diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7061 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.5,
+  "eval_steps": 2000,
+  "global_step": 10000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 20.5,
+      "learning_rate": 2.9999999999999997e-06,
+      "loss": 1.7805,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 19.625,
+      "learning_rate": 5.999999999999999e-06,
+      "loss": 1.8292,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 16.875,
+      "learning_rate": 8.999999999999999e-06,
+      "loss": 2.0677,
+      "step": 30
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 17.875,
+      "learning_rate": 1.1999999999999999e-05,
+      "loss": 1.6027,
+      "step": 40
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 21.625,
+      "learning_rate": 1.4999999999999999e-05,
+      "loss": 1.7499,
+      "step": 50
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 21.125,
+      "learning_rate": 1.7999999999999997e-05,
+      "loss": 2.0222,
+      "step": 60
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 22.75,
+      "learning_rate": 2.1e-05,
+      "loss": 1.4821,
+      "step": 70
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 25.75,
+      "learning_rate": 2.3999999999999997e-05,
+      "loss": 1.7637,
+      "step": 80
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 17.25,
+      "learning_rate": 2.6999999999999996e-05,
+      "loss": 1.6784,
+      "step": 90
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 11.9375,
+      "learning_rate": 2.9999999999999997e-05,
+      "loss": 1.6543,
+      "step": 100
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 25.875,
+      "learning_rate": 3.2999999999999996e-05,
+      "loss": 1.7161,
+      "step": 110
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 38.5,
+      "learning_rate": 3.5999999999999994e-05,
+      "loss": 1.8143,
+      "step": 120
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 19.375,
+      "learning_rate": 3.9e-05,
+      "loss": 1.5947,
+      "step": 130
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 16.5,
+      "learning_rate": 4.2e-05,
+      "loss": 2.1477,
+      "step": 140
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 18.875,
+      "learning_rate": 4.4999999999999996e-05,
+      "loss": 1.8095,
+      "step": 150
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 17.875,
+      "learning_rate": 4.7999999999999994e-05,
+      "loss": 1.7693,
+      "step": 160
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 12.9375,
+      "learning_rate": 5.1e-05,
+      "loss": 1.9624,
+      "step": 170
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 11.4375,
+      "learning_rate": 5.399999999999999e-05,
+      "loss": 1.7321,
+      "step": 180
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 18.125,
+      "learning_rate": 5.6999999999999996e-05,
+      "loss": 1.7955,
+      "step": 190
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 15.9375,
+      "learning_rate": 5.9999999999999995e-05,
+      "loss": 1.7011,
+      "step": 200
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 14.0625,
+      "learning_rate": 6.299999999999999e-05,
+      "loss": 1.5647,
+      "step": 210
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 12.6875,
+      "learning_rate": 6.599999999999999e-05,
+      "loss": 1.8383,
+      "step": 220
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 16.125,
+      "learning_rate": 6.9e-05,
+      "loss": 1.9608,
+      "step": 230
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 12.5,
+      "learning_rate": 7.199999999999999e-05,
+      "loss": 1.4068,
+      "step": 240
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 11.0,
+      "learning_rate": 7.5e-05,
+      "loss": 1.9903,
+      "step": 250
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 10.1875,
+      "learning_rate": 7.8e-05,
+      "loss": 2.0868,
+      "step": 260
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 14.8125,
+      "learning_rate": 8.1e-05,
+      "loss": 2.0174,
+      "step": 270
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 10.6875,
+      "learning_rate": 8.4e-05,
+      "loss": 2.0084,
+      "step": 280
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 10.375,
+      "learning_rate": 8.699999999999999e-05,
+      "loss": 1.9688,
+      "step": 290
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 10.625,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 1.9387,
+      "step": 300
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 10.875,
+      "learning_rate": 9.3e-05,
+      "loss": 2.1617,
+      "step": 310
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 10.5625,
+      "learning_rate": 9.599999999999999e-05,
+      "loss": 1.9997,
+      "step": 320
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 8.9375,
+      "learning_rate": 9.9e-05,
+      "loss": 1.5052,
+      "step": 330
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 10.4375,
+      "learning_rate": 0.000102,
+      "loss": 2.1021,
+      "step": 340
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 11.3125,
+      "learning_rate": 0.00010499999999999999,
+      "loss": 1.8825,
+      "step": 350
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 13.625,
+      "learning_rate": 0.00010799999999999998,
+      "loss": 1.9024,
+      "step": 360
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 18.125,
+      "learning_rate": 0.00011099999999999999,
+      "loss": 2.2218,
+      "step": 370
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 12.6875,
+      "learning_rate": 0.00011399999999999999,
+      "loss": 1.9759,
+      "step": 380
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 10.5625,
+      "learning_rate": 0.000117,
+      "loss": 1.9719,
+      "step": 390
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.09375,
+      "learning_rate": 0.00011999999999999999,
+      "loss": 1.7527,
+      "step": 400
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 8.5,
+      "learning_rate": 0.00012299999999999998,
+      "loss": 2.3111,
+      "step": 410
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.9375,
+      "learning_rate": 0.00012599999999999997,
+      "loss": 2.185,
+      "step": 420
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 8.375,
+      "learning_rate": 0.000129,
+      "loss": 2.0823,
+      "step": 430
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 9.5,
+      "learning_rate": 0.00013199999999999998,
+      "loss": 2.155,
+      "step": 440
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 8.5625,
+      "learning_rate": 0.000135,
+      "loss": 2.1027,
+      "step": 450
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.75,
+      "learning_rate": 0.000138,
+      "loss": 2.3035,
+      "step": 460
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.875,
+      "learning_rate": 0.00014099999999999998,
+      "loss": 1.9003,
+      "step": 470
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 9.125,
+      "learning_rate": 0.00014399999999999998,
+      "loss": 2.4984,
+      "step": 480
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 8.0,
+      "learning_rate": 0.000147,
+      "loss": 2.1406,
+      "step": 490
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 8.625,
+      "learning_rate": 0.00015,
+      "loss": 2.0724,
+      "step": 500
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.8125,
+      "learning_rate": 0.00014984210526315788,
+      "loss": 2.2684,
+      "step": 510
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.03125,
+      "learning_rate": 0.00014968421052631578,
+      "loss": 2.0504,
+      "step": 520
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 8.125,
+      "learning_rate": 0.00014952631578947368,
+      "loss": 2.3278,
+      "step": 530
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.5625,
+      "learning_rate": 0.00014936842105263157,
+      "loss": 2.1526,
+      "step": 540
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.90625,
+      "learning_rate": 0.00014921052631578947,
+      "loss": 2.3046,
+      "step": 550
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.625,
+      "learning_rate": 0.00014905263157894737,
+      "loss": 2.1977,
+      "step": 560
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.25,
+      "learning_rate": 0.00014889473684210526,
+      "loss": 2.0804,
+      "step": 570
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.65625,
+      "learning_rate": 0.00014873684210526313,
+      "loss": 1.9854,
+      "step": 580
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.46875,
+      "learning_rate": 0.00014857894736842103,
+      "loss": 2.319,
+      "step": 590
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.53125,
+      "learning_rate": 0.00014842105263157893,
+      "loss": 2.5894,
+      "step": 600
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.625,
+      "learning_rate": 0.00014826315789473682,
+      "loss": 1.7258,
+      "step": 610
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.3125,
+      "learning_rate": 0.00014810526315789472,
+      "loss": 2.1868,
+      "step": 620
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.125,
+      "learning_rate": 0.00014794736842105262,
+      "loss": 2.2646,
+      "step": 630
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.0001477894736842105,
+      "loss": 2.0835,
+      "step": 640
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.0625,
+      "learning_rate": 0.0001476315789473684,
+      "loss": 2.1409,
+      "step": 650
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.84375,
+      "learning_rate": 0.0001474736842105263,
+      "loss": 2.0042,
+      "step": 660
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.0001473157894736842,
+      "loss": 2.3385,
+      "step": 670
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 4.40625,
+      "learning_rate": 0.0001471578947368421,
+      "loss": 2.0918,
+      "step": 680
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.000147,
+      "loss": 1.9163,
+      "step": 690
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 9.1875,
+      "learning_rate": 0.0001468421052631579,
+      "loss": 2.2878,
+      "step": 700
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.0001466842105263158,
+      "loss": 2.0185,
+      "step": 710
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.03125,
+      "learning_rate": 0.00014652631578947366,
+      "loss": 2.1358,
+      "step": 720
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.00014636842105263158,
+      "loss": 2.4189,
+      "step": 730
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.40625,
+      "learning_rate": 0.00014621052631578945,
+      "loss": 1.9183,
+      "step": 740
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.40625,
+      "learning_rate": 0.00014605263157894735,
+      "loss": 2.2969,
+      "step": 750
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.625,
+      "learning_rate": 0.00014589473684210524,
+      "loss": 2.2832,
+      "step": 760
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.625,
+      "learning_rate": 0.00014573684210526314,
+      "loss": 2.1994,
+      "step": 770
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.5625,
+      "learning_rate": 0.00014557894736842104,
+      "loss": 2.0053,
+      "step": 780
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.65625,
+      "learning_rate": 0.00014542105263157893,
+      "loss": 2.2771,
+      "step": 790
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.4375,
+      "learning_rate": 0.00014526315789473683,
+      "loss": 2.1748,
+      "step": 800
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.59375,
+      "learning_rate": 0.00014510526315789473,
+      "loss": 1.9644,
+      "step": 810
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.09375,
+      "learning_rate": 0.00014494736842105262,
+      "loss": 1.6369,
+      "step": 820
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.1875,
+      "learning_rate": 0.00014478947368421052,
+      "loss": 2.2513,
+      "step": 830
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.6875,
+      "learning_rate": 0.00014463157894736842,
+      "loss": 2.0433,
+      "step": 840
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.46875,
+      "learning_rate": 0.00014447368421052631,
+      "loss": 1.8816,
+      "step": 850
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.75,
+      "learning_rate": 0.00014431578947368418,
+      "loss": 2.0584,
+      "step": 860
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.0001441578947368421,
+      "loss": 1.9437,
+      "step": 870
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.625,
+      "learning_rate": 0.00014399999999999998,
+      "loss": 2.1637,
+      "step": 880
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.34375,
+      "learning_rate": 0.00014384210526315787,
+      "loss": 1.6931,
+      "step": 890
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.15625,
+      "learning_rate": 0.00014368421052631577,
+      "loss": 2.1308,
+      "step": 900
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.75,
+      "learning_rate": 0.00014352631578947367,
+      "loss": 1.9802,
+      "step": 910
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.3125,
+      "learning_rate": 0.00014336842105263156,
+      "loss": 1.8052,
+      "step": 920
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.46875,
+      "learning_rate": 0.00014321052631578946,
+      "loss": 2.2819,
+      "step": 930
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.25,
+      "learning_rate": 0.00014305263157894736,
+      "loss": 1.9152,
+      "step": 940
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 7.03125,
+      "learning_rate": 0.00014289473684210525,
+      "loss": 1.6635,
+      "step": 950
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.00014273684210526315,
+      "loss": 1.8765,
+      "step": 960
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.21875,
+      "learning_rate": 0.00014257894736842105,
+      "loss": 2.5049,
+      "step": 970
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.28125,
+      "learning_rate": 0.00014242105263157894,
+      "loss": 1.8173,
+      "step": 980
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.609375,
+      "learning_rate": 0.00014226315789473684,
+      "loss": 1.8379,
+      "step": 990
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 7.40625,
+      "learning_rate": 0.0001421052631578947,
+      "loss": 2.0426,
+      "step": 1000
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 7.25,
+      "learning_rate": 0.00014194736842105263,
+      "loss": 2.1772,
+      "step": 1010
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.09375,
+      "learning_rate": 0.0001417894736842105,
+      "loss": 1.947,
+      "step": 1020
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.125,
+      "learning_rate": 0.00014163157894736843,
+      "loss": 2.2714,
+      "step": 1030
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.125,
+      "learning_rate": 0.0001414736842105263,
+      "loss": 1.8783,
+      "step": 1040
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.40625,
+      "learning_rate": 0.0001413157894736842,
+      "loss": 2.409,
+      "step": 1050
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.96875,
+      "learning_rate": 0.0001411578947368421,
+      "loss": 2.0452,
+      "step": 1060
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 4.75,
+      "learning_rate": 0.00014099999999999998,
+      "loss": 1.9155,
+      "step": 1070
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.65625,
+      "learning_rate": 0.00014084210526315788,
+      "loss": 1.9305,
+      "step": 1080
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.53125,
+      "learning_rate": 0.00014068421052631578,
+      "loss": 2.0101,
+      "step": 1090
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.9375,
+      "learning_rate": 0.00014052631578947367,
+      "loss": 2.1938,
+      "step": 1100
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.15625,
+      "learning_rate": 0.00014036842105263157,
+      "loss": 1.8036,
+      "step": 1110
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.65625,
+      "learning_rate": 0.00014021052631578947,
+      "loss": 1.8334,
+      "step": 1120
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.53125,
+      "learning_rate": 0.00014005263157894736,
+      "loss": 1.7881,
+      "step": 1130
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 7.5625,
+      "learning_rate": 0.00013989473684210523,
+      "loss": 1.9495,
+      "step": 1140
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 7.46875,
+      "learning_rate": 0.00013973684210526316,
+      "loss": 2.2284,
+      "step": 1150
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.28125,
+      "learning_rate": 0.00013957894736842103,
+      "loss": 2.3546,
+      "step": 1160
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.46875,
+      "learning_rate": 0.00013942105263157895,
+      "loss": 2.1805,
+      "step": 1170
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.78125,
+      "learning_rate": 0.00013926315789473682,
+      "loss": 2.1401,
+      "step": 1180
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.00013910526315789474,
+      "loss": 1.9132,
+      "step": 1190
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.9375,
+      "learning_rate": 0.00013894736842105261,
+      "loss": 1.9242,
+      "step": 1200
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.34375,
+      "learning_rate": 0.0001387894736842105,
+      "loss": 2.3808,
+      "step": 1210
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.0,
+      "learning_rate": 0.0001386315789473684,
+      "loss": 1.8303,
+      "step": 1220
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.34375,
+      "learning_rate": 0.0001384736842105263,
+      "loss": 2.1689,
+      "step": 1230
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.9375,
+      "learning_rate": 0.0001383157894736842,
+      "loss": 2.1055,
+      "step": 1240
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.4375,
+      "learning_rate": 0.0001381578947368421,
+      "loss": 2.1595,
+      "step": 1250
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.5625,
+      "learning_rate": 0.000138,
+      "loss": 2.2264,
+      "step": 1260
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 4.25,
+      "learning_rate": 0.0001378421052631579,
+      "loss": 2.0981,
+      "step": 1270
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.90625,
