diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4864 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.0,
+  "eval_steps": 500,
+  "global_step": 690,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.008705114254624592,
+      "grad_norm": 1.703125,
+      "learning_rate": 5e-05,
+      "loss": 0.8632,
+      "step": 1
+    },
+    {
+      "epoch": 0.017410228509249184,
+      "grad_norm": 2.015625,
+      "learning_rate": 4.9956521739130436e-05,
+      "loss": 0.7891,
+      "step": 2
+    },
+    {
+      "epoch": 0.026115342763873776,
+      "grad_norm": 8.375,
+      "learning_rate": 4.9913043478260876e-05,
+      "loss": 0.7896,
+      "step": 3
+    },
+    {
+      "epoch": 0.03482045701849837,
+      "grad_norm": 3.328125,
+      "learning_rate": 4.986956521739131e-05,
+      "loss": 0.9271,
+      "step": 4
+    },
+    {
+      "epoch": 0.04352557127312296,
+      "grad_norm": 1.640625,
+      "learning_rate": 4.9826086956521736e-05,
+      "loss": 0.9154,
+      "step": 5
+    },
+    {
+      "epoch": 0.05223068552774755,
+      "grad_norm": 1.5390625,
+      "learning_rate": 4.9782608695652176e-05,
+      "loss": 0.9267,
+      "step": 6
+    },
+    {
+      "epoch": 0.060935799782372145,
+      "grad_norm": 2.71875,
+      "learning_rate": 4.973913043478261e-05,
+      "loss": 0.8189,
+      "step": 7
+    },
+    {
+      "epoch": 0.06964091403699674,
+      "grad_norm": 1.5859375,
+      "learning_rate": 4.969565217391304e-05,
+      "loss": 0.8018,
+      "step": 8
+    },
+    {
+      "epoch": 0.07834602829162132,
+      "grad_norm": 1.359375,
+      "learning_rate": 4.9652173913043483e-05,
+      "loss": 0.9439,
+      "step": 9
+    },
+    {
+      "epoch": 0.08705114254624592,
+      "grad_norm": 2.375,
+      "learning_rate": 4.960869565217392e-05,
+      "loss": 0.8444,
+      "step": 10
+    },
+    {
+      "epoch": 0.0957562568008705,
+      "grad_norm": 2.125,
+      "learning_rate": 4.956521739130435e-05,
+      "loss": 1.013,
+      "step": 11
+    },
+    {
+      "epoch": 0.1044613710554951,
+      "grad_norm": 1.7109375,
+      "learning_rate": 4.9521739130434784e-05,
+      "loss": 1.1719,
+      "step": 12
+    },
+    {
+      "epoch": 0.11316648531011969,
+      "grad_norm": 1.4296875,
+      "learning_rate": 4.947826086956522e-05,
+      "loss": 1.0034,
+      "step": 13
+    },
+    {
+      "epoch": 0.12187159956474429,
+      "grad_norm": 1.25,
+      "learning_rate": 4.943478260869566e-05,
+      "loss": 1.0015,
+      "step": 14
+    },
+    {
+      "epoch": 0.1305767138193689,
+      "grad_norm": 1.25,
+      "learning_rate": 4.939130434782609e-05,
+      "loss": 0.9413,
+      "step": 15
+    },
+    {
+      "epoch": 0.13928182807399347,
+      "grad_norm": 1.5546875,
+      "learning_rate": 4.9347826086956524e-05,
+      "loss": 1.0435,
+      "step": 16
+    },
+    {
+      "epoch": 0.14798694232861806,
+      "grad_norm": 2.265625,
+      "learning_rate": 4.930434782608696e-05,
+      "loss": 0.9285,
+      "step": 17
+    },
+    {
+      "epoch": 0.15669205658324264,
+      "grad_norm": 1.3046875,
+      "learning_rate": 4.926086956521739e-05,
+      "loss": 0.91,
+      "step": 18
+    },
+    {
+      "epoch": 0.16539717083786726,
+      "grad_norm": 1.296875,
+      "learning_rate": 4.9217391304347824e-05,
+      "loss": 0.937,
+      "step": 19
+    },
+    {
+      "epoch": 0.17410228509249184,
+      "grad_norm": 1.28125,
+      "learning_rate": 4.9173913043478265e-05,
+      "loss": 0.9194,
+      "step": 20
+    },
+    {
+      "epoch": 0.18280739934711643,
+      "grad_norm": 1.203125,
+      "learning_rate": 4.91304347826087e-05,
+      "loss": 0.9392,
+      "step": 21
+    },
+    {
+      "epoch": 0.191512513601741,
+      "grad_norm": 1.3203125,
+      "learning_rate": 4.908695652173913e-05,
+      "loss": 0.935,
+      "step": 22
+    },
+    {
+      "epoch": 0.20021762785636563,
+      "grad_norm": 1.2421875,
+      "learning_rate": 4.904347826086957e-05,
+      "loss": 0.7352,
+      "step": 23
+    },
+    {
+      "epoch": 0.2089227421109902,
+      "grad_norm": 1.2578125,
+      "learning_rate": 4.9e-05,
+      "loss": 0.8006,
+      "step": 24
+    },
+    {
+      "epoch": 0.2176278563656148,
+      "grad_norm": 1.3046875,
+      "learning_rate": 4.895652173913044e-05,
+      "loss": 1.1003,
+      "step": 25
+    },
+    {
+      "epoch": 0.22633297062023938,
+      "grad_norm": 1.578125,
+      "learning_rate": 4.891304347826087e-05,
+      "loss": 0.8786,
+      "step": 26
+    },
+    {
+      "epoch": 0.235038084874864,
+      "grad_norm": 1.390625,
+      "learning_rate": 4.8869565217391305e-05,
+      "loss": 0.9133,
+      "step": 27
+    },
+    {
+      "epoch": 0.24374319912948858,
+      "grad_norm": 1.109375,
+      "learning_rate": 4.8826086956521746e-05,
+      "loss": 0.9822,
+      "step": 28
+    },
+    {
+      "epoch": 0.25244831338411317,
+      "grad_norm": 1.6640625,
+      "learning_rate": 4.878260869565218e-05,
+      "loss": 1.0016,
+      "step": 29
+    },
+    {
+      "epoch": 0.2611534276387378,
+      "grad_norm": 1.34375,
+      "learning_rate": 4.873913043478261e-05,
+      "loss": 1.1215,
+      "step": 30
+    },
+    {
+      "epoch": 0.26985854189336234,
+      "grad_norm": 1.171875,
+      "learning_rate": 4.8695652173913046e-05,
+      "loss": 1.1457,
+      "step": 31
+    },
+    {
+      "epoch": 0.27856365614798695,
+      "grad_norm": 1.3671875,
+      "learning_rate": 4.865217391304348e-05,
+      "loss": 1.0368,
+      "step": 32
+    },
+    {
+      "epoch": 0.28726877040261156,
+      "grad_norm": 1.140625,
+      "learning_rate": 4.860869565217391e-05,
+      "loss": 0.9539,
+      "step": 33
+    },
+    {
+      "epoch": 0.2959738846572361,
+      "grad_norm": 1.4296875,
+      "learning_rate": 4.856521739130435e-05,
+      "loss": 0.9165,
+      "step": 34
+    },
+    {
+      "epoch": 0.30467899891186073,
+      "grad_norm": 1.2265625,
+      "learning_rate": 4.8521739130434786e-05,
+      "loss": 1.03,
+      "step": 35
+    },
+    {
+      "epoch": 0.3133841131664853,
+      "grad_norm": 1.5703125,
+      "learning_rate": 4.847826086956522e-05,
+      "loss": 1.2848,
+      "step": 36
+    },
+    {
+      "epoch": 0.3220892274211099,
+      "grad_norm": 1.21875,
+      "learning_rate": 4.843478260869565e-05,
+      "loss": 0.8663,
+      "step": 37
+    },
+    {
+      "epoch": 0.3307943416757345,
+      "grad_norm": 1.125,
+      "learning_rate": 4.839130434782609e-05,
+      "loss": 0.9547,
+      "step": 38
+    },
+    {
+      "epoch": 0.3394994559303591,
+      "grad_norm": 1.328125,
+      "learning_rate": 4.834782608695652e-05,
+      "loss": 0.872,
+      "step": 39
+    },
+    {
+      "epoch": 0.3482045701849837,
+      "grad_norm": 1.140625,
+      "learning_rate": 4.830434782608696e-05,
+      "loss": 0.8424,
+      "step": 40
+    },
+    {
+      "epoch": 0.35690968443960824,
+      "grad_norm": 1.2578125,
+      "learning_rate": 4.8260869565217394e-05,
+      "loss": 0.8246,
+      "step": 41
+    },
+    {
+      "epoch": 0.36561479869423286,
+      "grad_norm": 1.3046875,
+      "learning_rate": 4.8217391304347834e-05,
+      "loss": 0.8326,
+      "step": 42
+    },
+    {
+      "epoch": 0.37431991294885747,
+      "grad_norm": 1.2109375,
+      "learning_rate": 4.817391304347826e-05,
+      "loss": 0.9043,
+      "step": 43
+    },
+    {
+      "epoch": 0.383025027203482,
+      "grad_norm": 1.296875,
+      "learning_rate": 4.8130434782608694e-05,
+      "loss": 0.8888,
+      "step": 44
+    },
+    {
+      "epoch": 0.39173014145810664,
+      "grad_norm": 1.2265625,
+      "learning_rate": 4.8086956521739134e-05,
+      "loss": 0.778,
+      "step": 45
+    },
+    {
+      "epoch": 0.40043525571273125,
+      "grad_norm": 1.0546875,
+      "learning_rate": 4.804347826086957e-05,
+      "loss": 0.8684,
+      "step": 46
+    },
+    {
+      "epoch": 0.4091403699673558,
+      "grad_norm": 1.5078125,
+      "learning_rate": 4.8e-05,
+      "loss": 0.9351,
+      "step": 47
+    },
+    {
+      "epoch": 0.4178454842219804,
+      "grad_norm": 1.2421875,
+      "learning_rate": 4.795652173913044e-05,
+      "loss": 0.8708,
+      "step": 48
+    },
+    {
+      "epoch": 0.426550598476605,
+      "grad_norm": 1.28125,
+      "learning_rate": 4.7913043478260875e-05,
+      "loss": 0.9919,
+      "step": 49
+    },
+    {
+      "epoch": 0.4352557127312296,
+      "grad_norm": 1.1953125,
+      "learning_rate": 4.78695652173913e-05,
+      "loss": 0.8166,
+      "step": 50
+    },
+    {
+      "epoch": 0.4439608269858542,
+      "grad_norm": 1.3203125,
+      "learning_rate": 4.782608695652174e-05,
+      "loss": 0.9542,
+      "step": 51
+    },
+    {
+      "epoch": 0.45266594124047876,
+      "grad_norm": 1.2109375,
+      "learning_rate": 4.7782608695652175e-05,
+      "loss": 1.0959,
+      "step": 52
+    },
+    {
+      "epoch": 0.4613710554951034,
+      "grad_norm": 1.140625,
+      "learning_rate": 4.773913043478261e-05,
+      "loss": 0.8695,
+      "step": 53
+    },
+    {
+      "epoch": 0.470076169749728,
+      "grad_norm": 1.5234375,
+      "learning_rate": 4.769565217391305e-05,
+      "loss": 1.1411,
+      "step": 54
+    },
+    {
+      "epoch": 0.47878128400435255,
+      "grad_norm": 1.125,
+      "learning_rate": 4.765217391304348e-05,
+      "loss": 0.8354,
+      "step": 55
+    },
+    {
+      "epoch": 0.48748639825897716,
+      "grad_norm": 1.390625,
+      "learning_rate": 4.7608695652173916e-05,
+      "loss": 0.9133,
+      "step": 56
+    },
+    {
+      "epoch": 0.4961915125136017,
+      "grad_norm": 1.2578125,
+      "learning_rate": 4.756521739130435e-05,
+      "loss": 0.9042,
+      "step": 57
+    },
+    {
+      "epoch": 0.5048966267682263,
+      "grad_norm": 1.28125,
+      "learning_rate": 4.752173913043478e-05,
+      "loss": 0.9707,
+      "step": 58
+    },
+    {
+      "epoch": 0.5136017410228509,
+      "grad_norm": 1.2734375,
+      "learning_rate": 4.747826086956522e-05,
+      "loss": 0.7845,
+      "step": 59
+    },
+    {
+      "epoch": 0.5223068552774756,
+      "grad_norm": 1.3046875,
+      "learning_rate": 4.7434782608695656e-05,
+      "loss": 0.955,
+      "step": 60
+    },
+    {
+      "epoch": 0.5310119695321001,
+      "grad_norm": 1.546875,
+      "learning_rate": 4.739130434782609e-05,
+      "loss": 0.9521,
+      "step": 61
+    },
+    {
+      "epoch": 0.5397170837867247,
+      "grad_norm": 1.2578125,
+      "learning_rate": 4.734782608695652e-05,
+      "loss": 1.0155,
+      "step": 62
+    },
+    {
+      "epoch": 0.5484221980413493,
+      "grad_norm": 1.3125,
+      "learning_rate": 4.7304347826086956e-05,
+      "loss": 1.0093,
+      "step": 63
+    },
+    {
+      "epoch": 0.5571273122959739,
+      "grad_norm": 1.234375,
+      "learning_rate": 4.726086956521739e-05,
+      "loss": 0.8372,
+      "step": 64
+    },
+    {
+      "epoch": 0.5658324265505985,
+      "grad_norm": 1.0703125,
+      "learning_rate": 4.721739130434783e-05,
+      "loss": 0.8776,
+      "step": 65
+    },
+    {
+      "epoch": 0.5745375408052231,
+      "grad_norm": 1.2265625,
+      "learning_rate": 4.7173913043478264e-05,
+      "loss": 0.8854,
+      "step": 66
+    },
+    {
+      "epoch": 0.5832426550598476,
+      "grad_norm": 1.1171875,
+      "learning_rate": 4.71304347826087e-05,
+      "loss": 0.8966,
+      "step": 67
+    },
+    {
+      "epoch": 0.5919477693144722,
+      "grad_norm": 1.296875,
+      "learning_rate": 4.708695652173914e-05,
+      "loss": 0.9694,
+      "step": 68
+    },
+    {
+      "epoch": 0.6006528835690969,
+      "grad_norm": 1.1875,
+      "learning_rate": 4.7043478260869564e-05,
+      "loss": 1.062,
+      "step": 69
+    },
+    {
+      "epoch": 0.6093579978237215,
+      "grad_norm": 1.078125,
+      "learning_rate": 4.7e-05,
+      "loss": 0.8369,
+      "step": 70
+    },
+    {
+      "epoch": 0.6180631120783461,
+      "grad_norm": 1.1953125,
+      "learning_rate": 4.695652173913044e-05,
+      "loss": 0.7927,
+      "step": 71
+    },
+    {
+      "epoch": 0.6267682263329706,
+      "grad_norm": 1.4140625,
+      "learning_rate": 4.691304347826087e-05,
+      "loss": 1.3626,
+      "step": 72
+    },
+    {
+      "epoch": 0.6354733405875952,
+      "grad_norm": 1.1328125,
+      "learning_rate": 4.686956521739131e-05,
+      "loss": 0.7737,
+      "step": 73
+    },
+    {
+      "epoch": 0.6441784548422198,
+      "grad_norm": 1.0859375,
+      "learning_rate": 4.6826086956521745e-05,
+      "loss": 0.9044,
+      "step": 74
+    },
+    {
+      "epoch": 0.6528835690968444,
+      "grad_norm": 1.078125,
+      "learning_rate": 4.678260869565218e-05,
+      "loss": 0.9451,
+      "step": 75
+    },
+    {
+      "epoch": 0.661588683351469,
+      "grad_norm": 1.234375,
+      "learning_rate": 4.673913043478261e-05,
+      "loss": 1.0073,
+      "step": 76
+    },
+    {
+      "epoch": 0.6702937976060935,
+      "grad_norm": 1.1484375,
+      "learning_rate": 4.6695652173913045e-05,
+      "loss": 1.0206,
+      "step": 77
+    },
+    {
+      "epoch": 0.6789989118607181,
+      "grad_norm": 1.25,
+      "learning_rate": 4.665217391304348e-05,
+      "loss": 0.7854,
+      "step": 78
+    },
+    {
+      "epoch": 0.6877040261153428,
+      "grad_norm": 1.0859375,
+      "learning_rate": 4.660869565217392e-05,
+      "loss": 0.9661,
+      "step": 79
+    },
+    {
+      "epoch": 0.6964091403699674,
+      "grad_norm": 1.296875,
+      "learning_rate": 4.656521739130435e-05,
+      "loss": 1.0657,
+      "step": 80
+    },
+    {
+      "epoch": 0.705114254624592,
+      "grad_norm": 1.0390625,
+      "learning_rate": 4.6521739130434785e-05,
+      "loss": 0.6992,
+      "step": 81
+    },
+    {
+      "epoch": 0.7138193688792165,
+      "grad_norm": 1.0703125,
+      "learning_rate": 4.647826086956522e-05,
+      "loss": 0.8135,
+      "step": 82
+    },
+    {
+      "epoch": 0.7225244831338411,
+      "grad_norm": 1.359375,
+      "learning_rate": 4.643478260869565e-05,
+      "loss": 1.1033,
+      "step": 83
+    },
+    {
+      "epoch": 0.7312295973884657,
+      "grad_norm": 1.3828125,
+      "learning_rate": 4.6391304347826086e-05,
+      "loss": 0.8077,
+      "step": 84
+    },
+    {
+      "epoch": 0.7399347116430903,
+      "grad_norm": 1.0859375,
+      "learning_rate": 4.6347826086956526e-05,
+      "loss": 0.8582,
+      "step": 85
+    },
+    {
+      "epoch": 0.7486398258977149,
