diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,9471 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 50.0,
+  "eval_steps": 500,
+  "global_step": 13500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.20531712472438812,
+      "learning_rate": 0.001998518518518519,
+      "loss": 2.6189,
+      "step": 10
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.22602832317352295,
+      "learning_rate": 0.001997037037037037,
+      "loss": 2.5836,
+      "step": 20
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.21022845804691315,
+      "learning_rate": 0.0019955555555555555,
+      "loss": 2.5439,
+      "step": 30
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.17906232178211212,
+      "learning_rate": 0.0019940740740740743,
+      "loss": 2.6242,
+      "step": 40
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.19077719748020172,
+      "learning_rate": 0.0019925925925925927,
+      "loss": 2.5929,
+      "step": 50
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.25262829661369324,
+      "learning_rate": 0.001991111111111111,
+      "loss": 2.5714,
+      "step": 60
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.32239940762519836,
+      "learning_rate": 0.00198962962962963,
+      "loss": 2.5695,
+      "step": 70
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.17224617302417755,
+      "learning_rate": 0.001988148148148148,
+      "loss": 2.5651,
+      "step": 80
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.17106947302818298,
+      "learning_rate": 0.0019866666666666665,
+      "loss": 2.6357,
+      "step": 90
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.1687716841697693,
+      "learning_rate": 0.0019851851851851853,
+      "loss": 2.5754,
+      "step": 100
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.17471371591091156,
+      "learning_rate": 0.0019837037037037037,
+      "loss": 2.5687,
+      "step": 110
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.16484995186328888,
+      "learning_rate": 0.0019822222222222225,
+      "loss": 2.5992,
+      "step": 120
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.16195523738861084,
+      "learning_rate": 0.001980740740740741,
+      "loss": 2.6108,
+      "step": 130
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.1896306574344635,
+      "learning_rate": 0.001979259259259259,
+      "loss": 2.6148,
+      "step": 140
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.1916884034872055,
+      "learning_rate": 0.001977777777777778,
+      "loss": 2.589,
+      "step": 150
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.20058733224868774,
+      "learning_rate": 0.0019762962962962963,
+      "loss": 2.5799,
+      "step": 160
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.16014626622200012,
+      "learning_rate": 0.0019748148148148147,
+      "loss": 2.5529,
+      "step": 170
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.15680766105651855,
+      "learning_rate": 0.0019733333333333334,
+      "loss": 2.562,
+      "step": 180
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.17019112408161163,
+      "learning_rate": 0.001971851851851852,
+      "loss": 2.579,
+      "step": 190
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.16960741579532623,
+      "learning_rate": 0.00197037037037037,
+      "loss": 2.5542,
+      "step": 200
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.17356687784194946,
+      "learning_rate": 0.001968888888888889,
+      "loss": 2.5641,
+      "step": 210
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.17307817935943604,
+      "learning_rate": 0.0019674074074074077,
+      "loss": 2.5962,
+      "step": 220
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.1674201637506485,
+      "learning_rate": 0.001965925925925926,
+      "loss": 2.5518,
+      "step": 230
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 0.15180547535419464,
+      "learning_rate": 0.0019644444444444444,
+      "loss": 2.5919,
+      "step": 240
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 0.166998952627182,
+      "learning_rate": 0.0019629629629629632,
+      "loss": 2.5684,
+      "step": 250
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.16316364705562592,
+      "learning_rate": 0.0019614814814814816,
+      "loss": 2.603,
+      "step": 260
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.2779152989387512,
+      "learning_rate": 0.00196,
+      "loss": 2.6232,
+      "step": 270
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 0.17026250064373016,
+      "learning_rate": 0.0019585185185185187,
+      "loss": 2.4226,
+      "step": 280
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 0.16963163018226624,
+      "learning_rate": 0.001957037037037037,
+      "loss": 2.4307,
+      "step": 290
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 0.1799466907978058,
+      "learning_rate": 0.0019555555555555554,
+      "loss": 2.4621,
+      "step": 300
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 0.17876045405864716,
+      "learning_rate": 0.0019540740740740742,
+      "loss": 2.4945,
+      "step": 310
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 0.1827176958322525,
+      "learning_rate": 0.0019525925925925928,
+      "loss": 2.4547,
+      "step": 320
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 0.20249758660793304,
+      "learning_rate": 0.0019511111111111111,
+      "loss": 2.4662,
+      "step": 330
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 0.19642537832260132,
+      "learning_rate": 0.0019496296296296297,
+      "loss": 2.5019,
+      "step": 340
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 0.2031635344028473,
+      "learning_rate": 0.0019481481481481483,
+      "loss": 2.4459,
+      "step": 350
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 0.20372791588306427,
+      "learning_rate": 0.0019466666666666669,
+      "loss": 2.477,
+      "step": 360
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 0.20827481150627136,
+      "learning_rate": 0.0019451851851851852,
+      "loss": 2.5037,
+      "step": 370
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 0.1974320113658905,
+      "learning_rate": 0.0019437037037037038,
+      "loss": 2.5155,
+      "step": 380
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.2353196144104004,
+      "learning_rate": 0.0019422222222222224,
+      "loss": 2.5645,
+      "step": 390
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 0.22184617817401886,
+      "learning_rate": 0.0019407407407407407,
+      "loss": 2.4933,
+      "step": 400
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 0.20222437381744385,
+      "learning_rate": 0.0019392592592592593,
+      "loss": 2.4861,
+      "step": 410
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 0.19691778719425201,
+      "learning_rate": 0.0019377777777777778,
+      "loss": 2.4869,
+      "step": 420
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 0.19557970762252808,
+      "learning_rate": 0.0019362962962962964,
+      "loss": 2.4886,
+      "step": 430
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 0.18064908683300018,
+      "learning_rate": 0.0019348148148148148,
+      "loss": 2.5016,
+      "step": 440
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 0.20412351191043854,
+      "learning_rate": 0.0019333333333333333,
+      "loss": 2.5193,
+      "step": 450
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 0.19166867434978485,
+      "learning_rate": 0.001931851851851852,
+      "loss": 2.5526,
+      "step": 460
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 0.18850873410701752,
+      "learning_rate": 0.0019303703703703703,
+      "loss": 2.5297,
+      "step": 470
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 0.1931956559419632,
+      "learning_rate": 0.0019288888888888888,
+      "loss": 2.5621,
+      "step": 480
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 0.20721589028835297,
+      "learning_rate": 0.0019274074074074074,
+      "loss": 2.5265,
+      "step": 490
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 0.17977182567119598,
+      "learning_rate": 0.0019259259259259258,
+      "loss": 2.4947,
+      "step": 500
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 0.19090428948402405,
+      "learning_rate": 0.0019244444444444443,
+      "loss": 2.5391,
+      "step": 510
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 0.1727808117866516,
+      "learning_rate": 0.0019229629629629631,
+      "loss": 2.5453,
+      "step": 520
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 0.20896504819393158,
+      "learning_rate": 0.0019214814814814817,
+      "loss": 2.5575,
+      "step": 530
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.3333798050880432,
+      "learning_rate": 0.00192,
+      "loss": 2.506,
+      "step": 540
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 0.2123243808746338,
+      "learning_rate": 0.0019185185185185186,
+      "loss": 2.3596,
+      "step": 550
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 0.20756013691425323,
+      "learning_rate": 0.0019170370370370372,
+      "loss": 2.3598,
+      "step": 560
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 0.25747150182724,
+      "learning_rate": 0.0019155555555555555,
+      "loss": 2.3519,
+      "step": 570
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 0.21804387867450714,
+      "learning_rate": 0.0019140740740740741,
+      "loss": 2.3626,
+      "step": 580
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 0.21577107906341553,
+      "learning_rate": 0.0019125925925925927,
+      "loss": 2.3083,
+      "step": 590
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 0.2492288202047348,
+      "learning_rate": 0.0019111111111111113,
+      "loss": 2.3581,
+      "step": 600
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 0.26941046118736267,
+      "learning_rate": 0.0019096296296296296,
+      "loss": 2.3776,
+      "step": 610
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 0.21193154156208038,
+      "learning_rate": 0.0019081481481481482,
+      "loss": 2.4152,
+      "step": 620
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 0.2306254655122757,
+      "learning_rate": 0.0019066666666666668,
+      "loss": 2.3869,
+      "step": 630
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 0.2227521389722824,
+      "learning_rate": 0.001905185185185185,
+      "loss": 2.3928,
+      "step": 640
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 0.2545499801635742,
+      "learning_rate": 0.0019037037037037037,
+      "loss": 2.3672,
+      "step": 650
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 0.24774783849716187,
+      "learning_rate": 0.0019022222222222222,
+      "loss": 2.3561,
+      "step": 660
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 0.22088710963726044,
+      "learning_rate": 0.0019007407407407408,
+      "loss": 2.4168,
+      "step": 670
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 0.2693859338760376,
+      "learning_rate": 0.0018992592592592592,
+      "loss": 2.4042,
+      "step": 680
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.22340671718120575,
+      "learning_rate": 0.0018977777777777777,
+      "loss": 2.4334,
+      "step": 690
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 0.24626752734184265,
+      "learning_rate": 0.0018962962962962963,
+      "loss": 2.4132,
+      "step": 700
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 0.22658760845661163,
+      "learning_rate": 0.0018948148148148147,
+      "loss": 2.4186,
+      "step": 710
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 0.22482579946517944,
+      "learning_rate": 0.0018933333333333335,
+      "loss": 2.4567,
+      "step": 720
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 0.2270815223455429,
+      "learning_rate": 0.001891851851851852,
+      "loss": 2.4324,
+      "step": 730
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.2389022558927536,
+      "learning_rate": 0.0018903703703703706,
+      "loss": 2.4357,
+      "step": 740
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.2350894808769226,
+      "learning_rate": 0.001888888888888889,
+      "loss": 2.4331,
+      "step": 750
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 0.21408489346504211,
+      "learning_rate": 0.0018874074074074075,
+      "loss": 2.4596,
+      "step": 760
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 0.21101674437522888,
+      "learning_rate": 0.001885925925925926,
+      "loss": 2.4607,
+      "step": 770
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 0.22046928107738495,
+      "learning_rate": 0.0018844444444444444,
+      "loss": 2.4497,
+      "step": 780
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 0.20029981434345245,
+      "learning_rate": 0.001882962962962963,
+      "loss": 2.4479,
+      "step": 790
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 0.20895542204380035,
+      "learning_rate": 0.0018814814814814816,
+      "loss": 2.4687,
+      "step": 800
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.5576801300048828,
+      "learning_rate": 0.00188,
+      "loss": 2.4626,
+      "step": 810
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.2836795449256897,
+      "learning_rate": 0.0018785185185185185,
+      "loss": 2.2582,
+      "step": 820
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.253925085067749,
+      "learning_rate": 0.001877037037037037,
+      "loss": 2.2339,
+      "step": 830
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 0.26034608483314514,
+      "learning_rate": 0.0018755555555555557,
+      "loss": 2.2487,
+      "step": 840
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 0.255710244178772,
+      "learning_rate": 0.001874074074074074,
+      "loss": 2.2417,
+      "step": 850
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 0.3075387477874756,
+      "learning_rate": 0.0018725925925925926,
+      "loss": 2.2657,
+      "step": 860
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 0.2657618224620819,
+      "learning_rate": 0.0018711111111111112,
+      "loss": 2.3137,
+      "step": 870
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 0.2990604639053345,
+      "learning_rate": 0.0018696296296296295,
+      "loss": 2.2845,
+      "step": 880
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 0.24910998344421387,
+      "learning_rate": 0.001868148148148148,
+      "loss": 2.2986,
+      "step": 890
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 0.27236661314964294,
+      "learning_rate": 0.0018666666666666666,
+      "loss": 2.3025,
+      "step": 900
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 0.27428188920021057,
+      "learning_rate": 0.0018651851851851852,
+      "loss": 2.2739,
+      "step": 910
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 0.26426592469215393,
+      "learning_rate": 0.0018637037037037036,
+      "loss": 2.3701,
+      "step": 920
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 0.24107518792152405,
+      "learning_rate": 0.0018622222222222224,
+      "loss": 2.3512,
+      "step": 930
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 0.26364269852638245,
+      "learning_rate": 0.001860740740740741,
+      "loss": 2.3045,
+      "step": 940
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 0.3026023209095001,
+      "learning_rate": 0.0018592592592592593,
+      "loss": 2.3437,
+      "step": 950
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 0.2444707602262497,
+      "learning_rate": 0.0018577777777777779,
+      "loss": 2.3434,
+      "step": 960
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 0.2680145502090454,
+      "learning_rate": 0.0018562962962962964,
+      "loss": 2.3478,
+      "step": 970
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 0.25497835874557495,
+      "learning_rate": 0.001854814814814815,
+      "loss": 2.3419,
+      "step": 980
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 0.2542170584201813,
+      "learning_rate": 0.0018533333333333334,
+      "loss": 2.3547,
+      "step": 990
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 0.2502199411392212,
+      "learning_rate": 0.001851851851851852,
+      "loss": 2.3492,
+      "step": 1000
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 0.23490004241466522,
+      "learning_rate": 0.0018503703703703705,
+      "loss": 2.3451,
+      "step": 1010
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 0.2701961100101471,
+      "learning_rate": 0.0018488888888888888,
+      "loss": 2.3778,
+      "step": 1020
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 0.27240490913391113,
+      "learning_rate": 0.0018474074074074074,
+      "loss": 2.3627,
+      "step": 1030
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 0.2597067952156067,
+      "learning_rate": 0.001845925925925926,
+      "loss": 2.3849,
+      "step": 1040
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 0.24304194748401642,
+      "learning_rate": 0.0018444444444444446,
+      "loss": 2.3623,
+      "step": 1050
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 0.24453286826610565,
+      "learning_rate": 0.001842962962962963,
+      "loss": 2.3752,
+      "step": 1060
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 0.24190236628055573,
+      "learning_rate": 0.0018414814814814815,
+      "loss": 2.4014,
+      "step": 1070
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.5276634693145752,
+      "learning_rate": 0.00184,
+      "loss": 2.3874,
+      "step": 1080
+    },
+    {
+      "epoch": 4.04,
+      "grad_norm": 0.26895132660865784,
+      "learning_rate": 0.0018385185185185184,
+      "loss": 2.1745,
+      "step": 1090
+    },
+    {
+      "epoch": 4.07,
+      "grad_norm": 0.2788925766944885,
+      "learning_rate": 0.001837037037037037,
+      "loss": 2.142,
+      "step": 1100
+    },
+    {
+      "epoch": 4.11,
+      "grad_norm": 0.28623485565185547,
+      "learning_rate": 0.0018355555555555556,
+      "loss": 2.166,
+      "step": 1110
+    },
+    {
+      "epoch": 4.15,
+      "grad_norm": 0.2977291941642761,
+      "learning_rate": 0.001834074074074074,
+      "loss": 2.1839,
+      "step": 1120
+    },
+    {
+      "epoch": 4.19,
+      "grad_norm": 0.30595630407333374,
+      "learning_rate": 0.0018325925925925927,
+      "loss": 2.2183,
+      "step": 1130
+    },
+    {
+      "epoch": 4.22,
+      "grad_norm": 0.30192676186561584,
+      "learning_rate": 0.0018311111111111113,
+      "loss": 2.1944,
+      "step": 1140
+    },
+    {
+      "epoch": 4.26,
+      "grad_norm": 0.29342707991600037,
+      "learning_rate": 0.0018296296296296298,
+      "loss": 2.2278,
+      "step": 1150
+    },
+    {
+      "epoch": 4.3,
+      "grad_norm": 0.28939688205718994,
+      "learning_rate": 0.0018281481481481482,
+      "loss": 2.2185,
+      "step": 1160
+    },
+    {
+      "epoch": 4.33,
+      "grad_norm": 0.30929163098335266,
+      "learning_rate": 0.0018266666666666668,
+      "loss": 2.2323,
+      "step": 1170
+    },
+    {
+      "epoch": 4.37,
+      "grad_norm": 0.2740519046783447,
+      "learning_rate": 0.0018251851851851853,
+      "loss": 2.2288,
+      "step": 1180
+    },
+    {
+      "epoch": 4.41,
+      "grad_norm": 0.27721187472343445,
+      "learning_rate": 0.0018237037037037037,
+      "loss": 2.2596,
+      "step": 1190
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 0.28971293568611145,
+      "learning_rate": 0.0018222222222222223,
+      "loss": 2.255,
+      "step": 1200
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 0.27458974719047546,
+      "learning_rate": 0.0018207407407407408,
+      "loss": 2.2531,
+      "step": 1210
+    },
+    {
+      "epoch": 4.52,
+      "grad_norm": 0.2680809795856476,
+      "learning_rate": 0.0018192592592592594,
+      "loss": 2.2601,
+      "step": 1220
+    },
+    {
+      "epoch": 4.56,
+      "grad_norm": 0.27959805727005005,
+      "learning_rate": 0.0018177777777777778,
+      "loss": 2.3105,
+      "step": 1230
+    },
+    {
+      "epoch": 4.59,
+      "grad_norm": 0.2920687794685364,
+      "learning_rate": 0.0018162962962962963,
+      "loss": 2.2636,
+      "step": 1240
+    },
+    {
+      "epoch": 4.63,
+      "grad_norm": 0.2851353585720062,
+      "learning_rate": 0.001814814814814815,
+      "loss": 2.2839,
+      "step": 1250
+    },
+    {
+      "epoch": 4.67,
+      "grad_norm": 0.2648279368877411,
+      "learning_rate": 0.0018133333333333332,
+      "loss": 2.2682,
+      "step": 1260
+    },
+    {
+      "epoch": 4.7,
+      "grad_norm": 0.29693925380706787,
+      "learning_rate": 0.0018118518518518518,
+      "loss": 2.3228,
+      "step": 1270
+    },
+    {
+      "epoch": 4.74,
+      "grad_norm": 0.2693846523761749,
+      "learning_rate": 0.0018103703703703704,
+      "loss": 2.2764,
+      "step": 1280
+    },
+    {
+      "epoch": 4.78,
+      "grad_norm": 0.29114627838134766,
+      "learning_rate": 0.001808888888888889,
+      "loss": 2.2741,
+      "step": 1290
+    },
+    {
+      "epoch": 4.81,
+      "grad_norm": 0.3028624951839447,
+      "learning_rate": 0.0018074074074074073,
+      "loss": 2.3024,
+      "step": 1300
+    },
+    {
+      "epoch": 4.85,
+      "grad_norm": 0.2751203775405884,
+      "learning_rate": 0.0018059259259259259,
+      "loss": 2.3401,
+      "step": 1310
+    },
+    {
+      "epoch": 4.89,
+      "grad_norm": 0.2969362437725067,
+      "learning_rate": 0.0018044444444444445,
+      "loss": 2.3181,
+      "step": 1320
+    },
+    {
+      "epoch": 4.93,
+      "grad_norm": 0.2908598780632019,
+      "learning_rate": 0.0018029629629629628,
+      "loss": 2.3443,
+      "step": 1330
+    },
+    {
+      "epoch": 4.96,
+      "grad_norm": 0.29255783557891846,
+      "learning_rate": 0.0018014814814814816,
+      "loss": 2.3673,
+      "step": 1340
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 0.54734867811203,
+      "learning_rate": 0.0018000000000000002,
+      "loss": 2.3092,
+      "step": 1350
+    },
+    {
+      "epoch": 5.04,
+      "grad_norm": 0.3392738997936249,
+      "learning_rate": 0.0017985185185185187,
+      "loss": 2.053,
+      "step": 1360
+    },
+    {
+      "epoch": 5.07,
+      "grad_norm": 0.3059435784816742,
+      "learning_rate": 0.001797037037037037,
+      "loss": 2.0656,
+      "step": 1370
+    },
+    {
+      "epoch": 5.11,
+      "grad_norm": 0.3064379394054413,
+      "learning_rate": 0.0017955555555555557,
+      "loss": 2.1344,
+      "step": 1380
+    },
+    {
+      "epoch": 5.15,
+      "grad_norm": 0.3037875294685364,
+      "learning_rate": 0.0017940740740740742,
+      "loss": 2.1175,
+      "step": 1390
+    },
+    {
+      "epoch": 5.19,
+      "grad_norm": 0.3414519727230072,
+      "learning_rate": 0.0017925925925925926,
+      "loss": 2.1025,
+      "step": 1400
+    },
+    {
+      "epoch": 5.22,
+      "grad_norm": 0.31483858823776245,
+      "learning_rate": 0.0017911111111111112,
+      "loss": 2.1406,
+      "step": 1410
+    },
+    {
+      "epoch": 5.26,
+      "grad_norm": 0.3169196546077728,
+      "learning_rate": 0.0017896296296296297,
+      "loss": 2.1202,
+      "step": 1420
+    },
+    {
+      "epoch": 5.3,
+      "grad_norm": 0.32005852460861206,
+      "learning_rate": 0.001788148148148148,
+      "loss": 2.165,
+      "step": 1430
+    },
+    {
+      "epoch": 5.33,
+      "grad_norm": 0.32935044169425964,
+      "learning_rate": 0.0017866666666666667,
+      "loss": 2.19,
+      "step": 1440
+    },
+    {
+      "epoch": 5.37,
+      "grad_norm": 0.3298720419406891,
+      "learning_rate": 0.0017851851851851852,
+      "loss": 2.2113,
+      "step": 1450
+    },
+    {
+      "epoch": 5.41,
+      "grad_norm": 0.3138920068740845,
+      "learning_rate": 0.0017837037037037038,
+      "loss": 2.1645,
+      "step": 1460
+    },
+    {
+      "epoch": 5.44,
+      "grad_norm": 0.31499677896499634,
+      "learning_rate": 0.0017822222222222222,
+      "loss": 2.1918,
+      "step": 1470
+    },
+    {
+      "epoch": 5.48,
+      "grad_norm": 0.31354495882987976,
+      "learning_rate": 0.0017807407407407407,
+      "loss": 2.2005,
+      "step": 1480
+    },
+    {
+      "epoch": 5.52,
+      "grad_norm": 0.2973974943161011,
+      "learning_rate": 0.0017792592592592593,
+      "loss": 2.2388,
+      "step": 1490
+    },
+    {
+      "epoch": 5.56,
+      "grad_norm": 0.29360270500183105,
+      "learning_rate": 0.0017777777777777776,
+      "loss": 2.2541,
+      "step": 1500
+    },
+    {
+      "epoch": 5.59,
+      "grad_norm": 0.29777848720550537,
+      "learning_rate": 0.0017762962962962962,
+      "loss": 2.2157,
+      "step": 1510
+    },
+    {
+      "epoch": 5.63,
+      "grad_norm": 0.3189815878868103,
+      "learning_rate": 0.0017748148148148148,
+      "loss": 2.2672,
+      "step": 1520
+    },
+    {
+      "epoch": 5.67,
+      "grad_norm": 0.30607324838638306,
+      "learning_rate": 0.0017733333333333334,
+      "loss": 2.2685,
+      "step": 1530
+    },
+    {
+      "epoch": 5.7,
+      "grad_norm": 0.3085130751132965,
+      "learning_rate": 0.001771851851851852,
+      "loss": 2.2189,
+      "step": 1540
+    },
+    {
+      "epoch": 5.74,
+      "grad_norm": 0.3023441433906555,
+      "learning_rate": 0.0017703703703703705,
+      "loss": 2.2097,
+      "step": 1550
+    },
+    {
+      "epoch": 5.78,
+      "grad_norm": 0.29119688272476196,
+      "learning_rate": 0.001768888888888889,
+      "loss": 2.225,
+      "step": 1560
+    },
+    {
+      "epoch": 5.81,
+      "grad_norm": 0.2987782657146454,
+      "learning_rate": 0.0017674074074074074,
+      "loss": 2.282,
+      "step": 1570
+    },
+    {
+      "epoch": 5.85,
+      "grad_norm": 0.31132954359054565,
+      "learning_rate": 0.001765925925925926,
+      "loss": 2.2924,
+      "step": 1580
+    },
+    {
+      "epoch": 5.89,
+      "grad_norm": 0.29663658142089844,
+      "learning_rate": 0.0017644444444444446,
+      "loss": 2.2517,
+      "step": 1590
+    },
+    {
+      "epoch": 5.93,
+      "grad_norm": 0.28728464245796204,
+      "learning_rate": 0.0017629629629629631,
+      "loss": 2.2853,
+      "step": 1600
+    },
+    {
+      "epoch": 5.96,
+      "grad_norm": 0.29688742756843567,
+      "learning_rate": 0.0017614814814814815,
+      "loss": 2.2909,
+      "step": 1610
+    },
+    {
+      "epoch": 6.0,
+      "grad_norm": 0.765344500541687,
+      "learning_rate": 0.00176,
+      "loss": 2.2834,
+      "step": 1620
+    },
+    {
+      "epoch": 6.04,
+      "grad_norm": 0.3417266309261322,
+      "learning_rate": 0.0017585185185185186,
+      "loss": 2.0025,
+      "step": 1630
+    },
+    {
+      "epoch": 6.07,
+      "grad_norm": 0.34385794401168823,
+      "learning_rate": 0.001757037037037037,
+      "loss": 1.9865,
+      "step": 1640
+    },
+    {
+      "epoch": 6.11,
+      "grad_norm": 0.31611743569374084,
+      "learning_rate": 0.0017555555555555556,
+      "loss": 2.0147,
+      "step": 1650
+    },
+    {
+      "epoch": 6.15,
+      "grad_norm": 0.33877480030059814,
+      "learning_rate": 0.0017540740740740741,
+      "loss": 2.0638,
+      "step": 1660
+    },
+    {
+      "epoch": 6.19,
+      "grad_norm": 0.33535125851631165,
+      "learning_rate": 0.0017525925925925927,
+      "loss": 2.0912,
+      "step": 1670
+    },
+    {
+      "epoch": 6.22,
+      "grad_norm": 0.34495416283607483,
+      "learning_rate": 0.001751111111111111,
+      "loss": 2.1032,
+      "step": 1680
+    },
+    {
+      "epoch": 6.26,
+      "grad_norm": 0.34369274973869324,
+      "learning_rate": 0.0017496296296296296,
+      "loss": 2.1167,
+      "step": 1690
+    },
+    {
+      "epoch": 6.3,
+      "grad_norm": 0.34757956862449646,
+      "learning_rate": 0.0017481481481481482,
+      "loss": 2.1166,
+      "step": 1700
+    },
+    {
+      "epoch": 6.33,
+      "grad_norm": 0.35368654131889343,
+      "learning_rate": 0.0017466666666666665,
+      "loss": 2.1397,
+      "step": 1710
+    },
+    {
+      "epoch": 6.37,
+      "grad_norm": 0.341086745262146,
