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": 37.5,
+      "learning_rate": 1.6129032258064518e-07,
+      "loss": 1.9582,
+      "step": 1
+    },
+    {
+      "epoch": 0.0032323232323232323,
+      "grad_norm": 38.25,
+      "learning_rate": 3.2258064516129035e-07,
+      "loss": 2.1833,
+      "step": 2
+    },
+    {
+      "epoch": 0.0048484848484848485,
+      "grad_norm": 37.75,
+      "learning_rate": 4.838709677419355e-07,
+      "loss": 1.8698,
+      "step": 3
+    },
+    {
+      "epoch": 0.006464646464646465,
+      "grad_norm": 38.25,
+      "learning_rate": 6.451612903225807e-07,
+      "loss": 2.1808,
+      "step": 4
+    },
+    {
+      "epoch": 0.00808080808080808,
+      "grad_norm": 34.0,
+      "learning_rate": 8.064516129032258e-07,
+      "loss": 1.9292,
+      "step": 5
+    },
+    {
+      "epoch": 0.009696969696969697,
+      "grad_norm": 31.5,
+      "learning_rate": 9.67741935483871e-07,
+      "loss": 1.8996,
+      "step": 6
+    },
+    {
+      "epoch": 0.011313131313131313,
+      "grad_norm": 29.5,
+      "learning_rate": 1.1290322580645162e-06,
+      "loss": 1.8762,
+      "step": 7
+    },
+    {
+      "epoch": 0.01292929292929293,
+      "grad_norm": 33.25,
+      "learning_rate": 1.2903225806451614e-06,
+      "loss": 1.7922,
+      "step": 8
+    },
+    {
+      "epoch": 0.014545454545454545,
+      "grad_norm": 22.5,
+      "learning_rate": 1.4516129032258066e-06,
+      "loss": 1.5962,
+      "step": 9
+    },
+    {
+      "epoch": 0.01616161616161616,
+      "grad_norm": 20.5,
+      "learning_rate": 1.6129032258064516e-06,
+      "loss": 1.7643,
+      "step": 10
+    },
+    {
+      "epoch": 0.017777777777777778,
+      "grad_norm": 19.0,
+      "learning_rate": 1.774193548387097e-06,
+      "loss": 1.3477,
+      "step": 11
+    },
+    {
+      "epoch": 0.019393939393939394,
+      "grad_norm": 19.0,
+      "learning_rate": 1.935483870967742e-06,
+      "loss": 1.4468,
+      "step": 12
+    },
+    {
+      "epoch": 0.02101010101010101,
+      "grad_norm": 18.25,
+      "learning_rate": 2.096774193548387e-06,
+      "loss": 1.2715,
+      "step": 13
+    },
+    {
+      "epoch": 0.022626262626262626,
+      "grad_norm": 17.875,
+      "learning_rate": 2.2580645161290324e-06,
+      "loss": 1.5449,
+      "step": 14
+    },
+    {
+      "epoch": 0.024242424242424242,
+      "grad_norm": 16.375,
+      "learning_rate": 2.4193548387096776e-06,
+      "loss": 1.194,
+      "step": 15
+    },
+    {
+      "epoch": 0.02585858585858586,
+      "grad_norm": 18.0,
+      "learning_rate": 2.580645161290323e-06,
+      "loss": 1.3124,
+      "step": 16
+    },
+    {
+      "epoch": 0.027474747474747475,
+      "grad_norm": 12.0625,
+      "learning_rate": 2.7419354838709676e-06,
+      "loss": 1.1663,
+      "step": 17
+    },
+    {
+      "epoch": 0.02909090909090909,
+      "grad_norm": 11.0625,
+      "learning_rate": 2.903225806451613e-06,
+      "loss": 1.1157,
+      "step": 18
+    },
+    {
+      "epoch": 0.030707070707070707,
+      "grad_norm": 10.1875,
+      "learning_rate": 3.0645161290322584e-06,
+      "loss": 1.3948,
+      "step": 19
+    },
+    {
+      "epoch": 0.03232323232323232,
+      "grad_norm": 11.4375,
+      "learning_rate": 3.225806451612903e-06,
+      "loss": 1.1091,
+      "step": 20
+    },
+    {
+      "epoch": 0.03393939393939394,
+      "grad_norm": 10.3125,
+      "learning_rate": 3.3870967741935484e-06,
+      "loss": 1.0059,
+      "step": 21
+    },
+    {
+      "epoch": 0.035555555555555556,
+      "grad_norm": 19.0,
+      "learning_rate": 3.548387096774194e-06,
+      "loss": 0.8881,
+      "step": 22
+    },
+    {
+      "epoch": 0.037171717171717175,
+      "grad_norm": 10.25,
+      "learning_rate": 3.7096774193548392e-06,
+      "loss": 1.0606,
+      "step": 23
+    },
+    {
+      "epoch": 0.03878787878787879,
+      "grad_norm": 9.875,
+      "learning_rate": 3.870967741935484e-06,
+      "loss": 0.7371,
+      "step": 24
+    },
+    {
+      "epoch": 0.04040404040404041,
+      "grad_norm": 9.75,
+      "learning_rate": 4.032258064516129e-06,
+      "loss": 0.7106,
+      "step": 25
+    },
+    {
+      "epoch": 0.04202020202020202,
+      "grad_norm": 11.4375,
+      "learning_rate": 4.193548387096774e-06,
+      "loss": 1.1739,
+      "step": 26
+    },
+    {
+      "epoch": 0.04363636363636364,
+      "grad_norm": 10.875,
+      "learning_rate": 4.35483870967742e-06,
+      "loss": 0.9232,
+      "step": 27
+    },
+    {
+      "epoch": 0.04525252525252525,
+      "grad_norm": 10.3125,
+      "learning_rate": 4.516129032258065e-06,
+      "loss": 0.8703,
+      "step": 28
+    },
+    {
+      "epoch": 0.04686868686868687,
+      "grad_norm": 9.9375,
+      "learning_rate": 4.67741935483871e-06,
+      "loss": 0.936,
+      "step": 29
+    },
+    {
+      "epoch": 0.048484848484848485,
+      "grad_norm": 8.75,
+      "learning_rate": 4.838709677419355e-06,
+      "loss": 0.9632,
+      "step": 30
+    },
+    {
+      "epoch": 0.050101010101010104,
+      "grad_norm": 8.5625,
+      "learning_rate": 5e-06,
+      "loss": 0.9792,
+      "step": 31
+    },
+    {
+      "epoch": 0.05171717171717172,
+      "grad_norm": 8.25,
+      "learning_rate": 5.161290322580646e-06,
+      "loss": 0.717,
+      "step": 32
+    },
+    {
+      "epoch": 0.05333333333333334,
+      "grad_norm": 8.4375,
+      "learning_rate": 5.322580645161291e-06,
+      "loss": 0.8609,
+      "step": 33
+    },
+    {
+      "epoch": 0.05494949494949495,
+      "grad_norm": 8.5,
+      "learning_rate": 5.483870967741935e-06,
+      "loss": 0.7129,
+      "step": 34
+    },
+    {
+      "epoch": 0.05656565656565657,
+      "grad_norm": 8.1875,
+      "learning_rate": 5.645161290322582e-06,
+      "loss": 0.9142,
+      "step": 35
+    },
+    {
+      "epoch": 0.05818181818181818,
+      "grad_norm": 7.8125,
+      "learning_rate": 5.806451612903226e-06,
+      "loss": 0.5824,
+      "step": 36
+    },
+    {
+      "epoch": 0.0597979797979798,
+      "grad_norm": 7.96875,
+      "learning_rate": 5.967741935483872e-06,
+      "loss": 0.6665,
+      "step": 37
+    },
+    {
+      "epoch": 0.061414141414141414,
+      "grad_norm": 8.875,
+      "learning_rate": 6.129032258064517e-06,
+      "loss": 0.8831,
+      "step": 38
+    },
+    {
+      "epoch": 0.06303030303030303,
+      "grad_norm": 9.6875,
+      "learning_rate": 6.290322580645162e-06,
+      "loss": 0.8367,
+      "step": 39
+    },
+    {
+      "epoch": 0.06464646464646465,
+      "grad_norm": 8.375,
+      "learning_rate": 6.451612903225806e-06,
+      "loss": 0.7128,
+      "step": 40
+    },
+    {
+      "epoch": 0.06626262626262626,
+      "grad_norm": 8.125,
+      "learning_rate": 6.612903225806452e-06,
+      "loss": 0.8643,
+      "step": 41
+    },
+    {
+      "epoch": 0.06787878787878789,
+      "grad_norm": 8.3125,
+      "learning_rate": 6.774193548387097e-06,
+      "loss": 0.7025,
+      "step": 42
+    },
+    {
+      "epoch": 0.0694949494949495,
+      "grad_norm": 7.96875,
+      "learning_rate": 6.935483870967743e-06,
+      "loss": 0.66,
+      "step": 43
+    },
+    {
+      "epoch": 0.07111111111111111,
+      "grad_norm": 8.625,
+      "learning_rate": 7.096774193548388e-06,
+      "loss": 0.8488,
+      "step": 44
+    },
+    {
+      "epoch": 0.07272727272727272,
+      "grad_norm": 7.84375,
+      "learning_rate": 7.258064516129033e-06,
+      "loss": 0.9772,
+      "step": 45
+    },
+    {
+      "epoch": 0.07434343434343435,
+      "grad_norm": 7.3125,
+      "learning_rate": 7.4193548387096784e-06,
+      "loss": 0.6694,
+      "step": 46
+    },
+    {
+      "epoch": 0.07595959595959596,
+      "grad_norm": 8.25,
+      "learning_rate": 7.580645161290323e-06,
+      "loss": 0.8924,
+      "step": 47
+    },
+    {
+      "epoch": 0.07757575757575758,
+      "grad_norm": 8.3125,
+      "learning_rate": 7.741935483870968e-06,
+      "loss": 0.982,
+      "step": 48
+    },
+    {
+      "epoch": 0.07919191919191919,
+      "grad_norm": 7.96875,
+      "learning_rate": 7.903225806451613e-06,
+      "loss": 0.8941,
+      "step": 49
+    },
+    {
+      "epoch": 0.08080808080808081,
+      "grad_norm": 7.8125,
+      "learning_rate": 8.064516129032258e-06,
+      "loss": 0.8727,
+      "step": 50
+    },
+    {
+      "epoch": 0.08242424242424243,
+      "grad_norm": 7.21875,
+      "learning_rate": 8.225806451612904e-06,
+      "loss": 0.7483,
+      "step": 51
+    },
+    {
+      "epoch": 0.08404040404040404,
+      "grad_norm": 7.875,
+      "learning_rate": 8.387096774193549e-06,
+      "loss": 0.733,
+      "step": 52
+    },
+    {
+      "epoch": 0.08565656565656565,
+      "grad_norm": 7.53125,
+      "learning_rate": 8.548387096774194e-06,
+      "loss": 0.7931,
+      "step": 53
+    },
+    {
+      "epoch": 0.08727272727272728,
+      "grad_norm": 7.28125,
+      "learning_rate": 8.70967741935484e-06,
+      "loss": 0.7543,
+      "step": 54
+    },
+    {
+      "epoch": 0.08888888888888889,
+      "grad_norm": 7.90625,
+      "learning_rate": 8.870967741935484e-06,
+      "loss": 0.8684,
+      "step": 55
+    },
+    {
+      "epoch": 0.0905050505050505,
+      "grad_norm": 7.5,
+      "learning_rate": 9.03225806451613e-06,
+      "loss": 0.5935,
+      "step": 56
+    },
+    {
+      "epoch": 0.09212121212121212,
+      "grad_norm": 8.375,
+      "learning_rate": 9.193548387096775e-06,
+      "loss": 1.094,
+      "step": 57
+    },
+    {
+      "epoch": 0.09373737373737374,
+      "grad_norm": 7.40625,
+      "learning_rate": 9.35483870967742e-06,
+      "loss": 0.9814,
+      "step": 58
+    },
+    {
+      "epoch": 0.09535353535353536,
+      "grad_norm": 6.75,
+      "learning_rate": 9.516129032258065e-06,
+      "loss": 0.5332,
+      "step": 59
+    },
+    {
+      "epoch": 0.09696969696969697,
+      "grad_norm": 7.21875,
+      "learning_rate": 9.67741935483871e-06,
+      "loss": 0.7308,
+      "step": 60
+    },
+    {
+      "epoch": 0.09858585858585858,
+      "grad_norm": 8.375,
+      "learning_rate": 9.838709677419356e-06,
+      "loss": 0.9279,
+      "step": 61
+    },
+    {
+      "epoch": 0.10020202020202021,
+      "grad_norm": 7.15625,
+      "learning_rate": 1e-05,
+      "loss": 0.7172,
+      "step": 62
+    },
+    {
+      "epoch": 0.10181818181818182,
+      "grad_norm": 7.34375,
+      "learning_rate": 9.999920184173449e-06,
+      "loss": 0.8192,
+      "step": 63
+    },
+    {
+      "epoch": 0.10343434343434343,
+      "grad_norm": 8.25,
+      "learning_rate": 9.999680739242022e-06,
+      "loss": 0.882,
+      "step": 64
+    },
+    {
+      "epoch": 0.10505050505050505,
+      "grad_norm": 9.125,
+      "learning_rate": 9.999281672850317e-06,
+      "loss": 0.8104,
+      "step": 65
+    },
+    {
+      "epoch": 0.10666666666666667,
+      "grad_norm": 7.90625,
+      "learning_rate": 9.99872299773906e-06,
+      "loss": 0.5911,
+      "step": 66
+    },
+    {
+      "epoch": 0.10828282828282829,
+      "grad_norm": 6.3125,
+      "learning_rate": 9.998004731744696e-06,
+      "loss": 0.6357,
+      "step": 67
+    },
+    {
+      "epoch": 0.1098989898989899,
+      "grad_norm": 8.9375,
+      "learning_rate": 9.997126897798826e-06,
+      "loss": 0.693,
+      "step": 68
+    },
+    {
+      "epoch": 0.11151515151515151,
+      "grad_norm": 6.65625,
+      "learning_rate": 9.996089523927461e-06,
+      "loss": 0.5648,
+      "step": 69
+    },
+    {
+      "epoch": 0.11313131313131314,
+      "grad_norm": 10.5,
+      "learning_rate": 9.994892643250147e-06,
+      "loss": 1.0413,
+      "step": 70
+    },
+    {
+      "epoch": 0.11474747474747475,
+      "grad_norm": 6.28125,
+      "learning_rate": 9.993536293978892e-06,
+      "loss": 0.6363,
+      "step": 71
+    },
+    {
+      "epoch": 0.11636363636363636,
+      "grad_norm": 6.84375,
+      "learning_rate": 9.99202051941695e-06,
+      "loss": 0.7609,
+      "step": 72
+    },
+    {
+      "epoch": 0.11797979797979798,
+      "grad_norm": 7.21875,
+      "learning_rate": 9.99034536795744e-06,
+      "loss": 0.6991,
+      "step": 73
+    },
+    {
+      "epoch": 0.1195959595959596,
+      "grad_norm": 6.53125,
+      "learning_rate": 9.9885108930818e-06,
+      "loss": 0.6906,
+      "step": 74
+    },
+    {
+      "epoch": 0.12121212121212122,
+      "grad_norm": 6.4375,
+      "learning_rate": 9.986517153358086e-06,
+      "loss": 0.5406,
+      "step": 75
+    },
+    {
+      "epoch": 0.12282828282828283,
+      "grad_norm": 6.375,
+      "learning_rate": 9.984364212439089e-06,
+      "loss": 0.7722,
+      "step": 76
+    },
+    {
+      "epoch": 0.12444444444444444,
+      "grad_norm": 7.0,
+      "learning_rate": 9.982052139060312e-06,
+      "loss": 0.6162,
+      "step": 77
+    },
+    {
+      "epoch": 0.12606060606060607,
+      "grad_norm": 6.84375,
+      "learning_rate": 9.979581007037776e-06,
+      "loss": 0.806,
+      "step": 78
+    },
+    {
+      "epoch": 0.12767676767676767,
+      "grad_norm": 7.25,
+      "learning_rate": 9.976950895265657e-06,
