diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5034 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 25.64,
+  "eval_steps": 500,
+  "global_step": 5000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.0512,
+      "grad_norm": 0.86328125,
+      "learning_rate": 3.076923076923077e-05,
+      "loss": 12.4822,
+      "router_z_loss": 0.0,
+      "step": 10
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.1024,
+      "grad_norm": 0.90625,
+      "learning_rate": 6.153846153846154e-05,
+      "loss": 12.2583,
+      "router_z_loss": 0.0,
+      "step": 20
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.1536,
+      "grad_norm": 0.84765625,
+      "learning_rate": 9.230769230769232e-05,
+      "loss": 11.7958,
+      "router_z_loss": 0.0,
+      "step": 30
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.2048,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.00012307692307692307,
+      "loss": 11.1369,
+      "router_z_loss": 0.0,
+      "step": 40
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.256,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.00015384615384615385,
+      "loss": 10.2261,
+      "router_z_loss": 0.0,
+      "step": 50
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.3072,
+      "grad_norm": 0.9140625,
+      "learning_rate": 0.00018461538461538463,
+      "loss": 9.3587,
+      "router_z_loss": 0.0,
+      "step": 60
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.3584,
+      "grad_norm": 0.5546875,
+      "learning_rate": 0.0002153846153846154,
+      "loss": 8.7066,
+      "router_z_loss": 0.0,
+      "step": 70
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.4096,
+      "grad_norm": 4.03125,
+      "learning_rate": 0.00024615384615384614,
+      "loss": 8.2878,
+      "router_z_loss": 0.0,
+      "step": 80
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.4608,
+      "grad_norm": 2.375,
+      "learning_rate": 0.00027692307692307695,
+      "loss": 7.9893,
+      "router_z_loss": 0.0,
+      "step": 90
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.512,
+      "grad_norm": 2.109375,
+      "learning_rate": 0.0003076923076923077,
+      "loss": 7.6368,
+      "router_z_loss": 0.0,
+      "step": 100
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.5632,
+      "grad_norm": 0.5078125,
+      "learning_rate": 0.00033846153846153846,
+      "loss": 7.1774,
+      "router_z_loss": 0.0,
+      "step": 110
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.6144,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.00036923076923076927,
+      "loss": 6.6833,
+      "router_z_loss": 0.0,
+      "step": 120
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.6656,
+      "grad_norm": 1.609375,
+      "learning_rate": 0.0004,
+      "loss": 6.2922,
+      "router_z_loss": 0.0,
+      "step": 130
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.7168,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.0004307692307692308,
+      "loss": 6.0386,
+      "router_z_loss": 0.0,
+      "step": 140
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.768,
+      "grad_norm": 0.71484375,
+      "learning_rate": 0.0004615384615384616,
+      "loss": 5.8256,
+      "router_z_loss": 0.0,
+      "step": 150
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.8192,
+      "grad_norm": 1.1875,
+      "learning_rate": 0.0004923076923076923,
+      "loss": 5.6412,
+      "router_z_loss": 0.0,
+      "step": 160
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.8704,
+      "grad_norm": 0.91015625,
+      "learning_rate": 0.0005230769230769231,
+      "loss": 5.4734,
+      "router_z_loss": 0.0,
+      "step": 170
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.9216,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.0005538461538461539,
+      "loss": 5.3141,
+      "router_z_loss": 0.0,
+      "step": 180
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 0.9728,
+      "grad_norm": 0.9375,
+      "learning_rate": 0.0005846153846153846,
+      "loss": 5.1415,
+      "router_z_loss": 0.0,
+      "step": 190
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.0256,
+      "grad_norm": 0.81640625,
+      "learning_rate": 0.0006153846153846154,
+      "loss": 5.5058,
+      "router_z_loss": 0.0,
+      "step": 200
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.0768,
+      "grad_norm": 0.5546875,
+      "learning_rate": 0.0006461538461538462,
+      "loss": 4.8577,
+      "router_z_loss": 0.0,
+      "step": 210
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.1280000000000001,
+      "grad_norm": 0.859375,
+      "learning_rate": 0.0006769230769230769,
+      "loss": 4.7405,
+      "router_z_loss": 0.0,
+      "step": 220
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.1792,
+      "grad_norm": 0.75,
+      "learning_rate": 0.0007076923076923077,
+      "loss": 4.6311,
+      "router_z_loss": 0.0,
+      "step": 230
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.2304,
+      "grad_norm": 1.5625,
+      "learning_rate": 0.0007384615384615385,
+      "loss": 4.4939,
+      "router_z_loss": 0.0,
+      "step": 240
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.2816,
+      "grad_norm": 1.8046875,
+      "learning_rate": 0.0007692307692307692,
+      "loss": 4.4278,
+      "router_z_loss": 0.0,
+      "step": 250
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.3328,
+      "grad_norm": 1.4453125,
+      "learning_rate": 0.0008,
+      "loss": 4.3309,
+      "router_z_loss": 0.0,
+      "step": 260
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.384,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.0008307692307692308,
+      "loss": 4.2308,
+      "router_z_loss": 0.0,
+      "step": 270
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.4352,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.0008615384615384615,
+      "loss": 4.1648,
+      "router_z_loss": 0.0,
+      "step": 280
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.4864,
+      "grad_norm": 1.0859375,
+      "learning_rate": 0.0008923076923076924,
+      "loss": 4.0882,
+      "router_z_loss": 0.0,
+      "step": 290
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.5375999999999999,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.0009230769230769232,
+      "loss": 4.0039,
+      "router_z_loss": 0.0,
+      "step": 300
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.5888,
+      "grad_norm": 0.68359375,
+      "learning_rate": 0.0009538461538461538,
+      "loss": 3.9642,
+      "router_z_loss": 0.0,
+      "step": 310
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.6400000000000001,
+      "grad_norm": 0.77734375,
+      "learning_rate": 0.0009846153846153846,
+      "loss": 3.8901,
+      "router_z_loss": 0.0,
+      "step": 320
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.6912,
+      "grad_norm": 0.9453125,
+      "learning_rate": 0.0010153846153846155,
+      "loss": 3.8254,
+      "router_z_loss": 0.0,
+      "step": 330
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.7424,
+      "grad_norm": 0.71484375,
+      "learning_rate": 0.0010461538461538462,
+      "loss": 3.7841,
+      "router_z_loss": 0.0,
+      "step": 340
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.7936,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.0010769230769230769,
+      "loss": 3.751,
+      "router_z_loss": 0.0,
+      "step": 350
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.8448,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.0011076923076923078,
+      "loss": 3.6586,
+      "router_z_loss": 0.0,
+      "step": 360
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.896,
+      "grad_norm": 1.4140625,
+      "learning_rate": 0.0011384615384615385,
+      "loss": 3.6627,
+      "router_z_loss": 0.0,
+      "step": 370
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.9472,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.0011692307692307692,
+      "loss": 3.5893,
+      "router_z_loss": 0.0,
+      "step": 380
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 1.9984,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.0012000000000000001,
+      "loss": 3.5727,
+      "router_z_loss": 0.0,
+      "step": 390
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.0512,
+      "grad_norm": 0.400390625,
+      "learning_rate": 0.0012307692307692308,
+      "loss": 3.8897,
+      "router_z_loss": 0.0,
+      "step": 400
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.1024,
+      "grad_norm": 0.486328125,
+      "learning_rate": 0.0012615384615384615,
+      "loss": 3.501,
+      "router_z_loss": 0.0,
+      "step": 410
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.1536,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.0012923076923076924,
+      "loss": 3.4462,
+      "router_z_loss": 0.0,
+      "step": 420
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.2048,
+      "grad_norm": 0.5078125,
+      "learning_rate": 0.0013230769230769231,
+      "loss": 3.4335,
+      "router_z_loss": 0.0,
+      "step": 430
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.2560000000000002,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.0013538461538461538,
+      "loss": 3.3947,
+      "router_z_loss": 0.0,
+      "step": 440
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.3072,
+      "grad_norm": 0.66796875,
+      "learning_rate": 0.0013846153846153847,
+      "loss": 3.37,
+      "router_z_loss": 0.0,
+      "step": 450
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.3584,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.0014153846153846154,
+      "loss": 3.345,
+      "router_z_loss": 0.0,
+      "step": 460
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.4096,
+      "grad_norm": 0.447265625,
+      "learning_rate": 0.0014461538461538461,
+      "loss": 3.355,
+      "router_z_loss": 0.0,
+      "step": 470
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.4608,
+      "grad_norm": 0.8671875,
+      "learning_rate": 0.001476923076923077,
+      "loss": 3.3138,
+      "router_z_loss": 0.0,
+      "step": 480
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.512,
+      "grad_norm": 0.6171875,
+      "learning_rate": 0.0015076923076923078,
+      "loss": 3.2843,
+      "router_z_loss": 0.0,
+      "step": 490
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.5632,
+      "grad_norm": 0.60546875,
+      "learning_rate": 0.0015384615384615385,
+      "loss": 3.2832,
+      "router_z_loss": 0.0,
+      "step": 500
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.6144,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.0015692307692307694,
+      "loss": 3.2416,
+      "router_z_loss": 0.0,
+      "step": 510
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.6656,
+      "grad_norm": 0.421875,
+      "learning_rate": 0.0016,
+      "loss": 3.2343,
+      "router_z_loss": 0.0,
+      "step": 520
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.7168,
+      "grad_norm": 0.404296875,
+      "learning_rate": 0.0016307692307692308,
+      "loss": 3.2167,
+      "router_z_loss": 0.0,
+      "step": 530
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.768,
+      "grad_norm": 0.4765625,
+      "learning_rate": 0.0016615384615384617,
+      "loss": 3.1902,
+      "router_z_loss": 0.0,
+      "step": 540
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.8192,
+      "grad_norm": 0.4921875,
+      "learning_rate": 0.0016923076923076924,
+      "loss": 3.2147,
+      "router_z_loss": 0.0,
+      "step": 550
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.8704,
+      "grad_norm": 0.640625,
+      "learning_rate": 0.001723076923076923,
+      "loss": 3.1914,
+      "router_z_loss": 0.0,
+      "step": 560
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.9215999999999998,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.001753846153846154,
+      "loss": 3.1648,
+      "router_z_loss": 0.0,
+      "step": 570
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 2.9728,
+      "grad_norm": 0.56640625,
+      "learning_rate": 0.0017846153846153847,
+      "loss": 3.1473,
+      "router_z_loss": 0.0,
+      "step": 580
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.0256,
