diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4333 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 6160,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.7635692954063416,
+      "learning_rate": 0.00019967532467532467,
+      "loss": 1.7962,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.8835197687149048,
+      "learning_rate": 0.00019935064935064936,
+      "loss": 0.6672,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.4420289695262909,
+      "learning_rate": 0.00019902597402597403,
+      "loss": 0.4401,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.43932077288627625,
+      "learning_rate": 0.00019870129870129872,
+      "loss": 0.3919,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.4500159025192261,
+      "learning_rate": 0.0001983766233766234,
+      "loss": 0.3576,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.4480540156364441,
+      "learning_rate": 0.00019805194805194807,
+      "loss": 0.3356,
+      "step": 60
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.46007585525512695,
+      "learning_rate": 0.00019772727272727273,
+      "loss": 0.3216,
+      "step": 70
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.4019887149333954,
+      "learning_rate": 0.00019740259740259742,
+      "loss": 0.3303,
+      "step": 80
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.4473404884338379,
+      "learning_rate": 0.00019707792207792208,
+      "loss": 0.3143,
+      "step": 90
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.4089139401912689,
+      "learning_rate": 0.00019675324675324675,
+      "loss": 0.2977,
+      "step": 100
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.40341681241989136,
+      "learning_rate": 0.00019642857142857144,
+      "loss": 0.2922,
+      "step": 110
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.364865243434906,
+      "learning_rate": 0.00019610389610389613,
+      "loss": 0.291,
+      "step": 120
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.3587811589241028,
+      "learning_rate": 0.0001957792207792208,
+      "loss": 0.307,
+      "step": 130
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.46341705322265625,
+      "learning_rate": 0.00019545454545454548,
+      "loss": 0.2949,
+      "step": 140
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.3727189898490906,
+      "learning_rate": 0.00019512987012987014,
+      "loss": 0.2853,
+      "step": 150
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.4104022681713104,
+      "learning_rate": 0.0001948051948051948,
+      "loss": 0.288,
+      "step": 160
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.44043171405792236,
+      "learning_rate": 0.0001944805194805195,
+      "loss": 0.2889,
+      "step": 170
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.29318371415138245,
+      "learning_rate": 0.00019415584415584416,
+      "loss": 0.3016,
+      "step": 180
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.32713964581489563,
+      "learning_rate": 0.00019383116883116882,
+      "loss": 0.293,
+      "step": 190
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.3267592489719391,
+      "learning_rate": 0.00019350649350649354,
+      "loss": 0.2918,
+      "step": 200
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.33080804347991943,
+      "learning_rate": 0.0001931818181818182,
+      "loss": 0.2725,
+      "step": 210
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.3360421061515808,
+      "learning_rate": 0.00019285714285714286,
+      "loss": 0.2742,
+      "step": 220
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.28836965560913086,
+      "learning_rate": 0.00019253246753246755,
+      "loss": 0.2827,
+      "step": 230
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.31561553478240967,
+      "learning_rate": 0.00019220779220779222,
+      "loss": 0.2812,
+      "step": 240
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.2840307950973511,
+      "learning_rate": 0.00019188311688311688,
+      "loss": 0.2717,
+      "step": 250
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.27172142267227173,
+      "learning_rate": 0.00019155844155844157,
+      "loss": 0.2734,
+      "step": 260
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.316268652677536,
+      "learning_rate": 0.00019123376623376626,
+      "loss": 0.2711,
+      "step": 270
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.3135654628276825,
+      "learning_rate": 0.00019090909090909092,
+      "loss": 0.2691,
+      "step": 280
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.25730589032173157,
+      "learning_rate": 0.0001905844155844156,
+      "loss": 0.2705,
+      "step": 290
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.31735968589782715,
+      "learning_rate": 0.00019025974025974027,
+      "loss": 0.2704,
+      "step": 300
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.30799657106399536,
+      "learning_rate": 0.00018993506493506494,
+      "loss": 0.2647,
+      "step": 310
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.3381764590740204,
+      "learning_rate": 0.00018961038961038963,
+      "loss": 0.2694,
+      "step": 320
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.2561868131160736,
+      "learning_rate": 0.0001892857142857143,
+      "loss": 0.2647,
+      "step": 330
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.31571149826049805,
+      "learning_rate": 0.00018896103896103895,
+      "loss": 0.2699,
+      "step": 340
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.2856205105781555,
+      "learning_rate": 0.00018863636363636364,
+      "loss": 0.2665,
+      "step": 350
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.26820528507232666,
+      "learning_rate": 0.00018831168831168833,
+      "loss": 0.266,
+      "step": 360
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.27866330742836,
+      "learning_rate": 0.000187987012987013,
+      "loss": 0.2602,
+      "step": 370
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.26675528287887573,
+      "learning_rate": 0.00018766233766233769,
+      "loss": 0.2682,
+      "step": 380
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.32583510875701904,
+      "learning_rate": 0.00018733766233766235,
+      "loss": 0.2554,
+      "step": 390
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.2768939733505249,
+      "learning_rate": 0.000187012987012987,
+      "loss": 0.261,
+      "step": 400
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.32414308190345764,
+      "learning_rate": 0.0001866883116883117,
+      "loss": 0.2669,
+      "step": 410
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.29801836609840393,
+      "learning_rate": 0.00018636363636363636,
+      "loss": 0.2667,
+      "step": 420
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.27965185046195984,
+      "learning_rate": 0.00018603896103896105,
+      "loss": 0.2483,
+      "step": 430
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.26572081446647644,
+      "learning_rate": 0.00018571428571428572,
+      "loss": 0.2644,
+      "step": 440
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.27148863673210144,
+      "learning_rate": 0.0001853896103896104,
+      "loss": 0.2556,
+      "step": 450
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.2528524696826935,
+      "learning_rate": 0.00018506493506493507,
+      "loss": 0.2508,
+      "step": 460
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.24941977858543396,
+      "learning_rate": 0.00018474025974025976,
+      "loss": 0.25,
+      "step": 470
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.25681182742118835,
+      "learning_rate": 0.00018441558441558442,
+      "loss": 0.2601,
+      "step": 480
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.34882852435112,
+      "learning_rate": 0.00018409090909090909,
+      "loss": 0.2528,
+      "step": 490
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.2678586542606354,
+      "learning_rate": 0.00018376623376623378,
+      "loss": 0.2619,
+      "step": 500
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.2930823266506195,
+      "learning_rate": 0.00018344155844155847,
+      "loss": 0.2713,
+      "step": 510
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.2575713098049164,
+      "learning_rate": 0.00018311688311688313,
+      "loss": 0.2543,
+      "step": 520
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.27224481105804443,
+      "learning_rate": 0.0001827922077922078,
+      "loss": 0.2538,
+      "step": 530
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.26867520809173584,
+      "learning_rate": 0.00018246753246753248,
+      "loss": 0.2627,
+      "step": 540
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.31610989570617676,
+      "learning_rate": 0.00018214285714285714,
+      "loss": 0.2589,
+      "step": 550
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.25024908781051636,
+      "learning_rate": 0.00018181818181818183,
+      "loss": 0.2472,
+      "step": 560
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.25710681080818176,
+      "learning_rate": 0.0001814935064935065,
+      "loss": 0.2528,
+      "step": 570
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.2248850166797638,
+      "learning_rate": 0.0001811688311688312,
+      "loss": 0.2526,
+      "step": 580
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.24048291146755219,
+      "learning_rate": 0.00018084415584415585,
+      "loss": 0.2546,
+      "step": 590
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.36109045147895813,
+      "learning_rate": 0.00018051948051948054,
+      "loss": 0.246,
+      "step": 600
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.22476331889629364,
+      "learning_rate": 0.0001801948051948052,
+      "loss": 0.2509,
+      "step": 610
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.26300567388534546,
+      "learning_rate": 0.00017987012987012987,
+      "loss": 0.2454,
+      "step": 620
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.2843801975250244,
+      "learning_rate": 0.00017954545454545456,
+      "loss": 0.2505,
+      "step": 630
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.23380717635154724,
+      "learning_rate": 0.00017922077922077922,
+      "loss": 0.2553,
+      "step": 640
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.24327170848846436,
+      "learning_rate": 0.0001788961038961039,
+      "loss": 0.2474,
+      "step": 650
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.21787764132022858,
+      "learning_rate": 0.0001785714285714286,
+      "loss": 0.2446,
+      "step": 660
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.2602842450141907,
+      "learning_rate": 0.00017824675324675326,
+      "loss": 0.2431,
+      "step": 670
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.2135908156633377,
+      "learning_rate": 0.00017792207792207792,
+      "loss": 0.2622,
+      "step": 680
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.24802103638648987,
+      "learning_rate": 0.00017759740259740261,
+      "loss": 0.2634,
+      "step": 690
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.21606019139289856,
+      "learning_rate": 0.00017727272727272728,
+      "loss": 0.2467,
+      "step": 700
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.23391862213611603,
+      "learning_rate": 0.00017694805194805194,
+      "loss": 0.2578,
+      "step": 710
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.22003191709518433,
+      "learning_rate": 0.00017662337662337663,
+      "loss": 0.2475,
+      "step": 720
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.24106952548027039,
+      "learning_rate": 0.0001762987012987013,
+      "loss": 0.2486,
+      "step": 730
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.19571684300899506,
+      "learning_rate": 0.00017597402597402598,
+      "loss": 0.2432,
+      "step": 740
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.1917944699525833,
+      "learning_rate": 0.00017564935064935067,
+      "loss": 0.2521,
+      "step": 750
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.19699664413928986,
+      "learning_rate": 0.00017532467532467534,
+      "loss": 0.2564,
+      "step": 760
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.2014484703540802,
+      "learning_rate": 0.000175,
+      "loss": 0.2537,
+      "step": 770
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.22416658699512482,