+      "loss": 0.8358,
+      "step": 79
+    },
+    {
+      "epoch": 0.1292929292929293,
+      "grad_norm": 7.25,
+      "learning_rate": 9.974161887713775e-06,
+      "loss": 0.894,
+      "step": 80
+    },
+    {
+      "epoch": 0.13090909090909092,
+      "grad_norm": 7.5,
+      "learning_rate": 9.971214073424906e-06,
+      "loss": 1.2015,
+      "step": 81
+    },
+    {
+      "epoch": 0.13252525252525252,
+      "grad_norm": 6.84375,
+      "learning_rate": 9.968107546511942e-06,
+      "loss": 0.8696,
+      "step": 82
+    },
+    {
+      "epoch": 0.13414141414141414,
+      "grad_norm": 6.71875,
+      "learning_rate": 9.964842406154892e-06,
+      "loss": 0.6243,
+      "step": 83
+    },
+    {
+      "epoch": 0.13575757575757577,
+      "grad_norm": 5.65625,
+      "learning_rate": 9.961418756597703e-06,
+      "loss": 0.55,
+      "step": 84
+    },
+    {
+      "epoch": 0.13737373737373737,
+      "grad_norm": 5.84375,
+      "learning_rate": 9.957836707144943e-06,
+      "loss": 0.6568,
+      "step": 85
+    },
+    {
+      "epoch": 0.138989898989899,
+      "grad_norm": 6.03125,
+      "learning_rate": 9.95409637215831e-06,
+      "loss": 0.785,
+      "step": 86
+    },
+    {
+      "epoch": 0.1406060606060606,
+      "grad_norm": 5.5625,
+      "learning_rate": 9.950197871052974e-06,
+      "loss": 0.4307,
+      "step": 87
+    },
+    {
+      "epoch": 0.14222222222222222,
+      "grad_norm": 6.03125,
+      "learning_rate": 9.94614132829377e-06,
+      "loss": 0.7334,
+      "step": 88
+    },
+    {
+      "epoch": 0.14383838383838385,
+      "grad_norm": 6.40625,
+      "learning_rate": 9.941926873391223e-06,
+      "loss": 0.4968,
+      "step": 89
+    },
+    {
+      "epoch": 0.14545454545454545,
+      "grad_norm": 7.875,
+      "learning_rate": 9.937554640897414e-06,
+      "loss": 0.7624,
+      "step": 90
+    },
+    {
+      "epoch": 0.14707070707070707,
+      "grad_norm": 7.15625,
+      "learning_rate": 9.933024770401682e-06,
+      "loss": 0.7453,
+      "step": 91
+    },
+    {
+      "epoch": 0.1486868686868687,
+      "grad_norm": 15.375,
+      "learning_rate": 9.928337406526172e-06,
+      "loss": 0.7507,
+      "step": 92
+    },
+    {
+      "epoch": 0.1503030303030303,
+      "grad_norm": 6.78125,
+      "learning_rate": 9.923492698921214e-06,
+      "loss": 0.5031,
+      "step": 93
+    },
+    {
+      "epoch": 0.15191919191919193,
+      "grad_norm": 7.0,
+      "learning_rate": 9.918490802260538e-06,
+      "loss": 0.9447,
+      "step": 94
+    },
+    {
+      "epoch": 0.15353535353535352,
+      "grad_norm": 7.5625,
+      "learning_rate": 9.913331876236358e-06,
+      "loss": 0.9043,
+      "step": 95
+    },
+    {
+      "epoch": 0.15515151515151515,
+      "grad_norm": 6.65625,
+      "learning_rate": 9.90801608555425e-06,
+      "loss": 0.7444,
+      "step": 96
+    },
+    {
+      "epoch": 0.15676767676767678,
+      "grad_norm": 5.96875,
+      "learning_rate": 9.902543599927903e-06,
+      "loss": 0.5291,
+      "step": 97
+    },
+    {
+      "epoch": 0.15838383838383838,
+      "grad_norm": 8.0,
+      "learning_rate": 9.896914594073703e-06,
+      "loss": 0.6361,
+      "step": 98
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.625,
+      "learning_rate": 9.891129247705153e-06,
+      "loss": 0.7495,
+      "step": 99
+    },
+    {
+      "epoch": 0.16161616161616163,
+      "grad_norm": 6.875,
+      "learning_rate": 9.885187745527132e-06,
+      "loss": 0.6287,
+      "step": 100
+    },
+    {
+      "epoch": 0.16323232323232323,
+      "grad_norm": 5.21875,
+      "learning_rate": 9.879090277230005e-06,
+      "loss": 0.7185,
+      "step": 101
+    },
+    {
+      "epoch": 0.16484848484848486,
+      "grad_norm": 7.25,
+      "learning_rate": 9.87283703748356e-06,
+      "loss": 0.8769,
+      "step": 102
+    },
+    {
+      "epoch": 0.16646464646464645,
+      "grad_norm": 6.375,
+      "learning_rate": 9.866428225930798e-06,
+      "loss": 0.7113,
+      "step": 103
+    },
+    {
+      "epoch": 0.16808080808080808,
+      "grad_norm": 6.90625,
+      "learning_rate": 9.859864047181551e-06,
+      "loss": 0.6993,
+      "step": 104
+    },
+    {
+      "epoch": 0.1696969696969697,
+      "grad_norm": 6.8125,
+      "learning_rate": 9.853144710805966e-06,
+      "loss": 0.8039,
+      "step": 105
+    },
+    {
+      "epoch": 0.1713131313131313,
+      "grad_norm": 5.5625,
+      "learning_rate": 9.846270431327793e-06,
+      "loss": 0.7498,
+      "step": 106
+    },
+    {
+      "epoch": 0.17292929292929293,
+      "grad_norm": 5.15625,
+      "learning_rate": 9.839241428217553e-06,
+      "loss": 0.6026,
+      "step": 107
+    },
+    {
+      "epoch": 0.17454545454545456,
+      "grad_norm": 7.09375,
+      "learning_rate": 9.832057925885526e-06,
+      "loss": 0.6578,
+      "step": 108
+    },
+    {
+      "epoch": 0.17616161616161616,
+      "grad_norm": 6.1875,
+      "learning_rate": 9.824720153674578e-06,
+      "loss": 0.6443,
+      "step": 109
+    },
+    {
+      "epoch": 0.17777777777777778,
+      "grad_norm": 7.25,
+      "learning_rate": 9.817228345852853e-06,
+      "loss": 0.7546,
+      "step": 110
+    },
+    {
+      "epoch": 0.17939393939393938,
+      "grad_norm": 5.4375,
+      "learning_rate": 9.809582741606283e-06,
+      "loss": 0.702,
+      "step": 111
+    },
+    {
+      "epoch": 0.181010101010101,
+      "grad_norm": 5.875,
+      "learning_rate": 9.801783585030959e-06,
+      "loss": 0.674,
+      "step": 112
+    },
+    {
+      "epoch": 0.18262626262626264,
+      "grad_norm": 7.65625,
+      "learning_rate": 9.79383112512533e-06,
+      "loss": 1.2071,
+      "step": 113
+    },
+    {
+      "epoch": 0.18424242424242424,
+      "grad_norm": 5.1875,
+      "learning_rate": 9.785725615782262e-06,
+      "loss": 0.4847,
+      "step": 114
+    },
+    {
+      "epoch": 0.18585858585858586,
+      "grad_norm": 6.1875,
+      "learning_rate": 9.777467315780926e-06,
+      "loss": 0.8333,
+      "step": 115
+    },
+    {
+      "epoch": 0.1874747474747475,
+      "grad_norm": 6.875,
+      "learning_rate": 9.769056488778538e-06,
+      "loss": 0.6039,
+      "step": 116
+    },
+    {
+      "epoch": 0.1890909090909091,
+      "grad_norm": 6.96875,
+      "learning_rate": 9.760493403301941e-06,
+      "loss": 0.9114,
+      "step": 117
+    },
+    {
+      "epoch": 0.1907070707070707,
+      "grad_norm": 4.9375,
+      "learning_rate": 9.751778332739033e-06,
+      "loss": 0.5002,
+      "step": 118
+    },
+    {
+      "epoch": 0.1923232323232323,
+      "grad_norm": 6.65625,
+      "learning_rate": 9.74291155533004e-06,
+      "loss": 0.8994,
+      "step": 119
+    },
+    {
+      "epoch": 0.19393939393939394,
+      "grad_norm": 5.9375,
+      "learning_rate": 9.733893354158628e-06,
+      "loss": 0.7578,
+      "step": 120
+    },
+    {
+      "epoch": 0.19555555555555557,
+      "grad_norm": 6.21875,
+      "learning_rate": 9.724724017142869e-06,
+      "loss": 0.6673,
+      "step": 121
+    },
+    {
+      "epoch": 0.19717171717171716,
+      "grad_norm": 6.1875,
+      "learning_rate": 9.715403837026046e-06,
+      "loss": 0.7346,
+      "step": 122
+    },
+    {
+      "epoch": 0.1987878787878788,
+      "grad_norm": 5.8125,
+      "learning_rate": 9.705933111367314e-06,
+      "loss": 0.7654,
+      "step": 123
+    },
+    {
+      "epoch": 0.1987878787878788,
+      "eval_loss": 0.6892983913421631,
+      "eval_runtime": 36.0789,
+      "eval_samples_per_second": 30.489,
+      "eval_steps_per_second": 3.825,
+      "step": 123
+    },
+    {
+      "epoch": 0.20040404040404042,
+      "grad_norm": 6.03125,
+      "learning_rate": 9.69631214253219e-06,
+      "loss": 0.8745,
+      "step": 124
+    },
+    {
+      "epoch": 0.20202020202020202,
+      "grad_norm": 7.0625,
+      "learning_rate": 9.68654123768291e-06,
+      "loss": 0.5199,
+      "step": 125
+    },
+    {
+      "epoch": 0.20363636363636364,
+      "grad_norm": 6.59375,
+      "learning_rate": 9.676620708768608e-06,
+      "loss": 0.7167,
+      "step": 126
+    },
+    {
+      "epoch": 0.20525252525252524,
+      "grad_norm": 6.40625,
+      "learning_rate": 9.666550872515367e-06,
+      "loss": 0.7242,
+      "step": 127
+    },
+    {
+      "epoch": 0.20686868686868687,
+      "grad_norm": 6.4375,
+      "learning_rate": 9.656332050416118e-06,
+      "loss": 0.8338,
+      "step": 128
+    },
+    {
+      "epoch": 0.2084848484848485,
+      "grad_norm": 6.3125,
+      "learning_rate": 9.645964568720345e-06,
+      "loss": 0.6658,
+      "step": 129
+    },
+    {
+      "epoch": 0.2101010101010101,
+      "grad_norm": 6.03125,
+      "learning_rate": 9.635448758423703e-06,
+      "loss": 0.6096,
+      "step": 130
+    },
+    {
+      "epoch": 0.21171717171717172,
+      "grad_norm": 5.25,
+      "learning_rate": 9.624784955257423e-06,
+      "loss": 0.7533,
+      "step": 131
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 7.0,
+      "learning_rate": 9.613973499677613e-06,
+      "loss": 0.8072,
+      "step": 132
+    },
+    {
+      "epoch": 0.21494949494949495,
+      "grad_norm": 5.875,
+      "learning_rate": 9.60301473685438e-06,
+      "loss": 0.6549,
+      "step": 133
+    },
+    {
+      "epoch": 0.21656565656565657,
+      "grad_norm": 6.0,
+      "learning_rate": 9.591909016660806e-06,
+      "loss": 0.637,
+      "step": 134
+    },
+    {
+      "epoch": 0.21818181818181817,
+      "grad_norm": 5.65625,
+      "learning_rate": 9.580656693661787e-06,
+      "loss": 0.6146,
+      "step": 135
+    },
+    {
+      "epoch": 0.2197979797979798,
+      "grad_norm": 5.75,
+      "learning_rate": 9.569258127102708e-06,
+      "loss": 0.6695,
+      "step": 136
+    },
+    {
+      "epoch": 0.22141414141414142,
+      "grad_norm": 5.59375,
+      "learning_rate": 9.55771368089797e-06,
+      "loss": 0.6858,
+      "step": 137
+    },
+    {
+      "epoch": 0.22303030303030302,
+      "grad_norm": 5.71875,
+      "learning_rate": 9.546023723619387e-06,
+      "loss": 0.5637,
+      "step": 138
+    },
+    {
+      "epoch": 0.22464646464646465,
+      "grad_norm": 5.875,
+      "learning_rate": 9.534188628484391e-06,
+      "loss": 0.9502,
+      "step": 139
+    },
+    {
+      "epoch": 0.22626262626262628,
+      "grad_norm": 6.53125,
+      "learning_rate": 9.522208773344147e-06,
+      "loss": 0.8898,
+      "step": 140
+    },
+    {
+      "epoch": 0.22787878787878788,
+      "grad_norm": 6.0625,
+      "learning_rate": 9.510084540671471e-06,
+      "loss": 0.8036,
+      "step": 141
+    },
+    {
+      "epoch": 0.2294949494949495,
+      "grad_norm": 6.28125,
+      "learning_rate": 9.497816317548625e-06,
+      "loss": 0.9315,
+      "step": 142
+    },
+    {
+      "epoch": 0.2311111111111111,
+      "grad_norm": 6.28125,
+      "learning_rate": 9.485404495654953e-06,
+      "loss": 0.55,
+      "step": 143
+    },
+    {
+      "epoch": 0.23272727272727273,
+      "grad_norm": 6.0,
+      "learning_rate": 9.472849471254386e-06,
+      "loss": 0.4922,
+      "step": 144
+    },
+    {
+      "epoch": 0.23434343434343435,
+      "grad_norm": 6.03125,
+      "learning_rate": 9.460151645182784e-06,
+      "loss": 0.9106,
+      "step": 145
+    },
+    {
+      "epoch": 0.23595959595959595,
+      "grad_norm": 5.59375,
+      "learning_rate": 9.447311422835141e-06,
+      "loss": 0.7088,
+      "step": 146
+    },
+    {
+      "epoch": 0.23757575757575758,
+      "grad_norm": 6.34375,
+      "learning_rate": 9.43432921415264e-06,
+      "loss": 0.7527,
+      "step": 147
+    },
+    {
+      "epoch": 0.2391919191919192,
+      "grad_norm": 7.53125,
+      "learning_rate": 9.421205433609568e-06,
+      "loss": 0.5364,
+      "step": 148
+    },
+    {
+      "epoch": 0.2408080808080808,
+      "grad_norm": 5.625,
+      "learning_rate": 9.407940500200082e-06,
+      "loss": 0.6517,
+      "step": 149
+    },
+    {
+      "epoch": 0.24242424242424243,
+      "grad_norm": 5.875,
+      "learning_rate": 9.39453483742483e-06,
+      "loss": 0.5657,
+      "step": 150
+    },
+    {
+      "epoch": 0.24404040404040403,
+      "grad_norm": 5.5,
+      "learning_rate": 9.380988873277436e-06,
+      "loss": 0.8081,
+      "step": 151
+    },
+    {
+      "epoch": 0.24565656565656566,
+      "grad_norm": 6.46875,
+      "learning_rate": 9.367303040230828e-06,
+      "loss": 0.7384,
+      "step": 152
+    },
+    {
+      "epoch": 0.24727272727272728,
+      "grad_norm": 7.5,
+      "learning_rate": 9.35347777522344e-06,
+      "loss": 0.8627,
+      "step": 153
+    },
+    {
+      "epoch": 0.24888888888888888,
+      "grad_norm": 5.03125,
+      "learning_rate": 9.33951351964525e-06,
+      "loss": 0.6216,
+      "step": 154
+    },
+    {
+      "epoch": 0.2505050505050505,
+      "grad_norm": 5.34375,
+      "learning_rate": 9.3254107193237e-06,
+      "loss": 0.6768,
+      "step": 155
+    },
+    {
+      "epoch": 0.25212121212121213,
+      "grad_norm": 5.53125,
+      "learning_rate": 9.311169824509454e-06,
