diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,1881 +1,4093 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 0.9999059177721329,
+  "epoch": 1.0,
   "eval_steps": 20,
-  "global_step": 5314,
+  "global_step": 11621,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 0.0037632891146862357,
-      "grad_norm": 0.15685315430164337,
+      "epoch": 0.0017211333663217229,
+      "grad_norm": 0.512730062007904,
       "learning_rate": 0.0002,
-      "loss": 0.4618,
+      "loss": 1.7072,
       "step": 20
     },
     {
-      "epoch": 0.0075265782293724715,
-      "grad_norm": 0.10632659494876862,
+      "epoch": 0.0034422667326434457,
+      "grad_norm": 0.3541152775287628,
       "learning_rate": 0.0002,
-      "loss": 0.2985,
+      "loss": 1.3871,
       "step": 40
     },
     {
-      "epoch": 0.011289867344058707,
-      "grad_norm": 0.12228264659643173,
+      "epoch": 0.005163400098965169,
+      "grad_norm": 0.40835729241371155,
       "learning_rate": 0.0002,
-      "loss": 0.2689,
+      "loss": 1.2637,
       "step": 60
     },
     {
-      "epoch": 0.015053156458744943,
-      "grad_norm": 0.09272768348455429,
+      "epoch": 0.0068845334652868915,
+      "grad_norm": 0.3723342716693878,
       "learning_rate": 0.0002,
-      "loss": 0.2249,
+      "loss": 1.2378,
       "step": 80
     },
     {
-      "epoch": 0.01881644557343118,
-      "grad_norm": 0.11586301028728485,
+      "epoch": 0.008605666831608615,
+      "grad_norm": 0.3391265571117401,
       "learning_rate": 0.0002,
-      "loss": 0.23,
+      "loss": 1.1991,
       "step": 100
     },
     {
-      "epoch": 0.022579734688117414,
-      "grad_norm": 0.08657937496900558,
+      "epoch": 0.010326800197930338,
+      "grad_norm": 0.3430226445198059,
       "learning_rate": 0.0002,
-      "loss": 0.2189,
+      "loss": 1.1191,
       "step": 120
     },
     {
-      "epoch": 0.02634302380280365,
-      "grad_norm": 0.08374184370040894,
+      "epoch": 0.01204793356425206,
+      "grad_norm": 0.39266398549079895,
       "learning_rate": 0.0002,
-      "loss": 0.1987,
+      "loss": 1.1358,
       "step": 140
     },
     {
-      "epoch": 0.030106312917489886,
-      "grad_norm": 0.08528616279363632,
+      "epoch": 0.013769066930573783,
+      "grad_norm": 0.38184505701065063,
       "learning_rate": 0.0002,
-      "loss": 0.2065,
+      "loss": 1.0851,
       "step": 160
     },
     {
-      "epoch": 0.03386960203217612,
-      "grad_norm": 0.17550894618034363,
+      "epoch": 0.015490200296895506,
+      "grad_norm": 0.3323766887187958,
       "learning_rate": 0.0002,
-      "loss": 0.1904,
+      "loss": 1.0519,
       "step": 180
     },
     {
-      "epoch": 0.03763289114686236,
-      "grad_norm": 0.09562012553215027,
+      "epoch": 0.01721133366321723,
+      "grad_norm": 0.3520768880844116,
       "learning_rate": 0.0002,
-      "loss": 0.1824,
+      "loss": 0.9788,
       "step": 200
     },
     {
-      "epoch": 0.04139618026154859,
-      "grad_norm": 0.12333519756793976,
+      "epoch": 0.018932467029538953,
+      "grad_norm": 0.4066319465637207,
       "learning_rate": 0.0002,
-      "loss": 0.1702,
+      "loss": 0.9647,
       "step": 220
     },
     {
-      "epoch": 0.04515946937623483,
-      "grad_norm": 0.11404936760663986,
+      "epoch": 0.020653600395860675,
