diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5633 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.56,
+  "eval_steps": 500,
+  "global_step": 8000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0032,
+      "grad_norm": 4.3523268699646,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 0.7342,
+      "step": 10
+    },
+    {
+      "epoch": 0.0064,
+      "grad_norm": 4.409250259399414,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 0.7042,
+      "step": 20
+    },
+    {
+      "epoch": 0.0096,
+      "grad_norm": 6.984313488006592,
+      "learning_rate": 3e-06,
+      "loss": 0.6975,
+      "step": 30
+    },
+    {
+      "epoch": 0.0128,
+      "grad_norm": 4.685063362121582,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.6833,
+      "step": 40
+    },
+    {
+      "epoch": 0.016,
+      "grad_norm": 8.828569412231445,
+      "learning_rate": 5e-06,
+      "loss": 0.7077,
+      "step": 50
+    },
+    {
+      "epoch": 0.0192,
+      "grad_norm": 5.112845420837402,
+      "learning_rate": 6e-06,
+      "loss": 0.66,
+      "step": 60
+    },
+    {
+      "epoch": 0.0224,
+      "grad_norm": 6.451657772064209,
+      "learning_rate": 7.000000000000001e-06,
+      "loss": 0.6894,
+      "step": 70
+    },
+    {
+      "epoch": 0.0256,
+      "grad_norm": 8.004484176635742,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.6722,
+      "step": 80
+    },
+    {
+      "epoch": 0.0288,
+      "grad_norm": 4.263449668884277,
+      "learning_rate": 9e-06,
+      "loss": 0.6639,
+      "step": 90
+    },
+    {
+      "epoch": 0.032,
+      "grad_norm": 6.908220291137695,
+      "learning_rate": 1e-05,
+      "loss": 0.6664,
+      "step": 100
+    },
+    {
+      "epoch": 0.0352,
+      "grad_norm": 9.965716361999512,
+      "learning_rate": 1.1000000000000001e-05,
+      "loss": 0.6425,
+      "step": 110
+    },
+    {
+      "epoch": 0.0384,
+      "grad_norm": 9.22775650024414,
+      "learning_rate": 1.2e-05,
+      "loss": 0.5719,
+      "step": 120
+    },
+    {
+      "epoch": 0.0416,
+      "grad_norm": 8.442060470581055,
+      "learning_rate": 1.3000000000000001e-05,
+      "loss": 0.5148,
+      "step": 130
+    },
+    {
+      "epoch": 0.0448,
+      "grad_norm": 3.89926815032959,
+      "learning_rate": 1.4000000000000001e-05,
+      "loss": 0.4995,
+      "step": 140
+    },
+    {
+      "epoch": 0.048,
+      "grad_norm": 10.715374946594238,
+      "learning_rate": 1.5e-05,
+      "loss": 0.445,
+      "step": 150
+    },
+    {
+      "epoch": 0.0512,
+      "grad_norm": 9.363809585571289,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 0.5027,
+      "step": 160
+    },
+    {
+      "epoch": 0.0544,
+      "grad_norm": 12.60737419128418,
+      "learning_rate": 1.7000000000000003e-05,
+      "loss": 0.3606,
+      "step": 170
+    },
+    {
+      "epoch": 0.0576,
+      "grad_norm": 6.228287696838379,
+      "learning_rate": 1.8e-05,
+      "loss": 0.4815,
+      "step": 180
+    },
+    {
+      "epoch": 0.0608,
+      "grad_norm": 21.15777015686035,
+      "learning_rate": 1.9e-05,
+      "loss": 0.3848,
+      "step": 190
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 7.5884809494018555,
+      "learning_rate": 2e-05,
+      "loss": 0.4474,
+      "step": 200
+    },
+    {
+      "epoch": 0.0672,
+      "grad_norm": 14.817816734313965,
+      "learning_rate": 2.1e-05,
+      "loss": 0.3291,
+      "step": 210
+    },
+    {
+      "epoch": 0.0704,
+      "grad_norm": 25.30421257019043,
+      "learning_rate": 2.2000000000000003e-05,
+      "loss": 0.4788,
+      "step": 220
+    },
+    {
+      "epoch": 0.0736,
+      "grad_norm": 5.897189617156982,
+      "learning_rate": 2.3000000000000003e-05,
+      "loss": 0.4708,
+      "step": 230
+    },
+    {
+      "epoch": 0.0768,
+      "grad_norm": 5.656806468963623,
+      "learning_rate": 2.4e-05,
+      "loss": 0.4214,
+      "step": 240
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 39.75941467285156,
+      "learning_rate": 2.5e-05,
+      "loss": 0.5151,
+      "step": 250
+    },
+    {
+      "epoch": 0.0832,
+      "grad_norm": 9.505982398986816,
+      "learning_rate": 2.6000000000000002e-05,
+      "loss": 0.5456,
+      "step": 260
+    },
+    {
+      "epoch": 0.0864,
+      "grad_norm": 9.837905883789062,
+      "learning_rate": 2.7000000000000002e-05,
+      "loss": 0.3981,
+      "step": 270
+    },
+    {
+      "epoch": 0.0896,
+      "grad_norm": 6.425085544586182,
+      "learning_rate": 2.8000000000000003e-05,
+      "loss": 0.3444,
+      "step": 280
+    },
+    {
+      "epoch": 0.0928,
+      "grad_norm": 8.601673126220703,
+      "learning_rate": 2.9e-05,
+      "loss": 0.3116,
+      "step": 290
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 24.99056625366211,
+      "learning_rate": 3e-05,
+      "loss": 0.4125,
+      "step": 300
+    },
+    {
+      "epoch": 0.0992,
+      "grad_norm": 4.368201732635498,
+      "learning_rate": 3.1e-05,
+      "loss": 0.2946,
+      "step": 310
+    },
+    {
+      "epoch": 0.1024,
+      "grad_norm": 7.49916934967041,
+      "learning_rate": 3.2000000000000005e-05,
+      "loss": 0.4568,
+      "step": 320
+    },
+    {
+      "epoch": 0.1056,
+      "grad_norm": 6.2486138343811035,
+      "learning_rate": 3.3e-05,
+      "loss": 0.4596,
+      "step": 330
+    },
+    {
+      "epoch": 0.1088,
+      "grad_norm": 5.9687886238098145,
+      "learning_rate": 3.4000000000000007e-05,
+      "loss": 0.4197,
+      "step": 340
+    },
+    {
+      "epoch": 0.112,
+      "grad_norm": 5.545505046844482,
+      "learning_rate": 3.5e-05,
+      "loss": 0.3072,
+      "step": 350
+    },
+    {
+      "epoch": 0.1152,
+      "grad_norm": 29.903961181640625,
+      "learning_rate": 3.6e-05,
+      "loss": 0.5313,
+      "step": 360
+    },
+    {
+      "epoch": 0.1184,
+      "grad_norm": 7.169201850891113,
+      "learning_rate": 3.7e-05,
+      "loss": 0.4665,
+      "step": 370
+    },
+    {
+      "epoch": 0.1216,
+      "grad_norm": 11.079299926757812,
+      "learning_rate": 3.8e-05,
+      "loss": 0.5497,
+      "step": 380
+    },
+    {
+      "epoch": 0.1248,
+      "grad_norm": 4.827323913574219,
+      "learning_rate": 3.9000000000000006e-05,
+      "loss": 0.4849,
+      "step": 390
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 6.925987720489502,
+      "learning_rate": 4e-05,
+      "loss": 0.4411,
+      "step": 400
+    },
+    {
+      "epoch": 0.1312,
+      "grad_norm": 8.159820556640625,
+      "learning_rate": 4.1e-05,
+      "loss": 0.4872,
+      "step": 410
+    },
+    {
+      "epoch": 0.1344,
+      "grad_norm": 10.454991340637207,
+      "learning_rate": 4.2e-05,
+      "loss": 0.3407,
+      "step": 420
+    },
+    {
+      "epoch": 0.1376,
+      "grad_norm": 7.866086959838867,
+      "learning_rate": 4.3e-05,
+      "loss": 0.5081,
+      "step": 430
+    },
+    {
+      "epoch": 0.1408,
+      "grad_norm": 11.918012619018555,
+      "learning_rate": 4.4000000000000006e-05,
+      "loss": 0.5015,
+      "step": 440
+    },
+    {
+      "epoch": 0.144,
+      "grad_norm": 14.668400764465332,
+      "learning_rate": 4.5e-05,
+      "loss": 0.3979,
+      "step": 450
+    },
+    {
+      "epoch": 0.1472,
+      "grad_norm": 13.602070808410645,
+      "learning_rate": 4.600000000000001e-05,
+      "loss": 0.4356,
+      "step": 460
+    },
+    {
+      "epoch": 0.1504,
+      "grad_norm": 16.491836547851562,
+      "learning_rate": 4.7e-05,
+      "loss": 0.5188,
+      "step": 470
+    },
+    {
+      "epoch": 0.1536,
+      "grad_norm": 7.220741271972656,
+      "learning_rate": 4.8e-05,
+      "loss": 0.3895,
+      "step": 480
+    },
+    {
+      "epoch": 0.1568,
+      "grad_norm": 11.220433235168457,
+      "learning_rate": 4.9e-05,
+      "loss": 0.4704,
+      "step": 490
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 19.75952911376953,
+      "learning_rate": 5e-05,
+      "loss": 0.3499,
+      "step": 500
+    },
+    {
+      "epoch": 0.1632,
+      "grad_norm": 6.53499174118042,
+      "learning_rate": 4.994366197183099e-05,
+      "loss": 0.5274,
+      "step": 510
+    },
+    {
+      "epoch": 0.1664,
+      "grad_norm": 7.6956000328063965,
+      "learning_rate": 4.9887323943661973e-05,
+      "loss": 0.3979,
+      "step": 520
+    },
+    {
+      "epoch": 0.1696,
+      "grad_norm": 21.266582489013672,
+      "learning_rate": 4.983098591549296e-05,
+      "loss": 0.3423,
+      "step": 530
+    },
+    {
+      "epoch": 0.1728,
+      "grad_norm": 1.1490899324417114,
+      "learning_rate": 4.9774647887323944e-05,
+      "loss": 0.3753,
+      "step": 540
+    },
+    {
+      "epoch": 0.176,
+      "grad_norm": 10.279012680053711,
+      "learning_rate": 4.971830985915493e-05,
+      "loss": 0.5932,
+      "step": 550
+    },
+    {
+      "epoch": 0.1792,
+      "grad_norm": 6.127996444702148,
+      "learning_rate": 4.966197183098592e-05,
+      "loss": 0.608,
+      "step": 560
+    },
+    {
+      "epoch": 0.1824,
+      "grad_norm": 6.2316718101501465,
+      "learning_rate": 4.96056338028169e-05,
+      "loss": 0.4464,
+      "step": 570
+    },
+    {
+      "epoch": 0.1856,
+      "grad_norm": 2.966583251953125,
+      "learning_rate": 4.954929577464789e-05,
+      "loss": 0.4239,
+      "step": 580
+    },
+    {
+      "epoch": 0.1888,
+      "grad_norm": 13.743029594421387,
+      "learning_rate": 4.949295774647887e-05,
+      "loss": 0.5533,
+      "step": 590
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 18.420978546142578,
+      "learning_rate": 4.9436619718309864e-05,
+      "loss": 0.4474,
+      "step": 600
+    },
+    {
+      "epoch": 0.1952,
+      "grad_norm": 7.505041122436523,
+      "learning_rate": 4.938028169014084e-05,
+      "loss": 0.443,
+      "step": 610
+    },
+    {
+      "epoch": 0.1984,
+      "grad_norm": 12.293984413146973,
+      "learning_rate": 4.9323943661971835e-05,
+      "loss": 0.4537,
+      "step": 620
+    },
+    {
+      "epoch": 0.2016,
+      "grad_norm": 4.876405239105225,
+      "learning_rate": 4.926760563380282e-05,
+      "loss": 0.5191,
+      "step": 630
+    },
+    {
+      "epoch": 0.2048,
+      "grad_norm": 8.690363883972168,
+      "learning_rate": 4.9211267605633806e-05,
+      "loss": 0.556,
+      "step": 640
+    },
+    {
+      "epoch": 0.208,
+      "grad_norm": 15.03184700012207,
+      "learning_rate": 4.915492957746479e-05,
+      "loss": 0.3694,
+      "step": 650
+    },
+    {
+      "epoch": 0.2112,
+      "grad_norm": 5.66799259185791,
+      "learning_rate": 4.909859154929578e-05,
+      "loss": 0.5219,
+      "step": 660
+    },
+    {
+      "epoch": 0.2144,
+      "grad_norm": 12.765690803527832,
+      "learning_rate": 4.904225352112676e-05,
+      "loss": 0.4007,
+      "step": 670
+    },
+    {
+      "epoch": 0.2176,
+      "grad_norm": 3.547962188720703,
+      "learning_rate": 4.898591549295775e-05,
+      "loss": 0.2905,
+      "step": 680
+    },
+    {
+      "epoch": 0.2208,
+      "grad_norm": 3.7791709899902344,
+      "learning_rate": 4.8929577464788734e-05,
+      "loss": 0.4592,
+      "step": 690
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 18.622026443481445,
+      "learning_rate": 4.887323943661972e-05,
+      "loss": 0.4226,
+      "step": 700
+    },
+    {
+      "epoch": 0.2272,
+      "grad_norm": 7.643071174621582,
+      "learning_rate": 4.8816901408450705e-05,
+      "loss": 0.3244,
+      "step": 710
+    },
+    {
+      "epoch": 0.2304,
+      "grad_norm": 27.778474807739258,
+      "learning_rate": 4.876056338028169e-05,
+      "loss": 0.3863,
+      "step": 720
+    },
+    {
+      "epoch": 0.2336,
+      "grad_norm": 4.5000834465026855,
+      "learning_rate": 4.8704225352112676e-05,
+      "loss": 0.3431,
+      "step": 730
+    },
+    {
+      "epoch": 0.2368,
+      "grad_norm": 44.82728576660156,
+      "learning_rate": 4.864788732394366e-05,
+      "loss": 0.4113,
+      "step": 740
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 29.374372482299805,
+      "learning_rate": 4.8591549295774653e-05,
+      "loss": 0.5123,
+      "step": 750
+    },
+    {
+      "epoch": 0.2432,
+      "grad_norm": 5.750948429107666,
+      "learning_rate": 4.853521126760563e-05,
+      "loss": 0.3296,
+      "step": 760
+    },
+    {
+      "epoch": 0.2464,
+      "grad_norm": 15.520541191101074,
+      "learning_rate": 4.8478873239436624e-05,
+      "loss": 0.3358,
+      "step": 770
+    },
+    {
+      "epoch": 0.2496,
+      "grad_norm": 5.127716541290283,
+      "learning_rate": 4.84225352112676e-05,
+      "loss": 0.3469,
+      "step": 780
+    },
+    {
+      "epoch": 0.2528,
+      "grad_norm": 20.350370407104492,
+      "learning_rate": 4.8366197183098595e-05,
+      "loss": 0.3067,
+      "step": 790
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 11.152381896972656,
+      "learning_rate": 4.830985915492958e-05,
+      "loss": 0.5645,
+      "step": 800
+    },
+    {
+      "epoch": 0.2592,
+      "grad_norm": 19.948450088500977,
+      "learning_rate": 4.8253521126760566e-05,
+      "loss": 0.6015,
+      "step": 810
+    },
+    {
+      "epoch": 0.2624,
+      "grad_norm": 7.729649543762207,
+      "learning_rate": 4.819718309859155e-05,
+      "loss": 0.3755,
+      "step": 820
+    },
+    {
+      "epoch": 0.2656,
+      "grad_norm": 9.633999824523926,
+      "learning_rate": 4.814084507042254e-05,
+      "loss": 0.4016,
+      "step": 830
+    },
+    {
+      "epoch": 0.2688,
+      "grad_norm": 10.421425819396973,
+      "learning_rate": 4.808450704225352e-05,
+      "loss": 0.3536,
+      "step": 840
+    },
+    {
+      "epoch": 0.272,
+      "grad_norm": 7.2195515632629395,
+      "learning_rate": 4.8028169014084515e-05,
+      "loss": 0.4654,
+      "step": 850
+    },
+    {
+      "epoch": 0.2752,
+      "grad_norm": 17.624547958374023,
+      "learning_rate": 4.7971830985915494e-05,
+      "loss": 0.3924,
+      "step": 860
+    },
+    {
+      "epoch": 0.2784,
+      "grad_norm": 12.011979103088379,
+      "learning_rate": 4.791549295774648e-05,
+      "loss": 0.3513,
+      "step": 870
+    },
+    {
+      "epoch": 0.2816,
+      "grad_norm": 5.497714519500732,
+      "learning_rate": 4.7859154929577465e-05,
+      "loss": 0.3268,
+      "step": 880
+    },
+    {
+      "epoch": 0.2848,
+      "grad_norm": 24.21810531616211,
+      "learning_rate": 4.780281690140845e-05,
+      "loss": 0.4526,
+      "step": 890
+    },
+    {
+      "epoch": 0.288,
+      "grad_norm": 20.337770462036133,
+      "learning_rate": 4.7746478873239436e-05,
+      "loss": 0.3238,
+      "step": 900
+    },
+    {
+      "epoch": 0.2912,
+      "grad_norm": 3.2851223945617676,
+      "learning_rate": 4.769014084507042e-05,
+      "loss": 0.3168,
+      "step": 910
+    },
+    {
+      "epoch": 0.2944,
+      "grad_norm": 37.14635467529297,
+      "learning_rate": 4.7633802816901414e-05,
+      "loss": 0.4194,
+      "step": 920
+    },
+    {
+      "epoch": 0.2976,
+      "grad_norm": 16.439712524414062,
+      "learning_rate": 4.757746478873239e-05,
+      "loss": 0.4244,
+      "step": 930
+    },
+    {
+      "epoch": 0.3008,
+      "grad_norm": 20.88243293762207,
+      "learning_rate": 4.7521126760563385e-05,
+      "loss": 0.5273,
+      "step": 940
+    },
+    {
+      "epoch": 0.304,
+      "grad_norm": 7.504245758056641,
+      "learning_rate": 4.7464788732394363e-05,
+      "loss": 0.4809,
+      "step": 950
+    },
+    {
+      "epoch": 0.3072,
+      "grad_norm": 4.482902526855469,
+      "learning_rate": 4.7408450704225356e-05,
+      "loss": 0.3758,
+      "step": 960
+    },
+    {
+      "epoch": 0.3104,
+      "grad_norm": 7.861649513244629,
+      "learning_rate": 4.735211267605634e-05,
+      "loss": 0.2721,
+      "step": 970
+    },
+    {
+      "epoch": 0.3136,
+      "grad_norm": 7.620471000671387,
+      "learning_rate": 4.729577464788733e-05,
+      "loss": 0.4302,
+      "step": 980
+    },
+    {
+      "epoch": 0.3168,
+      "grad_norm": 9.727157592773438,
+      "learning_rate": 4.723943661971831e-05,
+      "loss": 0.3482,
+      "step": 990
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 20.1031494140625,
+      "learning_rate": 4.71830985915493e-05,
+      "loss": 0.3377,
+      "step": 1000
+    },
+    {
+      "epoch": 0.3232,
+      "grad_norm": 5.509233474731445,
+      "learning_rate": 4.712676056338028e-05,
+      "loss": 0.583,
+      "step": 1010
+    },
+    {
+      "epoch": 0.3264,
+      "grad_norm": 7.83276891708374,
+      "learning_rate": 4.707042253521127e-05,
+      "loss": 0.4061,
+      "step": 1020
+    },
+    {
+      "epoch": 0.3296,
+      "grad_norm": 11.982447624206543,
+      "learning_rate": 4.7014084507042254e-05,
+      "loss": 0.3287,
+      "step": 1030
+    },
+    {
+      "epoch": 0.3328,
