diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5494 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.0,
+  "eval_steps": 500,
+  "global_step": 780,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.005148005148005148,
+      "grad_norm": 30.033884048461914,
+      "learning_rate": 0.0,
+      "loss": 0.7068,
+      "step": 1
+    },
+    {
+      "epoch": 0.010296010296010296,
+      "grad_norm": 44.104042053222656,
+      "learning_rate": 2.061855670103093e-06,
+      "loss": 1.3248,
+      "step": 2
+    },
+    {
+      "epoch": 0.015444015444015444,
+      "grad_norm": 32.23274612426758,
+      "learning_rate": 4.123711340206186e-06,
+      "loss": 0.9166,
+      "step": 3
+    },
+    {
+      "epoch": 0.02059202059202059,
+      "grad_norm": 92.9388427734375,
+      "learning_rate": 6.185567010309279e-06,
+      "loss": 1.1103,
+      "step": 4
+    },
+    {
+      "epoch": 0.02574002574002574,
+      "grad_norm": 32.028648376464844,
+      "learning_rate": 8.247422680412371e-06,
+      "loss": 1.3597,
+      "step": 5
+    },
+    {
+      "epoch": 0.03088803088803089,
+      "grad_norm": 34.139163970947266,
+      "learning_rate": 1.0309278350515464e-05,
+      "loss": 1.2249,
+      "step": 6
+    },
+    {
+      "epoch": 0.036036036036036036,
+      "grad_norm": 33.96830368041992,
+      "learning_rate": 1.2371134020618558e-05,
+      "loss": 0.5991,
+      "step": 7
+    },
+    {
+      "epoch": 0.04118404118404118,
+      "grad_norm": 16.034446716308594,
+      "learning_rate": 1.4432989690721649e-05,
+      "loss": 0.4935,
+      "step": 8
+    },
+    {
+      "epoch": 0.04633204633204633,
+      "grad_norm": 26.77821159362793,
+      "learning_rate": 1.6494845360824743e-05,
+      "loss": 0.5649,
+      "step": 9
+    },
+    {
+      "epoch": 0.05148005148005148,
+      "grad_norm": 37.52692794799805,
+      "learning_rate": 1.8556701030927837e-05,
+      "loss": 0.3199,
+      "step": 10
+    },
+    {
+      "epoch": 0.05662805662805663,
+      "grad_norm": 14.663997650146484,
+      "learning_rate": 2.0618556701030927e-05,
+      "loss": 0.4138,
+      "step": 11
+    },
+    {
+      "epoch": 0.06177606177606178,
+      "grad_norm": 10.800355911254883,
+      "learning_rate": 2.268041237113402e-05,
+      "loss": 0.2719,
+      "step": 12
+    },
+    {
+      "epoch": 0.06692406692406692,
+      "grad_norm": 10.981304168701172,
+      "learning_rate": 2.4742268041237116e-05,
+      "loss": 0.3439,
+      "step": 13
+    },
+    {
+      "epoch": 0.07207207207207207,
+      "grad_norm": 18.021671295166016,
+      "learning_rate": 2.6804123711340206e-05,
+      "loss": 0.4027,
+      "step": 14
+    },
+    {
+      "epoch": 0.07722007722007722,
+      "grad_norm": 16.514860153198242,
+      "learning_rate": 2.8865979381443297e-05,
+      "loss": 0.3471,
+      "step": 15
+    },
+    {
+      "epoch": 0.08236808236808237,
+      "grad_norm": 9.517114639282227,
+      "learning_rate": 3.0927835051546395e-05,
+      "loss": 0.3127,
+      "step": 16
+    },
+    {
+      "epoch": 0.08751608751608751,
+      "grad_norm": 5.607189178466797,
+      "learning_rate": 3.2989690721649485e-05,
+      "loss": 0.2454,
+      "step": 17
+    },
+    {
+      "epoch": 0.09266409266409266,
+      "grad_norm": 11.989208221435547,
+      "learning_rate": 3.5051546391752576e-05,
+      "loss": 0.4909,
+      "step": 18
+    },
+    {
+      "epoch": 0.0978120978120978,
+      "grad_norm": 8.628355026245117,
+      "learning_rate": 3.7113402061855674e-05,
+      "loss": 0.2838,
+      "step": 19
+    },
+    {
+      "epoch": 0.10296010296010295,
+      "grad_norm": 5.067573547363281,
+      "learning_rate": 3.9175257731958764e-05,
+      "loss": 0.1443,
+      "step": 20
+    },
+    {
+      "epoch": 0.10810810810810811,
+      "grad_norm": 14.469016075134277,
+      "learning_rate": 4.1237113402061855e-05,
+      "loss": 0.2668,
+      "step": 21
+    },
+    {
+      "epoch": 0.11325611325611326,
+      "grad_norm": 11.140338897705078,
+      "learning_rate": 4.329896907216495e-05,
+      "loss": 0.6323,
+      "step": 22
+    },
+    {
+      "epoch": 0.11840411840411841,
+      "grad_norm": 6.364691257476807,
+      "learning_rate": 4.536082474226804e-05,
+      "loss": 0.1889,
+      "step": 23
+    },
+    {
+      "epoch": 0.12355212355212356,
+      "grad_norm": 4.331116676330566,
+      "learning_rate": 4.7422680412371134e-05,
+      "loss": 0.0984,
+      "step": 24
+    },
+    {
+      "epoch": 0.1287001287001287,
+      "grad_norm": 8.377317428588867,
+      "learning_rate": 4.948453608247423e-05,
+      "loss": 0.1589,
+      "step": 25
+    },
+    {
+      "epoch": 0.13384813384813385,
+      "grad_norm": 85.1617660522461,
+      "learning_rate": 5.1546391752577315e-05,
+      "loss": 0.4123,
+      "step": 26
+    },
+    {
+      "epoch": 0.138996138996139,
+      "grad_norm": 4.7678303718566895,
+      "learning_rate": 5.360824742268041e-05,
+      "loss": 0.1103,
+      "step": 27
+    },
+    {
+      "epoch": 0.14414414414414414,
+      "grad_norm": 10.029397964477539,
+      "learning_rate": 5.567010309278351e-05,
+      "loss": 0.3802,
+      "step": 28
+    },
+    {
+      "epoch": 0.1492921492921493,
+      "grad_norm": 8.106687545776367,
+      "learning_rate": 5.7731958762886594e-05,
+      "loss": 0.3531,
+      "step": 29
+    },
+    {
+      "epoch": 0.15444015444015444,
+      "grad_norm": 3.803243398666382,
+      "learning_rate": 5.979381443298969e-05,
+      "loss": 0.1492,
+      "step": 30
+    },
+    {
+      "epoch": 0.15958815958815958,
+      "grad_norm": 5.68239164352417,
+      "learning_rate": 6.185567010309279e-05,
+      "loss": 0.2362,
+      "step": 31
+    },
+    {
+      "epoch": 0.16473616473616473,
+      "grad_norm": 12.547609329223633,
+      "learning_rate": 6.391752577319587e-05,
+      "loss": 0.2034,
+      "step": 32
+    },
+    {
+      "epoch": 0.16988416988416988,
+      "grad_norm": 3.7096502780914307,
+      "learning_rate": 6.597938144329897e-05,
+      "loss": 0.137,
+      "step": 33
+    },
+    {
+      "epoch": 0.17503217503217502,
+      "grad_norm": 23.394716262817383,
+      "learning_rate": 6.804123711340207e-05,
+      "loss": 0.4918,
+      "step": 34
+    },
+    {
+      "epoch": 0.18018018018018017,
+      "grad_norm": 42.18636703491211,
+      "learning_rate": 7.010309278350515e-05,
+      "loss": 0.4892,
+      "step": 35
+    },
+    {
+      "epoch": 0.18532818532818532,
+      "grad_norm": 7.870489597320557,
+      "learning_rate": 7.216494845360825e-05,
+      "loss": 0.158,
+      "step": 36
+    },
+    {
+      "epoch": 0.19047619047619047,
+      "grad_norm": 9.5137939453125,
+      "learning_rate": 7.422680412371135e-05,
+      "loss": 0.473,
+      "step": 37
+    },
+    {
+      "epoch": 0.1956241956241956,
+      "grad_norm": 5.566166877746582,
+      "learning_rate": 7.628865979381443e-05,
+      "loss": 0.2825,
+      "step": 38
+    },
+    {
+      "epoch": 0.20077220077220076,
+      "grad_norm": 8.659141540527344,
+      "learning_rate": 7.835051546391753e-05,
+      "loss": 0.2889,
+      "step": 39
+    },
+    {
+      "epoch": 0.2059202059202059,
+      "grad_norm": 8.041424751281738,
+      "learning_rate": 8.041237113402063e-05,
+      "loss": 0.358,
+      "step": 40
+    },
+    {
+      "epoch": 0.21106821106821108,
+      "grad_norm": 3.9942994117736816,
+      "learning_rate": 8.247422680412371e-05,
+      "loss": 0.1863,
+      "step": 41
+    },
+    {
+      "epoch": 0.21621621621621623,
+      "grad_norm": 12.74262809753418,
+      "learning_rate": 8.453608247422681e-05,
+      "loss": 0.6432,
+      "step": 42
+    },
+    {
+      "epoch": 0.22136422136422138,
+      "grad_norm": 13.1393461227417,
+      "learning_rate": 8.65979381443299e-05,
+      "loss": 0.4567,
+      "step": 43
+    },
+    {
+      "epoch": 0.22651222651222652,
+      "grad_norm": 10.950636863708496,
+      "learning_rate": 8.865979381443299e-05,
+      "loss": 0.3148,
+      "step": 44
+    },
+    {
+      "epoch": 0.23166023166023167,
+      "grad_norm": 4.720818519592285,
+      "learning_rate": 9.072164948453609e-05,
+      "loss": 0.2311,
+      "step": 45
+    },
+    {
+      "epoch": 0.23680823680823682,
+      "grad_norm": 3.089749336242676,
+      "learning_rate": 9.278350515463918e-05,
+      "loss": 0.2013,
+      "step": 46
+    },
+    {
+      "epoch": 0.24195624195624196,
+      "grad_norm": 1.8420156240463257,
+      "learning_rate": 9.484536082474227e-05,
+      "loss": 0.0402,
+      "step": 47
+    },
+    {
+      "epoch": 0.2471042471042471,
+      "grad_norm": 8.579316139221191,
+      "learning_rate": 9.690721649484537e-05,
+      "loss": 0.2985,
+      "step": 48
+    },
+    {
+      "epoch": 0.25225225225225223,
+      "grad_norm": 5.109619140625,
+      "learning_rate": 9.896907216494846e-05,
+      "loss": 0.1551,
+      "step": 49
+    },
+    {
+      "epoch": 0.2574002574002574,
+      "grad_norm": 2.1247408390045166,
+      "learning_rate": 0.00010103092783505155,
+      "loss": 0.0796,
+      "step": 50
+    },
+    {
+      "epoch": 0.2625482625482625,
+      "grad_norm": 9.897648811340332,
+      "learning_rate": 0.00010309278350515463,
+      "loss": 0.3831,
+      "step": 51
+    },
+    {
+      "epoch": 0.2676962676962677,
+      "grad_norm": 9.84504222869873,
+      "learning_rate": 0.00010515463917525774,
+      "loss": 0.2121,
+      "step": 52
+    },
+    {
+      "epoch": 0.2728442728442728,
+      "grad_norm": 4.189642906188965,
+      "learning_rate": 0.00010721649484536083,
+      "loss": 0.064,
+      "step": 53
+    },
+    {
+      "epoch": 0.277992277992278,
+      "grad_norm": 12.464111328125,
+      "learning_rate": 0.00010927835051546391,
+      "loss": 0.1176,
+      "step": 54
+    },
+    {
+      "epoch": 0.28314028314028317,
+      "grad_norm": 3.741978168487549,
+      "learning_rate": 0.00011134020618556702,
+      "loss": 0.1495,
+      "step": 55
+    },
+    {
+      "epoch": 0.2882882882882883,
+      "grad_norm": 6.447311878204346,
+      "learning_rate": 0.0001134020618556701,
+      "loss": 0.2374,
+      "step": 56
+    },
+    {
+      "epoch": 0.29343629343629346,
+      "grad_norm": 8.749642372131348,
+      "learning_rate": 0.00011546391752577319,
+      "loss": 0.2342,
+      "step": 57
+    },
+    {
+      "epoch": 0.2985842985842986,
+      "grad_norm": 4.203049659729004,
+      "learning_rate": 0.0001175257731958763,
+      "loss": 0.1135,
+      "step": 58
+    },
+    {
+      "epoch": 0.30373230373230375,
+      "grad_norm": 12.240730285644531,
+      "learning_rate": 0.00011958762886597938,
+      "loss": 0.1644,
+      "step": 59
+    },
+    {
+      "epoch": 0.3088803088803089,
+      "grad_norm": 2.5101993083953857,
+      "learning_rate": 0.00012164948453608247,
+      "loss": 0.0574,
+      "step": 60
+    },
+    {
+      "epoch": 0.31402831402831405,
+      "grad_norm": 4.4976043701171875,
+      "learning_rate": 0.00012371134020618558,
+      "loss": 0.1284,
+      "step": 61
+    },
+    {
+      "epoch": 0.31917631917631917,
+      "grad_norm": 12.019474983215332,
+      "learning_rate": 0.00012577319587628866,
+      "loss": 0.5396,
+      "step": 62
+    },
+    {
+      "epoch": 0.32432432432432434,
+      "grad_norm": 4.723063945770264,
+      "learning_rate": 0.00012783505154639175,
+      "loss": 0.143,
+      "step": 63
+    },
+    {
+      "epoch": 0.32947232947232946,
+      "grad_norm": 19.409879684448242,
+      "learning_rate": 0.00012989690721649486,
+      "loss": 0.181,
+      "step": 64
+    },
+    {
+      "epoch": 0.33462033462033464,
+      "grad_norm": 7.176737308502197,
+      "learning_rate": 0.00013195876288659794,
+      "loss": 0.3821,
+      "step": 65
+    },
+    {
+      "epoch": 0.33976833976833976,
+      "grad_norm": 11.919947624206543,
+      "learning_rate": 0.00013402061855670103,
+      "loss": 0.3292,
+      "step": 66
+    },
+    {
+      "epoch": 0.34491634491634493,
+      "grad_norm": 22.050472259521484,
+      "learning_rate": 0.00013608247422680414,
+      "loss": 0.5224,
+      "step": 67
+    },
+    {
+      "epoch": 0.35006435006435005,
+      "grad_norm": 7.246392250061035,
+      "learning_rate": 0.00013814432989690722,
+      "loss": 0.4381,
+      "step": 68
+    },
+    {
+      "epoch": 0.3552123552123552,
+      "grad_norm": 4.389172077178955,
+      "learning_rate": 0.0001402061855670103,
+      "loss": 0.2367,
+      "step": 69
+    },
+    {
+      "epoch": 0.36036036036036034,
+      "grad_norm": 3.4686994552612305,
+      "learning_rate": 0.00014226804123711342,
+      "loss": 0.1812,
+      "step": 70
+    },
+    {
+      "epoch": 0.3655083655083655,
+      "grad_norm": 2.404954433441162,
+      "learning_rate": 0.0001443298969072165,
+      "loss": 0.0997,
+      "step": 71
+    },
+    {
+      "epoch": 0.37065637065637064,
+      "grad_norm": 2.276104211807251,
+      "learning_rate": 0.00014639175257731958,
+      "loss": 0.1073,
+      "step": 72
+    },
+    {
+      "epoch": 0.3758043758043758,
+      "grad_norm": 3.871577262878418,
+      "learning_rate": 0.0001484536082474227,
+      "loss": 0.0802,
+      "step": 73
+    },
+    {
+      "epoch": 0.38095238095238093,
+      "grad_norm": 3.5958058834075928,
+      "learning_rate": 0.00015051546391752578,
+      "loss": 0.08,
+      "step": 74
+    },
+    {
+      "epoch": 0.3861003861003861,
+      "grad_norm": 13.763484954833984,
+      "learning_rate": 0.00015257731958762886,
+      "loss": 0.7291,
+      "step": 75
+    },
+    {
+      "epoch": 0.3912483912483912,
+      "grad_norm": 10.148221015930176,
+      "learning_rate": 0.00015463917525773197,
+      "loss": 0.6071,
+      "step": 76
+    },
+    {
+      "epoch": 0.3963963963963964,
+      "grad_norm": 2.279761552810669,
+      "learning_rate": 0.00015670103092783506,
+      "loss": 0.0242,
+      "step": 77
+    },
+    {
+      "epoch": 0.4015444015444015,
+      "grad_norm": 9.17572021484375,
+      "learning_rate": 0.00015876288659793814,
+      "loss": 0.341,
+      "step": 78
+    },
+    {
+      "epoch": 0.4066924066924067,
+      "grad_norm": 1.0897811651229858,
+      "learning_rate": 0.00016082474226804125,
+      "loss": 0.0143,
+      "step": 79
+    },
+    {
+      "epoch": 0.4118404118404118,
+      "grad_norm": 11.125105857849121,
+      "learning_rate": 0.00016288659793814434,
+      "loss": 0.6037,
+      "step": 80
+    },
+    {
+      "epoch": 0.416988416988417,
+      "grad_norm": 6.2225661277771,
+      "learning_rate": 0.00016494845360824742,
+      "loss": 0.3567,
+      "step": 81
+    },
+    {
+      "epoch": 0.42213642213642216,
+      "grad_norm": 9.033268928527832,
+      "learning_rate": 0.00016701030927835053,
+      "loss": 0.2895,
+      "step": 82
+    },
+    {
+      "epoch": 0.4272844272844273,
+      "grad_norm": 20.892881393432617,
+      "learning_rate": 0.00016907216494845361,
+      "loss": 0.1748,
+      "step": 83
+    },
+    {
+      "epoch": 0.43243243243243246,
+      "grad_norm": 13.1263427734375,
+      "learning_rate": 0.0001711340206185567,
+      "loss": 0.3571,
+      "step": 84
+    },
+    {
+      "epoch": 0.4375804375804376,
+      "grad_norm": 6.22101354598999,
+      "learning_rate": 0.0001731958762886598,
+      "loss": 0.1215,
+      "step": 85
+    },
+    {
+      "epoch": 0.44272844272844275,
+      "grad_norm": 5.855807304382324,
+      "learning_rate": 0.0001752577319587629,
+      "loss": 0.292,
+      "step": 86
+    },
+    {
+      "epoch": 0.44787644787644787,
+      "grad_norm": 23.522438049316406,
+      "learning_rate": 0.00017731958762886598,
+      "loss": 0.5337,
+      "step": 87
+    },
+    {
+      "epoch": 0.45302445302445304,
+      "grad_norm": 4.277527332305908,
+      "learning_rate": 0.0001793814432989691,
+      "loss": 0.0837,
+      "step": 88
+    },
+    {
+      "epoch": 0.45817245817245816,
+      "grad_norm": 4.319547176361084,
+      "learning_rate": 0.00018144329896907217,
+      "loss": 0.1263,
+      "step": 89
+    },
+    {
+      "epoch": 0.46332046332046334,
+      "grad_norm": 11.513567924499512,
+      "learning_rate": 0.00018350515463917526,
+      "loss": 0.3003,
+      "step": 90
+    },
+    {
+      "epoch": 0.46846846846846846,
+      "grad_norm": 11.12172794342041,
+      "learning_rate": 0.00018556701030927837,
+      "loss": 0.237,
+      "step": 91
+    },
+    {
+      "epoch": 0.47361647361647363,
+      "grad_norm": 23.389734268188477,
+      "learning_rate": 0.00018762886597938145,
+      "loss": 0.1702,
+      "step": 92
+    },
+    {
+      "epoch": 0.47876447876447875,
+      "grad_norm": 6.61017370223999,
+      "learning_rate": 0.00018969072164948454,
+      "loss": 0.7371,
+      "step": 93
+    },
+    {
+      "epoch": 0.4839124839124839,
+      "grad_norm": 5.59273099899292,
+      "learning_rate": 0.00019175257731958765,
+      "loss": 0.1608,
+      "step": 94
+    },
+    {
+      "epoch": 0.48906048906048905,
+      "grad_norm": 11.069086074829102,
+      "learning_rate": 0.00019381443298969073,
+      "loss": 0.2826,
+      "step": 95
+    },
+    {
+      "epoch": 0.4942084942084942,
+      "grad_norm": 7.276542663574219,
+      "learning_rate": 0.00019587628865979381,
+      "loss": 0.2525,
+      "step": 96
+    },
+    {
+      "epoch": 0.49935649935649934,
+      "grad_norm": 2.1657791137695312,
+      "learning_rate": 0.00019793814432989693,
+      "loss": 0.0535,
+      "step": 97
+    },
