diff --git "a/idefics2/checkpoint-9000/trainer_state.json" "b/idefics2/checkpoint-9000/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/idefics2/checkpoint-9000/trainer_state.json"
@@ -0,0 +1,6405 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.1212068666551067,
+  "eval_steps": 1000,
+  "global_step": 9000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0034680076296167853,
+      "grad_norm": 0.7528897523880005,
+      "learning_rate": 0.00019982,
+      "loss": 4.9086,
+      "step": 10
+    },
+    {
+      "epoch": 0.006936015259233571,
+      "grad_norm": 0.17007838189601898,
+      "learning_rate": 0.00019962000000000002,
+      "loss": 0.1029,
+      "step": 20
+    },
+    {
+      "epoch": 0.010404022888850355,
+      "grad_norm": 0.12162350863218307,
+      "learning_rate": 0.00019942,
+      "loss": 0.0687,
+      "step": 30
+    },
+    {
+      "epoch": 0.013872030518467141,
+      "grad_norm": 0.08915918320417404,
+      "learning_rate": 0.00019922,
+      "loss": 0.0636,
+      "step": 40
+    },
+    {
+      "epoch": 0.017340038148083926,
+      "grad_norm": 0.11356709897518158,
+      "learning_rate": 0.00019902,
+      "loss": 0.0623,
+      "step": 50
+    },
+    {
+      "epoch": 0.02080804577770071,
+      "grad_norm": 0.09400122612714767,
+      "learning_rate": 0.00019882,
+      "loss": 0.0561,
+      "step": 60
+    },
+    {
+      "epoch": 0.024276053407317495,
+      "grad_norm": 0.1102517619729042,
+      "learning_rate": 0.00019862000000000002,
+      "loss": 0.0603,
+      "step": 70
+    },
+    {
+      "epoch": 0.027744061036934282,
+      "grad_norm": 0.09301582723855972,
+      "learning_rate": 0.00019842000000000001,
+      "loss": 0.0559,
+      "step": 80
+    },
+    {
+      "epoch": 0.031212068666551067,
+      "grad_norm": 0.08400452882051468,
+      "learning_rate": 0.00019822,
+      "loss": 0.0645,
+      "step": 90
+    },
+    {
+      "epoch": 0.03468007629616785,
+      "grad_norm": 0.10444998741149902,
+      "learning_rate": 0.00019802,
+      "loss": 0.0644,
+      "step": 100
+    },
+    {
+      "epoch": 0.038148083925784636,
+      "grad_norm": 0.06524047255516052,
+      "learning_rate": 0.00019782,
+      "loss": 0.0599,
+      "step": 110
+    },
+    {
+      "epoch": 0.04161609155540142,
+      "grad_norm": 0.0640910267829895,
+      "learning_rate": 0.00019762,
+      "loss": 0.0542,
+      "step": 120
+    },
+    {
+      "epoch": 0.045084099185018205,
+      "grad_norm": 0.07332012802362442,
+      "learning_rate": 0.00019742000000000002,
+      "loss": 0.0567,
+      "step": 130
+    },
+    {
+      "epoch": 0.04855210681463499,
+      "grad_norm": 4.02905797958374,
+      "learning_rate": 0.00019722,
+      "loss": 0.0592,
+      "step": 140
+    },
+    {
+      "epoch": 0.05202011444425178,
+      "grad_norm": 0.09805350750684738,
+      "learning_rate": 0.00019702,
+      "loss": 0.0657,
+      "step": 150
+    },
+    {
+      "epoch": 0.055488122073868565,
+      "grad_norm": 0.056836508214473724,
+      "learning_rate": 0.00019682,
+      "loss": 0.0523,
+      "step": 160
+    },
+    {
+      "epoch": 0.05895612970348535,
+      "grad_norm": 0.08670804649591446,
+      "learning_rate": 0.00019662,
+      "loss": 0.0585,
+      "step": 170
+    },
+    {
+      "epoch": 0.062424137333102134,
+      "grad_norm": 0.08405158668756485,
+      "learning_rate": 0.00019642,
+      "loss": 0.0537,
+      "step": 180
+    },
+    {
+      "epoch": 0.06589214496271892,
+      "grad_norm": 0.0825357437133789,
+      "learning_rate": 0.00019622000000000002,
+      "loss": 0.055,
+      "step": 190
+    },
+    {
+      "epoch": 0.0693601525923357,
+      "grad_norm": 0.06040720269083977,
+      "learning_rate": 0.00019602,
+      "loss": 0.0554,
+      "step": 200
+    },
+    {
+      "epoch": 0.07282816022195249,
+      "grad_norm": 0.08198798447847366,
+      "learning_rate": 0.00019582,
+      "loss": 0.0599,
+      "step": 210
+    },
+    {
+      "epoch": 0.07629616785156927,
+      "grad_norm": 0.05527138710021973,
+      "learning_rate": 0.00019562,
+      "loss": 0.0533,
+      "step": 220
+    },
+    {
+      "epoch": 0.07976417548118606,
+      "grad_norm": 0.07315631955862045,
+      "learning_rate": 0.00019542,
+      "loss": 0.0528,
+      "step": 230
+    },
+    {
+      "epoch": 0.08323218311080284,
+      "grad_norm": 0.08420062810182571,
+      "learning_rate": 0.00019522,
+      "loss": 0.0518,
+      "step": 240
+    },
+    {
+      "epoch": 0.08670019074041962,
+      "grad_norm": 0.0529639795422554,
+      "learning_rate": 0.00019502,
+      "loss": 0.0568,
+      "step": 250
+    },
+    {
+      "epoch": 0.09016819837003641,
+      "grad_norm": 0.09339221566915512,
+      "learning_rate": 0.00019482,
+      "loss": 0.0557,
+      "step": 260
+    },
+    {
+      "epoch": 0.0936362059996532,
+      "grad_norm": 0.06453025341033936,
+      "learning_rate": 0.00019462,
+      "loss": 0.062,
+      "step": 270
+    },
+    {
+      "epoch": 0.09710421362926998,
+      "grad_norm": 0.07090363651514053,
+      "learning_rate": 0.00019442,
+      "loss": 0.0609,
+      "step": 280
+    },
+    {
+      "epoch": 0.10057222125888678,
+      "grad_norm": 0.1122497022151947,
+      "learning_rate": 0.00019422,
+      "loss": 0.0541,
+      "step": 290
+    },
+    {
+      "epoch": 0.10404022888850356,
+      "grad_norm": 0.09782398492097855,
+      "learning_rate": 0.00019402,
+      "loss": 0.0553,
+      "step": 300
+    },
+    {
+      "epoch": 0.10750823651812035,
+      "grad_norm": 0.06216060370206833,
+      "learning_rate": 0.00019382,
+      "loss": 0.0483,
+      "step": 310
+    },
+    {
+      "epoch": 0.11097624414773713,
+      "grad_norm": 0.08817891031503677,
+      "learning_rate": 0.00019362,
+      "loss": 0.058,
+      "step": 320
+    },
+    {
+      "epoch": 0.11444425177735391,
+      "grad_norm": 0.07251620292663574,
+      "learning_rate": 0.00019342,
+      "loss": 0.0477,
+      "step": 330
+    },
+    {
+      "epoch": 0.1179122594069707,
+      "grad_norm": 0.059537626802921295,
+      "learning_rate": 0.00019322,
+      "loss": 0.0559,
+      "step": 340
+    },
+    {
+      "epoch": 0.12138026703658748,
+      "grad_norm": 0.06329932063817978,
+      "learning_rate": 0.00019302,
+      "loss": 0.0557,
+      "step": 350
+    },
+    {
+      "epoch": 0.12484827466620427,
+      "grad_norm": 0.06601905822753906,
+      "learning_rate": 0.00019282000000000001,
+      "loss": 0.0575,
+      "step": 360
+    },
+    {
+      "epoch": 0.12831628229582104,
+      "grad_norm": 0.0639985054731369,
+      "learning_rate": 0.00019262,
+      "loss": 0.0519,
+      "step": 370
+    },
+    {
+      "epoch": 0.13178428992543784,
+      "grad_norm": 0.08187698572874069,
+      "learning_rate": 0.00019242,
+      "loss": 0.0545,
+      "step": 380
+    },
+    {
+      "epoch": 0.13525229755505463,
+      "grad_norm": 0.06790990382432938,
+      "learning_rate": 0.00019222,
+      "loss": 0.0474,
+      "step": 390
+    },
+    {
+      "epoch": 0.1387203051846714,
+      "grad_norm": 0.04906002804636955,
+      "learning_rate": 0.00019202,
+      "loss": 0.0579,
+      "step": 400
+    },
+    {
+      "epoch": 0.1421883128142882,
+      "grad_norm": 0.06125594303011894,
+      "learning_rate": 0.00019182,
+      "loss": 0.0421,
+      "step": 410
+    },
+    {
+      "epoch": 0.14565632044390497,
+      "grad_norm": 0.06252908706665039,
+      "learning_rate": 0.00019162,
+      "loss": 0.0577,
+      "step": 420
+    },
+    {
+      "epoch": 0.14912432807352177,
+      "grad_norm": 0.09376012533903122,
+      "learning_rate": 0.00019142,
+      "loss": 0.0546,
+      "step": 430
+    },
+    {
+      "epoch": 0.15259233570313854,
+      "grad_norm": 0.05025137588381767,
+      "learning_rate": 0.00019122,
+      "loss": 0.0505,
+      "step": 440
+    },
+    {
+      "epoch": 0.15606034333275534,
+      "grad_norm": 0.07034559547901154,
+      "learning_rate": 0.00019102,
+      "loss": 0.0495,
+      "step": 450
+    },
+    {
+      "epoch": 0.1595283509623721,
+      "grad_norm": 0.07475865632295609,
+      "learning_rate": 0.00019082,
+      "loss": 0.0533,
+      "step": 460
+    },
+    {
+      "epoch": 0.1629963585919889,
+      "grad_norm": 0.07459491491317749,
+      "learning_rate": 0.00019062,
+      "loss": 0.0551,
+      "step": 470
+    },
+    {
+      "epoch": 0.16646436622160568,
+      "grad_norm": 0.052578963339328766,
+      "learning_rate": 0.00019042,
+      "loss": 0.0564,
+      "step": 480
+    },
+    {
+      "epoch": 0.16993237385122248,
+      "grad_norm": 0.05785336345434189,
+      "learning_rate": 0.00019022,
+      "loss": 0.0549,
+      "step": 490
+    },
+    {
+      "epoch": 0.17340038148083925,
+      "grad_norm": 0.06621215492486954,
+      "learning_rate": 0.00019002,
+      "loss": 0.0528,
+      "step": 500
+    },
+    {
+      "epoch": 0.17686838911045605,
+      "grad_norm": 0.07456778734922409,
+      "learning_rate": 0.00018982000000000002,
+      "loss": 0.0496,
+      "step": 510
+    },
+    {
+      "epoch": 0.18033639674007282,
+      "grad_norm": 0.06371001899242401,
+      "learning_rate": 0.00018962000000000002,
+      "loss": 0.0509,
+      "step": 520
+    },
+    {
+      "epoch": 0.18380440436968962,
+      "grad_norm": 0.052791863679885864,
+      "learning_rate": 0.00018942,
+      "loss": 0.0504,
+      "step": 530
+    },
+    {
+      "epoch": 0.1872724119993064,
+      "grad_norm": 0.06491260975599289,
+      "learning_rate": 0.00018922,
+      "loss": 0.0527,
+      "step": 540
+    },
+    {
+      "epoch": 0.19074041962892319,
+      "grad_norm": 0.07835149765014648,
+      "learning_rate": 0.00018902000000000003,
+      "loss": 0.0564,
+      "step": 550
+    },
+    {
+      "epoch": 0.19420842725853996,
+      "grad_norm": 0.05977100506424904,
+      "learning_rate": 0.00018882000000000003,
+      "loss": 0.0477,
+      "step": 560
+    },
+    {
+      "epoch": 0.19767643488815675,
+      "grad_norm": 0.07109620422124863,
+      "learning_rate": 0.00018862000000000002,
+      "loss": 0.0537,
+      "step": 570
+    },
+    {
+      "epoch": 0.20114444251777355,
+      "grad_norm": 0.10347943007946014,
+      "learning_rate": 0.00018842000000000002,
+      "loss": 0.0542,
+      "step": 580
+    },
+    {
+      "epoch": 0.20461245014739032,
+      "grad_norm": 0.06503281742334366,
+      "learning_rate": 0.00018822,
+      "loss": 0.0484,
+      "step": 590
+    },
+    {
+      "epoch": 0.20808045777700712,
+      "grad_norm": 0.07903438806533813,
+      "learning_rate": 0.00018802,
+      "loss": 0.0597,
+      "step": 600
+    },
+    {
+      "epoch": 0.2115484654066239,
+      "grad_norm": 0.0713895708322525,
+      "learning_rate": 0.00018782000000000003,
+      "loss": 0.0498,
+      "step": 610
+    },
+    {
+      "epoch": 0.2150164730362407,
+      "grad_norm": 0.061313629150390625,
+      "learning_rate": 0.00018762000000000002,
+      "loss": 0.053,
+      "step": 620
+    },
+    {
+      "epoch": 0.21848448066585746,
+      "grad_norm": 0.07045572996139526,
+      "learning_rate": 0.00018742000000000002,
+      "loss": 0.0541,
+      "step": 630
+    },
+    {
+      "epoch": 0.22195248829547426,
+      "grad_norm": 0.07118247449398041,
+      "learning_rate": 0.00018722,
+      "loss": 0.0586,
+      "step": 640
+    },
+    {
+      "epoch": 0.22542049592509103,
+      "grad_norm": 0.05364071577787399,
+      "learning_rate": 0.00018702,
+      "loss": 0.0497,
+      "step": 650
+    },
+    {
+      "epoch": 0.22888850355470783,
+      "grad_norm": 0.07208040356636047,
+      "learning_rate": 0.00018682000000000003,
+      "loss": 0.0559,
+      "step": 660
+    },
+    {
+      "epoch": 0.2323565111843246,
+      "grad_norm": 0.07200731337070465,
+      "learning_rate": 0.00018662000000000003,
+      "loss": 0.048,
+      "step": 670
+    },
+    {
+      "epoch": 0.2358245188139414,
+      "grad_norm": 0.05730220302939415,
+      "learning_rate": 0.00018642000000000002,
+      "loss": 0.0495,
+      "step": 680
+    },
+    {
+      "epoch": 0.23929252644355817,
+      "grad_norm": 0.06378819793462753,
+      "learning_rate": 0.00018622000000000002,
+      "loss": 0.0552,
+      "step": 690
+    },
+    {
+      "epoch": 0.24276053407317497,
+      "grad_norm": 0.05866115912795067,
+      "learning_rate": 0.00018602,
+      "loss": 0.0541,
+      "step": 700
+    },
+    {
+      "epoch": 0.24622854170279174,
+      "grad_norm": 0.07053161412477493,
+      "learning_rate": 0.00018582,
+      "loss": 0.0533,
+      "step": 710
+    },
+    {
+      "epoch": 0.24969654933240854,
+      "grad_norm": 0.07450433820486069,
+      "learning_rate": 0.00018562000000000003,
+      "loss": 0.0527,
+      "step": 720
+    },
+    {
+      "epoch": 0.25316455696202533,
+      "grad_norm": 0.07172481715679169,
+      "learning_rate": 0.00018542000000000002,
+      "loss": 0.0583,
+      "step": 730
+    },
+    {
+      "epoch": 0.2566325645916421,
+      "grad_norm": 0.0536239892244339,
+      "learning_rate": 0.00018522000000000002,
+      "loss": 0.0486,
+      "step": 740
+    },
+    {
+      "epoch": 0.2601005722212589,
+      "grad_norm": 0.06321065127849579,
+      "learning_rate": 0.00018502000000000001,
+      "loss": 0.0412,
+      "step": 750
+    },
+    {
+      "epoch": 0.2635685798508757,
+      "grad_norm": 0.056946441531181335,
+      "learning_rate": 0.00018482,
+      "loss": 0.051,
+      "step": 760
+    },
+    {
+      "epoch": 0.26703658748049247,
+      "grad_norm": 0.05819573253393173,
+      "learning_rate": 0.00018462,
+      "loss": 0.0517,
+      "step": 770
+    },
+    {
+      "epoch": 0.27050459511010927,
+      "grad_norm": 0.05857665091753006,
+      "learning_rate": 0.00018442000000000003,
+      "loss": 0.0529,
+      "step": 780
+    },
+    {
+      "epoch": 0.273972602739726,
+      "grad_norm": 0.06014329940080643,
+      "learning_rate": 0.00018422000000000002,
+      "loss": 0.0544,
+      "step": 790
+    },
+    {
+      "epoch": 0.2774406103693428,
+      "grad_norm": 0.09187959134578705,
+      "learning_rate": 0.00018402000000000002,
+      "loss": 0.0439,
+      "step": 800
+    },
+    {
+      "epoch": 0.2809086179989596,
+      "grad_norm": 0.056131429970264435,
+      "learning_rate": 0.00018382,
+      "loss": 0.0389,
+      "step": 810
+    },
+    {
+      "epoch": 0.2843766256285764,
+      "grad_norm": 0.04884343966841698,
+      "learning_rate": 0.00018362,
+      "loss": 0.0475,
+      "step": 820
+    },
+    {
+      "epoch": 0.28784463325819315,
+      "grad_norm": 0.09032488614320755,
+      "learning_rate": 0.00018342,
+      "loss": 0.056,
+      "step": 830
+    },
+    {
+      "epoch": 0.29131264088780995,
+      "grad_norm": 0.09926522523164749,
+      "learning_rate": 0.00018322000000000002,
+      "loss": 0.0515,
+      "step": 840
+    },
+    {
+      "epoch": 0.29478064851742675,
+      "grad_norm": 0.04553750529885292,
+      "learning_rate": 0.00018302000000000002,
+      "loss": 0.0505,
+      "step": 850
+    },
+    {
+      "epoch": 0.29824865614704354,
+      "grad_norm": 0.058485984802246094,
+      "learning_rate": 0.00018282000000000001,
+      "loss": 0.0453,
+      "step": 860
+    },
+    {
+      "epoch": 0.3017166637766603,
+      "grad_norm": 0.052825070917606354,
+      "learning_rate": 0.00018262,
+      "loss": 0.053,
+      "step": 870
+    },
+    {
+      "epoch": 0.3051846714062771,
+      "grad_norm": 0.07626510411500931,
+      "learning_rate": 0.00018242,
+      "loss": 0.0474,
+      "step": 880
+    },
+    {
+      "epoch": 0.3086526790358939,
+      "grad_norm": 0.07244163751602173,
+      "learning_rate": 0.00018222,
+      "loss": 0.0456,
+      "step": 890
+    },
+    {
+      "epoch": 0.3121206866655107,
+      "grad_norm": 0.09651289880275726,
+      "learning_rate": 0.00018202000000000002,
+      "loss": 0.0499,
+      "step": 900
+    },
+    {
+      "epoch": 0.3155886942951274,
+      "grad_norm": 0.0737752839922905,
+      "learning_rate": 0.00018182000000000002,
+      "loss": 0.0517,
+      "step": 910
+    },
+    {
+      "epoch": 0.3190567019247442,
+      "grad_norm": 0.06631263345479965,
+      "learning_rate": 0.00018162,
+      "loss": 0.0499,
+      "step": 920
+    },
+    {
+      "epoch": 0.322524709554361,
+      "grad_norm": 0.10696552693843842,
+      "learning_rate": 0.00018142,
+      "loss": 0.0524,
+      "step": 930
+    },
+    {
+      "epoch": 0.3259927171839778,
+      "grad_norm": 0.06753025203943253,
+      "learning_rate": 0.00018122,
+      "loss": 0.0458,
+      "step": 940
+    },
+    {
+      "epoch": 0.32946072481359456,
+      "grad_norm": 0.05970798432826996,
+      "learning_rate": 0.00018102000000000003,
+      "loss": 0.0471,
+      "step": 950
+    },
+    {
+      "epoch": 0.33292873244321136,
+      "grad_norm": 0.05324438214302063,
+      "learning_rate": 0.00018082000000000002,
+      "loss": 0.0468,
+      "step": 960
+    },
+    {
+      "epoch": 0.33639674007282816,
+      "grad_norm": 0.0749637559056282,
+      "learning_rate": 0.00018062000000000002,
+      "loss": 0.0507,
+      "step": 970
+    },
+    {
+      "epoch": 0.33986474770244496,
+      "grad_norm": 0.07317759841680527,
+      "learning_rate": 0.00018042,
+      "loss": 0.0482,
+      "step": 980
+    },
+    {
+      "epoch": 0.34333275533206176,
+      "grad_norm": 0.11602938175201416,
+      "learning_rate": 0.00018022,
+      "loss": 0.0558,
+      "step": 990
+    },
+    {
+      "epoch": 0.3468007629616785,
+      "grad_norm": 0.07047244161367416,
+      "learning_rate": 0.00018002,
+      "loss": 0.0547,
+      "step": 1000
+    },
+    {
+      "epoch": 0.3468007629616785,
+      "eval_loss": 0.062305010855197906,
+      "eval_runtime": 714.4647,
+      "eval_samples_per_second": 12.852,
+      "eval_steps_per_second": 1.607,
+      "step": 1000
+    },
+    {
+      "epoch": 0.3502687705912953,
+      "grad_norm": 0.07715447247028351,
+      "learning_rate": 0.00017982000000000002,
+      "loss": 0.0445,
+      "step": 1010
+    },
+    {
+      "epoch": 0.3537367782209121,
+      "grad_norm": 0.08189492672681808,
+      "learning_rate": 0.00017962000000000002,
+      "loss": 0.0529,
+      "step": 1020
+    },
+    {
+      "epoch": 0.3572047858505289,
+      "grad_norm": 0.06011577695608139,
+      "learning_rate": 0.00017942,
+      "loss": 0.0469,
+      "step": 1030
+    },
+    {
+      "epoch": 0.36067279348014564,
+      "grad_norm": 0.06397314369678497,
+      "learning_rate": 0.00017922,
+      "loss": 0.0542,
+      "step": 1040
+    },
+    {
+      "epoch": 0.36414080110976244,
+      "grad_norm": 0.06121763586997986,
+      "learning_rate": 0.00017902,
+      "loss": 0.0474,
+      "step": 1050
+    },
+    {
+      "epoch": 0.36760880873937923,
+      "grad_norm": 0.0765228345990181,
+      "learning_rate": 0.00017882,
+      "loss": 0.0476,
+      "step": 1060
+    },
+    {
+      "epoch": 0.37107681636899603,
+      "grad_norm": 0.0813635066151619,
+      "learning_rate": 0.00017862000000000002,
+      "loss": 0.0488,
+      "step": 1070
+    },
+    {
+      "epoch": 0.3745448239986128,
+      "grad_norm": 0.06827688962221146,
+      "learning_rate": 0.00017842000000000002,
+      "loss": 0.0431,
+      "step": 1080
+    },
+    {
+      "epoch": 0.3780128316282296,
+      "grad_norm": 0.06176091730594635,
+      "learning_rate": 0.00017822,
+      "loss": 0.0513,
+      "step": 1090
+    },
+    {
+      "epoch": 0.38148083925784637,
+      "grad_norm": 0.07062922418117523,
+      "learning_rate": 0.00017802,
+      "loss": 0.0499,
+      "step": 1100
+    },
+    {
+      "epoch": 0.38494884688746317,
+      "grad_norm": 0.059431031346321106,
+      "learning_rate": 0.00017782,
