diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5653 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.8689259458490226,
+  "eval_steps": 2000,
+  "global_step": 8000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 5.21875,
+      "learning_rate": 3.0000000000000004e-08,
+      "loss": 1.4288,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 9.25,
+      "learning_rate": 6.000000000000001e-08,
+      "loss": 1.5985,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 6.90625,
+      "learning_rate": 9e-08,
+      "loss": 1.3356,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.2000000000000002e-07,
+      "loss": 1.3789,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 6.28125,
+      "learning_rate": 1.5000000000000002e-07,
+      "loss": 1.362,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.8125,
+      "learning_rate": 1.8e-07,
+      "loss": 1.3439,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.65625,
+      "learning_rate": 2.1000000000000003e-07,
+      "loss": 1.3946,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 5.5,
+      "learning_rate": 2.4000000000000003e-07,
+      "loss": 1.3411,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.3125,
+      "learning_rate": 2.7e-07,
+      "loss": 1.3421,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.96875,
+      "learning_rate": 3.0000000000000004e-07,
+      "loss": 1.4355,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 6.59375,
+      "learning_rate": 3.3e-07,
+      "loss": 1.5107,
+      "step": 110
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.09375,
+      "learning_rate": 3.6e-07,
+      "loss": 1.2753,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 7.21875,
+      "learning_rate": 3.9e-07,
+      "loss": 1.3473,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 7.40625,
+      "learning_rate": 4.2000000000000006e-07,
+      "loss": 1.4292,
+      "step": 140
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 7.875,
+      "learning_rate": 4.5e-07,
+      "loss": 1.4616,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.8125,
+      "learning_rate": 4.800000000000001e-07,
+      "loss": 1.3935,
+      "step": 160
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.625,
+      "learning_rate": 5.100000000000001e-07,
+      "loss": 1.4381,
+      "step": 170
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 7.21875,
+      "learning_rate": 5.4e-07,
+      "loss": 1.3679,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 7.375,
+      "learning_rate": 5.7e-07,
+      "loss": 1.4046,
+      "step": 190
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 7.0,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 1.3214,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.59375,
+      "learning_rate": 6.3e-07,
+      "loss": 1.328,
+      "step": 210
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.25,
+      "learning_rate": 6.6e-07,
+      "loss": 1.3994,
+      "step": 220
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 7.375,
+      "learning_rate": 6.900000000000001e-07,
+      "loss": 1.3372,
+      "step": 230
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.65625,
+      "learning_rate": 7.2e-07,
+      "loss": 1.4327,
+      "step": 240
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.46875,
+      "learning_rate": 7.5e-07,
+      "loss": 1.3006,
+      "step": 250
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 8.5625,
+      "learning_rate": 7.8e-07,
+      "loss": 1.4955,
+      "step": 260
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.84375,
+      "learning_rate": 8.100000000000001e-07,
+      "loss": 1.2202,
+      "step": 270
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.09375,
+      "learning_rate": 8.400000000000001e-07,
+      "loss": 1.3608,
+      "step": 280
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 6.75,
+      "learning_rate": 8.699999999999999e-07,
+      "loss": 1.3799,
+      "step": 290
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.09375,
+      "learning_rate": 9e-07,
+      "loss": 1.3552,
+      "step": 300
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.90625,
+      "learning_rate": 9.3e-07,
+      "loss": 1.4482,
+      "step": 310
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.53125,
+      "learning_rate": 9.600000000000001e-07,
+      "loss": 1.4208,
+      "step": 320
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.09375,
+      "learning_rate": 9.9e-07,
+      "loss": 1.373,
+      "step": 330
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.40625,
+      "learning_rate": 1.0200000000000002e-06,
+      "loss": 1.3638,
+      "step": 340
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.875,
+      "learning_rate": 1.05e-06,
+      "loss": 1.4984,
+      "step": 350
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.25,
+      "learning_rate": 1.08e-06,
+      "loss": 1.4193,
+      "step": 360
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 7.5625,
+      "learning_rate": 1.11e-06,
+      "loss": 1.4644,
+      "step": 370
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.625,
+      "learning_rate": 1.14e-06,
+      "loss": 1.2974,
+      "step": 380
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 7.0,
+      "learning_rate": 1.17e-06,
+      "loss": 1.4013,
+      "step": 390
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 1.5707,
+      "step": 400
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.5,
+      "learning_rate": 1.2299999999999999e-06,
+      "loss": 1.4111,
+      "step": 410
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.71875,
+      "learning_rate": 1.26e-06,
+      "loss": 1.3019,
+      "step": 420
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 7.625,
+      "learning_rate": 1.29e-06,
+      "loss": 1.4128,
+      "step": 430
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.15625,
+      "learning_rate": 1.32e-06,
+      "loss": 1.2438,
+      "step": 440
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.25,
+      "learning_rate": 1.35e-06,
+      "loss": 1.3703,
+      "step": 450
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.15625,
+      "learning_rate": 1.3800000000000001e-06,
+      "loss": 1.3877,
+      "step": 460
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.41e-06,
+      "loss": 1.3418,
+      "step": 470
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 7.09375,
+      "learning_rate": 1.44e-06,
+      "loss": 1.3734,
+      "step": 480
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.34375,
+      "learning_rate": 1.4700000000000001e-06,
+      "loss": 1.3055,
+      "step": 490
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.65625,
+      "learning_rate": 1.5e-06,
+      "loss": 1.4607,
+      "step": 500
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.90625,
+      "learning_rate": 1.498421052631579e-06,
+      "loss": 1.3326,
+      "step": 510
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.75,
+      "learning_rate": 1.4968421052631579e-06,
+      "loss": 1.3793,
+      "step": 520
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 7.0625,
+      "learning_rate": 1.4952631578947368e-06,
+      "loss": 1.3474,
+      "step": 530
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.15625,
+      "learning_rate": 1.4936842105263159e-06,
+      "loss": 1.3504,
+      "step": 540
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.953125,
+      "learning_rate": 1.4921052631578948e-06,
+      "loss": 1.29,
+      "step": 550
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.625,
+      "learning_rate": 1.4905263157894737e-06,
+      "loss": 1.3552,
+      "step": 560
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.28125,
+      "learning_rate": 1.4889473684210526e-06,
+      "loss": 1.2646,
+      "step": 570
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.15625,
+      "learning_rate": 1.4873684210526315e-06,
+      "loss": 1.2683,
+      "step": 580
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 7.03125,
+      "learning_rate": 1.4857894736842106e-06,
+      "loss": 1.3948,
+      "step": 590
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 7.0,
+      "learning_rate": 1.4842105263157895e-06,
+      "loss": 1.2919,
+      "step": 600
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.15625,
+      "learning_rate": 1.4826315789473684e-06,
+      "loss": 1.3664,
+      "step": 610
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.75,
+      "learning_rate": 1.4810526315789473e-06,
+      "loss": 1.3198,
+      "step": 620
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.0625,
+      "learning_rate": 1.4794736842105265e-06,
+      "loss": 1.3457,
+      "step": 630
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 6.28125,
+      "learning_rate": 1.4778947368421054e-06,
+      "loss": 1.2258,
+      "step": 640
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.78125,
+      "learning_rate": 1.4763157894736843e-06,
+      "loss": 1.2654,
+      "step": 650
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.5,
+      "learning_rate": 1.4747368421052632e-06,
+      "loss": 1.2623,
+      "step": 660
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.8125,
+      "learning_rate": 1.473157894736842e-06,
+      "loss": 1.3726,
+      "step": 670
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.4715789473684212e-06,
+      "loss": 1.2427,
+      "step": 680
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.4700000000000001e-06,
+      "loss": 1.2959,
+      "step": 690
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.09375,
+      "learning_rate": 1.468421052631579e-06,
+      "loss": 1.2209,
+      "step": 700
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.84375,
+      "learning_rate": 1.466842105263158e-06,
+      "loss": 1.297,
+      "step": 710
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.625,
+      "learning_rate": 1.4652631578947368e-06,
+      "loss": 1.3002,
+      "step": 720
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 7.96875,
+      "learning_rate": 1.463684210526316e-06,
+      "loss": 1.3578,
+      "step": 730
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.4375,
+      "learning_rate": 1.4621052631578946e-06,
+      "loss": 1.3381,
+      "step": 740
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.25,
+      "learning_rate": 1.4605263157894738e-06,
+      "loss": 1.2755,
+      "step": 750
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.28125,
+      "learning_rate": 1.4589473684210527e-06,
+      "loss": 1.3443,
+      "step": 760
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.4573684210526316e-06,
+      "loss": 1.3165,
+      "step": 770
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.125,
+      "learning_rate": 1.4557894736842105e-06,
+      "loss": 1.3626,
+      "step": 780
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.625,
+      "learning_rate": 1.4542105263157894e-06,
+      "loss": 1.3729,
+      "step": 790
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.125,
+      "learning_rate": 1.4526315789473685e-06,
+      "loss": 1.3209,
+      "step": 800
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.96875,
+      "learning_rate": 1.4510526315789474e-06,
+      "loss": 1.2652,
+      "step": 810
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 3.796875,
+      "learning_rate": 1.4494736842105263e-06,
+      "loss": 1.2958,
+      "step": 820
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.84375,
+      "learning_rate": 1.4478947368421052e-06,
+      "loss": 1.271,
+      "step": 830
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 3.5,
+      "learning_rate": 1.4463157894736844e-06,
+      "loss": 1.2985,
+      "step": 840
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.4447368421052633e-06,
+      "loss": 1.3285,
+      "step": 850
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.3125,
+      "learning_rate": 1.4431578947368422e-06,
+      "loss": 1.2987,
+      "step": 860
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 4.8125,
+      "learning_rate": 1.441578947368421e-06,
+      "loss": 1.3378,
+      "step": 870
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.875,
+      "learning_rate": 1.44e-06,
+      "loss": 1.2942,
+      "step": 880
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4.5625,
+      "learning_rate": 1.438421052631579e-06,
+      "loss": 1.259,
+      "step": 890
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4.34375,
+      "learning_rate": 1.436842105263158e-06,
+      "loss": 1.2991,
+      "step": 900
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 4.9375,
+      "learning_rate": 1.435263157894737e-06,
+      "loss": 1.2565,
+      "step": 910
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.4336842105263158e-06,
+      "loss": 1.1737,
+      "step": 920
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 4.875,
+      "learning_rate": 1.4321052631578947e-06,
+      "loss": 1.2779,
+      "step": 930
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.375,
+      "learning_rate": 1.4305263157894738e-06,
+      "loss": 1.3104,
+      "step": 940
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.53125,
+      "learning_rate": 1.4289473684210525e-06,
+      "loss": 1.2818,
+      "step": 950
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.4273684210526317e-06,
+      "loss": 1.3147,
+      "step": 960
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.59375,
+      "learning_rate": 1.4257894736842106e-06,
+      "loss": 1.2595,
+      "step": 970
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.125,
+      "learning_rate": 1.4242105263157895e-06,
+      "loss": 1.3695,
+      "step": 980
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.96875,
+      "learning_rate": 1.4226315789473684e-06,
+      "loss": 1.1634,
+      "step": 990
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 4.65625,
+      "learning_rate": 1.4210526315789473e-06,
+      "loss": 1.2587,
+      "step": 1000
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 7.625,
+      "learning_rate": 1.4194736842105264e-06,
+      "loss": 1.2994,
+      "step": 1010
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 4.9375,
+      "learning_rate": 1.4178947368421053e-06,
+      "loss": 1.3351,
+      "step": 1020
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.21875,
+      "learning_rate": 1.4163157894736842e-06,
+      "loss": 1.2404,
+      "step": 1030
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.0,
+      "learning_rate": 1.4147368421052631e-06,
+      "loss": 1.3211,
+      "step": 1040
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.4131578947368422e-06,
+      "loss": 1.2811,
+      "step": 1050
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.4115789473684211e-06,
