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": 3.5861574323112784,
+  "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": 5.03125,
+      "learning_rate": 3.0000000000000004e-07,
+      "loss": 1.4133,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 9.0,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 1.5815,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.5625,
+      "learning_rate": 9e-07,
+      "loss": 1.3232,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.875,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 1.3635,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.21875,
+      "learning_rate": 1.5e-06,
+      "loss": 1.3467,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.4375,
+      "learning_rate": 1.8e-06,
+      "loss": 1.3232,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.9375,
+      "learning_rate": 2.1000000000000002e-06,
+      "loss": 1.3657,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.0625,
+      "learning_rate": 2.4000000000000003e-06,
+      "loss": 1.3112,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.53125,
+      "learning_rate": 2.7e-06,
+      "loss": 1.3052,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.34375,
+      "learning_rate": 3e-06,
+      "loss": 1.3857,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.9375,
+      "learning_rate": 3.3e-06,
+      "loss": 1.4426,
+      "step": 110
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.875,
+      "learning_rate": 3.6e-06,
+      "loss": 1.2005,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.65625,
+      "learning_rate": 3.9e-06,
+      "loss": 1.2581,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.09375,
+      "learning_rate": 4.2000000000000004e-06,
+      "loss": 1.3267,
+      "step": 140
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.46875,
+      "learning_rate": 4.5e-06,
+      "loss": 1.3335,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.59375,
+      "learning_rate": 4.800000000000001e-06,
+      "loss": 1.2575,
+      "step": 160
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.8125,
+      "learning_rate": 5.1e-06,
+      "loss": 1.2955,
+      "step": 170
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.1875,
+      "learning_rate": 5.4e-06,
+      "loss": 1.1892,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.09375,
+      "learning_rate": 5.7000000000000005e-06,
+      "loss": 1.1546,
+      "step": 190
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.0,
+      "learning_rate": 6e-06,
+      "loss": 1.0558,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.65625,
+      "learning_rate": 6.3e-06,
+      "loss": 1.0267,
+      "step": 210
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.3125,
+      "learning_rate": 6.6e-06,
+      "loss": 1.1234,
+      "step": 220
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.0,
+      "learning_rate": 6.900000000000001e-06,
+      "loss": 1.0501,
+      "step": 230
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.15625,
+      "learning_rate": 7.2e-06,
+      "loss": 1.1591,
+      "step": 240
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.53125,
+      "learning_rate": 7.5e-06,
+      "loss": 0.9678,
+      "step": 250
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.0625,
+      "learning_rate": 7.8e-06,
+      "loss": 1.1697,
+      "step": 260
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.53125,
+      "learning_rate": 8.1e-06,
+      "loss": 0.8972,
+      "step": 270
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.78125,
+      "learning_rate": 8.400000000000001e-06,
+      "loss": 1.0512,
+      "step": 280
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.9375,
+      "learning_rate": 8.7e-06,
+      "loss": 0.9511,
+      "step": 290
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.09375,
+      "learning_rate": 9e-06,
+      "loss": 1.0023,
+      "step": 300
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.1875,
+      "learning_rate": 9.3e-06,
+      "loss": 1.0908,
+      "step": 310
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.875,
+      "learning_rate": 9.600000000000001e-06,
+      "loss": 1.0276,
+      "step": 320
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.15625,
+      "learning_rate": 9.9e-06,
+      "loss": 0.9734,
+      "step": 330
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.625,
+      "learning_rate": 1.02e-05,
+      "loss": 0.8525,
+      "step": 340
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.546875,
+      "learning_rate": 1.05e-05,
+      "loss": 1.0789,
+      "step": 350
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.21875,
+      "learning_rate": 1.08e-05,
+      "loss": 0.981,
+      "step": 360
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.5625,
+      "learning_rate": 1.11e-05,
+      "loss": 1.0341,
+      "step": 370
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.8125,
+      "learning_rate": 1.1400000000000001e-05,
+      "loss": 0.8214,
+      "step": 380
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.96875,
+      "learning_rate": 1.1700000000000001e-05,
+      "loss": 0.9675,
+      "step": 390
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.125,
+      "learning_rate": 1.2e-05,
+      "loss": 1.2041,
+      "step": 400
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.15625,
+      "learning_rate": 1.2299999999999999e-05,
+      "loss": 1.0061,
+      "step": 410
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.328125,
+      "learning_rate": 1.26e-05,
+      "loss": 0.8642,
+      "step": 420
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.29e-05,
+      "loss": 1.0072,
+      "step": 430
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.34375,
+      "learning_rate": 1.32e-05,
+      "loss": 0.9096,
+      "step": 440
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.125,
+      "learning_rate": 1.3500000000000001e-05,
+      "loss": 1.002,
+      "step": 450
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.78125,
+      "learning_rate": 1.3800000000000002e-05,
+      "loss": 0.8901,
+      "step": 460
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.3125,
+      "learning_rate": 1.4099999999999999e-05,
+      "loss": 0.8891,
+      "step": 470
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.44e-05,
+      "loss": 0.9389,
+      "step": 480
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.28125,
+      "learning_rate": 1.47e-05,
+      "loss": 0.8838,
+      "step": 490
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.765625,
+      "learning_rate": 1.5e-05,
+      "loss": 1.0191,
+      "step": 500
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.328125,
+      "learning_rate": 1.4984210526315789e-05,
+      "loss": 0.9197,
+      "step": 510
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.90625,
+      "learning_rate": 1.496842105263158e-05,
+      "loss": 0.9614,
+      "step": 520
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.84375,
+      "learning_rate": 1.4952631578947368e-05,
+      "loss": 0.8612,
+      "step": 530
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 3.765625,
+      "learning_rate": 1.4936842105263158e-05,
+      "loss": 0.91,
+      "step": 540
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.15625,
+      "learning_rate": 1.4921052631578947e-05,
+      "loss": 0.8851,
+      "step": 550
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.4905263157894737e-05,
+      "loss": 0.9448,
+      "step": 560
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.859375,
+      "learning_rate": 1.4889473684210526e-05,
+      "loss": 0.8091,
+      "step": 570
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.4873684210526315e-05,
+      "loss": 0.8932,
+      "step": 580
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.375,
+      "learning_rate": 1.4857894736842107e-05,
+      "loss": 0.966,
+      "step": 590
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 3.875,
+      "learning_rate": 1.4842105263157895e-05,
+      "loss": 0.8036,
+      "step": 600
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 3.9375,
+      "learning_rate": 1.4826315789473686e-05,
+      "loss": 0.9743,
+      "step": 610
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 3.09375,
+      "learning_rate": 1.4810526315789474e-05,
+      "loss": 0.8383,
+      "step": 620
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.4794736842105265e-05,
+      "loss": 0.8982,
+      "step": 630
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.90625,
+      "learning_rate": 1.4778947368421053e-05,
+      "loss": 0.8112,
+      "step": 640
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.96875,
+      "learning_rate": 1.4763157894736842e-05,
+      "loss": 0.8426,
+      "step": 650
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.4747368421052632e-05,
+      "loss": 0.8399,
+      "step": 660
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.125,
+      "learning_rate": 1.4731578947368421e-05,
+      "loss": 0.9043,
+      "step": 670
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.5,
+      "learning_rate": 1.4715789473684211e-05,
+      "loss": 0.8521,
+      "step": 680
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.46875,
+      "learning_rate": 1.47e-05,
+      "loss": 0.9079,
+      "step": 690
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.468421052631579e-05,
+      "loss": 0.7998,
+      "step": 700
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.46875,
+      "learning_rate": 1.4668421052631579e-05,
+      "loss": 0.8204,
+      "step": 710
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 3.3125,
+      "learning_rate": 1.4652631578947367e-05,
+      "loss": 0.8764,
+      "step": 720
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.28125,
+      "learning_rate": 1.4636842105263158e-05,
+      "loss": 0.9226,
+      "step": 730
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 3.921875,
+      "learning_rate": 1.4621052631578946e-05,
+      "loss": 0.8921,
+      "step": 740
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 3.671875,
+      "learning_rate": 1.4605263157894737e-05,
+      "loss": 0.8764,
+      "step": 750
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 3.671875,
+      "learning_rate": 1.4589473684210527e-05,
+      "loss": 0.8475,
+      "step": 760
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.921875,
+      "learning_rate": 1.4573684210526317e-05,
+      "loss": 0.9297,
+      "step": 770
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.109375,
+      "learning_rate": 1.4557894736842106e-05,
+      "loss": 0.9052,
+      "step": 780
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.4542105263157895e-05,
+      "loss": 0.8695,
+      "step": 790
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 4.25,
+      "learning_rate": 1.4526315789473685e-05,
+      "loss": 0.9214,
+      "step": 800
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.25,
+      "learning_rate": 1.4510526315789474e-05,
+      "loss": 0.8403,
+      "step": 810
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.046875,
+      "learning_rate": 1.4494736842105264e-05,
+      "loss": 0.918,
+      "step": 820
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.53125,
+      "learning_rate": 1.4478947368421053e-05,
+      "loss": 0.8599,
+      "step": 830
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.875,
+      "learning_rate": 1.4463157894736843e-05,
+      "loss": 0.954,
+      "step": 840
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.4447368421052632e-05,
+      "loss": 0.8568,
+      "step": 850
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.443157894736842e-05,
+      "loss": 0.9043,
+      "step": 860
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 4.0,
+      "learning_rate": 1.441578947368421e-05,
+      "loss": 0.9108,
+      "step": 870
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.44e-05,
+      "loss": 0.8723,
+      "step": 880
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 3.859375,
+      "learning_rate": 1.438421052631579e-05,
+      "loss": 0.8514,
+      "step": 890
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 2.59375,
+      "learning_rate": 1.4368421052631578e-05,
+      "loss": 0.8584,
+      "step": 900
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 3.9375,
+      "learning_rate": 1.4352631578947369e-05,
+      "loss": 0.8763,
+      "step": 910
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.4336842105263159e-05,
+      "loss": 0.7777,
+      "step": 920
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 3.671875,
+      "learning_rate": 1.4321052631578948e-05,
+      "loss": 0.9353,
+      "step": 930
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 3.4375,
+      "learning_rate": 1.4305263157894738e-05,
+      "loss": 0.8708,
+      "step": 940
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 3.828125,
+      "learning_rate": 1.4289473684210527e-05,
+      "loss": 0.8801,
+      "step": 950
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 3.5,
+      "learning_rate": 1.4273684210526317e-05,
+      "loss": 0.8625,
+      "step": 960
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.03125,
+      "learning_rate": 1.4257894736842106e-05,
+      "loss": 0.875,
+      "step": 970
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.21875,
+      "learning_rate": 1.4242105263157896e-05,
+      "loss": 0.9445,
+      "step": 980
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 3.421875,
+      "learning_rate": 1.4226315789473685e-05,
+      "loss": 0.7317,
+      "step": 990
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 3.6875,
+      "learning_rate": 1.4210526315789473e-05,
+      "loss": 0.8303,
+      "step": 1000
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.4194736842105264e-05,
+      "loss": 0.8433,
+      "step": 1010
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 3.484375,
+      "learning_rate": 1.4178947368421052e-05,
+      "loss": 0.8713,
+      "step": 1020
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 3.890625,
+      "learning_rate": 1.4163157894736843e-05,
+      "loss": 0.8411,
+      "step": 1030
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.4147368421052631e-05,
+      "loss": 0.8757,
+      "step": 1040
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.3125,
+      "learning_rate": 1.4131578947368422e-05,
+      "loss": 0.7574,
+      "step": 1050
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.21875,
+      "learning_rate": 1.411578947368421e-05,
+      "loss": 0.8089,
+      "step": 1060
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 2.890625,
+      "learning_rate": 1.4099999999999999e-05,
+      "loss": 0.7259,
+      "step": 1070
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.125,
+      "learning_rate": 1.408421052631579e-05,
+      "loss": 0.7544,
+      "step": 1080
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.375,
+      "learning_rate": 1.406842105263158e-05,
+      "loss": 0.9522,
+      "step": 1090
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.78125,
+      "learning_rate": 1.405263157894737e-05,
+      "loss": 0.8574,
+      "step": 1100
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.4036842105263158e-05,
+      "loss": 0.9539,
+      "step": 1110
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 3.34375,
+      "learning_rate": 1.4021052631578949e-05,
+      "loss": 0.8365,
+      "step": 1120
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.34375,
+      "learning_rate": 1.4005263157894737e-05,
+      "loss": 0.7868,
+      "step": 1130
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 3.515625,
+      "learning_rate": 1.3989473684210526e-05,
+      "loss": 0.9054,
+      "step": 1140
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 3.015625,
+      "learning_rate": 1.3973684210526316e-05,
+      "loss": 0.841,
+      "step": 1150
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.3957894736842105e-05,
+      "loss": 0.8132,
+      "step": 1160
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 3.515625,
+      "learning_rate": 1.3942105263157895e-05,
+      "loss": 0.8678,
+      "step": 1170
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 3.4375,
+      "learning_rate": 1.3926315789473684e-05,
+      "loss": 0.8434,
+      "step": 1180
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.3910526315789474e-05,
+      "loss": 0.7466,
+      "step": 1190
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 3.5625,
+      "learning_rate": 1.3894736842105263e-05,
+      "loss": 0.9028,
+      "step": 1200
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 3.203125,
+      "learning_rate": 1.3878947368421052e-05,
+      "loss": 0.9181,
+      "step": 1210
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 3.703125,
+      "learning_rate": 1.3863157894736842e-05,
+      "loss": 0.9427,
+      "step": 1220
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 2.609375,
+      "learning_rate": 1.384736842105263e-05,
+      "loss": 0.7861,
+      "step": 1230
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.71875,
+      "learning_rate": 1.3831578947368421e-05,
+      "loss": 0.8857,
+      "step": 1240
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 3.140625,
+      "learning_rate": 1.3815789473684211e-05,
+      "loss": 0.8447,
+      "step": 1250
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 3.140625,
+      "learning_rate": 1.3800000000000002e-05,
+      "loss": 0.7963,
+      "step": 1260
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.378421052631579e-05,
+      "loss": 0.7099,
+      "step": 1270
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 3.765625,
+      "learning_rate": 1.3768421052631579e-05,
+      "loss": 0.8737,
+      "step": 1280
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 3.921875,
+      "learning_rate": 1.375263157894737e-05,
+      "loss": 0.8896,
+      "step": 1290
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 3.640625,
+      "learning_rate": 1.3736842105263158e-05,
+      "loss": 0.8224,
+      "step": 1300
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 3.796875,
+      "learning_rate": 1.3721052631578948e-05,
+      "loss": 0.7736,
+      "step": 1310
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 4.65625,
+      "learning_rate": 1.3705263157894737e-05,
+      "loss": 0.8416,
+      "step": 1320
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.3689473684210527e-05,
