diff --git "a/progen2-large/cls/trainer_state.json" "b/progen2-large/cls/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/progen2-large/cls/trainer_state.json"
@@ -0,0 +1,6153 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9912030727295255,
+  "eval_steps": 100,
+  "global_step": 8750,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.13297522068023682,
+      "learning_rate": 0.0001,
+      "loss": 8.7348,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.16867296397686005,
+      "learning_rate": 0.0001,
+      "loss": 8.941,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.16890430450439453,
+      "learning_rate": 0.0001,
+      "loss": 8.88,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.19192273914813995,
+      "learning_rate": 0.0001,
+      "loss": 8.8443,
+      "step": 30
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.23721852898597717,
+      "learning_rate": 0.0001,
+      "loss": 8.8711,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.3551916480064392,
+      "learning_rate": 0.0001,
+      "loss": 8.8673,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.5221909880638123,
+      "learning_rate": 0.0001,
+      "loss": 8.8649,
+      "step": 60
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.8092646598815918,
+      "learning_rate": 0.0001,
+      "loss": 8.7519,
+      "step": 70
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.4652483463287354,
+      "learning_rate": 0.0001,
+      "loss": 8.6941,
+      "step": 80
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.9412227869033813,
+      "learning_rate": 0.0001,
+      "loss": 8.5262,
+      "step": 90
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.83344304561615,
+      "learning_rate": 0.0001,
+      "loss": 8.5971,
+      "step": 100
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.094378709793091,
+      "learning_rate": 0.0001,
+      "loss": 8.4853,
+      "step": 110
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 2.055784225463867,
+      "learning_rate": 0.0001,
+      "loss": 8.5176,
+      "step": 120
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.9372063875198364,
+      "learning_rate": 0.0001,
+      "loss": 8.354,
+      "step": 130
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.129876136779785,
+      "learning_rate": 0.0001,
+      "loss": 8.4071,
+      "step": 140
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.973355770111084,
+      "learning_rate": 0.0001,
+      "loss": 8.3377,
+      "step": 150
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.0081779956817627,
+      "learning_rate": 0.0001,
+      "loss": 8.2952,
+      "step": 160
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.0090904235839844,
+      "learning_rate": 0.0001,
+      "loss": 8.2046,
+      "step": 170
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.922044277191162,
+      "learning_rate": 0.0001,
+      "loss": 8.3795,
+      "step": 180
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.9839919805526733,
+      "learning_rate": 0.0001,
+      "loss": 8.3883,
+      "step": 190
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.92957603931427,
+      "learning_rate": 0.0001,
+      "loss": 8.382,
+      "step": 200
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.0117273330688477,
+      "learning_rate": 0.0001,
+      "loss": 8.2335,
+      "step": 210
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.016472816467285,
+      "learning_rate": 0.0001,
+      "loss": 8.2139,
+      "step": 220
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.095881938934326,
+      "learning_rate": 0.0001,
+      "loss": 8.2523,
+      "step": 230
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.027923107147217,
+      "learning_rate": 0.0001,
+      "loss": 8.2233,
+      "step": 240
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.1539618968963623,
+      "learning_rate": 0.0001,
+      "loss": 8.2608,
+      "step": 250
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.0893354415893555,
+      "learning_rate": 0.0001,
+      "loss": 8.079,
+      "step": 260
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.130445957183838,
+      "learning_rate": 0.0001,
+      "loss": 8.1396,
+      "step": 270
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.1687655448913574,
+      "learning_rate": 0.0001,
+      "loss": 8.2298,
+      "step": 280
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.2181570529937744,
+      "learning_rate": 0.0001,
+      "loss": 8.1175,
+      "step": 290
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.2125351428985596,
+      "learning_rate": 0.0001,
+      "loss": 8.218,
+      "step": 300
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.2724785804748535,
+      "learning_rate": 0.0001,
+      "loss": 8.1262,
+      "step": 310
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.1586649417877197,
+      "learning_rate": 0.0001,
+      "loss": 8.1425,
+      "step": 320
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.3413655757904053,
+      "learning_rate": 0.0001,
+      "loss": 8.2228,
+      "step": 330
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.6243274211883545,
+      "learning_rate": 0.0001,
+      "loss": 8.1692,
+      "step": 340
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.6347498893737793,
+      "learning_rate": 0.0001,
+      "loss": 8.0803,
+      "step": 350
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.480410099029541,
+      "learning_rate": 0.0001,
+      "loss": 7.9656,
+      "step": 360
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.5455801486968994,
+      "learning_rate": 0.0001,
+      "loss": 8.104,
+      "step": 370
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.529554843902588,
+      "learning_rate": 0.0001,
+      "loss": 7.7275,
+      "step": 380
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 2.7491965293884277,
+      "learning_rate": 0.0001,
+      "loss": 8.0563,
+      "step": 390
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.432471990585327,
+      "learning_rate": 0.0001,
+      "loss": 7.8892,
+      "step": 400
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.631291151046753,
+      "learning_rate": 0.0001,
+      "loss": 8.0527,
+      "step": 410
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.9655838012695312,
+      "learning_rate": 0.0001,
+      "loss": 7.9904,
+      "step": 420
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.5145082473754883,
+      "learning_rate": 0.0001,
+      "loss": 8.2301,
+      "step": 430
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.496493101119995,
+      "learning_rate": 0.0001,
+      "loss": 7.9915,
+      "step": 440
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.9066152572631836,
+      "learning_rate": 0.0001,
+      "loss": 7.9225,
+      "step": 450
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.7459330558776855,
+      "learning_rate": 0.0001,
+      "loss": 7.8367,
+      "step": 460
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.8107752799987793,
+      "learning_rate": 0.0001,
+      "loss": 7.7504,
+      "step": 470
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.2546205520629883,
+      "learning_rate": 0.0001,
+      "loss": 7.6948,
+      "step": 480
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.727996826171875,
+      "learning_rate": 0.0001,
+      "loss": 7.8125,
+      "step": 490
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.6076881885528564,
+      "learning_rate": 0.0001,
+      "loss": 7.9309,
+      "step": 500
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.5891363620758057,
+      "learning_rate": 0.0001,
+      "loss": 7.8744,
+      "step": 510
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.8208518028259277,
+      "learning_rate": 0.0001,
+      "loss": 7.9102,
+      "step": 520
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.871469736099243,
+      "learning_rate": 0.0001,
+      "loss": 7.7643,
+      "step": 530
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.9549624919891357,
+      "learning_rate": 0.0001,
+      "loss": 7.7503,
+      "step": 540
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.624119281768799,
+      "learning_rate": 0.0001,
+      "loss": 7.8581,
+      "step": 550
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.895923137664795,
+      "learning_rate": 0.0001,
+      "loss": 7.8364,
+      "step": 560
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 3.53452205657959,
+      "learning_rate": 0.0001,
+      "loss": 7.7302,
+      "step": 570
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.2762277126312256,
+      "learning_rate": 0.0001,
+      "loss": 7.7466,
+      "step": 580
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.998901128768921,
+      "learning_rate": 0.0001,
+      "loss": 7.8137,
+      "step": 590
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.803013324737549,
+      "learning_rate": 0.0001,
+      "loss": 7.7706,
+      "step": 600
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.7393321990966797,
+      "learning_rate": 0.0001,
+      "loss": 7.7073,
+      "step": 610
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.095567464828491,
+      "learning_rate": 0.0001,
+      "loss": 7.7535,
+      "step": 620
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.023447275161743,
+      "learning_rate": 0.0001,
+      "loss": 7.767,
+      "step": 630
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.525261640548706,
+      "learning_rate": 0.0001,
+      "loss": 7.5876,
+      "step": 640
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 4.5561981201171875,
+      "learning_rate": 0.0001,
+      "loss": 7.6317,
+      "step": 650
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 3.14880633354187,
+      "learning_rate": 0.0001,
+      "loss": 7.4775,
+      "step": 660
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.4303388595581055,
+      "learning_rate": 0.0001,
+      "loss": 7.7653,
+      "step": 670
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.303943157196045,
+      "learning_rate": 0.0001,
+      "loss": 7.6533,
+      "step": 680
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.96913480758667,
+      "learning_rate": 0.0001,
+      "loss": 7.6354,
+      "step": 690
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.2788634300231934,
+      "learning_rate": 0.0001,
+      "loss": 7.8131,
+      "step": 700
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.0674147605895996,
+      "learning_rate": 0.0001,
+      "loss": 7.7653,
+      "step": 710
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.29290771484375,
+      "learning_rate": 0.0001,
+      "loss": 7.7904,
+      "step": 720
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.1886162757873535,
+      "learning_rate": 0.0001,
+      "loss": 7.6222,
+      "step": 730
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.1626453399658203,
+      "learning_rate": 0.0001,
+      "loss": 7.572,
+      "step": 740
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 3.583855152130127,
+      "learning_rate": 0.0001,
+      "loss": 7.6754,
+      "step": 750
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.819612503051758,
+      "learning_rate": 0.0001,
+      "loss": 7.7297,
+      "step": 760
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.806347846984863,
+      "learning_rate": 0.0001,
+      "loss": 7.7799,
+      "step": 770
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.675781726837158,
+      "learning_rate": 0.0001,
+      "loss": 7.2738,
+      "step": 780
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.470928192138672,
+      "learning_rate": 0.0001,
+      "loss": 7.5812,
+      "step": 790
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.193761110305786,
+      "learning_rate": 0.0001,
+      "loss": 7.4136,
+      "step": 800
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 4.080562591552734,
+      "learning_rate": 0.0001,
+      "loss": 7.5431,
+      "step": 810
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.260985851287842,
+      "learning_rate": 0.0001,
+      "loss": 7.5088,
+      "step": 820
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.8062641620635986,
+      "learning_rate": 0.0001,
+      "loss": 7.1512,
+      "step": 830
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.346114158630371,
+      "learning_rate": 0.0001,
+      "loss": 7.5035,
+      "step": 840
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.501239061355591,
+      "learning_rate": 0.0001,
+      "loss": 7.5003,
+      "step": 850
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.6613893508911133,
+      "learning_rate": 0.0001,
+      "loss": 7.5195,
+      "step": 860
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.71455454826355,
+      "learning_rate": 0.0001,
+      "loss": 7.2193,
+      "step": 870
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.5135610103607178,
+      "learning_rate": 0.0001,
+      "loss": 7.4728,
+      "step": 880
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.6381916999816895,
+      "learning_rate": 0.0001,
+      "loss": 7.2997,
+      "step": 890
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.9867775440216064,
+      "learning_rate": 0.0001,
+      "loss": 7.4964,
+      "step": 900
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 3.5973315238952637,
+      "learning_rate": 0.0001,
+      "loss": 7.3361,
+      "step": 910
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4.139585494995117,
+      "learning_rate": 0.0001,
+      "loss": 7.6401,
+      "step": 920
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.9018635749816895,
+      "learning_rate": 0.0001,
+      "loss": 7.3081,
+      "step": 930
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.454687595367432,
+      "learning_rate": 0.0001,
+      "loss": 7.6361,
+      "step": 940
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.974452018737793,
+      "learning_rate": 0.0001,
+      "loss": 7.5548,
+      "step": 950
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.2566089630126953,
+      "learning_rate": 0.0001,
+      "loss": 7.2521,
+      "step": 960
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.986081123352051,
+      "learning_rate": 0.0001,
+      "loss": 7.4253,
+      "step": 970
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.088415145874023,
+      "learning_rate": 0.0001,
+      "loss": 7.4853,
+      "step": 980
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.8717310428619385,
+      "learning_rate": 0.0001,
+      "loss": 7.3578,
+      "step": 990
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 4.0846147537231445,
+      "learning_rate": 0.0001,
+      "loss": 7.3095,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 3.92598295211792,
+      "learning_rate": 0.0001,
+      "loss": 7.3403,
+      "step": 1010
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.901190757751465,
+      "learning_rate": 0.0001,
+      "loss": 7.3783,
+      "step": 1020
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.8988521099090576,
+      "learning_rate": 0.0001,
+      "loss": 7.3554,
+      "step": 1030
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.7809934616088867,
+      "learning_rate": 0.0001,
+      "loss": 7.5183,
+      "step": 1040
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.2939453125,
+      "learning_rate": 0.0001,
+      "loss": 7.5551,
+      "step": 1050
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.896609306335449,
+      "learning_rate": 0.0001,
+      "loss": 7.0902,
+      "step": 1060
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.5009660720825195,
+      "learning_rate": 0.0001,
+      "loss": 7.3456,
+      "step": 1070
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 4.345921516418457,
+      "learning_rate": 0.0001,
+      "loss": 7.4538,
+      "step": 1080
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.0528645515441895,
