diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,9941 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.975609756097561,
+  "eval_steps": 100,
+  "global_step": 1400,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.222031839173859,
+      "learning_rate": 2.0000000000000002e-07,
+      "loss": 1.0419,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "eval_loss": 1.111273169517517,
+      "eval_runtime": 187.0422,
+      "eval_samples_per_second": 216.272,
+      "eval_steps_per_second": 27.037,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.978909941838748,
+      "learning_rate": 4.0000000000000003e-07,
+      "loss": 1.1665,
+      "step": 2
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 6.130580056448725,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 1.0667,
+      "step": 3
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 6.089935092381693,
+      "learning_rate": 8.000000000000001e-07,
+      "loss": 1.1201,
+      "step": 4
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.442869698529515,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 1.1546,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.28471945792447,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 1.0699,
+      "step": 6
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.161065961031279,
+      "learning_rate": 1.4000000000000001e-06,
+      "loss": 1.0906,
+      "step": 7
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.4182824757873225,
+      "learning_rate": 1.6000000000000001e-06,
+      "loss": 1.0971,
+      "step": 8
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.306106150153299,
+      "learning_rate": 1.8000000000000001e-06,
+      "loss": 1.0531,
+      "step": 9
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.191198685577769,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 0.9869,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.533581010023776,
+      "learning_rate": 2.2e-06,
+      "loss": 0.9584,
+      "step": 11
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.0578264756667672,
+      "learning_rate": 2.4000000000000003e-06,
+      "loss": 1.0656,
+      "step": 12
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.8914865841383306,
+      "learning_rate": 2.6e-06,
+      "loss": 0.9628,
+      "step": 13
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.7786338729768105,
+      "learning_rate": 2.8000000000000003e-06,
+      "loss": 0.9298,
+      "step": 14
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.7245515964220006,
+      "learning_rate": 3e-06,
+      "loss": 0.9603,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.247401176863816,
+      "learning_rate": 3.2000000000000003e-06,
+      "loss": 0.9844,
+      "step": 16
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.9693650667856115,
+      "learning_rate": 3.4000000000000005e-06,
+      "loss": 1.0114,
+      "step": 17
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.5493157164665008,
+      "learning_rate": 3.6000000000000003e-06,
+      "loss": 0.9769,
+      "step": 18
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.5312837230851732,
+      "learning_rate": 3.8000000000000005e-06,
+      "loss": 0.9671,
+      "step": 19
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.438348358445216,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.9968,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.6205393019694059,
+      "learning_rate": 4.2000000000000004e-06,
+      "loss": 0.9493,
+      "step": 21
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.8998328673915148,
+      "learning_rate": 4.4e-06,
+      "loss": 0.9248,
+      "step": 22
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.4772584854244761,
+      "learning_rate": 4.600000000000001e-06,
+      "loss": 0.9606,
+      "step": 23
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.6492344751154981,
+      "learning_rate": 4.800000000000001e-06,
+      "loss": 0.9367,
+      "step": 24
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.4492135752311766,
+      "learning_rate": 5e-06,
+      "loss": 0.9681,
+      "step": 25
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.9345029453812452,
+      "learning_rate": 5.2e-06,
+      "loss": 0.9767,
+      "step": 26
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.4303746347311266,
+      "learning_rate": 5.400000000000001e-06,
+      "loss": 1.0061,
+      "step": 27
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.315283188205805,
+      "learning_rate": 5.600000000000001e-06,
+      "loss": 0.8702,
+      "step": 28
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.3343374527628737,
+      "learning_rate": 5.8e-06,
+      "loss": 0.9541,
+      "step": 29
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.4954136888328653,
+      "learning_rate": 6e-06,
+      "loss": 0.9801,
+      "step": 30
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.5130617654914862,
+      "learning_rate": 6.200000000000001e-06,
+      "loss": 0.9215,
+      "step": 31
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.499775954243391,
+      "learning_rate": 6.4000000000000006e-06,
+      "loss": 0.9385,
+      "step": 32
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.7786154279558195,
+      "learning_rate": 6.600000000000001e-06,
+      "loss": 0.929,
+      "step": 33
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.7042536424473338,
+      "learning_rate": 6.800000000000001e-06,
+      "loss": 0.9697,
+      "step": 34
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.2627033504811447,
+      "learning_rate": 7e-06,
+      "loss": 0.9032,
+      "step": 35
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.2805435720774605,
+      "learning_rate": 7.2000000000000005e-06,
+      "loss": 0.8599,
+      "step": 36
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.4840232476266473,
+      "learning_rate": 7.4e-06,
+      "loss": 0.9369,
+      "step": 37
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.4458756643238664,
+      "learning_rate": 7.600000000000001e-06,
+      "loss": 0.8746,
+      "step": 38
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.3994007419423393,
+      "learning_rate": 7.800000000000002e-06,
+      "loss": 0.8775,
+      "step": 39
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.089121632396219,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.8899,
+      "step": 40
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.592387928741599,
+      "learning_rate": 8.2e-06,
+      "loss": 0.8429,
+      "step": 41
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.4183237137083025,
+      "learning_rate": 8.400000000000001e-06,
+      "loss": 0.9994,
+      "step": 42
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.253020204127273,
+      "learning_rate": 8.6e-06,
+      "loss": 0.9008,
+      "step": 43
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.2426930310189384,
+      "learning_rate": 8.8e-06,
+      "loss": 0.8955,
+      "step": 44
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.9803308262802113,
+      "learning_rate": 9e-06,
+      "loss": 0.953,
+      "step": 45
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.2733579391179952,
+      "learning_rate": 9.200000000000002e-06,
+      "loss": 0.9422,
+      "step": 46
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.1807015822635625,
+      "learning_rate": 9.4e-06,
+      "loss": 0.9207,
+      "step": 47
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.2510051340626769,
+      "learning_rate": 9.600000000000001e-06,
+      "loss": 0.8431,
+      "step": 48
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.2611410009201918,
+      "learning_rate": 9.800000000000001e-06,
+      "loss": 0.8627,
+      "step": 49
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.588503959328431,
+      "learning_rate": 1e-05,
+      "loss": 0.937,
+      "step": 50
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.472726233992422,
+      "learning_rate": 1.02e-05,
+      "loss": 0.915,
+      "step": 51
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.2477329691005965,
+      "learning_rate": 1.04e-05,
+      "loss": 0.9082,
+      "step": 52
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.9798238525349643,
+      "learning_rate": 1.0600000000000002e-05,
+      "loss": 0.909,
+      "step": 53
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.4023041981907929,
+      "learning_rate": 1.0800000000000002e-05,
+      "loss": 0.8508,
+      "step": 54
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.38789783036251,
+      "learning_rate": 1.1000000000000001e-05,
+      "loss": 0.962,
+      "step": 55
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.6046775651575935,
+      "learning_rate": 1.1200000000000001e-05,
+      "loss": 0.8629,
+      "step": 56
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.211890766417864,
+      "learning_rate": 1.14e-05,
+      "loss": 0.8936,
+      "step": 57
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.5305064382435207,
+      "learning_rate": 1.16e-05,
+      "loss": 0.9087,
+      "step": 58
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.3858807358761336,
+      "learning_rate": 1.18e-05,
+      "loss": 0.8888,
+      "step": 59
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.4160996675215114,
+      "learning_rate": 1.2e-05,
+      "loss": 0.9709,
+      "step": 60
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.612657711675939,
+      "learning_rate": 1.22e-05,
+      "loss": 0.9553,
+      "step": 61
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.563228725731034,
+      "learning_rate": 1.2400000000000002e-05,
+      "loss": 0.9759,
+      "step": 62
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.549307417934899,
+      "learning_rate": 1.2600000000000001e-05,
+      "loss": 0.9167,
+      "step": 63
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.6084211112800015,
+      "learning_rate": 1.2800000000000001e-05,
+      "loss": 0.8917,
+      "step": 64
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.3125970489166219,
+      "learning_rate": 1.3000000000000001e-05,
+      "loss": 0.83,
+      "step": 65
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.5194204769356594,
+      "learning_rate": 1.3200000000000002e-05,
+      "loss": 0.8531,
+      "step": 66
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.3199910396453893,
+      "learning_rate": 1.3400000000000002e-05,
+      "loss": 0.9043,
+      "step": 67
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.3154362295691828,
+      "learning_rate": 1.3600000000000002e-05,
+      "loss": 0.839,
+      "step": 68
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.4154491863536813,
+      "learning_rate": 1.38e-05,
+      "loss": 0.8477,
+      "step": 69
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.434434391565484,
+      "learning_rate": 1.4e-05,
+      "loss": 0.8615,
+      "step": 70
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.4770964651688123,
+      "learning_rate": 1.4200000000000001e-05,
+      "loss": 0.8751,
+      "step": 71
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.2822341452960506,
+      "learning_rate": 1.4400000000000001e-05,
+      "loss": 0.8426,
+      "step": 72
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.7793171841690287,
+      "learning_rate": 1.46e-05,
+      "loss": 0.8814,
+      "step": 73
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.3354120614602836,
+      "learning_rate": 1.48e-05,
+      "loss": 0.9254,
+      "step": 74
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.3501540008431716,
+      "learning_rate": 1.5000000000000002e-05,
+      "loss": 0.8709,
+      "step": 75
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.266376412273164,
+      "learning_rate": 1.5200000000000002e-05,
+      "loss": 0.8844,
+      "step": 76
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.281004813368152,
+      "learning_rate": 1.54e-05,
+      "loss": 0.9125,
+      "step": 77
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.351899638153987,
+      "learning_rate": 1.5600000000000003e-05,
+      "loss": 0.947,
+      "step": 78
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.230662844786248,
+      "learning_rate": 1.58e-05,
+      "loss": 0.8643,
+      "step": 79
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.248647661067115,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 0.8728,
+      "step": 80
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.4342385839949727,
+      "learning_rate": 1.62e-05,
+      "loss": 0.896,
+      "step": 81
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.2641773466744164,
+      "learning_rate": 1.64e-05,
+      "loss": 0.8514,
+      "step": 82
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.5090046658368796,
+      "learning_rate": 1.66e-05,
+      "loss": 0.8948,
+      "step": 83
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.077918114752302,
+      "learning_rate": 1.6800000000000002e-05,
+      "loss": 0.9363,
+      "step": 84
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.4114006422721,
+      "learning_rate": 1.7e-05,
+      "loss": 0.8664,
+      "step": 85
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.4095249133167944,
+      "learning_rate": 1.72e-05,
+      "loss": 0.8569,
+      "step": 86
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.3912369318560927,
+      "learning_rate": 1.7400000000000003e-05,
+      "loss": 0.9265,
+      "step": 87
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.591904092465942,
+      "learning_rate": 1.76e-05,
+      "loss": 0.8981,
+      "step": 88
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.3641569225962626,
+      "learning_rate": 1.7800000000000002e-05,
+      "loss": 0.9682,
+      "step": 89
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.7178690467888635,
+      "learning_rate": 1.8e-05,
+      "loss": 0.9528,
+      "step": 90
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.2120969466340743,
+      "learning_rate": 1.8200000000000002e-05,
+      "loss": 0.8879,
+      "step": 91
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.4194691633551582,
+      "learning_rate": 1.8400000000000003e-05,
+      "loss": 0.916,
+      "step": 92
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.2552224627285886,
+      "learning_rate": 1.86e-05,
+      "loss": 0.8385,
+      "step": 93
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.5744253728758688,
+      "learning_rate": 1.88e-05,
+      "loss": 0.9253,
+      "step": 94
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.28754739044814,
+      "learning_rate": 1.9e-05,
+      "loss": 0.9304,
+      "step": 95
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.3937275944005247,
+      "learning_rate": 1.9200000000000003e-05,
+      "loss": 0.7894,
+      "step": 96
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.7203298354877872,
+      "learning_rate": 1.94e-05,
+      "loss": 0.8396,
+      "step": 97
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.3582005168065812,
+      "learning_rate": 1.9600000000000002e-05,
+      "loss": 0.8391,
+      "step": 98
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.7784040830385452,
+      "learning_rate": 1.98e-05,
+      "loss": 0.9852,
+      "step": 99
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.8125475091606469,
+      "learning_rate": 2e-05,
+      "loss": 0.9179,
+      "step": 100
+    },
+    {
+      "epoch": 0.07,
+      "eval_loss": 0.8885624408721924,
+      "eval_runtime": 187.1128,
+      "eval_samples_per_second": 216.19,
+      "eval_steps_per_second": 27.026,
+      "step": 100
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.583544229383965,
+      "learning_rate": 1.9999993568531234e-05,
+      "loss": 0.9191,
+      "step": 101
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.312290801935943,
+      "learning_rate": 1.99999742741332e-05,
+      "loss": 0.866,
+      "step": 102
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.5049576941839968,
+      "learning_rate": 1.999994211683072e-05,
+      "loss": 0.8443,
+      "step": 103
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.449632940843496,
+      "learning_rate": 1.9999897096665158e-05,
+      "loss": 0.9668,
+      "step": 104
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.393761122315979,
+      "learning_rate": 1.999983921369442e-05,
+      "loss": 0.9582,
+      "step": 105
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.4243225596755649,
+      "learning_rate": 1.9999768467992964e-05,
+      "loss": 0.8816,
+      "step": 106
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.6681837751811386,
+      "learning_rate": 1.999968485965179e-05,
+      "loss": 1.0217,
+      "step": 107
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.6052796096952295,
+      "learning_rate": 1.999958838877844e-05,
+      "loss": 0.9085,
+      "step": 108
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.3634731163174791,
+      "learning_rate": 1.999947905549701e-05,
+      "loss": 0.8114,
+      "step": 109
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.563483750334544,
+      "learning_rate": 1.9999356859948126e-05,
+      "loss": 0.844,
+      "step": 110
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.309832232231074,
+      "learning_rate": 1.999922180228897e-05,
+      "loss": 0.8668,
+      "step": 111
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.5223471077188282,
+      "learning_rate": 1.999907388269327e-05,
+      "loss": 0.8465,
+      "step": 112
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.3448768039546044,
+      "learning_rate": 1.999891310135129e-05,
+      "loss": 0.96,
+      "step": 113
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.4862383791563054,
+      "learning_rate": 1.9998739458469846e-05,
+      "loss": 0.9533,
+      "step": 114
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.2017931806414803,
+      "learning_rate": 1.999855295427229e-05,
+      "loss": 0.8375,
+      "step": 115
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.2583376428869173,
+      "learning_rate": 1.999835358899852e-05,
+      "loss": 0.805,
+      "step": 116
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.2260058170108115,
+      "learning_rate": 1.9998141362904978e-05,
+      "loss": 0.8608,
+      "step": 117
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.437771413281763,
+      "learning_rate": 1.999791627626466e-05,
+      "loss": 0.886,
+      "step": 118
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.2184693792099177,
+      "learning_rate": 1.999767832936708e-05,
+      "loss": 0.8407,
+      "step": 119
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.306478784713779,
+      "learning_rate": 1.9997427522518315e-05,
+      "loss": 0.8752,
+      "step": 120
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.1534611511399118,
+      "learning_rate": 1.9997163856040974e-05,
+      "loss": 0.932,
+      "step": 121
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3792596997556017,
+      "learning_rate": 1.999688733027421e-05,
+      "loss": 0.923,
+      "step": 122
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.8415988707474704,
+      "learning_rate": 1.9996597945573714e-05,
+      "loss": 0.8956,
+      "step": 123
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3351124075493161,
+      "learning_rate": 1.9996295702311724e-05,
+      "loss": 0.9524,
+      "step": 124
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3103430169447072,
+      "learning_rate": 1.999598060087701e-05,
+      "loss": 0.9206,
+      "step": 125
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3641483437726287,
+      "learning_rate": 1.999565264167489e-05,
+      "loss": 0.8577,
+      "step": 126
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3543902897202786,
+      "learning_rate": 1.9995311825127206e-05,
+      "loss": 0.8494,
+      "step": 127
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.1381528151420175,
+      "learning_rate": 1.9994958151672358e-05,
+      "loss": 0.8701,
+      "step": 128
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.4344875566309996,
+      "learning_rate": 1.9994591621765272e-05,
+      "loss": 0.8702,
+      "step": 129
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.1956730998726899,
+      "learning_rate": 1.9994212235877407e-05,
+      "loss": 0.9015,
+      "step": 130
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.2029518457227866,
+      "learning_rate": 1.9993819994496776e-05,
+      "loss": 0.9442,
+      "step": 131
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.1604274764251628,
+      "learning_rate": 1.9993414898127905e-05,
+      "loss": 0.8042,
+      "step": 132
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3038943540300167,
+      "learning_rate": 1.9992996947291876e-05,
+      "loss": 0.9018,
+      "step": 133
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.1960860187114335,
+      "learning_rate": 1.999256614252629e-05,
+      "loss": 0.9428,
+      "step": 134
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3913155509430453,
+      "learning_rate": 1.999212248438529e-05,
+      "loss": 0.9788,
+      "step": 135
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.3052468844211198,
+      "learning_rate": 1.999166597343955e-05,
+      "loss": 0.9086,
+      "step": 136
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.7145477455732823,
+      "learning_rate": 1.9991196610276283e-05,
+      "loss": 0.9369,
+      "step": 137
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.3431672968207697,
+      "learning_rate": 1.9990714395499222e-05,
+      "loss": 0.9278,
+      "step": 138
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.1560484700856692,
+      "learning_rate": 1.999021932972864e-05,
+      "loss": 0.871,
+      "step": 139
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.137982958645585,
+      "learning_rate": 1.9989711413601332e-05,
+      "loss": 0.7911,
+      "step": 140
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.41243915234061,
+      "learning_rate": 1.9989190647770633e-05,
+      "loss": 1.0343,
+      "step": 141
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.6313054425561822,
+      "learning_rate": 1.99886570329064e-05,
+      "loss": 0.9113,
+      "step": 142
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.1822708557044925,
+      "learning_rate": 1.9988110569695017e-05,
+      "loss": 0.8441,
+      "step": 143
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.1087613022251992,
+      "learning_rate": 1.9987551258839392e-05,
+      "loss": 0.8759,
+      "step": 144
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.1203723621386166,
+      "learning_rate": 1.9986979101058972e-05,
+      "loss": 0.8872,
+      "step": 145
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.5757934934202387,
+      "learning_rate": 1.9986394097089714e-05,
+      "loss": 0.8804,
+      "step": 146
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.1456220385730522,
+      "learning_rate": 1.998579624768411e-05,
+      "loss": 0.7979,
+      "step": 147
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.097480086108459,
+      "learning_rate": 1.998518555361116e-05,
+      "loss": 0.8817,
+      "step": 148
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.2141863942408255,
+      "learning_rate": 1.9984562015656406e-05,
+      "loss": 0.8688,
+      "step": 149
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.165899291397342,
+      "learning_rate": 1.9983925634621894e-05,
+      "loss": 0.8389,
+      "step": 150
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1224242061916174,
+      "learning_rate": 1.9983276411326206e-05,
+      "loss": 0.8456,
+      "step": 151
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1718864519831262,
+      "learning_rate": 1.9982614346604424e-05,
+      "loss": 0.9252,
+      "step": 152
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1633450486139567,
+      "learning_rate": 1.9981939441308166e-05,
+      "loss": 0.9022,
+      "step": 153
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.8675156572488367,
+      "learning_rate": 1.998125169630555e-05,
+      "loss": 0.9143,
+      "step": 154
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.3275288997192118,
+      "learning_rate": 1.9980551112481224e-05,
+      "loss": 0.975,
+      "step": 155
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1366645435061968,
+      "learning_rate": 1.997983769073634e-05,
+      "loss": 0.9066,
+      "step": 156
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.2539240061165282,
+      "learning_rate": 1.9979111431988575e-05,
+      "loss": 0.8686,
+      "step": 157
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1830638013438253,
+      "learning_rate": 1.997837233717211e-05,
+      "loss": 0.9105,
+      "step": 158
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.2289851798766707,
+      "learning_rate": 1.9977620407237627e-05,
+      "loss": 0.97,
+      "step": 159
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1894635911968074,
+      "learning_rate": 1.997685564315234e-05,
+      "loss": 0.941,
+      "step": 160
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.439765259122199,
+      "learning_rate": 1.997607804589996e-05,
+      "loss": 1.0221,
+      "step": 161
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.1495993551137769,
+      "learning_rate": 1.9975287616480702e-05,
+      "loss": 0.8823,
+      "step": 162
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.2383675419211015,
+      "learning_rate": 1.9974484355911287e-05,
+      "loss": 0.8684,
+      "step": 163
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.219442395631165,
+      "learning_rate": 1.997366826522495e-05,
+      "loss": 0.9084,
+      "step": 164
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.2050137099782028,
+      "learning_rate": 1.9972839345471423e-05,
+      "loss": 0.8461,
+      "step": 165
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.1610243840319072,
+      "learning_rate": 1.997199759771694e-05,
+      "loss": 0.843,
+      "step": 166
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.1154404343775486,
+      "learning_rate": 1.997114302304423e-05,
+      "loss": 0.8247,
+      "step": 167
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.194607031155807,
+      "learning_rate": 1.9970275622552536e-05,
+      "loss": 0.8432,
+      "step": 168
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.4022970990428032,
+      "learning_rate": 1.9969395397357584e-05,
+      "loss": 0.9016,
+      "step": 169
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.1858220432102802,
+      "learning_rate": 1.99685023485916e-05,
+      "loss": 0.9437,
+      "step": 170
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.1155321667076121,
+      "learning_rate": 1.9967596477403318e-05,
+      "loss": 0.8602,
+      "step": 171
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.2905373990572484,
+      "learning_rate": 1.996667778495794e-05,
+      "loss": 0.9333,
+      "step": 172
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.1931787552579112,
+      "learning_rate": 1.9965746272437187e-05,
+      "loss": 0.7826,
+      "step": 173
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.2630219016898117,
+      "learning_rate": 1.996480194103925e-05,
+      "loss": 0.8185,
+      "step": 174
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.204987733919362,
+      "learning_rate": 1.9963844791978818e-05,
+      "loss": 0.8742,
+      "step": 175
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.2437041912683955,
+      "learning_rate": 1.996287482648707e-05,
+      "loss": 0.9834,
+      "step": 176
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.254870155842108,
+      "learning_rate": 1.9961892045811663e-05,
+      "loss": 0.8836,
+      "step": 177
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.2090034933034504,
+      "learning_rate": 1.9960896451216738e-05,
+      "loss": 0.8667,
+      "step": 178
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.1598314715618245,
+      "learning_rate": 1.9959888043982927e-05,
+      "loss": 0.8849,
+      "step": 179
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.252477055176123,
