diff --git "a/last-checkpoint/trainer_state.json" "b/last-checkpoint/trainer_state.json"
--- "a/last-checkpoint/trainer_state.json"
+++ "b/last-checkpoint/trainer_state.json"
@@ -2,6384 +2,720 @@
   "best_global_step": null,
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 0.9345794392523364,
+  "epoch": 0.10384215991692627,
   "eval_steps": 100,
-  "global_step": 900,
+  "global_step": 100,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.0010384215991692627,
-      "grad_norm": 1.6015625,
+      "grad_norm": 1.7265625,
       "learning_rate": 0.0,
-      "loss": 10.8837,
+      "loss": 10.8665,
       "step": 1
     },
     {
       "epoch": 0.0020768431983385254,
-      "grad_norm": 1.4765625,
+      "grad_norm": 1.78125,
       "learning_rate": 2.0408163265306123e-05,
-      "loss": 10.8726,
+      "loss": 10.8771,
       "step": 2
     },
     {
       "epoch": 0.003115264797507788,
-      "grad_norm": 1.515625,
+      "grad_norm": 1.71875,
       "learning_rate": 4.0816326530612245e-05,
-      "loss": 10.8856,
+      "loss": 10.8809,
       "step": 3
     },
     {
       "epoch": 0.004153686396677051,
-      "grad_norm": 1.625,
+      "grad_norm": 1.6953125,
       "learning_rate": 6.122448979591836e-05,
-      "loss": 10.8768,
+      "loss": 10.881,
       "step": 4
     },
     {
       "epoch": 0.005192107995846314,
-      "grad_norm": 2.65625,
+      "grad_norm": 1.7109375,
       "learning_rate": 8.163265306122449e-05,
-      "loss": 10.8674,
+      "loss": 10.8689,
       "step": 5
     },
     {
       "epoch": 0.006230529595015576,
-      "grad_norm": 1.46875,
+      "grad_norm": 1.6015625,
       "learning_rate": 0.00010204081632653062,
-      "loss": 10.8608,
+      "loss": 10.8573,
       "step": 6
     },
     {
       "epoch": 0.007268951194184839,
-      "grad_norm": 2.046875,
+      "grad_norm": 1.5390625,
       "learning_rate": 0.00012244897959183673,
-      "loss": 10.8457,
+      "loss": 10.8544,
       "step": 7
     },
     {
       "epoch": 0.008307372793354102,
-      "grad_norm": 1.6171875,
+      "grad_norm": 1.5234375,
       "learning_rate": 0.00014285714285714284,
-      "loss": 10.8527,
+      "loss": 10.828,
       "step": 8
     },
     {
       "epoch": 0.009345794392523364,
-      "grad_norm": 1.7578125,
+      "grad_norm": 1.7421875,
       "learning_rate": 0.00016326530612244898,
-      "loss": 10.8386,
+      "loss": 10.8025,
       "step": 9
     },
     {
       "epoch": 0.010384215991692628,
-      "grad_norm": 1.875,
+      "grad_norm": 1.6484375,
       "learning_rate": 0.00018367346938775512,
-      "loss": 10.8283,
+      "loss": 10.7755,
       "step": 10
     },
     {
       "epoch": 0.01142263759086189,
-      "grad_norm": 1.8515625,
+      "grad_norm": 1.890625,
       "learning_rate": 0.00020408163265306123,
-      "loss": 10.7993,
+      "loss": 10.762,
       "step": 11
     },
     {
       "epoch": 0.012461059190031152,
-      "grad_norm": 1.7890625,
+      "grad_norm": 1.859375,
       "learning_rate": 0.00022448979591836734,
-      "loss": 10.7665,
+      "loss": 10.7136,
       "step": 12
     },
     {
       "epoch": 0.013499480789200415,
-      "grad_norm": 9.5,
+      "grad_norm": 1.921875,
       "learning_rate": 0.00024489795918367346,
-      "loss": 10.815,
+      "loss": 10.652,
       "step": 13
     },
     {
       "epoch": 0.014537902388369679,
-      "grad_norm": 1.984375,
+      "grad_norm": 2.125,
       "learning_rate": 0.0002653061224489796,
-      "loss": 10.7245,
+      "loss": 10.5645,
       "step": 14
     },
     {
       "epoch": 0.01557632398753894,
-      "grad_norm": 2.25,
+      "grad_norm": 2.296875,
       "learning_rate": 0.0002857142857142857,
-      "loss": 10.6816,
+      "loss": 10.4862,
       "step": 15
     },
     {
       "epoch": 0.016614745586708203,
-      "grad_norm": 2.3125,
+      "grad_norm": 2.390625,
       "learning_rate": 0.0003061224489795919,
-      "loss": 10.6417,
+      "loss": 10.4193,
       "step": 16
     },
     {
       "epoch": 0.017653167185877467,
-      "grad_norm": 2.6875,
+      "grad_norm": 2.5,
       "learning_rate": 0.00032653061224489796,
-      "loss": 10.5831,
+      "loss": 10.2264,
       "step": 17
     },
     {
       "epoch": 0.018691588785046728,
-      "grad_norm": 3.109375,
+      "grad_norm": 2.546875,
       "learning_rate": 0.0003469387755102041,
-      "loss": 10.549,
+      "loss": 10.1162,
       "step": 18
     },
     {
       "epoch": 0.01973001038421599,
-      "grad_norm": 3.21875,
+      "grad_norm": 2.734375,
       "learning_rate": 0.00036734693877551024,
-      "loss": 10.4233,
+      "loss": 9.9658,
       "step": 19
     },
     {
       "epoch": 0.020768431983385256,
-      "grad_norm": 3.578125,
+      "grad_norm": 2.40625,
       "learning_rate": 0.0003877551020408163,
-      "loss": 10.3826,
+      "loss": 9.7941,
       "step": 20
     },
     {
       "epoch": 0.021806853582554516,
-      "grad_norm": 3.59375,
+      "grad_norm": 2.34375,
       "learning_rate": 0.00040816326530612246,
-      "loss": 10.3051,
+      "loss": 9.5882,
       "step": 21
     },
     {
       "epoch": 0.02284527518172378,
-      "grad_norm": 4.65625,
+      "grad_norm": 2.28125,
       "learning_rate": 0.00042857142857142855,
-      "loss": 10.1497,
+      "loss": 9.4191,
       "step": 22
     },
     {
       "epoch": 0.023883696780893044,
-      "grad_norm": 4.40625,
+      "grad_norm": 2.078125,
       "learning_rate": 0.0004489795918367347,
-      "loss": 10.043,
+      "loss": 9.2386,
       "step": 23
     },
     {
       "epoch": 0.024922118380062305,
-      "grad_norm": 4.78125,
+      "grad_norm": 1.6328125,
       "learning_rate": 0.00046938775510204083,
-      "loss": 9.9011,
+      "loss": 9.1101,
       "step": 24
     },
     {
       "epoch": 0.02596053997923157,
-      "grad_norm": 4.5,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.0004897959183673469,
-      "loss": 9.803,
+      "loss": 8.9569,
       "step": 25
     },
     {
       "epoch": 0.02699896157840083,
-      "grad_norm": 4.5625,
+      "grad_norm": 1.3671875,
       "learning_rate": 0.0005102040816326531,
-      "loss": 9.6678,
+      "loss": 8.8046,
       "step": 26
     },
     {
       "epoch": 0.028037383177570093,
-      "grad_norm": 4.8125,
+      "grad_norm": 0.98046875,
       "learning_rate": 0.0005306122448979592,
-      "loss": 9.5148,
+      "loss": 8.6394,
       "step": 27
     },
     {
       "epoch": 0.029075804776739357,
-      "grad_norm": 4.71875,
+      "grad_norm": 0.7578125,
       "learning_rate": 0.0005510204081632653,
-      "loss": 9.3753,
+      "loss": 8.5832,
       "step": 28
     },
     {
       "epoch": 0.030114226375908618,
-      "grad_norm": 4.875,
+      "grad_norm": 0.79296875,
       "learning_rate": 0.0005714285714285714,
-      "loss": 9.2481,
+      "loss": 8.4092,
       "step": 29
     },
     {
       "epoch": 0.03115264797507788,
-      "grad_norm": 4.53125,
+      "grad_norm": 0.66015625,
       "learning_rate": 0.0005918367346938776,
-      "loss": 9.1792,
+      "loss": 8.364,
       "step": 30
     },
     {
       "epoch": 0.032191069574247146,
-      "grad_norm": 4.40625,
+      "grad_norm": 0.875,
       "learning_rate": 0.0006122448979591838,
-      "loss": 9.1323,
+      "loss": 8.2433,
       "step": 31
     },
     {
       "epoch": 0.033229491173416406,
-      "grad_norm": 4.15625,
+      "grad_norm": 8.875,
       "learning_rate": 0.0006326530612244898,
-      "loss": 8.8994,
+      "loss": 8.5265,
       "step": 32
     },
     {
       "epoch": 0.03426791277258567,
-      "grad_norm": 9.4375,
+      "grad_norm": 0.73046875,
       "learning_rate": 0.0006530612244897959,
-      "loss": 8.7817,
+      "loss": 8.2092,
       "step": 33
     },
     {
       "epoch": 0.035306334371754934,
-      "grad_norm": 3.546875,
+      "grad_norm": 0.70703125,
       "learning_rate": 0.000673469387755102,
-      "loss": 8.6807,
+      "loss": 8.2644,
       "step": 34
     },
     {
       "epoch": 0.036344755970924195,
-      "grad_norm": 3.65625,
+      "grad_norm": 0.671875,
       "learning_rate": 0.0006938775510204082,
-      "loss": 8.5546,
+      "loss": 8.3096,
       "step": 35
     },
     {
       "epoch": 0.037383177570093455,
-      "grad_norm": 3.0,
+      "grad_norm": 0.71484375,
       "learning_rate": 0.0007142857142857143,
-      "loss": 8.5261,
+      "loss": 8.2509,
       "step": 36
     },
     {
       "epoch": 0.03842159916926272,
-      "grad_norm": 2.453125,
+      "grad_norm": 0.7109375,
       "learning_rate": 0.0007346938775510205,
-      "loss": 8.4449,
+      "loss": 8.3452,
       "step": 37
     },
     {
       "epoch": 0.03946002076843198,
-      "grad_norm": 1.8984375,
+      "grad_norm": 0.5625,
       "learning_rate": 0.0007551020408163265,
-      "loss": 8.3237,
+      "loss": 8.1782,
       "step": 38
     },
     {
       "epoch": 0.040498442367601244,
-      "grad_norm": 1.7578125,
+      "grad_norm": 0.5390625,
       "learning_rate": 0.0007755102040816326,
-      "loss": 8.3935,
+      "loss": 8.2422,
       "step": 39
     },
     {
       "epoch": 0.04153686396677051,
-      "grad_norm": 2.28125,
+      "grad_norm": 0.6640625,
       "learning_rate": 0.0007959183673469387,
-      "loss": 8.2964,
+      "loss": 8.2454,
       "step": 40
     },
     {
       "epoch": 0.04257528556593977,
-      "grad_norm": 2.21875,
+      "grad_norm": 0.6328125,
       "learning_rate": 0.0008163265306122449,
-      "loss": 8.4733,
+      "loss": 8.1751,
       "step": 41
     },
     {
       "epoch": 0.04361370716510903,
-      "grad_norm": 2.265625,
+      "grad_norm": 0.4609375,
       "learning_rate": 0.0008367346938775511,
-      "loss": 8.3357,
+      "loss": 8.1872,
       "step": 42
     },
     {
       "epoch": 0.0446521287642783,
-      "grad_norm": 2.203125,
+      "grad_norm": 6.03125,
       "learning_rate": 0.0008571428571428571,
-      "loss": 8.2853,
+      "loss": 8.0279,
       "step": 43
     },
     {
       "epoch": 0.04569055036344756,
-      "grad_norm": 2.15625,
+      "grad_norm": 0.57421875,
       "learning_rate": 0.0008775510204081633,
-      "loss": 8.4411,
+      "loss": 8.1868,
       "step": 44
     },
     {
       "epoch": 0.04672897196261682,
-      "grad_norm": 2.0625,
+      "grad_norm": 1.640625,
       "learning_rate": 0.0008979591836734694,
-      "loss": 8.3876,
+      "loss": 8.1969,
       "step": 45
     },
     {
       "epoch": 0.04776739356178609,
-      "grad_norm": 2.078125,
+      "grad_norm": 1.25,
       "learning_rate": 0.0009183673469387756,
-      "loss": 8.5309,
+      "loss": 8.1478,
       "step": 46
     },
     {
       "epoch": 0.04880581516095535,
-      "grad_norm": 1.8828125,
+      "grad_norm": 0.8984375,
       "learning_rate": 0.0009387755102040817,
-      "loss": 8.3942,
+      "loss": 8.1154,
       "step": 47
     },
     {
       "epoch": 0.04984423676012461,
-      "grad_norm": 1.703125,
+      "grad_norm": 1.109375,
       "learning_rate": 0.0009591836734693877,
-      "loss": 8.3333,
+      "loss": 8.1156,
       "step": 48
     },
     {
       "epoch": 0.05088265835929388,
-      "grad_norm": 1.6171875,
+      "grad_norm": 1.328125,
       "learning_rate": 0.0009795918367346938,
-      "loss": 8.3561,
+      "loss": 8.0631,
       "step": 49
     },
     {
       "epoch": 0.05192107995846314,
-      "grad_norm": 1.5234375,
+      "grad_norm": 0.82421875,
       "learning_rate": 0.001,
-      "loss": 8.3608,
+      "loss": 8.0943,
       "step": 50
     },
     {
       "epoch": 0.0529595015576324,
-      "grad_norm": 1.3125,
+      "grad_norm": 0.9140625,
       "learning_rate": 0.0009999970464322657,
-      "loss": 8.3252,
+      "loss": 7.9754,
       "step": 51
     },
     {
       "epoch": 0.05399792315680166,
-      "grad_norm": 1.1171875,
+      "grad_norm": 0.6953125,
       "learning_rate": 0.0009999881857639566,
-      "loss": 8.2845,
+      "loss": 8.0153,
       "step": 52
     },
     {
       "epoch": 0.055036344755970926,
-      "grad_norm": 1.1484375,
+      "grad_norm": 0.53125,
       "learning_rate": 0.0009999734180997555,
-      "loss": 8.3072,
+      "loss": 7.9965,
       "step": 53
     },
     {
       "epoch": 0.056074766355140186,
-      "grad_norm": 1.203125,
+      "grad_norm": 0.66796875,
       "learning_rate": 0.0009999527436141311,
-      "loss": 8.2587,
+      "loss": 7.911,
       "step": 54
     },
     {
       "epoch": 0.05711318795430945,
-      "grad_norm": 1.3125,
+      "grad_norm": 0.82421875,
       "learning_rate": 0.0009999261625513378,
-      "loss": 8.3019,
+      "loss": 7.9039,
       "step": 55
     },
     {
       "epoch": 0.058151609553478714,
-      "grad_norm": 1.2265625,
+      "grad_norm": 0.7421875,
       "learning_rate": 0.0009998936752254111,
-      "loss": 8.3933,
+      "loss": 7.9147,
       "step": 56
     },
     {
       "epoch": 0.059190031152647975,
-      "grad_norm": 1.125,
+      "grad_norm": 0.80859375,
       "learning_rate": 0.0009998552820201655,
-      "loss": 8.2921,
+      "loss": 7.9948,
       "step": 57
     },
     {
       "epoch": 0.060228452751817235,
-      "grad_norm": 1.265625,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.0009998109833891883,
-      "loss": 8.1804,
+      "loss": 7.9213,
       "step": 58
     },
     {
       "epoch": 0.0612668743509865,
-      "grad_norm": 1.171875,
+      "grad_norm": 0.84375,
       "learning_rate": 0.0009997607798558359,
-      "loss": 8.3127,
+      "loss": 7.8629,
       "step": 59
     },
     {
       "epoch": 0.06230529595015576,
-      "grad_norm": 1.2421875,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.0009997046720132262,
-      "loss": 8.1936,
+      "loss": 7.8418,
       "step": 60
     },
     {
       "epoch": 0.06334371754932502,
-      "grad_norm": 1.0,
+      "grad_norm": 0.7265625,
       "learning_rate": 0.0009996426605242323,
-      "loss": 8.2456,
+      "loss": 7.9071,
       "step": 61
     },
     {
       "epoch": 0.06438213914849429,
-      "grad_norm": 1.1484375,
+      "grad_norm": 1.0078125,
       "learning_rate": 0.0009995747461214752,
-      "loss": 8.2582,
+      "loss": 7.8148,
       "step": 62
     },
     {
       "epoch": 0.06542056074766354,
-      "grad_norm": 1.09375,
+      "grad_norm": 0.82421875,
       "learning_rate": 0.0009995009296073138,
-      "loss": 8.2961,
+      "loss": 7.8949,
       "step": 63
     },
     {
       "epoch": 0.06645898234683281,
-      "grad_norm": 1.2421875,
+      "grad_norm": 0.8515625,
       "learning_rate": 0.0009994212118538365,
-      "loss": 8.1676,
+      "loss": 7.7507,
       "step": 64
     },
     {
       "epoch": 0.06749740394600208,
-      "grad_norm": 7.9375,
+      "grad_norm": 0.74609375,
       "learning_rate": 0.00099933559380285,
-      "loss": 8.1897,
+      "loss": 7.8288,
       "step": 65
     },
     {
       "epoch": 0.06853582554517133,
-      "grad_norm": 1.1953125,
+      "grad_norm": 0.75390625,
       "learning_rate": 0.0009992440764658697,
-      "loss": 8.3123,
+      "loss": 7.6728,
       "step": 66
     },
     {
       "epoch": 0.0695742471443406,
-      "grad_norm": 0.9453125,
+      "grad_norm": 0.8984375,
       "learning_rate": 0.000999146660924106,
-      "loss": 8.2322,
+      "loss": 7.65,
       "step": 67
     },
     {
       "epoch": 0.07061266874350987,
-      "grad_norm": 1.1796875,
+      "grad_norm": 1.2890625,
       "learning_rate": 0.0009990433483284527,
-      "loss": 8.2263,
+      "loss": 7.5908,
       "step": 68
     },
     {
       "epoch": 0.07165109034267912,
-      "grad_norm": 1.0703125,
+      "grad_norm": 0.7109375,
       "learning_rate": 0.0009989341398994724,
-      "loss": 8.2091,
+      "loss": 7.5341,
       "step": 69
     },
     {
       "epoch": 0.07268951194184839,
-      "grad_norm": 1.046875,
+      "grad_norm": 0.9375,
       "learning_rate": 0.0009988190369273832,
-      "loss": 8.1462,
+      "loss": 7.4824,
       "step": 70
     },
     {
       "epoch": 0.07372793354101766,
-      "grad_norm": 1.0390625,
+      "grad_norm": 0.91015625,
       "learning_rate": 0.000998698040772043,
-      "loss": 8.183,
+      "loss": 7.489,
       "step": 71
     },
     {
       "epoch": 0.07476635514018691,
-      "grad_norm": 0.9921875,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.000998571152862933,
-      "loss": 8.1936,
+      "loss": 7.3756,
       "step": 72
     },
     {
       "epoch": 0.07580477673935618,
-      "grad_norm": 1.0,
+      "grad_norm": 0.5625,
       "learning_rate": 0.0009984383746991417,
-      "loss": 8.2025,
+      "loss": 7.3947,
       "step": 73
     },
     {
       "epoch": 0.07684319833852545,
-      "grad_norm": 0.984375,
+      "grad_norm": 0.6171875,
       "learning_rate": 0.0009982997078493456,
-      "loss": 8.1422,
+      "loss": 7.3367,
       "step": 74
     },
     {
       "epoch": 0.0778816199376947,
-      "grad_norm": 1.0703125,
+      "grad_norm": 0.90625,
       "learning_rate": 0.0009981551539517929,
-      "loss": 8.1703,
+      "loss": 7.359,
       "step": 75
     },
     {
       "epoch": 0.07892004153686397,
-      "grad_norm": 1.078125,
+      "grad_norm": 0.74609375,
       "learning_rate": 0.0009980047147142824,
-      "loss": 8.1197,
+      "loss": 7.1188,
       "step": 76
     },
     {
       "epoch": 0.07995846313603323,
-      "grad_norm": 1.1015625,
+      "grad_norm": 0.66796875,
       "learning_rate": 0.0009978483919141441,
-      "loss": 8.2123,
+      "loss": 7.3683,
       "step": 77
     },
     {
       "epoch": 0.08099688473520249,
-      "grad_norm": 1.0234375,
+      "grad_norm": 0.92578125,
       "learning_rate": 0.0009976861873982176,
-      "loss": 8.1096,
+      "loss": 7.2359,
       "step": 78
     },
     {
       "epoch": 0.08203530633437175,
-      "grad_norm": 1.5546875,
+      "grad_norm": 1.46875,
       "learning_rate": 0.0009975181030828317,
-      "loss": 8.266,
+      "loss": 7.1375,
       "step": 79
     },
     {
       "epoch": 0.08307372793354102,
-      "grad_norm": 1.2421875,
+      "grad_norm": 0.52734375,
       "learning_rate": 0.0009973441409537793,
-      "loss": 8.1913,
+      "loss": 7.0958,
       "step": 80
     },
     {
       "epoch": 0.08411214953271028,
-      "grad_norm": 1.03125,
+      "grad_norm": 0.62890625,
       "learning_rate": 0.0009971643030662965,
-      "loss": 8.1207,
+      "loss": 7.1365,
       "step": 81
     },
     {
       "epoch": 0.08515057113187954,
-      "grad_norm": 1.03125,
+      "grad_norm": 0.8046875,
       "learning_rate": 0.0009969785915450366,
-      "loss": 8.1482,
+      "loss": 7.122,
       "step": 82
     },
     {
       "epoch": 0.08618899273104881,
-      "grad_norm": 1.203125,
+      "grad_norm": 1.0546875,
       "learning_rate": 0.0009967870085840463,
-      "loss": 8.1108,
+      "loss": 7.0849,
       "step": 83
     },
     {
       "epoch": 0.08722741433021806,
-      "grad_norm": 0.99609375,
+      "grad_norm": 0.458984375,
       "learning_rate": 0.000996589556446738,
-      "loss": 8.082,
+      "loss": 7.0815,
       "step": 84
     },
     {
       "epoch": 0.08826583592938733,
-      "grad_norm": 1.109375,
+      "grad_norm": 0.72265625,
       "learning_rate": 0.000996386237465865,
-      "loss": 8.0509,
+      "loss": 7.109,
       "step": 85
     },
     {
       "epoch": 0.0893042575285566,
-      "grad_norm": 1.0390625,
+      "grad_norm": 0.7734375,
       "learning_rate": 0.000996177054043493,
-      "loss": 8.0733,
+      "loss": 6.9565,
       "step": 86
     },
     {
       "epoch": 0.09034267912772585,
-      "grad_norm": 9.5,
+      "grad_norm": 0.609375,
       "learning_rate": 0.0009959620086509714,
-      "loss": 7.9885,
+      "loss": 7.0025,
       "step": 87
     },
     {
       "epoch": 0.09138110072689512,
-      "grad_norm": 1.0859375,
+      "grad_norm": 0.546875,
       "learning_rate": 0.0009957411038289047,
-      "loss": 7.9692,
+      "loss": 7.002,
       "step": 88
     },
     {
       "epoch": 0.09241952232606439,
-      "grad_norm": 1.03125,