+      "learning_rate": 0.0001,
+      "loss": 7.1884,
+      "step": 1090
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.5573432445526123,
+      "learning_rate": 0.0001,
+      "loss": 7.1504,
+      "step": 1100
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.174637317657471,
+      "learning_rate": 0.0001,
+      "loss": 7.5281,
+      "step": 1110
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.023937225341797,
+      "learning_rate": 0.0001,
+      "loss": 7.243,
+      "step": 1120
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.240757465362549,
+      "learning_rate": 0.0001,
+      "loss": 7.1576,
+      "step": 1130
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.350215911865234,
+      "learning_rate": 0.0001,
+      "loss": 7.4068,
+      "step": 1140
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.183955669403076,
+      "learning_rate": 0.0001,
+      "loss": 7.355,
+      "step": 1150
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 3.797305107116699,
+      "learning_rate": 0.0001,
+      "loss": 7.3564,
+      "step": 1160
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.141573429107666,
+      "learning_rate": 0.0001,
+      "loss": 7.3712,
+      "step": 1170
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.167743682861328,
+      "learning_rate": 0.0001,
+      "loss": 7.2091,
+      "step": 1180
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.940791130065918,
+      "learning_rate": 0.0001,
+      "loss": 7.3867,
+      "step": 1190
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.129706382751465,
+      "learning_rate": 0.0001,
+      "loss": 7.3293,
+      "step": 1200
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.8427541255950928,
+      "learning_rate": 0.0001,
+      "loss": 7.2904,
+      "step": 1210
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.395451068878174,
+      "learning_rate": 0.0001,
+      "loss": 7.3325,
+      "step": 1220
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.780239582061768,
+      "learning_rate": 0.0001,
+      "loss": 7.3046,
+      "step": 1230
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.4168381690979,
+      "learning_rate": 0.0001,
+      "loss": 7.2068,
+      "step": 1240
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.216266632080078,
+      "learning_rate": 0.0001,
+      "loss": 7.3161,
+      "step": 1250
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.174491882324219,
+      "learning_rate": 0.0001,
+      "loss": 7.049,
+      "step": 1260
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.3526835441589355,
+      "learning_rate": 0.0001,
+      "loss": 7.1545,
+      "step": 1270
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.346323013305664,
+      "learning_rate": 0.0001,
+      "loss": 6.9883,
+      "step": 1280
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.120974063873291,
+      "learning_rate": 0.0001,
+      "loss": 7.2313,
+      "step": 1290
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.537820816040039,
+      "learning_rate": 0.0001,
+      "loss": 7.1379,
+      "step": 1300
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.454488754272461,
+      "learning_rate": 0.0001,
+      "loss": 6.9717,
+      "step": 1310
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.880317687988281,
+      "learning_rate": 0.0001,
+      "loss": 6.9814,
+      "step": 1320
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.789674758911133,
+      "learning_rate": 0.0001,
+      "loss": 7.1174,
+      "step": 1330
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.854316711425781,
+      "learning_rate": 0.0001,
+      "loss": 7.0875,
+      "step": 1340
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.365521430969238,
+      "learning_rate": 0.0001,
+      "loss": 7.159,
+      "step": 1350
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 4.074314594268799,
+      "learning_rate": 0.0001,
+      "loss": 7.0675,
+      "step": 1360
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.003274440765381,
+      "learning_rate": 0.0001,
+      "loss": 7.1666,
+      "step": 1370
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.538660049438477,
+      "learning_rate": 0.0001,
+      "loss": 7.1847,
+      "step": 1380
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.598351001739502,
+      "learning_rate": 0.0001,
+      "loss": 7.066,
+      "step": 1390
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.349963188171387,
+      "learning_rate": 0.0001,
+      "loss": 7.087,
+      "step": 1400
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.83491849899292,
+      "learning_rate": 0.0001,
+      "loss": 6.9987,
+      "step": 1410
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.384171009063721,
+      "learning_rate": 0.0001,
+      "loss": 7.103,
+      "step": 1420
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.429465293884277,
+      "learning_rate": 0.0001,
+      "loss": 7.5216,
+      "step": 1430
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.560487270355225,
+      "learning_rate": 0.0001,
+      "loss": 6.9828,
+      "step": 1440
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 4.715222358703613,
+      "learning_rate": 0.0001,
+      "loss": 7.2415,
+      "step": 1450
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.314020156860352,
+      "learning_rate": 0.0001,
+      "loss": 7.0882,
+      "step": 1460
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.258492469787598,
+      "learning_rate": 0.0001,
+      "loss": 6.8067,
+      "step": 1470
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.436995983123779,
+      "learning_rate": 0.0001,
+      "loss": 7.0345,
+      "step": 1480
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.494552135467529,
+      "learning_rate": 0.0001,
+      "loss": 7.0223,
+      "step": 1490
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.3929033279418945,
+      "learning_rate": 0.0001,
+      "loss": 7.2191,
+      "step": 1500
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.227630138397217,
+      "learning_rate": 0.0001,
+      "loss": 7.1492,
+      "step": 1510
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.707615375518799,
+      "learning_rate": 0.0001,
+      "loss": 7.0272,
+      "step": 1520
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.752662658691406,
+      "learning_rate": 0.0001,
+      "loss": 7.1189,
+      "step": 1530
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 4.955765247344971,
+      "learning_rate": 0.0001,
+      "loss": 7.0189,
+      "step": 1540
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.985854148864746,
+      "learning_rate": 0.0001,
+      "loss": 6.9964,
+      "step": 1550
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.682392120361328,
+      "learning_rate": 0.0001,
+      "loss": 7.2356,
+      "step": 1560
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.694766044616699,
+      "learning_rate": 0.0001,
+      "loss": 7.2553,
+      "step": 1570
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.26613712310791,
+      "learning_rate": 0.0001,
+      "loss": 7.0534,
+      "step": 1580
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.842898368835449,
+      "learning_rate": 0.0001,
+      "loss": 6.8348,
+      "step": 1590
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.8272480964660645,
+      "learning_rate": 0.0001,
+      "loss": 7.0437,
+      "step": 1600
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.158578872680664,
+      "learning_rate": 0.0001,
+      "loss": 7.0812,
+      "step": 1610
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 4.662472248077393,
+      "learning_rate": 0.0001,
+      "loss": 6.7437,
+      "step": 1620
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.311957836151123,
+      "learning_rate": 0.0001,
+      "loss": 6.886,
+      "step": 1630
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.214154243469238,
+      "learning_rate": 0.0001,
+      "loss": 6.9165,
+      "step": 1640
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.116697311401367,
+      "learning_rate": 0.0001,
+      "loss": 7.1303,
+      "step": 1650
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.14206600189209,
+      "learning_rate": 0.0001,
+      "loss": 6.6644,
+      "step": 1660
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.349717617034912,
+      "learning_rate": 0.0001,
+      "loss": 6.7909,
+      "step": 1670
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.650986671447754,
+      "learning_rate": 0.0001,
+      "loss": 6.8895,
+      "step": 1680
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.30741024017334,
+      "learning_rate": 0.0001,
+      "loss": 6.8483,
+      "step": 1690
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.038286209106445,
+      "learning_rate": 0.0001,
+      "loss": 6.6968,
+      "step": 1700
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 4.951850891113281,
+      "learning_rate": 0.0001,
+      "loss": 6.9279,
+      "step": 1710
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.617232799530029,
+      "learning_rate": 0.0001,
+      "loss": 6.9467,
+      "step": 1720
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.944021701812744,
+      "learning_rate": 0.0001,
+      "loss": 6.8199,
+      "step": 1730
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.753108501434326,
+      "learning_rate": 0.0001,
+      "loss": 6.6966,
+      "step": 1740
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.107850551605225,
+      "learning_rate": 0.0001,
+      "loss": 6.8714,
+      "step": 1750
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.417325496673584,
+      "learning_rate": 0.0001,
+      "loss": 6.7952,
+      "step": 1760
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.707197189331055,
+      "learning_rate": 0.0001,
+      "loss": 6.9271,
+      "step": 1770
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.911560535430908,
+      "learning_rate": 0.0001,
+      "loss": 6.991,
+      "step": 1780
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 4.661187648773193,
+      "learning_rate": 0.0001,
+      "loss": 6.6251,
+      "step": 1790
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.884848117828369,
+      "learning_rate": 0.0001,
+      "loss": 6.8899,
+      "step": 1800
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.706746578216553,
+      "learning_rate": 0.0001,
+      "loss": 6.4484,
+      "step": 1810
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.115345001220703,
+      "learning_rate": 0.0001,
+      "loss": 6.8768,
+      "step": 1820
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.78771448135376,
+      "learning_rate": 0.0001,
+      "loss": 6.8661,
+      "step": 1830
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.827949523925781,
+      "learning_rate": 0.0001,
+      "loss": 7.0969,
+      "step": 1840
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.353413105010986,
+      "learning_rate": 0.0001,
+      "loss": 6.7536,
+      "step": 1850
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.2046661376953125,
+      "learning_rate": 0.0001,
+      "loss": 6.5023,
+      "step": 1860
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.460099697113037,
+      "learning_rate": 0.0001,
+      "loss": 6.9073,
+      "step": 1870
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.963456630706787,
+      "learning_rate": 0.0001,
+      "loss": 6.7439,
+      "step": 1880
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 4.732865810394287,
+      "learning_rate": 0.0001,
+      "loss": 7.0059,
+      "step": 1890
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.471370220184326,
+      "learning_rate": 0.0001,
+      "loss": 6.7639,
+      "step": 1900
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.5579023361206055,
+      "learning_rate": 0.0001,
+      "loss": 6.4475,
+      "step": 1910
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.0699944496154785,
+      "learning_rate": 0.0001,
+      "loss": 6.8788,
+      "step": 1920
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.1162590980529785,
+      "learning_rate": 0.0001,
+      "loss": 6.8587,
+      "step": 1930
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.13946008682251,
+      "learning_rate": 0.0001,
+      "loss": 6.5843,
+      "step": 1940
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.0145745277404785,
+      "learning_rate": 0.0001,
+      "loss": 6.3852,
+      "step": 1950
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.8105058670043945,
+      "learning_rate": 0.0001,
+      "loss": 6.8755,
+      "step": 1960
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.958199501037598,
+      "learning_rate": 0.0001,
+      "loss": 6.7635,
+      "step": 1970
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 4.654933452606201,
+      "learning_rate": 0.0001,
+      "loss": 6.6911,
+      "step": 1980
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.084805011749268,
+      "learning_rate": 0.0001,
+      "loss": 6.8646,
+      "step": 1990
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.228176116943359,
+      "learning_rate": 0.0001,
+      "loss": 6.7352,
+      "step": 2000
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.022884368896484,
+      "learning_rate": 0.0001,
+      "loss": 6.4302,
+      "step": 2010
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.891600131988525,
+      "learning_rate": 0.0001,
+      "loss": 6.6442,
+      "step": 2020
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.488617897033691,
+      "learning_rate": 0.0001,
+      "loss": 6.7842,
+      "step": 2030
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.55974006652832,
+      "learning_rate": 0.0001,
+      "loss": 6.799,
+      "step": 2040
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 4.813317775726318,
+      "learning_rate": 0.0001,
+      "loss": 6.8495,
+      "step": 2050
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.716929912567139,
+      "learning_rate": 0.0001,
+      "loss": 7.0349,
+      "step": 2060
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.431642532348633,
+      "learning_rate": 0.0001,
+      "loss": 6.6589,
+      "step": 2070
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.8946356773376465,
+      "learning_rate": 0.0001,
+      "loss": 6.7466,
+      "step": 2080
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.600463390350342,
+      "learning_rate": 0.0001,
+      "loss": 6.9385,
+      "step": 2090
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 4.970430850982666,
+      "learning_rate": 0.0001,
+      "loss": 6.734,
+      "step": 2100
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.003845691680908,
+      "learning_rate": 0.0001,
+      "loss": 6.5599,
+      "step": 2110
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.170951843261719,
+      "learning_rate": 0.0001,
+      "loss": 6.6936,
+      "step": 2120
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.419890403747559,
+      "learning_rate": 0.0001,
+      "loss": 6.7239,
+      "step": 2130
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.742493629455566,
+      "learning_rate": 0.0001,
+      "loss": 6.589,
+      "step": 2140
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.559704303741455,
+      "learning_rate": 0.0001,
+      "loss": 6.5758,
+      "step": 2150
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.219899654388428,
+      "learning_rate": 0.0001,
+      "loss": 6.6135,
+      "step": 2160
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.38388729095459,
+      "learning_rate": 0.0001,
+      "loss": 6.701,
+      "step": 2170
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.994665622711182,
+      "learning_rate": 0.0001,
+      "loss": 6.5586,
+      "step": 2180
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.255434989929199,
+      "learning_rate": 0.0001,
+      "loss": 6.2539,
+      "step": 2190