+      "loss": 0.4977,
+      "step": 156
+    },
+    {
+      "epoch": 0.25373737373737376,
+      "grad_norm": 5.09375,
+      "learning_rate": 9.296791289862031e-06,
+      "loss": 0.5076,
+      "step": 157
+    },
+    {
+      "epoch": 0.25535353535353533,
+      "grad_norm": 6.96875,
+      "learning_rate": 9.28227557443528e-06,
+      "loss": 0.7456,
+      "step": 158
+    },
+    {
+      "epoch": 0.25696969696969696,
+      "grad_norm": 5.59375,
+      "learning_rate": 9.267623141662734e-06,
+      "loss": 0.7384,
+      "step": 159
+    },
+    {
+      "epoch": 0.2585858585858586,
+      "grad_norm": 6.28125,
+      "learning_rate": 9.252834459342801e-06,
+      "loss": 0.8759,
+      "step": 160
+    },
+    {
+      "epoch": 0.2602020202020202,
+      "grad_norm": 5.53125,
+      "learning_rate": 9.237909999623847e-06,
+      "loss": 0.6948,
+      "step": 161
+    },
+    {
+      "epoch": 0.26181818181818184,
+      "grad_norm": 5.875,
+      "learning_rate": 9.222850238989104e-06,
+      "loss": 0.5993,
+      "step": 162
+    },
+    {
+      "epoch": 0.2634343434343434,
+      "grad_norm": 6.0,
+      "learning_rate": 9.207655658241469e-06,
+      "loss": 0.5063,
+      "step": 163
+    },
+    {
+      "epoch": 0.26505050505050504,
+      "grad_norm": 6.03125,
+      "learning_rate": 9.192326742488153e-06,
+      "loss": 0.6991,
+      "step": 164
+    },
+    {
+      "epoch": 0.26666666666666666,
+      "grad_norm": 5.53125,
+      "learning_rate": 9.176863981125185e-06,
+      "loss": 0.5182,
+      "step": 165
+    },
+    {
+      "epoch": 0.2682828282828283,
+      "grad_norm": 5.28125,
+      "learning_rate": 9.161267867821802e-06,
+      "loss": 0.8789,
+      "step": 166
+    },
+    {
+      "epoch": 0.2698989898989899,
+      "grad_norm": 6.03125,
+      "learning_rate": 9.14553890050467e-06,
+      "loss": 0.7351,
+      "step": 167
+    },
+    {
+      "epoch": 0.27151515151515154,
+      "grad_norm": 5.4375,
+      "learning_rate": 9.129677581342e-06,
+      "loss": 0.776,
+      "step": 168
+    },
+    {
+      "epoch": 0.2731313131313131,
+      "grad_norm": 6.90625,
+      "learning_rate": 9.113684416727511e-06,
+      "loss": 0.7084,
+      "step": 169
+    },
+    {
+      "epoch": 0.27474747474747474,
+      "grad_norm": 6.375,
+      "learning_rate": 9.097559917264268e-06,
+      "loss": 0.7935,
+      "step": 170
+    },
+    {
+      "epoch": 0.27636363636363637,
+      "grad_norm": 5.3125,
+      "learning_rate": 9.081304597748366e-06,
+      "loss": 0.6118,
+      "step": 171
+    },
+    {
+      "epoch": 0.277979797979798,
+      "grad_norm": 5.71875,
+      "learning_rate": 9.064918977152517e-06,
+      "loss": 0.7662,
+      "step": 172
+    },
+    {
+      "epoch": 0.2795959595959596,
+      "grad_norm": 5.09375,
+      "learning_rate": 9.048403578609454e-06,
+      "loss": 0.7373,
+      "step": 173
+    },
+    {
+      "epoch": 0.2812121212121212,
+      "grad_norm": 6.0,
+      "learning_rate": 9.031758929395259e-06,
+      "loss": 0.8678,
+      "step": 174
+    },
+    {
+      "epoch": 0.2828282828282828,
+      "grad_norm": 4.6875,
+      "learning_rate": 9.014985560912499e-06,
+      "loss": 0.424,
+      "step": 175
+    },
+    {
+      "epoch": 0.28444444444444444,
+      "grad_norm": 6.84375,
+      "learning_rate": 8.998084008673284e-06,
+      "loss": 0.6046,
+      "step": 176
+    },
+    {
+      "epoch": 0.28606060606060607,
+      "grad_norm": 5.0625,
+      "learning_rate": 8.981054812282162e-06,
+      "loss": 0.6064,
+      "step": 177
+    },
+    {
+      "epoch": 0.2876767676767677,
+      "grad_norm": 6.96875,
+      "learning_rate": 8.963898515418885e-06,
+      "loss": 0.5947,
+      "step": 178
+    },
+    {
+      "epoch": 0.28929292929292927,
+      "grad_norm": 5.9375,
+      "learning_rate": 8.946615665821059e-06,
+      "loss": 0.6578,
+      "step": 179
+    },
+    {
+      "epoch": 0.2909090909090909,
+      "grad_norm": 6.1875,
+      "learning_rate": 8.929206815266653e-06,
+      "loss": 0.5758,
+      "step": 180
+    },
+    {
+      "epoch": 0.2925252525252525,
+      "grad_norm": 5.78125,
+      "learning_rate": 8.911672519556386e-06,
+      "loss": 0.6474,
+      "step": 181
+    },
+    {
+      "epoch": 0.29414141414141415,
+      "grad_norm": 6.1875,
+      "learning_rate": 8.89401333849598e-06,
+      "loss": 0.5758,
+      "step": 182
+    },
+    {
+      "epoch": 0.2957575757575758,
+      "grad_norm": 5.4375,
+      "learning_rate": 8.87622983587829e-06,
+      "loss": 0.9669,
+      "step": 183
+    },
+    {
+      "epoch": 0.2973737373737374,
+      "grad_norm": 5.03125,
+      "learning_rate": 8.8583225794653e-06,
+      "loss": 0.8864,
+      "step": 184
+    },
+    {
+      "epoch": 0.298989898989899,
+      "grad_norm": 4.84375,
+      "learning_rate": 8.840292140969995e-06,
+      "loss": 0.5196,
+      "step": 185
+    },
+    {
+      "epoch": 0.3006060606060606,
+      "grad_norm": 5.65625,
+      "learning_rate": 8.82213909603812e-06,
+      "loss": 0.932,
+      "step": 186
+    },
+    {
+      "epoch": 0.3022222222222222,
+      "grad_norm": 5.53125,
+      "learning_rate": 8.803864024229786e-06,
+      "loss": 0.7756,
+      "step": 187
+    },
+    {
+      "epoch": 0.30383838383838385,
+      "grad_norm": 5.5,
+      "learning_rate": 8.78546750900098e-06,
+      "loss": 0.5672,
+      "step": 188
+    },
+    {
+      "epoch": 0.3054545454545455,
+      "grad_norm": 5.15625,
+      "learning_rate": 8.766950137684929e-06,
+      "loss": 0.4919,
+      "step": 189
+    },
+    {
+      "epoch": 0.30707070707070705,
+      "grad_norm": 6.5625,
+      "learning_rate": 8.748312501473351e-06,
+      "loss": 0.7069,
+      "step": 190
+    },
+    {
+      "epoch": 0.3086868686868687,
+      "grad_norm": 5.8125,
+      "learning_rate": 8.72955519539758e-06,
+      "loss": 0.513,
+      "step": 191
+    },
+    {
+      "epoch": 0.3103030303030303,
+      "grad_norm": 5.1875,
+      "learning_rate": 8.710678818309576e-06,
+      "loss": 0.6311,
+      "step": 192
+    },
+    {
+      "epoch": 0.31191919191919193,
+      "grad_norm": 5.625,
+      "learning_rate": 8.691683972862792e-06,
+      "loss": 0.651,
+      "step": 193
+    },
+    {
+      "epoch": 0.31353535353535356,
+      "grad_norm": 6.0625,
+      "learning_rate": 8.672571265492944e-06,
+      "loss": 1.1743,
+      "step": 194
+    },
+    {
+      "epoch": 0.3151515151515151,
+      "grad_norm": 6.46875,
+      "learning_rate": 8.653341306398645e-06,
+      "loss": 0.7795,
+      "step": 195
+    },
+    {
+      "epoch": 0.31676767676767675,
+      "grad_norm": 4.90625,
+      "learning_rate": 8.63399470952193e-06,
+      "loss": 0.7834,
+      "step": 196
+    },
+    {
+      "epoch": 0.3183838383838384,
+      "grad_norm": 5.65625,
+      "learning_rate": 8.614532092528645e-06,
+      "loss": 0.6692,
+      "step": 197
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.53125,
+      "learning_rate": 8.594954076788736e-06,
+      "loss": 0.8049,
+      "step": 198
+    },
+    {
+      "epoch": 0.32161616161616163,
+      "grad_norm": 6.21875,
+      "learning_rate": 8.575261287356407e-06,
+      "loss": 0.6907,
+      "step": 199
+    },
+    {
+      "epoch": 0.32323232323232326,
+      "grad_norm": 5.1875,
+      "learning_rate": 8.555454352950161e-06,
+      "loss": 0.5804,
+      "step": 200
+    },
+    {
+      "epoch": 0.32484848484848483,
+      "grad_norm": 5.40625,
+      "learning_rate": 8.535533905932739e-06,
+      "loss": 0.7927,
+      "step": 201
+    },
+    {
+      "epoch": 0.32646464646464646,
+      "grad_norm": 5.59375,
+      "learning_rate": 8.515500582290914e-06,
+      "loss": 0.5515,
+      "step": 202
+    },
+    {
+      "epoch": 0.3280808080808081,
+      "grad_norm": 5.96875,
+      "learning_rate": 8.495355021615204e-06,
+      "loss": 0.6509,
+      "step": 203
+    },
+    {
+      "epoch": 0.3296969696969697,
+      "grad_norm": 4.90625,
+      "learning_rate": 8.475097867079437e-06,
+      "loss": 0.8164,
+      "step": 204
+    },
+    {
+      "epoch": 0.33131313131313134,
+      "grad_norm": 4.5,
+      "learning_rate": 8.454729765420228e-06,
+      "loss": 0.592,
+      "step": 205
+    },
+    {
+      "epoch": 0.3329292929292929,
+      "grad_norm": 5.125,
+      "learning_rate": 8.434251366916323e-06,
+      "loss": 0.5169,
+      "step": 206
+    },
+    {
+      "epoch": 0.33454545454545453,
+      "grad_norm": 5.4375,
+      "learning_rate": 8.413663325367845e-06,
+      "loss": 0.6679,
+      "step": 207
+    },
+    {
+      "epoch": 0.33616161616161616,
+      "grad_norm": 4.5625,
+      "learning_rate": 8.392966298075413e-06,
+      "loss": 0.5196,
+      "step": 208
+    },
+    {
+      "epoch": 0.3377777777777778,
+      "grad_norm": 5.1875,
+      "learning_rate": 8.372160945819164e-06,
+      "loss": 0.4692,
+      "step": 209
+    },
+    {
+      "epoch": 0.3393939393939394,
+      "grad_norm": 5.28125,
+      "learning_rate": 8.351247932837655e-06,
+      "loss": 0.5804,
+      "step": 210
+    },
+    {
+      "epoch": 0.341010101010101,
+      "grad_norm": 5.25,
+      "learning_rate": 8.330227926806652e-06,
+      "loss": 0.5084,
+      "step": 211
+    },
+    {
+      "epoch": 0.3426262626262626,
+      "grad_norm": 5.1875,
+      "learning_rate": 8.309101598817812e-06,
+      "loss": 0.7203,
+      "step": 212
+    },
+    {
+      "epoch": 0.34424242424242424,
+      "grad_norm": 6.0,
+      "learning_rate": 8.287869623357275e-06,
+      "loss": 0.7598,
+      "step": 213
+    },
+    {
+      "epoch": 0.34585858585858587,
+      "grad_norm": 5.5625,
+      "learning_rate": 8.266532678284103e-06,
+      "loss": 0.6887,
+      "step": 214
+    },
+    {
+      "epoch": 0.3474747474747475,
+      "grad_norm": 5.96875,
+      "learning_rate": 8.245091444808663e-06,
+      "loss": 0.6851,
+      "step": 215
+    },
+    {
+      "epoch": 0.3490909090909091,
+      "grad_norm": 5.0,
+      "learning_rate": 8.223546607470863e-06,
+      "loss": 0.6598,
+      "step": 216
+    },
+    {
+      "epoch": 0.3507070707070707,
+      "grad_norm": 5.03125,
+      "learning_rate": 8.201898854118301e-06,
+      "loss": 0.601,
+      "step": 217
+    },
+    {
+      "epoch": 0.3523232323232323,
+      "grad_norm": 5.8125,
+      "learning_rate": 8.18014887588431e-06,
+      "loss": 0.6152,
+      "step": 218
+    },
+    {
+      "epoch": 0.35393939393939394,
+      "grad_norm": 7.71875,
+      "learning_rate": 8.158297367165885e-06,
+      "loss": 0.7415,
+      "step": 219
+    },
+    {
+      "epoch": 0.35555555555555557,
+      "grad_norm": 5.40625,
+      "learning_rate": 8.13634502560152e-06,
+      "loss": 0.7995,
+      "step": 220
+    },
+    {
+      "epoch": 0.3571717171717172,
+      "grad_norm": 4.6875,
+      "learning_rate": 8.114292552048925e-06,
+      "loss": 0.5419,
+      "step": 221
+    },
+    {
+      "epoch": 0.35878787878787877,
+      "grad_norm": 4.90625,
+      "learning_rate": 8.092140650562665e-06,
+      "loss": 0.7207,
+      "step": 222
+    },
+    {
+      "epoch": 0.3604040404040404,
+      "grad_norm": 5.28125,
+      "learning_rate": 8.069890028371672e-06,
+      "loss": 0.7831,
+      "step": 223
+    },
+    {
+      "epoch": 0.362020202020202,
+      "grad_norm": 5.09375,
+      "learning_rate": 8.047541395856661e-06,
+      "loss": 0.6833,
+      "step": 224
+    },
+    {
+      "epoch": 0.36363636363636365,
+      "grad_norm": 4.9375,
+      "learning_rate": 8.025095466527468e-06,
+      "loss": 0.8531,
+      "step": 225
+    },
+    {
+      "epoch": 0.3652525252525253,
+      "grad_norm": 5.21875,
+      "learning_rate": 8.002552957000254e-06,
+      "loss": 0.7006,
+      "step": 226
+    },
+    {
+      "epoch": 0.36686868686868684,
+      "grad_norm": 5.0625,
+      "learning_rate": 7.979914586974628e-06,
+      "loss": 0.8985,
+      "step": 227
+    },
+    {
+      "epoch": 0.36848484848484847,
+      "grad_norm": 5.375,
+      "learning_rate": 7.957181079210676e-06,
+      "loss": 0.5145,
+      "step": 228
+    },
+    {
+      "epoch": 0.3701010101010101,
+      "grad_norm": 6.125,
+      "learning_rate": 7.934353159505885e-06,
+      "loss": 0.6869,
+      "step": 229
+    },
+    {
+      "epoch": 0.3717171717171717,
+      "grad_norm": 11.375,
+      "learning_rate": 7.911431556671967e-06,
+      "loss": 0.4694,
+      "step": 230
+    },
+    {
+      "epoch": 0.37333333333333335,
+      "grad_norm": 5.40625,
+      "learning_rate": 7.888417002511592e-06,
+      "loss": 0.5565,
+      "step": 231
+    },
+    {
+      "epoch": 0.374949494949495,
+      "grad_norm": 4.375,
+      "learning_rate": 7.865310231795026e-06,
+      "loss": 0.4966,
+      "step": 232
+    },
+    {
+      "epoch": 0.37656565656565655,
+      "grad_norm": 5.0625,
+      "learning_rate": 7.842111982236669e-06,
+      "loss": 0.6889,
+      "step": 233
+    },
+    {
+      "epoch": 0.3781818181818182,
+      "grad_norm": 6.28125,
+      "learning_rate": 7.818822994471504e-06,