+      "learning_rate": 1.995886682540734e-05,
+      "loss": 0.8851,
+      "step": 180
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.1256837247964426,
+      "learning_rate": 1.995783279680356e-05,
+      "loss": 0.8993,
+      "step": 181
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.2572839567534193,
+      "learning_rate": 1.995678595950165e-05,
+      "loss": 0.8838,
+      "step": 182
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.2788480841418937,
+      "learning_rate": 1.9955726314848154e-05,
+      "loss": 0.9657,
+      "step": 183
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.4351399334715638,
+      "learning_rate": 1.9954653864206088e-05,
+      "loss": 0.8965,
+      "step": 184
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.148517952301927,
+      "learning_rate": 1.9953568608954934e-05,
+      "loss": 0.9614,
+      "step": 185
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.0899591202732373,
+      "learning_rate": 1.995247055049065e-05,
+      "loss": 1.0451,
+      "step": 186
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.1956254586029718,
+      "learning_rate": 1.9951359690225662e-05,
+      "loss": 0.8135,
+      "step": 187
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.1207108959665357,
+      "learning_rate": 1.9950236029588863e-05,
+      "loss": 0.8822,
+      "step": 188
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.1899112755126982,
+      "learning_rate": 1.9949099570025613e-05,
+      "loss": 0.869,
+      "step": 189
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.2114377382527095,
+      "learning_rate": 1.994795031299773e-05,
+      "loss": 0.8621,
+      "step": 190
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.1290548904860946,
+      "learning_rate": 1.9946788259983493e-05,
+      "loss": 0.9085,
+      "step": 191
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.101382636702055,
+      "learning_rate": 1.9945613412477652e-05,
+      "loss": 0.8717,
+      "step": 192
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.2652298684318162,
+      "learning_rate": 1.99444257719914e-05,
+      "loss": 0.914,
+      "step": 193
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.1661768680774713,
+      "learning_rate": 1.9943225340052392e-05,
+      "loss": 1.0426,
+      "step": 194
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.3501980135490719,
+      "learning_rate": 1.9942012118204738e-05,
+      "loss": 0.8917,
+      "step": 195
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.2049018991454707,
+      "learning_rate": 1.9940786108009e-05,
+      "loss": 0.8631,
+      "step": 196
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.215922883123164,
+      "learning_rate": 1.9939547311042177e-05,
+      "loss": 0.9292,
+      "step": 197
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.1267422886076772,
+      "learning_rate": 1.9938295728897735e-05,
+      "loss": 0.8138,
+      "step": 198
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.1497998538174055,
+      "learning_rate": 1.9937031363185573e-05,
+      "loss": 0.8412,
+      "step": 199
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.3726576421968895,
+      "learning_rate": 1.993575421553204e-05,
+      "loss": 1.0123,
+      "step": 200
+    },
+    {
+      "epoch": 0.14,
+      "eval_loss": 0.8822317719459534,
+      "eval_runtime": 187.1914,
+      "eval_samples_per_second": 216.1,
+      "eval_steps_per_second": 27.015,
+      "step": 200
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.2637805473747161,
+      "learning_rate": 1.9934464287579922e-05,
+      "loss": 0.9358,
+      "step": 201
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.2221016247650263,
+      "learning_rate": 1.9933161580988444e-05,
+      "loss": 0.8732,
+      "step": 202
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.2362316473986823,
+      "learning_rate": 1.9931846097433265e-05,
+      "loss": 0.8965,
+      "step": 203
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.3892342291259105,
+      "learning_rate": 1.9930517838606492e-05,
+      "loss": 0.8785,
+      "step": 204
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.2568068018376466,
+      "learning_rate": 1.992917680621665e-05,
+      "loss": 0.8879,
+      "step": 205
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.2887830429527976,
+      "learning_rate": 1.9927823001988703e-05,
+      "loss": 0.8982,
+      "step": 206
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.0952878042078324,
+      "learning_rate": 1.9926456427664038e-05,
+      "loss": 0.8321,
+      "step": 207
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.1488105022621649,
+      "learning_rate": 1.9925077085000475e-05,
+      "loss": 0.8938,
+      "step": 208
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.2001692579551484,
+      "learning_rate": 1.992368497577225e-05,
+      "loss": 0.9604,
+      "step": 209
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.5520591983316103,
+      "learning_rate": 1.992228010177003e-05,
+      "loss": 0.8965,
+      "step": 210
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.1538845325628375,
+      "learning_rate": 1.992086246480089e-05,
+      "loss": 0.9079,
+      "step": 211
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.3086118735118826,
+      "learning_rate": 1.991943206668833e-05,
+      "loss": 0.8822,
+      "step": 212
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.3066154322169896,
+      "learning_rate": 1.9917988909272263e-05,
+      "loss": 0.9422,
+      "step": 213
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.3788975402648815,
+      "learning_rate": 1.9916532994409012e-05,
+      "loss": 0.8661,
+      "step": 214
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.157353475407855,
+      "learning_rate": 1.991506432397131e-05,
+      "loss": 0.8708,
+      "step": 215
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.0668946879406427,
+      "learning_rate": 1.99135828998483e-05,
+      "loss": 0.8281,
+      "step": 216
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.0586925383296883,
+      "learning_rate": 1.9912088723945527e-05,
+      "loss": 0.8748,
+      "step": 217
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.2243244443889567,
+      "learning_rate": 1.9910581798184947e-05,
+      "loss": 0.8876,
+      "step": 218
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.153241330865434,
+      "learning_rate": 1.99090621245049e-05,
+      "loss": 1.0029,
+      "step": 219
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.3289456136734883,
+      "learning_rate": 1.990752970486014e-05,
+      "loss": 0.9331,
+      "step": 220
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.1658780585501232,
+      "learning_rate": 1.99059845412218e-05,
+      "loss": 0.8539,
+      "step": 221
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.0841474719145574,
+      "learning_rate": 1.9904426635577426e-05,
+      "loss": 0.7989,
+      "step": 222
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2100420613683809,
+      "learning_rate": 1.990285598993093e-05,
+      "loss": 0.9099,
+      "step": 223
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2327481765852046,
+      "learning_rate": 1.9901272606302635e-05,
+      "loss": 0.8782,
+      "step": 224
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.1374571970583345,
+      "learning_rate": 1.9899676486729227e-05,
+      "loss": 0.9215,
+      "step": 225
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2454011716677853,
+      "learning_rate": 1.9898067633263795e-05,
+      "loss": 0.9805,
+      "step": 226
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.051858806753117,
+      "learning_rate": 1.989644604797579e-05,
+      "loss": 0.8915,
+      "step": 227
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.6709690832913764,
+      "learning_rate": 1.989481173295105e-05,
+      "loss": 0.9316,
+      "step": 228
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2921955310126267,
+      "learning_rate": 1.9893164690291783e-05,
+      "loss": 0.911,
+      "step": 229
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.4306151791171935,
+      "learning_rate": 1.9891504922116572e-05,
+      "loss": 0.8217,
+      "step": 230
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.4490313892439401,
+      "learning_rate": 1.988983243056036e-05,
+      "loss": 0.9171,
+      "step": 231
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2261731033069914,
+      "learning_rate": 1.988814721777447e-05,
+      "loss": 0.857,
+      "step": 232
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.1936230768093887,
+      "learning_rate": 1.988644928592658e-05,
+      "loss": 0.9604,
+      "step": 233
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2805860141107388,
+      "learning_rate": 1.988473863720072e-05,
+      "loss": 0.8911,
+      "step": 234
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2391969410884403,
+      "learning_rate": 1.9883015273797302e-05,
+      "loss": 0.8648,
+      "step": 235
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.1576939443779746,
+      "learning_rate": 1.988127919793306e-05,
+      "loss": 0.8933,
+      "step": 236
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.3904763872990524,
+      "learning_rate": 1.9879530411841114e-05,
+      "loss": 0.8738,
+      "step": 237
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.6162138215699062,
+      "learning_rate": 1.987776891777091e-05,
+      "loss": 0.8614,
+      "step": 238
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.3562904123393382,
+      "learning_rate": 1.9875994717988238e-05,
+      "loss": 0.8795,
+      "step": 239
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.5730263292883433,
+      "learning_rate": 1.9874207814775252e-05,
+      "loss": 0.9134,
+      "step": 240
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.7831810147386489,
+      "learning_rate": 1.9872408210430433e-05,
+      "loss": 0.8516,
+      "step": 241
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.1832426844117234,
+      "learning_rate": 1.9870595907268596e-05,
+      "loss": 0.8199,
+      "step": 242
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.350941258232983,
+      "learning_rate": 1.98687709076209e-05,
+      "loss": 0.9702,
+      "step": 243
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.2867689498331254,
+      "learning_rate": 1.9866933213834824e-05,
+      "loss": 0.8643,
+      "step": 244
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.4677460519503693,
+      "learning_rate": 1.986508282827419e-05,
+      "loss": 0.8778,
+      "step": 245
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.4363310224409451,
+      "learning_rate": 1.9863219753319135e-05,
+      "loss": 0.8522,
+      "step": 246
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.1699677293775574,
+      "learning_rate": 1.9861343991366115e-05,
+      "loss": 0.802,
+      "step": 247
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.2924807238317337,
+      "learning_rate": 1.9859455544827918e-05,
+      "loss": 0.9509,
+      "step": 248
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.2521477373576704,
+      "learning_rate": 1.9857554416133638e-05,
+      "loss": 0.8525,
+      "step": 249
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.1517606654171464,
+      "learning_rate": 1.9855640607728684e-05,
+      "loss": 0.8671,
+      "step": 250
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.1846979830932318,
+      "learning_rate": 1.985371412207478e-05,
+      "loss": 0.8872,
+      "step": 251
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.0494510402617383,
+      "learning_rate": 1.9851774961649947e-05,
+      "loss": 0.8404,
+      "step": 252
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.2000573806870802,
+      "learning_rate": 1.984982312894852e-05,
+      "loss": 0.858,
+      "step": 253
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.2135277830431934,
+      "learning_rate": 1.984785862648112e-05,
+      "loss": 0.8937,
+      "step": 254
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.1101162005595753,
+      "learning_rate": 1.984588145677469e-05,
+      "loss": 0.8475,
+      "step": 255
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.2142650007241018,
+      "learning_rate": 1.9843891622372434e-05,
+      "loss": 0.9388,
+      "step": 256
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.1822666855469028,
+      "learning_rate": 1.9841889125833877e-05,
+      "loss": 0.8937,
+      "step": 257
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.5588956230747333,
+      "learning_rate": 1.9839873969734813e-05,
+      "loss": 0.8999,
+      "step": 258
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.2813736208840643,
+      "learning_rate": 1.983784615666732e-05,
+      "loss": 1.0269,
+      "step": 259
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.0937396851014907,
+      "learning_rate": 1.983580568923977e-05,
+      "loss": 0.8552,
+      "step": 260
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.0789428924402067,
+      "learning_rate": 1.9833752570076794e-05,
+      "loss": 0.8626,
+      "step": 261
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.0716959865947704,
+      "learning_rate": 1.983168680181931e-05,
+      "loss": 0.9014,
+      "step": 262
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.5586891426715053,
+      "learning_rate": 1.9829608387124512e-05,
+      "loss": 0.8257,
+      "step": 263
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.159299773943287,
+      "learning_rate": 1.9827517328665835e-05,
+      "loss": 0.9158,
+      "step": 264
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.614928320766928,
+      "learning_rate": 1.9825413629133007e-05,
+      "loss": 0.905,
+      "step": 265
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.2759728656235267,
+      "learning_rate": 1.9823297291232003e-05,
+      "loss": 0.8884,
+      "step": 266
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.130197334786531,
+      "learning_rate": 1.9821168317685048e-05,
+      "loss": 0.9401,
+      "step": 267
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.1071953972918522,
+      "learning_rate": 1.981902671123063e-05,
+      "loss": 0.8998,
+      "step": 268
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.4143214889252662,
+      "learning_rate": 1.981687247462349e-05,
+      "loss": 1.0354,
+      "step": 269
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.1936018368917338,
+      "learning_rate": 1.9814705610634602e-05,
+      "loss": 0.8934,
+      "step": 270
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.9588028448015303,
+      "learning_rate": 1.981252612205119e-05,
+      "loss": 0.7949,
+      "step": 271
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.0367090915141624,
+      "learning_rate": 1.981033401167672e-05,
+      "loss": 0.85,
+      "step": 272
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.1632134055825956,
+      "learning_rate": 1.9808129282330895e-05,
+      "loss": 0.8961,
+      "step": 273
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.1644937246368534,
+      "learning_rate": 1.980591193684963e-05,
+      "loss": 0.9055,
+      "step": 274
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.1695602344439167,
+      "learning_rate": 1.9803681978085095e-05,
+      "loss": 0.9705,
+      "step": 275
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.0729503600381038,
+      "learning_rate": 1.9801439408905663e-05,
+      "loss": 0.8347,
+      "step": 276
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.0999879962808221,
+      "learning_rate": 1.9799184232195947e-05,
+      "loss": 0.9158,
+      "step": 277
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.0865411615717533,
+      "learning_rate": 1.979691645085676e-05,
+      "loss": 0.8844,
+      "step": 278
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.012147198379429,
+      "learning_rate": 1.9794636067805137e-05,
+      "loss": 0.9223,
+      "step": 279
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.412178875932858,
+      "learning_rate": 1.9792343085974316e-05,
+      "loss": 0.9005,
+      "step": 280
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.1197488002533464,
+      "learning_rate": 1.979003750831375e-05,
+      "loss": 0.8507,
+      "step": 281
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.2057779133550202,
+      "learning_rate": 1.978771933778909e-05,
+      "loss": 0.9847,
+      "step": 282
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.1213946217031967,
+      "learning_rate": 1.9785388577382183e-05,
+      "loss": 0.824,
+      "step": 283
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.1036782471126692,
+      "learning_rate": 1.978304523009107e-05,
+      "loss": 0.8095,
+      "step": 284
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.0945851242489233,
+      "learning_rate": 1.9780689298929984e-05,
+      "loss": 0.8134,
+      "step": 285
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.3077751312491577,
+      "learning_rate": 1.9778320786929347e-05,
+      "loss": 0.9936,
+      "step": 286
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.2408368443934399,
+      "learning_rate": 1.9775939697135758e-05,
+      "loss": 0.9111,
+      "step": 287
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.0957840590624228,
+      "learning_rate": 1.9773546032611997e-05,
+      "loss": 0.8965,
+      "step": 288
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.0147868519584988,
+      "learning_rate": 1.9771139796437028e-05,
+      "loss": 0.7851,
+      "step": 289
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.0973117222342668,
+      "learning_rate": 1.976872099170597e-05,
+      "loss": 0.8744,
+      "step": 290
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.3287791694807247,
+      "learning_rate": 1.9766289621530118e-05,
+      "loss": 0.8534,
+      "step": 291
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.133246313776563,
+      "learning_rate": 1.9763845689036927e-05,
+      "loss": 0.8312,
+      "step": 292
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.1419361652462952,
+      "learning_rate": 1.9761389197370014e-05,
+      "loss": 0.8366,
+      "step": 293
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.0821551951711113,
+      "learning_rate": 1.9758920149689146e-05,
+      "loss": 0.8487,
+      "step": 294
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.1556389890148893,
+      "learning_rate": 1.975643854917025e-05,
+      "loss": 0.8578,
+      "step": 295
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.2067794152264784,
+      "learning_rate": 1.9753944399005388e-05,
+      "loss": 0.8771,
+      "step": 296
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.1417102044118401,
+      "learning_rate": 1.9751437702402773e-05,
+      "loss": 0.9566,
+      "step": 297
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.7660938523689576,
+      "learning_rate": 1.9748918462586752e-05,
+      "loss": 0.876,
+      "step": 298
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.1220879011792553,
+      "learning_rate": 1.9746386682797802e-05,
+      "loss": 0.8612,
+      "step": 299
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.0816131747151747,
+      "learning_rate": 1.9743842366292544e-05,
+      "loss": 0.9106,
+      "step": 300
+    },
+    {
+      "epoch": 0.21,
+      "eval_loss": 0.8701280951499939,
+      "eval_runtime": 187.4205,
+      "eval_samples_per_second": 215.836,
+      "eval_steps_per_second": 26.982,
+      "step": 300
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.0299750746890193,
+      "learning_rate": 1.974128551634371e-05,
+      "loss": 0.8134,
+      "step": 301
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.0355435348681352,
+      "learning_rate": 1.9738716136240166e-05,
+      "loss": 0.8523,
+      "step": 302
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.2446494879448189,
+      "learning_rate": 1.9736134229286884e-05,
+      "loss": 0.9613,
+      "step": 303
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.2659564391872702,
+      "learning_rate": 1.9733539798804958e-05,
+      "loss": 0.8897,
+      "step": 304
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.2668653568603956,
+      "learning_rate": 1.9730932848131585e-05,
+      "loss": 0.8957,
+      "step": 305
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.1810413523703902,
+      "learning_rate": 1.9728313380620073e-05,
+      "loss": 0.8298,
+      "step": 306
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.1504498545031272,
+      "learning_rate": 1.9725681399639826e-05,
+      "loss": 0.9032,
+      "step": 307
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.1109923815534024,
+      "learning_rate": 1.9723036908576344e-05,
+      "loss": 0.8851,
+      "step": 308
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.1153797340529337,
+      "learning_rate": 1.972037991083122e-05,
+      "loss": 0.9025,
+      "step": 309
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.0547112750314187,
+      "learning_rate": 1.971771040982213e-05,
+      "loss": 0.867,
+      "step": 310
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.9896043321667425,
+      "learning_rate": 1.971502840898284e-05,
+      "loss": 0.8251,
+      "step": 311
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.0919583168522662,
+      "learning_rate": 1.9712333911763186e-05,
+      "loss": 0.9165,
+      "step": 312
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.2904601705665253,
+      "learning_rate": 1.970962692162909e-05,
+      "loss": 0.8716,
+      "step": 313
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.0305914092123785,
+      "learning_rate": 1.9706907442062536e-05,
+      "loss": 0.8143,
+      "step": 314
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.1005551182409805,
+      "learning_rate": 1.9704175476561568e-05,
+      "loss": 0.8313,
+      "step": 315
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.1201052731592884,
+      "learning_rate": 1.97014310286403e-05,
+      "loss": 0.8986,
+      "step": 316
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.046546723321121,
+      "learning_rate": 1.9698674101828895e-05,
+      "loss": 0.9152,
+      "step": 317
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.018976681773729,
+      "learning_rate": 1.969590469967358e-05,
+      "loss": 0.8272,
+      "step": 318
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.9825109312046809,
+      "learning_rate": 1.9693122825736607e-05,
+      "loss": 0.8734,
+      "step": 319
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.027432613849138,
+      "learning_rate": 1.9690328483596287e-05,
+      "loss": 0.8852,
+      "step": 320
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.1532549677069504,
+      "learning_rate": 1.9687521676846974e-05,
+      "loss": 0.8848,
+      "step": 321
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.1500913753153383,
+      "learning_rate": 1.968470240909903e-05,
+      "loss": 0.9482,
+      "step": 322
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.0861728022209993,
+      "learning_rate": 1.9681870683978878e-05,
+      "loss": 0.885,
+      "step": 323
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.1508385225727733,
+      "learning_rate": 1.967902650512894e-05,
+      "loss": 0.8525,
+      "step": 324
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.1091611254738105,
+      "learning_rate": 1.9676169876207664e-05,
+      "loss": 0.9763,
+      "step": 325
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.012395214833452,
+      "learning_rate": 1.9673300800889513e-05,
+      "loss": 1.0059,
+      "step": 326
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.0709485032246597,
+      "learning_rate": 1.9670419282864968e-05,
+      "loss": 0.8361,
+      "step": 327
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.067182605716564,
+      "learning_rate": 1.96675253258405e-05,
+      "loss": 0.8886,
+      "step": 328
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.0898056998348664,
+      "learning_rate": 1.9664618933538593e-05,
+      "loss": 0.8371,
+      "step": 329
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.0874145605191532,
+      "learning_rate": 1.9661700109697718e-05,
+      "loss": 0.9242,
+      "step": 330
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.2537141356402897,
+      "learning_rate": 1.9658768858072346e-05,
+      "loss": 0.9072,
+      "step": 331
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.1244052821739037,
+      "learning_rate": 1.965582518243292e-05,
+      "loss": 0.9079,
+      "step": 332
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.5132621278229172,
+      "learning_rate": 1.9652869086565874e-05,
+      "loss": 0.8279,
+      "step": 333
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.1445681558432512,
+      "learning_rate": 1.9649900574273616e-05,
+      "loss": 0.8482,
+      "step": 334
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.801049653373737,
+      "learning_rate": 1.9646919649374524e-05,
+      "loss": 0.8787,
+      "step": 335
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.1852812751336075,
+      "learning_rate": 1.9643926315702947e-05,
+      "loss": 0.9122,
+      "step": 336
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.1555348167729393,
+      "learning_rate": 1.9640920577109183e-05,
+      "loss": 0.8472,
+      "step": 337
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.4430203433228617,
+      "learning_rate": 1.9637902437459506e-05,
+      "loss": 0.9194,
+      "step": 338
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.0544540038634749,
+      "learning_rate": 1.9634871900636124e-05,
+      "loss": 0.9071,
+      "step": 339
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.1435050397743178,
+      "learning_rate": 1.9631828970537196e-05,
+      "loss": 0.8863,
+      "step": 340
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.268170090315426,
+      "learning_rate": 1.9628773651076825e-05,
+      "loss": 1.0548,
+      "step": 341
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.0486670138134782,
+      "learning_rate": 1.9625705946185053e-05,
+      "loss": 0.8845,
+      "step": 342
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.1109602735025481,
+      "learning_rate": 1.9622625859807843e-05,
+      "loss": 0.8888,
+      "step": 343
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.0327094310140685,
+      "learning_rate": 1.96195333959071e-05,
+      "loss": 0.8448,
+      "step": 344
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 12.81851288731012,
+      "learning_rate": 1.9616428558460634e-05,
+      "loss": 0.8091,
+      "step": 345
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.3904475342636298,
+      "learning_rate": 1.961331135146218e-05,
+      "loss": 0.8578,
+      "step": 346
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.0316055568361129,
+      "learning_rate": 1.961018177892138e-05,
+      "loss": 0.8227,
+      "step": 347
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.301685178122554,
+      "learning_rate": 1.9607039844863788e-05,
+      "loss": 0.9027,
+      "step": 348
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.1850865906075192,
+      "learning_rate": 1.960388555333085e-05,
+      "loss": 0.8745,
+      "step": 349
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.171726269820577,
+      "learning_rate": 1.960071890837991e-05,
+      "loss": 0.8424,
+      "step": 350
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 1.2171194717004472,
+      "learning_rate": 1.959753991408421e-05,
+      "loss": 0.9305,
+      "step": 351
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.2097775475792434,
+      "learning_rate": 1.9594348574532868e-05,
+      "loss": 0.7946,
+      "step": 352
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.1659533100233976,
+      "learning_rate": 1.9591144893830884e-05,
+      "loss": 0.9041,
+      "step": 353
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.361702782618829,
+      "learning_rate": 1.9587928876099126e-05,
+      "loss": 0.8343,
+      "step": 354
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.2483023246254354,
