diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7061 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.5,
+  "eval_steps": 2000,
+  "global_step": 10000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 6.03125,
+      "learning_rate": 3.0000000000000004e-07,
+      "loss": 0.5688,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.5,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 0.4803,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 4.6875,
+      "learning_rate": 9e-07,
+      "loss": 0.6168,
+      "step": 30
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 6.40625,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 0.5197,
+      "step": 40
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 7.0,
+      "learning_rate": 1.5e-06,
+      "loss": 0.5441,
+      "step": 50
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.8e-06,
+      "loss": 0.6031,
+      "step": 60
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.4375,
+      "learning_rate": 2.1000000000000002e-06,
+      "loss": 0.4926,
+      "step": 70
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 6.4375,
+      "learning_rate": 2.4000000000000003e-06,
+      "loss": 0.5795,
+      "step": 80
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 4.75,
+      "learning_rate": 2.7e-06,
+      "loss": 0.5599,
+      "step": 90
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.125,
+      "learning_rate": 3e-06,
+      "loss": 0.5279,
+      "step": 100
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.96875,
+      "learning_rate": 3.3e-06,
+      "loss": 0.5455,
+      "step": 110
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.09375,
+      "learning_rate": 3.6e-06,
+      "loss": 0.5655,
+      "step": 120
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.25,
+      "learning_rate": 3.9e-06,
+      "loss": 0.4635,
+      "step": 130
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.8125,
+      "learning_rate": 4.2000000000000004e-06,
+      "loss": 0.64,
+      "step": 140
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.9375,
+      "learning_rate": 4.5e-06,
+      "loss": 0.4565,
+      "step": 150
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.09375,
+      "learning_rate": 4.800000000000001e-06,
+      "loss": 0.5357,
+      "step": 160
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.59375,
+      "learning_rate": 5.1e-06,
+      "loss": 0.4885,
+      "step": 170
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.9375,
+      "learning_rate": 5.4e-06,
+      "loss": 0.5125,
+      "step": 180
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.21875,
+      "learning_rate": 5.7000000000000005e-06,
+      "loss": 0.4882,
+      "step": 190
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 3.625,
+      "learning_rate": 6e-06,
+      "loss": 0.4504,
+      "step": 200
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.15625,
+      "learning_rate": 6.3e-06,
+      "loss": 0.404,
+      "step": 210
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.1875,
+      "learning_rate": 6.6e-06,
+      "loss": 0.4566,
+      "step": 220
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.125,
+      "learning_rate": 6.900000000000001e-06,
+      "loss": 0.4597,
+      "step": 230
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.0625,
+      "learning_rate": 7.2e-06,
+      "loss": 0.3678,
+      "step": 240
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.09375,
+      "learning_rate": 7.5e-06,
+      "loss": 0.5107,
+      "step": 250
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.15625,
+      "learning_rate": 7.8e-06,
+      "loss": 0.524,
+      "step": 260
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.78125,
+      "learning_rate": 8.1e-06,
+      "loss": 0.4228,
+      "step": 270
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.71875,
+      "learning_rate": 8.400000000000001e-06,
+      "loss": 0.4605,
+      "step": 280
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.4375,
+      "learning_rate": 8.7e-06,
+      "loss": 0.4211,
+      "step": 290
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.84375,
+      "learning_rate": 9e-06,
+      "loss": 0.438,
+      "step": 300
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.375,
+      "learning_rate": 9.3e-06,
+      "loss": 0.4466,
+      "step": 310
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.265625,
+      "learning_rate": 9.600000000000001e-06,
+      "loss": 0.3544,
+      "step": 320
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.703125,
+      "learning_rate": 9.9e-06,
+      "loss": 0.3532,
+      "step": 330
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.25,
+      "learning_rate": 1.02e-05,
+      "loss": 0.355,
+      "step": 340
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.25,
+      "learning_rate": 1.05e-05,
+      "loss": 0.3628,
+      "step": 350
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.9375,
+      "learning_rate": 1.08e-05,
+      "loss": 0.3914,
+      "step": 360
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.3125,
+      "learning_rate": 1.11e-05,
+      "loss": 0.4033,
+      "step": 370
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.625,
+      "learning_rate": 1.1400000000000001e-05,
+      "loss": 0.329,
+      "step": 380
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.75,
+      "learning_rate": 1.1700000000000001e-05,
+      "loss": 0.3501,
+      "step": 390
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.21875,
+      "learning_rate": 1.2e-05,
+      "loss": 0.3402,
+      "step": 400
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.703125,
+      "learning_rate": 1.2299999999999999e-05,
+      "loss": 0.3807,
+      "step": 410
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.26e-05,
+      "loss": 0.38,
+      "step": 420
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.29e-05,
+      "loss": 0.3419,
+      "step": 430
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 5.96875,
+      "learning_rate": 1.32e-05,
+      "loss": 0.4231,
+      "step": 440
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.3500000000000001e-05,
+      "loss": 0.3869,
+      "step": 450
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.375,
+      "learning_rate": 1.3800000000000002e-05,
+      "loss": 0.3628,
+      "step": 460
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.59375,
+      "learning_rate": 1.4099999999999999e-05,
+      "loss": 0.3268,
+      "step": 470
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.15625,
+      "learning_rate": 1.44e-05,
+      "loss": 0.4242,
+      "step": 480
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 5.40625,
+      "learning_rate": 1.47e-05,
+      "loss": 0.3388,
+      "step": 490
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.5e-05,
+      "loss": 0.3188,
+      "step": 500
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.03125,
+      "learning_rate": 1.4984210526315789e-05,
+      "loss": 0.3138,
+      "step": 510
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.9375,
+      "learning_rate": 1.496842105263158e-05,
+      "loss": 0.3381,
+      "step": 520
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.0625,
+      "learning_rate": 1.4952631578947368e-05,
+      "loss": 0.3579,
+      "step": 530
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.40625,
+      "learning_rate": 1.4936842105263158e-05,
+      "loss": 0.308,
+      "step": 540
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.125,
+      "learning_rate": 1.4921052631578947e-05,
+      "loss": 0.4211,
+      "step": 550
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.375,
+      "learning_rate": 1.4905263157894737e-05,
+      "loss": 0.3596,
+      "step": 560
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.4375,
+      "learning_rate": 1.4889473684210526e-05,
+      "loss": 0.3154,
+      "step": 570
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.28125,
+      "learning_rate": 1.4873684210526315e-05,
+      "loss": 0.3132,
+      "step": 580
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.9375,
+      "learning_rate": 1.4857894736842107e-05,
+      "loss": 0.3773,
+      "step": 590
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.484375,
+      "learning_rate": 1.4842105263157895e-05,
+      "loss": 0.3634,
+      "step": 600
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.875,
+      "learning_rate": 1.4826315789473686e-05,
+      "loss": 0.2933,
+      "step": 610
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.515625,
+      "learning_rate": 1.4810526315789474e-05,
+      "loss": 0.3159,
+      "step": 620
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.4794736842105265e-05,
+      "loss": 0.3357,
+      "step": 630
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.875,
+      "learning_rate": 1.4778947368421053e-05,
+      "loss": 0.3478,
+      "step": 640
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.4763157894736842e-05,
+      "loss": 0.3873,
+      "step": 650
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.03125,
+      "learning_rate": 1.4747368421052632e-05,
+      "loss": 0.2899,
+      "step": 660
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.78125,
+      "learning_rate": 1.4731578947368421e-05,
+      "loss": 0.3353,
+      "step": 670
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.0625,
+      "learning_rate": 1.4715789473684211e-05,
+      "loss": 0.3488,
+      "step": 680
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 8.625,
+      "learning_rate": 1.47e-05,
+      "loss": 0.3456,
+      "step": 690
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.125,
+      "learning_rate": 1.468421052631579e-05,
+      "loss": 0.3656,
+      "step": 700
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.4668421052631579e-05,
+      "loss": 0.3494,
+      "step": 710
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.28125,
+      "learning_rate": 1.4652631578947367e-05,
+      "loss": 0.2928,
+      "step": 720
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.625,
+      "learning_rate": 1.4636842105263158e-05,
+      "loss": 0.391,
+      "step": 730
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.4621052631578946e-05,
+      "loss": 0.3042,
+      "step": 740
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.625,
+      "learning_rate": 1.4605263157894737e-05,
+      "loss": 0.2822,
+      "step": 750
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.4589473684210527e-05,
+      "loss": 0.3103,
+      "step": 760
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.9375,
+      "learning_rate": 1.4573684210526317e-05,
+      "loss": 0.2982,
+      "step": 770
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.4557894736842106e-05,
+      "loss": 0.2873,
+      "step": 780
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 11.375,
+      "learning_rate": 1.4542105263157895e-05,
+      "loss": 0.346,
+      "step": 790
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.71875,
+      "learning_rate": 1.4526315789473685e-05,
+      "loss": 0.3055,
+      "step": 800
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.96875,
+      "learning_rate": 1.4510526315789474e-05,
+      "loss": 0.2721,
+      "step": 810
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.4494736842105264e-05,
+      "loss": 0.2343,
+      "step": 820
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.90625,
+      "learning_rate": 1.4478947368421053e-05,
+      "loss": 0.2855,
+      "step": 830
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.875,
+      "learning_rate": 1.4463157894736843e-05,
+      "loss": 0.2685,
+      "step": 840
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.4447368421052632e-05,
+      "loss": 0.2675,
+      "step": 850
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.84375,
+      "learning_rate": 1.443157894736842e-05,
+      "loss": 0.3448,
+      "step": 860
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.5625,
+      "learning_rate": 1.441578947368421e-05,
+      "loss": 0.2843,
+      "step": 870
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.44e-05,
+      "loss": 0.3069,
+      "step": 880
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.75,
+      "learning_rate": 1.438421052631579e-05,
+      "loss": 0.2324,
+      "step": 890
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.8125,
+      "learning_rate": 1.4368421052631578e-05,
+      "loss": 0.3389,
+      "step": 900
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.96875,
+      "learning_rate": 1.4352631578947369e-05,
+      "loss": 0.3082,
+      "step": 910
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.4336842105263159e-05,
+      "loss": 0.2889,
+      "step": 920
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.109375,
+      "learning_rate": 1.4321052631578948e-05,
+      "loss": 0.2842,
+      "step": 930
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.96875,
+      "learning_rate": 1.4305263157894738e-05,
+      "loss": 0.2695,
+      "step": 940
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.875,
+      "learning_rate": 1.4289473684210527e-05,
+      "loss": 0.2021,
+      "step": 950
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.03125,
+      "learning_rate": 1.4273684210526317e-05,
+      "loss": 0.3382,
+      "step": 960
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.4257894736842106e-05,
+      "loss": 0.3721,
+      "step": 970
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.921875,
+      "learning_rate": 1.4242105263157896e-05,
+      "loss": 0.2715,
+      "step": 980
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.4226315789473685e-05,
+      "loss": 0.2261,
+      "step": 990
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.40625,
+      "learning_rate": 1.4210526315789473e-05,
+      "loss": 0.2718,
+      "step": 1000
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.4194736842105264e-05,
+      "loss": 0.355,
+      "step": 1010
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.0625,
+      "learning_rate": 1.4178947368421052e-05,
+      "loss": 0.3133,
+      "step": 1020
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.734375,
+      "learning_rate": 1.4163157894736843e-05,
+      "loss": 0.3427,
+      "step": 1030
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.5625,
+      "learning_rate": 1.4147368421052631e-05,
+      "loss": 0.2373,
+      "step": 1040
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.4131578947368422e-05,
+      "loss": 0.3505,
+      "step": 1050
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.411578947368421e-05,
+      "loss": 0.3211,
+      "step": 1060
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.984375,
+      "learning_rate": 1.4099999999999999e-05,
+      "loss": 0.3006,
+      "step": 1070
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.765625,
+      "learning_rate": 1.408421052631579e-05,
+      "loss": 0.31,
+      "step": 1080
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.406842105263158e-05,
+      "loss": 0.2641,
+      "step": 1090
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 9.375,
+      "learning_rate": 1.405263157894737e-05,
+      "loss": 0.3475,
+      "step": 1100
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.09375,
+      "learning_rate": 1.4036842105263158e-05,
+      "loss": 0.2827,
+      "step": 1110
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.46875,
+      "learning_rate": 1.4021052631578949e-05,
+      "loss": 0.2408,
+      "step": 1120
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.84375,
+      "learning_rate": 1.4005263157894737e-05,
+      "loss": 0.2428,
+      "step": 1130
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.375,
+      "learning_rate": 1.3989473684210526e-05,
+      "loss": 0.2533,
+      "step": 1140
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.84375,
+      "learning_rate": 1.3973684210526316e-05,
+      "loss": 0.3098,
+      "step": 1150
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.875,
+      "learning_rate": 1.3957894736842105e-05,
+      "loss": 0.3235,
+      "step": 1160
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.3942105263157895e-05,
+      "loss": 0.2959,
+      "step": 1170
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.125,
+      "learning_rate": 1.3926315789473684e-05,
+      "loss": 0.2873,
+      "step": 1180
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.515625,
+      "learning_rate": 1.3910526315789474e-05,
+      "loss": 0.3158,
+      "step": 1190
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.3894736842105263e-05,
+      "loss": 0.3004,
+      "step": 1200
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.34375,
+      "learning_rate": 1.3878947368421052e-05,
+      "loss": 0.3596,
+      "step": 1210
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.453125,
+      "learning_rate": 1.3863157894736842e-05,
+      "loss": 0.2345,
+      "step": 1220
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.384736842105263e-05,
+      "loss": 0.3332,
+      "step": 1230
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.921875,
+      "learning_rate": 1.3831578947368421e-05,
+      "loss": 0.2898,
+      "step": 1240
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.890625,
+      "learning_rate": 1.3815789473684211e-05,
+      "loss": 0.2854,
+      "step": 1250
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.3800000000000002e-05,
+      "loss": 0.2932,
+      "step": 1260
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.378421052631579e-05,
+      "loss": 0.3153,
+      "step": 1270
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.21875,
+      "learning_rate": 1.3768421052631579e-05,
+      "loss": 0.2947,
+      "step": 1280
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.234375,
+      "learning_rate": 1.375263157894737e-05,
+      "loss": 0.2211,
+      "step": 1290
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.15625,
+      "learning_rate": 1.3736842105263158e-05,
+      "loss": 0.296,
+      "step": 1300
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.21875,
+      "learning_rate": 1.3721052631578948e-05,
+      "loss": 0.3276,
+      "step": 1310
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.46875,
