diff --git "a/checkpoint-618/trainer_state.json" "b/checkpoint-618/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-618/trainer_state.json"
@@ -0,0 +1,4399 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9987878787878788,
+  "eval_steps": 123,
+  "global_step": 618,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0016161616161616162,
+      "grad_norm": 280.0,
+      "learning_rate": 8.064516129032259e-08,
+      "loss": 4.116,
+      "step": 1
+    },
+    {
+      "epoch": 0.0032323232323232323,
+      "grad_norm": 268.0,
+      "learning_rate": 1.6129032258064518e-07,
+      "loss": 4.16,
+      "step": 2
+    },
+    {
+      "epoch": 0.0048484848484848485,
+      "grad_norm": 316.0,
+      "learning_rate": 2.4193548387096775e-07,
+      "loss": 4.0967,
+      "step": 3
+    },
+    {
+      "epoch": 0.006464646464646465,
+      "grad_norm": 280.0,
+      "learning_rate": 3.2258064516129035e-07,
+      "loss": 4.1925,
+      "step": 4
+    },
+    {
+      "epoch": 0.00808080808080808,
+      "grad_norm": 300.0,
+      "learning_rate": 4.032258064516129e-07,
+      "loss": 4.0364,
+      "step": 5
+    },
+    {
+      "epoch": 0.009696969696969697,
+      "grad_norm": 290.0,
+      "learning_rate": 4.838709677419355e-07,
+      "loss": 3.9787,
+      "step": 6
+    },
+    {
+      "epoch": 0.011313131313131313,
+      "grad_norm": 272.0,
+      "learning_rate": 5.645161290322581e-07,
+      "loss": 3.9483,
+      "step": 7
+    },
+    {
+      "epoch": 0.01292929292929293,
+      "grad_norm": 212.0,
+      "learning_rate": 6.451612903225807e-07,
+      "loss": 3.8711,
+      "step": 8
+    },
+    {
+      "epoch": 0.014545454545454545,
+      "grad_norm": 195.0,
+      "learning_rate": 7.258064516129033e-07,
+      "loss": 3.6116,
+      "step": 9
+    },
+    {
+      "epoch": 0.01616161616161616,
+      "grad_norm": 183.0,
+      "learning_rate": 8.064516129032258e-07,
+      "loss": 3.6654,
+      "step": 10
+    },
+    {
+      "epoch": 0.017777777777777778,
+      "grad_norm": 151.0,
+      "learning_rate": 8.870967741935485e-07,
+      "loss": 3.3067,
+      "step": 11
+    },
+    {
+      "epoch": 0.019393939393939394,
+      "grad_norm": 130.0,
+      "learning_rate": 9.67741935483871e-07,
+      "loss": 3.2813,
+      "step": 12
+    },
+    {
+      "epoch": 0.02101010101010101,
+      "grad_norm": 126.0,
+      "learning_rate": 1.0483870967741936e-06,
+      "loss": 3.1032,
+      "step": 13
+    },
+    {
+      "epoch": 0.022626262626262626,
+      "grad_norm": 94.5,
+      "learning_rate": 1.1290322580645162e-06,
+      "loss": 3.2868,
+      "step": 14
+    },
+    {
+      "epoch": 0.024242424242424242,
+      "grad_norm": 98.5,
+      "learning_rate": 1.2096774193548388e-06,
+      "loss": 2.9182,
+      "step": 15
+    },
+    {
+      "epoch": 0.02585858585858586,
+      "grad_norm": 82.0,
+      "learning_rate": 1.2903225806451614e-06,
+      "loss": 2.9145,
+      "step": 16
+    },
+    {
+      "epoch": 0.027474747474747475,
+      "grad_norm": 98.5,
+      "learning_rate": 1.3709677419354838e-06,
+      "loss": 2.7479,
+      "step": 17
+    },
+    {
+      "epoch": 0.02909090909090909,
+      "grad_norm": 94.5,
+      "learning_rate": 1.4516129032258066e-06,
+      "loss": 2.5529,
+      "step": 18
+    },
+    {
+      "epoch": 0.030707070707070707,
+      "grad_norm": 71.0,
+      "learning_rate": 1.5322580645161292e-06,
+      "loss": 2.6872,
+      "step": 19
+    },
+    {
+      "epoch": 0.03232323232323232,
+      "grad_norm": 63.25,
+      "learning_rate": 1.6129032258064516e-06,
+      "loss": 2.3578,
+      "step": 20
+    },
+    {
+      "epoch": 0.03393939393939394,
+      "grad_norm": 76.0,
+      "learning_rate": 1.6935483870967742e-06,
+      "loss": 2.2456,
+      "step": 21
+    },
+    {
+      "epoch": 0.035555555555555556,
+      "grad_norm": 111.5,
+      "learning_rate": 1.774193548387097e-06,
+      "loss": 2.0884,
+      "step": 22
+    },
+    {
+      "epoch": 0.037171717171717175,
+      "grad_norm": 49.5,
+      "learning_rate": 1.8548387096774196e-06,
+      "loss": 2.1233,
+      "step": 23
+    },
+    {
+      "epoch": 0.03878787878787879,
+      "grad_norm": 47.75,
+      "learning_rate": 1.935483870967742e-06,
+      "loss": 1.8103,
+      "step": 24
+    },
+    {
+      "epoch": 0.04040404040404041,
+      "grad_norm": 49.5,
+      "learning_rate": 2.0161290322580646e-06,
+      "loss": 1.6413,
+      "step": 25
+    },
+    {
+      "epoch": 0.04202020202020202,
+      "grad_norm": 45.75,
+      "learning_rate": 2.096774193548387e-06,
+      "loss": 2.0003,
+      "step": 26
+    },
+    {
+      "epoch": 0.04363636363636364,
+      "grad_norm": 43.5,
+      "learning_rate": 2.17741935483871e-06,
+      "loss": 1.7371,
+      "step": 27
+    },
+    {
+      "epoch": 0.04525252525252525,
+      "grad_norm": 40.0,
+      "learning_rate": 2.2580645161290324e-06,
+      "loss": 1.6431,
+      "step": 28
+    },
+    {
+      "epoch": 0.04686868686868687,
+      "grad_norm": 33.75,
+      "learning_rate": 2.338709677419355e-06,
+      "loss": 1.7026,
+      "step": 29
+    },
+    {
+      "epoch": 0.048484848484848485,
+      "grad_norm": 38.5,
+      "learning_rate": 2.4193548387096776e-06,
+      "loss": 1.6614,
+      "step": 30
+    },
+    {
+      "epoch": 0.050101010101010104,
+      "grad_norm": 33.25,
+      "learning_rate": 2.5e-06,
+      "loss": 1.6833,
+      "step": 31
+    },
+    {
+      "epoch": 0.05171717171717172,
+      "grad_norm": 34.75,
+      "learning_rate": 2.580645161290323e-06,
+      "loss": 1.3788,
+      "step": 32
+    },
+    {
+      "epoch": 0.05333333333333334,
+      "grad_norm": 30.5,
+      "learning_rate": 2.6612903225806454e-06,
+      "loss": 1.4505,
+      "step": 33
+    },
+    {
+      "epoch": 0.05494949494949495,
+      "grad_norm": 34.5,
+      "learning_rate": 2.7419354838709676e-06,
+      "loss": 1.3686,
+      "step": 34
+    },
+    {
+      "epoch": 0.05656565656565657,
+      "grad_norm": 29.125,
+      "learning_rate": 2.822580645161291e-06,
+      "loss": 1.5094,
+      "step": 35
+    },
+    {
+      "epoch": 0.05818181818181818,
+      "grad_norm": 27.625,
+      "learning_rate": 2.903225806451613e-06,
+      "loss": 1.1449,
+      "step": 36
+    },
+    {
+      "epoch": 0.0597979797979798,
+      "grad_norm": 26.0,
+      "learning_rate": 2.983870967741936e-06,
+      "loss": 1.2051,
+      "step": 37
+    },
+    {
+      "epoch": 0.061414141414141414,
+      "grad_norm": 30.5,
+      "learning_rate": 3.0645161290322584e-06,
+      "loss": 1.4705,
+      "step": 38
+    },
+    {
+      "epoch": 0.06303030303030303,
+      "grad_norm": 27.125,
+      "learning_rate": 3.145161290322581e-06,
+      "loss": 1.4405,
+      "step": 39
+    },
+    {
+      "epoch": 0.06464646464646465,
+      "grad_norm": 26.5,
+      "learning_rate": 3.225806451612903e-06,
+      "loss": 1.2536,
+      "step": 40
+    },
+    {
+      "epoch": 0.06626262626262626,
+      "grad_norm": 28.375,
+      "learning_rate": 3.306451612903226e-06,
+      "loss": 1.3682,
+      "step": 41
+    },
+    {
+      "epoch": 0.06787878787878789,
+      "grad_norm": 28.375,
+      "learning_rate": 3.3870967741935484e-06,
+      "loss": 1.2269,
+      "step": 42
+    },
+    {
+      "epoch": 0.0694949494949495,
+      "grad_norm": 25.75,
+      "learning_rate": 3.4677419354838714e-06,
+      "loss": 1.155,
+      "step": 43
+    },
+    {
+      "epoch": 0.07111111111111111,
+      "grad_norm": 23.75,
+      "learning_rate": 3.548387096774194e-06,
+      "loss": 1.3323,
+      "step": 44
+    },
+    {
+      "epoch": 0.07272727272727272,
+      "grad_norm": 27.125,
+      "learning_rate": 3.6290322580645166e-06,
+      "loss": 1.4642,
+      "step": 45
+    },
+    {
+      "epoch": 0.07434343434343435,
+      "grad_norm": 25.875,
+      "learning_rate": 3.7096774193548392e-06,
+      "loss": 1.1611,
+      "step": 46
+    },
+    {
+      "epoch": 0.07595959595959596,
+      "grad_norm": 25.375,
+      "learning_rate": 3.7903225806451614e-06,
+      "loss": 1.36,
+      "step": 47
+    },
+    {
+      "epoch": 0.07757575757575758,
+      "grad_norm": 25.5,
+      "learning_rate": 3.870967741935484e-06,
+      "loss": 1.463,
+      "step": 48
+    },
+    {
+      "epoch": 0.07919191919191919,
+      "grad_norm": 28.25,
+      "learning_rate": 3.951612903225807e-06,
+      "loss": 1.3702,
+      "step": 49
+    },
+    {
+      "epoch": 0.08080808080808081,
+      "grad_norm": 25.625,
+      "learning_rate": 4.032258064516129e-06,
+      "loss": 1.3247,
+      "step": 50
+    },
+    {
+      "epoch": 0.08242424242424243,
+      "grad_norm": 19.625,
+      "learning_rate": 4.112903225806452e-06,
+      "loss": 1.1308,
+      "step": 51
+    },
+    {
+      "epoch": 0.08404040404040404,
+      "grad_norm": 22.0,
+      "learning_rate": 4.193548387096774e-06,
+      "loss": 1.1602,
+      "step": 52
+    },
+    {
+      "epoch": 0.08565656565656565,
+      "grad_norm": 22.0,
+      "learning_rate": 4.274193548387097e-06,
+      "loss": 1.2107,
+      "step": 53
+    },
+    {
+      "epoch": 0.08727272727272728,
+      "grad_norm": 22.0,
+      "learning_rate": 4.35483870967742e-06,
+      "loss": 1.1575,
+      "step": 54
+    },
+    {
+      "epoch": 0.08888888888888889,
+      "grad_norm": 23.5,
+      "learning_rate": 4.435483870967742e-06,
+      "loss": 1.3044,
+      "step": 55
+    },
+    {
+      "epoch": 0.0905050505050505,
+      "grad_norm": 20.625,
+      "learning_rate": 4.516129032258065e-06,
+      "loss": 0.9626,
+      "step": 56
+    },
+    {
+      "epoch": 0.09212121212121212,
+      "grad_norm": 23.0,
+      "learning_rate": 4.596774193548387e-06,
+      "loss": 1.5338,
+      "step": 57
+    },
+    {
+      "epoch": 0.09373737373737374,
+      "grad_norm": 21.5,
+      "learning_rate": 4.67741935483871e-06,
+      "loss": 1.3822,
+      "step": 58
+    },
+    {
+      "epoch": 0.09535353535353536,
+      "grad_norm": 18.0,
+      "learning_rate": 4.758064516129033e-06,
+      "loss": 0.8616,
+      "step": 59
+    },
+    {
+      "epoch": 0.09696969696969697,
+      "grad_norm": 20.75,
+      "learning_rate": 4.838709677419355e-06,
+      "loss": 1.133,
+      "step": 60
+    },
+    {
+      "epoch": 0.09858585858585858,
+      "grad_norm": 21.125,
+      "learning_rate": 4.919354838709678e-06,
+      "loss": 1.2883,
+      "step": 61
+    },
+    {
+      "epoch": 0.10020202020202021,
+      "grad_norm": 16.875,
+      "learning_rate": 5e-06,
+      "loss": 1.0195,
+      "step": 62
+    },
+    {
+      "epoch": 0.10181818181818182,
+      "grad_norm": 16.25,
+      "learning_rate": 5.080645161290323e-06,
+      "loss": 1.1219,
+      "step": 63
+    },
+    {
+      "epoch": 0.10343434343434343,
+      "grad_norm": 16.75,
+      "learning_rate": 5.161290322580646e-06,
+      "loss": 1.183,
+      "step": 64
+    },
+    {
+      "epoch": 0.10505050505050505,
+      "grad_norm": 20.375,
+      "learning_rate": 5.241935483870968e-06,
+      "loss": 1.1628,
+      "step": 65
+    },
+    {
+      "epoch": 0.10666666666666667,
+      "grad_norm": 18.75,
+      "learning_rate": 5.322580645161291e-06,
+      "loss": 0.9526,
+      "step": 66
+    },
+    {
+      "epoch": 0.10828282828282829,
+      "grad_norm": 18.25,
+      "learning_rate": 5.4032258064516126e-06,
+      "loss": 0.9687,
+      "step": 67
+    },
+    {
+      "epoch": 0.1098989898989899,
+      "grad_norm": 17.625,
+      "learning_rate": 5.483870967741935e-06,
+      "loss": 0.9895,
+      "step": 68
+    },
+    {
+      "epoch": 0.11151515151515151,
+      "grad_norm": 18.75,
+      "learning_rate": 5.564516129032258e-06,
+      "loss": 0.9116,
+      "step": 69
+    },
+    {
+      "epoch": 0.11313131313131314,
+      "grad_norm": 19.625,
+      "learning_rate": 5.645161290322582e-06,
+      "loss": 1.3456,
+      "step": 70
+    },
+    {
+      "epoch": 0.11474747474747475,
+      "grad_norm": 19.375,
+      "learning_rate": 5.725806451612904e-06,
+      "loss": 0.9474,
+      "step": 71
+    },
+    {
+      "epoch": 0.11636363636363636,
+      "grad_norm": 15.75,
+      "learning_rate": 5.806451612903226e-06,
+      "loss": 1.0595,
+      "step": 72
+    },
+    {
+      "epoch": 0.11797979797979798,
+      "grad_norm": 18.0,
+      "learning_rate": 5.887096774193549e-06,
+      "loss": 1.0755,
+      "step": 73
+    },
+    {
+      "epoch": 0.1195959595959596,
+      "grad_norm": 17.5,
+      "learning_rate": 5.967741935483872e-06,
+      "loss": 0.9928,
+      "step": 74
+    },
+    {
+      "epoch": 0.12121212121212122,
+      "grad_norm": 25.125,
+      "learning_rate": 6.048387096774194e-06,
+      "loss": 0.8585,
+      "step": 75
+    },
+    {
+      "epoch": 0.12282828282828283,
+      "grad_norm": 17.0,
+      "learning_rate": 6.129032258064517e-06,
+      "loss": 1.0376,
+      "step": 76
+    },
+    {
+      "epoch": 0.12444444444444444,
+      "grad_norm": 18.875,
+      "learning_rate": 6.209677419354839e-06,
+      "loss": 0.8824,
+      "step": 77
