
+      "loss": 0.1405,
+      "step": 1026
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.14099909365177155,
+      "learning_rate": 0.00034018169890880225,
+      "loss": 0.1342,
+      "step": 1027
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.24455852806568146,
+      "learning_rate": 0.00033929865810231264,
+      "loss": 0.1482,
+      "step": 1028
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.23751892149448395,
+      "learning_rate": 0.0003384161759194658,
+      "loss": 0.1509,
+      "step": 1029
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.40307796001434326,
+      "learning_rate": 0.00033753425542791104,
+      "loss": 0.1542,
+      "step": 1030
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 0.123422771692276,
+      "learning_rate": 0.0003366528996933458,
+      "loss": 0.129,
+      "step": 1031
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.15652324259281158,
+      "learning_rate": 0.00033577211177950386,
+      "loss": 0.1277,
+      "step": 1032
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.2987327575683594,
+      "learning_rate": 0.0003348918947481452,
+      "loss": 0.1395,
+      "step": 1033
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.2202194184064865,
+      "learning_rate": 0.00033401225165904556,
+      "loss": 0.1287,
+      "step": 1034
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.10783470422029495,
+      "learning_rate": 0.0003331331855699852,
+      "loss": 0.1423,
+      "step": 1035
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.26680612564086914,
+      "learning_rate": 0.0003322546995367394,
+      "loss": 0.1307,
+      "step": 1036
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 0.14138604700565338,
+      "learning_rate": 0.00033137679661306575,
+      "loss": 0.1273,
+      "step": 1037
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.2755129039287567,
+      "learning_rate": 0.0003304994798506962,
+      "loss": 0.1485,
+      "step": 1038
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.15637388825416565,
+      "learning_rate": 0.00032962275229932446,
+      "loss": 0.1233,
+      "step": 1039
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.10934972018003464,
+      "learning_rate": 0.00032874661700659587,
+      "loss": 0.1438,
+      "step": 1040
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.16785453259944916,
+      "learning_rate": 0.00032787107701809755,
+      "loss": 0.1257,
+      "step": 1041
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 0.2813141644001007,
+      "learning_rate": 0.0003269961353773469,
+      "loss": 0.1594,
+      "step": 1042
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.321236789226532,
+      "learning_rate": 0.00032612179512578126,
+      "loss": 0.1476,
+      "step": 1043
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.35957199335098267,
+      "learning_rate": 0.0003252480593027478,
+      "loss": 0.1614,
+      "step": 1044
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.2529314160346985,
+      "learning_rate": 0.0003243749309454922,
+      "loss": 0.1436,
+      "step": 1045
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.36090412735939026,
+      "learning_rate": 0.00032350241308914864,
+      "loss": 0.1578,
+      "step": 1046
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.47541487216949463,
+      "learning_rate": 0.0003226305087667295,
+      "loss": 0.1247,
+      "step": 1047
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.23453806340694427,
+      "learning_rate": 0.0003217592210091137,
+      "loss": 0.1435,
+      "step": 1048
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.12492989748716354,
+      "learning_rate": 0.0003208885528450376,
+      "loss": 0.1229,
+      "step": 1049
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.19712020456790924,
+      "learning_rate": 0.00032001850730108307,
+      "loss": 0.1292,
+      "step": 1050
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.09731408208608627,
+      "learning_rate": 0.00031914908740166795,
+      "loss": 0.1333,
+      "step": 1051
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.06944354623556137,
+      "learning_rate": 0.0003182802961690357,
+      "loss": 0.1292,
+      "step": 1052
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.07448045909404755,
+      "learning_rate": 0.00031741213662324363,
+      "loss": 0.1349,
+      "step": 1053
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 0.28523683547973633,
+      "learning_rate": 0.0003165446117821538,
+      "loss": 0.1452,
+      "step": 1054
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.09108186513185501,
+      "learning_rate": 0.0003156777246614215,
+      "loss": 0.1361,
+      "step": 1055
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.13375020027160645,
+      "learning_rate": 0.0003148114782744855,
+      "loss": 0.1381,
+      "step": 1056
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.06716307252645493,
+      "learning_rate": 0.00031394587563255755,
+      "loss": 0.1383,
+      "step": 1057
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.11596639454364777,
+      "learning_rate": 0.00031308091974461064,
+      "loss": 0.1313,
+      "step": 1058
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 0.39337942004203796,
+      "learning_rate": 0.00031221661361737065,
+      "loss": 0.1359,
+      "step": 1059
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.07525162398815155,
+      "learning_rate": 0.00031135296025530424,
+      "loss": 0.1326,
+      "step": 1060
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.020530417561531067,
+      "learning_rate": 0.0003104899626606088,
+      "loss": 0.1368,
+      "step": 1061
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.17400570213794708,
+      "learning_rate": 0.00030962762383320285,
+      "loss": 0.1309,
+      "step": 1062
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.1574063003063202,
+      "learning_rate": 0.00030876594677071404,
+      "loss": 0.1365,
+      "step": 1063
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.21009187400341034,
+      "learning_rate": 0.0003079049344684702,
+      "loss": 0.1382,
+      "step": 1064
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 0.047014713287353516,
+      "learning_rate": 0.00030704458991948844,
+      "loss": 0.1327,
+      "step": 1065
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.11032029241323471,
+      "learning_rate": 0.0003061849161144641,
+      "loss": 0.1252,
+      "step": 1066
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.08818018436431885,
+      "learning_rate": 0.0003053259160417613,
+      "loss": 0.1495,
+      "step": 1067
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.08112979680299759,
+      "learning_rate": 0.0003044675926874023,
+      "loss": 0.1408,
+      "step": 1068
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.11870189756155014,
+      "learning_rate": 0.00030360994903505653,
+      "loss": 0.1342,
+      "step": 1069
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.16511432826519012,
+      "learning_rate": 0.000302752988066031,
+      "loss": 0.148,
+      "step": 1070
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.31104427576065063,
+      "learning_rate": 0.0003018967127592595,
+      "loss": 0.15,
+      "step": 1071
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.1434139758348465,
+      "learning_rate": 0.0003010411260912922,
+      "loss": 0.142,
+      "step": 1072
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.29530733823776245,
+      "learning_rate": 0.00030018623103628594,
+      "loss": 0.1284,
+      "step": 1073
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.06294587254524231,
+      "learning_rate": 0.00029933203056599274,
+      "loss": 0.1379,
+      "step": 1074
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.0581539049744606,
+      "learning_rate": 0.0002984785276497507,
+      "loss": 0.1278,
+      "step": 1075
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 0.09157228469848633,
+      "learning_rate": 0.0002976257252544726,
+      "loss": 0.1348,
+      "step": 1076
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.10196894407272339,
+      "learning_rate": 0.00029677362634463643,
+      "loss": 0.1386,
+      "step": 1077
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.1152050793170929,
+      "learning_rate": 0.00029592223388227504,
+      "loss": 0.1337,
+      "step": 1078
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.34997934103012085,
+      "learning_rate": 0.0002950715508269648,
+      "loss": 0.1346,
+      "step": 1079
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.21548382937908173,
+      "learning_rate": 0.00029422158013581656,
+      "loss": 0.1309,
+      "step": 1080
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.3693360686302185,
+      "learning_rate": 0.000293372324763465,
+      "loss": 0.1587,
+      "step": 1081
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 0.2655669152736664,
+      "learning_rate": 0.0002925237876620576,
+      "loss": 0.1285,
+      "step": 1082
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.08570755273103714,
+      "learning_rate": 0.00029167597178124583,
+      "loss": 0.1359,
+      "step": 1083
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.2622168958187103,
+      "learning_rate": 0.00029082888006817364,
+      "loss": 0.1315,
+      "step": 1084
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.020985718816518784,
+      "learning_rate": 0.0002899825154674674,
+      "loss": 0.1308,
+      "step": 1085
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.15370753407478333,
+      "learning_rate": 0.00028913688092122665,
+      "loss": 0.1269,
+      "step": 1086
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.1030414029955864,
+      "learning_rate": 0.0002882919793690123,
+      "loss": 0.1143,
+      "step": 1087
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.31633636355400085,
+      "learning_rate": 0.00028744781374783813,
+      "loss": 0.1563,
+      "step": 1088
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.03694160282611847,
+      "learning_rate": 0.00028660438699215895,
+      "loss": 0.1276,
+      "step": 1089
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.14471565186977386,
+      "learning_rate": 0.0002857617020338614,
+      "loss": 0.137,
+      "step": 1090
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.08222481608390808,
+      "learning_rate": 0.0002849197618022539,
+      "loss": 0.1279,
+      "step": 1091
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.11603690683841705,
+      "learning_rate": 0.00028407856922405526,
+      "loss": 0.138,
+      "step": 1092
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.18144792318344116,
+      "learning_rate": 0.0002832381272233864,
+      "loss": 0.1481,
+      "step": 1093
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.05054265260696411,
+      "learning_rate": 0.00028239843872175814,
+      "loss": 0.1363,
+      "step": 1094
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 2.8396074771881104,
+      "learning_rate": 0.00028155950663806236,
+      "loss": 0.1345,
+      "step": 1095
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.30984073877334595,
+      "learning_rate": 0.0002807213338885619,
+      "loss": 0.1338,
+      "step": 1096
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.34585434198379517,
+      "learning_rate": 0.00027988392338687925,
+      "loss": 0.1389,
+      "step": 1097
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.0312834158539772,
+      "learning_rate": 0.0002790472780439881,
+      "loss": 0.1338,
+      "step": 1098
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 0.5297988057136536,
+      "learning_rate": 0.0002782114007682016,
+      "loss": 0.1321,
+      "step": 1099
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.10174748301506042,
+      "learning_rate": 0.0002773762944651632,
+      "loss": 0.1202,
+      "step": 1100
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.0418776273727417,
+      "learning_rate": 0.0002765419620378366,
+      "loss": 0.1257,
+      "step": 1101
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.3508782386779785,
+      "learning_rate": 0.0002757084063864949,
+      "loss": 0.1612,
+      "step": 1102
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.2867041230201721,
+      "learning_rate": 0.00027487563040871145,
+      "loss": 0.1574,
+      "step": 1103
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.24160172045230865,
+      "learning_rate": 0.00027404363699934907,
+      "loss": 0.1175,
+      "step": 1104
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 0.03382538631558418,
+      "learning_rate": 0.0002732124290505501,
+      "loss": 0.1363,
+      "step": 1105
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.05089818313717842,
+      "learning_rate": 0.000272382009451727,
+      "loss": 0.1384,
+      "step": 1106
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.08688928186893463,
+      "learning_rate": 0.0002715523810895515,
+      "loss": 0.1415,
+      "step": 1107
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.03926026448607445,
+      "learning_rate": 0.00027072354684794486,
+      "loss": 0.1357,
+      "step": 1108
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.05058757960796356,
+      "learning_rate": 0.0002698955096080677,
+      "loss": 0.1356,
+      "step": 1109
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.0489107221364975,
+      "learning_rate": 0.00026906827224831023,
+      "loss": 0.1394,
+      "step": 1110
