diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,10968 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9090644635792413,
+  "eval_steps": 500,
+  "global_step": 15625,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 6.432150460863581e-05,
+      "grad_norm": 2.25,
+      "learning_rate": 2e-05,
+      "loss": 0.9123,
+      "step": 10
+    },
+    {
+      "epoch": 0.00012864300921727162,
+      "grad_norm": 2.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.5268,
+      "step": 20
+    },
+    {
+      "epoch": 0.00019296451382590742,
+      "grad_norm": 2.125,
+      "learning_rate": 2e-05,
+      "loss": 0.5193,
+      "step": 30
+    },
+    {
+      "epoch": 0.00025728601843454324,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.5167,
+      "step": 40
+    },
+    {
+      "epoch": 0.00032160752304317904,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.5129,
+      "step": 50
+    },
+    {
+      "epoch": 0.00038592902765181484,
+      "grad_norm": 2.78125,
+      "learning_rate": 2e-05,
+      "loss": 0.5131,
+      "step": 60
+    },
+    {
+      "epoch": 0.00045025053226045063,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.5065,
+      "step": 70
+    },
+    {
+      "epoch": 0.0005145720368690865,
+      "grad_norm": 2.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.5123,
+      "step": 80
+    },
+    {
+      "epoch": 0.0005788935414777223,
+      "grad_norm": 2.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.4953,
+      "step": 90
+    },
+    {
+      "epoch": 0.0006432150460863581,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.4935,
+      "step": 100
+    },
+    {
+      "epoch": 0.0007075365506949939,
+      "grad_norm": 2.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.487,
+      "step": 110
+    },
+    {
+      "epoch": 0.0007718580553036297,
+      "grad_norm": 2.0,
+      "learning_rate": 2e-05,
+      "loss": 0.4784,
+      "step": 120
+    },
+    {
+      "epoch": 0.0008361795599122655,
+      "grad_norm": 2.3125,
+      "learning_rate": 2e-05,
+      "loss": 0.4745,
+      "step": 130
+    },
+    {
+      "epoch": 0.0009005010645209013,
+      "grad_norm": 2.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.4811,
+      "step": 140
+    },
+    {
+      "epoch": 0.0009648225691295371,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.4681,
+      "step": 150
+    },
+    {
+      "epoch": 0.001029144073738173,
+      "grad_norm": 1.9921875,
+      "learning_rate": 2e-05,
+      "loss": 0.4657,
+      "step": 160
+    },
+    {
+      "epoch": 0.0010934655783468087,
+      "grad_norm": 2.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.4622,
+      "step": 170
+    },
+    {
+      "epoch": 0.0011577870829554446,
+      "grad_norm": 2.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.4551,
+      "step": 180
+    },
+    {
+      "epoch": 0.0012221085875640803,
+      "grad_norm": 2.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.4548,
+      "step": 190
+    },
+    {
+      "epoch": 0.0012864300921727162,
+      "grad_norm": 2.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.443,
+      "step": 200
+    },
+    {
+      "epoch": 0.0013507515967813518,
+      "grad_norm": 3.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.4357,
+      "step": 210
+    },
+    {
+      "epoch": 0.0014150731013899878,
+      "grad_norm": 3.296875,
+      "learning_rate": 2e-05,
+      "loss": 0.4224,
+      "step": 220
+    },
+    {
+      "epoch": 0.0014793946059986234,
+      "grad_norm": 2.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.4226,
+      "step": 230
+    },
+    {
+      "epoch": 0.0015437161106072593,
+      "grad_norm": 2.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.4331,
+      "step": 240
+    },
+    {
+      "epoch": 0.001608037615215895,
+      "grad_norm": 3.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.4166,
+      "step": 250
+    },
+    {
+      "epoch": 0.001672359119824531,
+      "grad_norm": 2.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.4132,
+      "step": 260
+    },
+    {
+      "epoch": 0.0017366806244331668,
+      "grad_norm": 2.234375,
+      "learning_rate": 2e-05,
+      "loss": 0.4078,
+      "step": 270
+    },
+    {
+      "epoch": 0.0018010021290418025,
+      "grad_norm": 3.28125,
+      "learning_rate": 2e-05,
+      "loss": 0.4001,
+      "step": 280
+    },
+    {
+      "epoch": 0.0018653236336504384,
+      "grad_norm": 3.25,
+      "learning_rate": 2e-05,
+      "loss": 0.3942,
+      "step": 290
+    },
+    {
+      "epoch": 0.0019296451382590741,
+      "grad_norm": 3.25,
+      "learning_rate": 2e-05,
+      "loss": 0.3827,
+      "step": 300
+    },
+    {
+      "epoch": 0.00199396664286771,
+      "grad_norm": 3.109375,
+      "learning_rate": 2e-05,
+      "loss": 0.3871,
+      "step": 310
+    },
+    {
+      "epoch": 0.002058288147476346,
+      "grad_norm": 2.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.3694,
+      "step": 320
+    },
+    {
+      "epoch": 0.0021226096520849816,
+      "grad_norm": 2.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.377,
+      "step": 330
+    },
+    {
+      "epoch": 0.0021869311566936173,
+      "grad_norm": 2.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.3603,
+      "step": 340
+    },
+    {
+      "epoch": 0.002251252661302253,
+      "grad_norm": 3.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.3604,
+      "step": 350
+    },
+    {
+      "epoch": 0.002315574165910889,
+      "grad_norm": 5.3125,
+      "learning_rate": 2e-05,
+      "loss": 0.3606,
+      "step": 360
+    },
+    {
+      "epoch": 0.002379895670519525,
+      "grad_norm": 2.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.3552,
+      "step": 370
+    },
+    {
+      "epoch": 0.0024442171751281605,
+      "grad_norm": 4.09375,
+      "learning_rate": 2e-05,
+      "loss": 0.3466,
+      "step": 380
+    },
+    {
+      "epoch": 0.002508538679736796,
+      "grad_norm": 3.21875,
+      "learning_rate": 2e-05,
+      "loss": 0.3506,
+      "step": 390
+    },
+    {
+      "epoch": 0.0025728601843454323,
+      "grad_norm": 3.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.3492,
+      "step": 400
+    },
+    {
+      "epoch": 0.002637181688954068,
+      "grad_norm": 3.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.3467,
+      "step": 410
+    },
+    {
+      "epoch": 0.0027015031935627037,
+      "grad_norm": 3.0,
+      "learning_rate": 2e-05,
+      "loss": 0.3269,
+      "step": 420
+    },
+    {
+      "epoch": 0.00276582469817134,
+      "grad_norm": 2.90625,
+      "learning_rate": 2e-05,
+      "loss": 0.3392,
+      "step": 430
+    },
+    {
+      "epoch": 0.0028301462027799755,
+      "grad_norm": 3.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.3333,
+      "step": 440
+    },
+    {
+      "epoch": 0.002894467707388611,
+      "grad_norm": 3.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.3236,
+      "step": 450
+    },
+    {
+      "epoch": 0.002958789211997247,
+      "grad_norm": 3.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.3229,
+      "step": 460
+    },
+    {
+      "epoch": 0.003023110716605883,
+      "grad_norm": 3.3125,
+      "learning_rate": 2e-05,
+      "loss": 0.321,
+      "step": 470
+    },
+    {
+      "epoch": 0.0030874322212145187,
+      "grad_norm": 2.921875,
+      "learning_rate": 2e-05,
+      "loss": 0.3187,
+      "step": 480
+    },
+    {
+      "epoch": 0.0031517537258231544,
+      "grad_norm": 2.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.3205,
+      "step": 490
+    },
+    {
+      "epoch": 0.00321607523043179,
+      "grad_norm": 4.09375,
+      "learning_rate": 2e-05,
+      "loss": 0.3166,
+      "step": 500
+    },
+    {
+      "epoch": 0.003280396735040426,
+      "grad_norm": 2.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.3093,
+      "step": 510
+    },
+    {
+      "epoch": 0.003344718239649062,
+      "grad_norm": 3.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.3118,
+      "step": 520
+    },
+    {
+      "epoch": 0.0034090397442576976,
+      "grad_norm": 3.125,
+      "learning_rate": 2e-05,
+      "loss": 0.3092,
+      "step": 530
+    },
+    {
+      "epoch": 0.0034733612488663337,
+      "grad_norm": 2.96875,
+      "learning_rate": 2e-05,
+      "loss": 0.3123,
+      "step": 540
+    },
+    {
+      "epoch": 0.0035376827534749694,
+      "grad_norm": 2.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.3019,
+      "step": 550
+    },
+    {
+      "epoch": 0.003602004258083605,
+      "grad_norm": 2.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.301,
+      "step": 560
+    },
+    {
+      "epoch": 0.0036663257626922408,
+      "grad_norm": 2.265625,
+      "learning_rate": 2e-05,
+      "loss": 0.3011,
+      "step": 570
+    },
+    {
+      "epoch": 0.003730647267300877,
+      "grad_norm": 3.21875,
+      "learning_rate": 2e-05,
+      "loss": 0.2949,
+      "step": 580
+    },
+    {
+      "epoch": 0.0037949687719095126,
+      "grad_norm": 2.96875,
+      "learning_rate": 2e-05,
+      "loss": 0.3023,
+      "step": 590
+    },
+    {
+      "epoch": 0.0038592902765181483,
+      "grad_norm": 3.109375,
+      "learning_rate": 2e-05,
+      "loss": 0.3025,
+      "step": 600
+    },
+    {
+      "epoch": 0.003923611781126784,
+      "grad_norm": 3.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.297,
+      "step": 610
+    },
+    {
+      "epoch": 0.00398793328573542,
+      "grad_norm": 2.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2962,
+      "step": 620
+    },
+    {
+      "epoch": 0.004052254790344056,
+      "grad_norm": 3.171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2929,
+      "step": 630
+    },
+    {
+      "epoch": 0.004116576294952692,
+      "grad_norm": 3.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.3011,
+      "step": 640
+    },
+    {
+      "epoch": 0.004180897799561327,
+      "grad_norm": 3.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2968,
+      "step": 650
+    },
+    {
+      "epoch": 0.004245219304169963,
+      "grad_norm": 2.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2969,
+      "step": 660
+    },
+    {
+      "epoch": 0.004309540808778599,
+      "grad_norm": 2.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2965,
+      "step": 670
+    },
+    {
+      "epoch": 0.004373862313387235,
+      "grad_norm": 2.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.2882,
+      "step": 680
+    },
+    {
+      "epoch": 0.004438183817995871,
+      "grad_norm": 2.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2901,
+      "step": 690
+    },
+    {
+      "epoch": 0.004502505322604506,
+      "grad_norm": 2.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2945,
+      "step": 700
+    },
+    {
+      "epoch": 0.004566826827213142,
+      "grad_norm": 3.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2879,
+      "step": 710
+    },
+    {
+      "epoch": 0.004631148331821778,
+      "grad_norm": 3.5,
+      "learning_rate": 2e-05,
+      "loss": 0.2824,
+      "step": 720
+    },
+    {
+      "epoch": 0.0046954698364304135,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2934,
+      "step": 730
+    },
+    {
+      "epoch": 0.00475979134103905,
+      "grad_norm": 3.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2977,
+      "step": 740
+    },
+    {
+      "epoch": 0.004824112845647686,
+      "grad_norm": 2.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2924,
+      "step": 750
+    },
+    {
+      "epoch": 0.004888434350256321,
+      "grad_norm": 2.375,
+      "learning_rate": 2e-05,
+      "loss": 0.287,
+      "step": 760
+    },
+    {
+      "epoch": 0.004952755854864957,
+      "grad_norm": 2.9375,
+      "learning_rate": 2e-05,
+      "loss": 0.2881,
+      "step": 770
+    },
+    {
+      "epoch": 0.005017077359473592,
+      "grad_norm": 2.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.289,
+      "step": 780
+    },
+    {
+      "epoch": 0.0050813988640822285,
+      "grad_norm": 2.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2873,
+      "step": 790
+    },
+    {
+      "epoch": 0.005145720368690865,
+      "grad_norm": 3.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.2805,
+      "step": 800
+    },
+    {
+      "epoch": 0.0052100418732995,
+      "grad_norm": 2.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2847,
+      "step": 810
+    },
+    {
+      "epoch": 0.005274363377908136,
+      "grad_norm": 2.9375,
+      "learning_rate": 2e-05,
+      "loss": 0.2881,
+      "step": 820
+    },
+    {
+      "epoch": 0.005338684882516772,
+      "grad_norm": 2.296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2876,
+      "step": 830
+    },
+    {
+      "epoch": 0.005403006387125407,
+      "grad_norm": 2.828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2773,
+      "step": 840
+    },
+    {
+      "epoch": 0.0054673278917340435,
+      "grad_norm": 2.9375,
+      "learning_rate": 2e-05,
+      "loss": 0.2873,
+      "step": 850
+    },
+    {
+      "epoch": 0.00553164939634268,
+      "grad_norm": 2.625,
+      "learning_rate": 2e-05,
+      "loss": 0.284,
+      "step": 860
+    },
+    {
+      "epoch": 0.005595970900951315,
+      "grad_norm": 3.109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2856,
+      "step": 870
+    },
+    {
+      "epoch": 0.005660292405559951,
+      "grad_norm": 2.375,
+      "learning_rate": 2e-05,
+      "loss": 0.2769,
+      "step": 880
+    },
+    {
+      "epoch": 0.005724613910168586,
+      "grad_norm": 2.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.2811,
+      "step": 890
+    },
+    {
+      "epoch": 0.005788935414777222,
+      "grad_norm": 2.203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2757,
+      "step": 900
+    },
+    {
+      "epoch": 0.0058532569193858585,
+      "grad_norm": 2.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.2798,
+      "step": 910
+    },
+    {
+      "epoch": 0.005917578423994494,
+      "grad_norm": 2.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.2801,
+      "step": 920
+    },
+    {
+      "epoch": 0.00598189992860313,
+      "grad_norm": 2.25,
+      "learning_rate": 2e-05,
+      "loss": 0.285,
+      "step": 930
+    },
+    {
+      "epoch": 0.006046221433211766,
+      "grad_norm": 2.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2713,
+      "step": 940
+    },
+    {
+      "epoch": 0.006110542937820401,
+      "grad_norm": 2.3125,
+      "learning_rate": 2e-05,
+      "loss": 0.2736,
+      "step": 950
+    },
+    {
+      "epoch": 0.006174864442429037,
+      "grad_norm": 2.234375,
+      "learning_rate": 2e-05,
+      "loss": 0.279,
+      "step": 960
+    },
+    {
+      "epoch": 0.0062391859470376735,
+      "grad_norm": 2.375,
+      "learning_rate": 2e-05,
+      "loss": 0.2795,
+      "step": 970
+    },
+    {
+      "epoch": 0.006303507451646309,
+      "grad_norm": 3.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2772,
+      "step": 980
+    },
+    {
+      "epoch": 0.006367828956254945,
+      "grad_norm": 2.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.2763,
+      "step": 990
+    },
+    {
+      "epoch": 0.00643215046086358,
+      "grad_norm": 2.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2732,
+      "step": 1000
+    },
+    {
+      "epoch": 0.006496471965472216,
+      "grad_norm": 2.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2658,
+      "step": 1010
+    },
+    {
+      "epoch": 0.006560793470080852,
+      "grad_norm": 2.203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2762,
+      "step": 1020
+    },
+    {
+      "epoch": 0.006625114974689488,
+      "grad_norm": 2.5,
+      "learning_rate": 2e-05,
+      "loss": 0.2647,
+      "step": 1030
+    },
+    {
+      "epoch": 0.006689436479298124,
+      "grad_norm": 2.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2738,
+      "step": 1040
+    },
+    {
+      "epoch": 0.00675375798390676,
+      "grad_norm": 2.171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2768,
+      "step": 1050
+    },
+    {
+      "epoch": 0.006818079488515395,
+      "grad_norm": 2.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.2667,
+      "step": 1060
+    },
+    {
+      "epoch": 0.006882400993124031,
+      "grad_norm": 2.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2734,
+      "step": 1070
+    },
+    {
+      "epoch": 0.006946722497732667,
+      "grad_norm": 2.3125,
+      "learning_rate": 2e-05,
+      "loss": 0.27,
+      "step": 1080
+    },
+    {
+      "epoch": 0.007011044002341303,
+      "grad_norm": 3.09375,
+      "learning_rate": 2e-05,
+      "loss": 0.2683,
+      "step": 1090
+    },
+    {
+      "epoch": 0.007075365506949939,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2683,
+      "step": 1100
+    },
+    {
+      "epoch": 0.007139687011558574,
+      "grad_norm": 2.34375,
+      "learning_rate": 2e-05,
+      "loss": 0.2715,
+      "step": 1110
+    },
+    {
+      "epoch": 0.00720400851616721,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2756,
+      "step": 1120
+    },
+    {
+      "epoch": 0.007268330020775846,
+      "grad_norm": 2.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2689,
+      "step": 1130
+    },
+    {
+      "epoch": 0.0073326515253844815,
+      "grad_norm": 2.1875,
+      "learning_rate": 2e-05,
+      "loss": 0.2684,
+      "step": 1140
+    },
+    {
+      "epoch": 0.007396973029993118,
+      "grad_norm": 1.9375,
+      "learning_rate": 2e-05,
+      "loss": 0.2662,
+      "step": 1150
+    },
+    {
+      "epoch": 0.007461294534601754,
+      "grad_norm": 2.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2605,
+      "step": 1160
+    },
+    {
+      "epoch": 0.007525616039210389,
+      "grad_norm": 1.8671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2664,
+      "step": 1170
+    },
+    {
+      "epoch": 0.007589937543819025,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.2659,
+      "step": 1180
+    },
+    {
+      "epoch": 0.007654259048427661,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2703,
+      "step": 1190
+    },
+    {
+      "epoch": 0.0077185805530362965,
+      "grad_norm": 2.921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2714,
+      "step": 1200
+    },
+    {
+      "epoch": 0.007782902057644933,
+      "grad_norm": 3.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2651,
+      "step": 1210
+    },
+    {
+      "epoch": 0.007847223562253569,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2624,
+      "step": 1220
+    },
+    {
+      "epoch": 0.007911545066862204,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2681,
+      "step": 1230
+    },
+    {
+      "epoch": 0.00797586657147084,
+      "grad_norm": 2.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2654,
+      "step": 1240
+    },
+    {
+      "epoch": 0.008040188076079476,
+      "grad_norm": 2.375,
+      "learning_rate": 2e-05,
+      "loss": 0.2575,
+      "step": 1250
+    },
+    {
+      "epoch": 0.008104509580688112,
+      "grad_norm": 2.21875,
+      "learning_rate": 2e-05,
+      "loss": 0.255,
+      "step": 1260
+    },
+    {
+      "epoch": 0.008168831085296747,
+      "grad_norm": 2.375,
+      "learning_rate": 2e-05,
+      "loss": 0.2593,
+      "step": 1270
+    },
+    {
+      "epoch": 0.008233152589905384,
+      "grad_norm": 2.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.2623,
+      "step": 1280
+    },
+    {
+      "epoch": 0.008297474094514019,
+      "grad_norm": 2.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2591,
+      "step": 1290
+    },
+    {
+      "epoch": 0.008361795599122654,
+      "grad_norm": 2.1875,
+      "learning_rate": 2e-05,
+      "loss": 0.2573,
+      "step": 1300
+    },
+    {
+      "epoch": 0.008426117103731291,
+      "grad_norm": 2.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2611,
+      "step": 1310
+    },
+    {
+      "epoch": 0.008490438608339927,
+      "grad_norm": 2.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2591,
+      "step": 1320
+    },
+    {
+      "epoch": 0.008554760112948562,
+      "grad_norm": 2.234375,
+      "learning_rate": 2e-05,
+      "loss": 0.2599,
+      "step": 1330
+    },
+    {
+      "epoch": 0.008619081617557199,
+      "grad_norm": 2.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.259,
+      "step": 1340
+    },
+    {
+      "epoch": 0.008683403122165834,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2661,
+      "step": 1350
+    },
+    {
+      "epoch": 0.00874772462677447,
+      "grad_norm": 2.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.2639,
+      "step": 1360
+    },
+    {
+      "epoch": 0.008812046131383105,
+      "grad_norm": 2.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2621,
+      "step": 1370
+    },
+    {
+      "epoch": 0.008876367635991742,
+      "grad_norm": 2.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2537,
+      "step": 1380
+    },
+    {
+      "epoch": 0.008940689140600377,
+      "grad_norm": 2.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2596,
+      "step": 1390
+    },
+    {
+      "epoch": 0.009005010645209012,
+      "grad_norm": 2.09375,
+      "learning_rate": 2e-05,
+      "loss": 0.255,
+      "step": 1400
+    },
+    {
+      "epoch": 0.009069332149817649,
+      "grad_norm": 2.171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2601,
+      "step": 1410
+    },
+    {
+      "epoch": 0.009133653654426284,
+      "grad_norm": 1.9140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2516,
+      "step": 1420
+    },
+    {
+      "epoch": 0.00919797515903492,
+      "grad_norm": 2.171875,
+      "learning_rate": 2e-05,
+      "loss": 0.26,
+      "step": 1430
+    },
+    {
+      "epoch": 0.009262296663643556,
+      "grad_norm": 2.375,
+      "learning_rate": 2e-05,
+      "loss": 0.2524,
+      "step": 1440
+    },
+    {
+      "epoch": 0.009326618168252192,
+      "grad_norm": 2.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.2551,
+      "step": 1450
+    },
+    {
+      "epoch": 0.009390939672860827,
+      "grad_norm": 2.296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2593,
+      "step": 1460
+    },
+    {
+      "epoch": 0.009455261177469464,
+      "grad_norm": 2.203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2612,
+      "step": 1470
+    },
+    {
+      "epoch": 0.0095195826820781,
+      "grad_norm": 2.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2579,
+      "step": 1480
+    },
+    {
+      "epoch": 0.009583904186686735,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2564,
+      "step": 1490
+    },
+    {
+      "epoch": 0.009648225691295371,
+      "grad_norm": 2.25,
+      "learning_rate": 2e-05,
+      "loss": 0.252,
+      "step": 1500
+    },
+    {
+      "epoch": 0.009712547195904007,
+      "grad_norm": 2.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.259,
+      "step": 1510
+    },
+    {
+      "epoch": 0.009776868700512642,
+      "grad_norm": 2.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.2543,
+      "step": 1520
+    },
+    {
+      "epoch": 0.009841190205121279,
+      "grad_norm": 2.203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2513,
+      "step": 1530
+    },
+    {
+      "epoch": 0.009905511709729914,
+      "grad_norm": 2.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2601,
+      "step": 1540
+    },
+    {
+      "epoch": 0.00996983321433855,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.2475,
+      "step": 1550
+    },
+    {
+      "epoch": 0.010034154718947185,
+      "grad_norm": 2.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.2579,
+      "step": 1560
+    },
+    {
+      "epoch": 0.010098476223555822,
+      "grad_norm": 2.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2547,
+      "step": 1570
+    },
+    {
+      "epoch": 0.010162797728164457,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.2558,
+      "step": 1580
+    },
+    {
+      "epoch": 0.010227119232773092,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2532,
+      "step": 1590
+    },
+    {
+      "epoch": 0.01029144073738173,
+      "grad_norm": 1.96875,
+      "learning_rate": 2e-05,
+      "loss": 0.249,
+      "step": 1600
+    },
+    {
+      "epoch": 0.010355762241990365,
+      "grad_norm": 2.28125,
+      "learning_rate": 2e-05,
+      "loss": 0.2578,
+      "step": 1610
+    },
+    {
+      "epoch": 0.010420083746599,
+      "grad_norm": 2.28125,
+      "learning_rate": 2e-05,
+      "loss": 0.2536,
+      "step": 1620
+    },
+    {
+      "epoch": 0.010484405251207637,
+      "grad_norm": 2.78125,
+      "learning_rate": 2e-05,
+      "loss": 0.2551,
+      "step": 1630
+    },
+    {
+      "epoch": 0.010548726755816272,
+      "grad_norm": 1.9921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2538,
+      "step": 1640
+    },
+    {
+      "epoch": 0.010613048260424907,
+      "grad_norm": 2.5,
+      "learning_rate": 2e-05,
+      "loss": 0.2616,
+      "step": 1650
+    },
+    {
+      "epoch": 0.010677369765033544,
+      "grad_norm": 2.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2421,
+      "step": 1660
+    },
+    {
+      "epoch": 0.01074169126964218,
+      "grad_norm": 2.34375,
+      "learning_rate": 2e-05,
+      "loss": 0.2531,
+      "step": 1670
+    },
+    {
+      "epoch": 0.010806012774250815,
+      "grad_norm": 2.265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2531,
+      "step": 1680
+    },
+    {
+      "epoch": 0.010870334278859452,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2456,
+      "step": 1690
+    },
+    {
+      "epoch": 0.010934655783468087,
+      "grad_norm": 2.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2455,
+      "step": 1700
+    },
+    {
+      "epoch": 0.010998977288076722,
+      "grad_norm": 1.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.2521,
+      "step": 1710
+    },
+    {
+      "epoch": 0.01106329879268536,
+      "grad_norm": 2.625,
+      "learning_rate": 2e-05,
+      "loss": 0.2469,
+      "step": 1720
+    },
+    {
+      "epoch": 0.011127620297293995,
+      "grad_norm": 2.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.2522,
+      "step": 1730
+    },
+    {
+      "epoch": 0.01119194180190263,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2552,
+      "step": 1740
+    },
+    {
+      "epoch": 0.011256263306511267,
+      "grad_norm": 1.8671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2596,
+      "step": 1750
+    },
+    {
+      "epoch": 0.011320584811119902,
+      "grad_norm": 3.984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2493,
+      "step": 1760
+    },
+    {
+      "epoch": 0.011384906315728537,
+      "grad_norm": 1.90625,
+      "learning_rate": 2e-05,
+      "loss": 0.2504,
+      "step": 1770
+    },
+    {
+      "epoch": 0.011449227820337173,
+      "grad_norm": 2.1875,
+      "learning_rate": 2e-05,
+      "loss": 0.2472,
+      "step": 1780
+    },
+    {
+      "epoch": 0.01151354932494581,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2513,