+      "learning_rate": 1.9584700525474347e-05,
+      "loss": 0.8595,
+      "step": 355
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.1656310027795045,
+      "learning_rate": 1.9581459846109153e-05,
+      "loss": 0.8499,
+      "step": 356
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.2008528979815898,
+      "learning_rate": 1.9578206842172003e-05,
+      "loss": 0.9706,
+      "step": 357
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.1352069526109516,
+      "learning_rate": 1.9574941517847223e-05,
+      "loss": 0.8912,
+      "step": 358
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.1495509789253435,
+      "learning_rate": 1.9571663877334975e-05,
+      "loss": 0.8749,
+      "step": 359
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.1074439603505601,
+      "learning_rate": 1.9568373924851267e-05,
+      "loss": 0.9433,
+      "step": 360
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.2169160927046383,
+      "learning_rate": 1.9565071664627947e-05,
+      "loss": 0.9967,
+      "step": 361
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.1445655760646407,
+      "learning_rate": 1.9561757100912692e-05,
+      "loss": 0.9123,
+      "step": 362
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.034825339597779,
+      "learning_rate": 1.9558430237968998e-05,
+      "loss": 0.8477,
+      "step": 363
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.1684005466502694,
+      "learning_rate": 1.95550910800762e-05,
+      "loss": 0.9228,
+      "step": 364
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.0611745934297807,
+      "learning_rate": 1.9551739631529427e-05,
+      "loss": 0.8413,
+      "step": 365
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.164478242612286,
+      "learning_rate": 1.9548375896639627e-05,
+      "loss": 0.8819,
+      "step": 366
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.069365191756458,
+      "learning_rate": 1.9544999879733556e-05,
+      "loss": 0.9306,
+      "step": 367
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.2354599047635832,
+      "learning_rate": 1.9541611585153758e-05,
+      "loss": 0.846,
+      "step": 368
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.051996239556798,
+      "learning_rate": 1.9538211017258573e-05,
+      "loss": 0.8362,
+      "step": 369
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.0239770659963459,
+      "learning_rate": 1.953479818042214e-05,
+      "loss": 0.8912,
+      "step": 370
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.0915734894988451,
+      "learning_rate": 1.953137307903436e-05,
+      "loss": 0.8258,
+      "step": 371
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.4240207885170275,
+      "learning_rate": 1.952793571750093e-05,
+      "loss": 0.9152,
+      "step": 372
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.1917328080665825,
+      "learning_rate": 1.9524486100243297e-05,
+      "loss": 0.8451,
+      "step": 373
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.5484400558582907,
+      "learning_rate": 1.9521024231698684e-05,
+      "loss": 0.8355,
+      "step": 374
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.1437245836456504,
+      "learning_rate": 1.9517550116320074e-05,
+      "loss": 0.963,
+      "step": 375
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.1576928591561433,
+      "learning_rate": 1.9514063758576198e-05,
+      "loss": 0.8982,
+      "step": 376
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.2350762795576822,
+      "learning_rate": 1.9510565162951538e-05,
+      "loss": 0.8619,
+      "step": 377
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.013440484146166,
+      "learning_rate": 1.9507054333946313e-05,
+      "loss": 0.8565,
+      "step": 378
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.1573947595135938,
+      "learning_rate": 1.9503531276076483e-05,
+      "loss": 0.883,
+      "step": 379
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.0506730903652184,
+      "learning_rate": 1.949999599387373e-05,
+      "loss": 0.8537,
+      "step": 380
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.2671573201994177,
+      "learning_rate": 1.9496448491885473e-05,
+      "loss": 0.9342,
+      "step": 381
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.1837608740018646,
+      "learning_rate": 1.9492888774674837e-05,
+      "loss": 0.8467,
+      "step": 382
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.12550000000519,
+      "learning_rate": 1.9489316846820668e-05,
+      "loss": 0.8109,
+      "step": 383
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.0726587171258777,
+      "learning_rate": 1.9485732712917507e-05,
+      "loss": 0.7991,
+      "step": 384
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.2691153809839144,
+      "learning_rate": 1.9482136377575612e-05,
+      "loss": 1.0423,
+      "step": 385
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.0637626583859996,
+      "learning_rate": 1.9478527845420915e-05,
+      "loss": 0.8993,
+      "step": 386
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.0884178299273755,
+      "learning_rate": 1.9474907121095064e-05,
+      "loss": 0.8942,
+      "step": 387
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.0391015249601647,
+      "learning_rate": 1.947127420925536e-05,
+      "loss": 0.8135,
+      "step": 388
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.1516781892264725,
+      "learning_rate": 1.9467629114574805e-05,
+      "loss": 0.8714,
+      "step": 389
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.1314783254435155,
+      "learning_rate": 1.9463971841742057e-05,
+      "loss": 0.9036,
+      "step": 390
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.025454709514132,
+      "learning_rate": 1.9460302395461437e-05,
+      "loss": 0.8731,
+      "step": 391
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.2270327905246061,
+      "learning_rate": 1.9456620780452943e-05,
+      "loss": 0.8698,
+      "step": 392
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.0616296964610366,
+      "learning_rate": 1.9452927001452203e-05,
+      "loss": 0.7885,
+      "step": 393
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.151475949060371,
+      "learning_rate": 1.9449221063210512e-05,
+      "loss": 0.8256,
+      "step": 394
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.423361234847695,
+      "learning_rate": 1.9445502970494787e-05,
+      "loss": 0.8914,
+      "step": 395
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.1433125351711972,
+      "learning_rate": 1.944177272808759e-05,
+      "loss": 0.8687,
+      "step": 396
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.184275686610942,
+      "learning_rate": 1.943803034078711e-05,
+      "loss": 0.7931,
+      "step": 397
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.0661356068578263,
+      "learning_rate": 1.943427581340715e-05,
+      "loss": 0.7678,
+      "step": 398
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.2228382448553605,
+      "learning_rate": 1.9430509150777144e-05,
+      "loss": 0.8018,
+      "step": 399
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.88748184451058,
+      "learning_rate": 1.9426730357742123e-05,
+      "loss": 0.8992,
+      "step": 400
+    },
+    {
+      "epoch": 0.28,
+      "eval_loss": 0.8636618852615356,
+      "eval_runtime": 187.3213,
+      "eval_samples_per_second": 215.95,
+      "eval_steps_per_second": 26.996,
+      "step": 400
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.2624347636274722,
+      "learning_rate": 1.9422939439162724e-05,
+      "loss": 0.9155,
+      "step": 401
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.1698553557184548,
+      "learning_rate": 1.941913639991518e-05,
+      "loss": 0.86,
+      "step": 402
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.2896541603131255,
+      "learning_rate": 1.9415321244891324e-05,
+      "loss": 0.8662,
+      "step": 403
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.0328968599567645,
+      "learning_rate": 1.9411493978998556e-05,
+      "loss": 0.8603,
+      "step": 404
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.1164191505312102,
+      "learning_rate": 1.9407654607159872e-05,
+      "loss": 0.8241,
+      "step": 405
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.0333308388183566,
+      "learning_rate": 1.940380313431383e-05,
+      "loss": 0.9102,
+      "step": 406
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.2102188125348146,
+      "learning_rate": 1.9399939565414554e-05,
+      "loss": 0.8601,
+      "step": 407
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.0252791341841436,
+      "learning_rate": 1.939606390543173e-05,
+      "loss": 0.9156,
+      "step": 408
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.0237005282287415,
+      "learning_rate": 1.9392176159350592e-05,
+      "loss": 0.8721,
+      "step": 409
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.179791811608601,
+      "learning_rate": 1.938827633217193e-05,
+      "loss": 0.8506,
+      "step": 410
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.2083149029506115,
+      "learning_rate": 1.938436442891206e-05,
+      "loss": 0.9396,
+      "step": 411
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.304928466464715,
+      "learning_rate": 1.938044045460284e-05,
+      "loss": 0.848,
+      "step": 412
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.245493747578827,
+      "learning_rate": 1.9376504414291662e-05,
+      "loss": 1.016,
+      "step": 413
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.435385045917715,
+      "learning_rate": 1.9372556313041423e-05,
+      "loss": 1.0092,
+      "step": 414
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.0903396597636805,
+      "learning_rate": 1.9368596155930538e-05,
+      "loss": 0.7966,
+      "step": 415
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.1738401247123667,
+      "learning_rate": 1.936462394805294e-05,
+      "loss": 0.8328,
+      "step": 416
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.0633958378072337,
+      "learning_rate": 1.9360639694518047e-05,
+      "loss": 0.8777,
+      "step": 417
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.0396714883459242,
+      "learning_rate": 1.935664340045079e-05,
+      "loss": 0.7648,
+      "step": 418
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.0106668472141556,
+      "learning_rate": 1.9352635070991567e-05,
+      "loss": 0.8712,
+      "step": 419
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.1381217667885855,
+      "learning_rate": 1.934861471129627e-05,
+      "loss": 0.8539,
+      "step": 420
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.1151430876376536,
+      "learning_rate": 1.9344582326536265e-05,
+      "loss": 0.8892,
+      "step": 421
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.371300072001641,
+      "learning_rate": 1.934053792189838e-05,
+      "loss": 0.848,
+      "step": 422
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.068226316452399,
+      "learning_rate": 1.9336481502584914e-05,
+      "loss": 0.858,
+      "step": 423
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.0487313154852826,
+      "learning_rate": 1.9332413073813606e-05,
+      "loss": 0.8205,
+      "step": 424
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.0254282471589573,
+      "learning_rate": 1.9328332640817657e-05,
+      "loss": 0.86,
+      "step": 425
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.0351254744835727,
+      "learning_rate": 1.9324240208845694e-05,
+      "loss": 0.9067,
+      "step": 426
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.0150706385574908,
+      "learning_rate": 1.9320135783161795e-05,
+      "loss": 0.7885,
+      "step": 427
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.0164480002903071,
+      "learning_rate": 1.9316019369045453e-05,
+      "loss": 0.8156,
+      "step": 428
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.0441180688201646,
+      "learning_rate": 1.9311890971791586e-05,
+      "loss": 0.8825,
+      "step": 429
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.044772972990732,
+      "learning_rate": 1.930775059671053e-05,
+      "loss": 0.7897,
+      "step": 430
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.9632658012924297,
+      "learning_rate": 1.9303598249128016e-05,
+      "loss": 0.7772,
+      "step": 431
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.0341143803884538,
+      "learning_rate": 1.929943393438519e-05,
+      "loss": 0.9062,
+      "step": 432
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.094263199803778,
+      "learning_rate": 1.9295257657838578e-05,
+      "loss": 0.859,
+      "step": 433
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.643732102287827,
+      "learning_rate": 1.9291069424860103e-05,
+      "loss": 0.9486,
+      "step": 434
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.1173987376473988,
+      "learning_rate": 1.928686924083706e-05,
+      "loss": 0.8217,
+      "step": 435
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.029920328225798,
+      "learning_rate": 1.9282657111172122e-05,
+      "loss": 0.898,
+      "step": 436
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.095170152800852,
+      "learning_rate": 1.927843304128332e-05,
+      "loss": 0.8778,
+      "step": 437
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.1961572487379104,
+      "learning_rate": 1.9274197036604056e-05,
+      "loss": 0.9524,
+      "step": 438
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.1804576808566998,
+      "learning_rate": 1.9269949102583073e-05,
+      "loss": 0.8683,
+      "step": 439
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.9733805716493049,
+      "learning_rate": 1.926568924468446e-05,
+      "loss": 0.8669,
+      "step": 440
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.0706964473279539,
+      "learning_rate": 1.926141746838765e-05,
+      "loss": 0.8367,
+      "step": 441
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.0033805905176616,
+      "learning_rate": 1.9257133779187396e-05,
+      "loss": 0.7884,
+      "step": 442
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.1611745396135442,
+      "learning_rate": 1.9252838182593785e-05,
+      "loss": 0.8934,
+      "step": 443
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.954029454505071,
+      "learning_rate": 1.9248530684132217e-05,
+      "loss": 0.842,
+      "step": 444
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.1689253660362962,
+      "learning_rate": 1.9244211289343397e-05,
+      "loss": 0.8226,
+      "step": 445
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.0553358851077157,
+      "learning_rate": 1.923988000378334e-05,
+      "loss": 0.8474,
+      "step": 446
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.2187486629066102,
+      "learning_rate": 1.9235536833023345e-05,
+      "loss": 0.8575,
+      "step": 447
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.0348731817900605,
+      "learning_rate": 1.923118178265001e-05,
+      "loss": 0.8526,
+      "step": 448
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.0842630814194327,
+      "learning_rate": 1.922681485826521e-05,
+      "loss": 0.9687,
+      "step": 449
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.113876223105261,
+      "learning_rate": 1.922243606548609e-05,
+      "loss": 0.891,
+      "step": 450
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.1371794908578412,
+      "learning_rate": 1.9218045409945066e-05,
+      "loss": 0.8671,
+      "step": 451
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.1093753268810582,
+      "learning_rate": 1.9213642897289807e-05,
+      "loss": 0.8499,
+      "step": 452
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.0311878548667242,
+      "learning_rate": 1.920922853318324e-05,
+      "loss": 0.8404,
+      "step": 453
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.5676170591497518,
+      "learning_rate": 1.9204802323303533e-05,
+      "loss": 0.8505,
+      "step": 454
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.105576515631698,
+      "learning_rate": 1.9200364273344096e-05,
+      "loss": 0.8631,
+      "step": 455
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.1197845348326956,
+      "learning_rate": 1.9195914389013557e-05,
+      "loss": 0.8843,
+      "step": 456
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.387690470463562,
+      "learning_rate": 1.9191452676035785e-05,
+      "loss": 0.7525,
+      "step": 457
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.1987114359010893,
+      "learning_rate": 1.9186979140149845e-05,
+      "loss": 0.8599,
+      "step": 458
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.217781571332019,
+      "learning_rate": 1.9182493787110027e-05,
+      "loss": 0.8816,
+      "step": 459
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.048348069880512,
+      "learning_rate": 1.91779966226858e-05,
+      "loss": 0.8502,
+      "step": 460
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.0833712301083942,
+      "learning_rate": 1.9173487652661847e-05,
+      "loss": 0.8391,
+      "step": 461
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.2372405354323792,
+      "learning_rate": 1.9168966882838025e-05,
+      "loss": 0.8567,
+      "step": 462
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.0414088831972983,
+      "learning_rate": 1.9164434319029378e-05,
+      "loss": 0.8912,
+      "step": 463
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.252034938210078,
+      "learning_rate": 1.9159889967066104e-05,
+      "loss": 0.8629,
+      "step": 464
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.4159470689143745,
+      "learning_rate": 1.915533383279358e-05,
+      "loss": 0.8894,
+      "step": 465
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.089863489159817,
+      "learning_rate": 1.915076592207234e-05,
+      "loss": 0.8936,
+      "step": 466
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.0763910526561575,
+      "learning_rate": 1.9146186240778048e-05,
+      "loss": 0.8483,
+      "step": 467
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.9653415746089279,
+      "learning_rate": 1.9141594794801525e-05,
+      "loss": 0.8416,
+      "step": 468
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.0979290386531497,
+      "learning_rate": 1.9136991590048713e-05,
+      "loss": 0.8149,
+      "step": 469
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.1134257431959327,
+      "learning_rate": 1.91323766324407e-05,
+      "loss": 0.9428,
+      "step": 470
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.2475415888372035,
+      "learning_rate": 1.912774992791366e-05,
+      "loss": 0.8359,
+      "step": 471
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.1322862514712104,
+      "learning_rate": 1.9123111482418908e-05,
+      "loss": 0.8473,
+      "step": 472
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.0202073961215956,
+      "learning_rate": 1.9118461301922837e-05,
+      "loss": 0.8386,
+      "step": 473
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.1772015279152943,
+      "learning_rate": 1.911379939240695e-05,
+      "loss": 0.8364,
+      "step": 474
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.9781738963360617,
+      "learning_rate": 1.9109125759867834e-05,
+      "loss": 0.931,
+      "step": 475
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.9690542279329114,
+      "learning_rate": 1.9104440410317148e-05,
+      "loss": 0.7959,
+      "step": 476
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.8264203526007248,
+      "learning_rate": 1.9099743349781633e-05,
+      "loss": 0.8622,
+      "step": 477
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.0483364415879783,
+      "learning_rate": 1.9095034584303086e-05,
+      "loss": 0.8971,
+      "step": 478
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.1658429627200173,
+      "learning_rate": 1.9090314119938362e-05,
+      "loss": 0.8295,
+      "step": 479
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.1851938754575262,
+      "learning_rate": 1.9085581962759366e-05,
+      "loss": 0.8943,
+      "step": 480
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.9887334244222568,
+      "learning_rate": 1.9080838118853043e-05,
+      "loss": 0.8942,
+      "step": 481
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.996368417491876,
+      "learning_rate": 1.9076082594321368e-05,
+      "loss": 0.8431,
+      "step": 482
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.1072723640538134,
+      "learning_rate": 1.9071315395281344e-05,
+      "loss": 0.8338,
+      "step": 483
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.1114720146010824,
+      "learning_rate": 1.906653652786499e-05,
+      "loss": 0.8687,
+      "step": 484
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.9795411708341635,
+      "learning_rate": 1.906174599821933e-05,
+      "loss": 0.8558,
+      "step": 485
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.0358155237590705,
+      "learning_rate": 1.9056943812506395e-05,
+      "loss": 0.8539,
+      "step": 486
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.0328294723383766,
+      "learning_rate": 1.9052129976903206e-05,
+      "loss": 0.9297,
+      "step": 487
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.947373757200239,
+      "learning_rate": 1.904730449760177e-05,
+      "loss": 0.7847,
+      "step": 488
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.0349224123359808,
+      "learning_rate": 1.9042467380809068e-05,
+      "loss": 0.8638,
+      "step": 489
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.958310460108622,
+      "learning_rate": 1.903761863274706e-05,
+      "loss": 0.7945,
+      "step": 490
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.4179625401019371,
+      "learning_rate": 1.903275825965265e-05,
+      "loss": 0.8623,
+      "step": 491
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.9984916683954115,
+      "learning_rate": 1.902788626777772e-05,
+      "loss": 0.8547,
+      "step": 492
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.9696717947291953,
+      "learning_rate": 1.902300266338907e-05,
+      "loss": 0.9136,
+      "step": 493
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.0442302283722409,
+      "learning_rate": 1.901810745276845e-05,
+      "loss": 0.9077,
+      "step": 494
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.0615973187793764,
+      "learning_rate": 1.9013200642212547e-05,
+      "loss": 0.9122,
+      "step": 495
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.0278429898382326,
+      "learning_rate": 1.9008282238032955e-05,
+      "loss": 0.8604,
+      "step": 496
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.9677095617976222,
+      "learning_rate": 1.9003352246556187e-05,
+      "loss": 0.9101,
+      "step": 497
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.9735541688566568,
+      "learning_rate": 1.8998410674123665e-05,
+      "loss": 0.8234,
+      "step": 498
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.1110533273357168,
+      "learning_rate": 1.8993457527091696e-05,
+      "loss": 0.8805,
+      "step": 499
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.1119953319593077,
+      "learning_rate": 1.8988492811831485e-05,
+      "loss": 0.7915,
+      "step": 500
+    },
+    {
+      "epoch": 0.35,
+      "eval_loss": 0.8526943325996399,
+      "eval_runtime": 187.0435,
+      "eval_samples_per_second": 216.271,
+      "eval_steps_per_second": 27.036,
+      "step": 500
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.9717769254859505,
+      "learning_rate": 1.8983516534729117e-05,
+      "loss": 0.8541,
+      "step": 501
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.1567304120011481,
+      "learning_rate": 1.8978528702185547e-05,
+      "loss": 0.9735,
+      "step": 502
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.045800467752564,
+      "learning_rate": 1.8973529320616583e-05,
+      "loss": 0.7481,
+      "step": 503
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.9988848948131525,
+      "learning_rate": 1.896851839645291e-05,
+      "loss": 0.8684,
+      "step": 504
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.052167675344499,
+      "learning_rate": 1.8963495936140036e-05,
+      "loss": 0.8679,
+      "step": 505
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.1729615157564766,
+      "learning_rate": 1.8958461946138335e-05,
+      "loss": 0.8659,
+      "step": 506
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.4141646096076652,
+      "learning_rate": 1.8953416432922986e-05,
+      "loss": 0.9133,
+      "step": 507
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.0173125987989802,
+      "learning_rate": 1.8948359402984006e-05,
+      "loss": 0.8216,
+      "step": 508
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.0734401456100644,
+      "learning_rate": 1.894329086282622e-05,
+      "loss": 0.8061,
+      "step": 509
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.0470846911584983,
+      "learning_rate": 1.8938210818969257e-05,
+      "loss": 0.8438,
+      "step": 510
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.0472618723061844,
+      "learning_rate": 1.8933119277947548e-05,
+      "loss": 0.8508,
+      "step": 511
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.0077323188601064,
+      "learning_rate": 1.892801624631031e-05,
+      "loss": 0.9165,
+      "step": 512
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.0844160173693849,
+      "learning_rate": 1.8922901730621544e-05,
+      "loss": 0.8232,
+      "step": 513
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.9761080617342631,
+      "learning_rate": 1.8917775737460015e-05,
+      "loss": 0.9321,
+      "step": 514
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.0110333862570609,
+      "learning_rate": 1.8912638273419257e-05,
+      "loss": 0.9282,
+      "step": 515
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.1126860660977669,
+      "learning_rate": 1.890748934510756e-05,
+      "loss": 0.8755,
+      "step": 516
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.974615001889175,
+      "learning_rate": 1.8902328959147953e-05,
+      "loss": 0.833,
+      "step": 517
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.04420546634678,
+      "learning_rate": 1.8897157122178216e-05,
+      "loss": 0.9264,
+      "step": 518
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.9882139360408511,
+      "learning_rate": 1.8891973840850844e-05,
+      "loss": 0.8147,
+      "step": 519
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.9698350901525421,
+      "learning_rate": 1.8886779121833065e-05,
+      "loss": 0.797,
+      "step": 520
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.0892458752220342,
+      "learning_rate": 1.8881572971806808e-05,
+      "loss": 0.8928,
+      "step": 521
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.9279387036040393,
+      "learning_rate": 1.8876355397468714e-05,
+      "loss": 0.8401,
+      "step": 522
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.0113032414744425,
+      "learning_rate": 1.8871126405530113e-05,
+      "loss": 0.9115,
+      "step": 523
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.9129421884681735,
+      "learning_rate": 1.8865886002717034e-05,
+      "loss": 0.8143,
+      "step": 524
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.3098098146866635,
+      "learning_rate": 1.8860634195770165e-05,
+      "loss": 1.0118,
+      "step": 525
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.9903655628930513,
+      "learning_rate": 1.8855370991444876e-05,
+      "loss": 0.8416,
+      "step": 526
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.0163609363523551,
+      "learning_rate": 1.8850096396511196e-05,
+      "loss": 0.8967,
+      "step": 527
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.058332850451146,
+      "learning_rate": 1.8844810417753797e-05,
+      "loss": 0.8814,
+      "step": 528
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.0760460518670376,
+      "learning_rate": 1.8839513061972007e-05,
+      "loss": 0.8574,
+      "step": 529