+      "grad_norm": 0.3871042728424072,
       "learning_rate": 0.0002,
-      "loss": 0.18,
+      "loss": 0.9981,
       "step": 240
     },
     {
-      "epoch": 0.048922758490921064,
-      "grad_norm": 0.08656694740056992,
+      "epoch": 0.022374733762182398,
+      "grad_norm": 0.3759310841560364,
       "learning_rate": 0.0002,
-      "loss": 0.1749,
+      "loss": 0.9527,
       "step": 260
     },
     {
-      "epoch": 0.0526860476056073,
-      "grad_norm": 0.09797225147485733,
+      "epoch": 0.02409586712850412,
+      "grad_norm": 0.411364883184433,
       "learning_rate": 0.0002,
-      "loss": 0.1731,
+      "loss": 1.0079,
       "step": 280
     },
     {
-      "epoch": 0.056449336720293536,
-      "grad_norm": 0.09765412658452988,
+      "epoch": 0.025817000494825843,
+      "grad_norm": 0.38831627368927,
       "learning_rate": 0.0002,
-      "loss": 0.1578,
+      "loss": 1.0129,
       "step": 300
     },
     {
-      "epoch": 0.06021262583497977,
-      "grad_norm": 0.07540671527385712,
+      "epoch": 0.027538133861147566,
+      "grad_norm": 0.36452218890190125,
       "learning_rate": 0.0002,
-      "loss": 0.1693,
+      "loss": 0.9021,
       "step": 320
     },
     {
-      "epoch": 0.06397591494966601,
-      "grad_norm": 0.08590289205312729,
+      "epoch": 0.02925926722746929,
+      "grad_norm": 0.40188145637512207,
       "learning_rate": 0.0002,
-      "loss": 0.1607,
+      "loss": 0.9138,
       "step": 340
     },
     {
-      "epoch": 0.06773920406435224,
-      "grad_norm": 0.09767664223909378,
+      "epoch": 0.03098040059379101,
+      "grad_norm": 0.3994237184524536,
       "learning_rate": 0.0002,
-      "loss": 0.1419,
+      "loss": 0.9164,
       "step": 360
     },
     {
-      "epoch": 0.07150249317903848,
-      "grad_norm": 0.10479151457548141,
+      "epoch": 0.03270153396011274,
+      "grad_norm": 0.4226379692554474,
       "learning_rate": 0.0002,
-      "loss": 0.1629,
+      "loss": 0.8986,
       "step": 380
     },
     {
-      "epoch": 0.07526578229372471,
-      "grad_norm": 0.08791118115186691,
+      "epoch": 0.03442266732643446,
+      "grad_norm": 0.4331601560115814,
       "learning_rate": 0.0002,
-      "loss": 0.1484,
+      "loss": 0.8443,
       "step": 400
     },
     {
-      "epoch": 0.07902907140841095,
-      "grad_norm": 0.10221686214208603,
+      "epoch": 0.03614380069275618,
+      "grad_norm": 0.373415470123291,
       "learning_rate": 0.0002,
-      "loss": 0.1499,
+      "loss": 0.8437,
       "step": 420
     },
     {
-      "epoch": 0.08279236052309719,
-      "grad_norm": 0.09131903946399689,
+      "epoch": 0.037864934059077905,
+      "grad_norm": 0.49005845189094543,
       "learning_rate": 0.0002,
-      "loss": 0.1485,
+      "loss": 0.8387,
       "step": 440
     },
     {
-      "epoch": 0.08655564963778342,
-      "grad_norm": 0.10372031480073929,
+      "epoch": 0.03958606742539963,
+      "grad_norm": 0.5034841895103455,
       "learning_rate": 0.0002,
-      "loss": 0.1441,
+      "loss": 0.8108,
       "step": 460
     },
     {
-      "epoch": 0.09031893875246966,
-      "grad_norm": 0.09649350494146347,
+      "epoch": 0.04130720079172135,
+      "grad_norm": 0.3401569426059723,
       "learning_rate": 0.0002,
-      "loss": 0.147,