+      "grad_norm": 1.140625,
+      "learning_rate": 4.630434782608696e-05,
+      "loss": 1.0223,
+      "step": 86
+    },
+    {
+      "epoch": 0.7573449401523396,
+      "grad_norm": 1.2734375,
+      "learning_rate": 4.62608695652174e-05,
+      "loss": 0.7866,
+      "step": 87
+    },
+    {
+      "epoch": 0.766050054406964,
+      "grad_norm": 1.1484375,
+      "learning_rate": 4.6217391304347826e-05,
+      "loss": 0.804,
+      "step": 88
+    },
+    {
+      "epoch": 0.7747551686615887,
+      "grad_norm": 0.96484375,
+      "learning_rate": 4.617391304347826e-05,
+      "loss": 0.681,
+      "step": 89
+    },
+    {
+      "epoch": 0.7834602829162133,
+      "grad_norm": 1.0078125,
+      "learning_rate": 4.61304347826087e-05,
+      "loss": 0.9594,
+      "step": 90
+    },
+    {
+      "epoch": 0.7921653971708379,
+      "grad_norm": 1.0546875,
+      "learning_rate": 4.608695652173913e-05,
+      "loss": 0.9842,
+      "step": 91
+    },
+    {
+      "epoch": 0.8008705114254625,
+      "grad_norm": 1.09375,
+      "learning_rate": 4.6043478260869567e-05,
+      "loss": 0.9142,
+      "step": 92
+    },
+    {
+      "epoch": 0.809575625680087,
+      "grad_norm": 1.15625,
+      "learning_rate": 4.600000000000001e-05,
+      "loss": 0.7379,
+      "step": 93
+    },
+    {
+      "epoch": 0.8182807399347116,
+      "grad_norm": 1.3984375,
+      "learning_rate": 4.595652173913044e-05,
+      "loss": 1.0656,
+      "step": 94
+    },
+    {
+      "epoch": 0.8269858541893362,
+      "grad_norm": 1.15625,
+      "learning_rate": 4.591304347826087e-05,
+      "loss": 0.9404,
+      "step": 95
+    },
+    {
+      "epoch": 0.8356909684439608,
+      "grad_norm": 1.109375,
+      "learning_rate": 4.586956521739131e-05,
+      "loss": 1.0705,
+      "step": 96
+    },
+    {
+      "epoch": 0.8443960826985855,
+      "grad_norm": 1.1171875,
+      "learning_rate": 4.582608695652174e-05,
+      "loss": 1.1848,
+      "step": 97
+    },
+    {
+      "epoch": 0.85310119695321,
+      "grad_norm": 1.15625,
+      "learning_rate": 4.5782608695652174e-05,
+      "loss": 0.8771,
+      "step": 98
+    },
+    {
+      "epoch": 0.8618063112078346,
+      "grad_norm": 1.265625,
+      "learning_rate": 4.5739130434782614e-05,
+      "loss": 0.9586,
+      "step": 99
+    },
+    {
+      "epoch": 0.8705114254624592,
+      "grad_norm": 1.1015625,
+      "learning_rate": 4.569565217391305e-05,
+      "loss": 0.7516,
+      "step": 100
+    },
+    {
+      "epoch": 0.8792165397170838,
+      "grad_norm": 1.3125,
+      "learning_rate": 4.565217391304348e-05,
+      "loss": 0.8131,
+      "step": 101
+    },
+    {
+      "epoch": 0.8879216539717084,
+      "grad_norm": 1.109375,
+      "learning_rate": 4.5608695652173914e-05,
+      "loss": 0.9118,
+      "step": 102
+    },
+    {
+      "epoch": 0.8966267682263329,
+      "grad_norm": 1.4140625,
+      "learning_rate": 4.556521739130435e-05,
+      "loss": 0.8855,
+      "step": 103
+    },
+    {
+      "epoch": 0.9053318824809575,
+      "grad_norm": 1.1796875,
+      "learning_rate": 4.552173913043479e-05,
+      "loss": 0.7721,
+      "step": 104
+    },
+    {
+      "epoch": 0.9140369967355821,
+      "grad_norm": 1.2265625,
+      "learning_rate": 4.547826086956522e-05,
+      "loss": 0.7768,
+      "step": 105
+    },
+    {
+      "epoch": 0.9227421109902068,
+      "grad_norm": 1.0703125,
+      "learning_rate": 4.5434782608695655e-05,
+      "loss": 0.9557,
+      "step": 106
+    },
+    {
+      "epoch": 0.9314472252448314,
+      "grad_norm": 1.0625,
+      "learning_rate": 4.539130434782609e-05,
+      "loss": 0.8868,
+      "step": 107
+    },
+    {
+      "epoch": 0.940152339499456,
+      "grad_norm": 1.15625,
+      "learning_rate": 4.534782608695652e-05,
+      "loss": 0.8889,
+      "step": 108
+    },
+    {
+      "epoch": 0.9488574537540805,
+      "grad_norm": 1.15625,
+      "learning_rate": 4.5304347826086955e-05,
+      "loss": 1.1122,
+      "step": 109
+    },
+    {
+      "epoch": 0.9575625680087051,
+      "grad_norm": 1.1875,
+      "learning_rate": 4.5260869565217395e-05,
+      "loss": 0.7954,
+      "step": 110
+    },
+    {
+      "epoch": 0.9662676822633297,
+      "grad_norm": 1.1796875,
+      "learning_rate": 4.521739130434783e-05,
+      "loss": 0.8551,
+      "step": 111
+    },
+    {
+      "epoch": 0.9749727965179543,
+      "grad_norm": 1.1953125,
+      "learning_rate": 4.517391304347826e-05,
+      "loss": 1.0285,
+      "step": 112
+    },
+    {
+      "epoch": 0.9836779107725789,
+      "grad_norm": 1.25,
+      "learning_rate": 4.51304347826087e-05,
+      "loss": 1.0737,
+      "step": 113
+    },
+    {
+      "epoch": 0.9923830250272034,
+      "grad_norm": 1.015625,
+      "learning_rate": 4.508695652173913e-05,
+      "loss": 0.8448,
+      "step": 114
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.2421875,
+      "learning_rate": 4.504347826086956e-05,
+      "loss": 0.9171,
+      "step": 115
+    },
+    {
+      "epoch": 1.0087051142546246,
+      "grad_norm": 1.234375,
+      "learning_rate": 4.5e-05,
+      "loss": 0.7003,
+      "step": 116
+    },
+    {
+      "epoch": 1.0174102285092492,
+      "grad_norm": 1.5078125,
+      "learning_rate": 4.4956521739130436e-05,
+      "loss": 0.7938,
+      "step": 117
+    },
+    {
+      "epoch": 1.0261153427638738,
+      "grad_norm": 1.09375,
+      "learning_rate": 4.4913043478260876e-05,
+      "loss": 0.7323,
+      "step": 118
+    },
+    {
+      "epoch": 1.0348204570184985,
+      "grad_norm": 1.2421875,
+      "learning_rate": 4.486956521739131e-05,
+      "loss": 0.5957,
+      "step": 119
+    },
+    {
+      "epoch": 1.043525571273123,
+      "grad_norm": 1.453125,
+      "learning_rate": 4.482608695652174e-05,
+      "loss": 0.7519,
+      "step": 120
+    },
+    {
+      "epoch": 1.0522306855277475,
+      "grad_norm": 1.4453125,
+      "learning_rate": 4.478260869565218e-05,
+      "loss": 0.7003,
+      "step": 121
+    },
+    {
+      "epoch": 1.060935799782372,
+      "grad_norm": 1.4296875,
+      "learning_rate": 4.473913043478261e-05,
+      "loss": 0.8537,
+      "step": 122
+    },
+    {
+      "epoch": 1.0696409140369967,
+      "grad_norm": 1.2265625,
+      "learning_rate": 4.4695652173913044e-05,
+      "loss": 0.7305,
+      "step": 123
+    },
+    {
+      "epoch": 1.0783460282916213,
+      "grad_norm": 1.3203125,
+      "learning_rate": 4.4652173913043484e-05,
+      "loss": 0.6122,
+      "step": 124
+    },
+    {
+      "epoch": 1.087051142546246,
+      "grad_norm": 1.296875,
+      "learning_rate": 4.460869565217392e-05,
+      "loss": 0.9224,
+      "step": 125
+    },
+    {
+      "epoch": 1.0957562568008705,
+      "grad_norm": 0.94921875,
+      "learning_rate": 4.456521739130435e-05,
+      "loss": 0.6261,
+      "step": 126
+    },
+    {
+      "epoch": 1.1044613710554951,
+      "grad_norm": 1.125,
+      "learning_rate": 4.4521739130434784e-05,
+      "loss": 0.7336,
+      "step": 127
+    },
+    {
+      "epoch": 1.1131664853101197,
+      "grad_norm": 1.125,
+      "learning_rate": 4.447826086956522e-05,
+      "loss": 0.6978,
+      "step": 128
+    },
+    {
+      "epoch": 1.1218715995647444,
+      "grad_norm": 1.125,
+      "learning_rate": 4.443478260869565e-05,
+      "loss": 0.6352,
+      "step": 129
+    },
+    {
+      "epoch": 1.130576713819369,
+      "grad_norm": 1.125,
+      "learning_rate": 4.439130434782609e-05,
+      "loss": 0.4918,
+      "step": 130
+    },
+    {
+      "epoch": 1.1392818280739934,
+      "grad_norm": 1.25,
+      "learning_rate": 4.4347826086956525e-05,
+      "loss": 0.9074,
+      "step": 131
+    },
+    {
+      "epoch": 1.147986942328618,
+      "grad_norm": 1.296875,
+      "learning_rate": 4.430434782608696e-05,
+      "loss": 0.7653,
+      "step": 132
+    },
+    {
+      "epoch": 1.1566920565832426,
+      "grad_norm": 1.046875,
+      "learning_rate": 4.426086956521739e-05,
+      "loss": 0.6195,
+      "step": 133
+    },
+    {
+      "epoch": 1.1653971708378672,
+      "grad_norm": 1.1953125,
+      "learning_rate": 4.4217391304347825e-05,
+      "loss": 0.5887,
+      "step": 134
+    },
+    {
+      "epoch": 1.1741022850924918,
+      "grad_norm": 2.15625,
+      "learning_rate": 4.4173913043478265e-05,
+      "loss": 0.7727,
+      "step": 135
+    },
+    {
+      "epoch": 1.1828073993471164,
+      "grad_norm": 1.0703125,
+      "learning_rate": 4.41304347826087e-05,
+      "loss": 0.598,
+      "step": 136
+    },
+    {
+      "epoch": 1.191512513601741,
+      "grad_norm": 1.2265625,
+      "learning_rate": 4.408695652173913e-05,
+      "loss": 0.7572,
+      "step": 137
+    },
+    {
+      "epoch": 1.2002176278563657,
+      "grad_norm": 1.2265625,
+      "learning_rate": 4.404347826086957e-05,
+      "loss": 0.6791,
+      "step": 138
+    },
+    {
+      "epoch": 1.2089227421109903,
+      "grad_norm": 1.125,
+      "learning_rate": 4.4000000000000006e-05,
+      "loss": 0.7195,
+      "step": 139
+    },
+    {
+      "epoch": 1.2176278563656149,
+      "grad_norm": 0.984375,
+      "learning_rate": 4.395652173913043e-05,
+      "loss": 0.5021,
+      "step": 140
+    },
+    {
+      "epoch": 1.2263329706202395,
+      "grad_norm": 1.2578125,
+      "learning_rate": 4.391304347826087e-05,
+      "loss": 0.6567,
+      "step": 141
+    },
+    {
+      "epoch": 1.2350380848748639,
+      "grad_norm": 1.1875,
+      "learning_rate": 4.3869565217391306e-05,
+      "loss": 0.8551,
+      "step": 142
+    },
+    {
+      "epoch": 1.2437431991294885,
+      "grad_norm": 1.046875,
+      "learning_rate": 4.382608695652174e-05,
+      "loss": 0.7298,
+      "step": 143
+    },
+    {
+      "epoch": 1.252448313384113,
+      "grad_norm": 1.3203125,
+      "learning_rate": 4.378260869565218e-05,
+      "loss": 0.9639,
+      "step": 144
+    },
+    {
+      "epoch": 1.2611534276387377,
+      "grad_norm": 1.0546875,
+      "learning_rate": 4.373913043478261e-05,
+      "loss": 0.7321,
+      "step": 145
+    },
+    {
+      "epoch": 1.2698585418933623,
+      "grad_norm": 0.9921875,
+      "learning_rate": 4.3695652173913046e-05,
+      "loss": 0.6488,
+      "step": 146
+    },
+    {
+      "epoch": 1.278563656147987,
+      "grad_norm": 1.0703125,
+      "learning_rate": 4.365217391304348e-05,
+      "loss": 0.7783,
+      "step": 147
+    },
+    {
+      "epoch": 1.2872687704026116,
+      "grad_norm": 1.1484375,
+      "learning_rate": 4.360869565217391e-05,
+      "loss": 0.6042,
+      "step": 148
+    },
+    {
+      "epoch": 1.2959738846572362,
+      "grad_norm": 1.09375,
+      "learning_rate": 4.3565217391304353e-05,
+      "loss": 0.7951,
+      "step": 149
+    },
+    {
+      "epoch": 1.3046789989118608,
+      "grad_norm": 1.09375,
+      "learning_rate": 4.352173913043479e-05,
+      "loss": 0.6017,
+      "step": 150
+    },
+    {
+      "epoch": 1.3133841131664852,
+      "grad_norm": 1.2109375,
+      "learning_rate": 4.347826086956522e-05,
+      "loss": 0.6653,
+      "step": 151
+    },
+    {
+      "epoch": 1.32208922742111,
+      "grad_norm": 1.3828125,
+      "learning_rate": 4.3434782608695654e-05,
+      "loss": 0.9719,
+      "step": 152
+    },
+    {
+      "epoch": 1.3307943416757344,
+      "grad_norm": 1.015625,
+      "learning_rate": 4.339130434782609e-05,
+      "loss": 0.6013,
+      "step": 153
+    },
+    {
+      "epoch": 1.339499455930359,
+      "grad_norm": 1.1015625,
+      "learning_rate": 4.334782608695652e-05,
+      "loss": 0.6483,
+      "step": 154
+    },
+    {
+      "epoch": 1.3482045701849836,
+      "grad_norm": 1.0859375,
+      "learning_rate": 4.330434782608696e-05,
+      "loss": 0.6805,
+      "step": 155
+    },
+    {
+      "epoch": 1.3569096844396082,
+      "grad_norm": 1.140625,
+      "learning_rate": 4.3260869565217394e-05,
+      "loss": 0.7598,
+      "step": 156
+    },
+    {
+      "epoch": 1.3656147986942329,
+      "grad_norm": 1.046875,
+      "learning_rate": 4.321739130434783e-05,
+      "loss": 0.569,
+      "step": 157
+    },
+    {
+      "epoch": 1.3743199129488575,
+      "grad_norm": 1.0625,
+      "learning_rate": 4.317391304347827e-05,
+      "loss": 0.7223,
+      "step": 158
+    },
+    {
+      "epoch": 1.383025027203482,
+      "grad_norm": 1.1328125,
+      "learning_rate": 4.3130434782608695e-05,
+      "loss": 0.7309,
+      "step": 159
+    },
+    {
+      "epoch": 1.3917301414581067,
+      "grad_norm": 1.21875,
+      "learning_rate": 4.308695652173913e-05,
+      "loss": 0.9004,
+      "step": 160
+    },
+    {
+      "epoch": 1.4004352557127313,
+      "grad_norm": 1.125,
+      "learning_rate": 4.304347826086957e-05,
+      "loss": 0.6035,
+      "step": 161
+    },
+    {
+      "epoch": 1.4091403699673557,
+      "grad_norm": 1.2578125,
+      "learning_rate": 4.3e-05,
+      "loss": 0.957,
+      "step": 162
+    },
+    {
+      "epoch": 1.4178454842219805,
+      "grad_norm": 2.6875,
+      "learning_rate": 4.2956521739130435e-05,
+      "loss": 0.5841,
+      "step": 163
+    },
+    {
+      "epoch": 1.426550598476605,
+      "grad_norm": 1.1640625,
+      "learning_rate": 4.2913043478260875e-05,
+      "loss": 0.8018,
+      "step": 164
+    },
+    {
+      "epoch": 1.4352557127312295,
+      "grad_norm": 1.078125,
+      "learning_rate": 4.286956521739131e-05,
+      "loss": 0.6603,
+      "step": 165
+    },
+    {
+      "epoch": 1.4439608269858542,
+      "grad_norm": 1.40625,
+      "learning_rate": 4.282608695652174e-05,
+      "loss": 0.7107,
+      "step": 166
+    },
+    {
+      "epoch": 1.4526659412404788,
+      "grad_norm": 1.0390625,
+      "learning_rate": 4.2782608695652176e-05,
+      "loss": 0.7577,
+      "step": 167
+    },
+    {
+      "epoch": 1.4613710554951034,
+      "grad_norm": 1.0859375,
+      "learning_rate": 4.273913043478261e-05,
+      "loss": 0.7573,
+      "step": 168
+    },
+    {
+      "epoch": 1.470076169749728,
+      "grad_norm": 1.265625,
+      "learning_rate": 4.269565217391305e-05,
+      "loss": 0.7008,
+      "step": 169
+    },
+    {
+      "epoch": 1.4787812840043526,
+      "grad_norm": 1.1953125,
+      "learning_rate": 4.265217391304348e-05,
+      "loss": 0.6966,
+      "step": 170
+    },
+    {
+      "epoch": 1.4874863982589772,
+      "grad_norm": 1.1015625,
+      "learning_rate": 4.2608695652173916e-05,
+      "loss": 0.6855,
+      "step": 171
+    },
+    {
+      "epoch": 1.4961915125136018,
+      "grad_norm": 1.109375,