+      "loss": 1.2131,
+      "step": 1060
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.41e-06,
+      "loss": 1.1713,
+      "step": 1070
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.408421052631579e-06,
+      "loss": 1.2058,
+      "step": 1080
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 3.6875,
+      "learning_rate": 1.4068421052631579e-06,
+      "loss": 1.3375,
+      "step": 1090
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.21875,
+      "learning_rate": 1.405263157894737e-06,
+      "loss": 1.3032,
+      "step": 1100
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 3.359375,
+      "learning_rate": 1.4036842105263159e-06,
+      "loss": 1.3644,
+      "step": 1110
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.65625,
+      "learning_rate": 1.4021052631578948e-06,
+      "loss": 1.2786,
+      "step": 1120
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.8125,
+      "learning_rate": 1.4005263157894737e-06,
+      "loss": 1.2067,
+      "step": 1130
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 3.90625,
+      "learning_rate": 1.3989473684210526e-06,
+      "loss": 1.3127,
+      "step": 1140
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 4.0,
+      "learning_rate": 1.3973684210526317e-06,
+      "loss": 1.2442,
+      "step": 1150
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.875,
+      "learning_rate": 1.3957894736842104e-06,
+      "loss": 1.2307,
+      "step": 1160
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.3942105263157895e-06,
+      "loss": 1.2828,
+      "step": 1170
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.3926315789473685e-06,
+      "loss": 1.2676,
+      "step": 1180
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.3910526315789476e-06,
+      "loss": 1.1911,
+      "step": 1190
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.3894736842105263e-06,
+      "loss": 1.3177,
+      "step": 1200
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.40625,
+      "learning_rate": 1.3878947368421052e-06,
+      "loss": 1.2891,
+      "step": 1210
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.90625,
+      "learning_rate": 1.3863157894736843e-06,
+      "loss": 1.303,
+      "step": 1220
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.125,
+      "learning_rate": 1.3847368421052632e-06,
+      "loss": 1.2586,
+      "step": 1230
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 6.8125,
+      "learning_rate": 1.383157894736842e-06,
+      "loss": 1.3141,
+      "step": 1240
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.381578947368421e-06,
+      "loss": 1.2174,
+      "step": 1250
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 4.625,
+      "learning_rate": 1.3800000000000001e-06,
+      "loss": 1.2609,
+      "step": 1260
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.75,
+      "learning_rate": 1.378421052631579e-06,
+      "loss": 1.1323,
+      "step": 1270
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 4.625,
+      "learning_rate": 1.376842105263158e-06,
+      "loss": 1.2328,
+      "step": 1280
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.3752631578947368e-06,
+      "loss": 1.3087,
+      "step": 1290
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 4.90625,
+      "learning_rate": 1.3736842105263158e-06,
+      "loss": 1.2428,
+      "step": 1300
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.46875,
+      "learning_rate": 1.3721052631578949e-06,
+      "loss": 1.235,
+      "step": 1310
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.46875,
+      "learning_rate": 1.3705263157894738e-06,
+      "loss": 1.2108,
+      "step": 1320
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.875,
+      "learning_rate": 1.3689473684210527e-06,
+      "loss": 1.1704,
+      "step": 1330
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 4.625,
+      "learning_rate": 1.3673684210526316e-06,
+      "loss": 1.3411,
+      "step": 1340
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.3657894736842107e-06,
+      "loss": 1.239,
+      "step": 1350
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.3642105263157896e-06,
+      "loss": 1.1661,
+      "step": 1360
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.3626315789473683e-06,
+      "loss": 1.338,
+      "step": 1370
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.3610526315789474e-06,
+      "loss": 1.2926,
+      "step": 1380
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 2.875,
+      "learning_rate": 1.3594736842105263e-06,
+      "loss": 1.2866,
+      "step": 1390
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 4.96875,
+      "learning_rate": 1.3578947368421055e-06,
+      "loss": 1.3333,
+      "step": 1400
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.53125,
+      "learning_rate": 1.3563157894736842e-06,
+      "loss": 1.2472,
+      "step": 1410
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 3.390625,
+      "learning_rate": 1.3547368421052633e-06,
+      "loss": 1.1796,
+      "step": 1420
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.125,
+      "learning_rate": 1.3531578947368422e-06,
+      "loss": 1.321,
+      "step": 1430
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 4.5625,
+      "learning_rate": 1.351578947368421e-06,
+      "loss": 1.2529,
+      "step": 1440
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 5.40625,
+      "learning_rate": 1.35e-06,
+      "loss": 1.2682,
+      "step": 1450
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 4.84375,
+      "learning_rate": 1.348421052631579e-06,
+      "loss": 1.358,
+      "step": 1460
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 5.21875,
+      "learning_rate": 1.346842105263158e-06,
+      "loss": 1.3318,
+      "step": 1470
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 5.65625,
+      "learning_rate": 1.345263157894737e-06,
+      "loss": 1.2531,
+      "step": 1480
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 5.46875,
+      "learning_rate": 1.3436842105263158e-06,
+      "loss": 1.2796,
+      "step": 1490
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 3.203125,
+      "learning_rate": 1.3421052631578947e-06,
+      "loss": 1.2315,
+      "step": 1500
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 4.71875,
+      "learning_rate": 1.3405263157894736e-06,
+      "loss": 1.2501,
+      "step": 1510
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 5.03125,
+      "learning_rate": 1.3389473684210528e-06,
+      "loss": 1.3444,
+      "step": 1520
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 6.3125,
+      "learning_rate": 1.3373684210526317e-06,
+      "loss": 1.2065,
+      "step": 1530
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 4.59375,
+      "learning_rate": 1.3357894736842106e-06,
+      "loss": 1.3091,
+      "step": 1540
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 4.625,
+      "learning_rate": 1.3342105263157895e-06,
+      "loss": 1.1932,
+      "step": 1550
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 3.859375,
+      "learning_rate": 1.3326315789473686e-06,
+      "loss": 1.1837,
+      "step": 1560
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 5.40625,
+      "learning_rate": 1.3310526315789475e-06,
+      "loss": 1.1756,
+      "step": 1570
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 4.65625,
+      "learning_rate": 1.3294736842105262e-06,
+      "loss": 1.3163,
+      "step": 1580
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 3.921875,
+      "learning_rate": 1.3278947368421053e-06,
+      "loss": 1.1937,
+      "step": 1590
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.3263157894736842e-06,
+      "loss": 1.3058,
+      "step": 1600
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 3.609375,
+      "learning_rate": 1.3247368421052633e-06,
+      "loss": 1.2718,
+      "step": 1610
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 4.875,
+      "learning_rate": 1.323157894736842e-06,
+      "loss": 1.2027,
+      "step": 1620
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 6.15625,
+      "learning_rate": 1.3215789473684212e-06,
+      "loss": 1.2548,
+      "step": 1630
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 5.875,
+      "learning_rate": 1.32e-06,
+      "loss": 1.3534,
+      "step": 1640
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 4.125,
+      "learning_rate": 1.318421052631579e-06,
+      "loss": 1.2223,
+      "step": 1650
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 4.875,
+      "learning_rate": 1.3168421052631579e-06,
+      "loss": 1.2684,
+      "step": 1660
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 5.5625,
+      "learning_rate": 1.3152631578947368e-06,
+      "loss": 1.2611,
+      "step": 1670
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 4.71875,
+      "learning_rate": 1.313684210526316e-06,
+      "loss": 1.3116,
+      "step": 1680
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 5.5,
+      "learning_rate": 1.3121052631578948e-06,
+      "loss": 1.303,
+      "step": 1690
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.3105263157894737e-06,
+      "loss": 1.2399,
+      "step": 1700
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.3089473684210526e-06,
+      "loss": 1.2954,
+      "step": 1710
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 5.125,
+      "learning_rate": 1.3073684210526315e-06,
+      "loss": 1.2689,
+      "step": 1720
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 5.625,
+      "learning_rate": 1.3057894736842107e-06,
+      "loss": 1.3605,
+      "step": 1730
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 4.125,
+      "learning_rate": 1.3042105263157896e-06,
+      "loss": 1.2873,
+      "step": 1740
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 5.5,
+      "learning_rate": 1.3026315789473685e-06,
+      "loss": 1.2686,
+      "step": 1750
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 5.5625,
+      "learning_rate": 1.3010526315789474e-06,
+      "loss": 1.244,
+      "step": 1760
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.2994736842105265e-06,
+      "loss": 1.2044,
+      "step": 1770
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 3.5625,
+      "learning_rate": 1.2978947368421054e-06,
+      "loss": 1.2026,
+      "step": 1780
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.296315789473684e-06,
+      "loss": 1.306,
+      "step": 1790
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 4.125,
+      "learning_rate": 1.2947368421052632e-06,
+      "loss": 1.2968,
+      "step": 1800
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.75,
+      "learning_rate": 1.2931578947368421e-06,
+      "loss": 1.2308,
+      "step": 1810
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 5.9375,
+      "learning_rate": 1.2915789473684212e-06,
+      "loss": 1.3022,
+      "step": 1820
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 6.34375,
+      "learning_rate": 1.29e-06,
+      "loss": 1.2298,
+      "step": 1830
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 4.8125,
+      "learning_rate": 1.288421052631579e-06,
+      "loss": 1.1421,
+      "step": 1840
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 5.0,
+      "learning_rate": 1.286842105263158e-06,
+      "loss": 1.2018,
+      "step": 1850
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 5.84375,
+      "learning_rate": 1.2852631578947369e-06,
+      "loss": 1.2329,
+      "step": 1860
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 5.375,
+      "learning_rate": 1.2836842105263158e-06,
+      "loss": 1.2509,
+      "step": 1870
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 4.34375,
+      "learning_rate": 1.2821052631578947e-06,
+      "loss": 1.2556,
+      "step": 1880
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 4.90625,
+      "learning_rate": 1.2805263157894738e-06,
+      "loss": 1.3504,
+      "step": 1890
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.2789473684210527e-06,
+      "loss": 1.1525,
+      "step": 1900
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 5.71875,
+      "learning_rate": 1.2773684210526316e-06,
+      "loss": 1.2227,
+      "step": 1910
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 3.65625,
+      "learning_rate": 1.2757894736842105e-06,
+      "loss": 1.2962,
+      "step": 1920
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 3.765625,
+      "learning_rate": 1.2742105263157894e-06,
+      "loss": 1.1717,
+      "step": 1930
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 6.65625,
+      "learning_rate": 1.2726315789473685e-06,
+      "loss": 1.2543,
+      "step": 1940
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 3.984375,
+      "learning_rate": 1.2710526315789474e-06,
+      "loss": 1.2092,
+      "step": 1950
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 4.9375,
+      "learning_rate": 1.2694736842105264e-06,
+      "loss": 1.1855,
+      "step": 1960
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.2678947368421053e-06,
+      "loss": 1.2698,
+      "step": 1970
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 5.625,
+      "learning_rate": 1.2663157894736844e-06,
+      "loss": 1.2615,
+      "step": 1980
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 4.59375,
+      "learning_rate": 1.2647368421052633e-06,
+      "loss": 1.2156,
+      "step": 1990
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 6.875,
+      "learning_rate": 1.263157894736842e-06,
+      "loss": 1.2328,
+      "step": 2000
+    },
+    {
+      "epoch": 0.72,
+      "eval_loss": 1.3270684480667114,
+      "eval_runtime": 64.5706,
+      "eval_samples_per_second": 15.487,
+      "eval_steps_per_second": 15.487,
+      "step": 2000
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.261578947368421e-06,
+      "loss": 1.1687,
+      "step": 2010
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 5.15625,
+      "learning_rate": 1.26e-06,
+      "loss": 1.2967,
+      "step": 2020
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 5.21875,
+      "learning_rate": 1.2584210526315791e-06,
+      "loss": 1.2625,
+      "step": 2030
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 4.5625,
+      "learning_rate": 1.2568421052631578e-06,
+      "loss": 1.2231,
+      "step": 2040
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 6.21875,
+      "learning_rate": 1.255263157894737e-06,
+      "loss": 1.2561,
+      "step": 2050
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.2536842105263158e-06,
+      "loss": 1.1732,
+      "step": 2060
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 3.21875,
+      "learning_rate": 1.2521052631578948e-06,
+      "loss": 1.1991,
+      "step": 2070
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.2505263157894737e-06,