+      "learning_rate": 1.3705263157894737e-05,
+      "loss": 0.2716,
+      "step": 1320
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.3689473684210527e-05,
+      "loss": 0.2471,
+      "step": 1330
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.5,
+      "learning_rate": 1.3673684210526316e-05,
+      "loss": 0.3197,
+      "step": 1340
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.953125,
+      "learning_rate": 1.3657894736842106e-05,
+      "loss": 0.275,
+      "step": 1350
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.90625,
+      "learning_rate": 1.3642105263157895e-05,
+      "loss": 0.2567,
+      "step": 1360
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.890625,
+      "learning_rate": 1.3626315789473684e-05,
+      "loss": 0.2856,
+      "step": 1370
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 8.5,
+      "learning_rate": 1.3610526315789474e-05,
+      "loss": 0.3118,
+      "step": 1380
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.8125,
+      "learning_rate": 1.3594736842105263e-05,
+      "loss": 0.2669,
+      "step": 1390
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.375,
+      "learning_rate": 1.3578947368421053e-05,
+      "loss": 0.2718,
+      "step": 1400
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.90625,
+      "learning_rate": 1.3563157894736842e-05,
+      "loss": 0.3,
+      "step": 1410
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 7.59375,
+      "learning_rate": 1.3547368421052634e-05,
+      "loss": 0.2698,
+      "step": 1420
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.578125,
+      "learning_rate": 1.3531578947368422e-05,
+      "loss": 0.2467,
+      "step": 1430
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.3515789473684211e-05,
+      "loss": 0.2349,
+      "step": 1440
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 7.15625,
+      "learning_rate": 1.3500000000000001e-05,
+      "loss": 0.2565,
+      "step": 1450
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.375,
+      "learning_rate": 1.348421052631579e-05,
+      "loss": 0.2881,
+      "step": 1460
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.9375,
+      "learning_rate": 1.346842105263158e-05,
+      "loss": 0.2777,
+      "step": 1470
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.828125,
+      "learning_rate": 1.3452631578947369e-05,
+      "loss": 0.3411,
+      "step": 1480
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.343684210526316e-05,
+      "loss": 0.3102,
+      "step": 1490
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.859375,
+      "learning_rate": 1.3421052631578948e-05,
+      "loss": 0.2269,
+      "step": 1500
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.71875,
+      "learning_rate": 1.3405263157894736e-05,
+      "loss": 0.2404,
+      "step": 1510
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.21875,
+      "learning_rate": 1.3389473684210527e-05,
+      "loss": 0.2708,
+      "step": 1520
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.171875,
+      "learning_rate": 1.3373684210526315e-05,
+      "loss": 0.2771,
+      "step": 1530
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.34375,
+      "learning_rate": 1.3357894736842106e-05,
+      "loss": 0.3431,
+      "step": 1540
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.3342105263157894e-05,
+      "loss": 0.2593,
+      "step": 1550
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.375,
+      "learning_rate": 1.3326315789473685e-05,
+      "loss": 0.3764,
+      "step": 1560
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.5625,
+      "learning_rate": 1.3310526315789473e-05,
+      "loss": 0.2739,
+      "step": 1570
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.3294736842105262e-05,
+      "loss": 0.3306,
+      "step": 1580
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.546875,
+      "learning_rate": 1.3278947368421054e-05,
+      "loss": 0.2687,
+      "step": 1590
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.5,
+      "learning_rate": 1.3263157894736843e-05,
+      "loss": 0.288,
+      "step": 1600
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.390625,
+      "learning_rate": 1.3247368421052633e-05,
+      "loss": 0.2626,
+      "step": 1610
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.0,
+      "learning_rate": 1.3231578947368422e-05,
+      "loss": 0.2635,
+      "step": 1620
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.3215789473684212e-05,
+      "loss": 0.2567,
+      "step": 1630
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.32e-05,
+      "loss": 0.2756,
+      "step": 1640
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.21875,
+      "learning_rate": 1.318421052631579e-05,
+      "loss": 0.32,
+      "step": 1650
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.65625,
+      "learning_rate": 1.316842105263158e-05,
+      "loss": 0.2499,
+      "step": 1660
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.125,
+      "learning_rate": 1.3152631578947368e-05,
+      "loss": 0.2084,
+      "step": 1670
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.3136842105263159e-05,
+      "loss": 0.2263,
+      "step": 1680
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.71875,
+      "learning_rate": 1.3121052631578947e-05,
+      "loss": 0.3104,
+      "step": 1690
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 7.21875,
+      "learning_rate": 1.3105263157894738e-05,
+      "loss": 0.3948,
+      "step": 1700
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.921875,
+      "learning_rate": 1.3089473684210526e-05,
+      "loss": 0.2561,
+      "step": 1710
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.5,
+      "learning_rate": 1.3073684210526315e-05,
+      "loss": 0.2774,
+      "step": 1720
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.328125,
+      "learning_rate": 1.3057894736842105e-05,
+      "loss": 0.2165,
+      "step": 1730
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.046875,
+      "learning_rate": 1.3042105263157894e-05,
+      "loss": 0.2819,
+      "step": 1740
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.3026315789473684e-05,
+      "loss": 0.2111,
+      "step": 1750
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.21875,
+      "learning_rate": 1.3010526315789475e-05,
+      "loss": 0.2293,
+      "step": 1760
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.53125,
+      "learning_rate": 1.2994736842105265e-05,
+      "loss": 0.2363,
+      "step": 1770
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.296875,
+      "learning_rate": 1.2978947368421054e-05,
+      "loss": 0.3236,
+      "step": 1780
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.2963157894736842e-05,
+      "loss": 0.2785,
+      "step": 1790
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.265625,
+      "learning_rate": 1.2947368421052633e-05,
+      "loss": 0.238,
+      "step": 1800
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.828125,
+      "learning_rate": 1.2931578947368421e-05,
+      "loss": 0.2736,
+      "step": 1810
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.65625,
+      "learning_rate": 1.2915789473684212e-05,
+      "loss": 0.3267,
+      "step": 1820
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.29e-05,
+      "loss": 0.2644,
+      "step": 1830
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.875,
+      "learning_rate": 1.288421052631579e-05,
+      "loss": 0.2847,
+      "step": 1840
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.5625,
+      "learning_rate": 1.2868421052631579e-05,
+      "loss": 0.3246,
+      "step": 1850
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.328125,
+      "learning_rate": 1.2852631578947368e-05,
+      "loss": 0.2835,
+      "step": 1860
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.9375,
+      "learning_rate": 1.2836842105263158e-05,
+      "loss": 0.2773,
+      "step": 1870
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.2821052631578947e-05,
+      "loss": 0.2967,
+      "step": 1880
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.5625,
+      "learning_rate": 1.2805263157894737e-05,
+      "loss": 0.3291,
+      "step": 1890
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.875,
+      "learning_rate": 1.2789473684210526e-05,
+      "loss": 0.2619,
+      "step": 1900
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.15625,
+      "learning_rate": 1.2773684210526316e-05,
+      "loss": 0.2467,
+      "step": 1910
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.25,
+      "learning_rate": 1.2757894736842106e-05,
+      "loss": 0.224,
+      "step": 1920
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 7.09375,
+      "learning_rate": 1.2742105263157895e-05,
+      "loss": 0.2697,
+      "step": 1930
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.2726315789473685e-05,
+      "loss": 0.245,
+      "step": 1940
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.65625,
+      "learning_rate": 1.2710526315789474e-05,
+      "loss": 0.22,
+      "step": 1950
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.2694736842105264e-05,
+      "loss": 0.3149,
+      "step": 1960
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.15625,
+      "learning_rate": 1.2678947368421053e-05,
+      "loss": 0.2345,
+      "step": 1970
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.0,
+      "learning_rate": 1.2663157894736843e-05,
+      "loss": 0.2053,
+      "step": 1980
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.84375,
+      "learning_rate": 1.2647368421052632e-05,
+      "loss": 0.2364,
+      "step": 1990
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.0,
+      "learning_rate": 1.263157894736842e-05,
+      "loss": 0.3397,
+      "step": 2000
+    },
+    {
+      "epoch": 0.1,
+      "eval_loss": 0.28045549988746643,
+      "eval_runtime": 64.869,
+      "eval_samples_per_second": 15.416,
+      "eval_steps_per_second": 15.416,
+      "step": 2000
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.9375,
+      "learning_rate": 1.2615789473684211e-05,
+      "loss": 0.2323,
+      "step": 2010
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.625,
+      "learning_rate": 1.26e-05,
+      "loss": 0.2569,
+      "step": 2020
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.28125,
+      "learning_rate": 1.258421052631579e-05,
+      "loss": 0.2124,
+      "step": 2030
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.296875,
+      "learning_rate": 1.2568421052631579e-05,
+      "loss": 0.2332,
+      "step": 2040
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.78125,
+      "learning_rate": 1.2552631578947369e-05,
+      "loss": 0.3277,
+      "step": 2050
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.34375,
+      "learning_rate": 1.2536842105263158e-05,
+      "loss": 0.2017,
+      "step": 2060
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.59375,
+      "learning_rate": 1.2521052631578946e-05,
+      "loss": 0.2663,
+      "step": 2070
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.53125,
+      "learning_rate": 1.2505263157894737e-05,
+      "loss": 0.2702,
+      "step": 2080
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.875,
+      "learning_rate": 1.2489473684210527e-05,
+      "loss": 0.2721,
+      "step": 2090
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.2473684210526317e-05,
+      "loss": 0.2631,
+      "step": 2100
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.78125,
+      "learning_rate": 1.2457894736842106e-05,
+      "loss": 0.2694,
+      "step": 2110
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.609375,
+      "learning_rate": 1.2442105263157896e-05,
+      "loss": 0.2236,
+      "step": 2120
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.2426315789473685e-05,
+      "loss": 0.2442,
+      "step": 2130
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.2410526315789474e-05,
+      "loss": 0.2605,
+      "step": 2140
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.2394736842105264e-05,
+      "loss": 0.2723,
+      "step": 2150
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.125,
+      "learning_rate": 1.2378947368421053e-05,
+      "loss": 0.2138,
+      "step": 2160
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.2363157894736843e-05,
+      "loss": 0.2413,
+      "step": 2170
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.28125,
+      "learning_rate": 1.2347368421052631e-05,
+      "loss": 0.2137,
+      "step": 2180
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.96875,
+      "learning_rate": 1.2331578947368422e-05,
+      "loss": 0.318,
+      "step": 2190
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.40625,
+      "learning_rate": 1.231578947368421e-05,
+      "loss": 0.267,
+      "step": 2200
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.625,
+      "learning_rate": 1.2299999999999999e-05,
+      "loss": 0.2593,
+      "step": 2210
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.078125,
+      "learning_rate": 1.228421052631579e-05,
+      "loss": 0.2626,
+      "step": 2220
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.015625,
+      "learning_rate": 1.2268421052631578e-05,
+      "loss": 0.3095,
+      "step": 2230
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 7.03125,
+      "learning_rate": 1.2252631578947368e-05,
+      "loss": 0.2364,
+      "step": 2240
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.84375,
+      "learning_rate": 1.2236842105263159e-05,
+      "loss": 0.3079,
+      "step": 2250
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.2221052631578949e-05,
+      "loss": 0.2233,
+      "step": 2260
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.796875,
+      "learning_rate": 1.2205263157894738e-05,
+      "loss": 0.3072,
+      "step": 2270
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.59375,
+      "learning_rate": 1.2189473684210526e-05,
+      "loss": 0.2579,
+      "step": 2280
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.890625,
+      "learning_rate": 1.2173684210526317e-05,
+      "loss": 0.2265,
+      "step": 2290
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.875,
+      "learning_rate": 1.2157894736842105e-05,
+      "loss": 0.2932,
+      "step": 2300
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.2142105263157896e-05,
+      "loss": 0.3118,
+      "step": 2310
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.34375,
+      "learning_rate": 1.2126315789473684e-05,
+      "loss": 0.3196,
+      "step": 2320
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.59375,
+      "learning_rate": 1.2110526315789475e-05,
+      "loss": 0.3374,
+      "step": 2330
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.75,
+      "learning_rate": 1.2094736842105263e-05,
+      "loss": 0.2429,
+      "step": 2340
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.5625,
+      "learning_rate": 1.2078947368421052e-05,
+      "loss": 0.2552,
+      "step": 2350
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.2063157894736842e-05,
+      "loss": 0.25,
+      "step": 2360
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.96875,
+      "learning_rate": 1.2047368421052631e-05,
+      "loss": 0.2549,
+      "step": 2370
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.2031578947368421e-05,
+      "loss": 0.2079,
+      "step": 2380
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.96875,
+      "learning_rate": 1.201578947368421e-05,
+      "loss": 0.269,
+      "step": 2390
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 7.28125,
+      "learning_rate": 1.2e-05,
+      "loss": 0.2658,
+      "step": 2400
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.0625,
+      "learning_rate": 1.1984210526315789e-05,
+      "loss": 0.3198,
+      "step": 2410
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.196842105263158e-05,
+      "loss": 0.2094,
+      "step": 2420
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.84375,
+      "learning_rate": 1.195263157894737e-05,
+      "loss": 0.2877,
+      "step": 2430
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.15625,
+      "learning_rate": 1.1936842105263158e-05,
+      "loss": 0.2458,
+      "step": 2440
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.1875,
+      "learning_rate": 1.1921052631578949e-05,
+      "loss": 0.2186,
+      "step": 2450
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.71875,
+      "learning_rate": 1.1905263157894737e-05,
+      "loss": 0.2657,
+      "step": 2460
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.1889473684210528e-05,
+      "loss": 0.2838,
+      "step": 2470
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.0,
+      "learning_rate": 1.1873684210526316e-05,
+      "loss": 0.283,
+      "step": 2480
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.34375,
+      "learning_rate": 1.1857894736842105e-05,
+      "loss": 0.2818,
+      "step": 2490
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.0,
+      "learning_rate": 1.1842105263157895e-05,
+      "loss": 0.2842,
+      "step": 2500
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.921875,
+      "learning_rate": 1.1826315789473684e-05,
+      "loss": 0.2352,
+      "step": 2510
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 8.25,
+      "learning_rate": 1.1810526315789474e-05,
+      "loss": 0.2655,
+      "step": 2520
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.53125,
+      "learning_rate": 1.1794736842105263e-05,
+      "loss": 0.2457,
+      "step": 2530
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.96875,
+      "learning_rate": 1.1778947368421053e-05,
+      "loss": 0.2637,
+      "step": 2540
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.1763157894736842e-05,
+      "loss": 0.2808,
+      "step": 2550
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.174736842105263e-05,
+      "loss": 0.2925,
+      "step": 2560
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.6875,
+      "learning_rate": 1.173157894736842e-05,
+      "loss": 0.2698,
+      "step": 2570
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.171578947368421e-05,
+      "loss": 0.2744,
+      "step": 2580
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.546875,
+      "learning_rate": 1.1700000000000001e-05,
+      "loss": 0.2291,
+      "step": 2590