+      "grad_norm": 0.5078125,
       "learning_rate": 0.0009955143421871227,
-      "loss": 8.05,
+      "loss": 6.6959,
       "step": 89
     },
     {
       "epoch": 0.09345794392523364,
-      "grad_norm": 1.0078125,
+      "grad_norm": 0.53515625,
       "learning_rate": 0.0009952817264046484,
-      "loss": 8.1264,
+      "loss": 6.7453,
       "step": 90
     },
     {
       "epoch": 0.09449636552440291,
-      "grad_norm": 1.6796875,
+      "grad_norm": 0.5546875,
       "learning_rate": 0.000995043259229668,
-      "loss": 8.0639,
+      "loss": 6.734,
       "step": 91
     },
     {
       "epoch": 0.09553478712357218,
-      "grad_norm": 1.1328125,
+      "grad_norm": 0.52734375,
       "learning_rate": 0.0009947989434794973,
-      "loss": 8.0093,
+      "loss": 6.8088,
       "step": 92
     },
     {
       "epoch": 0.09657320872274143,
-      "grad_norm": 1.1484375,
+      "grad_norm": 0.80859375,
       "learning_rate": 0.0009945487820405485,
-      "loss": 8.1015,
+      "loss": 6.8563,
       "step": 93
     },
     {
       "epoch": 0.0976116303219107,
-      "grad_norm": 0.91796875,
+      "grad_norm": 0.7734375,
       "learning_rate": 0.0009942927778682968,
-      "loss": 8.0984,
+      "loss": 6.5181,
       "step": 94
     },
     {
       "epoch": 0.09865005192107996,
-      "grad_norm": 1.203125,
+      "grad_norm": 0.60546875,
       "learning_rate": 0.0009940309339872449,
-      "loss": 7.9496,
+      "loss": 6.8268,
       "step": 95
     },
     {
       "epoch": 0.09968847352024922,
-      "grad_norm": 1.25,
+      "grad_norm": 0.625,
       "learning_rate": 0.000993763253490887,
-      "loss": 8.0393,
+      "loss": 6.7739,
       "step": 96
     },
     {
       "epoch": 0.10072689511941849,
-      "grad_norm": 1.109375,
+      "grad_norm": 0.54296875,
       "learning_rate": 0.0009934897395416737,
-      "loss": 8.0389,
+      "loss": 6.6782,
       "step": 97
     },
     {
       "epoch": 0.10176531671858775,
-      "grad_norm": 1.078125,
+      "grad_norm": 0.5625,
       "learning_rate": 0.0009932103953709723,
-      "loss": 7.9644,
+      "loss": 6.727,
       "step": 98
     },
     {
       "epoch": 0.102803738317757,
-      "grad_norm": 1.2578125,
+      "grad_norm": 0.5625,
       "learning_rate": 0.0009929252242790309,
-      "loss": 7.9443,
+      "loss": 6.7222,
       "step": 99
     },
     {
       "epoch": 0.10384215991692627,
-      "grad_norm": 1.390625,
+      "grad_norm": 0.5234375,
       "learning_rate": 0.0009926342296349379,
-      "loss": 7.9222,
+      "loss": 6.5565,
       "step": 100
     },
     {
       "epoch": 0.10384215991692627,
-      "eval_loss": 7.957639694213867,
-      "eval_runtime": 2.6128,
-      "eval_samples_per_second": 6.124,
-      "eval_steps_per_second": 0.765,
+      "eval_loss": 6.583271503448486,
+      "eval_runtime": 2.5831,
+      "eval_samples_per_second": 6.194,
+      "eval_steps_per_second": 0.774,
       "step": 100
-    },
-    {
-      "epoch": 0.10488058151609553,
-      "grad_norm": 1.2890625,
-      "learning_rate": 0.0009923374148765826,
-      "loss": 7.926,
-      "step": 101
-    },
-    {
-      "epoch": 0.1059190031152648,
-      "grad_norm": 1.0703125,
-      "learning_rate": 0.0009920347835106153,
-      "loss": 8.0019,
-      "step": 102
-    },
-    {
-      "epoch": 0.10695742471443406,
-      "grad_norm": 1.4609375,
-      "learning_rate": 0.0009917263391124047,
-      "loss": 7.8873,
-      "step": 103
-    },
-    {
-      "epoch": 0.10799584631360332,
-      "grad_norm": 1.2734375,
-      "learning_rate": 0.0009914120853259966,
-      "loss": 7.9428,
-      "step": 104
-    },
-    {
-      "epoch": 0.10903426791277258,
-      "grad_norm": 5.625,
-      "learning_rate": 0.0009910920258640707,
-      "loss": 8.0591,
-      "step": 105
-    },
-    {
-      "epoch": 0.11007268951194185,
-      "grad_norm": 1.1875,
-      "learning_rate": 0.0009907661645078957,
-      "loss": 7.9165,
-      "step": 106
-    },
-    {
-      "epoch": 0.1111111111111111,
-      "grad_norm": 1.6796875,
-      "learning_rate": 0.0009904345051072862,
-      "loss": 7.997,
-      "step": 107
-    },
-    {
-      "epoch": 0.11214953271028037,
-      "grad_norm": 1.09375,
-      "learning_rate": 0.0009900970515805563,
-      "loss": 7.9629,
-      "step": 108
-    },
-    {
-      "epoch": 0.11318795430944964,
-      "grad_norm": 1.015625,
-      "learning_rate": 0.0009897538079144733,
-      "loss": 7.9483,
-      "step": 109
-    },
-    {
-      "epoch": 0.1142263759086189,
-      "grad_norm": 1.171875,
-      "learning_rate": 0.0009894047781642107,
-      "loss": 7.8955,
-      "step": 110
-    },
-    {
-      "epoch": 0.11526479750778816,
-      "grad_norm": 1.09375,
-      "learning_rate": 0.0009890499664533007,
-      "loss": 7.908,
-      "step": 111
-    },
-    {
-      "epoch": 0.11630321910695743,
-      "grad_norm": 1.9375,
-      "learning_rate": 0.0009886893769735852,
-      "loss": 7.8703,
-      "step": 112
-    },
-    {
-      "epoch": 0.11734164070612668,
-      "grad_norm": 0.99609375,
-      "learning_rate": 0.0009883230139851656,
-      "loss": 7.8814,
-      "step": 113
-    },
-    {
-      "epoch": 0.11838006230529595,
-      "grad_norm": 1.4609375,
-      "learning_rate": 0.0009879508818163537,
-      "loss": 7.9023,
-      "step": 114
-    },
-    {
-      "epoch": 0.11941848390446522,
-      "grad_norm": 1.265625,
-      "learning_rate": 0.0009875729848636196,
-      "loss": 7.7985,
-      "step": 115
-    },
-    {
-      "epoch": 0.12045690550363447,
-      "grad_norm": 1.2890625,
-      "learning_rate": 0.0009871893275915408,
-      "loss": 7.9216,
-      "step": 116
-    },
-    {
-      "epoch": 0.12149532710280374,
-      "grad_norm": 1.2890625,
-      "learning_rate": 0.0009867999145327475,
-      "loss": 7.8029,
-      "step": 117
-    },
-    {
-      "epoch": 0.122533748701973,
-      "grad_norm": 1.125,
-      "learning_rate": 0.0009864047502878717,
-      "loss": 7.8441,
-      "step": 118
-    },
-    {
-      "epoch": 0.12357217030114226,
-      "grad_norm": 1.1171875,
-      "learning_rate": 0.0009860038395254906,
-      "loss": 7.8356,
-      "step": 119
-    },
-    {
-      "epoch": 0.12461059190031153,
-      "grad_norm": 1.1171875,
-      "learning_rate": 0.0009855971869820726,
-      "loss": 7.8124,
-      "step": 120
-    },
-    {
-      "epoch": 0.1256490134994808,
-      "grad_norm": 1.1328125,
-      "learning_rate": 0.000985184797461921,
-      "loss": 7.758,
-      "step": 121
-    },
-    {
-      "epoch": 0.12668743509865005,
-      "grad_norm": 1.2578125,
-      "learning_rate": 0.0009847666758371174,
-      "loss": 7.8092,
-      "step": 122
-    },
-    {
-      "epoch": 0.1277258566978193,
-      "grad_norm": 1.21875,
-      "learning_rate": 0.000984342827047464,
-      "loss": 7.7203,
-      "step": 123
-    },
-    {
-      "epoch": 0.12876427829698858,
-      "grad_norm": 1.2109375,
-      "learning_rate": 0.0009839132561004248,
-      "loss": 7.7885,
-      "step": 124
-    },
-    {
-      "epoch": 0.12980269989615784,
-      "grad_norm": 1.4375,
-      "learning_rate": 0.0009834779680710681,
-      "loss": 7.722,
-      "step": 125
-    },
-    {
-      "epoch": 0.1308411214953271,
-      "grad_norm": 1.4375,
-      "learning_rate": 0.0009830369681020042,
-      "loss": 7.733,
-      "step": 126
-    },
-    {
-      "epoch": 0.13187954309449637,
-      "grad_norm": 2.40625,
-      "learning_rate": 0.000982590261403326,
-      "loss": 7.8116,
-      "step": 127
-    },
-    {
-      "epoch": 0.13291796469366562,
-      "grad_norm": 1.828125,
-      "learning_rate": 0.0009821378532525477,
-      "loss": 7.8408,
-      "step": 128
-    },
-    {
-      "epoch": 0.13395638629283488,
-      "grad_norm": 1.3515625,
-      "learning_rate": 0.000981679748994542,
-      "loss": 7.8247,
-      "step": 129
-    },
-    {
-      "epoch": 0.13499480789200416,
-      "grad_norm": 1.2421875,
-      "learning_rate": 0.0009812159540414764,
-      "loss": 7.6884,
-      "step": 130
-    },
-    {
-      "epoch": 0.1360332294911734,
-      "grad_norm": 1.6640625,
-      "learning_rate": 0.0009807464738727503,
-      "loss": 7.7018,
-      "step": 131
-    },
-    {
-      "epoch": 0.13707165109034267,
-      "grad_norm": 1.2890625,
-      "learning_rate": 0.0009802713140349294,
-      "loss": 7.6106,
-      "step": 132
-    },
-    {
-      "epoch": 0.13811007268951195,
-      "grad_norm": 1.921875,
-      "learning_rate": 0.000979790480141681,
-      "loss": 7.6482,
-      "step": 133
-    },
-    {
-      "epoch": 0.1391484942886812,
-      "grad_norm": 1.4296875,
-      "learning_rate": 0.0009793039778737069,
-      "loss": 7.6953,
-      "step": 134
-    },
-    {
-      "epoch": 0.14018691588785046,
-      "grad_norm": 1.2890625,
-      "learning_rate": 0.0009788118129786768,
-      "loss": 7.5737,
-      "step": 135
-    },
-    {
-      "epoch": 0.14122533748701974,
-      "grad_norm": 1.25,
-      "learning_rate": 0.0009783139912711597,
-      "loss": 7.7097,
-      "step": 136
-    },
-    {
-      "epoch": 0.142263759086189,
-      "grad_norm": 1.4765625,
-      "learning_rate": 0.0009778105186325566,
-      "loss": 7.6463,
-      "step": 137
-    },
-    {
-      "epoch": 0.14330218068535824,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.0009773014010110298,
-      "loss": 7.5251,
-      "step": 138
-    },
-    {
-      "epoch": 0.14434060228452752,
-      "grad_norm": 1.34375,
-      "learning_rate": 0.0009767866444214321,
-      "loss": 7.5961,
-      "step": 139
-    },
-    {
-      "epoch": 0.14537902388369678,
-      "grad_norm": 5.875,
-      "learning_rate": 0.0009762662549452379,
-      "loss": 7.3664,
-      "step": 140
-    },
-    {
-      "epoch": 0.14641744548286603,
-      "grad_norm": 1.4375,
-      "learning_rate": 0.0009757402387304694,
-      "loss": 7.5663,
-      "step": 141
-    },
-    {
-      "epoch": 0.1474558670820353,
-      "grad_norm": 1.5859375,
-      "learning_rate": 0.0009752086019916245,
-      "loss": 7.6656,
-      "step": 142
-    },
-    {
-      "epoch": 0.14849428868120457,
-      "grad_norm": 1.40625,
-      "learning_rate": 0.000974671351009604,
-      "loss": 7.5884,
-      "step": 143
-    },
-    {
-      "epoch": 0.14953271028037382,
-      "grad_norm": 1.3359375,
-      "learning_rate": 0.000974128492131636,
-      "loss": 7.5191,
-      "step": 144
-    },
-    {
-      "epoch": 0.1505711318795431,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.0009735800317712028,
-      "loss": 7.4686,
-      "step": 145
-    },
-    {
-      "epoch": 0.15160955347871236,
-      "grad_norm": 1.7421875,
-      "learning_rate": 0.0009730259764079637,
-      "loss": 7.4524,
-      "step": 146
-    },
-    {
-      "epoch": 0.1526479750778816,
-      "grad_norm": 1.828125,
-      "learning_rate": 0.0009724663325876786,
-      "loss": 7.46,
-      "step": 147
-    },
-    {
-      "epoch": 0.1536863966770509,
-      "grad_norm": 1.7578125,
-      "learning_rate": 0.0009719011069221315,
-      "loss": 7.4372,
-      "step": 148
-    },
-    {
-      "epoch": 0.15472481827622014,
-      "grad_norm": 1.7890625,
-      "learning_rate": 0.0009713303060890514,
-      "loss": 7.3809,
-      "step": 149
-    },
-    {
-      "epoch": 0.1557632398753894,
-      "grad_norm": 2.375,
-      "learning_rate": 0.000970753936832034,
-      "loss": 7.4196,
-      "step": 150
-    },
-    {
-      "epoch": 0.15680166147455868,
-      "grad_norm": 1.3828125,
-      "learning_rate": 0.000970172005960462,
-      "loss": 7.4732,
-      "step": 151
-    },
-    {
-      "epoch": 0.15784008307372793,
-      "grad_norm": 1.5390625,
-      "learning_rate": 0.0009695845203494243,
-      "loss": 7.4662,
-      "step": 152
-    },
-    {
-      "epoch": 0.1588785046728972,
-      "grad_norm": 1.484375,
-      "learning_rate": 0.0009689914869396349,
-      "loss": 7.4651,
-      "step": 153
-    },
-    {
-      "epoch": 0.15991692627206647,
-      "grad_norm": 1.5625,
-      "learning_rate": 0.0009683929127373514,
-      "loss": 7.4005,
-      "step": 154
-    },
-    {
-      "epoch": 0.16095534787123572,
-      "grad_norm": 1.34375,
-      "learning_rate": 0.0009677888048142915,
-      "loss": 7.3895,
-      "step": 155
-    },
-    {
-      "epoch": 0.16199376947040497,
-      "grad_norm": 1.578125,
-      "learning_rate": 0.00096717917030755,
-      "loss": 7.3578,
-      "step": 156
-    },
-    {
-      "epoch": 0.16303219106957426,
-      "grad_norm": 1.578125,
-      "learning_rate": 0.0009665640164195142,
-      "loss": 7.358,
-      "step": 157
-    },
-    {
-      "epoch": 0.1640706126687435,
-      "grad_norm": 6.46875,
-      "learning_rate": 0.0009659433504177785,
-      "loss": 7.0409,
-      "step": 158
-    },
-    {
-      "epoch": 0.16510903426791276,
-      "grad_norm": 1.7265625,
-      "learning_rate": 0.0009653171796350593,
-      "loss": 7.4103,
-      "step": 159
-    },
-    {
-      "epoch": 0.16614745586708204,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.000964685511469108,
-      "loss": 7.3658,
-      "step": 160
-    },
-    {
-      "epoch": 0.1671858774662513,
-      "grad_norm": 1.328125,
-      "learning_rate": 0.0009640483533826234,
-      "loss": 7.2973,
-      "step": 161
-    },
-    {
-      "epoch": 0.16822429906542055,
-      "grad_norm": 1.4765625,
-      "learning_rate": 0.0009634057129031639,
-      "loss": 7.3486,
-      "step": 162
-    },
-    {
-      "epoch": 0.16926272066458983,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.0009627575976230579,
-      "loss": 7.2357,
-      "step": 163
-    },
-    {
-      "epoch": 0.1703011422637591,
-      "grad_norm": 1.640625,
-      "learning_rate": 0.0009621040151993152,
-      "loss": 7.2312,
-      "step": 164
-    },
-    {
-      "epoch": 0.17133956386292834,
-      "grad_norm": 1.5390625,
-      "learning_rate": 0.0009614449733535357,
-      "loss": 7.2768,
-      "step": 165
-    },
-    {
-      "epoch": 0.17237798546209762,
-      "grad_norm": 1.375,
-      "learning_rate": 0.000960780479871818,
-      "loss": 7.282,
-      "step": 166
-    },
-    {
-      "epoch": 0.17341640706126688,
-      "grad_norm": 1.359375,
-      "learning_rate": 0.0009601105426046684,
-      "loss": 7.2627,
-      "step": 167
-    },
-    {
-      "epoch": 0.17445482866043613,
-      "grad_norm": 1.5078125,
-      "learning_rate": 0.000959435169466907,
-      "loss": 7.2436,
-      "step": 168
-    },
-    {
-      "epoch": 0.1754932502596054,
-      "grad_norm": 1.296875,
-      "learning_rate": 0.0009587543684375755,
-      "loss": 7.1775,
-      "step": 169
-    },
-    {
-      "epoch": 0.17653167185877466,
-      "grad_norm": 1.3515625,
-      "learning_rate": 0.0009580681475598413,
-      "loss": 7.2909,
-      "step": 170
-    },
-    {
-      "epoch": 0.17757009345794392,
-      "grad_norm": 1.6015625,
-      "learning_rate": 0.000957376514940904,
-      "loss": 7.2195,
-      "step": 171
-    },
-    {
-      "epoch": 0.1786085150571132,
-      "grad_norm": 1.5859375,
-      "learning_rate": 0.0009566794787518986,
-      "loss": 7.3209,
-      "step": 172
-    },
-    {
-      "epoch": 0.17964693665628245,
-      "grad_norm": 1.4765625,
-      "learning_rate": 0.0009559770472277996,
-      "loss": 7.1312,
-      "step": 173
-    },
-    {
-      "epoch": 0.1806853582554517,
-      "grad_norm": 1.265625,
-      "learning_rate": 0.0009552692286673232,
-      "loss": 7.1432,
-      "step": 174
-    },
-    {
-      "epoch": 0.181723779854621,
-      "grad_norm": 1.4375,
-      "learning_rate": 0.0009545560314328297,
-      "loss": 7.1383,
-      "step": 175
-    },
-    {
-      "epoch": 0.18276220145379024,
-      "grad_norm": 1.7265625,
-      "learning_rate": 0.0009538374639502247,
-      "loss": 7.0784,
-      "step": 176
-    },
-    {
-      "epoch": 0.1838006230529595,
-      "grad_norm": 1.7890625,
-      "learning_rate": 0.0009531135347088589,
-      "loss": 7.2013,
-      "step": 177
-    },
-    {
-      "epoch": 0.18483904465212878,
-      "grad_norm": 1.75,
-      "learning_rate": 0.0009523842522614285,
-      "loss": 7.1522,
-      "step": 178
-    },
-    {
-      "epoch": 0.18587746625129803,
-      "grad_norm": 1.71875,
-      "learning_rate": 0.0009516496252238738,
-      "loss": 7.262,
-      "step": 179
-    },
-    {
-      "epoch": 0.18691588785046728,
-      "grad_norm": 1.6953125,
-      "learning_rate": 0.000950909662275278,
-      "loss": 7.1672,
-      "step": 180
-    },
-    {
-      "epoch": 0.18795430944963656,
-      "grad_norm": 1.8203125,
-      "learning_rate": 0.0009501643721577637,
-      "loss": 7.132,
-      "step": 181
-    },
-    {
-      "epoch": 0.18899273104880582,
-      "grad_norm": 2.171875,
-      "learning_rate": 0.00094941376367639,
-      "loss": 7.0703,
-      "step": 182
-    },
-    {
-      "epoch": 0.19003115264797507,
-      "grad_norm": 1.5234375,
-      "learning_rate": 0.0009486578456990493,
-      "loss": 7.1446,
-      "step": 183
-    },
-    {
-      "epoch": 0.19106957424714435,
-      "grad_norm": 1.578125,
-      "learning_rate": 0.0009478966271563613,
-      "loss": 7.1094,
-      "step": 184
-    },
-    {
-      "epoch": 0.1921079958463136,
-      "grad_norm": 1.8828125,
-      "learning_rate": 0.0009471301170415679,
-      "loss": 7.1068,
-      "step": 185
-    },
-    {
-      "epoch": 0.19314641744548286,
-      "grad_norm": 1.6328125,
-      "learning_rate": 0.0009463583244104273,
-      "loss": 7.1385,
-      "step": 186
-    },
-    {
-      "epoch": 0.19418483904465214,
-      "grad_norm": 1.3671875,
-      "learning_rate": 0.000945581258381107,
-      "loss": 7.0947,
-      "step": 187
-    },
-    {
-      "epoch": 0.1952232606438214,
-      "grad_norm": 1.6328125,
-      "learning_rate": 0.0009447989281340753,
-      "loss": 7.1821,
-      "step": 188
-    },
-    {
-      "epoch": 0.19626168224299065,
-      "grad_norm": 1.75,
-      "learning_rate": 0.0009440113429119939,
-      "loss": 7.0452,
-      "step": 189
-    },
-    {
-      "epoch": 0.19730010384215993,
-      "grad_norm": 1.6015625,
-      "learning_rate": 0.0009432185120196079,
-      "loss": 7.0751,
-      "step": 190
-    },
-    {
-      "epoch": 0.19833852544132918,
-      "grad_norm": 1.75,
-      "learning_rate": 0.0009424204448236364,
-      "loss": 7.2093,
-      "step": 191
-    },
-    {
-      "epoch": 0.19937694704049844,
-      "grad_norm": 2.171875,
-      "learning_rate": 0.0009416171507526614,
-      "loss": 6.9811,
-      "step": 192
-    },
-    {
-      "epoch": 0.20041536863966772,
-      "grad_norm": 1.7421875,
-      "learning_rate": 0.0009408086392970167,
-      "loss": 7.0301,
-      "step": 193
-    },
-    {
-      "epoch": 0.20145379023883697,
-      "grad_norm": 1.6640625,
-      "learning_rate": 0.0009399949200086756,
-      "loss": 6.9737,
-      "step": 194
-    },
-    {
-      "epoch": 0.20249221183800623,
-      "grad_norm": 2.0,
-      "learning_rate": 0.0009391760025011384,
-      "loss": 6.9886,
-      "step": 195
-    },
-    {
-      "epoch": 0.2035306334371755,
-      "grad_norm": 1.359375,
-      "learning_rate": 0.0009383518964493182,
-      "loss": 6.9123,
-      "step": 196
-    },
-    {
-      "epoch": 0.20456905503634476,
-      "grad_norm": 1.5546875,
-      "learning_rate": 0.0009375226115894275,
-      "loss": 7.0419,
-      "step": 197
-    },
-    {
-      "epoch": 0.205607476635514,
-      "grad_norm": 1.3984375,
-      "learning_rate": 0.0009366881577188619,
-      "loss": 7.0344,
-      "step": 198
-    },
-    {
-      "epoch": 0.2066458982346833,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.0009358485446960861,
-      "loss": 7.0015,
-      "step": 199
-    },
-    {
-      "epoch": 0.20768431983385255,
-      "grad_norm": 4.59375,
-      "learning_rate": 0.0009350037824405151,
-      "loss": 7.1113,
-      "step": 200
-    },
-    {
-      "epoch": 0.20768431983385255,
-      "eval_loss": 6.959992408752441,
-      "eval_runtime": 1.5744,
-      "eval_samples_per_second": 10.163,