+      "loss": 1.2405,
+      "step": 2080
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 3.921875,
+      "learning_rate": 1.2489473684210526e-06,
+      "loss": 1.2658,
+      "step": 2090
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 5.46875,
+      "learning_rate": 1.2473684210526317e-06,
+      "loss": 1.214,
+      "step": 2100
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 4.375,
+      "learning_rate": 1.2457894736842106e-06,
+      "loss": 1.2522,
+      "step": 2110
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 3.140625,
+      "learning_rate": 1.2442105263157895e-06,
+      "loss": 1.2873,
+      "step": 2120
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 4.71875,
+      "learning_rate": 1.2426315789473684e-06,
+      "loss": 1.2513,
+      "step": 2130
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.2410526315789473e-06,
+      "loss": 1.2378,
+      "step": 2140
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 3.859375,
+      "learning_rate": 1.2394736842105264e-06,
+      "loss": 1.2217,
+      "step": 2150
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 5.65625,
+      "learning_rate": 1.2378947368421053e-06,
+      "loss": 1.1462,
+      "step": 2160
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.2363157894736842e-06,
+      "loss": 1.2306,
+      "step": 2170
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.2347368421052631e-06,
+      "loss": 1.2744,
+      "step": 2180
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.2331578947368423e-06,
+      "loss": 1.1246,
+      "step": 2190
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 5.40625,
+      "learning_rate": 1.2315789473684212e-06,
+      "loss": 1.2965,
+      "step": 2200
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 5.875,
+      "learning_rate": 1.2299999999999999e-06,
+      "loss": 1.1981,
+      "step": 2210
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 4.78125,
+      "learning_rate": 1.228421052631579e-06,
+      "loss": 1.1496,
+      "step": 2220
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 5.03125,
+      "learning_rate": 1.226842105263158e-06,
+      "loss": 1.2732,
+      "step": 2230
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 3.4375,
+      "learning_rate": 1.225263157894737e-06,
+      "loss": 1.2618,
+      "step": 2240
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 4.125,
+      "learning_rate": 1.2236842105263157e-06,
+      "loss": 1.1922,
+      "step": 2250
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 4.9375,
+      "learning_rate": 1.2221052631578948e-06,
+      "loss": 1.1998,
+      "step": 2260
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 3.953125,
+      "learning_rate": 1.2205263157894737e-06,
+      "loss": 1.237,
+      "step": 2270
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 5.84375,
+      "learning_rate": 1.2189473684210526e-06,
+      "loss": 1.2583,
+      "step": 2280
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.15625,
+      "learning_rate": 1.2173684210526315e-06,
+      "loss": 1.2544,
+      "step": 2290
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.2157894736842105e-06,
+      "loss": 1.3154,
+      "step": 2300
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 4.90625,
+      "learning_rate": 1.2142105263157896e-06,
+      "loss": 1.248,
+      "step": 2310
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 6.5,
+      "learning_rate": 1.2126315789473685e-06,
+      "loss": 1.2039,
+      "step": 2320
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 3.796875,
+      "learning_rate": 1.2110526315789474e-06,
+      "loss": 1.328,
+      "step": 2330
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 6.0625,
+      "learning_rate": 1.2094736842105263e-06,
+      "loss": 1.2407,
+      "step": 2340
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 5.5625,
+      "learning_rate": 1.2078947368421052e-06,
+      "loss": 1.1751,
+      "step": 2350
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 3.828125,
+      "learning_rate": 1.2063157894736843e-06,
+      "loss": 1.2525,
+      "step": 2360
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 4.65625,
+      "learning_rate": 1.2047368421052632e-06,
+      "loss": 1.2554,
+      "step": 2370
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 6.03125,
+      "learning_rate": 1.2031578947368421e-06,
+      "loss": 1.2803,
+      "step": 2380
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 3.59375,
+      "learning_rate": 1.201578947368421e-06,
+      "loss": 1.1723,
+      "step": 2390
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 6.65625,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 1.2068,
+      "step": 2400
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 5.5625,
+      "learning_rate": 1.198421052631579e-06,
+      "loss": 1.2599,
+      "step": 2410
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.1968421052631578e-06,
+      "loss": 1.327,
+      "step": 2420
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.1952631578947369e-06,
+      "loss": 1.1712,
+      "step": 2430
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 4.96875,
+      "learning_rate": 1.1936842105263158e-06,
+      "loss": 1.303,
+      "step": 2440
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 4.5,
+      "learning_rate": 1.192105263157895e-06,
+      "loss": 1.2574,
+      "step": 2450
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.90625,
+      "learning_rate": 1.1905263157894736e-06,
+      "loss": 1.3576,
+      "step": 2460
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 6.21875,
+      "learning_rate": 1.1889473684210527e-06,
+      "loss": 1.2322,
+      "step": 2470
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 4.78125,
+      "learning_rate": 1.1873684210526316e-06,
+      "loss": 1.2198,
+      "step": 2480
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 4.9375,
+      "learning_rate": 1.1857894736842105e-06,
+      "loss": 1.2068,
+      "step": 2490
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.1842105263157894e-06,
+      "loss": 1.2278,
+      "step": 2500
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.1826315789473683e-06,
+      "loss": 1.2847,
+      "step": 2510
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 5.34375,
+      "learning_rate": 1.1810526315789475e-06,
+      "loss": 1.1646,
+      "step": 2520
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 4.09375,
+      "learning_rate": 1.1794736842105264e-06,
+      "loss": 1.201,
+      "step": 2530
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 6.34375,
+      "learning_rate": 1.1778947368421053e-06,
+      "loss": 1.1445,
+      "step": 2540
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.1763157894736842e-06,
+      "loss": 1.2555,
+      "step": 2550
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 4.125,
+      "learning_rate": 1.174736842105263e-06,
+      "loss": 1.1944,
+      "step": 2560
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 4.65625,
+      "learning_rate": 1.1731578947368422e-06,
+      "loss": 1.1288,
+      "step": 2570
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 4.9375,
+      "learning_rate": 1.1715789473684211e-06,
+      "loss": 1.2126,
+      "step": 2580
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.17e-06,
+      "loss": 1.3395,
+      "step": 2590
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 4.875,
+      "learning_rate": 1.168421052631579e-06,
+      "loss": 1.1955,
+      "step": 2600
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.166842105263158e-06,
+      "loss": 1.1931,
+      "step": 2610
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 4.96875,
+      "learning_rate": 1.165263157894737e-06,
+      "loss": 1.2561,
+      "step": 2620
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.1636842105263156e-06,
+      "loss": 1.2339,
+      "step": 2630
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 5.53125,
+      "learning_rate": 1.1621052631578948e-06,
+      "loss": 1.1738,
+      "step": 2640
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.1605263157894737e-06,
+      "loss": 1.1535,
+      "step": 2650
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 3.703125,
+      "learning_rate": 1.1589473684210528e-06,
+      "loss": 1.2578,
+      "step": 2660
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.1573684210526315e-06,
+      "loss": 1.229,
+      "step": 2670
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 3.859375,
+      "learning_rate": 1.1557894736842106e-06,
+      "loss": 1.2044,
+      "step": 2680
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 6.0,
+      "learning_rate": 1.1542105263157895e-06,
+      "loss": 1.1518,
+      "step": 2690
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 4.625,
+      "learning_rate": 1.1526315789473684e-06,
+      "loss": 1.1951,
+      "step": 2700
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 4.375,
+      "learning_rate": 1.1510526315789473e-06,
+      "loss": 1.2361,
+      "step": 2710
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 5.125,
+      "learning_rate": 1.1494736842105262e-06,
+      "loss": 1.2218,
+      "step": 2720
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.1478947368421054e-06,
+      "loss": 1.1337,
+      "step": 2730
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 5.21875,
+      "learning_rate": 1.1463157894736843e-06,
+      "loss": 1.2582,
+      "step": 2740
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 3.34375,
+      "learning_rate": 1.1447368421052632e-06,
+      "loss": 1.2717,
+      "step": 2750
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.143157894736842e-06,
+      "loss": 1.2379,
+      "step": 2760
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 3.8125,
+      "learning_rate": 1.141578947368421e-06,
+      "loss": 1.1829,
+      "step": 2770
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 5.625,
+      "learning_rate": 1.14e-06,
+      "loss": 1.3186,
+      "step": 2780
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 5.96875,
+      "learning_rate": 1.138421052631579e-06,
+      "loss": 1.1503,
+      "step": 2790
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.136842105263158e-06,
+      "loss": 1.347,
+      "step": 2800
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 5.53125,
+      "learning_rate": 1.1352631578947368e-06,
+      "loss": 1.187,
+      "step": 2810
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 6.09375,
+      "learning_rate": 1.133684210526316e-06,
+      "loss": 1.2084,
+      "step": 2820
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 5.1875,
+      "learning_rate": 1.1321052631578948e-06,
+      "loss": 1.1274,
+      "step": 2830
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 4.375,
+      "learning_rate": 1.1305263157894735e-06,
+      "loss": 1.2603,
+      "step": 2840
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 3.5,
+      "learning_rate": 1.1289473684210527e-06,
+      "loss": 1.2914,
+      "step": 2850
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.1273684210526316e-06,
+      "loss": 1.2076,
+      "step": 2860
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 5.96875,
+      "learning_rate": 1.1257894736842107e-06,
+      "loss": 1.3274,
+      "step": 2870
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 4.40625,
+      "learning_rate": 1.1242105263157894e-06,
+      "loss": 1.3251,
+      "step": 2880
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 5.59375,
+      "learning_rate": 1.1226315789473685e-06,
+      "loss": 1.16,
+      "step": 2890
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 5.71875,
+      "learning_rate": 1.1210526315789474e-06,
+      "loss": 1.2715,
+      "step": 2900
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.1194736842105265e-06,
+      "loss": 1.2301,
+      "step": 2910
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 4.46875,
+      "learning_rate": 1.1178947368421052e-06,
+      "loss": 1.2531,
+      "step": 2920
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 7.25,
+      "learning_rate": 1.1163157894736841e-06,
+      "loss": 1.2059,
+      "step": 2930
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.1147368421052632e-06,
+      "loss": 1.3348,
+      "step": 2940
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 5.25,
+      "learning_rate": 1.1131578947368421e-06,
+      "loss": 1.2387,
+      "step": 2950
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.111578947368421e-06,
+      "loss": 1.2128,
+      "step": 2960
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.11e-06,
+      "loss": 1.1845,
+      "step": 2970
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 4.75,
+      "learning_rate": 1.108421052631579e-06,
+      "loss": 1.2829,
+      "step": 2980
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 5.65625,
+      "learning_rate": 1.106842105263158e-06,
+      "loss": 1.185,
+      "step": 2990
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 4.96875,
+      "learning_rate": 1.1052631578947369e-06,
+      "loss": 1.3094,
+      "step": 3000
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.1036842105263158e-06,
+      "loss": 1.3127,
+      "step": 3010
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 5.03125,
+      "learning_rate": 1.1021052631578947e-06,
+      "loss": 1.3655,
+      "step": 3020
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 6.28125,
+      "learning_rate": 1.1005263157894738e-06,
+      "loss": 1.272,
+      "step": 3030
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 4.5625,
+      "learning_rate": 1.0989473684210527e-06,
+      "loss": 1.2848,
+      "step": 3040
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 5.4375,
+      "learning_rate": 1.0973684210526316e-06,
+      "loss": 1.2372,
+      "step": 3050
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.0957894736842105e-06,
+      "loss": 1.1784,
+      "step": 3060
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 6.875,
+      "learning_rate": 1.0942105263157895e-06,
+      "loss": 1.245,
+      "step": 3070
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 3.484375,
+      "learning_rate": 1.0926315789473686e-06,
+      "loss": 1.2199,
+      "step": 3080
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 4.34375,
+      "learning_rate": 1.0910526315789473e-06,
+      "loss": 1.2262,
+      "step": 3090
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 4.5625,
+      "learning_rate": 1.0894736842105264e-06,
+      "loss": 1.1917,
+      "step": 3100
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 3.984375,
+      "learning_rate": 1.0878947368421053e-06,
+      "loss": 1.2455,
+      "step": 3110
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 5.21875,