+      "step": 1790
+    },
+    {
+      "epoch": 0.011577870829554445,
+      "grad_norm": 2.796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2456,
+      "step": 1800
+    },
+    {
+      "epoch": 0.01164219233416308,
+      "grad_norm": 3.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2467,
+      "step": 1810
+    },
+    {
+      "epoch": 0.011706513838771717,
+      "grad_norm": 1.921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2566,
+      "step": 1820
+    },
+    {
+      "epoch": 0.011770835343380352,
+      "grad_norm": 2.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2547,
+      "step": 1830
+    },
+    {
+      "epoch": 0.011835156847988988,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2538,
+      "step": 1840
+    },
+    {
+      "epoch": 0.011899478352597625,
+      "grad_norm": 2.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2501,
+      "step": 1850
+    },
+    {
+      "epoch": 0.01196379985720626,
+      "grad_norm": 2.109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2499,
+      "step": 1860
+    },
+    {
+      "epoch": 0.012028121361814895,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.2383,
+      "step": 1870
+    },
+    {
+      "epoch": 0.012092442866423532,
+      "grad_norm": 1.9765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2397,
+      "step": 1880
+    },
+    {
+      "epoch": 0.012156764371032167,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2449,
+      "step": 1890
+    },
+    {
+      "epoch": 0.012221085875640803,
+      "grad_norm": 2.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2503,
+      "step": 1900
+    },
+    {
+      "epoch": 0.01228540738024944,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2447,
+      "step": 1910
+    },
+    {
+      "epoch": 0.012349728884858075,
+      "grad_norm": 2.15625,
+      "learning_rate": 2e-05,
+      "loss": 0.2509,
+      "step": 1920
+    },
+    {
+      "epoch": 0.01241405038946671,
+      "grad_norm": 2.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2462,
+      "step": 1930
+    },
+    {
+      "epoch": 0.012478371894075347,
+      "grad_norm": 2.203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2487,
+      "step": 1940
+    },
+    {
+      "epoch": 0.012542693398683982,
+      "grad_norm": 2.953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2439,
+      "step": 1950
+    },
+    {
+      "epoch": 0.012607014903292618,
+      "grad_norm": 2.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.2536,
+      "step": 1960
+    },
+    {
+      "epoch": 0.012671336407901253,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2498,
+      "step": 1970
+    },
+    {
+      "epoch": 0.01273565791250989,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2521,
+      "step": 1980
+    },
+    {
+      "epoch": 0.012799979417118525,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2522,
+      "step": 1990
+    },
+    {
+      "epoch": 0.01286430092172716,
+      "grad_norm": 2.25,
+      "learning_rate": 2e-05,
+      "loss": 0.2494,
+      "step": 2000
+    },
+    {
+      "epoch": 0.012928622426335797,
+      "grad_norm": 2.34375,
+      "learning_rate": 2e-05,
+      "loss": 0.2402,
+      "step": 2010
+    },
+    {
+      "epoch": 0.012992943930944433,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2477,
+      "step": 2020
+    },
+    {
+      "epoch": 0.013057265435553068,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2429,
+      "step": 2030
+    },
+    {
+      "epoch": 0.013121586940161705,
+      "grad_norm": 2.25,
+      "learning_rate": 2e-05,
+      "loss": 0.2416,
+      "step": 2040
+    },
+    {
+      "epoch": 0.01318590844477034,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.2443,
+      "step": 2050
+    },
+    {
+      "epoch": 0.013250229949378975,
+      "grad_norm": 2.953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2457,
+      "step": 2060
+    },
+    {
+      "epoch": 0.013314551453987612,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2428,
+      "step": 2070
+    },
+    {
+      "epoch": 0.013378872958596248,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2491,
+      "step": 2080
+    },
+    {
+      "epoch": 0.013443194463204883,
+      "grad_norm": 2.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2471,
+      "step": 2090
+    },
+    {
+      "epoch": 0.01350751596781352,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2398,
+      "step": 2100
+    },
+    {
+      "epoch": 0.013571837472422155,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2489,
+      "step": 2110
+    },
+    {
+      "epoch": 0.01363615897703079,
+      "grad_norm": 1.78125,
+      "learning_rate": 2e-05,
+      "loss": 0.2425,
+      "step": 2120
+    },
+    {
+      "epoch": 0.013700480481639427,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.244,
+      "step": 2130
+    },
+    {
+      "epoch": 0.013764801986248063,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2e-05,
+      "loss": 0.24,
+      "step": 2140
+    },
+    {
+      "epoch": 0.013829123490856698,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2399,
+      "step": 2150
+    },
+    {
+      "epoch": 0.013893444995465335,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.2418,
+      "step": 2160
+    },
+    {
+      "epoch": 0.01395776650007397,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.2427,
+      "step": 2170
+    },
+    {
+      "epoch": 0.014022088004682605,
+      "grad_norm": 2.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.2403,
+      "step": 2180
+    },
+    {
+      "epoch": 0.01408640950929124,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2457,
+      "step": 2190
+    },
+    {
+      "epoch": 0.014150731013899878,
+      "grad_norm": 1.96875,
+      "learning_rate": 2e-05,
+      "loss": 0.2491,
+      "step": 2200
+    },
+    {
+      "epoch": 0.014419093228245763,
+      "grad_norm": 1.953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2406,
+      "step": 2210
+    },
+    {
+      "epoch": 0.014484337993984433,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.2421,
+      "step": 2220
+    },
+    {
+      "epoch": 0.0145495827597231,
+      "grad_norm": 2.203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2342,
+      "step": 2230
+    },
+    {
+      "epoch": 0.01461482752546177,
+      "grad_norm": 2.234375,
+      "learning_rate": 2e-05,
+      "loss": 0.2452,
+      "step": 2240
+    },
+    {
+      "epoch": 0.014680072291200438,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2409,
+      "step": 2250
+    },
+    {
+      "epoch": 0.014745317056939107,
+      "grad_norm": 2.09375,
+      "learning_rate": 2e-05,
+      "loss": 0.2417,
+      "step": 2260
+    },
+    {
+      "epoch": 0.014810561822677776,
+      "grad_norm": 2.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2374,
+      "step": 2270
+    },
+    {
+      "epoch": 0.014875806588416444,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2435,
+      "step": 2280
+    },
+    {
+      "epoch": 0.014941051354155114,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2358,
+      "step": 2290
+    },
+    {
+      "epoch": 0.015006296119893781,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.2439,
+      "step": 2300
+    },
+    {
+      "epoch": 0.01507154088563245,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2432,
+      "step": 2310
+    },
+    {
+      "epoch": 0.015136785651371118,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2418,
+      "step": 2320
+    },
+    {
+      "epoch": 0.015202030417109788,
+      "grad_norm": 2.25,
+      "learning_rate": 2e-05,
+      "loss": 0.235,
+      "step": 2330
+    },
+    {
+      "epoch": 0.015267275182848455,
+      "grad_norm": 2.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2367,
+      "step": 2340
+    },
+    {
+      "epoch": 0.015332519948587125,
+      "grad_norm": 2.09375,
+      "learning_rate": 2e-05,
+      "loss": 0.2425,
+      "step": 2350
+    },
+    {
+      "epoch": 0.015397764714325793,
+      "grad_norm": 2.109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2417,
+      "step": 2360
+    },
+    {
+      "epoch": 0.015463009480064462,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2352,
+      "step": 2370
+    },
+    {
+      "epoch": 0.01552825424580313,
+      "grad_norm": 2.3125,
+      "learning_rate": 2e-05,
+      "loss": 0.2418,
+      "step": 2380
+    },
+    {
+      "epoch": 0.015593499011541799,
+      "grad_norm": 2.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2358,
+      "step": 2390
+    },
+    {
+      "epoch": 0.01565874377728047,
+      "grad_norm": 1.90625,
+      "learning_rate": 2e-05,
+      "loss": 0.2424,
+      "step": 2400
+    },
+    {
+      "epoch": 0.015723988543019138,
+      "grad_norm": 2.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2359,
+      "step": 2410
+    },
+    {
+      "epoch": 0.015789233308757804,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2435,
+      "step": 2420
+    },
+    {
+      "epoch": 0.015854478074496473,
+      "grad_norm": 2.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2349,
+      "step": 2430
+    },
+    {
+      "epoch": 0.015919722840235143,
+      "grad_norm": 1.8359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2382,
+      "step": 2440
+    },
+    {
+      "epoch": 0.015984967605973812,
+      "grad_norm": 1.9453125,
+      "learning_rate": 2e-05,
+      "loss": 0.2419,
+      "step": 2450
+    },
+    {
+      "epoch": 0.016050212371712478,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2395,
+      "step": 2460
+    },
+    {
+      "epoch": 0.016115457137451147,
+      "grad_norm": 2.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2358,
+      "step": 2470
+    },
+    {
+      "epoch": 0.016180701903189817,
+      "grad_norm": 2.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2371,
+      "step": 2480
+    },
+    {
+      "epoch": 0.016245946668928486,
+      "grad_norm": 2.171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2401,
+      "step": 2490
+    },
+    {
+      "epoch": 0.016311191434667152,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2e-05,
+      "loss": 0.2348,
+      "step": 2500
+    },
+    {
+      "epoch": 0.01637643620040582,
+      "grad_norm": 1.8359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2318,
+      "step": 2510
+    },
+    {
+      "epoch": 0.01644168096614449,
+      "grad_norm": 2.1875,
+      "learning_rate": 2e-05,
+      "loss": 0.2375,
+      "step": 2520
+    },
+    {
+      "epoch": 0.01650692573188316,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2409,
+      "step": 2530
+    },
+    {
+      "epoch": 0.01657217049762183,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2397,
+      "step": 2540
+    },
+    {
+      "epoch": 0.016637415263360496,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2346,
+      "step": 2550
+    },
+    {
+      "epoch": 0.016702660029099165,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2462,
+      "step": 2560
+    },
+    {
+      "epoch": 0.016767904794837835,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2367,
+      "step": 2570
+    },
+    {
+      "epoch": 0.016833149560576504,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2401,
+      "step": 2580
+    },
+    {
+      "epoch": 0.01689839432631517,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2372,
+      "step": 2590
+    },
+    {
+      "epoch": 0.01696363909205384,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2372,
+      "step": 2600
+    },
+    {
+      "epoch": 0.01702888385779251,
+      "grad_norm": 2.25,
+      "learning_rate": 2e-05,
+      "loss": 0.2307,
+      "step": 2610
+    },
+    {
+      "epoch": 0.017094128623531178,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2411,
+      "step": 2620
+    },
+    {
+      "epoch": 0.017159373389269844,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2332,
+      "step": 2630
+    },
+    {
+      "epoch": 0.017224618155008514,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2387,
+      "step": 2640
+    },
+    {
+      "epoch": 0.017289862920747183,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2399,
+      "step": 2650
+    },
+    {
+      "epoch": 0.017355107686485852,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.2386,
+      "step": 2660
+    },
+    {
+      "epoch": 0.017420352452224522,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2388,
+      "step": 2670
+    },
+    {
+      "epoch": 0.017485597217963188,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2361,
+      "step": 2680
+    },
+    {
+      "epoch": 0.017550841983701857,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.237,
+      "step": 2690
+    },
+    {
+      "epoch": 0.017616086749440527,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.2332,
+      "step": 2700
+    },
+    {
+      "epoch": 0.017739317134478426,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2319,
+      "step": 2710
+    },
+    {
+      "epoch": 0.017804775869291998,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2451,
+      "step": 2720
+    },
+    {
+      "epoch": 0.017870234604105573,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2381,
+      "step": 2730
+    },
+    {
+      "epoch": 0.017935693338919145,
+      "grad_norm": 2.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2286,
+      "step": 2740
+    },
+    {
+      "epoch": 0.018001152073732717,
+      "grad_norm": 2.109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2274,
+      "step": 2750
+    },
+    {
+      "epoch": 0.018066610808546293,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.2359,
+      "step": 2760
+    },
+    {
+      "epoch": 0.018132069543359865,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2348,
+      "step": 2770
+    },
+    {
+      "epoch": 0.01819752827817344,
+      "grad_norm": 2.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2371,
+      "step": 2780
+    },
+    {
+      "epoch": 0.018262987012987012,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2417,
+      "step": 2790
+    },
+    {
+      "epoch": 0.018328445747800588,
+      "grad_norm": 2.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2347,
+      "step": 2800
+    },
+    {
+      "epoch": 0.01839390448261416,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.2355,
+      "step": 2810
+    },
+    {
+      "epoch": 0.018459363217427735,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2e-05,
+      "loss": 0.2443,
+      "step": 2820
+    },
+    {
+      "epoch": 0.018524821952241307,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2333,
+      "step": 2830
+    },
+    {
+      "epoch": 0.01859028068705488,
+      "grad_norm": 1.8671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2326,
+      "step": 2840
+    },
+    {
+      "epoch": 0.018655739421868454,
+      "grad_norm": 2.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2357,
+      "step": 2850
+    },
+    {
+      "epoch": 0.018721198156682026,
+      "grad_norm": 2.203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2298,
+      "step": 2860
+    },
+    {
+      "epoch": 0.018786656891495602,
+      "grad_norm": 1.9453125,
+      "learning_rate": 2e-05,
+      "loss": 0.2286,
+      "step": 2870
+    },
+    {
+      "epoch": 0.018852115626309174,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2e-05,
+      "loss": 0.231,
+      "step": 2880
+    },
+    {
+      "epoch": 0.01891757436112275,
+      "grad_norm": 2.125,
+      "learning_rate": 2e-05,
+      "loss": 0.2347,
+      "step": 2890
+    },
+    {
+      "epoch": 0.01898303309593632,
+      "grad_norm": 1.9140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2294,
+      "step": 2900
+    },
+    {
+      "epoch": 0.019048491830749897,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.2298,
+      "step": 2910
+    },
+    {
+      "epoch": 0.01911395056556347,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.2277,
+      "step": 2920
+    },
+    {
+      "epoch": 0.01917940930037704,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2363,
+      "step": 2930
+    },
+    {
+      "epoch": 0.019244868035190616,
+      "grad_norm": 1.8671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2303,
+      "step": 2940
+    },
+    {
+      "epoch": 0.019310326770004188,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.238,
+      "step": 2950
+    },
+    {
+      "epoch": 0.019375785504817764,
+      "grad_norm": 2.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2296,
+      "step": 2960
+    },
+    {
+      "epoch": 0.019441244239631335,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2328,
+      "step": 2970
+    },
+    {
+      "epoch": 0.01950670297444491,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2331,
+      "step": 2980
+    },
+    {
+      "epoch": 0.019572161709258483,
+      "grad_norm": 2.234375,
+      "learning_rate": 2e-05,
+      "loss": 0.2358,
+      "step": 2990
+    },
+    {
+      "epoch": 0.01963762044407206,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2312,
+      "step": 3000
+    },
+    {
+      "epoch": 0.019741717988574728,
+      "grad_norm": 1.921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2357,
+      "step": 3010
+    },
+    {
+      "epoch": 0.0198073050915268,
+      "grad_norm": 3.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2317,
+      "step": 3020
+    },
+    {
+      "epoch": 0.019872892194478877,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2379,
+      "step": 3030
+    },
+    {
+      "epoch": 0.019938479297430953,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.2291,
+      "step": 3040
+    },
+    {
+      "epoch": 0.02000406640038303,
+      "grad_norm": 1.8359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2318,
+      "step": 3050
+    },
+    {
+      "epoch": 0.020069653503335103,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2317,
+      "step": 3060
+    },
+    {
+      "epoch": 0.02013524060628718,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2308,
+      "step": 3070
+    },
+    {
+      "epoch": 0.020200827709239255,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2313,
+      "step": 3080
+    },
+    {
+      "epoch": 0.02026641481219133,
+      "grad_norm": 1.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.2299,
+      "step": 3090
+    },
+    {
+      "epoch": 0.020332001915143408,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2319,
+      "step": 3100
+    },
+    {
+      "epoch": 0.02039758901809548,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2327,
+      "step": 3110
+    },
+    {
+      "epoch": 0.020463176121047557,
+      "grad_norm": 2.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2278,
+      "step": 3120
+    },
+    {
+      "epoch": 0.020528763223999633,
+      "grad_norm": 2.21875,
+      "learning_rate": 2e-05,
+      "loss": 0.2313,
+      "step": 3130
+    },
+    {
+      "epoch": 0.02059435032695171,
+      "grad_norm": 2.34375,
+      "learning_rate": 2e-05,
+      "loss": 0.2291,
+      "step": 3140
+    },
+    {
+      "epoch": 0.020659937429903782,
+      "grad_norm": 1.921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2295,
+      "step": 3150
+    },
+    {
+      "epoch": 0.02072552453285586,
+      "grad_norm": 1.75,
+      "learning_rate": 2e-05,
+      "loss": 0.2335,
+      "step": 3160
+    },
+    {
+      "epoch": 0.020791111635807935,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2271,
+      "step": 3170
+    },
+    {
+      "epoch": 0.02085669873876001,
+      "grad_norm": 3.34375,
+      "learning_rate": 2e-05,
+      "loss": 0.237,
+      "step": 3180
+    },
+    {
+      "epoch": 0.020922285841712084,
+      "grad_norm": 2.34375,
+      "learning_rate": 2e-05,
+      "loss": 0.2256,
+      "step": 3190
+    },
+    {
+      "epoch": 0.02098787294466416,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2293,
+      "step": 3200
+    },
+    {
+      "epoch": 0.021053460047616237,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2291,
+      "step": 3210
+    },
+    {
+      "epoch": 0.021119047150568313,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2243,
+      "step": 3220
+    },
+    {
+      "epoch": 0.021184634253520386,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2261,
+      "step": 3230
+    },
+    {
+      "epoch": 0.021250221356472462,
+      "grad_norm": 2.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2289,
+      "step": 3240
+    },
+    {
+      "epoch": 0.02131580845942454,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2386,
+      "step": 3250
+    },
+    {
+      "epoch": 0.021381395562376615,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2255,
+      "step": 3260
+    },
+    {
+      "epoch": 0.02144698266532869,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2214,
+      "step": 3270
+    },
+    {
+      "epoch": 0.021512569768280764,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2278,
+      "step": 3280
+    },
+    {
+      "epoch": 0.02157815687123284,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.2233,
+      "step": 3290
+    },
+    {
+      "epoch": 0.021643743974184917,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2244,
+      "step": 3300
+    },
+    {
+      "epoch": 0.021709331077136993,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2291,
+      "step": 3310
+    },
+    {
+      "epoch": 0.021774918180089066,
+      "grad_norm": 2.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2297,
+      "step": 3320
+    },
+    {
+      "epoch": 0.021840505283041142,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2239,
+      "step": 3330
+    },
+    {
+      "epoch": 0.02190609238599322,
+      "grad_norm": 2.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2287,
+      "step": 3340
+    },
+    {
+      "epoch": 0.021971679488945295,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2269,
+      "step": 3350
+    },
+    {
+      "epoch": 0.022037266591897368,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2356,
+      "step": 3360
+    },
+    {
+      "epoch": 0.022102853694849444,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.23,
+      "step": 3370
+    },
+    {
+      "epoch": 0.02216844079780152,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2339,
+      "step": 3380
+    },
+    {
+      "epoch": 0.022234027900753597,
+      "grad_norm": 4.15625,
+      "learning_rate": 2e-05,
+      "loss": 0.2293,
+      "step": 3390
+    },
+    {
+      "epoch": 0.022299615003705673,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2296,
+      "step": 3400
+    },
+    {
+      "epoch": 0.022365202106657746,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2345,
+      "step": 3410
+    },
+    {
+      "epoch": 0.022430789209609822,
+      "grad_norm": 2.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2311,
+      "step": 3420
+    },
+    {
+      "epoch": 0.0224963763125619,
+      "grad_norm": 2.234375,
+      "learning_rate": 2e-05,
+      "loss": 0.2231,
+      "step": 3430
+    },
+    {
+      "epoch": 0.022561963415513975,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2298,
+      "step": 3440
+    },
+    {
+      "epoch": 0.022627550518466048,
+      "grad_norm": 1.9921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2273,
+      "step": 3450
+    },
+    {
+      "epoch": 0.022693137621418124,
+      "grad_norm": 2.171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2333,
+      "step": 3460
+    },
+    {
+      "epoch": 0.0227587247243702,
+      "grad_norm": 1.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2202,
+      "step": 3470
+    },
+    {
+      "epoch": 0.022824311827322277,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2343,
+      "step": 3480
+    },
+    {
+      "epoch": 0.02288989893027435,
+      "grad_norm": 2.296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2285,
+      "step": 3490
+    },
+    {
+      "epoch": 0.022955486033226426,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2276,
+      "step": 3500
+    },
+    {
+      "epoch": 0.20421224109844077,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2247,
+      "step": 3510
+    },
+    {
+      "epoch": 0.2047940423551315,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2226,
+      "step": 3520
+    },
+    {
+      "epoch": 0.2053758436118222,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2225,
+      "step": 3530
+    },
+    {
+      "epoch": 0.20595764486851292,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2274,
+      "step": 3540
+    },
+    {
+      "epoch": 0.20653944612520364,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.2256,
+      "step": 3550
+    },
+    {
+      "epoch": 0.20712124738189436,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.2218,
+      "step": 3560
+    },
+    {
+      "epoch": 0.20770304863858505,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2284,
+      "step": 3570
+    },
+    {
+      "epoch": 0.20828484989527576,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2217,
+      "step": 3580
+    },
+    {
+      "epoch": 0.20886665115196648,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2186,
+      "step": 3590
+    },
+    {
+      "epoch": 0.2094484524086572,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.225,
+      "step": 3600
+    },
+    {
+      "epoch": 0.21003025366534792,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2289,
+      "step": 3610
+    },
+    {
+      "epoch": 0.21061205492203863,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2227,
+      "step": 3620
+    },
+    {
+      "epoch": 0.21119385617872935,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2286,
+      "step": 3630
+    },
+    {
+      "epoch": 0.21177565743542007,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.2192,
+      "step": 3640
+    },
+    {
+      "epoch": 0.21235745869211078,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.227,
+      "step": 3650
+    },
+    {
+      "epoch": 0.2129392599488015,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2275,
+      "step": 3660
+    },
+    {
+      "epoch": 0.2135210612054922,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2183,
+      "step": 3670
+    },
+    {
+      "epoch": 0.2141028624621829,
+      "grad_norm": 2.09375,
+      "learning_rate": 2e-05,
+      "loss": 0.2294,
+      "step": 3680
+    },
+    {
+      "epoch": 0.21468466371887363,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2215,
+      "step": 3690
+    },
+    {
+      "epoch": 0.21526646497556434,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2256,
+      "step": 3700
+    },
+    {
+      "epoch": 0.21584826623225506,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2344,
+      "step": 3710
+    },
+    {
+      "epoch": 0.21643006748894578,
+      "grad_norm": 2.296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2215,
+      "step": 3720
+    },
+    {
+      "epoch": 0.2170118687456365,
+      "grad_norm": 1.9375,
