diff --git "a/checkpoint-3200/trainer_state.json" "b/checkpoint-3200/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-3200/trainer_state.json"
@@ -0,0 +1,19513 @@
+{
+  "best_metric": 6.423073768615723,
+  "best_model_checkpoint": "./output_v2/34bCodellama_CodeLlama-34b-Python-hf_unnatural-instructions_standardized/checkpoint-3200",
+  "epoch": 0.024444274692536856,
+  "global_step": 3200,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.0808,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.8773,
+      "step": 2
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.1965,
+      "step": 3
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.118,
+      "step": 4
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.1773,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.1165,
+      "step": 6
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.2666,
+      "step": 7
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.3704,
+      "step": 8
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.9976,
+      "step": 9
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.985,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.0541,
+      "step": 11
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.6228,
+      "step": 12
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.3651,
+      "step": 13
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.0867,
+      "step": 14
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.4422,
+      "step": 15
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.7759,
+      "step": 16
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.1446,
+      "step": 17
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.0007,
+      "step": 18
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.0894,
+      "step": 19
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2424,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.1343,
+      "step": 21
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.5354,
+      "step": 22
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.1887,
+      "step": 23
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.6652,
+      "step": 24
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.964,
+      "step": 25
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.1872,
+      "step": 26
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.4722,
+      "step": 27
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.1462,
+      "step": 28
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.0485,
+      "step": 29
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.148,
+      "step": 30
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.7274,
+      "step": 31
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.6689,
+      "step": 32
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.3384,
+      "step": 33
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.5354,
+      "step": 34
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.1976,
+      "step": 35
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.8593,
+      "step": 36
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.9302,
+      "step": 37
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.5968,
+      "step": 38
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.3169,
+      "step": 39
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.1793,
+      "step": 40
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.8457,
+      "step": 41
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.5177,
+      "step": 42
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.003,
+      "step": 43
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.9928,
+      "step": 44
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 15.2574,
+      "step": 45
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 0.3915,
+      "step": 46
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.4105,
+      "step": 47
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.1184,
+      "step": 48
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.72,
+      "step": 49
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.9628,
+      "step": 50
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.2372,
+      "step": 51
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3733,
+      "step": 52
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.8936,
+      "step": 53
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.5353,
+      "step": 54
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.0754,
+      "step": 55
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.6685,
+      "step": 56
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.8984,
+      "step": 57
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.2265,
+      "step": 58
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.7696,
+      "step": 59
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.7349,
+      "step": 60
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.0221,
+      "step": 61
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 15.1901,
+      "step": 62
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.387,
+      "step": 63
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.7323,
+      "step": 64
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.2077,
+      "step": 65
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.3155,
+      "step": 66
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.1656,
+      "step": 67
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 13.0828,
+      "step": 68
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.5295,
+      "step": 69
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.4575,
+      "step": 70
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 14.7654,
+      "step": 71
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.6263,
+      "step": 72
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 24.8238,
+      "step": 73
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 15.0654,
+      "step": 74
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 28.1046,
+      "step": 75
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 14.3232,
+      "step": 76
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 22.9712,
+      "step": 77
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 18.8529,
+      "step": 78
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 15.8356,
+      "step": 79
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 16.472,
+      "step": 80
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 12.2369,
+      "step": 81
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 14.0731,
+      "step": 82
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.8853,
+      "step": 83
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.5438,
+      "step": 84
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.2665,
+      "step": 85
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 11.5484,
+      "step": 86
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.7546,
+      "step": 87
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.4309,
+      "step": 88
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 11.5593,
+      "step": 89
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.3822,
+      "step": 90
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.6315,
+      "step": 91
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.6116,
+      "step": 92
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.2288,
+      "step": 93
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.0053,
+      "step": 94
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 12.359,
+      "step": 95
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9235,
+      "step": 96
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 31.9845,
+      "step": 97
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 11.1385,
+      "step": 98
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.6161,
+      "step": 99
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.8096,
+      "step": 100
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.9918,
+      "step": 101
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.344,
+      "step": 102
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.1607,
+      "step": 103
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.4834,
+      "step": 104
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.704,
+      "step": 105
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.1238,
+      "step": 106
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.8066,
+      "step": 107
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9656,
+      "step": 108
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.1979,
+      "step": 109
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2294,
+      "step": 110
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.066,
+      "step": 111
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7914,
+      "step": 112
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.7344,
+      "step": 113
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.6703,
+      "step": 114
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.8817,
+      "step": 115
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 11.7733,
+      "step": 116
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.469,
+      "step": 117
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.1304,
+      "step": 118
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.871,
+      "step": 119
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.5353,
+      "step": 120
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.9055,
+      "step": 121
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.6142,
+      "step": 122
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.0201,
+      "step": 123
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.3805,
+      "step": 124
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.6825,
+      "step": 125
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.7166,
+      "step": 126
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.7747,
+      "step": 127
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.7695,
+      "step": 128
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.7291,
+      "step": 129
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.1296,
+      "step": 130
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.5374,
+      "step": 131
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.1854,
+      "step": 132
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.434,
+      "step": 133
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.438,
+      "step": 134
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3027,
+      "step": 135
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.382,
+      "step": 136
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.9277,
+      "step": 137
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.223,
+      "step": 138
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.3042,
+      "step": 139
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.6361,
+      "step": 140
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.3547,
+      "step": 141
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.7181,
+      "step": 142
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.7528,
+      "step": 143
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 11.4316,
+      "step": 144
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.2219,
+      "step": 145
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.7788,
+      "step": 146
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.2749,
+      "step": 147
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.2397,
+      "step": 148
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.6243,
+      "step": 149
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.145,
+      "step": 150
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.7951,
+      "step": 151
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.1862,
+      "step": 152
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.1305,
+      "step": 153
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.5766,
+      "step": 154
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9232,
+      "step": 155
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.9936,
+      "step": 156
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.9692,
+      "step": 157
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.2772,
+      "step": 158
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.302,
+      "step": 159
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9931,
+      "step": 160
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.9675,
+      "step": 161
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.8536,
+      "step": 162
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.6589,
+      "step": 163
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.932,
+      "step": 164
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.0301,
+      "step": 165
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4861,
+      "step": 166
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.1354,
+      "step": 167
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0717,
+      "step": 168
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.9346,
+      "step": 169
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9373,
+      "step": 170
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.8777,
+      "step": 171
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.4193,
+      "step": 172
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.6831,
+      "step": 173
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4175,
+      "step": 174
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.3629,
+      "step": 175
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.118,
+      "step": 176
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.633,
+      "step": 177
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.8355,
+      "step": 178
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.4522,
+      "step": 179
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.9272,
+      "step": 180
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.4631,
+      "step": 181
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.2987,
+      "step": 182
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.1183,
+      "step": 183
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.9976,
+      "step": 184
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.0668,
+      "step": 185
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.6291,
+      "step": 186
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.5937,
+      "step": 187
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.7382,
+      "step": 188
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.7677,
+      "step": 189
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.0293,
+      "step": 190
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.6407,
+      "step": 191
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.9508,
+      "step": 192
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.5053,
+      "step": 193
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.5718,
+      "step": 194
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.5211,
+      "step": 195
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9557,
+      "step": 196
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.1609,
+      "step": 197
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.8505,
+      "step": 198
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.8278,
+      "step": 199
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.8447,
+      "step": 200
+    },
+    {
+      "epoch": 0.0,
+      "eval_loss": 7.883856773376465,
+      "eval_runtime": 22.4254,
+      "eval_samples_per_second": 2.23,
+      "eval_steps_per_second": 1.115,
+      "step": 200
+    },
+    {
+      "epoch": 0.0,
+      "mmlu_eval_accuracy": 0.2525477994227994,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.07142857142857142,
+      "mmlu_eval_accuracy_astronomy": 0.3125,
+      "mmlu_eval_accuracy_business_ethics": 0.4444444444444444,
+      "mmlu_loss": 4.629522514343262,
+      "step": 200
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.3249,
+      "step": 201
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.352,
+      "step": 202
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2984,
+      "step": 203
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 11.2734,
+      "step": 204
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.1,
+      "step": 205
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.448,
+      "step": 206
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.2387,
+      "step": 207
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.861,
+      "step": 208
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.603,
+      "step": 209
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.29,
+      "step": 210
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2105,
+      "step": 211
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.1949,
+      "step": 212
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.0538,
+      "step": 213
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0343,
+      "step": 214
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.7794,
+      "step": 215
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.5532,
+      "step": 216
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2676,
+      "step": 217
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.566,
+      "step": 218
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.0432,
+      "step": 219
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9391,
+      "step": 220
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.724,
+      "step": 221
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.229,
+      "step": 222
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.3462,
+      "step": 223
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0752,
+      "step": 224
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.1966,
+      "step": 225
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7279,
+      "step": 226
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.8484,
+      "step": 227
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7291,
+      "step": 228
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.2665,
+      "step": 229
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.3551,
+      "step": 230
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.7338,
+      "step": 231
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.8407,
+      "step": 232
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.3581,
+      "step": 233
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.441,
+      "step": 234
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.0788,
+      "step": 235
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.8404,
+      "step": 236
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.4314,
+      "step": 237
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.8426,
+      "step": 238
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.0205,
+      "step": 239
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4162,
+      "step": 240
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.7515,
+      "step": 241
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.1442,
+      "step": 242
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.5868,
+      "step": 243
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.6514,
+      "step": 244
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2683,
+      "step": 245
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.31,
+      "step": 246
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0161,
+      "step": 247
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.484,
+      "step": 248
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.9726,
+      "step": 249
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.0926,
+      "step": 250
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.5279,
+      "step": 251
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.0017,
+      "step": 252
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.5684,
+      "step": 253
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.3875,
+      "step": 254
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.9489,
+      "step": 255
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.8948,
+      "step": 256
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.0856,
+      "step": 257
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.599,
+      "step": 258
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.1575,
+      "step": 259
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.3701,
+      "step": 260
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.464,
+      "step": 261
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9193,
+      "step": 262
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.5679,
+      "step": 263
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9424,
+      "step": 264
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.6689,
+      "step": 265
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.6475,
+      "step": 266
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.4311,
+      "step": 267
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.7426,
+      "step": 268
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.5191,
+      "step": 269
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.3059,
+      "step": 270
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.0142,
+      "step": 271
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.4509,
+      "step": 272
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.0831,
+      "step": 273
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.6977,
+      "step": 274
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.4236,
+      "step": 275
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2129,
+      "step": 276
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.1394,
+      "step": 277
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.685,
+      "step": 278
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.0275,
+      "step": 279
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 11.3215,
+      "step": 280
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.6542,
+      "step": 281
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.7614,
+      "step": 282
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.2996,
+      "step": 283
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.6275,
+      "step": 284
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.8736,
+      "step": 285
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.4667,
+      "step": 286
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.8486,
+      "step": 287
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2125,
+      "step": 288
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.4523,
+      "step": 289
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.551,
+      "step": 290
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 11.7158,
+      "step": 291
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.5092,
+      "step": 292
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9169,
+      "step": 293
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.5333,
+      "step": 294
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9949,
+      "step": 295
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.7189,
+      "step": 296
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.2366,
+      "step": 297
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.4745,
+      "step": 298
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.2439,
+      "step": 299
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4176,
+      "step": 300
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.9365,
+      "step": 301
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.5309,
+      "step": 302
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2201,
+      "step": 303
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.0312,
+      "step": 304
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.4173,
+      "step": 305
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.4856,
+      "step": 306
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.5041,
+      "step": 307
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.3597,
+      "step": 308
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.8395,
+      "step": 309
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.0776,
+      "step": 310
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.7566,
+      "step": 311
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.9767,
+      "step": 312
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.3804,
+      "step": 313
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 11.5327,
+      "step": 314
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.5293,
+      "step": 315
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4531,
+      "step": 316
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.3961,
+      "step": 317
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.5669,
+      "step": 318
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.8559,
+      "step": 319
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.117,
+      "step": 320
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.4279,
+      "step": 321
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.7977,
+      "step": 322
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.955,
+      "step": 323
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0164,
+      "step": 324
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 12.0495,
+      "step": 325
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.2768,
+      "step": 326
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.3162,
+      "step": 327
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.88,
+      "step": 328
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.2157,
+      "step": 329
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.8427,
+      "step": 330
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.9729,
+      "step": 331
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.1779,
+      "step": 332
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.1302,
+      "step": 333
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7705,
+      "step": 334
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.523,
+      "step": 335
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9375,
+      "step": 336
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.1409,
+      "step": 337
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.633,
+      "step": 338
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.6481,
+      "step": 339
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.933,
+      "step": 340
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.9179,
+      "step": 341
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.9332,
+      "step": 342
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.6553,
+      "step": 343
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.7412,
+      "step": 344
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.849,
+      "step": 345
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.7321,
+      "step": 346
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9717,
+      "step": 347
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.3465,
+      "step": 348
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4535,
+      "step": 349
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.2376,
+      "step": 350
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9025,
+      "step": 351
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.916,
+      "step": 352
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.3785,
+      "step": 353
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0576,
+      "step": 354
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.5081,
+      "step": 355
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.1303,
+      "step": 356
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.3854,
+      "step": 357
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 11.5553,
+      "step": 358
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9627,
+      "step": 359
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.402,
+      "step": 360
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3484,
+      "step": 361
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.5428,
+      "step": 362
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9128,
+      "step": 363
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3934,
+      "step": 364
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.4812,
+      "step": 365
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.5395,
+      "step": 366
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.6304,
+      "step": 367
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.5626,
+      "step": 368
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.5693,
+      "step": 369
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.3458,
+      "step": 370
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.6254,
+      "step": 371
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.8706,
+      "step": 372
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.6076,
+      "step": 373
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.2912,
+      "step": 374
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.3326,
+      "step": 375
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.3735,
+      "step": 376
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.4916,
+      "step": 377
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.5553,
+      "step": 378
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.6241,
+      "step": 379
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.6106,
+      "step": 380
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.266,
+      "step": 381
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7738,
+      "step": 382
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.4988,
+      "step": 383
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2968,
+      "step": 384
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.8512,
+      "step": 385
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0341,
+      "step": 386
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.898,
+      "step": 387
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.23,
+      "step": 388
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.9608,
+      "step": 389
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.3679,
+      "step": 390
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.7074,
+      "step": 391
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9903,
+      "step": 392
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.5845,
+      "step": 393
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.6493,
+      "step": 394
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7962,
+      "step": 395
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4865,
+      "step": 396
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.3418,
+      "step": 397
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.3942,
+      "step": 398
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.4715,
+      "step": 399
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.2073,
+      "step": 400
+    },
+    {
+      "epoch": 0.0,
+      "eval_loss": 7.106412410736084,
+      "eval_runtime": 22.5667,
+      "eval_samples_per_second": 2.216,
+      "eval_steps_per_second": 1.108,
+      "step": 400
+    },
+    {
+      "epoch": 0.0,
+      "mmlu_eval_accuracy": 0.2525477994227994,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.07142857142857142,
+      "mmlu_eval_accuracy_astronomy": 0.3125,
+      "mmlu_eval_accuracy_business_ethics": 0.4444444444444444,