+      "learning_rate": 1.0863157894736844e-06,
+      "loss": 1.1966,
+      "step": 3120
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 6.0,
+      "learning_rate": 1.084736842105263e-06,
+      "loss": 1.396,
+      "step": 3130
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.083157894736842e-06,
+      "loss": 1.2836,
+      "step": 3140
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 4.75,
+      "learning_rate": 1.0815789473684211e-06,
+      "loss": 1.2193,
+      "step": 3150
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 6.3125,
+      "learning_rate": 1.08e-06,
+      "loss": 1.2754,
+      "step": 3160
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 5.25,
+      "learning_rate": 1.078421052631579e-06,
+      "loss": 1.3408,
+      "step": 3170
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 5.53125,
+      "learning_rate": 1.0768421052631578e-06,
+      "loss": 1.2585,
+      "step": 3180
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 4.75,
+      "learning_rate": 1.075263157894737e-06,
+      "loss": 1.2207,
+      "step": 3190
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 5.25,
+      "learning_rate": 1.0736842105263159e-06,
+      "loss": 1.3081,
+      "step": 3200
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 5.9375,
+      "learning_rate": 1.0721052631578948e-06,
+      "loss": 1.2992,
+      "step": 3210
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 4.125,
+      "learning_rate": 1.0705263157894737e-06,
+      "loss": 1.2481,
+      "step": 3220
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 6.21875,
+      "learning_rate": 1.0689473684210526e-06,
+      "loss": 1.2883,
+      "step": 3230
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 6.25,
+      "learning_rate": 1.0673684210526317e-06,
+      "loss": 1.2565,
+      "step": 3240
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.0657894736842106e-06,
+      "loss": 1.1712,
+      "step": 3250
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 6.03125,
+      "learning_rate": 1.0642105263157895e-06,
+      "loss": 1.1657,
+      "step": 3260
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 5.90625,
+      "learning_rate": 1.0626315789473684e-06,
+      "loss": 1.3871,
+      "step": 3270
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 5.0,
+      "learning_rate": 1.0610526315789473e-06,
+      "loss": 1.2952,
+      "step": 3280
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 5.59375,
+      "learning_rate": 1.0594736842105265e-06,
+      "loss": 1.2694,
+      "step": 3290
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 6.375,
+      "learning_rate": 1.0578947368421052e-06,
+      "loss": 1.2199,
+      "step": 3300
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 5.65625,
+      "learning_rate": 1.0563157894736843e-06,
+      "loss": 1.3025,
+      "step": 3310
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 5.28125,
+      "learning_rate": 1.0547368421052632e-06,
+      "loss": 1.3357,
+      "step": 3320
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 3.65625,
+      "learning_rate": 1.0531578947368423e-06,
+      "loss": 1.2073,
+      "step": 3330
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 6.125,
+      "learning_rate": 1.051578947368421e-06,
+      "loss": 1.2984,
+      "step": 3340
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 5.0,
+      "learning_rate": 1.05e-06,
+      "loss": 1.2861,
+      "step": 3350
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 4.6875,
+      "learning_rate": 1.048421052631579e-06,
+      "loss": 1.2146,
+      "step": 3360
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 4.625,
+      "learning_rate": 1.046842105263158e-06,
+      "loss": 1.3299,
+      "step": 3370
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 6.28125,
+      "learning_rate": 1.0452631578947368e-06,
+      "loss": 1.2183,
+      "step": 3380
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 5.09375,
+      "learning_rate": 1.0436842105263157e-06,
+      "loss": 1.2435,
+      "step": 3390
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 4.59375,
+      "learning_rate": 1.0421052631578949e-06,
+      "loss": 1.2189,
+      "step": 3400
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.0405263157894738e-06,
+      "loss": 1.2964,
+      "step": 3410
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 6.03125,
+      "learning_rate": 1.0389473684210527e-06,
+      "loss": 1.1613,
+      "step": 3420
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 5.625,
+      "learning_rate": 1.0373684210526316e-06,
+      "loss": 1.2126,
+      "step": 3430
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 5.40625,
+      "learning_rate": 1.0357894736842105e-06,
+      "loss": 1.2181,
+      "step": 3440
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 4.0,
+      "learning_rate": 1.0342105263157896e-06,
+      "loss": 1.2717,
+      "step": 3450
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 4.96875,
+      "learning_rate": 1.0326315789473685e-06,
+      "loss": 1.161,
+      "step": 3460
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 4.5,
+      "learning_rate": 1.0310526315789474e-06,
+      "loss": 1.3256,
+      "step": 3470
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 3.453125,
+      "learning_rate": 1.0294736842105263e-06,
+      "loss": 1.2079,
+      "step": 3480
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 5.40625,
+      "learning_rate": 1.0278947368421052e-06,
+      "loss": 1.2473,
+      "step": 3490
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 4.53125,
+      "learning_rate": 1.0263157894736843e-06,
+      "loss": 1.2761,
+      "step": 3500
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 3.984375,
+      "learning_rate": 1.024736842105263e-06,
+      "loss": 1.1995,
+      "step": 3510
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 6.1875,
+      "learning_rate": 1.0231578947368422e-06,
+      "loss": 1.2516,
+      "step": 3520
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 4.9375,
+      "learning_rate": 1.021578947368421e-06,
+      "loss": 1.3408,
+      "step": 3530
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 6.09375,
+      "learning_rate": 1.0200000000000002e-06,
+      "loss": 1.1742,
+      "step": 3540
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.0184210526315789e-06,
+      "loss": 1.22,
+      "step": 3550
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 3.9375,
+      "learning_rate": 1.0168421052631578e-06,
+      "loss": 1.1872,
+      "step": 3560
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.015263157894737e-06,
+      "loss": 1.255,
+      "step": 3570
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 5.0,
+      "learning_rate": 1.0136842105263158e-06,
+      "loss": 1.1664,
+      "step": 3580
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 6.75,
+      "learning_rate": 1.0121052631578947e-06,
+      "loss": 1.1648,
+      "step": 3590
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 5.5625,
+      "learning_rate": 1.0105263157894736e-06,
+      "loss": 1.3138,
+      "step": 3600
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 5.625,
+      "learning_rate": 1.0089473684210527e-06,
+      "loss": 1.2173,
+      "step": 3610
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 5.0625,
+      "learning_rate": 1.0073684210526317e-06,
+      "loss": 1.2128,
+      "step": 3620
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 4.3125,
+      "learning_rate": 1.0057894736842106e-06,
+      "loss": 1.2562,
+      "step": 3630
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 5.34375,
+      "learning_rate": 1.0042105263157895e-06,
+      "loss": 1.1745,
+      "step": 3640
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 6.875,
+      "learning_rate": 1.0026315789473684e-06,
+      "loss": 1.329,
+      "step": 3650
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 4.5625,
+      "learning_rate": 1.0010526315789475e-06,
+      "loss": 1.2953,
+      "step": 3660
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 4.5625,
+      "learning_rate": 9.994736842105264e-07,
+      "loss": 1.2442,
+      "step": 3670
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 5.0,
+      "learning_rate": 9.978947368421053e-07,
+      "loss": 1.1968,
+      "step": 3680
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 6.375,
+      "learning_rate": 9.963157894736842e-07,
+      "loss": 1.205,
+      "step": 3690
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 3.203125,
+      "learning_rate": 9.947368421052631e-07,
+      "loss": 1.2721,
+      "step": 3700
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 5.3125,
+      "learning_rate": 9.931578947368422e-07,
+      "loss": 1.3076,
+      "step": 3710
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 5.0625,
+      "learning_rate": 9.91578947368421e-07,
+      "loss": 1.1861,
+      "step": 3720
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 6.5,
+      "learning_rate": 9.9e-07,
+      "loss": 1.2757,
+      "step": 3730
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 5.03125,
+      "learning_rate": 9.88421052631579e-07,
+      "loss": 1.251,
+      "step": 3740
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 4.5625,
+      "learning_rate": 9.86842105263158e-07,
+      "loss": 1.3049,
+      "step": 3750
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 5.875,
+      "learning_rate": 9.852631578947368e-07,
+      "loss": 1.1694,
+      "step": 3760
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 4.0625,
+      "learning_rate": 9.836842105263157e-07,
+      "loss": 1.2924,
+      "step": 3770
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 3.765625,
+      "learning_rate": 9.821052631578948e-07,
+      "loss": 1.1178,
+      "step": 3780
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 8.5,
+      "learning_rate": 9.805263157894737e-07,
+      "loss": 1.4241,
+      "step": 3790
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 5.375,
+      "learning_rate": 9.789473684210526e-07,
+      "loss": 1.2019,
+      "step": 3800
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 6.53125,
+      "learning_rate": 9.773684210526315e-07,
+      "loss": 1.3219,
+      "step": 3810
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 4.40625,
+      "learning_rate": 9.757894736842106e-07,
+      "loss": 1.2338,
+      "step": 3820
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 4.75,
+      "learning_rate": 9.742105263157895e-07,
+      "loss": 1.2872,
+      "step": 3830
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 5.09375,
+      "learning_rate": 9.726315789473684e-07,
+      "loss": 1.2748,
+      "step": 3840
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 5.15625,
+      "learning_rate": 9.710526315789474e-07,
+      "loss": 1.2021,
+      "step": 3850
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 5.40625,
+      "learning_rate": 9.694736842105263e-07,
+      "loss": 1.2684,
+      "step": 3860
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 5.71875,
+      "learning_rate": 9.678947368421054e-07,
+      "loss": 1.2757,
+      "step": 3870
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 4.65625,
+      "learning_rate": 9.663157894736843e-07,
+      "loss": 1.1384,
+      "step": 3880
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 3.375,
+      "learning_rate": 9.647368421052632e-07,
+      "loss": 1.2843,
+      "step": 3890
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 5.0,
+      "learning_rate": 9.63157894736842e-07,
+      "loss": 1.3287,
+      "step": 3900
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 4.15625,
+      "learning_rate": 9.61578947368421e-07,
+      "loss": 1.1495,
+      "step": 3910
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 5.71875,
+      "learning_rate": 9.600000000000001e-07,
+      "loss": 1.3465,
+      "step": 3920
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 5.28125,
+      "learning_rate": 9.584210526315788e-07,
+      "loss": 1.2577,
+      "step": 3930
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 6.125,
+      "learning_rate": 9.56842105263158e-07,
+      "loss": 1.2289,
+      "step": 3940
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 5.84375,
+      "learning_rate": 9.552631578947368e-07,
+      "loss": 1.279,
+      "step": 3950
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 4.375,
+      "learning_rate": 9.53684210526316e-07,
+      "loss": 1.1106,
+      "step": 3960
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 6.09375,
+      "learning_rate": 9.521052631578948e-07,
+      "loss": 1.2996,
+      "step": 3970
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 4.875,
+      "learning_rate": 9.505263157894737e-07,
+      "loss": 1.1866,
+      "step": 3980
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 3.78125,
+      "learning_rate": 9.489473684210527e-07,
+      "loss": 1.2004,
+      "step": 3990
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 5.0625,
+      "learning_rate": 9.473684210526316e-07,
+      "loss": 1.2039,
+      "step": 4000
+    },
+    {
+      "epoch": 1.43,
+      "eval_loss": 1.2978060245513916,
+      "eval_runtime": 64.5441,
+      "eval_samples_per_second": 15.493,
+      "eval_steps_per_second": 15.493,
+      "step": 4000
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 5.46875,
+      "learning_rate": 9.457894736842106e-07,
+      "loss": 1.2033,
+      "step": 4010
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 5.6875,
+      "learning_rate": 9.442105263157895e-07,
+      "loss": 1.2614,
+      "step": 4020
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 4.65625,
+      "learning_rate": 9.426315789473685e-07,
+      "loss": 1.3363,
+      "step": 4030
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 4.125,
+      "learning_rate": 9.410526315789474e-07,
+      "loss": 1.3142,
+      "step": 4040
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 5.1875,
+      "learning_rate": 9.394736842105262e-07,
+      "loss": 1.2824,
+      "step": 4050
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 4.34375,
+      "learning_rate": 9.378947368421054e-07,
+      "loss": 1.2548,
+      "step": 4060
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 3.71875,
+      "learning_rate": 9.363157894736842e-07,
+      "loss": 1.2731,
+      "step": 4070
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 4.625,
+      "learning_rate": 9.347368421052633e-07,
+      "loss": 1.2233,
+      "step": 4080
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 2.875,
+      "learning_rate": 9.331578947368421e-07,
+      "loss": 1.3257,
+      "step": 4090
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 4.53125,