+      "learning_rate": 2e-05,
+      "loss": 0.2224,
+      "step": 3730
+    },
+    {
+      "epoch": 0.2175936700023272,
+      "grad_norm": 2.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2228,
+      "step": 3740
+    },
+    {
+      "epoch": 0.21817547125901793,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2241,
+      "step": 3750
+    },
+    {
+      "epoch": 0.21875727251570865,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2257,
+      "step": 3760
+    },
+    {
+      "epoch": 0.21933907377239936,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.218,
+      "step": 3770
+    },
+    {
+      "epoch": 0.21992087502909005,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2204,
+      "step": 3780
+    },
+    {
+      "epoch": 0.22050267628578077,
+      "grad_norm": 1.96875,
+      "learning_rate": 2e-05,
+      "loss": 0.2274,
+      "step": 3790
+    },
+    {
+      "epoch": 0.22108447754247149,
+      "grad_norm": 2.171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2224,
+      "step": 3800
+    },
+    {
+      "epoch": 0.2216662787991622,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2237,
+      "step": 3810
+    },
+    {
+      "epoch": 0.22224808005585292,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2268,
+      "step": 3820
+    },
+    {
+      "epoch": 0.22282988131254364,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2249,
+      "step": 3830
+    },
+    {
+      "epoch": 0.22341168256923435,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2334,
+      "step": 3840
+    },
+    {
+      "epoch": 0.22399348382592507,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.2211,
+      "step": 3850
+    },
+    {
+      "epoch": 0.2245752850826158,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.2173,
+      "step": 3860
+    },
+    {
+      "epoch": 0.2251570863393065,
+      "grad_norm": 1.78125,
+      "learning_rate": 2e-05,
+      "loss": 0.2233,
+      "step": 3870
+    },
+    {
+      "epoch": 0.2257388875959972,
+      "grad_norm": 2.09375,
+      "learning_rate": 2e-05,
+      "loss": 0.2258,
+      "step": 3880
+    },
+    {
+      "epoch": 0.2263206888526879,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.2222,
+      "step": 3890
+    },
+    {
+      "epoch": 0.22690249010937863,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2223,
+      "step": 3900
+    },
+    {
+      "epoch": 0.22748429136606935,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2226,
+      "step": 3910
+    },
+    {
+      "epoch": 0.22806609262276006,
+      "grad_norm": 1.9375,
+      "learning_rate": 2e-05,
+      "loss": 0.2295,
+      "step": 3920
+    },
+    {
+      "epoch": 0.22864789387945078,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.229,
+      "step": 3930
+    },
+    {
+      "epoch": 0.2292296951361415,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2219,
+      "step": 3940
+    },
+    {
+      "epoch": 0.22981149639283222,
+      "grad_norm": 2.28125,
+      "learning_rate": 2e-05,
+      "loss": 0.2253,
+      "step": 3950
+    },
+    {
+      "epoch": 0.23039329764952293,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2229,
+      "step": 3960
+    },
+    {
+      "epoch": 0.23097509890621365,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2242,
+      "step": 3970
+    },
+    {
+      "epoch": 0.23155690016290434,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2212,
+      "step": 3980
+    },
+    {
+      "epoch": 0.23213870141959506,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.2234,
+      "step": 3990
+    },
+    {
+      "epoch": 0.23272050267628577,
+      "grad_norm": 2.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2216,
+      "step": 4000
+    },
+    {
+      "epoch": 0.2333023039329765,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2213,
+      "step": 4010
+    },
+    {
+      "epoch": 0.2338841051896672,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2243,
+      "step": 4020
+    },
+    {
+      "epoch": 0.23446590644635792,
+      "grad_norm": 3.25,
+      "learning_rate": 2e-05,
+      "loss": 0.2163,
+      "step": 4030
+    },
+    {
+      "epoch": 0.23504770770304864,
+      "grad_norm": 1.75,
+      "learning_rate": 2e-05,
+      "loss": 0.2256,
+      "step": 4040
+    },
+    {
+      "epoch": 0.23562950895973936,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.2178,
+      "step": 4050
+    },
+    {
+      "epoch": 0.23621131021643008,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2191,
+      "step": 4060
+    },
+    {
+      "epoch": 0.2367931114731208,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.2236,
+      "step": 4070
+    },
+    {
+      "epoch": 0.2373749127298115,
+      "grad_norm": 1.8359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2297,
+      "step": 4080
+    },
+    {
+      "epoch": 0.2379567139865022,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.2242,
+      "step": 4090
+    },
+    {
+      "epoch": 0.23853851524319292,
+      "grad_norm": 1.8515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2249,
+      "step": 4100
+    },
+    {
+      "epoch": 0.23912031649988363,
+      "grad_norm": 1.90625,
+      "learning_rate": 2e-05,
+      "loss": 0.2225,
+      "step": 4110
+    },
+    {
+      "epoch": 0.23970211775657435,
+      "grad_norm": 1.984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2232,
+      "step": 4120
+    },
+    {
+      "epoch": 0.24028391901326507,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2235,
+      "step": 4130
+    },
+    {
+      "epoch": 0.24086572026995579,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2183,
+      "step": 4140
+    },
+    {
+      "epoch": 0.2414475215266465,
+      "grad_norm": 1.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2216,
+      "step": 4150
+    },
+    {
+      "epoch": 0.24202932278333722,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2218,
+      "step": 4160
+    },
+    {
+      "epoch": 0.24261112404002794,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2167,
+      "step": 4170
+    },
+    {
+      "epoch": 0.24319292529671865,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2258,
+      "step": 4180
+    },
+    {
+      "epoch": 0.24377472655340934,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2174,
+      "step": 4190
+    },
+    {
+      "epoch": 0.24435652781010006,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2179,
+      "step": 4200
+    },
+    {
+      "epoch": 0.24493832906679078,
+      "grad_norm": 1.953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2234,
+      "step": 4210
+    },
+    {
+      "epoch": 0.2455201303234815,
+      "grad_norm": 1.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2168,
+      "step": 4220
+    },
+    {
+      "epoch": 0.2461019315801722,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2173,
+      "step": 4230
+    },
+    {
+      "epoch": 0.24668373283686293,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2187,
+      "step": 4240
+    },
+    {
+      "epoch": 0.24726553409355365,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2e-05,
+      "loss": 0.2195,
+      "step": 4250
+    },
+    {
+      "epoch": 0.24784733535024436,
+      "grad_norm": 1.828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2169,
+      "step": 4260
+    },
+    {
+      "epoch": 0.24842913660693508,
+      "grad_norm": 2.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.219,
+      "step": 4270
+    },
+    {
+      "epoch": 0.2490109378636258,
+      "grad_norm": 2.21875,
+      "learning_rate": 2e-05,
+      "loss": 0.2176,
+      "step": 4280
+    },
+    {
+      "epoch": 0.2495927391203165,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2157,
+      "step": 4290
+    },
+    {
+      "epoch": 0.2501745403770072,
+      "grad_norm": 2.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2242,
+      "step": 4300
+    },
+    {
+      "epoch": 0.2507563416336979,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2243,
+      "step": 4310
+    },
+    {
+      "epoch": 0.25133814289038864,
+      "grad_norm": 1.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.2192,
+      "step": 4320
+    },
+    {
+      "epoch": 0.25191994414707936,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2216,
+      "step": 4330
+    },
+    {
+      "epoch": 0.2525017454037701,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.2163,
+      "step": 4340
+    },
+    {
+      "epoch": 0.2530835466604608,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.2249,
+      "step": 4350
+    },
+    {
+      "epoch": 0.2536653479171515,
+      "grad_norm": 1.3125,
+      "learning_rate": 2e-05,
+      "loss": 0.2249,
+      "step": 4360
+    },
+    {
+      "epoch": 0.2542471491738422,
+      "grad_norm": 2.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2227,
+      "step": 4370
+    },
+    {
+      "epoch": 0.25482895043053294,
+      "grad_norm": 3.953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2163,
+      "step": 4380
+    },
+    {
+      "epoch": 0.25541075168722366,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2168,
+      "step": 4390
+    },
+    {
+      "epoch": 0.2559925529439144,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2164,
+      "step": 4400
+    },
+    {
+      "epoch": 0.2565743542006051,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2e-05,
+      "loss": 0.213,
+      "step": 4410
+    },
+    {
+      "epoch": 0.2571561554572958,
+      "grad_norm": 1.9765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2203,
+      "step": 4420
+    },
+    {
+      "epoch": 0.2577379567139865,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2154,
+      "step": 4430
+    },
+    {
+      "epoch": 0.25831975797067724,
+      "grad_norm": 1.90625,
+      "learning_rate": 2e-05,
+      "loss": 0.2214,
+      "step": 4440
+    },
+    {
+      "epoch": 0.2589015592273679,
+      "grad_norm": 2.28125,
+      "learning_rate": 2e-05,
+      "loss": 0.2136,
+      "step": 4450
+    },
+    {
+      "epoch": 0.2594833604840586,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2204,
+      "step": 4460
+    },
+    {
+      "epoch": 0.26006516174074934,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.219,
+      "step": 4470
+    },
+    {
+      "epoch": 0.26064696299744006,
+      "grad_norm": 1.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.2112,
+      "step": 4480
+    },
+    {
+      "epoch": 0.2612287642541308,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2184,
+      "step": 4490
+    },
+    {
+      "epoch": 0.2618105655108215,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2219,
+      "step": 4500
+    },
+    {
+      "epoch": 0.2623923667675122,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2205,
+      "step": 4510
+    },
+    {
+      "epoch": 0.2629741680242029,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.216,
+      "step": 4520
+    },
+    {
+      "epoch": 0.26355596928089364,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.2275,
+      "step": 4530
+    },
+    {
+      "epoch": 0.26413777053758436,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2199,
+      "step": 4540
+    },
+    {
+      "epoch": 0.2647195717942751,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.2193,
+      "step": 4550
+    },
+    {
+      "epoch": 0.2653013730509658,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2123,
+      "step": 4560
+    },
+    {
+      "epoch": 0.2658831743076565,
+      "grad_norm": 2.25,
+      "learning_rate": 2e-05,
+      "loss": 0.2244,
+      "step": 4570
+    },
+    {
+      "epoch": 0.26646497556434723,
+      "grad_norm": 2.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2165,
+      "step": 4580
+    },
+    {
+      "epoch": 0.26704677682103795,
+      "grad_norm": 2.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2167,
+      "step": 4590
+    },
+    {
+      "epoch": 0.26762857807772866,
+      "grad_norm": 1.75,
+      "learning_rate": 2e-05,
+      "loss": 0.2146,
+      "step": 4600
+    },
+    {
+      "epoch": 0.2682103793344194,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2166,
+      "step": 4610
+    },
+    {
+      "epoch": 0.2687921805911101,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.2148,
+      "step": 4620
+    },
+    {
+      "epoch": 0.2693739818478008,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2174,
+      "step": 4630
+    },
+    {
+      "epoch": 0.26995578310449153,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2229,
+      "step": 4640
+    },
+    {
+      "epoch": 0.2705375843611822,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.2139,
+      "step": 4650
+    },
+    {
+      "epoch": 0.2711193856178729,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.213,
+      "step": 4660
+    },
+    {
+      "epoch": 0.2717011868745636,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2109,
+      "step": 4670
+    },
+    {
+      "epoch": 0.27228298813125434,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2168,
+      "step": 4680
+    },
+    {
+      "epoch": 0.27286478938794506,
+      "grad_norm": 1.96875,
+      "learning_rate": 2e-05,
+      "loss": 0.2185,
+      "step": 4690
+    },
+    {
+      "epoch": 0.2734465906446358,
+      "grad_norm": 2.171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2209,
+      "step": 4700
+    },
+    {
+      "epoch": 0.2740283919013265,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.2132,
+      "step": 4710
+    },
+    {
+      "epoch": 0.2746101931580172,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2109,
+      "step": 4720
+    },
+    {
+      "epoch": 0.27519199441470793,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2212,
+      "step": 4730
+    },
+    {
+      "epoch": 0.27577379567139865,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2227,
+      "step": 4740
+    },
+    {
+      "epoch": 0.27635559692808936,
+      "grad_norm": 1.8515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2187,
+      "step": 4750
+    },
+    {
+      "epoch": 0.2769373981847801,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2176,
+      "step": 4760
+    },
+    {
+      "epoch": 0.2775191994414708,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2e-05,
+      "loss": 0.217,
+      "step": 4770
+    },
+    {
+      "epoch": 0.2781010006981615,
+      "grad_norm": 1.34375,
+      "learning_rate": 2e-05,
+      "loss": 0.2158,
+      "step": 4780
+    },
+    {
+      "epoch": 0.27868280195485223,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.2147,
+      "step": 4790
+    },
+    {
+      "epoch": 0.27926460321154295,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.221,
+      "step": 4800
+    },
+    {
+      "epoch": 0.27984640446823367,
+      "grad_norm": 1.90625,
+      "learning_rate": 2e-05,
+      "loss": 0.2137,
+      "step": 4810
+    },
+    {
+      "epoch": 0.2804282057249244,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.2216,
+      "step": 4820
+    },
+    {
+      "epoch": 0.2810100069816151,
+      "grad_norm": 1.75,
+      "learning_rate": 2e-05,
+      "loss": 0.2122,
+      "step": 4830
+    },
+    {
+      "epoch": 0.2815918082383058,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2156,
+      "step": 4840
+    },
+    {
+      "epoch": 0.28217360949499654,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2129,
+      "step": 4850
+    },
+    {
+      "epoch": 0.2827554107516872,
+      "grad_norm": 1.265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2167,
+      "step": 4860
+    },
+    {
+      "epoch": 0.2833372120083779,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2152,
+      "step": 4870
+    },
+    {
+      "epoch": 0.28391901326506863,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2119,
+      "step": 4880
+    },
+    {
+      "epoch": 0.28450081452175935,
+      "grad_norm": 1.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2137,
+      "step": 4890
+    },
+    {
+      "epoch": 0.28508261577845007,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2e-05,
+      "loss": 0.22,
+      "step": 4900
+    },
+    {
+      "epoch": 0.2856644170351408,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2184,
+      "step": 4910
+    },
+    {
+      "epoch": 0.2862462182918315,
+      "grad_norm": 1.8515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2243,
+      "step": 4920
+    },
+    {
+      "epoch": 0.2868280195485222,
+      "grad_norm": 1.953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2173,
+      "step": 4930
+    },
+    {
+      "epoch": 0.28740982080521293,
+      "grad_norm": 1.9765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2135,
+      "step": 4940
+    },
+    {
+      "epoch": 0.28799162206190365,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2184,
+      "step": 4950
+    },
+    {
+      "epoch": 0.28857342331859437,
+      "grad_norm": 1.921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2161,
+      "step": 4960
+    },
+    {
+      "epoch": 0.2891552245752851,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2181,
+      "step": 4970
+    },
+    {
+      "epoch": 0.2897370258319758,
+      "grad_norm": 1.9140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2139,
+      "step": 4980
+    },
+    {
+      "epoch": 0.2903188270886665,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2135,
+      "step": 4990
+    },
+    {
+      "epoch": 0.29090062834535724,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.2145,
+      "step": 5000
+    },
+    {
+      "epoch": 0.29148242960204795,
+      "grad_norm": 1.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.215,
+      "step": 5010
+    },
+    {
+      "epoch": 0.29206423085873867,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2147,
+      "step": 5020
+    },
+    {
+      "epoch": 0.2926460321154294,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2142,
+      "step": 5030
+    },
+    {
+      "epoch": 0.2932278333721201,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.2148,
+      "step": 5040
+    },
+    {
+      "epoch": 0.2938096346288108,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2099,
+      "step": 5050
+    },
+    {
+      "epoch": 0.29439143588550154,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.2139,
+      "step": 5060
+    },
+    {
+      "epoch": 0.2949732371421922,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.213,
+      "step": 5070
+    },
+    {
+      "epoch": 0.2955550383988829,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.2142,
+      "step": 5080
+    },
+    {
+      "epoch": 0.29613683965557364,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.216,
+      "step": 5090
+    },
+    {
+      "epoch": 0.29671864091226435,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2194,
+      "step": 5100
+    },
+    {
+      "epoch": 0.29730044216895507,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2099,
+      "step": 5110
+    },
+    {
+      "epoch": 0.2978822434256458,
+      "grad_norm": 1.953125,
+      "learning_rate": 2e-05,
+      "loss": 0.216,
+      "step": 5120
+    },
+    {
+      "epoch": 0.2984640446823365,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.2113,
+      "step": 5130
+    },
+    {
+      "epoch": 0.2990458459390272,
+      "grad_norm": 1.921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2155,
+      "step": 5140
+    },
+    {
+      "epoch": 0.29962764719571794,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2083,
+      "step": 5150
+    },
+    {
+      "epoch": 0.30020944845240866,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2179,
+      "step": 5160
+    },
+    {
+      "epoch": 0.3007912497090994,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.2137,
+      "step": 5170
+    },
+    {
+      "epoch": 0.3013730509657901,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.2129,
+      "step": 5180
+    },
+    {
+      "epoch": 0.3019548522224808,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2079,
+      "step": 5190
+    },
+    {
+      "epoch": 0.3025366534791715,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2175,
+      "step": 5200
+    },
+    {
+      "epoch": 0.30311845473586224,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2186,
+      "step": 5210
+    },
+    {
+      "epoch": 0.30370025599255296,
+      "grad_norm": 1.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2091,
+      "step": 5220
+    },
+    {
+      "epoch": 0.3042820572492437,
+      "grad_norm": 1.828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2132,
+      "step": 5230
+    },
+    {
+      "epoch": 0.3048638585059344,
+      "grad_norm": 1.96875,
+      "learning_rate": 2e-05,
+      "loss": 0.2139,
+      "step": 5240
+    },
+    {
+      "epoch": 0.3054456597626251,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2201,
+      "step": 5250
+    },
+    {
+      "epoch": 0.3060274610193158,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.2159,
+      "step": 5260
+    },
+    {
+      "epoch": 0.3066092622760065,
+      "grad_norm": 1.8515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2146,
+      "step": 5270
+    },
+    {
+      "epoch": 0.3071910635326972,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2075,
+      "step": 5280
+    },
+    {
+      "epoch": 0.3077728647893879,
+      "grad_norm": 1.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.215,
+      "step": 5290
+    },
+    {
+      "epoch": 0.30835466604607864,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.2197,
+      "step": 5300
+    },
+    {
+      "epoch": 0.30893646730276936,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2151,
+      "step": 5310
+    },
+    {
+      "epoch": 0.3095182685594601,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2208,
+      "step": 5320
+    },
+    {
+      "epoch": 0.3101000698161508,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2121,
+      "step": 5330
+    },
+    {
+      "epoch": 0.3106818710728415,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.2063,
+      "step": 5340
+    },
+    {
+      "epoch": 0.3112636723295322,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.2118,
+      "step": 5350
+    },
+    {
+      "epoch": 0.31184547358622294,
+      "grad_norm": 2.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2156,
+      "step": 5360
+    },
+    {
+      "epoch": 0.31242727484291366,
+      "grad_norm": 1.796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2199,
+      "step": 5370
+    },
+    {
+      "epoch": 0.3130090760996044,
+      "grad_norm": 1.921875,
+      "learning_rate": 2e-05,
+      "loss": 0.211,
+      "step": 5380
+    },
+    {
+      "epoch": 0.3135908773562951,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2158,
+      "step": 5390
+    },
+    {
+      "epoch": 0.3141726786129858,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2078,
+      "step": 5400
+    },
+    {
+      "epoch": 0.31475447986967653,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2168,
+      "step": 5410
+    },
+    {
+      "epoch": 0.31533628112636725,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2191,
+      "step": 5420
+    },
+    {
+      "epoch": 0.31591808238305796,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2117,
+      "step": 5430
+    },
+    {
+      "epoch": 0.3164998836397487,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2034,
+      "step": 5440
+    },
+    {
+      "epoch": 0.3170816848964394,
+      "grad_norm": 2.21875,
+      "learning_rate": 2e-05,
+      "loss": 0.2146,
+      "step": 5450
+    },
+    {
+      "epoch": 0.3176634861531301,
+      "grad_norm": 1.2734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2086,
+      "step": 5460
+    },
+    {
+      "epoch": 0.31824528740982083,
+      "grad_norm": 1.78125,
+      "learning_rate": 2e-05,
+      "loss": 0.2155,
+      "step": 5470
+    },
+    {
+      "epoch": 0.3188270886665115,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2146,
+      "step": 5480
+    },
+    {
+      "epoch": 0.3194088899232022,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.2133,
+      "step": 5490
+    },
+    {
+      "epoch": 0.3199906911798929,
+      "grad_norm": 1.78125,
+      "learning_rate": 2e-05,
+      "loss": 0.2191,
+      "step": 5500
+    },
+    {
+      "epoch": 0.32057249243658364,
+      "grad_norm": 2.265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2131,
+      "step": 5510
+    },
+    {
+      "epoch": 0.32115429369327436,
+      "grad_norm": 1.90625,
+      "learning_rate": 2e-05,
+      "loss": 0.214,
+      "step": 5520
+    },
+    {
+      "epoch": 0.3217360949499651,
+      "grad_norm": 1.78125,
+      "learning_rate": 2e-05,
+      "loss": 0.2091,
+      "step": 5530
+    },
+    {
+      "epoch": 0.3223178962066558,
+      "grad_norm": 1.953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2077,
+      "step": 5540
+    },
+    {
+      "epoch": 0.3228996974633465,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.2137,
+      "step": 5550
+    },
+    {
+      "epoch": 0.32348149872003723,
+      "grad_norm": 2.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2149,
+      "step": 5560
+    },
+    {
+      "epoch": 0.32406329997672795,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.2111,
+      "step": 5570
+    },
+    {
+      "epoch": 0.32464510123341866,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2136,
+      "step": 5580
+    },
+    {
+      "epoch": 0.3252269024901094,
+      "grad_norm": 1.796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2099,
+      "step": 5590
+    },
+    {
+      "epoch": 0.3258087037468001,
+      "grad_norm": 2.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2114,
+      "step": 5600
+    },
+    {
+      "epoch": 0.3263905050034908,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2152,
+      "step": 5610
+    },
+    {
+      "epoch": 0.32697230626018153,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2153,
+      "step": 5620
+    },
+    {
+      "epoch": 0.32755410751687225,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.207,
+      "step": 5630
+    },
+    {
+      "epoch": 0.32813590877356297,
+      "grad_norm": 1.9453125,
+      "learning_rate": 2e-05,
+      "loss": 0.215,
+      "step": 5640
+    },
+    {
+      "epoch": 0.3287177100302537,
+      "grad_norm": 1.9765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2084,
+      "step": 5650
+    },
+    {
+      "epoch": 0.3292995112869444,
+      "grad_norm": 1.78125,
+      "learning_rate": 2e-05,
+      "loss": 0.217,
+      "step": 5660
+    },
+    {
+      "epoch": 0.3298813125436351,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.2094,
+      "step": 5670
+    },
+    {
+      "epoch": 0.33046311380032584,
+      "grad_norm": 1.96875,
+      "learning_rate": 2e-05,
+      "loss": 0.2118,
+      "step": 5680
+    },
+    {
+      "epoch": 0.3310449150570165,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2092,
+      "step": 5690
+    },