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.021824359893799,
+      "learning_rate": 0.0001,
+      "loss": 6.6047,
+      "step": 2200
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.989778518676758,
+      "learning_rate": 0.0001,
+      "loss": 6.6875,
+      "step": 2210
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.963884353637695,
+      "learning_rate": 0.0001,
+      "loss": 6.5153,
+      "step": 2220
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.188070774078369,
+      "learning_rate": 0.0001,
+      "loss": 6.8159,
+      "step": 2230
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.5286641120910645,
+      "learning_rate": 0.0001,
+      "loss": 6.4946,
+      "step": 2240
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.531112194061279,
+      "learning_rate": 0.0001,
+      "loss": 6.5821,
+      "step": 2250
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.289617538452148,
+      "learning_rate": 0.0001,
+      "loss": 6.2002,
+      "step": 2260
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.274537563323975,
+      "learning_rate": 0.0001,
+      "loss": 6.4972,
+      "step": 2270
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.150925159454346,
+      "learning_rate": 0.0001,
+      "loss": 6.4208,
+      "step": 2280
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.425180435180664,
+      "learning_rate": 0.0001,
+      "loss": 6.5705,
+      "step": 2290
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 4.959611415863037,
+      "learning_rate": 0.0001,
+      "loss": 6.5698,
+      "step": 2300
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.738392353057861,
+      "learning_rate": 0.0001,
+      "loss": 6.8422,
+      "step": 2310
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.691156387329102,
+      "learning_rate": 0.0001,
+      "loss": 6.5405,
+      "step": 2320
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.606560230255127,
+      "learning_rate": 0.0001,
+      "loss": 6.4569,
+      "step": 2330
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.126870632171631,
+      "learning_rate": 0.0001,
+      "loss": 6.6748,
+      "step": 2340
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.772897720336914,
+      "learning_rate": 0.0001,
+      "loss": 6.4001,
+      "step": 2350
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.502315044403076,
+      "learning_rate": 0.0001,
+      "loss": 6.7553,
+      "step": 2360
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.4864959716796875,
+      "learning_rate": 0.0001,
+      "loss": 6.7791,
+      "step": 2370
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.814062595367432,
+      "learning_rate": 0.0001,
+      "loss": 6.5445,
+      "step": 2380
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.651641368865967,
+      "learning_rate": 0.0001,
+      "loss": 6.3983,
+      "step": 2390
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.141276836395264,
+      "learning_rate": 0.0001,
+      "loss": 6.6345,
+      "step": 2400
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.182980060577393,
+      "learning_rate": 0.0001,
+      "loss": 6.4681,
+      "step": 2410
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.8619065284729,
+      "learning_rate": 0.0001,
+      "loss": 6.7078,
+      "step": 2420
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 7.372335910797119,
+      "learning_rate": 0.0001,
+      "loss": 6.6238,
+      "step": 2430
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.698265552520752,
+      "learning_rate": 0.0001,
+      "loss": 6.7048,
+      "step": 2440
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.471320152282715,
+      "learning_rate": 0.0001,
+      "loss": 6.5252,
+      "step": 2450
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.720361232757568,
+      "learning_rate": 0.0001,
+      "loss": 6.5005,
+      "step": 2460
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.11624813079834,
+      "learning_rate": 0.0001,
+      "loss": 6.5593,
+      "step": 2470
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.4028639793396,
+      "learning_rate": 0.0001,
+      "loss": 6.4823,
+      "step": 2480
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.128964900970459,
+      "learning_rate": 0.0001,
+      "loss": 6.4828,
+      "step": 2490
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.734837055206299,
+      "learning_rate": 0.0001,
+      "loss": 6.4272,
+      "step": 2500
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.629927635192871,
+      "learning_rate": 0.0001,
+      "loss": 6.4034,
+      "step": 2510
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.910006523132324,
+      "learning_rate": 0.0001,
+      "loss": 6.1226,
+      "step": 2520
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.990917682647705,
+      "learning_rate": 0.0001,
+      "loss": 6.6915,
+      "step": 2530
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.546036720275879,
+      "learning_rate": 0.0001,
+      "loss": 6.548,
+      "step": 2540
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.954886436462402,
+      "learning_rate": 0.0001,
+      "loss": 6.5445,
+      "step": 2550
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 7.880654335021973,
+      "learning_rate": 0.0001,
+      "loss": 6.3416,
+      "step": 2560
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 7.332695484161377,
+      "learning_rate": 0.0001,
+      "loss": 6.147,
+      "step": 2570
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.7025957107543945,
+      "learning_rate": 0.0001,
+      "loss": 6.5129,
+      "step": 2580
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.3153533935546875,
+      "learning_rate": 0.0001,
+      "loss": 6.4226,
+      "step": 2590
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.144418239593506,
+      "learning_rate": 0.0001,
+      "loss": 6.321,
+      "step": 2600
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.722589015960693,
+      "learning_rate": 0.0001,
+      "loss": 6.5143,
+      "step": 2610
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.166811466217041,
+      "learning_rate": 0.0001,
+      "loss": 6.295,
+      "step": 2620
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.112163543701172,
+      "learning_rate": 0.0001,
+      "loss": 6.4611,
+      "step": 2630
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.27703857421875,
+      "learning_rate": 0.0001,
+      "loss": 6.384,
+      "step": 2640
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.082823753356934,
+      "learning_rate": 0.0001,
+      "loss": 6.7442,
+      "step": 2650
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.528651237487793,
+      "learning_rate": 0.0001,
+      "loss": 6.3007,
+      "step": 2660
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.324984550476074,
+      "learning_rate": 0.0001,
+      "loss": 6.305,
+      "step": 2670
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.834125518798828,
+      "learning_rate": 0.0001,
+      "loss": 6.6715,
+      "step": 2680
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.2475385665893555,
+      "learning_rate": 0.0001,
+      "loss": 6.5477,
+      "step": 2690
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.854740142822266,
+      "learning_rate": 0.0001,
+      "loss": 6.3708,
+      "step": 2700
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.970745086669922,
+      "learning_rate": 0.0001,
+      "loss": 6.3584,
+      "step": 2710
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 7.371762275695801,
+      "learning_rate": 0.0001,
+      "loss": 6.1323,
+      "step": 2720
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.290621757507324,
+      "learning_rate": 0.0001,
+      "loss": 6.2741,
+      "step": 2730
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.945145606994629,
+      "learning_rate": 0.0001,
+      "loss": 6.3291,
+      "step": 2740
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 6.711264133453369,
+      "learning_rate": 0.0001,
+      "loss": 6.1622,
+      "step": 2750
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.326553821563721,
+      "learning_rate": 0.0001,
+      "loss": 6.2276,
+      "step": 2760
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.685893535614014,
+      "learning_rate": 0.0001,
+      "loss": 6.1257,
+      "step": 2770
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 6.339507579803467,
+      "learning_rate": 0.0001,
+      "loss": 6.3386,
+      "step": 2780
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.355484962463379,
+      "learning_rate": 0.0001,
+      "loss": 6.2942,
+      "step": 2790
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.826178073883057,
+      "learning_rate": 0.0001,
+      "loss": 6.2307,
+      "step": 2800
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.861305236816406,
+      "learning_rate": 0.0001,
+      "loss": 6.4896,
+      "step": 2810
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.401525497436523,
+      "learning_rate": 0.0001,
+      "loss": 6.383,
+      "step": 2820
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.831990718841553,
+      "learning_rate": 0.0001,
+      "loss": 6.4586,
+      "step": 2830
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.628566265106201,
+      "learning_rate": 0.0001,
+      "loss": 6.3588,
+      "step": 2840
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.475712776184082,
+      "learning_rate": 0.0001,
+      "loss": 6.3179,
+      "step": 2850
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.779538631439209,
+      "learning_rate": 0.0001,
+      "loss": 6.1582,
+      "step": 2860
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.988759517669678,
+      "learning_rate": 0.0001,
+      "loss": 6.4682,
+      "step": 2870
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.633481502532959,
+      "learning_rate": 0.0001,
+      "loss": 6.1482,
+      "step": 2880
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.781832695007324,
+      "learning_rate": 0.0001,
+      "loss": 6.6348,
+      "step": 2890
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.584262371063232,
+      "learning_rate": 0.0001,
+      "loss": 5.9724,
+      "step": 2900
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.023799896240234,
+      "learning_rate": 0.0001,
+      "loss": 6.0828,
+      "step": 2910
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.146210193634033,
+      "learning_rate": 0.0001,
+      "loss": 6.3467,
+      "step": 2920
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.341283798217773,
+      "learning_rate": 0.0001,
+      "loss": 6.3819,
+      "step": 2930
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.72313928604126,
+      "learning_rate": 0.0001,
+      "loss": 6.1114,
+      "step": 2940
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.976281642913818,
+      "learning_rate": 0.0001,
+      "loss": 6.4512,
+      "step": 2950
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.852048873901367,
+      "learning_rate": 0.0001,
+      "loss": 6.2642,
+      "step": 2960
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.98239278793335,
+      "learning_rate": 0.0001,
+      "loss": 6.5138,
+      "step": 2970
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.225486755371094,
+      "learning_rate": 0.0001,
+      "loss": 6.1986,
+      "step": 2980
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.385969161987305,
+      "learning_rate": 0.0001,
+      "loss": 6.246,
+      "step": 2990
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.716956615447998,
+      "learning_rate": 0.0001,
+      "loss": 6.3674,
+      "step": 3000
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.050891399383545,
+      "learning_rate": 0.0001,
+      "loss": 6.0153,
+      "step": 3010
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.074161529541016,
+      "learning_rate": 0.0001,
+      "loss": 5.9909,
+      "step": 3020
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.282413005828857,
+      "learning_rate": 0.0001,
+      "loss": 6.418,
+      "step": 3030
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.029009819030762,
+      "learning_rate": 0.0001,
+      "loss": 6.1026,
+      "step": 3040
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.274483680725098,
+      "learning_rate": 0.0001,
+      "loss": 6.2792,
+      "step": 3050
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.048450469970703,
+      "learning_rate": 0.0001,
+      "loss": 5.9428,
+      "step": 3060
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.735396862030029,
+      "learning_rate": 0.0001,
+      "loss": 6.1837,
+      "step": 3070
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.277106285095215,
+      "learning_rate": 0.0001,
+      "loss": 6.2583,
+      "step": 3080
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.375043869018555,
+      "learning_rate": 0.0001,
+      "loss": 5.7969,
+      "step": 3090
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.384618282318115,
+      "learning_rate": 0.0001,
+      "loss": 6.4965,
+      "step": 3100
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.7526535987854,
+      "learning_rate": 0.0001,
+      "loss": 6.0606,
+      "step": 3110
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 6.359055519104004,
+      "learning_rate": 0.0001,
+      "loss": 6.1342,
+      "step": 3120
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 7.263937473297119,
+      "learning_rate": 0.0001,
+      "loss": 6.3009,
+      "step": 3130
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.001373767852783,
+      "learning_rate": 0.0001,
+      "loss": 6.1135,
+      "step": 3140
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.04130744934082,
+      "learning_rate": 0.0001,
+      "loss": 6.1888,
+      "step": 3150
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.760219573974609,
+      "learning_rate": 0.0001,
+      "loss": 6.0767,
+      "step": 3160
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.233415603637695,
+      "learning_rate": 0.0001,
+      "loss": 5.9785,
+      "step": 3170
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.705065727233887,
+      "learning_rate": 0.0001,
+      "loss": 6.1772,
+      "step": 3180
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.345217704772949,
+      "learning_rate": 0.0001,
+      "loss": 5.9939,
+      "step": 3190
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.34625244140625,
+      "learning_rate": 0.0001,
+      "loss": 6.1603,
+      "step": 3200
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.51918888092041,
+      "learning_rate": 0.0001,
+      "loss": 6.0723,
+      "step": 3210
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.3529558181762695,
+      "learning_rate": 0.0001,
+      "loss": 6.2052,
+      "step": 3220
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 7.890661716461182,
+      "learning_rate": 0.0001,
+      "loss": 5.9315,
+      "step": 3230
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.765235424041748,
+      "learning_rate": 0.0001,
+      "loss": 6.3738,
+      "step": 3240
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.84202766418457,
+      "learning_rate": 0.0001,
+      "loss": 6.0678,
+      "step": 3250
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.379911422729492,
+      "learning_rate": 0.0001,
+      "loss": 6.1275,
+      "step": 3260
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.704864025115967,
+      "learning_rate": 0.0001,
+      "loss": 6.215,
+      "step": 3270
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.76657772064209,
+      "learning_rate": 0.0001,
+      "loss": 6.0536,
+      "step": 3280
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.307955265045166,
+      "learning_rate": 0.0001,
+      "loss": 6.4189,
+      "step": 3290
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.170482635498047,