+      "learning_rate": 0.0001746753246753247,
+      "loss": 0.2615,
+      "step": 780
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.18095014989376068,
+      "learning_rate": 0.00017435064935064935,
+      "loss": 0.2389,
+      "step": 790
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.15944726765155792,
+      "learning_rate": 0.00017402597402597401,
+      "loss": 0.2441,
+      "step": 800
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.23109684884548187,
+      "learning_rate": 0.00017370129870129873,
+      "loss": 0.2412,
+      "step": 810
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.2189890742301941,
+      "learning_rate": 0.0001733766233766234,
+      "loss": 0.2487,
+      "step": 820
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.19159863889217377,
+      "learning_rate": 0.00017305194805194806,
+      "loss": 0.2371,
+      "step": 830
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.2036837339401245,
+      "learning_rate": 0.00017272727272727275,
+      "loss": 0.249,
+      "step": 840
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.1821984201669693,
+      "learning_rate": 0.0001724025974025974,
+      "loss": 0.2427,
+      "step": 850
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.24790024757385254,
+      "learning_rate": 0.00017207792207792207,
+      "loss": 0.2392,
+      "step": 860
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.2032424509525299,
+      "learning_rate": 0.00017175324675324676,
+      "loss": 0.2426,
+      "step": 870
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.16505777835845947,
+      "learning_rate": 0.00017142857142857143,
+      "loss": 0.2375,
+      "step": 880
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.2213476002216339,
+      "learning_rate": 0.00017110389610389612,
+      "loss": 0.2504,
+      "step": 890
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.21578127145767212,
+      "learning_rate": 0.0001707792207792208,
+      "loss": 0.2559,
+      "step": 900
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.20231306552886963,
+      "learning_rate": 0.00017045454545454547,
+      "loss": 0.2431,
+      "step": 910
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.18694159388542175,
+      "learning_rate": 0.00017012987012987013,
+      "loss": 0.2414,
+      "step": 920
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.18938755989074707,
+      "learning_rate": 0.00016980519480519482,
+      "loss": 0.2527,
+      "step": 930
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.17312529683113098,
+      "learning_rate": 0.00016948051948051948,
+      "loss": 0.2436,
+      "step": 940
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.160197913646698,
+      "learning_rate": 0.00016915584415584415,
+      "loss": 0.2347,
+      "step": 950
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.18281523883342743,
+      "learning_rate": 0.00016883116883116884,
+      "loss": 0.24,
+      "step": 960
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.1861046701669693,
+      "learning_rate": 0.00016850649350649353,
+      "loss": 0.2439,
+      "step": 970
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.21592392027378082,
+      "learning_rate": 0.0001681818181818182,
+      "loss": 0.2355,
+      "step": 980
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.18106144666671753,
+      "learning_rate": 0.00016785714285714288,
+      "loss": 0.2449,
+      "step": 990
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.18148744106292725,
+      "learning_rate": 0.00016753246753246754,
+      "loss": 0.2453,
+      "step": 1000
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.24214132130146027,
+      "learning_rate": 0.0001672077922077922,
+      "loss": 0.2429,
+      "step": 1010
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.18021221458911896,
+      "learning_rate": 0.0001668831168831169,
+      "loss": 0.2501,
+      "step": 1020
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.19619709253311157,
+      "learning_rate": 0.00016655844155844156,
+      "loss": 0.2509,
+      "step": 1030
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.1765470653772354,
+      "learning_rate": 0.00016623376623376625,
+      "loss": 0.2394,
+      "step": 1040
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.16927708685398102,
+      "learning_rate": 0.00016590909090909094,
+      "loss": 0.2375,
+      "step": 1050
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.21551525592803955,
+      "learning_rate": 0.0001655844155844156,
+      "loss": 0.239,
+      "step": 1060
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.1624051332473755,
+      "learning_rate": 0.00016525974025974026,
+      "loss": 0.2363,
+      "step": 1070
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.17235176265239716,
+      "learning_rate": 0.00016493506493506495,
+      "loss": 0.2472,
+      "step": 1080
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.1943296194076538,
+      "learning_rate": 0.00016461038961038962,
+      "loss": 0.2484,
+      "step": 1090
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.17198781669139862,
+      "learning_rate": 0.00016428571428571428,
+      "loss": 0.2477,
+      "step": 1100
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.15986420214176178,
+      "learning_rate": 0.00016396103896103897,
+      "loss": 0.2351,
+      "step": 1110
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.19471360743045807,
+      "learning_rate": 0.00016363636363636366,
+      "loss": 0.2383,
+      "step": 1120
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.18861734867095947,
+      "learning_rate": 0.00016331168831168832,
+      "loss": 0.249,
+      "step": 1130
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.16632774472236633,
+      "learning_rate": 0.000162987012987013,
+      "loss": 0.241,
+      "step": 1140
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.16405850648880005,
+      "learning_rate": 0.00016266233766233767,
+      "loss": 0.2431,
+      "step": 1150
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.22708448767662048,
+      "learning_rate": 0.00016233766233766234,
+      "loss": 0.2342,
+      "step": 1160
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.19473408162593842,
+      "learning_rate": 0.00016201298701298703,
+      "loss": 0.2351,
+      "step": 1170
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.16566310822963715,
+      "learning_rate": 0.0001616883116883117,
+      "loss": 0.2541,
+      "step": 1180
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.24258695542812347,
+      "learning_rate": 0.00016136363636363635,
+      "loss": 0.236,
+      "step": 1190
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.17347609996795654,
+      "learning_rate": 0.00016103896103896104,
+      "loss": 0.2285,
+      "step": 1200
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.1914084255695343,
+      "learning_rate": 0.00016071428571428573,
+      "loss": 0.2349,
+      "step": 1210
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.25977787375450134,
+      "learning_rate": 0.0001603896103896104,
+      "loss": 0.2476,
+      "step": 1220
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.18098042905330658,
+      "learning_rate": 0.00016006493506493509,
+      "loss": 0.2483,
+      "step": 1230
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.1897449642419815,
+      "learning_rate": 0.00015974025974025975,
+      "loss": 0.2434,
+      "step": 1240
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.14767146110534668,
+      "learning_rate": 0.0001594155844155844,
+      "loss": 0.2427,
+      "step": 1250
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.17135238647460938,
+      "learning_rate": 0.0001590909090909091,
+      "loss": 0.2348,
+      "step": 1260
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.19989098608493805,
+      "learning_rate": 0.00015876623376623376,
+      "loss": 0.2383,
+      "step": 1270
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.14210206270217896,
+      "learning_rate": 0.00015844155844155845,
+      "loss": 0.2495,
+      "step": 1280
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.1692097932100296,
+      "learning_rate": 0.00015811688311688312,
+      "loss": 0.2383,
+      "step": 1290
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.14251230657100677,
+      "learning_rate": 0.0001577922077922078,
+      "loss": 0.2431,
+      "step": 1300
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.15610291063785553,
+      "learning_rate": 0.00015746753246753247,
+      "loss": 0.254,
+      "step": 1310
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.17254161834716797,
+      "learning_rate": 0.00015714285714285716,
+      "loss": 0.2435,
+      "step": 1320
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.1727110743522644,
+      "learning_rate": 0.00015681818181818182,
+      "loss": 0.2476,
+      "step": 1330
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.18663935363292694,
+      "learning_rate": 0.00015649350649350649,
+      "loss": 0.2404,
+      "step": 1340
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.152663916349411,
+      "learning_rate": 0.00015616883116883118,
+      "loss": 0.2391,
+      "step": 1350
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.1640835851430893,
+      "learning_rate": 0.00015584415584415587,
+      "loss": 0.236,
+      "step": 1360
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.1654743254184723,
+      "learning_rate": 0.00015551948051948053,
+      "loss": 0.259,
+      "step": 1370
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.22310000658035278,
+      "learning_rate": 0.0001551948051948052,
+      "loss": 0.2314,
+      "step": 1380
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.17061170935630798,
+      "learning_rate": 0.00015487012987012988,
+      "loss": 0.2421,
+      "step": 1390
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.17354106903076172,
+      "learning_rate": 0.00015454545454545454,
+      "loss": 0.246,
+      "step": 1400
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.1683073192834854,
+      "learning_rate": 0.00015422077922077923,
+      "loss": 0.2393,
+      "step": 1410
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.19657649099826813,
+      "learning_rate": 0.0001538961038961039,
+      "loss": 0.2361,
+      "step": 1420
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.16683852672576904,
+      "learning_rate": 0.0001535714285714286,
+      "loss": 0.2357,
+      "step": 1430
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.14492397010326385,
+      "learning_rate": 0.00015324675324675325,
+      "loss": 0.2461,
+      "step": 1440
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.16166526079177856,
+      "learning_rate": 0.00015292207792207794,
+      "loss": 0.2369,
+      "step": 1450
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.13416822254657745,
+      "learning_rate": 0.0001525974025974026,
+      "loss": 0.2351,
+      "step": 1460
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.15957880020141602,
+      "learning_rate": 0.00015227272727272727,
+      "loss": 0.2452,
+      "step": 1470
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.1618945598602295,
+      "learning_rate": 0.00015194805194805196,
+      "loss": 0.2318,
+      "step": 1480
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.16433538496494293,
+      "learning_rate": 0.00015162337662337662,
+      "loss": 0.2417,
+      "step": 1490
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.1811251938343048,
+      "learning_rate": 0.0001512987012987013,
+      "loss": 0.2373,
+      "step": 1500
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.1398647129535675,
+      "learning_rate": 0.000150974025974026,
+      "loss": 0.2345,
+      "step": 1510
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.15805283188819885,
+      "learning_rate": 0.00015064935064935066,
+      "loss": 0.237,
+      "step": 1520
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.14151178300380707,
+      "learning_rate": 0.00015032467532467532,
+      "loss": 0.2371,
+      "step": 1530
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.1412213295698166,
+      "learning_rate": 0.00015000000000000001,
+      "loss": 0.2366,
+      "step": 1540
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.1680016666650772,
+      "learning_rate": 0.00014967532467532468,