+      "loss": 0.0556,
+      "step": 1110
+    },
+    {
+      "epoch": 0.3884168545170799,
+      "grad_norm": 0.06391894072294235,
+      "learning_rate": 0.00017762,
+      "loss": 0.0535,
+      "step": 1120
+    },
+    {
+      "epoch": 0.3918848621466967,
+      "grad_norm": 0.08487355709075928,
+      "learning_rate": 0.00017742000000000002,
+      "loss": 0.0509,
+      "step": 1130
+    },
+    {
+      "epoch": 0.3953528697763135,
+      "grad_norm": 0.06291911005973816,
+      "learning_rate": 0.00017722000000000001,
+      "loss": 0.0462,
+      "step": 1140
+    },
+    {
+      "epoch": 0.3988208774059303,
+      "grad_norm": 0.06936580687761307,
+      "learning_rate": 0.00017702,
+      "loss": 0.0465,
+      "step": 1150
+    },
+    {
+      "epoch": 0.4022888850355471,
+      "grad_norm": 0.06751543283462524,
+      "learning_rate": 0.00017682,
+      "loss": 0.0553,
+      "step": 1160
+    },
+    {
+      "epoch": 0.40575689266516385,
+      "grad_norm": 0.08026771247386932,
+      "learning_rate": 0.00017662,
+      "loss": 0.0503,
+      "step": 1170
+    },
+    {
+      "epoch": 0.40922490029478065,
+      "grad_norm": 0.05316636711359024,
+      "learning_rate": 0.00017642,
+      "loss": 0.0426,
+      "step": 1180
+    },
+    {
+      "epoch": 0.41269290792439745,
+      "grad_norm": 0.15491995215415955,
+      "learning_rate": 0.00017622000000000002,
+      "loss": 0.0481,
+      "step": 1190
+    },
+    {
+      "epoch": 0.41616091555401424,
+      "grad_norm": 0.059617578983306885,
+      "learning_rate": 0.00017602,
+      "loss": 0.0443,
+      "step": 1200
+    },
+    {
+      "epoch": 0.419628923183631,
+      "grad_norm": 0.08931437879800797,
+      "learning_rate": 0.00017582,
+      "loss": 0.0462,
+      "step": 1210
+    },
+    {
+      "epoch": 0.4230969308132478,
+      "grad_norm": 1.4069609642028809,
+      "learning_rate": 0.00017562,
+      "loss": 0.0577,
+      "step": 1220
+    },
+    {
+      "epoch": 0.4265649384428646,
+      "grad_norm": 7.056313514709473,
+      "learning_rate": 0.00017542,
+      "loss": 0.3501,
+      "step": 1230
+    },
+    {
+      "epoch": 0.4300329460724814,
+      "grad_norm": 0.8197808265686035,
+      "learning_rate": 0.00017522000000000002,
+      "loss": 0.1137,
+      "step": 1240
+    },
+    {
+      "epoch": 0.4335009537020981,
+      "grad_norm": 0.4104197025299072,
+      "learning_rate": 0.00017502000000000001,
+      "loss": 0.1134,
+      "step": 1250
+    },
+    {
+      "epoch": 0.4369689613317149,
+      "grad_norm": 1.3801881074905396,
+      "learning_rate": 0.00017482,
+      "loss": 0.0869,
+      "step": 1260
+    },
+    {
+      "epoch": 0.4404369689613317,
+      "grad_norm": 3.6437034606933594,
+      "learning_rate": 0.00017462,
+      "loss": 0.0544,
+      "step": 1270
+    },
+    {
+      "epoch": 0.4439049765909485,
+      "grad_norm": 1.1371792554855347,
+      "learning_rate": 0.00017442,
+      "loss": 0.0929,
+      "step": 1280
+    },
+    {
+      "epoch": 0.44737298422056526,
+      "grad_norm": 2.9385204315185547,
+      "learning_rate": 0.00017422,
+      "loss": 0.1166,
+      "step": 1290
+    },
+    {
+      "epoch": 0.45084099185018206,
+      "grad_norm": 0.8243001699447632,
+      "learning_rate": 0.00017402000000000002,
+      "loss": 0.0907,
+      "step": 1300
+    },
+    {
+      "epoch": 0.45430899947979886,
+      "grad_norm": 1.3375756740570068,
+      "learning_rate": 0.00017382,
+      "loss": 0.0641,
+      "step": 1310
+    },
+    {
+      "epoch": 0.45777700710941566,
+      "grad_norm": 0.7172356843948364,
+      "learning_rate": 0.00017362,
+      "loss": 0.0575,
+      "step": 1320
+    },
+    {
+      "epoch": 0.4612450147390324,
+      "grad_norm": 0.18255668878555298,
+      "learning_rate": 0.00017342,
+      "loss": 0.0628,
+      "step": 1330
+    },
+    {
+      "epoch": 0.4647130223686492,
+      "grad_norm": 0.11856569349765778,
+      "learning_rate": 0.00017322,
+      "loss": 0.1111,
+      "step": 1340
+    },
+    {
+      "epoch": 0.468181029998266,
+      "grad_norm": 0.16621063649654388,
+      "learning_rate": 0.00017302,
+      "loss": 0.0446,
+      "step": 1350
+    },
+    {
+      "epoch": 0.4716490376278828,
+      "grad_norm": 0.10158076882362366,
+      "learning_rate": 0.00017282000000000002,
+      "loss": 0.0532,
+      "step": 1360
+    },
+    {
+      "epoch": 0.4751170452574996,
+      "grad_norm": 0.08783379942178726,
+      "learning_rate": 0.00017262,
+      "loss": 0.0491,
+      "step": 1370
+    },
+    {
+      "epoch": 0.47858505288711634,
+      "grad_norm": 0.08198387920856476,
+      "learning_rate": 0.00017242,
+      "loss": 0.0485,
+      "step": 1380
+    },
+    {
+      "epoch": 0.48205306051673313,
+      "grad_norm": 0.09546195715665817,
+      "learning_rate": 0.00017222,
+      "loss": 0.0486,
+      "step": 1390
+    },
+    {
+      "epoch": 0.48552106814634993,
+      "grad_norm": 0.13258413970470428,
+      "learning_rate": 0.00017202,
+      "loss": 0.0544,
+      "step": 1400
+    },
+    {
+      "epoch": 0.48898907577596673,
+      "grad_norm": 0.09508573263883591,
+      "learning_rate": 0.00017182,
+      "loss": 0.0486,
+      "step": 1410
+    },
+    {
+      "epoch": 0.4924570834055835,
+      "grad_norm": 0.07258226722478867,
+      "learning_rate": 0.00017162000000000001,
+      "loss": 0.0448,
+      "step": 1420
+    },
+    {
+      "epoch": 0.49592509103520027,
+      "grad_norm": 0.05399150773882866,
+      "learning_rate": 0.00017142,
+      "loss": 0.0436,
+      "step": 1430
+    },
+    {
+      "epoch": 0.49939309866481707,
+      "grad_norm": 0.06922592222690582,
+      "learning_rate": 0.00017122,
+      "loss": 0.0451,
+      "step": 1440
+    },
+    {
+      "epoch": 0.5028611062944338,
+      "grad_norm": 0.05059856176376343,
+      "learning_rate": 0.00017102,
+      "loss": 0.0413,
+      "step": 1450
+    },
+    {
+      "epoch": 0.5063291139240507,
+      "grad_norm": 0.06339547783136368,
+      "learning_rate": 0.00017082,
+      "loss": 0.0495,
+      "step": 1460
+    },
+    {
+      "epoch": 0.5097971215536674,
+      "grad_norm": 0.0521874763071537,
+      "learning_rate": 0.00017062,
+      "loss": 0.0489,
+      "step": 1470
+    },
+    {
+      "epoch": 0.5132651291832842,
+      "grad_norm": 0.15334908664226532,
+      "learning_rate": 0.00017042,
+      "loss": 0.04,
+      "step": 1480
+    },
+    {
+      "epoch": 0.516733136812901,
+      "grad_norm": 0.07494404166936874,
+      "learning_rate": 0.00017022,
+      "loss": 0.0537,
+      "step": 1490
+    },
+    {
+      "epoch": 0.5202011444425177,
+      "grad_norm": 0.052238237112760544,
+      "learning_rate": 0.00017002,
+      "loss": 0.0492,
+      "step": 1500
+    },
+    {
+      "epoch": 0.5236691520721346,
+      "grad_norm": 0.06761351227760315,
+      "learning_rate": 0.00016982,
+      "loss": 0.051,
+      "step": 1510
+    },
+    {
+      "epoch": 0.5271371597017513,
+      "grad_norm": 0.07580805569887161,
+      "learning_rate": 0.00016962,
+      "loss": 0.0498,
+      "step": 1520
+    },
+    {
+      "epoch": 0.5306051673313681,
+      "grad_norm": 0.07397795468568802,
+      "learning_rate": 0.00016942000000000001,
+      "loss": 0.0471,
+      "step": 1530
+    },
+    {
+      "epoch": 0.5340731749609849,
+      "grad_norm": 0.04779529199004173,
+      "learning_rate": 0.00016922,
+      "loss": 0.0465,
+      "step": 1540
+    },
+    {
+      "epoch": 0.5375411825906017,
+      "grad_norm": 0.08508727699518204,
+      "learning_rate": 0.00016902,
+      "loss": 0.046,
+      "step": 1550
+    },
+    {
+      "epoch": 0.5410091902202185,
+      "grad_norm": 0.06656762957572937,
+      "learning_rate": 0.00016882,
+      "loss": 0.0433,
+      "step": 1560
+    },
+    {
+      "epoch": 0.5444771978498353,
+      "grad_norm": 0.05622195079922676,
+      "learning_rate": 0.00016862,
+      "loss": 0.0455,
+      "step": 1570
+    },
+    {
+      "epoch": 0.547945205479452,
+      "grad_norm": 0.06422954052686691,
+      "learning_rate": 0.00016842,
+      "loss": 0.0428,
+      "step": 1580
+    },
+    {
+      "epoch": 0.5514132131090689,
+      "grad_norm": 0.08891351521015167,
+      "learning_rate": 0.00016822,
+      "loss": 0.0495,
+      "step": 1590
+    },
+    {
+      "epoch": 0.5548812207386856,
+      "grad_norm": 0.08472294360399246,
+      "learning_rate": 0.00016802,
+      "loss": 0.0525,
+      "step": 1600
+    },
+    {
+      "epoch": 0.5583492283683024,
+      "grad_norm": 0.08518495410680771,
+      "learning_rate": 0.00016782,
+      "loss": 0.0518,
+      "step": 1610
+    },
+    {
+      "epoch": 0.5618172359979192,
+      "grad_norm": 0.1859533041715622,
+      "learning_rate": 0.00016762,
+      "loss": 0.0478,
+      "step": 1620
+    },
+    {
+      "epoch": 0.565285243627536,
+      "grad_norm": 0.05245356634259224,
+      "learning_rate": 0.00016742,
+      "loss": 0.0418,
+      "step": 1630
+    },
+    {
+      "epoch": 0.5687532512571528,
+      "grad_norm": 0.05982668697834015,
+      "learning_rate": 0.00016722,
+      "loss": 0.0458,
+      "step": 1640
+    },
+    {
+      "epoch": 0.5722212588867696,
+      "grad_norm": 0.0750059187412262,
+      "learning_rate": 0.00016702,
+      "loss": 0.0489,
+      "step": 1650
+    },
+    {
+      "epoch": 0.5756892665163863,
+      "grad_norm": 0.061683133244514465,
+      "learning_rate": 0.00016682,
+      "loss": 0.0461,
+      "step": 1660
+    },
+    {
+      "epoch": 0.5791572741460032,
+      "grad_norm": 0.06833604723215103,
+      "learning_rate": 0.00016662,
+      "loss": 0.0466,
+      "step": 1670
+    },
+    {
+      "epoch": 0.5826252817756199,
+      "grad_norm": 0.05507722124457359,
+      "learning_rate": 0.00016642,
+      "loss": 0.0416,
+      "step": 1680
+    },
+    {
+      "epoch": 0.5860932894052366,
+      "grad_norm": 0.09453442692756653,
+      "learning_rate": 0.00016622,
+      "loss": 0.0418,
+      "step": 1690
+    },
+    {
+      "epoch": 0.5895612970348535,
+      "grad_norm": 0.0574457123875618,
+      "learning_rate": 0.00016601999999999999,
+      "loss": 0.0495,
+      "step": 1700
+    },
+    {
+      "epoch": 0.5930293046644702,
+      "grad_norm": 0.0829281136393547,
+      "learning_rate": 0.00016582,
+      "loss": 0.0435,
+      "step": 1710
+    },
+    {
+      "epoch": 0.5964973122940871,
+      "grad_norm": 0.0569952018558979,
+      "learning_rate": 0.00016562,
+      "loss": 0.0476,
+      "step": 1720
+    },
+    {
+      "epoch": 0.5999653199237038,
+      "grad_norm": 0.09291055798530579,
+      "learning_rate": 0.00016542,
+      "loss": 0.0446,
+      "step": 1730
+    },
+    {
+      "epoch": 0.6034333275533206,
+      "grad_norm": 0.03767919912934303,
+      "learning_rate": 0.00016522,
+      "loss": 0.0426,
+      "step": 1740
+    },
+    {
+      "epoch": 0.6069013351829374,
+      "grad_norm": 0.07064680010080338,
+      "learning_rate": 0.00016502,
+      "loss": 0.0491,
+      "step": 1750
+    },
+    {
+      "epoch": 0.6103693428125542,
+      "grad_norm": 0.05687623471021652,
+      "learning_rate": 0.00016482,
+      "loss": 0.0454,
+      "step": 1760
+    },
+    {
+      "epoch": 0.613837350442171,
+      "grad_norm": 0.04933289438486099,
+      "learning_rate": 0.00016462,
+      "loss": 0.0385,
+      "step": 1770
+    },
+    {
+      "epoch": 0.6173053580717878,
+      "grad_norm": 0.0763295516371727,
+      "learning_rate": 0.00016442000000000003,
+      "loss": 0.0448,
+      "step": 1780
+    },
+    {
+      "epoch": 0.6207733657014045,
+      "grad_norm": 0.04926356300711632,
+      "learning_rate": 0.00016422000000000002,
+      "loss": 0.0484,
+      "step": 1790
+    },
+    {
+      "epoch": 0.6242413733310214,
+      "grad_norm": 0.07130167633295059,
+      "learning_rate": 0.00016402000000000002,
+      "loss": 0.0403,
+      "step": 1800
+    },
+    {
+      "epoch": 0.6277093809606381,
+      "grad_norm": 0.06025327742099762,
+      "learning_rate": 0.00016382000000000001,
+      "loss": 0.0483,
+      "step": 1810
+    },
+    {
+      "epoch": 0.6311773885902549,
+      "grad_norm": 0.06522911787033081,
+      "learning_rate": 0.00016362,
+      "loss": 0.0433,
+      "step": 1820
+    },
+    {
+      "epoch": 0.6346453962198717,
+      "grad_norm": 0.09565310180187225,
+      "learning_rate": 0.00016342,
+      "loss": 0.0539,
+      "step": 1830
+    },
+    {
+      "epoch": 0.6381134038494884,
+      "grad_norm": 0.08908990025520325,
+      "learning_rate": 0.00016322000000000003,
+      "loss": 0.0479,
+      "step": 1840
+    },
+    {
+      "epoch": 0.6415814114791053,
+      "grad_norm": 0.05405285581946373,
+      "learning_rate": 0.00016302000000000002,
+      "loss": 0.0416,
+      "step": 1850
+    },
+    {
+      "epoch": 0.645049419108722,
+      "grad_norm": 0.0656275674700737,
+      "learning_rate": 0.00016282000000000002,
+      "loss": 0.0455,
+      "step": 1860
+    },
+    {
+      "epoch": 0.6485174267383388,
+      "grad_norm": 0.30592814087867737,
+      "learning_rate": 0.00016262,
+      "loss": 0.0471,
+      "step": 1870
+    },
+    {
+      "epoch": 0.6519854343679556,
+      "grad_norm": 0.06433047354221344,
+      "learning_rate": 0.00016242,
+      "loss": 0.045,
+      "step": 1880
+    },
+    {
+      "epoch": 0.6554534419975724,
+      "grad_norm": 0.05807631090283394,
+      "learning_rate": 0.00016222000000000003,
+      "loss": 0.0502,
+      "step": 1890
+    },
+    {
+      "epoch": 0.6589214496271891,
+      "grad_norm": 0.09822454303503036,
+      "learning_rate": 0.00016202000000000002,
+      "loss": 0.0465,
+      "step": 1900
+    },
+    {
+      "epoch": 0.662389457256806,
+      "grad_norm": 0.06036192551255226,
+      "learning_rate": 0.00016182000000000002,
+      "loss": 0.0455,
+      "step": 1910
+    },
+    {
+      "epoch": 0.6658574648864227,
+      "grad_norm": 0.05637621134519577,
+      "learning_rate": 0.00016162000000000001,
+      "loss": 0.0459,
+      "step": 1920
+    },
+    {
+      "epoch": 0.6693254725160396,
+      "grad_norm": 0.062302861362695694,
+      "learning_rate": 0.00016142,
+      "loss": 0.0461,
+      "step": 1930
+    },
+    {
+      "epoch": 0.6727934801456563,
+      "grad_norm": 0.05843142420053482,
+      "learning_rate": 0.00016122,
+      "loss": 0.0455,
+      "step": 1940
+    },
+    {
+      "epoch": 0.6762614877752731,
+      "grad_norm": 0.052565112709999084,
+      "learning_rate": 0.00016102000000000003,
+      "loss": 0.0432,
+      "step": 1950
+    },
+    {
+      "epoch": 0.6797294954048899,
+      "grad_norm": 0.059093691408634186,
+      "learning_rate": 0.00016082000000000002,
+      "loss": 0.0433,
+      "step": 1960
+    },
+    {
+      "epoch": 0.6831975030345067,
+      "grad_norm": 0.07197156548500061,
+      "learning_rate": 0.00016062000000000002,
+      "loss": 0.0465,
+      "step": 1970
+    },
+    {
+      "epoch": 0.6866655106641235,
+      "grad_norm": 0.039968665689229965,
+      "learning_rate": 0.00016042,
+      "loss": 0.0381,
+      "step": 1980
+    },
+    {
+      "epoch": 0.6901335182937403,
+      "grad_norm": 0.05672089383006096,
+      "learning_rate": 0.00016022,
+      "loss": 0.0422,
+      "step": 1990
+    },
+    {
+      "epoch": 0.693601525923357,
+      "grad_norm": 0.10138531029224396,
+      "learning_rate": 0.00016002,
+      "loss": 0.0464,
+      "step": 2000
+    },
+    {
+      "epoch": 0.693601525923357,
+      "eval_loss": 0.06039171665906906,
+      "eval_runtime": 710.8114,
+      "eval_samples_per_second": 12.918,
+      "eval_steps_per_second": 1.615,
+      "step": 2000
+    },
+    {
+      "epoch": 0.6970695335529739,
+      "grad_norm": 0.0607437863945961,
+      "learning_rate": 0.00015982000000000002,
+      "loss": 0.0481,
+      "step": 2010
+    },
+    {
+      "epoch": 0.7005375411825906,
+      "grad_norm": 0.06681676208972931,
+      "learning_rate": 0.00015962000000000002,
+      "loss": 0.049,
+      "step": 2020
+    },
+    {
+      "epoch": 0.7040055488122073,
+      "grad_norm": 0.07225602120161057,
+      "learning_rate": 0.00015942000000000002,
+      "loss": 0.0465,
+      "step": 2030
+    },
+    {
+      "epoch": 0.7074735564418242,
+      "grad_norm": 0.056885506957769394,
+      "learning_rate": 0.00015922,
+      "loss": 0.0452,
+      "step": 2040
+    },
+    {
+      "epoch": 0.7109415640714409,
+      "grad_norm": 0.05572199076414108,
+      "learning_rate": 0.00015902,
+      "loss": 0.0423,
+      "step": 2050
+    },
+    {
+      "epoch": 0.7144095717010578,
+      "grad_norm": 0.0516788586974144,
+      "learning_rate": 0.00015882,
+      "loss": 0.0404,
+      "step": 2060
+    },
+    {
+      "epoch": 0.7178775793306745,
+      "grad_norm": 0.05214313417673111,
+      "learning_rate": 0.00015862000000000002,
+      "loss": 0.0428,
+      "step": 2070
+    },
+    {
+      "epoch": 0.7213455869602913,
+      "grad_norm": 0.06317329406738281,
+      "learning_rate": 0.00015842000000000002,
+      "loss": 0.0447,
+      "step": 2080
+    },
+    {
+      "epoch": 0.7248135945899081,
+      "grad_norm": 0.07574247568845749,
+      "learning_rate": 0.00015822,
+      "loss": 0.0395,
+      "step": 2090
+    },
+    {
+      "epoch": 0.7282816022195249,
+      "grad_norm": 0.06367363035678864,
+      "learning_rate": 0.00015802,
+      "loss": 0.0439,
+      "step": 2100
+    },
+    {
+      "epoch": 0.7317496098491416,
+      "grad_norm": 0.06190785393118858,
+      "learning_rate": 0.00015782,
+      "loss": 0.0404,
+      "step": 2110
+    },
+    {
+      "epoch": 0.7352176174787585,
+      "grad_norm": 0.07103675603866577,
+      "learning_rate": 0.00015762,
+      "loss": 0.0468,
+      "step": 2120
+    },
+    {
+      "epoch": 0.7386856251083752,
+      "grad_norm": 0.04957522451877594,
+      "learning_rate": 0.00015742000000000002,
+      "loss": 0.0453,
+      "step": 2130
+    },
+    {
+      "epoch": 0.7421536327379921,
+      "grad_norm": 0.07643826305866241,
+      "learning_rate": 0.00015722000000000002,
+      "loss": 0.0404,
+      "step": 2140
+    },
+    {
+      "epoch": 0.7456216403676088,
+      "grad_norm": 0.04660920053720474,
+      "learning_rate": 0.00015702,
+      "loss": 0.0373,
+      "step": 2150
+    },
+    {
+      "epoch": 0.7490896479972255,
+      "grad_norm": 0.06206243112683296,
+      "learning_rate": 0.00015682,
+      "loss": 0.0406,
+      "step": 2160
+    },
+    {
+      "epoch": 0.7525576556268424,
+      "grad_norm": 0.07452013343572617,
+      "learning_rate": 0.00015662,
+      "loss": 0.043,
+      "step": 2170
+    },
+    {
+      "epoch": 0.7560256632564591,
+      "grad_norm": 0.04379798844456673,
+      "learning_rate": 0.00015642000000000002,
+      "loss": 0.0494,
+      "step": 2180
+    },
+    {
+      "epoch": 0.759493670886076,
+      "grad_norm": 0.05709415674209595,
+      "learning_rate": 0.00015622000000000002,
+      "loss": 0.041,
+      "step": 2190
+    },
+    {
+      "epoch": 0.7629616785156927,
+      "grad_norm": 0.07468123733997345,
+      "learning_rate": 0.00015602000000000001,
+      "loss": 0.0403,
+      "step": 2200
+    },
+    {
+      "epoch": 0.7664296861453095,
+      "grad_norm": 0.08100239932537079,
+      "learning_rate": 0.00015582,
+      "loss": 0.0481,
+      "step": 2210
+    },
+    {
+      "epoch": 0.7698976937749263,
+      "grad_norm": 0.07269110530614853,
+      "learning_rate": 0.00015562,
+      "loss": 0.0394,
+      "step": 2220
+    },
+    {
+      "epoch": 0.7733657014045431,
+      "grad_norm": 0.060352034866809845,
+      "learning_rate": 0.00015542,
+      "loss": 0.04,
+      "step": 2230
+    },
+    {
+      "epoch": 0.7768337090341598,
+      "grad_norm": 0.05698138475418091,
