diff --git "a/checkpoint-1980/trainer_state.json" "b/checkpoint-1980/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-1980/trainer_state.json"
@@ -0,0 +1,13975 @@
+{
+  "best_metric": 0.1333775818347931,
+  "best_model_checkpoint": "segformer-b0-finetuned-segments-sidewalk-outputs/checkpoint-1380",
+  "epoch": 49.5,
+  "global_step": 1980,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.03,
+      "learning_rate": 5.9970000000000004e-05,
+      "loss": 1.2082,
+      "step": 1
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 5.994e-05,
+      "loss": 1.0322,
+      "step": 2
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 5.991000000000001e-05,
+      "loss": 0.9391,
+      "step": 3
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 5.988e-05,
+      "loss": 0.9601,
+      "step": 4
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 5.9850000000000005e-05,
+      "loss": 0.9436,
+      "step": 5
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 5.982e-05,
+      "loss": 0.8901,
+      "step": 6
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 5.979e-05,
+      "loss": 0.9326,
+      "step": 7
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 5.9760000000000004e-05,
+      "loss": 0.9164,
+      "step": 8
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 5.9730000000000006e-05,
+      "loss": 0.8523,
+      "step": 9
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 5.97e-05,
+      "loss": 0.7235,
+      "step": 10
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 5.967e-05,
+      "loss": 0.9828,
+      "step": 11
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 5.964e-05,
+      "loss": 0.7795,
+      "step": 12
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 5.961000000000001e-05,
+      "loss": 0.8967,
+      "step": 13
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 5.958e-05,
+      "loss": 0.8618,
+      "step": 14
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 5.9550000000000004e-05,
+      "loss": 0.8513,
+      "step": 15
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.952e-05,
+      "loss": 0.7067,
+      "step": 16
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.949e-05,
+      "loss": 0.7935,
+      "step": 17
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.946e-05,
+      "loss": 0.9116,
+      "step": 18
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.9430000000000005e-05,
+      "loss": 0.6582,
+      "step": 19
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.94e-05,
+      "loss": 0.6827,
+      "step": 20
+    },
+    {
+      "epoch": 0.5,
+      "eval_loss": 0.8834543228149414,
+      "eval_mean_accuracy": 0.6718939356205333,
+      "eval_mean_iou": 0.2607108677425002,
+      "eval_overall_accuracy": 0.5466109819427459,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.44492668085864034,
+        0.8988611903824264
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.43746558076697917,
+        0.3446670224605214
+      ],
+      "eval_runtime": 47.0306,
+      "eval_samples_per_second": 0.425,
+      "eval_steps_per_second": 0.213,
+      "step": 20
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 5.937e-05,
+      "loss": 1.1641,
+      "step": 21
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 5.934e-05,
+      "loss": 0.7001,
+      "step": 22
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 5.9310000000000006e-05,
+      "loss": 0.6366,
+      "step": 23
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 5.928e-05,
+      "loss": 0.7333,
+      "step": 24
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 5.9250000000000004e-05,
+      "loss": 0.731,
+      "step": 25
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 5.922e-05,
+      "loss": 0.6953,
+      "step": 26
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 5.919e-05,
+      "loss": 0.7026,
+      "step": 27
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 5.916e-05,
+      "loss": 0.633,
+      "step": 28
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 5.9130000000000005e-05,
+      "loss": 0.6653,
+      "step": 29
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 5.91e-05,
+      "loss": 0.6987,
+      "step": 30
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 5.907e-05,
+      "loss": 0.6367,
+      "step": 31
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 5.9040000000000004e-05,
+      "loss": 0.5351,
+      "step": 32
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 5.9010000000000006e-05,
+      "loss": 0.7612,
+      "step": 33
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 5.898e-05,
+      "loss": 1.0138,
+      "step": 34
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.895e-05,
+      "loss": 0.6202,
+      "step": 35
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.892e-05,
+      "loss": 0.6921,
+      "step": 36
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 5.889000000000001e-05,
+      "loss": 0.5186,
+      "step": 37
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 5.886e-05,
+      "loss": 0.5955,
+      "step": 38
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 5.8830000000000004e-05,
+      "loss": 0.6696,
+      "step": 39
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5.88e-05,
+      "loss": 0.6717,
+      "step": 40
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.7762032747268677,
+      "eval_mean_accuracy": 0.756401809675905,
+      "eval_mean_iou": 0.3105024097830028,
+      "eval_overall_accuracy": 0.6408774372137014,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.5471135633014456,
+        0.9656900560503644
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.5429513256944513,
+        0.38855590365455717
+      ],
+      "eval_runtime": 47.4699,
+      "eval_samples_per_second": 0.421,
+      "eval_steps_per_second": 0.211,
+      "step": 40
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 5.877e-05,
+      "loss": 0.5692,
+      "step": 41
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 5.8740000000000003e-05,
+      "loss": 0.5096,
+      "step": 42
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 5.8710000000000005e-05,
+      "loss": 0.5457,
+      "step": 43
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 5.868e-05,
+      "loss": 0.4582,
+      "step": 44
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 5.865e-05,
+      "loss": 0.6143,
+      "step": 45
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 5.862e-05,
+      "loss": 0.6951,
+      "step": 46
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 5.8590000000000007e-05,
+      "loss": 0.5025,
+      "step": 47
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 5.856e-05,
+      "loss": 1.1675,
+      "step": 48
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 5.8530000000000004e-05,
+      "loss": 1.0585,
+      "step": 49
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 5.85e-05,
+      "loss": 0.6964,
+      "step": 50
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 5.847e-05,
+      "loss": 0.5112,
+      "step": 51
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 5.844e-05,
+      "loss": 0.822,
+      "step": 52
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 5.8410000000000005e-05,
+      "loss": 0.547,
+      "step": 53
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 5.838e-05,
+      "loss": 0.443,
+      "step": 54
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 5.835e-05,
+      "loss": 0.7045,
+      "step": 55
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 5.832e-05,
+      "loss": 0.5225,
+      "step": 56
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 5.8290000000000006e-05,
+      "loss": 0.4225,
+      "step": 57
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 5.826e-05,
+      "loss": 0.5549,
+      "step": 58
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 5.823e-05,
+      "loss": 0.4284,
+      "step": 59
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 5.82e-05,
+      "loss": 0.5073,
+      "step": 60
+    },
+    {
+      "epoch": 1.5,
+      "eval_loss": 0.5137239694595337,
+      "eval_mean_accuracy": 0.8241622297344804,
+      "eval_mean_iou": 0.48694262570265306,
+      "eval_overall_accuracy": 0.8814629163262906,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9279702830244996,
+        0.7203541764444612
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8606411961780306,
+        0.6001866809299288
+      ],
+      "eval_runtime": 48.2702,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 60
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 5.817e-05,
+      "loss": 0.4515,
+      "step": 61
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 5.814e-05,
+      "loss": 0.9959,
+      "step": 62
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 5.8110000000000004e-05,
+      "loss": 0.7347,
+      "step": 63
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 5.808e-05,
+      "loss": 0.6901,
+      "step": 64
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 5.805e-05,
+      "loss": 0.3837,
+      "step": 65
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 5.802e-05,
+      "loss": 0.6317,
+      "step": 66
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 5.7990000000000006e-05,
+      "loss": 0.5058,
+      "step": 67
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 5.796e-05,
+      "loss": 0.403,
+      "step": 68
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 5.793e-05,
+      "loss": 0.5889,
+      "step": 69
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 5.79e-05,
+      "loss": 0.421,
+      "step": 70
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 5.787e-05,
+      "loss": 0.5823,
+      "step": 71
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 5.784e-05,
+      "loss": 0.5088,
+      "step": 72
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 5.7810000000000004e-05,
+      "loss": 0.6356,
+      "step": 73
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 5.778e-05,
+      "loss": 0.433,
+      "step": 74
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 5.775e-05,
+      "loss": 0.301,
+      "step": 75
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 5.7719999999999996e-05,
+      "loss": 0.4938,
+      "step": 76
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 5.7690000000000005e-05,
+      "loss": 0.4548,
+      "step": 77
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 5.766e-05,
+      "loss": 0.3456,
+      "step": 78
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 5.763e-05,
+      "loss": 0.4227,
+      "step": 79
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 5.76e-05,
+      "loss": 0.645,
+      "step": 80
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 0.5222005844116211,
+      "eval_mean_accuracy": 0.8593003570537622,
+      "eval_mean_iou": 0.4324267875542314,
+      "eval_overall_accuracy": 0.7904938796752369,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.734647985585855,
+        0.9839527285216694
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.7340596529537314,
+        0.5632207097089628
+      ],
+      "eval_runtime": 48.1367,
+      "eval_samples_per_second": 0.415,
+      "eval_steps_per_second": 0.208,
+      "step": 80
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 5.757e-05,
+      "loss": 0.3101,
+      "step": 81
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 5.754e-05,
+      "loss": 0.4324,
+      "step": 82
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 5.751e-05,
+      "loss": 0.5271,
+      "step": 83
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 5.748e-05,
+      "loss": 0.4749,
+      "step": 84
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 5.745e-05,
+      "loss": 0.4237,
+      "step": 85
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 5.742e-05,
+      "loss": 0.5763,
+      "step": 86
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 5.7390000000000004e-05,
+      "loss": 0.4159,
+      "step": 87
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 5.736e-05,
+      "loss": 0.5616,
+      "step": 88
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 5.733e-05,
+      "loss": 0.5792,
+      "step": 89
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 5.73e-05,
+      "loss": 0.4864,
+      "step": 90
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 5.7270000000000006e-05,
+      "loss": 0.4517,
+      "step": 91
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 5.724e-05,
+      "loss": 0.4227,
+      "step": 92
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 5.721e-05,
+      "loss": 1.0034,
+      "step": 93
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 5.718e-05,
+      "loss": 0.5311,
+      "step": 94
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 5.715e-05,
+      "loss": 0.4184,
+      "step": 95
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 5.712e-05,
+      "loss": 0.465,
+      "step": 96
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 5.7090000000000004e-05,
+      "loss": 0.7336,
+      "step": 97
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 5.706e-05,
+      "loss": 0.7865,
+      "step": 98
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 5.703e-05,
+      "loss": 0.4739,
+      "step": 99
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 5.6999999999999996e-05,
+      "loss": 0.4173,
+      "step": 100
+    },
+    {
+      "epoch": 2.5,
+      "eval_loss": 0.34641411900520325,
+      "eval_mean_accuracy": 0.7393900107293994,
+      "eval_mean_iou": 0.4519346247035367,
+      "eval_overall_accuracy": 0.8488291062247336,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9376539432040742,
+        0.5411260782547244
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8348604687385176,
+        0.5209434053720925
+      ],
+      "eval_runtime": 48.5002,
+      "eval_samples_per_second": 0.412,
+      "eval_steps_per_second": 0.206,
+      "step": 100
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 5.6970000000000005e-05,
+      "loss": 0.5154,
+      "step": 101
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 5.694e-05,
+      "loss": 0.4903,
+      "step": 102
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 5.691e-05,
+      "loss": 0.4885,
+      "step": 103
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 5.688e-05,
+      "loss": 0.4041,
+      "step": 104
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 5.685e-05,
+      "loss": 0.3674,
+      "step": 105
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 5.682e-05,
+      "loss": 0.2295,
+      "step": 106
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 5.6790000000000003e-05,
+      "loss": 1.0179,
+      "step": 107
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 5.676e-05,
+      "loss": 0.2604,
+      "step": 108
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 5.673e-05,
+      "loss": 0.3548,
+      "step": 109
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 5.6699999999999996e-05,
+      "loss": 0.4074,
+      "step": 110
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 5.6670000000000005e-05,
+      "loss": 0.6856,
+      "step": 111
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 5.664e-05,
+      "loss": 1.1289,
+      "step": 112
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 5.661e-05,
+      "loss": 0.4405,
+      "step": 113
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 5.658e-05,
+      "loss": 0.8337,
+      "step": 114
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 5.655e-05,
+      "loss": 0.3255,
+      "step": 115
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 5.652e-05,
+      "loss": 0.5069,
+      "step": 116
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 5.649e-05,
+      "loss": 0.4483,
+      "step": 117
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 5.646e-05,
+      "loss": 0.368,
+      "step": 118
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 5.643e-05,
+      "loss": 0.7829,
+      "step": 119
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 5.6399999999999995e-05,
+      "loss": 0.8881,
+      "step": 120
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 0.3434440791606903,
+      "eval_mean_accuracy": 0.8174777092760303,
+      "eval_mean_iou": 0.5078240370773638,
+      "eval_overall_accuracy": 0.8651856882525534,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9039072569572595,
+        0.731048161594801
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8416629247588947,
+        0.6818091864731967
+      ],
+      "eval_runtime": 49.2759,
+      "eval_samples_per_second": 0.406,
+      "eval_steps_per_second": 0.203,
+      "step": 120
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 5.6370000000000004e-05,
+      "loss": 0.4869,
+      "step": 121
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 5.634e-05,
+      "loss": 0.2544,
+      "step": 122
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 5.631e-05,
+      "loss": 0.5558,
+      "step": 123
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 5.6279999999999996e-05,
+      "loss": 0.5816,
+      "step": 124
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 5.625e-05,
+      "loss": 0.5193,
+      "step": 125
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 5.622000000000001e-05,
+      "loss": 0.4635,
+      "step": 126
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 5.619e-05,
+      "loss": 0.4327,
+      "step": 127
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 5.6160000000000004e-05,
+      "loss": 0.3984,
+      "step": 128
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 5.613e-05,
+      "loss": 0.391,
+      "step": 129
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 5.61e-05,
+      "loss": 0.6207,
+      "step": 130
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 5.6070000000000004e-05,
+      "loss": 0.3424,
+      "step": 131
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 5.6040000000000006e-05,
+      "loss": 0.5947,
+      "step": 132
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 5.601e-05,
+      "loss": 0.2676,
+      "step": 133
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 5.598e-05,
+      "loss": 0.3732,
+      "step": 134
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 5.595e-05,
+      "loss": 0.523,
+      "step": 135
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 5.592000000000001e-05,
+      "loss": 0.3125,
+      "step": 136
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 5.589e-05,
+      "loss": 0.4795,
+      "step": 137
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 5.5860000000000004e-05,
+      "loss": 0.2888,
+      "step": 138
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 5.583e-05,
+      "loss": 0.5451,
+      "step": 139
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 5.58e-05,
+      "loss": 0.3846,
+      "step": 140
+    },
+    {
+      "epoch": 3.5,
+      "eval_loss": 0.31755417585372925,
+      "eval_mean_accuracy": 0.9216071847563998,
+      "eval_mean_iou": 0.569468718641803,
+      "eval_overall_accuracy": 0.9143545260756287,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9084679987604084,
+        0.9347463707523912
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8941303774730092,
+        0.8142757784524
+      ],
+      "eval_runtime": 49.4017,
+      "eval_samples_per_second": 0.405,
+      "eval_steps_per_second": 0.202,
+      "step": 140
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 5.577e-05,
+      "loss": 0.5607,
+      "step": 141
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 5.5740000000000005e-05,
+      "loss": 0.5292,
+      "step": 142
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 5.571e-05,
+      "loss": 0.2441,
+      "step": 143
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 5.568e-05,
+      "loss": 0.2737,
+      "step": 144
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 5.5650000000000004e-05,
+      "loss": 0.4429,
+      "step": 145
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 5.5620000000000006e-05,
+      "loss": 0.4026,
+      "step": 146
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 5.559e-05,
+      "loss": 0.4234,
+      "step": 147
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 5.556e-05,
+      "loss": 0.4091,
+      "step": 148
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 5.553e-05,
+      "loss": 0.6148,
+      "step": 149
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 5.550000000000001e-05,
+      "loss": 1.3302,
+      "step": 150
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 5.547e-05,
+      "loss": 0.2761,
+      "step": 151
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 5.5440000000000005e-05,
+      "loss": 0.3663,
+      "step": 152
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 5.541e-05,
+      "loss": 0.2674,
+      "step": 153
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 5.538e-05,
+      "loss": 0.2528,
+      "step": 154
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 5.5350000000000004e-05,
+      "loss": 0.6118,
+      "step": 155
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 5.5320000000000006e-05,
+      "loss": 0.3095,
+      "step": 156
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 5.529e-05,
+      "loss": 0.3554,
+      "step": 157
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 5.526e-05,
+      "loss": 0.4051,
+      "step": 158
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 5.523e-05,
+      "loss": 0.315,
+      "step": 159
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 5.520000000000001e-05,
+      "loss": 0.9319,
+      "step": 160
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 0.2767624855041504,
+      "eval_mean_accuracy": 0.9090526773584855,
+      "eval_mean_iou": 0.5241215752108322,
+      "eval_overall_accuracy": 0.8693237754113918,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.8370783221494974,
+        0.9810270325674735
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8342241618348791,
+        0.7381405637976174
+      ],
+      "eval_runtime": 48.9178,
+      "eval_samples_per_second": 0.409,
+      "eval_steps_per_second": 0.204,
+      "step": 160
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 5.517e-05,
+      "loss": 0.384,
+      "step": 161
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 5.5140000000000004e-05,
+      "loss": 0.2424,
+      "step": 162
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 5.511e-05,
+      "loss": 0.5975,
+      "step": 163
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 5.508e-05,
+      "loss": 0.1757,
+      "step": 164
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 5.505e-05,
+      "loss": 0.3585,
+      "step": 165
+    },
+    {
+      "epoch": 4.15,
+      "learning_rate": 5.5020000000000005e-05,
+      "loss": 0.349,
+      "step": 166
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 5.499e-05,
+      "loss": 0.243,
+      "step": 167
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 5.496e-05,
+      "loss": 0.4824,
+      "step": 168
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 5.493e-05,
+      "loss": 0.2798,
+      "step": 169
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 5.4900000000000006e-05,
+      "loss": 0.3875,
+      "step": 170
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 5.487e-05,
+      "loss": 0.7034,
+      "step": 171
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 5.4840000000000003e-05,
+      "loss": 0.1977,
+      "step": 172
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 5.481e-05,
+      "loss": 0.1849,
+      "step": 173
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 5.478e-05,
+      "loss": 0.2269,
+      "step": 174
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 5.475e-05,
+      "loss": 0.2172,
+      "step": 175
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 5.4720000000000005e-05,
+      "loss": 0.9048,
+      "step": 176
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 5.469e-05,
+      "loss": 0.2553,
+      "step": 177
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 5.466e-05,
+      "loss": 0.4218,
+      "step": 178
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 5.463e-05,
+      "loss": 0.288,
+      "step": 179
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 5.4600000000000006e-05,
+      "loss": 0.4806,
+      "step": 180
+    },
+    {
+      "epoch": 4.5,
+      "eval_loss": 0.38951587677001953,
+      "eval_mean_accuracy": 0.9439787770292487,
+      "eval_mean_iou": 0.5620683451115345,
+      "eval_overall_accuracy": 0.9275893154935977,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9142870194311686,
+        0.9736705346273289
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9096082836264106,
+        0.776596751708193
+      ],
+      "eval_runtime": 48.7001,
+      "eval_samples_per_second": 0.411,
+      "eval_steps_per_second": 0.205,
+      "step": 180
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 5.457e-05,
+      "loss": 0.5392,
+      "step": 181
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 5.454e-05,
+      "loss": 0.37,
+      "step": 182
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 5.451e-05,
+      "loss": 0.3083,
+      "step": 183
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 5.448e-05,
+      "loss": 0.5515,
+      "step": 184
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 5.445e-05,
+      "loss": 0.4385,
+      "step": 185
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 5.4420000000000004e-05,
+      "loss": 0.252,
+      "step": 186
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 5.439e-05,
+      "loss": 0.4565,
+      "step": 187
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 5.436e-05,
+      "loss": 0.2629,
+      "step": 188
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 5.4329999999999997e-05,
+      "loss": 0.4152,
+      "step": 189
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 5.4300000000000005e-05,
+      "loss": 0.3362,
+      "step": 190
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 5.427e-05,
+      "loss": 0.6633,
+      "step": 191
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 5.424e-05,
+      "loss": 0.6759,
+      "step": 192
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 5.421e-05,
+      "loss": 0.9825,
+      "step": 193
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 5.418e-05,
+      "loss": 0.376,
+      "step": 194
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 5.415e-05,
+      "loss": 0.3648,
+      "step": 195
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 5.4120000000000004e-05,
+      "loss": 0.2648,
+      "step": 196
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 5.409e-05,
+      "loss": 0.6202,
+      "step": 197
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 5.406e-05,
+      "loss": 0.6997,
+      "step": 198
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 5.403e-05,
+      "loss": 0.4324,
+      "step": 199
+    },
+    {
+      "epoch": 5.0,
+      "learning_rate": 5.4000000000000005e-05,
+      "loss": 0.3945,
+      "step": 200
+    },
+    {
+      "epoch": 5.0,
+      "eval_loss": 0.27207431197166443,
+      "eval_mean_accuracy": 0.8545890809494763,
+      "eval_mean_iou": 0.5138796076397971,
+      "eval_overall_accuracy": 0.8242902050635074,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.7996985114000377,
+        0.9094796504989148
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.7903048161453703,
+        0.7513340067740211
+      ],
+      "eval_runtime": 48.403,
+      "eval_samples_per_second": 0.413,
+      "eval_steps_per_second": 0.207,
+      "step": 200
+    },
+    {
+      "epoch": 5.03,
+      "learning_rate": 5.397e-05,
+      "loss": 0.455,
+      "step": 201
+    },
+    {
+      "epoch": 5.05,
+      "learning_rate": 5.394e-05,
+      "loss": 0.2955,
+      "step": 202
+    },
+    {
+      "epoch": 5.08,
+      "learning_rate": 5.391e-05,
+      "loss": 0.1744,
+      "step": 203
+    },
+    {
+      "epoch": 5.1,
+      "learning_rate": 5.3880000000000006e-05,
+      "loss": 0.1787,
+      "step": 204
+    },
+    {
+      "epoch": 5.12,
+      "learning_rate": 5.385e-05,
+      "loss": 0.2821,
+      "step": 205
+    },
+    {
+      "epoch": 5.15,
+      "learning_rate": 5.382e-05,
+      "loss": 0.6153,
+      "step": 206
+    },
+    {
+      "epoch": 5.17,
+      "learning_rate": 5.379e-05,
+      "loss": 0.3641,
+      "step": 207
+    },
+    {
+      "epoch": 5.2,
+      "learning_rate": 5.376e-05,
+      "loss": 0.3918,
+      "step": 208
+    },
+    {
+      "epoch": 5.22,
+      "learning_rate": 5.373e-05,
+      "loss": 0.268,
+      "step": 209
+    },
+    {
+      "epoch": 5.25,
+      "learning_rate": 5.3700000000000004e-05,
+      "loss": 0.4003,
+      "step": 210
+    },
+    {
+      "epoch": 5.28,
+      "learning_rate": 5.367e-05,
+      "loss": 0.2636,
+      "step": 211
+    },
+    {
+      "epoch": 5.3,
+      "learning_rate": 5.364e-05,
+      "loss": 0.3127,
+      "step": 212
+    },
+    {
+      "epoch": 5.33,
+      "learning_rate": 5.361e-05,
+      "loss": 0.3199,
+      "step": 213
+    },
+    {
+      "epoch": 5.35,
+      "learning_rate": 5.3580000000000005e-05,
+      "loss": 0.4172,
+      "step": 214
+    },
+    {
+      "epoch": 5.38,
+      "learning_rate": 5.355e-05,
+      "loss": 1.0095,
+      "step": 215
+    },
+    {
+      "epoch": 5.4,
+      "learning_rate": 5.352e-05,
+      "loss": 1.1924,
+      "step": 216
+    },
+    {
+      "epoch": 5.42,
+      "learning_rate": 5.349e-05,
+      "loss": 0.2096,
+      "step": 217
+    },
+    {
+      "epoch": 5.45,
+      "learning_rate": 5.346e-05,
+      "loss": 0.1999,
+      "step": 218
+    },
+    {
+      "epoch": 5.47,
+      "learning_rate": 5.343e-05,
+      "loss": 0.4955,
+      "step": 219
+    },
+    {
+      "epoch": 5.5,
+      "learning_rate": 5.3400000000000004e-05,
+      "loss": 0.481,
+      "step": 220
+    },
+    {
+      "epoch": 5.5,
+      "eval_loss": 0.3197166323661804,
+      "eval_mean_accuracy": 0.8925474390283793,
+      "eval_mean_iou": 0.5617007834875217,
+      "eval_overall_accuracy": 0.933504608974328,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.966746970190107,
+        0.8183479078666516
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9219297160027523,
+        0.763172634459813
+      ],
+      "eval_runtime": 48.9978,
+      "eval_samples_per_second": 0.408,
+      "eval_steps_per_second": 0.204,
+      "step": 220
+    },
+    {
+      "epoch": 5.53,
+      "learning_rate": 5.337e-05,
+      "loss": 0.2962,
+      "step": 221
+    },
+    {
+      "epoch": 5.55,
+      "learning_rate": 5.334e-05,
+      "loss": 0.1479,
+      "step": 222
+    },
+    {
+      "epoch": 5.58,
+      "learning_rate": 5.3309999999999996e-05,
+      "loss": 0.2301,
+      "step": 223
+    },
+    {
+      "epoch": 5.6,
+      "learning_rate": 5.3280000000000005e-05,
+      "loss": 0.3265,
+      "step": 224
+    },
+    {
+      "epoch": 5.62,
+      "learning_rate": 5.325e-05,
+      "loss": 0.2636,
+      "step": 225
+    },
+    {
+      "epoch": 5.65,
+      "learning_rate": 5.322e-05,
+      "loss": 0.3857,
+      "step": 226
+    },
+    {
+      "epoch": 5.67,
+      "learning_rate": 5.319e-05,
+      "loss": 0.3996,
+      "step": 227
+    },
+    {
+      "epoch": 5.7,
+      "learning_rate": 5.316e-05,
+      "loss": 0.2629,
+      "step": 228
+    },
+    {
+      "epoch": 5.72,
+      "learning_rate": 5.313e-05,
+      "loss": 0.3125,
+      "step": 229
+    },
+    {
+      "epoch": 5.75,
+      "learning_rate": 5.31e-05,
+      "loss": 0.3026,
+      "step": 230
+    },
+    {
+      "epoch": 5.78,
+      "learning_rate": 5.307e-05,
+      "loss": 0.3461,
+      "step": 231
+    },
+    {
+      "epoch": 5.8,
+      "learning_rate": 5.304e-05,
+      "loss": 0.3292,
+      "step": 232
+    },
+    {
+      "epoch": 5.83,
+      "learning_rate": 5.3009999999999996e-05,
+      "loss": 0.8407,
+      "step": 233
+    },
+    {
+      "epoch": 5.85,
+      "learning_rate": 5.2980000000000004e-05,
+      "loss": 0.3062,
+      "step": 234
+    },
+    {
+      "epoch": 5.88,
+      "learning_rate": 5.295e-05,
+      "loss": 0.138,
+      "step": 235
+    },
+    {
+      "epoch": 5.9,
+      "learning_rate": 5.292e-05,
+      "loss": 0.169,
+      "step": 236
+    },
+    {
+      "epoch": 5.92,
+      "learning_rate": 5.289e-05,
+      "loss": 0.443,
+      "step": 237
+    },
+    {
+      "epoch": 5.95,
+      "learning_rate": 5.286e-05,
+      "loss": 0.3298,
+      "step": 238
+    },
+    {
+      "epoch": 5.97,
+      "learning_rate": 5.283e-05,
+      "loss": 0.3776,
+      "step": 239
+    },
+    {
+      "epoch": 6.0,
+      "learning_rate": 5.28e-05,
+      "loss": 0.3986,
+      "step": 240
+    },
+    {
+      "epoch": 6.0,
+      "eval_loss": 0.20506949722766876,
+      "eval_mean_accuracy": 0.9171030416015327,
+      "eval_mean_iou": 0.5573306667281388,
+      "eval_overall_accuracy": 0.9116858969091434,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.907289140959936,
+        0.9269169422431295
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.891243490977623,
+        0.7807485092067935
+      ],
+      "eval_runtime": 48.3589,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 240
+    },
+    {
+      "epoch": 6.03,
+      "learning_rate": 5.277e-05,
+      "loss": 0.9028,
+      "step": 241
+    },
+    {
+      "epoch": 6.05,
+      "learning_rate": 5.274e-05,
+      "loss": 0.2695,
+      "step": 242
+    },
+    {
+      "epoch": 6.08,
+      "learning_rate": 5.2709999999999995e-05,
+      "loss": 0.5083,
+      "step": 243
+    },
+    {
+      "epoch": 6.1,
+      "learning_rate": 5.2680000000000004e-05,
+      "loss": 0.2162,
+      "step": 244
+    },
+    {
+      "epoch": 6.12,
+      "learning_rate": 5.265e-05,
+      "loss": 0.4195,
+      "step": 245
+    },
+    {
+      "epoch": 6.15,
+      "learning_rate": 5.262e-05,
+      "loss": 0.6899,
+      "step": 246
+    },
+    {
+      "epoch": 6.17,
+      "learning_rate": 5.2589999999999996e-05,
+      "loss": 0.3309,
+      "step": 247
+    },
+    {
+      "epoch": 6.2,
+      "learning_rate": 5.256e-05,
+      "loss": 0.2429,
+      "step": 248
+    },
+    {
+      "epoch": 6.22,
+      "learning_rate": 5.253e-05,
+      "loss": 0.2289,
+      "step": 249
+    },
+    {
+      "epoch": 6.25,
+      "learning_rate": 5.25e-05,
+      "loss": 0.2147,
+      "step": 250
+    },
+    {
+      "epoch": 6.28,
+      "learning_rate": 5.2470000000000004e-05,
+      "loss": 0.2161,
+      "step": 251
+    },
+    {
+      "epoch": 6.3,
+      "learning_rate": 5.244e-05,
+      "loss": 0.3912,
+      "step": 252
+    },
+    {
+      "epoch": 6.33,
+      "learning_rate": 5.241e-05,
+      "loss": 0.654,
+      "step": 253
+    },
+    {
+      "epoch": 6.35,
+      "learning_rate": 5.238e-05,
+      "loss": 0.3199,
+      "step": 254
+    },
+    {
+      "epoch": 6.38,
+      "learning_rate": 5.2350000000000005e-05,
+      "loss": 0.2043,
+      "step": 255
+    },
+    {
+      "epoch": 6.4,
+      "learning_rate": 5.232e-05,
+      "loss": 0.1392,
+      "step": 256
+    },
+    {
+      "epoch": 6.42,
+      "learning_rate": 5.229e-05,
+      "loss": 0.1945,
+      "step": 257
+    },
+    {
+      "epoch": 6.45,
+      "learning_rate": 5.2260000000000004e-05,
+      "loss": 0.3461,
+      "step": 258
+    },
+    {
+      "epoch": 6.47,
+      "learning_rate": 5.2230000000000006e-05,
+      "loss": 0.4144,
+      "step": 259
+    },
+    {
+      "epoch": 6.5,
+      "learning_rate": 5.22e-05,
+      "loss": 0.282,
+      "step": 260