+      "loss": 0.7975,
       "step": 480
     },
     {
-      "epoch": 0.0940822278671559,
-      "grad_norm": 0.09961670637130737,
+      "epoch": 0.04302833415804307,
+      "grad_norm": 0.37207746505737305,
       "learning_rate": 0.0002,
-      "loss": 0.1465,
+      "loss": 0.8135,
       "step": 500
     },
     {
-      "epoch": 0.09784551698184213,
-      "grad_norm": 0.08490657806396484,
+      "epoch": 0.044749467524364796,
+      "grad_norm": 0.39017221331596375,
       "learning_rate": 0.0002,
-      "loss": 0.132,
+      "loss": 0.7053,
       "step": 520
     },
     {
-      "epoch": 0.10160880609652836,
-      "grad_norm": 0.08765380829572678,
+      "epoch": 0.04647060089068652,
+      "grad_norm": 0.42643848061561584,
       "learning_rate": 0.0002,
-      "loss": 0.1283,
+      "loss": 0.8119,
       "step": 540
     },
     {
-      "epoch": 0.1053720952112146,
-      "grad_norm": 0.09319768846035004,
+      "epoch": 0.04819173425700824,
+      "grad_norm": 0.4266549050807953,
       "learning_rate": 0.0002,
-      "loss": 0.1382,
+      "loss": 0.7676,
       "step": 560
     },
     {
-      "epoch": 0.10913538432590084,
-      "grad_norm": 0.08941628783941269,
+      "epoch": 0.049912867623329964,
+      "grad_norm": 0.37006089091300964,
       "learning_rate": 0.0002,
-      "loss": 0.1335,
+      "loss": 0.7675,
       "step": 580
     },
     {
-      "epoch": 0.11289867344058707,
-      "grad_norm": 0.0971933901309967,
+      "epoch": 0.051634000989651686,
+      "grad_norm": 0.3692554235458374,
       "learning_rate": 0.0002,
-      "loss": 0.137,
+      "loss": 0.7843,
       "step": 600
     },
     {
-      "epoch": 0.11666196255527331,
-      "grad_norm": 0.07488075643777847,
+      "epoch": 0.05335513435597341,
+      "grad_norm": 0.41451218724250793,
       "learning_rate": 0.0002,
-      "loss": 0.1283,
+      "loss": 0.7111,
       "step": 620
     },
     {
-      "epoch": 0.12042525166995954,
-      "grad_norm": 0.08711710572242737,
+      "epoch": 0.05507626772229513,
+      "grad_norm": 0.46577128767967224,
       "learning_rate": 0.0002,
-      "loss": 0.1292,
+      "loss": 0.7248,
       "step": 640
     },
     {
-      "epoch": 0.12418854078464578,
-      "grad_norm": 0.08043856918811798,
+      "epoch": 0.056797401088616854,
+      "grad_norm": 0.4749889373779297,
       "learning_rate": 0.0002,
-      "loss": 0.1269,
+      "loss": 0.7454,
       "step": 660
     },
     {
-      "epoch": 0.12795182989933201,
-      "grad_norm": 0.07097792625427246,
+      "epoch": 0.05851853445493858,
+      "grad_norm": 0.44663748145103455,
       "learning_rate": 0.0002,
-      "loss": 0.1364,
+      "loss": 0.6829,
       "step": 680
     },
     {
-      "epoch": 0.13171511901401825,
-      "grad_norm": 0.0742156058549881,
+      "epoch": 0.0602396678212603,
+      "grad_norm": 0.3434062600135803,
       "learning_rate": 0.0002,
-      "loss": 0.1277,
+      "loss": 0.6625,
       "step": 700
     },
     {
-      "epoch": 0.1354784081287045,
-      "grad_norm": 0.09956187009811401,
+      "epoch": 0.06196080118758202,
+      "grad_norm": 0.45600661635398865,
       "learning_rate": 0.0002,
-      "loss": 0.1314,
+      "loss": 0.6623,
       "step": 720
     },
     {
-      "epoch": 0.13924169724339072,
-      "grad_norm": 0.08661571145057678,