+      "grad_norm": 1.984375,
+      "learning_rate": 0.0018153846153846154,
+      "loss": 3.555,
+      "router_z_loss": 0.0,
+      "step": 590
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.0768,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.0018461538461538463,
+      "loss": 3.2716,
+      "router_z_loss": 0.0,
+      "step": 600
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.128,
+      "grad_norm": 0.3515625,
+      "learning_rate": 0.001876923076923077,
+      "loss": 3.1852,
+      "router_z_loss": 0.0,
+      "step": 610
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.1792,
+      "grad_norm": 0.345703125,
+      "learning_rate": 0.0019076923076923075,
+      "loss": 3.1403,
+      "router_z_loss": 0.0,
+      "step": 620
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.2304,
+      "grad_norm": 0.4296875,
+      "learning_rate": 0.0019384615384615386,
+      "loss": 3.1375,
+      "router_z_loss": 0.0,
+      "step": 630
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.2816,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.001969230769230769,
+      "loss": 3.1062,
+      "router_z_loss": 0.0,
+      "step": 640
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.3327999999999998,
+      "grad_norm": 0.486328125,
+      "learning_rate": 0.002,
+      "loss": 3.0988,
+      "router_z_loss": 0.0,
+      "step": 650
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.384,
+      "grad_norm": 0.458984375,
+      "learning_rate": 0.002030769230769231,
+      "loss": 3.0966,
+      "router_z_loss": 0.0,
+      "step": 660
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.4352,
+      "grad_norm": 0.33203125,
+      "learning_rate": 0.0020615384615384614,
+      "loss": 3.0842,
+      "router_z_loss": 0.0,
+      "step": 670
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.4864,
+      "grad_norm": 0.306640625,
+      "learning_rate": 0.0020923076923076924,
+      "loss": 3.0998,
+      "router_z_loss": 0.0,
+      "step": 680
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.5376,
+      "grad_norm": 0.4140625,
+      "learning_rate": 0.0021230769230769233,
+      "loss": 3.074,
+      "router_z_loss": 0.0,
+      "step": 690
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.5888,
+      "grad_norm": 0.408203125,
+      "learning_rate": 0.0021538461538461538,
+      "loss": 3.0512,
+      "router_z_loss": 0.0,
+      "step": 700
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.64,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.0021846153846153847,
+      "loss": 3.0719,
+      "router_z_loss": 0.0,
+      "step": 710
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.6912000000000003,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.0022153846153846156,
+      "loss": 3.0421,
+      "router_z_loss": 0.0,
+      "step": 720
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.7424,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.002246153846153846,
+      "loss": 3.061,
+      "router_z_loss": 0.0,
+      "step": 730
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.7936,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.002276923076923077,
+      "loss": 3.0319,
+      "router_z_loss": 0.0,
+      "step": 740
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.8448,
+      "grad_norm": 1.453125,
+      "learning_rate": 0.002307692307692308,
+      "loss": 3.0467,
+      "router_z_loss": 0.0,
+      "step": 750
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.896,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0023384615384615384,
+      "loss": 3.0424,
+      "router_z_loss": 0.0,
+      "step": 760
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.9472,
+      "grad_norm": 0.333984375,
+      "learning_rate": 0.0023692307692307693,
+      "loss": 3.0409,
+      "router_z_loss": 0.0,
+      "step": 770
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 3.9984,
+      "grad_norm": 0.41015625,
+      "learning_rate": 0.0024000000000000002,
+      "loss": 3.0104,
+      "router_z_loss": 0.0,
+      "step": 780
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.0512,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0024307692307692307,
+      "loss": 3.3914,
+      "router_z_loss": 0.0,
+      "step": 790
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.1024,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0024615384615384616,
+      "loss": 2.9982,
+      "router_z_loss": 0.0,
+      "step": 800
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.1536,
+      "grad_norm": 0.2890625,
+      "learning_rate": 0.0024923076923076925,
+      "loss": 3.0196,
+      "router_z_loss": 0.0,
+      "step": 810
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.2048,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.002523076923076923,
+      "loss": 2.9782,
+      "router_z_loss": 0.0,
+      "step": 820
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.256,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.002553846153846154,
+      "loss": 2.9809,
+      "router_z_loss": 0.0,
+      "step": 830
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.3072,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.002584615384615385,
+      "loss": 2.9719,
+      "router_z_loss": 0.0,
+      "step": 840
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.3584,
+      "grad_norm": 0.275390625,
+      "learning_rate": 0.0026153846153846153,
+      "loss": 2.967,
+      "router_z_loss": 0.0,
+      "step": 850
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.4096,
+      "grad_norm": 0.2490234375,
+      "learning_rate": 0.0026461538461538463,
+      "loss": 2.951,
+      "router_z_loss": 0.0,
+      "step": 860
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.4608,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.002676923076923077,
+      "loss": 2.9363,
+      "router_z_loss": 0.0,
+      "step": 870
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.5120000000000005,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0027076923076923077,
+      "loss": 2.9139,
+      "router_z_loss": 0.0,
+      "step": 880
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.5632,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0027384615384615386,
+      "loss": 2.9451,
+      "router_z_loss": 0.0,
+      "step": 890
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.6144,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0027692307692307695,
+      "loss": 2.9361,
+      "router_z_loss": 0.0,
+      "step": 900
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.6655999999999995,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0028,
+      "loss": 2.9243,
+      "router_z_loss": 0.0,
+      "step": 910
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.7168,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.002830769230769231,
+      "loss": 2.9238,
+      "router_z_loss": 0.0,
+      "step": 920
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.768,
+      "grad_norm": 0.2177734375,
+      "learning_rate": 0.002861538461538462,
+      "loss": 2.9097,
+      "router_z_loss": 0.0,
+      "step": 930
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.8192,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.0028923076923076923,
+      "loss": 2.9009,
+      "router_z_loss": 0.0,
+      "step": 940
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.8704,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.002923076923076923,
+      "loss": 2.9211,
+      "router_z_loss": 0.0,
+      "step": 950
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.9216,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.002953846153846154,
+      "loss": 2.8885,
+      "router_z_loss": 0.0,
+      "step": 960
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 4.9728,
+      "grad_norm": 0.34765625,
+      "learning_rate": 0.0029846153846153846,
+      "loss": 2.8904,
+      "router_z_loss": 0.0,
+      "step": 970
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.0256,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.003,
+      "loss": 3.1845,
+      "router_z_loss": 0.0,
+      "step": 980
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.0768,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.003,
+      "loss": 2.8775,
+      "router_z_loss": 0.0,
+      "step": 990
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.128,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.003,
+      "loss": 2.8545,
+      "router_z_loss": 0.0,
+      "step": 1000
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.1792,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.003,
+      "loss": 2.8552,
+      "router_z_loss": 0.0,
+      "step": 1010
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.2304,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.003,
+      "loss": 2.8244,
+      "router_z_loss": 0.0,
+      "step": 1020
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.2816,
+      "grad_norm": 0.30078125,
+      "learning_rate": 0.003,
+      "loss": 2.836,
+      "router_z_loss": 0.0,
+      "step": 1030
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.3328,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.003,
+      "loss": 2.8203,
+      "router_z_loss": 0.0,
+      "step": 1040
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.384,
+      "grad_norm": 0.330078125,
+      "learning_rate": 0.003,
+      "loss": 2.8179,
+      "router_z_loss": 0.0,
+      "step": 1050
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.4352,
+      "grad_norm": 0.375,
+      "learning_rate": 0.003,
+      "loss": 2.8163,
+      "router_z_loss": 0.0,
+      "step": 1060
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.4864,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.003,
+      "loss": 2.8234,
+      "router_z_loss": 0.0,
+      "step": 1070
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.5376,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.003,
+      "loss": 2.7997,
+      "router_z_loss": 0.0,
+      "step": 1080
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.5888,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.003,
+      "loss": 2.7923,
+      "router_z_loss": 0.0,
+      "step": 1090
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.64,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.003,
+      "loss": 2.7918,
+      "router_z_loss": 0.0,
+      "step": 1100
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.6912,
+      "grad_norm": 0.2353515625,
+      "learning_rate": 0.003,
+      "loss": 2.7794,
+      "router_z_loss": 0.0,
+      "step": 1110
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.7424,
+      "grad_norm": 0.361328125,
+      "learning_rate": 0.003,
+      "loss": 2.7793,
+      "router_z_loss": 0.0,
+      "step": 1120
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.7936,
+      "grad_norm": 0.20703125,
+      "learning_rate": 0.003,
+      "loss": 2.793,
+      "router_z_loss": 0.0,
+      "step": 1130
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.8448,
+      "grad_norm": 0.2060546875,
+      "learning_rate": 0.003,
+      "loss": 2.7649,
+      "router_z_loss": 0.0,
+      "step": 1140
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.896,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.003,
+      "loss": 2.7741,
+      "router_z_loss": 0.0,
+      "step": 1150
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.9472000000000005,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.003,
+      "loss": 2.7515,
+      "router_z_loss": 0.0,
+      "step": 1160
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 5.9984,
+      "grad_norm": 0.19921875,
+      "learning_rate": 0.003,
+      "loss": 2.7619,
+      "router_z_loss": 0.0,
+      "step": 1170
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.0512,
+      "grad_norm": 0.2421875,
+      "learning_rate": 0.003,
+      "loss": 3.0268,
+      "router_z_loss": 0.0,
+      "step": 1180
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.1024,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.003,
+      "loss": 2.7404,
+      "router_z_loss": 0.0,
+      "step": 1190
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.1536,