-      "eval_steps_per_second": 1.27,
-      "step": 200
-    },
-    {
-      "epoch": 0.2087227414330218,
-      "grad_norm": 1.890625,
-      "learning_rate": 0.0009341538809323997,
-      "loss": 6.9605,
-      "step": 201
-    },
-    {
-      "epoch": 0.20976116303219106,
-      "grad_norm": 1.53125,
-      "learning_rate": 0.0009332988502127062,
-      "loss": 7.1006,
-      "step": 202
-    },
-    {
-      "epoch": 0.21079958463136034,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.0009324387003829993,
-      "loss": 7.0535,
-      "step": 203
-    },
-    {
-      "epoch": 0.2118380062305296,
-      "grad_norm": 2.046875,
-      "learning_rate": 0.0009315734416053222,
-      "loss": 7.1432,
-      "step": 204
-    },
-    {
-      "epoch": 0.21287642782969884,
-      "grad_norm": 4.90625,
-      "learning_rate": 0.0009307030841020765,
-      "loss": 6.8602,
-      "step": 205
-    },
-    {
-      "epoch": 0.21391484942886813,
-      "grad_norm": 2.265625,
-      "learning_rate": 0.0009298276381559014,
-      "loss": 6.8611,
-      "step": 206
-    },
-    {
-      "epoch": 0.21495327102803738,
-      "grad_norm": 1.484375,
-      "learning_rate": 0.0009289471141095528,
-      "loss": 6.9153,
-      "step": 207
-    },
-    {
-      "epoch": 0.21599169262720663,
-      "grad_norm": 2.015625,
-      "learning_rate": 0.0009280615223657801,
-      "loss": 6.88,
-      "step": 208
-    },
-    {
-      "epoch": 0.21703011422637591,
-      "grad_norm": 1.875,
-      "learning_rate": 0.0009271708733872042,
-      "loss": 6.9271,
-      "step": 209
-    },
-    {
-      "epoch": 0.21806853582554517,
-      "grad_norm": 1.375,
-      "learning_rate": 0.0009262751776961935,
-      "loss": 6.9736,
-      "step": 210
-    },
-    {
-      "epoch": 0.21910695742471442,
-      "grad_norm": 1.75,
-      "learning_rate": 0.0009253744458747394,
-      "loss": 6.9327,
-      "step": 211
-    },
-    {
-      "epoch": 0.2201453790238837,
-      "grad_norm": 1.390625,
-      "learning_rate": 0.0009244686885643319,
-      "loss": 6.9902,
-      "step": 212
-    },
-    {
-      "epoch": 0.22118380062305296,
-      "grad_norm": 1.5390625,
-      "learning_rate": 0.000923557916465833,
-      "loss": 6.8993,
-      "step": 213
-    },
-    {
-      "epoch": 0.2222222222222222,
-      "grad_norm": 1.640625,
-      "learning_rate": 0.0009226421403393512,
-      "loss": 6.7531,
-      "step": 214
-    },
-    {
-      "epoch": 0.2232606438213915,
-      "grad_norm": 1.3515625,
-      "learning_rate": 0.0009217213710041138,
-      "loss": 6.9481,
-      "step": 215
-    },
-    {
-      "epoch": 0.22429906542056074,
-      "grad_norm": 1.5859375,
-      "learning_rate": 0.0009207956193383392,
-      "loss": 6.9571,
-      "step": 216
-    },
-    {
-      "epoch": 0.22533748701973,
-      "grad_norm": 1.4609375,
-      "learning_rate": 0.0009198648962791083,
-      "loss": 6.7456,
-      "step": 217
-    },
-    {
-      "epoch": 0.22637590861889928,
-      "grad_norm": 1.2890625,
-      "learning_rate": 0.0009189292128222355,
-      "loss": 6.8709,
-      "step": 218
-    },
-    {
-      "epoch": 0.22741433021806853,
-      "grad_norm": 1.7890625,
-      "learning_rate": 0.0009179885800221388,
-      "loss": 6.9518,
-      "step": 219
-    },
-    {
-      "epoch": 0.2284527518172378,
-      "grad_norm": 1.8046875,
-      "learning_rate": 0.0009170430089917088,
-      "loss": 6.8989,
-      "step": 220
-    },
-    {
-      "epoch": 0.22949117341640707,
-      "grad_norm": 1.4453125,
-      "learning_rate": 0.0009160925109021781,
-      "loss": 6.9521,
-      "step": 221
-    },
-    {
-      "epoch": 0.23052959501557632,
-      "grad_norm": 1.859375,
-      "learning_rate": 0.0009151370969829882,
-      "loss": 6.7814,
-      "step": 222
-    },
-    {
-      "epoch": 0.23156801661474558,
-      "grad_norm": 1.625,
-      "learning_rate": 0.0009141767785216584,
-      "loss": 6.808,
-      "step": 223
-    },
-    {
-      "epoch": 0.23260643821391486,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.0009132115668636511,
-      "loss": 6.9264,
-      "step": 224
-    },
-    {
-      "epoch": 0.2336448598130841,
-      "grad_norm": 1.34375,
-      "learning_rate": 0.0009122414734122383,
-      "loss": 6.8812,
-      "step": 225
-    },
-    {
-      "epoch": 0.23468328141225336,
-      "grad_norm": 1.59375,
-      "learning_rate": 0.0009112665096283667,
-      "loss": 6.7264,
-      "step": 226
-    },
-    {
-      "epoch": 0.23572170301142265,
-      "grad_norm": 2.1875,
-      "learning_rate": 0.0009102866870305231,
-      "loss": 6.8659,
-      "step": 227
-    },
-    {
-      "epoch": 0.2367601246105919,
-      "grad_norm": 1.8984375,
-      "learning_rate": 0.0009093020171945965,
-      "loss": 6.8242,
-      "step": 228
-    },
-    {
-      "epoch": 0.23779854620976115,
-      "grad_norm": 1.609375,
-      "learning_rate": 0.0009083125117537436,
-      "loss": 6.7649,
-      "step": 229
-    },
-    {
-      "epoch": 0.23883696780893043,
-      "grad_norm": 1.4765625,
-      "learning_rate": 0.0009073181823982494,
-      "loss": 6.9016,
-      "step": 230
-    },
-    {
-      "epoch": 0.2398753894080997,
-      "grad_norm": 1.625,
-      "learning_rate": 0.0009063190408753908,
-      "loss": 6.794,
-      "step": 231
-    },
-    {
-      "epoch": 0.24091381100726894,
-      "grad_norm": 1.5859375,
-      "learning_rate": 0.000905315098989296,
-      "loss": 6.7241,
-      "step": 232
-    },
-    {
-      "epoch": 0.24195223260643822,
-      "grad_norm": 1.734375,
-      "learning_rate": 0.0009043063686008066,
-      "loss": 6.9263,
-      "step": 233
-    },
-    {
-      "epoch": 0.24299065420560748,
-      "grad_norm": 1.765625,
-      "learning_rate": 0.0009032928616273368,
-      "loss": 6.8098,
-      "step": 234
-    },
-    {
-      "epoch": 0.24402907580477673,
-      "grad_norm": 5.96875,
-      "learning_rate": 0.0009022745900427324,
-      "loss": 6.7429,
-      "step": 235
-    },
-    {
-      "epoch": 0.245067497403946,
-      "grad_norm": 1.7578125,
-      "learning_rate": 0.00090125156587713,
-      "loss": 6.7389,
-      "step": 236
-    },
-    {
-      "epoch": 0.24610591900311526,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.0009002238012168143,
-      "loss": 6.6653,
-      "step": 237
-    },
-    {
-      "epoch": 0.24714434060228452,
-      "grad_norm": 1.703125,
-      "learning_rate": 0.0008991913082040751,
-      "loss": 6.8711,
-      "step": 238
-    },
-    {
-      "epoch": 0.2481827622014538,
-      "grad_norm": 2.015625,
-      "learning_rate": 0.0008981540990370649,
-      "loss": 6.6424,
-      "step": 239
-    },
-    {
-      "epoch": 0.24922118380062305,
-      "grad_norm": 1.609375,
-      "learning_rate": 0.0008971121859696538,
-      "loss": 6.7559,
-      "step": 240
-    },
-    {
-      "epoch": 0.2502596053997923,
-      "grad_norm": 1.703125,
-      "learning_rate": 0.000896065581311285,
-      "loss": 6.7114,
-      "step": 241
-    },
-    {
-      "epoch": 0.2512980269989616,
-      "grad_norm": 1.8203125,
-      "learning_rate": 0.0008950142974268293,
-      "loss": 6.7799,
-      "step": 242
-    },
-    {
-      "epoch": 0.2523364485981308,
-      "grad_norm": 1.3984375,
-      "learning_rate": 0.0008939583467364398,
-      "loss": 6.7178,
-      "step": 243
-    },
-    {
-      "epoch": 0.2533748701973001,
-      "grad_norm": 1.5859375,
-      "learning_rate": 0.0008928977417154037,
-      "loss": 6.731,
-      "step": 244
-    },
-    {
-      "epoch": 0.2544132917964694,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.0008918324948939962,
-      "loss": 6.7293,
-      "step": 245
-    },
-    {
-      "epoch": 0.2554517133956386,
-      "grad_norm": 1.7265625,
-      "learning_rate": 0.0008907626188573319,
-      "loss": 6.8015,
-      "step": 246
-    },
-    {
-      "epoch": 0.2564901349948079,
-      "grad_norm": 2.28125,
-      "learning_rate": 0.000889688126245216,
-      "loss": 6.7796,
-      "step": 247
-    },
-    {
-      "epoch": 0.25752855659397716,
-      "grad_norm": 3.234375,
-      "learning_rate": 0.0008886090297519955,
-      "loss": 6.6777,
-      "step": 248
-    },
-    {
-      "epoch": 0.2585669781931464,
-      "grad_norm": 2.234375,
-      "learning_rate": 0.0008875253421264087,
-      "loss": 6.5925,
-      "step": 249
-    },
-    {
-      "epoch": 0.25960539979231567,
-      "grad_norm": 1.8828125,
-      "learning_rate": 0.0008864370761714347,
-      "loss": 6.7802,
-      "step": 250
-    },
-    {
-      "epoch": 0.26064382139148495,
-      "grad_norm": 2.109375,
-      "learning_rate": 0.0008853442447441426,
-      "loss": 6.6427,
-      "step": 251
-    },
-    {
-      "epoch": 0.2616822429906542,
-      "grad_norm": 1.7265625,
-      "learning_rate": 0.0008842468607555389,
-      "loss": 6.7946,
-      "step": 252
-    },
-    {
-      "epoch": 0.26272066458982346,
-      "grad_norm": 1.90625,
-      "learning_rate": 0.000883144937170415,
-      "loss": 6.6118,
-      "step": 253
-    },
-    {
-      "epoch": 0.26375908618899274,
-      "grad_norm": 3.78125,
-      "learning_rate": 0.0008820384870071951,
-      "loss": 6.8264,
-      "step": 254
-    },
-    {
-      "epoch": 0.26479750778816197,
-      "grad_norm": 1.4375,
-      "learning_rate": 0.0008809275233377813,
-      "loss": 6.6543,
-      "step": 255
-    },
-    {
-      "epoch": 0.26583592938733125,
-      "grad_norm": 1.578125,
-      "learning_rate": 0.0008798120592873989,
-      "loss": 6.6704,
-      "step": 256
-    },
-    {
-      "epoch": 0.26687435098650053,
-      "grad_norm": 1.53125,
-      "learning_rate": 0.0008786921080344428,
-      "loss": 6.7579,
-      "step": 257
-    },
-    {
-      "epoch": 0.26791277258566976,
-      "grad_norm": 1.8125,
-      "learning_rate": 0.0008775676828103205,
-      "loss": 6.6997,
-      "step": 258
-    },
-    {
-      "epoch": 0.26895119418483904,
-      "grad_norm": 1.59375,
-      "learning_rate": 0.0008764387968992961,
-      "loss": 6.8118,
-      "step": 259
-    },
-    {
-      "epoch": 0.2699896157840083,
-      "grad_norm": 3.578125,
-      "learning_rate": 0.0008753054636383336,
-      "loss": 6.4752,
-      "step": 260
-    },
-    {
-      "epoch": 0.27102803738317754,
-      "grad_norm": 2.046875,
-      "learning_rate": 0.0008741676964169394,
-      "loss": 6.8343,
-      "step": 261
-    },
-    {
-      "epoch": 0.2720664589823468,
-      "grad_norm": 1.9921875,
-      "learning_rate": 0.0008730255086770036,
-      "loss": 6.8082,
-      "step": 262
-    },
-    {
-      "epoch": 0.2731048805815161,
-      "grad_norm": 2.296875,
-      "learning_rate": 0.0008718789139126416,
-      "loss": 6.5756,
-      "step": 263
-    },
-    {
-      "epoch": 0.27414330218068533,
-      "grad_norm": 2.25,
-      "learning_rate": 0.0008707279256700347,
-      "loss": 6.6768,
-      "step": 264
-    },
-    {
-      "epoch": 0.2751817237798546,
-      "grad_norm": 1.671875,
-      "learning_rate": 0.0008695725575472697,
-      "loss": 6.6729,
-      "step": 265
-    },
-    {
-      "epoch": 0.2762201453790239,
-      "grad_norm": 2.46875,
-      "learning_rate": 0.0008684128231941788,
-      "loss": 6.7176,
-      "step": 266
-    },
-    {
-      "epoch": 0.2772585669781931,
-      "grad_norm": 2.296875,
-      "learning_rate": 0.0008672487363121777,
-      "loss": 6.6639,
-      "step": 267
-    },
-    {
-      "epoch": 0.2782969885773624,
-      "grad_norm": 1.9453125,
-      "learning_rate": 0.0008660803106541043,
-      "loss": 6.7007,
-      "step": 268
-    },
-    {
-      "epoch": 0.2793354101765317,
-      "grad_norm": 2.0625,
-      "learning_rate": 0.0008649075600240559,
-      "loss": 6.6715,
-      "step": 269
-    },
-    {
-      "epoch": 0.2803738317757009,
-      "grad_norm": 1.890625,
-      "learning_rate": 0.0008637304982772262,
-      "loss": 6.6705,
-      "step": 270
-    },
-    {
-      "epoch": 0.2814122533748702,
-      "grad_norm": 1.8671875,
-      "learning_rate": 0.0008625491393197415,
-      "loss": 6.7322,
-      "step": 271
-    },
-    {
-      "epoch": 0.2824506749740395,
-      "grad_norm": 1.6875,
-      "learning_rate": 0.0008613634971084967,
-      "loss": 6.7985,
-      "step": 272
-    },
-    {
-      "epoch": 0.2834890965732087,
-      "grad_norm": 2.09375,
-      "learning_rate": 0.0008601735856509902,
-      "loss": 6.4347,
-      "step": 273
-    },
-    {
-      "epoch": 0.284527518172378,
-      "grad_norm": 1.5234375,
-      "learning_rate": 0.0008589794190051581,
-      "loss": 6.5613,
-      "step": 274
-    },
-    {
-      "epoch": 0.28556593977154726,
-      "grad_norm": 1.6171875,
-      "learning_rate": 0.0008577810112792089,
-      "loss": 6.5693,
-      "step": 275
-    },
-    {
-      "epoch": 0.2866043613707165,
-      "grad_norm": 1.796875,
-      "learning_rate": 0.000856578376631456,
-      "loss": 6.5121,
-      "step": 276
-    },
-    {
-      "epoch": 0.28764278296988577,
-      "grad_norm": 1.6328125,
-      "learning_rate": 0.0008553715292701512,
-      "loss": 6.5373,
-      "step": 277
-    },
-    {
-      "epoch": 0.28868120456905505,
-      "grad_norm": 1.515625,
-      "learning_rate": 0.0008541604834533159,
-      "loss": 6.6374,
-      "step": 278
-    },
-    {
-      "epoch": 0.2897196261682243,
-      "grad_norm": 1.5859375,
-      "learning_rate": 0.0008529452534885738,
-      "loss": 6.6681,
-      "step": 279
-    },
-    {
-      "epoch": 0.29075804776739356,
-      "grad_norm": 1.6640625,
-      "learning_rate": 0.0008517258537329808,
-      "loss": 6.5905,
-      "step": 280
-    },
-    {
-      "epoch": 0.29179646936656284,
-      "grad_norm": 1.4609375,
-      "learning_rate": 0.0008505022985928559,
-      "loss": 6.5741,
-      "step": 281
-    },
-    {
-      "epoch": 0.29283489096573206,
-      "grad_norm": 1.59375,
-      "learning_rate": 0.0008492746025236113,
-      "loss": 6.5985,
-      "step": 282
-    },
-    {
-      "epoch": 0.29387331256490135,
-      "grad_norm": 1.5,
-      "learning_rate": 0.0008480427800295809,
-      "loss": 6.6211,
-      "step": 283
-    },
-    {
-      "epoch": 0.2949117341640706,
-      "grad_norm": 1.7890625,
-      "learning_rate": 0.0008468068456638491,
-      "loss": 6.5773,
-      "step": 284
-    },
-    {
-      "epoch": 0.29595015576323985,
-      "grad_norm": 1.75,
-      "learning_rate": 0.0008455668140280798,
-      "loss": 6.6446,
-      "step": 285
-    },
-    {
-      "epoch": 0.29698857736240913,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.0008443226997723424,
-      "loss": 6.672,
-      "step": 286
-    },
-    {
-      "epoch": 0.2980269989615784,
-      "grad_norm": 1.9765625,
-      "learning_rate": 0.0008430745175949399,
-      "loss": 6.3888,
-      "step": 287
-    },
-    {
-      "epoch": 0.29906542056074764,
-      "grad_norm": 1.7265625,
-      "learning_rate": 0.0008418222822422348,
-      "loss": 6.583,
-      "step": 288
-    },
-    {
-      "epoch": 0.3001038421599169,
-      "grad_norm": 3.109375,
-      "learning_rate": 0.0008405660085084748,
-      "loss": 6.5971,
-      "step": 289
-    },
-    {
-      "epoch": 0.3011422637590862,
-      "grad_norm": 1.703125,
-      "learning_rate": 0.0008393057112356181,
-      "loss": 6.683,
-      "step": 290
-    },
-    {
-      "epoch": 0.30218068535825543,
-      "grad_norm": 1.734375,
-      "learning_rate": 0.000838041405313158,
-      "loss": 6.6866,
-      "step": 291
-    },
-    {
-      "epoch": 0.3032191069574247,
-      "grad_norm": 1.5234375,
-      "learning_rate": 0.0008367731056779475,
-      "loss": 6.5641,
-      "step": 292
-    },
-    {
-      "epoch": 0.304257528556594,
-      "grad_norm": 1.3515625,
-      "learning_rate": 0.0008355008273140221,
-      "loss": 6.6613,
-      "step": 293
-    },
-    {
-      "epoch": 0.3052959501557632,
-      "grad_norm": 1.6640625,
-      "learning_rate": 0.0008342245852524229,
-      "loss": 6.4916,
-      "step": 294
-    },
-    {
-      "epoch": 0.3063343717549325,
-      "grad_norm": 1.5390625,
-      "learning_rate": 0.0008329443945710196,
-      "loss": 6.5246,
-      "step": 295
-    },
-    {
-      "epoch": 0.3073727933541018,
-      "grad_norm": 2.375,
-      "learning_rate": 0.0008316602703943314,
-      "loss": 6.2067,
-      "step": 296
-    },
-    {
-      "epoch": 0.308411214953271,
-      "grad_norm": 1.8984375,
-      "learning_rate": 0.0008303722278933497,
-      "loss": 6.7264,
-      "step": 297
-    },
-    {
-      "epoch": 0.3094496365524403,
-      "grad_norm": 1.453125,
-      "learning_rate": 0.0008290802822853575,
-      "loss": 6.5385,
-      "step": 298
-    },
-    {
-      "epoch": 0.31048805815160957,
-      "grad_norm": 1.765625,
-      "learning_rate": 0.00082778444883375,
-      "loss": 6.4497,
-      "step": 299
-    },
-    {
-      "epoch": 0.3115264797507788,
-      "grad_norm": 1.5390625,
-      "learning_rate": 0.0008264847428478549,
-      "loss": 6.6755,
-      "step": 300
-    },
-    {
-      "epoch": 0.3115264797507788,
-      "eval_loss": 6.586581230163574,
-      "eval_runtime": 1.5864,
-      "eval_samples_per_second": 10.086,
-      "eval_steps_per_second": 1.261,
-      "step": 300
-    },
-    {
-      "epoch": 0.3125649013499481,
-      "grad_norm": 1.7890625,
-      "learning_rate": 0.0008251811796827506,
-      "loss": 6.7036,
-      "step": 301
-    },
-    {
-      "epoch": 0.31360332294911736,
-      "grad_norm": 1.734375,
-      "learning_rate": 0.0008238737747390859,
-      "loss": 6.641,
-      "step": 302
-    },
-    {
-      "epoch": 0.3146417445482866,
-      "grad_norm": 1.6875,
-      "learning_rate": 0.0008225625434628969,
-      "loss": 6.4883,
-      "step": 303
-    },
-    {
-      "epoch": 0.31568016614745587,
-      "grad_norm": 1.6796875,
-      "learning_rate": 0.0008212475013454248,
-      "loss": 6.5421,
-      "step": 304
-    },
-    {
-      "epoch": 0.31671858774662515,
-      "grad_norm": 1.6171875,
-      "learning_rate": 0.0008199286639229339,
-      "loss": 6.5171,
-      "step": 305
-    },
-    {
-      "epoch": 0.3177570093457944,
-      "grad_norm": 1.5859375,
-      "learning_rate": 0.0008186060467765268,
-      "loss": 6.4292,
-      "step": 306
-    },
-    {
-      "epoch": 0.31879543094496365,
-      "grad_norm": 1.6953125,
-      "learning_rate": 0.0008172796655319606,
-      "loss": 6.604,
-      "step": 307
-    },
-    {
-      "epoch": 0.31983385254413293,
-      "grad_norm": 1.6953125,
-      "learning_rate": 0.0008159495358594628,
-      "loss": 6.5217,
-      "step": 308
-    },
-    {
-      "epoch": 0.32087227414330216,
-      "grad_norm": 1.453125,
-      "learning_rate": 0.0008146156734735457,
-      "loss": 6.5842,
-      "step": 309
-    },
-    {
-      "epoch": 0.32191069574247144,
-      "grad_norm": 1.640625,
-      "learning_rate": 0.0008132780941328211,
-      "loss": 6.4846,
-      "step": 310
-    },
-    {
-      "epoch": 0.3229491173416407,
-      "grad_norm": 1.8046875,
-      "learning_rate": 0.0008119368136398133,
-      "loss": 6.488,
-      "step": 311
-    },
-    {
-      "epoch": 0.32398753894080995,
-      "grad_norm": 1.890625,
-      "learning_rate": 0.0008105918478407739,
-      "loss": 6.6186,
-      "step": 312
-    },
-    {
-      "epoch": 0.32502596053997923,
-      "grad_norm": 2.109375,
-      "learning_rate": 0.0008092432126254931,
-      "loss": 6.5442,
-      "step": 313
-    },
-    {
-      "epoch": 0.3260643821391485,
-      "grad_norm": 1.609375,
-      "learning_rate": 0.0008078909239271126,
-      "loss": 6.5886,
-      "step": 314
-    },
-    {
-      "epoch": 0.32710280373831774,
-      "grad_norm": 1.796875,
-      "learning_rate": 0.0008065349977219379,
-      "loss": 6.5669,
-      "step": 315
-    },
-    {
-      "epoch": 0.328141225337487,
-      "grad_norm": 2.109375,
-      "learning_rate": 0.0008051754500292479,
-      "loss": 6.6405,
-      "step": 316
-    },
-    {
-      "epoch": 0.3291796469366563,
-      "grad_norm": 1.8671875,
-      "learning_rate": 0.0008038122969111079,