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.09765997529029846,
+      "learning_rate": 0.00026824183764428223,
+      "loss": 0.1388,
+      "step": 1111
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.06232646107673645,
+      "learning_rate": 0.00026741620866880335,
+      "loss": 0.1322,
+      "step": 1112
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.08469201624393463,
+      "learning_rate": 0.0002665913881918921,
+      "loss": 0.1369,
+      "step": 1113
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.08312228322029114,
+      "learning_rate": 0.00026576737908075667,
+      "loss": 0.137,
+      "step": 1114
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.13206493854522705,
+      "learning_rate": 0.00026494418419978485,
+      "loss": 0.127,
+      "step": 1115
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 0.050156012177467346,
+      "learning_rate": 0.0002641218064105341,
+      "loss": 0.1304,
+      "step": 1116
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.22207175195217133,
+      "learning_rate": 0.0002633002485717219,
+      "loss": 0.1406,
+      "step": 1117
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.25118494033813477,
+      "learning_rate": 0.0002624795135392148,
+      "loss": 0.1373,
+      "step": 1118
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.09831628948450089,
+      "learning_rate": 0.00026165960416601943,
+      "loss": 0.1459,
+      "step": 1119
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.037262722849845886,
+      "learning_rate": 0.00026084052330227237,
+      "loss": 0.1314,
+      "step": 1120
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.05748564377427101,
+      "learning_rate": 0.0002600222737952299,
+      "loss": 0.1439,
+      "step": 1121
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 0.06702205538749695,
+      "learning_rate": 0.00025920485848925914,
+      "loss": 0.1265,
+      "step": 1122
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.11122670769691467,
+      "learning_rate": 0.00025838828022582596,
+      "loss": 0.1275,
+      "step": 1123
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.1843162626028061,
+      "learning_rate": 0.0002575725418434878,
+      "loss": 0.1419,
+      "step": 1124
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.11396101117134094,
+      "learning_rate": 0.00025675764617788234,
+      "loss": 0.1342,
+      "step": 1125
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.043184638023376465,
+      "learning_rate": 0.00025594359606171725,
+      "loss": 0.1295,
+      "step": 1126
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 0.15845736861228943,
+      "learning_rate": 0.0002551303943247619,
+      "loss": 0.1429,
+      "step": 1127
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.1975352019071579,
+      "learning_rate": 0.0002543180437938352,
+      "loss": 0.1153,
+      "step": 1128
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.13739655911922455,
+      "learning_rate": 0.00025350654729279834,
+      "loss": 0.1347,
+      "step": 1129
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.2761506140232086,
+      "learning_rate": 0.0002526959076425434,
+      "loss": 0.147,
+      "step": 1130
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.09016864746809006,
+      "learning_rate": 0.0002518861276609837,
+      "loss": 0.1298,
+      "step": 1131
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.15213081240653992,
+      "learning_rate": 0.00025107721016304424,
+      "loss": 0.126,
+      "step": 1132
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.10915911197662354,
+      "learning_rate": 0.00025026915796065233,
+      "loss": 0.1533,
+      "step": 1133
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.21597544848918915,
+      "learning_rate": 0.0002494619738627266,
+      "loss": 0.1352,
+      "step": 1134
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.14955481886863708,
+      "learning_rate": 0.00024865566067516895,
+      "loss": 0.115,
+      "step": 1135
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.07370211184024811,
+      "learning_rate": 0.000247850221200853,
+      "loss": 0.1262,
+      "step": 1136
+    },
+    {
+      "epoch": 2.01,
+      "eval_loss": 0.13425856828689575,
+      "eval_runtime": 14.0067,
+      "eval_samples_per_second": 34.055,
+      "eval_steps_per_second": 8.567,
+      "step": 1136
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.22561423480510712,
+      "learning_rate": 0.00024704565823961564,
+      "loss": 0.1442,
+      "step": 1137
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 0.1496383398771286,
+      "learning_rate": 0.0002462419745882469,
+      "loss": 0.1367,
+      "step": 1138
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.1709650307893753,
+      "learning_rate": 0.00024543917304047986,
+      "loss": 0.1334,
+      "step": 1139
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.2110009342432022,
+      "learning_rate": 0.0002446372563869818,
+      "loss": 0.1387,
+      "step": 1140
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.26138541102409363,
+      "learning_rate": 0.00024383622741534343,
+      "loss": 0.1366,
+      "step": 1141
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.13308700919151306,
+      "learning_rate": 0.00024303608891006984,
+      "loss": 0.1272,
+      "step": 1142
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.31121209263801575,
+      "learning_rate": 0.0002422368436525711,
+      "loss": 0.158,
+      "step": 1143
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 0.09018420428037643,
+      "learning_rate": 0.00024143849442115158,
+      "loss": 0.1333,
+      "step": 1144
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.19516992568969727,
+      "learning_rate": 0.00024064104399100167,
+      "loss": 0.1276,
+      "step": 1145
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.22115693986415863,
+      "learning_rate": 0.00023984449513418687,
+      "loss": 0.1184,
+      "step": 1146
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.08551298081874847,
+      "learning_rate": 0.00023904885061963843,
+      "loss": 0.1318,
+      "step": 1147
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.27814337611198425,
+      "learning_rate": 0.00023825411321314489,
+      "loss": 0.1454,
+      "step": 1148
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.39333808422088623,
+      "learning_rate": 0.0002374602856773404,
+      "loss": 0.1491,
+      "step": 1149
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 0.1820652335882187,
+      "learning_rate": 0.00023666737077169726,
+      "loss": 0.1464,
+      "step": 1150
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.21041136980056763,
+      "learning_rate": 0.00023587537125251468,
+      "loss": 0.1437,
+      "step": 1151
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.22204861044883728,
+      "learning_rate": 0.00023508428987290987,
+      "loss": 0.1375,
+      "step": 1152
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.1012771725654602,
+      "learning_rate": 0.00023429412938280898,
+      "loss": 0.1271,
+      "step": 1153
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.23188555240631104,
+      "learning_rate": 0.0002335048925289362,
+      "loss": 0.1254,
+      "step": 1154
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.22195938229560852,
+      "learning_rate": 0.00023271658205480588,
+      "loss": 0.1374,
+      "step": 1155
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.24288509786128998,
+      "learning_rate": 0.00023192920070071144,
+      "loss": 0.1386,
+      "step": 1156
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.0804426372051239,
+      "learning_rate": 0.00023114275120371657,
+      "loss": 0.128,
+      "step": 1157
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.2349734604358673,
+      "learning_rate": 0.00023035723629764615,
+      "loss": 0.139,
+      "step": 1158
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.21826253831386566,
+      "learning_rate": 0.0002295726587130761,
+      "loss": 0.1483,
+      "step": 1159
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.12499336153268814,
+      "learning_rate": 0.0002287890211773238,
+      "loss": 0.1446,
+      "step": 1160
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 0.22501634061336517,
+      "learning_rate": 0.00022800632641443902,
+      "loss": 0.1288,
+      "step": 1161
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.123602956533432,
+      "learning_rate": 0.00022722457714519418,
+      "loss": 0.1287,
+      "step": 1162
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.1110440120100975,
+      "learning_rate": 0.00022644377608707522,
+      "loss": 0.1375,
+      "step": 1163
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.10080868005752563,
+      "learning_rate": 0.00022566392595427216,
+      "loss": 0.1236,
+      "step": 1164
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.11740677803754807,
+      "learning_rate": 0.00022488502945766892,
+      "loss": 0.1362,
+      "step": 1165
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.08558549731969833,
+      "learning_rate": 0.00022410708930483463,
+      "loss": 0.1525,
+      "step": 1166
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 0.4037264883518219,
+      "learning_rate": 0.00022333010820001394,
+      "loss": 0.1366,
+      "step": 1167
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.12114354968070984,
+      "learning_rate": 0.00022255408884411793,
+      "loss": 0.1386,
+      "step": 1168
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.13077768683433533,
+      "learning_rate": 0.0002217790339347146,
+      "loss": 0.1281,
+      "step": 1169
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.15883688628673553,
+      "learning_rate": 0.0002210049461660189,
+      "loss": 0.129,
+      "step": 1170
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.1601915806531906,
+      "learning_rate": 0.00022023182822888398,
+      "loss": 0.1119,
+      "step": 1171
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.17367267608642578,
+      "learning_rate": 0.0002194596828107921,
+      "loss": 0.147,
+      "step": 1172
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.2555027902126312,
+      "learning_rate": 0.00021868851259584427,
+      "loss": 0.1495,
+      "step": 1173
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.10203690081834793,
+      "learning_rate": 0.00021791832026475238,
+      "loss": 0.1348,
+      "step": 1174
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.19506026804447174,
+      "learning_rate": 0.00021714910849482776,
+      "loss": 0.1332,
+      "step": 1175
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.1914072483778,
+      "learning_rate": 0.00021638087995997442,
+      "loss": 0.1545,
+      "step": 1176
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.1700693666934967,
+      "learning_rate": 0.00021561363733067795,
+      "loss": 0.1435,
+      "step": 1177
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.11551807075738907,
+      "learning_rate": 0.00021484738327399682,
+      "loss": 0.1335,
+      "step": 1178
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.039073869585990906,
+      "learning_rate": 0.0002140821204535529,
+      "loss": 0.1292,
+      "step": 1179
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.1664426028728485,
+      "learning_rate": 0.00021331785152952244,
+      "loss": 0.1322,
+      "step": 1180
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.10394078493118286,
+      "learning_rate": 0.00021255457915862692,
+      "loss": 0.1355,
+      "step": 1181
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.23373393714427948,
+      "learning_rate": 0.00021179230599412374,
+      "loss": 0.1345,
+      "step": 1182
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.19787871837615967,
+      "learning_rate": 0.00021103103468579653,
+      "loss": 0.1341,
+      "step": 1183
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 0.1043769121170044,
+      "learning_rate": 0.00021027076787994632,
+      "loss": 0.1305,
+      "step": 1184
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.14168131351470947,
+      "learning_rate": 0.00020951150821938276,
+      "loss": 0.1399,
+      "step": 1185
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.2587750554084778,
+      "learning_rate": 0.0002087532583434139,
+      "loss": 0.1381,
+      "step": 1186
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.13129281997680664,
+      "learning_rate": 0.00020799602088783837,
+      "loss": 0.1191,
+      "step": 1187
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.1005392074584961,
+      "learning_rate": 0.00020723979848493473,
+      "loss": 0.1342,
+      "step": 1188
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 0.10267619788646698,
+      "learning_rate": 0.0002064845937634533,
+      "loss": 0.1355,
+      "step": 1189
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.06988910585641861,
+      "learning_rate": 0.00020573040934860715,
+      "loss": 0.1301,
+      "step": 1190
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.10177844017744064,
+      "learning_rate": 0.00020497724786206183,
+      "loss": 0.1309,
+      "step": 1191
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.10639530420303345,
+      "learning_rate": 0.000204225111921928,
+      "loss": 0.1201,
+      "step": 1192
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.176835834980011,
+      "learning_rate": 0.00020347400414275058,
+      "loss": 0.134,
+      "step": 1193
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.0938374251127243,