+      "mmlu_loss": 2.9128687667846678,
+      "step": 400
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.3984,
+      "step": 401
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.7983,
+      "step": 402
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.8589,
+      "step": 403
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.9884,
+      "step": 404
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.4427,
+      "step": 405
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0374,
+      "step": 406
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.7999,
+      "step": 407
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.2437,
+      "step": 408
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.6902,
+      "step": 409
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.81,
+      "step": 410
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.8979,
+      "step": 411
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0211,
+      "step": 412
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.3945,
+      "step": 413
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.5807,
+      "step": 414
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.1433,
+      "step": 415
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9466,
+      "step": 416
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.6276,
+      "step": 417
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.4945,
+      "step": 418
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.6215,
+      "step": 419
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.3919,
+      "step": 420
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.7915,
+      "step": 421
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.3284,
+      "step": 422
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.8723,
+      "step": 423
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.0149,
+      "step": 424
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.979,
+      "step": 425
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.9175,
+      "step": 426
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.4994,
+      "step": 427
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.9791,
+      "step": 428
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.1156,
+      "step": 429
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.5813,
+      "step": 430
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.1882,
+      "step": 431
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.9956,
+      "step": 432
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.6189,
+      "step": 433
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.9624,
+      "step": 434
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.5387,
+      "step": 435
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4605,
+      "step": 436
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.474,
+      "step": 437
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0497,
+      "step": 438
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.5705,
+      "step": 439
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.275,
+      "step": 440
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.9638,
+      "step": 441
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.4857,
+      "step": 442
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3067,
+      "step": 443
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.8152,
+      "step": 444
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.1668,
+      "step": 445
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.5293,
+      "step": 446
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3981,
+      "step": 447
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.4787,
+      "step": 448
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.5981,
+      "step": 449
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.3569,
+      "step": 450
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4088,
+      "step": 451
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.3677,
+      "step": 452
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.4686,
+      "step": 453
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.3552,
+      "step": 454
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.7931,
+      "step": 455
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.9285,
+      "step": 456
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.0554,
+      "step": 457
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.7277,
+      "step": 458
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2474,
+      "step": 459
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.9274,
+      "step": 460
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.2558,
+      "step": 461
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.7547,
+      "step": 462
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.1264,
+      "step": 463
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.2124,
+      "step": 464
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.8751,
+      "step": 465
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.7317,
+      "step": 466
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3697,
+      "step": 467
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0021,
+      "step": 468
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3761,
+      "step": 469
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2291,
+      "step": 470
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7968,
+      "step": 471
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.9454,
+      "step": 472
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.0194,
+      "step": 473
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.5048,
+      "step": 474
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.6837,
+      "step": 475
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.1066,
+      "step": 476
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.3501,
+      "step": 477
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.5071,
+      "step": 478
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.1086,
+      "step": 479
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.7269,
+      "step": 480
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.5419,
+      "step": 481
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.2974,
+      "step": 482
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.1433,
+      "step": 483
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0869,
+      "step": 484
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.032,
+      "step": 485
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.0946,
+      "step": 486
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.7162,
+      "step": 487
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.0406,
+      "step": 488
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.9048,
+      "step": 489
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.2231,
+      "step": 490
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.6524,
+      "step": 491
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.1151,
+      "step": 492
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.591,
+      "step": 493
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.1628,
+      "step": 494
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.0757,
+      "step": 495
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.3471,
+      "step": 496
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9385,
+      "step": 497
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9362,
+      "step": 498
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.2252,
+      "step": 499
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.359,
+      "step": 500
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0497,
+      "step": 501
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.0484,
+      "step": 502
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.5773,
+      "step": 503
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.39,
+      "step": 504
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.5923,
+      "step": 505
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2,
+      "step": 506
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.5536,
+      "step": 507
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.8958,
+      "step": 508
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7763,
+      "step": 509
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.2045,
+      "step": 510
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.4219,
+      "step": 511
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.6305,
+      "step": 512
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.4243,
+      "step": 513
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7842,
+      "step": 514
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.8769,
+      "step": 515
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.8903,
+      "step": 516
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.0489,
+      "step": 517
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.1314,
+      "step": 518
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.5973,
+      "step": 519
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.8022,
+      "step": 520
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3539,
+      "step": 521
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.222,
+      "step": 522
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.5403,
+      "step": 523
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.1323,
+      "step": 524
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7813,
+      "step": 525
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.4982,
+      "step": 526
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2426,
+      "step": 527
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.0142,
+      "step": 528
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.8996,
+      "step": 529
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.8671,
+      "step": 530
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.4139,
+      "step": 531
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9478,
+      "step": 532
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.7062,
+      "step": 533
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.0098,
+      "step": 534
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.9195,
+      "step": 535
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.0255,
+      "step": 536
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.6291,
+      "step": 537
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.3245,
+      "step": 538
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.6382,
+      "step": 539
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 1.8076,
+      "step": 540
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.6725,
+      "step": 541
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.0563,
+      "step": 542
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.6178,
+      "step": 543
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.7974,
+      "step": 544
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.7535,
+      "step": 545
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4948,
+      "step": 546
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.8941,
+      "step": 547
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.6496,
+      "step": 548
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.9084,
+      "step": 549
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.65,
+      "step": 550
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7673,
+      "step": 551
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 10.2221,
+      "step": 552
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.14,
+      "step": 553
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.6747,
+      "step": 554
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.8009,
+      "step": 555
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.7307,
+      "step": 556
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.0143,
+      "step": 557
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.8098,
+      "step": 558
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.026,
+      "step": 559
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.4572,
+      "step": 560
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.7913,
+      "step": 561
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9962,
+      "step": 562
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.767,
+      "step": 563
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.9497,
+      "step": 564
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.9626,
+      "step": 565
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2536,
+      "step": 566
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.0421,
+      "step": 567
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.8177,
+      "step": 568
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9241,
+      "step": 569
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.0162,
+      "step": 570
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.3368,
+      "step": 571
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7515,
+      "step": 572
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.6389,
+      "step": 573
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.662,
+      "step": 574
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.8097,
+      "step": 575
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9346,
+      "step": 576
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.3154,
+      "step": 577
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.7724,
+      "step": 578
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3685,
+      "step": 579
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.2775,
+      "step": 580
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.106,
+      "step": 581
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.4733,
+      "step": 582
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.2334,
+      "step": 583
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9478,
+      "step": 584
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.0013,
+      "step": 585
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.7242,
+      "step": 586
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.922,
+      "step": 587
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.1418,
+      "step": 588
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4472,
+      "step": 589
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.4785,
+      "step": 590
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.783,
+      "step": 591
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.0706,
+      "step": 592
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.4136,
+      "step": 593
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.5969,
+      "step": 594
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.5157,
+      "step": 595
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.5658,
+      "step": 596
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.4647,
+      "step": 597
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.2028,
+      "step": 598
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.6913,
+      "step": 599
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.7239,
+      "step": 600
+    },
+    {
+      "epoch": 0.0,
+      "eval_loss": 7.012163162231445,
+      "eval_runtime": 22.5807,
+      "eval_samples_per_second": 2.214,
+      "eval_steps_per_second": 1.107,
+      "step": 600
+    },
+    {
+      "epoch": 0.0,
+      "mmlu_eval_accuracy": 0.3260281385281385,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.35714285714285715,
+      "mmlu_eval_accuracy_astronomy": 0.25,
+      "mmlu_eval_accuracy_business_ethics": 0.3333333333333333,
+      "mmlu_loss": 4.24488224029541,
+      "step": 600
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.5253,
+      "step": 601
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0392,
+      "step": 602
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.447,
+      "step": 603
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.9441,
+      "step": 604
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.1874,
+      "step": 605
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.7817,
+      "step": 606
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.0348,
+      "step": 607
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.5593,
+      "step": 608
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.9361,
+      "step": 609
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3534,
+      "step": 610
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.476,
+      "step": 611
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.0937,
+      "step": 612
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3027,
+      "step": 613
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.5586,
+      "step": 614
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3796,
+      "step": 615
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.676,
+      "step": 616
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.5321,
+      "step": 617
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.0059,
+      "step": 618
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.6139,
+      "step": 619
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.2391,
+      "step": 620
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.0636,
+      "step": 621
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.0895,
+      "step": 622
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.62,
+      "step": 623
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.0469,
+      "step": 624
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.2173,
+      "step": 625
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 3.9432,
+      "step": 626
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.3928,
+      "step": 627
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.0959,
+      "step": 628
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.1197,
+      "step": 629
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 2.4277,
+      "step": 630
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.418,
+      "step": 631
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.8687,
+      "step": 632
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.0156,
+      "step": 633
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.573,
+      "step": 634
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.112,
+      "step": 635
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.8954,
+      "step": 636
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.36,
+      "step": 637
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.924,
+      "step": 638
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.4625,
+      "step": 639
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.2023,
+      "step": 640
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.0685,
+      "step": 641
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.5304,
+      "step": 642
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4456,
+      "step": 643
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.7271,
+      "step": 644
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.6011,
+      "step": 645
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.895,
+      "step": 646
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 4.864,
+      "step": 647
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.3452,
+      "step": 648
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 5.8978,
+      "step": 649
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.2253,
+      "step": 650
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 9.2813,
+      "step": 651
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 7.7248,
+      "step": 652
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 8.4283,
+      "step": 653
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0004,
+      "loss": 6.4304,
+      "step": 654
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3893,
+      "step": 655
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1115,
+      "step": 656
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.5892,
+      "step": 657
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6572,
+      "step": 658
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.925,
+      "step": 659
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4431,
+      "step": 660
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7711,
+      "step": 661
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9439,
+      "step": 662
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3781,
+      "step": 663
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5573,
+      "step": 664
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 10.4476,
+      "step": 665
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0057,
+      "step": 666
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2702,
+      "step": 667
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5717,
+      "step": 668
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2242,
+      "step": 669
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1,
+      "step": 670
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0517,
+      "step": 671
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6543,
+      "step": 672
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1138,
+      "step": 673
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.461,
+      "step": 674
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.7094,
+      "step": 675
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.521,
+      "step": 676
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7116,
+      "step": 677
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6343,
+      "step": 678
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3762,
+      "step": 679
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3603,
+      "step": 680
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7144,
+      "step": 681
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4545,
+      "step": 682
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8188,
+      "step": 683
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7965,
+      "step": 684
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.4675,
+      "step": 685
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0436,
+      "step": 686
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1219,
+      "step": 687
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.4517,
+      "step": 688
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8476,
+      "step": 689
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 1.9284,
+      "step": 690
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7405,
+      "step": 691
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7142,
+      "step": 692
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3979,
+      "step": 693
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 1.3285,
+      "step": 694
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3418,
+      "step": 695
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.4472,
+      "step": 696
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7355,
+      "step": 697
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7982,
+      "step": 698
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.4516,
+      "step": 699
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.2532,
+      "step": 700
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.9959,
+      "step": 701
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.0418,
+      "step": 702
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 10.7767,
+      "step": 703
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.774,
+      "step": 704
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8912,
+      "step": 705
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2178,
+      "step": 706
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.6197,
+      "step": 707
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4755,
+      "step": 708
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8276,
+      "step": 709
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.2925,
+      "step": 710
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.3887,
+      "step": 711
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1465,
+      "step": 712
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.5806,
+      "step": 713
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3063,
+      "step": 714
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6066,
+      "step": 715
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1536,
+      "step": 716
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5582,
+      "step": 717
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0353,
+      "step": 718
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6415,
+      "step": 719
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8291,
+      "step": 720
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 10.7575,
+      "step": 721
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9141,
+      "step": 722
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5217,
+      "step": 723
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4549,
+      "step": 724
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8112,
+      "step": 725
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2729,
+      "step": 726
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8515,
+      "step": 727
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.9712,
+      "step": 728
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.097,
+      "step": 729
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0208,
+      "step": 730
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1377,
+      "step": 731
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4019,
+      "step": 732
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.9869,
+      "step": 733
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2954,
+      "step": 734
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4144,
+      "step": 735
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8053,
+      "step": 736
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8891,
+      "step": 737
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.812,
+      "step": 738
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2657,
+      "step": 739
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3747,
+      "step": 740
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0364,
+      "step": 741
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8845,
+      "step": 742
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.887,
+      "step": 743
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0706,
+      "step": 744
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6619,
+      "step": 745
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2941,
+      "step": 746
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9192,
+      "step": 747
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9947,
+      "step": 748
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6376,
+      "step": 749
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.0358,
+      "step": 750
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4578,
+      "step": 751
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7784,
+      "step": 752
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.632,
+      "step": 753
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.8649,
+      "step": 754
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7951,
+      "step": 755
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3841,
+      "step": 756
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.4558,
+      "step": 757
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7638,
+      "step": 758
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9413,
+      "step": 759
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0916,
+      "step": 760
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1351,
+      "step": 761
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6078,
+      "step": 762
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7982,
+      "step": 763
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6132,
+      "step": 764
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.551,
+      "step": 765
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3301,
+      "step": 766
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4888,
+      "step": 767
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1476,
+      "step": 768
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4244,
+      "step": 769
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6025,
+      "step": 770
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.102,
+      "step": 771
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.017,
+      "step": 772
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.4101,
+      "step": 773
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1741,
+      "step": 774
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1256,
+      "step": 775
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5164,
+      "step": 776
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.6959,
+      "step": 777
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7666,
+      "step": 778
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4336,
+      "step": 779
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 11.8478,
+      "step": 780
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8382,
+      "step": 781
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1792,
+      "step": 782
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4424,
+      "step": 783
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.345,
+      "step": 784
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6887,
+      "step": 785
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9867,
+      "step": 786
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6152,
+      "step": 787
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7283,
+      "step": 788
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.0157,
+      "step": 789
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6044,
+      "step": 790
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.4132,
+      "step": 791
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.735,
+      "step": 792
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3631,
+      "step": 793
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2308,
+      "step": 794
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2184,
+      "step": 795
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.4661,
+      "step": 796
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9637,
+      "step": 797
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4178,
+      "step": 798
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.5909,
+      "step": 799
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.1482,
+      "step": 800
+    },
+    {
+      "epoch": 0.01,
+      "eval_loss": 7.355834484100342,
+      "eval_runtime": 22.6252,
+      "eval_samples_per_second": 2.21,
+      "eval_steps_per_second": 1.105,
+      "step": 800
+    },
+    {
+      "epoch": 0.01,
+      "mmlu_eval_accuracy": 0.2525477994227994,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.07142857142857142,
+      "mmlu_eval_accuracy_astronomy": 0.3125,
+      "mmlu_eval_accuracy_business_ethics": 0.4444444444444444,
+      "mmlu_loss": 5.191131496429444,
+      "step": 800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 10.0427,
+      "step": 801
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.2669,
+      "step": 802
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.8026,