+      "loss": 0.2302,
+      "step": 1550
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.14209860563278198,
+      "learning_rate": 0.00014935064935064934,
+      "loss": 0.2402,
+      "step": 1560
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.19270756840705872,
+      "learning_rate": 0.00014902597402597403,
+      "loss": 0.2407,
+      "step": 1570
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.15573273599147797,
+      "learning_rate": 0.00014870129870129872,
+      "loss": 0.2328,
+      "step": 1580
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.14654339849948883,
+      "learning_rate": 0.00014837662337662338,
+      "loss": 0.2338,
+      "step": 1590
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.14677156507968903,
+      "learning_rate": 0.00014805194805194807,
+      "loss": 0.2355,
+      "step": 1600
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.13458256423473358,
+      "learning_rate": 0.00014772727272727274,
+      "loss": 0.2371,
+      "step": 1610
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.15569452941417694,
+      "learning_rate": 0.0001474025974025974,
+      "loss": 0.2392,
+      "step": 1620
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.14752694964408875,
+      "learning_rate": 0.0001470779220779221,
+      "loss": 0.2381,
+      "step": 1630
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.13646119832992554,
+      "learning_rate": 0.00014675324675324675,
+      "loss": 0.2378,
+      "step": 1640
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.13957124948501587,
+      "learning_rate": 0.00014642857142857141,
+      "loss": 0.2427,
+      "step": 1650
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.2876950204372406,
+      "learning_rate": 0.00014610389610389613,
+      "loss": 0.2349,
+      "step": 1660
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.27290669083595276,
+      "learning_rate": 0.0001457792207792208,
+      "loss": 0.2407,
+      "step": 1670
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.1855621188879013,
+      "learning_rate": 0.00014545454545454546,
+      "loss": 0.2467,
+      "step": 1680
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.18681453168392181,
+      "learning_rate": 0.00014512987012987015,
+      "loss": 0.2367,
+      "step": 1690
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.1627301722764969,
+      "learning_rate": 0.0001448051948051948,
+      "loss": 0.2485,
+      "step": 1700
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.14202873408794403,
+      "learning_rate": 0.00014448051948051947,
+      "loss": 0.232,
+      "step": 1710
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.15968629717826843,
+      "learning_rate": 0.00014415584415584416,
+      "loss": 0.229,
+      "step": 1720
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.14707553386688232,
+      "learning_rate": 0.00014383116883116883,
+      "loss": 0.2402,
+      "step": 1730
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.1813046634197235,
+      "learning_rate": 0.00014350649350649352,
+      "loss": 0.232,
+      "step": 1740
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.2322862595319748,
+      "learning_rate": 0.0001431818181818182,
+      "loss": 0.24,
+      "step": 1750
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.130840003490448,
+      "learning_rate": 0.00014285714285714287,
+      "loss": 0.2364,
+      "step": 1760
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.1930732727050781,
+      "learning_rate": 0.00014253246753246753,
+      "loss": 0.2471,
+      "step": 1770
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.15140008926391602,
+      "learning_rate": 0.00014220779220779222,
+      "loss": 0.2379,
+      "step": 1780
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.1689486801624298,
+      "learning_rate": 0.00014188311688311688,
+      "loss": 0.2338,
+      "step": 1790
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.17393773794174194,
+      "learning_rate": 0.00014155844155844155,
+      "loss": 0.2356,
+      "step": 1800
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.1497236043214798,
+      "learning_rate": 0.00014123376623376626,
+      "loss": 0.234,
+      "step": 1810
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.19287332892417908,
+      "learning_rate": 0.00014090909090909093,
+      "loss": 0.2292,
+      "step": 1820
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.1971684992313385,
+      "learning_rate": 0.0001405844155844156,
+      "loss": 0.2424,
+      "step": 1830
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.15559417009353638,
+      "learning_rate": 0.00014025974025974028,
+      "loss": 0.252,
+      "step": 1840
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.17173197865486145,
+      "learning_rate": 0.00013993506493506494,
+      "loss": 0.2409,
+      "step": 1850
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.18045447766780853,
+      "learning_rate": 0.0001396103896103896,
+      "loss": 0.2383,
+      "step": 1860
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.12264901399612427,
+      "learning_rate": 0.0001392857142857143,
+      "loss": 0.2429,
+      "step": 1870
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.15579771995544434,
+      "learning_rate": 0.00013896103896103896,
+      "loss": 0.2469,
+      "step": 1880
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.15319515764713287,
+      "learning_rate": 0.00013863636363636365,
+      "loss": 0.2375,
+      "step": 1890
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.1389021873474121,
+      "learning_rate": 0.00013831168831168834,
+      "loss": 0.2305,
+      "step": 1900
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.14433075487613678,
+      "learning_rate": 0.000137987012987013,
+      "loss": 0.2338,
+      "step": 1910
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.15670061111450195,
+      "learning_rate": 0.00013766233766233766,
+      "loss": 0.2382,
+      "step": 1920
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.1808587610721588,
+      "learning_rate": 0.00013733766233766235,
+      "loss": 0.2227,
+      "step": 1930
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.141093447804451,
+      "learning_rate": 0.00013701298701298702,
+      "loss": 0.2458,
+      "step": 1940
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.1257307231426239,
+      "learning_rate": 0.00013668831168831168,
+      "loss": 0.2422,
+      "step": 1950
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.14036783576011658,
+      "learning_rate": 0.00013636363636363637,
+      "loss": 0.2326,
+      "step": 1960
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.1454552710056305,
+      "learning_rate": 0.00013603896103896106,
+      "loss": 0.2385,
+      "step": 1970
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.14793068170547485,
+      "learning_rate": 0.00013571428571428572,
+      "loss": 0.2361,
+      "step": 1980
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.18030133843421936,
+      "learning_rate": 0.0001353896103896104,
+      "loss": 0.2371,
+      "step": 1990
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.15282848477363586,
+      "learning_rate": 0.00013506493506493507,
+      "loss": 0.2436,
+      "step": 2000
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.13970941305160522,
+      "learning_rate": 0.00013474025974025974,
+      "loss": 0.241,
+      "step": 2010
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.12444699555635452,
+      "learning_rate": 0.00013441558441558443,
+      "loss": 0.2381,
+      "step": 2020
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.13609889149665833,
+      "learning_rate": 0.0001340909090909091,
+      "loss": 0.247,
+      "step": 2030
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.14910899102687836,
+      "learning_rate": 0.00013376623376623375,
+      "loss": 0.2338,
+      "step": 2040
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.12678007781505585,
+      "learning_rate": 0.00013344155844155844,
+      "loss": 0.2435,
+      "step": 2050
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.1263749897480011,
+      "learning_rate": 0.00013311688311688313,
+      "loss": 0.228,
+      "step": 2060
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.1846819370985031,
+      "learning_rate": 0.0001327922077922078,
+      "loss": 0.235,
+      "step": 2070
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.15016387403011322,
+      "learning_rate": 0.00013246753246753249,
+      "loss": 0.2447,
+      "step": 2080
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.12729190289974213,
+      "learning_rate": 0.00013214285714285715,
+      "loss": 0.2368,
+      "step": 2090
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.17617157101631165,
+      "learning_rate": 0.0001318181818181818,
+      "loss": 0.2337,
+      "step": 2100
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.13113613426685333,
+      "learning_rate": 0.0001314935064935065,
+      "loss": 0.237,
+      "step": 2110
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.13761071860790253,
+      "learning_rate": 0.0001311688311688312,
+      "loss": 0.2339,
+      "step": 2120
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.13893531262874603,
+      "learning_rate": 0.00013084415584415585,
+      "loss": 0.2282,
+      "step": 2130
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.13437838852405548,
+      "learning_rate": 0.00013051948051948052,
+      "loss": 0.2416,
+      "step": 2140
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.13162140548229218,
+      "learning_rate": 0.0001301948051948052,
+      "loss": 0.2434,
+      "step": 2150
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.12518011033535004,
+      "learning_rate": 0.00012987012987012987,
+      "loss": 0.2355,
+      "step": 2160
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.12965276837348938,
+      "learning_rate": 0.00012954545454545456,
+      "loss": 0.2344,
+      "step": 2170
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.1295609027147293,
+      "learning_rate": 0.00012922077922077922,
+      "loss": 0.2346,
+      "step": 2180
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.14101965725421906,
+      "learning_rate": 0.00012889610389610389,
+      "loss": 0.2338,
+      "step": 2190
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.14208084344863892,
+      "learning_rate": 0.00012857142857142858,
+      "loss": 0.2435,
+      "step": 2200
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.10847887396812439,
+      "learning_rate": 0.00012824675324675327,
+      "loss": 0.2286,
+      "step": 2210
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.13687682151794434,
+      "learning_rate": 0.00012792207792207793,
+      "loss": 0.2405,
+      "step": 2220
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.12724491953849792,
+      "learning_rate": 0.0001275974025974026,
+      "loss": 0.2272,
+      "step": 2230
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.1222841814160347,
+      "learning_rate": 0.00012727272727272728,
+      "loss": 0.2339,
+      "step": 2240
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.16014225780963898,
+      "learning_rate": 0.00012694805194805194,
+      "loss": 0.2346,
+      "step": 2250
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.1270248293876648,
+      "learning_rate": 0.00012662337662337663,
+      "loss": 0.2331,
+      "step": 2260
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.1222517117857933,
+      "learning_rate": 0.0001262987012987013,
+      "loss": 0.2325,
+      "step": 2270
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.11582531034946442,
+      "learning_rate": 0.000125974025974026,
+      "loss": 0.2294,
+      "step": 2280
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.11494652181863785,
+      "learning_rate": 0.00012564935064935065,
+      "loss": 0.229,
+      "step": 2290
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.12617026269435883,
+      "learning_rate": 0.00012532467532467534,
+      "loss": 0.2332,
+      "step": 2300
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.14116579294204712,
+      "learning_rate": 0.000125,
+      "loss": 0.2374,
+      "step": 2310
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.14407993853092194,
+      "learning_rate": 0.00012467532467532467,