+      "grad_norm": 0.2333984375,
+      "learning_rate": 0.003,
+      "loss": 2.7276,
+      "router_z_loss": 0.0,
+      "step": 1200
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.2048,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.003,
+      "loss": 2.6999,
+      "router_z_loss": 0.0,
+      "step": 1210
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.256,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.003,
+      "loss": 2.7228,
+      "router_z_loss": 0.0,
+      "step": 1220
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.3072,
+      "grad_norm": 0.22265625,
+      "learning_rate": 0.003,
+      "loss": 2.7162,
+      "router_z_loss": 0.0,
+      "step": 1230
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.3584,
+      "grad_norm": 0.419921875,
+      "learning_rate": 0.003,
+      "loss": 2.7036,
+      "router_z_loss": 0.0,
+      "step": 1240
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.4096,
+      "grad_norm": 0.201171875,
+      "learning_rate": 0.003,
+      "loss": 2.7148,
+      "router_z_loss": 0.0,
+      "step": 1250
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.4608,
+      "grad_norm": 0.21875,
+      "learning_rate": 0.003,
+      "loss": 2.6904,
+      "router_z_loss": 0.0,
+      "step": 1260
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.5120000000000005,
+      "grad_norm": 0.298828125,
+      "learning_rate": 0.003,
+      "loss": 2.6952,
+      "router_z_loss": 0.0,
+      "step": 1270
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.5632,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.003,
+      "loss": 2.6898,
+      "router_z_loss": 0.0,
+      "step": 1280
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.6144,
+      "grad_norm": 0.2041015625,
+      "learning_rate": 0.003,
+      "loss": 2.6835,
+      "router_z_loss": 0.0,
+      "step": 1290
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.6655999999999995,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.003,
+      "loss": 2.6932,
+      "router_z_loss": 0.0,
+      "step": 1300
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.7168,
+      "grad_norm": 0.2119140625,
+      "learning_rate": 0.003,
+      "loss": 2.6754,
+      "router_z_loss": 0.0,
+      "step": 1310
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.768,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.003,
+      "loss": 2.6842,
+      "router_z_loss": 0.0,
+      "step": 1320
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.8192,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.003,
+      "loss": 2.6741,
+      "router_z_loss": 0.0,
+      "step": 1330
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.8704,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.003,
+      "loss": 2.6605,
+      "router_z_loss": 0.0,
+      "step": 1340
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.9216,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.003,
+      "loss": 2.6674,
+      "router_z_loss": 0.0,
+      "step": 1350
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 6.9728,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.003,
+      "loss": 2.6727,
+      "router_z_loss": 0.0,
+      "step": 1360
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.0256,
+      "grad_norm": 0.275390625,
+      "learning_rate": 0.003,
+      "loss": 2.9389,
+      "router_z_loss": 0.0,
+      "step": 1370
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.0768,
+      "grad_norm": 0.1962890625,
+      "learning_rate": 0.003,
+      "loss": 2.6637,
+      "router_z_loss": 0.0,
+      "step": 1380
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.128,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.003,
+      "loss": 2.6444,
+      "router_z_loss": 0.0,
+      "step": 1390
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.1792,
+      "grad_norm": 0.2314453125,
+      "learning_rate": 0.003,
+      "loss": 2.6353,
+      "router_z_loss": 0.0,
+      "step": 1400
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.2304,
+      "grad_norm": 0.2109375,
+      "learning_rate": 0.003,
+      "loss": 2.631,
+      "router_z_loss": 0.0,
+      "step": 1410
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.2816,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.003,
+      "loss": 2.6191,
+      "router_z_loss": 0.0,
+      "step": 1420
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.3328,
+      "grad_norm": 0.2353515625,
+      "learning_rate": 0.003,
+      "loss": 2.627,
+      "router_z_loss": 0.0,
+      "step": 1430
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.384,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.003,
+      "loss": 2.6287,
+      "router_z_loss": 0.0,
+      "step": 1440
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.4352,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.003,
+      "loss": 2.6258,
+      "router_z_loss": 0.0,
+      "step": 1450
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.4864,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.003,
+      "loss": 2.6219,
+      "router_z_loss": 0.0,
+      "step": 1460
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.5376,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.003,
+      "loss": 2.6155,
+      "router_z_loss": 0.0,
+      "step": 1470
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.5888,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.003,
+      "loss": 2.6125,
+      "router_z_loss": 0.0,
+      "step": 1480
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.64,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.003,
+      "loss": 2.616,
+      "router_z_loss": 0.0,
+      "step": 1490
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.6912,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.003,
+      "loss": 2.6035,
+      "router_z_loss": 0.0,
+      "step": 1500
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.7424,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.003,
+      "loss": 2.6112,
+      "router_z_loss": 0.0,
+      "step": 1510
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.7936,
+      "grad_norm": 0.2431640625,
+      "learning_rate": 0.003,
+      "loss": 2.6033,
+      "router_z_loss": 0.0,
+      "step": 1520
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.8448,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.003,
+      "loss": 2.5985,
+      "router_z_loss": 0.0,
+      "step": 1530
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.896,
+      "grad_norm": 0.21875,
+      "learning_rate": 0.003,
+      "loss": 2.6088,
+      "router_z_loss": 0.0,
+      "step": 1540
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.9472000000000005,
+      "grad_norm": 0.2119140625,
+      "learning_rate": 0.003,
+      "loss": 2.5988,
+      "router_z_loss": 0.0,
+      "step": 1550
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 7.9984,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.003,
+      "loss": 2.5964,
+      "router_z_loss": 0.0,
+      "step": 1560
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.0512,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.003,
+      "loss": 2.8688,
+      "router_z_loss": 0.0,
+      "step": 1570
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.1024,
+      "grad_norm": 0.1923828125,
+      "learning_rate": 0.003,
+      "loss": 2.5877,
+      "router_z_loss": 0.0,
+      "step": 1580
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.1536,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.003,
+      "loss": 2.5778,
+      "router_z_loss": 0.0,
+      "step": 1590
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.2048,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.003,
+      "loss": 2.5813,
+      "router_z_loss": 0.0,
+      "step": 1600
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.256,
+      "grad_norm": 0.2216796875,
+      "learning_rate": 0.003,
+      "loss": 2.5646,
+      "router_z_loss": 0.0,
+      "step": 1610
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.3072,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.003,
+      "loss": 2.5666,
+      "router_z_loss": 0.0,
+      "step": 1620
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.3584,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.003,
+      "loss": 2.5681,
+      "router_z_loss": 0.0,
+      "step": 1630
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.4096,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.003,
+      "loss": 2.5768,
+      "router_z_loss": 0.0,
+      "step": 1640
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.4608,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.003,
+      "loss": 2.5612,
+      "router_z_loss": 0.0,
+      "step": 1650
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.512,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.003,
+      "loss": 2.562,
+      "router_z_loss": 0.0,
+      "step": 1660
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.5632,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.003,
+      "loss": 2.571,
+      "router_z_loss": 0.0,
+      "step": 1670
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.6144,
+      "grad_norm": 0.21875,
+      "learning_rate": 0.003,
+      "loss": 2.5554,
+      "router_z_loss": 0.0,
+      "step": 1680
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.6656,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.003,
+      "loss": 2.5633,
+      "router_z_loss": 0.0,
+      "step": 1690
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.7168,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.003,
+      "loss": 2.5639,
+      "router_z_loss": 0.0,
+      "step": 1700
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.768,
+      "grad_norm": 0.201171875,
+      "learning_rate": 0.003,
+      "loss": 2.546,
+      "router_z_loss": 0.0,
+      "step": 1710
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.8192,
+      "grad_norm": 0.2314453125,
+      "learning_rate": 0.003,
+      "loss": 2.5578,
+      "router_z_loss": 0.0,
+      "step": 1720
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.8704,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.003,
+      "loss": 2.5567,
+      "router_z_loss": 0.0,
+      "step": 1730
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.9216,
+      "grad_norm": 0.208984375,
+      "learning_rate": 0.003,
+      "loss": 2.5517,
+      "router_z_loss": 0.0,
+      "step": 1740
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 8.9728,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.003,
+      "loss": 2.5523,
+      "router_z_loss": 0.0,
+      "step": 1750
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.0256,
+      "grad_norm": 0.2890625,
+      "learning_rate": 0.003,
+      "loss": 2.816,
+      "router_z_loss": 0.0,
+      "step": 1760
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.0768,
+      "grad_norm": 0.1884765625,
+      "learning_rate": 0.003,
+      "loss": 2.5529,
+      "router_z_loss": 0.0,
+      "step": 1770
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.128,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.003,
+      "loss": 2.5302,
+      "router_z_loss": 0.0,
+      "step": 1780
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.1792,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.003,
+      "loss": 2.526,
+      "router_z_loss": 0.0,
+      "step": 1790
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.2304,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.003,
+      "loss": 2.5318,
+      "router_z_loss": 0.0,
+      "step": 1800
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.2816,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.003,
+      "loss": 2.5167,
+      "router_z_loss": 0.0,
+      "step": 1810
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.3328,
+      "grad_norm": 0.2373046875,
+      "learning_rate": 0.003,
+      "loss": 2.5291,
+      "router_z_loss": 0.0,
+      "step": 1820
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.384,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.003,
+      "loss": 2.5245,
+      "router_z_loss": 0.0,
+      "step": 1830