+      "learning_rate": 4.256521739130435e-05,
+      "loss": 0.67,
+      "step": 172
+    },
+    {
+      "epoch": 1.5048966267682262,
+      "grad_norm": 1.15625,
+      "learning_rate": 4.252173913043478e-05,
+      "loss": 0.8798,
+      "step": 173
+    },
+    {
+      "epoch": 1.513601741022851,
+      "grad_norm": 1.15625,
+      "learning_rate": 4.2478260869565216e-05,
+      "loss": 0.7921,
+      "step": 174
+    },
+    {
+      "epoch": 1.5223068552774754,
+      "grad_norm": 1.1171875,
+      "learning_rate": 4.2434782608695657e-05,
+      "loss": 0.6351,
+      "step": 175
+    },
+    {
+      "epoch": 1.5310119695321,
+      "grad_norm": 1.609375,
+      "learning_rate": 4.239130434782609e-05,
+      "loss": 0.6681,
+      "step": 176
+    },
+    {
+      "epoch": 1.5397170837867247,
+      "grad_norm": 1.265625,
+      "learning_rate": 4.2347826086956523e-05,
+      "loss": 0.7897,
+      "step": 177
+    },
+    {
+      "epoch": 1.5484221980413493,
+      "grad_norm": 1.2734375,
+      "learning_rate": 4.230434782608696e-05,
+      "loss": 0.772,
+      "step": 178
+    },
+    {
+      "epoch": 1.557127312295974,
+      "grad_norm": 1.0546875,
+      "learning_rate": 4.226086956521739e-05,
+      "loss": 0.5812,
+      "step": 179
+    },
+    {
+      "epoch": 1.5658324265505985,
+      "grad_norm": 1.0703125,
+      "learning_rate": 4.221739130434783e-05,
+      "loss": 0.5453,
+      "step": 180
+    },
+    {
+      "epoch": 1.5745375408052231,
+      "grad_norm": 1.0078125,
+      "learning_rate": 4.2173913043478264e-05,
+      "loss": 0.6831,
+      "step": 181
+    },
+    {
+      "epoch": 1.5832426550598475,
+      "grad_norm": 1.0625,
+      "learning_rate": 4.21304347826087e-05,
+      "loss": 0.767,
+      "step": 182
+    },
+    {
+      "epoch": 1.5919477693144723,
+      "grad_norm": 1.1484375,
+      "learning_rate": 4.208695652173914e-05,
+      "loss": 0.6145,
+      "step": 183
+    },
+    {
+      "epoch": 1.6006528835690967,
+      "grad_norm": 1.1015625,
+      "learning_rate": 4.204347826086957e-05,
+      "loss": 0.8171,
+      "step": 184
+    },
+    {
+      "epoch": 1.6093579978237216,
+      "grad_norm": 1.4609375,
+      "learning_rate": 4.2e-05,
+      "loss": 0.6039,
+      "step": 185
+    },
+    {
+      "epoch": 1.618063112078346,
+      "grad_norm": 1.1171875,
+      "learning_rate": 4.195652173913044e-05,
+      "loss": 0.7126,
+      "step": 186
+    },
+    {
+      "epoch": 1.6267682263329706,
+      "grad_norm": 1.1328125,
+      "learning_rate": 4.191304347826087e-05,
+      "loss": 0.5621,
+      "step": 187
+    },
+    {
+      "epoch": 1.6354733405875952,
+      "grad_norm": 1.0703125,
+      "learning_rate": 4.1869565217391305e-05,
+      "loss": 0.7605,
+      "step": 188
+    },
+    {
+      "epoch": 1.6441784548422198,
+      "grad_norm": 1.0625,
+      "learning_rate": 4.1826086956521745e-05,
+      "loss": 0.8704,
+      "step": 189
+    },
+    {
+      "epoch": 1.6528835690968444,
+      "grad_norm": 1.359375,
+      "learning_rate": 4.178260869565218e-05,
+      "loss": 0.7108,
+      "step": 190
+    },
+    {
+      "epoch": 1.661588683351469,
+      "grad_norm": 1.0859375,
+      "learning_rate": 4.1739130434782605e-05,
+      "loss": 0.67,
+      "step": 191
+    },
+    {
+      "epoch": 1.6702937976060936,
+      "grad_norm": 1.1171875,
+      "learning_rate": 4.1695652173913045e-05,
+      "loss": 0.9325,
+      "step": 192
+    },
+    {
+      "epoch": 1.678998911860718,
+      "grad_norm": 1.2578125,
+      "learning_rate": 4.165217391304348e-05,
+      "loss": 0.8226,
+      "step": 193
+    },
+    {
+      "epoch": 1.6877040261153429,
+      "grad_norm": 1.03125,
+      "learning_rate": 4.160869565217391e-05,
+      "loss": 0.8652,
+      "step": 194
+    },
+    {
+      "epoch": 1.6964091403699673,
+      "grad_norm": 1.171875,
+      "learning_rate": 4.156521739130435e-05,
+      "loss": 0.6873,
+      "step": 195
+    },
+    {
+      "epoch": 1.705114254624592,
+      "grad_norm": 1.1953125,
+      "learning_rate": 4.1521739130434786e-05,
+      "loss": 0.7681,
+      "step": 196
+    },
+    {
+      "epoch": 1.7138193688792165,
+      "grad_norm": 1.0625,
+      "learning_rate": 4.147826086956522e-05,
+      "loss": 0.6852,
+      "step": 197
+    },
+    {
+      "epoch": 1.722524483133841,
+      "grad_norm": 1.0546875,
+      "learning_rate": 4.143478260869565e-05,
+      "loss": 0.8035,
+      "step": 198
+    },
+    {
+      "epoch": 1.7312295973884657,
+      "grad_norm": 1.140625,
+      "learning_rate": 4.1391304347826086e-05,
+      "loss": 0.6638,
+      "step": 199
+    },
+    {
+      "epoch": 1.7399347116430903,
+      "grad_norm": 1.2265625,
+      "learning_rate": 4.1347826086956526e-05,
+      "loss": 0.8636,
+      "step": 200
+    },
+    {
+      "epoch": 1.748639825897715,
+      "grad_norm": 1.140625,
+      "learning_rate": 4.130434782608696e-05,
+      "loss": 0.9844,
+      "step": 201
+    },
+    {
+      "epoch": 1.7573449401523396,
+      "grad_norm": 1.1796875,
+      "learning_rate": 4.126086956521739e-05,
+      "loss": 0.7461,
+      "step": 202
+    },
+    {
+      "epoch": 1.7660500544069642,
+      "grad_norm": 1.1484375,
+      "learning_rate": 4.1217391304347827e-05,
+      "loss": 0.677,
+      "step": 203
+    },
+    {
+      "epoch": 1.7747551686615886,
+      "grad_norm": 1.234375,
+      "learning_rate": 4.117391304347826e-05,
+      "loss": 0.5882,
+      "step": 204
+    },
+    {
+      "epoch": 1.7834602829162134,
+      "grad_norm": 1.2109375,
+      "learning_rate": 4.1130434782608693e-05,
+      "loss": 0.7051,
+      "step": 205
+    },
+    {
+      "epoch": 1.7921653971708378,
+      "grad_norm": 1.2109375,
+      "learning_rate": 4.1086956521739134e-05,
+      "loss": 0.7658,
+      "step": 206
+    },
+    {
+      "epoch": 1.8008705114254626,
+      "grad_norm": 1.328125,
+      "learning_rate": 4.104347826086957e-05,
+      "loss": 0.8158,
+      "step": 207
+    },
+    {
+      "epoch": 1.809575625680087,
+      "grad_norm": 1.4765625,
+      "learning_rate": 4.1e-05,
+      "loss": 0.7081,
+      "step": 208
+    },
+    {
+      "epoch": 1.8182807399347116,
+      "grad_norm": 1.046875,
+      "learning_rate": 4.095652173913044e-05,
+      "loss": 0.648,
+      "step": 209
+    },
+    {
+      "epoch": 1.8269858541893362,
+      "grad_norm": 1.2734375,
+      "learning_rate": 4.091304347826087e-05,
+      "loss": 0.7721,
+      "step": 210
+    },
+    {
+      "epoch": 1.8356909684439608,
+      "grad_norm": 1.03125,
+      "learning_rate": 4.086956521739131e-05,
+      "loss": 0.5612,
+      "step": 211
+    },
+    {
+      "epoch": 1.8443960826985855,
+      "grad_norm": 1.1015625,
+      "learning_rate": 4.082608695652174e-05,
+      "loss": 0.635,
+      "step": 212
+    },
+    {
+      "epoch": 1.8531011969532099,
+      "grad_norm": 1.1015625,
+      "learning_rate": 4.0782608695652174e-05,
+      "loss": 0.7699,
+      "step": 213
+    },
+    {
+      "epoch": 1.8618063112078347,
+      "grad_norm": 1.234375,
+      "learning_rate": 4.0739130434782615e-05,
+      "loss": 0.7569,
+      "step": 214
+    },
+    {
+      "epoch": 1.870511425462459,
+      "grad_norm": 1.203125,
+      "learning_rate": 4.069565217391305e-05,
+      "loss": 0.7783,
+      "step": 215
+    },
+    {
+      "epoch": 1.879216539717084,
+      "grad_norm": 1.203125,
+      "learning_rate": 4.065217391304348e-05,
+      "loss": 0.7102,
+      "step": 216
+    },
+    {
+      "epoch": 1.8879216539717083,
+      "grad_norm": 2.53125,
+      "learning_rate": 4.0608695652173915e-05,
+      "loss": 1.0202,
+      "step": 217
+    },
+    {
+      "epoch": 1.896626768226333,
+      "grad_norm": 1.2109375,
+      "learning_rate": 4.056521739130435e-05,
+      "loss": 0.682,
+      "step": 218
+    },
+    {
+      "epoch": 1.9053318824809575,
+      "grad_norm": 1.03125,
+      "learning_rate": 4.052173913043478e-05,
+      "loss": 0.648,
+      "step": 219
+    },
+    {
+      "epoch": 1.9140369967355821,
+      "grad_norm": 1.234375,
+      "learning_rate": 4.047826086956522e-05,
+      "loss": 0.8244,
+      "step": 220
+    },
+    {
+      "epoch": 1.9227421109902068,
+      "grad_norm": 1.078125,
+      "learning_rate": 4.0434782608695655e-05,
+      "loss": 0.6392,
+      "step": 221
+    },
+    {
+      "epoch": 1.9314472252448314,
+      "grad_norm": 1.09375,
+      "learning_rate": 4.039130434782609e-05,
+      "loss": 0.7142,
+      "step": 222
+    },
+    {
+      "epoch": 1.940152339499456,
+      "grad_norm": 1.1328125,
+      "learning_rate": 4.034782608695652e-05,
+      "loss": 0.637,
+      "step": 223
+    },
+    {
+      "epoch": 1.9488574537540804,
+      "grad_norm": 1.03125,
+      "learning_rate": 4.0304347826086956e-05,
+      "loss": 0.7092,
+      "step": 224
+    },
+    {
+      "epoch": 1.9575625680087052,
+      "grad_norm": 1.0390625,
+      "learning_rate": 4.026086956521739e-05,
+      "loss": 0.6927,
+      "step": 225
+    },
+    {
+      "epoch": 1.9662676822633296,
+      "grad_norm": 1.125,
+      "learning_rate": 4.021739130434783e-05,
+      "loss": 0.744,
+      "step": 226
+    },
+    {
+      "epoch": 1.9749727965179544,
+      "grad_norm": 1.1484375,
+      "learning_rate": 4.017391304347826e-05,
+      "loss": 0.6604,
+      "step": 227
+    },
+    {
+      "epoch": 1.9836779107725788,
+      "grad_norm": 1.0625,
+      "learning_rate": 4.01304347826087e-05,
+      "loss": 0.7406,
+      "step": 228
+    },
+    {
+      "epoch": 1.9923830250272034,
+      "grad_norm": 1.078125,
+      "learning_rate": 4.008695652173913e-05,
+      "loss": 0.645,
+      "step": 229
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 1.4765625,
+      "learning_rate": 4.004347826086956e-05,
+      "loss": 0.7339,
+      "step": 230
+    },
+    {
+      "epoch": 2.0087051142546244,
+      "grad_norm": 1.2578125,
+      "learning_rate": 4e-05,
+      "loss": 0.6646,
+      "step": 231
+    },
+    {
+      "epoch": 2.0174102285092492,
+      "grad_norm": 1.21875,
+      "learning_rate": 3.995652173913044e-05,
+      "loss": 0.4969,
+      "step": 232
+    },
+    {
+      "epoch": 2.0261153427638736,
+      "grad_norm": 1.1953125,
+      "learning_rate": 3.991304347826087e-05,
+      "loss": 0.5202,
+      "step": 233
+    },
+    {
+      "epoch": 2.0348204570184985,
+      "grad_norm": 0.97265625,
+      "learning_rate": 3.986956521739131e-05,
+      "loss": 0.671,
+      "step": 234
+    },
+    {
+      "epoch": 2.043525571273123,
+      "grad_norm": 1.1875,
+      "learning_rate": 3.9826086956521744e-05,
+      "loss": 0.4306,
+      "step": 235
+    },
+    {
+      "epoch": 2.0522306855277477,
+      "grad_norm": 1.75,
+      "learning_rate": 3.978260869565217e-05,
+      "loss": 0.5232,
+      "step": 236
+    },
+    {
+      "epoch": 2.060935799782372,
+      "grad_norm": 1.6953125,
+      "learning_rate": 3.973913043478261e-05,
+      "loss": 0.498,
+      "step": 237
+    },
+    {
+      "epoch": 2.069640914036997,
+      "grad_norm": 1.6484375,
+      "learning_rate": 3.9695652173913044e-05,
+      "loss": 0.6858,
+      "step": 238
+    },
+    {
+      "epoch": 2.0783460282916213,
+      "grad_norm": 1.5703125,
+      "learning_rate": 3.965217391304348e-05,
+      "loss": 0.525,
+      "step": 239
+    },
+    {
+      "epoch": 2.087051142546246,
+      "grad_norm": 1.203125,
+      "learning_rate": 3.960869565217392e-05,
+      "loss": 0.4226,
+      "step": 240
+    },
+    {
+      "epoch": 2.0957562568008705,
+      "grad_norm": 1.1953125,
+      "learning_rate": 3.956521739130435e-05,
+      "loss": 0.5474,
+      "step": 241
+    },
+    {
+      "epoch": 2.104461371055495,
+      "grad_norm": 1.109375,
+      "learning_rate": 3.9521739130434785e-05,
+      "loss": 0.4487,
+      "step": 242
+    },
+    {
+      "epoch": 2.1131664853101197,
+      "grad_norm": 1.125,
+      "learning_rate": 3.947826086956522e-05,
+      "loss": 0.404,
+      "step": 243
+    },
+    {
+      "epoch": 2.121871599564744,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.943478260869565e-05,
+      "loss": 0.4901,
+      "step": 244
+    },
+    {
+      "epoch": 2.130576713819369,
+      "grad_norm": 1.1875,
+      "learning_rate": 3.939130434782609e-05,
+      "loss": 0.417,
+      "step": 245
+    },
+    {
+      "epoch": 2.1392818280739934,
+      "grad_norm": 1.1171875,
+      "learning_rate": 3.9347826086956525e-05,
+      "loss": 0.7131,
+      "step": 246
+    },
+    {
+      "epoch": 2.147986942328618,
+      "grad_norm": 1.296875,
+      "learning_rate": 3.930434782608696e-05,
+      "loss": 0.5996,
+      "step": 247
+    },
+    {
+      "epoch": 2.1566920565832426,
+      "grad_norm": 1.171875,
+      "learning_rate": 3.926086956521739e-05,
+      "loss": 0.6252,
+      "step": 248
+    },
+    {
+      "epoch": 2.1653971708378674,
+      "grad_norm": 1.28125,
+      "learning_rate": 3.9217391304347825e-05,
+      "loss": 0.4977,
+      "step": 249
+    },
+    {
+      "epoch": 2.174102285092492,
+      "grad_norm": 1.1640625,
+      "learning_rate": 3.917391304347826e-05,
+      "loss": 0.7226,
+      "step": 250
+    },
+    {
+      "epoch": 2.1828073993471167,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.91304347826087e-05,
+      "loss": 0.7221,
+      "step": 251
+    },
+    {
+      "epoch": 2.191512513601741,
+      "grad_norm": 1.171875,
+      "learning_rate": 3.908695652173913e-05,
+      "loss": 0.6118,
+      "step": 252
+    },
+    {
+      "epoch": 2.2002176278563654,
+      "grad_norm": 1.109375,
+      "learning_rate": 3.9043478260869566e-05,
+      "loss": 0.8739,
+      "step": 253
+    },
+    {
+      "epoch": 2.2089227421109903,
+      "grad_norm": 1.2734375,
+      "learning_rate": 3.9000000000000006e-05,
+      "loss": 0.5692,
+      "step": 254
+    },
+    {
+      "epoch": 2.2176278563656147,
+      "grad_norm": 1.2890625,
+      "learning_rate": 3.895652173913043e-05,
+      "loss": 0.527,
+      "step": 255
+    },
+    {
+      "epoch": 2.2263329706202395,
+      "grad_norm": 1.1015625,
+      "learning_rate": 3.8913043478260866e-05,
+      "loss": 0.4093,
+      "step": 256
+    },
+    {
+      "epoch": 2.235038084874864,
+      "grad_norm": 1.3828125,
+      "learning_rate": 3.8869565217391306e-05,
+      "loss": 0.57,
+      "step": 257
+    },
+    {
+      "epoch": 2.2437431991294887,
+      "grad_norm": 1.15625,
+      "learning_rate": 3.882608695652174e-05,
+      "loss": 0.4616,