+      "learning_rate": 0.00020272392713550048,
+      "loss": 0.1424,
+      "step": 1194
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.0616929791867733,
+      "learning_rate": 0.00020197488350756615,
+      "loss": 0.1422,
+      "step": 1195
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.07180429995059967,
+      "learning_rate": 0.00020122687586274297,
+      "loss": 0.1355,
+      "step": 1196
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.1905076801776886,
+      "learning_rate": 0.00020047990680122584,
+      "loss": 0.1155,
+      "step": 1197
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.3793567717075348,
+      "learning_rate": 0.00019973397891959893,
+      "loss": 0.1582,
+      "step": 1198
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.27994590997695923,
+      "learning_rate": 0.000198989094810827,
+      "loss": 0.1186,
+      "step": 1199
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.0879906564950943,
+      "learning_rate": 0.000198245257064247,
+      "loss": 0.1313,
+      "step": 1200
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 0.20328675210475922,
+      "learning_rate": 0.000197502468265558,
+      "loss": 0.1494,
+      "step": 1201
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.12098411470651627,
+      "learning_rate": 0.00019676073099681335,
+      "loss": 0.1201,
+      "step": 1202
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.1231808289885521,
+      "learning_rate": 0.00019602004783641054,
+      "loss": 0.1374,
+      "step": 1203
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.14193227887153625,
+      "learning_rate": 0.00019528042135908292,
+      "loss": 0.14,
+      "step": 1204
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.2550235688686371,
+      "learning_rate": 0.0001945418541358911,
+      "loss": 0.1289,
+      "step": 1205
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 0.05161747708916664,
+      "learning_rate": 0.00019380434873421292,
+      "loss": 0.1208,
+      "step": 1206
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.40697818994522095,
+      "learning_rate": 0.00019306790771773575,
+      "loss": 0.107,
+      "step": 1207
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.09957734495401382,
+      "learning_rate": 0.00019233253364644653,
+      "loss": 0.1258,
+      "step": 1208
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.15506863594055176,
+      "learning_rate": 0.00019159822907662333,
+      "loss": 0.137,
+      "step": 1209
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.11825218796730042,
+      "learning_rate": 0.00019086499656082684,
+      "loss": 0.1299,
+      "step": 1210
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.3029904365539551,
+      "learning_rate": 0.00019013283864789104,
+      "loss": 0.1136,
+      "step": 1211
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.11233114451169968,
+      "learning_rate": 0.00018940175788291407,
+      "loss": 0.125,
+      "step": 1212
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.16029077768325806,
+      "learning_rate": 0.00018867175680725002,
+      "loss": 0.1713,
+      "step": 1213
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.1274372637271881,
+      "learning_rate": 0.0001879428379584995,
+      "loss": 0.0996,
+      "step": 1214
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.1937287449836731,
+      "learning_rate": 0.0001872150038705015,
+      "loss": 0.1246,
+      "step": 1215
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.1734904944896698,
+      "learning_rate": 0.00018648825707332433,
+      "loss": 0.1698,
+      "step": 1216
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.31292834877967834,
+      "learning_rate": 0.0001857626000932562,
+      "loss": 0.1284,
+      "step": 1217
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.11059171706438065,
+      "learning_rate": 0.0001850380354527972,
+      "loss": 0.1039,
+      "step": 1218
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.12312529236078262,
+      "learning_rate": 0.00018431456567065003,
+      "loss": 0.1436,
+      "step": 1219
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.2583165168762207,
+      "learning_rate": 0.0001835921932617119,
+      "loss": 0.113,
+      "step": 1220
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.2463943064212799,
+      "learning_rate": 0.00018287092073706557,
+      "loss": 0.1373,
+      "step": 1221
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.23069016635417938,
+      "learning_rate": 0.0001821507506039693,
+      "loss": 0.1473,
+      "step": 1222
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 0.057799965143203735,
+      "learning_rate": 0.0001814316853658503,
+      "loss": 0.1089,
+      "step": 1223
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.14288663864135742,
+      "learning_rate": 0.00018071372752229497,
+      "loss": 0.1168,
+      "step": 1224
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.1829814910888672,
+      "learning_rate": 0.00017999687956903954,
+      "loss": 0.1232,
+      "step": 1225
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.07779528200626373,
+      "learning_rate": 0.00017928114399796296,
+      "loss": 0.1319,
+      "step": 1226
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.12610167264938354,
+      "learning_rate": 0.00017856652329707623,
+      "loss": 0.1499,
+      "step": 1227
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.08324690163135529,
+      "learning_rate": 0.00017785301995051588,
+      "loss": 0.1164,
+      "step": 1228
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 0.16664673388004303,
+      "learning_rate": 0.00017714063643853423,
+      "loss": 0.1102,
+      "step": 1229
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.47252926230430603,
+      "learning_rate": 0.00017642937523749036,
+      "loss": 0.1267,
+      "step": 1230
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.24932174384593964,
+      "learning_rate": 0.00017571923881984236,
+      "loss": 0.0999,
+      "step": 1231
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.1122673973441124,
+      "learning_rate": 0.0001750102296541382,
+      "loss": 0.1149,
+      "step": 1232
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.03139616549015045,
+      "learning_rate": 0.00017430235020500756,
+      "loss": 0.14,
+      "step": 1233
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.1314571052789688,
+      "learning_rate": 0.0001735956029331532,
+      "loss": 0.1316,
+      "step": 1234
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 0.12405366450548172,
+      "learning_rate": 0.00017288999029534176,
+      "loss": 0.1066,
+      "step": 1235
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.12566903233528137,
+      "learning_rate": 0.00017218551474439586,
+      "loss": 0.1217,
+      "step": 1236
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.136549711227417,
+      "learning_rate": 0.0001714821787291858,
+      "loss": 0.1242,
+      "step": 1237
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.049642208963632584,
+      "learning_rate": 0.00017077998469462009,
+      "loss": 0.1343,
+      "step": 1238
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.2866723835468292,
+      "learning_rate": 0.0001700789350816382,
+      "loss": 0.1365,
+      "step": 1239
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.12442098557949066,
+      "learning_rate": 0.00016937903232720075,
+      "loss": 0.1512,
+      "step": 1240
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.11928176134824753,
+      "learning_rate": 0.00016868027886428194,
+      "loss": 0.147,
+      "step": 1241
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.09463697671890259,
+      "learning_rate": 0.00016798267712186122,
+      "loss": 0.117,
+      "step": 1242
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.06501590460538864,
+      "learning_rate": 0.0001672862295249138,
+      "loss": 0.1508,
+      "step": 1243
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.30164363980293274,
+      "learning_rate": 0.00016659093849440354,
+      "loss": 0.1501,
+      "step": 1244
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.15526002645492554,
+      "learning_rate": 0.00016589680644727345,
+      "loss": 0.1242,
+      "step": 1245
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.07917926460504532,
+      "learning_rate": 0.00016520383579643767,
+      "loss": 0.1314,
+      "step": 1246
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.10616712272167206,
+      "learning_rate": 0.00016451202895077378,
+      "loss": 0.1462,
+      "step": 1247
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.1220119372010231,
+      "learning_rate": 0.0001638213883151129,
+      "loss": 0.1191,
+      "step": 1248
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.10847750306129456,
+      "learning_rate": 0.0001631319162902331,
+      "loss": 0.0983,
+      "step": 1249
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.03508616238832474,
+      "learning_rate": 0.0001624436152728495,
+      "loss": 0.1238,
+      "step": 1250
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.05642473325133324,
+      "learning_rate": 0.0001617564876556067,
+      "loss": 0.1381,
+      "step": 1251
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 0.12747938930988312,
+      "learning_rate": 0.0001610705358270708,
+      "loss": 0.1165,
+      "step": 1252
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.030638879165053368,
+      "learning_rate": 0.0001603857621717202,
+      "loss": 0.1415,
+      "step": 1253
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.08363982290029526,
+      "learning_rate": 0.00015970216906993818,
+      "loss": 0.1193,
+      "step": 1254
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.04466724768280983,
+      "learning_rate": 0.00015901975889800386,
+      "loss": 0.1214,
+      "step": 1255
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.1267317682504654,
+      "learning_rate": 0.00015833853402808434,
+      "loss": 0.1391,
+      "step": 1256
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.18954621255397797,
+      "learning_rate": 0.00015765849682822686,
+      "loss": 0.1665,
+      "step": 1257
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.10698127746582031,
+      "learning_rate": 0.00015697964966234946,
+      "loss": 0.1019,
+      "step": 1258
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.12424101680517197,
+      "learning_rate": 0.00015630199489023415,
+      "loss": 0.1591,
+      "step": 1259
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.0653344914317131,
+      "learning_rate": 0.0001556255348675174,
+      "loss": 0.1536,
+      "step": 1260
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.04556593671441078,
+      "learning_rate": 0.0001549502719456827,
+      "loss": 0.1502,
+      "step": 1261
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.10966439545154572,
+      "learning_rate": 0.00015427620847205238,
+      "loss": 0.1322,
+      "step": 1262
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 0.21205949783325195,
+      "learning_rate": 0.00015360334678977933,
+      "loss": 0.1374,
+      "step": 1263
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.3078778088092804,
+      "learning_rate": 0.00015293168923783857,
+      "loss": 0.1303,
+      "step": 1264
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.22372427582740784,
+      "learning_rate": 0.0001522612381510195,
+      "loss": 0.1334,
+      "step": 1265
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.07522010058164597,
+      "learning_rate": 0.00015159199585991744,
+      "loss": 0.1315,
+      "step": 1266
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.10713458061218262,
+      "learning_rate": 0.00015092396469092617,
+      "loss": 0.1181,
+      "step": 1267
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.14708364009857178,
+      "learning_rate": 0.00015025714696622933,
+      "loss": 0.1246,
+      "step": 1268
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.050081461668014526,
+      "learning_rate": 0.00014959154500379212,
+      "loss": 0.1329,
+      "step": 1269
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.15830601751804352,
+      "learning_rate": 0.00014892716111735376,
+      "loss": 0.1372,
+      "step": 1270
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.0707254484295845,
+      "learning_rate": 0.00014826399761641907,
+      "loss": 0.1235,
+      "step": 1271
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.1088636964559555,
+      "learning_rate": 0.00014760205680625083,
+      "loss": 0.1439,
+      "step": 1272
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.07059821486473083,
+      "learning_rate": 0.00014694134098786182,
+      "loss": 0.1359,
+      "step": 1273
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 0.2451217770576477,
+      "learning_rate": 0.0001462818524580057,
+      "loss": 0.1592,
+      "step": 1274
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.08877355605363846,
+      "learning_rate": 0.00014562359350917055,
+      "loss": 0.1267,
+      "step": 1275
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.09591105580329895,
+      "learning_rate": 0.00014496656642957025,
+      "loss": 0.1461,
+      "step": 1276
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.12581078708171844,
+      "learning_rate": 0.0001443107735031361,
+      "loss": 0.1357,
+      "step": 1277
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.04613172635436058,
+      "learning_rate": 0.00014365621700950987,
