diff --git "a/checkpoint-654/trainer_state.json" "b/checkpoint-654/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-654/trainer_state.json"
@@ -0,0 +1,4611 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.951219512195122,
+  "eval_steps": 500,
+  "global_step": 654,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.009146341463414634,
+      "grad_norm": 40.05127716064453,
+      "learning_rate": 5.0000000000000004e-08,
+      "loss": 5.1282,
+      "step": 1
+    },
+    {
+      "epoch": 0.018292682926829267,
+      "grad_norm": 39.01039505004883,
+      "learning_rate": 1.0000000000000001e-07,
+      "loss": 5.0586,
+      "step": 2
+    },
+    {
+      "epoch": 0.027439024390243903,
+      "grad_norm": 40.20766067504883,
+      "learning_rate": 1.5000000000000002e-07,
+      "loss": 5.0311,
+      "step": 3
+    },
+    {
+      "epoch": 0.036585365853658534,
+      "grad_norm": 38.70350646972656,
+      "learning_rate": 2.0000000000000002e-07,
+      "loss": 5.1598,
+      "step": 4
+    },
+    {
+      "epoch": 0.04573170731707317,
+      "grad_norm": 39.230979919433594,
+      "learning_rate": 2.5000000000000004e-07,
+      "loss": 5.0812,
+      "step": 5
+    },
+    {
+      "epoch": 0.054878048780487805,
+      "grad_norm": 39.45293045043945,
+      "learning_rate": 3.0000000000000004e-07,
+      "loss": 5.0841,
+      "step": 6
+    },
+    {
+      "epoch": 0.06402439024390244,
+      "grad_norm": 38.92170333862305,
+      "learning_rate": 3.5000000000000004e-07,
+      "loss": 5.022,
+      "step": 7
+    },
+    {
+      "epoch": 0.07317073170731707,
+      "grad_norm": 38.932369232177734,
+      "learning_rate": 4.0000000000000003e-07,
+      "loss": 5.0876,
+      "step": 8
+    },
+    {
+      "epoch": 0.08231707317073171,
+      "grad_norm": 39.60245132446289,
+      "learning_rate": 4.5000000000000003e-07,
+      "loss": 5.0798,
+      "step": 9
+    },
+    {
+      "epoch": 0.09146341463414634,
+      "grad_norm": 39.53607177734375,
+      "learning_rate": 5.000000000000001e-07,
+      "loss": 5.0105,
+      "step": 10
+    },
+    {
+      "epoch": 0.10060975609756098,
+      "grad_norm": 39.50700378417969,
+      "learning_rate": 5.5e-07,
+      "loss": 4.9951,
+      "step": 11
+    },
+    {
+      "epoch": 0.10975609756097561,
+      "grad_norm": 39.52376174926758,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 4.9163,
+      "step": 12
+    },
+    {
+      "epoch": 0.11890243902439024,
+      "grad_norm": 39.15214157104492,
+      "learning_rate": 6.5e-07,
+      "loss": 4.9634,
+      "step": 13
+    },
+    {
+      "epoch": 0.12804878048780488,
+      "grad_norm": 37.795066833496094,
+      "learning_rate": 7.000000000000001e-07,
+      "loss": 4.819,
+      "step": 14
+    },
+    {
+      "epoch": 0.13719512195121952,
+      "grad_norm": 38.23014450073242,
+      "learning_rate": 7.5e-07,
+      "loss": 4.8756,
+      "step": 15
+    },
+    {
+      "epoch": 0.14634146341463414,
+      "grad_norm": 38.57551574707031,
+      "learning_rate": 8.000000000000001e-07,
+      "loss": 4.81,
+      "step": 16
+    },
+    {
+      "epoch": 0.15548780487804878,
+      "grad_norm": 36.06121063232422,
+      "learning_rate": 8.500000000000001e-07,
+      "loss": 4.7901,
+      "step": 17
+    },
+    {
+      "epoch": 0.16463414634146342,
+      "grad_norm": 34.38008117675781,
+      "learning_rate": 9.000000000000001e-07,
+      "loss": 4.5775,
+      "step": 18
+    },
+    {
+      "epoch": 0.17378048780487804,
+      "grad_norm": 32.60890579223633,
+      "learning_rate": 9.500000000000001e-07,
+      "loss": 4.3737,
+      "step": 19
+    },
+    {
+      "epoch": 0.18292682926829268,
+      "grad_norm": 31.983110427856445,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 4.3773,
+      "step": 20
+    },
+    {
+      "epoch": 0.19207317073170732,
+      "grad_norm": 27.69571876525879,
+      "learning_rate": 1.0500000000000001e-06,
+      "loss": 4.1276,
+      "step": 21
+    },
+    {
+      "epoch": 0.20121951219512196,
+      "grad_norm": 26.542314529418945,
+      "learning_rate": 1.1e-06,
+      "loss": 4.0641,
+      "step": 22
+    },
+    {
+      "epoch": 0.21036585365853658,
+      "grad_norm": 23.26603126525879,
+      "learning_rate": 1.1500000000000002e-06,
+      "loss": 4.0197,
+      "step": 23
+    },
+    {
+      "epoch": 0.21951219512195122,
+      "grad_norm": 21.04385757446289,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 4.0179,
+      "step": 24
+    },
+    {
+      "epoch": 0.22865853658536586,
+      "grad_norm": 19.790103912353516,
+      "learning_rate": 1.25e-06,
+      "loss": 3.8334,
+      "step": 25
+    },
+    {
+      "epoch": 0.23780487804878048,
+      "grad_norm": 18.2396183013916,
+      "learning_rate": 1.3e-06,
+      "loss": 3.7027,
+      "step": 26
+    },
+    {
+      "epoch": 0.24695121951219512,
+      "grad_norm": 18.725860595703125,
+      "learning_rate": 1.3500000000000002e-06,
+      "loss": 3.7268,
+      "step": 27
+    },
+    {
+      "epoch": 0.25609756097560976,
+      "grad_norm": 18.91561508178711,
+      "learning_rate": 1.4000000000000001e-06,
+      "loss": 3.52,
+      "step": 28
+    },
+    {
+      "epoch": 0.2652439024390244,
+      "grad_norm": 18.196651458740234,
+      "learning_rate": 1.45e-06,
+      "loss": 3.4765,
+      "step": 29
+    },
+    {
+      "epoch": 0.27439024390243905,
+      "grad_norm": 17.778606414794922,
+      "learning_rate": 1.5e-06,
+      "loss": 3.3573,
+      "step": 30
+    },
+    {
+      "epoch": 0.28353658536585363,
+      "grad_norm": 16.23012924194336,
+      "learning_rate": 1.5500000000000002e-06,
+      "loss": 3.2783,
+      "step": 31
+    },
+    {
+      "epoch": 0.2926829268292683,
+      "grad_norm": 15.836565017700195,
+      "learning_rate": 1.6000000000000001e-06,
+      "loss": 3.2319,
+      "step": 32
+    },
+    {
+      "epoch": 0.3018292682926829,
+      "grad_norm": 11.919289588928223,
+      "learning_rate": 1.6500000000000003e-06,
+      "loss": 3.1345,
+      "step": 33
+    },
+    {
+      "epoch": 0.31097560975609756,
+      "grad_norm": 11.086843490600586,
+      "learning_rate": 1.7000000000000002e-06,
+      "loss": 3.1153,
+      "step": 34
+    },
+    {
+      "epoch": 0.3201219512195122,
+      "grad_norm": 7.563486576080322,
+      "learning_rate": 1.75e-06,
+      "loss": 2.9784,
+      "step": 35
+    },
+    {
+      "epoch": 0.32926829268292684,
+      "grad_norm": 7.60545539855957,
+      "learning_rate": 1.8000000000000001e-06,
+      "loss": 2.8568,
+      "step": 36
+    },
+    {
+      "epoch": 0.3384146341463415,
+      "grad_norm": 7.071218013763428,
+      "learning_rate": 1.85e-06,
+      "loss": 2.8879,
+      "step": 37
+    },
+    {
+      "epoch": 0.3475609756097561,
+      "grad_norm": 8.200672149658203,
+      "learning_rate": 1.9000000000000002e-06,
+      "loss": 2.7437,
+      "step": 38
+    },
+    {
+      "epoch": 0.3567073170731707,
+      "grad_norm": 12.079374313354492,
+      "learning_rate": 1.9500000000000004e-06,
+      "loss": 2.7589,
+      "step": 39
+    },
+    {
+      "epoch": 0.36585365853658536,
+      "grad_norm": 12.21938705444336,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 2.7098,
+      "step": 40
+    },
+    {
+      "epoch": 0.375,
+      "grad_norm": 10.224228858947754,
+      "learning_rate": 2.05e-06,
+      "loss": 2.6803,
+      "step": 41
+    },
+    {
+      "epoch": 0.38414634146341464,
+      "grad_norm": 10.054616928100586,
+      "learning_rate": 2.1000000000000002e-06,
+      "loss": 2.4658,
+      "step": 42
+    },
+    {
+      "epoch": 0.3932926829268293,
+      "grad_norm": 8.074049949645996,
+      "learning_rate": 2.15e-06,
+      "loss": 2.6341,
+      "step": 43
+    },
+    {
+      "epoch": 0.4024390243902439,
+      "grad_norm": 5.721987724304199,
+      "learning_rate": 2.2e-06,
+      "loss": 2.3848,
+      "step": 44
+    },
+    {
+      "epoch": 0.4115853658536585,
+      "grad_norm": 4.646566390991211,
+      "learning_rate": 2.25e-06,
+      "loss": 2.4323,
+      "step": 45
+    },
+    {
+      "epoch": 0.42073170731707316,
+      "grad_norm": 4.591104984283447,
+      "learning_rate": 2.3000000000000004e-06,
+      "loss": 2.4022,
+      "step": 46
+    },
+    {
+      "epoch": 0.4298780487804878,
+      "grad_norm": 4.822715759277344,
+      "learning_rate": 2.35e-06,
+      "loss": 2.2768,
+      "step": 47
+    },
+    {
+      "epoch": 0.43902439024390244,
+      "grad_norm": 5.176568984985352,
+      "learning_rate": 2.4000000000000003e-06,
+      "loss": 2.281,
+      "step": 48
+    },
+    {
+      "epoch": 0.4481707317073171,
+      "grad_norm": 4.731242656707764,
+      "learning_rate": 2.4500000000000003e-06,
+      "loss": 2.235,
+      "step": 49
+    },
+    {
+      "epoch": 0.4573170731707317,
+      "grad_norm": 4.6343865394592285,
+      "learning_rate": 2.5e-06,
+      "loss": 2.1465,
+      "step": 50
+    },
+    {
+      "epoch": 0.46646341463414637,
+      "grad_norm": 4.669382095336914,
+      "learning_rate": 2.55e-06,
+      "loss": 2.0969,
+      "step": 51
+    },
+    {
+      "epoch": 0.47560975609756095,
+      "grad_norm": 4.207769870758057,
+      "learning_rate": 2.6e-06,
+      "loss": 2.0308,
+      "step": 52
+    },
+    {
+      "epoch": 0.4847560975609756,
+      "grad_norm": 5.222470760345459,
+      "learning_rate": 2.6500000000000005e-06,
+      "loss": 1.9445,
+      "step": 53
+    },
+    {
+      "epoch": 0.49390243902439024,
+      "grad_norm": 5.10925817489624,
+      "learning_rate": 2.7000000000000004e-06,
+      "loss": 1.7763,
+      "step": 54
+    },
+    {
+      "epoch": 0.5030487804878049,
+      "grad_norm": 4.8151044845581055,
+      "learning_rate": 2.7500000000000004e-06,
+      "loss": 1.7661,
+      "step": 55
+    },
+    {
+      "epoch": 0.5121951219512195,
+      "grad_norm": 6.155700206756592,
+      "learning_rate": 2.8000000000000003e-06,
+      "loss": 1.6665,
+      "step": 56
+    },
+    {
+      "epoch": 0.5213414634146342,
+      "grad_norm": 4.719021320343018,
+      "learning_rate": 2.85e-06,
+      "loss": 1.5532,
+      "step": 57
+    },
+    {
+      "epoch": 0.5304878048780488,
+      "grad_norm": 4.591799259185791,
+      "learning_rate": 2.9e-06,
+      "loss": 1.5244,
+      "step": 58
+    },
+    {
+      "epoch": 0.5396341463414634,
+      "grad_norm": 5.483795642852783,
+      "learning_rate": 2.95e-06,
+      "loss": 1.3816,
+      "step": 59
+    },
+    {
+      "epoch": 0.5487804878048781,
+      "grad_norm": 4.32553768157959,
+      "learning_rate": 3e-06,
+      "loss": 1.4268,
+      "step": 60
+    },
+    {
+      "epoch": 0.5579268292682927,
+      "grad_norm": 8.296231269836426,
+      "learning_rate": 3.05e-06,
+      "loss": 1.2887,
+      "step": 61
+    },
+    {
+      "epoch": 0.5670731707317073,
+      "grad_norm": 5.07594108581543,
+      "learning_rate": 3.1000000000000004e-06,
+      "loss": 1.228,
+      "step": 62
+    },
+    {
+      "epoch": 0.5762195121951219,
+      "grad_norm": 8.427435874938965,
+      "learning_rate": 3.1500000000000003e-06,
+      "loss": 1.1724,
+      "step": 63
+    },
+    {
+      "epoch": 0.5853658536585366,
+      "grad_norm": 22.45092010498047,
+      "learning_rate": 3.2000000000000003e-06,
+      "loss": 1.0871,
+      "step": 64
+    },
+    {
+      "epoch": 0.5945121951219512,
+      "grad_norm": 20.67412567138672,
+      "learning_rate": 3.2500000000000002e-06,
+      "loss": 1.0108,
+      "step": 65
+    },
+    {
+      "epoch": 0.6036585365853658,
+      "grad_norm": 4.921634674072266,
+      "learning_rate": 3.3000000000000006e-06,
+      "loss": 0.9775,
+      "step": 66
+    },
+    {
+      "epoch": 0.6128048780487805,
+      "grad_norm": 4.477286338806152,
+      "learning_rate": 3.3500000000000005e-06,
+      "loss": 0.9554,
+      "step": 67
+    },
+    {
+      "epoch": 0.6219512195121951,
+      "grad_norm": 8.190194129943848,
+      "learning_rate": 3.4000000000000005e-06,
+      "loss": 0.8583,
+      "step": 68
+    },
+    {
+      "epoch": 0.6310975609756098,
+      "grad_norm": 29.21201515197754,
+      "learning_rate": 3.45e-06,
+      "loss": 0.8499,
+      "step": 69
+    },
+    {
+      "epoch": 0.6402439024390244,
+      "grad_norm": 33.81047439575195,
+      "learning_rate": 3.5e-06,
+      "loss": 0.858,
+      "step": 70
+    },
+    {
+      "epoch": 0.649390243902439,
+      "grad_norm": 3.5125606060028076,
+      "learning_rate": 3.5500000000000003e-06,
+      "loss": 0.7227,
+      "step": 71
+    },
+    {
+      "epoch": 0.6585365853658537,
+      "grad_norm": 18.493423461914062,
+      "learning_rate": 3.6000000000000003e-06,
+      "loss": 0.7447,
+      "step": 72
+    },
+    {
+      "epoch": 0.6676829268292683,
+      "grad_norm": 24.923927307128906,
+      "learning_rate": 3.65e-06,
+      "loss": 0.8423,
+      "step": 73
+    },
+    {
+      "epoch": 0.676829268292683,
+      "grad_norm": 9.870631217956543,
+      "learning_rate": 3.7e-06,
+      "loss": 0.6934,
+      "step": 74
+    },
+    {
+      "epoch": 0.6859756097560976,
+      "grad_norm": 20.598569869995117,
+      "learning_rate": 3.7500000000000005e-06,
+      "loss": 0.7005,
+      "step": 75
+    },
+    {
+      "epoch": 0.6951219512195121,
+      "grad_norm": 30.662466049194336,
+      "learning_rate": 3.8000000000000005e-06,
+      "loss": 0.6226,
+      "step": 76
+    },
+    {
+      "epoch": 0.7042682926829268,
+      "grad_norm": 28.930774688720703,
+      "learning_rate": 3.85e-06,
+      "loss": 0.6357,
+      "step": 77
+    },
+    {
+      "epoch": 0.7134146341463414,
+      "grad_norm": 7.140467166900635,
+      "learning_rate": 3.900000000000001e-06,
+      "loss": 0.6129,
+      "step": 78
+    },
+    {
+      "epoch": 0.7225609756097561,
+      "grad_norm": 6.322911739349365,
+      "learning_rate": 3.95e-06,
+      "loss": 0.6086,
+      "step": 79
+    },
+    {
+      "epoch": 0.7317073170731707,
+      "grad_norm": 5.404141426086426,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.5986,
+      "step": 80
+    },
+    {
+      "epoch": 0.7408536585365854,
+      "grad_norm": 24.445228576660156,
+      "learning_rate": 4.05e-06,
+      "loss": 0.5932,
+      "step": 81
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 14.275381088256836,
+      "learning_rate": 4.1e-06,
+      "loss": 0.5286,
+      "step": 82
+    },
+    {
+      "epoch": 0.7591463414634146,
+      "grad_norm": 9.77132511138916,
+      "learning_rate": 4.15e-06,
+      "loss": 0.4814,