+      "loss": 0.4771,
+      "step": 234
+    },
+    {
+      "epoch": 0.3797979797979798,
+      "grad_norm": 6.0,
+      "learning_rate": 7.79544401203146e-06,
+      "loss": 0.7411,
+      "step": 235
+    },
+    {
+      "epoch": 0.3814141414141414,
+      "grad_norm": 5.03125,
+      "learning_rate": 7.771975781321655e-06,
+      "loss": 0.7665,
+      "step": 236
+    },
+    {
+      "epoch": 0.38303030303030305,
+      "grad_norm": 5.21875,
+      "learning_rate": 7.748419051596586e-06,
+      "loss": 0.7166,
+      "step": 237
+    },
+    {
+      "epoch": 0.3846464646464646,
+      "grad_norm": 4.8125,
+      "learning_rate": 7.72477457493619e-06,
+      "loss": 0.4777,
+      "step": 238
+    },
+    {
+      "epoch": 0.38626262626262625,
+      "grad_norm": 5.21875,
+      "learning_rate": 7.701043106221847e-06,
+      "loss": 0.5036,
+      "step": 239
+    },
+    {
+      "epoch": 0.3878787878787879,
+      "grad_norm": 5.5625,
+      "learning_rate": 7.677225403112277e-06,
+      "loss": 0.6378,
+      "step": 240
+    },
+    {
+      "epoch": 0.3894949494949495,
+      "grad_norm": 5.3125,
+      "learning_rate": 7.653322226019341e-06,
+      "loss": 0.6353,
+      "step": 241
+    },
+    {
+      "epoch": 0.39111111111111113,
+      "grad_norm": 5.21875,
+      "learning_rate": 7.629334338083774e-06,
+      "loss": 0.7203,
+      "step": 242
+    },
+    {
+      "epoch": 0.3927272727272727,
+      "grad_norm": 4.96875,
+      "learning_rate": 7.605262505150819e-06,
+      "loss": 0.6799,
+      "step": 243
+    },
+    {
+      "epoch": 0.39434343434343433,
+      "grad_norm": 4.09375,
+      "learning_rate": 7.58110749574577e-06,
+      "loss": 0.3474,
+      "step": 244
+    },
+    {
+      "epoch": 0.39595959595959596,
+      "grad_norm": 5.9375,
+      "learning_rate": 7.556870081049444e-06,
+      "loss": 0.6227,
+      "step": 245
+    },
+    {
+      "epoch": 0.3975757575757576,
+      "grad_norm": 4.71875,
+      "learning_rate": 7.532551034873558e-06,
+      "loss": 0.664,
+      "step": 246
+    },
+    {
+      "epoch": 0.3975757575757576,
+      "eval_loss": 0.6342406272888184,
+      "eval_runtime": 36.056,
+      "eval_samples_per_second": 30.508,
+      "eval_steps_per_second": 3.827,
+      "step": 246
+    },
+    {
+      "epoch": 0.3991919191919192,
+      "grad_norm": 4.96875,
+      "learning_rate": 7.50815113363602e-06,
+      "loss": 0.6073,
+      "step": 247
+    },
+    {
+      "epoch": 0.40080808080808084,
+      "grad_norm": 5.78125,
+      "learning_rate": 7.483671156336142e-06,
+      "loss": 0.6488,
+      "step": 248
+    },
+    {
+      "epoch": 0.4024242424242424,
+      "grad_norm": 5.4375,
+      "learning_rate": 7.459111884529775e-06,
+      "loss": 0.6575,
+      "step": 249
+    },
+    {
+      "epoch": 0.40404040404040403,
+      "grad_norm": 5.71875,
+      "learning_rate": 7.43447410230435e-06,
+      "loss": 0.5836,
+      "step": 250
+    },
+    {
+      "epoch": 0.40565656565656566,
+      "grad_norm": 5.15625,
+      "learning_rate": 7.409758596253849e-06,
+      "loss": 0.589,
+      "step": 251
+    },
+    {
+      "epoch": 0.4072727272727273,
+      "grad_norm": 4.625,
+      "learning_rate": 7.384966155453686e-06,
+      "loss": 0.6821,
+      "step": 252
+    },
+    {
+      "epoch": 0.4088888888888889,
+      "grad_norm": 6.375,
+      "learning_rate": 7.360097571435527e-06,
+      "loss": 0.6267,
+      "step": 253
+    },
+    {
+      "epoch": 0.4105050505050505,
+      "grad_norm": 5.15625,
+      "learning_rate": 7.335153638162005e-06,
+      "loss": 0.6578,
+      "step": 254
+    },
+    {
+      "epoch": 0.4121212121212121,
+      "grad_norm": 5.65625,
+      "learning_rate": 7.310135152001381e-06,
+      "loss": 0.9352,
+      "step": 255
+    },
+    {
+      "epoch": 0.41373737373737374,
+      "grad_norm": 6.28125,
+      "learning_rate": 7.285042911702116e-06,
+      "loss": 0.7077,
+      "step": 256
+    },
+    {
+      "epoch": 0.41535353535353536,
+      "grad_norm": 6.46875,
+      "learning_rate": 7.259877718367372e-06,
+      "loss": 0.9397,
+      "step": 257
+    },
+    {
+      "epoch": 0.416969696969697,
+      "grad_norm": 5.8125,
+      "learning_rate": 7.234640375429427e-06,
+      "loss": 0.6653,
+      "step": 258
+    },
+    {
+      "epoch": 0.41858585858585856,
+      "grad_norm": 5.03125,
+      "learning_rate": 7.209331688624039e-06,
+      "loss": 0.7109,
+      "step": 259
+    },
+    {
+      "epoch": 0.4202020202020202,
+      "grad_norm": 4.71875,
+      "learning_rate": 7.183952465964711e-06,
+      "loss": 0.4387,
+      "step": 260
+    },
+    {
+      "epoch": 0.4218181818181818,
+      "grad_norm": 5.0,
+      "learning_rate": 7.158503517716894e-06,
+      "loss": 0.5614,
+      "step": 261
+    },
+    {
+      "epoch": 0.42343434343434344,
+      "grad_norm": 4.625,
+      "learning_rate": 7.132985656372126e-06,
+      "loss": 0.6086,
+      "step": 262
+    },
+    {
+      "epoch": 0.42505050505050507,
+      "grad_norm": 5.78125,
+      "learning_rate": 7.1073996966220835e-06,
+      "loss": 0.5876,
+      "step": 263
+    },
+    {
+      "epoch": 0.4266666666666667,
+      "grad_norm": 5.71875,
+      "learning_rate": 7.0817464553325764e-06,
+      "loss": 0.8125,
+      "step": 264
+    },
+    {
+      "epoch": 0.42828282828282827,
+      "grad_norm": 5.03125,
+      "learning_rate": 7.0560267515174685e-06,
+      "loss": 0.601,
+      "step": 265
+    },
+    {
+      "epoch": 0.4298989898989899,
+      "grad_norm": 5.4375,
+      "learning_rate": 7.030241406312528e-06,
+      "loss": 0.7071,
+      "step": 266
+    },
+    {
+      "epoch": 0.4315151515151515,
+      "grad_norm": 5.9375,
+      "learning_rate": 7.004391242949209e-06,
+      "loss": 0.8666,
+      "step": 267
+    },
+    {
+      "epoch": 0.43313131313131314,
+      "grad_norm": 5.53125,
+      "learning_rate": 6.978477086728375e-06,
+      "loss": 0.6213,
+      "step": 268
+    },
+    {
+      "epoch": 0.43474747474747477,
+      "grad_norm": 5.625,
+      "learning_rate": 6.952499764993945e-06,
+      "loss": 0.6405,
+      "step": 269
+    },
+    {
+      "epoch": 0.43636363636363634,
+      "grad_norm": 4.90625,
+      "learning_rate": 6.926460107106483e-06,
+      "loss": 0.7085,
+      "step": 270
+    },
+    {
+      "epoch": 0.43797979797979797,
+      "grad_norm": 5.28125,
+      "learning_rate": 6.900358944416712e-06,
+      "loss": 0.5896,
+      "step": 271
+    },
+    {
+      "epoch": 0.4395959595959596,
+      "grad_norm": 6.9375,
+      "learning_rate": 6.874197110238986e-06,
+      "loss": 0.7036,
+      "step": 272
+    },
+    {
+      "epoch": 0.4412121212121212,
+      "grad_norm": 5.625,
+      "learning_rate": 6.847975439824669e-06,
+      "loss": 0.9067,
+      "step": 273
+    },
+    {
+      "epoch": 0.44282828282828285,
+      "grad_norm": 5.0,
+      "learning_rate": 6.8216947703354815e-06,
+      "loss": 0.6876,
+      "step": 274
+    },
+    {
+      "epoch": 0.4444444444444444,
+      "grad_norm": 6.03125,
+      "learning_rate": 6.795355940816768e-06,
+      "loss": 0.8371,
+      "step": 275
+    },
+    {
+      "epoch": 0.44606060606060605,
+      "grad_norm": 5.9375,
+      "learning_rate": 6.7689597921707065e-06,
+      "loss": 0.8827,
+      "step": 276
+    },
+    {
+      "epoch": 0.4476767676767677,
+      "grad_norm": 4.59375,
+      "learning_rate": 6.742507167129465e-06,
+      "loss": 0.557,
+      "step": 277
+    },
+    {
+      "epoch": 0.4492929292929293,
+      "grad_norm": 4.5625,
+      "learning_rate": 6.715998910228296e-06,
+      "loss": 0.5257,
+      "step": 278
+    },
+    {
+      "epoch": 0.4509090909090909,
+      "grad_norm": 4.8125,
+      "learning_rate": 6.689435867778578e-06,
+      "loss": 0.6005,
+      "step": 279
+    },
+    {
+      "epoch": 0.45252525252525255,
+      "grad_norm": 5.78125,
+      "learning_rate": 6.6628188878407806e-06,
+      "loss": 0.6156,
+      "step": 280
+    },
+    {
+      "epoch": 0.4541414141414141,
+      "grad_norm": 4.8125,
+      "learning_rate": 6.636148820197409e-06,
+      "loss": 0.3638,
+      "step": 281
+    },
+    {
+      "epoch": 0.45575757575757575,
+      "grad_norm": 7.1875,
+      "learning_rate": 6.609426516325859e-06,
+      "loss": 0.6019,
+      "step": 282
+    },
+    {
+      "epoch": 0.4573737373737374,
+      "grad_norm": 4.8125,
+      "learning_rate": 6.58265282937124e-06,
+      "loss": 0.4772,
+      "step": 283
+    },
+    {
+      "epoch": 0.458989898989899,
+      "grad_norm": 5.1875,
+      "learning_rate": 6.555828614119132e-06,
+      "loss": 0.8364,
+      "step": 284
+    },
+    {
+      "epoch": 0.46060606060606063,
+      "grad_norm": 5.375,
+      "learning_rate": 6.528954726968302e-06,
+      "loss": 0.5237,
+      "step": 285
+    },
+    {
+      "epoch": 0.4622222222222222,
+      "grad_norm": 5.65625,
+      "learning_rate": 6.502032025903356e-06,
+      "loss": 0.7171,
+      "step": 286
+    },
+    {
+      "epoch": 0.4638383838383838,
+      "grad_norm": 4.71875,
+      "learning_rate": 6.475061370467346e-06,
+      "loss": 0.5332,
+      "step": 287
+    },
+    {
+      "epoch": 0.46545454545454545,
+      "grad_norm": 5.84375,
+      "learning_rate": 6.4480436217343366e-06,
+      "loss": 0.9313,
+      "step": 288
+    },
+    {
+      "epoch": 0.4670707070707071,
+      "grad_norm": 5.25,
+      "learning_rate": 6.420979642281909e-06,
+      "loss": 0.5537,
+      "step": 289
+    },
+    {
+      "epoch": 0.4686868686868687,
+      "grad_norm": 4.71875,
+      "learning_rate": 6.393870296163616e-06,
+      "loss": 0.5569,
+      "step": 290
+    },
+    {
+      "epoch": 0.4703030303030303,
+      "grad_norm": 5.625,
+      "learning_rate": 6.366716448881407e-06,
+      "loss": 0.4858,
+      "step": 291
+    },
+    {
+      "epoch": 0.4719191919191919,
+      "grad_norm": 5.625,
+      "learning_rate": 6.339518967357985e-06,
+      "loss": 0.6206,
+      "step": 292
+    },
+    {
+      "epoch": 0.47353535353535353,
+      "grad_norm": 5.03125,
+      "learning_rate": 6.312278719909138e-06,
+      "loss": 0.6686,
+      "step": 293
+    },
+    {
+      "epoch": 0.47515151515151516,
+      "grad_norm": 6.375,
+      "learning_rate": 6.284996576216014e-06,
+      "loss": 1.0407,
+      "step": 294
+    },
+    {
+      "epoch": 0.4767676767676768,
+      "grad_norm": 4.90625,
+      "learning_rate": 6.257673407297352e-06,
+      "loss": 0.6464,
+      "step": 295
+    },
+    {
+      "epoch": 0.4783838383838384,
+      "grad_norm": 5.9375,
+      "learning_rate": 6.230310085481677e-06,
+      "loss": 0.5525,
+      "step": 296
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.84375,
+      "learning_rate": 6.2029074843794445e-06,
+      "loss": 0.6877,
+      "step": 297
+    },
+    {
+      "epoch": 0.4816161616161616,
+      "grad_norm": 5.75,
+      "learning_rate": 6.175466478855161e-06,
+      "loss": 0.647,
+      "step": 298
+    },
+    {
+      "epoch": 0.48323232323232324,
+      "grad_norm": 5.5,
+      "learning_rate": 6.147987944999441e-06,
+      "loss": 0.8442,
+      "step": 299
+    },
+    {
+      "epoch": 0.48484848484848486,
+      "grad_norm": 5.5625,
+      "learning_rate": 6.1204727601010396e-06,
+      "loss": 0.5415,
+      "step": 300
+    },
+    {
+      "epoch": 0.4864646464646465,
+      "grad_norm": 5.1875,
+      "learning_rate": 6.092921802618849e-06,
+      "loss": 0.4224,
+      "step": 301
+    },
+    {
+      "epoch": 0.48808080808080806,
+      "grad_norm": 4.59375,
+      "learning_rate": 6.065335952153846e-06,
+      "loss": 0.5432,
+      "step": 302
+    },
+    {
+      "epoch": 0.4896969696969697,
+      "grad_norm": 4.25,
+      "learning_rate": 6.037716089421011e-06,
+      "loss": 0.4561,
+      "step": 303
+    },
+    {
+      "epoch": 0.4913131313131313,
+      "grad_norm": 5.34375,
+      "learning_rate": 6.010063096221215e-06,
+      "loss": 0.7473,
+      "step": 304
+    },
+    {
+      "epoch": 0.49292929292929294,
+      "grad_norm": 5.21875,
+      "learning_rate": 5.982377855413063e-06,
+      "loss": 0.5236,
+      "step": 305
+    },
+    {
+      "epoch": 0.49454545454545457,
+      "grad_norm": 4.96875,
+      "learning_rate": 5.954661250884704e-06,
+      "loss": 0.5116,
+      "step": 306
+    },
+    {
+      "epoch": 0.49616161616161614,
+      "grad_norm": 5.25,
+      "learning_rate": 5.926914167525618e-06,
+      "loss": 0.6585,
+      "step": 307
+    },
+    {
+      "epoch": 0.49777777777777776,
+      "grad_norm": 6.25,
+      "learning_rate": 5.899137491198364e-06,
+      "loss": 0.4622,
+      "step": 308
+    },
+    {
+      "epoch": 0.4993939393939394,
+      "grad_norm": 5.0,
+      "learning_rate": 5.871332108710292e-06,
+      "loss": 0.5889,
+      "step": 309
+    },
+    {
+      "epoch": 0.501010101010101,
+      "grad_norm": 4.59375,
+      "learning_rate": 5.843498907785236e-06,
+      "loss": 0.613,
+      "step": 310
+    },
+    {
+      "epoch": 0.5026262626262626,
+      "grad_norm": 5.9375,