+      "grad_norm": 5.372861385345459,
+      "learning_rate": 4.6957746478873247e-05,
+      "loss": 0.2898,
+      "step": 1040
+    },
+    {
+      "epoch": 0.336,
+      "grad_norm": 6.437655448913574,
+      "learning_rate": 4.6901408450704225e-05,
+      "loss": 0.3084,
+      "step": 1050
+    },
+    {
+      "epoch": 0.3392,
+      "grad_norm": 14.815783500671387,
+      "learning_rate": 4.684507042253522e-05,
+      "loss": 0.444,
+      "step": 1060
+    },
+    {
+      "epoch": 0.3424,
+      "grad_norm": 5.138498306274414,
+      "learning_rate": 4.6788732394366196e-05,
+      "loss": 0.3657,
+      "step": 1070
+    },
+    {
+      "epoch": 0.3456,
+      "grad_norm": 12.453933715820312,
+      "learning_rate": 4.673239436619719e-05,
+      "loss": 0.6394,
+      "step": 1080
+    },
+    {
+      "epoch": 0.3488,
+      "grad_norm": 12.436538696289062,
+      "learning_rate": 4.6676056338028174e-05,
+      "loss": 0.3662,
+      "step": 1090
+    },
+    {
+      "epoch": 0.352,
+      "grad_norm": 8.125226974487305,
+      "learning_rate": 4.661971830985915e-05,
+      "loss": 0.486,
+      "step": 1100
+    },
+    {
+      "epoch": 0.3552,
+      "grad_norm": 12.688572883605957,
+      "learning_rate": 4.6563380281690145e-05,
+      "loss": 0.5292,
+      "step": 1110
+    },
+    {
+      "epoch": 0.3584,
+      "grad_norm": 8.127309799194336,
+      "learning_rate": 4.6507042253521124e-05,
+      "loss": 0.2529,
+      "step": 1120
+    },
+    {
+      "epoch": 0.3616,
+      "grad_norm": 14.914011001586914,
+      "learning_rate": 4.6450704225352116e-05,
+      "loss": 0.381,
+      "step": 1130
+    },
+    {
+      "epoch": 0.3648,
+      "grad_norm": 15.199287414550781,
+      "learning_rate": 4.63943661971831e-05,
+      "loss": 0.5236,
+      "step": 1140
+    },
+    {
+      "epoch": 0.368,
+      "grad_norm": 6.843156814575195,
+      "learning_rate": 4.633802816901409e-05,
+      "loss": 0.2929,
+      "step": 1150
+    },
+    {
+      "epoch": 0.3712,
+      "grad_norm": 12.875916481018066,
+      "learning_rate": 4.628169014084507e-05,
+      "loss": 0.5681,
+      "step": 1160
+    },
+    {
+      "epoch": 0.3744,
+      "grad_norm": 7.395442485809326,
+      "learning_rate": 4.622535211267606e-05,
+      "loss": 0.4717,
+      "step": 1170
+    },
+    {
+      "epoch": 0.3776,
+      "grad_norm": 5.369324207305908,
+      "learning_rate": 4.6169014084507044e-05,
+      "loss": 0.4466,
+      "step": 1180
+    },
+    {
+      "epoch": 0.3808,
+      "grad_norm": 5.074844837188721,
+      "learning_rate": 4.611267605633803e-05,
+      "loss": 0.4263,
+      "step": 1190
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 15.802391052246094,
+      "learning_rate": 4.6056338028169015e-05,
+      "loss": 0.351,
+      "step": 1200
+    },
+    {
+      "epoch": 0.3872,
+      "grad_norm": 20.11571502685547,
+      "learning_rate": 4.600000000000001e-05,
+      "loss": 0.3114,
+      "step": 1210
+    },
+    {
+      "epoch": 0.3904,
+      "grad_norm": 20.25322723388672,
+      "learning_rate": 4.5943661971830986e-05,
+      "loss": 0.4008,
+      "step": 1220
+    },
+    {
+      "epoch": 0.3936,
+      "grad_norm": 4.877046585083008,
+      "learning_rate": 4.588732394366198e-05,
+      "loss": 0.6073,
+      "step": 1230
+    },
+    {
+      "epoch": 0.3968,
+      "grad_norm": 6.517822742462158,
+      "learning_rate": 4.5830985915492957e-05,
+      "loss": 0.4318,
+      "step": 1240
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.672747611999512,
+      "learning_rate": 4.577464788732395e-05,
+      "loss": 0.414,
+      "step": 1250
+    },
+    {
+      "epoch": 0.4032,
+      "grad_norm": 4.382776260375977,
+      "learning_rate": 4.5718309859154934e-05,
+      "loss": 0.3432,
+      "step": 1260
+    },
+    {
+      "epoch": 0.4064,
+      "grad_norm": 9.080897331237793,
+      "learning_rate": 4.566197183098592e-05,
+      "loss": 0.4862,
+      "step": 1270
+    },
+    {
+      "epoch": 0.4096,
+      "grad_norm": 5.132823944091797,
+      "learning_rate": 4.5605633802816905e-05,
+      "loss": 0.4707,
+      "step": 1280
+    },
+    {
+      "epoch": 0.4128,
+      "grad_norm": 4.521566867828369,
+      "learning_rate": 4.554929577464789e-05,
+      "loss": 0.4951,
+      "step": 1290
+    },
+    {
+      "epoch": 0.416,
+      "grad_norm": 9.381317138671875,
+      "learning_rate": 4.5492957746478876e-05,
+      "loss": 0.3341,
+      "step": 1300
+    },
+    {
+      "epoch": 0.4192,
+      "grad_norm": 10.4902982711792,
+      "learning_rate": 4.543661971830986e-05,
+      "loss": 0.4471,
+      "step": 1310
+    },
+    {
+      "epoch": 0.4224,
+      "grad_norm": 5.194609642028809,
+      "learning_rate": 4.538028169014085e-05,
+      "loss": 0.4406,
+      "step": 1320
+    },
+    {
+      "epoch": 0.4256,
+      "grad_norm": 13.40365982055664,
+      "learning_rate": 4.532394366197183e-05,
+      "loss": 0.3805,
+      "step": 1330
+    },
+    {
+      "epoch": 0.4288,
+      "grad_norm": 15.255276679992676,
+      "learning_rate": 4.526760563380282e-05,
+      "loss": 0.2003,
+      "step": 1340
+    },
+    {
+      "epoch": 0.432,
+      "grad_norm": 13.552937507629395,
+      "learning_rate": 4.5211267605633804e-05,
+      "loss": 0.4048,
+      "step": 1350
+    },
+    {
+      "epoch": 0.4352,
+      "grad_norm": 24.772096633911133,
+      "learning_rate": 4.515492957746479e-05,
+      "loss": 0.586,
+      "step": 1360
+    },
+    {
+      "epoch": 0.4384,
+      "grad_norm": 29.092702865600586,
+      "learning_rate": 4.5098591549295775e-05,
+      "loss": 0.6368,
+      "step": 1370
+    },
+    {
+      "epoch": 0.4416,
+      "grad_norm": 15.915087699890137,
+      "learning_rate": 4.504225352112677e-05,
+      "loss": 0.3818,
+      "step": 1380
+    },
+    {
+      "epoch": 0.4448,
+      "grad_norm": 14.00623607635498,
+      "learning_rate": 4.4985915492957746e-05,
+      "loss": 0.4146,
+      "step": 1390
+    },
+    {
+      "epoch": 0.448,
+      "grad_norm": 9.716373443603516,
+      "learning_rate": 4.492957746478874e-05,
+      "loss": 0.3421,
+      "step": 1400
+    },
+    {
+      "epoch": 0.4512,
+      "grad_norm": 5.982295989990234,
+      "learning_rate": 4.487323943661972e-05,
+      "loss": 0.4263,
+      "step": 1410
+    },
+    {
+      "epoch": 0.4544,
+      "grad_norm": 10.845952987670898,
+      "learning_rate": 4.481690140845071e-05,
+      "loss": 0.2946,
+      "step": 1420
+    },
+    {
+      "epoch": 0.4576,
+      "grad_norm": 17.834733963012695,
+      "learning_rate": 4.4760563380281695e-05,
+      "loss": 0.3593,
+      "step": 1430
+    },
+    {
+      "epoch": 0.4608,
+      "grad_norm": 7.576904296875,
+      "learning_rate": 4.470422535211268e-05,
+      "loss": 0.4492,
+      "step": 1440
+    },
+    {
+      "epoch": 0.464,
+      "grad_norm": 7.559220790863037,
+      "learning_rate": 4.4647887323943666e-05,
+      "loss": 0.4127,
+      "step": 1450
+    },
+    {
+      "epoch": 0.4672,
+      "grad_norm": 4.112594127655029,
+      "learning_rate": 4.459154929577465e-05,
+      "loss": 0.3307,
+      "step": 1460
+    },
+    {
+      "epoch": 0.4704,
+      "grad_norm": 2.598599910736084,
+      "learning_rate": 4.4535211267605637e-05,
+      "loss": 0.2452,
+      "step": 1470
+    },
+    {
+      "epoch": 0.4736,
+      "grad_norm": 5.336888790130615,
+      "learning_rate": 4.447887323943662e-05,
+      "loss": 0.4106,
+      "step": 1480
+    },
+    {
+      "epoch": 0.4768,
+      "grad_norm": 7.816699028015137,
+      "learning_rate": 4.442253521126761e-05,
+      "loss": 0.492,
+      "step": 1490
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 14.75847053527832,
+      "learning_rate": 4.436619718309859e-05,
+      "loss": 0.4283,
+      "step": 1500
+    },
+    {
+      "epoch": 0.4832,
+      "grad_norm": 5.2887959480285645,
+      "learning_rate": 4.430985915492958e-05,
+      "loss": 0.3547,
+      "step": 1510
+    },
+    {
+      "epoch": 0.4864,
+      "grad_norm": 3.928128242492676,
+      "learning_rate": 4.4253521126760564e-05,
+      "loss": 0.3868,
+      "step": 1520
+    },
+    {
+      "epoch": 0.4896,
+      "grad_norm": 6.465476036071777,
+      "learning_rate": 4.419718309859155e-05,
+      "loss": 0.3789,
+      "step": 1530
+    },
+    {
+      "epoch": 0.4928,
+      "grad_norm": 5.5618743896484375,
+      "learning_rate": 4.4140845070422535e-05,
+      "loss": 0.4552,
+      "step": 1540
+    },
+    {
+      "epoch": 0.496,
+      "grad_norm": 6.504174709320068,
+      "learning_rate": 4.408450704225353e-05,
+      "loss": 0.4515,
+      "step": 1550
+    },
+    {
+      "epoch": 0.4992,
+      "grad_norm": 11.997910499572754,
+      "learning_rate": 4.4028169014084506e-05,
+      "loss": 0.2976,
+      "step": 1560
+    },
+    {
+      "epoch": 0.5024,
+      "grad_norm": 5.83452844619751,
+      "learning_rate": 4.39718309859155e-05,
+      "loss": 0.3886,
+      "step": 1570
+    },
+    {
+      "epoch": 0.5056,
+      "grad_norm": 11.406950950622559,
+      "learning_rate": 4.391549295774648e-05,
+      "loss": 0.3743,
+      "step": 1580
+    },
+    {
+      "epoch": 0.5088,
+      "grad_norm": 4.557556629180908,
+      "learning_rate": 4.385915492957747e-05,
+      "loss": 0.3984,
+      "step": 1590
+    },
+    {
+      "epoch": 0.512,
+      "grad_norm": 10.334356307983398,
+      "learning_rate": 4.3802816901408455e-05,
+      "loss": 0.3593,
+      "step": 1600
+    },
+    {
+      "epoch": 0.5152,
+      "grad_norm": 10.674864768981934,
+      "learning_rate": 4.374647887323944e-05,
+      "loss": 0.1878,
+      "step": 1610
+    },
+    {
+      "epoch": 0.5184,
+      "grad_norm": 3.705169916152954,
+      "learning_rate": 4.3690140845070426e-05,
+      "loss": 0.446,
+      "step": 1620
+    },
+    {
+      "epoch": 0.5216,
+      "grad_norm": 5.510721683502197,
+      "learning_rate": 4.363380281690141e-05,
+      "loss": 0.4812,
+      "step": 1630
+    },
+    {
+      "epoch": 0.5248,
+      "grad_norm": 2.3618953227996826,
+      "learning_rate": 4.35774647887324e-05,
+      "loss": 0.4004,
+      "step": 1640
+    },
+    {
+      "epoch": 0.528,
+      "grad_norm": 10.285249710083008,
+      "learning_rate": 4.352112676056338e-05,
+      "loss": 0.3904,
+      "step": 1650
+    },
+    {
+      "epoch": 0.5312,
+      "grad_norm": 15.25522518157959,
+      "learning_rate": 4.346478873239437e-05,
+      "loss": 0.3363,
+      "step": 1660
+    },
+    {
+      "epoch": 0.5344,
+      "grad_norm": 10.684788703918457,
+      "learning_rate": 4.340845070422535e-05,
+      "loss": 0.5174,
+      "step": 1670
+    },
+    {
+      "epoch": 0.5376,
+      "grad_norm": 4.573671340942383,
+      "learning_rate": 4.335211267605634e-05,
+      "loss": 0.2947,
+      "step": 1680
+    },
+    {
+      "epoch": 0.5408,
+      "grad_norm": 13.247304916381836,
+      "learning_rate": 4.3295774647887324e-05,
+      "loss": 0.4169,
+      "step": 1690
+    },
+    {
+      "epoch": 0.544,
+      "grad_norm": 16.0648250579834,
+      "learning_rate": 4.323943661971831e-05,
+      "loss": 0.2454,
+      "step": 1700
+    },
+    {
+      "epoch": 0.5472,
+      "grad_norm": 7.58563232421875,
+      "learning_rate": 4.3183098591549295e-05,
+      "loss": 0.4982,
+      "step": 1710
+    },
+    {
+      "epoch": 0.5504,
+      "grad_norm": 4.593902587890625,
+      "learning_rate": 4.312676056338029e-05,
+      "loss": 0.4422,
+      "step": 1720
+    },
+    {
+      "epoch": 0.5536,
+      "grad_norm": 6.4184370040893555,
+      "learning_rate": 4.3070422535211266e-05,
+      "loss": 0.3147,
+      "step": 1730
+    },
+    {
+      "epoch": 0.5568,
+      "grad_norm": 16.60883140563965,
+      "learning_rate": 4.301408450704226e-05,
+      "loss": 0.3153,
+      "step": 1740
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 2.526179552078247,
+      "learning_rate": 4.295774647887324e-05,
+      "loss": 0.3074,
+      "step": 1750
+    },
+    {
+      "epoch": 0.5632,
+      "grad_norm": 17.307958602905273,
+      "learning_rate": 4.290140845070423e-05,
+      "loss": 0.3178,
+      "step": 1760
+    },
+    {
+      "epoch": 0.5664,
+      "grad_norm": 9.892918586730957,
+      "learning_rate": 4.284507042253521e-05,
+      "loss": 0.2585,
+      "step": 1770
+    },
+    {
+      "epoch": 0.5696,
+      "grad_norm": 11.281522750854492,
+      "learning_rate": 4.27887323943662e-05,
+      "loss": 0.2706,
+      "step": 1780
+    },
+    {
+      "epoch": 0.5728,
+      "grad_norm": 22.379169464111328,
+      "learning_rate": 4.2732394366197186e-05,
+      "loss": 0.4405,
+      "step": 1790
+    },
+    {
+      "epoch": 0.576,
+      "grad_norm": 7.15933084487915,
+      "learning_rate": 4.267605633802817e-05,
+      "loss": 0.3592,
+      "step": 1800
+    },
+    {
+      "epoch": 0.5792,
+      "grad_norm": 6.169369220733643,
+      "learning_rate": 4.261971830985916e-05,
+      "loss": 0.3866,
+      "step": 1810
+    },
+    {
+      "epoch": 0.5824,
+      "grad_norm": 4.099594593048096,
+      "learning_rate": 4.256338028169014e-05,
+      "loss": 0.4223,
+      "step": 1820
+    },
+    {
+      "epoch": 0.5856,
+      "grad_norm": 18.78368377685547,
+      "learning_rate": 4.250704225352113e-05,
+      "loss": 0.376,
+      "step": 1830
+    },
+    {
+      "epoch": 0.5888,
+      "grad_norm": 2.630387306213379,
+      "learning_rate": 4.2450704225352114e-05,
+      "loss": 0.2327,
+      "step": 1840
+    },
+    {
+      "epoch": 0.592,
+      "grad_norm": 17.63787269592285,
+      "learning_rate": 4.23943661971831e-05,
+      "loss": 0.5726,
+      "step": 1850
+    },
+    {
+      "epoch": 0.5952,
+      "grad_norm": 6.631278038024902,
+      "learning_rate": 4.2338028169014085e-05,
+      "loss": 0.3042,
+      "step": 1860
+    },
+    {
+      "epoch": 0.5984,
+      "grad_norm": 10.951118469238281,
+      "learning_rate": 4.228169014084507e-05,
+      "loss": 0.5125,
+      "step": 1870
+    },
+    {
+      "epoch": 0.6016,
+      "grad_norm": 8.784004211425781,
+      "learning_rate": 4.2225352112676056e-05,
+      "loss": 0.4033,
+      "step": 1880
+    },
+    {
+      "epoch": 0.6048,
+      "grad_norm": 4.034893989562988,
+      "learning_rate": 4.216901408450705e-05,
+      "loss": 0.3788,
+      "step": 1890
+    },
+    {
+      "epoch": 0.608,
+      "grad_norm": 4.547167778015137,
+      "learning_rate": 4.211267605633803e-05,
+      "loss": 0.3065,
+      "step": 1900
+    },
+    {
+      "epoch": 0.6112,
+      "grad_norm": 8.01904582977295,
+      "learning_rate": 4.205633802816902e-05,
+      "loss": 0.3656,
+      "step": 1910
+    },
+    {
+      "epoch": 0.6144,
+      "grad_norm": 3.676229953765869,
+      "learning_rate": 4.2e-05,
+      "loss": 0.4528,
+      "step": 1920
+    },
+    {
+      "epoch": 0.6176,
+      "grad_norm": 14.89476203918457,
+      "learning_rate": 4.194366197183099e-05,
+      "loss": 0.3974,
+      "step": 1930
+    },
+    {
+      "epoch": 0.6208,
+      "grad_norm": 6.517081260681152,
+      "learning_rate": 4.188732394366197e-05,
+      "loss": 0.3502,
+      "step": 1940
+    },
+    {
+      "epoch": 0.624,
+      "grad_norm": 9.692541122436523,
+      "learning_rate": 4.183098591549296e-05,
+      "loss": 0.2676,
+      "step": 1950
+    },
+    {
+      "epoch": 0.6272,
+      "grad_norm": 2.047581434249878,
+      "learning_rate": 4.1774647887323946e-05,
+      "loss": 0.3422,
+      "step": 1960
+    },
+    {
+      "epoch": 0.6304,
+      "grad_norm": 0.5576546788215637,
+      "learning_rate": 4.171830985915493e-05,
+      "loss": 0.2076,
+      "step": 1970
+    },
+    {
+      "epoch": 0.6336,
+      "grad_norm": 3.5656802654266357,
+      "learning_rate": 4.166197183098592e-05,
+      "loss": 0.4356,
+      "step": 1980
+    },
+    {
+      "epoch": 0.6368,
+      "grad_norm": 1.7690439224243164,
+      "learning_rate": 4.16056338028169e-05,
+      "loss": 0.2592,
+      "step": 1990
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 21.8055362701416,
+      "learning_rate": 4.154929577464789e-05,
+      "loss": 0.4841,
+      "step": 2000
+    },
+    {
+      "epoch": 0.6432,
+      "grad_norm": 2.1585135459899902,
+      "learning_rate": 4.149295774647888e-05,
+      "loss": 0.2427,
+      "step": 2010
+    },
+    {
+      "epoch": 0.6464,
+      "grad_norm": 22.61993980407715,
+      "learning_rate": 4.143661971830986e-05,
+      "loss": 0.3264,
+      "step": 2020
+    },
+    {
+      "epoch": 0.6496,
+      "grad_norm": 3.826843500137329,
+      "learning_rate": 4.138028169014085e-05,
+      "loss": 0.3896,
+      "step": 2030
+    },
+    {
+      "epoch": 0.6528,
+      "grad_norm": 14.643287658691406,
+      "learning_rate": 4.132394366197183e-05,
+      "loss": 0.5588,
+      "step": 2040
+    },
+    {