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.15625,
+      "learning_rate": 1.168421052631579e-05,
+      "loss": 0.2584,
+      "step": 2600
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.40625,
+      "learning_rate": 1.166842105263158e-05,
+      "loss": 0.197,
+      "step": 2610
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 7.96875,
+      "learning_rate": 1.1652631578947369e-05,
+      "loss": 0.2722,
+      "step": 2620
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.15625,
+      "learning_rate": 1.1636842105263158e-05,
+      "loss": 0.2192,
+      "step": 2630
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.375,
+      "learning_rate": 1.1621052631578948e-05,
+      "loss": 0.1978,
+      "step": 2640
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.78125,
+      "learning_rate": 1.1605263157894737e-05,
+      "loss": 0.2161,
+      "step": 2650
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.1589473684210527e-05,
+      "loss": 0.2509,
+      "step": 2660
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.546875,
+      "learning_rate": 1.1573684210526316e-05,
+      "loss": 0.2299,
+      "step": 2670
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.1557894736842106e-05,
+      "loss": 0.286,
+      "step": 2680
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.296875,
+      "learning_rate": 1.1542105263157895e-05,
+      "loss": 0.2221,
+      "step": 2690
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.1526315789473683e-05,
+      "loss": 0.2128,
+      "step": 2700
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.75,
+      "learning_rate": 1.1510526315789474e-05,
+      "loss": 0.2578,
+      "step": 2710
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.796875,
+      "learning_rate": 1.1494736842105262e-05,
+      "loss": 0.3137,
+      "step": 2720
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 1.8515625,
+      "learning_rate": 1.1478947368421053e-05,
+      "loss": 0.2952,
+      "step": 2730
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.34375,
+      "learning_rate": 1.1463157894736841e-05,
+      "loss": 0.2424,
+      "step": 2740
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 7.03125,
+      "learning_rate": 1.1447368421052632e-05,
+      "loss": 0.1941,
+      "step": 2750
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.1431578947368422e-05,
+      "loss": 0.2441,
+      "step": 2760
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.71875,
+      "learning_rate": 1.141578947368421e-05,
+      "loss": 0.2584,
+      "step": 2770
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.03125,
+      "learning_rate": 1.1400000000000001e-05,
+      "loss": 0.2769,
+      "step": 2780
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.875,
+      "learning_rate": 1.138421052631579e-05,
+      "loss": 0.2841,
+      "step": 2790
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 8.4375,
+      "learning_rate": 1.136842105263158e-05,
+      "loss": 0.2043,
+      "step": 2800
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.1352631578947369e-05,
+      "loss": 0.2642,
+      "step": 2810
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.90625,
+      "learning_rate": 1.1336842105263159e-05,
+      "loss": 0.1911,
+      "step": 2820
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.1321052631578948e-05,
+      "loss": 0.3403,
+      "step": 2830
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.0,
+      "learning_rate": 1.1305263157894736e-05,
+      "loss": 0.224,
+      "step": 2840
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.25,
+      "learning_rate": 1.1289473684210527e-05,
+      "loss": 0.2674,
+      "step": 2850
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.1273684210526315e-05,
+      "loss": 0.2787,
+      "step": 2860
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.8125,
+      "learning_rate": 1.1257894736842106e-05,
+      "loss": 0.2718,
+      "step": 2870
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.09375,
+      "learning_rate": 1.1242105263157894e-05,
+      "loss": 0.2747,
+      "step": 2880
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.875,
+      "learning_rate": 1.1226315789473685e-05,
+      "loss": 0.2384,
+      "step": 2890
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 7.53125,
+      "learning_rate": 1.1210526315789473e-05,
+      "loss": 0.2541,
+      "step": 2900
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.1194736842105264e-05,
+      "loss": 0.2794,
+      "step": 2910
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.0625,
+      "learning_rate": 1.1178947368421054e-05,
+      "loss": 0.2691,
+      "step": 2920
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.34375,
+      "learning_rate": 1.1163157894736842e-05,
+      "loss": 0.2865,
+      "step": 2930
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.40625,
+      "learning_rate": 1.1147368421052633e-05,
+      "loss": 0.2395,
+      "step": 2940
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.921875,
+      "learning_rate": 1.1131578947368421e-05,
+      "loss": 0.2228,
+      "step": 2950
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.375,
+      "learning_rate": 1.1115789473684212e-05,
+      "loss": 0.2599,
+      "step": 2960
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.421875,
+      "learning_rate": 1.11e-05,
+      "loss": 0.2064,
+      "step": 2970
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.34375,
+      "learning_rate": 1.108421052631579e-05,
+      "loss": 0.2317,
+      "step": 2980
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.4375,
+      "learning_rate": 1.106842105263158e-05,
+      "loss": 0.2188,
+      "step": 2990
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.15625,
+      "learning_rate": 1.1052631578947368e-05,
+      "loss": 0.2675,
+      "step": 3000
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.71875,
+      "learning_rate": 1.1036842105263158e-05,
+      "loss": 0.1853,
+      "step": 3010
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.75,
+      "learning_rate": 1.1021052631578947e-05,
+      "loss": 0.2659,
+      "step": 3020
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.21875,
+      "learning_rate": 1.1005263157894737e-05,
+      "loss": 0.296,
+      "step": 3030
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.75,
+      "learning_rate": 1.0989473684210526e-05,
+      "loss": 0.2744,
+      "step": 3040
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.75,
+      "learning_rate": 1.0973684210526316e-05,
+      "loss": 0.2558,
+      "step": 3050
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.0957894736842105e-05,
+      "loss": 0.3046,
+      "step": 3060
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.0942105263157894e-05,
+      "loss": 0.2861,
+      "step": 3070
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.0926315789473684e-05,
+      "loss": 0.237,
+      "step": 3080
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.90625,
+      "learning_rate": 1.0910526315789474e-05,
+      "loss": 0.2734,
+      "step": 3090
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.84375,
+      "learning_rate": 1.0894736842105265e-05,
+      "loss": 0.3093,
+      "step": 3100
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.25,
+      "learning_rate": 1.0878947368421053e-05,
+      "loss": 0.2508,
+      "step": 3110
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.0863157894736844e-05,
+      "loss": 0.264,
+      "step": 3120
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.03125,
+      "learning_rate": 1.0847368421052632e-05,
+      "loss": 0.1805,
+      "step": 3130
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.15625,
+      "learning_rate": 1.0831578947368421e-05,
+      "loss": 0.2598,
+      "step": 3140
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.25,
+      "learning_rate": 1.0815789473684211e-05,
+      "loss": 0.2245,
+      "step": 3150
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.08e-05,
+      "loss": 0.2477,
+      "step": 3160
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.8125,
+      "learning_rate": 1.078421052631579e-05,
+      "loss": 0.309,
+      "step": 3170
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.34375,
+      "learning_rate": 1.0768421052631579e-05,
+      "loss": 0.2718,
+      "step": 3180
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.46875,
+      "learning_rate": 1.075263157894737e-05,
+      "loss": 0.2231,
+      "step": 3190
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.375,
+      "learning_rate": 1.0736842105263158e-05,
+      "loss": 0.2859,
+      "step": 3200
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.9375,
+      "learning_rate": 1.0721052631578947e-05,
+      "loss": 0.276,
+      "step": 3210
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.0705263157894737e-05,
+      "loss": 0.2616,
+      "step": 3220
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.609375,
+      "learning_rate": 1.0689473684210526e-05,
+      "loss": 0.2865,
+      "step": 3230
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.0673684210526316e-05,
+      "loss": 0.2276,
+      "step": 3240
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.8125,
+      "learning_rate": 1.0657894736842106e-05,
+      "loss": 0.2579,
+      "step": 3250
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.5,
+      "learning_rate": 1.0642105263157897e-05,
+      "loss": 0.3116,
+      "step": 3260
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.546875,
+      "learning_rate": 1.0626315789473685e-05,
+      "loss": 0.2143,
+      "step": 3270
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.0610526315789474e-05,
+      "loss": 0.2593,
+      "step": 3280
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.0625,
+      "learning_rate": 1.0594736842105264e-05,
+      "loss": 0.232,
+      "step": 3290
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.71875,
+      "learning_rate": 1.0578947368421053e-05,
+      "loss": 0.229,
+      "step": 3300
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.921875,
+      "learning_rate": 1.0563157894736843e-05,
+      "loss": 0.2862,
+      "step": 3310
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.140625,
+      "learning_rate": 1.0547368421052632e-05,
+      "loss": 0.3062,
+      "step": 3320
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.625,
+      "learning_rate": 1.0531578947368422e-05,
+      "loss": 0.2486,
+      "step": 3330
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.90625,
+      "learning_rate": 1.051578947368421e-05,
+      "loss": 0.3082,
+      "step": 3340
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.15625,
+      "learning_rate": 1.05e-05,
+      "loss": 0.2006,
+      "step": 3350
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.048421052631579e-05,
+      "loss": 0.246,
+      "step": 3360
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.75,
+      "learning_rate": 1.0468421052631578e-05,
+      "loss": 0.201,
+      "step": 3370
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.84375,
+      "learning_rate": 1.0452631578947369e-05,
+      "loss": 0.1955,
+      "step": 3380
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.375,
+      "learning_rate": 1.0436842105263157e-05,
+      "loss": 0.243,
+      "step": 3390
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.53125,
+      "learning_rate": 1.0421052631578948e-05,
+      "loss": 0.2789,
+      "step": 3400
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.625,
+      "learning_rate": 1.0405263157894736e-05,
+      "loss": 0.1762,
+      "step": 3410
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.84375,
+      "learning_rate": 1.0389473684210527e-05,
+      "loss": 0.214,
+      "step": 3420
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.78125,
+      "learning_rate": 1.0373684210526317e-05,
+      "loss": 0.2191,
+      "step": 3430
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.8125,
+      "learning_rate": 1.0357894736842106e-05,
+      "loss": 0.292,
+      "step": 3440
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.875,
+      "learning_rate": 1.0342105263157896e-05,
+      "loss": 0.2538,
+      "step": 3450
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.328125,
+      "learning_rate": 1.0326315789473685e-05,
+      "loss": 0.258,
+      "step": 3460
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.375,
+      "learning_rate": 1.0310526315789475e-05,
+      "loss": 0.2307,
+      "step": 3470
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.0294736842105264e-05,
+      "loss": 0.2486,
+      "step": 3480
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.8828125,
+      "learning_rate": 1.0278947368421052e-05,
+      "loss": 0.143,
+      "step": 3490
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.0,
+      "learning_rate": 1.0263157894736843e-05,
+      "loss": 0.2726,
+      "step": 3500
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.0247368421052631e-05,
+      "loss": 0.2666,
+      "step": 3510
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.59375,
+      "learning_rate": 1.0231578947368422e-05,
+      "loss": 0.2403,
+      "step": 3520
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 8.875,
+      "learning_rate": 1.021578947368421e-05,
+      "loss": 0.3262,
+      "step": 3530
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.9375,
+      "learning_rate": 1.02e-05,
+      "loss": 0.2611,
+      "step": 3540
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.84375,
+      "learning_rate": 1.018421052631579e-05,
+      "loss": 0.2457,
+      "step": 3550
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.0168421052631578e-05,
+      "loss": 0.2983,
+      "step": 3560
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.984375,
+      "learning_rate": 1.0152631578947368e-05,
+      "loss": 0.2505,
+      "step": 3570
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.0136842105263157e-05,
+      "loss": 0.2786,
+      "step": 3580
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.25,
+      "learning_rate": 1.0121052631578949e-05,
+      "loss": 0.3497,
+      "step": 3590
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.5,
+      "learning_rate": 1.0105263157894738e-05,
+      "loss": 0.2734,
+      "step": 3600
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.25,
+      "learning_rate": 1.0089473684210528e-05,
+      "loss": 0.2685,
+      "step": 3610
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 8.3125,
+      "learning_rate": 1.0073684210526317e-05,
+      "loss": 0.3044,
+      "step": 3620
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.40625,
+      "learning_rate": 1.0057894736842105e-05,
+      "loss": 0.2584,
+      "step": 3630
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.46875,
+      "learning_rate": 1.0042105263157896e-05,
+      "loss": 0.2911,
+      "step": 3640
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.859375,
+      "learning_rate": 1.0026315789473684e-05,
+      "loss": 0.2277,
+      "step": 3650
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.0010526315789474e-05,
+      "loss": 0.2325,
+      "step": 3660
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.21875,
+      "learning_rate": 9.994736842105263e-06,
+      "loss": 0.296,
+      "step": 3670
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.59375,
+      "learning_rate": 9.978947368421053e-06,
+      "loss": 0.2246,
+      "step": 3680
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.796875,
+      "learning_rate": 9.963157894736842e-06,
+      "loss": 0.2403,
+      "step": 3690
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.4375,
+      "learning_rate": 9.94736842105263e-06,
+      "loss": 0.2477,
+      "step": 3700
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.6875,
+      "learning_rate": 9.931578947368421e-06,
+      "loss": 0.2639,
+      "step": 3710
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.6875,
+      "learning_rate": 9.91578947368421e-06,
+      "loss": 0.2638,
+      "step": 3720
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.5625,
+      "learning_rate": 9.9e-06,
+      "loss": 0.2519,
+      "step": 3730
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 3.34375,
+      "learning_rate": 9.884210526315789e-06,
+      "loss": 0.2468,
+      "step": 3740
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 3.453125,
+      "learning_rate": 9.868421052631579e-06,
+      "loss": 0.2896,
+      "step": 3750
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 3.796875,
+      "learning_rate": 9.85263157894737e-06,
+      "loss": 0.2912,
+      "step": 3760
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.625,
+      "learning_rate": 9.836842105263158e-06,
+      "loss": 0.3592,
+      "step": 3770
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 3.375,
+      "learning_rate": 9.821052631578948e-06,
+      "loss": 0.289,
+      "step": 3780
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.0625,
+      "learning_rate": 9.805263157894737e-06,
+      "loss": 0.168,
+      "step": 3790
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.96875,
+      "learning_rate": 9.789473684210527e-06,
+      "loss": 0.2772,
+      "step": 3800
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.625,
+      "learning_rate": 9.773684210526316e-06,
+      "loss": 0.2263,
+      "step": 3810
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.5,
+      "learning_rate": 9.757894736842106e-06,
+      "loss": 0.3336,
+      "step": 3820
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.0625,
+      "learning_rate": 9.742105263157895e-06,
+      "loss": 0.1924,
+      "step": 3830
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.15625,
+      "learning_rate": 9.726315789473684e-06,
+      "loss": 0.2615,
+      "step": 3840
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.28125,
+      "learning_rate": 9.710526315789474e-06,
+      "loss": 0.2536,
+      "step": 3850
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 3.890625,
+      "learning_rate": 9.694736842105263e-06,
+      "loss": 0.2672,
+      "step": 3860
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.03125,
+      "learning_rate": 9.678947368421053e-06,
+      "loss": 0.2703,