+    },
+    {
+      "epoch": 6.5,
+      "eval_loss": 0.388349711894989,
+      "eval_mean_accuracy": 0.8259209922494022,
+      "eval_mean_iou": 0.38122471501239064,
+      "eval_overall_accuracy": 0.7367579599669372,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.6643899294086868,
+        0.9874520550901176
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.662672740913094,
+        0.481001404124078
+      ],
+      "eval_runtime": 48.6412,
+      "eval_samples_per_second": 0.411,
+      "eval_steps_per_second": 0.206,
+      "step": 260
+    },
+    {
+      "epoch": 6.53,
+      "learning_rate": 5.2170000000000004e-05,
+      "loss": 0.3829,
+      "step": 261
+    },
+    {
+      "epoch": 6.55,
+      "learning_rate": 5.214e-05,
+      "loss": 0.3072,
+      "step": 262
+    },
+    {
+      "epoch": 6.58,
+      "learning_rate": 5.211000000000001e-05,
+      "loss": 0.321,
+      "step": 263
+    },
+    {
+      "epoch": 6.6,
+      "learning_rate": 5.208e-05,
+      "loss": 0.2191,
+      "step": 264
+    },
+    {
+      "epoch": 6.62,
+      "learning_rate": 5.2050000000000005e-05,
+      "loss": 0.3312,
+      "step": 265
+    },
+    {
+      "epoch": 6.65,
+      "learning_rate": 5.202e-05,
+      "loss": 0.7948,
+      "step": 266
+    },
+    {
+      "epoch": 6.67,
+      "learning_rate": 5.199e-05,
+      "loss": 0.3981,
+      "step": 267
+    },
+    {
+      "epoch": 6.7,
+      "learning_rate": 5.1960000000000004e-05,
+      "loss": 0.2188,
+      "step": 268
+    },
+    {
+      "epoch": 6.72,
+      "learning_rate": 5.1930000000000006e-05,
+      "loss": 0.3235,
+      "step": 269
+    },
+    {
+      "epoch": 6.75,
+      "learning_rate": 5.19e-05,
+      "loss": 0.1482,
+      "step": 270
+    },
+    {
+      "epoch": 6.78,
+      "learning_rate": 5.187e-05,
+      "loss": 0.1747,
+      "step": 271
+    },
+    {
+      "epoch": 6.8,
+      "learning_rate": 5.184e-05,
+      "loss": 0.2945,
+      "step": 272
+    },
+    {
+      "epoch": 6.83,
+      "learning_rate": 5.181000000000001e-05,
+      "loss": 0.3469,
+      "step": 273
+    },
+    {
+      "epoch": 6.85,
+      "learning_rate": 5.178e-05,
+      "loss": 0.5744,
+      "step": 274
+    },
+    {
+      "epoch": 6.88,
+      "learning_rate": 5.1750000000000004e-05,
+      "loss": 0.1551,
+      "step": 275
+    },
+    {
+      "epoch": 6.9,
+      "learning_rate": 5.172e-05,
+      "loss": 0.135,
+      "step": 276
+    },
+    {
+      "epoch": 6.92,
+      "learning_rate": 5.169e-05,
+      "loss": 1.0639,
+      "step": 277
+    },
+    {
+      "epoch": 6.95,
+      "learning_rate": 5.1660000000000003e-05,
+      "loss": 0.2964,
+      "step": 278
+    },
+    {
+      "epoch": 6.97,
+      "learning_rate": 5.1630000000000005e-05,
+      "loss": 0.2584,
+      "step": 279
+    },
+    {
+      "epoch": 7.0,
+      "learning_rate": 5.16e-05,
+      "loss": 0.3275,
+      "step": 280
+    },
+    {
+      "epoch": 7.0,
+      "eval_loss": 0.24394457042217255,
+      "eval_mean_accuracy": 0.8693783875093231,
+      "eval_mean_iou": 0.5201932822181216,
+      "eval_overall_accuracy": 0.8368297410563663,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.8104120499927068,
+        0.9283447250259392
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.7959839091242811,
+        0.7645959375300835
+      ],
+      "eval_runtime": 47.6324,
+      "eval_samples_per_second": 0.42,
+      "eval_steps_per_second": 0.21,
+      "step": 280
+    },
+    {
+      "epoch": 7.03,
+      "learning_rate": 5.157e-05,
+      "loss": 0.4328,
+      "step": 281
+    },
+    {
+      "epoch": 7.05,
+      "learning_rate": 5.154e-05,
+      "loss": 0.5671,
+      "step": 282
+    },
+    {
+      "epoch": 7.08,
+      "learning_rate": 5.1510000000000007e-05,
+      "loss": 0.3449,
+      "step": 283
+    },
+    {
+      "epoch": 7.1,
+      "learning_rate": 5.148e-05,
+      "loss": 0.3505,
+      "step": 284
+    },
+    {
+      "epoch": 7.12,
+      "learning_rate": 5.1450000000000004e-05,
+      "loss": 0.7337,
+      "step": 285
+    },
+    {
+      "epoch": 7.15,
+      "learning_rate": 5.142e-05,
+      "loss": 0.2992,
+      "step": 286
+    },
+    {
+      "epoch": 7.17,
+      "learning_rate": 5.139e-05,
+      "loss": 0.2461,
+      "step": 287
+    },
+    {
+      "epoch": 7.2,
+      "learning_rate": 5.136e-05,
+      "loss": 0.1222,
+      "step": 288
+    },
+    {
+      "epoch": 7.22,
+      "learning_rate": 5.1330000000000005e-05,
+      "loss": 0.4672,
+      "step": 289
+    },
+    {
+      "epoch": 7.25,
+      "learning_rate": 5.13e-05,
+      "loss": 0.3195,
+      "step": 290
+    },
+    {
+      "epoch": 7.28,
+      "learning_rate": 5.127e-05,
+      "loss": 0.3162,
+      "step": 291
+    },
+    {
+      "epoch": 7.3,
+      "learning_rate": 5.124e-05,
+      "loss": 0.4218,
+      "step": 292
+    },
+    {
+      "epoch": 7.33,
+      "learning_rate": 5.1210000000000006e-05,
+      "loss": 0.2803,
+      "step": 293
+    },
+    {
+      "epoch": 7.35,
+      "learning_rate": 5.118e-05,
+      "loss": 0.3924,
+      "step": 294
+    },
+    {
+      "epoch": 7.38,
+      "learning_rate": 5.115e-05,
+      "loss": 0.5085,
+      "step": 295
+    },
+    {
+      "epoch": 7.4,
+      "learning_rate": 5.112e-05,
+      "loss": 0.1882,
+      "step": 296
+    },
+    {
+      "epoch": 7.42,
+      "learning_rate": 5.109e-05,
+      "loss": 0.4084,
+      "step": 297
+    },
+    {
+      "epoch": 7.45,
+      "learning_rate": 5.106e-05,
+      "loss": 0.3493,
+      "step": 298
+    },
+    {
+      "epoch": 7.47,
+      "learning_rate": 5.1030000000000004e-05,
+      "loss": 0.2055,
+      "step": 299
+    },
+    {
+      "epoch": 7.5,
+      "learning_rate": 5.1e-05,
+      "loss": 0.3243,
+      "step": 300
+    },
+    {
+      "epoch": 7.5,
+      "eval_loss": 0.2048894464969635,
+      "eval_mean_accuracy": 0.8435106512197734,
+      "eval_mean_iou": 0.5263646497051458,
+      "eval_overall_accuracy": 0.8976785141717301,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9416431647789888,
+        0.7453781376605582
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8818220751562738,
+        0.6972718739591639
+      ],
+      "eval_runtime": 48.7394,
+      "eval_samples_per_second": 0.41,
+      "eval_steps_per_second": 0.205,
+      "step": 300
+    },
+    {
+      "epoch": 7.53,
+      "learning_rate": 5.097e-05,
+      "loss": 0.135,
+      "step": 301
+    },
+    {
+      "epoch": 7.55,
+      "learning_rate": 5.094e-05,
+      "loss": 0.2282,
+      "step": 302
+    },
+    {
+      "epoch": 7.58,
+      "learning_rate": 5.0910000000000006e-05,
+      "loss": 0.4108,
+      "step": 303
+    },
+    {
+      "epoch": 7.6,
+      "learning_rate": 5.088e-05,
+      "loss": 0.2306,
+      "step": 304
+    },
+    {
+      "epoch": 7.62,
+      "learning_rate": 5.085e-05,
+      "loss": 0.2897,
+      "step": 305
+    },
+    {
+      "epoch": 7.65,
+      "learning_rate": 5.082e-05,
+      "loss": 0.9242,
+      "step": 306
+    },
+    {
+      "epoch": 7.67,
+      "learning_rate": 5.079e-05,
+      "loss": 0.2076,
+      "step": 307
+    },
+    {
+      "epoch": 7.7,
+      "learning_rate": 5.076e-05,
+      "loss": 0.5449,
+      "step": 308
+    },
+    {
+      "epoch": 7.72,
+      "learning_rate": 5.0730000000000004e-05,
+      "loss": 0.4291,
+      "step": 309
+    },
+    {
+      "epoch": 7.75,
+      "learning_rate": 5.07e-05,
+      "loss": 0.4141,
+      "step": 310
+    },
+    {
+      "epoch": 7.78,
+      "learning_rate": 5.067e-05,
+      "loss": 0.4648,
+      "step": 311
+    },
+    {
+      "epoch": 7.8,
+      "learning_rate": 5.0639999999999996e-05,
+      "loss": 0.8204,
+      "step": 312
+    },
+    {
+      "epoch": 7.83,
+      "learning_rate": 5.0610000000000005e-05,
+      "loss": 0.403,
+      "step": 313
+    },
+    {
+      "epoch": 7.85,
+      "learning_rate": 5.058e-05,
+      "loss": 0.3946,
+      "step": 314
+    },
+    {
+      "epoch": 7.88,
+      "learning_rate": 5.055e-05,
+      "loss": 0.2797,
+      "step": 315
+    },
+    {
+      "epoch": 7.9,
+      "learning_rate": 5.052e-05,
+      "loss": 0.2285,
+      "step": 316
+    },
+    {
+      "epoch": 7.92,
+      "learning_rate": 5.0490000000000006e-05,
+      "loss": 0.3405,
+      "step": 317
+    },
+    {
+      "epoch": 7.95,
+      "learning_rate": 5.046e-05,
+      "loss": 0.1786,
+      "step": 318
+    },
+    {
+      "epoch": 7.97,
+      "learning_rate": 5.043e-05,
+      "loss": 0.1665,
+      "step": 319
+    },
+    {
+      "epoch": 8.0,
+      "learning_rate": 5.04e-05,
+      "loss": 0.2753,
+      "step": 320
+    },
+    {
+      "epoch": 8.0,
+      "eval_loss": 0.22952759265899658,
+      "eval_mean_accuracy": 0.9036801208276177,
+      "eval_mean_iou": 0.5711426371831221,
+      "eval_overall_accuracy": 0.9283498244545524,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9483726727161642,
+        0.8589875689390711
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9135066161058577,
+        0.7999212954435085
+      ],
+      "eval_runtime": 48.1711,
+      "eval_samples_per_second": 0.415,
+      "eval_steps_per_second": 0.208,
+      "step": 320
+    },
+    {
+      "epoch": 8.03,
+      "learning_rate": 5.037e-05,
+      "loss": 0.1384,
+      "step": 321
+    },
+    {
+      "epoch": 8.05,
+      "learning_rate": 5.034e-05,
+      "loss": 0.3371,
+      "step": 322
+    },
+    {
+      "epoch": 8.07,
+      "learning_rate": 5.0310000000000005e-05,
+      "loss": 0.1202,
+      "step": 323
+    },
+    {
+      "epoch": 8.1,
+      "learning_rate": 5.028e-05,
+      "loss": 0.145,
+      "step": 324
+    },
+    {
+      "epoch": 8.12,
+      "learning_rate": 5.025e-05,
+      "loss": 0.2253,
+      "step": 325
+    },
+    {
+      "epoch": 8.15,
+      "learning_rate": 5.022e-05,
+      "loss": 0.2934,
+      "step": 326
+    },
+    {
+      "epoch": 8.18,
+      "learning_rate": 5.0190000000000006e-05,
+      "loss": 0.258,
+      "step": 327
+    },
+    {
+      "epoch": 8.2,
+      "learning_rate": 5.016e-05,
+      "loss": 0.5661,
+      "step": 328
+    },
+    {
+      "epoch": 8.22,
+      "learning_rate": 5.013e-05,
+      "loss": 0.1907,
+      "step": 329
+    },
+    {
+      "epoch": 8.25,
+      "learning_rate": 5.01e-05,
+      "loss": 1.0441,
+      "step": 330
+    },
+    {
+      "epoch": 8.28,
+      "learning_rate": 5.007e-05,
+      "loss": 0.2594,
+      "step": 331
+    },
+    {
+      "epoch": 8.3,
+      "learning_rate": 5.004e-05,
+      "loss": 0.1567,
+      "step": 332
+    },
+    {
+      "epoch": 8.32,
+      "learning_rate": 5.0010000000000004e-05,
+      "loss": 0.1027,
+      "step": 333
+    },
+    {
+      "epoch": 8.35,
+      "learning_rate": 4.998e-05,
+      "loss": 0.7182,
+      "step": 334
+    },
+    {
+      "epoch": 8.38,
+      "learning_rate": 4.995e-05,
+      "loss": 0.2658,
+      "step": 335
+    },
+    {
+      "epoch": 8.4,
+      "learning_rate": 4.9919999999999996e-05,
+      "loss": 0.2252,
+      "step": 336
+    },
+    {
+      "epoch": 8.43,
+      "learning_rate": 4.9890000000000005e-05,
+      "loss": 0.3155,
+      "step": 337
+    },
+    {
+      "epoch": 8.45,
+      "learning_rate": 4.986e-05,
+      "loss": 0.1659,
+      "step": 338
+    },
+    {
+      "epoch": 8.47,
+      "learning_rate": 4.983e-05,
+      "loss": 1.1028,
+      "step": 339
+    },
+    {
+      "epoch": 8.5,
+      "learning_rate": 4.98e-05,
+      "loss": 0.2368,
+      "step": 340
+    },
+    {
+      "epoch": 8.5,
+      "eval_loss": 0.19084516167640686,
+      "eval_mean_accuracy": 0.9140556027707284,
+      "eval_mean_iou": 0.5612282269733403,
+      "eval_overall_accuracy": 0.9094506934004901,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9057131778237694,
+        0.9223980277176874
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8891001341699122,
+        0.7945845467501085
+      ],
+      "eval_runtime": 48.1614,
+      "eval_samples_per_second": 0.415,
+      "eval_steps_per_second": 0.208,
+      "step": 340
+    },
+    {
+      "epoch": 8.53,
+      "learning_rate": 4.977e-05,
+      "loss": 0.2577,
+      "step": 341
+    },
+    {
+      "epoch": 8.55,
+      "learning_rate": 4.974e-05,
+      "loss": 0.3297,
+      "step": 342
+    },
+    {
+      "epoch": 8.57,
+      "learning_rate": 4.9710000000000003e-05,
+      "loss": 0.313,
+      "step": 343
+    },
+    {
+      "epoch": 8.6,
+      "learning_rate": 4.968e-05,
+      "loss": 0.2803,
+      "step": 344
+    },
+    {
+      "epoch": 8.62,
+      "learning_rate": 4.965e-05,
+      "loss": 0.2727,
+      "step": 345
+    },
+    {
+      "epoch": 8.65,
+      "learning_rate": 4.9619999999999996e-05,
+      "loss": 0.5579,
+      "step": 346
+    },
+    {
+      "epoch": 8.68,
+      "learning_rate": 4.9590000000000005e-05,
+      "loss": 0.276,
+      "step": 347
+    },
+    {
+      "epoch": 8.7,
+      "learning_rate": 4.956e-05,
+      "loss": 0.3912,
+      "step": 348
+    },
+    {
+      "epoch": 8.72,
+      "learning_rate": 4.953e-05,
+      "loss": 0.24,
+      "step": 349
+    },
+    {
+      "epoch": 8.75,
+      "learning_rate": 4.95e-05,
+      "loss": 0.4231,
+      "step": 350
+    },
+    {
+      "epoch": 8.78,
+      "learning_rate": 4.947e-05,
+      "loss": 0.3451,
+      "step": 351
+    },
+    {
+      "epoch": 8.8,
+      "learning_rate": 4.944e-05,
+      "loss": 0.3634,
+      "step": 352
+    },
+    {
+      "epoch": 8.82,
+      "learning_rate": 4.941e-05,
+      "loss": 0.5014,
+      "step": 353
+    },
+    {
+      "epoch": 8.85,
+      "learning_rate": 4.938e-05,
+      "loss": 0.2758,
+      "step": 354
+    },
+    {
+      "epoch": 8.88,
+      "learning_rate": 4.935e-05,
+      "loss": 0.3193,
+      "step": 355
+    },
+    {
+      "epoch": 8.9,
+      "learning_rate": 4.9319999999999995e-05,
+      "loss": 0.3801,
+      "step": 356
+    },
+    {
+      "epoch": 8.93,
+      "learning_rate": 4.9290000000000004e-05,
+      "loss": 0.2007,
+      "step": 357
+    },
+    {
+      "epoch": 8.95,
+      "learning_rate": 4.926e-05,
+      "loss": 0.2294,
+      "step": 358
+    },
+    {
+      "epoch": 8.97,
+      "learning_rate": 4.923e-05,
+      "loss": 0.3337,
+      "step": 359
+    },
+    {
+      "epoch": 9.0,
+      "learning_rate": 4.9199999999999997e-05,
+      "loss": 0.9901,
+      "step": 360
+    },
+    {
+      "epoch": 9.0,
+      "eval_loss": 0.24918393790721893,
+      "eval_mean_accuracy": 0.8951753051498105,
+      "eval_mean_iou": 0.4812836580752557,
+      "eval_overall_accuracy": 0.8465435512223731,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.8070722124373391,
+        0.9832783978622819
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.803806401181222,
+        0.640044573044545
+      ],
+      "eval_runtime": 49.5136,
+      "eval_samples_per_second": 0.404,
+      "eval_steps_per_second": 0.202,
+      "step": 360
+    },
+    {
+      "epoch": 9.03,
+      "learning_rate": 4.917e-05,
+      "loss": 0.1101,
+      "step": 361
+    },
+    {
+      "epoch": 9.05,
+      "learning_rate": 4.914e-05,
+      "loss": 0.1172,
+      "step": 362
+    },
+    {
+      "epoch": 9.07,
+      "learning_rate": 4.911e-05,
+      "loss": 0.1172,
+      "step": 363
+    },
+    {
+      "epoch": 9.1,
+      "learning_rate": 4.908e-05,
+      "loss": 0.1692,
+      "step": 364
+    },
+    {
+      "epoch": 9.12,
+      "learning_rate": 4.905e-05,
+      "loss": 0.1617,
+      "step": 365
+    },
+    {
+      "epoch": 9.15,
+      "learning_rate": 4.9019999999999995e-05,
+      "loss": 0.4696,
+      "step": 366
+    },
+    {
+      "epoch": 9.18,
+      "learning_rate": 4.8990000000000004e-05,
+      "loss": 0.3077,
+      "step": 367
+    },
+    {
+      "epoch": 9.2,
+      "learning_rate": 4.896e-05,
+      "loss": 0.3329,
+      "step": 368
+    },
+    {
+      "epoch": 9.22,
+      "learning_rate": 4.893e-05,
+      "loss": 0.6601,
+      "step": 369
+    },
+    {
+      "epoch": 9.25,
+      "learning_rate": 4.8899999999999996e-05,
+      "loss": 0.4652,
+      "step": 370
+    },
+    {
+      "epoch": 9.28,
+      "learning_rate": 4.887e-05,
+      "loss": 0.1901,
+      "step": 371
+    },
+    {
+      "epoch": 9.3,
+      "learning_rate": 4.884e-05,
+      "loss": 0.1811,
+      "step": 372
+    },
+    {
+      "epoch": 9.32,
+      "learning_rate": 4.881e-05,
+      "loss": 0.3049,
+      "step": 373
+    },
+    {
+      "epoch": 9.35,
+      "learning_rate": 4.878e-05,
+      "loss": 0.2756,
+      "step": 374
+    },
+    {
+      "epoch": 9.38,
+      "learning_rate": 4.875e-05,
+      "loss": 0.4155,
+      "step": 375
+    },
+    {
+      "epoch": 9.4,
+      "learning_rate": 4.872000000000001e-05,
+      "loss": 0.3104,
+      "step": 376
+    },
+    {
+      "epoch": 9.43,
+      "learning_rate": 4.869e-05,
+      "loss": 0.1151,
+      "step": 377
+    },
+    {
+      "epoch": 9.45,
+      "learning_rate": 4.8660000000000005e-05,
+      "loss": 0.2135,
+      "step": 378
+    },
+    {
+      "epoch": 9.47,
+      "learning_rate": 4.863e-05,
+      "loss": 0.5245,
+      "step": 379
+    },
+    {
+      "epoch": 9.5,
+      "learning_rate": 4.86e-05,
+      "loss": 0.2519,
+      "step": 380
+    },
+    {
+      "epoch": 9.5,
+      "eval_loss": 0.23749105632305145,
+      "eval_mean_accuracy": 0.9468136114330645,
+      "eval_mean_iou": 0.581027164032012,
+      "eval_overall_accuracy": 0.9390303832997389,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.93271322609932,
+        0.9609139967668092
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9236989144275697,
+        0.8193825776684663
+      ],
+      "eval_runtime": 48.4577,
+      "eval_samples_per_second": 0.413,
+      "eval_steps_per_second": 0.206,
+      "step": 380
+    },
+    {
+      "epoch": 9.53,
+      "learning_rate": 4.8570000000000004e-05,
+      "loss": 0.7302,
+      "step": 381
+    },
+    {
+      "epoch": 9.55,
+      "learning_rate": 4.8540000000000006e-05,
+      "loss": 0.2297,
+      "step": 382
+    },
+    {
+      "epoch": 9.57,
+      "learning_rate": 4.851e-05,
+      "loss": 0.1605,
+      "step": 383
+    },
+    {
+      "epoch": 9.6,
+      "learning_rate": 4.8480000000000003e-05,
+      "loss": 0.2559,
+      "step": 384
+    },
+    {
+      "epoch": 9.62,
+      "learning_rate": 4.845e-05,
+      "loss": 0.2862,
+      "step": 385
+    },
+    {
+      "epoch": 9.65,
+      "learning_rate": 4.842000000000001e-05,
+      "loss": 0.2086,
+      "step": 386
+    },
+    {
+      "epoch": 9.68,
+      "learning_rate": 4.839e-05,
+      "loss": 0.2762,
+      "step": 387
+    },
+    {
+      "epoch": 9.7,
+      "learning_rate": 4.8360000000000005e-05,
+      "loss": 0.3488,
+      "step": 388
+    },
+    {
+      "epoch": 9.72,
+      "learning_rate": 4.833e-05,
+      "loss": 0.2638,
+      "step": 389
+    },
+    {
+      "epoch": 9.75,
+      "learning_rate": 4.83e-05,
+      "loss": 0.1605,
+      "step": 390
+    },
+    {
+      "epoch": 9.78,
+      "learning_rate": 4.8270000000000004e-05,
+      "loss": 0.2095,
+      "step": 391
+    },
+    {
+      "epoch": 9.8,
+      "learning_rate": 4.8240000000000006e-05,
+      "loss": 0.5525,
+      "step": 392
+    },
+    {
+      "epoch": 9.82,
+      "learning_rate": 4.821e-05,
+      "loss": 0.3645,
+      "step": 393
+    },
+    {
+      "epoch": 9.85,
+      "learning_rate": 4.818e-05,
+      "loss": 0.119,
+      "step": 394
+    },
+    {
+      "epoch": 9.88,
+      "learning_rate": 4.815e-05,
+      "loss": 0.2868,
+      "step": 395
+    },
+    {
+      "epoch": 9.9,
+      "learning_rate": 4.812000000000001e-05,
+      "loss": 0.8835,
+      "step": 396
+    },
+    {
+      "epoch": 9.93,
+      "learning_rate": 4.809e-05,
+      "loss": 0.122,
+      "step": 397
+    },
+    {
+      "epoch": 9.95,
+      "learning_rate": 4.8060000000000004e-05,
+      "loss": 0.1681,
+      "step": 398
+    },
+    {
+      "epoch": 9.97,
+      "learning_rate": 4.803e-05,
+      "loss": 0.281,
+      "step": 399
+    },
+    {
+      "epoch": 10.0,
+      "learning_rate": 4.8e-05,
+      "loss": 0.4478,
+      "step": 400
+    },
+    {
+      "epoch": 10.0,
+      "eval_loss": 0.19183433055877686,
+      "eval_mean_accuracy": 0.8839528075533328,
+      "eval_mean_iou": 0.5615844872126089,
+      "eval_overall_accuracy": 0.9199415278173876,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9491513106292899,
+        0.8187543044773757
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9064106123096447,
+        0.7783428493281822
+      ],
+      "eval_runtime": 49.2471,
+      "eval_samples_per_second": 0.406,
+      "eval_steps_per_second": 0.203,
+      "step": 400
+    },
+    {
+      "epoch": 10.03,
+      "learning_rate": 4.797e-05,
+      "loss": 0.2364,
+      "step": 401
+    },
+    {
+      "epoch": 10.05,
+      "learning_rate": 4.7940000000000005e-05,
+      "loss": 0.186,
+      "step": 402
+    },
+    {
+      "epoch": 10.07,
+      "learning_rate": 4.791e-05,
+      "loss": 0.5067,
+      "step": 403
+    },
+    {
+      "epoch": 10.1,
+      "learning_rate": 4.788e-05,
+      "loss": 0.2468,
+      "step": 404
+    },
+    {
+      "epoch": 10.12,
+      "learning_rate": 4.785e-05,
+      "loss": 0.2668,
+      "step": 405
+    },
+    {
+      "epoch": 10.15,
+      "learning_rate": 4.7820000000000006e-05,
+      "loss": 0.2371,
+      "step": 406
+    },
+    {
+      "epoch": 10.18,
+      "learning_rate": 4.779e-05,
+      "loss": 0.3081,
+      "step": 407
+    },
+    {
+      "epoch": 10.2,
+      "learning_rate": 4.7760000000000004e-05,
+      "loss": 0.1102,
+      "step": 408
+    },
+    {
+      "epoch": 10.22,
+      "learning_rate": 4.773e-05,
+      "loss": 0.4816,
+      "step": 409
+    },
+    {
+      "epoch": 10.25,
+      "learning_rate": 4.77e-05,
+      "loss": 0.2743,
+      "step": 410
+    },
+    {
+      "epoch": 10.28,
+      "learning_rate": 4.767e-05,
+      "loss": 0.3725,
+      "step": 411
+    },
+    {
+      "epoch": 10.3,
+      "learning_rate": 4.7640000000000005e-05,
+      "loss": 0.2469,
+      "step": 412
+    },
+    {
+      "epoch": 10.32,
+      "learning_rate": 4.761e-05,
+      "loss": 0.2175,
+      "step": 413
+    },
+    {
+      "epoch": 10.35,
+      "learning_rate": 4.758e-05,
+      "loss": 0.3623,
+      "step": 414
+    },
+    {
+      "epoch": 10.38,
+      "learning_rate": 4.755e-05,
+      "loss": 0.3651,
+      "step": 415
+    },
+    {
+      "epoch": 10.4,
+      "learning_rate": 4.7520000000000006e-05,
+      "loss": 0.2813,
+      "step": 416
+    },
+    {
+      "epoch": 10.43,
+      "learning_rate": 4.749e-05,
+      "loss": 0.1988,
+      "step": 417
+    },
+    {
+      "epoch": 10.45,
+      "learning_rate": 4.746e-05,
+      "loss": 0.0948,
+      "step": 418
+    },
+    {
+      "epoch": 10.47,
+      "learning_rate": 4.743e-05,
+      "loss": 0.1196,
+      "step": 419
+    },
+    {
+      "epoch": 10.5,
+      "learning_rate": 4.74e-05,
+      "loss": 0.2322,
+      "step": 420
+    },
+    {
+      "epoch": 10.5,
+      "eval_loss": 0.18487034738063812,
+      "eval_mean_accuracy": 0.9393724907811466,
+      "eval_mean_iou": 0.5821265257663112,
+      "eval_overall_accuracy": 0.9355605611653833,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9324666574268301,
+        0.9462783241354631
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9208515742743779,
+        0.8255280030245556
+      ],
+      "eval_runtime": 48.2088,
+      "eval_samples_per_second": 0.415,
+      "eval_steps_per_second": 0.207,
+      "step": 420
+    },
+    {
+      "epoch": 10.53,
+      "learning_rate": 4.737e-05,
+      "loss": 0.1669,
+      "step": 421
+    },
+    {
+      "epoch": 10.55,
+      "learning_rate": 4.7340000000000004e-05,
+      "loss": 0.6335,
+      "step": 422
+    },
+    {
+      "epoch": 10.57,
+      "learning_rate": 4.731e-05,
+      "loss": 0.1626,
+      "step": 423
+    },
+    {
+      "epoch": 10.6,
+      "learning_rate": 4.728e-05,
+      "loss": 0.5899,
+      "step": 424
+    },
+    {
+      "epoch": 10.62,
+      "learning_rate": 4.7249999999999997e-05,
+      "loss": 0.236,
+      "step": 425
+    },
+    {
+      "epoch": 10.65,
+      "learning_rate": 4.7220000000000005e-05,
+      "loss": 0.2995,
+      "step": 426
+    },
+    {
+      "epoch": 10.68,
+      "learning_rate": 4.719e-05,
+      "loss": 0.232,
+      "step": 427
+    },
+    {
+      "epoch": 10.7,
+      "learning_rate": 4.716e-05,
+      "loss": 0.1955,
+      "step": 428
+    },
+    {
+      "epoch": 10.72,
+      "learning_rate": 4.713e-05,
+      "loss": 0.2382,
+      "step": 429
+    },
+    {
+      "epoch": 10.75,
+      "learning_rate": 4.7100000000000006e-05,
+      "loss": 0.1718,
+      "step": 430
+    },
+    {
+      "epoch": 10.78,
+      "learning_rate": 4.707e-05,
+      "loss": 0.1884,
+      "step": 431
+    },
+    {
+      "epoch": 10.8,
+      "learning_rate": 4.7040000000000004e-05,
+      "loss": 1.017,
+      "step": 432
+    },
+    {
+      "epoch": 10.82,
+      "learning_rate": 4.701e-05,
+      "loss": 0.1411,
+      "step": 433
+    },
+    {
+      "epoch": 10.85,
+      "learning_rate": 4.698e-05,
+      "loss": 0.2267,
+      "step": 434
+    },
+    {
+      "epoch": 10.88,
+      "learning_rate": 4.695e-05,
+      "loss": 0.2418,
+      "step": 435
+    },
+    {
+      "epoch": 10.9,
+      "learning_rate": 4.6920000000000005e-05,
+      "loss": 0.3928,
+      "step": 436
+    },
+    {
+      "epoch": 10.93,
+      "learning_rate": 4.689e-05,
+      "loss": 0.2357,
+      "step": 437
+    },
+    {
+      "epoch": 10.95,
+      "learning_rate": 4.686e-05,
+      "loss": 0.1766,
+      "step": 438
+    },
+    {
+      "epoch": 10.97,
+      "learning_rate": 4.683e-05,
+      "loss": 0.2089,
+      "step": 439
+    },
+    {
+      "epoch": 11.0,
+      "learning_rate": 4.6800000000000006e-05,
+      "loss": 0.1698,
+      "step": 440
+    },
+    {
+      "epoch": 11.0,
+      "eval_loss": 0.19902649521827698,
+      "eval_mean_accuracy": 0.9215499899486547,
+      "eval_mean_iou": 0.5553175130512749,
+      "eval_overall_accuracy": 0.8984438568760805,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.8796900605832205,
+        0.9634099193140888
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8726210024783129,
+        0.7933315366755117
+      ],
+      "eval_runtime": 49.031,
+      "eval_samples_per_second": 0.408,
+      "eval_steps_per_second": 0.204,
+      "step": 440
+    },
+    {
+      "epoch": 11.03,
+      "learning_rate": 4.677e-05,
+      "loss": 0.533,
+      "step": 441
+    },
+    {
+      "epoch": 11.05,
+      "learning_rate": 4.674e-05,
+      "loss": 0.6826,
+      "step": 442
+    },
+    {
+      "epoch": 11.07,
+      "learning_rate": 4.671e-05,
+      "loss": 0.1957,
+      "step": 443
+    },
+    {
+      "epoch": 11.1,
+      "learning_rate": 4.668e-05,
+      "loss": 0.1768,
+      "step": 444
+    },
+    {
+      "epoch": 11.12,
+      "learning_rate": 4.665e-05,
+      "loss": 0.6004,
+      "step": 445
+    },
+    {
+      "epoch": 11.15,
+      "learning_rate": 4.6620000000000004e-05,
+      "loss": 0.314,
+      "step": 446
+    },
+    {
+      "epoch": 11.18,
+      "learning_rate": 4.659e-05,
+      "loss": 1.1865,
+      "step": 447
+    },
+    {
+      "epoch": 11.2,
+      "learning_rate": 4.656e-05,
+      "loss": 0.0892,
+      "step": 448
+    },
+    {
+      "epoch": 11.22,
+      "learning_rate": 4.653e-05,
+      "loss": 0.1509,
+      "step": 449
+    },
+    {
+      "epoch": 11.25,
+      "learning_rate": 4.6500000000000005e-05,
+      "loss": 0.3381,
+      "step": 450
+    },
+    {
+      "epoch": 11.28,
+      "learning_rate": 4.647e-05,
+      "loss": 1.3277,
+      "step": 451
+    },
+    {
+      "epoch": 11.3,
+      "learning_rate": 4.644e-05,
+      "loss": 0.1961,
+      "step": 452
+    },
+    {
+      "epoch": 11.32,
+      "learning_rate": 4.641e-05,
+      "loss": 0.2051,
+      "step": 453
+    },
+    {
+      "epoch": 11.35,
+      "learning_rate": 4.638e-05,
+      "loss": 0.2301,
+      "step": 454
+    },
+    {
+      "epoch": 11.38,
+      "learning_rate": 4.635e-05,
+      "loss": 0.2205,
+      "step": 455
+    },
+    {
+      "epoch": 11.4,
+      "learning_rate": 4.6320000000000004e-05,
+      "loss": 0.1577,
+      "step": 456
+    },
+    {
+      "epoch": 11.43,
+      "learning_rate": 4.629e-05,
+      "loss": 0.264,
+      "step": 457
+    },
+    {
+      "epoch": 11.45,
+      "learning_rate": 4.626e-05,
+      "loss": 0.223,
+      "step": 458
+    },
+    {
+      "epoch": 11.47,
+      "learning_rate": 4.6229999999999996e-05,
+      "loss": 0.3168,
+      "step": 459
+    },
+    {
+      "epoch": 11.5,
+      "learning_rate": 4.6200000000000005e-05,
+      "loss": 0.2747,
+      "step": 460
+    },
+    {
+      "epoch": 11.5,
+      "eval_loss": 0.1631600707769394,
+      "eval_mean_accuracy": 0.9303702950626429,
+      "eval_mean_iou": 0.5869421148668948,
+      "eval_overall_accuracy": 0.9340532388497624,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9370424528962994,
+        0.9236981372289865
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9195581410749489,
+        0.8412682035257355
+      ],
+      "eval_runtime": 48.2511,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 460
+    },
+    {
+      "epoch": 11.53,
+      "learning_rate": 4.617e-05,
+      "loss": 0.2376,
+      "step": 461
+    },
+    {
+      "epoch": 11.55,
+      "learning_rate": 4.614e-05,
+      "loss": 0.3034,
+      "step": 462
+    },
+    {
+      "epoch": 11.57,
+      "learning_rate": 4.611e-05,
+      "loss": 0.0951,
+      "step": 463
+    },
+    {
+      "epoch": 11.6,
+      "learning_rate": 4.608e-05,
+      "loss": 0.3745,
+      "step": 464
+    },
+    {
+      "epoch": 11.62,
+      "learning_rate": 4.605e-05,
+      "loss": 0.1535,
+      "step": 465
+    },
+    {
+      "epoch": 11.65,
+      "learning_rate": 4.602e-05,
+      "loss": 0.2144,
+      "step": 466
+    },
+    {
+      "epoch": 11.68,
+      "learning_rate": 4.599e-05,
+      "loss": 0.2269,
+      "step": 467
+    },
+    {
+      "epoch": 11.7,
+      "learning_rate": 4.596e-05,
+      "loss": 0.2681,
+      "step": 468
+    },
+    {
+      "epoch": 11.72,
+      "learning_rate": 4.5929999999999996e-05,
+      "loss": 0.2469,
+      "step": 469
+    },
+    {
+      "epoch": 11.75,
+      "learning_rate": 4.5900000000000004e-05,
+      "loss": 0.1947,
+      "step": 470
+    },
+    {
+      "epoch": 11.78,
+      "learning_rate": 4.587e-05,
+      "loss": 0.2222,
+      "step": 471
+    },
+    {
+      "epoch": 11.8,
+      "learning_rate": 4.584e-05,
+      "loss": 0.2049,
+      "step": 472
+    },
+    {
+      "epoch": 11.82,
+      "learning_rate": 4.581e-05,
+      "loss": 0.1804,
+      "step": 473
+    },
+    {
+      "epoch": 11.85,
+      "learning_rate": 4.578e-05,
+      "loss": 0.1954,
+      "step": 474
+    },
+    {
+      "epoch": 11.88,
+      "learning_rate": 4.575e-05,
+      "loss": 0.2368,
+      "step": 475
+    },
+    {
+      "epoch": 11.9,
+      "learning_rate": 4.572e-05,
+      "loss": 0.3448,
+      "step": 476
+    },
+    {
+      "epoch": 11.93,
+      "learning_rate": 4.569e-05,
+      "loss": 0.3405,
+      "step": 477
+    },
+    {
+      "epoch": 11.95,
+      "learning_rate": 4.566e-05,
+      "loss": 0.1396,
+      "step": 478
+    },
+    {
+      "epoch": 11.97,
+      "learning_rate": 4.5629999999999995e-05,
+      "loss": 0.2008,
+      "step": 479
+    },
+    {
+      "epoch": 12.0,
+      "learning_rate": 4.5600000000000004e-05,
+      "loss": 0.4617,
+      "step": 480
+    },
+    {
+      "epoch": 12.0,
+      "eval_loss": 0.19759610295295715,
+      "eval_mean_accuracy": 0.9295030231770169,
+      "eval_mean_iou": 0.5369163642640026,
+      "eval_overall_accuracy": 0.8968608059139239,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.8703671693033942,
+        0.9886388770506396
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8686163566323581,
+        0.7421327361596495
+      ],
+      "eval_runtime": 48.2743,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 480
+    },
+    {
+      "epoch": 12.03,
+      "learning_rate": 4.557e-05,
+      "loss": 0.172,
+      "step": 481
+    },
+    {
+      "epoch": 12.05,
+      "learning_rate": 4.554e-05,
+      "loss": 0.1316,
+      "step": 482
+    },
+    {
+      "epoch": 12.07,
+      "learning_rate": 4.5509999999999996e-05,
+      "loss": 0.2638,
+      "step": 483
+    },
+    {
+      "epoch": 12.1,
+      "learning_rate": 4.548e-05,
+      "loss": 0.4121,
+      "step": 484
+    },
+    {
+      "epoch": 12.12,
+      "learning_rate": 4.545e-05,
+      "loss": 0.9736,
+      "step": 485
+    },
+    {
+      "epoch": 12.15,
+      "learning_rate": 4.542e-05,
+      "loss": 0.2648,
+      "step": 486
+    },
+    {
+      "epoch": 12.18,
+      "learning_rate": 4.539e-05,
+      "loss": 0.2924,
+      "step": 487
+    },
+    {
+      "epoch": 12.2,
+      "learning_rate": 4.536e-05,
+      "loss": 0.0925,
+      "step": 488
+    },
+    {
+      "epoch": 12.22,
+      "learning_rate": 4.533e-05,
+      "loss": 0.3587,
+      "step": 489
+    },
+    {
+      "epoch": 12.25,
+      "learning_rate": 4.53e-05,
+      "loss": 0.4169,
+      "step": 490
+    },
+    {
+      "epoch": 12.28,
+      "learning_rate": 4.527e-05,
+      "loss": 0.1362,
+      "step": 491
+    },
+    {
+      "epoch": 12.3,
+      "learning_rate": 4.524e-05,
+      "loss": 0.8605,
+      "step": 492
+    },
+    {
+      "epoch": 12.32,
+      "learning_rate": 4.5209999999999996e-05,
+      "loss": 0.2946,
+      "step": 493
+    },
+    {
+      "epoch": 12.35,
+      "learning_rate": 4.5180000000000004e-05,
+      "loss": 0.5967,
+      "step": 494
+    },
+    {
+      "epoch": 12.38,
+      "learning_rate": 4.515e-05,
+      "loss": 0.0964,
+      "step": 495
+    },
+    {
+      "epoch": 12.4,
+      "learning_rate": 4.512e-05,
+      "loss": 0.2239,
+      "step": 496
+    },
+    {
+      "epoch": 12.43,
+      "learning_rate": 4.509e-05,
+      "loss": 0.1887,
+      "step": 497
+    },
+    {
+      "epoch": 12.45,
+      "learning_rate": 4.506e-05,
+      "loss": 0.2777,
+      "step": 498
+    },
+    {
+      "epoch": 12.47,
+      "learning_rate": 4.503e-05,
+      "loss": 0.2711,
+      "step": 499
+    },
+    {
+      "epoch": 12.5,
+      "learning_rate": 4.5e-05,
+      "loss": 0.2906,
+      "step": 500
+    },
+    {
+      "epoch": 12.5,
+      "eval_loss": 0.19325587153434753,
+      "eval_mean_accuracy": 0.9273299026552342,
+      "eval_mean_iou": 0.5883427062855762,
+      "eval_overall_accuracy": 0.9495417611260688,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.96756973091831,
+        0.8870900743921584
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.939462057439274,
+        0.8255660614174546
+      ],
+      "eval_runtime": 48.3895,
+      "eval_samples_per_second": 0.413,
+      "eval_steps_per_second": 0.207,
+      "step": 500
+    },
+    {
+      "epoch": 12.53,
+      "learning_rate": 4.4970000000000005e-05,
+      "loss": 0.1969,
+      "step": 501
+    },
+    {
+      "epoch": 12.55,
+      "learning_rate": 4.494e-05,
+      "loss": 0.1268,
+      "step": 502
+    },
+    {
+      "epoch": 12.57,
+      "learning_rate": 4.491e-05,
+      "loss": 0.1312,
+      "step": 503
+    },
+    {
+      "epoch": 12.6,
+      "learning_rate": 4.4880000000000004e-05,
+      "loss": 0.9626,
+      "step": 504
+    },
+    {
+      "epoch": 12.62,
+      "learning_rate": 4.4850000000000006e-05,
+      "loss": 0.2582,
+      "step": 505