+      "epoch": 0.656,
+      "grad_norm": 1.5682073831558228,
+      "learning_rate": 4.126760563380282e-05,
+      "loss": 0.4475,
+      "step": 2050
+    },
+    {
+      "epoch": 0.6592,
+      "grad_norm": 11.66586971282959,
+      "learning_rate": 4.12112676056338e-05,
+      "loss": 0.3633,
+      "step": 2060
+    },
+    {
+      "epoch": 0.6624,
+      "grad_norm": 11.989961624145508,
+      "learning_rate": 4.115492957746479e-05,
+      "loss": 0.4635,
+      "step": 2070
+    },
+    {
+      "epoch": 0.6656,
+      "grad_norm": 3.4878032207489014,
+      "learning_rate": 4.109859154929578e-05,
+      "loss": 0.2872,
+      "step": 2080
+    },
+    {
+      "epoch": 0.6688,
+      "grad_norm": 3.756565570831299,
+      "learning_rate": 4.104225352112676e-05,
+      "loss": 0.4745,
+      "step": 2090
+    },
+    {
+      "epoch": 0.672,
+      "grad_norm": 6.970531940460205,
+      "learning_rate": 4.098591549295775e-05,
+      "loss": 0.3618,
+      "step": 2100
+    },
+    {
+      "epoch": 0.6752,
+      "grad_norm": 16.83983039855957,
+      "learning_rate": 4.092957746478873e-05,
+      "loss": 0.3934,
+      "step": 2110
+    },
+    {
+      "epoch": 0.6784,
+      "grad_norm": 11.064716339111328,
+      "learning_rate": 4.087323943661972e-05,
+      "loss": 0.2683,
+      "step": 2120
+    },
+    {
+      "epoch": 0.6816,
+      "grad_norm": 18.883390426635742,
+      "learning_rate": 4.081690140845071e-05,
+      "loss": 0.3461,
+      "step": 2130
+    },
+    {
+      "epoch": 0.6848,
+      "grad_norm": 4.280035972595215,
+      "learning_rate": 4.076056338028169e-05,
+      "loss": 0.372,
+      "step": 2140
+    },
+    {
+      "epoch": 0.688,
+      "grad_norm": 10.117981910705566,
+      "learning_rate": 4.070422535211268e-05,
+      "loss": 0.2614,
+      "step": 2150
+    },
+    {
+      "epoch": 0.6912,
+      "grad_norm": 11.863015174865723,
+      "learning_rate": 4.064788732394366e-05,
+      "loss": 0.3622,
+      "step": 2160
+    },
+    {
+      "epoch": 0.6944,
+      "grad_norm": 7.1603875160217285,
+      "learning_rate": 4.059154929577465e-05,
+      "loss": 0.278,
+      "step": 2170
+    },
+    {
+      "epoch": 0.6976,
+      "grad_norm": 9.962820053100586,
+      "learning_rate": 4.053521126760564e-05,
+      "loss": 0.544,
+      "step": 2180
+    },
+    {
+      "epoch": 0.7008,
+      "grad_norm": 7.794748306274414,
+      "learning_rate": 4.047887323943662e-05,
+      "loss": 0.3095,
+      "step": 2190
+    },
+    {
+      "epoch": 0.704,
+      "grad_norm": 20.568464279174805,
+      "learning_rate": 4.042253521126761e-05,
+      "loss": 0.3264,
+      "step": 2200
+    },
+    {
+      "epoch": 0.7072,
+      "grad_norm": 4.824507236480713,
+      "learning_rate": 4.036619718309859e-05,
+      "loss": 0.3093,
+      "step": 2210
+    },
+    {
+      "epoch": 0.7104,
+      "grad_norm": 6.159689426422119,
+      "learning_rate": 4.030985915492958e-05,
+      "loss": 0.3073,
+      "step": 2220
+    },
+    {
+      "epoch": 0.7136,
+      "grad_norm": 22.985084533691406,
+      "learning_rate": 4.025352112676056e-05,
+      "loss": 0.2162,
+      "step": 2230
+    },
+    {
+      "epoch": 0.7168,
+      "grad_norm": 19.654817581176758,
+      "learning_rate": 4.0197183098591554e-05,
+      "loss": 0.6091,
+      "step": 2240
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 6.315866947174072,
+      "learning_rate": 4.014084507042254e-05,
+      "loss": 0.459,
+      "step": 2250
+    },
+    {
+      "epoch": 0.7232,
+      "grad_norm": 7.05145788192749,
+      "learning_rate": 4.0084507042253525e-05,
+      "loss": 0.3191,
+      "step": 2260
+    },
+    {
+      "epoch": 0.7264,
+      "grad_norm": 11.295578956604004,
+      "learning_rate": 4.002816901408451e-05,
+      "loss": 0.3457,
+      "step": 2270
+    },
+    {
+      "epoch": 0.7296,
+      "grad_norm": 4.964128494262695,
+      "learning_rate": 3.9971830985915496e-05,
+      "loss": 0.3634,
+      "step": 2280
+    },
+    {
+      "epoch": 0.7328,
+      "grad_norm": 9.028850555419922,
+      "learning_rate": 3.991549295774648e-05,
+      "loss": 0.4049,
+      "step": 2290
+    },
+    {
+      "epoch": 0.736,
+      "grad_norm": 7.955386161804199,
+      "learning_rate": 3.985915492957747e-05,
+      "loss": 0.3013,
+      "step": 2300
+    },
+    {
+      "epoch": 0.7392,
+      "grad_norm": 9.309741020202637,
+      "learning_rate": 3.980281690140845e-05,
+      "loss": 0.3583,
+      "step": 2310
+    },
+    {
+      "epoch": 0.7424,
+      "grad_norm": 13.393871307373047,
+      "learning_rate": 3.974647887323944e-05,
+      "loss": 0.433,
+      "step": 2320
+    },
+    {
+      "epoch": 0.7456,
+      "grad_norm": 13.058290481567383,
+      "learning_rate": 3.9690140845070424e-05,
+      "loss": 0.5419,
+      "step": 2330
+    },
+    {
+      "epoch": 0.7488,
+      "grad_norm": 5.2141900062561035,
+      "learning_rate": 3.963380281690141e-05,
+      "loss": 0.2244,
+      "step": 2340
+    },
+    {
+      "epoch": 0.752,
+      "grad_norm": 10.393515586853027,
+      "learning_rate": 3.9577464788732395e-05,
+      "loss": 0.4972,
+      "step": 2350
+    },
+    {
+      "epoch": 0.7552,
+      "grad_norm": 2.1989641189575195,
+      "learning_rate": 3.952112676056338e-05,
+      "loss": 0.395,
+      "step": 2360
+    },
+    {
+      "epoch": 0.7584,
+      "grad_norm": 10.207283973693848,
+      "learning_rate": 3.946478873239437e-05,
+      "loss": 0.5946,
+      "step": 2370
+    },
+    {
+      "epoch": 0.7616,
+      "grad_norm": 5.437625408172607,
+      "learning_rate": 3.940845070422535e-05,
+      "loss": 0.2024,
+      "step": 2380
+    },
+    {
+      "epoch": 0.7648,
+      "grad_norm": 5.534853458404541,
+      "learning_rate": 3.935211267605634e-05,
+      "loss": 0.3462,
+      "step": 2390
+    },
+    {
+      "epoch": 0.768,
+      "grad_norm": 14.348875045776367,
+      "learning_rate": 3.929577464788732e-05,
+      "loss": 0.2925,
+      "step": 2400
+    },
+    {
+      "epoch": 0.7712,
+      "grad_norm": 1.4454820156097412,
+      "learning_rate": 3.9239436619718314e-05,
+      "loss": 0.3436,
+      "step": 2410
+    },
+    {
+      "epoch": 0.7744,
+      "grad_norm": 11.598908424377441,
+      "learning_rate": 3.91830985915493e-05,
+      "loss": 0.3707,
+      "step": 2420
+    },
+    {
+      "epoch": 0.7776,
+      "grad_norm": 23.476207733154297,
+      "learning_rate": 3.9126760563380285e-05,
+      "loss": 0.5327,
+      "step": 2430
+    },
+    {
+      "epoch": 0.7808,
+      "grad_norm": 11.371722221374512,
+      "learning_rate": 3.907042253521127e-05,
+      "loss": 0.4592,
+      "step": 2440
+    },
+    {
+      "epoch": 0.784,
+      "grad_norm": 7.114121437072754,
+      "learning_rate": 3.9014084507042256e-05,
+      "loss": 0.2222,
+      "step": 2450
+    },
+    {
+      "epoch": 0.7872,
+      "grad_norm": 9.982179641723633,
+      "learning_rate": 3.895774647887324e-05,
+      "loss": 0.4403,
+      "step": 2460
+    },
+    {
+      "epoch": 0.7904,
+      "grad_norm": 4.670555114746094,
+      "learning_rate": 3.890140845070423e-05,
+      "loss": 0.3811,
+      "step": 2470
+    },
+    {
+      "epoch": 0.7936,
+      "grad_norm": 4.608165740966797,
+      "learning_rate": 3.884507042253521e-05,
+      "loss": 0.2737,
+      "step": 2480
+    },
+    {
+      "epoch": 0.7968,
+      "grad_norm": 10.740816116333008,
+      "learning_rate": 3.87887323943662e-05,
+      "loss": 0.3186,
+      "step": 2490
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 21.781532287597656,
+      "learning_rate": 3.8732394366197184e-05,
+      "loss": 0.2801,
+      "step": 2500
+    },
+    {
+      "epoch": 0.8032,
+      "grad_norm": 18.888141632080078,
+      "learning_rate": 3.867605633802817e-05,
+      "loss": 0.2524,
+      "step": 2510
+    },
+    {
+      "epoch": 0.8064,
+      "grad_norm": 14.424897193908691,
+      "learning_rate": 3.8619718309859155e-05,
+      "loss": 0.4049,
+      "step": 2520
+    },
+    {
+      "epoch": 0.8096,
+      "grad_norm": 2.566080093383789,
+      "learning_rate": 3.856338028169014e-05,
+      "loss": 0.542,
+      "step": 2530
+    },
+    {
+      "epoch": 0.8128,
+      "grad_norm": 20.100793838500977,
+      "learning_rate": 3.850704225352113e-05,
+      "loss": 0.331,
+      "step": 2540
+    },
+    {
+      "epoch": 0.816,
+      "grad_norm": 17.21098518371582,
+      "learning_rate": 3.845070422535211e-05,
+      "loss": 0.5059,
+      "step": 2550
+    },
+    {
+      "epoch": 0.8192,
+      "grad_norm": 26.911142349243164,
+      "learning_rate": 3.8394366197183104e-05,
+      "loss": 0.3813,
+      "step": 2560
+    },
+    {
+      "epoch": 0.8224,
+      "grad_norm": 8.0260591506958,
+      "learning_rate": 3.833802816901408e-05,
+      "loss": 0.3696,
+      "step": 2570
+    },
+    {
+      "epoch": 0.8256,
+      "grad_norm": 12.544899940490723,
+      "learning_rate": 3.8281690140845075e-05,
+      "loss": 0.2778,
+      "step": 2580
+    },
+    {
+      "epoch": 0.8288,
+      "grad_norm": 8.61177921295166,
+      "learning_rate": 3.822535211267606e-05,
+      "loss": 0.3647,
+      "step": 2590
+    },
+    {
+      "epoch": 0.832,
+      "grad_norm": 2.9444468021392822,
+      "learning_rate": 3.8169014084507046e-05,
+      "loss": 0.3391,
+      "step": 2600
+    },
+    {
+      "epoch": 0.8352,
+      "grad_norm": 12.346148490905762,
+      "learning_rate": 3.811267605633803e-05,
+      "loss": 0.3134,
+      "step": 2610
+    },
+    {
+      "epoch": 0.8384,
+      "grad_norm": 16.827272415161133,
+      "learning_rate": 3.8056338028169017e-05,
+      "loss": 0.2674,
+      "step": 2620
+    },
+    {
+      "epoch": 0.8416,
+      "grad_norm": 16.586444854736328,
+      "learning_rate": 3.8e-05,
+      "loss": 0.2912,
+      "step": 2630
+    },
+    {
+      "epoch": 0.8448,
+      "grad_norm": 6.491583347320557,
+      "learning_rate": 3.794366197183099e-05,
+      "loss": 0.3261,
+      "step": 2640
+    },
+    {
+      "epoch": 0.848,
+      "grad_norm": 35.861572265625,
+      "learning_rate": 3.788732394366197e-05,
+      "loss": 0.2646,
+      "step": 2650
+    },
+    {
+      "epoch": 0.8512,
+      "grad_norm": 8.829320907592773,
+      "learning_rate": 3.783098591549296e-05,
+      "loss": 0.5009,
+      "step": 2660
+    },
+    {
+      "epoch": 0.8544,
+      "grad_norm": 1.658776879310608,
+      "learning_rate": 3.7774647887323944e-05,
+      "loss": 0.4636,
+      "step": 2670
+    },
+    {
+      "epoch": 0.8576,
+      "grad_norm": 2.2379770278930664,
+      "learning_rate": 3.771830985915493e-05,
+      "loss": 0.3915,
+      "step": 2680
+    },
+    {
+      "epoch": 0.8608,
+      "grad_norm": 4.984516143798828,
+      "learning_rate": 3.7661971830985915e-05,
+      "loss": 0.1507,
+      "step": 2690
+    },
+    {
+      "epoch": 0.864,
+      "grad_norm": 5.562011241912842,
+      "learning_rate": 3.76056338028169e-05,
+      "loss": 0.5398,
+      "step": 2700
+    },
+    {
+      "epoch": 0.8672,
+      "grad_norm": 21.320629119873047,
+      "learning_rate": 3.754929577464789e-05,
+      "loss": 0.4201,
+      "step": 2710
+    },
+    {
+      "epoch": 0.8704,
+      "grad_norm": 19.99195671081543,
+      "learning_rate": 3.749295774647887e-05,
+      "loss": 0.4603,
+      "step": 2720
+    },
+    {
+      "epoch": 0.8736,
+      "grad_norm": 3.42061185836792,
+      "learning_rate": 3.7436619718309864e-05,
+      "loss": 0.2968,
+      "step": 2730
+    },
+    {
+      "epoch": 0.8768,
+      "grad_norm": 27.126548767089844,
+      "learning_rate": 3.738028169014084e-05,
+      "loss": 0.3913,
+      "step": 2740
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 11.521971702575684,
+      "learning_rate": 3.7323943661971835e-05,
+      "loss": 0.2858,
+      "step": 2750
+    },
+    {
+      "epoch": 0.8832,
+      "grad_norm": 20.222986221313477,
+      "learning_rate": 3.726760563380282e-05,
+      "loss": 0.3669,
+      "step": 2760
+    },
+    {
+      "epoch": 0.8864,
+      "grad_norm": 1.7271472215652466,
+      "learning_rate": 3.7211267605633806e-05,
+      "loss": 0.3193,
+      "step": 2770
+    },
+    {
+      "epoch": 0.8896,
+      "grad_norm": 29.57312774658203,
+      "learning_rate": 3.715492957746479e-05,
+      "loss": 0.5223,
+      "step": 2780
+    },
+    {
+      "epoch": 0.8928,
+      "grad_norm": 6.471129894256592,
+      "learning_rate": 3.709859154929578e-05,
+      "loss": 0.3132,
+      "step": 2790
+    },
+    {
+      "epoch": 0.896,
+      "grad_norm": 13.332724571228027,
+      "learning_rate": 3.704225352112676e-05,
+      "loss": 0.5883,
+      "step": 2800
+    },
+    {
+      "epoch": 0.8992,
+      "grad_norm": 16.782337188720703,
+      "learning_rate": 3.698591549295775e-05,
+      "loss": 0.3876,
+      "step": 2810
+    },
+    {
+      "epoch": 0.9024,
+      "grad_norm": 12.669254302978516,
+      "learning_rate": 3.692957746478873e-05,
+      "loss": 0.5085,
+      "step": 2820
+    },
+    {
+      "epoch": 0.9056,
+      "grad_norm": 6.967997074127197,
+      "learning_rate": 3.687323943661972e-05,
+      "loss": 0.3927,
+      "step": 2830
+    },
+    {
+      "epoch": 0.9088,
+      "grad_norm": 11.67349624633789,
+      "learning_rate": 3.6816901408450704e-05,
+      "loss": 0.2962,
+      "step": 2840
+    },
+    {
+      "epoch": 0.912,
+      "grad_norm": 9.875104904174805,
+      "learning_rate": 3.676056338028169e-05,
+      "loss": 0.3898,
+      "step": 2850
+    },
+    {
+      "epoch": 0.9152,
+      "grad_norm": 1.1998872756958008,
+      "learning_rate": 3.6704225352112675e-05,
+      "loss": 0.3745,
+      "step": 2860
+    },
+    {
+      "epoch": 0.9184,
+      "grad_norm": 6.544206619262695,
+      "learning_rate": 3.664788732394366e-05,
+      "loss": 0.2814,
+      "step": 2870
+    },
+    {
+      "epoch": 0.9216,
+      "grad_norm": 27.720895767211914,
+      "learning_rate": 3.659154929577465e-05,
+      "loss": 0.5197,
+      "step": 2880
+    },
+    {
+      "epoch": 0.9248,
+      "grad_norm": 5.2081403732299805,
+      "learning_rate": 3.653521126760563e-05,
+      "loss": 0.2418,
+      "step": 2890
+    },
+    {
+      "epoch": 0.928,
+      "grad_norm": 25.75909996032715,
+      "learning_rate": 3.6478873239436624e-05,
+      "loss": 0.43,
+      "step": 2900
+    },
+    {
+      "epoch": 0.9312,
+      "grad_norm": 11.020965576171875,
+      "learning_rate": 3.64225352112676e-05,
+      "loss": 0.2788,
+      "step": 2910
+    },
+    {
+      "epoch": 0.9344,
+      "grad_norm": 5.504922866821289,
+      "learning_rate": 3.6366197183098595e-05,
+      "loss": 0.2734,
+      "step": 2920
+    },
+    {
+      "epoch": 0.9376,
+      "grad_norm": 10.418407440185547,
+      "learning_rate": 3.630985915492958e-05,
+      "loss": 0.4713,
+      "step": 2930
+    },
+    {
+      "epoch": 0.9408,
+      "grad_norm": 7.805202960968018,
+      "learning_rate": 3.6253521126760566e-05,
+      "loss": 0.2821,
+      "step": 2940
+    },
+    {
+      "epoch": 0.944,
+      "grad_norm": 7.880125045776367,
+      "learning_rate": 3.619718309859155e-05,
+      "loss": 0.1627,
+      "step": 2950
+    },
+    {
+      "epoch": 0.9472,
+      "grad_norm": 33.17704772949219,
+      "learning_rate": 3.614084507042254e-05,
+      "loss": 0.4046,
+      "step": 2960
+    },
+    {
+      "epoch": 0.9504,
+      "grad_norm": 3.1542086601257324,
+      "learning_rate": 3.608450704225352e-05,
+      "loss": 0.3558,
+      "step": 2970
+    },
+    {
+      "epoch": 0.9536,
+      "grad_norm": 21.562021255493164,
+      "learning_rate": 3.602816901408451e-05,
+      "loss": 0.3107,
+      "step": 2980
+    },
+    {
+      "epoch": 0.9568,
+      "grad_norm": 17.724111557006836,
+      "learning_rate": 3.5971830985915494e-05,
+      "loss": 0.2855,
+      "step": 2990
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 16.4515323638916,
+      "learning_rate": 3.5915492957746486e-05,
+      "loss": 0.3282,
+      "step": 3000
+    },
+    {
+      "epoch": 0.9632,
+      "grad_norm": 3.998889684677124,
+      "learning_rate": 3.5859154929577465e-05,
+      "loss": 0.4798,
+      "step": 3010
+    },
+    {
+      "epoch": 0.9664,
+      "grad_norm": 13.89487361907959,
+      "learning_rate": 3.580281690140846e-05,
+      "loss": 0.3655,
+      "step": 3020
+    },
+    {
+      "epoch": 0.9696,
+      "grad_norm": 12.545125007629395,
+      "learning_rate": 3.5746478873239436e-05,
+      "loss": 0.374,
+      "step": 3030
+    },
+    {
+      "epoch": 0.9728,
+      "grad_norm": 16.73505210876465,
+      "learning_rate": 3.569014084507042e-05,
+      "loss": 0.4029,
+      "step": 3040
+    },
+    {
+      "epoch": 0.976,
+      "grad_norm": 3.9983644485473633,
+      "learning_rate": 3.5633802816901413e-05,