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.9547071896987184,
+      "learning_rate": 1.8834204335979777e-05,
+      "loss": 0.8741,
+      "step": 530
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.9830111500512667,
+      "learning_rate": 1.8828884246605696e-05,
+      "loss": 0.857,
+      "step": 531
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.9666641431423578,
+      "learning_rate": 1.882355280069295e-05,
+      "loss": 0.8321,
+      "step": 532
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.4401922028564462,
+      "learning_rate": 1.8818210005099355e-05,
+      "loss": 0.8167,
+      "step": 533
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.0669102521388445,
+      "learning_rate": 1.881285586669731e-05,
+      "loss": 0.7776,
+      "step": 534
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.0093813108688519,
+      "learning_rate": 1.8807490392373808e-05,
+      "loss": 0.841,
+      "step": 535
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.7570201266726655,
+      "learning_rate": 1.880211358903043e-05,
+      "loss": 0.9326,
+      "step": 536
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.9953583010172994,
+      "learning_rate": 1.879672546358332e-05,
+      "loss": 0.8339,
+      "step": 537
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.0508471775378638,
+      "learning_rate": 1.8791326022963197e-05,
+      "loss": 0.8643,
+      "step": 538
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.07762751139997,
+      "learning_rate": 1.8785915274115322e-05,
+      "loss": 0.8097,
+      "step": 539
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.0601387141355354,
+      "learning_rate": 1.8780493223999508e-05,
+      "loss": 0.834,
+      "step": 540
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.021216968150058,
+      "learning_rate": 1.8775059879590106e-05,
+      "loss": 0.8466,
+      "step": 541
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.092057770038313,
+      "learning_rate": 1.876961524787599e-05,
+      "loss": 0.8418,
+      "step": 542
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.0761804463517708,
+      "learning_rate": 1.876415933586056e-05,
+      "loss": 0.8096,
+      "step": 543
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.0382523513195845,
+      "learning_rate": 1.8758692150561718e-05,
+      "loss": 0.8371,
+      "step": 544
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.0560965482503217,
+      "learning_rate": 1.8753213699011877e-05,
+      "loss": 0.9391,
+      "step": 545
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.5019300766603845,
+      "learning_rate": 1.8747723988257924e-05,
+      "loss": 0.8927,
+      "step": 546
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.053684928562445,
+      "learning_rate": 1.8742223025361245e-05,
+      "loss": 0.9148,
+      "step": 547
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.2566833905549095,
+      "learning_rate": 1.8736710817397697e-05,
+      "loss": 0.8452,
+      "step": 548
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.9516384274941435,
+      "learning_rate": 1.8731187371457594e-05,
+      "loss": 0.8479,
+      "step": 549
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 1.244319475578046,
+      "learning_rate": 1.8725652694645714e-05,
+      "loss": 0.806,
+      "step": 550
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.9965107061476687,
+      "learning_rate": 1.8720106794081276e-05,
+      "loss": 0.8408,
+      "step": 551
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.9923816429015672,
+      "learning_rate": 1.8714549676897936e-05,
+      "loss": 0.8249,
+      "step": 552
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.0276121053745104,
+      "learning_rate": 1.870898135024378e-05,
+      "loss": 0.9268,
+      "step": 553
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.9716000925703704,
+      "learning_rate": 1.8703401821281317e-05,
+      "loss": 0.8661,
+      "step": 554
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.028051340486763,
+      "learning_rate": 1.8697811097187445e-05,
+      "loss": 0.8388,
+      "step": 555
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.2861670965261067,
+      "learning_rate": 1.8692209185153493e-05,
+      "loss": 0.8534,
+      "step": 556
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.0530293580854682,
+      "learning_rate": 1.868659609238516e-05,
+      "loss": 0.8347,
+      "step": 557
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.9979681252925467,
+      "learning_rate": 1.868097182610253e-05,
+      "loss": 0.8898,
+      "step": 558
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.05569167326003,
+      "learning_rate": 1.8675336393540065e-05,
+      "loss": 0.9221,
+      "step": 559
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.0355101234684094,
+      "learning_rate": 1.8669689801946585e-05,
+      "loss": 0.8189,
+      "step": 560
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.0206561288788434,
+      "learning_rate": 1.8664032058585265e-05,
+      "loss": 0.8029,
+      "step": 561
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.1234982856261049,
+      "learning_rate": 1.8658363170733625e-05,
+      "loss": 0.8269,
+      "step": 562
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.0286971576376855,
+      "learning_rate": 1.8652683145683523e-05,
+      "loss": 0.9344,
+      "step": 563
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.0455324514804116,
+      "learning_rate": 1.8646991990741138e-05,
+      "loss": 0.889,
+      "step": 564
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.0032390735498653,
+      "learning_rate": 1.8641289713226963e-05,
+      "loss": 0.8517,
+      "step": 565
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 1.1131893195292746,
+      "learning_rate": 1.863557632047581e-05,
+      "loss": 0.8676,
+      "step": 566
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.0714530814917393,
+      "learning_rate": 1.8629851819836774e-05,
+      "loss": 0.8465,
+      "step": 567
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.944334257350204,
+      "learning_rate": 1.862411621867324e-05,
+      "loss": 0.7745,
+      "step": 568
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.0785324680314694,
+      "learning_rate": 1.8618369524362895e-05,
+      "loss": 0.8505,
+      "step": 569
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.0253026699762982,
+      "learning_rate": 1.861261174429765e-05,
+      "loss": 0.8498,
+      "step": 570
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.1073123137203067,
+      "learning_rate": 1.8606842885883724e-05,
+      "loss": 0.8425,
+      "step": 571
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.0138475687542496,
+      "learning_rate": 1.860106295654155e-05,
+      "loss": 0.8327,
+      "step": 572
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.983905689468371,
+      "learning_rate": 1.8595271963705822e-05,
+      "loss": 0.8286,
+      "step": 573
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.9978766241312532,
+      "learning_rate": 1.8589469914825457e-05,
+      "loss": 0.8048,
+      "step": 574
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.9888483691868927,
+      "learning_rate": 1.858365681736359e-05,
+      "loss": 0.8162,
+      "step": 575
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.9966944146838984,
+      "learning_rate": 1.8577832678797575e-05,
+      "loss": 0.8313,
+      "step": 576
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.069934140934502,
+      "learning_rate": 1.8571997506618964e-05,
+      "loss": 0.8822,
+      "step": 577
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.9827128570955065,
+      "learning_rate": 1.8566151308333502e-05,
+      "loss": 0.8465,
+      "step": 578
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.1466769116887388,
+      "learning_rate": 1.856029409146112e-05,
+      "loss": 0.9171,
+      "step": 579
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.0198004108553271,
+      "learning_rate": 1.8554425863535915e-05,
+      "loss": 0.9252,
+      "step": 580
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.0242842691126035,
+      "learning_rate": 1.854854663210616e-05,
+      "loss": 0.7665,
+      "step": 581
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.0890534784704105,
+      "learning_rate": 1.8542656404734264e-05,
+      "loss": 0.8269,
+      "step": 582
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.9015971958546473,
+      "learning_rate": 1.8536755188996797e-05,
+      "loss": 0.8683,
+      "step": 583
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.0291552133995123,
+      "learning_rate": 1.853084299248445e-05,
+      "loss": 0.8375,
+      "step": 584
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.0379854385313003,
+      "learning_rate": 1.852491982280205e-05,
+      "loss": 0.8745,
+      "step": 585
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.0475983349262212,
+      "learning_rate": 1.851898568756853e-05,
+      "loss": 0.8367,
+      "step": 586
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.0198449464309953,
+      "learning_rate": 1.8513040594416934e-05,
+      "loss": 0.8291,
+      "step": 587
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.0449883854278477,
+      "learning_rate": 1.850708455099439e-05,
+      "loss": 0.8851,
+      "step": 588
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.0197024852454293,
+      "learning_rate": 1.850111756496213e-05,
+      "loss": 0.8032,
+      "step": 589
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.9918341986182831,
+      "learning_rate": 1.849513964399545e-05,
+      "loss": 0.8112,
+      "step": 590
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.0484201108935518,
+      "learning_rate": 1.8489150795783702e-05,
+      "loss": 0.8965,
+      "step": 591
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.9689254913631651,
+      "learning_rate": 1.848315102803031e-05,
+      "loss": 0.8887,
+      "step": 592
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.9706717251970238,
+      "learning_rate": 1.8477140348452745e-05,
+      "loss": 0.8358,
+      "step": 593
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.0842401799268369,
+      "learning_rate": 1.8471118764782497e-05,
+      "loss": 0.8892,
+      "step": 594
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.001371823170473,
+      "learning_rate": 1.8465086284765093e-05,
+      "loss": 0.8611,
+      "step": 595
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.9558206663655661,
+      "learning_rate": 1.8459042916160077e-05,
+      "loss": 0.8917,
+      "step": 596
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.0018787110080574,
+      "learning_rate": 1.8452988666740993e-05,
+      "loss": 0.8895,
+      "step": 597
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.9565424323384727,
+      "learning_rate": 1.844692354429539e-05,
+      "loss": 0.847,
+      "step": 598
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.9944052423681737,
+      "learning_rate": 1.8440847556624794e-05,
+      "loss": 0.8445,
+      "step": 599
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.0571874002170982,
+      "learning_rate": 1.8434760711544707e-05,
+      "loss": 0.9123,
+      "step": 600
+    },
+    {
+      "epoch": 0.42,
+      "eval_loss": 0.8448428511619568,
+      "eval_runtime": 187.3148,
+      "eval_samples_per_second": 215.957,
+      "eval_steps_per_second": 26.997,
+      "step": 600
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.0692919112422168,
+      "learning_rate": 1.8428663016884606e-05,
+      "loss": 0.835,
+      "step": 601
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.0624352078303714,
+      "learning_rate": 1.842255448048791e-05,
+      "loss": 0.92,
+      "step": 602
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.0217814818022022,
+      "learning_rate": 1.8416435110211998e-05,
+      "loss": 0.8442,
+      "step": 603
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.0079541176167188,
+      "learning_rate": 1.841030491392817e-05,
+      "loss": 0.8381,
+      "step": 604
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.0605990654308264,
+      "learning_rate": 1.8404163899521666e-05,
+      "loss": 0.8171,
+      "step": 605
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.0802017142498224,
+      "learning_rate": 1.8398012074891636e-05,
+      "loss": 0.9261,
+      "step": 606
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.9980562068741823,
+      "learning_rate": 1.8391849447951128e-05,
+      "loss": 0.7851,
+      "step": 607
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.0262857177043048,
+      "learning_rate": 1.838567602662709e-05,
+      "loss": 0.8495,
+      "step": 608
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.0383479353156029,
+      "learning_rate": 1.837949181886036e-05,
+      "loss": 0.8686,
+      "step": 609
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.9725276530954041,
+      "learning_rate": 1.8373296832605647e-05,
+      "loss": 0.8217,
+      "step": 610
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.96459871670791,
+      "learning_rate": 1.8367091075831517e-05,
+      "loss": 0.8748,
+      "step": 611
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.9652669926432812,
+      "learning_rate": 1.83608745565204e-05,
+      "loss": 0.8963,
+      "step": 612
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.9725870173994274,
+      "learning_rate": 1.8354647282668566e-05,
+      "loss": 0.8629,
+      "step": 613
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.0224955249723027,
+      "learning_rate": 1.8348409262286117e-05,
+      "loss": 0.8649,
+      "step": 614
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.96997495024294,
+      "learning_rate": 1.8342160503396975e-05,
+      "loss": 0.8625,
+      "step": 615
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.9528776424597024,
+      "learning_rate": 1.833590101403889e-05,
+      "loss": 0.7936,
+      "step": 616
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.0092126509995532,
+      "learning_rate": 1.8329630802263397e-05,
+      "loss": 0.8944,
+      "step": 617
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.1003855463170757,
+      "learning_rate": 1.8323349876135836e-05,
+      "loss": 0.9236,
+      "step": 618
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.9882424235256009,
+      "learning_rate": 1.8317058243735312e-05,
+      "loss": 0.8451,
+      "step": 619
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.963270254631444,
+      "learning_rate": 1.8310755913154726e-05,
+      "loss": 0.9003,
+      "step": 620
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.070834999042204,
+      "learning_rate": 1.8304442892500716e-05,
+      "loss": 0.8523,
+      "step": 621
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.9218568344531317,
+      "learning_rate": 1.8298119189893686e-05,
+      "loss": 0.814,
+      "step": 622
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.0121387565853985,
+      "learning_rate": 1.8291784813467775e-05,
+      "loss": 0.8424,
+      "step": 623
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.6096637215697438,
+      "learning_rate": 1.828543977137085e-05,
+      "loss": 0.9638,
+      "step": 624
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.0093617823228238,
+      "learning_rate": 1.82790840717645e-05,
+      "loss": 0.8431,
+      "step": 625
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.0192552554698213,
+      "learning_rate": 1.8272717722824024e-05,
+      "loss": 0.8596,
+      "step": 626
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.3776610014995394,
+      "learning_rate": 1.826634073273841e-05,
+      "loss": 1.0204,
+      "step": 627
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.0675434029261415,
+      "learning_rate": 1.825995310971035e-05,
+      "loss": 0.8003,
+      "step": 628
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.9842725904044134,
+      "learning_rate": 1.8253554861956196e-05,
+      "loss": 0.8682,
+      "step": 629
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.0144622853869512,
+      "learning_rate": 1.8247145997705977e-05,
+      "loss": 0.8425,
+      "step": 630
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.9797681977185918,
+      "learning_rate": 1.8240726525203373e-05,
+      "loss": 0.7792,
+      "step": 631
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.0253288961176894,
+      "learning_rate": 1.8234296452705715e-05,
+      "loss": 0.8911,
+      "step": 632
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.9908479790600693,
+      "learning_rate": 1.8227855788483966e-05,
+      "loss": 0.7824,
+      "step": 633
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.0055635922446198,
+      "learning_rate": 1.8221404540822706e-05,
+      "loss": 0.8183,
+      "step": 634
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.0186612448588224,
+      "learning_rate": 1.8214942718020138e-05,
+      "loss": 0.8881,
+      "step": 635
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.1193248624591543,
+      "learning_rate": 1.8208470328388065e-05,
+      "loss": 0.9409,
+      "step": 636
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.0941512225566314,
+      "learning_rate": 1.820198738025188e-05,
+      "loss": 0.7906,
+      "step": 637
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.1150873133380623,
+      "learning_rate": 1.8195493881950566e-05,
+      "loss": 0.8611,
+      "step": 638
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.9378843998364783,
+      "learning_rate": 1.8188989841836656e-05,
+      "loss": 0.773,
+      "step": 639
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.9753210492248877,
+      "learning_rate": 1.8182475268276265e-05,
+      "loss": 0.8462,
+      "step": 640
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.0828463958224648,
+      "learning_rate": 1.8175950169649048e-05,
+      "loss": 0.8576,
+      "step": 641
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.9769233721699588,
+      "learning_rate": 1.8169414554348194e-05,
+      "loss": 0.7836,
+      "step": 642
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.0461794610793929,
+      "learning_rate": 1.8162868430780426e-05,
+      "loss": 0.8385,
+      "step": 643
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.003015728656599,
+      "learning_rate": 1.8156311807365985e-05,
+      "loss": 0.8552,
+      "step": 644
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.0186870649283732,
+      "learning_rate": 1.814974469253861e-05,
+      "loss": 0.8379,
+      "step": 645
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.0451464039674105,
+      "learning_rate": 1.8143167094745544e-05,
+      "loss": 0.9206,
+      "step": 646
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.9836969096682053,
+      "learning_rate": 1.8136579022447506e-05,
+      "loss": 0.8315,
+      "step": 647
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.1918878087256983,
+      "learning_rate": 1.81299804841187e-05,
+      "loss": 0.7708,
+      "step": 648
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.13014118954486,
+      "learning_rate": 1.8123371488246776e-05,
+      "loss": 0.8264,
+      "step": 649
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.095613581352552,
+      "learning_rate": 1.8116752043332848e-05,
+      "loss": 0.8275,
+      "step": 650
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.001887759694024,
+      "learning_rate": 1.8110122157891466e-05,
+      "loss": 0.8993,
+      "step": 651
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.0591263293529547,
+      "learning_rate": 1.810348184045061e-05,
+      "loss": 0.8239,
+      "step": 652
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.9867565259786909,
+      "learning_rate": 1.8096831099551675e-05,
+      "loss": 0.8347,
+      "step": 653
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.9573471269079379,
+      "learning_rate": 1.8090169943749477e-05,
+      "loss": 0.8389,
+      "step": 654
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.0815591626860135,
+      "learning_rate": 1.808349838161221e-05,
+      "loss": 0.8124,
+      "step": 655
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.0660151285755348,
+      "learning_rate": 1.807681642172147e-05,
+      "loss": 0.9354,
+      "step": 656
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.9862775055482695,
+      "learning_rate": 1.8070124072672212e-05,
+      "loss": 0.8433,
+      "step": 657
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.9895614964471511,
+      "learning_rate": 1.806342134307277e-05,
+      "loss": 0.8612,
+      "step": 658
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.9589334452418762,
+      "learning_rate": 1.805670824154482e-05,
+      "loss": 0.8296,
+      "step": 659
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.9501049834625723,
+      "learning_rate": 1.8049984776723383e-05,
+      "loss": 0.8283,
+      "step": 660
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.1151775277350218,
+      "learning_rate": 1.804325095725681e-05,
+      "loss": 0.795,
+      "step": 661
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.0790263646019147,
+      "learning_rate": 1.803650679180677e-05,
+      "loss": 0.9071,
+      "step": 662
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.9658054663092771,
+      "learning_rate": 1.8029752289048245e-05,
+      "loss": 0.7951,
+      "step": 663
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.06270294763586,
+      "learning_rate": 1.8022987457669504e-05,
+      "loss": 0.8664,
+      "step": 664
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.039694994665679,
+      "learning_rate": 1.801621230637211e-05,
+      "loss": 0.8906,
+      "step": 665
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.9165308822871485,
+      "learning_rate": 1.8009426843870897e-05,
+      "loss": 0.8651,
+      "step": 666
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.106059285409371,
+      "learning_rate": 1.8002631078893964e-05,
+      "loss": 0.8719,
+      "step": 667
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.9388909424266827,
+      "learning_rate": 1.7995825020182664e-05,
+      "loss": 0.8676,
+      "step": 668
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.9428941948815046,
+      "learning_rate": 1.7989008676491576e-05,
+      "loss": 0.7718,
+      "step": 669
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.9435866134778035,
+      "learning_rate": 1.7982182056588536e-05,
+      "loss": 0.9058,
+      "step": 670
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.9484400640960485,
+      "learning_rate": 1.7975345169254574e-05,
+      "loss": 0.8145,
+      "step": 671
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.0135613748858736,
+      "learning_rate": 1.7968498023283937e-05,
+      "loss": 0.8617,
+      "step": 672
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.9907037798755128,
+      "learning_rate": 1.7961640627484064e-05,
+      "loss": 0.8254,
+      "step": 673
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.0539851003236405,
+      "learning_rate": 1.7954772990675583e-05,
+      "loss": 0.8105,
+      "step": 674
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.5109070638813087,
+      "learning_rate": 1.7947895121692293e-05,
+      "loss": 0.8989,
+      "step": 675
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.0570034476843884,
+      "learning_rate": 1.794100702938115e-05,
+      "loss": 0.8437,
+      "step": 676
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.0062280180181036,
+      "learning_rate": 1.7934108722602266e-05,
+      "loss": 0.8234,
+      "step": 677
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.0254412892858618,
+      "learning_rate": 1.7927200210228895e-05,
+      "loss": 0.8789,
+      "step": 678
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.009355678226713,
+      "learning_rate": 1.7920281501147406e-05,
+      "loss": 0.8942,
+      "step": 679
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.9707762185300899,
+      "learning_rate": 1.791335260425729e-05,
+      "loss": 0.8253,
+      "step": 680
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.0507998746245961,
+      "learning_rate": 1.7906413528471155e-05,
+      "loss": 0.8184,
+      "step": 681
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.014499078320948,
+      "learning_rate": 1.7899464282714683e-05,
+      "loss": 0.8316,
+      "step": 682
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.2690613471208891,
+      "learning_rate": 1.7892504875926643e-05,
+      "loss": 0.8476,
+      "step": 683
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.015777608061474,
+      "learning_rate": 1.7885535317058882e-05,
+      "loss": 0.8758,
+      "step": 684
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.0097208382065235,
+      "learning_rate": 1.7878555615076294e-05,
+      "loss": 0.7964,
+      "step": 685
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.0403303018421468,
+      "learning_rate": 1.7871565778956836e-05,
+      "loss": 0.8279,
+      "step": 686
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.0185895791071147,
+      "learning_rate": 1.7864565817691478e-05,
+      "loss": 0.9441,
+      "step": 687
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.219088999355827,
+      "learning_rate": 1.7857555740284234e-05,
+      "loss": 0.9177,
+      "step": 688
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.021606981348204,
+      "learning_rate": 1.7850535555752127e-05,
+      "loss": 0.8538,
+      "step": 689
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.2146520398195872,
+      "learning_rate": 1.7843505273125164e-05,
+      "loss": 0.9698,
+      "step": 690
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.9344553257619521,
+      "learning_rate": 1.7836464901446363e-05,
+      "loss": 0.8474,
+      "step": 691
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.0275115933940366,
+      "learning_rate": 1.782941444977171e-05,
+      "loss": 0.7603,
+      "step": 692
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.0505661615764563,
+      "learning_rate": 1.7822353927170154e-05,
+      "loss": 0.8694,
+      "step": 693
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.090766269758975,
+      "learning_rate": 1.78152833427236e-05,
+      "loss": 0.8273,
+      "step": 694
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.0174580986492128,
+      "learning_rate": 1.78082027055269e-05,
+      "loss": 0.7529,
+      "step": 695
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.0021861850530214,
+      "learning_rate": 1.780111202468783e-05,
+      "loss": 0.9338,
+      "step": 696
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.0609059318746341,
+      "learning_rate": 1.7794011309327094e-05,
+      "loss": 0.8543,
+      "step": 697
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.9551542638515328,
+      "learning_rate": 1.778690056857829e-05,
+      "loss": 0.8321,
+      "step": 698
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.1289162721981014,
+      "learning_rate": 1.7779779811587927e-05,
+      "loss": 0.8622,
+      "step": 699
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.0578729963198346,
+      "learning_rate": 1.7772649047515384e-05,
+      "loss": 0.7849,
+      "step": 700
+    },
+    {
+      "epoch": 0.49,
+      "eval_loss": 0.8381435871124268,
+      "eval_runtime": 187.1931,
+      "eval_samples_per_second": 216.098,
+      "eval_steps_per_second": 27.015,
+      "step": 700
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.9279732332051667,
+      "learning_rate": 1.7765508285532923e-05,
+      "loss": 0.7979,
+      "step": 701
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.0143917337734483,
+      "learning_rate": 1.7758357534825658e-05,
+      "loss": 0.8383,
+      "step": 702
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.0124719670852789,
+      "learning_rate": 1.7751196804591554e-05,
+      "loss": 0.9105,
+      "step": 703
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.0404174429801545,
+      "learning_rate": 1.7744026104041422e-05,