+      "learning_rate": 0.00015522000000000002,
+      "loss": 0.0383,
+      "step": 2240
+    },
+    {
+      "epoch": 0.7803017166637767,
+      "grad_norm": 0.06030441075563431,
+      "learning_rate": 0.00015502000000000002,
+      "loss": 0.0456,
+      "step": 2250
+    },
+    {
+      "epoch": 0.7837697242933934,
+      "grad_norm": 0.08272086083889008,
+      "learning_rate": 0.00015482,
+      "loss": 0.0443,
+      "step": 2260
+    },
+    {
+      "epoch": 0.7872377319230103,
+      "grad_norm": 0.06746231019496918,
+      "learning_rate": 0.00015462,
+      "loss": 0.04,
+      "step": 2270
+    },
+    {
+      "epoch": 0.790705739552627,
+      "grad_norm": 0.08844996243715286,
+      "learning_rate": 0.00015442,
+      "loss": 0.0418,
+      "step": 2280
+    },
+    {
+      "epoch": 0.7941737471822438,
+      "grad_norm": 0.04906987398862839,
+      "learning_rate": 0.00015422,
+      "loss": 0.0452,
+      "step": 2290
+    },
+    {
+      "epoch": 0.7976417548118606,
+      "grad_norm": 0.053966376930475235,
+      "learning_rate": 0.00015402000000000002,
+      "loss": 0.0404,
+      "step": 2300
+    },
+    {
+      "epoch": 0.8011097624414774,
+      "grad_norm": 0.07681586593389511,
+      "learning_rate": 0.00015382000000000001,
+      "loss": 0.05,
+      "step": 2310
+    },
+    {
+      "epoch": 0.8045777700710942,
+      "grad_norm": 0.059431836009025574,
+      "learning_rate": 0.00015362,
+      "loss": 0.0496,
+      "step": 2320
+    },
+    {
+      "epoch": 0.808045777700711,
+      "grad_norm": 0.058449339121580124,
+      "learning_rate": 0.00015342,
+      "loss": 0.0419,
+      "step": 2330
+    },
+    {
+      "epoch": 0.8115137853303277,
+      "grad_norm": 0.05404651537537575,
+      "learning_rate": 0.00015322,
+      "loss": 0.0382,
+      "step": 2340
+    },
+    {
+      "epoch": 0.8149817929599446,
+      "grad_norm": 0.06095472350716591,
+      "learning_rate": 0.00015302,
+      "loss": 0.0487,
+      "step": 2350
+    },
+    {
+      "epoch": 0.8184498005895613,
+      "grad_norm": 0.07456117123365402,
+      "learning_rate": 0.00015282000000000002,
+      "loss": 0.0386,
+      "step": 2360
+    },
+    {
+      "epoch": 0.821917808219178,
+      "grad_norm": 0.06098613142967224,
+      "learning_rate": 0.00015262,
+      "loss": 0.0425,
+      "step": 2370
+    },
+    {
+      "epoch": 0.8253858158487949,
+      "grad_norm": 0.07008852809667587,
+      "learning_rate": 0.00015242,
+      "loss": 0.0457,
+      "step": 2380
+    },
+    {
+      "epoch": 0.8288538234784116,
+      "grad_norm": 0.06413611769676208,
+      "learning_rate": 0.00015222,
+      "loss": 0.0433,
+      "step": 2390
+    },
+    {
+      "epoch": 0.8323218311080285,
+      "grad_norm": 0.05168429762125015,
+      "learning_rate": 0.00015202,
+      "loss": 0.0441,
+      "step": 2400
+    },
+    {
+      "epoch": 0.8357898387376452,
+      "grad_norm": 0.0708162784576416,
+      "learning_rate": 0.00015182,
+      "loss": 0.0393,
+      "step": 2410
+    },
+    {
+      "epoch": 0.839257846367262,
+      "grad_norm": 0.05856487527489662,
+      "learning_rate": 0.00015162000000000002,
+      "loss": 0.0437,
+      "step": 2420
+    },
+    {
+      "epoch": 0.8427258539968788,
+      "grad_norm": 0.06069020926952362,
+      "learning_rate": 0.00015142,
+      "loss": 0.0484,
+      "step": 2430
+    },
+    {
+      "epoch": 0.8461938616264956,
+      "grad_norm": 0.0641227513551712,
+      "learning_rate": 0.00015122,
+      "loss": 0.046,
+      "step": 2440
+    },
+    {
+      "epoch": 0.8496618692561123,
+      "grad_norm": 0.06422239542007446,
+      "learning_rate": 0.00015102,
+      "loss": 0.043,
+      "step": 2450
+    },
+    {
+      "epoch": 0.8531298768857292,
+      "grad_norm": 0.07488572597503662,
+      "learning_rate": 0.00015082,
+      "loss": 0.0517,
+      "step": 2460
+    },
+    {
+      "epoch": 0.8565978845153459,
+      "grad_norm": 0.06784242391586304,
+      "learning_rate": 0.00015062000000000002,
+      "loss": 0.0437,
+      "step": 2470
+    },
+    {
+      "epoch": 0.8600658921449628,
+      "grad_norm": 0.07483550906181335,
+      "learning_rate": 0.00015042,
+      "loss": 0.0491,
+      "step": 2480
+    },
+    {
+      "epoch": 0.8635338997745795,
+      "grad_norm": 0.06362838298082352,
+      "learning_rate": 0.00015022,
+      "loss": 0.043,
+      "step": 2490
+    },
+    {
+      "epoch": 0.8670019074041962,
+      "grad_norm": 0.06175532937049866,
+      "learning_rate": 0.00015002,
+      "loss": 0.0425,
+      "step": 2500
+    },
+    {
+      "epoch": 0.8704699150338131,
+      "grad_norm": 0.08987358957529068,
+      "learning_rate": 0.00014982,
+      "loss": 0.0479,
+      "step": 2510
+    },
+    {
+      "epoch": 0.8739379226634298,
+      "grad_norm": 0.05209062620997429,
+      "learning_rate": 0.00014962,
+      "loss": 0.045,
+      "step": 2520
+    },
+    {
+      "epoch": 0.8774059302930467,
+      "grad_norm": 0.055044736713171005,
+      "learning_rate": 0.00014942000000000002,
+      "loss": 0.0469,
+      "step": 2530
+    },
+    {
+      "epoch": 0.8808739379226634,
+      "grad_norm": 0.0646812692284584,
+      "learning_rate": 0.00014922,
+      "loss": 0.0426,
+      "step": 2540
+    },
+    {
+      "epoch": 0.8843419455522802,
+      "grad_norm": 0.048389263451099396,
+      "learning_rate": 0.00014902,
+      "loss": 0.0417,
+      "step": 2550
+    },
+    {
+      "epoch": 0.887809953181897,
+      "grad_norm": 0.07316736876964569,
+      "learning_rate": 0.00014882,
+      "loss": 0.0412,
+      "step": 2560
+    },
+    {
+      "epoch": 0.8912779608115138,
+      "grad_norm": 0.06801818311214447,
+      "learning_rate": 0.00014862,
+      "loss": 0.0446,
+      "step": 2570
+    },
+    {
+      "epoch": 0.8947459684411305,
+      "grad_norm": 0.08190831542015076,
+      "learning_rate": 0.00014842,
+      "loss": 0.0423,
+      "step": 2580
+    },
+    {
+      "epoch": 0.8982139760707474,
+      "grad_norm": 0.05850045010447502,
+      "learning_rate": 0.00014822000000000001,
+      "loss": 0.0418,
+      "step": 2590
+    },
+    {
+      "epoch": 0.9016819837003641,
+      "grad_norm": 0.07633431255817413,
+      "learning_rate": 0.00014802,
+      "loss": 0.0439,
+      "step": 2600
+    },
+    {
+      "epoch": 0.905149991329981,
+      "grad_norm": 0.07250861078500748,
+      "learning_rate": 0.00014782,
+      "loss": 0.0456,
+      "step": 2610
+    },
+    {
+      "epoch": 0.9086179989595977,
+      "grad_norm": 0.05381698161363602,
+      "learning_rate": 0.00014762,
+      "loss": 0.0398,
+      "step": 2620
+    },
+    {
+      "epoch": 0.9120860065892145,
+      "grad_norm": 0.05847073718905449,
+      "learning_rate": 0.00014742,
+      "loss": 0.0426,
+      "step": 2630
+    },
+    {
+      "epoch": 0.9155540142188313,
+      "grad_norm": 0.04121188446879387,
+      "learning_rate": 0.00014722,
+      "loss": 0.0379,
+      "step": 2640
+    },
+    {
+      "epoch": 0.9190220218484481,
+      "grad_norm": 0.05500589683651924,
+      "learning_rate": 0.00014702,
+      "loss": 0.0429,
+      "step": 2650
+    },
+    {
+      "epoch": 0.9224900294780648,
+      "grad_norm": 0.07371719181537628,
+      "learning_rate": 0.00014682,
+      "loss": 0.0404,
+      "step": 2660
+    },
+    {
+      "epoch": 0.9259580371076817,
+      "grad_norm": 0.07182417809963226,
+      "learning_rate": 0.00014662,
+      "loss": 0.0425,
+      "step": 2670
+    },
+    {
+      "epoch": 0.9294260447372984,
+      "grad_norm": 0.07057616114616394,
+      "learning_rate": 0.00014642,
+      "loss": 0.0447,
+      "step": 2680
+    },
+    {
+      "epoch": 0.9328940523669152,
+      "grad_norm": 0.06059495732188225,
+      "learning_rate": 0.00014622,
+      "loss": 0.0467,
+      "step": 2690
+    },
+    {
+      "epoch": 0.936362059996532,
+      "grad_norm": 0.05471622198820114,
+      "learning_rate": 0.00014602,
+      "loss": 0.0429,
+      "step": 2700
+    },
+    {
+      "epoch": 0.9398300676261487,
+      "grad_norm": 0.2795173227787018,
+      "learning_rate": 0.00014582,
+      "loss": 0.0441,
+      "step": 2710
+    },
+    {
+      "epoch": 0.9432980752557656,
+      "grad_norm": 0.05505786091089249,
+      "learning_rate": 0.00014562,
+      "loss": 0.0481,
+      "step": 2720
+    },
+    {
+      "epoch": 0.9467660828853823,
+      "grad_norm": 0.06604549288749695,
+      "learning_rate": 0.00014542,
+      "loss": 0.0436,
+      "step": 2730
+    },
+    {
+      "epoch": 0.9502340905149992,
+      "grad_norm": 0.04876833036541939,
+      "learning_rate": 0.00014522,
+      "loss": 0.0427,
+      "step": 2740
+    },
+    {
+      "epoch": 0.9537020981446159,
+      "grad_norm": 0.06586755067110062,
+      "learning_rate": 0.00014502,
+      "loss": 0.0376,
+      "step": 2750
+    },
+    {
+      "epoch": 0.9571701057742327,
+      "grad_norm": 0.05991548299789429,
+      "learning_rate": 0.00014482,
+      "loss": 0.0425,
+      "step": 2760
+    },
+    {
+      "epoch": 0.9606381134038495,
+      "grad_norm": 0.05464167147874832,
+      "learning_rate": 0.00014462,
+      "loss": 0.0475,
+      "step": 2770
+    },
+    {
+      "epoch": 0.9641061210334663,
+      "grad_norm": 0.08428128808736801,
+      "learning_rate": 0.00014442,
+      "loss": 0.0407,
+      "step": 2780
+    },
+    {
+      "epoch": 0.967574128663083,
+      "grad_norm": 0.05811979994177818,
+      "learning_rate": 0.00014422,
+      "loss": 0.0417,
+      "step": 2790
+    },
+    {
+      "epoch": 0.9710421362926999,
+      "grad_norm": 0.06777170300483704,
+      "learning_rate": 0.00014402,
+      "loss": 0.0394,
+      "step": 2800
+    },
+    {
+      "epoch": 0.9745101439223166,
+      "grad_norm": 0.07404989749193192,
+      "learning_rate": 0.00014382,
+      "loss": 0.0451,
+      "step": 2810
+    },
+    {
+      "epoch": 0.9779781515519335,
+      "grad_norm": 0.08595024049282074,
+      "learning_rate": 0.00014362,
+      "loss": 0.0518,
+      "step": 2820
+    },
+    {
+      "epoch": 0.9814461591815502,
+      "grad_norm": 0.07936517149209976,
+      "learning_rate": 0.00014342,
+      "loss": 0.0368,
+      "step": 2830
+    },
+    {
+      "epoch": 0.984914166811167,
+      "grad_norm": 0.08811389654874802,
+      "learning_rate": 0.00014322,
+      "loss": 0.0444,
+      "step": 2840
+    },
+    {
+      "epoch": 0.9883821744407838,
+      "grad_norm": 0.06864507496356964,
+      "learning_rate": 0.00014302,
+      "loss": 0.0445,
+      "step": 2850
+    },
+    {
+      "epoch": 0.9918501820704005,
+      "grad_norm": 0.05267275124788284,
+      "learning_rate": 0.00014282,
+      "loss": 0.0485,
+      "step": 2860
+    },
+    {
+      "epoch": 0.9953181897000173,
+      "grad_norm": 0.051028452813625336,
+      "learning_rate": 0.00014261999999999999,
+      "loss": 0.0416,
+      "step": 2870
+    },
+    {
+      "epoch": 0.9987861973296341,
+      "grad_norm": 0.047300126403570175,
+      "learning_rate": 0.00014242,
+      "loss": 0.0443,
+      "step": 2880
+    },
+    {
+      "epoch": 1.0022542049592509,
+      "grad_norm": 0.06848949193954468,
+      "learning_rate": 0.00014222,
+      "loss": 0.0422,
+      "step": 2890
+    },
+    {
+      "epoch": 1.0057222125888676,
+      "grad_norm": 0.05861698463559151,
+      "learning_rate": 0.00014202,
+      "loss": 0.0298,
+      "step": 2900
+    },
+    {
+      "epoch": 1.0091902202184846,
+      "grad_norm": 0.0671234130859375,
+      "learning_rate": 0.00014182,
+      "loss": 0.0271,
+      "step": 2910
+    },
+    {
+      "epoch": 1.0126582278481013,
+      "grad_norm": 0.07007008045911789,
+      "learning_rate": 0.00014162,
+      "loss": 0.0279,
+      "step": 2920
+    },
+    {
+      "epoch": 1.016126235477718,
+      "grad_norm": 0.07409070432186127,
+      "learning_rate": 0.00014141999999999998,
+      "loss": 0.0332,
+      "step": 2930
+    },
+    {
+      "epoch": 1.0195942431073348,
+      "grad_norm": 0.0820993110537529,
+      "learning_rate": 0.00014122,
+      "loss": 0.032,
+      "step": 2940
+    },
+    {
+      "epoch": 1.0230622507369516,
+      "grad_norm": 0.0683741420507431,
+      "learning_rate": 0.00014102,
+      "loss": 0.0285,
+      "step": 2950
+    },
+    {
+      "epoch": 1.0265302583665683,
+      "grad_norm": 0.09559917449951172,
+      "learning_rate": 0.00014082,
+      "loss": 0.0328,
+      "step": 2960
+    },
+    {
+      "epoch": 1.0299982659961853,
+      "grad_norm": 0.07927672564983368,
+      "learning_rate": 0.00014062,
+      "loss": 0.0317,
+      "step": 2970
+    },
+    {
+      "epoch": 1.033466273625802,
+      "grad_norm": 0.06615123897790909,
+      "learning_rate": 0.00014042,
+      "loss": 0.0289,
+      "step": 2980
+    },
+    {
+      "epoch": 1.0369342812554188,
+      "grad_norm": 0.051615212112665176,
+      "learning_rate": 0.00014022,
+      "loss": 0.0274,
+      "step": 2990
+    },
+    {
+      "epoch": 1.0404022888850355,
+      "grad_norm": 0.060853827744722366,
+      "learning_rate": 0.00014002,
+      "loss": 0.033,
+      "step": 3000
+    },
+    {
+      "epoch": 1.0404022888850355,
+      "eval_loss": 0.060293715447187424,
+      "eval_runtime": 713.1843,
+      "eval_samples_per_second": 12.875,
+      "eval_steps_per_second": 1.61,
+      "step": 3000
+    },
+    {
+      "epoch": 1.0438702965146522,
+      "grad_norm": 0.05449477955698967,
+      "learning_rate": 0.00013982000000000003,
+      "loss": 0.0308,
+      "step": 3010
+    },
+    {
+      "epoch": 1.0473383041442692,
+      "grad_norm": 0.06479578465223312,
+      "learning_rate": 0.00013962000000000002,
+      "loss": 0.0343,
+      "step": 3020
+    },
+    {
+      "epoch": 1.050806311773886,
+      "grad_norm": 0.060166362673044205,
+      "learning_rate": 0.00013942000000000002,
+      "loss": 0.0366,
+      "step": 3030
+    },
+    {
+      "epoch": 1.0542743194035027,
+      "grad_norm": 0.07008329033851624,
+      "learning_rate": 0.00013922,
+      "loss": 0.0327,
+      "step": 3040
+    },
+    {
+      "epoch": 1.0577423270331194,
+      "grad_norm": 0.07188612222671509,
+      "learning_rate": 0.00013902,
+      "loss": 0.0317,
+      "step": 3050
+    },
+    {
+      "epoch": 1.0612103346627362,
+      "grad_norm": 0.06554035097360611,
+      "learning_rate": 0.00013882000000000003,
+      "loss": 0.031,
+      "step": 3060
+    },
+    {
+      "epoch": 1.0646783422923531,
+      "grad_norm": 0.0675990879535675,
+      "learning_rate": 0.00013862000000000002,
+      "loss": 0.0341,
+      "step": 3070
+    },
+    {
+      "epoch": 1.0681463499219699,
+      "grad_norm": 0.061157677322626114,
+      "learning_rate": 0.00013842000000000002,
+      "loss": 0.0301,
+      "step": 3080
+    },
+    {
+      "epoch": 1.0716143575515866,
+      "grad_norm": 0.07951588183641434,
+      "learning_rate": 0.00013822000000000001,
+      "loss": 0.0289,
+      "step": 3090
+    },
+    {
+      "epoch": 1.0750823651812034,
+      "grad_norm": 0.1063622459769249,
+      "learning_rate": 0.00013802,
+      "loss": 0.0325,
+      "step": 3100
+    },
+    {
+      "epoch": 1.0785503728108201,
+      "grad_norm": 0.07916730642318726,
+      "learning_rate": 0.00013782,
+      "loss": 0.0338,
+      "step": 3110
+    },
+    {
+      "epoch": 1.082018380440437,
+      "grad_norm": 0.0792151466012001,
+      "learning_rate": 0.00013762000000000003,
+      "loss": 0.0287,
+      "step": 3120
+    },
+    {
+      "epoch": 1.0854863880700538,
+      "grad_norm": 0.0631512925028801,
+      "learning_rate": 0.00013742000000000002,
+      "loss": 0.031,
+      "step": 3130
+    },
+    {
+      "epoch": 1.0889543956996706,
+      "grad_norm": 0.07032682001590729,
+      "learning_rate": 0.00013722000000000002,
+      "loss": 0.0284,
+      "step": 3140
+    },
+    {
+      "epoch": 1.0924224033292873,
+      "grad_norm": 0.07017088681459427,
+      "learning_rate": 0.00013702,
+      "loss": 0.0292,
+      "step": 3150
+    },
+    {
+      "epoch": 1.095890410958904,
+      "grad_norm": 0.06100435182452202,
+      "learning_rate": 0.00013682,
+      "loss": 0.0359,
+      "step": 3160
+    },
+    {
+      "epoch": 1.0993584185885208,
+      "grad_norm": 0.05606581270694733,
+      "learning_rate": 0.00013662,
+      "loss": 0.032,
+      "step": 3170
+    },
+    {
+      "epoch": 1.1028264262181378,
+      "grad_norm": 0.07687368988990784,
+      "learning_rate": 0.00013642000000000003,
+      "loss": 0.0353,
+      "step": 3180
+    },
+    {
+      "epoch": 1.1062944338477545,
+      "grad_norm": 0.05796977877616882,
+      "learning_rate": 0.00013622000000000002,
+      "loss": 0.0286,
+      "step": 3190
+    },
+    {
+      "epoch": 1.1097624414773712,
+      "grad_norm": 0.08586996048688889,
+      "learning_rate": 0.00013602000000000002,
+      "loss": 0.0285,
+      "step": 3200
+    },
+    {
+      "epoch": 1.113230449106988,
+      "grad_norm": 0.08199802041053772,
+      "learning_rate": 0.00013582,
+      "loss": 0.0346,
+      "step": 3210
+    },
+    {
+      "epoch": 1.1166984567366047,
+      "grad_norm": 0.06581319123506546,
+      "learning_rate": 0.00013562,
+      "loss": 0.029,
+      "step": 3220
+    },
+    {
+      "epoch": 1.1201664643662217,
+      "grad_norm": 0.06755177676677704,
+      "learning_rate": 0.00013542,
+      "loss": 0.0323,
+      "step": 3230
+    },
+    {
+      "epoch": 1.1236344719958384,
+      "grad_norm": 0.07359416782855988,
+      "learning_rate": 0.00013522000000000002,
+      "loss": 0.031,
+      "step": 3240
+    },
+    {
+      "epoch": 1.1271024796254552,
+      "grad_norm": 0.07140175998210907,
+      "learning_rate": 0.00013502000000000002,
+      "loss": 0.0341,
+      "step": 3250
+    },
+    {
+      "epoch": 1.130570487255072,
+      "grad_norm": 0.07570434361696243,
+      "learning_rate": 0.00013482000000000001,
+      "loss": 0.0307,
+      "step": 3260
+    },
+    {
+      "epoch": 1.1340384948846887,
+      "grad_norm": 0.0646577849984169,
+      "learning_rate": 0.00013462,
+      "loss": 0.0353,
+      "step": 3270
+    },
+    {
+      "epoch": 1.1375065025143056,
+      "grad_norm": 0.07016121596097946,
+      "learning_rate": 0.00013442,
+      "loss": 0.0367,
+      "step": 3280
+    },
+    {
+      "epoch": 1.1409745101439224,
+      "grad_norm": 0.06299825757741928,
+      "learning_rate": 0.00013422,
+      "loss": 0.0355,
+      "step": 3290
+    },
+    {
+      "epoch": 1.1444425177735391,
+      "grad_norm": 0.07232199609279633,
+      "learning_rate": 0.00013402000000000002,
+      "loss": 0.0302,
+      "step": 3300
+    },
+    {
+      "epoch": 1.1479105254031559,
+      "grad_norm": 0.06672387570142746,
+      "learning_rate": 0.00013382000000000002,
+      "loss": 0.0314,
+      "step": 3310
+    },
+    {
+      "epoch": 1.1513785330327726,
+      "grad_norm": 0.06597165018320084,
+      "learning_rate": 0.00013362,
+      "loss": 0.0316,
+      "step": 3320
+    },
+    {
+      "epoch": 1.1548465406623896,
+      "grad_norm": 0.07945774495601654,
+      "learning_rate": 0.00013342,
+      "loss": 0.0304,
+      "step": 3330
+    },
+    {
+      "epoch": 1.1583145482920063,
+      "grad_norm": 0.0883309543132782,
+      "learning_rate": 0.00013322,
+      "loss": 0.0313,
+      "step": 3340
+    },
+    {
+      "epoch": 1.161782555921623,
+      "grad_norm": 0.1610005646944046,
+      "learning_rate": 0.00013302000000000002,
+      "loss": 0.0379,
+      "step": 3350
+    },
+    {
+      "epoch": 1.1652505635512398,
+      "grad_norm": 0.08350630104541779,
+      "learning_rate": 0.00013282000000000002,
+      "loss": 0.0283,
+      "step": 3360
+    },
+    {
+      "epoch": 1.1687185711808565,
+      "grad_norm": 0.08908521384000778,
+      "learning_rate": 0.00013262000000000001,