+      "learning_rate": 0.00013768421052631576,
+      "loss": 2.0902,
+      "step": 1280
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.125,
+      "learning_rate": 0.00013752631578947368,
+      "loss": 1.9157,
+      "step": 1290
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.5625,
+      "learning_rate": 0.00013736842105263155,
+      "loss": 1.7149,
+      "step": 1300
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.00013721052631578948,
+      "loss": 2.0319,
+      "step": 1310
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.00013705263157894735,
+      "loss": 1.9478,
+      "step": 1320
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.0,
+      "learning_rate": 0.00013689473684210527,
+      "loss": 1.7866,
+      "step": 1330
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.1875,
+      "learning_rate": 0.00013673684210526314,
+      "loss": 2.0339,
+      "step": 1340
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.125,
+      "learning_rate": 0.00013657894736842104,
+      "loss": 1.7241,
+      "step": 1350
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.25,
+      "learning_rate": 0.00013642105263157893,
+      "loss": 1.8559,
+      "step": 1360
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.21875,
+      "learning_rate": 0.00013626315789473683,
+      "loss": 1.847,
+      "step": 1370
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 7.0625,
+      "learning_rate": 0.00013610526315789473,
+      "loss": 1.8128,
+      "step": 1380
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.71875,
+      "learning_rate": 0.00013594736842105262,
+      "loss": 2.0117,
+      "step": 1390
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.1875,
+      "learning_rate": 0.00013578947368421052,
+      "loss": 2.319,
+      "step": 1400
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 10.5,
+      "learning_rate": 0.00013563157894736842,
+      "loss": 2.0572,
+      "step": 1410
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.0001354736842105263,
+      "loss": 1.9633,
+      "step": 1420
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.046875,
+      "learning_rate": 0.0001353157894736842,
+      "loss": 1.9709,
+      "step": 1430
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.1875,
+      "learning_rate": 0.00013515789473684208,
+      "loss": 1.9295,
+      "step": 1440
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.125,
+      "learning_rate": 0.000135,
+      "loss": 1.7119,
+      "step": 1450
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.21875,
+      "learning_rate": 0.00013484210526315787,
+      "loss": 1.8842,
+      "step": 1460
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.90625,
+      "learning_rate": 0.0001346842105263158,
+      "loss": 1.8175,
+      "step": 1470
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 7.28125,
+      "learning_rate": 0.00013452631578947366,
+      "loss": 2.3633,
+      "step": 1480
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.875,
+      "learning_rate": 0.0001343684210526316,
+      "loss": 2.2177,
+      "step": 1490
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.28125,
+      "learning_rate": 0.00013421052631578946,
+      "loss": 1.9366,
+      "step": 1500
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.09375,
+      "learning_rate": 0.00013405263157894735,
+      "loss": 1.9658,
+      "step": 1510
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.00013389473684210525,
+      "loss": 1.9321,
+      "step": 1520
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.03125,
+      "learning_rate": 0.00013373684210526315,
+      "loss": 1.8851,
+      "step": 1530
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.00013357894736842104,
+      "loss": 2.3404,
+      "step": 1540
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.34375,
+      "learning_rate": 0.00013342105263157894,
+      "loss": 1.9765,
+      "step": 1550
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.40625,
+      "learning_rate": 0.00013326315789473684,
+      "loss": 2.2503,
+      "step": 1560
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.53125,
+      "learning_rate": 0.00013310526315789473,
+      "loss": 1.9346,
+      "step": 1570
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.75,
+      "learning_rate": 0.0001329473684210526,
+      "loss": 2.0895,
+      "step": 1580
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.4375,
+      "learning_rate": 0.00013278947368421053,
+      "loss": 1.956,
+      "step": 1590
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.4375,
+      "learning_rate": 0.0001326315789473684,
+      "loss": 1.8528,
+      "step": 1600
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.00013247368421052632,
+      "loss": 1.9809,
+      "step": 1610
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.03125,
+      "learning_rate": 0.0001323157894736842,
+      "loss": 1.8755,
+      "step": 1620
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.78125,
+      "learning_rate": 0.0001321578947368421,
+      "loss": 1.9421,
+      "step": 1630
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.0625,
+      "learning_rate": 0.00013199999999999998,
+      "loss": 1.8252,
+      "step": 1640
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.1875,
+      "learning_rate": 0.00013184210526315788,
+      "loss": 2.0638,
+      "step": 1650
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.34375,
+      "learning_rate": 0.00013168421052631578,
+      "loss": 1.9106,
+      "step": 1660
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.40625,
+      "learning_rate": 0.00013152631578947367,
+      "loss": 1.7636,
+      "step": 1670
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.65625,
+      "learning_rate": 0.00013136842105263157,
+      "loss": 1.7,
+      "step": 1680
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 4.53125,
+      "learning_rate": 0.00013121052631578947,
+      "loss": 1.9783,
+      "step": 1690
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.75,
+      "learning_rate": 0.00013105263157894736,
+      "loss": 1.8911,
+      "step": 1700
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.00013089473684210526,
+      "loss": 1.6132,
+      "step": 1710
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.96875,
+      "learning_rate": 0.00013073684210526313,
+      "loss": 1.825,
+      "step": 1720
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.78125,
+      "learning_rate": 0.00013057894736842105,
+      "loss": 1.8831,
+      "step": 1730
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.8125,
+      "learning_rate": 0.00013042105263157892,
+      "loss": 2.0333,
+      "step": 1740
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.00013026315789473685,
+      "loss": 1.5719,
+      "step": 1750
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.28125,
+      "learning_rate": 0.00013010526315789472,
+      "loss": 1.6725,
+      "step": 1760
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.625,
+      "learning_rate": 0.00012994736842105264,
+      "loss": 1.5617,
+      "step": 1770
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.296875,
+      "learning_rate": 0.0001297894736842105,
+      "loss": 1.969,
+      "step": 1780
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.46875,
+      "learning_rate": 0.0001296315789473684,
+      "loss": 1.9816,
+      "step": 1790
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.40625,
+      "learning_rate": 0.0001294736842105263,
+      "loss": 1.8061,
+      "step": 1800
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.78125,
+      "learning_rate": 0.0001293157894736842,
+      "loss": 1.796,
+      "step": 1810
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.4375,
+      "learning_rate": 0.0001291578947368421,
+      "loss": 1.9946,
+      "step": 1820
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.0,
+      "learning_rate": 0.000129,
+      "loss": 1.9254,
+      "step": 1830
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.84375,
+      "learning_rate": 0.0001288421052631579,
+      "loss": 1.9615,
+      "step": 1840
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.25,
+      "learning_rate": 0.00012868421052631578,
+      "loss": 1.9225,
+      "step": 1850
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.00012852631578947365,
+      "loss": 1.7678,
+      "step": 1860
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.921875,
+      "learning_rate": 0.00012836842105263158,
+      "loss": 2.0424,
+      "step": 1870
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.125,
+      "learning_rate": 0.00012821052631578945,
+      "loss": 1.6981,
+      "step": 1880
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.0625,
+      "learning_rate": 0.00012805263157894737,
+      "loss": 2.013,
+      "step": 1890
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.46875,
+      "learning_rate": 0.00012789473684210524,
+      "loss": 1.7872,
+      "step": 1900
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.84375,
+      "learning_rate": 0.00012773684210526316,
+      "loss": 1.7634,
+      "step": 1910
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.78125,
+      "learning_rate": 0.00012757894736842103,
+      "loss": 1.4218,
+      "step": 1920
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.125,
+      "learning_rate": 0.00012742105263157893,
+      "loss": 1.9503,
+      "step": 1930
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.00012726315789473683,
+      "loss": 1.6645,
+      "step": 1940
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.21875,
+      "learning_rate": 0.00012710526315789472,
+      "loss": 1.6245,
+      "step": 1950
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.09375,
+      "learning_rate": 0.00012694736842105262,
+      "loss": 1.934,
+      "step": 1960
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.28125,
+      "learning_rate": 0.00012678947368421052,
+      "loss": 1.696,
+      "step": 1970
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.59375,
+      "learning_rate": 0.0001266315789473684,
+      "loss": 1.5946,
+      "step": 1980
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.0625,
+      "learning_rate": 0.0001264736842105263,
+      "loss": 1.6951,
+      "step": 1990
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.921875,
+      "learning_rate": 0.00012631578947368418,
+      "loss": 2.1469,
+      "step": 2000
+    },
+    {
+      "epoch": 0.1,
+      "eval_loss": 1.864747405052185,
+      "eval_runtime": 30.7527,
+      "eval_samples_per_second": 32.517,
+      "eval_steps_per_second": 32.517,
+      "step": 2000
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.3125,
+      "learning_rate": 0.0001261578947368421,
+      "loss": 1.7456,
+      "step": 2010
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.00012599999999999997,
+      "loss": 1.9554,
+      "step": 2020
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 5.09375,
+      "learning_rate": 0.0001258421052631579,
+      "loss": 1.8303,
+      "step": 2030
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.375,
+      "learning_rate": 0.00012568421052631577,
+      "loss": 1.7321,
+      "step": 2040
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.5625,
+      "learning_rate": 0.0001255263157894737,
+      "loss": 2.0467,
+      "step": 2050
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.625,
+      "learning_rate": 0.00012536842105263156,
+      "loss": 1.5569,
+      "step": 2060
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.3125,
+      "learning_rate": 0.00012521052631578946,
+      "loss": 1.6945,
+      "step": 2070
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.25,
+      "learning_rate": 0.00012505263157894735,
+      "loss": 1.9183,
+      "step": 2080
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.921875,
+      "learning_rate": 0.00012489473684210525,
+      "loss": 1.8488,
+      "step": 2090
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.53125,
+      "learning_rate": 0.00012473684210526315,
+      "loss": 1.7148,
+      "step": 2100
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.625,
+      "learning_rate": 0.00012457894736842104,
+      "loss": 1.9033,
+      "step": 2110
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.53125,
+      "learning_rate": 0.00012442105263157894,
+      "loss": 1.4302,
+      "step": 2120
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.59375,
+      "learning_rate": 0.00012426315789473684,
+      "loss": 1.7065,
+      "step": 2130
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.59375,
+      "learning_rate": 0.00012410526315789473,
+      "loss": 1.9168,
+      "step": 2140
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.3125,
+      "learning_rate": 0.00012394736842105263,
+      "loss": 1.993,
+      "step": 2150
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.875,
+      "learning_rate": 0.0001237894736842105,
+      "loss": 1.7291,
+      "step": 2160
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.5625,
+      "learning_rate": 0.00012363157894736842,
+      "loss": 1.6801,
+      "step": 2170
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.9375,
+      "learning_rate": 0.0001234736842105263,
+      "loss": 1.4205,
+      "step": 2180
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.78125,
+      "learning_rate": 0.00012331578947368421,
+      "loss": 1.8141,
+      "step": 2190
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.75,
+      "learning_rate": 0.00012315789473684208,
+      "loss": 1.9035,
+      "step": 2200
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.15625,
+      "learning_rate": 0.00012299999999999998,
+      "loss": 1.846,
+      "step": 2210
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.921875,
+      "learning_rate": 0.00012284210526315788,
+      "loss": 1.9915,
+      "step": 2220
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.8125,
+      "learning_rate": 0.00012268421052631577,
+      "loss": 1.9927,
+      "step": 2230
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.71875,
+      "learning_rate": 0.00012252631578947367,
+      "loss": 1.5576,
+      "step": 2240
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.09375,
+      "learning_rate": 0.00012236842105263157,
+      "loss": 1.7842,
+      "step": 2250
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.00012221052631578946,
+      "loss": 1.6442,
+      "step": 2260
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.625,
+      "learning_rate": 0.00012205263157894736,
+      "loss": 2.0708,
+      "step": 2270
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.3125,
+      "learning_rate": 0.00012189473684210524,
+      "loss": 1.7107,
+      "step": 2280
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.0,
+      "learning_rate": 0.00012173684210526315,
+      "loss": 1.6121,
+      "step": 2290
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.5625,
+      "learning_rate": 0.00012157894736842104,
+      "loss": 1.7322,
+      "step": 2300
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.9375,
+      "learning_rate": 0.00012142105263157895,
+      "loss": 1.9742,
+      "step": 2310
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.78125,
+      "learning_rate": 0.00012126315789473683,
+      "loss": 2.2318,
+      "step": 2320
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.1875,
+      "learning_rate": 0.00012110526315789473,
+      "loss": 2.1988,
+      "step": 2330
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.53125,
+      "learning_rate": 0.00012094736842105262,
+      "loss": 1.7186,
+      "step": 2340
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.375,
+      "learning_rate": 0.0001207894736842105,
+      "loss": 1.7444,
+      "step": 2350
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.875,
+      "learning_rate": 0.00012063157894736842,
+      "loss": 1.6583,
+      "step": 2360
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.0625,
+      "learning_rate": 0.0001204736842105263,
+      "loss": 1.6891,
+      "step": 2370
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.875,
+      "learning_rate": 0.00012031578947368421,
+      "loss": 1.5208,
+      "step": 2380
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.78125,
+      "learning_rate": 0.00012015789473684209,
+      "loss": 1.9399,
+      "step": 2390
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.0,
+      "learning_rate": 0.00011999999999999999,
+      "loss": 1.9137,
+      "step": 2400
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.71875,
+      "learning_rate": 0.00011984210526315789,
+      "loss": 1.8328,
+      "step": 2410
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.9375,
+      "learning_rate": 0.00011968421052631577,
+      "loss": 1.7356,
+      "step": 2420
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.875,
+      "learning_rate": 0.00011952631578947368,
+      "loss": 2.0099,
+      "step": 2430
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.8125,
+      "learning_rate": 0.00011936842105263156,
+      "loss": 1.8541,
+      "step": 2440
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.65625,
+      "learning_rate": 0.00011921052631578947,
+      "loss": 1.6947,
+      "step": 2450
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.0,
+      "learning_rate": 0.00011905263157894736,
+      "loss": 1.7433,
+      "step": 2460
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.00011889473684210527,
+      "loss": 1.6962,
+      "step": 2470
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.34375,
+      "learning_rate": 0.00011873684210526315,
+      "loss": 2.0064,
+      "step": 2480
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.00011857894736842103,
+      "loss": 1.8146,
+      "step": 2490
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.25,
+      "learning_rate": 0.00011842105263157894,
+      "loss": 1.6956,
+      "step": 2500
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.5,
+      "learning_rate": 0.00011826315789473683,
+      "loss": 1.7244,
+      "step": 2510
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.53125,
+      "learning_rate": 0.00011810526315789474,
+      "loss": 1.7706,
+      "step": 2520
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.375,
+      "learning_rate": 0.00011794736842105262,
+      "loss": 1.7227,
+      "step": 2530
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.640625,
+      "learning_rate": 0.00011778947368421053,