-      "loss": 6.4938,
-      "step": 317
-    },
-    {
-      "epoch": 0.3302180685358255,
-      "grad_norm": 1.5390625,
-      "learning_rate": 0.0008024455544721779,
-      "loss": 6.5688,
-      "step": 318
-    },
-    {
-      "epoch": 0.3312564901349948,
-      "grad_norm": 1.9296875,
-      "learning_rate": 0.0008010752388595235,
-      "loss": 6.5988,
-      "step": 319
-    },
-    {
-      "epoch": 0.3322949117341641,
-      "grad_norm": 1.640625,
-      "learning_rate": 0.0007997013662624246,
-      "loss": 6.4479,
-      "step": 320
-    },
-    {
-      "epoch": 0.3333333333333333,
-      "grad_norm": 1.8828125,
-      "learning_rate": 0.0007983239529121843,
-      "loss": 6.3366,
-      "step": 321
-    },
-    {
-      "epoch": 0.3343717549325026,
-      "grad_norm": 1.6171875,
-      "learning_rate": 0.0007969430150819372,
-      "loss": 6.4521,
-      "step": 322
-    },
-    {
-      "epoch": 0.3354101765316719,
-      "grad_norm": 1.6796875,
-      "learning_rate": 0.0007955585690864566,
-      "loss": 6.2812,
-      "step": 323
-    },
-    {
-      "epoch": 0.3364485981308411,
-      "grad_norm": 1.921875,
-      "learning_rate": 0.0007941706312819631,
-      "loss": 6.4711,
-      "step": 324
-    },
-    {
-      "epoch": 0.3374870197300104,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.0007927792180659296,
-      "loss": 6.4105,
-      "step": 325
-    },
-    {
-      "epoch": 0.33852544132917967,
-      "grad_norm": 1.7890625,
-      "learning_rate": 0.0007913843458768891,
-      "loss": 6.4175,
-      "step": 326
-    },
-    {
-      "epoch": 0.3395638629283489,
-      "grad_norm": 1.890625,
-      "learning_rate": 0.0007899860311942394,
-      "loss": 6.493,
-      "step": 327
-    },
-    {
-      "epoch": 0.3406022845275182,
-      "grad_norm": 1.6171875,
-      "learning_rate": 0.000788584290538049,
-      "loss": 6.553,
-      "step": 328
-    },
-    {
-      "epoch": 0.34164070612668745,
-      "grad_norm": 1.65625,
-      "learning_rate": 0.0007871791404688617,
-      "loss": 6.512,
-      "step": 329
-    },
-    {
-      "epoch": 0.3426791277258567,
-      "grad_norm": 1.6328125,
-      "learning_rate": 0.0007857705975875015,
-      "loss": 6.4949,
-      "step": 330
-    },
-    {
-      "epoch": 0.34371754932502596,
-      "grad_norm": 1.7265625,
-      "learning_rate": 0.0007843586785348752,
-      "loss": 6.4785,
-      "step": 331
-    },
-    {
-      "epoch": 0.34475597092419524,
-      "grad_norm": 1.5234375,
-      "learning_rate": 0.0007829433999917773,
-      "loss": 6.452,
-      "step": 332
-    },
-    {
-      "epoch": 0.34579439252336447,
-      "grad_norm": 1.765625,
-      "learning_rate": 0.0007815247786786919,
-      "loss": 6.6337,
-      "step": 333
-    },
-    {
-      "epoch": 0.34683281412253375,
-      "grad_norm": 1.40625,
-      "learning_rate": 0.0007801028313555953,
-      "loss": 6.4057,
-      "step": 334
-    },
-    {
-      "epoch": 0.34787123572170303,
-      "grad_norm": 1.5625,
-      "learning_rate": 0.000778677574821759,
-      "loss": 6.3247,
-      "step": 335
-    },
-    {
-      "epoch": 0.34890965732087226,
-      "grad_norm": 1.671875,
-      "learning_rate": 0.0007772490259155494,
-      "loss": 6.4594,
-      "step": 336
-    },
-    {
-      "epoch": 0.34994807892004154,
-      "grad_norm": 1.5078125,
-      "learning_rate": 0.0007758172015142302,
-      "loss": 6.4916,
-      "step": 337
-    },
-    {
-      "epoch": 0.3509865005192108,
-      "grad_norm": 1.75,
-      "learning_rate": 0.0007743821185337633,
-      "loss": 6.6281,
-      "step": 338
-    },
-    {
-      "epoch": 0.35202492211838005,
-      "grad_norm": 1.8125,
-      "learning_rate": 0.0007729437939286074,
-      "loss": 6.4948,
-      "step": 339
-    },
-    {
-      "epoch": 0.3530633437175493,
-      "grad_norm": 1.6640625,
-      "learning_rate": 0.0007715022446915194,
-      "loss": 6.6097,
-      "step": 340
-    },
-    {
-      "epoch": 0.3541017653167186,
-      "grad_norm": 1.609375,
-      "learning_rate": 0.0007700574878533524,
-      "loss": 6.5094,
-      "step": 341
-    },
-    {
-      "epoch": 0.35514018691588783,
-      "grad_norm": 1.5703125,
-      "learning_rate": 0.0007686095404828553,
-      "loss": 6.5109,
-      "step": 342
-    },
-    {
-      "epoch": 0.3561786085150571,
-      "grad_norm": 2.0625,
-      "learning_rate": 0.0007671584196864702,
-      "loss": 6.519,
-      "step": 343
-    },
-    {
-      "epoch": 0.3572170301142264,
-      "grad_norm": 1.5234375,
-      "learning_rate": 0.0007657041426081319,
-      "loss": 6.5761,
-      "step": 344
-    },
-    {
-      "epoch": 0.3582554517133956,
-      "grad_norm": 1.8671875,
-      "learning_rate": 0.0007642467264290636,
-      "loss": 6.6179,
-      "step": 345
-    },
-    {
-      "epoch": 0.3592938733125649,
-      "grad_norm": 1.625,
-      "learning_rate": 0.0007627861883675748,
-      "loss": 6.5245,
-      "step": 346
-    },
-    {
-      "epoch": 0.3603322949117342,
-      "grad_norm": 1.8046875,
-      "learning_rate": 0.0007613225456788579,
-      "loss": 6.3649,
-      "step": 347
-    },
-    {
-      "epoch": 0.3613707165109034,
-      "grad_norm": 1.9296875,
-      "learning_rate": 0.0007598558156547841,
-      "loss": 6.3155,
-      "step": 348
-    },
-    {
-      "epoch": 0.3624091381100727,
-      "grad_norm": 1.5078125,
-      "learning_rate": 0.0007583860156236997,
-      "loss": 6.4403,
-      "step": 349
-    },
-    {
-      "epoch": 0.363447559709242,
-      "grad_norm": 1.4296875,
-      "learning_rate": 0.0007569131629502201,
-      "loss": 6.5533,
-      "step": 350
-    },
-    {
-      "epoch": 0.3644859813084112,
-      "grad_norm": 1.6953125,
-      "learning_rate": 0.0007554372750350259,
-      "loss": 6.4351,
-      "step": 351
-    },
-    {
-      "epoch": 0.3655244029075805,
-      "grad_norm": 1.28125,
-      "learning_rate": 0.000753958369314657,
-      "loss": 6.511,
-      "step": 352
-    },
-    {
-      "epoch": 0.36656282450674976,
-      "grad_norm": 2.421875,
-      "learning_rate": 0.000752476463261306,
-      "loss": 6.4314,
-      "step": 353
-    },
-    {
-      "epoch": 0.367601246105919,
-      "grad_norm": 1.421875,
-      "learning_rate": 0.0007509915743826128,
-      "loss": 6.4206,
-      "step": 354
-    },
-    {
-      "epoch": 0.36863966770508827,
-      "grad_norm": 1.5546875,
-      "learning_rate": 0.0007495037202214565,
-      "loss": 6.5012,
-      "step": 355
-    },
-    {
-      "epoch": 0.36967808930425755,
-      "grad_norm": 1.3203125,
-      "learning_rate": 0.0007480129183557498,
-      "loss": 6.4068,
-      "step": 356
-    },
-    {
-      "epoch": 0.3707165109034268,
-      "grad_norm": 1.4140625,
-      "learning_rate": 0.0007465191863982295,
-      "loss": 6.5955,
-      "step": 357
-    },
-    {
-      "epoch": 0.37175493250259606,
-      "grad_norm": 1.484375,
-      "learning_rate": 0.0007450225419962498,
-      "loss": 6.4707,
-      "step": 358
-    },
-    {
-      "epoch": 0.37279335410176534,
-      "grad_norm": 1.7109375,
-      "learning_rate": 0.0007435230028315731,
-      "loss": 6.4684,
-      "step": 359
-    },
-    {
-      "epoch": 0.37383177570093457,
-      "grad_norm": 1.2265625,
-      "learning_rate": 0.0007420205866201614,
-      "loss": 6.4594,
-      "step": 360
-    },
-    {
-      "epoch": 0.37487019730010385,
-      "grad_norm": 1.75,
-      "learning_rate": 0.0007405153111119668,
-      "loss": 6.4225,
-      "step": 361
-    },
-    {
-      "epoch": 0.37590861889927313,
-      "grad_norm": 1.65625,
-      "learning_rate": 0.0007390071940907222,
-      "loss": 6.394,
-      "step": 362
-    },
-    {
-      "epoch": 0.37694704049844235,
-      "grad_norm": 1.796875,
-      "learning_rate": 0.0007374962533737304,
-      "loss": 6.3531,
-      "step": 363
-    },
-    {
-      "epoch": 0.37798546209761164,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.0007359825068116548,
-      "loss": 6.2633,
-      "step": 364
-    },
-    {
-      "epoch": 0.3790238836967809,
-      "grad_norm": 1.25,
-      "learning_rate": 0.0007344659722883072,
-      "loss": 6.5055,
-      "step": 365
-    },
-    {
-      "epoch": 0.38006230529595014,
-      "grad_norm": 1.3046875,
-      "learning_rate": 0.000732946667720437,
-      "loss": 6.4877,
-      "step": 366
-    },
-    {
-      "epoch": 0.3811007268951194,
-      "grad_norm": 2.8125,
-      "learning_rate": 0.0007314246110575206,
-      "loss": 6.5176,
-      "step": 367
-    },
-    {
-      "epoch": 0.3821391484942887,
-      "grad_norm": 1.75,
-      "learning_rate": 0.0007298998202815473,
-      "loss": 6.5267,
-      "step": 368
-    },
-    {
-      "epoch": 0.38317757009345793,
-      "grad_norm": 1.5,
-      "learning_rate": 0.0007283723134068089,
-      "loss": 6.4728,
-      "step": 369
-    },
-    {
-      "epoch": 0.3842159916926272,
-      "grad_norm": 1.953125,
-      "learning_rate": 0.0007268421084796851,
-      "loss": 6.3112,
-      "step": 370
-    },
-    {
-      "epoch": 0.3852544132917965,
-      "grad_norm": 1.6953125,
-      "learning_rate": 0.0007253092235784317,
-      "loss": 6.3693,
-      "step": 371
-    },
-    {
-      "epoch": 0.3862928348909657,
-      "grad_norm": 1.796875,
-      "learning_rate": 0.0007237736768129663,
-      "loss": 6.4253,
-      "step": 372
-    },
-    {
-      "epoch": 0.387331256490135,
-      "grad_norm": 1.578125,
-      "learning_rate": 0.0007222354863246542,
-      "loss": 6.566,
-      "step": 373
-    },
-    {
-      "epoch": 0.3883696780893043,
-      "grad_norm": 1.5625,
-      "learning_rate": 0.0007206946702860947,
-      "loss": 6.49,
-      "step": 374
-    },
-    {
-      "epoch": 0.3894080996884735,
-      "grad_norm": 1.4453125,
-      "learning_rate": 0.000719151246900906,
-      "loss": 6.5474,
-      "step": 375
-    },
-    {
-      "epoch": 0.3904465212876428,
-      "grad_norm": 1.84375,
-      "learning_rate": 0.00071760523440351,
-      "loss": 6.111,
-      "step": 376
-    },
-    {
-      "epoch": 0.39148494288681207,
-      "grad_norm": 3.21875,
-      "learning_rate": 0.0007160566510589174,
-      "loss": 6.1774,
-      "step": 377
-    },
-    {
-      "epoch": 0.3925233644859813,
-      "grad_norm": 1.5390625,
-      "learning_rate": 0.0007145055151625113,
-      "loss": 6.4977,
-      "step": 378
-    },
-    {
-      "epoch": 0.3935617860851506,
-      "grad_norm": 1.4453125,
-      "learning_rate": 0.0007129518450398314,
-      "loss": 6.3905,
-      "step": 379
-    },
-    {
-      "epoch": 0.39460020768431986,
-      "grad_norm": 1.9609375,
-      "learning_rate": 0.0007113956590463575,
-      "loss": 6.2437,
-      "step": 380
-    },
-    {
-      "epoch": 0.3956386292834891,
-      "grad_norm": 1.4140625,
-      "learning_rate": 0.0007098369755672926,
-      "loss": 6.3716,
-      "step": 381
-    },
-    {
-      "epoch": 0.39667705088265837,
-      "grad_norm": 1.546875,
-      "learning_rate": 0.0007082758130173456,
-      "loss": 6.4034,
-      "step": 382
-    },
-    {
-      "epoch": 0.39771547248182765,
-      "grad_norm": 1.796875,
-      "learning_rate": 0.0007067121898405138,
-      "loss": 6.4803,
-      "step": 383
-    },
-    {
-      "epoch": 0.3987538940809969,
-      "grad_norm": 1.5859375,
-      "learning_rate": 0.0007051461245098653,
-      "loss": 6.2609,
-      "step": 384
-    },
-    {
-      "epoch": 0.39979231568016615,
-      "grad_norm": 1.8515625,
-      "learning_rate": 0.0007035776355273199,
-      "loss": 6.3421,
-      "step": 385
-    },
-    {
-      "epoch": 0.40083073727933544,
-      "grad_norm": 1.3515625,
-      "learning_rate": 0.0007020067414234315,
-      "loss": 6.3368,
-      "step": 386
-    },
-    {
-      "epoch": 0.40186915887850466,
-      "grad_norm": 1.3515625,
-      "learning_rate": 0.0007004334607571687,
-      "loss": 6.2582,
-      "step": 387
-    },
-    {
-      "epoch": 0.40290758047767394,
-      "grad_norm": 1.96875,
-      "learning_rate": 0.0006988578121156955,
-      "loss": 6.4665,
-      "step": 388
-    },
-    {
-      "epoch": 0.4039460020768432,
-      "grad_norm": 1.515625,
-      "learning_rate": 0.000697279814114152,
-      "loss": 6.486,
-      "step": 389
-    },
-    {
-      "epoch": 0.40498442367601245,
-      "grad_norm": 1.734375,
-      "learning_rate": 0.0006956994853954342,
-      "loss": 6.4766,
-      "step": 390
-    },
-    {
-      "epoch": 0.40602284527518173,
-      "grad_norm": 1.5703125,
-      "learning_rate": 0.0006941168446299733,
-      "loss": 6.235,
-      "step": 391
-    },
-    {
-      "epoch": 0.407061266874351,
-      "grad_norm": 2.875,
-      "learning_rate": 0.0006925319105155164,
-      "loss": 6.0013,
-      "step": 392
-    },
-    {
-      "epoch": 0.40809968847352024,
-      "grad_norm": 1.796875,
-      "learning_rate": 0.0006909447017769046,
-      "loss": 6.2921,
-      "step": 393
-    },
-    {
-      "epoch": 0.4091381100726895,
-      "grad_norm": 1.5078125,
-      "learning_rate": 0.000689355237165852,
-      "loss": 6.3846,
-      "step": 394
-    },
-    {
-      "epoch": 0.4101765316718588,
-      "grad_norm": 1.7890625,
-      "learning_rate": 0.0006877635354607238,
-      "loss": 6.365,
-      "step": 395
-    },
-    {
-      "epoch": 0.411214953271028,
-      "grad_norm": 1.328125,
-      "learning_rate": 0.0006861696154663158,
-      "loss": 6.4089,
-      "step": 396
-    },
-    {
-      "epoch": 0.4122533748701973,
-      "grad_norm": 1.1796875,
-      "learning_rate": 0.0006845734960136301,
-      "loss": 6.4685,
-      "step": 397
-    },
-    {
-      "epoch": 0.4132917964693666,
-      "grad_norm": 1.3046875,
-      "learning_rate": 0.0006829751959596544,
-      "loss": 6.4737,
-      "step": 398
-    },
-    {
-      "epoch": 0.4143302180685358,
-      "grad_norm": 2.09375,
-      "learning_rate": 0.000681374734187139,
-      "loss": 5.9592,
-      "step": 399
-    },
-    {
-      "epoch": 0.4153686396677051,
-      "grad_norm": 1.7109375,
-      "learning_rate": 0.0006797721296043726,
-      "loss": 6.5409,
-      "step": 400
-    },
-    {
-      "epoch": 0.4153686396677051,
-      "eval_loss": 6.42236328125,
-      "eval_runtime": 1.6609,
-      "eval_samples_per_second": 9.633,
-      "eval_steps_per_second": 1.204,
-      "step": 400
-    },
-    {
-      "epoch": 0.4164070612668744,
-      "grad_norm": 1.875,
-      "learning_rate": 0.0006781674011449602,
-      "loss": 6.3909,
-      "step": 401
-    },
-    {
-      "epoch": 0.4174454828660436,
-      "grad_norm": 1.6328125,
-      "learning_rate": 0.0006765605677675981,
-      "loss": 6.3576,
-      "step": 402
-    },
-    {
-      "epoch": 0.4184839044652129,
-      "grad_norm": 1.40625,
-      "learning_rate": 0.0006749516484558518,
-      "loss": 6.3327,
-      "step": 403
-    },
-    {
-      "epoch": 0.4195223260643821,
-      "grad_norm": 1.5546875,
-      "learning_rate": 0.0006733406622179294,
-      "loss": 6.4531,
-      "step": 404
-    },
-    {
-      "epoch": 0.4205607476635514,
-      "grad_norm": 1.3515625,
-      "learning_rate": 0.000671727628086459,
-      "loss": 6.4259,
-      "step": 405
-    },
-    {
-      "epoch": 0.4215991692627207,
-      "grad_norm": 1.421875,
-      "learning_rate": 0.0006701125651182631,
-      "loss": 6.3521,
-      "step": 406
-    },
-    {
-      "epoch": 0.4226375908618899,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.000668495492394133,
-      "loss": 6.4586,
-      "step": 407
-    },
-    {
-      "epoch": 0.4236760124610592,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.0006668764290186039,
-      "loss": 6.3092,
-      "step": 408
-    },
-    {
-      "epoch": 0.42471443406022846,
-      "grad_norm": 1.328125,
-      "learning_rate": 0.0006652553941197294,
-      "loss": 6.4544,
-      "step": 409
-    },
-    {
-      "epoch": 0.4257528556593977,
-      "grad_norm": 2.078125,
-      "learning_rate": 0.0006636324068488547,
-      "loss": 6.4361,
-      "step": 410
-    },
-    {
-      "epoch": 0.42679127725856697,
-      "grad_norm": 1.7109375,
-      "learning_rate": 0.0006620074863803913,
-      "loss": 6.3362,
-      "step": 411
-    },
-    {
-      "epoch": 0.42782969885773625,
-      "grad_norm": 1.8203125,
-      "learning_rate": 0.0006603806519115899,
-      "loss": 6.4031,
-      "step": 412
-    },
-    {
-      "epoch": 0.4288681204569055,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.0006587519226623137,
-      "loss": 6.3723,
-      "step": 413
-    },
-    {
-      "epoch": 0.42990654205607476,
-      "grad_norm": 2.09375,
-      "learning_rate": 0.0006571213178748112,
-      "loss": 6.5301,
-      "step": 414
-    },
-    {
-      "epoch": 0.43094496365524404,
-      "grad_norm": 1.609375,
-      "learning_rate": 0.0006554888568134894,
-      "loss": 6.3084,
-      "step": 415
-    },
-    {
-      "epoch": 0.43198338525441327,
-      "grad_norm": 1.9921875,
-      "learning_rate": 0.0006538545587646854,
-      "loss": 6.4097,
-      "step": 416
-    },
-    {
-      "epoch": 0.43302180685358255,
-      "grad_norm": 2.09375,
-      "learning_rate": 0.0006522184430364391,
-      "loss": 6.2246,
-      "step": 417
-    },
-    {
-      "epoch": 0.43406022845275183,
-      "grad_norm": 1.875,
-      "learning_rate": 0.0006505805289582649,
-      "loss": 6.4737,
-      "step": 418
-    },
-    {
-      "epoch": 0.43509865005192105,
-      "grad_norm": 1.53125,
-      "learning_rate": 0.0006489408358809239,
-      "loss": 6.4345,
-      "step": 419
-    },
-    {
-      "epoch": 0.43613707165109034,
-      "grad_norm": 1.421875,
-      "learning_rate": 0.000647299383176194,
-      "loss": 6.2794,
-      "step": 420
-    },
-    {
-      "epoch": 0.4371754932502596,
-      "grad_norm": 1.578125,
-      "learning_rate": 0.0006456561902366424,
-      "loss": 6.3705,
-      "step": 421
-    },
-    {
-      "epoch": 0.43821391484942884,
-      "grad_norm": 1.3828125,
-      "learning_rate": 0.0006440112764753956,
-      "loss": 6.3456,
-      "step": 422
-    },
-    {
-      "epoch": 0.4392523364485981,
-      "grad_norm": 1.7109375,
-      "learning_rate": 0.0006423646613259104,
-      "loss": 6.4904,
-      "step": 423
-    },
-    {
-      "epoch": 0.4402907580477674,
-      "grad_norm": 1.5390625,
-      "learning_rate": 0.0006407163642417442,
-      "loss": 6.2549,
-      "step": 424
-    },
-    {
-      "epoch": 0.44132917964693663,
-      "grad_norm": 1.9375,
-      "learning_rate": 0.0006390664046963256,
-      "loss": 6.4673,
-      "step": 425
-    },
-    {
-      "epoch": 0.4423676012461059,
-      "grad_norm": 1.578125,
-      "learning_rate": 0.0006374148021827237,
-      "loss": 6.4597,
-      "step": 426
-    },
-    {
-      "epoch": 0.4434060228452752,
-      "grad_norm": 1.4296875,
-      "learning_rate": 0.0006357615762134178,
-      "loss": 6.3055,
-      "step": 427
-    },
-    {
-      "epoch": 0.4444444444444444,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.0006341067463200677,
-      "loss": 6.3134,
-      "step": 428
-    },
-    {
-      "epoch": 0.4454828660436137,
-      "grad_norm": 1.8359375,
-      "learning_rate": 0.0006324503320532819,
-      "loss": 5.9713,
-      "step": 429
-    },
-    {
-      "epoch": 0.446521287642783,
-      "grad_norm": 1.4765625,
-      "learning_rate": 0.0006307923529823876,
-      "loss": 6.3577,
-      "step": 430
-    },
-    {
-      "epoch": 0.4475597092419522,
-      "grad_norm": 3.546875,
-      "learning_rate": 0.0006291328286951985,
-      "loss": 6.227,
-      "step": 431
-    },
-    {
-      "epoch": 0.4485981308411215,
-      "grad_norm": 1.796875,
-      "learning_rate": 0.0006274717787977842,
-      "loss": 6.1991,
-      "step": 432
-    },
-    {
-      "epoch": 0.44963655244029077,
-      "grad_norm": 1.5625,
-      "learning_rate": 0.0006258092229142382,
-      "loss": 6.4064,
-      "step": 433
-    },
-    {
-      "epoch": 0.45067497403946,
-      "grad_norm": 1.734375,
-      "learning_rate": 0.0006241451806864465,
-      "loss": 6.4659,