+    {
+      "epoch": 0.3316267163137072,
+      "grad_norm": 1.796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2086,
+      "step": 5700
+    },
+    {
+      "epoch": 0.33220851757039793,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2126,
+      "step": 5710
+    },
+    {
+      "epoch": 0.33279031882708865,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2143,
+      "step": 5720
+    },
+    {
+      "epoch": 0.33337212008377937,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2077,
+      "step": 5730
+    },
+    {
+      "epoch": 0.3339539213404701,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2101,
+      "step": 5740
+    },
+    {
+      "epoch": 0.3345357225971608,
+      "grad_norm": 1.828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2088,
+      "step": 5750
+    },
+    {
+      "epoch": 0.3351175238538515,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.2186,
+      "step": 5760
+    },
+    {
+      "epoch": 0.33569932511054223,
+      "grad_norm": 1.28125,
+      "learning_rate": 2e-05,
+      "loss": 0.2052,
+      "step": 5770
+    },
+    {
+      "epoch": 0.33628112636723295,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2074,
+      "step": 5780
+    },
+    {
+      "epoch": 0.33686292762392367,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2087,
+      "step": 5790
+    },
+    {
+      "epoch": 0.3374447288806144,
+      "grad_norm": 1.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2156,
+      "step": 5800
+    },
+    {
+      "epoch": 0.3380265301373051,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2118,
+      "step": 5810
+    },
+    {
+      "epoch": 0.3386083313939958,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2093,
+      "step": 5820
+    },
+    {
+      "epoch": 0.33919013265068654,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2071,
+      "step": 5830
+    },
+    {
+      "epoch": 0.33977193390737725,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2179,
+      "step": 5840
+    },
+    {
+      "epoch": 0.34035373516406797,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2113,
+      "step": 5850
+    },
+    {
+      "epoch": 0.3409355364207587,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2038,
+      "step": 5860
+    },
+    {
+      "epoch": 0.3415173376774494,
+      "grad_norm": 1.796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2165,
+      "step": 5870
+    },
+    {
+      "epoch": 0.3420991389341401,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2154,
+      "step": 5880
+    },
+    {
+      "epoch": 0.3426809401908308,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2146,
+      "step": 5890
+    },
+    {
+      "epoch": 0.3432627414475215,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2077,
+      "step": 5900
+    },
+    {
+      "epoch": 0.3438445427042122,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2105,
+      "step": 5910
+    },
+    {
+      "epoch": 0.34442634396090294,
+      "grad_norm": 1.2578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2076,
+      "step": 5920
+    },
+    {
+      "epoch": 0.34500814521759365,
+      "grad_norm": 2.171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2107,
+      "step": 5930
+    },
+    {
+      "epoch": 0.34558994647428437,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.209,
+      "step": 5940
+    },
+    {
+      "epoch": 0.3461717477309751,
+      "grad_norm": 2.21875,
+      "learning_rate": 2e-05,
+      "loss": 0.2101,
+      "step": 5950
+    },
+    {
+      "epoch": 0.3467535489876658,
+      "grad_norm": 1.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.2137,
+      "step": 5960
+    },
+    {
+      "epoch": 0.3473353502443565,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2067,
+      "step": 5970
+    },
+    {
+      "epoch": 0.34791715150104724,
+      "grad_norm": 1.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2203,
+      "step": 5980
+    },
+    {
+      "epoch": 0.34849895275773796,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2115,
+      "step": 5990
+    },
+    {
+      "epoch": 0.3490807540144287,
+      "grad_norm": 1.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.21,
+      "step": 6000
+    },
+    {
+      "epoch": 0.3496625552711194,
+      "grad_norm": 1.28125,
+      "learning_rate": 2e-05,
+      "loss": 0.2184,
+      "step": 6010
+    },
+    {
+      "epoch": 0.3502443565278101,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2033,
+      "step": 6020
+    },
+    {
+      "epoch": 0.3508261577845008,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2103,
+      "step": 6030
+    },
+    {
+      "epoch": 0.35140795904119154,
+      "grad_norm": 1.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.2085,
+      "step": 6040
+    },
+    {
+      "epoch": 0.35198976029788226,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.213,
+      "step": 6050
+    },
+    {
+      "epoch": 0.352571561554573,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2067,
+      "step": 6060
+    },
+    {
+      "epoch": 0.3531533628112637,
+      "grad_norm": 1.8359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2107,
+      "step": 6070
+    },
+    {
+      "epoch": 0.3537351640679544,
+      "grad_norm": 1.75,
+      "learning_rate": 2e-05,
+      "loss": 0.2057,
+      "step": 6080
+    },
+    {
+      "epoch": 0.35431696532464513,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.2135,
+      "step": 6090
+    },
+    {
+      "epoch": 0.3548987665813358,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2066,
+      "step": 6100
+    },
+    {
+      "epoch": 0.3554805678380265,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2162,
+      "step": 6110
+    },
+    {
+      "epoch": 0.3560623690947172,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2062,
+      "step": 6120
+    },
+    {
+      "epoch": 0.35664417035140794,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2115,
+      "step": 6130
+    },
+    {
+      "epoch": 0.35722597160809866,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2108,
+      "step": 6140
+    },
+    {
+      "epoch": 0.3578077728647894,
+      "grad_norm": 1.296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2096,
+      "step": 6150
+    },
+    {
+      "epoch": 0.3583895741214801,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.207,
+      "step": 6160
+    },
+    {
+      "epoch": 0.3589713753781708,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2071,
+      "step": 6170
+    },
+    {
+      "epoch": 0.3595531766348615,
+      "grad_norm": 1.9375,
+      "learning_rate": 2e-05,
+      "loss": 0.2109,
+      "step": 6180
+    },
+    {
+      "epoch": 0.36013497789155224,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2118,
+      "step": 6190
+    },
+    {
+      "epoch": 0.36071677914824296,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2092,
+      "step": 6200
+    },
+    {
+      "epoch": 0.3612985804049337,
+      "grad_norm": 2.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2032,
+      "step": 6210
+    },
+    {
+      "epoch": 0.3618803816616244,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2169,
+      "step": 6220
+    },
+    {
+      "epoch": 0.3624621829183151,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2051,
+      "step": 6230
+    },
+    {
+      "epoch": 0.36304398417500583,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2112,
+      "step": 6240
+    },
+    {
+      "epoch": 0.36362578543169655,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2064,
+      "step": 6250
+    },
+    {
+      "epoch": 0.36420758668838726,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.2055,
+      "step": 6260
+    },
+    {
+      "epoch": 0.364789387945078,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2018,
+      "step": 6270
+    },
+    {
+      "epoch": 0.3653711892017687,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.2051,
+      "step": 6280
+    },
+    {
+      "epoch": 0.3659529904584594,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2094,
+      "step": 6290
+    },
+    {
+      "epoch": 0.36653479171515013,
+      "grad_norm": 1.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2119,
+      "step": 6300
+    },
+    {
+      "epoch": 0.3671165929718408,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2134,
+      "step": 6310
+    },
+    {
+      "epoch": 0.3676983942285315,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2078,
+      "step": 6320
+    },
+    {
+      "epoch": 0.36828019548522223,
+      "grad_norm": 1.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.2012,
+      "step": 6330
+    },
+    {
+      "epoch": 0.36886199674191295,
+      "grad_norm": 1.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.2051,
+      "step": 6340
+    },
+    {
+      "epoch": 0.36944379799860366,
+      "grad_norm": 2.015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2086,
+      "step": 6350
+    },
+    {
+      "epoch": 0.3700255992552944,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2e-05,
+      "loss": 0.2118,
+      "step": 6360
+    },
+    {
+      "epoch": 0.3706074005119851,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2086,
+      "step": 6370
+    },
+    {
+      "epoch": 0.3711892017686758,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2185,
+      "step": 6380
+    },
+    {
+      "epoch": 0.37177100302536653,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2094,
+      "step": 6390
+    },
+    {
+      "epoch": 0.37235280428205725,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2116,
+      "step": 6400
+    },
+    {
+      "epoch": 0.37293460553874797,
+      "grad_norm": 1.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.2097,
+      "step": 6410
+    },
+    {
+      "epoch": 0.3735164067954387,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2116,
+      "step": 6420
+    },
+    {
+      "epoch": 0.3740982080521294,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.2068,
+      "step": 6430
+    },
+    {
+      "epoch": 0.3746800093088201,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2094,
+      "step": 6440
+    },
+    {
+      "epoch": 0.37526181056551083,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.2103,
+      "step": 6450
+    },
+    {
+      "epoch": 0.37584361182220155,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.201,
+      "step": 6460
+    },
+    {
+      "epoch": 0.37642541307889227,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2099,
+      "step": 6470
+    },
+    {
+      "epoch": 0.377007214335583,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.2075,
+      "step": 6480
+    },
+    {
+      "epoch": 0.3775890155922737,
+      "grad_norm": 1.2578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2045,
+      "step": 6490
+    },
+    {
+      "epoch": 0.3781708168489644,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2085,
+      "step": 6500
+    },
+    {
+      "epoch": 0.3787526181056551,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2156,
+      "step": 6510
+    },
+    {
+      "epoch": 0.3793344193623458,
+      "grad_norm": 1.796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2152,
+      "step": 6520
+    },
+    {
+      "epoch": 0.3799162206190365,
+      "grad_norm": 1.21875,
+      "learning_rate": 2e-05,
+      "loss": 0.2085,
+      "step": 6530
+    },
+    {
+      "epoch": 0.38049802187572723,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2068,
+      "step": 6540
+    },
+    {
+      "epoch": 0.38107982313241795,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2148,
+      "step": 6550
+    },
+    {
+      "epoch": 0.38166162438910867,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2052,
+      "step": 6560
+    },
+    {
+      "epoch": 0.3822434256457994,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2098,
+      "step": 6570
+    },
+    {
+      "epoch": 0.3828252269024901,
+      "grad_norm": 1.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.2048,
+      "step": 6580
+    },
+    {
+      "epoch": 0.3834070281591808,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.2031,
+      "step": 6590
+    },
+    {
+      "epoch": 0.38398882941587154,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.2028,
+      "step": 6600
+    },
+    {
+      "epoch": 0.38457063067256225,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.2072,
+      "step": 6610
+    },
+    {
+      "epoch": 0.38515243192925297,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2026,
+      "step": 6620
+    },
+    {
+      "epoch": 0.3857342331859437,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.2031,
+      "step": 6630
+    },
+    {
+      "epoch": 0.3863160344426344,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2104,
+      "step": 6640
+    },
+    {
+      "epoch": 0.3868978356993251,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2065,
+      "step": 6650
+    },
+    {
+      "epoch": 0.38747963695601584,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2067,
+      "step": 6660
+    },
+    {
+      "epoch": 0.38806143821270656,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.2109,
+      "step": 6670
+    },
+    {
+      "epoch": 0.3886432394693973,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2045,
+      "step": 6680
+    },
+    {
+      "epoch": 0.389225040726088,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2026,
+      "step": 6690
+    },
+    {
+      "epoch": 0.3898068419827787,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2036,
+      "step": 6700
+    },
+    {
+      "epoch": 0.3903886432394694,
+      "grad_norm": 1.984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2152,
+      "step": 6710
+    },
+    {
+      "epoch": 0.3909704444961601,
+      "grad_norm": 3.0,
+      "learning_rate": 2e-05,
+      "loss": 0.2053,
+      "step": 6720
+    },
+    {
+      "epoch": 0.3915522457528508,
+      "grad_norm": 2.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.21,
+      "step": 6730
+    },
+    {
+      "epoch": 0.3921340470095415,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.2068,
+      "step": 6740
+    },
+    {
+      "epoch": 0.39271584826623224,
+      "grad_norm": 1.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2063,
+      "step": 6750
+    },
+    {
+      "epoch": 0.39329764952292295,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2093,
+      "step": 6760
+    },
+    {
+      "epoch": 0.39387945077961367,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.2043,
+      "step": 6770
+    },
+    {
+      "epoch": 0.3944612520363044,
+      "grad_norm": 2.15625,
+      "learning_rate": 2e-05,
+      "loss": 0.211,
+      "step": 6780
+    },
+    {
+      "epoch": 0.3950430532929951,
+      "grad_norm": 1.796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2111,
+      "step": 6790
+    },
+    {
+      "epoch": 0.3956248545496858,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2022,
+      "step": 6800
+    },
+    {
+      "epoch": 0.39620665580637654,
+      "grad_norm": 2.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2028,
+      "step": 6810
+    },
+    {
+      "epoch": 0.39678845706306726,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2143,
+      "step": 6820
+    },
+    {
+      "epoch": 0.397370258319758,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2124,
+      "step": 6830
+    },
+    {
+      "epoch": 0.3979520595764487,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2185,
+      "step": 6840
+    },
+    {
+      "epoch": 0.3985338608331394,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2118,
+      "step": 6850
+    },
+    {
+      "epoch": 0.3991156620898301,
+      "grad_norm": 1.3515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2091,
+      "step": 6860
+    },
+    {
+      "epoch": 0.39969746334652084,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2086,
+      "step": 6870
+    },
+    {
+      "epoch": 0.40027926460321156,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2025,
+      "step": 6880
+    },
+    {
+      "epoch": 0.4008610658599023,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2035,
+      "step": 6890
+    },
+    {
+      "epoch": 0.401442867116593,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.2086,
+      "step": 6900
+    },
+    {
+      "epoch": 0.4020246683732837,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2096,
+      "step": 6910
+    },
+    {
+      "epoch": 0.40260646962997443,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.207,
+      "step": 6920
+    },
+    {
+      "epoch": 0.4031882708866651,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2113,
+      "step": 6930
+    },
+    {
+      "epoch": 0.4037700721433558,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2077,
+      "step": 6940
+    },
+    {
+      "epoch": 0.4043518734000465,
+      "grad_norm": 1.96875,
+      "learning_rate": 2e-05,
+      "loss": 0.1999,
+      "step": 6950
+    },
+    {
+      "epoch": 0.40493367465673724,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2014,
+      "step": 6960
+    },
+    {
+      "epoch": 0.40551547591342796,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2058,
+      "step": 6970
+    },
+    {
+      "epoch": 0.4060972771701187,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.209,
+      "step": 6980
+    },
+    {
+      "epoch": 0.4066790784268094,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2103,
+      "step": 6990
+    },
+    {
+      "epoch": 0.4072608796835001,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2117,
+      "step": 7000
+    },
+    {
+      "epoch": 0.4078426809401908,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.2055,
+      "step": 7010
+    },
+    {
+      "epoch": 0.40842448219688154,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2129,
+      "step": 7020
+    },
+    {
+      "epoch": 0.40900628345357226,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.2038,
+      "step": 7030
+    },
+    {
+      "epoch": 0.409588084710263,
+      "grad_norm": 1.9375,
+      "learning_rate": 2e-05,
+      "loss": 0.2028,
+      "step": 7040
+    },
+    {
+      "epoch": 0.4101698859669537,
+      "grad_norm": 2.203125,
+      "learning_rate": 2e-05,
+      "loss": 0.2085,
+      "step": 7050
+    },
+    {
+      "epoch": 0.4107516872236444,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2076,
+      "step": 7060
+    },
+    {
+      "epoch": 0.41133348848033513,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.2047,
+      "step": 7070
+    },
+    {
+      "epoch": 0.41191528973702585,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2079,
+      "step": 7080
+    },
+    {
+      "epoch": 0.41249709099371656,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2016,
+      "step": 7090
+    },
+    {
+      "epoch": 0.4130788922504073,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2084,
+      "step": 7100
+    },
+    {
+      "epoch": 0.413660693507098,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2076,
+      "step": 7110
+    },
+    {
+      "epoch": 0.4142424947637887,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2084,
+      "step": 7120
+    },
+    {
+      "epoch": 0.4148242960204794,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.2003,
+      "step": 7130
+    },
+    {
+      "epoch": 0.4154060972771701,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2059,
+      "step": 7140
+    },
+    {
+      "epoch": 0.4159878985338608,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2067,
+      "step": 7150
+    },
+    {
+      "epoch": 0.41656969979055153,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.2028,
+      "step": 7160
+    },
+    {
+      "epoch": 0.41715150104724225,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2015,
+      "step": 7170
+    },
+    {
+      "epoch": 0.41773330230393296,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2007,
+      "step": 7180
+    },
+    {
+      "epoch": 0.4183151035606237,
+      "grad_norm": 1.375,
+      "learning_rate": 2e-05,
+      "loss": 0.2061,
+      "step": 7190
+    },
+    {
+      "epoch": 0.4188969048173144,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.2026,
+      "step": 7200
+    },
+    {
+      "epoch": 0.4194787060740051,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2088,
+      "step": 7210
+    },
+    {
+      "epoch": 0.42006050733069583,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.2037,
+      "step": 7220
+    },
+    {
+      "epoch": 0.42064230858738655,
+      "grad_norm": 2.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2063,
+      "step": 7230
+    },
+    {
+      "epoch": 0.42122410984407727,
+      "grad_norm": 1.796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2086,
+      "step": 7240
+    },
+    {
+      "epoch": 0.421805911100768,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2048,
+      "step": 7250
+    },
+    {
+      "epoch": 0.4223877123574587,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2002,
+      "step": 7260
+    },
+    {
+      "epoch": 0.4229695136141494,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1988,
+      "step": 7270
+    },
+    {
+      "epoch": 0.42355131487084013,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.2082,
+      "step": 7280
+    },
+    {
+      "epoch": 0.42413311612753085,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2086,
+      "step": 7290
+    },
+    {
+      "epoch": 0.42471491738422157,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1968,
+      "step": 7300
+    },
+    {
+      "epoch": 0.4252967186409123,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.1984,
+      "step": 7310
+    },
+    {
+      "epoch": 0.425878519897603,
+      "grad_norm": 2.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2068,
+      "step": 7320
+    },
+    {
+      "epoch": 0.4264603211542937,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2029,
+      "step": 7330
+    },
+    {
+      "epoch": 0.4270421224109844,
+      "grad_norm": 2.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.2093,
+      "step": 7340
+    },
+    {
+      "epoch": 0.4276239236676751,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.2055,
+      "step": 7350
+    },
+    {
+      "epoch": 0.4282057249243658,
+      "grad_norm": 3.625,
+      "learning_rate": 2e-05,
+      "loss": 0.2018,
+      "step": 7360
+    },
+    {
+      "epoch": 0.42878752618105653,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2011,
+      "step": 7370
+    },
+    {
+      "epoch": 0.42936932743774725,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2059,
+      "step": 7380
+    },
+    {
+      "epoch": 0.42995112869443797,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.2051,
+      "step": 7390
+    },
+    {
+      "epoch": 0.4305329299511287,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1991,
+      "step": 7400
+    },
+    {
+      "epoch": 0.4311147312078194,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.2114,
+      "step": 7410
+    },
+    {
+      "epoch": 0.4316965324645101,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1999,
+      "step": 7420
+    },
+    {
+      "epoch": 0.43227833372120084,
+      "grad_norm": 2.09375,
+      "learning_rate": 2e-05,
+      "loss": 0.2031,
+      "step": 7430
+    },
+    {
+      "epoch": 0.43286013497789155,
+      "grad_norm": 1.375,
+      "learning_rate": 2e-05,
+      "loss": 0.2055,
+      "step": 7440
+    },
+    {
+      "epoch": 0.43344193623458227,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2042,
+      "step": 7450
+    },
+    {
+      "epoch": 0.434023737491273,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.2036,
+      "step": 7460
+    },
+    {
+      "epoch": 0.4346055387479637,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2049,
+      "step": 7470
+    },
+    {
+      "epoch": 0.4351873400046544,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2002,
+      "step": 7480
+    },
+    {
+      "epoch": 0.43576914126134514,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2091,
+      "step": 7490
+    },
+    {
+      "epoch": 0.43635094251803586,
+      "grad_norm": 2.265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2114,
+      "step": 7500
+    },
+    {
+      "epoch": 0.4369327437747266,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2053,
+      "step": 7510
+    },
+    {
+      "epoch": 0.4375145450314173,
+      "grad_norm": 1.90625,
+      "learning_rate": 2e-05,
+      "loss": 0.1966,
+      "step": 7520
+    },
+    {
+      "epoch": 0.438096346288108,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1968,
+      "step": 7530
+    },
+    {
+      "epoch": 0.4386781475447987,
+      "grad_norm": 1.9453125,
+      "learning_rate": 2e-05,
+      "loss": 0.2038,
+      "step": 7540
+    },
+    {
+      "epoch": 0.4392599488014894,
+      "grad_norm": 1.78125,
+      "learning_rate": 2e-05,
+      "loss": 0.2028,
+      "step": 7550
+    },
+    {
+      "epoch": 0.4398417500581801,
+      "grad_norm": 1.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2027,
+      "step": 7560
+    },
+    {
+      "epoch": 0.4404235513148708,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2e-05,
+      "loss": 0.201,
+      "step": 7570
+    },
+    {
+      "epoch": 0.44100535257156154,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2057,
+      "step": 7580
+    },
+    {
+      "epoch": 0.44158715382825225,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.2059,
+      "step": 7590
+    },
+    {
+      "epoch": 0.44216895508494297,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1953,
+      "step": 7600
+    },
+    {
+      "epoch": 0.4427507563416337,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2039,
+      "step": 7610
+    },
+    {
+      "epoch": 0.4433325575983244,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2079,
+      "step": 7620
+    },
+    {
+      "epoch": 0.4439143588550151,
+      "grad_norm": 1.796875,
+      "learning_rate": 2e-05,
+      "loss": 0.202,
+      "step": 7630
+    },
+    {
+      "epoch": 0.44449616011170584,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2022,
+      "step": 7640
+    },
+    {
+      "epoch": 0.44507796136839656,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2013,
+      "step": 7650
+    },
+    {
+      "epoch": 0.4456597626250873,
+      "grad_norm": 1.921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2021,
+      "step": 7660
+    },
+    {
+      "epoch": 0.446241563881778,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.206,
+      "step": 7670
+    },
+    {
+      "epoch": 0.4468233651384687,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.2018,
+      "step": 7680
+    },
+    {
+      "epoch": 0.4474051663951594,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1979,
+      "step": 7690
+    },
+    {
+      "epoch": 0.44798696765185014,
+      "grad_norm": 2.0625,
+      "learning_rate": 2e-05,
+      "loss": 0.2063,