+      "loss": 0.1769,
+      "step": 3050
+    },
+    {
+      "epoch": 0.9792,
+      "grad_norm": 19.214399337768555,
+      "learning_rate": 3.557746478873239e-05,
+      "loss": 0.335,
+      "step": 3060
+    },
+    {
+      "epoch": 0.9824,
+      "grad_norm": 27.911151885986328,
+      "learning_rate": 3.5521126760563384e-05,
+      "loss": 0.3406,
+      "step": 3070
+    },
+    {
+      "epoch": 0.9856,
+      "grad_norm": 8.778318405151367,
+      "learning_rate": 3.546478873239436e-05,
+      "loss": 0.4602,
+      "step": 3080
+    },
+    {
+      "epoch": 0.9888,
+      "grad_norm": 5.281238555908203,
+      "learning_rate": 3.5408450704225355e-05,
+      "loss": 0.3499,
+      "step": 3090
+    },
+    {
+      "epoch": 0.992,
+      "grad_norm": 7.649629592895508,
+      "learning_rate": 3.5352112676056334e-05,
+      "loss": 0.2717,
+      "step": 3100
+    },
+    {
+      "epoch": 0.9952,
+      "grad_norm": 8.83627986907959,
+      "learning_rate": 3.5295774647887326e-05,
+      "loss": 0.3544,
+      "step": 3110
+    },
+    {
+      "epoch": 0.9984,
+      "grad_norm": 13.762328147888184,
+      "learning_rate": 3.523943661971831e-05,
+      "loss": 0.3039,
+      "step": 3120
+    },
+    {
+      "epoch": 1.0016,
+      "grad_norm": 28.637189865112305,
+      "learning_rate": 3.51830985915493e-05,
+      "loss": 0.1381,
+      "step": 3130
+    },
+    {
+      "epoch": 1.0048,
+      "grad_norm": 6.5435791015625,
+      "learning_rate": 3.512676056338028e-05,
+      "loss": 0.4339,
+      "step": 3140
+    },
+    {
+      "epoch": 1.008,
+      "grad_norm": 8.024589538574219,
+      "learning_rate": 3.507042253521127e-05,
+      "loss": 0.3516,
+      "step": 3150
+    },
+    {
+      "epoch": 1.0112,
+      "grad_norm": 0.35201606154441833,
+      "learning_rate": 3.5014084507042254e-05,
+      "loss": 0.232,
+      "step": 3160
+    },
+    {
+      "epoch": 1.0144,
+      "grad_norm": 0.8222833275794983,
+      "learning_rate": 3.4957746478873246e-05,
+      "loss": 0.555,
+      "step": 3170
+    },
+    {
+      "epoch": 1.0176,
+      "grad_norm": 0.6108925342559814,
+      "learning_rate": 3.4901408450704225e-05,
+      "loss": 0.4073,
+      "step": 3180
+    },
+    {
+      "epoch": 1.0208,
+      "grad_norm": 33.10356521606445,
+      "learning_rate": 3.484507042253522e-05,
+      "loss": 0.2513,
+      "step": 3190
+    },
+    {
+      "epoch": 1.024,
+      "grad_norm": 0.9735102653503418,
+      "learning_rate": 3.4788732394366196e-05,
+      "loss": 0.2464,
+      "step": 3200
+    },
+    {
+      "epoch": 1.0272,
+      "grad_norm": 28.903047561645508,
+      "learning_rate": 3.473239436619719e-05,
+      "loss": 0.2955,
+      "step": 3210
+    },
+    {
+      "epoch": 1.0304,
+      "grad_norm": 2.8575825691223145,
+      "learning_rate": 3.4676056338028174e-05,
+      "loss": 0.1999,
+      "step": 3220
+    },
+    {
+      "epoch": 1.0336,
+      "grad_norm": 20.94749641418457,
+      "learning_rate": 3.461971830985916e-05,
+      "loss": 0.2652,
+      "step": 3230
+    },
+    {
+      "epoch": 1.0368,
+      "grad_norm": 56.3437385559082,
+      "learning_rate": 3.4563380281690145e-05,
+      "loss": 0.2871,
+      "step": 3240
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.698115885257721,
+      "learning_rate": 3.450704225352113e-05,
+      "loss": 0.1267,
+      "step": 3250
+    },
+    {
+      "epoch": 1.0432,
+      "grad_norm": 0.10409284383058548,
+      "learning_rate": 3.4450704225352116e-05,
+      "loss": 0.3235,
+      "step": 3260
+    },
+    {
+      "epoch": 1.0464,
+      "grad_norm": 22.576404571533203,
+      "learning_rate": 3.4394366197183094e-05,
+      "loss": 0.3016,
+      "step": 3270
+    },
+    {
+      "epoch": 1.0496,
+      "grad_norm": 15.452841758728027,
+      "learning_rate": 3.433802816901409e-05,
+      "loss": 0.43,
+      "step": 3280
+    },
+    {
+      "epoch": 1.0528,
+      "grad_norm": 1.0227371454238892,
+      "learning_rate": 3.428169014084507e-05,
+      "loss": 0.261,
+      "step": 3290
+    },
+    {
+      "epoch": 1.056,
+      "grad_norm": 12.007558822631836,
+      "learning_rate": 3.422535211267606e-05,
+      "loss": 0.1861,
+      "step": 3300
+    },
+    {
+      "epoch": 1.0592,
+      "grad_norm": 4.2839484214782715,
+      "learning_rate": 3.416901408450704e-05,
+      "loss": 0.2487,
+      "step": 3310
+    },
+    {
+      "epoch": 1.0624,
+      "grad_norm": 31.287580490112305,
+      "learning_rate": 3.411267605633803e-05,
+      "loss": 0.2447,
+      "step": 3320
+    },
+    {
+      "epoch": 1.0656,
+      "grad_norm": 4.608018398284912,
+      "learning_rate": 3.4056338028169014e-05,
+      "loss": 0.3264,
+      "step": 3330
+    },
+    {
+      "epoch": 1.0688,
+      "grad_norm": 10.261385917663574,
+      "learning_rate": 3.4000000000000007e-05,
+      "loss": 0.1514,
+      "step": 3340
+    },
+    {
+      "epoch": 1.072,
+      "grad_norm": 28.37779426574707,
+      "learning_rate": 3.3943661971830985e-05,
+      "loss": 0.3787,
+      "step": 3350
+    },
+    {
+      "epoch": 1.0752,
+      "grad_norm": 4.77971076965332,
+      "learning_rate": 3.388732394366198e-05,
+      "loss": 0.2528,
+      "step": 3360
+    },
+    {
+      "epoch": 1.0784,
+      "grad_norm": 5.056029319763184,
+      "learning_rate": 3.3830985915492956e-05,
+      "loss": 0.1816,
+      "step": 3370
+    },
+    {
+      "epoch": 1.0816,
+      "grad_norm": 1.7177847623825073,
+      "learning_rate": 3.377464788732395e-05,
+      "loss": 0.1776,
+      "step": 3380
+    },
+    {
+      "epoch": 1.0848,
+      "grad_norm": 28.095943450927734,
+      "learning_rate": 3.371830985915493e-05,
+      "loss": 0.5433,
+      "step": 3390
+    },
+    {
+      "epoch": 1.088,
+      "grad_norm": 18.927244186401367,
+      "learning_rate": 3.366197183098592e-05,
+      "loss": 0.1047,
+      "step": 3400
+    },
+    {
+      "epoch": 1.0912,
+      "grad_norm": 13.517168998718262,
+      "learning_rate": 3.3605633802816905e-05,
+      "loss": 0.4499,
+      "step": 3410
+    },
+    {
+      "epoch": 1.0944,
+      "grad_norm": 8.441484451293945,
+      "learning_rate": 3.354929577464789e-05,
+      "loss": 0.1514,
+      "step": 3420
+    },
+    {
+      "epoch": 1.0976,
+      "grad_norm": 12.520185470581055,
+      "learning_rate": 3.3492957746478876e-05,
+      "loss": 0.329,
+      "step": 3430
+    },
+    {
+      "epoch": 1.1008,
+      "grad_norm": 5.51774263381958,
+      "learning_rate": 3.343661971830986e-05,
+      "loss": 0.4714,
+      "step": 3440
+    },
+    {
+      "epoch": 1.104,
+      "grad_norm": 0.26915284991264343,
+      "learning_rate": 3.338028169014085e-05,
+      "loss": 0.0801,
+      "step": 3450
+    },
+    {
+      "epoch": 1.1072,
+      "grad_norm": 8.110269546508789,
+      "learning_rate": 3.332394366197183e-05,
+      "loss": 0.2196,
+      "step": 3460
+    },
+    {
+      "epoch": 1.1104,
+      "grad_norm": 24.571348190307617,
+      "learning_rate": 3.326760563380282e-05,
+      "loss": 0.2151,
+      "step": 3470
+    },
+    {
+      "epoch": 1.1136,
+      "grad_norm": 4.834783554077148,
+      "learning_rate": 3.3211267605633804e-05,
+      "loss": 0.2286,
+      "step": 3480
+    },
+    {
+      "epoch": 1.1168,
+      "grad_norm": 0.4705199599266052,
+      "learning_rate": 3.315492957746479e-05,
+      "loss": 0.262,
+      "step": 3490
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 10.442275047302246,
+      "learning_rate": 3.3098591549295775e-05,
+      "loss": 0.1664,
+      "step": 3500
+    },
+    {
+      "epoch": 1.1232,
+      "grad_norm": 18.814613342285156,
+      "learning_rate": 3.304225352112677e-05,
+      "loss": 0.2115,
+      "step": 3510
+    },
+    {
+      "epoch": 1.1264,
+      "grad_norm": 0.24957086145877838,
+      "learning_rate": 3.2985915492957746e-05,
+      "loss": 0.3388,
+      "step": 3520
+    },
+    {
+      "epoch": 1.1296,
+      "grad_norm": 0.16773709654808044,
+      "learning_rate": 3.292957746478874e-05,
+      "loss": 0.0763,
+      "step": 3530
+    },
+    {
+      "epoch": 1.1328,
+      "grad_norm": 13.98716926574707,
+      "learning_rate": 3.2873239436619717e-05,
+      "loss": 0.1653,
+      "step": 3540
+    },
+    {
+      "epoch": 1.1360000000000001,
+      "grad_norm": 0.13978277146816254,
+      "learning_rate": 3.281690140845071e-05,
+      "loss": 0.4153,
+      "step": 3550
+    },
+    {
+      "epoch": 1.1392,
+      "grad_norm": 1.202662706375122,
+      "learning_rate": 3.276056338028169e-05,
+      "loss": 0.1792,
+      "step": 3560
+    },
+    {
+      "epoch": 1.1424,
+      "grad_norm": 0.2656092643737793,
+      "learning_rate": 3.270422535211268e-05,
+      "loss": 0.1704,
+      "step": 3570
+    },
+    {
+      "epoch": 1.1456,
+      "grad_norm": 0.1266532689332962,
+      "learning_rate": 3.2647887323943665e-05,
+      "loss": 0.1174,
+      "step": 3580
+    },
+    {
+      "epoch": 1.1488,
+      "grad_norm": 5.823769569396973,
+      "learning_rate": 3.259154929577465e-05,
+      "loss": 0.4392,
+      "step": 3590
+    },
+    {
+      "epoch": 1.152,
+      "grad_norm": 0.7768305540084839,
+      "learning_rate": 3.2535211267605636e-05,
+      "loss": 0.5019,
+      "step": 3600
+    },
+    {
+      "epoch": 1.1552,
+      "grad_norm": 4.988490104675293,
+      "learning_rate": 3.247887323943662e-05,
+      "loss": 0.1525,
+      "step": 3610
+    },
+    {
+      "epoch": 1.1584,
+      "grad_norm": 7.760066509246826,
+      "learning_rate": 3.242253521126761e-05,
+      "loss": 0.2967,
+      "step": 3620
+    },
+    {
+      "epoch": 1.1616,
+      "grad_norm": 0.3713386356830597,
+      "learning_rate": 3.236619718309859e-05,
+      "loss": 0.2972,
+      "step": 3630
+    },
+    {
+      "epoch": 1.1648,
+      "grad_norm": 0.1917293667793274,
+      "learning_rate": 3.230985915492958e-05,
+      "loss": 0.1505,
+      "step": 3640
+    },
+    {
+      "epoch": 1.168,
+      "grad_norm": 0.31380757689476013,
+      "learning_rate": 3.2253521126760564e-05,
+      "loss": 0.3006,
+      "step": 3650
+    },
+    {
+      "epoch": 1.1712,
+      "grad_norm": 18.34935188293457,
+      "learning_rate": 3.219718309859155e-05,
+      "loss": 0.5165,
+      "step": 3660
+    },
+    {
+      "epoch": 1.1743999999999999,
+      "grad_norm": 1.479733943939209,
+      "learning_rate": 3.2140845070422535e-05,
+      "loss": 0.104,
+      "step": 3670
+    },
+    {
+      "epoch": 1.1776,
+      "grad_norm": 29.272424697875977,
+      "learning_rate": 3.208450704225353e-05,
+      "loss": 0.1871,
+      "step": 3680
+    },
+    {
+      "epoch": 1.1808,
+      "grad_norm": 31.237834930419922,
+      "learning_rate": 3.2028169014084506e-05,
+      "loss": 0.3294,
+      "step": 3690
+    },
+    {
+      "epoch": 1.184,
+      "grad_norm": 0.5057691335678101,
+      "learning_rate": 3.19718309859155e-05,
+      "loss": 0.3202,
+      "step": 3700
+    },
+    {
+      "epoch": 1.1872,
+      "grad_norm": 35.599571228027344,
+      "learning_rate": 3.191549295774648e-05,
+      "loss": 0.5781,
+      "step": 3710
+    },
+    {
+      "epoch": 1.1904,
+      "grad_norm": 18.676931381225586,
+      "learning_rate": 3.185915492957747e-05,
+      "loss": 0.2537,
+      "step": 3720
+    },
+    {
+      "epoch": 1.1936,
+      "grad_norm": 17.747034072875977,
+      "learning_rate": 3.180281690140845e-05,
+      "loss": 0.2176,
+      "step": 3730
+    },
+    {
+      "epoch": 1.1968,
+      "grad_norm": 20.384511947631836,
+      "learning_rate": 3.174647887323944e-05,
+      "loss": 0.3224,
+      "step": 3740
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 1.9572980403900146,
+      "learning_rate": 3.1690140845070426e-05,
+      "loss": 0.3297,
+      "step": 3750
+    },
+    {
+      "epoch": 1.2032,
+      "grad_norm": 11.518452644348145,
+      "learning_rate": 3.163380281690141e-05,
+      "loss": 0.0754,
+      "step": 3760
+    },
+    {
+      "epoch": 1.2064,
+      "grad_norm": 21.269893646240234,
+      "learning_rate": 3.1577464788732397e-05,
+      "loss": 0.3698,
+      "step": 3770
+    },
+    {
+      "epoch": 1.2096,
+      "grad_norm": 6.943552494049072,
+      "learning_rate": 3.152112676056338e-05,
+      "loss": 0.3419,
+      "step": 3780
+    },
+    {
+      "epoch": 1.2128,
+      "grad_norm": 2.637138605117798,
+      "learning_rate": 3.146478873239437e-05,
+      "loss": 0.4475,
+      "step": 3790
+    },
+    {
+      "epoch": 1.216,
+      "grad_norm": 0.2345736175775528,
+      "learning_rate": 3.140845070422535e-05,
+      "loss": 0.26,
+      "step": 3800
+    },
+    {
+      "epoch": 1.2192,
+      "grad_norm": 23.990619659423828,
+      "learning_rate": 3.135211267605634e-05,
+      "loss": 0.4548,
+      "step": 3810
+    },
+    {
+      "epoch": 1.2224,
+      "grad_norm": 30.698591232299805,
+      "learning_rate": 3.1295774647887324e-05,
+      "loss": 0.2941,
+      "step": 3820
+    },
+    {
+      "epoch": 1.2256,
+      "grad_norm": 8.31469440460205,
+      "learning_rate": 3.123943661971831e-05,
+      "loss": 0.2317,
+      "step": 3830
+    },
+    {
+      "epoch": 1.2288000000000001,
+      "grad_norm": 0.8447297215461731,
+      "learning_rate": 3.1183098591549295e-05,
+      "loss": 0.1829,
+      "step": 3840
+    },
+    {
+      "epoch": 1.232,
+      "grad_norm": 1.337220549583435,
+      "learning_rate": 3.112676056338028e-05,
+      "loss": 0.1399,
+      "step": 3850
+    },
+    {
+      "epoch": 1.2352,
+      "grad_norm": 0.0955493152141571,
+      "learning_rate": 3.1070422535211266e-05,
+      "loss": 0.4618,
+      "step": 3860
+    },
+    {
+      "epoch": 1.2384,
+      "grad_norm": 0.7334279417991638,
+      "learning_rate": 3.101408450704226e-05,
+      "loss": 0.0653,
+      "step": 3870
+    },
+    {
+      "epoch": 1.2416,
+      "grad_norm": 0.8991394639015198,
+      "learning_rate": 3.095774647887324e-05,
+      "loss": 0.6064,
+      "step": 3880
+    },
+    {
+      "epoch": 1.2448,
+      "grad_norm": 30.786052703857422,
+      "learning_rate": 3.090140845070423e-05,
+      "loss": 0.2257,
+      "step": 3890
+    },
+    {
+      "epoch": 1.248,
+      "grad_norm": 8.322766304016113,
+      "learning_rate": 3.084507042253521e-05,
+      "loss": 0.3758,
+      "step": 3900
+    },
+    {
+      "epoch": 1.2511999999999999,
+      "grad_norm": 1.9371285438537598,
+      "learning_rate": 3.07887323943662e-05,
+      "loss": 0.1619,
+      "step": 3910
+    },
+    {
+      "epoch": 1.2544,
+      "grad_norm": 30.936664581298828,
+      "learning_rate": 3.0732394366197186e-05,
+      "loss": 0.226,
+      "step": 3920
+    },
+    {
+      "epoch": 1.2576,
+      "grad_norm": 18.97284507751465,
+      "learning_rate": 3.067605633802817e-05,
+      "loss": 0.2106,
+      "step": 3930
+    },
+    {
+      "epoch": 1.2608,
+      "grad_norm": 0.17548918724060059,
+      "learning_rate": 3.061971830985916e-05,
+      "loss": 0.1827,
+      "step": 3940
+    },
+    {
+      "epoch": 1.264,
+      "grad_norm": 53.99245834350586,
+      "learning_rate": 3.056338028169014e-05,
+      "loss": 0.1693,
+      "step": 3950
+    },
+    {
+      "epoch": 1.2671999999999999,
+      "grad_norm": 35.83251190185547,
+      "learning_rate": 3.0507042253521128e-05,
+      "loss": 0.2246,
+      "step": 3960
+    },
+    {
+      "epoch": 1.2704,
+      "grad_norm": 0.14614693820476532,
+      "learning_rate": 3.0450704225352117e-05,
+      "loss": 0.0399,
+      "step": 3970
+    },
+    {
+      "epoch": 1.2736,
+      "grad_norm": 0.09585163742303848,
+      "learning_rate": 3.03943661971831e-05,
+      "loss": 0.3103,
+      "step": 3980
+    },
+    {
+      "epoch": 1.2768,
+      "grad_norm": 0.35296738147735596,
+      "learning_rate": 3.0338028169014088e-05,
+      "loss": 0.4458,
+      "step": 3990
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 25.493444442749023,
+      "learning_rate": 3.028169014084507e-05,
+      "loss": 0.2397,
+      "step": 4000
+    },
+    {
+      "epoch": 1.2832,
+      "grad_norm": 21.99680519104004,
+      "learning_rate": 3.022535211267606e-05,
+      "loss": 0.3189,
+      "step": 4010
+    },
+    {
+      "epoch": 1.2864,
+      "grad_norm": 1.8091436624526978,
+      "learning_rate": 3.016901408450704e-05,
+      "loss": 0.1739,
+      "step": 4020
+    },
+    {
+      "epoch": 1.2896,
+      "grad_norm": 0.42829862236976624,
+      "learning_rate": 3.011267605633803e-05,
+      "loss": 0.1976,
+      "step": 4030
+    },
+    {
+      "epoch": 1.2928,
+      "grad_norm": 0.08488719165325165,
+      "learning_rate": 3.005633802816902e-05,
+      "loss": 0.2067,
+      "step": 4040
+    },
+    {
+      "epoch": 1.296,
+      "grad_norm": 9.230164527893066,
+      "learning_rate": 3e-05,
+      "loss": 0.2783,
+      "step": 4050