+      "loss": 0.2315,
+      "step": 2320
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.12318301200866699,
+      "learning_rate": 0.00012435064935064936,
+      "loss": 0.2345,
+      "step": 2330
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.13765230774879456,
+      "learning_rate": 0.00012402597402597402,
+      "loss": 0.2352,
+      "step": 2340
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.12083020061254501,
+      "learning_rate": 0.0001237012987012987,
+      "loss": 0.2284,
+      "step": 2350
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.12487612664699554,
+      "learning_rate": 0.0001233766233766234,
+      "loss": 0.2379,
+      "step": 2360
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.1041794940829277,
+      "learning_rate": 0.00012305194805194806,
+      "loss": 0.2387,
+      "step": 2370
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.1314081996679306,
+      "learning_rate": 0.00012272727272727272,
+      "loss": 0.2392,
+      "step": 2380
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.1306692659854889,
+      "learning_rate": 0.00012240259740259741,
+      "loss": 0.2342,
+      "step": 2390
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.10851772129535675,
+      "learning_rate": 0.00012207792207792208,
+      "loss": 0.2412,
+      "step": 2400
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.12694145739078522,
+      "learning_rate": 0.00012175324675324675,
+      "loss": 0.2404,
+      "step": 2410
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.1565706729888916,
+      "learning_rate": 0.00012142857142857143,
+      "loss": 0.2422,
+      "step": 2420
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.12534740567207336,
+      "learning_rate": 0.00012110389610389612,
+      "loss": 0.2401,
+      "step": 2430
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.1272275745868683,
+      "learning_rate": 0.0001207792207792208,
+      "loss": 0.2308,
+      "step": 2440
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.14152082800865173,
+      "learning_rate": 0.00012045454545454546,
+      "loss": 0.2494,
+      "step": 2450
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.13283218443393707,
+      "learning_rate": 0.00012012987012987014,
+      "loss": 0.2459,
+      "step": 2460
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.12492993474006653,
+      "learning_rate": 0.00011980519480519481,
+      "loss": 0.2354,
+      "step": 2470
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.13865774869918823,
+      "learning_rate": 0.00011948051948051949,
+      "loss": 0.2383,
+      "step": 2480
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.12774866819381714,
+      "learning_rate": 0.00011915584415584415,
+      "loss": 0.2305,
+      "step": 2490
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.11156503856182098,
+      "learning_rate": 0.00011883116883116883,
+      "loss": 0.2325,
+      "step": 2500
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.16965094208717346,
+      "learning_rate": 0.00011850649350649352,
+      "loss": 0.242,
+      "step": 2510
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.1178189367055893,
+      "learning_rate": 0.0001181818181818182,
+      "loss": 0.2316,
+      "step": 2520
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.11801028996706009,
+      "learning_rate": 0.00011785714285714287,
+      "loss": 0.2351,
+      "step": 2530
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.1737416535615921,
+      "learning_rate": 0.00011753246753246753,
+      "loss": 0.23,
+      "step": 2540
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.15001146495342255,
+      "learning_rate": 0.00011720779220779221,
+      "loss": 0.2325,
+      "step": 2550
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.14643049240112305,
+      "learning_rate": 0.00011688311688311689,
+      "loss": 0.242,
+      "step": 2560
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.1325196772813797,
+      "learning_rate": 0.00011655844155844156,
+      "loss": 0.2354,
+      "step": 2570
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.1217784732580185,
+      "learning_rate": 0.00011623376623376625,
+      "loss": 0.2347,
+      "step": 2580
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.14024858176708221,
+      "learning_rate": 0.00011590909090909093,
+      "loss": 0.2293,
+      "step": 2590
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.12446459382772446,
+      "learning_rate": 0.00011558441558441559,
+      "loss": 0.2336,
+      "step": 2600
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.16514386236667633,
+      "learning_rate": 0.00011525974025974027,
+      "loss": 0.2462,
+      "step": 2610
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.15729106962680817,
+      "learning_rate": 0.00011493506493506494,
+      "loss": 0.2351,
+      "step": 2620
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.1146213635802269,
+      "learning_rate": 0.00011461038961038961,
+      "loss": 0.2412,
+      "step": 2630
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.15709975361824036,
+      "learning_rate": 0.00011428571428571428,
+      "loss": 0.241,
+      "step": 2640
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.12694121897220612,
+      "learning_rate": 0.00011396103896103896,
+      "loss": 0.2354,
+      "step": 2650
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.14171747863292694,
+      "learning_rate": 0.00011363636363636365,
+      "loss": 0.2256,
+      "step": 2660
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.16765841841697693,
+      "learning_rate": 0.00011331168831168833,
+      "loss": 0.2321,
+      "step": 2670
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.18386825919151306,
+      "learning_rate": 0.000112987012987013,
+      "loss": 0.2418,
+      "step": 2680
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.16596557199954987,
+      "learning_rate": 0.00011266233766233767,
+      "loss": 0.2309,
+      "step": 2690
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.13755054771900177,
+      "learning_rate": 0.00011233766233766234,
+      "loss": 0.234,
+      "step": 2700
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.13105453550815582,
+      "learning_rate": 0.00011201298701298702,
+      "loss": 0.2333,
+      "step": 2710
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.1286339908838272,
+      "learning_rate": 0.00011168831168831168,
+      "loss": 0.2368,
+      "step": 2720
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.12714940309524536,
+      "learning_rate": 0.00011136363636363636,
+      "loss": 0.2274,
+      "step": 2730
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.10796003043651581,
+      "learning_rate": 0.00011103896103896105,
+      "loss": 0.2275,
+      "step": 2740
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.12860055267810822,
+      "learning_rate": 0.00011071428571428572,
+      "loss": 0.2351,
+      "step": 2750
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.11882119625806808,
+      "learning_rate": 0.0001103896103896104,
+      "loss": 0.2333,
+      "step": 2760
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.11222593486309052,
+      "learning_rate": 0.00011006493506493508,
+      "loss": 0.233,
+      "step": 2770
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.11925679445266724,
+      "learning_rate": 0.00010974025974025974,
+      "loss": 0.2398,
+      "step": 2780
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.11611876636743546,
+      "learning_rate": 0.00010941558441558442,
+      "loss": 0.2408,
+      "step": 2790
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.12408678978681564,
+      "learning_rate": 0.00010909090909090909,
+      "loss": 0.2308,
+      "step": 2800
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.4648301899433136,
+      "learning_rate": 0.00010876623376623376,
+      "loss": 0.2365,
+      "step": 2810
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.1081792488694191,
+      "learning_rate": 0.00010844155844155846,
+      "loss": 0.2373,
+      "step": 2820
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.1397983580827713,
+      "learning_rate": 0.00010811688311688312,
+      "loss": 0.2481,
+      "step": 2830
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.20857000350952148,
+      "learning_rate": 0.0001077922077922078,
+      "loss": 0.2295,
+      "step": 2840
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.11216516047716141,
+      "learning_rate": 0.00010746753246753247,
+      "loss": 0.2351,
+      "step": 2850
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.11065667122602463,
+      "learning_rate": 0.00010714285714285715,
+      "loss": 0.2256,
+      "step": 2860
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.12144267559051514,
+      "learning_rate": 0.00010681818181818181,
+      "loss": 0.2357,
+      "step": 2870
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.11921383440494537,
+      "learning_rate": 0.00010649350649350649,
+      "loss": 0.2309,
+      "step": 2880
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.12087354063987732,
+      "learning_rate": 0.00010616883116883118,
+      "loss": 0.2377,
+      "step": 2890
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.12411858886480331,
+      "learning_rate": 0.00010584415584415586,
+      "loss": 0.225,
+      "step": 2900
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.1320037543773651,
+      "learning_rate": 0.00010551948051948053,
+      "loss": 0.2424,
+      "step": 2910
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.1209094449877739,
+      "learning_rate": 0.0001051948051948052,
+      "loss": 0.2399,
+      "step": 2920
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.1084308847784996,
+      "learning_rate": 0.00010487012987012987,
+      "loss": 0.2192,
+      "step": 2930
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.12274836748838425,
+      "learning_rate": 0.00010454545454545455,
+      "loss": 0.2273,
+      "step": 2940
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.11587496101856232,
+      "learning_rate": 0.00010422077922077923,
+      "loss": 0.2329,
+      "step": 2950
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.11308249086141586,
+      "learning_rate": 0.00010389610389610389,
+      "loss": 0.2268,
+      "step": 2960
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.13320808112621307,
+      "learning_rate": 0.00010357142857142859,
+      "loss": 0.2236,
+      "step": 2970
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.11789855360984802,
+      "learning_rate": 0.00010324675324675325,
+      "loss": 0.2244,
+      "step": 2980
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.11310496926307678,
+      "learning_rate": 0.00010292207792207793,
+      "loss": 0.2335,
+      "step": 2990
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.142527237534523,
+      "learning_rate": 0.00010259740259740261,
+      "loss": 0.233,
+      "step": 3000
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.12672272324562073,
+      "learning_rate": 0.00010227272727272727,
+      "loss": 0.2367,
+      "step": 3010
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.1247657984495163,
+      "learning_rate": 0.00010194805194805195,
+      "loss": 0.2237,
+      "step": 3020
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.16208265721797943,
+      "learning_rate": 0.00010162337662337662,
+      "loss": 0.2293,
+      "step": 3030
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.11926627904176712,
+      "learning_rate": 0.0001012987012987013,
+      "loss": 0.2264,
+      "step": 3040
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.12271419167518616,
+      "learning_rate": 0.00010097402597402599,
+      "loss": 0.2267,
+      "step": 3050
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.1259615421295166,
+      "learning_rate": 0.00010064935064935067,
+      "loss": 0.2335,
+      "step": 3060
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.11754319071769714,
+      "learning_rate": 0.00010032467532467533,
+      "loss": 0.2261,
+      "step": 3070
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.12118161469697952,
+      "learning_rate": 0.0001,
+      "loss": 0.2291,
+      "step": 3080
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.10783757269382477,
+      "learning_rate": 9.967532467532468e-05,