+      "step": 3870
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.9375,
+      "learning_rate": 9.663157894736842e-06,
+      "loss": 0.2596,
+      "step": 3880
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.96875,
+      "learning_rate": 9.647368421052632e-06,
+      "loss": 0.2443,
+      "step": 3890
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.96875,
+      "learning_rate": 9.63157894736842e-06,
+      "loss": 0.2118,
+      "step": 3900
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.96875,
+      "learning_rate": 9.61578947368421e-06,
+      "loss": 0.2678,
+      "step": 3910
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.59375,
+      "learning_rate": 9.600000000000001e-06,
+      "loss": 0.2817,
+      "step": 3920
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.59375,
+      "learning_rate": 9.58421052631579e-06,
+      "loss": 0.225,
+      "step": 3930
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.1875,
+      "learning_rate": 9.56842105263158e-06,
+      "loss": 0.2452,
+      "step": 3940
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.84375,
+      "learning_rate": 9.552631578947369e-06,
+      "loss": 0.1778,
+      "step": 3950
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.96875,
+      "learning_rate": 9.53684210526316e-06,
+      "loss": 0.312,
+      "step": 3960
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.5625,
+      "learning_rate": 9.521052631578948e-06,
+      "loss": 0.2525,
+      "step": 3970
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.03125,
+      "learning_rate": 9.505263157894737e-06,
+      "loss": 0.2373,
+      "step": 3980
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.8125,
+      "learning_rate": 9.489473684210527e-06,
+      "loss": 0.2765,
+      "step": 3990
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.171875,
+      "learning_rate": 9.473684210526315e-06,
+      "loss": 0.2565,
+      "step": 4000
+    },
+    {
+      "epoch": 0.2,
+      "eval_loss": 0.2541459798812866,
+      "eval_runtime": 64.6611,
+      "eval_samples_per_second": 15.465,
+      "eval_steps_per_second": 15.465,
+      "step": 4000
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.3125,
+      "learning_rate": 9.457894736842106e-06,
+      "loss": 0.258,
+      "step": 4010
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.90625,
+      "learning_rate": 9.442105263157894e-06,
+      "loss": 0.254,
+      "step": 4020
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.234375,
+      "learning_rate": 9.426315789473685e-06,
+      "loss": 0.2508,
+      "step": 4030
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.25,
+      "learning_rate": 9.410526315789473e-06,
+      "loss": 0.2337,
+      "step": 4040
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.0625,
+      "learning_rate": 9.394736842105262e-06,
+      "loss": 0.2471,
+      "step": 4050
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.3125,
+      "learning_rate": 9.378947368421052e-06,
+      "loss": 0.268,
+      "step": 4060
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.625,
+      "learning_rate": 9.363157894736841e-06,
+      "loss": 0.28,
+      "step": 4070
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.46875,
+      "learning_rate": 9.347368421052631e-06,
+      "loss": 0.3035,
+      "step": 4080
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.34375,
+      "learning_rate": 9.331578947368422e-06,
+      "loss": 0.2386,
+      "step": 4090
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.71875,
+      "learning_rate": 9.315789473684212e-06,
+      "loss": 0.2449,
+      "step": 4100
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.5,
+      "learning_rate": 9.3e-06,
+      "loss": 0.2623,
+      "step": 4110
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.28125,
+      "learning_rate": 9.28421052631579e-06,
+      "loss": 0.2727,
+      "step": 4120
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.09375,
+      "learning_rate": 9.26842105263158e-06,
+      "loss": 0.2507,
+      "step": 4130
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.34375,
+      "learning_rate": 9.252631578947368e-06,
+      "loss": 0.3181,
+      "step": 4140
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.8125,
+      "learning_rate": 9.236842105263159e-06,
+      "loss": 0.2642,
+      "step": 4150
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 2.71875,
+      "learning_rate": 9.221052631578947e-06,
+      "loss": 0.2584,
+      "step": 4160
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.546875,
+      "learning_rate": 9.205263157894738e-06,
+      "loss": 0.2338,
+      "step": 4170
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.5,
+      "learning_rate": 9.189473684210526e-06,
+      "loss": 0.297,
+      "step": 4180
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.71875,
+      "learning_rate": 9.173684210526315e-06,
+      "loss": 0.3134,
+      "step": 4190
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 8.6875,
+      "learning_rate": 9.157894736842105e-06,
+      "loss": 0.2647,
+      "step": 4200
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 2.90625,
+      "learning_rate": 9.142105263157894e-06,
+      "loss": 0.301,
+      "step": 4210
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.4375,
+      "learning_rate": 9.126315789473684e-06,
+      "loss": 0.2248,
+      "step": 4220
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.390625,
+      "learning_rate": 9.110526315789473e-06,
+      "loss": 0.2932,
+      "step": 4230
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.640625,
+      "learning_rate": 9.094736842105263e-06,
+      "loss": 0.2674,
+      "step": 4240
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 2.25,
+      "learning_rate": 9.078947368421054e-06,
+      "loss": 0.263,
+      "step": 4250
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.78125,
+      "learning_rate": 9.063157894736842e-06,
+      "loss": 0.2053,
+      "step": 4260
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.90625,
+      "learning_rate": 9.047368421052633e-06,
+      "loss": 0.2831,
+      "step": 4270
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.5625,
+      "learning_rate": 9.031578947368421e-06,
+      "loss": 0.2645,
+      "step": 4280
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 10.0625,
+      "learning_rate": 9.015789473684212e-06,
+      "loss": 0.2411,
+      "step": 4290
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.5625,
+      "learning_rate": 9e-06,
+      "loss": 0.3289,
+      "step": 4300
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 3.671875,
+      "learning_rate": 8.98421052631579e-06,
+      "loss": 0.2813,
+      "step": 4310
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.0625,
+      "learning_rate": 8.96842105263158e-06,
+      "loss": 0.2614,
+      "step": 4320
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 3.4375,
+      "learning_rate": 8.952631578947368e-06,
+      "loss": 0.166,
+      "step": 4330
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.15625,
+      "learning_rate": 8.936842105263158e-06,
+      "loss": 0.2643,
+      "step": 4340
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 6.0625,
+      "learning_rate": 8.921052631578947e-06,
+      "loss": 0.3431,
+      "step": 4350
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.46875,
+      "learning_rate": 8.905263157894737e-06,
+      "loss": 0.2632,
+      "step": 4360
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.296875,
+      "learning_rate": 8.889473684210526e-06,
+      "loss": 0.2012,
+      "step": 4370
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.5,
+      "learning_rate": 8.873684210526316e-06,
+      "loss": 0.274,
+      "step": 4380
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.5,
+      "learning_rate": 8.857894736842105e-06,
+      "loss": 0.2305,
+      "step": 4390
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.34375,
+      "learning_rate": 8.842105263157893e-06,
+      "loss": 0.2416,
+      "step": 4400
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.28125,
+      "learning_rate": 8.826315789473684e-06,
+      "loss": 0.2294,
+      "step": 4410
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.5,
+      "learning_rate": 8.810526315789474e-06,
+      "loss": 0.3028,
+      "step": 4420
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.75,
+      "learning_rate": 8.794736842105264e-06,
+      "loss": 0.2739,
+      "step": 4430
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.65625,
+      "learning_rate": 8.778947368421053e-06,
+      "loss": 0.2835,
+      "step": 4440
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.75,
+      "learning_rate": 8.763157894736843e-06,
+      "loss": 0.2755,
+      "step": 4450
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.0625,
+      "learning_rate": 8.747368421052632e-06,
+      "loss": 0.1984,
+      "step": 4460
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.625,
+      "learning_rate": 8.731578947368422e-06,
+      "loss": 0.2497,
+      "step": 4470
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 6.3125,
+      "learning_rate": 8.715789473684211e-06,
+      "loss": 0.2464,
+      "step": 4480
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.28125,
+      "learning_rate": 8.7e-06,
+      "loss": 0.2134,
+      "step": 4490
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 2.046875,
+      "learning_rate": 8.68421052631579e-06,
+      "loss": 0.2548,
+      "step": 4500
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.6875,
+      "learning_rate": 8.668421052631579e-06,
+      "loss": 0.2385,
+      "step": 4510
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.53125,
+      "learning_rate": 8.652631578947369e-06,
+      "loss": 0.2266,
+      "step": 4520
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.1875,
+      "learning_rate": 8.636842105263158e-06,
+      "loss": 0.2529,
+      "step": 4530
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.6875,
+      "learning_rate": 8.621052631578948e-06,
+      "loss": 0.2372,
+      "step": 4540
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.390625,
+      "learning_rate": 8.605263157894737e-06,
+      "loss": 0.2801,
+      "step": 4550
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.5625,
+      "learning_rate": 8.589473684210525e-06,
+      "loss": 0.2418,
+      "step": 4560
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.5625,
+      "learning_rate": 8.573684210526316e-06,
+      "loss": 0.2292,
+      "step": 4570
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.703125,
+      "learning_rate": 8.557894736842104e-06,
+      "loss": 0.2036,
+      "step": 4580
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.6875,
+      "learning_rate": 8.542105263157896e-06,
+      "loss": 0.206,
+      "step": 4590
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.375,
+      "learning_rate": 8.526315789473685e-06,
+      "loss": 0.291,
+      "step": 4600
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.609375,
+      "learning_rate": 8.510526315789475e-06,
+      "loss": 0.2777,
+      "step": 4610
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.25,
+      "learning_rate": 8.494736842105264e-06,
+      "loss": 0.333,
+      "step": 4620
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.03125,
+      "learning_rate": 8.478947368421053e-06,
+      "loss": 0.277,
+      "step": 4630
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.90625,
+      "learning_rate": 8.463157894736843e-06,
+      "loss": 0.1894,
+      "step": 4640
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.8125,
+      "learning_rate": 8.447368421052632e-06,
+      "loss": 0.2772,
+      "step": 4650
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 2.21875,
+      "learning_rate": 8.431578947368422e-06,
+      "loss": 0.2534,
+      "step": 4660
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.9375,
+      "learning_rate": 8.41578947368421e-06,
+      "loss": 0.1995,
+      "step": 4670
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 8.3125,
+      "learning_rate": 8.400000000000001e-06,
+      "loss": 0.3163,
+      "step": 4680
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.625,
+      "learning_rate": 8.38421052631579e-06,
+      "loss": 0.2608,
+      "step": 4690
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.3125,
+      "learning_rate": 8.368421052631578e-06,
+      "loss": 0.2886,
+      "step": 4700
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.8125,
+      "learning_rate": 8.352631578947369e-06,
+      "loss": 0.2614,
+      "step": 4710
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.03125,
+      "learning_rate": 8.336842105263157e-06,
+      "loss": 0.2443,
+      "step": 4720
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.15625,
+      "learning_rate": 8.321052631578947e-06,
+      "loss": 0.2985,
+      "step": 4730
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.78125,
+      "learning_rate": 8.305263157894736e-06,
+      "loss": 0.2702,
+      "step": 4740
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.375,
+      "learning_rate": 8.289473684210526e-06,
+      "loss": 0.2466,
+      "step": 4750
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.140625,
+      "learning_rate": 8.273684210526317e-06,
+      "loss": 0.2397,
+      "step": 4760
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.3125,
+      "learning_rate": 8.257894736842105e-06,
+      "loss": 0.2904,
+      "step": 4770
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.40625,
+      "learning_rate": 8.242105263157896e-06,
+      "loss": 0.2848,
+      "step": 4780
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.34375,
+      "learning_rate": 8.226315789473684e-06,
+      "loss": 0.2291,
+      "step": 4790
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.34375,
+      "learning_rate": 8.210526315789475e-06,
+      "loss": 0.2031,
+      "step": 4800
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.65625,
+      "learning_rate": 8.194736842105263e-06,
+      "loss": 0.2391,
+      "step": 4810
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.34375,
+      "learning_rate": 8.178947368421054e-06,
+      "loss": 0.3645,
+      "step": 4820
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.6875,
+      "learning_rate": 8.163157894736842e-06,
+      "loss": 0.2322,
+      "step": 4830
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.75,
+      "learning_rate": 8.147368421052631e-06,
+      "loss": 0.2871,
+      "step": 4840
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.609375,
+      "learning_rate": 8.131578947368421e-06,
+      "loss": 0.2738,
+      "step": 4850
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.625,
+      "learning_rate": 8.11578947368421e-06,
+      "loss": 0.2776,
+      "step": 4860
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.453125,
+      "learning_rate": 8.1e-06,
+      "loss": 0.2701,
+      "step": 4870
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.96875,
+      "learning_rate": 8.084210526315789e-06,
+      "loss": 0.3132,
+      "step": 4880
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.734375,
+      "learning_rate": 8.06842105263158e-06,
+      "loss": 0.2406,
+      "step": 4890
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 7.46875,
+      "learning_rate": 8.052631578947368e-06,
+      "loss": 0.2348,
+      "step": 4900
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.734375,
+      "learning_rate": 8.036842105263157e-06,
+      "loss": 0.2231,
+      "step": 4910
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.3125,
+      "learning_rate": 8.021052631578949e-06,
+      "loss": 0.2324,
+      "step": 4920
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.515625,
+      "learning_rate": 8.005263157894737e-06,
+      "loss": 0.2207,
+      "step": 4930
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.75,
+      "learning_rate": 7.989473684210528e-06,
+      "loss": 0.2985,
+      "step": 4940
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.1875,
+      "learning_rate": 7.973684210526316e-06,
+      "loss": 0.2828,
+      "step": 4950
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.375,
+      "learning_rate": 7.957894736842107e-06,
+      "loss": 0.2745,
+      "step": 4960
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.1875,
+      "learning_rate": 7.942105263157895e-06,
+      "loss": 0.2668,
+      "step": 4970
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.03125,
+      "learning_rate": 7.926315789473684e-06,
+      "loss": 0.2334,
+      "step": 4980
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.53125,
+      "learning_rate": 7.910526315789474e-06,
+      "loss": 0.2461,
+      "step": 4990
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.6875,
+      "learning_rate": 7.894736842105263e-06,
+      "loss": 0.2325,
+      "step": 5000
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.828125,
+      "learning_rate": 7.878947368421053e-06,
+      "loss": 0.2599,
+      "step": 5010
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.34375,
+      "learning_rate": 7.863157894736842e-06,
+      "loss": 0.31,
+      "step": 5020
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.125,
+      "learning_rate": 7.847368421052632e-06,
+      "loss": 0.2413,
+      "step": 5030
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.40625,
+      "learning_rate": 7.831578947368421e-06,
+      "loss": 0.2332,
+      "step": 5040
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.46875,
+      "learning_rate": 7.81578947368421e-06,
+      "loss": 0.262,
+      "step": 5050
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.25,
+      "learning_rate": 7.8e-06,
+      "loss": 0.3235,
+      "step": 5060
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.34375,
+      "learning_rate": 7.784210526315789e-06,
+      "loss": 0.175,
+      "step": 5070
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.71875,
+      "learning_rate": 7.768421052631579e-06,
+      "loss": 0.2356,
+      "step": 5080
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.125,
+      "learning_rate": 7.75263157894737e-06,
+      "loss": 0.2758,
+      "step": 5090