+    },
+    {
+      "epoch": 12.65,
+      "learning_rate": 4.482e-05,
+      "loss": 0.2529,
+      "step": 506
+    },
+    {
+      "epoch": 12.68,
+      "learning_rate": 4.479e-05,
+      "loss": 0.1435,
+      "step": 507
+    },
+    {
+      "epoch": 12.7,
+      "learning_rate": 4.476e-05,
+      "loss": 0.2151,
+      "step": 508
+    },
+    {
+      "epoch": 12.72,
+      "learning_rate": 4.473000000000001e-05,
+      "loss": 0.3817,
+      "step": 509
+    },
+    {
+      "epoch": 12.75,
+      "learning_rate": 4.47e-05,
+      "loss": 0.6182,
+      "step": 510
+    },
+    {
+      "epoch": 12.78,
+      "learning_rate": 4.4670000000000004e-05,
+      "loss": 0.1625,
+      "step": 511
+    },
+    {
+      "epoch": 12.8,
+      "learning_rate": 4.464e-05,
+      "loss": 0.2269,
+      "step": 512
+    },
+    {
+      "epoch": 12.82,
+      "learning_rate": 4.461e-05,
+      "loss": 0.1072,
+      "step": 513
+    },
+    {
+      "epoch": 12.85,
+      "learning_rate": 4.4580000000000003e-05,
+      "loss": 0.1614,
+      "step": 514
+    },
+    {
+      "epoch": 12.88,
+      "learning_rate": 4.4550000000000005e-05,
+      "loss": 0.2633,
+      "step": 515
+    },
+    {
+      "epoch": 12.9,
+      "learning_rate": 4.452e-05,
+      "loss": 0.0781,
+      "step": 516
+    },
+    {
+      "epoch": 12.93,
+      "learning_rate": 4.449e-05,
+      "loss": 0.2842,
+      "step": 517
+    },
+    {
+      "epoch": 12.95,
+      "learning_rate": 4.446e-05,
+      "loss": 0.1952,
+      "step": 518
+    },
+    {
+      "epoch": 12.97,
+      "learning_rate": 4.4430000000000007e-05,
+      "loss": 0.2178,
+      "step": 519
+    },
+    {
+      "epoch": 13.0,
+      "learning_rate": 4.44e-05,
+      "loss": 0.143,
+      "step": 520
+    },
+    {
+      "epoch": 13.0,
+      "eval_loss": 0.16903528571128845,
+      "eval_mean_accuracy": 0.9235514768344331,
+      "eval_mean_iou": 0.5737554191549884,
+      "eval_overall_accuracy": 0.9139972318762819,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9062426516046949,
+        0.9408603020641712
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8926643773418198,
+        0.8286018801231455
+      ],
+      "eval_runtime": 48.337,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 520
+    },
+    {
+      "epoch": 13.03,
+      "learning_rate": 4.4370000000000004e-05,
+      "loss": 0.2364,
+      "step": 521
+    },
+    {
+      "epoch": 13.05,
+      "learning_rate": 4.434e-05,
+      "loss": 0.146,
+      "step": 522
+    },
+    {
+      "epoch": 13.07,
+      "learning_rate": 4.431e-05,
+      "loss": 0.5023,
+      "step": 523
+    },
+    {
+      "epoch": 13.1,
+      "learning_rate": 4.428e-05,
+      "loss": 0.2974,
+      "step": 524
+    },
+    {
+      "epoch": 13.12,
+      "learning_rate": 4.4250000000000005e-05,
+      "loss": 0.1238,
+      "step": 525
+    },
+    {
+      "epoch": 13.15,
+      "learning_rate": 4.422e-05,
+      "loss": 0.096,
+      "step": 526
+    },
+    {
+      "epoch": 13.18,
+      "learning_rate": 4.419e-05,
+      "loss": 0.2125,
+      "step": 527
+    },
+    {
+      "epoch": 13.2,
+      "learning_rate": 4.416e-05,
+      "loss": 0.167,
+      "step": 528
+    },
+    {
+      "epoch": 13.22,
+      "learning_rate": 4.4130000000000006e-05,
+      "loss": 0.0964,
+      "step": 529
+    },
+    {
+      "epoch": 13.25,
+      "learning_rate": 4.41e-05,
+      "loss": 0.4862,
+      "step": 530
+    },
+    {
+      "epoch": 13.28,
+      "learning_rate": 4.407e-05,
+      "loss": 0.197,
+      "step": 531
+    },
+    {
+      "epoch": 13.3,
+      "learning_rate": 4.404e-05,
+      "loss": 0.2174,
+      "step": 532
+    },
+    {
+      "epoch": 13.32,
+      "learning_rate": 4.401e-05,
+      "loss": 0.165,
+      "step": 533
+    },
+    {
+      "epoch": 13.35,
+      "learning_rate": 4.398e-05,
+      "loss": 0.2082,
+      "step": 534
+    },
+    {
+      "epoch": 13.38,
+      "learning_rate": 4.3950000000000004e-05,
+      "loss": 0.2521,
+      "step": 535
+    },
+    {
+      "epoch": 13.4,
+      "learning_rate": 4.392e-05,
+      "loss": 0.2875,
+      "step": 536
+    },
+    {
+      "epoch": 13.43,
+      "learning_rate": 4.389e-05,
+      "loss": 0.2588,
+      "step": 537
+    },
+    {
+      "epoch": 13.45,
+      "learning_rate": 4.386e-05,
+      "loss": 0.4512,
+      "step": 538
+    },
+    {
+      "epoch": 13.47,
+      "learning_rate": 4.3830000000000006e-05,
+      "loss": 0.093,
+      "step": 539
+    },
+    {
+      "epoch": 13.5,
+      "learning_rate": 4.38e-05,
+      "loss": 0.226,
+      "step": 540
+    },
+    {
+      "epoch": 13.5,
+      "eval_loss": 0.1501137614250183,
+      "eval_mean_accuracy": 0.9168172988787836,
+      "eval_mean_iou": 0.5812350863104522,
+      "eval_overall_accuracy": 0.9313205625832813,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9430919503893449,
+        0.8905426473682223
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9173144580162539,
+        0.8263908009151028
+      ],
+      "eval_runtime": 48.653,
+      "eval_samples_per_second": 0.411,
+      "eval_steps_per_second": 0.206,
+      "step": 540
+    },
+    {
+      "epoch": 13.53,
+      "learning_rate": 4.377e-05,
+      "loss": 0.1305,
+      "step": 541
+    },
+    {
+      "epoch": 13.55,
+      "learning_rate": 4.374e-05,
+      "loss": 0.0774,
+      "step": 542
+    },
+    {
+      "epoch": 13.57,
+      "learning_rate": 4.371e-05,
+      "loss": 0.6389,
+      "step": 543
+    },
+    {
+      "epoch": 13.6,
+      "learning_rate": 4.368e-05,
+      "loss": 0.2436,
+      "step": 544
+    },
+    {
+      "epoch": 13.62,
+      "learning_rate": 4.3650000000000004e-05,
+      "loss": 0.331,
+      "step": 545
+    },
+    {
+      "epoch": 13.65,
+      "learning_rate": 4.362e-05,
+      "loss": 0.2941,
+      "step": 546
+    },
+    {
+      "epoch": 13.68,
+      "learning_rate": 4.359e-05,
+      "loss": 1.1999,
+      "step": 547
+    },
+    {
+      "epoch": 13.7,
+      "learning_rate": 4.356e-05,
+      "loss": 0.3663,
+      "step": 548
+    },
+    {
+      "epoch": 13.72,
+      "learning_rate": 4.3530000000000005e-05,
+      "loss": 0.7002,
+      "step": 549
+    },
+    {
+      "epoch": 13.75,
+      "learning_rate": 4.35e-05,
+      "loss": 0.2695,
+      "step": 550
+    },
+    {
+      "epoch": 13.78,
+      "learning_rate": 4.347e-05,
+      "loss": 0.0617,
+      "step": 551
+    },
+    {
+      "epoch": 13.8,
+      "learning_rate": 4.344e-05,
+      "loss": 0.1818,
+      "step": 552
+    },
+    {
+      "epoch": 13.82,
+      "learning_rate": 4.3410000000000006e-05,
+      "loss": 0.2036,
+      "step": 553
+    },
+    {
+      "epoch": 13.85,
+      "learning_rate": 4.338e-05,
+      "loss": 0.1806,
+      "step": 554
+    },
+    {
+      "epoch": 13.88,
+      "learning_rate": 4.335e-05,
+      "loss": 0.2204,
+      "step": 555
+    },
+    {
+      "epoch": 13.9,
+      "learning_rate": 4.332e-05,
+      "loss": 0.0898,
+      "step": 556
+    },
+    {
+      "epoch": 13.93,
+      "learning_rate": 4.329e-05,
+      "loss": 0.1312,
+      "step": 557
+    },
+    {
+      "epoch": 13.95,
+      "learning_rate": 4.326e-05,
+      "loss": 0.2342,
+      "step": 558
+    },
+    {
+      "epoch": 13.97,
+      "learning_rate": 4.3230000000000005e-05,
+      "loss": 0.3738,
+      "step": 559
+    },
+    {
+      "epoch": 14.0,
+      "learning_rate": 4.32e-05,
+      "loss": 0.4826,
+      "step": 560
+    },
+    {
+      "epoch": 14.0,
+      "eval_loss": 0.22021767497062683,
+      "eval_mean_accuracy": 0.9476397824599877,
+      "eval_mean_iou": 0.590360191297902,
+      "eval_overall_accuracy": 0.9489838665757923,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9500747751942572,
+        0.9452047897257182
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9366055231315997,
+        0.8344750507621063
+      ],
+      "eval_runtime": 48.3176,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 560
+    },
+    {
+      "epoch": 14.03,
+      "learning_rate": 4.317e-05,
+      "loss": 0.659,
+      "step": 561
+    },
+    {
+      "epoch": 14.05,
+      "learning_rate": 4.314e-05,
+      "loss": 0.7442,
+      "step": 562
+    },
+    {
+      "epoch": 14.07,
+      "learning_rate": 4.3110000000000006e-05,
+      "loss": 0.2401,
+      "step": 563
+    },
+    {
+      "epoch": 14.1,
+      "learning_rate": 4.308e-05,
+      "loss": 0.1112,
+      "step": 564
+    },
+    {
+      "epoch": 14.12,
+      "learning_rate": 4.305e-05,
+      "loss": 1.1314,
+      "step": 565
+    },
+    {
+      "epoch": 14.15,
+      "learning_rate": 4.302e-05,
+      "loss": 0.2401,
+      "step": 566
+    },
+    {
+      "epoch": 14.18,
+      "learning_rate": 4.299e-05,
+      "loss": 0.1054,
+      "step": 567
+    },
+    {
+      "epoch": 14.2,
+      "learning_rate": 4.296e-05,
+      "loss": 0.0728,
+      "step": 568
+    },
+    {
+      "epoch": 14.22,
+      "learning_rate": 4.2930000000000004e-05,
+      "loss": 0.2716,
+      "step": 569
+    },
+    {
+      "epoch": 14.25,
+      "learning_rate": 4.29e-05,
+      "loss": 0.1054,
+      "step": 570
+    },
+    {
+      "epoch": 14.28,
+      "learning_rate": 4.287e-05,
+      "loss": 0.2411,
+      "step": 571
+    },
+    {
+      "epoch": 14.3,
+      "learning_rate": 4.2839999999999996e-05,
+      "loss": 0.2392,
+      "step": 572
+    },
+    {
+      "epoch": 14.32,
+      "learning_rate": 4.2810000000000005e-05,
+      "loss": 0.331,
+      "step": 573
+    },
+    {
+      "epoch": 14.35,
+      "learning_rate": 4.278e-05,
+      "loss": 0.1585,
+      "step": 574
+    },
+    {
+      "epoch": 14.38,
+      "learning_rate": 4.275e-05,
+      "loss": 0.1978,
+      "step": 575
+    },
+    {
+      "epoch": 14.4,
+      "learning_rate": 4.272e-05,
+      "loss": 0.1201,
+      "step": 576
+    },
+    {
+      "epoch": 14.43,
+      "learning_rate": 4.269e-05,
+      "loss": 0.1685,
+      "step": 577
+    },
+    {
+      "epoch": 14.45,
+      "learning_rate": 4.266e-05,
+      "loss": 0.2495,
+      "step": 578
+    },
+    {
+      "epoch": 14.47,
+      "learning_rate": 4.2630000000000004e-05,
+      "loss": 0.1228,
+      "step": 579
+    },
+    {
+      "epoch": 14.5,
+      "learning_rate": 4.26e-05,
+      "loss": 0.0776,
+      "step": 580
+    },
+    {
+      "epoch": 14.5,
+      "eval_loss": 0.20236675441265106,
+      "eval_mean_accuracy": 0.8875189182338821,
+      "eval_mean_iou": 0.5520159873990785,
+      "eval_overall_accuracy": 0.8935424410726399,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.8984313560602167,
+        0.8766064804075475
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.87157656518941,
+        0.7844713970078256
+      ],
+      "eval_runtime": 48.5407,
+      "eval_samples_per_second": 0.412,
+      "eval_steps_per_second": 0.206,
+      "step": 580
+    },
+    {
+      "epoch": 14.53,
+      "learning_rate": 4.257e-05,
+      "loss": 0.1925,
+      "step": 581
+    },
+    {
+      "epoch": 14.55,
+      "learning_rate": 4.2539999999999996e-05,
+      "loss": 0.7366,
+      "step": 582
+    },
+    {
+      "epoch": 14.57,
+      "learning_rate": 4.2510000000000005e-05,
+      "loss": 0.1727,
+      "step": 583
+    },
+    {
+      "epoch": 14.6,
+      "learning_rate": 4.248e-05,
+      "loss": 0.2376,
+      "step": 584
+    },
+    {
+      "epoch": 14.62,
+      "learning_rate": 4.245e-05,
+      "loss": 0.1822,
+      "step": 585
+    },
+    {
+      "epoch": 14.65,
+      "learning_rate": 4.242e-05,
+      "loss": 0.1792,
+      "step": 586
+    },
+    {
+      "epoch": 14.68,
+      "learning_rate": 4.239e-05,
+      "loss": 0.6131,
+      "step": 587
+    },
+    {
+      "epoch": 14.7,
+      "learning_rate": 4.236e-05,
+      "loss": 1.037,
+      "step": 588
+    },
+    {
+      "epoch": 14.72,
+      "learning_rate": 4.233e-05,
+      "loss": 0.4163,
+      "step": 589
+    },
+    {
+      "epoch": 14.75,
+      "learning_rate": 4.23e-05,
+      "loss": 0.2271,
+      "step": 590
+    },
+    {
+      "epoch": 14.78,
+      "learning_rate": 4.227e-05,
+      "loss": 0.2849,
+      "step": 591
+    },
+    {
+      "epoch": 14.8,
+      "learning_rate": 4.2239999999999995e-05,
+      "loss": 0.26,
+      "step": 592
+    },
+    {
+      "epoch": 14.82,
+      "learning_rate": 4.2210000000000004e-05,
+      "loss": 0.08,
+      "step": 593
+    },
+    {
+      "epoch": 14.85,
+      "learning_rate": 4.218e-05,
+      "loss": 1.0104,
+      "step": 594
+    },
+    {
+      "epoch": 14.88,
+      "learning_rate": 4.215e-05,
+      "loss": 0.3434,
+      "step": 595
+    },
+    {
+      "epoch": 14.9,
+      "learning_rate": 4.2119999999999997e-05,
+      "loss": 0.2438,
+      "step": 596
+    },
+    {
+      "epoch": 14.93,
+      "learning_rate": 4.209e-05,
+      "loss": 0.3231,
+      "step": 597
+    },
+    {
+      "epoch": 14.95,
+      "learning_rate": 4.206e-05,
+      "loss": 0.2779,
+      "step": 598
+    },
+    {
+      "epoch": 14.97,
+      "learning_rate": 4.203e-05,
+      "loss": 0.1847,
+      "step": 599
+    },
+    {
+      "epoch": 15.0,
+      "learning_rate": 4.2e-05,
+      "loss": 0.2478,
+      "step": 600
+    },
+    {
+      "epoch": 15.0,
+      "eval_loss": 0.31428685784339905,
+      "eval_mean_accuracy": 0.9244168410195777,
+      "eval_mean_iou": 0.5165786682460257,
+      "eval_overall_accuracy": 0.8922635131324752,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.8661666773080515,
+        0.9826670047311039
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8628620362091799,
+        0.6868739685288975
+      ],
+      "eval_runtime": 49.0378,
+      "eval_samples_per_second": 0.408,
+      "eval_steps_per_second": 0.204,
+      "step": 600
+    },
+    {
+      "epoch": 15.03,
+      "learning_rate": 4.197e-05,
+      "loss": 0.0973,
+      "step": 601
+    },
+    {
+      "epoch": 15.05,
+      "learning_rate": 4.194e-05,
+      "loss": 0.125,
+      "step": 602
+    },
+    {
+      "epoch": 15.07,
+      "learning_rate": 4.1910000000000004e-05,
+      "loss": 0.0719,
+      "step": 603
+    },
+    {
+      "epoch": 15.1,
+      "learning_rate": 4.188e-05,
+      "loss": 0.535,
+      "step": 604
+    },
+    {
+      "epoch": 15.12,
+      "learning_rate": 4.185e-05,
+      "loss": 0.6612,
+      "step": 605
+    },
+    {
+      "epoch": 15.15,
+      "learning_rate": 4.1819999999999996e-05,
+      "loss": 0.0919,
+      "step": 606
+    },
+    {
+      "epoch": 15.18,
+      "learning_rate": 4.1790000000000005e-05,
+      "loss": 0.2264,
+      "step": 607
+    },
+    {
+      "epoch": 15.2,
+      "learning_rate": 4.176e-05,
+      "loss": 0.124,
+      "step": 608
+    },
+    {
+      "epoch": 15.22,
+      "learning_rate": 4.173e-05,
+      "loss": 1.0385,
+      "step": 609
+    },
+    {
+      "epoch": 15.25,
+      "learning_rate": 4.17e-05,
+      "loss": 0.08,
+      "step": 610
+    },
+    {
+      "epoch": 15.28,
+      "learning_rate": 4.167e-05,
+      "loss": 0.1723,
+      "step": 611
+    },
+    {
+      "epoch": 15.3,
+      "learning_rate": 4.164e-05,
+      "loss": 0.2141,
+      "step": 612
+    },
+    {
+      "epoch": 15.32,
+      "learning_rate": 4.161e-05,
+      "loss": 0.0843,
+      "step": 613
+    },
+    {
+      "epoch": 15.35,
+      "learning_rate": 4.158e-05,
+      "loss": 0.3209,
+      "step": 614
+    },
+    {
+      "epoch": 15.38,
+      "learning_rate": 4.155e-05,
+      "loss": 0.1918,
+      "step": 615
+    },
+    {
+      "epoch": 15.4,
+      "learning_rate": 4.1519999999999995e-05,
+      "loss": 0.2061,
+      "step": 616
+    },
+    {
+      "epoch": 15.43,
+      "learning_rate": 4.1490000000000004e-05,
+      "loss": 0.1689,
+      "step": 617
+    },
+    {
+      "epoch": 15.45,
+      "learning_rate": 4.146e-05,
+      "loss": 0.0737,
+      "step": 618
+    },
+    {
+      "epoch": 15.47,
+      "learning_rate": 4.143e-05,
+      "loss": 0.0635,
+      "step": 619
+    },
+    {
+      "epoch": 15.5,
+      "learning_rate": 4.14e-05,
+      "loss": 0.2655,
+      "step": 620
+    },
+    {
+      "epoch": 15.5,
+      "eval_loss": 0.19317014515399933,
+      "eval_mean_accuracy": 0.8589319719879445,
+      "eval_mean_iou": 0.5429208488927418,
+      "eval_overall_accuracy": 0.8903078611168848,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9157736987532968,
+        0.8020902452225921
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8719395715980975,
+        0.7568229750801281
+      ],
+      "eval_runtime": 48.5099,
+      "eval_samples_per_second": 0.412,
+      "eval_steps_per_second": 0.206,
+      "step": 620
+    },
+    {
+      "epoch": 15.53,
+      "learning_rate": 4.137e-05,
+      "loss": 1.1966,
+      "step": 621
+    },
+    {
+      "epoch": 15.55,
+      "learning_rate": 4.134e-05,
+      "loss": 0.41,
+      "step": 622
+    },
+    {
+      "epoch": 15.57,
+      "learning_rate": 4.131e-05,
+      "loss": 0.3119,
+      "step": 623
+    },
+    {
+      "epoch": 15.6,
+      "learning_rate": 4.128e-05,
+      "loss": 0.4695,
+      "step": 624
+    },
+    {
+      "epoch": 15.62,
+      "learning_rate": 4.125e-05,
+      "loss": 0.2726,
+      "step": 625
+    },
+    {
+      "epoch": 15.65,
+      "learning_rate": 4.122e-05,
+      "loss": 0.4765,
+      "step": 626
+    },
+    {
+      "epoch": 15.68,
+      "learning_rate": 4.1190000000000004e-05,
+      "loss": 0.1646,
+      "step": 627
+    },
+    {
+      "epoch": 15.7,
+      "learning_rate": 4.1160000000000006e-05,
+      "loss": 0.1446,
+      "step": 628
+    },
+    {
+      "epoch": 15.72,
+      "learning_rate": 4.113e-05,
+      "loss": 0.1289,
+      "step": 629
+    },
+    {
+      "epoch": 15.75,
+      "learning_rate": 4.11e-05,
+      "loss": 0.2353,
+      "step": 630
+    },
+    {
+      "epoch": 15.78,
+      "learning_rate": 4.107e-05,
+      "loss": 0.109,
+      "step": 631
+    },
+    {
+      "epoch": 15.8,
+      "learning_rate": 4.104000000000001e-05,
+      "loss": 0.167,
+      "step": 632
+    },
+    {
+      "epoch": 15.82,
+      "learning_rate": 4.101e-05,
+      "loss": 0.307,
+      "step": 633
+    },
+    {
+      "epoch": 15.85,
+      "learning_rate": 4.0980000000000004e-05,
+      "loss": 0.1727,
+      "step": 634
+    },
+    {
+      "epoch": 15.88,
+      "learning_rate": 4.095e-05,
+      "loss": 0.1382,
+      "step": 635
+    },
+    {
+      "epoch": 15.9,
+      "learning_rate": 4.092e-05,
+      "loss": 0.3089,
+      "step": 636
+    },
+    {
+      "epoch": 15.93,
+      "learning_rate": 4.089e-05,
+      "loss": 0.138,
+      "step": 637
+    },
+    {
+      "epoch": 15.95,
+      "learning_rate": 4.0860000000000005e-05,
+      "loss": 0.4216,
+      "step": 638
+    },
+    {
+      "epoch": 15.97,
+      "learning_rate": 4.083e-05,
+      "loss": 0.1724,
+      "step": 639
+    },
+    {
+      "epoch": 16.0,
+      "learning_rate": 4.08e-05,
+      "loss": 0.176,
+      "step": 640
+    },
+    {
+      "epoch": 16.0,
+      "eval_loss": 0.1851038783788681,
+      "eval_mean_accuracy": 0.9435937525042573,
+      "eval_mean_iou": 0.5721625613288966,
+      "eval_overall_accuracy": 0.9333205239133342,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9249823898058648,
+        0.9622051152026498
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9165148934878891,
+        0.799972790498801
+      ],
+      "eval_runtime": 48.6944,
+      "eval_samples_per_second": 0.411,
+      "eval_steps_per_second": 0.205,
+      "step": 640
+    },
+    {
+      "epoch": 16.02,
+      "learning_rate": 4.077e-05,
+      "loss": 0.1323,
+      "step": 641
+    },
+    {
+      "epoch": 16.05,
+      "learning_rate": 4.0740000000000006e-05,
+      "loss": 0.1133,
+      "step": 642
+    },
+    {
+      "epoch": 16.07,
+      "learning_rate": 4.071e-05,
+      "loss": 0.1772,
+      "step": 643
+    },
+    {
+      "epoch": 16.1,
+      "learning_rate": 4.0680000000000004e-05,
+      "loss": 0.1973,
+      "step": 644
+    },
+    {
+      "epoch": 16.12,
+      "learning_rate": 4.065e-05,
+      "loss": 0.1182,
+      "step": 645
+    },
+    {
+      "epoch": 16.15,
+      "learning_rate": 4.062e-05,
+      "loss": 0.1707,
+      "step": 646
+    },
+    {
+      "epoch": 16.18,
+      "learning_rate": 4.059e-05,
+      "loss": 0.1126,
+      "step": 647
+    },
+    {
+      "epoch": 16.2,
+      "learning_rate": 4.0560000000000005e-05,
+      "loss": 0.1887,
+      "step": 648
+    },
+    {
+      "epoch": 16.23,
+      "learning_rate": 4.053e-05,
+      "loss": 0.0731,
+      "step": 649
+    },
+    {
+      "epoch": 16.25,
+      "learning_rate": 4.05e-05,
+      "loss": 0.2288,
+      "step": 650
+    },
+    {
+      "epoch": 16.27,
+      "learning_rate": 4.047e-05,
+      "loss": 0.0975,
+      "step": 651
+    },
+    {
+      "epoch": 16.3,
+      "learning_rate": 4.0440000000000006e-05,
+      "loss": 0.6953,
+      "step": 652
+    },
+    {
+      "epoch": 16.32,
+      "learning_rate": 4.041e-05,
+      "loss": 0.3601,
+      "step": 653
+    },
+    {
+      "epoch": 16.35,
+      "learning_rate": 4.038e-05,
+      "loss": 0.3311,
+      "step": 654
+    },
+    {
+      "epoch": 16.38,
+      "learning_rate": 4.035e-05,
+      "loss": 0.1751,
+      "step": 655
+    },
+    {
+      "epoch": 16.4,
+      "learning_rate": 4.032e-05,
+      "loss": 0.0658,
+      "step": 656
+    },
+    {
+      "epoch": 16.43,
+      "learning_rate": 4.029e-05,
+      "loss": 0.0907,
+      "step": 657
+    },
+    {
+      "epoch": 16.45,
+      "learning_rate": 4.0260000000000004e-05,
+      "loss": 0.0836,
+      "step": 658
+    },
+    {
+      "epoch": 16.48,
+      "learning_rate": 4.023e-05,
+      "loss": 0.2174,
+      "step": 659
+    },
+    {
+      "epoch": 16.5,
+      "learning_rate": 4.02e-05,
+      "loss": 1.2637,
+      "step": 660
+    },
+    {
+      "epoch": 16.5,
+      "eval_loss": 0.17593476176261902,
+      "eval_mean_accuracy": 0.9422583491499954,
+      "eval_mean_iou": 0.5795477089728069,
+      "eval_overall_accuracy": 0.9374328311073945,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9335162613337237,
+        0.9510004369662672
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9226226113759224,
+        0.8160205155424984
+      ],
+      "eval_runtime": 48.2239,
+      "eval_samples_per_second": 0.415,
+      "eval_steps_per_second": 0.207,
+      "step": 660
+    },
+    {
+      "epoch": 16.52,
+      "learning_rate": 4.017e-05,
+      "loss": 0.527,
+      "step": 661
+    },
+    {
+      "epoch": 16.55,
+      "learning_rate": 4.0140000000000005e-05,
+      "loss": 0.098,
+      "step": 662
+    },
+    {
+      "epoch": 16.57,
+      "learning_rate": 4.011e-05,
+      "loss": 0.4374,
+      "step": 663
+    },
+    {
+      "epoch": 16.6,
+      "learning_rate": 4.008e-05,
+      "loss": 0.138,
+      "step": 664
+    },
+    {
+      "epoch": 16.62,
+      "learning_rate": 4.005e-05,
+      "loss": 0.1325,
+      "step": 665
+    },
+    {
+      "epoch": 16.65,
+      "learning_rate": 4.0020000000000006e-05,
+      "loss": 0.0577,
+      "step": 666
+    },
+    {
+      "epoch": 16.68,
+      "learning_rate": 3.999e-05,
+      "loss": 0.338,
+      "step": 667
+    },
+    {
+      "epoch": 16.7,
+      "learning_rate": 3.9960000000000004e-05,
+      "loss": 0.1206,
+      "step": 668
+    },
+    {
+      "epoch": 16.73,
+      "learning_rate": 3.993e-05,
+      "loss": 0.2343,
+      "step": 669
+    },
+    {
+      "epoch": 16.75,
+      "learning_rate": 3.99e-05,
+      "loss": 0.1541,
+      "step": 670
+    },
+    {
+      "epoch": 16.77,
+      "learning_rate": 3.987e-05,
+      "loss": 0.2382,
+      "step": 671
+    },
+    {
+      "epoch": 16.8,
+      "learning_rate": 3.9840000000000005e-05,
+      "loss": 0.0979,
+      "step": 672
+    },
+    {
+      "epoch": 16.82,
+      "learning_rate": 3.981e-05,
+      "loss": 0.2911,
+      "step": 673
+    },
+    {
+      "epoch": 16.85,
+      "learning_rate": 3.978e-05,
+      "loss": 0.1072,
+      "step": 674
+    },
+    {
+      "epoch": 16.88,
+      "learning_rate": 3.975e-05,
+      "loss": 0.0793,
+      "step": 675
+    },
+    {
+      "epoch": 16.9,
+      "learning_rate": 3.9720000000000006e-05,
+      "loss": 0.1476,
+      "step": 676
+    },
+    {
+      "epoch": 16.93,
+      "learning_rate": 3.969e-05,
+      "loss": 0.2238,
+      "step": 677
+    },
+    {
+      "epoch": 16.95,
+      "learning_rate": 3.966e-05,
+      "loss": 0.2254,
+      "step": 678
+    },
+    {
+      "epoch": 16.98,
+      "learning_rate": 3.963e-05,
+      "loss": 0.2471,
+      "step": 679
+    },
+    {
+      "epoch": 17.0,
+      "learning_rate": 3.96e-05,
+      "loss": 0.3748,
+      "step": 680
+    },
+    {
+      "epoch": 17.0,
+      "eval_loss": 0.1429952085018158,
+      "eval_mean_accuracy": 0.9260606241904736,
+      "eval_mean_iou": 0.5748151114628661,
+      "eval_overall_accuracy": 0.9240236241152235,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9223703191532987,
+        0.9297509292276487
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9080772684274999,
+        0.8163680659610985
+      ],
+      "eval_runtime": 47.4552,
+      "eval_samples_per_second": 0.421,
+      "eval_steps_per_second": 0.211,
+      "step": 680
+    },
+    {
+      "epoch": 17.02,
+      "learning_rate": 3.957e-05,
+      "loss": 0.3839,
+      "step": 681
+    },
+    {
+      "epoch": 17.05,
+      "learning_rate": 3.9540000000000004e-05,
+      "loss": 0.0602,
+      "step": 682
+    },
+    {
+      "epoch": 17.07,
+      "learning_rate": 3.951e-05,
+      "loss": 0.0754,
+      "step": 683
+    },
+    {
+      "epoch": 17.1,
+      "learning_rate": 3.948e-05,
+      "loss": 0.1139,
+      "step": 684
+    },
+    {
+      "epoch": 17.12,
+      "learning_rate": 3.945e-05,
+      "loss": 0.2653,
+      "step": 685
+    },
+    {
+      "epoch": 17.15,
+      "learning_rate": 3.9420000000000005e-05,
+      "loss": 0.2246,
+      "step": 686
+    },
+    {
+      "epoch": 17.18,
+      "learning_rate": 3.939e-05,
+      "loss": 0.0618,
+      "step": 687
+    },
+    {
+      "epoch": 17.2,
+      "learning_rate": 3.936e-05,
+      "loss": 0.0635,
+      "step": 688
+    },
+    {
+      "epoch": 17.23,
+      "learning_rate": 3.933e-05,
+      "loss": 0.0675,
+      "step": 689
+    },
+    {
+      "epoch": 17.25,
+      "learning_rate": 3.93e-05,
+      "loss": 0.0761,
+      "step": 690
+    },
+    {
+      "epoch": 17.27,
+      "learning_rate": 3.927e-05,
+      "loss": 0.0886,
+      "step": 691
+    },
+    {
+      "epoch": 17.3,
+      "learning_rate": 3.9240000000000004e-05,
+      "loss": 0.1016,
+      "step": 692
+    },
+    {
+      "epoch": 17.32,
+      "learning_rate": 3.921e-05,
+      "loss": 0.4037,
+      "step": 693
+    },
+    {
+      "epoch": 17.35,
+      "learning_rate": 3.918e-05,
+      "loss": 0.2332,
+      "step": 694
+    },
+    {
+      "epoch": 17.38,
+      "learning_rate": 3.9149999999999996e-05,
+      "loss": 0.1594,
+      "step": 695
+    },
+    {
+      "epoch": 17.4,
+      "learning_rate": 3.9120000000000005e-05,
+      "loss": 0.4753,
+      "step": 696
+    },
+    {
+      "epoch": 17.43,
+      "learning_rate": 3.909e-05,
+      "loss": 0.382,
+      "step": 697
+    },
+    {
+      "epoch": 17.45,
+      "learning_rate": 3.906e-05,
+      "loss": 0.5817,
+      "step": 698
+    },
+    {
+      "epoch": 17.48,
+      "learning_rate": 3.903e-05,
+      "loss": 0.1298,
+      "step": 699
+    },
+    {
+      "epoch": 17.5,
+      "learning_rate": 3.9e-05,
+      "loss": 0.9875,
+      "step": 700
+    },
+    {
+      "epoch": 17.5,
+      "eval_loss": 0.16197028756141663,
+      "eval_mean_accuracy": 0.9293318312587986,
+      "eval_mean_iou": 0.563817892491426,
+      "eval_overall_accuracy": 0.9149937886397362,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9033565004067086,
+        0.9553071621108887
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8939333699071477,
+        0.7975203075671302
+      ],
+      "eval_runtime": 48.691,
+      "eval_samples_per_second": 0.411,
+      "eval_steps_per_second": 0.205,
+      "step": 700
+    },
+    {
+      "epoch": 17.52,
+      "learning_rate": 3.897e-05,
+      "loss": 0.1113,
+      "step": 701
+    },
+    {
+      "epoch": 17.55,
+      "learning_rate": 3.894e-05,
+      "loss": 0.2021,
+      "step": 702
+    },
+    {
+      "epoch": 17.57,
+      "learning_rate": 3.891e-05,
+      "loss": 0.1874,
+      "step": 703
+    },
+    {
+      "epoch": 17.6,
+      "learning_rate": 3.888e-05,
+      "loss": 0.5318,
+      "step": 704
+    },
+    {
+      "epoch": 17.62,
+      "learning_rate": 3.8849999999999996e-05,
+      "loss": 0.1391,
+      "step": 705
+    },
+    {
+      "epoch": 17.65,
+      "learning_rate": 3.8820000000000004e-05,
+      "loss": 0.3109,
+      "step": 706
+    },
+    {
+      "epoch": 17.68,
+      "learning_rate": 3.879e-05,
+      "loss": 0.7604,
+      "step": 707
+    },
+    {
+      "epoch": 17.7,
+      "learning_rate": 3.876e-05,
+      "loss": 0.1211,
+      "step": 708
+    },
+    {
+      "epoch": 17.73,
+      "learning_rate": 3.873e-05,
+      "loss": 0.0583,
+      "step": 709
+    },
+    {
+      "epoch": 17.75,
+      "learning_rate": 3.87e-05,
+      "loss": 0.1001,
+      "step": 710
+    },
+    {
+      "epoch": 17.77,
+      "learning_rate": 3.867e-05,
+      "loss": 0.2483,
+      "step": 711
+    },
+    {
+      "epoch": 17.8,
+      "learning_rate": 3.864e-05,
+      "loss": 0.3419,
+      "step": 712
+    },
+    {
+      "epoch": 17.82,
+      "learning_rate": 3.861e-05,
+      "loss": 0.1669,
+      "step": 713
+    },
+    {
+      "epoch": 17.85,
+      "learning_rate": 3.858e-05,
+      "loss": 0.2238,
+      "step": 714
+    },
+    {
+      "epoch": 17.88,
+      "learning_rate": 3.855e-05,
+      "loss": 0.1803,
+      "step": 715
+    },
+    {
+      "epoch": 17.9,
+      "learning_rate": 3.8520000000000004e-05,
+      "loss": 0.6072,
+      "step": 716
+    },
+    {
+      "epoch": 17.93,
+      "learning_rate": 3.849e-05,
+      "loss": 0.1341,
+      "step": 717
+    },
+    {
+      "epoch": 17.95,
+      "learning_rate": 3.846e-05,
+      "loss": 0.0749,
+      "step": 718
+    },
+    {
+      "epoch": 17.98,
+      "learning_rate": 3.8429999999999996e-05,
+      "loss": 0.1464,
+      "step": 719
+    },
+    {
+      "epoch": 18.0,
+      "learning_rate": 3.8400000000000005e-05,
+      "loss": 0.16,
+      "step": 720
+    },
+    {
+      "epoch": 18.0,
+      "eval_loss": 0.15552404522895813,
+      "eval_mean_accuracy": 0.9247679225016672,
+      "eval_mean_iou": 0.5778008229677836,
+      "eval_overall_accuracy": 0.9264860135634839,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9278804801185191,
+        0.9216553648848154
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9099117121849809,
+        0.8234907567183699
+      ],
+      "eval_runtime": 48.7573,
+      "eval_samples_per_second": 0.41,
+      "eval_steps_per_second": 0.205,
+      "step": 720
+    },
+    {
+      "epoch": 18.02,
+      "learning_rate": 3.837e-05,
+      "loss": 0.1312,
+      "step": 721
+    },
+    {
+      "epoch": 18.05,
+      "learning_rate": 3.834e-05,
+      "loss": 0.2353,
+      "step": 722
+    },
+    {
+      "epoch": 18.07,
+      "learning_rate": 3.831e-05,
+      "loss": 0.1771,
+      "step": 723
+    },
+    {
+      "epoch": 18.1,
+      "learning_rate": 3.828e-05,
+      "loss": 0.1728,
+      "step": 724
+    },
+    {
+      "epoch": 18.12,
+      "learning_rate": 3.825e-05,
+      "loss": 0.263,
+      "step": 725
+    },
+    {
+      "epoch": 18.15,
+      "learning_rate": 3.822e-05,
+      "loss": 0.0744,
+      "step": 726
+    },
+    {
+      "epoch": 18.18,
+      "learning_rate": 3.819e-05,
+      "loss": 0.0975,
+      "step": 727
+    },
+    {
+      "epoch": 18.2,
+      "learning_rate": 3.816e-05,
+      "loss": 0.1179,
+      "step": 728
+    },
+    {
+      "epoch": 18.23,
+      "learning_rate": 3.8129999999999996e-05,
+      "loss": 0.3915,
+      "step": 729
+    },
+    {
+      "epoch": 18.25,
+      "learning_rate": 3.8100000000000005e-05,
+      "loss": 0.5075,
+      "step": 730
+    },
+    {
+      "epoch": 18.27,
+      "learning_rate": 3.807e-05,
+      "loss": 0.2501,
+      "step": 731
+    },
+    {
+      "epoch": 18.3,
+      "learning_rate": 3.804e-05,
+      "loss": 0.081,
+      "step": 732
+    },
+    {
+      "epoch": 18.32,
+      "learning_rate": 3.801e-05,
+      "loss": 0.125,
+      "step": 733
+    },
+    {
+      "epoch": 18.35,
+      "learning_rate": 3.798e-05,
+      "loss": 0.1165,
+      "step": 734
+    },
+    {
+      "epoch": 18.38,
+      "learning_rate": 3.795e-05,
+      "loss": 0.1679,
+      "step": 735
+    },
+    {
+      "epoch": 18.4,
+      "learning_rate": 3.792e-05,
+      "loss": 0.7162,
+      "step": 736
+    },
+    {
+      "epoch": 18.43,
+      "learning_rate": 3.789e-05,
+      "loss": 0.3411,
+      "step": 737
+    },
+    {
+      "epoch": 18.45,
+      "learning_rate": 3.786e-05,
+      "loss": 0.2077,
+      "step": 738
+    },
+    {
+      "epoch": 18.48,
+      "learning_rate": 3.7829999999999995e-05,
+      "loss": 0.2529,
+      "step": 739
+    },
+    {
+      "epoch": 18.5,
+      "learning_rate": 3.7800000000000004e-05,
+      "loss": 0.3697,
+      "step": 740
+    },
+    {
+      "epoch": 18.5,
+      "eval_loss": 0.15720753371715546,
+      "eval_mean_accuracy": 0.9430400754892173,
+      "eval_mean_iou": 0.5705525377970841,
+      "eval_overall_accuracy": 0.929726635698484,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9189209531981514,
+        0.9671591977802833
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9117028284079924,
+        0.79995478498326
+      ],
+      "eval_runtime": 49.0295,
+      "eval_samples_per_second": 0.408,
+      "eval_steps_per_second": 0.204,
+      "step": 740
+    },
+    {
+      "epoch": 18.52,
+      "learning_rate": 3.777e-05,
+      "loss": 0.2686,
+      "step": 741
+    },
+    {
+      "epoch": 18.55,
+      "learning_rate": 3.774e-05,
+      "loss": 0.269,
+      "step": 742
+    },
+    {
+      "epoch": 18.57,
+      "learning_rate": 3.7709999999999996e-05,
+      "loss": 0.4568,
+      "step": 743
+    },
+    {
+      "epoch": 18.6,
+      "learning_rate": 3.768e-05,
+      "loss": 0.1015,
+      "step": 744
+    },
+    {
+      "epoch": 18.62,
+      "learning_rate": 3.765e-05,
+      "loss": 0.0878,
+      "step": 745
+    },
+    {
+      "epoch": 18.65,
+      "learning_rate": 3.762e-05,
+      "loss": 0.1061,
+      "step": 746
+    },
+    {
+      "epoch": 18.68,
+      "learning_rate": 3.759e-05,
+      "loss": 0.1726,
+      "step": 747
+    },
+    {
+      "epoch": 18.7,
+      "learning_rate": 3.756e-05,
+      "loss": 0.152,
+      "step": 748
+    },
+    {
+      "epoch": 18.73,
+      "learning_rate": 3.7529999999999995e-05,
+      "loss": 0.0627,
+      "step": 749
+    },
+    {
+      "epoch": 18.75,
+      "learning_rate": 3.7500000000000003e-05,
+      "loss": 0.0597,
+      "step": 750
+    },
+    {
+      "epoch": 18.77,
+      "learning_rate": 3.7470000000000005e-05,
+      "loss": 0.2041,
+      "step": 751
+    },
+    {
+      "epoch": 18.8,
+      "learning_rate": 3.744e-05,
+      "loss": 0.1792,
+      "step": 752
+    },
+    {
+      "epoch": 18.82,
+      "learning_rate": 3.741e-05,
+      "loss": 0.1539,
+      "step": 753
+    },
+    {
+      "epoch": 18.85,
+      "learning_rate": 3.738e-05,
+      "loss": 0.1403,
+      "step": 754
+    },
+    {
+      "epoch": 18.88,
+      "learning_rate": 3.735000000000001e-05,
+      "loss": 0.2269,