+    {
+      "epoch": 0.5045045045045045,
+      "grad_norm": 4.67576265335083,
+      "learning_rate": 0.0002,
+      "loss": 0.2473,
+      "step": 98
+    },
+    {
+      "epoch": 0.5096525096525096,
+      "grad_norm": 4.956919193267822,
+      "learning_rate": 0.00019999935249847796,
+      "loss": 0.329,
+      "step": 99
+    },
+    {
+      "epoch": 0.5148005148005148,
+      "grad_norm": 10.064221382141113,
+      "learning_rate": 0.00019999741000229694,
+      "loss": 0.2484,
+      "step": 100
+    },
+    {
+      "epoch": 0.51994851994852,
+      "grad_norm": 8.970917701721191,
+      "learning_rate": 0.00019999417253661235,
+      "loss": 0.1831,
+      "step": 101
+    },
+    {
+      "epoch": 0.525096525096525,
+      "grad_norm": 2.5499494075775146,
+      "learning_rate": 0.00019998964014334946,
+      "loss": 0.0392,
+      "step": 102
+    },
+    {
+      "epoch": 0.5302445302445302,
+      "grad_norm": 4.503442287445068,
+      "learning_rate": 0.00019998381288120295,
+      "loss": 0.5484,
+      "step": 103
+    },
+    {
+      "epoch": 0.5353925353925354,
+      "grad_norm": 11.03059196472168,
+      "learning_rate": 0.00019997669082563597,
+      "loss": 0.7209,
+      "step": 104
+    },
+    {
+      "epoch": 0.5405405405405406,
+      "grad_norm": 6.97821044921875,
+      "learning_rate": 0.0001999682740688794,
+      "loss": 0.1959,
+      "step": 105
+    },
+    {
+      "epoch": 0.5456885456885456,
+      "grad_norm": 1.7897404432296753,
+      "learning_rate": 0.0001999585627199305,
+      "loss": 0.0666,
+      "step": 106
+    },
+    {
+      "epoch": 0.5508365508365508,
+      "grad_norm": 3.4306905269622803,
+      "learning_rate": 0.00019994755690455152,
+      "loss": 0.2432,
+      "step": 107
+    },
+    {
+      "epoch": 0.555984555984556,
+      "grad_norm": 6.274919509887695,
+      "learning_rate": 0.00019993525676526805,
+      "loss": 0.2695,
+      "step": 108
+    },
+    {
+      "epoch": 0.5611325611325612,
+      "grad_norm": 4.889686584472656,
+      "learning_rate": 0.00019992166246136738,
+      "loss": 0.2152,
+      "step": 109
+    },
+    {
+      "epoch": 0.5662805662805663,
+      "grad_norm": 6.967682838439941,
+      "learning_rate": 0.00019990677416889608,
+      "loss": 0.4346,
+      "step": 110
+    },
+    {
+      "epoch": 0.5714285714285714,
+      "grad_norm": 7.712956428527832,
+      "learning_rate": 0.000199890592080658,
+      "loss": 0.2464,
+      "step": 111
+    },
+    {
+      "epoch": 0.5765765765765766,
+      "grad_norm": 6.341039657592773,
+      "learning_rate": 0.00019987311640621173,
+      "loss": 0.2062,
+      "step": 112
+    },
+    {
+      "epoch": 0.5817245817245817,
+      "grad_norm": 5.007922649383545,
+      "learning_rate": 0.0001998543473718677,
+      "loss": 0.2128,
+      "step": 113
+    },
+    {
+      "epoch": 0.5868725868725869,
+      "grad_norm": 6.674788475036621,
+      "learning_rate": 0.00019983428522068556,
+      "loss": 0.1843,
+      "step": 114
+    },
+    {
+      "epoch": 0.592020592020592,
+      "grad_norm": 6.240501880645752,
+      "learning_rate": 0.0001998129302124707,
+      "loss": 0.2693,
+      "step": 115
+    },
+    {
+      "epoch": 0.5971685971685972,
+      "grad_norm": 8.363272666931152,
+      "learning_rate": 0.00019979028262377118,
+      "loss": 0.1871,
+      "step": 116
+    },
+    {
+      "epoch": 0.6023166023166023,
+      "grad_norm": 5.811964988708496,
+      "learning_rate": 0.00019976634274787393,
+      "loss": 0.3791,
+      "step": 117
+    },
+    {
+      "epoch": 0.6074646074646075,
+      "grad_norm": 1.003872036933899,
+      "learning_rate": 0.00019974111089480112,
+      "loss": 0.0208,
+      "step": 118
+    },
+    {
+      "epoch": 0.6126126126126126,
+      "grad_norm": 3.447148084640503,
+      "learning_rate": 0.00019971458739130598,
+      "loss": 0.0435,
+      "step": 119
+    },
+    {
+      "epoch": 0.6177606177606177,
+      "grad_norm": 5.764918804168701,
+      "learning_rate": 0.00019968677258086866,
+      "loss": 0.081,
+      "step": 120
+    },
+    {
+      "epoch": 0.6229086229086229,
+      "grad_norm": 13.532068252563477,
+      "learning_rate": 0.00019965766682369186,
+      "loss": 0.0599,
+      "step": 121
+    },
+    {
+      "epoch": 0.6280566280566281,
+      "grad_norm": 13.2008638381958,
+      "learning_rate": 0.000199627270496696,
+      "loss": 0.7599,
+      "step": 122
+    },
+    {
+      "epoch": 0.6332046332046332,
+      "grad_norm": 9.765315055847168,
+      "learning_rate": 0.00019959558399351444,
+      "loss": 0.0808,
+      "step": 123
+    },
+    {
+      "epoch": 0.6383526383526383,
+      "grad_norm": 21.461593627929688,
+      "learning_rate": 0.00019956260772448833,
+      "loss": 0.775,
+      "step": 124
+    },
+    {
+      "epoch": 0.6435006435006435,
+      "grad_norm": 5.510493278503418,
+      "learning_rate": 0.0001995283421166614,
+      "loss": 0.0483,
+      "step": 125
+    },
+    {
+      "epoch": 0.6486486486486487,
+      "grad_norm": 7.267975330352783,
+      "learning_rate": 0.0001994927876137743,
+      "loss": 0.2109,
+      "step": 126
+    },
+    {
+      "epoch": 0.6537966537966537,
+      "grad_norm": 48.387664794921875,
+      "learning_rate": 0.00019945594467625895,
+      "loss": 0.3282,
+      "step": 127
+    },
+    {
+      "epoch": 0.6589446589446589,
+      "grad_norm": 7.07933235168457,
+      "learning_rate": 0.00019941781378123244,
+      "loss": 0.0902,
+      "step": 128
+    },
+    {
+      "epoch": 0.6640926640926641,
+      "grad_norm": 8.03501033782959,
+      "learning_rate": 0.00019937839542249108,
+      "loss": 0.1174,
+      "step": 129
+    },
+    {
+      "epoch": 0.6692406692406693,
+      "grad_norm": 8.442209243774414,
+      "learning_rate": 0.0001993376901105038,
+      "loss": 0.0573,
+      "step": 130
+    },
+    {
+      "epoch": 0.6743886743886743,
+      "grad_norm": 4.149614334106445,
+      "learning_rate": 0.00019929569837240564,
+      "loss": 0.0608,
+      "step": 131
+    },
+    {
+      "epoch": 0.6795366795366795,
+      "grad_norm": 8.750178337097168,
+      "learning_rate": 0.0001992524207519909,
+      "loss": 0.6886,
+      "step": 132
+    },
+    {
+      "epoch": 0.6846846846846847,
+      "grad_norm": 2.661649703979492,
+      "learning_rate": 0.00019920785780970604,
+      "loss": 0.0466,
+      "step": 133
+    },
+    {
+      "epoch": 0.6898326898326899,
+      "grad_norm": 16.77633285522461,
+      "learning_rate": 0.00019916201012264254,
+      "loss": 0.9548,
+      "step": 134
+    },
+    {
+      "epoch": 0.694980694980695,
+      "grad_norm": 7.334346294403076,
+      "learning_rate": 0.00019911487828452932,
+      "loss": 0.2075,
+      "step": 135
+    },
+    {
+      "epoch": 0.7001287001287001,
+      "grad_norm": 3.5749030113220215,
+      "learning_rate": 0.00019906646290572514,
+      "loss": 0.1089,
+      "step": 136
+    },
+    {
+      "epoch": 0.7052767052767053,
+      "grad_norm": 1.5380395650863647,
+      "learning_rate": 0.00019901676461321068,
+      "loss": 0.035,
+      "step": 137
+    },
+    {
+      "epoch": 0.7104247104247104,
+      "grad_norm": 8.098177909851074,
+      "learning_rate": 0.00019896578405058027,
+      "loss": 0.3158,
+      "step": 138
+    },
+    {
+      "epoch": 0.7155727155727156,
+      "grad_norm": 6.486197471618652,
+      "learning_rate": 0.00019891352187803376,
+      "loss": 0.2323,
+      "step": 139
+    },
+    {
+      "epoch": 0.7207207207207207,
+      "grad_norm": 3.673724889755249,
+      "learning_rate": 0.00019885997877236788,
+      "loss": 0.0765,
+      "step": 140
+    },
+    {
+      "epoch": 0.7258687258687259,
+      "grad_norm": 0.7524272203445435,
+      "learning_rate": 0.0001988051554269675,
+      "loss": 0.0143,
+      "step": 141
+    },
+    {
+      "epoch": 0.731016731016731,
+      "grad_norm": 8.265993118286133,
+      "learning_rate": 0.0001987490525517966,
+      "loss": 0.2867,
+      "step": 142
+    },
+    {
+      "epoch": 0.7361647361647362,
+      "grad_norm": 1.3028749227523804,
+      "learning_rate": 0.00019869167087338907,
+      "loss": 0.0154,
+      "step": 143
+    },
+    {
+      "epoch": 0.7413127413127413,
+      "grad_norm": 2.202714204788208,
+      "learning_rate": 0.00019863301113483948,
+      "loss": 0.0362,
+      "step": 144
+    },
+    {
+      "epoch": 0.7464607464607464,
+      "grad_norm": 5.332104206085205,
+      "learning_rate": 0.00019857307409579318,
+      "loss": 0.0697,
+      "step": 145
+    },
+    {
+      "epoch": 0.7516087516087516,
+      "grad_norm": 8.443161964416504,
+      "learning_rate": 0.00019851186053243666,
+      "loss": 0.2722,
+      "step": 146
+    },
+    {
+      "epoch": 0.7567567567567568,
+      "grad_norm": 13.74390983581543,
+      "learning_rate": 0.0001984493712374874,
+      "loss": 0.3142,
+      "step": 147
+    },
+    {
+      "epoch": 0.7619047619047619,
+      "grad_norm": 12.741813659667969,
+      "learning_rate": 0.00019838560702018374,
+      "loss": 0.5427,
+      "step": 148
+    },
+    {
+      "epoch": 0.767052767052767,
+      "grad_norm": 7.313443183898926,
+      "learning_rate": 0.00019832056870627417,
+      "loss": 0.0729,
+      "step": 149
+    },
+    {
+      "epoch": 0.7722007722007722,
+      "grad_norm": 1.7197271585464478,
+      "learning_rate": 0.00019825425713800688,
+      "loss": 0.009,
+      "step": 150
+    },
+    {
+      "epoch": 0.7773487773487774,
+      "grad_norm": 16.335969924926758,
+      "learning_rate": 0.00019818667317411865,
+      "loss": 0.2342,
+      "step": 151
+    },
+    {
+      "epoch": 0.7824967824967825,
+      "grad_norm": 5.488504886627197,
+      "learning_rate": 0.0001981178176898239,
+      "loss": 0.012,
+      "step": 152
+    },
+    {
+      "epoch": 0.7876447876447876,
+      "grad_norm": 8.866789817810059,
+      "learning_rate": 0.00019804769157680328,
+      "loss": 0.2584,
+      "step": 153
+    },
+    {
+      "epoch": 0.7927927927927928,
+      "grad_norm": 2.2538764476776123,
+      "learning_rate": 0.000197976295743192,
+      "loss": 0.0206,
+      "step": 154
+    },
+    {
+      "epoch": 0.797940797940798,
+      "grad_norm": 3.658780097961426,
+      "learning_rate": 0.00019790363111356837,
+      "loss": 0.0682,
+      "step": 155
+    },
+    {
+      "epoch": 0.803088803088803,
+      "grad_norm": 5.6560468673706055,
+      "learning_rate": 0.0001978296986289415,
+      "loss": 0.0411,
+      "step": 156
+    },
+    {
+      "epoch": 0.8082368082368082,
+      "grad_norm": 6.51859712600708,
+      "learning_rate": 0.00019775449924673934,
+      "loss": 0.0903,
+      "step": 157
+    },
+    {
+      "epoch": 0.8133848133848134,
+      "grad_norm": 3.372549295425415,
+      "learning_rate": 0.00019767803394079615,
+      "loss": 0.1594,
+      "step": 158
+    },
+    {
+      "epoch": 0.8185328185328186,
+      "grad_norm": 9.931015014648438,
+      "learning_rate": 0.00019760030370134004,
+      "loss": 0.1853,
+      "step": 159
+    },
+    {
+      "epoch": 0.8236808236808236,
+      "grad_norm": 11.29722785949707,
+      "learning_rate": 0.0001975213095349799,
+      "loss": 0.0855,
+      "step": 160
+    },
+    {
+      "epoch": 0.8288288288288288,
+      "grad_norm": 3.9944028854370117,
+      "learning_rate": 0.00019744105246469263,
+      "loss": 0.2021,
+      "step": 161
+    },
+    {
+      "epoch": 0.833976833976834,
+      "grad_norm": 2.6088204383850098,
+      "learning_rate": 0.0001973595335298097,
+      "loss": 0.0377,
+      "step": 162
+    },
+    {
+      "epoch": 0.8391248391248392,
+      "grad_norm": 5.811506271362305,
+      "learning_rate": 0.0001972767537860038,
+      "loss": 0.0626,
+      "step": 163
+    },
+    {
+      "epoch": 0.8442728442728443,
+      "grad_norm": 2.458167314529419,
+      "learning_rate": 0.0001971927143052752,
+      "loss": 0.0416,
+      "step": 164
+    },
+    {
+      "epoch": 0.8494208494208494,
+      "grad_norm": 5.227564334869385,
+      "learning_rate": 0.00019710741617593767,
+      "loss": 0.0489,
+      "step": 165
+    },
+    {
+      "epoch": 0.8545688545688546,
+      "grad_norm": 7.393219470977783,
+      "learning_rate": 0.00019702086050260456,
+      "loss": 0.1106,
+      "step": 166
+    },
+    {
+      "epoch": 0.8597168597168597,
+      "grad_norm": 5.9372053146362305,
+      "learning_rate": 0.00019693304840617457,
+      "loss": 0.0727,
+      "step": 167
+    },
+    {
+      "epoch": 0.8648648648648649,
+      "grad_norm": 11.54053783416748,
+      "learning_rate": 0.00019684398102381694,
+      "loss": 0.2974,
+      "step": 168
+    },
+    {
+      "epoch": 0.87001287001287,
+      "grad_norm": 2.5348217487335205,
+      "learning_rate": 0.000196753659508957,
+      "loss": 0.0342,
+      "step": 169
+    },
+    {
+      "epoch": 0.8751608751608752,
+      "grad_norm": 3.2081925868988037,
+      "learning_rate": 0.00019666208503126112,
+      "loss": 0.0324,
+      "step": 170
+    },
+    {
+      "epoch": 0.8803088803088803,
+      "grad_norm": 15.922517776489258,
+      "learning_rate": 0.0001965692587766216,
+      "loss": 0.5598,
+      "step": 171
+    },
+    {
+      "epoch": 0.8854568854568855,
+      "grad_norm": 0.7327184677124023,
+      "learning_rate": 0.00019647518194714124,
+      "loss": 0.01,
+      "step": 172
+    },
+    {
+      "epoch": 0.8906048906048906,
+      "grad_norm": 9.27102279663086,
+      "learning_rate": 0.00019637985576111778,
+      "loss": 0.0124,
+      "step": 173
+    },
+    {
+      "epoch": 0.8957528957528957,
+      "grad_norm": 10.692123413085938,
+      "learning_rate": 0.00019628328145302837,
+      "loss": 0.173,
+      "step": 174
+    },
+    {
+      "epoch": 0.9009009009009009,
+      "grad_norm": 7.879785537719727,
+      "learning_rate": 0.00019618546027351316,
+      "loss": 0.0765,
+      "step": 175
+    },
+    {
+      "epoch": 0.9060489060489061,
+      "grad_norm": 9.62712574005127,
+      "learning_rate": 0.0001960863934893594,
+      "loss": 0.2634,
+      "step": 176
+    },
+    {
+      "epoch": 0.9111969111969112,
+      "grad_norm": 6.596358776092529,
+      "learning_rate": 0.00019598608238348493,
+      "loss": 0.619,
+      "step": 177
+    },
+    {
+      "epoch": 0.9163449163449163,
+      "grad_norm": 5.179719924926758,
+      "learning_rate": 0.0001958845282549217,
+      "loss": 0.1794,
+      "step": 178
+    },
+    {
+      "epoch": 0.9214929214929215,
+      "grad_norm": 9.3878173828125,
+      "learning_rate": 0.00019578173241879872,
+      "loss": 0.3477,
+      "step": 179
+    },
+    {
+      "epoch": 0.9266409266409267,
+      "grad_norm": 7.18900728225708,
+      "learning_rate": 0.0001956776962063252,
+      "loss": 0.1673,
+      "step": 180
+    },
+    {
+      "epoch": 0.9317889317889317,
+      "grad_norm": 2.2193684577941895,
+      "learning_rate": 0.00019557242096477327,
+      "loss": 0.0267,
+      "step": 181
+    },
+    {
+      "epoch": 0.9369369369369369,
+      "grad_norm": 5.781956195831299,
+      "learning_rate": 0.00019546590805746052,
+      "loss": 0.2195,
+      "step": 182
+    },
+    {
+      "epoch": 0.9420849420849421,
+      "grad_norm": 15.61030101776123,
+      "learning_rate": 0.00019535815886373233,
+      "loss": 0.3574,
+      "step": 183
+    },
+    {
+      "epoch": 0.9472329472329473,
+      "grad_norm": 0.5154070258140564,
+      "learning_rate": 0.00019524917477894404,
+      "loss": 0.0086,
+      "step": 184
+    },
+    {
+      "epoch": 0.9523809523809523,
+      "grad_norm": 5.88370418548584,
+      "learning_rate": 0.00019513895721444286,
+      "loss": 0.4382,
+      "step": 185
+    },
+    {
+      "epoch": 0.9575289575289575,
+      "grad_norm": 1.9130569696426392,
+      "learning_rate": 0.00019502750759754962,
+      "loss": 0.0576,
+      "step": 186
+    },
+    {
+      "epoch": 0.9626769626769627,
+      "grad_norm": 2.4080262184143066,
+      "learning_rate": 0.0001949148273715402,
+      "loss": 0.0744,
+      "step": 187
+    },
+    {
+      "epoch": 0.9678249678249679,
+      "grad_norm": 3.948603391647339,
+      "learning_rate": 0.00019480091799562704,
+      "loss": 0.088,
+      "step": 188
+    },
+    {
+      "epoch": 0.972972972972973,
+      "grad_norm": 8.048215866088867,
+      "learning_rate": 0.00019468578094493998,
+      "loss": 0.4397,
+      "step": 189
+    },
+    {
+      "epoch": 0.9781209781209781,
+      "grad_norm": 0.7454734444618225,
+      "learning_rate": 0.00019456941771050734,
+      "loss": 0.0193,
+      "step": 190
+    },
+    {
+      "epoch": 0.9832689832689833,
+      "grad_norm": 6.949063777923584,
+      "learning_rate": 0.00019445182979923654,
+      "loss": 0.4304,
+      "step": 191
+    },
+    {
+      "epoch": 0.9884169884169884,
+      "grad_norm": 7.762242317199707,
+      "learning_rate": 0.00019433301873389463,
+      "loss": 0.2241,
+      "step": 192
+    },
+    {
+      "epoch": 0.9935649935649936,
+      "grad_norm": 6.118680000305176,
+      "learning_rate": 0.00019421298605308847,
+      "loss": 0.2356,
+      "step": 193
+    },
+    {
+      "epoch": 0.9987129987129987,
+      "grad_norm": 6.023503303527832,
+      "learning_rate": 0.000194091733311245,
+      "loss": 0.1854,
+      "step": 194
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 10.273907661437988,
+      "learning_rate": 0.00019396926207859084,
+      "loss": 0.3042,
+      "step": 195
+    },