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 3.03125,
+      "learning_rate": 7.73684210526316e-06,
+      "loss": 0.2448,
+      "step": 5100
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.21875,
+      "learning_rate": 7.721052631578948e-06,
+      "loss": 0.201,
+      "step": 5110
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.375,
+      "learning_rate": 7.705263157894737e-06,
+      "loss": 0.2505,
+      "step": 5120
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 7.4375,
+      "learning_rate": 7.689473684210527e-06,
+      "loss": 0.2609,
+      "step": 5130
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.90625,
+      "learning_rate": 7.673684210526316e-06,
+      "loss": 0.2542,
+      "step": 5140
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.2421875,
+      "learning_rate": 7.657894736842106e-06,
+      "loss": 0.2889,
+      "step": 5150
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.78125,
+      "learning_rate": 7.642105263157895e-06,
+      "loss": 0.3311,
+      "step": 5160
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.9375,
+      "learning_rate": 7.626315789473685e-06,
+      "loss": 0.2113,
+      "step": 5170
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 3.203125,
+      "learning_rate": 7.610526315789474e-06,
+      "loss": 0.2652,
+      "step": 5180
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.84375,
+      "learning_rate": 7.594736842105262e-06,
+      "loss": 0.2129,
+      "step": 5190
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 3.1875,
+      "learning_rate": 7.578947368421053e-06,
+      "loss": 0.2205,
+      "step": 5200
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.8125,
+      "learning_rate": 7.563157894736842e-06,
+      "loss": 0.2249,
+      "step": 5210
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.09375,
+      "learning_rate": 7.5473684210526326e-06,
+      "loss": 0.2124,
+      "step": 5220
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 7.5,
+      "learning_rate": 7.531578947368421e-06,
+      "loss": 0.2514,
+      "step": 5230
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.5625,
+      "learning_rate": 7.5157894736842115e-06,
+      "loss": 0.2329,
+      "step": 5240
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.78125,
+      "learning_rate": 7.5e-06,
+      "loss": 0.2487,
+      "step": 5250
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.75,
+      "learning_rate": 7.48421052631579e-06,
+      "loss": 0.275,
+      "step": 5260
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.59375,
+      "learning_rate": 7.468421052631579e-06,
+      "loss": 0.2123,
+      "step": 5270
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 3.90625,
+      "learning_rate": 7.452631578947369e-06,
+      "loss": 0.2539,
+      "step": 5280
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.96875,
+      "learning_rate": 7.436842105263157e-06,
+      "loss": 0.2035,
+      "step": 5290
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.875,
+      "learning_rate": 7.421052631578948e-06,
+      "loss": 0.2246,
+      "step": 5300
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 3.265625,
+      "learning_rate": 7.405263157894737e-06,
+      "loss": 0.2217,
+      "step": 5310
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 3.96875,
+      "learning_rate": 7.389473684210527e-06,
+      "loss": 0.283,
+      "step": 5320
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.109375,
+      "learning_rate": 7.373684210526316e-06,
+      "loss": 0.2569,
+      "step": 5330
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 3.46875,
+      "learning_rate": 7.357894736842106e-06,
+      "loss": 0.2087,
+      "step": 5340
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.5625,
+      "learning_rate": 7.342105263157895e-06,
+      "loss": 0.2625,
+      "step": 5350
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 3.453125,
+      "learning_rate": 7.326315789473684e-06,
+      "loss": 0.1765,
+      "step": 5360
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 11.9375,
+      "learning_rate": 7.310526315789473e-06,
+      "loss": 0.2767,
+      "step": 5370
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.25,
+      "learning_rate": 7.2947368421052636e-06,
+      "loss": 0.3017,
+      "step": 5380
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.03125,
+      "learning_rate": 7.278947368421053e-06,
+      "loss": 0.2989,
+      "step": 5390
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.0625,
+      "learning_rate": 7.2631578947368426e-06,
+      "loss": 0.2727,
+      "step": 5400
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.953125,
+      "learning_rate": 7.247368421052632e-06,
+      "loss": 0.2545,
+      "step": 5410
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 3.109375,
+      "learning_rate": 7.2315789473684215e-06,
+      "loss": 0.1987,
+      "step": 5420
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 7.0,
+      "learning_rate": 7.21578947368421e-06,
+      "loss": 0.2395,
+      "step": 5430
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 3.703125,
+      "learning_rate": 7.2e-06,
+      "loss": 0.2155,
+      "step": 5440
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.9375,
+      "learning_rate": 7.184210526315789e-06,
+      "loss": 0.2378,
+      "step": 5450
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.828125,
+      "learning_rate": 7.1684210526315795e-06,
+      "loss": 0.2106,
+      "step": 5460
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.28125,
+      "learning_rate": 7.152631578947369e-06,
+      "loss": 0.1986,
+      "step": 5470
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 3.390625,
+      "learning_rate": 7.1368421052631585e-06,
+      "loss": 0.2271,
+      "step": 5480
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.4375,
+      "learning_rate": 7.121052631578948e-06,
+      "loss": 0.2553,
+      "step": 5490
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.0625,
+      "learning_rate": 7.105263157894737e-06,
+      "loss": 0.2238,
+      "step": 5500
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.0625,
+      "learning_rate": 7.089473684210526e-06,
+      "loss": 0.2171,
+      "step": 5510
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.375,
+      "learning_rate": 7.073684210526316e-06,
+      "loss": 0.2435,
+      "step": 5520
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.28125,
+      "learning_rate": 7.057894736842105e-06,
+      "loss": 0.229,
+      "step": 5530
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.53125,
+      "learning_rate": 7.042105263157895e-06,
+      "loss": 0.2702,
+      "step": 5540
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 2.640625,
+      "learning_rate": 7.026315789473685e-06,
+      "loss": 0.2847,
+      "step": 5550
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.375,
+      "learning_rate": 7.010526315789474e-06,
+      "loss": 0.2038,
+      "step": 5560
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.125,
+      "learning_rate": 6.994736842105263e-06,
+      "loss": 0.2042,
+      "step": 5570
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.28125,
+      "learning_rate": 6.9789473684210525e-06,
+      "loss": 0.2213,
+      "step": 5580
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.890625,
+      "learning_rate": 6.963157894736842e-06,
+      "loss": 0.1767,
+      "step": 5590
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.546875,
+      "learning_rate": 6.9473684210526315e-06,
+      "loss": 0.2288,
+      "step": 5600
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.84375,
+      "learning_rate": 6.931578947368421e-06,
+      "loss": 0.1857,
+      "step": 5610
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.171875,
+      "learning_rate": 6.9157894736842105e-06,
+      "loss": 0.2064,
+      "step": 5620
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.078125,
+      "learning_rate": 6.900000000000001e-06,
+      "loss": 0.2454,
+      "step": 5630
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 16.75,
+      "learning_rate": 6.8842105263157895e-06,
+      "loss": 0.2678,
+      "step": 5640
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.40625,
+      "learning_rate": 6.868421052631579e-06,
+      "loss": 0.2492,
+      "step": 5650
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.90625,
+      "learning_rate": 6.8526315789473685e-06,
+      "loss": 0.216,
+      "step": 5660
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.5625,
+      "learning_rate": 6.836842105263158e-06,
+      "loss": 0.2555,
+      "step": 5670
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 7.5625,
+      "learning_rate": 6.8210526315789475e-06,
+      "loss": 0.326,
+      "step": 5680
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.453125,
+      "learning_rate": 6.805263157894737e-06,
+      "loss": 0.2912,
+      "step": 5690
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.88671875,
+      "learning_rate": 6.7894736842105264e-06,
+      "loss": 0.2688,
+      "step": 5700
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.09375,
+      "learning_rate": 6.773684210526317e-06,
+      "loss": 0.2003,
+      "step": 5710
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.59375,
+      "learning_rate": 6.7578947368421054e-06,
+      "loss": 0.2319,
+      "step": 5720
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 4.875,
+      "learning_rate": 6.742105263157895e-06,
+      "loss": 0.2731,
+      "step": 5730
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 4.40625,
+      "learning_rate": 6.726315789473684e-06,
+      "loss": 0.2665,
+      "step": 5740
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.40625,
+      "learning_rate": 6.710526315789474e-06,
+      "loss": 0.2487,
+      "step": 5750
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.28125,
+      "learning_rate": 6.694736842105263e-06,
+      "loss": 0.1795,
+      "step": 5760
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.609375,
+      "learning_rate": 6.678947368421053e-06,
+      "loss": 0.2613,
+      "step": 5770
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.203125,
+      "learning_rate": 6.663157894736842e-06,
+      "loss": 0.3303,
+      "step": 5780
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.25,
+      "learning_rate": 6.647368421052631e-06,
+      "loss": 0.2843,
+      "step": 5790
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 4.4375,
+      "learning_rate": 6.631578947368421e-06,
+      "loss": 0.3081,
+      "step": 5800
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 4.28125,
+      "learning_rate": 6.615789473684211e-06,
+      "loss": 0.2842,
+      "step": 5810
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.421875,
+      "learning_rate": 6.6e-06,
+      "loss": 0.1844,
+      "step": 5820
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.4375,
+      "learning_rate": 6.58421052631579e-06,
+      "loss": 0.2641,
+      "step": 5830
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.21875,
+      "learning_rate": 6.568421052631579e-06,
+      "loss": 0.2444,
+      "step": 5840
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 4.34375,
+      "learning_rate": 6.552631578947369e-06,
+      "loss": 0.1837,
+      "step": 5850
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.796875,
+      "learning_rate": 6.5368421052631575e-06,
+      "loss": 0.1657,
+      "step": 5860
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.59375,
+      "learning_rate": 6.521052631578947e-06,
+      "loss": 0.3079,
+      "step": 5870
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.90625,
+      "learning_rate": 6.505263157894737e-06,
+      "loss": 0.2566,
+      "step": 5880
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.28125,
+      "learning_rate": 6.489473684210527e-06,
+      "loss": 0.1638,
+      "step": 5890
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 9.4375,
+      "learning_rate": 6.473684210526316e-06,
+      "loss": 0.2744,
+      "step": 5900
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.75,
+      "learning_rate": 6.457894736842106e-06,
+      "loss": 0.2462,
+      "step": 5910
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 3.546875,
+      "learning_rate": 6.442105263157895e-06,
+      "loss": 0.2306,
+      "step": 5920
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 3.5,
+      "learning_rate": 6.426315789473684e-06,
+      "loss": 0.2583,
+      "step": 5930
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.1875,
+      "learning_rate": 6.410526315789473e-06,
+      "loss": 0.26,
+      "step": 5940
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.9375,
+      "learning_rate": 6.394736842105263e-06,
+      "loss": 0.2382,
+      "step": 5950
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.15625,
+      "learning_rate": 6.378947368421053e-06,
+      "loss": 0.2357,
+      "step": 5960
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.03125,
+      "learning_rate": 6.363157894736843e-06,
+      "loss": 0.2727,
+      "step": 5970
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.25,
+      "learning_rate": 6.347368421052632e-06,
+      "loss": 0.2614,
+      "step": 5980
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.90625,
+      "learning_rate": 6.331578947368422e-06,
+      "loss": 0.1835,
+      "step": 5990
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.59375,
+      "learning_rate": 6.31578947368421e-06,
+      "loss": 0.2649,
+      "step": 6000
+    },
+    {
+      "epoch": 0.3,
+      "eval_loss": 0.24383759498596191,
+      "eval_runtime": 64.6283,
+      "eval_samples_per_second": 15.473,
+      "eval_steps_per_second": 15.473,
+      "step": 6000
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.625,
+      "learning_rate": 6.3e-06,
+      "loss": 0.2733,
+      "step": 6010
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.28125,
+      "learning_rate": 6.284210526315789e-06,
+      "loss": 0.2434,
+      "step": 6020
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.0625,
+      "learning_rate": 6.268421052631579e-06,
+      "loss": 0.243,
+      "step": 6030
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.34375,
+      "learning_rate": 6.252631578947368e-06,
+      "loss": 0.257,
+      "step": 6040
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 7.1875,
+      "learning_rate": 6.236842105263159e-06,
+      "loss": 0.3039,
+      "step": 6050
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 7.4375,
+      "learning_rate": 6.221052631578948e-06,
+      "loss": 0.2401,
+      "step": 6060
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 11.1875,
+      "learning_rate": 6.205263157894737e-06,
+      "loss": 0.2435,
+      "step": 6070
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 3.484375,
+      "learning_rate": 6.189473684210526e-06,
+      "loss": 0.3069,
+      "step": 6080
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.53125,
+      "learning_rate": 6.173684210526316e-06,
+      "loss": 0.2593,
+      "step": 6090
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.03125,
+      "learning_rate": 6.157894736842105e-06,
+      "loss": 0.2392,
+      "step": 6100
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 8.625,
+      "learning_rate": 6.142105263157895e-06,
+      "loss": 0.1976,
+      "step": 6110
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 2.84375,
+      "learning_rate": 6.126315789473684e-06,
+      "loss": 0.2546,
+      "step": 6120
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 7.21875,
+      "learning_rate": 6.1105263157894746e-06,
+      "loss": 0.2016,
+      "step": 6130
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 4.84375,
+      "learning_rate": 6.094736842105263e-06,
+      "loss": 0.1514,
+      "step": 6140
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 3.125,
+      "learning_rate": 6.078947368421053e-06,
+      "loss": 0.2071,
+      "step": 6150
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 3.796875,
+      "learning_rate": 6.063157894736842e-06,
+      "loss": 0.2284,
+      "step": 6160
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 6.0,
+      "learning_rate": 6.047368421052632e-06,
+      "loss": 0.2323,
+      "step": 6170
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 9.625,
+      "learning_rate": 6.031578947368421e-06,
+      "loss": 0.3641,
+      "step": 6180
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.0625,
+      "learning_rate": 6.015789473684211e-06,
+      "loss": 0.2162,
+      "step": 6190
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 7.25,
+      "learning_rate": 6e-06,
+      "loss": 0.2786,
+      "step": 6200
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 3.640625,
+      "learning_rate": 5.98421052631579e-06,
+      "loss": 0.2747,
+      "step": 6210
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 3.15625,
+      "learning_rate": 5.968421052631579e-06,
+      "loss": 0.3007,
+      "step": 6220
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 3.546875,
+      "learning_rate": 5.952631578947369e-06,
+      "loss": 0.3041,
+      "step": 6230
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 3.0625,
+      "learning_rate": 5.936842105263158e-06,
+      "loss": 0.2188,
+      "step": 6240
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 2.96875,
+      "learning_rate": 5.921052631578948e-06,
+      "loss": 0.209,
+      "step": 6250
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 3.203125,
+      "learning_rate": 5.905263157894737e-06,
+      "loss": 0.2435,
+      "step": 6260
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.25,
+      "learning_rate": 5.889473684210527e-06,
+      "loss": 0.2715,
+      "step": 6270
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 4.40625,
+      "learning_rate": 5.873684210526315e-06,
+      "loss": 0.3081,
+      "step": 6280
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.6875,
+      "learning_rate": 5.857894736842105e-06,
+      "loss": 0.2746,
+      "step": 6290
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.40625,
+      "learning_rate": 5.842105263157895e-06,
+      "loss": 0.2818,
+      "step": 6300
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 3.4375,
+      "learning_rate": 5.8263157894736846e-06,