+      "step": 803
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4949,
+      "step": 804
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4491,
+      "step": 805
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0383,
+      "step": 806
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.1213,
+      "step": 807
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.5158,
+      "step": 808
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5648,
+      "step": 809
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9932,
+      "step": 810
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6441,
+      "step": 811
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8661,
+      "step": 812
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3609,
+      "step": 813
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6828,
+      "step": 814
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9693,
+      "step": 815
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3733,
+      "step": 816
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6286,
+      "step": 817
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4349,
+      "step": 818
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6706,
+      "step": 819
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3089,
+      "step": 820
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2394,
+      "step": 821
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.963,
+      "step": 822
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6564,
+      "step": 823
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.997,
+      "step": 824
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 1.9261,
+      "step": 825
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1421,
+      "step": 826
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2335,
+      "step": 827
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3432,
+      "step": 828
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0154,
+      "step": 829
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.5135,
+      "step": 830
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6226,
+      "step": 831
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1672,
+      "step": 832
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0853,
+      "step": 833
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1213,
+      "step": 834
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7815,
+      "step": 835
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8916,
+      "step": 836
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6464,
+      "step": 837
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3307,
+      "step": 838
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8165,
+      "step": 839
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.886,
+      "step": 840
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4781,
+      "step": 841
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8351,
+      "step": 842
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.358,
+      "step": 843
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6501,
+      "step": 844
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0864,
+      "step": 845
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2922,
+      "step": 846
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.9847,
+      "step": 847
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.2558,
+      "step": 848
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.0195,
+      "step": 849
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.996,
+      "step": 850
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.5705,
+      "step": 851
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.4136,
+      "step": 852
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6302,
+      "step": 853
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8761,
+      "step": 854
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4995,
+      "step": 855
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4762,
+      "step": 856
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5749,
+      "step": 857
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0273,
+      "step": 858
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8258,
+      "step": 859
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.1836,
+      "step": 860
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.5493,
+      "step": 861
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1891,
+      "step": 862
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.7392,
+      "step": 863
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1655,
+      "step": 864
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.5218,
+      "step": 865
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3759,
+      "step": 866
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2497,
+      "step": 867
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5901,
+      "step": 868
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.0624,
+      "step": 869
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 10.2452,
+      "step": 870
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5649,
+      "step": 871
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0826,
+      "step": 872
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.2703,
+      "step": 873
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.9088,
+      "step": 874
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3875,
+      "step": 875
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2511,
+      "step": 876
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4065,
+      "step": 877
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.175,
+      "step": 878
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8358,
+      "step": 879
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3208,
+      "step": 880
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2049,
+      "step": 881
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8251,
+      "step": 882
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4262,
+      "step": 883
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2227,
+      "step": 884
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1062,
+      "step": 885
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9417,
+      "step": 886
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3315,
+      "step": 887
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0012,
+      "step": 888
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6386,
+      "step": 889
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0377,
+      "step": 890
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6707,
+      "step": 891
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.4955,
+      "step": 892
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7343,
+      "step": 893
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8305,
+      "step": 894
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.7016,
+      "step": 895
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7149,
+      "step": 896
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5649,
+      "step": 897
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.815,
+      "step": 898
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6135,
+      "step": 899
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8776,
+      "step": 900
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.7288,
+      "step": 901
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8019,
+      "step": 902
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.0921,
+      "step": 903
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.824,
+      "step": 904
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7151,
+      "step": 905
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.5054,
+      "step": 906
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8095,
+      "step": 907
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.3218,
+      "step": 908
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9993,
+      "step": 909
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.4433,
+      "step": 910
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5863,
+      "step": 911
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.505,
+      "step": 912
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.9734,
+      "step": 913
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1792,
+      "step": 914
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4574,
+      "step": 915
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2787,
+      "step": 916
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8201,
+      "step": 917
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2537,
+      "step": 918
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1387,
+      "step": 919
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7161,
+      "step": 920
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2207,
+      "step": 921
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7953,
+      "step": 922
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9949,
+      "step": 923
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9173,
+      "step": 924
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7903,
+      "step": 925
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4784,
+      "step": 926
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2264,
+      "step": 927
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.566,
+      "step": 928
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0686,
+      "step": 929
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.791,
+      "step": 930
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8393,
+      "step": 931
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4387,
+      "step": 932
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2374,
+      "step": 933
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9598,
+      "step": 934
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1597,
+      "step": 935
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.0403,
+      "step": 936
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3301,
+      "step": 937
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.072,
+      "step": 938
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4788,
+      "step": 939
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0656,
+      "step": 940
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9647,
+      "step": 941
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1168,
+      "step": 942
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0293,
+      "step": 943
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3622,
+      "step": 944
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8957,
+      "step": 945
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.4,
+      "step": 946
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.6626,
+      "step": 947
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8212,
+      "step": 948
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8638,
+      "step": 949
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6406,
+      "step": 950
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7069,
+      "step": 951
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1384,
+      "step": 952
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.612,
+      "step": 953
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7201,
+      "step": 954
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3532,
+      "step": 955
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1266,
+      "step": 956
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6192,
+      "step": 957
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.826,
+      "step": 958
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9338,
+      "step": 959
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.4487,
+      "step": 960
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.872,
+      "step": 961
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8601,
+      "step": 962
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7401,
+      "step": 963
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5412,
+      "step": 964
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2501,
+      "step": 965
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6837,
+      "step": 966
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6494,
+      "step": 967
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.604,
+      "step": 968
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.837,
+      "step": 969
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3957,
+      "step": 970
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3281,
+      "step": 971
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8264,
+      "step": 972
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6554,
+      "step": 973
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.5768,
+      "step": 974
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4187,
+      "step": 975
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8479,
+      "step": 976
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9849,
+      "step": 977
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6471,
+      "step": 978
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8041,
+      "step": 979
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8876,
+      "step": 980
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6423,
+      "step": 981
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5329,
+      "step": 982
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2801,
+      "step": 983
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1699,
+      "step": 984
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6469,
+      "step": 985
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6766,
+      "step": 986
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7538,
+      "step": 987
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9606,
+      "step": 988
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0713,
+      "step": 989
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4965,
+      "step": 990
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3408,
+      "step": 991
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4007,
+      "step": 992
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8921,
+      "step": 993
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8681,
+      "step": 994
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 1.8867,
+      "step": 995
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.467,
+      "step": 996
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7895,
+      "step": 997
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0523,
+      "step": 998
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.4032,
+      "step": 999
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7719,
+      "step": 1000
+    },
+    {
+      "epoch": 0.01,
+      "eval_loss": 6.766034126281738,
+      "eval_runtime": 22.4042,
+      "eval_samples_per_second": 2.232,
+      "eval_steps_per_second": 1.116,
+      "step": 1000
+    },
+    {
+      "epoch": 0.01,
+      "mmlu_eval_accuracy": 0.2525477994227994,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.07142857142857142,
+      "mmlu_eval_accuracy_astronomy": 0.3125,
+      "mmlu_eval_accuracy_business_ethics": 0.4444444444444444,
+      "mmlu_loss": 4.338861379623413,
+      "step": 1000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0285,
+      "step": 1001
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4571,
+      "step": 1002
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7721,
+      "step": 1003
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5108,
+      "step": 1004
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3813,
+      "step": 1005
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.7963,
+      "step": 1006
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1101,
+      "step": 1007
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.021,
+      "step": 1008
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5916,
+      "step": 1009
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8813,
+      "step": 1010
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1193,
+      "step": 1011
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5092,
+      "step": 1012
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8569,
+      "step": 1013
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.119,
+      "step": 1014
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3247,
+      "step": 1015
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.2358,
+      "step": 1016
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2795,
+      "step": 1017
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3466,
+      "step": 1018
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5443,
+      "step": 1019
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7296,
+      "step": 1020
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0412,
+      "step": 1021
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.4829,
+      "step": 1022
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7901,
+      "step": 1023
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8077,
+      "step": 1024
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4887,
+      "step": 1025
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3095,
+      "step": 1026
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3235,
+      "step": 1027
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6315,
+      "step": 1028
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4294,
+      "step": 1029
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8457,
+      "step": 1030
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7583,
+      "step": 1031
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3129,
+      "step": 1032
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1832,
+      "step": 1033
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.1764,
+      "step": 1034
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0101,
+      "step": 1035
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6524,
+      "step": 1036
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2825,
+      "step": 1037
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2262,
+      "step": 1038
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2533,
+      "step": 1039
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8794,
+      "step": 1040
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7901,
+      "step": 1041
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8351,
+      "step": 1042
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.5888,
+      "step": 1043
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8932,
+      "step": 1044
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2999,
+      "step": 1045
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8396,
+      "step": 1046
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.4209,
+      "step": 1047
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.1524,
+      "step": 1048
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7784,
+      "step": 1049
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0179,
+      "step": 1050
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1153,
+      "step": 1051
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2149,
+      "step": 1052
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0117,
+      "step": 1053
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.9693,
+      "step": 1054
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5656,
+      "step": 1055
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5,
+      "step": 1056
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.102,
+      "step": 1057
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3079,
+      "step": 1058
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.5754,
+      "step": 1059
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6989,
+      "step": 1060
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.9597,
+      "step": 1061
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3743,
+      "step": 1062
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8887,
+      "step": 1063
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.3779,
+      "step": 1064
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5001,
+      "step": 1065
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4095,
+      "step": 1066
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5073,
+      "step": 1067
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1331,
+      "step": 1068
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.323,
+      "step": 1069
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6116,
+      "step": 1070
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.1212,
+      "step": 1071
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.0951,
+      "step": 1072
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.2463,
+      "step": 1073
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.4488,
+      "step": 1074
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.279,
+      "step": 1075
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.5728,
+      "step": 1076
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1362,
+      "step": 1077
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6648,
+      "step": 1078
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.427,
+      "step": 1079
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8145,
+      "step": 1080
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.5308,
+      "step": 1081
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.974,
+      "step": 1082
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1965,
+      "step": 1083
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8749,
+      "step": 1084
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7352,
+      "step": 1085
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.7934,
+      "step": 1086
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6003,
+      "step": 1087
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.5775,
+      "step": 1088
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.519,
+      "step": 1089
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7403,
+      "step": 1090
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8145,
+      "step": 1091
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5776,
+      "step": 1092
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3753,
+      "step": 1093
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.9586,
+      "step": 1094
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7263,
+      "step": 1095
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.7034,
+      "step": 1096
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0579,
+      "step": 1097
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8419,
+      "step": 1098
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0751,
+      "step": 1099
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6438,
+      "step": 1100
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8744,
+      "step": 1101
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4992,
+      "step": 1102
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.8094,
+      "step": 1103
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.162,
+      "step": 1104
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.8351,
+      "step": 1105
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8845,
+      "step": 1106
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1894,
+      "step": 1107
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 10.8333,
+      "step": 1108
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4226,
+      "step": 1109
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0712,
+      "step": 1110
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9981,
+      "step": 1111
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.5885,
+      "step": 1112
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.1915,
+      "step": 1113
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8003,
+      "step": 1114
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5566,
+      "step": 1115
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4085,
+      "step": 1116
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0793,
+      "step": 1117
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0909,
+      "step": 1118
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2273,
+      "step": 1119
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8273,
+      "step": 1120
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.0231,
+      "step": 1121
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7229,
+      "step": 1122
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4479,
+      "step": 1123
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2178,
+      "step": 1124
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9038,
+      "step": 1125
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2653,
+      "step": 1126
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2974,
+      "step": 1127
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3003,
+      "step": 1128
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7853,
+      "step": 1129
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9143,
+      "step": 1130
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2573,
+      "step": 1131
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7091,
+      "step": 1132
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3372,
+      "step": 1133
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4165,
+      "step": 1134
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.4422,
+      "step": 1135
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7693,
+      "step": 1136
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7802,
+      "step": 1137
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7263,
+      "step": 1138
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.6749,
+      "step": 1139
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9459,
+      "step": 1140
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9697,
+      "step": 1141
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4506,
+      "step": 1142
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5099,
+      "step": 1143
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1475,
+      "step": 1144
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3769,
+      "step": 1145
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2035,
+      "step": 1146
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6017,
+      "step": 1147
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.463,
+      "step": 1148
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3844,
+      "step": 1149
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.5306,
+      "step": 1150
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.5502,
+      "step": 1151
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7394,
+      "step": 1152
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5626,
+      "step": 1153
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1618,
+      "step": 1154
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.5174,
+      "step": 1155
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1038,
+      "step": 1156
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3789,
+      "step": 1157
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2985,
+      "step": 1158
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.4763,
+      "step": 1159
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.5071,
+      "step": 1160
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0827,
+      "step": 1161
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7349,
+      "step": 1162
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.798,
+      "step": 1163
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3176,
+      "step": 1164
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8114,
+      "step": 1165
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3379,
+      "step": 1166
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1157,
+      "step": 1167
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.4675,
+      "step": 1168
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2721,
+      "step": 1169
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.0603,
+      "step": 1170
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6358,
+      "step": 1171
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0865,
+      "step": 1172
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.709,
+      "step": 1173
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7705,
+      "step": 1174
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7677,
+      "step": 1175
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2418,
+      "step": 1176
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7114,
+      "step": 1177
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1165,
+      "step": 1178
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9654,
+      "step": 1179
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0672,
+      "step": 1180
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1738,
+      "step": 1181
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7604,
+      "step": 1182
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8426,
+      "step": 1183
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0231,
+      "step": 1184
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2938,
+      "step": 1185
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.783,
+      "step": 1186
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3328,
+      "step": 1187
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.321,
+      "step": 1188
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6368,
+      "step": 1189
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.101,
+      "step": 1190
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6777,
+      "step": 1191
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0831,
+      "step": 1192
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5853,
+      "step": 1193
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7923,
+      "step": 1194
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3734,
+      "step": 1195
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4268,
+      "step": 1196
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6796,
+      "step": 1197
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.9028,
+      "step": 1198
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3716,
+      "step": 1199
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6761,
+      "step": 1200
+    },
+    {
+      "epoch": 0.01,
+      "eval_loss": 6.9188361167907715,
+      "eval_runtime": 22.426,
+      "eval_samples_per_second": 2.23,
+      "eval_steps_per_second": 1.115,
+      "step": 1200
+    },
+    {
+      "epoch": 0.01,
+      "mmlu_eval_accuracy": 0.3260281385281385,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.35714285714285715,
+      "mmlu_eval_accuracy_astronomy": 0.25,
+      "mmlu_eval_accuracy_business_ethics": 0.3333333333333333,
+      "mmlu_loss": 3.3686839294433595,
+      "step": 1200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8855,
+      "step": 1201
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.8206,
+      "step": 1202
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4401,
+      "step": 1203
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2366,
+      "step": 1204
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9885,
+      "step": 1205
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.5444,