+      "loss": 0.7371,
+      "step": 1330
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 3.3125,
+      "learning_rate": 1.3673684210526316e-05,
+      "loss": 0.9159,
+      "step": 1340
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.625,
+      "learning_rate": 1.3657894736842106e-05,
+      "loss": 0.8392,
+      "step": 1350
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 3.609375,
+      "learning_rate": 1.3642105263157895e-05,
+      "loss": 0.7774,
+      "step": 1360
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.3626315789473684e-05,
+      "loss": 0.9497,
+      "step": 1370
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.3610526315789474e-05,
+      "loss": 0.8797,
+      "step": 1380
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 2.265625,
+      "learning_rate": 1.3594736842105263e-05,
+      "loss": 0.8666,
+      "step": 1390
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 3.96875,
+      "learning_rate": 1.3578947368421053e-05,
+      "loss": 0.9691,
+      "step": 1400
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 5.21875,
+      "learning_rate": 1.3563157894736842e-05,
+      "loss": 0.796,
+      "step": 1410
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 2.296875,
+      "learning_rate": 1.3547368421052634e-05,
+      "loss": 0.8152,
+      "step": 1420
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.3531578947368422e-05,
+      "loss": 0.8702,
+      "step": 1430
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 3.765625,
+      "learning_rate": 1.3515789473684211e-05,
+      "loss": 0.8582,
+      "step": 1440
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.3500000000000001e-05,
+      "loss": 0.8282,
+      "step": 1450
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 3.859375,
+      "learning_rate": 1.348421052631579e-05,
+      "loss": 0.9256,
+      "step": 1460
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.346842105263158e-05,
+      "loss": 0.9328,
+      "step": 1470
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 3.5,
+      "learning_rate": 1.3452631578947369e-05,
+      "loss": 0.8571,
+      "step": 1480
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 3.609375,
+      "learning_rate": 1.343684210526316e-05,
+      "loss": 0.8293,
+      "step": 1490
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 2.96875,
+      "learning_rate": 1.3421052631578948e-05,
+      "loss": 0.8282,
+      "step": 1500
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 3.8125,
+      "learning_rate": 1.3405263157894736e-05,
+      "loss": 0.8202,
+      "step": 1510
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 3.609375,
+      "learning_rate": 1.3389473684210527e-05,
+      "loss": 0.9545,
+      "step": 1520
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 4.625,
+      "learning_rate": 1.3373684210526315e-05,
+      "loss": 0.8083,
+      "step": 1530
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 3.84375,
+      "learning_rate": 1.3357894736842106e-05,
+      "loss": 0.8865,
+      "step": 1540
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 3.875,
+      "learning_rate": 1.3342105263157894e-05,
+      "loss": 0.7867,
+      "step": 1550
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 3.203125,
+      "learning_rate": 1.3326315789473685e-05,
+      "loss": 0.7882,
+      "step": 1560
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 3.90625,
+      "learning_rate": 1.3310526315789473e-05,
+      "loss": 0.7837,
+      "step": 1570
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 5.125,
+      "learning_rate": 1.3294736842105262e-05,
+      "loss": 0.9198,
+      "step": 1580
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 3.4375,
+      "learning_rate": 1.3278947368421054e-05,
+      "loss": 0.849,
+      "step": 1590
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.3263157894736843e-05,
+      "loss": 0.894,
+      "step": 1600
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 2.953125,
+      "learning_rate": 1.3247368421052633e-05,
+      "loss": 0.8318,
+      "step": 1610
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 3.703125,
+      "learning_rate": 1.3231578947368422e-05,
+      "loss": 0.7787,
+      "step": 1620
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 5.0,
+      "learning_rate": 1.3215789473684212e-05,
+      "loss": 0.8853,
+      "step": 1630
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.32e-05,
+      "loss": 0.9163,
+      "step": 1640
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 3.6875,
+      "learning_rate": 1.318421052631579e-05,
+      "loss": 0.8755,
+      "step": 1650
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 3.71875,
+      "learning_rate": 1.316842105263158e-05,
+      "loss": 0.8891,
+      "step": 1660
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 3.984375,
+      "learning_rate": 1.3152631578947368e-05,
+      "loss": 0.8621,
+      "step": 1670
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 3.28125,
+      "learning_rate": 1.3136842105263159e-05,
+      "loss": 0.8613,
+      "step": 1680
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.3121052631578947e-05,
+      "loss": 0.8886,
+      "step": 1690
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 3.53125,
+      "learning_rate": 1.3105263157894738e-05,
+      "loss": 0.843,
+      "step": 1700
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 3.75,
+      "learning_rate": 1.3089473684210526e-05,
+      "loss": 0.8877,
+      "step": 1710
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.3073684210526315e-05,
+      "loss": 0.8785,
+      "step": 1720
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 3.78125,
+      "learning_rate": 1.3057894736842105e-05,
+      "loss": 0.8878,
+      "step": 1730
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 3.828125,
+      "learning_rate": 1.3042105263157894e-05,
+      "loss": 0.8794,
+      "step": 1740
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 4.75,
+      "learning_rate": 1.3026315789473684e-05,
+      "loss": 0.8821,
+      "step": 1750
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 4.34375,
+      "learning_rate": 1.3010526315789475e-05,
+      "loss": 0.8146,
+      "step": 1760
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 3.953125,
+      "learning_rate": 1.2994736842105265e-05,
+      "loss": 0.8322,
+      "step": 1770
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 2.8125,
+      "learning_rate": 1.2978947368421054e-05,
+      "loss": 0.8197,
+      "step": 1780
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 4.59375,
+      "learning_rate": 1.2963157894736842e-05,
+      "loss": 0.8883,
+      "step": 1790
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 3.609375,
+      "learning_rate": 1.2947368421052633e-05,
+      "loss": 0.9259,
+      "step": 1800
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 4.125,
+      "learning_rate": 1.2931578947368421e-05,
+      "loss": 0.7667,
+      "step": 1810
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 5.3125,
+      "learning_rate": 1.2915789473684212e-05,
+      "loss": 0.9434,
+      "step": 1820
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 3.3125,
+      "learning_rate": 1.29e-05,
+      "loss": 0.8081,
+      "step": 1830
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 3.875,
+      "learning_rate": 1.288421052631579e-05,
+      "loss": 0.767,
+      "step": 1840
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.2868421052631579e-05,
+      "loss": 0.8339,
+      "step": 1850
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 4.0,
+      "learning_rate": 1.2852631578947368e-05,
+      "loss": 0.8083,
+      "step": 1860
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.2836842105263158e-05,
+      "loss": 0.8255,
+      "step": 1870
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 3.65625,
+      "learning_rate": 1.2821052631578947e-05,
+      "loss": 0.8618,
+      "step": 1880
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 4.15625,
+      "learning_rate": 1.2805263157894737e-05,
+      "loss": 0.913,
+      "step": 1890
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 4.125,
+      "learning_rate": 1.2789473684210526e-05,
+      "loss": 0.7233,
+      "step": 1900
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 4.5625,
+      "learning_rate": 1.2773684210526316e-05,
+      "loss": 0.839,
+      "step": 1910
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.2757894736842106e-05,
+      "loss": 0.9282,
+      "step": 1920
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 3.3125,
+      "learning_rate": 1.2742105263157895e-05,
+      "loss": 0.8012,
+      "step": 1930
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 6.3125,
+      "learning_rate": 1.2726315789473685e-05,
+      "loss": 0.7934,
+      "step": 1940
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 2.921875,
+      "learning_rate": 1.2710526315789474e-05,
+      "loss": 0.8342,
+      "step": 1950
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 2.96875,
+      "learning_rate": 1.2694736842105264e-05,
+      "loss": 0.8319,
+      "step": 1960
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 3.4375,
+      "learning_rate": 1.2678947368421053e-05,
+      "loss": 0.855,
+      "step": 1970
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.2663157894736843e-05,
+      "loss": 0.8794,
+      "step": 1980
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.2647368421052632e-05,
+      "loss": 0.8106,
+      "step": 1990
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 4.84375,
+      "learning_rate": 1.263157894736842e-05,
+      "loss": 0.7944,
+      "step": 2000
+    },
+    {
+      "epoch": 0.72,
+      "eval_loss": 0.964277446269989,
+      "eval_runtime": 64.3098,
+      "eval_samples_per_second": 15.55,
+      "eval_steps_per_second": 15.55,
+      "step": 2000
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 3.890625,
+      "learning_rate": 1.2615789473684211e-05,
+      "loss": 0.7848,
+      "step": 2010
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.26e-05,
+      "loss": 0.9305,
+      "step": 2020
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 3.125,
+      "learning_rate": 1.258421052631579e-05,
+      "loss": 0.8509,
+      "step": 2030
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 3.078125,
+      "learning_rate": 1.2568421052631579e-05,
+      "loss": 0.7565,
+      "step": 2040
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 7.125,
+      "learning_rate": 1.2552631578947369e-05,
+      "loss": 0.892,
+      "step": 2050
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 4.625,
+      "learning_rate": 1.2536842105263158e-05,
+      "loss": 0.7951,
+      "step": 2060
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 2.90625,
+      "learning_rate": 1.2521052631578946e-05,
+      "loss": 0.7561,
+      "step": 2070
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 3.15625,
+      "learning_rate": 1.2505263157894737e-05,
+      "loss": 0.8431,
+      "step": 2080
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 3.875,
+      "learning_rate": 1.2489473684210527e-05,
+      "loss": 0.8761,
+      "step": 2090
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.2473684210526317e-05,
+      "loss": 0.8271,
+      "step": 2100
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 3.796875,
+      "learning_rate": 1.2457894736842106e-05,
+      "loss": 0.8749,
+      "step": 2110
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 2.578125,
+      "learning_rate": 1.2442105263157896e-05,
+      "loss": 0.8827,
+      "step": 2120
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 3.390625,
+      "learning_rate": 1.2426315789473685e-05,
+      "loss": 0.8542,
+      "step": 2130
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 3.34375,
+      "learning_rate": 1.2410526315789474e-05,
+      "loss": 0.8718,
+      "step": 2140
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 3.234375,
+      "learning_rate": 1.2394736842105264e-05,
+      "loss": 0.804,
+      "step": 2150
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 4.375,
+      "learning_rate": 1.2378947368421053e-05,
+      "loss": 0.6931,
+      "step": 2160
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.2363157894736843e-05,
+      "loss": 0.8037,
+      "step": 2170
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 3.09375,
+      "learning_rate": 1.2347368421052631e-05,
+      "loss": 0.8514,
+      "step": 2180
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 3.65625,
+      "learning_rate": 1.2331578947368422e-05,
+      "loss": 0.7054,
+      "step": 2190
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 4.125,
+      "learning_rate": 1.231578947368421e-05,
+      "loss": 0.8655,
+      "step": 2200
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 4.59375,
+      "learning_rate": 1.2299999999999999e-05,
+      "loss": 0.7915,
+      "step": 2210
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 3.875,
+      "learning_rate": 1.228421052631579e-05,
+      "loss": 0.7686,
+      "step": 2220
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 4.25,
+      "learning_rate": 1.2268421052631578e-05,
+      "loss": 0.8301,
+      "step": 2230
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 2.953125,
+      "learning_rate": 1.2252631578947368e-05,
+      "loss": 0.8714,
+      "step": 2240
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 3.1875,
+      "learning_rate": 1.2236842105263159e-05,
+      "loss": 0.7753,
+      "step": 2250
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 3.15625,
+      "learning_rate": 1.2221052631578949e-05,
+      "loss": 0.763,
+      "step": 2260
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 3.015625,
+      "learning_rate": 1.2205263157894738e-05,
+      "loss": 0.862,
+      "step": 2270
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 4.59375,
+      "learning_rate": 1.2189473684210526e-05,
+      "loss": 0.7985,
+      "step": 2280
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.2173684210526317e-05,
+      "loss": 0.8335,
+      "step": 2290
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 3.140625,
+      "learning_rate": 1.2157894736842105e-05,
+      "loss": 0.9334,
+      "step": 2300
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.2142105263157896e-05,
+      "loss": 0.8609,
+      "step": 2310
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 4.25,
+      "learning_rate": 1.2126315789473684e-05,
+      "loss": 0.8096,
+      "step": 2320
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 3.1875,
+      "learning_rate": 1.2110526315789475e-05,
+      "loss": 0.9687,
+      "step": 2330
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.2094736842105263e-05,
+      "loss": 0.842,
+      "step": 2340
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 3.703125,
+      "learning_rate": 1.2078947368421052e-05,
+      "loss": 0.7848,
+      "step": 2350
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 3.625,
+      "learning_rate": 1.2063157894736842e-05,
+      "loss": 0.8813,
+      "step": 2360
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 3.3125,
+      "learning_rate": 1.2047368421052631e-05,
+      "loss": 0.8663,
+      "step": 2370
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 3.625,
+      "learning_rate": 1.2031578947368421e-05,
+      "loss": 0.8555,
+      "step": 2380
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 3.25,
+      "learning_rate": 1.201578947368421e-05,
+      "loss": 0.773,
+      "step": 2390
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 5.03125,
+      "learning_rate": 1.2e-05,
+      "loss": 0.8283,
+      "step": 2400
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 4.375,
+      "learning_rate": 1.1984210526315789e-05,
+      "loss": 0.8516,
+      "step": 2410
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 3.546875,
+      "learning_rate": 1.196842105263158e-05,
+      "loss": 0.8842,
+      "step": 2420
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 3.65625,
+      "learning_rate": 1.195263157894737e-05,
+      "loss": 0.8055,
+      "step": 2430
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.1936842105263158e-05,
+      "loss": 0.9313,
+      "step": 2440
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 3.515625,
+      "learning_rate": 1.1921052631578949e-05,
+      "loss": 0.8737,
+      "step": 2450
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 5.71875,
+      "learning_rate": 1.1905263157894737e-05,
+      "loss": 0.9073,
+      "step": 2460
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 3.46875,
+      "learning_rate": 1.1889473684210528e-05,
+      "loss": 0.8106,
+      "step": 2470
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 3.09375,
+      "learning_rate": 1.1873684210526316e-05,
+      "loss": 0.8151,
+      "step": 2480
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 3.5625,
+      "learning_rate": 1.1857894736842105e-05,
+      "loss": 0.7122,
+      "step": 2490
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 4.71875,
+      "learning_rate": 1.1842105263157895e-05,
+      "loss": 0.8079,
+      "step": 2500
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 3.1875,
+      "learning_rate": 1.1826315789473684e-05,
+      "loss": 0.8578,
+      "step": 2510
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.1810526315789474e-05,
+      "loss": 0.7484,
+      "step": 2520
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 2.609375,
+      "learning_rate": 1.1794736842105263e-05,
+      "loss": 0.8379,
+      "step": 2530
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 5.125,
+      "learning_rate": 1.1778947368421053e-05,
+      "loss": 0.6939,
+      "step": 2540
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 3.78125,
+      "learning_rate": 1.1763157894736842e-05,
+      "loss": 0.8335,
+      "step": 2550
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 3.8125,
+      "learning_rate": 1.174736842105263e-05,
+      "loss": 0.8134,
+      "step": 2560
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 4.375,
+      "learning_rate": 1.173157894736842e-05,
+      "loss": 0.7114,
+      "step": 2570
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 3.9375,
+      "learning_rate": 1.171578947368421e-05,
+      "loss": 0.8255,
+      "step": 2580
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 3.09375,
+      "learning_rate": 1.1700000000000001e-05,