+      "step": 83
+    },
+    {
+      "epoch": 0.7682926829268293,
+      "grad_norm": 11.043930053710938,
+      "learning_rate": 4.2000000000000004e-06,
+      "loss": 0.4953,
+      "step": 84
+    },
+    {
+      "epoch": 0.7774390243902439,
+      "grad_norm": 3.1333775520324707,
+      "learning_rate": 4.25e-06,
+      "loss": 0.4794,
+      "step": 85
+    },
+    {
+      "epoch": 0.7865853658536586,
+      "grad_norm": 9.50448989868164,
+      "learning_rate": 4.3e-06,
+      "loss": 0.4451,
+      "step": 86
+    },
+    {
+      "epoch": 0.7957317073170732,
+      "grad_norm": 5.867074489593506,
+      "learning_rate": 4.350000000000001e-06,
+      "loss": 0.4224,
+      "step": 87
+    },
+    {
+      "epoch": 0.8048780487804879,
+      "grad_norm": 12.556106567382812,
+      "learning_rate": 4.4e-06,
+      "loss": 0.4224,
+      "step": 88
+    },
+    {
+      "epoch": 0.8140243902439024,
+      "grad_norm": 7.916622638702393,
+      "learning_rate": 4.450000000000001e-06,
+      "loss": 0.4091,
+      "step": 89
+    },
+    {
+      "epoch": 0.823170731707317,
+      "grad_norm": 2.970932722091675,
+      "learning_rate": 4.5e-06,
+      "loss": 0.4181,
+      "step": 90
+    },
+    {
+      "epoch": 0.8323170731707317,
+      "grad_norm": 17.468870162963867,
+      "learning_rate": 4.5500000000000005e-06,
+      "loss": 0.4031,
+      "step": 91
+    },
+    {
+      "epoch": 0.8414634146341463,
+      "grad_norm": 15.861083984375,
+      "learning_rate": 4.600000000000001e-06,
+      "loss": 0.3604,
+      "step": 92
+    },
+    {
+      "epoch": 0.850609756097561,
+      "grad_norm": 6.715958118438721,
+      "learning_rate": 4.65e-06,
+      "loss": 0.3489,
+      "step": 93
+    },
+    {
+      "epoch": 0.8597560975609756,
+      "grad_norm": 5.439844608306885,
+      "learning_rate": 4.7e-06,
+      "loss": 0.3339,
+      "step": 94
+    },
+    {
+      "epoch": 0.8689024390243902,
+      "grad_norm": 4.650434494018555,
+      "learning_rate": 4.75e-06,
+      "loss": 0.3189,
+      "step": 95
+    },
+    {
+      "epoch": 0.8780487804878049,
+      "grad_norm": 15.319528579711914,
+      "learning_rate": 4.800000000000001e-06,
+      "loss": 0.3347,
+      "step": 96
+    },
+    {
+      "epoch": 0.8871951219512195,
+      "grad_norm": 5.151925563812256,
+      "learning_rate": 4.85e-06,
+      "loss": 0.3043,
+      "step": 97
+    },
+    {
+      "epoch": 0.8963414634146342,
+      "grad_norm": 6.351178169250488,
+      "learning_rate": 4.9000000000000005e-06,
+      "loss": 0.2939,
+      "step": 98
+    },
+    {
+      "epoch": 0.9054878048780488,
+      "grad_norm": 3.942911148071289,
+      "learning_rate": 4.95e-06,
+      "loss": 0.2887,
+      "step": 99
+    },
+    {
+      "epoch": 0.9146341463414634,
+      "grad_norm": 3.0885493755340576,
+      "learning_rate": 5e-06,
+      "loss": 0.2992,
+      "step": 100
+    },
+    {
+      "epoch": 0.9237804878048781,
+      "grad_norm": 7.023274898529053,
+      "learning_rate": 4.999959803423605e-06,
+      "loss": 0.2607,
+      "step": 101
+    },
+    {
+      "epoch": 0.9329268292682927,
+      "grad_norm": 8.774216651916504,
+      "learning_rate": 4.9998392149870315e-06,
+      "loss": 0.2674,
+      "step": 102
+    },
+    {
+      "epoch": 0.9420731707317073,
+      "grad_norm": 3.5439791679382324,
+      "learning_rate": 4.999638238568072e-06,
+      "loss": 0.2602,
+      "step": 103
+    },
+    {
+      "epoch": 0.9512195121951219,
+      "grad_norm": 7.669381618499756,
+      "learning_rate": 4.999356880629579e-06,
+      "loss": 0.2718,
+      "step": 104
+    },
+    {
+      "epoch": 0.9603658536585366,
+      "grad_norm": 6.9407854080200195,
+      "learning_rate": 4.998995150219252e-06,
+      "loss": 0.2494,
+      "step": 105
+    },
+    {
+      "epoch": 0.9695121951219512,
+      "grad_norm": 12.144411087036133,
+      "learning_rate": 4.998553058969352e-06,
+      "loss": 0.2297,
+      "step": 106
+    },
+    {
+      "epoch": 0.9786585365853658,
+      "grad_norm": 6.956138610839844,
+      "learning_rate": 4.998030621096322e-06,
+      "loss": 0.2386,
+      "step": 107
+    },
+    {
+      "epoch": 0.9878048780487805,
+      "grad_norm": 8.304037094116211,
+      "learning_rate": 4.997427853400333e-06,
+      "loss": 0.23,
+      "step": 108
+    },
+    {
+      "epoch": 0.9969512195121951,
+      "grad_norm": 7.031365394592285,
+      "learning_rate": 4.996744775264743e-06,
+      "loss": 0.2346,
+      "step": 109
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 7.031365394592285,
+      "learning_rate": 4.9959814086554735e-06,
+      "loss": 0.0739,
+      "step": 110
+    },
+    {
+      "epoch": 1.0091463414634145,
+      "grad_norm": 2.084608793258667,
+      "learning_rate": 4.995137778120306e-06,
+      "loss": 0.2422,
+      "step": 111
+    },
+    {
+      "epoch": 1.0182926829268293,
+      "grad_norm": 38.12245178222656,
+      "learning_rate": 4.994213910788085e-06,
+      "loss": 0.1992,
+      "step": 112
+    },
+    {
+      "epoch": 1.0274390243902438,
+      "grad_norm": 2.817565441131592,
+      "learning_rate": 4.993209836367856e-06,
+      "loss": 0.1882,
+      "step": 113
+    },
+    {
+      "epoch": 1.0365853658536586,
+      "grad_norm": 11.774249076843262,
+      "learning_rate": 4.9921255871479e-06,
+      "loss": 0.2235,
+      "step": 114
+    },
+    {
+      "epoch": 1.045731707317073,
+      "grad_norm": 2.3362019062042236,
+      "learning_rate": 4.9909611979947046e-06,
+      "loss": 0.1875,
+      "step": 115
+    },
+    {
+      "epoch": 1.0548780487804879,
+      "grad_norm": 1.9516818523406982,
+      "learning_rate": 4.989716706351835e-06,
+      "loss": 0.1858,
+      "step": 116
+    },
+    {
+      "epoch": 1.0640243902439024,
+      "grad_norm": 1.5484645366668701,
+      "learning_rate": 4.9883921522387336e-06,
+      "loss": 0.1845,
+      "step": 117
+    },
+    {
+      "epoch": 1.0731707317073171,
+      "grad_norm": 7.430396556854248,
+      "learning_rate": 4.986987578249433e-06,
+      "loss": 0.184,
+      "step": 118
+    },
+    {
+      "epoch": 1.0823170731707317,
+      "grad_norm": 2.0725948810577393,
+      "learning_rate": 4.9855030295511845e-06,
+      "loss": 0.177,
+      "step": 119
+    },
+    {
+      "epoch": 1.0914634146341464,
+      "grad_norm": 4.836520671844482,
+      "learning_rate": 4.98393855388301e-06,
+      "loss": 0.1744,
+      "step": 120
+    },
+    {
+      "epoch": 1.100609756097561,
+      "grad_norm": 5.22581672668457,
+      "learning_rate": 4.982294201554162e-06,
+      "loss": 0.1599,
+      "step": 121
+    },
+    {
+      "epoch": 1.1097560975609757,
+      "grad_norm": 5.406208038330078,
+      "learning_rate": 4.980570025442507e-06,
+      "loss": 0.1645,
+      "step": 122
+    },
+    {
+      "epoch": 1.1189024390243902,
+      "grad_norm": 8.736640930175781,
+      "learning_rate": 4.978766080992827e-06,
+      "loss": 0.1943,
+      "step": 123
+    },
+    {
+      "epoch": 1.1280487804878048,
+      "grad_norm": 9.184198379516602,
+      "learning_rate": 4.976882426215035e-06,
+      "loss": 0.1753,
+      "step": 124
+    },
+    {
+      "epoch": 1.1371951219512195,
+      "grad_norm": 5.317800521850586,
+      "learning_rate": 4.974919121682307e-06,
+      "loss": 0.1897,
+      "step": 125
+    },
+    {
+      "epoch": 1.146341463414634,
+      "grad_norm": 11.803364753723145,
+      "learning_rate": 4.972876230529143e-06,
+      "loss": 0.1516,
+      "step": 126
+    },
+    {
+      "epoch": 1.1554878048780488,
+      "grad_norm": 31.74310302734375,
+      "learning_rate": 4.9707538184493255e-06,
+      "loss": 0.1947,
+      "step": 127
+    },
+    {
+      "epoch": 1.1646341463414633,
+      "grad_norm": 1.956932783126831,
+      "learning_rate": 4.968551953693813e-06,
+      "loss": 0.1708,
+      "step": 128
+    },
+    {
+      "epoch": 1.173780487804878,
+      "grad_norm": 4.64445686340332,
+      "learning_rate": 4.966270707068548e-06,
+      "loss": 0.1711,
+      "step": 129
+    },
+    {
+      "epoch": 1.1829268292682926,
+      "grad_norm": 4.442819595336914,
+      "learning_rate": 4.963910151932171e-06,
+      "loss": 0.1572,
+      "step": 130
+    },
+    {
+      "epoch": 1.1920731707317074,
+      "grad_norm": 1.9702759981155396,
+      "learning_rate": 4.961470364193673e-06,
+      "loss": 0.1364,
+      "step": 131
+    },
+    {
+      "epoch": 1.201219512195122,
+      "grad_norm": 7.816587448120117,
+      "learning_rate": 4.9589514223099425e-06,
+      "loss": 0.1624,
+      "step": 132
+    },
+    {
+      "epoch": 1.2103658536585367,
+      "grad_norm": 61.18424606323242,
+      "learning_rate": 4.956353407283252e-06,
+      "loss": 0.26,
+      "step": 133
+    },
+    {
+      "epoch": 1.2195121951219512,
+      "grad_norm": 15.649847984313965,
+      "learning_rate": 4.9536764026586505e-06,
+      "loss": 0.1543,
+      "step": 134
+    },
+    {
+      "epoch": 1.228658536585366,
+      "grad_norm": 2.249363660812378,
+      "learning_rate": 4.950920494521274e-06,
+      "loss": 0.152,
+      "step": 135
+    },
+    {
+      "epoch": 1.2378048780487805,
+      "grad_norm": 4.055661678314209,
+      "learning_rate": 4.948085771493579e-06,
+      "loss": 0.1402,
+      "step": 136
+    },
+    {
+      "epoch": 1.2469512195121952,
+      "grad_norm": 2.2803292274475098,
+      "learning_rate": 4.945172324732496e-06,
+      "loss": 0.1427,
+      "step": 137
+    },
+    {
+      "epoch": 1.2560975609756098,
+      "grad_norm": 2.5540456771850586,
+      "learning_rate": 4.942180247926492e-06,
+      "loss": 0.143,
+      "step": 138
+    },
+    {
+      "epoch": 1.2652439024390243,
+      "grad_norm": 8.80340576171875,
+      "learning_rate": 4.939109637292563e-06,
+      "loss": 0.1569,
+      "step": 139
+    },
+    {
+      "epoch": 1.274390243902439,
+      "grad_norm": 3.1077523231506348,
+      "learning_rate": 4.935960591573136e-06,
+      "loss": 0.1313,
+      "step": 140
+    },
+    {
+      "epoch": 1.2835365853658536,
+      "grad_norm": 1.2506815195083618,
+      "learning_rate": 4.932733212032896e-06,
+      "loss": 0.137,
+      "step": 141
+    },
+    {
+      "epoch": 1.2926829268292683,
+      "grad_norm": 2.4397666454315186,
+      "learning_rate": 4.929427602455532e-06,
+      "loss": 0.1408,
+      "step": 142
+    },
+    {
+      "epoch": 1.3018292682926829,
+      "grad_norm": 1.7658663988113403,
+      "learning_rate": 4.926043869140392e-06,
+      "loss": 0.133,
+      "step": 143
+    },
+    {
+      "epoch": 1.3109756097560976,
+      "grad_norm": 1.0473108291625977,
+      "learning_rate": 4.922582120899072e-06,
+      "loss": 0.1358,
+      "step": 144
+    },
+    {
+      "epoch": 1.3201219512195121,
+      "grad_norm": 5.566607475280762,
+      "learning_rate": 4.919042469051916e-06,
+      "loss": 0.1237,
+      "step": 145
+    },
+    {
+      "epoch": 1.329268292682927,
+      "grad_norm": 1.0360139608383179,
+      "learning_rate": 4.915425027424429e-06,
+      "loss": 0.1274,
+      "step": 146
+    },
+    {
+      "epoch": 1.3384146341463414,
+      "grad_norm": 1.9091176986694336,
+      "learning_rate": 4.911729912343631e-06,
+      "loss": 0.1439,
+      "step": 147
+    },
+    {
+      "epoch": 1.3475609756097562,
+      "grad_norm": 1.5727638006210327,
+      "learning_rate": 4.907957242634299e-06,
+      "loss": 0.1259,
+      "step": 148
+    },
+    {
+      "epoch": 1.3567073170731707,
+      "grad_norm": 1.0574792623519897,
+      "learning_rate": 4.9041071396151585e-06,
+      "loss": 0.142,
+      "step": 149
+    },
+    {
+      "epoch": 1.3658536585365852,
+      "grad_norm": 8.020130157470703,
+      "learning_rate": 4.900179727094978e-06,
+      "loss": 0.1392,
+      "step": 150
+    },
+    {
+      "epoch": 1.375,
+      "grad_norm": 3.156255006790161,
+      "learning_rate": 4.896175131368587e-06,
+      "loss": 0.152,
+      "step": 151
+    },
+    {
+      "epoch": 1.3841463414634148,
+      "grad_norm": 2.3956990242004395,
+      "learning_rate": 4.892093481212817e-06,
+      "loss": 0.1244,
+      "step": 152
+    },
+    {
+      "epoch": 1.3932926829268293,
+      "grad_norm": 2.6058578491210938,
+      "learning_rate": 4.887934907882357e-06,
+      "loss": 0.1271,
+      "step": 153
+    },
+    {
+      "epoch": 1.4024390243902438,
+      "grad_norm": 1.7173289060592651,
+      "learning_rate": 4.883699545105535e-06,
+      "loss": 0.1257,
+      "step": 154
+    },
+    {
+      "epoch": 1.4115853658536586,
+      "grad_norm": 2.2237367630004883,
+      "learning_rate": 4.8793875290800186e-06,
+      "loss": 0.127,
+      "step": 155
+    },
+    {
+      "epoch": 1.420731707317073,
+      "grad_norm": 6.997623920440674,
+      "learning_rate": 4.874998998468433e-06,
+      "loss": 0.1454,
+      "step": 156
+    },
+    {
+      "epoch": 1.4298780487804879,
+      "grad_norm": 10.645139694213867,
+      "learning_rate": 4.870534094393903e-06,
+      "loss": 0.1402,
+      "step": 157
+    },
+    {
+      "epoch": 1.4390243902439024,
+      "grad_norm": 1.5808072090148926,
+      "learning_rate": 4.865992960435514e-06,
+      "loss": 0.1156,
+      "step": 158
+    },
+    {
+      "epoch": 1.4481707317073171,
+      "grad_norm": 0.990803062915802,
+      "learning_rate": 4.861375742623697e-06,
+      "loss": 0.1286,
+      "step": 159
+    },
+    {
+      "epoch": 1.4573170731707317,
+      "grad_norm": 1.5008662939071655,
+      "learning_rate": 4.856682589435531e-06,
+      "loss": 0.1282,
+      "step": 160
+    },
+    {
+      "epoch": 1.4664634146341464,
+      "grad_norm": 1.772304892539978,
+      "learning_rate": 4.851913651789968e-06,
+      "loss": 0.1308,
+      "step": 161
+    },
+    {
+      "epoch": 1.475609756097561,
+      "grad_norm": 1.201061487197876,
+      "learning_rate": 4.847069083042982e-06,
+      "loss": 0.1198,
+      "step": 162
+    },
+    {
+      "epoch": 1.4847560975609757,
+      "grad_norm": 0.6147828698158264,
+      "learning_rate": 4.8421490389826344e-06,
+      "loss": 0.1138,
+      "step": 163
+    },
+    {
+      "epoch": 1.4939024390243902,
+      "grad_norm": 1.3990256786346436,
+      "learning_rate": 4.837153677824068e-06,
+      "loss": 0.1205,
+      "step": 164
+    },
+    {
+      "epoch": 1.5030487804878048,
+      "grad_norm": 1.6977574825286865,
+      "learning_rate": 4.832083160204414e-06,
+      "loss": 0.1128,
+      "step": 165
+    },
+    {
+      "epoch": 1.5121951219512195,
+      "grad_norm": 2.4056453704833984,