+    {
+      "epoch": 1.005148005148005,
+      "grad_norm": 4.8654327392578125,
+      "learning_rate": 0.00019384557394113228,
+      "loss": 0.3831,
+      "step": 196
+    },
+    {
+      "epoch": 1.0102960102960103,
+      "grad_norm": 2.4801714420318604,
+      "learning_rate": 0.00019372067050063438,
+      "loss": 0.0892,
+      "step": 197
+    },
+    {
+      "epoch": 1.0154440154440154,
+      "grad_norm": 1.2199461460113525,
+      "learning_rate": 0.00019359455337460054,
+      "loss": 0.0405,
+      "step": 198
+    },
+    {
+      "epoch": 1.0205920205920207,
+      "grad_norm": 2.447180986404419,
+      "learning_rate": 0.00019346722419625136,
+      "loss": 0.0688,
+      "step": 199
+    },
+    {
+      "epoch": 1.0257400257400258,
+      "grad_norm": 0.522395670413971,
+      "learning_rate": 0.0001933386846145036,
+      "loss": 0.013,
+      "step": 200
+    },
+    {
+      "epoch": 1.0308880308880308,
+      "grad_norm": 0.43746644258499146,
+      "learning_rate": 0.00019320893629394873,
+      "loss": 0.0082,
+      "step": 201
+    },
+    {
+      "epoch": 1.0360360360360361,
+      "grad_norm": 6.228302001953125,
+      "learning_rate": 0.00019307798091483145,
+      "loss": 0.0846,
+      "step": 202
+    },
+    {
+      "epoch": 1.0411840411840412,
+      "grad_norm": 2.9194703102111816,
+      "learning_rate": 0.00019294582017302797,
+      "loss": 0.0419,
+      "step": 203
+    },
+    {
+      "epoch": 1.0463320463320462,
+      "grad_norm": 5.213253498077393,
+      "learning_rate": 0.0001928124557800238,
+      "loss": 0.2058,
+      "step": 204
+    },
+    {
+      "epoch": 1.0514800514800515,
+      "grad_norm": 8.838007926940918,
+      "learning_rate": 0.000192677889462892,
+      "loss": 0.1115,
+      "step": 205
+    },
+    {
+      "epoch": 1.0566280566280566,
+      "grad_norm": 5.529503345489502,
+      "learning_rate": 0.00019254212296427044,
+      "loss": 0.0317,
+      "step": 206
+    },
+    {
+      "epoch": 1.0617760617760619,
+      "grad_norm": 9.995556831359863,
+      "learning_rate": 0.00019240515804233937,
+      "loss": 0.3377,
+      "step": 207
+    },
+    {
+      "epoch": 1.066924066924067,
+      "grad_norm": 8.233309745788574,
+      "learning_rate": 0.00019226699647079874,
+      "loss": 0.1549,
+      "step": 208
+    },
+    {
+      "epoch": 1.072072072072072,
+      "grad_norm": 0.22019803524017334,
+      "learning_rate": 0.0001921276400388451,
+      "loss": 0.0015,
+      "step": 209
+    },
+    {
+      "epoch": 1.0772200772200773,
+      "grad_norm": 4.7017822265625,
+      "learning_rate": 0.0001919870905511485,
+      "loss": 0.0334,
+      "step": 210
+    },
+    {
+      "epoch": 1.0823680823680824,
+      "grad_norm": 6.453871726989746,
+      "learning_rate": 0.00019184534982782904,
+      "loss": 0.1731,
+      "step": 211
+    },
+    {
+      "epoch": 1.0875160875160874,
+      "grad_norm": 3.2720937728881836,
+      "learning_rate": 0.00019170241970443343,
+      "loss": 0.0385,
+      "step": 212
+    },
+    {
+      "epoch": 1.0926640926640927,
+      "grad_norm": 3.0594441890716553,
+      "learning_rate": 0.00019155830203191113,
+      "loss": 0.0567,
+      "step": 213
+    },
+    {
+      "epoch": 1.0978120978120978,
+      "grad_norm": 2.0646886825561523,
+      "learning_rate": 0.00019141299867659036,
+      "loss": 0.0374,
+      "step": 214
+    },
+    {
+      "epoch": 1.102960102960103,
+      "grad_norm": 7.499727249145508,
+      "learning_rate": 0.00019126651152015403,
+      "loss": 0.0685,
+      "step": 215
+    },
+    {
+      "epoch": 1.1081081081081081,
+      "grad_norm": 24.703758239746094,
+      "learning_rate": 0.00019111884245961522,
+      "loss": 0.8812,
+      "step": 216
+    },
+    {
+      "epoch": 1.1132561132561132,
+      "grad_norm": 2.5881152153015137,
+      "learning_rate": 0.00019096999340729282,
+      "loss": 0.0383,
+      "step": 217
+    },
+    {
+      "epoch": 1.1184041184041185,
+      "grad_norm": 14.550262451171875,
+      "learning_rate": 0.00019081996629078657,
+      "loss": 0.5696,
+      "step": 218
+    },
+    {
+      "epoch": 1.1235521235521235,
+      "grad_norm": 10.270796775817871,
+      "learning_rate": 0.00019066876305295216,
+      "loss": 0.1258,
+      "step": 219
+    },
+    {
+      "epoch": 1.1287001287001286,
+      "grad_norm": 8.031951904296875,
+      "learning_rate": 0.00019051638565187616,
+      "loss": 0.6005,
+      "step": 220
+    },
+    {
+      "epoch": 1.1338481338481339,
+      "grad_norm": 21.7176570892334,
+      "learning_rate": 0.00019036283606085053,
+      "loss": 0.6916,
+      "step": 221
+    },
+    {
+      "epoch": 1.138996138996139,
+      "grad_norm": 4.141570568084717,
+      "learning_rate": 0.0001902081162683472,
+      "loss": 0.0989,
+      "step": 222
+    },
+    {
+      "epoch": 1.1441441441441442,
+      "grad_norm": 3.6948814392089844,
+      "learning_rate": 0.00019005222827799212,
+      "loss": 0.0722,
+      "step": 223
+    },
+    {
+      "epoch": 1.1492921492921493,
+      "grad_norm": 10.094687461853027,
+      "learning_rate": 0.00018989517410853955,
+      "loss": 0.1136,
+      "step": 224
+    },
+    {
+      "epoch": 1.1544401544401544,
+      "grad_norm": 2.8889827728271484,
+      "learning_rate": 0.00018973695579384576,
+      "loss": 0.0158,
+      "step": 225
+    },
+    {
+      "epoch": 1.1595881595881596,
+      "grad_norm": 16.844894409179688,
+      "learning_rate": 0.00018957757538284273,
+      "loss": 0.2917,
+      "step": 226
+    },
+    {
+      "epoch": 1.1647361647361647,
+      "grad_norm": 8.093831062316895,
+      "learning_rate": 0.00018941703493951164,
+      "loss": 0.0753,
+      "step": 227
+    },
+    {
+      "epoch": 1.16988416988417,
+      "grad_norm": 4.4795918464660645,
+      "learning_rate": 0.00018925533654285615,
+      "loss": 0.1687,
+      "step": 228
+    },
+    {
+      "epoch": 1.175032175032175,
+      "grad_norm": 5.258548259735107,
+      "learning_rate": 0.00018909248228687536,
+      "loss": 0.1261,
+      "step": 229
+    },
+    {
+      "epoch": 1.1801801801801801,
+      "grad_norm": 2.162417411804199,
+      "learning_rate": 0.00018892847428053693,
+      "loss": 0.0345,
+      "step": 230
+    },
+    {
+      "epoch": 1.1853281853281854,
+      "grad_norm": 1.22999107837677,
+      "learning_rate": 0.00018876331464774945,
+      "loss": 0.0141,
+      "step": 231
+    },
+    {
+      "epoch": 1.1904761904761905,
+      "grad_norm": 5.745759010314941,
+      "learning_rate": 0.00018859700552733516,
+      "loss": 0.1698,
+      "step": 232
+    },
+    {
+      "epoch": 1.1956241956241955,
+      "grad_norm": 6.1637654304504395,
+      "learning_rate": 0.00018842954907300236,
+      "loss": 0.035,
+      "step": 233
+    },
+    {
+      "epoch": 1.2007722007722008,
+      "grad_norm": 3.9793756008148193,
+      "learning_rate": 0.00018826094745331715,
+      "loss": 0.0734,
+      "step": 234
+    },
+    {
+      "epoch": 1.2059202059202059,
+      "grad_norm": 3.5555646419525146,
+      "learning_rate": 0.00018809120285167565,
+      "loss": 0.0565,
+      "step": 235
+    },
+    {
+      "epoch": 1.2110682110682112,
+      "grad_norm": 0.7398817539215088,
+      "learning_rate": 0.00018792031746627563,
+      "loss": 0.0066,
+      "step": 236
+    },
+    {
+      "epoch": 1.2162162162162162,
+      "grad_norm": 7.219451427459717,
+      "learning_rate": 0.000187748293510088,
+      "loss": 0.2503,
+      "step": 237
+    },
+    {
+      "epoch": 1.2213642213642213,
+      "grad_norm": 7.229645252227783,
+      "learning_rate": 0.0001875751332108283,
+      "loss": 0.1628,
+      "step": 238
+    },
+    {
+      "epoch": 1.2265122265122266,
+      "grad_norm": 17.265424728393555,
+      "learning_rate": 0.0001874008388109276,
+      "loss": 0.0473,
+      "step": 239
+    },
+    {
+      "epoch": 1.2316602316602316,
+      "grad_norm": 8.940932273864746,
+      "learning_rate": 0.0001872254125675037,
+      "loss": 0.4562,
+      "step": 240
+    },
+    {
+      "epoch": 1.236808236808237,
+      "grad_norm": 5.684496879577637,
+      "learning_rate": 0.0001870488567523318,
+      "loss": 0.2874,
+      "step": 241
+    },
+    {
+      "epoch": 1.241956241956242,
+      "grad_norm": 3.7431609630584717,
+      "learning_rate": 0.00018687117365181512,
+      "loss": 0.0413,
+      "step": 242
+    },
+    {
+      "epoch": 1.247104247104247,
+      "grad_norm": 0.8493585586547852,
+      "learning_rate": 0.00018669236556695515,
+      "loss": 0.0064,
+      "step": 243
+    },
+    {
+      "epoch": 1.2522522522522523,
+      "grad_norm": 6.103120803833008,
+      "learning_rate": 0.00018651243481332213,
+      "loss": 0.3538,
+      "step": 244
+    },
+    {
+      "epoch": 1.2574002574002574,
+      "grad_norm": 3.9342570304870605,
+      "learning_rate": 0.00018633138372102468,
+      "loss": 0.1866,
+      "step": 245
+    },
+    {
+      "epoch": 1.2625482625482625,
+      "grad_norm": 10.490875244140625,
+      "learning_rate": 0.00018614921463468002,
+      "loss": 0.1791,
+      "step": 246
+    },
+    {
+      "epoch": 1.2676962676962678,
+      "grad_norm": 5.010405540466309,
+      "learning_rate": 0.00018596592991338334,
+      "loss": 0.138,
+      "step": 247
+    },
+    {
+      "epoch": 1.2728442728442728,
+      "grad_norm": 6.557880401611328,
+      "learning_rate": 0.00018578153193067745,
+      "loss": 0.2603,
+      "step": 248
+    },
+    {
+      "epoch": 1.2779922779922779,
+      "grad_norm": 9.848848342895508,
+      "learning_rate": 0.0001855960230745217,
+      "loss": 0.2449,
+      "step": 249
+    },
+    {
+      "epoch": 1.2831402831402832,
+      "grad_norm": 1.8260072469711304,
+      "learning_rate": 0.0001854094057472615,
+      "loss": 0.041,
+      "step": 250
+    },
+    {
+      "epoch": 1.2882882882882882,
+      "grad_norm": 1.8941277265548706,
+      "learning_rate": 0.00018522168236559695,
+      "loss": 0.0414,
+      "step": 251
+    },
+    {
+      "epoch": 1.2934362934362935,
+      "grad_norm": 2.751856803894043,
+      "learning_rate": 0.00018503285536055147,
+      "loss": 0.0419,
+      "step": 252
+    },
+    {
+      "epoch": 1.2985842985842986,
+      "grad_norm": 6.298224925994873,
+      "learning_rate": 0.0001848429271774406,
+      "loss": 0.2005,
+      "step": 253
+    },
+    {
+      "epoch": 1.3037323037323039,
+      "grad_norm": 5.027242660522461,
+      "learning_rate": 0.00018465190027584005,
+      "loss": 0.1076,
+      "step": 254
+    },
+    {
+      "epoch": 1.308880308880309,
+      "grad_norm": 3.2702324390411377,
+      "learning_rate": 0.00018445977712955403,
+      "loss": 0.1478,
+      "step": 255
+    },
+    {
+      "epoch": 1.314028314028314,
+      "grad_norm": 0.35357093811035156,
+      "learning_rate": 0.0001842665602265831,
+      "loss": 0.0038,
+      "step": 256
+    },
+    {
+      "epoch": 1.3191763191763193,
+      "grad_norm": 5.566512107849121,
+      "learning_rate": 0.00018407225206909208,
+      "loss": 0.352,
+      "step": 257
+    },
+    {
+      "epoch": 1.3243243243243243,
+      "grad_norm": 14.227112770080566,
+      "learning_rate": 0.0001838768551733775,
+      "loss": 0.4711,
+      "step": 258
+    },
+    {
+      "epoch": 1.3294723294723294,
+      "grad_norm": 7.369393348693848,
+      "learning_rate": 0.00018368037206983515,
+      "loss": 0.0975,
+      "step": 259
+    },
+    {
+      "epoch": 1.3346203346203347,
+      "grad_norm": 1.104875922203064,
+      "learning_rate": 0.00018348280530292713,
+      "loss": 0.0147,
+      "step": 260
+    },
+    {
+      "epoch": 1.3397683397683398,
+      "grad_norm": 7.377956390380859,
+      "learning_rate": 0.00018328415743114912,
+      "loss": 0.1585,
+      "step": 261
+    },
+    {
+      "epoch": 1.3449163449163448,
+      "grad_norm": 3.0821692943573,
+      "learning_rate": 0.00018308443102699713,
+      "loss": 0.2665,
+      "step": 262
+    },
+    {
+      "epoch": 1.35006435006435,
+      "grad_norm": 5.267409324645996,
+      "learning_rate": 0.00018288362867693414,
+      "loss": 0.049,
+      "step": 263
+    },
+    {
+      "epoch": 1.3552123552123552,
+      "grad_norm": 0.19192488491535187,
+      "learning_rate": 0.0001826817529813567,
+      "loss": 0.0046,
+      "step": 264
+    },
+    {
+      "epoch": 1.3603603603603602,
+      "grad_norm": 0.8637943863868713,
+      "learning_rate": 0.00018247880655456125,
+      "loss": 0.0152,
+      "step": 265
+    },
+    {
+      "epoch": 1.3655083655083655,
+      "grad_norm": 1.4999817609786987,
+      "learning_rate": 0.00018227479202471015,
+      "loss": 0.0181,
+      "step": 266
+    },
+    {
+      "epoch": 1.3706563706563706,
+      "grad_norm": 8.225967407226562,
+      "learning_rate": 0.00018206971203379773,
+      "loss": 0.1363,
+      "step": 267
+    },
+    {
+      "epoch": 1.3758043758043759,
+      "grad_norm": 2.5319197177886963,
+      "learning_rate": 0.00018186356923761626,
+      "loss": 0.0381,
+      "step": 268
+    },
+    {
+      "epoch": 1.380952380952381,
+      "grad_norm": 4.564790725708008,
+      "learning_rate": 0.0001816563663057211,
+      "loss": 0.0186,
+      "step": 269
+    },
+    {
+      "epoch": 1.3861003861003862,
+      "grad_norm": 0.6686348915100098,
+      "learning_rate": 0.00018144810592139656,
+      "loss": 0.0074,
+      "step": 270
+    },
+    {
+      "epoch": 1.3912483912483913,
+      "grad_norm": 7.826875686645508,
+      "learning_rate": 0.00018123879078162097,
+      "loss": 0.1727,
+      "step": 271
+    },
+    {
+      "epoch": 1.3963963963963963,
+      "grad_norm": 172.96063232421875,
+      "learning_rate": 0.00018102842359703176,
+      "loss": 0.2324,
+      "step": 272
+    },
+    {
+      "epoch": 1.4015444015444016,
+      "grad_norm": 14.447083473205566,
+      "learning_rate": 0.0001808170070918904,
+      "loss": 0.3839,
+      "step": 273
+    },
+    {
+      "epoch": 1.4066924066924067,
+      "grad_norm": 4.334786415100098,
+      "learning_rate": 0.00018060454400404695,
+      "loss": 0.0619,
+      "step": 274
+    },
+    {
+      "epoch": 1.4118404118404118,
+      "grad_norm": 18.446399688720703,
+      "learning_rate": 0.000180391037084905,
+      "loss": 0.1132,
+      "step": 275
+    },
+    {
+      "epoch": 1.416988416988417,
+      "grad_norm": 1.0036085844039917,
+      "learning_rate": 0.0001801764890993856,
+      "loss": 0.0039,
+      "step": 276
+    },
+    {
+      "epoch": 1.422136422136422,
+      "grad_norm": 4.464222431182861,
+      "learning_rate": 0.0001799609028258917,
+      "loss": 0.0361,
+      "step": 277
+    },
+    {
+      "epoch": 1.4272844272844272,
+      "grad_norm": 16.352134704589844,
+      "learning_rate": 0.00017974428105627208,
+      "loss": 0.3171,
+      "step": 278
+    },
+    {
+      "epoch": 1.4324324324324325,
+      "grad_norm": 14.250722885131836,
+      "learning_rate": 0.0001795266265957853,
+      "loss": 0.2031,
+      "step": 279
+    },
+    {
+      "epoch": 1.4375804375804375,
+      "grad_norm": 10.581098556518555,
+      "learning_rate": 0.0001793079422630632,
+      "loss": 0.086,
+      "step": 280
+    },
+    {
+      "epoch": 1.4427284427284428,
+      "grad_norm": 9.860410690307617,
+      "learning_rate": 0.00017908823089007457,
+      "loss": 0.1219,
+      "step": 281
+    },
+    {
+      "epoch": 1.4478764478764479,
+      "grad_norm": 12.166980743408203,
+      "learning_rate": 0.00017886749532208837,
+      "loss": 0.0866,
+      "step": 282
+    },
+    {
+      "epoch": 1.4530244530244532,
+      "grad_norm": 2.718438148498535,
+      "learning_rate": 0.00017864573841763694,
+      "loss": 0.0611,
+      "step": 283
+    },
+    {
+      "epoch": 1.4581724581724582,
+      "grad_norm": 2.4248666763305664,
+      "learning_rate": 0.00017842296304847893,
+      "loss": 0.0378,
+      "step": 284
+    },
+    {
+      "epoch": 1.4633204633204633,
+      "grad_norm": 2.257319211959839,
+      "learning_rate": 0.00017819917209956215,
+      "loss": 0.0195,
+      "step": 285
+    },
+    {
+      "epoch": 1.4684684684684686,
+      "grad_norm": 8.68844223022461,
+      "learning_rate": 0.00017797436846898619,
+      "loss": 0.2221,
+      "step": 286
+    },
+    {
+      "epoch": 1.4736164736164736,
+      "grad_norm": 33.7538948059082,
+      "learning_rate": 0.00017774855506796496,
+      "loss": 0.1283,
+      "step": 287
+    },
+    {
+      "epoch": 1.4787644787644787,
+      "grad_norm": 20.501745223999023,
+      "learning_rate": 0.0001775217348207888,
+      "loss": 0.8959,
+      "step": 288
+    },
+    {
+      "epoch": 1.483912483912484,
+      "grad_norm": 28.74897575378418,
+      "learning_rate": 0.00017729391066478688,
+      "loss": 0.1545,
+      "step": 289
+    },
+    {
+      "epoch": 1.489060489060489,
+      "grad_norm": 7.425793170928955,
+      "learning_rate": 0.00017706508555028893,
+      "loss": 0.1349,
+      "step": 290
+    },
+    {
+      "epoch": 1.494208494208494,
+      "grad_norm": 20.139829635620117,
+      "learning_rate": 0.00017683526244058716,
+      "loss": 0.1303,
+      "step": 291
+    },
+    {
+      "epoch": 1.4993564993564994,
+      "grad_norm": 11.224247932434082,
+      "learning_rate": 0.0001766044443118978,
+      "loss": 0.097,
+      "step": 292
+    },
+    {
+      "epoch": 1.5045045045045045,
+      "grad_norm": 12.70542049407959,
+      "learning_rate": 0.0001763726341533227,