+      "learning_rate": 0.0017451851851851851,
+      "loss": 2.1334,
+      "step": 1720
+    },
+    {
+      "epoch": 6.41,
+      "grad_norm": 0.36962762475013733,
+      "learning_rate": 0.0017437037037037037,
+      "loss": 2.141,
+      "step": 1730
+    },
+    {
+      "epoch": 6.44,
+      "grad_norm": 0.3078432083129883,
+      "learning_rate": 0.001742222222222222,
+      "loss": 2.1514,
+      "step": 1740
+    },
+    {
+      "epoch": 6.48,
+      "grad_norm": 0.34287720918655396,
+      "learning_rate": 0.0017407407407407408,
+      "loss": 2.1918,
+      "step": 1750
+    },
+    {
+      "epoch": 6.52,
+      "grad_norm": 0.315064936876297,
+      "learning_rate": 0.0017392592592592594,
+      "loss": 2.1685,
+      "step": 1760
+    },
+    {
+      "epoch": 6.56,
+      "grad_norm": 0.31131112575531006,
+      "learning_rate": 0.001737777777777778,
+      "loss": 2.1827,
+      "step": 1770
+    },
+    {
+      "epoch": 6.59,
+      "grad_norm": 0.3834725022315979,
+      "learning_rate": 0.0017362962962962963,
+      "loss": 2.1857,
+      "step": 1780
+    },
+    {
+      "epoch": 6.63,
+      "grad_norm": 0.34319067001342773,
+      "learning_rate": 0.001734814814814815,
+      "loss": 2.2084,
+      "step": 1790
+    },
+    {
+      "epoch": 6.67,
+      "grad_norm": 0.29598793387413025,
+      "learning_rate": 0.0017333333333333335,
+      "loss": 2.1508,
+      "step": 1800
+    },
+    {
+      "epoch": 6.7,
+      "grad_norm": 0.3391283452510834,
+      "learning_rate": 0.0017318518518518518,
+      "loss": 2.1681,
+      "step": 1810
+    },
+    {
+      "epoch": 6.74,
+      "grad_norm": 0.3115655183792114,
+      "learning_rate": 0.0017303703703703704,
+      "loss": 2.2049,
+      "step": 1820
+    },
+    {
+      "epoch": 6.78,
+      "grad_norm": 0.3322307765483856,
+      "learning_rate": 0.001728888888888889,
+      "loss": 2.2095,
+      "step": 1830
+    },
+    {
+      "epoch": 6.81,
+      "grad_norm": 0.32754597067832947,
+      "learning_rate": 0.0017274074074074075,
+      "loss": 2.2035,
+      "step": 1840
+    },
+    {
+      "epoch": 6.85,
+      "grad_norm": 0.31123578548431396,
+      "learning_rate": 0.0017259259259259259,
+      "loss": 2.2203,
+      "step": 1850
+    },
+    {
+      "epoch": 6.89,
+      "grad_norm": 0.32000821828842163,
+      "learning_rate": 0.0017244444444444445,
+      "loss": 2.1779,
+      "step": 1860
+    },
+    {
+      "epoch": 6.93,
+      "grad_norm": 0.3137577474117279,
+      "learning_rate": 0.001722962962962963,
+      "loss": 2.2027,
+      "step": 1870
+    },
+    {
+      "epoch": 6.96,
+      "grad_norm": 0.31178486347198486,
+      "learning_rate": 0.0017214814814814814,
+      "loss": 2.2265,
+      "step": 1880
+    },
+    {
+      "epoch": 7.0,
+      "grad_norm": 0.5970402956008911,
+      "learning_rate": 0.00172,
+      "loss": 2.184,
+      "step": 1890
+    },
+    {
+      "epoch": 7.04,
+      "grad_norm": 0.351868212223053,
+      "learning_rate": 0.0017185185185185185,
+      "loss": 1.9941,
+      "step": 1900
+    },
+    {
+      "epoch": 7.07,
+      "grad_norm": 0.37474524974823,
+      "learning_rate": 0.001717037037037037,
+      "loss": 1.9817,
+      "step": 1910
+    },
+    {
+      "epoch": 7.11,
+      "grad_norm": 0.40310508012771606,
+      "learning_rate": 0.0017155555555555555,
+      "loss": 2.0109,
+      "step": 1920
+    },
+    {
+      "epoch": 7.15,
+      "grad_norm": 0.40088704228401184,
+      "learning_rate": 0.001714074074074074,
+      "loss": 2.0127,
+      "step": 1930
+    },
+    {
+      "epoch": 7.19,
+      "grad_norm": 0.3924960196018219,
+      "learning_rate": 0.0017125925925925926,
+      "loss": 2.012,
+      "step": 1940
+    },
+    {
+      "epoch": 7.22,
+      "grad_norm": 0.3628038167953491,
+      "learning_rate": 0.0017111111111111112,
+      "loss": 2.016,
+      "step": 1950
+    },
+    {
+      "epoch": 7.26,
+      "grad_norm": 0.34143421053886414,
+      "learning_rate": 0.0017096296296296297,
+      "loss": 2.0582,
+      "step": 1960
+    },
+    {
+      "epoch": 7.3,
+      "grad_norm": 0.3582156002521515,
+      "learning_rate": 0.0017081481481481483,
+      "loss": 2.0677,
+      "step": 1970
+    },
+    {
+      "epoch": 7.33,
+      "grad_norm": 0.3526308536529541,
+      "learning_rate": 0.0017066666666666669,
+      "loss": 2.0938,
+      "step": 1980
+    },
+    {
+      "epoch": 7.37,
+      "grad_norm": 0.3765566051006317,
+      "learning_rate": 0.0017051851851851852,
+      "loss": 2.0936,
+      "step": 1990
+    },
+    {
+      "epoch": 7.41,
+      "grad_norm": 0.35167813301086426,
+      "learning_rate": 0.0017037037037037038,
+      "loss": 2.1124,
+      "step": 2000
+    },
+    {
+      "epoch": 7.44,
+      "grad_norm": 0.3823408782482147,
+      "learning_rate": 0.0017022222222222224,
+      "loss": 2.0656,
+      "step": 2010
+    },
+    {
+      "epoch": 7.48,
+      "grad_norm": 0.33811891078948975,
+      "learning_rate": 0.0017007407407407407,
+      "loss": 2.1003,
+      "step": 2020
+    },
+    {
+      "epoch": 7.52,
+      "grad_norm": 0.36594629287719727,
+      "learning_rate": 0.0016992592592592593,
+      "loss": 2.1514,
+      "step": 2030
+    },
+    {
+      "epoch": 7.56,
+      "grad_norm": 0.3879795968532562,
+      "learning_rate": 0.0016977777777777779,
+      "loss": 2.1274,
+      "step": 2040
+    },
+    {
+      "epoch": 7.59,
+      "grad_norm": 0.3433550298213959,
+      "learning_rate": 0.0016962962962962962,
+      "loss": 2.102,
+      "step": 2050
+    },
+    {
+      "epoch": 7.63,
+      "grad_norm": 0.34211045503616333,
+      "learning_rate": 0.0016948148148148148,
+      "loss": 2.0929,
+      "step": 2060
+    },
+    {
+      "epoch": 7.67,
+      "grad_norm": 0.3362341821193695,
+      "learning_rate": 0.0016933333333333334,
+      "loss": 2.1426,
+      "step": 2070
+    },
+    {
+      "epoch": 7.7,
+      "grad_norm": 0.3737432658672333,
+      "learning_rate": 0.001691851851851852,
+      "loss": 2.1574,
+      "step": 2080
+    },
+    {
+      "epoch": 7.74,
+      "grad_norm": 0.33443060517311096,
+      "learning_rate": 0.0016903703703703703,
+      "loss": 2.1441,
+      "step": 2090
+    },
+    {
+      "epoch": 7.78,
+      "grad_norm": 0.3313807547092438,
+      "learning_rate": 0.0016888888888888889,
+      "loss": 2.1376,
+      "step": 2100
+    },
+    {
+      "epoch": 7.81,
+      "grad_norm": 0.3413750231266022,
+      "learning_rate": 0.0016874074074074074,
+      "loss": 2.1227,
+      "step": 2110
+    },
+    {
+      "epoch": 7.85,
+      "grad_norm": 0.356935054063797,
+      "learning_rate": 0.0016859259259259258,
+      "loss": 2.1511,
+      "step": 2120
+    },
+    {
+      "epoch": 7.89,
+      "grad_norm": 0.30882149934768677,
+      "learning_rate": 0.0016844444444444444,
+      "loss": 2.1363,
+      "step": 2130
+    },
+    {
+      "epoch": 7.93,
+      "grad_norm": 0.3477705121040344,
+      "learning_rate": 0.001682962962962963,
+      "loss": 2.1855,
+      "step": 2140
+    },
+    {
+      "epoch": 7.96,
+      "grad_norm": 0.32417911291122437,
+      "learning_rate": 0.0016814814814814817,
+      "loss": 2.169,
+      "step": 2150
+    },
+    {
+      "epoch": 8.0,
+      "grad_norm": 0.6133363842964172,
+      "learning_rate": 0.00168,
+      "loss": 2.1631,
+      "step": 2160
+    },
+    {
+      "epoch": 8.04,
+      "grad_norm": 0.3599429130554199,
+      "learning_rate": 0.0016785185185185186,
+      "loss": 1.8783,
+      "step": 2170
+    },
+    {
+      "epoch": 8.07,
+      "grad_norm": 0.40336427092552185,
+      "learning_rate": 0.0016770370370370372,
+      "loss": 1.9111,
+      "step": 2180
+    },
+    {
+      "epoch": 8.11,
+      "grad_norm": 0.38553932309150696,
+      "learning_rate": 0.0016755555555555556,
+      "loss": 1.9616,
+      "step": 2190
+    },
+    {
+      "epoch": 8.15,
+      "grad_norm": 0.41913771629333496,
+      "learning_rate": 0.0016740740740740741,
+      "loss": 1.9623,
+      "step": 2200
+    },
+    {
+      "epoch": 8.19,
+      "grad_norm": 0.38957417011260986,
+      "learning_rate": 0.0016725925925925927,
+      "loss": 1.9991,
+      "step": 2210
+    },
+    {
+      "epoch": 8.22,
+      "grad_norm": 0.35183995962142944,
+      "learning_rate": 0.0016711111111111113,
+      "loss": 1.9689,
+      "step": 2220
+    },
+    {
+      "epoch": 8.26,
+      "grad_norm": 0.3833864629268646,
+      "learning_rate": 0.0016696296296296296,
+      "loss": 1.9944,
+      "step": 2230
+    },
+    {
+      "epoch": 8.3,
+      "grad_norm": 0.38431793451309204,
+      "learning_rate": 0.0016681481481481482,
+      "loss": 2.0343,
+      "step": 2240
+    },
+    {
+      "epoch": 8.33,
+      "grad_norm": 0.35817861557006836,
+      "learning_rate": 0.0016666666666666668,
+      "loss": 2.0205,
+      "step": 2250
+    },
+    {
+      "epoch": 8.37,
+      "grad_norm": 0.35562562942504883,
+      "learning_rate": 0.0016651851851851851,
+      "loss": 2.0322,
+      "step": 2260
+    },
+    {
+      "epoch": 8.41,
+      "grad_norm": 0.40019509196281433,
+      "learning_rate": 0.0016637037037037037,
+      "loss": 2.0804,
+      "step": 2270
+    },
+    {
+      "epoch": 8.44,
+      "grad_norm": 0.3636208772659302,
+      "learning_rate": 0.0016622222222222223,
+      "loss": 2.068,
+      "step": 2280
+    },
+    {
+      "epoch": 8.48,
+      "grad_norm": 0.4120703637599945,
+      "learning_rate": 0.0016607407407407408,
+      "loss": 2.0357,
+      "step": 2290
+    },
+    {
+      "epoch": 8.52,
+      "grad_norm": 0.3651635944843292,
+      "learning_rate": 0.0016592592592592592,
+      "loss": 2.0502,
+      "step": 2300
+    },
+    {
+      "epoch": 8.56,
+      "grad_norm": 0.36248978972435,
+      "learning_rate": 0.0016577777777777778,
+      "loss": 2.0824,
+      "step": 2310
+    },
+    {
+      "epoch": 8.59,
+      "grad_norm": 0.36948567628860474,
+      "learning_rate": 0.0016562962962962963,
+      "loss": 2.094,
+      "step": 2320
+    },
+    {
+      "epoch": 8.63,
+      "grad_norm": 0.3564412295818329,
+      "learning_rate": 0.0016548148148148147,
+      "loss": 2.1282,
+      "step": 2330
+    },
+    {
+      "epoch": 8.67,
+      "grad_norm": 0.3488951623439789,
+      "learning_rate": 0.0016533333333333333,
+      "loss": 2.081,
+      "step": 2340
+    },
+    {
+      "epoch": 8.7,
+      "grad_norm": 0.385596364736557,
+      "learning_rate": 0.0016518518518518518,
+      "loss": 2.1045,
+      "step": 2350
+    },
+    {
+      "epoch": 8.74,
+      "grad_norm": 0.35894304513931274,
+      "learning_rate": 0.0016503703703703704,
+      "loss": 2.1105,
+      "step": 2360
+    },
+    {
+      "epoch": 8.78,
+      "grad_norm": 0.34181079268455505,
+      "learning_rate": 0.001648888888888889,
+      "loss": 2.1062,
+      "step": 2370
+    },
+    {
+      "epoch": 8.81,
+      "grad_norm": 0.3692533075809479,
+      "learning_rate": 0.0016474074074074075,
+      "loss": 2.1161,
+      "step": 2380
+    },
+    {
+      "epoch": 8.85,
+      "grad_norm": 0.37401172518730164,
+      "learning_rate": 0.0016459259259259261,
+      "loss": 2.1134,
+      "step": 2390
+    },
+    {
+      "epoch": 8.89,
+      "grad_norm": 0.38438746333122253,
+      "learning_rate": 0.0016444444444444445,
+      "loss": 2.1237,
+      "step": 2400
+    },
+    {
+      "epoch": 8.93,
+      "grad_norm": 0.35017186403274536,
+      "learning_rate": 0.001642962962962963,
+      "loss": 2.1042,
+      "step": 2410
+    },
+    {
+      "epoch": 8.96,
+      "grad_norm": 0.3776445984840393,
+      "learning_rate": 0.0016414814814814816,
+      "loss": 2.1506,
+      "step": 2420
+    },
+    {
+      "epoch": 9.0,
+      "grad_norm": 0.6395183205604553,
+      "learning_rate": 0.00164,
+      "loss": 2.1215,
+      "step": 2430
+    },
+    {
+      "epoch": 9.04,
+      "grad_norm": 0.4051409661769867,
+      "learning_rate": 0.0016385185185185185,
+      "loss": 1.8761,
+      "step": 2440
+    },
+    {
+      "epoch": 9.07,
+      "grad_norm": 0.41390544176101685,
+      "learning_rate": 0.001637037037037037,
+      "loss": 1.8756,
+      "step": 2450
+    },
+    {
+      "epoch": 9.11,
+      "grad_norm": 0.4349888265132904,
+      "learning_rate": 0.0016355555555555557,
+      "loss": 1.8841,
+      "step": 2460
+    },
+    {
+      "epoch": 9.15,
+      "grad_norm": 0.4476017355918884,
+      "learning_rate": 0.001634074074074074,
+      "loss": 1.9162,
+      "step": 2470
+    },
+    {
+      "epoch": 9.19,
+      "grad_norm": 0.37201741337776184,
+      "learning_rate": 0.0016325925925925926,
+      "loss": 1.9246,
+      "step": 2480
+    },
+    {
+      "epoch": 9.22,
+      "grad_norm": 0.3889928162097931,
+      "learning_rate": 0.0016311111111111112,
+      "loss": 1.9684,
+      "step": 2490
+    },
+    {
+      "epoch": 9.26,
+      "grad_norm": 0.43831804394721985,
+      "learning_rate": 0.0016296296296296295,
+      "loss": 1.9903,
+      "step": 2500
+    },
+    {
+      "epoch": 9.3,
+      "grad_norm": 0.397524356842041,
+      "learning_rate": 0.001628148148148148,
+      "loss": 1.9662,
+      "step": 2510
+    },
+    {
+      "epoch": 9.33,
+      "grad_norm": 0.40506064891815186,
+      "learning_rate": 0.0016266666666666667,
+      "loss": 1.9938,
+      "step": 2520
+    },
+    {
+      "epoch": 9.37,
+      "grad_norm": 0.42983636260032654,
+      "learning_rate": 0.0016251851851851852,
+      "loss": 1.9802,
+      "step": 2530
+    },
+    {
+      "epoch": 9.41,
+      "grad_norm": 0.39103442430496216,
+      "learning_rate": 0.0016237037037037036,
+      "loss": 1.9937,
+      "step": 2540
+    },
+    {
+      "epoch": 9.44,
+      "grad_norm": 0.4009477496147156,
+      "learning_rate": 0.0016222222222222222,
+      "loss": 2.0031,
+      "step": 2550
+    },
+    {
+      "epoch": 9.48,
+      "grad_norm": 0.36088138818740845,
+      "learning_rate": 0.001620740740740741,
+      "loss": 2.0663,
+      "step": 2560
+    },
+    {
+      "epoch": 9.52,
+      "grad_norm": 0.3654880225658417,
+      "learning_rate": 0.0016192592592592593,
+      "loss": 2.0053,
+      "step": 2570
+    },
+    {
+      "epoch": 9.56,
+      "grad_norm": 0.35059937834739685,
+      "learning_rate": 0.0016177777777777779,
+      "loss": 2.0549,
+      "step": 2580
+    },
+    {
+      "epoch": 9.59,
+      "grad_norm": 0.37335512042045593,
+      "learning_rate": 0.0016162962962962964,
+      "loss": 2.0331,
+      "step": 2590
+    },
+    {
+      "epoch": 9.63,
+      "grad_norm": 0.36852195858955383,
+      "learning_rate": 0.001614814814814815,
+      "loss": 2.042,
+      "step": 2600
+    },
+    {
+      "epoch": 9.67,
+      "grad_norm": 0.3899456262588501,
+      "learning_rate": 0.0016133333333333334,
+      "loss": 2.0552,
+      "step": 2610
+    },
+    {
+      "epoch": 9.7,
+      "grad_norm": 0.390413373708725,
+      "learning_rate": 0.001611851851851852,
+      "loss": 2.0535,
+      "step": 2620
+    },
+    {
+      "epoch": 9.74,
+      "grad_norm": 0.3956015110015869,
+      "learning_rate": 0.0016103703703703705,
+      "loss": 2.0662,
+      "step": 2630
+    },
+    {
+      "epoch": 9.78,
+      "grad_norm": 0.39609503746032715,
+      "learning_rate": 0.0016088888888888889,
+      "loss": 2.0453,
+      "step": 2640
+    },
+    {
+      "epoch": 9.81,
+      "grad_norm": 0.38103851675987244,
+      "learning_rate": 0.0016074074074074074,
+      "loss": 2.068,
+      "step": 2650
+    },
+    {
+      "epoch": 9.85,
+      "grad_norm": 0.40990570187568665,
+      "learning_rate": 0.001605925925925926,
+      "loss": 2.0881,
+      "step": 2660
+    },
+    {
+      "epoch": 9.89,
+      "grad_norm": 0.37008753418922424,
+      "learning_rate": 0.0016044444444444444,
+      "loss": 2.0609,
+      "step": 2670
+    },
+    {
+      "epoch": 9.93,
+      "grad_norm": 0.4039746820926666,
+      "learning_rate": 0.001602962962962963,
+      "loss": 2.0809,
+      "step": 2680
+    },
+    {
+      "epoch": 9.96,
+      "grad_norm": 0.38472408056259155,
+      "learning_rate": 0.0016014814814814815,
+      "loss": 2.0827,
+      "step": 2690
+    },
+    {
+      "epoch": 10.0,
+      "grad_norm": 0.7517541646957397,
+      "learning_rate": 0.0016,
+      "loss": 2.058,
+      "step": 2700
+    },
+    {
+      "epoch": 10.04,
+      "grad_norm": 0.4214729368686676,
+      "learning_rate": 0.0015985185185185184,
+      "loss": 1.8799,
+      "step": 2710
+    },
+    {
+      "epoch": 10.07,
+      "grad_norm": 0.41175585985183716,
+      "learning_rate": 0.001597037037037037,
+      "loss": 1.8452,
+      "step": 2720
+    },
+    {
+      "epoch": 10.11,
+      "grad_norm": 0.4495772421360016,
+      "learning_rate": 0.0015955555555555556,
+      "loss": 1.8456,
+      "step": 2730
+    },
+    {
+      "epoch": 10.15,
+      "grad_norm": 0.4272764325141907,
+      "learning_rate": 0.001594074074074074,
+      "loss": 1.8606,
+      "step": 2740
+    },
+    {
+      "epoch": 10.19,
+      "grad_norm": 0.422076940536499,
+      "learning_rate": 0.0015925925925925925,
+      "loss": 1.8831,
+      "step": 2750
+    },
+    {
+      "epoch": 10.22,
+      "grad_norm": 0.4146776795387268,
+      "learning_rate": 0.001591111111111111,
+      "loss": 1.9224,
+      "step": 2760
+    },
+    {
+      "epoch": 10.26,
+      "grad_norm": 0.4350244998931885,
+      "learning_rate": 0.0015896296296296299,
+      "loss": 1.9407,
+      "step": 2770
+    },
+    {
+      "epoch": 10.3,
+      "grad_norm": 0.40515050292015076,
+      "learning_rate": 0.0015881481481481482,
+      "loss": 1.8997,
+      "step": 2780
+    },
+    {
+      "epoch": 10.33,
+      "grad_norm": 0.4228098690509796,
+      "learning_rate": 0.0015866666666666668,
+      "loss": 1.9657,
+      "step": 2790
+    },
+    {
+      "epoch": 10.37,
+      "grad_norm": 0.46530553698539734,
+      "learning_rate": 0.0015851851851851854,
+      "loss": 1.9677,
+      "step": 2800
+    },
+    {
+      "epoch": 10.41,
+      "grad_norm": 0.4347987174987793,
+      "learning_rate": 0.0015837037037037037,
+      "loss": 1.9443,
+      "step": 2810
+    },
+    {
+      "epoch": 10.44,
+      "grad_norm": 0.4162721633911133,
+      "learning_rate": 0.0015822222222222223,
+      "loss": 1.9851,
+      "step": 2820
+    },
+    {
+      "epoch": 10.48,
+      "grad_norm": 0.4888492822647095,
+      "learning_rate": 0.0015807407407407408,
+      "loss": 1.9795,
+      "step": 2830
+    },
+    {
+      "epoch": 10.52,
+      "grad_norm": 0.44969916343688965,
+      "learning_rate": 0.0015792592592592594,
+      "loss": 1.9934,
+      "step": 2840
+    },
+    {
+      "epoch": 10.56,
+      "grad_norm": 0.38131090998649597,
+      "learning_rate": 0.0015777777777777778,
+      "loss": 1.9955,
+      "step": 2850
+    },
+    {
+      "epoch": 10.59,
+      "grad_norm": 0.4034533202648163,
+      "learning_rate": 0.0015762962962962963,
+      "loss": 1.9994,
+      "step": 2860
+    },
+    {
+      "epoch": 10.63,
+      "grad_norm": 0.4201233386993408,
+      "learning_rate": 0.001574814814814815,
+      "loss": 2.0357,
+      "step": 2870
+    },
+    {
+      "epoch": 10.67,
+      "grad_norm": 0.3957751989364624,
+      "learning_rate": 0.0015733333333333333,
+      "loss": 2.0283,
+      "step": 2880
+    },
+    {
+      "epoch": 10.7,
+      "grad_norm": 0.40386030077934265,
+      "learning_rate": 0.0015718518518518518,
+      "loss": 2.0228,
+      "step": 2890
+    },
+    {
+      "epoch": 10.74,
+      "grad_norm": 0.4089337885379791,
+      "learning_rate": 0.0015703703703703704,
+      "loss": 2.0192,
+      "step": 2900
+    },
+    {
+      "epoch": 10.78,
+      "grad_norm": 0.38744863867759705,
+      "learning_rate": 0.001568888888888889,
+      "loss": 2.0336,
+      "step": 2910
+    },
+    {
+      "epoch": 10.81,
+      "grad_norm": 0.41213229298591614,
+      "learning_rate": 0.0015674074074074073,
+      "loss": 2.0287,
+      "step": 2920
+    },
+    {
+      "epoch": 10.85,
+      "grad_norm": 0.4100775718688965,
+      "learning_rate": 0.001565925925925926,
+      "loss": 2.0622,
+      "step": 2930
+    },
+    {
+      "epoch": 10.89,
+      "grad_norm": 0.3857145607471466,
+      "learning_rate": 0.0015644444444444445,
+      "loss": 2.0241,
+      "step": 2940
+    },
+    {
+      "epoch": 10.93,
+      "grad_norm": 0.38610953092575073,
+      "learning_rate": 0.0015629629629629628,
+      "loss": 2.0367,
+      "step": 2950
+    },
+    {
+      "epoch": 10.96,
+      "grad_norm": 0.38510599732398987,
+      "learning_rate": 0.0015614814814814814,
+      "loss": 2.0345,
+      "step": 2960
+    },
+    {
+      "epoch": 11.0,
+      "grad_norm": 0.8309366703033447,
+      "learning_rate": 0.0015600000000000002,
+      "loss": 2.0176,
+      "step": 2970
+    },
+    {
+      "epoch": 11.04,
+      "grad_norm": 0.38769230246543884,
+      "learning_rate": 0.0015585185185185185,
+      "loss": 1.8074,
+      "step": 2980
+    },
+    {
+      "epoch": 11.07,
+      "grad_norm": 0.47605445981025696,
+      "learning_rate": 0.0015570370370370371,
+      "loss": 1.8336,
+      "step": 2990
+    },
+    {
+      "epoch": 11.11,
+      "grad_norm": 0.4387536346912384,
+      "learning_rate": 0.0015555555555555557,
+      "loss": 1.7895,
+      "step": 3000
+    },
+    {
+      "epoch": 11.15,
+      "grad_norm": 0.4390158951282501,
+      "learning_rate": 0.0015540740740740743,
+      "loss": 1.8829,
+      "step": 3010
+    },
+    {
+      "epoch": 11.19,
+      "grad_norm": 0.40687650442123413,
+      "learning_rate": 0.0015525925925925926,
+      "loss": 1.8919,
+      "step": 3020
+    },
+    {
+      "epoch": 11.22,
+      "grad_norm": 0.44035059213638306,
+      "learning_rate": 0.0015511111111111112,
+      "loss": 1.8691,
+      "step": 3030
+    },
+    {
+      "epoch": 11.26,
+      "grad_norm": 0.4709152579307556,
+      "learning_rate": 0.0015496296296296298,
+      "loss": 1.8916,
+      "step": 3040
+    },
+    {
+      "epoch": 11.3,
+      "grad_norm": 0.40621572732925415,
+      "learning_rate": 0.001548148148148148,
+      "loss": 1.8913,
+      "step": 3050
+    },
+    {
+      "epoch": 11.33,
+      "grad_norm": 0.4792878329753876,
+      "learning_rate": 0.0015466666666666667,
+      "loss": 1.9039,
+      "step": 3060
+    },
+    {
+      "epoch": 11.37,
+      "grad_norm": 0.4478456974029541,
+      "learning_rate": 0.0015451851851851852,
+      "loss": 1.8723,
+      "step": 3070
+    },
+    {
+      "epoch": 11.41,
+      "grad_norm": 0.4035521447658539,
+      "learning_rate": 0.0015437037037037038,
+      "loss": 1.9437,
+      "step": 3080
+    },
+    {
+      "epoch": 11.44,
+      "grad_norm": 0.4260112941265106,
+      "learning_rate": 0.0015422222222222222,
+      "loss": 1.9769,
+      "step": 3090
+    },
+    {
+      "epoch": 11.48,
+      "grad_norm": 0.43007922172546387,
+      "learning_rate": 0.0015407407407407407,
+      "loss": 1.9229,
+      "step": 3100
+    },
+    {
+      "epoch": 11.52,
+      "grad_norm": 0.3983825147151947,
+      "learning_rate": 0.0015392592592592593,
+      "loss": 1.92,
+      "step": 3110
+    },
+    {
+      "epoch": 11.56,
+      "grad_norm": 0.42792314291000366,
+      "learning_rate": 0.0015377777777777777,
+      "loss": 1.9274,
+      "step": 3120
+    },
+    {
+      "epoch": 11.59,
+      "grad_norm": 0.4552850127220154,
+      "learning_rate": 0.0015362962962962962,
+      "loss": 1.95,
+      "step": 3130
+    },
+    {
+      "epoch": 11.63,
+      "grad_norm": 0.4422164261341095,
+      "learning_rate": 0.0015348148148148148,
+      "loss": 1.9896,
+      "step": 3140
+    },
+    {
+      "epoch": 11.67,
+      "grad_norm": 0.4225594103336334,
+      "learning_rate": 0.0015333333333333334,
+      "loss": 1.9788,
+      "step": 3150
+    },
+    {
+      "epoch": 11.7,
+      "grad_norm": 0.4389062821865082,
+      "learning_rate": 0.0015318518518518517,
+      "loss": 1.9784,
+      "step": 3160
+    },
+    {
+      "epoch": 11.74,
+      "grad_norm": 0.44201141595840454,
+      "learning_rate": 0.0015303703703703703,
+      "loss": 1.9676,
+      "step": 3170
+    },
+    {
+      "epoch": 11.78,
+      "grad_norm": 0.377491295337677,
+      "learning_rate": 0.001528888888888889,
+      "loss": 2.0079,
+      "step": 3180
+    },
+    {
+      "epoch": 11.81,
+      "grad_norm": 0.4113980531692505,
+      "learning_rate": 0.0015274074074074074,
+      "loss": 1.9995,
+      "step": 3190
+    },
+    {
+      "epoch": 11.85,
+      "grad_norm": 0.44535115361213684,
+      "learning_rate": 0.001525925925925926,
+      "loss": 1.9633,
+      "step": 3200
+    },
+    {
+      "epoch": 11.89,
+      "grad_norm": 0.3994745910167694,
+      "learning_rate": 0.0015244444444444446,
+      "loss": 2.0275,
+      "step": 3210
+    },
+    {
+      "epoch": 11.93,
+      "grad_norm": 0.43397390842437744,
+      "learning_rate": 0.0015229629629629632,
+      "loss": 1.9791,
+      "step": 3220
+    },
+    {
+      "epoch": 11.96,
+      "grad_norm": 0.4297787547111511,
+      "learning_rate": 0.0015214814814814815,
+      "loss": 1.9933,
+      "step": 3230
+    },
+    {
+      "epoch": 12.0,
+      "grad_norm": 0.6857362389564514,
+      "learning_rate": 0.00152,
+      "loss": 2.0339,
+      "step": 3240
+    },
+    {
+      "epoch": 12.04,
+      "grad_norm": 0.4128698706626892,
+      "learning_rate": 0.0015185185185185187,
+      "loss": 1.7699,
+      "step": 3250
+    },
+    {
+      "epoch": 12.07,
+      "grad_norm": 0.47724348306655884,
+      "learning_rate": 0.001517037037037037,
+      "loss": 1.7503,
+      "step": 3260
+    },
+    {
+      "epoch": 12.11,
+      "grad_norm": 0.46561479568481445,
+      "learning_rate": 0.0015155555555555556,
+      "loss": 1.7583,
+      "step": 3270
+    },
+    {
+      "epoch": 12.15,
+      "grad_norm": 0.4969346821308136,
+      "learning_rate": 0.0015140740740740742,
+      "loss": 1.7912,
+      "step": 3280
+    },
+    {
+      "epoch": 12.19,
+      "grad_norm": 0.4789353013038635,
+      "learning_rate": 0.0015125925925925925,
+      "loss": 1.81,
+      "step": 3290
+    },
+    {
+      "epoch": 12.22,
+      "grad_norm": 0.43490344285964966,
+      "learning_rate": 0.001511111111111111,
+      "loss": 1.8208,
+      "step": 3300
+    },
+    {
+      "epoch": 12.26,
+      "grad_norm": 0.4532495141029358,
+      "learning_rate": 0.0015096296296296296,
+      "loss": 1.8588,
+      "step": 3310
+    },
+    {
+      "epoch": 12.3,
+      "grad_norm": 0.468681275844574,
+      "learning_rate": 0.0015081481481481482,
+      "loss": 1.8655,
+      "step": 3320
+    },
+    {
+      "epoch": 12.33,
+      "grad_norm": 0.4549664556980133,
+      "learning_rate": 0.0015066666666666666,
+      "loss": 1.9103,
+      "step": 3330
+    },
+    {
+      "epoch": 12.37,
+      "grad_norm": 0.45052316784858704,
+      "learning_rate": 0.0015051851851851851,
+      "loss": 1.8733,
+      "step": 3340
+    },
+    {
+      "epoch": 12.41,
+      "grad_norm": 0.5009397268295288,
+      "learning_rate": 0.0015037037037037037,
+      "loss": 1.9003,
+      "step": 3350
+    },
+    {
+      "epoch": 12.44,
+      "grad_norm": 0.45916783809661865,
+      "learning_rate": 0.001502222222222222,
+      "loss": 1.9181,
+      "step": 3360
+    },
+    {
+      "epoch": 12.48,
+      "grad_norm": 0.4555447995662689,
+      "learning_rate": 0.0015007407407407406,
+      "loss": 1.9055,
+      "step": 3370
+    },
+    {
+      "epoch": 12.52,
+      "grad_norm": 0.42393794655799866,
+      "learning_rate": 0.0014992592592592594,
+      "loss": 1.8778,
+      "step": 3380
+    },
+    {
+      "epoch": 12.56,
+      "grad_norm": 0.41853970289230347,
+      "learning_rate": 0.001497777777777778,
+      "loss": 1.9385,
+      "step": 3390
+    },
+    {
+      "epoch": 12.59,
+      "grad_norm": 0.44111889600753784,
+      "learning_rate": 0.0014962962962962963,
+      "loss": 1.9134,
+      "step": 3400
+    },
+    {
+      "epoch": 12.63,
+      "grad_norm": 0.4621415138244629,