+      "loss": 1.8638,
+      "step": 2540
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.96875,
+      "learning_rate": 0.00011763157894736841,
+      "loss": 1.8712,
+      "step": 2550
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.9375,
+      "learning_rate": 0.0001174736842105263,
+      "loss": 1.8399,
+      "step": 2560
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.9375,
+      "learning_rate": 0.0001173157894736842,
+      "loss": 1.7568,
+      "step": 2570
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.1875,
+      "learning_rate": 0.00011715789473684209,
+      "loss": 1.4781,
+      "step": 2580
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.5,
+      "learning_rate": 0.000117,
+      "loss": 1.6916,
+      "step": 2590
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.40625,
+      "learning_rate": 0.00011684210526315788,
+      "loss": 1.8282,
+      "step": 2600
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 7.46875,
+      "learning_rate": 0.00011668421052631579,
+      "loss": 1.7929,
+      "step": 2610
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.75,
+      "learning_rate": 0.00011652631578947367,
+      "loss": 1.6475,
+      "step": 2620
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.765625,
+      "learning_rate": 0.00011636842105263156,
+      "loss": 1.6647,
+      "step": 2630
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.00011621052631578947,
+      "loss": 1.5357,
+      "step": 2640
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.125,
+      "learning_rate": 0.00011605263157894735,
+      "loss": 1.8162,
+      "step": 2650
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.125,
+      "learning_rate": 0.00011589473684210526,
+      "loss": 1.7852,
+      "step": 2660
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.4375,
+      "learning_rate": 0.00011573684210526314,
+      "loss": 1.4552,
+      "step": 2670
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.00011557894736842105,
+      "loss": 1.7193,
+      "step": 2680
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.9375,
+      "learning_rate": 0.00011542105263157894,
+      "loss": 1.6664,
+      "step": 2690
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.0,
+      "learning_rate": 0.00011526315789473682,
+      "loss": 1.7187,
+      "step": 2700
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.1875,
+      "learning_rate": 0.00011510526315789473,
+      "loss": 1.7427,
+      "step": 2710
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.90625,
+      "learning_rate": 0.00011494736842105261,
+      "loss": 1.9004,
+      "step": 2720
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.15625,
+      "learning_rate": 0.00011478947368421052,
+      "loss": 2.0091,
+      "step": 2730
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.28125,
+      "learning_rate": 0.0001146315789473684,
+      "loss": 1.7475,
+      "step": 2740
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.3125,
+      "learning_rate": 0.00011447368421052632,
+      "loss": 1.4338,
+      "step": 2750
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.375,
+      "learning_rate": 0.0001143157894736842,
+      "loss": 1.7836,
+      "step": 2760
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.828125,
+      "learning_rate": 0.00011415789473684208,
+      "loss": 1.6579,
+      "step": 2770
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.03125,
+      "learning_rate": 0.00011399999999999999,
+      "loss": 1.9117,
+      "step": 2780
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.5,
+      "learning_rate": 0.00011384210526315788,
+      "loss": 1.9963,
+      "step": 2790
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.15625,
+      "learning_rate": 0.00011368421052631579,
+      "loss": 1.2637,
+      "step": 2800
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.984375,
+      "learning_rate": 0.00011352631578947367,
+      "loss": 1.7052,
+      "step": 2810
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.65625,
+      "learning_rate": 0.00011336842105263158,
+      "loss": 1.3585,
+      "step": 2820
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.6875,
+      "learning_rate": 0.00011321052631578946,
+      "loss": 2.0309,
+      "step": 2830
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.3125,
+      "learning_rate": 0.00011305263157894735,
+      "loss": 1.527,
+      "step": 2840
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.00011289473684210526,
+      "loss": 1.7948,
+      "step": 2850
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.78125,
+      "learning_rate": 0.00011273684210526314,
+      "loss": 1.8825,
+      "step": 2860
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.15625,
+      "learning_rate": 0.00011257894736842105,
+      "loss": 1.7341,
+      "step": 2870
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.03125,
+      "learning_rate": 0.00011242105263157893,
+      "loss": 1.9529,
+      "step": 2880
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 8.5625,
+      "learning_rate": 0.00011226315789473684,
+      "loss": 1.6345,
+      "step": 2890
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.5,
+      "learning_rate": 0.00011210526315789472,
+      "loss": 1.6036,
+      "step": 2900
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.125,
+      "learning_rate": 0.00011194736842105264,
+      "loss": 1.7851,
+      "step": 2910
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.625,
+      "learning_rate": 0.00011178947368421052,
+      "loss": 1.9272,
+      "step": 2920
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.0001116315789473684,
+      "loss": 1.9839,
+      "step": 2930
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.00011147368421052631,
+      "loss": 1.5799,
+      "step": 2940
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.34375,
+      "learning_rate": 0.0001113157894736842,
+      "loss": 1.599,
+      "step": 2950
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 7.1875,
+      "learning_rate": 0.0001111578947368421,
+      "loss": 1.8219,
+      "step": 2960
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.84375,
+      "learning_rate": 0.00011099999999999999,
+      "loss": 1.5486,
+      "step": 2970
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.84375,
+      "learning_rate": 0.0001108421052631579,
+      "loss": 1.419,
+      "step": 2980
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.03125,
+      "learning_rate": 0.00011068421052631578,
+      "loss": 1.7484,
+      "step": 2990
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.09375,
+      "learning_rate": 0.00011052631578947366,
+      "loss": 1.7116,
+      "step": 3000
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.3125,
+      "learning_rate": 0.00011036842105263157,
+      "loss": 1.3924,
+      "step": 3010
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.125,
+      "learning_rate": 0.00011021052631578946,
+      "loss": 1.4203,
+      "step": 3020
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.25,
+      "learning_rate": 0.00011005263157894737,
+      "loss": 2.0427,
+      "step": 3030
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.09375,
+      "learning_rate": 0.00010989473684210525,
+      "loss": 1.7216,
+      "step": 3040
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.00010973684210526316,
+      "loss": 1.9249,
+      "step": 3050
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.15625,
+      "learning_rate": 0.00010957894736842104,
+      "loss": 1.9464,
+      "step": 3060
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.859375,
+      "learning_rate": 0.00010942105263157893,
+      "loss": 1.9662,
+      "step": 3070
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.5,
+      "learning_rate": 0.00010926315789473684,
+      "loss": 1.5929,
+      "step": 3080
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.28125,
+      "learning_rate": 0.00010910526315789472,
+      "loss": 1.8971,
+      "step": 3090
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.0625,
+      "learning_rate": 0.00010894736842105263,
+      "loss": 1.834,
+      "step": 3100
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.46875,
+      "learning_rate": 0.00010878947368421051,
+      "loss": 1.8009,
+      "step": 3110
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.09375,
+      "learning_rate": 0.00010863157894736842,
+      "loss": 1.7885,
+      "step": 3120
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.5,
+      "learning_rate": 0.0001084736842105263,
+      "loss": 1.3803,
+      "step": 3130
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.6875,
+      "learning_rate": 0.00010831578947368419,
+      "loss": 1.6063,
+      "step": 3140
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.34375,
+      "learning_rate": 0.0001081578947368421,
+      "loss": 1.6058,
+      "step": 3150
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.00010799999999999998,
+      "loss": 1.7769,
+      "step": 3160
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.78125,
+      "learning_rate": 0.00010784210526315789,
+      "loss": 1.8996,
+      "step": 3170
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.375,
+      "learning_rate": 0.00010768421052631578,
+      "loss": 1.5069,
+      "step": 3180
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.96875,
+      "learning_rate": 0.00010752631578947369,
+      "loss": 1.4708,
+      "step": 3190
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.59375,
+      "learning_rate": 0.00010736842105263157,
+      "loss": 1.7289,
+      "step": 3200
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.75,
+      "learning_rate": 0.00010721052631578945,
+      "loss": 1.509,
+      "step": 3210
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.375,
+      "learning_rate": 0.00010705263157894736,
+      "loss": 1.5922,
+      "step": 3220
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.00010689473684210525,
+      "loss": 1.8839,
+      "step": 3230
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.00010673684210526316,
+      "loss": 1.6906,
+      "step": 3240
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.921875,
+      "learning_rate": 0.00010657894736842104,
+      "loss": 1.6712,
+      "step": 3250
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.00010642105263157895,
+      "loss": 1.8901,
+      "step": 3260
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.1875,
+      "learning_rate": 0.00010626315789473683,
+      "loss": 1.4667,
+      "step": 3270
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.40625,
+      "learning_rate": 0.00010610526315789471,
+      "loss": 1.8619,
+      "step": 3280
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.40625,
+      "learning_rate": 0.00010594736842105262,
+      "loss": 1.6954,
+      "step": 3290
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 3.703125,
+      "learning_rate": 0.00010578947368421051,
+      "loss": 1.5175,
+      "step": 3300
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.96875,
+      "learning_rate": 0.00010563157894736842,
+      "loss": 1.7939,
+      "step": 3310
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.8125,
+      "learning_rate": 0.0001054736842105263,
+      "loss": 1.6237,
+      "step": 3320
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.15625,
+      "learning_rate": 0.00010531578947368421,
+      "loss": 1.6149,
+      "step": 3330
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.5625,
+      "learning_rate": 0.0001051578947368421,
+      "loss": 1.7133,
+      "step": 3340
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.5,
+      "learning_rate": 0.00010499999999999999,
+      "loss": 1.3957,
+      "step": 3350
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.28125,
+      "learning_rate": 0.00010484210526315789,
+      "loss": 1.9061,
+      "step": 3360
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.25,
+      "learning_rate": 0.00010468421052631577,
+      "loss": 1.6927,
+      "step": 3370
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.9375,
+      "learning_rate": 0.00010452631578947368,
+      "loss": 1.6389,
+      "step": 3380
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.00010436842105263156,
+      "loss": 1.6523,
+      "step": 3390
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.40625,
+      "learning_rate": 0.00010421052631578947,
+      "loss": 1.5598,
+      "step": 3400
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.59375,
+      "learning_rate": 0.00010405263157894736,
+      "loss": 1.2804,
+      "step": 3410
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.21875,
+      "learning_rate": 0.00010389473684210525,
+      "loss": 1.2989,
+      "step": 3420
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.65625,
+      "learning_rate": 0.00010373684210526315,
+      "loss": 1.4779,
+      "step": 3430
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.78125,
+      "learning_rate": 0.00010357894736842103,
+      "loss": 1.7402,
+      "step": 3440
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.09375,
+      "learning_rate": 0.00010342105263157894,
+      "loss": 1.6356,
+      "step": 3450
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.625,
+      "learning_rate": 0.00010326315789473683,
+      "loss": 1.9104,
+      "step": 3460
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 8.75,
+      "learning_rate": 0.00010310526315789474,
+      "loss": 1.5426,
+      "step": 3470
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.375,
+      "learning_rate": 0.00010294736842105262,
+      "loss": 1.5519,
+      "step": 3480
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.125,
+      "learning_rate": 0.00010278947368421052,
+      "loss": 1.3926,
+      "step": 3490
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.71875,
+      "learning_rate": 0.00010263157894736841,
+      "loss": 2.0501,
+      "step": 3500
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.21875,
+      "learning_rate": 0.0001024736842105263,
+      "loss": 1.8807,
+      "step": 3510
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.65625,
+      "learning_rate": 0.0001023157894736842,
+      "loss": 1.8071,
+      "step": 3520
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.0,
+      "learning_rate": 0.00010215789473684209,
+      "loss": 2.0167,
+      "step": 3530
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.5,
+      "learning_rate": 0.000102,
+      "loss": 1.8896,
+      "step": 3540
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.15625,
+      "learning_rate": 0.00010184210526315788,
+      "loss": 1.8863,
+      "step": 3550
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.53125,
+      "learning_rate": 0.00010168421052631578,
+      "loss": 1.7345,
+      "step": 3560
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.0,
+      "learning_rate": 0.00010152631578947368,
+      "loss": 1.4917,
+      "step": 3570
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.8125,
+      "learning_rate": 0.00010136842105263157,
+      "loss": 1.6955,
+      "step": 3580
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.4375,
+      "learning_rate": 0.00010121052631578947,
+      "loss": 2.1419,
+      "step": 3590
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.00010105263157894735,
+      "loss": 1.7818,
+      "step": 3600
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.40625,
+      "learning_rate": 0.00010089473684210526,
+      "loss": 1.8861,
+      "step": 3610
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.625,
+      "learning_rate": 0.00010073684210526314,
+      "loss": 1.8296,
+      "step": 3620
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.5,
+      "learning_rate": 0.00010057894736842104,
+      "loss": 1.4939,
+      "step": 3630
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.421875,
+      "learning_rate": 0.00010042105263157894,
+      "loss": 1.6977,
+      "step": 3640
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.90625,
+      "learning_rate": 0.00010026315789473683,
+      "loss": 1.6357,
+      "step": 3650
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.75,
+      "learning_rate": 0.00010010526315789473,
+      "loss": 1.4076,
+      "step": 3660
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.4375,
+      "learning_rate": 9.994736842105261e-05,
+      "loss": 1.9151,
+      "step": 3670
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.25,
+      "learning_rate": 9.978947368421052e-05,
+      "loss": 1.5325,
+      "step": 3680
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.453125,
+      "learning_rate": 9.963157894736841e-05,
+      "loss": 1.6342,
+      "step": 3690
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.9375,
+      "learning_rate": 9.94736842105263e-05,
+      "loss": 1.3291,
+      "step": 3700
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.3125,
+      "learning_rate": 9.93157894736842e-05,
+      "loss": 1.7364,
+      "step": 3710
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.21875,
+      "learning_rate": 9.91578947368421e-05,
+      "loss": 1.4092,
+      "step": 3720
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.21875,
+      "learning_rate": 9.9e-05,
+      "loss": 1.515,
+      "step": 3730
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.625,
+      "learning_rate": 9.884210526315788e-05,
+      "loss": 1.4621,
+      "step": 3740
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.09375,
+      "learning_rate": 9.868421052631579e-05,
+      "loss": 1.7776,
+      "step": 3750
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.46875,
+      "learning_rate": 9.852631578947367e-05,
+      "loss": 1.8841,
+      "step": 3760
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.53125,
+      "learning_rate": 9.836842105263157e-05,
+      "loss": 2.119,
+      "step": 3770
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.78125,
+      "learning_rate": 9.821052631578946e-05,
+      "loss": 1.4981,
+      "step": 3780
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.46875,
+      "learning_rate": 9.805263157894736e-05,
+      "loss": 1.3002,
+      "step": 3790
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.9375,
+      "learning_rate": 9.789473684210526e-05,
+      "loss": 1.5721,
+      "step": 3800
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.4375,
+      "learning_rate": 9.773684210526315e-05,
+      "loss": 1.6575,
+      "step": 3810