+      "epoch": 0.06368193455390375,
+      "grad_norm": 0.41260451078414917,
       "learning_rate": 0.0002,
-      "loss": 0.124,
+      "loss": 0.6497,
       "step": 740
     },
     {
-      "epoch": 0.14300498635807696,
-      "grad_norm": 0.08070897310972214,
+      "epoch": 0.06540306792022547,
+      "grad_norm": 0.3827366232872009,
       "learning_rate": 0.0002,
-      "loss": 0.1147,
+      "loss": 0.6133,
       "step": 760
     },
     {
-      "epoch": 0.1467682754727632,
-      "grad_norm": 0.09516704082489014,
+      "epoch": 0.0671242012865472,
+      "grad_norm": 0.44087541103363037,
       "learning_rate": 0.0002,
-      "loss": 0.1168,
+      "loss": 0.6782,
       "step": 780
     },
     {
-      "epoch": 0.15053156458744943,
-      "grad_norm": 0.09076276421546936,
+      "epoch": 0.06884533465286892,
+      "grad_norm": 0.3596974313259125,
       "learning_rate": 0.0002,
-      "loss": 0.1178,
+      "loss": 0.6294,
       "step": 800
     },
     {
-      "epoch": 0.15429485370213566,
-      "grad_norm": 0.0896017774939537,
+      "epoch": 0.07056646801919064,
+      "grad_norm": 0.42401421070098877,
       "learning_rate": 0.0002,
-      "loss": 0.1192,
+      "loss": 0.6761,
       "step": 820
     },
     {
-      "epoch": 0.1580581428168219,
-      "grad_norm": 0.07477965205907822,
+      "epoch": 0.07228760138551237,
+      "grad_norm": 0.34142622351646423,
       "learning_rate": 0.0002,
-      "loss": 0.1127,
+      "loss": 0.6274,
       "step": 840
     },
     {
-      "epoch": 0.16182143193150814,
-      "grad_norm": 0.08405464887619019,
+      "epoch": 0.07400873475183409,
+      "grad_norm": 0.3514570891857147,
       "learning_rate": 0.0002,
-      "loss": 0.119,
+      "loss": 0.6053,
       "step": 860
     },
     {
-      "epoch": 0.16558472104619437,
-      "grad_norm": 0.07539790868759155,
+      "epoch": 0.07572986811815581,
+      "grad_norm": 0.3995071053504944,
       "learning_rate": 0.0002,
-      "loss": 0.1213,
+      "loss": 0.5953,
       "step": 880
     },
     {
-      "epoch": 0.1693480101608806,
-      "grad_norm": 0.08806908130645752,
+      "epoch": 0.07745100148447753,
+      "grad_norm": 0.3967471420764923,
       "learning_rate": 0.0002,
-      "loss": 0.1192,
+      "loss": 0.6375,
       "step": 900
     },
     {
-      "epoch": 0.17311129927556684,
-      "grad_norm": 0.08064749836921692,
+      "epoch": 0.07917213485079926,
+      "grad_norm": 0.434865266084671,
       "learning_rate": 0.0002,
-      "loss": 0.1141,
+      "loss": 0.6053,
       "step": 920
     },
     {
-      "epoch": 0.17687458839025308,
-      "grad_norm": 0.09680119901895523,
+      "epoch": 0.08089326821712098,
+      "grad_norm": 0.42927464842796326,
       "learning_rate": 0.0002,
-      "loss": 0.1042,
+      "loss": 0.5947,
       "step": 940
     },
     {
-      "epoch": 0.18063787750493931,
-      "grad_norm": 0.09092500060796738,
+      "epoch": 0.0826144015834427,
+      "grad_norm": 0.6075530052185059,
       "learning_rate": 0.0002,
-      "loss": 0.1067,
+      "loss": 0.6331,
       "step": 960
     },
     {
-      "epoch": 0.18440116661962555,
-      "grad_norm": 0.07870171219110489,
+      "epoch": 0.08433553494976442,
+      "grad_norm": 0.46515706181526184,
       "learning_rate": 0.0002,