+      "loss": 0.0271,
+      "step": 3370
+    },
+    {
+      "epoch": 1.1721865788104733,
+      "grad_norm": 0.06909502297639847,
+      "learning_rate": 0.00013242,
+      "loss": 0.0312,
+      "step": 3380
+    },
+    {
+      "epoch": 1.1756545864400902,
+      "grad_norm": 0.0837428942322731,
+      "learning_rate": 0.00013222,
+      "loss": 0.0351,
+      "step": 3390
+    },
+    {
+      "epoch": 1.179122594069707,
+      "grad_norm": 0.06636606156826019,
+      "learning_rate": 0.00013202,
+      "loss": 0.0284,
+      "step": 3400
+    },
+    {
+      "epoch": 1.1825906016993237,
+      "grad_norm": 0.0990837961435318,
+      "learning_rate": 0.00013182000000000002,
+      "loss": 0.0301,
+      "step": 3410
+    },
+    {
+      "epoch": 1.1860586093289405,
+      "grad_norm": 0.08313869684934616,
+      "learning_rate": 0.00013162000000000002,
+      "loss": 0.0335,
+      "step": 3420
+    },
+    {
+      "epoch": 1.1895266169585572,
+      "grad_norm": 0.07330479472875595,
+      "learning_rate": 0.00013142,
+      "loss": 0.0328,
+      "step": 3430
+    },
+    {
+      "epoch": 1.1929946245881742,
+      "grad_norm": 0.05378459393978119,
+      "learning_rate": 0.00013122,
+      "loss": 0.0306,
+      "step": 3440
+    },
+    {
+      "epoch": 1.196462632217791,
+      "grad_norm": 0.06030990183353424,
+      "learning_rate": 0.00013102,
+      "loss": 0.0303,
+      "step": 3450
+    },
+    {
+      "epoch": 1.1999306398474077,
+      "grad_norm": 0.07298003882169724,
+      "learning_rate": 0.00013082,
+      "loss": 0.0312,
+      "step": 3460
+    },
+    {
+      "epoch": 1.2033986474770244,
+      "grad_norm": 0.06307482719421387,
+      "learning_rate": 0.00013062000000000002,
+      "loss": 0.0324,
+      "step": 3470
+    },
+    {
+      "epoch": 1.2068666551066412,
+      "grad_norm": 0.06007950380444527,
+      "learning_rate": 0.00013042000000000002,
+      "loss": 0.0339,
+      "step": 3480
+    },
+    {
+      "epoch": 1.2103346627362581,
+      "grad_norm": 0.07138363271951675,
+      "learning_rate": 0.00013022,
+      "loss": 0.0371,
+      "step": 3490
+    },
+    {
+      "epoch": 1.2138026703658749,
+      "grad_norm": 0.06266158819198608,
+      "learning_rate": 0.00013002,
+      "loss": 0.0329,
+      "step": 3500
+    },
+    {
+      "epoch": 1.2172706779954916,
+      "grad_norm": 0.06397438049316406,
+      "learning_rate": 0.00012982,
+      "loss": 0.0296,
+      "step": 3510
+    },
+    {
+      "epoch": 1.2207386856251083,
+      "grad_norm": 0.061814188957214355,
+      "learning_rate": 0.00012962,
+      "loss": 0.0306,
+      "step": 3520
+    },
+    {
+      "epoch": 1.224206693254725,
+      "grad_norm": 0.060092389583587646,
+      "learning_rate": 0.00012942000000000002,
+      "loss": 0.0323,
+      "step": 3530
+    },
+    {
+      "epoch": 1.227674700884342,
+      "grad_norm": 0.10667088627815247,
+      "learning_rate": 0.00012922,
+      "loss": 0.0298,
+      "step": 3540
+    },
+    {
+      "epoch": 1.2311427085139588,
+      "grad_norm": 0.09048482030630112,
+      "learning_rate": 0.00012902,
+      "loss": 0.0319,
+      "step": 3550
+    },
+    {
+      "epoch": 1.2346107161435755,
+      "grad_norm": 0.09124518185853958,
+      "learning_rate": 0.00012882,
+      "loss": 0.0361,
+      "step": 3560
+    },
+    {
+      "epoch": 1.2380787237731923,
+      "grad_norm": 0.05594000220298767,
+      "learning_rate": 0.00012862,
+      "loss": 0.0341,
+      "step": 3570
+    },
+    {
+      "epoch": 1.241546731402809,
+      "grad_norm": 0.06354895979166031,
+      "learning_rate": 0.00012842,
+      "loss": 0.0281,
+      "step": 3580
+    },
+    {
+      "epoch": 1.2450147390324258,
+      "grad_norm": 0.059312548488378525,
+      "learning_rate": 0.00012822000000000002,
+      "loss": 0.0276,
+      "step": 3590
+    },
+    {
+      "epoch": 1.2484827466620427,
+      "grad_norm": 0.06291409581899643,
+      "learning_rate": 0.00012802,
+      "loss": 0.0315,
+      "step": 3600
+    },
+    {
+      "epoch": 1.2519507542916595,
+      "grad_norm": 0.05183565244078636,
+      "learning_rate": 0.00012782,
+      "loss": 0.0294,
+      "step": 3610
+    },
+    {
+      "epoch": 1.2554187619212762,
+      "grad_norm": 0.06372030079364777,
+      "learning_rate": 0.00012762,
+      "loss": 0.03,
+      "step": 3620
+    },
+    {
+      "epoch": 1.258886769550893,
+      "grad_norm": 0.1327325701713562,
+      "learning_rate": 0.00012742,
+      "loss": 0.0366,
+      "step": 3630
+    },
+    {
+      "epoch": 1.26235477718051,
+      "grad_norm": 0.07337796688079834,
+      "learning_rate": 0.00012722000000000002,
+      "loss": 0.0324,
+      "step": 3640
+    },
+    {
+      "epoch": 1.2658227848101267,
+      "grad_norm": 0.06626396626234055,
+      "learning_rate": 0.00012702000000000001,
+      "loss": 0.0322,
+      "step": 3650
+    },
+    {
+      "epoch": 1.2692907924397434,
+      "grad_norm": 0.07255198061466217,
+      "learning_rate": 0.00012682,
+      "loss": 0.036,
+      "step": 3660
+    },
+    {
+      "epoch": 1.2727588000693602,
+      "grad_norm": 0.0766686201095581,
+      "learning_rate": 0.00012662,
+      "loss": 0.0315,
+      "step": 3670
+    },
+    {
+      "epoch": 1.276226807698977,
+      "grad_norm": 0.06377042084932327,
+      "learning_rate": 0.00012642,
+      "loss": 0.0322,
+      "step": 3680
+    },
+    {
+      "epoch": 1.2796948153285936,
+      "grad_norm": 0.0723329707980156,
+      "learning_rate": 0.00012622,
+      "loss": 0.0356,
+      "step": 3690
+    },
+    {
+      "epoch": 1.2831628229582104,
+      "grad_norm": 0.0876326933503151,
+      "learning_rate": 0.00012602000000000002,
+      "loss": 0.0346,
+      "step": 3700
+    },
+    {
+      "epoch": 1.2866308305878273,
+      "grad_norm": 0.07476814091205597,
+      "learning_rate": 0.00012582,
+      "loss": 0.0316,
+      "step": 3710
+    },
+    {
+      "epoch": 1.290098838217444,
+      "grad_norm": 0.07503268122673035,
+      "learning_rate": 0.00012562,
+      "loss": 0.0318,
+      "step": 3720
+    },
+    {
+      "epoch": 1.2935668458470608,
+      "grad_norm": 0.10708837956190109,
+      "learning_rate": 0.00012542,
+      "loss": 0.0349,
+      "step": 3730
+    },
+    {
+      "epoch": 1.2970348534766776,
+      "grad_norm": 0.08280046284198761,
+      "learning_rate": 0.00012522,
+      "loss": 0.0324,
+      "step": 3740
+    },
+    {
+      "epoch": 1.3005028611062945,
+      "grad_norm": 0.06630868464708328,
+      "learning_rate": 0.00012502,
+      "loss": 0.0305,
+      "step": 3750
+    },
+    {
+      "epoch": 1.3039708687359113,
+      "grad_norm": 0.09755595773458481,
+      "learning_rate": 0.00012482000000000001,
+      "loss": 0.0338,
+      "step": 3760
+    },
+    {
+      "epoch": 1.307438876365528,
+      "grad_norm": 0.08033673465251923,
+      "learning_rate": 0.00012462,
+      "loss": 0.0312,
+      "step": 3770
+    },
+    {
+      "epoch": 1.3109068839951448,
+      "grad_norm": 0.0811261385679245,
+      "learning_rate": 0.00012442,
+      "loss": 0.0322,
+      "step": 3780
+    },
+    {
+      "epoch": 1.3143748916247615,
+      "grad_norm": 0.06264316290616989,
+      "learning_rate": 0.00012422,
+      "loss": 0.0292,
+      "step": 3790
+    },
+    {
+      "epoch": 1.3178428992543783,
+      "grad_norm": 0.07748369127511978,
+      "learning_rate": 0.00012402,
+      "loss": 0.0302,
+      "step": 3800
+    },
+    {
+      "epoch": 1.3213109068839952,
+      "grad_norm": 0.0690523013472557,
+      "learning_rate": 0.00012382,
+      "loss": 0.0336,
+      "step": 3810
+    },
+    {
+      "epoch": 1.324778914513612,
+      "grad_norm": 0.09423090517520905,
+      "learning_rate": 0.00012362,
+      "loss": 0.0295,
+      "step": 3820
+    },
+    {
+      "epoch": 1.3282469221432287,
+      "grad_norm": 0.08562049269676208,
+      "learning_rate": 0.00012342,
+      "loss": 0.0292,
+      "step": 3830
+    },
+    {
+      "epoch": 1.3317149297728454,
+      "grad_norm": 0.05059509724378586,
+      "learning_rate": 0.00012322,
+      "loss": 0.0266,
+      "step": 3840
+    },
+    {
+      "epoch": 1.3351829374024624,
+      "grad_norm": 0.06358881294727325,
+      "learning_rate": 0.00012302,
+      "loss": 0.0309,
+      "step": 3850
+    },
+    {
+      "epoch": 1.3386509450320792,
+      "grad_norm": 0.1298878937959671,
+      "learning_rate": 0.00012282,
+      "loss": 0.0356,
+      "step": 3860
+    },
+    {
+      "epoch": 1.342118952661696,
+      "grad_norm": 0.06405311822891235,
+      "learning_rate": 0.00012262,
+      "loss": 0.0317,
+      "step": 3870
+    },
+    {
+      "epoch": 1.3455869602913126,
+      "grad_norm": 0.07615106552839279,
+      "learning_rate": 0.00012242,
+      "loss": 0.0285,
+      "step": 3880
+    },
+    {
+      "epoch": 1.3490549679209294,
+      "grad_norm": 0.08331302553415298,
+      "learning_rate": 0.00012222,
+      "loss": 0.0359,
+      "step": 3890
+    },
+    {
+      "epoch": 1.3525229755505461,
+      "grad_norm": 0.06869524717330933,
+      "learning_rate": 0.00012202,
+      "loss": 0.0319,
+      "step": 3900
+    },
+    {
+      "epoch": 1.3559909831801629,
+      "grad_norm": 0.08540484309196472,
+      "learning_rate": 0.00012182,
+      "loss": 0.0328,
+      "step": 3910
+    },
+    {
+      "epoch": 1.3594589908097798,
+      "grad_norm": 0.07091011852025986,
+      "learning_rate": 0.00012162,
+      "loss": 0.0368,
+      "step": 3920
+    },
+    {
+      "epoch": 1.3629269984393966,
+      "grad_norm": 0.075434111058712,
+      "learning_rate": 0.00012142,
+      "loss": 0.0336,
+      "step": 3930
+    },
+    {
+      "epoch": 1.3663950060690133,
+      "grad_norm": 0.06716951727867126,
+      "learning_rate": 0.00012122,
+      "loss": 0.0335,
+      "step": 3940
+    },
+    {
+      "epoch": 1.36986301369863,
+      "grad_norm": 0.09082087874412537,
+      "learning_rate": 0.00012102,
+      "loss": 0.0345,
+      "step": 3950
+    },
+    {
+      "epoch": 1.373331021328247,
+      "grad_norm": 0.0903453379869461,
+      "learning_rate": 0.00012082,
+      "loss": 0.0415,
+      "step": 3960
+    },
+    {
+      "epoch": 1.3767990289578638,
+      "grad_norm": 0.052235305309295654,
+      "learning_rate": 0.00012062,
+      "loss": 0.0282,
+      "step": 3970
+    },
+    {
+      "epoch": 1.3802670365874805,
+      "grad_norm": 0.07253699749708176,
+      "learning_rate": 0.00012042,
+      "loss": 0.0316,
+      "step": 3980
+    },
+    {
+      "epoch": 1.3837350442170973,
+      "grad_norm": 0.0548410601913929,
+      "learning_rate": 0.00012022,
+      "loss": 0.0303,
+      "step": 3990
+    },
+    {
+      "epoch": 1.387203051846714,
+      "grad_norm": 0.08785740286111832,
+      "learning_rate": 0.00012001999999999999,
+      "loss": 0.0337,
+      "step": 4000
+    },
+    {
+      "epoch": 1.387203051846714,
+      "eval_loss": 0.060144323855638504,
+      "eval_runtime": 708.9302,
+      "eval_samples_per_second": 12.952,
+      "eval_steps_per_second": 1.619,
+      "step": 4000
+    },
+    {
+      "epoch": 1.3906710594763307,
+      "grad_norm": 0.0650157481431961,
+      "learning_rate": 0.00011982,
+      "loss": 0.0327,
+      "step": 4010
+    },
+    {
+      "epoch": 1.3941390671059477,
+      "grad_norm": 0.06679214537143707,
+      "learning_rate": 0.00011962,
+      "loss": 0.0352,
+      "step": 4020
+    },
+    {
+      "epoch": 1.3976070747355644,
+      "grad_norm": 0.08368890732526779,
+      "learning_rate": 0.00011942,
+      "loss": 0.0346,
+      "step": 4030
+    },
+    {
+      "epoch": 1.4010750823651812,
+      "grad_norm": 0.09027834981679916,
+      "learning_rate": 0.00011922,
+      "loss": 0.0304,
+      "step": 4040
+    },
+    {
+      "epoch": 1.404543089994798,
+      "grad_norm": 0.0602988600730896,
+      "learning_rate": 0.00011902,
+      "loss": 0.0278,
+      "step": 4050
+    },
+    {
+      "epoch": 1.408011097624415,
+      "grad_norm": 0.08348573744297028,
+      "learning_rate": 0.00011882,
+      "loss": 0.0316,
+      "step": 4060
+    },
+    {
+      "epoch": 1.4114791052540316,
+      "grad_norm": 0.065241239964962,
+      "learning_rate": 0.00011862,
+      "loss": 0.0308,
+      "step": 4070
+    },
+    {
+      "epoch": 1.4149471128836484,
+      "grad_norm": 0.058722469955682755,
+      "learning_rate": 0.00011842,
+      "loss": 0.0312,
+      "step": 4080
+    },
+    {
+      "epoch": 1.4184151205132651,
+      "grad_norm": 0.06701633334159851,
+      "learning_rate": 0.00011822,
+      "loss": 0.0303,
+      "step": 4090
+    },
+    {
+      "epoch": 1.4218831281428819,
+      "grad_norm": 0.06035483255982399,
+      "learning_rate": 0.00011802,
+      "loss": 0.0263,
+      "step": 4100
+    },
+    {
+      "epoch": 1.4253511357724986,
+      "grad_norm": 0.07209423929452896,
+      "learning_rate": 0.00011782,
+      "loss": 0.0299,
+      "step": 4110
+    },
+    {
+      "epoch": 1.4288191434021154,
+      "grad_norm": 0.08608460426330566,
+      "learning_rate": 0.00011762,
+      "loss": 0.0299,
+      "step": 4120
+    },
+    {
+      "epoch": 1.4322871510317323,
+      "grad_norm": 0.06970153748989105,
+      "learning_rate": 0.00011742,
+      "loss": 0.0345,
+      "step": 4130
+    },
+    {
+      "epoch": 1.435755158661349,
+      "grad_norm": 0.05995609238743782,
+      "learning_rate": 0.00011721999999999999,
+      "loss": 0.0284,
+      "step": 4140
+    },
+    {
+      "epoch": 1.4392231662909658,
+      "grad_norm": 0.03222940117120743,
+      "learning_rate": 0.00011702,
+      "loss": 0.0293,
+      "step": 4150
+    },
+    {
+      "epoch": 1.4426911739205825,
+      "grad_norm": 0.07076498866081238,
+      "learning_rate": 0.00011682,
+      "loss": 0.0362,
+      "step": 4160
+    },
+    {
+      "epoch": 1.4461591815501995,
+      "grad_norm": 0.07425186783075333,
+      "learning_rate": 0.00011661999999999999,
+      "loss": 0.0281,
+      "step": 4170
+    },
+    {
+      "epoch": 1.4496271891798163,
+      "grad_norm": 0.04875819757580757,
+      "learning_rate": 0.00011642,
+      "loss": 0.0361,
+      "step": 4180
+    },
+    {
+      "epoch": 1.453095196809433,
+      "grad_norm": 0.06577154994010925,
+      "learning_rate": 0.00011622,
+      "loss": 0.0291,
+      "step": 4190
+    },
+    {
+      "epoch": 1.4565632044390497,
+      "grad_norm": 0.08174604177474976,
+      "learning_rate": 0.00011601999999999999,
+      "loss": 0.0284,
+      "step": 4200
+    },
+    {
+      "epoch": 1.4600312120686665,
+      "grad_norm": 0.08212857693433762,
+      "learning_rate": 0.00011582,
+      "loss": 0.03,
+      "step": 4210
+    },
+    {
+      "epoch": 1.4634992196982832,
+      "grad_norm": 0.06090838089585304,
+      "learning_rate": 0.00011562,
+      "loss": 0.028,
+      "step": 4220
+    },
+    {
+      "epoch": 1.4669672273279002,
+      "grad_norm": 0.07029874622821808,
+      "learning_rate": 0.00011541999999999999,
+      "loss": 0.033,
+      "step": 4230
+    },
+    {
+      "epoch": 1.470435234957517,
+      "grad_norm": 0.10370688140392303,
+      "learning_rate": 0.00011522,
+      "loss": 0.0341,
+      "step": 4240
+    },
+    {
+      "epoch": 1.4739032425871337,
+      "grad_norm": 0.06497831642627716,
+      "learning_rate": 0.00011501999999999999,
+      "loss": 0.0355,
+      "step": 4250
+    },
+    {
+      "epoch": 1.4773712502167504,
+      "grad_norm": 0.047859255224466324,
+      "learning_rate": 0.00011482000000000002,
+      "loss": 0.0312,
+      "step": 4260
+    },
+    {
+      "epoch": 1.4808392578463674,
+      "grad_norm": 0.044814929366111755,
+      "learning_rate": 0.00011462000000000001,
+      "loss": 0.0325,
+      "step": 4270
+    },
+    {
+      "epoch": 1.4843072654759841,
+      "grad_norm": 0.08687663078308105,
+      "learning_rate": 0.00011442000000000002,
+      "loss": 0.0346,
+      "step": 4280
+    },
+    {
+      "epoch": 1.4877752731056009,
+      "grad_norm": 0.06687606126070023,
+      "learning_rate": 0.00011422000000000001,
+      "loss": 0.0318,
+      "step": 4290
+    },
+    {
+      "epoch": 1.4912432807352176,
+      "grad_norm": 0.0769667997956276,
+      "learning_rate": 0.00011402000000000001,
+      "loss": 0.0314,
+      "step": 4300
+    },
+    {
+      "epoch": 1.4947112883648344,
+      "grad_norm": 0.07912110537290573,
+      "learning_rate": 0.00011382000000000002,
+      "loss": 0.0269,
+      "step": 4310
+    },
+    {
+      "epoch": 1.498179295994451,
+      "grad_norm": 0.06801219284534454,
+      "learning_rate": 0.00011362000000000001,
+      "loss": 0.032,
+      "step": 4320
+    },
+    {
+      "epoch": 1.5016473036240678,
+      "grad_norm": 0.07353610545396805,
+      "learning_rate": 0.00011342000000000001,
+      "loss": 0.0328,
+      "step": 4330
+    },
+    {
+      "epoch": 1.5051153112536848,
+      "grad_norm": 0.05926644429564476,
+      "learning_rate": 0.00011322000000000002,
+      "loss": 0.0268,
+      "step": 4340
+    },
+    {
+      "epoch": 1.5085833188833015,
+      "grad_norm": 0.07942460477352142,
+      "learning_rate": 0.00011302000000000001,
+      "loss": 0.0299,
+      "step": 4350
+    },
+    {
+      "epoch": 1.5120513265129183,
+      "grad_norm": 0.09032566100358963,
+      "learning_rate": 0.00011282000000000002,
+      "loss": 0.0344,
+      "step": 4360
+    },
+    {
+      "epoch": 1.5155193341425353,
+      "grad_norm": 0.08992986381053925,
+      "learning_rate": 0.00011262000000000002,
+      "loss": 0.0282,
+      "step": 4370
+    },
+    {
+      "epoch": 1.518987341772152,
+      "grad_norm": 0.10055962204933167,
+      "learning_rate": 0.00011242000000000001,
+      "loss": 0.0368,
+      "step": 4380
+    },
+    {
+      "epoch": 1.5224553494017687,
+      "grad_norm": 0.06706701964139938,
+      "learning_rate": 0.00011222000000000002,
+      "loss": 0.0348,
+      "step": 4390
+    },
+    {
+      "epoch": 1.5259233570313855,
+      "grad_norm": 0.07412678748369217,
+      "learning_rate": 0.00011202000000000002,
+      "loss": 0.0322,
+      "step": 4400
+    },
+    {
+      "epoch": 1.5293913646610022,
+      "grad_norm": 0.0761900544166565,
+      "learning_rate": 0.00011182000000000001,
+      "loss": 0.0334,
+      "step": 4410
+    },
+    {
+      "epoch": 1.532859372290619,
+      "grad_norm": 0.06172578036785126,
+      "learning_rate": 0.00011162000000000002,
+      "loss": 0.0326,
+      "step": 4420
+    },
+    {
+      "epoch": 1.5363273799202357,
+      "grad_norm": 0.06953331083059311,
+      "learning_rate": 0.00011142000000000001,
+      "loss": 0.0298,
+      "step": 4430
+    },
+    {
+      "epoch": 1.5397953875498525,
+      "grad_norm": 0.07618329674005508,
+      "learning_rate": 0.00011122000000000001,
+      "loss": 0.0283,
+      "step": 4440
+    },
+    {
+      "epoch": 1.5432633951794694,
+      "grad_norm": 0.07265307009220123,
+      "learning_rate": 0.00011102000000000002,
+      "loss": 0.032,
+      "step": 4450
+    },
+    {
+      "epoch": 1.5467314028090862,
+      "grad_norm": 0.07409724593162537,
+      "learning_rate": 0.00011082000000000001,
+      "loss": 0.0303,
+      "step": 4460
+    },
+    {
+      "epoch": 1.5501994104387031,
+      "grad_norm": 0.05352557823061943,
+      "learning_rate": 0.00011062000000000001,
+      "loss": 0.0304,
+      "step": 4470
+    },
+    {
+      "epoch": 1.5536674180683199,
+      "grad_norm": 0.10606401413679123,
+      "learning_rate": 0.00011042000000000002,
+      "loss": 0.0338,
+      "step": 4480
+    },
+    {
+      "epoch": 1.5571354256979366,
+      "grad_norm": 0.07364092022180557,
+      "learning_rate": 0.00011022000000000001,
+      "loss": 0.0308,
+      "step": 4490
+    },
+    {
+      "epoch": 1.5606034333275534,