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.46875,
+      "learning_rate": 9.757894736842105e-05,
+      "loss": 1.9991,
+      "step": 3820
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.65625,
+      "learning_rate": 9.742105263157893e-05,
+      "loss": 1.4024,
+      "step": 3830
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.0625,
+      "learning_rate": 9.726315789473683e-05,
+      "loss": 1.7038,
+      "step": 3840
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.9375,
+      "learning_rate": 9.710526315789473e-05,
+      "loss": 1.6067,
+      "step": 3850
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.875,
+      "learning_rate": 9.694736842105262e-05,
+      "loss": 1.6456,
+      "step": 3860
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.34375,
+      "learning_rate": 9.678947368421052e-05,
+      "loss": 1.8511,
+      "step": 3870
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.4375,
+      "learning_rate": 9.663157894736842e-05,
+      "loss": 1.7312,
+      "step": 3880
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.71875,
+      "learning_rate": 9.647368421052631e-05,
+      "loss": 1.6937,
+      "step": 3890
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.125,
+      "learning_rate": 9.63157894736842e-05,
+      "loss": 1.2875,
+      "step": 3900
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.3125,
+      "learning_rate": 9.615789473684209e-05,
+      "loss": 1.6645,
+      "step": 3910
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.875,
+      "learning_rate": 9.599999999999999e-05,
+      "loss": 1.8489,
+      "step": 3920
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.40625,
+      "learning_rate": 9.584210526315789e-05,
+      "loss": 1.6001,
+      "step": 3930
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.46875,
+      "learning_rate": 9.568421052631578e-05,
+      "loss": 1.546,
+      "step": 3940
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.53125,
+      "learning_rate": 9.552631578947368e-05,
+      "loss": 1.2363,
+      "step": 3950
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.0,
+      "learning_rate": 9.536842105263158e-05,
+      "loss": 1.6655,
+      "step": 3960
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.0,
+      "learning_rate": 9.521052631578946e-05,
+      "loss": 1.5445,
+      "step": 3970
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.625,
+      "learning_rate": 9.505263157894735e-05,
+      "loss": 1.4026,
+      "step": 3980
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.1875,
+      "learning_rate": 9.489473684210525e-05,
+      "loss": 1.6248,
+      "step": 3990
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.1875,
+      "learning_rate": 9.473684210526315e-05,
+      "loss": 1.9961,
+      "step": 4000
+    },
+    {
+      "epoch": 0.2,
+      "eval_loss": 1.6376639604568481,
+      "eval_runtime": 30.5452,
+      "eval_samples_per_second": 32.738,
+      "eval_steps_per_second": 32.738,
+      "step": 4000
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.15625,
+      "learning_rate": 9.457894736842104e-05,
+      "loss": 1.4331,
+      "step": 4010
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.921875,
+      "learning_rate": 9.442105263157894e-05,
+      "loss": 1.5178,
+      "step": 4020
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.953125,
+      "learning_rate": 9.426315789473684e-05,
+      "loss": 1.7301,
+      "step": 4030
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.75,
+      "learning_rate": 9.410526315789473e-05,
+      "loss": 1.5035,
+      "step": 4040
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.6875,
+      "learning_rate": 9.394736842105262e-05,
+      "loss": 1.5749,
+      "step": 4050
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.34375,
+      "learning_rate": 9.378947368421051e-05,
+      "loss": 1.7267,
+      "step": 4060
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.25,
+      "learning_rate": 9.363157894736841e-05,
+      "loss": 1.872,
+      "step": 4070
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.0625,
+      "learning_rate": 9.347368421052631e-05,
+      "loss": 1.991,
+      "step": 4080
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.75,
+      "learning_rate": 9.33157894736842e-05,
+      "loss": 1.4165,
+      "step": 4090
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.84375,
+      "learning_rate": 9.31578947368421e-05,
+      "loss": 1.5004,
+      "step": 4100
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.09375,
+      "learning_rate": 9.3e-05,
+      "loss": 1.4191,
+      "step": 4110
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 7.0,
+      "learning_rate": 9.284210526315788e-05,
+      "loss": 1.7576,
+      "step": 4120
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.96875,
+      "learning_rate": 9.268421052631578e-05,
+      "loss": 1.5489,
+      "step": 4130
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.6875,
+      "learning_rate": 9.252631578947367e-05,
+      "loss": 1.8809,
+      "step": 4140
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.59375,
+      "learning_rate": 9.236842105263157e-05,
+      "loss": 1.8825,
+      "step": 4150
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.0,
+      "learning_rate": 9.221052631578947e-05,
+      "loss": 1.5439,
+      "step": 4160
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.78125,
+      "learning_rate": 9.205263157894736e-05,
+      "loss": 1.5064,
+      "step": 4170
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.34375,
+      "learning_rate": 9.189473684210526e-05,
+      "loss": 1.756,
+      "step": 4180
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.0,
+      "learning_rate": 9.173684210526314e-05,
+      "loss": 1.683,
+      "step": 4190
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.125,
+      "learning_rate": 9.157894736842104e-05,
+      "loss": 1.6729,
+      "step": 4200
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 7.71875,
+      "learning_rate": 9.142105263157894e-05,
+      "loss": 1.6747,
+      "step": 4210
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.15625,
+      "learning_rate": 9.126315789473683e-05,
+      "loss": 1.4337,
+      "step": 4220
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.625,
+      "learning_rate": 9.110526315789473e-05,
+      "loss": 1.6269,
+      "step": 4230
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.65625,
+      "learning_rate": 9.094736842105263e-05,
+      "loss": 1.7594,
+      "step": 4240
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.75,
+      "learning_rate": 9.078947368421052e-05,
+      "loss": 1.6758,
+      "step": 4250
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.875,
+      "learning_rate": 9.06315789473684e-05,
+      "loss": 1.6468,
+      "step": 4260
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.625,
+      "learning_rate": 9.047368421052632e-05,
+      "loss": 1.9438,
+      "step": 4270
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.890625,
+      "learning_rate": 9.03157894736842e-05,
+      "loss": 1.3541,
+      "step": 4280
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.28125,
+      "learning_rate": 9.01578947368421e-05,
+      "loss": 1.5609,
+      "step": 4290
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 7.15625,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 1.846,
+      "step": 4300
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 6.6875,
+      "learning_rate": 8.984210526315789e-05,
+      "loss": 1.6363,
+      "step": 4310
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.875,
+      "learning_rate": 8.968421052631579e-05,
+      "loss": 1.7652,
+      "step": 4320
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.4375,
+      "learning_rate": 8.952631578947367e-05,
+      "loss": 1.3733,
+      "step": 4330
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.0,
+      "learning_rate": 8.936842105263158e-05,
+      "loss": 1.4821,
+      "step": 4340
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.3125,
+      "learning_rate": 8.921052631578946e-05,
+      "loss": 1.6647,
+      "step": 4350
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.09375,
+      "learning_rate": 8.905263157894736e-05,
+      "loss": 1.5257,
+      "step": 4360
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.75,
+      "learning_rate": 8.889473684210525e-05,
+      "loss": 1.4503,
+      "step": 4370
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.03125,
+      "learning_rate": 8.873684210526315e-05,
+      "loss": 1.7228,
+      "step": 4380
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.4375,
+      "learning_rate": 8.857894736842105e-05,
+      "loss": 1.4673,
+      "step": 4390
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.96875,
+      "learning_rate": 8.842105263157893e-05,
+      "loss": 1.7307,
+      "step": 4400
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.25,
+      "learning_rate": 8.826315789473684e-05,
+      "loss": 1.6506,
+      "step": 4410
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.125,
+      "learning_rate": 8.810526315789472e-05,
+      "loss": 1.8352,
+      "step": 4420
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.9375,
+      "learning_rate": 8.794736842105262e-05,
+      "loss": 1.8344,
+      "step": 4430
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 3.84375,
+      "learning_rate": 8.778947368421052e-05,
+      "loss": 1.5421,
+      "step": 4440
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.875,
+      "learning_rate": 8.763157894736841e-05,
+      "loss": 1.7763,
+      "step": 4450
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.84375,
+      "learning_rate": 8.747368421052631e-05,
+      "loss": 1.3265,
+      "step": 4460
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.5,
+      "learning_rate": 8.731578947368421e-05,
+      "loss": 1.4655,
+      "step": 4470
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.15625,
+      "learning_rate": 8.71578947368421e-05,
+      "loss": 1.4654,
+      "step": 4480
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 10.75,
+      "learning_rate": 8.699999999999999e-05,
+      "loss": 1.4957,
+      "step": 4490
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.09375,
+      "learning_rate": 8.68421052631579e-05,
+      "loss": 1.4691,
+      "step": 4500
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.4375,
+      "learning_rate": 8.668421052631578e-05,
+      "loss": 1.4734,
+      "step": 4510
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.09375,
+      "learning_rate": 8.652631578947368e-05,
+      "loss": 1.5839,
+      "step": 4520
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.25,
+      "learning_rate": 8.636842105263157e-05,
+      "loss": 1.4329,
+      "step": 4530
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.609375,
+      "learning_rate": 8.621052631578947e-05,
+      "loss": 1.6005,
+      "step": 4540
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.53125,
+      "learning_rate": 8.605263157894737e-05,
+      "loss": 1.6974,
+      "step": 4550
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.875,
+      "learning_rate": 8.589473684210525e-05,
+      "loss": 1.633,
+      "step": 4560
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.28125,
+      "learning_rate": 8.573684210526316e-05,
+      "loss": 1.6251,
+      "step": 4570
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.1875,
+      "learning_rate": 8.557894736842104e-05,
+      "loss": 1.3481,
+      "step": 4580
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.96875,
+      "learning_rate": 8.542105263157894e-05,
+      "loss": 1.0885,
+      "step": 4590
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.9375,
+      "learning_rate": 8.526315789473684e-05,
+      "loss": 1.6525,
+      "step": 4600
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.78125,
+      "learning_rate": 8.510526315789473e-05,
+      "loss": 1.7211,
+      "step": 4610
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.4375,
+      "learning_rate": 8.494736842105263e-05,
+      "loss": 2.0992,
+      "step": 4620
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.4375,
+      "learning_rate": 8.478947368421051e-05,
+      "loss": 1.8482,
+      "step": 4630
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.84375,
+      "learning_rate": 8.463157894736842e-05,
+      "loss": 1.2918,
+      "step": 4640
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.9375,
+      "learning_rate": 8.44736842105263e-05,
+      "loss": 1.7271,
+      "step": 4650
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.53125,
+      "learning_rate": 8.43157894736842e-05,
+      "loss": 1.6719,
+      "step": 4660
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.28125,
+      "learning_rate": 8.41578947368421e-05,
+      "loss": 1.305,
+      "step": 4670
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.4375,
+      "learning_rate": 8.4e-05,
+      "loss": 1.7411,
+      "step": 4680
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.0,
+      "learning_rate": 8.384210526315789e-05,
+      "loss": 1.5826,
+      "step": 4690
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.96875,
+      "learning_rate": 8.368421052631578e-05,
+      "loss": 1.687,
+      "step": 4700
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.59375,
+      "learning_rate": 8.352631578947369e-05,
+      "loss": 1.4675,
+      "step": 4710
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.28125,
+      "learning_rate": 8.336842105263157e-05,
+      "loss": 1.6521,
+      "step": 4720
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.21875,
+      "learning_rate": 8.321052631578948e-05,
+      "loss": 1.7566,
+      "step": 4730
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.9375,
+      "learning_rate": 8.305263157894736e-05,
+      "loss": 1.6841,
+      "step": 4740
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.578125,
+      "learning_rate": 8.289473684210526e-05,
+      "loss": 1.6672,
+      "step": 4750
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.53125,
+      "learning_rate": 8.273684210526315e-05,
+      "loss": 1.5411,
+      "step": 4760
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 7.4375,
+      "learning_rate": 8.257894736842104e-05,
+      "loss": 1.721,
+      "step": 4770
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.953125,
+      "learning_rate": 8.242105263157895e-05,
+      "loss": 1.7061,
+      "step": 4780
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.953125,
+      "learning_rate": 8.226315789473683e-05,
+      "loss": 1.5112,
+      "step": 4790
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.75,
+      "learning_rate": 8.210526315789474e-05,
+      "loss": 1.3537,
+      "step": 4800
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.125,
+      "learning_rate": 8.194736842105262e-05,
+      "loss": 1.3723,
+      "step": 4810
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.8125,
+      "learning_rate": 8.178947368421052e-05,
+      "loss": 1.9272,
+      "step": 4820
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 13.3125,
+      "learning_rate": 8.163157894736842e-05,
+      "loss": 1.3583,
+      "step": 4830
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 10.25,
+      "learning_rate": 8.14736842105263e-05,
+      "loss": 1.7233,
+      "step": 4840
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.3125,
+      "learning_rate": 8.131578947368421e-05,
+      "loss": 1.6048,
+      "step": 4850
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.4375,
+      "learning_rate": 8.11578947368421e-05,
+      "loss": 1.7952,
+      "step": 4860
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.6875,
+      "learning_rate": 8.1e-05,
+      "loss": 1.6838,
+      "step": 4870
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.546875,
+      "learning_rate": 8.084210526315789e-05,
+      "loss": 1.9311,
+      "step": 4880
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.28125,
+      "learning_rate": 8.068421052631578e-05,
+      "loss": 1.5036,
+      "step": 4890
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.90625,
+      "learning_rate": 8.052631578947368e-05,
+      "loss": 1.55,
+      "step": 4900
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.21875,
+      "learning_rate": 8.036842105263156e-05,
+      "loss": 1.5131,
+      "step": 4910
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.21875,
+      "learning_rate": 8.021052631578947e-05,
+      "loss": 1.5671,
+      "step": 4920
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.71875,
+      "learning_rate": 8.005263157894736e-05,
+      "loss": 1.6518,
+      "step": 4930
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.46875,
+      "learning_rate": 7.989473684210527e-05,
+      "loss": 1.7936,
+      "step": 4940
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 9.4375,
+      "learning_rate": 7.973684210526315e-05,
+      "loss": 1.5896,
+      "step": 4950
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.40625,
+      "learning_rate": 7.957894736842106e-05,
+      "loss": 1.6038,
+      "step": 4960
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.28125,
+      "learning_rate": 7.942105263157894e-05,
+      "loss": 1.4422,
+      "step": 4970
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.625,
+      "learning_rate": 7.926315789473683e-05,
+      "loss": 1.4781,
+      "step": 4980
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.90625,
+      "learning_rate": 7.910526315789474e-05,
+      "loss": 1.554,
+      "step": 4990
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.25,
+      "learning_rate": 7.894736842105262e-05,
+      "loss": 1.4183,
+      "step": 5000
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.3125,
+      "learning_rate": 7.878947368421053e-05,
+      "loss": 1.5773,
+      "step": 5010
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 8.9375,
+      "learning_rate": 7.863157894736841e-05,
+      "loss": 1.7739,
+      "step": 5020
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.6875,
+      "learning_rate": 7.847368421052632e-05,
+      "loss": 1.6497,
+      "step": 5030
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.296875,
+      "learning_rate": 7.83157894736842e-05,
+      "loss": 1.6733,
+      "step": 5040
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.09375,
+      "learning_rate": 7.815789473684209e-05,
+      "loss": 1.509,
+      "step": 5050