-      "loss": 0.1101,
+      "loss": 0.569,
       "step": 980
     },
     {
-      "epoch": 0.1881644557343118,
-      "grad_norm": 0.06842092424631119,
+      "epoch": 0.08605666831608615,
+      "grad_norm": 0.2592894732952118,
       "learning_rate": 0.0002,
-      "loss": 0.1045,
+      "loss": 0.5572,
       "step": 1000
     },
     {
-      "epoch": 0.19192774484899802,
-      "grad_norm": 0.08229291439056396,
+      "epoch": 0.08777780168240787,
+      "grad_norm": 0.4780791699886322,
       "learning_rate": 0.0002,
-      "loss": 0.114,
+      "loss": 0.6014,
       "step": 1020
     },
     {
-      "epoch": 0.19569103396368426,
-      "grad_norm": 0.07617371529340744,
+      "epoch": 0.08949893504872959,
+      "grad_norm": 0.37773773074150085,
       "learning_rate": 0.0002,
-      "loss": 0.1077,
+      "loss": 0.5783,
       "step": 1040
     },
     {
-      "epoch": 0.1994543230783705,
-      "grad_norm": 0.09529408067464828,
+      "epoch": 0.09122006841505131,
+      "grad_norm": 0.5149890780448914,
       "learning_rate": 0.0002,
-      "loss": 0.101,
+      "loss": 0.5485,
       "step": 1060
     },
     {
-      "epoch": 0.20321761219305673,
-      "grad_norm": 0.07952335476875305,
+      "epoch": 0.09294120178137304,
+      "grad_norm": 0.522530734539032,
       "learning_rate": 0.0002,
-      "loss": 0.1065,
+      "loss": 0.5707,
       "step": 1080
     },
     {
-      "epoch": 0.20698090130774296,
-      "grad_norm": 0.07625720649957657,
+      "epoch": 0.09466233514769476,
+      "grad_norm": 0.42418375611305237,
       "learning_rate": 0.0002,
-      "loss": 0.0987,
+      "loss": 0.5817,
       "step": 1100
     },
     {
-      "epoch": 0.2107441904224292,
-      "grad_norm": 0.07906854152679443,
+      "epoch": 0.09638346851401648,
+      "grad_norm": 0.3230820596218109,
       "learning_rate": 0.0002,
-      "loss": 0.0986,
+      "loss": 0.5426,
       "step": 1120
     },
     {
-      "epoch": 0.21450747953711544,
-      "grad_norm": 0.08430849015712738,
+      "epoch": 0.0981046018803382,
+      "grad_norm": 0.46259307861328125,
       "learning_rate": 0.0002,
-      "loss": 0.1085,
+      "loss": 0.5304,
       "step": 1140
     },
     {
-      "epoch": 0.21827076865180167,
-      "grad_norm": 0.0861297994852066,
+      "epoch": 0.09982573524665993,
+      "grad_norm": 0.41511914134025574,
       "learning_rate": 0.0002,
-      "loss": 0.1066,
+      "loss": 0.5343,
       "step": 1160
     },
     {
-      "epoch": 0.2220340577664879,
-      "grad_norm": 0.07576191425323486,
+      "epoch": 0.10154686861298165,
+      "grad_norm": 0.5695448517799377,
       "learning_rate": 0.0002,
-      "loss": 0.1051,
+      "loss": 0.5773,
       "step": 1180
     },
     {
-      "epoch": 0.22579734688117414,
-      "grad_norm": 0.08132428675889969,
+      "epoch": 0.10326800197930337,
+      "grad_norm": 0.3354031443595886,
       "learning_rate": 0.0002,
-      "loss": 0.1036,
+      "loss": 0.5377,
       "step": 1200
     },
     {
-      "epoch": 0.22956063599586038,
-      "grad_norm": 0.06960251182317734,
+      "epoch": 0.1049891353456251,
+      "grad_norm": 0.5198836922645569,
       "learning_rate": 0.0002,
-      "loss": 0.1039,
+      "loss": 0.5454,
       "step": 1220
     },
     {
-      "epoch": 0.23332392511054661,