+      "learning_rate": 0.0001,
+      "loss": 6.1157,
+      "step": 3300
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.046316623687744,
+      "learning_rate": 0.0001,
+      "loss": 6.4298,
+      "step": 3310
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 6.438562870025635,
+      "learning_rate": 0.0001,
+      "loss": 6.0918,
+      "step": 3320
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 6.435429096221924,
+      "learning_rate": 0.0001,
+      "loss": 6.2071,
+      "step": 3330
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 8.441079139709473,
+      "learning_rate": 0.0001,
+      "loss": 5.8003,
+      "step": 3340
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 6.558011054992676,
+      "learning_rate": 0.0001,
+      "loss": 5.8361,
+      "step": 3350
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 7.0438408851623535,
+      "learning_rate": 0.0001,
+      "loss": 6.3103,
+      "step": 3360
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 6.44441556930542,
+      "learning_rate": 0.0001,
+      "loss": 6.2944,
+      "step": 3370
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 6.227231025695801,
+      "learning_rate": 0.0001,
+      "loss": 6.1181,
+      "step": 3380
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 9.047100067138672,
+      "learning_rate": 0.0001,
+      "loss": 6.0345,
+      "step": 3390
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.155642509460449,
+      "learning_rate": 0.0001,
+      "loss": 6.2486,
+      "step": 3400
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.88898229598999,
+      "learning_rate": 0.0001,
+      "loss": 5.944,
+      "step": 3410
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.548526763916016,
+      "learning_rate": 0.0001,
+      "loss": 5.7387,
+      "step": 3420
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.032217502593994,
+      "learning_rate": 0.0001,
+      "loss": 6.224,
+      "step": 3430
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.76143741607666,
+      "learning_rate": 0.0001,
+      "loss": 6.1826,
+      "step": 3440
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.028382778167725,
+      "learning_rate": 0.0001,
+      "loss": 5.7295,
+      "step": 3450
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.807711601257324,
+      "learning_rate": 0.0001,
+      "loss": 5.8964,
+      "step": 3460
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.4100141525268555,
+      "learning_rate": 0.0001,
+      "loss": 6.1186,
+      "step": 3470
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.271136283874512,
+      "learning_rate": 0.0001,
+      "loss": 6.0327,
+      "step": 3480
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.835449695587158,
+      "learning_rate": 0.0001,
+      "loss": 5.8633,
+      "step": 3490
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.215906620025635,
+      "learning_rate": 0.0001,
+      "loss": 6.222,
+      "step": 3500
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.858363628387451,
+      "learning_rate": 0.0001,
+      "loss": 6.0412,
+      "step": 3510
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.011862277984619,
+      "learning_rate": 0.0001,
+      "loss": 6.1466,
+      "step": 3520
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.894974708557129,
+      "learning_rate": 0.0001,
+      "loss": 5.7592,
+      "step": 3530
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.660064697265625,
+      "learning_rate": 0.0001,
+      "loss": 6.0596,
+      "step": 3540
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.597464084625244,
+      "learning_rate": 0.0001,
+      "loss": 6.1758,
+      "step": 3550
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 7.283897399902344,
+      "learning_rate": 0.0001,
+      "loss": 5.9196,
+      "step": 3560
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.924407482147217,
+      "learning_rate": 0.0001,
+      "loss": 5.9528,
+      "step": 3570
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 6.309545516967773,
+      "learning_rate": 0.0001,
+      "loss": 5.6401,
+      "step": 3580
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 7.5026936531066895,
+      "learning_rate": 0.0001,
+      "loss": 5.9035,
+      "step": 3590
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 6.599335670471191,
+      "learning_rate": 0.0001,
+      "loss": 6.1346,
+      "step": 3600
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 6.5905280113220215,
+      "learning_rate": 0.0001,
+      "loss": 5.8073,
+      "step": 3610
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 7.569963455200195,
+      "learning_rate": 0.0001,
+      "loss": 5.7717,
+      "step": 3620
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 6.435553550720215,
+      "learning_rate": 0.0001,
+      "loss": 6.0474,
+      "step": 3630
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 6.190580368041992,
+      "learning_rate": 0.0001,
+      "loss": 5.7462,
+      "step": 3640
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 7.496687889099121,
+      "learning_rate": 0.0001,
+      "loss": 6.1665,
+      "step": 3650
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 6.533522605895996,
+      "learning_rate": 0.0001,
+      "loss": 5.8899,
+      "step": 3660
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.648139476776123,
+      "learning_rate": 0.0001,
+      "loss": 5.9665,
+      "step": 3670
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.2325005531311035,
+      "learning_rate": 0.0001,
+      "loss": 5.9779,
+      "step": 3680
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.726931095123291,
+      "learning_rate": 0.0001,
+      "loss": 5.6561,
+      "step": 3690
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.116617679595947,
+      "learning_rate": 0.0001,
+      "loss": 6.0688,
+      "step": 3700
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.392375469207764,
+      "learning_rate": 0.0001,
+      "loss": 5.8622,
+      "step": 3710
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.5375261306762695,
+      "learning_rate": 0.0001,
+      "loss": 6.0916,
+      "step": 3720
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.521157264709473,
+      "learning_rate": 0.0001,
+      "loss": 6.2655,
+      "step": 3730
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.832451343536377,
+      "learning_rate": 0.0001,
+      "loss": 6.1409,
+      "step": 3740
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.7112531661987305,
+      "learning_rate": 0.0001,
+      "loss": 5.7163,
+      "step": 3750
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.3935723304748535,
+      "learning_rate": 0.0001,
+      "loss": 6.0775,
+      "step": 3760
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.631297588348389,
+      "learning_rate": 0.0001,
+      "loss": 6.0298,
+      "step": 3770
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.299158096313477,
+      "learning_rate": 0.0001,
+      "loss": 5.5674,
+      "step": 3780
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.034112453460693,
+      "learning_rate": 0.0001,
+      "loss": 5.8177,
+      "step": 3790
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.987868785858154,
+      "learning_rate": 0.0001,
+      "loss": 5.7568,
+      "step": 3800
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 7.073724269866943,
+      "learning_rate": 0.0001,
+      "loss": 5.7159,
+      "step": 3810
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.987926006317139,
+      "learning_rate": 0.0001,
+      "loss": 5.9485,
+      "step": 3820
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.584548473358154,
+      "learning_rate": 0.0001,
+      "loss": 6.0265,
+      "step": 3830
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.466135501861572,
+      "learning_rate": 0.0001,
+      "loss": 6.0759,
+      "step": 3840
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 7.089422225952148,
+      "learning_rate": 0.0001,
+      "loss": 5.5964,
+      "step": 3850
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 7.915441513061523,
+      "learning_rate": 0.0001,
+      "loss": 6.2671,
+      "step": 3860
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.672188758850098,
+      "learning_rate": 0.0001,
+      "loss": 5.799,
+      "step": 3870
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 6.141867160797119,
+      "learning_rate": 0.0001,
+      "loss": 5.7663,
+      "step": 3880
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 6.205798149108887,
+      "learning_rate": 0.0001,
+      "loss": 5.6474,
+      "step": 3890
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 6.506004810333252,
+      "learning_rate": 0.0001,
+      "loss": 6.0889,
+      "step": 3900
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 7.304274559020996,
+      "learning_rate": 0.0001,
+      "loss": 5.636,
+      "step": 3910
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 6.640396595001221,
+      "learning_rate": 0.0001,
+      "loss": 5.5718,
+      "step": 3920
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.762118339538574,
+      "learning_rate": 0.0001,
+      "loss": 5.7206,
+      "step": 3930
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.803691864013672,
+      "learning_rate": 0.0001,
+      "loss": 5.6831,
+      "step": 3940
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 8.11018180847168,
+      "learning_rate": 0.0001,
+      "loss": 6.1698,
+      "step": 3950
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 5.925243377685547,
+      "learning_rate": 0.0001,
+      "loss": 5.6979,
+      "step": 3960
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.497583866119385,
+      "learning_rate": 0.0001,
+      "loss": 5.7107,
+      "step": 3970
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 8.120863914489746,
+      "learning_rate": 0.0001,
+      "loss": 5.5901,
+      "step": 3980
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.712709426879883,
+      "learning_rate": 0.0001,
+      "loss": 5.7548,
+      "step": 3990
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.161996841430664,
+      "learning_rate": 0.0001,
+      "loss": 5.52,
+      "step": 4000
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.447730541229248,
+      "learning_rate": 0.0001,
+      "loss": 5.7965,
+      "step": 4010
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.468249320983887,
+      "learning_rate": 0.0001,
+      "loss": 5.703,
+      "step": 4020
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.79154109954834,
+      "learning_rate": 0.0001,
+      "loss": 6.1059,
+      "step": 4030
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.422266960144043,
+      "learning_rate": 0.0001,
+      "loss": 5.9269,
+      "step": 4040
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.792859077453613,
+      "learning_rate": 0.0001,
+      "loss": 5.9643,
+      "step": 4050
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.374022006988525,
+      "learning_rate": 0.0001,
+      "loss": 5.659,
+      "step": 4060
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.154322147369385,
+      "learning_rate": 0.0001,
+      "loss": 5.5323,
+      "step": 4070
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.72569465637207,
+      "learning_rate": 0.0001,
+      "loss": 5.5219,
+      "step": 4080
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.635045528411865,
+      "learning_rate": 0.0001,
+      "loss": 6.0205,
+      "step": 4090
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.9341936111450195,
+      "learning_rate": 0.0001,
+      "loss": 5.8778,
+      "step": 4100
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 7.054350852966309,
+      "learning_rate": 0.0001,
+      "loss": 5.3647,
+      "step": 4110
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 7.247239589691162,
+      "learning_rate": 0.0001,
+      "loss": 5.6047,
+      "step": 4120
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.3277482986450195,
+      "learning_rate": 0.0001,
+      "loss": 5.5076,
+      "step": 4130
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.428859710693359,
+      "learning_rate": 0.0001,
+      "loss": 5.5991,
+      "step": 4140
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.407506465911865,
+      "learning_rate": 0.0001,
+      "loss": 6.0102,
+      "step": 4150
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.268433094024658,
+      "learning_rate": 0.0001,
+      "loss": 5.6661,
+      "step": 4160
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.008368492126465,
+      "learning_rate": 0.0001,
+      "loss": 5.6884,
+      "step": 4170
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 7.153110027313232,
+      "learning_rate": 0.0001,
+      "loss": 5.5301,
+      "step": 4180
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 7.091159343719482,
+      "learning_rate": 0.0001,
+      "loss": 5.7876,
+      "step": 4190
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.633502960205078,
+      "learning_rate": 0.0001,
+      "loss": 5.6609,
+      "step": 4200
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.497228622436523,
+      "learning_rate": 0.0001,
+      "loss": 5.9563,
+      "step": 4210
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.343845844268799,
+      "learning_rate": 0.0001,
+      "loss": 5.4256,
+      "step": 4220
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 7.208377838134766,
+      "learning_rate": 0.0001,
+      "loss": 5.5382,
+      "step": 4230
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.902278423309326,
+      "learning_rate": 0.0001,
+      "loss": 5.8672,
+      "step": 4240
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.505051612854004,
+      "learning_rate": 0.0001,
+      "loss": 5.5479,
+      "step": 4250
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.518232822418213,
+      "learning_rate": 0.0001,
+      "loss": 5.892,
+      "step": 4260
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 7.474423408508301,
+      "learning_rate": 0.0001,
+      "loss": 6.0323,
+      "step": 4270
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.5289483070373535,
+      "learning_rate": 0.0001,
+      "loss": 5.5301,
+      "step": 4280
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 7.282980918884277,
+      "learning_rate": 0.0001,
+      "loss": 5.6104,
+      "step": 4290
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.84410285949707,
+      "learning_rate": 0.0001,
+      "loss": 5.6046,
+      "step": 4300
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.337357997894287,
+      "learning_rate": 0.0001,
+      "loss": 5.6134,
+      "step": 4310
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.4376540184021,
+      "learning_rate": 0.0001,
+      "loss": 5.472,
+      "step": 4320
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.562799453735352,
+      "learning_rate": 0.0001,
+      "loss": 5.6591,
+      "step": 4330
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.455559253692627,
+      "learning_rate": 0.0001,
+      "loss": 6.0554,
+      "step": 4340
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.826608657836914,
+      "learning_rate": 0.0001,
+      "loss": 5.9704,
+      "step": 4350
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.765090465545654,
+      "learning_rate": 0.0001,
+      "loss": 5.4823,
+      "step": 4360
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.721391677856445,
+      "learning_rate": 0.0001,
+      "loss": 5.7044,
+      "step": 4370
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.924474716186523,
+      "learning_rate": 0.0001,
+      "loss": 5.7179,
+      "step": 4380
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 7.533623695373535,
+      "learning_rate": 0.0001,
+      "loss": 5.6416,
+      "step": 4390
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 7.464598178863525,
+      "learning_rate": 0.0001,