+      "step": 1206
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4259,
+      "step": 1207
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.5369,
+      "step": 1208
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0839,
+      "step": 1209
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7622,
+      "step": 1210
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.8979,
+      "step": 1211
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5508,
+      "step": 1212
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6439,
+      "step": 1213
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6249,
+      "step": 1214
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.495,
+      "step": 1215
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.0642,
+      "step": 1216
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8997,
+      "step": 1217
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6275,
+      "step": 1218
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.3317,
+      "step": 1219
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4635,
+      "step": 1220
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5197,
+      "step": 1221
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.5928,
+      "step": 1222
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.2363,
+      "step": 1223
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0266,
+      "step": 1224
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3356,
+      "step": 1225
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7927,
+      "step": 1226
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6952,
+      "step": 1227
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8878,
+      "step": 1228
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7472,
+      "step": 1229
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6454,
+      "step": 1230
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4972,
+      "step": 1231
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.3347,
+      "step": 1232
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1631,
+      "step": 1233
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.4708,
+      "step": 1234
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.5697,
+      "step": 1235
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8218,
+      "step": 1236
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.269,
+      "step": 1237
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4165,
+      "step": 1238
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3653,
+      "step": 1239
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0152,
+      "step": 1240
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.9157,
+      "step": 1241
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4086,
+      "step": 1242
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2493,
+      "step": 1243
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8279,
+      "step": 1244
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.6649,
+      "step": 1245
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4405,
+      "step": 1246
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.1992,
+      "step": 1247
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2055,
+      "step": 1248
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4395,
+      "step": 1249
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2475,
+      "step": 1250
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8205,
+      "step": 1251
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1053,
+      "step": 1252
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7494,
+      "step": 1253
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7387,
+      "step": 1254
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8983,
+      "step": 1255
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5614,
+      "step": 1256
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7617,
+      "step": 1257
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.2445,
+      "step": 1258
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3043,
+      "step": 1259
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4214,
+      "step": 1260
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1384,
+      "step": 1261
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3914,
+      "step": 1262
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3287,
+      "step": 1263
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2174,
+      "step": 1264
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4397,
+      "step": 1265
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6875,
+      "step": 1266
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4512,
+      "step": 1267
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2834,
+      "step": 1268
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7651,
+      "step": 1269
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9263,
+      "step": 1270
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6721,
+      "step": 1271
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.9178,
+      "step": 1272
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7967,
+      "step": 1273
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5242,
+      "step": 1274
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7794,
+      "step": 1275
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4256,
+      "step": 1276
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5788,
+      "step": 1277
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7586,
+      "step": 1278
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.964,
+      "step": 1279
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0749,
+      "step": 1280
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6248,
+      "step": 1281
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2465,
+      "step": 1282
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.1591,
+      "step": 1283
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4328,
+      "step": 1284
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.534,
+      "step": 1285
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.523,
+      "step": 1286
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.5672,
+      "step": 1287
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9162,
+      "step": 1288
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.1089,
+      "step": 1289
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.3287,
+      "step": 1290
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2499,
+      "step": 1291
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9645,
+      "step": 1292
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3903,
+      "step": 1293
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.5322,
+      "step": 1294
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2211,
+      "step": 1295
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2788,
+      "step": 1296
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1862,
+      "step": 1297
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2678,
+      "step": 1298
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.5399,
+      "step": 1299
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7935,
+      "step": 1300
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.0391,
+      "step": 1301
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1049,
+      "step": 1302
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.365,
+      "step": 1303
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.8809,
+      "step": 1304
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2335,
+      "step": 1305
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 10.5135,
+      "step": 1306
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2378,
+      "step": 1307
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9265,
+      "step": 1308
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.641,
+      "step": 1309
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9822,
+      "step": 1310
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3369,
+      "step": 1311
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3735,
+      "step": 1312
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.2618,
+      "step": 1313
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6854,
+      "step": 1314
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3748,
+      "step": 1315
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9206,
+      "step": 1316
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1969,
+      "step": 1317
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1245,
+      "step": 1318
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9977,
+      "step": 1319
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5319,
+      "step": 1320
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4431,
+      "step": 1321
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7264,
+      "step": 1322
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.05,
+      "step": 1323
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3118,
+      "step": 1324
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4575,
+      "step": 1325
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.593,
+      "step": 1326
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0061,
+      "step": 1327
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2481,
+      "step": 1328
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8017,
+      "step": 1329
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8617,
+      "step": 1330
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7036,
+      "step": 1331
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0091,
+      "step": 1332
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9687,
+      "step": 1333
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3925,
+      "step": 1334
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1127,
+      "step": 1335
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8163,
+      "step": 1336
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.0639,
+      "step": 1337
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8929,
+      "step": 1338
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5011,
+      "step": 1339
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.033,
+      "step": 1340
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0526,
+      "step": 1341
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4428,
+      "step": 1342
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3477,
+      "step": 1343
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.881,
+      "step": 1344
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.5276,
+      "step": 1345
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4183,
+      "step": 1346
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4943,
+      "step": 1347
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9187,
+      "step": 1348
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1003,
+      "step": 1349
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1187,
+      "step": 1350
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8081,
+      "step": 1351
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4695,
+      "step": 1352
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.5761,
+      "step": 1353
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9635,
+      "step": 1354
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2133,
+      "step": 1355
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2611,
+      "step": 1356
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.6885,
+      "step": 1357
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1157,
+      "step": 1358
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4421,
+      "step": 1359
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2128,
+      "step": 1360
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6978,
+      "step": 1361
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9804,
+      "step": 1362
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.3426,
+      "step": 1363
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2676,
+      "step": 1364
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.325,
+      "step": 1365
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1263,
+      "step": 1366
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7481,
+      "step": 1367
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6891,
+      "step": 1368
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8568,
+      "step": 1369
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9893,
+      "step": 1370
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0302,
+      "step": 1371
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3204,
+      "step": 1372
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9008,
+      "step": 1373
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2624,
+      "step": 1374
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6234,
+      "step": 1375
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2286,
+      "step": 1376
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3426,
+      "step": 1377
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1962,
+      "step": 1378
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3142,
+      "step": 1379
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.414,
+      "step": 1380
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0191,
+      "step": 1381
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.4953,
+      "step": 1382
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6694,
+      "step": 1383
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8611,
+      "step": 1384
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.86,
+      "step": 1385
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6519,
+      "step": 1386
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.394,
+      "step": 1387
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2117,
+      "step": 1388
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.9924,
+      "step": 1389
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.58,
+      "step": 1390
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.4415,
+      "step": 1391
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7196,
+      "step": 1392
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7388,
+      "step": 1393
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4784,
+      "step": 1394
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.496,
+      "step": 1395
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8706,
+      "step": 1396
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1858,
+      "step": 1397
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.9038,
+      "step": 1398
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4852,
+      "step": 1399
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2717,
+      "step": 1400
+    },
+    {
+      "epoch": 0.01,
+      "eval_loss": 6.97923469543457,
+      "eval_runtime": 22.472,
+      "eval_samples_per_second": 2.225,
+      "eval_steps_per_second": 1.112,
+      "step": 1400
+    },
+    {
+      "epoch": 0.01,
+      "mmlu_eval_accuracy": 0.2525477994227994,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.07142857142857142,
+      "mmlu_eval_accuracy_astronomy": 0.3125,
+      "mmlu_eval_accuracy_business_ethics": 0.4444444444444444,
+      "mmlu_loss": 3.657382688522339,
+      "step": 1400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.843,
+      "step": 1401
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.5611,
+      "step": 1402
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2434,
+      "step": 1403
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3136,
+      "step": 1404
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.686,
+      "step": 1405
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6365,
+      "step": 1406
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.1811,
+      "step": 1407
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7537,
+      "step": 1408
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.2949,
+      "step": 1409
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4827,
+      "step": 1410
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0965,
+      "step": 1411
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.007,
+      "step": 1412
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2861,
+      "step": 1413
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1774,
+      "step": 1414
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7777,
+      "step": 1415
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.0259,
+      "step": 1416
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.9024,
+      "step": 1417
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4786,
+      "step": 1418
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5873,
+      "step": 1419
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2744,
+      "step": 1420
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9484,
+      "step": 1421
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2093,
+      "step": 1422
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3394,
+      "step": 1423
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1247,
+      "step": 1424
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0691,
+      "step": 1425
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.559,
+      "step": 1426
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1518,
+      "step": 1427
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4143,
+      "step": 1428
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0287,
+      "step": 1429
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8112,
+      "step": 1430
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2625,
+      "step": 1431
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3528,
+      "step": 1432
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2715,
+      "step": 1433
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7849,
+      "step": 1434
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2002,
+      "step": 1435
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0658,
+      "step": 1436
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0671,
+      "step": 1437
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2577,
+      "step": 1438
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.803,
+      "step": 1439
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2974,
+      "step": 1440
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0897,
+      "step": 1441
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0805,
+      "step": 1442
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7681,
+      "step": 1443
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.6565,
+      "step": 1444
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0174,
+      "step": 1445
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8507,
+      "step": 1446
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2105,
+      "step": 1447
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.572,
+      "step": 1448
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2904,
+      "step": 1449
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.4623,
+      "step": 1450
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4774,
+      "step": 1451
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1277,
+      "step": 1452
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6204,
+      "step": 1453
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.3219,
+      "step": 1454
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2517,
+      "step": 1455
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.3026,
+      "step": 1456
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4016,
+      "step": 1457
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5256,
+      "step": 1458
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.9316,
+      "step": 1459
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.631,
+      "step": 1460
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2888,
+      "step": 1461
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5511,
+      "step": 1462
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.9799,
+      "step": 1463
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6982,
+      "step": 1464
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4923,
+      "step": 1465
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8329,
+      "step": 1466
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2733,
+      "step": 1467
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8221,
+      "step": 1468
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.363,
+      "step": 1469
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6348,
+      "step": 1470
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3319,
+      "step": 1471
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6768,
+      "step": 1472
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1985,
+      "step": 1473
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6109,
+      "step": 1474
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.974,
+      "step": 1475
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8902,
+      "step": 1476
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6762,
+      "step": 1477
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8541,
+      "step": 1478
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3867,
+      "step": 1479
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.9624,
+      "step": 1480
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8768,
+      "step": 1481
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7598,
+      "step": 1482
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6522,
+      "step": 1483
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8156,
+      "step": 1484
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3791,
+      "step": 1485
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2178,
+      "step": 1486
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8448,
+      "step": 1487
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5377,
+      "step": 1488
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7407,
+      "step": 1489
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7636,
+      "step": 1490
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4325,
+      "step": 1491
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8966,
+      "step": 1492
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0626,
+      "step": 1493
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.255,
+      "step": 1494
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2802,
+      "step": 1495
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.894,
+      "step": 1496
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6482,
+      "step": 1497
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8903,
+      "step": 1498
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8672,
+      "step": 1499
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6079,
+      "step": 1500
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6217,
+      "step": 1501
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2361,
+      "step": 1502
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3962,
+      "step": 1503
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.0036,
+      "step": 1504
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5926,
+      "step": 1505
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.114,
+      "step": 1506
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.4419,
+      "step": 1507
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7838,
+      "step": 1508
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6635,
+      "step": 1509
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.2906,
+      "step": 1510
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4186,
+      "step": 1511
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.4783,
+      "step": 1512
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1226,
+      "step": 1513
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2458,
+      "step": 1514
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5302,
+      "step": 1515
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1515,
+      "step": 1516
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4182,
+      "step": 1517
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8248,
+      "step": 1518
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2349,
+      "step": 1519
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9314,
+      "step": 1520
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1161,
+      "step": 1521
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4183,
+      "step": 1522
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4454,
+      "step": 1523
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.5588,
+      "step": 1524
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8026,
+      "step": 1525
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7695,
+      "step": 1526
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3636,
+      "step": 1527
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2776,
+      "step": 1528
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5386,
+      "step": 1529
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.521,
+      "step": 1530
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8388,
+      "step": 1531
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3561,
+      "step": 1532
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.9606,
+      "step": 1533
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9438,
+      "step": 1534
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7665,
+      "step": 1535
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5826,
+      "step": 1536
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.0798,
+      "step": 1537
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8545,
+      "step": 1538
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.302,
+      "step": 1539
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1092,
+      "step": 1540
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.5021,
+      "step": 1541
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9384,
+      "step": 1542
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8761,
+      "step": 1543
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3316,
+      "step": 1544
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.2051,
+      "step": 1545
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7907,
+      "step": 1546
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.2534,
+      "step": 1547
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2274,
+      "step": 1548
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9226,
+      "step": 1549
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2502,
+      "step": 1550
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2703,
+      "step": 1551
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4359,
+      "step": 1552
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.128,
+      "step": 1553
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3147,
+      "step": 1554
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.026,
+      "step": 1555
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.9393,
+      "step": 1556
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7753,
+      "step": 1557
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.9049,
+      "step": 1558
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0538,
+      "step": 1559
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.8691,
+      "step": 1560
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9377,
+      "step": 1561
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8385,
+      "step": 1562
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.939,
+      "step": 1563
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.727,
+      "step": 1564
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.7866,
+      "step": 1565
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2439,
+      "step": 1566
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9607,
+      "step": 1567
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3505,
+      "step": 1568
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7729,
+      "step": 1569
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4669,
+      "step": 1570
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8178,
+      "step": 1571
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2173,
+      "step": 1572
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2136,
+      "step": 1573
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.2888,
+      "step": 1574
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0386,
+      "step": 1575
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.9041,
+      "step": 1576
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7544,
+      "step": 1577
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 10.3229,
+      "step": 1578
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4203,
+      "step": 1579
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.497,
+      "step": 1580
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8253,
+      "step": 1581
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0801,
+      "step": 1582
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1585,
+      "step": 1583
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6965,
+      "step": 1584
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.498,
+      "step": 1585
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8697,
+      "step": 1586
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2663,
+      "step": 1587
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7004,
+      "step": 1588
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6561,
+      "step": 1589
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.785,
+      "step": 1590
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5708,
+      "step": 1591
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.326,
+      "step": 1592
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2974,
+      "step": 1593
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1408,
+      "step": 1594
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6526,
+      "step": 1595
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4116,
+      "step": 1596
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0484,
+      "step": 1597
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3162,
+      "step": 1598
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.3806,
+      "step": 1599
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0251,
+      "step": 1600
+    },
+    {
+      "epoch": 0.01,
+      "eval_loss": 6.617897987365723,
+      "eval_runtime": 22.4646,
+      "eval_samples_per_second": 2.226,
+      "eval_steps_per_second": 1.113,
+      "step": 1600
+    },
+    {
+      "epoch": 0.01,
+      "mmlu_eval_accuracy": 0.3260281385281385,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.35714285714285715,
+      "mmlu_eval_accuracy_astronomy": 0.25,
+      "mmlu_eval_accuracy_business_ethics": 0.3333333333333333,
+      "mmlu_loss": 4.160770101547241,
+      "step": 1600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9137,
+      "step": 1601
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2188,
+      "step": 1602
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7688,
+      "step": 1603
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.9799,
+      "step": 1604
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5429,
+      "step": 1605
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8559,
+      "step": 1606
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.3985,