+    },
+    {
+      "epoch": 0.12606060606060607,
+      "grad_norm": 19.875,
+      "learning_rate": 6.290322580645162e-06,
+      "loss": 1.0713,
+      "step": 78
+    },
+    {
+      "epoch": 0.12767676767676767,
+      "grad_norm": 16.875,
+      "learning_rate": 6.370967741935485e-06,
+      "loss": 1.1012,
+      "step": 79
+    },
+    {
+      "epoch": 0.1292929292929293,
+      "grad_norm": 20.5,
+      "learning_rate": 6.451612903225806e-06,
+      "loss": 1.2101,
+      "step": 80
+    },
+    {
+      "epoch": 0.13090909090909092,
+      "grad_norm": 16.375,
+      "learning_rate": 6.532258064516129e-06,
+      "loss": 1.4869,
+      "step": 81
+    },
+    {
+      "epoch": 0.13252525252525252,
+      "grad_norm": 18.375,
+      "learning_rate": 6.612903225806452e-06,
+      "loss": 1.1957,
+      "step": 82
+    },
+    {
+      "epoch": 0.13414141414141414,
+      "grad_norm": 17.0,
+      "learning_rate": 6.693548387096774e-06,
+      "loss": 0.9294,
+      "step": 83
+    },
+    {
+      "epoch": 0.13575757575757577,
+      "grad_norm": 14.8125,
+      "learning_rate": 6.774193548387097e-06,
+      "loss": 0.7969,
+      "step": 84
+    },
+    {
+      "epoch": 0.13737373737373737,
+      "grad_norm": 16.625,
+      "learning_rate": 6.854838709677419e-06,
+      "loss": 0.9239,
+      "step": 85
+    },
+    {
+      "epoch": 0.138989898989899,
+      "grad_norm": 13.625,
+      "learning_rate": 6.935483870967743e-06,
+      "loss": 0.9914,
+      "step": 86
+    },
+    {
+      "epoch": 0.1406060606060606,
+      "grad_norm": 13.6875,
+      "learning_rate": 7.0161290322580654e-06,
+      "loss": 0.643,
+      "step": 87
+    },
+    {
+      "epoch": 0.14222222222222222,
+      "grad_norm": 15.0625,
+      "learning_rate": 7.096774193548388e-06,
+      "loss": 1.0152,
+      "step": 88
+    },
+    {
+      "epoch": 0.14383838383838385,
+      "grad_norm": 14.3125,
+      "learning_rate": 7.177419354838711e-06,
+      "loss": 0.7406,
+      "step": 89
+    },
+    {
+      "epoch": 0.14545454545454545,
+      "grad_norm": 16.875,
+      "learning_rate": 7.258064516129033e-06,
+      "loss": 1.0667,
+      "step": 90
+    },
+    {
+      "epoch": 0.14707070707070707,
+      "grad_norm": 14.8125,
+      "learning_rate": 7.338709677419356e-06,
+      "loss": 0.958,
+      "step": 91
+    },
+    {
+      "epoch": 0.1486868686868687,
+      "grad_norm": 15.4375,
+      "learning_rate": 7.4193548387096784e-06,
+      "loss": 0.9992,
+      "step": 92
+    },
+    {
+      "epoch": 0.1503030303030303,
+      "grad_norm": 16.25,
+      "learning_rate": 7.500000000000001e-06,
+      "loss": 0.7043,
+      "step": 93
+    },
+    {
+      "epoch": 0.15191919191919193,
+      "grad_norm": 16.875,
+      "learning_rate": 7.580645161290323e-06,
+      "loss": 1.2198,
+      "step": 94
+    },
+    {
+      "epoch": 0.15353535353535352,
+      "grad_norm": 19.5,
+      "learning_rate": 7.661290322580646e-06,
+      "loss": 1.2321,
+      "step": 95
+    },
+    {
+      "epoch": 0.15515151515151515,
+      "grad_norm": 15.4375,
+      "learning_rate": 7.741935483870968e-06,
+      "loss": 0.9691,
+      "step": 96
+    },
+    {
+      "epoch": 0.15676767676767678,
+      "grad_norm": 15.625,
+      "learning_rate": 7.822580645161291e-06,
+      "loss": 0.7549,
+      "step": 97
+    },
+    {
+      "epoch": 0.15838383838383838,
+      "grad_norm": 14.6875,
+      "learning_rate": 7.903225806451613e-06,
+      "loss": 0.8696,
+      "step": 98
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 14.0,
+      "learning_rate": 7.983870967741935e-06,
+      "loss": 0.9764,
+      "step": 99
+    },
+    {
+      "epoch": 0.16161616161616163,
+      "grad_norm": 14.5,
+      "learning_rate": 8.064516129032258e-06,
+      "loss": 0.8763,
+      "step": 100
+    },
+    {
+      "epoch": 0.16323232323232323,
+      "grad_norm": 13.5625,
+      "learning_rate": 8.145161290322582e-06,
+      "loss": 0.9208,
+      "step": 101
+    },
+    {
+      "epoch": 0.16484848484848486,
+      "grad_norm": 16.0,
+      "learning_rate": 8.225806451612904e-06,
+      "loss": 1.1447,
+      "step": 102
+    },
+    {
+      "epoch": 0.16646464646464645,
+      "grad_norm": 16.25,
+      "learning_rate": 8.306451612903227e-06,
+      "loss": 0.9742,
+      "step": 103
+    },
+    {
+      "epoch": 0.16808080808080808,
+      "grad_norm": 17.125,
+      "learning_rate": 8.387096774193549e-06,
+      "loss": 0.9766,
+      "step": 104
+    },
+    {
+      "epoch": 0.1696969696969697,
+      "grad_norm": 15.4375,
+      "learning_rate": 8.467741935483872e-06,
+      "loss": 1.0782,
+      "step": 105
+    },
+    {
+      "epoch": 0.1713131313131313,
+      "grad_norm": 13.75,
+      "learning_rate": 8.548387096774194e-06,
+      "loss": 0.9839,
+      "step": 106
+    },
+    {
+      "epoch": 0.17292929292929293,
+      "grad_norm": 13.75,
+      "learning_rate": 8.629032258064517e-06,
+      "loss": 0.8056,
+      "step": 107
+    },
+    {
+      "epoch": 0.17454545454545456,
+      "grad_norm": 13.875,
+      "learning_rate": 8.70967741935484e-06,
+      "loss": 0.8916,
+      "step": 108
+    },
+    {
+      "epoch": 0.17616161616161616,
+      "grad_norm": 14.6875,
+      "learning_rate": 8.790322580645163e-06,
+      "loss": 0.8902,
+      "step": 109
+    },
+    {
+      "epoch": 0.17777777777777778,
+      "grad_norm": 18.125,
+      "learning_rate": 8.870967741935484e-06,
+      "loss": 1.027,
+      "step": 110
+    },
+    {
+      "epoch": 0.17939393939393938,
+      "grad_norm": 13.3125,
+      "learning_rate": 8.951612903225806e-06,
+      "loss": 0.9191,
+      "step": 111
+    },
+    {
+      "epoch": 0.181010101010101,
+      "grad_norm": 14.0625,
+      "learning_rate": 9.03225806451613e-06,
+      "loss": 0.9115,
+      "step": 112
+    },
+    {
+      "epoch": 0.18262626262626264,
+      "grad_norm": 14.8125,
+      "learning_rate": 9.112903225806451e-06,
+      "loss": 1.4586,
+      "step": 113
+    },
+    {
+      "epoch": 0.18424242424242424,
+      "grad_norm": 14.3125,
+      "learning_rate": 9.193548387096775e-06,
+      "loss": 0.6925,
+      "step": 114
+    },
+    {
+      "epoch": 0.18585858585858586,
+      "grad_norm": 13.875,
+      "learning_rate": 9.274193548387097e-06,
+      "loss": 1.0962,
+      "step": 115
+    },
+    {
+      "epoch": 0.1874747474747475,
+      "grad_norm": 14.25,
+      "learning_rate": 9.35483870967742e-06,
+      "loss": 0.8336,
+      "step": 116
+    },
+    {
+      "epoch": 0.1890909090909091,
+      "grad_norm": 24.375,
+      "learning_rate": 9.435483870967743e-06,
+      "loss": 1.1903,
+      "step": 117
+    },
+    {
+      "epoch": 0.1907070707070707,
+      "grad_norm": 11.9375,
+      "learning_rate": 9.516129032258065e-06,
+      "loss": 0.6861,
+      "step": 118
+    },
+    {
+      "epoch": 0.1923232323232323,
+      "grad_norm": 19.25,
+      "learning_rate": 9.596774193548389e-06,
+      "loss": 1.1889,
+      "step": 119
+    },
+    {
+      "epoch": 0.19393939393939394,
+      "grad_norm": 14.4375,
+      "learning_rate": 9.67741935483871e-06,
+      "loss": 1.0123,
+      "step": 120
+    },
+    {
+      "epoch": 0.19555555555555557,
+      "grad_norm": 15.625,
+      "learning_rate": 9.758064516129034e-06,
+      "loss": 0.9163,
+      "step": 121
+    },
+    {
+      "epoch": 0.19717171717171716,
+      "grad_norm": 14.25,
+      "learning_rate": 9.838709677419356e-06,
+      "loss": 0.9699,
+      "step": 122
+    },
+    {
+      "epoch": 0.1987878787878788,
+      "grad_norm": 13.25,
+      "learning_rate": 9.919354838709679e-06,
+      "loss": 0.9863,
+      "step": 123
+    },
+    {
+      "epoch": 0.1987878787878788,
+      "eval_loss": 0.8790417313575745,
+      "eval_runtime": 35.1712,
+      "eval_samples_per_second": 31.276,
+      "eval_steps_per_second": 3.924,
+      "step": 123
+    },
+    {
+      "epoch": 0.20040404040404042,
+      "grad_norm": 14.1875,
+      "learning_rate": 1e-05,
+      "loss": 1.1048,
+      "step": 124
+    },
+    {
+      "epoch": 0.20202020202020202,
+      "grad_norm": 13.375,
+      "learning_rate": 9.999980046003546e-06,
+      "loss": 0.7441,
+      "step": 125
+    },
+    {
+      "epoch": 0.20363636363636364,
+      "grad_norm": 12.9375,
+      "learning_rate": 9.999920184173449e-06,
+      "loss": 0.9276,
+      "step": 126
+    },
+    {
+      "epoch": 0.20525252525252524,
+      "grad_norm": 13.125,
+      "learning_rate": 9.999820414987502e-06,
+      "loss": 0.9395,
+      "step": 127
+    },
+    {
+      "epoch": 0.20686868686868687,
+      "grad_norm": 14.625,
+      "learning_rate": 9.999680739242022e-06,
+      "loss": 1.0927,
+      "step": 128
+    },
+    {
+      "epoch": 0.2084848484848485,
+      "grad_norm": 13.875,
+      "learning_rate": 9.999501158051846e-06,
+      "loss": 0.8983,
+      "step": 129
+    },
+    {
+      "epoch": 0.2101010101010101,
+      "grad_norm": 12.9375,
+      "learning_rate": 9.999281672850317e-06,
+      "loss": 0.8245,
+      "step": 130
+    },
+    {
+      "epoch": 0.21171717171717172,
+      "grad_norm": 15.5625,
+      "learning_rate": 9.99902228538928e-06,
+      "loss": 0.8986,
+      "step": 131
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 15.625,
+      "learning_rate": 9.99872299773906e-06,
+      "loss": 1.0636,
+      "step": 132
+    },
+    {
+      "epoch": 0.21494949494949495,
+      "grad_norm": 13.125,
+      "learning_rate": 9.998383812288451e-06,
+      "loss": 0.8439,
+      "step": 133
+    },
+    {
+      "epoch": 0.21656565656565657,
+      "grad_norm": 12.25,
+      "learning_rate": 9.998004731744696e-06,
+      "loss": 0.8364,
+      "step": 134
+    },
+    {
+      "epoch": 0.21818181818181817,
+      "grad_norm": 12.0,
+      "learning_rate": 9.997585759133463e-06,
+      "loss": 0.8303,
+      "step": 135
+    },
+    {
+      "epoch": 0.2197979797979798,
+      "grad_norm": 11.5625,
+      "learning_rate": 9.997126897798826e-06,
+      "loss": 0.8335,
+      "step": 136
+    },
+    {
+      "epoch": 0.22141414141414142,
+      "grad_norm": 13.625,
+      "learning_rate": 9.996628151403226e-06,
+      "loss": 0.9116,
+      "step": 137
+    },
+    {
+      "epoch": 0.22303030303030302,
+      "grad_norm": 10.8125,
+      "learning_rate": 9.996089523927461e-06,
+      "loss": 0.7399,
+      "step": 138
+    },
+    {
+      "epoch": 0.22464646464646465,
+      "grad_norm": 9.9375,
+      "learning_rate": 9.995511019670639e-06,
+      "loss": 1.1413,
+      "step": 139
+    },
+    {
+      "epoch": 0.22626262626262628,
+      "grad_norm": 11.625,
+      "learning_rate": 9.994892643250147e-06,
+      "loss": 1.0564,
+      "step": 140
+    },
+    {
+      "epoch": 0.22787878787878788,
+      "grad_norm": 11.6875,
+      "learning_rate": 9.99423439960162e-06,
+      "loss": 1.013,
+      "step": 141
+    },
+    {
+      "epoch": 0.2294949494949495,
+      "grad_norm": 11.5,
+      "learning_rate": 9.993536293978892e-06,
+      "loss": 1.154,
+      "step": 142
+    },
+    {
+      "epoch": 0.2311111111111111,
+      "grad_norm": 11.4375,
+      "learning_rate": 9.992798331953962e-06,
+      "loss": 0.7173,
+      "step": 143
+    },
+    {
+      "epoch": 0.23272727272727273,
+      "grad_norm": 9.9375,
+      "learning_rate": 9.99202051941695e-06,
+      "loss": 0.6389,
+      "step": 144
+    },
+    {
+      "epoch": 0.23434343434343435,
+      "grad_norm": 11.5,
+      "learning_rate": 9.991202862576036e-06,
+      "loss": 1.1159,
+      "step": 145
+    },
+    {
+      "epoch": 0.23595959595959595,
+      "grad_norm": 12.625,
+      "learning_rate": 9.99034536795744e-06,
+      "loss": 0.9197,
+      "step": 146
+    },
+    {
+      "epoch": 0.23757575757575758,
+      "grad_norm": 13.0,
+      "learning_rate": 9.989448042405328e-06,
+      "loss": 0.9929,
+      "step": 147
+    },
+    {
+      "epoch": 0.2391919191919192,
+      "grad_norm": 11.875,
+      "learning_rate": 9.9885108930818e-06,
+      "loss": 0.7332,
+      "step": 148
+    },
+    {
+      "epoch": 0.2408080808080808,
+      "grad_norm": 11.9375,
+      "learning_rate": 9.987533927466804e-06,
+      "loss": 0.8305,
+      "step": 149
+    },
+    {
+      "epoch": 0.24242424242424243,
+      "grad_norm": 12.375,
+      "learning_rate": 9.986517153358086e-06,
+      "loss": 0.7982,
+      "step": 150
+    },
+    {
+      "epoch": 0.24404040404040403,
+      "grad_norm": 10.5625,
+      "learning_rate": 9.98546057887113e-06,
+      "loss": 1.0044,
+      "step": 151
+    },
+    {
+      "epoch": 0.24565656565656566,
+      "grad_norm": 12.75,
+      "learning_rate": 9.984364212439089e-06,
+      "loss": 0.9286,
+      "step": 152
+    },
+    {
+      "epoch": 0.24727272727272728,
+      "grad_norm": 11.5,
+      "learning_rate": 9.98322806281272e-06,
+      "loss": 1.0573,
+      "step": 153
+    },
+    {
+      "epoch": 0.24888888888888888,
+      "grad_norm": 11.0,
+      "learning_rate": 9.982052139060312e-06,
+      "loss": 0.8163,
+      "step": 154
+    },