+    },
+    {
+      "epoch": 1.2992,
+      "grad_norm": 2.309288263320923,
+      "learning_rate": 2.994366197183099e-05,
+      "loss": 0.2829,
+      "step": 4060
+    },
+    {
+      "epoch": 1.3024,
+      "grad_norm": 38.344730377197266,
+      "learning_rate": 2.9887323943661972e-05,
+      "loss": 0.2297,
+      "step": 4070
+    },
+    {
+      "epoch": 1.3056,
+      "grad_norm": 6.4838337898254395,
+      "learning_rate": 2.983098591549296e-05,
+      "loss": 0.4281,
+      "step": 4080
+    },
+    {
+      "epoch": 1.3088,
+      "grad_norm": 30.31648826599121,
+      "learning_rate": 2.9774647887323946e-05,
+      "loss": 0.1629,
+      "step": 4090
+    },
+    {
+      "epoch": 1.312,
+      "grad_norm": 2.2914836406707764,
+      "learning_rate": 2.971830985915493e-05,
+      "loss": 0.2485,
+      "step": 4100
+    },
+    {
+      "epoch": 1.3152,
+      "grad_norm": 0.811667263507843,
+      "learning_rate": 2.9661971830985917e-05,
+      "loss": 0.096,
+      "step": 4110
+    },
+    {
+      "epoch": 1.3184,
+      "grad_norm": 10.037412643432617,
+      "learning_rate": 2.9605633802816903e-05,
+      "loss": 0.4057,
+      "step": 4120
+    },
+    {
+      "epoch": 1.3216,
+      "grad_norm": 15.876739501953125,
+      "learning_rate": 2.9549295774647888e-05,
+      "loss": 0.2493,
+      "step": 4130
+    },
+    {
+      "epoch": 1.3248,
+      "grad_norm": 0.10714894533157349,
+      "learning_rate": 2.9492957746478874e-05,
+      "loss": 0.1766,
+      "step": 4140
+    },
+    {
+      "epoch": 1.328,
+      "grad_norm": 0.9898651242256165,
+      "learning_rate": 2.943661971830986e-05,
+      "loss": 0.1064,
+      "step": 4150
+    },
+    {
+      "epoch": 1.3312,
+      "grad_norm": 17.39281463623047,
+      "learning_rate": 2.9380281690140848e-05,
+      "loss": 0.1863,
+      "step": 4160
+    },
+    {
+      "epoch": 1.3344,
+      "grad_norm": 4.467952728271484,
+      "learning_rate": 2.932394366197183e-05,
+      "loss": 0.3616,
+      "step": 4170
+    },
+    {
+      "epoch": 1.3376000000000001,
+      "grad_norm": 0.1744953840970993,
+      "learning_rate": 2.926760563380282e-05,
+      "loss": 0.3911,
+      "step": 4180
+    },
+    {
+      "epoch": 1.3408,
+      "grad_norm": 11.605582237243652,
+      "learning_rate": 2.92112676056338e-05,
+      "loss": 0.086,
+      "step": 4190
+    },
+    {
+      "epoch": 1.3439999999999999,
+      "grad_norm": 6.277223110198975,
+      "learning_rate": 2.915492957746479e-05,
+      "loss": 0.1406,
+      "step": 4200
+    },
+    {
+      "epoch": 1.3472,
+      "grad_norm": 55.23845672607422,
+      "learning_rate": 2.909859154929578e-05,
+      "loss": 0.3971,
+      "step": 4210
+    },
+    {
+      "epoch": 1.3504,
+      "grad_norm": 14.972460746765137,
+      "learning_rate": 2.904225352112676e-05,
+      "loss": 0.2301,
+      "step": 4220
+    },
+    {
+      "epoch": 1.3536000000000001,
+      "grad_norm": 0.19094644486904144,
+      "learning_rate": 2.898591549295775e-05,
+      "loss": 0.3753,
+      "step": 4230
+    },
+    {
+      "epoch": 1.3568,
+      "grad_norm": 4.376794815063477,
+      "learning_rate": 2.8929577464788732e-05,
+      "loss": 0.1735,
+      "step": 4240
+    },
+    {
+      "epoch": 1.3599999999999999,
+      "grad_norm": 50.90641403198242,
+      "learning_rate": 2.887323943661972e-05,
+      "loss": 0.22,
+      "step": 4250
+    },
+    {
+      "epoch": 1.3632,
+      "grad_norm": 10.364707946777344,
+      "learning_rate": 2.881690140845071e-05,
+      "loss": 0.1155,
+      "step": 4260
+    },
+    {
+      "epoch": 1.3664,
+      "grad_norm": 1.0661869049072266,
+      "learning_rate": 2.8760563380281692e-05,
+      "loss": 0.3827,
+      "step": 4270
+    },
+    {
+      "epoch": 1.3696,
+      "grad_norm": 17.855276107788086,
+      "learning_rate": 2.870422535211268e-05,
+      "loss": 0.3932,
+      "step": 4280
+    },
+    {
+      "epoch": 1.3728,
+      "grad_norm": 27.09354019165039,
+      "learning_rate": 2.8647887323943663e-05,
+      "loss": 0.3695,
+      "step": 4290
+    },
+    {
+      "epoch": 1.376,
+      "grad_norm": 0.13813486695289612,
+      "learning_rate": 2.859154929577465e-05,
+      "loss": 0.2251,
+      "step": 4300
+    },
+    {
+      "epoch": 1.3792,
+      "grad_norm": 16.627347946166992,
+      "learning_rate": 2.8535211267605634e-05,
+      "loss": 0.3089,
+      "step": 4310
+    },
+    {
+      "epoch": 1.3824,
+      "grad_norm": 0.3655046820640564,
+      "learning_rate": 2.847887323943662e-05,
+      "loss": 0.2134,
+      "step": 4320
+    },
+    {
+      "epoch": 1.3856,
+      "grad_norm": 38.9260368347168,
+      "learning_rate": 2.842253521126761e-05,
+      "loss": 0.3283,
+      "step": 4330
+    },
+    {
+      "epoch": 1.3888,
+      "grad_norm": 26.279027938842773,
+      "learning_rate": 2.836619718309859e-05,
+      "loss": 0.2663,
+      "step": 4340
+    },
+    {
+      "epoch": 1.392,
+      "grad_norm": 49.378902435302734,
+      "learning_rate": 2.830985915492958e-05,
+      "loss": 0.2607,
+      "step": 4350
+    },
+    {
+      "epoch": 1.3952,
+      "grad_norm": 0.32010015845298767,
+      "learning_rate": 2.825352112676056e-05,
+      "loss": 0.423,
+      "step": 4360
+    },
+    {
+      "epoch": 1.3984,
+      "grad_norm": 5.871781349182129,
+      "learning_rate": 2.819718309859155e-05,
+      "loss": 0.1471,
+      "step": 4370
+    },
+    {
+      "epoch": 1.4016,
+      "grad_norm": 3.814654588699341,
+      "learning_rate": 2.814084507042254e-05,
+      "loss": 0.2127,
+      "step": 4380
+    },
+    {
+      "epoch": 1.4048,
+      "grad_norm": 0.4913332164287567,
+      "learning_rate": 2.808450704225352e-05,
+      "loss": 0.4049,
+      "step": 4390
+    },
+    {
+      "epoch": 1.408,
+      "grad_norm": 1.195360779762268,
+      "learning_rate": 2.802816901408451e-05,
+      "loss": 0.3488,
+      "step": 4400
+    },
+    {
+      "epoch": 1.4112,
+      "grad_norm": 30.266616821289062,
+      "learning_rate": 2.7971830985915492e-05,
+      "loss": 0.2471,
+      "step": 4410
+    },
+    {
+      "epoch": 1.4144,
+      "grad_norm": 0.1005750447511673,
+      "learning_rate": 2.791549295774648e-05,
+      "loss": 0.1442,
+      "step": 4420
+    },
+    {
+      "epoch": 1.4176,
+      "grad_norm": 19.245065689086914,
+      "learning_rate": 2.7859154929577463e-05,
+      "loss": 0.3952,
+      "step": 4430
+    },
+    {
+      "epoch": 1.4208,
+      "grad_norm": 0.5281161069869995,
+      "learning_rate": 2.7802816901408452e-05,
+      "loss": 0.1281,
+      "step": 4440
+    },
+    {
+      "epoch": 1.424,
+      "grad_norm": 0.20444443821907043,
+      "learning_rate": 2.774647887323944e-05,
+      "loss": 0.3929,
+      "step": 4450
+    },
+    {
+      "epoch": 1.4272,
+      "grad_norm": 0.12726615369319916,
+      "learning_rate": 2.7690140845070423e-05,
+      "loss": 0.1062,
+      "step": 4460
+    },
+    {
+      "epoch": 1.4304000000000001,
+      "grad_norm": 108.31690216064453,
+      "learning_rate": 2.7633802816901412e-05,
+      "loss": 0.1228,
+      "step": 4470
+    },
+    {
+      "epoch": 1.4336,
+      "grad_norm": 25.274940490722656,
+      "learning_rate": 2.7577464788732394e-05,
+      "loss": 0.2112,
+      "step": 4480
+    },
+    {
+      "epoch": 1.4368,
+      "grad_norm": 7.059344291687012,
+      "learning_rate": 2.7521126760563383e-05,
+      "loss": 0.4667,
+      "step": 4490
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 24.798084259033203,
+      "learning_rate": 2.746478873239437e-05,
+      "loss": 0.3644,
+      "step": 4500
+    },
+    {
+      "epoch": 1.4432,
+      "grad_norm": 11.129374504089355,
+      "learning_rate": 2.7408450704225354e-05,
+      "loss": 0.4076,
+      "step": 4510
+    },
+    {
+      "epoch": 1.4464000000000001,
+      "grad_norm": 6.293646335601807,
+      "learning_rate": 2.735211267605634e-05,
+      "loss": 0.2754,
+      "step": 4520
+    },
+    {
+      "epoch": 1.4496,
+      "grad_norm": 22.136383056640625,
+      "learning_rate": 2.7295774647887322e-05,
+      "loss": 0.4665,
+      "step": 4530
+    },
+    {
+      "epoch": 1.4527999999999999,
+      "grad_norm": 36.15532684326172,
+      "learning_rate": 2.723943661971831e-05,
+      "loss": 0.2789,
+      "step": 4540
+    },
+    {
+      "epoch": 1.456,
+      "grad_norm": 58.91614532470703,
+      "learning_rate": 2.71830985915493e-05,
+      "loss": 0.2236,
+      "step": 4550
+    },
+    {
+      "epoch": 1.4592,
+      "grad_norm": 5.225749492645264,
+      "learning_rate": 2.712676056338028e-05,
+      "loss": 0.459,
+      "step": 4560
+    },
+    {
+      "epoch": 1.4624,
+      "grad_norm": 11.404582977294922,
+      "learning_rate": 2.707042253521127e-05,
+      "loss": 0.2615,
+      "step": 4570
+    },
+    {
+      "epoch": 1.4656,
+      "grad_norm": 15.184187889099121,
+      "learning_rate": 2.7014084507042253e-05,
+      "loss": 0.242,
+      "step": 4580
+    },
+    {
+      "epoch": 1.4687999999999999,
+      "grad_norm": 29.988828659057617,
+      "learning_rate": 2.695774647887324e-05,
+      "loss": 0.1761,
+      "step": 4590
+    },
+    {
+      "epoch": 1.472,
+      "grad_norm": 0.3717154562473297,
+      "learning_rate": 2.6901408450704224e-05,
+      "loss": 0.1711,
+      "step": 4600
+    },
+    {
+      "epoch": 1.4752,
+      "grad_norm": 0.5243228077888489,
+      "learning_rate": 2.6845070422535213e-05,
+      "loss": 0.3551,
+      "step": 4610
+    },
+    {
+      "epoch": 1.4784,
+      "grad_norm": 0.14952997863292694,
+      "learning_rate": 2.67887323943662e-05,
+      "loss": 0.2119,
+      "step": 4620
+    },
+    {
+      "epoch": 1.4816,
+      "grad_norm": 0.5155125856399536,
+      "learning_rate": 2.6732394366197184e-05,
+      "loss": 0.1516,
+      "step": 4630
+    },
+    {
+      "epoch": 1.4848,
+      "grad_norm": 16.079330444335938,
+      "learning_rate": 2.6676056338028172e-05,
+      "loss": 0.3477,
+      "step": 4640
+    },
+    {
+      "epoch": 1.488,
+      "grad_norm": 0.5251998901367188,
+      "learning_rate": 2.6619718309859155e-05,
+      "loss": 0.3483,
+      "step": 4650
+    },
+    {
+      "epoch": 1.4912,
+      "grad_norm": 12.90518569946289,
+      "learning_rate": 2.6563380281690143e-05,
+      "loss": 0.3346,
+      "step": 4660
+    },
+    {
+      "epoch": 1.4944,
+      "grad_norm": 3.163393259048462,
+      "learning_rate": 2.650704225352113e-05,
+      "loss": 0.4457,
+      "step": 4670
+    },
+    {
+      "epoch": 1.4976,
+      "grad_norm": 0.8409318327903748,
+      "learning_rate": 2.6450704225352114e-05,
+      "loss": 0.1024,
+      "step": 4680
+    },
+    {
+      "epoch": 1.5008,
+      "grad_norm": 0.23881012201309204,
+      "learning_rate": 2.63943661971831e-05,
+      "loss": 0.2065,
+      "step": 4690
+    },
+    {
+      "epoch": 1.504,
+      "grad_norm": 31.078039169311523,
+      "learning_rate": 2.6338028169014085e-05,
+      "loss": 0.2194,
+      "step": 4700
+    },
+    {
+      "epoch": 1.5072,
+      "grad_norm": 0.28144362568855286,
+      "learning_rate": 2.628169014084507e-05,
+      "loss": 0.0813,
+      "step": 4710
+    },
+    {
+      "epoch": 1.5104,
+      "grad_norm": 10.667701721191406,
+      "learning_rate": 2.6225352112676056e-05,
+      "loss": 0.3627,
+      "step": 4720
+    },
+    {
+      "epoch": 1.5135999999999998,
+      "grad_norm": 0.4722766578197479,
+      "learning_rate": 2.6169014084507042e-05,
+      "loss": 0.3747,
+      "step": 4730
+    },
+    {
+      "epoch": 1.5168,
+      "grad_norm": 13.2311429977417,
+      "learning_rate": 2.611267605633803e-05,
+      "loss": 0.1583,
+      "step": 4740
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 1.4900763034820557,
+      "learning_rate": 2.6056338028169013e-05,
+      "loss": 0.2564,
+      "step": 4750
+    },
+    {
+      "epoch": 1.5232,
+      "grad_norm": 32.169681549072266,
+      "learning_rate": 2.6000000000000002e-05,
+      "loss": 0.2757,
+      "step": 4760
+    },
+    {
+      "epoch": 1.5264,
+      "grad_norm": 25.864120483398438,
+      "learning_rate": 2.5943661971830984e-05,
+      "loss": 0.2264,
+      "step": 4770
+    },
+    {
+      "epoch": 1.5295999999999998,
+      "grad_norm": 5.702986717224121,
+      "learning_rate": 2.5887323943661973e-05,
+      "loss": 0.1946,
+      "step": 4780
+    },
+    {
+      "epoch": 1.5328,
+      "grad_norm": 0.28651973605155945,
+      "learning_rate": 2.583098591549296e-05,
+      "loss": 0.0838,
+      "step": 4790
+    },
+    {
+      "epoch": 1.536,
+      "grad_norm": 0.4322168529033661,
+      "learning_rate": 2.5774647887323944e-05,
+      "loss": 0.2331,
+      "step": 4800
+    },
+    {
+      "epoch": 1.5392000000000001,
+      "grad_norm": 0.17899250984191895,
+      "learning_rate": 2.5718309859154933e-05,
+      "loss": 0.2475,
+      "step": 4810
+    },
+    {
+      "epoch": 1.5424,
+      "grad_norm": 30.139265060424805,
+      "learning_rate": 2.5661971830985915e-05,
+      "loss": 0.4416,
+      "step": 4820
+    },
+    {
+      "epoch": 1.5455999999999999,
+      "grad_norm": 0.23452678322792053,
+      "learning_rate": 2.5605633802816904e-05,
+      "loss": 0.2825,
+      "step": 4830
+    },
+    {
+      "epoch": 1.5488,
+      "grad_norm": 0.22751548886299133,
+      "learning_rate": 2.5549295774647893e-05,
+      "loss": 0.2478,
+      "step": 4840
+    },
+    {
+      "epoch": 1.552,
+      "grad_norm": 6.262718200683594,
+      "learning_rate": 2.5492957746478875e-05,
+      "loss": 0.2214,
+      "step": 4850
+    },
+    {
+      "epoch": 1.5552000000000001,
+      "grad_norm": 0.2855672836303711,
+      "learning_rate": 2.5436619718309864e-05,
+      "loss": 0.1745,
+      "step": 4860
+    },
+    {
+      "epoch": 1.5584,
+      "grad_norm": 4.517999649047852,
+      "learning_rate": 2.5380281690140846e-05,
+      "loss": 0.2678,
+      "step": 4870
+    },
+    {
+      "epoch": 1.5615999999999999,
+      "grad_norm": 31.07318115234375,
+      "learning_rate": 2.5323943661971835e-05,
+      "loss": 0.2528,
+      "step": 4880
+    },
+    {
+      "epoch": 1.5648,
+      "grad_norm": 1.6451767683029175,
+      "learning_rate": 2.5267605633802817e-05,
+      "loss": 0.1002,
+      "step": 4890
+    },
+    {
+      "epoch": 1.568,
+      "grad_norm": 32.15398406982422,
+      "learning_rate": 2.5211267605633802e-05,
+      "loss": 0.3257,
+      "step": 4900
+    },
+    {
+      "epoch": 1.5712000000000002,
+      "grad_norm": 7.450695037841797,
+      "learning_rate": 2.515492957746479e-05,
+      "loss": 0.2022,
+      "step": 4910
+    },
+    {
+      "epoch": 1.5744,
+      "grad_norm": 14.619000434875488,
+      "learning_rate": 2.5098591549295773e-05,
+      "loss": 0.346,
+      "step": 4920
+    },
+    {
+      "epoch": 1.5776,
+      "grad_norm": 0.7622524499893188,
+      "learning_rate": 2.5042253521126762e-05,
+      "loss": 0.1638,
+      "step": 4930
+    },
+    {
+      "epoch": 1.5808,
+      "grad_norm": 2.6016695499420166,
+      "learning_rate": 2.4985915492957748e-05,
+      "loss": 0.0822,
+      "step": 4940
+    },
+    {
+      "epoch": 1.584,
+      "grad_norm": 2.1474409103393555,
+      "learning_rate": 2.4929577464788733e-05,
+      "loss": 0.1602,
+      "step": 4950
+    },
+    {
+      "epoch": 1.5872000000000002,
+      "grad_norm": 65.45417785644531,
+      "learning_rate": 2.487323943661972e-05,
+      "loss": 0.1883,
+      "step": 4960
+    },
+    {
+      "epoch": 1.5904,
+      "grad_norm": 13.360310554504395,
+      "learning_rate": 2.4816901408450704e-05,
+      "loss": 0.2619,
+      "step": 4970
+    },
+    {
+      "epoch": 1.5936,
+      "grad_norm": 0.6584329009056091,
+      "learning_rate": 2.476056338028169e-05,
+      "loss": 0.2612,
+      "step": 4980
+    },
+    {
+      "epoch": 1.5968,
+      "grad_norm": 54.88881301879883,
+      "learning_rate": 2.470422535211268e-05,
+      "loss": 0.6558,
+      "step": 4990
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.08469274640083313,
+      "learning_rate": 2.4647887323943664e-05,
+      "loss": 0.2935,
+      "step": 5000
+    },
+    {
+      "epoch": 1.6032,
+      "grad_norm": 38.17769241333008,
+      "learning_rate": 2.459154929577465e-05,
+      "loss": 0.2391,
+      "step": 5010
+    },
+    {
+      "epoch": 1.6064,
+      "grad_norm": 0.1355709284543991,
+      "learning_rate": 2.4535211267605635e-05,
+      "loss": 0.1574,
+      "step": 5020
+    },
+    {
+      "epoch": 1.6096,
+      "grad_norm": 7.013975143432617,
+      "learning_rate": 2.447887323943662e-05,
+      "loss": 0.1076,
+      "step": 5030
+    },
+    {
+      "epoch": 1.6128,
+      "grad_norm": 13.909317970275879,
+      "learning_rate": 2.442253521126761e-05,
+      "loss": 0.4274,
+      "step": 5040