+      "step": 755
+    },
+    {
+      "epoch": 18.9,
+      "learning_rate": 3.732e-05,
+      "loss": 0.0968,
+      "step": 756
+    },
+    {
+      "epoch": 18.93,
+      "learning_rate": 3.7290000000000004e-05,
+      "loss": 0.1496,
+      "step": 757
+    },
+    {
+      "epoch": 18.95,
+      "learning_rate": 3.726e-05,
+      "loss": 0.3167,
+      "step": 758
+    },
+    {
+      "epoch": 18.98,
+      "learning_rate": 3.723e-05,
+      "loss": 0.1403,
+      "step": 759
+    },
+    {
+      "epoch": 19.0,
+      "learning_rate": 3.72e-05,
+      "loss": 0.1338,
+      "step": 760
+    },
+    {
+      "epoch": 19.0,
+      "eval_loss": 0.14783106744289398,
+      "eval_mean_accuracy": 0.9454575459882153,
+      "eval_mean_iou": 0.5758130853178719,
+      "eval_overall_accuracy": 0.9243466792988494,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9072123153525076,
+        0.9837027766239231
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9046942157145423,
+        0.8227450402390731
+      ],
+      "eval_runtime": 48.5789,
+      "eval_samples_per_second": 0.412,
+      "eval_steps_per_second": 0.206,
+      "step": 760
+    },
+    {
+      "epoch": 19.02,
+      "learning_rate": 3.7170000000000005e-05,
+      "loss": 0.1798,
+      "step": 761
+    },
+    {
+      "epoch": 19.05,
+      "learning_rate": 3.714e-05,
+      "loss": 0.0672,
+      "step": 762
+    },
+    {
+      "epoch": 19.07,
+      "learning_rate": 3.711e-05,
+      "loss": 1.237,
+      "step": 763
+    },
+    {
+      "epoch": 19.1,
+      "learning_rate": 3.708e-05,
+      "loss": 0.2717,
+      "step": 764
+    },
+    {
+      "epoch": 19.12,
+      "learning_rate": 3.7050000000000006e-05,
+      "loss": 0.2913,
+      "step": 765
+    },
+    {
+      "epoch": 19.15,
+      "learning_rate": 3.702e-05,
+      "loss": 0.1676,
+      "step": 766
+    },
+    {
+      "epoch": 19.18,
+      "learning_rate": 3.699e-05,
+      "loss": 0.1547,
+      "step": 767
+    },
+    {
+      "epoch": 19.2,
+      "learning_rate": 3.696e-05,
+      "loss": 0.1109,
+      "step": 768
+    },
+    {
+      "epoch": 19.23,
+      "learning_rate": 3.693e-05,
+      "loss": 0.1951,
+      "step": 769
+    },
+    {
+      "epoch": 19.25,
+      "learning_rate": 3.69e-05,
+      "loss": 0.077,
+      "step": 770
+    },
+    {
+      "epoch": 19.27,
+      "learning_rate": 3.6870000000000004e-05,
+      "loss": 0.1245,
+      "step": 771
+    },
+    {
+      "epoch": 19.3,
+      "learning_rate": 3.684e-05,
+      "loss": 0.0799,
+      "step": 772
+    },
+    {
+      "epoch": 19.32,
+      "learning_rate": 3.681e-05,
+      "loss": 0.2119,
+      "step": 773
+    },
+    {
+      "epoch": 19.35,
+      "learning_rate": 3.6780000000000004e-05,
+      "loss": 0.3093,
+      "step": 774
+    },
+    {
+      "epoch": 19.38,
+      "learning_rate": 3.6750000000000006e-05,
+      "loss": 0.0728,
+      "step": 775
+    },
+    {
+      "epoch": 19.4,
+      "learning_rate": 3.672e-05,
+      "loss": 0.1206,
+      "step": 776
+    },
+    {
+      "epoch": 19.43,
+      "learning_rate": 3.669e-05,
+      "loss": 0.1638,
+      "step": 777
+    },
+    {
+      "epoch": 19.45,
+      "learning_rate": 3.666e-05,
+      "loss": 0.3086,
+      "step": 778
+    },
+    {
+      "epoch": 19.48,
+      "learning_rate": 3.663000000000001e-05,
+      "loss": 0.712,
+      "step": 779
+    },
+    {
+      "epoch": 19.5,
+      "learning_rate": 3.66e-05,
+      "loss": 0.7522,
+      "step": 780
+    },
+    {
+      "epoch": 19.5,
+      "eval_loss": 0.30090320110321045,
+      "eval_mean_accuracy": 0.9207649216566469,
+      "eval_mean_iou": 0.5125294777072046,
+      "eval_overall_accuracy": 0.8864320045372738,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.8585661330920594,
+        0.9829637102212344
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8555010673222496,
+        0.6820873657993641
+      ],
+      "eval_runtime": 47.7776,
+      "eval_samples_per_second": 0.419,
+      "eval_steps_per_second": 0.209,
+      "step": 780
+    },
+    {
+      "epoch": 19.52,
+      "learning_rate": 3.6570000000000004e-05,
+      "loss": 0.1087,
+      "step": 781
+    },
+    {
+      "epoch": 19.55,
+      "learning_rate": 3.654e-05,
+      "loss": 0.3194,
+      "step": 782
+    },
+    {
+      "epoch": 19.57,
+      "learning_rate": 3.651e-05,
+      "loss": 0.1143,
+      "step": 783
+    },
+    {
+      "epoch": 19.6,
+      "learning_rate": 3.648e-05,
+      "loss": 0.1599,
+      "step": 784
+    },
+    {
+      "epoch": 19.62,
+      "learning_rate": 3.6450000000000005e-05,
+      "loss": 0.1602,
+      "step": 785
+    },
+    {
+      "epoch": 19.65,
+      "learning_rate": 3.642e-05,
+      "loss": 0.1478,
+      "step": 786
+    },
+    {
+      "epoch": 19.68,
+      "learning_rate": 3.639e-05,
+      "loss": 0.2748,
+      "step": 787
+    },
+    {
+      "epoch": 19.7,
+      "learning_rate": 3.636e-05,
+      "loss": 0.1033,
+      "step": 788
+    },
+    {
+      "epoch": 19.73,
+      "learning_rate": 3.6330000000000006e-05,
+      "loss": 0.0757,
+      "step": 789
+    },
+    {
+      "epoch": 19.75,
+      "learning_rate": 3.63e-05,
+      "loss": 0.7663,
+      "step": 790
+    },
+    {
+      "epoch": 19.77,
+      "learning_rate": 3.6270000000000003e-05,
+      "loss": 0.0485,
+      "step": 791
+    },
+    {
+      "epoch": 19.8,
+      "learning_rate": 3.624e-05,
+      "loss": 0.0576,
+      "step": 792
+    },
+    {
+      "epoch": 19.82,
+      "learning_rate": 3.621e-05,
+      "loss": 0.1733,
+      "step": 793
+    },
+    {
+      "epoch": 19.85,
+      "learning_rate": 3.618e-05,
+      "loss": 0.1801,
+      "step": 794
+    },
+    {
+      "epoch": 19.88,
+      "learning_rate": 3.6150000000000005e-05,
+      "loss": 0.6399,
+      "step": 795
+    },
+    {
+      "epoch": 19.9,
+      "learning_rate": 3.612e-05,
+      "loss": 0.2138,
+      "step": 796
+    },
+    {
+      "epoch": 19.93,
+      "learning_rate": 3.609e-05,
+      "loss": 1.1095,
+      "step": 797
+    },
+    {
+      "epoch": 19.95,
+      "learning_rate": 3.606e-05,
+      "loss": 0.2221,
+      "step": 798
+    },
+    {
+      "epoch": 19.98,
+      "learning_rate": 3.6030000000000006e-05,
+      "loss": 0.1944,
+      "step": 799
+    },
+    {
+      "epoch": 20.0,
+      "learning_rate": 3.6e-05,
+      "loss": 0.1449,
+      "step": 800
+    },
+    {
+      "epoch": 20.0,
+      "eval_loss": 0.161110058426857,
+      "eval_mean_accuracy": 0.9330079294552827,
+      "eval_mean_iou": 0.5745221298756912,
+      "eval_overall_accuracy": 0.9217948655977649,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9126939262609133,
+        0.9533219326496519
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9032388150253235,
+        0.8203275746017501
+      ],
+      "eval_runtime": 48.6183,
+      "eval_samples_per_second": 0.411,
+      "eval_steps_per_second": 0.206,
+      "step": 800
+    },
+    {
+      "epoch": 20.02,
+      "learning_rate": 3.597e-05,
+      "loss": 0.1831,
+      "step": 801
+    },
+    {
+      "epoch": 20.05,
+      "learning_rate": 3.594e-05,
+      "loss": 0.2915,
+      "step": 802
+    },
+    {
+      "epoch": 20.07,
+      "learning_rate": 3.591e-05,
+      "loss": 0.2098,
+      "step": 803
+    },
+    {
+      "epoch": 20.1,
+      "learning_rate": 3.588e-05,
+      "loss": 0.1522,
+      "step": 804
+    },
+    {
+      "epoch": 20.12,
+      "learning_rate": 3.5850000000000004e-05,
+      "loss": 0.1788,
+      "step": 805
+    },
+    {
+      "epoch": 20.15,
+      "learning_rate": 3.582e-05,
+      "loss": 0.0554,
+      "step": 806
+    },
+    {
+      "epoch": 20.18,
+      "learning_rate": 3.579e-05,
+      "loss": 0.0983,
+      "step": 807
+    },
+    {
+      "epoch": 20.2,
+      "learning_rate": 3.5759999999999996e-05,
+      "loss": 0.1853,
+      "step": 808
+    },
+    {
+      "epoch": 20.23,
+      "learning_rate": 3.5730000000000005e-05,
+      "loss": 0.23,
+      "step": 809
+    },
+    {
+      "epoch": 20.25,
+      "learning_rate": 3.57e-05,
+      "loss": 0.1719,
+      "step": 810
+    },
+    {
+      "epoch": 20.27,
+      "learning_rate": 3.567e-05,
+      "loss": 0.0485,
+      "step": 811
+    },
+    {
+      "epoch": 20.3,
+      "learning_rate": 3.564e-05,
+      "loss": 0.1964,
+      "step": 812
+    },
+    {
+      "epoch": 20.32,
+      "learning_rate": 3.561e-05,
+      "loss": 0.1697,
+      "step": 813
+    },
+    {
+      "epoch": 20.35,
+      "learning_rate": 3.558e-05,
+      "loss": 0.0638,
+      "step": 814
+    },
+    {
+      "epoch": 20.38,
+      "learning_rate": 3.5550000000000004e-05,
+      "loss": 0.1374,
+      "step": 815
+    },
+    {
+      "epoch": 20.4,
+      "learning_rate": 3.552e-05,
+      "loss": 0.045,
+      "step": 816
+    },
+    {
+      "epoch": 20.43,
+      "learning_rate": 3.549e-05,
+      "loss": 0.1264,
+      "step": 817
+    },
+    {
+      "epoch": 20.45,
+      "learning_rate": 3.5459999999999996e-05,
+      "loss": 0.2464,
+      "step": 818
+    },
+    {
+      "epoch": 20.48,
+      "learning_rate": 3.5430000000000005e-05,
+      "loss": 0.234,
+      "step": 819
+    },
+    {
+      "epoch": 20.5,
+      "learning_rate": 3.54e-05,
+      "loss": 0.2661,
+      "step": 820
+    },
+    {
+      "epoch": 20.5,
+      "eval_loss": 0.15575893223285675,
+      "eval_mean_accuracy": 0.9424208546546318,
+      "eval_mean_iou": 0.5733499196680856,
+      "eval_overall_accuracy": 0.9244240191931837,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9098171187178844,
+        0.975024590591379
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9057178734574919,
+        0.8143318855467647
+      ],
+      "eval_runtime": 48.847,
+      "eval_samples_per_second": 0.409,
+      "eval_steps_per_second": 0.205,
+      "step": 820
+    },
+    {
+      "epoch": 20.52,
+      "learning_rate": 3.537e-05,
+      "loss": 1.111,
+      "step": 821
+    },
+    {
+      "epoch": 20.55,
+      "learning_rate": 3.534e-05,
+      "loss": 0.1015,
+      "step": 822
+    },
+    {
+      "epoch": 20.57,
+      "learning_rate": 3.531e-05,
+      "loss": 0.1468,
+      "step": 823
+    },
+    {
+      "epoch": 20.6,
+      "learning_rate": 3.528e-05,
+      "loss": 0.1679,
+      "step": 824
+    },
+    {
+      "epoch": 20.62,
+      "learning_rate": 3.525e-05,
+      "loss": 0.1482,
+      "step": 825
+    },
+    {
+      "epoch": 20.65,
+      "learning_rate": 3.522e-05,
+      "loss": 0.1424,
+      "step": 826
+    },
+    {
+      "epoch": 20.68,
+      "learning_rate": 3.519e-05,
+      "loss": 0.0609,
+      "step": 827
+    },
+    {
+      "epoch": 20.7,
+      "learning_rate": 3.516e-05,
+      "loss": 0.1189,
+      "step": 828
+    },
+    {
+      "epoch": 20.73,
+      "learning_rate": 3.5130000000000004e-05,
+      "loss": 0.0904,
+      "step": 829
+    },
+    {
+      "epoch": 20.75,
+      "learning_rate": 3.51e-05,
+      "loss": 0.1175,
+      "step": 830
+    },
+    {
+      "epoch": 20.77,
+      "learning_rate": 3.507e-05,
+      "loss": 0.1293,
+      "step": 831
+    },
+    {
+      "epoch": 20.8,
+      "learning_rate": 3.5039999999999997e-05,
+      "loss": 0.0566,
+      "step": 832
+    },
+    {
+      "epoch": 20.82,
+      "learning_rate": 3.5010000000000005e-05,
+      "loss": 0.0742,
+      "step": 833
+    },
+    {
+      "epoch": 20.85,
+      "learning_rate": 3.498e-05,
+      "loss": 0.1171,
+      "step": 834
+    },
+    {
+      "epoch": 20.88,
+      "learning_rate": 3.495e-05,
+      "loss": 0.193,
+      "step": 835
+    },
+    {
+      "epoch": 20.9,
+      "learning_rate": 3.492e-05,
+      "loss": 0.154,
+      "step": 836
+    },
+    {
+      "epoch": 20.93,
+      "learning_rate": 3.489e-05,
+      "loss": 0.1993,
+      "step": 837
+    },
+    {
+      "epoch": 20.95,
+      "learning_rate": 3.486e-05,
+      "loss": 0.1253,
+      "step": 838
+    },
+    {
+      "epoch": 20.98,
+      "learning_rate": 3.4830000000000004e-05,
+      "loss": 0.0621,
+      "step": 839
+    },
+    {
+      "epoch": 21.0,
+      "learning_rate": 3.48e-05,
+      "loss": 0.1951,
+      "step": 840
+    },
+    {
+      "epoch": 21.0,
+      "eval_loss": 0.2067786008119583,
+      "eval_mean_accuracy": 0.954513781309211,
+      "eval_mean_iou": 0.5791708070121998,
+      "eval_overall_accuracy": 0.9421984992838004,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9322029587202515,
+        0.9768246038981707
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9270619743496028,
+        0.8104504466869968
+      ],
+      "eval_runtime": 48.2927,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 840
+    },
+    {
+      "epoch": 21.02,
+      "learning_rate": 3.477e-05,
+      "loss": 0.1603,
+      "step": 841
+    },
+    {
+      "epoch": 21.05,
+      "learning_rate": 3.4739999999999996e-05,
+      "loss": 0.0336,
+      "step": 842
+    },
+    {
+      "epoch": 21.07,
+      "learning_rate": 3.4710000000000005e-05,
+      "loss": 0.2433,
+      "step": 843
+    },
+    {
+      "epoch": 21.1,
+      "learning_rate": 3.468e-05,
+      "loss": 0.0526,
+      "step": 844
+    },
+    {
+      "epoch": 21.12,
+      "learning_rate": 3.465e-05,
+      "loss": 0.1717,
+      "step": 845
+    },
+    {
+      "epoch": 21.15,
+      "learning_rate": 3.462e-05,
+      "loss": 0.1157,
+      "step": 846
+    },
+    {
+      "epoch": 21.18,
+      "learning_rate": 3.459e-05,
+      "loss": 0.5927,
+      "step": 847
+    },
+    {
+      "epoch": 21.2,
+      "learning_rate": 3.456e-05,
+      "loss": 0.1884,
+      "step": 848
+    },
+    {
+      "epoch": 21.23,
+      "learning_rate": 3.453e-05,
+      "loss": 0.1366,
+      "step": 849
+    },
+    {
+      "epoch": 21.25,
+      "learning_rate": 3.45e-05,
+      "loss": 0.0697,
+      "step": 850
+    },
+    {
+      "epoch": 21.27,
+      "learning_rate": 3.447e-05,
+      "loss": 0.0812,
+      "step": 851
+    },
+    {
+      "epoch": 21.3,
+      "learning_rate": 3.4439999999999996e-05,
+      "loss": 0.1395,
+      "step": 852
+    },
+    {
+      "epoch": 21.32,
+      "learning_rate": 3.4410000000000004e-05,
+      "loss": 0.1977,
+      "step": 853
+    },
+    {
+      "epoch": 21.35,
+      "learning_rate": 3.438e-05,
+      "loss": 0.1142,
+      "step": 854
+    },
+    {
+      "epoch": 21.38,
+      "learning_rate": 3.435e-05,
+      "loss": 0.1353,
+      "step": 855
+    },
+    {
+      "epoch": 21.4,
+      "learning_rate": 3.432e-05,
+      "loss": 0.5259,
+      "step": 856
+    },
+    {
+      "epoch": 21.43,
+      "learning_rate": 3.429e-05,
+      "loss": 0.3162,
+      "step": 857
+    },
+    {
+      "epoch": 21.45,
+      "learning_rate": 3.426e-05,
+      "loss": 0.3085,
+      "step": 858
+    },
+    {
+      "epoch": 21.48,
+      "learning_rate": 3.423e-05,
+      "loss": 0.0997,
+      "step": 859
+    },
+    {
+      "epoch": 21.5,
+      "learning_rate": 3.42e-05,
+      "loss": 0.1015,
+      "step": 860
+    },
+    {
+      "epoch": 21.5,
+      "eval_loss": 0.1526772677898407,
+      "eval_mean_accuracy": 0.9402454093097647,
+      "eval_mean_iou": 0.5674436765724752,
+      "eval_overall_accuracy": 0.9198090026859501,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9032220555937088,
+        0.9772687630258206
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9001446953405203,
+        0.802186334376905
+      ],
+      "eval_runtime": 47.7529,
+      "eval_samples_per_second": 0.419,
+      "eval_steps_per_second": 0.209,
+      "step": 860
+    },
+    {
+      "epoch": 21.52,
+      "learning_rate": 3.417e-05,
+      "loss": 0.1098,
+      "step": 861
+    },
+    {
+      "epoch": 21.55,
+      "learning_rate": 3.4139999999999995e-05,
+      "loss": 0.0643,
+      "step": 862
+    },
+    {
+      "epoch": 21.57,
+      "learning_rate": 3.4110000000000004e-05,
+      "loss": 0.1593,
+      "step": 863
+    },
+    {
+      "epoch": 21.6,
+      "learning_rate": 3.408e-05,
+      "loss": 0.9459,
+      "step": 864
+    },
+    {
+      "epoch": 21.62,
+      "learning_rate": 3.405e-05,
+      "loss": 0.0629,
+      "step": 865
+    },
+    {
+      "epoch": 21.65,
+      "learning_rate": 3.4019999999999996e-05,
+      "loss": 0.1155,
+      "step": 866
+    },
+    {
+      "epoch": 21.68,
+      "learning_rate": 3.399e-05,
+      "loss": 0.2053,
+      "step": 867
+    },
+    {
+      "epoch": 21.7,
+      "learning_rate": 3.396e-05,
+      "loss": 0.1684,
+      "step": 868
+    },
+    {
+      "epoch": 21.73,
+      "learning_rate": 3.393e-05,
+      "loss": 0.2342,
+      "step": 869
+    },
+    {
+      "epoch": 21.75,
+      "learning_rate": 3.39e-05,
+      "loss": 0.0851,
+      "step": 870
+    },
+    {
+      "epoch": 21.77,
+      "learning_rate": 3.387e-05,
+      "loss": 0.114,
+      "step": 871
+    },
+    {
+      "epoch": 21.8,
+      "learning_rate": 3.3839999999999994e-05,
+      "loss": 0.0905,
+      "step": 872
+    },
+    {
+      "epoch": 21.82,
+      "learning_rate": 3.381e-05,
+      "loss": 0.1668,
+      "step": 873
+    },
+    {
+      "epoch": 21.85,
+      "learning_rate": 3.378e-05,
+      "loss": 0.0753,
+      "step": 874
+    },
+    {
+      "epoch": 21.88,
+      "learning_rate": 3.375e-05,
+      "loss": 0.24,
+      "step": 875
+    },
+    {
+      "epoch": 21.9,
+      "learning_rate": 3.372e-05,
+      "loss": 0.0768,
+      "step": 876
+    },
+    {
+      "epoch": 21.93,
+      "learning_rate": 3.369e-05,
+      "loss": 0.1481,
+      "step": 877
+    },
+    {
+      "epoch": 21.95,
+      "learning_rate": 3.3660000000000006e-05,
+      "loss": 0.0601,
+      "step": 878
+    },
+    {
+      "epoch": 21.98,
+      "learning_rate": 3.363e-05,
+      "loss": 0.1339,
+      "step": 879
+    },
+    {
+      "epoch": 22.0,
+      "learning_rate": 3.3600000000000004e-05,
+      "loss": 0.293,
+      "step": 880
+    },
+    {
+      "epoch": 22.0,
+      "eval_loss": 0.14977200329303741,
+      "eval_mean_accuracy": 0.9319219947774842,
+      "eval_mean_iou": 0.5899828627725737,
+      "eval_overall_accuracy": 0.9393002673060098,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9452887474882439,
+        0.9185552420667246
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9254936739096737,
+        0.8444549144080475
+      ],
+      "eval_runtime": 47.9376,
+      "eval_samples_per_second": 0.417,
+      "eval_steps_per_second": 0.209,
+      "step": 880
+    },
+    {
+      "epoch": 22.02,
+      "learning_rate": 3.357e-05,
+      "loss": 0.161,
+      "step": 881
+    },
+    {
+      "epoch": 22.05,
+      "learning_rate": 3.354e-05,
+      "loss": 0.0778,
+      "step": 882
+    },
+    {
+      "epoch": 22.07,
+      "learning_rate": 3.351e-05,
+      "loss": 0.1908,
+      "step": 883
+    },
+    {
+      "epoch": 22.1,
+      "learning_rate": 3.3480000000000005e-05,
+      "loss": 0.2216,
+      "step": 884
+    },
+    {
+      "epoch": 22.12,
+      "learning_rate": 3.345e-05,
+      "loss": 0.5083,
+      "step": 885
+    },
+    {
+      "epoch": 22.15,
+      "learning_rate": 3.342e-05,
+      "loss": 0.1677,
+      "step": 886
+    },
+    {
+      "epoch": 22.18,
+      "learning_rate": 3.3390000000000004e-05,
+      "loss": 0.0513,
+      "step": 887
+    },
+    {
+      "epoch": 22.2,
+      "learning_rate": 3.3360000000000006e-05,
+      "loss": 0.2139,
+      "step": 888
+    },
+    {
+      "epoch": 22.23,
+      "learning_rate": 3.333e-05,
+      "loss": 0.152,
+      "step": 889
+    },
+    {
+      "epoch": 22.25,
+      "learning_rate": 3.33e-05,
+      "loss": 0.2707,
+      "step": 890
+    },
+    {
+      "epoch": 22.27,
+      "learning_rate": 3.327e-05,
+      "loss": 0.1546,
+      "step": 891
+    },
+    {
+      "epoch": 22.3,
+      "learning_rate": 3.324000000000001e-05,
+      "loss": 0.0585,
+      "step": 892
+    },
+    {
+      "epoch": 22.32,
+      "learning_rate": 3.321e-05,
+      "loss": 0.425,
+      "step": 893
+    },
+    {
+      "epoch": 22.35,
+      "learning_rate": 3.3180000000000004e-05,
+      "loss": 0.1308,
+      "step": 894
+    },
+    {
+      "epoch": 22.38,
+      "learning_rate": 3.315e-05,
+      "loss": 0.1866,
+      "step": 895
+    },
+    {
+      "epoch": 22.4,
+      "learning_rate": 3.312e-05,
+      "loss": 0.05,
+      "step": 896
+    },
+    {
+      "epoch": 22.43,
+      "learning_rate": 3.309e-05,
+      "loss": 0.1705,
+      "step": 897
+    },
+    {
+      "epoch": 22.45,
+      "learning_rate": 3.3060000000000005e-05,
+      "loss": 0.1573,
+      "step": 898
+    },
+    {
+      "epoch": 22.48,
+      "learning_rate": 3.303e-05,
+      "loss": 0.081,
+      "step": 899
+    },
+    {
+      "epoch": 22.5,
+      "learning_rate": 3.3e-05,
+      "loss": 0.1246,
+      "step": 900
+    },
+    {
+      "epoch": 22.5,
+      "eval_loss": 0.18260009586811066,
+      "eval_mean_accuracy": 0.9329330838917222,
+      "eval_mean_iou": 0.5442009612433875,
+      "eval_overall_accuracy": 0.909885730306121,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.8911796416293051,
+        0.9746865261541394
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8858857774156546,
+        0.7467171063145076
+      ],
+      "eval_runtime": 48.5884,
+      "eval_samples_per_second": 0.412,
+      "eval_steps_per_second": 0.206,
+      "step": 900
+    },
+    {
+      "epoch": 22.52,
+      "learning_rate": 3.297e-05,
+      "loss": 0.1514,
+      "step": 901
+    },
+    {
+      "epoch": 22.55,
+      "learning_rate": 3.2940000000000006e-05,
+      "loss": 0.2313,
+      "step": 902
+    },
+    {
+      "epoch": 22.57,
+      "learning_rate": 3.291e-05,
+      "loss": 0.1051,
+      "step": 903
+    },
+    {
+      "epoch": 22.6,
+      "learning_rate": 3.2880000000000004e-05,
+      "loss": 0.045,
+      "step": 904
+    },
+    {
+      "epoch": 22.62,
+      "learning_rate": 3.285e-05,
+      "loss": 0.1758,
+      "step": 905
+    },
+    {
+      "epoch": 22.65,
+      "learning_rate": 3.282e-05,
+      "loss": 0.0755,
+      "step": 906
+    },
+    {
+      "epoch": 22.68,
+      "learning_rate": 3.279e-05,
+      "loss": 0.8215,
+      "step": 907
+    },
+    {
+      "epoch": 22.7,
+      "learning_rate": 3.2760000000000005e-05,
+      "loss": 0.039,
+      "step": 908
+    },
+    {
+      "epoch": 22.73,
+      "learning_rate": 3.273e-05,
+      "loss": 0.1883,
+      "step": 909
+    },
+    {
+      "epoch": 22.75,
+      "learning_rate": 3.27e-05,
+      "loss": 0.1038,
+      "step": 910
+    },
+    {
+      "epoch": 22.77,
+      "learning_rate": 3.267e-05,
+      "loss": 0.5631,
+      "step": 911
+    },
+    {
+      "epoch": 22.8,
+      "learning_rate": 3.2640000000000006e-05,
+      "loss": 0.3108,
+      "step": 912
+    },
+    {
+      "epoch": 22.82,
+      "learning_rate": 3.261e-05,
+      "loss": 0.1469,
+      "step": 913
+    },
+    {
+      "epoch": 22.85,
+      "learning_rate": 3.258e-05,
+      "loss": 0.0762,
+      "step": 914
+    },
+    {
+      "epoch": 22.88,
+      "learning_rate": 3.255e-05,
+      "loss": 0.1664,
+      "step": 915
+    },
+    {
+      "epoch": 22.9,
+      "learning_rate": 3.252e-05,
+      "loss": 0.1231,
+      "step": 916
+    },
+    {
+      "epoch": 22.93,
+      "learning_rate": 3.249e-05,
+      "loss": 0.5659,
+      "step": 917
+    },
+    {
+      "epoch": 22.95,
+      "learning_rate": 3.2460000000000004e-05,
+      "loss": 0.159,
+      "step": 918
+    },
+    {
+      "epoch": 22.98,
+      "learning_rate": 3.243e-05,
+      "loss": 0.0483,
+      "step": 919
+    },
+    {
+      "epoch": 23.0,
+      "learning_rate": 3.24e-05,
+      "loss": 0.1243,
+      "step": 920
+    },
+    {
+      "epoch": 23.0,
+      "eval_loss": 0.16654913127422333,
+      "eval_mean_accuracy": 0.9321520220158931,
+      "eval_mean_iou": 0.562641953931424,
+      "eval_overall_accuracy": 0.9077226301364566,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.8878948278197982,
+        0.976409216211988
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8845905410374336,
+        0.8033353207568384
+      ],
+      "eval_runtime": 48.2046,
+      "eval_samples_per_second": 0.415,
+      "eval_steps_per_second": 0.207,
+      "step": 920
+    },
+    {
+      "epoch": 23.02,
+      "learning_rate": 3.237e-05,
+      "loss": 0.0707,
+      "step": 921
+    },
+    {
+      "epoch": 23.05,
+      "learning_rate": 3.2340000000000005e-05,
+      "loss": 0.0467,
+      "step": 922
+    },
+    {
+      "epoch": 23.07,
+      "learning_rate": 3.231e-05,
+      "loss": 0.2368,
+      "step": 923
+    },
+    {
+      "epoch": 23.1,
+      "learning_rate": 3.228e-05,
+      "loss": 0.1229,
+      "step": 924
+    },
+    {
+      "epoch": 23.12,
+      "learning_rate": 3.225e-05,
+      "loss": 0.0503,
+      "step": 925
+    },
+    {
+      "epoch": 23.15,
+      "learning_rate": 3.222e-05,
+      "loss": 0.0769,
+      "step": 926
+    },
+    {
+      "epoch": 23.18,
+      "learning_rate": 3.219e-05,
+      "loss": 0.0734,
+      "step": 927
+    },
+    {
+      "epoch": 23.2,
+      "learning_rate": 3.2160000000000004e-05,
+      "loss": 0.1102,
+      "step": 928
+    },
+    {
+      "epoch": 23.23,
+      "learning_rate": 3.213e-05,
+      "loss": 0.2293,
+      "step": 929
+    },
+    {
+      "epoch": 23.25,
+      "learning_rate": 3.21e-05,
+      "loss": 1.006,
+      "step": 930
+    },
+    {
+      "epoch": 23.27,
+      "learning_rate": 3.2069999999999996e-05,
+      "loss": 0.182,
+      "step": 931
+    },
+    {
+      "epoch": 23.3,
+      "learning_rate": 3.2040000000000005e-05,
+      "loss": 0.1221,
+      "step": 932
+    },
+    {
+      "epoch": 23.32,
+      "learning_rate": 3.201e-05,
+      "loss": 0.0659,
+      "step": 933
+    },
+    {
+      "epoch": 23.35,
+      "learning_rate": 3.198e-05,
+      "loss": 0.1934,
+      "step": 934
+    },
+    {
+      "epoch": 23.38,
+      "learning_rate": 3.195e-05,
+      "loss": 0.3662,
+      "step": 935
+    },
+    {
+      "epoch": 23.4,
+      "learning_rate": 3.192e-05,
+      "loss": 0.1094,
+      "step": 936
+    },
+    {
+      "epoch": 23.43,
+      "learning_rate": 3.189e-05,
+      "loss": 0.3466,
+      "step": 937
+    },
+    {
+      "epoch": 23.45,
+      "learning_rate": 3.186e-05,
+      "loss": 0.1565,
+      "step": 938
+    },
+    {
+      "epoch": 23.48,
+      "learning_rate": 3.183e-05,
+      "loss": 0.0585,
+      "step": 939
+    },
+    {
+      "epoch": 23.5,
+      "learning_rate": 3.18e-05,
+      "loss": 0.2825,
+      "step": 940
+    },
+    {
+      "epoch": 23.5,
+      "eval_loss": 0.256104052066803,
+      "eval_mean_accuracy": 0.8992670882287088,
+      "eval_mean_iou": 0.4865682121604448,
+      "eval_overall_accuracy": 0.8505583779246162,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.8110245784843657,
+        0.9875095979730519
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8092426358962087,
+        0.6504620005851257
+      ],
+      "eval_runtime": 48.9285,
+      "eval_samples_per_second": 0.409,
+      "eval_steps_per_second": 0.204,
+      "step": 940
+    },
+    {
+      "epoch": 23.52,
+      "learning_rate": 3.1769999999999996e-05,
+      "loss": 0.1726,
+      "step": 941
+    },
+    {
+      "epoch": 23.55,
+      "learning_rate": 3.1740000000000004e-05,
+      "loss": 0.0788,
+      "step": 942
+    },
+    {
+      "epoch": 23.57,
+      "learning_rate": 3.171e-05,
+      "loss": 0.2621,
+      "step": 943
+    },
+    {
+      "epoch": 23.6,
+      "learning_rate": 3.168e-05,
+      "loss": 0.0422,
+      "step": 944
+    },
+    {
+      "epoch": 23.62,
+      "learning_rate": 3.165e-05,
+      "loss": 0.1793,
+      "step": 945
+    },
+    {
+      "epoch": 23.65,
+      "learning_rate": 3.1620000000000006e-05,
+      "loss": 0.0958,
+      "step": 946
+    },
+    {
+      "epoch": 23.68,
+      "learning_rate": 3.159e-05,
+      "loss": 0.1503,
+      "step": 947
+    },
+    {
+      "epoch": 23.7,
+      "learning_rate": 3.156e-05,
+      "loss": 0.1169,
+      "step": 948
+    },
+    {
+      "epoch": 23.73,
+      "learning_rate": 3.153e-05,
+      "loss": 0.2745,
+      "step": 949
+    },
+    {
+      "epoch": 23.75,
+      "learning_rate": 3.15e-05,
+      "loss": 0.2336,
+      "step": 950
+    },
+    {
+      "epoch": 23.77,
+      "learning_rate": 3.147e-05,
+      "loss": 0.1181,
+      "step": 951
+    },
+    {
+      "epoch": 23.8,
+      "learning_rate": 3.1440000000000004e-05,
+      "loss": 0.0454,
+      "step": 952
+    },
+    {
+      "epoch": 23.82,
+      "learning_rate": 3.141e-05,
+      "loss": 0.2745,
+      "step": 953
+    },
+    {
+      "epoch": 23.85,
+      "learning_rate": 3.138e-05,
+      "loss": 0.1712,
+      "step": 954
+    },
+    {
+      "epoch": 23.88,
+      "learning_rate": 3.1349999999999996e-05,
+      "loss": 0.1363,
+      "step": 955
+    },
+    {
+      "epoch": 23.9,
+      "learning_rate": 3.1320000000000005e-05,
+      "loss": 0.2775,
+      "step": 956
+    },
+    {
+      "epoch": 23.93,
+      "learning_rate": 3.129e-05,
+      "loss": 0.2212,
+      "step": 957
+    },
+    {
+      "epoch": 23.95,
+      "learning_rate": 3.126e-05,
+      "loss": 0.0593,
+      "step": 958
+    },
+    {
+      "epoch": 23.98,
+      "learning_rate": 3.123e-05,
+      "loss": 0.0591,
+      "step": 959
+    },
+    {
+      "epoch": 24.0,
+      "learning_rate": 3.12e-05,
+      "loss": 0.1744,
+      "step": 960
+    },
+    {
+      "epoch": 24.0,
+      "eval_loss": 0.16813531517982483,
+      "eval_mean_accuracy": 0.9478273629765915,
+      "eval_mean_iou": 0.5834502854952942,
+      "eval_overall_accuracy": 0.9403922905015331,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9343577093718417,
+        0.9612970165813414
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9259582204169315,
+        0.8243926360689513
+      ],
+      "eval_runtime": 47.5753,
+      "eval_samples_per_second": 0.42,
+      "eval_steps_per_second": 0.21,
+      "step": 960
+    },
+    {
+      "epoch": 24.02,
+      "learning_rate": 3.117e-05,
+      "loss": 0.1811,
+      "step": 961
+    },
+    {
+      "epoch": 24.05,
+      "learning_rate": 3.1140000000000003e-05,
+      "loss": 0.0841,
+      "step": 962
+    },
+    {
+      "epoch": 24.07,
+      "learning_rate": 3.111e-05,
+      "loss": 0.0302,
+      "step": 963
+    },
+    {
+      "epoch": 24.1,
+      "learning_rate": 3.108e-05,
+      "loss": 0.2162,
+      "step": 964
+    },
+    {
+      "epoch": 24.12,
+      "learning_rate": 3.1049999999999996e-05,
+      "loss": 0.1507,
+      "step": 965
+    },
+    {
+      "epoch": 24.15,
+      "learning_rate": 3.1020000000000005e-05,
+      "loss": 0.0737,
+      "step": 966
+    },
+    {
+      "epoch": 24.18,
+      "learning_rate": 3.099e-05,
+      "loss": 0.2242,
+      "step": 967
+    },
+    {
+      "epoch": 24.2,
+      "learning_rate": 3.096e-05,
+      "loss": 0.2893,
+      "step": 968
+    },
+    {
+      "epoch": 24.23,
+      "learning_rate": 3.093e-05,
+      "loss": 0.0528,
+      "step": 969
+    },
+    {
+      "epoch": 24.25,
+      "learning_rate": 3.09e-05,
+      "loss": 0.1138,
+      "step": 970
+    },
+    {
+      "epoch": 24.27,
+      "learning_rate": 3.087e-05,
+      "loss": 0.0636,
+      "step": 971
+    },
+    {
+      "epoch": 24.3,
+      "learning_rate": 3.084e-05,
+      "loss": 0.1288,
+      "step": 972
+    },
+    {
+      "epoch": 24.32,
+      "learning_rate": 3.081e-05,
+      "loss": 0.1742,
+      "step": 973
+    },
+    {
+      "epoch": 24.35,
+      "learning_rate": 3.078e-05,
+      "loss": 0.1587,
+      "step": 974
+    },
+    {
+      "epoch": 24.38,
+      "learning_rate": 3.0749999999999995e-05,
+      "loss": 0.1458,
+      "step": 975
+    },
+    {
+      "epoch": 24.4,
+      "learning_rate": 3.0720000000000004e-05,
+      "loss": 0.0764,
+      "step": 976
+    },
+    {
+      "epoch": 24.43,
+      "learning_rate": 3.069e-05,
+      "loss": 0.194,
+      "step": 977
+    },
+    {
+      "epoch": 24.45,
+      "learning_rate": 3.066e-05,
+      "loss": 0.0569,
+      "step": 978
+    },
+    {
+      "epoch": 24.48,
+      "learning_rate": 3.0629999999999996e-05,
+      "loss": 0.0761,
+      "step": 979
+    },
+    {
+      "epoch": 24.5,
+      "learning_rate": 3.06e-05,
+      "loss": 0.5815,
+      "step": 980
+    },
+    {
+      "epoch": 24.5,
+      "eval_loss": 0.14417901635169983,
+      "eval_mean_accuracy": 0.9263959317675383,
+      "eval_mean_iou": 0.5814876190615167,
+      "eval_overall_accuracy": 0.9300380093355698,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9329940548399873,
+        0.9197978086950892
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9139827267771568,
+        0.8304801304073932
+      ],
+      "eval_runtime": 48.6861,
+      "eval_samples_per_second": 0.411,
+      "eval_steps_per_second": 0.205,
+      "step": 980
+    },
+    {
+      "epoch": 24.52,
+      "learning_rate": 3.057e-05,
+      "loss": 0.7866,
+      "step": 981
+    },
+    {
+      "epoch": 24.55,
+      "learning_rate": 3.054e-05,
+      "loss": 0.1626,
+      "step": 982
+    },
+    {
+      "epoch": 24.57,
+      "learning_rate": 3.0509999999999998e-05,
+      "loss": 0.149,
+      "step": 983
+    },
+    {
+      "epoch": 24.6,
+      "learning_rate": 3.048e-05,
+      "loss": 0.233,
+      "step": 984
+    },
+    {
+      "epoch": 24.62,
+      "learning_rate": 3.0449999999999998e-05,
+      "loss": 0.1141,
+      "step": 985
+    },
+    {
+      "epoch": 24.65,
+      "learning_rate": 3.042e-05,
+      "loss": 0.1139,
+      "step": 986
+    },
+    {
+      "epoch": 24.68,
+      "learning_rate": 3.039e-05,
+      "loss": 0.0739,
+      "step": 987
+    },
+    {
+      "epoch": 24.7,
+      "learning_rate": 3.036e-05,
+      "loss": 0.1264,
+      "step": 988
+    },
+    {
+      "epoch": 24.73,
+      "learning_rate": 3.033e-05,
+      "loss": 0.1145,
+      "step": 989
+    },
+    {
+      "epoch": 24.75,
+      "learning_rate": 3.03e-05,
+      "loss": 0.9447,
+      "step": 990
+    },
+    {
+      "epoch": 24.77,
+      "learning_rate": 3.0269999999999996e-05,
+      "loss": 0.1732,
+      "step": 991
+    },
+    {
+      "epoch": 24.8,
+      "learning_rate": 3.0240000000000002e-05,
+      "loss": 0.1101,
+      "step": 992
+    },
+    {
+      "epoch": 24.82,
+      "learning_rate": 3.0209999999999997e-05,
+      "loss": 0.144,
+      "step": 993
+    },
+    {
+      "epoch": 24.85,
+      "learning_rate": 3.0180000000000002e-05,
+      "loss": 0.377,
+      "step": 994
+    },
+    {
+      "epoch": 24.88,
+      "learning_rate": 3.0149999999999998e-05,
+      "loss": 0.1945,
+      "step": 995
+    },
+    {
+      "epoch": 24.9,
+      "learning_rate": 3.012e-05,
+      "loss": 0.057,
+      "step": 996
+    },
+    {
+      "epoch": 24.93,
+      "learning_rate": 3.0089999999999998e-05,
+      "loss": 0.1585,
+      "step": 997
+    },
+    {
+      "epoch": 24.95,
+      "learning_rate": 3.006e-05,
+      "loss": 0.2699,
+      "step": 998
+    },
+    {
+      "epoch": 24.98,
+      "learning_rate": 3.003e-05,
+      "loss": 0.173,
+      "step": 999
+    },
+    {
+      "epoch": 25.0,
+      "learning_rate": 3e-05,
+      "loss": 0.1304,
+      "step": 1000
+    },
+    {
+      "epoch": 25.0,
+      "eval_loss": 0.1669236570596695,
+      "eval_mean_accuracy": 0.9456798873018607,
+      "eval_mean_iou": 0.5816811024664563,
+      "eval_overall_accuracy": 0.9389155818940862,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.933425420243859,
+        0.9579343543598624
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9242141792859011,
+        0.8208291281134678