+      "learning_rate": 9.315789473684212e-07,
+      "loss": 1.2334,
+      "step": 4100
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 5.15625,
+      "learning_rate": 9.3e-07,
+      "loss": 1.1002,
+      "step": 4110
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 5.1875,
+      "learning_rate": 9.284210526315789e-07,
+      "loss": 1.2399,
+      "step": 4120
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 4.65625,
+      "learning_rate": 9.268421052631579e-07,
+      "loss": 1.2203,
+      "step": 4130
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 3.71875,
+      "learning_rate": 9.252631578947368e-07,
+      "loss": 1.3164,
+      "step": 4140
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 5.5,
+      "learning_rate": 9.236842105263158e-07,
+      "loss": 1.1929,
+      "step": 4150
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 5.40625,
+      "learning_rate": 9.221052631578947e-07,
+      "loss": 1.1785,
+      "step": 4160
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 5.34375,
+      "learning_rate": 9.205263157894737e-07,
+      "loss": 1.1197,
+      "step": 4170
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 5.5,
+      "learning_rate": 9.189473684210527e-07,
+      "loss": 1.2186,
+      "step": 4180
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 4.46875,
+      "learning_rate": 9.173684210526316e-07,
+      "loss": 1.2606,
+      "step": 4190
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 5.15625,
+      "learning_rate": 9.157894736842106e-07,
+      "loss": 1.2619,
+      "step": 4200
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 5.40625,
+      "learning_rate": 9.142105263157895e-07,
+      "loss": 1.2624,
+      "step": 4210
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 7.28125,
+      "learning_rate": 9.126315789473685e-07,
+      "loss": 1.2749,
+      "step": 4220
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 4.34375,
+      "learning_rate": 9.110526315789474e-07,
+      "loss": 1.3602,
+      "step": 4230
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 4.15625,
+      "learning_rate": 9.094736842105264e-07,
+      "loss": 1.162,
+      "step": 4240
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 3.875,
+      "learning_rate": 9.078947368421053e-07,
+      "loss": 1.269,
+      "step": 4250
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 5.28125,
+      "learning_rate": 9.063157894736841e-07,
+      "loss": 1.1874,
+      "step": 4260
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 6.28125,
+      "learning_rate": 9.047368421052632e-07,
+      "loss": 1.2483,
+      "step": 4270
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 4.0625,
+      "learning_rate": 9.03157894736842e-07,
+      "loss": 1.2523,
+      "step": 4280
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 3.09375,
+      "learning_rate": 9.015789473684212e-07,
+      "loss": 1.2586,
+      "step": 4290
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 3.5625,
+      "learning_rate": 9e-07,
+      "loss": 1.1993,
+      "step": 4300
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 6.6875,
+      "learning_rate": 8.984210526315791e-07,
+      "loss": 1.2135,
+      "step": 4310
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 5.90625,
+      "learning_rate": 8.968421052631579e-07,
+      "loss": 1.2604,
+      "step": 4320
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 6.75,
+      "learning_rate": 8.952631578947368e-07,
+      "loss": 1.317,
+      "step": 4330
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 6.125,
+      "learning_rate": 8.936842105263158e-07,
+      "loss": 1.1593,
+      "step": 4340
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 4.03125,
+      "learning_rate": 8.921052631578947e-07,
+      "loss": 1.2535,
+      "step": 4350
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 5.21875,
+      "learning_rate": 8.905263157894737e-07,
+      "loss": 1.2596,
+      "step": 4360
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 6.625,
+      "learning_rate": 8.889473684210526e-07,
+      "loss": 1.2864,
+      "step": 4370
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 2.796875,
+      "learning_rate": 8.873684210526316e-07,
+      "loss": 1.2415,
+      "step": 4380
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 5.0625,
+      "learning_rate": 8.857894736842105e-07,
+      "loss": 1.2328,
+      "step": 4390
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 5.71875,
+      "learning_rate": 8.842105263157895e-07,
+      "loss": 1.2419,
+      "step": 4400
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 3.625,
+      "learning_rate": 8.826315789473685e-07,
+      "loss": 1.3862,
+      "step": 4410
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 4.78125,
+      "learning_rate": 8.810526315789474e-07,
+      "loss": 1.2003,
+      "step": 4420
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 6.1875,
+      "learning_rate": 8.794736842105264e-07,
+      "loss": 1.3076,
+      "step": 4430
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 5.40625,
+      "learning_rate": 8.778947368421053e-07,
+      "loss": 1.2327,
+      "step": 4440
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 5.28125,
+      "learning_rate": 8.763157894736843e-07,
+      "loss": 1.2707,
+      "step": 4450
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 4.78125,
+      "learning_rate": 8.747368421052632e-07,
+      "loss": 1.2497,
+      "step": 4460
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 4.34375,
+      "learning_rate": 8.731578947368422e-07,
+      "loss": 1.3623,
+      "step": 4470
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 6.1875,
+      "learning_rate": 8.715789473684211e-07,
+      "loss": 1.1402,
+      "step": 4480
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 4.78125,
+      "learning_rate": 8.699999999999999e-07,
+      "loss": 1.2146,
+      "step": 4490
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 4.03125,
+      "learning_rate": 8.68421052631579e-07,
+      "loss": 1.217,
+      "step": 4500
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 4.28125,
+      "learning_rate": 8.668421052631578e-07,
+      "loss": 1.1123,
+      "step": 4510
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 3.296875,
+      "learning_rate": 8.65263157894737e-07,
+      "loss": 1.2955,
+      "step": 4520
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 4.3125,
+      "learning_rate": 8.636842105263158e-07,
+      "loss": 1.1202,
+      "step": 4530
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 3.8125,
+      "learning_rate": 8.621052631578949e-07,
+      "loss": 1.2172,
+      "step": 4540
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 5.875,
+      "learning_rate": 8.605263157894737e-07,
+      "loss": 1.1838,
+      "step": 4550
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 4.75,
+      "learning_rate": 8.589473684210526e-07,
+      "loss": 1.2317,
+      "step": 4560
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 6.375,
+      "learning_rate": 8.573684210526316e-07,
+      "loss": 1.3236,
+      "step": 4570
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 3.984375,
+      "learning_rate": 8.557894736842105e-07,
+      "loss": 1.2875,
+      "step": 4580
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 5.3125,
+      "learning_rate": 8.542105263157895e-07,
+      "loss": 1.3036,
+      "step": 4590
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 5.0625,
+      "learning_rate": 8.526315789473684e-07,
+      "loss": 1.2511,
+      "step": 4600
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 6.40625,
+      "learning_rate": 8.510526315789474e-07,
+      "loss": 1.2276,
+      "step": 4610
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 5.53125,
+      "learning_rate": 8.494736842105264e-07,
+      "loss": 1.2911,
+      "step": 4620
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 4.84375,
+      "learning_rate": 8.478947368421053e-07,
+      "loss": 1.2284,
+      "step": 4630
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 5.53125,
+      "learning_rate": 8.463157894736843e-07,
+      "loss": 1.1752,
+      "step": 4640
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 5.40625,
+      "learning_rate": 8.447368421052632e-07,
+      "loss": 1.2767,
+      "step": 4650
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 3.859375,
+      "learning_rate": 8.431578947368422e-07,
+      "loss": 1.2495,
+      "step": 4660
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 4.1875,
+      "learning_rate": 8.415789473684211e-07,
+      "loss": 1.1838,
+      "step": 4670
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 3.5,
+      "learning_rate": 8.400000000000001e-07,
+      "loss": 1.2218,
+      "step": 4680
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 5.5625,
+      "learning_rate": 8.38421052631579e-07,
+      "loss": 1.1463,
+      "step": 4690
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 3.78125,
+      "learning_rate": 8.368421052631578e-07,
+      "loss": 1.1506,
+      "step": 4700
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 6.375,
+      "learning_rate": 8.352631578947369e-07,
+      "loss": 1.2432,
+      "step": 4710
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 5.125,
+      "learning_rate": 8.336842105263157e-07,
+      "loss": 1.2779,
+      "step": 4720
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 6.25,
+      "learning_rate": 8.321052631578949e-07,
+      "loss": 1.189,
+      "step": 4730
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 5.25,
+      "learning_rate": 8.305263157894737e-07,
+      "loss": 1.3471,
+      "step": 4740
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 7.03125,
+      "learning_rate": 8.289473684210528e-07,
+      "loss": 1.3651,
+      "step": 4750
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 4.65625,
+      "learning_rate": 8.273684210526316e-07,
+      "loss": 1.1558,
+      "step": 4760
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 5.125,
+      "learning_rate": 8.257894736842105e-07,
+      "loss": 1.2075,
+      "step": 4770
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 5.28125,
+      "learning_rate": 8.242105263157895e-07,
+      "loss": 1.2764,
+      "step": 4780
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 4.96875,
+      "learning_rate": 8.226315789473684e-07,
+      "loss": 1.2569,
+      "step": 4790
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 6.40625,
+      "learning_rate": 8.210526315789474e-07,
+      "loss": 1.1777,
+      "step": 4800
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 4.6875,
+      "learning_rate": 8.194736842105263e-07,
+      "loss": 1.3637,
+      "step": 4810
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 6.96875,
+      "learning_rate": 8.178947368421053e-07,
+      "loss": 1.3181,
+      "step": 4820
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 4.90625,
+      "learning_rate": 8.163157894736842e-07,
+      "loss": 1.2333,
+      "step": 4830
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 6.4375,
+      "learning_rate": 8.147368421052631e-07,
+      "loss": 1.1743,
+      "step": 4840
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 5.96875,
+      "learning_rate": 8.131578947368422e-07,
+      "loss": 1.2702,
+      "step": 4850
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 6.5625,
+      "learning_rate": 8.115789473684211e-07,
+      "loss": 1.1836,
+      "step": 4860
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 4.9375,
+      "learning_rate": 8.100000000000001e-07,
+      "loss": 1.226,
+      "step": 4870
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 3.640625,
+      "learning_rate": 8.08421052631579e-07,
+      "loss": 1.1897,
+      "step": 4880
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 5.65625,
+      "learning_rate": 8.06842105263158e-07,
+      "loss": 1.3081,
+      "step": 4890
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 5.25,
+      "learning_rate": 8.052631578947369e-07,
+      "loss": 1.2547,
+      "step": 4900
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 4.59375,
+      "learning_rate": 8.036842105263157e-07,
+      "loss": 1.1939,
+      "step": 4910
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 4.40625,
+      "learning_rate": 8.021052631578948e-07,
+      "loss": 1.3548,
+      "step": 4920
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 6.53125,
+      "learning_rate": 8.005263157894736e-07,
+      "loss": 1.3371,
+      "step": 4930
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 6.03125,
+      "learning_rate": 7.989473684210527e-07,
+      "loss": 1.2122,
+      "step": 4940
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 3.8125,
+      "learning_rate": 7.973684210526315e-07,
+      "loss": 1.2003,
+      "step": 4950
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 5.34375,
+      "learning_rate": 7.957894736842107e-07,
+      "loss": 1.1898,
+      "step": 4960
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 5.6875,
+      "learning_rate": 7.942105263157895e-07,
+      "loss": 1.2971,
+      "step": 4970
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 6.46875,
+      "learning_rate": 7.926315789473684e-07,
+      "loss": 1.1603,
+      "step": 4980
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 4.0625,
+      "learning_rate": 7.910526315789474e-07,
+      "loss": 1.2873,
+      "step": 4990
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 5.375,
+      "learning_rate": 7.894736842105263e-07,
+      "loss": 1.2465,
+      "step": 5000
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 4.3125,
+      "learning_rate": 7.878947368421053e-07,
+      "loss": 1.2559,
+      "step": 5010
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 4.21875,
+      "learning_rate": 7.863157894736842e-07,
+      "loss": 1.1589,
+      "step": 5020
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 4.3125,
+      "learning_rate": 7.847368421052632e-07,
+      "loss": 1.2557,
+      "step": 5030
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 4.25,
+      "learning_rate": 7.831578947368421e-07,
+      "loss": 1.1851,
+      "step": 5040
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 6.46875,
+      "learning_rate": 7.81578947368421e-07,
+      "loss": 1.2658,
+      "step": 5050
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 5.15625,
+      "learning_rate": 7.8e-07,
+      "loss": 1.2452,
+      "step": 5060
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 4.4375,
+      "learning_rate": 7.78421052631579e-07,
+      "loss": 1.1701,
+      "step": 5070
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 3.890625,
+      "learning_rate": 7.76842105263158e-07,
+      "loss": 1.3408,
+      "step": 5080