+      "loss": 0.962,
+      "step": 2590
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 3.890625,
+      "learning_rate": 1.168421052631579e-05,
+      "loss": 0.7826,
+      "step": 2600
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 2.953125,
+      "learning_rate": 1.166842105263158e-05,
+      "loss": 0.7567,
+      "step": 2610
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 3.109375,
+      "learning_rate": 1.1652631578947369e-05,
+      "loss": 0.8709,
+      "step": 2620
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 3.71875,
+      "learning_rate": 1.1636842105263158e-05,
+      "loss": 0.8485,
+      "step": 2630
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 4.125,
+      "learning_rate": 1.1621052631578948e-05,
+      "loss": 0.7735,
+      "step": 2640
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.1605263157894737e-05,
+      "loss": 0.7513,
+      "step": 2650
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 3.484375,
+      "learning_rate": 1.1589473684210527e-05,
+      "loss": 0.8615,
+      "step": 2660
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 3.90625,
+      "learning_rate": 1.1573684210526316e-05,
+      "loss": 0.7475,
+      "step": 2670
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 3.078125,
+      "learning_rate": 1.1557894736842106e-05,
+      "loss": 0.7589,
+      "step": 2680
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 4.8125,
+      "learning_rate": 1.1542105263157895e-05,
+      "loss": 0.7037,
+      "step": 2690
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 2.96875,
+      "learning_rate": 1.1526315789473683e-05,
+      "loss": 0.7924,
+      "step": 2700
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 3.140625,
+      "learning_rate": 1.1510526315789474e-05,
+      "loss": 0.7881,
+      "step": 2710
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.1494736842105262e-05,
+      "loss": 0.8313,
+      "step": 2720
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 3.3125,
+      "learning_rate": 1.1478947368421053e-05,
+      "loss": 0.7263,
+      "step": 2730
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.1463157894736841e-05,
+      "loss": 0.894,
+      "step": 2740
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.1447368421052632e-05,
+      "loss": 0.8584,
+      "step": 2750
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 3.921875,
+      "learning_rate": 1.1431578947368422e-05,
+      "loss": 0.8449,
+      "step": 2760
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 3.0625,
+      "learning_rate": 1.141578947368421e-05,
+      "loss": 0.7279,
+      "step": 2770
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 4.625,
+      "learning_rate": 1.1400000000000001e-05,
+      "loss": 0.8696,
+      "step": 2780
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 3.875,
+      "learning_rate": 1.138421052631579e-05,
+      "loss": 0.7332,
+      "step": 2790
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 2.875,
+      "learning_rate": 1.136842105263158e-05,
+      "loss": 0.9731,
+      "step": 2800
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 3.875,
+      "learning_rate": 1.1352631578947369e-05,
+      "loss": 0.7976,
+      "step": 2810
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 3.4375,
+      "learning_rate": 1.1336842105263159e-05,
+      "loss": 0.7155,
+      "step": 2820
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 3.75,
+      "learning_rate": 1.1321052631578948e-05,
+      "loss": 0.6819,
+      "step": 2830
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.1305263157894736e-05,
+      "loss": 0.8241,
+      "step": 2840
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 3.0625,
+      "learning_rate": 1.1289473684210527e-05,
+      "loss": 0.8655,
+      "step": 2850
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 3.234375,
+      "learning_rate": 1.1273684210526315e-05,
+      "loss": 0.7401,
+      "step": 2860
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.1257894736842106e-05,
+      "loss": 0.8938,
+      "step": 2870
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 3.796875,
+      "learning_rate": 1.1242105263157894e-05,
+      "loss": 0.9024,
+      "step": 2880
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.1226315789473685e-05,
+      "loss": 0.7119,
+      "step": 2890
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.1210526315789473e-05,
+      "loss": 0.8225,
+      "step": 2900
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 3.4375,
+      "learning_rate": 1.1194736842105264e-05,
+      "loss": 0.7932,
+      "step": 2910
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 3.96875,
+      "learning_rate": 1.1178947368421054e-05,
+      "loss": 0.8156,
+      "step": 2920
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 5.15625,
+      "learning_rate": 1.1163157894736842e-05,
+      "loss": 0.7501,
+      "step": 2930
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 4.5,
+      "learning_rate": 1.1147368421052633e-05,
+      "loss": 0.8522,
+      "step": 2940
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 3.21875,
+      "learning_rate": 1.1131578947368421e-05,
+      "loss": 0.7594,
+      "step": 2950
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 2.609375,
+      "learning_rate": 1.1115789473684212e-05,
+      "loss": 0.7883,
+      "step": 2960
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 3.453125,
+      "learning_rate": 1.11e-05,
+      "loss": 0.7622,
+      "step": 2970
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.108421052631579e-05,
+      "loss": 0.7797,
+      "step": 2980
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 3.78125,
+      "learning_rate": 1.106842105263158e-05,
+      "loss": 0.7898,
+      "step": 2990
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 4.0,
+      "learning_rate": 1.1052631578947368e-05,
+      "loss": 0.8001,
+      "step": 3000
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 3.59375,
+      "learning_rate": 1.1036842105263158e-05,
+      "loss": 0.8653,
+      "step": 3010
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 3.921875,
+      "learning_rate": 1.1021052631578947e-05,
+      "loss": 0.9141,
+      "step": 3020
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.1005263157894737e-05,
+      "loss": 0.8174,
+      "step": 3030
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 4.0,
+      "learning_rate": 1.0989473684210526e-05,
+      "loss": 0.8196,
+      "step": 3040
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 3.8125,
+      "learning_rate": 1.0973684210526316e-05,
+      "loss": 0.7277,
+      "step": 3050
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 4.375,
+      "learning_rate": 1.0957894736842105e-05,
+      "loss": 0.7426,
+      "step": 3060
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.0942105263157894e-05,
+      "loss": 0.7751,
+      "step": 3070
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 3.140625,
+      "learning_rate": 1.0926315789473684e-05,
+      "loss": 0.7859,
+      "step": 3080
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 2.96875,
+      "learning_rate": 1.0910526315789474e-05,
+      "loss": 0.7445,
+      "step": 3090
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 3.84375,
+      "learning_rate": 1.0894736842105265e-05,
+      "loss": 0.7919,
+      "step": 3100
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 3.734375,
+      "learning_rate": 1.0878947368421053e-05,
+      "loss": 0.8279,
+      "step": 3110
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.0863157894736844e-05,
+      "loss": 0.7749,
+      "step": 3120
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 4.875,
+      "learning_rate": 1.0847368421052632e-05,
+      "loss": 0.9269,
+      "step": 3130
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 3.53125,
+      "learning_rate": 1.0831578947368421e-05,
+      "loss": 0.819,
+      "step": 3140
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 3.40625,
+      "learning_rate": 1.0815789473684211e-05,
+      "loss": 0.7356,
+      "step": 3150
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.08e-05,
+      "loss": 0.8045,
+      "step": 3160
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 4.125,
+      "learning_rate": 1.078421052631579e-05,
+      "loss": 0.8847,
+      "step": 3170
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 3.78125,
+      "learning_rate": 1.0768421052631579e-05,
+      "loss": 0.8183,
+      "step": 3180
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 3.46875,
+      "learning_rate": 1.075263157894737e-05,
+      "loss": 0.756,
+      "step": 3190
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 4.0,
+      "learning_rate": 1.0736842105263158e-05,
+      "loss": 0.7879,
+      "step": 3200
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.0721052631578947e-05,
+      "loss": 0.8386,
+      "step": 3210
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 3.09375,
+      "learning_rate": 1.0705263157894737e-05,
+      "loss": 0.798,
+      "step": 3220
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 4.96875,
+      "learning_rate": 1.0689473684210526e-05,
+      "loss": 0.8175,
+      "step": 3230
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 3.9375,
+      "learning_rate": 1.0673684210526316e-05,
+      "loss": 0.7822,
+      "step": 3240
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 3.09375,
+      "learning_rate": 1.0657894736842106e-05,
+      "loss": 0.6839,
+      "step": 3250
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 3.90625,
+      "learning_rate": 1.0642105263157897e-05,
+      "loss": 0.67,
+      "step": 3260
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.0626315789473685e-05,
+      "loss": 0.9422,
+      "step": 3270
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 3.890625,
+      "learning_rate": 1.0610526315789474e-05,
+      "loss": 0.7891,
+      "step": 3280
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.0594736842105264e-05,
+      "loss": 0.7829,
+      "step": 3290
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.0578947368421053e-05,
+      "loss": 0.7444,
+      "step": 3300
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.0563157894736843e-05,
+      "loss": 0.8278,
+      "step": 3310
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 4.34375,
+      "learning_rate": 1.0547368421052632e-05,
+      "loss": 0.8772,
+      "step": 3320
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 2.40625,
+      "learning_rate": 1.0531578947368422e-05,
+      "loss": 0.757,
+      "step": 3330
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 3.90625,
+      "learning_rate": 1.051578947368421e-05,
+      "loss": 0.8061,
+      "step": 3340
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 3.65625,
+      "learning_rate": 1.05e-05,
+      "loss": 0.7796,
+      "step": 3350
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.048421052631579e-05,
+      "loss": 0.7227,
+      "step": 3360
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 3.984375,
+      "learning_rate": 1.0468421052631578e-05,
+      "loss": 0.8646,
+      "step": 3370
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 4.125,
+      "learning_rate": 1.0452631578947369e-05,
+      "loss": 0.7302,
+      "step": 3380
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.0436842105263157e-05,
+      "loss": 0.7628,
+      "step": 3390
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 3.875,
+      "learning_rate": 1.0421052631578948e-05,
+      "loss": 0.7485,
+      "step": 3400
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 3.171875,
+      "learning_rate": 1.0405263157894736e-05,
+      "loss": 0.8089,
+      "step": 3410
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 3.8125,
+      "learning_rate": 1.0389473684210527e-05,
+      "loss": 0.7009,
+      "step": 3420
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 3.421875,
+      "learning_rate": 1.0373684210526317e-05,
+      "loss": 0.7743,
+      "step": 3430
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 3.828125,
+      "learning_rate": 1.0357894736842106e-05,
+      "loss": 0.7733,
+      "step": 3440
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 3.390625,
+      "learning_rate": 1.0342105263157896e-05,
+      "loss": 0.7991,
+      "step": 3450
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 3.125,
+      "learning_rate": 1.0326315789473685e-05,
+      "loss": 0.646,
+      "step": 3460
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 3.359375,
+      "learning_rate": 1.0310526315789475e-05,
+      "loss": 0.8139,
+      "step": 3470
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 2.90625,
+      "learning_rate": 1.0294736842105264e-05,
+      "loss": 0.7434,
+      "step": 3480
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.0278947368421052e-05,
+      "loss": 0.841,
+      "step": 3490
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.0263157894736843e-05,
+      "loss": 0.8342,
+      "step": 3500
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 2.9375,
+      "learning_rate": 1.0247368421052631e-05,
+      "loss": 0.7194,
+      "step": 3510
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 4.125,
+      "learning_rate": 1.0231578947368422e-05,
+      "loss": 0.7291,
+      "step": 3520
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 4.0,
+      "learning_rate": 1.021578947368421e-05,
+      "loss": 0.941,
+      "step": 3530
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.02e-05,
+      "loss": 0.6547,
+      "step": 3540
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 4.65625,
+      "learning_rate": 1.018421052631579e-05,
+      "loss": 0.792,
+      "step": 3550
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 3.65625,
+      "learning_rate": 1.0168421052631578e-05,
+      "loss": 0.7671,
+      "step": 3560
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 3.75,
+      "learning_rate": 1.0152631578947368e-05,
+      "loss": 0.8252,
+      "step": 3570
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 3.796875,
+      "learning_rate": 1.0136842105263157e-05,
+      "loss": 0.7255,
+      "step": 3580
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 4.75,
+      "learning_rate": 1.0121052631578949e-05,
+      "loss": 0.6673,
+      "step": 3590
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.0105263157894738e-05,
+      "loss": 0.8474,
+      "step": 3600
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 4.96875,
+      "learning_rate": 1.0089473684210528e-05,
+      "loss": 0.7781,
+      "step": 3610
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.0073684210526317e-05,
+      "loss": 0.7888,
+      "step": 3620
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 4.15625,
+      "learning_rate": 1.0057894736842105e-05,
+      "loss": 0.759,
+      "step": 3630
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 3.46875,
+      "learning_rate": 1.0042105263157896e-05,
+      "loss": 0.7179,
+      "step": 3640
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 4.75,
+      "learning_rate": 1.0026315789473684e-05,
+      "loss": 0.8623,
+      "step": 3650
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 3.25,
+      "learning_rate": 1.0010526315789474e-05,
+      "loss": 0.8619,
+      "step": 3660
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 4.15625,
+      "learning_rate": 9.994736842105263e-06,
+      "loss": 0.8069,
+      "step": 3670
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 3.84375,
+      "learning_rate": 9.978947368421053e-06,
+      "loss": 0.7284,
+      "step": 3680
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 4.375,
+      "learning_rate": 9.963157894736842e-06,
+      "loss": 0.7016,
+      "step": 3690
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 2.96875,
+      "learning_rate": 9.94736842105263e-06,
+      "loss": 0.8813,
+      "step": 3700
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 3.203125,
+      "learning_rate": 9.931578947368421e-06,
+      "loss": 0.805,
+      "step": 3710
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 4.03125,
+      "learning_rate": 9.91578947368421e-06,
+      "loss": 0.8019,
+      "step": 3720
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 4.65625,
+      "learning_rate": 9.9e-06,
+      "loss": 0.8296,
+      "step": 3730
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 3.203125,
+      "learning_rate": 9.884210526315789e-06,
+      "loss": 0.7661,
+      "step": 3740
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 3.734375,
+      "learning_rate": 9.868421052631579e-06,
+      "loss": 0.8698,
+      "step": 3750
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 3.09375,
+      "learning_rate": 9.85263157894737e-06,
+      "loss": 0.6393,
+      "step": 3760
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 3.09375,
+      "learning_rate": 9.836842105263158e-06,
+      "loss": 0.8669,
+      "step": 3770
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 3.484375,
+      "learning_rate": 9.821052631578948e-06,
+      "loss": 0.6591,
+      "step": 3780
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 6.09375,
+      "learning_rate": 9.805263157894737e-06,
+      "loss": 0.9513,
+      "step": 3790
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 3.53125,
+      "learning_rate": 9.789473684210527e-06,
+      "loss": 0.7733,
+      "step": 3800
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 4.5,
+      "learning_rate": 9.773684210526316e-06,
+      "loss": 0.8546,
+      "step": 3810
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 3.71875,
+      "learning_rate": 9.757894736842106e-06,
+      "loss": 0.7754,
+      "step": 3820
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 3.65625,
+      "learning_rate": 9.742105263157895e-06,
+      "loss": 0.8153,
+      "step": 3830
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 3.53125,
+      "learning_rate": 9.726315789473684e-06,
+      "loss": 0.8027,
+      "step": 3840
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 4.09375,
+      "learning_rate": 9.710526315789474e-06,
+      "loss": 0.7553,
+      "step": 3850
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 3.5625,
+      "learning_rate": 9.694736842105263e-06,
+      "loss": 0.792,
+      "step": 3860
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 4.5,
+      "learning_rate": 9.678947368421053e-06,
+      "loss": 0.8321,