+      "grad_norm": 0.08737417310476303,
+      "learning_rate": 0.00011002000000000001,
+      "loss": 0.0299,
+      "step": 4500
+    },
+    {
+      "epoch": 1.56407144095717,
+      "grad_norm": 0.07062090188264847,
+      "learning_rate": 0.00010982000000000002,
+      "loss": 0.0314,
+      "step": 4510
+    },
+    {
+      "epoch": 1.5675394485867868,
+      "grad_norm": 0.0711718276143074,
+      "learning_rate": 0.00010962000000000001,
+      "loss": 0.0322,
+      "step": 4520
+    },
+    {
+      "epoch": 1.5710074562164036,
+      "grad_norm": 0.08718711882829666,
+      "learning_rate": 0.00010942,
+      "loss": 0.0297,
+      "step": 4530
+    },
+    {
+      "epoch": 1.5744754638460203,
+      "grad_norm": 0.06502439081668854,
+      "learning_rate": 0.00010922000000000001,
+      "loss": 0.0385,
+      "step": 4540
+    },
+    {
+      "epoch": 1.5779434714756373,
+      "grad_norm": 0.07162761688232422,
+      "learning_rate": 0.00010902000000000001,
+      "loss": 0.0301,
+      "step": 4550
+    },
+    {
+      "epoch": 1.581411479105254,
+      "grad_norm": 0.07721313089132309,
+      "learning_rate": 0.00010882,
+      "loss": 0.0284,
+      "step": 4560
+    },
+    {
+      "epoch": 1.5848794867348708,
+      "grad_norm": 0.07071566581726074,
+      "learning_rate": 0.00010862000000000001,
+      "loss": 0.0352,
+      "step": 4570
+    },
+    {
+      "epoch": 1.5883474943644877,
+      "grad_norm": 0.1029210165143013,
+      "learning_rate": 0.00010842000000000001,
+      "loss": 0.0319,
+      "step": 4580
+    },
+    {
+      "epoch": 1.5918155019941045,
+      "grad_norm": 0.05683687701821327,
+      "learning_rate": 0.00010822,
+      "loss": 0.0343,
+      "step": 4590
+    },
+    {
+      "epoch": 1.5952835096237212,
+      "grad_norm": 0.05821290984749794,
+      "learning_rate": 0.00010802000000000001,
+      "loss": 0.039,
+      "step": 4600
+    },
+    {
+      "epoch": 1.598751517253338,
+      "grad_norm": 0.07931312173604965,
+      "learning_rate": 0.00010782000000000001,
+      "loss": 0.0294,
+      "step": 4610
+    },
+    {
+      "epoch": 1.6022195248829547,
+      "grad_norm": 0.06197603419423103,
+      "learning_rate": 0.00010762,
+      "loss": 0.0322,
+      "step": 4620
+    },
+    {
+      "epoch": 1.6056875325125715,
+      "grad_norm": 0.06312838196754456,
+      "learning_rate": 0.00010742000000000001,
+      "loss": 0.0298,
+      "step": 4630
+    },
+    {
+      "epoch": 1.6091555401421882,
+      "grad_norm": 0.07855828106403351,
+      "learning_rate": 0.00010722000000000001,
+      "loss": 0.0303,
+      "step": 4640
+    },
+    {
+      "epoch": 1.612623547771805,
+      "grad_norm": 0.055718790739774704,
+      "learning_rate": 0.00010702000000000002,
+      "loss": 0.0311,
+      "step": 4650
+    },
+    {
+      "epoch": 1.616091555401422,
+      "grad_norm": 0.07305306941270828,
+      "learning_rate": 0.00010682000000000001,
+      "loss": 0.0303,
+      "step": 4660
+    },
+    {
+      "epoch": 1.6195595630310387,
+      "grad_norm": 0.07300154864788055,
+      "learning_rate": 0.00010662,
+      "loss": 0.0301,
+      "step": 4670
+    },
+    {
+      "epoch": 1.6230275706606556,
+      "grad_norm": 0.06121309846639633,
+      "learning_rate": 0.00010642000000000001,
+      "loss": 0.029,
+      "step": 4680
+    },
+    {
+      "epoch": 1.6264955782902724,
+      "grad_norm": 0.060993146151304245,
+      "learning_rate": 0.00010622000000000001,
+      "loss": 0.0297,
+      "step": 4690
+    },
+    {
+      "epoch": 1.629963585919889,
+      "grad_norm": 0.07691816985607147,
+      "learning_rate": 0.00010602,
+      "loss": 0.0339,
+      "step": 4700
+    },
+    {
+      "epoch": 1.6334315935495058,
+      "grad_norm": 0.07278670370578766,
+      "learning_rate": 0.00010582000000000001,
+      "loss": 0.0262,
+      "step": 4710
+    },
+    {
+      "epoch": 1.6368996011791226,
+      "grad_norm": 0.05704551190137863,
+      "learning_rate": 0.00010562000000000001,
+      "loss": 0.028,
+      "step": 4720
+    },
+    {
+      "epoch": 1.6403676088087393,
+      "grad_norm": 0.07973553240299225,
+      "learning_rate": 0.00010542,
+      "loss": 0.0328,
+      "step": 4730
+    },
+    {
+      "epoch": 1.643835616438356,
+      "grad_norm": 0.05720138177275658,
+      "learning_rate": 0.00010522000000000001,
+      "loss": 0.0309,
+      "step": 4740
+    },
+    {
+      "epoch": 1.6473036240679728,
+      "grad_norm": 0.07787197083234787,
+      "learning_rate": 0.00010502000000000001,
+      "loss": 0.0294,
+      "step": 4750
+    },
+    {
+      "epoch": 1.6507716316975898,
+      "grad_norm": 0.08834118396043777,
+      "learning_rate": 0.00010482,
+      "loss": 0.0301,
+      "step": 4760
+    },
+    {
+      "epoch": 1.6542396393272065,
+      "grad_norm": 0.08383214473724365,
+      "learning_rate": 0.00010462000000000001,
+      "loss": 0.0322,
+      "step": 4770
+    },
+    {
+      "epoch": 1.6577076469568233,
+      "grad_norm": 0.08266714960336685,
+      "learning_rate": 0.00010442,
+      "loss": 0.0323,
+      "step": 4780
+    },
+    {
+      "epoch": 1.6611756545864402,
+      "grad_norm": 0.06535809487104416,
+      "learning_rate": 0.00010422,
+      "loss": 0.0334,
+      "step": 4790
+    },
+    {
+      "epoch": 1.664643662216057,
+      "grad_norm": 0.07224865257740021,
+      "learning_rate": 0.00010402000000000001,
+      "loss": 0.0312,
+      "step": 4800
+    },
+    {
+      "epoch": 1.6681116698456737,
+      "grad_norm": 0.07816470414400101,
+      "learning_rate": 0.00010382,
+      "loss": 0.0342,
+      "step": 4810
+    },
+    {
+      "epoch": 1.6715796774752905,
+      "grad_norm": 0.48284128308296204,
+      "learning_rate": 0.00010362,
+      "loss": 0.0387,
+      "step": 4820
+    },
+    {
+      "epoch": 1.6750476851049072,
+      "grad_norm": 0.15331751108169556,
+      "learning_rate": 0.00010342000000000001,
+      "loss": 0.0289,
+      "step": 4830
+    },
+    {
+      "epoch": 1.678515692734524,
+      "grad_norm": 0.08506326377391815,
+      "learning_rate": 0.00010322,
+      "loss": 0.0299,
+      "step": 4840
+    },
+    {
+      "epoch": 1.6819837003641407,
+      "grad_norm": 0.06808125227689743,
+      "learning_rate": 0.00010302,
+      "loss": 0.0266,
+      "step": 4850
+    },
+    {
+      "epoch": 1.6854517079937574,
+      "grad_norm": 0.07349207252264023,
+      "learning_rate": 0.00010282000000000001,
+      "loss": 0.0297,
+      "step": 4860
+    },
+    {
+      "epoch": 1.6889197156233744,
+      "grad_norm": 0.3372306525707245,
+      "learning_rate": 0.00010262,
+      "loss": 0.0342,
+      "step": 4870
+    },
+    {
+      "epoch": 1.6923877232529911,
+      "grad_norm": 0.10550106316804886,
+      "learning_rate": 0.00010242,
+      "loss": 0.0338,
+      "step": 4880
+    },
+    {
+      "epoch": 1.695855730882608,
+      "grad_norm": 0.0708487331867218,
+      "learning_rate": 0.00010222000000000001,
+      "loss": 0.0294,
+      "step": 4890
+    },
+    {
+      "epoch": 1.6993237385122248,
+      "grad_norm": 0.5124090313911438,
+      "learning_rate": 0.00010202,
+      "loss": 0.0321,
+      "step": 4900
+    },
+    {
+      "epoch": 1.7027917461418416,
+      "grad_norm": 0.08598774671554565,
+      "learning_rate": 0.00010182,
+      "loss": 0.0326,
+      "step": 4910
+    },
+    {
+      "epoch": 1.7062597537714583,
+      "grad_norm": 0.06315886229276657,
+      "learning_rate": 0.00010162,
+      "loss": 0.0358,
+      "step": 4920
+    },
+    {
+      "epoch": 1.709727761401075,
+      "grad_norm": 0.2513497769832611,
+      "learning_rate": 0.00010142,
+      "loss": 0.0352,
+      "step": 4930
+    },
+    {
+      "epoch": 1.7131957690306918,
+      "grad_norm": 0.06378067284822464,
+      "learning_rate": 0.00010122000000000001,
+      "loss": 0.0277,
+      "step": 4940
+    },
+    {
+      "epoch": 1.7166637766603086,
+      "grad_norm": 0.08322855085134506,
+      "learning_rate": 0.00010102,
+      "loss": 0.0303,
+      "step": 4950
+    },
+    {
+      "epoch": 1.7201317842899253,
+      "grad_norm": 0.1097235381603241,
+      "learning_rate": 0.00010082,
+      "loss": 0.0328,
+      "step": 4960
+    },
+    {
+      "epoch": 1.7235997919195423,
+      "grad_norm": 0.06668414920568466,
+      "learning_rate": 0.00010062000000000001,
+      "loss": 0.034,
+      "step": 4970
+    },
+    {
+      "epoch": 1.727067799549159,
+      "grad_norm": 0.08046268671751022,
+      "learning_rate": 0.00010042,
+      "loss": 0.0315,
+      "step": 4980
+    },
+    {
+      "epoch": 1.7305358071787758,
+      "grad_norm": 0.06834772229194641,
+      "learning_rate": 0.00010022,
+      "loss": 0.0304,
+      "step": 4990
+    },
+    {
+      "epoch": 1.7340038148083927,
+      "grad_norm": 0.06567910313606262,
+      "learning_rate": 0.00010002000000000001,
+      "loss": 0.0338,
+      "step": 5000
+    },
+    {
+      "epoch": 1.7340038148083927,
+      "eval_loss": 0.059129249304533005,
+      "eval_runtime": 713.2468,
+      "eval_samples_per_second": 12.874,
+      "eval_steps_per_second": 1.61,
+      "step": 5000
+    },
+    {
+      "epoch": 1.7374718224380095,
+      "grad_norm": 0.0854811891913414,
+      "learning_rate": 9.982e-05,
+      "loss": 0.0308,
+      "step": 5010
+    },
+    {
+      "epoch": 1.7409398300676262,
+      "grad_norm": 0.07272527366876602,
+      "learning_rate": 9.962e-05,
+      "loss": 0.0333,
+      "step": 5020
+    },
+    {
+      "epoch": 1.744407837697243,
+      "grad_norm": 0.0846826583147049,
+      "learning_rate": 9.942000000000001e-05,
+      "loss": 0.0342,
+      "step": 5030
+    },
+    {
+      "epoch": 1.7478758453268597,
+      "grad_norm": 0.06776320934295654,
+      "learning_rate": 9.922e-05,
+      "loss": 0.0333,
+      "step": 5040
+    },
+    {
+      "epoch": 1.7513438529564764,
+      "grad_norm": 0.07815729081630707,
+      "learning_rate": 9.902e-05,
+      "loss": 0.0355,
+      "step": 5050
+    },
+    {
+      "epoch": 1.7548118605860932,
+      "grad_norm": 0.13363681733608246,
+      "learning_rate": 9.882e-05,
+      "loss": 0.032,
+      "step": 5060
+    },
+    {
+      "epoch": 1.75827986821571,
+      "grad_norm": 0.05876624956727028,
+      "learning_rate": 9.862e-05,
+      "loss": 0.033,
+      "step": 5070
+    },
+    {
+      "epoch": 1.7617478758453269,
+      "grad_norm": 0.10773160308599472,
+      "learning_rate": 9.842e-05,
+      "loss": 0.0309,
+      "step": 5080
+    },
+    {
+      "epoch": 1.7652158834749436,
+      "grad_norm": 0.05928561091423035,
+      "learning_rate": 9.822e-05,
+      "loss": 0.0288,
+      "step": 5090
+    },
+    {
+      "epoch": 1.7686838911045606,
+      "grad_norm": 0.058999065309762955,
+      "learning_rate": 9.802e-05,
+      "loss": 0.0251,
+      "step": 5100
+    },
+    {
+      "epoch": 1.7721518987341773,
+      "grad_norm": 0.0768052414059639,
+      "learning_rate": 9.782e-05,
+      "loss": 0.0292,
+      "step": 5110
+    },
+    {
+      "epoch": 1.775619906363794,
+      "grad_norm": 0.1691245585680008,
+      "learning_rate": 9.762e-05,
+      "loss": 0.0327,
+      "step": 5120
+    },
+    {
+      "epoch": 1.7790879139934108,
+      "grad_norm": 0.08563978224992752,
+      "learning_rate": 9.742e-05,
+      "loss": 0.0321,
+      "step": 5130
+    },
+    {
+      "epoch": 1.7825559216230276,
+      "grad_norm": 0.07125357538461685,
+      "learning_rate": 9.722e-05,
+      "loss": 0.0281,
+      "step": 5140
+    },
+    {
+      "epoch": 1.7860239292526443,
+      "grad_norm": 0.10519967973232269,
+      "learning_rate": 9.702e-05,
+      "loss": 0.03,
+      "step": 5150
+    },
+    {
+      "epoch": 1.789491936882261,
+      "grad_norm": 0.08440076559782028,
+      "learning_rate": 9.682e-05,
+      "loss": 0.0309,
+      "step": 5160
+    },
+    {
+      "epoch": 1.7929599445118778,
+      "grad_norm": 0.09944937378168106,
+      "learning_rate": 9.661999999999999e-05,
+      "loss": 0.0323,
+      "step": 5170
+    },
+    {
+      "epoch": 1.7964279521414948,
+      "grad_norm": 0.07523104548454285,
+      "learning_rate": 9.642e-05,
+      "loss": 0.026,
+      "step": 5180
+    },
+    {
+      "epoch": 1.7998959597711115,
+      "grad_norm": 0.09478747099637985,
+      "learning_rate": 9.622000000000001e-05,
+      "loss": 0.0304,
+      "step": 5190
+    },
+    {
+      "epoch": 1.8033639674007282,
+      "grad_norm": 0.0627417042851448,
+      "learning_rate": 9.602e-05,
+      "loss": 0.0353,
+      "step": 5200
+    },
+    {
+      "epoch": 1.8068319750303452,
+      "grad_norm": 0.06294772773981094,
+      "learning_rate": 9.582000000000001e-05,
+      "loss": 0.0301,
+      "step": 5210
+    },
+    {
+      "epoch": 1.810299982659962,
+      "grad_norm": 0.06453125923871994,
+      "learning_rate": 9.562000000000001e-05,
+      "loss": 0.0372,
+      "step": 5220
+    },
+    {
+      "epoch": 1.8137679902895787,
+      "grad_norm": 0.08603645861148834,
+      "learning_rate": 9.542e-05,
+      "loss": 0.0313,
+      "step": 5230
+    },
+    {
+      "epoch": 1.8172359979191954,
+      "grad_norm": 0.0609930120408535,
+      "learning_rate": 9.522000000000001e-05,
+      "loss": 0.0337,
+      "step": 5240
+    },
+    {
+      "epoch": 1.8207040055488122,
+      "grad_norm": 0.06789145618677139,
+      "learning_rate": 9.502000000000001e-05,
+      "loss": 0.0291,
+      "step": 5250
+    },
+    {
+      "epoch": 1.824172013178429,
+      "grad_norm": 0.062086399644613266,
+      "learning_rate": 9.482e-05,
+      "loss": 0.0284,
+      "step": 5260
+    },
+    {
+      "epoch": 1.8276400208080457,
+      "grad_norm": 0.0755184143781662,
+      "learning_rate": 9.462000000000001e-05,
+      "loss": 0.0311,
+      "step": 5270
+    },
+    {
+      "epoch": 1.8311080284376624,
+      "grad_norm": 0.06724268943071365,
+      "learning_rate": 9.442000000000001e-05,
+      "loss": 0.0288,
+      "step": 5280
+    },
+    {
+      "epoch": 1.8345760360672794,
+      "grad_norm": 0.08856779336929321,
+      "learning_rate": 9.422e-05,
+      "loss": 0.0313,
+      "step": 5290
+    },
+    {
+      "epoch": 1.8380440436968961,
+      "grad_norm": 0.08593250811100006,
+      "learning_rate": 9.402000000000001e-05,
+      "loss": 0.0299,
+      "step": 5300
+    },
+    {
+      "epoch": 1.841512051326513,
+      "grad_norm": 0.07944291085004807,
+      "learning_rate": 9.382e-05,
+      "loss": 0.0309,
+      "step": 5310
+    },
+    {
+      "epoch": 1.8449800589561298,
+      "grad_norm": 0.06297358870506287,
+      "learning_rate": 9.362e-05,
+      "loss": 0.0322,
+      "step": 5320
+    },
+    {
+      "epoch": 1.8484480665857466,
+      "grad_norm": 0.0893145203590393,
+      "learning_rate": 9.342000000000001e-05,
+      "loss": 0.0297,
+      "step": 5330
+    },
+    {
+      "epoch": 1.8519160742153633,
+      "grad_norm": 0.06626788526773453,
+      "learning_rate": 9.322e-05,
+      "loss": 0.0359,
+      "step": 5340
+    },
+    {
+      "epoch": 1.85538408184498,
+      "grad_norm": 0.10941380262374878,
+      "learning_rate": 9.302e-05,
+      "loss": 0.0322,
+      "step": 5350
+    },
+    {
+      "epoch": 1.8588520894745968,
+      "grad_norm": 0.08435889333486557,
+      "learning_rate": 9.282000000000001e-05,
+      "loss": 0.0361,
+      "step": 5360
+    },
+    {
+      "epoch": 1.8623200971042135,
+      "grad_norm": 0.08986232429742813,
+      "learning_rate": 9.262e-05,
+      "loss": 0.0288,
+      "step": 5370
+    },
+    {
+      "epoch": 1.8657881047338303,
+      "grad_norm": 0.0800371989607811,
+      "learning_rate": 9.242000000000001e-05,
+      "loss": 0.0339,
+      "step": 5380
+    },
+    {
+      "epoch": 1.8692561123634472,
+      "grad_norm": 0.08191009610891342,
+      "learning_rate": 9.222000000000001e-05,
+      "loss": 0.0284,
+      "step": 5390
+    },
+    {
+      "epoch": 1.872724119993064,
+      "grad_norm": 0.10277281701564789,
+      "learning_rate": 9.202e-05,
+      "loss": 0.0307,
+      "step": 5400
+    },
+    {
+      "epoch": 1.8761921276226807,
+      "grad_norm": 0.08141244202852249,
+      "learning_rate": 9.182000000000001e-05,
+      "loss": 0.0327,
+      "step": 5410
+    },
+    {
+      "epoch": 1.8796601352522977,
+      "grad_norm": 0.07554444670677185,
+      "learning_rate": 9.162000000000001e-05,
+      "loss": 0.031,
+      "step": 5420
+    },
+    {
+      "epoch": 1.8831281428819144,
+      "grad_norm": 0.0898871198296547,
+      "learning_rate": 9.142e-05,
+      "loss": 0.0306,
+      "step": 5430
+    },
+    {
+      "epoch": 1.8865961505115312,
+      "grad_norm": 0.06314833462238312,
+      "learning_rate": 9.122000000000001e-05,
+      "loss": 0.0348,
+      "step": 5440
+    },
+    {
+      "epoch": 1.890064158141148,
+      "grad_norm": 0.06320305913686752,
+      "learning_rate": 9.102e-05,
+      "loss": 0.0277,
+      "step": 5450
+    },
+    {
+      "epoch": 1.8935321657707647,
+      "grad_norm": 0.09586924314498901,
+      "learning_rate": 9.082e-05,
+      "loss": 0.0361,
+      "step": 5460
+    },
+    {
+      "epoch": 1.8970001734003814,
+      "grad_norm": 0.0577247217297554,
+      "learning_rate": 9.062000000000001e-05,
+      "loss": 0.0295,
+      "step": 5470
+    },
+    {
+      "epoch": 1.9004681810299981,
+      "grad_norm": 0.06483156979084015,
+      "learning_rate": 9.042e-05,
+      "loss": 0.0311,
+      "step": 5480
+    },
+    {
+      "epoch": 1.903936188659615,
+      "grad_norm": 0.0748353898525238,
+      "learning_rate": 9.022e-05,
+      "loss": 0.0324,
+      "step": 5490
+    },
+    {
+      "epoch": 1.9074041962892319,
+      "grad_norm": 0.09499184042215347,
+      "learning_rate": 9.002000000000001e-05,
+      "loss": 0.0307,
+      "step": 5500
+    },
+    {
+      "epoch": 1.9108722039188486,
+      "grad_norm": 0.07709678262472153,
+      "learning_rate": 8.982e-05,
+      "loss": 0.0337,
+      "step": 5510
+    },
+    {
+      "epoch": 1.9143402115484656,
+      "grad_norm": 0.06375749409198761,
+      "learning_rate": 8.962e-05,
+      "loss": 0.0231,
+      "step": 5520
+    },
+    {
+      "epoch": 1.9178082191780823,
+      "grad_norm": 0.04287609085440636,
+      "learning_rate": 8.942000000000001e-05,
+      "loss": 0.0243,
+      "step": 5530
+    },
+    {
+      "epoch": 1.921276226807699,
+      "grad_norm": 0.07707373052835464,
+      "learning_rate": 8.922e-05,
+      "loss": 0.0304,
+      "step": 5540
+    },
+    {
+      "epoch": 1.9247442344373158,
+      "grad_norm": 0.06724567711353302,
+      "learning_rate": 8.902e-05,
+      "loss": 0.0323,
+      "step": 5550
+    },
+    {
+      "epoch": 1.9282122420669325,
+      "grad_norm": 0.06895706802606583,
+      "learning_rate": 8.882000000000001e-05,
+      "loss": 0.0346,
+      "step": 5560
+    },
+    {
+      "epoch": 1.9316802496965493,
+      "grad_norm": 0.0870724767446518,
+      "learning_rate": 8.862e-05,
+      "loss": 0.0326,
+      "step": 5570
+    },
+    {
+      "epoch": 1.935148257326166,
+      "grad_norm": 0.061669524759054184,
+      "learning_rate": 8.842e-05,
+      "loss": 0.0287,
+      "step": 5580
+    },
+    {
+      "epoch": 1.9386162649557828,
+      "grad_norm": 0.06450454145669937,
+      "learning_rate": 8.822e-05,
+      "loss": 0.0295,
+      "step": 5590
+    },
+    {
+      "epoch": 1.9420842725853997,
+      "grad_norm": 0.07439760863780975,
+      "learning_rate": 8.802e-05,
+      "loss": 0.0325,
+      "step": 5600
+    },
+    {
+      "epoch": 1.9455522802150165,
+      "grad_norm": 0.08620608597993851,
+      "learning_rate": 8.782e-05,
+      "loss": 0.0306,
+      "step": 5610
+    },
+    {
+      "epoch": 1.9490202878446332,
+      "grad_norm": 0.08417027443647385,
+      "learning_rate": 8.762e-05,