+      "loss": 0.12,
+      "step": 1278
+    },
+    {
+      "epoch": 2.26,
+      "eval_loss": 0.13804392516613007,
+      "eval_runtime": 14.0836,
+      "eval_samples_per_second": 33.869,
+      "eval_steps_per_second": 8.521,
+      "step": 1278
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.0844263806939125,
+      "learning_rate": 0.0001430028992240344,
+      "loss": 0.1243,
+      "step": 1279
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.22574763000011444,
+      "learning_rate": 0.0001423508224177474,
+      "loss": 0.1461,
+      "step": 1280
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.2293427437543869,
+      "learning_rate": 0.00014169998885737257,
+      "loss": 0.1612,
+      "step": 1281
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.05698062479496002,
+      "learning_rate": 0.00014105040080531161,
+      "loss": 0.1343,
+      "step": 1282
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.13387730717658997,
+      "learning_rate": 0.00014040206051963678,
+      "loss": 0.1132,
+      "step": 1283
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.24633343517780304,
+      "learning_rate": 0.00013975497025408285,
+      "loss": 0.1122,
+      "step": 1284
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 0.03243758901953697,
+      "learning_rate": 0.00013910913225803945,
+      "loss": 0.1229,
+      "step": 1285
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.10440926253795624,
+      "learning_rate": 0.00013846454877654318,
+      "loss": 0.1347,
+      "step": 1286
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.1835152506828308,
+      "learning_rate": 0.00013782122205026958,
+      "loss": 0.1554,
+      "step": 1287
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.02530871145427227,
+      "learning_rate": 0.0001371791543155253,
+      "loss": 0.1306,
+      "step": 1288
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.02593812346458435,
+      "learning_rate": 0.00013653834780424112,
+      "loss": 0.1425,
+      "step": 1289
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.13932591676712036,
+      "learning_rate": 0.000135898804743963,
+      "loss": 0.1314,
+      "step": 1290
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 0.06134510040283203,
+      "learning_rate": 0.00013526052735784538,
+      "loss": 0.1215,
+      "step": 1291
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.12832637131214142,
+      "learning_rate": 0.00013462351786464283,
+      "loss": 0.1293,
+      "step": 1292
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.1946830302476883,
+      "learning_rate": 0.00013398777847870236,
+      "loss": 0.1383,
+      "step": 1293
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.07782842963933945,
+      "learning_rate": 0.00013335331140995626,
+      "loss": 0.1397,
+      "step": 1294
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.12014558911323547,
+      "learning_rate": 0.00013272011886391366,
+      "loss": 0.1342,
+      "step": 1295
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.3111260235309601,
+      "learning_rate": 0.00013208820304165365,
+      "loss": 0.1425,
+      "step": 1296
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 0.10395296663045883,
+      "learning_rate": 0.0001314575661398168,
+      "loss": 0.1355,
+      "step": 1297
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.14676041901111603,
+      "learning_rate": 0.00013082821035059812,
+      "loss": 0.1315,
+      "step": 1298
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.1285468190908432,
+      "learning_rate": 0.00013020013786173946,
+      "loss": 0.1381,
+      "step": 1299
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.1933162659406662,
+      "learning_rate": 0.0001295733508565213,
+      "loss": 0.1389,
+      "step": 1300
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.10214361548423767,
+      "learning_rate": 0.0001289478515137561,
+      "loss": 0.1317,
+      "step": 1301
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.07621518522500992,
+      "learning_rate": 0.0001283236420077798,
+      "loss": 0.1316,
+      "step": 1302
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.19626323878765106,
+      "learning_rate": 0.0001277007245084446,
+      "loss": 0.1365,
+      "step": 1303
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.14391857385635376,
+      "learning_rate": 0.00012707910118111193,
+      "loss": 0.1284,
+      "step": 1304
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.346746563911438,
+      "learning_rate": 0.0001264587741866439,
+      "loss": 0.1354,
+      "step": 1305
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.14903244376182556,
+      "learning_rate": 0.00012583974568139699,
+      "loss": 0.1351,
+      "step": 1306
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.11248177289962769,
+      "learning_rate": 0.00012522201781721336,
+      "loss": 0.1264,
+      "step": 1307
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 0.16379296779632568,
+      "learning_rate": 0.00012460559274141407,
+      "loss": 0.1302,
+      "step": 1308
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.07437599450349808,
+      "learning_rate": 0.00012399047259679182,
+      "loss": 0.1368,
+      "step": 1309
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.035767342895269394,
+      "learning_rate": 0.00012337665952160266,
+      "loss": 0.1329,
+      "step": 1310
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.1937914788722992,
+      "learning_rate": 0.00012276415564955952,
+      "loss": 0.1353,
+      "step": 1311
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.08204084634780884,
+      "learning_rate": 0.00012215296310982404,
+      "loss": 0.1349,
+      "step": 1312
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.08185072988271713,
+      "learning_rate": 0.00012154308402699932,
+      "loss": 0.1279,
+      "step": 1313
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.12454118579626083,
+      "learning_rate": 0.00012093452052112308,
+      "loss": 0.143,
+      "step": 1314
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.047802336513996124,
+      "learning_rate": 0.0001203272747076598,
+      "loss": 0.1268,
+      "step": 1315
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.1269330382347107,
+      "learning_rate": 0.00011972134869749324,
+      "loss": 0.1307,
+      "step": 1316
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.04310709610581398,
+      "learning_rate": 0.0001191167445969193,
+      "loss": 0.1361,
+      "step": 1317
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.11329284310340881,
+      "learning_rate": 0.00011851346450763878,
+      "loss": 0.1275,
+      "step": 1318
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.19605499505996704,
+      "learning_rate": 0.0001179115105267502,
+      "loss": 0.1276,
+      "step": 1319
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.07777590304613113,
+      "learning_rate": 0.00011731088474674234,
+      "loss": 0.1266,
+      "step": 1320
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.36323878169059753,
+      "learning_rate": 0.00011671158925548624,
+      "loss": 0.1324,
+      "step": 1321
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.10665354877710342,
+      "learning_rate": 0.0001161136261362296,
+      "loss": 0.1274,
+      "step": 1322
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.10488320887088776,
+      "learning_rate": 0.00011551699746758787,
+      "loss": 0.1244,
+      "step": 1323
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.0471271350979805,
+      "learning_rate": 0.00011492170532353813,
+      "loss": 0.1326,
+      "step": 1324
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 0.09352532774209976,
+      "learning_rate": 0.00011432775177341165,
+      "loss": 0.1407,
+      "step": 1325
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.03449404612183571,
+      "learning_rate": 0.00011373513888188564,
+      "loss": 0.1328,
+      "step": 1326
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.09462067484855652,
+      "learning_rate": 0.00011314386870897792,
+      "loss": 0.141,
+      "step": 1327
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.06903573870658875,
+      "learning_rate": 0.00011255394331003854,
+      "loss": 0.1169,
+      "step": 1328
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.030416639521718025,
+      "learning_rate": 0.00011196536473574276,
+      "loss": 0.1235,
+      "step": 1329
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.2395412176847458,
+      "learning_rate": 0.00011137813503208399,
+      "loss": 0.145,
+      "step": 1330
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 0.2817445695400238,
+      "learning_rate": 0.00011079225624036688,
+      "loss": 0.1249,
+      "step": 1331
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.17817607522010803,
+      "learning_rate": 0.00011020773039720017,
+      "loss": 0.1458,
+      "step": 1332
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.03026886098086834,
+      "learning_rate": 0.00010962455953448952,
+      "loss": 0.1258,
+      "step": 1333
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.07572410255670547,
+      "learning_rate": 0.0001090427456794304,
+      "loss": 0.1217,
+      "step": 1334
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.06871062517166138,
+      "learning_rate": 0.00010846229085450099,
+      "loss": 0.1402,
+      "step": 1335
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 0.2113131284713745,
+      "learning_rate": 0.00010788319707745525,
+      "loss": 0.1276,
+      "step": 1336
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.20129992067813873,
+      "learning_rate": 0.00010730546636131621,
+      "loss": 0.1118,
+      "step": 1337
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.05704864114522934,
+      "learning_rate": 0.00010672910071436865,
+      "loss": 0.1203,
+      "step": 1338
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.058348219841718674,
+      "learning_rate": 0.00010615410214015186,
+      "loss": 0.1218,
+      "step": 1339
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.15510503947734833,
+      "learning_rate": 0.00010558047263745297,
+      "loss": 0.1207,
+      "step": 1340
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.2903478741645813,
+      "learning_rate": 0.00010500821420030049,
+      "loss": 0.1317,
+      "step": 1341
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.18940383195877075,
+      "learning_rate": 0.00010443732881795614,
+      "loss": 0.1333,
+      "step": 1342
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.08819776773452759,
+      "learning_rate": 0.0001038678184749095,
+      "loss": 0.1396,
+      "step": 1343
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.09822792559862137,
+      "learning_rate": 0.00010329968515086969,
+      "loss": 0.1326,
+      "step": 1344
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.06949262320995331,
+      "learning_rate": 0.00010273293082075913,
+      "loss": 0.1459,
+      "step": 1345
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.1197618618607521,
+      "learning_rate": 0.00010216755745470701,
+      "loss": 0.1171,
+      "step": 1346
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.12239718437194824,
+      "learning_rate": 0.00010160356701804169,
+      "loss": 0.1496,
+      "step": 1347
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 0.04563026875257492,
+      "learning_rate": 0.00010104096147128455,
+      "loss": 0.1425,
+      "step": 1348
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.14311982691287994,
+      "learning_rate": 0.00010047974277014266,
+      "loss": 0.1313,
+      "step": 1349
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.05187974497675896,
+      "learning_rate": 9.991991286550207e-05,
+      "loss": 0.1278,
+      "step": 1350
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.14110100269317627,
+      "learning_rate": 9.936147370342164e-05,
+      "loss": 0.1132,
+      "step": 1351
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.05993328616023064,
+      "learning_rate": 9.880442722512518e-05,
+      "loss": 0.1409,
+      "step": 1352
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 0.12421682476997375,
+      "learning_rate": 9.824877536699584e-05,
+      "loss": 0.1564,
+      "step": 1353
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.11717087775468826,
+      "learning_rate": 9.769452006056856e-05,
+      "loss": 0.1205,
+      "step": 1354
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.13294954597949982,
+      "learning_rate": 9.714166323252349e-05,
+      "loss": 0.1439,
+      "step": 1355
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.03808373585343361,
+      "learning_rate": 9.659020680468e-05,
+      "loss": 0.1139,
+      "step": 1356
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.07248345017433167,
+      "learning_rate": 9.604015269398874e-05,
+      "loss": 0.1314,
+      "step": 1357
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.16012993454933167,
+      "learning_rate": 9.549150281252633e-05,
+      "loss": 0.1408,
+      "step": 1358
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.07186062633991241,
+      "learning_rate": 9.49442590674876e-05,
+      "loss": 0.1515,
+      "step": 1359
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.06986256688833237,
+      "learning_rate": 9.439842336117954e-05,
+      "loss": 0.1054,
+      "step": 1360
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.06326832622289658,
+      "learning_rate": 9.385399759101481e-05,