+      "step": 258
+    },
+    {
+      "epoch": 2.252448313384113,
+      "grad_norm": 1.2265625,
+      "learning_rate": 3.878260869565218e-05,
+      "loss": 0.5554,
+      "step": 259
+    },
+    {
+      "epoch": 2.261153427638738,
+      "grad_norm": 1.0078125,
+      "learning_rate": 3.8739130434782613e-05,
+      "loss": 0.4197,
+      "step": 260
+    },
+    {
+      "epoch": 2.2698585418933623,
+      "grad_norm": 1.140625,
+      "learning_rate": 3.869565217391305e-05,
+      "loss": 0.4527,
+      "step": 261
+    },
+    {
+      "epoch": 2.2785636561479867,
+      "grad_norm": 1.28125,
+      "learning_rate": 3.865217391304348e-05,
+      "loss": 0.6683,
+      "step": 262
+    },
+    {
+      "epoch": 2.2872687704026116,
+      "grad_norm": 1.203125,
+      "learning_rate": 3.8608695652173914e-05,
+      "loss": 0.4808,
+      "step": 263
+    },
+    {
+      "epoch": 2.295973884657236,
+      "grad_norm": 1.109375,
+      "learning_rate": 3.856521739130435e-05,
+      "loss": 0.4871,
+      "step": 264
+    },
+    {
+      "epoch": 2.304678998911861,
+      "grad_norm": 1.2421875,
+      "learning_rate": 3.852173913043479e-05,
+      "loss": 0.6857,
+      "step": 265
+    },
+    {
+      "epoch": 2.313384113166485,
+      "grad_norm": 1.1328125,
+      "learning_rate": 3.847826086956522e-05,
+      "loss": 0.6002,
+      "step": 266
+    },
+    {
+      "epoch": 2.32208922742111,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.8434782608695654e-05,
+      "loss": 0.5997,
+      "step": 267
+    },
+    {
+      "epoch": 2.3307943416757344,
+      "grad_norm": 1.1640625,
+      "learning_rate": 3.839130434782609e-05,
+      "loss": 0.4839,
+      "step": 268
+    },
+    {
+      "epoch": 2.3394994559303592,
+      "grad_norm": 1.1640625,
+      "learning_rate": 3.834782608695652e-05,
+      "loss": 0.5754,
+      "step": 269
+    },
+    {
+      "epoch": 2.3482045701849836,
+      "grad_norm": 1.21875,
+      "learning_rate": 3.8304347826086955e-05,
+      "loss": 0.5072,
+      "step": 270
+    },
+    {
+      "epoch": 2.356909684439608,
+      "grad_norm": 1.453125,
+      "learning_rate": 3.8260869565217395e-05,
+      "loss": 0.6542,
+      "step": 271
+    },
+    {
+      "epoch": 2.365614798694233,
+      "grad_norm": 1.2265625,
+      "learning_rate": 3.821739130434783e-05,
+      "loss": 0.6175,
+      "step": 272
+    },
+    {
+      "epoch": 2.3743199129488577,
+      "grad_norm": 1.0859375,
+      "learning_rate": 3.817391304347827e-05,
+      "loss": 0.5063,
+      "step": 273
+    },
+    {
+      "epoch": 2.383025027203482,
+      "grad_norm": 1.1484375,
+      "learning_rate": 3.8130434782608695e-05,
+      "loss": 0.5095,
+      "step": 274
+    },
+    {
+      "epoch": 2.3917301414581065,
+      "grad_norm": 1.5078125,
+      "learning_rate": 3.808695652173913e-05,
+      "loss": 0.5236,
+      "step": 275
+    },
+    {
+      "epoch": 2.4004352557127313,
+      "grad_norm": 1.25,
+      "learning_rate": 3.804347826086957e-05,
+      "loss": 0.6417,
+      "step": 276
+    },
+    {
+      "epoch": 2.4091403699673557,
+      "grad_norm": 1.2421875,
+      "learning_rate": 3.8e-05,
+      "loss": 0.5618,
+      "step": 277
+    },
+    {
+      "epoch": 2.4178454842219805,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.7956521739130436e-05,
+      "loss": 0.409,
+      "step": 278
+    },
+    {
+      "epoch": 2.426550598476605,
+      "grad_norm": 1.2421875,
+      "learning_rate": 3.7913043478260876e-05,
+      "loss": 0.4815,
+      "step": 279
+    },
+    {
+      "epoch": 2.4352557127312298,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.786956521739131e-05,
+      "loss": 0.5533,
+      "step": 280
+    },
+    {
+      "epoch": 2.443960826985854,
+      "grad_norm": 1.1875,
+      "learning_rate": 3.7826086956521736e-05,
+      "loss": 0.4977,
+      "step": 281
+    },
+    {
+      "epoch": 2.452665941240479,
+      "grad_norm": 1.265625,
+      "learning_rate": 3.7782608695652176e-05,
+      "loss": 0.5402,
+      "step": 282
+    },
+    {
+      "epoch": 2.4613710554951034,
+      "grad_norm": 1.234375,
+      "learning_rate": 3.773913043478261e-05,
+      "loss": 0.4572,
+      "step": 283
+    },
+    {
+      "epoch": 2.4700761697497278,
+      "grad_norm": 1.1953125,
+      "learning_rate": 3.769565217391304e-05,
+      "loss": 0.5009,
+      "step": 284
+    },
+    {
+      "epoch": 2.4787812840043526,
+      "grad_norm": 1.140625,
+      "learning_rate": 3.765217391304348e-05,
+      "loss": 0.508,
+      "step": 285
+    },
+    {
+      "epoch": 2.487486398258977,
+      "grad_norm": 1.3359375,
+      "learning_rate": 3.7608695652173917e-05,
+      "loss": 0.5922,
+      "step": 286
+    },
+    {
+      "epoch": 2.496191512513602,
+      "grad_norm": 1.15625,
+      "learning_rate": 3.756521739130435e-05,
+      "loss": 0.5631,
+      "step": 287
+    },
+    {
+      "epoch": 2.504896626768226,
+      "grad_norm": 1.1953125,
+      "learning_rate": 3.752173913043478e-05,
+      "loss": 0.5335,
+      "step": 288
+    },
+    {
+      "epoch": 2.513601741022851,
+      "grad_norm": 1.28125,
+      "learning_rate": 3.747826086956522e-05,
+      "loss": 0.6231,
+      "step": 289
+    },
+    {
+      "epoch": 2.5223068552774754,
+      "grad_norm": 1.4765625,
+      "learning_rate": 3.743478260869566e-05,
+      "loss": 0.5049,
+      "step": 290
+    },
+    {
+      "epoch": 2.5310119695321003,
+      "grad_norm": 1.25,
+      "learning_rate": 3.739130434782609e-05,
+      "loss": 0.6165,
+      "step": 291
+    },
+    {
+      "epoch": 2.5397170837867247,
+      "grad_norm": 1.2265625,
+      "learning_rate": 3.7347826086956524e-05,
+      "loss": 0.5344,
+      "step": 292
+    },
+    {
+      "epoch": 2.548422198041349,
+      "grad_norm": 1.078125,
+      "learning_rate": 3.730434782608696e-05,
+      "loss": 0.5464,
+      "step": 293
+    },
+    {
+      "epoch": 2.557127312295974,
+      "grad_norm": 1.1640625,
+      "learning_rate": 3.726086956521739e-05,
+      "loss": 0.571,
+      "step": 294
+    },
+    {
+      "epoch": 2.5658324265505987,
+      "grad_norm": 1.1328125,
+      "learning_rate": 3.7217391304347824e-05,
+      "loss": 0.4843,
+      "step": 295
+    },
+    {
+      "epoch": 2.574537540805223,
+      "grad_norm": 1.4453125,
+      "learning_rate": 3.7173913043478264e-05,
+      "loss": 0.4947,
+      "step": 296
+    },
+    {
+      "epoch": 2.5832426550598475,
+      "grad_norm": 1.171875,
+      "learning_rate": 3.71304347826087e-05,
+      "loss": 0.5202,
+      "step": 297
+    },
+    {
+      "epoch": 2.5919477693144723,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.708695652173913e-05,
+      "loss": 0.4633,
+      "step": 298
+    },
+    {
+      "epoch": 2.6006528835690967,
+      "grad_norm": 1.3125,
+      "learning_rate": 3.704347826086957e-05,
+      "loss": 0.4723,
+      "step": 299
+    },
+    {
+      "epoch": 2.6093579978237216,
+      "grad_norm": 1.265625,
+      "learning_rate": 3.7e-05,
+      "loss": 0.4811,
+      "step": 300
+    },
+    {
+      "epoch": 2.618063112078346,
+      "grad_norm": 1.34375,
+      "learning_rate": 3.695652173913043e-05,
+      "loss": 0.7662,
+      "step": 301
+    },
+    {
+      "epoch": 2.6267682263329704,
+      "grad_norm": 1.3125,
+      "learning_rate": 3.691304347826087e-05,
+      "loss": 0.67,
+      "step": 302
+    },
+    {
+      "epoch": 2.635473340587595,
+      "grad_norm": 1.3671875,
+      "learning_rate": 3.6869565217391305e-05,
+      "loss": 0.6694,
+      "step": 303
+    },
+    {
+      "epoch": 2.64417845484222,
+      "grad_norm": 0.9921875,
+      "learning_rate": 3.682608695652174e-05,
+      "loss": 0.5212,
+      "step": 304
+    },
+    {
+      "epoch": 2.6528835690968444,
+      "grad_norm": 1.09375,
+      "learning_rate": 3.678260869565218e-05,
+      "loss": 0.4885,
+      "step": 305
+    },
+    {
+      "epoch": 2.661588683351469,
+      "grad_norm": 1.15625,
+      "learning_rate": 3.673913043478261e-05,
+      "loss": 0.6483,
+      "step": 306
+    },
+    {
+      "epoch": 2.6702937976060936,
+      "grad_norm": 1.25,
+      "learning_rate": 3.6695652173913046e-05,
+      "loss": 0.557,
+      "step": 307
+    },
+    {
+      "epoch": 2.678998911860718,
+      "grad_norm": 1.109375,
+      "learning_rate": 3.665217391304348e-05,
+      "loss": 0.5384,
+      "step": 308
+    },
+    {
+      "epoch": 2.687704026115343,
+      "grad_norm": 1.1484375,
+      "learning_rate": 3.660869565217391e-05,
+      "loss": 0.3727,
+      "step": 309
+    },
+    {
+      "epoch": 2.6964091403699673,
+      "grad_norm": 1.3515625,
+      "learning_rate": 3.656521739130435e-05,
+      "loss": 0.5383,
+      "step": 310
+    },
+    {
+      "epoch": 2.705114254624592,
+      "grad_norm": 1.328125,
+      "learning_rate": 3.6521739130434786e-05,
+      "loss": 0.6103,
+      "step": 311
+    },
+    {
+      "epoch": 2.7138193688792165,
+      "grad_norm": 1.2421875,
+      "learning_rate": 3.647826086956522e-05,
+      "loss": 0.5469,
+      "step": 312
+    },
+    {
+      "epoch": 2.7225244831338413,
+      "grad_norm": 1.1953125,
+      "learning_rate": 3.643478260869565e-05,
+      "loss": 0.5393,
+      "step": 313
+    },
+    {
+      "epoch": 2.7312295973884657,
+      "grad_norm": 1.1953125,
+      "learning_rate": 3.6391304347826086e-05,
+      "loss": 0.6248,
+      "step": 314
+    },
+    {
+      "epoch": 2.73993471164309,
+      "grad_norm": 1.2421875,
+      "learning_rate": 3.634782608695652e-05,
+      "loss": 0.5939,
+      "step": 315
+    },
+    {
+      "epoch": 2.748639825897715,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.630434782608696e-05,
+      "loss": 0.5281,
+      "step": 316
+    },
+    {
+      "epoch": 2.7573449401523398,
+      "grad_norm": 1.09375,
+      "learning_rate": 3.6260869565217394e-05,
+      "loss": 0.6094,
+      "step": 317
+    },
+    {
+      "epoch": 2.766050054406964,
+      "grad_norm": 1.625,
+      "learning_rate": 3.621739130434783e-05,
+      "loss": 0.5353,
+      "step": 318
+    },
+    {
+      "epoch": 2.7747551686615886,
+      "grad_norm": 1.296875,
+      "learning_rate": 3.617391304347826e-05,
+      "loss": 0.8688,
+      "step": 319
+    },
+    {
+      "epoch": 2.7834602829162134,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.6130434782608694e-05,
+      "loss": 0.4359,
+      "step": 320
+    },
+    {
+      "epoch": 2.792165397170838,
+      "grad_norm": 1.34375,
+      "learning_rate": 3.6086956521739134e-05,
+      "loss": 0.5274,
+      "step": 321
+    },
+    {
+      "epoch": 2.8008705114254626,
+      "grad_norm": 1.0703125,
+      "learning_rate": 3.604347826086957e-05,
+      "loss": 0.5489,
+      "step": 322
+    },
+    {
+      "epoch": 2.809575625680087,
+      "grad_norm": 1.4375,
+      "learning_rate": 3.6e-05,
+      "loss": 0.4897,
+      "step": 323
+    },
+    {
+      "epoch": 2.8182807399347114,
+      "grad_norm": 1.2109375,
+      "learning_rate": 3.595652173913044e-05,
+      "loss": 0.4663,
+      "step": 324
+    },
+    {
+      "epoch": 2.8269858541893362,
+      "grad_norm": 1.125,
+      "learning_rate": 3.5913043478260875e-05,
+      "loss": 0.6313,
+      "step": 325
+    },
+    {
+      "epoch": 2.835690968443961,
+      "grad_norm": 1.1875,
+      "learning_rate": 3.58695652173913e-05,
+      "loss": 0.8168,
+      "step": 326
+    },
+    {
+      "epoch": 2.8443960826985855,
+      "grad_norm": 1.203125,
+      "learning_rate": 3.582608695652174e-05,
+      "loss": 0.8116,
+      "step": 327
+    },
+    {
+      "epoch": 2.85310119695321,
+      "grad_norm": 1.1015625,
+      "learning_rate": 3.5782608695652175e-05,
+      "loss": 0.6349,
+      "step": 328
+    },
+    {
+      "epoch": 2.8618063112078347,
+      "grad_norm": 1.0859375,
+      "learning_rate": 3.573913043478261e-05,
+      "loss": 0.5963,
+      "step": 329
+    },
+    {
+      "epoch": 2.870511425462459,
+      "grad_norm": 1.09375,
+      "learning_rate": 3.569565217391305e-05,
+      "loss": 0.6366,
+      "step": 330
+    },
+    {
+      "epoch": 2.879216539717084,
+      "grad_norm": 1.09375,
+      "learning_rate": 3.565217391304348e-05,
+      "loss": 0.5822,
+      "step": 331
+    },
+    {
+      "epoch": 2.8879216539717083,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.5608695652173915e-05,
+      "loss": 0.4839,
+      "step": 332
+    },
+    {
+      "epoch": 2.8966267682263327,
+      "grad_norm": 1.0859375,
+      "learning_rate": 3.556521739130435e-05,
+      "loss": 0.6158,
+      "step": 333
+    },
+    {
+      "epoch": 2.9053318824809575,
+      "grad_norm": 1.2109375,
+      "learning_rate": 3.552173913043478e-05,
+      "loss": 0.4699,
+      "step": 334
+    },
+    {
+      "epoch": 2.9140369967355824,
+      "grad_norm": 1.4765625,
+      "learning_rate": 3.5478260869565216e-05,
+      "loss": 0.6381,
+      "step": 335
+    },
+    {
+      "epoch": 2.9227421109902068,
+      "grad_norm": 1.2421875,
+      "learning_rate": 3.5434782608695656e-05,
+      "loss": 0.5853,
+      "step": 336
+    },
+    {
+      "epoch": 2.931447225244831,
+      "grad_norm": 1.234375,
+      "learning_rate": 3.539130434782609e-05,
+      "loss": 0.6238,
+      "step": 337
+    },
+    {
+      "epoch": 2.940152339499456,
+      "grad_norm": 1.3046875,
+      "learning_rate": 3.534782608695652e-05,
+      "loss": 0.6795,
+      "step": 338
+    },
+    {
+      "epoch": 2.9488574537540804,
+      "grad_norm": 1.0859375,
+      "learning_rate": 3.5304347826086956e-05,
+      "loss": 0.546,
+      "step": 339
+    },
+    {
+      "epoch": 2.957562568008705,
+      "grad_norm": 1.1171875,
+      "learning_rate": 3.526086956521739e-05,
+      "loss": 0.4245,
+      "step": 340
+    },
+    {
+      "epoch": 2.9662676822633296,
+      "grad_norm": 1.265625,
+      "learning_rate": 3.521739130434783e-05,
+      "loss": 0.5433,
+      "step": 341
+    },
+    {
+      "epoch": 2.9749727965179544,
+      "grad_norm": 1.2578125,
+      "learning_rate": 3.517391304347826e-05,
+      "loss": 0.4675,
+      "step": 342
+    },
+    {
+      "epoch": 2.983677910772579,
+      "grad_norm": 1.28125,
+      "learning_rate": 3.51304347826087e-05,
+      "loss": 0.6289,
+      "step": 343
+    },
+    {
+      "epoch": 2.9923830250272037,
+      "grad_norm": 1.21875,
+      "learning_rate": 3.508695652173914e-05,
+      "loss": 0.6289,
+      "step": 344