+      "loss": 0.2459,
+      "step": 3090
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.1269863098859787,
+      "learning_rate": 9.935064935064936e-05,
+      "loss": 0.2286,
+      "step": 3100
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.1457485854625702,
+      "learning_rate": 9.902597402597403e-05,
+      "loss": 0.2368,
+      "step": 3110
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.11660724133253098,
+      "learning_rate": 9.870129870129871e-05,
+      "loss": 0.2467,
+      "step": 3120
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.11261265724897385,
+      "learning_rate": 9.837662337662337e-05,
+      "loss": 0.2345,
+      "step": 3130
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.10939832776784897,
+      "learning_rate": 9.805194805194806e-05,
+      "loss": 0.2238,
+      "step": 3140
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.11895822733640671,
+      "learning_rate": 9.772727272727274e-05,
+      "loss": 0.2274,
+      "step": 3150
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.10990051180124283,
+      "learning_rate": 9.74025974025974e-05,
+      "loss": 0.2418,
+      "step": 3160
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.18286411464214325,
+      "learning_rate": 9.707792207792208e-05,
+      "loss": 0.236,
+      "step": 3170
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.12277409434318542,
+      "learning_rate": 9.675324675324677e-05,
+      "loss": 0.2346,
+      "step": 3180
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.1168995350599289,
+      "learning_rate": 9.642857142857143e-05,
+      "loss": 0.2346,
+      "step": 3190
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.6002693176269531,
+      "learning_rate": 9.610389610389611e-05,
+      "loss": 0.2355,
+      "step": 3200
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.13241051137447357,
+      "learning_rate": 9.577922077922078e-05,
+      "loss": 0.2371,
+      "step": 3210
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.11275602877140045,
+      "learning_rate": 9.545454545454546e-05,
+      "loss": 0.2341,
+      "step": 3220
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.13273973762989044,
+      "learning_rate": 9.512987012987014e-05,
+      "loss": 0.2355,
+      "step": 3230
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.11785340309143066,
+      "learning_rate": 9.480519480519481e-05,
+      "loss": 0.2298,
+      "step": 3240
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.12268821150064468,
+      "learning_rate": 9.448051948051948e-05,
+      "loss": 0.2358,
+      "step": 3250
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.15601615607738495,
+      "learning_rate": 9.415584415584417e-05,
+      "loss": 0.2464,
+      "step": 3260
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.14406079053878784,
+      "learning_rate": 9.383116883116884e-05,
+      "loss": 0.2231,
+      "step": 3270
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.12585234642028809,
+      "learning_rate": 9.35064935064935e-05,
+      "loss": 0.2557,
+      "step": 3280
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.14403945207595825,
+      "learning_rate": 9.318181818181818e-05,
+      "loss": 0.229,
+      "step": 3290
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.11052271723747253,
+      "learning_rate": 9.285714285714286e-05,
+      "loss": 0.2316,
+      "step": 3300
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.11795616894960403,
+      "learning_rate": 9.253246753246754e-05,
+      "loss": 0.2367,
+      "step": 3310
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.11179772764444351,
+      "learning_rate": 9.220779220779221e-05,
+      "loss": 0.2366,
+      "step": 3320
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.11838437616825104,
+      "learning_rate": 9.188311688311689e-05,
+      "loss": 0.2333,
+      "step": 3330
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.11461728811264038,
+      "learning_rate": 9.155844155844156e-05,
+      "loss": 0.2287,
+      "step": 3340
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.10472013801336288,
+      "learning_rate": 9.123376623376624e-05,
+      "loss": 0.2286,
+      "step": 3350
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.1110708937048912,
+      "learning_rate": 9.090909090909092e-05,
+      "loss": 0.2415,
+      "step": 3360
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.13132265210151672,
+      "learning_rate": 9.05844155844156e-05,
+      "loss": 0.2294,
+      "step": 3370
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.12723708152770996,
+      "learning_rate": 9.025974025974027e-05,
+      "loss": 0.2259,
+      "step": 3380
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.1238892674446106,
+      "learning_rate": 8.993506493506493e-05,
+      "loss": 0.2451,
+      "step": 3390
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.10817281901836395,
+      "learning_rate": 8.961038961038961e-05,
+      "loss": 0.2311,
+      "step": 3400
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.12980854511260986,
+      "learning_rate": 8.92857142857143e-05,
+      "loss": 0.2428,
+      "step": 3410
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.1358005255460739,
+      "learning_rate": 8.896103896103896e-05,
+      "loss": 0.2333,
+      "step": 3420
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.1351250410079956,
+      "learning_rate": 8.863636363636364e-05,
+      "loss": 0.2395,
+      "step": 3430
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.11637056618928909,
+      "learning_rate": 8.831168831168831e-05,
+      "loss": 0.2409,
+      "step": 3440
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.15907792747020721,
+      "learning_rate": 8.798701298701299e-05,
+      "loss": 0.2325,
+      "step": 3450
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.1116669550538063,
+      "learning_rate": 8.766233766233767e-05,
+      "loss": 0.2361,
+      "step": 3460
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.12072435766458511,
+      "learning_rate": 8.733766233766234e-05,
+      "loss": 0.2326,
+      "step": 3470
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.11513634026050568,
+      "learning_rate": 8.701298701298701e-05,
+      "loss": 0.2282,
+      "step": 3480
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.1277078241109848,
+      "learning_rate": 8.66883116883117e-05,
+      "loss": 0.2293,
+      "step": 3490
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.12794838845729828,
+      "learning_rate": 8.636363636363637e-05,
+      "loss": 0.2408,
+      "step": 3500
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.12294359505176544,
+      "learning_rate": 8.603896103896104e-05,
+      "loss": 0.2309,
+      "step": 3510
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.1393396109342575,
+      "learning_rate": 8.571428571428571e-05,
+      "loss": 0.2346,
+      "step": 3520
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.12883618474006653,
+      "learning_rate": 8.53896103896104e-05,
+      "loss": 0.2259,
+      "step": 3530
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.11025449633598328,
+      "learning_rate": 8.506493506493507e-05,
+      "loss": 0.2222,
+      "step": 3540
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.10652487725019455,
+      "learning_rate": 8.474025974025974e-05,
+      "loss": 0.2267,
+      "step": 3550
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.14078746736049652,
+      "learning_rate": 8.441558441558442e-05,
+      "loss": 0.2384,
+      "step": 3560
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.10785678774118423,
+      "learning_rate": 8.40909090909091e-05,
+      "loss": 0.2419,
+      "step": 3570
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.1360328048467636,
+      "learning_rate": 8.376623376623377e-05,
+      "loss": 0.2269,
+      "step": 3580
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.12993495166301727,
+      "learning_rate": 8.344155844155845e-05,
+      "loss": 0.2353,
+      "step": 3590
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.11193781346082687,
+      "learning_rate": 8.311688311688312e-05,
+      "loss": 0.2255,
+      "step": 3600
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.12667816877365112,
+      "learning_rate": 8.27922077922078e-05,
+      "loss": 0.2391,
+      "step": 3610
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.10947450250387192,
+      "learning_rate": 8.246753246753248e-05,
+      "loss": 0.2352,
+      "step": 3620
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.10795076936483383,
+      "learning_rate": 8.214285714285714e-05,
+      "loss": 0.2308,
+      "step": 3630
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.10383721441030502,
+      "learning_rate": 8.181818181818183e-05,
+      "loss": 0.2233,
+      "step": 3640
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.11570408195257187,
+      "learning_rate": 8.14935064935065e-05,
+      "loss": 0.2286,
+      "step": 3650
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.11449676007032394,
+      "learning_rate": 8.116883116883117e-05,
+      "loss": 0.2361,
+      "step": 3660
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.14948555827140808,
+      "learning_rate": 8.084415584415585e-05,
+      "loss": 0.2407,
+      "step": 3670
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.12643155455589294,
+      "learning_rate": 8.051948051948052e-05,
+      "loss": 0.2413,
+      "step": 3680
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.11452844738960266,
+      "learning_rate": 8.01948051948052e-05,
+      "loss": 0.2308,
+      "step": 3690
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.13087721168994904,
+      "learning_rate": 7.987012987012987e-05,
+      "loss": 0.2317,
+      "step": 3700
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.1411277949810028,
+      "learning_rate": 7.954545454545455e-05,
+      "loss": 0.243,
+      "step": 3710
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.1330225020647049,
+      "learning_rate": 7.922077922077923e-05,
+      "loss": 0.2288,
+      "step": 3720
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.135909765958786,
+      "learning_rate": 7.88961038961039e-05,
+      "loss": 0.2332,
+      "step": 3730
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.12682045996189117,
+      "learning_rate": 7.857142857142858e-05,
+      "loss": 0.2253,
+      "step": 3740
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.10853119194507599,
+      "learning_rate": 7.824675324675324e-05,
+      "loss": 0.2233,
+      "step": 3750
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.11555913835763931,
+      "learning_rate": 7.792207792207793e-05,
+      "loss": 0.2314,
+      "step": 3760
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.11717897653579712,
+      "learning_rate": 7.75974025974026e-05,
+      "loss": 0.2262,
+      "step": 3770
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.11001992970705032,
+      "learning_rate": 7.727272727272727e-05,
+      "loss": 0.2263,
+      "step": 3780
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.12689049541950226,
+      "learning_rate": 7.694805194805195e-05,
+      "loss": 0.2331,
+      "step": 3790
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.10925900191068649,
+      "learning_rate": 7.662337662337662e-05,
+      "loss": 0.2431,
+      "step": 3800
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.1141931340098381,
+      "learning_rate": 7.62987012987013e-05,
+      "loss": 0.235,
+      "step": 3810
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.1575770378112793,
+      "learning_rate": 7.597402597402598e-05,
+      "loss": 0.2256,
+      "step": 3820
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.13379736244678497,
+      "learning_rate": 7.564935064935065e-05,
+      "loss": 0.2313,
+      "step": 3830
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.11195072531700134,
+      "learning_rate": 7.532467532467533e-05,
+      "loss": 0.2284,
+      "step": 3840
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.13220836222171783,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 0.2453,
+      "step": 3850
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.10971071571111679,
+      "learning_rate": 7.467532467532467e-05,
+      "loss": 0.2234,
+      "step": 3860