+      "step": 7700
+    },
+    {
+      "epoch": 0.44856876890854086,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2001,
+      "step": 7710
+    },
+    {
+      "epoch": 0.4491505701652316,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.2038,
+      "step": 7720
+    },
+    {
+      "epoch": 0.4497323714219223,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2037,
+      "step": 7730
+    },
+    {
+      "epoch": 0.450314172678613,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2e-05,
+      "loss": 0.209,
+      "step": 7740
+    },
+    {
+      "epoch": 0.4508959739353037,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2001,
+      "step": 7750
+    },
+    {
+      "epoch": 0.4514777751919944,
+      "grad_norm": 1.75,
+      "learning_rate": 2e-05,
+      "loss": 0.2054,
+      "step": 7760
+    },
+    {
+      "epoch": 0.4520595764486851,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.204,
+      "step": 7770
+    },
+    {
+      "epoch": 0.4526413777053758,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2025,
+      "step": 7780
+    },
+    {
+      "epoch": 0.45322317896206654,
+      "grad_norm": 2.375,
+      "learning_rate": 2e-05,
+      "loss": 0.1992,
+      "step": 7790
+    },
+    {
+      "epoch": 0.45380498021875726,
+      "grad_norm": 2.25,
+      "learning_rate": 2e-05,
+      "loss": 0.1998,
+      "step": 7800
+    },
+    {
+      "epoch": 0.454386781475448,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.2055,
+      "step": 7810
+    },
+    {
+      "epoch": 0.4549685827321387,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.206,
+      "step": 7820
+    },
+    {
+      "epoch": 0.4555503839888294,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2024,
+      "step": 7830
+    },
+    {
+      "epoch": 0.4561321852455201,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.1985,
+      "step": 7840
+    },
+    {
+      "epoch": 0.45671398650221084,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2086,
+      "step": 7850
+    },
+    {
+      "epoch": 0.45729578775890156,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2052,
+      "step": 7860
+    },
+    {
+      "epoch": 0.4578775890155923,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.204,
+      "step": 7870
+    },
+    {
+      "epoch": 0.458459390272283,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.2028,
+      "step": 7880
+    },
+    {
+      "epoch": 0.4590411915289737,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1982,
+      "step": 7890
+    },
+    {
+      "epoch": 0.45962299278566443,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.1986,
+      "step": 7900
+    },
+    {
+      "epoch": 0.46020479404235515,
+      "grad_norm": 2.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1971,
+      "step": 7910
+    },
+    {
+      "epoch": 0.46078659529904586,
+      "grad_norm": 1.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2011,
+      "step": 7920
+    },
+    {
+      "epoch": 0.4613683965557366,
+      "grad_norm": 2.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.207,
+      "step": 7930
+    },
+    {
+      "epoch": 0.4619501978124273,
+      "grad_norm": 1.9765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2032,
+      "step": 7940
+    },
+    {
+      "epoch": 0.462531999069118,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2053,
+      "step": 7950
+    },
+    {
+      "epoch": 0.4631138003258087,
+      "grad_norm": 1.90625,
+      "learning_rate": 2e-05,
+      "loss": 0.1991,
+      "step": 7960
+    },
+    {
+      "epoch": 0.4636956015824994,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2059,
+      "step": 7970
+    },
+    {
+      "epoch": 0.4642774028391901,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1986,
+      "step": 7980
+    },
+    {
+      "epoch": 0.46485920409588083,
+      "grad_norm": 1.828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2033,
+      "step": 7990
+    },
+    {
+      "epoch": 0.46544100535257155,
+      "grad_norm": 1.75,
+      "learning_rate": 2e-05,
+      "loss": 0.2005,
+      "step": 8000
+    },
+    {
+      "epoch": 0.46602280660926226,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2101,
+      "step": 8010
+    },
+    {
+      "epoch": 0.466604607865953,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1976,
+      "step": 8020
+    },
+    {
+      "epoch": 0.4671864091226437,
+      "grad_norm": 1.8671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1992,
+      "step": 8030
+    },
+    {
+      "epoch": 0.4677682103793344,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2069,
+      "step": 8040
+    },
+    {
+      "epoch": 0.46835001163602513,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.2055,
+      "step": 8050
+    },
+    {
+      "epoch": 0.46893181289271585,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.203,
+      "step": 8060
+    },
+    {
+      "epoch": 0.46951361414940657,
+      "grad_norm": 1.8515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2023,
+      "step": 8070
+    },
+    {
+      "epoch": 0.4700954154060973,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.2084,
+      "step": 8080
+    },
+    {
+      "epoch": 0.470677216662788,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2094,
+      "step": 8090
+    },
+    {
+      "epoch": 0.4712590179194787,
+      "grad_norm": 2.15625,
+      "learning_rate": 2e-05,
+      "loss": 0.202,
+      "step": 8100
+    },
+    {
+      "epoch": 0.47184081917616943,
+      "grad_norm": 1.28125,
+      "learning_rate": 2e-05,
+      "loss": 0.1977,
+      "step": 8110
+    },
+    {
+      "epoch": 0.47242262043286015,
+      "grad_norm": 1.8359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2017,
+      "step": 8120
+    },
+    {
+      "epoch": 0.47300442168955087,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.193,
+      "step": 8130
+    },
+    {
+      "epoch": 0.4735862229462416,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2109,
+      "step": 8140
+    },
+    {
+      "epoch": 0.4741680242029323,
+      "grad_norm": 1.234375,
+      "learning_rate": 2e-05,
+      "loss": 0.1937,
+      "step": 8150
+    },
+    {
+      "epoch": 0.474749825459623,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2015,
+      "step": 8160
+    },
+    {
+      "epoch": 0.4753316267163137,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2024,
+      "step": 8170
+    },
+    {
+      "epoch": 0.4759134279730044,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2028,
+      "step": 8180
+    },
+    {
+      "epoch": 0.4764952292296951,
+      "grad_norm": 2.109375,
+      "learning_rate": 2e-05,
+      "loss": 0.1949,
+      "step": 8190
+    },
+    {
+      "epoch": 0.47707703048638583,
+      "grad_norm": 1.296875,
+      "learning_rate": 2e-05,
+      "loss": 0.2067,
+      "step": 8200
+    },
+    {
+      "epoch": 0.47765883174307655,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1973,
+      "step": 8210
+    },
+    {
+      "epoch": 0.47824063299976727,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.197,
+      "step": 8220
+    },
+    {
+      "epoch": 0.478822434256458,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1956,
+      "step": 8230
+    },
+    {
+      "epoch": 0.4794042355131487,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1982,
+      "step": 8240
+    },
+    {
+      "epoch": 0.4799860367698394,
+      "grad_norm": 2.15625,
+      "learning_rate": 2e-05,
+      "loss": 0.1975,
+      "step": 8250
+    },
+    {
+      "epoch": 0.48056783802653014,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1969,
+      "step": 8260
+    },
+    {
+      "epoch": 0.48114963928322085,
+      "grad_norm": 1.828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2013,
+      "step": 8270
+    },
+    {
+      "epoch": 0.48173144053991157,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2e-05,
+      "loss": 0.1965,
+      "step": 8280
+    },
+    {
+      "epoch": 0.4823132417966023,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2035,
+      "step": 8290
+    },
+    {
+      "epoch": 0.482895043053293,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1969,
+      "step": 8300
+    },
+    {
+      "epoch": 0.4834768443099837,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1949,
+      "step": 8310
+    },
+    {
+      "epoch": 0.48405864556667444,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1987,
+      "step": 8320
+    },
+    {
+      "epoch": 0.48464044682336516,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.2047,
+      "step": 8330
+    },
+    {
+      "epoch": 0.4852222480800559,
+      "grad_norm": 1.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2075,
+      "step": 8340
+    },
+    {
+      "epoch": 0.4858040493367466,
+      "grad_norm": 2.125,
+      "learning_rate": 2e-05,
+      "loss": 0.2021,
+      "step": 8350
+    },
+    {
+      "epoch": 0.4863858505934373,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2014,
+      "step": 8360
+    },
+    {
+      "epoch": 0.48696765185012797,
+      "grad_norm": 1.9375,
+      "learning_rate": 2e-05,
+      "loss": 0.2019,
+      "step": 8370
+    },
+    {
+      "epoch": 0.4875494531068187,
+      "grad_norm": 1.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.2022,
+      "step": 8380
+    },
+    {
+      "epoch": 0.4881312543635094,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2055,
+      "step": 8390
+    },
+    {
+      "epoch": 0.4887130556202001,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.2015,
+      "step": 8400
+    },
+    {
+      "epoch": 0.48929485687689084,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.2027,
+      "step": 8410
+    },
+    {
+      "epoch": 0.48987665813358156,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2009,
+      "step": 8420
+    },
+    {
+      "epoch": 0.49045845939027227,
+      "grad_norm": 1.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.2031,
+      "step": 8430
+    },
+    {
+      "epoch": 0.491040260646963,
+      "grad_norm": 1.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.2014,
+      "step": 8440
+    },
+    {
+      "epoch": 0.4916220619036537,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.2051,
+      "step": 8450
+    },
+    {
+      "epoch": 0.4922038631603444,
+      "grad_norm": 1.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2,
+      "step": 8460
+    },
+    {
+      "epoch": 0.49278566441703514,
+      "grad_norm": 1.9375,
+      "learning_rate": 2e-05,
+      "loss": 0.1986,
+      "step": 8470
+    },
+    {
+      "epoch": 0.49336746567372586,
+      "grad_norm": 1.34375,
+      "learning_rate": 2e-05,
+      "loss": 0.1997,
+      "step": 8480
+    },
+    {
+      "epoch": 0.4939492669304166,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.2012,
+      "step": 8490
+    },
+    {
+      "epoch": 0.4945310681871073,
+      "grad_norm": 2.28125,
+      "learning_rate": 2e-05,
+      "loss": 0.1993,
+      "step": 8500
+    },
+    {
+      "epoch": 0.495112869443798,
+      "grad_norm": 3.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.2046,
+      "step": 8510
+    },
+    {
+      "epoch": 0.4956946707004887,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2006,
+      "step": 8520
+    },
+    {
+      "epoch": 0.49627647195717944,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2005,
+      "step": 8530
+    },
+    {
+      "epoch": 0.49685827321387016,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2065,
+      "step": 8540
+    },
+    {
+      "epoch": 0.4974400744705609,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.2061,
+      "step": 8550
+    },
+    {
+      "epoch": 0.4980218757272516,
+      "grad_norm": 1.796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2032,
+      "step": 8560
+    },
+    {
+      "epoch": 0.4986036769839423,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.2103,
+      "step": 8570
+    },
+    {
+      "epoch": 0.499185478240633,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.1934,
+      "step": 8580
+    },
+    {
+      "epoch": 0.4997672794973237,
+      "grad_norm": 2.140625,
+      "learning_rate": 2e-05,
+      "loss": 0.203,
+      "step": 8590
+    },
+    {
+      "epoch": 0.5003490807540144,
+      "grad_norm": 1.3515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2059,
+      "step": 8600
+    },
+    {
+      "epoch": 0.5009308820107051,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1978,
+      "step": 8610
+    },
+    {
+      "epoch": 0.5015126832673958,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1987,
+      "step": 8620
+    },
+    {
+      "epoch": 0.5020944845240866,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1923,
+      "step": 8630
+    },
+    {
+      "epoch": 0.5026762857807773,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1964,
+      "step": 8640
+    },
+    {
+      "epoch": 0.503258087037468,
+      "grad_norm": 1.3515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2076,
+      "step": 8650
+    },
+    {
+      "epoch": 0.5038398882941587,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2042,
+      "step": 8660
+    },
+    {
+      "epoch": 0.5044216895508494,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2003,
+      "step": 8670
+    },
+    {
+      "epoch": 0.5050034908075401,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1981,
+      "step": 8680
+    },
+    {
+      "epoch": 0.5055852920642309,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1957,
+      "step": 8690
+    },
+    {
+      "epoch": 0.5061670933209216,
+      "grad_norm": 1.828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2042,
+      "step": 8700
+    },
+    {
+      "epoch": 0.5067488945776123,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1998,
+      "step": 8710
+    },
+    {
+      "epoch": 0.507330695834303,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.2107,
+      "step": 8720
+    },
+    {
+      "epoch": 0.5079124970909937,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.199,
+      "step": 8730
+    },
+    {
+      "epoch": 0.5084942983476844,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2059,
+      "step": 8740
+    },
+    {
+      "epoch": 0.5090760996043752,
+      "grad_norm": 2.125,
+      "learning_rate": 2e-05,
+      "loss": 0.2057,
+      "step": 8750
+    },
+    {
+      "epoch": 0.5096579008610659,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.198,
+      "step": 8760
+    },
+    {
+      "epoch": 0.5102397021177566,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1962,
+      "step": 8770
+    },
+    {
+      "epoch": 0.5108215033744473,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.199,
+      "step": 8780
+    },
+    {
+      "epoch": 0.511403304631138,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.2023,
+      "step": 8790
+    },
+    {
+      "epoch": 0.5119851058878288,
+      "grad_norm": 1.8359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2043,
+      "step": 8800
+    },
+    {
+      "epoch": 0.5125669071445195,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1959,
+      "step": 8810
+    },
+    {
+      "epoch": 0.5131487084012102,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2e-05,
+      "loss": 0.2019,
+      "step": 8820
+    },
+    {
+      "epoch": 0.5137305096579009,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.2009,
+      "step": 8830
+    },
+    {
+      "epoch": 0.5143123109145916,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.2055,
+      "step": 8840
+    },
+    {
+      "epoch": 0.5148941121712823,
+      "grad_norm": 2.046875,
+      "learning_rate": 2e-05,
+      "loss": 0.2037,
+      "step": 8850
+    },
+    {
+      "epoch": 0.515475913427973,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1999,
+      "step": 8860
+    },
+    {
+      "epoch": 0.5160577146846638,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.1957,
+      "step": 8870
+    },
+    {
+      "epoch": 0.5166395159413545,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2015,
+      "step": 8880
+    },
+    {
+      "epoch": 0.5172213171980451,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2038,
+      "step": 8890
+    },
+    {
+      "epoch": 0.5178031184547358,
+      "grad_norm": 1.296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1999,
+      "step": 8900
+    },
+    {
+      "epoch": 0.5183849197114265,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1965,
+      "step": 8910
+    },
+    {
+      "epoch": 0.5189667209681172,
+      "grad_norm": 1.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1953,
+      "step": 8920
+    },
+    {
+      "epoch": 0.519548522224808,
+      "grad_norm": 2.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.2033,
+      "step": 8930
+    },
+    {
+      "epoch": 0.5201303234814987,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1939,
+      "step": 8940
+    },
+    {
+      "epoch": 0.5207121247381894,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2023,
+      "step": 8950
+    },
+    {
+      "epoch": 0.5212939259948801,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.2116,
+      "step": 8960
+    },
+    {
+      "epoch": 0.5218757272515708,
+      "grad_norm": 1.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1955,
+      "step": 8970
+    },
+    {
+      "epoch": 0.5224575285082615,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.1979,
+      "step": 8980
+    },
+    {
+      "epoch": 0.5230393297649523,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.2002,
+      "step": 8990
+    },
+    {
+      "epoch": 0.523621131021643,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1995,
+      "step": 9000
+    },
+    {
+      "epoch": 0.5242029322783337,
+      "grad_norm": 2.171875,
+      "learning_rate": 2e-05,
+      "loss": 0.202,
+      "step": 9010
+    },
+    {
+      "epoch": 0.5247847335350244,
+      "grad_norm": 1.9921875,
+      "learning_rate": 2e-05,
+      "loss": 0.1997,
+      "step": 9020
+    },
+    {
+      "epoch": 0.5253665347917151,
+      "grad_norm": 1.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1984,
+      "step": 9030
+    },
+    {
+      "epoch": 0.5259483360484059,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1953,
+      "step": 9040
+    },
+    {
+      "epoch": 0.5265301373050966,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.1966,
+      "step": 9050
+    },
+    {
+      "epoch": 0.5271119385617873,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2021,
+      "step": 9060
+    },
+    {
+      "epoch": 0.527693739818478,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2032,
+      "step": 9070
+    },
+    {
+      "epoch": 0.5282755410751687,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.1982,
+      "step": 9080
+    },
+    {
+      "epoch": 0.5288573423318594,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1945,
+      "step": 9090
+    },
+    {
+      "epoch": 0.5294391435885502,
+      "grad_norm": 1.8515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1889,
+      "step": 9100
+    },
+    {
+      "epoch": 0.5300209448452409,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.204,
+      "step": 9110
+    },
+    {
+      "epoch": 0.5306027461019316,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2003,
+      "step": 9120
+    },
+    {
+      "epoch": 0.5311845473586223,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.2015,
+      "step": 9130
+    },
+    {
+      "epoch": 0.531766348615313,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2016,
+      "step": 9140
+    },
+    {
+      "epoch": 0.5323481498720037,
+      "grad_norm": 1.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.2051,
+      "step": 9150
+    },
+    {
+      "epoch": 0.5329299511286945,
+      "grad_norm": 2.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.2016,
+      "step": 9160
+    },
+    {
+      "epoch": 0.5335117523853852,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.2005,
+      "step": 9170
+    },
+    {
+      "epoch": 0.5340935536420759,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2001,
+      "step": 9180
+    },
+    {
+      "epoch": 0.5346753548987666,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.196,
+      "step": 9190
+    },
+    {
+      "epoch": 0.5352571561554573,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.1957,
+      "step": 9200
+    },
+    {
+      "epoch": 0.535838957412148,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.1972,
+      "step": 9210
+    },
+    {
+      "epoch": 0.5364207586688388,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1928,
+      "step": 9220
+    },
+    {
+      "epoch": 0.5370025599255295,
+      "grad_norm": 1.9375,
+      "learning_rate": 2e-05,
+      "loss": 0.1996,
+      "step": 9230
+    },
+    {
+      "epoch": 0.5375843611822202,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.1943,
+      "step": 9240
+    },
+    {
+      "epoch": 0.5381661624389109,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1901,
+      "step": 9250
+    },
+    {
+      "epoch": 0.5387479636956016,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.2016,
+      "step": 9260
+    },
+    {
+      "epoch": 0.5393297649522923,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.1996,
+      "step": 9270
+    },
+    {
+      "epoch": 0.5399115662089831,
+      "grad_norm": 1.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1972,
+      "step": 9280
+    },
+    {
+      "epoch": 0.5404933674656738,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2063,
+      "step": 9290
+    },
+    {
+      "epoch": 0.5410751687223644,
+      "grad_norm": 1.8359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1986,
+      "step": 9300
+    },
+    {
+      "epoch": 0.5416569699790551,
+      "grad_norm": 1.3125,
+      "learning_rate": 2e-05,
+      "loss": 0.2049,
+      "step": 9310
+    },
+    {
+      "epoch": 0.5422387712357458,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1968,
+      "step": 9320
+    },
+    {
+      "epoch": 0.5428205724924365,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.1941,
+      "step": 9330
+    },
+    {
+      "epoch": 0.5434023737491273,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2014,
+      "step": 9340
+    },
+    {
+      "epoch": 0.543984175005818,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1961,
+      "step": 9350
+    },
+    {
+      "epoch": 0.5445659762625087,
+      "grad_norm": 1.8671875,
+      "learning_rate": 2e-05,
+      "loss": 0.205,
+      "step": 9360
+    },
+    {
+      "epoch": 0.5451477775191994,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2042,
+      "step": 9370
+    },
+    {
+      "epoch": 0.5457295787758901,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1927,
+      "step": 9380
+    },
+    {
+      "epoch": 0.5463113800325808,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.205,
+      "step": 9390
+    },
+    {
+      "epoch": 0.5468931812892716,
+      "grad_norm": 1.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.1973,
+      "step": 9400
+    },
+    {
+      "epoch": 0.5474749825459623,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.2069,
+      "step": 9410
+    },
+    {
+      "epoch": 0.548056783802653,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.1974,
+      "step": 9420
+    },
+    {
+      "epoch": 0.5486385850593437,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.194,
+      "step": 9430
+    },
+    {
+      "epoch": 0.5492203863160344,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1969,
+      "step": 9440
+    },
+    {
+      "epoch": 0.5498021875727251,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2019,
+      "step": 9450
+    },
+    {
+      "epoch": 0.5503839888294159,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.1983,
+      "step": 9460
+    },
+    {
+      "epoch": 0.5509657900861066,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1945,
+      "step": 9470
+    },
+    {
+      "epoch": 0.5515475913427973,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.2043,
+      "step": 9480
+    },
+    {
+      "epoch": 0.552129392599488,
+      "grad_norm": 2.171875,
+      "learning_rate": 2e-05,
+      "loss": 0.1923,
+      "step": 9490
+    },
+    {
+      "epoch": 0.5527111938561787,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.203,
+      "step": 9500
+    },
+    {
+      "epoch": 0.5532929951128694,
+      "grad_norm": 1.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.2024,
+      "step": 9510
+    },
+    {
+      "epoch": 0.5538747963695602,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.1938,
+      "step": 9520
+    },
+    {
+      "epoch": 0.5544565976262509,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1949,
+      "step": 9530
+    },
+    {
+      "epoch": 0.5550383988829416,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1989,
+      "step": 9540
+    },
+    {
+      "epoch": 0.5556202001396323,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1998,
+      "step": 9550
+    },
+    {
+      "epoch": 0.556202001396323,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.2009,
+      "step": 9560
+    },
+    {
+      "epoch": 0.5567838026530137,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.1976,
+      "step": 9570
+    },
+    {
+      "epoch": 0.5573656039097045,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.1979,
+      "step": 9580
+    },
+    {
+      "epoch": 0.5579474051663952,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1874,
+      "step": 9590
+    },
+    {
+      "epoch": 0.5585292064230859,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.2006,
+      "step": 9600
+    },
+    {
+      "epoch": 0.5591110076797766,
+      "grad_norm": 1.78125,
+      "learning_rate": 2e-05,
+      "loss": 0.1976,
+      "step": 9610
+    },
+    {
+      "epoch": 0.5596928089364673,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1918,
+      "step": 9620
+    },
+    {
+      "epoch": 0.560274610193158,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1916,
+      "step": 9630
+    },
+    {
+      "epoch": 0.5608564114498488,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.1925,
+      "step": 9640
+    },
+    {
+      "epoch": 0.5614382127065395,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.2056,
+      "step": 9650
+    },
+    {
+      "epoch": 0.5620200139632302,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1989,
+      "step": 9660
+    },
+    {
+      "epoch": 0.5626018152199209,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.2061,
+      "step": 9670
+    },
+    {
+      "epoch": 0.5631836164766116,
+      "grad_norm": 1.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1993,
+      "step": 9680
+    },
+    {
+      "epoch": 0.5637654177333024,
+      "grad_norm": 1.3515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1957,
+      "step": 9690
+    },
+    {
+      "epoch": 0.5643472189899931,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1978,