-      "grad_norm": 0.08230841159820557,
+      "epoch": 0.10671026871194682,
+      "grad_norm": 0.3364623188972473,
       "learning_rate": 0.0002,
-      "loss": 0.1046,
+      "loss": 0.5057,
       "step": 1240
     },
     {
-      "epoch": 0.23708721422523285,
-      "grad_norm": 0.07119760662317276,
+      "epoch": 0.10843140207826854,
+      "grad_norm": 0.3844848871231079,
       "learning_rate": 0.0002,
-      "loss": 0.1064,
+      "loss": 0.571,
       "step": 1260
     },
     {
-      "epoch": 0.2408505033399191,
-      "grad_norm": 0.06965576857328415,
+      "epoch": 0.11015253544459026,
+      "grad_norm": 0.45747458934783936,
       "learning_rate": 0.0002,
-      "loss": 0.0964,
+      "loss": 0.5425,
       "step": 1280
     },
     {
-      "epoch": 0.24461379245460532,
-      "grad_norm": 0.07533243298530579,
+      "epoch": 0.11187366881091199,
+      "grad_norm": 0.4278349280357361,
       "learning_rate": 0.0002,
-      "loss": 0.0973,
+      "loss": 0.4814,
       "step": 1300
     },
     {
-      "epoch": 0.24837708156929156,
-      "grad_norm": 0.07530753314495087,
+      "epoch": 0.11359480217723371,
+      "grad_norm": 0.4377232491970062,
       "learning_rate": 0.0002,
-      "loss": 0.1002,
+      "loss": 0.5083,
       "step": 1320
     },
     {
-      "epoch": 0.2521403706839778,
-      "grad_norm": 0.0701604038476944,
+      "epoch": 0.11531593554355543,
+      "grad_norm": 0.48636892437934875,
       "learning_rate": 0.0002,
-      "loss": 0.1014,
+      "loss": 0.5317,
       "step": 1340
     },
     {
-      "epoch": 0.25590365979866403,
-      "grad_norm": 0.08768032491207123,
+      "epoch": 0.11703706890987715,
+      "grad_norm": 0.39063599705696106,
       "learning_rate": 0.0002,
-      "loss": 0.099,
+      "loss": 0.482,
       "step": 1360
     },
     {
-      "epoch": 0.2596669489133503,
-      "grad_norm": 0.0789860337972641,
+      "epoch": 0.11875820227619888,
+      "grad_norm": 0.45347273349761963,
       "learning_rate": 0.0002,
-      "loss": 0.0947,
+      "loss": 0.4754,
       "step": 1380
     },
     {
-      "epoch": 0.2634302380280365,
-      "grad_norm": 0.09132009744644165,
+      "epoch": 0.1204793356425206,
+      "grad_norm": 0.44759300351142883,
       "learning_rate": 0.0002,
-      "loss": 0.0975,
+      "loss": 0.4773,
       "step": 1400
     },
     {
-      "epoch": 0.26719352714272276,
-      "grad_norm": 0.07386859506368637,
+      "epoch": 0.12220046900884232,
+      "grad_norm": 0.44881579279899597,
       "learning_rate": 0.0002,
-      "loss": 0.0997,
+      "loss": 0.5236,
       "step": 1420
     },
     {
-      "epoch": 0.270956816257409,
-      "grad_norm": 0.07243089377880096,
+      "epoch": 0.12392160237516404,
+      "grad_norm": 0.5386276245117188,
       "learning_rate": 0.0002,
-      "loss": 0.0956,
+      "loss": 0.5021,
       "step": 1440
     },
     {
-      "epoch": 0.27472010537209524,
-      "grad_norm": 0.0814971774816513,
+      "epoch": 0.12564273574148577,
+      "grad_norm": 0.48680126667022705,
       "learning_rate": 0.0002,
-      "loss": 0.0869,
+      "loss": 0.4784,
       "step": 1460
     },
     {
-      "epoch": 0.27848339448678144,
-      "grad_norm": 0.07246191054582596,
+      "epoch": 0.1273638691078075,