+      "loss": 0.121,
+      "step": 1361
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.2067011445760727,
+      "learning_rate": 9.331098364950453e-05,
+      "loss": 0.094,
+      "step": 1362
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.03115193173289299,
+      "learning_rate": 9.276938342425245e-05,
+      "loss": 0.1194,
+      "step": 1363
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.050981197506189346,
+      "learning_rate": 9.222919879794772e-05,
+      "loss": 0.1483,
+      "step": 1364
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.15227286517620087,
+      "learning_rate": 9.169043164835867e-05,
+      "loss": 0.1301,
+      "step": 1365
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.18392544984817505,
+      "learning_rate": 9.115308384832638e-05,
+      "loss": 0.1309,
+      "step": 1366
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.09350251406431198,
+      "learning_rate": 9.061715726575825e-05,
+      "loss": 0.1098,
+      "step": 1367
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.19781382381916046,
+      "learning_rate": 9.008265376362079e-05,
+      "loss": 0.0971,
+      "step": 1368
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.06422741711139679,
+      "learning_rate": 8.954957519993401e-05,
+      "loss": 0.116,
+      "step": 1369
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 0.04665152728557587,
+      "learning_rate": 8.901792342776438e-05,
+      "loss": 0.102,
+      "step": 1370
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.23038771748542786,
+      "learning_rate": 8.848770029521874e-05,
+      "loss": 0.1434,
+      "step": 1371
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.06798720359802246,
+      "learning_rate": 8.795890764543818e-05,
+      "loss": 0.1251,
+      "step": 1372
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.11675478518009186,
+      "learning_rate": 8.74315473165902e-05,
+      "loss": 0.1236,
+      "step": 1373
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.0576016791164875,
+      "learning_rate": 8.690562114186423e-05,
+      "loss": 0.1292,
+      "step": 1374
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.060148756951093674,
+      "learning_rate": 8.638113094946381e-05,
+      "loss": 0.0988,
+      "step": 1375
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 0.18434777855873108,
+      "learning_rate": 8.585807856260108e-05,
+      "loss": 0.1204,
+      "step": 1376
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.20784330368041992,
+      "learning_rate": 8.533646579949034e-05,
+      "loss": 0.1306,
+      "step": 1377
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.14157481491565704,
+      "learning_rate": 8.481629447334066e-05,
+      "loss": 0.1467,
+      "step": 1378
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.06923433393239975,
+      "learning_rate": 8.429756639235136e-05,
+      "loss": 0.1189,
+      "step": 1379
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.24702174961566925,
+      "learning_rate": 8.37802833597045e-05,
+      "loss": 0.1472,
+      "step": 1380
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.0554979182779789,
+      "learning_rate": 8.326444717355874e-05,
+      "loss": 0.1087,
+      "step": 1381
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.11092767864465714,
+      "learning_rate": 8.275005962704346e-05,
+      "loss": 0.1191,
+      "step": 1382
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.09958731383085251,
+      "learning_rate": 8.223712250825216e-05,
+      "loss": 0.1345,
+      "step": 1383
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.19082792103290558,
+      "learning_rate": 8.172563760023665e-05,
+      "loss": 0.1056,
+      "step": 1384
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.06456390023231506,
+      "learning_rate": 8.121560668100064e-05,
+      "loss": 0.1055,
+      "step": 1385
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.10609104484319687,
+      "learning_rate": 8.070703152349334e-05,
+      "loss": 0.1202,
+      "step": 1386
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 0.13664673268795013,
+      "learning_rate": 8.019991389560349e-05,
+      "loss": 0.1486,
+      "step": 1387
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.24724584817886353,
+      "learning_rate": 7.969425556015325e-05,
+      "loss": 0.1453,
+      "step": 1388
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.07616293430328369,
+      "learning_rate": 7.919005827489228e-05,
+      "loss": 0.1203,
+      "step": 1389
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.10011781007051468,
+      "learning_rate": 7.868732379249122e-05,
+      "loss": 0.0969,
+      "step": 1390
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.19106784462928772,
+      "learning_rate": 7.818605386053573e-05,
+      "loss": 0.1426,
+      "step": 1391
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.17057418823242188,
+      "learning_rate": 7.768625022152037e-05,
+      "loss": 0.1412,
+      "step": 1392
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 0.10925137251615524,
+      "learning_rate": 7.718791461284303e-05,
+      "loss": 0.1177,
+      "step": 1393
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.16444699466228485,
+      "learning_rate": 7.669104876679795e-05,
+      "loss": 0.1238,
+      "step": 1394
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.09140460193157196,
+      "learning_rate": 7.619565441057075e-05,
+      "loss": 0.127,
+      "step": 1395
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.2647579312324524,
+      "learning_rate": 7.570173326623153e-05,
+      "loss": 0.1404,
+      "step": 1396
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.11721226572990417,
+      "learning_rate": 7.520928705072938e-05,
+      "loss": 0.1432,
+      "step": 1397
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 0.08702779561281204,
+      "learning_rate": 7.471831747588653e-05,
+      "loss": 0.0986,
+      "step": 1398
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.09286800026893616,
+      "learning_rate": 7.422882624839178e-05,
+      "loss": 0.1113,
+      "step": 1399
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.10280603915452957,
+      "learning_rate": 7.37408150697953e-05,
+      "loss": 0.1048,
+      "step": 1400
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.20054741203784943,
+      "learning_rate": 7.325428563650222e-05,
+      "loss": 0.1496,
+      "step": 1401
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.21980531513690948,
+      "learning_rate": 7.276923963976667e-05,
+      "loss": 0.1246,
+      "step": 1402
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.16484405100345612,
+      "learning_rate": 7.228567876568665e-05,
+      "loss": 0.1257,
+      "step": 1403
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.219330832362175,
+      "learning_rate": 7.180360469519714e-05,
+      "loss": 0.1329,
+      "step": 1404
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.17407891154289246,
+      "learning_rate": 7.132301910406502e-05,
+      "loss": 0.1585,
+      "step": 1405
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.09703540056943893,
+      "learning_rate": 7.084392366288295e-05,
+      "loss": 0.1161,
+      "step": 1406
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.21986158192157745,
+      "learning_rate": 7.036632003706328e-05,
+      "loss": 0.1408,
+      "step": 1407
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.16698057949543,
+      "learning_rate": 6.989020988683314e-05,
+      "loss": 0.1183,
+      "step": 1408
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.09089485555887222,
+      "learning_rate": 6.941559486722748e-05,
+      "loss": 0.1138,
+      "step": 1409
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 0.3572080731391907,
+      "learning_rate": 6.894247662808457e-05,
+      "loss": 0.1148,
+      "step": 1410
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.21670278906822205,
+      "learning_rate": 6.847085681403914e-05,
+      "loss": 0.1081,
+      "step": 1411
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.1353166103363037,
+      "learning_rate": 6.800073706451721e-05,
+      "loss": 0.1335,
+      "step": 1412
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.10845667868852615,
+      "learning_rate": 6.753211901373064e-05,
+      "loss": 0.1462,
+      "step": 1413
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.13182678818702698,
+      "learning_rate": 6.706500429067075e-05,
+      "loss": 0.137,
+      "step": 1414
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.11666212975978851,
+      "learning_rate": 6.659939451910341e-05,
+      "loss": 0.1245,
+      "step": 1415
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.19230018556118011,
+      "learning_rate": 6.613529131756285e-05,
+      "loss": 0.1392,
+      "step": 1416
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.12369673699140549,
+      "learning_rate": 6.567269629934613e-05,
+      "loss": 0.152,
+      "step": 1417
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.4312630891799927,
+      "learning_rate": 6.521161107250778e-05,
+      "loss": 0.1539,
+      "step": 1418
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.12075989693403244,
+      "learning_rate": 6.475203723985418e-05,
+      "loss": 0.1429,
+      "step": 1419
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 0.19320321083068848,
+      "learning_rate": 6.429397639893758e-05,
+      "loss": 0.146,
+      "step": 1420
+    },
+    {
+      "epoch": 2.51,
+      "eval_loss": 0.13353538513183594,
+      "eval_runtime": 14.1007,
+      "eval_samples_per_second": 33.828,
+      "eval_steps_per_second": 8.51,
+      "step": 1420
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 2.2152562141418457,
+      "learning_rate": 6.38374301420509e-05,
+      "loss": 0.1597,
+      "step": 1421
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.04021691530942917,
+      "learning_rate": 6.338240005622209e-05,
+      "loss": 0.1246,
+      "step": 1422
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.06637288630008698,
+      "learning_rate": 6.292888772320882e-05,
+      "loss": 0.1187,
+      "step": 1423
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.16920767724514008,
+      "learning_rate": 6.247689471949291e-05,
+      "loss": 0.1268,
+      "step": 1424
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.22192606329917908,
+      "learning_rate": 6.202642261627411e-05,
+      "loss": 0.1477,
+      "step": 1425
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.08232539147138596,
+      "learning_rate": 6.157747297946608e-05,
+      "loss": 0.1264,
+      "step": 1426
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.08191093057394028,
+      "learning_rate": 6.113004736968953e-05,
+      "loss": 0.1459,
+      "step": 1427
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.12319450080394745,
+      "learning_rate": 6.068414734226774e-05,
+      "loss": 0.105,
+      "step": 1428
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.2650308907032013,
+      "learning_rate": 6.023977444722095e-05,
+      "loss": 0.1232,
+      "step": 1429
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.13269290328025818,
+      "learning_rate": 5.979693022926025e-05,
+      "loss": 0.1118,
+      "step": 1430
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.145817831158638,
+      "learning_rate": 5.935561622778335e-05,
+      "loss": 0.1142,
+      "step": 1431
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 0.09145599603652954,
+      "learning_rate": 5.891583397686862e-05,
+      "loss": 0.1285,
+      "step": 1432
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.11915401369333267,
+      "learning_rate": 5.8477585005269564e-05,
+      "loss": 0.1172,
+      "step": 1433
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.13061662018299103,
+      "learning_rate": 5.804087083641002e-05,
+      "loss": 0.1401,
+      "step": 1434
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.08747705817222595,
+      "learning_rate": 5.760569298837826e-05,
+      "loss": 0.1315,
+      "step": 1435
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.08562975376844406,
+      "learning_rate": 5.717205297392247e-05,
+      "loss": 0.122,
+      "step": 1436
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.12241534143686295,
+      "learning_rate": 5.673995230044498e-05,
+      "loss": 0.1247,
+      "step": 1437
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 0.044354431331157684,
+      "learning_rate": 5.6309392469996944e-05,
+      "loss": 0.123,
+      "step": 1438
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.06913258135318756,
+      "learning_rate": 5.5880374979273395e-05,
+      "loss": 0.1275,
+      "step": 1439
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.0650060847401619,
+      "learning_rate": 5.5452901319607894e-05,
+      "loss": 0.1066,
+      "step": 1440
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.14583538472652435,
+      "learning_rate": 5.502697297696746e-05,
+      "loss": 0.157,
+      "step": 1441
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.12845255434513092,
+      "learning_rate": 5.4602591431947514e-05,
+      "loss": 0.1498,
+      "step": 1442
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.08573470264673233,
+      "learning_rate": 5.417975815976628e-05,
+      "loss": 0.1014,
+      "step": 1443
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.21065488457679749,
+      "learning_rate": 5.37584746302599e-05,
+      "loss": 0.168,
+      "step": 1444