+      "step": 3870
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 3.296875,
+      "learning_rate": 9.663157894736842e-06,
+      "loss": 0.7507,
+      "step": 3880
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 3.34375,
+      "learning_rate": 9.647368421052632e-06,
+      "loss": 0.8785,
+      "step": 3890
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 3.40625,
+      "learning_rate": 9.63157894736842e-06,
+      "loss": 0.8377,
+      "step": 3900
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 3.125,
+      "learning_rate": 9.61578947368421e-06,
+      "loss": 0.7123,
+      "step": 3910
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 3.796875,
+      "learning_rate": 9.600000000000001e-06,
+      "loss": 0.8589,
+      "step": 3920
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 4.6875,
+      "learning_rate": 9.58421052631579e-06,
+      "loss": 0.7271,
+      "step": 3930
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 4.25,
+      "learning_rate": 9.56842105263158e-06,
+      "loss": 0.7599,
+      "step": 3940
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 4.65625,
+      "learning_rate": 9.552631578947369e-06,
+      "loss": 0.8152,
+      "step": 3950
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 3.21875,
+      "learning_rate": 9.53684210526316e-06,
+      "loss": 0.6405,
+      "step": 3960
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 3.921875,
+      "learning_rate": 9.521052631578948e-06,
+      "loss": 0.8223,
+      "step": 3970
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 4.25,
+      "learning_rate": 9.505263157894737e-06,
+      "loss": 0.7384,
+      "step": 3980
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 2.984375,
+      "learning_rate": 9.489473684210527e-06,
+      "loss": 0.7385,
+      "step": 3990
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 3.421875,
+      "learning_rate": 9.473684210526315e-06,
+      "loss": 0.7129,
+      "step": 4000
+    },
+    {
+      "epoch": 1.43,
+      "eval_loss": 0.9182843565940857,
+      "eval_runtime": 64.588,
+      "eval_samples_per_second": 15.483,
+      "eval_steps_per_second": 15.483,
+      "step": 4000
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 3.859375,
+      "learning_rate": 9.457894736842106e-06,
+      "loss": 0.723,
+      "step": 4010
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 3.921875,
+      "learning_rate": 9.442105263157894e-06,
+      "loss": 0.7264,
+      "step": 4020
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 4.21875,
+      "learning_rate": 9.426315789473685e-06,
+      "loss": 0.9201,
+      "step": 4030
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 3.4375,
+      "learning_rate": 9.410526315789473e-06,
+      "loss": 0.8538,
+      "step": 4040
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 4.09375,
+      "learning_rate": 9.394736842105262e-06,
+      "loss": 0.8243,
+      "step": 4050
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 3.796875,
+      "learning_rate": 9.378947368421052e-06,
+      "loss": 0.7478,
+      "step": 4060
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 2.90625,
+      "learning_rate": 9.363157894736841e-06,
+      "loss": 0.817,
+      "step": 4070
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 5.0625,
+      "learning_rate": 9.347368421052631e-06,
+      "loss": 0.8185,
+      "step": 4080
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 2.53125,
+      "learning_rate": 9.331578947368422e-06,
+      "loss": 0.8513,
+      "step": 4090
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 3.03125,
+      "learning_rate": 9.315789473684212e-06,
+      "loss": 0.7128,
+      "step": 4100
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 3.921875,
+      "learning_rate": 9.3e-06,
+      "loss": 0.5976,
+      "step": 4110
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 3.90625,
+      "learning_rate": 9.28421052631579e-06,
+      "loss": 0.7252,
+      "step": 4120
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 3.609375,
+      "learning_rate": 9.26842105263158e-06,
+      "loss": 0.7928,
+      "step": 4130
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 3.125,
+      "learning_rate": 9.252631578947368e-06,
+      "loss": 0.8765,
+      "step": 4140
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 4.40625,
+      "learning_rate": 9.236842105263159e-06,
+      "loss": 0.7312,
+      "step": 4150
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 3.78125,
+      "learning_rate": 9.221052631578947e-06,
+      "loss": 0.6883,
+      "step": 4160
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 4.40625,
+      "learning_rate": 9.205263157894738e-06,
+      "loss": 0.655,
+      "step": 4170
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 3.609375,
+      "learning_rate": 9.189473684210526e-06,
+      "loss": 0.7601,
+      "step": 4180
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 3.578125,
+      "learning_rate": 9.173684210526315e-06,
+      "loss": 0.7925,
+      "step": 4190
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 3.640625,
+      "learning_rate": 9.157894736842105e-06,
+      "loss": 0.7922,
+      "step": 4200
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 4.59375,
+      "learning_rate": 9.142105263157894e-06,
+      "loss": 0.8186,
+      "step": 4210
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 4.40625,
+      "learning_rate": 9.126315789473684e-06,
+      "loss": 0.8376,
+      "step": 4220
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 3.90625,
+      "learning_rate": 9.110526315789473e-06,
+      "loss": 0.9098,
+      "step": 4230
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 3.96875,
+      "learning_rate": 9.094736842105263e-06,
+      "loss": 0.7075,
+      "step": 4240
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 3.171875,
+      "learning_rate": 9.078947368421054e-06,
+      "loss": 0.8315,
+      "step": 4250
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 3.84375,
+      "learning_rate": 9.063157894736842e-06,
+      "loss": 0.7757,
+      "step": 4260
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 3.296875,
+      "learning_rate": 9.047368421052633e-06,
+      "loss": 0.8147,
+      "step": 4270
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 3.109375,
+      "learning_rate": 9.031578947368421e-06,
+      "loss": 0.8017,
+      "step": 4280
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 2.90625,
+      "learning_rate": 9.015789473684212e-06,
+      "loss": 0.8293,
+      "step": 4290
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 2.65625,
+      "learning_rate": 9e-06,
+      "loss": 0.7904,
+      "step": 4300
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 5.15625,
+      "learning_rate": 8.98421052631579e-06,
+      "loss": 0.7352,
+      "step": 4310
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 4.28125,
+      "learning_rate": 8.96842105263158e-06,
+      "loss": 0.8582,
+      "step": 4320
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 4.1875,
+      "learning_rate": 8.952631578947368e-06,
+      "loss": 0.7915,
+      "step": 4330
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 3.3125,
+      "learning_rate": 8.936842105263158e-06,
+      "loss": 0.6807,
+      "step": 4340
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 3.390625,
+      "learning_rate": 8.921052631578947e-06,
+      "loss": 0.7701,
+      "step": 4350
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 4.8125,
+      "learning_rate": 8.905263157894737e-06,
+      "loss": 0.8392,
+      "step": 4360
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 4.375,
+      "learning_rate": 8.889473684210526e-06,
+      "loss": 0.788,
+      "step": 4370
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 2.5625,
+      "learning_rate": 8.873684210526316e-06,
+      "loss": 0.7791,
+      "step": 4380
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 4.09375,
+      "learning_rate": 8.857894736842105e-06,
+      "loss": 0.7486,
+      "step": 4390
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 3.15625,
+      "learning_rate": 8.842105263157893e-06,
+      "loss": 0.6972,
+      "step": 4400
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 3.546875,
+      "learning_rate": 8.826315789473684e-06,
+      "loss": 0.8958,
+      "step": 4410
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 3.59375,
+      "learning_rate": 8.810526315789474e-06,
+      "loss": 0.7414,
+      "step": 4420
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 5.03125,
+      "learning_rate": 8.794736842105264e-06,
+      "loss": 0.8639,
+      "step": 4430
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 4.46875,
+      "learning_rate": 8.778947368421053e-06,
+      "loss": 0.7731,
+      "step": 4440
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 4.15625,
+      "learning_rate": 8.763157894736843e-06,
+      "loss": 0.8201,
+      "step": 4450
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 4.375,
+      "learning_rate": 8.747368421052632e-06,
+      "loss": 0.8138,
+      "step": 4460
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 3.90625,
+      "learning_rate": 8.731578947368422e-06,
+      "loss": 0.8884,
+      "step": 4470
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 4.75,
+      "learning_rate": 8.715789473684211e-06,
+      "loss": 0.6786,
+      "step": 4480
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 4.0625,
+      "learning_rate": 8.7e-06,
+      "loss": 0.7384,
+      "step": 4490
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 3.046875,
+      "learning_rate": 8.68421052631579e-06,
+      "loss": 0.6972,
+      "step": 4500
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 4.15625,
+      "learning_rate": 8.668421052631579e-06,
+      "loss": 0.6961,
+      "step": 4510
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 2.65625,
+      "learning_rate": 8.652631578947369e-06,
+      "loss": 0.8085,
+      "step": 4520
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 3.046875,
+      "learning_rate": 8.636842105263158e-06,
+      "loss": 0.621,
+      "step": 4530
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 3.265625,
+      "learning_rate": 8.621052631578948e-06,
+      "loss": 0.7796,
+      "step": 4540
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 3.859375,
+      "learning_rate": 8.605263157894737e-06,
+      "loss": 0.712,
+      "step": 4550
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 2.796875,
+      "learning_rate": 8.589473684210525e-06,
+      "loss": 0.7597,
+      "step": 4560
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 4.5625,
+      "learning_rate": 8.573684210526316e-06,
+      "loss": 0.8231,
+      "step": 4570
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 3.65625,
+      "learning_rate": 8.557894736842104e-06,
+      "loss": 0.8502,
+      "step": 4580
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 3.765625,
+      "learning_rate": 8.542105263157896e-06,
+      "loss": 0.8364,
+      "step": 4590
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 3.953125,
+      "learning_rate": 8.526315789473685e-06,
+      "loss": 0.8229,
+      "step": 4600
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 4.5,
+      "learning_rate": 8.510526315789475e-06,
+      "loss": 0.7733,
+      "step": 4610
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 3.359375,
+      "learning_rate": 8.494736842105264e-06,
+      "loss": 0.8358,
+      "step": 4620
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 3.4375,
+      "learning_rate": 8.478947368421053e-06,
+      "loss": 0.7853,
+      "step": 4630
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 3.65625,
+      "learning_rate": 8.463157894736843e-06,
+      "loss": 0.7147,
+      "step": 4640
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 3.375,
+      "learning_rate": 8.447368421052632e-06,
+      "loss": 0.7554,
+      "step": 4650
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 3.515625,
+      "learning_rate": 8.431578947368422e-06,
+      "loss": 0.7443,
+      "step": 4660
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 3.046875,
+      "learning_rate": 8.41578947368421e-06,
+      "loss": 0.7417,
+      "step": 4670
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 2.6875,
+      "learning_rate": 8.400000000000001e-06,
+      "loss": 0.8205,
+      "step": 4680
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 4.5,
+      "learning_rate": 8.38421052631579e-06,
+      "loss": 0.6986,
+      "step": 4690
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 3.359375,
+      "learning_rate": 8.368421052631578e-06,
+      "loss": 0.713,
+      "step": 4700
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 3.953125,
+      "learning_rate": 8.352631578947369e-06,
+      "loss": 0.7703,
+      "step": 4710
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 4.125,
+      "learning_rate": 8.336842105263157e-06,
+      "loss": 0.7835,
+      "step": 4720
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 4.34375,
+      "learning_rate": 8.321052631578947e-06,
+      "loss": 0.7691,
+      "step": 4730
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 4.46875,
+      "learning_rate": 8.305263157894736e-06,
+      "loss": 0.8923,
+      "step": 4740
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 4.6875,
+      "learning_rate": 8.289473684210526e-06,
+      "loss": 0.8753,
+      "step": 4750
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 3.53125,
+      "learning_rate": 8.273684210526317e-06,
+      "loss": 0.7422,
+      "step": 4760
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 4.78125,
+      "learning_rate": 8.257894736842105e-06,
+      "loss": 0.7183,
+      "step": 4770
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 4.34375,
+      "learning_rate": 8.242105263157896e-06,
+      "loss": 0.7278,
+      "step": 4780
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 4.3125,
+      "learning_rate": 8.226315789473684e-06,
+      "loss": 0.7935,
+      "step": 4790
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 4.21875,
+      "learning_rate": 8.210526315789475e-06,
+      "loss": 0.6831,
+      "step": 4800
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 4.25,
+      "learning_rate": 8.194736842105263e-06,
+      "loss": 0.9393,
+      "step": 4810
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 3.875,
+      "learning_rate": 8.178947368421054e-06,
+      "loss": 0.8149,
+      "step": 4820
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 4.28125,
+      "learning_rate": 8.163157894736842e-06,
+      "loss": 0.7412,
+      "step": 4830
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 4.375,
+      "learning_rate": 8.147368421052631e-06,
+      "loss": 0.7156,
+      "step": 4840
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 3.734375,
+      "learning_rate": 8.131578947368421e-06,
+      "loss": 0.7877,
+      "step": 4850
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 5.875,
+      "learning_rate": 8.11578947368421e-06,
+      "loss": 0.7018,
+      "step": 4860
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 4.59375,
+      "learning_rate": 8.1e-06,
+      "loss": 0.7845,
+      "step": 4870
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 3.046875,
+      "learning_rate": 8.084210526315789e-06,
+      "loss": 0.6962,
+      "step": 4880
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 3.71875,
+      "learning_rate": 8.06842105263158e-06,
+      "loss": 0.8147,
+      "step": 4890
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 3.859375,
+      "learning_rate": 8.052631578947368e-06,
+      "loss": 0.75,
+      "step": 4900
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 3.140625,
+      "learning_rate": 8.036842105263157e-06,
+      "loss": 0.7435,
+      "step": 4910
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 3.34375,
+      "learning_rate": 8.021052631578949e-06,
+      "loss": 0.7902,
+      "step": 4920
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 5.5,
+      "learning_rate": 8.005263157894737e-06,
+      "loss": 0.8582,
+      "step": 4930
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 5.25,
+      "learning_rate": 7.989473684210528e-06,
+      "loss": 0.73,
+      "step": 4940
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 2.796875,
+      "learning_rate": 7.973684210526316e-06,
+      "loss": 0.7446,
+      "step": 4950
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 5.0625,
+      "learning_rate": 7.957894736842107e-06,
+      "loss": 0.7384,
+      "step": 4960
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 3.671875,
+      "learning_rate": 7.942105263157895e-06,
+      "loss": 0.8652,
+      "step": 4970
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 4.3125,
+      "learning_rate": 7.926315789473684e-06,
+      "loss": 0.6912,
+      "step": 4980
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 3.21875,
+      "learning_rate": 7.910526315789474e-06,
+      "loss": 0.8403,
+      "step": 4990
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 3.734375,
+      "learning_rate": 7.894736842105263e-06,
+      "loss": 0.7563,
+      "step": 5000
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 3.484375,
+      "learning_rate": 7.878947368421053e-06,
+      "loss": 0.7618,
+      "step": 5010
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 3.671875,
+      "learning_rate": 7.863157894736842e-06,
+      "loss": 0.6903,
+      "step": 5020
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 3.53125,
+      "learning_rate": 7.847368421052632e-06,
+      "loss": 0.787,
+      "step": 5030
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 3.515625,
+      "learning_rate": 7.831578947368421e-06,
+      "loss": 0.7276,
+      "step": 5040
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 4.5,
+      "learning_rate": 7.81578947368421e-06,
+      "loss": 0.8304,
+      "step": 5050
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 4.875,
+      "learning_rate": 7.8e-06,
+      "loss": 0.802,
+      "step": 5060
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 3.890625,
+      "learning_rate": 7.784210526315789e-06,
+      "loss": 0.72,
+      "step": 5070
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 2.953125,
+      "learning_rate": 7.768421052631579e-06,
+      "loss": 0.8963,
+      "step": 5080
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 4.25,
+      "learning_rate": 7.75263157894737e-06,