+    {
+      "epoch": 0.2505050505050505,
+      "grad_norm": 10.0625,
+      "learning_rate": 9.980836450567619e-06,
+      "loss": 0.8431,
+      "step": 155
+    },
+    {
+      "epoch": 0.25212121212121213,
+      "grad_norm": 12.5,
+      "learning_rate": 9.979581007037776e-06,
+      "loss": 0.6704,
+      "step": 156
+    },
+    {
+      "epoch": 0.25373737373737376,
+      "grad_norm": 10.875,
+      "learning_rate": 9.978285818491232e-06,
+      "loss": 0.6866,
+      "step": 157
+    },
+    {
+      "epoch": 0.25535353535353533,
+      "grad_norm": 14.375,
+      "learning_rate": 9.976950895265657e-06,
+      "loss": 1.0205,
+      "step": 158
+    },
+    {
+      "epoch": 0.25696969696969696,
+      "grad_norm": 9.75,
+      "learning_rate": 9.975576248015878e-06,
+      "loss": 0.9211,
+      "step": 159
+    },
+    {
+      "epoch": 0.2585858585858586,
+      "grad_norm": 11.9375,
+      "learning_rate": 9.974161887713775e-06,
+      "loss": 1.1265,
+      "step": 160
+    },
+    {
+      "epoch": 0.2602020202020202,
+      "grad_norm": 9.0,
+      "learning_rate": 9.972707825648204e-06,
+      "loss": 0.8746,
+      "step": 161
+    },
+    {
+      "epoch": 0.26181818181818184,
+      "grad_norm": 12.8125,
+      "learning_rate": 9.971214073424906e-06,
+      "loss": 0.846,
+      "step": 162
+    },
+    {
+      "epoch": 0.2634343434343434,
+      "grad_norm": 10.1875,
+      "learning_rate": 9.969680642966409e-06,
+      "loss": 0.6882,
+      "step": 163
+    },
+    {
+      "epoch": 0.26505050505050504,
+      "grad_norm": 13.5625,
+      "learning_rate": 9.968107546511942e-06,
+      "loss": 0.9636,
+      "step": 164
+    },
+    {
+      "epoch": 0.26666666666666666,
+      "grad_norm": 11.1875,
+      "learning_rate": 9.966494796617328e-06,
+      "loss": 0.6918,
+      "step": 165
+    },
+    {
+      "epoch": 0.2682828282828283,
+      "grad_norm": 9.375,
+      "learning_rate": 9.964842406154892e-06,
+      "loss": 1.0313,
+      "step": 166
+    },
+    {
+      "epoch": 0.2698989898989899,
+      "grad_norm": 10.625,
+      "learning_rate": 9.963150388313347e-06,
+      "loss": 0.9354,
+      "step": 167
+    },
+    {
+      "epoch": 0.27151515151515154,
+      "grad_norm": 12.3125,
+      "learning_rate": 9.961418756597703e-06,
+      "loss": 0.9895,
+      "step": 168
+    },
+    {
+      "epoch": 0.2731313131313131,
+      "grad_norm": 12.3125,
+      "learning_rate": 9.959647524829148e-06,
+      "loss": 0.9479,
+      "step": 169
+    },
+    {
+      "epoch": 0.27474747474747474,
+      "grad_norm": 11.1875,
+      "learning_rate": 9.957836707144943e-06,
+      "loss": 1.0128,
+      "step": 170
+    },
+    {
+      "epoch": 0.27636363636363637,
+      "grad_norm": 9.6875,
+      "learning_rate": 9.95598631799831e-06,
+      "loss": 0.7783,
+      "step": 171
+    },
+    {
+      "epoch": 0.277979797979798,
+      "grad_norm": 11.8125,
+      "learning_rate": 9.95409637215831e-06,
+      "loss": 0.9492,
+      "step": 172
+    },
+    {
+      "epoch": 0.2795959595959596,
+      "grad_norm": 8.8125,
+      "learning_rate": 9.952166884709735e-06,
+      "loss": 0.8773,
+      "step": 173
+    },
+    {
+      "epoch": 0.2812121212121212,
+      "grad_norm": 9.6875,
+      "learning_rate": 9.950197871052974e-06,
+      "loss": 1.0219,
+      "step": 174
+    },
+    {
+      "epoch": 0.2828282828282828,
+      "grad_norm": 10.25,
+      "learning_rate": 9.94818934690391e-06,
+      "loss": 0.6188,
+      "step": 175
+    },
+    {
+      "epoch": 0.28444444444444444,
+      "grad_norm": 12.1875,
+      "learning_rate": 9.94614132829377e-06,
+      "loss": 0.7879,
+      "step": 176
+    },
+    {
+      "epoch": 0.28606060606060607,
+      "grad_norm": 10.5625,
+      "learning_rate": 9.94405383156902e-06,
+      "loss": 0.7926,
+      "step": 177
+    },
+    {
+      "epoch": 0.2876767676767677,
+      "grad_norm": 10.1875,
+      "learning_rate": 9.941926873391223e-06,
+      "loss": 0.7642,
+      "step": 178
+    },
+    {
+      "epoch": 0.28929292929292927,
+      "grad_norm": 11.9375,
+      "learning_rate": 9.939760470736902e-06,
+      "loss": 0.8427,
+      "step": 179
+    },
+    {
+      "epoch": 0.2909090909090909,
+      "grad_norm": 12.5,
+      "learning_rate": 9.937554640897414e-06,
+      "loss": 0.7917,
+      "step": 180
+    },
+    {
+      "epoch": 0.2925252525252525,
+      "grad_norm": 11.875,
+      "learning_rate": 9.935309401478808e-06,
+      "loss": 0.8325,
+      "step": 181
+    },
+    {
+      "epoch": 0.29414141414141415,
+      "grad_norm": 10.0,
+      "learning_rate": 9.933024770401682e-06,
+      "loss": 0.7708,
+      "step": 182
+    },
+    {
+      "epoch": 0.2957575757575758,
+      "grad_norm": 9.5625,
+      "learning_rate": 9.930700765901046e-06,
+      "loss": 1.1573,
+      "step": 183
+    },
+    {
+      "epoch": 0.2973737373737374,
+      "grad_norm": 10.4375,
+      "learning_rate": 9.928337406526172e-06,
+      "loss": 1.0751,
+      "step": 184
+    },
+    {
+      "epoch": 0.298989898989899,
+      "grad_norm": 8.25,
+      "learning_rate": 9.925934711140444e-06,
+      "loss": 0.6648,
+      "step": 185
+    },
+    {
+      "epoch": 0.3006060606060606,
+      "grad_norm": 9.625,
+      "learning_rate": 9.923492698921214e-06,
+      "loss": 1.0982,
+      "step": 186
+    },
+    {
+      "epoch": 0.3022222222222222,
+      "grad_norm": 9.5,
+      "learning_rate": 9.921011389359638e-06,
+      "loss": 0.9648,
+      "step": 187
+    },
+    {
+      "epoch": 0.30383838383838385,
+      "grad_norm": 13.4375,
+      "learning_rate": 9.918490802260538e-06,
+      "loss": 0.7589,
+      "step": 188
+    },
+    {
+      "epoch": 0.3054545454545455,
+      "grad_norm": 10.4375,
+      "learning_rate": 9.915930957742228e-06,
+      "loss": 0.6699,
+      "step": 189
+    },
+    {
+      "epoch": 0.30707070707070705,
+      "grad_norm": 10.9375,
+      "learning_rate": 9.913331876236358e-06,
+      "loss": 0.9228,
+      "step": 190
+    },
+    {
+      "epoch": 0.3086868686868687,
+      "grad_norm": 9.75,
+      "learning_rate": 9.910693578487755e-06,
+      "loss": 0.6985,
+      "step": 191
+    },
+    {
+      "epoch": 0.3103030303030303,
+      "grad_norm": 9.0625,
+      "learning_rate": 9.90801608555425e-06,
+      "loss": 0.8004,
+      "step": 192
+    },
+    {
+      "epoch": 0.31191919191919193,
+      "grad_norm": 9.25,
+      "learning_rate": 9.905299418806517e-06,
+      "loss": 0.8232,
+      "step": 193
+    },
+    {
+      "epoch": 0.31353535353535356,
+      "grad_norm": 10.625,
+      "learning_rate": 9.902543599927903e-06,
+      "loss": 1.3909,
+      "step": 194
+    },
+    {
+      "epoch": 0.3151515151515151,
+      "grad_norm": 9.25,
+      "learning_rate": 9.899748650914245e-06,
+      "loss": 0.9673,
+      "step": 195
+    },
+    {
+      "epoch": 0.31676767676767675,
+      "grad_norm": 12.0,
+      "learning_rate": 9.896914594073703e-06,
+      "loss": 0.9208,
+      "step": 196
+    },
+    {
+      "epoch": 0.3183838383838384,
+      "grad_norm": 9.4375,
+      "learning_rate": 9.894041452026584e-06,
+      "loss": 0.8361,
+      "step": 197
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 9.6875,
+      "learning_rate": 9.891129247705153e-06,
+      "loss": 1.0054,
+      "step": 198
+    },
+    {
+      "epoch": 0.32161616161616163,
+      "grad_norm": 10.0625,
+      "learning_rate": 9.888178004353456e-06,
+      "loss": 0.8643,
+      "step": 199
+    },
+    {
+      "epoch": 0.32323232323232326,
+      "grad_norm": 8.75,
+      "learning_rate": 9.885187745527132e-06,
+      "loss": 0.738,
+      "step": 200
+    },
+    {
+      "epoch": 0.32484848484848483,
+      "grad_norm": 10.0,
+      "learning_rate": 9.88215849509323e-06,
+      "loss": 0.9793,
+      "step": 201
+    },
+    {
+      "epoch": 0.32646464646464646,
+      "grad_norm": 10.4375,
+      "learning_rate": 9.879090277230005e-06,
+      "loss": 0.7254,
+      "step": 202
+    },
+    {
+      "epoch": 0.3280808080808081,
+      "grad_norm": 11.5,
+      "learning_rate": 9.875983116426745e-06,
+      "loss": 0.8806,
+      "step": 203
+    },
+    {
+      "epoch": 0.3296969696969697,
+      "grad_norm": 8.9375,
+      "learning_rate": 9.87283703748356e-06,
+      "loss": 0.9895,
+      "step": 204
+    },
+    {
+      "epoch": 0.33131313131313134,
+      "grad_norm": 8.4375,
+      "learning_rate": 9.869652065511188e-06,
+      "loss": 0.7317,
+      "step": 205
+    },
+    {
+      "epoch": 0.3329292929292929,
+      "grad_norm": 9.5,
+      "learning_rate": 9.866428225930798e-06,
+      "loss": 0.6969,
+      "step": 206
+    },
+    {
+      "epoch": 0.33454545454545453,
+      "grad_norm": 10.1875,
+      "learning_rate": 9.863165544473781e-06,
+      "loss": 0.8589,
+      "step": 207
+    },
+    {
+      "epoch": 0.33616161616161616,
+      "grad_norm": 8.5625,
+      "learning_rate": 9.859864047181551e-06,
+      "loss": 0.6515,
+      "step": 208
+    },
+    {
+      "epoch": 0.3377777777777778,
+      "grad_norm": 7.84375,
+      "learning_rate": 9.856523760405339e-06,
+      "loss": 0.6014,
+      "step": 209
+    },
+    {
+      "epoch": 0.3393939393939394,
+      "grad_norm": 8.8125,
+      "learning_rate": 9.853144710805966e-06,
+      "loss": 0.7315,
+      "step": 210
+    },
+    {
+      "epoch": 0.341010101010101,
+      "grad_norm": 11.25,
+      "learning_rate": 9.849726925353655e-06,
+      "loss": 0.721,
+      "step": 211
+    },
+    {
+      "epoch": 0.3426262626262626,
+      "grad_norm": 9.8125,
+      "learning_rate": 9.846270431327793e-06,
+      "loss": 0.8893,
+      "step": 212
+    },
+    {
+      "epoch": 0.34424242424242424,
+      "grad_norm": 8.9375,
+      "learning_rate": 9.842775256316732e-06,
+      "loss": 0.9498,
+      "step": 213
+    },
+    {
+      "epoch": 0.34585858585858587,
+      "grad_norm": 11.125,
+      "learning_rate": 9.839241428217553e-06,
+      "loss": 0.8841,
+      "step": 214
+    },
+    {
+      "epoch": 0.3474747474747475,
+      "grad_norm": 10.125,
+      "learning_rate": 9.835668975235857e-06,
+      "loss": 0.9036,
+      "step": 215
+    },
+    {
+      "epoch": 0.3490909090909091,
+      "grad_norm": 9.6875,
+      "learning_rate": 9.832057925885526e-06,
+      "loss": 0.8493,
+      "step": 216
+    },
+    {
+      "epoch": 0.3507070707070707,
+      "grad_norm": 9.375,
+      "learning_rate": 9.828408308988506e-06,
+      "loss": 0.8034,
+      "step": 217
+    },
+    {
+      "epoch": 0.3523232323232323,
+      "grad_norm": 8.9375,
+      "learning_rate": 9.824720153674578e-06,
+      "loss": 0.7762,
+      "step": 218
+    },
+    {
+      "epoch": 0.35393939393939394,
+      "grad_norm": 11.6875,
+      "learning_rate": 9.820993489381114e-06,
+      "loss": 0.986,
+      "step": 219
+    },
+    {
+      "epoch": 0.35555555555555557,
+      "grad_norm": 9.8125,
+      "learning_rate": 9.817228345852853e-06,
+      "loss": 0.9599,
+      "step": 220
+    },
+    {
+      "epoch": 0.3571717171717172,
+      "grad_norm": 9.1875,
+      "learning_rate": 9.813424753141658e-06,
+      "loss": 0.7152,
+      "step": 221
+    },
+    {
+      "epoch": 0.35878787878787877,
+      "grad_norm": 7.75,
+      "learning_rate": 9.809582741606283e-06,
+      "loss": 0.828,
+      "step": 222
+    },
+    {
+      "epoch": 0.3604040404040404,
+      "grad_norm": 9.3125,
+      "learning_rate": 9.805702341912117e-06,
+      "loss": 0.9487,
+      "step": 223
+    },
+    {
+      "epoch": 0.362020202020202,
+      "grad_norm": 9.125,
+      "learning_rate": 9.801783585030959e-06,
+      "loss": 0.8378,
+      "step": 224
+    },
+    {
+      "epoch": 0.36363636363636365,
+      "grad_norm": 10.5625,
+      "learning_rate": 9.797826502240746e-06,
+      "loss": 1.0327,
+      "step": 225
+    },
+    {
+      "epoch": 0.3652525252525253,
+      "grad_norm": 38.75,
+      "learning_rate": 9.79383112512533e-06,
+      "loss": 0.877,
+      "step": 226
+    },
+    {
+      "epoch": 0.36686868686868684,
+      "grad_norm": 8.6875,
+      "learning_rate": 9.789797485574204e-06,
+      "loss": 1.0547,
+      "step": 227
+    },
+    {
+      "epoch": 0.36848484848484847,
+      "grad_norm": 9.1875,
+      "learning_rate": 9.785725615782262e-06,
+      "loss": 0.6674,
+      "step": 228
+    },
+    {
+      "epoch": 0.3701010101010101,
+      "grad_norm": 14.75,
+      "learning_rate": 9.781615548249533e-06,
+      "loss": 0.8929,
+      "step": 229
+    },
+    {
+      "epoch": 0.3717171717171717,
+      "grad_norm": 8.5,
+      "learning_rate": 9.777467315780926e-06,
+      "loss": 0.6194,
+      "step": 230
+    },
+    {
+      "epoch": 0.37333333333333335,
+      "grad_norm": 9.6875,
+      "learning_rate": 9.773280951485967e-06,
+      "loss": 0.7332,
+      "step": 231
+    },
+    {
+      "epoch": 0.374949494949495,
+      "grad_norm": 7.5625,
+      "learning_rate": 9.769056488778538e-06,
+      "loss": 0.6245,
+      "step": 232
+    },
+    {
+      "epoch": 0.37656565656565655,