+      "loss": 0.0677,
+      "step": 293
+    },
+    {
+      "epoch": 1.5096525096525095,
+      "grad_norm": 19.43762969970703,
+      "learning_rate": 0.00017613983496681044,
+      "loss": 0.2573,
+      "step": 294
+    },
+    {
+      "epoch": 1.5148005148005148,
+      "grad_norm": 1.649288296699524,
+      "learning_rate": 0.00017590604976711754,
+      "loss": 0.014,
+      "step": 295
+    },
+    {
+      "epoch": 1.51994851994852,
+      "grad_norm": 3.48229718208313,
+      "learning_rate": 0.00017567128158176953,
+      "loss": 0.0964,
+      "step": 296
+    },
+    {
+      "epoch": 1.525096525096525,
+      "grad_norm": 4.940736770629883,
+      "learning_rate": 0.00017543553345102152,
+      "loss": 0.1336,
+      "step": 297
+    },
+    {
+      "epoch": 1.5302445302445302,
+      "grad_norm": 3.744666576385498,
+      "learning_rate": 0.00017519880842781894,
+      "loss": 0.0416,
+      "step": 298
+    },
+    {
+      "epoch": 1.5353925353925355,
+      "grad_norm": 11.016815185546875,
+      "learning_rate": 0.0001749611095777581,
+      "loss": 0.3351,
+      "step": 299
+    },
+    {
+      "epoch": 1.5405405405405406,
+      "grad_norm": 4.5957746505737305,
+      "learning_rate": 0.00017472243997904626,
+      "loss": 0.0394,
+      "step": 300
+    },
+    {
+      "epoch": 1.5456885456885456,
+      "grad_norm": 3.2034809589385986,
+      "learning_rate": 0.00017448280272246212,
+      "loss": 0.1119,
+      "step": 301
+    },
+    {
+      "epoch": 1.550836550836551,
+      "grad_norm": 12.639411926269531,
+      "learning_rate": 0.00017424220091131535,
+      "loss": 0.1371,
+      "step": 302
+    },
+    {
+      "epoch": 1.555984555984556,
+      "grad_norm": 0.4791819751262665,
+      "learning_rate": 0.00017400063766140678,
+      "loss": 0.0061,
+      "step": 303
+    },
+    {
+      "epoch": 1.561132561132561,
+      "grad_norm": 5.1950788497924805,
+      "learning_rate": 0.0001737581161009878,
+      "loss": 0.0235,
+      "step": 304
+    },
+    {
+      "epoch": 1.5662805662805663,
+      "grad_norm": 9.007669448852539,
+      "learning_rate": 0.00017351463937072004,
+      "loss": 0.1716,
+      "step": 305
+    },
+    {
+      "epoch": 1.5714285714285714,
+      "grad_norm": 1.012742280960083,
+      "learning_rate": 0.00017327021062363458,
+      "loss": 0.0108,
+      "step": 306
+    },
+    {
+      "epoch": 1.5765765765765765,
+      "grad_norm": 4.6199164390563965,
+      "learning_rate": 0.00017302483302509108,
+      "loss": 0.1989,
+      "step": 307
+    },
+    {
+      "epoch": 1.5817245817245817,
+      "grad_norm": 11.968474388122559,
+      "learning_rate": 0.00017277850975273696,
+      "loss": 0.0341,
+      "step": 308
+    },
+    {
+      "epoch": 1.586872586872587,
+      "grad_norm": 6.559664249420166,
+      "learning_rate": 0.00017253124399646606,
+      "loss": 0.2667,
+      "step": 309
+    },
+    {
+      "epoch": 1.5920205920205919,
+      "grad_norm": 0.1562490612268448,
+      "learning_rate": 0.00017228303895837748,
+      "loss": 0.003,
+      "step": 310
+    },
+    {
+      "epoch": 1.5971685971685972,
+      "grad_norm": 5.0401291847229,
+      "learning_rate": 0.000172033897852734,
+      "loss": 0.1008,
+      "step": 311
+    },
+    {
+      "epoch": 1.6023166023166024,
+      "grad_norm": 2.4609522819519043,
+      "learning_rate": 0.00017178382390592057,
+      "loss": 0.0623,
+      "step": 312
+    },
+    {
+      "epoch": 1.6074646074646075,
+      "grad_norm": 2.4802753925323486,
+      "learning_rate": 0.0001715328203564023,
+      "loss": 0.2339,
+      "step": 313
+    },
+    {
+      "epoch": 1.6126126126126126,
+      "grad_norm": 5.051590442657471,
+      "learning_rate": 0.00017128089045468294,
+      "loss": 0.0395,
+      "step": 314
+    },
+    {
+      "epoch": 1.6177606177606179,
+      "grad_norm": 13.369756698608398,
+      "learning_rate": 0.00017102803746326227,
+      "loss": 0.299,
+      "step": 315
+    },
+    {
+      "epoch": 1.622908622908623,
+      "grad_norm": 0.9210929274559021,
+      "learning_rate": 0.00017077426465659433,
+      "loss": 0.0249,
+      "step": 316
+    },
+    {
+      "epoch": 1.628056628056628,
+      "grad_norm": 4.471125602722168,
+      "learning_rate": 0.0001705195753210446,
+      "loss": 0.1232,
+      "step": 317
+    },
+    {
+      "epoch": 1.6332046332046333,
+      "grad_norm": 1.5548120737075806,
+      "learning_rate": 0.00017026397275484773,
+      "loss": 0.0195,
+      "step": 318
+    },
+    {
+      "epoch": 1.6383526383526383,
+      "grad_norm": 3.820404291152954,
+      "learning_rate": 0.0001700074602680648,
+      "loss": 0.0377,
+      "step": 319
+    },
+    {
+      "epoch": 1.6435006435006434,
+      "grad_norm": 5.681390285491943,
+      "learning_rate": 0.0001697500411825403,
+      "loss": 0.2897,
+      "step": 320
+    },
+    {
+      "epoch": 1.6486486486486487,
+      "grad_norm": 13.289894104003906,
+      "learning_rate": 0.00016949171883185918,
+      "loss": 0.0439,
+      "step": 321
+    },
+    {
+      "epoch": 1.6537966537966537,
+      "grad_norm": 5.805883884429932,
+      "learning_rate": 0.0001692324965613038,
+      "loss": 0.1332,
+      "step": 322
+    },
+    {
+      "epoch": 1.6589446589446588,
+      "grad_norm": 5.065412998199463,
+      "learning_rate": 0.00016897237772781044,
+      "loss": 0.0258,
+      "step": 323
+    },
+    {
+      "epoch": 1.664092664092664,
+      "grad_norm": 3.332307815551758,
+      "learning_rate": 0.00016871136569992587,
+      "loss": 0.0235,
+      "step": 324
+    },
+    {
+      "epoch": 1.6692406692406694,
+      "grad_norm": 5.160305976867676,
+      "learning_rate": 0.00016844946385776384,
+      "loss": 0.2321,
+      "step": 325
+    },
+    {
+      "epoch": 1.6743886743886742,
+      "grad_norm": 1.2170029878616333,
+      "learning_rate": 0.0001681866755929612,
+      "loss": 0.0138,
+      "step": 326
+    },
+    {
+      "epoch": 1.6795366795366795,
+      "grad_norm": 3.085536003112793,
+      "learning_rate": 0.00016792300430863396,
+      "loss": 0.0593,
+      "step": 327
+    },
+    {
+      "epoch": 1.6846846846846848,
+      "grad_norm": 24.154245376586914,
+      "learning_rate": 0.0001676584534193332,
+      "loss": 0.0421,
+      "step": 328
+    },
+    {
+      "epoch": 1.6898326898326899,
+      "grad_norm": 5.640690326690674,
+      "learning_rate": 0.00016739302635100108,
+      "loss": 0.0342,
+      "step": 329
+    },
+    {
+      "epoch": 1.694980694980695,
+      "grad_norm": 3.8991475105285645,
+      "learning_rate": 0.00016712672654092622,
+      "loss": 0.3529,
+      "step": 330
+    },
+    {
+      "epoch": 1.7001287001287002,
+      "grad_norm": 4.105321407318115,
+      "learning_rate": 0.0001668595574376992,
+      "loss": 0.1234,
+      "step": 331
+    },
+    {
+      "epoch": 1.7052767052767053,
+      "grad_norm": 11.61262321472168,
+      "learning_rate": 0.00016659152250116812,
+      "loss": 0.224,
+      "step": 332
+    },
+    {
+      "epoch": 1.7104247104247103,
+      "grad_norm": 33.88574981689453,
+      "learning_rate": 0.0001663226252023935,
+      "loss": 0.2226,
+      "step": 333
+    },
+    {
+      "epoch": 1.7155727155727156,
+      "grad_norm": 7.029160976409912,
+      "learning_rate": 0.00016605286902360357,
+      "loss": 0.3167,
+      "step": 334
+    },
+    {
+      "epoch": 1.7207207207207207,
+      "grad_norm": 9.228235244750977,
+      "learning_rate": 0.00016578225745814907,
+      "loss": 0.3091,
+      "step": 335
+    },
+    {
+      "epoch": 1.7258687258687258,
+      "grad_norm": 6.143121242523193,
+      "learning_rate": 0.000165510794010458,
+      "loss": 0.1115,
+      "step": 336
+    },
+    {
+      "epoch": 1.731016731016731,
+      "grad_norm": 10.932745933532715,
+      "learning_rate": 0.00016523848219599023,
+      "loss": 0.338,
+      "step": 337
+    },
+    {
+      "epoch": 1.7361647361647363,
+      "grad_norm": 25.598955154418945,
+      "learning_rate": 0.00016496532554119214,
+      "loss": 0.4376,
+      "step": 338
+    },
+    {
+      "epoch": 1.7413127413127412,
+      "grad_norm": 4.426016807556152,
+      "learning_rate": 0.0001646913275834506,
+      "loss": 0.0712,
+      "step": 339
+    },
+    {
+      "epoch": 1.7464607464607464,
+      "grad_norm": 6.740967750549316,
+      "learning_rate": 0.00016441649187104763,
+      "loss": 0.0637,
+      "step": 340
+    },
+    {
+      "epoch": 1.7516087516087517,
+      "grad_norm": 4.395668983459473,
+      "learning_rate": 0.000164140821963114,
+      "loss": 0.1922,
+      "step": 341
+    },
+    {
+      "epoch": 1.7567567567567568,
+      "grad_norm": 0.5163567662239075,
+      "learning_rate": 0.00016386432142958342,
+      "loss": 0.0135,
+      "step": 342
+    },
+    {
+      "epoch": 1.7619047619047619,
+      "grad_norm": 1.5225406885147095,
+      "learning_rate": 0.00016358699385114625,
+      "loss": 0.0354,
+      "step": 343
+    },
+    {
+      "epoch": 1.7670527670527671,
+      "grad_norm": 0.5186464190483093,
+      "learning_rate": 0.000163308842819203,
+      "loss": 0.0086,
+      "step": 344
+    },
+    {
+      "epoch": 1.7722007722007722,
+      "grad_norm": 6.920417785644531,
+      "learning_rate": 0.00016302987193581806,
+      "loss": 0.3046,
+      "step": 345
+    },
+    {
+      "epoch": 1.7773487773487773,
+      "grad_norm": 5.082959175109863,
+      "learning_rate": 0.00016275008481367287,
+      "loss": 0.3969,
+      "step": 346
+    },
+    {
+      "epoch": 1.7824967824967826,
+      "grad_norm": 3.6133973598480225,
+      "learning_rate": 0.00016246948507601914,
+      "loss": 0.0636,
+      "step": 347
+    },
+    {
+      "epoch": 1.7876447876447876,
+      "grad_norm": 0.09072360396385193,
+      "learning_rate": 0.00016218807635663202,
+      "loss": 0.0019,
+      "step": 348
+    },
+    {
+      "epoch": 1.7927927927927927,
+      "grad_norm": 20.693056106567383,
+      "learning_rate": 0.00016190586229976304,
+      "loss": 0.3171,
+      "step": 349
+    },
+    {
+      "epoch": 1.797940797940798,
+      "grad_norm": 5.184165000915527,
+      "learning_rate": 0.00016162284656009274,
+      "loss": 0.1358,
+      "step": 350
+    },
+    {
+      "epoch": 1.803088803088803,
+      "grad_norm": 5.272857189178467,
+      "learning_rate": 0.00016133903280268362,
+      "loss": 0.1494,
+      "step": 351
+    },
+    {
+      "epoch": 1.808236808236808,
+      "grad_norm": 1.6124505996704102,
+      "learning_rate": 0.0001610544247029325,
+      "loss": 0.0436,
+      "step": 352
+    },
+    {
+      "epoch": 1.8133848133848134,
+      "grad_norm": 4.880893230438232,
+      "learning_rate": 0.0001607690259465229,
+      "loss": 0.079,
+      "step": 353
+    },
+    {
+      "epoch": 1.8185328185328187,
+      "grad_norm": 1.2332196235656738,
+      "learning_rate": 0.0001604828402293774,
+      "loss": 0.0256,
+      "step": 354
+    },
+    {
+      "epoch": 1.8236808236808235,
+      "grad_norm": 9.343794822692871,
+      "learning_rate": 0.00016019587125760978,
+      "loss": 0.1764,
+      "step": 355
+    },
+    {
+      "epoch": 1.8288288288288288,
+      "grad_norm": 3.8978259563446045,
+      "learning_rate": 0.00015990812274747692,
+      "loss": 0.2402,
+      "step": 356
+    },
+    {
+      "epoch": 1.833976833976834,
+      "grad_norm": 4.8559770584106445,
+      "learning_rate": 0.00015961959842533083,
+      "loss": 0.0427,
+      "step": 357
+    },
+    {
+      "epoch": 1.8391248391248392,
+      "grad_norm": 17.083404541015625,
+      "learning_rate": 0.00015933030202757022,
+      "loss": 0.1369,
+      "step": 358
+    },
+    {
+      "epoch": 1.8442728442728442,
+      "grad_norm": 3.8209121227264404,
+      "learning_rate": 0.00015904023730059228,
+      "loss": 0.0522,
+      "step": 359
+    },
+    {
+      "epoch": 1.8494208494208495,
+      "grad_norm": 34.773990631103516,
+      "learning_rate": 0.00015874940800074402,
+      "loss": 0.0258,
+      "step": 360
+    },
+    {
+      "epoch": 1.8545688545688546,
+      "grad_norm": 1.1415127515792847,
+      "learning_rate": 0.00015845781789427377,
+      "loss": 0.0051,
+      "step": 361
+    },
+    {
+      "epoch": 1.8597168597168596,
+      "grad_norm": 0.8682407736778259,
+      "learning_rate": 0.00015816547075728226,
+      "loss": 0.0155,
+      "step": 362
+    },
+    {
+      "epoch": 1.864864864864865,
+      "grad_norm": 6.977367877960205,
+      "learning_rate": 0.00015787237037567385,
+      "loss": 0.0353,
+      "step": 363
+    },
+    {
+      "epoch": 1.87001287001287,
+      "grad_norm": 24.50101089477539,
+      "learning_rate": 0.0001575785205451073,
+      "loss": 0.2093,
+      "step": 364
+    },
+    {
+      "epoch": 1.875160875160875,
+      "grad_norm": 3.2389304637908936,
+      "learning_rate": 0.000157283925070947,
+      "loss": 0.0761,
+      "step": 365
+    },
+    {
+      "epoch": 1.8803088803088803,
+      "grad_norm": 0.12845732271671295,
+      "learning_rate": 0.0001569885877682132,
+      "loss": 0.0014,
+      "step": 366
+    },
+    {
+      "epoch": 1.8854568854568856,
+      "grad_norm": 15.371975898742676,
+      "learning_rate": 0.000156692512461533,
+      "loss": 0.0787,
+      "step": 367
+    },
+    {
+      "epoch": 1.8906048906048905,
+      "grad_norm": 10.21326732635498,
+      "learning_rate": 0.00015639570298509064,
+      "loss": 0.2458,
+      "step": 368
+    },
+    {
+      "epoch": 1.8957528957528957,
+      "grad_norm": 0.1647750437259674,
+      "learning_rate": 0.00015609816318257788,
+      "loss": 0.0021,
+      "step": 369
+    },
+    {
+      "epoch": 1.900900900900901,
+      "grad_norm": 6.113330364227295,
+      "learning_rate": 0.00015579989690714423,
+      "loss": 0.0783,
+      "step": 370
+    },
+    {
+      "epoch": 1.906048906048906,
+      "grad_norm": 20.247173309326172,
+      "learning_rate": 0.000155500908021347,
+      "loss": 0.1167,
+      "step": 371
+    },
+    {
+      "epoch": 1.9111969111969112,
+      "grad_norm": 0.4571623206138611,
+      "learning_rate": 0.00015520120039710139,
+      "loss": 0.0036,
+      "step": 372
+    },
+    {
+      "epoch": 1.9163449163449164,
+      "grad_norm": 8.424338340759277,
+      "learning_rate": 0.0001549007779156302,
+      "loss": 0.0414,
+      "step": 373
+    },
+    {
+      "epoch": 1.9214929214929215,
+      "grad_norm": 20.99398422241211,
+      "learning_rate": 0.00015459964446741382,
+      "loss": 0.6867,
+      "step": 374
+    },
+    {
+      "epoch": 1.9266409266409266,
+      "grad_norm": 26.535310745239258,
+      "learning_rate": 0.0001542978039521395,
+      "loss": 0.9056,
+      "step": 375
+    },
+    {
+      "epoch": 1.9317889317889319,
+      "grad_norm": 0.012314067222177982,
+      "learning_rate": 0.00015399526027865107,
+      "loss": 0.0002,
+      "step": 376
+    },
+    {
+      "epoch": 1.936936936936937,
+      "grad_norm": 6.414860248565674,
+      "learning_rate": 0.0001536920173648984,
+      "loss": 0.4392,
+      "step": 377
+    },
+    {
+      "epoch": 1.942084942084942,
+      "grad_norm": 21.50543975830078,
+      "learning_rate": 0.00015338807913788636,
+      "loss": 0.2503,
+      "step": 378
+    },
+    {
+      "epoch": 1.9472329472329473,
+      "grad_norm": 12.399309158325195,
+      "learning_rate": 0.0001530834495336243,
+      "loss": 0.2085,
+      "step": 379
+    },
+    {
+      "epoch": 1.9523809523809523,
+      "grad_norm": 13.036598205566406,
+      "learning_rate": 0.00015277813249707487,
+      "loss": 0.1387,
+      "step": 380
+    },
+    {
+      "epoch": 1.9575289575289574,
+      "grad_norm": 4.081547737121582,
+      "learning_rate": 0.000152472131982103,
+      "loss": 0.0301,
+      "step": 381
+    },
+    {
+      "epoch": 1.9626769626769627,
+      "grad_norm": 0.05565841868519783,
+      "learning_rate": 0.0001521654519514246,
+      "loss": 0.0009,
+      "step": 382
+    },
+    {
+      "epoch": 1.967824967824968,
+      "grad_norm": 0.7645258903503418,
+      "learning_rate": 0.0001518580963765555,
+      "loss": 0.0122,
+      "step": 383
+    },
+    {
+      "epoch": 1.972972972972973,
+      "grad_norm": 0.9463894963264465,
+      "learning_rate": 0.00015155006923775965,
+      "loss": 0.0071,
+      "step": 384
+    },
+    {
+      "epoch": 1.978120978120978,
+      "grad_norm": 10.104920387268066,
+      "learning_rate": 0.00015124137452399795,
+      "loss": 0.53,
+      "step": 385
+    },
+    {
+      "epoch": 1.9832689832689834,
+      "grad_norm": 9.295433044433594,
+      "learning_rate": 0.00015093201623287631,
+      "loss": 0.0379,
+      "step": 386
+    },
+    {
+      "epoch": 1.9884169884169884,
+      "grad_norm": 29.789033889770508,
+      "learning_rate": 0.00015062199837059405,
+      "loss": 0.2779,
+      "step": 387
+    },
+    {
+      "epoch": 1.9935649935649935,
+      "grad_norm": 6.143742084503174,
+      "learning_rate": 0.00015031132495189187,
+      "loss": 0.2134,
+      "step": 388
+    },
+    {
+      "epoch": 1.9987129987129988,
+      "grad_norm": 0.6877163052558899,
+      "learning_rate": 0.00015000000000000001,
+      "loss": 0.0068,
+      "step": 389
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 2.2970244884490967,
+      "learning_rate": 0.00014968802754658614,
+      "loss": 0.012,
+      "step": 390
+    },
+    {
+      "epoch": 2.0051480051480053,
+      "grad_norm": 2.1872456073760986,