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.4352,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.003,
+      "loss": 2.5227,
+      "router_z_loss": 0.0,
+      "step": 1840
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.4864,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.003,
+      "loss": 2.5194,
+      "router_z_loss": 0.0,
+      "step": 1850
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.5376,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.003,
+      "loss": 2.5335,
+      "router_z_loss": 0.0,
+      "step": 1860
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.588799999999999,
+      "grad_norm": 0.2373046875,
+      "learning_rate": 0.003,
+      "loss": 2.5289,
+      "router_z_loss": 0.0,
+      "step": 1870
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.64,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.003,
+      "loss": 2.5138,
+      "router_z_loss": 0.0,
+      "step": 1880
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.6912,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.003,
+      "loss": 2.5216,
+      "router_z_loss": 0.0,
+      "step": 1890
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.7424,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.003,
+      "loss": 2.5238,
+      "router_z_loss": 0.0,
+      "step": 1900
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.7936,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.003,
+      "loss": 2.5161,
+      "router_z_loss": 0.0,
+      "step": 1910
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.8448,
+      "grad_norm": 0.1953125,
+      "learning_rate": 0.003,
+      "loss": 2.5058,
+      "router_z_loss": 0.0,
+      "step": 1920
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.896,
+      "grad_norm": 0.318359375,
+      "learning_rate": 0.003,
+      "loss": 2.508,
+      "router_z_loss": 0.0,
+      "step": 1930
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.9472,
+      "grad_norm": 0.228515625,
+      "learning_rate": 0.003,
+      "loss": 2.5147,
+      "router_z_loss": 0.0,
+      "step": 1940
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 9.9984,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.003,
+      "loss": 2.5169,
+      "router_z_loss": 0.0,
+      "step": 1950
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.0512,
+      "grad_norm": 0.228515625,
+      "learning_rate": 0.003,
+      "loss": 2.7845,
+      "router_z_loss": 0.0,
+      "step": 1960
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.1024,
+      "grad_norm": 0.1884765625,
+      "learning_rate": 0.003,
+      "loss": 2.5004,
+      "router_z_loss": 0.0,
+      "step": 1970
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.1536,
+      "grad_norm": 0.2001953125,
+      "learning_rate": 0.003,
+      "loss": 2.4885,
+      "router_z_loss": 0.0,
+      "step": 1980
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.2048,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.003,
+      "loss": 2.4883,
+      "router_z_loss": 0.0,
+      "step": 1990
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.256,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.003,
+      "loss": 2.4822,
+      "router_z_loss": 0.0,
+      "step": 2000
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.3072,
+      "grad_norm": 0.25,
+      "learning_rate": 0.003,
+      "loss": 2.4836,
+      "router_z_loss": 0.0,
+      "step": 2010
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.3584,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.003,
+      "loss": 2.4835,
+      "router_z_loss": 0.0,
+      "step": 2020
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.4096,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.003,
+      "loss": 2.488,
+      "router_z_loss": 0.0,
+      "step": 2030
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.4608,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.003,
+      "loss": 2.4858,
+      "router_z_loss": 0.0,
+      "step": 2040
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.512,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.003,
+      "loss": 2.4848,
+      "router_z_loss": 0.0,
+      "step": 2050
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.5632,
+      "grad_norm": 0.23828125,
+      "learning_rate": 0.003,
+      "loss": 2.4799,
+      "router_z_loss": 0.0,
+      "step": 2060
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.6144,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.003,
+      "loss": 2.4868,
+      "router_z_loss": 0.0,
+      "step": 2070
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.6656,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.003,
+      "loss": 2.4811,
+      "router_z_loss": 0.0,
+      "step": 2080
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.7168,
+      "grad_norm": 0.294921875,
+      "learning_rate": 0.003,
+      "loss": 2.4766,
+      "router_z_loss": 0.0,
+      "step": 2090
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.768,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.003,
+      "loss": 2.4846,
+      "router_z_loss": 0.0,
+      "step": 2100
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.8192,
+      "grad_norm": 0.212890625,
+      "learning_rate": 0.003,
+      "loss": 2.4844,
+      "router_z_loss": 0.0,
+      "step": 2110
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.8704,
+      "grad_norm": 0.330078125,
+      "learning_rate": 0.003,
+      "loss": 2.4765,
+      "router_z_loss": 0.0,
+      "step": 2120
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.9216,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.003,
+      "loss": 2.4752,
+      "router_z_loss": 0.0,
+      "step": 2130
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 10.9728,
+      "grad_norm": 0.2158203125,
+      "learning_rate": 0.003,
+      "loss": 2.4769,
+      "router_z_loss": 0.0,
+      "step": 2140
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.0256,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.003,
+      "loss": 2.733,
+      "router_z_loss": 0.0,
+      "step": 2150
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.0768,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.003,
+      "loss": 2.4789,
+      "router_z_loss": 0.0,
+      "step": 2160
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.128,
+      "grad_norm": 0.1982421875,
+      "learning_rate": 0.003,
+      "loss": 2.4601,
+      "router_z_loss": 0.0,
+      "step": 2170
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.1792,
+      "grad_norm": 0.2119140625,
+      "learning_rate": 0.003,
+      "loss": 2.4514,
+      "router_z_loss": 0.0,
+      "step": 2180
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.2304,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.003,
+      "loss": 2.4503,
+      "router_z_loss": 0.0,
+      "step": 2190
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.2816,
+      "grad_norm": 0.27734375,
+      "learning_rate": 0.003,
+      "loss": 2.4582,
+      "router_z_loss": 0.0,
+      "step": 2200
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.3328,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.003,
+      "loss": 2.4509,
+      "router_z_loss": 0.0,
+      "step": 2210
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.384,
+      "grad_norm": 0.2060546875,
+      "learning_rate": 0.003,
+      "loss": 2.4592,
+      "router_z_loss": 0.0,
+      "step": 2220
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.4352,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.003,
+      "loss": 2.453,
+      "router_z_loss": 0.0,
+      "step": 2230
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.4864,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.003,
+      "loss": 2.4599,
+      "router_z_loss": 0.0,
+      "step": 2240
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.5376,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.003,
+      "loss": 2.4608,
+      "router_z_loss": 0.0,
+      "step": 2250
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.588799999999999,
+      "grad_norm": 0.2021484375,
+      "learning_rate": 0.003,
+      "loss": 2.4552,
+      "router_z_loss": 0.0,
+      "step": 2260
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.64,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.003,
+      "loss": 2.4493,
+      "router_z_loss": 0.0,
+      "step": 2270
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.6912,
+      "grad_norm": 0.2421875,
+      "learning_rate": 0.003,
+      "loss": 2.4593,
+      "router_z_loss": 0.0,
+      "step": 2280
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.7424,
+      "grad_norm": 0.2080078125,
+      "learning_rate": 0.003,
+      "loss": 2.4547,
+      "router_z_loss": 0.0,
+      "step": 2290
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.7936,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.003,
+      "loss": 2.4519,
+      "router_z_loss": 0.0,
+      "step": 2300
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.8448,
+      "grad_norm": 0.20703125,
+      "learning_rate": 0.003,
+      "loss": 2.4577,
+      "router_z_loss": 0.0,
+      "step": 2310
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.896,
+      "grad_norm": 0.18359375,
+      "learning_rate": 0.003,
+      "loss": 2.4436,
+      "router_z_loss": 0.0,
+      "step": 2320
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.9472,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.003,
+      "loss": 2.4479,
+      "router_z_loss": 0.0,
+      "step": 2330
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 11.9984,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.003,
+      "loss": 2.4643,
+      "router_z_loss": 0.0,
+      "step": 2340
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.0512,
+      "grad_norm": 0.236328125,
+      "learning_rate": 0.003,
+      "loss": 2.7028,
+      "router_z_loss": 0.0,
+      "step": 2350
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.1024,
+      "grad_norm": 0.1796875,
+      "learning_rate": 0.003,
+      "loss": 2.4387,
+      "router_z_loss": 0.0,
+      "step": 2360
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.1536,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.003,
+      "loss": 2.4258,
+      "router_z_loss": 0.0,
+      "step": 2370
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.2048,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.003,
+      "loss": 2.4298,
+      "router_z_loss": 0.0,
+      "step": 2380
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.256,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.003,
+      "loss": 2.4336,
+      "router_z_loss": 0.0,
+      "step": 2390
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.3072,
+      "grad_norm": 0.2021484375,
+      "learning_rate": 0.003,
+      "loss": 2.4322,
+      "router_z_loss": 0.0,
+      "step": 2400
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.3584,
+      "grad_norm": 0.25,
+      "learning_rate": 0.003,
+      "loss": 2.4283,
+      "router_z_loss": 0.0,
+      "step": 2410
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.4096,
+      "grad_norm": 0.228515625,
+      "learning_rate": 0.003,
+      "loss": 2.4257,
+      "router_z_loss": 0.0,
+      "step": 2420
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.4608,
+      "grad_norm": 0.216796875,
+      "learning_rate": 0.003,
+      "loss": 2.4435,
+      "router_z_loss": 0.0,
+      "step": 2430
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.512,
+      "grad_norm": 0.2470703125,
+      "learning_rate": 0.003,
+      "loss": 2.4284,
+      "router_z_loss": 0.0,
+      "step": 2440
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.5632,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.003,
+      "loss": 2.4325,
+      "router_z_loss": 0.0,
+      "step": 2450
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.6144,
+      "grad_norm": 0.201171875,
+      "learning_rate": 0.003,
+      "loss": 2.4266,
+      "router_z_loss": 0.0,
+      "step": 2460
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.6656,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.003,
+      "loss": 2.4236,