+      ],
+      "eval_runtime": 47.8829,
+      "eval_samples_per_second": 0.418,
+      "eval_steps_per_second": 0.209,
+      "step": 1000
+    },
+    {
+      "epoch": 25.02,
+      "learning_rate": 2.997e-05,
+      "loss": 0.1638,
+      "step": 1001
+    },
+    {
+      "epoch": 25.05,
+      "learning_rate": 2.994e-05,
+      "loss": 0.2992,
+      "step": 1002
+    },
+    {
+      "epoch": 25.07,
+      "learning_rate": 2.991e-05,
+      "loss": 0.2946,
+      "step": 1003
+    },
+    {
+      "epoch": 25.1,
+      "learning_rate": 2.9880000000000002e-05,
+      "loss": 0.2507,
+      "step": 1004
+    },
+    {
+      "epoch": 25.12,
+      "learning_rate": 2.985e-05,
+      "loss": 0.139,
+      "step": 1005
+    },
+    {
+      "epoch": 25.15,
+      "learning_rate": 2.982e-05,
+      "loss": 0.1165,
+      "step": 1006
+    },
+    {
+      "epoch": 25.18,
+      "learning_rate": 2.979e-05,
+      "loss": 0.1621,
+      "step": 1007
+    },
+    {
+      "epoch": 25.2,
+      "learning_rate": 2.976e-05,
+      "loss": 0.091,
+      "step": 1008
+    },
+    {
+      "epoch": 25.23,
+      "learning_rate": 2.973e-05,
+      "loss": 0.0911,
+      "step": 1009
+    },
+    {
+      "epoch": 25.25,
+      "learning_rate": 2.97e-05,
+      "loss": 0.0695,
+      "step": 1010
+    },
+    {
+      "epoch": 25.27,
+      "learning_rate": 2.967e-05,
+      "loss": 0.0873,
+      "step": 1011
+    },
+    {
+      "epoch": 25.3,
+      "learning_rate": 2.964e-05,
+      "loss": 0.0346,
+      "step": 1012
+    },
+    {
+      "epoch": 25.32,
+      "learning_rate": 2.961e-05,
+      "loss": 0.1843,
+      "step": 1013
+    },
+    {
+      "epoch": 25.35,
+      "learning_rate": 2.958e-05,
+      "loss": 0.0884,
+      "step": 1014
+    },
+    {
+      "epoch": 25.38,
+      "learning_rate": 2.955e-05,
+      "loss": 0.2015,
+      "step": 1015
+    },
+    {
+      "epoch": 25.4,
+      "learning_rate": 2.9520000000000002e-05,
+      "loss": 0.1084,
+      "step": 1016
+    },
+    {
+      "epoch": 25.43,
+      "learning_rate": 2.949e-05,
+      "loss": 0.0809,
+      "step": 1017
+    },
+    {
+      "epoch": 25.45,
+      "learning_rate": 2.946e-05,
+      "loss": 0.0906,
+      "step": 1018
+    },
+    {
+      "epoch": 25.48,
+      "learning_rate": 2.943e-05,
+      "loss": 0.2197,
+      "step": 1019
+    },
+    {
+      "epoch": 25.5,
+      "learning_rate": 2.94e-05,
+      "loss": 0.065,
+      "step": 1020
+    },
+    {
+      "epoch": 25.5,
+      "eval_loss": 0.1728467345237732,
+      "eval_mean_accuracy": 0.9262948263485297,
+      "eval_mean_iou": 0.566378377855929,
+      "eval_overall_accuracy": 0.9134864663241153,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9030907253323616,
+        0.9494989273646978
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8919851355550018,
+        0.8071499980127854
+      ],
+      "eval_runtime": 48.3367,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 1020
+    },
+    {
+      "epoch": 25.52,
+      "learning_rate": 2.9370000000000002e-05,
+      "loss": 0.1145,
+      "step": 1021
+    },
+    {
+      "epoch": 25.55,
+      "learning_rate": 2.934e-05,
+      "loss": 0.1756,
+      "step": 1022
+    },
+    {
+      "epoch": 25.57,
+      "learning_rate": 2.931e-05,
+      "loss": 0.0636,
+      "step": 1023
+    },
+    {
+      "epoch": 25.6,
+      "learning_rate": 2.928e-05,
+      "loss": 0.1047,
+      "step": 1024
+    },
+    {
+      "epoch": 25.62,
+      "learning_rate": 2.925e-05,
+      "loss": 0.0842,
+      "step": 1025
+    },
+    {
+      "epoch": 25.65,
+      "learning_rate": 2.922e-05,
+      "loss": 0.8233,
+      "step": 1026
+    },
+    {
+      "epoch": 25.68,
+      "learning_rate": 2.919e-05,
+      "loss": 0.0665,
+      "step": 1027
+    },
+    {
+      "epoch": 25.7,
+      "learning_rate": 2.916e-05,
+      "loss": 0.1181,
+      "step": 1028
+    },
+    {
+      "epoch": 25.73,
+      "learning_rate": 2.913e-05,
+      "loss": 0.1675,
+      "step": 1029
+    },
+    {
+      "epoch": 25.75,
+      "learning_rate": 2.91e-05,
+      "loss": 0.0643,
+      "step": 1030
+    },
+    {
+      "epoch": 25.77,
+      "learning_rate": 2.907e-05,
+      "loss": 0.1912,
+      "step": 1031
+    },
+    {
+      "epoch": 25.8,
+      "learning_rate": 2.904e-05,
+      "loss": 0.1314,
+      "step": 1032
+    },
+    {
+      "epoch": 25.82,
+      "learning_rate": 2.901e-05,
+      "loss": 0.1107,
+      "step": 1033
+    },
+    {
+      "epoch": 25.85,
+      "learning_rate": 2.898e-05,
+      "loss": 0.1791,
+      "step": 1034
+    },
+    {
+      "epoch": 25.88,
+      "learning_rate": 2.895e-05,
+      "loss": 0.0753,
+      "step": 1035
+    },
+    {
+      "epoch": 25.9,
+      "learning_rate": 2.892e-05,
+      "loss": 0.1201,
+      "step": 1036
+    },
+    {
+      "epoch": 25.93,
+      "learning_rate": 2.889e-05,
+      "loss": 0.1003,
+      "step": 1037
+    },
+    {
+      "epoch": 25.95,
+      "learning_rate": 2.8859999999999998e-05,
+      "loss": 0.0946,
+      "step": 1038
+    },
+    {
+      "epoch": 25.98,
+      "learning_rate": 2.883e-05,
+      "loss": 0.139,
+      "step": 1039
+    },
+    {
+      "epoch": 26.0,
+      "learning_rate": 2.88e-05,
+      "loss": 0.2734,
+      "step": 1040
+    },
+    {
+      "epoch": 26.0,
+      "eval_loss": 0.16094741225242615,
+      "eval_mean_accuracy": 0.9361853274982819,
+      "eval_mean_iou": 0.5693030045598508,
+      "eval_overall_accuracy": 0.9267023235804505,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9190055651847111,
+        0.9533650898118528
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9087456004057097,
+        0.7991634132738428
+      ],
+      "eval_runtime": 48.458,
+      "eval_samples_per_second": 0.413,
+      "eval_steps_per_second": 0.206,
+      "step": 1040
+    },
+    {
+      "epoch": 26.02,
+      "learning_rate": 2.877e-05,
+      "loss": 0.124,
+      "step": 1041
+    },
+    {
+      "epoch": 26.05,
+      "learning_rate": 2.874e-05,
+      "loss": 0.1015,
+      "step": 1042
+    },
+    {
+      "epoch": 26.07,
+      "learning_rate": 2.871e-05,
+      "loss": 0.1019,
+      "step": 1043
+    },
+    {
+      "epoch": 26.1,
+      "learning_rate": 2.868e-05,
+      "loss": 0.0594,
+      "step": 1044
+    },
+    {
+      "epoch": 26.12,
+      "learning_rate": 2.865e-05,
+      "loss": 0.1333,
+      "step": 1045
+    },
+    {
+      "epoch": 26.15,
+      "learning_rate": 2.862e-05,
+      "loss": 0.109,
+      "step": 1046
+    },
+    {
+      "epoch": 26.18,
+      "learning_rate": 2.859e-05,
+      "loss": 0.0929,
+      "step": 1047
+    },
+    {
+      "epoch": 26.2,
+      "learning_rate": 2.856e-05,
+      "loss": 0.3913,
+      "step": 1048
+    },
+    {
+      "epoch": 26.23,
+      "learning_rate": 2.853e-05,
+      "loss": 0.1095,
+      "step": 1049
+    },
+    {
+      "epoch": 26.25,
+      "learning_rate": 2.8499999999999998e-05,
+      "loss": 0.1029,
+      "step": 1050
+    },
+    {
+      "epoch": 26.27,
+      "learning_rate": 2.847e-05,
+      "loss": 0.113,
+      "step": 1051
+    },
+    {
+      "epoch": 26.3,
+      "learning_rate": 2.844e-05,
+      "loss": 0.0627,
+      "step": 1052
+    },
+    {
+      "epoch": 26.32,
+      "learning_rate": 2.841e-05,
+      "loss": 0.1225,
+      "step": 1053
+    },
+    {
+      "epoch": 26.35,
+      "learning_rate": 2.838e-05,
+      "loss": 0.3177,
+      "step": 1054
+    },
+    {
+      "epoch": 26.38,
+      "learning_rate": 2.8349999999999998e-05,
+      "loss": 0.164,
+      "step": 1055
+    },
+    {
+      "epoch": 26.4,
+      "learning_rate": 2.832e-05,
+      "loss": 0.0786,
+      "step": 1056
+    },
+    {
+      "epoch": 26.43,
+      "learning_rate": 2.829e-05,
+      "loss": 0.0325,
+      "step": 1057
+    },
+    {
+      "epoch": 26.45,
+      "learning_rate": 2.826e-05,
+      "loss": 0.2409,
+      "step": 1058
+    },
+    {
+      "epoch": 26.48,
+      "learning_rate": 2.823e-05,
+      "loss": 0.2664,
+      "step": 1059
+    },
+    {
+      "epoch": 26.5,
+      "learning_rate": 2.8199999999999998e-05,
+      "loss": 0.1941,
+      "step": 1060
+    },
+    {
+      "epoch": 26.5,
+      "eval_loss": 0.1887330263853073,
+      "eval_mean_accuracy": 0.9286438591581025,
+      "eval_mean_iou": 0.5598981742260135,
+      "eval_overall_accuracy": 0.9052537956970016,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.8862695509491336,
+        0.9710181673670715
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8815020187313487,
+        0.7981925039466916
+      ],
+      "eval_runtime": 48.4501,
+      "eval_samples_per_second": 0.413,
+      "eval_steps_per_second": 0.206,
+      "step": 1060
+    },
+    {
+      "epoch": 26.52,
+      "learning_rate": 2.817e-05,
+      "loss": 0.5032,
+      "step": 1061
+    },
+    {
+      "epoch": 26.55,
+      "learning_rate": 2.8139999999999998e-05,
+      "loss": 0.0904,
+      "step": 1062
+    },
+    {
+      "epoch": 26.57,
+      "learning_rate": 2.8110000000000004e-05,
+      "loss": 0.0895,
+      "step": 1063
+    },
+    {
+      "epoch": 26.6,
+      "learning_rate": 2.8080000000000002e-05,
+      "loss": 0.1405,
+      "step": 1064
+    },
+    {
+      "epoch": 26.62,
+      "learning_rate": 2.805e-05,
+      "loss": 0.1647,
+      "step": 1065
+    },
+    {
+      "epoch": 26.65,
+      "learning_rate": 2.8020000000000003e-05,
+      "loss": 0.1523,
+      "step": 1066
+    },
+    {
+      "epoch": 26.68,
+      "learning_rate": 2.799e-05,
+      "loss": 0.0459,
+      "step": 1067
+    },
+    {
+      "epoch": 26.7,
+      "learning_rate": 2.7960000000000003e-05,
+      "loss": 0.0915,
+      "step": 1068
+    },
+    {
+      "epoch": 26.73,
+      "learning_rate": 2.7930000000000002e-05,
+      "loss": 0.2523,
+      "step": 1069
+    },
+    {
+      "epoch": 26.75,
+      "learning_rate": 2.79e-05,
+      "loss": 0.0447,
+      "step": 1070
+    },
+    {
+      "epoch": 26.77,
+      "learning_rate": 2.7870000000000003e-05,
+      "loss": 0.186,
+      "step": 1071
+    },
+    {
+      "epoch": 26.8,
+      "learning_rate": 2.784e-05,
+      "loss": 0.1905,
+      "step": 1072
+    },
+    {
+      "epoch": 26.82,
+      "learning_rate": 2.7810000000000003e-05,
+      "loss": 0.2621,
+      "step": 1073
+    },
+    {
+      "epoch": 26.85,
+      "learning_rate": 2.778e-05,
+      "loss": 0.0945,
+      "step": 1074
+    },
+    {
+      "epoch": 26.88,
+      "learning_rate": 2.7750000000000004e-05,
+      "loss": 0.1728,
+      "step": 1075
+    },
+    {
+      "epoch": 26.9,
+      "learning_rate": 2.7720000000000002e-05,
+      "loss": 0.0536,
+      "step": 1076
+    },
+    {
+      "epoch": 26.93,
+      "learning_rate": 2.769e-05,
+      "loss": 0.1973,
+      "step": 1077
+    },
+    {
+      "epoch": 26.95,
+      "learning_rate": 2.7660000000000003e-05,
+      "loss": 0.2486,
+      "step": 1078
+    },
+    {
+      "epoch": 26.98,
+      "learning_rate": 2.763e-05,
+      "loss": 0.0485,
+      "step": 1079
+    },
+    {
+      "epoch": 27.0,
+      "learning_rate": 2.7600000000000003e-05,
+      "loss": 0.2251,
+      "step": 1080
+    },
+    {
+      "epoch": 27.0,
+      "eval_loss": 0.15258844196796417,
+      "eval_mean_accuracy": 0.9275157112111172,
+      "eval_mean_iou": 0.5864343359742814,
+      "eval_overall_accuracy": 0.9324786469385486,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9365067500120688,
+        0.9185246724101657
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9172787790258706,
+        0.8420242288969736
+      ],
+      "eval_runtime": 50.2818,
+      "eval_samples_per_second": 0.398,
+      "eval_steps_per_second": 0.199,
+      "step": 1080
+    },
+    {
+      "epoch": 27.02,
+      "learning_rate": 2.7570000000000002e-05,
+      "loss": 0.0773,
+      "step": 1081
+    },
+    {
+      "epoch": 27.05,
+      "learning_rate": 2.754e-05,
+      "loss": 0.0622,
+      "step": 1082
+    },
+    {
+      "epoch": 27.07,
+      "learning_rate": 2.7510000000000003e-05,
+      "loss": 0.3959,
+      "step": 1083
+    },
+    {
+      "epoch": 27.1,
+      "learning_rate": 2.748e-05,
+      "loss": 0.2353,
+      "step": 1084
+    },
+    {
+      "epoch": 27.12,
+      "learning_rate": 2.7450000000000003e-05,
+      "loss": 0.101,
+      "step": 1085
+    },
+    {
+      "epoch": 27.15,
+      "learning_rate": 2.7420000000000002e-05,
+      "loss": 0.1155,
+      "step": 1086
+    },
+    {
+      "epoch": 27.18,
+      "learning_rate": 2.739e-05,
+      "loss": 0.0893,
+      "step": 1087
+    },
+    {
+      "epoch": 27.2,
+      "learning_rate": 2.7360000000000002e-05,
+      "loss": 0.0436,
+      "step": 1088
+    },
+    {
+      "epoch": 27.23,
+      "learning_rate": 2.733e-05,
+      "loss": 0.1927,
+      "step": 1089
+    },
+    {
+      "epoch": 27.25,
+      "learning_rate": 2.7300000000000003e-05,
+      "loss": 0.1195,
+      "step": 1090
+    },
+    {
+      "epoch": 27.27,
+      "learning_rate": 2.727e-05,
+      "loss": 0.1076,
+      "step": 1091
+    },
+    {
+      "epoch": 27.3,
+      "learning_rate": 2.724e-05,
+      "loss": 0.2382,
+      "step": 1092
+    },
+    {
+      "epoch": 27.32,
+      "learning_rate": 2.7210000000000002e-05,
+      "loss": 0.1401,
+      "step": 1093
+    },
+    {
+      "epoch": 27.35,
+      "learning_rate": 2.718e-05,
+      "loss": 0.1697,
+      "step": 1094
+    },
+    {
+      "epoch": 27.38,
+      "learning_rate": 2.7150000000000003e-05,
+      "loss": 0.2192,
+      "step": 1095
+    },
+    {
+      "epoch": 27.4,
+      "learning_rate": 2.712e-05,
+      "loss": 0.1683,
+      "step": 1096
+    },
+    {
+      "epoch": 27.43,
+      "learning_rate": 2.709e-05,
+      "loss": 0.1108,
+      "step": 1097
+    },
+    {
+      "epoch": 27.45,
+      "learning_rate": 2.7060000000000002e-05,
+      "loss": 0.0669,
+      "step": 1098
+    },
+    {
+      "epoch": 27.48,
+      "learning_rate": 2.703e-05,
+      "loss": 0.2028,
+      "step": 1099
+    },
+    {
+      "epoch": 27.5,
+      "learning_rate": 2.7000000000000002e-05,
+      "loss": 0.0538,
+      "step": 1100
+    },
+    {
+      "epoch": 27.5,
+      "eval_loss": 0.15644869208335876,
+      "eval_mean_accuracy": 0.9438489255368437,
+      "eval_mean_iou": 0.5867915271754648,
+      "eval_overall_accuracy": 0.9422081667705922,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.940876465980531,
+        0.9468213850931565
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9294100936925702,
+        0.8309644878338242
+      ],
+      "eval_runtime": 48.8799,
+      "eval_samples_per_second": 0.409,
+      "eval_steps_per_second": 0.205,
+      "step": 1100
+    },
+    {
+      "epoch": 27.52,
+      "learning_rate": 2.697e-05,
+      "loss": 0.2306,
+      "step": 1101
+    },
+    {
+      "epoch": 27.55,
+      "learning_rate": 2.6940000000000003e-05,
+      "loss": 0.0404,
+      "step": 1102
+    },
+    {
+      "epoch": 27.57,
+      "learning_rate": 2.691e-05,
+      "loss": 0.1266,
+      "step": 1103
+    },
+    {
+      "epoch": 27.6,
+      "learning_rate": 2.688e-05,
+      "loss": 0.1199,
+      "step": 1104
+    },
+    {
+      "epoch": 27.62,
+      "learning_rate": 2.6850000000000002e-05,
+      "loss": 0.0922,
+      "step": 1105
+    },
+    {
+      "epoch": 27.65,
+      "learning_rate": 2.682e-05,
+      "loss": 0.3366,
+      "step": 1106
+    },
+    {
+      "epoch": 27.68,
+      "learning_rate": 2.6790000000000003e-05,
+      "loss": 0.0693,
+      "step": 1107
+    },
+    {
+      "epoch": 27.7,
+      "learning_rate": 2.676e-05,
+      "loss": 0.2045,
+      "step": 1108
+    },
+    {
+      "epoch": 27.73,
+      "learning_rate": 2.673e-05,
+      "loss": 0.1242,
+      "step": 1109
+    },
+    {
+      "epoch": 27.75,
+      "learning_rate": 2.6700000000000002e-05,
+      "loss": 0.2138,
+      "step": 1110
+    },
+    {
+      "epoch": 27.77,
+      "learning_rate": 2.667e-05,
+      "loss": 0.0715,
+      "step": 1111
+    },
+    {
+      "epoch": 27.8,
+      "learning_rate": 2.6640000000000002e-05,
+      "loss": 0.0564,
+      "step": 1112
+    },
+    {
+      "epoch": 27.82,
+      "learning_rate": 2.661e-05,
+      "loss": 0.1358,
+      "step": 1113
+    },
+    {
+      "epoch": 27.85,
+      "learning_rate": 2.658e-05,
+      "loss": 0.2516,
+      "step": 1114
+    },
+    {
+      "epoch": 27.88,
+      "learning_rate": 2.655e-05,
+      "loss": 0.1498,
+      "step": 1115
+    },
+    {
+      "epoch": 27.9,
+      "learning_rate": 2.652e-05,
+      "loss": 0.2401,
+      "step": 1116
+    },
+    {
+      "epoch": 27.93,
+      "learning_rate": 2.6490000000000002e-05,
+      "loss": 0.137,
+      "step": 1117
+    },
+    {
+      "epoch": 27.95,
+      "learning_rate": 2.646e-05,
+      "loss": 0.1819,
+      "step": 1118
+    },
+    {
+      "epoch": 27.98,
+      "learning_rate": 2.643e-05,
+      "loss": 0.0706,
+      "step": 1119
+    },
+    {
+      "epoch": 28.0,
+      "learning_rate": 2.64e-05,
+      "loss": 0.1654,
+      "step": 1120
+    },
+    {
+      "epoch": 28.0,
+      "eval_loss": 0.1621515303850174,
+      "eval_mean_accuracy": 0.9355972044866241,
+      "eval_mean_iou": 0.5692378473169154,
+      "eval_overall_accuracy": 0.9203213794859153,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9079229522212203,
+        0.963271456752028
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9000634499807797,
+        0.8076500919699665
+      ],
+      "eval_runtime": 48.0753,
+      "eval_samples_per_second": 0.416,
+      "eval_steps_per_second": 0.208,
+      "step": 1120
+    },
+    {
+      "epoch": 28.02,
+      "learning_rate": 2.637e-05,
+      "loss": 0.2287,
+      "step": 1121
+    },
+    {
+      "epoch": 28.05,
+      "learning_rate": 2.6340000000000002e-05,
+      "loss": 0.1468,
+      "step": 1122
+    },
+    {
+      "epoch": 28.07,
+      "learning_rate": 2.631e-05,
+      "loss": 0.1446,
+      "step": 1123
+    },
+    {
+      "epoch": 28.1,
+      "learning_rate": 2.628e-05,
+      "loss": 0.092,
+      "step": 1124
+    },
+    {
+      "epoch": 28.12,
+      "learning_rate": 2.625e-05,
+      "loss": 0.2824,
+      "step": 1125
+    },
+    {
+      "epoch": 28.15,
+      "learning_rate": 2.622e-05,
+      "loss": 0.1425,
+      "step": 1126
+    },
+    {
+      "epoch": 28.18,
+      "learning_rate": 2.619e-05,
+      "loss": 0.0563,
+      "step": 1127
+    },
+    {
+      "epoch": 28.2,
+      "learning_rate": 2.616e-05,
+      "loss": 0.1111,
+      "step": 1128
+    },
+    {
+      "epoch": 28.23,
+      "learning_rate": 2.6130000000000002e-05,
+      "loss": 0.1351,
+      "step": 1129
+    },
+    {
+      "epoch": 28.25,
+      "learning_rate": 2.61e-05,
+      "loss": 0.1631,
+      "step": 1130
+    },
+    {
+      "epoch": 28.27,
+      "learning_rate": 2.607e-05,
+      "loss": 0.2917,
+      "step": 1131
+    },
+    {
+      "epoch": 28.3,
+      "learning_rate": 2.604e-05,
+      "loss": 0.1254,
+      "step": 1132
+    },
+    {
+      "epoch": 28.32,
+      "learning_rate": 2.601e-05,
+      "loss": 0.0973,
+      "step": 1133
+    },
+    {
+      "epoch": 28.35,
+      "learning_rate": 2.5980000000000002e-05,
+      "loss": 0.0614,
+      "step": 1134
+    },
+    {
+      "epoch": 28.38,
+      "learning_rate": 2.595e-05,
+      "loss": 0.251,
+      "step": 1135
+    },
+    {
+      "epoch": 28.4,
+      "learning_rate": 2.592e-05,
+      "loss": 0.0512,
+      "step": 1136
+    },
+    {
+      "epoch": 28.43,
+      "learning_rate": 2.589e-05,
+      "loss": 0.1044,
+      "step": 1137
+    },
+    {
+      "epoch": 28.45,
+      "learning_rate": 2.586e-05,
+      "loss": 0.1466,
+      "step": 1138
+    },
+    {
+      "epoch": 28.48,
+      "learning_rate": 2.5830000000000002e-05,
+      "loss": 0.1983,
+      "step": 1139
+    },
+    {
+      "epoch": 28.5,
+      "learning_rate": 2.58e-05,
+      "loss": 0.1489,
+      "step": 1140
+    },
+    {
+      "epoch": 28.5,
+      "eval_loss": 0.15565548837184906,
+      "eval_mean_accuracy": 0.9209178913121763,
+      "eval_mean_iou": 0.5668986820631488,
+      "eval_overall_accuracy": 0.9245754764862553,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9275441085400488,
+        0.9142916740843039
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9072242990796557,
+        0.7934717471097905
+      ],
+      "eval_runtime": 48.5499,
+      "eval_samples_per_second": 0.412,
+      "eval_steps_per_second": 0.206,
+      "step": 1140
+    },
+    {
+      "epoch": 28.52,
+      "learning_rate": 2.577e-05,
+      "loss": 0.0775,
+      "step": 1141
+    },
+    {
+      "epoch": 28.55,
+      "learning_rate": 2.574e-05,
+      "loss": 0.1085,
+      "step": 1142
+    },
+    {
+      "epoch": 28.57,
+      "learning_rate": 2.571e-05,
+      "loss": 0.0434,
+      "step": 1143
+    },
+    {
+      "epoch": 28.6,
+      "learning_rate": 2.568e-05,
+      "loss": 0.0625,
+      "step": 1144
+    },
+    {
+      "epoch": 28.62,
+      "learning_rate": 2.565e-05,
+      "loss": 0.1077,
+      "step": 1145
+    },
+    {
+      "epoch": 28.65,
+      "learning_rate": 2.562e-05,
+      "loss": 0.0601,
+      "step": 1146
+    },
+    {
+      "epoch": 28.68,
+      "learning_rate": 2.559e-05,
+      "loss": 0.0678,
+      "step": 1147
+    },
+    {
+      "epoch": 28.7,
+      "learning_rate": 2.556e-05,
+      "loss": 0.1832,
+      "step": 1148
+    },
+    {
+      "epoch": 28.73,
+      "learning_rate": 2.553e-05,
+      "loss": 0.1015,
+      "step": 1149
+    },
+    {
+      "epoch": 28.75,
+      "learning_rate": 2.55e-05,
+      "loss": 0.2185,
+      "step": 1150
+    },
+    {
+      "epoch": 28.77,
+      "learning_rate": 2.547e-05,
+      "loss": 0.1194,
+      "step": 1151
+    },
+    {
+      "epoch": 28.8,
+      "learning_rate": 2.544e-05,
+      "loss": 0.0534,
+      "step": 1152
+    },
+    {
+      "epoch": 28.82,
+      "learning_rate": 2.541e-05,
+      "loss": 0.144,
+      "step": 1153
+    },
+    {
+      "epoch": 28.85,
+      "learning_rate": 2.538e-05,
+      "loss": 0.039,
+      "step": 1154
+    },
+    {
+      "epoch": 28.88,
+      "learning_rate": 2.535e-05,
+      "loss": 0.0868,
+      "step": 1155
+    },
+    {
+      "epoch": 28.9,
+      "learning_rate": 2.5319999999999998e-05,
+      "loss": 1.4289,
+      "step": 1156
+    },
+    {
+      "epoch": 28.93,
+      "learning_rate": 2.529e-05,
+      "loss": 0.1481,
+      "step": 1157
+    },
+    {
+      "epoch": 28.95,
+      "learning_rate": 2.526e-05,
+      "loss": 0.2125,
+      "step": 1158
+    },
+    {
+      "epoch": 28.98,
+      "learning_rate": 2.523e-05,
+      "loss": 0.0757,
+      "step": 1159
+    },
+    {
+      "epoch": 29.0,
+      "learning_rate": 2.52e-05,
+      "loss": 0.1042,
+      "step": 1160
+    },
+    {
+      "epoch": 29.0,
+      "eval_loss": 0.15596379339694977,
+      "eval_mean_accuracy": 0.9337504296205468,
+      "eval_mean_iou": 0.5750669188738581,
+      "eval_overall_accuracy": 0.9230653344869868,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9143929141873538,
+        0.9531079450537396
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9047500915523738,
+        0.8204506650692003
+      ],
+      "eval_runtime": 49.0758,
+      "eval_samples_per_second": 0.408,
+      "eval_steps_per_second": 0.204,
+      "step": 1160
+    },
+    {
+      "epoch": 29.02,
+      "learning_rate": 2.517e-05,
+      "loss": 0.1404,
+      "step": 1161
+    },
+    {
+      "epoch": 29.05,
+      "learning_rate": 2.514e-05,
+      "loss": 0.1236,
+      "step": 1162
+    },
+    {
+      "epoch": 29.07,
+      "learning_rate": 2.511e-05,
+      "loss": 0.1096,
+      "step": 1163
+    },
+    {
+      "epoch": 29.1,
+      "learning_rate": 2.508e-05,
+      "loss": 0.0628,
+      "step": 1164
+    },
+    {
+      "epoch": 29.12,
+      "learning_rate": 2.505e-05,
+      "loss": 0.0571,
+      "step": 1165
+    },
+    {
+      "epoch": 29.15,
+      "learning_rate": 2.502e-05,
+      "loss": 0.0366,
+      "step": 1166
+    },
+    {
+      "epoch": 29.18,
+      "learning_rate": 2.499e-05,
+      "loss": 0.2344,
+      "step": 1167
+    },
+    {
+      "epoch": 29.2,
+      "learning_rate": 2.4959999999999998e-05,
+      "loss": 0.1239,
+      "step": 1168
+    },
+    {
+      "epoch": 29.23,
+      "learning_rate": 2.493e-05,
+      "loss": 0.0831,
+      "step": 1169
+    },
+    {
+      "epoch": 29.25,
+      "learning_rate": 2.49e-05,
+      "loss": 0.2479,
+      "step": 1170
+    },
+    {
+      "epoch": 29.27,
+      "learning_rate": 2.487e-05,
+      "loss": 0.1274,
+      "step": 1171
+    },
+    {
+      "epoch": 29.3,
+      "learning_rate": 2.484e-05,
+      "loss": 0.1368,
+      "step": 1172
+    },
+    {
+      "epoch": 29.32,
+      "learning_rate": 2.4809999999999998e-05,
+      "loss": 0.0378,
+      "step": 1173
+    },
+    {
+      "epoch": 29.35,
+      "learning_rate": 2.478e-05,
+      "loss": 0.0683,
+      "step": 1174
+    },
+    {
+      "epoch": 29.38,
+      "learning_rate": 2.475e-05,
+      "loss": 0.1213,
+      "step": 1175
+    },
+    {
+      "epoch": 29.4,
+      "learning_rate": 2.472e-05,
+      "loss": 0.0405,
+      "step": 1176
+    },
+    {
+      "epoch": 29.43,
+      "learning_rate": 2.469e-05,
+      "loss": 0.1525,
+      "step": 1177
+    },
+    {
+      "epoch": 29.45,
+      "learning_rate": 2.4659999999999998e-05,
+      "loss": 0.2468,
+      "step": 1178
+    },
+    {
+      "epoch": 29.48,
+      "learning_rate": 2.463e-05,
+      "loss": 0.0625,
+      "step": 1179
+    },
+    {
+      "epoch": 29.5,
+      "learning_rate": 2.4599999999999998e-05,
+      "loss": 0.2673,
+      "step": 1180
+    },
+    {
+      "epoch": 29.5,
+      "eval_loss": 0.14658869802951813,
+      "eval_mean_accuracy": 0.9183105779793636,
+      "eval_mean_iou": 0.5774838423825798,
+      "eval_overall_accuracy": 0.9306454497556542,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9406568900890294,
+        0.8959642658696977
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.915568638090893,
+        0.8168828890568463
+      ],
+      "eval_runtime": 48.0493,
+      "eval_samples_per_second": 0.416,
+      "eval_steps_per_second": 0.208,
+      "step": 1180
+    },
+    {
+      "epoch": 29.52,
+      "learning_rate": 2.457e-05,
+      "loss": 0.0919,
+      "step": 1181
+    },
+    {
+      "epoch": 29.55,
+      "learning_rate": 2.454e-05,
+      "loss": 0.1101,
+      "step": 1182
+    },
+    {
+      "epoch": 29.57,
+      "learning_rate": 2.4509999999999997e-05,
+      "loss": 0.1841,
+      "step": 1183
+    },
+    {
+      "epoch": 29.6,
+      "learning_rate": 2.448e-05,
+      "loss": 0.0482,
+      "step": 1184
+    },
+    {
+      "epoch": 29.62,
+      "learning_rate": 2.4449999999999998e-05,
+      "loss": 0.047,
+      "step": 1185
+    },
+    {
+      "epoch": 29.65,
+      "learning_rate": 2.442e-05,
+      "loss": 0.0609,
+      "step": 1186
+    },
+    {
+      "epoch": 29.68,
+      "learning_rate": 2.439e-05,
+      "loss": 0.038,
+      "step": 1187
+    },
+    {
+      "epoch": 29.7,
+      "learning_rate": 2.4360000000000004e-05,
+      "loss": 0.1407,
+      "step": 1188
+    },
+    {
+      "epoch": 29.73,
+      "learning_rate": 2.4330000000000003e-05,
+      "loss": 0.0914,
+      "step": 1189
+    },
+    {
+      "epoch": 29.75,
+      "learning_rate": 2.43e-05,
+      "loss": 0.2476,
+      "step": 1190
+    },
+    {
+      "epoch": 29.77,
+      "learning_rate": 2.4270000000000003e-05,
+      "loss": 0.0884,
+      "step": 1191
+    },
+    {
+      "epoch": 29.8,
+      "learning_rate": 2.4240000000000002e-05,
+      "loss": 0.172,
+      "step": 1192
+    },
+    {
+      "epoch": 29.82,
+      "learning_rate": 2.4210000000000004e-05,
+      "loss": 0.0579,
+      "step": 1193
+    },
+    {
+      "epoch": 29.85,
+      "learning_rate": 2.4180000000000002e-05,
+      "loss": 0.1086,
+      "step": 1194
+    },
+    {
+      "epoch": 29.88,
+      "learning_rate": 2.415e-05,
+      "loss": 0.0723,
+      "step": 1195
+    },
+    {
+      "epoch": 29.9,
+      "learning_rate": 2.4120000000000003e-05,
+      "loss": 0.0915,
+      "step": 1196
+    },
+    {
+      "epoch": 29.93,
+      "learning_rate": 2.409e-05,
+      "loss": 0.1965,
+      "step": 1197
+    },
+    {
+      "epoch": 29.95,
+      "learning_rate": 2.4060000000000003e-05,
+      "loss": 0.222,
+      "step": 1198
+    },
+    {
+      "epoch": 29.98,
+      "learning_rate": 2.4030000000000002e-05,
+      "loss": 0.1071,
+      "step": 1199
+    },
+    {
+      "epoch": 30.0,
+      "learning_rate": 2.4e-05,
+      "loss": 0.2407,
+      "step": 1200
+    },
+    {
+      "epoch": 30.0,
+      "eval_loss": 0.13716813921928406,
+      "eval_mean_accuracy": 0.9278948485402951,
+      "eval_mean_iou": 0.5774629255691858,
+      "eval_overall_accuracy": 0.9275224487099545,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9272201951681884,
+        0.9285695019124017
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9111556483707967,
+        0.8212331283367605
+      ],
+      "eval_runtime": 48.011,
+      "eval_samples_per_second": 0.417,
+      "eval_steps_per_second": 0.208,
+      "step": 1200
+    },
+    {
+      "epoch": 30.02,
+      "learning_rate": 2.3970000000000003e-05,
+      "loss": 0.1321,
+      "step": 1201
+    },
+    {
+      "epoch": 30.05,
+      "learning_rate": 2.394e-05,
+      "loss": 0.0936,
+      "step": 1202
+    },
+    {
+      "epoch": 30.07,
+      "learning_rate": 2.3910000000000003e-05,
+      "loss": 0.059,
+      "step": 1203
+    },
+    {
+      "epoch": 30.1,
+      "learning_rate": 2.3880000000000002e-05,
+      "loss": 0.1303,
+      "step": 1204
+    },
+    {
+      "epoch": 30.12,
+      "learning_rate": 2.385e-05,
+      "loss": 0.1556,
+      "step": 1205
+    },
+    {
+      "epoch": 30.15,
+      "learning_rate": 2.3820000000000002e-05,
+      "loss": 0.9338,
+      "step": 1206
+    },
+    {
+      "epoch": 30.18,
+      "learning_rate": 2.379e-05,
+      "loss": 0.2618,
+      "step": 1207
+    },
+    {
+      "epoch": 30.2,
+      "learning_rate": 2.3760000000000003e-05,
+      "loss": 0.3154,
+      "step": 1208
+    },
+    {
+      "epoch": 30.23,
+      "learning_rate": 2.373e-05,
+      "loss": 0.1943,
+      "step": 1209
+    },
+    {
+      "epoch": 30.25,
+      "learning_rate": 2.37e-05,
+      "loss": 0.0261,
+      "step": 1210
+    },
+    {
+      "epoch": 30.27,
+      "learning_rate": 2.3670000000000002e-05,
+      "loss": 0.1178,
+      "step": 1211
+    },
+    {
+      "epoch": 30.3,
+      "learning_rate": 2.364e-05,
+      "loss": 0.0772,
+      "step": 1212
+    },
+    {
+      "epoch": 30.32,
+      "learning_rate": 2.3610000000000003e-05,
+      "loss": 0.2255,
+      "step": 1213
+    },
+    {
+      "epoch": 30.35,
+      "learning_rate": 2.358e-05,
+      "loss": 0.1229,
+      "step": 1214
+    },
+    {
+      "epoch": 30.38,
+      "learning_rate": 2.3550000000000003e-05,
+      "loss": 0.0847,
+      "step": 1215
+    },
+    {
+      "epoch": 30.4,
+      "learning_rate": 2.3520000000000002e-05,
+      "loss": 0.1285,
+      "step": 1216
+    },
+    {
+      "epoch": 30.43,
+      "learning_rate": 2.349e-05,
+      "loss": 0.1394,
+      "step": 1217
+    },
+    {
+      "epoch": 30.45,
+      "learning_rate": 2.3460000000000002e-05,
+      "loss": 0.1079,
+      "step": 1218
+    },
+    {
+      "epoch": 30.48,
+      "learning_rate": 2.343e-05,
+      "loss": 0.08,
+      "step": 1219
+    },
+    {
+      "epoch": 30.5,
+      "learning_rate": 2.3400000000000003e-05,
+      "loss": 0.156,
+      "step": 1220
+    },
+    {
+      "epoch": 30.5,
+      "eval_loss": 0.14172598719596863,
+      "eval_mean_accuracy": 0.9438137205101318,
+      "eval_mean_iou": 0.5857059186137794,
+      "eval_overall_accuracy": 0.9402666131732398,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9373876490377853,
+        0.9502397919824782
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9253977014462942,
+        0.831720054395044
+      ],
+      "eval_runtime": 48.2687,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 1220
+    },
+    {
+      "epoch": 30.52,
+      "learning_rate": 2.337e-05,
+      "loss": 0.1637,
+      "step": 1221
+    },
+    {
+      "epoch": 30.55,
+      "learning_rate": 2.334e-05,
+      "loss": 0.0296,
+      "step": 1222
+    },
+    {
+      "epoch": 30.57,
+      "learning_rate": 2.3310000000000002e-05,
+      "loss": 0.0958,
+      "step": 1223
+    },
+    {
+      "epoch": 30.6,
+      "learning_rate": 2.328e-05,
+      "loss": 0.3513,
+      "step": 1224
+    },
+    {
+      "epoch": 30.62,
+      "learning_rate": 2.3250000000000003e-05,
+      "loss": 0.0307,
+      "step": 1225
+    },
+    {
+      "epoch": 30.65,
+      "learning_rate": 2.322e-05,
+      "loss": 0.0679,
+      "step": 1226
+    },
+    {
+      "epoch": 30.68,
+      "learning_rate": 2.319e-05,
+      "loss": 0.0917,
+      "step": 1227
+    },
+    {
+      "epoch": 30.7,
+      "learning_rate": 2.3160000000000002e-05,
+      "loss": 0.0461,
+      "step": 1228
+    },
+    {
+      "epoch": 30.73,
+      "learning_rate": 2.313e-05,
+      "loss": 0.0671,
+      "step": 1229
+    },
+    {
+      "epoch": 30.75,
+      "learning_rate": 2.3100000000000002e-05,
+      "loss": 0.1705,
+      "step": 1230
+    },
+    {
+      "epoch": 30.77,
+      "learning_rate": 2.307e-05,
+      "loss": 0.102,
+      "step": 1231
+    },
+    {
+      "epoch": 30.8,
+      "learning_rate": 2.304e-05,
+      "loss": 0.0605,
+      "step": 1232
+    },
+    {
+      "epoch": 30.82,
+      "learning_rate": 2.301e-05,
+      "loss": 0.0652,
+      "step": 1233
+    },
+    {
+      "epoch": 30.85,
+      "learning_rate": 2.298e-05,
+      "loss": 0.1386,
+      "step": 1234
+    },
+    {
+      "epoch": 30.88,
+      "learning_rate": 2.2950000000000002e-05,
+      "loss": 0.1284,
+      "step": 1235
+    },
+    {
+      "epoch": 30.9,
+      "learning_rate": 2.292e-05,
+      "loss": 0.2326,
+      "step": 1236
+    },
+    {
+      "epoch": 30.93,
+      "learning_rate": 2.289e-05,
+      "loss": 0.0967,
+      "step": 1237
+    },
+    {
+      "epoch": 30.95,
+      "learning_rate": 2.286e-05,
+      "loss": 0.0923,
+      "step": 1238
+    },
+    {
+      "epoch": 30.98,
+      "learning_rate": 2.283e-05,
+      "loss": 0.0836,
+      "step": 1239
+    },
+    {
+      "epoch": 31.0,
+      "learning_rate": 2.2800000000000002e-05,
+      "loss": 0.2055,
+      "step": 1240
+    },
+    {
+      "epoch": 31.0,
+      "eval_loss": 0.13539931178092957,
+      "eval_mean_accuracy": 0.9464552948800316,
+      "eval_mean_iou": 0.587195085273711,
+      "eval_overall_accuracy": 0.9394420571122895,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9337498527076614,
+        0.9591607370524018
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9252739587919888,
+        0.8363112970291441
+      ],
+      "eval_runtime": 48.8168,
+      "eval_samples_per_second": 0.41,
+      "eval_steps_per_second": 0.205,
+      "step": 1240
+    },
+    {
+      "epoch": 31.02,
+      "learning_rate": 2.277e-05,
+      "loss": 0.0609,
+      "step": 1241
+    },
+    {
+      "epoch": 31.05,
+      "learning_rate": 2.274e-05,
+      "loss": 0.0757,
+      "step": 1242
+    },
+    {
+      "epoch": 31.07,
+      "learning_rate": 2.271e-05,
+      "loss": 0.0706,
+      "step": 1243
+    },
+    {
+      "epoch": 31.1,
+      "learning_rate": 2.268e-05,
+      "loss": 0.0559,
+      "step": 1244
+    },
+    {
+      "epoch": 31.12,
+      "learning_rate": 2.265e-05,
+      "loss": 0.1757,
+      "step": 1245