+      "learning_rate": 4.826937649177632e-06,
+      "loss": 0.1184,
+      "step": 166
+    },
+    {
+      "epoch": 1.5213414634146343,
+      "grad_norm": 2.7455694675445557,
+      "learning_rate": 4.821717310209265e-06,
+      "loss": 0.123,
+      "step": 167
+    },
+    {
+      "epoch": 1.5304878048780488,
+      "grad_norm": 1.8243584632873535,
+      "learning_rate": 4.816422311171115e-06,
+      "loss": 0.1198,
+      "step": 168
+    },
+    {
+      "epoch": 1.5396341463414633,
+      "grad_norm": 0.8920577764511108,
+      "learning_rate": 4.811052822335849e-06,
+      "loss": 0.1088,
+      "step": 169
+    },
+    {
+      "epoch": 1.548780487804878,
+      "grad_norm": 1.6639117002487183,
+      "learning_rate": 4.805609016371522e-06,
+      "loss": 0.1231,
+      "step": 170
+    },
+    {
+      "epoch": 1.5579268292682928,
+      "grad_norm": 1.383026123046875,
+      "learning_rate": 4.800091068336024e-06,
+      "loss": 0.1207,
+      "step": 171
+    },
+    {
+      "epoch": 1.5670731707317072,
+      "grad_norm": 1.6599416732788086,
+      "learning_rate": 4.79449915567145e-06,
+      "loss": 0.1206,
+      "step": 172
+    },
+    {
+      "epoch": 1.576219512195122,
+      "grad_norm": 1.5868676900863647,
+      "learning_rate": 4.788833458198395e-06,
+      "loss": 0.1158,
+      "step": 173
+    },
+    {
+      "epoch": 1.5853658536585367,
+      "grad_norm": 1.0154211521148682,
+      "learning_rate": 4.783094158110175e-06,
+      "loss": 0.1091,
+      "step": 174
+    },
+    {
+      "epoch": 1.5945121951219512,
+      "grad_norm": 2.1514394283294678,
+      "learning_rate": 4.7772814399669585e-06,
+      "loss": 0.1109,
+      "step": 175
+    },
+    {
+      "epoch": 1.6036585365853657,
+      "grad_norm": 0.7475085258483887,
+      "learning_rate": 4.7713954906898415e-06,
+      "loss": 0.1162,
+      "step": 176
+    },
+    {
+      "epoch": 1.6128048780487805,
+      "grad_norm": 0.8523163795471191,
+      "learning_rate": 4.7654364995548325e-06,
+      "loss": 0.1238,
+      "step": 177
+    },
+    {
+      "epoch": 1.6219512195121952,
+      "grad_norm": 1.5456233024597168,
+      "learning_rate": 4.759404658186765e-06,
+      "loss": 0.1163,
+      "step": 178
+    },
+    {
+      "epoch": 1.6310975609756098,
+      "grad_norm": 1.123340129852295,
+      "learning_rate": 4.753300160553137e-06,
+      "loss": 0.1104,
+      "step": 179
+    },
+    {
+      "epoch": 1.6402439024390243,
+      "grad_norm": 1.16841721534729,
+      "learning_rate": 4.747123202957871e-06,
+      "loss": 0.1099,
+      "step": 180
+    },
+    {
+      "epoch": 1.649390243902439,
+      "grad_norm": 2.025221109390259,
+      "learning_rate": 4.740873984035009e-06,
+      "loss": 0.1157,
+      "step": 181
+    },
+    {
+      "epoch": 1.6585365853658538,
+      "grad_norm": 3.876469373703003,
+      "learning_rate": 4.734552704742314e-06,
+      "loss": 0.1122,
+      "step": 182
+    },
+    {
+      "epoch": 1.6676829268292683,
+      "grad_norm": 1.5562145709991455,
+      "learning_rate": 4.728159568354814e-06,
+      "loss": 0.1127,
+      "step": 183
+    },
+    {
+      "epoch": 1.6768292682926829,
+      "grad_norm": 1.4691755771636963,
+      "learning_rate": 4.721694780458266e-06,
+      "loss": 0.1182,
+      "step": 184
+    },
+    {
+      "epoch": 1.6859756097560976,
+      "grad_norm": 1.9529973268508911,
+      "learning_rate": 4.715158548942541e-06,
+      "loss": 0.1153,
+      "step": 185
+    },
+    {
+      "epoch": 1.6951219512195121,
+      "grad_norm": 1.9065771102905273,
+      "learning_rate": 4.708551083994944e-06,
+      "loss": 0.1163,
+      "step": 186
+    },
+    {
+      "epoch": 1.7042682926829267,
+      "grad_norm": 1.0964735746383667,
+      "learning_rate": 4.701872598093452e-06,
+      "loss": 0.114,
+      "step": 187
+    },
+    {
+      "epoch": 1.7134146341463414,
+      "grad_norm": 1.0885926485061646,
+      "learning_rate": 4.695123305999877e-06,
+      "loss": 0.114,
+      "step": 188
+    },
+    {
+      "epoch": 1.7225609756097562,
+      "grad_norm": 1.1777565479278564,
+      "learning_rate": 4.688303424752969e-06,
+      "loss": 0.1116,
+      "step": 189
+    },
+    {
+      "epoch": 1.7317073170731707,
+      "grad_norm": 1.3686976432800293,
+      "learning_rate": 4.6814131736614286e-06,
+      "loss": 0.1081,
+      "step": 190
+    },
+    {
+      "epoch": 1.7408536585365852,
+      "grad_norm": 0.8220391869544983,
+      "learning_rate": 4.674452774296861e-06,
+      "loss": 0.1091,
+      "step": 191
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 1.073935866355896,
+      "learning_rate": 4.667422450486646e-06,
+      "loss": 0.1148,
+      "step": 192
+    },
+    {
+      "epoch": 1.7591463414634148,
+      "grad_norm": 1.389614224433899,
+      "learning_rate": 4.660322428306741e-06,
+      "loss": 0.1134,
+      "step": 193
+    },
+    {
+      "epoch": 1.7682926829268293,
+      "grad_norm": 1.349023699760437,
+      "learning_rate": 4.653152936074413e-06,
+      "loss": 0.1056,
+      "step": 194
+    },
+    {
+      "epoch": 1.7774390243902438,
+      "grad_norm": 1.0334906578063965,
+      "learning_rate": 4.645914204340898e-06,
+      "loss": 0.113,
+      "step": 195
+    },
+    {
+      "epoch": 1.7865853658536586,
+      "grad_norm": 18.6312313079834,
+      "learning_rate": 4.638606465883979e-06,
+      "loss": 0.1351,
+      "step": 196
+    },
+    {
+      "epoch": 1.7957317073170733,
+      "grad_norm": 0.7669345736503601,
+      "learning_rate": 4.6312299557005125e-06,
+      "loss": 0.1097,
+      "step": 197
+    },
+    {
+      "epoch": 1.8048780487804879,
+      "grad_norm": 2.1091127395629883,
+      "learning_rate": 4.623784910998862e-06,
+      "loss": 0.1122,
+      "step": 198
+    },
+    {
+      "epoch": 1.8140243902439024,
+      "grad_norm": 0.8031914830207825,
+      "learning_rate": 4.616271571191273e-06,
+      "loss": 0.1109,
+      "step": 199
+    },
+    {
+      "epoch": 1.8231707317073171,
+      "grad_norm": 2.1154661178588867,
+      "learning_rate": 4.608690177886177e-06,
+      "loss": 0.1131,
+      "step": 200
+    },
+    {
+      "epoch": 1.8323170731707317,
+      "grad_norm": 1.4946709871292114,
+      "learning_rate": 4.6010409748804165e-06,
+      "loss": 0.1136,
+      "step": 201
+    },
+    {
+      "epoch": 1.8414634146341462,
+      "grad_norm": 1.0509642362594604,
+      "learning_rate": 4.593324208151412e-06,
+      "loss": 0.1162,
+      "step": 202
+    },
+    {
+      "epoch": 1.850609756097561,
+      "grad_norm": 0.8742370009422302,
+      "learning_rate": 4.585540125849244e-06,
+      "loss": 0.1102,
+      "step": 203
+    },
+    {
+      "epoch": 1.8597560975609757,
+      "grad_norm": 1.1283056735992432,
+      "learning_rate": 4.577688978288681e-06,
+      "loss": 0.1104,
+      "step": 204
+    },
+    {
+      "epoch": 1.8689024390243902,
+      "grad_norm": 1.0081048011779785,
+      "learning_rate": 4.569771017941125e-06,
+      "loss": 0.1119,
+      "step": 205
+    },
+    {
+      "epoch": 1.8780487804878048,
+      "grad_norm": 0.7291656732559204,
+      "learning_rate": 4.561786499426494e-06,
+      "loss": 0.1103,
+      "step": 206
+    },
+    {
+      "epoch": 1.8871951219512195,
+      "grad_norm": 1.144096851348877,
+      "learning_rate": 4.5537356795050346e-06,
+      "loss": 0.1125,
+      "step": 207
+    },
+    {
+      "epoch": 1.8963414634146343,
+      "grad_norm": 0.7317456603050232,
+      "learning_rate": 4.545618817069066e-06,
+      "loss": 0.1105,
+      "step": 208
+    },
+    {
+      "epoch": 1.9054878048780488,
+      "grad_norm": 1.1636486053466797,
+      "learning_rate": 4.537436173134653e-06,
+      "loss": 0.1091,
+      "step": 209
+    },
+    {
+      "epoch": 1.9146341463414633,
+      "grad_norm": 0.6319037675857544,
+      "learning_rate": 4.529188010833212e-06,
+      "loss": 0.1088,
+      "step": 210
+    },
+    {
+      "epoch": 1.923780487804878,
+      "grad_norm": 0.7535589933395386,
+      "learning_rate": 4.520874595403053e-06,
+      "loss": 0.1041,
+      "step": 211
+    },
+    {
+      "epoch": 1.9329268292682928,
+      "grad_norm": 1.0461041927337646,
+      "learning_rate": 4.512496194180846e-06,
+      "loss": 0.1129,
+      "step": 212
+    },
+    {
+      "epoch": 1.9420731707317072,
+      "grad_norm": 0.6124874949455261,
+      "learning_rate": 4.5040530765930275e-06,
+      "loss": 0.1113,
+      "step": 213
+    },
+    {
+      "epoch": 1.951219512195122,
+      "grad_norm": 0.8516119122505188,
+      "learning_rate": 4.495545514147134e-06,
+      "loss": 0.1112,
+      "step": 214
+    },
+    {
+      "epoch": 1.9603658536585367,
+      "grad_norm": 0.8919075131416321,
+      "learning_rate": 4.486973780423073e-06,
+      "loss": 0.1089,
+      "step": 215
+    },
+    {
+      "epoch": 1.9695121951219512,
+      "grad_norm": 1.2183221578598022,
+      "learning_rate": 4.478338151064323e-06,
+      "loss": 0.1156,
+      "step": 216
+    },
+    {
+      "epoch": 1.9786585365853657,
+      "grad_norm": 1.0123646259307861,
+      "learning_rate": 4.4696389037690734e-06,
+      "loss": 0.1077,
+      "step": 217
+    },
+    {
+      "epoch": 1.9878048780487805,
+      "grad_norm": 0.8304896950721741,
+      "learning_rate": 4.460876318281291e-06,
+      "loss": 0.1051,
+      "step": 218
+    },
+    {
+      "epoch": 1.9969512195121952,
+      "grad_norm": 0.5879438519477844,
+      "learning_rate": 4.452050676381725e-06,
+      "loss": 0.1064,
+      "step": 219
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.5879438519477844,
+      "learning_rate": 4.443162261878846e-06,
+      "loss": 0.0344,
+      "step": 220
+    },
+    {
+      "epoch": 2.0091463414634148,
+      "grad_norm": 0.8598087430000305,
+      "learning_rate": 4.434211360599721e-06,
+      "loss": 0.1071,
+      "step": 221
+    },
+    {
+      "epoch": 2.018292682926829,
+      "grad_norm": 0.5539699196815491,
+      "learning_rate": 4.425198260380818e-06,
+      "loss": 0.1095,
+      "step": 222
+    },
+    {
+      "epoch": 2.027439024390244,
+      "grad_norm": 0.4581831395626068,
+      "learning_rate": 4.416123251058756e-06,
+      "loss": 0.104,
+      "step": 223
+    },
+    {
+      "epoch": 2.0365853658536586,
+      "grad_norm": 0.550529420375824,
+      "learning_rate": 4.406986624460979e-06,
+      "loss": 0.1049,
+      "step": 224
+    },
+    {
+      "epoch": 2.0457317073170733,
+      "grad_norm": 0.749291181564331,
+      "learning_rate": 4.397788674396374e-06,
+      "loss": 0.1091,
+      "step": 225
+    },
+    {
+      "epoch": 2.0548780487804876,
+      "grad_norm": 0.7612536549568176,
+      "learning_rate": 4.3885296966458225e-06,
+      "loss": 0.1045,
+      "step": 226
+    },
+    {
+      "epoch": 2.0640243902439024,
+      "grad_norm": 0.6671517491340637,
+      "learning_rate": 4.37920998895269e-06,
+      "loss": 0.1057,
+      "step": 227
+    },
+    {
+      "epoch": 2.073170731707317,
+      "grad_norm": 0.5838347673416138,
+      "learning_rate": 4.369829851013251e-06,
+      "loss": 0.1061,
+      "step": 228
+    },
+    {
+      "epoch": 2.082317073170732,
+      "grad_norm": 1.937151312828064,
+      "learning_rate": 4.36038958446705e-06,
+      "loss": 0.1074,
+      "step": 229
+    },
+    {
+      "epoch": 2.091463414634146,
+      "grad_norm": 0.8083934187889099,
+      "learning_rate": 4.350889492887202e-06,
+      "loss": 0.1096,
+      "step": 230
+    },
+    {
+      "epoch": 2.100609756097561,
+      "grad_norm": 0.6810253262519836,
+      "learning_rate": 4.341329881770635e-06,
+      "loss": 0.1024,
+      "step": 231
+    },
+    {
+      "epoch": 2.1097560975609757,
+      "grad_norm": 0.8901084065437317,
+      "learning_rate": 4.331711058528259e-06,
+      "loss": 0.1064,
+      "step": 232
+    },
+    {
+      "epoch": 2.1189024390243905,
+      "grad_norm": 0.34419459104537964,
+      "learning_rate": 4.322033332475084e-06,
+      "loss": 0.1074,
+      "step": 233
+    },
+    {
+      "epoch": 2.1280487804878048,
+      "grad_norm": 0.8596321940422058,
+      "learning_rate": 4.312297014820275e-06,
+      "loss": 0.1061,
+      "step": 234
+    },
+    {
+      "epoch": 2.1371951219512195,
+      "grad_norm": 0.8391798138618469,
+      "learning_rate": 4.302502418657138e-06,
+      "loss": 0.107,
+      "step": 235
+    },
+    {
+      "epoch": 2.1463414634146343,
+      "grad_norm": 0.7193841338157654,
+      "learning_rate": 4.292649858953063e-06,
+      "loss": 0.1083,
+      "step": 236
+    },
+    {
+      "epoch": 2.1554878048780486,
+      "grad_norm": 1.5975165367126465,
+      "learning_rate": 4.282739652539384e-06,
+      "loss": 0.109,
+      "step": 237
+    },
+    {
+      "epoch": 2.1646341463414633,
+      "grad_norm": 0.9933533072471619,
+      "learning_rate": 4.272772118101195e-06,
+      "loss": 0.1073,
+      "step": 238
+    },
+    {
+      "epoch": 2.173780487804878,
+      "grad_norm": 0.9281901717185974,
+      "learning_rate": 4.262747576167106e-06,
+      "loss": 0.1066,
+      "step": 239
+    },
+    {
+      "epoch": 2.182926829268293,
+      "grad_norm": 3.2005884647369385,
+      "learning_rate": 4.2526663490989264e-06,
+      "loss": 0.1106,
+      "step": 240
+    },
+    {
+      "epoch": 2.192073170731707,
+      "grad_norm": 1.084905982017517,
+      "learning_rate": 4.24252876108131e-06,
+      "loss": 0.1047,
+      "step": 241
+    },
+    {
+      "epoch": 2.201219512195122,
+      "grad_norm": 0.5345124006271362,
+      "learning_rate": 4.232335138111321e-06,
+      "loss": 0.1027,
+      "step": 242
+    },
+    {
+      "epoch": 2.2103658536585367,
+      "grad_norm": 0.6235584020614624,
+      "learning_rate": 4.222085807987954e-06,
+      "loss": 0.1032,
+      "step": 243
+    },
+    {
+      "epoch": 2.2195121951219514,
+      "grad_norm": 0.9797384738922119,
+      "learning_rate": 4.211781100301596e-06,
+      "loss": 0.1082,
+      "step": 244
+    },
+    {
+      "epoch": 2.2286585365853657,
+      "grad_norm": 0.3665165901184082,
+      "learning_rate": 4.201421346423421e-06,
+      "loss": 0.1057,
+      "step": 245
+    },
+    {
+      "epoch": 2.2378048780487805,
+      "grad_norm": 1.232487440109253,
+      "learning_rate": 4.19100687949474e-06,
+      "loss": 0.1088,
+      "step": 246
+    },
+    {
+      "epoch": 2.2469512195121952,
+      "grad_norm": 0.8028481602668762,
+      "learning_rate": 4.180538034416287e-06,
+      "loss": 0.1119,
+      "step": 247
+    },
+    {
+      "epoch": 2.2560975609756095,