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.11772307008504868,
+      "learning_rate": 7.435064935064936e-05,
+      "loss": 0.22,
+      "step": 3870
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.12884116172790527,
+      "learning_rate": 7.402597402597404e-05,
+      "loss": 0.2413,
+      "step": 3880
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.11188474297523499,
+      "learning_rate": 7.37012987012987e-05,
+      "loss": 0.2308,
+      "step": 3890
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.11030302196741104,
+      "learning_rate": 7.337662337662338e-05,
+      "loss": 0.2283,
+      "step": 3900
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.10029084980487823,
+      "learning_rate": 7.305194805194807e-05,
+      "loss": 0.2361,
+      "step": 3910
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.11051477491855621,
+      "learning_rate": 7.272727272727273e-05,
+      "loss": 0.2301,
+      "step": 3920
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.1116577535867691,
+      "learning_rate": 7.24025974025974e-05,
+      "loss": 0.2358,
+      "step": 3930
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.09377572685480118,
+      "learning_rate": 7.207792207792208e-05,
+      "loss": 0.223,
+      "step": 3940
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.13364656269550323,
+      "learning_rate": 7.175324675324676e-05,
+      "loss": 0.2388,
+      "step": 3950
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.14000515639781952,
+      "learning_rate": 7.142857142857143e-05,
+      "loss": 0.2414,
+      "step": 3960
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.11138896644115448,
+      "learning_rate": 7.110389610389611e-05,
+      "loss": 0.229,
+      "step": 3970
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.1375998854637146,
+      "learning_rate": 7.077922077922077e-05,
+      "loss": 0.2414,
+      "step": 3980
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.12048943340778351,
+      "learning_rate": 7.045454545454546e-05,
+      "loss": 0.241,
+      "step": 3990
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.10084032267332077,
+      "learning_rate": 7.012987012987014e-05,
+      "loss": 0.2327,
+      "step": 4000
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.11470773071050644,
+      "learning_rate": 6.98051948051948e-05,
+      "loss": 0.2249,
+      "step": 4010
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.1473914235830307,
+      "learning_rate": 6.948051948051948e-05,
+      "loss": 0.2293,
+      "step": 4020
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.12179002910852432,
+      "learning_rate": 6.915584415584417e-05,
+      "loss": 0.232,
+      "step": 4030
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.10279525071382523,
+      "learning_rate": 6.883116883116883e-05,
+      "loss": 0.2267,
+      "step": 4040
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.12297099083662033,
+      "learning_rate": 6.850649350649351e-05,
+      "loss": 0.2384,
+      "step": 4050
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.10803364217281342,
+      "learning_rate": 6.818181818181818e-05,
+      "loss": 0.2234,
+      "step": 4060
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.10421587526798248,
+      "learning_rate": 6.785714285714286e-05,
+      "loss": 0.2311,
+      "step": 4070
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.10371170192956924,
+      "learning_rate": 6.753246753246754e-05,
+      "loss": 0.2321,
+      "step": 4080
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.09133719652891159,
+      "learning_rate": 6.720779220779221e-05,
+      "loss": 0.2279,
+      "step": 4090
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.12619513273239136,
+      "learning_rate": 6.688311688311688e-05,
+      "loss": 0.2222,
+      "step": 4100
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.11133179813623428,
+      "learning_rate": 6.655844155844157e-05,
+      "loss": 0.2261,
+      "step": 4110
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.10881369560956955,
+      "learning_rate": 6.623376623376624e-05,
+      "loss": 0.2262,
+      "step": 4120
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.11743563413619995,
+      "learning_rate": 6.59090909090909e-05,
+      "loss": 0.2263,
+      "step": 4130
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.1423397809267044,
+      "learning_rate": 6.55844155844156e-05,
+      "loss": 0.2397,
+      "step": 4140
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.11401114612817764,
+      "learning_rate": 6.525974025974026e-05,
+      "loss": 0.2279,
+      "step": 4150
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.11780420690774918,
+      "learning_rate": 6.493506493506494e-05,
+      "loss": 0.233,
+      "step": 4160
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.11325237900018692,
+      "learning_rate": 6.461038961038961e-05,
+      "loss": 0.2254,
+      "step": 4170
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.10473790764808655,
+      "learning_rate": 6.428571428571429e-05,
+      "loss": 0.2286,
+      "step": 4180
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.10390175879001617,
+      "learning_rate": 6.396103896103896e-05,
+      "loss": 0.2236,
+      "step": 4190
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.10519735515117645,
+      "learning_rate": 6.363636363636364e-05,
+      "loss": 0.2349,
+      "step": 4200
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.10875875502824783,
+      "learning_rate": 6.331168831168832e-05,
+      "loss": 0.2289,
+      "step": 4210
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.1587696373462677,
+      "learning_rate": 6.2987012987013e-05,
+      "loss": 0.2261,
+      "step": 4220
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.12190727144479752,
+      "learning_rate": 6.266233766233767e-05,
+      "loss": 0.2275,
+      "step": 4230
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.11576837301254272,
+      "learning_rate": 6.233766233766233e-05,
+      "loss": 0.2292,
+      "step": 4240
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.14066992700099945,
+      "learning_rate": 6.201298701298701e-05,
+      "loss": 0.2244,
+      "step": 4250
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.11332803219556808,
+      "learning_rate": 6.16883116883117e-05,
+      "loss": 0.2358,
+      "step": 4260
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.10989506542682648,
+      "learning_rate": 6.136363636363636e-05,
+      "loss": 0.2296,
+      "step": 4270
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.1303674578666687,
+      "learning_rate": 6.103896103896104e-05,
+      "loss": 0.2336,
+      "step": 4280
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.1250188648700714,
+      "learning_rate": 6.0714285714285715e-05,
+      "loss": 0.2392,
+      "step": 4290
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.21410736441612244,
+      "learning_rate": 6.03896103896104e-05,
+      "loss": 0.2319,
+      "step": 4300
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.1308501809835434,
+      "learning_rate": 6.006493506493507e-05,
+      "loss": 0.2433,
+      "step": 4310
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.11900628358125687,
+      "learning_rate": 5.9740259740259744e-05,
+      "loss": 0.2257,
+      "step": 4320
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.12066376209259033,
+      "learning_rate": 5.9415584415584414e-05,
+      "loss": 0.2336,
+      "step": 4330
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.09623558074235916,
+      "learning_rate": 5.90909090909091e-05,
+      "loss": 0.2276,
+      "step": 4340
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.1280377358198166,
+      "learning_rate": 5.8766233766233766e-05,
+      "loss": 0.2267,
+      "step": 4350
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.10485294461250305,
+      "learning_rate": 5.844155844155844e-05,
+      "loss": 0.2271,
+      "step": 4360
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.10915611684322357,
+      "learning_rate": 5.8116883116883126e-05,
+      "loss": 0.2426,
+      "step": 4370
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.12123510241508484,
+      "learning_rate": 5.7792207792207796e-05,
+      "loss": 0.2304,
+      "step": 4380
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.10410697013139725,
+      "learning_rate": 5.746753246753247e-05,
+      "loss": 0.2279,
+      "step": 4390
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.1267937868833542,
+      "learning_rate": 5.714285714285714e-05,
+      "loss": 0.2357,
+      "step": 4400
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.14639367163181305,
+      "learning_rate": 5.6818181818181825e-05,
+      "loss": 0.2308,
+      "step": 4410
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.11761369556188583,
+      "learning_rate": 5.64935064935065e-05,
+      "loss": 0.2318,
+      "step": 4420
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.1099245548248291,
+      "learning_rate": 5.616883116883117e-05,
+      "loss": 0.2399,
+      "step": 4430
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.13074526190757751,
+      "learning_rate": 5.584415584415584e-05,
+      "loss": 0.2318,
+      "step": 4440
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.11614665389060974,
+      "learning_rate": 5.5519480519480524e-05,
+      "loss": 0.2435,
+      "step": 4450
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.16329911351203918,
+      "learning_rate": 5.51948051948052e-05,
+      "loss": 0.2296,
+      "step": 4460
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.10937174409627914,
+      "learning_rate": 5.487012987012987e-05,
+      "loss": 0.2415,
+      "step": 4470
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.10211482644081116,
+      "learning_rate": 5.4545454545454546e-05,
+      "loss": 0.224,
+      "step": 4480
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.1312587410211563,
+      "learning_rate": 5.422077922077923e-05,
+      "loss": 0.2252,
+      "step": 4490
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.1484302431344986,
+      "learning_rate": 5.38961038961039e-05,
+      "loss": 0.2321,
+      "step": 4500
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.13507643342018127,
+      "learning_rate": 5.3571428571428575e-05,
+      "loss": 0.2384,
+      "step": 4510
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.124076709151268,
+      "learning_rate": 5.3246753246753245e-05,
+      "loss": 0.2308,
+      "step": 4520
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.11308033764362335,
+      "learning_rate": 5.292207792207793e-05,
+      "loss": 0.2229,
+      "step": 4530
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.11671579629182816,
+      "learning_rate": 5.25974025974026e-05,
+      "loss": 0.2256,
+      "step": 4540
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.11172166466712952,
+      "learning_rate": 5.2272727272727274e-05,
+      "loss": 0.2298,
+      "step": 4550
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.19812701642513275,
+      "learning_rate": 5.1948051948051944e-05,
+      "loss": 0.227,
+      "step": 4560
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.10457318276166916,
+      "learning_rate": 5.162337662337663e-05,
+      "loss": 0.224,
+      "step": 4570
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.13775014877319336,
+      "learning_rate": 5.1298701298701304e-05,
+      "loss": 0.2232,
+      "step": 4580
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.10171498358249664,
+      "learning_rate": 5.097402597402597e-05,
+      "loss": 0.2408,
+      "step": 4590
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.1511399894952774,
+      "learning_rate": 5.064935064935065e-05,
+      "loss": 0.2324,
+      "step": 4600
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.11420062929391861,
+      "learning_rate": 5.032467532467533e-05,
+      "loss": 0.2238,
+      "step": 4610
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.12045635282993317,
+      "learning_rate": 5e-05,
+      "loss": 0.2302,
+      "step": 4620
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.14395587146282196,
+      "learning_rate": 4.967532467532468e-05,
+      "loss": 0.2276,
+      "step": 4630
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.09929464012384415,