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 1.28125,
+      "learning_rate": 3.5043478260869564e-05,
+      "loss": 0.4109,
+      "step": 345
+    },
+    {
+      "epoch": 3.0087051142546244,
+      "grad_norm": 1.421875,
+      "learning_rate": 3.5e-05,
+      "loss": 0.4183,
+      "step": 346
+    },
+    {
+      "epoch": 3.0174102285092492,
+      "grad_norm": 1.1953125,
+      "learning_rate": 3.495652173913044e-05,
+      "loss": 0.3592,
+      "step": 347
+    },
+    {
+      "epoch": 3.0261153427638736,
+      "grad_norm": 1.5234375,
+      "learning_rate": 3.491304347826087e-05,
+      "loss": 0.5384,
+      "step": 348
+    },
+    {
+      "epoch": 3.0348204570184985,
+      "grad_norm": 1.125,
+      "learning_rate": 3.4869565217391304e-05,
+      "loss": 0.5139,
+      "step": 349
+    },
+    {
+      "epoch": 3.043525571273123,
+      "grad_norm": 1.7421875,
+      "learning_rate": 3.4826086956521744e-05,
+      "loss": 0.9164,
+      "step": 350
+    },
+    {
+      "epoch": 3.0522306855277477,
+      "grad_norm": 1.8125,
+      "learning_rate": 3.478260869565218e-05,
+      "loss": 0.3615,
+      "step": 351
+    },
+    {
+      "epoch": 3.060935799782372,
+      "grad_norm": 2.046875,
+      "learning_rate": 3.473913043478261e-05,
+      "loss": 0.3857,
+      "step": 352
+    },
+    {
+      "epoch": 3.069640914036997,
+      "grad_norm": 1.84375,
+      "learning_rate": 3.4695652173913045e-05,
+      "loss": 0.4078,
+      "step": 353
+    },
+    {
+      "epoch": 3.0783460282916213,
+      "grad_norm": 1.796875,
+      "learning_rate": 3.465217391304348e-05,
+      "loss": 0.4181,
+      "step": 354
+    },
+    {
+      "epoch": 3.087051142546246,
+      "grad_norm": 1.6015625,
+      "learning_rate": 3.460869565217392e-05,
+      "loss": 0.3059,
+      "step": 355
+    },
+    {
+      "epoch": 3.0957562568008705,
+      "grad_norm": 1.390625,
+      "learning_rate": 3.456521739130435e-05,
+      "loss": 0.4606,
+      "step": 356
+    },
+    {
+      "epoch": 3.104461371055495,
+      "grad_norm": 1.28125,
+      "learning_rate": 3.4521739130434785e-05,
+      "loss": 0.4711,
+      "step": 357
+    },
+    {
+      "epoch": 3.1131664853101197,
+      "grad_norm": 1.28125,
+      "learning_rate": 3.447826086956522e-05,
+      "loss": 0.4096,
+      "step": 358
+    },
+    {
+      "epoch": 3.121871599564744,
+      "grad_norm": 1.25,
+      "learning_rate": 3.443478260869565e-05,
+      "loss": 0.6088,
+      "step": 359
+    },
+    {
+      "epoch": 3.130576713819369,
+      "grad_norm": 1.171875,
+      "learning_rate": 3.4391304347826085e-05,
+      "loss": 0.3165,
+      "step": 360
+    },
+    {
+      "epoch": 3.1392818280739934,
+      "grad_norm": 1.2421875,
+      "learning_rate": 3.4347826086956526e-05,
+      "loss": 0.306,
+      "step": 361
+    },
+    {
+      "epoch": 3.147986942328618,
+      "grad_norm": 1.2890625,
+      "learning_rate": 3.430434782608696e-05,
+      "loss": 0.3672,
+      "step": 362
+    },
+    {
+      "epoch": 3.1566920565832426,
+      "grad_norm": 1.1875,
+      "learning_rate": 3.426086956521739e-05,
+      "loss": 0.4125,
+      "step": 363
+    },
+    {
+      "epoch": 3.1653971708378674,
+      "grad_norm": 1.5,
+      "learning_rate": 3.4217391304347826e-05,
+      "loss": 0.4383,
+      "step": 364
+    },
+    {
+      "epoch": 3.174102285092492,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.417391304347826e-05,
+      "loss": 0.3679,
+      "step": 365
+    },
+    {
+      "epoch": 3.1828073993471167,
+      "grad_norm": 1.453125,
+      "learning_rate": 3.413043478260869e-05,
+      "loss": 0.4468,
+      "step": 366
+    },
+    {
+      "epoch": 3.191512513601741,
+      "grad_norm": 1.34375,
+      "learning_rate": 3.408695652173913e-05,
+      "loss": 0.3335,
+      "step": 367
+    },
+    {
+      "epoch": 3.2002176278563654,
+      "grad_norm": 1.359375,
+      "learning_rate": 3.4043478260869566e-05,
+      "loss": 0.3439,
+      "step": 368
+    },
+    {
+      "epoch": 3.2089227421109903,
+      "grad_norm": 1.4765625,
+      "learning_rate": 3.4000000000000007e-05,
+      "loss": 0.4858,
+      "step": 369
+    },
+    {
+      "epoch": 3.2176278563656147,
+      "grad_norm": 1.453125,
+      "learning_rate": 3.395652173913044e-05,
+      "loss": 0.3231,
+      "step": 370
+    },
+    {
+      "epoch": 3.2263329706202395,
+      "grad_norm": 1.6171875,
+      "learning_rate": 3.3913043478260867e-05,
+      "loss": 0.5224,
+      "step": 371
+    },
+    {
+      "epoch": 3.235038084874864,
+      "grad_norm": 1.328125,
+      "learning_rate": 3.386956521739131e-05,
+      "loss": 0.3573,
+      "step": 372
+    },
+    {
+      "epoch": 3.2437431991294887,
+      "grad_norm": 1.359375,
+      "learning_rate": 3.382608695652174e-05,
+      "loss": 0.3516,
+      "step": 373
+    },
+    {
+      "epoch": 3.252448313384113,
+      "grad_norm": 1.3515625,
+      "learning_rate": 3.3782608695652174e-05,
+      "loss": 0.3656,
+      "step": 374
+    },
+    {
+      "epoch": 3.261153427638738,
+      "grad_norm": 1.4453125,
+      "learning_rate": 3.3739130434782614e-05,
+      "loss": 0.5266,
+      "step": 375
+    },
+    {
+      "epoch": 3.2698585418933623,
+      "grad_norm": 1.3515625,
+      "learning_rate": 3.369565217391305e-05,
+      "loss": 0.3544,
+      "step": 376
+    },
+    {
+      "epoch": 3.2785636561479867,
+      "grad_norm": 1.2890625,
+      "learning_rate": 3.365217391304348e-05,
+      "loss": 0.3623,
+      "step": 377
+    },
+    {
+      "epoch": 3.2872687704026116,
+      "grad_norm": 1.3203125,
+      "learning_rate": 3.3608695652173914e-05,
+      "loss": 0.3814,
+      "step": 378
+    },
+    {
+      "epoch": 3.295973884657236,
+      "grad_norm": 1.390625,
+      "learning_rate": 3.356521739130435e-05,
+      "loss": 0.42,
+      "step": 379
+    },
+    {
+      "epoch": 3.304678998911861,
+      "grad_norm": 1.2890625,
+      "learning_rate": 3.352173913043478e-05,
+      "loss": 0.4972,
+      "step": 380
+    },
+    {
+      "epoch": 3.313384113166485,
+      "grad_norm": 1.6015625,
+      "learning_rate": 3.347826086956522e-05,
+      "loss": 0.5218,
+      "step": 381
+    },
+    {
+      "epoch": 3.32208922742111,
+      "grad_norm": 1.1953125,
+      "learning_rate": 3.3434782608695655e-05,
+      "loss": 0.3364,
+      "step": 382
+    },
+    {
+      "epoch": 3.3307943416757344,
+      "grad_norm": 1.2890625,
+      "learning_rate": 3.339130434782609e-05,
+      "loss": 0.363,
+      "step": 383
+    },
+    {
+      "epoch": 3.3394994559303592,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.334782608695652e-05,
+      "loss": 0.3453,
+      "step": 384
+    },
+    {
+      "epoch": 3.3482045701849836,
+      "grad_norm": 1.4765625,
+      "learning_rate": 3.3304347826086955e-05,
+      "loss": 0.3866,
+      "step": 385
+    },
+    {
+      "epoch": 3.356909684439608,
+      "grad_norm": 1.34375,
+      "learning_rate": 3.3260869565217395e-05,
+      "loss": 0.3148,
+      "step": 386
+    },
+    {
+      "epoch": 3.365614798694233,
+      "grad_norm": 1.2578125,
+      "learning_rate": 3.321739130434783e-05,
+      "loss": 0.6014,
+      "step": 387
+    },
+    {
+      "epoch": 3.3743199129488577,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.317391304347826e-05,
+      "loss": 0.3306,
+      "step": 388
+    },
+    {
+      "epoch": 3.383025027203482,
+      "grad_norm": 1.3125,
+      "learning_rate": 3.31304347826087e-05,
+      "loss": 0.4507,
+      "step": 389
+    },
+    {
+      "epoch": 3.3917301414581065,
+      "grad_norm": 1.3984375,
+      "learning_rate": 3.308695652173913e-05,
+      "loss": 0.4535,
+      "step": 390
+    },
+    {
+      "epoch": 3.4004352557127313,
+      "grad_norm": 1.296875,
+      "learning_rate": 3.304347826086956e-05,
+      "loss": 0.4736,
+      "step": 391
+    },
+    {
+      "epoch": 3.4091403699673557,
+      "grad_norm": 1.2890625,
+      "learning_rate": 3.3e-05,
+      "loss": 0.3137,
+      "step": 392
+    },
+    {
+      "epoch": 3.4178454842219805,
+      "grad_norm": 1.2109375,
+      "learning_rate": 3.2956521739130436e-05,
+      "loss": 0.335,
+      "step": 393
+    },
+    {
+      "epoch": 3.426550598476605,
+      "grad_norm": 1.328125,
+      "learning_rate": 3.291304347826087e-05,
+      "loss": 0.3644,
+      "step": 394
+    },
+    {
+      "epoch": 3.4352557127312298,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.286956521739131e-05,
+      "loss": 0.3236,
+      "step": 395
+    },
+    {
+      "epoch": 3.443960826985854,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.282608695652174e-05,
+      "loss": 0.6159,
+      "step": 396
+    },
+    {
+      "epoch": 3.452665941240479,
+      "grad_norm": 1.4140625,
+      "learning_rate": 3.278260869565217e-05,
+      "loss": 0.4811,
+      "step": 397
+    },
+    {
+      "epoch": 3.4613710554951034,
+      "grad_norm": 1.1640625,
+      "learning_rate": 3.273913043478261e-05,
+      "loss": 0.3592,
+      "step": 398
+    },
+    {
+      "epoch": 3.4700761697497278,
+      "grad_norm": 1.203125,
+      "learning_rate": 3.269565217391304e-05,
+      "loss": 0.31,
+      "step": 399
+    },
+    {
+      "epoch": 3.4787812840043526,
+      "grad_norm": 1.46875,
+      "learning_rate": 3.2652173913043484e-05,
+      "loss": 0.3305,
+      "step": 400
+    },
+    {
+      "epoch": 3.487486398258977,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.260869565217392e-05,
+      "loss": 0.4076,
+      "step": 401
+    },
+    {
+      "epoch": 3.496191512513602,
+      "grad_norm": 1.46875,
+      "learning_rate": 3.256521739130435e-05,
+      "loss": 0.4706,
+      "step": 402
+    },
+    {
+      "epoch": 3.504896626768226,
+      "grad_norm": 1.2578125,
+      "learning_rate": 3.2521739130434784e-05,
+      "loss": 0.2767,
+      "step": 403
+    },
+    {
+      "epoch": 3.513601741022851,
+      "grad_norm": 1.28125,
+      "learning_rate": 3.247826086956522e-05,
+      "loss": 0.3916,
+      "step": 404
+    },
+    {
+      "epoch": 3.5223068552774754,
+      "grad_norm": 1.234375,
+      "learning_rate": 3.243478260869565e-05,
+      "loss": 0.3626,
+      "step": 405
+    },
+    {
+      "epoch": 3.5310119695321003,
+      "grad_norm": 1.265625,
+      "learning_rate": 3.239130434782609e-05,
+      "loss": 0.3815,
+      "step": 406
+    },
+    {
+      "epoch": 3.5397170837867247,
+      "grad_norm": 1.1484375,
+      "learning_rate": 3.2347826086956524e-05,
+      "loss": 0.3034,
+      "step": 407
+    },
+    {
+      "epoch": 3.548422198041349,
+      "grad_norm": 1.3046875,
+      "learning_rate": 3.230434782608696e-05,
+      "loss": 0.374,
+      "step": 408
+    },
+    {
+      "epoch": 3.557127312295974,
+      "grad_norm": 1.3828125,
+      "learning_rate": 3.226086956521739e-05,
+      "loss": 0.3537,
+      "step": 409
+    },
+    {
+      "epoch": 3.5658324265505987,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.2217391304347825e-05,
+      "loss": 0.5427,
+      "step": 410
+    },
+    {
+      "epoch": 3.574537540805223,
+      "grad_norm": 1.3359375,
+      "learning_rate": 3.217391304347826e-05,
+      "loss": 0.3466,
+      "step": 411
+    },
+    {
+      "epoch": 3.5832426550598475,
+      "grad_norm": 1.4453125,
+      "learning_rate": 3.21304347826087e-05,
+      "loss": 0.5072,
+      "step": 412
+    },
+    {
+      "epoch": 3.5919477693144723,
+      "grad_norm": 1.3984375,
+      "learning_rate": 3.208695652173913e-05,
+      "loss": 0.3162,
+      "step": 413
+    },
+    {
+      "epoch": 3.6006528835690967,
+      "grad_norm": 1.1953125,
+      "learning_rate": 3.204347826086957e-05,
+      "loss": 0.51,
+      "step": 414
+    },
+    {
+      "epoch": 3.6093579978237216,
+      "grad_norm": 1.25,
+      "learning_rate": 3.2000000000000005e-05,
+      "loss": 0.4698,
+      "step": 415
+    },
+    {
+      "epoch": 3.618063112078346,
+      "grad_norm": 1.296875,
+      "learning_rate": 3.195652173913043e-05,
+      "loss": 0.3544,
+      "step": 416
+    },
+    {
+      "epoch": 3.6267682263329704,
+      "grad_norm": 1.453125,
+      "learning_rate": 3.191304347826087e-05,
+      "loss": 0.3747,
+      "step": 417
+    },
+    {
+      "epoch": 3.635473340587595,
+      "grad_norm": 1.328125,
+      "learning_rate": 3.1869565217391306e-05,
+      "loss": 0.4795,
+      "step": 418
+    },
+    {
+      "epoch": 3.64417845484222,
+      "grad_norm": 1.2734375,
+      "learning_rate": 3.182608695652174e-05,
+      "loss": 0.5804,
+      "step": 419
+    },
+    {
+      "epoch": 3.6528835690968444,
+      "grad_norm": 1.3125,
+      "learning_rate": 3.178260869565218e-05,
+      "loss": 0.4708,
+      "step": 420
+    },
+    {
+      "epoch": 3.661588683351469,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.173913043478261e-05,
+      "loss": 0.4494,
+      "step": 421
+    },
+    {
+      "epoch": 3.6702937976060936,
+      "grad_norm": 1.34375,
+      "learning_rate": 3.1695652173913046e-05,
+      "loss": 0.4712,
+      "step": 422
+    },
+    {
+      "epoch": 3.678998911860718,
+      "grad_norm": 1.6953125,
+      "learning_rate": 3.165217391304348e-05,
+      "loss": 0.5315,
+      "step": 423
+    },
+    {
+      "epoch": 3.687704026115343,
+      "grad_norm": 1.2578125,
+      "learning_rate": 3.160869565217391e-05,
+      "loss": 0.3474,
+      "step": 424
+    },
+    {
+      "epoch": 3.6964091403699673,
+      "grad_norm": 1.328125,
+      "learning_rate": 3.1565217391304346e-05,
+      "loss": 0.408,
+      "step": 425
+    },
+    {
+      "epoch": 3.705114254624592,
+      "grad_norm": 1.3828125,
+      "learning_rate": 3.152173913043479e-05,
+      "loss": 0.3594,
+      "step": 426
+    },
+    {
+      "epoch": 3.7138193688792165,
+      "grad_norm": 1.296875,
+      "learning_rate": 3.147826086956522e-05,
+      "loss": 0.3591,
+      "step": 427
+    },
+    {
+      "epoch": 3.7225244831338413,
+      "grad_norm": 1.359375,
+      "learning_rate": 3.1434782608695653e-05,
+      "loss": 0.4131,
+      "step": 428
+    },
+    {
+      "epoch": 3.7312295973884657,
+      "grad_norm": 1.2265625,
+      "learning_rate": 3.139130434782609e-05,
+      "loss": 0.59,
+      "step": 429
+    },
+    {
+      "epoch": 3.73993471164309,
+      "grad_norm": 1.515625,
+      "learning_rate": 3.134782608695652e-05,
+      "loss": 0.5576,
+      "step": 430
+    },
+    {
+      "epoch": 3.748639825897715,
+      "grad_norm": 1.3828125,