+      "learning_rate": 0.001494814814814815,
+      "loss": 1.9066,
+      "step": 3410
+    },
+    {
+      "epoch": 12.67,
+      "grad_norm": 0.4527168869972229,
+      "learning_rate": 0.0014933333333333335,
+      "loss": 1.9641,
+      "step": 3420
+    },
+    {
+      "epoch": 12.7,
+      "grad_norm": 0.4204464554786682,
+      "learning_rate": 0.0014918518518518518,
+      "loss": 1.9363,
+      "step": 3430
+    },
+    {
+      "epoch": 12.74,
+      "grad_norm": 0.44409334659576416,
+      "learning_rate": 0.0014903703703703704,
+      "loss": 1.9602,
+      "step": 3440
+    },
+    {
+      "epoch": 12.78,
+      "grad_norm": 0.42912226915359497,
+      "learning_rate": 0.001488888888888889,
+      "loss": 1.9766,
+      "step": 3450
+    },
+    {
+      "epoch": 12.81,
+      "grad_norm": 0.440065860748291,
+      "learning_rate": 0.0014874074074074076,
+      "loss": 1.9631,
+      "step": 3460
+    },
+    {
+      "epoch": 12.85,
+      "grad_norm": 0.528200626373291,
+      "learning_rate": 0.001485925925925926,
+      "loss": 1.9108,
+      "step": 3470
+    },
+    {
+      "epoch": 12.89,
+      "grad_norm": 0.41785484552383423,
+      "learning_rate": 0.0014844444444444445,
+      "loss": 1.9691,
+      "step": 3480
+    },
+    {
+      "epoch": 12.93,
+      "grad_norm": 0.4229365885257721,
+      "learning_rate": 0.001482962962962963,
+      "loss": 1.9762,
+      "step": 3490
+    },
+    {
+      "epoch": 12.96,
+      "grad_norm": 0.43063077330589294,
+      "learning_rate": 0.0014814814814814814,
+      "loss": 1.9718,
+      "step": 3500
+    },
+    {
+      "epoch": 13.0,
+      "grad_norm": 0.9820219278335571,
+      "learning_rate": 0.00148,
+      "loss": 1.9837,
+      "step": 3510
+    },
+    {
+      "epoch": 13.04,
+      "grad_norm": 0.4445718824863434,
+      "learning_rate": 0.0014785185185185185,
+      "loss": 1.6957,
+      "step": 3520
+    },
+    {
+      "epoch": 13.07,
+      "grad_norm": 0.49743878841400146,
+      "learning_rate": 0.001477037037037037,
+      "loss": 1.7397,
+      "step": 3530
+    },
+    {
+      "epoch": 13.11,
+      "grad_norm": 0.45971688628196716,
+      "learning_rate": 0.0014755555555555555,
+      "loss": 1.7366,
+      "step": 3540
+    },
+    {
+      "epoch": 13.15,
+      "grad_norm": 0.4774453639984131,
+      "learning_rate": 0.001474074074074074,
+      "loss": 1.778,
+      "step": 3550
+    },
+    {
+      "epoch": 13.19,
+      "grad_norm": 0.4667341411113739,
+      "learning_rate": 0.0014725925925925926,
+      "loss": 1.7641,
+      "step": 3560
+    },
+    {
+      "epoch": 13.22,
+      "grad_norm": 0.5121150612831116,
+      "learning_rate": 0.001471111111111111,
+      "loss": 1.8191,
+      "step": 3570
+    },
+    {
+      "epoch": 13.26,
+      "grad_norm": 0.49322351813316345,
+      "learning_rate": 0.0014696296296296298,
+      "loss": 1.8065,
+      "step": 3580
+    },
+    {
+      "epoch": 13.3,
+      "grad_norm": 0.4913088083267212,
+      "learning_rate": 0.0014681481481481483,
+      "loss": 1.834,
+      "step": 3590
+    },
+    {
+      "epoch": 13.33,
+      "grad_norm": 0.45339229702949524,
+      "learning_rate": 0.0014666666666666667,
+      "loss": 1.8561,
+      "step": 3600
+    },
+    {
+      "epoch": 13.37,
+      "grad_norm": 0.4763079285621643,
+      "learning_rate": 0.0014651851851851853,
+      "loss": 1.844,
+      "step": 3610
+    },
+    {
+      "epoch": 13.41,
+      "grad_norm": 0.5057812929153442,
+      "learning_rate": 0.0014637037037037038,
+      "loss": 1.8698,
+      "step": 3620
+    },
+    {
+      "epoch": 13.44,
+      "grad_norm": 0.4484459459781647,
+      "learning_rate": 0.0014622222222222224,
+      "loss": 1.854,
+      "step": 3630
+    },
+    {
+      "epoch": 13.48,
+      "grad_norm": 0.45381781458854675,
+      "learning_rate": 0.0014607407407407407,
+      "loss": 1.8449,
+      "step": 3640
+    },
+    {
+      "epoch": 13.52,
+      "grad_norm": 0.49549397826194763,
+      "learning_rate": 0.0014592592592592593,
+      "loss": 1.8873,
+      "step": 3650
+    },
+    {
+      "epoch": 13.56,
+      "grad_norm": 0.44379979372024536,
+      "learning_rate": 0.0014577777777777779,
+      "loss": 1.8636,
+      "step": 3660
+    },
+    {
+      "epoch": 13.59,
+      "grad_norm": 0.5110834836959839,
+      "learning_rate": 0.0014562962962962962,
+      "loss": 1.8945,
+      "step": 3670
+    },
+    {
+      "epoch": 13.63,
+      "grad_norm": 0.45052438974380493,
+      "learning_rate": 0.0014548148148148148,
+      "loss": 1.9166,
+      "step": 3680
+    },
+    {
+      "epoch": 13.67,
+      "grad_norm": 0.44467103481292725,
+      "learning_rate": 0.0014533333333333334,
+      "loss": 1.8971,
+      "step": 3690
+    },
+    {
+      "epoch": 13.7,
+      "grad_norm": 0.46108537912368774,
+      "learning_rate": 0.001451851851851852,
+      "loss": 1.8986,
+      "step": 3700
+    },
+    {
+      "epoch": 13.74,
+      "grad_norm": 0.4929622411727905,
+      "learning_rate": 0.0014503703703703703,
+      "loss": 1.9173,
+      "step": 3710
+    },
+    {
+      "epoch": 13.78,
+      "grad_norm": 0.45322754979133606,
+      "learning_rate": 0.0014488888888888889,
+      "loss": 1.9376,
+      "step": 3720
+    },
+    {
+      "epoch": 13.81,
+      "grad_norm": 0.43895524740219116,
+      "learning_rate": 0.0014474074074074075,
+      "loss": 1.9394,
+      "step": 3730
+    },
+    {
+      "epoch": 13.85,
+      "grad_norm": 0.43144819140434265,
+      "learning_rate": 0.0014459259259259258,
+      "loss": 1.892,
+      "step": 3740
+    },
+    {
+      "epoch": 13.89,
+      "grad_norm": 0.45700499415397644,
+      "learning_rate": 0.0014444444444444444,
+      "loss": 1.9435,
+      "step": 3750
+    },
+    {
+      "epoch": 13.93,
+      "grad_norm": 0.46102723479270935,
+      "learning_rate": 0.001442962962962963,
+      "loss": 1.9424,
+      "step": 3760
+    },
+    {
+      "epoch": 13.96,
+      "grad_norm": 0.4725353717803955,
+      "learning_rate": 0.0014414814814814815,
+      "loss": 1.9018,
+      "step": 3770
+    },
+    {
+      "epoch": 14.0,
+      "grad_norm": 0.9278309345245361,
+      "learning_rate": 0.0014399999999999999,
+      "loss": 1.9556,
+      "step": 3780
+    },
+    {
+      "epoch": 14.04,
+      "grad_norm": 0.528106153011322,
+      "learning_rate": 0.0014385185185185187,
+      "loss": 1.7123,
+      "step": 3790
+    },
+    {
+      "epoch": 14.07,
+      "grad_norm": 0.5264120101928711,
+      "learning_rate": 0.0014370370370370372,
+      "loss": 1.6959,
+      "step": 3800
+    },
+    {
+      "epoch": 14.11,
+      "grad_norm": 0.5138201117515564,
+      "learning_rate": 0.0014355555555555556,
+      "loss": 1.7195,
+      "step": 3810
+    },
+    {
+      "epoch": 14.15,
+      "grad_norm": 0.5557966232299805,
+      "learning_rate": 0.0014340740740740742,
+      "loss": 1.7021,
+      "step": 3820
+    },
+    {
+      "epoch": 14.19,
+      "grad_norm": 0.4654610753059387,
+      "learning_rate": 0.0014325925925925927,
+      "loss": 1.7411,
+      "step": 3830
+    },
+    {
+      "epoch": 14.22,
+      "grad_norm": 0.49798938632011414,
+      "learning_rate": 0.001431111111111111,
+      "loss": 1.742,
+      "step": 3840
+    },
+    {
+      "epoch": 14.26,
+      "grad_norm": 0.4929085969924927,
+      "learning_rate": 0.0014296296296296297,
+      "loss": 1.7655,
+      "step": 3850
+    },
+    {
+      "epoch": 14.3,
+      "grad_norm": 0.49013039469718933,
+      "learning_rate": 0.0014281481481481482,
+      "loss": 1.7726,
+      "step": 3860
+    },
+    {
+      "epoch": 14.33,
+      "grad_norm": 0.5224959254264832,
+      "learning_rate": 0.0014266666666666668,
+      "loss": 1.7967,
+      "step": 3870
+    },
+    {
+      "epoch": 14.37,
+      "grad_norm": 0.5195993185043335,
+      "learning_rate": 0.0014251851851851851,
+      "loss": 1.8479,
+      "step": 3880
+    },
+    {
+      "epoch": 14.41,
+      "grad_norm": 0.4698856770992279,
+      "learning_rate": 0.0014237037037037037,
+      "loss": 1.8365,
+      "step": 3890
+    },
+    {
+      "epoch": 14.44,
+      "grad_norm": 0.5218276977539062,
+      "learning_rate": 0.0014222222222222223,
+      "loss": 1.8239,
+      "step": 3900
+    },
+    {
+      "epoch": 14.48,
+      "grad_norm": 0.4560031294822693,
+      "learning_rate": 0.0014207407407407406,
+      "loss": 1.8485,
+      "step": 3910
+    },
+    {
+      "epoch": 14.52,
+      "grad_norm": 0.47998443245887756,
+      "learning_rate": 0.0014192592592592592,
+      "loss": 1.8399,
+      "step": 3920
+    },
+    {
+      "epoch": 14.56,
+      "grad_norm": 0.4536432921886444,
+      "learning_rate": 0.0014177777777777778,
+      "loss": 1.8454,
+      "step": 3930
+    },
+    {
+      "epoch": 14.59,
+      "grad_norm": 0.4415784776210785,
+      "learning_rate": 0.0014162962962962964,
+      "loss": 1.8784,
+      "step": 3940
+    },
+    {
+      "epoch": 14.63,
+      "grad_norm": 0.4850631356239319,
+      "learning_rate": 0.0014148148148148147,
+      "loss": 1.8635,
+      "step": 3950
+    },
+    {
+      "epoch": 14.67,
+      "grad_norm": 0.4737834930419922,
+      "learning_rate": 0.0014133333333333333,
+      "loss": 1.8486,
+      "step": 3960
+    },
+    {
+      "epoch": 14.7,
+      "grad_norm": 0.4832935631275177,
+      "learning_rate": 0.0014118518518518519,
+      "loss": 1.8626,
+      "step": 3970
+    },
+    {
+      "epoch": 14.74,
+      "grad_norm": 0.4913930594921112,
+      "learning_rate": 0.0014103703703703702,
+      "loss": 1.9059,
+      "step": 3980
+    },
+    {
+      "epoch": 14.78,
+      "grad_norm": 0.4726288616657257,
+      "learning_rate": 0.001408888888888889,
+      "loss": 1.893,
+      "step": 3990
+    },
+    {
+      "epoch": 14.81,
+      "grad_norm": 0.47808048129081726,
+      "learning_rate": 0.0014074074074074076,
+      "loss": 1.864,
+      "step": 4000
+    },
+    {
+      "epoch": 14.85,
+      "grad_norm": 0.493651419878006,
+      "learning_rate": 0.0014059259259259261,
+      "loss": 1.907,
+      "step": 4010
+    },
+    {
+      "epoch": 14.89,
+      "grad_norm": 0.47489118576049805,
+      "learning_rate": 0.0014044444444444445,
+      "loss": 1.901,
+      "step": 4020
+    },
+    {
+      "epoch": 14.93,
+      "grad_norm": 0.45440223813056946,
+      "learning_rate": 0.001402962962962963,
+      "loss": 1.8882,
+      "step": 4030
+    },
+    {
+      "epoch": 14.96,
+      "grad_norm": 0.4635215103626251,
+      "learning_rate": 0.0014014814814814816,
+      "loss": 1.9037,
+      "step": 4040
+    },
+    {
+      "epoch": 15.0,
+      "grad_norm": 0.9060683846473694,
+      "learning_rate": 0.0014,
+      "loss": 1.8878,
+      "step": 4050
+    },
+    {
+      "epoch": 15.04,
+      "grad_norm": 0.4977479577064514,
+      "learning_rate": 0.0013985185185185186,
+      "loss": 1.6462,
+      "step": 4060
+    },
+    {
+      "epoch": 15.07,
+      "grad_norm": 0.5147350430488586,
+      "learning_rate": 0.0013970370370370371,
+      "loss": 1.6374,
+      "step": 4070
+    },
+    {
+      "epoch": 15.11,
+      "grad_norm": 0.5145394206047058,
+      "learning_rate": 0.0013955555555555557,
+      "loss": 1.6771,
+      "step": 4080
+    },
+    {
+      "epoch": 15.15,
+      "grad_norm": 0.5721226930618286,
+      "learning_rate": 0.001394074074074074,
+      "loss": 1.686,
+      "step": 4090
+    },
+    {
+      "epoch": 15.19,
+      "grad_norm": 0.4931713938713074,
+      "learning_rate": 0.0013925925925925926,
+      "loss": 1.6849,
+      "step": 4100
+    },
+    {
+      "epoch": 15.22,
+      "grad_norm": 0.5027201175689697,
+      "learning_rate": 0.0013911111111111112,
+      "loss": 1.7257,
+      "step": 4110
+    },
+    {
+      "epoch": 15.26,
+      "grad_norm": 0.5069899559020996,
+      "learning_rate": 0.0013896296296296295,
+      "loss": 1.766,
+      "step": 4120
+    },
+    {
+      "epoch": 15.3,
+      "grad_norm": 0.538689136505127,
+      "learning_rate": 0.0013881481481481481,
+      "loss": 1.7425,
+      "step": 4130
+    },
+    {
+      "epoch": 15.33,
+      "grad_norm": 0.4966994822025299,
+      "learning_rate": 0.0013866666666666667,
+      "loss": 1.7532,
+      "step": 4140
+    },
+    {
+      "epoch": 15.37,
+      "grad_norm": 0.504673421382904,
+      "learning_rate": 0.001385185185185185,
+      "loss": 1.7633,
+      "step": 4150
+    },
+    {
+      "epoch": 15.41,
+      "grad_norm": 0.547132134437561,
+      "learning_rate": 0.0013837037037037036,
+      "loss": 1.8001,
+      "step": 4160
+    },
+    {
+      "epoch": 15.44,
+      "grad_norm": 0.5524857044219971,
+      "learning_rate": 0.0013822222222222222,
+      "loss": 1.7945,
+      "step": 4170
+    },
+    {
+      "epoch": 15.48,
+      "grad_norm": 0.5102695822715759,
+      "learning_rate": 0.0013807407407407408,
+      "loss": 1.8037,
+      "step": 4180
+    },
+    {
+      "epoch": 15.52,
+      "grad_norm": 0.48458731174468994,
+      "learning_rate": 0.0013792592592592591,
+      "loss": 1.8401,
+      "step": 4190
+    },
+    {
+      "epoch": 15.56,
+      "grad_norm": 0.4723525643348694,
+      "learning_rate": 0.001377777777777778,
+      "loss": 1.7892,
+      "step": 4200
+    },
+    {
+      "epoch": 15.59,
+      "grad_norm": 0.5020443201065063,
+      "learning_rate": 0.0013762962962962965,
+      "loss": 1.8237,
+      "step": 4210
+    },
+    {
+      "epoch": 15.63,
+      "grad_norm": 0.47924599051475525,
+      "learning_rate": 0.0013748148148148148,
+      "loss": 1.8486,
+      "step": 4220
+    },
+    {
+      "epoch": 15.67,
+      "grad_norm": 0.5095358490943909,
+      "learning_rate": 0.0013733333333333334,
+      "loss": 1.8571,
+      "step": 4230
+    },
+    {
+      "epoch": 15.7,
+      "grad_norm": 0.4978972375392914,
+      "learning_rate": 0.001371851851851852,
+      "loss": 1.8439,
+      "step": 4240
+    },
+    {
+      "epoch": 15.74,
+      "grad_norm": 0.5177919268608093,
+      "learning_rate": 0.0013703703703703705,
+      "loss": 1.8682,
+      "step": 4250
+    },
+    {
+      "epoch": 15.78,
+      "grad_norm": 0.5087350010871887,
+      "learning_rate": 0.0013688888888888889,
+      "loss": 1.8264,
+      "step": 4260
+    },
+    {
+      "epoch": 15.81,
+      "grad_norm": 0.5243315100669861,
+      "learning_rate": 0.0013674074074074075,
+      "loss": 1.8701,
+      "step": 4270
+    },
+    {
+      "epoch": 15.85,
+      "grad_norm": 0.5063629746437073,
+      "learning_rate": 0.001365925925925926,
+      "loss": 1.8565,
+      "step": 4280
+    },
+    {
+      "epoch": 15.89,
+      "grad_norm": 0.4717652201652527,
+      "learning_rate": 0.0013644444444444444,
+      "loss": 1.8555,
+      "step": 4290
+    },
+    {
+      "epoch": 15.93,
+      "grad_norm": 0.4810638725757599,
+      "learning_rate": 0.001362962962962963,
+      "loss": 1.8492,
+      "step": 4300
+    },
+    {
+      "epoch": 15.96,
+      "grad_norm": 0.4636698365211487,
+      "learning_rate": 0.0013614814814814815,
+      "loss": 1.8597,
+      "step": 4310
+    },
+    {
+      "epoch": 16.0,
+      "grad_norm": 0.9389230608940125,
+      "learning_rate": 0.00136,
+      "loss": 1.8731,
+      "step": 4320
+    },
+    {
+      "epoch": 16.04,
+      "grad_norm": 0.5125114917755127,
+      "learning_rate": 0.0013585185185185185,
+      "loss": 1.6402,
+      "step": 4330
+    },
+    {
+      "epoch": 16.07,
+      "grad_norm": 0.5031929612159729,
+      "learning_rate": 0.001357037037037037,
+      "loss": 1.6566,
+      "step": 4340
+    },
+    {
+      "epoch": 16.11,
+      "grad_norm": 0.4980713427066803,
+      "learning_rate": 0.0013555555555555556,
+      "loss": 1.6759,
+      "step": 4350
+    },
+    {
+      "epoch": 16.15,
+      "grad_norm": 0.49954333901405334,
+      "learning_rate": 0.001354074074074074,
+      "loss": 1.6649,
+      "step": 4360
+    },
+    {
+      "epoch": 16.19,
+      "grad_norm": 0.5035192966461182,
+      "learning_rate": 0.0013525925925925925,
+      "loss": 1.6801,
+      "step": 4370
+    },
+    {
+      "epoch": 16.22,
+      "grad_norm": 0.5428767800331116,
+      "learning_rate": 0.001351111111111111,
+      "loss": 1.6574,
+      "step": 4380
+    },
+    {
+      "epoch": 16.26,
+      "grad_norm": 0.5608072280883789,
+      "learning_rate": 0.0013496296296296297,
+      "loss": 1.6877,
+      "step": 4390
+    },
+    {
+      "epoch": 16.3,
+      "grad_norm": 0.5320686101913452,
+      "learning_rate": 0.0013481481481481482,
+      "loss": 1.6884,
+      "step": 4400
+    },
+    {
+      "epoch": 16.33,
+      "grad_norm": 0.5398997664451599,
+      "learning_rate": 0.0013466666666666668,
+      "loss": 1.717,
+      "step": 4410
+    },
+    {
+      "epoch": 16.37,
+      "grad_norm": 0.4949876368045807,
+      "learning_rate": 0.0013451851851851854,
+      "loss": 1.763,
+      "step": 4420
+    },
+    {
+      "epoch": 16.41,
+      "grad_norm": 0.4945550262928009,
+      "learning_rate": 0.0013437037037037037,
+      "loss": 1.7689,
+      "step": 4430
+    },
+    {
+      "epoch": 16.44,
+      "grad_norm": 0.5274211764335632,
+      "learning_rate": 0.0013422222222222223,
+      "loss": 1.7568,
+      "step": 4440
+    },
+    {
+      "epoch": 16.48,
+      "grad_norm": 0.5125142931938171,
+      "learning_rate": 0.0013407407407407409,
+      "loss": 1.782,
+      "step": 4450
+    },
+    {
+      "epoch": 16.52,
+      "grad_norm": 0.4983424246311188,
+      "learning_rate": 0.0013392592592592592,
+      "loss": 1.8036,
+      "step": 4460
+    },
+    {
+      "epoch": 16.56,
+      "grad_norm": 0.49390140175819397,
+      "learning_rate": 0.0013377777777777778,
+      "loss": 1.7774,
+      "step": 4470
+    },
+    {
+      "epoch": 16.59,
+      "grad_norm": 0.4962531626224518,
+      "learning_rate": 0.0013362962962962964,
+      "loss": 1.7785,
+      "step": 4480
+    },
+    {
+      "epoch": 16.63,
+      "grad_norm": 0.4956791400909424,
+      "learning_rate": 0.001334814814814815,
+      "loss": 1.7814,
+      "step": 4490
+    },
+    {
+      "epoch": 16.67,
+      "grad_norm": 0.5397014617919922,
+      "learning_rate": 0.0013333333333333333,
+      "loss": 1.8125,
+      "step": 4500
+    },
+    {
+      "epoch": 16.7,
+      "grad_norm": 0.4916256070137024,
+      "learning_rate": 0.0013318518518518519,
+      "loss": 1.7915,
+      "step": 4510
+    },
+    {
+      "epoch": 16.74,
+      "grad_norm": 0.5432267189025879,
+      "learning_rate": 0.0013303703703703704,
+      "loss": 1.8132,
+      "step": 4520
+    },
+    {
+      "epoch": 16.78,
+      "grad_norm": 0.480970174074173,
+      "learning_rate": 0.0013288888888888888,
+      "loss": 1.7865,
+      "step": 4530
+    },
+    {
+      "epoch": 16.81,
+      "grad_norm": 0.4810551702976227,
+      "learning_rate": 0.0013274074074074074,
+      "loss": 1.8132,
+      "step": 4540
+    },
+    {
+      "epoch": 16.85,
+      "grad_norm": 0.4746386706829071,
+      "learning_rate": 0.001325925925925926,
+      "loss": 1.8493,
+      "step": 4550
+    },
+    {
+      "epoch": 16.89,
+      "grad_norm": 0.5167309641838074,
+      "learning_rate": 0.0013244444444444445,
+      "loss": 1.8126,
+      "step": 4560
+    },
+    {
+      "epoch": 16.93,
+      "grad_norm": 0.49187368154525757,
+      "learning_rate": 0.0013229629629629629,
+      "loss": 1.8484,
+      "step": 4570
+    },
+    {
+      "epoch": 16.96,
+      "grad_norm": 0.5493544340133667,
+      "learning_rate": 0.0013214814814814814,
+      "loss": 1.81,
+      "step": 4580
+    },
+    {
+      "epoch": 17.0,
+      "grad_norm": 1.0746710300445557,
+      "learning_rate": 0.00132,
+      "loss": 1.8312,
+      "step": 4590
+    },
+    {
+      "epoch": 17.04,
+      "grad_norm": 0.5131716132164001,
+      "learning_rate": 0.0013185185185185183,
+      "loss": 1.6069,
+      "step": 4600
+    },
+    {
+      "epoch": 17.07,
+      "grad_norm": 0.5575482845306396,
+      "learning_rate": 0.0013170370370370371,
+      "loss": 1.6018,
+      "step": 4610
+    },
+    {
+      "epoch": 17.11,
+      "grad_norm": 0.5470114350318909,
+      "learning_rate": 0.0013155555555555557,
+      "loss": 1.6125,
+      "step": 4620
+    },
+    {
+      "epoch": 17.15,
+      "grad_norm": 0.5711544752120972,
+      "learning_rate": 0.0013140740740740743,
+      "loss": 1.6186,
+      "step": 4630
+    },
+    {
+      "epoch": 17.19,
+      "grad_norm": 0.550546407699585,
+      "learning_rate": 0.0013125925925925926,
+      "loss": 1.6509,
+      "step": 4640
+    },
+    {
+      "epoch": 17.22,
+      "grad_norm": 0.5357831120491028,
+      "learning_rate": 0.0013111111111111112,
+      "loss": 1.6721,
+      "step": 4650
+    },
+    {
+      "epoch": 17.26,
+      "grad_norm": 0.598581850528717,
+      "learning_rate": 0.0013096296296296298,
+      "loss": 1.6795,
+      "step": 4660
+    },
+    {
+      "epoch": 17.3,
+      "grad_norm": 0.5127149820327759,
+      "learning_rate": 0.0013081481481481481,
+      "loss": 1.6937,
+      "step": 4670
+    },
+    {
+      "epoch": 17.33,
+      "grad_norm": 0.523218035697937,
+      "learning_rate": 0.0013066666666666667,
+      "loss": 1.707,
+      "step": 4680
+    },
+    {
+      "epoch": 17.37,
+      "grad_norm": 0.5636884570121765,
+      "learning_rate": 0.0013051851851851853,
+      "loss": 1.7077,
+      "step": 4690
+    },
+    {
+      "epoch": 17.41,
+      "grad_norm": 0.5461720824241638,
+      "learning_rate": 0.0013037037037037038,
+      "loss": 1.7289,
+      "step": 4700
+    },
+    {
+      "epoch": 17.44,
+      "grad_norm": 0.569513201713562,
+      "learning_rate": 0.0013022222222222222,
+      "loss": 1.7377,
+      "step": 4710
+    },
+    {
+      "epoch": 17.48,
+      "grad_norm": 0.526752769947052,
+      "learning_rate": 0.0013007407407407408,
+      "loss": 1.7277,
+      "step": 4720
+    },
+    {
+      "epoch": 17.52,
+      "grad_norm": 0.5269809365272522,
+      "learning_rate": 0.0012992592592592593,
+      "loss": 1.7526,
+      "step": 4730
+    },
+    {
+      "epoch": 17.56,
+      "grad_norm": 0.5578441023826599,
+      "learning_rate": 0.0012977777777777777,
+      "loss": 1.7372,
+      "step": 4740
+    },
+    {
+      "epoch": 17.59,
+      "grad_norm": 0.5600742101669312,
+      "learning_rate": 0.0012962962962962963,
+      "loss": 1.7519,
+      "step": 4750
+    },
+    {
+      "epoch": 17.63,
+      "grad_norm": 0.5667843818664551,
+      "learning_rate": 0.0012948148148148148,
+      "loss": 1.7624,
+      "step": 4760
+    },
+    {
+      "epoch": 17.67,
+      "grad_norm": 0.5641533732414246,
+      "learning_rate": 0.0012933333333333332,
+      "loss": 1.7674,
+      "step": 4770
+    },
+    {
+      "epoch": 17.7,
+      "grad_norm": 0.49737048149108887,
+      "learning_rate": 0.0012918518518518518,
+      "loss": 1.8026,
+      "step": 4780
+    },
+    {
+      "epoch": 17.74,
+      "grad_norm": 0.5009216666221619,
+      "learning_rate": 0.0012903703703703703,
+      "loss": 1.7699,
+      "step": 4790
+    },
+    {
+      "epoch": 17.78,
+      "grad_norm": 0.5463806986808777,
+      "learning_rate": 0.001288888888888889,
+      "loss": 1.7999,
+      "step": 4800
+    },
+    {
+      "epoch": 17.81,
+      "grad_norm": 0.48484352231025696,
+      "learning_rate": 0.0012874074074074075,
+      "loss": 1.7801,
+      "step": 4810
+    },
+    {
+      "epoch": 17.85,
+      "grad_norm": 0.5290578007698059,
+      "learning_rate": 0.001285925925925926,
+      "loss": 1.771,
+      "step": 4820
+    },
+    {
+      "epoch": 17.89,
+      "grad_norm": 0.5166163444519043,
+      "learning_rate": 0.0012844444444444446,
+      "loss": 1.7913,
+      "step": 4830
+    },
+    {
+      "epoch": 17.93,
+      "grad_norm": 0.502242922782898,
+      "learning_rate": 0.001282962962962963,
+      "loss": 1.7949,
+      "step": 4840
+    },
+    {
+      "epoch": 17.96,
+      "grad_norm": 0.5369807481765747,
+      "learning_rate": 0.0012814814814814815,
+      "loss": 1.7944,
+      "step": 4850
+    },
+    {
+      "epoch": 18.0,
+      "grad_norm": 1.1158689260482788,
+      "learning_rate": 0.00128,
+      "loss": 1.7905,
+      "step": 4860
+    },
+    {
+      "epoch": 18.04,
+      "grad_norm": 0.5336357951164246,
+      "learning_rate": 0.0012785185185185187,
+      "loss": 1.5305,
+      "step": 4870
+    },
+    {
+      "epoch": 18.07,
+      "grad_norm": 0.5469295382499695,
+      "learning_rate": 0.001277037037037037,
+      "loss": 1.5331,
+      "step": 4880
+    },
+    {
+      "epoch": 18.11,
+      "grad_norm": 0.5697037577629089,
+      "learning_rate": 0.0012755555555555556,
+      "loss": 1.5679,
+      "step": 4890
+    },
+    {
+      "epoch": 18.15,
+      "grad_norm": 0.5730404853820801,
+      "learning_rate": 0.0012740740740740742,
+      "loss": 1.6015,
+      "step": 4900
+    },
+    {
+      "epoch": 18.19,
+      "grad_norm": 0.6694100499153137,
+      "learning_rate": 0.0012725925925925925,
+      "loss": 1.6017,
+      "step": 4910
+    },
+    {
+      "epoch": 18.22,
+      "grad_norm": 0.5456265807151794,
+      "learning_rate": 0.001271111111111111,
+      "loss": 1.6246,
+      "step": 4920
+    },
+    {
+      "epoch": 18.26,
+      "grad_norm": 0.6163968443870544,
+      "learning_rate": 0.0012696296296296297,
+      "loss": 1.6349,
+      "step": 4930
+    },
+    {
+      "epoch": 18.3,
+      "grad_norm": 0.5508036613464355,
+      "learning_rate": 0.0012681481481481482,
+      "loss": 1.6618,
+      "step": 4940
+    },
+    {
+      "epoch": 18.33,
+      "grad_norm": 0.5920714735984802,
+      "learning_rate": 0.0012666666666666666,
+      "loss": 1.6871,
+      "step": 4950
+    },
+    {
+      "epoch": 18.37,
+      "grad_norm": 0.5981717109680176,
+      "learning_rate": 0.0012651851851851852,
+      "loss": 1.656,
+      "step": 4960
+    },
+    {
+      "epoch": 18.41,
+      "grad_norm": 0.5558621287345886,
+      "learning_rate": 0.0012637037037037037,
+      "loss": 1.687,
+      "step": 4970
+    },
+    {
+      "epoch": 18.44,
+      "grad_norm": 0.5336385369300842,
+      "learning_rate": 0.001262222222222222,
+      "loss": 1.6954,
+      "step": 4980
+    },
+    {
+      "epoch": 18.48,
+      "grad_norm": 0.56126469373703,
+      "learning_rate": 0.0012607407407407407,
+      "loss": 1.7021,
+      "step": 4990
+    },
+    {
+      "epoch": 18.52,
+      "grad_norm": 0.5533148646354675,
+      "learning_rate": 0.0012592592592592592,
+      "loss": 1.6883,
+      "step": 5000
+    },
+    {
+      "epoch": 18.56,
+      "grad_norm": 0.5625854730606079,
+      "learning_rate": 0.001257777777777778,
+      "loss": 1.7399,
+      "step": 5010
+    },
+    {
+      "epoch": 18.59,
+      "grad_norm": 0.5190121531486511,
+      "learning_rate": 0.0012562962962962964,
+      "loss": 1.728,
+      "step": 5020
+    },
+    {
+      "epoch": 18.63,
+      "grad_norm": 0.5855059027671814,
+      "learning_rate": 0.001254814814814815,
+      "loss": 1.7204,
+      "step": 5030
+    },
+    {
+      "epoch": 18.67,
+      "grad_norm": 0.575351893901825,
+      "learning_rate": 0.0012533333333333335,
+      "loss": 1.7359,
+      "step": 5040
+    },
+    {
+      "epoch": 18.7,
+      "grad_norm": 0.49753502011299133,
+      "learning_rate": 0.0012518518518518519,
+      "loss": 1.7195,
+      "step": 5050
+    },
+    {
+      "epoch": 18.74,
+      "grad_norm": 0.5619398355484009,
+      "learning_rate": 0.0012503703703703704,
+      "loss": 1.7515,
+      "step": 5060
+    },
+    {
+      "epoch": 18.78,
+      "grad_norm": 0.5172871351242065,
+      "learning_rate": 0.001248888888888889,
+      "loss": 1.7777,
+      "step": 5070
+    },
+    {
+      "epoch": 18.81,
+      "grad_norm": 0.5624790191650391,
+      "learning_rate": 0.0012474074074074074,
+      "loss": 1.782,
+      "step": 5080
+    },
+    {
+      "epoch": 18.85,
+      "grad_norm": 0.5403957366943359,
+      "learning_rate": 0.001245925925925926,
+      "loss": 1.7426,
+      "step": 5090
+    },
+    {
+      "epoch": 18.89,
+      "grad_norm": 0.5816079378128052,