+      "loss": 0.8057,
+      "step": 704
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.0203141485006675,
+      "learning_rate": 1.7736845442398884e-05,
+      "loss": 0.8641,
+      "step": 705
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.0338735850682625,
+      "learning_rate": 1.7729654828900372e-05,
+      "loss": 0.7828,
+      "step": 706
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.9554435548551307,
+      "learning_rate": 1.7722454272795142e-05,
+      "loss": 0.7999,
+      "step": 707
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.0675538600899108,
+      "learning_rate": 1.7715243783345215e-05,
+      "loss": 0.8248,
+      "step": 708
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.0334356374016562,
+      "learning_rate": 1.77080233698254e-05,
+      "loss": 0.9482,
+      "step": 709
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.9625543396191655,
+      "learning_rate": 1.7700793041523272e-05,
+      "loss": 0.807,
+      "step": 710
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.1920966108340543,
+      "learning_rate": 1.7693552807739155e-05,
+      "loss": 1.0113,
+      "step": 711
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.1397337490346742,
+      "learning_rate": 1.7686302677786118e-05,
+      "loss": 0.8753,
+      "step": 712
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.0365348300829782,
+      "learning_rate": 1.7679042660989956e-05,
+      "loss": 0.8879,
+      "step": 713
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.0343554642761457,
+      "learning_rate": 1.7671772766689185e-05,
+      "loss": 0.8311,
+      "step": 714
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.0824083716448243,
+      "learning_rate": 1.7664493004235023e-05,
+      "loss": 1.0056,
+      "step": 715
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.0099315138119533,
+      "learning_rate": 1.7657203382991384e-05,
+      "loss": 0.8656,
+      "step": 716
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.976404160597241,
+      "learning_rate": 1.7649903912334864e-05,
+      "loss": 0.7945,
+      "step": 717
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.2084848684695084,
+      "learning_rate": 1.7642594601654726e-05,
+      "loss": 0.8635,
+      "step": 718
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.9971992358419232,
+      "learning_rate": 1.7635275460352885e-05,
+      "loss": 0.8541,
+      "step": 719
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.0726438623468066,
+      "learning_rate": 1.7627946497843917e-05,
+      "loss": 1.0002,
+      "step": 720
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.0289865886566527,
+      "learning_rate": 1.762060772355501e-05,
+      "loss": 0.795,
+      "step": 721
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.973501695572963,
+      "learning_rate": 1.761325914692599e-05,
+      "loss": 0.7983,
+      "step": 722
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.9356583400580673,
+      "learning_rate": 1.7605900777409282e-05,
+      "loss": 0.8379,
+      "step": 723
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.9969102865233515,
+      "learning_rate": 1.7598532624469916e-05,
+      "loss": 0.8237,
+      "step": 724
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.9193759607412801,
+      "learning_rate": 1.7591154697585496e-05,
+      "loss": 0.7915,
+      "step": 725
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.0241626821655354,
+      "learning_rate": 1.7583767006246203e-05,
+      "loss": 0.8324,
+      "step": 726
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.9535841484168424,
+      "learning_rate": 1.757636955995478e-05,
+      "loss": 0.82,
+      "step": 727
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.0646472607324255,
+      "learning_rate": 1.7568962368226517e-05,
+      "loss": 0.8231,
+      "step": 728
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.02174942247018,
+      "learning_rate": 1.7561545440589237e-05,
+      "loss": 0.9426,
+      "step": 729
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.931456362698152,
+      "learning_rate": 1.755411878658329e-05,
+      "loss": 0.7884,
+      "step": 730
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.084058916314809,
+      "learning_rate": 1.754668241576153e-05,
+      "loss": 0.8664,
+      "step": 731
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.996047425069412,
+      "learning_rate": 1.7539236337689317e-05,
+      "loss": 0.8128,
+      "step": 732
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.9554163336960624,
+      "learning_rate": 1.75317805619445e-05,
+      "loss": 0.8238,
+      "step": 733
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.9330046866060917,
+      "learning_rate": 1.7524315098117385e-05,
+      "loss": 0.8176,
+      "step": 734
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.9760340454661153,
+      "learning_rate": 1.751683995581076e-05,
+      "loss": 0.7941,
+      "step": 735
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.9449691883644274,
+      "learning_rate": 1.7509355144639853e-05,
+      "loss": 0.8796,
+      "step": 736
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.8772646626675931,
+      "learning_rate": 1.7501860674232327e-05,
+      "loss": 0.7869,
+      "step": 737
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.9696546113821329,
+      "learning_rate": 1.7494356554228277e-05,
+      "loss": 0.8105,
+      "step": 738
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.9528515867071141,
+      "learning_rate": 1.7486842794280198e-05,
+      "loss": 0.8852,
+      "step": 739
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.962161515570676,
+      "learning_rate": 1.7479319404053004e-05,
+      "loss": 0.8502,
+      "step": 740
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.9023054126262993,
+      "learning_rate": 1.7471786393223973e-05,
+      "loss": 0.8616,
+      "step": 741
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.0941424697568807,
+      "learning_rate": 1.7464243771482778e-05,
+      "loss": 0.8815,
+      "step": 742
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.9951238405318317,
+      "learning_rate": 1.7456691548531442e-05,
+      "loss": 0.8328,
+      "step": 743
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.9178551178716932,
+      "learning_rate": 1.744912973408434e-05,
+      "loss": 0.8078,
+      "step": 744
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.9496311013052559,
+      "learning_rate": 1.74415583378682e-05,
+      "loss": 0.8138,
+      "step": 745
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.9235589923434312,
+      "learning_rate": 1.7433977369622044e-05,
+      "loss": 0.8794,
+      "step": 746
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.9272481982717993,
+      "learning_rate": 1.7426386839097233e-05,
+      "loss": 0.8177,
+      "step": 747
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 1.0040310234682746,
+      "learning_rate": 1.741878675605742e-05,
+      "loss": 0.8455,
+      "step": 748
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.9249804834714497,
+      "learning_rate": 1.741117713027854e-05,
+      "loss": 0.798,
+      "step": 749
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.9297848707023268,
+      "learning_rate": 1.740355797154881e-05,
+      "loss": 0.8504,
+      "step": 750
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.9926542817126969,
+      "learning_rate": 1.7395929289668707e-05,
+      "loss": 0.858,
+      "step": 751
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.9213349718735717,
+      "learning_rate": 1.7388291094450953e-05,
+      "loss": 0.8135,
+      "step": 752
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.981486353205901,
+      "learning_rate": 1.7380643395720517e-05,
+      "loss": 0.8525,
+      "step": 753
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9204065547943485,
+      "learning_rate": 1.7372986203314578e-05,
+      "loss": 0.755,
+      "step": 754
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9738300309612525,
+      "learning_rate": 1.736531952708254e-05,
+      "loss": 0.8743,
+      "step": 755
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9452690514025663,
+      "learning_rate": 1.7357643376886004e-05,
+      "loss": 0.837,
+      "step": 756
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9727629030140199,
+      "learning_rate": 1.7349957762598745e-05,
+      "loss": 0.8241,
+      "step": 757
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9099279720365867,
+      "learning_rate": 1.734226269410673e-05,
+      "loss": 0.7765,
+      "step": 758
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.018538174707947,
+      "learning_rate": 1.733455818130807e-05,
+      "loss": 0.8626,
+      "step": 759
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9232631930298351,
+      "learning_rate": 1.7326844234113037e-05,
+      "loss": 0.8506,
+      "step": 760
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9405169893976543,
+      "learning_rate": 1.731912086244403e-05,
+      "loss": 0.9013,
+      "step": 761
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9350119648542045,
+      "learning_rate": 1.7311388076235576e-05,
+      "loss": 0.8907,
+      "step": 762
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9474704446032522,
+      "learning_rate": 1.7303645885434305e-05,
+      "loss": 0.8539,
+      "step": 763
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.8704134009300898,
+      "learning_rate": 1.7295894299998953e-05,
+      "loss": 0.745,
+      "step": 764
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9615910463364002,
+      "learning_rate": 1.7288133329900337e-05,
+      "loss": 0.8242,
+      "step": 765
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.9919439198316098,
+      "learning_rate": 1.728036298512134e-05,
+      "loss": 0.8495,
+      "step": 766
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.877565931650257,
+      "learning_rate": 1.727258327565691e-05,
+      "loss": 0.8206,
+      "step": 767
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.9878950331460611,
+      "learning_rate": 1.7264794211514042e-05,
+      "loss": 0.7938,
+      "step": 768
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.9787049714101482,
+      "learning_rate": 1.725699580271176e-05,
+      "loss": 0.7985,
+      "step": 769
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.9885640942295232,
+      "learning_rate": 1.72491880592811e-05,
+      "loss": 0.8215,
+      "step": 770
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.9808580950113753,
+      "learning_rate": 1.724137099126512e-05,
+      "loss": 0.8952,
+      "step": 771
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.0343691086853437,
+      "learning_rate": 1.7233544608718866e-05,
+      "loss": 0.9449,
+      "step": 772
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.0211962127955467,
+      "learning_rate": 1.7225708921709363e-05,
+      "loss": 0.9369,
+      "step": 773
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.0294479359588957,
+      "learning_rate": 1.721786394031561e-05,
+      "loss": 0.8511,
+      "step": 774
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.9820784552509588,
+      "learning_rate": 1.7210009674628553e-05,
+      "loss": 0.7821,
+      "step": 775
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.1866701805556474,
+      "learning_rate": 1.720214613475109e-05,
+      "loss": 0.9516,
+      "step": 776
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.9445534415444331,
+      "learning_rate": 1.719427333079804e-05,
+      "loss": 0.879,
+      "step": 777
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.0884563423631934,
+      "learning_rate": 1.718639127289614e-05,
+      "loss": 0.854,
+      "step": 778
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.3147407731744265,
+      "learning_rate": 1.7178499971184032e-05,
+      "loss": 0.8875,
+      "step": 779
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.9689204655821562,
+      "learning_rate": 1.7170599435812253e-05,
+      "loss": 0.8946,
+      "step": 780
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.047014670498161,
+      "learning_rate": 1.7162689676943206e-05,
+      "loss": 0.8802,
+      "step": 781
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.0306720035023686,
+      "learning_rate": 1.7154770704751173e-05,
+      "loss": 0.8289,
+      "step": 782
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.9451205937730558,
+      "learning_rate": 1.7146842529422268e-05,
+      "loss": 0.8247,
+      "step": 783
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.151775737564422,
+      "learning_rate": 1.713890516115446e-05,
+      "loss": 0.8381,
+      "step": 784
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.8978775908811445,
+      "learning_rate": 1.7130958610157535e-05,
+      "loss": 0.8475,
+      "step": 785
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.9936108721694572,
+      "learning_rate": 1.712300288665309e-05,
+      "loss": 0.8586,
+      "step": 786
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.0384341707306826,
+      "learning_rate": 1.7115038000874526e-05,
+      "loss": 0.9073,
+      "step": 787
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.0698844445150597,
+      "learning_rate": 1.7107063963067025e-05,
+      "loss": 0.8908,
+      "step": 788
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.2170675037604572,
+      "learning_rate": 1.7099080783487535e-05,
+      "loss": 0.8864,
+      "step": 789
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.9251427056637105,
+      "learning_rate": 1.709108847240478e-05,
+      "loss": 0.7828,
+      "step": 790
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.0492859628400695,
+      "learning_rate": 1.7083087040099218e-05,
+      "loss": 0.8347,
+      "step": 791
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.9566010306136947,
+      "learning_rate": 1.7075076496863035e-05,
+      "loss": 0.785,
+      "step": 792
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.0395573578976345,
+      "learning_rate": 1.7067056853000146e-05,
+      "loss": 0.8052,
+      "step": 793
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.0356951969375556,
+      "learning_rate": 1.7059028118826173e-05,
+      "loss": 0.7623,
+      "step": 794
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.8986456959574445,
+      "learning_rate": 1.7050990304668423e-05,
+      "loss": 0.8003,
+      "step": 795
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.963288211867447,
+      "learning_rate": 1.7042943420865884e-05,
+      "loss": 0.8653,
+      "step": 796
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.019609500131109,
+      "learning_rate": 1.7034887477769215e-05,
+      "loss": 0.7268,
+      "step": 797
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.939994639028726,
+      "learning_rate": 1.7026822485740725e-05,
+      "loss": 0.7512,
+      "step": 798
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.0241026823320927,
+      "learning_rate": 1.7018748455154365e-05,
+      "loss": 0.8455,
+      "step": 799
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.100237354106455,
+      "learning_rate": 1.7010665396395706e-05,
+      "loss": 0.8381,
+      "step": 800
+    },
+    {
+      "epoch": 0.56,
+      "eval_loss": 0.8343697190284729,
+      "eval_runtime": 187.1773,
+      "eval_samples_per_second": 216.116,
+      "eval_steps_per_second": 27.017,
+      "step": 800
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.9781956877736848,
+      "learning_rate": 1.700257331986194e-05,
+      "loss": 0.8316,
+      "step": 801
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.1589816575272422,
+      "learning_rate": 1.699447223596185e-05,
+      "loss": 0.9097,
+      "step": 802
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.9441271210063006,
+      "learning_rate": 1.6986362155115812e-05,
+      "loss": 0.7765,
+      "step": 803
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.8544844261366595,
+      "learning_rate": 1.6978243087755775e-05,
+      "loss": 0.8429,
+      "step": 804
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.9275645374138108,
+      "learning_rate": 1.697011504432524e-05,
+      "loss": 0.7922,
+      "step": 805
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.9188009246134373,
+      "learning_rate": 1.6961978035279262e-05,
+      "loss": 0.7762,
+      "step": 806
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.969499301955575,
+      "learning_rate": 1.6953832071084422e-05,
+      "loss": 0.7442,
+      "step": 807
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.980154184825822,
+      "learning_rate": 1.6945677162218824e-05,
+      "loss": 0.7951,
+      "step": 808
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.0230622276939902,
+      "learning_rate": 1.6937513319172078e-05,
+      "loss": 0.8171,
+      "step": 809
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.8991431113287323,
+      "learning_rate": 1.6929340552445283e-05,
+      "loss": 0.8356,
+      "step": 810
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.0510503515631233,
+      "learning_rate": 1.692115887255102e-05,
+      "loss": 0.827,
+      "step": 811
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.9178701479375049,
+      "learning_rate": 1.6912968290013325e-05,
+      "loss": 0.7745,
+      "step": 812
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.9360018185950565,
+      "learning_rate": 1.69047688153677e-05,
+      "loss": 0.7774,
+      "step": 813
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.316296501512368,
+      "learning_rate": 1.689656045916108e-05,
+      "loss": 0.8717,
+      "step": 814
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.9527242777681525,
+      "learning_rate": 1.6888343231951817e-05,
+      "loss": 0.8446,
+      "step": 815
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.9559347218205472,
+      "learning_rate": 1.6880117144309682e-05,
+      "loss": 0.7686,
+      "step": 816
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.9243473396928792,
+      "learning_rate": 1.6871882206815838e-05,
+      "loss": 0.843,
+      "step": 817
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.0471388055538455,
+      "learning_rate": 1.6863638430062835e-05,
+      "loss": 0.8021,
+      "step": 818
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.0681533849169953,
+      "learning_rate": 1.6855385824654595e-05,
+      "loss": 0.9617,
+      "step": 819
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.9907503662351598,
+      "learning_rate": 1.6847124401206384e-05,
+      "loss": 0.8398,
+      "step": 820
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.1267861144590767,
+      "learning_rate": 1.6838854170344827e-05,
+      "loss": 0.9212,
+      "step": 821
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.9371696927779393,
+      "learning_rate": 1.6830575142707866e-05,
+      "loss": 0.8774,
+      "step": 822
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.963308008855645,
+      "learning_rate": 1.6822287328944765e-05,
+      "loss": 0.8295,
+      "step": 823
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.9109515253684652,
+      "learning_rate": 1.681399073971609e-05,
+      "loss": 0.7968,
+      "step": 824
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.9512690574893296,
+      "learning_rate": 1.6805685385693684e-05,
+      "loss": 0.7919,
+      "step": 825
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.9277861987245727,
+      "learning_rate": 1.679737127756068e-05,
+      "loss": 0.8529,
+      "step": 826
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.9786371869517354,
+      "learning_rate": 1.6789048426011456e-05,
+      "loss": 0.8313,
+      "step": 827
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.8669153191576762,
+      "learning_rate": 1.678071684175165e-05,
+      "loss": 0.7609,
+      "step": 828
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.9367989371618103,
+      "learning_rate": 1.6772376535498122e-05,
+      "loss": 0.9188,
+      "step": 829
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.9319488948306222,
+      "learning_rate": 1.676402751797896e-05,
+      "loss": 0.8388,
+      "step": 830
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.9060314101494492,
+      "learning_rate": 1.6755669799933455e-05,
+      "loss": 0.8088,
+      "step": 831
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.9133531832761467,
+      "learning_rate": 1.674730339211208e-05,
+      "loss": 0.8333,
+      "step": 832
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.8858572520830194,
+      "learning_rate": 1.6738928305276498e-05,
+      "loss": 0.7837,
+      "step": 833
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.9748551828519176,
+      "learning_rate": 1.673054455019953e-05,
+      "loss": 0.8224,
+      "step": 834
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.9837368048123739,
+      "learning_rate": 1.6722152137665147e-05,
+      "loss": 0.7404,
+      "step": 835
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.9613075813336643,
+      "learning_rate": 1.671375107846846e-05,
+      "loss": 0.8452,
+      "step": 836
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.9022543493122955,
+      "learning_rate": 1.6705341383415694e-05,
+      "loss": 0.7479,
+      "step": 837
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.9381842639325487,
+      "learning_rate": 1.6696923063324192e-05,
+      "loss": 0.8381,
+      "step": 838
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.0225858550563163,
+      "learning_rate": 1.668849612902238e-05,
+      "loss": 0.8341,
+      "step": 839
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.9251227060209148,
+      "learning_rate": 1.6680060591349774e-05,
+      "loss": 0.8935,
+      "step": 840
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.9775603659563589,
+      "learning_rate": 1.6671616461156963e-05,
+      "loss": 0.9297,
+      "step": 841
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.1795119299407328,
+      "learning_rate": 1.6663163749305568e-05,
+      "loss": 0.8387,
+      "step": 842
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.0465229174302006,
+      "learning_rate": 1.665470246666826e-05,
+      "loss": 0.8438,
+      "step": 843
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.98481295597245,
+      "learning_rate": 1.6646232624128737e-05,
+      "loss": 0.8428,
+      "step": 844
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.8961490284253136,
+      "learning_rate": 1.66377542325817e-05,
+      "loss": 0.7532,
+      "step": 845
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.9818574745716003,
+      "learning_rate": 1.662926730293286e-05,
+      "loss": 0.8756,
+      "step": 846
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.1230544122671045,
+      "learning_rate": 1.6620771846098893e-05,
+      "loss": 0.9136,
+      "step": 847
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 1.0068717380378494,
+      "learning_rate": 1.6612267873007456e-05,
+      "loss": 0.9546,
+      "step": 848
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.96385028656652,
+      "learning_rate": 1.6603755394597155e-05,
+      "loss": 0.9825,
+      "step": 849
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.9111777420506347,
+      "learning_rate": 1.659523442181754e-05,
+      "loss": 0.7414,
+      "step": 850
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.9688593070694009,
+      "learning_rate": 1.6586704965629082e-05,
+      "loss": 0.8562,
+      "step": 851
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.9412426968470645,
+      "learning_rate": 1.6578167037003173e-05,
+      "loss": 0.8365,
+      "step": 852
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.9955700458307191,
+      "learning_rate": 1.656962064692209e-05,
+      "loss": 0.8033,
+      "step": 853
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.9691385337948485,
+      "learning_rate": 1.6561065806379005e-05,
+      "loss": 0.8894,
+      "step": 854
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.272726220321745,
+      "learning_rate": 1.6552502526377952e-05,
+      "loss": 0.9017,
+      "step": 855
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.1010411361156522,
+      "learning_rate": 1.6543930817933833e-05,
+      "loss": 0.9577,
+      "step": 856
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.9585360635545656,
+      "learning_rate": 1.6535350692072377e-05,
+      "loss": 0.7982,
+      "step": 857
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.9322358583855594,
+      "learning_rate": 1.6526762159830145e-05,
+      "loss": 0.9181,
+      "step": 858
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.9944886470410574,
+      "learning_rate": 1.6518165232254515e-05,
+      "loss": 0.8202,
+      "step": 859
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.0730618597851427,
+      "learning_rate": 1.6509559920403663e-05,
+      "loss": 0.8973,
+      "step": 860
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.971148988153191,
+      "learning_rate": 1.6500946235346545e-05,
+      "loss": 0.8835,
+      "step": 861
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.1248576590356607,
+      "learning_rate": 1.649232418816289e-05,
+      "loss": 0.8985,
+      "step": 862
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.9562539651810333,
+      "learning_rate": 1.6483693789943184e-05,
+      "loss": 0.8613,
+      "step": 863
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.9405416826009309,
+      "learning_rate": 1.6475055051788656e-05,
+      "loss": 0.7449,
+      "step": 864
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.9773248641215648,
+      "learning_rate": 1.646640798481126e-05,
+      "loss": 0.7838,
+      "step": 865
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.9352806966087698,
+      "learning_rate": 1.645775260013366e-05,
+      "loss": 0.9275,
+      "step": 866
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.3131597709452303,
+      "learning_rate": 1.644908890888923e-05,
+      "loss": 0.829,
+      "step": 867
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.0087536169626201,
+      "learning_rate": 1.644041692222202e-05,
+      "loss": 0.8893,
+      "step": 868
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.9344144260658866,
+      "learning_rate": 1.6431736651286753e-05,
+      "loss": 0.801,
+      "step": 869
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.0242357186289224,
+      "learning_rate": 1.64230481072488e-05,
+      "loss": 0.8546,
+      "step": 870
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.0407673296307454,
+      "learning_rate": 1.6414351301284194e-05,
+      "loss": 0.8561,
+      "step": 871
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.3359537210790293,
+      "learning_rate": 1.640564624457957e-05,
+      "loss": 0.7949,
+      "step": 872
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.9652079053552713,
+      "learning_rate": 1.6396932948332192e-05,
+      "loss": 0.8802,
+      "step": 873
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.9971827000934618,
+      "learning_rate": 1.6388211423749926e-05,
+      "loss": 0.8079,
+      "step": 874
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.9873924164513037,
+      "learning_rate": 1.63794816820512e-05,
+      "loss": 0.9431,
+      "step": 875
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.035708221786434,
+      "learning_rate": 1.6370743734465037e-05,
+      "loss": 0.8462,
+      "step": 876
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.0659960068581555,
+      "learning_rate": 1.6361997592231002e-05,
+      "loss": 0.7802,
+      "step": 877
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.95199115267153,
+      "learning_rate": 1.6353243266599204e-05,
+      "loss": 0.8402,
+      "step": 878
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.0352302934148574,
+      "learning_rate": 1.6344480768830272e-05,
+      "loss": 0.8664,
+      "step": 879
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.0727757180183821,
+      "learning_rate": 1.633571011019536e-05,
+      "loss": 0.8469,