+      "step": 1607
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.9139,
+      "step": 1608
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3303,
+      "step": 1609
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.5168,
+      "step": 1610
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5194,
+      "step": 1611
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9557,
+      "step": 1612
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7102,
+      "step": 1613
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8961,
+      "step": 1614
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6123,
+      "step": 1615
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7808,
+      "step": 1616
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.4919,
+      "step": 1617
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.0723,
+      "step": 1618
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2931,
+      "step": 1619
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8478,
+      "step": 1620
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7126,
+      "step": 1621
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6622,
+      "step": 1622
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3861,
+      "step": 1623
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9856,
+      "step": 1624
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.5076,
+      "step": 1625
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4168,
+      "step": 1626
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2825,
+      "step": 1627
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7497,
+      "step": 1628
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.5672,
+      "step": 1629
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4095,
+      "step": 1630
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.649,
+      "step": 1631
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.3418,
+      "step": 1632
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1337,
+      "step": 1633
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3829,
+      "step": 1634
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0294,
+      "step": 1635
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.2164,
+      "step": 1636
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.3294,
+      "step": 1637
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7047,
+      "step": 1638
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.5622,
+      "step": 1639
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4873,
+      "step": 1640
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6641,
+      "step": 1641
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3943,
+      "step": 1642
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2266,
+      "step": 1643
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.0471,
+      "step": 1644
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5658,
+      "step": 1645
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6489,
+      "step": 1646
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3851,
+      "step": 1647
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7921,
+      "step": 1648
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.4581,
+      "step": 1649
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1407,
+      "step": 1650
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.2919,
+      "step": 1651
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4061,
+      "step": 1652
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3081,
+      "step": 1653
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0527,
+      "step": 1654
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.8729,
+      "step": 1655
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.029,
+      "step": 1656
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6632,
+      "step": 1657
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7047,
+      "step": 1658
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6288,
+      "step": 1659
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8466,
+      "step": 1660
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7726,
+      "step": 1661
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.245,
+      "step": 1662
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0538,
+      "step": 1663
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3611,
+      "step": 1664
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.011,
+      "step": 1665
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6491,
+      "step": 1666
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3409,
+      "step": 1667
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.262,
+      "step": 1668
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.781,
+      "step": 1669
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.8025,
+      "step": 1670
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7529,
+      "step": 1671
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2322,
+      "step": 1672
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4527,
+      "step": 1673
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9457,
+      "step": 1674
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.859,
+      "step": 1675
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.9207,
+      "step": 1676
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.5378,
+      "step": 1677
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6585,
+      "step": 1678
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9523,
+      "step": 1679
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1348,
+      "step": 1680
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9582,
+      "step": 1681
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.416,
+      "step": 1682
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.8214,
+      "step": 1683
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8833,
+      "step": 1684
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.1021,
+      "step": 1685
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7392,
+      "step": 1686
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2616,
+      "step": 1687
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.325,
+      "step": 1688
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3814,
+      "step": 1689
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2816,
+      "step": 1690
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.033,
+      "step": 1691
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.5742,
+      "step": 1692
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0841,
+      "step": 1693
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.2888,
+      "step": 1694
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.9263,
+      "step": 1695
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7552,
+      "step": 1696
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.4163,
+      "step": 1697
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6207,
+      "step": 1698
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.938,
+      "step": 1699
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2925,
+      "step": 1700
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0401,
+      "step": 1701
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1536,
+      "step": 1702
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.2754,
+      "step": 1703
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6765,
+      "step": 1704
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.63,
+      "step": 1705
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6902,
+      "step": 1706
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6434,
+      "step": 1707
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2283,
+      "step": 1708
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.9377,
+      "step": 1709
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.371,
+      "step": 1710
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.6569,
+      "step": 1711
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.2221,
+      "step": 1712
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.5375,
+      "step": 1713
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2189,
+      "step": 1714
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.769,
+      "step": 1715
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0089,
+      "step": 1716
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6402,
+      "step": 1717
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4812,
+      "step": 1718
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9754,
+      "step": 1719
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8435,
+      "step": 1720
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9424,
+      "step": 1721
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.5465,
+      "step": 1722
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.477,
+      "step": 1723
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2254,
+      "step": 1724
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3663,
+      "step": 1725
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.663,
+      "step": 1726
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6951,
+      "step": 1727
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.856,
+      "step": 1728
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0652,
+      "step": 1729
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6929,
+      "step": 1730
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8069,
+      "step": 1731
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.02,
+      "step": 1732
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0782,
+      "step": 1733
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0236,
+      "step": 1734
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2769,
+      "step": 1735
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.7126,
+      "step": 1736
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2746,
+      "step": 1737
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8726,
+      "step": 1738
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.7962,
+      "step": 1739
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.7602,
+      "step": 1740
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 10.3105,
+      "step": 1741
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0771,
+      "step": 1742
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4738,
+      "step": 1743
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2952,
+      "step": 1744
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.2692,
+      "step": 1745
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7201,
+      "step": 1746
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2978,
+      "step": 1747
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.518,
+      "step": 1748
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.659,
+      "step": 1749
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.9101,
+      "step": 1750
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.8397,
+      "step": 1751
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.0451,
+      "step": 1752
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7696,
+      "step": 1753
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1377,
+      "step": 1754
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.2621,
+      "step": 1755
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2143,
+      "step": 1756
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4378,
+      "step": 1757
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8631,
+      "step": 1758
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.019,
+      "step": 1759
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.7475,
+      "step": 1760
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6228,
+      "step": 1761
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0703,
+      "step": 1762
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3426,
+      "step": 1763
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.0842,
+      "step": 1764
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1032,
+      "step": 1765
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6321,
+      "step": 1766
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7804,
+      "step": 1767
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6566,
+      "step": 1768
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4985,
+      "step": 1769
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1129,
+      "step": 1770
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.8081,
+      "step": 1771
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8441,
+      "step": 1772
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.4054,
+      "step": 1773
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6334,
+      "step": 1774
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.4323,
+      "step": 1775
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.18,
+      "step": 1776
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.7573,
+      "step": 1777
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4642,
+      "step": 1778
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.038,
+      "step": 1779
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3785,
+      "step": 1780
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.5088,
+      "step": 1781
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0139,
+      "step": 1782
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0999,
+      "step": 1783
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3224,
+      "step": 1784
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.634,
+      "step": 1785
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1264,
+      "step": 1786
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.317,
+      "step": 1787
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1279,
+      "step": 1788
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2364,
+      "step": 1789
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0627,
+      "step": 1790
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2471,
+      "step": 1791
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8407,
+      "step": 1792
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7083,
+      "step": 1793
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4522,
+      "step": 1794
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0308,
+      "step": 1795
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6915,
+      "step": 1796
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.082,
+      "step": 1797
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7956,
+      "step": 1798
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7007,
+      "step": 1799
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9197,
+      "step": 1800
+    },
+    {
+      "epoch": 0.01,
+      "eval_loss": 6.619495868682861,
+      "eval_runtime": 22.4352,
+      "eval_samples_per_second": 2.229,
+      "eval_steps_per_second": 1.114,
+      "step": 1800
+    },
+    {
+      "epoch": 0.01,
+      "mmlu_eval_accuracy": 0.3260281385281385,
+      "mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
+      "mmlu_eval_accuracy_anatomy": 0.35714285714285715,
+      "mmlu_eval_accuracy_astronomy": 0.25,
+      "mmlu_eval_accuracy_business_ethics": 0.3333333333333333,
+      "mmlu_loss": 4.238778591156006,
+      "step": 1800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1537,
+      "step": 1801
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.684,
+      "step": 1802
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7862,
+      "step": 1803
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.3518,
+      "step": 1804
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.1795,
+      "step": 1805
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.0054,
+      "step": 1806
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8997,
+      "step": 1807
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9002,
+      "step": 1808
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2805,
+      "step": 1809
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1203,
+      "step": 1810
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0206,
+      "step": 1811
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0151,
+      "step": 1812
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.3864,
+      "step": 1813
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1117,
+      "step": 1814
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8487,
+      "step": 1815
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.59,
+      "step": 1816
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.1615,
+      "step": 1817
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7362,
+      "step": 1818
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2294,
+      "step": 1819
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5622,
+      "step": 1820
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5437,
+      "step": 1821
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.093,
+      "step": 1822
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0343,
+      "step": 1823
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4454,
+      "step": 1824
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.5138,
+      "step": 1825
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5605,
+      "step": 1826
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.322,
+      "step": 1827
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6489,
+      "step": 1828
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.331,
+      "step": 1829
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6462,
+      "step": 1830
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.932,
+      "step": 1831
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.9058,
+      "step": 1832
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.3433,
+      "step": 1833
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.4365,
+      "step": 1834
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.3282,
+      "step": 1835
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.448,
+      "step": 1836
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.5369,
+      "step": 1837
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.177,
+      "step": 1838
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.3552,
+      "step": 1839
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.4568,
+      "step": 1840
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0602,
+      "step": 1841
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7449,
+      "step": 1842
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.2675,
+      "step": 1843
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.0317,
+      "step": 1844
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.4342,
+      "step": 1845
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8688,
+      "step": 1846
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.3571,
+      "step": 1847
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.3776,
+      "step": 1848
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.2248,
+      "step": 1849
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.6073,
+      "step": 1850
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8425,
+      "step": 1851
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.5954,
+      "step": 1852
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4197,
+      "step": 1853
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8624,
+      "step": 1854
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9652,
+      "step": 1855
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7145,
+      "step": 1856
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5309,
+      "step": 1857
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.4356,
+      "step": 1858
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6508,
+      "step": 1859
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0955,
+      "step": 1860
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6886,
+      "step": 1861
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7644,
+      "step": 1862
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.5709,
+      "step": 1863
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6212,
+      "step": 1864
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6325,
+      "step": 1865
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.6805,
+      "step": 1866
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.1464,
+      "step": 1867
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9244,
+      "step": 1868
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.336,
+      "step": 1869
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.8783,
+      "step": 1870
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8236,
+      "step": 1871
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.084,
+      "step": 1872
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9639,
+      "step": 1873
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4173,
+      "step": 1874
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0042,
+      "step": 1875
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.2519,
+      "step": 1876
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4656,
+      "step": 1877
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5136,
+      "step": 1878
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.3918,
+      "step": 1879
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9696,
+      "step": 1880
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.9736,
+      "step": 1881
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.6192,
+      "step": 1882
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.3476,
+      "step": 1883
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3048,
+      "step": 1884
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1116,
+      "step": 1885
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.971,
+      "step": 1886
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.0741,
+      "step": 1887
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1418,
+      "step": 1888
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3487,
+      "step": 1889
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.38,
+      "step": 1890
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6561,
+      "step": 1891
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.5606,
+      "step": 1892
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.8623,
+      "step": 1893
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.2984,
+      "step": 1894
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6179,
+      "step": 1895
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8625,
+      "step": 1896
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.8596,
+      "step": 1897
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.7205,
+      "step": 1898
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.6727,
+      "step": 1899
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.016,
+      "step": 1900
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.9868,
+      "step": 1901
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.632,
+      "step": 1902
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.5133,
+      "step": 1903
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7476,
+      "step": 1904
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4174,
+      "step": 1905
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.6789,
+      "step": 1906
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.4534,
+      "step": 1907
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3335,
+      "step": 1908
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.7921,
+      "step": 1909
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9567,
+      "step": 1910
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.1739,
+      "step": 1911
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7514,
+      "step": 1912
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.3858,
+      "step": 1913
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0462,
+      "step": 1914
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3817,
+      "step": 1915
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 1.9739,
+      "step": 1916
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.1122,
+      "step": 1917
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.3361,
+      "step": 1918
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.3184,
+      "step": 1919
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.7342,
+      "step": 1920
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.375,
+      "step": 1921
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.6841,
+      "step": 1922
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0773,
+      "step": 1923
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.8916,
+      "step": 1924
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.7176,
+      "step": 1925
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8841,
+      "step": 1926
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.8345,
+      "step": 1927
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.561,
+      "step": 1928
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.5392,
+      "step": 1929
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.1627,
+      "step": 1930
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.0657,
+      "step": 1931
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.7385,
+      "step": 1932
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.5533,
+      "step": 1933
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0925,
+      "step": 1934
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8752,
+      "step": 1935
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.4039,
+      "step": 1936
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.6472,
+      "step": 1937
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.1819,
+      "step": 1938
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.5919,
+      "step": 1939
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.6527,
+      "step": 1940
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.5188,
+      "step": 1941
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.9856,
+      "step": 1942
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.7038,
+      "step": 1943
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.911,
+      "step": 1944
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.497,
+      "step": 1945
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.1804,
+      "step": 1946
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 2.3949,
+      "step": 1947
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 6.0433,
+      "step": 1948
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.4706,
+      "step": 1949
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.5896,
+      "step": 1950
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.557,
+      "step": 1951
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.34,
+      "step": 1952
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.7865,
+      "step": 1953
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 8.0797,
+      "step": 1954
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.2896,
+      "step": 1955
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.4096,
+      "step": 1956
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 3.9538,
+      "step": 1957
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.2778,
+      "step": 1958
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 7.4968,
+      "step": 1959
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.8328,
+      "step": 1960
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 9.4597,
+      "step": 1961
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 4.6776,
+      "step": 1962
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0004,
+      "loss": 5.4861,
+      "step": 1963
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.5831,
+      "step": 1964
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.4585,
+      "step": 1965
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7898,
+      "step": 1966
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8714,
+      "step": 1967
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.752,
+      "step": 1968
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9024,
+      "step": 1969
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.058,
+      "step": 1970
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1745,
+      "step": 1971
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2162,
+      "step": 1972
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2668,
+      "step": 1973
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3307,
+      "step": 1974
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3285,
+      "step": 1975
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1353,
+      "step": 1976
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8069,
+      "step": 1977
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6885,
+      "step": 1978
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.5946,
+      "step": 1979
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6828,
+      "step": 1980
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6516,
+      "step": 1981
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.261,
+      "step": 1982
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.524,
+      "step": 1983
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.585,
+      "step": 1984
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.8883,
+      "step": 1985
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.689,
+      "step": 1986
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1083,
+      "step": 1987
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1606,
+      "step": 1988
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9243,
+      "step": 1989
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6597,
+      "step": 1990
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2849,
+      "step": 1991
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3715,
+      "step": 1992
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7262,
+      "step": 1993
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6862,
+      "step": 1994
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5412,
+      "step": 1995
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.7483,
+      "step": 1996
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3391,
+      "step": 1997
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2642,
+      "step": 1998
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1519,
+      "step": 1999
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7098,
+      "step": 2000
+    },
+    {
+      "epoch": 0.02,
+      "eval_loss": 6.762476921081543,
+      "eval_runtime": 22.4899,
+      "eval_samples_per_second": 2.223,
+      "eval_steps_per_second": 1.112,
+      "step": 2000
+    },
+    {
+      "epoch": 0.02,
+      "mmlu_eval_accuracy": 0.2525477994227994,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.07142857142857142,
+      "mmlu_eval_accuracy_astronomy": 0.3125,
+      "mmlu_eval_accuracy_business_ethics": 0.4444444444444444,
+      "mmlu_loss": 3.4606559085845947,
+      "step": 2000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.8099,
+      "step": 2001
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0567,
+      "step": 2002
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.2981,
+      "step": 2003
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2668,
+      "step": 2004
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.139,
+      "step": 2005
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.903,