+      "loss": 0.2525,
+      "step": 6310
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4.46875,
+      "learning_rate": 5.810526315789474e-06,
+      "loss": 0.2082,
+      "step": 6320
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.5546875,
+      "learning_rate": 5.7947368421052635e-06,
+      "loss": 0.3066,
+      "step": 6330
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4.75,
+      "learning_rate": 5.778947368421053e-06,
+      "loss": 0.3205,
+      "step": 6340
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 1.359375,
+      "learning_rate": 5.763157894736842e-06,
+      "loss": 0.2384,
+      "step": 6350
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 3.1875,
+      "learning_rate": 5.747368421052631e-06,
+      "loss": 0.2319,
+      "step": 6360
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4.1875,
+      "learning_rate": 5.731578947368421e-06,
+      "loss": 0.2959,
+      "step": 6370
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.28125,
+      "learning_rate": 5.715789473684211e-06,
+      "loss": 0.228,
+      "step": 6380
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 3.921875,
+      "learning_rate": 5.7000000000000005e-06,
+      "loss": 0.2771,
+      "step": 6390
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4.4375,
+      "learning_rate": 5.68421052631579e-06,
+      "loss": 0.2381,
+      "step": 6400
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 3.859375,
+      "learning_rate": 5.6684210526315795e-06,
+      "loss": 0.2612,
+      "step": 6410
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 2.96875,
+      "learning_rate": 5.652631578947368e-06,
+      "loss": 0.2266,
+      "step": 6420
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.125,
+      "learning_rate": 5.636842105263158e-06,
+      "loss": 0.2402,
+      "step": 6430
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 2.703125,
+      "learning_rate": 5.621052631578947e-06,
+      "loss": 0.24,
+      "step": 6440
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.4375,
+      "learning_rate": 5.605263157894737e-06,
+      "loss": 0.2476,
+      "step": 6450
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 7.375,
+      "learning_rate": 5.589473684210527e-06,
+      "loss": 0.2585,
+      "step": 6460
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.0,
+      "learning_rate": 5.573684210526316e-06,
+      "loss": 0.3165,
+      "step": 6470
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4.03125,
+      "learning_rate": 5.557894736842106e-06,
+      "loss": 0.2045,
+      "step": 6480
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 3.890625,
+      "learning_rate": 5.542105263157895e-06,
+      "loss": 0.2327,
+      "step": 6490
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.8125,
+      "learning_rate": 5.526315789473684e-06,
+      "loss": 0.2356,
+      "step": 6500
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 4.0,
+      "learning_rate": 5.5105263157894735e-06,
+      "loss": 0.2103,
+      "step": 6510
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 4.90625,
+      "learning_rate": 5.494736842105263e-06,
+      "loss": 0.2519,
+      "step": 6520
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.3125,
+      "learning_rate": 5.4789473684210525e-06,
+      "loss": 0.3197,
+      "step": 6530
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.953125,
+      "learning_rate": 5.463157894736842e-06,
+      "loss": 0.1994,
+      "step": 6540
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.875,
+      "learning_rate": 5.447368421052632e-06,
+      "loss": 0.2387,
+      "step": 6550
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.78125,
+      "learning_rate": 5.431578947368422e-06,
+      "loss": 0.2801,
+      "step": 6560
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 3.59375,
+      "learning_rate": 5.4157894736842105e-06,
+      "loss": 0.2357,
+      "step": 6570
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 3.625,
+      "learning_rate": 5.4e-06,
+      "loss": 0.2463,
+      "step": 6580
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 4.125,
+      "learning_rate": 5.3842105263157895e-06,
+      "loss": 0.2565,
+      "step": 6590
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 7.0,
+      "learning_rate": 5.368421052631579e-06,
+      "loss": 0.3018,
+      "step": 6600
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.828125,
+      "learning_rate": 5.3526315789473684e-06,
+      "loss": 0.2322,
+      "step": 6610
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.5625,
+      "learning_rate": 5.336842105263158e-06,
+      "loss": 0.2985,
+      "step": 6620
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.40625,
+      "learning_rate": 5.321052631578948e-06,
+      "loss": 0.2486,
+      "step": 6630
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.3125,
+      "learning_rate": 5.305263157894737e-06,
+      "loss": 0.2096,
+      "step": 6640
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.59375,
+      "learning_rate": 5.289473684210526e-06,
+      "loss": 0.1774,
+      "step": 6650
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 3.71875,
+      "learning_rate": 5.273684210526316e-06,
+      "loss": 0.2052,
+      "step": 6660
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 4.0,
+      "learning_rate": 5.257894736842105e-06,
+      "loss": 0.3246,
+      "step": 6670
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.15625,
+      "learning_rate": 5.242105263157895e-06,
+      "loss": 0.2495,
+      "step": 6680
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 3.359375,
+      "learning_rate": 5.226315789473684e-06,
+      "loss": 0.2657,
+      "step": 6690
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 3.78125,
+      "learning_rate": 5.210526315789474e-06,
+      "loss": 0.2189,
+      "step": 6700
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.84375,
+      "learning_rate": 5.194736842105263e-06,
+      "loss": 0.2425,
+      "step": 6710
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 2.375,
+      "learning_rate": 5.178947368421053e-06,
+      "loss": 0.258,
+      "step": 6720
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.28125,
+      "learning_rate": 5.163157894736842e-06,
+      "loss": 0.2181,
+      "step": 6730
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 4.78125,
+      "learning_rate": 5.147368421052632e-06,
+      "loss": 0.2184,
+      "step": 6740
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.5625,
+      "learning_rate": 5.131578947368421e-06,
+      "loss": 0.2479,
+      "step": 6750
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 3.890625,
+      "learning_rate": 5.115789473684211e-06,
+      "loss": 0.2786,
+      "step": 6760
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 3.203125,
+      "learning_rate": 5.1e-06,
+      "loss": 0.2686,
+      "step": 6770
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.03125,
+      "learning_rate": 5.084210526315789e-06,
+      "loss": 0.2323,
+      "step": 6780
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.625,
+      "learning_rate": 5.0684210526315784e-06,
+      "loss": 0.2765,
+      "step": 6790
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 4.09375,
+      "learning_rate": 5.052631578947369e-06,
+      "loss": 0.2438,
+      "step": 6800
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.25,
+      "learning_rate": 5.036842105263158e-06,
+      "loss": 0.2885,
+      "step": 6810
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 4.03125,
+      "learning_rate": 5.021052631578948e-06,
+      "loss": 0.2571,
+      "step": 6820
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 4.75,
+      "learning_rate": 5.005263157894737e-06,
+      "loss": 0.2096,
+      "step": 6830
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.4375,
+      "learning_rate": 4.989473684210527e-06,
+      "loss": 0.2472,
+      "step": 6840
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 3.859375,
+      "learning_rate": 4.973684210526315e-06,
+      "loss": 0.2378,
+      "step": 6850
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 7.8125,
+      "learning_rate": 4.957894736842105e-06,
+      "loss": 0.2411,
+      "step": 6860
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 2.859375,
+      "learning_rate": 4.942105263157894e-06,
+      "loss": 0.2572,
+      "step": 6870
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 4.4375,
+      "learning_rate": 4.926315789473685e-06,
+      "loss": 0.2721,
+      "step": 6880
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 4.0625,
+      "learning_rate": 4.910526315789474e-06,
+      "loss": 0.2846,
+      "step": 6890
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 3.3125,
+      "learning_rate": 4.894736842105264e-06,
+      "loss": 0.2659,
+      "step": 6900
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.3125,
+      "learning_rate": 4.878947368421053e-06,
+      "loss": 0.1965,
+      "step": 6910
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.65625,
+      "learning_rate": 4.863157894736842e-06,
+      "loss": 0.2682,
+      "step": 6920
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.375,
+      "learning_rate": 4.847368421052631e-06,
+      "loss": 0.2748,
+      "step": 6930
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.5625,
+      "learning_rate": 4.831578947368421e-06,
+      "loss": 0.2024,
+      "step": 6940
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.75,
+      "learning_rate": 4.81578947368421e-06,
+      "loss": 0.2885,
+      "step": 6950
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 3.84375,
+      "learning_rate": 4.800000000000001e-06,
+      "loss": 0.2016,
+      "step": 6960
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.34375,
+      "learning_rate": 4.78421052631579e-06,
+      "loss": 0.2577,
+      "step": 6970
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 2.53125,
+      "learning_rate": 4.76842105263158e-06,
+      "loss": 0.2533,
+      "step": 6980
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 3.5625,
+      "learning_rate": 4.752631578947368e-06,
+      "loss": 0.1985,
+      "step": 6990
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.96875,
+      "learning_rate": 4.736842105263158e-06,
+      "loss": 0.2907,
+      "step": 7000
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.90625,
+      "learning_rate": 4.721052631578947e-06,
+      "loss": 0.1911,
+      "step": 7010
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 3.8125,
+      "learning_rate": 4.705263157894737e-06,
+      "loss": 0.278,
+      "step": 7020
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.78125,
+      "learning_rate": 4.689473684210526e-06,
+      "loss": 0.2479,
+      "step": 7030
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 2.90625,
+      "learning_rate": 4.673684210526316e-06,
+      "loss": 0.2663,
+      "step": 7040
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.53125,
+      "learning_rate": 4.657894736842106e-06,
+      "loss": 0.2356,
+      "step": 7050
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 3.5625,
+      "learning_rate": 4.642105263157895e-06,
+      "loss": 0.2588,
+      "step": 7060
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.9375,
+      "learning_rate": 4.626315789473684e-06,
+      "loss": 0.2908,
+      "step": 7070
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 3.03125,
+      "learning_rate": 4.610526315789474e-06,
+      "loss": 0.1783,
+      "step": 7080
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.594736842105263e-06,
+      "loss": 0.2988,
+      "step": 7090
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 3.84375,
+      "learning_rate": 4.578947368421053e-06,
+      "loss": 0.2591,
+      "step": 7100
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 3.875,
+      "learning_rate": 4.563157894736842e-06,
+      "loss": 0.2693,
+      "step": 7110
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 3.9375,
+      "learning_rate": 4.547368421052632e-06,
+      "loss": 0.2101,
+      "step": 7120
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.5625,
+      "learning_rate": 4.531578947368421e-06,
+      "loss": 0.3042,
+      "step": 7130
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 3.078125,
+      "learning_rate": 4.515789473684211e-06,
+      "loss": 0.1868,
+      "step": 7140
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 4.6875,
+      "learning_rate": 4.5e-06,
+      "loss": 0.2599,
+      "step": 7150
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.15625,
+      "learning_rate": 4.48421052631579e-06,
+      "loss": 0.2096,
+      "step": 7160
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 4.875,
+      "learning_rate": 4.468421052631579e-06,
+      "loss": 0.2822,
+      "step": 7170
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 7.375,
+      "learning_rate": 4.452631578947369e-06,
+      "loss": 0.2097,
+      "step": 7180
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 4.5625,
+      "learning_rate": 4.436842105263158e-06,
+      "loss": 0.234,
+      "step": 7190
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 3.296875,
+      "learning_rate": 4.421052631578947e-06,
+      "loss": 0.2525,
+      "step": 7200
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.90625,
+      "learning_rate": 4.405263157894737e-06,
+      "loss": 0.2767,
+      "step": 7210
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 3.4375,
+      "learning_rate": 4.3894736842105266e-06,
+      "loss": 0.2579,
+      "step": 7220
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.373684210526316e-06,
+      "loss": 0.2289,
+      "step": 7230
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 3.75,
+      "learning_rate": 4.3578947368421055e-06,
+      "loss": 0.2551,
+      "step": 7240
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.482421875,
+      "learning_rate": 4.342105263157895e-06,
+      "loss": 0.2185,
+      "step": 7250
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 4.96875,
+      "learning_rate": 4.3263157894736845e-06,
+      "loss": 0.2201,
+      "step": 7260
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.65625,
+      "learning_rate": 4.310526315789474e-06,
+      "loss": 0.2462,
+      "step": 7270
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 2.28125,
+      "learning_rate": 4.294736842105263e-06,
+      "loss": 0.1894,
+      "step": 7280
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 4.46875,
+      "learning_rate": 4.278947368421052e-06,
+      "loss": 0.2699,
+      "step": 7290
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.875,
+      "learning_rate": 4.2631578947368425e-06,
+      "loss": 0.2688,
+      "step": 7300
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 4.34375,
+      "learning_rate": 4.247368421052632e-06,
+      "loss": 0.2514,
+      "step": 7310
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.34375,
+      "learning_rate": 4.2315789473684215e-06,
+      "loss": 0.2109,
+      "step": 7320
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.59375,
+      "learning_rate": 4.215789473684211e-06,
+      "loss": 0.2608,
+      "step": 7330
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.875,
+      "learning_rate": 4.2000000000000004e-06,
+      "loss": 0.1881,
+      "step": 7340
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 4.71875,
+      "learning_rate": 4.184210526315789e-06,
+      "loss": 0.29,
+      "step": 7350
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.46875,
+      "learning_rate": 4.168421052631579e-06,
+      "loss": 0.2948,
+      "step": 7360
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 3.890625,
+      "learning_rate": 4.152631578947368e-06,
+      "loss": 0.2215,
+      "step": 7370
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 4.875,
+      "learning_rate": 4.136842105263158e-06,
+      "loss": 0.2139,
+      "step": 7380
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 4.03125,
+      "learning_rate": 4.121052631578948e-06,
+      "loss": 0.2168,
+      "step": 7390
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.875,
+      "learning_rate": 4.105263157894737e-06,
+      "loss": 0.2291,
+      "step": 7400
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 3.8125,
+      "learning_rate": 4.089473684210527e-06,
+      "loss": 0.3034,
+      "step": 7410
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 3.515625,
+      "learning_rate": 4.0736842105263155e-06,
+      "loss": 0.2004,
+      "step": 7420
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 3.5625,
+      "learning_rate": 4.057894736842105e-06,
+      "loss": 0.2953,
+      "step": 7430
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 4.0625,
+      "learning_rate": 4.0421052631578945e-06,
+      "loss": 0.255,
+      "step": 7440
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 4.71875,
+      "learning_rate": 4.026315789473684e-06,
+      "loss": 0.2377,
+      "step": 7450
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.90625,
+      "learning_rate": 4.010526315789474e-06,
+      "loss": 0.2082,
+      "step": 7460
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 4.5625,
+      "learning_rate": 3.994736842105264e-06,
+      "loss": 0.1915,
+      "step": 7470
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 8.9375,
+      "learning_rate": 3.978947368421053e-06,
+      "loss": 0.2735,
+      "step": 7480
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 10.625,
+      "learning_rate": 3.963157894736842e-06,
+      "loss": 0.3229,
+      "step": 7490
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.671875,
+      "learning_rate": 3.9473684210526315e-06,
+      "loss": 0.2364,
+      "step": 7500
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 9.3125,
+      "learning_rate": 3.931578947368421e-06,
+      "loss": 0.2088,
+      "step": 7510
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 4.125,
+      "learning_rate": 3.9157894736842104e-06,
+      "loss": 0.2352,
+      "step": 7520
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.34375,
+      "learning_rate": 3.9e-06,
+      "loss": 0.2463,
+      "step": 7530
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.75,
+      "learning_rate": 3.884210526315789e-06,