-      "step": 434
-    },
-    {
-      "epoch": 0.4517133956386293,
-      "grad_norm": 1.3359375,
-      "learning_rate": 0.0006224796717738544,
-      "loss": 6.3568,
-      "step": 435
-    },
-    {
-      "epoch": 0.45275181723779856,
-      "grad_norm": 1.40625,
-      "learning_rate": 0.0006208127158532357,
-      "loss": 6.2009,
-      "step": 436
-    },
-    {
-      "epoch": 0.4537902388369678,
-      "grad_norm": 1.9765625,
-      "learning_rate": 0.0006191443326184591,
-      "loss": 6.3971,
-      "step": 437
-    },
-    {
-      "epoch": 0.45482866043613707,
-      "grad_norm": 1.625,
-      "learning_rate": 0.0006174745417802563,
-      "loss": 6.5802,
-      "step": 438
-    },
-    {
-      "epoch": 0.45586708203530635,
-      "grad_norm": 3.25,
-      "learning_rate": 0.0006158033630659886,
-      "loss": 5.9341,
-      "step": 439
-    },
-    {
-      "epoch": 0.4569055036344756,
-      "grad_norm": 1.6171875,
-      "learning_rate": 0.0006141308162194142,
-      "loss": 6.2119,
-      "step": 440
-    },
-    {
-      "epoch": 0.45794392523364486,
-      "grad_norm": 1.5859375,
-      "learning_rate": 0.0006124569210004547,
-      "loss": 6.3936,
-      "step": 441
-    },
-    {
-      "epoch": 0.45898234683281414,
-      "grad_norm": 1.9296875,
-      "learning_rate": 0.0006107816971849616,
-      "loss": 6.1906,
-      "step": 442
-    },
-    {
-      "epoch": 0.46002076843198336,
-      "grad_norm": 1.484375,
-      "learning_rate": 0.000609105164564483,
-      "loss": 6.1966,
-      "step": 443
-    },
-    {
-      "epoch": 0.46105919003115264,
-      "grad_norm": 1.3984375,
-      "learning_rate": 0.0006074273429460295,
-      "loss": 6.358,
-      "step": 444
-    },
-    {
-      "epoch": 0.4620976116303219,
-      "grad_norm": 1.8828125,
-      "learning_rate": 0.0006057482521518403,
-      "loss": 6.4207,
-      "step": 445
-    },
-    {
-      "epoch": 0.46313603322949115,
-      "grad_norm": 1.546875,
-      "learning_rate": 0.0006040679120191491,
-      "loss": 6.3122,
-      "step": 446
-    },
-    {
-      "epoch": 0.46417445482866043,
-      "grad_norm": 1.7734375,
-      "learning_rate": 0.0006023863423999496,
-      "loss": 6.2674,
-      "step": 447
-    },
-    {
-      "epoch": 0.4652128764278297,
-      "grad_norm": 1.6171875,
-      "learning_rate": 0.0006007035631607605,
-      "loss": 6.285,
-      "step": 448
-    },
-    {
-      "epoch": 0.46625129802699894,
-      "grad_norm": 1.546875,
-      "learning_rate": 0.0005990195941823916,
-      "loss": 6.2897,
-      "step": 449
-    },
-    {
-      "epoch": 0.4672897196261682,
-      "grad_norm": 1.703125,
-      "learning_rate": 0.0005973344553597091,
-      "loss": 6.3552,
-      "step": 450
-    },
-    {
-      "epoch": 0.4683281412253375,
-      "grad_norm": 1.671875,
-      "learning_rate": 0.0005956481666013993,
-      "loss": 6.3373,
-      "step": 451
-    },
-    {
-      "epoch": 0.46936656282450673,
-      "grad_norm": 1.6171875,
-      "learning_rate": 0.0005939607478297346,
-      "loss": 6.2862,
-      "step": 452
-    },
-    {
-      "epoch": 0.470404984423676,
-      "grad_norm": 1.8984375,
-      "learning_rate": 0.0005922722189803378,
-      "loss": 6.253,
-      "step": 453
-    },
-    {
-      "epoch": 0.4714434060228453,
-      "grad_norm": 1.4140625,
-      "learning_rate": 0.0005905826000019458,
-      "loss": 6.3616,
-      "step": 454
-    },
-    {
-      "epoch": 0.4724818276220145,
-      "grad_norm": 4.125,
-      "learning_rate": 0.0005888919108561749,
-      "loss": 6.4911,
-      "step": 455
-    },
-    {
-      "epoch": 0.4735202492211838,
-      "grad_norm": 1.6953125,
-      "learning_rate": 0.0005872001715172853,
-      "loss": 6.3549,
-      "step": 456
-    },
-    {
-      "epoch": 0.4745586708203531,
-      "grad_norm": 1.796875,
-      "learning_rate": 0.0005855074019719433,
-      "loss": 6.3192,
-      "step": 457
-    },
-    {
-      "epoch": 0.4755970924195223,
-      "grad_norm": 1.5078125,
-      "learning_rate": 0.0005838136222189874,
-      "loss": 6.2864,
-      "step": 458
-    },
-    {
-      "epoch": 0.4766355140186916,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.0005821188522691903,
-      "loss": 6.4379,
-      "step": 459
-    },
-    {
-      "epoch": 0.47767393561786087,
-      "grad_norm": 1.25,
-      "learning_rate": 0.0005804231121450236,
-      "loss": 6.4266,
-      "step": 460
-    },
-    {
-      "epoch": 0.4787123572170301,
-      "grad_norm": 1.3125,
-      "learning_rate": 0.0005787264218804201,
-      "loss": 6.3983,
-      "step": 461
-    },
-    {
-      "epoch": 0.4797507788161994,
-      "grad_norm": 1.6015625,
-      "learning_rate": 0.0005770288015205385,
-      "loss": 6.3447,
-      "step": 462
-    },
-    {
-      "epoch": 0.48078920041536866,
-      "grad_norm": 1.984375,
-      "learning_rate": 0.0005753302711215259,
-      "loss": 6.1843,
-      "step": 463
-    },
-    {
-      "epoch": 0.4818276220145379,
-      "grad_norm": 1.984375,
-      "learning_rate": 0.0005736308507502804,
-      "loss": 6.4501,
-      "step": 464
-    },
-    {
-      "epoch": 0.48286604361370716,
-      "grad_norm": 1.484375,
-      "learning_rate": 0.0005719305604842148,
-      "loss": 6.3637,
-      "step": 465
-    },
-    {
-      "epoch": 0.48390446521287644,
-      "grad_norm": 1.6796875,
-      "learning_rate": 0.000570229420411019,
-      "loss": 6.2897,
-      "step": 466
-    },
-    {
-      "epoch": 0.48494288681204567,
-      "grad_norm": 1.640625,
-      "learning_rate": 0.0005685274506284226,
-      "loss": 6.3343,
-      "step": 467
-    },
-    {
-      "epoch": 0.48598130841121495,
-      "grad_norm": 1.3671875,
-      "learning_rate": 0.0005668246712439578,
-      "loss": 6.4351,
-      "step": 468
-    },
-    {
-      "epoch": 0.48701973001038423,
-      "grad_norm": 1.9765625,
-      "learning_rate": 0.0005651211023747216,
-      "loss": 6.0338,
-      "step": 469
-    },
-    {
-      "epoch": 0.48805815160955346,
-      "grad_norm": 1.421875,
-      "learning_rate": 0.0005634167641471383,
-      "loss": 6.1922,
-      "step": 470
-    },
-    {
-      "epoch": 0.48909657320872274,
-      "grad_norm": 1.3203125,
-      "learning_rate": 0.0005617116766967212,
-      "loss": 6.6165,
-      "step": 471
-    },
-    {
-      "epoch": 0.490134994807892,
-      "grad_norm": 2.515625,
-      "learning_rate": 0.0005600058601678358,
-      "loss": 6.311,
-      "step": 472
-    },
-    {
-      "epoch": 0.49117341640706125,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.0005582993347134603,
-      "loss": 6.4132,
-      "step": 473
-    },
-    {
-      "epoch": 0.49221183800623053,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.0005565921204949492,
-      "loss": 6.2585,
-      "step": 474
-    },
-    {
-      "epoch": 0.4932502596053998,
-      "grad_norm": 2.03125,
-      "learning_rate": 0.0005548842376817935,
-      "loss": 6.3609,
-      "step": 475
-    },
-    {
-      "epoch": 0.49428868120456904,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.0005531757064513835,
-      "loss": 6.3305,
-      "step": 476
-    },
-    {
-      "epoch": 0.4953271028037383,
-      "grad_norm": 1.421875,
-      "learning_rate": 0.0005514665469887705,
-      "loss": 6.2568,
-      "step": 477
-    },
-    {
-      "epoch": 0.4963655244029076,
-      "grad_norm": 1.75,
-      "learning_rate": 0.000549756779486427,
-      "loss": 6.2414,
-      "step": 478
-    },
-    {
-      "epoch": 0.4974039460020768,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.0005480464241440096,
-      "loss": 6.3746,
-      "step": 479
-    },
-    {
-      "epoch": 0.4984423676012461,
-      "grad_norm": 1.484375,
-      "learning_rate": 0.0005463355011681199,
-      "loss": 6.3126,
-      "step": 480
-    },
-    {
-      "epoch": 0.4994807892004154,
-      "grad_norm": 1.3359375,
-      "learning_rate": 0.0005446240307720653,
-      "loss": 6.3338,
-      "step": 481
-    },
-    {
-      "epoch": 0.5005192107995846,
-      "grad_norm": 1.3359375,
-      "learning_rate": 0.0005429120331756208,
-      "loss": 6.3524,
-      "step": 482
-    },
-    {
-      "epoch": 0.5015576323987538,
-      "grad_norm": 1.71875,
-      "learning_rate": 0.00054119952860479,
-      "loss": 6.2864,
-      "step": 483
-    },
-    {
-      "epoch": 0.5025960539979232,
-      "grad_norm": 1.734375,
-      "learning_rate": 0.0005394865372915656,
-      "loss": 6.3503,
-      "step": 484
-    },
-    {
-      "epoch": 0.5036344755970924,
-      "grad_norm": 1.78125,
-      "learning_rate": 0.0005377730794736914,
-      "loss": 6.1829,
-      "step": 485
-    },
-    {
-      "epoch": 0.5046728971962616,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.000536059175394422,
-      "loss": 6.2451,
-      "step": 486
-    },
-    {
-      "epoch": 0.505711318795431,
-      "grad_norm": 2.265625,
-      "learning_rate": 0.0005343448453022847,
-      "loss": 5.8713,
-      "step": 487
-    },
-    {
-      "epoch": 0.5067497403946002,
-      "grad_norm": 1.3203125,
-      "learning_rate": 0.0005326301094508399,
-      "loss": 6.2755,
-      "step": 488
-    },
-    {
-      "epoch": 0.5077881619937694,
-      "grad_norm": 7.125,
-      "learning_rate": 0.0005309149880984411,
-      "loss": 6.4164,
-      "step": 489
-    },
-    {
-      "epoch": 0.5088265835929388,
-      "grad_norm": 2.0,
-      "learning_rate": 0.0005291995015079969,
-      "loss": 6.4206,
-      "step": 490
-    },
-    {
-      "epoch": 0.509865005192108,
-      "grad_norm": 1.734375,
-      "learning_rate": 0.0005274836699467307,
-      "loss": 6.2462,
-      "step": 491
-    },
-    {
-      "epoch": 0.5109034267912772,
-      "grad_norm": 2.234375,
-      "learning_rate": 0.0005257675136859414,
-      "loss": 6.0642,
-      "step": 492
-    },
-    {
-      "epoch": 0.5119418483904465,
-      "grad_norm": 2.53125,
-      "learning_rate": 0.000524051053000764,
-      "loss": 6.1371,
-      "step": 493
-    },
-    {
-      "epoch": 0.5129802699896158,
-      "grad_norm": 1.875,
-      "learning_rate": 0.0005223343081699302,
-      "loss": 6.4626,
-      "step": 494
-    },
-    {
-      "epoch": 0.514018691588785,
-      "grad_norm": 2.703125,
-      "learning_rate": 0.0005206172994755284,
-      "loss": 6.3176,
-      "step": 495
-    },
-    {
-      "epoch": 0.5150571131879543,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.0005189000472027645,
-      "loss": 6.2687,
-      "step": 496
-    },
-    {
-      "epoch": 0.5160955347871236,
-      "grad_norm": 1.296875,
-      "learning_rate": 0.0005171825716397222,
-      "loss": 6.3934,
-      "step": 497
-    },
-    {
-      "epoch": 0.5171339563862928,
-      "grad_norm": 1.609375,
-      "learning_rate": 0.0005154648930771234,
-      "loss": 6.3839,
-      "step": 498
-    },
-    {
-      "epoch": 0.5181723779854621,
-      "grad_norm": 1.734375,
-      "learning_rate": 0.0005137470318080876,
-      "loss": 6.2297,
-      "step": 499
-    },
-    {
-      "epoch": 0.5192107995846313,
-      "grad_norm": 2.0625,
-      "learning_rate": 0.0005120290081278934,
-      "loss": 6.214,
-      "step": 500
-    },
-    {
-      "epoch": 0.5192107995846313,
-      "eval_loss": 6.333878040313721,
-      "eval_runtime": 1.7224,
-      "eval_samples_per_second": 9.289,
-      "eval_steps_per_second": 1.161,
-      "step": 500
-    },
-    {
-      "epoch": 0.5202492211838006,
-      "grad_norm": 1.6640625,
-      "learning_rate": 0.0005103108423337382,
-      "loss": 6.1978,
-      "step": 501
-    },
-    {
-      "epoch": 0.5212876427829699,
-      "grad_norm": 1.515625,
-      "learning_rate": 0.0005085925547244979,
-      "loss": 6.3541,
-      "step": 502
-    },
-    {
-      "epoch": 0.5223260643821391,
-      "grad_norm": 1.7265625,
-      "learning_rate": 0.000506874165600488,
-      "loss": 6.3855,
-      "step": 503
-    },
-    {
-      "epoch": 0.5233644859813084,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.0005051556952632235,
-      "loss": 6.1905,
-      "step": 504
-    },
-    {
-      "epoch": 0.5244029075804777,
-      "grad_norm": 1.7578125,
-      "learning_rate": 0.0005034371640151781,
-      "loss": 6.2397,
-      "step": 505
-    },
-    {
-      "epoch": 0.5254413291796469,
-      "grad_norm": 1.671875,
-      "learning_rate": 0.0005017185921595457,
-      "loss": 6.3751,
-      "step": 506
-    },
-    {
-      "epoch": 0.5264797507788161,
-      "grad_norm": 2.109375,
-      "learning_rate": 0.0005,
-      "loss": 6.5016,
-      "step": 507
-    },
-    {
-      "epoch": 0.5275181723779855,
-      "grad_norm": 1.5625,
-      "learning_rate": 0.0004982814078404544,
-      "loss": 6.1849,
-      "step": 508
-    },
-    {
-      "epoch": 0.5285565939771547,
-      "grad_norm": 3.90625,
-      "learning_rate": 0.000496562835984822,
-      "loss": 5.6967,
-      "step": 509
-    },
-    {
-      "epoch": 0.5295950155763239,
-      "grad_norm": 1.5078125,
-      "learning_rate": 0.0004948443047367768,
-      "loss": 6.3165,
-      "step": 510
-    },
-    {
-      "epoch": 0.5306334371754933,
-      "grad_norm": 1.3515625,
-      "learning_rate": 0.0004931258343995119,
-      "loss": 6.3457,
-      "step": 511
-    },
-    {
-      "epoch": 0.5316718587746625,
-      "grad_norm": 1.5546875,
-      "learning_rate": 0.0004914074452755021,
-      "loss": 6.2912,
-      "step": 512
-    },
-    {
-      "epoch": 0.5327102803738317,
-      "grad_norm": 2.875,
-      "learning_rate": 0.000489689157666262,
-      "loss": 6.436,
-      "step": 513
-    },
-    {
-      "epoch": 0.5337487019730011,
-      "grad_norm": 1.9140625,
-      "learning_rate": 0.0004879709918721067,
-      "loss": 6.388,
-      "step": 514
-    },
-    {
-      "epoch": 0.5347871235721703,
-      "grad_norm": 1.5390625,
-      "learning_rate": 0.0004862529681919125,
-      "loss": 6.3625,
-      "step": 515
-    },
-    {
-      "epoch": 0.5358255451713395,
-      "grad_norm": 1.421875,
-      "learning_rate": 0.0004845351069228767,
-      "loss": 6.3582,
-      "step": 516
-    },
-    {
-      "epoch": 0.5368639667705088,
-      "grad_norm": 1.3671875,
-      "learning_rate": 0.0004828174283602778,
-      "loss": 6.2882,
-      "step": 517
-    },
-    {
-      "epoch": 0.5379023883696781,
-      "grad_norm": 1.3515625,
-      "learning_rate": 0.0004810999527972355,
-      "loss": 6.3509,
-      "step": 518
-    },
-    {
-      "epoch": 0.5389408099688473,
-      "grad_norm": 1.8203125,
-      "learning_rate": 0.0004793827005244717,
-      "loss": 6.1571,
-      "step": 519
-    },
-    {
-      "epoch": 0.5399792315680166,
-      "grad_norm": 1.8984375,
-      "learning_rate": 0.0004776656918300699,
-      "loss": 6.4146,
-      "step": 520
-    },
-    {
-      "epoch": 0.5410176531671859,
-      "grad_norm": 1.8828125,
-      "learning_rate": 0.00047594894699923605,
-      "loss": 6.2866,
-      "step": 521
-    },
-    {
-      "epoch": 0.5420560747663551,
-      "grad_norm": 1.390625,
-      "learning_rate": 0.0004742324863140586,
-      "loss": 6.3875,
-      "step": 522
-    },
-    {
-      "epoch": 0.5430944963655244,
-      "grad_norm": 1.8203125,
-      "learning_rate": 0.0004725163300532693,
-      "loss": 6.3196,
-      "step": 523
-    },
-    {
-      "epoch": 0.5441329179646937,
-      "grad_norm": 1.796875,
-      "learning_rate": 0.000470800498492003,
-      "loss": 6.1896,
-      "step": 524
-    },
-    {
-      "epoch": 0.5451713395638629,
-      "grad_norm": 1.890625,
-      "learning_rate": 0.0004690850119015589,
-      "loss": 6.1498,
-      "step": 525
-    },
-    {
-      "epoch": 0.5462097611630322,
-      "grad_norm": 2.359375,
-      "learning_rate": 0.00046736989054916016,
-      "loss": 5.9399,
-      "step": 526
-    },
-    {
-      "epoch": 0.5472481827622014,
-      "grad_norm": 1.96875,
-      "learning_rate": 0.00046565515469771524,
-      "loss": 6.011,
-      "step": 527
-    },
-    {
-      "epoch": 0.5482866043613707,
-      "grad_norm": 1.4453125,
-      "learning_rate": 0.0004639408246055781,
-      "loss": 6.2529,
-      "step": 528
-    },
-    {
-      "epoch": 0.54932502596054,
-      "grad_norm": 1.375,
-      "learning_rate": 0.00046222692052630876,
-      "loss": 6.2216,
-      "step": 529
-    },
-    {
-      "epoch": 0.5503634475597092,
-      "grad_norm": 1.59375,
-      "learning_rate": 0.00046051346270843446,
-      "loss": 6.2293,
-      "step": 530
-    },
-    {
-      "epoch": 0.5514018691588785,
-      "grad_norm": 1.796875,
-      "learning_rate": 0.00045880047139521,
-      "loss": 6.2045,
-      "step": 531
-    },
-    {
-      "epoch": 0.5524402907580478,
-      "grad_norm": 1.5078125,
-      "learning_rate": 0.0004570879668243792,
-      "loss": 6.163,
-      "step": 532
-    },
-    {
-      "epoch": 0.553478712357217,
-      "grad_norm": 1.5234375,
-      "learning_rate": 0.0004553759692279347,
-      "loss": 6.2428,
-      "step": 533
-    },
-    {
-      "epoch": 0.5545171339563862,
-      "grad_norm": 1.484375,
-      "learning_rate": 0.00045366449883188015,
-      "loss": 6.3414,
-      "step": 534
-    },
-    {
-      "epoch": 0.5555555555555556,
-      "grad_norm": 1.6953125,
-      "learning_rate": 0.0004519535758559904,
-      "loss": 6.3549,
-      "step": 535
-    },
-    {
-      "epoch": 0.5565939771547248,
-      "grad_norm": 1.609375,
-      "learning_rate": 0.0004502432205135731,
-      "loss": 6.205,
-      "step": 536
-    },
-    {
-      "epoch": 0.557632398753894,
-      "grad_norm": 2.78125,
-      "learning_rate": 0.0004485334530112297,
-      "loss": 6.3195,
-      "step": 537
-    },
-    {
-      "epoch": 0.5586708203530634,
-      "grad_norm": 1.359375,
-      "learning_rate": 0.00044682429354861637,
-      "loss": 6.1937,
-      "step": 538
-    },
-    {
-      "epoch": 0.5597092419522326,
-      "grad_norm": 1.8671875,
-      "learning_rate": 0.00044511576231820655,
-      "loss": 6.0656,
-      "step": 539
-    },
-    {
-      "epoch": 0.5607476635514018,
-      "grad_norm": 1.9296875,
-      "learning_rate": 0.0004434078795050509,
-      "loss": 6.3324,
-      "step": 540
-    },
-    {
-      "epoch": 0.5617860851505712,
-      "grad_norm": 1.7890625,
-      "learning_rate": 0.00044170066528653973,
-      "loss": 6.2444,
-      "step": 541
-    },
-    {
-      "epoch": 0.5628245067497404,
-      "grad_norm": 1.5,
-      "learning_rate": 0.0004399941398321644,
-      "loss": 6.2952,
-      "step": 542
-    },
-    {
-      "epoch": 0.5638629283489096,
-      "grad_norm": 1.5390625,
-      "learning_rate": 0.00043828832330327895,
-      "loss": 6.3273,
-      "step": 543
-    },
-    {
-      "epoch": 0.564901349948079,
-      "grad_norm": 1.9296875,
-      "learning_rate": 0.00043658323585286175,
-      "loss": 6.1395,
-      "step": 544
-    },
-    {
-      "epoch": 0.5659397715472482,
-      "grad_norm": 1.90625,
-      "learning_rate": 0.00043487889762527834,
-      "loss": 6.4269,
-      "step": 545
-    },
-    {
-      "epoch": 0.5669781931464174,
-      "grad_norm": 1.40625,
-      "learning_rate": 0.0004331753287560422,
-      "loss": 6.3326,
-      "step": 546
-    },
-    {
-      "epoch": 0.5680166147455867,
-      "grad_norm": 1.6328125,
-      "learning_rate": 0.00043147254937157744,
-      "loss": 6.2105,
-      "step": 547
-    },
-    {
-      "epoch": 0.569055036344756,
-      "grad_norm": 4.3125,
-      "learning_rate": 0.00042977057958898103,
-      "loss": 6.5283,
-      "step": 548
-    },
-    {
-      "epoch": 0.5700934579439252,
-      "grad_norm": 1.8046875,
-      "learning_rate": 0.0004280694395157853,
-      "loss": 6.3104,
-      "step": 549
-    },
-    {
-      "epoch": 0.5711318795430945,
-      "grad_norm": 1.5546875,
-      "learning_rate": 0.0004263691492497197,
-      "loss": 6.2183,
-      "step": 550
-    },
-    {
-      "epoch": 0.5721703011422637,
-      "grad_norm": 1.609375,
-      "learning_rate": 0.00042466972887847404,
-      "loss": 6.276,
-      "step": 551
-    },
-    {
-      "epoch": 0.573208722741433,