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.9375,
+      "learning_rate": 7.8e-05,
+      "loss": 1.7451,
+      "step": 5060
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.984375,
+      "learning_rate": 7.784210526315788e-05,
+      "loss": 1.1635,
+      "step": 5070
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.71875,
+      "learning_rate": 7.768421052631579e-05,
+      "loss": 1.3425,
+      "step": 5080
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.71875,
+      "learning_rate": 7.752631578947367e-05,
+      "loss": 1.6612,
+      "step": 5090
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.4375,
+      "learning_rate": 7.736842105263159e-05,
+      "loss": 1.5136,
+      "step": 5100
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.375,
+      "learning_rate": 7.721052631578947e-05,
+      "loss": 1.4066,
+      "step": 5110
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.15625,
+      "learning_rate": 7.705263157894735e-05,
+      "loss": 1.5721,
+      "step": 5120
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.75,
+      "learning_rate": 7.689473684210526e-05,
+      "loss": 1.7248,
+      "step": 5130
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.40625,
+      "learning_rate": 7.673684210526314e-05,
+      "loss": 1.5728,
+      "step": 5140
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.40625,
+      "learning_rate": 7.657894736842105e-05,
+      "loss": 1.6477,
+      "step": 5150
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.1875,
+      "learning_rate": 7.642105263157894e-05,
+      "loss": 1.7276,
+      "step": 5160
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.1875,
+      "learning_rate": 7.626315789473685e-05,
+      "loss": 1.5607,
+      "step": 5170
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.8125,
+      "learning_rate": 7.610526315789473e-05,
+      "loss": 1.7054,
+      "step": 5180
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.5625,
+      "learning_rate": 7.594736842105261e-05,
+      "loss": 1.5041,
+      "step": 5190
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.28125,
+      "learning_rate": 7.578947368421052e-05,
+      "loss": 1.3939,
+      "step": 5200
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.28125,
+      "learning_rate": 7.563157894736841e-05,
+      "loss": 1.3901,
+      "step": 5210
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.84375,
+      "learning_rate": 7.547368421052632e-05,
+      "loss": 1.3635,
+      "step": 5220
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.5,
+      "learning_rate": 7.53157894736842e-05,
+      "loss": 1.6156,
+      "step": 5230
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.5625,
+      "learning_rate": 7.515789473684211e-05,
+      "loss": 1.4224,
+      "step": 5240
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.6875,
+      "learning_rate": 7.5e-05,
+      "loss": 1.5507,
+      "step": 5250
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.5,
+      "learning_rate": 7.484210526315789e-05,
+      "loss": 1.5214,
+      "step": 5260
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.21875,
+      "learning_rate": 7.468421052631579e-05,
+      "loss": 1.4005,
+      "step": 5270
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 9.8125,
+      "learning_rate": 7.452631578947368e-05,
+      "loss": 1.649,
+      "step": 5280
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.25,
+      "learning_rate": 7.436842105263157e-05,
+      "loss": 1.3222,
+      "step": 5290
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.875,
+      "learning_rate": 7.421052631578946e-05,
+      "loss": 1.4316,
+      "step": 5300
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.59375,
+      "learning_rate": 7.405263157894736e-05,
+      "loss": 1.4701,
+      "step": 5310
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.96875,
+      "learning_rate": 7.389473684210526e-05,
+      "loss": 1.8238,
+      "step": 5320
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.375,
+      "learning_rate": 7.373684210526315e-05,
+      "loss": 1.4409,
+      "step": 5330
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.5,
+      "learning_rate": 7.357894736842105e-05,
+      "loss": 1.4107,
+      "step": 5340
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.0,
+      "learning_rate": 7.342105263157895e-05,
+      "loss": 1.5908,
+      "step": 5350
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.46875,
+      "learning_rate": 7.326315789473683e-05,
+      "loss": 1.1515,
+      "step": 5360
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.625,
+      "learning_rate": 7.310526315789473e-05,
+      "loss": 1.8387,
+      "step": 5370
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 7.34375,
+      "learning_rate": 7.294736842105262e-05,
+      "loss": 1.7581,
+      "step": 5380
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.40625,
+      "learning_rate": 7.278947368421052e-05,
+      "loss": 1.8786,
+      "step": 5390
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.09375,
+      "learning_rate": 7.263157894736842e-05,
+      "loss": 1.5486,
+      "step": 5400
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.375,
+      "learning_rate": 7.247368421052631e-05,
+      "loss": 1.6973,
+      "step": 5410
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.53125,
+      "learning_rate": 7.231578947368421e-05,
+      "loss": 1.1965,
+      "step": 5420
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.375,
+      "learning_rate": 7.215789473684209e-05,
+      "loss": 1.4671,
+      "step": 5430
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.03125,
+      "learning_rate": 7.199999999999999e-05,
+      "loss": 1.4568,
+      "step": 5440
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.0625,
+      "learning_rate": 7.184210526315788e-05,
+      "loss": 1.4513,
+      "step": 5450
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.4375,
+      "learning_rate": 7.168421052631578e-05,
+      "loss": 1.2652,
+      "step": 5460
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.125,
+      "learning_rate": 7.152631578947368e-05,
+      "loss": 1.3024,
+      "step": 5470
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 4.3125,
+      "learning_rate": 7.136842105263157e-05,
+      "loss": 1.3969,
+      "step": 5480
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.96875,
+      "learning_rate": 7.121052631578947e-05,
+      "loss": 1.5425,
+      "step": 5490
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.78125,
+      "learning_rate": 7.105263157894735e-05,
+      "loss": 1.5634,
+      "step": 5500
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.03125,
+      "learning_rate": 7.089473684210525e-05,
+      "loss": 1.3567,
+      "step": 5510
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.25,
+      "learning_rate": 7.073684210526315e-05,
+      "loss": 1.5746,
+      "step": 5520
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.875,
+      "learning_rate": 7.057894736842104e-05,
+      "loss": 1.6727,
+      "step": 5530
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.15625,
+      "learning_rate": 7.042105263157894e-05,
+      "loss": 1.709,
+      "step": 5540
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.4375,
+      "learning_rate": 7.026315789473684e-05,
+      "loss": 1.6377,
+      "step": 5550
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.75,
+      "learning_rate": 7.010526315789473e-05,
+      "loss": 1.368,
+      "step": 5560
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 11.0,
+      "learning_rate": 6.994736842105262e-05,
+      "loss": 1.1897,
+      "step": 5570
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.3125,
+      "learning_rate": 6.978947368421051e-05,
+      "loss": 1.4087,
+      "step": 5580
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.53125,
+      "learning_rate": 6.963157894736841e-05,
+      "loss": 1.4641,
+      "step": 5590
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.78125,
+      "learning_rate": 6.947368421052631e-05,
+      "loss": 1.2938,
+      "step": 5600
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.390625,
+      "learning_rate": 6.93157894736842e-05,
+      "loss": 1.2211,
+      "step": 5610
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.40625,
+      "learning_rate": 6.91578947368421e-05,
+      "loss": 1.239,
+      "step": 5620
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.375,
+      "learning_rate": 6.9e-05,
+      "loss": 1.4557,
+      "step": 5630
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 10.1875,
+      "learning_rate": 6.884210526315788e-05,
+      "loss": 1.5778,
+      "step": 5640
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.1875,
+      "learning_rate": 6.868421052631578e-05,
+      "loss": 1.5556,
+      "step": 5650
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.15625,
+      "learning_rate": 6.852631578947367e-05,
+      "loss": 1.2642,
+      "step": 5660
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.53125,
+      "learning_rate": 6.836842105263157e-05,
+      "loss": 1.3846,
+      "step": 5670
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 4.84375,
+      "learning_rate": 6.821052631578947e-05,
+      "loss": 1.6962,
+      "step": 5680
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 7.03125,
+      "learning_rate": 6.805263157894736e-05,
+      "loss": 1.7847,
+      "step": 5690
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 3.734375,
+      "learning_rate": 6.789473684210526e-05,
+      "loss": 1.618,
+      "step": 5700
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 7.59375,
+      "learning_rate": 6.773684210526316e-05,
+      "loss": 1.2276,
+      "step": 5710
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 4.0,
+      "learning_rate": 6.757894736842104e-05,
+      "loss": 1.4253,
+      "step": 5720
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.34375,
+      "learning_rate": 6.742105263157894e-05,
+      "loss": 1.5658,
+      "step": 5730
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.875,
+      "learning_rate": 6.726315789473683e-05,
+      "loss": 1.6033,
+      "step": 5740
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.90625,
+      "learning_rate": 6.710526315789473e-05,
+      "loss": 1.3583,
+      "step": 5750
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.15625,
+      "learning_rate": 6.694736842105263e-05,
+      "loss": 1.4659,
+      "step": 5760
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.09375,
+      "learning_rate": 6.678947368421052e-05,
+      "loss": 1.7884,
+      "step": 5770
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.0625,
+      "learning_rate": 6.663157894736842e-05,
+      "loss": 2.0443,
+      "step": 5780
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.90625,
+      "learning_rate": 6.64736842105263e-05,
+      "loss": 1.7151,
+      "step": 5790
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.453125,
+      "learning_rate": 6.63157894736842e-05,
+      "loss": 1.6927,
+      "step": 5800
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.0625,
+      "learning_rate": 6.61578947368421e-05,
+      "loss": 1.4678,
+      "step": 5810
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.171875,
+      "learning_rate": 6.599999999999999e-05,
+      "loss": 1.2995,
+      "step": 5820
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.0625,
+      "learning_rate": 6.584210526315789e-05,
+      "loss": 1.503,
+      "step": 5830
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.84375,
+      "learning_rate": 6.568421052631578e-05,
+      "loss": 1.3394,
+      "step": 5840
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 4.46875,
+      "learning_rate": 6.552631578947368e-05,
+      "loss": 1.1829,
+      "step": 5850
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.21875,
+      "learning_rate": 6.536842105263156e-05,
+      "loss": 1.1117,
+      "step": 5860
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.0,
+      "learning_rate": 6.521052631578946e-05,
+      "loss": 1.7312,
+      "step": 5870
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.4375,
+      "learning_rate": 6.505263157894736e-05,
+      "loss": 1.459,
+      "step": 5880
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 4.84375,
+      "learning_rate": 6.489473684210525e-05,
+      "loss": 1.179,
+      "step": 5890
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.78125,
+      "learning_rate": 6.473684210526315e-05,
+      "loss": 1.4745,
+      "step": 5900
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.4375,
+      "learning_rate": 6.457894736842105e-05,
+      "loss": 1.7963,
+      "step": 5910
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.9375,
+      "learning_rate": 6.442105263157894e-05,
+      "loss": 1.23,
+      "step": 5920
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 3.609375,
+      "learning_rate": 6.426315789473683e-05,
+      "loss": 1.3367,
+      "step": 5930
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.15625,
+      "learning_rate": 6.410526315789472e-05,
+      "loss": 1.4904,
+      "step": 5940
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.375,
+      "learning_rate": 6.394736842105262e-05,
+      "loss": 1.2374,
+      "step": 5950
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.21875,
+      "learning_rate": 6.378947368421052e-05,
+      "loss": 1.4946,
+      "step": 5960
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.6875,
+      "learning_rate": 6.363157894736841e-05,
+      "loss": 1.4075,
+      "step": 5970
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.59375,
+      "learning_rate": 6.347368421052631e-05,
+      "loss": 1.6144,
+      "step": 5980
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.71875,
+      "learning_rate": 6.33157894736842e-05,
+      "loss": 1.2361,
+      "step": 5990
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.625,
+      "learning_rate": 6.315789473684209e-05,
+      "loss": 1.4764,
+      "step": 6000
+    },
+    {
+      "epoch": 0.3,
+      "eval_loss": 1.4902899265289307,
+      "eval_runtime": 30.9163,
+      "eval_samples_per_second": 32.345,
+      "eval_steps_per_second": 32.345,
+      "step": 6000
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.875,
+      "learning_rate": 6.299999999999999e-05,
+      "loss": 1.568,
+      "step": 6010
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.8125,
+      "learning_rate": 6.284210526315788e-05,
+      "loss": 1.3994,
+      "step": 6020
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.71875,
+      "learning_rate": 6.268421052631578e-05,
+      "loss": 1.5673,
+      "step": 6030
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.5625,
+      "learning_rate": 6.252631578947368e-05,
+      "loss": 1.4264,
+      "step": 6040
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.34375,
+      "learning_rate": 6.236842105263157e-05,
+      "loss": 1.6823,
+      "step": 6050
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 3.84375,
+      "learning_rate": 6.221052631578947e-05,
+      "loss": 1.4544,
+      "step": 6060
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.9375,
+      "learning_rate": 6.205263157894737e-05,
+      "loss": 1.4708,
+      "step": 6070
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.0,
+      "learning_rate": 6.189473684210525e-05,
+      "loss": 1.4171,
+      "step": 6080
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.90625,
+      "learning_rate": 6.173684210526315e-05,
+      "loss": 1.8525,
+      "step": 6090
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.4375,
+      "learning_rate": 6.157894736842104e-05,
+      "loss": 1.386,
+      "step": 6100
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.09375,
+      "learning_rate": 6.142105263157894e-05,
+      "loss": 1.2856,
+      "step": 6110
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 4.71875,
+      "learning_rate": 6.126315789473684e-05,
+      "loss": 1.4263,
+      "step": 6120
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 8.0,
+      "learning_rate": 6.110526315789473e-05,
+      "loss": 1.4666,
+      "step": 6130
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.875,
+      "learning_rate": 6.094736842105262e-05,
+      "loss": 1.3614,
+      "step": 6140
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.03125,
+      "learning_rate": 6.078947368421052e-05,
+      "loss": 1.3097,
+      "step": 6150
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.15625,
+      "learning_rate": 6.0631578947368415e-05,
+      "loss": 1.5042,
+      "step": 6160
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.1875,
+      "learning_rate": 6.047368421052631e-05,
+      "loss": 1.0817,
+      "step": 6170
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.6875,
+      "learning_rate": 6.031578947368421e-05,
+      "loss": 1.5722,
+      "step": 6180
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.9375,
+      "learning_rate": 6.0157894736842105e-05,
+      "loss": 1.366,
+      "step": 6190
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 4.1875,
+      "learning_rate": 5.9999999999999995e-05,
+      "loss": 1.6359,
+      "step": 6200
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.125,
+      "learning_rate": 5.9842105263157885e-05,
+      "loss": 1.7341,
+      "step": 6210
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 3.625,
+      "learning_rate": 5.968421052631578e-05,
+      "loss": 1.7589,
+      "step": 6220
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.21875,
+      "learning_rate": 5.952631578947368e-05,
+      "loss": 1.6357,
+      "step": 6230
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 7.25,
+      "learning_rate": 5.9368421052631574e-05,
+      "loss": 1.4806,
+      "step": 6240
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.09375,
+      "learning_rate": 5.921052631578947e-05,
+      "loss": 1.5461,
+      "step": 6250
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 6.5625,
+      "learning_rate": 5.905263157894737e-05,
+      "loss": 1.3267,
+      "step": 6260
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 4.90625,
+      "learning_rate": 5.8894736842105264e-05,
+      "loss": 1.5908,
+      "step": 6270
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 3.890625,
+      "learning_rate": 5.873684210526315e-05,
+      "loss": 1.5187,
+      "step": 6280
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.875,
+      "learning_rate": 5.8578947368421044e-05,
+      "loss": 1.5094,
+      "step": 6290