+      "router_z_loss": 0.0,
+      "step": 2470
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.7168,
+      "grad_norm": 0.27734375,
+      "learning_rate": 0.003,
+      "loss": 2.4234,
+      "router_z_loss": 0.0,
+      "step": 2480
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.768,
+      "grad_norm": 0.236328125,
+      "learning_rate": 0.003,
+      "loss": 2.4256,
+      "router_z_loss": 0.0,
+      "step": 2490
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.8192,
+      "grad_norm": 0.205078125,
+      "learning_rate": 0.003,
+      "loss": 2.4317,
+      "router_z_loss": 0.0,
+      "step": 2500
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.8704,
+      "grad_norm": 0.24609375,
+      "learning_rate": 0.003,
+      "loss": 2.4258,
+      "router_z_loss": 0.0,
+      "step": 2510
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.9216,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.003,
+      "loss": 2.4369,
+      "router_z_loss": 0.0,
+      "step": 2520
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 12.9728,
+      "grad_norm": 0.203125,
+      "learning_rate": 0.003,
+      "loss": 2.427,
+      "router_z_loss": 0.0,
+      "step": 2530
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.0256,
+      "grad_norm": 0.46484375,
+      "learning_rate": 0.003,
+      "loss": 2.6928,
+      "router_z_loss": 0.0,
+      "step": 2540
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.0768,
+      "grad_norm": 0.1787109375,
+      "learning_rate": 0.003,
+      "loss": 2.4329,
+      "router_z_loss": 0.0,
+      "step": 2550
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.128,
+      "grad_norm": 0.18359375,
+      "learning_rate": 0.003,
+      "loss": 2.4122,
+      "router_z_loss": 0.0,
+      "step": 2560
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.1792,
+      "grad_norm": 0.23046875,
+      "learning_rate": 0.003,
+      "loss": 2.4012,
+      "router_z_loss": 0.0,
+      "step": 2570
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.2304,
+      "grad_norm": 0.2333984375,
+      "learning_rate": 0.003,
+      "loss": 2.4096,
+      "router_z_loss": 0.0,
+      "step": 2580
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.2816,
+      "grad_norm": 0.275390625,
+      "learning_rate": 0.003,
+      "loss": 2.4102,
+      "router_z_loss": 0.0,
+      "step": 2590
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.3328,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.003,
+      "loss": 2.4039,
+      "router_z_loss": 0.0,
+      "step": 2600
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.384,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.003,
+      "loss": 2.4086,
+      "router_z_loss": 0.0,
+      "step": 2610
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.4352,
+      "grad_norm": 0.228515625,
+      "learning_rate": 0.003,
+      "loss": 2.4067,
+      "router_z_loss": 0.0,
+      "step": 2620
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.4864,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.003,
+      "loss": 2.4154,
+      "router_z_loss": 0.0,
+      "step": 2630
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.5376,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.003,
+      "loss": 2.4111,
+      "router_z_loss": 0.0,
+      "step": 2640
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.588799999999999,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.003,
+      "loss": 2.4139,
+      "router_z_loss": 0.0,
+      "step": 2650
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.64,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.003,
+      "loss": 2.4113,
+      "router_z_loss": 0.0,
+      "step": 2660
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.6912,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.003,
+      "loss": 2.4106,
+      "router_z_loss": 0.0,
+      "step": 2670
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.7424,
+      "grad_norm": 0.216796875,
+      "learning_rate": 0.003,
+      "loss": 2.411,
+      "router_z_loss": 0.0,
+      "step": 2680
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.7936,
+      "grad_norm": 0.302734375,
+      "learning_rate": 0.003,
+      "loss": 2.4019,
+      "router_z_loss": 0.0,
+      "step": 2690
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.8448,
+      "grad_norm": 0.23046875,
+      "learning_rate": 0.003,
+      "loss": 2.4074,
+      "router_z_loss": 0.0,
+      "step": 2700
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.896,
+      "grad_norm": 0.2333984375,
+      "learning_rate": 0.003,
+      "loss": 2.4129,
+      "router_z_loss": 0.0,
+      "step": 2710
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.9472,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.003,
+      "loss": 2.3988,
+      "router_z_loss": 0.0,
+      "step": 2720
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 13.9984,
+      "grad_norm": 0.2216796875,
+      "learning_rate": 0.003,
+      "loss": 2.4069,
+      "router_z_loss": 0.0,
+      "step": 2730
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.0512,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.003,
+      "loss": 2.6531,
+      "router_z_loss": 0.0,
+      "step": 2740
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.1024,
+      "grad_norm": 0.2060546875,
+      "learning_rate": 0.003,
+      "loss": 2.3899,
+      "router_z_loss": 0.0,
+      "step": 2750
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.1536,
+      "grad_norm": 0.20703125,
+      "learning_rate": 0.003,
+      "loss": 2.376,
+      "router_z_loss": 0.0,
+      "step": 2760
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.2048,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.003,
+      "loss": 2.3885,
+      "router_z_loss": 0.0,
+      "step": 2770
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.256,
+      "grad_norm": 0.24609375,
+      "learning_rate": 0.003,
+      "loss": 2.3924,
+      "router_z_loss": 0.0,
+      "step": 2780
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.3072,
+      "grad_norm": 0.2177734375,
+      "learning_rate": 0.003,
+      "loss": 2.3908,
+      "router_z_loss": 0.0,
+      "step": 2790
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.3584,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.003,
+      "loss": 2.3913,
+      "router_z_loss": 0.0,
+      "step": 2800
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.4096,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.003,
+      "loss": 2.3867,
+      "router_z_loss": 0.0,
+      "step": 2810
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.4608,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.003,
+      "loss": 2.3902,
+      "router_z_loss": 0.0,
+      "step": 2820
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.512,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.003,
+      "loss": 2.3953,
+      "router_z_loss": 0.0,
+      "step": 2830
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.5632,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.003,
+      "loss": 2.3928,
+      "router_z_loss": 0.0,
+      "step": 2840
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.6144,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.003,
+      "loss": 2.3981,
+      "router_z_loss": 0.0,
+      "step": 2850
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.6656,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.003,
+      "loss": 2.393,
+      "router_z_loss": 0.0,
+      "step": 2860
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.7168,
+      "grad_norm": 0.1845703125,
+      "learning_rate": 0.003,
+      "loss": 2.3859,
+      "router_z_loss": 0.0,
+      "step": 2870
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.768,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.003,
+      "loss": 2.3941,
+      "router_z_loss": 0.0,
+      "step": 2880
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.8192,
+      "grad_norm": 0.2392578125,
+      "learning_rate": 0.003,
+      "loss": 2.3908,
+      "router_z_loss": 0.0,
+      "step": 2890
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.8704,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.003,
+      "loss": 2.392,
+      "router_z_loss": 0.0,
+      "step": 2900
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.9216,
+      "grad_norm": 0.18359375,
+      "learning_rate": 0.003,
+      "loss": 2.3869,
+      "router_z_loss": 0.0,
+      "step": 2910
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 14.9728,
+      "grad_norm": 0.306640625,
+      "learning_rate": 0.003,
+      "loss": 2.3867,
+      "router_z_loss": 0.0,
+      "step": 2920
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.0256,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.003,
+      "loss": 2.6412,
+      "router_z_loss": 0.0,
+      "step": 2930
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.0768,
+      "grad_norm": 0.1943359375,
+      "learning_rate": 0.003,
+      "loss": 2.3859,
+      "router_z_loss": 0.0,
+      "step": 2940
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.128,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.003,
+      "loss": 2.3677,
+      "router_z_loss": 0.0,
+      "step": 2950
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.1792,
+      "grad_norm": 0.2158203125,
+      "learning_rate": 0.003,
+      "loss": 2.366,
+      "router_z_loss": 0.0,
+      "step": 2960
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.2304,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.003,
+      "loss": 2.3677,
+      "router_z_loss": 0.0,
+      "step": 2970
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.2816,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.003,
+      "loss": 2.3678,
+      "router_z_loss": 0.0,
+      "step": 2980
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.3328,
+      "grad_norm": 0.291015625,
+      "learning_rate": 0.003,
+      "loss": 2.3698,
+      "router_z_loss": 0.0,
+      "step": 2990
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.384,
+      "grad_norm": 0.2119140625,
+      "learning_rate": 0.003,
+      "loss": 2.3727,
+      "router_z_loss": 0.0,
+      "step": 3000
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.4352,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.003,
+      "loss": 2.3714,
+      "router_z_loss": 0.0,
+      "step": 3010
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.4864,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.003,
+      "loss": 2.3734,
+      "router_z_loss": 0.0,
+      "step": 3020
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.5376,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.003,
+      "loss": 2.3697,
+      "router_z_loss": 0.0,
+      "step": 3030
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.588799999999999,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.003,
+      "loss": 2.3835,
+      "router_z_loss": 0.0,
+      "step": 3040
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.64,
+      "grad_norm": 0.2734375,
+      "learning_rate": 0.003,
+      "loss": 2.3769,
+      "router_z_loss": 0.0,
+      "step": 3050
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.6912,
+      "grad_norm": 0.2421875,
+      "learning_rate": 0.003,
+      "loss": 2.3736,
+      "router_z_loss": 0.0,
+      "step": 3060
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.7424,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.003,
+      "loss": 2.3659,
+      "router_z_loss": 0.0,
+      "step": 3070
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.7936,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.003,
+      "loss": 2.3654,
+      "router_z_loss": 0.0,
+      "step": 3080
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.8448,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.003,
+      "loss": 2.3832,
+      "router_z_loss": 0.0,
+      "step": 3090
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.896,
+      "grad_norm": 0.1943359375,
+      "learning_rate": 0.003,
+      "loss": 2.3691,
+      "router_z_loss": 0.0,
+      "step": 3100