+      "grad_norm": 1.094025731086731,
+      "learning_rate": 4.1700151478374436e-06,
+      "loss": 0.1033,
+      "step": 248
+    },
+    {
+      "epoch": 2.2652439024390243,
+      "grad_norm": 1.0119608640670776,
+      "learning_rate": 4.159438558145425e-06,
+      "loss": 0.108,
+      "step": 249
+    },
+    {
+      "epoch": 2.274390243902439,
+      "grad_norm": 0.6864210963249207,
+      "learning_rate": 4.148808605454385e-06,
+      "loss": 0.1072,
+      "step": 250
+    },
+    {
+      "epoch": 2.283536585365854,
+      "grad_norm": 0.8415966629981995,
+      "learning_rate": 4.138125631594488e-06,
+      "loss": 0.1052,
+      "step": 251
+    },
+    {
+      "epoch": 2.292682926829268,
+      "grad_norm": 0.7150280475616455,
+      "learning_rate": 4.127389980100916e-06,
+      "loss": 0.1049,
+      "step": 252
+    },
+    {
+      "epoch": 2.301829268292683,
+      "grad_norm": 1.2521971464157104,
+      "learning_rate": 4.116601996202815e-06,
+      "loss": 0.1078,
+      "step": 253
+    },
+    {
+      "epoch": 2.3109756097560976,
+      "grad_norm": 0.8204450011253357,
+      "learning_rate": 4.1057620268122e-06,
+      "loss": 0.1078,
+      "step": 254
+    },
+    {
+      "epoch": 2.3201219512195124,
+      "grad_norm": 1.0992320775985718,
+      "learning_rate": 4.0948704205128e-06,
+      "loss": 0.1063,
+      "step": 255
+    },
+    {
+      "epoch": 2.3292682926829267,
+      "grad_norm": 0.4928065240383148,
+      "learning_rate": 4.08392752754884e-06,
+      "loss": 0.1076,
+      "step": 256
+    },
+    {
+      "epoch": 2.3384146341463414,
+      "grad_norm": 1.1588791608810425,
+      "learning_rate": 4.072933699813788e-06,
+      "loss": 0.1075,
+      "step": 257
+    },
+    {
+      "epoch": 2.347560975609756,
+      "grad_norm": 0.5810270309448242,
+      "learning_rate": 4.061889290839032e-06,
+      "loss": 0.107,
+      "step": 258
+    },
+    {
+      "epoch": 2.3567073170731705,
+      "grad_norm": 0.8428746461868286,
+      "learning_rate": 4.050794655782515e-06,
+      "loss": 0.1057,
+      "step": 259
+    },
+    {
+      "epoch": 2.3658536585365852,
+      "grad_norm": 1.4898464679718018,
+      "learning_rate": 4.039650151417316e-06,
+      "loss": 0.1043,
+      "step": 260
+    },
+    {
+      "epoch": 2.375,
+      "grad_norm": 0.8613987565040588,
+      "learning_rate": 4.028456136120169e-06,
+      "loss": 0.1011,
+      "step": 261
+    },
+    {
+      "epoch": 2.3841463414634148,
+      "grad_norm": 0.7771743535995483,
+      "learning_rate": 4.017212969859948e-06,
+      "loss": 0.1073,
+      "step": 262
+    },
+    {
+      "epoch": 2.3932926829268295,
+      "grad_norm": 0.4884893596172333,
+      "learning_rate": 4.0059210141860875e-06,
+      "loss": 0.104,
+      "step": 263
+    },
+    {
+      "epoch": 2.402439024390244,
+      "grad_norm": 0.8188875317573547,
+      "learning_rate": 3.994580632216952e-06,
+      "loss": 0.1043,
+      "step": 264
+    },
+    {
+      "epoch": 2.4115853658536586,
+      "grad_norm": 0.814239501953125,
+      "learning_rate": 3.9831921886281675e-06,
+      "loss": 0.1073,
+      "step": 265
+    },
+    {
+      "epoch": 2.4207317073170733,
+      "grad_norm": 1.029391884803772,
+      "learning_rate": 3.971756049640888e-06,
+      "loss": 0.103,
+      "step": 266
+    },
+    {
+      "epoch": 2.4298780487804876,
+      "grad_norm": 0.3826993405818939,
+      "learning_rate": 3.96027258301002e-06,
+      "loss": 0.1089,
+      "step": 267
+    },
+    {
+      "epoch": 2.4390243902439024,
+      "grad_norm": 0.5306946039199829,
+      "learning_rate": 3.9487421580123995e-06,
+      "loss": 0.1086,
+      "step": 268
+    },
+    {
+      "epoch": 2.448170731707317,
+      "grad_norm": 1.286549687385559,
+      "learning_rate": 3.937165145434914e-06,
+      "loss": 0.1085,
+      "step": 269
+    },
+    {
+      "epoch": 2.457317073170732,
+      "grad_norm": 1.5009950399398804,
+      "learning_rate": 3.92554191756258e-06,
+      "loss": 0.1068,
+      "step": 270
+    },
+    {
+      "epoch": 2.466463414634146,
+      "grad_norm": 0.9225317239761353,
+      "learning_rate": 3.9138728481665696e-06,
+      "loss": 0.1059,
+      "step": 271
+    },
+    {
+      "epoch": 2.475609756097561,
+      "grad_norm": 0.5062177181243896,
+      "learning_rate": 3.902158312492196e-06,
+      "loss": 0.1089,
+      "step": 272
+    },
+    {
+      "epoch": 2.4847560975609757,
+      "grad_norm": 0.6272971630096436,
+      "learning_rate": 3.8903986872468416e-06,
+      "loss": 0.1099,
+      "step": 273
+    },
+    {
+      "epoch": 2.4939024390243905,
+      "grad_norm": 1.207027554512024,
+      "learning_rate": 3.878594350587844e-06,
+      "loss": 0.1086,
+      "step": 274
+    },
+    {
+      "epoch": 2.5030487804878048,
+      "grad_norm": 0.8417999744415283,
+      "learning_rate": 3.866745682110343e-06,
+      "loss": 0.1091,
+      "step": 275
+    },
+    {
+      "epoch": 2.5121951219512195,
+      "grad_norm": 0.5926170945167542,
+      "learning_rate": 3.854853062835062e-06,
+      "loss": 0.1081,
+      "step": 276
+    },
+    {
+      "epoch": 2.5213414634146343,
+      "grad_norm": 0.5806312561035156,
+      "learning_rate": 3.842916875196066e-06,
+      "loss": 0.1057,
+      "step": 277
+    },
+    {
+      "epoch": 2.5304878048780486,
+      "grad_norm": 1.2453746795654297,
+      "learning_rate": 3.830937503028457e-06,
+      "loss": 0.1086,
+      "step": 278
+    },
+    {
+      "epoch": 2.5396341463414633,
+      "grad_norm": 0.32152408361434937,
+      "learning_rate": 3.818915331556032e-06,
+      "loss": 0.1076,
+      "step": 279
+    },
+    {
+      "epoch": 2.548780487804878,
+      "grad_norm": 0.5756719708442688,
+      "learning_rate": 3.8068507473789016e-06,
+      "loss": 0.1057,
+      "step": 280
+    },
+    {
+      "epoch": 2.557926829268293,
+      "grad_norm": 0.9345139861106873,
+      "learning_rate": 3.7947441384610473e-06,
+      "loss": 0.1043,
+      "step": 281
+    },
+    {
+      "epoch": 2.567073170731707,
+      "grad_norm": 0.6475419998168945,
+      "learning_rate": 3.7825958941178538e-06,
+      "loss": 0.1046,
+      "step": 282
+    },
+    {
+      "epoch": 2.576219512195122,
+      "grad_norm": 0.7345425486564636,
+      "learning_rate": 3.770406405003586e-06,
+      "loss": 0.1042,
+      "step": 283
+    },
+    {
+      "epoch": 2.5853658536585367,
+      "grad_norm": 0.769500195980072,
+      "learning_rate": 3.758176063098829e-06,
+      "loss": 0.1048,
+      "step": 284
+    },
+    {
+      "epoch": 2.5945121951219514,
+      "grad_norm": 0.5934444665908813,
+      "learning_rate": 3.745905261697881e-06,
+      "loss": 0.1086,
+      "step": 285
+    },
+    {
+      "epoch": 2.6036585365853657,
+      "grad_norm": 0.5718421936035156,
+      "learning_rate": 3.7335943953961063e-06,
+      "loss": 0.1027,
+      "step": 286
+    },
+    {
+      "epoch": 2.6128048780487805,
+      "grad_norm": 0.6114512085914612,
+      "learning_rate": 3.721243860077247e-06,
+      "loss": 0.1091,
+      "step": 287
+    },
+    {
+      "epoch": 2.6219512195121952,
+      "grad_norm": 0.5620512366294861,
+      "learning_rate": 3.708854052900693e-06,
+      "loss": 0.1088,
+      "step": 288
+    },
+    {
+      "epoch": 2.6310975609756095,
+      "grad_norm": 0.6152999997138977,
+      "learning_rate": 3.696425372288708e-06,
+      "loss": 0.1035,
+      "step": 289
+    },
+    {
+      "epoch": 2.6402439024390243,
+      "grad_norm": 0.9731057286262512,
+      "learning_rate": 3.68395821791362e-06,
+      "loss": 0.1066,
+      "step": 290
+    },
+    {
+      "epoch": 2.649390243902439,
+      "grad_norm": 0.49733713269233704,
+      "learning_rate": 3.6714529906849683e-06,
+      "loss": 0.1045,
+      "step": 291
+    },
+    {
+      "epoch": 2.658536585365854,
+      "grad_norm": 0.49706941843032837,
+      "learning_rate": 3.65891009273661e-06,
+      "loss": 0.1035,
+      "step": 292
+    },
+    {
+      "epoch": 2.6676829268292686,
+      "grad_norm": 0.32593467831611633,
+      "learning_rate": 3.6463299274137883e-06,
+      "loss": 0.1017,
+      "step": 293
+    },
+    {
+      "epoch": 2.676829268292683,
+      "grad_norm": 0.4569077491760254,
+      "learning_rate": 3.6337128992601665e-06,
+      "loss": 0.1061,
+      "step": 294
+    },
+    {
+      "epoch": 2.6859756097560976,
+      "grad_norm": 0.5540208220481873,
+      "learning_rate": 3.621059414004811e-06,
+      "loss": 0.106,
+      "step": 295
+    },
+    {
+      "epoch": 2.6951219512195124,
+      "grad_norm": 0.9319402575492859,
+      "learning_rate": 3.6083698785491537e-06,
+      "loss": 0.1042,
+      "step": 296
+    },
+    {
+      "epoch": 2.7042682926829267,
+      "grad_norm": 0.795711874961853,
+      "learning_rate": 3.595644700953898e-06,
+      "loss": 0.104,
+      "step": 297
+    },
+    {
+      "epoch": 2.7134146341463414,
+      "grad_norm": 0.4707932472229004,
+      "learning_rate": 3.582884290425903e-06,
+      "loss": 0.1087,
+      "step": 298
+    },
+    {
+      "epoch": 2.722560975609756,
+      "grad_norm": 0.7709797620773315,
+      "learning_rate": 3.5700890573050224e-06,
+      "loss": 0.107,
+      "step": 299
+    },
+    {
+      "epoch": 2.7317073170731705,
+      "grad_norm": 1.1342726945877075,
+      "learning_rate": 3.5572594130509074e-06,
+      "loss": 0.1029,
+      "step": 300
+    },
+    {
+      "epoch": 2.7408536585365852,
+      "grad_norm": 0.9055543541908264,
+      "learning_rate": 3.544395770229781e-06,
+      "loss": 0.1087,
+      "step": 301
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 1.084656000137329,
+      "learning_rate": 3.531498542501161e-06,
+      "loss": 0.1044,
+      "step": 302
+    },
+    {
+      "epoch": 2.7591463414634148,
+      "grad_norm": 0.5991970300674438,
+      "learning_rate": 3.51856814460457e-06,
+      "loss": 0.1064,
+      "step": 303
+    },
+    {
+      "epoch": 2.7682926829268295,
+      "grad_norm": 1.7364325523376465,
+      "learning_rate": 3.5056049923461873e-06,
+      "loss": 0.1079,
+      "step": 304
+    },
+    {
+      "epoch": 2.777439024390244,
+      "grad_norm": 0.4037036895751953,
+      "learning_rate": 3.4926095025854863e-06,
+      "loss": 0.1058,
+      "step": 305
+    },
+    {
+      "epoch": 2.7865853658536586,
+      "grad_norm": 0.8944678902626038,
+      "learning_rate": 3.479582093221824e-06,
+      "loss": 0.1064,
+      "step": 306
+    },
+    {
+      "epoch": 2.7957317073170733,
+      "grad_norm": 0.5730510950088501,
+      "learning_rate": 3.4665231831810055e-06,
+      "loss": 0.1055,
+      "step": 307
+    },
+    {
+      "epoch": 2.8048780487804876,
+      "grad_norm": 0.519359827041626,
+      "learning_rate": 3.45343319240181e-06,
+      "loss": 0.1069,
+      "step": 308
+    },
+    {
+      "epoch": 2.8140243902439024,
+      "grad_norm": 0.6074658632278442,
+      "learning_rate": 3.44031254182249e-06,
+      "loss": 0.1058,
+      "step": 309
+    },
+    {
+      "epoch": 2.823170731707317,
+      "grad_norm": 0.37584108114242554,
+      "learning_rate": 3.4271616533672312e-06,
+      "loss": 0.1052,
+      "step": 310
+    },
+    {
+      "epoch": 2.8323170731707314,
+      "grad_norm": 0.7730855941772461,
+      "learning_rate": 3.413980949932589e-06,
+      "loss": 0.1037,
+      "step": 311
+    },
+    {
+      "epoch": 2.841463414634146,
+      "grad_norm": 0.9912458062171936,
+      "learning_rate": 3.4007708553738837e-06,
+      "loss": 0.1039,
+      "step": 312
+    },
+    {
+      "epoch": 2.850609756097561,
+      "grad_norm": 0.36861947178840637,
+      "learning_rate": 3.3875317944915765e-06,
+      "loss": 0.1083,
+      "step": 313
+    },
+    {
+      "epoch": 2.8597560975609757,
+      "grad_norm": 0.6069959402084351,
+      "learning_rate": 3.3742641930176044e-06,
+      "loss": 0.1063,
+      "step": 314
+    },
+    {
+      "epoch": 2.8689024390243905,
+      "grad_norm": 0.5478480458259583,
+      "learning_rate": 3.3609684776016932e-06,
+      "loss": 0.1033,
+      "step": 315
+    },
+    {
+      "epoch": 2.8780487804878048,
+      "grad_norm": 0.8352198004722595,
+      "learning_rate": 3.3476450757976353e-06,
+      "loss": 0.1026,
+      "step": 316
+    },
+    {
+      "epoch": 2.8871951219512195,
+      "grad_norm": 0.32641175389289856,
+      "learning_rate": 3.3342944160495405e-06,
+      "loss": 0.1065,
+      "step": 317
+    },
+    {
+      "epoch": 2.8963414634146343,
+      "grad_norm": 0.8076167702674866,
+      "learning_rate": 3.3209169276780612e-06,
+      "loss": 0.105,
+      "step": 318
+    },
+    {
+      "epoch": 2.9054878048780486,
+      "grad_norm": 0.3602111041545868,
+      "learning_rate": 3.307513040866584e-06,
+      "loss": 0.106,
+      "step": 319
+    },
+    {
+      "epoch": 2.9146341463414633,
+      "grad_norm": 0.227371945977211,
+      "learning_rate": 3.2940831866473967e-06,
+      "loss": 0.1029,
+      "step": 320
+    },
+    {
+      "epoch": 2.923780487804878,
+      "grad_norm": 0.5158879160881042,
+      "learning_rate": 3.280627796887828e-06,
+      "loss": 0.1046,
+      "step": 321
+    },
+    {
+      "epoch": 2.932926829268293,
+      "grad_norm": 1.0091784000396729,
+      "learning_rate": 3.2671473042763603e-06,
+      "loss": 0.1063,
+      "step": 322
+    },
+    {
+      "epoch": 2.942073170731707,
+      "grad_norm": 0.9310367107391357,
+      "learning_rate": 3.2536421423087137e-06,
+      "loss": 0.1015,
+      "step": 323
+    },
+    {
+      "epoch": 2.951219512195122,
+      "grad_norm": 0.7602329254150391,
+      "learning_rate": 3.2401127452739106e-06,
+      "loss": 0.1061,
+      "step": 324
+    },
+    {
+      "epoch": 2.9603658536585367,
+      "grad_norm": 0.3374696671962738,
+      "learning_rate": 3.2265595482403002e-06,
+      "loss": 0.1049,
+      "step": 325
+    },
+    {
+      "epoch": 2.9695121951219514,
+      "grad_norm": 0.8469128012657166,
+      "learning_rate": 3.212982987041582e-06,
+      "loss": 0.1059,
+      "step": 326
+    },
+    {
+      "epoch": 2.9786585365853657,
+      "grad_norm": 0.7546894550323486,
+      "learning_rate": 3.199383498262777e-06,
+      "loss": 0.1067,
+      "step": 327
+    },
+    {
+      "epoch": 2.9878048780487805,
+      "grad_norm": 0.471538245677948,
+      "learning_rate": 3.185761519226199e-06,
+      "loss": 0.1054,
+      "step": 328
+    },
+    {
+      "epoch": 2.9969512195121952,
+      "grad_norm": 1.0999424457550049,
+      "learning_rate": 3.1721174879773825e-06,