+      "loss": 0.8178,
+      "step": 5090
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 3.53125,
+      "learning_rate": 7.73684210526316e-06,
+      "loss": 0.791,
+      "step": 5100
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 3.328125,
+      "learning_rate": 7.721052631578948e-06,
+      "loss": 0.803,
+      "step": 5110
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 4.15625,
+      "learning_rate": 7.705263157894737e-06,
+      "loss": 0.8666,
+      "step": 5120
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 3.546875,
+      "learning_rate": 7.689473684210527e-06,
+      "loss": 0.9192,
+      "step": 5130
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 4.59375,
+      "learning_rate": 7.673684210526316e-06,
+      "loss": 0.7325,
+      "step": 5140
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 6.15625,
+      "learning_rate": 7.657894736842106e-06,
+      "loss": 0.8095,
+      "step": 5150
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 3.515625,
+      "learning_rate": 7.642105263157895e-06,
+      "loss": 0.7512,
+      "step": 5160
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 3.109375,
+      "learning_rate": 7.626315789473685e-06,
+      "loss": 0.7789,
+      "step": 5170
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 4.25,
+      "learning_rate": 7.610526315789474e-06,
+      "loss": 0.728,
+      "step": 5180
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 3.890625,
+      "learning_rate": 7.594736842105262e-06,
+      "loss": 0.8256,
+      "step": 5190
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 3.71875,
+      "learning_rate": 7.578947368421053e-06,
+      "loss": 0.8819,
+      "step": 5200
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 4.15625,
+      "learning_rate": 7.563157894736842e-06,
+      "loss": 0.7565,
+      "step": 5210
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 2.765625,
+      "learning_rate": 7.5473684210526326e-06,
+      "loss": 0.5771,
+      "step": 5220
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 4.71875,
+      "learning_rate": 7.531578947368421e-06,
+      "loss": 0.84,
+      "step": 5230
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 5.375,
+      "learning_rate": 7.5157894736842115e-06,
+      "loss": 0.7282,
+      "step": 5240
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 3.140625,
+      "learning_rate": 7.5e-06,
+      "loss": 0.7246,
+      "step": 5250
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 3.515625,
+      "learning_rate": 7.48421052631579e-06,
+      "loss": 0.7668,
+      "step": 5260
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 2.984375,
+      "learning_rate": 7.468421052631579e-06,
+      "loss": 0.7756,
+      "step": 5270
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 4.0,
+      "learning_rate": 7.452631578947369e-06,
+      "loss": 0.8317,
+      "step": 5280
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 2.421875,
+      "learning_rate": 7.436842105263157e-06,
+      "loss": 0.6991,
+      "step": 5290
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 5.59375,
+      "learning_rate": 7.421052631578948e-06,
+      "loss": 0.7847,
+      "step": 5300
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 3.21875,
+      "learning_rate": 7.405263157894737e-06,
+      "loss": 0.8201,
+      "step": 5310
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 4.8125,
+      "learning_rate": 7.389473684210527e-06,
+      "loss": 0.8239,
+      "step": 5320
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 3.546875,
+      "learning_rate": 7.373684210526316e-06,
+      "loss": 0.8543,
+      "step": 5330
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 3.859375,
+      "learning_rate": 7.357894736842106e-06,
+      "loss": 0.7703,
+      "step": 5340
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 3.53125,
+      "learning_rate": 7.342105263157895e-06,
+      "loss": 0.7758,
+      "step": 5350
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 4.5625,
+      "learning_rate": 7.326315789473684e-06,
+      "loss": 0.7798,
+      "step": 5360
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 4.53125,
+      "learning_rate": 7.310526315789473e-06,
+      "loss": 0.77,
+      "step": 5370
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 3.28125,
+      "learning_rate": 7.2947368421052636e-06,
+      "loss": 0.707,
+      "step": 5380
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 5.34375,
+      "learning_rate": 7.278947368421053e-06,
+      "loss": 0.753,
+      "step": 5390
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 4.4375,
+      "learning_rate": 7.2631578947368426e-06,
+      "loss": 0.7354,
+      "step": 5400
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 4.0625,
+      "learning_rate": 7.247368421052632e-06,
+      "loss": 0.7857,
+      "step": 5410
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 3.203125,
+      "learning_rate": 7.2315789473684215e-06,
+      "loss": 0.7155,
+      "step": 5420
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 4.03125,
+      "learning_rate": 7.21578947368421e-06,
+      "loss": 0.7654,
+      "step": 5430
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 4.59375,
+      "learning_rate": 7.2e-06,
+      "loss": 0.7981,
+      "step": 5440
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 4.28125,
+      "learning_rate": 7.184210526315789e-06,
+      "loss": 0.6829,
+      "step": 5450
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 3.453125,
+      "learning_rate": 7.1684210526315795e-06,
+      "loss": 0.7282,
+      "step": 5460
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 4.28125,
+      "learning_rate": 7.152631578947369e-06,
+      "loss": 0.8982,
+      "step": 5470
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 2.875,
+      "learning_rate": 7.1368421052631585e-06,
+      "loss": 0.7478,
+      "step": 5480
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 4.09375,
+      "learning_rate": 7.121052631578948e-06,
+      "loss": 0.7686,
+      "step": 5490
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 3.421875,
+      "learning_rate": 7.105263157894737e-06,
+      "loss": 0.7631,
+      "step": 5500
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 3.8125,
+      "learning_rate": 7.089473684210526e-06,
+      "loss": 0.7765,
+      "step": 5510
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 3.46875,
+      "learning_rate": 7.073684210526316e-06,
+      "loss": 0.8143,
+      "step": 5520
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 4.9375,
+      "learning_rate": 7.057894736842105e-06,
+      "loss": 0.7423,
+      "step": 5530
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 4.5625,
+      "learning_rate": 7.042105263157895e-06,
+      "loss": 0.8368,
+      "step": 5540
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 3.921875,
+      "learning_rate": 7.026315789473685e-06,
+      "loss": 0.8404,
+      "step": 5550
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 4.75,
+      "learning_rate": 7.010526315789474e-06,
+      "loss": 0.7315,
+      "step": 5560
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 4.0,
+      "learning_rate": 6.994736842105263e-06,
+      "loss": 0.7595,
+      "step": 5570
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 3.8125,
+      "learning_rate": 6.9789473684210525e-06,
+      "loss": 0.7103,
+      "step": 5580
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 3.21875,
+      "learning_rate": 6.963157894736842e-06,
+      "loss": 0.7027,
+      "step": 5590
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 3.421875,
+      "learning_rate": 6.9473684210526315e-06,
+      "loss": 0.7695,
+      "step": 5600
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 3.21875,
+      "learning_rate": 6.931578947368421e-06,
+      "loss": 0.7225,
+      "step": 5610
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 4.4375,
+      "learning_rate": 6.9157894736842105e-06,
+      "loss": 0.8339,
+      "step": 5620
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 3.59375,
+      "learning_rate": 6.900000000000001e-06,
+      "loss": 0.712,
+      "step": 5630
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 4.6875,
+      "learning_rate": 6.8842105263157895e-06,
+      "loss": 0.7968,
+      "step": 5640
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 4.65625,
+      "learning_rate": 6.868421052631579e-06,
+      "loss": 0.7415,
+      "step": 5650
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 3.6875,
+      "learning_rate": 6.8526315789473685e-06,
+      "loss": 0.7186,
+      "step": 5660
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 3.484375,
+      "learning_rate": 6.836842105263158e-06,
+      "loss": 0.7947,
+      "step": 5670
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 3.21875,
+      "learning_rate": 6.8210526315789475e-06,
+      "loss": 0.7821,
+      "step": 5680
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 5.0625,
+      "learning_rate": 6.805263157894737e-06,
+      "loss": 0.8232,
+      "step": 5690
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 4.59375,
+      "learning_rate": 6.7894736842105264e-06,
+      "loss": 0.7705,
+      "step": 5700
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 3.40625,
+      "learning_rate": 6.773684210526317e-06,
+      "loss": 0.8094,
+      "step": 5710
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 3.65625,
+      "learning_rate": 6.7578947368421054e-06,
+      "loss": 0.6984,
+      "step": 5720
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 5.6875,
+      "learning_rate": 6.742105263157895e-06,
+      "loss": 0.6577,
+      "step": 5730
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 4.125,
+      "learning_rate": 6.726315789473684e-06,
+      "loss": 0.7978,
+      "step": 5740
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 3.078125,
+      "learning_rate": 6.710526315789474e-06,
+      "loss": 0.8812,
+      "step": 5750
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 7.0,
+      "learning_rate": 6.694736842105263e-06,
+      "loss": 0.866,
+      "step": 5760
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 4.6875,
+      "learning_rate": 6.678947368421053e-06,
+      "loss": 0.76,
+      "step": 5770
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 3.640625,
+      "learning_rate": 6.663157894736842e-06,
+      "loss": 0.7124,
+      "step": 5780
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 3.96875,
+      "learning_rate": 6.647368421052631e-06,
+      "loss": 0.7763,
+      "step": 5790
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 4.3125,
+      "learning_rate": 6.631578947368421e-06,
+      "loss": 0.7461,
+      "step": 5800
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 3.4375,
+      "learning_rate": 6.615789473684211e-06,
+      "loss": 0.7609,
+      "step": 5810
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 2.90625,
+      "learning_rate": 6.6e-06,
+      "loss": 0.7333,
+      "step": 5820
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 3.03125,
+      "learning_rate": 6.58421052631579e-06,
+      "loss": 0.7595,
+      "step": 5830
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 4.40625,
+      "learning_rate": 6.568421052631579e-06,
+      "loss": 0.8614,
+      "step": 5840
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 3.609375,
+      "learning_rate": 6.552631578947369e-06,
+      "loss": 0.7724,
+      "step": 5850
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 3.125,
+      "learning_rate": 6.5368421052631575e-06,
+      "loss": 0.7704,
+      "step": 5860
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 3.96875,
+      "learning_rate": 6.521052631578947e-06,
+      "loss": 0.8044,
+      "step": 5870
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 3.875,
+      "learning_rate": 6.505263157894737e-06,
+      "loss": 0.7528,
+      "step": 5880
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 3.75,
+      "learning_rate": 6.489473684210527e-06,
+      "loss": 0.7674,
+      "step": 5890
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 4.125,
+      "learning_rate": 6.473684210526316e-06,
+      "loss": 0.7433,
+      "step": 5900
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 3.625,
+      "learning_rate": 6.457894736842106e-06,
+      "loss": 0.7442,
+      "step": 5910
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 4.46875,
+      "learning_rate": 6.442105263157895e-06,
+      "loss": 0.8057,
+      "step": 5920
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 3.828125,
+      "learning_rate": 6.426315789473684e-06,
+      "loss": 0.7341,
+      "step": 5930
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 4.53125,
+      "learning_rate": 6.410526315789473e-06,
+      "loss": 0.8244,
+      "step": 5940
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 5.34375,
+      "learning_rate": 6.394736842105263e-06,
+      "loss": 0.7761,
+      "step": 5950
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 3.765625,
+      "learning_rate": 6.378947368421053e-06,
+      "loss": 0.7798,
+      "step": 5960
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 3.828125,
+      "learning_rate": 6.363157894736843e-06,
+      "loss": 0.7711,
+      "step": 5970
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 3.25,
+      "learning_rate": 6.347368421052632e-06,
+      "loss": 0.7087,
+      "step": 5980
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 3.015625,
+      "learning_rate": 6.331578947368422e-06,
+      "loss": 0.7082,
+      "step": 5990
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 4.0,
+      "learning_rate": 6.31578947368421e-06,
+      "loss": 0.7282,
+      "step": 6000
+    },
+    {
+      "epoch": 2.15,
+      "eval_loss": 0.8967013955116272,
+      "eval_runtime": 64.6575,
+      "eval_samples_per_second": 15.466,
+      "eval_steps_per_second": 15.466,
+      "step": 6000
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 3.546875,
+      "learning_rate": 6.3e-06,
+      "loss": 0.7036,
+      "step": 6010
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 2.8125,
+      "learning_rate": 6.284210526315789e-06,
+      "loss": 0.6933,
+      "step": 6020
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 3.1875,
+      "learning_rate": 6.268421052631579e-06,
+      "loss": 0.8185,
+      "step": 6030
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 3.953125,
+      "learning_rate": 6.252631578947368e-06,
+      "loss": 0.7636,
+      "step": 6040
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 4.1875,
+      "learning_rate": 6.236842105263159e-06,
+      "loss": 0.8563,
+      "step": 6050
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 3.765625,
+      "learning_rate": 6.221052631578948e-06,
+      "loss": 0.8117,
+      "step": 6060
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 3.234375,
+      "learning_rate": 6.205263157894737e-06,
+      "loss": 0.8126,
+      "step": 6070
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 3.0,
+      "learning_rate": 6.189473684210526e-06,
+      "loss": 0.8573,
+      "step": 6080
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 2.984375,
+      "learning_rate": 6.173684210526316e-06,
+      "loss": 0.767,
+      "step": 6090
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 3.46875,
+      "learning_rate": 6.157894736842105e-06,
+      "loss": 0.7295,
+      "step": 6100
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 4.09375,
+      "learning_rate": 6.142105263157895e-06,
+      "loss": 0.8018,
+      "step": 6110
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 4.1875,
+      "learning_rate": 6.126315789473684e-06,
+      "loss": 0.7006,
+      "step": 6120
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 5.0,
+      "learning_rate": 6.1105263157894746e-06,
+      "loss": 0.6981,
+      "step": 6130
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 5.6875,
+      "learning_rate": 6.094736842105263e-06,
+      "loss": 0.74,
+      "step": 6140
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 3.59375,
+      "learning_rate": 6.078947368421053e-06,
+      "loss": 0.7163,
+      "step": 6150
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 5.25,
+      "learning_rate": 6.063157894736842e-06,
+      "loss": 0.709,
+      "step": 6160
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 4.15625,
+      "learning_rate": 6.047368421052632e-06,
+      "loss": 0.7817,
+      "step": 6170
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 4.03125,
+      "learning_rate": 6.031578947368421e-06,
+      "loss": 0.7428,
+      "step": 6180
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 3.828125,
+      "learning_rate": 6.015789473684211e-06,
+      "loss": 0.7936,
+      "step": 6190
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 3.90625,
+      "learning_rate": 6e-06,
+      "loss": 0.7811,
+      "step": 6200
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 3.234375,
+      "learning_rate": 5.98421052631579e-06,
+      "loss": 0.7288,
+      "step": 6210
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 3.0625,
+      "learning_rate": 5.968421052631579e-06,
+      "loss": 0.7112,
+      "step": 6220
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 3.5625,
+      "learning_rate": 5.952631578947369e-06,
+      "loss": 0.706,
+      "step": 6230
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 4.625,
+      "learning_rate": 5.936842105263158e-06,
+      "loss": 0.8553,
+      "step": 6240
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 3.84375,
+      "learning_rate": 5.921052631578948e-06,
+      "loss": 0.7414,
+      "step": 6250
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 3.734375,
+      "learning_rate": 5.905263157894737e-06,
+      "loss": 0.7544,
+      "step": 6260
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 3.890625,
+      "learning_rate": 5.889473684210527e-06,
+      "loss": 0.8527,
+      "step": 6270
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 2.71875,
+      "learning_rate": 5.873684210526315e-06,
+      "loss": 0.693,
+      "step": 6280
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 5.25,
+      "learning_rate": 5.857894736842105e-06,
+      "loss": 0.7309,
+      "step": 6290
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 4.0,
+      "learning_rate": 5.842105263157895e-06,
+      "loss": 0.8126,
+      "step": 6300
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 3.515625,