+      "learning_rate": 0.0012444444444444445,
+      "loss": 1.8058,
+      "step": 5100
+    },
+    {
+      "epoch": 18.93,
+      "grad_norm": 0.5828419327735901,
+      "learning_rate": 0.001242962962962963,
+      "loss": 1.8004,
+      "step": 5110
+    },
+    {
+      "epoch": 18.96,
+      "grad_norm": 0.5489016175270081,
+      "learning_rate": 0.0012414814814814814,
+      "loss": 1.7684,
+      "step": 5120
+    },
+    {
+      "epoch": 19.0,
+      "grad_norm": 1.1415126323699951,
+      "learning_rate": 0.00124,
+      "loss": 1.7936,
+      "step": 5130
+    },
+    {
+      "epoch": 19.04,
+      "grad_norm": 0.5856507420539856,
+      "learning_rate": 0.0012385185185185186,
+      "loss": 1.5181,
+      "step": 5140
+    },
+    {
+      "epoch": 19.07,
+      "grad_norm": 0.5820109248161316,
+      "learning_rate": 0.001237037037037037,
+      "loss": 1.5436,
+      "step": 5150
+    },
+    {
+      "epoch": 19.11,
+      "grad_norm": 0.591582715511322,
+      "learning_rate": 0.0012355555555555555,
+      "loss": 1.5365,
+      "step": 5160
+    },
+    {
+      "epoch": 19.15,
+      "grad_norm": 0.5697777271270752,
+      "learning_rate": 0.001234074074074074,
+      "loss": 1.5799,
+      "step": 5170
+    },
+    {
+      "epoch": 19.19,
+      "grad_norm": 0.5221427083015442,
+      "learning_rate": 0.0012325925925925926,
+      "loss": 1.606,
+      "step": 5180
+    },
+    {
+      "epoch": 19.22,
+      "grad_norm": 0.5956182479858398,
+      "learning_rate": 0.001231111111111111,
+      "loss": 1.6102,
+      "step": 5190
+    },
+    {
+      "epoch": 19.26,
+      "grad_norm": 0.5815330743789673,
+      "learning_rate": 0.0012296296296296296,
+      "loss": 1.6191,
+      "step": 5200
+    },
+    {
+      "epoch": 19.3,
+      "grad_norm": 0.5627025365829468,
+      "learning_rate": 0.0012281481481481481,
+      "loss": 1.629,
+      "step": 5210
+    },
+    {
+      "epoch": 19.33,
+      "grad_norm": 0.5641065835952759,
+      "learning_rate": 0.0012266666666666667,
+      "loss": 1.6447,
+      "step": 5220
+    },
+    {
+      "epoch": 19.37,
+      "grad_norm": 0.5593487024307251,
+      "learning_rate": 0.0012251851851851853,
+      "loss": 1.6571,
+      "step": 5230
+    },
+    {
+      "epoch": 19.41,
+      "grad_norm": 0.570100724697113,
+      "learning_rate": 0.0012237037037037038,
+      "loss": 1.6696,
+      "step": 5240
+    },
+    {
+      "epoch": 19.44,
+      "grad_norm": 0.5685579776763916,
+      "learning_rate": 0.0012222222222222224,
+      "loss": 1.664,
+      "step": 5250
+    },
+    {
+      "epoch": 19.48,
+      "grad_norm": 0.5899671912193298,
+      "learning_rate": 0.0012207407407407408,
+      "loss": 1.6643,
+      "step": 5260
+    },
+    {
+      "epoch": 19.52,
+      "grad_norm": 0.5786183476448059,
+      "learning_rate": 0.0012192592592592593,
+      "loss": 1.6632,
+      "step": 5270
+    },
+    {
+      "epoch": 19.56,
+      "grad_norm": 0.5458264946937561,
+      "learning_rate": 0.001217777777777778,
+      "loss": 1.6924,
+      "step": 5280
+    },
+    {
+      "epoch": 19.59,
+      "grad_norm": 0.5607069730758667,
+      "learning_rate": 0.0012162962962962963,
+      "loss": 1.6996,
+      "step": 5290
+    },
+    {
+      "epoch": 19.63,
+      "grad_norm": 0.5821597576141357,
+      "learning_rate": 0.0012148148148148148,
+      "loss": 1.7002,
+      "step": 5300
+    },
+    {
+      "epoch": 19.67,
+      "grad_norm": 0.5918251872062683,
+      "learning_rate": 0.0012133333333333334,
+      "loss": 1.6962,
+      "step": 5310
+    },
+    {
+      "epoch": 19.7,
+      "grad_norm": 0.5013391971588135,
+      "learning_rate": 0.001211851851851852,
+      "loss": 1.7107,
+      "step": 5320
+    },
+    {
+      "epoch": 19.74,
+      "grad_norm": 0.5495138168334961,
+      "learning_rate": 0.0012103703703703703,
+      "loss": 1.7032,
+      "step": 5330
+    },
+    {
+      "epoch": 19.78,
+      "grad_norm": 0.565083384513855,
+      "learning_rate": 0.001208888888888889,
+      "loss": 1.7078,
+      "step": 5340
+    },
+    {
+      "epoch": 19.81,
+      "grad_norm": 0.6654185652732849,
+      "learning_rate": 0.0012074074074074075,
+      "loss": 1.7146,
+      "step": 5350
+    },
+    {
+      "epoch": 19.85,
+      "grad_norm": 0.5463594198226929,
+      "learning_rate": 0.0012059259259259258,
+      "loss": 1.7316,
+      "step": 5360
+    },
+    {
+      "epoch": 19.89,
+      "grad_norm": 0.5753750205039978,
+      "learning_rate": 0.0012044444444444444,
+      "loss": 1.7339,
+      "step": 5370
+    },
+    {
+      "epoch": 19.93,
+      "grad_norm": 0.5591763854026794,
+      "learning_rate": 0.001202962962962963,
+      "loss": 1.7212,
+      "step": 5380
+    },
+    {
+      "epoch": 19.96,
+      "grad_norm": 0.5575348734855652,
+      "learning_rate": 0.0012014814814814813,
+      "loss": 1.7412,
+      "step": 5390
+    },
+    {
+      "epoch": 20.0,
+      "grad_norm": 1.3188482522964478,
+      "learning_rate": 0.0012,
+      "loss": 1.7287,
+      "step": 5400
+    },
+    {
+      "epoch": 20.04,
+      "grad_norm": 0.5789737701416016,
+      "learning_rate": 0.0011985185185185185,
+      "loss": 1.4605,
+      "step": 5410
+    },
+    {
+      "epoch": 20.07,
+      "grad_norm": 0.6385830044746399,
+      "learning_rate": 0.0011970370370370373,
+      "loss": 1.4666,
+      "step": 5420
+    },
+    {
+      "epoch": 20.11,
+      "grad_norm": 0.6019684076309204,
+      "learning_rate": 0.0011955555555555556,
+      "loss": 1.5268,
+      "step": 5430
+    },
+    {
+      "epoch": 20.15,
+      "grad_norm": 0.5914923548698425,
+      "learning_rate": 0.0011940740740740742,
+      "loss": 1.5409,
+      "step": 5440
+    },
+    {
+      "epoch": 20.19,
+      "grad_norm": 0.6487402319908142,
+      "learning_rate": 0.0011925925925925927,
+      "loss": 1.5568,
+      "step": 5450
+    },
+    {
+      "epoch": 20.22,
+      "grad_norm": 0.6002733707427979,
+      "learning_rate": 0.001191111111111111,
+      "loss": 1.5492,
+      "step": 5460
+    },
+    {
+      "epoch": 20.26,
+      "grad_norm": 0.6476941108703613,
+      "learning_rate": 0.0011896296296296297,
+      "loss": 1.5677,
+      "step": 5470
+    },
+    {
+      "epoch": 20.3,
+      "grad_norm": 0.5633615851402283,
+      "learning_rate": 0.0011881481481481482,
+      "loss": 1.6167,
+      "step": 5480
+    },
+    {
+      "epoch": 20.33,
+      "grad_norm": 0.5733054876327515,
+      "learning_rate": 0.0011866666666666668,
+      "loss": 1.6335,
+      "step": 5490
+    },
+    {
+      "epoch": 20.37,
+      "grad_norm": 0.5960119366645813,
+      "learning_rate": 0.0011851851851851852,
+      "loss": 1.5886,
+      "step": 5500
+    },
+    {
+      "epoch": 20.41,
+      "grad_norm": 0.5275907516479492,
+      "learning_rate": 0.0011837037037037037,
+      "loss": 1.624,
+      "step": 5510
+    },
+    {
+      "epoch": 20.44,
+      "grad_norm": 0.6697429418563843,
+      "learning_rate": 0.0011822222222222223,
+      "loss": 1.6462,
+      "step": 5520
+    },
+    {
+      "epoch": 20.48,
+      "grad_norm": 0.5955538749694824,
+      "learning_rate": 0.0011807407407407407,
+      "loss": 1.6311,
+      "step": 5530
+    },
+    {
+      "epoch": 20.52,
+      "grad_norm": 0.6113746166229248,
+      "learning_rate": 0.0011792592592592592,
+      "loss": 1.6212,
+      "step": 5540
+    },
+    {
+      "epoch": 20.56,
+      "grad_norm": 0.5709012150764465,
+      "learning_rate": 0.0011777777777777778,
+      "loss": 1.6609,
+      "step": 5550
+    },
+    {
+      "epoch": 20.59,
+      "grad_norm": 0.6236965656280518,
+      "learning_rate": 0.0011762962962962964,
+      "loss": 1.6794,
+      "step": 5560
+    },
+    {
+      "epoch": 20.63,
+      "grad_norm": 0.5546655654907227,
+      "learning_rate": 0.0011748148148148147,
+      "loss": 1.6809,
+      "step": 5570
+    },
+    {
+      "epoch": 20.67,
+      "grad_norm": 0.5919294357299805,
+      "learning_rate": 0.0011733333333333333,
+      "loss": 1.6551,
+      "step": 5580
+    },
+    {
+      "epoch": 20.7,
+      "grad_norm": 0.5872615575790405,
+      "learning_rate": 0.0011718518518518519,
+      "loss": 1.6593,
+      "step": 5590
+    },
+    {
+      "epoch": 20.74,
+      "grad_norm": 0.557774543762207,
+      "learning_rate": 0.0011703703703703702,
+      "loss": 1.6974,
+      "step": 5600
+    },
+    {
+      "epoch": 20.78,
+      "grad_norm": 0.559350311756134,
+      "learning_rate": 0.0011688888888888888,
+      "loss": 1.6906,
+      "step": 5610
+    },
+    {
+      "epoch": 20.81,
+      "grad_norm": 0.59498530626297,
+      "learning_rate": 0.0011674074074074074,
+      "loss": 1.7249,
+      "step": 5620
+    },
+    {
+      "epoch": 20.85,
+      "grad_norm": 0.6189757585525513,
+      "learning_rate": 0.0011659259259259262,
+      "loss": 1.6822,
+      "step": 5630
+    },
+    {
+      "epoch": 20.89,
+      "grad_norm": 0.5750965476036072,
+      "learning_rate": 0.0011644444444444445,
+      "loss": 1.6976,
+      "step": 5640
+    },
+    {
+      "epoch": 20.93,
+      "grad_norm": 0.6109630465507507,
+      "learning_rate": 0.001162962962962963,
+      "loss": 1.704,
+      "step": 5650
+    },
+    {
+      "epoch": 20.96,
+      "grad_norm": 0.5818935036659241,
+      "learning_rate": 0.0011614814814814817,
+      "loss": 1.701,
+      "step": 5660
+    },
+    {
+      "epoch": 21.0,
+      "grad_norm": 1.055619716644287,
+      "learning_rate": 0.00116,
+      "loss": 1.738,
+      "step": 5670
+    },
+    {
+      "epoch": 21.04,
+      "grad_norm": 0.6038652062416077,
+      "learning_rate": 0.0011585185185185186,
+      "loss": 1.4224,
+      "step": 5680
+    },
+    {
+      "epoch": 21.07,
+      "grad_norm": 0.7057866454124451,
+      "learning_rate": 0.0011570370370370371,
+      "loss": 1.4838,
+      "step": 5690
+    },
+    {
+      "epoch": 21.11,
+      "grad_norm": 0.6207330226898193,
+      "learning_rate": 0.0011555555555555555,
+      "loss": 1.5158,
+      "step": 5700
+    },
+    {
+      "epoch": 21.15,
+      "grad_norm": 0.6148425340652466,
+      "learning_rate": 0.001154074074074074,
+      "loss": 1.4815,
+      "step": 5710
+    },
+    {
+      "epoch": 21.19,
+      "grad_norm": 0.6946470737457275,
+      "learning_rate": 0.0011525925925925926,
+      "loss": 1.5331,
+      "step": 5720
+    },
+    {
+      "epoch": 21.22,
+      "grad_norm": 0.6209535002708435,
+      "learning_rate": 0.0011511111111111112,
+      "loss": 1.5187,
+      "step": 5730
+    },
+    {
+      "epoch": 21.26,
+      "grad_norm": 0.6300204396247864,
+      "learning_rate": 0.0011496296296296296,
+      "loss": 1.5572,
+      "step": 5740
+    },
+    {
+      "epoch": 21.3,
+      "grad_norm": 0.594424843788147,
+      "learning_rate": 0.0011481481481481481,
+      "loss": 1.5599,
+      "step": 5750
+    },
+    {
+      "epoch": 21.33,
+      "grad_norm": 0.6507193446159363,
+      "learning_rate": 0.0011466666666666667,
+      "loss": 1.5839,
+      "step": 5760
+    },
+    {
+      "epoch": 21.37,
+      "grad_norm": 0.5893404483795166,
+      "learning_rate": 0.001145185185185185,
+      "loss": 1.5907,
+      "step": 5770
+    },
+    {
+      "epoch": 21.41,
+      "grad_norm": 0.6263285875320435,
+      "learning_rate": 0.0011437037037037036,
+      "loss": 1.6002,
+      "step": 5780
+    },
+    {
+      "epoch": 21.44,
+      "grad_norm": 0.6052868366241455,
+      "learning_rate": 0.0011422222222222222,
+      "loss": 1.5703,
+      "step": 5790
+    },
+    {
+      "epoch": 21.48,
+      "grad_norm": 0.6054526567459106,
+      "learning_rate": 0.0011407407407407408,
+      "loss": 1.6044,
+      "step": 5800
+    },
+    {
+      "epoch": 21.52,
+      "grad_norm": 0.6268815994262695,
+      "learning_rate": 0.0011392592592592591,
+      "loss": 1.5775,
+      "step": 5810
+    },
+    {
+      "epoch": 21.56,
+      "grad_norm": 0.6130894422531128,
+      "learning_rate": 0.0011377777777777777,
+      "loss": 1.6355,
+      "step": 5820
+    },
+    {
+      "epoch": 21.59,
+      "grad_norm": 0.5609670281410217,
+      "learning_rate": 0.0011362962962962965,
+      "loss": 1.6457,
+      "step": 5830
+    },
+    {
+      "epoch": 21.63,
+      "grad_norm": 0.5807933807373047,
+      "learning_rate": 0.0011348148148148148,
+      "loss": 1.6233,
+      "step": 5840
+    },
+    {
+      "epoch": 21.67,
+      "grad_norm": 0.6003103256225586,
+      "learning_rate": 0.0011333333333333334,
+      "loss": 1.6452,
+      "step": 5850
+    },
+    {
+      "epoch": 21.7,
+      "grad_norm": 0.5617521405220032,
+      "learning_rate": 0.001131851851851852,
+      "loss": 1.6673,
+      "step": 5860
+    },
+    {
+      "epoch": 21.74,
+      "grad_norm": 0.6276643872261047,
+      "learning_rate": 0.0011303703703703706,
+      "loss": 1.6296,
+      "step": 5870
+    },
+    {
+      "epoch": 21.78,
+      "grad_norm": 0.6255550980567932,
+      "learning_rate": 0.001128888888888889,
+      "loss": 1.666,
+      "step": 5880
+    },
+    {
+      "epoch": 21.81,
+      "grad_norm": 0.625771701335907,
+      "learning_rate": 0.0011274074074074075,
+      "loss": 1.677,
+      "step": 5890
+    },
+    {
+      "epoch": 21.85,
+      "grad_norm": 0.5867087841033936,
+      "learning_rate": 0.001125925925925926,
+      "loss": 1.6738,
+      "step": 5900
+    },
+    {
+      "epoch": 21.89,
+      "grad_norm": 0.6343807578086853,
+      "learning_rate": 0.0011244444444444444,
+      "loss": 1.6722,
+      "step": 5910
+    },
+    {
+      "epoch": 21.93,
+      "grad_norm": 0.5979423522949219,
+      "learning_rate": 0.001122962962962963,
+      "loss": 1.6602,
+      "step": 5920
+    },
+    {
+      "epoch": 21.96,
+      "grad_norm": 0.5572076439857483,
+      "learning_rate": 0.0011214814814814815,
+      "loss": 1.7001,
+      "step": 5930
+    },
+    {
+      "epoch": 22.0,
+      "grad_norm": 1.0608664751052856,
+      "learning_rate": 0.0011200000000000001,
+      "loss": 1.703,
+      "step": 5940
+    },
+    {
+      "epoch": 22.04,
+      "grad_norm": 0.6108244061470032,
+      "learning_rate": 0.0011185185185185185,
+      "loss": 1.4507,
+      "step": 5950
+    },
+    {
+      "epoch": 22.07,
+      "grad_norm": 0.6707803010940552,
+      "learning_rate": 0.001117037037037037,
+      "loss": 1.4117,
+      "step": 5960
+    },
+    {
+      "epoch": 22.11,
+      "grad_norm": 0.6366927623748779,
+      "learning_rate": 0.0011155555555555556,
+      "loss": 1.4474,
+      "step": 5970
+    },
+    {
+      "epoch": 22.15,
+      "grad_norm": 0.5935031175613403,
+      "learning_rate": 0.001114074074074074,
+      "loss": 1.4372,
+      "step": 5980
+    },
+    {
+      "epoch": 22.19,
+      "grad_norm": 0.6634930372238159,
+      "learning_rate": 0.0011125925925925925,
+      "loss": 1.4715,
+      "step": 5990
+    },
+    {
+      "epoch": 22.22,
+      "grad_norm": 0.6681028008460999,
+      "learning_rate": 0.0011111111111111111,
+      "loss": 1.5219,
+      "step": 6000
+    },
+    {
+      "epoch": 22.26,
+      "grad_norm": 0.6176914572715759,
+      "learning_rate": 0.0011096296296296295,
+      "loss": 1.4901,
+      "step": 6010
+    },
+    {
+      "epoch": 22.3,
+      "grad_norm": 0.6895126104354858,
+      "learning_rate": 0.001108148148148148,
+      "loss": 1.5168,
+      "step": 6020
+    },
+    {
+      "epoch": 22.33,
+      "grad_norm": 0.633429765701294,
+      "learning_rate": 0.0011066666666666666,
+      "loss": 1.5659,
+      "step": 6030
+    },
+    {
+      "epoch": 22.37,
+      "grad_norm": 0.6452445387840271,
+      "learning_rate": 0.0011051851851851854,
+      "loss": 1.556,
+      "step": 6040
+    },
+    {
+      "epoch": 22.41,
+      "grad_norm": 0.5586777329444885,
+      "learning_rate": 0.0011037037037037037,
+      "loss": 1.5652,
+      "step": 6050
+    },
+    {
+      "epoch": 22.44,
+      "grad_norm": 0.6635491251945496,
+      "learning_rate": 0.0011022222222222223,
+      "loss": 1.6015,
+      "step": 6060
+    },
+    {
+      "epoch": 22.48,
+      "grad_norm": 0.6505566835403442,
+      "learning_rate": 0.0011007407407407409,
+      "loss": 1.5544,
+      "step": 6070
+    },
+    {
+      "epoch": 22.52,
+      "grad_norm": 0.635430097579956,
+      "learning_rate": 0.0010992592592592592,
+      "loss": 1.5827,
+      "step": 6080
+    },
+    {
+      "epoch": 22.56,
+      "grad_norm": 0.602946937084198,
+      "learning_rate": 0.0010977777777777778,
+      "loss": 1.5972,
+      "step": 6090
+    },
+    {
+      "epoch": 22.59,
+      "grad_norm": 0.6203567981719971,
+      "learning_rate": 0.0010962962962962964,
+      "loss": 1.6029,
+      "step": 6100
+    },
+    {
+      "epoch": 22.63,
+      "grad_norm": 0.6036391258239746,
+      "learning_rate": 0.001094814814814815,
+      "loss": 1.6075,
+      "step": 6110
+    },
+    {
+      "epoch": 22.67,
+      "grad_norm": 0.6256942749023438,
+      "learning_rate": 0.0010933333333333333,
+      "loss": 1.6119,
+      "step": 6120
+    },
+    {
+      "epoch": 22.7,
+      "grad_norm": 0.6039266586303711,
+      "learning_rate": 0.0010918518518518519,
+      "loss": 1.5997,
+      "step": 6130
+    },
+    {
+      "epoch": 22.74,
+      "grad_norm": 0.6471481323242188,
+      "learning_rate": 0.0010903703703703705,
+      "loss": 1.6425,
+      "step": 6140
+    },
+    {
+      "epoch": 22.78,
+      "grad_norm": 0.6322785019874573,
+      "learning_rate": 0.0010888888888888888,
+      "loss": 1.595,
+      "step": 6150
+    },
+    {
+      "epoch": 22.81,
+      "grad_norm": 0.6034554839134216,
+      "learning_rate": 0.0010874074074074074,
+      "loss": 1.6375,
+      "step": 6160
+    },
+    {
+      "epoch": 22.85,
+      "grad_norm": 0.636727511882782,
+      "learning_rate": 0.001085925925925926,
+      "loss": 1.6694,
+      "step": 6170
+    },
+    {
+      "epoch": 22.89,
+      "grad_norm": 0.6205847859382629,
+      "learning_rate": 0.0010844444444444445,
+      "loss": 1.6411,
+      "step": 6180
+    },
+    {
+      "epoch": 22.93,
+      "grad_norm": 0.6346337795257568,
+      "learning_rate": 0.0010829629629629629,
+      "loss": 1.6344,
+      "step": 6190
+    },
+    {
+      "epoch": 22.96,
+      "grad_norm": 0.6167390942573547,
+      "learning_rate": 0.0010814814814814814,
+      "loss": 1.6659,
+      "step": 6200
+    },
+    {
+      "epoch": 23.0,
+      "grad_norm": 1.001674771308899,
+      "learning_rate": 0.00108,
+      "loss": 1.6554,
+      "step": 6210
+    },
+    {
+      "epoch": 23.04,
+      "grad_norm": 0.6658300757408142,
+      "learning_rate": 0.0010785185185185184,
+      "loss": 1.4168,
+      "step": 6220
+    },
+    {
+      "epoch": 23.07,
+      "grad_norm": 0.6166800856590271,
+      "learning_rate": 0.001077037037037037,
+      "loss": 1.4117,
+      "step": 6230
+    },
+    {
+      "epoch": 23.11,
+      "grad_norm": 0.6343493461608887,
+      "learning_rate": 0.0010755555555555557,
+      "loss": 1.4265,
+      "step": 6240
+    },
+    {
+      "epoch": 23.15,
+      "grad_norm": 0.6020140647888184,
+      "learning_rate": 0.0010740740740740743,
+      "loss": 1.4226,
+      "step": 6250
+    },
+    {
+      "epoch": 23.19,
+      "grad_norm": 0.6626644134521484,
+      "learning_rate": 0.0010725925925925927,
+      "loss": 1.4645,
+      "step": 6260
+    },
+    {
+      "epoch": 23.22,
+      "grad_norm": 0.7044302821159363,
+      "learning_rate": 0.0010711111111111112,
+      "loss": 1.4651,
+      "step": 6270
+    },
+    {
+      "epoch": 23.26,
+      "grad_norm": 0.7018380165100098,
+      "learning_rate": 0.0010696296296296298,
+      "loss": 1.4781,
+      "step": 6280
+    },
+    {
+      "epoch": 23.3,
+      "grad_norm": 0.6937040686607361,
+      "learning_rate": 0.0010681481481481481,
+      "loss": 1.5295,
+      "step": 6290
+    },
+    {
+      "epoch": 23.33,
+      "grad_norm": 0.6271573901176453,
+      "learning_rate": 0.0010666666666666667,
+      "loss": 1.5178,
+      "step": 6300
+    },
+    {
+      "epoch": 23.37,
+      "grad_norm": 0.6487247347831726,
+      "learning_rate": 0.0010651851851851853,
+      "loss": 1.5118,
+      "step": 6310
+    },
+    {
+      "epoch": 23.41,
+      "grad_norm": 0.6135371327400208,
+      "learning_rate": 0.0010637037037037036,
+      "loss": 1.5239,
+      "step": 6320
+    },
+    {
+      "epoch": 23.44,
+      "grad_norm": 0.6121995449066162,
+      "learning_rate": 0.0010622222222222222,
+      "loss": 1.538,
+      "step": 6330
+    },
+    {
+      "epoch": 23.48,
+      "grad_norm": 0.619559645652771,
+      "learning_rate": 0.0010607407407407408,
+      "loss": 1.5498,
+      "step": 6340
+    },
+    {
+      "epoch": 23.52,
+      "grad_norm": 0.6271837949752808,
+      "learning_rate": 0.0010592592592592594,
+      "loss": 1.5702,
+      "step": 6350
+    },
+    {
+      "epoch": 23.56,
+      "grad_norm": 0.6492928266525269,
+      "learning_rate": 0.0010577777777777777,
+      "loss": 1.5458,
+      "step": 6360
+    },
+    {
+      "epoch": 23.59,
+      "grad_norm": 0.6269136071205139,
+      "learning_rate": 0.0010562962962962963,
+      "loss": 1.5948,
+      "step": 6370
+    },
+    {
+      "epoch": 23.63,
+      "grad_norm": 0.6717014908790588,
+      "learning_rate": 0.0010548148148148149,
+      "loss": 1.5589,
+      "step": 6380
+    },
+    {
+      "epoch": 23.67,
+      "grad_norm": 0.6429439783096313,
+      "learning_rate": 0.0010533333333333332,
+      "loss": 1.5839,
+      "step": 6390
+    },
+    {
+      "epoch": 23.7,
+      "grad_norm": 0.6275211572647095,
+      "learning_rate": 0.0010518518518518518,
+      "loss": 1.5769,
+      "step": 6400
+    },
+    {
+      "epoch": 23.74,
+      "grad_norm": 0.6340334415435791,
+      "learning_rate": 0.0010503703703703703,
+      "loss": 1.5764,
+      "step": 6410
+    },
+    {
+      "epoch": 23.78,
+      "grad_norm": 0.6724662780761719,
+      "learning_rate": 0.001048888888888889,
+      "loss": 1.6057,
+      "step": 6420
+    },
+    {
+      "epoch": 23.81,
+      "grad_norm": 0.6650535464286804,
+      "learning_rate": 0.0010474074074074073,
+      "loss": 1.6207,
+      "step": 6430
+    },
+    {
+      "epoch": 23.85,
+      "grad_norm": 0.6134965419769287,
+      "learning_rate": 0.0010459259259259258,
+      "loss": 1.5891,
+      "step": 6440
+    },
+    {
+      "epoch": 23.89,
+      "grad_norm": 0.6735835671424866,
+      "learning_rate": 0.0010444444444444446,
+      "loss": 1.5959,
+      "step": 6450
+    },
+    {
+      "epoch": 23.93,
+      "grad_norm": 0.6774803996086121,
+      "learning_rate": 0.001042962962962963,
+      "loss": 1.6223,
+      "step": 6460
+    },
+    {
+      "epoch": 23.96,
+      "grad_norm": 0.6173749566078186,
+      "learning_rate": 0.0010414814814814816,
+      "loss": 1.6134,
+      "step": 6470
+    },
+    {
+      "epoch": 24.0,
+      "grad_norm": 1.0137830972671509,
+      "learning_rate": 0.0010400000000000001,
+      "loss": 1.6228,
+      "step": 6480
+    },
+    {
+      "epoch": 24.04,
+      "grad_norm": 0.6443659067153931,
+      "learning_rate": 0.0010385185185185187,
+      "loss": 1.3545,
+      "step": 6490
+    },
+    {
+      "epoch": 24.07,
+      "grad_norm": 0.6179562211036682,
+      "learning_rate": 0.001037037037037037,
+      "loss": 1.4058,
+      "step": 6500
+    },
+    {
+      "epoch": 24.11,
+      "grad_norm": 0.6297662258148193,
+      "learning_rate": 0.0010355555555555556,
+      "loss": 1.386,
+      "step": 6510
+    },
+    {
+      "epoch": 24.15,
+      "grad_norm": 0.6821028590202332,
+      "learning_rate": 0.0010340740740740742,
+      "loss": 1.3864,
+      "step": 6520
+    },
+    {
+      "epoch": 24.19,
+      "grad_norm": 0.7033368349075317,
+      "learning_rate": 0.0010325925925925925,
+      "loss": 1.3966,
+      "step": 6530
+    },
+    {
+      "epoch": 24.22,
+      "grad_norm": 0.6669091582298279,
+      "learning_rate": 0.0010311111111111111,
+      "loss": 1.4178,
+      "step": 6540
+    },
+    {
+      "epoch": 24.26,
+      "grad_norm": 0.6988611817359924,
+      "learning_rate": 0.0010296296296296297,
+      "loss": 1.4648,
+      "step": 6550
+    },
+    {
+      "epoch": 24.3,
+      "grad_norm": 0.6371246576309204,
+      "learning_rate": 0.0010281481481481483,
+      "loss": 1.4565,
+      "step": 6560
+    },
+    {
+      "epoch": 24.33,
+      "grad_norm": 0.6623414754867554,
+      "learning_rate": 0.0010266666666666666,
+      "loss": 1.4721,
+      "step": 6570
+    },
+    {
+      "epoch": 24.37,
+      "grad_norm": 0.6429168581962585,
+      "learning_rate": 0.0010251851851851852,
+      "loss": 1.444,
+      "step": 6580
+    },
+    {
+      "epoch": 24.41,
+      "grad_norm": 0.6748073101043701,
+      "learning_rate": 0.0010237037037037038,
+      "loss": 1.5222,
+      "step": 6590
+    },
+    {
+      "epoch": 24.44,
+      "grad_norm": 0.6635433435440063,
+      "learning_rate": 0.0010222222222222221,
+      "loss": 1.5101,
+      "step": 6600
+    },
+    {
+      "epoch": 24.48,
+      "grad_norm": 0.7208848595619202,
+      "learning_rate": 0.0010207407407407407,
+      "loss": 1.5472,
+      "step": 6610
+    },
+    {
+      "epoch": 24.52,
+      "grad_norm": 0.6758795380592346,
+      "learning_rate": 0.0010192592592592593,
+      "loss": 1.504,
+      "step": 6620
+    },
+    {
+      "epoch": 24.56,
+      "grad_norm": 0.7414839863777161,
+      "learning_rate": 0.0010177777777777776,
+      "loss": 1.5277,
+      "step": 6630
+    },
+    {
+      "epoch": 24.59,
+      "grad_norm": 0.6956931948661804,
+      "learning_rate": 0.0010162962962962962,
+      "loss": 1.5416,
+      "step": 6640
+    },
+    {
+      "epoch": 24.63,
+      "grad_norm": 0.669710099697113,
+      "learning_rate": 0.001014814814814815,
+      "loss": 1.5459,
+      "step": 6650
+    },
+    {
+      "epoch": 24.67,
+      "grad_norm": 0.6642506122589111,
+      "learning_rate": 0.0010133333333333335,
+      "loss": 1.5502,
+      "step": 6660
+    },
+    {
+      "epoch": 24.7,
+      "grad_norm": 0.7576358914375305,
+      "learning_rate": 0.0010118518518518519,
+      "loss": 1.5497,
+      "step": 6670
+    },
+    {
+      "epoch": 24.74,
+      "grad_norm": 0.6992085576057434,
+      "learning_rate": 0.0010103703703703705,
+      "loss": 1.5709,
+      "step": 6680
+    },
+    {
+      "epoch": 24.78,
+      "grad_norm": 0.694860577583313,
+      "learning_rate": 0.001008888888888889,
+      "loss": 1.5626,
+      "step": 6690
+    },
+    {
+      "epoch": 24.81,
+      "grad_norm": 0.6058640480041504,
+      "learning_rate": 0.0010074074074074074,
+      "loss": 1.5897,
+      "step": 6700
+    },
+    {
+      "epoch": 24.85,
+      "grad_norm": 0.7034140229225159,
+      "learning_rate": 0.001005925925925926,
+      "loss": 1.5878,
+      "step": 6710
+    },
+    {
+      "epoch": 24.89,
+      "grad_norm": 0.6275174617767334,
+      "learning_rate": 0.0010044444444444445,
+      "loss": 1.5975,
+      "step": 6720
+    },
+    {
+      "epoch": 24.93,
+      "grad_norm": 0.6417464017868042,
+      "learning_rate": 0.001002962962962963,
+      "loss": 1.5888,
+      "step": 6730
+    },
+    {
+      "epoch": 24.96,
+      "grad_norm": 0.5986396074295044,
+      "learning_rate": 0.0010014814814814814,
+      "loss": 1.6057,
+      "step": 6740
+    },
+    {
+      "epoch": 25.0,
+      "grad_norm": 1.4795289039611816,
+      "learning_rate": 0.001,
+      "loss": 1.5858,
+      "step": 6750
+    },
+    {
+      "epoch": 25.04,
+      "grad_norm": 0.6829366087913513,
+      "learning_rate": 0.0009985185185185186,
+      "loss": 1.3255,
+      "step": 6760
+    },
+    {
+      "epoch": 25.07,
+      "grad_norm": 0.6618345975875854,
+      "learning_rate": 0.0009970370370370372,
+      "loss": 1.3511,
+      "step": 6770
+    },
+    {
+      "epoch": 25.11,
+      "grad_norm": 0.6847849488258362,
+      "learning_rate": 0.0009955555555555555,
+      "loss": 1.3619,
+      "step": 6780
+    },
+    {
+      "epoch": 25.15,
+      "grad_norm": 0.7433702945709229,