+      "learning_rate": 0.0001493754116317029,
+      "loss": 0.1649,
+      "step": 391
+    },
+    {
+      "epoch": 2.01029601029601,
+      "grad_norm": 4.704517364501953,
+      "learning_rate": 0.00014906215630373606,
+      "loss": 0.106,
+      "step": 392
+    },
+    {
+      "epoch": 2.0154440154440154,
+      "grad_norm": 1.443876028060913,
+      "learning_rate": 0.00014874826561935155,
+      "loss": 0.0131,
+      "step": 393
+    },
+    {
+      "epoch": 2.0205920205920207,
+      "grad_norm": 1.24201238155365,
+      "learning_rate": 0.00014843374364344333,
+      "loss": 0.0291,
+      "step": 394
+    },
+    {
+      "epoch": 2.0257400257400255,
+      "grad_norm": 4.6277265548706055,
+      "learning_rate": 0.00014811859444908052,
+      "loss": 0.0769,
+      "step": 395
+    },
+    {
+      "epoch": 2.030888030888031,
+      "grad_norm": 22.727767944335938,
+      "learning_rate": 0.0001478028221174548,
+      "loss": 0.0536,
+      "step": 396
+    },
+    {
+      "epoch": 2.036036036036036,
+      "grad_norm": 1.8225324153900146,
+      "learning_rate": 0.00014748643073782752,
+      "loss": 0.0275,
+      "step": 397
+    },
+    {
+      "epoch": 2.0411840411840414,
+      "grad_norm": 2.86083984375,
+      "learning_rate": 0.00014716942440747664,
+      "loss": 0.0402,
+      "step": 398
+    },
+    {
+      "epoch": 2.0463320463320462,
+      "grad_norm": 0.07705119997262955,
+      "learning_rate": 0.00014685180723164376,
+      "loss": 0.0014,
+      "step": 399
+    },
+    {
+      "epoch": 2.0514800514800515,
+      "grad_norm": 0.39398977160453796,
+      "learning_rate": 0.000146533583323481,
+      "loss": 0.0069,
+      "step": 400
+    },
+    {
+      "epoch": 2.056628056628057,
+      "grad_norm": 7.6373162269592285,
+      "learning_rate": 0.0001462147568039977,
+      "loss": 0.0387,
+      "step": 401
+    },
+    {
+      "epoch": 2.0617760617760617,
+      "grad_norm": 2.058821439743042,
+      "learning_rate": 0.00014589533180200693,
+      "loss": 0.0197,
+      "step": 402
+    },
+    {
+      "epoch": 2.066924066924067,
+      "grad_norm": 2.1784212589263916,
+      "learning_rate": 0.00014557531245407225,
+      "loss": 0.0057,
+      "step": 403
+    },
+    {
+      "epoch": 2.0720720720720722,
+      "grad_norm": 2.5718722343444824,
+      "learning_rate": 0.00014525470290445392,
+      "loss": 0.027,
+      "step": 404
+    },
+    {
+      "epoch": 2.077220077220077,
+      "grad_norm": 3.490748882293701,
+      "learning_rate": 0.00014493350730505533,
+      "loss": 0.0438,
+      "step": 405
+    },
+    {
+      "epoch": 2.0823680823680824,
+      "grad_norm": 4.078834056854248,
+      "learning_rate": 0.0001446117298153693,
+      "loss": 0.0282,
+      "step": 406
+    },
+    {
+      "epoch": 2.0875160875160876,
+      "grad_norm": 1.2998112440109253,
+      "learning_rate": 0.00014428937460242417,
+      "loss": 0.0077,
+      "step": 407
+    },
+    {
+      "epoch": 2.0926640926640925,
+      "grad_norm": 3.3758530616760254,
+      "learning_rate": 0.00014396644584072972,
+      "loss": 0.0196,
+      "step": 408
+    },
+    {
+      "epoch": 2.0978120978120978,
+      "grad_norm": 2.455396890640259,
+      "learning_rate": 0.00014364294771222324,
+      "loss": 0.015,
+      "step": 409
+    },
+    {
+      "epoch": 2.102960102960103,
+      "grad_norm": 0.5789293050765991,
+      "learning_rate": 0.00014331888440621533,
+      "loss": 0.004,
+      "step": 410
+    },
+    {
+      "epoch": 2.108108108108108,
+      "grad_norm": 19.402442932128906,
+      "learning_rate": 0.00014299426011933568,
+      "loss": 0.2045,
+      "step": 411
+    },
+    {
+      "epoch": 2.113256113256113,
+      "grad_norm": 7.968266010284424,
+      "learning_rate": 0.0001426690790554787,
+      "loss": 0.142,
+      "step": 412
+    },
+    {
+      "epoch": 2.1184041184041185,
+      "grad_norm": 0.7839431762695312,
+      "learning_rate": 0.00014234334542574906,
+      "loss": 0.0019,
+      "step": 413
+    },
+    {
+      "epoch": 2.1235521235521237,
+      "grad_norm": 4.259623050689697,
+      "learning_rate": 0.00014201706344840712,
+      "loss": 0.0511,
+      "step": 414
+    },
+    {
+      "epoch": 2.1287001287001286,
+      "grad_norm": 5.2883687019348145,
+      "learning_rate": 0.00014169023734881452,
+      "loss": 0.0085,
+      "step": 415
+    },
+    {
+      "epoch": 2.133848133848134,
+      "grad_norm": 5.6633453369140625,
+      "learning_rate": 0.00014136287135937915,
+      "loss": 0.0274,
+      "step": 416
+    },
+    {
+      "epoch": 2.138996138996139,
+      "grad_norm": 2.8423304557800293,
+      "learning_rate": 0.00014103496971950053,
+      "loss": 0.0123,
+      "step": 417
+    },
+    {
+      "epoch": 2.144144144144144,
+      "grad_norm": 1.3557080030441284,
+      "learning_rate": 0.0001407065366755149,
+      "loss": 0.0051,
+      "step": 418
+    },
+    {
+      "epoch": 2.1492921492921493,
+      "grad_norm": 6.159451007843018,
+      "learning_rate": 0.00014037757648064018,
+      "loss": 0.2243,
+      "step": 419
+    },
+    {
+      "epoch": 2.1544401544401546,
+      "grad_norm": 11.175186157226562,
+      "learning_rate": 0.00014004809339492088,
+      "loss": 0.3641,
+      "step": 420
+    },
+    {
+      "epoch": 2.1595881595881594,
+      "grad_norm": 4.087236404418945,
+      "learning_rate": 0.00013971809168517298,
+      "loss": 0.0304,
+      "step": 421
+    },
+    {
+      "epoch": 2.1647361647361647,
+      "grad_norm": 2.0897576808929443,
+      "learning_rate": 0.00013938757562492873,
+      "loss": 0.0208,
+      "step": 422
+    },
+    {
+      "epoch": 2.16988416988417,
+      "grad_norm": 2.677189350128174,
+      "learning_rate": 0.0001390565494943811,
+      "loss": 0.0114,
+      "step": 423
+    },
+    {
+      "epoch": 2.175032175032175,
+      "grad_norm": 1.6713389158248901,
+      "learning_rate": 0.00013872501758032863,
+      "loss": 0.0082,
+      "step": 424
+    },
+    {
+      "epoch": 2.18018018018018,
+      "grad_norm": 15.607181549072266,
+      "learning_rate": 0.00013839298417611963,
+      "loss": 0.162,
+      "step": 425
+    },
+    {
+      "epoch": 2.1853281853281854,
+      "grad_norm": 28.1248722076416,
+      "learning_rate": 0.00013806045358159683,
+      "loss": 0.3767,
+      "step": 426
+    },
+    {
+      "epoch": 2.1904761904761907,
+      "grad_norm": 10.008810043334961,
+      "learning_rate": 0.00013772743010304154,
+      "loss": 0.0227,
+      "step": 427
+    },
+    {
+      "epoch": 2.1956241956241955,
+      "grad_norm": 24.181835174560547,
+      "learning_rate": 0.00013739391805311793,
+      "loss": 0.1223,
+      "step": 428
+    },
+    {
+      "epoch": 2.200772200772201,
+      "grad_norm": 18.133371353149414,
+      "learning_rate": 0.00013705992175081728,
+      "loss": 0.2582,
+      "step": 429
+    },
+    {
+      "epoch": 2.205920205920206,
+      "grad_norm": 5.463319778442383,
+      "learning_rate": 0.00013672544552140176,
+      "loss": 0.0364,
+      "step": 430
+    },
+    {
+      "epoch": 2.211068211068211,
+      "grad_norm": 0.08443836867809296,
+      "learning_rate": 0.00013639049369634876,
+      "loss": 0.0021,
+      "step": 431
+    },
+    {
+      "epoch": 2.2162162162162162,
+      "grad_norm": 1.759372591972351,
+      "learning_rate": 0.00013605507061329464,
+      "loss": 0.004,
+      "step": 432
+    },
+    {
+      "epoch": 2.2213642213642215,
+      "grad_norm": 2.845254898071289,
+      "learning_rate": 0.0001357191806159785,
+      "loss": 0.0059,
+      "step": 433
+    },
+    {
+      "epoch": 2.2265122265122264,
+      "grad_norm": 0.06969447433948517,
+      "learning_rate": 0.0001353828280541861,
+      "loss": 0.001,
+      "step": 434
+    },
+    {
+      "epoch": 2.2316602316602316,
+      "grad_norm": 15.04641342163086,
+      "learning_rate": 0.00013504601728369327,
+      "loss": 0.0742,
+      "step": 435
+    },
+    {
+      "epoch": 2.236808236808237,
+      "grad_norm": 0.06809443980455399,
+      "learning_rate": 0.00013470875266620978,
+      "loss": 0.0012,
+      "step": 436
+    },
+    {
+      "epoch": 2.2419562419562418,
+      "grad_norm": 2.0466463565826416,
+      "learning_rate": 0.00013437103856932264,
+      "loss": 0.0338,
+      "step": 437
+    },
+    {
+      "epoch": 2.247104247104247,
+      "grad_norm": 6.784067630767822,
+      "learning_rate": 0.00013403287936643977,
+      "loss": 0.1072,
+      "step": 438
+    },
+    {
+      "epoch": 2.2522522522522523,
+      "grad_norm": 0.7220061421394348,
+      "learning_rate": 0.00013369427943673312,
+      "loss": 0.0047,
+      "step": 439
+    },
+    {
+      "epoch": 2.257400257400257,
+      "grad_norm": 15.778314590454102,
+      "learning_rate": 0.00013335524316508208,
+      "loss": 0.0888,
+      "step": 440
+    },
+    {
+      "epoch": 2.2625482625482625,
+      "grad_norm": 7.693014621734619,
+      "learning_rate": 0.00013301577494201664,
+      "loss": 0.1001,
+      "step": 441
+    },
+    {
+      "epoch": 2.2676962676962678,
+      "grad_norm": 5.649167537689209,
+      "learning_rate": 0.0001326758791636607,
+      "loss": 0.2558,
+      "step": 442
+    },
+    {
+      "epoch": 2.2728442728442726,
+      "grad_norm": 1.476900339126587,
+      "learning_rate": 0.00013233556023167485,
+      "loss": 0.0279,
+      "step": 443
+    },
+    {
+      "epoch": 2.277992277992278,
+      "grad_norm": 1.083785891532898,
+      "learning_rate": 0.0001319948225531997,
+      "loss": 0.0114,
+      "step": 444
+    },
+    {
+      "epoch": 2.283140283140283,
+      "grad_norm": 2.313983917236328,
+      "learning_rate": 0.0001316536705407985,
+      "loss": 0.0154,
+      "step": 445
+    },
+    {
+      "epoch": 2.2882882882882885,
+      "grad_norm": 10.615255355834961,
+      "learning_rate": 0.00013131210861240026,
+      "loss": 0.2956,
+      "step": 446
+    },
+    {
+      "epoch": 2.2934362934362933,
+      "grad_norm": 0.5474928021430969,
+      "learning_rate": 0.0001309701411912423,
+      "loss": 0.0054,
+      "step": 447
+    },
+    {
+      "epoch": 2.2985842985842986,
+      "grad_norm": 0.26078635454177856,
+      "learning_rate": 0.00013062777270581312,
+      "loss": 0.0018,
+      "step": 448
+    },
+    {
+      "epoch": 2.303732303732304,
+      "grad_norm": 1.3901201486587524,
+      "learning_rate": 0.00013028500758979506,
+      "loss": 0.0096,
+      "step": 449
+    },
+    {
+      "epoch": 2.3088803088803087,
+      "grad_norm": 3.0954666137695312,
+      "learning_rate": 0.00012994185028200684,
+      "loss": 0.0252,
+      "step": 450
+    },
+    {
+      "epoch": 2.314028314028314,
+      "grad_norm": 3.0988850593566895,
+      "learning_rate": 0.00012959830522634596,
+      "loss": 0.0192,
+      "step": 451
+    },
+    {
+      "epoch": 2.3191763191763193,
+      "grad_norm": 3.527113437652588,
+      "learning_rate": 0.00012925437687173142,
+      "loss": 0.159,
+      "step": 452
+    },
+    {
+      "epoch": 2.3243243243243246,
+      "grad_norm": 0.9381816983222961,
+      "learning_rate": 0.00012891006967204584,
+      "loss": 0.0053,
+      "step": 453
+    },
+    {
+      "epoch": 2.3294723294723294,
+      "grad_norm": 10.026618957519531,
+      "learning_rate": 0.00012856538808607795,
+      "loss": 0.0484,
+      "step": 454
+    },
+    {
+      "epoch": 2.3346203346203347,
+      "grad_norm": 0.5608552098274231,
+      "learning_rate": 0.00012822033657746478,
+      "loss": 0.0022,
+      "step": 455
+    },
+    {
+      "epoch": 2.33976833976834,
+      "grad_norm": 0.6499980092048645,
+      "learning_rate": 0.0001278749196146339,
+      "loss": 0.006,
+      "step": 456
+    },
+    {
+      "epoch": 2.344916344916345,
+      "grad_norm": 2.563431978225708,
+      "learning_rate": 0.0001275291416707454,
+      "loss": 0.0174,
+      "step": 457
+    },
+    {
+      "epoch": 2.35006435006435,
+      "grad_norm": 46.18597412109375,
+      "learning_rate": 0.0001271830072236343,
+      "loss": 0.4403,
+      "step": 458
+    },
+    {
+      "epoch": 2.3552123552123554,
+      "grad_norm": 7.0390214920043945,
+      "learning_rate": 0.00012683652075575218,
+      "loss": 0.1492,
+      "step": 459
+    },
+    {
+      "epoch": 2.3603603603603602,
+      "grad_norm": 6.817877292633057,
+      "learning_rate": 0.00012648968675410928,
+      "loss": 0.0186,
+      "step": 460
+    },
+    {
+      "epoch": 2.3655083655083655,
+      "grad_norm": 7.536839962005615,
+      "learning_rate": 0.00012614250971021657,
+      "loss": 0.0655,
+      "step": 461
+    },
+    {
+      "epoch": 2.370656370656371,
+      "grad_norm": 13.58019733428955,
+      "learning_rate": 0.0001257949941200273,
+      "loss": 0.3622,
+      "step": 462
+    },
+    {
+      "epoch": 2.3758043758043756,
+      "grad_norm": 0.5299798250198364,
+      "learning_rate": 0.00012544714448387893,
+      "loss": 0.0016,
+      "step": 463
+    },
+    {
+      "epoch": 2.380952380952381,
+      "grad_norm": 18.681720733642578,
+      "learning_rate": 0.00012509896530643488,
+      "loss": 0.1813,
+      "step": 464
+    },
+    {
+      "epoch": 2.386100386100386,
+      "grad_norm": 0.030531983822584152,
+      "learning_rate": 0.00012475046109662605,
+      "loss": 0.0007,
+      "step": 465
+    },
+    {
+      "epoch": 2.391248391248391,
+      "grad_norm": 0.040560245513916016,
+      "learning_rate": 0.0001244016363675926,
+      "loss": 0.0006,
+      "step": 466
+    },
+    {
+      "epoch": 2.3963963963963963,
+      "grad_norm": 1.3282872438430786,
+      "learning_rate": 0.00012405249563662537,
+      "loss": 0.0165,
+      "step": 467
+    },
+    {
+      "epoch": 2.4015444015444016,
+      "grad_norm": 3.2533833980560303,
+      "learning_rate": 0.00012370304342510747,
+      "loss": 0.1231,
+      "step": 468
+    },
+    {
+      "epoch": 2.4066924066924065,
+      "grad_norm": 5.422544479370117,
+      "learning_rate": 0.00012335328425845565,
+      "loss": 0.1672,
+      "step": 469
+    },
+    {
+      "epoch": 2.4118404118404118,
+      "grad_norm": 8.349811553955078,
+      "learning_rate": 0.00012300322266606178,
+      "loss": 0.0752,
+      "step": 470
+    },
+    {
+      "epoch": 2.416988416988417,
+      "grad_norm": 4.550008296966553,
+      "learning_rate": 0.00012265286318123415,
+      "loss": 0.022,
+      "step": 471
+    },
+    {
+      "epoch": 2.4221364221364223,
+      "grad_norm": 2.0059444904327393,
+      "learning_rate": 0.00012230221034113874,
+      "loss": 0.0191,
+      "step": 472
+    },
+    {
+      "epoch": 2.427284427284427,
+      "grad_norm": 9.578508377075195,
+      "learning_rate": 0.00012195126868674051,
+      "loss": 0.1949,
+      "step": 473
+    },
+    {
+      "epoch": 2.4324324324324325,
+      "grad_norm": 4.417119026184082,
+      "learning_rate": 0.00012160004276274453,
+      "loss": 0.0063,
+      "step": 474
+    },
+    {
+      "epoch": 2.4375804375804377,
+      "grad_norm": 0.7690457105636597,
+      "learning_rate": 0.00012124853711753727,
+      "loss": 0.0041,
+      "step": 475
+    },
+    {
+      "epoch": 2.4427284427284426,
+      "grad_norm": 11.609339714050293,
+      "learning_rate": 0.00012089675630312754,
+      "loss": 0.2607,
+      "step": 476
+    },
+    {
+      "epoch": 2.447876447876448,
+      "grad_norm": 3.6218950748443604,
+      "learning_rate": 0.00012054470487508751,
+      "loss": 0.0439,
+      "step": 477
+    },
+    {
+      "epoch": 2.453024453024453,
+      "grad_norm": 19.886072158813477,
+      "learning_rate": 0.00012019238739249397,
+      "loss": 0.2599,
+      "step": 478
+    },
+    {
+      "epoch": 2.458172458172458,
+      "grad_norm": 5.3012309074401855,
+      "learning_rate": 0.000119839808417869,
+      "loss": 0.1368,
+      "step": 479
+    },
+    {
+      "epoch": 2.4633204633204633,
+      "grad_norm": 2.9251902103424072,
+      "learning_rate": 0.00011948697251712109,
+      "loss": 0.1343,
+      "step": 480
+    },
+    {
+      "epoch": 2.4684684684684686,
+      "grad_norm": 0.6428108811378479,
+      "learning_rate": 0.00011913388425948584,
+      "loss": 0.0056,
+      "step": 481
+    },
+    {
+      "epoch": 2.473616473616474,
+      "grad_norm": 5.721848964691162,
+      "learning_rate": 0.00011878054821746703,
+      "loss": 0.0758,
+      "step": 482
+    },
+    {
+      "epoch": 2.4787644787644787,
+      "grad_norm": 1.6780121326446533,
+      "learning_rate": 0.00011842696896677708,
+      "loss": 0.022,
+      "step": 483
+    },
+    {
+      "epoch": 2.483912483912484,
+      "grad_norm": 4.217240810394287,
+      "learning_rate": 0.00011807315108627806,
+      "loss": 0.0262,
+      "step": 484
+    },
+    {
+      "epoch": 2.4890604890604893,
+      "grad_norm": 3.619426727294922,
+      "learning_rate": 0.0001177190991579223,
+      "loss": 0.0054,
+      "step": 485
+    },
+    {
+      "epoch": 2.494208494208494,
+      "grad_norm": 0.1457854062318802,
+      "learning_rate": 0.00011736481776669306,
+      "loss": 0.0021,
+      "step": 486
+    },
+    {
+      "epoch": 2.4993564993564994,
+      "grad_norm": 3.618464469909668,
+      "learning_rate": 0.0001170103115005451,
+      "loss": 0.0206,
+      "step": 487
+    },
+    {
+      "epoch": 2.5045045045045047,
+      "grad_norm": 1.354058027267456,
+      "learning_rate": 0.00011665558495034546,
+      "loss": 0.0137,
+      "step": 488
+    },
+    {
+      "epoch": 2.5096525096525095,