+      "learning_rate": 5.815638777035175e-06,
+      "loss": 0.9093,
+      "step": 311
+    },
+    {
+      "epoch": 0.5042424242424243,
+      "grad_norm": 5.15625,
+      "learning_rate": 5.78775260593185e-06,
+      "loss": 0.806,
+      "step": 312
+    },
+    {
+      "epoch": 0.5058585858585859,
+      "grad_norm": 5.71875,
+      "learning_rate": 5.759841284778379e-06,
+      "loss": 0.8169,
+      "step": 313
+    },
+    {
+      "epoch": 0.5074747474747475,
+      "grad_norm": 4.71875,
+      "learning_rate": 5.731905704680834e-06,
+      "loss": 0.4293,
+      "step": 314
+    },
+    {
+      "epoch": 0.509090909090909,
+      "grad_norm": 4.8125,
+      "learning_rate": 5.703946757519777e-06,
+      "loss": 0.5602,
+      "step": 315
+    },
+    {
+      "epoch": 0.5107070707070707,
+      "grad_norm": 5.6875,
+      "learning_rate": 5.6759653359218e-06,
+      "loss": 0.7587,
+      "step": 316
+    },
+    {
+      "epoch": 0.5123232323232323,
+      "grad_norm": 4.6875,
+      "learning_rate": 5.647962333231021e-06,
+      "loss": 0.4005,
+      "step": 317
+    },
+    {
+      "epoch": 0.5139393939393939,
+      "grad_norm": 4.78125,
+      "learning_rate": 5.6199386434805615e-06,
+      "loss": 0.4869,
+      "step": 318
+    },
+    {
+      "epoch": 0.5155555555555555,
+      "grad_norm": 5.375,
+      "learning_rate": 5.591895161364006e-06,
+      "loss": 0.6297,
+      "step": 319
+    },
+    {
+      "epoch": 0.5171717171717172,
+      "grad_norm": 5.625,
+      "learning_rate": 5.563832782206835e-06,
+      "loss": 0.7082,
+      "step": 320
+    },
+    {
+      "epoch": 0.5187878787878788,
+      "grad_norm": 3.96875,
+      "learning_rate": 5.535752401937846e-06,
+      "loss": 0.2927,
+      "step": 321
+    },
+    {
+      "epoch": 0.5204040404040404,
+      "grad_norm": 5.46875,
+      "learning_rate": 5.507654917060541e-06,
+      "loss": 0.752,
+      "step": 322
+    },
+    {
+      "epoch": 0.522020202020202,
+      "grad_norm": 5.09375,
+      "learning_rate": 5.4795412246245126e-06,
+      "loss": 0.7145,
+      "step": 323
+    },
+    {
+      "epoch": 0.5236363636363637,
+      "grad_norm": 6.15625,
+      "learning_rate": 5.451412222196801e-06,
+      "loss": 0.8012,
+      "step": 324
+    },
+    {
+      "epoch": 0.5252525252525253,
+      "grad_norm": 5.15625,
+      "learning_rate": 5.4232688078332384e-06,
+      "loss": 0.7388,
+      "step": 325
+    },
+    {
+      "epoch": 0.5268686868686868,
+      "grad_norm": 4.625,
+      "learning_rate": 5.395111880049775e-06,
+      "loss": 0.3386,
+      "step": 326
+    },
+    {
+      "epoch": 0.5284848484848484,
+      "grad_norm": 5.71875,
+      "learning_rate": 5.366942337793798e-06,
+      "loss": 0.8716,
+      "step": 327
+    },
+    {
+      "epoch": 0.5301010101010101,
+      "grad_norm": 5.375,
+      "learning_rate": 5.338761080415425e-06,
+      "loss": 0.5071,
+      "step": 328
+    },
+    {
+      "epoch": 0.5317171717171717,
+      "grad_norm": 4.84375,
+      "learning_rate": 5.310569007638795e-06,
+      "loss": 0.8588,
+      "step": 329
+    },
+    {
+      "epoch": 0.5333333333333333,
+      "grad_norm": 5.78125,
+      "learning_rate": 5.28236701953335e-06,
+      "loss": 0.7365,
+      "step": 330
+    },
+    {
+      "epoch": 0.534949494949495,
+      "grad_norm": 4.75,
+      "learning_rate": 5.254156016485082e-06,
+      "loss": 0.4938,
+      "step": 331
+    },
+    {
+      "epoch": 0.5365656565656566,
+      "grad_norm": 4.8125,
+      "learning_rate": 5.225936899167803e-06,
+      "loss": 0.5953,
+      "step": 332
+    },
+    {
+      "epoch": 0.5381818181818182,
+      "grad_norm": 5.375,
+      "learning_rate": 5.197710568514381e-06,
+      "loss": 0.7735,
+      "step": 333
+    },
+    {
+      "epoch": 0.5397979797979798,
+      "grad_norm": 4.9375,
+      "learning_rate": 5.169477925687981e-06,
+      "loss": 0.5992,
+      "step": 334
+    },
+    {
+      "epoch": 0.5414141414141415,
+      "grad_norm": 5.21875,
+      "learning_rate": 5.141239872053294e-06,
+      "loss": 0.4403,
+      "step": 335
+    },
+    {
+      "epoch": 0.5430303030303031,
+      "grad_norm": 5.46875,
+      "learning_rate": 5.112997309147753e-06,
+      "loss": 0.9552,
+      "step": 336
+    },
+    {
+      "epoch": 0.5446464646464646,
+      "grad_norm": 5.15625,
+      "learning_rate": 5.084751138652764e-06,
+      "loss": 0.5166,
+      "step": 337
+    },
+    {
+      "epoch": 0.5462626262626262,
+      "grad_norm": 4.96875,
+      "learning_rate": 5.0565022623649e-06,
+      "loss": 0.508,
+      "step": 338
+    },
+    {
+      "epoch": 0.5478787878787879,
+      "grad_norm": 6.15625,
+      "learning_rate": 5.028251582167126e-06,
+      "loss": 0.5759,
+      "step": 339
+    },
+    {
+      "epoch": 0.5494949494949495,
+      "grad_norm": 5.1875,
+      "learning_rate": 5e-06,
+      "loss": 0.5942,
+      "step": 340
+    },
+    {
+      "epoch": 0.5511111111111111,
+      "grad_norm": 4.5625,
+      "learning_rate": 4.971748417832876e-06,
+      "loss": 0.4949,
+      "step": 341
+    },
+    {
+      "epoch": 0.5527272727272727,
+      "grad_norm": 5.40625,
+      "learning_rate": 4.943497737635103e-06,
+      "loss": 0.7265,
+      "step": 342
+    },
+    {
+      "epoch": 0.5543434343434344,
+      "grad_norm": 4.40625,
+      "learning_rate": 4.915248861347239e-06,
+      "loss": 0.5739,
+      "step": 343
+    },
+    {
+      "epoch": 0.555959595959596,
+      "grad_norm": 4.34375,
+      "learning_rate": 4.887002690852249e-06,
+      "loss": 0.3617,
+      "step": 344
+    },
+    {
+      "epoch": 0.5575757575757576,
+      "grad_norm": 4.78125,
+      "learning_rate": 4.858760127946707e-06,
+      "loss": 0.4464,
+      "step": 345
+    },
+    {
+      "epoch": 0.5591919191919192,
+      "grad_norm": 5.1875,
+      "learning_rate": 4.830522074312019e-06,
+      "loss": 0.8132,
+      "step": 346
+    },
+    {
+      "epoch": 0.5608080808080808,
+      "grad_norm": 5.21875,
+      "learning_rate": 4.80228943148562e-06,
+      "loss": 0.6365,
+      "step": 347
+    },
+    {
+      "epoch": 0.5624242424242424,
+      "grad_norm": 5.125,
+      "learning_rate": 4.774063100832199e-06,
+      "loss": 0.5667,
+      "step": 348
+    },
+    {
+      "epoch": 0.564040404040404,
+      "grad_norm": 4.625,
+      "learning_rate": 4.745843983514919e-06,
+      "loss": 0.4285,
+      "step": 349
+    },
+    {
+      "epoch": 0.5656565656565656,
+      "grad_norm": 4.3125,
+      "learning_rate": 4.717632980466652e-06,
+      "loss": 0.3623,
+      "step": 350
+    },
+    {
+      "epoch": 0.5672727272727273,
+      "grad_norm": 5.40625,
+      "learning_rate": 4.6894309923612055e-06,
+      "loss": 0.6415,
+      "step": 351
+    },
+    {
+      "epoch": 0.5688888888888889,
+      "grad_norm": 4.15625,
+      "learning_rate": 4.661238919584578e-06,
+      "loss": 0.4874,
+      "step": 352
+    },
+    {
+      "epoch": 0.5705050505050505,
+      "grad_norm": 5.59375,
+      "learning_rate": 4.633057662206205e-06,
+      "loss": 0.8178,
+      "step": 353
+    },
+    {
+      "epoch": 0.5721212121212121,
+      "grad_norm": 6.5625,
+      "learning_rate": 4.6048881199502265e-06,
+      "loss": 0.9008,
+      "step": 354
+    },
+    {
+      "epoch": 0.5737373737373738,
+      "grad_norm": 5.09375,
+      "learning_rate": 4.576731192166762e-06,
+      "loss": 0.7421,
+      "step": 355
+    },
+    {
+      "epoch": 0.5753535353535354,
+      "grad_norm": 4.78125,
+      "learning_rate": 4.548587777803198e-06,
+      "loss": 0.451,
+      "step": 356
+    },
+    {
+      "epoch": 0.576969696969697,
+      "grad_norm": 5.0,
+      "learning_rate": 4.520458775375488e-06,
+      "loss": 0.6153,
+      "step": 357
+    },
+    {
+      "epoch": 0.5785858585858585,
+      "grad_norm": 4.6875,
+      "learning_rate": 4.49234508293946e-06,
+      "loss": 0.489,
+      "step": 358
+    },
+    {
+      "epoch": 0.5802020202020202,
+      "grad_norm": 5.5625,
+      "learning_rate": 4.464247598062156e-06,
+      "loss": 0.4607,
+      "step": 359
+    },
+    {
+      "epoch": 0.5818181818181818,
+      "grad_norm": 4.90625,
+      "learning_rate": 4.436167217793167e-06,
+      "loss": 0.464,
+      "step": 360
+    },
+    {
+      "epoch": 0.5834343434343434,
+      "grad_norm": 4.375,
+      "learning_rate": 4.408104838635996e-06,
+      "loss": 0.5131,
+      "step": 361
+    },
+    {
+      "epoch": 0.585050505050505,
+      "grad_norm": 5.46875,
+      "learning_rate": 4.38006135651944e-06,
+      "loss": 0.6232,
+      "step": 362
+    },
+    {
+      "epoch": 0.5866666666666667,
+      "grad_norm": 4.65625,
+      "learning_rate": 4.3520376667689815e-06,
+      "loss": 0.536,
+      "step": 363
+    },
+    {
+      "epoch": 0.5882828282828283,
+      "grad_norm": 4.78125,
+      "learning_rate": 4.3240346640782014e-06,
+      "loss": 0.5407,
+      "step": 364
+    },
+    {
+      "epoch": 0.5898989898989899,
+      "grad_norm": 5.21875,
+      "learning_rate": 4.2960532424802235e-06,
+      "loss": 0.6939,
+      "step": 365
+    },
+    {
+      "epoch": 0.5915151515151515,
+      "grad_norm": 5.46875,
+      "learning_rate": 4.268094295319167e-06,
+      "loss": 0.9133,
+      "step": 366
+    },
+    {
+      "epoch": 0.5931313131313132,
+      "grad_norm": 6.375,
+      "learning_rate": 4.240158715221621e-06,
+      "loss": 0.9438,
+      "step": 367
+    },
+    {
+      "epoch": 0.5947474747474748,
+      "grad_norm": 5.84375,
+      "learning_rate": 4.212247394068151e-06,
+      "loss": 0.622,
+      "step": 368
+    },
+    {
+      "epoch": 0.5963636363636363,
+      "grad_norm": 4.5,
+      "learning_rate": 4.184361222964827e-06,
+      "loss": 0.3833,
+      "step": 369
+    },
+    {
+      "epoch": 0.5963636363636363,
+      "eval_loss": 0.6024488210678101,
+      "eval_runtime": 36.0571,
+      "eval_samples_per_second": 30.507,
+      "eval_steps_per_second": 3.827,
+      "step": 369
+    },
+    {
+      "epoch": 0.597979797979798,
+      "grad_norm": 5.5625,
+      "learning_rate": 4.1565010922147644e-06,
+      "loss": 0.7982,
+      "step": 370
+    },
+    {
+      "epoch": 0.5995959595959596,
+      "grad_norm": 5.09375,
+      "learning_rate": 4.1286678912897095e-06,
+      "loss": 0.4394,
+      "step": 371
+    },
+    {
+      "epoch": 0.6012121212121212,
+      "grad_norm": 4.84375,
+      "learning_rate": 4.100862508801639e-06,
+      "loss": 0.569,
+      "step": 372
+    },
+    {
+      "epoch": 0.6028282828282828,
+      "grad_norm": 6.0625,
+      "learning_rate": 4.0730858324743845e-06,
+      "loss": 0.5745,
+      "step": 373
+    },
+    {
+      "epoch": 0.6044444444444445,
+      "grad_norm": 6.3125,
+      "learning_rate": 4.045338749115299e-06,
+      "loss": 0.7263,
+      "step": 374
+    },
+    {
+      "epoch": 0.6060606060606061,
+      "grad_norm": 4.84375,
+      "learning_rate": 4.017622144586938e-06,
+      "loss": 0.7118,
+      "step": 375
+    },
+    {
+      "epoch": 0.6076767676767677,
+      "grad_norm": 5.21875,
+      "learning_rate": 3.989936903778785e-06,
+      "loss": 0.5991,
+      "step": 376
+    },
+    {
+      "epoch": 0.6092929292929293,
+      "grad_norm": 4.65625,
+      "learning_rate": 3.962283910578991e-06,
+      "loss": 0.4636,
+      "step": 377
+    },
+    {
+      "epoch": 0.610909090909091,
+      "grad_norm": 5.0625,
+      "learning_rate": 3.934664047846157e-06,
+      "loss": 0.5579,
+      "step": 378
+    },
+    {
+      "epoch": 0.6125252525252525,
+      "grad_norm": 5.6875,
+      "learning_rate": 3.907078197381153e-06,
+      "loss": 0.8509,
+      "step": 379
+    },
+    {
+      "epoch": 0.6141414141414141,
+      "grad_norm": 4.75,
+      "learning_rate": 3.879527239898962e-06,
+      "loss": 0.7272,
+      "step": 380
+    },
+    {
+      "epoch": 0.6157575757575757,
+      "grad_norm": 5.53125,
+      "learning_rate": 3.85201205500056e-06,
+      "loss": 0.4632,
+      "step": 381
+    },
+    {
+      "epoch": 0.6173737373737374,
+      "grad_norm": 3.921875,
+      "learning_rate": 3.8245335211448404e-06,
+      "loss": 0.3335,
+      "step": 382
+    },
+    {
+      "epoch": 0.618989898989899,
+      "grad_norm": 5.1875,
+      "learning_rate": 3.797092515620557e-06,
+      "loss": 0.4935,
+      "step": 383
+    },
+    {
+      "epoch": 0.6206060606060606,
+      "grad_norm": 3.9375,
+      "learning_rate": 3.769689914518326e-06,
+      "loss": 0.5223,
+      "step": 384
+    },
+    {
+      "epoch": 0.6222222222222222,
+      "grad_norm": 4.34375,
+      "learning_rate": 3.7423265927026473e-06,
+      "loss": 0.3271,
+      "step": 385
+    },
+    {
+      "epoch": 0.6238383838383839,
+      "grad_norm": 6.15625,
+      "learning_rate": 3.715003423783986e-06,
+      "loss": 0.6514,
+      "step": 386
+    },
+    {
+      "epoch": 0.6254545454545455,
+      "grad_norm": 5.1875,
+      "learning_rate": 3.6877212800908625e-06,
+      "loss": 0.6649,
+      "step": 387
+    },
+    {
+      "epoch": 0.6270707070707071,