+      "step": 9700
+    },
+    {
+      "epoch": 0.5649290202466837,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1918,
+      "step": 9710
+    },
+    {
+      "epoch": 0.5655108215033744,
+      "grad_norm": 1.3515625,
+      "learning_rate": 2e-05,
+      "loss": 0.2032,
+      "step": 9720
+    },
+    {
+      "epoch": 0.5660926227600651,
+      "grad_norm": 1.953125,
+      "learning_rate": 2e-05,
+      "loss": 0.1951,
+      "step": 9730
+    },
+    {
+      "epoch": 0.5666744240167558,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1994,
+      "step": 9740
+    },
+    {
+      "epoch": 0.5672562252734465,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2e-05,
+      "loss": 0.2014,
+      "step": 9750
+    },
+    {
+      "epoch": 0.5678380265301373,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1957,
+      "step": 9760
+    },
+    {
+      "epoch": 0.568419827786828,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2084,
+      "step": 9770
+    },
+    {
+      "epoch": 0.5690016290435187,
+      "grad_norm": 1.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1968,
+      "step": 9780
+    },
+    {
+      "epoch": 0.5695834303002094,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.196,
+      "step": 9790
+    },
+    {
+      "epoch": 0.5701652315569001,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1956,
+      "step": 9800
+    },
+    {
+      "epoch": 0.5707470328135908,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.2006,
+      "step": 9810
+    },
+    {
+      "epoch": 0.5713288340702816,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.1954,
+      "step": 9820
+    },
+    {
+      "epoch": 0.5719106353269723,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.2012,
+      "step": 9830
+    },
+    {
+      "epoch": 0.572492436583663,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1908,
+      "step": 9840
+    },
+    {
+      "epoch": 0.5730742378403537,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1983,
+      "step": 9850
+    },
+    {
+      "epoch": 0.5736560390970444,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.2031,
+      "step": 9860
+    },
+    {
+      "epoch": 0.5742378403537352,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1971,
+      "step": 9870
+    },
+    {
+      "epoch": 0.5748196416104259,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1965,
+      "step": 9880
+    },
+    {
+      "epoch": 0.5754014428671166,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.1993,
+      "step": 9890
+    },
+    {
+      "epoch": 0.5759832441238073,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.1979,
+      "step": 9900
+    },
+    {
+      "epoch": 0.576565045380498,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1966,
+      "step": 9910
+    },
+    {
+      "epoch": 0.5771468466371887,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1889,
+      "step": 9920
+    },
+    {
+      "epoch": 0.5777286478938795,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1955,
+      "step": 9930
+    },
+    {
+      "epoch": 0.5783104491505702,
+      "grad_norm": 1.78125,
+      "learning_rate": 2e-05,
+      "loss": 0.1942,
+      "step": 9940
+    },
+    {
+      "epoch": 0.5788922504072609,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.2002,
+      "step": 9950
+    },
+    {
+      "epoch": 0.5794740516639516,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1961,
+      "step": 9960
+    },
+    {
+      "epoch": 0.5800558529206423,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.1976,
+      "step": 9970
+    },
+    {
+      "epoch": 0.580637654177333,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1981,
+      "step": 9980
+    },
+    {
+      "epoch": 0.5812194554340238,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.1927,
+      "step": 9990
+    },
+    {
+      "epoch": 0.5818012566907145,
+      "grad_norm": 1.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.1866,
+      "step": 10000
+    },
+    {
+      "epoch": 0.5823830579474052,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1975,
+      "step": 10010
+    },
+    {
+      "epoch": 0.5829648592040959,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.1944,
+      "step": 10020
+    },
+    {
+      "epoch": 0.5835466604607866,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2e-05,
+      "loss": 0.1929,
+      "step": 10030
+    },
+    {
+      "epoch": 0.5841284617174773,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 10040
+    },
+    {
+      "epoch": 0.5847102629741681,
+      "grad_norm": 1.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1965,
+      "step": 10050
+    },
+    {
+      "epoch": 0.5852920642308588,
+      "grad_norm": 1.78125,
+      "learning_rate": 2e-05,
+      "loss": 0.1919,
+      "step": 10060
+    },
+    {
+      "epoch": 0.5858738654875495,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1991,
+      "step": 10070
+    },
+    {
+      "epoch": 0.5864556667442402,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1966,
+      "step": 10080
+    },
+    {
+      "epoch": 0.5870374680009309,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1921,
+      "step": 10090
+    },
+    {
+      "epoch": 0.5876192692576216,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.202,
+      "step": 10100
+    },
+    {
+      "epoch": 0.5882010705143124,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1992,
+      "step": 10110
+    },
+    {
+      "epoch": 0.5887828717710031,
+      "grad_norm": 1.9921875,
+      "learning_rate": 2e-05,
+      "loss": 0.1981,
+      "step": 10120
+    },
+    {
+      "epoch": 0.5893646730276937,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 10130
+    },
+    {
+      "epoch": 0.5899464742843844,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1949,
+      "step": 10140
+    },
+    {
+      "epoch": 0.5905282755410751,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1994,
+      "step": 10150
+    },
+    {
+      "epoch": 0.5911100767977658,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.1982,
+      "step": 10160
+    },
+    {
+      "epoch": 0.5916918780544566,
+      "grad_norm": 1.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1963,
+      "step": 10170
+    },
+    {
+      "epoch": 0.5922736793111473,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.1936,
+      "step": 10180
+    },
+    {
+      "epoch": 0.592855480567838,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.1975,
+      "step": 10190
+    },
+    {
+      "epoch": 0.5934372818245287,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2004,
+      "step": 10200
+    },
+    {
+      "epoch": 0.5940190830812194,
+      "grad_norm": 2.09375,
+      "learning_rate": 2e-05,
+      "loss": 0.1942,
+      "step": 10210
+    },
+    {
+      "epoch": 0.5946008843379101,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1967,
+      "step": 10220
+    },
+    {
+      "epoch": 0.5951826855946009,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.1941,
+      "step": 10230
+    },
+    {
+      "epoch": 0.5957644868512916,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.1945,
+      "step": 10240
+    },
+    {
+      "epoch": 0.5963462881079823,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1944,
+      "step": 10250
+    },
+    {
+      "epoch": 0.596928089364673,
+      "grad_norm": 1.2890625,
+      "learning_rate": 2e-05,
+      "loss": 0.2036,
+      "step": 10260
+    },
+    {
+      "epoch": 0.5975098906213637,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1988,
+      "step": 10270
+    },
+    {
+      "epoch": 0.5980916918780544,
+      "grad_norm": 1.296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1969,
+      "step": 10280
+    },
+    {
+      "epoch": 0.5986734931347452,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1881,
+      "step": 10290
+    },
+    {
+      "epoch": 0.5992552943914359,
+      "grad_norm": 2.125,
+      "learning_rate": 2e-05,
+      "loss": 0.1914,
+      "step": 10300
+    },
+    {
+      "epoch": 0.5998370956481266,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2e-05,
+      "loss": 0.199,
+      "step": 10310
+    },
+    {
+      "epoch": 0.6004188969048173,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.2019,
+      "step": 10320
+    },
+    {
+      "epoch": 0.601000698161508,
+      "grad_norm": 1.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.1993,
+      "step": 10330
+    },
+    {
+      "epoch": 0.6015824994181987,
+      "grad_norm": 1.8671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2006,
+      "step": 10340
+    },
+    {
+      "epoch": 0.6021643006748895,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.1949,
+      "step": 10350
+    },
+    {
+      "epoch": 0.6027461019315802,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1885,
+      "step": 10360
+    },
+    {
+      "epoch": 0.6033279031882709,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.2039,
+      "step": 10370
+    },
+    {
+      "epoch": 0.6039097044449616,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1868,
+      "step": 10380
+    },
+    {
+      "epoch": 0.6044915057016523,
+      "grad_norm": 1.90625,
+      "learning_rate": 2e-05,
+      "loss": 0.2036,
+      "step": 10390
+    },
+    {
+      "epoch": 0.605073306958343,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2035,
+      "step": 10400
+    },
+    {
+      "epoch": 0.6056551082150338,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1957,
+      "step": 10410
+    },
+    {
+      "epoch": 0.6062369094717245,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.195,
+      "step": 10420
+    },
+    {
+      "epoch": 0.6068187107284152,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.1988,
+      "step": 10430
+    },
+    {
+      "epoch": 0.6074005119851059,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.1918,
+      "step": 10440
+    },
+    {
+      "epoch": 0.6079823132417966,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1978,
+      "step": 10450
+    },
+    {
+      "epoch": 0.6085641144984874,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2e-05,
+      "loss": 0.1996,
+      "step": 10460
+    },
+    {
+      "epoch": 0.6091459157551781,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2e-05,
+      "loss": 0.2012,
+      "step": 10470
+    },
+    {
+      "epoch": 0.6097277170118688,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1911,
+      "step": 10480
+    },
+    {
+      "epoch": 0.6103095182685595,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.1915,
+      "step": 10490
+    },
+    {
+      "epoch": 0.6108913195252502,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1956,
+      "step": 10500
+    },
+    {
+      "epoch": 0.6114731207819409,
+      "grad_norm": 1.75,
+      "learning_rate": 2e-05,
+      "loss": 0.1987,
+      "step": 10510
+    },
+    {
+      "epoch": 0.6120549220386317,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.2038,
+      "step": 10520
+    },
+    {
+      "epoch": 0.6126367232953224,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1991,
+      "step": 10530
+    },
+    {
+      "epoch": 0.613218524552013,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1942,
+      "step": 10540
+    },
+    {
+      "epoch": 0.6138003258087037,
+      "grad_norm": 2.09375,
+      "learning_rate": 2e-05,
+      "loss": 0.1916,
+      "step": 10550
+    },
+    {
+      "epoch": 0.6143821270653944,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 10560
+    },
+    {
+      "epoch": 0.6149639283220851,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1904,
+      "step": 10570
+    },
+    {
+      "epoch": 0.6155457295787758,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.19,
+      "step": 10580
+    },
+    {
+      "epoch": 0.6161275308354666,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1928,
+      "step": 10590
+    },
+    {
+      "epoch": 0.6167093320921573,
+      "grad_norm": 2.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.1963,
+      "step": 10600
+    },
+    {
+      "epoch": 0.617291133348848,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.1951,
+      "step": 10610
+    },
+    {
+      "epoch": 0.6178729346055387,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1987,
+      "step": 10620
+    },
+    {
+      "epoch": 0.6184547358622294,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1948,
+      "step": 10630
+    },
+    {
+      "epoch": 0.6190365371189201,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1929,
+      "step": 10640
+    },
+    {
+      "epoch": 0.6196183383756109,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.1983,
+      "step": 10650
+    },
+    {
+      "epoch": 0.6202001396323016,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.2026,
+      "step": 10660
+    },
+    {
+      "epoch": 0.6207819408889923,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.1898,
+      "step": 10670
+    },
+    {
+      "epoch": 0.621363742145683,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1921,
+      "step": 10680
+    },
+    {
+      "epoch": 0.6219455434023737,
+      "grad_norm": 2.125,
+      "learning_rate": 2e-05,
+      "loss": 0.194,
+      "step": 10690
+    },
+    {
+      "epoch": 0.6225273446590645,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.1968,
+      "step": 10700
+    },
+    {
+      "epoch": 0.6231091459157552,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.196,
+      "step": 10710
+    },
+    {
+      "epoch": 0.6236909471724459,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.1954,
+      "step": 10720
+    },
+    {
+      "epoch": 0.6242727484291366,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.2009,
+      "step": 10730
+    },
+    {
+      "epoch": 0.6248545496858273,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1904,
+      "step": 10740
+    },
+    {
+      "epoch": 0.625436350942518,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1912,
+      "step": 10750
+    },
+    {
+      "epoch": 0.6260181521992088,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.1967,
+      "step": 10760
+    },
+    {
+      "epoch": 0.6265999534558995,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2025,
+      "step": 10770
+    },
+    {
+      "epoch": 0.6271817547125902,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.1928,
+      "step": 10780
+    },
+    {
+      "epoch": 0.6277635559692809,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1945,
+      "step": 10790
+    },
+    {
+      "epoch": 0.6283453572259716,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1964,
+      "step": 10800
+    },
+    {
+      "epoch": 0.6289271584826623,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2e-05,
+      "loss": 0.194,
+      "step": 10810
+    },
+    {
+      "epoch": 0.6295089597393531,
+      "grad_norm": 1.828125,
+      "learning_rate": 2e-05,
+      "loss": 0.2006,
+      "step": 10820
+    },
+    {
+      "epoch": 0.6300907609960438,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1941,
+      "step": 10830
+    },
+    {
+      "epoch": 0.6306725622527345,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.1911,
+      "step": 10840
+    },
+    {
+      "epoch": 0.6312543635094252,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1847,
+      "step": 10850
+    },
+    {
+      "epoch": 0.6318361647661159,
+      "grad_norm": 1.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1973,
+      "step": 10860
+    },
+    {
+      "epoch": 0.6324179660228066,
+      "grad_norm": 1.2578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1971,
+      "step": 10870
+    },
+    {
+      "epoch": 0.6329997672794974,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.1932,
+      "step": 10880
+    },
+    {
+      "epoch": 0.6335815685361881,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1976,
+      "step": 10890
+    },
+    {
+      "epoch": 0.6341633697928788,
+      "grad_norm": 2.09375,
+      "learning_rate": 2e-05,
+      "loss": 0.1946,
+      "step": 10900
+    },
+    {
+      "epoch": 0.6347451710495695,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1941,
+      "step": 10910
+    },
+    {
+      "epoch": 0.6353269723062602,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1982,
+      "step": 10920
+    },
+    {
+      "epoch": 0.635908773562951,
+      "grad_norm": 1.375,
+      "learning_rate": 2e-05,
+      "loss": 0.1945,
+      "step": 10930
+    },
+    {
+      "epoch": 0.6364905748196417,
+      "grad_norm": 1.9921875,
+      "learning_rate": 2e-05,
+      "loss": 0.1909,
+      "step": 10940
+    },
+    {
+      "epoch": 0.6370723760763323,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1939,
+      "step": 10950
+    },
+    {
+      "epoch": 0.637654177333023,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.1909,
+      "step": 10960
+    },
+    {
+      "epoch": 0.6382359785897137,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1961,
+      "step": 10970
+    },
+    {
+      "epoch": 0.6388177798464044,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.1965,
+      "step": 10980
+    },
+    {
+      "epoch": 0.6393995811030951,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.194,
+      "step": 10990
+    },
+    {
+      "epoch": 0.6399813823597859,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.1931,
+      "step": 11000
+    },
+    {
+      "epoch": 0.6405631836164766,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1987,
+      "step": 11010
+    },
+    {
+      "epoch": 0.6411449848731673,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1927,
+      "step": 11020
+    },
+    {
+      "epoch": 0.641726786129858,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2,
+      "step": 11030
+    },
+    {
+      "epoch": 0.6423085873865487,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1961,
+      "step": 11040
+    },
+    {
+      "epoch": 0.6428903886432394,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1944,
+      "step": 11050
+    },
+    {
+      "epoch": 0.6434721898999302,
+      "grad_norm": 2.265625,
+      "learning_rate": 2e-05,
+      "loss": 0.2021,
+      "step": 11060
+    },
+    {
+      "epoch": 0.6440539911566209,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1967,
+      "step": 11070
+    },
+    {
+      "epoch": 0.6446357924133116,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1922,
+      "step": 11080
+    },
+    {
+      "epoch": 0.6452175936700023,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1909,
+      "step": 11090
+    },
+    {
+      "epoch": 0.645799394926693,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.185,
+      "step": 11100
+    },
+    {
+      "epoch": 0.6463811961833837,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 11110
+    },
+    {
+      "epoch": 0.6469629974400745,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.1938,
+      "step": 11120
+    },
+    {
+      "epoch": 0.6475447986967652,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1929,
+      "step": 11130
+    },
+    {
+      "epoch": 0.6481265999534559,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.2016,
+      "step": 11140
+    },
+    {
+      "epoch": 0.6487084012101466,
+      "grad_norm": 1.9765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1955,
+      "step": 11150
+    },
+    {
+      "epoch": 0.6492902024668373,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.2007,
+      "step": 11160
+    },
+    {
+      "epoch": 0.649872003723528,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1922,
+      "step": 11170
+    },
+    {
+      "epoch": 0.6504538049802188,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.199,
+      "step": 11180
+    },
+    {
+      "epoch": 0.6510356062369095,
+      "grad_norm": 1.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.1949,
+      "step": 11190
+    },
+    {
+      "epoch": 0.6516174074936002,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2e-05,
+      "loss": 0.1989,
+      "step": 11200
+    },
+    {
+      "epoch": 0.6521992087502909,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2e-05,
+      "loss": 0.1896,
+      "step": 11210
+    },
+    {
+      "epoch": 0.6527810100069816,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.1991,
+      "step": 11220
+    },
+    {
+      "epoch": 0.6533628112636723,
+      "grad_norm": 1.90625,
+      "learning_rate": 2e-05,
+      "loss": 0.2002,
+      "step": 11230
+    },
+    {
+      "epoch": 0.6539446125203631,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1963,
+      "step": 11240
+    },
+    {
+      "epoch": 0.6545264137770538,
+      "grad_norm": 1.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1899,
+      "step": 11250
+    },
+    {
+      "epoch": 0.6551082150337445,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.2011,
+      "step": 11260
+    },
+    {
+      "epoch": 0.6556900162904352,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2e-05,
+      "loss": 0.1949,
+      "step": 11270
+    },
+    {
+      "epoch": 0.6562718175471259,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1867,
+      "step": 11280
+    },
+    {
+      "epoch": 0.6568536188038167,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.1981,
+      "step": 11290
+    },
+    {
+      "epoch": 0.6574354200605074,
+      "grad_norm": 1.90625,
+      "learning_rate": 2e-05,
+      "loss": 0.1922,
+      "step": 11300
+    },
+    {
+      "epoch": 0.6580172213171981,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2e-05,
+      "loss": 0.2005,
+      "step": 11310
+    },
+    {
+      "epoch": 0.6585990225738888,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1973,
+      "step": 11320
+    },
+    {
+      "epoch": 0.6591808238305795,
+      "grad_norm": 1.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1991,
+      "step": 11330
+    },
+    {
+      "epoch": 0.6597626250872702,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1994,
+      "step": 11340
+    },
+    {
+      "epoch": 0.660344426343961,
+      "grad_norm": 1.75,
+      "learning_rate": 2e-05,
+      "loss": 0.1977,
+      "step": 11350
+    },
+    {
+      "epoch": 0.6609262276006517,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.1952,
+      "step": 11360
+    },
+    {
+      "epoch": 0.6615080288573423,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1918,
+      "step": 11370
+    },
+    {
+      "epoch": 0.662089830114033,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.1874,
+      "step": 11380
+    },
+    {
+      "epoch": 0.6626716313707237,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1966,
+      "step": 11390
+    },
+    {
+      "epoch": 0.6632534326274144,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1953,
+      "step": 11400
+    },
+    {
+      "epoch": 0.6638352338841051,
+      "grad_norm": 1.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1884,
+      "step": 11410
+    },
+    {
+      "epoch": 0.6644170351407959,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.1903,
+      "step": 11420
+    },
+    {
+      "epoch": 0.6649988363974866,
+      "grad_norm": 1.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1918,
+      "step": 11430
+    },
+    {
+      "epoch": 0.6655806376541773,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1956,
+      "step": 11440
+    },
+    {
+      "epoch": 0.666162438910868,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.1924,
+      "step": 11450
+    },
+    {
+      "epoch": 0.6667442401675587,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.1874,
+      "step": 11460
+    },
+    {
+      "epoch": 0.6673260414242494,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.1948,
+      "step": 11470
+    },
+    {
+      "epoch": 0.6679078426809402,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1908,
+      "step": 11480
+    },
+    {
+      "epoch": 0.6684896439376309,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1921,
+      "step": 11490
+    },
+    {
+      "epoch": 0.6690714451943216,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2e-05,
+      "loss": 0.1897,
+      "step": 11500
+    },
+    {
+      "epoch": 0.6696532464510123,
+      "grad_norm": 1.25,
+      "learning_rate": 2e-05,
+      "loss": 0.1964,
+      "step": 11510
+    },
+    {
+      "epoch": 0.670235047707703,
+      "grad_norm": 1.3515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1938,
+      "step": 11520
+    },
+    {
+      "epoch": 0.6708168489643938,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1921,
+      "step": 11530
+    },
+    {
+      "epoch": 0.6713986502210845,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.1881,
+      "step": 11540
+    },
+    {
+      "epoch": 0.6719804514777752,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 11550
+    },
+    {
+      "epoch": 0.6725622527344659,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.1948,
+      "step": 11560
+    },
+    {
+      "epoch": 0.6731440539911566,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.1885,
+      "step": 11570
+    },
+    {
+      "epoch": 0.6737258552478473,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.1906,
+      "step": 11580
+    },
+    {
+      "epoch": 0.674307656504538,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.1991,
+      "step": 11590
+    },
+    {
+      "epoch": 0.6748894577612288,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2e-05,
+      "loss": 0.189,
+      "step": 11600
+    },
+    {
+      "epoch": 0.6754712590179195,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 11610
+    },
+    {
+      "epoch": 0.6760530602746102,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2e-05,
+      "loss": 0.2006,
+      "step": 11620
+    },
+    {
+      "epoch": 0.6766348615313009,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1909,
+      "step": 11630
+    },
+    {
+      "epoch": 0.6772166627879916,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1944,
+      "step": 11640
+    },
+    {
+      "epoch": 0.6777984640446824,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1982,
+      "step": 11650
+    },
+    {
+      "epoch": 0.6783802653013731,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.1961,
+      "step": 11660
+    },
+    {
+      "epoch": 0.6789620665580638,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.1877,
+      "step": 11670
+    },
+    {
+      "epoch": 0.6795438678147545,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1929,
+      "step": 11680
+    },
+    {
+      "epoch": 0.6801256690714452,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1915,
+      "step": 11690
+    },
+    {
+      "epoch": 0.6807074703281359,
+      "grad_norm": 1.375,
+      "learning_rate": 2e-05,
+      "loss": 0.19,
+      "step": 11700
+    },
+    {
+      "epoch": 0.6812892715848267,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1923,
+      "step": 11710
+    },
+    {