+      "loss": 0.0339,
+      "step": 5620
+    },
+    {
+      "epoch": 1.9524882954742502,
+      "grad_norm": 0.0670836940407753,
+      "learning_rate": 8.742e-05,
+      "loss": 0.0318,
+      "step": 5630
+    },
+    {
+      "epoch": 1.955956303103867,
+      "grad_norm": 0.08096006512641907,
+      "learning_rate": 8.722e-05,
+      "loss": 0.0269,
+      "step": 5640
+    },
+    {
+      "epoch": 1.9594243107334837,
+      "grad_norm": 0.04559866338968277,
+      "learning_rate": 8.702e-05,
+      "loss": 0.0271,
+      "step": 5650
+    },
+    {
+      "epoch": 1.9628923183631004,
+      "grad_norm": 0.08112025260925293,
+      "learning_rate": 8.682e-05,
+      "loss": 0.036,
+      "step": 5660
+    },
+    {
+      "epoch": 1.9663603259927172,
+      "grad_norm": 0.07936326414346695,
+      "learning_rate": 8.662000000000001e-05,
+      "loss": 0.0323,
+      "step": 5670
+    },
+    {
+      "epoch": 1.969828333622334,
+      "grad_norm": 0.09269768744707108,
+      "learning_rate": 8.642e-05,
+      "loss": 0.0318,
+      "step": 5680
+    },
+    {
+      "epoch": 1.9732963412519506,
+      "grad_norm": 0.06654026359319687,
+      "learning_rate": 8.622e-05,
+      "loss": 0.0291,
+      "step": 5690
+    },
+    {
+      "epoch": 1.9767643488815674,
+      "grad_norm": 0.0971643254160881,
+      "learning_rate": 8.602e-05,
+      "loss": 0.031,
+      "step": 5700
+    },
+    {
+      "epoch": 1.9802323565111843,
+      "grad_norm": 0.06067187711596489,
+      "learning_rate": 8.582e-05,
+      "loss": 0.0306,
+      "step": 5710
+    },
+    {
+      "epoch": 1.983700364140801,
+      "grad_norm": 0.0945192202925682,
+      "learning_rate": 8.562e-05,
+      "loss": 0.0328,
+      "step": 5720
+    },
+    {
+      "epoch": 1.987168371770418,
+      "grad_norm": 0.07627417147159576,
+      "learning_rate": 8.542e-05,
+      "loss": 0.0352,
+      "step": 5730
+    },
+    {
+      "epoch": 1.9906363794000348,
+      "grad_norm": 0.08669853955507278,
+      "learning_rate": 8.522e-05,
+      "loss": 0.0337,
+      "step": 5740
+    },
+    {
+      "epoch": 1.9941043870296515,
+      "grad_norm": 0.06610149890184402,
+      "learning_rate": 8.502e-05,
+      "loss": 0.0288,
+      "step": 5750
+    },
+    {
+      "epoch": 1.9975723946592683,
+      "grad_norm": 0.06989070028066635,
+      "learning_rate": 8.482e-05,
+      "loss": 0.0306,
+      "step": 5760
+    },
+    {
+      "epoch": 2.001040402288885,
+      "grad_norm": 0.04675092548131943,
+      "learning_rate": 8.462e-05,
+      "loss": 0.0245,
+      "step": 5770
+    },
+    {
+      "epoch": 2.0045084099185018,
+      "grad_norm": 0.053521353751420975,
+      "learning_rate": 8.442e-05,
+      "loss": 0.0158,
+      "step": 5780
+    },
+    {
+      "epoch": 2.0079764175481185,
+      "grad_norm": 0.06453324854373932,
+      "learning_rate": 8.422e-05,
+      "loss": 0.0131,
+      "step": 5790
+    },
+    {
+      "epoch": 2.0114444251777353,
+      "grad_norm": 0.0763096958398819,
+      "learning_rate": 8.402e-05,
+      "loss": 0.0112,
+      "step": 5800
+    },
+    {
+      "epoch": 2.014912432807352,
+      "grad_norm": 0.0837683454155922,
+      "learning_rate": 8.382e-05,
+      "loss": 0.0135,
+      "step": 5810
+    },
+    {
+      "epoch": 2.018380440436969,
+      "grad_norm": 0.0868675634264946,
+      "learning_rate": 8.362000000000002e-05,
+      "loss": 0.0116,
+      "step": 5820
+    },
+    {
+      "epoch": 2.021848448066586,
+      "grad_norm": 0.06839966773986816,
+      "learning_rate": 8.342000000000001e-05,
+      "loss": 0.011,
+      "step": 5830
+    },
+    {
+      "epoch": 2.0253164556962027,
+      "grad_norm": 0.10859765112400055,
+      "learning_rate": 8.322e-05,
+      "loss": 0.0132,
+      "step": 5840
+    },
+    {
+      "epoch": 2.0287844633258194,
+      "grad_norm": 0.08824854344129562,
+      "learning_rate": 8.302000000000001e-05,
+      "loss": 0.0129,
+      "step": 5850
+    },
+    {
+      "epoch": 2.032252470955436,
+      "grad_norm": 0.0745101124048233,
+      "learning_rate": 8.282000000000001e-05,
+      "loss": 0.013,
+      "step": 5860
+    },
+    {
+      "epoch": 2.035720478585053,
+      "grad_norm": 0.06870684772729874,
+      "learning_rate": 8.262e-05,
+      "loss": 0.0126,
+      "step": 5870
+    },
+    {
+      "epoch": 2.0391884862146696,
+      "grad_norm": 0.08353777974843979,
+      "learning_rate": 8.242000000000001e-05,
+      "loss": 0.0127,
+      "step": 5880
+    },
+    {
+      "epoch": 2.0426564938442864,
+      "grad_norm": 0.07282493263483047,
+      "learning_rate": 8.222000000000001e-05,
+      "loss": 0.014,
+      "step": 5890
+    },
+    {
+      "epoch": 2.046124501473903,
+      "grad_norm": 0.07777410745620728,
+      "learning_rate": 8.202e-05,
+      "loss": 0.0108,
+      "step": 5900
+    },
+    {
+      "epoch": 2.04959250910352,
+      "grad_norm": 0.06361842900514603,
+      "learning_rate": 8.182000000000001e-05,
+      "loss": 0.0132,
+      "step": 5910
+    },
+    {
+      "epoch": 2.0530605167331366,
+      "grad_norm": 0.09724973887205124,
+      "learning_rate": 8.162000000000001e-05,
+      "loss": 0.0139,
+      "step": 5920
+    },
+    {
+      "epoch": 2.056528524362754,
+      "grad_norm": 0.09359490126371384,
+      "learning_rate": 8.142e-05,
+      "loss": 0.013,
+      "step": 5930
+    },
+    {
+      "epoch": 2.0599965319923705,
+      "grad_norm": 0.06116607412695885,
+      "learning_rate": 8.122000000000001e-05,
+      "loss": 0.0143,
+      "step": 5940
+    },
+    {
+      "epoch": 2.0634645396219873,
+      "grad_norm": 0.06787212938070297,
+      "learning_rate": 8.102000000000001e-05,
+      "loss": 0.0137,
+      "step": 5950
+    },
+    {
+      "epoch": 2.066932547251604,
+      "grad_norm": 0.06523068249225616,
+      "learning_rate": 8.082e-05,
+      "loss": 0.0116,
+      "step": 5960
+    },
+    {
+      "epoch": 2.0704005548812208,
+      "grad_norm": 0.09076279401779175,
+      "learning_rate": 8.062000000000001e-05,
+      "loss": 0.01,
+      "step": 5970
+    },
+    {
+      "epoch": 2.0738685625108375,
+      "grad_norm": 0.07997199892997742,
+      "learning_rate": 8.042e-05,
+      "loss": 0.0132,
+      "step": 5980
+    },
+    {
+      "epoch": 2.0773365701404543,
+      "grad_norm": 0.0561593659222126,
+      "learning_rate": 8.022e-05,
+      "loss": 0.0117,
+      "step": 5990
+    },
+    {
+      "epoch": 2.080804577770071,
+      "grad_norm": 0.08588268607854843,
+      "learning_rate": 8.002000000000001e-05,
+      "loss": 0.0125,
+      "step": 6000
+    },
+    {
+      "epoch": 2.080804577770071,
+      "eval_loss": 0.06557230651378632,
+      "eval_runtime": 707.733,
+      "eval_samples_per_second": 12.974,
+      "eval_steps_per_second": 1.622,
+      "step": 6000
+    },
+    {
+      "epoch": 2.0842725853996877,
+      "grad_norm": 0.09114642441272736,
+      "learning_rate": 7.982e-05,
+      "loss": 0.0107,
+      "step": 6010
+    },
+    {
+      "epoch": 2.0877405930293045,
+      "grad_norm": 0.06732100248336792,
+      "learning_rate": 7.962e-05,
+      "loss": 0.012,
+      "step": 6020
+    },
+    {
+      "epoch": 2.0912086006589217,
+      "grad_norm": 0.08021605014801025,
+      "learning_rate": 7.942000000000001e-05,
+      "loss": 0.0123,
+      "step": 6030
+    },
+    {
+      "epoch": 2.0946766082885384,
+      "grad_norm": 0.06761088222265244,
+      "learning_rate": 7.922e-05,
+      "loss": 0.0117,
+      "step": 6040
+    },
+    {
+      "epoch": 2.098144615918155,
+      "grad_norm": 0.11337202787399292,
+      "learning_rate": 7.902e-05,
+      "loss": 0.0131,
+      "step": 6050
+    },
+    {
+      "epoch": 2.101612623547772,
+      "grad_norm": 0.09859013557434082,
+      "learning_rate": 7.882000000000001e-05,
+      "loss": 0.0139,
+      "step": 6060
+    },
+    {
+      "epoch": 2.1050806311773886,
+      "grad_norm": 0.08745191991329193,
+      "learning_rate": 7.862e-05,
+      "loss": 0.0139,
+      "step": 6070
+    },
+    {
+      "epoch": 2.1085486388070054,
+      "grad_norm": 0.04903840273618698,
+      "learning_rate": 7.842e-05,
+      "loss": 0.0142,
+      "step": 6080
+    },
+    {
+      "epoch": 2.112016646436622,
+      "grad_norm": 0.07992135733366013,
+      "learning_rate": 7.822e-05,
+      "loss": 0.0123,
+      "step": 6090
+    },
+    {
+      "epoch": 2.115484654066239,
+      "grad_norm": 0.08518462628126144,
+      "learning_rate": 7.802e-05,
+      "loss": 0.0138,
+      "step": 6100
+    },
+    {
+      "epoch": 2.1189526616958556,
+      "grad_norm": 0.08434431999921799,
+      "learning_rate": 7.782000000000001e-05,
+      "loss": 0.0155,
+      "step": 6110
+    },
+    {
+      "epoch": 2.1224206693254724,
+      "grad_norm": 0.08734823763370514,
+      "learning_rate": 7.762e-05,
+      "loss": 0.0125,
+      "step": 6120
+    },
+    {
+      "epoch": 2.125888676955089,
+      "grad_norm": 0.08129347115755081,
+      "learning_rate": 7.742e-05,
+      "loss": 0.0118,
+      "step": 6130
+    },
+    {
+      "epoch": 2.1293566845847063,
+      "grad_norm": 0.09434698522090912,
+      "learning_rate": 7.722000000000001e-05,
+      "loss": 0.0128,
+      "step": 6140
+    },
+    {
+      "epoch": 2.132824692214323,
+      "grad_norm": 0.08226180821657181,
+      "learning_rate": 7.702e-05,
+      "loss": 0.0159,
+      "step": 6150
+    },
+    {
+      "epoch": 2.1362926998439398,
+      "grad_norm": 0.06351976841688156,
+      "learning_rate": 7.682e-05,
+      "loss": 0.0135,
+      "step": 6160
+    },
+    {
+      "epoch": 2.1397607074735565,
+      "grad_norm": 0.07568191736936569,
+      "learning_rate": 7.662000000000001e-05,
+      "loss": 0.0122,
+      "step": 6170
+    },
+    {
+      "epoch": 2.1432287151031733,
+      "grad_norm": 0.08052569627761841,
+      "learning_rate": 7.642e-05,
+      "loss": 0.0123,
+      "step": 6180
+    },
+    {
+      "epoch": 2.14669672273279,
+      "grad_norm": 0.08767958730459213,
+      "learning_rate": 7.622e-05,
+      "loss": 0.0145,
+      "step": 6190
+    },
+    {
+      "epoch": 2.1501647303624067,
+      "grad_norm": 0.09768091887235641,
+      "learning_rate": 7.602000000000001e-05,
+      "loss": 0.0113,
+      "step": 6200
+    },
+    {
+      "epoch": 2.1536327379920235,
+      "grad_norm": 0.09293138980865479,
+      "learning_rate": 7.582e-05,
+      "loss": 0.0111,
+      "step": 6210
+    },
+    {
+      "epoch": 2.1571007456216402,
+      "grad_norm": 0.07943341881036758,
+      "learning_rate": 7.562e-05,
+      "loss": 0.0133,
+      "step": 6220
+    },
+    {
+      "epoch": 2.160568753251257,
+      "grad_norm": 0.08316273242235184,
+      "learning_rate": 7.542e-05,
+      "loss": 0.0142,
+      "step": 6230
+    },
+    {
+      "epoch": 2.164036760880874,
+      "grad_norm": 0.05390129238367081,
+      "learning_rate": 7.522e-05,
+      "loss": 0.0128,
+      "step": 6240
+    },
+    {
+      "epoch": 2.167504768510491,
+      "grad_norm": 0.05505843088030815,
+      "learning_rate": 7.502e-05,
+      "loss": 0.0124,
+      "step": 6250
+    },
+    {
+      "epoch": 2.1709727761401076,
+      "grad_norm": 0.08427543193101883,
+      "learning_rate": 7.482e-05,
+      "loss": 0.0136,
+      "step": 6260
+    },
+    {
+      "epoch": 2.1744407837697244,
+      "grad_norm": 0.06911098957061768,
+      "learning_rate": 7.462e-05,
+      "loss": 0.0117,
+      "step": 6270
+    },
+    {
+      "epoch": 2.177908791399341,
+      "grad_norm": 0.08595360815525055,
+      "learning_rate": 7.442e-05,
+      "loss": 0.0155,
+      "step": 6280
+    },
+    {
+      "epoch": 2.181376799028958,
+      "grad_norm": 0.0827205702662468,
+      "learning_rate": 7.422e-05,
+      "loss": 0.0127,
+      "step": 6290
+    },
+    {
+      "epoch": 2.1848448066585746,
+      "grad_norm": 0.12057662755250931,
+      "learning_rate": 7.402e-05,
+      "loss": 0.0113,
+      "step": 6300
+    },
+    {
+      "epoch": 2.1883128142881914,
+      "grad_norm": 0.10641255229711533,
+      "learning_rate": 7.382e-05,
+      "loss": 0.0123,
+      "step": 6310
+    },
+    {
+      "epoch": 2.191780821917808,
+      "grad_norm": 0.11188361793756485,
+      "learning_rate": 7.362e-05,
+      "loss": 0.013,
+      "step": 6320
+    },
+    {
+      "epoch": 2.195248829547425,
+      "grad_norm": 0.06386271864175797,
+      "learning_rate": 7.342e-05,
+      "loss": 0.0114,
+      "step": 6330
+    },
+    {
+      "epoch": 2.1987168371770416,
+      "grad_norm": 0.09285665303468704,
+      "learning_rate": 7.322e-05,
+      "loss": 0.0156,
+      "step": 6340
+    },
+    {
+      "epoch": 2.2021848448066588,
+      "grad_norm": 0.0885651484131813,
+      "learning_rate": 7.302e-05,
+      "loss": 0.013,
+      "step": 6350
+    },
+    {
+      "epoch": 2.2056528524362755,
+      "grad_norm": 0.09753404557704926,
+      "learning_rate": 7.282e-05,
+      "loss": 0.0135,
+      "step": 6360
+    },
+    {
+      "epoch": 2.2091208600658923,
+      "grad_norm": 0.06870284676551819,
+      "learning_rate": 7.261999999999999e-05,
+      "loss": 0.0163,
+      "step": 6370
+    },
+    {
+      "epoch": 2.212588867695509,
+      "grad_norm": 0.07846927642822266,
+      "learning_rate": 7.242e-05,
+      "loss": 0.0147,
+      "step": 6380
+    },
+    {
+      "epoch": 2.2160568753251257,
+      "grad_norm": 0.06510089337825775,
+      "learning_rate": 7.222e-05,
+      "loss": 0.0129,
+      "step": 6390
+    },
+    {
+      "epoch": 2.2195248829547425,
+      "grad_norm": 0.13771073520183563,
+      "learning_rate": 7.202e-05,
+      "loss": 0.0151,
+      "step": 6400
+    },
+    {
+      "epoch": 2.2229928905843592,
+      "grad_norm": 0.08157498389482498,
+      "learning_rate": 7.182e-05,
+      "loss": 0.0136,
+      "step": 6410
+    },
+    {
+      "epoch": 2.226460898213976,
+      "grad_norm": 0.09444098174571991,
+      "learning_rate": 7.162e-05,
+      "loss": 0.016,
+      "step": 6420
+    },
+    {
+      "epoch": 2.2299289058435927,
+      "grad_norm": 0.0773581713438034,
+      "learning_rate": 7.142e-05,
+      "loss": 0.0132,
+      "step": 6430
+    },
+    {
+      "epoch": 2.2333969134732095,
+      "grad_norm": 0.10038639605045319,
+      "learning_rate": 7.122000000000001e-05,
+      "loss": 0.0142,
+      "step": 6440
+    },
+    {
+      "epoch": 2.2368649211028266,
+      "grad_norm": 0.10728047043085098,
+      "learning_rate": 7.102000000000001e-05,
+      "loss": 0.0152,
+      "step": 6450
+    },
+    {
+      "epoch": 2.2403329287324434,
+      "grad_norm": 0.10695026069879532,
+      "learning_rate": 7.082e-05,
+      "loss": 0.0163,
+      "step": 6460
+    },
+    {
+      "epoch": 2.24380093636206,
+      "grad_norm": 0.06996133178472519,
+      "learning_rate": 7.062000000000001e-05,
+      "loss": 0.0149,
+      "step": 6470
+    },
+    {
+      "epoch": 2.247268943991677,
+      "grad_norm": 0.04395058751106262,
+      "learning_rate": 7.042000000000001e-05,
+      "loss": 0.0127,
+      "step": 6480
+    },
+    {
+      "epoch": 2.2507369516212936,
+      "grad_norm": 0.06262753158807755,
+      "learning_rate": 7.022e-05,
+      "loss": 0.014,
+      "step": 6490
+    },
+    {
+      "epoch": 2.2542049592509104,
+      "grad_norm": 0.073272705078125,
+      "learning_rate": 7.002000000000001e-05,
+      "loss": 0.0121,
+      "step": 6500
+    },
+    {
+      "epoch": 2.257672966880527,
+      "grad_norm": 0.11415940523147583,
+      "learning_rate": 6.982e-05,
+      "loss": 0.0136,
+      "step": 6510
+    },
+    {
+      "epoch": 2.261140974510144,
+      "grad_norm": 0.09325289726257324,
+      "learning_rate": 6.962e-05,
+      "loss": 0.0121,
+      "step": 6520
+    },
+    {
+      "epoch": 2.2646089821397606,
+      "grad_norm": 0.07223498821258545,
+      "learning_rate": 6.942000000000001e-05,
+      "loss": 0.0143,
+      "step": 6530
+    },
+    {
+      "epoch": 2.2680769897693773,
+      "grad_norm": 0.08595094084739685,
+      "learning_rate": 6.922e-05,
+      "loss": 0.0137,
+      "step": 6540
+    },
+    {
+      "epoch": 2.271544997398994,
+      "grad_norm": 0.08120746910572052,
+      "learning_rate": 6.902000000000001e-05,
+      "loss": 0.0101,
+      "step": 6550
+    },
+    {
+      "epoch": 2.2750130050286113,
+      "grad_norm": 0.1085987389087677,
+      "learning_rate": 6.882000000000001e-05,
+      "loss": 0.0149,
+      "step": 6560
+    },
+    {
+      "epoch": 2.278481012658228,
+      "grad_norm": 0.07946083694696426,
+      "learning_rate": 6.862e-05,
+      "loss": 0.0133,
+      "step": 6570
+    },
+    {
+      "epoch": 2.2819490202878447,
+      "grad_norm": 0.05504854768514633,
+      "learning_rate": 6.842000000000001e-05,
+      "loss": 0.0139,
+      "step": 6580
+    },
+    {
+      "epoch": 2.2854170279174615,
+      "grad_norm": 0.07158561050891876,
+      "learning_rate": 6.822000000000001e-05,
+      "loss": 0.0135,
+      "step": 6590
+    },
+    {
+      "epoch": 2.2888850355470782,
+      "grad_norm": 0.06974880397319794,
+      "learning_rate": 6.802e-05,
+      "loss": 0.0123,
+      "step": 6600
+    },
+    {
+      "epoch": 2.292353043176695,
+      "grad_norm": 0.08541780710220337,
+      "learning_rate": 6.782000000000001e-05,
+      "loss": 0.0144,
+      "step": 6610
+    },
+    {
+      "epoch": 2.2958210508063117,
+      "grad_norm": 0.10203000158071518,
+      "learning_rate": 6.762e-05,
+      "loss": 0.0138,
+      "step": 6620
+    },
+    {
+      "epoch": 2.2992890584359285,
+      "grad_norm": 0.09905651211738586,
+      "learning_rate": 6.742e-05,
+      "loss": 0.0145,
+      "step": 6630
+    },
+    {
+      "epoch": 2.302757066065545,
+      "grad_norm": 0.04088059067726135,
+      "learning_rate": 6.722000000000001e-05,
+      "loss": 0.011,
+      "step": 6640
+    },
+    {
+      "epoch": 2.306225073695162,
+      "grad_norm": 0.10128574818372726,
+      "learning_rate": 6.702e-05,
+      "loss": 0.0159,
+      "step": 6650
+    },
+    {
+      "epoch": 2.309693081324779,
+      "grad_norm": 0.08455543220043182,
+      "learning_rate": 6.682e-05,
+      "loss": 0.0131,
+      "step": 6660
+    },
+    {
+      "epoch": 2.313161088954396,
+      "grad_norm": 0.1268063485622406,
+      "learning_rate": 6.662000000000001e-05,
+      "loss": 0.0129,
+      "step": 6670
+    },
+    {
+      "epoch": 2.3166290965840126,
+      "grad_norm": 0.11783897876739502,
+      "learning_rate": 6.642e-05,
+      "loss": 0.0106,
+      "step": 6680
+    },
+    {
+      "epoch": 2.3200971042136294,
+      "grad_norm": 0.09608971327543259,
+      "learning_rate": 6.622e-05,
+      "loss": 0.0137,
+      "step": 6690
+    },
+    {
+      "epoch": 2.323565111843246,
+      "grad_norm": 0.07495573163032532,
+      "learning_rate": 6.602000000000001e-05,
+      "loss": 0.0131,
+      "step": 6700
+    },
+    {
+      "epoch": 2.327033119472863,
+      "grad_norm": 0.07772886753082275,
+      "learning_rate": 6.582e-05,
+      "loss": 0.0124,
+      "step": 6710
+    },
+    {
+      "epoch": 2.3305011271024796,
+      "grad_norm": 0.07365009188652039,
+      "learning_rate": 6.562e-05,
+      "loss": 0.0112,
+      "step": 6720
+    },
+    {
+      "epoch": 2.3339691347320963,
+      "grad_norm": 0.0697893276810646,
+      "learning_rate": 6.542000000000001e-05,
+      "loss": 0.0105,
+      "step": 6730
+    },
+    {
+      "epoch": 2.337437142361713,
+      "grad_norm": 0.09057148545980453,
+      "learning_rate": 6.522e-05,
+      "loss": 0.0134,
+      "step": 6740
+    },
+    {
+      "epoch": 2.34090514999133,
+      "grad_norm": 0.09601489454507828,
+      "learning_rate": 6.502e-05,
+      "loss": 0.014,
+      "step": 6750
+    },
+    {
+      "epoch": 2.3443731576209466,
+      "grad_norm": 0.11969607323408127,