+      "learning_rate": 0.000994074074074074,
+      "loss": 1.3861,
+      "step": 6790
+    },
+    {
+      "epoch": 25.19,
+      "grad_norm": 0.7247262597084045,
+      "learning_rate": 0.0009925925925925927,
+      "loss": 1.4214,
+      "step": 6800
+    },
+    {
+      "epoch": 25.22,
+      "grad_norm": 0.6940606236457825,
+      "learning_rate": 0.0009911111111111112,
+      "loss": 1.4373,
+      "step": 6810
+    },
+    {
+      "epoch": 25.26,
+      "grad_norm": 0.6243559122085571,
+      "learning_rate": 0.0009896296296296296,
+      "loss": 1.419,
+      "step": 6820
+    },
+    {
+      "epoch": 25.3,
+      "grad_norm": 0.7156795859336853,
+      "learning_rate": 0.0009881481481481482,
+      "loss": 1.4338,
+      "step": 6830
+    },
+    {
+      "epoch": 25.33,
+      "grad_norm": 0.6805399060249329,
+      "learning_rate": 0.0009866666666666667,
+      "loss": 1.4278,
+      "step": 6840
+    },
+    {
+      "epoch": 25.37,
+      "grad_norm": 0.7315900325775146,
+      "learning_rate": 0.000985185185185185,
+      "loss": 1.4578,
+      "step": 6850
+    },
+    {
+      "epoch": 25.41,
+      "grad_norm": 0.6801655888557434,
+      "learning_rate": 0.0009837037037037039,
+      "loss": 1.4719,
+      "step": 6860
+    },
+    {
+      "epoch": 25.44,
+      "grad_norm": 0.664529025554657,
+      "learning_rate": 0.0009822222222222222,
+      "loss": 1.4752,
+      "step": 6870
+    },
+    {
+      "epoch": 25.48,
+      "grad_norm": 0.6608176231384277,
+      "learning_rate": 0.0009807407407407408,
+      "loss": 1.4745,
+      "step": 6880
+    },
+    {
+      "epoch": 25.52,
+      "grad_norm": 0.7137537002563477,
+      "learning_rate": 0.0009792592592592594,
+      "loss": 1.4934,
+      "step": 6890
+    },
+    {
+      "epoch": 25.56,
+      "grad_norm": 0.698538064956665,
+      "learning_rate": 0.0009777777777777777,
+      "loss": 1.4909,
+      "step": 6900
+    },
+    {
+      "epoch": 25.59,
+      "grad_norm": 0.6954187750816345,
+      "learning_rate": 0.0009762962962962964,
+      "loss": 1.4949,
+      "step": 6910
+    },
+    {
+      "epoch": 25.63,
+      "grad_norm": 0.7112638354301453,
+      "learning_rate": 0.0009748148148148149,
+      "loss": 1.5047,
+      "step": 6920
+    },
+    {
+      "epoch": 25.67,
+      "grad_norm": 0.6482926607131958,
+      "learning_rate": 0.0009733333333333334,
+      "loss": 1.541,
+      "step": 6930
+    },
+    {
+      "epoch": 25.7,
+      "grad_norm": 0.6862718462944031,
+      "learning_rate": 0.0009718518518518519,
+      "loss": 1.5217,
+      "step": 6940
+    },
+    {
+      "epoch": 25.74,
+      "grad_norm": 0.7731064558029175,
+      "learning_rate": 0.0009703703703703704,
+      "loss": 1.5248,
+      "step": 6950
+    },
+    {
+      "epoch": 25.78,
+      "grad_norm": 0.6840183138847351,
+      "learning_rate": 0.0009688888888888889,
+      "loss": 1.5454,
+      "step": 6960
+    },
+    {
+      "epoch": 25.81,
+      "grad_norm": 0.6388102769851685,
+      "learning_rate": 0.0009674074074074074,
+      "loss": 1.5506,
+      "step": 6970
+    },
+    {
+      "epoch": 25.85,
+      "grad_norm": 0.6708328127861023,
+      "learning_rate": 0.000965925925925926,
+      "loss": 1.5413,
+      "step": 6980
+    },
+    {
+      "epoch": 25.89,
+      "grad_norm": 0.7057149410247803,
+      "learning_rate": 0.0009644444444444444,
+      "loss": 1.5626,
+      "step": 6990
+    },
+    {
+      "epoch": 25.93,
+      "grad_norm": 0.6393432021141052,
+      "learning_rate": 0.0009629629629629629,
+      "loss": 1.5463,
+      "step": 7000
+    },
+    {
+      "epoch": 25.96,
+      "grad_norm": 0.6737004518508911,
+      "learning_rate": 0.0009614814814814816,
+      "loss": 1.5552,
+      "step": 7010
+    },
+    {
+      "epoch": 26.0,
+      "grad_norm": 1.0224347114562988,
+      "learning_rate": 0.00096,
+      "loss": 1.5871,
+      "step": 7020
+    },
+    {
+      "epoch": 26.04,
+      "grad_norm": 0.6330727934837341,
+      "learning_rate": 0.0009585185185185186,
+      "loss": 1.3241,
+      "step": 7030
+    },
+    {
+      "epoch": 26.07,
+      "grad_norm": 0.6640604138374329,
+      "learning_rate": 0.0009570370370370371,
+      "loss": 1.3282,
+      "step": 7040
+    },
+    {
+      "epoch": 26.11,
+      "grad_norm": 0.6334345936775208,
+      "learning_rate": 0.0009555555555555556,
+      "loss": 1.3204,
+      "step": 7050
+    },
+    {
+      "epoch": 26.15,
+      "grad_norm": 0.6973744034767151,
+      "learning_rate": 0.0009540740740740741,
+      "loss": 1.3403,
+      "step": 7060
+    },
+    {
+      "epoch": 26.19,
+      "grad_norm": 0.7151355147361755,
+      "learning_rate": 0.0009525925925925926,
+      "loss": 1.3856,
+      "step": 7070
+    },
+    {
+      "epoch": 26.22,
+      "grad_norm": 0.7948924899101257,
+      "learning_rate": 0.0009511111111111111,
+      "loss": 1.3786,
+      "step": 7080
+    },
+    {
+      "epoch": 26.26,
+      "grad_norm": 0.6959167122840881,
+      "learning_rate": 0.0009496296296296296,
+      "loss": 1.4085,
+      "step": 7090
+    },
+    {
+      "epoch": 26.3,
+      "grad_norm": 0.6958004236221313,
+      "learning_rate": 0.0009481481481481482,
+      "loss": 1.4115,
+      "step": 7100
+    },
+    {
+      "epoch": 26.33,
+      "grad_norm": 0.6725401282310486,
+      "learning_rate": 0.0009466666666666667,
+      "loss": 1.4026,
+      "step": 7110
+    },
+    {
+      "epoch": 26.37,
+      "grad_norm": 0.7028361558914185,
+      "learning_rate": 0.0009451851851851853,
+      "loss": 1.4275,
+      "step": 7120
+    },
+    {
+      "epoch": 26.41,
+      "grad_norm": 0.6960654258728027,
+      "learning_rate": 0.0009437037037037038,
+      "loss": 1.4398,
+      "step": 7130
+    },
+    {
+      "epoch": 26.44,
+      "grad_norm": 0.7060914039611816,
+      "learning_rate": 0.0009422222222222222,
+      "loss": 1.4417,
+      "step": 7140
+    },
+    {
+      "epoch": 26.48,
+      "grad_norm": 0.7186646461486816,
+      "learning_rate": 0.0009407407407407408,
+      "loss": 1.4622,
+      "step": 7150
+    },
+    {
+      "epoch": 26.52,
+      "grad_norm": 0.7396743893623352,
+      "learning_rate": 0.0009392592592592593,
+      "loss": 1.471,
+      "step": 7160
+    },
+    {
+      "epoch": 26.56,
+      "grad_norm": 0.6751599311828613,
+      "learning_rate": 0.0009377777777777778,
+      "loss": 1.4557,
+      "step": 7170
+    },
+    {
+      "epoch": 26.59,
+      "grad_norm": 0.6983835697174072,
+      "learning_rate": 0.0009362962962962963,
+      "loss": 1.4358,
+      "step": 7180
+    },
+    {
+      "epoch": 26.63,
+      "grad_norm": 0.6938909292221069,
+      "learning_rate": 0.0009348148148148148,
+      "loss": 1.4836,
+      "step": 7190
+    },
+    {
+      "epoch": 26.67,
+      "grad_norm": 0.6815834641456604,
+      "learning_rate": 0.0009333333333333333,
+      "loss": 1.4937,
+      "step": 7200
+    },
+    {
+      "epoch": 26.7,
+      "grad_norm": 0.699643611907959,
+      "learning_rate": 0.0009318518518518518,
+      "loss": 1.5198,
+      "step": 7210
+    },
+    {
+      "epoch": 26.74,
+      "grad_norm": 0.6944101452827454,
+      "learning_rate": 0.0009303703703703705,
+      "loss": 1.4792,
+      "step": 7220
+    },
+    {
+      "epoch": 26.78,
+      "grad_norm": 0.7341158390045166,
+      "learning_rate": 0.0009288888888888889,
+      "loss": 1.5256,
+      "step": 7230
+    },
+    {
+      "epoch": 26.81,
+      "grad_norm": 0.6798523664474487,
+      "learning_rate": 0.0009274074074074075,
+      "loss": 1.5082,
+      "step": 7240
+    },
+    {
+      "epoch": 26.85,
+      "grad_norm": 0.696607768535614,
+      "learning_rate": 0.000925925925925926,
+      "loss": 1.5197,
+      "step": 7250
+    },
+    {
+      "epoch": 26.89,
+      "grad_norm": 0.6994819641113281,
+      "learning_rate": 0.0009244444444444444,
+      "loss": 1.4907,
+      "step": 7260
+    },
+    {
+      "epoch": 26.93,
+      "grad_norm": 0.6782405972480774,
+      "learning_rate": 0.000922962962962963,
+      "loss": 1.5202,
+      "step": 7270
+    },
+    {
+      "epoch": 26.96,
+      "grad_norm": 0.701532781124115,
+      "learning_rate": 0.0009214814814814815,
+      "loss": 1.5186,
+      "step": 7280
+    },
+    {
+      "epoch": 27.0,
+      "grad_norm": 1.0853149890899658,
+      "learning_rate": 0.00092,
+      "loss": 1.5585,
+      "step": 7290
+    },
+    {
+      "epoch": 27.04,
+      "grad_norm": 0.6549865007400513,
+      "learning_rate": 0.0009185185185185185,
+      "loss": 1.2778,
+      "step": 7300
+    },
+    {
+      "epoch": 27.07,
+      "grad_norm": 0.7133400440216064,
+      "learning_rate": 0.000917037037037037,
+      "loss": 1.2791,
+      "step": 7310
+    },
+    {
+      "epoch": 27.11,
+      "grad_norm": 0.7217587828636169,
+      "learning_rate": 0.0009155555555555556,
+      "loss": 1.3171,
+      "step": 7320
+    },
+    {
+      "epoch": 27.15,
+      "grad_norm": 0.6442558169364929,
+      "learning_rate": 0.0009140740740740741,
+      "loss": 1.3219,
+      "step": 7330
+    },
+    {
+      "epoch": 27.19,
+      "grad_norm": 0.737054705619812,
+      "learning_rate": 0.0009125925925925927,
+      "loss": 1.3253,
+      "step": 7340
+    },
+    {
+      "epoch": 27.22,
+      "grad_norm": 0.7420579195022583,
+      "learning_rate": 0.0009111111111111111,
+      "loss": 1.3616,
+      "step": 7350
+    },
+    {
+      "epoch": 27.26,
+      "grad_norm": 0.745717465877533,
+      "learning_rate": 0.0009096296296296297,
+      "loss": 1.3606,
+      "step": 7360
+    },
+    {
+      "epoch": 27.3,
+      "grad_norm": 0.7737590074539185,
+      "learning_rate": 0.0009081481481481482,
+      "loss": 1.3716,
+      "step": 7370
+    },
+    {
+      "epoch": 27.33,
+      "grad_norm": 0.7196997404098511,
+      "learning_rate": 0.0009066666666666666,
+      "loss": 1.3784,
+      "step": 7380
+    },
+    {
+      "epoch": 27.37,
+      "grad_norm": 0.6725694537162781,
+      "learning_rate": 0.0009051851851851852,
+      "loss": 1.4092,
+      "step": 7390
+    },
+    {
+      "epoch": 27.41,
+      "grad_norm": 0.7940300703048706,
+      "learning_rate": 0.0009037037037037037,
+      "loss": 1.4181,
+      "step": 7400
+    },
+    {
+      "epoch": 27.44,
+      "grad_norm": 0.7204284071922302,
+      "learning_rate": 0.0009022222222222222,
+      "loss": 1.4147,
+      "step": 7410
+    },
+    {
+      "epoch": 27.48,
+      "grad_norm": 0.6743030548095703,
+      "learning_rate": 0.0009007407407407408,
+      "loss": 1.4266,
+      "step": 7420
+    },
+    {
+      "epoch": 27.52,
+      "grad_norm": 0.6738618612289429,
+      "learning_rate": 0.0008992592592592594,
+      "loss": 1.4105,
+      "step": 7430
+    },
+    {
+      "epoch": 27.56,
+      "grad_norm": 0.7223931550979614,
+      "learning_rate": 0.0008977777777777778,
+      "loss": 1.4684,
+      "step": 7440
+    },
+    {
+      "epoch": 27.59,
+      "grad_norm": 0.7275183200836182,
+      "learning_rate": 0.0008962962962962963,
+      "loss": 1.4572,
+      "step": 7450
+    },
+    {
+      "epoch": 27.63,
+      "grad_norm": 0.7375164031982422,
+      "learning_rate": 0.0008948148148148149,
+      "loss": 1.4505,
+      "step": 7460
+    },
+    {
+      "epoch": 27.67,
+      "grad_norm": 0.6762166023254395,
+      "learning_rate": 0.0008933333333333333,
+      "loss": 1.4677,
+      "step": 7470
+    },
+    {
+      "epoch": 27.7,
+      "grad_norm": 0.7304455041885376,
+      "learning_rate": 0.0008918518518518519,
+      "loss": 1.4723,
+      "step": 7480
+    },
+    {
+      "epoch": 27.74,
+      "grad_norm": 0.7225548028945923,
+      "learning_rate": 0.0008903703703703704,
+      "loss": 1.4739,
+      "step": 7490
+    },
+    {
+      "epoch": 27.78,
+      "grad_norm": 0.7559143304824829,
+      "learning_rate": 0.0008888888888888888,
+      "loss": 1.4707,
+      "step": 7500
+    },
+    {
+      "epoch": 27.81,
+      "grad_norm": 0.770778238773346,
+      "learning_rate": 0.0008874074074074074,
+      "loss": 1.4737,
+      "step": 7510
+    },
+    {
+      "epoch": 27.85,
+      "grad_norm": 0.7308626770973206,
+      "learning_rate": 0.000885925925925926,
+      "loss": 1.4682,
+      "step": 7520
+    },
+    {
+      "epoch": 27.89,
+      "grad_norm": 0.6600290536880493,
+      "learning_rate": 0.0008844444444444445,
+      "loss": 1.4645,
+      "step": 7530
+    },
+    {
+      "epoch": 27.93,
+      "grad_norm": 0.7288909554481506,
+      "learning_rate": 0.000882962962962963,
+      "loss": 1.4978,
+      "step": 7540
+    },
+    {
+      "epoch": 27.96,
+      "grad_norm": 0.7489798069000244,
+      "learning_rate": 0.0008814814814814816,
+      "loss": 1.4994,
+      "step": 7550
+    },
+    {
+      "epoch": 28.0,
+      "grad_norm": 1.4726680517196655,
+      "learning_rate": 0.00088,
+      "loss": 1.5065,
+      "step": 7560
+    },
+    {
+      "epoch": 28.04,
+      "grad_norm": 0.7046101093292236,
+      "learning_rate": 0.0008785185185185185,
+      "loss": 1.2712,
+      "step": 7570
+    },
+    {
+      "epoch": 28.07,
+      "grad_norm": 0.6764820218086243,
+      "learning_rate": 0.0008770370370370371,
+      "loss": 1.2717,
+      "step": 7580
+    },
+    {
+      "epoch": 28.11,
+      "grad_norm": 0.713107705116272,
+      "learning_rate": 0.0008755555555555555,
+      "loss": 1.2657,
+      "step": 7590
+    },
+    {
+      "epoch": 28.15,
+      "grad_norm": 0.7514472603797913,
+      "learning_rate": 0.0008740740740740741,
+      "loss": 1.2947,
+      "step": 7600
+    },
+    {
+      "epoch": 28.19,
+      "grad_norm": 0.7262618541717529,
+      "learning_rate": 0.0008725925925925926,
+      "loss": 1.3297,
+      "step": 7610
+    },
+    {
+      "epoch": 28.22,
+      "grad_norm": 0.8503907322883606,
+      "learning_rate": 0.000871111111111111,
+      "loss": 1.2874,
+      "step": 7620
+    },
+    {
+      "epoch": 28.26,
+      "grad_norm": 0.7442007064819336,
+      "learning_rate": 0.0008696296296296297,
+      "loss": 1.3066,
+      "step": 7630
+    },
+    {
+      "epoch": 28.3,
+      "grad_norm": 0.6790578365325928,
+      "learning_rate": 0.0008681481481481482,
+      "loss": 1.3362,
+      "step": 7640
+    },
+    {
+      "epoch": 28.33,
+      "grad_norm": 0.7511619925498962,
+      "learning_rate": 0.0008666666666666667,
+      "loss": 1.3415,
+      "step": 7650
+    },
+    {
+      "epoch": 28.37,
+      "grad_norm": 0.7373079657554626,
+      "learning_rate": 0.0008651851851851852,
+      "loss": 1.3483,
+      "step": 7660
+    },
+    {
+      "epoch": 28.41,
+      "grad_norm": 0.7537468075752258,
+      "learning_rate": 0.0008637037037037038,
+      "loss": 1.3976,
+      "step": 7670
+    },
+    {
+      "epoch": 28.44,
+      "grad_norm": 0.758500337600708,
+      "learning_rate": 0.0008622222222222222,
+      "loss": 1.3929,
+      "step": 7680
+    },
+    {
+      "epoch": 28.48,
+      "grad_norm": 0.7068962454795837,
+      "learning_rate": 0.0008607407407407407,
+      "loss": 1.3989,
+      "step": 7690
+    },
+    {
+      "epoch": 28.52,
+      "grad_norm": 0.7316468954086304,
+      "learning_rate": 0.0008592592592592593,
+      "loss": 1.4046,
+      "step": 7700
+    },
+    {
+      "epoch": 28.56,
+      "grad_norm": 0.7010143995285034,
+      "learning_rate": 0.0008577777777777777,
+      "loss": 1.4199,
+      "step": 7710
+    },
+    {
+      "epoch": 28.59,
+      "grad_norm": 0.7400304675102234,
+      "learning_rate": 0.0008562962962962963,
+      "loss": 1.4107,
+      "step": 7720
+    },
+    {
+      "epoch": 28.63,
+      "grad_norm": 0.7666124105453491,
+      "learning_rate": 0.0008548148148148149,
+      "loss": 1.4561,
+      "step": 7730
+    },
+    {
+      "epoch": 28.67,
+      "grad_norm": 0.7306181192398071,
+      "learning_rate": 0.0008533333333333334,
+      "loss": 1.4353,
+      "step": 7740
+    },
+    {
+      "epoch": 28.7,
+      "grad_norm": 0.7335529327392578,
+      "learning_rate": 0.0008518518518518519,
+      "loss": 1.4214,
+      "step": 7750
+    },
+    {
+      "epoch": 28.74,
+      "grad_norm": 0.7898188829421997,
+      "learning_rate": 0.0008503703703703704,
+      "loss": 1.4253,
+      "step": 7760
+    },
+    {
+      "epoch": 28.78,
+      "grad_norm": 0.7438576817512512,
+      "learning_rate": 0.0008488888888888889,
+      "loss": 1.4307,
+      "step": 7770
+    },
+    {
+      "epoch": 28.81,
+      "grad_norm": 0.7355637550354004,
+      "learning_rate": 0.0008474074074074074,
+      "loss": 1.4224,
+      "step": 7780
+    },
+    {
+      "epoch": 28.85,
+      "grad_norm": 0.756335973739624,
+      "learning_rate": 0.000845925925925926,
+      "loss": 1.4869,
+      "step": 7790
+    },
+    {
+      "epoch": 28.89,
+      "grad_norm": 0.6840737462043762,
+      "learning_rate": 0.0008444444444444444,
+      "loss": 1.4612,
+      "step": 7800
+    },
+    {
+      "epoch": 28.93,
+      "grad_norm": 0.7832042574882507,
+      "learning_rate": 0.0008429629629629629,
+      "loss": 1.4442,
+      "step": 7810
+    },
+    {
+      "epoch": 28.96,
+      "grad_norm": 0.7849950194358826,
+      "learning_rate": 0.0008414814814814815,
+      "loss": 1.4796,
+      "step": 7820
+    },
+    {
+      "epoch": 29.0,
+      "grad_norm": 1.6340038776397705,
+      "learning_rate": 0.00084,
+      "loss": 1.4447,
+      "step": 7830
+    },
+    {
+      "epoch": 29.04,
+      "grad_norm": 0.7272326350212097,
+      "learning_rate": 0.0008385185185185186,
+      "loss": 1.2243,
+      "step": 7840
+    },
+    {
+      "epoch": 29.07,
+      "grad_norm": 0.796206533908844,
+      "learning_rate": 0.0008370370370370371,
+      "loss": 1.2332,
+      "step": 7850
+    },
+    {
+      "epoch": 29.11,
+      "grad_norm": 0.7212488651275635,
+      "learning_rate": 0.0008355555555555556,
+      "loss": 1.2729,
+      "step": 7860
+    },
+    {
+      "epoch": 29.15,
+      "grad_norm": 0.7437489032745361,
+      "learning_rate": 0.0008340740740740741,
+      "loss": 1.2607,
+      "step": 7870
+    },
+    {
+      "epoch": 29.19,
+      "grad_norm": 0.7367982864379883,
+      "learning_rate": 0.0008325925925925926,
+      "loss": 1.2863,
+      "step": 7880
+    },
+    {
+      "epoch": 29.22,
+      "grad_norm": 0.688751220703125,
+      "learning_rate": 0.0008311111111111111,
+      "loss": 1.302,
+      "step": 7890
+    },
+    {
+      "epoch": 29.26,
+      "grad_norm": 0.7412949800491333,
+      "learning_rate": 0.0008296296296296296,
+      "loss": 1.2973,
+      "step": 7900
+    },
+    {
+      "epoch": 29.3,
+      "grad_norm": 0.7355971336364746,
+      "learning_rate": 0.0008281481481481482,
+      "loss": 1.3144,
+      "step": 7910
+    },
+    {
+      "epoch": 29.33,
+      "grad_norm": 0.7711012363433838,
+      "learning_rate": 0.0008266666666666666,
+      "loss": 1.3428,
+      "step": 7920
+    },
+    {
+      "epoch": 29.37,
+      "grad_norm": 0.7146042585372925,
+      "learning_rate": 0.0008251851851851852,
+      "loss": 1.3259,
+      "step": 7930
+    },
+    {
+      "epoch": 29.41,
+      "grad_norm": 0.7842440605163574,
+      "learning_rate": 0.0008237037037037038,
+      "loss": 1.3353,
+      "step": 7940
+    },
+    {
+      "epoch": 29.44,
+      "grad_norm": 0.7185430526733398,
+      "learning_rate": 0.0008222222222222222,
+      "loss": 1.3476,
+      "step": 7950
+    },
+    {
+      "epoch": 29.48,
+      "grad_norm": 0.7307253479957581,
+      "learning_rate": 0.0008207407407407408,
+      "loss": 1.3576,
+      "step": 7960
+    },
+    {
+      "epoch": 29.52,
+      "grad_norm": 0.7476086020469666,
+      "learning_rate": 0.0008192592592592593,
+      "loss": 1.3491,
+      "step": 7970
+    },
+    {
+      "epoch": 29.56,
+      "grad_norm": 0.773776113986969,
+      "learning_rate": 0.0008177777777777778,
+      "loss": 1.3646,
+      "step": 7980
+    },
+    {
+      "epoch": 29.59,
+      "grad_norm": 0.7761340141296387,
+      "learning_rate": 0.0008162962962962963,
+      "loss": 1.383,
+      "step": 7990
+    },
+    {
+      "epoch": 29.63,
+      "grad_norm": 0.7412744760513306,
+      "learning_rate": 0.0008148148148148148,
+      "loss": 1.4269,
+      "step": 8000
+    },
+    {
+      "epoch": 29.67,
+      "grad_norm": 0.7098486423492432,
+      "learning_rate": 0.0008133333333333333,
+      "loss": 1.3804,
+      "step": 8010
+    },
+    {
+      "epoch": 29.7,
+      "grad_norm": 0.7861642241477966,
+      "learning_rate": 0.0008118518518518518,
+      "loss": 1.4102,
+      "step": 8020
+    },
+    {
+      "epoch": 29.74,
+      "grad_norm": 0.7670470476150513,
+      "learning_rate": 0.0008103703703703705,
+      "loss": 1.3886,
+      "step": 8030
+    },
+    {
+      "epoch": 29.78,
+      "grad_norm": 0.840653121471405,
+      "learning_rate": 0.0008088888888888889,
+      "loss": 1.4109,
+      "step": 8040
+    },
+    {
+      "epoch": 29.81,
+      "grad_norm": 0.7671210169792175,
+      "learning_rate": 0.0008074074074074075,
+      "loss": 1.4307,
+      "step": 8050
+    },
+    {
+      "epoch": 29.85,
+      "grad_norm": 0.7228606343269348,
+      "learning_rate": 0.000805925925925926,
+      "loss": 1.4196,
+      "step": 8060
+    },
+    {
+      "epoch": 29.89,
+      "grad_norm": 0.6415588855743408,
+      "learning_rate": 0.0008044444444444444,
+      "loss": 1.421,
+      "step": 8070
+    },
+    {
+      "epoch": 29.93,
+      "grad_norm": 0.780892014503479,
+      "learning_rate": 0.000802962962962963,
+      "loss": 1.439,
+      "step": 8080
+    },
+    {
+      "epoch": 29.96,
+      "grad_norm": 0.7971543073654175,
+      "learning_rate": 0.0008014814814814815,
+      "loss": 1.425,
+      "step": 8090
+    },
+    {
+      "epoch": 30.0,
+      "grad_norm": 1.307153344154358,
+      "learning_rate": 0.0008,
+      "loss": 1.4416,
+      "step": 8100
+    },
+    {
+      "epoch": 30.04,
+      "grad_norm": 0.7271667122840881,
+      "learning_rate": 0.0007985185185185185,
+      "loss": 1.1745,
+      "step": 8110
+    },
+    {
+      "epoch": 30.07,
+      "grad_norm": 0.7366812825202942,
+      "learning_rate": 0.000797037037037037,
+      "loss": 1.2231,
+      "step": 8120
+    },
+    {
+      "epoch": 30.11,
+      "grad_norm": 0.8225249648094177,
+      "learning_rate": 0.0007955555555555555,
+      "loss": 1.2032,
+      "step": 8130
+    },
+    {
+      "epoch": 30.15,
+      "grad_norm": 0.7685932517051697,
+      "learning_rate": 0.0007940740740740741,
+      "loss": 1.2319,
+      "step": 8140
+    },
+    {
+      "epoch": 30.19,
+      "grad_norm": 0.6946758031845093,
+      "learning_rate": 0.0007925925925925927,
+      "loss": 1.2392,
+      "step": 8150
+    },
+    {
+      "epoch": 30.22,
+      "grad_norm": 0.7242510318756104,
+      "learning_rate": 0.0007911111111111111,
+      "loss": 1.2644,
+      "step": 8160
+    },
+    {
+      "epoch": 30.26,
+      "grad_norm": 0.7538670301437378,
+      "learning_rate": 0.0007896296296296297,
+      "loss": 1.285,
+      "step": 8170
+    },
+    {
+      "epoch": 30.3,
+      "grad_norm": 0.7838382720947266,
+      "learning_rate": 0.0007881481481481482,
+      "loss": 1.2723,
+      "step": 8180
+    },
+    {
+      "epoch": 30.33,
+      "grad_norm": 0.7739988565444946,
+      "learning_rate": 0.0007866666666666666,
+      "loss": 1.288,
+      "step": 8190
+    },
+    {
+      "epoch": 30.37,
+      "grad_norm": 0.758450448513031,
+      "learning_rate": 0.0007851851851851852,
+      "loss": 1.3066,
+      "step": 8200
+    },
+    {
+      "epoch": 30.41,
+      "grad_norm": 0.7945404052734375,
+      "learning_rate": 0.0007837037037037037,
+      "loss": 1.3066,
+      "step": 8210
+    },
+    {
+      "epoch": 30.44,
+      "grad_norm": 0.8125270009040833,
+      "learning_rate": 0.0007822222222222222,
+      "loss": 1.3481,
+      "step": 8220
+    },
+    {
+      "epoch": 30.48,
+      "grad_norm": 0.7620575428009033,
+      "learning_rate": 0.0007807407407407407,
+      "loss": 1.3177,
+      "step": 8230
+    },
+    {
+      "epoch": 30.52,
+      "grad_norm": 0.7492615580558777,
+      "learning_rate": 0.0007792592592592593,
+      "loss": 1.3379,
+      "step": 8240
+    },
+    {
+      "epoch": 30.56,
+      "grad_norm": 0.7885991334915161,
+      "learning_rate": 0.0007777777777777778,
+      "loss": 1.3671,
+      "step": 8250
+    },
+    {
+      "epoch": 30.59,
+      "grad_norm": 0.8088184595108032,
+      "learning_rate": 0.0007762962962962963,
+      "loss": 1.3576,
+      "step": 8260
+    },
+    {
+      "epoch": 30.63,
+      "grad_norm": 0.7370725274085999,
+      "learning_rate": 0.0007748148148148149,
+      "loss": 1.3592,
+      "step": 8270
+    },
+    {
+      "epoch": 30.67,
+      "grad_norm": 0.8647828102111816,
+      "learning_rate": 0.0007733333333333333,
+      "loss": 1.3471,
+      "step": 8280
+    },
+    {
+      "epoch": 30.7,
+      "grad_norm": 0.8166521787643433,
+      "learning_rate": 0.0007718518518518519,
+      "loss": 1.3605,
+      "step": 8290
+    },
+    {
+      "epoch": 30.74,
+      "grad_norm": 0.7414300441741943,
+      "learning_rate": 0.0007703703703703704,
+      "loss": 1.3433,
+      "step": 8300
+    },
+    {
+      "epoch": 30.78,
+      "grad_norm": 0.8628682494163513,
+      "learning_rate": 0.0007688888888888888,
+      "loss": 1.3999,
+      "step": 8310
+    },
+    {
+      "epoch": 30.81,
+      "grad_norm": 0.7411145567893982,
+      "learning_rate": 0.0007674074074074074,
+      "loss": 1.3844,
+      "step": 8320
+    },
+    {
+      "epoch": 30.85,
+      "grad_norm": 0.7552719116210938,
+      "learning_rate": 0.0007659259259259259,
+      "loss": 1.4042,
+      "step": 8330
+    },
+    {
+      "epoch": 30.89,
+      "grad_norm": 0.7058090567588806,
+      "learning_rate": 0.0007644444444444445,
+      "loss": 1.4025,
+      "step": 8340
+    },
+    {
+      "epoch": 30.93,
+      "grad_norm": 0.7776255011558533,
+      "learning_rate": 0.000762962962962963,
+      "loss": 1.3931,
+      "step": 8350
+    },
+    {
+      "epoch": 30.96,
+      "grad_norm": 0.7906094193458557,
+      "learning_rate": 0.0007614814814814816,
+      "loss": 1.4201,
+      "step": 8360
+    },
+    {
+      "epoch": 31.0,
+      "grad_norm": 1.9872336387634277,
+      "learning_rate": 0.00076,
+      "loss": 1.4128,
+      "step": 8370
+    },
+    {
+      "epoch": 31.04,
+      "grad_norm": 0.7633602619171143,
+      "learning_rate": 0.0007585185185185185,
+      "loss": 1.1678,
+      "step": 8380
+    },
+    {
+      "epoch": 31.07,
+      "grad_norm": 0.7451400756835938,
+      "learning_rate": 0.0007570370370370371,
+      "loss": 1.1833,
+      "step": 8390
+    },
+    {
+      "epoch": 31.11,
+      "grad_norm": 0.7801716327667236,
+      "learning_rate": 0.0007555555555555555,
+      "loss": 1.1863,
+      "step": 8400
+    },
+    {
+      "epoch": 31.15,
+      "grad_norm": 0.8019985556602478,
+      "learning_rate": 0.0007540740740740741,
+      "loss": 1.2048,
+      "step": 8410
+    },
+    {
+      "epoch": 31.19,
+      "grad_norm": 0.7380914092063904,
+      "learning_rate": 0.0007525925925925926,
+      "loss": 1.2158,
+      "step": 8420
+    },
+    {
+      "epoch": 31.22,
+      "grad_norm": 0.8214986324310303,
+      "learning_rate": 0.000751111111111111,
+      "loss": 1.2368,
+      "step": 8430
+    },
+    {
+      "epoch": 31.26,
+      "grad_norm": 0.8278571963310242,
+      "learning_rate": 0.0007496296296296297,
+      "loss": 1.2449,
+      "step": 8440
+    },
+    {
+      "epoch": 31.3,
+      "grad_norm": 0.8095946907997131,
+      "learning_rate": 0.0007481481481481482,
+      "loss": 1.2509,
+      "step": 8450
+    },
+    {
+      "epoch": 31.33,
+      "grad_norm": 0.701068639755249,
+      "learning_rate": 0.0007466666666666667,
+      "loss": 1.2798,
+      "step": 8460
+    },
+    {
+      "epoch": 31.37,
+      "grad_norm": 0.7845342755317688,
+      "learning_rate": 0.0007451851851851852,
+      "loss": 1.2811,
+      "step": 8470
+    },
+    {
+      "epoch": 31.41,
+      "grad_norm": 0.7937617301940918,