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 7.625,
+      "learning_rate": 5.842105263157894e-05,
+      "loss": 1.5578,
+      "step": 6300
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.65625,
+      "learning_rate": 5.826315789473684e-05,
+      "loss": 1.5853,
+      "step": 6310
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.25,
+      "learning_rate": 5.8105263157894734e-05,
+      "loss": 1.0628,
+      "step": 6320
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.0625,
+      "learning_rate": 5.794736842105263e-05,
+      "loss": 1.6518,
+      "step": 6330
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.25,
+      "learning_rate": 5.778947368421053e-05,
+      "loss": 1.7742,
+      "step": 6340
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4.625,
+      "learning_rate": 5.763157894736841e-05,
+      "loss": 1.3831,
+      "step": 6350
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4.15625,
+      "learning_rate": 5.7473684210526307e-05,
+      "loss": 1.2819,
+      "step": 6360
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 7.6875,
+      "learning_rate": 5.73157894736842e-05,
+      "loss": 1.7622,
+      "step": 6370
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.0625,
+      "learning_rate": 5.71578947368421e-05,
+      "loss": 1.3274,
+      "step": 6380
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.6875,
+      "learning_rate": 5.6999999999999996e-05,
+      "loss": 1.5017,
+      "step": 6390
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.40625,
+      "learning_rate": 5.684210526315789e-05,
+      "loss": 1.5522,
+      "step": 6400
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.21875,
+      "learning_rate": 5.668421052631579e-05,
+      "loss": 1.752,
+      "step": 6410
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 8.1875,
+      "learning_rate": 5.652631578947367e-05,
+      "loss": 1.3857,
+      "step": 6420
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 7.4375,
+      "learning_rate": 5.636842105263157e-05,
+      "loss": 1.4945,
+      "step": 6430
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4.4375,
+      "learning_rate": 5.6210526315789466e-05,
+      "loss": 1.5736,
+      "step": 6440
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.0625,
+      "learning_rate": 5.605263157894736e-05,
+      "loss": 1.4432,
+      "step": 6450
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 9.625,
+      "learning_rate": 5.589473684210526e-05,
+      "loss": 1.499,
+      "step": 6460
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.34375,
+      "learning_rate": 5.5736842105263156e-05,
+      "loss": 1.54,
+      "step": 6470
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.40625,
+      "learning_rate": 5.557894736842105e-05,
+      "loss": 1.3533,
+      "step": 6480
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 8.0625,
+      "learning_rate": 5.542105263157895e-05,
+      "loss": 1.335,
+      "step": 6490
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.875,
+      "learning_rate": 5.526315789473683e-05,
+      "loss": 1.498,
+      "step": 6500
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.3125,
+      "learning_rate": 5.510526315789473e-05,
+      "loss": 1.37,
+      "step": 6510
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.3125,
+      "learning_rate": 5.4947368421052625e-05,
+      "loss": 1.3711,
+      "step": 6520
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.5625,
+      "learning_rate": 5.478947368421052e-05,
+      "loss": 1.3383,
+      "step": 6530
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.09375,
+      "learning_rate": 5.463157894736842e-05,
+      "loss": 1.5205,
+      "step": 6540
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 7.09375,
+      "learning_rate": 5.4473684210526315e-05,
+      "loss": 1.4338,
+      "step": 6550
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 7.6875,
+      "learning_rate": 5.431578947368421e-05,
+      "loss": 1.4553,
+      "step": 6560
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.8125,
+      "learning_rate": 5.4157894736842095e-05,
+      "loss": 1.5879,
+      "step": 6570
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.0625,
+      "learning_rate": 5.399999999999999e-05,
+      "loss": 1.4582,
+      "step": 6580
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.0,
+      "learning_rate": 5.384210526315789e-05,
+      "loss": 1.3241,
+      "step": 6590
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.96875,
+      "learning_rate": 5.3684210526315784e-05,
+      "loss": 1.7072,
+      "step": 6600
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 7.34375,
+      "learning_rate": 5.352631578947368e-05,
+      "loss": 1.5041,
+      "step": 6610
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.28125,
+      "learning_rate": 5.336842105263158e-05,
+      "loss": 1.7467,
+      "step": 6620
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.375,
+      "learning_rate": 5.3210526315789474e-05,
+      "loss": 1.3781,
+      "step": 6630
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.0,
+      "learning_rate": 5.305263157894736e-05,
+      "loss": 1.3092,
+      "step": 6640
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.796875,
+      "learning_rate": 5.2894736842105254e-05,
+      "loss": 1.2497,
+      "step": 6650
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 4.9375,
+      "learning_rate": 5.273684210526315e-05,
+      "loss": 1.0076,
+      "step": 6660
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 11.0625,
+      "learning_rate": 5.257894736842105e-05,
+      "loss": 1.4536,
+      "step": 6670
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 7.09375,
+      "learning_rate": 5.2421052631578944e-05,
+      "loss": 1.3639,
+      "step": 6680
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 8.9375,
+      "learning_rate": 5.226315789473684e-05,
+      "loss": 1.5545,
+      "step": 6690
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.59375,
+      "learning_rate": 5.210526315789474e-05,
+      "loss": 1.2892,
+      "step": 6700
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.125,
+      "learning_rate": 5.194736842105263e-05,
+      "loss": 1.4664,
+      "step": 6710
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.875,
+      "learning_rate": 5.1789473684210517e-05,
+      "loss": 1.542,
+      "step": 6720
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.15625,
+      "learning_rate": 5.163157894736841e-05,
+      "loss": 1.6154,
+      "step": 6730
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 4.21875,
+      "learning_rate": 5.147368421052631e-05,
+      "loss": 1.289,
+      "step": 6740
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 4.75,
+      "learning_rate": 5.1315789473684206e-05,
+      "loss": 1.4106,
+      "step": 6750
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.625,
+      "learning_rate": 5.11578947368421e-05,
+      "loss": 1.4117,
+      "step": 6760
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.875,
+      "learning_rate": 5.1e-05,
+      "loss": 1.5164,
+      "step": 6770
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.25,
+      "learning_rate": 5.084210526315789e-05,
+      "loss": 1.3858,
+      "step": 6780
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 7.3125,
+      "learning_rate": 5.0684210526315786e-05,
+      "loss": 1.6531,
+      "step": 6790
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.125,
+      "learning_rate": 5.0526315789473676e-05,
+      "loss": 1.4197,
+      "step": 6800
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.625,
+      "learning_rate": 5.036842105263157e-05,
+      "loss": 1.6369,
+      "step": 6810
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 9.125,
+      "learning_rate": 5.021052631578947e-05,
+      "loss": 1.4371,
+      "step": 6820
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.25,
+      "learning_rate": 5.0052631578947366e-05,
+      "loss": 1.3158,
+      "step": 6830
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.25,
+      "learning_rate": 4.989473684210526e-05,
+      "loss": 1.421,
+      "step": 6840
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 4.75,
+      "learning_rate": 4.973684210526315e-05,
+      "loss": 1.3778,
+      "step": 6850
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 4.9375,
+      "learning_rate": 4.957894736842105e-05,
+      "loss": 1.4383,
+      "step": 6860
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 4.6875,
+      "learning_rate": 4.942105263157894e-05,
+      "loss": 1.4528,
+      "step": 6870
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.53125,
+      "learning_rate": 4.9263157894736835e-05,
+      "loss": 1.4387,
+      "step": 6880
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.125,
+      "learning_rate": 4.910526315789473e-05,
+      "loss": 1.4136,
+      "step": 6890
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.375,
+      "learning_rate": 4.894736842105263e-05,
+      "loss": 1.5636,
+      "step": 6900
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.6875,
+      "learning_rate": 4.8789473684210525e-05,
+      "loss": 1.3611,
+      "step": 6910
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 7.15625,
+      "learning_rate": 4.8631578947368415e-05,
+      "loss": 1.6587,
+      "step": 6920
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.78125,
+      "learning_rate": 4.847368421052631e-05,
+      "loss": 1.4203,
+      "step": 6930
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.0,
+      "learning_rate": 4.831578947368421e-05,
+      "loss": 1.3195,
+      "step": 6940
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.0625,
+      "learning_rate": 4.81578947368421e-05,
+      "loss": 1.5523,
+      "step": 6950
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.0,
+      "learning_rate": 4.7999999999999994e-05,
+      "loss": 1.2051,
+      "step": 6960
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.625,
+      "learning_rate": 4.784210526315789e-05,
+      "loss": 1.3228,
+      "step": 6970
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.53125,
+      "learning_rate": 4.768421052631579e-05,
+      "loss": 1.4439,
+      "step": 6980
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.875,
+      "learning_rate": 4.752631578947368e-05,
+      "loss": 1.2889,
+      "step": 6990
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.78125,
+      "learning_rate": 4.7368421052631574e-05,
+      "loss": 1.6213,
+      "step": 7000
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.8125,
+      "learning_rate": 4.721052631578947e-05,
+      "loss": 1.2401,
+      "step": 7010
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.125,
+      "learning_rate": 4.705263157894737e-05,
+      "loss": 1.6266,
+      "step": 7020
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.15625,
+      "learning_rate": 4.689473684210526e-05,
+      "loss": 1.3183,
+      "step": 7030
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.71875,
+      "learning_rate": 4.6736842105263154e-05,
+      "loss": 1.4612,
+      "step": 7040
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 7.8125,
+      "learning_rate": 4.657894736842105e-05,
+      "loss": 1.3345,
+      "step": 7050
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.6875,
+      "learning_rate": 4.642105263157894e-05,
+      "loss": 1.3447,
+      "step": 7060
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.53125,
+      "learning_rate": 4.626315789473684e-05,
+      "loss": 1.4236,
+      "step": 7070
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.3125,
+      "learning_rate": 4.610526315789473e-05,
+      "loss": 0.9894,
+      "step": 7080
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.9375,
+      "learning_rate": 4.594736842105263e-05,
+      "loss": 1.6438,
+      "step": 7090
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 4.125,
+      "learning_rate": 4.578947368421052e-05,
+      "loss": 1.432,
+      "step": 7100
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 8.0625,
+      "learning_rate": 4.5631578947368416e-05,
+      "loss": 1.4258,
+      "step": 7110
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.03125,
+      "learning_rate": 4.547368421052631e-05,
+      "loss": 1.2104,
+      "step": 7120
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.84375,
+      "learning_rate": 4.53157894736842e-05,
+      "loss": 1.5062,
+      "step": 7130
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 4.4375,
+      "learning_rate": 4.51578947368421e-05,
+      "loss": 1.1665,
+      "step": 7140
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 2.25,
+      "learning_rate": 4.4999999999999996e-05,
+      "loss": 1.4222,
+      "step": 7150
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.09375,
+      "learning_rate": 4.484210526315789e-05,
+      "loss": 1.2456,
+      "step": 7160
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 9.5625,
+      "learning_rate": 4.468421052631579e-05,
+      "loss": 1.6563,
+      "step": 7170
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.9375,
+      "learning_rate": 4.452631578947368e-05,
+      "loss": 1.4191,
+      "step": 7180
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.125,
+      "learning_rate": 4.4368421052631576e-05,
+      "loss": 1.4417,
+      "step": 7190
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.34375,
+      "learning_rate": 4.4210526315789466e-05,
+      "loss": 1.2115,
+      "step": 7200
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.0,
+      "learning_rate": 4.405263157894736e-05,
+      "loss": 1.5196,
+      "step": 7210
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.375,
+      "learning_rate": 4.389473684210526e-05,
+      "loss": 1.4436,
+      "step": 7220
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.90625,
+      "learning_rate": 4.3736842105263155e-05,
+      "loss": 1.3976,
+      "step": 7230
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 7.8125,
+      "learning_rate": 4.357894736842105e-05,
+      "loss": 1.5432,
+      "step": 7240
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 3.234375,
+      "learning_rate": 4.342105263157895e-05,
+      "loss": 1.2969,
+      "step": 7250
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.8125,
+      "learning_rate": 4.326315789473684e-05,
+      "loss": 1.3246,
+      "step": 7260
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 9.8125,
+      "learning_rate": 4.3105263157894735e-05,
+      "loss": 1.3552,
+      "step": 7270
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 3.84375,
+      "learning_rate": 4.2947368421052625e-05,
+      "loss": 1.1,
+      "step": 7280
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.25,
+      "learning_rate": 4.278947368421052e-05,
+      "loss": 1.449,
+      "step": 7290
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 8.5,
+      "learning_rate": 4.263157894736842e-05,
+      "loss": 1.572,
+      "step": 7300
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.1875,
+      "learning_rate": 4.2473684210526315e-05,
+      "loss": 1.2444,
+      "step": 7310
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.0,
+      "learning_rate": 4.231578947368421e-05,
+      "loss": 1.1988,
+      "step": 7320
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.4375,
+      "learning_rate": 4.21578947368421e-05,
+      "loss": 1.6043,
+      "step": 7330
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.3125,
+      "learning_rate": 4.2e-05,
+      "loss": 1.2237,
+      "step": 7340
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 7.4375,
+      "learning_rate": 4.184210526315789e-05,
+      "loss": 1.6211,
+      "step": 7350
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.28125,
+      "learning_rate": 4.1684210526315784e-05,
+      "loss": 1.5567,
+      "step": 7360
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 7.375,
+      "learning_rate": 4.152631578947368e-05,
+      "loss": 1.3958,
+      "step": 7370
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.5625,
+      "learning_rate": 4.136842105263158e-05,
+      "loss": 1.3257,
+      "step": 7380
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 8.875,
+      "learning_rate": 4.1210526315789474e-05,
+      "loss": 1.3897,
+      "step": 7390
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.0,
+      "learning_rate": 4.105263157894737e-05,
+      "loss": 1.3825,
+      "step": 7400
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 4.75,
+      "learning_rate": 4.089473684210526e-05,
+      "loss": 1.7153,
+      "step": 7410
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.21875,
+      "learning_rate": 4.073684210526315e-05,
+      "loss": 1.3951,
+      "step": 7420
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 7.34375,
+      "learning_rate": 4.057894736842105e-05,
+      "loss": 1.7072,
+      "step": 7430
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 4.8125,
+      "learning_rate": 4.0421052631578943e-05,
+      "loss": 1.3874,
+      "step": 7440
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 7.09375,
+      "learning_rate": 4.026315789473684e-05,
+      "loss": 1.3483,
+      "step": 7450
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 10.0625,
+      "learning_rate": 4.0105263157894737e-05,
+      "loss": 1.3684,
+      "step": 7460
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.03125,
+      "learning_rate": 3.994736842105263e-05,
+      "loss": 1.1862,
+      "step": 7470
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.53125,
+      "learning_rate": 3.978947368421053e-05,
+      "loss": 1.8852,
+      "step": 7480
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.5,
+      "learning_rate": 3.963157894736841e-05,
+      "loss": 1.5302,
+      "step": 7490
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 6.1875,
+      "learning_rate": 3.947368421052631e-05,
+      "loss": 1.2026,
+      "step": 7500
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.5,
+      "learning_rate": 3.9315789473684206e-05,
+      "loss": 1.3831,
+      "step": 7510
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 6.5625,
+      "learning_rate": 3.91578947368421e-05,
+      "loss": 1.4271,
+      "step": 7520
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 4.5,
+      "learning_rate": 3.9e-05,
+      "loss": 1.2867,
+      "step": 7530
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 2.828125,
+      "learning_rate": 3.8842105263157896e-05,
+      "loss": 0.9484,
+      "step": 7540