+      "learning_rate": 6.482e-05,
+      "loss": 0.0116,
+      "step": 6760
+    },
+    {
+      "epoch": 2.3478411652505637,
+      "grad_norm": 0.103757843375206,
+      "learning_rate": 6.462e-05,
+      "loss": 0.0088,
+      "step": 6770
+    },
+    {
+      "epoch": 2.3513091728801805,
+      "grad_norm": 0.09077152609825134,
+      "learning_rate": 6.442e-05,
+      "loss": 0.0111,
+      "step": 6780
+    },
+    {
+      "epoch": 2.3547771805097972,
+      "grad_norm": 0.06362780928611755,
+      "learning_rate": 6.422e-05,
+      "loss": 0.01,
+      "step": 6790
+    },
+    {
+      "epoch": 2.358245188139414,
+      "grad_norm": 0.10334885120391846,
+      "learning_rate": 6.402e-05,
+      "loss": 0.0144,
+      "step": 6800
+    },
+    {
+      "epoch": 2.3617131957690307,
+      "grad_norm": 0.0676029622554779,
+      "learning_rate": 6.382e-05,
+      "loss": 0.0131,
+      "step": 6810
+    },
+    {
+      "epoch": 2.3651812033986475,
+      "grad_norm": 0.06794017553329468,
+      "learning_rate": 6.362e-05,
+      "loss": 0.0136,
+      "step": 6820
+    },
+    {
+      "epoch": 2.368649211028264,
+      "grad_norm": 0.11005677282810211,
+      "learning_rate": 6.342e-05,
+      "loss": 0.0141,
+      "step": 6830
+    },
+    {
+      "epoch": 2.372117218657881,
+      "grad_norm": 0.07998326420783997,
+      "learning_rate": 6.322000000000001e-05,
+      "loss": 0.0119,
+      "step": 6840
+    },
+    {
+      "epoch": 2.3755852262874977,
+      "grad_norm": 0.07724535465240479,
+      "learning_rate": 6.302e-05,
+      "loss": 0.0149,
+      "step": 6850
+    },
+    {
+      "epoch": 2.3790532339171144,
+      "grad_norm": 0.07563537359237671,
+      "learning_rate": 6.282e-05,
+      "loss": 0.0145,
+      "step": 6860
+    },
+    {
+      "epoch": 2.3825212415467316,
+      "grad_norm": 0.048786722123622894,
+      "learning_rate": 6.262000000000001e-05,
+      "loss": 0.0122,
+      "step": 6870
+    },
+    {
+      "epoch": 2.3859892491763484,
+      "grad_norm": 0.09792380034923553,
+      "learning_rate": 6.242e-05,
+      "loss": 0.0133,
+      "step": 6880
+    },
+    {
+      "epoch": 2.389457256805965,
+      "grad_norm": 0.06414589285850525,
+      "learning_rate": 6.222e-05,
+      "loss": 0.0119,
+      "step": 6890
+    },
+    {
+      "epoch": 2.392925264435582,
+      "grad_norm": 0.0844031348824501,
+      "learning_rate": 6.202e-05,
+      "loss": 0.0139,
+      "step": 6900
+    },
+    {
+      "epoch": 2.3963932720651986,
+      "grad_norm": 0.0652434229850769,
+      "learning_rate": 6.182e-05,
+      "loss": 0.01,
+      "step": 6910
+    },
+    {
+      "epoch": 2.3998612796948153,
+      "grad_norm": 0.07981958985328674,
+      "learning_rate": 6.162e-05,
+      "loss": 0.0105,
+      "step": 6920
+    },
+    {
+      "epoch": 2.403329287324432,
+      "grad_norm": 0.064891017973423,
+      "learning_rate": 6.142e-05,
+      "loss": 0.0138,
+      "step": 6930
+    },
+    {
+      "epoch": 2.406797294954049,
+      "grad_norm": 0.07090698927640915,
+      "learning_rate": 6.122e-05,
+      "loss": 0.0103,
+      "step": 6940
+    },
+    {
+      "epoch": 2.4102653025836656,
+      "grad_norm": 0.09227363765239716,
+      "learning_rate": 6.102e-05,
+      "loss": 0.0146,
+      "step": 6950
+    },
+    {
+      "epoch": 2.4137333102132823,
+      "grad_norm": 0.056207556277513504,
+      "learning_rate": 6.082e-05,
+      "loss": 0.0143,
+      "step": 6960
+    },
+    {
+      "epoch": 2.417201317842899,
+      "grad_norm": 0.08221688121557236,
+      "learning_rate": 6.062e-05,
+      "loss": 0.0158,
+      "step": 6970
+    },
+    {
+      "epoch": 2.4206693254725162,
+      "grad_norm": 0.08280789107084274,
+      "learning_rate": 6.042e-05,
+      "loss": 0.0139,
+      "step": 6980
+    },
+    {
+      "epoch": 2.424137333102133,
+      "grad_norm": 0.06660090386867523,
+      "learning_rate": 6.0219999999999996e-05,
+      "loss": 0.013,
+      "step": 6990
+    },
+    {
+      "epoch": 2.4276053407317497,
+      "grad_norm": 0.10185576230287552,
+      "learning_rate": 6.002e-05,
+      "loss": 0.0153,
+      "step": 7000
+    },
+    {
+      "epoch": 2.4276053407317497,
+      "eval_loss": 0.0646032989025116,
+      "eval_runtime": 713.661,
+      "eval_samples_per_second": 12.866,
+      "eval_steps_per_second": 1.609,
+      "step": 7000
+    },
+    {
+      "epoch": 2.4310733483613665,
+      "grad_norm": 0.11863771080970764,
+      "learning_rate": 5.982e-05,
+      "loss": 0.0149,
+      "step": 7010
+    },
+    {
+      "epoch": 2.434541355990983,
+      "grad_norm": 0.08282611519098282,
+      "learning_rate": 5.9619999999999995e-05,
+      "loss": 0.0114,
+      "step": 7020
+    },
+    {
+      "epoch": 2.4380093636206,
+      "grad_norm": 0.10327205806970596,
+      "learning_rate": 5.942e-05,
+      "loss": 0.0113,
+      "step": 7030
+    },
+    {
+      "epoch": 2.4414773712502167,
+      "grad_norm": 0.07424825429916382,
+      "learning_rate": 5.922e-05,
+      "loss": 0.0149,
+      "step": 7040
+    },
+    {
+      "epoch": 2.4449453788798334,
+      "grad_norm": 0.07107970863580704,
+      "learning_rate": 5.902e-05,
+      "loss": 0.0133,
+      "step": 7050
+    },
+    {
+      "epoch": 2.44841338650945,
+      "grad_norm": 0.08729968965053558,
+      "learning_rate": 5.8819999999999996e-05,
+      "loss": 0.0136,
+      "step": 7060
+    },
+    {
+      "epoch": 2.451881394139067,
+      "grad_norm": 0.06300070136785507,
+      "learning_rate": 5.862000000000001e-05,
+      "loss": 0.01,
+      "step": 7070
+    },
+    {
+      "epoch": 2.455349401768684,
+      "grad_norm": 0.107483871281147,
+      "learning_rate": 5.8420000000000006e-05,
+      "loss": 0.0124,
+      "step": 7080
+    },
+    {
+      "epoch": 2.458817409398301,
+      "grad_norm": 0.06871318072080612,
+      "learning_rate": 5.822000000000001e-05,
+      "loss": 0.013,
+      "step": 7090
+    },
+    {
+      "epoch": 2.4622854170279176,
+      "grad_norm": 0.10566007345914841,
+      "learning_rate": 5.802000000000001e-05,
+      "loss": 0.0133,
+      "step": 7100
+    },
+    {
+      "epoch": 2.4657534246575343,
+      "grad_norm": 0.04511050507426262,
+      "learning_rate": 5.7820000000000005e-05,
+      "loss": 0.012,
+      "step": 7110
+    },
+    {
+      "epoch": 2.469221432287151,
+      "grad_norm": 0.09614109992980957,
+      "learning_rate": 5.762000000000001e-05,
+      "loss": 0.0115,
+      "step": 7120
+    },
+    {
+      "epoch": 2.472689439916768,
+      "grad_norm": 0.07863055169582367,
+      "learning_rate": 5.742000000000001e-05,
+      "loss": 0.0141,
+      "step": 7130
+    },
+    {
+      "epoch": 2.4761574475463846,
+      "grad_norm": 0.09857816249132156,
+      "learning_rate": 5.7220000000000004e-05,
+      "loss": 0.0131,
+      "step": 7140
+    },
+    {
+      "epoch": 2.4796254551760013,
+      "grad_norm": 0.11649773269891739,
+      "learning_rate": 5.7020000000000006e-05,
+      "loss": 0.014,
+      "step": 7150
+    },
+    {
+      "epoch": 2.483093462805618,
+      "grad_norm": 0.09917131811380386,
+      "learning_rate": 5.682000000000001e-05,
+      "loss": 0.0119,
+      "step": 7160
+    },
+    {
+      "epoch": 2.486561470435235,
+      "grad_norm": 0.090948186814785,
+      "learning_rate": 5.6620000000000003e-05,
+      "loss": 0.0131,
+      "step": 7170
+    },
+    {
+      "epoch": 2.4900294780648515,
+      "grad_norm": 0.10430017858743668,
+      "learning_rate": 5.6420000000000005e-05,
+      "loss": 0.0122,
+      "step": 7180
+    },
+    {
+      "epoch": 2.4934974856944687,
+      "grad_norm": 0.08023589104413986,
+      "learning_rate": 5.622000000000001e-05,
+      "loss": 0.0146,
+      "step": 7190
+    },
+    {
+      "epoch": 2.4969654933240855,
+      "grad_norm": 0.058440957218408585,
+      "learning_rate": 5.602000000000001e-05,
+      "loss": 0.0125,
+      "step": 7200
+    },
+    {
+      "epoch": 2.500433500953702,
+      "grad_norm": 0.07056768983602524,
+      "learning_rate": 5.5820000000000004e-05,
+      "loss": 0.0122,
+      "step": 7210
+    },
+    {
+      "epoch": 2.503901508583319,
+      "grad_norm": 0.09822002053260803,
+      "learning_rate": 5.5620000000000006e-05,
+      "loss": 0.0111,
+      "step": 7220
+    },
+    {
+      "epoch": 2.5073695162129357,
+      "grad_norm": 0.04473882168531418,
+      "learning_rate": 5.542000000000001e-05,
+      "loss": 0.0126,
+      "step": 7230
+    },
+    {
+      "epoch": 2.5108375238425524,
+      "grad_norm": 0.07200278341770172,
+      "learning_rate": 5.522e-05,
+      "loss": 0.0123,
+      "step": 7240
+    },
+    {
+      "epoch": 2.514305531472169,
+      "grad_norm": 0.054528553038835526,
+      "learning_rate": 5.5020000000000005e-05,
+      "loss": 0.0139,
+      "step": 7250
+    },
+    {
+      "epoch": 2.517773539101786,
+      "grad_norm": 0.06125443056225777,
+      "learning_rate": 5.482000000000001e-05,
+      "loss": 0.0136,
+      "step": 7260
+    },
+    {
+      "epoch": 2.5212415467314027,
+      "grad_norm": 0.0841405987739563,
+      "learning_rate": 5.462e-05,
+      "loss": 0.012,
+      "step": 7270
+    },
+    {
+      "epoch": 2.52470955436102,
+      "grad_norm": 0.05345413088798523,
+      "learning_rate": 5.4420000000000004e-05,
+      "loss": 0.0124,
+      "step": 7280
+    },
+    {
+      "epoch": 2.5281775619906366,
+      "grad_norm": 0.1266576498746872,
+      "learning_rate": 5.4220000000000006e-05,
+      "loss": 0.0115,
+      "step": 7290
+    },
+    {
+      "epoch": 2.5316455696202533,
+      "grad_norm": 0.06762862950563431,
+      "learning_rate": 5.402e-05,
+      "loss": 0.0144,
+      "step": 7300
+    },
+    {
+      "epoch": 2.53511357724987,
+      "grad_norm": 0.06965386867523193,
+      "learning_rate": 5.382e-05,
+      "loss": 0.0107,
+      "step": 7310
+    },
+    {
+      "epoch": 2.538581584879487,
+      "grad_norm": 0.13501428067684174,
+      "learning_rate": 5.3620000000000005e-05,
+      "loss": 0.0122,
+      "step": 7320
+    },
+    {
+      "epoch": 2.5420495925091036,
+      "grad_norm": 0.0995524674654007,
+      "learning_rate": 5.342e-05,
+      "loss": 0.0129,
+      "step": 7330
+    },
+    {
+      "epoch": 2.5455176001387203,
+      "grad_norm": 0.10781200975179672,
+      "learning_rate": 5.322e-05,
+      "loss": 0.0123,
+      "step": 7340
+    },
+    {
+      "epoch": 2.548985607768337,
+      "grad_norm": 0.08666220307350159,
+      "learning_rate": 5.3020000000000004e-05,
+      "loss": 0.0128,
+      "step": 7350
+    },
+    {
+      "epoch": 2.552453615397954,
+      "grad_norm": 0.0817803218960762,
+      "learning_rate": 5.2820000000000006e-05,
+      "loss": 0.0133,
+      "step": 7360
+    },
+    {
+      "epoch": 2.5559216230275705,
+      "grad_norm": 0.08346541970968246,
+      "learning_rate": 5.262e-05,
+      "loss": 0.0142,
+      "step": 7370
+    },
+    {
+      "epoch": 2.5593896306571873,
+      "grad_norm": 0.08062835782766342,
+      "learning_rate": 5.242e-05,
+      "loss": 0.0108,
+      "step": 7380
+    },
+    {
+      "epoch": 2.562857638286804,
+      "grad_norm": 0.07476343959569931,
+      "learning_rate": 5.2220000000000005e-05,
+      "loss": 0.0129,
+      "step": 7390
+    },
+    {
+      "epoch": 2.5663256459164208,
+      "grad_norm": 0.0925203412771225,
+      "learning_rate": 5.202e-05,
+      "loss": 0.0137,
+      "step": 7400
+    },
+    {
+      "epoch": 2.569793653546038,
+      "grad_norm": 0.06427571177482605,
+      "learning_rate": 5.182e-05,
+      "loss": 0.012,
+      "step": 7410
+    },
+    {
+      "epoch": 2.5732616611756547,
+      "grad_norm": 0.08569345623254776,
+      "learning_rate": 5.1620000000000004e-05,
+      "loss": 0.0103,
+      "step": 7420
+    },
+    {
+      "epoch": 2.5767296688052714,
+      "grad_norm": 0.09565524011850357,
+      "learning_rate": 5.142e-05,
+      "loss": 0.0099,
+      "step": 7430
+    },
+    {
+      "epoch": 2.580197676434888,
+      "grad_norm": 0.09814833849668503,
+      "learning_rate": 5.122e-05,
+      "loss": 0.0125,
+      "step": 7440
+    },
+    {
+      "epoch": 2.583665684064505,
+      "grad_norm": 0.09652476012706757,
+      "learning_rate": 5.102e-05,
+      "loss": 0.0139,
+      "step": 7450
+    },
+    {
+      "epoch": 2.5871336916941217,
+      "grad_norm": 0.09088584780693054,
+      "learning_rate": 5.082e-05,
+      "loss": 0.0129,
+      "step": 7460
+    },
+    {
+      "epoch": 2.5906016993237384,
+      "grad_norm": 0.0785548985004425,
+      "learning_rate": 5.062e-05,
+      "loss": 0.0119,
+      "step": 7470
+    },
+    {
+      "epoch": 2.594069706953355,
+      "grad_norm": 0.0708330050110817,
+      "learning_rate": 5.042e-05,
+      "loss": 0.0121,
+      "step": 7480
+    },
+    {
+      "epoch": 2.5975377145829723,
+      "grad_norm": 0.09456871449947357,
+      "learning_rate": 5.0220000000000004e-05,
+      "loss": 0.0127,
+      "step": 7490
+    },
+    {
+      "epoch": 2.601005722212589,
+      "grad_norm": 0.05804060399532318,
+      "learning_rate": 5.002e-05,
+      "loss": 0.012,
+      "step": 7500
+    },
+    {
+      "epoch": 2.604473729842206,
+      "grad_norm": 0.08347133547067642,
+      "learning_rate": 4.982e-05,
+      "loss": 0.0121,
+      "step": 7510
+    },
+    {
+      "epoch": 2.6079417374718226,
+      "grad_norm": 0.10097778588533401,
+      "learning_rate": 4.962e-05,
+      "loss": 0.0116,
+      "step": 7520
+    },
+    {
+      "epoch": 2.6114097451014393,
+      "grad_norm": 0.07667971402406693,
+      "learning_rate": 4.942e-05,
+      "loss": 0.0144,
+      "step": 7530
+    },
+    {
+      "epoch": 2.614877752731056,
+      "grad_norm": 0.09747060388326645,
+      "learning_rate": 4.9220000000000006e-05,
+      "loss": 0.0115,
+      "step": 7540
+    },
+    {
+      "epoch": 2.618345760360673,
+      "grad_norm": 0.07479839771986008,
+      "learning_rate": 4.902e-05,
+      "loss": 0.0127,
+      "step": 7550
+    },
+    {
+      "epoch": 2.6218137679902895,
+      "grad_norm": 0.11268935352563858,
+      "learning_rate": 4.8820000000000004e-05,
+      "loss": 0.0143,
+      "step": 7560
+    },
+    {
+      "epoch": 2.6252817756199063,
+      "grad_norm": 0.0846814289689064,
+      "learning_rate": 4.8620000000000005e-05,
+      "loss": 0.0133,
+      "step": 7570
+    },
+    {
+      "epoch": 2.628749783249523,
+      "grad_norm": 0.10621094703674316,
+      "learning_rate": 4.842000000000001e-05,
+      "loss": 0.013,
+      "step": 7580
+    },
+    {
+      "epoch": 2.6322177908791398,
+      "grad_norm": 0.07603222131729126,
+      "learning_rate": 4.822e-05,
+      "loss": 0.0124,
+      "step": 7590
+    },
+    {
+      "epoch": 2.6356857985087565,
+      "grad_norm": 0.08150995522737503,
+      "learning_rate": 4.8020000000000004e-05,
+      "loss": 0.013,
+      "step": 7600
+    },
+    {
+      "epoch": 2.6391538061383732,
+      "grad_norm": 0.08962032943964005,
+      "learning_rate": 4.7820000000000006e-05,
+      "loss": 0.0119,
+      "step": 7610
+    },
+    {
+      "epoch": 2.6426218137679904,
+      "grad_norm": 0.07465488463640213,
+      "learning_rate": 4.762e-05,
+      "loss": 0.0116,
+      "step": 7620
+    },
+    {
+      "epoch": 2.646089821397607,
+      "grad_norm": 0.1040426716208458,
+      "learning_rate": 4.742e-05,
+      "loss": 0.0115,
+      "step": 7630
+    },
+    {
+      "epoch": 2.649557829027224,
+      "grad_norm": 0.14476630091667175,
+      "learning_rate": 4.7220000000000005e-05,
+      "loss": 0.0114,
+      "step": 7640
+    },
+    {
+      "epoch": 2.6530258366568407,
+      "grad_norm": 0.07727625221014023,
+      "learning_rate": 4.702e-05,
+      "loss": 0.0122,
+      "step": 7650
+    },
+    {
+      "epoch": 2.6564938442864574,
+      "grad_norm": 0.0987255796790123,
+      "learning_rate": 4.682e-05,
+      "loss": 0.0132,
+      "step": 7660
+    },
+    {
+      "epoch": 2.659961851916074,
+      "grad_norm": 0.07335236668586731,
+      "learning_rate": 4.6620000000000004e-05,
+      "loss": 0.0126,
+      "step": 7670
+    },
+    {
+      "epoch": 2.663429859545691,
+      "grad_norm": 0.09524571150541306,
+      "learning_rate": 4.642e-05,
+      "loss": 0.0114,
+      "step": 7680
+    },
+    {
+      "epoch": 2.6668978671753076,
+      "grad_norm": 0.11405625939369202,
+      "learning_rate": 4.622e-05,
+      "loss": 0.0144,
+      "step": 7690
+    },
+    {
+      "epoch": 2.670365874804925,
+      "grad_norm": 0.09738898277282715,
+      "learning_rate": 4.602e-05,
+      "loss": 0.0139,
+      "step": 7700
+    },
+    {
+      "epoch": 2.6738338824345416,
+      "grad_norm": 0.08633222430944443,
+      "learning_rate": 4.5820000000000005e-05,
+      "loss": 0.0136,
+      "step": 7710
+    },
+    {
+      "epoch": 2.6773018900641583,
+      "grad_norm": 0.09597263485193253,
+      "learning_rate": 4.562e-05,
+      "loss": 0.015,
+      "step": 7720
+    },
+    {
+      "epoch": 2.680769897693775,
+      "grad_norm": 0.0945490151643753,
+      "learning_rate": 4.542e-05,
+      "loss": 0.012,
+      "step": 7730
+    },
+    {
+      "epoch": 2.684237905323392,
+      "grad_norm": 0.0953177884221077,
+      "learning_rate": 4.5220000000000004e-05,
+      "loss": 0.0121,
+      "step": 7740
+    },
+    {
+      "epoch": 2.6877059129530085,
+      "grad_norm": 0.09284798800945282,
+      "learning_rate": 4.502e-05,
+      "loss": 0.0128,
+      "step": 7750
+    },
+    {
+      "epoch": 2.6911739205826253,
+      "grad_norm": 0.10929513722658157,
+      "learning_rate": 4.482e-05,
+      "loss": 0.0122,
+      "step": 7760
+    },
+    {
+      "epoch": 2.694641928212242,
+      "grad_norm": 0.10034655779600143,
+      "learning_rate": 4.462e-05,
+      "loss": 0.0137,
+      "step": 7770
+    },
+    {
+      "epoch": 2.6981099358418588,
+      "grad_norm": 0.06497249752283096,
+      "learning_rate": 4.442e-05,
+      "loss": 0.0116,
+      "step": 7780
+    },
+    {
+      "epoch": 2.7015779434714755,
+      "grad_norm": 0.061711184680461884,
+      "learning_rate": 4.422e-05,
+      "loss": 0.0118,
+      "step": 7790
+    },
+    {
+      "epoch": 2.7050459511010922,
+      "grad_norm": 0.08474565297365189,
+      "learning_rate": 4.402e-05,
+      "loss": 0.0113,
+      "step": 7800
+    },
+    {
+      "epoch": 2.708513958730709,
+      "grad_norm": 0.06698320806026459,
+      "learning_rate": 4.382e-05,
+      "loss": 0.0116,
+      "step": 7810
+    },
+    {
+      "epoch": 2.7119819663603257,
+      "grad_norm": 0.11040494590997696,
+      "learning_rate": 4.362e-05,
+      "loss": 0.014,
+      "step": 7820
+    },
+    {
+      "epoch": 2.715449973989943,
+      "grad_norm": 0.077610544860363,
+      "learning_rate": 4.342e-05,
+      "loss": 0.0148,
+      "step": 7830
+    },
+    {
+      "epoch": 2.7189179816195597,
+      "grad_norm": 0.07504422217607498,
+      "learning_rate": 4.3219999999999996e-05,
+      "loss": 0.0109,
+      "step": 7840
+    },
+    {
+      "epoch": 2.7223859892491764,
+      "grad_norm": 0.083879753947258,
+      "learning_rate": 4.3020000000000005e-05,
+      "loss": 0.0136,
+      "step": 7850
+    },
+    {
+      "epoch": 2.725853996878793,
+      "grad_norm": 0.07689789682626724,
+      "learning_rate": 4.282000000000001e-05,
+      "loss": 0.013,
+      "step": 7860
+    },
+    {
+      "epoch": 2.72932200450841,
+      "grad_norm": 0.07872592657804489,
+      "learning_rate": 4.262e-05,
+      "loss": 0.0123,
+      "step": 7870
+    },
+    {
+      "epoch": 2.7327900121380266,
+      "grad_norm": 0.10761665552854538,
+      "learning_rate": 4.2420000000000004e-05,
+      "loss": 0.0141,
+      "step": 7880
+    },
+    {
+      "epoch": 2.7362580197676434,
+      "grad_norm": 0.07873303443193436,