-      "grad_norm": 1.40625,
-      "learning_rate": 0.0004229711984794614,
-      "loss": 6.1446,
-      "step": 552
-    },
-    {
-      "epoch": 0.5742471443406023,
-      "grad_norm": 1.5,
-      "learning_rate": 0.00042127357811958,
-      "loss": 6.3867,
-      "step": 553
-    },
-    {
-      "epoch": 0.5752855659397715,
-      "grad_norm": 1.453125,
-      "learning_rate": 0.0004195768878549766,
-      "loss": 6.4525,
-      "step": 554
-    },
-    {
-      "epoch": 0.5763239875389408,
-      "grad_norm": 1.203125,
-      "learning_rate": 0.00041788114773080984,
-      "loss": 6.3332,
-      "step": 555
-    },
-    {
-      "epoch": 0.5773624091381101,
-      "grad_norm": 1.21875,
-      "learning_rate": 0.0004161863777810128,
-      "loss": 6.2988,
-      "step": 556
-    },
-    {
-      "epoch": 0.5784008307372793,
-      "grad_norm": 1.2890625,
-      "learning_rate": 0.00041449259802805685,
-      "loss": 6.3041,
-      "step": 557
-    },
-    {
-      "epoch": 0.5794392523364486,
-      "grad_norm": 1.4140625,
-      "learning_rate": 0.0004127998284827148,
-      "loss": 6.1669,
-      "step": 558
-    },
-    {
-      "epoch": 0.5804776739356179,
-      "grad_norm": 1.8125,
-      "learning_rate": 0.00041110808914382506,
-      "loss": 6.033,
-      "step": 559
-    },
-    {
-      "epoch": 0.5815160955347871,
-      "grad_norm": 1.4375,
-      "learning_rate": 0.0004094173999980544,
-      "loss": 6.3417,
-      "step": 560
-    },
-    {
-      "epoch": 0.5825545171339563,
-      "grad_norm": 2.015625,
-      "learning_rate": 0.00040772778101966234,
-      "loss": 6.3482,
-      "step": 561
-    },
-    {
-      "epoch": 0.5835929387331257,
-      "grad_norm": 1.4453125,
-      "learning_rate": 0.00040603925217026543,
-      "loss": 6.1923,
-      "step": 562
-    },
-    {
-      "epoch": 0.5846313603322949,
-      "grad_norm": 2.125,
-      "learning_rate": 0.00040435183339860084,
-      "loss": 6.0967,
-      "step": 563
-    },
-    {
-      "epoch": 0.5856697819314641,
-      "grad_norm": 1.71875,
-      "learning_rate": 0.00040266554464029117,
-      "loss": 6.1855,
-      "step": 564
-    },
-    {
-      "epoch": 0.5867082035306335,
-      "grad_norm": 1.234375,
-      "learning_rate": 0.00040098040581760836,
-      "loss": 6.2101,
-      "step": 565
-    },
-    {
-      "epoch": 0.5877466251298027,
-      "grad_norm": 1.1875,
-      "learning_rate": 0.0003992964368392397,
-      "loss": 6.3314,
-      "step": 566
-    },
-    {
-      "epoch": 0.5887850467289719,
-      "grad_norm": 1.6015625,
-      "learning_rate": 0.00039761365760005053,
-      "loss": 6.1649,
-      "step": 567
-    },
-    {
-      "epoch": 0.5898234683281413,
-      "grad_norm": 1.40625,
-      "learning_rate": 0.0003959320879808509,
-      "loss": 6.3325,
-      "step": 568
-    },
-    {
-      "epoch": 0.5908618899273105,
-      "grad_norm": 1.7578125,
-      "learning_rate": 0.00039425174784815973,
-      "loss": 6.1956,
-      "step": 569
-    },
-    {
-      "epoch": 0.5919003115264797,
-      "grad_norm": 1.578125,
-      "learning_rate": 0.0003925726570539706,
-      "loss": 6.3196,
-      "step": 570
-    },
-    {
-      "epoch": 0.592938733125649,
-      "grad_norm": 1.6796875,
-      "learning_rate": 0.000390894835435517,
-      "loss": 6.2539,
-      "step": 571
-    },
-    {
-      "epoch": 0.5939771547248183,
-      "grad_norm": 1.4140625,
-      "learning_rate": 0.00038921830281503844,
-      "loss": 6.2395,
-      "step": 572
-    },
-    {
-      "epoch": 0.5950155763239875,
-      "grad_norm": 1.9296875,
-      "learning_rate": 0.00038754307899954536,
-      "loss": 6.3062,
-      "step": 573
-    },
-    {
-      "epoch": 0.5960539979231568,
-      "grad_norm": 1.9921875,
-      "learning_rate": 0.0003858691837805859,
-      "loss": 6.1366,
-      "step": 574
-    },
-    {
-      "epoch": 0.5970924195223261,
-      "grad_norm": 1.8984375,
-      "learning_rate": 0.0003841966369340115,
-      "loss": 6.1059,
-      "step": 575
-    },
-    {
-      "epoch": 0.5981308411214953,
-      "grad_norm": 1.8125,
-      "learning_rate": 0.0003825254582197438,
-      "loss": 6.0075,
-      "step": 576
-    },
-    {
-      "epoch": 0.5991692627206646,
-      "grad_norm": 1.8046875,
-      "learning_rate": 0.000380855667381541,
-      "loss": 6.1437,
-      "step": 577
-    },
-    {
-      "epoch": 0.6002076843198338,
-      "grad_norm": 1.59375,
-      "learning_rate": 0.0003791872841467643,
-      "loss": 6.3531,
-      "step": 578
-    },
-    {
-      "epoch": 0.6012461059190031,
-      "grad_norm": 2.5,
-      "learning_rate": 0.00037752032822614554,
-      "loss": 6.2369,
-      "step": 579
-    },
-    {
-      "epoch": 0.6022845275181724,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.0003758548193135536,
-      "loss": 6.2403,
-      "step": 580
-    },
-    {
-      "epoch": 0.6033229491173416,
-      "grad_norm": 1.40625,
-      "learning_rate": 0.0003741907770857618,
-      "loss": 6.2333,
-      "step": 581
-    },
-    {
-      "epoch": 0.6043613707165109,
-      "grad_norm": 1.4375,
-      "learning_rate": 0.0003725282212022159,
-      "loss": 6.2413,
-      "step": 582
-    },
-    {
-      "epoch": 0.6053997923156802,
-      "grad_norm": 1.6171875,
-      "learning_rate": 0.0003708671713048017,
-      "loss": 6.4416,
-      "step": 583
-    },
-    {
-      "epoch": 0.6064382139148494,
-      "grad_norm": 1.484375,
-      "learning_rate": 0.0003692076470176126,
-      "loss": 6.0833,
-      "step": 584
-    },
-    {
-      "epoch": 0.6074766355140186,
-      "grad_norm": 1.78125,
-      "learning_rate": 0.000367549667946718,
-      "loss": 6.1351,
-      "step": 585
-    },
-    {
-      "epoch": 0.608515057113188,
-      "grad_norm": 1.3828125,
-      "learning_rate": 0.0003658932536799324,
-      "loss": 6.3859,
-      "step": 586
-    },
-    {
-      "epoch": 0.6095534787123572,
-      "grad_norm": 1.2734375,
-      "learning_rate": 0.0003642384237865823,
-      "loss": 6.3225,
-      "step": 587
-    },
-    {
-      "epoch": 0.6105919003115264,
-      "grad_norm": 1.7421875,
-      "learning_rate": 0.0003625851978172765,
-      "loss": 6.2748,
-      "step": 588
-    },
-    {
-      "epoch": 0.6116303219106958,
-      "grad_norm": 1.5625,
-      "learning_rate": 0.00036093359530367454,
-      "loss": 6.2779,
-      "step": 589
-    },
-    {
-      "epoch": 0.612668743509865,
-      "grad_norm": 1.3125,
-      "learning_rate": 0.00035928363575825595,
-      "loss": 6.2115,
-      "step": 590
-    },
-    {
-      "epoch": 0.6137071651090342,
-      "grad_norm": 1.59375,
-      "learning_rate": 0.0003576353386740899,
-      "loss": 6.2217,
-      "step": 591
-    },
-    {
-      "epoch": 0.6147455867082036,
-      "grad_norm": 1.5078125,
-      "learning_rate": 0.00035598872352460455,
-      "loss": 6.3422,
-      "step": 592
-    },
-    {
-      "epoch": 0.6157840083073728,
-      "grad_norm": 1.8984375,
-      "learning_rate": 0.0003543438097633577,
-      "loss": 6.0398,
-      "step": 593
-    },
-    {
-      "epoch": 0.616822429906542,
-      "grad_norm": 1.90625,
-      "learning_rate": 0.00035270061682380607,
-      "loss": 6.2974,
-      "step": 594
-    },
-    {
-      "epoch": 0.6178608515057114,
-      "grad_norm": 1.453125,
-      "learning_rate": 0.0003510591641190762,
-      "loss": 6.2733,
-      "step": 595
-    },
-    {
-      "epoch": 0.6188992731048806,
-      "grad_norm": 1.7578125,
-      "learning_rate": 0.0003494194710417351,
-      "loss": 6.2583,
-      "step": 596
-    },
-    {
-      "epoch": 0.6199376947040498,
-      "grad_norm": 1.65625,
-      "learning_rate": 0.0003477815569635611,
-      "loss": 5.9226,
-      "step": 597
-    },
-    {
-      "epoch": 0.6209761163032191,
-      "grad_norm": 1.4765625,
-      "learning_rate": 0.00034614544123531474,
-      "loss": 6.2218,
-      "step": 598
-    },
-    {
-      "epoch": 0.6220145379023884,
-      "grad_norm": 2.015625,
-      "learning_rate": 0.00034451114318651064,
-      "loss": 6.2521,
-      "step": 599
-    },
-    {
-      "epoch": 0.6230529595015576,
-      "grad_norm": 1.6015625,
-      "learning_rate": 0.0003428786821251888,
-      "loss": 6.2548,
-      "step": 600
-    },
-    {
-      "epoch": 0.6230529595015576,
-      "eval_loss": 6.287681579589844,
-      "eval_runtime": 1.6799,
-      "eval_samples_per_second": 9.525,
-      "eval_steps_per_second": 1.191,
-      "step": 600
-    },
-    {
-      "epoch": 0.6240913811007269,
-      "grad_norm": 1.609375,
-      "learning_rate": 0.0003412480773376864,
-      "loss": 6.2849,
-      "step": 601
-    },
-    {
-      "epoch": 0.6251298026998962,
-      "grad_norm": 1.5234375,
-      "learning_rate": 0.00033961934808841024,
-      "loss": 6.1491,
-      "step": 602
-    },
-    {
-      "epoch": 0.6261682242990654,
-      "grad_norm": 1.3203125,
-      "learning_rate": 0.0003379925136196088,
-      "loss": 6.281,
-      "step": 603
-    },
-    {
-      "epoch": 0.6272066458982347,
-      "grad_norm": 1.5625,
-      "learning_rate": 0.0003363675931511455,
-      "loss": 6.2496,
-      "step": 604
-    },
-    {
-      "epoch": 0.6282450674974039,
-      "grad_norm": 1.3046875,
-      "learning_rate": 0.0003347446058802708,
-      "loss": 6.2776,
-      "step": 605
-    },
-    {
-      "epoch": 0.6292834890965732,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.00033312357098139617,
-      "loss": 6.3928,
-      "step": 606
-    },
-    {
-      "epoch": 0.6303219106957425,
-      "grad_norm": 1.34375,
-      "learning_rate": 0.0003315045076058671,
-      "loss": 6.2803,
-      "step": 607
-    },
-    {
-      "epoch": 0.6313603322949117,
-      "grad_norm": 1.4296875,
-      "learning_rate": 0.00032988743488173697,
-      "loss": 6.2263,
-      "step": 608
-    },
-    {
-      "epoch": 0.632398753894081,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.000328272371913541,
-      "loss": 6.1383,
-      "step": 609
-    },
-    {
-      "epoch": 0.6334371754932503,
-      "grad_norm": 1.265625,
-      "learning_rate": 0.0003266593377820708,
-      "loss": 6.2603,
-      "step": 610
-    },
-    {
-      "epoch": 0.6344755970924195,
-      "grad_norm": 2.25,
-      "learning_rate": 0.0003250483515441485,
-      "loss": 6.3359,
-      "step": 611
-    },
-    {
-      "epoch": 0.6355140186915887,
-      "grad_norm": 1.6015625,
-      "learning_rate": 0.0003234394322324019,
-      "loss": 6.1653,
-      "step": 612
-    },
-    {
-      "epoch": 0.6365524402907581,
-      "grad_norm": 2.125,
-      "learning_rate": 0.00032183259885504,
-      "loss": 6.2869,
-      "step": 613
-    },
-    {
-      "epoch": 0.6375908618899273,
-      "grad_norm": 1.3515625,
-      "learning_rate": 0.00032022787039562745,
-      "loss": 6.3017,
-      "step": 614
-    },
-    {
-      "epoch": 0.6386292834890965,
-      "grad_norm": 4.3125,
-      "learning_rate": 0.0003186252658128611,
-      "loss": 6.1045,
-      "step": 615
-    },
-    {
-      "epoch": 0.6396677050882659,
-      "grad_norm": 1.6171875,
-      "learning_rate": 0.00031702480404034565,
-      "loss": 6.3121,
-      "step": 616
-    },
-    {
-      "epoch": 0.6407061266874351,
-      "grad_norm": 1.8515625,
-      "learning_rate": 0.00031542650398637016,
-      "loss": 6.1043,
-      "step": 617
-    },
-    {
-      "epoch": 0.6417445482866043,
-      "grad_norm": 1.4375,
-      "learning_rate": 0.0003138303845336844,
-      "loss": 6.2402,
-      "step": 618
-    },
-    {
-      "epoch": 0.6427829698857737,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.0003122364645392762,
-      "loss": 6.2972,
-      "step": 619
-    },
-    {
-      "epoch": 0.6438213914849429,
-      "grad_norm": 2.234375,
-      "learning_rate": 0.0003106447628341482,
-      "loss": 6.2454,
-      "step": 620
-    },
-    {
-      "epoch": 0.6448598130841121,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.0003090552982230954,
-      "loss": 6.1745,
-      "step": 621
-    },
-    {
-      "epoch": 0.6458982346832814,
-      "grad_norm": 1.3984375,
-      "learning_rate": 0.00030746808948448366,
-      "loss": 6.224,
-      "step": 622
-    },
-    {
-      "epoch": 0.6469366562824507,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.0003058831553700268,
-      "loss": 6.3142,
-      "step": 623
-    },
-    {
-      "epoch": 0.6479750778816199,
-      "grad_norm": 1.9140625,
-      "learning_rate": 0.00030430051460456596,
-      "loss": 6.2258,
-      "step": 624
-    },
-    {
-      "epoch": 0.6490134994807892,
-      "grad_norm": 1.6171875,
-      "learning_rate": 0.0003027201858858479,
-      "loss": 6.1406,
-      "step": 625
-    },
-    {
-      "epoch": 0.6500519210799585,
-      "grad_norm": 1.59375,
-      "learning_rate": 0.00030114218788430437,
-      "loss": 5.7233,
-      "step": 626
-    },
-    {
-      "epoch": 0.6510903426791277,
-      "grad_norm": 1.5234375,
-      "learning_rate": 0.0002995665392428313,
-      "loss": 6.0472,
-      "step": 627
-    },
-    {
-      "epoch": 0.652128764278297,
-      "grad_norm": 2.265625,
-      "learning_rate": 0.00029799325857656855,
-      "loss": 6.161,
-      "step": 628
-    },
-    {
-      "epoch": 0.6531671858774662,
-      "grad_norm": 1.53125,
-      "learning_rate": 0.00029642236447268024,
-      "loss": 6.112,
-      "step": 629
-    },
-    {
-      "epoch": 0.6542056074766355,
-      "grad_norm": 1.4140625,
-      "learning_rate": 0.00029485387549013485,
-      "loss": 6.2132,
-      "step": 630
-    },
-    {
-      "epoch": 0.6552440290758048,
-      "grad_norm": 1.1875,
-      "learning_rate": 0.00029328781015948625,
-      "loss": 6.2657,
-      "step": 631
-    },
-    {
-      "epoch": 0.656282450674974,
-      "grad_norm": 1.5078125,
-      "learning_rate": 0.00029172418698265444,
-      "loss": 6.1711,
-      "step": 632
-    },
-    {
-      "epoch": 0.6573208722741433,
-      "grad_norm": 1.5859375,
-      "learning_rate": 0.0002901630244327075,
-      "loss": 6.1935,
-      "step": 633
-    },
-    {
-      "epoch": 0.6583592938733126,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.00028860434095364263,
-      "loss": 6.4055,
-      "step": 634
-    },
-    {
-      "epoch": 0.6593977154724818,
-      "grad_norm": 1.96875,
-      "learning_rate": 0.00028704815496016875,
-      "loss": 5.9916,
-      "step": 635
-    },
-    {
-      "epoch": 0.660436137071651,
-      "grad_norm": 1.3359375,
-      "learning_rate": 0.00028549448483748886,
-      "loss": 6.1303,
-      "step": 636
-    },
-    {
-      "epoch": 0.6614745586708204,
-      "grad_norm": 1.65625,
-      "learning_rate": 0.0002839433489410828,
-      "loss": 6.1363,
-      "step": 637
-    },
-    {
-      "epoch": 0.6625129802699896,
-      "grad_norm": 1.734375,
-      "learning_rate": 0.0002823947655964901,
-      "loss": 6.1624,
-      "step": 638
-    },
-    {
-      "epoch": 0.6635514018691588,
-      "grad_norm": 1.453125,
-      "learning_rate": 0.000280848753099094,
-      "loss": 5.9139,
-      "step": 639
-    },
-    {
-      "epoch": 0.6645898234683282,
-      "grad_norm": 2.625,
-      "learning_rate": 0.0002793053297139054,
-      "loss": 6.3661,
-      "step": 640
-    },
-    {
-      "epoch": 0.6656282450674974,
-      "grad_norm": 1.7578125,
-      "learning_rate": 0.0002777645136753459,
-      "loss": 6.3797,
-      "step": 641
-    },
-    {
-      "epoch": 0.6666666666666666,
-      "grad_norm": 1.8671875,
-      "learning_rate": 0.0002762263231870339,
-      "loss": 6.0671,
-      "step": 642
-    },
-    {
-      "epoch": 0.667705088265836,
-      "grad_norm": 1.6328125,
-      "learning_rate": 0.00027469077642156844,
-      "loss": 6.1843,
-      "step": 643
-    },
-    {
-      "epoch": 0.6687435098650052,
-      "grad_norm": 1.625,
-      "learning_rate": 0.000273157891520315,
-      "loss": 6.4473,
-      "step": 644
-    },
-    {
-      "epoch": 0.6697819314641744,
-      "grad_norm": 1.6953125,
-      "learning_rate": 0.00027162768659319114,
-      "loss": 6.3164,
-      "step": 645
-    },
-    {
-      "epoch": 0.6708203530633438,
-      "grad_norm": 1.6328125,
-      "learning_rate": 0.00027010017971845264,
-      "loss": 6.4004,
-      "step": 646
-    },
-    {
-      "epoch": 0.671858774662513,
-      "grad_norm": 1.6171875,
-      "learning_rate": 0.00026857538894247947,
-      "loss": 6.242,
-      "step": 647
-    },
-    {
-      "epoch": 0.6728971962616822,
-      "grad_norm": 1.28125,
-      "learning_rate": 0.00026705333227956303,
-      "loss": 6.2597,
-      "step": 648
-    },
-    {
-      "epoch": 0.6739356178608515,
-      "grad_norm": 1.2734375,
-      "learning_rate": 0.000265534027711693,
-      "loss": 6.3972,
-      "step": 649
-    },
-    {
-      "epoch": 0.6749740394600208,
-      "grad_norm": 1.90625,
-      "learning_rate": 0.00026401749318834527,
-      "loss": 6.4521,
-      "step": 650
-    },
-    {
-      "epoch": 0.67601246105919,
-      "grad_norm": 1.40625,
-      "learning_rate": 0.0002625037466262696,
-      "loss": 5.9594,
-      "step": 651
-    },
-    {
-      "epoch": 0.6770508826583593,
-      "grad_norm": 1.6015625,
-      "learning_rate": 0.0002609928059092779,
-      "loss": 6.2181,
-      "step": 652
-    },
-    {
-      "epoch": 0.6780893042575286,
-      "grad_norm": 1.3828125,
-      "learning_rate": 0.00025948468888803324,
-      "loss": 6.2781,
-      "step": 653
-    },
-    {
-      "epoch": 0.6791277258566978,
-      "grad_norm": 1.65625,
-      "learning_rate": 0.00025797941337983875,
-      "loss": 6.1757,
-      "step": 654
-    },
-    {
-      "epoch": 0.6801661474558671,
-      "grad_norm": 1.4609375,
-      "learning_rate": 0.0002564769971684271,
-      "loss": 6.266,
-      "step": 655
-    },
-    {
-      "epoch": 0.6812045690550363,
-      "grad_norm": 1.390625,
-      "learning_rate": 0.00025497745800375036,
-      "loss": 6.3151,
-      "step": 656
-    },
-    {
-      "epoch": 0.6822429906542056,
-      "grad_norm": 1.3671875,
-      "learning_rate": 0.0002534808136017707,
-      "loss": 6.2002,
-      "step": 657
-    },
-    {
-      "epoch": 0.6832814122533749,
-      "grad_norm": 1.3671875,
-      "learning_rate": 0.00025198708164425045,
-      "loss": 6.1243,
-      "step": 658
-    },
-    {
-      "epoch": 0.6843198338525441,
-      "grad_norm": 1.3671875,
-      "learning_rate": 0.0002504962797785435,
-      "loss": 6.2666,
-      "step": 659
-    },
-    {
-      "epoch": 0.6853582554517134,
-      "grad_norm": 4.4375,
-      "learning_rate": 0.00024900842561738736,
-      "loss": 5.9076,
-      "step": 660
-    },
-    {
-      "epoch": 0.6863966770508827,
-      "grad_norm": 1.484375,
-      "learning_rate": 0.00024752353673869405,
-      "loss": 6.1968,
-      "step": 661
-    },
-    {
-      "epoch": 0.6874350986500519,
-      "grad_norm": 1.203125,
-      "learning_rate": 0.00024604163068534315,
-      "loss": 6.2919,
-      "step": 662
-    },
-    {
-      "epoch": 0.6884735202492211,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.00024456272496497415,
-      "loss": 6.1599,
-      "step": 663
-    },
-    {
-      "epoch": 0.6895119418483905,
-      "grad_norm": 1.8984375,
-      "learning_rate": 0.00024308683704978002,
-      "loss": 6.2052,
-      "step": 664
-    },
-    {
-      "epoch": 0.6905503634475597,
-      "grad_norm": 1.375,
-      "learning_rate": 0.00024161398437630045,
-      "loss": 6.3025,
-      "step": 665
-    },
-    {
-      "epoch": 0.6915887850467289,
-      "grad_norm": 1.296875,
-      "learning_rate": 0.0002401441843452159,
-      "loss": 6.1938,
-      "step": 666
-    },
-    {
-      "epoch": 0.6926272066458983,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.0002386774543211423,
-      "loss": 6.3049,
-      "step": 667
-    },
-    {
-      "epoch": 0.6936656282450675,
-      "grad_norm": 1.8515625,
-      "learning_rate": 0.0002372138116324254,
-      "loss": 6.3624,
-      "step": 668
-    },
-    {
-      "epoch": 0.6947040498442367,
-      "grad_norm": 1.84375,