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 8.875,
+      "learning_rate": 3.868421052631579e-05,
+      "loss": 1.4929,
+      "step": 7550
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 4.09375,
+      "learning_rate": 3.8526315789473676e-05,
+      "loss": 1.4549,
+      "step": 7560
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.75,
+      "learning_rate": 3.836842105263157e-05,
+      "loss": 1.1515,
+      "step": 7570
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.3125,
+      "learning_rate": 3.821052631578947e-05,
+      "loss": 1.3083,
+      "step": 7580
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.8125,
+      "learning_rate": 3.8052631578947365e-05,
+      "loss": 1.1697,
+      "step": 7590
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.0625,
+      "learning_rate": 3.789473684210526e-05,
+      "loss": 1.4502,
+      "step": 7600
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.3125,
+      "learning_rate": 3.773684210526316e-05,
+      "loss": 1.4979,
+      "step": 7610
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.09375,
+      "learning_rate": 3.7578947368421055e-05,
+      "loss": 1.4675,
+      "step": 7620
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 7.1875,
+      "learning_rate": 3.7421052631578945e-05,
+      "loss": 1.3028,
+      "step": 7630
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 4.71875,
+      "learning_rate": 3.726315789473684e-05,
+      "loss": 1.337,
+      "step": 7640
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 8.4375,
+      "learning_rate": 3.710526315789473e-05,
+      "loss": 1.5052,
+      "step": 7650
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.6875,
+      "learning_rate": 3.694736842105263e-05,
+      "loss": 1.4187,
+      "step": 7660
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 6.3125,
+      "learning_rate": 3.6789473684210525e-05,
+      "loss": 1.2143,
+      "step": 7670
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 7.8125,
+      "learning_rate": 3.6631578947368414e-05,
+      "loss": 1.4787,
+      "step": 7680
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.34375,
+      "learning_rate": 3.647368421052631e-05,
+      "loss": 1.5673,
+      "step": 7690
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.828125,
+      "learning_rate": 3.631578947368421e-05,
+      "loss": 1.2965,
+      "step": 7700
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.46875,
+      "learning_rate": 3.6157894736842104e-05,
+      "loss": 1.3391,
+      "step": 7710
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.6875,
+      "learning_rate": 3.5999999999999994e-05,
+      "loss": 0.8979,
+      "step": 7720
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.25,
+      "learning_rate": 3.584210526315789e-05,
+      "loss": 1.3974,
+      "step": 7730
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.53125,
+      "learning_rate": 3.568421052631579e-05,
+      "loss": 1.562,
+      "step": 7740
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 7.84375,
+      "learning_rate": 3.552631578947368e-05,
+      "loss": 1.4908,
+      "step": 7750
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.03125,
+      "learning_rate": 3.5368421052631574e-05,
+      "loss": 1.5491,
+      "step": 7760
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 7.59375,
+      "learning_rate": 3.521052631578947e-05,
+      "loss": 1.5424,
+      "step": 7770
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.0,
+      "learning_rate": 3.505263157894737e-05,
+      "loss": 1.7824,
+      "step": 7780
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.84375,
+      "learning_rate": 3.489473684210526e-05,
+      "loss": 1.5703,
+      "step": 7790
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 8.4375,
+      "learning_rate": 3.4736842105263153e-05,
+      "loss": 1.079,
+      "step": 7800
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.5,
+      "learning_rate": 3.457894736842105e-05,
+      "loss": 1.438,
+      "step": 7810
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.28125,
+      "learning_rate": 3.442105263157894e-05,
+      "loss": 1.5293,
+      "step": 7820
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.09375,
+      "learning_rate": 3.4263157894736836e-05,
+      "loss": 1.2791,
+      "step": 7830
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.71875,
+      "learning_rate": 3.410526315789473e-05,
+      "loss": 1.2439,
+      "step": 7840
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.140625,
+      "learning_rate": 3.394736842105263e-05,
+      "loss": 1.1694,
+      "step": 7850
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.40625,
+      "learning_rate": 3.378947368421052e-05,
+      "loss": 1.5024,
+      "step": 7860
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.5625,
+      "learning_rate": 3.3631578947368416e-05,
+      "loss": 1.2139,
+      "step": 7870
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.9375,
+      "learning_rate": 3.347368421052631e-05,
+      "loss": 1.5477,
+      "step": 7880
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.625,
+      "learning_rate": 3.331578947368421e-05,
+      "loss": 1.4138,
+      "step": 7890
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.8125,
+      "learning_rate": 3.31578947368421e-05,
+      "loss": 1.2124,
+      "step": 7900
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 9.6875,
+      "learning_rate": 3.2999999999999996e-05,
+      "loss": 1.3678,
+      "step": 7910
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 3.703125,
+      "learning_rate": 3.284210526315789e-05,
+      "loss": 1.1943,
+      "step": 7920
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 10.5625,
+      "learning_rate": 3.268421052631578e-05,
+      "loss": 1.3115,
+      "step": 7930
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 7.5,
+      "learning_rate": 3.252631578947368e-05,
+      "loss": 1.3972,
+      "step": 7940
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.21875,
+      "learning_rate": 3.2368421052631575e-05,
+      "loss": 1.1899,
+      "step": 7950
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 8.125,
+      "learning_rate": 3.221052631578947e-05,
+      "loss": 1.439,
+      "step": 7960
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.09375,
+      "learning_rate": 3.205263157894736e-05,
+      "loss": 1.55,
+      "step": 7970
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.625,
+      "learning_rate": 3.189473684210526e-05,
+      "loss": 1.6196,
+      "step": 7980
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 7.96875,
+      "learning_rate": 3.1736842105263155e-05,
+      "loss": 1.1113,
+      "step": 7990
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.1875,
+      "learning_rate": 3.1578947368421045e-05,
+      "loss": 1.3067,
+      "step": 8000
+    },
+    {
+      "epoch": 0.4,
+      "eval_loss": 1.3947522640228271,
+      "eval_runtime": 30.8209,
+      "eval_samples_per_second": 32.445,
+      "eval_steps_per_second": 32.445,
+      "step": 8000
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.9375,
+      "learning_rate": 3.142105263157894e-05,
+      "loss": 1.4431,
+      "step": 8010
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.125,
+      "learning_rate": 3.126315789473684e-05,
+      "loss": 1.5649,
+      "step": 8020
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 8.4375,
+      "learning_rate": 3.1105263157894735e-05,
+      "loss": 1.6709,
+      "step": 8030
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.34375,
+      "learning_rate": 3.0947368421052625e-05,
+      "loss": 1.2212,
+      "step": 8040
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.90625,
+      "learning_rate": 3.078947368421052e-05,
+      "loss": 1.2499,
+      "step": 8050
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 7.15625,
+      "learning_rate": 3.063157894736842e-05,
+      "loss": 1.4421,
+      "step": 8060
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 4.625,
+      "learning_rate": 3.047368421052631e-05,
+      "loss": 1.4175,
+      "step": 8070
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.0625,
+      "learning_rate": 3.0315789473684208e-05,
+      "loss": 1.4997,
+      "step": 8080
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 4.8125,
+      "learning_rate": 3.0157894736842104e-05,
+      "loss": 1.6382,
+      "step": 8090
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 6.0,
+      "learning_rate": 2.9999999999999997e-05,
+      "loss": 1.3528,
+      "step": 8100
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 8.3125,
+      "learning_rate": 2.984210526315789e-05,
+      "loss": 1.4061,
+      "step": 8110
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.0625,
+      "learning_rate": 2.9684210526315787e-05,
+      "loss": 1.7639,
+      "step": 8120
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 3.859375,
+      "learning_rate": 2.9526315789473684e-05,
+      "loss": 1.152,
+      "step": 8130
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 8.75,
+      "learning_rate": 2.9368421052631574e-05,
+      "loss": 1.3679,
+      "step": 8140
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.84375,
+      "learning_rate": 2.921052631578947e-05,
+      "loss": 1.2758,
+      "step": 8150
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.28125,
+      "learning_rate": 2.9052631578947367e-05,
+      "loss": 1.2823,
+      "step": 8160
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 9.5625,
+      "learning_rate": 2.8894736842105263e-05,
+      "loss": 1.6093,
+      "step": 8170
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 8.0,
+      "learning_rate": 2.8736842105263153e-05,
+      "loss": 1.3733,
+      "step": 8180
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 7.0,
+      "learning_rate": 2.857894736842105e-05,
+      "loss": 1.6597,
+      "step": 8190
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 7.96875,
+      "learning_rate": 2.8421052631578946e-05,
+      "loss": 1.2709,
+      "step": 8200
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.65625,
+      "learning_rate": 2.8263157894736836e-05,
+      "loss": 1.3686,
+      "step": 8210
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 8.1875,
+      "learning_rate": 2.8105263157894733e-05,
+      "loss": 1.2215,
+      "step": 8220
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 4.875,
+      "learning_rate": 2.794736842105263e-05,
+      "loss": 1.5427,
+      "step": 8230
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.78125,
+      "learning_rate": 2.7789473684210526e-05,
+      "loss": 1.6579,
+      "step": 8240
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 6.53125,
+      "learning_rate": 2.7631578947368416e-05,
+      "loss": 1.3988,
+      "step": 8250
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 9.875,
+      "learning_rate": 2.7473684210526313e-05,
+      "loss": 1.4041,
+      "step": 8260
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.6875,
+      "learning_rate": 2.731578947368421e-05,
+      "loss": 1.126,
+      "step": 8270
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 3.0,
+      "learning_rate": 2.7157894736842106e-05,
+      "loss": 1.2544,
+      "step": 8280
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.1875,
+      "learning_rate": 2.6999999999999996e-05,
+      "loss": 1.3303,
+      "step": 8290
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 3.15625,
+      "learning_rate": 2.6842105263157892e-05,
+      "loss": 1.5063,
+      "step": 8300
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 8.6875,
+      "learning_rate": 2.668421052631579e-05,
+      "loss": 1.3768,
+      "step": 8310
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 8.1875,
+      "learning_rate": 2.652631578947368e-05,
+      "loss": 1.3025,
+      "step": 8320
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.03125,
+      "learning_rate": 2.6368421052631575e-05,
+      "loss": 1.2676,
+      "step": 8330
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.5625,
+      "learning_rate": 2.6210526315789472e-05,
+      "loss": 1.2945,
+      "step": 8340
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.875,
+      "learning_rate": 2.605263157894737e-05,
+      "loss": 1.4027,
+      "step": 8350
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 7.25,
+      "learning_rate": 2.5894736842105258e-05,
+      "loss": 0.8923,
+      "step": 8360
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.3125,
+      "learning_rate": 2.5736842105263155e-05,
+      "loss": 1.3464,
+      "step": 8370
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.03125,
+      "learning_rate": 2.557894736842105e-05,
+      "loss": 1.4385,
+      "step": 8380
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 8.6875,
+      "learning_rate": 2.5421052631578945e-05,
+      "loss": 1.283,
+      "step": 8390
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.1875,
+      "learning_rate": 2.5263157894736838e-05,
+      "loss": 1.1647,
+      "step": 8400
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 7.125,
+      "learning_rate": 2.5105263157894735e-05,
+      "loss": 1.3794,
+      "step": 8410
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 7.0,
+      "learning_rate": 2.494736842105263e-05,
+      "loss": 1.2719,
+      "step": 8420
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.40625,
+      "learning_rate": 2.4789473684210524e-05,
+      "loss": 1.3743,
+      "step": 8430
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 7.46875,
+      "learning_rate": 2.4631578947368418e-05,
+      "loss": 1.2803,
+      "step": 8440
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 7.84375,
+      "learning_rate": 2.4473684210526314e-05,
+      "loss": 1.3228,
+      "step": 8450
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.96875,
+      "learning_rate": 2.4315789473684207e-05,
+      "loss": 1.4926,
+      "step": 8460
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.5625,
+      "learning_rate": 2.4157894736842104e-05,
+      "loss": 1.3543,
+      "step": 8470
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 7.1875,
+      "learning_rate": 2.3999999999999997e-05,
+      "loss": 1.337,
+      "step": 8480
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 7.96875,
+      "learning_rate": 2.3842105263157894e-05,
+      "loss": 1.09,
+      "step": 8490
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 7.21875,
+      "learning_rate": 2.3684210526315787e-05,
+      "loss": 1.5419,
+      "step": 8500
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.0,
+      "learning_rate": 2.3526315789473684e-05,
+      "loss": 1.5295,
+      "step": 8510
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.5,
+      "learning_rate": 2.3368421052631577e-05,
+      "loss": 1.3311,
+      "step": 8520
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.75,
+      "learning_rate": 2.321052631578947e-05,
+      "loss": 1.4346,
+      "step": 8530
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 4.75,
+      "learning_rate": 2.3052631578947367e-05,
+      "loss": 1.1276,
+      "step": 8540
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 10.125,
+      "learning_rate": 2.289473684210526e-05,
+      "loss": 1.4764,
+      "step": 8550
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.09375,
+      "learning_rate": 2.2736842105263157e-05,
+      "loss": 1.3054,
+      "step": 8560
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 8.0,
+      "learning_rate": 2.257894736842105e-05,
+      "loss": 1.7071,
+      "step": 8570
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.21875,
+      "learning_rate": 2.2421052631578946e-05,
+      "loss": 1.1322,
+      "step": 8580
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.6875,
+      "learning_rate": 2.226315789473684e-05,
+      "loss": 1.2859,
+      "step": 8590
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.8125,
+      "learning_rate": 2.2105263157894733e-05,
+      "loss": 1.4777,
+      "step": 8600
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.0625,
+      "learning_rate": 2.194736842105263e-05,
+      "loss": 1.2259,
+      "step": 8610
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.125,
+      "learning_rate": 2.1789473684210526e-05,
+      "loss": 1.1973,
+      "step": 8620
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 7.75,
+      "learning_rate": 2.163157894736842e-05,
+      "loss": 1.3127,
+      "step": 8630
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.90625,
+      "learning_rate": 2.1473684210526312e-05,
+      "loss": 1.278,
+      "step": 8640
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.03125,
+      "learning_rate": 2.131578947368421e-05,
+      "loss": 1.4912,
+      "step": 8650
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.6875,
+      "learning_rate": 2.1157894736842106e-05,
+      "loss": 1.2192,
+      "step": 8660
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 7.0,
+      "learning_rate": 2.1e-05,
+      "loss": 1.387,
+      "step": 8670
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.78125,
+      "learning_rate": 2.0842105263157892e-05,
+      "loss": 1.327,
+      "step": 8680
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.625,
+      "learning_rate": 2.068421052631579e-05,
+      "loss": 1.2982,
+      "step": 8690
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.15625,
+      "learning_rate": 2.0526315789473685e-05,
+      "loss": 1.3495,
+      "step": 8700
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 6.65625,
+      "learning_rate": 2.0368421052631575e-05,
+      "loss": 1.4706,
+      "step": 8710
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.28125,
+      "learning_rate": 2.0210526315789472e-05,
+      "loss": 1.268,
+      "step": 8720
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 2.390625,
+      "learning_rate": 2.0052631578947368e-05,
+      "loss": 1.2314,
+      "step": 8730
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.84375,
+      "learning_rate": 1.9894736842105265e-05,
+      "loss": 1.3312,
+      "step": 8740
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 6.875,
+      "learning_rate": 1.9736842105263155e-05,
+      "loss": 1.3956,
+      "step": 8750
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.90625,
+      "learning_rate": 1.957894736842105e-05,
+      "loss": 1.4379,
+      "step": 8760
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 7.15625,
+      "learning_rate": 1.9421052631578948e-05,
+      "loss": 1.6044,