+      "grad_norm": 8.25,
+      "learning_rate": 9.764793961376602e-06,
+      "loss": 0.8382,
+      "step": 233
+    },
+    {
+      "epoch": 0.3781818181818182,
+      "grad_norm": 8.9375,
+      "learning_rate": 9.760493403301941e-06,
+      "loss": 0.6372,
+      "step": 234
+    },
+    {
+      "epoch": 0.3797979797979798,
+      "grad_norm": 11.875,
+      "learning_rate": 9.756154848879885e-06,
+      "loss": 0.9138,
+      "step": 235
+    },
+    {
+      "epoch": 0.3814141414141414,
+      "grad_norm": 8.25,
+      "learning_rate": 9.751778332739033e-06,
+      "loss": 0.9099,
+      "step": 236
+    },
+    {
+      "epoch": 0.38303030303030305,
+      "grad_norm": 9.1875,
+      "learning_rate": 9.74736388981098e-06,
+      "loss": 0.864,
+      "step": 237
+    },
+    {
+      "epoch": 0.3846464646464646,
+      "grad_norm": 8.4375,
+      "learning_rate": 9.74291155533004e-06,
+      "loss": 0.6037,
+      "step": 238
+    },
+    {
+      "epoch": 0.38626262626262625,
+      "grad_norm": 8.875,
+      "learning_rate": 9.738421364832956e-06,
+      "loss": 0.6612,
+      "step": 239
+    },
+    {
+      "epoch": 0.3878787878787879,
+      "grad_norm": 8.25,
+      "learning_rate": 9.733893354158628e-06,
+      "loss": 0.772,
+      "step": 240
+    },
+    {
+      "epoch": 0.3894949494949495,
+      "grad_norm": 8.6875,
+      "learning_rate": 9.729327559447817e-06,
+      "loss": 0.7761,
+      "step": 241
+    },
+    {
+      "epoch": 0.39111111111111113,
+      "grad_norm": 9.4375,
+      "learning_rate": 9.724724017142869e-06,
+      "loss": 0.8807,
+      "step": 242
+    },
+    {
+      "epoch": 0.3927272727272727,
+      "grad_norm": 8.5,
+      "learning_rate": 9.720082763987404e-06,
+      "loss": 0.8429,
+      "step": 243
+    },
+    {
+      "epoch": 0.39434343434343433,
+      "grad_norm": 7.5625,
+      "learning_rate": 9.715403837026046e-06,
+      "loss": 0.4624,
+      "step": 244
+    },
+    {
+      "epoch": 0.39595959595959596,
+      "grad_norm": 10.4375,
+      "learning_rate": 9.710687273604111e-06,
+      "loss": 0.865,
+      "step": 245
+    },
+    {
+      "epoch": 0.3975757575757576,
+      "grad_norm": 6.875,
+      "learning_rate": 9.705933111367314e-06,
+      "loss": 0.7918,
+      "step": 246
+    },
+    {
+      "epoch": 0.3975757575757576,
+      "eval_loss": 0.8323823809623718,
+      "eval_runtime": 35.1362,
+      "eval_samples_per_second": 31.307,
+      "eval_steps_per_second": 3.928,
+      "step": 246
+    },
+    {
+      "epoch": 0.3991919191919192,
+      "grad_norm": 8.375,
+      "learning_rate": 9.701141388261471e-06,
+      "loss": 0.7728,
+      "step": 247
+    },
+    {
+      "epoch": 0.40080808080808084,
+      "grad_norm": 9.5,
+      "learning_rate": 9.69631214253219e-06,
+      "loss": 0.8228,
+      "step": 248
+    },
+    {
+      "epoch": 0.4024242424242424,
+      "grad_norm": 10.8125,
+      "learning_rate": 9.691445412724576e-06,
+      "loss": 0.8661,
+      "step": 249
+    },
+    {
+      "epoch": 0.40404040404040403,
+      "grad_norm": 10.6875,
+      "learning_rate": 9.68654123768291e-06,
+      "loss": 0.8111,
+      "step": 250
+    },
+    {
+      "epoch": 0.40565656565656566,
+      "grad_norm": 8.5,
+      "learning_rate": 9.681599656550346e-06,
+      "loss": 0.7355,
+      "step": 251
+    },
+    {
+      "epoch": 0.4072727272727273,
+      "grad_norm": 8.1875,
+      "learning_rate": 9.676620708768608e-06,
+      "loss": 0.8194,
+      "step": 252
+    },
+    {
+      "epoch": 0.4088888888888889,
+      "grad_norm": 10.25,
+      "learning_rate": 9.67160443407765e-06,
+      "loss": 0.8502,
+      "step": 253
+    },
+    {
+      "epoch": 0.4105050505050505,
+      "grad_norm": 16.625,
+      "learning_rate": 9.666550872515367e-06,
+      "loss": 0.841,
+      "step": 254
+    },
+    {
+      "epoch": 0.4121212121212121,
+      "grad_norm": 9.3125,
+      "learning_rate": 9.661460064417263e-06,
+      "loss": 1.1208,
+      "step": 255
+    },
+    {
+      "epoch": 0.41373737373737374,
+      "grad_norm": 10.0625,
+      "learning_rate": 9.656332050416118e-06,
+      "loss": 0.9173,
+      "step": 256
+    },
+    {
+      "epoch": 0.41535353535353536,
+      "grad_norm": 9.5625,
+      "learning_rate": 9.651166871441684e-06,
+      "loss": 1.1702,
+      "step": 257
+    },
+    {
+      "epoch": 0.416969696969697,
+      "grad_norm": 9.875,
+      "learning_rate": 9.645964568720345e-06,
+      "loss": 0.8802,
+      "step": 258
+    },
+    {
+      "epoch": 0.41858585858585856,
+      "grad_norm": 8.3125,
+      "learning_rate": 9.640725183774797e-06,
+      "loss": 0.9126,
+      "step": 259
+    },
+    {
+      "epoch": 0.4202020202020202,
+      "grad_norm": 8.25,
+      "learning_rate": 9.635448758423703e-06,
+      "loss": 0.6081,
+      "step": 260
+    },
+    {
+      "epoch": 0.4218181818181818,
+      "grad_norm": 7.6875,
+      "learning_rate": 9.630135334781373e-06,
+      "loss": 0.7091,
+      "step": 261
+    },
+    {
+      "epoch": 0.42343434343434344,
+      "grad_norm": 9.8125,
+      "learning_rate": 9.624784955257423e-06,
+      "loss": 0.7537,
+      "step": 262
+    },
+    {
+      "epoch": 0.42505050505050507,
+      "grad_norm": 8.375,
+      "learning_rate": 9.619397662556434e-06,
+      "loss": 0.7439,
+      "step": 263
+    },
+    {
+      "epoch": 0.4266666666666667,
+      "grad_norm": 8.5625,
+      "learning_rate": 9.613973499677613e-06,
+      "loss": 0.9897,
+      "step": 264
+    },
+    {
+      "epoch": 0.42828282828282827,
+      "grad_norm": 9.9375,
+      "learning_rate": 9.608512509914453e-06,
+      "loss": 0.7775,
+      "step": 265
+    },
+    {
+      "epoch": 0.4298989898989899,
+      "grad_norm": 9.6875,
+      "learning_rate": 9.60301473685438e-06,
+      "loss": 0.8957,
+      "step": 266
+    },
+    {
+      "epoch": 0.4315151515151515,
+      "grad_norm": 11.0,
+      "learning_rate": 9.597480224378412e-06,
+      "loss": 1.0668,
+      "step": 267
+    },
+    {
+      "epoch": 0.43313131313131314,
+      "grad_norm": 10.6875,
+      "learning_rate": 9.591909016660806e-06,
+      "loss": 0.8081,
+      "step": 268
+    },
+    {
+      "epoch": 0.43474747474747477,
+      "grad_norm": 9.25,
+      "learning_rate": 9.586301158168706e-06,
+      "loss": 0.8119,
+      "step": 269
+    },
+    {
+      "epoch": 0.43636363636363634,
+      "grad_norm": 11.4375,
+      "learning_rate": 9.580656693661787e-06,
+      "loss": 0.8829,
+      "step": 270
+    },
+    {
+      "epoch": 0.43797979797979797,
+      "grad_norm": 9.8125,
+      "learning_rate": 9.574975668191899e-06,
+      "loss": 0.7646,
+      "step": 271
+    },
+    {
+      "epoch": 0.4395959595959596,
+      "grad_norm": 10.25,
+      "learning_rate": 9.569258127102708e-06,
+      "loss": 0.8685,
+      "step": 272
+    },
+    {
+      "epoch": 0.4412121212121212,
+      "grad_norm": 10.0,
+      "learning_rate": 9.56350411602933e-06,
+      "loss": 1.093,
+      "step": 273
+    },
+    {
+      "epoch": 0.44282828282828285,
+      "grad_norm": 51.5,
+      "learning_rate": 9.55771368089797e-06,
+      "loss": 0.8206,
+      "step": 274
+    },
+    {
+      "epoch": 0.4444444444444444,
+      "grad_norm": 8.5625,
+      "learning_rate": 9.551886867925565e-06,
+      "loss": 1.0301,
+      "step": 275
+    },
+    {
+      "epoch": 0.44606060606060605,
+      "grad_norm": 10.125,
+      "learning_rate": 9.546023723619387e-06,
+      "loss": 1.0908,
+      "step": 276
+    },
+    {
+      "epoch": 0.4476767676767677,
+      "grad_norm": 9.3125,
+      "learning_rate": 9.540124294776704e-06,
+      "loss": 0.7014,
+      "step": 277
+    },
+    {
+      "epoch": 0.4492929292929293,
+      "grad_norm": 7.84375,
+      "learning_rate": 9.534188628484391e-06,
+      "loss": 0.682,
+      "step": 278
+    },
+    {
+      "epoch": 0.4509090909090909,
+      "grad_norm": 7.59375,
+      "learning_rate": 9.528216772118553e-06,
+      "loss": 0.7439,
+      "step": 279
+    },
+    {
+      "epoch": 0.45252525252525255,
+      "grad_norm": 8.125,
+      "learning_rate": 9.522208773344147e-06,
+      "loss": 0.7983,
+      "step": 280
+    },
+    {
+      "epoch": 0.4541414141414141,
+      "grad_norm": 6.40625,
+      "learning_rate": 9.516164680114612e-06,
+      "loss": 0.4831,
+      "step": 281
+    },
+    {
+      "epoch": 0.45575757575757575,
+      "grad_norm": 10.375,
+      "learning_rate": 9.510084540671471e-06,
+      "loss": 0.7558,
+      "step": 282
+    },
+    {
+      "epoch": 0.4573737373737374,
+      "grad_norm": 9.8125,
+      "learning_rate": 9.50396840354396e-06,
+      "loss": 0.5978,
+      "step": 283
+    },
+    {
+      "epoch": 0.458989898989899,
+      "grad_norm": 8.0625,
+      "learning_rate": 9.497816317548625e-06,
+      "loss": 0.9693,
+      "step": 284
+    },
+    {
+      "epoch": 0.46060606060606063,
+      "grad_norm": 7.65625,
+      "learning_rate": 9.491628331788949e-06,
+      "loss": 0.6546,
+      "step": 285
+    },
+    {
+      "epoch": 0.4622222222222222,
+      "grad_norm": 8.9375,
+      "learning_rate": 9.485404495654953e-06,
+      "loss": 0.9247,
+      "step": 286
+    },
+    {
+      "epoch": 0.4638383838383838,
+      "grad_norm": 8.0,
+      "learning_rate": 9.479144858822795e-06,
+      "loss": 0.6692,
+      "step": 287
+    },
+    {
+      "epoch": 0.46545454545454545,
+      "grad_norm": 10.125,
+      "learning_rate": 9.472849471254386e-06,
+      "loss": 1.1385,
+      "step": 288
+    },
+    {
+      "epoch": 0.4670707070707071,
+      "grad_norm": 8.125,
+      "learning_rate": 9.466518383196981e-06,
+      "loss": 0.6915,
+      "step": 289
+    },
+    {
+      "epoch": 0.4686868686868687,
+      "grad_norm": 6.625,
+      "learning_rate": 9.460151645182784e-06,
+      "loss": 0.6693,
+      "step": 290
+    },
+    {
+      "epoch": 0.4703030303030303,
+      "grad_norm": 8.6875,
+      "learning_rate": 9.453749308028542e-06,
+      "loss": 0.641,
+      "step": 291
+    },
+    {
+      "epoch": 0.4719191919191919,
+      "grad_norm": 8.75,
+      "learning_rate": 9.447311422835141e-06,
+      "loss": 0.8019,
+      "step": 292
+    },
+    {
+      "epoch": 0.47353535353535353,
+      "grad_norm": 9.0,
+      "learning_rate": 9.440838040987194e-06,
+      "loss": 0.8253,
+      "step": 293
+    },
+    {
+      "epoch": 0.47515151515151516,
+      "grad_norm": 9.4375,
+      "learning_rate": 9.43432921415264e-06,
+      "loss": 1.1923,
+      "step": 294
+    },
+    {
+      "epoch": 0.4767676767676768,
+      "grad_norm": 7.90625,
+      "learning_rate": 9.427784994282318e-06,
+      "loss": 0.7972,
+      "step": 295
+    },
+    {
+      "epoch": 0.4783838383838384,
+      "grad_norm": 8.0,
+      "learning_rate": 9.421205433609568e-06,
+      "loss": 0.7207,
+      "step": 296
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 8.5625,
+      "learning_rate": 9.414590584649798e-06,
+      "loss": 0.861,
+      "step": 297
+    },
+    {
+      "epoch": 0.4816161616161616,
+      "grad_norm": 8.8125,
+      "learning_rate": 9.407940500200082e-06,
+      "loss": 0.8226,
+      "step": 298
+    },
+    {
+      "epoch": 0.48323232323232324,
+      "grad_norm": 7.75,
+      "learning_rate": 9.40125523333872e-06,
+      "loss": 0.9907,
+      "step": 299
+    },
+    {
+      "epoch": 0.48484848484848486,
+      "grad_norm": 8.3125,
+      "learning_rate": 9.39453483742483e-06,
+      "loss": 0.7025,
+      "step": 300
+    },
+    {
+      "epoch": 0.4864646464646465,
+      "grad_norm": 6.6875,
+      "learning_rate": 9.387779366097915e-06,
+      "loss": 0.5553,
+      "step": 301
+    },
+    {
+      "epoch": 0.48808080808080806,
+      "grad_norm": 6.40625,
+      "learning_rate": 9.380988873277436e-06,
+      "loss": 0.6758,
+      "step": 302
+    },
+    {
+      "epoch": 0.4896969696969697,
+      "grad_norm": 7.0625,
+      "learning_rate": 9.37416341316238e-06,
+      "loss": 0.5991,
+      "step": 303
+    },
+    {
+      "epoch": 0.4913131313131313,
+      "grad_norm": 8.8125,
+      "learning_rate": 9.367303040230828e-06,
+      "loss": 0.9227,
+      "step": 304
+    },
+    {
+      "epoch": 0.49292929292929294,
+      "grad_norm": 23.75,
+      "learning_rate": 9.360407809239527e-06,
+      "loss": 0.6559,
+      "step": 305
+    },
+    {
+      "epoch": 0.49454545454545457,
+      "grad_norm": 9.625,
+      "learning_rate": 9.35347777522344e-06,
+      "loss": 0.6758,
+      "step": 306
+    },
+    {
+      "epoch": 0.49616161616161614,
+      "grad_norm": 8.875,
+      "learning_rate": 9.346512993495315e-06,
+      "loss": 0.8509,
+      "step": 307
+    },
+    {
+      "epoch": 0.49777777777777776,
+      "grad_norm": 8.6875,
+      "learning_rate": 9.33951351964525e-06,
+      "loss": 0.6474,
+      "step": 308
+    },
+    {
+      "epoch": 0.4993939393939394,
+      "grad_norm": 8.0625,
+      "learning_rate": 9.33247940954023e-06,
+      "loss": 0.7196,
+      "step": 309
+    },
+    {
+      "epoch": 0.501010101010101,
+      "grad_norm": 8.125,