+      "grad_norm": 4.875,
+      "learning_rate": 3.660481032642016e-06,
+      "loss": 0.5735,
+      "step": 388
+    },
+    {
+      "epoch": 0.6286868686868687,
+      "grad_norm": 5.78125,
+      "learning_rate": 3.633283551118595e-06,
+      "loss": 0.59,
+      "step": 389
+    },
+    {
+      "epoch": 0.6303030303030303,
+      "grad_norm": 5.59375,
+      "learning_rate": 3.6061297038363853e-06,
+      "loss": 0.6853,
+      "step": 390
+    },
+    {
+      "epoch": 0.6319191919191919,
+      "grad_norm": 5.4375,
+      "learning_rate": 3.579020357718092e-06,
+      "loss": 0.6379,
+      "step": 391
+    },
+    {
+      "epoch": 0.6335353535353535,
+      "grad_norm": 4.4375,
+      "learning_rate": 3.5519563782656642e-06,
+      "loss": 0.7215,
+      "step": 392
+    },
+    {
+      "epoch": 0.6351515151515151,
+      "grad_norm": 4.59375,
+      "learning_rate": 3.524938629532656e-06,
+      "loss": 0.4573,
+      "step": 393
+    },
+    {
+      "epoch": 0.6367676767676768,
+      "grad_norm": 5.3125,
+      "learning_rate": 3.497967974096647e-06,
+      "loss": 0.4821,
+      "step": 394
+    },
+    {
+      "epoch": 0.6383838383838384,
+      "grad_norm": 5.6875,
+      "learning_rate": 3.4710452730316978e-06,
+      "loss": 0.5618,
+      "step": 395
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 4.25,
+      "learning_rate": 3.4441713858808684e-06,
+      "loss": 0.6161,
+      "step": 396
+    },
+    {
+      "epoch": 0.6416161616161616,
+      "grad_norm": 5.875,
+      "learning_rate": 3.4173471706287607e-06,
+      "loss": 0.5043,
+      "step": 397
+    },
+    {
+      "epoch": 0.6432323232323233,
+      "grad_norm": 5.1875,
+      "learning_rate": 3.3905734836741415e-06,
+      "loss": 0.7049,
+      "step": 398
+    },
+    {
+      "epoch": 0.6448484848484849,
+      "grad_norm": 5.5,
+      "learning_rate": 3.3638511798025927e-06,
+      "loss": 0.587,
+      "step": 399
+    },
+    {
+      "epoch": 0.6464646464646465,
+      "grad_norm": 5.03125,
+      "learning_rate": 3.3371811121592203e-06,
+      "loss": 0.5026,
+      "step": 400
+    },
+    {
+      "epoch": 0.648080808080808,
+      "grad_norm": 4.96875,
+      "learning_rate": 3.3105641322214238e-06,
+      "loss": 0.6927,
+      "step": 401
+    },
+    {
+      "epoch": 0.6496969696969697,
+      "grad_norm": 5.125,
+      "learning_rate": 3.2840010897717045e-06,
+      "loss": 0.749,
+      "step": 402
+    },
+    {
+      "epoch": 0.6513131313131313,
+      "grad_norm": 6.09375,
+      "learning_rate": 3.257492832870537e-06,
+      "loss": 0.8106,
+      "step": 403
+    },
+    {
+      "epoch": 0.6529292929292929,
+      "grad_norm": 4.9375,
+      "learning_rate": 3.2310402078292956e-06,
+      "loss": 0.7662,
+      "step": 404
+    },
+    {
+      "epoch": 0.6545454545454545,
+      "grad_norm": 4.75,
+      "learning_rate": 3.2046440591832323e-06,
+      "loss": 0.4449,
+      "step": 405
+    },
+    {
+      "epoch": 0.6561616161616162,
+      "grad_norm": 4.71875,
+      "learning_rate": 3.178305229664519e-06,
+      "loss": 0.5476,
+      "step": 406
+    },
+    {
+      "epoch": 0.6577777777777778,
+      "grad_norm": 5.6875,
+      "learning_rate": 3.152024560175332e-06,
+      "loss": 0.7195,
+      "step": 407
+    },
+    {
+      "epoch": 0.6593939393939394,
+      "grad_norm": 4.625,
+      "learning_rate": 3.125802889761016e-06,
+      "loss": 0.4362,
+      "step": 408
+    },
+    {
+      "epoch": 0.661010101010101,
+      "grad_norm": 4.28125,
+      "learning_rate": 3.099641055583289e-06,
+      "loss": 0.3943,
+      "step": 409
+    },
+    {
+      "epoch": 0.6626262626262627,
+      "grad_norm": 5.15625,
+      "learning_rate": 3.073539892893519e-06,
+      "loss": 0.6311,
+      "step": 410
+    },
+    {
+      "epoch": 0.6642424242424242,
+      "grad_norm": 5.15625,
+      "learning_rate": 3.047500235006056e-06,
+      "loss": 0.4944,
+      "step": 411
+    },
+    {
+      "epoch": 0.6658585858585858,
+      "grad_norm": 5.28125,
+      "learning_rate": 3.021522913271627e-06,
+      "loss": 0.5021,
+      "step": 412
+    },
+    {
+      "epoch": 0.6674747474747474,
+      "grad_norm": 4.875,
+      "learning_rate": 2.995608757050793e-06,
+      "loss": 0.7164,
+      "step": 413
+    },
+    {
+      "epoch": 0.6690909090909091,
+      "grad_norm": 4.71875,
+      "learning_rate": 2.969758593687475e-06,
+      "loss": 0.4361,
+      "step": 414
+    },
+    {
+      "epoch": 0.6707070707070707,
+      "grad_norm": 5.0,
+      "learning_rate": 2.9439732484825323e-06,
+      "loss": 0.5235,
+      "step": 415
+    },
+    {
+      "epoch": 0.6723232323232323,
+      "grad_norm": 4.78125,
+      "learning_rate": 2.9182535446674244e-06,
+      "loss": 0.5304,
+      "step": 416
+    },
+    {
+      "epoch": 0.673939393939394,
+      "grad_norm": 6.09375,
+      "learning_rate": 2.8926003033779194e-06,
+      "loss": 0.4838,
+      "step": 417
+    },
+    {
+      "epoch": 0.6755555555555556,
+      "grad_norm": 5.75,
+      "learning_rate": 2.8670143436278757e-06,
+      "loss": 1.1298,
+      "step": 418
+    },
+    {
+      "epoch": 0.6771717171717172,
+      "grad_norm": 5.625,
+      "learning_rate": 2.8414964822831063e-06,
+      "loss": 0.5968,
+      "step": 419
+    },
+    {
+      "epoch": 0.6787878787878788,
+      "grad_norm": 4.96875,
+      "learning_rate": 2.8160475340352913e-06,
+      "loss": 0.4014,
+      "step": 420
+    },
+    {
+      "epoch": 0.6804040404040405,
+      "grad_norm": 5.875,
+      "learning_rate": 2.790668311375962e-06,
+      "loss": 0.6859,
+      "step": 421
+    },
+    {
+      "epoch": 0.682020202020202,
+      "grad_norm": 4.5625,
+      "learning_rate": 2.765359624570574e-06,
+      "loss": 0.463,
+      "step": 422
+    },
+    {
+      "epoch": 0.6836363636363636,
+      "grad_norm": 4.96875,
+      "learning_rate": 2.7401222816326316e-06,
+      "loss": 0.6448,
+      "step": 423
+    },
+    {
+      "epoch": 0.6852525252525252,
+      "grad_norm": 3.875,
+      "learning_rate": 2.714957088297886e-06,
+      "loss": 0.375,
+      "step": 424
+    },
+    {
+      "epoch": 0.6868686868686869,
+      "grad_norm": 4.65625,
+      "learning_rate": 2.6898648479986187e-06,
+      "loss": 0.5391,
+      "step": 425
+    },
+    {
+      "epoch": 0.6884848484848485,
+      "grad_norm": 4.625,
+      "learning_rate": 2.664846361837997e-06,
+      "loss": 0.7007,
+      "step": 426
+    },
+    {
+      "epoch": 0.6901010101010101,
+      "grad_norm": 5.21875,
+      "learning_rate": 2.639902428564475e-06,
+      "loss": 0.5808,
+      "step": 427
+    },
+    {
+      "epoch": 0.6917171717171717,
+      "grad_norm": 4.8125,
+      "learning_rate": 2.6150338445463146e-06,
+      "loss": 0.4929,
+      "step": 428
+    },
+    {
+      "epoch": 0.6933333333333334,
+      "grad_norm": 5.0625,
+      "learning_rate": 2.5902414037461544e-06,
+      "loss": 0.4209,
+      "step": 429
+    },
+    {
+      "epoch": 0.694949494949495,
+      "grad_norm": 5.0625,
+      "learning_rate": 2.565525897695651e-06,
+      "loss": 0.8549,
+      "step": 430
+    },
+    {
+      "epoch": 0.6965656565656566,
+      "grad_norm": 4.625,
+      "learning_rate": 2.540888115470226e-06,
+      "loss": 0.4793,
+      "step": 431
+    },
+    {
+      "epoch": 0.6981818181818182,
+      "grad_norm": 5.21875,
+      "learning_rate": 2.51632884366386e-06,
+      "loss": 0.5831,
+      "step": 432
+    },
+    {
+      "epoch": 0.6997979797979798,
+      "grad_norm": 5.5625,
+      "learning_rate": 2.4918488663639824e-06,
+      "loss": 0.5122,
+      "step": 433
+    },
+    {
+      "epoch": 0.7014141414141414,
+      "grad_norm": 4.6875,
+      "learning_rate": 2.4674489651264433e-06,
+      "loss": 0.5828,
+      "step": 434
+    },
+    {
+      "epoch": 0.703030303030303,
+      "grad_norm": 5.65625,
+      "learning_rate": 2.4431299189505563e-06,
+      "loss": 0.6851,
+      "step": 435
+    },
+    {
+      "epoch": 0.7046464646464646,
+      "grad_norm": 4.78125,
+      "learning_rate": 2.418892504254231e-06,
+      "loss": 0.5535,
+      "step": 436
+    },
+    {
+      "epoch": 0.7062626262626263,
+      "grad_norm": 4.6875,
+      "learning_rate": 2.394737494849184e-06,
+      "loss": 0.4243,
+      "step": 437
+    },
+    {
+      "epoch": 0.7078787878787879,
+      "grad_norm": 5.4375,
+      "learning_rate": 2.3706656619162278e-06,
+      "loss": 0.6774,
+      "step": 438
+    },
+    {
+      "epoch": 0.7094949494949495,
+      "grad_norm": 5.5,
+      "learning_rate": 2.34667777398066e-06,
+      "loss": 0.5865,
+      "step": 439
+    },
+    {
+      "epoch": 0.7111111111111111,
+      "grad_norm": 4.96875,
+      "learning_rate": 2.322774596887726e-06,
+      "loss": 0.4611,
+      "step": 440
+    },
+    {
+      "epoch": 0.7127272727272728,
+      "grad_norm": 4.5625,
+      "learning_rate": 2.298956893778154e-06,
+      "loss": 0.4801,
+      "step": 441
+    },
+    {
+      "epoch": 0.7143434343434344,
+      "grad_norm": 5.1875,
+      "learning_rate": 2.275225425063813e-06,
+      "loss": 0.5611,
+      "step": 442
+    },
+    {
+      "epoch": 0.7159595959595959,
+      "grad_norm": 5.09375,
+      "learning_rate": 2.251580948403416e-06,
+      "loss": 0.5869,
+      "step": 443
+    },
+    {
+      "epoch": 0.7175757575757575,
+      "grad_norm": 4.5,
+      "learning_rate": 2.2280242186783473e-06,
+      "loss": 0.4296,
+      "step": 444
+    },
+    {
+      "epoch": 0.7191919191919192,
+      "grad_norm": 4.34375,
+      "learning_rate": 2.204555987968541e-06,
+      "loss": 0.5446,
+      "step": 445
+    },
+    {
+      "epoch": 0.7208080808080808,
+      "grad_norm": 4.78125,
+      "learning_rate": 2.1811770055284968e-06,
+      "loss": 0.4477,
+      "step": 446
+    },
+    {
+      "epoch": 0.7224242424242424,
+      "grad_norm": 5.09375,
+      "learning_rate": 2.157888017763333e-06,
+      "loss": 0.7658,
+      "step": 447
+    },
+    {
+      "epoch": 0.724040404040404,
+      "grad_norm": 4.84375,
+      "learning_rate": 2.134689768204975e-06,
+      "loss": 0.41,
+      "step": 448
+    },
+    {
+      "epoch": 0.7256565656565657,
+      "grad_norm": 4.59375,
+      "learning_rate": 2.1115829974884097e-06,
+      "loss": 0.5721,
+      "step": 449
+    },
+    {
+      "epoch": 0.7272727272727273,
+      "grad_norm": 4.53125,
+      "learning_rate": 2.0885684433280336e-06,
+      "loss": 0.5407,
+      "step": 450
+    },
+    {
+      "epoch": 0.7288888888888889,
+      "grad_norm": 5.84375,
+      "learning_rate": 2.065646840494115e-06,
+      "loss": 0.6501,
+      "step": 451
+    },
+    {
+      "epoch": 0.7305050505050505,
+      "grad_norm": 4.9375,
+      "learning_rate": 2.042818920789326e-06,
+      "loss": 0.4143,
+      "step": 452
+    },
+    {
+      "epoch": 0.7321212121212122,
+      "grad_norm": 4.28125,
+      "learning_rate": 2.020085413025375e-06,
+      "loss": 0.4846,
+      "step": 453
+    },
+    {
+      "epoch": 0.7337373737373737,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.9974470429997482e-06,
+      "loss": 0.7159,
+      "step": 454
+    },
+    {
+      "epoch": 0.7353535353535353,
+      "grad_norm": 4.75,
+      "learning_rate": 1.974904533472532e-06,
+      "loss": 0.5781,
+      "step": 455
+    },
+    {
+      "epoch": 0.7369696969696969,
+      "grad_norm": 5.03125,
+      "learning_rate": 1.9524586041433393e-06,
+      "loss": 0.5525,
+      "step": 456
+    },
+    {
+      "epoch": 0.7385858585858586,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.9301099716283293e-06,
+      "loss": 0.7763,
+      "step": 457
+    },
+    {
+      "epoch": 0.7402020202020202,
+      "grad_norm": 5.78125,
+      "learning_rate": 1.907859349437336e-06,
+      "loss": 0.8187,
+      "step": 458
+    },
+    {
+      "epoch": 0.7418181818181818,
+      "grad_norm": 5.125,
+      "learning_rate": 1.8857074479510761e-06,
+      "loss": 0.4967,
+      "step": 459
+    },
+    {
+      "epoch": 0.7434343434343434,
+      "grad_norm": 4.78125,
+      "learning_rate": 1.8636549743984815e-06,
+      "loss": 0.558,
+      "step": 460
+    },
+    {
+      "epoch": 0.7450505050505051,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.8417026328341158e-06,
+      "loss": 0.3578,
+      "step": 461
+    },
+    {
+      "epoch": 0.7466666666666667,
+      "grad_norm": 4.34375,
+      "learning_rate": 1.8198511241156902e-06,
+      "loss": 0.4788,
+      "step": 462
+    },
+    {
+      "epoch": 0.7482828282828283,
+      "grad_norm": 6.375,
+      "learning_rate": 1.7981011458816988e-06,
+      "loss": 0.8778,
+      "step": 463
+    },
+    {
+      "epoch": 0.74989898989899,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.776453392529139e-06,
+      "loss": 0.491,
+      "step": 464
+    },
+    {
+      "epoch": 0.7515151515151515,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.7549085551913358e-06,
+      "loss": 0.4935,
+      "step": 465
+    },
+    {
+      "epoch": 0.7531313131313131,