+      "loss": 0.1067,
+      "step": 329
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.9866787791252136,
+      "learning_rate": 3.1584518432710055e-06,
+      "loss": 0.0361,
+      "step": 330
+    },
+    {
+      "epoch": 3.0091463414634148,
+      "grad_norm": 0.4876601994037628,
+      "learning_rate": 3.1447650245567714e-06,
+      "loss": 0.1058,
+      "step": 331
+    },
+    {
+      "epoch": 3.018292682926829,
+      "grad_norm": 1.1444916725158691,
+      "learning_rate": 3.131057471965283e-06,
+      "loss": 0.1051,
+      "step": 332
+    },
+    {
+      "epoch": 3.027439024390244,
+      "grad_norm": 0.6264286637306213,
+      "learning_rate": 3.117329626293891e-06,
+      "loss": 0.1065,
+      "step": 333
+    },
+    {
+      "epoch": 3.0365853658536586,
+      "grad_norm": 0.49345099925994873,
+      "learning_rate": 3.10358192899251e-06,
+      "loss": 0.1034,
+      "step": 334
+    },
+    {
+      "epoch": 3.0457317073170733,
+      "grad_norm": 0.663642168045044,
+      "learning_rate": 3.089814822149435e-06,
+      "loss": 0.1063,
+      "step": 335
+    },
+    {
+      "epoch": 3.0548780487804876,
+      "grad_norm": 0.7982928156852722,
+      "learning_rate": 3.0760287484771124e-06,
+      "loss": 0.1047,
+      "step": 336
+    },
+    {
+      "epoch": 3.0640243902439024,
+      "grad_norm": 0.7954766750335693,
+      "learning_rate": 3.062224151297915e-06,
+      "loss": 0.1069,
+      "step": 337
+    },
+    {
+      "epoch": 3.073170731707317,
+      "grad_norm": 0.42658311128616333,
+      "learning_rate": 3.0484014745298784e-06,
+      "loss": 0.1043,
+      "step": 338
+    },
+    {
+      "epoch": 3.082317073170732,
+      "grad_norm": 0.8154664635658264,
+      "learning_rate": 3.0345611626724286e-06,
+      "loss": 0.1061,
+      "step": 339
+    },
+    {
+      "epoch": 3.091463414634146,
+      "grad_norm": 0.6405104994773865,
+      "learning_rate": 3.020703660792089e-06,
+      "loss": 0.1044,
+      "step": 340
+    },
+    {
+      "epoch": 3.100609756097561,
+      "grad_norm": 0.7365081906318665,
+      "learning_rate": 3.006829414508164e-06,
+      "loss": 0.1061,
+      "step": 341
+    },
+    {
+      "epoch": 3.1097560975609757,
+      "grad_norm": 0.322336345911026,
+      "learning_rate": 2.9929388699784163e-06,
+      "loss": 0.1067,
+      "step": 342
+    },
+    {
+      "epoch": 3.1189024390243905,
+      "grad_norm": 1.0929139852523804,
+      "learning_rate": 2.9790324738847117e-06,
+      "loss": 0.1084,
+      "step": 343
+    },
+    {
+      "epoch": 3.1280487804878048,
+      "grad_norm": 0.9065106511116028,
+      "learning_rate": 2.965110673418662e-06,
+      "loss": 0.1029,
+      "step": 344
+    },
+    {
+      "epoch": 3.1371951219512195,
+      "grad_norm": 0.7359282970428467,
+      "learning_rate": 2.9511739162672377e-06,
+      "loss": 0.1022,
+      "step": 345
+    },
+    {
+      "epoch": 3.1463414634146343,
+      "grad_norm": 0.5595942139625549,
+      "learning_rate": 2.9372226505983803e-06,
+      "loss": 0.1067,
+      "step": 346
+    },
+    {
+      "epoch": 3.1554878048780486,
+      "grad_norm": 0.9696919918060303,
+      "learning_rate": 2.923257325046581e-06,
+      "loss": 0.1085,
+      "step": 347
+    },
+    {
+      "epoch": 3.1646341463414633,
+      "grad_norm": 0.3462243676185608,
+      "learning_rate": 2.9092783886984615e-06,
+      "loss": 0.1022,
+      "step": 348
+    },
+    {
+      "epoch": 3.173780487804878,
+      "grad_norm": 0.5291770696640015,
+      "learning_rate": 2.895286291078327e-06,
+      "loss": 0.1053,
+      "step": 349
+    },
+    {
+      "epoch": 3.182926829268293,
+      "grad_norm": 0.667945921421051,
+      "learning_rate": 2.881281482133715e-06,
+      "loss": 0.1031,
+      "step": 350
+    },
+    {
+      "epoch": 3.192073170731707,
+      "grad_norm": 1.4323768615722656,
+      "learning_rate": 2.8672644122209224e-06,
+      "loss": 0.1081,
+      "step": 351
+    },
+    {
+      "epoch": 3.201219512195122,
+      "grad_norm": 0.6841598749160767,
+      "learning_rate": 2.8532355320905277e-06,
+      "loss": 0.106,
+      "step": 352
+    },
+    {
+      "epoch": 3.2103658536585367,
+      "grad_norm": 1.017216444015503,
+      "learning_rate": 2.839195292872892e-06,
+      "loss": 0.1087,
+      "step": 353
+    },
+    {
+      "epoch": 3.2195121951219514,
+      "grad_norm": 0.6531134843826294,
+      "learning_rate": 2.8251441460636542e-06,
+      "loss": 0.1046,
+      "step": 354
+    },
+    {
+      "epoch": 3.2286585365853657,
+      "grad_norm": 0.5456640124320984,
+      "learning_rate": 2.8110825435092108e-06,
+      "loss": 0.1039,
+      "step": 355
+    },
+    {
+      "epoch": 3.2378048780487805,
+      "grad_norm": 0.7548927068710327,
+      "learning_rate": 2.797010937392188e-06,
+      "loss": 0.1031,
+      "step": 356
+    },
+    {
+      "epoch": 3.2469512195121952,
+      "grad_norm": 0.3874664008617401,
+      "learning_rate": 2.782929780216896e-06,
+      "loss": 0.1044,
+      "step": 357
+    },
+    {
+      "epoch": 3.2560975609756095,
+      "grad_norm": 0.41182073950767517,
+      "learning_rate": 2.768839524794784e-06,
+      "loss": 0.1041,
+      "step": 358
+    },
+    {
+      "epoch": 3.2652439024390243,
+      "grad_norm": 0.9750642776489258,
+      "learning_rate": 2.754740624229875e-06,
+      "loss": 0.1074,
+      "step": 359
+    },
+    {
+      "epoch": 3.274390243902439,
+      "grad_norm": 0.5468454957008362,
+      "learning_rate": 2.740633531904196e-06,
+      "loss": 0.1056,
+      "step": 360
+    },
+    {
+      "epoch": 3.283536585365854,
+      "grad_norm": 0.6265747547149658,
+      "learning_rate": 2.7265187014631983e-06,
+      "loss": 0.0993,
+      "step": 361
+    },
+    {
+      "epoch": 3.292682926829268,
+      "grad_norm": 0.5447110533714294,
+      "learning_rate": 2.7123965868011693e-06,
+      "loss": 0.1023,
+      "step": 362
+    },
+    {
+      "epoch": 3.301829268292683,
+      "grad_norm": 0.5873153805732727,
+      "learning_rate": 2.698267642046639e-06,
+      "loss": 0.1043,
+      "step": 363
+    },
+    {
+      "epoch": 3.3109756097560976,
+      "grad_norm": 0.8410884141921997,
+      "learning_rate": 2.6841323215477716e-06,
+      "loss": 0.1026,
+      "step": 364
+    },
+    {
+      "epoch": 3.3201219512195124,
+      "grad_norm": 0.6379916667938232,
+      "learning_rate": 2.6699910798577612e-06,
+      "loss": 0.1043,
+      "step": 365
+    },
+    {
+      "epoch": 3.3292682926829267,
+      "grad_norm": 0.3435944616794586,
+      "learning_rate": 2.6558443717202077e-06,
+      "loss": 0.105,
+      "step": 366
+    },
+    {
+      "epoch": 3.3384146341463414,
+      "grad_norm": 1.0160895586013794,
+      "learning_rate": 2.6416926520545e-06,
+      "loss": 0.1029,
+      "step": 367
+    },
+    {
+      "epoch": 3.347560975609756,
+      "grad_norm": 0.6623978018760681,
+      "learning_rate": 2.6275363759411817e-06,
+      "loss": 0.1023,
+      "step": 368
+    },
+    {
+      "epoch": 3.3567073170731705,
+      "grad_norm": 0.42660218477249146,
+      "learning_rate": 2.61337599860732e-06,
+      "loss": 0.102,
+      "step": 369
+    },
+    {
+      "epoch": 3.3658536585365852,
+      "grad_norm": 0.992641270160675,
+      "learning_rate": 2.599211975411867e-06,
+      "loss": 0.1087,
+      "step": 370
+    },
+    {
+      "epoch": 3.375,
+      "grad_norm": 0.9266735911369324,
+      "learning_rate": 2.585044761831014e-06,
+      "loss": 0.102,
+      "step": 371
+    },
+    {
+      "epoch": 3.3841463414634148,
+      "grad_norm": 0.7853158712387085,
+      "learning_rate": 2.5708748134435503e-06,
+      "loss": 0.1062,
+      "step": 372
+    },
+    {
+      "epoch": 3.3932926829268295,
+      "grad_norm": 0.501575767993927,
+      "learning_rate": 2.556702585916202e-06,
+      "loss": 0.1034,
+      "step": 373
+    },
+    {
+      "epoch": 3.402439024390244,
+      "grad_norm": 0.4489349126815796,
+      "learning_rate": 2.5425285349889937e-06,
+      "loss": 0.1072,
+      "step": 374
+    },
+    {
+      "epoch": 3.4115853658536586,
+      "grad_norm": 0.6775074601173401,
+      "learning_rate": 2.52835311646058e-06,
+      "loss": 0.1069,
+      "step": 375
+    },
+    {
+      "epoch": 3.4207317073170733,
+      "grad_norm": 0.6220641732215881,
+      "learning_rate": 2.5141767861735976e-06,
+      "loss": 0.1057,
+      "step": 376
+    },
+    {
+      "epoch": 3.4298780487804876,
+      "grad_norm": 0.8748292326927185,
+      "learning_rate": 2.5e-06,
+      "loss": 0.1043,
+      "step": 377
+    },
+    {
+      "epoch": 3.4390243902439024,
+      "grad_norm": 0.9639992117881775,
+      "learning_rate": 2.485823213826403e-06,
+      "loss": 0.1086,
+      "step": 378
+    },
+    {
+      "epoch": 3.448170731707317,
+      "grad_norm": 0.7457119822502136,
+      "learning_rate": 2.47164688353942e-06,
+      "loss": 0.1066,
+      "step": 379
+    },
+    {
+      "epoch": 3.457317073170732,
+      "grad_norm": 0.9384257197380066,
+      "learning_rate": 2.457471465011007e-06,
+      "loss": 0.1045,
+      "step": 380
+    },
+    {
+      "epoch": 3.466463414634146,
+      "grad_norm": 1.3412197828292847,
+      "learning_rate": 2.4432974140837988e-06,
+      "loss": 0.1041,
+      "step": 381
+    },
+    {
+      "epoch": 3.475609756097561,
+      "grad_norm": 0.5068745017051697,
+      "learning_rate": 2.4291251865564505e-06,
+      "loss": 0.1032,
+      "step": 382
+    },
+    {
+      "epoch": 3.4847560975609757,
+      "grad_norm": 0.40219390392303467,
+      "learning_rate": 2.414955238168986e-06,
+      "loss": 0.1039,
+      "step": 383
+    },
+    {
+      "epoch": 3.4939024390243905,
+      "grad_norm": 1.0244395732879639,
+      "learning_rate": 2.4007880245881345e-06,
+      "loss": 0.1049,
+      "step": 384
+    },
+    {
+      "epoch": 3.5030487804878048,
+      "grad_norm": 0.7072446942329407,
+      "learning_rate": 2.386624001392681e-06,
+      "loss": 0.1062,
+      "step": 385
+    },
+    {
+      "epoch": 3.5121951219512195,
+      "grad_norm": 0.7629527449607849,
+      "learning_rate": 2.3724636240588196e-06,
+      "loss": 0.1034,
+      "step": 386
+    },
+    {
+      "epoch": 3.5213414634146343,
+      "grad_norm": 0.8184047937393188,
+      "learning_rate": 2.358307347945501e-06,
+      "loss": 0.1018,
+      "step": 387
+    },
+    {
+      "epoch": 3.5304878048780486,
+      "grad_norm": 0.353141188621521,
+      "learning_rate": 2.3441556282797935e-06,
+      "loss": 0.1077,
+      "step": 388
+    },
+    {
+      "epoch": 3.5396341463414633,
+      "grad_norm": 0.572818398475647,
+      "learning_rate": 2.33000892014224e-06,
+      "loss": 0.1047,
+      "step": 389
+    },
+    {
+      "epoch": 3.548780487804878,
+      "grad_norm": 0.7230311632156372,
+      "learning_rate": 2.315867678452229e-06,
+      "loss": 0.1025,
+      "step": 390
+    },
+    {
+      "epoch": 3.557926829268293,
+      "grad_norm": 1.1688281297683716,
+      "learning_rate": 2.3017323579533617e-06,
+      "loss": 0.1036,
+      "step": 391
+    },
+    {
+      "epoch": 3.567073170731707,
+      "grad_norm": 0.8227838277816772,
+      "learning_rate": 2.287603413198831e-06,
+      "loss": 0.104,
+      "step": 392
+    },
+    {
+      "epoch": 3.576219512195122,
+      "grad_norm": 0.45366042852401733,
+      "learning_rate": 2.2734812985368034e-06,
+      "loss": 0.1042,
+      "step": 393
+    },
+    {
+      "epoch": 3.5853658536585367,
+      "grad_norm": 0.31546878814697266,
+      "learning_rate": 2.2593664680958048e-06,
+      "loss": 0.0998,
+      "step": 394
+    },
+    {
+      "epoch": 3.5945121951219514,
+      "grad_norm": 0.3552347719669342,
+      "learning_rate": 2.2452593757701254e-06,
+      "loss": 0.1032,
+      "step": 395
+    },
+    {
+      "epoch": 3.6036585365853657,
+      "grad_norm": 0.38970789313316345,
+      "learning_rate": 2.231160475205216e-06,
+      "loss": 0.1014,
+      "step": 396
+    },
+    {
+      "epoch": 3.6128048780487805,
+      "grad_norm": 0.32332485914230347,
+      "learning_rate": 2.2170702197831052e-06,
+      "loss": 0.1042,
+      "step": 397
+    },
+    {
+      "epoch": 3.6219512195121952,
+      "grad_norm": 0.5800347924232483,
+      "learning_rate": 2.2029890626078133e-06,
+      "loss": 0.1025,
+      "step": 398
+    },
+    {
+      "epoch": 3.6310975609756095,
+      "grad_norm": 0.5717971920967102,
+      "learning_rate": 2.1889174564907897e-06,
+      "loss": 0.1022,
+      "step": 399
+    },
+    {
+      "epoch": 3.6402439024390243,
+      "grad_norm": 0.4129480719566345,
+      "learning_rate": 2.1748558539363458e-06,
+      "loss": 0.1043,
+      "step": 400
+    },
+    {
+      "epoch": 3.649390243902439,
+      "grad_norm": 0.8331505656242371,
+      "learning_rate": 2.1608047071271087e-06,
+      "loss": 0.103,
+      "step": 401
+    },
+    {
+      "epoch": 3.658536585365854,
+      "grad_norm": 0.7323583364486694,
+      "learning_rate": 2.1467644679094727e-06,
+      "loss": 0.1066,
+      "step": 402
+    },
+    {
+      "epoch": 3.6676829268292686,
+      "grad_norm": 0.828615665435791,
+      "learning_rate": 2.132735587779078e-06,
+      "loss": 0.1029,
+      "step": 403
+    },
+    {
+      "epoch": 3.676829268292683,
+      "grad_norm": 0.35115236043930054,
+      "learning_rate": 2.118718517866286e-06,
+      "loss": 0.1005,
+      "step": 404
+    },
+    {
+      "epoch": 3.6859756097560976,
+      "grad_norm": 0.45387038588523865,
+      "learning_rate": 2.1047137089216728e-06,
+      "loss": 0.1043,
+      "step": 405
+    },
+    {
+      "epoch": 3.6951219512195124,
+      "grad_norm": 0.6801460981369019,
+      "learning_rate": 2.09072161130154e-06,
+      "loss": 0.1013,
+      "step": 406
+    },
+    {
+      "epoch": 3.7042682926829267,
+      "grad_norm": 0.5863631367683411,
+      "learning_rate": 2.0767426749534194e-06,
+      "loss": 0.1008,
+      "step": 407
+    },
+    {
+      "epoch": 3.7134146341463414,
+      "grad_norm": 0.8629023432731628,
+      "learning_rate": 2.0627773494016205e-06,
+      "loss": 0.1058,
+      "step": 408
+    },
+    {
+      "epoch": 3.722560975609756,
+      "grad_norm": 0.929577112197876,
+      "learning_rate": 2.0488260837327627e-06,
+      "loss": 0.1043,
+      "step": 409
+    },
+    {
+      "epoch": 3.7317073170731705,
+      "grad_norm": 0.4356759488582611,
+      "learning_rate": 2.0348893265813397e-06,
+      "loss": 0.1029,
+      "step": 410
+    },
+    {
+      "epoch": 3.7408536585365852,