-      "learning_rate": 0.00023575327357093658,
-      "loss": 6.3294,
-      "step": 669
-    },
-    {
-      "epoch": 0.6957424714434061,
-      "grad_norm": 1.296875,
-      "learning_rate": 0.0002342958573918682,
-      "loss": 6.097,
-      "step": 670
-    },
-    {
-      "epoch": 0.6967808930425753,
-      "grad_norm": 1.8203125,
-      "learning_rate": 0.0002328415803135298,
-      "loss": 6.1121,
-      "step": 671
-    },
-    {
-      "epoch": 0.6978193146417445,
-      "grad_norm": 1.5859375,
-      "learning_rate": 0.0002313904595171447,
-      "loss": 6.1288,
-      "step": 672
-    },
-    {
-      "epoch": 0.6988577362409139,
-      "grad_norm": 1.5,
-      "learning_rate": 0.0002299425121466475,
-      "loss": 6.0291,
-      "step": 673
-    },
-    {
-      "epoch": 0.6998961578400831,
-      "grad_norm": 1.5,
-      "learning_rate": 0.00022849775530848056,
-      "loss": 6.3553,
-      "step": 674
-    },
-    {
-      "epoch": 0.7009345794392523,
-      "grad_norm": 1.4765625,
-      "learning_rate": 0.00022705620607139254,
-      "loss": 6.2154,
-      "step": 675
-    },
-    {
-      "epoch": 0.7019730010384216,
-      "grad_norm": 1.3984375,
-      "learning_rate": 0.00022561788146623679,
-      "loss": 6.1831,
-      "step": 676
-    },
-    {
-      "epoch": 0.7030114226375909,
-      "grad_norm": 1.34375,
-      "learning_rate": 0.0002241827984857698,
-      "loss": 6.1788,
-      "step": 677
-    },
-    {
-      "epoch": 0.7040498442367601,
-      "grad_norm": 1.6875,
-      "learning_rate": 0.00022275097408445076,
-      "loss": 6.3325,
-      "step": 678
-    },
-    {
-      "epoch": 0.7050882658359294,
-      "grad_norm": 1.3203125,
-      "learning_rate": 0.00022132242517824115,
-      "loss": 6.2826,
-      "step": 679
-    },
-    {
-      "epoch": 0.7061266874350987,
-      "grad_norm": 1.3359375,
-      "learning_rate": 0.0002198971686444047,
-      "loss": 6.1395,
-      "step": 680
-    },
-    {
-      "epoch": 0.7071651090342679,
-      "grad_norm": 1.4296875,
-      "learning_rate": 0.00021847522132130827,
-      "loss": 6.3967,
-      "step": 681
-    },
-    {
-      "epoch": 0.7082035306334372,
-      "grad_norm": 2.234375,
-      "learning_rate": 0.00021705660000822285,
-      "loss": 6.4163,
-      "step": 682
-    },
-    {
-      "epoch": 0.7092419522326064,
-      "grad_norm": 1.5546875,
-      "learning_rate": 0.00021564132146512495,
-      "loss": 6.1454,
-      "step": 683
-    },
-    {
-      "epoch": 0.7102803738317757,
-      "grad_norm": 1.9296875,
-      "learning_rate": 0.00021422940241249872,
-      "loss": 6.2889,
-      "step": 684
-    },
-    {
-      "epoch": 0.711318795430945,
-      "grad_norm": 1.4765625,
-      "learning_rate": 0.0002128208595311384,
-      "loss": 6.227,
-      "step": 685
-    },
-    {
-      "epoch": 0.7123572170301142,
-      "grad_norm": 1.5078125,
-      "learning_rate": 0.00021141570946195105,
-      "loss": 6.2655,
-      "step": 686
-    },
-    {
-      "epoch": 0.7133956386292835,
-      "grad_norm": 1.4140625,
-      "learning_rate": 0.00021001396880576063,
-      "loss": 6.1813,
-      "step": 687
-    },
-    {
-      "epoch": 0.7144340602284528,
-      "grad_norm": 1.1875,
-      "learning_rate": 0.0002086156541231109,
-      "loss": 6.3148,
-      "step": 688
-    },
-    {
-      "epoch": 0.715472481827622,
-      "grad_norm": 1.6015625,
-      "learning_rate": 0.00020722078193407035,
-      "loss": 6.2127,
-      "step": 689
-    },
-    {
-      "epoch": 0.7165109034267912,
-      "grad_norm": 1.53125,
-      "learning_rate": 0.00020582936871803693,
-      "loss": 6.2863,
-      "step": 690
-    },
-    {
-      "epoch": 0.7175493250259606,
-      "grad_norm": 1.203125,
-      "learning_rate": 0.0002044414309135434,
-      "loss": 6.236,
-      "step": 691
-    },
-    {
-      "epoch": 0.7185877466251298,
-      "grad_norm": 1.5546875,
-      "learning_rate": 0.00020305698491806295,
-      "loss": 6.3918,
-      "step": 692
-    },
-    {
-      "epoch": 0.719626168224299,
-      "grad_norm": 1.59375,
-      "learning_rate": 0.0002016760470878158,
-      "loss": 5.9227,
-      "step": 693
-    },
-    {
-      "epoch": 0.7206645898234684,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.0002002986337375755,
-      "loss": 6.0943,
-      "step": 694
-    },
-    {
-      "epoch": 0.7217030114226376,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.00019892476114047664,
-      "loss": 6.2113,
-      "step": 695
-    },
-    {
-      "epoch": 0.7227414330218068,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.00019755444552782225,
-      "loss": 6.3502,
-      "step": 696
-    },
-    {
-      "epoch": 0.7237798546209762,
-      "grad_norm": 1.515625,
-      "learning_rate": 0.00019618770308889227,
-      "loss": 5.9658,
-      "step": 697
-    },
-    {
-      "epoch": 0.7248182762201454,
-      "grad_norm": 1.5,
-      "learning_rate": 0.00019482454997075228,
-      "loss": 6.1518,
-      "step": 698
-    },
-    {
-      "epoch": 0.7258566978193146,
-      "grad_norm": 1.4609375,
-      "learning_rate": 0.00019346500227806218,
-      "loss": 6.1651,
-      "step": 699
-    },
-    {
-      "epoch": 0.726895119418484,
-      "grad_norm": 1.359375,
-      "learning_rate": 0.00019210907607288723,
-      "loss": 6.2656,
-      "step": 700
-    },
-    {
-      "epoch": 0.726895119418484,
-      "eval_loss": 6.260857105255127,
-      "eval_runtime": 1.646,
-      "eval_samples_per_second": 9.72,
-      "eval_steps_per_second": 1.215,
-      "step": 700
-    },
-    {
-      "epoch": 0.7279335410176532,
-      "grad_norm": 1.7109375,
-      "learning_rate": 0.00019075678737450686,
-      "loss": 6.1547,
-      "step": 701
-    },
-    {
-      "epoch": 0.7289719626168224,
-      "grad_norm": 1.7734375,
-      "learning_rate": 0.00018940815215922607,
-      "loss": 6.0599,
-      "step": 702
-    },
-    {
-      "epoch": 0.7300103842159917,
-      "grad_norm": 1.6171875,
-      "learning_rate": 0.00018806318636018665,
-      "loss": 6.2195,
-      "step": 703
-    },
-    {
-      "epoch": 0.731048805815161,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.00018672190586717908,
-      "loss": 6.4289,
-      "step": 704
-    },
-    {
-      "epoch": 0.7320872274143302,
-      "grad_norm": 1.203125,
-      "learning_rate": 0.00018538432652645437,
-      "loss": 6.2451,
-      "step": 705
-    },
-    {
-      "epoch": 0.7331256490134995,
-      "grad_norm": 1.1640625,
-      "learning_rate": 0.00018405046414053728,
-      "loss": 6.2281,
-      "step": 706
-    },
-    {
-      "epoch": 0.7341640706126688,
-      "grad_norm": 1.484375,
-      "learning_rate": 0.00018272033446803949,
-      "loss": 6.2168,
-      "step": 707
-    },
-    {
-      "epoch": 0.735202492211838,
-      "grad_norm": 1.7890625,
-      "learning_rate": 0.00018139395322347334,
-      "loss": 6.1276,
-      "step": 708
-    },
-    {
-      "epoch": 0.7362409138110073,
-      "grad_norm": 1.4453125,
-      "learning_rate": 0.00018007133607706615,
-      "loss": 6.0438,
-      "step": 709
-    },
-    {
-      "epoch": 0.7372793354101765,
-      "grad_norm": 1.875,
-      "learning_rate": 0.00017875249865457527,
-      "loss": 6.0624,
-      "step": 710
-    },
-    {
-      "epoch": 0.7383177570093458,
-      "grad_norm": 1.671875,
-      "learning_rate": 0.00017743745653710336,
-      "loss": 6.4648,
-      "step": 711
-    },
-    {
-      "epoch": 0.7393561786085151,
-      "grad_norm": 1.4375,
-      "learning_rate": 0.00017612622526091403,
-      "loss": 6.2431,
-      "step": 712
-    },
-    {
-      "epoch": 0.7403946002076843,
-      "grad_norm": 1.375,
-      "learning_rate": 0.00017481882031724927,
-      "loss": 6.3375,
-      "step": 713
-    },
-    {
-      "epoch": 0.7414330218068536,
-      "grad_norm": 1.703125,
-      "learning_rate": 0.0001735152571521451,
-      "loss": 6.3156,
-      "step": 714
-    },
-    {
-      "epoch": 0.7424714434060229,
-      "grad_norm": 1.1484375,
-      "learning_rate": 0.00017221555116625,
-      "loss": 6.2417,
-      "step": 715
-    },
-    {
-      "epoch": 0.7435098650051921,
-      "grad_norm": 1.09375,
-      "learning_rate": 0.0001709197177146425,
-      "loss": 6.3143,
-      "step": 716
-    },
-    {
-      "epoch": 0.7445482866043613,
-      "grad_norm": 1.34375,
-      "learning_rate": 0.0001696277721066502,
-      "loss": 6.1396,
-      "step": 717
-    },
-    {
-      "epoch": 0.7455867082035307,
-      "grad_norm": 1.2265625,
-      "learning_rate": 0.00016833972960566868,
-      "loss": 6.3164,
-      "step": 718
-    },
-    {
-      "epoch": 0.7466251298026999,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.00016705560542898051,
-      "loss": 6.2559,
-      "step": 719
-    },
-    {
-      "epoch": 0.7476635514018691,
-      "grad_norm": 1.5234375,
-      "learning_rate": 0.00016577541474757713,
-      "loss": 6.4104,
-      "step": 720
-    },
-    {
-      "epoch": 0.7487019730010385,
-      "grad_norm": 1.421875,
-      "learning_rate": 0.00016449917268597798,
-      "loss": 6.0631,
-      "step": 721
-    },
-    {
-      "epoch": 0.7497403946002077,
-      "grad_norm": 1.4140625,
-      "learning_rate": 0.00016322689432205252,
-      "loss": 6.3299,
-      "step": 722
-    },
-    {
-      "epoch": 0.7507788161993769,
-      "grad_norm": 1.5546875,
-      "learning_rate": 0.00016195859468684198,
-      "loss": 6.2053,
-      "step": 723
-    },
-    {
-      "epoch": 0.7518172377985463,
-      "grad_norm": 1.5546875,
-      "learning_rate": 0.00016069428876438202,
-      "loss": 6.3051,
-      "step": 724
-    },
-    {
-      "epoch": 0.7528556593977155,
-      "grad_norm": 1.359375,
-      "learning_rate": 0.00015943399149152533,
-      "loss": 6.1548,
-      "step": 725
-    },
-    {
-      "epoch": 0.7538940809968847,
-      "grad_norm": 1.375,
-      "learning_rate": 0.00015817771775776507,
-      "loss": 6.2009,
-      "step": 726
-    },
-    {
-      "epoch": 0.754932502596054,
-      "grad_norm": 1.3046875,
-      "learning_rate": 0.00015692548240506,
-      "loss": 6.2028,
-      "step": 727
-    },
-    {
-      "epoch": 0.7559709241952233,
-      "grad_norm": 1.25,
-      "learning_rate": 0.00015567730022765752,
-      "loss": 5.9373,
-      "step": 728
-    },
-    {
-      "epoch": 0.7570093457943925,
-      "grad_norm": 1.203125,
-      "learning_rate": 0.0001544331859719202,
-      "loss": 6.1761,
-      "step": 729
-    },
-    {
-      "epoch": 0.7580477673935618,
-      "grad_norm": 1.484375,
-      "learning_rate": 0.000153193154336151,
-      "loss": 6.2549,
-      "step": 730
-    },
-    {
-      "epoch": 0.7590861889927311,
-      "grad_norm": 1.609375,
-      "learning_rate": 0.00015195721997041933,
-      "loss": 6.1982,
-      "step": 731
-    },
-    {
-      "epoch": 0.7601246105919003,
-      "grad_norm": 1.59375,
-      "learning_rate": 0.00015072539747638887,
-      "loss": 6.2346,
-      "step": 732
-    },
-    {
-      "epoch": 0.7611630321910696,
-      "grad_norm": 1.765625,
-      "learning_rate": 0.00014949770140714407,
-      "loss": 5.5064,
-      "step": 733
-    },
-    {
-      "epoch": 0.7622014537902388,
-      "grad_norm": 1.296875,
-      "learning_rate": 0.0001482741462670193,
-      "loss": 6.3146,
-      "step": 734
-    },
-    {
-      "epoch": 0.7632398753894081,
-      "grad_norm": 1.578125,
-      "learning_rate": 0.0001470547465114263,
-      "loss": 6.2277,
-      "step": 735
-    },
-    {
-      "epoch": 0.7642782969885774,
-      "grad_norm": 1.328125,
-      "learning_rate": 0.00014583951654668415,
-      "loss": 6.3032,
-      "step": 736
-    },
-    {
-      "epoch": 0.7653167185877466,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.00014462847072984898,
-      "loss": 5.9154,
-      "step": 737
-    },
-    {
-      "epoch": 0.7663551401869159,
-      "grad_norm": 1.6484375,
-      "learning_rate": 0.0001434216233685441,
-      "loss": 6.0951,
-      "step": 738
-    },
-    {
-      "epoch": 0.7673935617860852,
-      "grad_norm": 1.578125,
-      "learning_rate": 0.00014221898872079108,
-      "loss": 6.0921,
-      "step": 739
-    },
-    {
-      "epoch": 0.7684319833852544,
-      "grad_norm": 1.5703125,
-      "learning_rate": 0.0001410205809948419,
-      "loss": 6.2295,
-      "step": 740
-    },
-    {
-      "epoch": 0.7694704049844237,
-      "grad_norm": 1.3515625,
-      "learning_rate": 0.00013982641434900984,
-      "loss": 6.229,
-      "step": 741
-    },
-    {
-      "epoch": 0.770508826583593,
-      "grad_norm": 1.5,
-      "learning_rate": 0.00013863650289150338,
-      "loss": 6.3173,
-      "step": 742
-    },
-    {
-      "epoch": 0.7715472481827622,
-      "grad_norm": 1.515625,
-      "learning_rate": 0.00013745086068025857,
-      "loss": 6.3666,
-      "step": 743
-    },
-    {
-      "epoch": 0.7725856697819314,
-      "grad_norm": 1.15625,
-      "learning_rate": 0.00013626950172277398,
-      "loss": 6.1824,
-      "step": 744
-    },
-    {
-      "epoch": 0.7736240913811008,
-      "grad_norm": 1.78125,
-      "learning_rate": 0.00013509243997594423,
-      "loss": 6.2045,
-      "step": 745
-    },
-    {
-      "epoch": 0.77466251298027,
-      "grad_norm": 1.359375,
-      "learning_rate": 0.00013391968934589572,
-      "loss": 6.295,
-      "step": 746
-    },
-    {
-      "epoch": 0.7757009345794392,
-      "grad_norm": 1.265625,
-      "learning_rate": 0.00013275126368782235,
-      "loss": 6.3082,
-      "step": 747
-    },
-    {
-      "epoch": 0.7767393561786086,
-      "grad_norm": 1.5703125,
-      "learning_rate": 0.00013158717680582127,
-      "loss": 6.2444,
-      "step": 748
-    },
-    {
-      "epoch": 0.7777777777777778,
-      "grad_norm": 1.4296875,
-      "learning_rate": 0.00013042744245273037,
-      "loss": 6.1545,
-      "step": 749
-    },
-    {
-      "epoch": 0.778816199376947,
-      "grad_norm": 1.75,
-      "learning_rate": 0.0001292720743299654,
-      "loss": 6.0637,
-      "step": 750
-    },
-    {
-      "epoch": 0.7798546209761164,
-      "grad_norm": 1.4453125,
-      "learning_rate": 0.00012812108608735846,
-      "loss": 6.0392,
-      "step": 751
-    },
-    {
-      "epoch": 0.7808930425752856,
-      "grad_norm": 1.6796875,
-      "learning_rate": 0.0001269744913229965,
-      "loss": 6.285,
-      "step": 752
-    },
-    {
-      "epoch": 0.7819314641744548,
-      "grad_norm": 1.21875,
-      "learning_rate": 0.00012583230358306053,
-      "loss": 6.3178,
-      "step": 753
-    },
-    {
-      "epoch": 0.7829698857736241,
-      "grad_norm": 1.578125,
-      "learning_rate": 0.00012469453636166643,
-      "loss": 6.1123,
-      "step": 754
-    },
-    {
-      "epoch": 0.7840083073727934,
-      "grad_norm": 1.453125,
-      "learning_rate": 0.00012356120310070407,
-      "loss": 6.379,
-      "step": 755
-    },
-    {
-      "epoch": 0.7850467289719626,
-      "grad_norm": 1.625,
-      "learning_rate": 0.00012243231718967967,
-      "loss": 6.127,
-      "step": 756
-    },
-    {
-      "epoch": 0.7860851505711319,
-      "grad_norm": 1.1796875,
-      "learning_rate": 0.0001213078919655573,
-      "loss": 6.222,
-      "step": 757
-    },
-    {
-      "epoch": 0.7871235721703012,
-      "grad_norm": 1.1875,
-      "learning_rate": 0.00012018794071260119,
-      "loss": 6.0595,
-      "step": 758
-    },
-    {
-      "epoch": 0.7881619937694704,
-      "grad_norm": 1.2421875,
-      "learning_rate": 0.00011907247666221893,
-      "loss": 6.1771,
-      "step": 759
-    },
-    {
-      "epoch": 0.7892004153686397,
-      "grad_norm": 1.5,
-      "learning_rate": 0.00011796151299280483,
-      "loss": 6.1493,
-      "step": 760
-    },
-    {
-      "epoch": 0.7902388369678089,
-      "grad_norm": 1.3046875,
-      "learning_rate": 0.00011685506282958496,
-      "loss": 6.3724,
-      "step": 761
-    },
-    {
-      "epoch": 0.7912772585669782,
-      "grad_norm": 1.8203125,
-      "learning_rate": 0.00011575313924446123,
-      "loss": 6.1028,
-      "step": 762
-    },
-    {
-      "epoch": 0.7923156801661475,
-      "grad_norm": 1.5234375,
-      "learning_rate": 0.00011465575525585741,
-      "loss": 6.1988,
-      "step": 763
-    },
-    {
-      "epoch": 0.7933541017653167,
-      "grad_norm": 3.03125,
-      "learning_rate": 0.00011356292382856532,
-      "loss": 6.1213,
-      "step": 764
-    },
-    {
-      "epoch": 0.794392523364486,
-      "grad_norm": 1.8515625,
-      "learning_rate": 0.0001124746578735914,
-      "loss": 6.3859,
-      "step": 765
-    },
-    {
-      "epoch": 0.7954309449636553,
-      "grad_norm": 1.8984375,
-      "learning_rate": 0.0001113909702480046,
-      "loss": 6.0653,
-      "step": 766
-    },
-    {
-      "epoch": 0.7964693665628245,
-      "grad_norm": 1.359375,
-      "learning_rate": 0.00011031187375478407,
-      "loss": 6.2933,
-      "step": 767
-    },
-    {
-      "epoch": 0.7975077881619937,
-      "grad_norm": 1.3203125,
-      "learning_rate": 0.00010923738114266823,
-      "loss": 6.0991,
-      "step": 768
-    },
-    {
-      "epoch": 0.7985462097611631,
-      "grad_norm": 1.4921875,
-      "learning_rate": 0.00010816750510600387,
-      "loss": 6.2484,
-      "step": 769
-    },
-    {
-      "epoch": 0.7995846313603323,
-      "grad_norm": 1.703125,
-      "learning_rate": 0.00010710225828459641,
-      "loss": 5.7827,
-      "step": 770
-    },
-    {
-      "epoch": 0.8006230529595015,
-      "grad_norm": 1.46875,
-      "learning_rate": 0.0001060416532635603,
-      "loss": 6.3373,
-      "step": 771
-    },
-    {
-      "epoch": 0.8016614745586709,
-      "grad_norm": 1.2265625,
-      "learning_rate": 0.00010498570257317076,
-      "loss": 6.3325,
-      "step": 772
-    },
-    {
-      "epoch": 0.8026998961578401,
-      "grad_norm": 1.578125,
-      "learning_rate": 0.00010393441868871506,
-      "loss": 6.1373,
-      "step": 773
-    },
-    {
-      "epoch": 0.8037383177570093,
-      "grad_norm": 1.609375,
-      "learning_rate": 0.00010288781403034619,
-      "loss": 6.06,
-      "step": 774
-    },
-    {
-      "epoch": 0.8047767393561787,
-      "grad_norm": 1.71875,
-      "learning_rate": 0.00010184590096293506,
-      "loss": 6.0622,
-      "step": 775
-    },
-    {
-      "epoch": 0.8058151609553479,
-      "grad_norm": 1.0546875,
-      "learning_rate": 0.0001008086917959249,
-      "loss": 6.2272,
-      "step": 776
-    },
-    {
-      "epoch": 0.8068535825545171,
-      "grad_norm": 1.3828125,
-      "learning_rate": 9.977619878318578e-05,
-      "loss": 6.2692,
-      "step": 777
-    },
-    {
-      "epoch": 0.8078920041536864,
-      "grad_norm": 1.2265625,
-      "learning_rate": 9.874843412286993e-05,
-      "loss": 6.3817,
-      "step": 778
-    },
-    {
-      "epoch": 0.8089304257528557,
-      "grad_norm": 1.9921875,
-      "learning_rate": 9.772540995726753e-05,
-      "loss": 6.4033,
-      "step": 779
-    },
-    {
-      "epoch": 0.8099688473520249,
-      "grad_norm": 1.625,
-      "learning_rate": 9.67071383726632e-05,
-      "loss": 6.0418,
-      "step": 780
-    },
-    {
-      "epoch": 0.8110072689511942,
-      "grad_norm": 1.3125,
-      "learning_rate": 9.569363139919341e-05,
-      "loss": 6.2407,
-      "step": 781
-    },
-    {
-      "epoch": 0.8120456905503635,
-      "grad_norm": 1.6796875,
-      "learning_rate": 9.468490101070409e-05,
-      "loss": 6.1643,
-      "step": 782
-    },
-    {
-      "epoch": 0.8130841121495327,
-      "grad_norm": 1.1796875,
-      "learning_rate": 9.368095912460934e-05,
-      "loss": 6.1331,
-      "step": 783
-    },
-    {
-      "epoch": 0.814122533748702,
-      "grad_norm": 1.2421875,
-      "learning_rate": 9.26818176017506e-05,
-      "loss": 6.363,
-      "step": 784
-    },
-    {
-      "epoch": 0.8151609553478713,
-      "grad_norm": 1.4609375,
-      "learning_rate": 9.168748824625655e-05,
-      "loss": 6.2178,
-      "step": 785
-    },
-    {
-      "epoch": 0.8161993769470405,
-      "grad_norm": 1.6328125,