+      "grad_norm": 5.335404872894287,
+      "learning_rate": 0.00011630064270981367,
+      "loss": 0.0286,
+      "step": 489
+    },
+    {
+      "epoch": 2.514800514800515,
+      "grad_norm": 0.21103155612945557,
+      "learning_rate": 0.00011594548937546257,
+      "loss": 0.0034,
+      "step": 490
+    },
+    {
+      "epoch": 2.51994851994852,
+      "grad_norm": 0.12383392453193665,
+      "learning_rate": 0.00011559012954653865,
+      "loss": 0.0019,
+      "step": 491
+    },
+    {
+      "epoch": 2.525096525096525,
+      "grad_norm": 0.8978340029716492,
+      "learning_rate": 0.00011523456782496253,
+      "loss": 0.0127,
+      "step": 492
+    },
+    {
+      "epoch": 2.53024453024453,
+      "grad_norm": 0.08338338136672974,
+      "learning_rate": 0.00011487880881526934,
+      "loss": 0.001,
+      "step": 493
+    },
+    {
+      "epoch": 2.5353925353925355,
+      "grad_norm": 0.04397974908351898,
+      "learning_rate": 0.00011452285712454904,
+      "loss": 0.001,
+      "step": 494
+    },
+    {
+      "epoch": 2.5405405405405403,
+      "grad_norm": 0.02892712503671646,
+      "learning_rate": 0.00011416671736238689,
+      "loss": 0.0004,
+      "step": 495
+    },
+    {
+      "epoch": 2.5456885456885456,
+      "grad_norm": 0.1170390322804451,
+      "learning_rate": 0.00011381039414080365,
+      "loss": 0.0006,
+      "step": 496
+    },
+    {
+      "epoch": 2.550836550836551,
+      "grad_norm": 0.01198833342641592,
+      "learning_rate": 0.00011345389207419588,
+      "loss": 0.0001,
+      "step": 497
+    },
+    {
+      "epoch": 2.5559845559845558,
+      "grad_norm": 0.05665574595332146,
+      "learning_rate": 0.00011309721577927619,
+      "loss": 0.0005,
+      "step": 498
+    },
+    {
+      "epoch": 2.561132561132561,
+      "grad_norm": 2.8404929637908936,
+      "learning_rate": 0.00011274036987501348,
+      "loss": 0.1128,
+      "step": 499
+    },
+    {
+      "epoch": 2.5662805662805663,
+      "grad_norm": 42.856910705566406,
+      "learning_rate": 0.00011238335898257304,
+      "loss": 0.4863,
+      "step": 500
+    },
+    {
+      "epoch": 2.571428571428571,
+      "grad_norm": 0.12049880623817444,
+      "learning_rate": 0.0001120261877252568,
+      "loss": 0.0008,
+      "step": 501
+    },
+    {
+      "epoch": 2.5765765765765765,
+      "grad_norm": 17.771467208862305,
+      "learning_rate": 0.00011166886072844342,
+      "loss": 0.1862,
+      "step": 502
+    },
+    {
+      "epoch": 2.5817245817245817,
+      "grad_norm": 10.894670486450195,
+      "learning_rate": 0.00011131138261952845,
+      "loss": 0.2314,
+      "step": 503
+    },
+    {
+      "epoch": 2.586872586872587,
+      "grad_norm": 6.98428201675415,
+      "learning_rate": 0.00011095375802786419,
+      "loss": 0.3374,
+      "step": 504
+    },
+    {
+      "epoch": 2.592020592020592,
+      "grad_norm": 0.03092469647526741,
+      "learning_rate": 0.00011059599158470002,
+      "loss": 0.0003,
+      "step": 505
+    },
+    {
+      "epoch": 2.597168597168597,
+      "grad_norm": 0.9625420570373535,
+      "learning_rate": 0.00011023808792312227,
+      "loss": 0.0048,
+      "step": 506
+    },
+    {
+      "epoch": 2.6023166023166024,
+      "grad_norm": 6.929542064666748,
+      "learning_rate": 0.00010988005167799427,
+      "loss": 0.2218,
+      "step": 507
+    },
+    {
+      "epoch": 2.6074646074646077,
+      "grad_norm": 5.105766296386719,
+      "learning_rate": 0.00010952188748589625,
+      "loss": 0.0989,
+      "step": 508
+    },
+    {
+      "epoch": 2.6126126126126126,
+      "grad_norm": 0.5411030650138855,
+      "learning_rate": 0.0001091635999850655,
+      "loss": 0.0046,
+      "step": 509
+    },
+    {
+      "epoch": 2.617760617760618,
+      "grad_norm": 0.1863740086555481,
+      "learning_rate": 0.00010880519381533592,
+      "loss": 0.0024,
+      "step": 510
+    },
+    {
+      "epoch": 2.622908622908623,
+      "grad_norm": 3.417442560195923,
+      "learning_rate": 0.00010844667361807842,
+      "loss": 0.3164,
+      "step": 511
+    },
+    {
+      "epoch": 2.628056628056628,
+      "grad_norm": 1.0131347179412842,
+      "learning_rate": 0.00010808804403614043,
+      "loss": 0.005,
+      "step": 512
+    },
+    {
+      "epoch": 2.6332046332046333,
+      "grad_norm": 4.6805033683776855,
+      "learning_rate": 0.00010772930971378596,
+      "loss": 0.032,
+      "step": 513
+    },
+    {
+      "epoch": 2.6383526383526386,
+      "grad_norm": 0.5987910628318787,
+      "learning_rate": 0.00010737047529663545,
+      "loss": 0.0018,
+      "step": 514
+    },
+    {
+      "epoch": 2.6435006435006434,
+      "grad_norm": 3.238391160964966,
+      "learning_rate": 0.00010701154543160541,
+      "loss": 0.1815,
+      "step": 515
+    },
+    {
+      "epoch": 2.6486486486486487,
+      "grad_norm": 0.4085369110107422,
+      "learning_rate": 0.00010665252476684864,
+      "loss": 0.0051,
+      "step": 516
+    },
+    {
+      "epoch": 2.653796653796654,
+      "grad_norm": 2.291940689086914,
+      "learning_rate": 0.0001062934179516936,
+      "loss": 0.0352,
+      "step": 517
+    },
+    {
+      "epoch": 2.658944658944659,
+      "grad_norm": 0.0635519027709961,
+      "learning_rate": 0.00010593422963658452,
+      "loss": 0.0012,
+      "step": 518
+    },
+    {
+      "epoch": 2.664092664092664,
+      "grad_norm": 1.7995437383651733,
+      "learning_rate": 0.00010557496447302102,
+      "loss": 0.005,
+      "step": 519
+    },
+    {
+      "epoch": 2.6692406692406694,
+      "grad_norm": 4.290238380432129,
+      "learning_rate": 0.00010521562711349788,
+      "loss": 0.1066,
+      "step": 520
+    },
+    {
+      "epoch": 2.6743886743886742,
+      "grad_norm": 1.1551425457000732,
+      "learning_rate": 0.00010485622221144484,
+      "loss": 0.0121,
+      "step": 521
+    },
+    {
+      "epoch": 2.6795366795366795,
+      "grad_norm": 0.09409017860889435,
+      "learning_rate": 0.00010449675442116634,
+      "loss": 0.0008,
+      "step": 522
+    },
+    {
+      "epoch": 2.684684684684685,
+      "grad_norm": 0.7674731016159058,
+      "learning_rate": 0.00010413722839778117,
+      "loss": 0.0022,
+      "step": 523
+    },
+    {
+      "epoch": 2.6898326898326896,
+      "grad_norm": 0.594530463218689,
+      "learning_rate": 0.00010377764879716234,
+      "loss": 0.0033,
+      "step": 524
+    },
+    {
+      "epoch": 2.694980694980695,
+      "grad_norm": 16.2296085357666,
+      "learning_rate": 0.00010341802027587659,
+      "loss": 0.22,
+      "step": 525
+    },
+    {
+      "epoch": 2.7001287001287,
+      "grad_norm": 3.139812707901001,
+      "learning_rate": 0.00010305834749112421,
+      "loss": 0.1727,
+      "step": 526
+    },
+    {
+      "epoch": 2.705276705276705,
+      "grad_norm": 24.274486541748047,
+      "learning_rate": 0.00010269863510067872,
+      "loss": 0.0916,
+      "step": 527
+    },
+    {
+      "epoch": 2.7104247104247103,
+      "grad_norm": 0.01993538811802864,
+      "learning_rate": 0.00010233888776282649,
+      "loss": 0.0004,
+      "step": 528
+    },
+    {
+      "epoch": 2.7155727155727156,
+      "grad_norm": 5.188899993896484,
+      "learning_rate": 0.00010197911013630659,
+      "loss": 0.0287,
+      "step": 529
+    },
+    {
+      "epoch": 2.7207207207207205,
+      "grad_norm": 2.857003927230835,
+      "learning_rate": 0.00010161930688025017,
+      "loss": 0.0129,
+      "step": 530
+    },
+    {
+      "epoch": 2.7258687258687258,
+      "grad_norm": 2.1086604595184326,
+      "learning_rate": 0.00010125948265412033,
+      "loss": 0.2294,
+      "step": 531
+    },
+    {
+      "epoch": 2.731016731016731,
+      "grad_norm": 5.304686069488525,
+      "learning_rate": 0.0001008996421176518,
+      "loss": 0.0386,
+      "step": 532
+    },
+    {
+      "epoch": 2.7361647361647363,
+      "grad_norm": 0.28571614623069763,
+      "learning_rate": 0.00010053978993079045,
+      "loss": 0.0022,
+      "step": 533
+    },
+    {
+      "epoch": 2.741312741312741,
+      "grad_norm": 8.396451950073242,
+      "learning_rate": 0.00010017993075363305,
+      "loss": 0.0193,
+      "step": 534
+    },
+    {
+      "epoch": 2.7464607464607464,
+      "grad_norm": 4.702949523925781,
+      "learning_rate": 9.982006924636697e-05,
+      "loss": 0.1092,
+      "step": 535
+    },
+    {
+      "epoch": 2.7516087516087517,
+      "grad_norm": 3.2389168739318848,
+      "learning_rate": 9.946021006920959e-05,
+      "loss": 0.0225,
+      "step": 536
+    },
+    {
+      "epoch": 2.756756756756757,
+      "grad_norm": 2.40317964553833,
+      "learning_rate": 9.910035788234822e-05,
+      "loss": 0.0144,
+      "step": 537
+    },
+    {
+      "epoch": 2.761904761904762,
+      "grad_norm": 1.8464491367340088,
+      "learning_rate": 9.874051734587968e-05,
+      "loss": 0.0202,
+      "step": 538
+    },
+    {
+      "epoch": 2.767052767052767,
+      "grad_norm": 2.658496141433716,
+      "learning_rate": 9.838069311974986e-05,
+      "loss": 0.0124,
+      "step": 539
+    },
+    {
+      "epoch": 2.7722007722007724,
+      "grad_norm": 7.585038185119629,
+      "learning_rate": 9.802088986369342e-05,
+      "loss": 0.1211,
+      "step": 540
+    },
+    {
+      "epoch": 2.7773487773487773,
+      "grad_norm": 2.9328129291534424,
+      "learning_rate": 9.766111223717352e-05,
+      "loss": 0.0349,
+      "step": 541
+    },
+    {
+      "epoch": 2.7824967824967826,
+      "grad_norm": 0.6864379644393921,
+      "learning_rate": 9.730136489932133e-05,
+      "loss": 0.0081,
+      "step": 542
+    },
+    {
+      "epoch": 2.787644787644788,
+      "grad_norm": 4.683409214019775,
+      "learning_rate": 9.694165250887584e-05,
+      "loss": 0.1795,
+      "step": 543
+    },
+    {
+      "epoch": 2.7927927927927927,
+      "grad_norm": 0.6130110025405884,
+      "learning_rate": 9.658197972412345e-05,
+      "loss": 0.0058,
+      "step": 544
+    },
+    {
+      "epoch": 2.797940797940798,
+      "grad_norm": 0.6490806341171265,
+      "learning_rate": 9.622235120283769e-05,
+      "loss": 0.0116,
+      "step": 545
+    },
+    {
+      "epoch": 2.8030888030888033,
+      "grad_norm": 1.1808786392211914,
+      "learning_rate": 9.586277160221884e-05,
+      "loss": 0.0114,
+      "step": 546
+    },
+    {
+      "epoch": 2.808236808236808,
+      "grad_norm": 2.684443235397339,
+      "learning_rate": 9.550324557883373e-05,
+      "loss": 0.2232,
+      "step": 547
+    },
+    {
+      "epoch": 2.8133848133848134,
+      "grad_norm": 10.013498306274414,
+      "learning_rate": 9.514377778855521e-05,
+      "loss": 0.1232,
+      "step": 548
+    },
+    {
+      "epoch": 2.8185328185328187,
+      "grad_norm": 0.043998826295137405,
+      "learning_rate": 9.478437288650213e-05,
+      "loss": 0.0009,
+      "step": 549
+    },
+    {
+      "epoch": 2.8236808236808235,
+      "grad_norm": 6.991347789764404,
+      "learning_rate": 9.442503552697899e-05,
+      "loss": 0.0247,
+      "step": 550
+    },
+    {
+      "epoch": 2.828828828828829,
+      "grad_norm": 0.24740318953990936,
+      "learning_rate": 9.406577036341548e-05,
+      "loss": 0.0057,
+      "step": 551
+    },
+    {
+      "epoch": 2.833976833976834,
+      "grad_norm": 0.46866661310195923,
+      "learning_rate": 9.37065820483064e-05,
+      "loss": 0.0106,
+      "step": 552
+    },
+    {
+      "epoch": 2.839124839124839,
+      "grad_norm": 0.47178831696510315,
+      "learning_rate": 9.334747523315137e-05,
+      "loss": 0.009,
+      "step": 553
+    },
+    {
+      "epoch": 2.844272844272844,
+      "grad_norm": 1.6347211599349976,
+      "learning_rate": 9.298845456839459e-05,
+      "loss": 0.007,
+      "step": 554
+    },
+    {
+      "epoch": 2.8494208494208495,
+      "grad_norm": 0.12780241668224335,
+      "learning_rate": 9.262952470336458e-05,
+      "loss": 0.002,
+      "step": 555
+    },
+    {
+      "epoch": 2.8545688545688543,
+      "grad_norm": 4.040054798126221,
+      "learning_rate": 9.227069028621406e-05,
+      "loss": 0.0206,
+      "step": 556
+    },
+    {
+      "epoch": 2.8597168597168596,
+      "grad_norm": 5.645758152008057,
+      "learning_rate": 9.19119559638596e-05,
+      "loss": 0.0945,
+      "step": 557
+    },
+    {
+      "epoch": 2.864864864864865,
+      "grad_norm": 2.331653118133545,
+      "learning_rate": 9.15533263819216e-05,
+      "loss": 0.0338,
+      "step": 558
+    },
+    {
+      "epoch": 2.8700128700128698,
+      "grad_norm": 1.8393348455429077,
+      "learning_rate": 9.119480618466409e-05,
+      "loss": 0.0185,
+      "step": 559
+    },
+    {
+      "epoch": 2.875160875160875,
+      "grad_norm": 126.81710815429688,
+      "learning_rate": 9.083640001493454e-05,
+      "loss": 0.1491,
+      "step": 560
+    },
+    {
+      "epoch": 2.8803088803088803,
+      "grad_norm": 1.250824213027954,
+      "learning_rate": 9.047811251410376e-05,
+      "loss": 0.0161,
+      "step": 561
+    },
+    {
+      "epoch": 2.8854568854568856,
+      "grad_norm": 23.800554275512695,
+      "learning_rate": 9.011994832200577e-05,
+      "loss": 0.0677,
+      "step": 562
+    },
+    {
+      "epoch": 2.8906048906048905,
+      "grad_norm": 3.137748956680298,
+      "learning_rate": 8.976191207687775e-05,
+      "loss": 0.0216,
+      "step": 563
+    },
+    {
+      "epoch": 2.8957528957528957,
+      "grad_norm": 11.139954566955566,
+      "learning_rate": 8.94040084153e-05,
+      "loss": 0.098,
+      "step": 564
+    },
+    {
+      "epoch": 2.900900900900901,
+      "grad_norm": 7.414366722106934,
+      "learning_rate": 8.904624197213585e-05,
+      "loss": 0.1656,
+      "step": 565
+    },
+    {
+      "epoch": 2.9060489060489063,
+      "grad_norm": 0.17657352983951569,
+      "learning_rate": 8.868861738047158e-05,
+      "loss": 0.0016,
+      "step": 566
+    },
+    {
+      "epoch": 2.911196911196911,
+      "grad_norm": 16.940473556518555,
+      "learning_rate": 8.83311392715566e-05,
+      "loss": 0.1336,
+      "step": 567
+    },
+    {
+      "epoch": 2.9163449163449164,
+      "grad_norm": 0.16067034006118774,
+      "learning_rate": 8.797381227474324e-05,
+      "loss": 0.0012,
+      "step": 568
+    },
+    {
+      "epoch": 2.9214929214929217,
+      "grad_norm": 0.563951313495636,
+      "learning_rate": 8.7616641017427e-05,
+      "loss": 0.0022,
+      "step": 569
+    },
+    {
+      "epoch": 2.9266409266409266,
+      "grad_norm": 5.007421016693115,
+      "learning_rate": 8.725963012498657e-05,
+      "loss": 0.0758,
+      "step": 570
+    },
+    {
+      "epoch": 2.931788931788932,
+      "grad_norm": 3.450300931930542,
+      "learning_rate": 8.690278422072384e-05,
+      "loss": 0.3816,
+      "step": 571
+    },
+    {
+      "epoch": 2.936936936936937,
+      "grad_norm": 7.5695109367370605,
+      "learning_rate": 8.654610792580415e-05,
+      "loss": 0.1364,
+      "step": 572
+    },
+    {
+      "epoch": 2.942084942084942,
+      "grad_norm": 13.636802673339844,
+      "learning_rate": 8.61896058591964e-05,
+      "loss": 0.1396,
+      "step": 573
+    },
+    {
+      "epoch": 2.9472329472329473,
+      "grad_norm": 2.10152006149292,
+      "learning_rate": 8.583328263761316e-05,
+      "loss": 0.0204,
+      "step": 574
+    },
+    {
+      "epoch": 2.9523809523809526,
+      "grad_norm": 44.20505905151367,
+      "learning_rate": 8.5477142875451e-05,
+      "loss": 0.1648,
+      "step": 575
+    },
+    {
+      "epoch": 2.9575289575289574,
+      "grad_norm": 2.720144271850586,
+      "learning_rate": 8.512119118473067e-05,
+      "loss": 0.0214,
+      "step": 576
+    },
+    {
+      "epoch": 2.9626769626769627,
+      "grad_norm": 21.961288452148438,
+      "learning_rate": 8.476543217503748e-05,
+      "loss": 0.0217,
+      "step": 577
+    },
+    {
+      "epoch": 2.967824967824968,
+      "grad_norm": 44.06306838989258,
+      "learning_rate": 8.440987045346134e-05,
+      "loss": 0.2875,
+      "step": 578
+    },
+    {
+      "epoch": 2.972972972972973,
+      "grad_norm": 8.869024276733398,
+      "learning_rate": 8.405451062453744e-05,
+      "loss": 0.0713,
+      "step": 579
+    },
+    {
+      "epoch": 2.978120978120978,
+      "grad_norm": 9.520781517028809,
+      "learning_rate": 8.369935729018634e-05,
+      "loss": 0.0662,
+      "step": 580
+    },
+    {
+      "epoch": 2.9832689832689834,
+      "grad_norm": 0.617495596408844,
+      "learning_rate": 8.334441504965455e-05,
+      "loss": 0.0105,
+      "step": 581
+    },
+    {
+      "epoch": 2.988416988416988,
+      "grad_norm": 2.0771758556365967,
+      "learning_rate": 8.29896884994549e-05,
+      "loss": 0.0791,
+      "step": 582
+    },
+    {
+      "epoch": 2.9935649935649935,
+      "grad_norm": 5.953162670135498,
+      "learning_rate": 8.263518223330697e-05,
+      "loss": 0.1749,
+      "step": 583
+    },
+    {
+      "epoch": 2.998712998712999,
+      "grad_norm": 4.602631568908691,
+      "learning_rate": 8.228090084207774e-05,
+      "loss": 0.0568,
+      "step": 584
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 1.2461541891098022,
+      "learning_rate": 8.192684891372198e-05,
+      "loss": 0.0141,
+      "step": 585
+    },
+    {
+      "epoch": 3.0051480051480053,
+      "grad_norm": 0.2215481400489807,
+      "learning_rate": 8.157303103322296e-05,
+      "loss": 0.0049,
+      "step": 586
+    },
+    {
+      "epoch": 3.01029601029601,