+      "grad_norm": 1.105850100517273,
+      "learning_rate": 2.0209675261152896e-06,
+      "loss": 0.104,
+      "step": 411
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 0.5208716988563538,
+      "learning_rate": 2.0070611300215846e-06,
+      "loss": 0.103,
+      "step": 412
+    },
+    {
+      "epoch": 3.7591463414634148,
+      "grad_norm": 0.6509872674942017,
+      "learning_rate": 1.993170585491836e-06,
+      "loss": 0.1032,
+      "step": 413
+    },
+    {
+      "epoch": 3.7682926829268295,
+      "grad_norm": 0.48762744665145874,
+      "learning_rate": 1.9792963392079125e-06,
+      "loss": 0.1065,
+      "step": 414
+    },
+    {
+      "epoch": 3.777439024390244,
+      "grad_norm": 0.4907792806625366,
+      "learning_rate": 1.9654388373275722e-06,
+      "loss": 0.1018,
+      "step": 415
+    },
+    {
+      "epoch": 3.7865853658536586,
+      "grad_norm": 0.6827414035797119,
+      "learning_rate": 1.951598525470122e-06,
+      "loss": 0.103,
+      "step": 416
+    },
+    {
+      "epoch": 3.7957317073170733,
+      "grad_norm": 0.915044367313385,
+      "learning_rate": 1.937775848702086e-06,
+      "loss": 0.1033,
+      "step": 417
+    },
+    {
+      "epoch": 3.8048780487804876,
+      "grad_norm": 0.7613033056259155,
+      "learning_rate": 1.923971251522888e-06,
+      "loss": 0.1047,
+      "step": 418
+    },
+    {
+      "epoch": 3.8140243902439024,
+      "grad_norm": 0.5619191527366638,
+      "learning_rate": 1.9101851778505666e-06,
+      "loss": 0.1031,
+      "step": 419
+    },
+    {
+      "epoch": 3.823170731707317,
+      "grad_norm": 0.8900383710861206,
+      "learning_rate": 1.8964180710074905e-06,
+      "loss": 0.1059,
+      "step": 420
+    },
+    {
+      "epoch": 3.8323170731707314,
+      "grad_norm": 1.0572234392166138,
+      "learning_rate": 1.8826703737061097e-06,
+      "loss": 0.1062,
+      "step": 421
+    },
+    {
+      "epoch": 3.841463414634146,
+      "grad_norm": 0.379605233669281,
+      "learning_rate": 1.8689425280347168e-06,
+      "loss": 0.1016,
+      "step": 422
+    },
+    {
+      "epoch": 3.850609756097561,
+      "grad_norm": 0.5158970355987549,
+      "learning_rate": 1.8552349754432303e-06,
+      "loss": 0.1044,
+      "step": 423
+    },
+    {
+      "epoch": 3.8597560975609757,
+      "grad_norm": 0.5448965430259705,
+      "learning_rate": 1.8415481567289956e-06,
+      "loss": 0.1055,
+      "step": 424
+    },
+    {
+      "epoch": 3.8689024390243905,
+      "grad_norm": 0.6450077891349792,
+      "learning_rate": 1.8278825120226179e-06,
+      "loss": 0.1015,
+      "step": 425
+    },
+    {
+      "epoch": 3.8780487804878048,
+      "grad_norm": 0.7701898813247681,
+      "learning_rate": 1.8142384807738023e-06,
+      "loss": 0.1019,
+      "step": 426
+    },
+    {
+      "epoch": 3.8871951219512195,
+      "grad_norm": 1.0020835399627686,
+      "learning_rate": 1.800616501737224e-06,
+      "loss": 0.1048,
+      "step": 427
+    },
+    {
+      "epoch": 3.8963414634146343,
+      "grad_norm": 0.8066243529319763,
+      "learning_rate": 1.7870170129584191e-06,
+      "loss": 0.1017,
+      "step": 428
+    },
+    {
+      "epoch": 3.9054878048780486,
+      "grad_norm": 0.6243605613708496,
+      "learning_rate": 1.7734404517597004e-06,
+      "loss": 0.1048,
+      "step": 429
+    },
+    {
+      "epoch": 3.9146341463414633,
+      "grad_norm": 0.7325178384780884,
+      "learning_rate": 1.7598872547260904e-06,
+      "loss": 0.1018,
+      "step": 430
+    },
+    {
+      "epoch": 3.923780487804878,
+      "grad_norm": 0.8298372030258179,
+      "learning_rate": 1.7463578576912859e-06,
+      "loss": 0.1061,
+      "step": 431
+    },
+    {
+      "epoch": 3.932926829268293,
+      "grad_norm": 0.3550826609134674,
+      "learning_rate": 1.7328526957236408e-06,
+      "loss": 0.105,
+      "step": 432
+    },
+    {
+      "epoch": 3.942073170731707,
+      "grad_norm": 0.6778271794319153,
+      "learning_rate": 1.7193722031121726e-06,
+      "loss": 0.1007,
+      "step": 433
+    },
+    {
+      "epoch": 3.951219512195122,
+      "grad_norm": 0.8127948641777039,
+      "learning_rate": 1.7059168133526043e-06,
+      "loss": 0.1009,
+      "step": 434
+    },
+    {
+      "epoch": 3.9603658536585367,
+      "grad_norm": 0.3050505518913269,
+      "learning_rate": 1.692486959133417e-06,
+      "loss": 0.1035,
+      "step": 435
+    },
+    {
+      "epoch": 3.9695121951219514,
+      "grad_norm": 0.6452750563621521,
+      "learning_rate": 1.6790830723219398e-06,
+      "loss": 0.1061,
+      "step": 436
+    },
+    {
+      "epoch": 3.9786585365853657,
+      "grad_norm": 0.72505784034729,
+      "learning_rate": 1.6657055839504601e-06,
+      "loss": 0.1041,
+      "step": 437
+    },
+    {
+      "epoch": 3.9878048780487805,
+      "grad_norm": 1.2222975492477417,
+      "learning_rate": 1.6523549242023657e-06,
+      "loss": 0.1025,
+      "step": 438
+    },
+    {
+      "epoch": 3.9969512195121952,
+      "grad_norm": 0.8575263619422913,
+      "learning_rate": 1.6390315223983068e-06,
+      "loss": 0.1004,
+      "step": 439
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.8575263619422913,
+      "learning_rate": 1.6257358069823965e-06,
+      "loss": 0.0363,
+      "step": 440
+    },
+    {
+      "epoch": 4.009146341463414,
+      "grad_norm": 0.9300926327705383,
+      "learning_rate": 1.6124682055084245e-06,
+      "loss": 0.1052,
+      "step": 441
+    },
+    {
+      "epoch": 4.0182926829268295,
+      "grad_norm": 0.7189462780952454,
+      "learning_rate": 1.599229144626117e-06,
+      "loss": 0.1051,
+      "step": 442
+    },
+    {
+      "epoch": 4.027439024390244,
+      "grad_norm": 0.5437363386154175,
+      "learning_rate": 1.5860190500674115e-06,
+      "loss": 0.1024,
+      "step": 443
+    },
+    {
+      "epoch": 4.036585365853658,
+      "grad_norm": 0.39869755506515503,
+      "learning_rate": 1.5728383466327685e-06,
+      "loss": 0.1039,
+      "step": 444
+    },
+    {
+      "epoch": 4.045731707317073,
+      "grad_norm": 0.9101700186729431,
+      "learning_rate": 1.5596874581775113e-06,
+      "loss": 0.104,
+      "step": 445
+    },
+    {
+      "epoch": 4.054878048780488,
+      "grad_norm": 0.8250009417533875,
+      "learning_rate": 1.5465668075981905e-06,
+      "loss": 0.1022,
+      "step": 446
+    },
+    {
+      "epoch": 4.064024390243903,
+      "grad_norm": 0.31336626410484314,
+      "learning_rate": 1.5334768168189953e-06,
+      "loss": 0.1027,
+      "step": 447
+    },
+    {
+      "epoch": 4.073170731707317,
+      "grad_norm": 0.807352602481842,
+      "learning_rate": 1.520417906778176e-06,
+      "loss": 0.1015,
+      "step": 448
+    },
+    {
+      "epoch": 4.0823170731707314,
+      "grad_norm": 1.00467848777771,
+      "learning_rate": 1.5073904974145145e-06,
+      "loss": 0.1035,
+      "step": 449
+    },
+    {
+      "epoch": 4.091463414634147,
+      "grad_norm": 0.42968934774398804,
+      "learning_rate": 1.4943950076538135e-06,
+      "loss": 0.102,
+      "step": 450
+    },
+    {
+      "epoch": 4.100609756097561,
+      "grad_norm": 0.573142409324646,
+      "learning_rate": 1.4814318553954307e-06,
+      "loss": 0.1064,
+      "step": 451
+    },
+    {
+      "epoch": 4.109756097560975,
+      "grad_norm": 0.7983995079994202,
+      "learning_rate": 1.4685014574988394e-06,
+      "loss": 0.106,
+      "step": 452
+    },
+    {
+      "epoch": 4.1189024390243905,
+      "grad_norm": 0.7572171092033386,
+      "learning_rate": 1.455604229770221e-06,
+      "loss": 0.1033,
+      "step": 453
+    },
+    {
+      "epoch": 4.128048780487805,
+      "grad_norm": 1.041406273841858,
+      "learning_rate": 1.4427405869490924e-06,
+      "loss": 0.1084,
+      "step": 454
+    },
+    {
+      "epoch": 4.137195121951219,
+      "grad_norm": 0.9685583114624023,
+      "learning_rate": 1.4299109426949784e-06,
+      "loss": 0.1005,
+      "step": 455
+    },
+    {
+      "epoch": 4.146341463414634,
+      "grad_norm": 0.9551640748977661,
+      "learning_rate": 1.4171157095740976e-06,
+      "loss": 0.0999,
+      "step": 456
+    },
+    {
+      "epoch": 4.155487804878049,
+      "grad_norm": 0.8675352334976196,
+      "learning_rate": 1.404355299046103e-06,
+      "loss": 0.1078,
+      "step": 457
+    },
+    {
+      "epoch": 4.164634146341464,
+      "grad_norm": 0.4849691390991211,
+      "learning_rate": 1.391630121450847e-06,
+      "loss": 0.1015,
+      "step": 458
+    },
+    {
+      "epoch": 4.173780487804878,
+      "grad_norm": 0.4850751757621765,
+      "learning_rate": 1.3789405859951894e-06,
+      "loss": 0.1029,
+      "step": 459
+    },
+    {
+      "epoch": 4.182926829268292,
+      "grad_norm": 0.4103759825229645,
+      "learning_rate": 1.3662871007398347e-06,
+      "loss": 0.1007,
+      "step": 460
+    },
+    {
+      "epoch": 4.192073170731708,
+      "grad_norm": 0.4612462520599365,
+      "learning_rate": 1.3536700725862115e-06,
+      "loss": 0.1036,
+      "step": 461
+    },
+    {
+      "epoch": 4.201219512195122,
+      "grad_norm": 0.4885130524635315,
+      "learning_rate": 1.3410899072633915e-06,
+      "loss": 0.1046,
+      "step": 462
+    },
+    {
+      "epoch": 4.210365853658536,
+      "grad_norm": 0.3064654767513275,
+      "learning_rate": 1.3285470093150328e-06,
+      "loss": 0.0992,
+      "step": 463
+    },
+    {
+      "epoch": 4.219512195121951,
+      "grad_norm": 0.76682448387146,
+      "learning_rate": 1.3160417820863808e-06,
+      "loss": 0.1051,
+      "step": 464
+    },
+    {
+      "epoch": 4.228658536585366,
+      "grad_norm": 0.771308958530426,
+      "learning_rate": 1.303574627711292e-06,
+      "loss": 0.1028,
+      "step": 465
+    },
+    {
+      "epoch": 4.237804878048781,
+      "grad_norm": 0.327519029378891,
+      "learning_rate": 1.2911459470993084e-06,
+      "loss": 0.1025,
+      "step": 466
+    },
+    {
+      "epoch": 4.246951219512195,
+      "grad_norm": 0.5206422209739685,
+      "learning_rate": 1.2787561399227545e-06,
+      "loss": 0.103,
+      "step": 467
+    },
+    {
+      "epoch": 4.2560975609756095,
+      "grad_norm": 0.7346555590629578,
+      "learning_rate": 1.2664056046038941e-06,
+      "loss": 0.1042,
+      "step": 468
+    },
+    {
+      "epoch": 4.265243902439025,
+      "grad_norm": 0.9764044880867004,
+      "learning_rate": 1.2540947383021197e-06,
+      "loss": 0.0985,
+      "step": 469
+    },
+    {
+      "epoch": 4.274390243902439,
+      "grad_norm": 0.8075615167617798,
+      "learning_rate": 1.2418239369011712e-06,
+      "loss": 0.1038,
+      "step": 470
+    },
+    {
+      "epoch": 4.283536585365853,
+      "grad_norm": 0.7572922706604004,
+      "learning_rate": 1.2295935949964143e-06,
+      "loss": 0.1044,
+      "step": 471
+    },
+    {
+      "epoch": 4.2926829268292686,
+      "grad_norm": 0.6829261183738708,
+      "learning_rate": 1.217404105882147e-06,
+      "loss": 0.101,
+      "step": 472
+    },
+    {
+      "epoch": 4.301829268292683,
+      "grad_norm": 0.2822739779949188,
+      "learning_rate": 1.2052558615389535e-06,
+      "loss": 0.1043,
+      "step": 473
+    },
+    {
+      "epoch": 4.310975609756097,
+      "grad_norm": 0.3203172981739044,
+      "learning_rate": 1.193149252621099e-06,
+      "loss": 0.1021,
+      "step": 474
+    },
+    {
+      "epoch": 4.320121951219512,
+      "grad_norm": 0.8120080828666687,
+      "learning_rate": 1.1810846684439682e-06,
+      "loss": 0.1035,
+      "step": 475
+    },
+    {
+      "epoch": 4.329268292682927,
+      "grad_norm": 0.5470507740974426,
+      "learning_rate": 1.1690624969715441e-06,
+      "loss": 0.1013,
+      "step": 476
+    },
+    {
+      "epoch": 4.338414634146342,
+      "grad_norm": 0.7421087026596069,
+      "learning_rate": 1.1570831248039347e-06,
+      "loss": 0.1029,
+      "step": 477
+    },
+    {
+      "epoch": 4.347560975609756,
+      "grad_norm": 1.4894626140594482,
+      "learning_rate": 1.1451469371649382e-06,
+      "loss": 0.1007,
+      "step": 478
+    },
+    {
+      "epoch": 4.3567073170731705,
+      "grad_norm": 0.8726413249969482,
+      "learning_rate": 1.1332543178896578e-06,
+      "loss": 0.1035,
+      "step": 479
+    },
+    {
+      "epoch": 4.365853658536586,
+      "grad_norm": 0.3398156464099884,
+      "learning_rate": 1.121405649412156e-06,
+      "loss": 0.1049,
+      "step": 480
+    },
+    {
+      "epoch": 4.375,
+      "grad_norm": 1.0846507549285889,
+      "learning_rate": 1.10960131275316e-06,
+      "loss": 0.1038,
+      "step": 481
+    },
+    {
+      "epoch": 4.384146341463414,
+      "grad_norm": 0.7278075814247131,
+      "learning_rate": 1.0978416875078043e-06,
+      "loss": 0.1045,
+      "step": 482
+    },
+    {
+      "epoch": 4.3932926829268295,
+      "grad_norm": 0.6897367835044861,
+      "learning_rate": 1.0861271518334306e-06,
+      "loss": 0.1026,
+      "step": 483
+    },
+    {
+      "epoch": 4.402439024390244,
+      "grad_norm": 0.5504060387611389,
+      "learning_rate": 1.0744580824374218e-06,
+      "loss": 0.0984,
+      "step": 484
+    },
+    {
+      "epoch": 4.411585365853658,
+      "grad_norm": 0.3768717646598816,
+      "learning_rate": 1.062834854565087e-06,
+      "loss": 0.1028,
+      "step": 485
+    },
+    {
+      "epoch": 4.420731707317073,
+      "grad_norm": 0.6275299191474915,
+      "learning_rate": 1.0512578419876005e-06,
+      "loss": 0.1022,
+      "step": 486
+    },
+    {
+      "epoch": 4.429878048780488,
+      "grad_norm": 0.6650844216346741,
+      "learning_rate": 1.0397274169899802e-06,
+      "loss": 0.1024,
+      "step": 487
+    },
+    {
+      "epoch": 4.439024390243903,
+      "grad_norm": 0.5434347987174988,
+      "learning_rate": 1.0282439503591135e-06,
+      "loss": 0.1013,
+      "step": 488
+    },
+    {
+      "epoch": 4.448170731707317,
+      "grad_norm": 1.1153298616409302,
+      "learning_rate": 1.0168078113718327e-06,
+      "loss": 0.1027,
+      "step": 489
+    },
+    {
+      "epoch": 4.4573170731707314,
+      "grad_norm": 0.6244189143180847,
+      "learning_rate": 1.005419367783048e-06,
+      "loss": 0.1009,
+      "step": 490
+    },
+    {
+      "epoch": 4.466463414634147,
+      "grad_norm": 0.46386802196502686,
+      "learning_rate": 9.940789858139134e-07,
+      "loss": 0.0998,
+      "step": 491
+    },
+    {
+      "epoch": 4.475609756097561,
+      "grad_norm": 0.8419226408004761,
+      "learning_rate": 9.827870301400528e-07,
+      "loss": 0.1041,