+      "grad_norm": 5.28125,
+      "learning_rate": 1.7334673217158976e-06,
+      "loss": 0.6893,
+      "step": 466
+    },
+    {
+      "epoch": 0.7547474747474747,
+      "grad_norm": 5.9375,
+      "learning_rate": 1.7121303766427266e-06,
+      "loss": 0.6041,
+      "step": 467
+    },
+    {
+      "epoch": 0.7563636363636363,
+      "grad_norm": 5.5,
+      "learning_rate": 1.6908984011821883e-06,
+      "loss": 0.5337,
+      "step": 468
+    },
+    {
+      "epoch": 0.757979797979798,
+      "grad_norm": 5.25,
+      "learning_rate": 1.669772073193352e-06,
+      "loss": 0.6738,
+      "step": 469
+    },
+    {
+      "epoch": 0.7595959595959596,
+      "grad_norm": 5.25,
+      "learning_rate": 1.6487520671623469e-06,
+      "loss": 0.5704,
+      "step": 470
+    },
+    {
+      "epoch": 0.7612121212121212,
+      "grad_norm": 5.53125,
+      "learning_rate": 1.6278390541808364e-06,
+      "loss": 0.7108,
+      "step": 471
+    },
+    {
+      "epoch": 0.7628282828282829,
+      "grad_norm": 5.71875,
+      "learning_rate": 1.6070337019245896e-06,
+      "loss": 0.5632,
+      "step": 472
+    },
+    {
+      "epoch": 0.7644444444444445,
+      "grad_norm": 4.8125,
+      "learning_rate": 1.5863366746321578e-06,
+      "loss": 0.7405,
+      "step": 473
+    },
+    {
+      "epoch": 0.7660606060606061,
+      "grad_norm": 5.53125,
+      "learning_rate": 1.5657486330836786e-06,
+      "loss": 0.7586,
+      "step": 474
+    },
+    {
+      "epoch": 0.7676767676767676,
+      "grad_norm": 5.59375,
+      "learning_rate": 1.5452702345797738e-06,
+      "loss": 0.7458,
+      "step": 475
+    },
+    {
+      "epoch": 0.7692929292929293,
+      "grad_norm": 5.9375,
+      "learning_rate": 1.5249021329205638e-06,
+      "loss": 0.6163,
+      "step": 476
+    },
+    {
+      "epoch": 0.7709090909090909,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.5046449783847965e-06,
+      "loss": 0.4016,
+      "step": 477
+    },
+    {
+      "epoch": 0.7725252525252525,
+      "grad_norm": 4.34375,
+      "learning_rate": 1.4844994177090871e-06,
+      "loss": 0.4217,
+      "step": 478
+    },
+    {
+      "epoch": 0.7741414141414141,
+      "grad_norm": 5.28125,
+      "learning_rate": 1.4644660940672628e-06,
+      "loss": 0.753,
+      "step": 479
+    },
+    {
+      "epoch": 0.7757575757575758,
+      "grad_norm": 4.84375,
+      "learning_rate": 1.4445456470498392e-06,
+      "loss": 0.5012,
+      "step": 480
+    },
+    {
+      "epoch": 0.7773737373737374,
+      "grad_norm": 4.625,
+      "learning_rate": 1.4247387126435957e-06,
+      "loss": 0.4383,
+      "step": 481
+    },
+    {
+      "epoch": 0.778989898989899,
+      "grad_norm": 5.34375,
+      "learning_rate": 1.4050459232112652e-06,
+      "loss": 0.7582,
+      "step": 482
+    },
+    {
+      "epoch": 0.7806060606060606,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.3854679074713557e-06,
+      "loss": 0.5607,
+      "step": 483
+    },
+    {
+      "epoch": 0.7822222222222223,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.3660052904780707e-06,
+      "loss": 0.6204,
+      "step": 484
+    },
+    {
+      "epoch": 0.7838383838383839,
+      "grad_norm": 5.84375,
+      "learning_rate": 1.3466586936013548e-06,
+      "loss": 0.7345,
+      "step": 485
+    },
+    {
+      "epoch": 0.7854545454545454,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.3274287345070564e-06,
+      "loss": 0.598,
+      "step": 486
+    },
+    {
+      "epoch": 0.787070707070707,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.3083160271372092e-06,
+      "loss": 0.4424,
+      "step": 487
+    },
+    {
+      "epoch": 0.7886868686868687,
+      "grad_norm": 5.59375,
+      "learning_rate": 1.2893211816904243e-06,
+      "loss": 0.4828,
+      "step": 488
+    },
+    {
+      "epoch": 0.7903030303030303,
+      "grad_norm": 4.9375,
+      "learning_rate": 1.2704448046024192e-06,
+      "loss": 0.4666,
+      "step": 489
+    },
+    {
+      "epoch": 0.7919191919191919,
+      "grad_norm": 5.0,
+      "learning_rate": 1.2516874985266508e-06,
+      "loss": 0.6226,
+      "step": 490
+    },
+    {
+      "epoch": 0.7935353535353535,
+      "grad_norm": 3.984375,
+      "learning_rate": 1.2330498623150722e-06,
+      "loss": 0.6046,
+      "step": 491
+    },
+    {
+      "epoch": 0.7951515151515152,
+      "grad_norm": 5.03125,
+      "learning_rate": 1.2145324909990202e-06,
+      "loss": 0.4401,
+      "step": 492
+    },
+    {
+      "epoch": 0.7951515151515152,
+      "eval_loss": 0.5906627178192139,
+      "eval_runtime": 36.0661,
+      "eval_samples_per_second": 30.5,
+      "eval_steps_per_second": 3.826,
+      "step": 492
+    },
+    {
+      "epoch": 0.7967676767676768,
+      "grad_norm": 4.75,
+      "learning_rate": 1.1961359757702151e-06,
+      "loss": 0.5237,
+      "step": 493
+    },
+    {
+      "epoch": 0.7983838383838384,
+      "grad_norm": 4.84375,
+      "learning_rate": 1.1778609039618804e-06,
+      "loss": 0.6893,
+      "step": 494
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 5.71875,
+      "learning_rate": 1.1597078590300054e-06,
+      "loss": 0.901,
+      "step": 495
+    },
+    {
+      "epoch": 0.8016161616161617,
+      "grad_norm": 6.625,
+      "learning_rate": 1.1416774205347015e-06,
+      "loss": 0.7245,
+      "step": 496
+    },
+    {
+      "epoch": 0.8032323232323232,
+      "grad_norm": 5.375,
+      "learning_rate": 1.1237701641217097e-06,
+      "loss": 0.4828,
+      "step": 497
+    },
+    {
+      "epoch": 0.8048484848484848,
+      "grad_norm": 5.25,
+      "learning_rate": 1.1059866615040205e-06,
+      "loss": 0.6147,
+      "step": 498
+    },
+    {
+      "epoch": 0.8064646464646464,
+      "grad_norm": 5.53125,
+      "learning_rate": 1.0883274804436155e-06,
+      "loss": 0.6404,
+      "step": 499
+    },
+    {
+      "epoch": 0.8080808080808081,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.0707931847333487e-06,
+      "loss": 0.6509,
+      "step": 500
+    },
+    {
+      "epoch": 0.8096969696969697,
+      "grad_norm": 6.65625,
+      "learning_rate": 1.053384334178944e-06,
+      "loss": 0.6269,
+      "step": 501
+    },
+    {
+      "epoch": 0.8113131313131313,
+      "grad_norm": 3.984375,
+      "learning_rate": 1.036101484581117e-06,
+      "loss": 0.4626,
+      "step": 502
+    },
+    {
+      "epoch": 0.812929292929293,
+      "grad_norm": 4.65625,
+      "learning_rate": 1.0189451877178386e-06,
+      "loss": 0.4138,
+      "step": 503
+    },
+    {
+      "epoch": 0.8145454545454546,
+      "grad_norm": 4.8125,
+      "learning_rate": 1.0019159913267156e-06,
+      "loss": 0.5092,
+      "step": 504
+    },
+    {
+      "epoch": 0.8161616161616162,
+      "grad_norm": 4.875,
+      "learning_rate": 9.850144390875022e-07,
+      "loss": 0.7465,
+      "step": 505
+    },
+    {
+      "epoch": 0.8177777777777778,
+      "grad_norm": 5.21875,
+      "learning_rate": 9.68241070604743e-07,
+      "loss": 0.5662,
+      "step": 506
+    },
+    {
+      "epoch": 0.8193939393939393,
+      "grad_norm": 5.875,
+      "learning_rate": 9.51596421390547e-07,
+      "loss": 1.1438,
+      "step": 507
+    },
+    {
+      "epoch": 0.821010101010101,
+      "grad_norm": 6.46875,
+      "learning_rate": 9.350810228474855e-07,
+      "loss": 0.614,
+      "step": 508
+    },
+    {
+      "epoch": 0.8226262626262626,
+      "grad_norm": 5.46875,
+      "learning_rate": 9.186954022516343e-07,
+      "loss": 0.6482,
+      "step": 509
+    },
+    {
+      "epoch": 0.8242424242424242,
+      "grad_norm": 5.78125,
+      "learning_rate": 9.024400827357344e-07,
+      "loss": 0.7774,
+      "step": 510
+    },
+    {
+      "epoch": 0.8258585858585858,
+      "grad_norm": 4.875,
+      "learning_rate": 8.863155832724895e-07,
+      "loss": 0.573,
+      "step": 511
+    },
+    {
+      "epoch": 0.8274747474747475,
+      "grad_norm": 5.65625,
+      "learning_rate": 8.703224186580012e-07,
+      "loss": 0.5287,
+      "step": 512
+    },
+    {
+      "epoch": 0.8290909090909091,
+      "grad_norm": 4.25,
+      "learning_rate": 8.544610994953317e-07,
+      "loss": 0.3565,
+      "step": 513
+    },
+    {
+      "epoch": 0.8307070707070707,
+      "grad_norm": 5.375,
+      "learning_rate": 8.387321321781977e-07,
+      "loss": 0.566,
+      "step": 514
+    },
+    {
+      "epoch": 0.8323232323232324,
+      "grad_norm": 5.46875,
+      "learning_rate": 8.23136018874815e-07,
+      "loss": 0.4459,
+      "step": 515
+    },
+    {
+      "epoch": 0.833939393939394,
+      "grad_norm": 4.28125,
+      "learning_rate": 8.07673257511849e-07,
+      "loss": 0.5429,
+      "step": 516
+    },
+    {
+      "epoch": 0.8355555555555556,
+      "grad_norm": 5.5,
+      "learning_rate": 7.923443417585324e-07,
+      "loss": 0.6166,
+      "step": 517
+    },
+    {
+      "epoch": 0.8371717171717171,
+      "grad_norm": 5.125,
+      "learning_rate": 7.771497610108981e-07,
+      "loss": 0.498,
+      "step": 518
+    },
+    {
+      "epoch": 0.8387878787878787,
+      "grad_norm": 5.625,
+      "learning_rate": 7.620900003761561e-07,
+      "loss": 0.9846,
+      "step": 519
+    },
+    {
+      "epoch": 0.8404040404040404,
+      "grad_norm": 4.875,
+      "learning_rate": 7.471655406572003e-07,
+      "loss": 0.745,
+      "step": 520
+    },
+    {
+      "epoch": 0.842020202020202,
+      "grad_norm": 4.28125,
+      "learning_rate": 7.323768583372675e-07,
+      "loss": 0.423,
+      "step": 521
+    },
+    {
+      "epoch": 0.8436363636363636,
+      "grad_norm": 5.625,
+      "learning_rate": 7.177244255647209e-07,
+      "loss": 0.5674,
+      "step": 522
+    },
+    {
+      "epoch": 0.8452525252525253,
+      "grad_norm": 5.53125,
+      "learning_rate": 7.032087101379703e-07,
+      "loss": 0.6423,
+      "step": 523
+    },
+    {
+      "epoch": 0.8468686868686869,
+      "grad_norm": 9.75,
+      "learning_rate": 6.888301754905469e-07,
+      "loss": 0.6539,
+      "step": 524
+    },
+    {
+      "epoch": 0.8484848484848485,
+      "grad_norm": 3.8125,
+      "learning_rate": 6.745892806763015e-07,
+      "loss": 0.3109,
+      "step": 525
+    },
+    {
+      "epoch": 0.8501010101010101,
+      "grad_norm": 4.40625,
+      "learning_rate": 6.604864803547511e-07,
+      "loss": 0.3802,
+      "step": 526
+    },
+    {
+      "epoch": 0.8517171717171718,
+      "grad_norm": 4.46875,
+      "learning_rate": 6.465222247765618e-07,
+      "loss": 0.452,
+      "step": 527
+    },
+    {
+      "epoch": 0.8533333333333334,
+      "grad_norm": 5.625,
+      "learning_rate": 6.326969597691724e-07,
+      "loss": 0.6779,
+      "step": 528
+    },
+    {
+      "epoch": 0.8549494949494949,
+      "grad_norm": 5.46875,
+      "learning_rate": 6.190111267225651e-07,
+      "loss": 0.7927,
+      "step": 529
+    },
+    {
+      "epoch": 0.8565656565656565,
+      "grad_norm": 4.9375,
+      "learning_rate": 6.054651625751717e-07,
+      "loss": 0.5197,
+      "step": 530
+    },
+    {
+      "epoch": 0.8581818181818182,
+      "grad_norm": 5.0,
+      "learning_rate": 5.920594997999202e-07,
+      "loss": 0.617,
+      "step": 531
+    },
+    {
+      "epoch": 0.8597979797979798,
+      "grad_norm": 5.5,
+      "learning_rate": 5.787945663904332e-07,
+      "loss": 0.5004,
+      "step": 532
+    },
+    {
+      "epoch": 0.8614141414141414,
+      "grad_norm": 4.65625,
+      "learning_rate": 5.65670785847362e-07,
+      "loss": 0.6185,
+      "step": 533
+    },
+    {
+      "epoch": 0.863030303030303,
+      "grad_norm": 4.90625,
+      "learning_rate": 5.526885771648599e-07,
+      "loss": 0.4621,
+      "step": 534
+    },
+    {
+      "epoch": 0.8646464646464647,
+      "grad_norm": 4.8125,
+      "learning_rate": 5.398483548172162e-07,
+      "loss": 0.7177,
+      "step": 535
+    },
+    {
+      "epoch": 0.8662626262626263,
+      "grad_norm": 5.0625,
+      "learning_rate": 5.271505287456153e-07,
+      "loss": 0.4255,
+      "step": 536
+    },
+    {
+      "epoch": 0.8678787878787879,
+      "grad_norm": 4.8125,
+      "learning_rate": 5.145955043450484e-07,
+      "loss": 0.3932,
+      "step": 537
+    },
+    {
+      "epoch": 0.8694949494949495,
+      "grad_norm": 5.75,
+      "learning_rate": 5.021836824513759e-07,
+      "loss": 0.758,
+      "step": 538
+    },
+    {
+      "epoch": 0.8711111111111111,
+      "grad_norm": 5.78125,
+      "learning_rate": 4.899154593285294e-07,
+      "loss": 0.5175,
+      "step": 539
+    },
+    {
+      "epoch": 0.8727272727272727,
+      "grad_norm": 4.90625,
+      "learning_rate": 4.777912266558532e-07,
+      "loss": 0.5595,
+      "step": 540
+    },
+    {
+      "epoch": 0.8743434343434343,
+      "grad_norm": 5.65625,
+      "learning_rate": 4.658113715156098e-07,
+      "loss": 0.7423,
+      "step": 541
+    },
+    {
+      "epoch": 0.8759595959595959,
+      "grad_norm": 5.03125,
+      "learning_rate": 4.5397627638061604e-07,
+      "loss": 0.456,
+      "step": 542
+    },
+    {
+      "epoch": 0.8775757575757576,