+      "learning_rate": 0.0007437037037037038,
+      "loss": 1.2757,
+      "step": 8480
+    },
+    {
+      "epoch": 31.44,
+      "grad_norm": 0.8310158848762512,
+      "learning_rate": 0.0007422222222222222,
+      "loss": 1.2953,
+      "step": 8490
+    },
+    {
+      "epoch": 31.48,
+      "grad_norm": 0.8204872012138367,
+      "learning_rate": 0.0007407407407407407,
+      "loss": 1.2792,
+      "step": 8500
+    },
+    {
+      "epoch": 31.52,
+      "grad_norm": 0.746772050857544,
+      "learning_rate": 0.0007392592592592593,
+      "loss": 1.2849,
+      "step": 8510
+    },
+    {
+      "epoch": 31.56,
+      "grad_norm": 0.7524877190589905,
+      "learning_rate": 0.0007377777777777777,
+      "loss": 1.3163,
+      "step": 8520
+    },
+    {
+      "epoch": 31.59,
+      "grad_norm": 0.8254786729812622,
+      "learning_rate": 0.0007362962962962963,
+      "loss": 1.3039,
+      "step": 8530
+    },
+    {
+      "epoch": 31.63,
+      "grad_norm": 0.7942031025886536,
+      "learning_rate": 0.0007348148148148149,
+      "loss": 1.3309,
+      "step": 8540
+    },
+    {
+      "epoch": 31.67,
+      "grad_norm": 0.746993899345398,
+      "learning_rate": 0.0007333333333333333,
+      "loss": 1.3345,
+      "step": 8550
+    },
+    {
+      "epoch": 31.7,
+      "grad_norm": 0.7152719497680664,
+      "learning_rate": 0.0007318518518518519,
+      "loss": 1.3433,
+      "step": 8560
+    },
+    {
+      "epoch": 31.74,
+      "grad_norm": 0.8452175855636597,
+      "learning_rate": 0.0007303703703703704,
+      "loss": 1.3443,
+      "step": 8570
+    },
+    {
+      "epoch": 31.78,
+      "grad_norm": 0.7857072949409485,
+      "learning_rate": 0.0007288888888888889,
+      "loss": 1.3459,
+      "step": 8580
+    },
+    {
+      "epoch": 31.81,
+      "grad_norm": 0.8009936809539795,
+      "learning_rate": 0.0007274074074074074,
+      "loss": 1.3455,
+      "step": 8590
+    },
+    {
+      "epoch": 31.85,
+      "grad_norm": 0.9043723940849304,
+      "learning_rate": 0.000725925925925926,
+      "loss": 1.3832,
+      "step": 8600
+    },
+    {
+      "epoch": 31.89,
+      "grad_norm": 0.7199929356575012,
+      "learning_rate": 0.0007244444444444444,
+      "loss": 1.362,
+      "step": 8610
+    },
+    {
+      "epoch": 31.93,
+      "grad_norm": 0.7644399404525757,
+      "learning_rate": 0.0007229629629629629,
+      "loss": 1.3514,
+      "step": 8620
+    },
+    {
+      "epoch": 31.96,
+      "grad_norm": 0.7020518183708191,
+      "learning_rate": 0.0007214814814814815,
+      "loss": 1.3749,
+      "step": 8630
+    },
+    {
+      "epoch": 32.0,
+      "grad_norm": 1.4921846389770508,
+      "learning_rate": 0.0007199999999999999,
+      "loss": 1.405,
+      "step": 8640
+    },
+    {
+      "epoch": 32.04,
+      "grad_norm": 0.7334415912628174,
+      "learning_rate": 0.0007185185185185186,
+      "loss": 1.1488,
+      "step": 8650
+    },
+    {
+      "epoch": 32.07,
+      "grad_norm": 0.7583827376365662,
+      "learning_rate": 0.0007170370370370371,
+      "loss": 1.175,
+      "step": 8660
+    },
+    {
+      "epoch": 32.11,
+      "grad_norm": 0.755535900592804,
+      "learning_rate": 0.0007155555555555555,
+      "loss": 1.1634,
+      "step": 8670
+    },
+    {
+      "epoch": 32.15,
+      "grad_norm": 0.7786100506782532,
+      "learning_rate": 0.0007140740740740741,
+      "loss": 1.1713,
+      "step": 8680
+    },
+    {
+      "epoch": 32.19,
+      "grad_norm": 0.7711645364761353,
+      "learning_rate": 0.0007125925925925926,
+      "loss": 1.1811,
+      "step": 8690
+    },
+    {
+      "epoch": 32.22,
+      "grad_norm": 0.8045185804367065,
+      "learning_rate": 0.0007111111111111111,
+      "loss": 1.2022,
+      "step": 8700
+    },
+    {
+      "epoch": 32.26,
+      "grad_norm": 0.8078054189682007,
+      "learning_rate": 0.0007096296296296296,
+      "loss": 1.2254,
+      "step": 8710
+    },
+    {
+      "epoch": 32.3,
+      "grad_norm": 0.7914540767669678,
+      "learning_rate": 0.0007081481481481482,
+      "loss": 1.2144,
+      "step": 8720
+    },
+    {
+      "epoch": 32.33,
+      "grad_norm": 0.7771093249320984,
+      "learning_rate": 0.0007066666666666666,
+      "loss": 1.2267,
+      "step": 8730
+    },
+    {
+      "epoch": 32.37,
+      "grad_norm": 0.8438934087753296,
+      "learning_rate": 0.0007051851851851851,
+      "loss": 1.2378,
+      "step": 8740
+    },
+    {
+      "epoch": 32.41,
+      "grad_norm": 0.8273505568504333,
+      "learning_rate": 0.0007037037037037038,
+      "loss": 1.2385,
+      "step": 8750
+    },
+    {
+      "epoch": 32.44,
+      "grad_norm": 0.7227649688720703,
+      "learning_rate": 0.0007022222222222222,
+      "loss": 1.2618,
+      "step": 8760
+    },
+    {
+      "epoch": 32.48,
+      "grad_norm": 0.7891519069671631,
+      "learning_rate": 0.0007007407407407408,
+      "loss": 1.2637,
+      "step": 8770
+    },
+    {
+      "epoch": 32.52,
+      "grad_norm": 0.8562486171722412,
+      "learning_rate": 0.0006992592592592593,
+      "loss": 1.292,
+      "step": 8780
+    },
+    {
+      "epoch": 32.56,
+      "grad_norm": 0.722166895866394,
+      "learning_rate": 0.0006977777777777778,
+      "loss": 1.2806,
+      "step": 8790
+    },
+    {
+      "epoch": 32.59,
+      "grad_norm": 0.8187007904052734,
+      "learning_rate": 0.0006962962962962963,
+      "loss": 1.2906,
+      "step": 8800
+    },
+    {
+      "epoch": 32.63,
+      "grad_norm": 0.8709007501602173,
+      "learning_rate": 0.0006948148148148148,
+      "loss": 1.2941,
+      "step": 8810
+    },
+    {
+      "epoch": 32.67,
+      "grad_norm": 0.8180269002914429,
+      "learning_rate": 0.0006933333333333333,
+      "loss": 1.2994,
+      "step": 8820
+    },
+    {
+      "epoch": 32.7,
+      "grad_norm": 0.8007955551147461,
+      "learning_rate": 0.0006918518518518518,
+      "loss": 1.2668,
+      "step": 8830
+    },
+    {
+      "epoch": 32.74,
+      "grad_norm": 0.8287192583084106,
+      "learning_rate": 0.0006903703703703704,
+      "loss": 1.32,
+      "step": 8840
+    },
+    {
+      "epoch": 32.78,
+      "grad_norm": 0.8008798956871033,
+      "learning_rate": 0.000688888888888889,
+      "loss": 1.3194,
+      "step": 8850
+    },
+    {
+      "epoch": 32.81,
+      "grad_norm": 0.7793354988098145,
+      "learning_rate": 0.0006874074074074074,
+      "loss": 1.3138,
+      "step": 8860
+    },
+    {
+      "epoch": 32.85,
+      "grad_norm": 0.8135443925857544,
+      "learning_rate": 0.000685925925925926,
+      "loss": 1.3349,
+      "step": 8870
+    },
+    {
+      "epoch": 32.89,
+      "grad_norm": 0.8550287485122681,
+      "learning_rate": 0.0006844444444444444,
+      "loss": 1.358,
+      "step": 8880
+    },
+    {
+      "epoch": 32.93,
+      "grad_norm": 0.8146428465843201,
+      "learning_rate": 0.000682962962962963,
+      "loss": 1.347,
+      "step": 8890
+    },
+    {
+      "epoch": 32.96,
+      "grad_norm": 0.8575021028518677,
+      "learning_rate": 0.0006814814814814815,
+      "loss": 1.3261,
+      "step": 8900
+    },
+    {
+      "epoch": 33.0,
+      "grad_norm": 1.6308609247207642,
+      "learning_rate": 0.00068,
+      "loss": 1.3611,
+      "step": 8910
+    },
+    {
+      "epoch": 33.04,
+      "grad_norm": 0.7819394469261169,
+      "learning_rate": 0.0006785185185185185,
+      "loss": 1.1404,
+      "step": 8920
+    },
+    {
+      "epoch": 33.07,
+      "grad_norm": 0.7906447052955627,
+      "learning_rate": 0.000677037037037037,
+      "loss": 1.1216,
+      "step": 8930
+    },
+    {
+      "epoch": 33.11,
+      "grad_norm": 0.7358431816101074,
+      "learning_rate": 0.0006755555555555555,
+      "loss": 1.1406,
+      "step": 8940
+    },
+    {
+      "epoch": 33.15,
+      "grad_norm": 0.778339147567749,
+      "learning_rate": 0.0006740740740740741,
+      "loss": 1.1622,
+      "step": 8950
+    },
+    {
+      "epoch": 33.19,
+      "grad_norm": 0.7508297562599182,
+      "learning_rate": 0.0006725925925925927,
+      "loss": 1.1778,
+      "step": 8960
+    },
+    {
+      "epoch": 33.22,
+      "grad_norm": 0.830898106098175,
+      "learning_rate": 0.0006711111111111111,
+      "loss": 1.1648,
+      "step": 8970
+    },
+    {
+      "epoch": 33.26,
+      "grad_norm": 0.8099901080131531,
+      "learning_rate": 0.0006696296296296296,
+      "loss": 1.1866,
+      "step": 8980
+    },
+    {
+      "epoch": 33.3,
+      "grad_norm": 0.8449798822402954,
+      "learning_rate": 0.0006681481481481482,
+      "loss": 1.1964,
+      "step": 8990
+    },
+    {
+      "epoch": 33.33,
+      "grad_norm": 0.7969886660575867,
+      "learning_rate": 0.0006666666666666666,
+      "loss": 1.1974,
+      "step": 9000
+    },
+    {
+      "epoch": 33.37,
+      "grad_norm": 0.809606671333313,
+      "learning_rate": 0.0006651851851851852,
+      "loss": 1.1999,
+      "step": 9010
+    },
+    {
+      "epoch": 33.41,
+      "grad_norm": 0.8491615653038025,
+      "learning_rate": 0.0006637037037037037,
+      "loss": 1.2246,
+      "step": 9020
+    },
+    {
+      "epoch": 33.44,
+      "grad_norm": 0.8338679671287537,
+      "learning_rate": 0.0006622222222222222,
+      "loss": 1.2387,
+      "step": 9030
+    },
+    {
+      "epoch": 33.48,
+      "grad_norm": 0.7784556746482849,
+      "learning_rate": 0.0006607407407407407,
+      "loss": 1.2544,
+      "step": 9040
+    },
+    {
+      "epoch": 33.52,
+      "grad_norm": 0.8017482757568359,
+      "learning_rate": 0.0006592592592592592,
+      "loss": 1.2187,
+      "step": 9050
+    },
+    {
+      "epoch": 33.56,
+      "grad_norm": 0.8610512018203735,
+      "learning_rate": 0.0006577777777777779,
+      "loss": 1.239,
+      "step": 9060
+    },
+    {
+      "epoch": 33.59,
+      "grad_norm": 0.7477060556411743,
+      "learning_rate": 0.0006562962962962963,
+      "loss": 1.2626,
+      "step": 9070
+    },
+    {
+      "epoch": 33.63,
+      "grad_norm": 0.8773489594459534,
+      "learning_rate": 0.0006548148148148149,
+      "loss": 1.2678,
+      "step": 9080
+    },
+    {
+      "epoch": 33.67,
+      "grad_norm": 0.8236984610557556,
+      "learning_rate": 0.0006533333333333333,
+      "loss": 1.275,
+      "step": 9090
+    },
+    {
+      "epoch": 33.7,
+      "grad_norm": 0.799561083316803,
+      "learning_rate": 0.0006518518518518519,
+      "loss": 1.273,
+      "step": 9100
+    },
+    {
+      "epoch": 33.74,
+      "grad_norm": 0.8354243040084839,
+      "learning_rate": 0.0006503703703703704,
+      "loss": 1.2556,
+      "step": 9110
+    },
+    {
+      "epoch": 33.78,
+      "grad_norm": 0.7642504572868347,
+      "learning_rate": 0.0006488888888888888,
+      "loss": 1.2839,
+      "step": 9120
+    },
+    {
+      "epoch": 33.81,
+      "grad_norm": 0.8058938384056091,
+      "learning_rate": 0.0006474074074074074,
+      "loss": 1.2844,
+      "step": 9130
+    },
+    {
+      "epoch": 33.85,
+      "grad_norm": 0.8240184187889099,
+      "learning_rate": 0.0006459259259259259,
+      "loss": 1.2444,
+      "step": 9140
+    },
+    {
+      "epoch": 33.89,
+      "grad_norm": 0.7965198755264282,
+      "learning_rate": 0.0006444444444444444,
+      "loss": 1.302,
+      "step": 9150
+    },
+    {
+      "epoch": 33.93,
+      "grad_norm": 0.807142436504364,
+      "learning_rate": 0.000642962962962963,
+      "loss": 1.2898,
+      "step": 9160
+    },
+    {
+      "epoch": 33.96,
+      "grad_norm": 0.8767658472061157,
+      "learning_rate": 0.0006414814814814815,
+      "loss": 1.3294,
+      "step": 9170
+    },
+    {
+      "epoch": 34.0,
+      "grad_norm": 1.232200264930725,
+      "learning_rate": 0.00064,
+      "loss": 1.3133,
+      "step": 9180
+    },
+    {
+      "epoch": 34.04,
+      "grad_norm": 0.7402983903884888,
+      "learning_rate": 0.0006385185185185185,
+      "loss": 1.0779,
+      "step": 9190
+    },
+    {
+      "epoch": 34.07,
+      "grad_norm": 0.8266010880470276,
+      "learning_rate": 0.0006370370370370371,
+      "loss": 1.1209,
+      "step": 9200
+    },
+    {
+      "epoch": 34.11,
+      "grad_norm": 0.7701414823532104,
+      "learning_rate": 0.0006355555555555555,
+      "loss": 1.1255,
+      "step": 9210
+    },
+    {
+      "epoch": 34.15,
+      "grad_norm": 0.7860375642776489,
+      "learning_rate": 0.0006340740740740741,
+      "loss": 1.1165,
+      "step": 9220
+    },
+    {
+      "epoch": 34.19,
+      "grad_norm": 0.8956103920936584,
+      "learning_rate": 0.0006325925925925926,
+      "loss": 1.0995,
+      "step": 9230
+    },
+    {
+      "epoch": 34.22,
+      "grad_norm": 0.7913826107978821,
+      "learning_rate": 0.000631111111111111,
+      "loss": 1.1349,
+      "step": 9240
+    },
+    {
+      "epoch": 34.26,
+      "grad_norm": 0.8237706422805786,
+      "learning_rate": 0.0006296296296296296,
+      "loss": 1.1396,
+      "step": 9250
+    },
+    {
+      "epoch": 34.3,
+      "grad_norm": 0.7774602770805359,
+      "learning_rate": 0.0006281481481481482,
+      "loss": 1.162,
+      "step": 9260
+    },
+    {
+      "epoch": 34.33,
+      "grad_norm": 0.8474662899971008,
+      "learning_rate": 0.0006266666666666668,
+      "loss": 1.1747,
+      "step": 9270
+    },
+    {
+      "epoch": 34.37,
+      "grad_norm": 0.9087252616882324,
+      "learning_rate": 0.0006251851851851852,
+      "loss": 1.1785,
+      "step": 9280
+    },
+    {
+      "epoch": 34.41,
+      "grad_norm": 0.8514711260795593,
+      "learning_rate": 0.0006237037037037037,
+      "loss": 1.2354,
+      "step": 9290
+    },
+    {
+      "epoch": 34.44,
+      "grad_norm": 0.8436718583106995,
+      "learning_rate": 0.0006222222222222223,
+      "loss": 1.1793,
+      "step": 9300
+    },
+    {
+      "epoch": 34.48,
+      "grad_norm": 0.9170453548431396,
+      "learning_rate": 0.0006207407407407407,
+      "loss": 1.1807,
+      "step": 9310
+    },
+    {
+      "epoch": 34.52,
+      "grad_norm": 0.9778527021408081,
+      "learning_rate": 0.0006192592592592593,
+      "loss": 1.2231,
+      "step": 9320
+    },
+    {
+      "epoch": 34.56,
+      "grad_norm": 0.8001771569252014,
+      "learning_rate": 0.0006177777777777777,
+      "loss": 1.2107,
+      "step": 9330
+    },
+    {
+      "epoch": 34.59,
+      "grad_norm": 0.7859574556350708,
+      "learning_rate": 0.0006162962962962963,
+      "loss": 1.2242,
+      "step": 9340
+    },
+    {
+      "epoch": 34.63,
+      "grad_norm": 0.8739088773727417,
+      "learning_rate": 0.0006148148148148148,
+      "loss": 1.2294,
+      "step": 9350
+    },
+    {
+      "epoch": 34.67,
+      "grad_norm": 0.8462000489234924,
+      "learning_rate": 0.0006133333333333334,
+      "loss": 1.2335,
+      "step": 9360
+    },
+    {
+      "epoch": 34.7,
+      "grad_norm": 0.9332739114761353,
+      "learning_rate": 0.0006118518518518519,
+      "loss": 1.2425,
+      "step": 9370
+    },
+    {
+      "epoch": 34.74,
+      "grad_norm": 0.8226819038391113,
+      "learning_rate": 0.0006103703703703704,
+      "loss": 1.255,
+      "step": 9380
+    },
+    {
+      "epoch": 34.78,
+      "grad_norm": 0.8331789374351501,
+      "learning_rate": 0.000608888888888889,
+      "loss": 1.2433,
+      "step": 9390
+    },
+    {
+      "epoch": 34.81,
+      "grad_norm": 0.8723573684692383,
+      "learning_rate": 0.0006074074074074074,
+      "loss": 1.2628,
+      "step": 9400
+    },
+    {
+      "epoch": 34.85,
+      "grad_norm": 0.8315209746360779,
+      "learning_rate": 0.000605925925925926,
+      "loss": 1.2484,
+      "step": 9410
+    },
+    {
+      "epoch": 34.89,
+      "grad_norm": 0.8821229338645935,
+      "learning_rate": 0.0006044444444444445,
+      "loss": 1.2668,
+      "step": 9420
+    },
+    {
+      "epoch": 34.93,
+      "grad_norm": 0.8588849306106567,
+      "learning_rate": 0.0006029629629629629,
+      "loss": 1.2643,
+      "step": 9430
+    },
+    {
+      "epoch": 34.96,
+      "grad_norm": 0.8674469590187073,
+      "learning_rate": 0.0006014814814814815,
+      "loss": 1.2944,
+      "step": 9440
+    },
+    {
+      "epoch": 35.0,
+      "grad_norm": 1.8903077840805054,
+      "learning_rate": 0.0006,
+      "loss": 1.2972,
+      "step": 9450
+    },
+    {
+      "epoch": 35.04,
+      "grad_norm": 0.7540388107299805,
+      "learning_rate": 0.0005985185185185186,
+      "loss": 1.0662,
+      "step": 9460
+    },
+    {
+      "epoch": 35.07,
+      "grad_norm": 0.8312435150146484,
+      "learning_rate": 0.0005970370370370371,
+      "loss": 1.0752,
+      "step": 9470
+    },
+    {
+      "epoch": 35.11,
+      "grad_norm": 0.8651759624481201,
+      "learning_rate": 0.0005955555555555556,
+      "loss": 1.0925,
+      "step": 9480
+    },
+    {
+      "epoch": 35.15,
+      "grad_norm": 0.7402911186218262,
+      "learning_rate": 0.0005940740740740741,
+      "loss": 1.0936,
+      "step": 9490
+    },
+    {
+      "epoch": 35.19,
+      "grad_norm": 0.7958086133003235,
+      "learning_rate": 0.0005925925925925926,
+      "loss": 1.0865,
+      "step": 9500
+    },
+    {
+      "epoch": 35.22,
+      "grad_norm": 0.773444652557373,
+      "learning_rate": 0.0005911111111111112,
+      "loss": 1.1231,
+      "step": 9510
+    },
+    {
+      "epoch": 35.26,
+      "grad_norm": 0.8964213132858276,
+      "learning_rate": 0.0005896296296296296,
+      "loss": 1.1239,
+      "step": 9520
+    },
+    {
+      "epoch": 35.3,
+      "grad_norm": 0.8228880167007446,
+      "learning_rate": 0.0005881481481481482,
+      "loss": 1.1283,
+      "step": 9530
+    },
+    {
+      "epoch": 35.33,
+      "grad_norm": 0.8290715217590332,
+      "learning_rate": 0.0005866666666666667,
+      "loss": 1.1549,
+      "step": 9540
+    },
+    {
+      "epoch": 35.37,
+      "grad_norm": 0.7984841465950012,
+      "learning_rate": 0.0005851851851851851,
+      "loss": 1.1152,
+      "step": 9550
+    },
+    {
+      "epoch": 35.41,
+      "grad_norm": 0.9492020606994629,
+      "learning_rate": 0.0005837037037037037,
+      "loss": 1.1375,
+      "step": 9560
+    },
+    {
+      "epoch": 35.44,
+      "grad_norm": 0.9132887721061707,
+      "learning_rate": 0.0005822222222222223,
+      "loss": 1.1848,
+      "step": 9570
+    },
+    {
+      "epoch": 35.48,
+      "grad_norm": 0.9160571098327637,
+      "learning_rate": 0.0005807407407407408,
+      "loss": 1.1619,
+      "step": 9580
+    },
+    {
+      "epoch": 35.52,
+      "grad_norm": 0.8078466057777405,
+      "learning_rate": 0.0005792592592592593,
+      "loss": 1.217,
+      "step": 9590
+    },
+    {
+      "epoch": 35.56,
+      "grad_norm": 0.8517169952392578,
+      "learning_rate": 0.0005777777777777778,
+      "loss": 1.1829,
+      "step": 9600
+    },
+    {
+      "epoch": 35.59,
+      "grad_norm": 0.8791046142578125,
+      "learning_rate": 0.0005762962962962963,
+      "loss": 1.1896,
+      "step": 9610
+    },
+    {
+      "epoch": 35.63,
+      "grad_norm": 0.9342435002326965,
+      "learning_rate": 0.0005748148148148148,
+      "loss": 1.2028,
+      "step": 9620
+    },
+    {
+      "epoch": 35.67,
+      "grad_norm": 0.884878933429718,
+      "learning_rate": 0.0005733333333333334,
+      "loss": 1.202,
+      "step": 9630
+    },
+    {
+      "epoch": 35.7,
+      "grad_norm": 0.8292385935783386,
+      "learning_rate": 0.0005718518518518518,
+      "loss": 1.2376,
+      "step": 9640
+    },
+    {
+      "epoch": 35.74,
+      "grad_norm": 0.8106871843338013,
+      "learning_rate": 0.0005703703703703704,
+      "loss": 1.1888,
+      "step": 9650
+    },
+    {
+      "epoch": 35.78,
+      "grad_norm": 0.8425321578979492,
+      "learning_rate": 0.0005688888888888889,
+      "loss": 1.2282,
+      "step": 9660
+    },
+    {
+      "epoch": 35.81,
+      "grad_norm": 0.8284263610839844,
+      "learning_rate": 0.0005674074074074074,
+      "loss": 1.2185,
+      "step": 9670
+    },
+    {
+      "epoch": 35.85,
+      "grad_norm": 0.8811337947845459,
+      "learning_rate": 0.000565925925925926,
+      "loss": 1.2189,
+      "step": 9680
+    },
+    {
+      "epoch": 35.89,
+      "grad_norm": 0.8189839124679565,
+      "learning_rate": 0.0005644444444444445,
+      "loss": 1.2378,
+      "step": 9690
+    },
+    {
+      "epoch": 35.93,
+      "grad_norm": 0.9154495000839233,
+      "learning_rate": 0.000562962962962963,
+      "loss": 1.2155,
+      "step": 9700
+    },
+    {
+      "epoch": 35.96,
+      "grad_norm": 0.84986412525177,
+      "learning_rate": 0.0005614814814814815,
+      "loss": 1.2282,
+      "step": 9710
+    },
+    {
+      "epoch": 36.0,
+      "grad_norm": 1.4053990840911865,
+      "learning_rate": 0.0005600000000000001,
+      "loss": 1.2341,
+      "step": 9720
+    },
+    {
+      "epoch": 36.04,
+      "grad_norm": 0.8471843004226685,
+      "learning_rate": 0.0005585185185185185,
+      "loss": 1.0177,
+      "step": 9730
+    },
+    {
+      "epoch": 36.07,
+      "grad_norm": 0.8182628154754639,
+      "learning_rate": 0.000557037037037037,
+      "loss": 1.0433,
+      "step": 9740
+    },
+    {
+      "epoch": 36.11,
+      "grad_norm": 0.8502013087272644,
+      "learning_rate": 0.0005555555555555556,
+      "loss": 1.102,
+      "step": 9750
+    },
+    {
+      "epoch": 36.15,
+      "grad_norm": 0.8067908883094788,
+      "learning_rate": 0.000554074074074074,
+      "loss": 1.0811,
+      "step": 9760
+    },
+    {
+      "epoch": 36.19,
+      "grad_norm": 0.8608354926109314,
+      "learning_rate": 0.0005525925925925927,
+      "loss": 1.0716,
+      "step": 9770
+    },
+    {
+      "epoch": 36.22,
+      "grad_norm": 0.8624645471572876,
+      "learning_rate": 0.0005511111111111112,
+      "loss": 1.0711,
+      "step": 9780
+    },
+    {
+      "epoch": 36.26,
+      "grad_norm": 0.8814002871513367,
+      "learning_rate": 0.0005496296296296296,
+      "loss": 1.0988,
+      "step": 9790
+    },
+    {
+      "epoch": 36.3,
+      "grad_norm": 0.796582043170929,
+      "learning_rate": 0.0005481481481481482,
+      "loss": 1.0714,
+      "step": 9800
+    },
+    {
+      "epoch": 36.33,
+      "grad_norm": 0.8638683557510376,
+      "learning_rate": 0.0005466666666666667,
+      "loss": 1.1021,
+      "step": 9810
+    },
+    {
+      "epoch": 36.37,
+      "grad_norm": 1.0243921279907227,
+      "learning_rate": 0.0005451851851851852,
+      "loss": 1.1143,
+      "step": 9820
+    },
+    {
+      "epoch": 36.41,
+      "grad_norm": 0.9122822284698486,
+      "learning_rate": 0.0005437037037037037,
+      "loss": 1.1485,
+      "step": 9830
+    },
+    {
+      "epoch": 36.44,
+      "grad_norm": 0.7532286047935486,
+      "learning_rate": 0.0005422222222222223,
+      "loss": 1.1323,
+      "step": 9840
+    },
+    {
+      "epoch": 36.48,
+      "grad_norm": 0.8187335729598999,
+      "learning_rate": 0.0005407407407407407,
+      "loss": 1.133,
+      "step": 9850
+    },
+    {
+      "epoch": 36.52,
+      "grad_norm": 0.8526188135147095,
+      "learning_rate": 0.0005392592592592592,
+      "loss": 1.1739,
+      "step": 9860
+    },
+    {
+      "epoch": 36.56,
+      "grad_norm": 0.8166966438293457,
+      "learning_rate": 0.0005377777777777779,
+      "loss": 1.1633,
+      "step": 9870
+    },
+    {
+      "epoch": 36.59,
+      "grad_norm": 0.8333300352096558,
+      "learning_rate": 0.0005362962962962963,
+      "loss": 1.1226,
+      "step": 9880
+    },
+    {
+      "epoch": 36.63,
+      "grad_norm": 0.8543572425842285,
+      "learning_rate": 0.0005348148148148149,
+      "loss": 1.1745,
+      "step": 9890
+    },
+    {
+      "epoch": 36.67,
+      "grad_norm": 0.8492584824562073,
+      "learning_rate": 0.0005333333333333334,
+      "loss": 1.1747,
+      "step": 9900
+    },
+    {
+      "epoch": 36.7,
+      "grad_norm": 0.8231162428855896,
+      "learning_rate": 0.0005318518518518518,
+      "loss": 1.1825,
+      "step": 9910
+    },
+    {
+      "epoch": 36.74,
+      "grad_norm": 0.8251094222068787,
+      "learning_rate": 0.0005303703703703704,
+      "loss": 1.1943,
+      "step": 9920
+    },
+    {
+      "epoch": 36.78,
+      "grad_norm": 0.8481724262237549,
+      "learning_rate": 0.0005288888888888889,
+      "loss": 1.1725,
+      "step": 9930
+    },
+    {
+      "epoch": 36.81,
+      "grad_norm": 0.9228334426879883,
+      "learning_rate": 0.0005274074074074074,
+      "loss": 1.1736,
+      "step": 9940
+    },
+    {
+      "epoch": 36.85,
+      "grad_norm": 0.8884844779968262,
+      "learning_rate": 0.0005259259259259259,
+      "loss": 1.1856,
+      "step": 9950
+    },
+    {
+      "epoch": 36.89,
+      "grad_norm": 0.8618705868721008,
+      "learning_rate": 0.0005244444444444445,
+      "loss": 1.2202,
+      "step": 9960
+    },
+    {
+      "epoch": 36.93,
+      "grad_norm": 1.0046896934509277,
+      "learning_rate": 0.0005229629629629629,
+      "loss": 1.2043,
+      "step": 9970
+    },
+    {
+      "epoch": 36.96,
+      "grad_norm": 0.8208348155021667,
+      "learning_rate": 0.0005214814814814815,
+      "loss": 1.1784,
+      "step": 9980
+    },
+    {
+      "epoch": 37.0,
+      "grad_norm": 1.7133288383483887,
+      "learning_rate": 0.0005200000000000001,
+      "loss": 1.1988,
+      "step": 9990
+    },
+    {
+      "epoch": 37.04,
+      "grad_norm": 0.8276928663253784,
+      "learning_rate": 0.0005185185185185185,
+      "loss": 0.9958,
+      "step": 10000
+    },
+    {
+      "epoch": 37.07,
+      "grad_norm": 0.8038898706436157,
+      "learning_rate": 0.0005170370370370371,
+      "loss": 1.0148,
+      "step": 10010
+    },
+    {
+      "epoch": 37.11,
+      "grad_norm": 0.8378689885139465,
+      "learning_rate": 0.0005155555555555556,
+      "loss": 1.0242,
+      "step": 10020
+    },
+    {
+      "epoch": 37.15,
+      "grad_norm": 0.826815664768219,
+      "learning_rate": 0.0005140740740740741,
+      "loss": 1.0133,
+      "step": 10030
+    },
+    {
+      "epoch": 37.19,
+      "grad_norm": 0.8740249872207642,
+      "learning_rate": 0.0005125925925925926,
+      "loss": 1.0245,
+      "step": 10040
+    },
+    {
+      "epoch": 37.22,
+      "grad_norm": 0.7812831401824951,
+      "learning_rate": 0.0005111111111111111,
+      "loss": 1.0451,
+      "step": 10050
+    },
+    {
+      "epoch": 37.26,
+      "grad_norm": 0.8525770306587219,
+      "learning_rate": 0.0005096296296296296,
+      "loss": 1.0525,
+      "step": 10060
+    },
+    {
+      "epoch": 37.3,
+      "grad_norm": 0.8188793063163757,
+      "learning_rate": 0.0005081481481481481,
+      "loss": 1.079,
+      "step": 10070
+    },
+    {
+      "epoch": 37.33,
+      "grad_norm": 0.8820340633392334,
+      "learning_rate": 0.0005066666666666668,
+      "loss": 1.0808,
+      "step": 10080
+    },
+    {
+      "epoch": 37.37,
+      "grad_norm": 0.9423266649246216,
+      "learning_rate": 0.0005051851851851852,
+      "loss": 1.0786,
+      "step": 10090
+    },
+    {
+      "epoch": 37.41,
+      "grad_norm": 0.8830150365829468,
+      "learning_rate": 0.0005037037037037037,
+      "loss": 1.107,
+      "step": 10100
+    },
+    {
+      "epoch": 37.44,
+      "grad_norm": 0.9641768932342529,
+      "learning_rate": 0.0005022222222222223,
+      "loss": 1.0972,
+      "step": 10110
+    },
+    {
+      "epoch": 37.48,
+      "grad_norm": 0.8334240317344666,
+      "learning_rate": 0.0005007407407407407,
+      "loss": 1.1229,
+      "step": 10120
+    },
+    {
+      "epoch": 37.52,
+      "grad_norm": 0.8352447748184204,
+      "learning_rate": 0.0004992592592592593,
+      "loss": 1.1144,
+      "step": 10130
+    },
+    {
+      "epoch": 37.56,
+      "grad_norm": 0.834314227104187,
+      "learning_rate": 0.0004977777777777778,
+      "loss": 1.1161,
+      "step": 10140
+    },
+    {
+      "epoch": 37.59,
+      "grad_norm": 0.8779057860374451,
+      "learning_rate": 0.0004962962962962963,
+      "loss": 1.133,
+      "step": 10150
+    },
+    {
+      "epoch": 37.63,
+      "grad_norm": 0.8864279389381409,
+      "learning_rate": 0.0004948148148148148,
+      "loss": 1.1581,
+      "step": 10160
+    },
+    {
+      "epoch": 37.67,