+      "step": 492
+    },
+    {
+      "epoch": 4.484756097560975,
+      "grad_norm": 0.4894380271434784,
+      "learning_rate": 9.715438638798315e-07,
+      "loss": 0.1029,
+      "step": 493
+    },
+    {
+      "epoch": 4.4939024390243905,
+      "grad_norm": 1.153964638710022,
+      "learning_rate": 9.60349848582685e-07,
+      "loss": 0.1057,
+      "step": 494
+    },
+    {
+      "epoch": 4.503048780487805,
+      "grad_norm": 0.3604268729686737,
+      "learning_rate": 9.492053442174851e-07,
+      "loss": 0.102,
+      "step": 495
+    },
+    {
+      "epoch": 4.512195121951219,
+      "grad_norm": 0.9208924770355225,
+      "learning_rate": 9.381107091609689e-07,
+      "loss": 0.1025,
+      "step": 496
+    },
+    {
+      "epoch": 4.521341463414634,
+      "grad_norm": 0.47151800990104675,
+      "learning_rate": 9.270663001862129e-07,
+      "loss": 0.1031,
+      "step": 497
+    },
+    {
+      "epoch": 4.530487804878049,
+      "grad_norm": 0.3480950891971588,
+      "learning_rate": 9.160724724511608e-07,
+      "loss": 0.0968,
+      "step": 498
+    },
+    {
+      "epoch": 4.539634146341464,
+      "grad_norm": 0.3497055470943451,
+      "learning_rate": 9.051295794872008e-07,
+      "loss": 0.102,
+      "step": 499
+    },
+    {
+      "epoch": 4.548780487804878,
+      "grad_norm": 0.48209884762763977,
+      "learning_rate": 8.942379731877992e-07,
+      "loss": 0.1047,
+      "step": 500
+    },
+    {
+      "epoch": 4.557926829268292,
+      "grad_norm": 0.4844468832015991,
+      "learning_rate": 8.833980037971862e-07,
+      "loss": 0.1007,
+      "step": 501
+    },
+    {
+      "epoch": 4.567073170731708,
+      "grad_norm": 0.5743635296821594,
+      "learning_rate": 8.726100198990853e-07,
+      "loss": 0.1013,
+      "step": 502
+    },
+    {
+      "epoch": 4.576219512195122,
+      "grad_norm": 0.8410227298736572,
+      "learning_rate": 8.618743684055125e-07,
+      "loss": 0.1021,
+      "step": 503
+    },
+    {
+      "epoch": 4.585365853658536,
+      "grad_norm": 0.7356903553009033,
+      "learning_rate": 8.511913945456151e-07,
+      "loss": 0.0994,
+      "step": 504
+    },
+    {
+      "epoch": 4.594512195121951,
+      "grad_norm": 0.7952869534492493,
+      "learning_rate": 8.40561441854576e-07,
+      "loss": 0.1039,
+      "step": 505
+    },
+    {
+      "epoch": 4.603658536585366,
+      "grad_norm": 0.7460613250732422,
+      "learning_rate": 8.299848521625564e-07,
+      "loss": 0.0997,
+      "step": 506
+    },
+    {
+      "epoch": 4.612804878048781,
+      "grad_norm": 0.4789368510246277,
+      "learning_rate": 8.194619655837135e-07,
+      "loss": 0.0988,
+      "step": 507
+    },
+    {
+      "epoch": 4.621951219512195,
+      "grad_norm": 0.4485442638397217,
+      "learning_rate": 8.089931205052598e-07,
+      "loss": 0.1014,
+      "step": 508
+    },
+    {
+      "epoch": 4.6310975609756095,
+      "grad_norm": 0.36737513542175293,
+      "learning_rate": 7.985786535765794e-07,
+      "loss": 0.0996,
+      "step": 509
+    },
+    {
+      "epoch": 4.640243902439025,
+      "grad_norm": 0.6417019963264465,
+      "learning_rate": 7.882188996984047e-07,
+      "loss": 0.1026,
+      "step": 510
+    },
+    {
+      "epoch": 4.649390243902439,
+      "grad_norm": 0.6503584980964661,
+      "learning_rate": 7.779141920120462e-07,
+      "loss": 0.0981,
+      "step": 511
+    },
+    {
+      "epoch": 4.658536585365853,
+      "grad_norm": 0.3724409341812134,
+      "learning_rate": 7.676648618886797e-07,
+      "loss": 0.1001,
+      "step": 512
+    },
+    {
+      "epoch": 4.6676829268292686,
+      "grad_norm": 0.8001664876937866,
+      "learning_rate": 7.574712389186906e-07,
+      "loss": 0.1014,
+      "step": 513
+    },
+    {
+      "epoch": 4.676829268292683,
+      "grad_norm": 0.48333311080932617,
+      "learning_rate": 7.473336509010742e-07,
+      "loss": 0.098,
+      "step": 514
+    },
+    {
+      "epoch": 4.685975609756097,
+      "grad_norm": 1.0942858457565308,
+      "learning_rate": 7.372524238328955e-07,
+      "loss": 0.1,
+      "step": 515
+    },
+    {
+      "epoch": 4.695121951219512,
+      "grad_norm": 0.48746439814567566,
+      "learning_rate": 7.272278818988054e-07,
+      "loss": 0.1015,
+      "step": 516
+    },
+    {
+      "epoch": 4.704268292682927,
+      "grad_norm": 0.539008617401123,
+      "learning_rate": 7.172603474606169e-07,
+      "loss": 0.1001,
+      "step": 517
+    },
+    {
+      "epoch": 4.713414634146341,
+      "grad_norm": 0.36583229899406433,
+      "learning_rate": 7.073501410469371e-07,
+      "loss": 0.1006,
+      "step": 518
+    },
+    {
+      "epoch": 4.722560975609756,
+      "grad_norm": 0.6509392261505127,
+      "learning_rate": 6.974975813428622e-07,
+      "loss": 0.1006,
+      "step": 519
+    },
+    {
+      "epoch": 4.7317073170731705,
+      "grad_norm": 0.5271562337875366,
+      "learning_rate": 6.877029851797265e-07,
+      "loss": 0.1021,
+      "step": 520
+    },
+    {
+      "epoch": 4.740853658536586,
+      "grad_norm": 0.530583918094635,
+      "learning_rate": 6.779666675249166e-07,
+      "loss": 0.1023,
+      "step": 521
+    },
+    {
+      "epoch": 4.75,
+      "grad_norm": 0.6522917747497559,
+      "learning_rate": 6.682889414717417e-07,
+      "loss": 0.0987,
+      "step": 522
+    },
+    {
+      "epoch": 4.759146341463414,
+      "grad_norm": 0.3822062611579895,
+      "learning_rate": 6.586701182293654e-07,
+      "loss": 0.1004,
+      "step": 523
+    },
+    {
+      "epoch": 4.7682926829268295,
+      "grad_norm": 0.38580814003944397,
+      "learning_rate": 6.491105071127985e-07,
+      "loss": 0.0971,
+      "step": 524
+    },
+    {
+      "epoch": 4.777439024390244,
+      "grad_norm": 0.43107372522354126,
+      "learning_rate": 6.396104155329508e-07,
+      "loss": 0.098,
+      "step": 525
+    },
+    {
+      "epoch": 4.786585365853659,
+      "grad_norm": 0.5299065113067627,
+      "learning_rate": 6.301701489867496e-07,
+      "loss": 0.0997,
+      "step": 526
+    },
+    {
+      "epoch": 4.795731707317073,
+      "grad_norm": 0.7257470488548279,
+      "learning_rate": 6.207900110473112e-07,
+      "loss": 0.0962,
+      "step": 527
+    },
+    {
+      "epoch": 4.804878048780488,
+      "grad_norm": 0.7482612133026123,
+      "learning_rate": 6.114703033541783e-07,
+      "loss": 0.1027,
+      "step": 528
+    },
+    {
+      "epoch": 4.814024390243903,
+      "grad_norm": 0.4249897301197052,
+      "learning_rate": 6.022113256036268e-07,
+      "loss": 0.102,
+      "step": 529
+    },
+    {
+      "epoch": 4.823170731707317,
+      "grad_norm": 0.34573644399642944,
+      "learning_rate": 5.930133755390216e-07,
+      "loss": 0.1018,
+      "step": 530
+    },
+    {
+      "epoch": 4.8323170731707314,
+      "grad_norm": 0.6430623531341553,
+      "learning_rate": 5.838767489412453e-07,
+      "loss": 0.0993,
+      "step": 531
+    },
+    {
+      "epoch": 4.841463414634147,
+      "grad_norm": 0.3944159150123596,
+      "learning_rate": 5.748017396191827e-07,
+      "loss": 0.0999,
+      "step": 532
+    },
+    {
+      "epoch": 4.850609756097561,
+      "grad_norm": 0.5905594229698181,
+      "learning_rate": 5.657886394002802e-07,
+      "loss": 0.1013,
+      "step": 533
+    },
+    {
+      "epoch": 4.859756097560975,
+      "grad_norm": 0.6324522495269775,
+      "learning_rate": 5.568377381211548e-07,
+      "loss": 0.1026,
+      "step": 534
+    },
+    {
+      "epoch": 4.8689024390243905,
+      "grad_norm": 0.6493394374847412,
+      "learning_rate": 5.479493236182759e-07,
+      "loss": 0.1015,
+      "step": 535
+    },
+    {
+      "epoch": 4.878048780487805,
+      "grad_norm": 0.7953840494155884,
+      "learning_rate": 5.391236817187096e-07,
+      "loss": 0.1026,
+      "step": 536
+    },
+    {
+      "epoch": 4.887195121951219,
+      "grad_norm": 0.46242570877075195,
+      "learning_rate": 5.303610962309266e-07,
+      "loss": 0.0993,
+      "step": 537
+    },
+    {
+      "epoch": 4.896341463414634,
+      "grad_norm": 0.6665740609169006,
+      "learning_rate": 5.216618489356773e-07,
+      "loss": 0.1015,
+      "step": 538
+    },
+    {
+      "epoch": 4.905487804878049,
+      "grad_norm": 0.5248611569404602,
+      "learning_rate": 5.130262195769273e-07,
+      "loss": 0.1029,
+      "step": 539
+    },
+    {
+      "epoch": 4.914634146341464,
+      "grad_norm": 0.5713643431663513,
+      "learning_rate": 5.044544858528668e-07,
+      "loss": 0.1021,
+      "step": 540
+    },
+    {
+      "epoch": 4.923780487804878,
+      "grad_norm": 0.7313759326934814,
+      "learning_rate": 4.959469234069735e-07,
+      "loss": 0.0989,
+      "step": 541
+    },
+    {
+      "epoch": 4.932926829268292,
+      "grad_norm": 0.7184482216835022,
+      "learning_rate": 4.87503805819155e-07,
+      "loss": 0.0981,
+      "step": 542
+    },
+    {
+      "epoch": 4.942073170731708,
+      "grad_norm": 0.4711061716079712,
+      "learning_rate": 4.791254045969477e-07,
+      "loss": 0.1,
+      "step": 543
+    },
+    {
+      "epoch": 4.951219512195122,
+      "grad_norm": 0.4654499888420105,
+      "learning_rate": 4.7081198916678924e-07,
+      "loss": 0.0992,
+      "step": 544
+    },
+    {
+      "epoch": 4.960365853658536,
+      "grad_norm": 1.0146021842956543,
+      "learning_rate": 4.6256382686534844e-07,
+      "loss": 0.0993,
+      "step": 545
+    },
+    {
+      "epoch": 4.969512195121951,
+      "grad_norm": 0.490422785282135,
+      "learning_rate": 4.543811829309341e-07,
+      "loss": 0.0999,
+      "step": 546
+    },
+    {
+      "epoch": 4.978658536585366,
+      "grad_norm": 0.8930878043174744,
+      "learning_rate": 4.462643204949654e-07,
+      "loss": 0.1051,
+      "step": 547
+    },
+    {
+      "epoch": 4.987804878048781,
+      "grad_norm": 0.4631318747997284,
+      "learning_rate": 4.382135005735072e-07,
+      "loss": 0.1,
+      "step": 548
+    },
+    {
+      "epoch": 4.996951219512195,
+      "grad_norm": 0.5027859210968018,
+      "learning_rate": 4.302289820588762e-07,
+      "loss": 0.1007,
+      "step": 549
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 0.5027859210968018,
+      "learning_rate": 4.223110217113191e-07,
+      "loss": 0.035,
+      "step": 550
+    },
+    {
+      "epoch": 5.009146341463414,
+      "grad_norm": 0.5998918414115906,
+      "learning_rate": 4.1445987415075634e-07,
+      "loss": 0.1008,
+      "step": 551
+    },
+    {
+      "epoch": 5.0182926829268295,
+      "grad_norm": 0.40694114565849304,
+      "learning_rate": 4.0667579184858865e-07,
+      "loss": 0.1019,
+      "step": 552
+    },
+    {
+      "epoch": 5.027439024390244,
+      "grad_norm": 0.6867284178733826,
+      "learning_rate": 3.989590251195835e-07,
+      "loss": 0.0983,
+      "step": 553
+    },
+    {
+      "epoch": 5.036585365853658,
+      "grad_norm": 0.45372405648231506,
+      "learning_rate": 3.9130982211382367e-07,
+      "loss": 0.0984,
+      "step": 554
+    },
+    {
+      "epoch": 5.045731707317073,
+      "grad_norm": 0.5498050451278687,
+      "learning_rate": 3.837284288087273e-07,
+      "loss": 0.0991,
+      "step": 555
+    },
+    {
+      "epoch": 5.054878048780488,
+      "grad_norm": 1.2504230737686157,
+      "learning_rate": 3.762150890011387e-07,
+      "loss": 0.1017,
+      "step": 556
+    },
+    {
+      "epoch": 5.064024390243903,
+      "grad_norm": 0.6257364153862,
+      "learning_rate": 3.687700442994879e-07,
+      "loss": 0.097,
+      "step": 557
+    },
+    {
+      "epoch": 5.073170731707317,
+      "grad_norm": 2.0217955112457275,
+      "learning_rate": 3.613935341160216e-07,
+      "loss": 0.105,
+      "step": 558
+    },
+    {
+      "epoch": 5.0823170731707314,
+      "grad_norm": 1.0320863723754883,
+      "learning_rate": 3.5408579565910345e-07,
+      "loss": 0.1004,
+      "step": 559
+    },
+    {
+      "epoch": 5.091463414634147,
+      "grad_norm": 0.504863977432251,
+      "learning_rate": 3.468470639255872e-07,
+      "loss": 0.1011,
+      "step": 560
+    },
+    {
+      "epoch": 5.100609756097561,
+      "grad_norm": 0.8961494565010071,
+      "learning_rate": 3.396775716932599e-07,
+      "loss": 0.1,
+      "step": 561
+    },
+    {
+      "epoch": 5.109756097560975,
+      "grad_norm": 0.5523324608802795,
+      "learning_rate": 3.3257754951335457e-07,
+      "loss": 0.0967,
+      "step": 562
+    },
+    {
+      "epoch": 5.1189024390243905,
+      "grad_norm": 0.6520503759384155,
+      "learning_rate": 3.255472257031392e-07,
+      "loss": 0.1025,
+      "step": 563
+    },
+    {
+      "epoch": 5.128048780487805,
+      "grad_norm": 0.47220784425735474,
+      "learning_rate": 3.1858682633857106e-07,
+      "loss": 0.1018,
+      "step": 564
+    },
+    {
+      "epoch": 5.137195121951219,
+      "grad_norm": 1.0287193059921265,
+      "learning_rate": 3.1169657524703164e-07,
+      "loss": 0.0992,
+      "step": 565
+    },
+    {
+      "epoch": 5.146341463414634,
+      "grad_norm": 0.486314058303833,
+      "learning_rate": 3.048766940001238e-07,
+      "loss": 0.0974,
+      "step": 566
+    },
+    {
+      "epoch": 5.155487804878049,
+      "grad_norm": 0.6423690319061279,
+      "learning_rate": 2.981274019065486e-07,
+      "loss": 0.1018,
+      "step": 567
+    },
+    {
+      "epoch": 5.164634146341464,
+      "grad_norm": 0.8213430047035217,
+      "learning_rate": 2.9144891600505543e-07,
+      "loss": 0.0975,
+      "step": 568
+    },
+    {
+      "epoch": 5.173780487804878,
+      "grad_norm": 0.6525201797485352,
+      "learning_rate": 2.8484145105745903e-07,
+      "loss": 0.0964,
+      "step": 569
+    },
+    {
+      "epoch": 5.182926829268292,
+      "grad_norm": 0.7462143898010254,
+      "learning_rate": 2.7830521954173546e-07,
+      "loss": 0.0992,
+      "step": 570
+    },
+    {
+      "epoch": 5.192073170731708,
+      "grad_norm": 0.9441603422164917,
+      "learning_rate": 2.7184043164518643e-07,
+      "loss": 0.0993,
+      "step": 571
+    },
+    {
+      "epoch": 5.201219512195122,
+      "grad_norm": 0.4851154386997223,
+      "learning_rate": 2.6544729525768645e-07,
+      "loss": 0.1052,
+      "step": 572
+    },
+    {
+      "epoch": 5.210365853658536,
+      "grad_norm": 1.151993989944458,
+      "learning_rate": 2.591260159649908e-07,
+      "loss": 0.1014,
+      "step": 573
+    },
+    {
+      "epoch": 5.219512195121951,
+      "grad_norm": 0.48957523703575134,
+      "learning_rate": 2.5287679704212835e-07,