+      "learning_rate": 5.8263157894736846e-06,
+      "loss": 0.6254,
+      "step": 6310
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 4.15625,
+      "learning_rate": 5.810526315789474e-06,
+      "loss": 0.7217,
+      "step": 6320
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 4.3125,
+      "learning_rate": 5.7947368421052635e-06,
+      "loss": 0.8655,
+      "step": 6330
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 3.578125,
+      "learning_rate": 5.778947368421053e-06,
+      "loss": 0.806,
+      "step": 6340
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 3.5,
+      "learning_rate": 5.763157894736842e-06,
+      "loss": 0.6672,
+      "step": 6350
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 3.859375,
+      "learning_rate": 5.747368421052631e-06,
+      "loss": 0.7941,
+      "step": 6360
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 3.515625,
+      "learning_rate": 5.731578947368421e-06,
+      "loss": 0.7278,
+      "step": 6370
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 4.6875,
+      "learning_rate": 5.715789473684211e-06,
+      "loss": 0.7937,
+      "step": 6380
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 4.03125,
+      "learning_rate": 5.7000000000000005e-06,
+      "loss": 0.7249,
+      "step": 6390
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 4.0,
+      "learning_rate": 5.68421052631579e-06,
+      "loss": 0.8656,
+      "step": 6400
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 2.984375,
+      "learning_rate": 5.6684210526315795e-06,
+      "loss": 0.8155,
+      "step": 6410
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 4.75,
+      "learning_rate": 5.652631578947368e-06,
+      "loss": 0.7534,
+      "step": 6420
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 3.015625,
+      "learning_rate": 5.636842105263158e-06,
+      "loss": 0.7301,
+      "step": 6430
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 4.40625,
+      "learning_rate": 5.621052631578947e-06,
+      "loss": 0.6654,
+      "step": 6440
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 3.171875,
+      "learning_rate": 5.605263157894737e-06,
+      "loss": 0.77,
+      "step": 6450
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 4.78125,
+      "learning_rate": 5.589473684210527e-06,
+      "loss": 0.8216,
+      "step": 6460
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 3.640625,
+      "learning_rate": 5.573684210526316e-06,
+      "loss": 0.7287,
+      "step": 6470
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 4.15625,
+      "learning_rate": 5.557894736842106e-06,
+      "loss": 0.7853,
+      "step": 6480
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 3.21875,
+      "learning_rate": 5.542105263157895e-06,
+      "loss": 0.8813,
+      "step": 6490
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 3.046875,
+      "learning_rate": 5.526315789473684e-06,
+      "loss": 0.7435,
+      "step": 6500
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 3.8125,
+      "learning_rate": 5.5105263157894735e-06,
+      "loss": 0.7538,
+      "step": 6510
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 2.984375,
+      "learning_rate": 5.494736842105263e-06,
+      "loss": 0.6783,
+      "step": 6520
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 3.71875,
+      "learning_rate": 5.4789473684210525e-06,
+      "loss": 0.7574,
+      "step": 6530
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 4.5,
+      "learning_rate": 5.463157894736842e-06,
+      "loss": 0.7444,
+      "step": 6540
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 3.21875,
+      "learning_rate": 5.447368421052632e-06,
+      "loss": 0.8801,
+      "step": 6550
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 3.515625,
+      "learning_rate": 5.431578947368422e-06,
+      "loss": 0.7584,
+      "step": 6560
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 3.0,
+      "learning_rate": 5.4157894736842105e-06,
+      "loss": 0.7801,
+      "step": 6570
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 4.125,
+      "learning_rate": 5.4e-06,
+      "loss": 0.7235,
+      "step": 6580
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 3.296875,
+      "learning_rate": 5.3842105263157895e-06,
+      "loss": 0.7289,
+      "step": 6590
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 4.84375,
+      "learning_rate": 5.368421052631579e-06,
+      "loss": 0.8056,
+      "step": 6600
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 4.1875,
+      "learning_rate": 5.3526315789473684e-06,
+      "loss": 0.7156,
+      "step": 6610
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 3.609375,
+      "learning_rate": 5.336842105263158e-06,
+      "loss": 0.8015,
+      "step": 6620
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 3.625,
+      "learning_rate": 5.321052631578948e-06,
+      "loss": 0.7601,
+      "step": 6630
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 4.25,
+      "learning_rate": 5.305263157894737e-06,
+      "loss": 0.7695,
+      "step": 6640
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 3.265625,
+      "learning_rate": 5.289473684210526e-06,
+      "loss": 0.7584,
+      "step": 6650
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 4.03125,
+      "learning_rate": 5.273684210526316e-06,
+      "loss": 0.8082,
+      "step": 6660
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 3.671875,
+      "learning_rate": 5.257894736842105e-06,
+      "loss": 0.7787,
+      "step": 6670
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 4.65625,
+      "learning_rate": 5.242105263157895e-06,
+      "loss": 0.8339,
+      "step": 6680
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 5.625,
+      "learning_rate": 5.226315789473684e-06,
+      "loss": 0.7868,
+      "step": 6690
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 3.8125,
+      "learning_rate": 5.210526315789474e-06,
+      "loss": 0.7713,
+      "step": 6700
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 4.25,
+      "learning_rate": 5.194736842105263e-06,
+      "loss": 0.7354,
+      "step": 6710
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 3.03125,
+      "learning_rate": 5.178947368421053e-06,
+      "loss": 0.6295,
+      "step": 6720
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 3.734375,
+      "learning_rate": 5.163157894736842e-06,
+      "loss": 0.7935,
+      "step": 6730
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 4.125,
+      "learning_rate": 5.147368421052632e-06,
+      "loss": 0.7819,
+      "step": 6740
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 3.9375,
+      "learning_rate": 5.131578947368421e-06,
+      "loss": 0.6854,
+      "step": 6750
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 4.90625,
+      "learning_rate": 5.115789473684211e-06,
+      "loss": 0.7765,
+      "step": 6760
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 4.09375,
+      "learning_rate": 5.1e-06,
+      "loss": 0.7446,
+      "step": 6770
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 4.3125,
+      "learning_rate": 5.084210526315789e-06,
+      "loss": 0.7907,
+      "step": 6780
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 2.265625,
+      "learning_rate": 5.0684210526315784e-06,
+      "loss": 0.7386,
+      "step": 6790
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 3.90625,
+      "learning_rate": 5.052631578947369e-06,
+      "loss": 0.8133,
+      "step": 6800
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 2.890625,
+      "learning_rate": 5.036842105263158e-06,
+      "loss": 0.7751,
+      "step": 6810
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 3.171875,
+      "learning_rate": 5.021052631578948e-06,
+      "loss": 0.6517,
+      "step": 6820
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 4.28125,
+      "learning_rate": 5.005263157894737e-06,
+      "loss": 0.7274,
+      "step": 6830
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 3.203125,
+      "learning_rate": 4.989473684210527e-06,
+      "loss": 0.7575,
+      "step": 6840
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 4.3125,
+      "learning_rate": 4.973684210526315e-06,
+      "loss": 0.8018,
+      "step": 6850
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 3.703125,
+      "learning_rate": 4.957894736842105e-06,
+      "loss": 0.7471,
+      "step": 6860
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 5.03125,
+      "learning_rate": 4.942105263157894e-06,
+      "loss": 0.7156,
+      "step": 6870
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 3.953125,
+      "learning_rate": 4.926315789473685e-06,
+      "loss": 0.7791,
+      "step": 6880
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 3.4375,
+      "learning_rate": 4.910526315789474e-06,
+      "loss": 0.7262,
+      "step": 6890
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 4.6875,
+      "learning_rate": 4.894736842105264e-06,
+      "loss": 0.7311,
+      "step": 6900
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 5.84375,
+      "learning_rate": 4.878947368421053e-06,
+      "loss": 0.7743,
+      "step": 6910
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 4.125,
+      "learning_rate": 4.863157894736842e-06,
+      "loss": 0.6631,
+      "step": 6920
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 3.546875,
+      "learning_rate": 4.847368421052631e-06,
+      "loss": 0.825,
+      "step": 6930
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 4.78125,
+      "learning_rate": 4.831578947368421e-06,
+      "loss": 0.7485,
+      "step": 6940
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 3.296875,
+      "learning_rate": 4.81578947368421e-06,
+      "loss": 0.6608,
+      "step": 6950
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 3.84375,
+      "learning_rate": 4.800000000000001e-06,
+      "loss": 0.7761,
+      "step": 6960
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 3.875,
+      "learning_rate": 4.78421052631579e-06,
+      "loss": 0.881,
+      "step": 6970
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 4.3125,
+      "learning_rate": 4.76842105263158e-06,
+      "loss": 0.8395,
+      "step": 6980
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 3.296875,
+      "learning_rate": 4.752631578947368e-06,
+      "loss": 0.8253,
+      "step": 6990
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 4.96875,
+      "learning_rate": 4.736842105263158e-06,
+      "loss": 0.8286,
+      "step": 7000
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 4.25,
+      "learning_rate": 4.721052631578947e-06,
+      "loss": 0.6905,
+      "step": 7010
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 4.625,
+      "learning_rate": 4.705263157894737e-06,
+      "loss": 0.7968,
+      "step": 7020
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 3.59375,
+      "learning_rate": 4.689473684210526e-06,
+      "loss": 0.7455,
+      "step": 7030
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 3.609375,
+      "learning_rate": 4.673684210526316e-06,
+      "loss": 0.757,
+      "step": 7040
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 3.359375,
+      "learning_rate": 4.657894736842106e-06,
+      "loss": 0.7665,
+      "step": 7050
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 4.375,
+      "learning_rate": 4.642105263157895e-06,
+      "loss": 0.7843,
+      "step": 7060
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 2.921875,
+      "learning_rate": 4.626315789473684e-06,
+      "loss": 0.8173,
+      "step": 7070
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 3.6875,
+      "learning_rate": 4.610526315789474e-06,
+      "loss": 0.7317,
+      "step": 7080
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 3.34375,
+      "learning_rate": 4.594736842105263e-06,
+      "loss": 0.8384,
+      "step": 7090
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 3.984375,
+      "learning_rate": 4.578947368421053e-06,
+      "loss": 0.7956,
+      "step": 7100
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 5.34375,
+      "learning_rate": 4.563157894736842e-06,
+      "loss": 0.6905,
+      "step": 7110
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 5.59375,
+      "learning_rate": 4.547368421052632e-06,
+      "loss": 0.8327,
+      "step": 7120
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 3.515625,
+      "learning_rate": 4.531578947368421e-06,
+      "loss": 0.6478,
+      "step": 7130
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 3.765625,
+      "learning_rate": 4.515789473684211e-06,
+      "loss": 0.7978,
+      "step": 7140
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 2.9375,
+      "learning_rate": 4.5e-06,
+      "loss": 0.6729,
+      "step": 7150
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 3.21875,
+      "learning_rate": 4.48421052631579e-06,
+      "loss": 0.6966,
+      "step": 7160
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 2.9375,
+      "learning_rate": 4.468421052631579e-06,
+      "loss": 0.7551,
+      "step": 7170
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 3.421875,
+      "learning_rate": 4.452631578947369e-06,
+      "loss": 0.7854,
+      "step": 7180
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 4.65625,
+      "learning_rate": 4.436842105263158e-06,
+      "loss": 0.7641,
+      "step": 7190
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 5.09375,
+      "learning_rate": 4.421052631578947e-06,
+      "loss": 0.6273,
+      "step": 7200
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 4.84375,
+      "learning_rate": 4.405263157894737e-06,
+      "loss": 0.7939,
+      "step": 7210
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 3.953125,
+      "learning_rate": 4.3894736842105266e-06,
+      "loss": 0.7322,
+      "step": 7220
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 3.8125,
+      "learning_rate": 4.373684210526316e-06,
+      "loss": 0.7427,
+      "step": 7230
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 3.296875,
+      "learning_rate": 4.3578947368421055e-06,
+      "loss": 0.8102,
+      "step": 7240
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 3.71875,
+      "learning_rate": 4.342105263157895e-06,
+      "loss": 0.8368,
+      "step": 7250
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 2.953125,
+      "learning_rate": 4.3263157894736845e-06,
+      "loss": 0.7338,
+      "step": 7260
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 3.90625,
+      "learning_rate": 4.310526315789474e-06,
+      "loss": 0.7885,
+      "step": 7270
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 4.5,
+      "learning_rate": 4.294736842105263e-06,
+      "loss": 0.7239,
+      "step": 7280
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 3.375,
+      "learning_rate": 4.278947368421052e-06,
+      "loss": 0.8039,
+      "step": 7290
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 3.9375,
+      "learning_rate": 4.2631578947368425e-06,
+      "loss": 0.8542,
+      "step": 7300
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 5.3125,
+      "learning_rate": 4.247368421052632e-06,
+      "loss": 0.7059,
+      "step": 7310
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 3.796875,
+      "learning_rate": 4.2315789473684215e-06,
+      "loss": 0.8158,
+      "step": 7320
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 4.65625,
+      "learning_rate": 4.215789473684211e-06,
+      "loss": 0.7445,
+      "step": 7330
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 4.03125,
+      "learning_rate": 4.2000000000000004e-06,
+      "loss": 0.7084,
+      "step": 7340
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 4.59375,
+      "learning_rate": 4.184210526315789e-06,
+      "loss": 0.6896,
+      "step": 7350
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 3.609375,
+      "learning_rate": 4.168421052631579e-06,
+      "loss": 0.7434,
+      "step": 7360
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 3.3125,
+      "learning_rate": 4.152631578947368e-06,
+      "loss": 0.7333,
+      "step": 7370
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 3.546875,
+      "learning_rate": 4.136842105263158e-06,
+      "loss": 0.7222,
+      "step": 7380
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 5.125,
+      "learning_rate": 4.121052631578948e-06,
+      "loss": 0.7963,
+      "step": 7390
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 6.40625,
+      "learning_rate": 4.105263157894737e-06,
+      "loss": 0.7662,
+      "step": 7400
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 3.671875,
+      "learning_rate": 4.089473684210527e-06,
+      "loss": 0.6625,
+      "step": 7410
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 3.421875,
+      "learning_rate": 4.0736842105263155e-06,
+      "loss": 0.8417,
+      "step": 7420
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 3.90625,
+      "learning_rate": 4.057894736842105e-06,
+      "loss": 0.7341,
+      "step": 7430
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 3.6875,
+      "learning_rate": 4.0421052631578945e-06,
+      "loss": 0.7993,
+      "step": 7440
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 3.9375,
+      "learning_rate": 4.026315789473684e-06,
+      "loss": 0.7611,
+      "step": 7450
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 4.46875,
+      "learning_rate": 4.010526315789474e-06,
+      "loss": 0.7935,
+      "step": 7460
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 3.703125,
+      "learning_rate": 3.994736842105264e-06,
+      "loss": 0.7611,
+      "step": 7470
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 4.78125,
+      "learning_rate": 3.978947368421053e-06,
+      "loss": 0.7544,
+      "step": 7480
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 3.53125,
+      "learning_rate": 3.963157894736842e-06,
+      "loss": 0.784,
+      "step": 7490
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 4.9375,
+      "learning_rate": 3.9473684210526315e-06,
+      "loss": 0.7721,
+      "step": 7500
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 3.921875,
+      "learning_rate": 3.931578947368421e-06,
+      "loss": 0.7488,
+      "step": 7510
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.9157894736842104e-06,
+      "loss": 0.8802,
+      "step": 7520
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 4.53125,
+      "learning_rate": 3.9e-06,
+      "loss": 0.7867,
+      "step": 7530
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 5.71875,