+    },
+    {
+      "epoch": 31.15,
+      "learning_rate": 2.262e-05,
+      "loss": 0.0531,
+      "step": 1246
+    },
+    {
+      "epoch": 31.18,
+      "learning_rate": 2.2590000000000002e-05,
+      "loss": 0.19,
+      "step": 1247
+    },
+    {
+      "epoch": 31.2,
+      "learning_rate": 2.256e-05,
+      "loss": 0.0815,
+      "step": 1248
+    },
+    {
+      "epoch": 31.23,
+      "learning_rate": 2.253e-05,
+      "loss": 0.2507,
+      "step": 1249
+    },
+    {
+      "epoch": 31.25,
+      "learning_rate": 2.25e-05,
+      "loss": 0.0954,
+      "step": 1250
+    },
+    {
+      "epoch": 31.27,
+      "learning_rate": 2.247e-05,
+      "loss": 0.366,
+      "step": 1251
+    },
+    {
+      "epoch": 31.3,
+      "learning_rate": 2.2440000000000002e-05,
+      "loss": 0.2605,
+      "step": 1252
+    },
+    {
+      "epoch": 31.32,
+      "learning_rate": 2.241e-05,
+      "loss": 0.2055,
+      "step": 1253
+    },
+    {
+      "epoch": 31.35,
+      "learning_rate": 2.238e-05,
+      "loss": 0.1824,
+      "step": 1254
+    },
+    {
+      "epoch": 31.38,
+      "learning_rate": 2.235e-05,
+      "loss": 0.0568,
+      "step": 1255
+    },
+    {
+      "epoch": 31.4,
+      "learning_rate": 2.232e-05,
+      "loss": 0.0994,
+      "step": 1256
+    },
+    {
+      "epoch": 31.43,
+      "learning_rate": 2.2290000000000002e-05,
+      "loss": 0.0737,
+      "step": 1257
+    },
+    {
+      "epoch": 31.45,
+      "learning_rate": 2.226e-05,
+      "loss": 0.1399,
+      "step": 1258
+    },
+    {
+      "epoch": 31.48,
+      "learning_rate": 2.223e-05,
+      "loss": 0.182,
+      "step": 1259
+    },
+    {
+      "epoch": 31.5,
+      "learning_rate": 2.22e-05,
+      "loss": 0.1991,
+      "step": 1260
+    },
+    {
+      "epoch": 31.5,
+      "eval_loss": 0.19345512986183167,
+      "eval_mean_accuracy": 0.9347760583245306,
+      "eval_mean_iou": 0.5441659817453899,
+      "eval_overall_accuracy": 0.9125732916342403,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.894552701068966,
+        0.9749994155800952
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8895013422759901,
+        0.7429966029601797
+      ],
+      "eval_runtime": 48.2394,
+      "eval_samples_per_second": 0.415,
+      "eval_steps_per_second": 0.207,
+      "step": 1260
+    },
+    {
+      "epoch": 31.52,
+      "learning_rate": 2.217e-05,
+      "loss": 0.1023,
+      "step": 1261
+    },
+    {
+      "epoch": 31.55,
+      "learning_rate": 2.214e-05,
+      "loss": 0.0606,
+      "step": 1262
+    },
+    {
+      "epoch": 31.57,
+      "learning_rate": 2.211e-05,
+      "loss": 0.1546,
+      "step": 1263
+    },
+    {
+      "epoch": 31.6,
+      "learning_rate": 2.208e-05,
+      "loss": 0.0425,
+      "step": 1264
+    },
+    {
+      "epoch": 31.62,
+      "learning_rate": 2.205e-05,
+      "loss": 0.1535,
+      "step": 1265
+    },
+    {
+      "epoch": 31.65,
+      "learning_rate": 2.202e-05,
+      "loss": 0.1127,
+      "step": 1266
+    },
+    {
+      "epoch": 31.68,
+      "learning_rate": 2.199e-05,
+      "loss": 0.2647,
+      "step": 1267
+    },
+    {
+      "epoch": 31.7,
+      "learning_rate": 2.196e-05,
+      "loss": 0.2571,
+      "step": 1268
+    },
+    {
+      "epoch": 31.73,
+      "learning_rate": 2.193e-05,
+      "loss": 0.1019,
+      "step": 1269
+    },
+    {
+      "epoch": 31.75,
+      "learning_rate": 2.19e-05,
+      "loss": 0.0443,
+      "step": 1270
+    },
+    {
+      "epoch": 31.77,
+      "learning_rate": 2.187e-05,
+      "loss": 0.1379,
+      "step": 1271
+    },
+    {
+      "epoch": 31.8,
+      "learning_rate": 2.184e-05,
+      "loss": 0.2044,
+      "step": 1272
+    },
+    {
+      "epoch": 31.82,
+      "learning_rate": 2.181e-05,
+      "loss": 0.0275,
+      "step": 1273
+    },
+    {
+      "epoch": 31.85,
+      "learning_rate": 2.178e-05,
+      "loss": 0.0409,
+      "step": 1274
+    },
+    {
+      "epoch": 31.88,
+      "learning_rate": 2.175e-05,
+      "loss": 0.068,
+      "step": 1275
+    },
+    {
+      "epoch": 31.9,
+      "learning_rate": 2.172e-05,
+      "loss": 0.0255,
+      "step": 1276
+    },
+    {
+      "epoch": 31.93,
+      "learning_rate": 2.169e-05,
+      "loss": 0.1377,
+      "step": 1277
+    },
+    {
+      "epoch": 31.95,
+      "learning_rate": 2.166e-05,
+      "loss": 0.1298,
+      "step": 1278
+    },
+    {
+      "epoch": 31.98,
+      "learning_rate": 2.163e-05,
+      "loss": 0.1004,
+      "step": 1279
+    },
+    {
+      "epoch": 32.0,
+      "learning_rate": 2.16e-05,
+      "loss": 0.0246,
+      "step": 1280
+    },
+    {
+      "epoch": 32.0,
+      "eval_loss": 0.13773247599601746,
+      "eval_mean_accuracy": 0.9310309811807711,
+      "eval_mean_iou": 0.5852262909654686,
+      "eval_overall_accuracy": 0.9354260219741974,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9389932004146506,
+        0.9230687619468915
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9211641944119524,
+        0.8345146784844534
+      ],
+      "eval_runtime": 48.3932,
+      "eval_samples_per_second": 0.413,
+      "eval_steps_per_second": 0.207,
+      "step": 1280
+    },
+    {
+      "epoch": 32.02,
+      "learning_rate": 2.157e-05,
+      "loss": 0.1107,
+      "step": 1281
+    },
+    {
+      "epoch": 32.05,
+      "learning_rate": 2.154e-05,
+      "loss": 0.1925,
+      "step": 1282
+    },
+    {
+      "epoch": 32.08,
+      "learning_rate": 2.151e-05,
+      "loss": 0.1466,
+      "step": 1283
+    },
+    {
+      "epoch": 32.1,
+      "learning_rate": 2.148e-05,
+      "loss": 0.0485,
+      "step": 1284
+    },
+    {
+      "epoch": 32.12,
+      "learning_rate": 2.145e-05,
+      "loss": 0.1816,
+      "step": 1285
+    },
+    {
+      "epoch": 32.15,
+      "learning_rate": 2.1419999999999998e-05,
+      "loss": 0.1174,
+      "step": 1286
+    },
+    {
+      "epoch": 32.17,
+      "learning_rate": 2.139e-05,
+      "loss": 0.4293,
+      "step": 1287
+    },
+    {
+      "epoch": 32.2,
+      "learning_rate": 2.136e-05,
+      "loss": 0.1377,
+      "step": 1288
+    },
+    {
+      "epoch": 32.23,
+      "learning_rate": 2.133e-05,
+      "loss": 0.0382,
+      "step": 1289
+    },
+    {
+      "epoch": 32.25,
+      "learning_rate": 2.13e-05,
+      "loss": 0.0932,
+      "step": 1290
+    },
+    {
+      "epoch": 32.27,
+      "learning_rate": 2.1269999999999998e-05,
+      "loss": 0.1005,
+      "step": 1291
+    },
+    {
+      "epoch": 32.3,
+      "learning_rate": 2.124e-05,
+      "loss": 0.1093,
+      "step": 1292
+    },
+    {
+      "epoch": 32.33,
+      "learning_rate": 2.121e-05,
+      "loss": 0.229,
+      "step": 1293
+    },
+    {
+      "epoch": 32.35,
+      "learning_rate": 2.118e-05,
+      "loss": 0.1214,
+      "step": 1294
+    },
+    {
+      "epoch": 32.38,
+      "learning_rate": 2.115e-05,
+      "loss": 0.0534,
+      "step": 1295
+    },
+    {
+      "epoch": 32.4,
+      "learning_rate": 2.1119999999999998e-05,
+      "loss": 0.0326,
+      "step": 1296
+    },
+    {
+      "epoch": 32.42,
+      "learning_rate": 2.109e-05,
+      "loss": 0.0355,
+      "step": 1297
+    },
+    {
+      "epoch": 32.45,
+      "learning_rate": 2.1059999999999998e-05,
+      "loss": 0.1252,
+      "step": 1298
+    },
+    {
+      "epoch": 32.48,
+      "learning_rate": 2.103e-05,
+      "loss": 0.1066,
+      "step": 1299
+    },
+    {
+      "epoch": 32.5,
+      "learning_rate": 2.1e-05,
+      "loss": 0.1121,
+      "step": 1300
+    },
+    {
+      "epoch": 32.5,
+      "eval_loss": 0.15950095653533936,
+      "eval_mean_accuracy": 0.9427759583816913,
+      "eval_mean_iou": 0.5842666722586219,
+      "eval_overall_accuracy": 0.935530350269159,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.929649545457141,
+        0.9559023713062414
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9197659954949357,
+        0.8330340212809302
+      ],
+      "eval_runtime": 49.099,
+      "eval_samples_per_second": 0.407,
+      "eval_steps_per_second": 0.204,
+      "step": 1300
+    },
+    {
+      "epoch": 32.52,
+      "learning_rate": 2.097e-05,
+      "loss": 0.1008,
+      "step": 1301
+    },
+    {
+      "epoch": 32.55,
+      "learning_rate": 2.094e-05,
+      "loss": 0.1884,
+      "step": 1302
+    },
+    {
+      "epoch": 32.58,
+      "learning_rate": 2.0909999999999998e-05,
+      "loss": 0.0282,
+      "step": 1303
+    },
+    {
+      "epoch": 32.6,
+      "learning_rate": 2.088e-05,
+      "loss": 0.0394,
+      "step": 1304
+    },
+    {
+      "epoch": 32.62,
+      "learning_rate": 2.085e-05,
+      "loss": 0.055,
+      "step": 1305
+    },
+    {
+      "epoch": 32.65,
+      "learning_rate": 2.082e-05,
+      "loss": 0.0894,
+      "step": 1306
+    },
+    {
+      "epoch": 32.67,
+      "learning_rate": 2.079e-05,
+      "loss": 0.0912,
+      "step": 1307
+    },
+    {
+      "epoch": 32.7,
+      "learning_rate": 2.0759999999999998e-05,
+      "loss": 0.1385,
+      "step": 1308
+    },
+    {
+      "epoch": 32.73,
+      "learning_rate": 2.073e-05,
+      "loss": 0.0427,
+      "step": 1309
+    },
+    {
+      "epoch": 32.75,
+      "learning_rate": 2.07e-05,
+      "loss": 0.1541,
+      "step": 1310
+    },
+    {
+      "epoch": 32.77,
+      "learning_rate": 2.067e-05,
+      "loss": 0.151,
+      "step": 1311
+    },
+    {
+      "epoch": 32.8,
+      "learning_rate": 2.064e-05,
+      "loss": 0.067,
+      "step": 1312
+    },
+    {
+      "epoch": 32.83,
+      "learning_rate": 2.061e-05,
+      "loss": 0.1293,
+      "step": 1313
+    },
+    {
+      "epoch": 32.85,
+      "learning_rate": 2.0580000000000003e-05,
+      "loss": 0.0954,
+      "step": 1314
+    },
+    {
+      "epoch": 32.88,
+      "learning_rate": 2.055e-05,
+      "loss": 0.2732,
+      "step": 1315
+    },
+    {
+      "epoch": 32.9,
+      "learning_rate": 2.0520000000000003e-05,
+      "loss": 0.098,
+      "step": 1316
+    },
+    {
+      "epoch": 32.92,
+      "learning_rate": 2.0490000000000002e-05,
+      "loss": 0.065,
+      "step": 1317
+    },
+    {
+      "epoch": 32.95,
+      "learning_rate": 2.046e-05,
+      "loss": 0.0508,
+      "step": 1318
+    },
+    {
+      "epoch": 32.98,
+      "learning_rate": 2.0430000000000003e-05,
+      "loss": 0.0714,
+      "step": 1319
+    },
+    {
+      "epoch": 33.0,
+      "learning_rate": 2.04e-05,
+      "loss": 0.1412,
+      "step": 1320
+    },
+    {
+      "epoch": 33.0,
+      "eval_loss": 0.15888680517673492,
+      "eval_mean_accuracy": 0.9359989685387133,
+      "eval_mean_iou": 0.5682434711799925,
+      "eval_overall_accuracy": 0.9230065239423366,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9124613730708597,
+        0.9595365640065671
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9037129387329121,
+        0.8010174748070655
+      ],
+      "eval_runtime": 48.665,
+      "eval_samples_per_second": 0.411,
+      "eval_steps_per_second": 0.205,
+      "step": 1320
+    },
+    {
+      "epoch": 33.02,
+      "learning_rate": 2.0370000000000003e-05,
+      "loss": 0.0426,
+      "step": 1321
+    },
+    {
+      "epoch": 33.05,
+      "learning_rate": 2.0340000000000002e-05,
+      "loss": 0.1127,
+      "step": 1322
+    },
+    {
+      "epoch": 33.08,
+      "learning_rate": 2.031e-05,
+      "loss": 0.0802,
+      "step": 1323
+    },
+    {
+      "epoch": 33.1,
+      "learning_rate": 2.0280000000000002e-05,
+      "loss": 0.1685,
+      "step": 1324
+    },
+    {
+      "epoch": 33.12,
+      "learning_rate": 2.025e-05,
+      "loss": 0.1417,
+      "step": 1325
+    },
+    {
+      "epoch": 33.15,
+      "learning_rate": 2.0220000000000003e-05,
+      "loss": 0.0906,
+      "step": 1326
+    },
+    {
+      "epoch": 33.17,
+      "learning_rate": 2.019e-05,
+      "loss": 0.1718,
+      "step": 1327
+    },
+    {
+      "epoch": 33.2,
+      "learning_rate": 2.016e-05,
+      "loss": 0.1187,
+      "step": 1328
+    },
+    {
+      "epoch": 33.23,
+      "learning_rate": 2.0130000000000002e-05,
+      "loss": 0.2979,
+      "step": 1329
+    },
+    {
+      "epoch": 33.25,
+      "learning_rate": 2.01e-05,
+      "loss": 0.1852,
+      "step": 1330
+    },
+    {
+      "epoch": 33.27,
+      "learning_rate": 2.0070000000000003e-05,
+      "loss": 0.0292,
+      "step": 1331
+    },
+    {
+      "epoch": 33.3,
+      "learning_rate": 2.004e-05,
+      "loss": 0.0987,
+      "step": 1332
+    },
+    {
+      "epoch": 33.33,
+      "learning_rate": 2.0010000000000003e-05,
+      "loss": 0.097,
+      "step": 1333
+    },
+    {
+      "epoch": 33.35,
+      "learning_rate": 1.9980000000000002e-05,
+      "loss": 0.0652,
+      "step": 1334
+    },
+    {
+      "epoch": 33.38,
+      "learning_rate": 1.995e-05,
+      "loss": 0.1437,
+      "step": 1335
+    },
+    {
+      "epoch": 33.4,
+      "learning_rate": 1.9920000000000002e-05,
+      "loss": 0.2821,
+      "step": 1336
+    },
+    {
+      "epoch": 33.42,
+      "learning_rate": 1.989e-05,
+      "loss": 0.134,
+      "step": 1337
+    },
+    {
+      "epoch": 33.45,
+      "learning_rate": 1.9860000000000003e-05,
+      "loss": 0.1025,
+      "step": 1338
+    },
+    {
+      "epoch": 33.48,
+      "learning_rate": 1.983e-05,
+      "loss": 0.0547,
+      "step": 1339
+    },
+    {
+      "epoch": 33.5,
+      "learning_rate": 1.98e-05,
+      "loss": 0.1543,
+      "step": 1340
+    },
+    {
+      "epoch": 33.5,
+      "eval_loss": 0.15789881348609924,
+      "eval_mean_accuracy": 0.9333310262617378,
+      "eval_mean_iou": 0.578741288722326,
+      "eval_overall_accuracy": 0.9306966068732608,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9285584141948806,
+        0.938103638328595
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9141493842516836,
+        0.8220744819152943
+      ],
+      "eval_runtime": 48.3937,
+      "eval_samples_per_second": 0.413,
+      "eval_steps_per_second": 0.207,
+      "step": 1340
+    },
+    {
+      "epoch": 33.52,
+      "learning_rate": 1.9770000000000002e-05,
+      "loss": 0.0497,
+      "step": 1341
+    },
+    {
+      "epoch": 33.55,
+      "learning_rate": 1.974e-05,
+      "loss": 0.0928,
+      "step": 1342
+    },
+    {
+      "epoch": 33.58,
+      "learning_rate": 1.9710000000000003e-05,
+      "loss": 0.0698,
+      "step": 1343
+    },
+    {
+      "epoch": 33.6,
+      "learning_rate": 1.968e-05,
+      "loss": 0.1718,
+      "step": 1344
+    },
+    {
+      "epoch": 33.62,
+      "learning_rate": 1.965e-05,
+      "loss": 0.1448,
+      "step": 1345
+    },
+    {
+      "epoch": 33.65,
+      "learning_rate": 1.9620000000000002e-05,
+      "loss": 0.0541,
+      "step": 1346
+    },
+    {
+      "epoch": 33.67,
+      "learning_rate": 1.959e-05,
+      "loss": 0.1672,
+      "step": 1347
+    },
+    {
+      "epoch": 33.7,
+      "learning_rate": 1.9560000000000002e-05,
+      "loss": 0.1103,
+      "step": 1348
+    },
+    {
+      "epoch": 33.73,
+      "learning_rate": 1.953e-05,
+      "loss": 0.1526,
+      "step": 1349
+    },
+    {
+      "epoch": 33.75,
+      "learning_rate": 1.95e-05,
+      "loss": 0.0793,
+      "step": 1350
+    },
+    {
+      "epoch": 33.77,
+      "learning_rate": 1.947e-05,
+      "loss": 0.1565,
+      "step": 1351
+    },
+    {
+      "epoch": 33.8,
+      "learning_rate": 1.944e-05,
+      "loss": 0.1673,
+      "step": 1352
+    },
+    {
+      "epoch": 33.83,
+      "learning_rate": 1.9410000000000002e-05,
+      "loss": 0.0978,
+      "step": 1353
+    },
+    {
+      "epoch": 33.85,
+      "learning_rate": 1.938e-05,
+      "loss": 0.0977,
+      "step": 1354
+    },
+    {
+      "epoch": 33.88,
+      "learning_rate": 1.935e-05,
+      "loss": 0.065,
+      "step": 1355
+    },
+    {
+      "epoch": 33.9,
+      "learning_rate": 1.932e-05,
+      "loss": 0.1086,
+      "step": 1356
+    },
+    {
+      "epoch": 33.92,
+      "learning_rate": 1.929e-05,
+      "loss": 0.1271,
+      "step": 1357
+    },
+    {
+      "epoch": 33.95,
+      "learning_rate": 1.9260000000000002e-05,
+      "loss": 0.1243,
+      "step": 1358
+    },
+    {
+      "epoch": 33.98,
+      "learning_rate": 1.923e-05,
+      "loss": 0.2135,
+      "step": 1359
+    },
+    {
+      "epoch": 34.0,
+      "learning_rate": 1.9200000000000003e-05,
+      "loss": 0.0749,
+      "step": 1360
+    },
+    {
+      "epoch": 34.0,
+      "eval_loss": 0.22385883331298828,
+      "eval_mean_accuracy": 0.9420340200496059,
+      "eval_mean_iou": 0.565621990022045,
+      "eval_overall_accuracy": 0.9319807713687711,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9238211811314232,
+        0.9602468589677886
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9151379118844872,
+        0.7817280581816478
+      ],
+      "eval_runtime": 47.8116,
+      "eval_samples_per_second": 0.418,
+      "eval_steps_per_second": 0.209,
+      "step": 1360
+    },
+    {
+      "epoch": 34.02,
+      "learning_rate": 1.917e-05,
+      "loss": 0.0622,
+      "step": 1361
+    },
+    {
+      "epoch": 34.05,
+      "learning_rate": 1.914e-05,
+      "loss": 0.0533,
+      "step": 1362
+    },
+    {
+      "epoch": 34.08,
+      "learning_rate": 1.911e-05,
+      "loss": 0.0981,
+      "step": 1363
+    },
+    {
+      "epoch": 34.1,
+      "learning_rate": 1.908e-05,
+      "loss": 0.1633,
+      "step": 1364
+    },
+    {
+      "epoch": 34.12,
+      "learning_rate": 1.9050000000000002e-05,
+      "loss": 0.1357,
+      "step": 1365
+    },
+    {
+      "epoch": 34.15,
+      "learning_rate": 1.902e-05,
+      "loss": 0.0699,
+      "step": 1366
+    },
+    {
+      "epoch": 34.17,
+      "learning_rate": 1.899e-05,
+      "loss": 0.0902,
+      "step": 1367
+    },
+    {
+      "epoch": 34.2,
+      "learning_rate": 1.896e-05,
+      "loss": 0.1042,
+      "step": 1368
+    },
+    {
+      "epoch": 34.23,
+      "learning_rate": 1.893e-05,
+      "loss": 0.0877,
+      "step": 1369
+    },
+    {
+      "epoch": 34.25,
+      "learning_rate": 1.8900000000000002e-05,
+      "loss": 0.12,
+      "step": 1370
+    },
+    {
+      "epoch": 34.27,
+      "learning_rate": 1.887e-05,
+      "loss": 0.0861,
+      "step": 1371
+    },
+    {
+      "epoch": 34.3,
+      "learning_rate": 1.884e-05,
+      "loss": 0.1076,
+      "step": 1372
+    },
+    {
+      "epoch": 34.33,
+      "learning_rate": 1.881e-05,
+      "loss": 0.1013,
+      "step": 1373
+    },
+    {
+      "epoch": 34.35,
+      "learning_rate": 1.878e-05,
+      "loss": 0.091,
+      "step": 1374
+    },
+    {
+      "epoch": 34.38,
+      "learning_rate": 1.8750000000000002e-05,
+      "loss": 0.0832,
+      "step": 1375
+    },
+    {
+      "epoch": 34.4,
+      "learning_rate": 1.872e-05,
+      "loss": 0.1889,
+      "step": 1376
+    },
+    {
+      "epoch": 34.42,
+      "learning_rate": 1.869e-05,
+      "loss": 0.0417,
+      "step": 1377
+    },
+    {
+      "epoch": 34.45,
+      "learning_rate": 1.866e-05,
+      "loss": 0.1096,
+      "step": 1378
+    },
+    {
+      "epoch": 34.48,
+      "learning_rate": 1.863e-05,
+      "loss": 0.0575,
+      "step": 1379
+    },
+    {
+      "epoch": 34.5,
+      "learning_rate": 1.86e-05,
+      "loss": 0.0426,
+      "step": 1380
+    },
+    {
+      "epoch": 34.5,
+      "eval_loss": 0.1333775818347931,
+      "eval_mean_accuracy": 0.9364623083249072,
+      "eval_mean_iou": 0.5894436338021803,
+      "eval_overall_accuracy": 0.9347106279516045,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.933288899063091,
+        0.9396357175867234
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9206012726121113,
+        0.8477296287944296
+      ],
+      "eval_runtime": 48.4729,
+      "eval_samples_per_second": 0.413,
+      "eval_steps_per_second": 0.206,
+      "step": 1380
+    },
+    {
+      "epoch": 34.52,
+      "learning_rate": 1.857e-05,
+      "loss": 0.1903,
+      "step": 1381
+    },
+    {
+      "epoch": 34.55,
+      "learning_rate": 1.854e-05,
+      "loss": 0.3042,
+      "step": 1382
+    },
+    {
+      "epoch": 34.58,
+      "learning_rate": 1.851e-05,
+      "loss": 0.2378,
+      "step": 1383
+    },
+    {
+      "epoch": 34.6,
+      "learning_rate": 1.848e-05,
+      "loss": 0.0362,
+      "step": 1384
+    },
+    {
+      "epoch": 34.62,
+      "learning_rate": 1.845e-05,
+      "loss": 0.0629,
+      "step": 1385
+    },
+    {
+      "epoch": 34.65,
+      "learning_rate": 1.842e-05,
+      "loss": 0.061,
+      "step": 1386
+    },
+    {
+      "epoch": 34.67,
+      "learning_rate": 1.8390000000000002e-05,
+      "loss": 0.1255,
+      "step": 1387
+    },
+    {
+      "epoch": 34.7,
+      "learning_rate": 1.836e-05,
+      "loss": 0.2017,
+      "step": 1388
+    },
+    {
+      "epoch": 34.73,
+      "learning_rate": 1.833e-05,
+      "loss": 0.1912,
+      "step": 1389
+    },
+    {
+      "epoch": 34.75,
+      "learning_rate": 1.83e-05,
+      "loss": 0.1635,
+      "step": 1390
+    },
+    {
+      "epoch": 34.77,
+      "learning_rate": 1.827e-05,
+      "loss": 0.0553,
+      "step": 1391
+    },
+    {
+      "epoch": 34.8,
+      "learning_rate": 1.824e-05,
+      "loss": 0.0589,
+      "step": 1392
+    },
+    {
+      "epoch": 34.83,
+      "learning_rate": 1.821e-05,
+      "loss": 0.0885,
+      "step": 1393
+    },
+    {
+      "epoch": 34.85,
+      "learning_rate": 1.818e-05,
+      "loss": 0.0911,
+      "step": 1394
+    },
+    {
+      "epoch": 34.88,
+      "learning_rate": 1.815e-05,
+      "loss": 0.1625,
+      "step": 1395
+    },
+    {
+      "epoch": 34.9,
+      "learning_rate": 1.812e-05,
+      "loss": 0.03,
+      "step": 1396
+    },
+    {
+      "epoch": 34.92,
+      "learning_rate": 1.809e-05,
+      "loss": 0.1078,
+      "step": 1397
+    },
+    {
+      "epoch": 34.95,
+      "learning_rate": 1.806e-05,
+      "loss": 0.0651,
+      "step": 1398
+    },
+    {
+      "epoch": 34.98,
+      "learning_rate": 1.803e-05,
+      "loss": 0.1961,
+      "step": 1399
+    },
+    {
+      "epoch": 35.0,
+      "learning_rate": 1.8e-05,
+      "loss": 0.0746,
+      "step": 1400
+    },
+    {
+      "epoch": 35.0,
+      "eval_loss": 0.14976564049720764,
+      "eval_mean_accuracy": 0.942508828234,
+      "eval_mean_iou": 0.5901978897189305,
+      "eval_overall_accuracy": 0.9407725449820105,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9393633129693565,
+        0.9456543434986433
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9261620349045498,
+        0.8444316342522417
+      ],
+      "eval_runtime": 47.832,
+      "eval_samples_per_second": 0.418,
+      "eval_steps_per_second": 0.209,
+      "step": 1400
+    },
+    {
+      "epoch": 35.02,
+      "learning_rate": 1.797e-05,
+      "loss": 0.1048,
+      "step": 1401
+    },
+    {
+      "epoch": 35.05,
+      "learning_rate": 1.794e-05,
+      "loss": 0.1245,
+      "step": 1402
+    },
+    {
+      "epoch": 35.08,
+      "learning_rate": 1.791e-05,
+      "loss": 0.0843,
+      "step": 1403
+    },
+    {
+      "epoch": 35.1,
+      "learning_rate": 1.7879999999999998e-05,
+      "loss": 0.1497,
+      "step": 1404
+    },
+    {
+      "epoch": 35.12,
+      "learning_rate": 1.785e-05,
+      "loss": 0.0681,
+      "step": 1405
+    },
+    {
+      "epoch": 35.15,
+      "learning_rate": 1.782e-05,
+      "loss": 0.0865,
+      "step": 1406
+    },
+    {
+      "epoch": 35.17,
+      "learning_rate": 1.779e-05,
+      "loss": 0.0424,
+      "step": 1407
+    },
+    {
+      "epoch": 35.2,
+      "learning_rate": 1.776e-05,
+      "loss": 0.0668,
+      "step": 1408
+    },
+    {
+      "epoch": 35.23,
+      "learning_rate": 1.7729999999999998e-05,
+      "loss": 0.0472,
+      "step": 1409
+    },
+    {
+      "epoch": 35.25,
+      "learning_rate": 1.77e-05,
+      "loss": 0.1185,
+      "step": 1410
+    },
+    {
+      "epoch": 35.27,
+      "learning_rate": 1.767e-05,
+      "loss": 0.1397,
+      "step": 1411
+    },
+    {
+      "epoch": 35.3,
+      "learning_rate": 1.764e-05,
+      "loss": 0.0682,
+      "step": 1412
+    },
+    {
+      "epoch": 35.33,
+      "learning_rate": 1.761e-05,
+      "loss": 0.0593,
+      "step": 1413
+    },
+    {
+      "epoch": 35.35,
+      "learning_rate": 1.758e-05,
+      "loss": 0.0931,
+      "step": 1414
+    },
+    {
+      "epoch": 35.38,
+      "learning_rate": 1.755e-05,
+      "loss": 0.0229,
+      "step": 1415
+    },
+    {
+      "epoch": 35.4,
+      "learning_rate": 1.7519999999999998e-05,
+      "loss": 0.0546,
+      "step": 1416
+    },
+    {
+      "epoch": 35.42,
+      "learning_rate": 1.749e-05,
+      "loss": 0.1259,
+      "step": 1417
+    },
+    {
+      "epoch": 35.45,
+      "learning_rate": 1.746e-05,
+      "loss": 0.076,
+      "step": 1418
+    },
+    {
+      "epoch": 35.48,
+      "learning_rate": 1.743e-05,
+      "loss": 0.0419,
+      "step": 1419
+    },
+    {
+      "epoch": 35.5,
+      "learning_rate": 1.74e-05,
+      "loss": 0.0867,
+      "step": 1420
+    },
+    {
+      "epoch": 35.5,
+      "eval_loss": 0.1530875712633133,
+      "eval_mean_accuracy": 0.937884059851001,
+      "eval_mean_iou": 0.5842564679460326,
+      "eval_overall_accuracy": 0.9370650637973565,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9364003361639417,
+        0.9393677835380602
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9220496170564254,
+        0.8307197867816722
+      ],
+      "eval_runtime": 48.147,
+      "eval_samples_per_second": 0.415,
+      "eval_steps_per_second": 0.208,
+      "step": 1420
+    },
+    {
+      "epoch": 35.52,
+      "learning_rate": 1.7369999999999998e-05,
+      "loss": 0.3758,
+      "step": 1421
+    },
+    {
+      "epoch": 35.55,
+      "learning_rate": 1.734e-05,
+      "loss": 0.2305,
+      "step": 1422
+    },
+    {
+      "epoch": 35.58,
+      "learning_rate": 1.731e-05,
+      "loss": 0.0796,
+      "step": 1423
+    },
+    {
+      "epoch": 35.6,
+      "learning_rate": 1.728e-05,
+      "loss": 0.0356,
+      "step": 1424
+    },
+    {
+      "epoch": 35.62,
+      "learning_rate": 1.725e-05,
+      "loss": 0.0649,
+      "step": 1425
+    },
+    {
+      "epoch": 35.65,
+      "learning_rate": 1.7219999999999998e-05,
+      "loss": 0.1014,
+      "step": 1426
+    },
+    {
+      "epoch": 35.67,
+      "learning_rate": 1.719e-05,
+      "loss": 0.2181,
+      "step": 1427
+    },
+    {
+      "epoch": 35.7,
+      "learning_rate": 1.716e-05,
+      "loss": 0.1608,
+      "step": 1428
+    },
+    {
+      "epoch": 35.73,
+      "learning_rate": 1.713e-05,
+      "loss": 0.0587,
+      "step": 1429
+    },
+    {
+      "epoch": 35.75,
+      "learning_rate": 1.71e-05,
+      "loss": 0.1245,
+      "step": 1430
+    },
+    {
+      "epoch": 35.77,
+      "learning_rate": 1.7069999999999998e-05,
+      "loss": 0.0723,
+      "step": 1431
+    },
+    {
+      "epoch": 35.8,
+      "learning_rate": 1.704e-05,
+      "loss": 0.1054,
+      "step": 1432
+    },
+    {
+      "epoch": 35.83,
+      "learning_rate": 1.7009999999999998e-05,
+      "loss": 0.0617,
+      "step": 1433
+    },
+    {
+      "epoch": 35.85,
+      "learning_rate": 1.698e-05,
+      "loss": 0.224,
+      "step": 1434
+    },
+    {
+      "epoch": 35.88,
+      "learning_rate": 1.695e-05,
+      "loss": 0.0697,
+      "step": 1435
+    },
+    {
+      "epoch": 35.9,
+      "learning_rate": 1.6919999999999997e-05,
+      "loss": 0.1029,
+      "step": 1436
+    },
+    {
+      "epoch": 35.92,
+      "learning_rate": 1.689e-05,
+      "loss": 0.0393,
+      "step": 1437
+    },
+    {
+      "epoch": 35.95,
+      "learning_rate": 1.686e-05,
+      "loss": 0.1017,
+      "step": 1438
+    },
+    {
+      "epoch": 35.98,
+      "learning_rate": 1.6830000000000003e-05,
+      "loss": 0.04,
+      "step": 1439
+    },
+    {
+      "epoch": 36.0,
+      "learning_rate": 1.6800000000000002e-05,
+      "loss": 0.1269,
+      "step": 1440
+    },
+    {
+      "epoch": 36.0,
+      "eval_loss": 0.14739394187927246,
+      "eval_mean_accuracy": 0.9302531657011144,
+      "eval_mean_iou": 0.5861619422050715,
+      "eval_overall_accuracy": 0.9336004782183467,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9363172814532082,
+        0.9241890499490206
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9186850462077554,
+        0.839800780407459
+      ],
+      "eval_runtime": 49.1789,
+      "eval_samples_per_second": 0.407,
+      "eval_steps_per_second": 0.203,
+      "step": 1440
+    },
+    {
+      "epoch": 36.02,
+      "learning_rate": 1.677e-05,
+      "loss": 0.0928,
+      "step": 1441
+    },
+    {
+      "epoch": 36.05,
+      "learning_rate": 1.6740000000000002e-05,
+      "loss": 0.1342,
+      "step": 1442
+    },
+    {
+      "epoch": 36.08,
+      "learning_rate": 1.671e-05,
+      "loss": 0.0428,
+      "step": 1443
+    },
+    {
+      "epoch": 36.1,
+      "learning_rate": 1.6680000000000003e-05,
+      "loss": 0.0644,
+      "step": 1444
+    },
+    {
+      "epoch": 36.12,
+      "learning_rate": 1.665e-05,
+      "loss": 0.1103,
+      "step": 1445
+    },
+    {
+      "epoch": 36.15,
+      "learning_rate": 1.6620000000000004e-05,
+      "loss": 0.0315,
+      "step": 1446
+    },
+    {
+      "epoch": 36.17,
+      "learning_rate": 1.6590000000000002e-05,
+      "loss": 0.1668,
+      "step": 1447
+    },
+    {
+      "epoch": 36.2,
+      "learning_rate": 1.656e-05,
+      "loss": 0.0641,
+      "step": 1448
+    },
+    {
+      "epoch": 36.23,
+      "learning_rate": 1.6530000000000003e-05,
+      "loss": 0.0403,
+      "step": 1449
+    },
+    {
+      "epoch": 36.25,
+      "learning_rate": 1.65e-05,
+      "loss": 0.1035,
+      "step": 1450
+    },
+    {
+      "epoch": 36.27,
+      "learning_rate": 1.6470000000000003e-05,
+      "loss": 0.0445,
+      "step": 1451
+    },
+    {
+      "epoch": 36.3,
+      "learning_rate": 1.6440000000000002e-05,
+      "loss": 0.054,
+      "step": 1452
+    },
+    {
+      "epoch": 36.33,
+      "learning_rate": 1.641e-05,
+      "loss": 0.1142,
+      "step": 1453
+    },
+    {
+      "epoch": 36.35,
+      "learning_rate": 1.6380000000000002e-05,
+      "loss": 0.0414,
+      "step": 1454
+    },
+    {
+      "epoch": 36.38,
+      "learning_rate": 1.635e-05,
+      "loss": 0.1737,
+      "step": 1455
+    },
+    {
+      "epoch": 36.4,
+      "learning_rate": 1.6320000000000003e-05,
+      "loss": 0.204,
+      "step": 1456
+    },
+    {
+      "epoch": 36.42,
+      "learning_rate": 1.629e-05,
+      "loss": 0.1368,
+      "step": 1457
+    },
+    {
+      "epoch": 36.45,
+      "learning_rate": 1.626e-05,
+      "loss": 0.215,
+      "step": 1458
+    },
+    {
+      "epoch": 36.48,
+      "learning_rate": 1.6230000000000002e-05,
+      "loss": 0.0336,
+      "step": 1459
+    },
+    {
+      "epoch": 36.5,
+      "learning_rate": 1.62e-05,
+      "loss": 0.1781,
+      "step": 1460
+    },
+    {
+      "epoch": 36.5,
+      "eval_loss": 0.13674582540988922,
+      "eval_mean_accuracy": 0.9394248202931195,
+      "eval_mean_iou": 0.5865482558773343,
+      "eval_overall_accuracy": 0.9346481920994075,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9307713031439842,
+        0.9480783374422548
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9195332271786706,
+        0.8401115404533323
+      ],
+      "eval_runtime": 48.7405,
+      "eval_samples_per_second": 0.41,
+      "eval_steps_per_second": 0.205,
+      "step": 1460
+    },
+    {
+      "epoch": 36.52,
+      "learning_rate": 1.6170000000000003e-05,
+      "loss": 0.2124,
+      "step": 1461
+    },
+    {
+      "epoch": 36.55,
+      "learning_rate": 1.614e-05,
+      "loss": 0.1168,
+      "step": 1462
+    },
+    {
+      "epoch": 36.58,
+      "learning_rate": 1.611e-05,
+      "loss": 0.0259,
+      "step": 1463
+    },
+    {
+      "epoch": 36.6,
+      "learning_rate": 1.6080000000000002e-05,
+      "loss": 0.1414,
+      "step": 1464
+    },
+    {
+      "epoch": 36.62,
+      "learning_rate": 1.605e-05,
+      "loss": 0.1541,
+      "step": 1465
+    },
+    {
+      "epoch": 36.65,
+      "learning_rate": 1.6020000000000002e-05,
+      "loss": 0.1184,
+      "step": 1466
+    },
+    {
+      "epoch": 36.67,
+      "learning_rate": 1.599e-05,
+      "loss": 0.1012,
+      "step": 1467
+    },
+    {
+      "epoch": 36.7,
+      "learning_rate": 1.596e-05,
+      "loss": 0.0599,
+      "step": 1468
+    },
+    {
+      "epoch": 36.73,
+      "learning_rate": 1.593e-05,
+      "loss": 0.1021,
+      "step": 1469
+    },
+    {
+      "epoch": 36.75,
+      "learning_rate": 1.59e-05,
+      "loss": 0.0441,
+      "step": 1470
+    },
+    {
+      "epoch": 36.77,
+      "learning_rate": 1.5870000000000002e-05,
+      "loss": 0.1371,
+      "step": 1471
+    },
+    {
+      "epoch": 36.8,
+      "learning_rate": 1.584e-05,
+      "loss": 0.05,
+      "step": 1472
+    },
+    {
+      "epoch": 36.83,
+      "learning_rate": 1.5810000000000003e-05,
+      "loss": 0.0781,
+      "step": 1473
+    },
+    {
+      "epoch": 36.85,
+      "learning_rate": 1.578e-05,
+      "loss": 0.125,
+      "step": 1474
+    },
+    {
+      "epoch": 36.88,
+      "learning_rate": 1.575e-05,
+      "loss": 0.0814,
+      "step": 1475
+    },
+    {
+      "epoch": 36.9,
+      "learning_rate": 1.5720000000000002e-05,
+      "loss": 0.137,
+      "step": 1476
+    },
+    {
+      "epoch": 36.92,
+      "learning_rate": 1.569e-05,
+      "loss": 0.0856,
+      "step": 1477
+    },
+    {
+      "epoch": 36.95,
+      "learning_rate": 1.5660000000000003e-05,
+      "loss": 0.0723,
+      "step": 1478
+    },
+    {
+      "epoch": 36.98,
+      "learning_rate": 1.563e-05,
+      "loss": 0.0817,
+      "step": 1479
+    },
+    {
+      "epoch": 37.0,
+      "learning_rate": 1.56e-05,
+      "loss": 0.1156,
+      "step": 1480
+    },
+    {
+      "epoch": 37.0,
+      "eval_loss": 0.1699506789445877,
+      "eval_mean_accuracy": 0.9400083419738761,
+      "eval_mean_iou": 0.5834442910001464,
+      "eval_overall_accuracy": 0.9436325098245835,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9465740191368436,
+        0.9334426648109087
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9301988314355468,
+        0.8201340415648926
+      ],
+      "eval_runtime": 48.4058,
+      "eval_samples_per_second": 0.413,
+      "eval_steps_per_second": 0.207,
+      "step": 1480
+    },
+    {
+      "epoch": 37.02,
+      "learning_rate": 1.5570000000000002e-05,
+      "loss": 0.134,
+      "step": 1481
+    },
+    {
+      "epoch": 37.05,
+      "learning_rate": 1.554e-05,
+      "loss": 0.1748,
+      "step": 1482
+    },
+    {
+      "epoch": 37.08,
+      "learning_rate": 1.5510000000000002e-05,
+      "loss": 0.0336,
+      "step": 1483
+    },
+    {
+      "epoch": 37.1,
+      "learning_rate": 1.548e-05,
+      "loss": 0.0543,
+      "step": 1484
+    },
+    {
+      "epoch": 37.12,
+      "learning_rate": 1.545e-05,
+      "loss": 0.1106,
+      "step": 1485
+    },
+    {
+      "epoch": 37.15,
+      "learning_rate": 1.542e-05,
+      "loss": 0.0342,
+      "step": 1486
+    },
+    {
+      "epoch": 37.17,
+      "learning_rate": 1.539e-05,
+      "loss": 0.1299,
+      "step": 1487
+    },
+    {
+      "epoch": 37.2,
+      "learning_rate": 1.5360000000000002e-05,
+      "loss": 0.1131,
+      "step": 1488
+    },
+    {
+      "epoch": 37.23,
+      "learning_rate": 1.533e-05,
+      "loss": 0.052,
+      "step": 1489
+    },
+    {
+      "epoch": 37.25,
+      "learning_rate": 1.53e-05,
+      "loss": 0.1341,
+      "step": 1490
+    },
+    {
+      "epoch": 37.27,
+      "learning_rate": 1.527e-05,
+      "loss": 0.1071,
+      "step": 1491
+    },
+    {
+      "epoch": 37.3,
+      "learning_rate": 1.524e-05,