+      "grad_norm": 0.8707226514816284,
+      "learning_rate": 0.0004933333333333334,
+      "loss": 1.135,
+      "step": 10170
+    },
+    {
+      "epoch": 37.7,
+      "grad_norm": 0.8655386567115784,
+      "learning_rate": 0.0004918518518518519,
+      "loss": 1.1366,
+      "step": 10180
+    },
+    {
+      "epoch": 37.74,
+      "grad_norm": 0.9207199811935425,
+      "learning_rate": 0.0004903703703703704,
+      "loss": 1.1788,
+      "step": 10190
+    },
+    {
+      "epoch": 37.78,
+      "grad_norm": 0.8524510264396667,
+      "learning_rate": 0.0004888888888888889,
+      "loss": 1.1285,
+      "step": 10200
+    },
+    {
+      "epoch": 37.81,
+      "grad_norm": 0.8669392466545105,
+      "learning_rate": 0.00048740740740740743,
+      "loss": 1.1691,
+      "step": 10210
+    },
+    {
+      "epoch": 37.85,
+      "grad_norm": 0.8475798964500427,
+      "learning_rate": 0.00048592592592592595,
+      "loss": 1.1825,
+      "step": 10220
+    },
+    {
+      "epoch": 37.89,
+      "grad_norm": 0.8444201350212097,
+      "learning_rate": 0.00048444444444444446,
+      "loss": 1.1717,
+      "step": 10230
+    },
+    {
+      "epoch": 37.93,
+      "grad_norm": 0.9086614847183228,
+      "learning_rate": 0.000482962962962963,
+      "loss": 1.1652,
+      "step": 10240
+    },
+    {
+      "epoch": 37.96,
+      "grad_norm": 0.9066885709762573,
+      "learning_rate": 0.00048148148148148144,
+      "loss": 1.1702,
+      "step": 10250
+    },
+    {
+      "epoch": 38.0,
+      "grad_norm": 1.5993976593017578,
+      "learning_rate": 0.00048,
+      "loss": 1.1431,
+      "step": 10260
+    },
+    {
+      "epoch": 38.04,
+      "grad_norm": 0.8458494544029236,
+      "learning_rate": 0.00047851851851851853,
+      "loss": 0.9829,
+      "step": 10270
+    },
+    {
+      "epoch": 38.07,
+      "grad_norm": 0.8219654560089111,
+      "learning_rate": 0.00047703703703703705,
+      "loss": 1.0083,
+      "step": 10280
+    },
+    {
+      "epoch": 38.11,
+      "grad_norm": 0.9130375385284424,
+      "learning_rate": 0.00047555555555555556,
+      "loss": 0.9749,
+      "step": 10290
+    },
+    {
+      "epoch": 38.15,
+      "grad_norm": 0.8476659655570984,
+      "learning_rate": 0.0004740740740740741,
+      "loss": 1.0061,
+      "step": 10300
+    },
+    {
+      "epoch": 38.19,
+      "grad_norm": 0.9237126708030701,
+      "learning_rate": 0.00047259259259259265,
+      "loss": 1.0001,
+      "step": 10310
+    },
+    {
+      "epoch": 38.22,
+      "grad_norm": 0.8230653405189514,
+      "learning_rate": 0.0004711111111111111,
+      "loss": 1.0353,
+      "step": 10320
+    },
+    {
+      "epoch": 38.26,
+      "grad_norm": 0.8375163078308105,
+      "learning_rate": 0.00046962962962962963,
+      "loss": 1.0583,
+      "step": 10330
+    },
+    {
+      "epoch": 38.3,
+      "grad_norm": 0.8810118436813354,
+      "learning_rate": 0.00046814814814814815,
+      "loss": 1.0576,
+      "step": 10340
+    },
+    {
+      "epoch": 38.33,
+      "grad_norm": 0.9873121380805969,
+      "learning_rate": 0.00046666666666666666,
+      "loss": 1.0278,
+      "step": 10350
+    },
+    {
+      "epoch": 38.37,
+      "grad_norm": 0.8619394898414612,
+      "learning_rate": 0.00046518518518518523,
+      "loss": 1.0851,
+      "step": 10360
+    },
+    {
+      "epoch": 38.41,
+      "grad_norm": 0.8966696262359619,
+      "learning_rate": 0.00046370370370370375,
+      "loss": 1.0599,
+      "step": 10370
+    },
+    {
+      "epoch": 38.44,
+      "grad_norm": 0.8234748840332031,
+      "learning_rate": 0.0004622222222222222,
+      "loss": 1.0872,
+      "step": 10380
+    },
+    {
+      "epoch": 38.48,
+      "grad_norm": 0.8345742225646973,
+      "learning_rate": 0.00046074074074074073,
+      "loss": 1.0703,
+      "step": 10390
+    },
+    {
+      "epoch": 38.52,
+      "grad_norm": 0.957393229007721,
+      "learning_rate": 0.00045925925925925925,
+      "loss": 1.0654,
+      "step": 10400
+    },
+    {
+      "epoch": 38.56,
+      "grad_norm": 0.8456324338912964,
+      "learning_rate": 0.0004577777777777778,
+      "loss": 1.082,
+      "step": 10410
+    },
+    {
+      "epoch": 38.59,
+      "grad_norm": 0.923425555229187,
+      "learning_rate": 0.00045629629629629633,
+      "loss": 1.0821,
+      "step": 10420
+    },
+    {
+      "epoch": 38.63,
+      "grad_norm": 0.8464405536651611,
+      "learning_rate": 0.00045481481481481485,
+      "loss": 1.1023,
+      "step": 10430
+    },
+    {
+      "epoch": 38.67,
+      "grad_norm": 0.8511638641357422,
+      "learning_rate": 0.0004533333333333333,
+      "loss": 1.0764,
+      "step": 10440
+    },
+    {
+      "epoch": 38.7,
+      "grad_norm": 0.8756155371665955,
+      "learning_rate": 0.00045185185185185183,
+      "loss": 1.0985,
+      "step": 10450
+    },
+    {
+      "epoch": 38.74,
+      "grad_norm": 0.9443466067314148,
+      "learning_rate": 0.0004503703703703704,
+      "loss": 1.1203,
+      "step": 10460
+    },
+    {
+      "epoch": 38.78,
+      "grad_norm": 0.9254005551338196,
+      "learning_rate": 0.0004488888888888889,
+      "loss": 1.1035,
+      "step": 10470
+    },
+    {
+      "epoch": 38.81,
+      "grad_norm": 0.8819862008094788,
+      "learning_rate": 0.00044740740740740743,
+      "loss": 1.1301,
+      "step": 10480
+    },
+    {
+      "epoch": 38.85,
+      "grad_norm": 0.9181599617004395,
+      "learning_rate": 0.00044592592592592595,
+      "loss": 1.1231,
+      "step": 10490
+    },
+    {
+      "epoch": 38.89,
+      "grad_norm": 0.8710459470748901,
+      "learning_rate": 0.0004444444444444444,
+      "loss": 1.1396,
+      "step": 10500
+    },
+    {
+      "epoch": 38.93,
+      "grad_norm": 0.8439869284629822,
+      "learning_rate": 0.000442962962962963,
+      "loss": 1.1424,
+      "step": 10510
+    },
+    {
+      "epoch": 38.96,
+      "grad_norm": 0.9218877553939819,
+      "learning_rate": 0.0004414814814814815,
+      "loss": 1.1299,
+      "step": 10520
+    },
+    {
+      "epoch": 39.0,
+      "grad_norm": 1.462300419807434,
+      "learning_rate": 0.00044,
+      "loss": 1.1464,
+      "step": 10530
+    },
+    {
+      "epoch": 39.04,
+      "grad_norm": 0.8511271476745605,
+      "learning_rate": 0.00043851851851851853,
+      "loss": 0.9556,
+      "step": 10540
+    },
+    {
+      "epoch": 39.07,
+      "grad_norm": 0.8737080097198486,
+      "learning_rate": 0.00043703703703703705,
+      "loss": 0.9775,
+      "step": 10550
+    },
+    {
+      "epoch": 39.11,
+      "grad_norm": 0.8078649640083313,
+      "learning_rate": 0.0004355555555555555,
+      "loss": 0.9902,
+      "step": 10560
+    },
+    {
+      "epoch": 39.15,
+      "grad_norm": 0.885040819644928,
+      "learning_rate": 0.0004340740740740741,
+      "loss": 0.9642,
+      "step": 10570
+    },
+    {
+      "epoch": 39.19,
+      "grad_norm": 0.7933640480041504,
+      "learning_rate": 0.0004325925925925926,
+      "loss": 0.9942,
+      "step": 10580
+    },
+    {
+      "epoch": 39.22,
+      "grad_norm": 0.8872569799423218,
+      "learning_rate": 0.0004311111111111111,
+      "loss": 0.9867,
+      "step": 10590
+    },
+    {
+      "epoch": 39.26,
+      "grad_norm": 0.9073619842529297,
+      "learning_rate": 0.00042962962962962963,
+      "loss": 0.9865,
+      "step": 10600
+    },
+    {
+      "epoch": 39.3,
+      "grad_norm": 0.8212401270866394,
+      "learning_rate": 0.00042814814814814815,
+      "loss": 1.0186,
+      "step": 10610
+    },
+    {
+      "epoch": 39.33,
+      "grad_norm": 0.9199197292327881,
+      "learning_rate": 0.0004266666666666667,
+      "loss": 1.036,
+      "step": 10620
+    },
+    {
+      "epoch": 39.37,
+      "grad_norm": 0.8489059805870056,
+      "learning_rate": 0.0004251851851851852,
+      "loss": 1.0283,
+      "step": 10630
+    },
+    {
+      "epoch": 39.41,
+      "grad_norm": 0.8785027861595154,
+      "learning_rate": 0.0004237037037037037,
+      "loss": 1.0366,
+      "step": 10640
+    },
+    {
+      "epoch": 39.44,
+      "grad_norm": 0.9491832256317139,
+      "learning_rate": 0.0004222222222222222,
+      "loss": 1.0272,
+      "step": 10650
+    },
+    {
+      "epoch": 39.48,
+      "grad_norm": 0.8814444541931152,
+      "learning_rate": 0.00042074074074074073,
+      "loss": 1.0486,
+      "step": 10660
+    },
+    {
+      "epoch": 39.52,
+      "grad_norm": 0.8941586017608643,
+      "learning_rate": 0.0004192592592592593,
+      "loss": 1.0587,
+      "step": 10670
+    },
+    {
+      "epoch": 39.56,
+      "grad_norm": 0.8640462756156921,
+      "learning_rate": 0.0004177777777777778,
+      "loss": 1.0482,
+      "step": 10680
+    },
+    {
+      "epoch": 39.59,
+      "grad_norm": 0.9424142241477966,
+      "learning_rate": 0.0004162962962962963,
+      "loss": 1.0624,
+      "step": 10690
+    },
+    {
+      "epoch": 39.63,
+      "grad_norm": 0.883101224899292,
+      "learning_rate": 0.0004148148148148148,
+      "loss": 1.0434,
+      "step": 10700
+    },
+    {
+      "epoch": 39.67,
+      "grad_norm": 0.9248467087745667,
+      "learning_rate": 0.0004133333333333333,
+      "loss": 1.0626,
+      "step": 10710
+    },
+    {
+      "epoch": 39.7,
+      "grad_norm": 0.9620226621627808,
+      "learning_rate": 0.0004118518518518519,
+      "loss": 1.0794,
+      "step": 10720
+    },
+    {
+      "epoch": 39.74,
+      "grad_norm": 0.8799591064453125,
+      "learning_rate": 0.0004103703703703704,
+      "loss": 1.072,
+      "step": 10730
+    },
+    {
+      "epoch": 39.78,
+      "grad_norm": 0.9275612831115723,
+      "learning_rate": 0.0004088888888888889,
+      "loss": 1.0961,
+      "step": 10740
+    },
+    {
+      "epoch": 39.81,
+      "grad_norm": 0.8838842511177063,
+      "learning_rate": 0.0004074074074074074,
+      "loss": 1.0729,
+      "step": 10750
+    },
+    {
+      "epoch": 39.85,
+      "grad_norm": 0.8704506158828735,
+      "learning_rate": 0.0004059259259259259,
+      "loss": 1.0743,
+      "step": 10760
+    },
+    {
+      "epoch": 39.89,
+      "grad_norm": 0.9030023217201233,
+      "learning_rate": 0.00040444444444444447,
+      "loss": 1.1022,
+      "step": 10770
+    },
+    {
+      "epoch": 39.93,
+      "grad_norm": 0.9882118701934814,
+      "learning_rate": 0.000402962962962963,
+      "loss": 1.1086,
+      "step": 10780
+    },
+    {
+      "epoch": 39.96,
+      "grad_norm": 0.848917543888092,
+      "learning_rate": 0.0004014814814814815,
+      "loss": 1.1086,
+      "step": 10790
+    },
+    {
+      "epoch": 40.0,
+      "grad_norm": 1.5683587789535522,
+      "learning_rate": 0.0004,
+      "loss": 1.1207,
+      "step": 10800
+    },
+    {
+      "epoch": 40.04,
+      "grad_norm": 0.8366634845733643,
+      "learning_rate": 0.0003985185185185185,
+      "loss": 0.9414,
+      "step": 10810
+    },
+    {
+      "epoch": 40.07,
+      "grad_norm": 0.876707911491394,
+      "learning_rate": 0.00039703703703703705,
+      "loss": 0.9212,
+      "step": 10820
+    },
+    {
+      "epoch": 40.11,
+      "grad_norm": 0.7773799300193787,
+      "learning_rate": 0.00039555555555555557,
+      "loss": 0.9462,
+      "step": 10830
+    },
+    {
+      "epoch": 40.15,
+      "grad_norm": 0.7996877431869507,
+      "learning_rate": 0.0003940740740740741,
+      "loss": 0.9493,
+      "step": 10840
+    },
+    {
+      "epoch": 40.19,
+      "grad_norm": 0.8463727831840515,
+      "learning_rate": 0.0003925925925925926,
+      "loss": 0.9419,
+      "step": 10850
+    },
+    {
+      "epoch": 40.22,
+      "grad_norm": 0.8150336742401123,
+      "learning_rate": 0.0003911111111111111,
+      "loss": 0.9619,
+      "step": 10860
+    },
+    {
+      "epoch": 40.26,
+      "grad_norm": 0.8403350114822388,
+      "learning_rate": 0.00038962962962962964,
+      "loss": 0.968,
+      "step": 10870
+    },
+    {
+      "epoch": 40.3,
+      "grad_norm": 0.8616166710853577,
+      "learning_rate": 0.00038814814814814815,
+      "loss": 0.9785,
+      "step": 10880
+    },
+    {
+      "epoch": 40.33,
+      "grad_norm": 0.8425131440162659,
+      "learning_rate": 0.00038666666666666667,
+      "loss": 0.9913,
+      "step": 10890
+    },
+    {
+      "epoch": 40.37,
+      "grad_norm": 0.8531786799430847,
+      "learning_rate": 0.0003851851851851852,
+      "loss": 0.9692,
+      "step": 10900
+    },
+    {
+      "epoch": 40.41,
+      "grad_norm": 0.9102742671966553,
+      "learning_rate": 0.0003837037037037037,
+      "loss": 1.0184,
+      "step": 10910
+    },
+    {
+      "epoch": 40.44,
+      "grad_norm": 0.8685367703437805,
+      "learning_rate": 0.0003822222222222223,
+      "loss": 1.0278,
+      "step": 10920
+    },
+    {
+      "epoch": 40.48,
+      "grad_norm": 0.8668144941329956,
+      "learning_rate": 0.0003807407407407408,
+      "loss": 0.9819,
+      "step": 10930
+    },
+    {
+      "epoch": 40.52,
+      "grad_norm": 0.9007130861282349,
+      "learning_rate": 0.00037925925925925925,
+      "loss": 0.9945,
+      "step": 10940
+    },
+    {
+      "epoch": 40.56,
+      "grad_norm": 0.93947434425354,
+      "learning_rate": 0.00037777777777777777,
+      "loss": 1.0076,
+      "step": 10950
+    },
+    {
+      "epoch": 40.59,
+      "grad_norm": 0.8986189961433411,
+      "learning_rate": 0.0003762962962962963,
+      "loss": 1.0123,
+      "step": 10960
+    },
+    {
+      "epoch": 40.63,
+      "grad_norm": 0.8962591290473938,
+      "learning_rate": 0.00037481481481481486,
+      "loss": 1.0446,
+      "step": 10970
+    },
+    {
+      "epoch": 40.67,
+      "grad_norm": 0.8676485419273376,
+      "learning_rate": 0.0003733333333333334,
+      "loss": 1.036,
+      "step": 10980
+    },
+    {
+      "epoch": 40.7,
+      "grad_norm": 0.9445183873176575,
+      "learning_rate": 0.0003718518518518519,
+      "loss": 1.0697,
+      "step": 10990
+    },
+    {
+      "epoch": 40.74,
+      "grad_norm": 0.9026569128036499,
+      "learning_rate": 0.00037037037037037035,
+      "loss": 1.0636,
+      "step": 11000
+    },
+    {
+      "epoch": 40.78,
+      "grad_norm": 0.8249679803848267,
+      "learning_rate": 0.00036888888888888887,
+      "loss": 1.0346,
+      "step": 11010
+    },
+    {
+      "epoch": 40.81,
+      "grad_norm": 0.8658695220947266,
+      "learning_rate": 0.00036740740740740744,
+      "loss": 1.0539,
+      "step": 11020
+    },
+    {
+      "epoch": 40.85,
+      "grad_norm": 0.9501706957817078,
+      "learning_rate": 0.00036592592592592596,
+      "loss": 1.0831,
+      "step": 11030
+    },
+    {
+      "epoch": 40.89,
+      "grad_norm": 0.9114291667938232,
+      "learning_rate": 0.00036444444444444447,
+      "loss": 1.0748,
+      "step": 11040
+    },
+    {
+      "epoch": 40.93,
+      "grad_norm": 0.8624236583709717,
+      "learning_rate": 0.000362962962962963,
+      "loss": 1.072,
+      "step": 11050
+    },
+    {
+      "epoch": 40.96,
+      "grad_norm": 0.8550862669944763,
+      "learning_rate": 0.00036148148148148145,
+      "loss": 1.1031,
+      "step": 11060
+    },
+    {
+      "epoch": 41.0,
+      "grad_norm": 1.7575430870056152,
+      "learning_rate": 0.00035999999999999997,
+      "loss": 1.0836,
+      "step": 11070
+    },
+    {
+      "epoch": 41.04,
+      "grad_norm": 0.8067315220832825,
+      "learning_rate": 0.00035851851851851854,
+      "loss": 0.8977,
+      "step": 11080
+    },
+    {
+      "epoch": 41.07,
+      "grad_norm": 0.8133816123008728,
+      "learning_rate": 0.00035703703703703706,
+      "loss": 0.906,
+      "step": 11090
+    },
+    {
+      "epoch": 41.11,
+      "grad_norm": 0.7989582419395447,
+      "learning_rate": 0.00035555555555555557,
+      "loss": 0.9014,
+      "step": 11100
+    },
+    {
+      "epoch": 41.15,
+      "grad_norm": 0.7977972626686096,
+      "learning_rate": 0.0003540740740740741,
+      "loss": 0.9254,
+      "step": 11110
+    },
+    {
+      "epoch": 41.19,
+      "grad_norm": 0.8233643770217896,
+      "learning_rate": 0.00035259259259259255,
+      "loss": 0.9437,
+      "step": 11120
+    },
+    {
+      "epoch": 41.22,
+      "grad_norm": 0.8295071125030518,
+      "learning_rate": 0.0003511111111111111,
+      "loss": 0.9336,
+      "step": 11130
+    },
+    {
+      "epoch": 41.26,
+      "grad_norm": 0.855749785900116,
+      "learning_rate": 0.00034962962962962964,
+      "loss": 0.9252,
+      "step": 11140
+    },
+    {
+      "epoch": 41.3,
+      "grad_norm": 0.8856293559074402,
+      "learning_rate": 0.00034814814814814816,
+      "loss": 0.9455,
+      "step": 11150
+    },
+    {
+      "epoch": 41.33,
+      "grad_norm": 0.7978120446205139,
+      "learning_rate": 0.00034666666666666667,
+      "loss": 0.9425,
+      "step": 11160
+    },
+    {
+      "epoch": 41.37,
+      "grad_norm": 0.979863166809082,
+      "learning_rate": 0.0003451851851851852,
+      "loss": 0.9542,
+      "step": 11170
+    },
+    {
+      "epoch": 41.41,
+      "grad_norm": 0.9663028717041016,
+      "learning_rate": 0.0003437037037037037,
+      "loss": 0.9512,
+      "step": 11180
+    },
+    {
+      "epoch": 41.44,
+      "grad_norm": 0.8670268058776855,
+      "learning_rate": 0.0003422222222222222,
+      "loss": 0.9682,
+      "step": 11190
+    },
+    {
+      "epoch": 41.48,
+      "grad_norm": 0.8599923849105835,
+      "learning_rate": 0.00034074074074074074,
+      "loss": 0.9807,
+      "step": 11200
+    },
+    {
+      "epoch": 41.52,
+      "grad_norm": 0.8521960377693176,
+      "learning_rate": 0.00033925925925925926,
+      "loss": 1.0195,
+      "step": 11210
+    },
+    {
+      "epoch": 41.56,
+      "grad_norm": 0.8760194778442383,
+      "learning_rate": 0.00033777777777777777,
+      "loss": 1.0155,
+      "step": 11220
+    },
+    {
+      "epoch": 41.59,
+      "grad_norm": 0.9435073733329773,
+      "learning_rate": 0.00033629629629629634,
+      "loss": 1.0335,
+      "step": 11230
+    },
+    {
+      "epoch": 41.63,
+      "grad_norm": 0.95516037940979,
+      "learning_rate": 0.0003348148148148148,
+      "loss": 1.0216,
+      "step": 11240
+    },
+    {
+      "epoch": 41.67,
+      "grad_norm": 0.9270447492599487,
+      "learning_rate": 0.0003333333333333333,
+      "loss": 0.9949,
+      "step": 11250
+    },
+    {
+      "epoch": 41.7,
+      "grad_norm": 0.9160760045051575,
+      "learning_rate": 0.00033185185185185184,
+      "loss": 1.0288,
+      "step": 11260
+    },
+    {
+      "epoch": 41.74,
+      "grad_norm": 0.9127193093299866,
+      "learning_rate": 0.00033037037037037036,
+      "loss": 1.0357,
+      "step": 11270
+    },
+    {
+      "epoch": 41.78,
+      "grad_norm": 0.993202269077301,
+      "learning_rate": 0.0003288888888888889,
+      "loss": 1.0154,
+      "step": 11280
+    },
+    {
+      "epoch": 41.81,
+      "grad_norm": 0.9545233845710754,
+      "learning_rate": 0.00032740740740740744,
+      "loss": 1.0064,
+      "step": 11290
+    },
+    {
+      "epoch": 41.85,
+      "grad_norm": 0.9582635760307312,
+      "learning_rate": 0.00032592592592592596,
+      "loss": 1.0345,
+      "step": 11300
+    },
+    {
+      "epoch": 41.89,
+      "grad_norm": 0.8648403882980347,
+      "learning_rate": 0.0003244444444444444,
+      "loss": 0.9964,
+      "step": 11310
+    },
+    {
+      "epoch": 41.93,
+      "grad_norm": 0.9680795669555664,
+      "learning_rate": 0.00032296296296296294,
+      "loss": 1.0256,
+      "step": 11320
+    },
+    {
+      "epoch": 41.96,
+      "grad_norm": 0.9373118281364441,
+      "learning_rate": 0.0003214814814814815,
+      "loss": 1.0366,
+      "step": 11330
+    },
+    {
+      "epoch": 42.0,
+      "grad_norm": 1.8185220956802368,
+      "learning_rate": 0.00032,
+      "loss": 1.036,
+      "step": 11340
+    },
+    {
+      "epoch": 42.04,
+      "grad_norm": 0.8170128464698792,
+      "learning_rate": 0.00031851851851851854,
+      "loss": 0.8674,
+      "step": 11350
+    },
+    {
+      "epoch": 42.07,
+      "grad_norm": 0.8094532489776611,
+      "learning_rate": 0.00031703703703703706,
+      "loss": 0.8846,
+      "step": 11360
+    },
+    {
+      "epoch": 42.11,
+      "grad_norm": 0.796272873878479,
+      "learning_rate": 0.0003155555555555555,
+      "loss": 0.8928,
+      "step": 11370
+    },
+    {
+      "epoch": 42.15,
+      "grad_norm": 0.8592316508293152,
+      "learning_rate": 0.0003140740740740741,
+      "loss": 0.8658,
+      "step": 11380
+    },
+    {
+      "epoch": 42.19,
+      "grad_norm": 0.9022819399833679,
+      "learning_rate": 0.0003125925925925926,
+      "loss": 0.9059,
+      "step": 11390
+    },
+    {
+      "epoch": 42.22,
+      "grad_norm": 0.8500738739967346,
+      "learning_rate": 0.0003111111111111111,
+      "loss": 0.8895,
+      "step": 11400
+    },
+    {
+      "epoch": 42.26,
+      "grad_norm": 0.9119003415107727,
+      "learning_rate": 0.00030962962962962964,
+      "loss": 0.9162,
+      "step": 11410
+    },
+    {
+      "epoch": 42.3,
+      "grad_norm": 0.8775156140327454,
+      "learning_rate": 0.00030814814814814816,
+      "loss": 0.9242,
+      "step": 11420
+    },
+    {
+      "epoch": 42.33,
+      "grad_norm": 0.882416307926178,
+      "learning_rate": 0.0003066666666666667,
+      "loss": 0.9143,
+      "step": 11430
+    },
+    {
+      "epoch": 42.37,
+      "grad_norm": 0.928913950920105,
+      "learning_rate": 0.0003051851851851852,
+      "loss": 0.9363,
+      "step": 11440
+    },
+    {
+      "epoch": 42.41,
+      "grad_norm": 0.8908606171607971,
+      "learning_rate": 0.0003037037037037037,
+      "loss": 0.9727,
+      "step": 11450
+    },
+    {
+      "epoch": 42.44,
+      "grad_norm": 0.8055885434150696,
+      "learning_rate": 0.0003022222222222222,
+      "loss": 0.9555,
+      "step": 11460
+    },
+    {
+      "epoch": 42.48,
+      "grad_norm": 0.8678398728370667,
+      "learning_rate": 0.00030074074074074074,
+      "loss": 0.9621,
+      "step": 11470
+    },
+    {
+      "epoch": 42.52,
+      "grad_norm": 0.9374486804008484,
+      "learning_rate": 0.0002992592592592593,
+      "loss": 0.9392,
+      "step": 11480
+    },
+    {
+      "epoch": 42.56,
+      "grad_norm": 0.8560148477554321,
+      "learning_rate": 0.0002977777777777778,
+      "loss": 0.969,
+      "step": 11490
+    },
+    {
+      "epoch": 42.59,
+      "grad_norm": 0.9379673004150391,
+      "learning_rate": 0.0002962962962962963,
+      "loss": 0.9561,
+      "step": 11500
+    },
+    {
+      "epoch": 42.63,
+      "grad_norm": 0.8736953735351562,
+      "learning_rate": 0.0002948148148148148,
+      "loss": 0.9687,
+      "step": 11510
+    },
+    {
+      "epoch": 42.67,
+      "grad_norm": 0.8709719777107239,
+      "learning_rate": 0.0002933333333333333,
+      "loss": 0.9725,
+      "step": 11520
+    },
+    {
+      "epoch": 42.7,
+      "grad_norm": 0.9811522364616394,
+      "learning_rate": 0.00029185185185185184,
+      "loss": 0.9799,
+      "step": 11530
+    },
+    {
+      "epoch": 42.74,
+      "grad_norm": 0.9396344423294067,
+      "learning_rate": 0.0002903703703703704,
+      "loss": 0.9782,
+      "step": 11540
+    },
+    {
+      "epoch": 42.78,
+      "grad_norm": 0.8983749747276306,
+      "learning_rate": 0.0002888888888888889,
+      "loss": 1.0041,
+      "step": 11550
+    },
+    {
+      "epoch": 42.81,
+      "grad_norm": 0.8971825242042542,
+      "learning_rate": 0.0002874074074074074,
+      "loss": 0.9945,
+      "step": 11560
+    },
+    {
+      "epoch": 42.85,
+      "grad_norm": 0.9464390277862549,
+      "learning_rate": 0.0002859259259259259,
+      "loss": 0.9863,
+      "step": 11570
+    },
+    {
+      "epoch": 42.89,
+      "grad_norm": 0.8920719027519226,
+      "learning_rate": 0.0002844444444444444,
+      "loss": 1.0018,
+      "step": 11580
+    },
+    {
+      "epoch": 42.93,
+      "grad_norm": 0.9418626427650452,
+      "learning_rate": 0.000282962962962963,
+      "loss": 1.0104,
+      "step": 11590
+    },
+    {
+      "epoch": 42.96,
+      "grad_norm": 0.9117964506149292,
+      "learning_rate": 0.0002814814814814815,
+      "loss": 1.011,
+      "step": 11600
+    },
+    {
+      "epoch": 43.0,
+      "grad_norm": 1.7632156610488892,
+      "learning_rate": 0.00028000000000000003,
+      "loss": 0.9915,
+      "step": 11610
+    },
+    {
+      "epoch": 43.04,
+      "grad_norm": 0.8767898082733154,
+      "learning_rate": 0.0002785185185185185,
+      "loss": 0.8568,
+      "step": 11620
+    },
+    {
+      "epoch": 43.07,
+      "grad_norm": 0.7765992283821106,
+      "learning_rate": 0.000277037037037037,
+      "loss": 0.8576,
+      "step": 11630
+    },
+    {
+      "epoch": 43.11,
+      "grad_norm": 0.8032299876213074,
+      "learning_rate": 0.0002755555555555556,
+      "loss": 0.8709,
+      "step": 11640
+    },
+    {
+      "epoch": 43.15,
+      "grad_norm": 0.7957133054733276,
+      "learning_rate": 0.0002740740740740741,
+      "loss": 0.875,
+      "step": 11650
+    },
+    {
+      "epoch": 43.19,
+      "grad_norm": 0.8329715728759766,
+      "learning_rate": 0.0002725925925925926,
+      "loss": 0.8807,
+      "step": 11660
+    },
+    {
+      "epoch": 43.22,
+      "grad_norm": 0.9103427529335022,
+      "learning_rate": 0.00027111111111111113,
+      "loss": 0.8938,
+      "step": 11670
+    },
+    {
+      "epoch": 43.26,
+      "grad_norm": 0.8969374299049377,
+      "learning_rate": 0.0002696296296296296,
+      "loss": 0.9047,
+      "step": 11680
+    },
+    {
+      "epoch": 43.3,
+      "grad_norm": 0.8393443822860718,
+      "learning_rate": 0.00026814814814814816,
+      "loss": 0.8889,
+      "step": 11690
+    },
+    {
+      "epoch": 43.33,
+      "grad_norm": 0.8780049681663513,
+      "learning_rate": 0.0002666666666666667,
+      "loss": 0.9081,
+      "step": 11700
+    },
+    {
+      "epoch": 43.37,
+      "grad_norm": 0.838235080242157,
+      "learning_rate": 0.0002651851851851852,
+      "loss": 0.8852,
+      "step": 11710
+    },
+    {
+      "epoch": 43.41,
+      "grad_norm": 0.9195196032524109,
+      "learning_rate": 0.0002637037037037037,
+      "loss": 0.9201,
+      "step": 11720
+    },
+    {
+      "epoch": 43.44,
+      "grad_norm": 0.8491339683532715,
+      "learning_rate": 0.00026222222222222223,
+      "loss": 0.9016,
+      "step": 11730
+    },
+    {
+      "epoch": 43.48,
+      "grad_norm": 0.861749529838562,
+      "learning_rate": 0.00026074074074074075,
+      "loss": 0.9327,
+      "step": 11740
+    },
+    {
+      "epoch": 43.52,
+      "grad_norm": 0.8553155660629272,
+      "learning_rate": 0.00025925925925925926,
+      "loss": 0.9025,
+      "step": 11750
+    },
+    {
+      "epoch": 43.56,
+      "grad_norm": 0.923421323299408,
+      "learning_rate": 0.0002577777777777778,
+      "loss": 0.9277,
+      "step": 11760
+    },
+    {
+      "epoch": 43.59,
+      "grad_norm": 0.949780285358429,
+      "learning_rate": 0.0002562962962962963,
+      "loss": 0.9283,
+      "step": 11770
+    },
+    {
+      "epoch": 43.63,
+      "grad_norm": 0.8996204733848572,
+      "learning_rate": 0.0002548148148148148,
+      "loss": 0.9325,
+      "step": 11780
+    },
+    {
+      "epoch": 43.67,
+      "grad_norm": 0.8850076198577881,
+      "learning_rate": 0.0002533333333333334,
+      "loss": 0.9396,
+      "step": 11790
+    },
+    {
+      "epoch": 43.7,
+      "grad_norm": 0.9062913656234741,
+      "learning_rate": 0.00025185185185185185,
+      "loss": 0.9394,
+      "step": 11800
+    },
+    {
+      "epoch": 43.74,
+      "grad_norm": 0.8735949397087097,
+      "learning_rate": 0.00025037037037037036,
+      "loss": 0.9475,
+      "step": 11810
+    },
+    {
+      "epoch": 43.78,
+      "grad_norm": 0.9367344379425049,
+      "learning_rate": 0.0002488888888888889,
+      "loss": 0.9474,
+      "step": 11820
+    },
+    {
+      "epoch": 43.81,
+      "grad_norm": 0.9842424988746643,
+      "learning_rate": 0.0002474074074074074,
+      "loss": 0.9767,
+      "step": 11830
+    },
+    {
+      "epoch": 43.85,
+      "grad_norm": 0.9207190275192261,
+      "learning_rate": 0.00024592592592592597,