+      "learning_rate": 9.3254107193237e-06,
+      "loss": 0.7781,
+      "step": 310
+    },
+    {
+      "epoch": 0.5026262626262626,
+      "grad_norm": 11.1875,
+      "learning_rate": 9.318307505415108e-06,
+      "loss": 1.1145,
+      "step": 311
+    },
+    {
+      "epoch": 0.5042424242424243,
+      "grad_norm": 8.6875,
+      "learning_rate": 9.311169824509454e-06,
+      "loss": 0.979,
+      "step": 312
+    },
+    {
+      "epoch": 0.5058585858585859,
+      "grad_norm": 9.6875,
+      "learning_rate": 9.303997733576846e-06,
+      "loss": 1.0086,
+      "step": 313
+    },
+    {
+      "epoch": 0.5074747474747475,
+      "grad_norm": 5.9375,
+      "learning_rate": 9.296791289862031e-06,
+      "loss": 0.5292,
+      "step": 314
+    },
+    {
+      "epoch": 0.509090909090909,
+      "grad_norm": 8.1875,
+      "learning_rate": 9.289550550883952e-06,
+      "loss": 0.7044,
+      "step": 315
+    },
+    {
+      "epoch": 0.5107070707070707,
+      "grad_norm": 8.5625,
+      "learning_rate": 9.28227557443528e-06,
+      "loss": 0.9252,
+      "step": 316
+    },
+    {
+      "epoch": 0.5123232323232323,
+      "grad_norm": 7.125,
+      "learning_rate": 9.274966418581958e-06,
+      "loss": 0.5309,
+      "step": 317
+    },
+    {
+      "epoch": 0.5139393939393939,
+      "grad_norm": 8.0,
+      "learning_rate": 9.267623141662734e-06,
+      "loss": 0.6308,
+      "step": 318
+    },
+    {
+      "epoch": 0.5155555555555555,
+      "grad_norm": 8.875,
+      "learning_rate": 9.260245802288694e-06,
+      "loss": 0.7667,
+      "step": 319
+    },
+    {
+      "epoch": 0.5171717171717172,
+      "grad_norm": 12.9375,
+      "learning_rate": 9.252834459342801e-06,
+      "loss": 0.8895,
+      "step": 320
+    },
+    {
+      "epoch": 0.5187878787878788,
+      "grad_norm": 7.03125,
+      "learning_rate": 9.24538917197942e-06,
+      "loss": 0.3923,
+      "step": 321
+    },
+    {
+      "epoch": 0.5204040404040404,
+      "grad_norm": 11.125,
+      "learning_rate": 9.237909999623847e-06,
+      "loss": 0.9289,
+      "step": 322
+    },
+    {
+      "epoch": 0.522020202020202,
+      "grad_norm": 10.0625,
+      "learning_rate": 9.230397001971829e-06,
+      "loss": 0.8841,
+      "step": 323
+    },
+    {
+      "epoch": 0.5236363636363637,
+      "grad_norm": 10.0,
+      "learning_rate": 9.222850238989104e-06,
+      "loss": 1.0293,
+      "step": 324
+    },
+    {
+      "epoch": 0.5252525252525253,
+      "grad_norm": 8.5625,
+      "learning_rate": 9.215269770910897e-06,
+      "loss": 0.9055,
+      "step": 325
+    },
+    {
+      "epoch": 0.5268686868686868,
+      "grad_norm": 10.0625,
+      "learning_rate": 9.207655658241469e-06,
+      "loss": 0.4496,
+      "step": 326
+    },
+    {
+      "epoch": 0.5284848484848484,
+      "grad_norm": 9.25,
+      "learning_rate": 9.200007961753605e-06,
+      "loss": 1.0731,
+      "step": 327
+    },
+    {
+      "epoch": 0.5301010101010101,
+      "grad_norm": 8.1875,
+      "learning_rate": 9.192326742488153e-06,
+      "loss": 0.672,
+      "step": 328
+    },
+    {
+      "epoch": 0.5317171717171717,
+      "grad_norm": 10.9375,
+      "learning_rate": 9.184612061753518e-06,
+      "loss": 1.0366,
+      "step": 329
+    },
+    {
+      "epoch": 0.5333333333333333,
+      "grad_norm": 8.625,
+      "learning_rate": 9.176863981125185e-06,
+      "loss": 0.9249,
+      "step": 330
+    },
+    {
+      "epoch": 0.534949494949495,
+      "grad_norm": 8.1875,
+      "learning_rate": 9.169082562445228e-06,
+      "loss": 0.6611,
+      "step": 331
+    },
+    {
+      "epoch": 0.5365656565656566,
+      "grad_norm": 7.625,
+      "learning_rate": 9.161267867821802e-06,
+      "loss": 0.7343,
+      "step": 332
+    },
+    {
+      "epoch": 0.5381818181818182,
+      "grad_norm": 8.75,
+      "learning_rate": 9.153419959628666e-06,
+      "loss": 0.9306,
+      "step": 333
+    },
+    {
+      "epoch": 0.5397979797979798,
+      "grad_norm": 6.75,
+      "learning_rate": 9.14553890050467e-06,
+      "loss": 0.7274,
+      "step": 334
+    },
+    {
+      "epoch": 0.5414141414141415,
+      "grad_norm": 9.1875,
+      "learning_rate": 9.137624753353268e-06,
+      "loss": 0.5954,
+      "step": 335
+    },
+    {
+      "epoch": 0.5430303030303031,
+      "grad_norm": 7.125,
+      "learning_rate": 9.129677581342e-06,
+      "loss": 1.0952,
+      "step": 336
+    },
+    {
+      "epoch": 0.5446464646464646,
+      "grad_norm": 8.375,
+      "learning_rate": 9.121697447902006e-06,
+      "loss": 0.6458,
+      "step": 337
+    },
+    {
+      "epoch": 0.5462626262626262,
+      "grad_norm": 8.1875,
+      "learning_rate": 9.113684416727511e-06,
+      "loss": 0.6578,
+      "step": 338
+    },
+    {
+      "epoch": 0.5478787878787879,
+      "grad_norm": 8.3125,
+      "learning_rate": 9.10563855177531e-06,
+      "loss": 0.6899,
+      "step": 339
+    },
+    {
+      "epoch": 0.5494949494949495,
+      "grad_norm": 10.375,
+      "learning_rate": 9.097559917264268e-06,
+      "loss": 0.7607,
+      "step": 340
+    },
+    {
+      "epoch": 0.5511111111111111,
+      "grad_norm": 9.875,
+      "learning_rate": 9.089448577674799e-06,
+      "loss": 0.6255,
+      "step": 341
+    },
+    {
+      "epoch": 0.5527272727272727,
+      "grad_norm": 8.6875,
+      "learning_rate": 9.081304597748366e-06,
+      "loss": 0.9227,
+      "step": 342
+    },
+    {
+      "epoch": 0.5543434343434344,
+      "grad_norm": 6.71875,
+      "learning_rate": 9.073128042486945e-06,
+      "loss": 0.6744,
+      "step": 343
+    },
+    {
+      "epoch": 0.555959595959596,
+      "grad_norm": 6.65625,
+      "learning_rate": 9.064918977152517e-06,
+      "loss": 0.4926,
+      "step": 344
+    },
+    {
+      "epoch": 0.5575757575757576,
+      "grad_norm": 6.90625,
+      "learning_rate": 9.056677467266543e-06,
+      "loss": 0.5658,
+      "step": 345
+    },
+    {
+      "epoch": 0.5591919191919192,
+      "grad_norm": 8.3125,
+      "learning_rate": 9.048403578609454e-06,
+      "loss": 0.9654,
+      "step": 346
+    },
+    {
+      "epoch": 0.5608080808080808,
+      "grad_norm": 9.4375,
+      "learning_rate": 9.040097377220103e-06,
+      "loss": 0.8399,
+      "step": 347
+    },
+    {
+      "epoch": 0.5624242424242424,
+      "grad_norm": 8.875,
+      "learning_rate": 9.031758929395259e-06,
+      "loss": 0.7202,
+      "step": 348
+    },
+    {
+      "epoch": 0.564040404040404,
+      "grad_norm": 9.3125,
+      "learning_rate": 9.02338830168906e-06,
+      "loss": 0.5778,
+      "step": 349
+    },
+    {
+      "epoch": 0.5656565656565656,
+      "grad_norm": 6.375,
+      "learning_rate": 9.014985560912499e-06,
+      "loss": 0.4703,
+      "step": 350
+    },
+    {
+      "epoch": 0.5672727272727273,
+      "grad_norm": 8.5625,
+      "learning_rate": 9.00655077413288e-06,
+      "loss": 0.8051,
+      "step": 351
+    },
+    {
+      "epoch": 0.5688888888888889,
+      "grad_norm": 5.9375,
+      "learning_rate": 8.998084008673284e-06,
+      "loss": 0.5963,
+      "step": 352
+    },
+    {
+      "epoch": 0.5705050505050505,
+      "grad_norm": 8.1875,
+      "learning_rate": 8.989585332112039e-06,
+      "loss": 0.9701,
+      "step": 353
+    },
+    {
+      "epoch": 0.5721212121212121,
+      "grad_norm": 9.9375,
+      "learning_rate": 8.981054812282162e-06,
+      "loss": 1.0809,
+      "step": 354
+    },
+    {
+      "epoch": 0.5737373737373738,
+      "grad_norm": 8.1875,
+      "learning_rate": 8.972492517270845e-06,
+      "loss": 0.8909,
+      "step": 355
+    },
+    {
+      "epoch": 0.5753535353535354,
+      "grad_norm": 7.71875,
+      "learning_rate": 8.963898515418885e-06,
+      "loss": 0.5967,
+      "step": 356
+    },
+    {
+      "epoch": 0.576969696969697,
+      "grad_norm": 7.34375,
+      "learning_rate": 8.955272875320159e-06,
+      "loss": 0.74,
+      "step": 357
+    },
+    {
+      "epoch": 0.5785858585858585,
+      "grad_norm": 7.71875,
+      "learning_rate": 8.946615665821059e-06,
+      "loss": 0.6306,
+      "step": 358
+    },
+    {
+      "epoch": 0.5802020202020202,
+      "grad_norm": 7.0625,
+      "learning_rate": 8.937926956019957e-06,
+      "loss": 0.5952,
+      "step": 359
+    },
+    {
+      "epoch": 0.5818181818181818,
+      "grad_norm": 9.0,
+      "learning_rate": 8.929206815266653e-06,
+      "loss": 0.5917,
+      "step": 360
+    },
+    {
+      "epoch": 0.5834343434343434,
+      "grad_norm": 6.9375,
+      "learning_rate": 8.920455313161801e-06,
+      "loss": 0.6248,
+      "step": 361
+    },
+    {
+      "epoch": 0.585050505050505,
+      "grad_norm": 8.375,
+      "learning_rate": 8.911672519556386e-06,
+      "loss": 0.7684,
+      "step": 362
+    },
+    {
+      "epoch": 0.5866666666666667,
+      "grad_norm": 7.125,
+      "learning_rate": 8.902858504551136e-06,
+      "loss": 0.6592,
+      "step": 363
+    },
+    {
+      "epoch": 0.5882828282828283,
+      "grad_norm": 8.6875,
+      "learning_rate": 8.89401333849598e-06,
+      "loss": 0.6805,
+      "step": 364
+    },
+    {
+      "epoch": 0.5898989898989899,
+      "grad_norm": 8.6875,
+      "learning_rate": 8.885137091989488e-06,
+      "loss": 0.8623,
+      "step": 365
+    },
+    {
+      "epoch": 0.5915151515151515,
+      "grad_norm": 8.1875,
+      "learning_rate": 8.87622983587829e-06,
+      "loss": 1.0672,
+      "step": 366
+    },
+    {
+      "epoch": 0.5931313131313132,
+      "grad_norm": 7.8125,
+      "learning_rate": 8.867291641256534e-06,
+      "loss": 1.091,
+      "step": 367
+    },
+    {
+      "epoch": 0.5947474747474748,
+      "grad_norm": 12.1875,
+      "learning_rate": 8.8583225794653e-06,
+      "loss": 0.7798,
+      "step": 368
+    },
+    {
+      "epoch": 0.5963636363636363,
+      "grad_norm": 6.53125,
+      "learning_rate": 8.849322722092037e-06,
+      "loss": 0.5133,
+      "step": 369
+    },
+    {
+      "epoch": 0.5963636363636363,
+      "eval_loss": 0.7914893627166748,
+      "eval_runtime": 35.1895,
+      "eval_samples_per_second": 31.259,
+      "eval_steps_per_second": 3.922,
+      "step": 369
+    },
+    {
+      "epoch": 0.597979797979798,
+      "grad_norm": 7.65625,
+      "learning_rate": 8.840292140969995e-06,
+      "loss": 0.9505,
+      "step": 370
+    },
+    {
+      "epoch": 0.5995959595959596,
+      "grad_norm": 8.1875,
+      "learning_rate": 8.83123090817765e-06,
+      "loss": 0.6006,
+      "step": 371
+    },
+    {
+      "epoch": 0.6012121212121212,
+      "grad_norm": 8.1875,
+      "learning_rate": 8.82213909603812e-06,
+      "loss": 0.7318,
+      "step": 372
+    },
+    {
+      "epoch": 0.6028282828282828,
+      "grad_norm": 8.125,
+      "learning_rate": 8.813016777118604e-06,
+      "loss": 0.725,
+      "step": 373
+    },
+    {
+      "epoch": 0.6044444444444445,
+      "grad_norm": 8.0,
+      "learning_rate": 8.803864024229786e-06,
+      "loss": 0.8758,
+      "step": 374
+    },
+    {
+      "epoch": 0.6060606060606061,
+      "grad_norm": 7.6875,
+      "learning_rate": 8.79468091042527e-06,
+      "loss": 0.8636,
+      "step": 375
+    },
+    {
+      "epoch": 0.6076767676767677,
+      "grad_norm": 9.75,
+      "learning_rate": 8.78546750900098e-06,
+      "loss": 0.7711,
+      "step": 376
+    },
+    {
+      "epoch": 0.6092929292929293,
+      "grad_norm": 7.21875,
+      "learning_rate": 8.77622389349459e-06,
+      "loss": 0.5907,
+      "step": 377
+    },
+    {
+      "epoch": 0.610909090909091,
+      "grad_norm": 6.15625,
+      "learning_rate": 8.766950137684929e-06,
+      "loss": 0.6627,
+      "step": 378
+    },
+    {
+      "epoch": 0.6125252525252525,
+      "grad_norm": 8.125,
+      "learning_rate": 8.757646315591391e-06,
+      "loss": 0.9963,
+      "step": 379
+    },
+    {
+      "epoch": 0.6141414141414141,
+      "grad_norm": 6.25,
+      "learning_rate": 8.748312501473351e-06,
+      "loss": 0.8233,
+      "step": 380
+    },
+    {
+      "epoch": 0.6157575757575757,
+      "grad_norm": 7.5,
+      "learning_rate": 8.738948769829565e-06,
+      "loss": 0.5933,
+      "step": 381
+    },
+    {
+      "epoch": 0.6173737373737374,
+      "grad_norm": 6.8125,
+      "learning_rate": 8.72955519539758e-06,
+      "loss": 0.4457,
+      "step": 382
+    },
+    {
+      "epoch": 0.618989898989899,
+      "grad_norm": 8.1875,
+      "learning_rate": 8.72013185315314e-06,
+      "loss": 0.6291,
+      "step": 383
+    },
+    {
+      "epoch": 0.6206060606060606,
+      "grad_norm": 4.6875,
+      "learning_rate": 8.710678818309576e-06,
+      "loss": 0.5969,
+      "step": 384
+    },
+    {
+      "epoch": 0.6222222222222222,
+      "grad_norm": 6.1875,
+      "learning_rate": 8.701196166317221e-06,
+      "loss": 0.4308,
+      "step": 385
+    },
+    {
+      "epoch": 0.6238383838383839,
+      "grad_norm": 9.5,
+      "learning_rate": 8.691683972862792e-06,
+      "loss": 0.8222,
+      "step": 386
+    },
+    {
+      "epoch": 0.6254545454545455,
+      "grad_norm": 8.125,
+      "learning_rate": 8.682142313868802e-06,