+      "loss": 0.0526,
+      "step": 1492
+    },
+    {
+      "epoch": 37.33,
+      "learning_rate": 1.521e-05,
+      "loss": 0.1031,
+      "step": 1493
+    },
+    {
+      "epoch": 37.35,
+      "learning_rate": 1.518e-05,
+      "loss": 0.2032,
+      "step": 1494
+    },
+    {
+      "epoch": 37.38,
+      "learning_rate": 1.515e-05,
+      "loss": 0.3003,
+      "step": 1495
+    },
+    {
+      "epoch": 37.4,
+      "learning_rate": 1.5120000000000001e-05,
+      "loss": 0.1506,
+      "step": 1496
+    },
+    {
+      "epoch": 37.42,
+      "learning_rate": 1.5090000000000001e-05,
+      "loss": 0.1489,
+      "step": 1497
+    },
+    {
+      "epoch": 37.45,
+      "learning_rate": 1.506e-05,
+      "loss": 0.126,
+      "step": 1498
+    },
+    {
+      "epoch": 37.48,
+      "learning_rate": 1.503e-05,
+      "loss": 0.0682,
+      "step": 1499
+    },
+    {
+      "epoch": 37.5,
+      "learning_rate": 1.5e-05,
+      "loss": 0.2491,
+      "step": 1500
+    },
+    {
+      "epoch": 37.5,
+      "eval_loss": 0.1416798233985901,
+      "eval_mean_accuracy": 0.9348448180461979,
+      "eval_mean_iou": 0.583733217018338,
+      "eval_overall_accuracy": 0.935467914416962,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9359736425875488,
+        0.9337159935048471
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9203475590677381,
+        0.8308520919872759
+      ],
+      "eval_runtime": 49.5728,
+      "eval_samples_per_second": 0.403,
+      "eval_steps_per_second": 0.202,
+      "step": 1500
+    },
+    {
+      "epoch": 37.52,
+      "learning_rate": 1.497e-05,
+      "loss": 0.034,
+      "step": 1501
+    },
+    {
+      "epoch": 37.55,
+      "learning_rate": 1.4940000000000001e-05,
+      "loss": 0.0597,
+      "step": 1502
+    },
+    {
+      "epoch": 37.58,
+      "learning_rate": 1.491e-05,
+      "loss": 0.0511,
+      "step": 1503
+    },
+    {
+      "epoch": 37.6,
+      "learning_rate": 1.488e-05,
+      "loss": 0.0459,
+      "step": 1504
+    },
+    {
+      "epoch": 37.62,
+      "learning_rate": 1.485e-05,
+      "loss": 0.1409,
+      "step": 1505
+    },
+    {
+      "epoch": 37.65,
+      "learning_rate": 1.482e-05,
+      "loss": 0.1254,
+      "step": 1506
+    },
+    {
+      "epoch": 37.67,
+      "learning_rate": 1.479e-05,
+      "loss": 0.0691,
+      "step": 1507
+    },
+    {
+      "epoch": 37.7,
+      "learning_rate": 1.4760000000000001e-05,
+      "loss": 0.1005,
+      "step": 1508
+    },
+    {
+      "epoch": 37.73,
+      "learning_rate": 1.473e-05,
+      "loss": 0.1332,
+      "step": 1509
+    },
+    {
+      "epoch": 37.75,
+      "learning_rate": 1.47e-05,
+      "loss": 0.1135,
+      "step": 1510
+    },
+    {
+      "epoch": 37.77,
+      "learning_rate": 1.467e-05,
+      "loss": 0.1385,
+      "step": 1511
+    },
+    {
+      "epoch": 37.8,
+      "learning_rate": 1.464e-05,
+      "loss": 0.0694,
+      "step": 1512
+    },
+    {
+      "epoch": 37.83,
+      "learning_rate": 1.461e-05,
+      "loss": 0.1426,
+      "step": 1513
+    },
+    {
+      "epoch": 37.85,
+      "learning_rate": 1.458e-05,
+      "loss": 0.3264,
+      "step": 1514
+    },
+    {
+      "epoch": 37.88,
+      "learning_rate": 1.455e-05,
+      "loss": 0.0749,
+      "step": 1515
+    },
+    {
+      "epoch": 37.9,
+      "learning_rate": 1.452e-05,
+      "loss": 0.084,
+      "step": 1516
+    },
+    {
+      "epoch": 37.92,
+      "learning_rate": 1.449e-05,
+      "loss": 0.0469,
+      "step": 1517
+    },
+    {
+      "epoch": 37.95,
+      "learning_rate": 1.446e-05,
+      "loss": 0.0312,
+      "step": 1518
+    },
+    {
+      "epoch": 37.98,
+      "learning_rate": 1.4429999999999999e-05,
+      "loss": 0.2118,
+      "step": 1519
+    },
+    {
+      "epoch": 38.0,
+      "learning_rate": 1.44e-05,
+      "loss": 0.0513,
+      "step": 1520
+    },
+    {
+      "epoch": 38.0,
+      "eval_loss": 0.17021945118904114,
+      "eval_mean_accuracy": 0.9346689759231479,
+      "eval_mean_iou": 0.5692587329056024,
+      "eval_overall_accuracy": 0.9224292944184765,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9124951140470952,
+        0.9568428377992005
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.903679454502646,
+        0.804096744214161
+      ],
+      "eval_runtime": 48.8208,
+      "eval_samples_per_second": 0.41,
+      "eval_steps_per_second": 0.205,
+      "step": 1520
+    },
+    {
+      "epoch": 38.02,
+      "learning_rate": 1.437e-05,
+      "loss": 0.0984,
+      "step": 1521
+    },
+    {
+      "epoch": 38.05,
+      "learning_rate": 1.434e-05,
+      "loss": 0.0513,
+      "step": 1522
+    },
+    {
+      "epoch": 38.08,
+      "learning_rate": 1.431e-05,
+      "loss": 0.0709,
+      "step": 1523
+    },
+    {
+      "epoch": 38.1,
+      "learning_rate": 1.428e-05,
+      "loss": 0.0715,
+      "step": 1524
+    },
+    {
+      "epoch": 38.12,
+      "learning_rate": 1.4249999999999999e-05,
+      "loss": 0.0209,
+      "step": 1525
+    },
+    {
+      "epoch": 38.15,
+      "learning_rate": 1.422e-05,
+      "loss": 0.2465,
+      "step": 1526
+    },
+    {
+      "epoch": 38.17,
+      "learning_rate": 1.419e-05,
+      "loss": 0.606,
+      "step": 1527
+    },
+    {
+      "epoch": 38.2,
+      "learning_rate": 1.416e-05,
+      "loss": 0.0553,
+      "step": 1528
+    },
+    {
+      "epoch": 38.23,
+      "learning_rate": 1.413e-05,
+      "loss": 0.0377,
+      "step": 1529
+    },
+    {
+      "epoch": 38.25,
+      "learning_rate": 1.4099999999999999e-05,
+      "loss": 0.1503,
+      "step": 1530
+    },
+    {
+      "epoch": 38.27,
+      "learning_rate": 1.4069999999999999e-05,
+      "loss": 0.0807,
+      "step": 1531
+    },
+    {
+      "epoch": 38.3,
+      "learning_rate": 1.4040000000000001e-05,
+      "loss": 0.0664,
+      "step": 1532
+    },
+    {
+      "epoch": 38.33,
+      "learning_rate": 1.4010000000000001e-05,
+      "loss": 0.1416,
+      "step": 1533
+    },
+    {
+      "epoch": 38.35,
+      "learning_rate": 1.3980000000000002e-05,
+      "loss": 0.0887,
+      "step": 1534
+    },
+    {
+      "epoch": 38.38,
+      "learning_rate": 1.395e-05,
+      "loss": 0.0687,
+      "step": 1535
+    },
+    {
+      "epoch": 38.4,
+      "learning_rate": 1.392e-05,
+      "loss": 0.0594,
+      "step": 1536
+    },
+    {
+      "epoch": 38.42,
+      "learning_rate": 1.389e-05,
+      "loss": 0.1011,
+      "step": 1537
+    },
+    {
+      "epoch": 38.45,
+      "learning_rate": 1.3860000000000001e-05,
+      "loss": 0.3874,
+      "step": 1538
+    },
+    {
+      "epoch": 38.48,
+      "learning_rate": 1.3830000000000001e-05,
+      "loss": 0.1075,
+      "step": 1539
+    },
+    {
+      "epoch": 38.5,
+      "learning_rate": 1.3800000000000002e-05,
+      "loss": 0.1152,
+      "step": 1540
+    },
+    {
+      "epoch": 38.5,
+      "eval_loss": 0.16474521160125732,
+      "eval_mean_accuracy": 0.9375231913789847,
+      "eval_mean_iou": 0.5693530557237917,
+      "eval_overall_accuracy": 0.9232578785989234,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9116796206060813,
+        0.963366762151888
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9043497738220596,
+        0.8037093933493155
+      ],
+      "eval_runtime": 48.9289,
+      "eval_samples_per_second": 0.409,
+      "eval_steps_per_second": 0.204,
+      "step": 1540
+    },
+    {
+      "epoch": 38.52,
+      "learning_rate": 1.377e-05,
+      "loss": 0.1256,
+      "step": 1541
+    },
+    {
+      "epoch": 38.55,
+      "learning_rate": 1.374e-05,
+      "loss": 0.1918,
+      "step": 1542
+    },
+    {
+      "epoch": 38.58,
+      "learning_rate": 1.3710000000000001e-05,
+      "loss": 0.1348,
+      "step": 1543
+    },
+    {
+      "epoch": 38.6,
+      "learning_rate": 1.3680000000000001e-05,
+      "loss": 0.0442,
+      "step": 1544
+    },
+    {
+      "epoch": 38.62,
+      "learning_rate": 1.3650000000000001e-05,
+      "loss": 0.0866,
+      "step": 1545
+    },
+    {
+      "epoch": 38.65,
+      "learning_rate": 1.362e-05,
+      "loss": 0.0558,
+      "step": 1546
+    },
+    {
+      "epoch": 38.67,
+      "learning_rate": 1.359e-05,
+      "loss": 0.1172,
+      "step": 1547
+    },
+    {
+      "epoch": 38.7,
+      "learning_rate": 1.356e-05,
+      "loss": 0.0327,
+      "step": 1548
+    },
+    {
+      "epoch": 38.73,
+      "learning_rate": 1.3530000000000001e-05,
+      "loss": 0.0357,
+      "step": 1549
+    },
+    {
+      "epoch": 38.75,
+      "learning_rate": 1.3500000000000001e-05,
+      "loss": 0.0719,
+      "step": 1550
+    },
+    {
+      "epoch": 38.77,
+      "learning_rate": 1.3470000000000001e-05,
+      "loss": 0.0593,
+      "step": 1551
+    },
+    {
+      "epoch": 38.8,
+      "learning_rate": 1.344e-05,
+      "loss": 0.0729,
+      "step": 1552
+    },
+    {
+      "epoch": 38.83,
+      "learning_rate": 1.341e-05,
+      "loss": 0.1262,
+      "step": 1553
+    },
+    {
+      "epoch": 38.85,
+      "learning_rate": 1.338e-05,
+      "loss": 0.1082,
+      "step": 1554
+    },
+    {
+      "epoch": 38.88,
+      "learning_rate": 1.3350000000000001e-05,
+      "loss": 0.1281,
+      "step": 1555
+    },
+    {
+      "epoch": 38.9,
+      "learning_rate": 1.3320000000000001e-05,
+      "loss": 0.0546,
+      "step": 1556
+    },
+    {
+      "epoch": 38.92,
+      "learning_rate": 1.329e-05,
+      "loss": 0.1394,
+      "step": 1557
+    },
+    {
+      "epoch": 38.95,
+      "learning_rate": 1.326e-05,
+      "loss": 0.068,
+      "step": 1558
+    },
+    {
+      "epoch": 38.98,
+      "learning_rate": 1.323e-05,
+      "loss": 0.1634,
+      "step": 1559
+    },
+    {
+      "epoch": 39.0,
+      "learning_rate": 1.32e-05,
+      "loss": 0.0377,
+      "step": 1560
+    },
+    {
+      "epoch": 39.0,
+      "eval_loss": 0.15060418844223022,
+      "eval_mean_accuracy": 0.9282203457503876,
+      "eval_mean_iou": 0.5837112341440569,
+      "eval_overall_accuracy": 0.9339702595881328,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9386371033423812,
+        0.917803588158394
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9191812596170122,
+        0.8319524428151585
+      ],
+      "eval_runtime": 48.5623,
+      "eval_samples_per_second": 0.412,
+      "eval_steps_per_second": 0.206,
+      "step": 1560
+    },
+    {
+      "epoch": 39.02,
+      "learning_rate": 1.3170000000000001e-05,
+      "loss": 0.2205,
+      "step": 1561
+    },
+    {
+      "epoch": 39.05,
+      "learning_rate": 1.314e-05,
+      "loss": 0.0386,
+      "step": 1562
+    },
+    {
+      "epoch": 39.08,
+      "learning_rate": 1.311e-05,
+      "loss": 0.1041,
+      "step": 1563
+    },
+    {
+      "epoch": 39.1,
+      "learning_rate": 1.308e-05,
+      "loss": 0.0255,
+      "step": 1564
+    },
+    {
+      "epoch": 39.12,
+      "learning_rate": 1.305e-05,
+      "loss": 0.034,
+      "step": 1565
+    },
+    {
+      "epoch": 39.15,
+      "learning_rate": 1.302e-05,
+      "loss": 0.0867,
+      "step": 1566
+    },
+    {
+      "epoch": 39.17,
+      "learning_rate": 1.2990000000000001e-05,
+      "loss": 0.1219,
+      "step": 1567
+    },
+    {
+      "epoch": 39.2,
+      "learning_rate": 1.296e-05,
+      "loss": 0.1316,
+      "step": 1568
+    },
+    {
+      "epoch": 39.23,
+      "learning_rate": 1.293e-05,
+      "loss": 0.0808,
+      "step": 1569
+    },
+    {
+      "epoch": 39.25,
+      "learning_rate": 1.29e-05,
+      "loss": 0.057,
+      "step": 1570
+    },
+    {
+      "epoch": 39.27,
+      "learning_rate": 1.287e-05,
+      "loss": 0.03,
+      "step": 1571
+    },
+    {
+      "epoch": 39.3,
+      "learning_rate": 1.284e-05,
+      "loss": 0.125,
+      "step": 1572
+    },
+    {
+      "epoch": 39.33,
+      "learning_rate": 1.281e-05,
+      "loss": 0.0966,
+      "step": 1573
+    },
+    {
+      "epoch": 39.35,
+      "learning_rate": 1.278e-05,
+      "loss": 0.1163,
+      "step": 1574
+    },
+    {
+      "epoch": 39.38,
+      "learning_rate": 1.275e-05,
+      "loss": 0.0494,
+      "step": 1575
+    },
+    {
+      "epoch": 39.4,
+      "learning_rate": 1.272e-05,
+      "loss": 0.0871,
+      "step": 1576
+    },
+    {
+      "epoch": 39.42,
+      "learning_rate": 1.269e-05,
+      "loss": 0.2024,
+      "step": 1577
+    },
+    {
+      "epoch": 39.45,
+      "learning_rate": 1.2659999999999999e-05,
+      "loss": 0.0404,
+      "step": 1578
+    },
+    {
+      "epoch": 39.48,
+      "learning_rate": 1.263e-05,
+      "loss": 0.1453,
+      "step": 1579
+    },
+    {
+      "epoch": 39.5,
+      "learning_rate": 1.26e-05,
+      "loss": 0.1739,
+      "step": 1580
+    },
+    {
+      "epoch": 39.5,
+      "eval_loss": 0.17252664268016815,
+      "eval_mean_accuracy": 0.9455844532848459,
+      "eval_mean_iou": 0.5759927550891428,
+      "eval_overall_accuracy": 0.9348298602887034,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9261010329410555,
+        0.9650678736286362
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9188780912904103,
+        0.8091001739770182
+      ],
+      "eval_runtime": 48.5914,
+      "eval_samples_per_second": 0.412,
+      "eval_steps_per_second": 0.206,
+      "step": 1580
+    },
+    {
+      "epoch": 39.52,
+      "learning_rate": 1.257e-05,
+      "loss": 0.0157,
+      "step": 1581
+    },
+    {
+      "epoch": 39.55,
+      "learning_rate": 1.254e-05,
+      "loss": 0.107,
+      "step": 1582
+    },
+    {
+      "epoch": 39.58,
+      "learning_rate": 1.251e-05,
+      "loss": 0.1293,
+      "step": 1583
+    },
+    {
+      "epoch": 39.6,
+      "learning_rate": 1.2479999999999999e-05,
+      "loss": 0.06,
+      "step": 1584
+    },
+    {
+      "epoch": 39.62,
+      "learning_rate": 1.245e-05,
+      "loss": 0.0913,
+      "step": 1585
+    },
+    {
+      "epoch": 39.65,
+      "learning_rate": 1.242e-05,
+      "loss": 0.1068,
+      "step": 1586
+    },
+    {
+      "epoch": 39.67,
+      "learning_rate": 1.239e-05,
+      "loss": 0.0989,
+      "step": 1587
+    },
+    {
+      "epoch": 39.7,
+      "learning_rate": 1.236e-05,
+      "loss": 0.1252,
+      "step": 1588
+    },
+    {
+      "epoch": 39.73,
+      "learning_rate": 1.2329999999999999e-05,
+      "loss": 0.1365,
+      "step": 1589
+    },
+    {
+      "epoch": 39.75,
+      "learning_rate": 1.2299999999999999e-05,
+      "loss": 0.1302,
+      "step": 1590
+    },
+    {
+      "epoch": 39.77,
+      "learning_rate": 1.227e-05,
+      "loss": 0.088,
+      "step": 1591
+    },
+    {
+      "epoch": 39.8,
+      "learning_rate": 1.224e-05,
+      "loss": 0.0972,
+      "step": 1592
+    },
+    {
+      "epoch": 39.83,
+      "learning_rate": 1.221e-05,
+      "loss": 0.1552,
+      "step": 1593
+    },
+    {
+      "epoch": 39.85,
+      "learning_rate": 1.2180000000000002e-05,
+      "loss": 0.0441,
+      "step": 1594
+    },
+    {
+      "epoch": 39.88,
+      "learning_rate": 1.215e-05,
+      "loss": 0.1692,
+      "step": 1595
+    },
+    {
+      "epoch": 39.9,
+      "learning_rate": 1.2120000000000001e-05,
+      "loss": 0.0646,
+      "step": 1596
+    },
+    {
+      "epoch": 39.92,
+      "learning_rate": 1.2090000000000001e-05,
+      "loss": 0.0185,
+      "step": 1597
+    },
+    {
+      "epoch": 39.95,
+      "learning_rate": 1.2060000000000001e-05,
+      "loss": 0.0201,
+      "step": 1598
+    },
+    {
+      "epoch": 39.98,
+      "learning_rate": 1.2030000000000002e-05,
+      "loss": 0.0976,
+      "step": 1599
+    },
+    {
+      "epoch": 40.0,
+      "learning_rate": 1.2e-05,
+      "loss": 0.1689,
+      "step": 1600
+    },
+    {
+      "epoch": 40.0,
+      "eval_loss": 0.15484194457530975,
+      "eval_mean_accuracy": 0.9263051408733539,
+      "eval_mean_iou": 0.5865586544006898,
+      "eval_overall_accuracy": 0.9452296592049781,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9605895015731082,
+        0.8920207801735996
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9340668160081408,
+        0.8256091471939285
+      ],
+      "eval_runtime": 48.1835,
+      "eval_samples_per_second": 0.415,
+      "eval_steps_per_second": 0.208,
+      "step": 1600
+    },
+    {
+      "epoch": 40.02,
+      "learning_rate": 1.197e-05,
+      "loss": 0.0417,
+      "step": 1601
+    },
+    {
+      "epoch": 40.05,
+      "learning_rate": 1.1940000000000001e-05,
+      "loss": 0.1956,
+      "step": 1602
+    },
+    {
+      "epoch": 40.08,
+      "learning_rate": 1.1910000000000001e-05,
+      "loss": 0.0775,
+      "step": 1603
+    },
+    {
+      "epoch": 40.1,
+      "learning_rate": 1.1880000000000001e-05,
+      "loss": 0.2231,
+      "step": 1604
+    },
+    {
+      "epoch": 40.12,
+      "learning_rate": 1.185e-05,
+      "loss": 0.0206,
+      "step": 1605
+    },
+    {
+      "epoch": 40.15,
+      "learning_rate": 1.182e-05,
+      "loss": 0.0827,
+      "step": 1606
+    },
+    {
+      "epoch": 40.17,
+      "learning_rate": 1.179e-05,
+      "loss": 0.1537,
+      "step": 1607
+    },
+    {
+      "epoch": 40.2,
+      "learning_rate": 1.1760000000000001e-05,
+      "loss": 0.1099,
+      "step": 1608
+    },
+    {
+      "epoch": 40.23,
+      "learning_rate": 1.1730000000000001e-05,
+      "loss": 0.2516,
+      "step": 1609
+    },
+    {
+      "epoch": 40.25,
+      "learning_rate": 1.1700000000000001e-05,
+      "loss": 0.0471,
+      "step": 1610
+    },
+    {
+      "epoch": 40.27,
+      "learning_rate": 1.167e-05,
+      "loss": 0.0874,
+      "step": 1611
+    },
+    {
+      "epoch": 40.3,
+      "learning_rate": 1.164e-05,
+      "loss": 0.0487,
+      "step": 1612
+    },
+    {
+      "epoch": 40.33,
+      "learning_rate": 1.161e-05,
+      "loss": 0.0814,
+      "step": 1613
+    },
+    {
+      "epoch": 40.35,
+      "learning_rate": 1.1580000000000001e-05,
+      "loss": 0.1377,
+      "step": 1614
+    },
+    {
+      "epoch": 40.38,
+      "learning_rate": 1.1550000000000001e-05,
+      "loss": 0.0841,
+      "step": 1615
+    },
+    {
+      "epoch": 40.4,
+      "learning_rate": 1.152e-05,
+      "loss": 0.1164,
+      "step": 1616
+    },
+    {
+      "epoch": 40.42,
+      "learning_rate": 1.149e-05,
+      "loss": 0.0625,
+      "step": 1617
+    },
+    {
+      "epoch": 40.45,
+      "learning_rate": 1.146e-05,
+      "loss": 0.0747,
+      "step": 1618
+    },
+    {
+      "epoch": 40.48,
+      "learning_rate": 1.143e-05,
+      "loss": 0.0291,
+      "step": 1619
+    },
+    {
+      "epoch": 40.5,
+      "learning_rate": 1.1400000000000001e-05,
+      "loss": 0.0554,
+      "step": 1620
+    },
+    {
+      "epoch": 40.5,
+      "eval_loss": 0.18766362965106964,
+      "eval_mean_accuracy": 0.9372510288672151,
+      "eval_mean_iou": 0.5852950516956295,
+      "eval_overall_accuracy": 0.9467301337174548,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9544237274850359,
+        0.9200783302493944
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9345147757387361,
+        0.8213703793481524
+      ],
+      "eval_runtime": 48.2812,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 1620
+    },
+    {
+      "epoch": 40.52,
+      "learning_rate": 1.137e-05,
+      "loss": 0.2487,
+      "step": 1621
+    },
+    {
+      "epoch": 40.55,
+      "learning_rate": 1.134e-05,
+      "loss": 0.1035,
+      "step": 1622
+    },
+    {
+      "epoch": 40.58,
+      "learning_rate": 1.131e-05,
+      "loss": 0.0741,
+      "step": 1623
+    },
+    {
+      "epoch": 40.6,
+      "learning_rate": 1.128e-05,
+      "loss": 0.12,
+      "step": 1624
+    },
+    {
+      "epoch": 40.62,
+      "learning_rate": 1.125e-05,
+      "loss": 0.0885,
+      "step": 1625
+    },
+    {
+      "epoch": 40.65,
+      "learning_rate": 1.1220000000000001e-05,
+      "loss": 0.077,
+      "step": 1626
+    },
+    {
+      "epoch": 40.67,
+      "learning_rate": 1.119e-05,
+      "loss": 0.0719,
+      "step": 1627
+    },
+    {
+      "epoch": 40.7,
+      "learning_rate": 1.116e-05,
+      "loss": 0.0861,
+      "step": 1628
+    },
+    {
+      "epoch": 40.73,
+      "learning_rate": 1.113e-05,
+      "loss": 0.1737,
+      "step": 1629
+    },
+    {
+      "epoch": 40.75,
+      "learning_rate": 1.11e-05,
+      "loss": 0.0647,
+      "step": 1630
+    },
+    {
+      "epoch": 40.77,
+      "learning_rate": 1.107e-05,
+      "loss": 0.0721,
+      "step": 1631
+    },
+    {
+      "epoch": 40.8,
+      "learning_rate": 1.104e-05,
+      "loss": 0.2314,
+      "step": 1632
+    },
+    {
+      "epoch": 40.83,
+      "learning_rate": 1.101e-05,
+      "loss": 0.0226,
+      "step": 1633
+    },
+    {
+      "epoch": 40.85,
+      "learning_rate": 1.098e-05,
+      "loss": 0.0976,
+      "step": 1634
+    },
+    {
+      "epoch": 40.88,
+      "learning_rate": 1.095e-05,
+      "loss": 0.148,
+      "step": 1635
+    },
+    {
+      "epoch": 40.9,
+      "learning_rate": 1.092e-05,
+      "loss": 0.0273,
+      "step": 1636
+    },
+    {
+      "epoch": 40.92,
+      "learning_rate": 1.089e-05,
+      "loss": 0.0631,
+      "step": 1637
+    },
+    {
+      "epoch": 40.95,
+      "learning_rate": 1.086e-05,
+      "loss": 0.1522,
+      "step": 1638
+    },
+    {
+      "epoch": 40.98,
+      "learning_rate": 1.083e-05,
+      "loss": 0.2047,
+      "step": 1639
+    },
+    {
+      "epoch": 41.0,
+      "learning_rate": 1.08e-05,
+      "loss": 0.1138,
+      "step": 1640
+    },
+    {
+      "epoch": 41.0,
+      "eval_loss": 0.22113797068595886,
+      "eval_mean_accuracy": 0.9447133266132,
+      "eval_mean_iou": 0.5833858856464879,
+      "eval_overall_accuracy": 0.9449601780106568,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9451605317785492,
+        0.9442661214478508
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.932104857931509,
+        0.8180527990079545
+      ],
+      "eval_runtime": 48.4069,
+      "eval_samples_per_second": 0.413,
+      "eval_steps_per_second": 0.207,
+      "step": 1640
+    },
+    {
+      "epoch": 41.02,
+      "learning_rate": 1.077e-05,
+      "loss": 0.131,
+      "step": 1641
+    },
+    {
+      "epoch": 41.05,
+      "learning_rate": 1.074e-05,
+      "loss": 0.0855,
+      "step": 1642
+    },
+    {
+      "epoch": 41.08,
+      "learning_rate": 1.0709999999999999e-05,
+      "loss": 0.0409,
+      "step": 1643
+    },
+    {
+      "epoch": 41.1,
+      "learning_rate": 1.068e-05,
+      "loss": 0.1722,
+      "step": 1644
+    },
+    {
+      "epoch": 41.12,
+      "learning_rate": 1.065e-05,
+      "loss": 0.0603,
+      "step": 1645
+    },
+    {
+      "epoch": 41.15,
+      "learning_rate": 1.062e-05,
+      "loss": 0.0655,
+      "step": 1646
+    },
+    {
+      "epoch": 41.17,
+      "learning_rate": 1.059e-05,
+      "loss": 0.0634,
+      "step": 1647
+    },
+    {
+      "epoch": 41.2,
+      "learning_rate": 1.0559999999999999e-05,
+      "loss": 0.14,
+      "step": 1648
+    },
+    {
+      "epoch": 41.23,
+      "learning_rate": 1.0529999999999999e-05,
+      "loss": 0.0512,
+      "step": 1649
+    },
+    {
+      "epoch": 41.25,
+      "learning_rate": 1.05e-05,
+      "loss": 0.1021,
+      "step": 1650
+    },
+    {
+      "epoch": 41.27,
+      "learning_rate": 1.047e-05,
+      "loss": 0.1156,
+      "step": 1651
+    },
+    {
+      "epoch": 41.3,
+      "learning_rate": 1.044e-05,
+      "loss": 0.1384,
+      "step": 1652
+    },
+    {
+      "epoch": 41.33,
+      "learning_rate": 1.041e-05,
+      "loss": 0.0354,
+      "step": 1653
+    },
+    {
+      "epoch": 41.35,
+      "learning_rate": 1.0379999999999999e-05,
+      "loss": 0.0358,
+      "step": 1654
+    },
+    {
+      "epoch": 41.38,
+      "learning_rate": 1.035e-05,
+      "loss": 0.1408,
+      "step": 1655
+    },
+    {
+      "epoch": 41.4,
+      "learning_rate": 1.032e-05,
+      "loss": 0.0478,
+      "step": 1656
+    },
+    {
+      "epoch": 41.42,
+      "learning_rate": 1.0290000000000001e-05,
+      "loss": 0.2455,
+      "step": 1657
+    },
+    {
+      "epoch": 41.45,
+      "learning_rate": 1.0260000000000002e-05,
+      "loss": 0.0977,
+      "step": 1658
+    },
+    {
+      "epoch": 41.48,
+      "learning_rate": 1.023e-05,
+      "loss": 0.059,
+      "step": 1659
+    },
+    {
+      "epoch": 41.5,
+      "learning_rate": 1.02e-05,
+      "loss": 0.0918,
+      "step": 1660
+    },
+    {
+      "epoch": 41.5,
+      "eval_loss": 0.15297159552574158,
+      "eval_mean_accuracy": 0.9390500783329355,
+      "eval_mean_iou": 0.5760834490653837,
+      "eval_overall_accuracy": 0.9268852002055952,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9170117330351669,
+        0.9610884236307041
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9088471905037024,
+        0.8194031566924487
+      ],
+      "eval_runtime": 47.2833,
+      "eval_samples_per_second": 0.423,
+      "eval_steps_per_second": 0.211,
+      "step": 1660
+    },
+    {
+      "epoch": 41.52,
+      "learning_rate": 1.0170000000000001e-05,
+      "loss": 0.0586,
+      "step": 1661
+    },
+    {
+      "epoch": 41.55,
+      "learning_rate": 1.0140000000000001e-05,
+      "loss": 0.075,
+      "step": 1662
+    },
+    {
+      "epoch": 41.58,
+      "learning_rate": 1.0110000000000001e-05,
+      "loss": 0.1022,
+      "step": 1663
+    },
+    {
+      "epoch": 41.6,
+      "learning_rate": 1.008e-05,
+      "loss": 0.0883,
+      "step": 1664
+    },
+    {
+      "epoch": 41.62,
+      "learning_rate": 1.005e-05,
+      "loss": 0.2111,
+      "step": 1665
+    },
+    {
+      "epoch": 41.65,
+      "learning_rate": 1.002e-05,
+      "loss": 0.0609,
+      "step": 1666
+    },
+    {
+      "epoch": 41.67,
+      "learning_rate": 9.990000000000001e-06,
+      "loss": 0.0365,
+      "step": 1667
+    },
+    {
+      "epoch": 41.7,
+      "learning_rate": 9.960000000000001e-06,
+      "loss": 0.0493,
+      "step": 1668
+    },
+    {
+      "epoch": 41.73,
+      "learning_rate": 9.930000000000001e-06,
+      "loss": 0.1346,
+      "step": 1669
+    },
+    {
+      "epoch": 41.75,
+      "learning_rate": 9.9e-06,
+      "loss": 0.0564,
+      "step": 1670
+    },
+    {
+      "epoch": 41.77,
+      "learning_rate": 9.87e-06,
+      "loss": 0.1128,
+      "step": 1671
+    },
+    {
+      "epoch": 41.8,
+      "learning_rate": 9.84e-06,
+      "loss": 0.1089,
+      "step": 1672
+    },
+    {
+      "epoch": 41.83,
+      "learning_rate": 9.810000000000001e-06,
+      "loss": 0.1535,
+      "step": 1673
+    },
+    {
+      "epoch": 41.85,
+      "learning_rate": 9.780000000000001e-06,
+      "loss": 0.0341,
+      "step": 1674
+    },
+    {
+      "epoch": 41.88,
+      "learning_rate": 9.75e-06,
+      "loss": 0.0231,
+      "step": 1675
+    },
+    {
+      "epoch": 41.9,
+      "learning_rate": 9.72e-06,
+      "loss": 0.0845,
+      "step": 1676
+    },
+    {
+      "epoch": 41.92,
+      "learning_rate": 9.69e-06,
+      "loss": 0.0239,
+      "step": 1677
+    },
+    {
+      "epoch": 41.95,
+      "learning_rate": 9.66e-06,
+      "loss": 0.0626,
+      "step": 1678
+    },
+    {
+      "epoch": 41.98,
+      "learning_rate": 9.630000000000001e-06,
+      "loss": 0.1257,
+      "step": 1679
+    },
+    {
+      "epoch": 42.0,
+      "learning_rate": 9.600000000000001e-06,
+      "loss": 0.0359,
+      "step": 1680
+    },
+    {
+      "epoch": 42.0,
+      "eval_loss": 0.14599038660526276,
+      "eval_mean_accuracy": 0.9393921557895011,
+      "eval_mean_iou": 0.5898119037201489,
+      "eval_overall_accuracy": 0.9414275172121546,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9430794921827349,
+        0.9357048193962673
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9275075761288636,
+        0.8419281350315833
+      ],
+      "eval_runtime": 49.0276,
+      "eval_samples_per_second": 0.408,
+      "eval_steps_per_second": 0.204,
+      "step": 1680
+    },
+    {
+      "epoch": 42.02,
+      "learning_rate": 9.57e-06,
+      "loss": 0.051,
+      "step": 1681
+    },
+    {
+      "epoch": 42.05,
+      "learning_rate": 9.54e-06,
+      "loss": 0.055,
+      "step": 1682
+    },
+    {
+      "epoch": 42.08,
+      "learning_rate": 9.51e-06,
+      "loss": 0.2484,
+      "step": 1683
+    },
+    {
+      "epoch": 42.1,
+      "learning_rate": 9.48e-06,
+      "loss": 0.0257,
+      "step": 1684
+    },
+    {
+      "epoch": 42.12,
+      "learning_rate": 9.450000000000001e-06,
+      "loss": 0.0349,
+      "step": 1685
+    },
+    {
+      "epoch": 42.15,
+      "learning_rate": 9.42e-06,
+      "loss": 0.0657,
+      "step": 1686
+    },
+    {
+      "epoch": 42.17,
+      "learning_rate": 9.39e-06,
+      "loss": 0.0777,
+      "step": 1687
+    },
+    {
+      "epoch": 42.2,
+      "learning_rate": 9.36e-06,
+      "loss": 0.2147,
+      "step": 1688
+    },
+    {
+      "epoch": 42.23,
+      "learning_rate": 9.33e-06,
+      "loss": 0.0289,
+      "step": 1689
+    },
+    {
+      "epoch": 42.25,
+      "learning_rate": 9.3e-06,
+      "loss": 0.0605,
+      "step": 1690
+    },
+    {
+      "epoch": 42.27,
+      "learning_rate": 9.27e-06,
+      "loss": 0.0459,
+      "step": 1691
+    },
+    {
+      "epoch": 42.3,
+      "learning_rate": 9.24e-06,
+      "loss": 0.0542,
+      "step": 1692
+    },
+    {
+      "epoch": 42.33,
+      "learning_rate": 9.21e-06,
+      "loss": 0.1233,
+      "step": 1693
+    },
+    {
+      "epoch": 42.35,
+      "learning_rate": 9.18e-06,
+      "loss": 0.0352,
+      "step": 1694
+    },
+    {
+      "epoch": 42.38,
+      "learning_rate": 9.15e-06,
+      "loss": 0.0286,
+      "step": 1695
+    },
+    {
+      "epoch": 42.4,
+      "learning_rate": 9.12e-06,
+      "loss": 0.0271,
+      "step": 1696
+    },
+    {
+      "epoch": 42.42,
+      "learning_rate": 9.09e-06,
+      "loss": 0.0366,
+      "step": 1697
+    },
+    {
+      "epoch": 42.45,
+      "learning_rate": 9.06e-06,
+      "loss": 0.102,
+      "step": 1698
+    },
+    {
+      "epoch": 42.48,
+      "learning_rate": 9.03e-06,
+      "loss": 0.0608,
+      "step": 1699
+    },
+    {
+      "epoch": 42.5,
+      "learning_rate": 9e-06,
+      "loss": 0.2009,
+      "step": 1700
+    },
+    {
+      "epoch": 42.5,
+      "eval_loss": 0.1421266496181488,
+      "eval_mean_accuracy": 0.9412907318947704,
+      "eval_mean_iou": 0.5838746378657184,
+      "eval_overall_accuracy": 0.9372294110728171,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9339330921632171,
+        0.9486483716263238
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.922518225606811,
+        0.829105687990344
+      ],
+      "eval_runtime": 48.4509,
+      "eval_samples_per_second": 0.413,
+      "eval_steps_per_second": 0.206,
+      "step": 1700
+    },
+    {
+      "epoch": 42.52,
+      "learning_rate": 8.97e-06,
+      "loss": 0.093,
+      "step": 1701
+    },
+    {
+      "epoch": 42.55,
+      "learning_rate": 8.939999999999999e-06,
+      "loss": 0.1562,
+      "step": 1702
+    },
+    {
+      "epoch": 42.58,
+      "learning_rate": 8.91e-06,
+      "loss": 0.0256,
+      "step": 1703
+    },
+    {
+      "epoch": 42.6,
+      "learning_rate": 8.88e-06,
+      "loss": 0.0778,
+      "step": 1704
+    },
+    {
+      "epoch": 42.62,
+      "learning_rate": 8.85e-06,
+      "loss": 0.1148,
+      "step": 1705
+    },
+    {
+      "epoch": 42.65,
+      "learning_rate": 8.82e-06,
+      "loss": 1.5135,
+      "step": 1706
+    },
+    {
+      "epoch": 42.67,
+      "learning_rate": 8.79e-06,
+      "loss": 0.1192,
+      "step": 1707
+    },
+    {
+      "epoch": 42.7,
+      "learning_rate": 8.759999999999999e-06,
+      "loss": 0.1799,
+      "step": 1708
+    },
+    {
+      "epoch": 42.73,
+      "learning_rate": 8.73e-06,
+      "loss": 0.1268,
+      "step": 1709
+    },
+    {
+      "epoch": 42.75,
+      "learning_rate": 8.7e-06,
+      "loss": 0.225,
+      "step": 1710
+    },
+    {
+      "epoch": 42.77,
+      "learning_rate": 8.67e-06,
+      "loss": 0.0298,
+      "step": 1711
+    },
+    {
+      "epoch": 42.8,
+      "learning_rate": 8.64e-06,
+      "loss": 0.0488,
+      "step": 1712
+    },
+    {
+      "epoch": 42.83,
+      "learning_rate": 8.609999999999999e-06,
+      "loss": 0.1025,
+      "step": 1713
+    },
+    {
+      "epoch": 42.85,
+      "learning_rate": 8.58e-06,
+      "loss": 0.041,
+      "step": 1714
+    },
+    {
+      "epoch": 42.88,
+      "learning_rate": 8.55e-06,
+      "loss": 0.0534,
+      "step": 1715
+    },
+    {
+      "epoch": 42.9,
+      "learning_rate": 8.52e-06,
+      "loss": 0.0614,
+      "step": 1716
+    },
+    {
+      "epoch": 42.92,
+      "learning_rate": 8.49e-06,
+      "loss": 0.0666,
+      "step": 1717
+    },
+    {
+      "epoch": 42.95,
+      "learning_rate": 8.459999999999999e-06,
+      "loss": 0.0889,
+      "step": 1718
+    },
+    {
+      "epoch": 42.98,
+      "learning_rate": 8.43e-06,
+      "loss": 0.0423,
+      "step": 1719
+    },
+    {
+      "epoch": 43.0,
+      "learning_rate": 8.400000000000001e-06,
+      "loss": 0.0574,
+      "step": 1720
+    },
+    {
+      "epoch": 43.0,
+      "eval_loss": 0.1485082358121872,
+      "eval_mean_accuracy": 0.9346000768787144,
+      "eval_mean_iou": 0.5704930768355303,
+      "eval_overall_accuracy": 0.9248018568019631,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9168492572572947,
+        0.9523508965001339
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9065900984529122,
+        0.8048891320536786
+      ],
+      "eval_runtime": 48.0387,
+      "eval_samples_per_second": 0.416,
+      "eval_steps_per_second": 0.208,
+      "step": 1720
+    },
+    {
+      "epoch": 43.02,
+      "learning_rate": 8.370000000000001e-06,
+      "loss": 0.0588,
+      "step": 1721
+    },
+    {
+      "epoch": 43.05,
+      "learning_rate": 8.340000000000001e-06,
+      "loss": 0.0348,
+      "step": 1722
+    },
+    {
+      "epoch": 43.08,
+      "learning_rate": 8.310000000000002e-06,
+      "loss": 0.0472,
+      "step": 1723
+    },
+    {
+      "epoch": 43.1,
+      "learning_rate": 8.28e-06,
+      "loss": 0.1446,
+      "step": 1724
+    },
+    {
+      "epoch": 43.12,
+      "learning_rate": 8.25e-06,
+      "loss": 0.0406,
+      "step": 1725
+    },
+    {
+      "epoch": 43.15,
+      "learning_rate": 8.220000000000001e-06,
+      "loss": 0.1688,
+      "step": 1726
+    },
+    {
+      "epoch": 43.17,
+      "learning_rate": 8.190000000000001e-06,
+      "loss": 0.0478,
+      "step": 1727
+    },
+    {
+      "epoch": 43.2,
+      "learning_rate": 8.160000000000001e-06,
+      "loss": 0.1769,
+      "step": 1728
+    },
+    {
+      "epoch": 43.23,
+      "learning_rate": 8.13e-06,
+      "loss": 0.1665,
+      "step": 1729
+    },
+    {
+      "epoch": 43.25,
+      "learning_rate": 8.1e-06,
+      "loss": 0.0912,
+      "step": 1730
+    },
+    {
+      "epoch": 43.27,
+      "learning_rate": 8.07e-06,
+      "loss": 0.0239,
+      "step": 1731
+    },
+    {
+      "epoch": 43.3,
+      "learning_rate": 8.040000000000001e-06,
+      "loss": 0.0439,
+      "step": 1732
+    },
+    {
+      "epoch": 43.33,
+      "learning_rate": 8.010000000000001e-06,
+      "loss": 0.1336,
+      "step": 1733
+    },
+    {
+      "epoch": 43.35,
+      "learning_rate": 7.98e-06,
+      "loss": 0.0304,
+      "step": 1734
+    },
+    {
+      "epoch": 43.38,
+      "learning_rate": 7.95e-06,
+      "loss": 0.0585,
+      "step": 1735
+    },
+    {
+      "epoch": 43.4,
+      "learning_rate": 7.92e-06,
+      "loss": 0.0715,
+      "step": 1736
+    },
+    {
+      "epoch": 43.42,
+      "learning_rate": 7.89e-06,
+      "loss": 0.1494,
+      "step": 1737
+    },
+    {
+      "epoch": 43.45,
+      "learning_rate": 7.860000000000001e-06,
+      "loss": 0.1154,
+      "step": 1738