+      "grad_norm": 0.21982088685035706,
+      "learning_rate": 8.1219451782533e-05,
+      "loss": 0.0018,
+      "step": 587
+    },
+    {
+      "epoch": 3.0154440154440154,
+      "grad_norm": 0.6997593641281128,
+      "learning_rate": 8.086611574051417e-05,
+      "loss": 0.0114,
+      "step": 588
+    },
+    {
+      "epoch": 3.0205920205920207,
+      "grad_norm": 1.940786361694336,
+      "learning_rate": 8.051302748287895e-05,
+      "loss": 0.0493,
+      "step": 589
+    },
+    {
+      "epoch": 3.0257400257400255,
+      "grad_norm": 0.13358527421951294,
+      "learning_rate": 8.016019158213101e-05,
+      "loss": 0.003,
+      "step": 590
+    },
+    {
+      "epoch": 3.030888030888031,
+      "grad_norm": 0.06366240233182907,
+      "learning_rate": 7.980761260750607e-05,
+      "loss": 0.0007,
+      "step": 591
+    },
+    {
+      "epoch": 3.036036036036036,
+      "grad_norm": 0.15043139457702637,
+      "learning_rate": 7.945529512491251e-05,
+      "loss": 0.0017,
+      "step": 592
+    },
+    {
+      "epoch": 3.0411840411840414,
+      "grad_norm": 0.009339330717921257,
+      "learning_rate": 7.91032436968725e-05,
+      "loss": 0.0003,
+      "step": 593
+    },
+    {
+      "epoch": 3.0463320463320462,
+      "grad_norm": 0.034088149666786194,
+      "learning_rate": 7.875146288246275e-05,
+      "loss": 0.0006,
+      "step": 594
+    },
+    {
+      "epoch": 3.0514800514800515,
+      "grad_norm": 0.06482572853565216,
+      "learning_rate": 7.839995723725548e-05,
+      "loss": 0.0007,
+      "step": 595
+    },
+    {
+      "epoch": 3.056628056628057,
+      "grad_norm": 4.102519989013672,
+      "learning_rate": 7.804873131325954e-05,
+      "loss": 0.0125,
+      "step": 596
+    },
+    {
+      "epoch": 3.0617760617760617,
+      "grad_norm": 0.008063066750764847,
+      "learning_rate": 7.76977896588613e-05,
+      "loss": 0.0002,
+      "step": 597
+    },
+    {
+      "epoch": 3.066924066924067,
+      "grad_norm": 0.050067584961652756,
+      "learning_rate": 7.734713681876589e-05,
+      "loss": 0.0005,
+      "step": 598
+    },
+    {
+      "epoch": 3.0720720720720722,
+      "grad_norm": 4.622978687286377,
+      "learning_rate": 7.699677733393826e-05,
+      "loss": 0.0283,
+      "step": 599
+    },
+    {
+      "epoch": 3.077220077220077,
+      "grad_norm": 3.009385585784912,
+      "learning_rate": 7.66467157415444e-05,
+      "loss": 0.0063,
+      "step": 600
+    },
+    {
+      "epoch": 3.0823680823680824,
+      "grad_norm": 8.337059020996094,
+      "learning_rate": 7.629695657489257e-05,
+      "loss": 0.1835,
+      "step": 601
+    },
+    {
+      "epoch": 3.0875160875160876,
+      "grad_norm": 0.011049388907849789,
+      "learning_rate": 7.594750436337467e-05,
+      "loss": 0.0002,
+      "step": 602
+    },
+    {
+      "epoch": 3.0926640926640925,
+      "grad_norm": 0.9146316647529602,
+      "learning_rate": 7.55983636324074e-05,
+      "loss": 0.0032,
+      "step": 603
+    },
+    {
+      "epoch": 3.0978120978120978,
+      "grad_norm": 0.1846778839826584,
+      "learning_rate": 7.524953890337395e-05,
+      "loss": 0.0005,
+      "step": 604
+    },
+    {
+      "epoch": 3.102960102960103,
+      "grad_norm": 0.017363158985972404,
+      "learning_rate": 7.490103469356513e-05,
+      "loss": 0.0004,
+      "step": 605
+    },
+    {
+      "epoch": 3.108108108108108,
+      "grad_norm": 0.09835785627365112,
+      "learning_rate": 7.455285551612105e-05,
+      "loss": 0.0008,
+      "step": 606
+    },
+    {
+      "epoch": 3.113256113256113,
+      "grad_norm": 0.04886562004685402,
+      "learning_rate": 7.42050058799727e-05,
+      "loss": 0.0002,
+      "step": 607
+    },
+    {
+      "epoch": 3.1184041184041185,
+      "grad_norm": 0.8650006055831909,
+      "learning_rate": 7.385749028978346e-05,
+      "loss": 0.0058,
+      "step": 608
+    },
+    {
+      "epoch": 3.1235521235521237,
+      "grad_norm": 0.3819728493690491,
+      "learning_rate": 7.351031324589074e-05,
+      "loss": 0.0005,
+      "step": 609
+    },
+    {
+      "epoch": 3.1287001287001286,
+      "grad_norm": 0.06521546840667725,
+      "learning_rate": 7.316347924424787e-05,
+      "loss": 0.0005,
+      "step": 610
+    },
+    {
+      "epoch": 3.133848133848134,
+      "grad_norm": 0.8945662379264832,
+      "learning_rate": 7.281699277636572e-05,
+      "loss": 0.005,
+      "step": 611
+    },
+    {
+      "epoch": 3.138996138996139,
+      "grad_norm": 0.04454834386706352,
+      "learning_rate": 7.24708583292546e-05,
+      "loss": 0.0006,
+      "step": 612
+    },
+    {
+      "epoch": 3.144144144144144,
+      "grad_norm": 0.5885918736457825,
+      "learning_rate": 7.212508038536613e-05,
+      "loss": 0.0026,
+      "step": 613
+    },
+    {
+      "epoch": 3.1492921492921493,
+      "grad_norm": 2.309579372406006,
+      "learning_rate": 7.177966342253524e-05,
+      "loss": 0.0133,
+      "step": 614
+    },
+    {
+      "epoch": 3.1544401544401546,
+      "grad_norm": 0.17220087349414825,
+      "learning_rate": 7.143461191392206e-05,
+      "loss": 0.0016,
+      "step": 615
+    },
+    {
+      "epoch": 3.1595881595881594,
+      "grad_norm": 2.7658121585845947,
+      "learning_rate": 7.108993032795418e-05,
+      "loss": 0.0096,
+      "step": 616
+    },
+    {
+      "epoch": 3.1647361647361647,
+      "grad_norm": 0.008501471020281315,
+      "learning_rate": 7.07456231282686e-05,
+      "loss": 0.0002,
+      "step": 617
+    },
+    {
+      "epoch": 3.16988416988417,
+      "grad_norm": 1.094897985458374,
+      "learning_rate": 7.040169477365403e-05,
+      "loss": 0.0068,
+      "step": 618
+    },
+    {
+      "epoch": 3.175032175032175,
+      "grad_norm": 0.05795695632696152,
+      "learning_rate": 7.005814971799318e-05,
+      "loss": 0.0005,
+      "step": 619
+    },
+    {
+      "epoch": 3.18018018018018,
+      "grad_norm": 1.3671921491622925,
+      "learning_rate": 6.971499241020495e-05,
+      "loss": 0.0051,
+      "step": 620
+    },
+    {
+      "epoch": 3.1853281853281854,
+      "grad_norm": 0.878653883934021,
+      "learning_rate": 6.93722272941869e-05,
+      "loss": 0.0027,
+      "step": 621
+    },
+    {
+      "epoch": 3.1904761904761907,
+      "grad_norm": 0.12594768404960632,
+      "learning_rate": 6.902985880875773e-05,
+      "loss": 0.0012,
+      "step": 622
+    },
+    {
+      "epoch": 3.1956241956241955,
+      "grad_norm": 0.1369112879037857,
+      "learning_rate": 6.868789138759976e-05,
+      "loss": 0.0004,
+      "step": 623
+    },
+    {
+      "epoch": 3.200772200772201,
+      "grad_norm": 0.029848698526620865,
+      "learning_rate": 6.83463294592015e-05,
+      "loss": 0.0003,
+      "step": 624
+    },
+    {
+      "epoch": 3.205920205920206,
+      "grad_norm": 0.13778966665267944,
+      "learning_rate": 6.800517744680032e-05,
+      "loss": 0.001,
+      "step": 625
+    },
+    {
+      "epoch": 3.211068211068211,
+      "grad_norm": 0.8360304236412048,
+      "learning_rate": 6.766443976832517e-05,
+      "loss": 0.0025,
+      "step": 626
+    },
+    {
+      "epoch": 3.2162162162162162,
+      "grad_norm": 0.016754303127527237,
+      "learning_rate": 6.732412083633936e-05,
+      "loss": 0.0002,
+      "step": 627
+    },
+    {
+      "epoch": 3.2213642213642215,
+      "grad_norm": 0.09218200296163559,
+      "learning_rate": 6.698422505798338e-05,
+      "loss": 0.0007,
+      "step": 628
+    },
+    {
+      "epoch": 3.2265122265122264,
+      "grad_norm": 0.13681890070438385,
+      "learning_rate": 6.664475683491796e-05,
+      "loss": 0.0004,
+      "step": 629
+    },
+    {
+      "epoch": 3.2316602316602316,
+      "grad_norm": 0.002995492657646537,
+      "learning_rate": 6.630572056326687e-05,
+      "loss": 0.0001,
+      "step": 630
+    },
+    {
+      "epoch": 3.236808236808237,
+      "grad_norm": 0.032443318516016006,
+      "learning_rate": 6.59671206335602e-05,
+      "loss": 0.0002,
+      "step": 631
+    },
+    {
+      "epoch": 3.2419562419562418,
+      "grad_norm": 0.00456043379381299,
+      "learning_rate": 6.562896143067734e-05,
+      "loss": 0.0,
+      "step": 632
+    },
+    {
+      "epoch": 3.247104247104247,
+      "grad_norm": 1.5936821699142456,
+      "learning_rate": 6.529124733379024e-05,
+      "loss": 0.0026,
+      "step": 633
+    },
+    {
+      "epoch": 3.2522522522522523,
+      "grad_norm": 0.16793234646320343,
+      "learning_rate": 6.495398271630675e-05,
+      "loss": 0.0002,
+      "step": 634
+    },
+    {
+      "epoch": 3.257400257400257,
+      "grad_norm": 0.22084961831569672,
+      "learning_rate": 6.461717194581393e-05,
+      "loss": 0.0006,
+      "step": 635
+    },
+    {
+      "epoch": 3.2625482625482625,
+      "grad_norm": 3.4965598583221436,
+      "learning_rate": 6.428081938402149e-05,
+      "loss": 0.2658,
+      "step": 636
+    },
+    {
+      "epoch": 3.2676962676962678,
+      "grad_norm": 0.010342913679778576,
+      "learning_rate": 6.394492938670538e-05,
+      "loss": 0.0001,
+      "step": 637
+    },
+    {
+      "epoch": 3.2728442728442726,
+      "grad_norm": 0.09224316477775574,
+      "learning_rate": 6.360950630365126e-05,
+      "loss": 0.0006,
+      "step": 638
+    },
+    {
+      "epoch": 3.277992277992278,
+      "grad_norm": 0.0012620487250387669,
+      "learning_rate": 6.327455447859827e-05,
+      "loss": 0.0,
+      "step": 639
+    },
+    {
+      "epoch": 3.283140283140283,
+      "grad_norm": 0.10412480682134628,
+      "learning_rate": 6.294007824918276e-05,
+      "loss": 0.0004,
+      "step": 640
+    },
+    {
+      "epoch": 3.2882882882882885,
+      "grad_norm": 0.008364715613424778,
+      "learning_rate": 6.260608194688206e-05,
+      "loss": 0.0001,
+      "step": 641
+    },
+    {
+      "epoch": 3.2934362934362933,
+      "grad_norm": 0.3181469440460205,
+      "learning_rate": 6.227256989695848e-05,
+      "loss": 0.0008,
+      "step": 642
+    },
+    {
+      "epoch": 3.2985842985842986,
+      "grad_norm": 0.15038099884986877,
+      "learning_rate": 6.193954641840318e-05,
+      "loss": 0.0007,
+      "step": 643
+    },
+    {
+      "epoch": 3.303732303732304,
+      "grad_norm": 8.061479568481445,
+      "learning_rate": 6.160701582388038e-05,
+      "loss": 0.044,
+      "step": 644
+    },
+    {
+      "epoch": 3.3088803088803087,
+      "grad_norm": 0.010255937464535236,
+      "learning_rate": 6.12749824196714e-05,
+      "loss": 0.0001,
+      "step": 645
+    },
+    {
+      "epoch": 3.314028314028314,
+      "grad_norm": 1.5667259693145752,
+      "learning_rate": 6.0943450505618917e-05,
+      "loss": 0.0338,
+      "step": 646
+    },
+    {
+      "epoch": 3.3191763191763193,
+      "grad_norm": 0.001371858175843954,
+      "learning_rate": 6.061242437507131e-05,
+      "loss": 0.0,
+      "step": 647
+    },
+    {
+      "epoch": 3.3243243243243246,
+      "grad_norm": 0.11614678055047989,
+      "learning_rate": 6.028190831482703e-05,
+      "loss": 0.0014,
+      "step": 648
+    },
+    {
+      "epoch": 3.3294723294723294,
+      "grad_norm": 0.006284735165536404,
+      "learning_rate": 5.995190660507915e-05,
+      "loss": 0.0001,
+      "step": 649
+    },
+    {
+      "epoch": 3.3346203346203347,
+      "grad_norm": 0.5939111113548279,
+      "learning_rate": 5.962242351935985e-05,
+      "loss": 0.0021,
+      "step": 650
+    },
+    {
+      "epoch": 3.33976833976834,
+      "grad_norm": 0.025645185261964798,
+      "learning_rate": 5.929346332448511e-05,
+      "loss": 0.0003,
+      "step": 651
+    },
+    {
+      "epoch": 3.344916344916345,
+      "grad_norm": 27.726987838745117,
+      "learning_rate": 5.89650302804995e-05,
+      "loss": 0.0904,
+      "step": 652
+    },
+    {
+      "epoch": 3.35006435006435,
+      "grad_norm": 0.05819752812385559,
+      "learning_rate": 5.863712864062089e-05,
+      "loss": 0.0003,
+      "step": 653
+    },
+    {
+      "epoch": 3.3552123552123554,
+      "grad_norm": 5.248758316040039,
+      "learning_rate": 5.8309762651185484e-05,
+      "loss": 0.0361,
+      "step": 654
+    },
+    {
+      "epoch": 3.3603603603603602,
+      "grad_norm": 0.12868645787239075,
+      "learning_rate": 5.7982936551592906e-05,
+      "loss": 0.0015,
+      "step": 655
+    },
+    {
+      "epoch": 3.3655083655083655,
+      "grad_norm": 1.610785961151123,
+      "learning_rate": 5.765665457425102e-05,
+      "loss": 0.0264,
+      "step": 656
+    },
+    {
+      "epoch": 3.370656370656371,
+      "grad_norm": 0.07218751311302185,
+      "learning_rate": 5.733092094452135e-05,
+      "loss": 0.0004,
+      "step": 657
+    },
+    {
+      "epoch": 3.3758043758043756,
+      "grad_norm": 0.02295105904340744,
+      "learning_rate": 5.700573988066433e-05,
+      "loss": 0.0004,
+      "step": 658
+    },
+    {
+      "epoch": 3.380952380952381,
+      "grad_norm": 0.24112549424171448,
+      "learning_rate": 5.668111559378471e-05,
+      "loss": 0.0009,
+      "step": 659
+    },
+    {
+      "epoch": 3.386100386100386,
+      "grad_norm": 0.0482708103954792,
+      "learning_rate": 5.6357052287776765e-05,
+      "loss": 0.0003,
+      "step": 660
+    },
+    {
+      "epoch": 3.391248391248391,
+      "grad_norm": 0.7052761316299438,
+      "learning_rate": 5.6033554159270294e-05,
+      "loss": 0.0092,
+      "step": 661
+    },
+    {
+      "epoch": 3.3963963963963963,
+      "grad_norm": 12.661201477050781,
+      "learning_rate": 5.571062539757581e-05,
+      "loss": 0.1111,
+      "step": 662
+    },
+    {
+      "epoch": 3.4015444015444016,
+      "grad_norm": 4.992003440856934,
+      "learning_rate": 5.538827018463069e-05,
+      "loss": 0.1267,
+      "step": 663
+    },
+    {
+      "epoch": 3.4066924066924065,
+      "grad_norm": 0.11709136515855789,
+      "learning_rate": 5.50664926949447e-05,
+      "loss": 0.0014,
+      "step": 664
+    },
+    {
+      "epoch": 3.4118404118404118,
+      "grad_norm": 29.206151962280273,
+      "learning_rate": 5.474529709554612e-05,
+      "loss": 0.1259,
+      "step": 665
+    },
+    {
+      "epoch": 3.416988416988417,
+      "grad_norm": 0.016309332102537155,
+      "learning_rate": 5.4424687545927776e-05,
+      "loss": 0.0002,
+      "step": 666
+    },
+    {
+      "epoch": 3.4221364221364223,
+      "grad_norm": 0.07664862275123596,
+      "learning_rate": 5.410466819799306e-05,
+      "loss": 0.0005,
+      "step": 667
+    },
+    {
+      "epoch": 3.427284427284427,
+      "grad_norm": 0.37740492820739746,
+      "learning_rate": 5.378524319600231e-05,
+      "loss": 0.0031,
+      "step": 668
+    },
+    {
+      "epoch": 3.4324324324324325,
+      "grad_norm": 0.0018758762162178755,
+      "learning_rate": 5.346641667651897e-05,
+      "loss": 0.0,
+      "step": 669
+    },
+    {
+      "epoch": 3.4375804375804377,
+      "grad_norm": 0.0038844633381813765,
+      "learning_rate": 5.314819276835625e-05,
+      "loss": 0.0001,
+      "step": 670
+    },
+    {
+      "epoch": 3.4427284427284426,
+      "grad_norm": 0.00700331898406148,
+      "learning_rate": 5.283057559252341e-05,
+      "loss": 0.0001,
+      "step": 671
+    },
+    {
+      "epoch": 3.447876447876448,
+      "grad_norm": 16.81180191040039,
+      "learning_rate": 5.25135692621725e-05,
+      "loss": 0.0209,
+      "step": 672
+    },
+    {
+      "epoch": 3.453024453024453,
+      "grad_norm": 0.05513580143451691,
+      "learning_rate": 5.219717788254521e-05,
+      "loss": 0.0003,
+      "step": 673
+    },
+    {
+      "epoch": 3.458172458172458,
+      "grad_norm": 1.1876192092895508,
+      "learning_rate": 5.1881405550919493e-05,
+      "loss": 0.0074,
+      "step": 674
+    },
+    {
+      "epoch": 3.4633204633204633,
+      "grad_norm": 19.85177230834961,
+      "learning_rate": 5.156625635655672e-05,
+      "loss": 0.1497,
+      "step": 675
+    },
+    {
+      "epoch": 3.4684684684684686,
+      "grad_norm": 0.004083602223545313,
+      "learning_rate": 5.12517343806485e-05,
+      "loss": 0.0001,
+      "step": 676
+    },
+    {
+      "epoch": 3.473616473616474,
+      "grad_norm": 0.012743712402880192,
+      "learning_rate": 5.0937843696263966e-05,
+      "loss": 0.0001,
+      "step": 677
+    },
+    {
+      "epoch": 3.4787644787644787,
+      "grad_norm": 1.3095192909240723,
+      "learning_rate": 5.062458836829711e-05,
+      "loss": 0.0018,
+      "step": 678
+    },
+    {
+      "epoch": 3.483912483912484,
+      "grad_norm": 0.02523154765367508,
+      "learning_rate": 5.03119724534139e-05,
+      "loss": 0.0002,
+      "step": 679
+    },
+    {
+      "epoch": 3.4890604890604893,
+      "grad_norm": 0.035427629947662354,
+      "learning_rate": 5.000000000000002e-05,
+      "loss": 0.0003,
+      "step": 680
+    },
+    {
+      "epoch": 3.494208494208494,
+      "grad_norm": 3.5284361839294434,
+      "learning_rate": 4.96886750481082e-05,
+      "loss": 0.0244,
+      "step": 681
+    },
+    {
+      "epoch": 3.4993564993564994,
+      "grad_norm": 0.5589154362678528,
+      "learning_rate": 4.9378001629406e-05,
+      "loss": 0.0014,
+      "step": 682
+    },
+    {
+      "epoch": 3.5045045045045047,
+      "grad_norm": 0.01958996057510376,
+      "learning_rate": 4.9067983767123736e-05,
+      "loss": 0.0002,
+      "step": 683
+    },
+    {
+      "epoch": 3.5096525096525095,
+      "grad_norm": 0.04676245152950287,
+      "learning_rate": 4.875862547600207e-05,