+      "step": 880
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.0103695746994772,
+      "learning_rate": 1.632693130197611e-05,
+      "loss": 0.8474,
+      "step": 881
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.9279816172567278,
+      "learning_rate": 1.6318144355464636e-05,
+      "loss": 0.8251,
+      "step": 882
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.0177667753191524,
+      "learning_rate": 1.6309349281963554e-05,
+      "loss": 0.8037,
+      "step": 883
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.0205024180657574,
+      "learning_rate": 1.6300546092785893e-05,
+      "loss": 0.8937,
+      "step": 884
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.8978191926449389,
+      "learning_rate": 1.629173479925515e-05,
+      "loss": 0.7662,
+      "step": 885
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.9794809130650164,
+      "learning_rate": 1.6282915412705237e-05,
+      "loss": 0.778,
+      "step": 886
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.1484736613051145,
+      "learning_rate": 1.627408794448047e-05,
+      "loss": 0.8501,
+      "step": 887
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.2006586269251032,
+      "learning_rate": 1.6265252405935567e-05,
+      "loss": 0.8275,
+      "step": 888
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.9565918259072778,
+      "learning_rate": 1.6256408808435632e-05,
+      "loss": 0.7764,
+      "step": 889
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.9230346491449858,
+      "learning_rate": 1.6247557163356127e-05,
+      "loss": 0.8537,
+      "step": 890
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.1385277004351848,
+      "learning_rate": 1.6238697482082865e-05,
+      "loss": 0.8683,
+      "step": 891
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.9525638773784072,
+      "learning_rate": 1.6229829776012e-05,
+      "loss": 0.8534,
+      "step": 892
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.9347976283055996,
+      "learning_rate": 1.6220954056550007e-05,
+      "loss": 0.7843,
+      "step": 893
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.0583805539492264,
+      "learning_rate": 1.621207033511367e-05,
+      "loss": 0.8629,
+      "step": 894
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.0048885232922549,
+      "learning_rate": 1.620317862313006e-05,
+      "loss": 0.9306,
+      "step": 895
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.9232667435607793,
+      "learning_rate": 1.6194278932036537e-05,
+      "loss": 0.819,
+      "step": 896
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.8682759371657264,
+      "learning_rate": 1.6185371273280716e-05,
+      "loss": 0.7695,
+      "step": 897
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.996774327536682,
+      "learning_rate": 1.6176455658320458e-05,
+      "loss": 0.8802,
+      "step": 898
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.9556030109023915,
+      "learning_rate": 1.616753209862387e-05,
+      "loss": 0.8849,
+      "step": 899
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.9688301385550866,
+      "learning_rate": 1.6158600605669264e-05,
+      "loss": 0.7652,
+      "step": 900
+    },
+    {
+      "epoch": 0.63,
+      "eval_loss": 0.822993814945221,
+      "eval_runtime": 187.3145,
+      "eval_samples_per_second": 215.958,
+      "eval_steps_per_second": 26.997,
+      "step": 900
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.9318473637952628,
+      "learning_rate": 1.6149661190945166e-05,
+      "loss": 0.8415,
+      "step": 901
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.9274136224074371,
+      "learning_rate": 1.6140713865950295e-05,
+      "loss": 0.8316,
+      "step": 902
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.9014362272472443,
+      "learning_rate": 1.6131758642193532e-05,
+      "loss": 0.8,
+      "step": 903
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.9216394797926182,
+      "learning_rate": 1.6122795531193928e-05,
+      "loss": 0.8132,
+      "step": 904
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.9725243755979751,
+      "learning_rate": 1.6113824544480677e-05,
+      "loss": 0.9002,
+      "step": 905
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.9882233530522131,
+      "learning_rate": 1.61048456935931e-05,
+      "loss": 0.8617,
+      "step": 906
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.9917887320236796,
+      "learning_rate": 1.6095858990080645e-05,
+      "loss": 0.8216,
+      "step": 907
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.0406878445820225,
+      "learning_rate": 1.608686444550285e-05,
+      "loss": 0.9715,
+      "step": 908
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.1217295732718597,
+      "learning_rate": 1.607786207142933e-05,
+      "loss": 0.8178,
+      "step": 909
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.1019645894046524,
+      "learning_rate": 1.606885187943979e-05,
+      "loss": 0.9135,
+      "step": 910
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.9552174332315572,
+      "learning_rate": 1.6059833881123992e-05,
+      "loss": 0.8409,
+      "step": 911
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.9611142730552733,
+      "learning_rate": 1.6050808088081724e-05,
+      "loss": 0.9057,
+      "step": 912
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.0288385302918484,
+      "learning_rate": 1.6041774511922805e-05,
+      "loss": 0.8507,
+      "step": 913
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.886330987251663,
+      "learning_rate": 1.6032733164267068e-05,
+      "loss": 0.7642,
+      "step": 914
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.951400710797225,
+      "learning_rate": 1.602368405674435e-05,
+      "loss": 0.8739,
+      "step": 915
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.9686000092526066,
+      "learning_rate": 1.601462720099445e-05,
+      "loss": 0.8591,
+      "step": 916
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.8930468558428761,
+      "learning_rate": 1.600556260866715e-05,
+      "loss": 0.8217,
+      "step": 917
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.9035192721578178,
+      "learning_rate": 1.599649029142218e-05,
+      "loss": 0.8376,
+      "step": 918
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.8646970486410438,
+      "learning_rate": 1.5987410260929207e-05,
+      "loss": 0.8474,
+      "step": 919
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.9248834234575523,
+      "learning_rate": 1.597832252886781e-05,
+      "loss": 0.8444,
+      "step": 920
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.9427172975191896,
+      "learning_rate": 1.5969227106927484e-05,
+      "loss": 0.826,
+      "step": 921
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.4077800946655261,
+      "learning_rate": 1.596012400680762e-05,
+      "loss": 0.7879,
+      "step": 922
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.837913547326813,
+      "learning_rate": 1.5951013240217476e-05,
+      "loss": 0.7558,
+      "step": 923
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.9175537997708259,
+      "learning_rate": 1.5941894818876167e-05,
+      "loss": 0.8003,
+      "step": 924
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.9341359649482986,
+      "learning_rate": 1.593276875451267e-05,
+      "loss": 0.9179,
+      "step": 925
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.8901143268702544,
+      "learning_rate": 1.592363505886578e-05,
+      "loss": 0.749,
+      "step": 926
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.955997180286808,
+      "learning_rate": 1.591449374368412e-05,
+      "loss": 0.8353,
+      "step": 927
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.9247328217729289,
+      "learning_rate": 1.5905344820726095e-05,
+      "loss": 0.891,
+      "step": 928
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.9606738684735865,
+      "learning_rate": 1.5896188301759915e-05,
+      "loss": 0.7897,
+      "step": 929
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.9863303377461871,
+      "learning_rate": 1.5887024198563552e-05,
+      "loss": 0.8119,
+      "step": 930
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.9680881639267663,
+      "learning_rate": 1.5877852522924733e-05,
+      "loss": 0.8187,
+      "step": 931
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.9137190159154377,
+      "learning_rate": 1.586867328664093e-05,
+      "loss": 0.7274,
+      "step": 932
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.8914334634486136,
+      "learning_rate": 1.5859486501519336e-05,
+      "loss": 0.7296,
+      "step": 933
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.918290906922166,
+      "learning_rate": 1.5850292179376853e-05,
+      "loss": 0.8544,
+      "step": 934
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.8983013443167193,
+      "learning_rate": 1.584109033204008e-05,
+      "loss": 0.8667,
+      "step": 935
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.8783351660772043,
+      "learning_rate": 1.58318809713453e-05,
+      "loss": 0.8418,
+      "step": 936
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.888234321129709,
+      "learning_rate": 1.5822664109138455e-05,
+      "loss": 0.8389,
+      "step": 937
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 1.082130346683686,
+      "learning_rate": 1.5813439757275136e-05,
+      "loss": 0.8873,
+      "step": 938
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.9420960129507885,
+      "learning_rate": 1.580420792762057e-05,
+      "loss": 0.8385,
+      "step": 939
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.8962298028357553,
+      "learning_rate": 1.5794968632049598e-05,
+      "loss": 0.8424,
+      "step": 940
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.9622507092912542,
+      "learning_rate": 1.5785721882446674e-05,
+      "loss": 0.9024,
+      "step": 941
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.924531664156692,
+      "learning_rate": 1.577646769070583e-05,
+      "loss": 0.8497,
+      "step": 942
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.8907729853195185,
+      "learning_rate": 1.576720606873068e-05,
+      "loss": 0.7535,
+      "step": 943
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.922725222248796,
+      "learning_rate": 1.5757937028434386e-05,
+      "loss": 0.8141,
+      "step": 944
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.0247896956566702,
+      "learning_rate": 1.5748660581739658e-05,
+      "loss": 0.8645,
+      "step": 945
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.9683488898803148,
+      "learning_rate": 1.5739376740578725e-05,
+      "loss": 0.8602,
+      "step": 946
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.9250842111817158,
+      "learning_rate": 1.573008551689335e-05,
+      "loss": 0.7781,
+      "step": 947
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.0417883831148922,
+      "learning_rate": 1.572078692263476e-05,
+      "loss": 0.8281,
+      "step": 948
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.9067729621768262,
+      "learning_rate": 1.5711480969763687e-05,
+      "loss": 0.8305,
+      "step": 949
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.9354864669464692,
+      "learning_rate": 1.570216767025032e-05,
+      "loss": 0.7685,
+      "step": 950
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.9522717869161624,
+      "learning_rate": 1.569284703607429e-05,
+      "loss": 0.8501,
+      "step": 951
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.9356497506936815,
+      "learning_rate": 1.568351907922468e-05,
+      "loss": 0.7932,
+      "step": 952
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.991045124767421,
+      "learning_rate": 1.5674183811699977e-05,
+      "loss": 0.7607,
+      "step": 953
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.9425625884087384,
+      "learning_rate": 1.5664841245508083e-05,
+      "loss": 0.8086,
+      "step": 954
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.9457733257889495,
+      "learning_rate": 1.5655491392666278e-05,
+      "loss": 0.8644,
+      "step": 955
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.870259271198405,
+      "learning_rate": 1.564613426520122e-05,
+      "loss": 0.7676,
+      "step": 956
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.9439558057808078,
+      "learning_rate": 1.5636769875148923e-05,
+      "loss": 0.8501,
+      "step": 957
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.9299410524152394,
+      "learning_rate": 1.5627398234554743e-05,
+      "loss": 0.7952,
+      "step": 958
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.97156518556245,
+      "learning_rate": 1.5618019355473365e-05,
+      "loss": 0.8358,
+      "step": 959
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.9100662530892337,
+      "learning_rate": 1.5608633249968783e-05,
+      "loss": 0.7722,
+      "step": 960
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.0511679090288089,
+      "learning_rate": 1.5599239930114284e-05,
+      "loss": 0.7877,
+      "step": 961
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.8761142337595833,
+      "learning_rate": 1.5589839407992438e-05,
+      "loss": 0.7341,
+      "step": 962
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.8990059427825337,
+      "learning_rate": 1.5580431695695074e-05,
+      "loss": 0.8201,
+      "step": 963
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.0026432046525355,
+      "learning_rate": 1.5571016805323277e-05,
+      "loss": 0.867,
+      "step": 964
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.9429348388707998,
+      "learning_rate": 1.5561594748987366e-05,
+      "loss": 0.9206,
+      "step": 965
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.9167946059522355,
+      "learning_rate": 1.5552165538806864e-05,
+      "loss": 0.7893,
+      "step": 966
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.1871295768545387,
+      "learning_rate": 1.554272918691051e-05,
+      "loss": 0.8828,
+      "step": 967
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.9679415215947258,
+      "learning_rate": 1.5533285705436223e-05,
+      "loss": 0.8616,
+      "step": 968
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.9402913050429529,
+      "learning_rate": 1.55238351065311e-05,
+      "loss": 0.8481,
+      "step": 969
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.0003446194946921,
+      "learning_rate": 1.5514377402351376e-05,
+      "loss": 0.8663,
+      "step": 970
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.8965443469538628,
+      "learning_rate": 1.550491260506245e-05,
+      "loss": 0.798,
+      "step": 971
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.9302594905643129,
+      "learning_rate": 1.5495440726838823e-05,
+      "loss": 0.8539,
+      "step": 972
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.87788264580982,
+      "learning_rate": 1.5485961779864113e-05,
+      "loss": 0.8012,
+      "step": 973
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.8959993564405986,
+      "learning_rate": 1.5476475776331038e-05,
+      "loss": 0.8234,
+      "step": 974
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.921332705440342,
+      "learning_rate": 1.5466982728441372e-05,
+      "loss": 0.7851,
+      "step": 975
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.9266184071321535,
+      "learning_rate": 1.5457482648405974e-05,
+      "loss": 0.8268,
+      "step": 976
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.2058651665186106,
+      "learning_rate": 1.5447975548444736e-05,
+      "loss": 0.8695,
+      "step": 977
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.0440445222540657,
+      "learning_rate": 1.5438461440786576e-05,
+      "loss": 0.8452,
+      "step": 978
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.1348865256068061,
+      "learning_rate": 1.5428940337669437e-05,
+      "loss": 0.8998,
+      "step": 979
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.9749698913377348,
+      "learning_rate": 1.541941225134025e-05,
+      "loss": 0.8256,
+      "step": 980
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.9220326956037203,
+      "learning_rate": 1.5409877194054933e-05,
+      "loss": 0.8253,
+      "step": 981
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 1.0765304215328484,
+      "learning_rate": 1.540033517807838e-05,
+      "loss": 0.8896,
+      "step": 982
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.9206753023806792,
+      "learning_rate": 1.5390786215684418e-05,
+      "loss": 0.7448,
+      "step": 983
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.9708651392708145,
+      "learning_rate": 1.5381230319155814e-05,
+      "loss": 0.9519,
+      "step": 984
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.218620357778404,
+      "learning_rate": 1.5371667500784263e-05,
+      "loss": 0.853,
+      "step": 985
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.1911032016704852,
+      "learning_rate": 1.5362097772870363e-05,
+      "loss": 0.9497,
+      "step": 986
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.9398864147894423,
+      "learning_rate": 1.5352521147723586e-05,
+      "loss": 0.8944,
+      "step": 987
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.9661960223002711,
+      "learning_rate": 1.5342937637662293e-05,
+      "loss": 0.7934,
+      "step": 988
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.8855725014661222,
+      "learning_rate": 1.5333347255013688e-05,
+      "loss": 0.7527,
+      "step": 989
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.0787207219460402,
+      "learning_rate": 1.532375001211383e-05,
+      "loss": 0.8123,
+      "step": 990
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.0384264021628293,
+      "learning_rate": 1.5314145921307575e-05,
+      "loss": 0.831,
+      "step": 991
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.8883493942021998,
+      "learning_rate": 1.5304534994948615e-05,
+      "loss": 0.8752,
+      "step": 992
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.9373183287986955,
+      "learning_rate": 1.5294917245399432e-05,
+      "loss": 0.8047,
+      "step": 993
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.978544050648968,
+      "learning_rate": 1.5285292685031264e-05,
+      "loss": 0.7891,
+      "step": 994
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.9645365407413318,
+      "learning_rate": 1.5275661326224128e-05,
+      "loss": 0.7789,
+      "step": 995
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.9920781081825699,
+      "learning_rate": 1.5266023181366785e-05,
+      "loss": 0.808,
+      "step": 996
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.8737892505774815,
+      "learning_rate": 1.5256378262856717e-05,
+      "loss": 0.8315,
+      "step": 997
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.946250945761465,
+      "learning_rate": 1.5246726583100117e-05,
+      "loss": 0.7663,
+      "step": 998
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.0186205498174532,
+      "learning_rate": 1.5237068154511892e-05,
+      "loss": 0.8631,
+      "step": 999
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.0311343742793326,
+      "learning_rate": 1.5227402989515607e-05,
+      "loss": 0.9006,
+      "step": 1000
+    },
+    {
+      "epoch": 0.7,
+      "eval_loss": 0.8166661858558655,
+      "eval_runtime": 187.1666,
+      "eval_samples_per_second": 216.128,
+      "eval_steps_per_second": 27.019,
+      "step": 1000
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.0366177745268226,
+      "learning_rate": 1.5217731100543507e-05,
+      "loss": 0.8218,
+      "step": 1001
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.9506885195850883,
+      "learning_rate": 1.5208052500036488e-05,
+      "loss": 0.7495,
+      "step": 1002
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.0338066376511579,
+      "learning_rate": 1.5198367200444067e-05,
+      "loss": 0.7877,
+      "step": 1003
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.9322333474694569,
+      "learning_rate": 1.518867521422439e-05,
+      "loss": 0.7568,
+      "step": 1004
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.9609534097048402,
+      "learning_rate": 1.517897655384419e-05,
+      "loss": 0.7611,
+      "step": 1005
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.0325720072776678,
+      "learning_rate": 1.5169271231778799e-05,
+      "loss": 0.7689,
+      "step": 1006
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.9418930343657563,
+      "learning_rate": 1.5159559260512109e-05,
+      "loss": 0.8998,
+      "step": 1007
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.0235955508964831,
+      "learning_rate": 1.5149840652536574e-05,
+      "loss": 0.8997,
+      "step": 1008
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.9414657504967248,
+      "learning_rate": 1.5140115420353173e-05,
+      "loss": 0.8484,
+      "step": 1009
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.9245083032735134,
+      "learning_rate": 1.5130383576471415e-05,
+      "loss": 0.7856,
+      "step": 1010
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.9161849096789572,
+      "learning_rate": 1.5120645133409304e-05,
+      "loss": 0.835,
+      "step": 1011
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.8943557174659821,
+      "learning_rate": 1.5110900103693346e-05,
+      "loss": 0.7928,
+      "step": 1012
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.0426274191582412,
+      "learning_rate": 1.510114849985851e-05,
+      "loss": 0.9187,
+      "step": 1013
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.9941986293419836,
+      "learning_rate": 1.509139033444822e-05,
+      "loss": 0.7255,
+      "step": 1014
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.004374685633853,
+      "learning_rate": 1.5081625620014344e-05,
+      "loss": 0.8519,
+      "step": 1015
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.8963139975673067,
+      "learning_rate": 1.5071854369117178e-05,
+      "loss": 0.8662,
+      "step": 1016
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.87070455564935,
+      "learning_rate": 1.506207659432541e-05,
+      "loss": 0.7996,
+      "step": 1017
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.278643651530359,
+      "learning_rate": 1.5052292308216143e-05,
+      "loss": 0.8961,
+      "step": 1018
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.885800971500993,
+      "learning_rate": 1.5042501523374834e-05,
+      "loss": 0.7309,
+      "step": 1019
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.0241069842563644,
+      "learning_rate": 1.5032704252395315e-05,
+      "loss": 0.8767,
+      "step": 1020
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.9138257247149999,
+      "learning_rate": 1.5022900507879752e-05,
+      "loss": 0.8001,
+      "step": 1021
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.941823978584876,
+      "learning_rate": 1.501309030243864e-05,
+      "loss": 0.8897,
+      "step": 1022
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.9469953491831391,
+      "learning_rate": 1.5003273648690786e-05,
+      "loss": 0.9473,
+      "step": 1023
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.9207848899089832,
+      "learning_rate": 1.4993450559263285e-05,
+      "loss": 0.8003,
+      "step": 1024
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.8926744152182166,
+      "learning_rate": 1.4983621046791524e-05,
+      "loss": 0.8505,
+      "step": 1025
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.9408245436209166,
+      "learning_rate": 1.4973785123919138e-05,
+      "loss": 0.8341,
+      "step": 1026
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.8629284716812735,
+      "learning_rate": 1.4963942803298015e-05,
+      "loss": 0.7872,
+      "step": 1027
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.8962927330501582,
+      "learning_rate": 1.495409409758827e-05,
+      "loss": 0.7843,
+      "step": 1028
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9258411057151813,
+      "learning_rate": 1.494423901945823e-05,
+      "loss": 0.7379,
+      "step": 1029
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9798123698284505,
+      "learning_rate": 1.4934377581584425e-05,
+      "loss": 0.7774,
+      "step": 1030
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9132130466817296,
+      "learning_rate": 1.4924509796651557e-05,
+      "loss": 0.7349,
+      "step": 1031
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9640135589319957,
+      "learning_rate": 1.4914635677352495e-05,
+      "loss": 0.7628,
+      "step": 1032
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9171028168227539,
+      "learning_rate": 1.4904755236388261e-05,
+      "loss": 0.7603,
+      "step": 1033
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9418343170159471,
+      "learning_rate": 1.4894868486468004e-05,
+      "loss": 0.8052,
+      "step": 1034
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9598503981723232,
+      "learning_rate": 1.4884975440308989e-05,
+      "loss": 0.8606,
+      "step": 1035
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.006904224011919,
+      "learning_rate": 1.4875076110636574e-05,
+      "loss": 0.8552,
+      "step": 1036
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9124164328650459,
+      "learning_rate": 1.4865170510184213e-05,
+      "loss": 0.7951,
+      "step": 1037
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9550560395782298,
+      "learning_rate": 1.4855258651693409e-05,
+      "loss": 0.8485,
+      "step": 1038
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.9775924030336234,
+      "learning_rate": 1.4845340547913729e-05,
+      "loss": 0.807,
+      "step": 1039
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.8750397902514163,
+      "learning_rate": 1.4835416211602771e-05,
+      "loss": 0.7876,
+      "step": 1040
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.8987320057700271,
+      "learning_rate": 1.4825485655526141e-05,
+      "loss": 0.7378,
+      "step": 1041
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.1034171103710384,
+      "learning_rate": 1.4815548892457454e-05,
+      "loss": 0.7968,
+      "step": 1042
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 4.840081865612383,
+      "learning_rate": 1.4805605935178304e-05,
+      "loss": 0.8786,
+      "step": 1043
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.003518046889941,
+      "learning_rate": 1.479565679647826e-05,
+      "loss": 0.8088,
+      "step": 1044
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.9524891013440205,
+      "learning_rate": 1.4785701489154831e-05,
+      "loss": 0.8254,
+      "step": 1045
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.9247188686906767,
+      "learning_rate": 1.477574002601347e-05,
+      "loss": 0.7715,
+      "step": 1046
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.8565668707747258,
+      "learning_rate": 1.4765772419867546e-05,
+      "loss": 0.6915,
+      "step": 1047
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.8846787014674027,
+      "learning_rate": 1.4755798683538326e-05,
+      "loss": 0.8063,
+      "step": 1048
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.9727973778751486,
+      "learning_rate": 1.4745818829854961e-05,
+      "loss": 0.8962,
+      "step": 1049
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.9056258797643352,
+      "learning_rate": 1.473583287165448e-05,
+      "loss": 0.7706,
+      "step": 1050
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.943869945595465,
+      "learning_rate": 1.472584082178176e-05,
+      "loss": 0.8285,
+      "step": 1051
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.9194718412835111,
+      "learning_rate": 1.4715842693089509e-05,
+      "loss": 0.8362,
+      "step": 1052
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.9223521065577772,
+      "learning_rate": 1.4705838498438257e-05,
+      "loss": 0.9504,
+      "step": 1053
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.8443051767638792,
+      "learning_rate": 1.469582825069634e-05,
+      "loss": 0.7688,
+      "step": 1054
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.0946829760149586,