-      "learning_rate": 9.069798280540348e-05,
-      "loss": 6.146,
-      "step": 786
-    },
-    {
-      "epoch": 0.8172377985462098,
-      "grad_norm": 1.4453125,
-      "learning_rate": 8.9713312969477e-05,
-      "loss": 6.1887,
-      "step": 787
-    },
-    {
-      "epoch": 0.818276220145379,
-      "grad_norm": 1.2421875,
-      "learning_rate": 8.87334903716332e-05,
-      "loss": 6.2521,
-      "step": 788
-    },
-    {
-      "epoch": 0.8193146417445483,
-      "grad_norm": 1.265625,
-      "learning_rate": 8.775852658776173e-05,
-      "loss": 6.3487,
-      "step": 789
-    },
-    {
-      "epoch": 0.8203530633437176,
-      "grad_norm": 1.5234375,
-      "learning_rate": 8.678843313634893e-05,
-      "loss": 6.2509,
-      "step": 790
-    },
-    {
-      "epoch": 0.8213914849428868,
-      "grad_norm": 1.71875,
-      "learning_rate": 8.58232214783416e-05,
-      "loss": 6.0586,
-      "step": 791
-    },
-    {
-      "epoch": 0.822429906542056,
-      "grad_norm": 1.4921875,
-      "learning_rate": 8.486290301701182e-05,
-      "loss": 6.293,
-      "step": 792
-    },
-    {
-      "epoch": 0.8234683281412254,
-      "grad_norm": 1.5,
-      "learning_rate": 8.390748909782204e-05,
-      "loss": 6.2504,
-      "step": 793
-    },
-    {
-      "epoch": 0.8245067497403946,
-      "grad_norm": 1.28125,
-      "learning_rate": 8.295699100829124e-05,
-      "loss": 6.2907,
-      "step": 794
-    },
-    {
-      "epoch": 0.8255451713395638,
-      "grad_norm": 1.125,
-      "learning_rate": 8.201141997786127e-05,
-      "loss": 6.2033,
-      "step": 795
-    },
-    {
-      "epoch": 0.8265835929387332,
-      "grad_norm": 1.2109375,
-      "learning_rate": 8.107078717776456e-05,
-      "loss": 6.3058,
-      "step": 796
-    },
-    {
-      "epoch": 0.8276220145379024,
-      "grad_norm": 1.4296875,
-      "learning_rate": 8.013510372089184e-05,
-      "loss": 6.1276,
-      "step": 797
-    },
-    {
-      "epoch": 0.8286604361370716,
-      "grad_norm": 1.4140625,
-      "learning_rate": 7.920438066166097e-05,
-      "loss": 6.4023,
-      "step": 798
-    },
-    {
-      "epoch": 0.829698857736241,
-      "grad_norm": 1.953125,
-      "learning_rate": 7.827862899588634e-05,
-      "loss": 6.1487,
-      "step": 799
-    },
-    {
-      "epoch": 0.8307372793354102,
-      "grad_norm": 1.3671875,
-      "learning_rate": 7.735785966064884e-05,
-      "loss": 5.9001,
-      "step": 800
-    },
-    {
-      "epoch": 0.8307372793354102,
-      "eval_loss": 6.255230903625488,
-      "eval_runtime": 1.6449,
-      "eval_samples_per_second": 9.727,
-      "eval_steps_per_second": 1.216,
-      "step": 800
-    },
-    {
-      "epoch": 0.8317757009345794,
-      "grad_norm": 1.3515625,
-      "learning_rate": 7.644208353416704e-05,
-      "loss": 6.1918,
-      "step": 801
-    },
-    {
-      "epoch": 0.8328141225337488,
-      "grad_norm": 2.328125,
-      "learning_rate": 7.553131143566822e-05,
-      "loss": 6.2873,
-      "step": 802
-    },
-    {
-      "epoch": 0.833852544132918,
-      "grad_norm": 1.390625,
-      "learning_rate": 7.462555412526062e-05,
-      "loss": 6.1316,
-      "step": 803
-    },
-    {
-      "epoch": 0.8348909657320872,
-      "grad_norm": 2.0,
-      "learning_rate": 7.372482230380656e-05,
-      "loss": 6.2054,
-      "step": 804
-    },
-    {
-      "epoch": 0.8359293873312564,
-      "grad_norm": 1.7578125,
-      "learning_rate": 7.282912661279584e-05,
-      "loss": 6.1334,
-      "step": 805
-    },
-    {
-      "epoch": 0.8369678089304258,
-      "grad_norm": 1.328125,
-      "learning_rate": 7.19384776342199e-05,
-      "loss": 6.1515,
-      "step": 806
-    },
-    {
-      "epoch": 0.838006230529595,
-      "grad_norm": 1.6328125,
-      "learning_rate": 7.105288589044723e-05,
-      "loss": 6.044,
-      "step": 807
-    },
-    {
-      "epoch": 0.8390446521287642,
-      "grad_norm": 1.265625,
-      "learning_rate": 7.017236184409858e-05,
-      "loss": 6.4329,
-      "step": 808
-    },
-    {
-      "epoch": 0.8400830737279336,
-      "grad_norm": 1.203125,
-      "learning_rate": 6.929691589792358e-05,
-      "loss": 6.2727,
-      "step": 809
-    },
-    {
-      "epoch": 0.8411214953271028,
-      "grad_norm": 1.6015625,
-      "learning_rate": 6.842655839467788e-05,
-      "loss": 6.1676,
-      "step": 810
-    },
-    {
-      "epoch": 0.842159916926272,
-      "grad_norm": 1.546875,
-      "learning_rate": 6.756129961700075e-05,
-      "loss": 5.9781,
-      "step": 811
-    },
-    {
-      "epoch": 0.8431983385254413,
-      "grad_norm": 1.8515625,
-      "learning_rate": 6.670114978729391e-05,
-      "loss": 5.9659,
-      "step": 812
-    },
-    {
-      "epoch": 0.8442367601246106,
-      "grad_norm": 1.6640625,
-      "learning_rate": 6.584611906760035e-05,
-      "loss": 5.9971,
-      "step": 813
-    },
-    {
-      "epoch": 0.8452751817237798,
-      "grad_norm": 1.7734375,
-      "learning_rate": 6.499621755948487e-05,
-      "loss": 6.3746,
-      "step": 814
-    },
-    {
-      "epoch": 0.8463136033229491,
-      "grad_norm": 1.296875,
-      "learning_rate": 6.415145530391403e-05,
-      "loss": 5.9012,
-      "step": 815
-    },
-    {
-      "epoch": 0.8473520249221184,
-      "grad_norm": 1.15625,
-      "learning_rate": 6.331184228113802e-05,
-      "loss": 6.3141,
-      "step": 816
-    },
-    {
-      "epoch": 0.8483904465212876,
-      "grad_norm": 1.5078125,
-      "learning_rate": 6.247738841057255e-05,
-      "loss": 6.2155,
-      "step": 817
-    },
-    {
-      "epoch": 0.8494288681204569,
-      "grad_norm": 1.65625,
-      "learning_rate": 6.164810355068179e-05,
-      "loss": 6.1955,
-      "step": 818
-    },
-    {
-      "epoch": 0.8504672897196262,
-      "grad_norm": 1.515625,
-      "learning_rate": 6.082399749886169e-05,
-      "loss": 6.3145,
-      "step": 819
-    },
-    {
-      "epoch": 0.8515057113187954,
-      "grad_norm": 1.3984375,
-      "learning_rate": 6.000507999132443e-05,
-      "loss": 6.173,
-      "step": 820
-    },
-    {
-      "epoch": 0.8525441329179647,
-      "grad_norm": 1.484375,
-      "learning_rate": 5.919136070298342e-05,
-      "loss": 6.0277,
-      "step": 821
-    },
-    {
-      "epoch": 0.8535825545171339,
-      "grad_norm": 1.4609375,
-      "learning_rate": 5.838284924733866e-05,
-      "loss": 6.163,
-      "step": 822
-    },
-    {
-      "epoch": 0.8546209761163032,
-      "grad_norm": 1.390625,
-      "learning_rate": 5.7579555176363654e-05,
-      "loss": 6.324,
-      "step": 823
-    },
-    {
-      "epoch": 0.8556593977154725,
-      "grad_norm": 1.5390625,
-      "learning_rate": 5.678148798039212e-05,
-      "loss": 6.3801,
-      "step": 824
-    },
-    {
-      "epoch": 0.8566978193146417,
-      "grad_norm": 1.5625,
-      "learning_rate": 5.598865708800616e-05,
-      "loss": 5.8392,
-      "step": 825
-    },
-    {
-      "epoch": 0.857736240913811,
-      "grad_norm": 1.609375,
-      "learning_rate": 5.520107186592477e-05,
-      "loss": 6.3217,
-      "step": 826
-    },
-    {
-      "epoch": 0.8587746625129803,
-      "grad_norm": 1.4609375,
-      "learning_rate": 5.441874161889304e-05,
-      "loss": 6.2988,
-      "step": 827
-    },
-    {
-      "epoch": 0.8598130841121495,
-      "grad_norm": 1.328125,
-      "learning_rate": 5.364167558957267e-05,
-      "loss": 6.3745,
-      "step": 828
-    },
-    {
-      "epoch": 0.8608515057113187,
-      "grad_norm": 1.296875,
-      "learning_rate": 5.286988295843215e-05,
-      "loss": 6.2074,
-      "step": 829
-    },
-    {
-      "epoch": 0.8618899273104881,
-      "grad_norm": 1.4296875,
-      "learning_rate": 5.2103372843638754e-05,
-      "loss": 6.2226,
-      "step": 830
-    },
-    {
-      "epoch": 0.8629283489096573,
-      "grad_norm": 1.59375,
-      "learning_rate": 5.134215430095068e-05,
-      "loss": 5.8549,
-      "step": 831
-    },
-    {
-      "epoch": 0.8639667705088265,
-      "grad_norm": 1.71875,
-      "learning_rate": 5.0586236323610034e-05,
-      "loss": 6.3173,
-      "step": 832
-    },
-    {
-      "epoch": 0.8650051921079959,
-      "grad_norm": 1.234375,
-      "learning_rate": 4.983562784223644e-05,
-      "loss": 6.2721,
-      "step": 833
-    },
-    {
-      "epoch": 0.8660436137071651,
-      "grad_norm": 1.40625,
-      "learning_rate": 4.909033772472205e-05,
-      "loss": 6.1832,
-      "step": 834
-    },
-    {
-      "epoch": 0.8670820353063343,
-      "grad_norm": 1.6640625,
-      "learning_rate": 4.835037477612619e-05,
-      "loss": 6.2962,
-      "step": 835
-    },
-    {
-      "epoch": 0.8681204569055037,
-      "grad_norm": 1.5625,
-      "learning_rate": 4.761574773857163e-05,
-      "loss": 6.3508,
-      "step": 836
-    },
-    {
-      "epoch": 0.8691588785046729,
-      "grad_norm": 1.5859375,
-      "learning_rate": 4.688646529114121e-05,
-      "loss": 6.0014,
-      "step": 837
-    },
-    {
-      "epoch": 0.8701973001038421,
-      "grad_norm": 1.5,
-      "learning_rate": 4.6162536049775385e-05,
-      "loss": 5.8404,
-      "step": 838
-    },
-    {
-      "epoch": 0.8712357217030114,
-      "grad_norm": 1.359375,
-      "learning_rate": 4.5443968567170314e-05,
-      "loss": 5.7522,
-      "step": 839
-    },
-    {
-      "epoch": 0.8722741433021807,
-      "grad_norm": 1.2890625,
-      "learning_rate": 4.4730771332676835e-05,
-      "loss": 6.0892,
-      "step": 840
-    },
-    {
-      "epoch": 0.8733125649013499,
-      "grad_norm": 1.328125,
-      "learning_rate": 4.402295277220048e-05,
-      "loss": 6.2618,
-      "step": 841
-    },
-    {
-      "epoch": 0.8743509865005192,
-      "grad_norm": 1.5390625,
-      "learning_rate": 4.3320521248101484e-05,
-      "loss": 5.9419,
-      "step": 842
-    },
-    {
-      "epoch": 0.8753894080996885,
-      "grad_norm": 1.75,
-      "learning_rate": 4.262348505909608e-05,
-      "loss": 5.9647,
-      "step": 843
-    },
-    {
-      "epoch": 0.8764278296988577,
-      "grad_norm": 1.703125,
-      "learning_rate": 4.1931852440158794e-05,
-      "loss": 6.2304,
-      "step": 844
-    },
-    {
-      "epoch": 0.877466251298027,
-      "grad_norm": 1.9765625,
-      "learning_rate": 4.124563156242467e-05,
-      "loss": 6.3681,
-      "step": 845
-    },
-    {
-      "epoch": 0.8785046728971962,
-      "grad_norm": 1.3828125,
-      "learning_rate": 4.056483053309301e-05,
-      "loss": 6.222,
-      "step": 846
-    },
-    {
-      "epoch": 0.8795430944963655,
-      "grad_norm": 1.625,
-      "learning_rate": 3.988945739533173e-05,
-      "loss": 6.1925,
-      "step": 847
-    },
-    {
-      "epoch": 0.8805815160955348,
-      "grad_norm": 1.4453125,
-      "learning_rate": 3.9219520128182086e-05,
-      "loss": 6.0481,
-      "step": 848
-    },
-    {
-      "epoch": 0.881619937694704,
-      "grad_norm": 1.3203125,
-      "learning_rate": 3.855502664646443e-05,
-      "loss": 6.1119,
-      "step": 849
-    },
-    {
-      "epoch": 0.8826583592938733,
-      "grad_norm": 1.40625,
-      "learning_rate": 3.789598480068479e-05,
-      "loss": 6.2874,
-      "step": 850
-    },
-    {
-      "epoch": 0.8836967808930426,
-      "grad_norm": 1.453125,
-      "learning_rate": 3.7242402376942096e-05,
-      "loss": 6.1814,
-      "step": 851
-    },
-    {
-      "epoch": 0.8847352024922118,
-      "grad_norm": 1.7734375,
-      "learning_rate": 3.659428709683621e-05,
-      "loss": 6.3491,
-      "step": 852
-    },
-    {
-      "epoch": 0.885773624091381,
-      "grad_norm": 1.171875,
-      "learning_rate": 3.59516466173766e-05,
-      "loss": 6.186,
-      "step": 853
-    },
-    {
-      "epoch": 0.8868120456905504,
-      "grad_norm": 1.390625,
-      "learning_rate": 3.531448853089192e-05,
-      "loss": 6.1761,
-      "step": 854
-    },
-    {
-      "epoch": 0.8878504672897196,
-      "grad_norm": 1.6640625,
-      "learning_rate": 3.4682820364940636e-05,
-      "loss": 6.235,
-      "step": 855
-    },
-    {
-      "epoch": 0.8888888888888888,
-      "grad_norm": 1.3046875,
-      "learning_rate": 3.40566495822216e-05,
-      "loss": 6.3703,
-      "step": 856
-    },
-    {
-      "epoch": 0.8899273104880582,
-      "grad_norm": 1.53125,
-      "learning_rate": 3.343598358048594e-05,
-      "loss": 6.2765,
-      "step": 857
-    },
-    {
-      "epoch": 0.8909657320872274,
-      "grad_norm": 1.703125,
-      "learning_rate": 3.2820829692449985e-05,
-      "loss": 6.4258,
-      "step": 858
-    },
-    {
-      "epoch": 0.8920041536863966,
-      "grad_norm": 1.6484375,
-      "learning_rate": 3.221119518570848e-05,
-      "loss": 5.9581,
-      "step": 859
-    },
-    {
-      "epoch": 0.893042575285566,
-      "grad_norm": 1.609375,
-      "learning_rate": 3.160708726264855e-05,
-      "loss": 6.1848,
-      "step": 860
-    },
-    {
-      "epoch": 0.8940809968847352,
-      "grad_norm": 1.640625,
-      "learning_rate": 3.100851306036512e-05,
-      "loss": 6.2575,
-      "step": 861
-    },
-    {
-      "epoch": 0.8951194184839044,
-      "grad_norm": 1.5859375,
-      "learning_rate": 3.0415479650575784e-05,
-      "loss": 5.9909,
-      "step": 862
-    },
-    {
-      "epoch": 0.8961578400830738,
-      "grad_norm": 1.3515625,
-      "learning_rate": 2.982799403953801e-05,
-      "loss": 5.9893,
-      "step": 863
-    },
-    {
-      "epoch": 0.897196261682243,
-      "grad_norm": 1.4921875,
-      "learning_rate": 2.9246063167965962e-05,
-      "loss": 6.1746,
-      "step": 864
-    },
-    {
-      "epoch": 0.8982346832814122,
-      "grad_norm": 1.4453125,
-      "learning_rate": 2.8669693910948646e-05,
-      "loss": 6.1166,
-      "step": 865
-    },
-    {
-      "epoch": 0.8992731048805815,
-      "grad_norm": 1.5,
-      "learning_rate": 2.809889307786856e-05,
-      "loss": 6.2788,
-      "step": 866
-    },
-    {
-      "epoch": 0.9003115264797508,
-      "grad_norm": 1.71875,
-      "learning_rate": 2.7533667412321385e-05,
-      "loss": 5.8872,
-      "step": 867
-    },
-    {
-      "epoch": 0.90134994807892,
-      "grad_norm": 1.75,
-      "learning_rate": 2.6974023592036378e-05,
-      "loss": 6.292,
-      "step": 868
-    },
-    {
-      "epoch": 0.9023883696780893,
-      "grad_norm": 1.5390625,
-      "learning_rate": 2.6419968228797274e-05,
-      "loss": 6.03,
-      "step": 869
-    },
-    {
-      "epoch": 0.9034267912772586,
-      "grad_norm": 1.34375,
-      "learning_rate": 2.5871507868364063e-05,
-      "loss": 6.2342,
-      "step": 870
-    },
-    {
-      "epoch": 0.9044652128764278,
-      "grad_norm": 1.453125,
-      "learning_rate": 2.532864899039622e-05,
-      "loss": 6.2455,
-      "step": 871
-    },
-    {
-      "epoch": 0.9055036344755971,
-      "grad_norm": 1.8125,
-      "learning_rate": 2.4791398008375542e-05,
-      "loss": 5.9972,
-      "step": 872
-    },
-    {
-      "epoch": 0.9065420560747663,
-      "grad_norm": 1.2421875,
-      "learning_rate": 2.4259761269530666e-05,
-      "loss": 6.218,
-      "step": 873
-    },
-    {
-      "epoch": 0.9075804776739356,
-      "grad_norm": 1.6796875,
-      "learning_rate": 2.3733745054762058e-05,
-      "loss": 5.6284,
-      "step": 874
-    },
-    {
-      "epoch": 0.9086188992731049,
-      "grad_norm": 1.3671875,
-      "learning_rate": 2.321335557856791e-05,
-      "loss": 6.2292,
-      "step": 875
-    },
-    {
-      "epoch": 0.9096573208722741,
-      "grad_norm": 1.3671875,
-      "learning_rate": 2.2698598988970422e-05,
-      "loss": 6.3235,
-      "step": 876
-    },
-    {
-      "epoch": 0.9106957424714434,
-      "grad_norm": 1.484375,
-      "learning_rate": 2.2189481367443366e-05,
-      "loss": 6.2901,
-      "step": 877
-    },
-    {
-      "epoch": 0.9117341640706127,
-      "grad_norm": 1.3828125,
-      "learning_rate": 2.16860087288403e-05,
-      "loss": 6.1946,
-      "step": 878
-    },
-    {
-      "epoch": 0.9127725856697819,
-      "grad_norm": 1.390625,
-      "learning_rate": 2.1188187021323413e-05,
-      "loss": 5.9255,
-      "step": 879
-    },
-    {
-      "epoch": 0.9138110072689511,
-      "grad_norm": 1.6171875,
-      "learning_rate": 2.0696022126293124e-05,
-      "loss": 5.9947,
-      "step": 880
-    },
-    {
-      "epoch": 0.9148494288681205,
-      "grad_norm": 1.3828125,
-      "learning_rate": 2.0209519858319037e-05,
-      "loss": 6.0533,
-      "step": 881
-    },
-    {
-      "epoch": 0.9158878504672897,
-      "grad_norm": 1.296875,
-      "learning_rate": 1.9728685965070602e-05,
-      "loss": 6.1479,
-      "step": 882
-    },
-    {
-      "epoch": 0.9169262720664589,
-      "grad_norm": 1.1875,
-      "learning_rate": 1.925352612724979e-05,
-      "loss": 6.1287,
-      "step": 883
-    },
-    {
-      "epoch": 0.9179646936656283,
-      "grad_norm": 1.4609375,
-      "learning_rate": 1.878404595852362e-05,
-      "loss": 6.1971,
-      "step": 884
-    },
-    {
-      "epoch": 0.9190031152647975,
-      "grad_norm": 2.34375,
-      "learning_rate": 1.8320251005457976e-05,
-      "loss": 6.3691,
-      "step": 885
-    },
-    {
-      "epoch": 0.9200415368639667,
-      "grad_norm": 1.4296875,
-      "learning_rate": 1.7862146747452178e-05,
-      "loss": 6.3665,
-      "step": 886
-    },
-    {
-      "epoch": 0.9210799584631361,
-      "grad_norm": 1.5625,
-      "learning_rate": 1.740973859667394e-05,
-      "loss": 6.278,
-      "step": 887
-    },
-    {
-      "epoch": 0.9221183800623053,
-      "grad_norm": 1.8125,
-      "learning_rate": 1.6963031897995862e-05,
-      "loss": 6.1343,
-      "step": 888
-    },
-    {
-      "epoch": 0.9231568016614745,
-      "grad_norm": 1.3046875,
-      "learning_rate": 1.652203192893187e-05,
-      "loss": 6.3423,
-      "step": 889
-    },
-    {
-      "epoch": 0.9241952232606438,
-      "grad_norm": 1.78125,
-      "learning_rate": 1.6086743899575042e-05,
-      "loss": 6.2271,
-      "step": 890
-    },
-    {
-      "epoch": 0.9252336448598131,
-      "grad_norm": 1.578125,
-      "learning_rate": 1.56571729525361e-05,
-      "loss": 6.1616,
-      "step": 891
-    },
-    {
-      "epoch": 0.9262720664589823,
-      "grad_norm": 1.1875,
-      "learning_rate": 1.5233324162882589e-05,
-      "loss": 6.3031,
-      "step": 892
-    },
-    {
-      "epoch": 0.9273104880581516,
-      "grad_norm": 1.3828125,
-      "learning_rate": 1.4815202538078998e-05,
-      "loss": 6.2343,
-      "step": 893
-    },
-    {
-      "epoch": 0.9283489096573209,
-      "grad_norm": 1.9609375,
-      "learning_rate": 1.4402813017927396e-05,
-      "loss": 6.1985,
-      "step": 894
-    },
-    {
-      "epoch": 0.9293873312564901,
-      "grad_norm": 1.3046875,
-      "learning_rate": 1.3996160474509411e-05,
-      "loss": 6.2852,
-      "step": 895
-    },
-    {
-      "epoch": 0.9304257528556594,
-      "grad_norm": 1.8828125,
-      "learning_rate": 1.3595249712128333e-05,
-      "loss": 6.39,
-      "step": 896
-    },
-    {
-      "epoch": 0.9314641744548287,
-      "grad_norm": 1.5078125,
-      "learning_rate": 1.3200085467252488e-05,
-      "loss": 6.1291,
-      "step": 897
-    },
-    {
-      "epoch": 0.9325025960539979,
-      "grad_norm": 1.734375,
-      "learning_rate": 1.28106724084594e-05,
-      "loss": 6.2431,
-      "step": 898
-    },
-    {
-      "epoch": 0.9335410176531672,
-      "grad_norm": 1.1953125,
-      "learning_rate": 1.2427015136380393e-05,
-      "loss": 6.2799,
-      "step": 899
-    },
-    {
-      "epoch": 0.9345794392523364,
-      "grad_norm": 1.3359375,
-      "learning_rate": 1.2049118183646401e-05,
-      "loss": 6.259,
-      "step": 900
-    },
-    {
-      "epoch": 0.9345794392523364,
-      "eval_loss": 6.254264831542969,
-      "eval_runtime": 1.7188,
-      "eval_samples_per_second": 9.309,
-      "eval_steps_per_second": 1.164,
-      "step": 900
     }
   ],
   "logging_steps": 1,
@@ -6399,7 +735,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 1.146256374104064e+17,
+  "total_flos": 1.27361819344896e+16,
   "train_batch_size": 16,
   "trial_name": null,
   "trial_params": null