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.9472,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.003,
+      "loss": 2.3783,
+      "router_z_loss": 0.0,
+      "step": 3110
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 15.9984,
+      "grad_norm": 0.2470703125,
+      "learning_rate": 0.003,
+      "loss": 2.371,
+      "router_z_loss": 0.0,
+      "step": 3120
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.0512,
+      "grad_norm": 0.24609375,
+      "learning_rate": 0.003,
+      "loss": 2.6061,
+      "router_z_loss": 0.0,
+      "step": 3130
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.1024,
+      "grad_norm": 0.2021484375,
+      "learning_rate": 0.003,
+      "loss": 2.3592,
+      "router_z_loss": 0.0,
+      "step": 3140
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.1536,
+      "grad_norm": 0.24609375,
+      "learning_rate": 0.003,
+      "loss": 2.3532,
+      "router_z_loss": 0.0,
+      "step": 3150
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.2048,
+      "grad_norm": 0.2392578125,
+      "learning_rate": 0.003,
+      "loss": 2.3589,
+      "router_z_loss": 0.0,
+      "step": 3160
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.256,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.003,
+      "loss": 2.3526,
+      "router_z_loss": 0.0,
+      "step": 3170
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.3072,
+      "grad_norm": 0.2734375,
+      "learning_rate": 0.003,
+      "loss": 2.3661,
+      "router_z_loss": 0.0,
+      "step": 3180
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.3584,
+      "grad_norm": 0.46875,
+      "learning_rate": 0.003,
+      "loss": 2.3769,
+      "router_z_loss": 0.0,
+      "step": 3190
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.4096,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.003,
+      "loss": 2.389,
+      "router_z_loss": 0.0,
+      "step": 3200
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.4608,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.003,
+      "loss": 2.3585,
+      "router_z_loss": 0.0,
+      "step": 3210
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.512,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.003,
+      "loss": 2.3547,
+      "router_z_loss": 0.0,
+      "step": 3220
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.5632,
+      "grad_norm": 0.236328125,
+      "learning_rate": 0.003,
+      "loss": 2.36,
+      "router_z_loss": 0.0,
+      "step": 3230
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.6144,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.003,
+      "loss": 2.3523,
+      "router_z_loss": 0.0,
+      "step": 3240
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.6656,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.003,
+      "loss": 2.3612,
+      "router_z_loss": 0.0,
+      "step": 3250
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.7168,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.003,
+      "loss": 2.3605,
+      "router_z_loss": 0.0,
+      "step": 3260
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.768,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.003,
+      "loss": 2.3628,
+      "router_z_loss": 0.0,
+      "step": 3270
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.8192,
+      "grad_norm": 0.208984375,
+      "learning_rate": 0.003,
+      "loss": 2.3562,
+      "router_z_loss": 0.0,
+      "step": 3280
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.8704,
+      "grad_norm": 0.203125,
+      "learning_rate": 0.003,
+      "loss": 2.364,
+      "router_z_loss": 0.0,
+      "step": 3290
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.9216,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.003,
+      "loss": 2.3577,
+      "router_z_loss": 0.0,
+      "step": 3300
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 16.9728,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.003,
+      "loss": 2.3625,
+      "router_z_loss": 0.0,
+      "step": 3310
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.0256,
+      "grad_norm": 0.3125,
+      "learning_rate": 0.003,
+      "loss": 2.5916,
+      "router_z_loss": 0.0,
+      "step": 3320
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.0768,
+      "grad_norm": 0.1943359375,
+      "learning_rate": 0.003,
+      "loss": 2.3568,
+      "router_z_loss": 0.0,
+      "step": 3330
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.128,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.003,
+      "loss": 2.3433,
+      "router_z_loss": 0.0,
+      "step": 3340
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.1792,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.003,
+      "loss": 2.3373,
+      "router_z_loss": 0.0,
+      "step": 3350
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.2304,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.003,
+      "loss": 2.3511,
+      "router_z_loss": 0.0,
+      "step": 3360
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.2816,
+      "grad_norm": 0.2490234375,
+      "learning_rate": 0.003,
+      "loss": 2.3433,
+      "router_z_loss": 0.0,
+      "step": 3370
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.3328,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.003,
+      "loss": 2.3534,
+      "router_z_loss": 0.0,
+      "step": 3380
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.384,
+      "grad_norm": 0.22265625,
+      "learning_rate": 0.003,
+      "loss": 2.3519,
+      "router_z_loss": 0.0,
+      "step": 3390
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.4352,
+      "grad_norm": 0.19921875,
+      "learning_rate": 0.003,
+      "loss": 2.3326,
+      "router_z_loss": 0.0,
+      "step": 3400
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.4864,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.003,
+      "loss": 2.3422,
+      "router_z_loss": 0.0,
+      "step": 3410
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.5376,
+      "grad_norm": 0.30859375,
+      "learning_rate": 0.003,
+      "loss": 2.3433,
+      "router_z_loss": 0.0,
+      "step": 3420
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.5888,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.003,
+      "loss": 2.3479,
+      "router_z_loss": 0.0,
+      "step": 3430
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.64,
+      "grad_norm": 0.23828125,
+      "learning_rate": 0.003,
+      "loss": 2.3461,
+      "router_z_loss": 0.0,
+      "step": 3440
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.6912,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.003,
+      "loss": 2.3472,
+      "router_z_loss": 0.0,
+      "step": 3450
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.7424,
+      "grad_norm": 0.23046875,
+      "learning_rate": 0.003,
+      "loss": 2.3466,
+      "router_z_loss": 0.0,
+      "step": 3460
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.7936,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.003,
+      "loss": 2.3429,
+      "router_z_loss": 0.0,
+      "step": 3470
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.8448,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.003,
+      "loss": 2.3371,
+      "router_z_loss": 0.0,
+      "step": 3480
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.896,
+      "grad_norm": 0.23046875,
+      "learning_rate": 0.003,
+      "loss": 2.3425,
+      "router_z_loss": 0.0,
+      "step": 3490
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.9472,
+      "grad_norm": 0.30078125,
+      "learning_rate": 0.003,
+      "loss": 2.3435,
+      "router_z_loss": 0.0,
+      "step": 3500
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 17.9984,
+      "grad_norm": 0.201171875,
+      "learning_rate": 0.003,
+      "loss": 2.346,
+      "router_z_loss": 0.0,
+      "step": 3510
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.0512,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.003,
+      "loss": 2.5893,
+      "router_z_loss": 0.0,
+      "step": 3520
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.1024,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.003,
+      "loss": 2.3269,
+      "router_z_loss": 0.0,
+      "step": 3530
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.1536,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.003,
+      "loss": 2.3263,
+      "router_z_loss": 0.0,
+      "step": 3540
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.2048,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.003,
+      "loss": 2.3229,
+      "router_z_loss": 0.0,
+      "step": 3550
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.256,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.003,
+      "loss": 2.3294,
+      "router_z_loss": 0.0,
+      "step": 3560
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.3072,
+      "grad_norm": 0.220703125,
+      "learning_rate": 0.003,
+      "loss": 2.3249,
+      "router_z_loss": 0.0,
+      "step": 3570
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.3584,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.003,
+      "loss": 2.3299,
+      "router_z_loss": 0.0,
+      "step": 3580
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.4096,
+      "grad_norm": 0.2373046875,
+      "learning_rate": 0.003,
+      "loss": 2.3279,
+      "router_z_loss": 0.0,
+      "step": 3590
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.4608,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.003,
+      "loss": 2.3334,
+      "router_z_loss": 0.0,
+      "step": 3600
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.512,
+      "grad_norm": 0.2333984375,
+      "learning_rate": 0.003,
+      "loss": 2.3275,
+      "router_z_loss": 0.0,
+      "step": 3610
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.5632,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.003,
+      "loss": 2.3335,
+      "router_z_loss": 0.0,
+      "step": 3620
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.6144,
+      "grad_norm": 0.224609375,
+      "learning_rate": 0.003,
+      "loss": 2.3309,
+      "router_z_loss": 0.0,
+      "step": 3630
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.6656,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.003,
+      "loss": 2.3376,
+      "router_z_loss": 0.0,
+      "step": 3640
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.7168,
+      "grad_norm": 0.30859375,
+      "learning_rate": 0.003,
+      "loss": 2.3377,
+      "router_z_loss": 0.0,
+      "step": 3650
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.768,
+      "grad_norm": 0.1904296875,
+      "learning_rate": 0.003,
+      "loss": 2.3373,
+      "router_z_loss": 0.0,
+      "step": 3660
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.8192,
+      "grad_norm": 0.203125,
+      "learning_rate": 0.003,
+      "loss": 2.3309,
+      "router_z_loss": 0.0,
+      "step": 3670
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.8704,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.003,
+      "loss": 2.3363,
+      "router_z_loss": 0.0,
+      "step": 3680
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.9216,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.003,
+      "loss": 2.3413,
+      "router_z_loss": 0.0,
+      "step": 3690
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 18.9728,
+      "grad_norm": 0.1953125,
+      "learning_rate": 0.003,
+      "loss": 2.3221,
+      "router_z_loss": 0.0,
+      "step": 3700
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.0256,
+      "grad_norm": 0.37109375,
+      "learning_rate": 0.003,
+      "loss": 2.5891,
+      "router_z_loss": 0.0,
+      "step": 3710
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.0768,
+      "grad_norm": 0.2080078125,
+      "learning_rate": 0.003,
+      "loss": 2.3331,
+      "router_z_loss": 0.0,
+      "step": 3720
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.128,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.003,
+      "loss": 2.3133,
+      "router_z_loss": 0.0,
+      "step": 3730
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.1792,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.003,
+      "loss": 2.3156,
+      "router_z_loss": 0.0,
+      "step": 3740