+      "loss": 0.1023,
+      "step": 574
+    },
+    {
+      "epoch": 5.228658536585366,
+      "grad_norm": 0.36362597346305847,
+      "learning_rate": 2.4669983944686393e-07,
+      "loss": 0.1013,
+      "step": 575
+    },
+    {
+      "epoch": 5.237804878048781,
+      "grad_norm": 0.6136613488197327,
+      "learning_rate": 2.405953418132359e-07,
+      "loss": 0.1001,
+      "step": 576
+    },
+    {
+      "epoch": 5.246951219512195,
+      "grad_norm": 0.5123737454414368,
+      "learning_rate": 2.3456350044516825e-07,
+      "loss": 0.1027,
+      "step": 577
+    },
+    {
+      "epoch": 5.2560975609756095,
+      "grad_norm": 0.8742106556892395,
+      "learning_rate": 2.2860450931015855e-07,
+      "loss": 0.1013,
+      "step": 578
+    },
+    {
+      "epoch": 5.265243902439025,
+      "grad_norm": 0.2665225565433502,
+      "learning_rate": 2.2271856003304253e-07,
+      "loss": 0.1001,
+      "step": 579
+    },
+    {
+      "epoch": 5.274390243902439,
+      "grad_norm": 0.5179498791694641,
+      "learning_rate": 2.1690584188982595e-07,
+      "loss": 0.1034,
+      "step": 580
+    },
+    {
+      "epoch": 5.283536585365853,
+      "grad_norm": 0.5693004727363586,
+      "learning_rate": 2.1116654180160512e-07,
+      "loss": 0.0988,
+      "step": 581
+    },
+    {
+      "epoch": 5.2926829268292686,
+      "grad_norm": 0.4897104501724243,
+      "learning_rate": 2.0550084432855055e-07,
+      "loss": 0.1004,
+      "step": 582
+    },
+    {
+      "epoch": 5.301829268292683,
+      "grad_norm": 0.9483790397644043,
+      "learning_rate": 1.9990893166397685e-07,
+      "loss": 0.098,
+      "step": 583
+    },
+    {
+      "epoch": 5.310975609756097,
+      "grad_norm": 0.5504937171936035,
+      "learning_rate": 1.9439098362847825e-07,
+      "loss": 0.0954,
+      "step": 584
+    },
+    {
+      "epoch": 5.320121951219512,
+      "grad_norm": 0.8426586985588074,
+      "learning_rate": 1.8894717766415076e-07,
+      "loss": 0.1014,
+      "step": 585
+    },
+    {
+      "epoch": 5.329268292682927,
+      "grad_norm": 0.7435939311981201,
+      "learning_rate": 1.835776888288851e-07,
+      "loss": 0.0982,
+      "step": 586
+    },
+    {
+      "epoch": 5.338414634146342,
+      "grad_norm": 0.3978838622570038,
+      "learning_rate": 1.7828268979073582e-07,
+      "loss": 0.1,
+      "step": 587
+    },
+    {
+      "epoch": 5.347560975609756,
+      "grad_norm": 0.46983951330184937,
+      "learning_rate": 1.730623508223686e-07,
+      "loss": 0.0974,
+      "step": 588
+    },
+    {
+      "epoch": 5.3567073170731705,
+      "grad_norm": 0.8042597770690918,
+      "learning_rate": 1.6791683979558687e-07,
+      "loss": 0.1003,
+      "step": 589
+    },
+    {
+      "epoch": 5.365853658536586,
+      "grad_norm": 0.40804341435432434,
+      "learning_rate": 1.6284632217593298e-07,
+      "loss": 0.1001,
+      "step": 590
+    },
+    {
+      "epoch": 5.375,
+      "grad_norm": 1.1025656461715698,
+      "learning_rate": 1.5785096101736592e-07,
+      "loss": 0.1013,
+      "step": 591
+    },
+    {
+      "epoch": 5.384146341463414,
+      "grad_norm": 1.113221287727356,
+      "learning_rate": 1.529309169570184e-07,
+      "loss": 0.0981,
+      "step": 592
+    },
+    {
+      "epoch": 5.3932926829268295,
+      "grad_norm": 0.5680567026138306,
+      "learning_rate": 1.4808634821003226e-07,
+      "loss": 0.1007,
+      "step": 593
+    },
+    {
+      "epoch": 5.402439024390244,
+      "grad_norm": 0.33911556005477905,
+      "learning_rate": 1.4331741056446968e-07,
+      "loss": 0.0992,
+      "step": 594
+    },
+    {
+      "epoch": 5.411585365853658,
+      "grad_norm": 0.6437206268310547,
+      "learning_rate": 1.3862425737630385e-07,
+      "loss": 0.1002,
+      "step": 595
+    },
+    {
+      "epoch": 5.420731707317073,
+      "grad_norm": 0.439963698387146,
+      "learning_rate": 1.3400703956448684e-07,
+      "loss": 0.0995,
+      "step": 596
+    },
+    {
+      "epoch": 5.429878048780488,
+      "grad_norm": 0.7596728205680847,
+      "learning_rate": 1.2946590560609819e-07,
+      "loss": 0.0995,
+      "step": 597
+    },
+    {
+      "epoch": 5.439024390243903,
+      "grad_norm": 0.42153531312942505,
+      "learning_rate": 1.250010015315678e-07,
+      "loss": 0.0974,
+      "step": 598
+    },
+    {
+      "epoch": 5.448170731707317,
+      "grad_norm": 0.478438138961792,
+      "learning_rate": 1.2061247091998213e-07,
+      "loss": 0.1013,
+      "step": 599
+    },
+    {
+      "epoch": 5.4573170731707314,
+      "grad_norm": 0.41593241691589355,
+      "learning_rate": 1.1630045489446573e-07,
+      "loss": 0.1038,
+      "step": 600
+    },
+    {
+      "epoch": 5.466463414634147,
+      "grad_norm": 0.45681267976760864,
+      "learning_rate": 1.1206509211764416e-07,
+      "loss": 0.1024,
+      "step": 601
+    },
+    {
+      "epoch": 5.475609756097561,
+      "grad_norm": 0.5991023778915405,
+      "learning_rate": 1.0790651878718389e-07,
+      "loss": 0.1021,
+      "step": 602
+    },
+    {
+      "epoch": 5.484756097560975,
+      "grad_norm": 0.5945449471473694,
+      "learning_rate": 1.0382486863141305e-07,
+      "loss": 0.1011,
+      "step": 603
+    },
+    {
+      "epoch": 5.4939024390243905,
+      "grad_norm": 0.7431889772415161,
+      "learning_rate": 9.982027290502238e-08,
+      "loss": 0.0983,
+      "step": 604
+    },
+    {
+      "epoch": 5.503048780487805,
+      "grad_norm": 0.46923258900642395,
+      "learning_rate": 9.589286038484219e-08,
+      "loss": 0.1021,
+      "step": 605
+    },
+    {
+      "epoch": 5.512195121951219,
+      "grad_norm": 0.5379825830459595,
+      "learning_rate": 9.20427573657015e-08,
+      "loss": 0.1008,
+      "step": 606
+    },
+    {
+      "epoch": 5.521341463414634,
+      "grad_norm": 1.3347440958023071,
+      "learning_rate": 8.827008765636941e-08,
+      "loss": 0.102,
+      "step": 607
+    },
+    {
+      "epoch": 5.530487804878049,
+      "grad_norm": 0.735386848449707,
+      "learning_rate": 8.457497257557041e-08,
+      "loss": 0.1029,
+      "step": 608
+    },
+    {
+      "epoch": 5.539634146341464,
+      "grad_norm": 0.45335713028907776,
+      "learning_rate": 8.095753094808505e-08,
+      "loss": 0.1023,
+      "step": 609
+    },
+    {
+      "epoch": 5.548780487804878,
+      "grad_norm": 0.5249711871147156,
+      "learning_rate": 7.741787910092808e-08,
+      "loss": 0.1033,
+      "step": 610
+    },
+    {
+      "epoch": 5.557926829268292,
+      "grad_norm": 0.5457543730735779,
+      "learning_rate": 7.395613085960874e-08,
+      "loss": 0.1025,
+      "step": 611
+    },
+    {
+      "epoch": 5.567073170731708,
+      "grad_norm": 0.6174072027206421,
+      "learning_rate": 7.05723975444686e-08,
+      "loss": 0.0997,
+      "step": 612
+    },
+    {
+      "epoch": 5.576219512195122,
+      "grad_norm": 0.5054624676704407,
+      "learning_rate": 6.726678796710378e-08,
+      "loss": 0.1029,
+      "step": 613
+    },
+    {
+      "epoch": 5.585365853658536,
+      "grad_norm": 0.6858559846878052,
+      "learning_rate": 6.403940842686473e-08,
+      "loss": 0.1004,
+      "step": 614
+    },
+    {
+      "epoch": 5.594512195121951,
+      "grad_norm": 0.3704938292503357,
+      "learning_rate": 6.089036270743754e-08,
+      "loss": 0.0994,
+      "step": 615
+    },
+    {
+      "epoch": 5.603658536585366,
+      "grad_norm": 0.4708552360534668,
+      "learning_rate": 5.7819752073508263e-08,
+      "loss": 0.1002,
+      "step": 616
+    },
+    {
+      "epoch": 5.612804878048781,
+      "grad_norm": 0.34890425205230713,
+      "learning_rate": 5.4827675267504213e-08,
+      "loss": 0.1015,
+      "step": 617
+    },
+    {
+      "epoch": 5.621951219512195,
+      "grad_norm": 0.6657831072807312,
+      "learning_rate": 5.191422850642114e-08,
+      "loss": 0.099,
+      "step": 618
+    },
+    {
+      "epoch": 5.6310975609756095,
+      "grad_norm": 0.6973615884780884,
+      "learning_rate": 4.9079505478726866e-08,
+      "loss": 0.0994,
+      "step": 619
+    },
+    {
+      "epoch": 5.640243902439025,
+      "grad_norm": 0.8337546586990356,
+      "learning_rate": 4.6323597341350055e-08,
+      "loss": 0.0985,
+      "step": 620
+    },
+    {
+      "epoch": 5.649390243902439,
+      "grad_norm": 0.4006950855255127,
+      "learning_rate": 4.3646592716748423e-08,
+      "loss": 0.0991,
+      "step": 621
+    },
+    {
+      "epoch": 5.658536585365853,
+      "grad_norm": 0.84935063123703,
+      "learning_rate": 4.104857769005876e-08,
+      "loss": 0.1001,
+      "step": 622
+    },
+    {
+      "epoch": 5.6676829268292686,
+      "grad_norm": 0.5318524837493896,
+      "learning_rate": 3.852963580632835e-08,
+      "loss": 0.098,
+      "step": 623
+    },
+    {
+      "epoch": 5.676829268292683,
+      "grad_norm": 0.7271676659584045,
+      "learning_rate": 3.608984806782928e-08,
+      "loss": 0.1009,
+      "step": 624
+    },
+    {
+      "epoch": 5.685975609756097,
+      "grad_norm": 0.33880215883255005,
+      "learning_rate": 3.372929293145283e-08,
+      "loss": 0.1015,
+      "step": 625
+    },
+    {
+      "epoch": 5.695121951219512,
+      "grad_norm": 0.4488051235675812,
+      "learning_rate": 3.144804630618725e-08,
+      "loss": 0.103,
+      "step": 626
+    },
+    {
+      "epoch": 5.704268292682927,
+      "grad_norm": 0.7648248076438904,
+      "learning_rate": 2.92461815506756e-08,
+      "loss": 0.1006,
+      "step": 627
+    },
+    {
+      "epoch": 5.713414634146341,
+      "grad_norm": 0.5335168242454529,
+      "learning_rate": 2.7123769470857596e-08,
+      "loss": 0.1005,
+      "step": 628
+    },
+    {
+      "epoch": 5.722560975609756,
+      "grad_norm": 0.672450602054596,
+      "learning_rate": 2.5080878317693126e-08,
+      "loss": 0.0967,
+      "step": 629
+    },
+    {
+      "epoch": 5.7317073170731705,
+      "grad_norm": 0.8220935463905334,
+      "learning_rate": 2.3117573784966206e-08,
+      "loss": 0.1024,
+      "step": 630
+    },
+    {
+      "epoch": 5.740853658536586,
+      "grad_norm": 0.5261617302894592,
+      "learning_rate": 2.1233919007173086e-08,
+      "loss": 0.1005,
+      "step": 631
+    },
+    {
+      "epoch": 5.75,
+      "grad_norm": 0.568458616733551,
+      "learning_rate": 1.9429974557493016e-08,
+      "loss": 0.0971,
+      "step": 632
+    },
+    {
+      "epoch": 5.759146341463414,
+      "grad_norm": 0.8193382024765015,
+      "learning_rate": 1.770579844583814e-08,
+      "loss": 0.0969,
+      "step": 633
+    },
+    {
+      "epoch": 5.7682926829268295,
+      "grad_norm": 0.9471128582954407,
+      "learning_rate": 1.606144611699001e-08,
+      "loss": 0.1029,
+      "step": 634
+    },
+    {
+      "epoch": 5.777439024390244,
+      "grad_norm": 0.6565077304840088,
+      "learning_rate": 1.4496970448815706e-08,
+      "loss": 0.1014,
+      "step": 635
+    },
+    {
+      "epoch": 5.786585365853659,
+      "grad_norm": 0.5950571298599243,
+      "learning_rate": 1.3012421750568105e-08,
+      "loss": 0.0995,
+      "step": 636
+    },
+    {
+      "epoch": 5.795731707317073,
+      "grad_norm": 0.5122717022895813,
+      "learning_rate": 1.1607847761267165e-08,
+      "loss": 0.102,
+      "step": 637
+    },
+    {
+      "epoch": 5.804878048780488,
+      "grad_norm": 0.5124772191047668,
+      "learning_rate": 1.0283293648165605e-08,
+      "loss": 0.101,
+      "step": 638
+    },
+    {
+      "epoch": 5.814024390243903,
+      "grad_norm": 0.4995802342891693,
+      "learning_rate": 9.03880200529561e-09,
+      "loss": 0.1002,
+      "step": 639
+    },
+    {
+      "epoch": 5.823170731707317,
+      "grad_norm": 0.41052404046058655,
+      "learning_rate": 7.874412852099944e-09,
+      "loss": 0.0998,
+      "step": 640
+    },
+    {
+      "epoch": 5.8323170731707314,
+      "grad_norm": 1.3058655261993408,
+      "learning_rate": 6.790163632144631e-09,
+      "loss": 0.0996,
+      "step": 641
+    },
+    {
+      "epoch": 5.841463414634147,
+      "grad_norm": 0.4820384383201599,
+      "learning_rate": 5.7860892119152025e-09,
+      "loss": 0.0983,
+      "step": 642
+    },
+    {
+      "epoch": 5.850609756097561,
+      "grad_norm": 0.6845442652702332,
+      "learning_rate": 4.862221879694817e-09,
+      "loss": 0.0991,
+      "step": 643
+    },
+    {
+      "epoch": 5.859756097560975,
+      "grad_norm": 0.8762632012367249,
+      "learning_rate": 4.018591344526479e-09,
+      "loss": 0.0986,
+      "step": 644
+    },
+    {
+      "epoch": 5.8689024390243905,
+      "grad_norm": 0.8684572577476501,
+      "learning_rate": 3.255224735257412e-09,
+      "loss": 0.0978,
+      "step": 645
+    },
+    {
+      "epoch": 5.878048780487805,
+      "grad_norm": 0.3539469242095947,
+      "learning_rate": 2.5721465996675352e-09,
+      "loss": 0.0991,
+      "step": 646
+    },
+    {
+      "epoch": 5.887195121951219,
+      "grad_norm": 0.5980482697486877,
+      "learning_rate": 1.9693789036781543e-09,
+      "loss": 0.0981,
+      "step": 647
+    },
+    {
+      "epoch": 5.896341463414634,
+      "grad_norm": 0.4726741909980774,
+      "learning_rate": 1.4469410306480747e-09,
+      "loss": 0.0997,
+      "step": 648
+    },
+    {
+      "epoch": 5.905487804878049,
+      "grad_norm": 0.4221959114074707,
+      "learning_rate": 1.0048497807479962e-09,
+      "loss": 0.0974,
+      "step": 649
+    },
+    {
+      "epoch": 5.914634146341464,
+      "grad_norm": 0.7107351422309875,
+      "learning_rate": 6.431193704217742e-10,
+      "loss": 0.1022,
+      "step": 650
+    },
+    {
+      "epoch": 5.923780487804878,
+      "grad_norm": 0.5341996550559998,
+      "learning_rate": 3.6176143192873105e-10,
+      "loss": 0.0984,
+      "step": 651
+    },
+    {
+      "epoch": 5.932926829268292,
+      "grad_norm": 0.4433972239494324,
+      "learning_rate": 1.6078501296951098e-10,
+      "loss": 0.1031,
+      "step": 652
+    },
+    {
+      "epoch": 5.942073170731708,
+      "grad_norm": 0.5114858746528625,
+      "learning_rate": 4.019657639520169e-11,
+      "loss": 0.1025,
+      "step": 653
+    },
+    {
+      "epoch": 5.951219512195122,
+      "grad_norm": 0.48689013719558716,
+      "learning_rate": 0.0,
+      "loss": 0.1027,
+      "step": 654
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 654,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 6,
+  "save_steps": 109,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.9748247438964556e+19,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}