+      "learning_rate": 3.130434782608696e-05,
+      "loss": 0.333,
+      "step": 431
+    },
+    {
+      "epoch": 3.7573449401523398,
+      "grad_norm": 1.25,
+      "learning_rate": 3.1260869565217394e-05,
+      "loss": 0.3218,
+      "step": 432
+    },
+    {
+      "epoch": 3.766050054406964,
+      "grad_norm": 1.328125,
+      "learning_rate": 3.121739130434783e-05,
+      "loss": 0.4542,
+      "step": 433
+    },
+    {
+      "epoch": 3.7747551686615886,
+      "grad_norm": 1.5234375,
+      "learning_rate": 3.117391304347827e-05,
+      "loss": 0.497,
+      "step": 434
+    },
+    {
+      "epoch": 3.7834602829162134,
+      "grad_norm": 1.40625,
+      "learning_rate": 3.1130434782608694e-05,
+      "loss": 0.3721,
+      "step": 435
+    },
+    {
+      "epoch": 3.792165397170838,
+      "grad_norm": 1.34375,
+      "learning_rate": 3.108695652173913e-05,
+      "loss": 0.4833,
+      "step": 436
+    },
+    {
+      "epoch": 3.8008705114254626,
+      "grad_norm": 1.484375,
+      "learning_rate": 3.104347826086957e-05,
+      "loss": 0.3413,
+      "step": 437
+    },
+    {
+      "epoch": 3.809575625680087,
+      "grad_norm": 1.1640625,
+      "learning_rate": 3.1e-05,
+      "loss": 0.3731,
+      "step": 438
+    },
+    {
+      "epoch": 3.8182807399347114,
+      "grad_norm": 1.2734375,
+      "learning_rate": 3.0956521739130435e-05,
+      "loss": 0.3438,
+      "step": 439
+    },
+    {
+      "epoch": 3.8269858541893362,
+      "grad_norm": 1.359375,
+      "learning_rate": 3.0913043478260875e-05,
+      "loss": 0.4249,
+      "step": 440
+    },
+    {
+      "epoch": 3.835690968443961,
+      "grad_norm": 1.25,
+      "learning_rate": 3.086956521739131e-05,
+      "loss": 0.4995,
+      "step": 441
+    },
+    {
+      "epoch": 3.8443960826985855,
+      "grad_norm": 1.484375,
+      "learning_rate": 3.0826086956521735e-05,
+      "loss": 0.4477,
+      "step": 442
+    },
+    {
+      "epoch": 3.85310119695321,
+      "grad_norm": 1.25,
+      "learning_rate": 3.0782608695652175e-05,
+      "loss": 0.4823,
+      "step": 443
+    },
+    {
+      "epoch": 3.8618063112078347,
+      "grad_norm": 1.3828125,
+      "learning_rate": 3.073913043478261e-05,
+      "loss": 0.5226,
+      "step": 444
+    },
+    {
+      "epoch": 3.870511425462459,
+      "grad_norm": 1.359375,
+      "learning_rate": 3.069565217391305e-05,
+      "loss": 0.4088,
+      "step": 445
+    },
+    {
+      "epoch": 3.879216539717084,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.065217391304348e-05,
+      "loss": 0.4232,
+      "step": 446
+    },
+    {
+      "epoch": 3.8879216539717083,
+      "grad_norm": 1.4375,
+      "learning_rate": 3.0608695652173916e-05,
+      "loss": 0.3985,
+      "step": 447
+    },
+    {
+      "epoch": 3.8966267682263327,
+      "grad_norm": 1.421875,
+      "learning_rate": 3.056521739130435e-05,
+      "loss": 0.4992,
+      "step": 448
+    },
+    {
+      "epoch": 3.9053318824809575,
+      "grad_norm": 1.4296875,
+      "learning_rate": 3.052173913043478e-05,
+      "loss": 0.4681,
+      "step": 449
+    },
+    {
+      "epoch": 3.9140369967355824,
+      "grad_norm": 1.34375,
+      "learning_rate": 3.0478260869565216e-05,
+      "loss": 0.3519,
+      "step": 450
+    },
+    {
+      "epoch": 3.9227421109902068,
+      "grad_norm": 1.15625,
+      "learning_rate": 3.0434782608695656e-05,
+      "loss": 0.38,
+      "step": 451
+    },
+    {
+      "epoch": 3.931447225244831,
+      "grad_norm": 1.421875,
+      "learning_rate": 3.039130434782609e-05,
+      "loss": 0.5022,
+      "step": 452
+    },
+    {
+      "epoch": 3.940152339499456,
+      "grad_norm": 1.2890625,
+      "learning_rate": 3.034782608695652e-05,
+      "loss": 0.4629,
+      "step": 453
+    },
+    {
+      "epoch": 3.9488574537540804,
+      "grad_norm": 1.328125,
+      "learning_rate": 3.030434782608696e-05,
+      "loss": 0.7028,
+      "step": 454
+    },
+    {
+      "epoch": 3.957562568008705,
+      "grad_norm": 1.15625,
+      "learning_rate": 3.0260869565217393e-05,
+      "loss": 0.3148,
+      "step": 455
+    },
+    {
+      "epoch": 3.9662676822633296,
+      "grad_norm": 1.3046875,
+      "learning_rate": 3.0217391304347827e-05,
+      "loss": 0.5369,
+      "step": 456
+    },
+    {
+      "epoch": 3.9749727965179544,
+      "grad_norm": 1.125,
+      "learning_rate": 3.0173913043478264e-05,
+      "loss": 0.3866,
+      "step": 457
+    },
+    {
+      "epoch": 3.983677910772579,
+      "grad_norm": 1.3203125,
+      "learning_rate": 3.0130434782608697e-05,
+      "loss": 0.5832,
+      "step": 458
+    },
+    {
+      "epoch": 3.9923830250272037,
+      "grad_norm": 1.359375,
+      "learning_rate": 3.008695652173913e-05,
+      "loss": 0.4359,
+      "step": 459
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 1.7265625,
+      "learning_rate": 3.0043478260869567e-05,
+      "loss": 0.3593,
+      "step": 460
+    },
+    {
+      "epoch": 4.008705114254624,
+      "grad_norm": 1.4140625,
+      "learning_rate": 3e-05,
+      "loss": 0.3197,
+      "step": 461
+    },
+    {
+      "epoch": 4.017410228509249,
+      "grad_norm": 1.15625,
+      "learning_rate": 2.9956521739130438e-05,
+      "loss": 0.3831,
+      "step": 462
+    },
+    {
+      "epoch": 4.026115342763874,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.991304347826087e-05,
+      "loss": 0.2362,
+      "step": 463
+    },
+    {
+      "epoch": 4.0348204570184985,
+      "grad_norm": 1.328125,
+      "learning_rate": 2.9869565217391304e-05,
+      "loss": 0.2624,
+      "step": 464
+    },
+    {
+      "epoch": 4.043525571273123,
+      "grad_norm": 1.234375,
+      "learning_rate": 2.982608695652174e-05,
+      "loss": 0.2093,
+      "step": 465
+    },
+    {
+      "epoch": 4.052230685527747,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2.9782608695652175e-05,
+      "loss": 0.4162,
+      "step": 466
+    },
+    {
+      "epoch": 4.0609357997823725,
+      "grad_norm": 1.9921875,
+      "learning_rate": 2.9739130434782608e-05,
+      "loss": 0.286,
+      "step": 467
+    },
+    {
+      "epoch": 4.069640914036997,
+      "grad_norm": 2.28125,
+      "learning_rate": 2.969565217391305e-05,
+      "loss": 0.3429,
+      "step": 468
+    },
+    {
+      "epoch": 4.078346028291621,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2.965217391304348e-05,
+      "loss": 0.2248,
+      "step": 469
+    },
+    {
+      "epoch": 4.087051142546246,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2.9608695652173912e-05,
+      "loss": 0.3577,
+      "step": 470
+    },
+    {
+      "epoch": 4.09575625680087,
+      "grad_norm": 1.453125,
+      "learning_rate": 2.9565217391304352e-05,
+      "loss": 0.2521,
+      "step": 471
+    },
+    {
+      "epoch": 4.104461371055495,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2.9521739130434782e-05,
+      "loss": 0.2246,
+      "step": 472
+    },
+    {
+      "epoch": 4.11316648531012,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2.9478260869565215e-05,
+      "loss": 0.4472,
+      "step": 473
+    },
+    {
+      "epoch": 4.121871599564744,
+      "grad_norm": 1.515625,
+      "learning_rate": 2.9434782608695656e-05,
+      "loss": 0.2405,
+      "step": 474
+    },
+    {
+      "epoch": 4.1305767138193685,
+      "grad_norm": 1.171875,
+      "learning_rate": 2.939130434782609e-05,
+      "loss": 0.2111,
+      "step": 475
+    },
+    {
+      "epoch": 4.139281828073994,
+      "grad_norm": 1.171875,
+      "learning_rate": 2.9347826086956526e-05,
+      "loss": 0.3421,
+      "step": 476
+    },
+    {
+      "epoch": 4.147986942328618,
+      "grad_norm": 1.359375,
+      "learning_rate": 2.930434782608696e-05,
+      "loss": 0.343,
+      "step": 477
+    },
+    {
+      "epoch": 4.156692056583243,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.9260869565217393e-05,
+      "loss": 0.1912,
+      "step": 478
+    },
+    {
+      "epoch": 4.165397170837867,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2.921739130434783e-05,
+      "loss": 0.3702,
+      "step": 479
+    },
+    {
+      "epoch": 4.174102285092492,
+      "grad_norm": 1.828125,
+      "learning_rate": 2.9173913043478263e-05,
+      "loss": 0.2608,
+      "step": 480
+    },
+    {
+      "epoch": 4.182807399347117,
+      "grad_norm": 1.4375,
+      "learning_rate": 2.9130434782608696e-05,
+      "loss": 0.2425,
+      "step": 481
+    },
+    {
+      "epoch": 4.191512513601741,
+      "grad_norm": 1.5,
+      "learning_rate": 2.9086956521739133e-05,
+      "loss": 0.2814,
+      "step": 482
+    },
+    {
+      "epoch": 4.200217627856365,
+      "grad_norm": 1.65625,
+      "learning_rate": 2.9043478260869567e-05,
+      "loss": 0.2904,
+      "step": 483
+    },
+    {
+      "epoch": 4.20892274211099,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2.9e-05,
+      "loss": 0.275,
+      "step": 484
+    },
+    {
+      "epoch": 4.217627856365615,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.8956521739130437e-05,
+      "loss": 0.2462,
+      "step": 485
+    },
+    {
+      "epoch": 4.2263329706202395,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2.891304347826087e-05,
+      "loss": 0.262,
+      "step": 486
+    },
+    {
+      "epoch": 4.235038084874864,
+      "grad_norm": 1.25,
+      "learning_rate": 2.8869565217391304e-05,
+      "loss": 0.2443,
+      "step": 487
+    },
+    {
+      "epoch": 4.243743199129488,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.882608695652174e-05,
+      "loss": 0.3418,
+      "step": 488
+    },
+    {
+      "epoch": 4.2524483133841136,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.8782608695652174e-05,
+      "loss": 0.2845,
+      "step": 489
+    },
+    {
+      "epoch": 4.261153427638738,
+      "grad_norm": 1.671875,
+      "learning_rate": 2.8739130434782608e-05,
+      "loss": 0.3489,
+      "step": 490
+    },
+    {
+      "epoch": 4.269858541893362,
+      "grad_norm": 1.421875,
+      "learning_rate": 2.8695652173913044e-05,
+      "loss": 0.2591,
+      "step": 491
+    },
+    {
+      "epoch": 4.278563656147987,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2.8652173913043478e-05,
+      "loss": 0.5734,
+      "step": 492
+    },
+    {
+      "epoch": 4.287268770402612,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2.8608695652173918e-05,
+      "loss": 0.2109,
+      "step": 493
+    },
+    {
+      "epoch": 4.295973884657236,
+      "grad_norm": 1.5,
+      "learning_rate": 2.8565217391304348e-05,
+      "loss": 0.2421,
+      "step": 494
+    },
+    {
+      "epoch": 4.304678998911861,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2.852173913043478e-05,
+      "loss": 0.4073,
+      "step": 495
+    },
+    {
+      "epoch": 4.313384113166485,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2.847826086956522e-05,
+      "loss": 0.2709,
+      "step": 496
+    },
+    {
+      "epoch": 4.32208922742111,
+      "grad_norm": 1.453125,
+      "learning_rate": 2.8434782608695655e-05,
+      "loss": 0.2862,
+      "step": 497
+    },
+    {
+      "epoch": 4.330794341675735,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.8391304347826085e-05,
+      "loss": 0.5668,
+      "step": 498
+    },
+    {
+      "epoch": 4.339499455930359,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2.8347826086956525e-05,
+      "loss": 0.53,
+      "step": 499
+    },
+    {
+      "epoch": 4.348204570184984,
+      "grad_norm": 1.40625,
+      "learning_rate": 2.830434782608696e-05,
+      "loss": 0.2408,
+      "step": 500
+    },
+    {
+      "epoch": 4.356909684439608,
+      "grad_norm": 1.265625,
+      "learning_rate": 2.826086956521739e-05,
+      "loss": 0.2201,
+      "step": 501
+    },
+    {
+      "epoch": 4.365614798694233,
+      "grad_norm": 1.2890625,
+      "learning_rate": 2.821739130434783e-05,
+      "loss": 0.2811,
+      "step": 502
+    },
+    {
+      "epoch": 4.374319912948858,
+      "grad_norm": 1.484375,
+      "learning_rate": 2.8173913043478262e-05,
+      "loss": 0.2138,
+      "step": 503
+    },
+    {
+      "epoch": 4.383025027203482,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2.8130434782608696e-05,
+      "loss": 0.2542,
+      "step": 504
+    },
+    {
+      "epoch": 4.3917301414581065,
+      "grad_norm": 1.640625,
+      "learning_rate": 2.8086956521739133e-05,
+      "loss": 0.289,
+      "step": 505
+    },
+    {
+      "epoch": 4.400435255712731,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2.8043478260869566e-05,
+      "loss": 0.2872,
+      "step": 506
+    },
+    {
+      "epoch": 4.409140369967356,
+      "grad_norm": 1.3515625,
+      "learning_rate": 2.8000000000000003e-05,
+      "loss": 0.3546,
+      "step": 507
+    },
+    {
+      "epoch": 4.4178454842219805,
+      "grad_norm": 1.546875,
+      "learning_rate": 2.7956521739130436e-05,
+      "loss": 0.2645,
+      "step": 508
+    },
+    {
+      "epoch": 4.426550598476605,
+      "grad_norm": 1.2890625,
+      "learning_rate": 2.791304347826087e-05,
+      "loss": 0.3288,
+      "step": 509
+    },
+    {
+      "epoch": 4.435255712731229,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2.7869565217391307e-05,
+      "loss": 0.3309,
+      "step": 510
+    },
+    {
+      "epoch": 4.443960826985855,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.782608695652174e-05,
+      "loss": 0.2221,
+      "step": 511
+    },
+    {
+      "epoch": 4.452665941240479,
+      "grad_norm": 1.1796875,
+      "learning_rate": 2.7782608695652174e-05,
+      "loss": 0.3022,
+      "step": 512
+    },
+    {
+      "epoch": 4.461371055495103,
+      "grad_norm": 1.484375,
+      "learning_rate": 2.773913043478261e-05,
+      "loss": 0.2131,
+      "step": 513
+    },
+    {
+      "epoch": 4.470076169749728,
+      "grad_norm": 1.3046875,
+      "learning_rate": 2.7695652173913044e-05,
+      "loss": 0.3289,
+      "step": 514
+    },
+    {
+      "epoch": 4.478781284004352,
+      "grad_norm": 1.2890625,
+      "learning_rate": 2.7652173913043477e-05,
+      "loss": 0.1957,
+      "step": 515
+    },
+    {
+      "epoch": 4.487486398258977,
+      "grad_norm": 1.578125,
+      "learning_rate": 2.7608695652173917e-05,
+      "loss": 0.3898,
+      "step": 516
+    },
+    {
+      "epoch": 4.496191512513602,
+      "grad_norm": 1.2890625,
+      "learning_rate": 2.7565217391304347e-05,
+      "loss": 0.28,
+      "step": 517
+    },
+    {
+      "epoch": 4.504896626768226,