+      "learning_rate": 3.884210526315789e-06,
+      "loss": 0.7267,
+      "step": 7540
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 3.109375,
+      "learning_rate": 3.86842105263158e-06,
+      "loss": 0.7654,
+      "step": 7550
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 2.8125,
+      "learning_rate": 3.852631578947368e-06,
+      "loss": 0.779,
+      "step": 7560
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 2.9375,
+      "learning_rate": 3.836842105263158e-06,
+      "loss": 0.7029,
+      "step": 7570
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 3.265625,
+      "learning_rate": 3.821052631578947e-06,
+      "loss": 0.6534,
+      "step": 7580
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 3.609375,
+      "learning_rate": 3.805263157894737e-06,
+      "loss": 0.7616,
+      "step": 7590
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 3.84375,
+      "learning_rate": 3.7894736842105264e-06,
+      "loss": 0.9068,
+      "step": 7600
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 6.78125,
+      "learning_rate": 3.7736842105263163e-06,
+      "loss": 0.8443,
+      "step": 7610
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 4.625,
+      "learning_rate": 3.7578947368421058e-06,
+      "loss": 0.7684,
+      "step": 7620
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 4.03125,
+      "learning_rate": 3.742105263157895e-06,
+      "loss": 0.857,
+      "step": 7630
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 4.53125,
+      "learning_rate": 3.7263157894736843e-06,
+      "loss": 0.783,
+      "step": 7640
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 3.8125,
+      "learning_rate": 3.710526315789474e-06,
+      "loss": 0.9224,
+      "step": 7650
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 3.703125,
+      "learning_rate": 3.6947368421052633e-06,
+      "loss": 0.7238,
+      "step": 7660
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 3.703125,
+      "learning_rate": 3.678947368421053e-06,
+      "loss": 0.6559,
+      "step": 7670
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 4.25,
+      "learning_rate": 3.663157894736842e-06,
+      "loss": 0.765,
+      "step": 7680
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 6.53125,
+      "learning_rate": 3.6473684210526318e-06,
+      "loss": 0.8451,
+      "step": 7690
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 4.46875,
+      "learning_rate": 3.6315789473684213e-06,
+      "loss": 0.8324,
+      "step": 7700
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 3.828125,
+      "learning_rate": 3.6157894736842108e-06,
+      "loss": 0.7819,
+      "step": 7710
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 2.375,
+      "learning_rate": 3.6e-06,
+      "loss": 0.8752,
+      "step": 7720
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 3.734375,
+      "learning_rate": 3.5842105263157898e-06,
+      "loss": 0.7615,
+      "step": 7730
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 3.5625,
+      "learning_rate": 3.5684210526315792e-06,
+      "loss": 0.8277,
+      "step": 7740
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 3.984375,
+      "learning_rate": 3.5526315789473683e-06,
+      "loss": 0.8103,
+      "step": 7750
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 4.875,
+      "learning_rate": 3.536842105263158e-06,
+      "loss": 0.6556,
+      "step": 7760
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 3.375,
+      "learning_rate": 3.5210526315789473e-06,
+      "loss": 0.8534,
+      "step": 7770
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 3.53125,
+      "learning_rate": 3.505263157894737e-06,
+      "loss": 0.6124,
+      "step": 7780
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 4.5,
+      "learning_rate": 3.4894736842105263e-06,
+      "loss": 0.8278,
+      "step": 7790
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 4.28125,
+      "learning_rate": 3.4736842105263158e-06,
+      "loss": 0.7901,
+      "step": 7800
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 3.5625,
+      "learning_rate": 3.4578947368421053e-06,
+      "loss": 0.6828,
+      "step": 7810
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 5.65625,
+      "learning_rate": 3.4421052631578947e-06,
+      "loss": 0.7066,
+      "step": 7820
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 2.703125,
+      "learning_rate": 3.4263157894736842e-06,
+      "loss": 0.7368,
+      "step": 7830
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 3.671875,
+      "learning_rate": 3.4105263157894737e-06,
+      "loss": 0.7422,
+      "step": 7840
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 3.0,
+      "learning_rate": 3.3947368421052632e-06,
+      "loss": 0.8336,
+      "step": 7850
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 3.953125,
+      "learning_rate": 3.3789473684210527e-06,
+      "loss": 0.8125,
+      "step": 7860
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 4.75,
+      "learning_rate": 3.363157894736842e-06,
+      "loss": 0.7668,
+      "step": 7870
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 3.46875,
+      "learning_rate": 3.3473684210526317e-06,
+      "loss": 0.75,
+      "step": 7880
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 3.65625,
+      "learning_rate": 3.331578947368421e-06,
+      "loss": 0.7404,
+      "step": 7890
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 6.15625,
+      "learning_rate": 3.3157894736842107e-06,
+      "loss": 0.6642,
+      "step": 7900
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 5.09375,
+      "learning_rate": 3.3e-06,
+      "loss": 0.7927,
+      "step": 7910
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 3.375,
+      "learning_rate": 3.2842105263157897e-06,
+      "loss": 0.7087,
+      "step": 7920
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 3.796875,
+      "learning_rate": 3.2684210526315787e-06,
+      "loss": 0.7991,
+      "step": 7930
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 5.25,
+      "learning_rate": 3.2526315789473686e-06,
+      "loss": 0.7497,
+      "step": 7940
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 4.5,
+      "learning_rate": 3.236842105263158e-06,
+      "loss": 0.7304,
+      "step": 7950
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 3.84375,
+      "learning_rate": 3.2210526315789476e-06,
+      "loss": 0.728,
+      "step": 7960
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 3.671875,
+      "learning_rate": 3.2052631578947367e-06,
+      "loss": 0.6906,
+      "step": 7970
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 4.6875,
+      "learning_rate": 3.1894736842105266e-06,
+      "loss": 0.8678,
+      "step": 7980
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 3.484375,
+      "learning_rate": 3.173684210526316e-06,
+      "loss": 0.7119,
+      "step": 7990
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 4.3125,
+      "learning_rate": 3.157894736842105e-06,
+      "loss": 0.7661,
+      "step": 8000
+    },
+    {
+      "epoch": 2.87,
+      "eval_loss": 0.9163448214530945,
+      "eval_runtime": 64.8831,
+      "eval_samples_per_second": 15.412,
+      "eval_steps_per_second": 15.412,
+      "step": 8000
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 3.625,
+      "learning_rate": 3.1421052631578947e-06,
+      "loss": 0.746,
+      "step": 8010
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 3.53125,
+      "learning_rate": 3.126315789473684e-06,
+      "loss": 0.7455,
+      "step": 8020
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 2.84375,
+      "learning_rate": 3.110526315789474e-06,
+      "loss": 0.7597,
+      "step": 8030
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 5.5,
+      "learning_rate": 3.094736842105263e-06,
+      "loss": 0.6805,
+      "step": 8040
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 2.796875,
+      "learning_rate": 3.0789473684210526e-06,
+      "loss": 0.7597,
+      "step": 8050
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 4.84375,
+      "learning_rate": 3.063157894736842e-06,
+      "loss": 0.8047,
+      "step": 8060
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 3.65625,
+      "learning_rate": 3.0473684210526316e-06,
+      "loss": 0.8595,
+      "step": 8070
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 3.9375,
+      "learning_rate": 3.031578947368421e-06,
+      "loss": 0.9178,
+      "step": 8080
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 4.21875,
+      "learning_rate": 3.0157894736842106e-06,
+      "loss": 0.7056,
+      "step": 8090
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 4.5,
+      "learning_rate": 3e-06,
+      "loss": 0.793,
+      "step": 8100
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 4.59375,
+      "learning_rate": 2.9842105263157896e-06,
+      "loss": 0.7367,
+      "step": 8110
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 4.40625,
+      "learning_rate": 2.968421052631579e-06,
+      "loss": 0.8232,
+      "step": 8120
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 3.609375,
+      "learning_rate": 2.9526315789473685e-06,
+      "loss": 0.6963,
+      "step": 8130
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 3.65625,
+      "learning_rate": 2.9368421052631576e-06,
+      "loss": 0.7976,
+      "step": 8140
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 3.65625,
+      "learning_rate": 2.9210526315789475e-06,
+      "loss": 0.8019,
+      "step": 8150
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 4.6875,
+      "learning_rate": 2.905263157894737e-06,
+      "loss": 0.7212,
+      "step": 8160
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 3.875,
+      "learning_rate": 2.8894736842105265e-06,
+      "loss": 0.7343,
+      "step": 8170
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 3.234375,
+      "learning_rate": 2.8736842105263156e-06,
+      "loss": 0.7321,
+      "step": 8180
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 3.375,
+      "learning_rate": 2.8578947368421055e-06,
+      "loss": 0.7918,
+      "step": 8190
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 3.9375,
+      "learning_rate": 2.842105263157895e-06,
+      "loss": 0.7019,
+      "step": 8200
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 3.8125,
+      "learning_rate": 2.826315789473684e-06,
+      "loss": 0.8198,
+      "step": 8210
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 2.6875,
+      "learning_rate": 2.8105263157894735e-06,
+      "loss": 0.7551,
+      "step": 8220
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 2.3125,
+      "learning_rate": 2.7947368421052635e-06,
+      "loss": 0.6969,
+      "step": 8230
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 4.3125,
+      "learning_rate": 2.778947368421053e-06,
+      "loss": 0.7906,
+      "step": 8240
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 3.921875,
+      "learning_rate": 2.763157894736842e-06,
+      "loss": 0.8364,
+      "step": 8250
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 4.625,
+      "learning_rate": 2.7473684210526315e-06,
+      "loss": 0.7917,
+      "step": 8260
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 3.765625,
+      "learning_rate": 2.731578947368421e-06,
+      "loss": 0.7971,
+      "step": 8270
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 5.03125,
+      "learning_rate": 2.715789473684211e-06,
+      "loss": 0.8622,
+      "step": 8280
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 2.875,
+      "learning_rate": 2.7e-06,
+      "loss": 0.7672,
+      "step": 8290
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 3.9375,
+      "learning_rate": 2.6842105263157895e-06,
+      "loss": 0.7785,
+      "step": 8300
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 3.71875,
+      "learning_rate": 2.668421052631579e-06,
+      "loss": 0.6706,
+      "step": 8310
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 3.484375,
+      "learning_rate": 2.6526315789473685e-06,
+      "loss": 0.7921,
+      "step": 8320
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 3.0,
+      "learning_rate": 2.636842105263158e-06,
+      "loss": 0.8154,
+      "step": 8330
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 4.21875,
+      "learning_rate": 2.6210526315789474e-06,
+      "loss": 0.7711,
+      "step": 8340
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 3.796875,
+      "learning_rate": 2.605263157894737e-06,
+      "loss": 0.8248,
+      "step": 8350
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 3.109375,
+      "learning_rate": 2.5894736842105264e-06,
+      "loss": 0.838,
+      "step": 8360
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 3.5,
+      "learning_rate": 2.573684210526316e-06,
+      "loss": 0.7573,
+      "step": 8370
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 5.21875,
+      "learning_rate": 2.5578947368421054e-06,
+      "loss": 0.7551,
+      "step": 8380
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 4.5,
+      "learning_rate": 2.5421052631578945e-06,
+      "loss": 0.7525,
+      "step": 8390
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 3.53125,
+      "learning_rate": 2.5263157894736844e-06,
+      "loss": 0.7289,
+      "step": 8400
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 4.625,
+      "learning_rate": 2.510526315789474e-06,
+      "loss": 0.7806,
+      "step": 8410
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 4.1875,
+      "learning_rate": 2.4947368421052634e-06,
+      "loss": 0.7476,
+      "step": 8420
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 3.171875,
+      "learning_rate": 2.4789473684210524e-06,
+      "loss": 0.7582,
+      "step": 8430
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 3.078125,
+      "learning_rate": 2.4631578947368424e-06,
+      "loss": 0.7983,
+      "step": 8440
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 3.828125,
+      "learning_rate": 2.447368421052632e-06,
+      "loss": 0.856,
+      "step": 8450
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 4.53125,
+      "learning_rate": 2.431578947368421e-06,
+      "loss": 0.7705,
+      "step": 8460
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 3.046875,
+      "learning_rate": 2.4157894736842104e-06,
+      "loss": 0.8032,
+      "step": 8470
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 3.890625,
+      "learning_rate": 2.4000000000000003e-06,
+      "loss": 0.7947,
+      "step": 8480
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 4.03125,
+      "learning_rate": 2.38421052631579e-06,
+      "loss": 0.8456,
+      "step": 8490
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 4.5,
+      "learning_rate": 2.368421052631579e-06,
+      "loss": 0.7104,
+      "step": 8500
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 3.890625,
+      "learning_rate": 2.3526315789473684e-06,
+      "loss": 0.711,
+      "step": 8510
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 4.65625,
+      "learning_rate": 2.336842105263158e-06,
+      "loss": 0.859,
+      "step": 8520
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 3.515625,
+      "learning_rate": 2.3210526315789473e-06,
+      "loss": 0.8214,
+      "step": 8530
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 4.34375,
+      "learning_rate": 2.305263157894737e-06,
+      "loss": 0.8534,
+      "step": 8540
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 3.109375,
+      "learning_rate": 2.2894736842105263e-06,
+      "loss": 0.7356,
+      "step": 8550
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 5.46875,
+      "learning_rate": 2.273684210526316e-06,
+      "loss": 0.6943,
+      "step": 8560
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 3.34375,
+      "learning_rate": 2.2578947368421053e-06,
+      "loss": 0.7523,
+      "step": 8570
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 3.78125,
+      "learning_rate": 2.242105263157895e-06,
+      "loss": 0.7882,
+      "step": 8580
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 4.6875,
+      "learning_rate": 2.2263157894736843e-06,
+      "loss": 0.7794,
+      "step": 8590
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 2.671875,
+      "learning_rate": 2.2105263157894734e-06,
+      "loss": 0.6316,
+      "step": 8600
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 3.296875,
+      "learning_rate": 2.1947368421052633e-06,
+      "loss": 0.7972,
+      "step": 8610
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 3.015625,
+      "learning_rate": 2.1789473684210528e-06,
+      "loss": 0.8178,
+      "step": 8620
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 3.875,
+      "learning_rate": 2.1631578947368423e-06,
+      "loss": 0.7696,
+      "step": 8630
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 3.125,
+      "learning_rate": 2.1473684210526313e-06,
+      "loss": 0.7796,
+      "step": 8640
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 3.625,
+      "learning_rate": 2.1315789473684212e-06,
+      "loss": 0.7402,
+      "step": 8650
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 3.1875,
+      "learning_rate": 2.1157894736842107e-06,
+      "loss": 0.6745,
+      "step": 8660
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 4.46875,
+      "learning_rate": 2.1000000000000002e-06,
+      "loss": 0.8412,
+      "step": 8670
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 5.03125,
+      "learning_rate": 2.0842105263157893e-06,
+      "loss": 0.7342,
+      "step": 8680
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 3.65625,
+      "learning_rate": 2.068421052631579e-06,
+      "loss": 0.7719,
+      "step": 8690
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 4.40625,
+      "learning_rate": 2.0526315789473687e-06,
+      "loss": 0.8248,
+      "step": 8700
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 4.21875,
+      "learning_rate": 2.0368421052631578e-06,
+      "loss": 0.7602,
+      "step": 8710
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 3.703125,
+      "learning_rate": 2.0210526315789473e-06,
+      "loss": 0.708,
+      "step": 8720
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 3.0,
+      "learning_rate": 2.005263157894737e-06,
+      "loss": 0.7468,
+      "step": 8730
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.9894736842105267e-06,
+      "loss": 0.6687,
+      "step": 8740
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 3.3125,
+      "learning_rate": 1.9736842105263157e-06,