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.2304,
+      "grad_norm": 0.1923828125,
+      "learning_rate": 0.003,
+      "loss": 2.3177,
+      "router_z_loss": 0.0,
+      "step": 3750
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.2816,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.003,
+      "loss": 2.3054,
+      "router_z_loss": 0.0,
+      "step": 3760
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.3328,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.003,
+      "loss": 2.3219,
+      "router_z_loss": 0.0,
+      "step": 3770
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.384,
+      "grad_norm": 0.19921875,
+      "learning_rate": 0.003,
+      "loss": 2.3164,
+      "router_z_loss": 0.0,
+      "step": 3780
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.4352,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.003,
+      "loss": 2.3115,
+      "router_z_loss": 0.0,
+      "step": 3790
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.4864,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.003,
+      "loss": 2.3204,
+      "router_z_loss": 0.0,
+      "step": 3800
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.5376,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.003,
+      "loss": 2.3235,
+      "router_z_loss": 0.0,
+      "step": 3810
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.5888,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.003,
+      "loss": 2.3196,
+      "router_z_loss": 0.0,
+      "step": 3820
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.64,
+      "grad_norm": 0.2119140625,
+      "learning_rate": 0.003,
+      "loss": 2.3186,
+      "router_z_loss": 0.0,
+      "step": 3830
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.6912,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.003,
+      "loss": 2.3255,
+      "router_z_loss": 0.0,
+      "step": 3840
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.7424,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.003,
+      "loss": 2.3227,
+      "router_z_loss": 0.0,
+      "step": 3850
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.7936,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.003,
+      "loss": 2.3226,
+      "router_z_loss": 0.0,
+      "step": 3860
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.8448,
+      "grad_norm": 0.2373046875,
+      "learning_rate": 0.003,
+      "loss": 2.321,
+      "router_z_loss": 0.0,
+      "step": 3870
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.896,
+      "grad_norm": 0.306640625,
+      "learning_rate": 0.003,
+      "loss": 2.3289,
+      "router_z_loss": 0.0,
+      "step": 3880
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.9472,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.003,
+      "loss": 2.3259,
+      "router_z_loss": 0.0,
+      "step": 3890
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 19.9984,
+      "grad_norm": 0.208984375,
+      "learning_rate": 0.003,
+      "loss": 2.3186,
+      "router_z_loss": 0.0,
+      "step": 3900
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.0512,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.003,
+      "loss": 2.5579,
+      "router_z_loss": 0.0,
+      "step": 3910
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.1024,
+      "grad_norm": 0.2001953125,
+      "learning_rate": 0.003,
+      "loss": 2.3122,
+      "router_z_loss": 0.0,
+      "step": 3920
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.1536,
+      "grad_norm": 0.1962890625,
+      "learning_rate": 0.003,
+      "loss": 2.3019,
+      "router_z_loss": 0.0,
+      "step": 3930
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.2048,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.003,
+      "loss": 2.3079,
+      "router_z_loss": 0.0,
+      "step": 3940
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.256,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.003,
+      "loss": 2.3077,
+      "router_z_loss": 0.0,
+      "step": 3950
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.3072,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.003,
+      "loss": 2.3094,
+      "router_z_loss": 0.0,
+      "step": 3960
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.3584,
+      "grad_norm": 0.216796875,
+      "learning_rate": 0.003,
+      "loss": 2.3071,
+      "router_z_loss": 0.0,
+      "step": 3970
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.4096,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.003,
+      "loss": 2.307,
+      "router_z_loss": 0.0,
+      "step": 3980
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.4608,
+      "grad_norm": 0.2392578125,
+      "learning_rate": 0.003,
+      "loss": 2.3117,
+      "router_z_loss": 0.0,
+      "step": 3990
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.512,
+      "grad_norm": 0.228515625,
+      "learning_rate": 0.003,
+      "loss": 2.305,
+      "router_z_loss": 0.0,
+      "step": 4000
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.5632,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.003,
+      "loss": 2.3096,
+      "router_z_loss": 0.0,
+      "step": 4010
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.6144,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.003,
+      "loss": 2.3166,
+      "router_z_loss": 0.0,
+      "step": 4020
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.6656,
+      "grad_norm": 0.216796875,
+      "learning_rate": 0.003,
+      "loss": 2.3105,
+      "router_z_loss": 0.0,
+      "step": 4030
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.7168,
+      "grad_norm": 0.2333984375,
+      "learning_rate": 0.003,
+      "loss": 2.3168,
+      "router_z_loss": 0.0,
+      "step": 4040
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.768,
+      "grad_norm": 0.203125,
+      "learning_rate": 0.003,
+      "loss": 2.3146,
+      "router_z_loss": 0.0,
+      "step": 4050
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.8192,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.003,
+      "loss": 2.3069,
+      "router_z_loss": 0.0,
+      "step": 4060
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.8704,
+      "grad_norm": 0.2041015625,
+      "learning_rate": 0.003,
+      "loss": 2.3131,
+      "router_z_loss": 0.0,
+      "step": 4070
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.9216,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.003,
+      "loss": 2.3062,
+      "router_z_loss": 0.0,
+      "step": 4080
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 20.9728,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.003,
+      "loss": 2.3078,
+      "router_z_loss": 0.0,
+      "step": 4090
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.0256,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.003,
+      "loss": 2.5517,
+      "router_z_loss": 0.0,
+      "step": 4100
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.0768,
+      "grad_norm": 0.189453125,
+      "learning_rate": 0.003,
+      "loss": 2.3115,
+      "router_z_loss": 0.0,
+      "step": 4110
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.128,
+      "grad_norm": 0.2216796875,
+      "learning_rate": 0.003,
+      "loss": 2.2996,
+      "router_z_loss": 0.0,
+      "step": 4120
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.1792,
+      "grad_norm": 0.306640625,
+      "learning_rate": 0.003,
+      "loss": 2.2861,
+      "router_z_loss": 0.0,
+      "step": 4130
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.2304,
+      "grad_norm": 0.212890625,
+      "learning_rate": 0.003,
+      "loss": 2.2964,
+      "router_z_loss": 0.0,
+      "step": 4140
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.2816,
+      "grad_norm": 0.330078125,
+      "learning_rate": 0.003,
+      "loss": 2.2974,
+      "router_z_loss": 0.0,
+      "step": 4150
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.3328,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.003,
+      "loss": 2.2996,
+      "router_z_loss": 0.0,
+      "step": 4160
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.384,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.003,
+      "loss": 2.2975,
+      "router_z_loss": 0.0,
+      "step": 4170
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.4352,
+      "grad_norm": 0.27734375,
+      "learning_rate": 0.003,
+      "loss": 2.2967,
+      "router_z_loss": 0.0,
+      "step": 4180
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.4864,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.003,
+      "loss": 2.3015,
+      "router_z_loss": 0.0,
+      "step": 4190
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.5376,
+      "grad_norm": 0.2431640625,
+      "learning_rate": 0.003,
+      "loss": 2.3039,
+      "router_z_loss": 0.0,
+      "step": 4200
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.5888,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.003,
+      "loss": 2.2986,
+      "router_z_loss": 0.0,
+      "step": 4210
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.64,
+      "grad_norm": 0.2236328125,
+      "learning_rate": 0.003,
+      "loss": 2.3019,
+      "router_z_loss": 0.0,
+      "step": 4220
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.6912,
+      "grad_norm": 0.384765625,
+      "learning_rate": 0.003,
+      "loss": 2.3001,
+      "router_z_loss": 0.0,
+      "step": 4230
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.7424,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.003,
+      "loss": 2.3052,
+      "router_z_loss": 0.0,
+      "step": 4240
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.7936,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.003,
+      "loss": 2.3046,
+      "router_z_loss": 0.0,
+      "step": 4250
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.8448,
+      "grad_norm": 0.2001953125,
+      "learning_rate": 0.003,
+      "loss": 2.2952,
+      "router_z_loss": 0.0,
+      "step": 4260
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.896,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.003,
+      "loss": 2.3044,
+      "router_z_loss": 0.0,
+      "step": 4270
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.9472,
+      "grad_norm": 0.23828125,
+      "learning_rate": 0.003,
+      "loss": 2.297,
+      "router_z_loss": 0.0,
+      "step": 4280
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 21.9984,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.003,
+      "loss": 2.3065,
+      "router_z_loss": 0.0,
+      "step": 4290
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.0512,
+      "grad_norm": 0.220703125,
+      "learning_rate": 0.003,
+      "loss": 2.5303,
+      "router_z_loss": 0.0,
+      "step": 4300
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.1024,
+      "grad_norm": 0.208984375,
+      "learning_rate": 0.003,
+      "loss": 2.2945,
+      "router_z_loss": 0.0,
+      "step": 4310
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.1536,
+      "grad_norm": 0.2001953125,
+      "learning_rate": 0.003,
+      "loss": 2.2929,
+      "router_z_loss": 0.0,
+      "step": 4320
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.2048,
+      "grad_norm": 0.1962890625,
+      "learning_rate": 0.003,
+      "loss": 2.2863,
+      "router_z_loss": 0.0,
+      "step": 4330
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.256,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.003,
+      "loss": 2.2848,
+      "router_z_loss": 0.0,
+      "step": 4340
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.3072,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.003,
+      "loss": 2.2808,
+      "router_z_loss": 0.0,
+      "step": 4350
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.3584,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.003,
+      "loss": 2.295,
+      "router_z_loss": 0.0,
+      "step": 4360
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.4096,
+      "grad_norm": 0.216796875,
+      "learning_rate": 0.003,
+      "loss": 2.2885,
+      "router_z_loss": 0.0,
+      "step": 4370
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.4608,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.003,
+      "loss": 2.293,