+      "epoch": 0.6818710728415174,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1969,
+      "step": 11720
+    },
+    {
+      "epoch": 0.6824528740982081,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1876,
+      "step": 11730
+    },
+    {
+      "epoch": 0.6830346753548988,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.1919,
+      "step": 11740
+    },
+    {
+      "epoch": 0.6836164766115895,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1939,
+      "step": 11750
+    },
+    {
+      "epoch": 0.6841982778682802,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1953,
+      "step": 11760
+    },
+    {
+      "epoch": 0.684780079124971,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.196,
+      "step": 11770
+    },
+    {
+      "epoch": 0.6853618803816616,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.1975,
+      "step": 11780
+    },
+    {
+      "epoch": 0.6859436816383523,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2e-05,
+      "loss": 0.1965,
+      "step": 11790
+    },
+    {
+      "epoch": 0.686525482895043,
+      "grad_norm": 1.375,
+      "learning_rate": 2e-05,
+      "loss": 0.1993,
+      "step": 11800
+    },
+    {
+      "epoch": 0.6871072841517337,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1959,
+      "step": 11810
+    },
+    {
+      "epoch": 0.6876890854084244,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.1962,
+      "step": 11820
+    },
+    {
+      "epoch": 0.6882708866651152,
+      "grad_norm": 1.375,
+      "learning_rate": 2e-05,
+      "loss": 0.1973,
+      "step": 11830
+    },
+    {
+      "epoch": 0.6888526879218059,
+      "grad_norm": 1.265625,
+      "learning_rate": 2e-05,
+      "loss": 0.1908,
+      "step": 11840
+    },
+    {
+      "epoch": 0.6894344891784966,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1892,
+      "step": 11850
+    },
+    {
+      "epoch": 0.6900162904351873,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2e-05,
+      "loss": 0.1985,
+      "step": 11860
+    },
+    {
+      "epoch": 0.690598091691878,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.1862,
+      "step": 11870
+    },
+    {
+      "epoch": 0.6911798929485687,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.1974,
+      "step": 11880
+    },
+    {
+      "epoch": 0.6917616942052595,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1938,
+      "step": 11890
+    },
+    {
+      "epoch": 0.6923434954619502,
+      "grad_norm": 1.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.1933,
+      "step": 11900
+    },
+    {
+      "epoch": 0.6929252967186409,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.194,
+      "step": 11910
+    },
+    {
+      "epoch": 0.6935070979753316,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.1906,
+      "step": 11920
+    },
+    {
+      "epoch": 0.6940888992320223,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.1994,
+      "step": 11930
+    },
+    {
+      "epoch": 0.694670700488713,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.1944,
+      "step": 11940
+    },
+    {
+      "epoch": 0.6952525017454038,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.1994,
+      "step": 11950
+    },
+    {
+      "epoch": 0.6958343030020945,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 11960
+    },
+    {
+      "epoch": 0.6964161042587852,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1899,
+      "step": 11970
+    },
+    {
+      "epoch": 0.6969979055154759,
+      "grad_norm": 1.1875,
+      "learning_rate": 2e-05,
+      "loss": 0.1932,
+      "step": 11980
+    },
+    {
+      "epoch": 0.6975797067721666,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.2018,
+      "step": 11990
+    },
+    {
+      "epoch": 0.6981615080288573,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1967,
+      "step": 12000
+    },
+    {
+      "epoch": 0.6987433092855481,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1952,
+      "step": 12010
+    },
+    {
+      "epoch": 0.6993251105422388,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.1867,
+      "step": 12020
+    },
+    {
+      "epoch": 0.6999069117989295,
+      "grad_norm": 1.8515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1951,
+      "step": 12030
+    },
+    {
+      "epoch": 0.7004887130556202,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1906,
+      "step": 12040
+    },
+    {
+      "epoch": 0.7010705143123109,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.1943,
+      "step": 12050
+    },
+    {
+      "epoch": 0.7016523155690016,
+      "grad_norm": 1.34375,
+      "learning_rate": 2e-05,
+      "loss": 0.1913,
+      "step": 12060
+    },
+    {
+      "epoch": 0.7022341168256924,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.1924,
+      "step": 12070
+    },
+    {
+      "epoch": 0.7028159180823831,
+      "grad_norm": 1.8046875,
+      "learning_rate": 2e-05,
+      "loss": 0.1917,
+      "step": 12080
+    },
+    {
+      "epoch": 0.7033977193390738,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1951,
+      "step": 12090
+    },
+    {
+      "epoch": 0.7039795205957645,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.192,
+      "step": 12100
+    },
+    {
+      "epoch": 0.7045613218524552,
+      "grad_norm": 1.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.1914,
+      "step": 12110
+    },
+    {
+      "epoch": 0.705143123109146,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.189,
+      "step": 12120
+    },
+    {
+      "epoch": 0.7057249243658367,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.1956,
+      "step": 12130
+    },
+    {
+      "epoch": 0.7063067256225274,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1871,
+      "step": 12140
+    },
+    {
+      "epoch": 0.7068885268792181,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1923,
+      "step": 12150
+    },
+    {
+      "epoch": 0.7074703281359088,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1964,
+      "step": 12160
+    },
+    {
+      "epoch": 0.7080521293925995,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1907,
+      "step": 12170
+    },
+    {
+      "epoch": 0.7086339306492903,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.189,
+      "step": 12180
+    },
+    {
+      "epoch": 0.7092157319059809,
+      "grad_norm": 1.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.1992,
+      "step": 12190
+    },
+    {
+      "epoch": 0.7097975331626716,
+      "grad_norm": 1.3515625,
+      "learning_rate": 2e-05,
+      "loss": 0.193,
+      "step": 12200
+    },
+    {
+      "epoch": 0.7103793344193623,
+      "grad_norm": 1.2578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1986,
+      "step": 12210
+    },
+    {
+      "epoch": 0.710961135676053,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1948,
+      "step": 12220
+    },
+    {
+      "epoch": 0.7115429369327437,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1925,
+      "step": 12230
+    },
+    {
+      "epoch": 0.7121247381894344,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1895,
+      "step": 12240
+    },
+    {
+      "epoch": 0.7127065394461252,
+      "grad_norm": 3.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1859,
+      "step": 12250
+    },
+    {
+      "epoch": 0.7132883407028159,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.187,
+      "step": 12260
+    },
+    {
+      "epoch": 0.7138701419595066,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.1889,
+      "step": 12270
+    },
+    {
+      "epoch": 0.7144519432161973,
+      "grad_norm": 1.3515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1868,
+      "step": 12280
+    },
+    {
+      "epoch": 0.715033744472888,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.2027,
+      "step": 12290
+    },
+    {
+      "epoch": 0.7156155457295788,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.1981,
+      "step": 12300
+    },
+    {
+      "epoch": 0.7161973469862695,
+      "grad_norm": 1.34375,
+      "learning_rate": 2e-05,
+      "loss": 0.189,
+      "step": 12310
+    },
+    {
+      "epoch": 0.7167791482429602,
+      "grad_norm": 1.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1937,
+      "step": 12320
+    },
+    {
+      "epoch": 0.7173609494996509,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1842,
+      "step": 12330
+    },
+    {
+      "epoch": 0.7179427507563416,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.2,
+      "step": 12340
+    },
+    {
+      "epoch": 0.7185245520130323,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1907,
+      "step": 12350
+    },
+    {
+      "epoch": 0.719106353269723,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.1923,
+      "step": 12360
+    },
+    {
+      "epoch": 0.7196881545264138,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.1911,
+      "step": 12370
+    },
+    {
+      "epoch": 0.7202699557831045,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1887,
+      "step": 12380
+    },
+    {
+      "epoch": 0.7208517570397952,
+      "grad_norm": 1.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.1894,
+      "step": 12390
+    },
+    {
+      "epoch": 0.7214335582964859,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 12400
+    },
+    {
+      "epoch": 0.7220153595531766,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1942,
+      "step": 12410
+    },
+    {
+      "epoch": 0.7225971608098674,
+      "grad_norm": 1.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1935,
+      "step": 12420
+    },
+    {
+      "epoch": 0.7231789620665581,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1869,
+      "step": 12430
+    },
+    {
+      "epoch": 0.7237607633232488,
+      "grad_norm": 1.2578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1995,
+      "step": 12440
+    },
+    {
+      "epoch": 0.7243425645799395,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1874,
+      "step": 12450
+    },
+    {
+      "epoch": 0.7249243658366302,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.191,
+      "step": 12460
+    },
+    {
+      "epoch": 0.7255061670933209,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1912,
+      "step": 12470
+    },
+    {
+      "epoch": 0.7260879683500117,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.1865,
+      "step": 12480
+    },
+    {
+      "epoch": 0.7266697696067024,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.192,
+      "step": 12490
+    },
+    {
+      "epoch": 0.7272515708633931,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1869,
+      "step": 12500
+    },
+    {
+      "epoch": 0.7278333721200838,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1923,
+      "step": 12510
+    },
+    {
+      "epoch": 0.7284151733767745,
+      "grad_norm": 1.875,
+      "learning_rate": 2e-05,
+      "loss": 0.1896,
+      "step": 12520
+    },
+    {
+      "epoch": 0.7289969746334652,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1864,
+      "step": 12530
+    },
+    {
+      "epoch": 0.729578775890156,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1922,
+      "step": 12540
+    },
+    {
+      "epoch": 0.7301605771468467,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 12550
+    },
+    {
+      "epoch": 0.7307423784035374,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1973,
+      "step": 12560
+    },
+    {
+      "epoch": 0.7313241796602281,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.186,
+      "step": 12570
+    },
+    {
+      "epoch": 0.7319059809169188,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2e-05,
+      "loss": 0.1958,
+      "step": 12580
+    },
+    {
+      "epoch": 0.7324877821736095,
+      "grad_norm": 1.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.191,
+      "step": 12590
+    },
+    {
+      "epoch": 0.7330695834303003,
+      "grad_norm": 1.265625,
+      "learning_rate": 2e-05,
+      "loss": 0.1886,
+      "step": 12600
+    },
+    {
+      "epoch": 0.7336513846869909,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.1925,
+      "step": 12610
+    },
+    {
+      "epoch": 0.7342331859436816,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1922,
+      "step": 12620
+    },
+    {
+      "epoch": 0.7348149872003723,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1931,
+      "step": 12630
+    },
+    {
+      "epoch": 0.735396788457063,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1939,
+      "step": 12640
+    },
+    {
+      "epoch": 0.7359785897137537,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1827,
+      "step": 12650
+    },
+    {
+      "epoch": 0.7365603909704445,
+      "grad_norm": 2.3125,
+      "learning_rate": 2e-05,
+      "loss": 0.1921,
+      "step": 12660
+    },
+    {
+      "epoch": 0.7371421922271352,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1901,
+      "step": 12670
+    },
+    {
+      "epoch": 0.7377239934838259,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1859,
+      "step": 12680
+    },
+    {
+      "epoch": 0.7383057947405166,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.196,
+      "step": 12690
+    },
+    {
+      "epoch": 0.7388875959972073,
+      "grad_norm": 1.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1915,
+      "step": 12700
+    },
+    {
+      "epoch": 0.739469397253898,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1862,
+      "step": 12710
+    },
+    {
+      "epoch": 0.7400511985105888,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1976,
+      "step": 12720
+    },
+    {
+      "epoch": 0.7406329997672795,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1869,
+      "step": 12730
+    },
+    {
+      "epoch": 0.7412148010239702,
+      "grad_norm": 1.25,
+      "learning_rate": 2e-05,
+      "loss": 0.1931,
+      "step": 12740
+    },
+    {
+      "epoch": 0.7417966022806609,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1889,
+      "step": 12750
+    },
+    {
+      "epoch": 0.7423784035373516,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1878,
+      "step": 12760
+    },
+    {
+      "epoch": 0.7429602047940423,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1846,
+      "step": 12770
+    },
+    {
+      "epoch": 0.7435420060507331,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.1889,
+      "step": 12780
+    },
+    {
+      "epoch": 0.7441238073074238,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1863,
+      "step": 12790
+    },
+    {
+      "epoch": 0.7447056085641145,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.198,
+      "step": 12800
+    },
+    {
+      "epoch": 0.7452874098208052,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.1848,
+      "step": 12810
+    },
+    {
+      "epoch": 0.7458692110774959,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.1983,
+      "step": 12820
+    },
+    {
+      "epoch": 0.7464510123341866,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1902,
+      "step": 12830
+    },
+    {
+      "epoch": 0.7470328135908774,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1951,
+      "step": 12840
+    },
+    {
+      "epoch": 0.7476146148475681,
+      "grad_norm": 2.796875,
+      "learning_rate": 2e-05,
+      "loss": 0.1877,
+      "step": 12850
+    },
+    {
+      "epoch": 0.7481964161042588,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1916,
+      "step": 12860
+    },
+    {
+      "epoch": 0.7487782173609495,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.1944,
+      "step": 12870
+    },
+    {
+      "epoch": 0.7493600186176402,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2e-05,
+      "loss": 0.1874,
+      "step": 12880
+    },
+    {
+      "epoch": 0.749941819874331,
+      "grad_norm": 2.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1953,
+      "step": 12890
+    },
+    {
+      "epoch": 0.7505236211310217,
+      "grad_norm": 1.265625,
+      "learning_rate": 2e-05,
+      "loss": 0.1946,
+      "step": 12900
+    },
+    {
+      "epoch": 0.7511054223877124,
+      "grad_norm": 1.2109375,
+      "learning_rate": 2e-05,
+      "loss": 0.1905,
+      "step": 12910
+    },
+    {
+      "epoch": 0.7516872236444031,
+      "grad_norm": 1.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1876,
+      "step": 12920
+    },
+    {
+      "epoch": 0.7522690249010938,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1909,
+      "step": 12930
+    },
+    {
+      "epoch": 0.7528508261577845,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2e-05,
+      "loss": 0.1899,
+      "step": 12940
+    },
+    {
+      "epoch": 0.7534326274144753,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.1895,
+      "step": 12950
+    },
+    {
+      "epoch": 0.754014428671166,
+      "grad_norm": 1.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1822,
+      "step": 12960
+    },
+    {
+      "epoch": 0.7545962299278567,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1909,
+      "step": 12970
+    },
+    {
+      "epoch": 0.7551780311845474,
+      "grad_norm": 1.65625,
+      "learning_rate": 2e-05,
+      "loss": 0.1845,
+      "step": 12980
+    },
+    {
+      "epoch": 0.7557598324412381,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.1923,
+      "step": 12990
+    },
+    {
+      "epoch": 0.7563416336979288,
+      "grad_norm": 1.953125,
+      "learning_rate": 2e-05,
+      "loss": 0.1929,
+      "step": 13000
+    },
+    {
+      "epoch": 0.7569234349546196,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1915,
+      "step": 13010
+    },
+    {
+      "epoch": 0.7575052362113102,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.1893,
+      "step": 13020
+    },
+    {
+      "epoch": 0.7580870374680009,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.2014,
+      "step": 13030
+    },
+    {
+      "epoch": 0.7586688387246916,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1985,
+      "step": 13040
+    },
+    {
+      "epoch": 0.7592506399813823,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1919,
+      "step": 13050
+    },
+    {
+      "epoch": 0.759832441238073,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1892,
+      "step": 13060
+    },
+    {
+      "epoch": 0.7604142424947637,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1849,
+      "step": 13070
+    },
+    {
+      "epoch": 0.7609960437514545,
+      "grad_norm": 1.34375,
+      "learning_rate": 2e-05,
+      "loss": 0.1904,
+      "step": 13080
+    },
+    {
+      "epoch": 0.7615778450081452,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.1896,
+      "step": 13090
+    },
+    {
+      "epoch": 0.7621596462648359,
+      "grad_norm": 1.7890625,
+      "learning_rate": 2e-05,
+      "loss": 0.1891,
+      "step": 13100
+    },
+    {
+      "epoch": 0.7627414475215266,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1914,
+      "step": 13110
+    },
+    {
+      "epoch": 0.7633232487782173,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.188,
+      "step": 13120
+    },
+    {
+      "epoch": 0.763905050034908,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.2015,
+      "step": 13130
+    },
+    {
+      "epoch": 0.7644868512915988,
+      "grad_norm": 1.375,
+      "learning_rate": 2e-05,
+      "loss": 0.1905,
+      "step": 13140
+    },
+    {
+      "epoch": 0.7650686525482895,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.1894,
+      "step": 13150
+    },
+    {
+      "epoch": 0.7656504538049802,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.1938,
+      "step": 13160
+    },
+    {
+      "epoch": 0.7662322550616709,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1898,
+      "step": 13170
+    },
+    {
+      "epoch": 0.7668140563183616,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1891,
+      "step": 13180
+    },
+    {
+      "epoch": 0.7673958575750524,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1957,
+      "step": 13190
+    },
+    {
+      "epoch": 0.7679776588317431,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1943,
+      "step": 13200
+    },
+    {
+      "epoch": 0.7685594600884338,
+      "grad_norm": 1.25,
+      "learning_rate": 2e-05,
+      "loss": 0.1898,
+      "step": 13210
+    },
+    {
+      "epoch": 0.7691412613451245,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1906,
+      "step": 13220
+    },
+    {
+      "epoch": 0.7697230626018152,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1904,
+      "step": 13230
+    },
+    {
+      "epoch": 0.7703048638585059,
+      "grad_norm": 2.171875,
+      "learning_rate": 2e-05,
+      "loss": 0.1967,
+      "step": 13240
+    },
+    {
+      "epoch": 0.7708866651151967,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.1881,
+      "step": 13250
+    },
+    {
+      "epoch": 0.7714684663718874,
+      "grad_norm": 1.375,
+      "learning_rate": 2e-05,
+      "loss": 0.193,
+      "step": 13260
+    },
+    {
+      "epoch": 0.7720502676285781,
+      "grad_norm": 4.96875,
+      "learning_rate": 2e-05,
+      "loss": 0.1843,
+      "step": 13270
+    },
+    {
+      "epoch": 0.7726320688852688,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1997,
+      "step": 13280
+    },
+    {
+      "epoch": 0.7732138701419595,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 13290
+    },
+    {
+      "epoch": 0.7737956713986502,
+      "grad_norm": 2.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1971,
+      "step": 13300
+    },
+    {
+      "epoch": 0.774377472655341,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1891,
+      "step": 13310
+    },
+    {
+      "epoch": 0.7749592739120317,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.19,
+      "step": 13320
+    },
+    {
+      "epoch": 0.7755410751687224,
+      "grad_norm": 1.1796875,
+      "learning_rate": 2e-05,
+      "loss": 0.1967,
+      "step": 13330
+    },
+    {
+      "epoch": 0.7761228764254131,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.1902,
+      "step": 13340
+    },
+    {
+      "epoch": 0.7767046776821038,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.1891,
+      "step": 13350
+    },
+    {
+      "epoch": 0.7772864789387945,
+      "grad_norm": 1.640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 13360
+    },
+    {
+      "epoch": 0.7778682801954853,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1928,
+      "step": 13370
+    },
+    {
+      "epoch": 0.778450081452176,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1904,
+      "step": 13380
+    },
+    {
+      "epoch": 0.7790318827088667,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2e-05,
+      "loss": 0.1981,
+      "step": 13390
+    },
+    {
+      "epoch": 0.7796136839655574,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1882,
+      "step": 13400
+    },
+    {
+      "epoch": 0.7801954852222481,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1871,
+      "step": 13410
+    },
+    {
+      "epoch": 0.7807772864789388,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1967,
+      "step": 13420
+    },
+    {
+      "epoch": 0.7813590877356295,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1957,
+      "step": 13430
+    },
+    {
+      "epoch": 0.7819408889923202,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1907,
+      "step": 13440
+    },
+    {
+      "epoch": 0.7825226902490109,
+      "grad_norm": 1.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1937,
+      "step": 13450
+    },
+    {
+      "epoch": 0.7831044915057016,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1971,
+      "step": 13460
+    },
+    {
+      "epoch": 0.7836862927623923,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.1944,
+      "step": 13470
+    },
+    {
+      "epoch": 0.784268094019083,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.1879,
+      "step": 13480
+    },
+    {
+      "epoch": 0.7848498952757738,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1884,
+      "step": 13490
+    },
+    {
+      "epoch": 0.7854316965324645,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1849,
+      "step": 13500
+    },
+    {
+      "epoch": 0.7860134977891552,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.1871,
+      "step": 13510
+    },
+    {
+      "epoch": 0.7865952990458459,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.1954,
+      "step": 13520
+    },
+    {
+      "epoch": 0.7871771003025366,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1938,
+      "step": 13530
+    },
+    {
+      "epoch": 0.7877589015592273,
+      "grad_norm": 2.125,
+      "learning_rate": 2e-05,
+      "loss": 0.1921,
+      "step": 13540
+    },
+    {
+      "epoch": 0.7883407028159181,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1841,
+      "step": 13550
+    },
+    {
+      "epoch": 0.7889225040726088,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1892,
+      "step": 13560
+    },
+    {
+      "epoch": 0.7895043053292995,
+      "grad_norm": 2.03125,
+      "learning_rate": 2e-05,
+      "loss": 0.1865,
+      "step": 13570
+    },
+    {
+      "epoch": 0.7900861065859902,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1921,
+      "step": 13580
+    },
+    {
+      "epoch": 0.7906679078426809,
+      "grad_norm": 1.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1855,
+      "step": 13590
+    },
+    {
+      "epoch": 0.7912497090993716,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.1925,
+      "step": 13600
+    },
+    {
+      "epoch": 0.7918315103560624,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.193,
+      "step": 13610
+    },
+    {
+      "epoch": 0.7924133116127531,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.1905,
+      "step": 13620
+    },
+    {
+      "epoch": 0.7929951128694438,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2e-05,
+      "loss": 0.1866,
+      "step": 13630
+    },
+    {
+      "epoch": 0.7935769141261345,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1917,
+      "step": 13640
+    },
+    {
+      "epoch": 0.7941587153828252,
+      "grad_norm": 1.8125,
+      "learning_rate": 2e-05,
+      "loss": 0.1843,
+      "step": 13650
+    },
+    {
+      "epoch": 0.794740516639516,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2e-05,
+      "loss": 0.1933,
+      "step": 13660
+    },
+    {
+      "epoch": 0.7953223178962067,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1946,
+      "step": 13670
+    },
+    {
+      "epoch": 0.7959041191528974,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.197,
+      "step": 13680
+    },
+    {
+      "epoch": 0.7964859204095881,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1929,
+      "step": 13690
+    },
+    {
+      "epoch": 0.7970677216662788,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.1894,
+      "step": 13700
+    },
+    {
+      "epoch": 0.7976495229229695,
+      "grad_norm": 1.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.193,
+      "step": 13710
+    },
+    {
+      "epoch": 0.7982313241796603,