+    },
+    {
+      "epoch": 1.616,
+      "grad_norm": 4.903537273406982,
+      "learning_rate": 2.4366197183098595e-05,
+      "loss": 0.2527,
+      "step": 5050
+    },
+    {
+      "epoch": 1.6192,
+      "grad_norm": 9.500699996948242,
+      "learning_rate": 2.430985915492958e-05,
+      "loss": 0.4478,
+      "step": 5060
+    },
+    {
+      "epoch": 1.6223999999999998,
+      "grad_norm": 47.62290954589844,
+      "learning_rate": 2.4253521126760566e-05,
+      "loss": 0.2439,
+      "step": 5070
+    },
+    {
+      "epoch": 1.6256,
+      "grad_norm": 0.21192322671413422,
+      "learning_rate": 2.419718309859155e-05,
+      "loss": 0.2524,
+      "step": 5080
+    },
+    {
+      "epoch": 1.6288,
+      "grad_norm": 3.06548810005188,
+      "learning_rate": 2.4140845070422537e-05,
+      "loss": 0.1995,
+      "step": 5090
+    },
+    {
+      "epoch": 1.6320000000000001,
+      "grad_norm": 36.12741470336914,
+      "learning_rate": 2.4084507042253522e-05,
+      "loss": 0.2553,
+      "step": 5100
+    },
+    {
+      "epoch": 1.6352,
+      "grad_norm": 9.318374633789062,
+      "learning_rate": 2.4028169014084508e-05,
+      "loss": 0.2452,
+      "step": 5110
+    },
+    {
+      "epoch": 1.6383999999999999,
+      "grad_norm": 27.07297134399414,
+      "learning_rate": 2.3971830985915493e-05,
+      "loss": 0.4859,
+      "step": 5120
+    },
+    {
+      "epoch": 1.6416,
+      "grad_norm": 17.92713165283203,
+      "learning_rate": 2.391549295774648e-05,
+      "loss": 0.2726,
+      "step": 5130
+    },
+    {
+      "epoch": 1.6448,
+      "grad_norm": 20.595443725585938,
+      "learning_rate": 2.3859154929577464e-05,
+      "loss": 0.4673,
+      "step": 5140
+    },
+    {
+      "epoch": 1.6480000000000001,
+      "grad_norm": 15.670424461364746,
+      "learning_rate": 2.380281690140845e-05,
+      "loss": 0.131,
+      "step": 5150
+    },
+    {
+      "epoch": 1.6512,
+      "grad_norm": 0.27188238501548767,
+      "learning_rate": 2.374647887323944e-05,
+      "loss": 0.0991,
+      "step": 5160
+    },
+    {
+      "epoch": 1.6543999999999999,
+      "grad_norm": 0.1418936550617218,
+      "learning_rate": 2.3690140845070424e-05,
+      "loss": 0.3724,
+      "step": 5170
+    },
+    {
+      "epoch": 1.6576,
+      "grad_norm": 14.037035942077637,
+      "learning_rate": 2.363380281690141e-05,
+      "loss": 0.3105,
+      "step": 5180
+    },
+    {
+      "epoch": 1.6608,
+      "grad_norm": 1.6368416547775269,
+      "learning_rate": 2.3577464788732395e-05,
+      "loss": 0.174,
+      "step": 5190
+    },
+    {
+      "epoch": 1.6640000000000001,
+      "grad_norm": 16.241477966308594,
+      "learning_rate": 2.352112676056338e-05,
+      "loss": 0.2609,
+      "step": 5200
+    },
+    {
+      "epoch": 1.6672,
+      "grad_norm": 0.27356526255607605,
+      "learning_rate": 2.3464788732394366e-05,
+      "loss": 0.1827,
+      "step": 5210
+    },
+    {
+      "epoch": 1.6703999999999999,
+      "grad_norm": 35.08028030395508,
+      "learning_rate": 2.3408450704225355e-05,
+      "loss": 0.2963,
+      "step": 5220
+    },
+    {
+      "epoch": 1.6736,
+      "grad_norm": 0.12633004784584045,
+      "learning_rate": 2.335211267605634e-05,
+      "loss": 0.1284,
+      "step": 5230
+    },
+    {
+      "epoch": 1.6768,
+      "grad_norm": 10.867715835571289,
+      "learning_rate": 2.3295774647887326e-05,
+      "loss": 0.3947,
+      "step": 5240
+    },
+    {
+      "epoch": 1.6800000000000002,
+      "grad_norm": 0.37642917037010193,
+      "learning_rate": 2.323943661971831e-05,
+      "loss": 0.0878,
+      "step": 5250
+    },
+    {
+      "epoch": 1.6832,
+      "grad_norm": 9.886677742004395,
+      "learning_rate": 2.3183098591549297e-05,
+      "loss": 0.1667,
+      "step": 5260
+    },
+    {
+      "epoch": 1.6864,
+      "grad_norm": 0.5025785565376282,
+      "learning_rate": 2.3126760563380283e-05,
+      "loss": 0.2399,
+      "step": 5270
+    },
+    {
+      "epoch": 1.6896,
+      "grad_norm": 0.07013744115829468,
+      "learning_rate": 2.3070422535211268e-05,
+      "loss": 0.2592,
+      "step": 5280
+    },
+    {
+      "epoch": 1.6928,
+      "grad_norm": 9.038287162780762,
+      "learning_rate": 2.3014084507042254e-05,
+      "loss": 0.3243,
+      "step": 5290
+    },
+    {
+      "epoch": 1.696,
+      "grad_norm": 0.15734457969665527,
+      "learning_rate": 2.295774647887324e-05,
+      "loss": 0.0506,
+      "step": 5300
+    },
+    {
+      "epoch": 1.6992,
+      "grad_norm": 0.21506910026073456,
+      "learning_rate": 2.2901408450704225e-05,
+      "loss": 0.2006,
+      "step": 5310
+    },
+    {
+      "epoch": 1.7024,
+      "grad_norm": 11.207597732543945,
+      "learning_rate": 2.284507042253521e-05,
+      "loss": 0.2125,
+      "step": 5320
+    },
+    {
+      "epoch": 1.7056,
+      "grad_norm": 7.165248394012451,
+      "learning_rate": 2.27887323943662e-05,
+      "loss": 0.1971,
+      "step": 5330
+    },
+    {
+      "epoch": 1.7088,
+      "grad_norm": 0.8289473056793213,
+      "learning_rate": 2.2732394366197185e-05,
+      "loss": 0.1824,
+      "step": 5340
+    },
+    {
+      "epoch": 1.712,
+      "grad_norm": 1.2633789777755737,
+      "learning_rate": 2.267605633802817e-05,
+      "loss": 0.2601,
+      "step": 5350
+    },
+    {
+      "epoch": 1.7151999999999998,
+      "grad_norm": 38.94256591796875,
+      "learning_rate": 2.2619718309859156e-05,
+      "loss": 0.345,
+      "step": 5360
+    },
+    {
+      "epoch": 1.7184,
+      "grad_norm": 0.10120674222707748,
+      "learning_rate": 2.256338028169014e-05,
+      "loss": 0.1882,
+      "step": 5370
+    },
+    {
+      "epoch": 1.7216,
+      "grad_norm": 17.41254425048828,
+      "learning_rate": 2.2507042253521127e-05,
+      "loss": 0.1648,
+      "step": 5380
+    },
+    {
+      "epoch": 1.7248,
+      "grad_norm": 0.33858543634414673,
+      "learning_rate": 2.2450704225352115e-05,
+      "loss": 0.3624,
+      "step": 5390
+    },
+    {
+      "epoch": 1.728,
+      "grad_norm": 0.3513981103897095,
+      "learning_rate": 2.23943661971831e-05,
+      "loss": 0.3238,
+      "step": 5400
+    },
+    {
+      "epoch": 1.7311999999999999,
+      "grad_norm": 0.7570049166679382,
+      "learning_rate": 2.2338028169014086e-05,
+      "loss": 0.3377,
+      "step": 5410
+    },
+    {
+      "epoch": 1.7344,
+      "grad_norm": 0.7027788162231445,
+      "learning_rate": 2.2281690140845072e-05,
+      "loss": 0.1963,
+      "step": 5420
+    },
+    {
+      "epoch": 1.7376,
+      "grad_norm": 55.278343200683594,
+      "learning_rate": 2.2225352112676057e-05,
+      "loss": 0.4997,
+      "step": 5430
+    },
+    {
+      "epoch": 1.7408000000000001,
+      "grad_norm": 2.759753704071045,
+      "learning_rate": 2.2169014084507043e-05,
+      "loss": 0.2161,
+      "step": 5440
+    },
+    {
+      "epoch": 1.744,
+      "grad_norm": 13.195887565612793,
+      "learning_rate": 2.2112676056338032e-05,
+      "loss": 0.2539,
+      "step": 5450
+    },
+    {
+      "epoch": 1.7471999999999999,
+      "grad_norm": 12.78817081451416,
+      "learning_rate": 2.2056338028169017e-05,
+      "loss": 0.2056,
+      "step": 5460
+    },
+    {
+      "epoch": 1.7504,
+      "grad_norm": 40.1257209777832,
+      "learning_rate": 2.2000000000000003e-05,
+      "loss": 0.2951,
+      "step": 5470
+    },
+    {
+      "epoch": 1.7536,
+      "grad_norm": 0.3393701910972595,
+      "learning_rate": 2.1943661971830985e-05,
+      "loss": 0.2428,
+      "step": 5480
+    },
+    {
+      "epoch": 1.7568000000000001,
+      "grad_norm": 13.551216125488281,
+      "learning_rate": 2.188732394366197e-05,
+      "loss": 0.3264,
+      "step": 5490
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 41.21603012084961,
+      "learning_rate": 2.1830985915492956e-05,
+      "loss": 0.4246,
+      "step": 5500
+    },
+    {
+      "epoch": 1.7631999999999999,
+      "grad_norm": 9.464485168457031,
+      "learning_rate": 2.1774647887323945e-05,
+      "loss": 0.1579,
+      "step": 5510
+    },
+    {
+      "epoch": 1.7664,
+      "grad_norm": 45.843814849853516,
+      "learning_rate": 2.171830985915493e-05,
+      "loss": 0.1919,
+      "step": 5520
+    },
+    {
+      "epoch": 1.7696,
+      "grad_norm": 1.6334397792816162,
+      "learning_rate": 2.1661971830985916e-05,
+      "loss": 0.2822,
+      "step": 5530
+    },
+    {
+      "epoch": 1.7728000000000002,
+      "grad_norm": 0.7097220420837402,
+      "learning_rate": 2.16056338028169e-05,
+      "loss": 0.1146,
+      "step": 5540
+    },
+    {
+      "epoch": 1.776,
+      "grad_norm": 11.706197738647461,
+      "learning_rate": 2.1549295774647887e-05,
+      "loss": 0.2456,
+      "step": 5550
+    },
+    {
+      "epoch": 1.7792,
+      "grad_norm": 0.8858042359352112,
+      "learning_rate": 2.1492957746478876e-05,
+      "loss": 0.461,
+      "step": 5560
+    },
+    {
+      "epoch": 1.7824,
+      "grad_norm": 2.2900185585021973,
+      "learning_rate": 2.143661971830986e-05,
+      "loss": 0.0824,
+      "step": 5570
+    },
+    {
+      "epoch": 1.7856,
+      "grad_norm": 1.9041435718536377,
+      "learning_rate": 2.1380281690140847e-05,
+      "loss": 0.1843,
+      "step": 5580
+    },
+    {
+      "epoch": 1.7888,
+      "grad_norm": 9.106405258178711,
+      "learning_rate": 2.1323943661971832e-05,
+      "loss": 0.1421,
+      "step": 5590
+    },
+    {
+      "epoch": 1.792,
+      "grad_norm": 19.528039932250977,
+      "learning_rate": 2.1267605633802818e-05,
+      "loss": 0.3623,
+      "step": 5600
+    },
+    {
+      "epoch": 1.7952,
+      "grad_norm": 0.16566412150859833,
+      "learning_rate": 2.1211267605633803e-05,
+      "loss": 0.4338,
+      "step": 5610
+    },
+    {
+      "epoch": 1.7984,
+      "grad_norm": 0.8347293138504028,
+      "learning_rate": 2.1154929577464792e-05,
+      "loss": 0.2157,
+      "step": 5620
+    },
+    {
+      "epoch": 1.8016,
+      "grad_norm": 22.129648208618164,
+      "learning_rate": 2.1098591549295778e-05,
+      "loss": 0.2917,
+      "step": 5630
+    },
+    {
+      "epoch": 1.8048,
+      "grad_norm": 0.25225210189819336,
+      "learning_rate": 2.1042253521126763e-05,
+      "loss": 0.2006,
+      "step": 5640
+    },
+    {
+      "epoch": 1.808,
+      "grad_norm": 0.3600423336029053,
+      "learning_rate": 2.098591549295775e-05,
+      "loss": 0.3905,
+      "step": 5650
+    },
+    {
+      "epoch": 1.8112,
+      "grad_norm": 22.235361099243164,
+      "learning_rate": 2.0929577464788734e-05,
+      "loss": 0.2482,
+      "step": 5660
+    },
+    {
+      "epoch": 1.8144,
+      "grad_norm": 9.403947830200195,
+      "learning_rate": 2.087323943661972e-05,
+      "loss": 0.2406,
+      "step": 5670
+    },
+    {
+      "epoch": 1.8176,
+      "grad_norm": 1.0296498537063599,
+      "learning_rate": 2.0816901408450705e-05,
+      "loss": 0.1626,
+      "step": 5680
+    },
+    {
+      "epoch": 1.8208,
+      "grad_norm": 25.019081115722656,
+      "learning_rate": 2.076056338028169e-05,
+      "loss": 0.1861,
+      "step": 5690
+    },
+    {
+      "epoch": 1.8239999999999998,
+      "grad_norm": 6.003271579742432,
+      "learning_rate": 2.0704225352112676e-05,
+      "loss": 0.4198,
+      "step": 5700
+    },
+    {
+      "epoch": 1.8272,
+      "grad_norm": 0.24664323031902313,
+      "learning_rate": 2.064788732394366e-05,
+      "loss": 0.2311,
+      "step": 5710
+    },
+    {
+      "epoch": 1.8304,
+      "grad_norm": 0.3681061863899231,
+      "learning_rate": 2.0591549295774647e-05,
+      "loss": 0.3282,
+      "step": 5720
+    },
+    {
+      "epoch": 1.8336000000000001,
+      "grad_norm": 1.1691765785217285,
+      "learning_rate": 2.0535211267605633e-05,
+      "loss": 0.1885,
+      "step": 5730
+    },
+    {
+      "epoch": 1.8368,
+      "grad_norm": 43.80043029785156,
+      "learning_rate": 2.047887323943662e-05,
+      "loss": 0.2497,
+      "step": 5740
+    },
+    {
+      "epoch": 1.8399999999999999,
+      "grad_norm": 22.248729705810547,
+      "learning_rate": 2.0422535211267607e-05,
+      "loss": 0.1009,
+      "step": 5750
+    },
+    {
+      "epoch": 1.8432,
+      "grad_norm": 0.1887352615594864,
+      "learning_rate": 2.0366197183098592e-05,
+      "loss": 0.2099,
+      "step": 5760
+    },
+    {
+      "epoch": 1.8464,
+      "grad_norm": 39.37889862060547,
+      "learning_rate": 2.0309859154929578e-05,
+      "loss": 0.2349,
+      "step": 5770
+    },
+    {
+      "epoch": 1.8496000000000001,
+      "grad_norm": 2.4076569080352783,
+      "learning_rate": 2.0253521126760563e-05,
+      "loss": 0.3262,
+      "step": 5780
+    },
+    {
+      "epoch": 1.8528,
+      "grad_norm": 0.38582533597946167,
+      "learning_rate": 2.019718309859155e-05,
+      "loss": 0.0311,
+      "step": 5790
+    },
+    {
+      "epoch": 1.8559999999999999,
+      "grad_norm": 0.12109358608722687,
+      "learning_rate": 2.0140845070422538e-05,
+      "loss": 0.311,
+      "step": 5800
+    },
+    {
+      "epoch": 1.8592,
+      "grad_norm": 6.223245620727539,
+      "learning_rate": 2.0084507042253523e-05,
+      "loss": 0.4537,
+      "step": 5810
+    },
+    {
+      "epoch": 1.8624,
+      "grad_norm": 0.16367606818675995,
+      "learning_rate": 2.002816901408451e-05,
+      "loss": 0.3076,
+      "step": 5820
+    },
+    {
+      "epoch": 1.8656000000000001,
+      "grad_norm": 30.715484619140625,
+      "learning_rate": 1.9971830985915494e-05,
+      "loss": 0.346,
+      "step": 5830
+    },
+    {
+      "epoch": 1.8688,
+      "grad_norm": 42.86243438720703,
+      "learning_rate": 1.991549295774648e-05,
+      "loss": 0.0792,
+      "step": 5840
+    },
+    {
+      "epoch": 1.8719999999999999,
+      "grad_norm": 15.401847839355469,
+      "learning_rate": 1.9859154929577465e-05,
+      "loss": 0.2927,
+      "step": 5850
+    },
+    {
+      "epoch": 1.8752,
+      "grad_norm": 0.9566125273704529,
+      "learning_rate": 1.980281690140845e-05,
+      "loss": 0.2851,
+      "step": 5860
+    },
+    {
+      "epoch": 1.8784,
+      "grad_norm": 11.325541496276855,
+      "learning_rate": 1.9746478873239436e-05,
+      "loss": 0.4101,
+      "step": 5870
+    },
+    {
+      "epoch": 1.8816000000000002,
+      "grad_norm": 0.608905553817749,
+      "learning_rate": 1.9690140845070422e-05,
+      "loss": 0.0911,
+      "step": 5880
+    },
+    {
+      "epoch": 1.8848,
+      "grad_norm": 14.199214935302734,
+      "learning_rate": 1.9633802816901407e-05,
+      "loss": 0.4375,
+      "step": 5890
+    },
+    {
+      "epoch": 1.888,
+      "grad_norm": 20.619394302368164,
+      "learning_rate": 1.9577464788732393e-05,
+      "loss": 0.3372,
+      "step": 5900
+    },
+    {
+      "epoch": 1.8912,
+      "grad_norm": 11.778953552246094,
+      "learning_rate": 1.9521126760563382e-05,
+      "loss": 0.2411,
+      "step": 5910
+    },
+    {
+      "epoch": 1.8944,
+      "grad_norm": 111.18775939941406,
+      "learning_rate": 1.9464788732394367e-05,
+      "loss": 0.239,
+      "step": 5920
+    },
+    {
+      "epoch": 1.8976,
+      "grad_norm": 0.6485037207603455,
+      "learning_rate": 1.9408450704225353e-05,
+      "loss": 0.1002,
+      "step": 5930
+    },
+    {
+      "epoch": 1.9008,
+      "grad_norm": 51.51342010498047,
+      "learning_rate": 1.9352112676056338e-05,
+      "loss": 0.3721,
+      "step": 5940
+    },
+    {
+      "epoch": 1.904,
+      "grad_norm": 9.155081748962402,
+      "learning_rate": 1.9295774647887324e-05,
+      "loss": 0.2457,
+      "step": 5950
+    },
+    {
+      "epoch": 1.9072,
+      "grad_norm": 31.439834594726562,
+      "learning_rate": 1.923943661971831e-05,
+      "loss": 0.3378,
+      "step": 5960
+    },
+    {
+      "epoch": 1.9104,
+      "grad_norm": 38.59767532348633,
+      "learning_rate": 1.9183098591549298e-05,
+      "loss": 0.0842,
+      "step": 5970
+    },
+    {
+      "epoch": 1.9136,
+      "grad_norm": 30.20688819885254,
+      "learning_rate": 1.9126760563380284e-05,
+      "loss": 0.1471,
+      "step": 5980
+    },
+    {
+      "epoch": 1.9167999999999998,
+      "grad_norm": 76.58573150634766,
+      "learning_rate": 1.907042253521127e-05,
+      "loss": 0.2427,
+      "step": 5990
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.44934359192848206,
+      "learning_rate": 1.9014084507042255e-05,
+      "loss": 0.2053,
+      "step": 6000
+    },
+    {
+      "epoch": 1.9232,
+      "grad_norm": 12.041892051696777,
+      "learning_rate": 1.895774647887324e-05,
+      "loss": 0.235,
+      "step": 6010
+    },
+    {
+      "epoch": 1.9264000000000001,
+      "grad_norm": 0.10604248195886612,
+      "learning_rate": 1.8901408450704226e-05,
+      "loss": 0.2931,
+      "step": 6020
+    },
+    {
+      "epoch": 1.9296,