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 6.03125,
+      "learning_rate": 7.752631578947369e-07,
+      "loss": 1.3047,
+      "step": 5090
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 3.984375,
+      "learning_rate": 7.736842105263159e-07,
+      "loss": 1.2437,
+      "step": 5100
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 4.25,
+      "learning_rate": 7.721052631578948e-07,
+      "loss": 1.2852,
+      "step": 5110
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 4.75,
+      "learning_rate": 7.705263157894736e-07,
+      "loss": 1.2755,
+      "step": 5120
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 4.65625,
+      "learning_rate": 7.689473684210527e-07,
+      "loss": 1.3391,
+      "step": 5130
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 5.96875,
+      "learning_rate": 7.673684210526315e-07,
+      "loss": 1.2166,
+      "step": 5140
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 6.21875,
+      "learning_rate": 7.657894736842106e-07,
+      "loss": 1.3108,
+      "step": 5150
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 5.125,
+      "learning_rate": 7.642105263157894e-07,
+      "loss": 1.2495,
+      "step": 5160
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 4.25,
+      "learning_rate": 7.626315789473686e-07,
+      "loss": 1.2511,
+      "step": 5170
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 5.96875,
+      "learning_rate": 7.610526315789474e-07,
+      "loss": 1.2249,
+      "step": 5180
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 5.5,
+      "learning_rate": 7.594736842105263e-07,
+      "loss": 1.2706,
+      "step": 5190
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 4.5625,
+      "learning_rate": 7.578947368421053e-07,
+      "loss": 1.3103,
+      "step": 5200
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 8.5625,
+      "learning_rate": 7.563157894736842e-07,
+      "loss": 1.3034,
+      "step": 5210
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 4.40625,
+      "learning_rate": 7.547368421052632e-07,
+      "loss": 1.068,
+      "step": 5220
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 5.25,
+      "learning_rate": 7.531578947368421e-07,
+      "loss": 1.2315,
+      "step": 5230
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 6.21875,
+      "learning_rate": 7.515789473684211e-07,
+      "loss": 1.2178,
+      "step": 5240
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 4.1875,
+      "learning_rate": 7.5e-07,
+      "loss": 1.2192,
+      "step": 5250
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 4.1875,
+      "learning_rate": 7.484210526315789e-07,
+      "loss": 1.1862,
+      "step": 5260
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 3.453125,
+      "learning_rate": 7.468421052631579e-07,
+      "loss": 1.2045,
+      "step": 5270
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 5.78125,
+      "learning_rate": 7.452631578947368e-07,
+      "loss": 1.2443,
+      "step": 5280
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 2.625,
+      "learning_rate": 7.436842105263158e-07,
+      "loss": 1.1414,
+      "step": 5290
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 7.4375,
+      "learning_rate": 7.421052631578948e-07,
+      "loss": 1.3059,
+      "step": 5300
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 4.625,
+      "learning_rate": 7.405263157894737e-07,
+      "loss": 1.2119,
+      "step": 5310
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 5.6875,
+      "learning_rate": 7.389473684210527e-07,
+      "loss": 1.323,
+      "step": 5320
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 3.796875,
+      "learning_rate": 7.373684210526316e-07,
+      "loss": 1.2824,
+      "step": 5330
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 6.34375,
+      "learning_rate": 7.357894736842106e-07,
+      "loss": 1.2775,
+      "step": 5340
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 4.0625,
+      "learning_rate": 7.342105263157895e-07,
+      "loss": 1.2661,
+      "step": 5350
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 4.96875,
+      "learning_rate": 7.326315789473684e-07,
+      "loss": 1.2221,
+      "step": 5360
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 5.9375,
+      "learning_rate": 7.310526315789473e-07,
+      "loss": 1.2545,
+      "step": 5370
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 4.625,
+      "learning_rate": 7.294736842105263e-07,
+      "loss": 1.173,
+      "step": 5380
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 7.28125,
+      "learning_rate": 7.278947368421052e-07,
+      "loss": 1.2131,
+      "step": 5390
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 5.3125,
+      "learning_rate": 7.263157894736843e-07,
+      "loss": 1.2319,
+      "step": 5400
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 6.59375,
+      "learning_rate": 7.247368421052632e-07,
+      "loss": 1.3161,
+      "step": 5410
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 4.125,
+      "learning_rate": 7.231578947368422e-07,
+      "loss": 1.1922,
+      "step": 5420
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 4.5625,
+      "learning_rate": 7.215789473684211e-07,
+      "loss": 1.2175,
+      "step": 5430
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 6.15625,
+      "learning_rate": 7.2e-07,
+      "loss": 1.2557,
+      "step": 5440
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 5.78125,
+      "learning_rate": 7.18421052631579e-07,
+      "loss": 1.1532,
+      "step": 5450
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 3.890625,
+      "learning_rate": 7.168421052631579e-07,
+      "loss": 1.1921,
+      "step": 5460
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 5.78125,
+      "learning_rate": 7.152631578947369e-07,
+      "loss": 1.3977,
+      "step": 5470
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 4.3125,
+      "learning_rate": 7.136842105263158e-07,
+      "loss": 1.2026,
+      "step": 5480
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 4.59375,
+      "learning_rate": 7.121052631578947e-07,
+      "loss": 1.2707,
+      "step": 5490
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 6.71875,
+      "learning_rate": 7.105263157894736e-07,
+      "loss": 1.2391,
+      "step": 5500
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 5.15625,
+      "learning_rate": 7.089473684210527e-07,
+      "loss": 1.2497,
+      "step": 5510
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 4.53125,
+      "learning_rate": 7.073684210526316e-07,
+      "loss": 1.2682,
+      "step": 5520
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 6.03125,
+      "learning_rate": 7.057894736842106e-07,
+      "loss": 1.2696,
+      "step": 5530
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 5.96875,
+      "learning_rate": 7.042105263157895e-07,
+      "loss": 1.2782,
+      "step": 5540
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 4.09375,
+      "learning_rate": 7.026315789473685e-07,
+      "loss": 1.2647,
+      "step": 5550
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 6.53125,
+      "learning_rate": 7.010526315789474e-07,
+      "loss": 1.2365,
+      "step": 5560
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 6.09375,
+      "learning_rate": 6.994736842105263e-07,
+      "loss": 1.2428,
+      "step": 5570
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 4.4375,
+      "learning_rate": 6.978947368421052e-07,
+      "loss": 1.2722,
+      "step": 5580
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 4.75,
+      "learning_rate": 6.963157894736842e-07,
+      "loss": 1.2367,
+      "step": 5590
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 5.46875,
+      "learning_rate": 6.947368421052631e-07,
+      "loss": 1.337,
+      "step": 5600
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 6.3125,
+      "learning_rate": 6.931578947368421e-07,
+      "loss": 1.2827,
+      "step": 5610
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 7.34375,
+      "learning_rate": 6.91578947368421e-07,
+      "loss": 1.3569,
+      "step": 5620
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 5.1875,
+      "learning_rate": 6.900000000000001e-07,
+      "loss": 1.1996,
+      "step": 5630
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 5.90625,
+      "learning_rate": 6.88421052631579e-07,
+      "loss": 1.3072,
+      "step": 5640
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 6.21875,
+      "learning_rate": 6.868421052631579e-07,
+      "loss": 1.2958,
+      "step": 5650
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 4.875,
+      "learning_rate": 6.852631578947369e-07,
+      "loss": 1.2732,
+      "step": 5660
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 5.46875,
+      "learning_rate": 6.836842105263158e-07,
+      "loss": 1.2906,
+      "step": 5670
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 4.09375,
+      "learning_rate": 6.821052631578948e-07,
+      "loss": 1.2793,
+      "step": 5680
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 6.375,
+      "learning_rate": 6.805263157894737e-07,
+      "loss": 1.289,
+      "step": 5690
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 6.625,
+      "learning_rate": 6.789473684210527e-07,
+      "loss": 1.2384,
+      "step": 5700
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 4.375,
+      "learning_rate": 6.773684210526316e-07,
+      "loss": 1.3608,
+      "step": 5710
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 6.03125,
+      "learning_rate": 6.757894736842105e-07,
+      "loss": 1.2554,
+      "step": 5720
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 7.59375,
+      "learning_rate": 6.742105263157894e-07,
+      "loss": 1.2276,
+      "step": 5730
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 5.25,
+      "learning_rate": 6.726315789473685e-07,
+      "loss": 1.2748,
+      "step": 5740
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 4.5,
+      "learning_rate": 6.710526315789474e-07,
+      "loss": 1.397,
+      "step": 5750
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 9.5,
+      "learning_rate": 6.694736842105264e-07,
+      "loss": 1.3647,
+      "step": 5760
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 4.71875,
+      "learning_rate": 6.678947368421053e-07,
+      "loss": 1.3448,
+      "step": 5770
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 5.28125,
+      "learning_rate": 6.663157894736843e-07,
+      "loss": 1.2532,
+      "step": 5780
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 5.03125,
+      "learning_rate": 6.647368421052631e-07,
+      "loss": 1.2502,
+      "step": 5790
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 6.25,
+      "learning_rate": 6.631578947368421e-07,
+      "loss": 1.2307,
+      "step": 5800
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 6.78125,
+      "learning_rate": 6.61578947368421e-07,
+      "loss": 1.2215,
+      "step": 5810
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 3.6875,
+      "learning_rate": 6.6e-07,
+      "loss": 1.2169,
+      "step": 5820
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 5.8125,
+      "learning_rate": 6.584210526315789e-07,
+      "loss": 1.2402,
+      "step": 5830
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 5.0625,
+      "learning_rate": 6.56842105263158e-07,
+      "loss": 1.3188,
+      "step": 5840
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 5.53125,
+      "learning_rate": 6.552631578947369e-07,
+      "loss": 1.3201,
+      "step": 5850
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 3.40625,
+      "learning_rate": 6.536842105263158e-07,
+      "loss": 1.2958,
+      "step": 5860
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 5.3125,
+      "learning_rate": 6.521052631578948e-07,
+      "loss": 1.3746,
+      "step": 5870
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 5.625,
+      "learning_rate": 6.505263157894737e-07,
+      "loss": 1.2334,
+      "step": 5880
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 5.0625,
+      "learning_rate": 6.489473684210527e-07,
+      "loss": 1.2365,
+      "step": 5890
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 6.75,
+      "learning_rate": 6.473684210526316e-07,
+      "loss": 1.3491,
+      "step": 5900
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 6.5625,
+      "learning_rate": 6.457894736842106e-07,
+      "loss": 1.2262,
+      "step": 5910
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 5.40625,
+      "learning_rate": 6.442105263157895e-07,
+      "loss": 1.3511,
+      "step": 5920
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 5.3125,
+      "learning_rate": 6.426315789473684e-07,
+      "loss": 1.2133,
+      "step": 5930
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 4.8125,
+      "learning_rate": 6.410526315789473e-07,
+      "loss": 1.291,
+      "step": 5940
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 6.6875,
+      "learning_rate": 6.394736842105264e-07,
+      "loss": 1.2519,
+      "step": 5950
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 4.96875,
+      "learning_rate": 6.378947368421053e-07,
+      "loss": 1.2712,
+      "step": 5960
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 5.0625,
+      "learning_rate": 6.363157894736843e-07,
+      "loss": 1.3061,
+      "step": 5970
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 3.578125,
+      "learning_rate": 6.347368421052632e-07,
+      "loss": 1.2029,
+      "step": 5980
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 5.5625,
+      "learning_rate": 6.331578947368422e-07,
+      "loss": 1.2404,
+      "step": 5990
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 6.46875,
+      "learning_rate": 6.31578947368421e-07,
+      "loss": 1.2706,
+      "step": 6000
+    },
+    {
+      "epoch": 2.15,
+      "eval_loss": 1.3080763816833496,
+      "eval_runtime": 64.5168,
+      "eval_samples_per_second": 15.5,
+      "eval_steps_per_second": 15.5,
+      "step": 6000
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 6.25,
+      "learning_rate": 6.3e-07,
+      "loss": 1.2568,
+      "step": 6010
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 4.25,
+      "learning_rate": 6.284210526315789e-07,
+      "loss": 1.1807,
+      "step": 6020
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 4.25,
+      "learning_rate": 6.268421052631579e-07,
+      "loss": 1.3481,
+      "step": 6030
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 7.09375,
+      "learning_rate": 6.252631578947368e-07,
+      "loss": 1.2661,
+      "step": 6040
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 4.625,
+      "learning_rate": 6.236842105263158e-07,
+      "loss": 1.3857,
+      "step": 6050
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 6.09375,