+      "loss": 0.9553,
+      "step": 11840
+    },
+    {
+      "epoch": 43.89,
+      "grad_norm": 0.8815023899078369,
+      "learning_rate": 0.00024444444444444443,
+      "loss": 0.947,
+      "step": 11850
+    },
+    {
+      "epoch": 43.93,
+      "grad_norm": 0.9600652456283569,
+      "learning_rate": 0.00024296296296296297,
+      "loss": 0.9737,
+      "step": 11860
+    },
+    {
+      "epoch": 43.96,
+      "grad_norm": 0.9066606760025024,
+      "learning_rate": 0.0002414814814814815,
+      "loss": 0.9744,
+      "step": 11870
+    },
+    {
+      "epoch": 44.0,
+      "grad_norm": 1.5773016214370728,
+      "learning_rate": 0.00024,
+      "loss": 0.9578,
+      "step": 11880
+    },
+    {
+      "epoch": 44.04,
+      "grad_norm": 0.8049682378768921,
+      "learning_rate": 0.00023851851851851852,
+      "loss": 0.8292,
+      "step": 11890
+    },
+    {
+      "epoch": 44.07,
+      "grad_norm": 0.7937451601028442,
+      "learning_rate": 0.00023703703703703704,
+      "loss": 0.8328,
+      "step": 11900
+    },
+    {
+      "epoch": 44.11,
+      "grad_norm": 0.8170547485351562,
+      "learning_rate": 0.00023555555555555556,
+      "loss": 0.8325,
+      "step": 11910
+    },
+    {
+      "epoch": 44.15,
+      "grad_norm": 0.8086414337158203,
+      "learning_rate": 0.00023407407407407407,
+      "loss": 0.8293,
+      "step": 11920
+    },
+    {
+      "epoch": 44.19,
+      "grad_norm": 0.8746134042739868,
+      "learning_rate": 0.00023259259259259262,
+      "loss": 0.8795,
+      "step": 11930
+    },
+    {
+      "epoch": 44.22,
+      "grad_norm": 0.8427841663360596,
+      "learning_rate": 0.0002311111111111111,
+      "loss": 0.8783,
+      "step": 11940
+    },
+    {
+      "epoch": 44.26,
+      "grad_norm": 0.8464419841766357,
+      "learning_rate": 0.00022962962962962962,
+      "loss": 0.8574,
+      "step": 11950
+    },
+    {
+      "epoch": 44.3,
+      "grad_norm": 0.947233259677887,
+      "learning_rate": 0.00022814814814814817,
+      "loss": 0.8892,
+      "step": 11960
+    },
+    {
+      "epoch": 44.33,
+      "grad_norm": 0.850940465927124,
+      "learning_rate": 0.00022666666666666666,
+      "loss": 0.8745,
+      "step": 11970
+    },
+    {
+      "epoch": 44.37,
+      "grad_norm": 0.8650044202804565,
+      "learning_rate": 0.0002251851851851852,
+      "loss": 0.8772,
+      "step": 11980
+    },
+    {
+      "epoch": 44.41,
+      "grad_norm": 0.8571513891220093,
+      "learning_rate": 0.00022370370370370372,
+      "loss": 0.8875,
+      "step": 11990
+    },
+    {
+      "epoch": 44.44,
+      "grad_norm": 0.8866828680038452,
+      "learning_rate": 0.0002222222222222222,
+      "loss": 0.8747,
+      "step": 12000
+    },
+    {
+      "epoch": 44.48,
+      "grad_norm": 0.9303273558616638,
+      "learning_rate": 0.00022074074074074075,
+      "loss": 0.9052,
+      "step": 12010
+    },
+    {
+      "epoch": 44.52,
+      "grad_norm": 0.8586504459381104,
+      "learning_rate": 0.00021925925925925927,
+      "loss": 0.8762,
+      "step": 12020
+    },
+    {
+      "epoch": 44.56,
+      "grad_norm": 0.8415549397468567,
+      "learning_rate": 0.00021777777777777776,
+      "loss": 0.8934,
+      "step": 12030
+    },
+    {
+      "epoch": 44.59,
+      "grad_norm": 0.894075334072113,
+      "learning_rate": 0.0002162962962962963,
+      "loss": 0.885,
+      "step": 12040
+    },
+    {
+      "epoch": 44.63,
+      "grad_norm": 0.9093772768974304,
+      "learning_rate": 0.00021481481481481482,
+      "loss": 0.9059,
+      "step": 12050
+    },
+    {
+      "epoch": 44.67,
+      "grad_norm": 1.0175001621246338,
+      "learning_rate": 0.00021333333333333336,
+      "loss": 0.9157,
+      "step": 12060
+    },
+    {
+      "epoch": 44.7,
+      "grad_norm": 0.9057775735855103,
+      "learning_rate": 0.00021185185185185185,
+      "loss": 0.9139,
+      "step": 12070
+    },
+    {
+      "epoch": 44.74,
+      "grad_norm": 0.8696966767311096,
+      "learning_rate": 0.00021037037037037037,
+      "loss": 0.9275,
+      "step": 12080
+    },
+    {
+      "epoch": 44.78,
+      "grad_norm": 1.0306414365768433,
+      "learning_rate": 0.0002088888888888889,
+      "loss": 0.9056,
+      "step": 12090
+    },
+    {
+      "epoch": 44.81,
+      "grad_norm": 0.8880541920661926,
+      "learning_rate": 0.0002074074074074074,
+      "loss": 0.9099,
+      "step": 12100
+    },
+    {
+      "epoch": 44.85,
+      "grad_norm": 0.9045748710632324,
+      "learning_rate": 0.00020592592592592594,
+      "loss": 0.9254,
+      "step": 12110
+    },
+    {
+      "epoch": 44.89,
+      "grad_norm": 0.9737614989280701,
+      "learning_rate": 0.00020444444444444446,
+      "loss": 0.9166,
+      "step": 12120
+    },
+    {
+      "epoch": 44.93,
+      "grad_norm": 0.838835597038269,
+      "learning_rate": 0.00020296296296296295,
+      "loss": 0.9455,
+      "step": 12130
+    },
+    {
+      "epoch": 44.96,
+      "grad_norm": 0.911868691444397,
+      "learning_rate": 0.0002014814814814815,
+      "loss": 0.9586,
+      "step": 12140
+    },
+    {
+      "epoch": 45.0,
+      "grad_norm": 1.629447340965271,
+      "learning_rate": 0.0002,
+      "loss": 0.9213,
+      "step": 12150
+    },
+    {
+      "epoch": 45.04,
+      "grad_norm": 0.8545764684677124,
+      "learning_rate": 0.00019851851851851853,
+      "loss": 0.8333,
+      "step": 12160
+    },
+    {
+      "epoch": 45.07,
+      "grad_norm": 0.8149832487106323,
+      "learning_rate": 0.00019703703703703704,
+      "loss": 0.7863,
+      "step": 12170
+    },
+    {
+      "epoch": 45.11,
+      "grad_norm": 0.8192715644836426,
+      "learning_rate": 0.00019555555555555556,
+      "loss": 0.8435,
+      "step": 12180
+    },
+    {
+      "epoch": 45.15,
+      "grad_norm": 0.8543215394020081,
+      "learning_rate": 0.00019407407407407408,
+      "loss": 0.8246,
+      "step": 12190
+    },
+    {
+      "epoch": 45.19,
+      "grad_norm": 0.8113214373588562,
+      "learning_rate": 0.0001925925925925926,
+      "loss": 0.8264,
+      "step": 12200
+    },
+    {
+      "epoch": 45.22,
+      "grad_norm": 0.8288257718086243,
+      "learning_rate": 0.00019111111111111114,
+      "loss": 0.8308,
+      "step": 12210
+    },
+    {
+      "epoch": 45.26,
+      "grad_norm": 0.8256189227104187,
+      "learning_rate": 0.00018962962962962963,
+      "loss": 0.8362,
+      "step": 12220
+    },
+    {
+      "epoch": 45.3,
+      "grad_norm": 0.8684147596359253,
+      "learning_rate": 0.00018814814814814814,
+      "loss": 0.8363,
+      "step": 12230
+    },
+    {
+      "epoch": 45.33,
+      "grad_norm": 0.850592315196991,
+      "learning_rate": 0.0001866666666666667,
+      "loss": 0.853,
+      "step": 12240
+    },
+    {
+      "epoch": 45.37,
+      "grad_norm": 0.894404411315918,
+      "learning_rate": 0.00018518518518518518,
+      "loss": 0.8639,
+      "step": 12250
+    },
+    {
+      "epoch": 45.41,
+      "grad_norm": 0.8467885255813599,
+      "learning_rate": 0.00018370370370370372,
+      "loss": 0.8461,
+      "step": 12260
+    },
+    {
+      "epoch": 45.44,
+      "grad_norm": 0.9240211844444275,
+      "learning_rate": 0.00018222222222222224,
+      "loss": 0.8751,
+      "step": 12270
+    },
+    {
+      "epoch": 45.48,
+      "grad_norm": 0.8148699998855591,
+      "learning_rate": 0.00018074074074074073,
+      "loss": 0.8744,
+      "step": 12280
+    },
+    {
+      "epoch": 45.52,
+      "grad_norm": 0.8215749859809875,
+      "learning_rate": 0.00017925925925925927,
+      "loss": 0.8629,
+      "step": 12290
+    },
+    {
+      "epoch": 45.56,
+      "grad_norm": 0.9265359044075012,
+      "learning_rate": 0.00017777777777777779,
+      "loss": 0.882,
+      "step": 12300
+    },
+    {
+      "epoch": 45.59,
+      "grad_norm": 0.8920613527297974,
+      "learning_rate": 0.00017629629629629628,
+      "loss": 0.8522,
+      "step": 12310
+    },
+    {
+      "epoch": 45.63,
+      "grad_norm": 0.964079737663269,
+      "learning_rate": 0.00017481481481481482,
+      "loss": 0.873,
+      "step": 12320
+    },
+    {
+      "epoch": 45.67,
+      "grad_norm": 0.8842025399208069,
+      "learning_rate": 0.00017333333333333334,
+      "loss": 0.8642,
+      "step": 12330
+    },
+    {
+      "epoch": 45.7,
+      "grad_norm": 0.9127376079559326,
+      "learning_rate": 0.00017185185185185185,
+      "loss": 0.8805,
+      "step": 12340
+    },
+    {
+      "epoch": 45.74,
+      "grad_norm": 0.8502327799797058,
+      "learning_rate": 0.00017037037037037037,
+      "loss": 0.8818,
+      "step": 12350
+    },
+    {
+      "epoch": 45.78,
+      "grad_norm": 0.9649406671524048,
+      "learning_rate": 0.00016888888888888889,
+      "loss": 0.8934,
+      "step": 12360
+    },
+    {
+      "epoch": 45.81,
+      "grad_norm": 0.9439674615859985,
+      "learning_rate": 0.0001674074074074074,
+      "loss": 0.9173,
+      "step": 12370
+    },
+    {
+      "epoch": 45.85,
+      "grad_norm": 0.8551307916641235,
+      "learning_rate": 0.00016592592592592592,
+      "loss": 0.8811,
+      "step": 12380
+    },
+    {
+      "epoch": 45.89,
+      "grad_norm": 0.9205163717269897,
+      "learning_rate": 0.00016444444444444446,
+      "loss": 0.8846,
+      "step": 12390
+    },
+    {
+      "epoch": 45.93,
+      "grad_norm": 0.90606290102005,
+      "learning_rate": 0.00016296296296296298,
+      "loss": 0.8941,
+      "step": 12400
+    },
+    {
+      "epoch": 45.96,
+      "grad_norm": 0.9254661798477173,
+      "learning_rate": 0.00016148148148148147,
+      "loss": 0.901,
+      "step": 12410
+    },
+    {
+      "epoch": 46.0,
+      "grad_norm": 1.6181832551956177,
+      "learning_rate": 0.00016,
+      "loss": 0.8908,
+      "step": 12420
+    },
+    {
+      "epoch": 46.04,
+      "grad_norm": 0.8116936683654785,
+      "learning_rate": 0.00015851851851851853,
+      "loss": 0.8005,
+      "step": 12430
+    },
+    {
+      "epoch": 46.07,
+      "grad_norm": 0.749098002910614,
+      "learning_rate": 0.00015703703703703705,
+      "loss": 0.7914,
+      "step": 12440
+    },
+    {
+      "epoch": 46.11,
+      "grad_norm": 0.7693229913711548,
+      "learning_rate": 0.00015555555555555556,
+      "loss": 0.8053,
+      "step": 12450
+    },
+    {
+      "epoch": 46.15,
+      "grad_norm": 0.7828912138938904,
+      "learning_rate": 0.00015407407407407408,
+      "loss": 0.7946,
+      "step": 12460
+    },
+    {
+      "epoch": 46.19,
+      "grad_norm": 0.8083990216255188,
+      "learning_rate": 0.0001525925925925926,
+      "loss": 0.817,
+      "step": 12470
+    },
+    {
+      "epoch": 46.22,
+      "grad_norm": 0.827688455581665,
+      "learning_rate": 0.0001511111111111111,
+      "loss": 0.7919,
+      "step": 12480
+    },
+    {
+      "epoch": 46.26,
+      "grad_norm": 0.8141525983810425,
+      "learning_rate": 0.00014962962962962966,
+      "loss": 0.8346,
+      "step": 12490
+    },
+    {
+      "epoch": 46.3,
+      "grad_norm": 0.823664128780365,
+      "learning_rate": 0.00014814814814814815,
+      "loss": 0.84,
+      "step": 12500
+    },
+    {
+      "epoch": 46.33,
+      "grad_norm": 0.8547099232673645,
+      "learning_rate": 0.00014666666666666666,
+      "loss": 0.803,
+      "step": 12510
+    },
+    {
+      "epoch": 46.37,
+      "grad_norm": 0.8899357914924622,
+      "learning_rate": 0.0001451851851851852,
+      "loss": 0.8316,
+      "step": 12520
+    },
+    {
+      "epoch": 46.41,
+      "grad_norm": 0.8477994203567505,
+      "learning_rate": 0.0001437037037037037,
+      "loss": 0.8129,
+      "step": 12530
+    },
+    {
+      "epoch": 46.44,
+      "grad_norm": 0.7684743404388428,
+      "learning_rate": 0.0001422222222222222,
+      "loss": 0.8405,
+      "step": 12540
+    },
+    {
+      "epoch": 46.48,
+      "grad_norm": 0.816189706325531,
+      "learning_rate": 0.00014074074074074076,
+      "loss": 0.8365,
+      "step": 12550
+    },
+    {
+      "epoch": 46.52,
+      "grad_norm": 0.844874382019043,
+      "learning_rate": 0.00013925925925925925,
+      "loss": 0.8149,
+      "step": 12560
+    },
+    {
+      "epoch": 46.56,
+      "grad_norm": 0.8706566095352173,
+      "learning_rate": 0.0001377777777777778,
+      "loss": 0.8676,
+      "step": 12570
+    },
+    {
+      "epoch": 46.59,
+      "grad_norm": 0.8781683444976807,
+      "learning_rate": 0.0001362962962962963,
+      "loss": 0.8667,
+      "step": 12580
+    },
+    {
+      "epoch": 46.63,
+      "grad_norm": 0.8802633881568909,
+      "learning_rate": 0.0001348148148148148,
+      "loss": 0.8314,
+      "step": 12590
+    },
+    {
+      "epoch": 46.67,
+      "grad_norm": 0.8872252106666565,
+      "learning_rate": 0.00013333333333333334,
+      "loss": 0.858,
+      "step": 12600
+    },
+    {
+      "epoch": 46.7,
+      "grad_norm": 0.8980612754821777,
+      "learning_rate": 0.00013185185185185186,
+      "loss": 0.8538,
+      "step": 12610
+    },
+    {
+      "epoch": 46.74,
+      "grad_norm": 0.8398934006690979,
+      "learning_rate": 0.00013037037037037037,
+      "loss": 0.8457,
+      "step": 12620
+    },
+    {
+      "epoch": 46.78,
+      "grad_norm": 0.8653325438499451,
+      "learning_rate": 0.0001288888888888889,
+      "loss": 0.8621,
+      "step": 12630
+    },
+    {
+      "epoch": 46.81,
+      "grad_norm": 0.9367448687553406,
+      "learning_rate": 0.0001274074074074074,
+      "loss": 0.8437,
+      "step": 12640
+    },
+    {
+      "epoch": 46.85,
+      "grad_norm": 0.9195182919502258,
+      "learning_rate": 0.00012592592592592592,
+      "loss": 0.8675,
+      "step": 12650
+    },
+    {
+      "epoch": 46.89,
+      "grad_norm": 0.8260027170181274,
+      "learning_rate": 0.00012444444444444444,
+      "loss": 0.8702,
+      "step": 12660
+    },
+    {
+      "epoch": 46.93,
+      "grad_norm": 0.8661918044090271,
+      "learning_rate": 0.00012296296296296298,
+      "loss": 0.8738,
+      "step": 12670
+    },
+    {
+      "epoch": 46.96,
+      "grad_norm": 0.9230578541755676,
+      "learning_rate": 0.00012148148148148149,
+      "loss": 0.8542,
+      "step": 12680
+    },
+    {
+      "epoch": 47.0,
+      "grad_norm": 1.741862177848816,
+      "learning_rate": 0.00012,
+      "loss": 0.8547,
+      "step": 12690
+    },
+    {
+      "epoch": 47.04,
+      "grad_norm": 0.864289402961731,
+      "learning_rate": 0.00011851851851851852,
+      "loss": 0.7575,
+      "step": 12700
+    },
+    {
+      "epoch": 47.07,
+      "grad_norm": 0.7751877307891846,
+      "learning_rate": 0.00011703703703703704,
+      "loss": 0.7825,
+      "step": 12710
+    },
+    {
+      "epoch": 47.11,
+      "grad_norm": 0.8020772933959961,
+      "learning_rate": 0.00011555555555555555,
+      "loss": 0.787,
+      "step": 12720
+    },
+    {
+      "epoch": 47.15,
+      "grad_norm": 0.8673202395439148,
+      "learning_rate": 0.00011407407407407408,
+      "loss": 0.7693,
+      "step": 12730
+    },
+    {
+      "epoch": 47.19,
+      "grad_norm": 0.7707006335258484,
+      "learning_rate": 0.0001125925925925926,
+      "loss": 0.779,
+      "step": 12740
+    },
+    {
+      "epoch": 47.22,
+      "grad_norm": 0.7779736518859863,
+      "learning_rate": 0.0001111111111111111,
+      "loss": 0.7879,
+      "step": 12750
+    },
+    {
+      "epoch": 47.26,
+      "grad_norm": 0.8151842951774597,
+      "learning_rate": 0.00010962962962962963,
+      "loss": 0.806,
+      "step": 12760
+    },
+    {
+      "epoch": 47.3,
+      "grad_norm": 0.7836940884590149,
+      "learning_rate": 0.00010814814814814815,
+      "loss": 0.7884,
+      "step": 12770
+    },
+    {
+      "epoch": 47.33,
+      "grad_norm": 0.826682448387146,
+      "learning_rate": 0.00010666666666666668,
+      "loss": 0.8061,
+      "step": 12780
+    },
+    {
+      "epoch": 47.37,
+      "grad_norm": 0.8258213996887207,
+      "learning_rate": 0.00010518518518518518,
+      "loss": 0.7968,
+      "step": 12790
+    },
+    {
+      "epoch": 47.41,
+      "grad_norm": 0.8098385334014893,
+      "learning_rate": 0.0001037037037037037,
+      "loss": 0.8034,
+      "step": 12800
+    },
+    {
+      "epoch": 47.44,
+      "grad_norm": 0.8096972107887268,
+      "learning_rate": 0.00010222222222222223,
+      "loss": 0.813,
+      "step": 12810
+    },
+    {
+      "epoch": 47.48,
+      "grad_norm": 0.8447617292404175,
+      "learning_rate": 0.00010074074074074075,
+      "loss": 0.8093,
+      "step": 12820
+    },
+    {
+      "epoch": 47.52,
+      "grad_norm": 0.81069016456604,
+      "learning_rate": 9.925925925925926e-05,
+      "loss": 0.8279,
+      "step": 12830
+    },
+    {
+      "epoch": 47.56,
+      "grad_norm": 0.8630595207214355,
+      "learning_rate": 9.777777777777778e-05,
+      "loss": 0.8246,
+      "step": 12840
+    },
+    {
+      "epoch": 47.59,
+      "grad_norm": 0.839095950126648,
+      "learning_rate": 9.62962962962963e-05,
+      "loss": 0.8453,
+      "step": 12850
+    },
+    {
+      "epoch": 47.63,
+      "grad_norm": 0.8493757843971252,
+      "learning_rate": 9.481481481481481e-05,
+      "loss": 0.8245,
+      "step": 12860
+    },
+    {
+      "epoch": 47.67,
+      "grad_norm": 0.8431633114814758,
+      "learning_rate": 9.333333333333334e-05,
+      "loss": 0.8121,
+      "step": 12870
+    },
+    {
+      "epoch": 47.7,
+      "grad_norm": 0.8720538020133972,
+      "learning_rate": 9.185185185185186e-05,
+      "loss": 0.8136,
+      "step": 12880
+    },
+    {
+      "epoch": 47.74,
+      "grad_norm": 0.8341265320777893,
+      "learning_rate": 9.037037037037036e-05,
+      "loss": 0.8345,
+      "step": 12890
+    },
+    {
+      "epoch": 47.78,
+      "grad_norm": 0.853097677230835,
+      "learning_rate": 8.888888888888889e-05,
+      "loss": 0.8462,
+      "step": 12900
+    },
+    {
+      "epoch": 47.81,
+      "grad_norm": 0.8465114831924438,
+      "learning_rate": 8.740740740740741e-05,
+      "loss": 0.8283,
+      "step": 12910
+    },
+    {
+      "epoch": 47.85,
+      "grad_norm": 0.8404960632324219,
+      "learning_rate": 8.592592592592593e-05,
+      "loss": 0.8428,
+      "step": 12920
+    },
+    {
+      "epoch": 47.89,
+      "grad_norm": 0.9160316586494446,
+      "learning_rate": 8.444444444444444e-05,
+      "loss": 0.8383,
+      "step": 12930
+    },
+    {
+      "epoch": 47.93,
+      "grad_norm": 0.8309608101844788,
+      "learning_rate": 8.296296296296296e-05,
+      "loss": 0.8282,
+      "step": 12940
+    },
+    {
+      "epoch": 47.96,
+      "grad_norm": 0.9367932677268982,
+      "learning_rate": 8.148148148148149e-05,
+      "loss": 0.8221,
+      "step": 12950
+    },
+    {
+      "epoch": 48.0,
+      "grad_norm": 1.5219368934631348,
+      "learning_rate": 8e-05,
+      "loss": 0.8323,
+      "step": 12960
+    },
+    {
+      "epoch": 48.04,
+      "grad_norm": 0.7839010953903198,
+      "learning_rate": 7.851851851851852e-05,
+      "loss": 0.7676,
+      "step": 12970
+    },
+    {
+      "epoch": 48.07,
+      "grad_norm": 0.7914835214614868,
+      "learning_rate": 7.703703703703704e-05,
+      "loss": 0.7613,
+      "step": 12980
+    },
+    {
+      "epoch": 48.11,
+      "grad_norm": 0.7833623886108398,
+      "learning_rate": 7.555555555555556e-05,
+      "loss": 0.7553,
+      "step": 12990
+    },
+    {
+      "epoch": 48.15,
+      "grad_norm": 0.7814247608184814,
+      "learning_rate": 7.407407407407407e-05,
+      "loss": 0.7786,
+      "step": 13000
+    },
+    {
+      "epoch": 48.19,
+      "grad_norm": 0.7823523283004761,
+      "learning_rate": 7.25925925925926e-05,
+      "loss": 0.7827,
+      "step": 13010
+    },
+    {
+      "epoch": 48.22,
+      "grad_norm": 0.792614758014679,
+      "learning_rate": 7.11111111111111e-05,
+      "loss": 0.7865,
+      "step": 13020
+    },
+    {
+      "epoch": 48.26,
+      "grad_norm": 0.793111264705658,
+      "learning_rate": 6.962962962962962e-05,
+      "loss": 0.7847,
+      "step": 13030
+    },
+    {
+      "epoch": 48.3,
+      "grad_norm": 0.8105340600013733,
+      "learning_rate": 6.814814814814815e-05,
+      "loss": 0.7835,
+      "step": 13040
+    },
+    {
+      "epoch": 48.33,
+      "grad_norm": 0.8111973404884338,
+      "learning_rate": 6.666666666666667e-05,
+      "loss": 0.7689,
+      "step": 13050
+    },
+    {
+      "epoch": 48.37,
+      "grad_norm": 0.8513317108154297,
+      "learning_rate": 6.518518518518519e-05,
+      "loss": 0.7717,
+      "step": 13060
+    },
+    {
+      "epoch": 48.41,
+      "grad_norm": 0.8038240075111389,
+      "learning_rate": 6.37037037037037e-05,
+      "loss": 0.795,
+      "step": 13070
+    },
+    {
+      "epoch": 48.44,
+      "grad_norm": 0.7720206379890442,
+      "learning_rate": 6.222222222222222e-05,
+      "loss": 0.7906,
+      "step": 13080
+    },
+    {
+      "epoch": 48.48,
+      "grad_norm": 0.8711926937103271,
+      "learning_rate": 6.074074074074074e-05,
+      "loss": 0.7849,
+      "step": 13090
+    },
+    {
+      "epoch": 48.52,
+      "grad_norm": 0.8024256229400635,
+      "learning_rate": 5.925925925925926e-05,
+      "loss": 0.7792,
+      "step": 13100
+    },
+    {
+      "epoch": 48.56,
+      "grad_norm": 0.8680411577224731,
+      "learning_rate": 5.7777777777777776e-05,
+      "loss": 0.7717,
+      "step": 13110
+    },
+    {
+      "epoch": 48.59,
+      "grad_norm": 0.8359400033950806,
+      "learning_rate": 5.62962962962963e-05,
+      "loss": 0.7976,
+      "step": 13120
+    },
+    {
+      "epoch": 48.63,
+      "grad_norm": 0.8003830313682556,
+      "learning_rate": 5.4814814814814817e-05,
+      "loss": 0.7872,
+      "step": 13130
+    },
+    {
+      "epoch": 48.67,
+      "grad_norm": 0.799286961555481,
+      "learning_rate": 5.333333333333334e-05,
+      "loss": 0.7912,
+      "step": 13140
+    },
+    {
+      "epoch": 48.7,
+      "grad_norm": 0.8358856439590454,
+      "learning_rate": 5.185185185185185e-05,
+      "loss": 0.8027,
+      "step": 13150
+    },
+    {
+      "epoch": 48.74,
+      "grad_norm": 0.8304854035377502,
+      "learning_rate": 5.037037037037037e-05,
+      "loss": 0.7769,
+      "step": 13160
+    },
+    {
+      "epoch": 48.78,
+      "grad_norm": 0.8182996511459351,
+      "learning_rate": 4.888888888888889e-05,
+      "loss": 0.7943,
+      "step": 13170
+    },
+    {
+      "epoch": 48.81,
+      "grad_norm": 0.8343393802642822,
+      "learning_rate": 4.7407407407407407e-05,
+      "loss": 0.798,
+      "step": 13180
+    },
+    {
+      "epoch": 48.85,
+      "grad_norm": 0.8523905277252197,
+      "learning_rate": 4.592592592592593e-05,
+      "loss": 0.8004,
+      "step": 13190
+    },
+    {
+      "epoch": 48.89,
+      "grad_norm": 0.8108332753181458,
+      "learning_rate": 4.4444444444444447e-05,
+      "loss": 0.7972,
+      "step": 13200
+    },
+    {
+      "epoch": 48.93,
+      "grad_norm": 0.8315889239311218,
+      "learning_rate": 4.296296296296296e-05,
+      "loss": 0.8369,
+      "step": 13210
+    },
+    {
+      "epoch": 48.96,
+      "grad_norm": 0.8833268880844116,
+      "learning_rate": 4.148148148148148e-05,
+      "loss": 0.8205,
+      "step": 13220
+    },
+    {
+      "epoch": 49.0,
+      "grad_norm": 1.7572418451309204,
+      "learning_rate": 4e-05,
+      "loss": 0.7834,
+      "step": 13230
+    },
+    {
+      "epoch": 49.04,
+      "grad_norm": 0.7579549551010132,
+      "learning_rate": 3.851851851851852e-05,
+      "loss": 0.7437,
+      "step": 13240
+    },
+    {
+      "epoch": 49.07,
+      "grad_norm": 0.8062260150909424,
+      "learning_rate": 3.7037037037037037e-05,
+      "loss": 0.7523,
+      "step": 13250
+    },
+    {
+      "epoch": 49.11,
+      "grad_norm": 0.758929967880249,
+      "learning_rate": 3.555555555555555e-05,
+      "loss": 0.7441,
+      "step": 13260
+    },
+    {
+      "epoch": 49.15,
+      "grad_norm": 0.8011849522590637,
+      "learning_rate": 3.4074074074074077e-05,
+      "loss": 0.7644,
+      "step": 13270
+    },
+    {
+      "epoch": 49.19,
+      "grad_norm": 0.7755334377288818,
+      "learning_rate": 3.259259259259259e-05,
+      "loss": 0.7656,
+      "step": 13280
+    },
+    {
+      "epoch": 49.22,
+      "grad_norm": 0.7567688226699829,
+      "learning_rate": 3.111111111111111e-05,
+      "loss": 0.7662,
+      "step": 13290
+    },
+    {
+      "epoch": 49.26,
+      "grad_norm": 0.7689881920814514,
+      "learning_rate": 2.962962962962963e-05,
+      "loss": 0.7661,
+      "step": 13300
+    },
+    {
+      "epoch": 49.3,
+      "grad_norm": 0.803693950176239,
+      "learning_rate": 2.814814814814815e-05,
+      "loss": 0.7798,
+      "step": 13310
+    },
+    {
+      "epoch": 49.33,
+      "grad_norm": 0.7691094875335693,
+      "learning_rate": 2.666666666666667e-05,
+      "loss": 0.7601,
+      "step": 13320
+    },
+    {
+      "epoch": 49.37,
+      "grad_norm": 0.7872125506401062,
+      "learning_rate": 2.5185185185185187e-05,
+      "loss": 0.7695,
+      "step": 13330
+    },
+    {
+      "epoch": 49.41,
+      "grad_norm": 0.7724753618240356,
+      "learning_rate": 2.3703703703703703e-05,
+      "loss": 0.7635,
+      "step": 13340
+    },
+    {
+      "epoch": 49.44,
+      "grad_norm": 0.7950810194015503,
+      "learning_rate": 2.2222222222222223e-05,
+      "loss": 0.7501,
+      "step": 13350
+    },
+    {
+      "epoch": 49.48,
+      "grad_norm": 0.8038039803504944,
+      "learning_rate": 2.074074074074074e-05,
+      "loss": 0.7426,
+      "step": 13360
+    },
+    {
+      "epoch": 49.52,
+      "grad_norm": 0.7851365804672241,
+      "learning_rate": 1.925925925925926e-05,
+      "loss": 0.7925,
+      "step": 13370
+    },
+    {
+      "epoch": 49.56,
+      "grad_norm": 0.797153115272522,
+      "learning_rate": 1.7777777777777777e-05,
+      "loss": 0.7748,
+      "step": 13380
+    },
+    {
+      "epoch": 49.59,
+      "grad_norm": 0.7819933891296387,
+      "learning_rate": 1.6296296296296297e-05,
+      "loss": 0.7626,
+      "step": 13390
+    },
+    {
+      "epoch": 49.63,
+      "grad_norm": 0.7707337737083435,
+      "learning_rate": 1.4814814814814815e-05,
+      "loss": 0.7607,
+      "step": 13400
+    },
+    {
+      "epoch": 49.67,
+      "grad_norm": 0.7730938792228699,
+      "learning_rate": 1.3333333333333335e-05,
+      "loss": 0.7886,
+      "step": 13410
+    },
+    {
+      "epoch": 49.7,
+      "grad_norm": 0.8007907867431641,
+      "learning_rate": 1.1851851851851852e-05,
+      "loss": 0.7844,
+      "step": 13420
+    },
+    {
+      "epoch": 49.74,
+      "grad_norm": 0.7911803126335144,
+      "learning_rate": 1.037037037037037e-05,
+      "loss": 0.8026,
+      "step": 13430
+    },
+    {
+      "epoch": 49.78,
+      "grad_norm": 0.8312400579452515,
+      "learning_rate": 8.888888888888888e-06,
+      "loss": 0.7486,
+      "step": 13440
+    },
+    {
+      "epoch": 49.81,
+      "grad_norm": 0.7924412488937378,
+      "learning_rate": 7.4074074074074075e-06,
+      "loss": 0.7762,
+      "step": 13450
+    },
+    {
+      "epoch": 49.85,
+      "grad_norm": 0.8182461261749268,
+      "learning_rate": 5.925925925925926e-06,
+      "loss": 0.7547,
+      "step": 13460
+    },
+    {
+      "epoch": 49.89,
+      "grad_norm": 0.843571126461029,
+      "learning_rate": 4.444444444444444e-06,
+      "loss": 0.7633,
+      "step": 13470
+    },
+    {
+      "epoch": 49.93,
+      "grad_norm": 0.7726001739501953,
+      "learning_rate": 2.962962962962963e-06,
+      "loss": 0.7705,
+      "step": 13480
+    },
+    {
+      "epoch": 49.96,
+      "grad_norm": 0.8144612312316895,
+      "learning_rate": 1.4814814814814815e-06,
+      "loss": 0.777,
+      "step": 13490
+    },
+    {
+      "epoch": 50.0,
+      "grad_norm": 1.684682846069336,
+      "learning_rate": 0.0,
+      "loss": 0.7724,
+      "step": 13500
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 13500,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 50,
+  "save_steps": 500,
+  "total_flos": 1.6910948327464694e+18,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}