+      "learning_rate": 4.2220000000000006e-05,
+      "loss": 0.0114,
+      "step": 7890
+    },
+    {
+      "epoch": 2.73972602739726,
+      "grad_norm": 0.05685030296444893,
+      "learning_rate": 4.202e-05,
+      "loss": 0.0102,
+      "step": 7900
+    },
+    {
+      "epoch": 2.7431940350268773,
+      "grad_norm": 0.10359474271535873,
+      "learning_rate": 4.182e-05,
+      "loss": 0.0131,
+      "step": 7910
+    },
+    {
+      "epoch": 2.746662042656494,
+      "grad_norm": 0.07630922645330429,
+      "learning_rate": 4.1620000000000005e-05,
+      "loss": 0.0115,
+      "step": 7920
+    },
+    {
+      "epoch": 2.750130050286111,
+      "grad_norm": 0.09103643894195557,
+      "learning_rate": 4.142000000000001e-05,
+      "loss": 0.012,
+      "step": 7930
+    },
+    {
+      "epoch": 2.7535980579157275,
+      "grad_norm": 0.08436847478151321,
+      "learning_rate": 4.122e-05,
+      "loss": 0.0115,
+      "step": 7940
+    },
+    {
+      "epoch": 2.7570660655453443,
+      "grad_norm": 0.09147916734218597,
+      "learning_rate": 4.1020000000000004e-05,
+      "loss": 0.0164,
+      "step": 7950
+    },
+    {
+      "epoch": 2.760534073174961,
+      "grad_norm": 0.1505836397409439,
+      "learning_rate": 4.0820000000000006e-05,
+      "loss": 0.0136,
+      "step": 7960
+    },
+    {
+      "epoch": 2.7640020808045778,
+      "grad_norm": 0.06953336298465729,
+      "learning_rate": 4.062e-05,
+      "loss": 0.0114,
+      "step": 7970
+    },
+    {
+      "epoch": 2.7674700884341945,
+      "grad_norm": 0.08808320015668869,
+      "learning_rate": 4.042e-05,
+      "loss": 0.0095,
+      "step": 7980
+    },
+    {
+      "epoch": 2.7709380960638113,
+      "grad_norm": 0.10280350595712662,
+      "learning_rate": 4.0220000000000005e-05,
+      "loss": 0.0095,
+      "step": 7990
+    },
+    {
+      "epoch": 2.774406103693428,
+      "grad_norm": 0.11173941940069199,
+      "learning_rate": 4.002e-05,
+      "loss": 0.0128,
+      "step": 8000
+    },
+    {
+      "epoch": 2.774406103693428,
+      "eval_loss": 0.06569719314575195,
+      "eval_runtime": 714.8383,
+      "eval_samples_per_second": 12.845,
+      "eval_steps_per_second": 1.606,
+      "step": 8000
+    },
+    {
+      "epoch": 2.7778741113230447,
+      "grad_norm": 0.0838409885764122,
+      "learning_rate": 3.982e-05,
+      "loss": 0.0132,
+      "step": 8010
+    },
+    {
+      "epoch": 2.7813421189526615,
+      "grad_norm": 0.04326736554503441,
+      "learning_rate": 3.9620000000000004e-05,
+      "loss": 0.011,
+      "step": 8020
+    },
+    {
+      "epoch": 2.7848101265822782,
+      "grad_norm": 0.054840609431266785,
+      "learning_rate": 3.942e-05,
+      "loss": 0.0104,
+      "step": 8030
+    },
+    {
+      "epoch": 2.7882781342118954,
+      "grad_norm": 0.055376045405864716,
+      "learning_rate": 3.922e-05,
+      "loss": 0.012,
+      "step": 8040
+    },
+    {
+      "epoch": 2.791746141841512,
+      "grad_norm": 0.10542810708284378,
+      "learning_rate": 3.902e-05,
+      "loss": 0.0143,
+      "step": 8050
+    },
+    {
+      "epoch": 2.795214149471129,
+      "grad_norm": 0.09201773256063461,
+      "learning_rate": 3.882e-05,
+      "loss": 0.0131,
+      "step": 8060
+    },
+    {
+      "epoch": 2.7986821571007456,
+      "grad_norm": 0.10081171244382858,
+      "learning_rate": 3.862e-05,
+      "loss": 0.0132,
+      "step": 8070
+    },
+    {
+      "epoch": 2.8021501647303624,
+      "grad_norm": 0.13332189619541168,
+      "learning_rate": 3.842e-05,
+      "loss": 0.0119,
+      "step": 8080
+    },
+    {
+      "epoch": 2.805618172359979,
+      "grad_norm": 0.060715481638908386,
+      "learning_rate": 3.822e-05,
+      "loss": 0.0132,
+      "step": 8090
+    },
+    {
+      "epoch": 2.809086179989596,
+      "grad_norm": 0.08016696572303772,
+      "learning_rate": 3.802e-05,
+      "loss": 0.0118,
+      "step": 8100
+    },
+    {
+      "epoch": 2.8125541876192126,
+      "grad_norm": 0.07697981595993042,
+      "learning_rate": 3.782e-05,
+      "loss": 0.0125,
+      "step": 8110
+    },
+    {
+      "epoch": 2.81602219524883,
+      "grad_norm": 0.10400475561618805,
+      "learning_rate": 3.762e-05,
+      "loss": 0.0096,
+      "step": 8120
+    },
+    {
+      "epoch": 2.8194902028784465,
+      "grad_norm": 0.09359320998191833,
+      "learning_rate": 3.742e-05,
+      "loss": 0.0124,
+      "step": 8130
+    },
+    {
+      "epoch": 2.8229582105080633,
+      "grad_norm": 0.08394026756286621,
+      "learning_rate": 3.722e-05,
+      "loss": 0.0101,
+      "step": 8140
+    },
+    {
+      "epoch": 2.82642621813768,
+      "grad_norm": 0.08613195270299911,
+      "learning_rate": 3.702e-05,
+      "loss": 0.0138,
+      "step": 8150
+    },
+    {
+      "epoch": 2.8298942257672968,
+      "grad_norm": 0.1419047713279724,
+      "learning_rate": 3.682e-05,
+      "loss": 0.0109,
+      "step": 8160
+    },
+    {
+      "epoch": 2.8333622333969135,
+      "grad_norm": 0.08159425854682922,
+      "learning_rate": 3.6620000000000005e-05,
+      "loss": 0.0117,
+      "step": 8170
+    },
+    {
+      "epoch": 2.8368302410265303,
+      "grad_norm": 0.07656218111515045,
+      "learning_rate": 3.642000000000001e-05,
+      "loss": 0.014,
+      "step": 8180
+    },
+    {
+      "epoch": 2.840298248656147,
+      "grad_norm": 0.07194588333368301,
+      "learning_rate": 3.622e-05,
+      "loss": 0.0097,
+      "step": 8190
+    },
+    {
+      "epoch": 2.8437662562857637,
+      "grad_norm": 0.09303127229213715,
+      "learning_rate": 3.6020000000000004e-05,
+      "loss": 0.0098,
+      "step": 8200
+    },
+    {
+      "epoch": 2.8472342639153805,
+      "grad_norm": 0.07557094097137451,
+      "learning_rate": 3.5820000000000006e-05,
+      "loss": 0.0105,
+      "step": 8210
+    },
+    {
+      "epoch": 2.8507022715449972,
+      "grad_norm": 0.07813210785388947,
+      "learning_rate": 3.562e-05,
+      "loss": 0.0108,
+      "step": 8220
+    },
+    {
+      "epoch": 2.854170279174614,
+      "grad_norm": 0.1083005964756012,
+      "learning_rate": 3.542e-05,
+      "loss": 0.0109,
+      "step": 8230
+    },
+    {
+      "epoch": 2.8576382868042307,
+      "grad_norm": 0.11126165091991425,
+      "learning_rate": 3.5220000000000005e-05,
+      "loss": 0.0117,
+      "step": 8240
+    },
+    {
+      "epoch": 2.861106294433848,
+      "grad_norm": 0.07165870070457458,
+      "learning_rate": 3.502e-05,
+      "loss": 0.0101,
+      "step": 8250
+    },
+    {
+      "epoch": 2.8645743020634646,
+      "grad_norm": 0.11130272597074509,
+      "learning_rate": 3.482e-05,
+      "loss": 0.0132,
+      "step": 8260
+    },
+    {
+      "epoch": 2.8680423096930814,
+      "grad_norm": 0.1115003302693367,
+      "learning_rate": 3.4620000000000004e-05,
+      "loss": 0.0157,
+      "step": 8270
+    },
+    {
+      "epoch": 2.871510317322698,
+      "grad_norm": 0.11023005098104477,
+      "learning_rate": 3.442e-05,
+      "loss": 0.0117,
+      "step": 8280
+    },
+    {
+      "epoch": 2.874978324952315,
+      "grad_norm": 0.11513439565896988,
+      "learning_rate": 3.422e-05,
+      "loss": 0.0132,
+      "step": 8290
+    },
+    {
+      "epoch": 2.8784463325819316,
+      "grad_norm": 0.07337312400341034,
+      "learning_rate": 3.402e-05,
+      "loss": 0.011,
+      "step": 8300
+    },
+    {
+      "epoch": 2.8819143402115484,
+      "grad_norm": 0.08071292191743851,
+      "learning_rate": 3.3820000000000005e-05,
+      "loss": 0.012,
+      "step": 8310
+    },
+    {
+      "epoch": 2.885382347841165,
+      "grad_norm": 0.09460859000682831,
+      "learning_rate": 3.362e-05,
+      "loss": 0.0118,
+      "step": 8320
+    },
+    {
+      "epoch": 2.8888503554707823,
+      "grad_norm": 0.08895740658044815,
+      "learning_rate": 3.342e-05,
+      "loss": 0.011,
+      "step": 8330
+    },
+    {
+      "epoch": 2.892318363100399,
+      "grad_norm": 0.08386581391096115,
+      "learning_rate": 3.3220000000000004e-05,
+      "loss": 0.0108,
+      "step": 8340
+    },
+    {
+      "epoch": 2.8957863707300158,
+      "grad_norm": 0.09314845502376556,
+      "learning_rate": 3.302e-05,
+      "loss": 0.0131,
+      "step": 8350
+    },
+    {
+      "epoch": 2.8992543783596325,
+      "grad_norm": 0.06327735632658005,
+      "learning_rate": 3.282e-05,
+      "loss": 0.0101,
+      "step": 8360
+    },
+    {
+      "epoch": 2.9027223859892493,
+      "grad_norm": 0.07077737152576447,
+      "learning_rate": 3.262e-05,
+      "loss": 0.0109,
+      "step": 8370
+    },
+    {
+      "epoch": 2.906190393618866,
+      "grad_norm": 0.07617966085672379,
+      "learning_rate": 3.242e-05,
+      "loss": 0.0104,
+      "step": 8380
+    },
+    {
+      "epoch": 2.9096584012484827,
+      "grad_norm": 0.10365297645330429,
+      "learning_rate": 3.222e-05,
+      "loss": 0.0114,
+      "step": 8390
+    },
+    {
+      "epoch": 2.9131264088780995,
+      "grad_norm": 0.07602348923683167,
+      "learning_rate": 3.202e-05,
+      "loss": 0.011,
+      "step": 8400
+    },
+    {
+      "epoch": 2.9165944165077162,
+      "grad_norm": 0.10498479008674622,
+      "learning_rate": 3.182e-05,
+      "loss": 0.0118,
+      "step": 8410
+    },
+    {
+      "epoch": 2.920062424137333,
+      "grad_norm": 0.08163636922836304,
+      "learning_rate": 3.162e-05,
+      "loss": 0.0121,
+      "step": 8420
+    },
+    {
+      "epoch": 2.9235304317669497,
+      "grad_norm": 0.09529927372932434,
+      "learning_rate": 3.142e-05,
+      "loss": 0.0128,
+      "step": 8430
+    },
+    {
+      "epoch": 2.9269984393965665,
+      "grad_norm": 0.10269023478031158,
+      "learning_rate": 3.122e-05,
+      "loss": 0.0102,
+      "step": 8440
+    },
+    {
+      "epoch": 2.930466447026183,
+      "grad_norm": 0.1097848117351532,
+      "learning_rate": 3.102e-05,
+      "loss": 0.0128,
+      "step": 8450
+    },
+    {
+      "epoch": 2.9339344546558004,
+      "grad_norm": 0.06423291563987732,
+      "learning_rate": 3.082e-05,
+      "loss": 0.0086,
+      "step": 8460
+    },
+    {
+      "epoch": 2.937402462285417,
+      "grad_norm": 0.10673278570175171,
+      "learning_rate": 3.062e-05,
+      "loss": 0.0128,
+      "step": 8470
+    },
+    {
+      "epoch": 2.940870469915034,
+      "grad_norm": 0.08257333934307098,
+      "learning_rate": 3.0420000000000004e-05,
+      "loss": 0.0122,
+      "step": 8480
+    },
+    {
+      "epoch": 2.9443384775446506,
+      "grad_norm": 0.08055904507637024,
+      "learning_rate": 3.0220000000000005e-05,
+      "loss": 0.0115,
+      "step": 8490
+    },
+    {
+      "epoch": 2.9478064851742674,
+      "grad_norm": 0.09221120178699493,
+      "learning_rate": 3.0020000000000004e-05,
+      "loss": 0.0117,
+      "step": 8500
+    },
+    {
+      "epoch": 2.951274492803884,
+      "grad_norm": 0.0831000804901123,
+      "learning_rate": 2.9820000000000002e-05,
+      "loss": 0.0108,
+      "step": 8510
+    },
+    {
+      "epoch": 2.954742500433501,
+      "grad_norm": 0.06930562108755112,
+      "learning_rate": 2.9620000000000004e-05,
+      "loss": 0.0117,
+      "step": 8520
+    },
+    {
+      "epoch": 2.9582105080631176,
+      "grad_norm": 0.07918361574411392,
+      "learning_rate": 2.9420000000000003e-05,
+      "loss": 0.0113,
+      "step": 8530
+    },
+    {
+      "epoch": 2.9616785156927348,
+      "grad_norm": 0.06908438354730606,
+      "learning_rate": 2.922e-05,
+      "loss": 0.0101,
+      "step": 8540
+    },
+    {
+      "epoch": 2.9651465233223515,
+      "grad_norm": 0.08593347668647766,
+      "learning_rate": 2.9020000000000003e-05,
+      "loss": 0.0127,
+      "step": 8550
+    },
+    {
+      "epoch": 2.9686145309519683,
+      "grad_norm": 0.09223376214504242,
+      "learning_rate": 2.8820000000000002e-05,
+      "loss": 0.0137,
+      "step": 8560
+    },
+    {
+      "epoch": 2.972082538581585,
+      "grad_norm": 0.09096557646989822,
+      "learning_rate": 2.8620000000000004e-05,
+      "loss": 0.0104,
+      "step": 8570
+    },
+    {
+      "epoch": 2.9755505462112017,
+      "grad_norm": 0.0837118849158287,
+      "learning_rate": 2.8420000000000002e-05,
+      "loss": 0.0113,
+      "step": 8580
+    },
+    {
+      "epoch": 2.9790185538408185,
+      "grad_norm": 0.060600802302360535,
+      "learning_rate": 2.822e-05,
+      "loss": 0.0119,
+      "step": 8590
+    },
+    {
+      "epoch": 2.9824865614704352,
+      "grad_norm": 0.09768325090408325,
+      "learning_rate": 2.8020000000000003e-05,
+      "loss": 0.0094,
+      "step": 8600
+    },
+    {
+      "epoch": 2.985954569100052,
+      "grad_norm": 0.10265090316534042,
+      "learning_rate": 2.782e-05,
+      "loss": 0.0117,
+      "step": 8610
+    },
+    {
+      "epoch": 2.9894225767296687,
+      "grad_norm": 0.11693336814641953,
+      "learning_rate": 2.762e-05,
+      "loss": 0.0126,
+      "step": 8620
+    },
+    {
+      "epoch": 2.9928905843592855,
+      "grad_norm": 0.06177123263478279,
+      "learning_rate": 2.7420000000000002e-05,
+      "loss": 0.0112,
+      "step": 8630
+    },
+    {
+      "epoch": 2.996358591988902,
+      "grad_norm": 0.07003209739923477,
+      "learning_rate": 2.722e-05,
+      "loss": 0.0111,
+      "step": 8640
+    },
+    {
+      "epoch": 2.999826599618519,
+      "grad_norm": 0.09619972854852676,
+      "learning_rate": 2.7020000000000002e-05,
+      "loss": 0.0113,
+      "step": 8650
+    },
+    {
+      "epoch": 3.003294607248136,
+      "grad_norm": 0.04991637170314789,
+      "learning_rate": 2.682e-05,
+      "loss": 0.0045,
+      "step": 8660
+    },
+    {
+      "epoch": 3.006762614877753,
+      "grad_norm": 0.054620299488306046,
+      "learning_rate": 2.662e-05,
+      "loss": 0.0029,
+      "step": 8670
+    },
+    {
+      "epoch": 3.0102306225073696,
+      "grad_norm": 0.031793851405382156,
+      "learning_rate": 2.642e-05,
+      "loss": 0.003,
+      "step": 8680
+    },
+    {
+      "epoch": 3.0136986301369864,
+      "grad_norm": 0.03903364762663841,
+      "learning_rate": 2.622e-05,
+      "loss": 0.0024,
+      "step": 8690
+    },
+    {
+      "epoch": 3.017166637766603,
+      "grad_norm": 0.05518382415175438,
+      "learning_rate": 2.602e-05,
+      "loss": 0.0028,
+      "step": 8700
+    },
+    {
+      "epoch": 3.02063464539622,
+      "grad_norm": 0.02921813167631626,
+      "learning_rate": 2.582e-05,
+      "loss": 0.0018,
+      "step": 8710
+    },
+    {
+      "epoch": 3.0241026530258366,
+      "grad_norm": 0.08628415316343307,
+      "learning_rate": 2.562e-05,
+      "loss": 0.0027,
+      "step": 8720
+    },
+    {
+      "epoch": 3.0275706606554533,
+      "grad_norm": 0.034173641353845596,
+      "learning_rate": 2.542e-05,
+      "loss": 0.002,
+      "step": 8730
+    },
+    {
+      "epoch": 3.03103866828507,
+      "grad_norm": 0.03343451768159866,
+      "learning_rate": 2.522e-05,
+      "loss": 0.0028,
+      "step": 8740
+    },
+    {
+      "epoch": 3.034506675914687,
+      "grad_norm": 0.0506439134478569,
+      "learning_rate": 2.5019999999999998e-05,
+      "loss": 0.002,
+      "step": 8750
+    },
+    {
+      "epoch": 3.037974683544304,
+      "grad_norm": 0.011204255744814873,
+      "learning_rate": 2.4820000000000003e-05,
+      "loss": 0.0026,
+      "step": 8760
+    },
+    {
+      "epoch": 3.0414426911739207,
+      "grad_norm": 0.05876510962843895,
+      "learning_rate": 2.462e-05,
+      "loss": 0.0023,
+      "step": 8770
+    },
+    {
+      "epoch": 3.0449106988035375,
+      "grad_norm": 0.017792142927646637,
+      "learning_rate": 2.442e-05,
+      "loss": 0.0014,
+      "step": 8780
+    },
+    {
+      "epoch": 3.0483787064331542,
+      "grad_norm": 0.0634002834558487,
+      "learning_rate": 2.4220000000000002e-05,
+      "loss": 0.002,
+      "step": 8790
+    },
+    {
+      "epoch": 3.051846714062771,
+      "grad_norm": 0.10669626295566559,
+      "learning_rate": 2.402e-05,
+      "loss": 0.002,
+      "step": 8800
+    },
+    {
+      "epoch": 3.0553147216923877,
+      "grad_norm": 0.02126486971974373,
+      "learning_rate": 2.3820000000000002e-05,
+      "loss": 0.002,
+      "step": 8810
+    },
+    {
+      "epoch": 3.0587827293220045,
+      "grad_norm": 0.13548636436462402,
+      "learning_rate": 2.362e-05,
+      "loss": 0.0034,
+      "step": 8820
+    },
+    {
+      "epoch": 3.062250736951621,
+      "grad_norm": 0.04129517078399658,
+      "learning_rate": 2.342e-05,
+      "loss": 0.0032,
+      "step": 8830
+    },
+    {
+      "epoch": 3.065718744581238,
+      "grad_norm": 0.0712270438671112,
+      "learning_rate": 2.322e-05,
+      "loss": 0.0025,
+      "step": 8840
+    },
+    {
+      "epoch": 3.0691867522108547,
+      "grad_norm": 0.03855830430984497,
+      "learning_rate": 2.302e-05,
+      "loss": 0.0022,
+      "step": 8850
+    },
+    {
+      "epoch": 3.0726547598404714,
+      "grad_norm": 0.06443187594413757,
+      "learning_rate": 2.282e-05,
+      "loss": 0.0028,
+      "step": 8860
+    },
+    {
+      "epoch": 3.0761227674700886,
+      "grad_norm": 0.09318159520626068,
+      "learning_rate": 2.2620000000000004e-05,
+      "loss": 0.0021,
+      "step": 8870
+    },
+    {
+      "epoch": 3.0795907750997054,
+      "grad_norm": 0.06871949136257172,
+      "learning_rate": 2.2420000000000002e-05,
+      "loss": 0.0033,
+      "step": 8880
+    },
+    {
+      "epoch": 3.083058782729322,
+      "grad_norm": 0.031738586723804474,
+      "learning_rate": 2.222e-05,
+      "loss": 0.0016,
+      "step": 8890
+    },
+    {
+      "epoch": 3.086526790358939,
+      "grad_norm": 0.028125595301389694,
+      "learning_rate": 2.2020000000000003e-05,
+      "loss": 0.0026,
+      "step": 8900
+    },
+    {
+      "epoch": 3.0899947979885556,
+      "grad_norm": 0.046880729496479034,
+      "learning_rate": 2.182e-05,
+      "loss": 0.0021,
+      "step": 8910
+    },
+    {
+      "epoch": 3.0934628056181723,
+      "grad_norm": 0.06030425429344177,
+      "learning_rate": 2.162e-05,
+      "loss": 0.0029,
+      "step": 8920
+    },
+    {
+      "epoch": 3.096930813247789,
+      "grad_norm": 0.028561096638441086,
+      "learning_rate": 2.142e-05,
+      "loss": 0.0023,
+      "step": 8930
+    },
+    {
+      "epoch": 3.100398820877406,
+      "grad_norm": 0.04128929600119591,
+      "learning_rate": 2.122e-05,
+      "loss": 0.0022,
+      "step": 8940
+    },
+    {
+      "epoch": 3.1038668285070226,
+      "grad_norm": 0.04722205549478531,
+      "learning_rate": 2.1020000000000002e-05,
+      "loss": 0.0027,
+      "step": 8950
+    },
+    {
+      "epoch": 3.1073348361366393,
+      "grad_norm": 0.057234570384025574,
+      "learning_rate": 2.082e-05,
+      "loss": 0.0016,
+      "step": 8960
+    },
+    {
+      "epoch": 3.1108028437662565,
+      "grad_norm": 0.010785787366330624,
+      "learning_rate": 2.062e-05,
+      "loss": 0.0016,
+      "step": 8970
+    },
+    {
+      "epoch": 3.1142708513958732,
+      "grad_norm": 0.08110074698925018,
+      "learning_rate": 2.042e-05,
+      "loss": 0.0022,
+      "step": 8980
+    },
+    {
+      "epoch": 3.11773885902549,
+      "grad_norm": 0.04953372851014137,
+      "learning_rate": 2.022e-05,
+      "loss": 0.0022,
+      "step": 8990
+    },
+    {
+      "epoch": 3.1212068666551067,
+      "grad_norm": 0.022535376250743866,
+      "learning_rate": 2.002e-05,
+      "loss": 0.0015,
+      "step": 9000
+    },
+    {
+      "epoch": 3.1212068666551067,
+      "eval_loss": 0.0743083506822586,
+      "eval_runtime": 714.2967,
+      "eval_samples_per_second": 12.855,
+      "eval_steps_per_second": 1.607,
+      "step": 9000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.85989473816475e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}