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.06427565217018127,
+      "learning_rate": 5.333874230787772e-05,
+      "loss": 0.1166,
+      "step": 1445
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.15018604695796967,
+      "learning_rate": 5.292056265167644e-05,
+      "loss": 0.1257,
+      "step": 1446
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.17715144157409668,
+      "learning_rate": 5.2503937115315816e-05,
+      "loss": 0.1457,
+      "step": 1447
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.19283372163772583,
+      "learning_rate": 5.208886714705291e-05,
+      "loss": 0.1431,
+      "step": 1448
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.1359899789094925,
+      "learning_rate": 5.16753541897374e-05,
+      "loss": 0.1388,
+      "step": 1449
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.09452392160892487,
+      "learning_rate": 5.126339968080695e-05,
+      "loss": 0.1245,
+      "step": 1450
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.043228164315223694,
+      "learning_rate": 5.085300505228124e-05,
+      "loss": 0.1291,
+      "step": 1451
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.04729638248682022,
+      "learning_rate": 5.0444171730758046e-05,
+      "loss": 0.1374,
+      "step": 1452
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.1768418550491333,
+      "learning_rate": 5.00369011374075e-05,
+      "loss": 0.1232,
+      "step": 1453
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.040266744792461395,
+      "learning_rate": 4.9631194687967394e-05,
+      "loss": 0.1337,
+      "step": 1454
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 0.0584387332201004,
+      "learning_rate": 4.9227053792738616e-05,
+      "loss": 0.151,
+      "step": 1455
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.06620542705059052,
+      "learning_rate": 4.882447985657956e-05,
+      "loss": 0.1306,
+      "step": 1456
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.2833789885044098,
+      "learning_rate": 4.842347427890198e-05,
+      "loss": 0.149,
+      "step": 1457
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.12224625051021576,
+      "learning_rate": 4.8024038453665544e-05,
+      "loss": 0.1403,
+      "step": 1458
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.027846721932291985,
+      "learning_rate": 4.762617376937312e-05,
+      "loss": 0.1287,
+      "step": 1459
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.26311802864074707,
+      "learning_rate": 4.722988160906638e-05,
+      "loss": 0.1335,
+      "step": 1460
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 0.3691308796405792,
+      "learning_rate": 4.6835163350320176e-05,
+      "loss": 0.145,
+      "step": 1461
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.028389999642968178,
+      "learning_rate": 4.644202036523881e-05,
+      "loss": 0.1248,
+      "step": 1462
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.04470786452293396,
+      "learning_rate": 4.605045402045022e-05,
+      "loss": 0.1381,
+      "step": 1463
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.21815143525600433,
+      "learning_rate": 4.566046567710169e-05,
+      "loss": 0.1258,
+      "step": 1464
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.14745290577411652,
+      "learning_rate": 4.527205669085549e-05,
+      "loss": 0.1403,
+      "step": 1465
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.2701740562915802,
+      "learning_rate": 4.488522841188336e-05,
+      "loss": 0.1228,
+      "step": 1466
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.1660449504852295,
+      "learning_rate": 4.449998218486262e-05,
+      "loss": 0.1529,
+      "step": 1467
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.060599759221076965,
+      "learning_rate": 4.411631934897092e-05,
+      "loss": 0.1211,
+      "step": 1468
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.2801418900489807,
+      "learning_rate": 4.3734241237881666e-05,
+      "loss": 0.137,
+      "step": 1469
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.06915499269962311,
+      "learning_rate": 4.335374917975982e-05,
+      "loss": 0.1358,
+      "step": 1470
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.12740154564380646,
+      "learning_rate": 4.297484449725691e-05,
+      "loss": 0.1347,
+      "step": 1471
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 0.1669289469718933,
+      "learning_rate": 4.259752850750609e-05,
+      "loss": 0.1301,
+      "step": 1472
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.07437172532081604,
+      "learning_rate": 4.222180252211849e-05,
+      "loss": 0.119,
+      "step": 1473
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.08567313104867935,
+      "learning_rate": 4.184766784717775e-05,
+      "loss": 0.1128,
+      "step": 1474
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.11972495913505554,
+      "learning_rate": 4.147512578323615e-05,
+      "loss": 0.1355,
+      "step": 1475
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.0828404352068901,
+      "learning_rate": 4.110417762530977e-05,
+      "loss": 0.1511,
+      "step": 1476
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.09823042154312134,
+      "learning_rate": 4.073482466287359e-05,
+      "loss": 0.1237,
+      "step": 1477
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 0.039382707327604294,
+      "learning_rate": 4.036706817985802e-05,
+      "loss": 0.1182,
+      "step": 1478
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.09079232066869736,
+      "learning_rate": 4.0000909454643406e-05,
+      "loss": 0.1207,
+      "step": 1479
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.15438657999038696,
+      "learning_rate": 3.9636349760056425e-05,
+      "loss": 0.1229,
+      "step": 1480
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.09138775616884232,
+      "learning_rate": 3.927339036336486e-05,
+      "loss": 0.1291,
+      "step": 1481
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.06647726148366928,
+      "learning_rate": 3.8912032526273846e-05,
+      "loss": 0.13,
+      "step": 1482
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 0.13314370810985565,
+      "learning_rate": 3.855227750492118e-05,
+      "loss": 0.1367,
+      "step": 1483
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.09128770977258682,
+      "learning_rate": 3.819412654987314e-05,
+      "loss": 0.1164,
+      "step": 1484
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.03929639607667923,
+      "learning_rate": 3.783758090611983e-05,
+      "loss": 0.1463,
+      "step": 1485
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.06576254218816757,
+      "learning_rate": 3.748264181307109e-05,
+      "loss": 0.1158,
+      "step": 1486
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.1495221108198166,
+      "learning_rate": 3.712931050455204e-05,
+      "loss": 0.1234,
+      "step": 1487
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.1673547774553299,
+      "learning_rate": 3.6777588208799116e-05,
+      "loss": 0.1176,
+      "step": 1488
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.14675819873809814,
+      "learning_rate": 3.6427476148455484e-05,
+      "loss": 0.1317,
+      "step": 1489
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.07739079743623734,
+      "learning_rate": 3.607897554056672e-05,
+      "loss": 0.1235,
+      "step": 1490
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.048171430826187134,
+      "learning_rate": 3.5732087596576866e-05,
+      "loss": 0.1408,
+      "step": 1491
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.04192957654595375,
+      "learning_rate": 3.538681352232403e-05,
+      "loss": 0.1361,
+      "step": 1492
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.0708787590265274,
+      "learning_rate": 3.50431545180363e-05,
+      "loss": 0.1345,
+      "step": 1493
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.047831833362579346,
+      "learning_rate": 3.470111177832758e-05,
+      "loss": 0.1179,
+      "step": 1494
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 0.07011017948389053,
+      "learning_rate": 3.436068649219326e-05,
+      "loss": 0.1206,
+      "step": 1495
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.20468172430992126,
+      "learning_rate": 3.402187984300614e-05,
+      "loss": 0.1155,
+      "step": 1496
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.5008040070533752,
+      "learning_rate": 3.368469300851262e-05,
+      "loss": 0.1518,
+      "step": 1497
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.12470618635416031,
+      "learning_rate": 3.334912716082811e-05,
+      "loss": 0.1476,
+      "step": 1498
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.04671414569020271,
+      "learning_rate": 3.30151834664334e-05,
+      "loss": 0.1516,
+      "step": 1499
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 0.06154852360486984,
+      "learning_rate": 3.2682863086170414e-05,
+      "loss": 0.1419,
+      "step": 1500
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.13698704540729523,
+      "learning_rate": 3.235216717523787e-05,
+      "loss": 0.14,
+      "step": 1501
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.25294530391693115,
+      "learning_rate": 3.2023096883188e-05,
+      "loss": 0.1392,
+      "step": 1502
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.0433129258453846,
+      "learning_rate": 3.169565335392183e-05,
+      "loss": 0.1203,
+      "step": 1503
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.10114753991365433,
+      "learning_rate": 3.136983772568569e-05,
+      "loss": 0.1179,
+      "step": 1504
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.058237019926309586,
+      "learning_rate": 3.104565113106689e-05,
+      "loss": 0.1491,
+      "step": 1505
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.07598286867141724,
+      "learning_rate": 3.0723094696990027e-05,
+      "loss": 0.1117,
+      "step": 1506
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.13283759355545044,
+      "learning_rate": 3.040216954471309e-05,
+      "loss": 0.1409,
+      "step": 1507
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.1821421980857849,
+      "learning_rate": 3.0082876789823245e-05,
+      "loss": 0.1257,
+      "step": 1508
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.04798609018325806,
+      "learning_rate": 2.9765217542233438e-05,
+      "loss": 0.1253,
+      "step": 1509
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.3384339213371277,
+      "learning_rate": 2.9449192906178203e-05,
+      "loss": 0.1432,
+      "step": 1510
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.08849462866783142,
+      "learning_rate": 2.9134803980209734e-05,
+      "loss": 0.1336,
+      "step": 1511
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.11560472846031189,
+      "learning_rate": 2.88220518571945e-05,
+      "loss": 0.11,
+      "step": 1512
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.2896404266357422,
+      "learning_rate": 2.8510937624308954e-05,
+      "loss": 0.163,
+      "step": 1513
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.15655553340911865,
+      "learning_rate": 2.8201462363036112e-05,
+      "loss": 0.1396,
+      "step": 1514
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.05299900099635124,
+      "learning_rate": 2.7893627149161717e-05,
+      "loss": 0.115,
+      "step": 1515
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.08437127619981766,
+      "learning_rate": 2.7587433052770115e-05,
+      "loss": 0.1149,
+      "step": 1516
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 0.18013562262058258,
+      "learning_rate": 2.72828811382414e-05,
+      "loss": 0.1164,
+      "step": 1517
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.09165755659341812,
+      "learning_rate": 2.6979972464246604e-05,
+      "loss": 0.1474,
+      "step": 1518
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.16448521614074707,
+      "learning_rate": 2.667870808374506e-05,
+      "loss": 0.1372,
+      "step": 1519
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.06156858801841736,
+      "learning_rate": 2.6379089043980064e-05,
+      "loss": 0.1421,
+      "step": 1520
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.13896550238132477,
+      "learning_rate": 2.6081116386475313e-05,
+      "loss": 0.1283,
+      "step": 1521
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.12410745024681091,
+      "learning_rate": 2.5784791147031638e-05,
+      "loss": 0.1273,
+      "step": 1522
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 0.0802113488316536,
+      "learning_rate": 2.5490114355723294e-05,
+      "loss": 0.1134,
+      "step": 1523
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.22037829458713531,
+      "learning_rate": 2.5197087036893774e-05,
+      "loss": 0.1457,
+      "step": 1524
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.09046395123004913,
+      "learning_rate": 2.490571020915322e-05,
+      "loss": 0.1279,
+      "step": 1525
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.08439705520868301,
+      "learning_rate": 2.4615984885374143e-05,
+      "loss": 0.0983,
+      "step": 1526
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.311358243227005,
+      "learning_rate": 2.4327912072688306e-05,
+      "loss": 0.1276,
+      "step": 1527
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.18210773169994354,
+      "learning_rate": 2.404149277248313e-05,
+      "loss": 0.1321,
+      "step": 1528
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.17474836111068726,
+      "learning_rate": 2.3756727980397742e-05,