+    },
+    {
+      "epoch": 43.48,
+      "learning_rate": 7.830000000000001e-06,
+      "loss": 0.0471,
+      "step": 1739
+    },
+    {
+      "epoch": 43.5,
+      "learning_rate": 7.8e-06,
+      "loss": 0.106,
+      "step": 1740
+    },
+    {
+      "epoch": 43.5,
+      "eval_loss": 0.14649371802806854,
+      "eval_mean_accuracy": 0.9331176082208781,
+      "eval_mean_iou": 0.5745408333682841,
+      "eval_overall_accuracy": 0.9258302357094405,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9199155333591841,
+        0.9463196830825722
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9081253343664103,
+        0.8154971657384421
+      ],
+      "eval_runtime": 47.5858,
+      "eval_samples_per_second": 0.42,
+      "eval_steps_per_second": 0.21,
+      "step": 1740
+    },
+    {
+      "epoch": 43.52,
+      "learning_rate": 7.77e-06,
+      "loss": 0.1116,
+      "step": 1741
+    },
+    {
+      "epoch": 43.55,
+      "learning_rate": 7.74e-06,
+      "loss": 0.1824,
+      "step": 1742
+    },
+    {
+      "epoch": 43.58,
+      "learning_rate": 7.71e-06,
+      "loss": 0.2358,
+      "step": 1743
+    },
+    {
+      "epoch": 43.6,
+      "learning_rate": 7.680000000000001e-06,
+      "loss": 0.0334,
+      "step": 1744
+    },
+    {
+      "epoch": 43.62,
+      "learning_rate": 7.65e-06,
+      "loss": 0.0929,
+      "step": 1745
+    },
+    {
+      "epoch": 43.65,
+      "learning_rate": 7.62e-06,
+      "loss": 0.0384,
+      "step": 1746
+    },
+    {
+      "epoch": 43.67,
+      "learning_rate": 7.59e-06,
+      "loss": 0.0364,
+      "step": 1747
+    },
+    {
+      "epoch": 43.7,
+      "learning_rate": 7.5600000000000005e-06,
+      "loss": 0.1077,
+      "step": 1748
+    },
+    {
+      "epoch": 43.73,
+      "learning_rate": 7.53e-06,
+      "loss": 0.1359,
+      "step": 1749
+    },
+    {
+      "epoch": 43.75,
+      "learning_rate": 7.5e-06,
+      "loss": 0.111,
+      "step": 1750
+    },
+    {
+      "epoch": 43.77,
+      "learning_rate": 7.4700000000000005e-06,
+      "loss": 0.1023,
+      "step": 1751
+    },
+    {
+      "epoch": 43.8,
+      "learning_rate": 7.44e-06,
+      "loss": 0.0789,
+      "step": 1752
+    },
+    {
+      "epoch": 43.83,
+      "learning_rate": 7.41e-06,
+      "loss": 0.2633,
+      "step": 1753
+    },
+    {
+      "epoch": 43.85,
+      "learning_rate": 7.3800000000000005e-06,
+      "loss": 0.094,
+      "step": 1754
+    },
+    {
+      "epoch": 43.88,
+      "learning_rate": 7.35e-06,
+      "loss": 0.049,
+      "step": 1755
+    },
+    {
+      "epoch": 43.9,
+      "learning_rate": 7.32e-06,
+      "loss": 0.0599,
+      "step": 1756
+    },
+    {
+      "epoch": 43.92,
+      "learning_rate": 7.29e-06,
+      "loss": 0.1121,
+      "step": 1757
+    },
+    {
+      "epoch": 43.95,
+      "learning_rate": 7.26e-06,
+      "loss": 0.1018,
+      "step": 1758
+    },
+    {
+      "epoch": 43.98,
+      "learning_rate": 7.23e-06,
+      "loss": 0.0733,
+      "step": 1759
+    },
+    {
+      "epoch": 44.0,
+      "learning_rate": 7.2e-06,
+      "loss": 0.1341,
+      "step": 1760
+    },
+    {
+      "epoch": 44.0,
+      "eval_loss": 0.16505067050457,
+      "eval_mean_accuracy": 0.9306736867728073,
+      "eval_mean_iou": 0.5589558450764843,
+      "eval_overall_accuracy": 0.9171597084930483,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9061912615024286,
+        0.955156112043186
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.8962234074503034,
+        0.7806441277791495
+      ],
+      "eval_runtime": 48.0696,
+      "eval_samples_per_second": 0.416,
+      "eval_steps_per_second": 0.208,
+      "step": 1760
+    },
+    {
+      "epoch": 44.02,
+      "learning_rate": 7.17e-06,
+      "loss": 0.0613,
+      "step": 1761
+    },
+    {
+      "epoch": 44.05,
+      "learning_rate": 7.14e-06,
+      "loss": 0.031,
+      "step": 1762
+    },
+    {
+      "epoch": 44.08,
+      "learning_rate": 7.11e-06,
+      "loss": 0.0271,
+      "step": 1763
+    },
+    {
+      "epoch": 44.1,
+      "learning_rate": 7.08e-06,
+      "loss": 0.0994,
+      "step": 1764
+    },
+    {
+      "epoch": 44.12,
+      "learning_rate": 7.049999999999999e-06,
+      "loss": 0.1717,
+      "step": 1765
+    },
+    {
+      "epoch": 44.15,
+      "learning_rate": 7.0200000000000006e-06,
+      "loss": 0.0398,
+      "step": 1766
+    },
+    {
+      "epoch": 44.17,
+      "learning_rate": 6.990000000000001e-06,
+      "loss": 0.0975,
+      "step": 1767
+    },
+    {
+      "epoch": 44.2,
+      "learning_rate": 6.96e-06,
+      "loss": 0.6028,
+      "step": 1768
+    },
+    {
+      "epoch": 44.23,
+      "learning_rate": 6.9300000000000006e-06,
+      "loss": 0.0812,
+      "step": 1769
+    },
+    {
+      "epoch": 44.25,
+      "learning_rate": 6.900000000000001e-06,
+      "loss": 0.0766,
+      "step": 1770
+    },
+    {
+      "epoch": 44.27,
+      "learning_rate": 6.87e-06,
+      "loss": 0.0472,
+      "step": 1771
+    },
+    {
+      "epoch": 44.3,
+      "learning_rate": 6.840000000000001e-06,
+      "loss": 0.0552,
+      "step": 1772
+    },
+    {
+      "epoch": 44.33,
+      "learning_rate": 6.81e-06,
+      "loss": 0.1588,
+      "step": 1773
+    },
+    {
+      "epoch": 44.35,
+      "learning_rate": 6.78e-06,
+      "loss": 0.1295,
+      "step": 1774
+    },
+    {
+      "epoch": 44.38,
+      "learning_rate": 6.750000000000001e-06,
+      "loss": 0.0774,
+      "step": 1775
+    },
+    {
+      "epoch": 44.4,
+      "learning_rate": 6.72e-06,
+      "loss": 0.0959,
+      "step": 1776
+    },
+    {
+      "epoch": 44.42,
+      "learning_rate": 6.69e-06,
+      "loss": 0.0386,
+      "step": 1777
+    },
+    {
+      "epoch": 44.45,
+      "learning_rate": 6.660000000000001e-06,
+      "loss": 0.0292,
+      "step": 1778
+    },
+    {
+      "epoch": 44.48,
+      "learning_rate": 6.63e-06,
+      "loss": 0.1,
+      "step": 1779
+    },
+    {
+      "epoch": 44.5,
+      "learning_rate": 6.6e-06,
+      "loss": 0.1738,
+      "step": 1780
+    },
+    {
+      "epoch": 44.5,
+      "eval_loss": 0.15625318884849548,
+      "eval_mean_accuracy": 0.9318418501476647,
+      "eval_mean_iou": 0.5740119547275936,
+      "eval_overall_accuracy": 0.9277452037181154,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9244202132325879,
+        0.9392634870627415
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9104285326477329,
+        0.8116073315350479
+      ],
+      "eval_runtime": 47.1442,
+      "eval_samples_per_second": 0.424,
+      "eval_steps_per_second": 0.212,
+      "step": 1780
+    },
+    {
+      "epoch": 44.52,
+      "learning_rate": 6.57e-06,
+      "loss": 0.1046,
+      "step": 1781
+    },
+    {
+      "epoch": 44.55,
+      "learning_rate": 6.54e-06,
+      "loss": 0.1925,
+      "step": 1782
+    },
+    {
+      "epoch": 44.58,
+      "learning_rate": 6.51e-06,
+      "loss": 0.0939,
+      "step": 1783
+    },
+    {
+      "epoch": 44.6,
+      "learning_rate": 6.48e-06,
+      "loss": 0.3321,
+      "step": 1784
+    },
+    {
+      "epoch": 44.62,
+      "learning_rate": 6.45e-06,
+      "loss": 0.0773,
+      "step": 1785
+    },
+    {
+      "epoch": 44.65,
+      "learning_rate": 6.42e-06,
+      "loss": 0.1684,
+      "step": 1786
+    },
+    {
+      "epoch": 44.67,
+      "learning_rate": 6.39e-06,
+      "loss": 0.0742,
+      "step": 1787
+    },
+    {
+      "epoch": 44.7,
+      "learning_rate": 6.36e-06,
+      "loss": 0.0528,
+      "step": 1788
+    },
+    {
+      "epoch": 44.73,
+      "learning_rate": 6.3299999999999995e-06,
+      "loss": 0.1353,
+      "step": 1789
+    },
+    {
+      "epoch": 44.75,
+      "learning_rate": 6.3e-06,
+      "loss": 0.0947,
+      "step": 1790
+    },
+    {
+      "epoch": 44.77,
+      "learning_rate": 6.27e-06,
+      "loss": 0.1515,
+      "step": 1791
+    },
+    {
+      "epoch": 44.8,
+      "learning_rate": 6.2399999999999995e-06,
+      "loss": 0.1119,
+      "step": 1792
+    },
+    {
+      "epoch": 44.83,
+      "learning_rate": 6.21e-06,
+      "loss": 0.0368,
+      "step": 1793
+    },
+    {
+      "epoch": 44.85,
+      "learning_rate": 6.18e-06,
+      "loss": 0.0461,
+      "step": 1794
+    },
+    {
+      "epoch": 44.88,
+      "learning_rate": 6.1499999999999996e-06,
+      "loss": 0.0454,
+      "step": 1795
+    },
+    {
+      "epoch": 44.9,
+      "learning_rate": 6.12e-06,
+      "loss": 0.0755,
+      "step": 1796
+    },
+    {
+      "epoch": 44.92,
+      "learning_rate": 6.090000000000001e-06,
+      "loss": 0.1835,
+      "step": 1797
+    },
+    {
+      "epoch": 44.95,
+      "learning_rate": 6.0600000000000004e-06,
+      "loss": 0.1545,
+      "step": 1798
+    },
+    {
+      "epoch": 44.98,
+      "learning_rate": 6.030000000000001e-06,
+      "loss": 0.202,
+      "step": 1799
+    },
+    {
+      "epoch": 45.0,
+      "learning_rate": 6e-06,
+      "loss": 0.0251,
+      "step": 1800
+    },
+    {
+      "epoch": 45.0,
+      "eval_loss": 0.15182256698608398,
+      "eval_mean_accuracy": 0.937307797490532,
+      "eval_mean_iou": 0.578238774842495,
+      "eval_overall_accuracy": 0.9316633555524405,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9270821167115941,
+        0.9475334782694698
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9154576111677516,
+        0.8192587133597334
+      ],
+      "eval_runtime": 48.1959,
+      "eval_samples_per_second": 0.415,
+      "eval_steps_per_second": 0.207,
+      "step": 1800
+    },
+    {
+      "epoch": 45.02,
+      "learning_rate": 5.9700000000000004e-06,
+      "loss": 0.0351,
+      "step": 1801
+    },
+    {
+      "epoch": 45.05,
+      "learning_rate": 5.940000000000001e-06,
+      "loss": 0.1755,
+      "step": 1802
+    },
+    {
+      "epoch": 45.08,
+      "learning_rate": 5.91e-06,
+      "loss": 0.161,
+      "step": 1803
+    },
+    {
+      "epoch": 45.1,
+      "learning_rate": 5.8800000000000005e-06,
+      "loss": 0.0729,
+      "step": 1804
+    },
+    {
+      "epoch": 45.12,
+      "learning_rate": 5.850000000000001e-06,
+      "loss": 0.0269,
+      "step": 1805
+    },
+    {
+      "epoch": 45.15,
+      "learning_rate": 5.82e-06,
+      "loss": 0.0656,
+      "step": 1806
+    },
+    {
+      "epoch": 45.17,
+      "learning_rate": 5.7900000000000005e-06,
+      "loss": 0.0598,
+      "step": 1807
+    },
+    {
+      "epoch": 45.2,
+      "learning_rate": 5.76e-06,
+      "loss": 0.0766,
+      "step": 1808
+    },
+    {
+      "epoch": 45.23,
+      "learning_rate": 5.73e-06,
+      "loss": 0.1156,
+      "step": 1809
+    },
+    {
+      "epoch": 45.25,
+      "learning_rate": 5.7000000000000005e-06,
+      "loss": 0.0996,
+      "step": 1810
+    },
+    {
+      "epoch": 45.27,
+      "learning_rate": 5.67e-06,
+      "loss": 0.0189,
+      "step": 1811
+    },
+    {
+      "epoch": 45.3,
+      "learning_rate": 5.64e-06,
+      "loss": 0.0563,
+      "step": 1812
+    },
+    {
+      "epoch": 45.33,
+      "learning_rate": 5.6100000000000005e-06,
+      "loss": 0.0349,
+      "step": 1813
+    },
+    {
+      "epoch": 45.35,
+      "learning_rate": 5.58e-06,
+      "loss": 0.0525,
+      "step": 1814
+    },
+    {
+      "epoch": 45.38,
+      "learning_rate": 5.55e-06,
+      "loss": 0.1407,
+      "step": 1815
+    },
+    {
+      "epoch": 45.4,
+      "learning_rate": 5.52e-06,
+      "loss": 0.0265,
+      "step": 1816
+    },
+    {
+      "epoch": 45.42,
+      "learning_rate": 5.49e-06,
+      "loss": 0.0289,
+      "step": 1817
+    },
+    {
+      "epoch": 45.45,
+      "learning_rate": 5.46e-06,
+      "loss": 0.0685,
+      "step": 1818
+    },
+    {
+      "epoch": 45.48,
+      "learning_rate": 5.43e-06,
+      "loss": 0.1497,
+      "step": 1819
+    },
+    {
+      "epoch": 45.5,
+      "learning_rate": 5.4e-06,
+      "loss": 0.0487,
+      "step": 1820
+    },
+    {
+      "epoch": 45.5,
+      "eval_loss": 0.15128853917121887,
+      "eval_mean_accuracy": 0.9431732620750322,
+      "eval_mean_iou": 0.5870457363192317,
+      "eval_overall_accuracy": 0.940875664841123,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9390108495406815,
+        0.9473356746093827
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9267849951891893,
+        0.8343522137685061
+      ],
+      "eval_runtime": 48.3548,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 1820
+    },
+    {
+      "epoch": 45.52,
+      "learning_rate": 5.37e-06,
+      "loss": 0.1315,
+      "step": 1821
+    },
+    {
+      "epoch": 45.55,
+      "learning_rate": 5.34e-06,
+      "loss": 0.1397,
+      "step": 1822
+    },
+    {
+      "epoch": 45.58,
+      "learning_rate": 5.31e-06,
+      "loss": 0.0832,
+      "step": 1823
+    },
+    {
+      "epoch": 45.6,
+      "learning_rate": 5.279999999999999e-06,
+      "loss": 0.0754,
+      "step": 1824
+    },
+    {
+      "epoch": 45.62,
+      "learning_rate": 5.25e-06,
+      "loss": 0.0928,
+      "step": 1825
+    },
+    {
+      "epoch": 45.65,
+      "learning_rate": 5.22e-06,
+      "loss": 0.155,
+      "step": 1826
+    },
+    {
+      "epoch": 45.67,
+      "learning_rate": 5.1899999999999994e-06,
+      "loss": 0.0413,
+      "step": 1827
+    },
+    {
+      "epoch": 45.7,
+      "learning_rate": 5.16e-06,
+      "loss": 0.1205,
+      "step": 1828
+    },
+    {
+      "epoch": 45.73,
+      "learning_rate": 5.130000000000001e-06,
+      "loss": 0.1194,
+      "step": 1829
+    },
+    {
+      "epoch": 45.75,
+      "learning_rate": 5.1e-06,
+      "loss": 0.0306,
+      "step": 1830
+    },
+    {
+      "epoch": 45.77,
+      "learning_rate": 5.070000000000001e-06,
+      "loss": 0.0694,
+      "step": 1831
+    },
+    {
+      "epoch": 45.8,
+      "learning_rate": 5.04e-06,
+      "loss": 0.1616,
+      "step": 1832
+    },
+    {
+      "epoch": 45.83,
+      "learning_rate": 5.01e-06,
+      "loss": 0.0481,
+      "step": 1833
+    },
+    {
+      "epoch": 45.85,
+      "learning_rate": 4.980000000000001e-06,
+      "loss": 0.0345,
+      "step": 1834
+    },
+    {
+      "epoch": 45.88,
+      "learning_rate": 4.95e-06,
+      "loss": 0.1863,
+      "step": 1835
+    },
+    {
+      "epoch": 45.9,
+      "learning_rate": 4.92e-06,
+      "loss": 0.0529,
+      "step": 1836
+    },
+    {
+      "epoch": 45.92,
+      "learning_rate": 4.890000000000001e-06,
+      "loss": 0.0424,
+      "step": 1837
+    },
+    {
+      "epoch": 45.95,
+      "learning_rate": 4.86e-06,
+      "loss": 0.0836,
+      "step": 1838
+    },
+    {
+      "epoch": 45.98,
+      "learning_rate": 4.83e-06,
+      "loss": 0.0306,
+      "step": 1839
+    },
+    {
+      "epoch": 46.0,
+      "learning_rate": 4.800000000000001e-06,
+      "loss": 0.0932,
+      "step": 1840
+    },
+    {
+      "epoch": 46.0,
+      "eval_loss": 0.1517506241798401,
+      "eval_mean_accuracy": 0.9437667305584003,
+      "eval_mean_iou": 0.5830809997646579,
+      "eval_overall_accuracy": 0.9379355404205679,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9332027298007051,
+        0.9543307313160957
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9227875806392004,
+        0.8264554186547733
+      ],
+      "eval_runtime": 48.2712,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 1840
+    },
+    {
+      "epoch": 46.02,
+      "learning_rate": 4.77e-06,
+      "loss": 0.0696,
+      "step": 1841
+    },
+    {
+      "epoch": 46.05,
+      "learning_rate": 4.74e-06,
+      "loss": 0.0915,
+      "step": 1842
+    },
+    {
+      "epoch": 46.08,
+      "learning_rate": 4.71e-06,
+      "loss": 0.0352,
+      "step": 1843
+    },
+    {
+      "epoch": 46.1,
+      "learning_rate": 4.68e-06,
+      "loss": 0.1473,
+      "step": 1844
+    },
+    {
+      "epoch": 46.12,
+      "learning_rate": 4.65e-06,
+      "loss": 0.0575,
+      "step": 1845
+    },
+    {
+      "epoch": 46.15,
+      "learning_rate": 4.62e-06,
+      "loss": 0.1291,
+      "step": 1846
+    },
+    {
+      "epoch": 46.17,
+      "learning_rate": 4.59e-06,
+      "loss": 0.0613,
+      "step": 1847
+    },
+    {
+      "epoch": 46.2,
+      "learning_rate": 4.56e-06,
+      "loss": 0.0619,
+      "step": 1848
+    },
+    {
+      "epoch": 46.23,
+      "learning_rate": 4.53e-06,
+      "loss": 0.0901,
+      "step": 1849
+    },
+    {
+      "epoch": 46.25,
+      "learning_rate": 4.5e-06,
+      "loss": 0.0596,
+      "step": 1850
+    },
+    {
+      "epoch": 46.27,
+      "learning_rate": 4.4699999999999996e-06,
+      "loss": 0.0411,
+      "step": 1851
+    },
+    {
+      "epoch": 46.3,
+      "learning_rate": 4.44e-06,
+      "loss": 0.1174,
+      "step": 1852
+    },
+    {
+      "epoch": 46.33,
+      "learning_rate": 4.41e-06,
+      "loss": 0.086,
+      "step": 1853
+    },
+    {
+      "epoch": 46.35,
+      "learning_rate": 4.3799999999999996e-06,
+      "loss": 0.076,
+      "step": 1854
+    },
+    {
+      "epoch": 46.38,
+      "learning_rate": 4.35e-06,
+      "loss": 0.0298,
+      "step": 1855
+    },
+    {
+      "epoch": 46.4,
+      "learning_rate": 4.32e-06,
+      "loss": 0.0885,
+      "step": 1856
+    },
+    {
+      "epoch": 46.42,
+      "learning_rate": 4.29e-06,
+      "loss": 0.0525,
+      "step": 1857
+    },
+    {
+      "epoch": 46.45,
+      "learning_rate": 4.26e-06,
+      "loss": 0.0937,
+      "step": 1858
+    },
+    {
+      "epoch": 46.48,
+      "learning_rate": 4.229999999999999e-06,
+      "loss": 0.0621,
+      "step": 1859
+    },
+    {
+      "epoch": 46.5,
+      "learning_rate": 4.2000000000000004e-06,
+      "loss": 0.0885,
+      "step": 1860
+    },
+    {
+      "epoch": 46.5,
+      "eval_loss": 0.16364796459674835,
+      "eval_mean_accuracy": 0.941614636883882,
+      "eval_mean_iou": 0.5908998284465385,
+      "eval_overall_accuracy": 0.9494446834462013,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9557998402235002,
+        0.927429433544264
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9379215775366475,
+        0.8347779078029681
+      ],
+      "eval_runtime": 47.8707,
+      "eval_samples_per_second": 0.418,
+      "eval_steps_per_second": 0.209,
+      "step": 1860
+    },
+    {
+      "epoch": 46.52,
+      "learning_rate": 4.170000000000001e-06,
+      "loss": 0.1059,
+      "step": 1861
+    },
+    {
+      "epoch": 46.55,
+      "learning_rate": 4.14e-06,
+      "loss": 0.0989,
+      "step": 1862
+    },
+    {
+      "epoch": 46.58,
+      "learning_rate": 4.1100000000000005e-06,
+      "loss": 0.1744,
+      "step": 1863
+    },
+    {
+      "epoch": 46.6,
+      "learning_rate": 4.080000000000001e-06,
+      "loss": 0.1014,
+      "step": 1864
+    },
+    {
+      "epoch": 46.62,
+      "learning_rate": 4.05e-06,
+      "loss": 0.1141,
+      "step": 1865
+    },
+    {
+      "epoch": 46.65,
+      "learning_rate": 4.0200000000000005e-06,
+      "loss": 0.0206,
+      "step": 1866
+    },
+    {
+      "epoch": 46.67,
+      "learning_rate": 3.99e-06,
+      "loss": 0.0408,
+      "step": 1867
+    },
+    {
+      "epoch": 46.7,
+      "learning_rate": 3.96e-06,
+      "loss": 0.1098,
+      "step": 1868
+    },
+    {
+      "epoch": 46.73,
+      "learning_rate": 3.9300000000000005e-06,
+      "loss": 0.0255,
+      "step": 1869
+    },
+    {
+      "epoch": 46.75,
+      "learning_rate": 3.9e-06,
+      "loss": 0.0371,
+      "step": 1870
+    },
+    {
+      "epoch": 46.77,
+      "learning_rate": 3.87e-06,
+      "loss": 0.0625,
+      "step": 1871
+    },
+    {
+      "epoch": 46.8,
+      "learning_rate": 3.8400000000000005e-06,
+      "loss": 0.0394,
+      "step": 1872
+    },
+    {
+      "epoch": 46.83,
+      "learning_rate": 3.81e-06,
+      "loss": 0.1487,
+      "step": 1873
+    },
+    {
+      "epoch": 46.85,
+      "learning_rate": 3.7800000000000002e-06,
+      "loss": 0.0385,
+      "step": 1874
+    },
+    {
+      "epoch": 46.88,
+      "learning_rate": 3.75e-06,
+      "loss": 0.1142,
+      "step": 1875
+    },
+    {
+      "epoch": 46.9,
+      "learning_rate": 3.72e-06,
+      "loss": 0.0552,
+      "step": 1876
+    },
+    {
+      "epoch": 46.92,
+      "learning_rate": 3.6900000000000002e-06,
+      "loss": 0.1162,
+      "step": 1877
+    },
+    {
+      "epoch": 46.95,
+      "learning_rate": 3.66e-06,
+      "loss": 0.1338,
+      "step": 1878
+    },
+    {
+      "epoch": 46.98,
+      "learning_rate": 3.63e-06,
+      "loss": 0.0906,
+      "step": 1879
+    },
+    {
+      "epoch": 47.0,
+      "learning_rate": 3.6e-06,
+      "loss": 0.0643,
+      "step": 1880
+    },
+    {
+      "epoch": 47.0,
+      "eval_loss": 0.1605922281742096,
+      "eval_mean_accuracy": 0.9404316678429687,
+      "eval_mean_iou": 0.5871183536514295,
+      "eval_overall_accuracy": 0.9455639931231944,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9497295790527714,
+        0.9311337566331659
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9331293938120437,
+        0.8282256671422448
+      ],
+      "eval_runtime": 48.1296,
+      "eval_samples_per_second": 0.416,
+      "eval_steps_per_second": 0.208,
+      "step": 1880
+    },
+    {
+      "epoch": 47.02,
+      "learning_rate": 3.57e-06,
+      "loss": 0.2083,
+      "step": 1881
+    },
+    {
+      "epoch": 47.05,
+      "learning_rate": 3.54e-06,
+      "loss": 0.1619,
+      "step": 1882
+    },
+    {
+      "epoch": 47.08,
+      "learning_rate": 3.5100000000000003e-06,
+      "loss": 0.2476,
+      "step": 1883
+    },
+    {
+      "epoch": 47.1,
+      "learning_rate": 3.48e-06,
+      "loss": 0.0717,
+      "step": 1884
+    },
+    {
+      "epoch": 47.12,
+      "learning_rate": 3.4500000000000004e-06,
+      "loss": 0.0679,
+      "step": 1885
+    },
+    {
+      "epoch": 47.15,
+      "learning_rate": 3.4200000000000003e-06,
+      "loss": 0.1976,
+      "step": 1886
+    },
+    {
+      "epoch": 47.17,
+      "learning_rate": 3.39e-06,
+      "loss": 0.0193,
+      "step": 1887
+    },
+    {
+      "epoch": 47.2,
+      "learning_rate": 3.36e-06,
+      "loss": 0.1,
+      "step": 1888
+    },
+    {
+      "epoch": 47.23,
+      "learning_rate": 3.3300000000000003e-06,
+      "loss": 0.1271,
+      "step": 1889
+    },
+    {
+      "epoch": 47.25,
+      "learning_rate": 3.3e-06,
+      "loss": 0.0526,
+      "step": 1890
+    },
+    {
+      "epoch": 47.27,
+      "learning_rate": 3.27e-06,
+      "loss": 0.0896,
+      "step": 1891
+    },
+    {
+      "epoch": 47.3,
+      "learning_rate": 3.24e-06,
+      "loss": 0.0718,
+      "step": 1892
+    },
+    {
+      "epoch": 47.33,
+      "learning_rate": 3.21e-06,
+      "loss": 0.0524,
+      "step": 1893
+    },
+    {
+      "epoch": 47.35,
+      "learning_rate": 3.18e-06,
+      "loss": 0.0587,
+      "step": 1894
+    },
+    {
+      "epoch": 47.38,
+      "learning_rate": 3.15e-06,
+      "loss": 0.2194,
+      "step": 1895
+    },
+    {
+      "epoch": 47.4,
+      "learning_rate": 3.1199999999999998e-06,
+      "loss": 0.0847,
+      "step": 1896
+    },
+    {
+      "epoch": 47.42,
+      "learning_rate": 3.09e-06,
+      "loss": 0.0721,
+      "step": 1897
+    },
+    {
+      "epoch": 47.45,
+      "learning_rate": 3.06e-06,
+      "loss": 0.1872,
+      "step": 1898
+    },
+    {
+      "epoch": 47.48,
+      "learning_rate": 3.0300000000000002e-06,
+      "loss": 0.132,
+      "step": 1899
+    },
+    {
+      "epoch": 47.5,
+      "learning_rate": 3e-06,
+      "loss": 0.0604,
+      "step": 1900
+    },
+    {
+      "epoch": 47.5,
+      "eval_loss": 0.15312716364860535,
+      "eval_mean_accuracy": 0.9393063785994833,
+      "eval_mean_iou": 0.5808654342069676,
+      "eval_overall_accuracy": 0.937570189982228,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.936161034778641,
+        0.9424517224203256
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9229050959745889,
+        0.819691206646314
+      ],
+      "eval_runtime": 48.4321,
+      "eval_samples_per_second": 0.413,
+      "eval_steps_per_second": 0.206,
+      "step": 1900
+    },
+    {
+      "epoch": 47.52,
+      "learning_rate": 2.9700000000000004e-06,
+      "loss": 0.0585,
+      "step": 1901
+    },
+    {
+      "epoch": 47.55,
+      "learning_rate": 2.9400000000000002e-06,
+      "loss": 0.0932,
+      "step": 1902
+    },
+    {
+      "epoch": 47.58,
+      "learning_rate": 2.91e-06,
+      "loss": 0.0359,
+      "step": 1903
+    },
+    {
+      "epoch": 47.6,
+      "learning_rate": 2.88e-06,
+      "loss": 0.0869,
+      "step": 1904
+    },
+    {
+      "epoch": 47.62,
+      "learning_rate": 2.8500000000000002e-06,
+      "loss": 0.1431,
+      "step": 1905
+    },
+    {
+      "epoch": 47.65,
+      "learning_rate": 2.82e-06,
+      "loss": 0.0727,
+      "step": 1906
+    },
+    {
+      "epoch": 47.67,
+      "learning_rate": 2.79e-06,
+      "loss": 0.038,
+      "step": 1907
+    },
+    {
+      "epoch": 47.7,
+      "learning_rate": 2.76e-06,
+      "loss": 0.0579,
+      "step": 1908
+    },
+    {
+      "epoch": 47.73,
+      "learning_rate": 2.73e-06,
+      "loss": 0.0511,
+      "step": 1909
+    },
+    {
+      "epoch": 47.75,
+      "learning_rate": 2.7e-06,
+      "loss": 0.1041,
+      "step": 1910
+    },
+    {
+      "epoch": 47.77,
+      "learning_rate": 2.67e-06,
+      "loss": 0.0608,
+      "step": 1911
+    },
+    {
+      "epoch": 47.8,
+      "learning_rate": 2.6399999999999997e-06,
+      "loss": 0.0439,
+      "step": 1912
+    },
+    {
+      "epoch": 47.83,
+      "learning_rate": 2.61e-06,
+      "loss": 0.1331,
+      "step": 1913
+    },
+    {
+      "epoch": 47.85,
+      "learning_rate": 2.58e-06,
+      "loss": 0.1337,
+      "step": 1914
+    },
+    {
+      "epoch": 47.88,
+      "learning_rate": 2.55e-06,
+      "loss": 0.1342,
+      "step": 1915
+    },
+    {
+      "epoch": 47.9,
+      "learning_rate": 2.52e-06,
+      "loss": 0.077,
+      "step": 1916
+    },
+    {
+      "epoch": 47.92,
+      "learning_rate": 2.4900000000000003e-06,
+      "loss": 0.1093,
+      "step": 1917
+    },
+    {
+      "epoch": 47.95,
+      "learning_rate": 2.46e-06,
+      "loss": 0.0577,
+      "step": 1918
+    },
+    {
+      "epoch": 47.98,
+      "learning_rate": 2.43e-06,
+      "loss": 0.0223,
+      "step": 1919
+    },
+    {
+      "epoch": 48.0,
+      "learning_rate": 2.4000000000000003e-06,
+      "loss": 0.0485,
+      "step": 1920
+    },
+    {
+      "epoch": 48.0,
+      "eval_loss": 0.15490178763866425,
+      "eval_mean_accuracy": 0.9361269118671846,
+      "eval_mean_iou": 0.5869765782436444,
+      "eval_overall_accuracy": 0.9440276683471981,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9504402159214842,
+        0.9218136078128849
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9313395559398764,
+        0.8295901787910566
+      ],
+      "eval_runtime": 48.2784,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 1920
+    },
+    {
+      "epoch": 48.02,
+      "learning_rate": 2.37e-06,
+      "loss": 0.0743,
+      "step": 1921
+    },
+    {
+      "epoch": 48.05,
+      "learning_rate": 2.34e-06,
+      "loss": 0.0524,
+      "step": 1922
+    },
+    {
+      "epoch": 48.08,
+      "learning_rate": 2.31e-06,
+      "loss": 0.1514,
+      "step": 1923
+    },
+    {
+      "epoch": 48.1,
+      "learning_rate": 2.28e-06,
+      "loss": 0.138,
+      "step": 1924
+    },
+    {
+      "epoch": 48.12,
+      "learning_rate": 2.25e-06,
+      "loss": 0.0687,
+      "step": 1925
+    },
+    {
+      "epoch": 48.15,
+      "learning_rate": 2.22e-06,
+      "loss": 0.1171,
+      "step": 1926
+    },
+    {
+      "epoch": 48.17,
+      "learning_rate": 2.1899999999999998e-06,
+      "loss": 0.0276,
+      "step": 1927
+    },
+    {
+      "epoch": 48.2,
+      "learning_rate": 2.16e-06,
+      "loss": 0.1263,
+      "step": 1928
+    },
+    {
+      "epoch": 48.23,
+      "learning_rate": 2.13e-06,
+      "loss": 0.1219,
+      "step": 1929
+    },
+    {
+      "epoch": 48.25,
+      "learning_rate": 2.1000000000000002e-06,
+      "loss": 0.0811,
+      "step": 1930
+    },
+    {
+      "epoch": 48.27,
+      "learning_rate": 2.07e-06,
+      "loss": 0.0449,
+      "step": 1931
+    },
+    {
+      "epoch": 48.3,
+      "learning_rate": 2.0400000000000004e-06,
+      "loss": 0.053,
+      "step": 1932
+    },
+    {
+      "epoch": 48.33,
+      "learning_rate": 2.0100000000000002e-06,
+      "loss": 0.0786,
+      "step": 1933
+    },
+    {
+      "epoch": 48.35,
+      "learning_rate": 1.98e-06,
+      "loss": 0.1659,
+      "step": 1934
+    },
+    {
+      "epoch": 48.38,
+      "learning_rate": 1.95e-06,
+      "loss": 0.1207,
+      "step": 1935
+    },
+    {
+      "epoch": 48.4,
+      "learning_rate": 1.9200000000000003e-06,
+      "loss": 0.0484,
+      "step": 1936
+    },
+    {
+      "epoch": 48.42,
+      "learning_rate": 1.8900000000000001e-06,
+      "loss": 0.0794,
+      "step": 1937
+    },
+    {
+      "epoch": 48.45,
+      "learning_rate": 1.86e-06,
+      "loss": 0.0955,
+      "step": 1938
+    },
+    {
+      "epoch": 48.48,
+      "learning_rate": 1.83e-06,
+      "loss": 0.1916,
+      "step": 1939
+    },
+    {
+      "epoch": 48.5,
+      "learning_rate": 1.8e-06,
+      "loss": 0.05,
+      "step": 1940
+    },
+    {
+      "epoch": 48.5,
+      "eval_loss": 0.1405608355998993,
+      "eval_mean_accuracy": 0.9307088391843652,
+      "eval_mean_iou": 0.5846660909659979,
+      "eval_overall_accuracy": 0.9344089218013106,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9374120463590632,
+        0.9240056320096672
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9201697393019961,
+        0.8338285335959977
+      ],
+      "eval_runtime": 48.0586,
+      "eval_samples_per_second": 0.416,
+      "eval_steps_per_second": 0.208,
+      "step": 1940
+    },
+    {
+      "epoch": 48.52,
+      "learning_rate": 1.77e-06,
+      "loss": 0.0521,
+      "step": 1941
+    },
+    {
+      "epoch": 48.55,
+      "learning_rate": 1.74e-06,
+      "loss": 0.0349,
+      "step": 1942
+    },
+    {
+      "epoch": 48.58,
+      "learning_rate": 1.7100000000000001e-06,
+      "loss": 0.0611,
+      "step": 1943
+    },
+    {
+      "epoch": 48.6,
+      "learning_rate": 1.68e-06,
+      "loss": 0.0601,
+      "step": 1944
+    },
+    {
+      "epoch": 48.62,
+      "learning_rate": 1.65e-06,
+      "loss": 0.0518,
+      "step": 1945
+    },
+    {
+      "epoch": 48.65,
+      "learning_rate": 1.62e-06,
+      "loss": 0.0306,
+      "step": 1946
+    },
+    {
+      "epoch": 48.67,
+      "learning_rate": 1.59e-06,
+      "loss": 0.0344,
+      "step": 1947
+    },
+    {
+      "epoch": 48.7,
+      "learning_rate": 1.5599999999999999e-06,
+      "loss": 0.0618,
+      "step": 1948
+    },
+    {
+      "epoch": 48.73,
+      "learning_rate": 1.53e-06,
+      "loss": 0.2589,
+      "step": 1949
+    },
+    {
+      "epoch": 48.75,
+      "learning_rate": 1.5e-06,
+      "loss": 0.0743,
+      "step": 1950
+    },
+    {
+      "epoch": 48.77,
+      "learning_rate": 1.4700000000000001e-06,
+      "loss": 0.0519,
+      "step": 1951
+    },
+    {
+      "epoch": 48.8,
+      "learning_rate": 1.44e-06,
+      "loss": 0.0538,
+      "step": 1952
+    },
+    {
+      "epoch": 48.83,
+      "learning_rate": 1.41e-06,
+      "loss": 0.1266,
+      "step": 1953
+    },
+    {
+      "epoch": 48.85,
+      "learning_rate": 1.38e-06,
+      "loss": 0.1463,
+      "step": 1954
+    },
+    {
+      "epoch": 48.88,
+      "learning_rate": 1.35e-06,
+      "loss": 0.0337,
+      "step": 1955
+    },
+    {
+      "epoch": 48.9,
+      "learning_rate": 1.3199999999999999e-06,
+      "loss": 0.0746,
+      "step": 1956
+    },
+    {
+      "epoch": 48.92,
+      "learning_rate": 1.29e-06,
+      "loss": 0.0386,
+      "step": 1957
+    },
+    {
+      "epoch": 48.95,
+      "learning_rate": 1.26e-06,
+      "loss": 0.1289,
+      "step": 1958
+    },
+    {
+      "epoch": 48.98,
+      "learning_rate": 1.23e-06,
+      "loss": 0.0454,
+      "step": 1959
+    },
+    {
+      "epoch": 49.0,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 0.3912,
+      "step": 1960
+    },
+    {
+      "epoch": 49.0,
+      "eval_loss": 0.15050022304058075,
+      "eval_mean_accuracy": 0.9407804151601167,
+      "eval_mean_iou": 0.5886050276123915,
+      "eval_overall_accuracy": 0.9436872922497369,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9460466217236864,
+        0.9355142085965471
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.9302946310985881,
+        0.8355204517385865
+      ],
+      "eval_runtime": 48.2548,
+      "eval_samples_per_second": 0.414,
+      "eval_steps_per_second": 0.207,
+      "step": 1960
+    },
+    {
+      "epoch": 49.02,
+      "learning_rate": 1.17e-06,
+      "loss": 0.1476,
+      "step": 1961
+    },
+    {
+      "epoch": 49.05,
+      "learning_rate": 1.14e-06,
+      "loss": 0.1131,
+      "step": 1962
+    },
+    {
+      "epoch": 49.08,
+      "learning_rate": 1.11e-06,
+      "loss": 0.2595,
+      "step": 1963
+    },
+    {
+      "epoch": 49.1,
+      "learning_rate": 1.08e-06,
+      "loss": 0.0544,
+      "step": 1964
+    },
+    {
+      "epoch": 49.12,
+      "learning_rate": 1.0500000000000001e-06,
+      "loss": 0.0743,
+      "step": 1965
+    },
+    {
+      "epoch": 49.15,
+      "learning_rate": 1.0200000000000002e-06,
+      "loss": 0.099,
+      "step": 1966
+    },
+    {
+      "epoch": 49.17,
+      "learning_rate": 9.9e-07,
+      "loss": 0.0957,
+      "step": 1967
+    },
+    {
+      "epoch": 49.2,
+      "learning_rate": 9.600000000000001e-07,
+      "loss": 0.048,
+      "step": 1968
+    },
+    {
+      "epoch": 49.23,
+      "learning_rate": 9.3e-07,
+      "loss": 0.0635,
+      "step": 1969
+    },
+    {
+      "epoch": 49.25,
+      "learning_rate": 9e-07,
+      "loss": 0.0263,
+      "step": 1970
+    },
+    {
+      "epoch": 49.27,
+      "learning_rate": 8.7e-07,
+      "loss": 0.0271,
+      "step": 1971
+    },
+    {
+      "epoch": 49.3,
+      "learning_rate": 8.4e-07,
+      "loss": 0.1153,
+      "step": 1972
+    },
+    {
+      "epoch": 49.33,
+      "learning_rate": 8.1e-07,
+      "loss": 0.0444,
+      "step": 1973
+    },
+    {
+      "epoch": 49.35,
+      "learning_rate": 7.799999999999999e-07,
+      "loss": 0.0717,
+      "step": 1974
+    },
+    {
+      "epoch": 49.38,
+      "learning_rate": 7.5e-07,
+      "loss": 0.0278,
+      "step": 1975
+    },
+    {
+      "epoch": 49.4,
+      "learning_rate": 7.2e-07,
+      "loss": 0.1145,
+      "step": 1976
+    },
+    {
+      "epoch": 49.42,
+      "learning_rate": 6.9e-07,
+      "loss": 0.0551,
+      "step": 1977
+    },
+    {
+      "epoch": 49.45,
+      "learning_rate": 6.599999999999999e-07,
+      "loss": 0.0458,
+      "step": 1978
+    },
+    {
+      "epoch": 49.48,
+      "learning_rate": 6.3e-07,
+      "loss": 0.0622,
+      "step": 1979
+    },
+    {
+      "epoch": 49.5,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 0.0401,
+      "step": 1980
+    },
+    {
+      "epoch": 49.5,
+      "eval_loss": 0.14851275086402893,
+      "eval_mean_accuracy": 0.9400991166570261,
+      "eval_mean_iou": 0.586200467391329,
+      "eval_overall_accuracy": 0.9410110096562081,
+      "eval_per_category_accuracy": [
+        NaN,
+        0.9417511359029422,
+        0.9384470974111098
+      ],
+      "eval_per_category_iou": [
+        0.0,
+        0.926960691893792,
+        0.8316407102801952
+      ],
+      "eval_runtime": 48.9351,
+      "eval_samples_per_second": 0.409,
+      "eval_steps_per_second": 0.204,
+      "step": 1980
+    }
+  ],
+  "max_steps": 2000,
+  "num_train_epochs": 50,
+  "total_flos": 6.94155060707328e+16,
+  "trial_name": null,
+  "trial_params": null
+}