+      "loss": 0.8346,
+      "step": 387
+    },
+    {
+      "epoch": 0.6270707070707071,
+      "grad_norm": 7.40625,
+      "learning_rate": 8.672571265492944e-06,
+      "loss": 0.7311,
+      "step": 388
+    },
+    {
+      "epoch": 0.6286868686868687,
+      "grad_norm": 8.0625,
+      "learning_rate": 8.66297090412748e-06,
+      "loss": 0.7654,
+      "step": 389
+    },
+    {
+      "epoch": 0.6303030303030303,
+      "grad_norm": 9.25,
+      "learning_rate": 8.653341306398645e-06,
+      "loss": 0.8764,
+      "step": 390
+    },
+    {
+      "epoch": 0.6319191919191919,
+      "grad_norm": 7.75,
+      "learning_rate": 8.643682549166021e-06,
+      "loss": 0.8139,
+      "step": 391
+    },
+    {
+      "epoch": 0.6335353535353535,
+      "grad_norm": 7.4375,
+      "learning_rate": 8.63399470952193e-06,
+      "loss": 0.8704,
+      "step": 392
+    },
+    {
+      "epoch": 0.6351515151515151,
+      "grad_norm": 6.6875,
+      "learning_rate": 8.62427786479082e-06,
+      "loss": 0.5697,
+      "step": 393
+    },
+    {
+      "epoch": 0.6367676767676768,
+      "grad_norm": 7.46875,
+      "learning_rate": 8.614532092528645e-06,
+      "loss": 0.6327,
+      "step": 394
+    },
+    {
+      "epoch": 0.6383838383838384,
+      "grad_norm": 7.75,
+      "learning_rate": 8.604757470522247e-06,
+      "loss": 0.7221,
+      "step": 395
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 7.28125,
+      "learning_rate": 8.594954076788736e-06,
+      "loss": 0.7422,
+      "step": 396
+    },
+    {
+      "epoch": 0.6416161616161616,
+      "grad_norm": 14.1875,
+      "learning_rate": 8.585121989574865e-06,
+      "loss": 0.6289,
+      "step": 397
+    },
+    {
+      "epoch": 0.6432323232323233,
+      "grad_norm": 7.65625,
+      "learning_rate": 8.575261287356407e-06,
+      "loss": 0.8722,
+      "step": 398
+    },
+    {
+      "epoch": 0.6448484848484849,
+      "grad_norm": 7.125,
+      "learning_rate": 8.565372048837529e-06,
+      "loss": 0.7178,
+      "step": 399
+    },
+    {
+      "epoch": 0.6464646464646465,
+      "grad_norm": 6.90625,
+      "learning_rate": 8.555454352950161e-06,
+      "loss": 0.6246,
+      "step": 400
+    },
+    {
+      "epoch": 0.648080808080808,
+      "grad_norm": 8.4375,
+      "learning_rate": 8.545508278853373e-06,
+      "loss": 0.8206,
+      "step": 401
+    },
+    {
+      "epoch": 0.6496969696969697,
+      "grad_norm": 10.125,
+      "learning_rate": 8.535533905932739e-06,
+      "loss": 0.9123,
+      "step": 402
+    },
+    {
+      "epoch": 0.6513131313131313,
+      "grad_norm": 8.5625,
+      "learning_rate": 8.525531313799693e-06,
+      "loss": 0.9784,
+      "step": 403
+    },
+    {
+      "epoch": 0.6529292929292929,
+      "grad_norm": 8.9375,
+      "learning_rate": 8.515500582290914e-06,
+      "loss": 0.9171,
+      "step": 404
+    },
+    {
+      "epoch": 0.6545454545454545,
+      "grad_norm": 5.84375,
+      "learning_rate": 8.505441791467674e-06,
+      "loss": 0.5419,
+      "step": 405
+    },
+    {
+      "epoch": 0.6561616161616162,
+      "grad_norm": 6.625,
+      "learning_rate": 8.495355021615204e-06,
+      "loss": 0.6661,
+      "step": 406
+    },
+    {
+      "epoch": 0.6577777777777778,
+      "grad_norm": 7.625,
+      "learning_rate": 8.48524035324205e-06,
+      "loss": 0.8925,
+      "step": 407
+    },
+    {
+      "epoch": 0.6593939393939394,
+      "grad_norm": 7.0625,
+      "learning_rate": 8.475097867079437e-06,
+      "loss": 0.5441,
+      "step": 408
+    },
+    {
+      "epoch": 0.661010101010101,
+      "grad_norm": 6.21875,
+      "learning_rate": 8.464927644080617e-06,
+      "loss": 0.5148,
+      "step": 409
+    },
+    {
+      "epoch": 0.6626262626262627,
+      "grad_norm": 8.125,
+      "learning_rate": 8.454729765420228e-06,
+      "loss": 0.7649,
+      "step": 410
+    },
+    {
+      "epoch": 0.6642424242424242,
+      "grad_norm": 7.65625,
+      "learning_rate": 8.444504312493642e-06,
+      "loss": 0.6308,
+      "step": 411
+    },
+    {
+      "epoch": 0.6658585858585858,
+      "grad_norm": 9.9375,
+      "learning_rate": 8.434251366916323e-06,
+      "loss": 0.6534,
+      "step": 412
+    },
+    {
+      "epoch": 0.6674747474747474,
+      "grad_norm": 7.46875,
+      "learning_rate": 8.423971010523165e-06,
+      "loss": 0.8665,
+      "step": 413
+    },
+    {
+      "epoch": 0.6690909090909091,
+      "grad_norm": 6.78125,
+      "learning_rate": 8.413663325367845e-06,
+      "loss": 0.5408,
+      "step": 414
+    },
+    {
+      "epoch": 0.6707070707070707,
+      "grad_norm": 6.59375,
+      "learning_rate": 8.403328393722169e-06,
+      "loss": 0.6455,
+      "step": 415
+    },
+    {
+      "epoch": 0.6723232323232323,
+      "grad_norm": 6.5625,
+      "learning_rate": 8.392966298075413e-06,
+      "loss": 0.6351,
+      "step": 416
+    },
+    {
+      "epoch": 0.673939393939394,
+      "grad_norm": 9.5625,
+      "learning_rate": 8.382577121133664e-06,
+      "loss": 0.6295,
+      "step": 417
+    },
+    {
+      "epoch": 0.6755555555555556,
+      "grad_norm": 7.625,
+      "learning_rate": 8.372160945819164e-06,
+      "loss": 1.2937,
+      "step": 418
+    },
+    {
+      "epoch": 0.6771717171717172,
+      "grad_norm": 9.25,
+      "learning_rate": 8.361717855269643e-06,
+      "loss": 0.731,
+      "step": 419
+    },
+    {
+      "epoch": 0.6787878787878788,
+      "grad_norm": 6.6875,
+      "learning_rate": 8.351247932837655e-06,
+      "loss": 0.5192,
+      "step": 420
+    },
+    {
+      "epoch": 0.6804040404040405,
+      "grad_norm": 8.125,
+      "learning_rate": 8.34075126208992e-06,
+      "loss": 0.8637,
+      "step": 421
+    },
+    {
+      "epoch": 0.682020202020202,
+      "grad_norm": 7.78125,
+      "learning_rate": 8.330227926806652e-06,
+      "loss": 0.6211,
+      "step": 422
+    },
+    {
+      "epoch": 0.6836363636363636,
+      "grad_norm": 7.71875,
+      "learning_rate": 8.319678010980883e-06,
+      "loss": 0.7818,
+      "step": 423
+    },
+    {
+      "epoch": 0.6852525252525252,
+      "grad_norm": 5.5625,
+      "learning_rate": 8.309101598817812e-06,
+      "loss": 0.4556,
+      "step": 424
+    },
+    {
+      "epoch": 0.6868686868686869,
+      "grad_norm": 5.875,
+      "learning_rate": 8.298498774734114e-06,
+      "loss": 0.6321,
+      "step": 425
+    },
+    {
+      "epoch": 0.6884848484848485,
+      "grad_norm": 7.375,
+      "learning_rate": 8.287869623357275e-06,
+      "loss": 0.8331,
+      "step": 426
+    },
+    {
+      "epoch": 0.6901010101010101,
+      "grad_norm": 8.625,
+      "learning_rate": 8.277214229524913e-06,
+      "loss": 0.7505,
+      "step": 427
+    },
+    {
+      "epoch": 0.6917171717171717,
+      "grad_norm": 6.34375,
+      "learning_rate": 8.266532678284103e-06,
+      "loss": 0.6113,
+      "step": 428
+    },
+    {
+      "epoch": 0.6933333333333334,
+      "grad_norm": 8.0,
+      "learning_rate": 8.255825054890705e-06,
+      "loss": 0.5578,
+      "step": 429
+    },
+    {
+      "epoch": 0.694949494949495,
+      "grad_norm": 9.0,
+      "learning_rate": 8.245091444808663e-06,
+      "loss": 1.0137,
+      "step": 430
+    },
+    {
+      "epoch": 0.6965656565656566,
+      "grad_norm": 7.34375,
+      "learning_rate": 8.234331933709353e-06,
+      "loss": 0.5748,
+      "step": 431
+    },
+    {
+      "epoch": 0.6981818181818182,
+      "grad_norm": 7.15625,
+      "learning_rate": 8.223546607470863e-06,
+      "loss": 0.6973,
+      "step": 432
+    },
+    {
+      "epoch": 0.6997979797979798,
+      "grad_norm": 8.375,
+      "learning_rate": 8.212735552177347e-06,
+      "loss": 0.6268,
+      "step": 433
+    },
+    {
+      "epoch": 0.7014141414141414,
+      "grad_norm": 6.875,
+      "learning_rate": 8.201898854118301e-06,
+      "loss": 0.6871,
+      "step": 434
+    },
+    {
+      "epoch": 0.703030303030303,
+      "grad_norm": 9.5,
+      "learning_rate": 8.191036599787908e-06,
+      "loss": 0.8489,
+      "step": 435
+    },
+    {
+      "epoch": 0.7046464646464646,
+      "grad_norm": 6.875,
+      "learning_rate": 8.18014887588431e-06,
+      "loss": 0.6502,
+      "step": 436
+    },
+    {
+      "epoch": 0.7062626262626263,
+      "grad_norm": 6.21875,
+      "learning_rate": 8.169235769308959e-06,
+      "loss": 0.5236,
+      "step": 437
+    },
+    {
+      "epoch": 0.7078787878787879,
+      "grad_norm": 10.4375,
+      "learning_rate": 8.158297367165885e-06,
+      "loss": 0.8233,
+      "step": 438
+    },
+    {
+      "epoch": 0.7094949494949495,
+      "grad_norm": 9.3125,
+      "learning_rate": 8.147333756761027e-06,
+      "loss": 0.7471,
+      "step": 439
+    },
+    {
+      "epoch": 0.7111111111111111,
+      "grad_norm": 9.1875,
+      "learning_rate": 8.13634502560152e-06,
+      "loss": 0.627,
+      "step": 440
+    },
+    {
+      "epoch": 0.7127272727272728,
+      "grad_norm": 6.6875,
+      "learning_rate": 8.125331261395004e-06,
+      "loss": 0.6162,
+      "step": 441
+    },
+    {
+      "epoch": 0.7143434343434344,
+      "grad_norm": 9.25,
+      "learning_rate": 8.114292552048925e-06,
+      "loss": 0.7379,
+      "step": 442
+    },
+    {
+      "epoch": 0.7159595959595959,
+      "grad_norm": 6.125,
+      "learning_rate": 8.10322898566983e-06,
+      "loss": 0.6932,
+      "step": 443
+    },
+    {
+      "epoch": 0.7175757575757575,
+      "grad_norm": 6.0,
+      "learning_rate": 8.092140650562665e-06,
+      "loss": 0.5414,
+      "step": 444
+    },
+    {
+      "epoch": 0.7191919191919192,
+      "grad_norm": 6.8125,
+      "learning_rate": 8.08102763523007e-06,
+      "loss": 0.6522,
+      "step": 445
+    },
+    {
+      "epoch": 0.7208080808080808,
+      "grad_norm": 6.15625,
+      "learning_rate": 8.069890028371672e-06,
+      "loss": 0.5511,
+      "step": 446
+    },
+    {
+      "epoch": 0.7224242424242424,
+      "grad_norm": 8.5,
+      "learning_rate": 8.058727918883376e-06,
+      "loss": 0.9194,
+      "step": 447
+    },
+    {
+      "epoch": 0.724040404040404,
+      "grad_norm": 6.15625,
+      "learning_rate": 8.047541395856661e-06,
+      "loss": 0.5101,
+      "step": 448
+    },
+    {
+      "epoch": 0.7256565656565657,
+      "grad_norm": 6.78125,
+      "learning_rate": 8.036330548577866e-06,
+      "loss": 0.664,
+      "step": 449
+    },
+    {
+      "epoch": 0.7272727272727273,
+      "grad_norm": 7.21875,
+      "learning_rate": 8.025095466527468e-06,
+      "loss": 0.6514,
+      "step": 450
+    },
+    {
+      "epoch": 0.7288888888888889,
+      "grad_norm": 8.25,
+      "learning_rate": 8.013836239379388e-06,
+      "loss": 0.8009,
+      "step": 451
+    },
+    {
+      "epoch": 0.7305050505050505,
+      "grad_norm": 6.59375,
+      "learning_rate": 8.002552957000254e-06,
+      "loss": 0.5222,
+      "step": 452
+    },
+    {
+      "epoch": 0.7321212121212122,
+      "grad_norm": 6.71875,
+      "learning_rate": 7.991245709448697e-06,
+      "loss": 0.6088,
+      "step": 453
+    },
+    {
+      "epoch": 0.7337373737373737,
+      "grad_norm": 7.65625,
+      "learning_rate": 7.979914586974628e-06,
+      "loss": 0.8188,
+      "step": 454
+    },
+    {
+      "epoch": 0.7353535353535353,
+      "grad_norm": 6.96875,
+      "learning_rate": 7.968559680018518e-06,
+      "loss": 0.696,
+      "step": 455
+    },
+    {
+      "epoch": 0.7369696969696969,
+      "grad_norm": 7.90625,
+      "learning_rate": 7.957181079210676e-06,
+      "loss": 0.7024,
+      "step": 456
+    },
+    {
+      "epoch": 0.7385858585858586,
+      "grad_norm": 9.25,
+      "learning_rate": 7.945778875370527e-06,
+      "loss": 0.9176,
+      "step": 457
+    },
+    {
+      "epoch": 0.7402020202020202,
+      "grad_norm": 7.46875,
+      "learning_rate": 7.934353159505885e-06,
+      "loss": 0.9633,
+      "step": 458
+    },
+    {
+      "epoch": 0.7418181818181818,
+      "grad_norm": 7.28125,
+      "learning_rate": 7.922904022812228e-06,
+      "loss": 0.6307,
+      "step": 459
+    },
+    {
+      "epoch": 0.7434343434343434,
+      "grad_norm": 6.84375,
+      "learning_rate": 7.911431556671967e-06,
+      "loss": 0.6905,
+      "step": 460
+    },
+    {
+      "epoch": 0.7450505050505051,
+      "grad_norm": 6.875,
+      "learning_rate": 7.899935852653723e-06,
+      "loss": 0.4554,
+      "step": 461
+    },
+    {
+      "epoch": 0.7466666666666667,
+      "grad_norm": 6.21875,
+      "learning_rate": 7.888417002511592e-06,
+      "loss": 0.5635,
+      "step": 462
+    },
+    {
+      "epoch": 0.7482828282828283,
+      "grad_norm": 8.875,
+      "learning_rate": 7.876875098184411e-06,
+      "loss": 1.0392,
+      "step": 463
+    },
+    {
+      "epoch": 0.74989898989899,
+      "grad_norm": 5.875,
+      "learning_rate": 7.865310231795026e-06,
+      "loss": 0.5778,
+      "step": 464
+    },
+    {
+      "epoch": 0.7515151515151515,
+      "grad_norm": 7.09375,
+      "learning_rate": 7.853722495649558e-06,
+      "loss": 0.6214,
+      "step": 465
+    },
+    {