+      "loss": 0.757,
+      "step": 8750
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 5.59375,
+      "learning_rate": 1.9578947368421052e-06,
+      "loss": 0.7564,
+      "step": 8760
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 5.375,
+      "learning_rate": 1.9421052631578947e-06,
+      "loss": 0.702,
+      "step": 8770
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 3.96875,
+      "learning_rate": 1.926315789473684e-06,
+      "loss": 0.7674,
+      "step": 8780
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 4.625,
+      "learning_rate": 1.9105263157894737e-06,
+      "loss": 0.7201,
+      "step": 8790
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 3.578125,
+      "learning_rate": 1.8947368421052632e-06,
+      "loss": 0.682,
+      "step": 8800
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 3.390625,
+      "learning_rate": 1.8789473684210529e-06,
+      "loss": 0.7114,
+      "step": 8810
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.8631578947368422e-06,
+      "loss": 0.7361,
+      "step": 8820
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 3.515625,
+      "learning_rate": 1.8473684210526317e-06,
+      "loss": 0.7711,
+      "step": 8830
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 4.5,
+      "learning_rate": 1.831578947368421e-06,
+      "loss": 0.7366,
+      "step": 8840
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 3.984375,
+      "learning_rate": 1.8157894736842106e-06,
+      "loss": 0.7851,
+      "step": 8850
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 4.125,
+      "learning_rate": 1.8e-06,
+      "loss": 0.7344,
+      "step": 8860
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 3.765625,
+      "learning_rate": 1.7842105263157896e-06,
+      "loss": 0.74,
+      "step": 8870
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 4.375,
+      "learning_rate": 1.768421052631579e-06,
+      "loss": 0.7402,
+      "step": 8880
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 2.96875,
+      "learning_rate": 1.7526315789473686e-06,
+      "loss": 0.7151,
+      "step": 8890
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.7368421052631579e-06,
+      "loss": 0.7165,
+      "step": 8900
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 3.296875,
+      "learning_rate": 1.7210526315789474e-06,
+      "loss": 0.8903,
+      "step": 8910
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 3.828125,
+      "learning_rate": 1.7052631578947369e-06,
+      "loss": 0.8315,
+      "step": 8920
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 6.21875,
+      "learning_rate": 1.6894736842105264e-06,
+      "loss": 0.7346,
+      "step": 8930
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.6736842105263158e-06,
+      "loss": 0.7919,
+      "step": 8940
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 3.515625,
+      "learning_rate": 1.6578947368421053e-06,
+      "loss": 0.7286,
+      "step": 8950
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 3.203125,
+      "learning_rate": 1.6421052631578948e-06,
+      "loss": 0.7419,
+      "step": 8960
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 3.21875,
+      "learning_rate": 1.6263157894736843e-06,
+      "loss": 0.8076,
+      "step": 8970
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 2.5625,
+      "learning_rate": 1.6105263157894738e-06,
+      "loss": 0.7703,
+      "step": 8980
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.5947368421052633e-06,
+      "loss": 0.6863,
+      "step": 8990
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 4.9375,
+      "learning_rate": 1.5789473684210526e-06,
+      "loss": 0.7375,
+      "step": 9000
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 4.15625,
+      "learning_rate": 1.563157894736842e-06,
+      "loss": 0.8044,
+      "step": 9010
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.5473684210526316e-06,
+      "loss": 0.7712,
+      "step": 9020
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.531578947368421e-06,
+      "loss": 0.7582,
+      "step": 9030
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.5157894736842105e-06,
+      "loss": 0.7764,
+      "step": 9040
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 3.765625,
+      "learning_rate": 1.5e-06,
+      "loss": 0.7165,
+      "step": 9050
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 3.3125,
+      "learning_rate": 1.4842105263157895e-06,
+      "loss": 0.6494,
+      "step": 9060
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 4.15625,
+      "learning_rate": 1.4684210526315788e-06,
+      "loss": 0.6998,
+      "step": 9070
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 5.375,
+      "learning_rate": 1.4526315789473685e-06,
+      "loss": 0.7737,
+      "step": 9080
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 3.1875,
+      "learning_rate": 1.4368421052631578e-06,
+      "loss": 0.8789,
+      "step": 9090
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 4.875,
+      "learning_rate": 1.4210526315789475e-06,
+      "loss": 0.7121,
+      "step": 9100
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 3.1875,
+      "learning_rate": 1.4052631578947368e-06,
+      "loss": 0.7486,
+      "step": 9110
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 4.125,
+      "learning_rate": 1.3894736842105265e-06,
+      "loss": 0.7483,
+      "step": 9120
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 3.828125,
+      "learning_rate": 1.3736842105263158e-06,
+      "loss": 0.8725,
+      "step": 9130
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 3.46875,
+      "learning_rate": 1.3578947368421055e-06,
+      "loss": 0.7745,
+      "step": 9140
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 3.296875,
+      "learning_rate": 1.3421052631578947e-06,
+      "loss": 0.7438,
+      "step": 9150
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.3263157894736842e-06,
+      "loss": 0.8,
+      "step": 9160
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 3.296875,
+      "learning_rate": 1.3105263157894737e-06,
+      "loss": 0.7779,
+      "step": 9170
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 3.03125,
+      "learning_rate": 1.2947368421052632e-06,
+      "loss": 0.7862,
+      "step": 9180
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 2.90625,
+      "learning_rate": 1.2789473684210527e-06,
+      "loss": 0.7818,
+      "step": 9190
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.2631578947368422e-06,
+      "loss": 0.7169,
+      "step": 9200
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.2473684210526317e-06,
+      "loss": 0.8331,
+      "step": 9210
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 4.375,
+      "learning_rate": 1.2315789473684212e-06,
+      "loss": 0.7177,
+      "step": 9220
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 3.46875,
+      "learning_rate": 1.2157894736842105e-06,
+      "loss": 0.7036,
+      "step": 9230
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 0.8276,
+      "step": 9240
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.1842105263157894e-06,
+      "loss": 0.8222,
+      "step": 9250
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.168421052631579e-06,
+      "loss": 0.7487,
+      "step": 9260
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 3.96875,
+      "learning_rate": 1.1526315789473684e-06,
+      "loss": 0.8026,
+      "step": 9270
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.136842105263158e-06,
+      "loss": 0.7733,
+      "step": 9280
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 3.484375,
+      "learning_rate": 1.1210526315789474e-06,
+      "loss": 0.8602,
+      "step": 9290
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.1052631578947367e-06,
+      "loss": 0.8904,
+      "step": 9300
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 2.6875,
+      "learning_rate": 1.0894736842105264e-06,
+      "loss": 0.7111,
+      "step": 9310
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 3.8125,
+      "learning_rate": 1.0736842105263157e-06,
+      "loss": 0.7556,
+      "step": 9320
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.0578947368421054e-06,
+      "loss": 0.7822,
+      "step": 9330
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.0421052631578946e-06,
+      "loss": 0.7508,
+      "step": 9340
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 5.375,
+      "learning_rate": 1.0263157894736843e-06,
+      "loss": 0.893,
+      "step": 9350
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 5.46875,
+      "learning_rate": 1.0105263157894736e-06,
+      "loss": 0.8094,
+      "step": 9360
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 3.6875,
+      "learning_rate": 9.947368421052633e-07,
+      "loss": 0.8216,
+      "step": 9370
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 4.21875,
+      "learning_rate": 9.789473684210526e-07,
+      "loss": 0.7432,
+      "step": 9380
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 3.515625,
+      "learning_rate": 9.63157894736842e-07,
+      "loss": 0.8025,
+      "step": 9390
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 3.515625,
+      "learning_rate": 9.473684210526316e-07,
+      "loss": 0.7242,
+      "step": 9400
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 4.59375,
+      "learning_rate": 9.315789473684211e-07,
+      "loss": 0.82,
+      "step": 9410
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 4.125,
+      "learning_rate": 9.157894736842105e-07,
+      "loss": 0.7063,
+      "step": 9420
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 4.8125,
+      "learning_rate": 9e-07,
+      "loss": 0.7271,
+      "step": 9430
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 4.03125,
+      "learning_rate": 8.842105263157895e-07,
+      "loss": 0.7798,
+      "step": 9440
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 4.40625,
+      "learning_rate": 8.684210526315789e-07,
+      "loss": 0.7326,
+      "step": 9450
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 4.1875,
+      "learning_rate": 8.526315789473684e-07,
+      "loss": 0.822,
+      "step": 9460
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 3.40625,
+      "learning_rate": 8.368421052631579e-07,
+      "loss": 0.8049,
+      "step": 9470
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 4.53125,
+      "learning_rate": 8.210526315789474e-07,
+      "loss": 0.8065,
+      "step": 9480
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 3.96875,
+      "learning_rate": 8.052631578947369e-07,
+      "loss": 0.7411,
+      "step": 9490
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 4.8125,
+      "learning_rate": 7.894736842105263e-07,
+      "loss": 0.8655,
+      "step": 9500
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 4.4375,
+      "learning_rate": 7.736842105263158e-07,
+      "loss": 0.7711,
+      "step": 9510
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 5.15625,
+      "learning_rate": 7.578947368421053e-07,
+      "loss": 0.79,
+      "step": 9520
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 4.9375,
+      "learning_rate": 7.421052631578948e-07,
+      "loss": 0.7385,
+      "step": 9530
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 3.46875,
+      "learning_rate": 7.263157894736843e-07,
+      "loss": 0.8351,
+      "step": 9540
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 3.40625,
+      "learning_rate": 7.105263157894737e-07,
+      "loss": 0.7739,
+      "step": 9550
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 3.578125,
+      "learning_rate": 6.947368421052632e-07,
+      "loss": 0.7365,
+      "step": 9560
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 2.890625,
+      "learning_rate": 6.789473684210527e-07,
+      "loss": 0.8042,
+      "step": 9570
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 3.984375,
+      "learning_rate": 6.631578947368421e-07,
+      "loss": 0.6655,
+      "step": 9580
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 4.59375,
+      "learning_rate": 6.473684210526316e-07,
+      "loss": 0.7557,
+      "step": 9590
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 4.1875,
+      "learning_rate": 6.315789473684211e-07,
+      "loss": 0.834,
+      "step": 9600
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 3.359375,
+      "learning_rate": 6.157894736842106e-07,
+      "loss": 0.7044,
+      "step": 9610
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 4.59375,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 0.786,
+      "step": 9620
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 5.65625,
+      "learning_rate": 5.842105263157895e-07,
+      "loss": 0.7075,
+      "step": 9630
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 3.78125,
+      "learning_rate": 5.68421052631579e-07,
+      "loss": 0.7572,
+      "step": 9640
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 2.765625,
+      "learning_rate": 5.526315789473683e-07,
+      "loss": 0.8421,
+      "step": 9650
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 3.3125,
+      "learning_rate": 5.368421052631578e-07,
+      "loss": 0.7933,
+      "step": 9660
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 4.4375,
+      "learning_rate": 5.210526315789473e-07,
+      "loss": 0.8427,
+      "step": 9670
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 3.875,
+      "learning_rate": 5.052631578947368e-07,
+      "loss": 0.7658,
+      "step": 9680
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 3.046875,
+      "learning_rate": 4.894736842105263e-07,
+      "loss": 0.7417,
+      "step": 9690
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 3.609375,
+      "learning_rate": 4.736842105263158e-07,
+      "loss": 0.826,
+      "step": 9700
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 3.546875,
+      "learning_rate": 4.5789473684210523e-07,
+      "loss": 0.7745,
+      "step": 9710
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 5.09375,
+      "learning_rate": 4.421052631578947e-07,
+      "loss": 0.8953,
+      "step": 9720
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 2.65625,
+      "learning_rate": 4.263157894736842e-07,
+      "loss": 0.7377,
+      "step": 9730
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 4.15625,
+      "learning_rate": 4.105263157894737e-07,
+      "loss": 0.7578,
+      "step": 9740
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 3.828125,
+      "learning_rate": 3.9473684210526315e-07,
+      "loss": 0.6716,
+      "step": 9750
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 4.15625,
+      "learning_rate": 3.7894736842105264e-07,
+      "loss": 0.7195,
+      "step": 9760
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 4.125,
+      "learning_rate": 3.6315789473684213e-07,
+      "loss": 0.8409,
+      "step": 9770
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 4.15625,
+      "learning_rate": 3.473684210526316e-07,
+      "loss": 0.7313,
+      "step": 9780
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 3.40625,
+      "learning_rate": 3.3157894736842106e-07,
+      "loss": 0.7575,
+      "step": 9790
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 4.25,
+      "learning_rate": 3.1578947368421055e-07,
+      "loss": 0.8704,
+      "step": 9800
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 4.59375,
+      "learning_rate": 3.0000000000000004e-07,
+      "loss": 0.7686,
+      "step": 9810
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 4.125,
+      "learning_rate": 2.842105263157895e-07,
+      "loss": 0.8014,
+      "step": 9820
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 3.328125,
+      "learning_rate": 2.684210526315789e-07,
+      "loss": 0.8103,
+      "step": 9830
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 4.625,
+      "learning_rate": 2.526315789473684e-07,
+      "loss": 0.6381,
+      "step": 9840
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 3.625,
+      "learning_rate": 2.368421052631579e-07,
+      "loss": 0.7611,
+      "step": 9850
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 4.9375,
+      "learning_rate": 2.2105263157894736e-07,
+      "loss": 0.724,
+      "step": 9860
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 3.875,
+      "learning_rate": 2.0526315789473685e-07,
+      "loss": 0.5638,
+      "step": 9870
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 3.203125,
+      "learning_rate": 1.8947368421052632e-07,
+      "loss": 0.811,
+      "step": 9880
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.736842105263158e-07,
+      "loss": 0.6505,
+      "step": 9890
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 4.625,
+      "learning_rate": 1.5789473684210527e-07,
+      "loss": 0.6778,
+      "step": 9900
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.4210526315789474e-07,
+      "loss": 0.7813,
+      "step": 9910
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.263157894736842e-07,
+      "loss": 0.6871,
+      "step": 9920
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.1052631578947368e-07,
+      "loss": 0.6846,
+      "step": 9930
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 3.375,
+      "learning_rate": 9.473684210526316e-08,
+      "loss": 0.7486,
+      "step": 9940
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 4.8125,
+      "learning_rate": 7.894736842105264e-08,
+      "loss": 0.8227,
+      "step": 9950
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 3.984375,
+      "learning_rate": 6.31578947368421e-08,
+      "loss": 0.6844,
+      "step": 9960
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 4.6875,
+      "learning_rate": 4.736842105263158e-08,
+      "loss": 0.8294,
+      "step": 9970
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 3.359375,
+      "learning_rate": 3.157894736842105e-08,
+      "loss": 0.838,
+      "step": 9980
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 3.59375,
+      "learning_rate": 1.5789473684210525e-08,
+      "loss": 0.694,
+      "step": 9990
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 4.125,
+      "learning_rate": 0.0,
+      "loss": 0.7115,
+      "step": 10000
+    },
+    {
+      "epoch": 3.59,
+      "eval_loss": 0.9110761880874634,
+      "eval_runtime": 64.8981,
+      "eval_samples_per_second": 15.409,
+      "eval_steps_per_second": 15.409,
+      "step": 10000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 2000,
+  "total_flos": 1.613922041856e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}