+      "learning_rate": 6.221052631578947e-07,
+      "loss": 1.3428,
+      "step": 6060
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 4.0625,
+      "learning_rate": 6.205263157894737e-07,
+      "loss": 1.2642,
+      "step": 6070
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 4.0625,
+      "learning_rate": 6.189473684210527e-07,
+      "loss": 1.2849,
+      "step": 6080
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 4.1875,
+      "learning_rate": 6.173684210526316e-07,
+      "loss": 1.2598,
+      "step": 6090
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 5.5625,
+      "learning_rate": 6.157894736842106e-07,
+      "loss": 1.304,
+      "step": 6100
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 5.03125,
+      "learning_rate": 6.142105263157895e-07,
+      "loss": 1.2041,
+      "step": 6110
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 6.84375,
+      "learning_rate": 6.126315789473685e-07,
+      "loss": 1.1821,
+      "step": 6120
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 7.78125,
+      "learning_rate": 6.110526315789474e-07,
+      "loss": 1.2831,
+      "step": 6130
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 8.8125,
+      "learning_rate": 6.094736842105263e-07,
+      "loss": 1.2871,
+      "step": 6140
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 5.1875,
+      "learning_rate": 6.078947368421052e-07,
+      "loss": 1.2752,
+      "step": 6150
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 8.4375,
+      "learning_rate": 6.063157894736842e-07,
+      "loss": 1.2712,
+      "step": 6160
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 5.3125,
+      "learning_rate": 6.047368421052631e-07,
+      "loss": 1.3152,
+      "step": 6170
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 5.21875,
+      "learning_rate": 6.031578947368422e-07,
+      "loss": 1.2677,
+      "step": 6180
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 6.40625,
+      "learning_rate": 6.015789473684211e-07,
+      "loss": 1.2663,
+      "step": 6190
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 4.90625,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 1.2564,
+      "step": 6200
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 6.40625,
+      "learning_rate": 5.984210526315789e-07,
+      "loss": 1.2493,
+      "step": 6210
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 5.4375,
+      "learning_rate": 5.968421052631579e-07,
+      "loss": 1.2984,
+      "step": 6220
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 5.71875,
+      "learning_rate": 5.952631578947368e-07,
+      "loss": 1.243,
+      "step": 6230
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 7.90625,
+      "learning_rate": 5.936842105263158e-07,
+      "loss": 1.3316,
+      "step": 6240
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 5.21875,
+      "learning_rate": 5.921052631578947e-07,
+      "loss": 1.1823,
+      "step": 6250
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 5.46875,
+      "learning_rate": 5.905263157894737e-07,
+      "loss": 1.3328,
+      "step": 6260
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 6.34375,
+      "learning_rate": 5.889473684210526e-07,
+      "loss": 1.3038,
+      "step": 6270
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 5.0,
+      "learning_rate": 5.873684210526315e-07,
+      "loss": 1.2292,
+      "step": 6280
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 8.375,
+      "learning_rate": 5.857894736842106e-07,
+      "loss": 1.2515,
+      "step": 6290
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 5.53125,
+      "learning_rate": 5.842105263157895e-07,
+      "loss": 1.2778,
+      "step": 6300
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 5.125,
+      "learning_rate": 5.826315789473685e-07,
+      "loss": 1.1158,
+      "step": 6310
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 7.28125,
+      "learning_rate": 5.810526315789474e-07,
+      "loss": 1.2386,
+      "step": 6320
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 5.1875,
+      "learning_rate": 5.794736842105264e-07,
+      "loss": 1.3403,
+      "step": 6330
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 5.0,
+      "learning_rate": 5.778947368421053e-07,
+      "loss": 1.3082,
+      "step": 6340
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 6.3125,
+      "learning_rate": 5.763157894736842e-07,
+      "loss": 1.1886,
+      "step": 6350
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 6.0,
+      "learning_rate": 5.747368421052631e-07,
+      "loss": 1.3394,
+      "step": 6360
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 6.5,
+      "learning_rate": 5.731578947368421e-07,
+      "loss": 1.2935,
+      "step": 6370
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 5.5,
+      "learning_rate": 5.71578947368421e-07,
+      "loss": 1.3251,
+      "step": 6380
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 5.40625,
+      "learning_rate": 5.7e-07,
+      "loss": 1.2237,
+      "step": 6390
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 6.625,
+      "learning_rate": 5.68421052631579e-07,
+      "loss": 1.3365,
+      "step": 6400
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 4.125,
+      "learning_rate": 5.66842105263158e-07,
+      "loss": 1.3294,
+      "step": 6410
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 7.375,
+      "learning_rate": 5.652631578947368e-07,
+      "loss": 1.326,
+      "step": 6420
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 5.46875,
+      "learning_rate": 5.636842105263158e-07,
+      "loss": 1.2297,
+      "step": 6430
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 7.84375,
+      "learning_rate": 5.621052631578947e-07,
+      "loss": 1.2347,
+      "step": 6440
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 3.40625,
+      "learning_rate": 5.605263157894737e-07,
+      "loss": 1.3048,
+      "step": 6450
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 7.5,
+      "learning_rate": 5.589473684210526e-07,
+      "loss": 1.3468,
+      "step": 6460
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 4.71875,
+      "learning_rate": 5.573684210526316e-07,
+      "loss": 1.2422,
+      "step": 6470
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 7.125,
+      "learning_rate": 5.557894736842105e-07,
+      "loss": 1.3711,
+      "step": 6480
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 4.1875,
+      "learning_rate": 5.542105263157895e-07,
+      "loss": 1.3617,
+      "step": 6490
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 5.28125,
+      "learning_rate": 5.526315789473684e-07,
+      "loss": 1.2929,
+      "step": 6500
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 4.625,
+      "learning_rate": 5.510526315789474e-07,
+      "loss": 1.2417,
+      "step": 6510
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 5.59375,
+      "learning_rate": 5.494736842105264e-07,
+      "loss": 1.2321,
+      "step": 6520
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 5.8125,
+      "learning_rate": 5.478947368421053e-07,
+      "loss": 1.2214,
+      "step": 6530
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 5.9375,
+      "learning_rate": 5.463157894736843e-07,
+      "loss": 1.2832,
+      "step": 6540
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 4.65625,
+      "learning_rate": 5.447368421052632e-07,
+      "loss": 1.3456,
+      "step": 6550
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 4.78125,
+      "learning_rate": 5.431578947368422e-07,
+      "loss": 1.2742,
+      "step": 6560
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 4.03125,
+      "learning_rate": 5.41578947368421e-07,
+      "loss": 1.2544,
+      "step": 6570
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 7.90625,
+      "learning_rate": 5.4e-07,
+      "loss": 1.2501,
+      "step": 6580
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 5.375,
+      "learning_rate": 5.384210526315789e-07,
+      "loss": 1.2487,
+      "step": 6590
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 5.90625,
+      "learning_rate": 5.368421052631579e-07,
+      "loss": 1.3756,
+      "step": 6600
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 5.9375,
+      "learning_rate": 5.352631578947368e-07,
+      "loss": 1.1941,
+      "step": 6610
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 5.1875,
+      "learning_rate": 5.336842105263159e-07,
+      "loss": 1.2885,
+      "step": 6620
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 6.59375,
+      "learning_rate": 5.321052631578948e-07,
+      "loss": 1.2311,
+      "step": 6630
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 6.25,
+      "learning_rate": 5.305263157894737e-07,
+      "loss": 1.224,
+      "step": 6640
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 5.09375,
+      "learning_rate": 5.289473684210526e-07,
+      "loss": 1.2278,
+      "step": 6650
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 5.15625,
+      "learning_rate": 5.273684210526316e-07,
+      "loss": 1.3222,
+      "step": 6660
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 5.59375,
+      "learning_rate": 5.257894736842105e-07,
+      "loss": 1.2766,
+      "step": 6670
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 6.625,
+      "learning_rate": 5.242105263157895e-07,
+      "loss": 1.3131,
+      "step": 6680
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 6.34375,
+      "learning_rate": 5.226315789473684e-07,
+      "loss": 1.2999,
+      "step": 6690
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 5.96875,
+      "learning_rate": 5.210526315789474e-07,
+      "loss": 1.2805,
+      "step": 6700
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 7.09375,
+      "learning_rate": 5.194736842105263e-07,
+      "loss": 1.2095,
+      "step": 6710
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 5.0625,
+      "learning_rate": 5.178947368421052e-07,
+      "loss": 1.1974,
+      "step": 6720
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 5.125,
+      "learning_rate": 5.163157894736843e-07,
+      "loss": 1.3493,
+      "step": 6730
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 5.40625,
+      "learning_rate": 5.147368421052632e-07,
+      "loss": 1.2289,
+      "step": 6740
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 7.3125,
+      "learning_rate": 5.131578947368422e-07,
+      "loss": 1.2266,
+      "step": 6750
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 5.9375,
+      "learning_rate": 5.115789473684211e-07,
+      "loss": 1.3067,
+      "step": 6760
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 5.75,
+      "learning_rate": 5.100000000000001e-07,
+      "loss": 1.37,
+      "step": 6770
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 6.5,
+      "learning_rate": 5.084210526315789e-07,
+      "loss": 1.3337,
+      "step": 6780
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 3.53125,
+      "learning_rate": 5.068421052631579e-07,
+      "loss": 1.2663,
+      "step": 6790
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 4.90625,
+      "learning_rate": 5.052631578947368e-07,
+      "loss": 1.2713,
+      "step": 6800
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 4.375,
+      "learning_rate": 5.036842105263158e-07,
+      "loss": 1.2998,
+      "step": 6810
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 6.53125,
+      "learning_rate": 5.021052631578947e-07,
+      "loss": 1.1186,
+      "step": 6820
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 6.0625,
+      "learning_rate": 5.005263157894737e-07,
+      "loss": 1.3082,
+      "step": 6830
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 3.953125,
+      "learning_rate": 4.989473684210527e-07,
+      "loss": 1.2321,
+      "step": 6840
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 5.0625,
+      "learning_rate": 4.973684210526316e-07,
+      "loss": 1.2674,
+      "step": 6850
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 5.125,
+      "learning_rate": 4.957894736842105e-07,
+      "loss": 1.2587,
+      "step": 6860
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 6.59375,
+      "learning_rate": 4.942105263157895e-07,
+      "loss": 1.2964,
+      "step": 6870
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 4.46875,
+      "learning_rate": 4.926315789473684e-07,
+      "loss": 1.3224,
+      "step": 6880
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 4.0,
+      "learning_rate": 4.910526315789474e-07,
+      "loss": 1.2625,
+      "step": 6890
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 5.90625,
+      "learning_rate": 4.894736842105263e-07,
+      "loss": 1.2987,
+      "step": 6900
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 7.75,
+      "learning_rate": 4.878947368421053e-07,
+      "loss": 1.2911,
+      "step": 6910
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 5.25,
+      "learning_rate": 4.863157894736842e-07,
+      "loss": 1.1386,
+      "step": 6920
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 4.90625,
+      "learning_rate": 4.847368421052631e-07,
+      "loss": 1.3652,
+      "step": 6930
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 6.875,
+      "learning_rate": 4.831578947368421e-07,
+      "loss": 1.2419,
+      "step": 6940
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 4.125,
+      "learning_rate": 4.81578947368421e-07,
+      "loss": 1.2013,
+      "step": 6950
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 4.625,
+      "learning_rate": 4.800000000000001e-07,
+      "loss": 1.2694,
+      "step": 6960
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 5.84375,
+      "learning_rate": 4.78421052631579e-07,
+      "loss": 1.4021,
+      "step": 6970
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 5.46875,
+      "learning_rate": 4.76842105263158e-07,
+      "loss": 1.3204,
+      "step": 6980
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 5.125,
+      "learning_rate": 4.7526315789473684e-07,
+      "loss": 1.3388,
+      "step": 6990
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 6.875,
+      "learning_rate": 4.736842105263158e-07,
+      "loss": 1.3317,
+      "step": 7000
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 5.6875,
+      "learning_rate": 4.7210526315789476e-07,
+      "loss": 1.2053,
+      "step": 7010
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 5.375,
+      "learning_rate": 4.705263157894737e-07,
+      "loss": 1.3581,
+      "step": 7020
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 4.59375,
+      "learning_rate": 4.689473684210527e-07,
+      "loss": 1.2331,
+      "step": 7030
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 4.3125,
+      "learning_rate": 4.6736842105263163e-07,
+      "loss": 1.2968,
+      "step": 7040