+      "step": 2006
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.2182,
+      "step": 2007
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2347,
+      "step": 2008
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.8383,
+      "step": 2009
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0211,
+      "step": 2010
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2572,
+      "step": 2011
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2877,
+      "step": 2012
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3577,
+      "step": 2013
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.2022,
+      "step": 2014
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2722,
+      "step": 2015
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0552,
+      "step": 2016
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.9857,
+      "step": 2017
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0519,
+      "step": 2018
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.7118,
+      "step": 2019
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4465,
+      "step": 2020
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3009,
+      "step": 2021
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3614,
+      "step": 2022
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3493,
+      "step": 2023
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.34,
+      "step": 2024
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0416,
+      "step": 2025
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.686,
+      "step": 2026
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6021,
+      "step": 2027
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.4161,
+      "step": 2028
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.0029,
+      "step": 2029
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 1.8579,
+      "step": 2030
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0247,
+      "step": 2031
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4184,
+      "step": 2032
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.4962,
+      "step": 2033
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.5137,
+      "step": 2034
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.6692,
+      "step": 2035
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7161,
+      "step": 2036
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.617,
+      "step": 2037
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.413,
+      "step": 2038
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3056,
+      "step": 2039
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9441,
+      "step": 2040
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9943,
+      "step": 2041
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.5703,
+      "step": 2042
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1881,
+      "step": 2043
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5763,
+      "step": 2044
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6389,
+      "step": 2045
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1717,
+      "step": 2046
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5482,
+      "step": 2047
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.9469,
+      "step": 2048
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7685,
+      "step": 2049
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1381,
+      "step": 2050
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6961,
+      "step": 2051
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6425,
+      "step": 2052
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5354,
+      "step": 2053
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2404,
+      "step": 2054
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1556,
+      "step": 2055
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7133,
+      "step": 2056
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.8166,
+      "step": 2057
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.5071,
+      "step": 2058
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5429,
+      "step": 2059
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0367,
+      "step": 2060
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.5386,
+      "step": 2061
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.5899,
+      "step": 2062
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2968,
+      "step": 2063
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9951,
+      "step": 2064
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.8608,
+      "step": 2065
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.4735,
+      "step": 2066
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5612,
+      "step": 2067
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7461,
+      "step": 2068
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.5887,
+      "step": 2069
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3426,
+      "step": 2070
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5589,
+      "step": 2071
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.498,
+      "step": 2072
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1306,
+      "step": 2073
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.3492,
+      "step": 2074
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2311,
+      "step": 2075
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.8798,
+      "step": 2076
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6799,
+      "step": 2077
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.5011,
+      "step": 2078
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8892,
+      "step": 2079
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6449,
+      "step": 2080
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9117,
+      "step": 2081
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1157,
+      "step": 2082
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.196,
+      "step": 2083
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.9364,
+      "step": 2084
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3618,
+      "step": 2085
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3755,
+      "step": 2086
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4564,
+      "step": 2087
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.4912,
+      "step": 2088
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.113,
+      "step": 2089
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.0588,
+      "step": 2090
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.668,
+      "step": 2091
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.08,
+      "step": 2092
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2042,
+      "step": 2093
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4134,
+      "step": 2094
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0456,
+      "step": 2095
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2245,
+      "step": 2096
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4936,
+      "step": 2097
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5158,
+      "step": 2098
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7269,
+      "step": 2099
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7077,
+      "step": 2100
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6008,
+      "step": 2101
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4652,
+      "step": 2102
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 1.918,
+      "step": 2103
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.5819,
+      "step": 2104
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7764,
+      "step": 2105
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.0525,
+      "step": 2106
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.5359,
+      "step": 2107
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.4925,
+      "step": 2108
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4857,
+      "step": 2109
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 1.9445,
+      "step": 2110
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.8494,
+      "step": 2111
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1513,
+      "step": 2112
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2552,
+      "step": 2113
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7229,
+      "step": 2114
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8571,
+      "step": 2115
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5968,
+      "step": 2116
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.8806,
+      "step": 2117
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.4641,
+      "step": 2118
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.6039,
+      "step": 2119
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1379,
+      "step": 2120
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6688,
+      "step": 2121
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.293,
+      "step": 2122
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5664,
+      "step": 2123
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0825,
+      "step": 2124
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9788,
+      "step": 2125
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.9641,
+      "step": 2126
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7799,
+      "step": 2127
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0619,
+      "step": 2128
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0022,
+      "step": 2129
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8022,
+      "step": 2130
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.5301,
+      "step": 2131
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.681,
+      "step": 2132
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7362,
+      "step": 2133
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5462,
+      "step": 2134
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2356,
+      "step": 2135
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.2259,
+      "step": 2136
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3646,
+      "step": 2137
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8765,
+      "step": 2138
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6487,
+      "step": 2139
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9622,
+      "step": 2140
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1761,
+      "step": 2141
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6922,
+      "step": 2142
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.0371,
+      "step": 2143
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7869,
+      "step": 2144
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3725,
+      "step": 2145
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8894,
+      "step": 2146
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6083,
+      "step": 2147
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.4451,
+      "step": 2148
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1149,
+      "step": 2149
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.8058,
+      "step": 2150
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1308,
+      "step": 2151
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.1447,
+      "step": 2152
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.208,
+      "step": 2153
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5193,
+      "step": 2154
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7729,
+      "step": 2155
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5019,
+      "step": 2156
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6092,
+      "step": 2157
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1853,
+      "step": 2158
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7,
+      "step": 2159
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1638,
+      "step": 2160
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.762,
+      "step": 2161
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7455,
+      "step": 2162
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9372,
+      "step": 2163
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4569,
+      "step": 2164
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6705,
+      "step": 2165
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1988,
+      "step": 2166
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2526,
+      "step": 2167
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.9066,
+      "step": 2168
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1365,
+      "step": 2169
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3422,
+      "step": 2170
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2691,
+      "step": 2171
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9008,
+      "step": 2172
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.2555,
+      "step": 2173
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0886,
+      "step": 2174
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0369,
+      "step": 2175
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5566,
+      "step": 2176
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2567,
+      "step": 2177
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0179,
+      "step": 2178
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5383,
+      "step": 2179
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4797,
+      "step": 2180
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0163,
+      "step": 2181
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2658,
+      "step": 2182
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1337,
+      "step": 2183
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3287,
+      "step": 2184
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7874,
+      "step": 2185
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7153,
+      "step": 2186
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7037,
+      "step": 2187
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4412,
+      "step": 2188
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3912,
+      "step": 2189
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.034,
+      "step": 2190
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4697,
+      "step": 2191
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6243,
+      "step": 2192
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1133,
+      "step": 2193
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 1.9005,
+      "step": 2194
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7386,
+      "step": 2195
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4169,
+      "step": 2196
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8139,
+      "step": 2197
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.3012,
+      "step": 2198
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8223,
+      "step": 2199
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3757,
+      "step": 2200
+    },
+    {
+      "epoch": 0.02,
+      "eval_loss": 6.580160140991211,
+      "eval_runtime": 22.4971,
+      "eval_samples_per_second": 2.223,
+      "eval_steps_per_second": 1.111,
+      "step": 2200
+    },
+    {
+      "epoch": 0.02,
+      "mmlu_eval_accuracy": 0.2525477994227994,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.07142857142857142,
+      "mmlu_eval_accuracy_astronomy": 0.3125,
+      "mmlu_eval_accuracy_business_ethics": 0.4444444444444444,
+      "mmlu_loss": 3.755114164352417,
+      "step": 2200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5282,
+      "step": 2201
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2478,
+      "step": 2202
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.916,
+      "step": 2203
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5069,
+      "step": 2204
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5952,
+      "step": 2205
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.5059,
+      "step": 2206
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.7434,
+      "step": 2207
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.625,
+      "step": 2208
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1674,
+      "step": 2209
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3937,
+      "step": 2210
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.8783,
+      "step": 2211
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5263,
+      "step": 2212
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7887,
+      "step": 2213
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8911,
+      "step": 2214
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7211,
+      "step": 2215
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.089,
+      "step": 2216
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6373,
+      "step": 2217
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7728,
+      "step": 2218
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6957,
+      "step": 2219
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.43,
+      "step": 2220
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9673,
+      "step": 2221
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.8942,
+      "step": 2222
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2893,
+      "step": 2223
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1505,
+      "step": 2224
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3702,
+      "step": 2225
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1731,
+      "step": 2226
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.997,
+      "step": 2227
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9531,
+      "step": 2228
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0748,
+      "step": 2229
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0642,
+      "step": 2230
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9469,
+      "step": 2231
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2265,
+      "step": 2232
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6461,
+      "step": 2233
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.064,
+      "step": 2234
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1414,
+      "step": 2235
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.5375,
+      "step": 2236
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.6348,
+      "step": 2237
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9975,
+      "step": 2238
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.5242,
+      "step": 2239
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3179,
+      "step": 2240
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6054,
+      "step": 2241
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1832,
+      "step": 2242
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.0572,
+      "step": 2243
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2049,
+      "step": 2244
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6348,
+      "step": 2245
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.67,
+      "step": 2246
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 1.5627,
+      "step": 2247
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1851,
+      "step": 2248
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6792,
+      "step": 2249
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6344,
+      "step": 2250
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.7603,
+      "step": 2251
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7737,
+      "step": 2252
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.5323,
+      "step": 2253
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.4059,
+      "step": 2254
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9343,
+      "step": 2255
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0156,
+      "step": 2256
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1851,
+      "step": 2257
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.44,
+      "step": 2258
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9079,
+      "step": 2259
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.4982,
+      "step": 2260
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 12.3777,
+      "step": 2261
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 10.1265,
+      "step": 2262
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1428,
+      "step": 2263
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8072,
+      "step": 2264
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.911,
+      "step": 2265
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9453,
+      "step": 2266
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0168,
+      "step": 2267
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2098,
+      "step": 2268
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4417,
+      "step": 2269
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.8449,
+      "step": 2270
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.394,
+      "step": 2271
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7642,
+      "step": 2272
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5555,
+      "step": 2273
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3576,
+      "step": 2274
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.386,
+      "step": 2275
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6677,
+      "step": 2276
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2385,
+      "step": 2277
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8074,
+      "step": 2278
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.2963,
+      "step": 2279
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3612,
+      "step": 2280
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1837,
+      "step": 2281
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5882,
+      "step": 2282
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0968,
+      "step": 2283
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.2376,
+      "step": 2284
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3835,
+      "step": 2285
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0143,
+      "step": 2286
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.36,
+      "step": 2287
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0121,
+      "step": 2288
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0144,
+      "step": 2289
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6807,
+      "step": 2290
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8854,
+      "step": 2291
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1727,
+      "step": 2292
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.533,
+      "step": 2293
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.9793,
+      "step": 2294
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.538,
+      "step": 2295
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.145,
+      "step": 2296
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.163,
+      "step": 2297
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1045,
+      "step": 2298
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0209,
+      "step": 2299
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9728,
+      "step": 2300
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8902,
+      "step": 2301
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3075,
+      "step": 2302
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 10.2194,
+      "step": 2303
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.7375,
+      "step": 2304
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3863,
+      "step": 2305
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1317,
+      "step": 2306
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1878,
+      "step": 2307
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.6124,
+      "step": 2308
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8843,
+      "step": 2309
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3988,
+      "step": 2310
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3523,
+      "step": 2311
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5766,
+      "step": 2312
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9096,
+      "step": 2313
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9315,
+      "step": 2314
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4044,
+      "step": 2315
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.6206,
+      "step": 2316
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2429,
+      "step": 2317
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0383,
+      "step": 2318
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4282,
+      "step": 2319
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8973,
+      "step": 2320
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1771,
+      "step": 2321
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.624,
+      "step": 2322
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.5197,
+      "step": 2323
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7313,
+      "step": 2324
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8234,
+      "step": 2325
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1702,
+      "step": 2326
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.536,
+      "step": 2327
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1904,
+      "step": 2328
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2077,
+      "step": 2329
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.891,
+      "step": 2330
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6784,
+      "step": 2331
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6611,
+      "step": 2332
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3402,
+      "step": 2333
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 11.1523,
+      "step": 2334
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.5547,
+      "step": 2335
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3485,
+      "step": 2336
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8289,
+      "step": 2337
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2558,
+      "step": 2338
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1794,
+      "step": 2339
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.8782,
+      "step": 2340
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.415,
+      "step": 2341
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5257,
+      "step": 2342
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4751,
+      "step": 2343
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.2259,
+      "step": 2344
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8681,
+      "step": 2345
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6307,
+      "step": 2346
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.1487,
+      "step": 2347
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.3949,
+      "step": 2348
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6988,
+      "step": 2349
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1299,
+      "step": 2350
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9938,
+      "step": 2351
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4176,
+      "step": 2352
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.0184,
+      "step": 2353
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2779,
+      "step": 2354
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0162,
+      "step": 2355
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2335,
+      "step": 2356
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.5505,
+      "step": 2357
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.6445,
+      "step": 2358
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.6513,
+      "step": 2359
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8503,
+      "step": 2360
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1817,
+      "step": 2361
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4376,
+      "step": 2362
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1351,
+      "step": 2363
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7566,
+      "step": 2364
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.626,
+      "step": 2365
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5818,
+      "step": 2366
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3033,
+      "step": 2367
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9289,
+      "step": 2368
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0301,
+      "step": 2369
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4713,
+      "step": 2370
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0931,
+      "step": 2371
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.5812,
+      "step": 2372
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.2272,
+      "step": 2373
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5174,
+      "step": 2374
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1849,
+      "step": 2375
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7496,
+      "step": 2376
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.776,
+      "step": 2377
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3555,
+      "step": 2378
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.688,
+      "step": 2379
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0143,
+      "step": 2380
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.7788,
+      "step": 2381
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7772,
+      "step": 2382
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.6875,
+      "step": 2383
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9944,
+      "step": 2384
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8363,
+      "step": 2385
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7276,
+      "step": 2386
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.4892,
+      "step": 2387
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1083,
+      "step": 2388
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.834,
+      "step": 2389
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8406,
+      "step": 2390
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1168,
+      "step": 2391
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2535,
+      "step": 2392
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9025,
+      "step": 2393
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4481,
+      "step": 2394
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7631,
+      "step": 2395
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2051,
+      "step": 2396
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7816,
+      "step": 2397
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2566,
+      "step": 2398
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1125,
+      "step": 2399
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5952,
+      "step": 2400
+    },
+    {
+      "epoch": 0.02,
+      "eval_loss": 6.616010665893555,
+      "eval_runtime": 22.4801,
+      "eval_samples_per_second": 2.224,
+      "eval_steps_per_second": 1.112,
+      "step": 2400
+    },
+    {
+      "epoch": 0.02,
+      "mmlu_eval_accuracy": 0.2525477994227994,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.07142857142857142,
+      "mmlu_eval_accuracy_astronomy": 0.3125,
+      "mmlu_eval_accuracy_business_ethics": 0.4444444444444444,
+      "mmlu_loss": 3.427501640319824,
+      "step": 2400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6081,
+      "step": 2401
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2359,
+      "step": 2402
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2628,
+      "step": 2403
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.8465,
+      "step": 2404
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.6746,
+      "step": 2405
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1066,