+      "learning_rate": 4.9350649350649355e-05,
+      "loss": 0.2358,
+      "step": 4640
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.14118945598602295,
+      "learning_rate": 4.902597402597403e-05,
+      "loss": 0.2411,
+      "step": 4650
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.13900473713874817,
+      "learning_rate": 4.87012987012987e-05,
+      "loss": 0.2228,
+      "step": 4660
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.10782819241285324,
+      "learning_rate": 4.8376623376623384e-05,
+      "loss": 0.2295,
+      "step": 4670
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.126652792096138,
+      "learning_rate": 4.8051948051948054e-05,
+      "loss": 0.2217,
+      "step": 4680
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.10030221194028854,
+      "learning_rate": 4.772727272727273e-05,
+      "loss": 0.2245,
+      "step": 4690
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.10515980422496796,
+      "learning_rate": 4.740259740259741e-05,
+      "loss": 0.2258,
+      "step": 4700
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.10597904771566391,
+      "learning_rate": 4.707792207792208e-05,
+      "loss": 0.23,
+      "step": 4710
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.12041573226451874,
+      "learning_rate": 4.675324675324675e-05,
+      "loss": 0.223,
+      "step": 4720
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.09887034446001053,
+      "learning_rate": 4.642857142857143e-05,
+      "loss": 0.2247,
+      "step": 4730
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.11537348479032516,
+      "learning_rate": 4.6103896103896106e-05,
+      "loss": 0.2228,
+      "step": 4740
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.12704965472221375,
+      "learning_rate": 4.577922077922078e-05,
+      "loss": 0.2271,
+      "step": 4750
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.12509785592556,
+      "learning_rate": 4.545454545454546e-05,
+      "loss": 0.2282,
+      "step": 4760
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.10524313896894455,
+      "learning_rate": 4.5129870129870135e-05,
+      "loss": 0.2295,
+      "step": 4770
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.10979527235031128,
+      "learning_rate": 4.4805194805194805e-05,
+      "loss": 0.232,
+      "step": 4780
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.10340957343578339,
+      "learning_rate": 4.448051948051948e-05,
+      "loss": 0.2322,
+      "step": 4790
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.11665787547826767,
+      "learning_rate": 4.415584415584416e-05,
+      "loss": 0.2261,
+      "step": 4800
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.11915656179189682,
+      "learning_rate": 4.3831168831168834e-05,
+      "loss": 0.2466,
+      "step": 4810
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.10494299978017807,
+      "learning_rate": 4.3506493506493503e-05,
+      "loss": 0.2266,
+      "step": 4820
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.10959453880786896,
+      "learning_rate": 4.318181818181819e-05,
+      "loss": 0.2268,
+      "step": 4830
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.10894802957773209,
+      "learning_rate": 4.2857142857142856e-05,
+      "loss": 0.2306,
+      "step": 4840
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.12208307534456253,
+      "learning_rate": 4.253246753246753e-05,
+      "loss": 0.2307,
+      "step": 4850
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.11236707121133804,
+      "learning_rate": 4.220779220779221e-05,
+      "loss": 0.2293,
+      "step": 4860
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.12207558751106262,
+      "learning_rate": 4.1883116883116886e-05,
+      "loss": 0.2375,
+      "step": 4870
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.10055721551179886,
+      "learning_rate": 4.155844155844156e-05,
+      "loss": 0.2383,
+      "step": 4880
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.11718643456697464,
+      "learning_rate": 4.123376623376624e-05,
+      "loss": 0.2359,
+      "step": 4890
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.09924653172492981,
+      "learning_rate": 4.0909090909090915e-05,
+      "loss": 0.2286,
+      "step": 4900
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.12111229449510574,
+      "learning_rate": 4.0584415584415584e-05,
+      "loss": 0.2219,
+      "step": 4910
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.1255057007074356,
+      "learning_rate": 4.025974025974026e-05,
+      "loss": 0.2305,
+      "step": 4920
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.10687470436096191,
+      "learning_rate": 3.993506493506494e-05,
+      "loss": 0.2284,
+      "step": 4930
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.11906068027019501,
+      "learning_rate": 3.9610389610389614e-05,
+      "loss": 0.2304,
+      "step": 4940
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.13518692553043365,
+      "learning_rate": 3.928571428571429e-05,
+      "loss": 0.2377,
+      "step": 4950
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.10135382413864136,
+      "learning_rate": 3.8961038961038966e-05,
+      "loss": 0.2271,
+      "step": 4960
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.09538806229829788,
+      "learning_rate": 3.8636363636363636e-05,
+      "loss": 0.2246,
+      "step": 4970
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.12435057014226913,
+      "learning_rate": 3.831168831168831e-05,
+      "loss": 0.2305,
+      "step": 4980
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.11401098221540451,
+      "learning_rate": 3.798701298701299e-05,
+      "loss": 0.2308,
+      "step": 4990
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.12875626981258392,
+      "learning_rate": 3.7662337662337665e-05,
+      "loss": 0.233,
+      "step": 5000
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.1109662801027298,
+      "learning_rate": 3.7337662337662335e-05,
+      "loss": 0.2331,
+      "step": 5010
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.10425057262182236,
+      "learning_rate": 3.701298701298702e-05,
+      "loss": 0.2356,
+      "step": 5020
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.11894793808460236,
+      "learning_rate": 3.668831168831169e-05,
+      "loss": 0.2334,
+      "step": 5030
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.1109088659286499,
+      "learning_rate": 3.6363636363636364e-05,
+      "loss": 0.2247,
+      "step": 5040
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.15378132462501526,
+      "learning_rate": 3.603896103896104e-05,
+      "loss": 0.2349,
+      "step": 5050
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.14902709424495697,
+      "learning_rate": 3.571428571428572e-05,
+      "loss": 0.2289,
+      "step": 5060
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.09531041979789734,
+      "learning_rate": 3.5389610389610387e-05,
+      "loss": 0.2205,
+      "step": 5070
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 0.11934591829776764,
+      "learning_rate": 3.506493506493507e-05,
+      "loss": 0.2323,
+      "step": 5080
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.11388123780488968,
+      "learning_rate": 3.474025974025974e-05,
+      "loss": 0.2285,
+      "step": 5090
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.11454147845506668,
+      "learning_rate": 3.4415584415584416e-05,
+      "loss": 0.2225,
+      "step": 5100
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.13935193419456482,
+      "learning_rate": 3.409090909090909e-05,
+      "loss": 0.2285,
+      "step": 5110
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.10745522379875183,
+      "learning_rate": 3.376623376623377e-05,
+      "loss": 0.2364,
+      "step": 5120
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.12464471161365509,
+      "learning_rate": 3.344155844155844e-05,
+      "loss": 0.2361,
+      "step": 5130
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.14242151379585266,
+      "learning_rate": 3.311688311688312e-05,
+      "loss": 0.238,
+      "step": 5140
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.1100350096821785,
+      "learning_rate": 3.27922077922078e-05,
+      "loss": 0.2307,
+      "step": 5150
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.10302827507257462,
+      "learning_rate": 3.246753246753247e-05,
+      "loss": 0.2281,
+      "step": 5160
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.12884677946567535,
+      "learning_rate": 3.2142857142857144e-05,
+      "loss": 0.2261,
+      "step": 5170
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.11658567190170288,
+      "learning_rate": 3.181818181818182e-05,
+      "loss": 0.2211,
+      "step": 5180
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.127898171544075,
+      "learning_rate": 3.14935064935065e-05,
+      "loss": 0.2287,
+      "step": 5190
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 0.11611378192901611,
+      "learning_rate": 3.1168831168831166e-05,
+      "loss": 0.2205,
+      "step": 5200
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.11631160974502563,
+      "learning_rate": 3.084415584415585e-05,
+      "loss": 0.2286,
+      "step": 5210
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.10242964327335358,
+      "learning_rate": 3.051948051948052e-05,
+      "loss": 0.2253,
+      "step": 5220
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.10427073389291763,
+      "learning_rate": 3.01948051948052e-05,
+      "loss": 0.2288,
+      "step": 5230
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.09799051284790039,
+      "learning_rate": 2.9870129870129872e-05,
+      "loss": 0.2346,
+      "step": 5240
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.12969501316547394,
+      "learning_rate": 2.954545454545455e-05,
+      "loss": 0.2308,
+      "step": 5250
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.15084834396839142,
+      "learning_rate": 2.922077922077922e-05,
+      "loss": 0.2269,
+      "step": 5260
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.10667992383241653,
+      "learning_rate": 2.8896103896103898e-05,
+      "loss": 0.2274,
+      "step": 5270
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.14059442281723022,
+      "learning_rate": 2.857142857142857e-05,
+      "loss": 0.2312,
+      "step": 5280
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.15748588740825653,
+      "learning_rate": 2.824675324675325e-05,
+      "loss": 0.2309,
+      "step": 5290
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.1318967044353485,
+      "learning_rate": 2.792207792207792e-05,
+      "loss": 0.2339,
+      "step": 5300
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.12507110834121704,
+      "learning_rate": 2.75974025974026e-05,
+      "loss": 0.2234,
+      "step": 5310
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 0.14823444187641144,
+      "learning_rate": 2.7272727272727273e-05,
+      "loss": 0.2318,
+      "step": 5320
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.11165734380483627,
+      "learning_rate": 2.694805194805195e-05,
+      "loss": 0.2229,
+      "step": 5330
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.13370725512504578,
+      "learning_rate": 2.6623376623376623e-05,
+      "loss": 0.2237,
+      "step": 5340
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.10128387063741684,
+      "learning_rate": 2.62987012987013e-05,
+      "loss": 0.2278,
+      "step": 5350
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.1412563920021057,
+      "learning_rate": 2.5974025974025972e-05,
+      "loss": 0.2398,
+      "step": 5360
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.13788406550884247,
+      "learning_rate": 2.5649350649350652e-05,
+      "loss": 0.2427,
+      "step": 5370
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 0.11670909076929092,
+      "learning_rate": 2.5324675324675325e-05,
+      "loss": 0.2267,
+      "step": 5380
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.11005653440952301,
+      "learning_rate": 2.5e-05,
+      "loss": 0.2259,
+      "step": 5390
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.14962653815746307,
+      "learning_rate": 2.4675324675324678e-05,
+      "loss": 0.224,
+      "step": 5400
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.14623549580574036,