+      "epoch": 0.7531313131313131,
+      "grad_norm": 7.1875,
+      "learning_rate": 7.842111982236669e-06,
+      "loss": 0.8281,
+      "step": 466
+    },
+    {
+      "epoch": 0.7547474747474747,
+      "grad_norm": 7.9375,
+      "learning_rate": 7.83047878422681e-06,
+      "loss": 0.7688,
+      "step": 467
+    },
+    {
+      "epoch": 0.7563636363636363,
+      "grad_norm": 9.0625,
+      "learning_rate": 7.818822994471504e-06,
+      "loss": 0.6517,
+      "step": 468
+    },
+    {
+      "epoch": 0.757979797979798,
+      "grad_norm": 6.65625,
+      "learning_rate": 7.807144706002582e-06,
+      "loss": 0.7734,
+      "step": 469
+    },
+    {
+      "epoch": 0.7595959595959596,
+      "grad_norm": 7.4375,
+      "learning_rate": 7.79544401203146e-06,
+      "loss": 0.7312,
+      "step": 470
+    },
+    {
+      "epoch": 0.7612121212121212,
+      "grad_norm": 6.5625,
+      "learning_rate": 7.783721005948374e-06,
+      "loss": 0.8063,
+      "step": 471
+    },
+    {
+      "epoch": 0.7628282828282829,
+      "grad_norm": 7.96875,
+      "learning_rate": 7.771975781321655e-06,
+      "loss": 0.6811,
+      "step": 472
+    },
+    {
+      "epoch": 0.7644444444444445,
+      "grad_norm": 6.4375,
+      "learning_rate": 7.760208431896971e-06,
+      "loss": 0.866,
+      "step": 473
+    },
+    {
+      "epoch": 0.7660606060606061,
+      "grad_norm": 7.0625,
+      "learning_rate": 7.748419051596586e-06,
+      "loss": 0.8808,
+      "step": 474
+    },
+    {
+      "epoch": 0.7676767676767676,
+      "grad_norm": 7.9375,
+      "learning_rate": 7.736607734518593e-06,
+      "loss": 0.9093,
+      "step": 475
+    },
+    {
+      "epoch": 0.7692929292929293,
+      "grad_norm": 8.3125,
+      "learning_rate": 7.72477457493619e-06,
+      "loss": 0.7615,
+      "step": 476
+    },
+    {
+      "epoch": 0.7709090909090909,
+      "grad_norm": 6.6875,
+      "learning_rate": 7.712919667296902e-06,
+      "loss": 0.5422,
+      "step": 477
+    },
+    {
+      "epoch": 0.7725252525252525,
+      "grad_norm": 5.8125,
+      "learning_rate": 7.701043106221847e-06,
+      "loss": 0.5013,
+      "step": 478
+    },
+    {
+      "epoch": 0.7741414141414141,
+      "grad_norm": 9.25,
+      "learning_rate": 7.689144986504966e-06,
+      "loss": 0.8879,
+      "step": 479
+    },
+    {
+      "epoch": 0.7757575757575758,
+      "grad_norm": 7.40625,
+      "learning_rate": 7.677225403112277e-06,
+      "loss": 0.5958,
+      "step": 480
+    },
+    {
+      "epoch": 0.7773737373737374,
+      "grad_norm": 6.59375,
+      "learning_rate": 7.665284451181106e-06,
+      "loss": 0.5568,
+      "step": 481
+    },
+    {
+      "epoch": 0.778989898989899,
+      "grad_norm": 8.25,
+      "learning_rate": 7.653322226019341e-06,
+      "loss": 0.8891,
+      "step": 482
+    },
+    {
+      "epoch": 0.7806060606060606,
+      "grad_norm": 6.59375,
+      "learning_rate": 7.64133882310466e-06,
+      "loss": 0.6576,
+      "step": 483
+    },
+    {
+      "epoch": 0.7822222222222223,
+      "grad_norm": 7.5,
+      "learning_rate": 7.629334338083774e-06,
+      "loss": 0.7541,
+      "step": 484
+    },
+    {
+      "epoch": 0.7838383838383839,
+      "grad_norm": 8.3125,
+      "learning_rate": 7.617308866771667e-06,
+      "loss": 0.9125,
+      "step": 485
+    },
+    {
+      "epoch": 0.7854545454545454,
+      "grad_norm": 7.25,
+      "learning_rate": 7.605262505150819e-06,
+      "loss": 0.7351,
+      "step": 486
+    },
+    {
+      "epoch": 0.787070707070707,
+      "grad_norm": 6.625,
+      "learning_rate": 7.5931953493704545e-06,
+      "loss": 0.5468,
+      "step": 487
+    },
+    {
+      "epoch": 0.7886868686868687,
+      "grad_norm": 7.21875,
+      "learning_rate": 7.58110749574577e-06,
+      "loss": 0.5995,
+      "step": 488
+    },
+    {
+      "epoch": 0.7903030303030303,
+      "grad_norm": 7.78125,
+      "learning_rate": 7.568999040757157e-06,
+      "loss": 0.6156,
+      "step": 489
+    },
+    {
+      "epoch": 0.7919191919191919,
+      "grad_norm": 6.46875,
+      "learning_rate": 7.556870081049444e-06,
+      "loss": 0.7599,
+      "step": 490
+    },
+    {
+      "epoch": 0.7935353535353535,
+      "grad_norm": 6.90625,
+      "learning_rate": 7.5447207134311195e-06,
+      "loss": 0.7025,
+      "step": 491
+    },
+    {
+      "epoch": 0.7951515151515152,
+      "grad_norm": 6.9375,
+      "learning_rate": 7.532551034873558e-06,
+      "loss": 0.5702,
+      "step": 492
+    },
+    {
+      "epoch": 0.7951515151515152,
+      "eval_loss": 0.7591387629508972,
+      "eval_runtime": 40.0722,
+      "eval_samples_per_second": 27.45,
+      "eval_steps_per_second": 3.444,
+      "step": 492
+    },
+    {
+      "epoch": 0.7967676767676768,
+      "grad_norm": 7.28125,
+      "learning_rate": 7.520361142510248e-06,
+      "loss": 0.6548,
+      "step": 493
+    },
+    {
+      "epoch": 0.7983838383838384,
+      "grad_norm": 8.9375,
+      "learning_rate": 7.50815113363602e-06,
+      "loss": 0.8229,
+      "step": 494
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 7.59375,
+      "learning_rate": 7.49592110570626e-06,
+      "loss": 1.0282,
+      "step": 495
+    },
+    {
+      "epoch": 0.8016161616161617,
+      "grad_norm": 9.375,
+      "learning_rate": 7.483671156336142e-06,
+      "loss": 0.9156,
+      "step": 496
+    },
+    {
+      "epoch": 0.8032323232323232,
+      "grad_norm": 7.25,
+      "learning_rate": 7.471401383299847e-06,
+      "loss": 0.6175,
+      "step": 497
+    },
+    {
+      "epoch": 0.8048484848484848,
+      "grad_norm": 7.78125,
+      "learning_rate": 7.459111884529775e-06,
+      "loss": 0.7255,
+      "step": 498
+    },
+    {
+      "epoch": 0.8064646464646464,
+      "grad_norm": 8.9375,
+      "learning_rate": 7.446802758115775e-06,
+      "loss": 0.8077,
+      "step": 499
+    },
+    {
+      "epoch": 0.8080808080808081,
+      "grad_norm": 5.34375,
+      "learning_rate": 7.43447410230435e-06,
+      "loss": 0.7292,
+      "step": 500
+    },
+    {
+      "epoch": 0.8096969696969697,
+      "grad_norm": 7.71875,
+      "learning_rate": 7.422126015497884e-06,
+      "loss": 0.7707,
+      "step": 501
+    },
+    {
+      "epoch": 0.8113131313131313,
+      "grad_norm": 5.53125,
+      "learning_rate": 7.409758596253849e-06,
+      "loss": 0.5303,
+      "step": 502
+    },
+    {
+      "epoch": 0.812929292929293,
+      "grad_norm": 6.59375,
+      "learning_rate": 7.397371943284017e-06,
+      "loss": 0.4893,
+      "step": 503
+    },
+    {
+      "epoch": 0.8145454545454546,
+      "grad_norm": 7.625,
+      "learning_rate": 7.384966155453686e-06,
+      "loss": 0.6416,
+      "step": 504
+    },
+    {
+      "epoch": 0.8161616161616162,
+      "grad_norm": 7.8125,
+      "learning_rate": 7.372541331780871e-06,
+      "loss": 0.8562,
+      "step": 505
+    },
+    {
+      "epoch": 0.8177777777777778,
+      "grad_norm": 6.71875,
+      "learning_rate": 7.360097571435527e-06,
+      "loss": 0.6703,
+      "step": 506
+    },
+    {
+      "epoch": 0.8193939393939393,
+      "grad_norm": 7.6875,
+      "learning_rate": 7.347634973738753e-06,
+      "loss": 1.2725,
+      "step": 507
+    },
+    {
+      "epoch": 0.821010101010101,
+      "grad_norm": 6.125,
+      "learning_rate": 7.335153638162005e-06,
+      "loss": 0.6887,
+      "step": 508
+    },
+    {
+      "epoch": 0.8226262626262626,
+      "grad_norm": 6.84375,
+      "learning_rate": 7.322653664326289e-06,
+      "loss": 0.7664,
+      "step": 509
+    },
+    {
+      "epoch": 0.8242424242424242,
+      "grad_norm": 8.25,
+      "learning_rate": 7.310135152001381e-06,
+      "loss": 0.9012,
+      "step": 510
+    },
+    {
+      "epoch": 0.8258585858585858,
+      "grad_norm": 6.96875,
+      "learning_rate": 7.29759820110502e-06,
+      "loss": 0.6616,
+      "step": 511
+    },
+    {
+      "epoch": 0.8274747474747475,
+      "grad_norm": 7.5625,
+      "learning_rate": 7.285042911702116e-06,
+      "loss": 0.6404,
+      "step": 512
+    },
+    {
+      "epoch": 0.8290909090909091,
+      "grad_norm": 5.90625,
+      "learning_rate": 7.27246938400395e-06,
+      "loss": 0.437,
+      "step": 513
+    },
+    {
+      "epoch": 0.8307070707070707,
+      "grad_norm": 8.75,
+      "learning_rate": 7.259877718367372e-06,
+      "loss": 0.7375,
+      "step": 514
+    },
+    {
+      "epoch": 0.8323232323232324,
+      "grad_norm": 9.0,
+      "learning_rate": 7.2472680152940015e-06,
+      "loss": 0.5881,
+      "step": 515
+    },
+    {
+      "epoch": 0.833939393939394,
+      "grad_norm": 5.90625,
+      "learning_rate": 7.234640375429427e-06,
+      "loss": 0.628,
+      "step": 516
+    },
+    {
+      "epoch": 0.8355555555555556,
+      "grad_norm": 9.6875,
+      "learning_rate": 7.2219948995624035e-06,
+      "loss": 0.7878,
+      "step": 517
+    },
+    {
+      "epoch": 0.8371717171717171,
+      "grad_norm": 7.40625,
+      "learning_rate": 7.209331688624039e-06,
+      "loss": 0.5814,
+      "step": 518
+    },
+    {
+      "epoch": 0.8387878787878787,
+      "grad_norm": 7.71875,
+      "learning_rate": 7.1966508436870044e-06,
+      "loss": 1.1205,
+      "step": 519
+    },
+    {
+      "epoch": 0.8404040404040404,
+      "grad_norm": 7.34375,
+      "learning_rate": 7.183952465964711e-06,
+      "loss": 0.8673,
+      "step": 520
+    },
+    {
+      "epoch": 0.842020202020202,
+      "grad_norm": 6.375,
+      "learning_rate": 7.171236656810513e-06,
+      "loss": 0.4973,
+      "step": 521
+    },
+    {
+      "epoch": 0.8436363636363636,
+      "grad_norm": 7.59375,
+      "learning_rate": 7.158503517716894e-06,
+      "loss": 0.6818,
+      "step": 522
+    },
+    {
+      "epoch": 0.8452525252525253,
+      "grad_norm": 8.375,
+      "learning_rate": 7.145753150314661e-06,
+      "loss": 0.763,
+      "step": 523
+    },
+    {
+      "epoch": 0.8468686868686869,
+      "grad_norm": 8.4375,
+      "learning_rate": 7.132985656372126e-06,
+      "loss": 0.7879,
+      "step": 524
+    },
+    {
+      "epoch": 0.8484848484848485,
+      "grad_norm": 6.25,
+      "learning_rate": 7.120201137794301e-06,
+      "loss": 0.3823,
+      "step": 525
+    },
+    {
+      "epoch": 0.8501010101010101,
+      "grad_norm": 5.84375,
+      "learning_rate": 7.1073996966220835e-06,
+      "loss": 0.4849,
+      "step": 526
+    },
+    {
+      "epoch": 0.8517171717171718,
+      "grad_norm": 6.625,
+      "learning_rate": 7.0945814350314346e-06,
+      "loss": 0.5307,
+      "step": 527
+    },
+    {
+      "epoch": 0.8533333333333334,
+      "grad_norm": 10.1875,
+      "learning_rate": 7.0817464553325764e-06,
+      "loss": 0.8415,
+      "step": 528
+    },
+    {
+      "epoch": 0.8549494949494949,
+      "grad_norm": 7.5,
+      "learning_rate": 7.068894859969162e-06,
+      "loss": 0.9093,
+      "step": 529
+    },
+    {
+      "epoch": 0.8565656565656565,
+      "grad_norm": 8.0625,
+      "learning_rate": 7.0560267515174685e-06,
+      "loss": 0.6398,
+      "step": 530
+    },
+    {
+      "epoch": 0.8581818181818182,
+      "grad_norm": 6.90625,
+      "learning_rate": 7.043142232685572e-06,
+      "loss": 0.7314,
+      "step": 531
+    },
+    {
+      "epoch": 0.8597979797979798,
+      "grad_norm": 7.0,
+      "learning_rate": 7.030241406312528e-06,
+      "loss": 0.6262,
+      "step": 532
+    },
+    {
+      "epoch": 0.8614141414141414,
+      "grad_norm": 6.3125,
+      "learning_rate": 7.0173243753675544e-06,
+      "loss": 0.7267,
+      "step": 533
+    },
+    {
+      "epoch": 0.863030303030303,
+      "grad_norm": 6.28125,
+      "learning_rate": 7.004391242949209e-06,
+      "loss": 0.5482,
+      "step": 534
+    },
+    {
+      "epoch": 0.8646464646464647,
+      "grad_norm": 7.46875,
+      "learning_rate": 6.991442112284561e-06,
+      "loss": 0.8512,
+      "step": 535
+    },
+    {
+      "epoch": 0.8662626262626263,
+      "grad_norm": 6.3125,
+      "learning_rate": 6.978477086728375e-06,
+      "loss": 0.4953,
+      "step": 536
+    },
+    {
+      "epoch": 0.8678787878787879,
+      "grad_norm": 7.71875,
+      "learning_rate": 6.96549626976228e-06,
+      "loss": 0.4961,
+      "step": 537
+    },
+    {
+      "epoch": 0.8694949494949495,
+      "grad_norm": 8.6875,
+      "learning_rate": 6.952499764993945e-06,
+      "loss": 0.9078,
+      "step": 538
+    },
+    {
+      "epoch": 0.8711111111111111,
+      "grad_norm": 7.59375,
+      "learning_rate": 6.9394876761562555e-06,
+      "loss": 0.631,
+      "step": 539
+    },
+    {
+      "epoch": 0.8727272727272727,
+      "grad_norm": 7.5625,
+      "learning_rate": 6.926460107106483e-06,
+      "loss": 0.6733,
+      "step": 540
+    },
+    {
+      "epoch": 0.8743434343434343,
+      "grad_norm": 6.6875,
+      "learning_rate": 6.913417161825449e-06,
+      "loss": 0.8478,
+      "step": 541
+    },
+    {
+      "epoch": 0.8759595959595959,
+      "grad_norm": 5.71875,
+      "learning_rate": 6.900358944416712e-06,
+      "loss": 0.503,
+      "step": 542
+    },
+    {