+      "loss": 5.5196,
+      "step": 4400
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 9.54333782196045,
+      "learning_rate": 0.0001,
+      "loss": 5.7953,
+      "step": 4410
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 7.002260208129883,
+      "learning_rate": 0.0001,
+      "loss": 5.5461,
+      "step": 4420
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.3030619621276855,
+      "learning_rate": 0.0001,
+      "loss": 5.6702,
+      "step": 4430
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.416571140289307,
+      "learning_rate": 0.0001,
+      "loss": 5.7862,
+      "step": 4440
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.733397960662842,
+      "learning_rate": 0.0001,
+      "loss": 5.7584,
+      "step": 4450
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.797712802886963,
+      "learning_rate": 0.0001,
+      "loss": 5.9497,
+      "step": 4460
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.799641132354736,
+      "learning_rate": 0.0001,
+      "loss": 5.7458,
+      "step": 4470
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.507017612457275,
+      "learning_rate": 0.0001,
+      "loss": 5.5486,
+      "step": 4480
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.6487555503845215,
+      "learning_rate": 0.0001,
+      "loss": 5.6338,
+      "step": 4490
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.402535438537598,
+      "learning_rate": 0.0001,
+      "loss": 5.6835,
+      "step": 4500
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.486717700958252,
+      "learning_rate": 0.0001,
+      "loss": 5.4333,
+      "step": 4510
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.968420505523682,
+      "learning_rate": 0.0001,
+      "loss": 5.4903,
+      "step": 4520
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 7.403675556182861,
+      "learning_rate": 0.0001,
+      "loss": 5.5241,
+      "step": 4530
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 7.072659969329834,
+      "learning_rate": 0.0001,
+      "loss": 5.6181,
+      "step": 4540
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 7.88189697265625,
+      "learning_rate": 0.0001,
+      "loss": 5.6995,
+      "step": 4550
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 6.864249229431152,
+      "learning_rate": 0.0001,
+      "loss": 5.7266,
+      "step": 4560
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 6.341947078704834,
+      "learning_rate": 0.0001,
+      "loss": 5.6144,
+      "step": 4570
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 6.7726054191589355,
+      "learning_rate": 0.0001,
+      "loss": 5.5888,
+      "step": 4580
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 6.8624982833862305,
+      "learning_rate": 0.0001,
+      "loss": 5.8554,
+      "step": 4590
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 7.2841572761535645,
+      "learning_rate": 0.0001,
+      "loss": 5.4318,
+      "step": 4600
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 6.269191265106201,
+      "learning_rate": 0.0001,
+      "loss": 5.3461,
+      "step": 4610
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 6.2866291999816895,
+      "learning_rate": 0.0001,
+      "loss": 5.7069,
+      "step": 4620
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 6.1209588050842285,
+      "learning_rate": 0.0001,
+      "loss": 5.6412,
+      "step": 4630
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 7.39185905456543,
+      "learning_rate": 0.0001,
+      "loss": 5.6238,
+      "step": 4640
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 8.489421844482422,
+      "learning_rate": 0.0001,
+      "loss": 5.6198,
+      "step": 4650
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 7.254124164581299,
+      "learning_rate": 0.0001,
+      "loss": 5.6361,
+      "step": 4660
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 6.5817060470581055,
+      "learning_rate": 0.0001,
+      "loss": 5.3398,
+      "step": 4670
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 7.115869045257568,
+      "learning_rate": 0.0001,
+      "loss": 5.497,
+      "step": 4680
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 6.503268718719482,
+      "learning_rate": 0.0001,
+      "loss": 5.3715,
+      "step": 4690
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 6.895274639129639,
+      "learning_rate": 0.0001,
+      "loss": 5.7279,
+      "step": 4700
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 7.182039737701416,
+      "learning_rate": 0.0001,
+      "loss": 5.4719,
+      "step": 4710
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 6.981015682220459,
+      "learning_rate": 0.0001,
+      "loss": 5.4246,
+      "step": 4720
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 6.482900142669678,
+      "learning_rate": 0.0001,
+      "loss": 5.4917,
+      "step": 4730
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 6.363368511199951,
+      "learning_rate": 0.0001,
+      "loss": 5.693,
+      "step": 4740
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 7.24528694152832,
+      "learning_rate": 0.0001,
+      "loss": 5.2239,
+      "step": 4750
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 7.49357271194458,
+      "learning_rate": 0.0001,
+      "loss": 5.6384,
+      "step": 4760
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 6.410263538360596,
+      "learning_rate": 0.0001,
+      "loss": 5.4937,
+      "step": 4770
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 6.979201793670654,
+      "learning_rate": 0.0001,
+      "loss": 5.328,
+      "step": 4780
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 6.748431205749512,
+      "learning_rate": 0.0001,
+      "loss": 5.6439,
+      "step": 4790
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 6.251645088195801,
+      "learning_rate": 0.0001,
+      "loss": 5.3381,
+      "step": 4800
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 6.668383598327637,
+      "learning_rate": 0.0001,
+      "loss": 5.5238,
+      "step": 4810
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 8.519705772399902,
+      "learning_rate": 0.0001,
+      "loss": 5.6398,
+      "step": 4820
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 6.976381301879883,
+      "learning_rate": 0.0001,
+      "loss": 5.5514,
+      "step": 4830
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 6.86808967590332,
+      "learning_rate": 0.0001,
+      "loss": 5.5531,
+      "step": 4840
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 6.151678562164307,
+      "learning_rate": 0.0001,
+      "loss": 5.2745,
+      "step": 4850
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 6.982157230377197,
+      "learning_rate": 0.0001,
+      "loss": 5.9195,
+      "step": 4860
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 6.579090118408203,
+      "learning_rate": 0.0001,
+      "loss": 5.289,
+      "step": 4870
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 7.235924243927002,
+      "learning_rate": 0.0001,
+      "loss": 5.6733,
+      "step": 4880
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 6.930136680603027,
+      "learning_rate": 0.0001,
+      "loss": 5.3142,
+      "step": 4890
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 6.419060230255127,
+      "learning_rate": 0.0001,
+      "loss": 5.4393,
+      "step": 4900
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 7.221414566040039,
+      "learning_rate": 0.0001,
+      "loss": 5.5647,
+      "step": 4910
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 8.7325439453125,
+      "learning_rate": 0.0001,
+      "loss": 5.2114,
+      "step": 4920
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 7.386293888092041,
+      "learning_rate": 0.0001,
+      "loss": 5.5191,
+      "step": 4930
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 7.684049606323242,
+      "learning_rate": 0.0001,
+      "loss": 5.3727,
+      "step": 4940
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 6.742431640625,
+      "learning_rate": 0.0001,
+      "loss": 5.2648,
+      "step": 4950
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 7.356019973754883,
+      "learning_rate": 0.0001,
+      "loss": 5.2329,
+      "step": 4960
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 6.576801300048828,
+      "learning_rate": 0.0001,
+      "loss": 5.5637,
+      "step": 4970
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 6.794112205505371,
+      "learning_rate": 0.0001,
+      "loss": 5.3871,
+      "step": 4980
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 6.230078220367432,
+      "learning_rate": 0.0001,
+      "loss": 5.1712,
+      "step": 4990
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 6.648759841918945,
+      "learning_rate": 0.0001,
+      "loss": 5.6136,
+      "step": 5000
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 7.076524257659912,
+      "learning_rate": 0.0001,
+      "loss": 5.5051,
+      "step": 5010
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 6.528186798095703,
+      "learning_rate": 0.0001,
+      "loss": 5.6209,
+      "step": 5020
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 7.032749176025391,
+      "learning_rate": 0.0001,
+      "loss": 5.2655,
+      "step": 5030
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 6.77689790725708,
+      "learning_rate": 0.0001,
+      "loss": 5.1884,
+      "step": 5040
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 6.60637903213501,
+      "learning_rate": 0.0001,
+      "loss": 5.1725,
+      "step": 5050
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 10.403483390808105,
+      "learning_rate": 0.0001,
+      "loss": 5.4331,
+      "step": 5060
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 6.705955505371094,
+      "learning_rate": 0.0001,
+      "loss": 5.2901,
+      "step": 5070
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 6.173685073852539,
+      "learning_rate": 0.0001,
+      "loss": 5.4682,
+      "step": 5080
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 6.7048563957214355,
+      "learning_rate": 0.0001,
+      "loss": 5.6131,
+      "step": 5090
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 7.498973846435547,
+      "learning_rate": 0.0001,
+      "loss": 5.3099,
+      "step": 5100
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 7.012170791625977,
+      "learning_rate": 0.0001,
+      "loss": 5.2665,
+      "step": 5110
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 9.003202438354492,
+      "learning_rate": 0.0001,
+      "loss": 5.373,
+      "step": 5120
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 6.776703834533691,
+      "learning_rate": 0.0001,
+      "loss": 5.4426,
+      "step": 5130
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 5.626941204071045,
+      "learning_rate": 0.0001,
+      "loss": 5.4379,
+      "step": 5140
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 6.805680751800537,
+      "learning_rate": 0.0001,
+      "loss": 5.2413,
+      "step": 5150
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 7.581198692321777,
+      "learning_rate": 0.0001,
+      "loss": 5.1331,
+      "step": 5160
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 6.171557903289795,
+      "learning_rate": 0.0001,
+      "loss": 5.2741,
+      "step": 5170
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 6.99574089050293,
+      "learning_rate": 0.0001,
+      "loss": 5.4297,
+      "step": 5180
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 7.390645980834961,
+      "learning_rate": 0.0001,
+      "loss": 5.4969,
+      "step": 5190
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 6.326279640197754,
+      "learning_rate": 0.0001,
+      "loss": 5.2271,
+      "step": 5200
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 6.351705074310303,
+      "learning_rate": 0.0001,
+      "loss": 5.5723,
+      "step": 5210
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 7.223125457763672,
+      "learning_rate": 0.0001,
+      "loss": 5.394,
+      "step": 5220
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 8.226743698120117,
+      "learning_rate": 0.0001,
+      "loss": 5.2404,
+      "step": 5230
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 6.265896320343018,
+      "learning_rate": 0.0001,
+      "loss": 5.3163,
+      "step": 5240
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 6.862743854522705,
+      "learning_rate": 0.0001,
+      "loss": 5.2028,
+      "step": 5250
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 6.5056843757629395,
+      "learning_rate": 0.0001,
+      "loss": 5.3644,
+      "step": 5260
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 6.918007850646973,
+      "learning_rate": 0.0001,
+      "loss": 5.0995,
+      "step": 5270
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 6.946686744689941,
+      "learning_rate": 0.0001,
+      "loss": 5.251,
+      "step": 5280
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 6.9408721923828125,
+      "learning_rate": 0.0001,
+      "loss": 5.4676,
+      "step": 5290
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 8.541976928710938,
+      "learning_rate": 0.0001,
+      "loss": 5.1055,
+      "step": 5300
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 6.621685981750488,
+      "learning_rate": 0.0001,
+      "loss": 5.2773,
+      "step": 5310
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 7.954686164855957,
+      "learning_rate": 0.0001,
+      "loss": 5.0552,
+      "step": 5320
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 7.047935485839844,
+      "learning_rate": 0.0001,
+      "loss": 5.178,
+      "step": 5330
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 7.0911736488342285,
+      "learning_rate": 0.0001,
+      "loss": 5.4347,
+      "step": 5340
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 7.138049125671387,
+      "learning_rate": 0.0001,
+      "loss": 5.1555,
+      "step": 5350
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 7.027943134307861,
+      "learning_rate": 0.0001,
+      "loss": 5.3039,
+      "step": 5360
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 6.798688888549805,
+      "learning_rate": 0.0001,
+      "loss": 5.5416,
+      "step": 5370
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 7.587523937225342,
+      "learning_rate": 0.0001,
+      "loss": 5.0948,
+      "step": 5380
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 6.437488079071045,
+      "learning_rate": 0.0001,
+      "loss": 5.6189,
+      "step": 5390
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 7.30016565322876,
+      "learning_rate": 0.0001,
+      "loss": 5.6532,
+      "step": 5400
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 7.223166465759277,
+      "learning_rate": 0.0001,
+      "loss": 5.1357,
+      "step": 5410
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 7.237817287445068,
+      "learning_rate": 0.0001,
+      "loss": 5.5414,
+      "step": 5420
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 6.707641124725342,
+      "learning_rate": 0.0001,
+      "loss": 5.3728,
+      "step": 5430
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 6.906318664550781,
+      "learning_rate": 0.0001,
+      "loss": 5.2936,
+      "step": 5440
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 6.379888534545898,
+      "learning_rate": 0.0001,
+      "loss": 5.1335,
+      "step": 5450
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 7.027223587036133,
+      "learning_rate": 0.0001,
+      "loss": 5.4858,
+      "step": 5460
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 6.945804595947266,
+      "learning_rate": 0.0001,
+      "loss": 5.2437,
+      "step": 5470
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 7.221512317657471,
+      "learning_rate": 0.0001,
+      "loss": 5.2099,
+      "step": 5480
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 6.348363399505615,
+      "learning_rate": 0.0001,
+      "loss": 5.4316,