+      "learning_rate": 2.435064935064935e-05,
+      "loss": 0.2294,
+      "step": 5410
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.11526806652545929,
+      "learning_rate": 2.4025974025974027e-05,
+      "loss": 0.2351,
+      "step": 5420
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.12139612436294556,
+      "learning_rate": 2.3701298701298703e-05,
+      "loss": 0.2241,
+      "step": 5430
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.14310401678085327,
+      "learning_rate": 2.3376623376623376e-05,
+      "loss": 0.2388,
+      "step": 5440
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.12777364253997803,
+      "learning_rate": 2.3051948051948053e-05,
+      "loss": 0.2254,
+      "step": 5450
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.10837443172931671,
+      "learning_rate": 2.272727272727273e-05,
+      "loss": 0.2261,
+      "step": 5460
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.10602117329835892,
+      "learning_rate": 2.2402597402597402e-05,
+      "loss": 0.2202,
+      "step": 5470
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.11905395984649658,
+      "learning_rate": 2.207792207792208e-05,
+      "loss": 0.2332,
+      "step": 5480
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.10377994924783707,
+      "learning_rate": 2.1753246753246752e-05,
+      "loss": 0.225,
+      "step": 5490
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.10552787780761719,
+      "learning_rate": 2.1428571428571428e-05,
+      "loss": 0.2297,
+      "step": 5500
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.10886229574680328,
+      "learning_rate": 2.1103896103896105e-05,
+      "loss": 0.2259,
+      "step": 5510
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.13297457993030548,
+      "learning_rate": 2.077922077922078e-05,
+      "loss": 0.2233,
+      "step": 5520
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.11828389763832092,
+      "learning_rate": 2.0454545454545457e-05,
+      "loss": 0.2349,
+      "step": 5530
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.1609041392803192,
+      "learning_rate": 2.012987012987013e-05,
+      "loss": 0.2255,
+      "step": 5540
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.1048947125673294,
+      "learning_rate": 1.9805194805194807e-05,
+      "loss": 0.2205,
+      "step": 5550
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.10608299821615219,
+      "learning_rate": 1.9480519480519483e-05,
+      "loss": 0.2377,
+      "step": 5560
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 0.14137326180934906,
+      "learning_rate": 1.9155844155844156e-05,
+      "loss": 0.2406,
+      "step": 5570
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.10666473954916,
+      "learning_rate": 1.8831168831168833e-05,
+      "loss": 0.231,
+      "step": 5580
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.10281306505203247,
+      "learning_rate": 1.850649350649351e-05,
+      "loss": 0.2287,
+      "step": 5590
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.12429114431142807,
+      "learning_rate": 1.8181818181818182e-05,
+      "loss": 0.227,
+      "step": 5600
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.12389946728944778,
+      "learning_rate": 1.785714285714286e-05,
+      "loss": 0.2266,
+      "step": 5610
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.14057426154613495,
+      "learning_rate": 1.7532467532467535e-05,
+      "loss": 0.2385,
+      "step": 5620
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 0.09822268038988113,
+      "learning_rate": 1.7207792207792208e-05,
+      "loss": 0.2284,
+      "step": 5630
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.09939773380756378,
+      "learning_rate": 1.6883116883116884e-05,
+      "loss": 0.2213,
+      "step": 5640
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.1136704757809639,
+      "learning_rate": 1.655844155844156e-05,
+      "loss": 0.2248,
+      "step": 5650
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.10240314155817032,
+      "learning_rate": 1.6233766233766234e-05,
+      "loss": 0.2229,
+      "step": 5660
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.10413914173841476,
+      "learning_rate": 1.590909090909091e-05,
+      "loss": 0.2243,
+      "step": 5670
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.11324857175350189,
+      "learning_rate": 1.5584415584415583e-05,
+      "loss": 0.2197,
+      "step": 5680
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 0.11112158745527267,
+      "learning_rate": 1.525974025974026e-05,
+      "loss": 0.2282,
+      "step": 5690
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.1014389768242836,
+      "learning_rate": 1.4935064935064936e-05,
+      "loss": 0.2262,
+      "step": 5700
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.12159611284732819,
+      "learning_rate": 1.461038961038961e-05,
+      "loss": 0.231,
+      "step": 5710
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.10437947511672974,
+      "learning_rate": 1.4285714285714285e-05,
+      "loss": 0.2263,
+      "step": 5720
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.13253456354141235,
+      "learning_rate": 1.396103896103896e-05,
+      "loss": 0.2182,
+      "step": 5730
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.15731987357139587,
+      "learning_rate": 1.3636363636363637e-05,
+      "loss": 0.2255,
+      "step": 5740
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.1070665642619133,
+      "learning_rate": 1.3311688311688311e-05,
+      "loss": 0.2344,
+      "step": 5750
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.1093362346291542,
+      "learning_rate": 1.2987012987012986e-05,
+      "loss": 0.2312,
+      "step": 5760
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.10909686982631683,
+      "learning_rate": 1.2662337662337662e-05,
+      "loss": 0.217,
+      "step": 5770
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.17574919760227203,
+      "learning_rate": 1.2337662337662339e-05,
+      "loss": 0.2309,
+      "step": 5780
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.11184772849082947,
+      "learning_rate": 1.2012987012987014e-05,
+      "loss": 0.2278,
+      "step": 5790
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.12411190569400787,
+      "learning_rate": 1.1688311688311688e-05,
+      "loss": 0.2244,
+      "step": 5800
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.1730908751487732,
+      "learning_rate": 1.1363636363636365e-05,
+      "loss": 0.2308,
+      "step": 5810
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 0.13045470416545868,
+      "learning_rate": 1.103896103896104e-05,
+      "loss": 0.2316,
+      "step": 5820
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.14824289083480835,
+      "learning_rate": 1.0714285714285714e-05,
+      "loss": 0.2283,
+      "step": 5830
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.12188534438610077,
+      "learning_rate": 1.038961038961039e-05,
+      "loss": 0.2306,
+      "step": 5840
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.11303505301475525,
+      "learning_rate": 1.0064935064935065e-05,
+      "loss": 0.2223,
+      "step": 5850
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.10529917478561401,
+      "learning_rate": 9.740259740259742e-06,
+      "loss": 0.2289,
+      "step": 5860
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.10578101873397827,
+      "learning_rate": 9.415584415584416e-06,
+      "loss": 0.2228,
+      "step": 5870
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 0.16038714349269867,
+      "learning_rate": 9.090909090909091e-06,
+      "loss": 0.2342,
+      "step": 5880
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.1213398203253746,
+      "learning_rate": 8.766233766233767e-06,
+      "loss": 0.2386,
+      "step": 5890
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.15008947253227234,
+      "learning_rate": 8.441558441558442e-06,
+      "loss": 0.2432,
+      "step": 5900
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.12029566615819931,
+      "learning_rate": 8.116883116883117e-06,
+      "loss": 0.2185,
+      "step": 5910
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.14414188265800476,
+      "learning_rate": 7.792207792207792e-06,
+      "loss": 0.2372,
+      "step": 5920
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.10464475303888321,
+      "learning_rate": 7.467532467532468e-06,
+      "loss": 0.2251,
+      "step": 5930
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.12148386240005493,
+      "learning_rate": 7.142857142857143e-06,
+      "loss": 0.2288,
+      "step": 5940
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.17013368010520935,
+      "learning_rate": 6.818181818181818e-06,
+      "loss": 0.2274,
+      "step": 5950
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.10602659732103348,
+      "learning_rate": 6.493506493506493e-06,
+      "loss": 0.2289,
+      "step": 5960
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.1121257022023201,
+      "learning_rate": 6.168831168831169e-06,
+      "loss": 0.2322,
+      "step": 5970
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.12117767333984375,
+      "learning_rate": 5.844155844155844e-06,
+      "loss": 0.2325,
+      "step": 5980
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.10778909921646118,
+      "learning_rate": 5.51948051948052e-06,
+      "loss": 0.2239,
+      "step": 5990
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 0.15755324065685272,
+      "learning_rate": 5.194805194805195e-06,
+      "loss": 0.2263,
+      "step": 6000
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.1577947735786438,
+      "learning_rate": 4.870129870129871e-06,
+      "loss": 0.2219,
+      "step": 6010
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.1447291076183319,
+      "learning_rate": 4.5454545454545455e-06,
+      "loss": 0.2362,
+      "step": 6020
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.18480731546878815,
+      "learning_rate": 4.220779220779221e-06,
+      "loss": 0.2301,
+      "step": 6030
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.11400249600410461,
+      "learning_rate": 3.896103896103896e-06,
+      "loss": 0.233,
+      "step": 6040
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.1243402361869812,
+      "learning_rate": 3.5714285714285714e-06,
+      "loss": 0.2204,
+      "step": 6050
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 0.10859899967908859,
+      "learning_rate": 3.2467532467532465e-06,
+      "loss": 0.2236,
+      "step": 6060
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.11620640754699707,
+      "learning_rate": 2.922077922077922e-06,
+      "loss": 0.2248,
+      "step": 6070
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.12846139073371887,
+      "learning_rate": 2.5974025974025976e-06,
+      "loss": 0.2245,
+      "step": 6080
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.10820233076810837,
+      "learning_rate": 2.2727272727272728e-06,
+      "loss": 0.2334,
+      "step": 6090
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.12337440997362137,
+      "learning_rate": 1.948051948051948e-06,
+      "loss": 0.2226,
+      "step": 6100
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.13436639308929443,
+      "learning_rate": 1.6233766233766232e-06,
+      "loss": 0.2288,
+      "step": 6110
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 0.1430235058069229,
+      "learning_rate": 1.2987012987012988e-06,
+      "loss": 0.2317,
+      "step": 6120
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.17122237384319305,
+      "learning_rate": 9.74025974025974e-07,
+      "loss": 0.2281,
+      "step": 6130
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.13612103462219238,
+      "learning_rate": 6.493506493506494e-07,
+      "loss": 0.2246,
+      "step": 6140
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.12664946913719177,
+      "learning_rate": 3.246753246753247e-07,
+      "loss": 0.2309,
+      "step": 6150
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.21350349485874176,
+      "learning_rate": 0.0,
+      "loss": 0.2254,
+      "step": 6160
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 6160,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 500,
+  "total_flos": 4.845807449387827e+18,
+  "train_batch_size": 32,
+  "trial_name": null,
+  "trial_params": null
+}