+      "step": 2406
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4981,
+      "step": 2407
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9194,
+      "step": 2408
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.239,
+      "step": 2409
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1749,
+      "step": 2410
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4597,
+      "step": 2411
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5164,
+      "step": 2412
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4122,
+      "step": 2413
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7398,
+      "step": 2414
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5009,
+      "step": 2415
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2757,
+      "step": 2416
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4202,
+      "step": 2417
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.108,
+      "step": 2418
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3303,
+      "step": 2419
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.4671,
+      "step": 2420
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5768,
+      "step": 2421
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9172,
+      "step": 2422
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7606,
+      "step": 2423
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0745,
+      "step": 2424
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2907,
+      "step": 2425
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6956,
+      "step": 2426
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4414,
+      "step": 2427
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.9823,
+      "step": 2428
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6486,
+      "step": 2429
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.5173,
+      "step": 2430
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4647,
+      "step": 2431
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9405,
+      "step": 2432
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.4226,
+      "step": 2433
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4334,
+      "step": 2434
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.9136,
+      "step": 2435
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.6707,
+      "step": 2436
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.6107,
+      "step": 2437
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5296,
+      "step": 2438
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.0736,
+      "step": 2439
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4398,
+      "step": 2440
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.5669,
+      "step": 2441
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.303,
+      "step": 2442
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2993,
+      "step": 2443
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9063,
+      "step": 2444
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3015,
+      "step": 2445
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3117,
+      "step": 2446
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6228,
+      "step": 2447
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6216,
+      "step": 2448
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6188,
+      "step": 2449
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8996,
+      "step": 2450
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5802,
+      "step": 2451
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2603,
+      "step": 2452
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.0921,
+      "step": 2453
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9377,
+      "step": 2454
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.0934,
+      "step": 2455
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9832,
+      "step": 2456
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1084,
+      "step": 2457
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2592,
+      "step": 2458
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8545,
+      "step": 2459
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4335,
+      "step": 2460
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5591,
+      "step": 2461
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.284,
+      "step": 2462
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.8171,
+      "step": 2463
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8541,
+      "step": 2464
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1355,
+      "step": 2465
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6885,
+      "step": 2466
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.311,
+      "step": 2467
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.72,
+      "step": 2468
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.007,
+      "step": 2469
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2275,
+      "step": 2470
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.129,
+      "step": 2471
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.9531,
+      "step": 2472
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7327,
+      "step": 2473
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5001,
+      "step": 2474
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.9443,
+      "step": 2475
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6005,
+      "step": 2476
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5551,
+      "step": 2477
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3044,
+      "step": 2478
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6257,
+      "step": 2479
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5029,
+      "step": 2480
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3717,
+      "step": 2481
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.5226,
+      "step": 2482
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2921,
+      "step": 2483
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7169,
+      "step": 2484
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.2185,
+      "step": 2485
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5393,
+      "step": 2486
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0286,
+      "step": 2487
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3738,
+      "step": 2488
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2249,
+      "step": 2489
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7828,
+      "step": 2490
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.87,
+      "step": 2491
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.681,
+      "step": 2492
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5446,
+      "step": 2493
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0769,
+      "step": 2494
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3587,
+      "step": 2495
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.9188,
+      "step": 2496
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9357,
+      "step": 2497
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3449,
+      "step": 2498
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.2753,
+      "step": 2499
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3077,
+      "step": 2500
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0085,
+      "step": 2501
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5436,
+      "step": 2502
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9096,
+      "step": 2503
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7288,
+      "step": 2504
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7344,
+      "step": 2505
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6385,
+      "step": 2506
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.6713,
+      "step": 2507
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.6065,
+      "step": 2508
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3456,
+      "step": 2509
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1633,
+      "step": 2510
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5557,
+      "step": 2511
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7075,
+      "step": 2512
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.4954,
+      "step": 2513
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5104,
+      "step": 2514
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5706,
+      "step": 2515
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7479,
+      "step": 2516
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7042,
+      "step": 2517
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9569,
+      "step": 2518
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7846,
+      "step": 2519
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.749,
+      "step": 2520
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5868,
+      "step": 2521
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3957,
+      "step": 2522
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2594,
+      "step": 2523
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.632,
+      "step": 2524
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.023,
+      "step": 2525
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0423,
+      "step": 2526
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1901,
+      "step": 2527
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.0025,
+      "step": 2528
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0252,
+      "step": 2529
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8165,
+      "step": 2530
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6864,
+      "step": 2531
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.1174,
+      "step": 2532
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.46,
+      "step": 2533
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3714,
+      "step": 2534
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1206,
+      "step": 2535
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3734,
+      "step": 2536
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7229,
+      "step": 2537
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.0124,
+      "step": 2538
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2105,
+      "step": 2539
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1127,
+      "step": 2540
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.1163,
+      "step": 2541
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.5255,
+      "step": 2542
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2058,
+      "step": 2543
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7425,
+      "step": 2544
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3621,
+      "step": 2545
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.7541,
+      "step": 2546
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9611,
+      "step": 2547
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3031,
+      "step": 2548
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.1756,
+      "step": 2549
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6146,
+      "step": 2550
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1434,
+      "step": 2551
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0786,
+      "step": 2552
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9498,
+      "step": 2553
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.8681,
+      "step": 2554
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5898,
+      "step": 2555
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7522,
+      "step": 2556
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3648,
+      "step": 2557
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.8895,
+      "step": 2558
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9903,
+      "step": 2559
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1113,
+      "step": 2560
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6541,
+      "step": 2561
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.8563,
+      "step": 2562
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 10.0685,
+      "step": 2563
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.59,
+      "step": 2564
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0927,
+      "step": 2565
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3792,
+      "step": 2566
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.208,
+      "step": 2567
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9275,
+      "step": 2568
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.25,
+      "step": 2569
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9524,
+      "step": 2570
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.556,
+      "step": 2571
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6414,
+      "step": 2572
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1749,
+      "step": 2573
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.4529,
+      "step": 2574
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9764,
+      "step": 2575
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1835,
+      "step": 2576
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.438,
+      "step": 2577
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.834,
+      "step": 2578
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8879,
+      "step": 2579
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1302,
+      "step": 2580
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.8966,
+      "step": 2581
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7319,
+      "step": 2582
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3287,
+      "step": 2583
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3322,
+      "step": 2584
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0278,
+      "step": 2585
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.5355,
+      "step": 2586
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2016,
+      "step": 2587
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8335,
+      "step": 2588
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.721,
+      "step": 2589
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4628,
+      "step": 2590
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7896,
+      "step": 2591
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.7226,
+      "step": 2592
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5834,
+      "step": 2593
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8944,
+      "step": 2594
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1519,
+      "step": 2595
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2173,
+      "step": 2596
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9751,
+      "step": 2597
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1408,
+      "step": 2598
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.2469,
+      "step": 2599
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3136,
+      "step": 2600
+    },
+    {
+      "epoch": 0.02,
+      "eval_loss": 6.580307483673096,
+      "eval_runtime": 22.5866,
+      "eval_samples_per_second": 2.214,
+      "eval_steps_per_second": 1.107,
+      "step": 2600
+    },
+    {
+      "epoch": 0.02,
+      "mmlu_eval_accuracy": 0.17715097402597402,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.21428571428571427,
+      "mmlu_eval_accuracy_astronomy": 0.3125,
+      "mmlu_eval_accuracy_business_ethics": 0.0,
+      "mmlu_loss": 3.684196367263794,
+      "step": 2600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4702,
+      "step": 2601
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2103,
+      "step": 2602
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1688,
+      "step": 2603
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0882,
+      "step": 2604
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4417,
+      "step": 2605
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.4982,
+      "step": 2606
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.3721,
+      "step": 2607
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.5558,
+      "step": 2608
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.251,
+      "step": 2609
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5517,
+      "step": 2610
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5841,
+      "step": 2611
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3849,
+      "step": 2612
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.5556,
+      "step": 2613
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.4158,
+      "step": 2614
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9386,
+      "step": 2615
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6376,
+      "step": 2616
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7976,
+      "step": 2617
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.291,
+      "step": 2618
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.8779,
+      "step": 2619
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.8159,
+      "step": 2620
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.1198,
+      "step": 2621
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9783,
+      "step": 2622
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0635,
+      "step": 2623
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8539,
+      "step": 2624
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5904,
+      "step": 2625
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7561,
+      "step": 2626
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3628,
+      "step": 2627
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.2452,
+      "step": 2628
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8438,
+      "step": 2629
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.7726,
+      "step": 2630
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 1.8356,
+      "step": 2631
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6763,
+      "step": 2632
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.9104,
+      "step": 2633
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1199,
+      "step": 2634
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4833,
+      "step": 2635
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.6164,
+      "step": 2636
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2284,
+      "step": 2637
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8304,
+      "step": 2638
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7706,
+      "step": 2639
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.638,
+      "step": 2640
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9576,
+      "step": 2641
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0828,
+      "step": 2642
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.5206,
+      "step": 2643
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.7744,
+      "step": 2644
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.5361,
+      "step": 2645
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9041,
+      "step": 2646
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6145,
+      "step": 2647
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9121,
+      "step": 2648
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.1322,
+      "step": 2649
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.1881,
+      "step": 2650
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6446,
+      "step": 2651
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9137,
+      "step": 2652
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4681,
+      "step": 2653
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9891,
+      "step": 2654
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3255,
+      "step": 2655
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.3909,
+      "step": 2656
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6869,
+      "step": 2657
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0611,
+      "step": 2658
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3314,
+      "step": 2659
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6415,
+      "step": 2660
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5881,
+      "step": 2661
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8889,
+      "step": 2662
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3336,
+      "step": 2663
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1282,
+      "step": 2664
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.158,
+      "step": 2665
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1773,
+      "step": 2666
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9617,
+      "step": 2667
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.5576,
+      "step": 2668
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8284,
+      "step": 2669
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.5936,
+      "step": 2670
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0931,
+      "step": 2671
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.07,
+      "step": 2672
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.778,
+      "step": 2673
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7786,
+      "step": 2674
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1279,
+      "step": 2675
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.463,
+      "step": 2676
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2192,
+      "step": 2677
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4147,
+      "step": 2678
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9957,
+      "step": 2679
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8919,
+      "step": 2680
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1558,
+      "step": 2681
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7069,
+      "step": 2682
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.487,
+      "step": 2683
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7098,
+      "step": 2684
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1032,
+      "step": 2685
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9937,
+      "step": 2686
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 10.0677,
+      "step": 2687
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.644,
+      "step": 2688
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5099,
+      "step": 2689
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6922,
+      "step": 2690
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7491,
+      "step": 2691
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.605,
+      "step": 2692
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1904,
+      "step": 2693
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.9175,
+      "step": 2694
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3251,
+      "step": 2695
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.315,
+      "step": 2696
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3052,
+      "step": 2697
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2122,
+      "step": 2698
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.9462,
+      "step": 2699
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3221,
+      "step": 2700
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3125,
+      "step": 2701
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.938,
+      "step": 2702
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0983,
+      "step": 2703
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8028,
+      "step": 2704
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4752,
+      "step": 2705
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.16,
+      "step": 2706
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2429,
+      "step": 2707
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.623,
+      "step": 2708
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9595,
+      "step": 2709
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5444,
+      "step": 2710
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.6245,
+      "step": 2711
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.919,
+      "step": 2712
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7332,
+      "step": 2713
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0067,
+      "step": 2714
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.6957,
+      "step": 2715
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.994,
+      "step": 2716
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7562,
+      "step": 2717
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6882,
+      "step": 2718
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.8732,
+      "step": 2719
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.6496,
+      "step": 2720
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4841,
+      "step": 2721
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4927,
+      "step": 2722
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7437,
+      "step": 2723
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9469,
+      "step": 2724
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1481,
+      "step": 2725
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7762,
+      "step": 2726
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8806,
+      "step": 2727
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8352,
+      "step": 2728
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9488,
+      "step": 2729
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1679,
+      "step": 2730
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2412,
+      "step": 2731
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6839,
+      "step": 2732
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7269,
+      "step": 2733
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.6139,
+      "step": 2734
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8754,
+      "step": 2735
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9007,
+      "step": 2736
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9134,
+      "step": 2737
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9469,
+      "step": 2738
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9293,
+      "step": 2739
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0489,
+      "step": 2740
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4197,
+      "step": 2741
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.3667,
+      "step": 2742
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8059,
+      "step": 2743
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.024,
+      "step": 2744
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0756,
+      "step": 2745
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0596,
+      "step": 2746
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.1416,
+      "step": 2747
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1271,
+      "step": 2748
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1452,
+      "step": 2749
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 1.9527,
+      "step": 2750
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9189,
+      "step": 2751
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4364,
+      "step": 2752
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.4173,
+      "step": 2753
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4034,
+      "step": 2754
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6678,
+      "step": 2755
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1092,
+      "step": 2756
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.7751,
+      "step": 2757
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0164,
+      "step": 2758
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.5796,
+      "step": 2759
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7851,
+      "step": 2760
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1784,
+      "step": 2761
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7417,
+      "step": 2762
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4268,
+      "step": 2763
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6919,
+      "step": 2764
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1838,
+      "step": 2765
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5592,
+      "step": 2766
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.425,
+      "step": 2767
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.848,
+      "step": 2768
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.5683,
+      "step": 2769
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0262,
+      "step": 2770
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.8063,
+      "step": 2771
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6139,
+      "step": 2772
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.3478,
+      "step": 2773
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1811,
+      "step": 2774
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4519,
+      "step": 2775
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0071,
+      "step": 2776
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7872,
+      "step": 2777
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2263,
+      "step": 2778
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8923,
+      "step": 2779
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2997,
+      "step": 2780
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.6857,
+      "step": 2781
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8874,
+      "step": 2782
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8203,
+      "step": 2783
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.9583,
+      "step": 2784
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0814,
+      "step": 2785
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.419,
+      "step": 2786
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.3063,
+      "step": 2787
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1285,
+      "step": 2788
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0039,
+      "step": 2789
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.671,
+      "step": 2790
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.5452,
+      "step": 2791
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.3885,
+      "step": 2792
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6282,
+      "step": 2793
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5913,
+      "step": 2794
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6189,
+      "step": 2795
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2968,
+      "step": 2796
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2825,
+      "step": 2797
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9897,
+      "step": 2798
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.8193,
+      "step": 2799
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7042,
+      "step": 2800
+    },
+    {
+      "epoch": 0.02,
+      "eval_loss": 6.604581832885742,
+      "eval_runtime": 22.516,
+      "eval_samples_per_second": 2.221,
+      "eval_steps_per_second": 1.11,
+      "step": 2800
+    },
+    {
+      "epoch": 0.02,
+      "mmlu_eval_accuracy": 0.2525477994227994,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.07142857142857142,
+      "mmlu_eval_accuracy_astronomy": 0.3125,
+      "mmlu_eval_accuracy_business_ethics": 0.4444444444444444,
+      "mmlu_loss": 4.006761569976806,
+      "step": 2800
+    },
+    {
+      "epoch": 0.02,
+      "step": 2800,
+      "total_flos": 4.660001608148582e+16,
+      "train_loss": 6.312225336258395,
+      "train_runtime": 7855.0688,
+      "train_samples_per_second": 3.819,
+      "train_steps_per_second": 3.819
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0757,
+      "step": 2801
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.8944,
+      "step": 2802
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8365,
+      "step": 2803