+      "loss": 0.1602,
+      "step": 1529
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.040003515779972076,
+      "learning_rate": 2.3473618686320474e-05,
+      "loss": 0.1371,
+      "step": 1530
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.25132429599761963,
+      "learning_rate": 2.319216587438455e-05,
+      "loss": 0.1293,
+      "step": 1531
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.07414573431015015,
+      "learning_rate": 2.291237052296513e-05,
+      "loss": 0.1316,
+      "step": 1532
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.050560012459754944,
+      "learning_rate": 2.2634233604675812e-05,
+      "loss": 0.1191,
+      "step": 1533
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 0.048617489635944366,
+      "learning_rate": 2.2357756086364924e-05,
+      "loss": 0.1106,
+      "step": 1534
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.09126359969377518,
+      "learning_rate": 2.2082938929112838e-05,
+      "loss": 0.1422,
+      "step": 1535
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.09898441284894943,
+      "learning_rate": 2.180978308822812e-05,
+      "loss": 0.1145,
+      "step": 1536
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.2153613418340683,
+      "learning_rate": 2.1538289513244212e-05,
+      "loss": 0.1339,
+      "step": 1537
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.07101116329431534,
+      "learning_rate": 2.126845914791631e-05,
+      "loss": 0.1397,
+      "step": 1538
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.23921048641204834,
+      "learning_rate": 2.100029293021799e-05,
+      "loss": 0.113,
+      "step": 1539
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 0.16103985905647278,
+      "learning_rate": 2.0733791792338197e-05,
+      "loss": 0.1372,
+      "step": 1540
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.05808824300765991,
+      "learning_rate": 2.046895666067755e-05,
+      "loss": 0.1549,
+      "step": 1541
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.044097110629081726,
+      "learning_rate": 2.0205788455845474e-05,
+      "loss": 0.1337,
+      "step": 1542
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.16746556758880615,
+      "learning_rate": 1.9944288092656858e-05,
+      "loss": 0.1096,
+      "step": 1543
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.3115890920162201,
+      "learning_rate": 1.9684456480128843e-05,
+      "loss": 0.1437,
+      "step": 1544
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 0.12141338735818863,
+      "learning_rate": 1.942629452147787e-05,
+      "loss": 0.1322,
+      "step": 1545
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.13122229278087616,
+      "learning_rate": 1.9169803114116313e-05,
+      "loss": 0.1057,
+      "step": 1546
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.12674184143543243,
+      "learning_rate": 1.8914983149649513e-05,
+      "loss": 0.1004,
+      "step": 1547
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.18313969671726227,
+      "learning_rate": 1.866183551387235e-05,
+      "loss": 0.1122,
+      "step": 1548
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.12356254458427429,
+      "learning_rate": 1.84103610867668e-05,
+      "loss": 0.1362,
+      "step": 1549
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.09115366637706757,
+      "learning_rate": 1.8160560742498223e-05,
+      "loss": 0.131,
+      "step": 1550
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.03777840733528137,
+      "learning_rate": 1.7912435349412726e-05,
+      "loss": 0.1339,
+      "step": 1551
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.026918258517980576,
+      "learning_rate": 1.7665985770033975e-05,
+      "loss": 0.1414,
+      "step": 1552
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.06117779389023781,
+      "learning_rate": 1.7421212861060133e-05,
+      "loss": 0.1121,
+      "step": 1553
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.05697258189320564,
+      "learning_rate": 1.7178117473361287e-05,
+      "loss": 0.1523,
+      "step": 1554
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.14216284453868866,
+      "learning_rate": 1.693670045197582e-05,
+      "loss": 0.1354,
+      "step": 1555
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.1293654590845108,
+      "learning_rate": 1.669696263610815e-05,
+      "loss": 0.1466,
+      "step": 1556
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.09816037863492966,
+      "learning_rate": 1.6458904859125322e-05,
+      "loss": 0.1344,
+      "step": 1557
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.1260695606470108,
+      "learning_rate": 1.622252794855433e-05,
+      "loss": 0.1464,
+      "step": 1558
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.04835040122270584,
+      "learning_rate": 1.5987832726079343e-05,
+      "loss": 0.1292,
+      "step": 1559
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.1723223179578781,
+      "learning_rate": 1.5754820007538473e-05,
+      "loss": 0.1474,
+      "step": 1560
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.04331492260098457,
+      "learning_rate": 1.5523490602921353e-05,
+      "loss": 0.1513,
+      "step": 1561
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 0.18255582451820374,
+      "learning_rate": 1.5293845316366185e-05,
+      "loss": 0.1502,
+      "step": 1562
+    },
+    {
+      "epoch": 2.76,
+      "eval_loss": 0.13326887786388397,
+      "eval_runtime": 14.1351,
+      "eval_samples_per_second": 33.746,
+      "eval_steps_per_second": 8.489,
+      "step": 1562
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.3445633351802826,
+      "learning_rate": 1.5065884946156682e-05,
+      "loss": 0.1229,
+      "step": 1563
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.0768701359629631,
+      "learning_rate": 1.483961028471975e-05,
+      "loss": 0.1318,
+      "step": 1564
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.09043741226196289,
+      "learning_rate": 1.4615022118622368e-05,
+      "loss": 0.1038,
+      "step": 1565
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.12818877398967743,
+      "learning_rate": 1.4392121228569088e-05,
+      "loss": 0.121,
+      "step": 1566
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.08737614750862122,
+      "learning_rate": 1.4170908389399107e-05,
+      "loss": 0.1169,
+      "step": 1567
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 0.041610416024923325,
+      "learning_rate": 1.3951384370083697e-05,
+      "loss": 0.139,
+      "step": 1568
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.13382531702518463,
+      "learning_rate": 1.3733549933723666e-05,
+      "loss": 0.1242,
+      "step": 1569
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.14836427569389343,
+      "learning_rate": 1.3517405837546404e-05,
+      "loss": 0.1292,
+      "step": 1570
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.16950459778308868,
+      "learning_rate": 1.3302952832903392e-05,
+      "loss": 0.1338,
+      "step": 1571
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.09254854172468185,
+      "learning_rate": 1.3090191665267814e-05,
+      "loss": 0.1016,
+      "step": 1572
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.04616238549351692,
+      "learning_rate": 1.2879123074231502e-05,
+      "loss": 0.1467,
+      "step": 1573
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.04659713804721832,
+      "learning_rate": 1.2669747793502828e-05,
+      "loss": 0.1246,
+      "step": 1574
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.0829511284828186,
+      "learning_rate": 1.2462066550903816e-05,
+      "loss": 0.153,
+      "step": 1575
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.12231741100549698,
+      "learning_rate": 1.225608006836776e-05,
+      "loss": 0.1125,
+      "step": 1576
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.0418907031416893,
+      "learning_rate": 1.2051789061936713e-05,
+      "loss": 0.1282,
+      "step": 1577
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.03682105615735054,
+      "learning_rate": 1.1849194241759009e-05,
+      "loss": 0.1337,
+      "step": 1578
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 0.21309414505958557,
+      "learning_rate": 1.1648296312086747e-05,
+      "loss": 0.11,
+      "step": 1579
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.06915529817342758,
+      "learning_rate": 1.1449095971273304e-05,
+      "loss": 0.1595,
+      "step": 1580
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.07684849202632904,
+      "learning_rate": 1.1251593911771052e-05,
+      "loss": 0.1307,
+      "step": 1581
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.07943115383386612,
+      "learning_rate": 1.1055790820128919e-05,
+      "loss": 0.1172,
+      "step": 1582
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.21117864549160004,
+      "learning_rate": 1.0861687376989671e-05,
+      "loss": 0.1199,
+      "step": 1583
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.18909168243408203,
+      "learning_rate": 1.0669284257088185e-05,
+      "loss": 0.1482,
+      "step": 1584
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.027331866323947906,
+      "learning_rate": 1.0478582129248516e-05,
+      "loss": 0.1238,
+      "step": 1585
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.12267359346151352,
+      "learning_rate": 1.0289581656381774e-05,
+      "loss": 0.1284,
+      "step": 1586
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.07290299981832504,
+      "learning_rate": 1.0102283495483977e-05,
+      "loss": 0.1385,
+      "step": 1587
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.10076455026865005,
+      "learning_rate": 9.916688297633647e-06,
+      "loss": 0.1455,
+      "step": 1588
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.2671211361885071,
+      "learning_rate": 9.732796707989377e-06,
+      "loss": 0.16,
+      "step": 1589
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.11416134238243103,
+      "learning_rate": 9.550609365787888e-06,
+      "loss": 0.1575,
+      "step": 1590
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.08602559566497803,
+      "learning_rate": 9.37012690434147e-06,
+      "loss": 0.1292,
+      "step": 1591
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.046941716223955154,
+      "learning_rate": 9.191349951036266e-06,
+      "loss": 0.1017,
+      "step": 1592
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.061204541474580765,
+      "learning_rate": 9.014279127329605e-06,
+      "loss": 0.1385,
+      "step": 1593
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.07410020381212234,
+      "learning_rate": 8.838915048748064e-06,
+      "loss": 0.1474,
+      "step": 1594
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.35706961154937744,
+      "learning_rate": 8.66525832488535e-06,
+      "loss": 0.1596,
+      "step": 1595
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 0.06470204889774323,
+      "learning_rate": 8.493309559399976e-06,
+      "loss": 0.1628,
+      "step": 1596
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.18308192491531372,
+      "learning_rate": 8.323069350013479e-06,
+      "loss": 0.1176,
+      "step": 1597
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.06461644172668457,
+      "learning_rate": 8.154538288508206e-06,
+      "loss": 0.126,
+      "step": 1598
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.2490461766719818,
+      "learning_rate": 7.987716960725144e-06,
+      "loss": 0.1468,
+      "step": 1599
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.11734851449728012,
+      "learning_rate": 7.822605946561923e-06,
+      "loss": 0.145,
+      "step": 1600
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.039642583578825,
+      "learning_rate": 7.659205819970927e-06,
+      "loss": 0.1268,
+      "step": 1601
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 0.04151405021548271,
+      "learning_rate": 7.497517148957244e-06,
+      "loss": 0.1301,
+      "step": 1602
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.1388237625360489,
+      "learning_rate": 7.3375404955766665e-06,
+      "loss": 0.1382,
+      "step": 1603
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.26035836338996887,
+      "learning_rate": 7.179276415933633e-06,
+      "loss": 0.1538,
+      "step": 1604
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.22738611698150635,
+      "learning_rate": 7.022725460179458e-06,
+      "loss": 0.1083,
+      "step": 1605
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.2859216630458832,
+      "learning_rate": 6.867888172510439e-06,
+      "loss": 0.1362,
+      "step": 1606
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.22834646701812744,
+      "learning_rate": 6.7147650911658086e-06,
+      "loss": 0.1289,
+      "step": 1607
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 0.05140649899840355,
+      "learning_rate": 6.5633567484259525e-06,
+      "loss": 0.1191,
+      "step": 1608
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.06531380861997604,
+      "learning_rate": 6.413663670610526e-06,
+      "loss": 0.1122,
+      "step": 1609
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.031297944486141205,
+      "learning_rate": 6.26568637807673e-06,
+      "loss": 0.1274,
+      "step": 1610
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.23136821389198303,
+      "learning_rate": 6.119425385217314e-06,
+      "loss": 0.1178,
+      "step": 1611
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.11101561784744263,
+      "learning_rate": 5.9748812004590255e-06,