+      "grad_norm": 1.828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1891,
+      "step": 13720
+    },
+    {
+      "epoch": 0.798813125436351,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1817,
+      "step": 13730
+    },
+    {
+      "epoch": 0.7993949266930417,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1896,
+      "step": 13740
+    },
+    {
+      "epoch": 0.7999767279497324,
+      "grad_norm": 1.3515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1872,
+      "step": 13750
+    },
+    {
+      "epoch": 0.8005585292064231,
+      "grad_norm": 1.34375,
+      "learning_rate": 2e-05,
+      "loss": 0.1987,
+      "step": 13760
+    },
+    {
+      "epoch": 0.8011403304631138,
+      "grad_norm": 1.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1835,
+      "step": 13770
+    },
+    {
+      "epoch": 0.8017221317198046,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1898,
+      "step": 13780
+    },
+    {
+      "epoch": 0.8023039329764953,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.193,
+      "step": 13790
+    },
+    {
+      "epoch": 0.802885734233186,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.1867,
+      "step": 13800
+    },
+    {
+      "epoch": 0.8034675354898767,
+      "grad_norm": 2.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.1959,
+      "step": 13810
+    },
+    {
+      "epoch": 0.8040493367465674,
+      "grad_norm": 1.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1922,
+      "step": 13820
+    },
+    {
+      "epoch": 0.8046311380032581,
+      "grad_norm": 1.8515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1799,
+      "step": 13830
+    },
+    {
+      "epoch": 0.8052129392599489,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.196,
+      "step": 13840
+    },
+    {
+      "epoch": 0.8057947405166395,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1941,
+      "step": 13850
+    },
+    {
+      "epoch": 0.8063765417733302,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1913,
+      "step": 13860
+    },
+    {
+      "epoch": 0.8069583430300209,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1911,
+      "step": 13870
+    },
+    {
+      "epoch": 0.8075401442867116,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1894,
+      "step": 13880
+    },
+    {
+      "epoch": 0.8081219455434023,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.1824,
+      "step": 13890
+    },
+    {
+      "epoch": 0.808703746800093,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1895,
+      "step": 13900
+    },
+    {
+      "epoch": 0.8092855480567838,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.1869,
+      "step": 13910
+    },
+    {
+      "epoch": 0.8098673493134745,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1888,
+      "step": 13920
+    },
+    {
+      "epoch": 0.8104491505701652,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.1905,
+      "step": 13930
+    },
+    {
+      "epoch": 0.8110309518268559,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 13940
+    },
+    {
+      "epoch": 0.8116127530835466,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.1901,
+      "step": 13950
+    },
+    {
+      "epoch": 0.8121945543402374,
+      "grad_norm": 1.96875,
+      "learning_rate": 2e-05,
+      "loss": 0.1904,
+      "step": 13960
+    },
+    {
+      "epoch": 0.8127763555969281,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1936,
+      "step": 13970
+    },
+    {
+      "epoch": 0.8133581568536188,
+      "grad_norm": 1.2109375,
+      "learning_rate": 2e-05,
+      "loss": 0.1947,
+      "step": 13980
+    },
+    {
+      "epoch": 0.8139399581103095,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 13990
+    },
+    {
+      "epoch": 0.8145217593670002,
+      "grad_norm": 1.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1941,
+      "step": 14000
+    },
+    {
+      "epoch": 0.8151035606236909,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1898,
+      "step": 14010
+    },
+    {
+      "epoch": 0.8156853618803817,
+      "grad_norm": 1.421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 14020
+    },
+    {
+      "epoch": 0.8162671631370724,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1872,
+      "step": 14030
+    },
+    {
+      "epoch": 0.8168489643937631,
+      "grad_norm": 1.28125,
+      "learning_rate": 2e-05,
+      "loss": 0.1846,
+      "step": 14040
+    },
+    {
+      "epoch": 0.8174307656504538,
+      "grad_norm": 1.578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1928,
+      "step": 14050
+    },
+    {
+      "epoch": 0.8180125669071445,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1904,
+      "step": 14060
+    },
+    {
+      "epoch": 0.8185943681638352,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.1855,
+      "step": 14070
+    },
+    {
+      "epoch": 0.819176169420526,
+      "grad_norm": 1.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1879,
+      "step": 14080
+    },
+    {
+      "epoch": 0.8197579706772167,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1915,
+      "step": 14090
+    },
+    {
+      "epoch": 0.8203397719339074,
+      "grad_norm": 1.8671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1897,
+      "step": 14100
+    },
+    {
+      "epoch": 0.8209215731905981,
+      "grad_norm": 1.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1784,
+      "step": 14110
+    },
+    {
+      "epoch": 0.8215033744472888,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.1934,
+      "step": 14120
+    },
+    {
+      "epoch": 0.8220851757039795,
+      "grad_norm": 1.2734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 14130
+    },
+    {
+      "epoch": 0.8226669769606703,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1863,
+      "step": 14140
+    },
+    {
+      "epoch": 0.823248778217361,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1911,
+      "step": 14150
+    },
+    {
+      "epoch": 0.8238305794740517,
+      "grad_norm": 1.28125,
+      "learning_rate": 2e-05,
+      "loss": 0.1875,
+      "step": 14160
+    },
+    {
+      "epoch": 0.8244123807307424,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.1837,
+      "step": 14170
+    },
+    {
+      "epoch": 0.8249941819874331,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1912,
+      "step": 14180
+    },
+    {
+      "epoch": 0.8255759832441238,
+      "grad_norm": 1.7109375,
+      "learning_rate": 2e-05,
+      "loss": 0.187,
+      "step": 14190
+    },
+    {
+      "epoch": 0.8261577845008146,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.1863,
+      "step": 14200
+    },
+    {
+      "epoch": 0.8267395857575053,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.2013,
+      "step": 14210
+    },
+    {
+      "epoch": 0.827321387014196,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.1905,
+      "step": 14220
+    },
+    {
+      "epoch": 0.8279031882708867,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.1896,
+      "step": 14230
+    },
+    {
+      "epoch": 0.8284849895275774,
+      "grad_norm": 1.765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1895,
+      "step": 14240
+    },
+    {
+      "epoch": 0.8290667907842681,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1905,
+      "step": 14250
+    },
+    {
+      "epoch": 0.8296485920409588,
+      "grad_norm": 1.75,
+      "learning_rate": 2e-05,
+      "loss": 0.1919,
+      "step": 14260
+    },
+    {
+      "epoch": 0.8302303932976495,
+      "grad_norm": 1.265625,
+      "learning_rate": 2e-05,
+      "loss": 0.1869,
+      "step": 14270
+    },
+    {
+      "epoch": 0.8308121945543402,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1921,
+      "step": 14280
+    },
+    {
+      "epoch": 0.8313939958110309,
+      "grad_norm": 1.6875,
+      "learning_rate": 2e-05,
+      "loss": 0.1859,
+      "step": 14290
+    },
+    {
+      "epoch": 0.8319757970677216,
+      "grad_norm": 1.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1941,
+      "step": 14300
+    },
+    {
+      "epoch": 0.8325575983244123,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1849,
+      "step": 14310
+    },
+    {
+      "epoch": 0.8331393995811031,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.1966,
+      "step": 14320
+    },
+    {
+      "epoch": 0.8337212008377938,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1867,
+      "step": 14330
+    },
+    {
+      "epoch": 0.8343030020944845,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1891,
+      "step": 14340
+    },
+    {
+      "epoch": 0.8348848033511752,
+      "grad_norm": 3.15625,
+      "learning_rate": 2e-05,
+      "loss": 0.1881,
+      "step": 14350
+    },
+    {
+      "epoch": 0.8354666046078659,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1873,
+      "step": 14360
+    },
+    {
+      "epoch": 0.8360484058645566,
+      "grad_norm": 1.59375,
+      "learning_rate": 2e-05,
+      "loss": 0.1908,
+      "step": 14370
+    },
+    {
+      "epoch": 0.8366302071212474,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2e-05,
+      "loss": 0.194,
+      "step": 14380
+    },
+    {
+      "epoch": 0.8372120083779381,
+      "grad_norm": 1.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1854,
+      "step": 14390
+    },
+    {
+      "epoch": 0.8377938096346288,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.1855,
+      "step": 14400
+    },
+    {
+      "epoch": 0.8383756108913195,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 14410
+    },
+    {
+      "epoch": 0.8389574121480102,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.1874,
+      "step": 14420
+    },
+    {
+      "epoch": 0.839539213404701,
+      "grad_norm": 2.390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1883,
+      "step": 14430
+    },
+    {
+      "epoch": 0.8401210146613917,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1828,
+      "step": 14440
+    },
+    {
+      "epoch": 0.8407028159180824,
+      "grad_norm": 1.4921875,
+      "learning_rate": 2e-05,
+      "loss": 0.1864,
+      "step": 14450
+    },
+    {
+      "epoch": 0.8412846171747731,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.1918,
+      "step": 14460
+    },
+    {
+      "epoch": 0.8418664184314638,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1891,
+      "step": 14470
+    },
+    {
+      "epoch": 0.8424482196881545,
+      "grad_norm": 1.3046875,
+      "learning_rate": 2e-05,
+      "loss": 0.1836,
+      "step": 14480
+    },
+    {
+      "epoch": 0.8430300209448452,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.1869,
+      "step": 14490
+    },
+    {
+      "epoch": 0.843611822201536,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1906,
+      "step": 14500
+    },
+    {
+      "epoch": 0.8441936234582267,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1908,
+      "step": 14510
+    },
+    {
+      "epoch": 0.8447754247149174,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1932,
+      "step": 14520
+    },
+    {
+      "epoch": 0.8453572259716081,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.1902,
+      "step": 14530
+    },
+    {
+      "epoch": 0.8459390272282988,
+      "grad_norm": 1.546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1856,
+      "step": 14540
+    },
+    {
+      "epoch": 0.8465208284849896,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1905,
+      "step": 14550
+    },
+    {
+      "epoch": 0.8471026297416803,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.1897,
+      "step": 14560
+    },
+    {
+      "epoch": 0.847684430998371,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1955,
+      "step": 14570
+    },
+    {
+      "epoch": 0.8482662322550617,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1886,
+      "step": 14580
+    },
+    {
+      "epoch": 0.8488480335117524,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2e-05,
+      "loss": 0.186,
+      "step": 14590
+    },
+    {
+      "epoch": 0.8494298347684431,
+      "grad_norm": 1.4375,
+      "learning_rate": 2e-05,
+      "loss": 0.1864,
+      "step": 14600
+    },
+    {
+      "epoch": 0.8500116360251339,
+      "grad_norm": 1.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.1841,
+      "step": 14610
+    },
+    {
+      "epoch": 0.8505934372818246,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1906,
+      "step": 14620
+    },
+    {
+      "epoch": 0.8511752385385153,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.19,
+      "step": 14630
+    },
+    {
+      "epoch": 0.851757039795206,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2e-05,
+      "loss": 0.1846,
+      "step": 14640
+    },
+    {
+      "epoch": 0.8523388410518967,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1845,
+      "step": 14650
+    },
+    {
+      "epoch": 0.8529206423085874,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1926,
+      "step": 14660
+    },
+    {
+      "epoch": 0.853502443565278,
+      "grad_norm": 1.6796875,
+      "learning_rate": 2e-05,
+      "loss": 0.1855,
+      "step": 14670
+    },
+    {
+      "epoch": 0.8540842448219688,
+      "grad_norm": 1.375,
+      "learning_rate": 2e-05,
+      "loss": 0.187,
+      "step": 14680
+    },
+    {
+      "epoch": 0.8546660460786595,
+      "grad_norm": 1.5390625,
+      "learning_rate": 2e-05,
+      "loss": 0.1936,
+      "step": 14690
+    },
+    {
+      "epoch": 0.8552478473353502,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1933,
+      "step": 14700
+    },
+    {
+      "epoch": 0.8558296485920409,
+      "grad_norm": 1.2890625,
+      "learning_rate": 2e-05,
+      "loss": 0.1855,
+      "step": 14710
+    },
+    {
+      "epoch": 0.8564114498487316,
+      "grad_norm": 1.40625,
+      "learning_rate": 2e-05,
+      "loss": 0.1878,
+      "step": 14720
+    },
+    {
+      "epoch": 0.8569932511054223,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1958,
+      "step": 14730
+    },
+    {
+      "epoch": 0.8575750523621131,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1857,
+      "step": 14740
+    },
+    {
+      "epoch": 0.8581568536188038,
+      "grad_norm": 1.9453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1895,
+      "step": 14750
+    },
+    {
+      "epoch": 0.8587386548754945,
+      "grad_norm": 1.5234375,
+      "learning_rate": 2e-05,
+      "loss": 0.193,
+      "step": 14760
+    },
+    {
+      "epoch": 0.8593204561321852,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1879,
+      "step": 14770
+    },
+    {
+      "epoch": 0.8599022573888759,
+      "grad_norm": 1.21875,
+      "learning_rate": 2e-05,
+      "loss": 0.188,
+      "step": 14780
+    },
+    {
+      "epoch": 0.8604840586455667,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.1912,
+      "step": 14790
+    },
+    {
+      "epoch": 0.8610658599022574,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1904,
+      "step": 14800
+    },
+    {
+      "epoch": 0.8616476611589481,
+      "grad_norm": 1.453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1896,
+      "step": 14810
+    },
+    {
+      "epoch": 0.8622294624156388,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1844,
+      "step": 14820
+    },
+    {
+      "epoch": 0.8628112636723295,
+      "grad_norm": 1.3203125,
+      "learning_rate": 2e-05,
+      "loss": 0.1924,
+      "step": 14830
+    },
+    {
+      "epoch": 0.8633930649290202,
+      "grad_norm": 1.734375,
+      "learning_rate": 2e-05,
+      "loss": 0.1774,
+      "step": 14840
+    },
+    {
+      "epoch": 0.863974866185711,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1862,
+      "step": 14850
+    },
+    {
+      "epoch": 0.8645566674424017,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1914,
+      "step": 14860
+    },
+    {
+      "epoch": 0.8651384686990924,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1817,
+      "step": 14870
+    },
+    {
+      "epoch": 0.8657202699557831,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1876,
+      "step": 14880
+    },
+    {
+      "epoch": 0.8663020712124738,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.1912,
+      "step": 14890
+    },
+    {
+      "epoch": 0.8668838724691645,
+      "grad_norm": 1.4296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1924,
+      "step": 14900
+    },
+    {
+      "epoch": 0.8674656737258553,
+      "grad_norm": 1.84375,
+      "learning_rate": 2e-05,
+      "loss": 0.1894,
+      "step": 14910
+    },
+    {
+      "epoch": 0.868047474982546,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1847,
+      "step": 14920
+    },
+    {
+      "epoch": 0.8686292762392367,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1827,
+      "step": 14930
+    },
+    {
+      "epoch": 0.8692110774959274,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1867,
+      "step": 14940
+    },
+    {
+      "epoch": 0.8697928787526181,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.1858,
+      "step": 14950
+    },
+    {
+      "epoch": 0.8703746800093088,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2e-05,
+      "loss": 0.1802,
+      "step": 14960
+    },
+    {
+      "epoch": 0.8709564812659996,
+      "grad_norm": 1.8828125,
+      "learning_rate": 2e-05,
+      "loss": 0.181,
+      "step": 14970
+    },
+    {
+      "epoch": 0.8715382825226903,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1909,
+      "step": 14980
+    },
+    {
+      "epoch": 0.872120083779381,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1865,
+      "step": 14990
+    },
+    {
+      "epoch": 0.8727018850360717,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1878,
+      "step": 15000
+    },
+    {
+      "epoch": 0.8732836862927624,
+      "grad_norm": 1.359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1841,
+      "step": 15010
+    },
+    {
+      "epoch": 0.8738654875494531,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1893,
+      "step": 15020
+    },
+    {
+      "epoch": 0.8744472888061439,
+      "grad_norm": 1.2265625,
+      "learning_rate": 2e-05,
+      "loss": 0.1857,
+      "step": 15030
+    },
+    {
+      "epoch": 0.8750290900628346,
+      "grad_norm": 1.671875,
+      "learning_rate": 2e-05,
+      "loss": 0.189,
+      "step": 15040
+    },
+    {
+      "epoch": 0.8756108913195253,
+      "grad_norm": 1.5078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1863,
+      "step": 15050
+    },
+    {
+      "epoch": 0.876192692576216,
+      "grad_norm": 1.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1923,
+      "step": 15060
+    },
+    {
+      "epoch": 0.8767744938329067,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.1885,
+      "step": 15070
+    },
+    {
+      "epoch": 0.8773562950895974,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.1851,
+      "step": 15080
+    },
+    {
+      "epoch": 0.877938096346288,
+      "grad_norm": 1.5,
+      "learning_rate": 2e-05,
+      "loss": 0.1876,
+      "step": 15090
+    },
+    {
+      "epoch": 0.8785198976029788,
+      "grad_norm": 1.625,
+      "learning_rate": 2e-05,
+      "loss": 0.1883,
+      "step": 15100
+    },
+    {
+      "epoch": 0.8791016988596695,
+      "grad_norm": 1.6640625,
+      "learning_rate": 2e-05,
+      "loss": 0.185,
+      "step": 15110
+    },
+    {
+      "epoch": 0.8796835001163602,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.1854,
+      "step": 15120
+    },
+    {
+      "epoch": 0.8802653013730509,
+      "grad_norm": 1.796875,
+      "learning_rate": 2e-05,
+      "loss": 0.1848,
+      "step": 15130
+    },
+    {
+      "epoch": 0.8808471026297416,
+      "grad_norm": 2.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1893,
+      "step": 15140
+    },
+    {
+      "epoch": 0.8814289038864324,
+      "grad_norm": 1.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1881,
+      "step": 15150
+    },
+    {
+      "epoch": 0.8820107051431231,
+      "grad_norm": 1.2578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1857,
+      "step": 15160
+    },
+    {
+      "epoch": 0.8825925063998138,
+      "grad_norm": 1.234375,
+      "learning_rate": 2e-05,
+      "loss": 0.191,
+      "step": 15170
+    },
+    {
+      "epoch": 0.8831743076565045,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.1885,
+      "step": 15180
+    },
+    {
+      "epoch": 0.8837561089131952,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1863,
+      "step": 15190
+    },
+    {
+      "epoch": 0.8843379101698859,
+      "grad_norm": 1.6328125,
+      "learning_rate": 2e-05,
+      "loss": 0.196,
+      "step": 15200
+    },
+    {
+      "epoch": 0.8849197114265767,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2e-05,
+      "loss": 0.1847,
+      "step": 15210
+    },
+    {
+      "epoch": 0.8855015126832674,
+      "grad_norm": 1.296875,
+      "learning_rate": 2e-05,
+      "loss": 0.1823,
+      "step": 15220
+    },
+    {
+      "epoch": 0.8860833139399581,
+      "grad_norm": 1.7265625,
+      "learning_rate": 2e-05,
+      "loss": 0.1871,
+      "step": 15230
+    },
+    {
+      "epoch": 0.8866651151966488,
+      "grad_norm": 1.3125,
+      "learning_rate": 2e-05,
+      "loss": 0.1875,
+      "step": 15240
+    },
+    {
+      "epoch": 0.8872469164533395,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.1899,
+      "step": 15250
+    },
+    {
+      "epoch": 0.8878287177100302,
+      "grad_norm": 1.4765625,
+      "learning_rate": 2e-05,
+      "loss": 0.1801,
+      "step": 15260
+    },
+    {
+      "epoch": 0.888410518966721,
+      "grad_norm": 1.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1883,
+      "step": 15270
+    },
+    {
+      "epoch": 0.8889923202234117,
+      "grad_norm": 1.484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1833,
+      "step": 15280
+    },
+    {
+      "epoch": 0.8895741214801024,
+      "grad_norm": 1.703125,
+      "learning_rate": 2e-05,
+      "loss": 0.193,
+      "step": 15290
+    },
+    {
+      "epoch": 0.8901559227367931,
+      "grad_norm": 1.859375,
+      "learning_rate": 2e-05,
+      "loss": 0.177,
+      "step": 15300
+    },
+    {
+      "epoch": 0.8907377239934838,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.1885,
+      "step": 15310
+    },
+    {
+      "epoch": 0.8913195252501745,
+      "grad_norm": 1.6484375,
+      "learning_rate": 2e-05,
+      "loss": 0.1882,
+      "step": 15320
+    },
+    {
+      "epoch": 0.8919013265068653,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.1845,
+      "step": 15330
+    },
+    {
+      "epoch": 0.892483127763556,
+      "grad_norm": 1.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1844,
+      "step": 15340
+    },
+    {
+      "epoch": 0.8930649290202467,
+      "grad_norm": 2.109375,
+      "learning_rate": 2e-05,
+      "loss": 0.1823,
+      "step": 15350
+    },
+    {
+      "epoch": 0.8936467302769374,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.1835,
+      "step": 15360
+    },
+    {
+      "epoch": 0.8942285315336281,
+      "grad_norm": 1.609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1876,
+      "step": 15370
+    },
+    {
+      "epoch": 0.8948103327903189,
+      "grad_norm": 1.53125,
+      "learning_rate": 2e-05,
+      "loss": 0.1857,
+      "step": 15380
+    },
+    {
+      "epoch": 0.8953921340470096,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1879,
+      "step": 15390
+    },
+    {
+      "epoch": 0.8959739353037003,
+      "grad_norm": 1.890625,
+      "learning_rate": 2e-05,
+      "loss": 0.1862,
+      "step": 15400
+    },
+    {
+      "epoch": 0.896555736560391,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.185,
+      "step": 15410
+    },
+    {
+      "epoch": 0.8971375378170817,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.1875,
+      "step": 15420
+    },
+    {
+      "epoch": 0.8977193390737724,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2e-05,
+      "loss": 0.1819,
+      "step": 15430
+    },
+    {
+      "epoch": 0.8983011403304632,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2e-05,
+      "loss": 0.1877,
+      "step": 15440
+    },
+    {
+      "epoch": 0.8988829415871539,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2e-05,
+      "loss": 0.187,
+      "step": 15450
+    },
+    {
+      "epoch": 0.8994647428438446,
+      "grad_norm": 1.5859375,
+      "learning_rate": 2e-05,
+      "loss": 0.1893,
+      "step": 15460
+    },
+    {
+      "epoch": 0.9000465441005353,
+      "grad_norm": 1.46875,
+      "learning_rate": 2e-05,
+      "loss": 0.192,
+      "step": 15470
+    },
+    {
+      "epoch": 0.900628345357226,
+      "grad_norm": 1.5703125,
+      "learning_rate": 2e-05,
+      "loss": 0.1831,
+      "step": 15480
+    },
+    {
+      "epoch": 0.9012101466139167,
+      "grad_norm": 1.7421875,
+      "learning_rate": 2e-05,
+      "loss": 0.1891,
+      "step": 15490
+    },
+    {
+      "epoch": 0.9017919478706073,
+      "grad_norm": 1.328125,
+      "learning_rate": 2e-05,
+      "loss": 0.1916,
+      "step": 15500
+    },
+    {
+      "epoch": 0.9023737491272981,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2e-05,
+      "loss": 0.1886,
+      "step": 15510
+    },
+    {
+      "epoch": 0.9029555503839888,
+      "grad_norm": 1.4453125,
+      "learning_rate": 2e-05,
+      "loss": 0.1851,
+      "step": 15520
+    },
+    {
+      "epoch": 0.9035373516406795,
+      "grad_norm": 2.078125,
+      "learning_rate": 2e-05,
+      "loss": 0.1871,
+      "step": 15530
+    },
+    {
+      "epoch": 0.9041191528973702,
+      "grad_norm": 1.7578125,
+      "learning_rate": 2e-05,
+      "loss": 0.1876,
+      "step": 15540
+    },
+    {
+      "epoch": 0.9047009541540609,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2e-05,
+      "loss": 0.1896,
+      "step": 15550
+    },
+    {
+      "epoch": 0.9052827554107516,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.185,
+      "step": 15560
+    },
+    {
+      "epoch": 0.9058645566674424,
+      "grad_norm": 2.375,
+      "learning_rate": 2e-05,
+      "loss": 0.1844,
+      "step": 15570
+    },
+    {
+      "epoch": 0.9064463579241331,
+      "grad_norm": 1.515625,
+      "learning_rate": 2e-05,
+      "loss": 0.1853,
+      "step": 15580
+    },
+    {
+      "epoch": 0.9070281591808238,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2e-05,
+      "loss": 0.1806,
+      "step": 15590
+    },
+    {
+      "epoch": 0.9076099604375145,
+      "grad_norm": 1.5625,
+      "learning_rate": 2e-05,
+      "loss": 0.1859,
+      "step": 15600
+    },
+    {
+      "epoch": 0.9081917616942052,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2e-05,
+      "loss": 0.1796,
+      "step": 15610
+    },
+    {
+      "epoch": 0.908773562950896,
+      "grad_norm": 1.71875,
+      "learning_rate": 2e-05,
+      "loss": 0.1854,
+      "step": 15620
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 15625,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.4853137926124995e+18,
+  "train_batch_size": 64,
+  "trial_name": null,
+  "trial_params": null
+}