+      "grad_norm": 52.336849212646484,
+      "learning_rate": 1.8845070422535215e-05,
+      "loss": 0.46,
+      "step": 6030
+    },
+    {
+      "epoch": 1.9327999999999999,
+      "grad_norm": 0.30715158581733704,
+      "learning_rate": 1.87887323943662e-05,
+      "loss": 0.086,
+      "step": 6040
+    },
+    {
+      "epoch": 1.936,
+      "grad_norm": 12.268070220947266,
+      "learning_rate": 1.8732394366197186e-05,
+      "loss": 0.3663,
+      "step": 6050
+    },
+    {
+      "epoch": 1.9392,
+      "grad_norm": 0.7918230891227722,
+      "learning_rate": 1.867605633802817e-05,
+      "loss": 0.2688,
+      "step": 6060
+    },
+    {
+      "epoch": 1.9424000000000001,
+      "grad_norm": 0.276155024766922,
+      "learning_rate": 1.8619718309859157e-05,
+      "loss": 0.2618,
+      "step": 6070
+    },
+    {
+      "epoch": 1.9456,
+      "grad_norm": 0.6748977899551392,
+      "learning_rate": 1.8563380281690142e-05,
+      "loss": 0.0704,
+      "step": 6080
+    },
+    {
+      "epoch": 1.9487999999999999,
+      "grad_norm": 27.20131492614746,
+      "learning_rate": 1.8507042253521128e-05,
+      "loss": 0.4101,
+      "step": 6090
+    },
+    {
+      "epoch": 1.952,
+      "grad_norm": 0.1357765942811966,
+      "learning_rate": 1.8450704225352113e-05,
+      "loss": 0.354,
+      "step": 6100
+    },
+    {
+      "epoch": 1.9552,
+      "grad_norm": 21.598268508911133,
+      "learning_rate": 1.83943661971831e-05,
+      "loss": 0.2213,
+      "step": 6110
+    },
+    {
+      "epoch": 1.9584000000000001,
+      "grad_norm": 0.2529258728027344,
+      "learning_rate": 1.8338028169014084e-05,
+      "loss": 0.3205,
+      "step": 6120
+    },
+    {
+      "epoch": 1.9616,
+      "grad_norm": 2.0377094745635986,
+      "learning_rate": 1.828169014084507e-05,
+      "loss": 0.2115,
+      "step": 6130
+    },
+    {
+      "epoch": 1.9647999999999999,
+      "grad_norm": 9.584620475769043,
+      "learning_rate": 1.822535211267606e-05,
+      "loss": 0.4502,
+      "step": 6140
+    },
+    {
+      "epoch": 1.968,
+      "grad_norm": 0.26572760939598083,
+      "learning_rate": 1.8169014084507044e-05,
+      "loss": 0.4565,
+      "step": 6150
+    },
+    {
+      "epoch": 1.9712,
+      "grad_norm": 0.4592236876487732,
+      "learning_rate": 1.811267605633803e-05,
+      "loss": 0.1688,
+      "step": 6160
+    },
+    {
+      "epoch": 1.9744000000000002,
+      "grad_norm": 2.714552879333496,
+      "learning_rate": 1.8056338028169015e-05,
+      "loss": 0.1334,
+      "step": 6170
+    },
+    {
+      "epoch": 1.9776,
+      "grad_norm": 1.833774209022522,
+      "learning_rate": 1.8e-05,
+      "loss": 0.2703,
+      "step": 6180
+    },
+    {
+      "epoch": 1.9808,
+      "grad_norm": 0.23094186186790466,
+      "learning_rate": 1.7943661971830986e-05,
+      "loss": 0.2316,
+      "step": 6190
+    },
+    {
+      "epoch": 1.984,
+      "grad_norm": 2.593341112136841,
+      "learning_rate": 1.7887323943661975e-05,
+      "loss": 0.1416,
+      "step": 6200
+    },
+    {
+      "epoch": 1.9872,
+      "grad_norm": 63.741641998291016,
+      "learning_rate": 1.783098591549296e-05,
+      "loss": 0.1999,
+      "step": 6210
+    },
+    {
+      "epoch": 1.9904,
+      "grad_norm": 59.86637878417969,
+      "learning_rate": 1.7774647887323946e-05,
+      "loss": 0.7514,
+      "step": 6220
+    },
+    {
+      "epoch": 1.9936,
+      "grad_norm": 1.3984020948410034,
+      "learning_rate": 1.771830985915493e-05,
+      "loss": 0.3885,
+      "step": 6230
+    },
+    {
+      "epoch": 1.9968,
+      "grad_norm": 25.51970863342285,
+      "learning_rate": 1.7661971830985917e-05,
+      "loss": 0.1954,
+      "step": 6240
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.26810941100120544,
+      "learning_rate": 1.7605633802816902e-05,
+      "loss": 0.2831,
+      "step": 6250
+    },
+    {
+      "epoch": 2.0032,
+      "grad_norm": 25.82304573059082,
+      "learning_rate": 1.7549295774647888e-05,
+      "loss": 0.1234,
+      "step": 6260
+    },
+    {
+      "epoch": 2.0064,
+      "grad_norm": 0.15285342931747437,
+      "learning_rate": 1.7492957746478873e-05,
+      "loss": 0.1264,
+      "step": 6270
+    },
+    {
+      "epoch": 2.0096,
+      "grad_norm": 0.1930648535490036,
+      "learning_rate": 1.743661971830986e-05,
+      "loss": 0.1633,
+      "step": 6280
+    },
+    {
+      "epoch": 2.0128,
+      "grad_norm": 10.310894966125488,
+      "learning_rate": 1.7380281690140844e-05,
+      "loss": 0.0062,
+      "step": 6290
+    },
+    {
+      "epoch": 2.016,
+      "grad_norm": 0.060901541262865067,
+      "learning_rate": 1.732394366197183e-05,
+      "loss": 0.0814,
+      "step": 6300
+    },
+    {
+      "epoch": 2.0192,
+      "grad_norm": 0.14364077150821686,
+      "learning_rate": 1.7267605633802815e-05,
+      "loss": 0.0068,
+      "step": 6310
+    },
+    {
+      "epoch": 2.0224,
+      "grad_norm": 0.5780632495880127,
+      "learning_rate": 1.7211267605633804e-05,
+      "loss": 0.0471,
+      "step": 6320
+    },
+    {
+      "epoch": 2.0256,
+      "grad_norm": 0.5313758850097656,
+      "learning_rate": 1.715492957746479e-05,
+      "loss": 0.0619,
+      "step": 6330
+    },
+    {
+      "epoch": 2.0288,
+      "grad_norm": 0.028105057775974274,
+      "learning_rate": 1.7098591549295775e-05,
+      "loss": 0.009,
+      "step": 6340
+    },
+    {
+      "epoch": 2.032,
+      "grad_norm": 1.1972317695617676,
+      "learning_rate": 1.704225352112676e-05,
+      "loss": 0.0622,
+      "step": 6350
+    },
+    {
+      "epoch": 2.0352,
+      "grad_norm": 0.027558835223317146,
+      "learning_rate": 1.6985915492957746e-05,
+      "loss": 0.2316,
+      "step": 6360
+    },
+    {
+      "epoch": 2.0384,
+      "grad_norm": 0.04284098371863365,
+      "learning_rate": 1.6929577464788735e-05,
+      "loss": 0.0582,
+      "step": 6370
+    },
+    {
+      "epoch": 2.0416,
+      "grad_norm": 0.1924617737531662,
+      "learning_rate": 1.687323943661972e-05,
+      "loss": 0.0691,
+      "step": 6380
+    },
+    {
+      "epoch": 2.0448,
+      "grad_norm": 0.036435432732105255,
+      "learning_rate": 1.6816901408450706e-05,
+      "loss": 0.1442,
+      "step": 6390
+    },
+    {
+      "epoch": 2.048,
+      "grad_norm": 0.8796645402908325,
+      "learning_rate": 1.676056338028169e-05,
+      "loss": 0.0757,
+      "step": 6400
+    },
+    {
+      "epoch": 2.0512,
+      "grad_norm": 0.6916587352752686,
+      "learning_rate": 1.6704225352112677e-05,
+      "loss": 0.1356,
+      "step": 6410
+    },
+    {
+      "epoch": 2.0544,
+      "grad_norm": 0.10934862494468689,
+      "learning_rate": 1.6647887323943663e-05,
+      "loss": 0.211,
+      "step": 6420
+    },
+    {
+      "epoch": 2.0576,
+      "grad_norm": 0.03238527849316597,
+      "learning_rate": 1.659154929577465e-05,
+      "loss": 0.0556,
+      "step": 6430
+    },
+    {
+      "epoch": 2.0608,
+      "grad_norm": 0.25189611315727234,
+      "learning_rate": 1.6535211267605634e-05,
+      "loss": 0.0208,
+      "step": 6440
+    },
+    {
+      "epoch": 2.064,
+      "grad_norm": 0.08050217479467392,
+      "learning_rate": 1.647887323943662e-05,
+      "loss": 0.0021,
+      "step": 6450
+    },
+    {
+      "epoch": 2.0672,
+      "grad_norm": 0.045152150094509125,
+      "learning_rate": 1.6422535211267605e-05,
+      "loss": 0.0743,
+      "step": 6460
+    },
+    {
+      "epoch": 2.0704,
+      "grad_norm": 0.036941394209861755,
+      "learning_rate": 1.636619718309859e-05,
+      "loss": 0.1484,
+      "step": 6470
+    },
+    {
+      "epoch": 2.0736,
+      "grad_norm": 0.024720242246985435,
+      "learning_rate": 1.6309859154929576e-05,
+      "loss": 0.1229,
+      "step": 6480
+    },
+    {
+      "epoch": 2.0768,
+      "grad_norm": 0.033186838030815125,
+      "learning_rate": 1.6253521126760565e-05,
+      "loss": 0.1691,
+      "step": 6490
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.04443328082561493,
+      "learning_rate": 1.619718309859155e-05,
+      "loss": 0.1783,
+      "step": 6500
+    },
+    {
+      "epoch": 2.0832,
+      "grad_norm": 134.47421264648438,
+      "learning_rate": 1.6140845070422536e-05,
+      "loss": 0.2483,
+      "step": 6510
+    },
+    {
+      "epoch": 2.0864,
+      "grad_norm": 5.727556228637695,
+      "learning_rate": 1.608450704225352e-05,
+      "loss": 0.1534,
+      "step": 6520
+    },
+    {
+      "epoch": 2.0896,
+      "grad_norm": 0.7954875230789185,
+      "learning_rate": 1.6028169014084507e-05,
+      "loss": 0.1282,
+      "step": 6530
+    },
+    {
+      "epoch": 2.0928,
+      "grad_norm": 0.08250103145837784,
+      "learning_rate": 1.5971830985915492e-05,
+      "loss": 0.2734,
+      "step": 6540
+    },
+    {
+      "epoch": 2.096,
+      "grad_norm": 0.04844718798995018,
+      "learning_rate": 1.591549295774648e-05,
+      "loss": 0.0514,
+      "step": 6550
+    },
+    {
+      "epoch": 2.0992,
+      "grad_norm": 0.04677910357713699,
+      "learning_rate": 1.5859154929577466e-05,
+      "loss": 0.0621,
+      "step": 6560
+    },
+    {
+      "epoch": 2.1024,
+      "grad_norm": 0.12014532834291458,
+      "learning_rate": 1.5802816901408452e-05,
+      "loss": 0.0712,
+      "step": 6570
+    },
+    {
+      "epoch": 2.1056,
+      "grad_norm": 0.18135568499565125,
+      "learning_rate": 1.5746478873239437e-05,
+      "loss": 0.0642,
+      "step": 6580
+    },
+    {
+      "epoch": 2.1088,
+      "grad_norm": 0.13500288128852844,
+      "learning_rate": 1.5690140845070423e-05,
+      "loss": 0.0566,
+      "step": 6590
+    },
+    {
+      "epoch": 2.112,
+      "grad_norm": 0.03971581906080246,
+      "learning_rate": 1.5633802816901412e-05,
+      "loss": 0.0039,
+      "step": 6600
+    },
+    {
+      "epoch": 2.1152,
+      "grad_norm": 0.12814994156360626,
+      "learning_rate": 1.5577464788732397e-05,
+      "loss": 0.1774,
+      "step": 6610
+    },
+    {
+      "epoch": 2.1184,
+      "grad_norm": 0.02763848565518856,
+      "learning_rate": 1.5521126760563383e-05,
+      "loss": 0.2017,
+      "step": 6620
+    },
+    {
+      "epoch": 2.1216,
+      "grad_norm": 0.16662102937698364,
+      "learning_rate": 1.546478873239437e-05,
+      "loss": 0.0964,
+      "step": 6630
+    },
+    {
+      "epoch": 2.1248,
+      "grad_norm": 0.0411493182182312,
+      "learning_rate": 1.5408450704225354e-05,
+      "loss": 0.055,
+      "step": 6640
+    },
+    {
+      "epoch": 2.128,
+      "grad_norm": 0.10390494018793106,
+      "learning_rate": 1.535211267605634e-05,
+      "loss": 0.1993,
+      "step": 6650
+    },
+    {
+      "epoch": 2.1312,
+      "grad_norm": 32.90834426879883,
+      "learning_rate": 1.5295774647887325e-05,
+      "loss": 0.2295,
+      "step": 6660
+    },
+    {
+      "epoch": 2.1344,
+      "grad_norm": 0.20629918575286865,
+      "learning_rate": 1.5239436619718312e-05,
+      "loss": 0.1365,
+      "step": 6670
+    },
+    {
+      "epoch": 2.1376,
+      "grad_norm": 0.06436211615800858,
+      "learning_rate": 1.5183098591549298e-05,
+      "loss": 0.2804,
+      "step": 6680
+    },
+    {
+      "epoch": 2.1408,
+      "grad_norm": 6.357541561126709,
+      "learning_rate": 1.5126760563380283e-05,
+      "loss": 0.1547,
+      "step": 6690
+    },
+    {
+      "epoch": 2.144,
+      "grad_norm": 0.031177496537566185,
+      "learning_rate": 1.5070422535211269e-05,
+      "loss": 0.1221,
+      "step": 6700
+    },
+    {
+      "epoch": 2.1471999999999998,
+      "grad_norm": 0.0401877760887146,
+      "learning_rate": 1.5014084507042252e-05,
+      "loss": 0.0467,
+      "step": 6710
+    },
+    {
+      "epoch": 2.1504,
+      "grad_norm": 0.04002746194601059,
+      "learning_rate": 1.4957746478873241e-05,
+      "loss": 0.2081,
+      "step": 6720
+    },
+    {
+      "epoch": 2.1536,
+      "grad_norm": 0.08599916845560074,
+      "learning_rate": 1.4901408450704227e-05,
+      "loss": 0.0094,
+      "step": 6730
+    },
+    {
+      "epoch": 2.1568,
+      "grad_norm": 792.8843994140625,
+      "learning_rate": 1.4845070422535212e-05,
+      "loss": 0.0313,
+      "step": 6740
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 49.27170944213867,
+      "learning_rate": 1.4788732394366198e-05,
+      "loss": 0.0488,
+      "step": 6750
+    },
+    {
+      "epoch": 2.1632,
+      "grad_norm": 0.027708498761057854,
+      "learning_rate": 1.4732394366197183e-05,
+      "loss": 0.2522,
+      "step": 6760
+    },
+    {
+      "epoch": 2.1664,
+      "grad_norm": 144.439697265625,
+      "learning_rate": 1.4676056338028169e-05,
+      "loss": 0.2231,
+      "step": 6770
+    },
+    {
+      "epoch": 2.1696,
+      "grad_norm": 0.05224217101931572,
+      "learning_rate": 1.4619718309859156e-05,
+      "loss": 0.0018,
+      "step": 6780
+    },
+    {
+      "epoch": 2.1728,
+      "grad_norm": 85.55796813964844,
+      "learning_rate": 1.4563380281690141e-05,
+      "loss": 0.218,
+      "step": 6790
+    },
+    {
+      "epoch": 2.176,
+      "grad_norm": 0.17730030417442322,
+      "learning_rate": 1.4507042253521127e-05,
+      "loss": 0.13,
+      "step": 6800
+    },
+    {
+      "epoch": 2.1792,
+      "grad_norm": 0.05483116954565048,
+      "learning_rate": 1.4450704225352112e-05,
+      "loss": 0.0386,
+      "step": 6810
+    },
+    {
+      "epoch": 2.1824,
+      "grad_norm": 0.03330325335264206,
+      "learning_rate": 1.4394366197183098e-05,
+      "loss": 0.0036,
+      "step": 6820
+    },
+    {
+      "epoch": 2.1856,
+      "grad_norm": 0.030421894043684006,
+      "learning_rate": 1.4338028169014083e-05,
+      "loss": 0.1214,
+      "step": 6830
+    },
+    {
+      "epoch": 2.1888,
+      "grad_norm": 0.037813425064086914,
+      "learning_rate": 1.4281690140845072e-05,
+      "loss": 0.002,
+      "step": 6840
+    },
+    {
+      "epoch": 2.192,
+      "grad_norm": 0.4608314335346222,
+      "learning_rate": 1.4225352112676058e-05,
+      "loss": 0.2043,
+      "step": 6850
+    },
+    {
+      "epoch": 2.1952,
+      "grad_norm": 0.16903652250766754,
+      "learning_rate": 1.4169014084507043e-05,
+      "loss": 0.2208,
+      "step": 6860
+    },
+    {
+      "epoch": 2.1984,
+      "grad_norm": 0.09764442592859268,
+      "learning_rate": 1.4112676056338029e-05,
+      "loss": 0.2925,
+      "step": 6870
+    },
+    {
+      "epoch": 2.2016,
+      "grad_norm": 0.04944216087460518,
+      "learning_rate": 1.4056338028169014e-05,
+      "loss": 0.0023,
+      "step": 6880
+    },
+    {
+      "epoch": 2.2048,
+      "grad_norm": 18.86257553100586,
+      "learning_rate": 1.4000000000000001e-05,
+      "loss": 0.1508,
+      "step": 6890
+    },
+    {
+      "epoch": 2.208,
+      "grad_norm": 5.072443962097168,
+      "learning_rate": 1.3943661971830987e-05,
+      "loss": 0.1069,
+      "step": 6900
+    },
+    {
+      "epoch": 2.2112,
+      "grad_norm": 0.12859505414962769,
+      "learning_rate": 1.3887323943661972e-05,
+      "loss": 0.0233,
+      "step": 6910
+    },
+    {
+      "epoch": 2.2144,
+      "grad_norm": 0.06567766517400742,
+      "learning_rate": 1.3830985915492958e-05,
+      "loss": 0.0834,
+      "step": 6920
+    },
+    {
+      "epoch": 2.2176,
+      "grad_norm": 15.95632266998291,
+      "learning_rate": 1.3774647887323943e-05,
+      "loss": 0.3579,
+      "step": 6930
+    },
+    {
+      "epoch": 2.2208,
+      "grad_norm": 0.274181067943573,
+      "learning_rate": 1.3718309859154929e-05,
+      "loss": 0.1515,
+      "step": 6940
+    },
+    {
+      "epoch": 2.224,
+      "grad_norm": 0.13101747632026672,
+      "learning_rate": 1.3661971830985918e-05,
+      "loss": 0.0393,
+      "step": 6950
+    },
+    {
+      "epoch": 2.2272,
+      "grad_norm": 1.139413595199585,
+      "learning_rate": 1.3605633802816903e-05,
+      "loss": 0.1484,
+      "step": 6960
+    },
+    {
+      "epoch": 2.2304,
+      "grad_norm": 0.057852111756801605,
+      "learning_rate": 1.3549295774647889e-05,
+      "loss": 0.0394,
+      "step": 6970
+    },
+    {
+      "epoch": 2.2336,
+      "grad_norm": 0.6658930778503418,
+      "learning_rate": 1.3492957746478874e-05,
+      "loss": 0.0675,
+      "step": 6980
+    },
+    {
+      "epoch": 2.2368,
+      "grad_norm": 0.057538602501153946,
+      "learning_rate": 1.343661971830986e-05,
+      "loss": 0.1998,
+      "step": 6990
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.08062786608934402,
+      "learning_rate": 1.3380281690140845e-05,
+      "loss": 0.1043,
+      "step": 7000
+    },
+    {
+      "epoch": 2.2432,
+      "grad_norm": 12.996604919433594,
+      "learning_rate": 1.3323943661971833e-05,
+      "loss": 0.0718,
+      "step": 7010
+    },
+    {
+      "epoch": 2.2464,
+      "grad_norm": 0.1011863648891449,
+      "learning_rate": 1.3267605633802818e-05,