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.873,
+      "step": 2804
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3793,
+      "step": 2805
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1923,
+      "step": 2806
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2306,
+      "step": 2807
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.4439,
+      "step": 2808
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3377,
+      "step": 2809
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8737,
+      "step": 2810
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4191,
+      "step": 2811
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.892,
+      "step": 2812
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4568,
+      "step": 2813
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0652,
+      "step": 2814
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6912,
+      "step": 2815
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9033,
+      "step": 2816
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4433,
+      "step": 2817
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7591,
+      "step": 2818
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4458,
+      "step": 2819
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3721,
+      "step": 2820
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4111,
+      "step": 2821
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0999,
+      "step": 2822
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.5154,
+      "step": 2823
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1967,
+      "step": 2824
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8437,
+      "step": 2825
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.759,
+      "step": 2826
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6223,
+      "step": 2827
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3442,
+      "step": 2828
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1991,
+      "step": 2829
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5302,
+      "step": 2830
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1725,
+      "step": 2831
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8107,
+      "step": 2832
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.7891,
+      "step": 2833
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5592,
+      "step": 2834
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8792,
+      "step": 2835
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2387,
+      "step": 2836
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9369,
+      "step": 2837
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2098,
+      "step": 2838
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6645,
+      "step": 2839
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2349,
+      "step": 2840
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8613,
+      "step": 2841
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5482,
+      "step": 2842
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5438,
+      "step": 2843
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6133,
+      "step": 2844
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9983,
+      "step": 2845
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8706,
+      "step": 2846
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9685,
+      "step": 2847
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.04,
+      "step": 2848
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.6498,
+      "step": 2849
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6786,
+      "step": 2850
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.789,
+      "step": 2851
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 10.1116,
+      "step": 2852
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.7085,
+      "step": 2853
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1083,
+      "step": 2854
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0795,
+      "step": 2855
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8677,
+      "step": 2856
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1663,
+      "step": 2857
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5379,
+      "step": 2858
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4923,
+      "step": 2859
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1456,
+      "step": 2860
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1686,
+      "step": 2861
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4153,
+      "step": 2862
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.17,
+      "step": 2863
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3301,
+      "step": 2864
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7106,
+      "step": 2865
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.463,
+      "step": 2866
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.614,
+      "step": 2867
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1451,
+      "step": 2868
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.6817,
+      "step": 2869
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.9577,
+      "step": 2870
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.6171,
+      "step": 2871
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5129,
+      "step": 2872
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3386,
+      "step": 2873
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1176,
+      "step": 2874
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9075,
+      "step": 2875
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.667,
+      "step": 2876
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.8097,
+      "step": 2877
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.7707,
+      "step": 2878
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.7571,
+      "step": 2879
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0732,
+      "step": 2880
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5586,
+      "step": 2881
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8045,
+      "step": 2882
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.4442,
+      "step": 2883
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.418,
+      "step": 2884
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7253,
+      "step": 2885
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4227,
+      "step": 2886
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.9878,
+      "step": 2887
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.8351,
+      "step": 2888
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1715,
+      "step": 2889
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1207,
+      "step": 2890
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0396,
+      "step": 2891
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7162,
+      "step": 2892
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2566,
+      "step": 2893
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4087,
+      "step": 2894
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.4509,
+      "step": 2895
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8165,
+      "step": 2896
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.9842,
+      "step": 2897
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.574,
+      "step": 2898
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4512,
+      "step": 2899
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9904,
+      "step": 2900
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6141,
+      "step": 2901
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9528,
+      "step": 2902
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.9028,
+      "step": 2903
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3868,
+      "step": 2904
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0461,
+      "step": 2905
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.5161,
+      "step": 2906
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.107,
+      "step": 2907
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7174,
+      "step": 2908
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7671,
+      "step": 2909
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6803,
+      "step": 2910
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5357,
+      "step": 2911
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6898,
+      "step": 2912
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8564,
+      "step": 2913
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1457,
+      "step": 2914
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3408,
+      "step": 2915
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.6095,
+      "step": 2916
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.866,
+      "step": 2917
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.7324,
+      "step": 2918
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4026,
+      "step": 2919
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1467,
+      "step": 2920
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2496,
+      "step": 2921
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.5265,
+      "step": 2922
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8246,
+      "step": 2923
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.5775,
+      "step": 2924
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2886,
+      "step": 2925
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3076,
+      "step": 2926
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7975,
+      "step": 2927
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9898,
+      "step": 2928
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7256,
+      "step": 2929
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7971,
+      "step": 2930
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5884,
+      "step": 2931
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0749,
+      "step": 2932
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.6183,
+      "step": 2933
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.0019,
+      "step": 2934
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1706,
+      "step": 2935
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.4075,
+      "step": 2936
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.4092,
+      "step": 2937
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.9424,
+      "step": 2938
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.9025,
+      "step": 2939
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7751,
+      "step": 2940
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.988,
+      "step": 2941
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1208,
+      "step": 2942
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1999,
+      "step": 2943
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2129,
+      "step": 2944
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4431,
+      "step": 2945
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1979,
+      "step": 2946
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8246,
+      "step": 2947
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.4876,
+      "step": 2948
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.7158,
+      "step": 2949
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3611,
+      "step": 2950
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.9919,
+      "step": 2951
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4136,
+      "step": 2952
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.801,
+      "step": 2953
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.6503,
+      "step": 2954
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.553,
+      "step": 2955
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3536,
+      "step": 2956
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8516,
+      "step": 2957
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.9344,
+      "step": 2958
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8727,
+      "step": 2959
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9155,
+      "step": 2960
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9684,
+      "step": 2961
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0399,
+      "step": 2962
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.4298,
+      "step": 2963
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.4559,
+      "step": 2964
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0361,
+      "step": 2965
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0081,
+      "step": 2966
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.6698,
+      "step": 2967
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.3355,
+      "step": 2968
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.7555,
+      "step": 2969
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.29,
+      "step": 2970
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4079,
+      "step": 2971
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0588,
+      "step": 2972
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2956,
+      "step": 2973
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7318,
+      "step": 2974
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8846,
+      "step": 2975
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5159,
+      "step": 2976
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7629,
+      "step": 2977
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2039,
+      "step": 2978
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.311,
+      "step": 2979
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9786,
+      "step": 2980
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7138,
+      "step": 2981
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4827,
+      "step": 2982
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5264,
+      "step": 2983
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8153,
+      "step": 2984
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3512,
+      "step": 2985
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1515,
+      "step": 2986
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1689,
+      "step": 2987
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8027,
+      "step": 2988
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7472,
+      "step": 2989
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0236,
+      "step": 2990
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1741,
+      "step": 2991
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8884,
+      "step": 2992
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.3912,
+      "step": 2993
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2109,
+      "step": 2994
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1693,
+      "step": 2995
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8166,
+      "step": 2996
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4902,
+      "step": 2997
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3921,
+      "step": 2998
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.8787,
+      "step": 2999
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1304,
+      "step": 3000
+    },
+    {
+      "epoch": 0.02,
+      "eval_loss": 6.659167289733887,
+      "eval_runtime": 22.4512,
+      "eval_samples_per_second": 2.227,
+      "eval_steps_per_second": 1.114,
+      "step": 3000
+    },
+    {
+      "epoch": 0.02,
+      "mmlu_eval_accuracy": 0.2525477994227994,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.07142857142857142,
+      "mmlu_eval_accuracy_astronomy": 0.3125,
+      "mmlu_eval_accuracy_business_ethics": 0.4444444444444444,
+      "mmlu_loss": 3.973116703033447,
+      "step": 3000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.4366,
+      "step": 3001
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1239,
+      "step": 3002
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.419,
+      "step": 3003
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7662,
+      "step": 3004
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1246,
+      "step": 3005
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3498,
+      "step": 3006
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1245,
+      "step": 3007
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6488,
+      "step": 3008
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3641,
+      "step": 3009
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7842,
+      "step": 3010
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.802,
+      "step": 3011
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1015,
+      "step": 3012
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.9015,
+      "step": 3013
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8768,
+      "step": 3014
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7296,
+      "step": 3015
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4175,
+      "step": 3016
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3568,
+      "step": 3017
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5869,
+      "step": 3018
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5304,
+      "step": 3019
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1129,
+      "step": 3020
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8128,
+      "step": 3021
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1541,
+      "step": 3022
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.3002,
+      "step": 3023
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0616,
+      "step": 3024
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.3152,
+      "step": 3025
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4433,
+      "step": 3026
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.8406,
+      "step": 3027
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2653,
+      "step": 3028
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7041,
+      "step": 3029
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3463,
+      "step": 3030
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7164,
+      "step": 3031
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.9969,
+      "step": 3032
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1306,
+      "step": 3033
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0204,
+      "step": 3034
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6743,
+      "step": 3035
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3186,
+      "step": 3036
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.5673,
+      "step": 3037
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1416,
+      "step": 3038
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1956,
+      "step": 3039
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6836,
+      "step": 3040
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0172,
+      "step": 3041
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.796,
+      "step": 3042
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6945,
+      "step": 3043
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.5079,
+      "step": 3044
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.399,
+      "step": 3045
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0071,
+      "step": 3046
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.4171,
+      "step": 3047
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0925,
+      "step": 3048
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.6842,
+      "step": 3049
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.2442,
+      "step": 3050
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8023,
+      "step": 3051
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7147,
+      "step": 3052
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.9681,
+      "step": 3053
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.1882,
+      "step": 3054
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9869,
+      "step": 3055
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0705,
+      "step": 3056
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8802,
+      "step": 3057
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.8399,
+      "step": 3058
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.6438,
+      "step": 3059
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0608,
+      "step": 3060
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.763,
+      "step": 3061
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.355,
+      "step": 3062
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5504,
+      "step": 3063
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1582,
+      "step": 3064
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1922,
+      "step": 3065
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.0706,
+      "step": 3066
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.807,
+      "step": 3067
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0305,
+      "step": 3068
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0774,
+      "step": 3069
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4469,
+      "step": 3070
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1158,
+      "step": 3071
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8087,
+      "step": 3072
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.5672,
+      "step": 3073
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.5864,
+      "step": 3074
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.7665,
+      "step": 3075
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2434,
+      "step": 3076
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3988,
+      "step": 3077
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0307,
+      "step": 3078
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6561,
+      "step": 3079
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8158,
+      "step": 3080
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8384,
+      "step": 3081
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5515,
+      "step": 3082
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8108,
+      "step": 3083
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2536,
+      "step": 3084
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2506,
+      "step": 3085
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1605,
+      "step": 3086
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.4572,
+      "step": 3087
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.3312,
+      "step": 3088
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1481,
+      "step": 3089
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.3304,
+      "step": 3090
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2617,
+      "step": 3091
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.3148,
+      "step": 3092
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.4781,
+      "step": 3093
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.327,
+      "step": 3094
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3288,
+      "step": 3095
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2354,
+      "step": 3096
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.4802,
+      "step": 3097
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.1659,
+      "step": 3098
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 1.9887,
+      "step": 3099
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 1.9497,
+      "step": 3100
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2668,
+      "step": 3101
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.458,
+      "step": 3102
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.9919,
+      "step": 3103
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0408,
+      "step": 3104
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9966,
+      "step": 3105
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 9.1371,
+      "step": 3106
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0939,
+      "step": 3107
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2532,
+      "step": 3108
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.7343,
+      "step": 3109
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.156,
+      "step": 3110
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2223,
+      "step": 3111
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6182,
+      "step": 3112
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4423,
+      "step": 3113
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3855,
+      "step": 3114
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.2115,
+      "step": 3115
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6719,
+      "step": 3116
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5285,
+      "step": 3117
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.0171,
+      "step": 3118
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2127,
+      "step": 3119
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.8153,
+      "step": 3120
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1838,
+      "step": 3121
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.711,
+      "step": 3122
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1568,
+      "step": 3123
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.3741,
+      "step": 3124
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2741,
+      "step": 3125
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.1653,
+      "step": 3126
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.9722,
+      "step": 3127
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9181,
+      "step": 3128
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.384,
+      "step": 3129
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.1491,
+      "step": 3130
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.8641,
+      "step": 3131
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.6125,
+      "step": 3132
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1702,
+      "step": 3133
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.4853,
+      "step": 3134
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7929,
+      "step": 3135
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8747,
+      "step": 3136
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.2659,
+      "step": 3137
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.0685,
+      "step": 3138
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2297,
+      "step": 3139
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0715,
+      "step": 3140
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2201,
+      "step": 3141
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2134,
+      "step": 3142
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.6098,
+      "step": 3143
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.2036,
+      "step": 3144
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.2481,
+      "step": 3145
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.4508,
+      "step": 3146
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.1454,
+      "step": 3147
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7573,
+      "step": 3148
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2946,
+      "step": 3149
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.0121,
+      "step": 3150
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.498,
+      "step": 3151
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.4016,
+      "step": 3152
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5934,
+      "step": 3153
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.737,
+      "step": 3154
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.9874,
+      "step": 3155
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7714,
+      "step": 3156
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.3529,
+      "step": 3157
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.7673,
+      "step": 3158
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.3835,
+      "step": 3159
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.0336,
+      "step": 3160
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.2998,
+      "step": 3161
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0446,
+      "step": 3162
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5225,
+      "step": 3163
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.1391,
+      "step": 3164
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.7669,
+      "step": 3165
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.118,
+      "step": 3166
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.7045,
+      "step": 3167
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.077,
+      "step": 3168
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.0697,
+      "step": 3169
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.8445,
+      "step": 3170
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.4044,
+      "step": 3171
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9817,
+      "step": 3172
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.2508,
+      "step": 3173
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6302,
+      "step": 3174
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.46,
+      "step": 3175
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.8094,
+      "step": 3176
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.1174,
+      "step": 3177
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.5756,
+      "step": 3178
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.4215,
+      "step": 3179
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 8.5957,
+      "step": 3180
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.3534,
+      "step": 3181
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.9342,
+      "step": 3182
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.8227,
+      "step": 3183
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.1404,
+      "step": 3184
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.9897,
+      "step": 3185
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7286,
+      "step": 3186
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.5414,
+      "step": 3187
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.7523,
+      "step": 3188
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.4033,
+      "step": 3189
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.8593,
+      "step": 3190
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 6.6095,
+      "step": 3191
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.504,
+      "step": 3192
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.6776,
+      "step": 3193
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 7.0174,
+      "step": 3194
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 5.7759,
+      "step": 3195
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.5685,
+      "step": 3196
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.2786,
+      "step": 3197
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 3.7794,
+      "step": 3198
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 2.5827,
+      "step": 3199
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0004,
+      "loss": 4.3282,
+      "step": 3200
+    },
+    {
+      "epoch": 0.02,
+      "eval_loss": 6.423073768615723,
+      "eval_runtime": 22.4644,
+      "eval_samples_per_second": 2.226,
+      "eval_steps_per_second": 1.113,
+      "step": 3200
+    },
+    {
+      "epoch": 0.02,
+      "mmlu_eval_accuracy": 0.2525477994227994,
+      "mmlu_eval_accuracy_abstract_algebra": 0.18181818181818182,
+      "mmlu_eval_accuracy_anatomy": 0.07142857142857142,
+      "mmlu_eval_accuracy_astronomy": 0.3125,
+      "mmlu_eval_accuracy_business_ethics": 0.4444444444444444,
+      "mmlu_loss": 3.9956862831115725,
+      "step": 3200
+    }
+  ],
+  "max_steps": 30000,
+  "num_train_epochs": 1,
+  "total_flos": 5.284691537608704e+16,
+  "trial_name": null,
+  "trial_params": null
+}