+      "grad_norm": 0.44588467478752136,
       "learning_rate": 0.0002,
-      "loss": 0.0958,
+      "loss": 0.4905,
       "step": 1480
     },
     {
-      "epoch": 0.2822466836014677,
-      "grad_norm": 0.08997531235218048,
+      "epoch": 0.1290850024741292,
+      "grad_norm": 0.4805804491043091,
       "learning_rate": 0.0002,
-      "loss": 0.0875,
+      "loss": 0.4866,
       "step": 1500
     },
     {
-      "epoch": 0.2860099727161539,
-      "grad_norm": 0.08498572558164597,
+      "epoch": 0.13080613584045095,
+      "grad_norm": 0.44360053539276123,
       "learning_rate": 0.0002,
-      "loss": 0.0872,
+      "loss": 0.4637,
       "step": 1520
     },
     {
-      "epoch": 0.2897732618308402,
-      "grad_norm": 0.08210768550634384,
+      "epoch": 0.13252726920677266,
+      "grad_norm": 0.3818538784980774,
       "learning_rate": 0.0002,
-      "loss": 0.085,
+      "loss": 0.4512,
       "step": 1540
     },
     {
-      "epoch": 0.2935365509455264,
-      "grad_norm": 0.09801402688026428,
+      "epoch": 0.1342484025730944,
+      "grad_norm": 0.44495511054992676,
       "learning_rate": 0.0002,
-      "loss": 0.0937,
+      "loss": 0.4691,
       "step": 1560
     },
     {
-      "epoch": 0.29729984006021265,
-      "grad_norm": 0.07310175150632858,
+      "epoch": 0.1359695359394161,
+      "grad_norm": 0.4303475618362427,
       "learning_rate": 0.0002,
-      "loss": 0.0898,
+      "loss": 0.4551,
       "step": 1580
     },
     {
-      "epoch": 0.30106312917489886,
-      "grad_norm": 0.0679851844906807,
+      "epoch": 0.13769066930573784,
+      "grad_norm": 0.4990542531013489,
       "learning_rate": 0.0002,
-      "loss": 0.0853,
+      "loss": 0.4426,
       "step": 1600
     },
     {
-      "epoch": 0.3048264182895851,
-      "grad_norm": 0.0764717161655426,
+      "epoch": 0.13941180267205955,
+      "grad_norm": 0.4822593331336975,
       "learning_rate": 0.0002,
-      "loss": 0.0966,
+      "loss": 0.4746,
       "step": 1620
     },
     {
-      "epoch": 0.30858970740427133,
-      "grad_norm": 0.0813748836517334,
+      "epoch": 0.14113293603838128,
+      "grad_norm": 0.39103543758392334,
       "learning_rate": 0.0002,
-      "loss": 0.0914,
+      "loss": 0.4503,
       "step": 1640
     },
     {
-      "epoch": 0.3123529965189576,
-      "grad_norm": 0.07560984790325165,
+      "epoch": 0.142854069404703,
+      "grad_norm": 0.3490641117095947,
       "learning_rate": 0.0002,
-      "loss": 0.0889,
+      "loss": 0.4869,
       "step": 1660
     },
     {
-      "epoch": 0.3161162856336438,
-      "grad_norm": 0.07131631672382355,
+      "epoch": 0.14457520277102473,
+      "grad_norm": 0.3039151728153229,
       "learning_rate": 0.0002,
-      "loss": 0.0862,
+      "loss": 0.4885,
       "step": 1680
     },
     {
-      "epoch": 0.31987957474833006,
-      "grad_norm": 0.07922184467315674,
+      "epoch": 0.14629633613734644,
+      "grad_norm": 0.45525145530700684,
       "learning_rate": 0.0002,
-      "loss": 0.0908,
+      "loss": 0.4465,
       "step": 1700
     },
     {
-      "epoch": 0.3236428638630163,
-      "grad_norm": 0.09043222665786743,
+      "epoch": 0.14801746950366818,
+      "grad_norm": 0.3779124915599823,
       "learning_rate": 0.0002,
-      "loss": 0.0887,
+      "loss": 0.4077,
       "step": 1720