+      "loss": 0.1959,
+      "step": 7540
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 9.375,
+      "learning_rate": 3.86842105263158e-06,
+      "loss": 0.2796,
+      "step": 7550
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 2.84375,
+      "learning_rate": 3.852631578947368e-06,
+      "loss": 0.2441,
+      "step": 7560
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.9375,
+      "learning_rate": 3.836842105263158e-06,
+      "loss": 0.215,
+      "step": 7570
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.75,
+      "learning_rate": 3.821052631578947e-06,
+      "loss": 0.2734,
+      "step": 7580
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 4.125,
+      "learning_rate": 3.805263157894737e-06,
+      "loss": 0.2116,
+      "step": 7590
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 4.28125,
+      "learning_rate": 3.7894736842105264e-06,
+      "loss": 0.2347,
+      "step": 7600
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.78125,
+      "learning_rate": 3.7736842105263163e-06,
+      "loss": 0.2971,
+      "step": 7610
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.921875,
+      "learning_rate": 3.7578947368421058e-06,
+      "loss": 0.2963,
+      "step": 7620
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.59375,
+      "learning_rate": 3.742105263157895e-06,
+      "loss": 0.219,
+      "step": 7630
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 2.578125,
+      "learning_rate": 3.7263157894736843e-06,
+      "loss": 0.2094,
+      "step": 7640
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.515625,
+      "learning_rate": 3.710526315789474e-06,
+      "loss": 0.2544,
+      "step": 7650
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 4.25,
+      "learning_rate": 3.6947368421052633e-06,
+      "loss": 0.2504,
+      "step": 7660
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 4.8125,
+      "learning_rate": 3.678947368421053e-06,
+      "loss": 0.2364,
+      "step": 7670
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 6.0,
+      "learning_rate": 3.663157894736842e-06,
+      "loss": 0.295,
+      "step": 7680
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 2.796875,
+      "learning_rate": 3.6473684210526318e-06,
+      "loss": 0.3208,
+      "step": 7690
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.375,
+      "learning_rate": 3.6315789473684213e-06,
+      "loss": 0.2307,
+      "step": 7700
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.03125,
+      "learning_rate": 3.6157894736842108e-06,
+      "loss": 0.2242,
+      "step": 7710
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 2.84375,
+      "learning_rate": 3.6e-06,
+      "loss": 0.1419,
+      "step": 7720
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.4375,
+      "learning_rate": 3.5842105263157898e-06,
+      "loss": 0.2643,
+      "step": 7730
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.96875,
+      "learning_rate": 3.5684210526315792e-06,
+      "loss": 0.3376,
+      "step": 7740
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.953125,
+      "learning_rate": 3.5526315789473683e-06,
+      "loss": 0.2768,
+      "step": 7750
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.015625,
+      "learning_rate": 3.536842105263158e-06,
+      "loss": 0.2669,
+      "step": 7760
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.15625,
+      "learning_rate": 3.5210526315789473e-06,
+      "loss": 0.2625,
+      "step": 7770
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.5625,
+      "learning_rate": 3.505263157894737e-06,
+      "loss": 0.2951,
+      "step": 7780
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.1875,
+      "learning_rate": 3.4894736842105263e-06,
+      "loss": 0.3062,
+      "step": 7790
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.96875,
+      "learning_rate": 3.4736842105263158e-06,
+      "loss": 0.1846,
+      "step": 7800
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.46875,
+      "learning_rate": 3.4578947368421053e-06,
+      "loss": 0.2531,
+      "step": 7810
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.4375,
+      "learning_rate": 3.4421052631578947e-06,
+      "loss": 0.2476,
+      "step": 7820
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.3125,
+      "learning_rate": 3.4263157894736842e-06,
+      "loss": 0.2389,
+      "step": 7830
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.125,
+      "learning_rate": 3.4105263157894737e-06,
+      "loss": 0.2231,
+      "step": 7840
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 2.09375,
+      "learning_rate": 3.3947368421052632e-06,
+      "loss": 0.2248,
+      "step": 7850
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 2.03125,
+      "learning_rate": 3.3789473684210527e-06,
+      "loss": 0.288,
+      "step": 7860
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.15625,
+      "learning_rate": 3.363157894736842e-06,
+      "loss": 0.2108,
+      "step": 7870
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.0625,
+      "learning_rate": 3.3473684210526317e-06,
+      "loss": 0.2839,
+      "step": 7880
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.9375,
+      "learning_rate": 3.331578947368421e-06,
+      "loss": 0.2332,
+      "step": 7890
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 4.25,
+      "learning_rate": 3.3157894736842107e-06,
+      "loss": 0.1818,
+      "step": 7900
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.875,
+      "learning_rate": 3.3e-06,
+      "loss": 0.2643,
+      "step": 7910
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 4.21875,
+      "learning_rate": 3.2842105263157897e-06,
+      "loss": 0.2253,
+      "step": 7920
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 7.78125,
+      "learning_rate": 3.2684210526315787e-06,
+      "loss": 0.1979,
+      "step": 7930
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 3.203125,
+      "learning_rate": 3.2526315789473686e-06,
+      "loss": 0.2705,
+      "step": 7940
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 3.84375,
+      "learning_rate": 3.236842105263158e-06,
+      "loss": 0.2321,
+      "step": 7950
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 4.0625,
+      "learning_rate": 3.2210526315789476e-06,
+      "loss": 0.2483,
+      "step": 7960
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 4.1875,
+      "learning_rate": 3.2052631578947367e-06,
+      "loss": 0.279,
+      "step": 7970
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.28125,
+      "learning_rate": 3.1894736842105266e-06,
+      "loss": 0.289,
+      "step": 7980
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 2.25,
+      "learning_rate": 3.173684210526316e-06,
+      "loss": 0.1856,
+      "step": 7990
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 2.453125,
+      "learning_rate": 3.157894736842105e-06,
+      "loss": 0.2089,
+      "step": 8000
+    },
+    {
+      "epoch": 0.4,
+      "eval_loss": 0.24291838705539703,
+      "eval_runtime": 64.8709,
+      "eval_samples_per_second": 15.415,
+      "eval_steps_per_second": 15.415,
+      "step": 8000
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 4.34375,
+      "learning_rate": 3.1421052631578947e-06,
+      "loss": 0.255,
+      "step": 8010
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 4.09375,
+      "learning_rate": 3.126315789473684e-06,
+      "loss": 0.2536,
+      "step": 8020
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.1875,
+      "learning_rate": 3.110526315789474e-06,
+      "loss": 0.2415,
+      "step": 8030
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 4.0,
+      "learning_rate": 3.094736842105263e-06,
+      "loss": 0.1958,
+      "step": 8040
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 3.71875,
+      "learning_rate": 3.0789473684210526e-06,
+      "loss": 0.2015,
+      "step": 8050
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 4.5,
+      "learning_rate": 3.063157894736842e-06,
+      "loss": 0.2034,
+      "step": 8060
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 3.796875,
+      "learning_rate": 3.0473684210526316e-06,
+      "loss": 0.2505,
+      "step": 8070
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.15625,
+      "learning_rate": 3.031578947368421e-06,
+      "loss": 0.254,
+      "step": 8080
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.9375,
+      "learning_rate": 3.0157894736842106e-06,
+      "loss": 0.3193,
+      "step": 8090
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 4.5625,
+      "learning_rate": 3e-06,
+      "loss": 0.2621,
+      "step": 8100
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 4.9375,
+      "learning_rate": 2.9842105263157896e-06,
+      "loss": 0.2749,
+      "step": 8110
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 4.5625,
+      "learning_rate": 2.968421052631579e-06,
+      "loss": 0.289,
+      "step": 8120
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 4.34375,
+      "learning_rate": 2.9526315789473685e-06,
+      "loss": 0.2269,
+      "step": 8130
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 9.5,
+      "learning_rate": 2.9368421052631576e-06,
+      "loss": 0.2318,
+      "step": 8140
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.21875,
+      "learning_rate": 2.9210526315789475e-06,
+      "loss": 0.2405,
+      "step": 8150
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.15625,
+      "learning_rate": 2.905263157894737e-06,
+      "loss": 0.2535,
+      "step": 8160
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 4.34375,
+      "learning_rate": 2.8894736842105265e-06,
+      "loss": 0.264,
+      "step": 8170
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.125,
+      "learning_rate": 2.8736842105263156e-06,
+      "loss": 0.2379,
+      "step": 8180
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 2.71875,
+      "learning_rate": 2.8578947368421055e-06,
+      "loss": 0.227,
+      "step": 8190
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 2.90625,
+      "learning_rate": 2.842105263157895e-06,
+      "loss": 0.2269,
+      "step": 8200
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 3.265625,
+      "learning_rate": 2.826315789473684e-06,
+      "loss": 0.2254,
+      "step": 8210
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 3.015625,
+      "learning_rate": 2.8105263157894735e-06,
+      "loss": 0.2029,
+      "step": 8220
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.46875,
+      "learning_rate": 2.7947368421052635e-06,
+      "loss": 0.3118,
+      "step": 8230
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 3.78125,
+      "learning_rate": 2.778947368421053e-06,
+      "loss": 0.3158,
+      "step": 8240
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 4.6875,
+      "learning_rate": 2.763157894736842e-06,
+      "loss": 0.2401,
+      "step": 8250
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.15625,
+      "learning_rate": 2.7473684210526315e-06,
+      "loss": 0.2376,
+      "step": 8260
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 4.25,
+      "learning_rate": 2.731578947368421e-06,
+      "loss": 0.2148,
+      "step": 8270
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 2.453125,
+      "learning_rate": 2.715789473684211e-06,
+      "loss": 0.2275,
+      "step": 8280
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 3.625,
+      "learning_rate": 2.7e-06,
+      "loss": 0.1917,
+      "step": 8290
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 2.296875,
+      "learning_rate": 2.6842105263157895e-06,
+      "loss": 0.2804,
+      "step": 8300
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 8.875,
+      "learning_rate": 2.668421052631579e-06,
+      "loss": 0.2542,
+      "step": 8310
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 7.625,
+      "learning_rate": 2.6526315789473685e-06,
+      "loss": 0.2382,
+      "step": 8320
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 7.03125,
+      "learning_rate": 2.636842105263158e-06,
+      "loss": 0.2495,
+      "step": 8330
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 3.46875,
+      "learning_rate": 2.6210526315789474e-06,
+      "loss": 0.2376,
+      "step": 8340
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.875,
+      "learning_rate": 2.605263157894737e-06,
+      "loss": 0.2132,
+      "step": 8350
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.90625,
+      "learning_rate": 2.5894736842105264e-06,
+      "loss": 0.1529,
+      "step": 8360
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 3.0625,
+      "learning_rate": 2.573684210526316e-06,
+      "loss": 0.2545,
+      "step": 8370
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.40625,
+      "learning_rate": 2.5578947368421054e-06,
+      "loss": 0.2317,
+      "step": 8380
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 2.890625,
+      "learning_rate": 2.5421052631578945e-06,
+      "loss": 0.2406,
+      "step": 8390
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 3.3125,
+      "learning_rate": 2.5263157894736844e-06,
+      "loss": 0.1859,
+      "step": 8400
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.8125,
+      "learning_rate": 2.510526315789474e-06,
+      "loss": 0.2577,
+      "step": 8410
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.125,
+      "learning_rate": 2.4947368421052634e-06,
+      "loss": 0.2007,
+      "step": 8420
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 3.828125,
+      "learning_rate": 2.4789473684210524e-06,
+      "loss": 0.2364,
+      "step": 8430
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.3125,
+      "learning_rate": 2.4631578947368424e-06,
+      "loss": 0.1901,
+      "step": 8440
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.5625,
+      "learning_rate": 2.447368421052632e-06,
+      "loss": 0.2532,
+      "step": 8450
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 3.546875,
+      "learning_rate": 2.431578947368421e-06,
+      "loss": 0.2875,
+      "step": 8460
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 2.546875,
+      "learning_rate": 2.4157894736842104e-06,
+      "loss": 0.2616,
+      "step": 8470
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.78125,
+      "learning_rate": 2.4000000000000003e-06,
+      "loss": 0.2019,
+      "step": 8480
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 9.0625,
+      "learning_rate": 2.38421052631579e-06,
+      "loss": 0.2144,
+      "step": 8490
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 3.671875,
+      "learning_rate": 2.368421052631579e-06,
+      "loss": 0.2477,
+      "step": 8500
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 3.984375,
+      "learning_rate": 2.3526315789473684e-06,
+      "loss": 0.2677,
+      "step": 8510
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.78125,
+      "learning_rate": 2.336842105263158e-06,
+      "loss": 0.2082,
+      "step": 8520
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.15625,
+      "learning_rate": 2.3210526315789473e-06,
+      "loss": 0.2629,
+      "step": 8530
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 3.796875,
+      "learning_rate": 2.305263157894737e-06,
+      "loss": 0.207,
+      "step": 8540
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.0625,
+      "learning_rate": 2.2894736842105263e-06,
+      "loss": 0.2521,
+      "step": 8550
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.9375,
+      "learning_rate": 2.273684210526316e-06,
+      "loss": 0.2628,
+      "step": 8560
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.71875,
+      "learning_rate": 2.2578947368421053e-06,
+      "loss": 0.2633,
+      "step": 8570
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 2.890625,
+      "learning_rate": 2.242105263157895e-06,
+      "loss": 0.2246,
+      "step": 8580
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 3.921875,
+      "learning_rate": 2.2263157894736843e-06,
+      "loss": 0.1778,
+      "step": 8590
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 7.125,
+      "learning_rate": 2.2105263157894734e-06,
+      "loss": 0.2543,
+      "step": 8600
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.03125,
+      "learning_rate": 2.1947368421052633e-06,
+      "loss": 0.207,
+      "step": 8610
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.5,
+      "learning_rate": 2.1789473684210528e-06,
+      "loss": 0.2041,
+      "step": 8620
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 4.1875,
+      "learning_rate": 2.1631578947368423e-06,
+      "loss": 0.2527,
+      "step": 8630
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 4.34375,
+      "learning_rate": 2.1473684210526313e-06,
+      "loss": 0.2039,
+      "step": 8640
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 4.59375,
+      "learning_rate": 2.1315789473684212e-06,
+      "loss": 0.2555,
+      "step": 8650
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 3.75,
+      "learning_rate": 2.1157894736842107e-06,
+      "loss": 0.2033,
+      "step": 8660
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 2.34375,
+      "learning_rate": 2.1000000000000002e-06,
+      "loss": 0.2344,
+      "step": 8670
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.75,
+      "learning_rate": 2.0842105263157893e-06,
+      "loss": 0.2626,
+      "step": 8680
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 3.421875,
+      "learning_rate": 2.068421052631579e-06,
+      "loss": 0.1754,
+      "step": 8690
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 4.625,
+      "learning_rate": 2.0526315789473687e-06,
+      "loss": 0.2125,
+      "step": 8700
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 3.03125,
+      "learning_rate": 2.0368421052631578e-06,
+      "loss": 0.2514,
+      "step": 8710
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 2.640625,
+      "learning_rate": 2.0210526315789473e-06,
+      "loss": 0.2206,
+      "step": 8720
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 3.375,
+      "learning_rate": 2.005263157894737e-06,
+      "loss": 0.2518,
+      "step": 8730
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 6.6875,
+      "learning_rate": 1.9894736842105267e-06,
+      "loss": 0.2383,
+      "step": 8740
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.90625,