+      "step": 5490
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 6.8490118980407715,
+      "learning_rate": 0.0001,
+      "loss": 5.3492,
+      "step": 5500
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 8.31968879699707,
+      "learning_rate": 0.0001,
+      "loss": 5.6592,
+      "step": 5510
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 7.203205108642578,
+      "learning_rate": 0.0001,
+      "loss": 5.4621,
+      "step": 5520
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 8.014796257019043,
+      "learning_rate": 0.0001,
+      "loss": 5.3714,
+      "step": 5530
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 7.151791572570801,
+      "learning_rate": 0.0001,
+      "loss": 5.3971,
+      "step": 5540
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 7.131547451019287,
+      "learning_rate": 0.0001,
+      "loss": 4.8536,
+      "step": 5550
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 6.682518005371094,
+      "learning_rate": 0.0001,
+      "loss": 5.2076,
+      "step": 5560
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 7.194185733795166,
+      "learning_rate": 0.0001,
+      "loss": 5.5714,
+      "step": 5570
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 7.9541449546813965,
+      "learning_rate": 0.0001,
+      "loss": 5.1713,
+      "step": 5580
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 6.886518955230713,
+      "learning_rate": 0.0001,
+      "loss": 5.1552,
+      "step": 5590
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 6.2614665031433105,
+      "learning_rate": 0.0001,
+      "loss": 5.2702,
+      "step": 5600
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 6.540952682495117,
+      "learning_rate": 0.0001,
+      "loss": 5.1899,
+      "step": 5610
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 6.343952655792236,
+      "learning_rate": 0.0001,
+      "loss": 5.1553,
+      "step": 5620
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 7.184519290924072,
+      "learning_rate": 0.0001,
+      "loss": 5.4724,
+      "step": 5630
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 7.3336334228515625,
+      "learning_rate": 0.0001,
+      "loss": 5.4738,
+      "step": 5640
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 6.974772930145264,
+      "learning_rate": 0.0001,
+      "loss": 5.2551,
+      "step": 5650
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 7.498238563537598,
+      "learning_rate": 0.0001,
+      "loss": 5.1405,
+      "step": 5660
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 6.700224876403809,
+      "learning_rate": 0.0001,
+      "loss": 5.5326,
+      "step": 5670
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 8.83165168762207,
+      "learning_rate": 0.0001,
+      "loss": 5.1342,
+      "step": 5680
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 6.80472993850708,
+      "learning_rate": 0.0001,
+      "loss": 5.2951,
+      "step": 5690
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.722287178039551,
+      "learning_rate": 0.0001,
+      "loss": 5.3762,
+      "step": 5700
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.404741287231445,
+      "learning_rate": 0.0001,
+      "loss": 5.4071,
+      "step": 5710
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.86545991897583,
+      "learning_rate": 0.0001,
+      "loss": 5.0799,
+      "step": 5720
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 7.489872455596924,
+      "learning_rate": 0.0001,
+      "loss": 5.2653,
+      "step": 5730
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 7.100029945373535,
+      "learning_rate": 0.0001,
+      "loss": 5.4452,
+      "step": 5740
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.86623477935791,
+      "learning_rate": 0.0001,
+      "loss": 5.1413,
+      "step": 5750
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.705371379852295,
+      "learning_rate": 0.0001,
+      "loss": 5.126,
+      "step": 5760
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.805570602416992,
+      "learning_rate": 0.0001,
+      "loss": 5.0279,
+      "step": 5770
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.753636360168457,
+      "learning_rate": 0.0001,
+      "loss": 4.9529,
+      "step": 5780
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 6.734082221984863,
+      "learning_rate": 0.0001,
+      "loss": 5.2117,
+      "step": 5790
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 6.719262599945068,
+      "learning_rate": 0.0001,
+      "loss": 5.2086,
+      "step": 5800
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 7.295538425445557,
+      "learning_rate": 0.0001,
+      "loss": 5.2241,
+      "step": 5810
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 7.679185390472412,
+      "learning_rate": 0.0001,
+      "loss": 5.3836,
+      "step": 5820
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 8.083413124084473,
+      "learning_rate": 0.0001,
+      "loss": 5.2678,
+      "step": 5830
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 6.216175079345703,
+      "learning_rate": 0.0001,
+      "loss": 5.1701,
+      "step": 5840
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 7.124741554260254,
+      "learning_rate": 0.0001,
+      "loss": 5.4819,
+      "step": 5850
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 7.185711860656738,
+      "learning_rate": 0.0001,
+      "loss": 5.3135,
+      "step": 5860
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 6.44232702255249,
+      "learning_rate": 0.0001,
+      "loss": 5.0285,
+      "step": 5870
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 6.8775315284729,
+      "learning_rate": 0.0001,
+      "loss": 5.204,
+      "step": 5880
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 6.708162784576416,
+      "learning_rate": 0.0001,
+      "loss": 5.5587,
+      "step": 5890
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 7.007978439331055,
+      "learning_rate": 0.0001,
+      "loss": 5.1236,
+      "step": 5900
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 7.04622745513916,
+      "learning_rate": 0.0001,
+      "loss": 4.7059,
+      "step": 5910
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 6.12811279296875,
+      "learning_rate": 0.0001,
+      "loss": 5.5787,
+      "step": 5920
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 6.867854595184326,
+      "learning_rate": 0.0001,
+      "loss": 5.2176,
+      "step": 5930
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 7.983065605163574,
+      "learning_rate": 0.0001,
+      "loss": 5.0681,
+      "step": 5940
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 6.563666820526123,
+      "learning_rate": 0.0001,
+      "loss": 4.9634,
+      "step": 5950
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 6.634238243103027,
+      "learning_rate": 0.0001,
+      "loss": 5.1227,
+      "step": 5960
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 7.8757758140563965,
+      "learning_rate": 0.0001,
+      "loss": 5.0123,
+      "step": 5970
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 7.366939067840576,
+      "learning_rate": 0.0001,
+      "loss": 5.1309,
+      "step": 5980
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 6.557621002197266,
+      "learning_rate": 0.0001,
+      "loss": 4.8772,
+      "step": 5990
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 6.19675874710083,
+      "learning_rate": 0.0001,
+      "loss": 5.2759,
+      "step": 6000
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 7.6579365730285645,
+      "learning_rate": 0.0001,
+      "loss": 5.1434,
+      "step": 6010
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 7.236229419708252,
+      "learning_rate": 0.0001,
+      "loss": 4.7847,
+      "step": 6020
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 7.326665878295898,
+      "learning_rate": 0.0001,
+      "loss": 5.1641,
+      "step": 6030
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 6.918865203857422,
+      "learning_rate": 0.0001,
+      "loss": 5.0501,
+      "step": 6040
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 6.534049034118652,
+      "learning_rate": 0.0001,
+      "loss": 5.1034,
+      "step": 6050
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 6.996879577636719,
+      "learning_rate": 0.0001,
+      "loss": 5.1989,
+      "step": 6060
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 7.177955627441406,
+      "learning_rate": 0.0001,
+      "loss": 5.3066,
+      "step": 6070
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 7.171303749084473,
+      "learning_rate": 0.0001,
+      "loss": 5.0173,
+      "step": 6080
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 7.790228843688965,
+      "learning_rate": 0.0001,
+      "loss": 5.0558,
+      "step": 6090
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 6.762297630310059,
+      "learning_rate": 0.0001,
+      "loss": 5.1411,
+      "step": 6100
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 6.509268760681152,
+      "learning_rate": 0.0001,
+      "loss": 5.1563,
+      "step": 6110
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 6.736473083496094,
+      "learning_rate": 0.0001,
+      "loss": 5.2034,
+      "step": 6120
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 7.680369853973389,
+      "learning_rate": 0.0001,
+      "loss": 5.2912,
+      "step": 6130
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 6.34407377243042,
+      "learning_rate": 0.0001,
+      "loss": 5.2843,
+      "step": 6140
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 7.142203330993652,
+      "learning_rate": 0.0001,
+      "loss": 5.3214,
+      "step": 6150
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 7.356980323791504,
+      "learning_rate": 0.0001,
+      "loss": 4.9488,
+      "step": 6160
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 8.108223915100098,
+      "learning_rate": 0.0001,
+      "loss": 5.2189,
+      "step": 6170
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 6.899957656860352,
+      "learning_rate": 0.0001,
+      "loss": 4.6238,
+      "step": 6180
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 8.982954978942871,
+      "learning_rate": 0.0001,
+      "loss": 5.3696,
+      "step": 6190
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 7.346027374267578,
+      "learning_rate": 0.0001,
+      "loss": 5.0404,
+      "step": 6200
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 7.053363800048828,
+      "learning_rate": 0.0001,
+      "loss": 5.1258,
+      "step": 6210
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 7.732638835906982,
+      "learning_rate": 0.0001,
+      "loss": 5.3278,
+      "step": 6220
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 6.653524398803711,
+      "learning_rate": 0.0001,
+      "loss": 5.1815,
+      "step": 6230
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 7.062413692474365,
+      "learning_rate": 0.0001,
+      "loss": 5.3047,
+      "step": 6240
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 7.648515701293945,
+      "learning_rate": 0.0001,
+      "loss": 5.0456,
+      "step": 6250
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 9.226594924926758,
+      "learning_rate": 0.0001,
+      "loss": 5.116,
+      "step": 6260
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 7.474520206451416,
+      "learning_rate": 0.0001,
+      "loss": 5.1683,
+      "step": 6270
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 7.2378458976745605,
+      "learning_rate": 0.0001,
+      "loss": 5.1531,
+      "step": 6280
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 7.207047939300537,
+      "learning_rate": 0.0001,
+      "loss": 4.6059,
+      "step": 6290
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 7.035859107971191,
+      "learning_rate": 0.0001,
+      "loss": 4.9462,
+      "step": 6300
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 7.3313493728637695,
+      "learning_rate": 0.0001,
+      "loss": 5.2748,
+      "step": 6310
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 11.886040687561035,
+      "learning_rate": 0.0001,
+      "loss": 4.961,
+      "step": 6320
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 6.866034984588623,
+      "learning_rate": 0.0001,
+      "loss": 5.4463,
+      "step": 6330
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 6.495908737182617,
+      "learning_rate": 0.0001,
+      "loss": 5.1977,
+      "step": 6340
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 8.516640663146973,
+      "learning_rate": 0.0001,
+      "loss": 5.0344,
+      "step": 6350
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 7.495053291320801,
+      "learning_rate": 0.0001,
+      "loss": 5.0095,
+      "step": 6360
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 7.087994575500488,
+      "learning_rate": 0.0001,
+      "loss": 4.9734,
+      "step": 6370
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 8.174721717834473,
+      "learning_rate": 0.0001,
+      "loss": 5.0688,
+      "step": 6380
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 7.081657886505127,
+      "learning_rate": 0.0001,
+      "loss": 4.8497,
+      "step": 6390
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 7.30198335647583,
+      "learning_rate": 0.0001,
+      "loss": 5.0896,
+      "step": 6400
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 6.629162311553955,
+      "learning_rate": 0.0001,
+      "loss": 5.1279,
+      "step": 6410
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 7.20672082901001,
+      "learning_rate": 0.0001,
+      "loss": 5.1451,
+      "step": 6420
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 7.274264335632324,
+      "learning_rate": 0.0001,
+      "loss": 5.1337,
+      "step": 6430
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 7.5309929847717285,
+      "learning_rate": 0.0001,
+      "loss": 4.853,
+      "step": 6440
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 7.425622940063477,
+      "learning_rate": 0.0001,
+      "loss": 4.9372,
+      "step": 6450
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 6.842046737670898,
+      "learning_rate": 0.0001,
+      "loss": 4.6174,
+      "step": 6460
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 6.7330098152160645,
+      "learning_rate": 0.0001,
+      "loss": 4.9141,
+      "step": 6470
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 8.499651908874512,
+      "learning_rate": 0.0001,
+      "loss": 5.0174,
+      "step": 6480
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 6.639512538909912,
+      "learning_rate": 0.0001,
+      "loss": 5.0263,
+      "step": 6490
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 6.421000957489014,
+      "learning_rate": 0.0001,
+      "loss": 5.1396,
+      "step": 6500
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 6.731082439422607,
+      "learning_rate": 0.0001,
+      "loss": 5.0498,
+      "step": 6510
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 7.022935390472412,
+      "learning_rate": 0.0001,
+      "loss": 5.1146,
+      "step": 6520
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 7.430096626281738,
+      "learning_rate": 0.0001,
+      "loss": 4.7408,
+      "step": 6530
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 8.261208534240723,
+      "learning_rate": 0.0001,
+      "loss": 5.088,
+      "step": 6540
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 5.971454620361328,
+      "learning_rate": 0.0001,
+      "loss": 5.2964,
+      "step": 6550
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 9.39877700805664,
+      "learning_rate": 0.0001,
+      "loss": 5.0784,
+      "step": 6560
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 8.58777904510498,
+      "learning_rate": 0.0001,
+      "loss": 5.4992,
+      "step": 6570
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 7.111948013305664,
+      "learning_rate": 0.0001,
+      "loss": 4.757,
+      "step": 6580