+      "loss": 0.1435,
+      "step": 1612
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.12405877560377121,
+      "learning_rate": 5.832054326260605e-06,
+      "loss": 0.1342,
+      "step": 1613
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.02779628150165081,
+      "learning_rate": 5.69094525911118e-06,
+      "loss": 0.1283,
+      "step": 1614
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.1838769018650055,
+      "learning_rate": 5.551554489528432e-06,
+      "loss": 0.1569,
+      "step": 1615
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.06827875226736069,
+      "learning_rate": 5.413882502057155e-06,
+      "loss": 0.1119,
+      "step": 1616
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.07234180718660355,
+      "learning_rate": 5.277929775267143e-06,
+      "loss": 0.1302,
+      "step": 1617
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.23422658443450928,
+      "learning_rate": 5.143696781751972e-06,
+      "loss": 0.1459,
+      "step": 1618
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 0.10591956228017807,
+      "learning_rate": 5.011183988127055e-06,
+      "loss": 0.1529,
+      "step": 1619
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.04643354192376137,
+      "learning_rate": 4.880391855028088e-06,
+      "loss": 0.1496,
+      "step": 1620
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.2558877766132355,
+      "learning_rate": 4.751320837109552e-06,
+      "loss": 0.1484,
+      "step": 1621
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.14346656203269958,
+      "learning_rate": 4.6239713830429354e-06,
+      "loss": 0.1295,
+      "step": 1622
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.06926041841506958,
+      "learning_rate": 4.498343935515348e-06,
+      "loss": 0.1275,
+      "step": 1623
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.10302523523569107,
+      "learning_rate": 4.374438931228075e-06,
+      "loss": 0.1432,
+      "step": 1624
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 0.12526661157608032,
+      "learning_rate": 4.252256800894694e-06,
+      "loss": 0.1271,
+      "step": 1625
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.08141893148422241,
+      "learning_rate": 4.131797969239903e-06,
+      "loss": 0.1352,
+      "step": 1626
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.03432145342230797,
+      "learning_rate": 4.013062854998028e-06,
+      "loss": 0.1307,
+      "step": 1627
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.04721994698047638,
+      "learning_rate": 3.896051870911188e-06,
+      "loss": 0.1333,
+      "step": 1628
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.07389276474714279,
+      "learning_rate": 3.7807654237284605e-06,
+      "loss": 0.1245,
+      "step": 1629
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.027808329090476036,
+      "learning_rate": 3.6672039142039425e-06,
+      "loss": 0.1354,
+      "step": 1630
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.18238960206508636,
+      "learning_rate": 3.5553677370957495e-06,
+      "loss": 0.1252,
+      "step": 1631
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.042762529104948044,
+      "learning_rate": 3.445257281164349e-06,
+      "loss": 0.1445,
+      "step": 1632
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.16484946012496948,
+      "learning_rate": 3.3368729291712863e-06,
+      "loss": 0.1459,
+      "step": 1633
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.04074009507894516,
+      "learning_rate": 3.2302150578780165e-06,
+      "loss": 0.1229,
+      "step": 1634
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.03902529180049896,
+      "learning_rate": 3.125284038044407e-06,
+      "loss": 0.1303,
+      "step": 1635
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.0906100794672966,
+      "learning_rate": 3.0220802344275157e-06,
+      "loss": 0.1367,
+      "step": 1636
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.07743262499570847,
+      "learning_rate": 2.9206040057802584e-06,
+      "loss": 0.1423,
+      "step": 1637
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.0563175305724144,
+      "learning_rate": 2.8208557048503556e-06,
+      "loss": 0.122,
+      "step": 1638
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.08874353021383286,
+      "learning_rate": 2.7228356783788876e-06,
+      "loss": 0.1037,
+      "step": 1639
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.05931072309613228,
+      "learning_rate": 2.626544267099129e-06,
+      "loss": 0.1155,
+      "step": 1640
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.03920688480138779,
+      "learning_rate": 2.531981805735606e-06,
+      "loss": 0.1462,
+      "step": 1641
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 0.09828366339206696,
+      "learning_rate": 2.4391486230024297e-06,
+      "loss": 0.1603,
+      "step": 1642
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.16284847259521484,
+      "learning_rate": 2.3480450416027423e-06,
+      "loss": 0.1222,
+      "step": 1643
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.0588088184595108,
+      "learning_rate": 2.2586713782272172e-06,
+      "loss": 0.1423,
+      "step": 1644
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.21331720054149628,
+      "learning_rate": 2.1710279435530057e-06,
+      "loss": 0.1567,
+      "step": 1645
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.1025921106338501,
+      "learning_rate": 2.0851150422427913e-06,
+      "loss": 0.1297,
+      "step": 1646
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 0.12087155133485794,
+      "learning_rate": 2.0009329729435146e-06,
+      "loss": 0.1207,
+      "step": 1647
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.12067259848117828,
+      "learning_rate": 1.9184820282855953e-06,
+      "loss": 0.1335,
+      "step": 1648
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.12102551758289337,
+      "learning_rate": 1.8377624948817673e-06,
+      "loss": 0.1483,
+      "step": 1649
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.07149103283882141,
+      "learning_rate": 1.7587746533260785e-06,
+      "loss": 0.1423,
+      "step": 1650
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.19505378603935242,
+      "learning_rate": 1.6815187781928921e-06,
+      "loss": 0.1045,
+      "step": 1651
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.14010155200958252,
+      "learning_rate": 1.6059951380359984e-06,
+      "loss": 0.1282,
+      "step": 1652
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 0.061695683747529984,
+      "learning_rate": 1.5322039953878374e-06,
+      "loss": 0.1342,
+      "step": 1653
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.24404413998126984,
+      "learning_rate": 1.4601456067580564e-06,
+      "loss": 0.1379,
+      "step": 1654
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.07468795031309128,
+      "learning_rate": 1.3898202226333423e-06,
+      "loss": 0.1219,
+      "step": 1655
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.061813950538635254,
+      "learning_rate": 1.3212280874759231e-06,
+      "loss": 0.1151,
+      "step": 1656
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.1875925213098526,
+      "learning_rate": 1.2543694397230686e-06,
+      "loss": 0.1474,
+      "step": 1657
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.06623123586177826,
+      "learning_rate": 1.1892445117862028e-06,
+      "loss": 0.0881,
+      "step": 1658
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.057737018913030624,
+      "learning_rate": 1.1258535300499583e-06,
+      "loss": 0.1452,
+      "step": 1659
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.03276927024126053,
+      "learning_rate": 1.0641967148716236e-06,
+      "loss": 0.1435,
+      "step": 1660
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.07240304350852966,
+      "learning_rate": 1.004274280580142e-06,
+      "loss": 0.1266,
+      "step": 1661
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.10652832686901093,
+      "learning_rate": 9.460864354755017e-07,
+      "loss": 0.1628,
+      "step": 1662
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.37316522002220154,
+      "learning_rate": 8.896333818280145e-07,
+      "loss": 0.1186,
+      "step": 1663
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.20198196172714233,
+      "learning_rate": 8.349153158774825e-07,
+      "loss": 0.1195,
+      "step": 1664
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.18203914165496826,
+      "learning_rate": 7.8193242783281e-07,
+      "loss": 0.1618,
+      "step": 1665
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.1698591411113739,
+      "learning_rate": 7.306849018708927e-07,
+      "loss": 0.1423,
+      "step": 1666
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.13877595961093903,
+      "learning_rate": 6.811729161363966e-07,
+      "loss": 0.1068,
+      "step": 1667
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.2812526524066925,
+      "learning_rate": 6.333966427409243e-07,
+      "loss": 0.1271,
+      "step": 1668
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.2276443988084793,
+      "learning_rate": 5.873562477624605e-07,
+      "loss": 0.1471,
+      "step": 1669
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 0.09518153220415115,
+      "learning_rate": 5.430518912448168e-07,
+      "loss": 0.1251,
+      "step": 1670
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.0484699048101902,
+      "learning_rate": 5.004837271970764e-07,
+      "loss": 0.115,
+      "step": 1671
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.09272460639476776,
+      "learning_rate": 4.596519035929281e-07,
+      "loss": 0.1178,
+      "step": 1672
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.1327453851699829,
+      "learning_rate": 4.2055656237038884e-07,
+      "loss": 0.1252,
+      "step": 1673
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.10774451494216919,
+      "learning_rate": 3.83197839431082e-07,
+      "loss": 0.1243,
+      "step": 1674
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.05414144694805145,
+      "learning_rate": 3.475758646400151e-07,
+      "loss": 0.1254,
+      "step": 1675
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.05690954253077507,
+      "learning_rate": 3.1369076182480305e-07,
+      "loss": 0.1227,
+      "step": 1676
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.02989388071000576,
+      "learning_rate": 2.815426487755568e-07,
+      "loss": 0.1144,
+      "step": 1677
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.07636480778455734,
+      "learning_rate": 2.5113163724427293e-07,
+      "loss": 0.1264,
+      "step": 1678
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.14926138520240784,
+      "learning_rate": 2.2245783294444488e-07,
+      "loss": 0.1386,
+      "step": 1679
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.20612908899784088,
+      "learning_rate": 1.9552133555084116e-07,
+      "loss": 0.1107,
+      "step": 1680
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 0.16899043321609497,
+      "learning_rate": 1.7032223869911656e-07,
+      "loss": 0.1392,
+      "step": 1681
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.03907778859138489,
+      "learning_rate": 1.4686062998525706e-07,
+      "loss": 0.1191,
+      "step": 1682
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.05882280319929123,
+      "learning_rate": 1.2513659096569097e-07,
+      "loss": 0.1223,
+      "step": 1683
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.09647829085588455,
+      "learning_rate": 1.0515019715656716e-07,
+      "loss": 0.1278,
+      "step": 1684
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.15963368117809296,
+      "learning_rate": 8.690151803386614e-08,
+      "loss": 0.1391,
+      "step": 1685
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.13670912384986877,
+      "learning_rate": 7.0390617032845e-08,
+      "loss": 0.1424,
+      "step": 1686
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 0.1605810821056366,
+      "learning_rate": 5.5617551548148294e-08,
+      "loss": 0.1438,
+      "step": 1687
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.15910910069942474,
+      "learning_rate": 4.258237293325307e-08,
+      "loss": 0.1518,
+      "step": 1688
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.09237740933895111,
+      "learning_rate": 3.1285126500579795e-08,
+      "loss": 0.1212,
+      "step": 1689
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.05058419331908226,
+      "learning_rate": 2.1725851521103845e-08,
+      "loss": 0.1282,
+      "step": 1690
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.03350934758782387,
+      "learning_rate": 1.3904581224410962e-08,
+      "loss": 0.1285,
+      "step": 1691
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 0.11487533152103424,
+      "learning_rate": 7.821342798530751e-09,
+      "loss": 0.1352,
+      "step": 1692
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.03711562231183052,
+      "learning_rate": 3.4761573897701404e-09,
+      "loss": 0.1145,
+      "step": 1693
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.0804651752114296,
+      "learning_rate": 8.690401026578698e-10,
+      "loss": 0.1144,
+      "step": 1694
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.13091543316841125,
+      "learning_rate": 0.0,
+      "loss": 0.127,
+      "step": 1695
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 1695,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 565,
+  "total_flos": 1.549439947809751e+17,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}