+      "epoch": 0.8775757575757576,
+      "grad_norm": 5.84375,
+      "learning_rate": 6.887285559105721e-06,
+      "loss": 0.8665,
+      "step": 543
+    },
+    {
+      "epoch": 0.8791919191919192,
+      "grad_norm": 6.65625,
+      "learning_rate": 6.874197110238986e-06,
+      "loss": 0.5614,
+      "step": 544
+    },
+    {
+      "epoch": 0.8808080808080808,
+      "grad_norm": 6.53125,
+      "learning_rate": 6.861093702283254e-06,
+      "loss": 0.7636,
+      "step": 545
+    },
+    {
+      "epoch": 0.8824242424242424,
+      "grad_norm": 11.6875,
+      "learning_rate": 6.847975439824669e-06,
+      "loss": 0.74,
+      "step": 546
+    },
+    {
+      "epoch": 0.8840404040404041,
+      "grad_norm": 7.46875,
+      "learning_rate": 6.834842427567934e-06,
+      "loss": 0.7544,
+      "step": 547
+    },
+    {
+      "epoch": 0.8856565656565657,
+      "grad_norm": 6.4375,
+      "learning_rate": 6.8216947703354815e-06,
+      "loss": 0.8037,
+      "step": 548
+    },
+    {
+      "epoch": 0.8872727272727273,
+      "grad_norm": 8.9375,
+      "learning_rate": 6.808532573066635e-06,
+      "loss": 0.7922,
+      "step": 549
+    },
+    {
+      "epoch": 0.8888888888888888,
+      "grad_norm": 6.78125,
+      "learning_rate": 6.795355940816768e-06,
+      "loss": 0.5742,
+      "step": 550
+    },
+    {
+      "epoch": 0.8905050505050505,
+      "grad_norm": 8.9375,
+      "learning_rate": 6.7821649787564706e-06,
+      "loss": 0.6695,
+      "step": 551
+    },
+    {
+      "epoch": 0.8921212121212121,
+      "grad_norm": 10.6875,
+      "learning_rate": 6.7689597921707065e-06,
+      "loss": 0.8931,
+      "step": 552
+    },
+    {
+      "epoch": 0.8937373737373737,
+      "grad_norm": 7.6875,
+      "learning_rate": 6.755740486457973e-06,
+      "loss": 0.8255,
+      "step": 553
+    },
+    {
+      "epoch": 0.8953535353535353,
+      "grad_norm": 7.65625,
+      "learning_rate": 6.742507167129465e-06,
+      "loss": 0.5539,
+      "step": 554
+    },
+    {
+      "epoch": 0.896969696969697,
+      "grad_norm": 7.9375,
+      "learning_rate": 6.729259939808223e-06,
+      "loss": 0.6744,
+      "step": 555
+    },
+    {
+      "epoch": 0.8985858585858586,
+      "grad_norm": 7.0625,
+      "learning_rate": 6.715998910228296e-06,
+      "loss": 0.5843,
+      "step": 556
+    },
+    {
+      "epoch": 0.9002020202020202,
+      "grad_norm": 6.4375,
+      "learning_rate": 6.702724184233904e-06,
+      "loss": 0.6715,
+      "step": 557
+    },
+    {
+      "epoch": 0.9018181818181819,
+      "grad_norm": 6.65625,
+      "learning_rate": 6.689435867778578e-06,
+      "loss": 0.5671,
+      "step": 558
+    },
+    {
+      "epoch": 0.9034343434343435,
+      "grad_norm": 11.875,
+      "learning_rate": 6.676134066924325e-06,
+      "loss": 0.8063,
+      "step": 559
+    },
+    {
+      "epoch": 0.9050505050505051,
+      "grad_norm": 7.03125,
+      "learning_rate": 6.6628188878407806e-06,
+      "loss": 0.4545,
+      "step": 560
+    },
+    {
+      "epoch": 0.9066666666666666,
+      "grad_norm": 7.09375,
+      "learning_rate": 6.64949043680436e-06,
+      "loss": 0.7163,
+      "step": 561
+    },
+    {
+      "epoch": 0.9082828282828282,
+      "grad_norm": 8.4375,
+      "learning_rate": 6.636148820197409e-06,
+      "loss": 0.8048,
+      "step": 562
+    },
+    {
+      "epoch": 0.9098989898989899,
+      "grad_norm": 7.65625,
+      "learning_rate": 6.622794144507356e-06,
+      "loss": 0.7068,
+      "step": 563
+    },
+    {
+      "epoch": 0.9115151515151515,
+      "grad_norm": 8.75,
+      "learning_rate": 6.609426516325859e-06,
+      "loss": 0.5749,
+      "step": 564
+    },
+    {
+      "epoch": 0.9131313131313131,
+      "grad_norm": 7.375,
+      "learning_rate": 6.596046042347964e-06,
+      "loss": 0.5476,
+      "step": 565
+    },
+    {
+      "epoch": 0.9147474747474748,
+      "grad_norm": 8.9375,
+      "learning_rate": 6.58265282937124e-06,
+      "loss": 0.9369,
+      "step": 566
+    },
+    {
+      "epoch": 0.9163636363636364,
+      "grad_norm": 9.75,
+      "learning_rate": 6.569246984294938e-06,
+      "loss": 0.9137,
+      "step": 567
+    },
+    {
+      "epoch": 0.917979797979798,
+      "grad_norm": 7.75,
+      "learning_rate": 6.555828614119132e-06,
+      "loss": 0.6025,
+      "step": 568
+    },
+    {
+      "epoch": 0.9195959595959596,
+      "grad_norm": 7.28125,
+      "learning_rate": 6.542397825943867e-06,
+      "loss": 0.6785,
+      "step": 569
+    },
+    {
+      "epoch": 0.9212121212121213,
+      "grad_norm": 8.5,
+      "learning_rate": 6.528954726968302e-06,
+      "loss": 0.7209,
+      "step": 570
+    },
+    {
+      "epoch": 0.9228282828282828,
+      "grad_norm": 6.8125,
+      "learning_rate": 6.515499424489857e-06,
+      "loss": 0.5772,
+      "step": 571
+    },
+    {
+      "epoch": 0.9244444444444444,
+      "grad_norm": 7.9375,
+      "learning_rate": 6.502032025903356e-06,
+      "loss": 0.7312,
+      "step": 572
+    },
+    {
+      "epoch": 0.926060606060606,
+      "grad_norm": 7.75,
+      "learning_rate": 6.4885526387001654e-06,
+      "loss": 0.8718,
+      "step": 573
+    },
+    {
+      "epoch": 0.9276767676767677,
+      "grad_norm": 6.25,
+      "learning_rate": 6.475061370467346e-06,
+      "loss": 0.7196,
+      "step": 574
+    },
+    {
+      "epoch": 0.9292929292929293,
+      "grad_norm": 7.84375,
+      "learning_rate": 6.461558328886786e-06,
+      "loss": 0.6602,
+      "step": 575
+    },
+    {
+      "epoch": 0.9309090909090909,
+      "grad_norm": 6.9375,
+      "learning_rate": 6.4480436217343366e-06,
+      "loss": 0.6234,
+      "step": 576
+    },
+    {
+      "epoch": 0.9325252525252525,
+      "grad_norm": 7.25,
+      "learning_rate": 6.434517356878974e-06,
+      "loss": 0.5093,
+      "step": 577
+    },
+    {
+      "epoch": 0.9341414141414142,
+      "grad_norm": 8.875,
+      "learning_rate": 6.420979642281909e-06,
+      "loss": 0.7573,
+      "step": 578
+    },
+    {
+      "epoch": 0.9357575757575758,
+      "grad_norm": 7.625,
+      "learning_rate": 6.4074305859957475e-06,
+      "loss": 1.0258,
+      "step": 579
+    },
+    {
+      "epoch": 0.9373737373737374,
+      "grad_norm": 5.9375,
+      "learning_rate": 6.393870296163616e-06,
+      "loss": 0.7432,
+      "step": 580
+    },
+    {
+      "epoch": 0.938989898989899,
+      "grad_norm": 8.8125,
+      "learning_rate": 6.380298881018307e-06,
+      "loss": 1.0014,
+      "step": 581
+    },
+    {
+      "epoch": 0.9406060606060606,
+      "grad_norm": 7.4375,
+      "learning_rate": 6.366716448881407e-06,
+      "loss": 0.7915,
+      "step": 582
+    },
+    {
+      "epoch": 0.9422222222222222,
+      "grad_norm": 5.90625,
+      "learning_rate": 6.353123108162435e-06,
+      "loss": 0.5849,
+      "step": 583
+    },
+    {
+      "epoch": 0.9438383838383838,
+      "grad_norm": 5.65625,
+      "learning_rate": 6.339518967357985e-06,
+      "loss": 0.785,
+      "step": 584
+    },
+    {
+      "epoch": 0.9454545454545454,
+      "grad_norm": 7.125,
+      "learning_rate": 6.325904135050844e-06,
+      "loss": 0.5567,
+      "step": 585
+    },
+    {
+      "epoch": 0.9470707070707071,
+      "grad_norm": 7.125,
+      "learning_rate": 6.312278719909138e-06,
+      "loss": 0.8426,
+      "step": 586
+    },
+    {
+      "epoch": 0.9486868686868687,
+      "grad_norm": 6.625,
+      "learning_rate": 6.298642830685464e-06,
+      "loss": 0.6369,
+      "step": 587
+    },
+    {
+      "epoch": 0.9503030303030303,
+      "grad_norm": 7.34375,
+      "learning_rate": 6.284996576216014e-06,
+      "loss": 0.6441,
+      "step": 588
+    },
+    {
+      "epoch": 0.9519191919191919,
+      "grad_norm": 7.0,
+      "learning_rate": 6.271340065419715e-06,
+      "loss": 0.7032,
+      "step": 589
+    },
+    {
+      "epoch": 0.9535353535353536,
+      "grad_norm": 6.71875,
+      "learning_rate": 6.257673407297352e-06,
+      "loss": 0.5248,
+      "step": 590
+    },
+    {
+      "epoch": 0.9551515151515152,
+      "grad_norm": 7.40625,
+      "learning_rate": 6.243996710930705e-06,
+      "loss": 0.5664,
+      "step": 591
+    },
+    {
+      "epoch": 0.9567676767676768,
+      "grad_norm": 7.3125,
+      "learning_rate": 6.230310085481677e-06,
+      "loss": 0.4842,
+      "step": 592
+    },
+    {
+      "epoch": 0.9583838383838383,
+      "grad_norm": 6.28125,
+      "learning_rate": 6.216613640191414e-06,
+      "loss": 0.5453,
+      "step": 593
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 8.0,
+      "learning_rate": 6.2029074843794445e-06,
+      "loss": 0.6635,
+      "step": 594
+    },
+    {
+      "epoch": 0.9616161616161616,
+      "grad_norm": 6.71875,
+      "learning_rate": 6.189191727442807e-06,
+      "loss": 0.6053,
+      "step": 595
+    },
+    {
+      "epoch": 0.9632323232323232,
+      "grad_norm": 6.15625,
+      "learning_rate": 6.175466478855161e-06,
+      "loss": 0.5201,
+      "step": 596
+    },
+    {
+      "epoch": 0.9648484848484848,
+      "grad_norm": 6.53125,
+      "learning_rate": 6.1617318481659364e-06,
+      "loss": 0.6634,
+      "step": 597
+    },
+    {
+      "epoch": 0.9664646464646465,
+      "grad_norm": 7.90625,
+      "learning_rate": 6.147987944999441e-06,
+      "loss": 0.7641,
+      "step": 598
+    },
+    {
+      "epoch": 0.9680808080808081,
+      "grad_norm": 7.3125,
+      "learning_rate": 6.13423487905399e-06,
+      "loss": 0.7179,
+      "step": 599
+    },
+    {
+      "epoch": 0.9696969696969697,
+      "grad_norm": 6.53125,
+      "learning_rate": 6.1204727601010396e-06,
+      "loss": 0.7291,
+      "step": 600
+    },
+    {
+      "epoch": 0.9713131313131314,
+      "grad_norm": 8.1875,
+      "learning_rate": 6.106701697984296e-06,
+      "loss": 0.7674,
+      "step": 601
+    },
+    {
+      "epoch": 0.972929292929293,
+      "grad_norm": 7.1875,
+      "learning_rate": 6.092921802618849e-06,
+      "loss": 0.7901,
+      "step": 602
+    },
+    {
+      "epoch": 0.9745454545454545,
+      "grad_norm": 5.75,
+      "learning_rate": 6.079133183990292e-06,
+      "loss": 0.6376,
+      "step": 603
+    },
+    {
+      "epoch": 0.9761616161616161,
+      "grad_norm": 7.125,
+      "learning_rate": 6.065335952153846e-06,
+      "loss": 0.7688,
+      "step": 604
+    },
+    {
+      "epoch": 0.9777777777777777,
+      "grad_norm": 8.8125,
+      "learning_rate": 6.051530217233473e-06,
+      "loss": 0.7441,
+      "step": 605
+    },
+    {
+      "epoch": 0.9793939393939394,
+      "grad_norm": 7.21875,
+      "learning_rate": 6.037716089421011e-06,
+      "loss": 0.5193,
+      "step": 606
+    },
+    {
+      "epoch": 0.981010101010101,
+      "grad_norm": 5.90625,
+      "learning_rate": 6.023893678975281e-06,
+      "loss": 0.6732,
+      "step": 607
+    },
+    {
+      "epoch": 0.9826262626262626,
+      "grad_norm": 6.625,
+      "learning_rate": 6.010063096221215e-06,
+      "loss": 1.2178,
+      "step": 608
+    },
+    {
+      "epoch": 0.9842424242424243,
+      "grad_norm": 6.0,
+      "learning_rate": 5.996224451548974e-06,
+      "loss": 0.432,
+      "step": 609
+    },
+    {
+      "epoch": 0.9858585858585859,
+      "grad_norm": 6.875,
+      "learning_rate": 5.982377855413063e-06,
+      "loss": 0.6985,
+      "step": 610
+    },
+    {
+      "epoch": 0.9874747474747475,
+      "grad_norm": 6.15625,
+      "learning_rate": 5.968523418331453e-06,
+      "loss": 0.4962,
+      "step": 611
+    },
+    {
+      "epoch": 0.9890909090909091,
+      "grad_norm": 6.0,
+      "learning_rate": 5.954661250884704e-06,
+      "loss": 0.5834,
+      "step": 612
+    },
+    {
+      "epoch": 0.9907070707070708,
+      "grad_norm": 8.3125,
+      "learning_rate": 5.940791463715068e-06,
+      "loss": 0.7957,
+      "step": 613
+    },
+    {
+      "epoch": 0.9923232323232323,
+      "grad_norm": 8.25,
+      "learning_rate": 5.926914167525618e-06,
+      "loss": 0.6498,
+      "step": 614
+    },
+    {
+      "epoch": 0.9939393939393939,
+      "grad_norm": 6.96875,
+      "learning_rate": 5.913029473079365e-06,
+      "loss": 0.7897,
+      "step": 615
+    },
+    {
+      "epoch": 0.9939393939393939,
+      "eval_loss": 0.6975554823875427,
+      "eval_runtime": 35.4933,
+      "eval_samples_per_second": 30.992,
+      "eval_steps_per_second": 3.888,
+      "step": 615
+    },
+    {
+      "epoch": 0.9955555555555555,
+      "grad_norm": 6.5,
+      "learning_rate": 5.899137491198364e-06,
+      "loss": 0.9353,
+      "step": 616
+    },
+    {
+      "epoch": 0.9971717171717172,
+      "grad_norm": 5.40625,
+      "learning_rate": 5.885238332762837e-06,
+      "loss": 0.6782,
+      "step": 617
+    },
+    {
+      "epoch": 0.9987878787878788,
+      "grad_norm": 6.75,
+      "learning_rate": 5.871332108710292e-06,
+      "loss": 0.6207,
+      "step": 618
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 1236,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.464714349398098e+17,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}