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 7.673144340515137,
+      "learning_rate": 0.0001,
+      "loss": 4.9628,
+      "step": 6590
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 6.778475761413574,
+      "learning_rate": 0.0001,
+      "loss": 4.8403,
+      "step": 6600
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 7.993399620056152,
+      "learning_rate": 0.0001,
+      "loss": 4.7514,
+      "step": 6610
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 7.678924083709717,
+      "learning_rate": 0.0001,
+      "loss": 5.0417,
+      "step": 6620
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 7.200578689575195,
+      "learning_rate": 0.0001,
+      "loss": 5.1593,
+      "step": 6630
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 7.279279708862305,
+      "learning_rate": 0.0001,
+      "loss": 4.8801,
+      "step": 6640
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 6.850567817687988,
+      "learning_rate": 0.0001,
+      "loss": 5.0193,
+      "step": 6650
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 7.572697639465332,
+      "learning_rate": 0.0001,
+      "loss": 4.927,
+      "step": 6660
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 6.584297180175781,
+      "learning_rate": 0.0001,
+      "loss": 5.101,
+      "step": 6670
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 7.785579204559326,
+      "learning_rate": 0.0001,
+      "loss": 4.9634,
+      "step": 6680
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 7.837070465087891,
+      "learning_rate": 0.0001,
+      "loss": 4.9013,
+      "step": 6690
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 6.908084869384766,
+      "learning_rate": 0.0001,
+      "loss": 5.1681,
+      "step": 6700
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 7.562796115875244,
+      "learning_rate": 0.0001,
+      "loss": 5.0718,
+      "step": 6710
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 7.111804008483887,
+      "learning_rate": 0.0001,
+      "loss": 4.908,
+      "step": 6720
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 6.899424076080322,
+      "learning_rate": 0.0001,
+      "loss": 4.8494,
+      "step": 6730
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 7.4407734870910645,
+      "learning_rate": 0.0001,
+      "loss": 4.9015,
+      "step": 6740
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 7.013999938964844,
+      "learning_rate": 0.0001,
+      "loss": 4.936,
+      "step": 6750
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 7.256265163421631,
+      "learning_rate": 0.0001,
+      "loss": 8.8813,
+      "step": 6760
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 7.74426794052124,
+      "learning_rate": 0.0001,
+      "loss": 8.789,
+      "step": 6770
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 7.401425361633301,
+      "learning_rate": 0.0001,
+      "loss": 8.6391,
+      "step": 6780
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 7.382867336273193,
+      "learning_rate": 0.0001,
+      "loss": 8.5164,
+      "step": 6790
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 7.277320384979248,
+      "learning_rate": 0.0001,
+      "loss": 8.3517,
+      "step": 6800
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 7.002625942230225,
+      "learning_rate": 0.0001,
+      "loss": 8.2622,
+      "step": 6810
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 6.867417335510254,
+      "learning_rate": 0.0001,
+      "loss": 8.112,
+      "step": 6820
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 6.698737144470215,
+      "learning_rate": 0.0001,
+      "loss": 7.9223,
+      "step": 6830
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 6.709685802459717,
+      "learning_rate": 0.0001,
+      "loss": 7.8155,
+      "step": 6840
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 7.325059413909912,
+      "learning_rate": 0.0001,
+      "loss": 8.0405,
+      "step": 6850
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 7.344608306884766,
+      "learning_rate": 0.0001,
+      "loss": 7.9361,
+      "step": 6860
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 6.841386795043945,
+      "learning_rate": 0.0001,
+      "loss": 7.8364,
+      "step": 6870
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 7.001560211181641,
+      "learning_rate": 0.0001,
+      "loss": 7.8292,
+      "step": 6880
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 6.741129398345947,
+      "learning_rate": 0.0001,
+      "loss": 7.8206,
+      "step": 6890
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 6.907050132751465,
+      "learning_rate": 0.0001,
+      "loss": 7.6831,
+      "step": 6900
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 6.826385498046875,
+      "learning_rate": 0.0001,
+      "loss": 7.6411,
+      "step": 6910
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 6.961606025695801,
+      "learning_rate": 0.0001,
+      "loss": 7.3522,
+      "step": 6920
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 6.590288162231445,
+      "learning_rate": 0.0001,
+      "loss": 7.5391,
+      "step": 6930
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 7.105891704559326,
+      "learning_rate": 0.0001,
+      "loss": 7.3581,
+      "step": 6940
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 6.910645484924316,
+      "learning_rate": 0.0001,
+      "loss": 7.3498,
+      "step": 6950
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 6.971954345703125,
+      "learning_rate": 0.0001,
+      "loss": 7.2248,
+      "step": 6960
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 7.983415603637695,
+      "learning_rate": 0.0001,
+      "loss": 7.4894,
+      "step": 6970
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 8.602989196777344,
+      "learning_rate": 0.0001,
+      "loss": 7.0532,
+      "step": 6980
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 6.9297661781311035,
+      "learning_rate": 0.0001,
+      "loss": 7.231,
+      "step": 6990
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 8.00120735168457,
+      "learning_rate": 0.0001,
+      "loss": 7.0867,
+      "step": 7000
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 6.988987922668457,
+      "learning_rate": 0.0001,
+      "loss": 7.2018,
+      "step": 7010
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 6.845702648162842,
+      "learning_rate": 0.0001,
+      "loss": 7.0467,
+      "step": 7020
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 6.841978073120117,
+      "learning_rate": 0.0001,
+      "loss": 7.0383,
+      "step": 7030
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 6.429774761199951,
+      "learning_rate": 0.0001,
+      "loss": 6.9796,
+      "step": 7040
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 7.009889602661133,
+      "learning_rate": 0.0001,
+      "loss": 7.012,
+      "step": 7050
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 7.076528549194336,
+      "learning_rate": 0.0001,
+      "loss": 7.0587,
+      "step": 7060
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 7.102022647857666,
+      "learning_rate": 0.0001,
+      "loss": 7.242,
+      "step": 7070
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 7.024888515472412,
+      "learning_rate": 0.0001,
+      "loss": 6.9737,
+      "step": 7080
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 7.168479919433594,
+      "learning_rate": 0.0001,
+      "loss": 6.8972,
+      "step": 7090
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 7.698929786682129,
+      "learning_rate": 0.0001,
+      "loss": 6.9607,
+      "step": 7100
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 6.60235595703125,
+      "learning_rate": 0.0001,
+      "loss": 6.9548,
+      "step": 7110
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 7.053347110748291,
+      "learning_rate": 0.0001,
+      "loss": 6.8202,
+      "step": 7120
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 7.509400367736816,
+      "learning_rate": 0.0001,
+      "loss": 7.0308,
+      "step": 7130
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 6.726016521453857,
+      "learning_rate": 0.0001,
+      "loss": 7.086,
+      "step": 7140
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 6.476027011871338,
+      "learning_rate": 0.0001,
+      "loss": 6.7349,
+      "step": 7150
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 7.360076427459717,
+      "learning_rate": 0.0001,
+      "loss": 6.8585,
+      "step": 7160
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 6.807623863220215,
+      "learning_rate": 0.0001,
+      "loss": 6.9266,
+      "step": 7170
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 6.753279209136963,
+      "learning_rate": 0.0001,
+      "loss": 6.8999,
+      "step": 7180
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 6.772765636444092,
+      "learning_rate": 0.0001,
+      "loss": 6.7618,
+      "step": 7190
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 7.3718390464782715,
+      "learning_rate": 0.0001,
+      "loss": 6.8284,
+      "step": 7200
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 7.098963260650635,
+      "learning_rate": 0.0001,
+      "loss": 6.5729,
+      "step": 7210
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 7.068120002746582,
+      "learning_rate": 0.0001,
+      "loss": 6.665,
+      "step": 7220
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.719849109649658,
+      "learning_rate": 0.0001,
+      "loss": 6.4873,
+      "step": 7230
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.053072929382324,
+      "learning_rate": 0.0001,
+      "loss": 6.5232,
+      "step": 7240
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 7.5951995849609375,
+      "learning_rate": 0.0001,
+      "loss": 6.5575,
+      "step": 7250
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.817147731781006,
+      "learning_rate": 0.0001,
+      "loss": 6.6633,
+      "step": 7260
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.7283525466918945,
+      "learning_rate": 0.0001,
+      "loss": 6.4893,
+      "step": 7270
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.114947319030762,
+      "learning_rate": 0.0001,
+      "loss": 6.6407,
+      "step": 7280
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 6.183058261871338,
+      "learning_rate": 0.0001,
+      "loss": 6.4875,
+      "step": 7290
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 6.9300079345703125,
+      "learning_rate": 0.0001,
+      "loss": 6.7692,
+      "step": 7300
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 7.071079730987549,
+      "learning_rate": 0.0001,
+      "loss": 6.555,
+      "step": 7310
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 6.521429538726807,
+      "learning_rate": 0.0001,
+      "loss": 6.5556,
+      "step": 7320
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 7.271949291229248,
+      "learning_rate": 0.0001,
+      "loss": 6.5699,
+      "step": 7330
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 6.988924980163574,
+      "learning_rate": 0.0001,
+      "loss": 6.6741,
+      "step": 7340
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 7.370090007781982,
+      "learning_rate": 0.0001,
+      "loss": 6.7025,
+      "step": 7350
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 6.888191223144531,
+      "learning_rate": 0.0001,
+      "loss": 6.5708,
+      "step": 7360
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 8.086533546447754,
+      "learning_rate": 0.0001,
+      "loss": 6.6636,
+      "step": 7370
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 7.17087459564209,
+      "learning_rate": 0.0001,
+      "loss": 6.5017,
+      "step": 7380
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 7.062439441680908,
+      "learning_rate": 0.0001,
+      "loss": 6.5913,
+      "step": 7390
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 7.160670280456543,
+      "learning_rate": 0.0001,
+      "loss": 6.3069,
+      "step": 7400
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 7.2284064292907715,
+      "learning_rate": 0.0001,
+      "loss": 6.454,
+      "step": 7410
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 7.462839126586914,
+      "learning_rate": 0.0001,
+      "loss": 6.477,
+      "step": 7420
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 7.273873805999756,
+      "learning_rate": 0.0001,
+      "loss": 6.4675,
+      "step": 7430
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 6.605666637420654,
+      "learning_rate": 0.0001,
+      "loss": 6.2261,
+      "step": 7440
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 6.699402332305908,
+      "learning_rate": 0.0001,
+      "loss": 6.3782,
+      "step": 7450
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 6.655623435974121,
+      "learning_rate": 0.0001,
+      "loss": 6.2115,
+      "step": 7460
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 7.639959812164307,
+      "learning_rate": 0.0001,
+      "loss": 6.354,
+      "step": 7470
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 7.908720016479492,
+      "learning_rate": 0.0001,
+      "loss": 6.1482,
+      "step": 7480
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 6.721646308898926,
+      "learning_rate": 0.0001,
+      "loss": 6.1049,
+      "step": 7490
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 7.452669143676758,
+      "learning_rate": 0.0001,
+      "loss": 6.1361,
+      "step": 7500
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 6.34222412109375,
+      "learning_rate": 0.0001,
+      "loss": 6.1876,
+      "step": 7510
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 6.90377950668335,
+      "learning_rate": 0.0001,
+      "loss": 6.1359,
+      "step": 7520
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 6.4843831062316895,
+      "learning_rate": 0.0001,
+      "loss": 6.3015,
+      "step": 7530
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 6.968750476837158,
+      "learning_rate": 0.0001,
+      "loss": 6.214,
+      "step": 7540
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 7.655303478240967,
+      "learning_rate": 0.0001,
+      "loss": 6.3313,
+      "step": 7550
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 7.0183329582214355,
+      "learning_rate": 0.0001,
+      "loss": 6.2484,
+      "step": 7560
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 7.4320244789123535,
+      "learning_rate": 0.0001,
+      "loss": 6.0566,
+      "step": 7570
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 7.375026226043701,
+      "learning_rate": 0.0001,
+      "loss": 6.1668,
+      "step": 7580
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 6.809161186218262,
+      "learning_rate": 0.0001,
+      "loss": 5.8226,
+      "step": 7590
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 7.361865520477295,
+      "learning_rate": 0.0001,
+      "loss": 6.3405,
+      "step": 7600
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 6.975151062011719,
+      "learning_rate": 0.0001,
+      "loss": 6.0716,
+      "step": 7610
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 6.741641521453857,
+      "learning_rate": 0.0001,
+      "loss": 6.4173,
+      "step": 7620
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 7.211063861846924,
+      "learning_rate": 0.0001,
+      "loss": 6.2764,
+      "step": 7630
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 6.990452289581299,
+      "learning_rate": 0.0001,
+      "loss": 6.1399,
+      "step": 7640
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 7.010167121887207,
+      "learning_rate": 0.0001,
+      "loss": 6.2753,
+      "step": 7650
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 7.327430725097656,
+      "learning_rate": 0.0001,
+      "loss": 6.0911,
+      "step": 7660
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 7.55251932144165,
+      "learning_rate": 0.0001,
+      "loss": 5.9088,
+      "step": 7670