+      "step": 8770
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.46875,
+      "learning_rate": 1.9263157894736838e-05,
+      "loss": 1.3037,
+      "step": 8780
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 8.875,
+      "learning_rate": 1.9105263157894734e-05,
+      "loss": 1.4253,
+      "step": 8790
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 8.125,
+      "learning_rate": 1.894736842105263e-05,
+      "loss": 1.3077,
+      "step": 8800
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 14.0625,
+      "learning_rate": 1.8789473684210528e-05,
+      "loss": 1.4148,
+      "step": 8810
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 6.9375,
+      "learning_rate": 1.863157894736842e-05,
+      "loss": 1.4789,
+      "step": 8820
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 8.9375,
+      "learning_rate": 1.8473684210526314e-05,
+      "loss": 1.3468,
+      "step": 8830
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 6.0,
+      "learning_rate": 1.8315789473684207e-05,
+      "loss": 1.4482,
+      "step": 8840
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.875,
+      "learning_rate": 1.8157894736842104e-05,
+      "loss": 1.5264,
+      "step": 8850
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.8125,
+      "learning_rate": 1.7999999999999997e-05,
+      "loss": 1.5689,
+      "step": 8860
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 7.0,
+      "learning_rate": 1.7842105263157894e-05,
+      "loss": 1.5605,
+      "step": 8870
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.5,
+      "learning_rate": 1.7684210526315787e-05,
+      "loss": 1.2151,
+      "step": 8880
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 7.53125,
+      "learning_rate": 1.7526315789473683e-05,
+      "loss": 1.3092,
+      "step": 8890
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 7.84375,
+      "learning_rate": 1.7368421052631577e-05,
+      "loss": 1.3631,
+      "step": 8900
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 5.84375,
+      "learning_rate": 1.721052631578947e-05,
+      "loss": 1.6111,
+      "step": 8910
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.78125,
+      "learning_rate": 1.7052631578947367e-05,
+      "loss": 1.389,
+      "step": 8920
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.689473684210526e-05,
+      "loss": 1.2737,
+      "step": 8930
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 5.5625,
+      "learning_rate": 1.6736842105263156e-05,
+      "loss": 1.2458,
+      "step": 8940
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.65625,
+      "learning_rate": 1.657894736842105e-05,
+      "loss": 1.5805,
+      "step": 8950
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.6421052631578946e-05,
+      "loss": 1.5468,
+      "step": 8960
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 5.5,
+      "learning_rate": 1.626315789473684e-05,
+      "loss": 1.3546,
+      "step": 8970
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 5.3125,
+      "learning_rate": 1.6105263157894736e-05,
+      "loss": 1.3426,
+      "step": 8980
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.71875,
+      "learning_rate": 1.594736842105263e-05,
+      "loss": 1.6354,
+      "step": 8990
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 10.75,
+      "learning_rate": 1.5789473684210522e-05,
+      "loss": 1.6335,
+      "step": 9000
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.0625,
+      "learning_rate": 1.563157894736842e-05,
+      "loss": 1.1367,
+      "step": 9010
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 7.0625,
+      "learning_rate": 1.5473684210526312e-05,
+      "loss": 1.6116,
+      "step": 9020
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 9.6875,
+      "learning_rate": 1.531578947368421e-05,
+      "loss": 1.094,
+      "step": 9030
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 7.0,
+      "learning_rate": 1.5157894736842104e-05,
+      "loss": 1.1548,
+      "step": 9040
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.59375,
+      "learning_rate": 1.4999999999999999e-05,
+      "loss": 1.4296,
+      "step": 9050
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.0,
+      "learning_rate": 1.4842105263157894e-05,
+      "loss": 1.3269,
+      "step": 9060
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.4684210526315787e-05,
+      "loss": 1.2124,
+      "step": 9070
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 7.125,
+      "learning_rate": 1.4526315789473683e-05,
+      "loss": 1.5491,
+      "step": 9080
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 10.625,
+      "learning_rate": 1.4368421052631577e-05,
+      "loss": 1.4008,
+      "step": 9090
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.75,
+      "learning_rate": 1.4210526315789473e-05,
+      "loss": 1.6327,
+      "step": 9100
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.65625,
+      "learning_rate": 1.4052631578947366e-05,
+      "loss": 1.5382,
+      "step": 9110
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.28125,
+      "learning_rate": 1.3894736842105263e-05,
+      "loss": 1.2475,
+      "step": 9120
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.5,
+      "learning_rate": 1.3736842105263156e-05,
+      "loss": 1.4219,
+      "step": 9130
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.3578947368421053e-05,
+      "loss": 1.5508,
+      "step": 9140
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.90625,
+      "learning_rate": 1.3421052631578946e-05,
+      "loss": 1.6626,
+      "step": 9150
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.326315789473684e-05,
+      "loss": 1.0473,
+      "step": 9160
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.34375,
+      "learning_rate": 1.3105263157894736e-05,
+      "loss": 1.7347,
+      "step": 9170
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 7.15625,
+      "learning_rate": 1.2947368421052629e-05,
+      "loss": 1.0952,
+      "step": 9180
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 10.8125,
+      "learning_rate": 1.2789473684210526e-05,
+      "loss": 1.5358,
+      "step": 9190
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.9375,
+      "learning_rate": 1.2631578947368419e-05,
+      "loss": 1.5643,
+      "step": 9200
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 8.9375,
+      "learning_rate": 1.2473684210526316e-05,
+      "loss": 1.4119,
+      "step": 9210
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.78125,
+      "learning_rate": 1.2315789473684209e-05,
+      "loss": 1.5058,
+      "step": 9220
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.875,
+      "learning_rate": 1.2157894736842104e-05,
+      "loss": 1.294,
+      "step": 9230
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 11.1875,
+      "learning_rate": 1.1999999999999999e-05,
+      "loss": 1.2713,
+      "step": 9240
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.5,
+      "learning_rate": 1.1842105263157894e-05,
+      "loss": 1.149,
+      "step": 9250
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 9.3125,
+      "learning_rate": 1.1684210526315788e-05,
+      "loss": 1.1069,
+      "step": 9260
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.8125,
+      "learning_rate": 1.1526315789473683e-05,
+      "loss": 1.6268,
+      "step": 9270
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.8125,
+      "learning_rate": 1.1368421052631578e-05,
+      "loss": 1.4964,
+      "step": 9280
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.84375,
+      "learning_rate": 1.1210526315789473e-05,
+      "loss": 1.5281,
+      "step": 9290
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.78125,
+      "learning_rate": 1.1052631578947366e-05,
+      "loss": 1.31,
+      "step": 9300
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 4.125,
+      "learning_rate": 1.0894736842105263e-05,
+      "loss": 1.2768,
+      "step": 9310
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.125,
+      "learning_rate": 1.0736842105263156e-05,
+      "loss": 1.2628,
+      "step": 9320
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.90625,
+      "learning_rate": 1.0578947368421053e-05,
+      "loss": 1.1824,
+      "step": 9330
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.59375,
+      "learning_rate": 1.0421052631578946e-05,
+      "loss": 1.4915,
+      "step": 9340
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.609375,
+      "learning_rate": 1.0263157894736843e-05,
+      "loss": 1.418,
+      "step": 9350
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.09375,
+      "learning_rate": 1.0105263157894736e-05,
+      "loss": 1.2996,
+      "step": 9360
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.5,
+      "learning_rate": 9.947368421052632e-06,
+      "loss": 1.6655,
+      "step": 9370
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.59375,
+      "learning_rate": 9.789473684210526e-06,
+      "loss": 1.5435,
+      "step": 9380
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 9.75,
+      "learning_rate": 9.631578947368419e-06,
+      "loss": 0.9931,
+      "step": 9390
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.5,
+      "learning_rate": 9.473684210526315e-06,
+      "loss": 1.3517,
+      "step": 9400
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.3125,
+      "learning_rate": 9.31578947368421e-06,
+      "loss": 1.5011,
+      "step": 9410
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 4.59375,
+      "learning_rate": 9.157894736842104e-06,
+      "loss": 1.4879,
+      "step": 9420
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.375,
+      "learning_rate": 8.999999999999999e-06,
+      "loss": 1.2799,
+      "step": 9430
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.625,
+      "learning_rate": 8.842105263157893e-06,
+      "loss": 1.3966,
+      "step": 9440
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 7.78125,
+      "learning_rate": 8.684210526315788e-06,
+      "loss": 1.3313,
+      "step": 9450
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 7.84375,
+      "learning_rate": 8.526315789473683e-06,
+      "loss": 1.2297,
+      "step": 9460
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.375,
+      "learning_rate": 8.368421052631578e-06,
+      "loss": 1.5841,
+      "step": 9470
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 3.71875,
+      "learning_rate": 8.210526315789473e-06,
+      "loss": 1.2167,
+      "step": 9480
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 8.625,
+      "learning_rate": 8.052631578947368e-06,
+      "loss": 1.2425,
+      "step": 9490
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 4.21875,
+      "learning_rate": 7.894736842105261e-06,
+      "loss": 1.0646,
+      "step": 9500
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.9375,
+      "learning_rate": 7.736842105263156e-06,
+      "loss": 1.0081,
+      "step": 9510
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.8125,
+      "learning_rate": 7.578947368421052e-06,
+      "loss": 1.4349,
+      "step": 9520
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 3.78125,
+      "learning_rate": 7.421052631578947e-06,
+      "loss": 1.1173,
+      "step": 9530
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 8.8125,
+      "learning_rate": 7.263157894736842e-06,
+      "loss": 1.5393,
+      "step": 9540
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.9375,
+      "learning_rate": 7.105263157894737e-06,
+      "loss": 1.1636,
+      "step": 9550
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 8.8125,
+      "learning_rate": 6.9473684210526315e-06,
+      "loss": 1.5191,
+      "step": 9560
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.8125,
+      "learning_rate": 6.7894736842105264e-06,
+      "loss": 1.2565,
+      "step": 9570
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 3.625,
+      "learning_rate": 6.63157894736842e-06,
+      "loss": 1.3454,
+      "step": 9580
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.90625,
+      "learning_rate": 6.4736842105263146e-06,
+      "loss": 1.206,
+      "step": 9590
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.4375,
+      "learning_rate": 6.3157894736842095e-06,
+      "loss": 1.3835,
+      "step": 9600
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.03125,
+      "learning_rate": 6.157894736842104e-06,
+      "loss": 1.4376,
+      "step": 9610
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.3125,
+      "learning_rate": 5.999999999999999e-06,
+      "loss": 1.5081,
+      "step": 9620
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 7.96875,
+      "learning_rate": 5.842105263157894e-06,
+      "loss": 1.5345,
+      "step": 9630
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 7.125,
+      "learning_rate": 5.684210526315789e-06,
+      "loss": 1.4824,
+      "step": 9640
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.96875,
+      "learning_rate": 5.526315789473683e-06,
+      "loss": 1.2643,
+      "step": 9650
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 7.46875,
+      "learning_rate": 5.368421052631578e-06,
+      "loss": 1.5927,
+      "step": 9660
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 3.90625,
+      "learning_rate": 5.210526315789473e-06,
+      "loss": 1.4431,
+      "step": 9670
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.78125,
+      "learning_rate": 5.052631578947368e-06,
+      "loss": 1.652,
+      "step": 9680
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 9.5625,
+      "learning_rate": 4.894736842105263e-06,
+      "loss": 1.5427,
+      "step": 9690
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.5,
+      "learning_rate": 4.736842105263158e-06,
+      "loss": 1.3163,
+      "step": 9700
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.65625,
+      "learning_rate": 4.578947368421052e-06,
+      "loss": 1.3303,
+      "step": 9710
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.625,
+      "learning_rate": 4.421052631578947e-06,
+      "loss": 1.3957,
+      "step": 9720
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 10.0625,
+      "learning_rate": 4.263157894736842e-06,
+      "loss": 1.3943,
+      "step": 9730
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 4.25,
+      "learning_rate": 4.1052631578947365e-06,
+      "loss": 1.6132,
+      "step": 9740
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 7.0625,
+      "learning_rate": 3.947368421052631e-06,
+      "loss": 1.4703,
+      "step": 9750
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.78125,
+      "learning_rate": 3.789473684210526e-06,
+      "loss": 1.4131,
+      "step": 9760
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 9.1875,
+      "learning_rate": 3.631578947368421e-06,
+      "loss": 1.255,
+      "step": 9770
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 8.1875,
+      "learning_rate": 3.4736842105263158e-06,
+      "loss": 1.3646,
+      "step": 9780
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 14.25,
+      "learning_rate": 3.31578947368421e-06,
+      "loss": 1.3096,
+      "step": 9790
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 7.375,
+      "learning_rate": 3.1578947368421047e-06,
+      "loss": 1.0666,
+      "step": 9800
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 2.875,
+      "learning_rate": 2.9999999999999997e-06,
+      "loss": 1.1713,
+      "step": 9810
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 3.953125,
+      "learning_rate": 2.8421052631578946e-06,
+      "loss": 1.3558,
+      "step": 9820
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.9375,
+      "learning_rate": 2.684210526315789e-06,
+      "loss": 1.658,
+      "step": 9830
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 7.09375,
+      "learning_rate": 2.526315789473684e-06,
+      "loss": 1.4965,
+      "step": 9840
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.5625,
+      "learning_rate": 2.368421052631579e-06,
+      "loss": 1.6693,
+      "step": 9850
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 4.96875,
+      "learning_rate": 2.2105263157894734e-06,
+      "loss": 1.5437,
+      "step": 9860
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 4.84375,
+      "learning_rate": 2.0526315789473683e-06,
+      "loss": 1.2938,
+      "step": 9870
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.0625,
+      "learning_rate": 1.894736842105263e-06,
+      "loss": 1.3221,
+      "step": 9880
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.7368421052631579e-06,
+      "loss": 1.446,
+      "step": 9890
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 7.59375,
+      "learning_rate": 1.5789473684210524e-06,
+      "loss": 1.0271,
+      "step": 9900
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 10.5625,
+      "learning_rate": 1.4210526315789473e-06,
+      "loss": 1.2976,
+      "step": 9910
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 8.125,
+      "learning_rate": 1.263157894736842e-06,
+      "loss": 1.2977,
+      "step": 9920
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 5.78125,
+      "learning_rate": 1.1052631578947367e-06,
+      "loss": 1.5963,
+      "step": 9930
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 8.75,
+      "learning_rate": 9.473684210526315e-07,
+      "loss": 1.1931,
+      "step": 9940
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.9375,
+      "learning_rate": 7.894736842105262e-07,
+      "loss": 1.4246,
+      "step": 9950
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 7.03125,
+      "learning_rate": 6.31578947368421e-07,
+      "loss": 1.2961,
+      "step": 9960
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 7.5,
+      "learning_rate": 4.7368421052631574e-07,
+      "loss": 1.456,
+      "step": 9970
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 10.5,
+      "learning_rate": 3.157894736842105e-07,
+      "loss": 1.5236,
+      "step": 9980
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 4.8125,
+      "learning_rate": 1.5789473684210525e-07,
+      "loss": 1.3689,
+      "step": 9990
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 5.5625,
+      "learning_rate": 0.0,
+      "loss": 1.5764,
+      "step": 10000
+    },
+    {
+      "epoch": 0.5,
+      "eval_loss": 1.4314745664596558,
+      "eval_runtime": 30.6597,
+      "eval_samples_per_second": 32.616,
+      "eval_steps_per_second": 32.616,
+      "step": 10000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 2000,
+  "total_flos": 8.06961020928e+16,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}