+      "loss": 0.0694,
+      "step": 7020
+    },
+    {
+      "epoch": 2.2496,
+      "grad_norm": 20.72796058654785,
+      "learning_rate": 1.3211267605633804e-05,
+      "loss": 0.1426,
+      "step": 7030
+    },
+    {
+      "epoch": 2.2528,
+      "grad_norm": 0.22724929451942444,
+      "learning_rate": 1.3154929577464789e-05,
+      "loss": 0.0025,
+      "step": 7040
+    },
+    {
+      "epoch": 2.2560000000000002,
+      "grad_norm": 229.4677734375,
+      "learning_rate": 1.3098591549295775e-05,
+      "loss": 0.2219,
+      "step": 7050
+    },
+    {
+      "epoch": 2.2592,
+      "grad_norm": 0.1337730884552002,
+      "learning_rate": 1.304225352112676e-05,
+      "loss": 0.1146,
+      "step": 7060
+    },
+    {
+      "epoch": 2.2624,
+      "grad_norm": 0.08331338316202164,
+      "learning_rate": 1.2985915492957749e-05,
+      "loss": 0.0336,
+      "step": 7070
+    },
+    {
+      "epoch": 2.2656,
+      "grad_norm": 0.047301776707172394,
+      "learning_rate": 1.2929577464788733e-05,
+      "loss": 0.0719,
+      "step": 7080
+    },
+    {
+      "epoch": 2.2688,
+      "grad_norm": 0.0657852441072464,
+      "learning_rate": 1.2873239436619718e-05,
+      "loss": 0.1904,
+      "step": 7090
+    },
+    {
+      "epoch": 2.2720000000000002,
+      "grad_norm": 0.033138763159513474,
+      "learning_rate": 1.2816901408450704e-05,
+      "loss": 0.0779,
+      "step": 7100
+    },
+    {
+      "epoch": 2.2752,
+      "grad_norm": 37.34537887573242,
+      "learning_rate": 1.276056338028169e-05,
+      "loss": 0.1858,
+      "step": 7110
+    },
+    {
+      "epoch": 2.2784,
+      "grad_norm": 0.10586226731538773,
+      "learning_rate": 1.2704225352112675e-05,
+      "loss": 0.002,
+      "step": 7120
+    },
+    {
+      "epoch": 2.2816,
+      "grad_norm": 0.0424388162791729,
+      "learning_rate": 1.2647887323943664e-05,
+      "loss": 0.1177,
+      "step": 7130
+    },
+    {
+      "epoch": 2.2848,
+      "grad_norm": 0.05596411973237991,
+      "learning_rate": 1.259154929577465e-05,
+      "loss": 0.2422,
+      "step": 7140
+    },
+    {
+      "epoch": 2.288,
+      "grad_norm": 0.05766447260975838,
+      "learning_rate": 1.2535211267605635e-05,
+      "loss": 0.0803,
+      "step": 7150
+    },
+    {
+      "epoch": 2.2912,
+      "grad_norm": 1.7550102472305298,
+      "learning_rate": 1.247887323943662e-05,
+      "loss": 0.1408,
+      "step": 7160
+    },
+    {
+      "epoch": 2.2944,
+      "grad_norm": 0.13066236674785614,
+      "learning_rate": 1.2422535211267607e-05,
+      "loss": 0.002,
+      "step": 7170
+    },
+    {
+      "epoch": 2.2976,
+      "grad_norm": 0.1156509518623352,
+      "learning_rate": 1.2366197183098593e-05,
+      "loss": 0.2522,
+      "step": 7180
+    },
+    {
+      "epoch": 2.3008,
+      "grad_norm": 0.09673482924699783,
+      "learning_rate": 1.2309859154929577e-05,
+      "loss": 0.0857,
+      "step": 7190
+    },
+    {
+      "epoch": 2.304,
+      "grad_norm": 0.8121844530105591,
+      "learning_rate": 1.2253521126760564e-05,
+      "loss": 0.193,
+      "step": 7200
+    },
+    {
+      "epoch": 2.3072,
+      "grad_norm": 0.15363769233226776,
+      "learning_rate": 1.219718309859155e-05,
+      "loss": 0.1038,
+      "step": 7210
+    },
+    {
+      "epoch": 2.3104,
+      "grad_norm": 0.10093524307012558,
+      "learning_rate": 1.2140845070422535e-05,
+      "loss": 0.2331,
+      "step": 7220
+    },
+    {
+      "epoch": 2.3136,
+      "grad_norm": 0.4434497058391571,
+      "learning_rate": 1.2084507042253522e-05,
+      "loss": 0.0804,
+      "step": 7230
+    },
+    {
+      "epoch": 2.3168,
+      "grad_norm": 8.138230323791504,
+      "learning_rate": 1.2028169014084508e-05,
+      "loss": 0.2648,
+      "step": 7240
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 45.24201583862305,
+      "learning_rate": 1.1971830985915493e-05,
+      "loss": 0.2238,
+      "step": 7250
+    },
+    {
+      "epoch": 2.3232,
+      "grad_norm": 0.13272492587566376,
+      "learning_rate": 1.191549295774648e-05,
+      "loss": 0.2196,
+      "step": 7260
+    },
+    {
+      "epoch": 2.3264,
+      "grad_norm": 89.29029846191406,
+      "learning_rate": 1.1859154929577466e-05,
+      "loss": 0.3195,
+      "step": 7270
+    },
+    {
+      "epoch": 2.3296,
+      "grad_norm": 0.09818959981203079,
+      "learning_rate": 1.1802816901408451e-05,
+      "loss": 0.2584,
+      "step": 7280
+    },
+    {
+      "epoch": 2.3327999999999998,
+      "grad_norm": 0.21061986684799194,
+      "learning_rate": 1.1746478873239437e-05,
+      "loss": 0.0191,
+      "step": 7290
+    },
+    {
+      "epoch": 2.336,
+      "grad_norm": 0.12225896865129471,
+      "learning_rate": 1.1690140845070422e-05,
+      "loss": 0.1498,
+      "step": 7300
+    },
+    {
+      "epoch": 2.3392,
+      "grad_norm": 0.10590647161006927,
+      "learning_rate": 1.163380281690141e-05,
+      "loss": 0.0103,
+      "step": 7310
+    },
+    {
+      "epoch": 2.3424,
+      "grad_norm": 0.0610116645693779,
+      "learning_rate": 1.1577464788732395e-05,
+      "loss": 0.1074,
+      "step": 7320
+    },
+    {
+      "epoch": 2.3456,
+      "grad_norm": 0.21642152965068817,
+      "learning_rate": 1.152112676056338e-05,
+      "loss": 0.0459,
+      "step": 7330
+    },
+    {
+      "epoch": 2.3487999999999998,
+      "grad_norm": 0.12459522485733032,
+      "learning_rate": 1.1464788732394368e-05,
+      "loss": 0.0251,
+      "step": 7340
+    },
+    {
+      "epoch": 2.352,
+      "grad_norm": 0.03281530365347862,
+      "learning_rate": 1.1408450704225353e-05,
+      "loss": 0.0669,
+      "step": 7350
+    },
+    {
+      "epoch": 2.3552,
+      "grad_norm": 26.73065757751465,
+      "learning_rate": 1.1352112676056339e-05,
+      "loss": 0.287,
+      "step": 7360
+    },
+    {
+      "epoch": 2.3584,
+      "grad_norm": 55.194541931152344,
+      "learning_rate": 1.1295774647887324e-05,
+      "loss": 0.034,
+      "step": 7370
+    },
+    {
+      "epoch": 2.3616,
+      "grad_norm": 2.0578792095184326,
+      "learning_rate": 1.123943661971831e-05,
+      "loss": 0.0027,
+      "step": 7380
+    },
+    {
+      "epoch": 2.3648,
+      "grad_norm": 4.148108005523682,
+      "learning_rate": 1.1183098591549295e-05,
+      "loss": 0.2128,
+      "step": 7390
+    },
+    {
+      "epoch": 2.368,
+      "grad_norm": 0.11812355369329453,
+      "learning_rate": 1.1126760563380282e-05,
+      "loss": 0.0027,
+      "step": 7400
+    },
+    {
+      "epoch": 2.3712,
+      "grad_norm": 0.03446757793426514,
+      "learning_rate": 1.1070422535211268e-05,
+      "loss": 0.2747,
+      "step": 7410
+    },
+    {
+      "epoch": 2.3744,
+      "grad_norm": 16.427898406982422,
+      "learning_rate": 1.1014084507042253e-05,
+      "loss": 0.0905,
+      "step": 7420
+    },
+    {
+      "epoch": 2.3776,
+      "grad_norm": 0.08660475164651871,
+      "learning_rate": 1.095774647887324e-05,
+      "loss": 0.0019,
+      "step": 7430
+    },
+    {
+      "epoch": 2.3808,
+      "grad_norm": 0.038691841065883636,
+      "learning_rate": 1.0901408450704226e-05,
+      "loss": 0.0015,
+      "step": 7440
+    },
+    {
+      "epoch": 2.384,
+      "grad_norm": 0.037728451192379,
+      "learning_rate": 1.0845070422535212e-05,
+      "loss": 0.0715,
+      "step": 7450
+    },
+    {
+      "epoch": 2.3872,
+      "grad_norm": 0.17045029997825623,
+      "learning_rate": 1.0788732394366199e-05,
+      "loss": 0.1117,
+      "step": 7460
+    },
+    {
+      "epoch": 2.3904,
+      "grad_norm": 0.061354752629995346,
+      "learning_rate": 1.0732394366197184e-05,
+      "loss": 0.1794,
+      "step": 7470
+    },
+    {
+      "epoch": 2.3936,
+      "grad_norm": 0.052094943821430206,
+      "learning_rate": 1.067605633802817e-05,
+      "loss": 0.0585,
+      "step": 7480
+    },
+    {
+      "epoch": 2.3968,
+      "grad_norm": 81.28450012207031,
+      "learning_rate": 1.0619718309859155e-05,
+      "loss": 0.0784,
+      "step": 7490
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 3.190966844558716,
+      "learning_rate": 1.056338028169014e-05,
+      "loss": 0.2101,
+      "step": 7500
+    },
+    {
+      "epoch": 2.4032,
+      "grad_norm": 0.030748562887310982,
+      "learning_rate": 1.0507042253521126e-05,
+      "loss": 0.0671,
+      "step": 7510
+    },
+    {
+      "epoch": 2.4064,
+      "grad_norm": 0.1670941859483719,
+      "learning_rate": 1.0450704225352113e-05,
+      "loss": 0.2274,
+      "step": 7520
+    },
+    {
+      "epoch": 2.4096,
+      "grad_norm": 0.06420325487852097,
+      "learning_rate": 1.0394366197183099e-05,
+      "loss": 0.0918,
+      "step": 7530
+    },
+    {
+      "epoch": 2.4128,
+      "grad_norm": 0.030242426320910454,
+      "learning_rate": 1.0338028169014086e-05,
+      "loss": 0.0561,
+      "step": 7540
+    },
+    {
+      "epoch": 2.416,
+      "grad_norm": 0.0714086964726448,
+      "learning_rate": 1.0281690140845072e-05,
+      "loss": 0.2359,
+      "step": 7550
+    },
+    {
+      "epoch": 2.4192,
+      "grad_norm": 0.17600581049919128,
+      "learning_rate": 1.0225352112676057e-05,
+      "loss": 0.0605,
+      "step": 7560
+    },
+    {
+      "epoch": 2.4224,
+      "grad_norm": 0.03392624855041504,
+      "learning_rate": 1.0169014084507043e-05,
+      "loss": 0.0405,
+      "step": 7570
+    },
+    {
+      "epoch": 2.4256,
+      "grad_norm": 0.02195708081126213,
+      "learning_rate": 1.0112676056338028e-05,
+      "loss": 0.0828,
+      "step": 7580
+    },
+    {
+      "epoch": 2.4288,
+      "grad_norm": 0.08389753103256226,
+      "learning_rate": 1.0056338028169014e-05,
+      "loss": 0.1063,
+      "step": 7590
+    },
+    {
+      "epoch": 2.432,
+      "grad_norm": 4.389575481414795,
+      "learning_rate": 1e-05,
+      "loss": 0.5413,
+      "step": 7600
+    },
+    {
+      "epoch": 2.4352,
+      "grad_norm": 0.2542371153831482,
+      "learning_rate": 9.943661971830986e-06,
+      "loss": 0.0079,
+      "step": 7610
+    },
+    {
+      "epoch": 2.4384,
+      "grad_norm": 0.2106814831495285,
+      "learning_rate": 9.887323943661972e-06,
+      "loss": 0.1727,
+      "step": 7620
+    },
+    {
+      "epoch": 2.4416,
+      "grad_norm": 212.36619567871094,
+      "learning_rate": 9.830985915492959e-06,
+      "loss": 0.0392,
+      "step": 7630
+    },
+    {
+      "epoch": 2.4448,
+      "grad_norm": 80.81587219238281,
+      "learning_rate": 9.774647887323945e-06,
+      "loss": 0.1838,
+      "step": 7640
+    },
+    {
+      "epoch": 2.448,
+      "grad_norm": 23.665437698364258,
+      "learning_rate": 9.71830985915493e-06,
+      "loss": 0.2827,
+      "step": 7650
+    },
+    {
+      "epoch": 2.4512,
+      "grad_norm": 0.037149883806705475,
+      "learning_rate": 9.661971830985917e-06,
+      "loss": 0.0753,
+      "step": 7660
+    },
+    {
+      "epoch": 2.4544,
+      "grad_norm": 0.2889798581600189,
+      "learning_rate": 9.605633802816901e-06,
+      "loss": 0.0021,
+      "step": 7670
+    },
+    {
+      "epoch": 2.4576000000000002,
+      "grad_norm": 39.32601547241211,
+      "learning_rate": 9.549295774647887e-06,
+      "loss": 0.1529,
+      "step": 7680
+    },
+    {
+      "epoch": 2.4608,
+      "grad_norm": 14.48507022857666,
+      "learning_rate": 9.492957746478874e-06,
+      "loss": 0.3042,
+      "step": 7690
+    },
+    {
+      "epoch": 2.464,
+      "grad_norm": 13.561240196228027,
+      "learning_rate": 9.43661971830986e-06,
+      "loss": 0.2097,
+      "step": 7700
+    },
+    {
+      "epoch": 2.4672,
+      "grad_norm": 0.1803174614906311,
+      "learning_rate": 9.380281690140845e-06,
+      "loss": 0.198,
+      "step": 7710
+    },
+    {
+      "epoch": 2.4704,
+      "grad_norm": 0.10523468255996704,
+      "learning_rate": 9.323943661971832e-06,
+      "loss": 0.138,
+      "step": 7720
+    },
+    {
+      "epoch": 2.4736000000000002,
+      "grad_norm": 0.13104431331157684,
+      "learning_rate": 9.267605633802817e-06,
+      "loss": 0.0508,
+      "step": 7730
+    },
+    {
+      "epoch": 2.4768,
+      "grad_norm": 0.6252680420875549,
+      "learning_rate": 9.211267605633803e-06,
+      "loss": 0.0026,
+      "step": 7740
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.3997954726219177,
+      "learning_rate": 9.15492957746479e-06,
+      "loss": 0.0091,
+      "step": 7750
+    },
+    {
+      "epoch": 2.4832,
+      "grad_norm": 0.063034288585186,
+      "learning_rate": 9.098591549295776e-06,
+      "loss": 0.0568,
+      "step": 7760
+    },
+    {
+      "epoch": 2.4864,
+      "grad_norm": 0.26265960931777954,
+      "learning_rate": 9.042253521126761e-06,
+      "loss": 0.1125,
+      "step": 7770
+    },
+    {
+      "epoch": 2.4896,
+      "grad_norm": 16.294443130493164,
+      "learning_rate": 8.985915492957747e-06,
+      "loss": 0.0943,
+      "step": 7780
+    },
+    {
+      "epoch": 2.4928,
+      "grad_norm": 0.042526353150606155,
+      "learning_rate": 8.929577464788732e-06,
+      "loss": 0.0812,
+      "step": 7790
+    },
+    {
+      "epoch": 2.496,
+      "grad_norm": 0.23846685886383057,
+      "learning_rate": 8.87323943661972e-06,
+      "loss": 0.2302,
+      "step": 7800
+    },
+    {
+      "epoch": 2.4992,
+      "grad_norm": 0.08233381807804108,
+      "learning_rate": 8.816901408450705e-06,
+      "loss": 0.2928,
+      "step": 7810
+    },
+    {
+      "epoch": 2.5023999999999997,
+      "grad_norm": 0.319055438041687,
+      "learning_rate": 8.76056338028169e-06,
+      "loss": 0.0741,
+      "step": 7820
+    },
+    {
+      "epoch": 2.5056000000000003,
+      "grad_norm": 4.767480850219727,
+      "learning_rate": 8.704225352112677e-06,
+      "loss": 0.2083,
+      "step": 7830
+    },
+    {
+      "epoch": 2.5088,
+      "grad_norm": 0.0436800941824913,
+      "learning_rate": 8.647887323943663e-06,
+      "loss": 0.084,
+      "step": 7840
+    },
+    {
+      "epoch": 2.512,
+      "grad_norm": 102.92646789550781,
+      "learning_rate": 8.591549295774648e-06,
+      "loss": 0.0866,
+      "step": 7850
+    },
+    {
+      "epoch": 2.5152,
+      "grad_norm": 0.14441460371017456,
+      "learning_rate": 8.535211267605634e-06,
+      "loss": 0.0812,
+      "step": 7860
+    },
+    {
+      "epoch": 2.5183999999999997,
+      "grad_norm": 43.7078971862793,
+      "learning_rate": 8.47887323943662e-06,
+      "loss": 0.186,
+      "step": 7870
+    },
+    {
+      "epoch": 2.5216,
+      "grad_norm": 0.21896377205848694,
+      "learning_rate": 8.422535211267605e-06,
+      "loss": 0.022,
+      "step": 7880
+    },
+    {
+      "epoch": 2.5248,
+      "grad_norm": 0.0571066252887249,
+      "learning_rate": 8.366197183098592e-06,
+      "loss": 0.0039,
+      "step": 7890
+    },
+    {
+      "epoch": 2.528,
+      "grad_norm": 0.02678661048412323,
+      "learning_rate": 8.309859154929578e-06,
+      "loss": 0.0926,
+      "step": 7900
+    },
+    {
+      "epoch": 2.5312,
+      "grad_norm": 11.297701835632324,
+      "learning_rate": 8.253521126760563e-06,
+      "loss": 0.1038,
+      "step": 7910
+    },
+    {
+      "epoch": 2.5343999999999998,
+      "grad_norm": 0.04962446540594101,
+      "learning_rate": 8.19718309859155e-06,
+      "loss": 0.0019,
+      "step": 7920
+    },
+    {
+      "epoch": 2.5376,
+      "grad_norm": 17.65663719177246,
+      "learning_rate": 8.140845070422536e-06,
+      "loss": 0.128,
+      "step": 7930
+    },
+    {
+      "epoch": 2.5408,
+      "grad_norm": 33.21250915527344,
+      "learning_rate": 8.084507042253521e-06,
+      "loss": 0.1849,
+      "step": 7940
+    },
+    {
+      "epoch": 2.544,
+      "grad_norm": 52.65444564819336,
+      "learning_rate": 8.028169014084509e-06,
+      "loss": 0.0707,
+      "step": 7950
+    },
+    {
+      "epoch": 2.5472,
+      "grad_norm": 18.572467803955078,
+      "learning_rate": 7.971830985915494e-06,
+      "loss": 0.1295,
+      "step": 7960
+    },
+    {
+      "epoch": 2.5504,
+      "grad_norm": 0.7327030897140503,
+      "learning_rate": 7.915492957746478e-06,
+      "loss": 0.0859,
+      "step": 7970
+    },
+    {
+      "epoch": 2.5536,
+      "grad_norm": 0.0810910239815712,
+      "learning_rate": 7.859154929577465e-06,
+      "loss": 0.0147,
+      "step": 7980
+    },
+    {
+      "epoch": 2.5568,
+      "grad_norm": 0.1518411636352539,
+      "learning_rate": 7.80281690140845e-06,
+      "loss": 0.0688,
+      "step": 7990
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.07638181000947952,
+      "learning_rate": 7.746478873239436e-06,
+      "loss": 0.2128,
+      "step": 8000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 9375,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4209776885760000.0,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}