+      "grad_norm": 1.328125,
+      "learning_rate": 2.752173913043478e-05,
+      "loss": 0.348,
+      "step": 518
+    },
+    {
+      "epoch": 4.513601741022851,
+      "grad_norm": 1.546875,
+      "learning_rate": 2.747826086956522e-05,
+      "loss": 0.4128,
+      "step": 519
+    },
+    {
+      "epoch": 4.522306855277476,
+      "grad_norm": 1.2421875,
+      "learning_rate": 2.743478260869565e-05,
+      "loss": 0.294,
+      "step": 520
+    },
+    {
+      "epoch": 4.5310119695321,
+      "grad_norm": 1.609375,
+      "learning_rate": 2.7391304347826085e-05,
+      "loss": 0.1959,
+      "step": 521
+    },
+    {
+      "epoch": 4.539717083786725,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2.7347826086956525e-05,
+      "loss": 0.2295,
+      "step": 522
+    },
+    {
+      "epoch": 4.548422198041349,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2.7304347826086958e-05,
+      "loss": 0.5485,
+      "step": 523
+    },
+    {
+      "epoch": 4.5571273122959735,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2.7260869565217395e-05,
+      "loss": 0.2739,
+      "step": 524
+    },
+    {
+      "epoch": 4.565832426550599,
+      "grad_norm": 1.421875,
+      "learning_rate": 2.721739130434783e-05,
+      "loss": 0.3562,
+      "step": 525
+    },
+    {
+      "epoch": 4.574537540805223,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2.7173913043478262e-05,
+      "loss": 0.4189,
+      "step": 526
+    },
+    {
+      "epoch": 4.5832426550598475,
+      "grad_norm": 1.453125,
+      "learning_rate": 2.71304347826087e-05,
+      "loss": 0.4434,
+      "step": 527
+    },
+    {
+      "epoch": 4.591947769314472,
+      "grad_norm": 1.453125,
+      "learning_rate": 2.7086956521739132e-05,
+      "loss": 0.3395,
+      "step": 528
+    },
+    {
+      "epoch": 4.600652883569097,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2.7043478260869566e-05,
+      "loss": 0.3857,
+      "step": 529
+    },
+    {
+      "epoch": 4.609357997823722,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2.7000000000000002e-05,
+      "loss": 0.2454,
+      "step": 530
+    },
+    {
+      "epoch": 4.618063112078346,
+      "grad_norm": 1.546875,
+      "learning_rate": 2.6956521739130436e-05,
+      "loss": 0.2801,
+      "step": 531
+    },
+    {
+      "epoch": 4.62676822633297,
+      "grad_norm": 1.796875,
+      "learning_rate": 2.691304347826087e-05,
+      "loss": 0.4674,
+      "step": 532
+    },
+    {
+      "epoch": 4.635473340587595,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2.6869565217391306e-05,
+      "loss": 0.2464,
+      "step": 533
+    },
+    {
+      "epoch": 4.64417845484222,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.682608695652174e-05,
+      "loss": 0.2842,
+      "step": 534
+    },
+    {
+      "epoch": 4.652883569096844,
+      "grad_norm": 1.3515625,
+      "learning_rate": 2.6782608695652173e-05,
+      "loss": 0.3065,
+      "step": 535
+    },
+    {
+      "epoch": 4.661588683351469,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2.673913043478261e-05,
+      "loss": 0.2565,
+      "step": 536
+    },
+    {
+      "epoch": 4.670293797606093,
+      "grad_norm": 1.46875,
+      "learning_rate": 2.6695652173913043e-05,
+      "loss": 0.2242,
+      "step": 537
+    },
+    {
+      "epoch": 4.6789989118607185,
+      "grad_norm": 1.3125,
+      "learning_rate": 2.6652173913043483e-05,
+      "loss": 0.3663,
+      "step": 538
+    },
+    {
+      "epoch": 4.687704026115343,
+      "grad_norm": 1.46875,
+      "learning_rate": 2.6608695652173913e-05,
+      "loss": 0.3065,
+      "step": 539
+    },
+    {
+      "epoch": 4.696409140369967,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2.6565217391304347e-05,
+      "loss": 0.3606,
+      "step": 540
+    },
+    {
+      "epoch": 4.705114254624592,
+      "grad_norm": 1.65625,
+      "learning_rate": 2.6521739130434787e-05,
+      "loss": 0.4462,
+      "step": 541
+    },
+    {
+      "epoch": 4.713819368879216,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.647826086956522e-05,
+      "loss": 0.257,
+      "step": 542
+    },
+    {
+      "epoch": 4.722524483133841,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2.643478260869565e-05,
+      "loss": 0.4824,
+      "step": 543
+    },
+    {
+      "epoch": 4.731229597388466,
+      "grad_norm": 1.359375,
+      "learning_rate": 2.639130434782609e-05,
+      "loss": 0.3473,
+      "step": 544
+    },
+    {
+      "epoch": 4.73993471164309,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2.6347826086956524e-05,
+      "loss": 0.4204,
+      "step": 545
+    },
+    {
+      "epoch": 4.748639825897715,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2.6304347826086954e-05,
+      "loss": 0.3148,
+      "step": 546
+    },
+    {
+      "epoch": 4.75734494015234,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2.6260869565217394e-05,
+      "loss": 0.368,
+      "step": 547
+    },
+    {
+      "epoch": 4.766050054406964,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2.6217391304347828e-05,
+      "loss": 0.3937,
+      "step": 548
+    },
+    {
+      "epoch": 4.7747551686615886,
+      "grad_norm": 1.4375,
+      "learning_rate": 2.617391304347826e-05,
+      "loss": 0.4529,
+      "step": 549
+    },
+    {
+      "epoch": 4.783460282916213,
+      "grad_norm": 1.2890625,
+      "learning_rate": 2.6130434782608698e-05,
+      "loss": 0.2479,
+      "step": 550
+    },
+    {
+      "epoch": 4.792165397170838,
+      "grad_norm": 1.5,
+      "learning_rate": 2.608695652173913e-05,
+      "loss": 0.348,
+      "step": 551
+    },
+    {
+      "epoch": 4.800870511425463,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.6043478260869565e-05,
+      "loss": 0.2708,
+      "step": 552
+    },
+    {
+      "epoch": 4.809575625680087,
+      "grad_norm": 1.359375,
+      "learning_rate": 2.6000000000000002e-05,
+      "loss": 0.3292,
+      "step": 553
+    },
+    {
+      "epoch": 4.818280739934711,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2.5956521739130435e-05,
+      "loss": 0.2596,
+      "step": 554
+    },
+    {
+      "epoch": 4.826985854189337,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2.5913043478260872e-05,
+      "loss": 0.4254,
+      "step": 555
+    },
+    {
+      "epoch": 4.835690968443961,
+      "grad_norm": 1.1484375,
+      "learning_rate": 2.5869565217391305e-05,
+      "loss": 0.2204,
+      "step": 556
+    },
+    {
+      "epoch": 4.8443960826985855,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2.582608695652174e-05,
+      "loss": 0.3869,
+      "step": 557
+    },
+    {
+      "epoch": 4.85310119695321,
+      "grad_norm": 1.796875,
+      "learning_rate": 2.5782608695652176e-05,
+      "loss": 0.3907,
+      "step": 558
+    },
+    {
+      "epoch": 4.861806311207834,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.573913043478261e-05,
+      "loss": 0.2937,
+      "step": 559
+    },
+    {
+      "epoch": 4.8705114254624595,
+      "grad_norm": 1.796875,
+      "learning_rate": 2.5695652173913043e-05,
+      "loss": 0.4582,
+      "step": 560
+    },
+    {
+      "epoch": 4.879216539717084,
+      "grad_norm": 1.5625,
+      "learning_rate": 2.5652173913043483e-05,
+      "loss": 0.5466,
+      "step": 561
+    },
+    {
+      "epoch": 4.887921653971708,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2.5608695652173913e-05,
+      "loss": 0.2614,
+      "step": 562
+    },
+    {
+      "epoch": 4.896626768226333,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2.5565217391304346e-05,
+      "loss": 0.3748,
+      "step": 563
+    },
+    {
+      "epoch": 4.905331882480958,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2.5521739130434786e-05,
+      "loss": 0.5146,
+      "step": 564
+    },
+    {
+      "epoch": 4.914036996735582,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2.5478260869565217e-05,
+      "loss": 0.2761,
+      "step": 565
+    },
+    {
+      "epoch": 4.922742110990207,
+      "grad_norm": 1.265625,
+      "learning_rate": 2.543478260869565e-05,
+      "loss": 0.2113,
+      "step": 566
+    },
+    {
+      "epoch": 4.931447225244831,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2.539130434782609e-05,
+      "loss": 0.2445,
+      "step": 567
+    },
+    {
+      "epoch": 4.9401523394994555,
+      "grad_norm": 1.3515625,
+      "learning_rate": 2.5347826086956524e-05,
+      "loss": 0.2419,
+      "step": 568
+    },
+    {
+      "epoch": 4.948857453754081,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2.530434782608696e-05,
+      "loss": 0.205,
+      "step": 569
+    },
+    {
+      "epoch": 4.957562568008705,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2.5260869565217394e-05,
+      "loss": 0.2996,
+      "step": 570
+    },
+    {
+      "epoch": 4.96626768226333,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2.5217391304347827e-05,
+      "loss": 0.4251,
+      "step": 571
+    },
+    {
+      "epoch": 4.974972796517954,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2.5173913043478264e-05,
+      "loss": 0.3702,
+      "step": 572
+    },
+    {
+      "epoch": 4.983677910772579,
+      "grad_norm": 2.015625,
+      "learning_rate": 2.5130434782608698e-05,
+      "loss": 0.3779,
+      "step": 573
+    },
+    {
+      "epoch": 4.992383025027204,
+      "grad_norm": 1.28125,
+      "learning_rate": 2.508695652173913e-05,
+      "loss": 0.4117,
+      "step": 574
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2.5043478260869568e-05,
+      "loss": 0.3301,
+      "step": 575
+    },
+    {
+      "epoch": 5.008705114254624,
+      "grad_norm": 1.2578125,
+      "learning_rate": 2.5e-05,
+      "loss": 0.3133,
+      "step": 576
+    },
+    {
+      "epoch": 5.017410228509249,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2.4956521739130438e-05,
+      "loss": 0.3616,
+      "step": 577
+    },
+    {
+      "epoch": 5.026115342763874,
+      "grad_norm": 1.375,
+      "learning_rate": 2.4913043478260868e-05,
+      "loss": 0.2147,
+      "step": 578
+    },
+    {
+      "epoch": 5.0348204570184985,
+      "grad_norm": 1.2734375,
+      "learning_rate": 2.4869565217391305e-05,
+      "loss": 0.169,
+      "step": 579
+    },
+    {
+      "epoch": 5.043525571273123,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2.4826086956521742e-05,
+      "loss": 0.2304,
+      "step": 580
+    },
+    {
+      "epoch": 5.052230685527747,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2.4782608695652175e-05,
+      "loss": 0.1854,
+      "step": 581
+    },
+    {
+      "epoch": 5.0609357997823725,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2.473913043478261e-05,
+      "loss": 0.3501,
+      "step": 582
+    },
+    {
+      "epoch": 5.069640914036997,
+      "grad_norm": 1.671875,
+      "learning_rate": 2.4695652173913045e-05,
+      "loss": 0.2407,
+      "step": 583
+    },
+    {
+      "epoch": 5.078346028291621,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2.465217391304348e-05,
+      "loss": 0.147,
+      "step": 584
+    },
+    {
+      "epoch": 5.087051142546246,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2.4608695652173912e-05,
+      "loss": 0.193,
+      "step": 585
+    },
+    {
+      "epoch": 5.09575625680087,
+      "grad_norm": 1.765625,
+      "learning_rate": 2.456521739130435e-05,
+      "loss": 0.1877,
+      "step": 586
+    },
+    {
+      "epoch": 5.104461371055495,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2.4521739130434786e-05,
+      "loss": 0.2971,
+      "step": 587
+    },
+    {
+      "epoch": 5.11316648531012,
+      "grad_norm": 1.453125,
+      "learning_rate": 2.447826086956522e-05,
+      "loss": 0.2993,
+      "step": 588
+    },
+    {
+      "epoch": 5.121871599564744,
+      "grad_norm": 1.515625,
+      "learning_rate": 2.4434782608695653e-05,
+      "loss": 0.163,
+      "step": 589
+    },
+    {
+      "epoch": 5.1305767138193685,
+      "grad_norm": 1.359375,
+      "learning_rate": 2.439130434782609e-05,
+      "loss": 0.2468,
+      "step": 590
+    },
+    {
+      "epoch": 5.139281828073994,
+      "grad_norm": 1.234375,
+      "learning_rate": 2.4347826086956523e-05,
+      "loss": 0.4714,
+      "step": 591
+    },
+    {
+      "epoch": 5.147986942328618,
+      "grad_norm": 1.2265625,
+      "learning_rate": 2.4304347826086956e-05,
+      "loss": 0.1436,
+      "step": 592
+    },
+    {
+      "epoch": 5.156692056583243,
+      "grad_norm": 1.171875,
+      "learning_rate": 2.4260869565217393e-05,
+      "loss": 0.1693,
+      "step": 593
+    },
+    {
+      "epoch": 5.165397170837867,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2.4217391304347827e-05,
+      "loss": 0.2572,
+      "step": 594
+    },
+    {
+      "epoch": 5.174102285092492,
+      "grad_norm": 2.421875,
+      "learning_rate": 2.417391304347826e-05,
+      "loss": 0.2728,
+      "step": 595
+    },
+    {
+      "epoch": 5.182807399347117,
+      "grad_norm": 1.328125,
+      "learning_rate": 2.4130434782608697e-05,
+      "loss": 0.1908,
+      "step": 596
+    },
+    {
+      "epoch": 5.191512513601741,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2.408695652173913e-05,
+      "loss": 0.2543,
+      "step": 597
+    },
+    {
+      "epoch": 5.200217627856365,
+      "grad_norm": 1.2890625,
+      "learning_rate": 2.4043478260869567e-05,
+      "loss": 0.1825,
+      "step": 598
+    },
+    {
+      "epoch": 5.20892274211099,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2.4e-05,
+      "loss": 0.2256,
+      "step": 599
+    },
+    {
+      "epoch": 5.217627856365615,
+      "grad_norm": 1.375,
+      "learning_rate": 2.3956521739130437e-05,
+      "loss": 0.2383,
+      "step": 600
+    },
+    {
+      "epoch": 5.2263329706202395,
+      "grad_norm": 1.484375,
+      "learning_rate": 2.391304347826087e-05,
+      "loss": 0.2923,
+      "step": 601
+    },
+    {
+      "epoch": 5.235038084874864,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2.3869565217391304e-05,
+      "loss": 0.3034,
+      "step": 602
+    },
+    {
+      "epoch": 5.243743199129488,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2.382608695652174e-05,
+      "loss": 0.1643,
+      "step": 603
+    },
+    {
+      "epoch": 5.2524483133841136,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2.3782608695652175e-05,
+      "loss": 0.2489,
+      "step": 604
+    },
+    {
+      "epoch": 5.261153427638738,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2.373913043478261e-05,