+      "grad_norm": 4.4375,
+      "learning_rate": 4.422863191020299e-07,
+      "loss": 0.7905,
+      "step": 543
+    },
+    {
+      "epoch": 0.8791919191919192,
+      "grad_norm": 4.96875,
+      "learning_rate": 4.307418728972934e-07,
+      "loss": 0.4821,
+      "step": 544
+    },
+    {
+      "epoch": 0.8808080808080808,
+      "grad_norm": 4.46875,
+      "learning_rate": 4.193433063382135e-07,
+      "loss": 0.7014,
+      "step": 545
+    },
+    {
+      "epoch": 0.8824242424242424,
+      "grad_norm": 4.78125,
+      "learning_rate": 4.080909833391944e-07,
+      "loss": 0.6715,
+      "step": 546
+    },
+    {
+      "epoch": 0.8840404040404041,
+      "grad_norm": 4.78125,
+      "learning_rate": 3.9698526314562114e-07,
+      "loss": 0.629,
+      "step": 547
+    },
+    {
+      "epoch": 0.8856565656565657,
+      "grad_norm": 5.03125,
+      "learning_rate": 3.8602650032238675e-07,
+      "loss": 0.7098,
+      "step": 548
+    },
+    {
+      "epoch": 0.8872727272727273,
+      "grad_norm": 5.3125,
+      "learning_rate": 3.752150447425773e-07,
+      "loss": 0.6205,
+      "step": 549
+    },
+    {
+      "epoch": 0.8888888888888888,
+      "grad_norm": 4.8125,
+      "learning_rate": 3.6455124157629805e-07,
+      "loss": 0.4824,
+      "step": 550
+    },
+    {
+      "epoch": 0.8905050505050505,
+      "grad_norm": 4.8125,
+      "learning_rate": 3.5403543127965514e-07,
+      "loss": 0.5424,
+      "step": 551
+    },
+    {
+      "epoch": 0.8921212121212121,
+      "grad_norm": 5.90625,
+      "learning_rate": 3.436679495838835e-07,
+      "loss": 0.791,
+      "step": 552
+    },
+    {
+      "epoch": 0.8937373737373737,
+      "grad_norm": 5.40625,
+      "learning_rate": 3.334491274846324e-07,
+      "loss": 0.7193,
+      "step": 553
+    },
+    {
+      "epoch": 0.8953535353535353,
+      "grad_norm": 5.09375,
+      "learning_rate": 3.233792912313943e-07,
+      "loss": 0.5068,
+      "step": 554
+    },
+    {
+      "epoch": 0.896969696969697,
+      "grad_norm": 5.21875,
+      "learning_rate": 3.134587623170909e-07,
+      "loss": 0.6015,
+      "step": 555
+    },
+    {
+      "epoch": 0.8985858585858586,
+      "grad_norm": 5.1875,
+      "learning_rate": 3.0368785746780925e-07,
+      "loss": 0.5218,
+      "step": 556
+    },
+    {
+      "epoch": 0.9002020202020202,
+      "grad_norm": 5.1875,
+      "learning_rate": 2.940668886326864e-07,
+      "loss": 0.5964,
+      "step": 557
+    },
+    {
+      "epoch": 0.9018181818181819,
+      "grad_norm": 4.9375,
+      "learning_rate": 2.8459616297395464e-07,
+      "loss": 0.4713,
+      "step": 558
+    },
+    {
+      "epoch": 0.9034343434343435,
+      "grad_norm": 5.78125,
+      "learning_rate": 2.7527598285713387e-07,
+      "loss": 0.6696,
+      "step": 559
+    },
+    {
+      "epoch": 0.9050505050505051,
+      "grad_norm": 4.6875,
+      "learning_rate": 2.6610664584137413e-07,
+      "loss": 0.3846,
+      "step": 560
+    },
+    {
+      "epoch": 0.9066666666666666,
+      "grad_norm": 5.3125,
+      "learning_rate": 2.570884446699612e-07,
+      "loss": 0.644,
+      "step": 561
+    },
+    {
+      "epoch": 0.9082828282828282,
+      "grad_norm": 5.71875,
+      "learning_rate": 2.482216672609677e-07,
+      "loss": 0.722,
+      "step": 562
+    },
+    {
+      "epoch": 0.9098989898989899,
+      "grad_norm": 4.75,
+      "learning_rate": 2.3950659669806033e-07,
+      "loss": 0.6158,
+      "step": 563
+    },
+    {
+      "epoch": 0.9115151515151515,
+      "grad_norm": 4.6875,
+      "learning_rate": 2.3094351122146307e-07,
+      "loss": 0.4624,
+      "step": 564
+    },
+    {
+      "epoch": 0.9131313131313131,
+      "grad_norm": 5.15625,
+      "learning_rate": 2.2253268421907503e-07,
+      "loss": 0.4587,
+      "step": 565
+    },
+    {
+      "epoch": 0.9147474747474748,
+      "grad_norm": 5.28125,
+      "learning_rate": 2.142743842177386e-07,
+      "loss": 0.8627,
+      "step": 566
+    },
+    {
+      "epoch": 0.9163636363636364,
+      "grad_norm": 5.75,
+      "learning_rate": 2.0616887487467096e-07,
+      "loss": 0.7492,
+      "step": 567
+    },
+    {
+      "epoch": 0.917979797979798,
+      "grad_norm": 4.875,
+      "learning_rate": 1.98216414969043e-07,
+      "loss": 0.5206,
+      "step": 568
+    },
+    {
+      "epoch": 0.9195959595959596,
+      "grad_norm": 4.96875,
+      "learning_rate": 1.9041725839371805e-07,
+      "loss": 0.5692,
+      "step": 569
+    },
+    {
+      "epoch": 0.9212121212121213,
+      "grad_norm": 5.15625,
+      "learning_rate": 1.8277165414714858e-07,
+      "loss": 0.5574,
+      "step": 570
+    },
+    {
+      "epoch": 0.9228282828282828,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.752798463254235e-07,
+      "loss": 0.5013,
+      "step": 571
+    },
+    {
+      "epoch": 0.9244444444444444,
+      "grad_norm": 4.875,
+      "learning_rate": 1.6794207411447548e-07,
+      "loss": 0.5984,
+      "step": 572
+    },
+    {
+      "epoch": 0.926060606060606,
+      "grad_norm": 5.5625,
+      "learning_rate": 1.6075857178244613e-07,
+      "loss": 0.7467,
+      "step": 573
+    },
+    {
+      "epoch": 0.9276767676767677,
+      "grad_norm": 4.25,
+      "learning_rate": 1.5372956867220678e-07,
+      "loss": 0.6308,
+      "step": 574
+    },
+    {
+      "epoch": 0.9292929292929293,
+      "grad_norm": 4.71875,
+      "learning_rate": 1.4685528919403446e-07,
+      "loss": 0.5585,
+      "step": 575
+    },
+    {
+      "epoch": 0.9309090909090909,
+      "grad_norm": 5.71875,
+      "learning_rate": 1.4013595281844872e-07,
+      "loss": 0.5267,
+      "step": 576
+    },
+    {
+      "epoch": 0.9325252525252525,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.335717740692033e-07,
+      "loss": 0.4043,
+      "step": 577
+    },
+    {
+      "epoch": 0.9341414141414142,
+      "grad_norm": 5.875,
+      "learning_rate": 1.2716296251644e-07,
+      "loss": 0.6355,
+      "step": 578
+    },
+    {
+      "epoch": 0.9357575757575758,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.2090972276999513e-07,
+      "loss": 0.9279,
+      "step": 579
+    },
+    {
+      "epoch": 0.9373737373737374,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.1481225447286803e-07,
+      "loss": 0.69,
+      "step": 580
+    },
+    {
+      "epoch": 0.938989898989899,
+      "grad_norm": 5.90625,
+      "learning_rate": 1.0887075229484789e-07,
+      "loss": 0.8458,
+      "step": 581
+    },
+    {
+      "epoch": 0.9406060606060606,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.0308540592629756e-07,
+      "loss": 0.7002,
+      "step": 582
+    },
+    {
+      "epoch": 0.9422222222222222,
+      "grad_norm": 4.46875,
+      "learning_rate": 9.745640007209844e-08,
+      "loss": 0.5164,
+      "step": 583
+    },
+    {
+      "epoch": 0.9438383838383838,
+      "grad_norm": 4.28125,
+      "learning_rate": 9.198391444575072e-08,
+      "loss": 0.7122,
+      "step": 584
+    },
+    {
+      "epoch": 0.9454545454545454,
+      "grad_norm": 5.15625,
+      "learning_rate": 8.666812376364187e-08,
+      "loss": 0.4656,
+      "step": 585
+    },
+    {
+      "epoch": 0.9470707070707071,
+      "grad_norm": 5.625,
+      "learning_rate": 8.150919773946165e-08,
+      "loss": 0.728,
+      "step": 586
+    },
+    {
+      "epoch": 0.9486868686868687,
+      "grad_norm": 5.1875,
+      "learning_rate": 7.650730107878812e-08,
+      "loss": 0.5756,
+      "step": 587
+    },
+    {
+      "epoch": 0.9503030303030303,
+      "grad_norm": 5.5,
+      "learning_rate": 7.166259347382854e-08,
+      "loss": 0.5356,
+      "step": 588
+    },
+    {
+      "epoch": 0.9519191919191919,
+      "grad_norm": 5.4375,
+      "learning_rate": 6.697522959831837e-08,
+      "loss": 0.6048,
+      "step": 589
+    },
+    {
+      "epoch": 0.9535353535353536,
+      "grad_norm": 4.46875,
+      "learning_rate": 6.244535910258697e-08,
+      "loss": 0.4466,
+      "step": 590
+    },
+    {
+      "epoch": 0.9551515151515152,
+      "grad_norm": 5.0625,
+      "learning_rate": 5.8073126608778064e-08,
+      "loss": 0.4927,
+      "step": 591
+    },
+    {
+      "epoch": 0.9567676767676768,
+      "grad_norm": 4.53125,
+      "learning_rate": 5.3858671706230605e-08,
+      "loss": 0.4057,
+      "step": 592
+    },
+    {
+      "epoch": 0.9583838383838383,
+      "grad_norm": 4.96875,
+      "learning_rate": 4.98021289470263e-08,
+      "loss": 0.4878,
+      "step": 593
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 5.5,
+      "learning_rate": 4.590362784169022e-08,
+      "loss": 0.5471,
+      "step": 594
+    },
+    {
+      "epoch": 0.9616161616161616,
+      "grad_norm": 5.09375,
+      "learning_rate": 4.2163292855056936e-08,
+      "loss": 0.5259,
+      "step": 595
+    },
+    {
+      "epoch": 0.9632323232323232,
+      "grad_norm": 4.65625,
+      "learning_rate": 3.858124340229863e-08,
+      "loss": 0.443,
+      "step": 596
+    },
+    {
+      "epoch": 0.9648484848484848,
+      "grad_norm": 4.78125,
+      "learning_rate": 3.515759384510986e-08,
+      "loss": 0.5902,
+      "step": 597
+    },
+    {
+      "epoch": 0.9664646464646465,
+      "grad_norm": 6.09375,
+      "learning_rate": 3.1892453488058803e-08,
+      "loss": 0.6485,
+      "step": 598
+    },
+    {
+      "epoch": 0.9680808080808081,
+      "grad_norm": 4.78125,
+      "learning_rate": 2.87859265750956e-08,
+      "loss": 0.6152,
+      "step": 599
+    },
+    {
+      "epoch": 0.9696969696969697,
+      "grad_norm": 5.03125,
+      "learning_rate": 2.5838112286226123e-08,
+      "loss": 0.62,
+      "step": 600
+    },
+    {
+      "epoch": 0.9713131313131314,
+      "grad_norm": 5.875,
+      "learning_rate": 2.304910473434341e-08,
+      "loss": 0.7103,
+      "step": 601
+    },
+    {
+      "epoch": 0.972929292929293,
+      "grad_norm": 7.03125,
+      "learning_rate": 2.0418992962224495e-08,
+      "loss": 0.6722,
+      "step": 602
+    },
+    {
+      "epoch": 0.9745454545454545,
+      "grad_norm": 4.71875,
+      "learning_rate": 1.7947860939688255e-08,
+      "loss": 0.5565,
+      "step": 603
+    },
+    {
+      "epoch": 0.9761616161616161,
+      "grad_norm": 5.78125,
+      "learning_rate": 1.563578756091144e-08,
+      "loss": 0.697,
+      "step": 604
+    },
+    {
+      "epoch": 0.9777777777777777,
+      "grad_norm": 5.75,
+      "learning_rate": 1.3482846641914572e-08,
+      "loss": 0.6675,
+      "step": 605
+    },
+    {
+      "epoch": 0.9793939393939394,
+      "grad_norm": 4.96875,
+      "learning_rate": 1.1489106918200487e-08,
+      "loss": 0.4355,
+      "step": 606
+    },
+    {
+      "epoch": 0.981010101010101,
+      "grad_norm": 4.625,
+      "learning_rate": 9.654632042562229e-09,
+      "loss": 0.6112,
+      "step": 607
+    },
+    {
+      "epoch": 0.9826262626262626,
+      "grad_norm": 5.59375,
+      "learning_rate": 7.979480583052423e-09,
+      "loss": 1.1165,
+      "step": 608
+    },
+    {
+      "epoch": 0.9842424242424243,
+      "grad_norm": 4.96875,
+      "learning_rate": 6.4637060211092395e-09,
+      "loss": 0.3665,
+      "step": 609
+    },
+    {
+      "epoch": 0.9858585858585859,
+      "grad_norm": 5.0,
+      "learning_rate": 5.107356749853298e-09,
+      "loss": 0.6373,
+      "step": 610
+    },
+    {
+      "epoch": 0.9874747474747475,
+      "grad_norm": 4.84375,
+      "learning_rate": 3.910476072539471e-09,
+      "loss": 0.4729,
+      "step": 611
+    },
+    {
+      "epoch": 0.9890909090909091,
+      "grad_norm": 4.40625,
+      "learning_rate": 2.8731022011757593e-09,
+      "loss": 0.493,
+      "step": 612
+    },
+    {
+      "epoch": 0.9907070707070708,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.9952682553042722e-09,
+      "loss": 0.6581,
+      "step": 613
+    },
+    {
+      "epoch": 0.9923232323232323,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.2770022609409628e-09,
+      "loss": 0.5642,
+      "step": 614
+    },
+    {
+      "epoch": 0.9939393939393939,
+      "grad_norm": 5.40625,
+      "learning_rate": 7.18327149683562e-10,
+      "loss": 0.6746,
+      "step": 615
+    },
+    {
+      "epoch": 0.9939393939393939,
+      "eval_loss": 0.5900062322616577,
+      "eval_runtime": 36.0854,
+      "eval_samples_per_second": 30.483,
+      "eval_steps_per_second": 3.824,
+      "step": 615
+    },
+    {
+      "epoch": 0.9955555555555555,
+      "grad_norm": 4.59375,
+      "learning_rate": 3.1926075797827914e-10,
+      "loss": 0.8563,
+      "step": 616
+    },
+    {
+      "epoch": 0.9971717171717172,
+      "grad_norm": 4.125,
+      "learning_rate": 7.9815826551366e-11,
+      "loss": 0.6263,
+      "step": 617
+    },
+    {
+      "epoch": 0.9987878787878788,
+      "grad_norm": 4.5625,
+      "learning_rate": 0.0,
+      "loss": 0.552,
+      "step": 618
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 618,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.5051307725309542e+17,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}