+      "learning_rate": 1.9736842105263157e-06,
+      "loss": 0.2372,
+      "step": 8750
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 2.890625,
+      "learning_rate": 1.9578947368421052e-06,
+      "loss": 0.2932,
+      "step": 8760
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.9421052631578947e-06,
+      "loss": 0.2537,
+      "step": 8770
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 3.375,
+      "learning_rate": 1.926315789473684e-06,
+      "loss": 0.2264,
+      "step": 8780
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.9105263157894737e-06,
+      "loss": 0.2428,
+      "step": 8790
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.8947368421052632e-06,
+      "loss": 0.2256,
+      "step": 8800
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 3.734375,
+      "learning_rate": 1.8789473684210529e-06,
+      "loss": 0.2106,
+      "step": 8810
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 3.84375,
+      "learning_rate": 1.8631578947368422e-06,
+      "loss": 0.2479,
+      "step": 8820
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 3.671875,
+      "learning_rate": 1.8473684210526317e-06,
+      "loss": 0.2271,
+      "step": 8830
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.831578947368421e-06,
+      "loss": 0.2244,
+      "step": 8840
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.8157894736842106e-06,
+      "loss": 0.2843,
+      "step": 8850
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.625,
+      "learning_rate": 1.8e-06,
+      "loss": 0.2672,
+      "step": 8860
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.125,
+      "learning_rate": 1.7842105263157896e-06,
+      "loss": 0.2829,
+      "step": 8870
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.768421052631579e-06,
+      "loss": 0.1935,
+      "step": 8880
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.7526315789473686e-06,
+      "loss": 0.2524,
+      "step": 8890
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 3.078125,
+      "learning_rate": 1.7368421052631579e-06,
+      "loss": 0.2422,
+      "step": 8900
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 3.109375,
+      "learning_rate": 1.7210526315789474e-06,
+      "loss": 0.2526,
+      "step": 8910
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.7052631578947369e-06,
+      "loss": 0.2312,
+      "step": 8920
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 3.390625,
+      "learning_rate": 1.6894736842105264e-06,
+      "loss": 0.2285,
+      "step": 8930
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.6736842105263158e-06,
+      "loss": 0.2095,
+      "step": 8940
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 5.59375,
+      "learning_rate": 1.6578947368421053e-06,
+      "loss": 0.2667,
+      "step": 8950
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 3.015625,
+      "learning_rate": 1.6421052631578948e-06,
+      "loss": 0.2673,
+      "step": 8960
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.625,
+      "learning_rate": 1.6263157894736843e-06,
+      "loss": 0.2892,
+      "step": 8970
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.6105263157894738e-06,
+      "loss": 0.2294,
+      "step": 8980
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.65625,
+      "learning_rate": 1.5947368421052633e-06,
+      "loss": 0.3837,
+      "step": 8990
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 8.0,
+      "learning_rate": 1.5789473684210526e-06,
+      "loss": 0.3285,
+      "step": 9000
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 3.703125,
+      "learning_rate": 1.563157894736842e-06,
+      "loss": 0.1831,
+      "step": 9010
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 3.265625,
+      "learning_rate": 1.5473684210526316e-06,
+      "loss": 0.2585,
+      "step": 9020
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.21875,
+      "learning_rate": 1.531578947368421e-06,
+      "loss": 0.2129,
+      "step": 9030
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 5.53125,
+      "learning_rate": 1.5157894736842105e-06,
+      "loss": 0.1851,
+      "step": 9040
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 3.59375,
+      "learning_rate": 1.5e-06,
+      "loss": 0.2222,
+      "step": 9050
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 2.265625,
+      "learning_rate": 1.4842105263157895e-06,
+      "loss": 0.2288,
+      "step": 9060
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 2.5625,
+      "learning_rate": 1.4684210526315788e-06,
+      "loss": 0.231,
+      "step": 9070
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.4526315789473685e-06,
+      "loss": 0.2897,
+      "step": 9080
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 3.75,
+      "learning_rate": 1.4368421052631578e-06,
+      "loss": 0.3617,
+      "step": 9090
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.4210526315789475e-06,
+      "loss": 0.2884,
+      "step": 9100
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.875,
+      "learning_rate": 1.4052631578947368e-06,
+      "loss": 0.2562,
+      "step": 9110
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 3.734375,
+      "learning_rate": 1.3894736842105265e-06,
+      "loss": 0.2229,
+      "step": 9120
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.25,
+      "learning_rate": 1.3736842105263158e-06,
+      "loss": 0.2655,
+      "step": 9130
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.3578947368421055e-06,
+      "loss": 0.2506,
+      "step": 9140
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.3421052631578947e-06,
+      "loss": 0.2823,
+      "step": 9150
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.3263157894736842e-06,
+      "loss": 0.2113,
+      "step": 9160
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.3105263157894737e-06,
+      "loss": 0.2753,
+      "step": 9170
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.2947368421052632e-06,
+      "loss": 0.2243,
+      "step": 9180
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 3.03125,
+      "learning_rate": 1.2789473684210527e-06,
+      "loss": 0.2673,
+      "step": 9190
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.28125,
+      "learning_rate": 1.2631578947368422e-06,
+      "loss": 0.2609,
+      "step": 9200
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 2.671875,
+      "learning_rate": 1.2473684210526317e-06,
+      "loss": 0.26,
+      "step": 9210
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 2.671875,
+      "learning_rate": 1.2315789473684212e-06,
+      "loss": 0.2548,
+      "step": 9220
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.2157894736842105e-06,
+      "loss": 0.2231,
+      "step": 9230
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.65625,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 0.2367,
+      "step": 9240
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.1842105263157894e-06,
+      "loss": 0.1978,
+      "step": 9250
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.59765625,
+      "learning_rate": 1.168421052631579e-06,
+      "loss": 0.1858,
+      "step": 9260
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.1526315789473684e-06,
+      "loss": 0.3199,
+      "step": 9270
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.9375,
+      "learning_rate": 1.136842105263158e-06,
+      "loss": 0.2436,
+      "step": 9280
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 9.125,
+      "learning_rate": 1.1210526315789474e-06,
+      "loss": 0.2774,
+      "step": 9290
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 4.75,
+      "learning_rate": 1.1052631578947367e-06,
+      "loss": 0.23,
+      "step": 9300
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.0894736842105264e-06,
+      "loss": 0.203,
+      "step": 9310
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.453125,
+      "learning_rate": 1.0736842105263157e-06,
+      "loss": 0.206,
+      "step": 9320
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.625,
+      "learning_rate": 1.0578947368421054e-06,
+      "loss": 0.1693,
+      "step": 9330
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.0421052631578946e-06,
+      "loss": 0.2524,
+      "step": 9340
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.953125,
+      "learning_rate": 1.0263157894736843e-06,
+      "loss": 0.2036,
+      "step": 9350
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 3.109375,
+      "learning_rate": 1.0105263157894736e-06,
+      "loss": 0.22,
+      "step": 9360
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.5,
+      "learning_rate": 9.947368421052633e-07,
+      "loss": 0.2724,
+      "step": 9370
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.5,
+      "learning_rate": 9.789473684210526e-07,
+      "loss": 0.3088,
+      "step": 9380
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 4.25,
+      "learning_rate": 9.63157894736842e-07,
+      "loss": 0.1978,
+      "step": 9390
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.0625,
+      "learning_rate": 9.473684210526316e-07,
+      "loss": 0.2206,
+      "step": 9400
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.40625,
+      "learning_rate": 9.315789473684211e-07,
+      "loss": 0.2956,
+      "step": 9410
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.0625,
+      "learning_rate": 9.157894736842105e-07,
+      "loss": 0.2731,
+      "step": 9420
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 4.71875,
+      "learning_rate": 9e-07,
+      "loss": 0.2478,
+      "step": 9430
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 3.125,
+      "learning_rate": 8.842105263157895e-07,
+      "loss": 0.2278,
+      "step": 9440
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.734375,
+      "learning_rate": 8.684210526315789e-07,
+      "loss": 0.2344,
+      "step": 9450
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.5625,
+      "learning_rate": 8.526315789473684e-07,
+      "loss": 0.2311,
+      "step": 9460
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 4.1875,
+      "learning_rate": 8.368421052631579e-07,
+      "loss": 0.3073,
+      "step": 9470
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.3125,
+      "learning_rate": 8.210526315789474e-07,
+      "loss": 0.2114,
+      "step": 9480
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 4.125,
+      "learning_rate": 8.052631578947369e-07,
+      "loss": 0.2151,
+      "step": 9490
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.328125,
+      "learning_rate": 7.894736842105263e-07,
+      "loss": 0.2408,
+      "step": 9500
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.65625,
+      "learning_rate": 7.736842105263158e-07,
+      "loss": 0.2003,
+      "step": 9510
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 2.796875,
+      "learning_rate": 7.578947368421053e-07,
+      "loss": 0.1941,
+      "step": 9520
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.4375,
+      "learning_rate": 7.421052631578948e-07,
+      "loss": 0.1859,
+      "step": 9530
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 2.859375,
+      "learning_rate": 7.263157894736843e-07,
+      "loss": 0.2542,
+      "step": 9540
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 2.796875,
+      "learning_rate": 7.105263157894737e-07,
+      "loss": 0.1741,
+      "step": 9550
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.90625,
+      "learning_rate": 6.947368421052632e-07,
+      "loss": 0.2572,
+      "step": 9560
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 2.171875,
+      "learning_rate": 6.789473684210527e-07,
+      "loss": 0.1887,
+      "step": 9570
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.21875,
+      "learning_rate": 6.631578947368421e-07,
+      "loss": 0.2924,
+      "step": 9580
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 3.125,
+      "learning_rate": 6.473684210526316e-07,
+      "loss": 0.2577,
+      "step": 9590
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.90625,
+      "learning_rate": 6.315789473684211e-07,
+      "loss": 0.2628,
+      "step": 9600
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.28125,
+      "learning_rate": 6.157894736842106e-07,
+      "loss": 0.2306,
+      "step": 9610
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 7.125,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 0.3013,
+      "step": 9620
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 3.859375,
+      "learning_rate": 5.842105263157895e-07,
+      "loss": 0.2663,
+      "step": 9630
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 3.640625,
+      "learning_rate": 5.68421052631579e-07,
+      "loss": 0.2698,
+      "step": 9640
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 9.1875,
+      "learning_rate": 5.526315789473683e-07,
+      "loss": 0.2508,
+      "step": 9650
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.625,
+      "learning_rate": 5.368421052631578e-07,
+      "loss": 0.2684,
+      "step": 9660
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.0625,
+      "learning_rate": 5.210526315789473e-07,
+      "loss": 0.2289,
+      "step": 9670
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.28125,
+      "learning_rate": 5.052631578947368e-07,
+      "loss": 0.2914,
+      "step": 9680
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.375,
+      "learning_rate": 4.894736842105263e-07,
+      "loss": 0.249,
+      "step": 9690
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.5625,
+      "learning_rate": 4.736842105263158e-07,
+      "loss": 0.225,
+      "step": 9700
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.59375,
+      "learning_rate": 4.5789473684210523e-07,
+      "loss": 0.231,
+      "step": 9710
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.65625,
+      "learning_rate": 4.421052631578947e-07,
+      "loss": 0.2524,
+      "step": 9720
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 4.34375,
+      "learning_rate": 4.263157894736842e-07,
+      "loss": 0.2368,
+      "step": 9730
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 7.1875,
+      "learning_rate": 4.105263157894737e-07,
+      "loss": 0.2755,
+      "step": 9740
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 3.1875,
+      "learning_rate": 3.9473684210526315e-07,
+      "loss": 0.2611,
+      "step": 9750
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 4.09375,
+      "learning_rate": 3.7894736842105264e-07,
+      "loss": 0.2149,
+      "step": 9760
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 2.046875,
+      "learning_rate": 3.6315789473684213e-07,
+      "loss": 0.238,
+      "step": 9770
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.0,
+      "learning_rate": 3.473684210526316e-07,
+      "loss": 0.2223,
+      "step": 9780
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.875,
+      "learning_rate": 3.3157894736842106e-07,
+      "loss": 0.2587,
+      "step": 9790
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 7.40625,
+      "learning_rate": 3.1578947368421055e-07,
+      "loss": 0.2041,
+      "step": 9800
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 3.375,
+      "learning_rate": 3.0000000000000004e-07,
+      "loss": 0.2226,
+      "step": 9810
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 4.4375,
+      "learning_rate": 2.842105263157895e-07,
+      "loss": 0.2312,
+      "step": 9820
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.875,
+      "learning_rate": 2.684210526315789e-07,
+      "loss": 0.3272,
+      "step": 9830
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 3.59375,
+      "learning_rate": 2.526315789473684e-07,
+      "loss": 0.2515,
+      "step": 9840
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 3.65625,
+      "learning_rate": 2.368421052631579e-07,
+      "loss": 0.2716,
+      "step": 9850
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.6875,
+      "learning_rate": 2.2105263157894736e-07,
+      "loss": 0.2898,
+      "step": 9860
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 3.0625,
+      "learning_rate": 2.0526315789473685e-07,
+      "loss": 0.2238,
+      "step": 9870
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.28125,
+      "learning_rate": 1.8947368421052632e-07,
+      "loss": 0.2226,
+      "step": 9880
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.736842105263158e-07,
+      "loss": 0.2741,
+      "step": 9890
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.5,
+      "learning_rate": 1.5789473684210527e-07,
+      "loss": 0.1991,
+      "step": 9900
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 5.21875,
+      "learning_rate": 1.4210526315789474e-07,
+      "loss": 0.2271,
+      "step": 9910
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 4.75,
+      "learning_rate": 1.263157894736842e-07,
+      "loss": 0.2208,
+      "step": 9920
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.1052631578947368e-07,
+      "loss": 0.2251,
+      "step": 9930
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 5.4375,
+      "learning_rate": 9.473684210526316e-08,
+      "loss": 0.1776,
+      "step": 9940
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 4.3125,
+      "learning_rate": 7.894736842105264e-08,
+      "loss": 0.26,
+      "step": 9950
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.09375,
+      "learning_rate": 6.31578947368421e-08,
+      "loss": 0.2694,
+      "step": 9960
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 7.5,
+      "learning_rate": 4.736842105263158e-08,
+      "loss": 0.3072,
+      "step": 9970
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 2.75,
+      "learning_rate": 3.157894736842105e-08,
+      "loss": 0.258,
+      "step": 9980
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.6953125,
+      "learning_rate": 1.5789473684210525e-08,
+      "loss": 0.2416,
+      "step": 9990
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 3.203125,
+      "learning_rate": 0.0,
+      "loss": 0.2035,
+      "step": 10000
+    },
+    {
+      "epoch": 0.5,
+      "eval_loss": 0.2470625340938568,
+      "eval_runtime": 65.0454,
+      "eval_samples_per_second": 15.374,
+      "eval_steps_per_second": 15.374,
+      "step": 10000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 2000,
+  "total_flos": 1.613922041856e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}