+      "router_z_loss": 0.0,
+      "step": 4380
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.512,
+      "grad_norm": 0.3046875,
+      "learning_rate": 0.003,
+      "loss": 2.2869,
+      "router_z_loss": 0.0,
+      "step": 4390
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.5632,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.003,
+      "loss": 2.2914,
+      "router_z_loss": 0.0,
+      "step": 4400
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.6144,
+      "grad_norm": 0.228515625,
+      "learning_rate": 0.003,
+      "loss": 2.2923,
+      "router_z_loss": 0.0,
+      "step": 4410
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.6656,
+      "grad_norm": 0.2421875,
+      "learning_rate": 0.003,
+      "loss": 2.2918,
+      "router_z_loss": 0.0,
+      "step": 4420
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.7168,
+      "grad_norm": 0.365234375,
+      "learning_rate": 0.003,
+      "loss": 2.2946,
+      "router_z_loss": 0.0,
+      "step": 4430
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.768,
+      "grad_norm": 0.224609375,
+      "learning_rate": 0.003,
+      "loss": 2.2953,
+      "router_z_loss": 0.0,
+      "step": 4440
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.8192,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.003,
+      "loss": 2.2919,
+      "router_z_loss": 0.0,
+      "step": 4450
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.8704,
+      "grad_norm": 0.37109375,
+      "learning_rate": 0.003,
+      "loss": 2.2963,
+      "router_z_loss": 0.0,
+      "step": 4460
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.9216,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.003,
+      "loss": 2.3011,
+      "router_z_loss": 0.0,
+      "step": 4470
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 22.9728,
+      "grad_norm": 0.2333984375,
+      "learning_rate": 0.003,
+      "loss": 2.288,
+      "router_z_loss": 0.0,
+      "step": 4480
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.0256,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.003,
+      "loss": 2.5194,
+      "router_z_loss": 0.0,
+      "step": 4490
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.0768,
+      "grad_norm": 0.2158203125,
+      "learning_rate": 0.003,
+      "loss": 2.29,
+      "router_z_loss": 0.0,
+      "step": 4500
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.128,
+      "grad_norm": 0.2470703125,
+      "learning_rate": 0.003,
+      "loss": 2.2739,
+      "router_z_loss": 0.0,
+      "step": 4510
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.1792,
+      "grad_norm": 0.2060546875,
+      "learning_rate": 0.003,
+      "loss": 2.2745,
+      "router_z_loss": 0.0,
+      "step": 4520
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.2304,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.003,
+      "loss": 2.2817,
+      "router_z_loss": 0.0,
+      "step": 4530
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.2816,
+      "grad_norm": 0.23828125,
+      "learning_rate": 0.003,
+      "loss": 2.2819,
+      "router_z_loss": 0.0,
+      "step": 4540
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.3328,
+      "grad_norm": 0.205078125,
+      "learning_rate": 0.003,
+      "loss": 2.2771,
+      "router_z_loss": 0.0,
+      "step": 4550
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.384,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.003,
+      "loss": 2.2759,
+      "router_z_loss": 0.0,
+      "step": 4560
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.4352,
+      "grad_norm": 0.2490234375,
+      "learning_rate": 0.003,
+      "loss": 2.2858,
+      "router_z_loss": 0.0,
+      "step": 4570
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.4864,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.003,
+      "loss": 2.2792,
+      "router_z_loss": 0.0,
+      "step": 4580
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.5376,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.003,
+      "loss": 2.2787,
+      "router_z_loss": 0.0,
+      "step": 4590
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.5888,
+      "grad_norm": 0.23828125,
+      "learning_rate": 0.003,
+      "loss": 2.2848,
+      "router_z_loss": 0.0,
+      "step": 4600
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.64,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.003,
+      "loss": 2.2912,
+      "router_z_loss": 0.0,
+      "step": 4610
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.6912,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.003,
+      "loss": 2.2804,
+      "router_z_loss": 0.0,
+      "step": 4620
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.7424,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.003,
+      "loss": 2.2858,
+      "router_z_loss": 0.0,
+      "step": 4630
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.7936,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.003,
+      "loss": 2.276,
+      "router_z_loss": 0.0,
+      "step": 4640
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.8448,
+      "grad_norm": 0.25,
+      "learning_rate": 0.003,
+      "loss": 2.2804,
+      "router_z_loss": 0.0,
+      "step": 4650
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.896,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.003,
+      "loss": 2.2768,
+      "router_z_loss": 0.0,
+      "step": 4660
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.9472,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.003,
+      "loss": 2.2873,
+      "router_z_loss": 0.0,
+      "step": 4670
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 23.9984,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.003,
+      "loss": 2.2838,
+      "router_z_loss": 0.0,
+      "step": 4680
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.0512,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.003,
+      "loss": 2.5179,
+      "router_z_loss": 0.0,
+      "step": 4690
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.1024,
+      "grad_norm": 0.2060546875,
+      "learning_rate": 0.003,
+      "loss": 2.2732,
+      "router_z_loss": 0.0,
+      "step": 4700
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.1536,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.003,
+      "loss": 2.2682,
+      "router_z_loss": 0.0,
+      "step": 4710
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.2048,
+      "grad_norm": 0.2041015625,
+      "learning_rate": 0.003,
+      "loss": 2.2655,
+      "router_z_loss": 0.0,
+      "step": 4720
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.256,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.003,
+      "loss": 2.2717,
+      "router_z_loss": 0.0,
+      "step": 4730
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.3072,
+      "grad_norm": 0.2236328125,
+      "learning_rate": 0.003,
+      "loss": 2.2683,
+      "router_z_loss": 0.0,
+      "step": 4740
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.3584,
+      "grad_norm": 0.193359375,
+      "learning_rate": 0.003,
+      "loss": 2.2719,
+      "router_z_loss": 0.0,
+      "step": 4750
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.4096,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.003,
+      "loss": 2.2755,
+      "router_z_loss": 0.0,
+      "step": 4760
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.4608,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.003,
+      "loss": 2.2796,
+      "router_z_loss": 0.0,
+      "step": 4770
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.512,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.003,
+      "loss": 2.2727,
+      "router_z_loss": 0.0,
+      "step": 4780
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.5632,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.003,
+      "loss": 2.28,
+      "router_z_loss": 0.0,
+      "step": 4790
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.6144,
+      "grad_norm": 0.291015625,
+      "learning_rate": 0.003,
+      "loss": 2.2707,
+      "router_z_loss": 0.0,
+      "step": 4800
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.6656,
+      "grad_norm": 0.1982421875,
+      "learning_rate": 0.003,
+      "loss": 2.2735,
+      "router_z_loss": 0.0,
+      "step": 4810
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.7168,
+      "grad_norm": 0.30078125,
+      "learning_rate": 0.003,
+      "loss": 2.2702,
+      "router_z_loss": 0.0,
+      "step": 4820
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.768,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.003,
+      "loss": 2.2765,
+      "router_z_loss": 0.0,
+      "step": 4830
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.8192,
+      "grad_norm": 0.353515625,
+      "learning_rate": 0.003,
+      "loss": 2.2724,
+      "router_z_loss": 0.0,
+      "step": 4840
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.8704,
+      "grad_norm": 0.220703125,
+      "learning_rate": 0.003,
+      "loss": 2.2781,
+      "router_z_loss": 0.0,
+      "step": 4850
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.9216,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.003,
+      "loss": 2.2737,
+      "router_z_loss": 0.0,
+      "step": 4860
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 24.9728,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.003,
+      "loss": 2.2763,
+      "router_z_loss": 0.0,
+      "step": 4870
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.0256,
+      "grad_norm": 0.30859375,
+      "learning_rate": 0.003,
+      "loss": 2.5145,
+      "router_z_loss": 0.0,
+      "step": 4880
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.0768,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.003,
+      "loss": 2.273,
+      "router_z_loss": 0.0,
+      "step": 4890
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.128,
+      "grad_norm": 0.2392578125,
+      "learning_rate": 0.003,
+      "loss": 2.2639,
+      "router_z_loss": 0.0,
+      "step": 4900
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.1792,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.003,
+      "loss": 2.2609,
+      "router_z_loss": 0.0,
+      "step": 4910
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.2304,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.003,
+      "loss": 2.2573,
+      "router_z_loss": 0.0,
+      "step": 4920
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.2816,
+      "grad_norm": 0.201171875,
+      "learning_rate": 0.003,
+      "loss": 2.2669,
+      "router_z_loss": 0.0,
+      "step": 4930
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.3328,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.003,
+      "loss": 2.2681,
+      "router_z_loss": 0.0,
+      "step": 4940
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.384,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.003,
+      "loss": 2.2659,
+      "router_z_loss": 0.0,
+      "step": 4950
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.4352,
+      "grad_norm": 0.3046875,
+      "learning_rate": 0.003,
+      "loss": 2.2657,
+      "router_z_loss": 0.0,
+      "step": 4960
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.4864,
+      "grad_norm": 0.32421875,
+      "learning_rate": 0.003,
+      "loss": 2.2681,
+      "router_z_loss": 0.0,
+      "step": 4970
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.5376,
+      "grad_norm": 0.2021484375,
+      "learning_rate": 0.003,
+      "loss": 2.2655,
+      "router_z_loss": 0.0,
+      "step": 4980
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.5888,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.003,
+      "loss": 2.2683,
+      "router_z_loss": 0.0,
+      "step": 4990
+    },
+    {
+      "balancing_entropy": 0.0,
+      "balancing_loss": 0.0,
+      "epoch": 25.64,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.003,
+      "loss": 2.2736,
+      "router_z_loss": 0.0,
+      "step": 5000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 19500,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 100,
+  "save_steps": 5000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.066322671894528e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}