+      "learning_rate": 1.4685811962739873e-05,
+      "loss": 0.7208,
+      "step": 1055
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.9671049669919355,
+      "learning_rate": 1.467578964745275e-05,
+      "loss": 0.7694,
+      "step": 1056
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.0050083456532306,
+      "learning_rate": 1.4665761317726608e-05,
+      "loss": 0.8395,
+      "step": 1057
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.9401037736089004,
+      "learning_rate": 1.4655726986460825e-05,
+      "loss": 0.8505,
+      "step": 1058
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.9271070392150399,
+      "learning_rate": 1.4645686666562503e-05,
+      "loss": 0.8382,
+      "step": 1059
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.9161677661842693,
+      "learning_rate": 1.463564037094644e-05,
+      "loss": 0.7854,
+      "step": 1060
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.8785511543929789,
+      "learning_rate": 1.462558811253512e-05,
+      "loss": 0.877,
+      "step": 1061
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.902800739075103,
+      "learning_rate": 1.4615529904258705e-05,
+      "loss": 0.7724,
+      "step": 1062
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.061019400757448,
+      "learning_rate": 1.4605465759055006e-05,
+      "loss": 0.8561,
+      "step": 1063
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.9825523333215745,
+      "learning_rate": 1.4595395689869464e-05,
+      "loss": 0.8538,
+      "step": 1064
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.9181139164136977,
+      "learning_rate": 1.4585319709655153e-05,
+      "loss": 0.845,
+      "step": 1065
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.9735509559194173,
+      "learning_rate": 1.457523783137274e-05,
+      "loss": 0.8189,
+      "step": 1066
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.9841500939652149,
+      "learning_rate": 1.4565150067990483e-05,
+      "loss": 0.8327,
+      "step": 1067
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.8988091678342344,
+      "learning_rate": 1.4555056432484207e-05,
+      "loss": 0.7679,
+      "step": 1068
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.944014867434341,
+      "learning_rate": 1.4544956937837298e-05,
+      "loss": 0.8342,
+      "step": 1069
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.1056206144151361,
+      "learning_rate": 1.4534851597040666e-05,
+      "loss": 0.8465,
+      "step": 1070
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.9660105954153827,
+      "learning_rate": 1.452474042309275e-05,
+      "loss": 0.8517,
+      "step": 1071
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.0472463649236523,
+      "learning_rate": 1.451462342899949e-05,
+      "loss": 0.8548,
+      "step": 1072
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.9762588730513893,
+      "learning_rate": 1.4504500627774313e-05,
+      "loss": 0.8829,
+      "step": 1073
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.9489649675731793,
+      "learning_rate": 1.4494372032438116e-05,
+      "loss": 0.7979,
+      "step": 1074
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.990120487046462,
+      "learning_rate": 1.4484237656019244e-05,
+      "loss": 0.8817,
+      "step": 1075
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.9075790189381905,
+      "learning_rate": 1.447409751155349e-05,
+      "loss": 0.7941,
+      "step": 1076
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.8831467966088242,
+      "learning_rate": 1.4463951612084048e-05,
+      "loss": 0.8214,
+      "step": 1077
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.0125327855835362,
+      "learning_rate": 1.445379997066153e-05,
+      "loss": 0.7625,
+      "step": 1078
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.9101408924011467,
+      "learning_rate": 1.4443642600343932e-05,
+      "loss": 0.7662,
+      "step": 1079
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.9019559452891296,
+      "learning_rate": 1.4433479514196615e-05,
+      "loss": 0.7902,
+      "step": 1080
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.9207943304282776,
+      "learning_rate": 1.4423310725292288e-05,
+      "loss": 0.875,
+      "step": 1081
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.929234724837283,
+      "learning_rate": 1.4413136246711008e-05,
+      "loss": 0.8082,
+      "step": 1082
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.8907862985352285,
+      "learning_rate": 1.4402956091540137e-05,
+      "loss": 0.8322,
+      "step": 1083
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.1412941737168578,
+      "learning_rate": 1.4392770272874344e-05,
+      "loss": 0.9219,
+      "step": 1084
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.9762146877555116,
+      "learning_rate": 1.4382578803815592e-05,
+      "loss": 0.8423,
+      "step": 1085
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.9177873129682679,
+      "learning_rate": 1.4372381697473098e-05,
+      "loss": 0.8468,
+      "step": 1086
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.9079263274942797,
+      "learning_rate": 1.4362178966963335e-05,
+      "loss": 0.85,
+      "step": 1087
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.9099542658185245,
+      "learning_rate": 1.435197062541002e-05,
+      "loss": 0.8523,
+      "step": 1088
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.9510183315678092,
+      "learning_rate": 1.4341756685944069e-05,
+      "loss": 0.84,
+      "step": 1089
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.9423797325098769,
+      "learning_rate": 1.4331537161703612e-05,
+      "loss": 0.9125,
+      "step": 1090
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.9666830124658934,
+      "learning_rate": 1.4321312065833961e-05,
+      "loss": 0.8334,
+      "step": 1091
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.050044401434885,
+      "learning_rate": 1.4311081411487593e-05,
+      "loss": 0.818,
+      "step": 1092
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.971299073600437,
+      "learning_rate": 1.4300845211824133e-05,
+      "loss": 0.7817,
+      "step": 1093
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 1.0622022924767303,
+      "learning_rate": 1.4290603480010337e-05,
+      "loss": 0.8204,
+      "step": 1094
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.8940801885856055,
+      "learning_rate": 1.428035622922009e-05,
+      "loss": 0.8421,
+      "step": 1095
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.9711244830016025,
+      "learning_rate": 1.427010347263436e-05,
+      "loss": 0.7441,
+      "step": 1096
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.9108635630363782,
+      "learning_rate": 1.4259845223441201e-05,
+      "loss": 0.8546,
+      "step": 1097
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.9957713741461913,
+      "learning_rate": 1.4249581494835744e-05,
+      "loss": 0.7939,
+      "step": 1098
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.947865938918299,
+      "learning_rate": 1.4239312300020155e-05,
+      "loss": 0.9002,
+      "step": 1099
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.955212444373132,
+      "learning_rate": 1.422903765220363e-05,
+      "loss": 0.8589,
+      "step": 1100
+    },
+    {
+      "epoch": 0.77,
+      "eval_loss": 0.8088098764419556,
+      "eval_runtime": 187.3476,
+      "eval_samples_per_second": 215.919,
+      "eval_steps_per_second": 26.993,
+      "step": 1100
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.8964662083850382,
+      "learning_rate": 1.4218757564602395e-05,
+      "loss": 0.8541,
+      "step": 1101
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.8530014267313194,
+      "learning_rate": 1.4208472050439653e-05,
+      "loss": 0.7937,
+      "step": 1102
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.9305448294362012,
+      "learning_rate": 1.4198181122945596e-05,
+      "loss": 0.8638,
+      "step": 1103
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.0396841586497925,
+      "learning_rate": 1.4187884795357385e-05,
+      "loss": 0.9494,
+      "step": 1104
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.9137449553212285,
+      "learning_rate": 1.4177583080919124e-05,
+      "loss": 0.8144,
+      "step": 1105
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.0190854848529411,
+      "learning_rate": 1.4167275992881835e-05,
+      "loss": 0.928,
+      "step": 1106
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.9353683263098286,
+      "learning_rate": 1.415696354450347e-05,
+      "loss": 0.7958,
+      "step": 1107
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.9209680060709873,
+      "learning_rate": 1.4146645749048861e-05,
+      "loss": 0.8689,
+      "step": 1108
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.898167043125677,
+      "learning_rate": 1.4136322619789729e-05,
+      "loss": 0.882,
+      "step": 1109
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.0433577130243574,
+      "learning_rate": 1.4125994170004644e-05,
+      "loss": 0.8298,
+      "step": 1110
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.8853306203175052,
+      "learning_rate": 1.4115660412979033e-05,
+      "loss": 0.8184,
+      "step": 1111
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.9242770972057193,
+      "learning_rate": 1.410532136200514e-05,
+      "loss": 0.833,
+      "step": 1112
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.8551822597963529,
+      "learning_rate": 1.409497703038202e-05,
+      "loss": 0.8164,
+      "step": 1113
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.9074799292741689,
+      "learning_rate": 1.4084627431415525e-05,
+      "loss": 0.7843,
+      "step": 1114
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.9030612446473509,
+      "learning_rate": 1.407427257841828e-05,
+      "loss": 0.8201,
+      "step": 1115
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.8554120431464576,
+      "learning_rate": 1.4063912484709663e-05,
+      "loss": 0.7776,
+      "step": 1116
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.9501159195331209,
+      "learning_rate": 1.4053547163615803e-05,
+      "loss": 0.8334,
+      "step": 1117
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.8938629752691346,
+      "learning_rate": 1.4043176628469545e-05,
+      "loss": 0.8048,
+      "step": 1118
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.0428868146721986,
+      "learning_rate": 1.4032800892610446e-05,
+      "loss": 0.8879,
+      "step": 1119
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.9516786058647616,
+      "learning_rate": 1.402241996938475e-05,
+      "loss": 0.8268,
+      "step": 1120
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.9922315164671542,
+      "learning_rate": 1.401203387214537e-05,
+      "loss": 0.7855,
+      "step": 1121
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.9060023281261937,
+      "learning_rate": 1.4001642614251881e-05,
+      "loss": 0.8324,
+      "step": 1122
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.9601763856903994,
+      "learning_rate": 1.3991246209070493e-05,
+      "loss": 0.8077,
+      "step": 1123
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.9577849169044219,
+      "learning_rate": 1.3980844669974036e-05,
+      "loss": 0.8355,
+      "step": 1124
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.8960036099129043,
+      "learning_rate": 1.3970438010341945e-05,
+      "loss": 0.8218,
+      "step": 1125
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.9461892853385439,
+      "learning_rate": 1.3960026243560243e-05,
+      "loss": 0.8302,
+      "step": 1126
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.9086947040687626,
+      "learning_rate": 1.3949609383021518e-05,
+      "loss": 0.7887,
+      "step": 1127
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.0253390198732695,
+      "learning_rate": 1.3939187442124915e-05,
+      "loss": 0.7914,
+      "step": 1128
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.039432871340379,
+      "learning_rate": 1.3928760434276108e-05,
+      "loss": 0.8485,
+      "step": 1129
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.8970385765648137,
+      "learning_rate": 1.3918328372887295e-05,
+      "loss": 0.8225,
+      "step": 1130
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.8726470987900731,
+      "learning_rate": 1.3907891271377174e-05,
+      "loss": 0.8193,
+      "step": 1131
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.3511389366858273,
+      "learning_rate": 1.3897449143170915e-05,
+      "loss": 0.7433,
+      "step": 1132
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.9003522292869762,
+      "learning_rate": 1.3887002001700171e-05,
+      "loss": 0.7925,
+      "step": 1133
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.9604476147261559,
+      "learning_rate": 1.3876549860403034e-05,
+      "loss": 0.7373,
+      "step": 1134
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.8931730019645792,
+      "learning_rate": 1.3866092732724022e-05,
+      "loss": 0.8209,
+      "step": 1135
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.9533626103452278,
+      "learning_rate": 1.3855630632114077e-05,
+      "loss": 0.8119,
+      "step": 1136
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.9210708748170129,
+      "learning_rate": 1.3845163572030533e-05,
+      "loss": 0.8493,
+      "step": 1137
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.88266502750559,
+      "learning_rate": 1.3834691565937104e-05,
+      "loss": 0.7441,
+      "step": 1138
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.9714788778711877,
+      "learning_rate": 1.3824214627303865e-05,
+      "loss": 0.9199,
+      "step": 1139
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.93445787773671,
+      "learning_rate": 1.381373276960724e-05,
+      "loss": 0.8291,
+      "step": 1140
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.9064124219434208,
+      "learning_rate": 1.3803246006329975e-05,
+      "loss": 0.7146,
+      "step": 1141
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.8611879842958952,
+      "learning_rate": 1.3792754350961126e-05,
+      "loss": 0.7505,
+      "step": 1142
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.8957401093812204,
+      "learning_rate": 1.3782257816996047e-05,
+      "loss": 0.8717,
+      "step": 1143
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.9580044205886977,
+      "learning_rate": 1.3771756417936363e-05,
+      "loss": 0.7724,
+      "step": 1144
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.8208381090776238,
+      "learning_rate": 1.376125016728996e-05,
+      "loss": 0.7815,
+      "step": 1145
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.8539973246995288,
+      "learning_rate": 1.3750739078570958e-05,
+      "loss": 0.7544,
+      "step": 1146
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.0014874732651948,
+      "learning_rate": 1.3740223165299708e-05,
+      "loss": 0.849,
+      "step": 1147
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.9126804758407212,
+      "learning_rate": 1.3729702441002763e-05,
+      "loss": 0.8151,
+      "step": 1148
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.9687131763677577,
+      "learning_rate": 1.3719176919212865e-05,
+      "loss": 0.9047,
+      "step": 1149
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.969129888421451,
+      "learning_rate": 1.3708646613468925e-05,
+      "loss": 0.8256,
+      "step": 1150
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.9653248623855282,
+      "learning_rate": 1.3698111537316013e-05,
+      "loss": 0.8613,
+      "step": 1151
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.9477030668227989,
+      "learning_rate": 1.3687571704305329e-05,
+      "loss": 0.8741,
+      "step": 1152
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.9467403055169439,
+      "learning_rate": 1.3677027127994195e-05,
+      "loss": 0.8158,
+      "step": 1153
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.8881445660059051,
+      "learning_rate": 1.3666477821946036e-05,
+      "loss": 0.7622,
+      "step": 1154
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.9823672949148402,
+      "learning_rate": 1.3655923799730355e-05,
+      "loss": 0.8495,
+      "step": 1155
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.9247600537788654,
+      "learning_rate": 1.3645365074922728e-05,
+      "loss": 0.793,
+      "step": 1156
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.9356835238185092,
+      "learning_rate": 1.3634801661104771e-05,
+      "loss": 0.8562,
+      "step": 1157
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.8899179313721481,
+      "learning_rate": 1.3624233571864143e-05,
+      "loss": 0.8039,
+      "step": 1158
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.8907403674153812,
+      "learning_rate": 1.361366082079451e-05,
+      "loss": 0.7767,
+      "step": 1159
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.8609601412851831,
+      "learning_rate": 1.3603083421495535e-05,
+      "loss": 0.757,
+      "step": 1160
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.854048536932743,
+      "learning_rate": 1.3592501387572858e-05,
+      "loss": 0.8074,
+      "step": 1161
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.8583038832209344,
+      "learning_rate": 1.3581914732638088e-05,
+      "loss": 0.7514,
+      "step": 1162
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.9102036881133101,
+      "learning_rate": 1.3571323470308772e-05,
+      "loss": 0.879,
+      "step": 1163
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.9892591027955606,
+      "learning_rate": 1.3560727614208382e-05,
+      "loss": 0.7995,
+      "step": 1164
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.8587977399936928,
+      "learning_rate": 1.3550127177966306e-05,
+      "loss": 0.7842,
+      "step": 1165
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.8595666226056676,
+      "learning_rate": 1.3539522175217815e-05,
+      "loss": 0.848,
+      "step": 1166
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.9459028126824949,
+      "learning_rate": 1.3528912619604055e-05,
+      "loss": 0.8037,
+      "step": 1167
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.9332031814722181,
+      "learning_rate": 1.3518298524772037e-05,
+      "loss": 0.7559,
+      "step": 1168
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.9494114970027568,
+      "learning_rate": 1.3507679904374605e-05,
+      "loss": 0.8466,
+      "step": 1169
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.8704421497086452,
+      "learning_rate": 1.3497056772070417e-05,
+      "loss": 0.8155,
+      "step": 1170
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.918802130165072,
+      "learning_rate": 1.3486429141523952e-05,
+      "loss": 0.7543,
+      "step": 1171
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.9429001916650661,
+      "learning_rate": 1.3475797026405455e-05,
+      "loss": 0.8587,
+      "step": 1172
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.9468204443661803,
+      "learning_rate": 1.3465160440390956e-05,
+      "loss": 0.8496,
+      "step": 1173
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.9994607399005049,
+      "learning_rate": 1.3454519397162224e-05,
+      "loss": 0.8622,
+      "step": 1174
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.1004175782584324,
+      "learning_rate": 1.3443873910406773e-05,
+      "loss": 0.9329,
+      "step": 1175
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.0020293054378908,
+      "learning_rate": 1.3433223993817821e-05,
+      "loss": 0.8327,
+      "step": 1176
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.9863748418961857,
+      "learning_rate": 1.3422569661094292e-05,
+      "loss": 0.8051,
+      "step": 1177
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.9320271692137702,
+      "learning_rate": 1.3411910925940785e-05,
+      "loss": 0.79,
+      "step": 1178
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.9135413110414086,
+      "learning_rate": 1.3401247802067565e-05,
+      "loss": 0.884,
+      "step": 1179
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.052077980452545,
+      "learning_rate": 1.3390580303190541e-05,
+      "loss": 0.8115,
+      "step": 1180
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.9349164080270702,
+      "learning_rate": 1.3379908443031253e-05,
+      "loss": 0.7915,
+      "step": 1181
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.2324139315928029,
+      "learning_rate": 1.3369232235316847e-05,
+      "loss": 0.8856,
+      "step": 1182
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.9594440807814407,
+      "learning_rate": 1.335855169378006e-05,
+      "loss": 0.8418,
+      "step": 1183
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.9197962337492925,
+      "learning_rate": 1.3347866832159206e-05,
+      "loss": 0.783,
+      "step": 1184
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.9305460243082693,
+      "learning_rate": 1.3337177664198162e-05,
+      "loss": 0.782,
+      "step": 1185
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.9373693721632484,
+      "learning_rate": 1.3326484203646333e-05,
+      "loss": 0.8484,
+      "step": 1186
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.916846515507665,
+      "learning_rate": 1.3315786464258647e-05,
+      "loss": 0.8364,
+      "step": 1187
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.3099634206251927,
+      "learning_rate": 1.3305084459795547e-05,
+      "loss": 0.86,
+      "step": 1188
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.9211071839446942,
+      "learning_rate": 1.3294378204022948e-05,
+      "loss": 0.8358,
+      "step": 1189
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.9267300515820535,
+      "learning_rate": 1.3283667710712245e-05,
+      "loss": 0.7812,
+      "step": 1190
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.9469209502966243,
+      "learning_rate": 1.3272952993640276e-05,
+      "loss": 0.8247,
+      "step": 1191
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.8804259262659825,
+      "learning_rate": 1.3262234066589315e-05,
+      "loss": 0.7344,
+      "step": 1192
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.9132281206728566,
+      "learning_rate": 1.3251510943347049e-05,
+      "loss": 0.7078,
+      "step": 1193
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.9892610920445006,
+      "learning_rate": 1.3240783637706566e-05,
+      "loss": 0.8737,
+      "step": 1194
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.1053484676015917,
+      "learning_rate": 1.3230052163466337e-05,
+      "loss": 0.789,
+      "step": 1195
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.8906000246500937,
+      "learning_rate": 1.321931653443018e-05,
+      "loss": 0.8797,
+      "step": 1196
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.8979403780314781,
+      "learning_rate": 1.3208576764407277e-05,
+      "loss": 0.7897,
+      "step": 1197
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 1.123038434121297,
+      "learning_rate": 1.3197832867212123e-05,
+      "loss": 0.8288,
+      "step": 1198
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.8854053827854093,
+      "learning_rate": 1.3187084856664524e-05,
+      "loss": 0.7818,
+      "step": 1199
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.8361990472972701,
+      "learning_rate": 1.3176332746589587e-05,
+      "loss": 0.7635,
+      "step": 1200
+    },
+    {
+      "epoch": 0.84,
+      "eval_loss": 0.8016307353973389,
+      "eval_runtime": 187.26,
+      "eval_samples_per_second": 216.02,
+      "eval_steps_per_second": 27.005,
+      "step": 1200
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.909024655853674,
+      "learning_rate": 1.3165576550817674e-05,
+      "loss": 0.8462,
+      "step": 1201
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.8313849727666298,
+      "learning_rate": 1.3154816283184416e-05,
+      "loss": 0.7439,
+      "step": 1202
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.9042935007013845,
+      "learning_rate": 1.314405195753068e-05,
+      "loss": 0.8269,
+      "step": 1203
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.9956573032713342,
+      "learning_rate": 1.313328358770255e-05,
+      "loss": 0.8411,
+      "step": 1204
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.8692638810097056,
+      "learning_rate": 1.3122511187551313e-05,
+      "loss": 0.7821,
+      "step": 1205
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.8861664890808165,
+      "learning_rate": 1.3111734770933435e-05,
+      "loss": 0.8507,
+      "step": 1206
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.010454663093981,
+      "learning_rate": 1.3100954351710561e-05,
+      "loss": 0.87,
+      "step": 1207
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.9296535384437415,
+      "learning_rate": 1.3090169943749475e-05,
+      "loss": 0.8421,
+      "step": 1208
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.8936969609941104,
+      "learning_rate": 1.307938156092209e-05,
+      "loss": 0.8661,
+      "step": 1209
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.9227907157257895,
+      "learning_rate": 1.3068589217105441e-05,
+      "loss": 0.7934,
+      "step": 1210
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.9335664869611315,
+      "learning_rate": 1.3057792926181648e-05,
+      "loss": 0.7751,
+      "step": 1211
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.9019312696351173,
+      "learning_rate": 1.3046992702037912e-05,
+      "loss": 0.8335,
+      "step": 1212
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.8899699042126639,
+      "learning_rate": 1.3036188558566498e-05,
+      "loss": 0.7718,
+      "step": 1213
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.8694994902176645,
+      "learning_rate": 1.3025380509664704e-05,
+      "loss": 0.8224,
+      "step": 1214
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.8679155573621159,
+      "learning_rate": 1.3014568569234855e-05,
+      "loss": 0.793,
+      "step": 1215
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.8731683514157172,
+      "learning_rate": 1.300375275118429e-05,
+      "loss": 0.7145,
+      "step": 1216
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.9617633270628331,
+      "learning_rate": 1.2992933069425322e-05,
+      "loss": 0.8323,
+      "step": 1217
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.9205318690136027,
+      "learning_rate": 1.2982109537875242e-05,
+      "loss": 0.7919,
+      "step": 1218
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.8720538824550501,
+      "learning_rate": 1.2971282170456287e-05,
+      "loss": 0.739,
+      "step": 1219
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.9832029171959736,
+      "learning_rate": 1.2960450981095643e-05,
+      "loss": 0.8244,
+      "step": 1220
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.8770451456283266,
+      "learning_rate": 1.2949615983725386e-05,
+      "loss": 0.7719,
+      "step": 1221
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.8840598304118039,
+      "learning_rate": 1.2938777192282516e-05,
+      "loss": 0.8434,
+      "step": 1222
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.9055448234498211,
+      "learning_rate": 1.2927934620708902e-05,
+      "loss": 0.8374,
+      "step": 1223
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.8729195849353666,
+      "learning_rate": 1.291708828295127e-05,
+      "loss": 0.774,
+      "step": 1224
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.9284381671456617,
+      "learning_rate": 1.2906238192961201e-05,
+      "loss": 0.9232,
+      "step": 1225
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.8999733124576376,
+      "learning_rate": 1.28953843646951e-05,
+      "loss": 0.8288,
+      "step": 1226
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.9245147181563812,
+      "learning_rate": 1.2884526812114176e-05,
+      "loss": 0.8119,
+      "step": 1227
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.8530109921661233,
+      "learning_rate": 1.2873665549184429e-05,