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 7.749558925628662,
+      "learning_rate": 0.0001,
+      "loss": 6.0838,
+      "step": 7680
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 8.06988525390625,
+      "learning_rate": 0.0001,
+      "loss": 6.2276,
+      "step": 7690
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 7.404271602630615,
+      "learning_rate": 0.0001,
+      "loss": 6.0319,
+      "step": 7700
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 7.118124485015869,
+      "learning_rate": 0.0001,
+      "loss": 6.2328,
+      "step": 7710
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 5.857346057891846,
+      "learning_rate": 0.0001,
+      "loss": 6.1652,
+      "step": 7720
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.439952850341797,
+      "learning_rate": 0.0001,
+      "loss": 5.7766,
+      "step": 7730
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 7.161487102508545,
+      "learning_rate": 0.0001,
+      "loss": 5.9695,
+      "step": 7740
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 8.715397834777832,
+      "learning_rate": 0.0001,
+      "loss": 6.215,
+      "step": 7750
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.895458698272705,
+      "learning_rate": 0.0001,
+      "loss": 6.2879,
+      "step": 7760
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.521083354949951,
+      "learning_rate": 0.0001,
+      "loss": 6.2056,
+      "step": 7770
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.712365627288818,
+      "learning_rate": 0.0001,
+      "loss": 6.1096,
+      "step": 7780
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 7.358547687530518,
+      "learning_rate": 0.0001,
+      "loss": 5.8294,
+      "step": 7790
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.958637714385986,
+      "learning_rate": 0.0001,
+      "loss": 6.0671,
+      "step": 7800
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.4223527908325195,
+      "learning_rate": 0.0001,
+      "loss": 6.0847,
+      "step": 7810
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 7.301987171173096,
+      "learning_rate": 0.0001,
+      "loss": 5.9785,
+      "step": 7820
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 7.92888069152832,
+      "learning_rate": 0.0001,
+      "loss": 6.192,
+      "step": 7830
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 8.519168853759766,
+      "learning_rate": 0.0001,
+      "loss": 6.1531,
+      "step": 7840
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 6.962387561798096,
+      "learning_rate": 0.0001,
+      "loss": 6.0309,
+      "step": 7850
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 9.192756652832031,
+      "learning_rate": 0.0001,
+      "loss": 6.1146,
+      "step": 7860
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 7.554586410522461,
+      "learning_rate": 0.0001,
+      "loss": 5.9243,
+      "step": 7870
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 6.524350643157959,
+      "learning_rate": 0.0001,
+      "loss": 5.7512,
+      "step": 7880
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 6.445712089538574,
+      "learning_rate": 0.0001,
+      "loss": 5.7798,
+      "step": 7890
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 7.094208240509033,
+      "learning_rate": 0.0001,
+      "loss": 6.0569,
+      "step": 7900
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 8.0321683883667,
+      "learning_rate": 0.0001,
+      "loss": 5.9737,
+      "step": 7910
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 7.475709438323975,
+      "learning_rate": 0.0001,
+      "loss": 5.938,
+      "step": 7920
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 7.048959255218506,
+      "learning_rate": 0.0001,
+      "loss": 6.0372,
+      "step": 7930
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 7.509370803833008,
+      "learning_rate": 0.0001,
+      "loss": 6.1764,
+      "step": 7940
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 7.586709499359131,
+      "learning_rate": 0.0001,
+      "loss": 5.9189,
+      "step": 7950
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 6.576029300689697,
+      "learning_rate": 0.0001,
+      "loss": 5.9605,
+      "step": 7960
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 6.362544536590576,
+      "learning_rate": 0.0001,
+      "loss": 6.0666,
+      "step": 7970
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 9.308067321777344,
+      "learning_rate": 0.0001,
+      "loss": 6.0342,
+      "step": 7980
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 7.298825263977051,
+      "learning_rate": 0.0001,
+      "loss": 6.0839,
+      "step": 7990
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 7.239445686340332,
+      "learning_rate": 0.0001,
+      "loss": 6.0886,
+      "step": 8000
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 7.520422458648682,
+      "learning_rate": 0.0001,
+      "loss": 6.0614,
+      "step": 8010
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 6.770960807800293,
+      "learning_rate": 0.0001,
+      "loss": 6.0054,
+      "step": 8020
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 7.698451995849609,
+      "learning_rate": 0.0001,
+      "loss": 5.9751,
+      "step": 8030
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 7.468120098114014,
+      "learning_rate": 0.0001,
+      "loss": 5.4905,
+      "step": 8040
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 7.01066255569458,
+      "learning_rate": 0.0001,
+      "loss": 5.9105,
+      "step": 8050
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 7.036108016967773,
+      "learning_rate": 0.0001,
+      "loss": 5.8238,
+      "step": 8060
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 7.004483699798584,
+      "learning_rate": 0.0001,
+      "loss": 6.1893,
+      "step": 8070
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 7.315797328948975,
+      "learning_rate": 0.0001,
+      "loss": 5.6716,
+      "step": 8080
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 6.802061557769775,
+      "learning_rate": 0.0001,
+      "loss": 5.7308,
+      "step": 8090
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 6.446535110473633,
+      "learning_rate": 0.0001,
+      "loss": 6.0949,
+      "step": 8100
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 7.042125701904297,
+      "learning_rate": 0.0001,
+      "loss": 5.6031,
+      "step": 8110
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 6.511283874511719,
+      "learning_rate": 0.0001,
+      "loss": 5.8619,
+      "step": 8120
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 7.169569969177246,
+      "learning_rate": 0.0001,
+      "loss": 5.478,
+      "step": 8130
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 6.882211208343506,
+      "learning_rate": 0.0001,
+      "loss": 5.9039,
+      "step": 8140
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 8.79712963104248,
+      "learning_rate": 0.0001,
+      "loss": 5.8303,
+      "step": 8150
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 12.268855094909668,
+      "learning_rate": 0.0001,
+      "loss": 5.5183,
+      "step": 8160
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 7.2880682945251465,
+      "learning_rate": 0.0001,
+      "loss": 5.849,
+      "step": 8170
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 8.324126243591309,
+      "learning_rate": 0.0001,
+      "loss": 5.9732,
+      "step": 8180
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 6.885074615478516,
+      "learning_rate": 0.0001,
+      "loss": 5.8009,
+      "step": 8190
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 6.152491092681885,
+      "learning_rate": 0.0001,
+      "loss": 5.6438,
+      "step": 8200
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 6.755161285400391,
+      "learning_rate": 0.0001,
+      "loss": 5.4413,
+      "step": 8210
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 6.749188423156738,
+      "learning_rate": 0.0001,
+      "loss": 5.7069,
+      "step": 8220
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 6.857624053955078,
+      "learning_rate": 0.0001,
+      "loss": 5.7234,
+      "step": 8230
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 7.235221862792969,
+      "learning_rate": 0.0001,
+      "loss": 5.8452,
+      "step": 8240
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 7.165960788726807,
+      "learning_rate": 0.0001,
+      "loss": 5.6126,
+      "step": 8250
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 7.479679107666016,
+      "learning_rate": 0.0001,
+      "loss": 5.8875,
+      "step": 8260
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 7.460275650024414,
+      "learning_rate": 0.0001,
+      "loss": 5.4776,
+      "step": 8270
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 6.884261608123779,
+      "learning_rate": 0.0001,
+      "loss": 5.604,
+      "step": 8280
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 6.93402099609375,
+      "learning_rate": 0.0001,
+      "loss": 5.5675,
+      "step": 8290
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 6.882469177246094,
+      "learning_rate": 0.0001,
+      "loss": 5.4436,
+      "step": 8300
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 6.561320781707764,
+      "learning_rate": 0.0001,
+      "loss": 5.6808,
+      "step": 8310
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 8.458192825317383,
+      "learning_rate": 0.0001,
+      "loss": 5.8178,
+      "step": 8320
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 8.324678421020508,
+      "learning_rate": 0.0001,
+      "loss": 5.4394,
+      "step": 8330
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 6.853770732879639,
+      "learning_rate": 0.0001,
+      "loss": 5.5402,
+      "step": 8340
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 6.634481430053711,
+      "learning_rate": 0.0001,
+      "loss": 5.7143,
+      "step": 8350
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 6.843596935272217,
+      "learning_rate": 0.0001,
+      "loss": 5.4863,
+      "step": 8360
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 8.276290893554688,
+      "learning_rate": 0.0001,
+      "loss": 5.7036,
+      "step": 8370
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 7.795248985290527,
+      "learning_rate": 0.0001,
+      "loss": 5.7043,
+      "step": 8380
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 8.418254852294922,
+      "learning_rate": 0.0001,
+      "loss": 5.5217,
+      "step": 8390
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 6.733100414276123,
+      "learning_rate": 0.0001,
+      "loss": 5.5014,
+      "step": 8400
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 7.750212669372559,
+      "learning_rate": 0.0001,
+      "loss": 5.5834,
+      "step": 8410
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 6.9276323318481445,
+      "learning_rate": 0.0001,
+      "loss": 5.4844,
+      "step": 8420
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 7.04974889755249,
+      "learning_rate": 0.0001,
+      "loss": 5.6058,
+      "step": 8430
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 8.230254173278809,
+      "learning_rate": 0.0001,
+      "loss": 5.7526,
+      "step": 8440
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 7.268505573272705,
+      "learning_rate": 0.0001,
+      "loss": 5.7203,
+      "step": 8450
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 7.4106574058532715,
+      "learning_rate": 0.0001,
+      "loss": 5.3592,
+      "step": 8460
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 6.8834404945373535,
+      "learning_rate": 0.0001,
+      "loss": 5.5183,
+      "step": 8470
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 7.1408209800720215,
+      "learning_rate": 0.0001,
+      "loss": 5.9501,
+      "step": 8480
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 7.075471878051758,
+      "learning_rate": 0.0001,
+      "loss": 5.7628,
+      "step": 8490
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 7.672554969787598,
+      "learning_rate": 0.0001,
+      "loss": 5.7017,
+      "step": 8500
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 6.123058319091797,
+      "learning_rate": 0.0001,
+      "loss": 5.4458,
+      "step": 8510
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 10.512075424194336,
+      "learning_rate": 0.0001,
+      "loss": 5.1682,
+      "step": 8520
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 7.003477573394775,
+      "learning_rate": 0.0001,
+      "loss": 5.4874,
+      "step": 8530
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 7.111389636993408,
+      "learning_rate": 0.0001,
+      "loss": 5.9687,
+      "step": 8540
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 6.513724327087402,
+      "learning_rate": 0.0001,
+      "loss": 5.896,
+      "step": 8550
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 7.711248397827148,
+      "learning_rate": 0.0001,
+      "loss": 5.2314,
+      "step": 8560
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 6.907662391662598,
+      "learning_rate": 0.0001,
+      "loss": 5.3432,
+      "step": 8570
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 6.691112041473389,
+      "learning_rate": 0.0001,
+      "loss": 5.5133,
+      "step": 8580
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 8.064393043518066,
+      "learning_rate": 0.0001,
+      "loss": 5.5594,
+      "step": 8590
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 7.452256679534912,
+      "learning_rate": 0.0001,
+      "loss": 5.6432,
+      "step": 8600
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 7.371987342834473,
+      "learning_rate": 0.0001,
+      "loss": 5.6172,
+      "step": 8610
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 6.757103443145752,
+      "learning_rate": 0.0001,
+      "loss": 5.4583,
+      "step": 8620
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 9.567061424255371,
+      "learning_rate": 0.0001,
+      "loss": 5.6373,
+      "step": 8630
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 6.892947196960449,
+      "learning_rate": 0.0001,
+      "loss": 5.1379,
+      "step": 8640
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 7.586696147918701,
+      "learning_rate": 0.0001,
+      "loss": 5.4885,
+      "step": 8650
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 6.56678581237793,
+      "learning_rate": 0.0001,
+      "loss": 5.3281,
+      "step": 8660
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 7.0721330642700195,
+      "learning_rate": 0.0001,
+      "loss": 5.7122,
+      "step": 8670
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 7.059956073760986,
+      "learning_rate": 0.0001,
+      "loss": 5.5002,
+      "step": 8680
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 7.983617782592773,
+      "learning_rate": 0.0001,
+      "loss": 5.6252,
+      "step": 8690
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 6.5815839767456055,
+      "learning_rate": 0.0001,
+      "loss": 5.2709,
+      "step": 8700
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 7.360924243927002,
+      "learning_rate": 0.0001,
+      "loss": 5.4033,
+      "step": 8710
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 7.26482629776001,
+      "learning_rate": 0.0001,
+      "loss": 5.4748,
+      "step": 8720
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 7.413302898406982,
+      "learning_rate": 0.0001,
+      "loss": 5.3392,
+      "step": 8730
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 7.516079902648926,
+      "learning_rate": 0.0001,
+      "loss": 5.4799,
+      "step": 8740
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 7.392667770385742,
+      "learning_rate": 0.0001,
+      "loss": 5.58,
+      "step": 8750
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 8827,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 250,
+  "total_flos": 4.37471633743872e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}