+      "loss": 0.0005,
+      "step": 684
+    },
+    {
+      "epoch": 3.514800514800515,
+      "grad_norm": 0.06450199335813522,
+      "learning_rate": 4.8449930762240355e-05,
+      "loss": 0.0003,
+      "step": 685
+    },
+    {
+      "epoch": 3.51994851994852,
+      "grad_norm": 29.790546417236328,
+      "learning_rate": 4.814190362344454e-05,
+      "loss": 0.1438,
+      "step": 686
+    },
+    {
+      "epoch": 3.525096525096525,
+      "grad_norm": 9.640891075134277,
+      "learning_rate": 4.783454804857539e-05,
+      "loss": 0.036,
+      "step": 687
+    },
+    {
+      "epoch": 3.53024453024453,
+      "grad_norm": 0.020610546693205833,
+      "learning_rate": 4.752786801789703e-05,
+      "loss": 0.0001,
+      "step": 688
+    },
+    {
+      "epoch": 3.5353925353925355,
+      "grad_norm": 0.04118812084197998,
+      "learning_rate": 4.722186750292511e-05,
+      "loss": 0.0003,
+      "step": 689
+    },
+    {
+      "epoch": 3.5405405405405403,
+      "grad_norm": 0.10585196316242218,
+      "learning_rate": 4.6916550466375684e-05,
+      "loss": 0.0006,
+      "step": 690
+    },
+    {
+      "epoch": 3.5456885456885456,
+      "grad_norm": 0.169553741812706,
+      "learning_rate": 4.661192086211366e-05,
+      "loss": 0.0012,
+      "step": 691
+    },
+    {
+      "epoch": 3.550836550836551,
+      "grad_norm": 8.701114654541016,
+      "learning_rate": 4.630798263510162e-05,
+      "loss": 0.0198,
+      "step": 692
+    },
+    {
+      "epoch": 3.5559845559845558,
+      "grad_norm": 0.13898779451847076,
+      "learning_rate": 4.600473972134894e-05,
+      "loss": 0.0001,
+      "step": 693
+    },
+    {
+      "epoch": 3.561132561132561,
+      "grad_norm": 0.1630486696958542,
+      "learning_rate": 4.570219604786051e-05,
+      "loss": 0.0007,
+      "step": 694
+    },
+    {
+      "epoch": 3.5662805662805663,
+      "grad_norm": 0.12859463691711426,
+      "learning_rate": 4.540035553258619e-05,
+      "loss": 0.0006,
+      "step": 695
+    },
+    {
+      "epoch": 3.571428571428571,
+      "grad_norm": 0.49533209204673767,
+      "learning_rate": 4.5099222084369805e-05,
+      "loss": 0.0005,
+      "step": 696
+    },
+    {
+      "epoch": 3.5765765765765765,
+      "grad_norm": 19.686153411865234,
+      "learning_rate": 4.479879960289863e-05,
+      "loss": 0.0773,
+      "step": 697
+    },
+    {
+      "epoch": 3.5817245817245817,
+      "grad_norm": 0.2063945233821869,
+      "learning_rate": 4.449909197865303e-05,
+      "loss": 0.0008,
+      "step": 698
+    },
+    {
+      "epoch": 3.586872586872587,
+      "grad_norm": 18.023893356323242,
+      "learning_rate": 4.420010309285577e-05,
+      "loss": 0.0506,
+      "step": 699
+    },
+    {
+      "epoch": 3.592020592020592,
+      "grad_norm": 0.052725061774253845,
+      "learning_rate": 4.3901836817422124e-05,
+      "loss": 0.0006,
+      "step": 700
+    },
+    {
+      "epoch": 3.597168597168597,
+      "grad_norm": 0.021397685632109642,
+      "learning_rate": 4.360429701490934e-05,
+      "loss": 0.0001,
+      "step": 701
+    },
+    {
+      "epoch": 3.6023166023166024,
+      "grad_norm": 0.012758452445268631,
+      "learning_rate": 4.3307487538467006e-05,
+      "loss": 0.0001,
+      "step": 702
+    },
+    {
+      "epoch": 3.6074646074646077,
+      "grad_norm": 0.050585564225912094,
+      "learning_rate": 4.301141223178684e-05,
+      "loss": 0.0005,
+      "step": 703
+    },
+    {
+      "epoch": 3.6126126126126126,
+      "grad_norm": 1.5477937459945679,
+      "learning_rate": 4.271607492905303e-05,
+      "loss": 0.0045,
+      "step": 704
+    },
+    {
+      "epoch": 3.617760617760618,
+      "grad_norm": 6.680271148681641,
+      "learning_rate": 4.242147945489272e-05,
+      "loss": 0.1734,
+      "step": 705
+    },
+    {
+      "epoch": 3.622908622908623,
+      "grad_norm": 0.2742946147918701,
+      "learning_rate": 4.212762962432619e-05,
+      "loss": 0.0009,
+      "step": 706
+    },
+    {
+      "epoch": 3.628056628056628,
+      "grad_norm": 1.242712140083313,
+      "learning_rate": 4.183452924271776e-05,
+      "loss": 0.0076,
+      "step": 707
+    },
+    {
+      "epoch": 3.6332046332046333,
+      "grad_norm": 0.8588606715202332,
+      "learning_rate": 4.154218210572627e-05,
+      "loss": 0.0036,
+      "step": 708
+    },
+    {
+      "epoch": 3.6383526383526386,
+      "grad_norm": 0.001759879756718874,
+      "learning_rate": 4.125059199925599e-05,
+      "loss": 0.0,
+      "step": 709
+    },
+    {
+      "epoch": 3.6435006435006434,
+      "grad_norm": 0.01626988872885704,
+      "learning_rate": 4.0959762699407766e-05,
+      "loss": 0.0001,
+      "step": 710
+    },
+    {
+      "epoch": 3.6486486486486487,
+      "grad_norm": 0.901871383190155,
+      "learning_rate": 4.06696979724298e-05,
+      "loss": 0.0046,
+      "step": 711
+    },
+    {
+      "epoch": 3.653796653796654,
+      "grad_norm": 0.21007761359214783,
+      "learning_rate": 4.038040157466918e-05,
+      "loss": 0.0003,
+      "step": 712
+    },
+    {
+      "epoch": 3.658944658944659,
+      "grad_norm": 0.0374271459877491,
+      "learning_rate": 4.009187725252309e-05,
+      "loss": 0.0002,
+      "step": 713
+    },
+    {
+      "epoch": 3.664092664092664,
+      "grad_norm": 0.7672528028488159,
+      "learning_rate": 3.980412874239021e-05,
+      "loss": 0.0008,
+      "step": 714
+    },
+    {
+      "epoch": 3.6692406692406694,
+      "grad_norm": 0.7226448059082031,
+      "learning_rate": 3.95171597706226e-05,
+      "loss": 0.0033,
+      "step": 715
+    },
+    {
+      "epoch": 3.6743886743886742,
+      "grad_norm": 8.923495292663574,
+      "learning_rate": 3.9230974053477086e-05,
+      "loss": 0.0061,
+      "step": 716
+    },
+    {
+      "epoch": 3.6795366795366795,
+      "grad_norm": 0.010472980327904224,
+      "learning_rate": 3.8945575297067506e-05,
+      "loss": 0.0001,
+      "step": 717
+    },
+    {
+      "epoch": 3.684684684684685,
+      "grad_norm": 0.003836965188384056,
+      "learning_rate": 3.866096719731639e-05,
+      "loss": 0.0001,
+      "step": 718
+    },
+    {
+      "epoch": 3.6898326898326896,
+      "grad_norm": 0.6405398845672607,
+      "learning_rate": 3.8377153439907266e-05,
+      "loss": 0.0016,
+      "step": 719
+    },
+    {
+      "epoch": 3.694980694980695,
+      "grad_norm": 0.00374055583961308,
+      "learning_rate": 3.809413770023701e-05,
+      "loss": 0.0001,
+      "step": 720
+    },
+    {
+      "epoch": 3.7001287001287,
+      "grad_norm": 0.0010400298051536083,
+      "learning_rate": 3.7811923643367974e-05,
+      "loss": 0.0,
+      "step": 721
+    },
+    {
+      "epoch": 3.705276705276705,
+      "grad_norm": 0.17661187052726746,
+      "learning_rate": 3.7530514923980884e-05,
+      "loss": 0.0005,
+      "step": 722
+    },
+    {
+      "epoch": 3.7104247104247103,
+      "grad_norm": 0.3646957278251648,
+      "learning_rate": 3.724991518632717e-05,
+      "loss": 0.0011,
+      "step": 723
+    },
+    {
+      "epoch": 3.7155727155727156,
+      "grad_norm": 0.045435935258865356,
+      "learning_rate": 3.697012806418194e-05,
+      "loss": 0.0003,
+      "step": 724
+    },
+    {
+      "epoch": 3.7207207207207205,
+      "grad_norm": 0.013008499518036842,
+      "learning_rate": 3.669115718079702e-05,
+      "loss": 0.0001,
+      "step": 725
+    },
+    {
+      "epoch": 3.7258687258687258,
+      "grad_norm": 8.1923828125,
+      "learning_rate": 3.641300614885378e-05,
+      "loss": 0.0495,
+      "step": 726
+    },
+    {
+      "epoch": 3.731016731016731,
+      "grad_norm": 3.7366905212402344,
+      "learning_rate": 3.61356785704166e-05,
+      "loss": 0.0443,
+      "step": 727
+    },
+    {
+      "epoch": 3.7361647361647363,
+      "grad_norm": 0.062481578439474106,
+      "learning_rate": 3.585917803688603e-05,
+      "loss": 0.0003,
+      "step": 728
+    },
+    {
+      "epoch": 3.741312741312741,
+      "grad_norm": 0.04548823460936546,
+      "learning_rate": 3.558350812895238e-05,
+      "loss": 0.0002,
+      "step": 729
+    },
+    {
+      "epoch": 3.7464607464607464,
+      "grad_norm": 1.8862141370773315,
+      "learning_rate": 3.530867241654942e-05,
+      "loss": 0.0037,
+      "step": 730
+    },
+    {
+      "epoch": 3.7516087516087517,
+      "grad_norm": 1.1617075204849243,
+      "learning_rate": 3.503467445880789e-05,
+      "loss": 0.0097,
+      "step": 731
+    },
+    {
+      "epoch": 3.756756756756757,
+      "grad_norm": 0.831436038017273,
+      "learning_rate": 3.476151780400979e-05,
+      "loss": 0.0054,
+      "step": 732
+    },
+    {
+      "epoch": 3.761904761904762,
+      "grad_norm": 0.015881504863500595,
+      "learning_rate": 3.448920598954203e-05,
+      "loss": 0.0002,
+      "step": 733
+    },
+    {
+      "epoch": 3.767052767052767,
+      "grad_norm": 0.028170155361294746,
+      "learning_rate": 3.421774254185096e-05,
+      "loss": 0.0,
+      "step": 734
+    },
+    {
+      "epoch": 3.7722007722007724,
+      "grad_norm": 0.0074317120015621185,
+      "learning_rate": 3.394713097639647e-05,
+      "loss": 0.0001,
+      "step": 735
+    },
+    {
+      "epoch": 3.7773487773487773,
+      "grad_norm": 0.001211655791848898,
+      "learning_rate": 3.367737479760652e-05,
+      "loss": 0.0,
+      "step": 736
+    },
+    {
+      "epoch": 3.7824967824967826,
+      "grad_norm": 0.004705388098955154,
+      "learning_rate": 3.340847749883191e-05,
+      "loss": 0.0001,
+      "step": 737
+    },
+    {
+      "epoch": 3.787644787644788,
+      "grad_norm": 7.889760971069336,
+      "learning_rate": 3.31404425623008e-05,
+      "loss": 0.0922,
+      "step": 738
+    },
+    {
+      "epoch": 3.7927927927927927,
+      "grad_norm": 7.054792404174805,
+      "learning_rate": 3.287327345907381e-05,
+      "loss": 0.0503,
+      "step": 739
+    },
+    {
+      "epoch": 3.797940797940798,
+      "grad_norm": 5.205869674682617,
+      "learning_rate": 3.2606973648998915e-05,
+      "loss": 0.2434,
+      "step": 740
+    },
+    {
+      "epoch": 3.8030888030888033,
+      "grad_norm": 0.2362300604581833,
+      "learning_rate": 3.2341546580666796e-05,
+      "loss": 0.0012,
+      "step": 741
+    },
+    {
+      "epoch": 3.808236808236808,
+      "grad_norm": 0.012846952304244041,
+      "learning_rate": 3.207699569136608e-05,
+      "loss": 0.0001,
+      "step": 742
+    },
+    {
+      "epoch": 3.8133848133848134,
+      "grad_norm": 1.303697943687439,
+      "learning_rate": 3.1813324407038825e-05,
+      "loss": 0.0013,
+      "step": 743
+    },
+    {
+      "epoch": 3.8185328185328187,
+      "grad_norm": 0.0019087812397629023,
+      "learning_rate": 3.1550536142236145e-05,
+      "loss": 0.0,
+      "step": 744
+    },
+    {
+      "epoch": 3.8236808236808235,
+      "grad_norm": 0.02671913057565689,
+      "learning_rate": 3.128863430007414e-05,
+      "loss": 0.0002,
+      "step": 745
+    },
+    {
+      "epoch": 3.828828828828829,
+      "grad_norm": 0.04236476495862007,
+      "learning_rate": 3.102762227218957e-05,
+      "loss": 0.0002,
+      "step": 746
+    },
+    {
+      "epoch": 3.833976833976834,
+      "grad_norm": 0.023901907727122307,
+      "learning_rate": 3.0767503438696213e-05,
+      "loss": 0.0001,
+      "step": 747
+    },
+    {
+      "epoch": 3.839124839124839,
+      "grad_norm": 0.12795507907867432,
+      "learning_rate": 3.0508281168140806e-05,
+      "loss": 0.0005,
+      "step": 748
+    },
+    {
+      "epoch": 3.844272844272844,
+      "grad_norm": 0.031604088842868805,
+      "learning_rate": 3.0249958817459722e-05,
+      "loss": 0.0001,
+      "step": 749
+    },
+    {
+      "epoch": 3.8494208494208495,
+      "grad_norm": 0.05081813409924507,
+      "learning_rate": 2.999253973193522e-05,
+      "loss": 0.0003,
+      "step": 750
+    },
+    {
+      "epoch": 3.8545688545688543,
+      "grad_norm": 0.5902686715126038,
+      "learning_rate": 2.9736027245152275e-05,
+      "loss": 0.0009,
+      "step": 751
+    },
+    {
+      "epoch": 3.8597168597168596,
+      "grad_norm": 0.046482257544994354,
+      "learning_rate": 2.9480424678955443e-05,
+      "loss": 0.0003,
+      "step": 752
+    },
+    {
+      "epoch": 3.864864864864865,
+      "grad_norm": 0.25589659810066223,
+      "learning_rate": 2.9225735343405693e-05,
+      "loss": 0.0018,
+      "step": 753
+    },
+    {
+      "epoch": 3.8700128700128698,
+      "grad_norm": 0.01878608949482441,
+      "learning_rate": 2.897196253673773e-05,
+      "loss": 0.0002,
+      "step": 754
+    },
+    {
+      "epoch": 3.875160875160875,
+      "grad_norm": 4.430696964263916,
+      "learning_rate": 2.8719109545317103e-05,
+      "loss": 0.118,
+      "step": 755
+    },
+    {
+      "epoch": 3.8803088803088803,
+      "grad_norm": 0.000966802064795047,
+      "learning_rate": 2.8467179643597697e-05,
+      "loss": 0.0,
+      "step": 756
+    },
+    {
+      "epoch": 3.8854568854568856,
+      "grad_norm": 0.0021949531510472298,
+      "learning_rate": 2.8216176094079482e-05,
+      "loss": 0.0,
+      "step": 757
+    },
+    {
+      "epoch": 3.8906048906048905,
+      "grad_norm": 1.5945611000061035,
+      "learning_rate": 2.7966102147265994e-05,
+      "loss": 0.0073,
+      "step": 758
+    },
+    {
+      "epoch": 3.8957528957528957,
+      "grad_norm": 2.5020394325256348,
+      "learning_rate": 2.7716961041622534e-05,
+      "loss": 0.2119,
+      "step": 759
+    },
+    {
+      "epoch": 3.900900900900901,
+      "grad_norm": 0.051576703786849976,
+      "learning_rate": 2.746875600353398e-05,
+      "loss": 0.0004,
+      "step": 760
+    },
+    {
+      "epoch": 3.9060489060489063,
+      "grad_norm": 2.3891408443450928,
+      "learning_rate": 2.722149024726307e-05,
+      "loss": 0.0035,
+      "step": 761
+    },
+    {
+      "epoch": 3.911196911196911,
+      "grad_norm": 0.01467575877904892,
+      "learning_rate": 2.697516697490896e-05,
+      "loss": 0.0001,
+      "step": 762
+    },
+    {
+      "epoch": 3.9163449163449164,
+      "grad_norm": 0.041044168174266815,
+      "learning_rate": 2.6729789376365456e-05,
+      "loss": 0.0002,
+      "step": 763
+    },
+    {
+      "epoch": 3.9214929214929217,
+      "grad_norm": 2.140899181365967,
+      "learning_rate": 2.6485360629279987e-05,
+      "loss": 0.2384,
+      "step": 764
+    },
+    {
+      "epoch": 3.9266409266409266,
+      "grad_norm": 0.41608694195747375,
+      "learning_rate": 2.624188389901221e-05,
+      "loss": 0.0023,
+      "step": 765
+    },
+    {
+      "epoch": 3.931788931788932,
+      "grad_norm": 0.0035576452501118183,
+      "learning_rate": 2.599936233859326e-05,
+      "loss": 0.0001,
+      "step": 766
+    },
+    {
+      "epoch": 3.936936936936937,
+      "grad_norm": 0.07718081772327423,
+      "learning_rate": 2.5757799088684654e-05,
+      "loss": 0.0005,
+      "step": 767
+    },
+    {
+      "epoch": 3.942084942084942,
+      "grad_norm": 0.4290231466293335,
+      "learning_rate": 2.5517197277537886e-05,
+      "loss": 0.001,
+      "step": 768
+    },
+    {
+      "epoch": 3.9472329472329473,
+      "grad_norm": 61.477691650390625,
+      "learning_rate": 2.527756002095373e-05,
+      "loss": 0.1683,
+      "step": 769
+    },
+    {
+      "epoch": 3.9523809523809526,
+      "grad_norm": 0.2100103348493576,
+      "learning_rate": 2.5038890422241958e-05,
+      "loss": 0.0017,
+      "step": 770
+    },
+    {
+      "epoch": 3.9575289575289574,
+      "grad_norm": 0.06918445229530334,
+      "learning_rate": 2.480119157218108e-05,
+      "loss": 0.0005,
+      "step": 771
+    },
+    {
+      "epoch": 3.9626769626769627,
+      "grad_norm": 0.5728282928466797,
+      "learning_rate": 2.4564466548978525e-05,
+      "loss": 0.0021,
+      "step": 772
+    },
+    {
+      "epoch": 3.967824967824968,
+      "grad_norm": 0.032326120883226395,
+      "learning_rate": 2.432871841823047e-05,
+      "loss": 0.0003,
+      "step": 773
+    },
+    {
+      "epoch": 3.972972972972973,
+      "grad_norm": 0.11446495354175568,
+      "learning_rate": 2.4093950232882456e-05,
+      "loss": 0.0005,
+      "step": 774
+    },
+    {
+      "epoch": 3.978120978120978,
+      "grad_norm": 0.012360852211713791,
+      "learning_rate": 2.3860165033189587e-05,
+      "loss": 0.0001,
+      "step": 775
+    },
+    {
+      "epoch": 3.9832689832689834,
+      "grad_norm": 9.299793243408203,
+      "learning_rate": 2.3627365846677306e-05,
+      "loss": 0.1779,
+      "step": 776
+    },
+    {
+      "epoch": 3.988416988416988,
+      "grad_norm": 7.160345077514648,
+      "learning_rate": 2.339555568810221e-05,
+      "loss": 0.0124,
+      "step": 777
+    },
+    {
+      "epoch": 3.9935649935649935,
+      "grad_norm": 0.28910279273986816,
+      "learning_rate": 2.3164737559412854e-05,
+      "loss": 0.0015,
+      "step": 778
+    },
+    {
+      "epoch": 3.998712998712999,
+      "grad_norm": 1.926042079925537,
+      "learning_rate": 2.2934914449711087e-05,
+      "loss": 0.0111,
+      "step": 779
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.014931468293070793,
+      "learning_rate": 2.2706089335213122e-05,
+      "loss": 0.0001,
+      "step": 780
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 970,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 5,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.0784491173701962e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}