+      "loss": 0.7753,
+      "step": 1228
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.9043600405219798,
+      "learning_rate": 1.2862800589876636e-05,
+      "loss": 0.8343,
+      "step": 1229
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.8830059417691293,
+      "learning_rate": 1.2851931948166328e-05,
+      "loss": 0.7843,
+      "step": 1230
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.9418973868303533,
+      "learning_rate": 1.2841059638033765e-05,
+      "loss": 0.7896,
+      "step": 1231
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.9100590025127928,
+      "learning_rate": 1.2830183673463941e-05,
+      "loss": 0.7616,
+      "step": 1232
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.8495301196659513,
+      "learning_rate": 1.2819304068446532e-05,
+      "loss": 0.7689,
+      "step": 1233
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.8780282740804941,
+      "learning_rate": 1.280842083697591e-05,
+      "loss": 0.7213,
+      "step": 1234
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.8601052861886926,
+      "learning_rate": 1.2797533993051108e-05,
+      "loss": 0.8038,
+      "step": 1235
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.873800855570753,
+      "learning_rate": 1.2786643550675806e-05,
+      "loss": 0.9166,
+      "step": 1236
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.872977358615432,
+      "learning_rate": 1.277574952385831e-05,
+      "loss": 0.7841,
+      "step": 1237
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.9348029069693068,
+      "learning_rate": 1.2764851926611538e-05,
+      "loss": 0.8556,
+      "step": 1238
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.9177313188133553,
+      "learning_rate": 1.2753950772953006e-05,
+      "loss": 0.8374,
+      "step": 1239
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.8880377226520811,
+      "learning_rate": 1.2743046076904795e-05,
+      "loss": 0.807,
+      "step": 1240
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.8581317781991079,
+      "learning_rate": 1.2732137852493548e-05,
+      "loss": 0.7839,
+      "step": 1241
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.9423518364761677,
+      "learning_rate": 1.2721226113750449e-05,
+      "loss": 0.9337,
+      "step": 1242
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.8697716259429181,
+      "learning_rate": 1.2710310874711195e-05,
+      "loss": 0.7732,
+      "step": 1243
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.030589559664667,
+      "learning_rate": 1.2699392149415992e-05,
+      "loss": 0.8249,
+      "step": 1244
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.9171711323829616,
+      "learning_rate": 1.268846995190953e-05,
+      "loss": 0.8395,
+      "step": 1245
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.8315506767883373,
+      "learning_rate": 1.2677544296240963e-05,
+      "loss": 0.7798,
+      "step": 1246
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.20140255611588,
+      "learning_rate": 1.2666615196463887e-05,
+      "loss": 0.8241,
+      "step": 1247
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.9073000748254678,
+      "learning_rate": 1.2655682666636346e-05,
+      "loss": 0.854,
+      "step": 1248
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.9417372875604748,
+      "learning_rate": 1.2644746720820778e-05,
+      "loss": 0.9372,
+      "step": 1249
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.9290306561838421,
+      "learning_rate": 1.2633807373084022e-05,
+      "loss": 0.8309,
+      "step": 1250
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.9534748535916682,
+      "learning_rate": 1.262286463749729e-05,
+      "loss": 0.906,
+      "step": 1251
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.8950263175691467,
+      "learning_rate": 1.261191852813616e-05,
+      "loss": 0.8991,
+      "step": 1252
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.9129258423461022,
+      "learning_rate": 1.2600969059080543e-05,
+      "loss": 0.8243,
+      "step": 1253
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.8634972435621588,
+      "learning_rate": 1.2590016244414674e-05,
+      "loss": 0.781,
+      "step": 1254
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.903443853461887,
+      "learning_rate": 1.2579060098227086e-05,
+      "loss": 0.8297,
+      "step": 1255
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.946311697477589,
+      "learning_rate": 1.25681006346106e-05,
+      "loss": 0.8786,
+      "step": 1256
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.8728360680570422,
+      "learning_rate": 1.255713786766231e-05,
+      "loss": 0.7831,
+      "step": 1257
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.9486738553484381,
+      "learning_rate": 1.2546171811483555e-05,
+      "loss": 0.7885,
+      "step": 1258
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.852399743727923,
+      "learning_rate": 1.2535202480179902e-05,
+      "loss": 0.7437,
+      "step": 1259
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.8446167680879512,
+      "learning_rate": 1.2524229887861132e-05,
+      "loss": 0.8096,
+      "step": 1260
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.9119806793650314,
+      "learning_rate": 1.2513254048641228e-05,
+      "loss": 0.835,
+      "step": 1261
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.9060150023697798,
+      "learning_rate": 1.2502274976638337e-05,
+      "loss": 0.8107,
+      "step": 1262
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.9178164241596967,
+      "learning_rate": 1.2491292685974775e-05,
+      "loss": 0.8105,
+      "step": 1263
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.9694234877661287,
+      "learning_rate": 1.2480307190776994e-05,
+      "loss": 0.9413,
+      "step": 1264
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.054215876059854,
+      "learning_rate": 1.2469318505175564e-05,
+      "loss": 0.8096,
+      "step": 1265
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.9302411914965589,
+      "learning_rate": 1.2458326643305162e-05,
+      "loss": 0.836,
+      "step": 1266
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.9092781470891533,
+      "learning_rate": 1.244733161930456e-05,
+      "loss": 0.9085,
+      "step": 1267
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.0185778379552477,
+      "learning_rate": 1.2436333447316578e-05,
+      "loss": 0.8225,
+      "step": 1268
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.008752810569094,
+      "learning_rate": 1.2425332141488103e-05,
+      "loss": 0.7972,
+      "step": 1269
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.9508043331539013,
+      "learning_rate": 1.241432771597004e-05,
+      "loss": 0.7727,
+      "step": 1270
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.9081696570803016,
+      "learning_rate": 1.240332018491732e-05,
+      "loss": 0.8256,
+      "step": 1271
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.9184233825599799,
+      "learning_rate": 1.2392309562488857e-05,
+      "loss": 0.771,
+      "step": 1272
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.8776440619902922,
+      "learning_rate": 1.2381295862847546e-05,
+      "loss": 0.7685,
+      "step": 1273
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.900259780582409,
+      "learning_rate": 1.2370279100160243e-05,
+      "loss": 0.7493,
+      "step": 1274
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.0195338622101757,
+      "learning_rate": 1.235925928859774e-05,
+      "loss": 0.8172,
+      "step": 1275
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.9777459968555523,
+      "learning_rate": 1.2348236442334747e-05,
+      "loss": 0.8224,
+      "step": 1276
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.8132773288609149,
+      "learning_rate": 1.2337210575549885e-05,
+      "loss": 0.7546,
+      "step": 1277
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.978595342667741,
+      "learning_rate": 1.2326181702425662e-05,
+      "loss": 0.7756,
+      "step": 1278
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.916777195194068,
+      "learning_rate": 1.2315149837148441e-05,
+      "loss": 0.7982,
+      "step": 1279
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.981152620310292,
+      "learning_rate": 1.230411499390845e-05,
+      "loss": 0.7198,
+      "step": 1280
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.8969732111862679,
+      "learning_rate": 1.229307718689973e-05,
+      "loss": 0.7637,
+      "step": 1281
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.8998245989558533,
+      "learning_rate": 1.2282036430320153e-05,
+      "loss": 0.8069,
+      "step": 1282
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.9532879266081471,
+      "learning_rate": 1.2270992738371363e-05,
+      "loss": 0.7618,
+      "step": 1283
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.8787836684734067,
+      "learning_rate": 1.2259946125258804e-05,
+      "loss": 0.7614,
+      "step": 1284
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.9396395297610475,
+      "learning_rate": 1.224889660519166e-05,
+      "loss": 0.805,
+      "step": 1285
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.8943880825019749,
+      "learning_rate": 1.223784419238286e-05,
+      "loss": 0.7378,
+      "step": 1286
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.8440758161057578,
+      "learning_rate": 1.2226788901049052e-05,
+      "loss": 0.7209,
+      "step": 1287
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.9473861048709283,
+      "learning_rate": 1.221573074541059e-05,
+      "loss": 0.7762,
+      "step": 1288
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.8593533374009497,
+      "learning_rate": 1.2204669739691508e-05,
+      "loss": 0.7285,
+      "step": 1289
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.8949062108663332,
+      "learning_rate": 1.2193605898119513e-05,
+      "loss": 0.8688,
+      "step": 1290
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.9090979861560906,
+      "learning_rate": 1.2182539234925954e-05,
+      "loss": 0.8169,
+      "step": 1291
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.0789848804501179,
+      "learning_rate": 1.2171469764345807e-05,
+      "loss": 0.8544,
+      "step": 1292
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.9176123781134722,
+      "learning_rate": 1.216039750061767e-05,
+      "loss": 0.8013,
+      "step": 1293
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.8972025672926027,
+      "learning_rate": 1.2149322457983719e-05,
+      "loss": 0.7344,
+      "step": 1294
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.9696364696159007,
+      "learning_rate": 1.2138244650689714e-05,
+      "loss": 0.8064,
+      "step": 1295
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.010581794626841,
+      "learning_rate": 1.2127164092984976e-05,
+      "loss": 0.8848,
+      "step": 1296
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.9026898328330966,
+      "learning_rate": 1.2116080799122349e-05,
+      "loss": 0.7996,
+      "step": 1297
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.1285869333851108,
+      "learning_rate": 1.2104994783358206e-05,
+      "loss": 0.9334,
+      "step": 1298
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.819826433384388,
+      "learning_rate": 1.2093906059952425e-05,
+      "loss": 0.7214,
+      "step": 1299
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.017003481795163,
+      "learning_rate": 1.2082814643168357e-05,
+      "loss": 0.7696,
+      "step": 1300
+    },
+    {
+      "epoch": 0.91,
+      "eval_loss": 0.795089840888977,
+      "eval_runtime": 187.5096,
+      "eval_samples_per_second": 215.733,
+      "eval_steps_per_second": 26.969,
+      "step": 1300
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.8798817267157177,
+      "learning_rate": 1.2071720547272823e-05,
+      "loss": 0.7564,
+      "step": 1301
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.90666438626132,
+      "learning_rate": 1.2060623786536086e-05,
+      "loss": 0.8346,
+      "step": 1302
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.8789513496907703,
+      "learning_rate": 1.2049524375231848e-05,
+      "loss": 0.817,
+      "step": 1303
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.9112276995271548,
+      "learning_rate": 1.2038422327637208e-05,
+      "loss": 0.8376,
+      "step": 1304
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.9175604567901359,
+      "learning_rate": 1.2027317658032655e-05,
+      "loss": 0.8326,
+      "step": 1305
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.8946435401851419,
+      "learning_rate": 1.2016210380702068e-05,
+      "loss": 0.7805,
+      "step": 1306
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.9434366572160074,
+      "learning_rate": 1.2005100509932662e-05,
+      "loss": 0.7713,
+      "step": 1307
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.9641160978432078,
+      "learning_rate": 1.199398806001499e-05,
+      "loss": 0.8679,
+      "step": 1308
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.9441666498195864,
+      "learning_rate": 1.1982873045242933e-05,
+      "loss": 0.8775,
+      "step": 1309
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.8583055752214402,
+      "learning_rate": 1.1971755479913665e-05,
+      "loss": 0.8123,
+      "step": 1310
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.9304765536666078,
+      "learning_rate": 1.1960635378327636e-05,
+      "loss": 0.7299,
+      "step": 1311
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.8234687024912583,
+      "learning_rate": 1.194951275478857e-05,
+      "loss": 0.7957,
+      "step": 1312
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.8546894141514437,
+      "learning_rate": 1.1938387623603425e-05,
+      "loss": 0.8257,
+      "step": 1313
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.8632822306104739,
+      "learning_rate": 1.1927259999082386e-05,
+      "loss": 0.7314,
+      "step": 1314
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.896871727091561,
+      "learning_rate": 1.1916129895538847e-05,
+      "loss": 0.7087,
+      "step": 1315
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.9393014655617864,
+      "learning_rate": 1.1904997327289391e-05,
+      "loss": 0.728,
+      "step": 1316
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.8769371873165606,
+      "learning_rate": 1.1893862308653777e-05,
+      "loss": 0.736,
+      "step": 1317
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.8852596329458032,
+      "learning_rate": 1.1882724853954902e-05,
+      "loss": 0.7668,
+      "step": 1318
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.8617739696699027,
+      "learning_rate": 1.1871584977518806e-05,
+      "loss": 0.7659,
+      "step": 1319
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.9314434034383864,
+      "learning_rate": 1.1860442693674648e-05,
+      "loss": 0.8394,
+      "step": 1320
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.861142003308729,
+      "learning_rate": 1.1849298016754669e-05,
+      "loss": 0.7768,
+      "step": 1321
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.8769070771121391,
+      "learning_rate": 1.1838150961094204e-05,
+      "loss": 0.8732,
+      "step": 1322
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.8603644023677578,
+      "learning_rate": 1.182700154103164e-05,
+      "loss": 0.6549,
+      "step": 1323
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.9638201294806444,
+      "learning_rate": 1.1815849770908403e-05,
+      "loss": 0.8464,
+      "step": 1324
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.8306462320605537,
+      "learning_rate": 1.180469566506895e-05,
+      "loss": 0.7991,
+      "step": 1325
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.9863044155217197,
+      "learning_rate": 1.1793539237860738e-05,
+      "loss": 0.8572,
+      "step": 1326
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.9702494386049709,
+      "learning_rate": 1.1782380503634203e-05,
+      "loss": 1.0178,
+      "step": 1327
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.8563480076569994,
+      "learning_rate": 1.177121947674276e-05,
+      "loss": 0.8421,
+      "step": 1328
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.9823180165930772,
+      "learning_rate": 1.1760056171542773e-05,
+      "loss": 0.821,
+      "step": 1329
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.906116648047096,
+      "learning_rate": 1.1748890602393521e-05,
+      "loss": 0.7529,
+      "step": 1330
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.8593895588191588,
+      "learning_rate": 1.1737722783657213e-05,
+      "loss": 0.8545,
+      "step": 1331
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.8499393110247033,
+      "learning_rate": 1.172655272969894e-05,
+      "loss": 0.8143,
+      "step": 1332
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.9164004760981641,
+      "learning_rate": 1.171538045488668e-05,
+      "loss": 0.78,
+      "step": 1333
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.9122326263339436,
+      "learning_rate": 1.170420597359125e-05,
+      "loss": 0.8505,
+      "step": 1334
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.9549519243654508,
+      "learning_rate": 1.1693029300186324e-05,
+      "loss": 0.7455,
+      "step": 1335
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.0516468035890543,
+      "learning_rate": 1.1681850449048387e-05,
+      "loss": 0.8577,
+      "step": 1336
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.8758025212436614,
+      "learning_rate": 1.1670669434556716e-05,
+      "loss": 0.8231,
+      "step": 1337
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.8884706110774874,
+      "learning_rate": 1.1659486271093392e-05,
+      "loss": 0.8108,
+      "step": 1338
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.9218324642854692,
+      "learning_rate": 1.1648300973043245e-05,
+      "loss": 0.7995,
+      "step": 1339
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.8654985510951089,
+      "learning_rate": 1.1637113554793846e-05,
+      "loss": 0.7501,
+      "step": 1340
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.8380612233059123,
+      "learning_rate": 1.1625924030735511e-05,
+      "loss": 0.686,
+      "step": 1341
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.8848563947968523,
+      "learning_rate": 1.161473241526125e-05,
+      "loss": 0.8025,
+      "step": 1342
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.8821522053418632,
+      "learning_rate": 1.1603538722766767e-05,
+      "loss": 0.7701,
+      "step": 1343
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.8692396513709321,
+      "learning_rate": 1.1592342967650445e-05,
+      "loss": 0.831,
+      "step": 1344
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.8726825340572187,
+      "learning_rate": 1.1581145164313307e-05,
+      "loss": 0.8207,
+      "step": 1345
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.9591798533346475,
+      "learning_rate": 1.1569945327159023e-05,
+      "loss": 0.8628,
+      "step": 1346
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.9596109009779479,
+      "learning_rate": 1.155874347059387e-05,
+      "loss": 0.814,
+      "step": 1347
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.9027226890146384,
+      "learning_rate": 1.1547539609026726e-05,
+      "loss": 0.8522,
+      "step": 1348
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.8971455922709491,
+      "learning_rate": 1.153633375686905e-05,
+      "loss": 0.7769,
+      "step": 1349
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.9048377154051395,
+      "learning_rate": 1.152512592853486e-05,
+      "loss": 0.7424,
+      "step": 1350
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.8397557960034163,
+      "learning_rate": 1.1513916138440712e-05,
+      "loss": 0.7782,
+      "step": 1351
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.8503308719220157,
+      "learning_rate": 1.1502704401005694e-05,
+      "loss": 0.7616,
+      "step": 1352
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.8467495732273495,
+      "learning_rate": 1.1491490730651389e-05,
+      "loss": 0.7797,
+      "step": 1353
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.899994053064031,
+      "learning_rate": 1.1480275141801873e-05,
+      "loss": 0.7726,
+      "step": 1354
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.8583068406846777,
+      "learning_rate": 1.146905764888369e-05,
+      "loss": 0.7654,
+      "step": 1355
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.9301024028790893,
+      "learning_rate": 1.1457838266325828e-05,
+      "loss": 0.7648,
+      "step": 1356
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.8523423465764791,
+      "learning_rate": 1.1446617008559713e-05,
+      "loss": 0.7613,
+      "step": 1357
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.9179001752750603,
+      "learning_rate": 1.1435393890019173e-05,
+      "loss": 0.7407,
+      "step": 1358
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.8807224051681771,
+      "learning_rate": 1.1424168925140438e-05,
+      "loss": 0.7672,
+      "step": 1359
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.9213553252352924,
+      "learning_rate": 1.141294212836211e-05,
+      "loss": 0.7772,
+      "step": 1360
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.9143784577248558,
+      "learning_rate": 1.1401713514125152e-05,
+      "loss": 0.7812,
+      "step": 1361
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.8653978990179322,
+      "learning_rate": 1.1390483096872851e-05,
+      "loss": 0.8557,
+      "step": 1362
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.0061464612362694,
+      "learning_rate": 1.137925089105083e-05,
+      "loss": 0.926,
+      "step": 1363
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.1393025054238606,
+      "learning_rate": 1.1368016911107004e-05,
+      "loss": 0.9341,
+      "step": 1364
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.9080538249481118,
+      "learning_rate": 1.1356781171491569e-05,
+      "loss": 0.8169,
+      "step": 1365
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.8933884113869166,
+      "learning_rate": 1.1345543686656989e-05,
+      "loss": 0.8321,
+      "step": 1366
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.8700460854854123,
+      "learning_rate": 1.1334304471057968e-05,
+      "loss": 0.7425,
+      "step": 1367
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.9531764227643872,
+      "learning_rate": 1.1323063539151442e-05,
+      "loss": 0.8056,
+      "step": 1368
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.902921855333192,
+      "learning_rate": 1.1311820905396547e-05,
+      "loss": 0.878,
+      "step": 1369
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.8272338173452627,
+      "learning_rate": 1.1300576584254617e-05,
+      "loss": 0.7549,
+      "step": 1370
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.8555435739973453,
+      "learning_rate": 1.1289330590189152e-05,
+      "loss": 0.8163,
+      "step": 1371
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.895793435047801,
+      "learning_rate": 1.12780829376658e-05,
+      "loss": 0.8736,
+      "step": 1372
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.8643206423346252,
+      "learning_rate": 1.126683364115235e-05,
+      "loss": 0.813,
+      "step": 1373
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.9454856642959599,
+      "learning_rate": 1.1255582715118701e-05,
+      "loss": 0.7898,
+      "step": 1374
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.8681278075027504,
+      "learning_rate": 1.1244330174036843e-05,
+      "loss": 0.6867,
+      "step": 1375
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.9328868275202035,
+      "learning_rate": 1.123307603238086e-05,
+      "loss": 0.824,
+      "step": 1376
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.9073248755158357,
+      "learning_rate": 1.1221820304626878e-05,
+      "loss": 0.8881,
+      "step": 1377
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.9364571025151188,
+      "learning_rate": 1.1210563005253067e-05,
+      "loss": 0.8359,
+      "step": 1378
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.9404626028617894,
+      "learning_rate": 1.1199304148739627e-05,
+      "loss": 0.8087,
+      "step": 1379
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.260358311619401,
+      "learning_rate": 1.1188043749568752e-05,
+      "loss": 0.7809,
+      "step": 1380
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.9276597929451739,
+      "learning_rate": 1.117678182222462e-05,
+      "loss": 0.8202,
+      "step": 1381
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.8692262060914322,
+      "learning_rate": 1.1165518381193379e-05,
+      "loss": 0.7998,
+      "step": 1382
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.9063084872578175,
+      "learning_rate": 1.115425344096313e-05,
+      "loss": 0.8227,
+      "step": 1383
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.9570539636008829,
+      "learning_rate": 1.1142987016023887e-05,
+      "loss": 0.7911,
+      "step": 1384
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.9148160271160868,
+      "learning_rate": 1.1131719120867584e-05,
+      "loss": 0.8319,
+      "step": 1385
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.8706583871071989,
+      "learning_rate": 1.1120449769988043e-05,
+      "loss": 0.778,
+      "step": 1386
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.8691907155679381,
+      "learning_rate": 1.1109178977880966e-05,
+      "loss": 0.7964,
+      "step": 1387
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.8493717758867598,
+      "learning_rate": 1.1097906759043896e-05,
+      "loss": 0.6965,
+      "step": 1388
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.9992890399936691,
+      "learning_rate": 1.108663312797622e-05,
+      "loss": 0.834,
+      "step": 1389
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.8756559048622335,
+      "learning_rate": 1.1075358099179136e-05,
+      "loss": 0.7885,
+      "step": 1390
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.8429751025054221,
+      "learning_rate": 1.1064081687155647e-05,
+      "loss": 0.7318,
+      "step": 1391
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.8655765925457657,
+      "learning_rate": 1.1052803906410531e-05,
+      "loss": 0.7906,
+      "step": 1392
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.8853165227750313,
+      "learning_rate": 1.1041524771450329e-05,
+      "loss": 0.8022,
+      "step": 1393
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.8899368333997295,
+      "learning_rate": 1.1030244296783316e-05,
+      "loss": 0.8456,
+      "step": 1394
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.8465449912567415,
+      "learning_rate": 1.10189624969195e-05,
+      "loss": 0.7837,
+      "step": 1395
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.9150904249953385,
+      "learning_rate": 1.1007679386370591e-05,
+      "loss": 0.8145,
+      "step": 1396
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.9457916211995286,
+      "learning_rate": 1.0996394979649977e-05,
+      "loss": 0.8615,
+      "step": 1397
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.823661666782617,
+      "learning_rate": 1.0985109291272729e-05,
+      "loss": 0.7733,
+      "step": 1398
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.0025967815723742,
+      "learning_rate": 1.0973822335755551e-05,
+      "loss": 0.7727,
+      "step": 1399
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.0641070270245732,
+      "learning_rate": 1.0962534127616784e-05,
+      "loss": 0.8476,
+      "step": 1400
+    },
+    {
+      "epoch": 0.98,
+      "eval_loss": 0.7878583073616028,
+      "eval_runtime": 187.2349,
+      "eval_samples_per_second": 216.049,
+      "eval_steps_per_second": 27.009,
+      "step": 1400
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 2870,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 100,
+  "total_flos": 1.6525916175020851e+19,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}