+      "loss": 0.3017,
+      "step": 605
+    },
+    {
+      "epoch": 5.269858541893362,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.3695652173913045e-05,
+      "loss": 0.1751,
+      "step": 606
+    },
+    {
+      "epoch": 5.278563656147987,
+      "grad_norm": 1.421875,
+      "learning_rate": 2.3652173913043478e-05,
+      "loss": 0.1485,
+      "step": 607
+    },
+    {
+      "epoch": 5.287268770402612,
+      "grad_norm": 1.578125,
+      "learning_rate": 2.3608695652173915e-05,
+      "loss": 0.1679,
+      "step": 608
+    },
+    {
+      "epoch": 5.295973884657236,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2.356521739130435e-05,
+      "loss": 0.162,
+      "step": 609
+    },
+    {
+      "epoch": 5.304678998911861,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2.3521739130434782e-05,
+      "loss": 0.1554,
+      "step": 610
+    },
+    {
+      "epoch": 5.313384113166485,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2.347826086956522e-05,
+      "loss": 0.2891,
+      "step": 611
+    },
+    {
+      "epoch": 5.32208922742111,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2.3434782608695656e-05,
+      "loss": 0.3698,
+      "step": 612
+    },
+    {
+      "epoch": 5.330794341675735,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.339130434782609e-05,
+      "loss": 0.1886,
+      "step": 613
+    },
+    {
+      "epoch": 5.339499455930359,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2.3347826086956522e-05,
+      "loss": 0.1689,
+      "step": 614
+    },
+    {
+      "epoch": 5.348204570184984,
+      "grad_norm": 1.734375,
+      "learning_rate": 2.330434782608696e-05,
+      "loss": 0.2938,
+      "step": 615
+    },
+    {
+      "epoch": 5.356909684439608,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2.3260869565217393e-05,
+      "loss": 0.2641,
+      "step": 616
+    },
+    {
+      "epoch": 5.365614798694233,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.3217391304347826e-05,
+      "loss": 0.1735,
+      "step": 617
+    },
+    {
+      "epoch": 5.374319912948858,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2.3173913043478263e-05,
+      "loss": 0.3248,
+      "step": 618
+    },
+    {
+      "epoch": 5.383025027203482,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2.31304347826087e-05,
+      "loss": 0.1592,
+      "step": 619
+    },
+    {
+      "epoch": 5.3917301414581065,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2.308695652173913e-05,
+      "loss": 0.1805,
+      "step": 620
+    },
+    {
+      "epoch": 5.400435255712731,
+      "grad_norm": 1.53125,
+      "learning_rate": 2.3043478260869567e-05,
+      "loss": 0.2743,
+      "step": 621
+    },
+    {
+      "epoch": 5.409140369967356,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2.3000000000000003e-05,
+      "loss": 0.2918,
+      "step": 622
+    },
+    {
+      "epoch": 5.4178454842219805,
+      "grad_norm": 2.28125,
+      "learning_rate": 2.2956521739130433e-05,
+      "loss": 0.3951,
+      "step": 623
+    },
+    {
+      "epoch": 5.426550598476605,
+      "grad_norm": 1.375,
+      "learning_rate": 2.291304347826087e-05,
+      "loss": 0.2319,
+      "step": 624
+    },
+    {
+      "epoch": 5.435255712731229,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2.2869565217391307e-05,
+      "loss": 0.2565,
+      "step": 625
+    },
+    {
+      "epoch": 5.443960826985855,
+      "grad_norm": 1.609375,
+      "learning_rate": 2.282608695652174e-05,
+      "loss": 0.1877,
+      "step": 626
+    },
+    {
+      "epoch": 5.452665941240479,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2.2782608695652174e-05,
+      "loss": 0.1451,
+      "step": 627
+    },
+    {
+      "epoch": 5.461371055495103,
+      "grad_norm": 1.53125,
+      "learning_rate": 2.273913043478261e-05,
+      "loss": 0.2011,
+      "step": 628
+    },
+    {
+      "epoch": 5.470076169749728,
+      "grad_norm": 1.40625,
+      "learning_rate": 2.2695652173913044e-05,
+      "loss": 0.1698,
+      "step": 629
+    },
+    {
+      "epoch": 5.478781284004352,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2.2652173913043478e-05,
+      "loss": 0.1567,
+      "step": 630
+    },
+    {
+      "epoch": 5.487486398258977,
+      "grad_norm": 1.296875,
+      "learning_rate": 2.2608695652173914e-05,
+      "loss": 0.1489,
+      "step": 631
+    },
+    {
+      "epoch": 5.496191512513602,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2.256521739130435e-05,
+      "loss": 0.1715,
+      "step": 632
+    },
+    {
+      "epoch": 5.504896626768226,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2.252173913043478e-05,
+      "loss": 0.2442,
+      "step": 633
+    },
+    {
+      "epoch": 5.513601741022851,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2.2478260869565218e-05,
+      "loss": 0.5451,
+      "step": 634
+    },
+    {
+      "epoch": 5.522306855277476,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2.2434782608695655e-05,
+      "loss": 0.312,
+      "step": 635
+    },
+    {
+      "epoch": 5.5310119695321,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2.239130434782609e-05,
+      "loss": 0.4152,
+      "step": 636
+    },
+    {
+      "epoch": 5.539717083786725,
+      "grad_norm": 1.375,
+      "learning_rate": 2.2347826086956522e-05,
+      "loss": 0.5363,
+      "step": 637
+    },
+    {
+      "epoch": 5.548422198041349,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2.230434782608696e-05,
+      "loss": 0.1844,
+      "step": 638
+    },
+    {
+      "epoch": 5.5571273122959735,
+      "grad_norm": 1.9140625,
+      "learning_rate": 2.2260869565217392e-05,
+      "loss": 0.1978,
+      "step": 639
+    },
+    {
+      "epoch": 5.565832426550599,
+      "grad_norm": 1.703125,
+      "learning_rate": 2.2217391304347825e-05,
+      "loss": 0.2888,
+      "step": 640
+    },
+    {
+      "epoch": 5.574537540805223,
+      "grad_norm": 1.484375,
+      "learning_rate": 2.2173913043478262e-05,
+      "loss": 0.149,
+      "step": 641
+    },
+    {
+      "epoch": 5.5832426550598475,
+      "grad_norm": 1.53125,
+      "learning_rate": 2.2130434782608696e-05,
+      "loss": 0.3461,
+      "step": 642
+    },
+    {
+      "epoch": 5.591947769314472,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2.2086956521739133e-05,
+      "loss": 0.4538,
+      "step": 643
+    },
+    {
+      "epoch": 5.600652883569097,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2.2043478260869566e-05,
+      "loss": 0.1924,
+      "step": 644
+    },
+    {
+      "epoch": 5.609357997823722,
+      "grad_norm": 1.890625,
+      "learning_rate": 2.2000000000000003e-05,
+      "loss": 0.2456,
+      "step": 645
+    },
+    {
+      "epoch": 5.618063112078346,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2.1956521739130436e-05,
+      "loss": 0.1566,
+      "step": 646
+    },
+    {
+      "epoch": 5.62676822633297,
+      "grad_norm": 1.390625,
+      "learning_rate": 2.191304347826087e-05,
+      "loss": 0.2101,
+      "step": 647
+    },
+    {
+      "epoch": 5.635473340587595,
+      "grad_norm": 1.421875,
+      "learning_rate": 2.1869565217391306e-05,
+      "loss": 0.266,
+      "step": 648
+    },
+    {
+      "epoch": 5.64417845484222,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2.182608695652174e-05,
+      "loss": 0.4792,
+      "step": 649
+    },
+    {
+      "epoch": 5.652883569096844,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2.1782608695652177e-05,
+      "loss": 0.1806,
+      "step": 650
+    },
+    {
+      "epoch": 5.661588683351469,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2.173913043478261e-05,
+      "loss": 0.2714,
+      "step": 651
+    },
+    {
+      "epoch": 5.670293797606093,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2.1695652173913044e-05,
+      "loss": 0.2635,
+      "step": 652
+    },
+    {
+      "epoch": 5.6789989118607185,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2.165217391304348e-05,
+      "loss": 0.1836,
+      "step": 653
+    },
+    {
+      "epoch": 5.687704026115343,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2.1608695652173914e-05,
+      "loss": 0.4014,
+      "step": 654
+    },
+    {
+      "epoch": 5.696409140369967,
+      "grad_norm": 1.171875,
+      "learning_rate": 2.1565217391304347e-05,
+      "loss": 0.1997,
+      "step": 655
+    },
+    {
+      "epoch": 5.705114254624592,
+      "grad_norm": 1.34375,
+      "learning_rate": 2.1521739130434784e-05,
+      "loss": 0.1798,
+      "step": 656
+    },
+    {
+      "epoch": 5.713819368879216,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.1478260869565218e-05,
+      "loss": 0.1653,
+      "step": 657
+    },
+    {
+      "epoch": 5.722524483133841,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2.1434782608695654e-05,
+      "loss": 0.1735,
+      "step": 658
+    },
+    {
+      "epoch": 5.731229597388466,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2.1391304347826088e-05,
+      "loss": 0.1909,
+      "step": 659
+    },
+    {
+      "epoch": 5.73993471164309,
+      "grad_norm": 1.359375,
+      "learning_rate": 2.1347826086956525e-05,
+      "loss": 0.239,
+      "step": 660
+    },
+    {
+      "epoch": 5.748639825897715,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2.1304347826086958e-05,
+      "loss": 0.1403,
+      "step": 661
+    },
+    {
+      "epoch": 5.75734494015234,
+      "grad_norm": 1.234375,
+      "learning_rate": 2.126086956521739e-05,
+      "loss": 0.2705,
+      "step": 662
+    },
+    {
+      "epoch": 5.766050054406964,
+      "grad_norm": 1.3046875,
+      "learning_rate": 2.1217391304347828e-05,
+      "loss": 0.3939,
+      "step": 663
+    },
+    {
+      "epoch": 5.7747551686615886,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2.1173913043478262e-05,
+      "loss": 0.2337,
+      "step": 664
+    },
+    {
+      "epoch": 5.783460282916213,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.1130434782608695e-05,
+      "loss": 0.1415,
+      "step": 665
+    },
+    {
+      "epoch": 5.792165397170838,
+      "grad_norm": 1.578125,
+      "learning_rate": 2.1086956521739132e-05,
+      "loss": 0.3502,
+      "step": 666
+    },
+    {
+      "epoch": 5.800870511425463,
+      "grad_norm": 1.4375,
+      "learning_rate": 2.104347826086957e-05,
+      "loss": 0.2582,
+      "step": 667
+    },
+    {
+      "epoch": 5.809575625680087,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2.1e-05,
+      "loss": 0.2606,
+      "step": 668
+    },
+    {
+      "epoch": 5.818280739934711,
+      "grad_norm": 1.453125,
+      "learning_rate": 2.0956521739130436e-05,
+      "loss": 0.2181,
+      "step": 669
+    },
+    {
+      "epoch": 5.826985854189337,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2.0913043478260872e-05,
+      "loss": 0.2898,
+      "step": 670
+    },
+    {
+      "epoch": 5.835690968443961,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2.0869565217391303e-05,
+      "loss": 0.313,
+      "step": 671
+    },
+    {
+      "epoch": 5.8443960826985855,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2.082608695652174e-05,
+      "loss": 0.2565,
+      "step": 672
+    },
+    {
+      "epoch": 5.85310119695321,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2.0782608695652176e-05,
+      "loss": 0.1933,
+      "step": 673
+    },
+    {
+      "epoch": 5.861806311207834,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2.073913043478261e-05,
+      "loss": 0.1794,
+      "step": 674
+    },
+    {
+      "epoch": 5.8705114254624595,
+      "grad_norm": 1.625,
+      "learning_rate": 2.0695652173913043e-05,
+      "loss": 0.3617,
+      "step": 675
+    },
+    {
+      "epoch": 5.879216539717084,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.065217391304348e-05,
+      "loss": 0.258,
+      "step": 676
+    },
+    {
+      "epoch": 5.887921653971708,
+      "grad_norm": 1.28125,
+      "learning_rate": 2.0608695652173913e-05,
+      "loss": 0.2287,
+      "step": 677
+    },
+    {
+      "epoch": 5.896626768226333,
+      "grad_norm": 1.34375,
+      "learning_rate": 2.0565217391304347e-05,
+      "loss": 0.2913,
+      "step": 678
+    },
+    {
+      "epoch": 5.905331882480958,
+      "grad_norm": 1.796875,
+      "learning_rate": 2.0521739130434784e-05,
+      "loss": 0.4219,
+      "step": 679
+    },
+    {
+      "epoch": 5.914036996735582,
+      "grad_norm": 1.421875,
+      "learning_rate": 2.047826086956522e-05,
+      "loss": 0.2396,
+      "step": 680
+    },
+    {
+      "epoch": 5.922742110990207,
+      "grad_norm": 1.3046875,
+      "learning_rate": 2.0434782608695654e-05,
+      "loss": 0.152,
+      "step": 681
+    },
+    {
+      "epoch": 5.931447225244831,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.0391304347826087e-05,
+      "loss": 0.2087,
+      "step": 682
+    },
+    {
+      "epoch": 5.9401523394994555,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2.0347826086956524e-05,
+      "loss": 0.1789,
+      "step": 683
+    },
+    {
+      "epoch": 5.948857453754081,
+      "grad_norm": 1.2421875,
+      "learning_rate": 2.0304347826086957e-05,
+      "loss": 0.1685,
+      "step": 684
+    },
+    {
+      "epoch": 5.957562568008705,
+      "grad_norm": 1.515625,
+      "learning_rate": 2.026086956521739e-05,
+      "loss": 0.2052,
+      "step": 685
+    },
+    {
+      "epoch": 5.96626768226333,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2.0217391304347828e-05,
+      "loss": 0.1638,
+      "step": 686
+    },
+    {
+      "epoch": 5.974972796517954,
+      "grad_norm": 1.4375,
+      "learning_rate": 2.017391304347826e-05,
+      "loss": 0.1529,
+      "step": 687
+    },
+    {
+      "epoch": 5.983677910772579,
+      "grad_norm": 1.671875,
+      "learning_rate": 2.0130434782608695e-05,
+      "loss": 0.2103,
+      "step": 688
+    },
+    {
+      "epoch": 5.992383025027204,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2.008695652173913e-05,
+      "loss": 0.2846,
+      "step": 689
+    },
+    {
+      "epoch": 6.0,
+      "grad_norm": 1.53125,
+      "learning_rate": 2.0043478260869565e-05,
+      "loss": 0.1696,
+      "step": 690
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 1150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "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": 3.551339206661898e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}