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 4.46875,
+      "learning_rate": 4.657894736842106e-07,
+      "loss": 1.3134,
+      "step": 7050
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 5.78125,
+      "learning_rate": 4.6421052631578945e-07,
+      "loss": 1.3078,
+      "step": 7060
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 4.09375,
+      "learning_rate": 4.626315789473684e-07,
+      "loss": 1.3564,
+      "step": 7070
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 5.5625,
+      "learning_rate": 4.6105263157894737e-07,
+      "loss": 1.3339,
+      "step": 7080
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 4.625,
+      "learning_rate": 4.5947368421052633e-07,
+      "loss": 1.3372,
+      "step": 7090
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 5.03125,
+      "learning_rate": 4.578947368421053e-07,
+      "loss": 1.2462,
+      "step": 7100
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 8.5,
+      "learning_rate": 4.5631578947368425e-07,
+      "loss": 1.2572,
+      "step": 7110
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 5.96875,
+      "learning_rate": 4.547368421052632e-07,
+      "loss": 1.3266,
+      "step": 7120
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 5.28125,
+      "learning_rate": 4.5315789473684206e-07,
+      "loss": 1.2053,
+      "step": 7130
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 4.0,
+      "learning_rate": 4.51578947368421e-07,
+      "loss": 1.3004,
+      "step": 7140
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 4.65625,
+      "learning_rate": 4.5e-07,
+      "loss": 1.1984,
+      "step": 7150
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 3.71875,
+      "learning_rate": 4.4842105263157894e-07,
+      "loss": 1.1182,
+      "step": 7160
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 4.15625,
+      "learning_rate": 4.468421052631579e-07,
+      "loss": 1.2745,
+      "step": 7170
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 4.40625,
+      "learning_rate": 4.4526315789473686e-07,
+      "loss": 1.3204,
+      "step": 7180
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 7.09375,
+      "learning_rate": 4.436842105263158e-07,
+      "loss": 1.2898,
+      "step": 7190
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 5.96875,
+      "learning_rate": 4.421052631578947e-07,
+      "loss": 1.1829,
+      "step": 7200
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 5.4375,
+      "learning_rate": 4.405263157894737e-07,
+      "loss": 1.25,
+      "step": 7210
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 5.6875,
+      "learning_rate": 4.3894736842105264e-07,
+      "loss": 1.2121,
+      "step": 7220
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 4.21875,
+      "learning_rate": 4.373684210526316e-07,
+      "loss": 1.2617,
+      "step": 7230
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 4.5625,
+      "learning_rate": 4.3578947368421056e-07,
+      "loss": 1.3154,
+      "step": 7240
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 4.34375,
+      "learning_rate": 4.342105263157895e-07,
+      "loss": 1.3184,
+      "step": 7250
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 5.1875,
+      "learning_rate": 4.326315789473685e-07,
+      "loss": 1.3071,
+      "step": 7260
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 4.53125,
+      "learning_rate": 4.3105263157894744e-07,
+      "loss": 1.3033,
+      "step": 7270
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 6.34375,
+      "learning_rate": 4.294736842105263e-07,
+      "loss": 1.2635,
+      "step": 7280
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 4.90625,
+      "learning_rate": 4.2789473684210526e-07,
+      "loss": 1.3287,
+      "step": 7290
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 6.90625,
+      "learning_rate": 4.263157894736842e-07,
+      "loss": 1.353,
+      "step": 7300
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 8.8125,
+      "learning_rate": 4.247368421052632e-07,
+      "loss": 1.2535,
+      "step": 7310
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 5.3125,
+      "learning_rate": 4.2315789473684214e-07,
+      "loss": 1.3146,
+      "step": 7320
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 8.875,
+      "learning_rate": 4.215789473684211e-07,
+      "loss": 1.3312,
+      "step": 7330
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 6.8125,
+      "learning_rate": 4.2000000000000006e-07,
+      "loss": 1.2497,
+      "step": 7340
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 5.84375,
+      "learning_rate": 4.184210526315789e-07,
+      "loss": 1.2423,
+      "step": 7350
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 6.46875,
+      "learning_rate": 4.1684210526315787e-07,
+      "loss": 1.27,
+      "step": 7360
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 5.1875,
+      "learning_rate": 4.1526315789473683e-07,
+      "loss": 1.2337,
+      "step": 7370
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 5.21875,
+      "learning_rate": 4.136842105263158e-07,
+      "loss": 1.3373,
+      "step": 7380
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 6.75,
+      "learning_rate": 4.1210526315789475e-07,
+      "loss": 1.2874,
+      "step": 7390
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 7.78125,
+      "learning_rate": 4.105263157894737e-07,
+      "loss": 1.3553,
+      "step": 7400
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 4.40625,
+      "learning_rate": 4.0894736842105267e-07,
+      "loss": 1.1787,
+      "step": 7410
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 4.8125,
+      "learning_rate": 4.073684210526316e-07,
+      "loss": 1.2681,
+      "step": 7420
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 6.375,
+      "learning_rate": 4.0578947368421053e-07,
+      "loss": 1.3252,
+      "step": 7430
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 5.28125,
+      "learning_rate": 4.042105263157895e-07,
+      "loss": 1.4007,
+      "step": 7440
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 4.96875,
+      "learning_rate": 4.0263157894736845e-07,
+      "loss": 1.3017,
+      "step": 7450
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 6.0625,
+      "learning_rate": 4.010526315789474e-07,
+      "loss": 1.3029,
+      "step": 7460
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 6.125,
+      "learning_rate": 3.9947368421052637e-07,
+      "loss": 1.2494,
+      "step": 7470
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 5.1875,
+      "learning_rate": 3.9789473684210533e-07,
+      "loss": 1.3023,
+      "step": 7480
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 4.375,
+      "learning_rate": 3.963157894736842e-07,
+      "loss": 1.3009,
+      "step": 7490
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 6.46875,
+      "learning_rate": 3.9473684210526315e-07,
+      "loss": 1.2782,
+      "step": 7500
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 4.625,
+      "learning_rate": 3.931578947368421e-07,
+      "loss": 1.2013,
+      "step": 7510
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 6.96875,
+      "learning_rate": 3.9157894736842107e-07,
+      "loss": 1.408,
+      "step": 7520
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 6.21875,
+      "learning_rate": 3.9e-07,
+      "loss": 1.2942,
+      "step": 7530
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 7.125,
+      "learning_rate": 3.88421052631579e-07,
+      "loss": 1.2746,
+      "step": 7540
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 3.9375,
+      "learning_rate": 3.8684210526315794e-07,
+      "loss": 1.3276,
+      "step": 7550
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 5.15625,
+      "learning_rate": 3.852631578947368e-07,
+      "loss": 1.2782,
+      "step": 7560
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 4.75,
+      "learning_rate": 3.8368421052631576e-07,
+      "loss": 1.1563,
+      "step": 7570
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 4.40625,
+      "learning_rate": 3.821052631578947e-07,
+      "loss": 1.1357,
+      "step": 7580
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 4.46875,
+      "learning_rate": 3.805263157894737e-07,
+      "loss": 1.2978,
+      "step": 7590
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 4.9375,
+      "learning_rate": 3.7894736842105264e-07,
+      "loss": 1.3815,
+      "step": 7600
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 8.6875,
+      "learning_rate": 3.773684210526316e-07,
+      "loss": 1.3365,
+      "step": 7610
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 5.40625,
+      "learning_rate": 3.7578947368421056e-07,
+      "loss": 1.312,
+      "step": 7620
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 4.84375,
+      "learning_rate": 3.7421052631578946e-07,
+      "loss": 1.3392,
+      "step": 7630
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 6.875,
+      "learning_rate": 3.726315789473684e-07,
+      "loss": 1.3296,
+      "step": 7640
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 4.375,
+      "learning_rate": 3.710526315789474e-07,
+      "loss": 1.3867,
+      "step": 7650
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 5.78125,
+      "learning_rate": 3.6947368421052634e-07,
+      "loss": 1.2468,
+      "step": 7660
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 7.28125,
+      "learning_rate": 3.678947368421053e-07,
+      "loss": 1.1845,
+      "step": 7670
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 6.09375,
+      "learning_rate": 3.663157894736842e-07,
+      "loss": 1.2763,
+      "step": 7680
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 6.78125,
+      "learning_rate": 3.6473684210526317e-07,
+      "loss": 1.3962,
+      "step": 7690
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 6.34375,
+      "learning_rate": 3.6315789473684213e-07,
+      "loss": 1.3092,
+      "step": 7700
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 5.03125,
+      "learning_rate": 3.615789473684211e-07,
+      "loss": 1.227,
+      "step": 7710
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 3.34375,
+      "learning_rate": 3.6e-07,
+      "loss": 1.3445,
+      "step": 7720
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 4.03125,
+      "learning_rate": 3.5842105263157895e-07,
+      "loss": 1.3133,
+      "step": 7730
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 4.25,
+      "learning_rate": 3.568421052631579e-07,
+      "loss": 1.2684,
+      "step": 7740
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 5.1875,
+      "learning_rate": 3.552631578947368e-07,
+      "loss": 1.3025,
+      "step": 7750
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 7.65625,
+      "learning_rate": 3.536842105263158e-07,
+      "loss": 1.1968,
+      "step": 7760
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 4.125,
+      "learning_rate": 3.5210526315789474e-07,
+      "loss": 1.3351,
+      "step": 7770
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 3.890625,
+      "learning_rate": 3.505263157894737e-07,
+      "loss": 1.1521,
+      "step": 7780
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 7.125,
+      "learning_rate": 3.489473684210526e-07,
+      "loss": 1.3631,
+      "step": 7790
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 5.65625,
+      "learning_rate": 3.4736842105263157e-07,
+      "loss": 1.2709,
+      "step": 7800
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 6.90625,
+      "learning_rate": 3.457894736842105e-07,
+      "loss": 1.186,
+      "step": 7810
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 8.3125,
+      "learning_rate": 3.442105263157895e-07,
+      "loss": 1.2198,
+      "step": 7820
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 3.265625,
+      "learning_rate": 3.4263157894736845e-07,
+      "loss": 1.1781,
+      "step": 7830
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 4.9375,
+      "learning_rate": 3.410526315789474e-07,
+      "loss": 1.1974,
+      "step": 7840
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 5.71875,
+      "learning_rate": 3.3947368421052636e-07,
+      "loss": 1.3283,
+      "step": 7850
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 5.15625,
+      "learning_rate": 3.3789473684210527e-07,
+      "loss": 1.2897,
+      "step": 7860
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 5.84375,
+      "learning_rate": 3.3631578947368423e-07,
+      "loss": 1.278,
+      "step": 7870
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 4.96875,
+      "learning_rate": 3.347368421052632e-07,
+      "loss": 1.2842,
+      "step": 7880
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 5.28125,
+      "learning_rate": 3.3315789473684215e-07,
+      "loss": 1.2207,
+      "step": 7890
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 7.90625,
+      "learning_rate": 3.3157894736842106e-07,
+      "loss": 1.1945,
+      "step": 7900
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 8.4375,
+      "learning_rate": 3.3e-07,
+      "loss": 1.3554,
+      "step": 7910
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 5.0,
+      "learning_rate": 3.28421052631579e-07,
+      "loss": 1.2055,
+      "step": 7920
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 4.65625,
+      "learning_rate": 3.268421052631579e-07,
+      "loss": 1.3286,
+      "step": 7930
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 7.59375,
+      "learning_rate": 3.2526315789473684e-07,
+      "loss": 1.3289,
+      "step": 7940
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 5.28125,
+      "learning_rate": 3.236842105263158e-07,
+      "loss": 1.2589,
+      "step": 7950
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 8.0,
+      "learning_rate": 3.2210526315789476e-07,
+      "loss": 1.2188,
+      "step": 7960
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 6.59375,
+      "learning_rate": 3.2052631578947367e-07,
+      "loss": 1.2005,
+      "step": 7970
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 6.03125,
+      "learning_rate": 3.1894736842105263e-07,
+      "loss": 1.3808,
+      "step": 7980
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 7.1875,
+      "learning_rate": 3.173684210526316e-07,
+      "loss": 1.2131,
+      "step": 7990
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 6.75,
+      "learning_rate": 3.157894736842105e-07,
+      "loss": 1.2658,
+      "step": 8000
+    },
+    {
+      "epoch": 2.87,
+      "eval_loss": 1.3140743970870972,
+      "eval_runtime": 64.7365,
+      "eval_samples_per_second": 15.447,
+      "eval_steps_per_second": 15.447,
+      "step": 8000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 2000,
+  "total_flos": 1.2911376334848e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}