     },
     {
-      "epoch": 0.32740615297770254,
-      "grad_norm": 0.08574160188436508,
+      "epoch": 0.14973860286998988,
+      "grad_norm": 0.3984091579914093,
       "learning_rate": 0.0002,
-      "loss": 0.0895,
+      "loss": 0.4212,
       "step": 1740
     },
     {
-      "epoch": 0.33116944209238874,
-      "grad_norm": 0.0885438472032547,
+      "epoch": 0.15145973623631162,
+      "grad_norm": 0.37952274084091187,
       "learning_rate": 0.0002,
-      "loss": 0.0886,
+      "loss": 0.4158,
       "step": 1760
     },
     {
-      "epoch": 0.334932731207075,
-      "grad_norm": 0.08546678721904755,
+      "epoch": 0.15318086960263333,
+      "grad_norm": 0.34842512011528015,
       "learning_rate": 0.0002,
-      "loss": 0.0871,
+      "loss": 0.4188,
       "step": 1780
     },
     {
-      "epoch": 0.3386960203217612,
-      "grad_norm": 0.0922817662358284,
+      "epoch": 0.15490200296895507,
+      "grad_norm": 0.36399173736572266,
       "learning_rate": 0.0002,
-      "loss": 0.0877,
+      "loss": 0.4347,
       "step": 1800
     },
     {
-      "epoch": 0.3424593094364475,
-      "grad_norm": 0.08879639208316803,
+      "epoch": 0.15662313633527677,
+      "grad_norm": 0.38256826996803284,
       "learning_rate": 0.0002,
-      "loss": 0.0877,
+      "loss": 0.4179,
       "step": 1820
     },
     {
-      "epoch": 0.3462225985511337,
-      "grad_norm": 0.09202056378126144,
+      "epoch": 0.1583442697015985,
+      "grad_norm": 0.3715302050113678,
       "learning_rate": 0.0002,
-      "loss": 0.0827,
+      "loss": 0.4222,
       "step": 1840
     },
     {
-      "epoch": 0.34998588766581995,
-      "grad_norm": 0.0998295396566391,
+      "epoch": 0.16006540306792022,
+      "grad_norm": 0.3852146565914154,
       "learning_rate": 0.0002,
-      "loss": 0.0835,
+      "loss": 0.4005,
       "step": 1860
     },
     {
-      "epoch": 0.35374917678050616,
-      "grad_norm": 0.09490591287612915,
+      "epoch": 0.16178653643424196,
+      "grad_norm": 0.3589220643043518,
       "learning_rate": 0.0002,
-      "loss": 0.0862,
+      "loss": 0.4099,
       "step": 1880
     },
     {
-      "epoch": 0.3575124658951924,
-      "grad_norm": 0.08920489251613617,
+      "epoch": 0.16350766980056367,
+      "grad_norm": 0.5060593485832214,
       "learning_rate": 0.0002,
-      "loss": 0.0859,
+      "loss": 0.4158,
       "step": 1900
     },
     {
-      "epoch": 0.36127575500987863,
-      "grad_norm": 0.0803561583161354,
+      "epoch": 0.1652288031668854,
+      "grad_norm": 0.3826269507408142,
       "learning_rate": 0.0002,
-      "loss": 0.081,
+      "loss": 0.3874,
       "step": 1920
     },
     {
-      "epoch": 0.3650390441245649,
-      "grad_norm": 0.08496900647878647,
+      "epoch": 0.1669499365332071,
+      "grad_norm": 0.3576675057411194,
       "learning_rate": 0.0002,
-      "loss": 0.0835,
+      "loss": 0.4226,
       "step": 1940
     },
     {
-      "epoch": 0.3688023332392511,
-      "grad_norm": 0.08658849447965622,
+      "epoch": 0.16867106989952885,
+      "grad_norm": 0.37530502676963806,
       "learning_rate": 0.0002,
-      "loss": 0.0883,
+      "loss": 0.4014,
       "step": 1960
     },
     {
-      "epoch": 0.37256562235393736,
-      "grad_norm": 0.08497461676597595,
+      "epoch": 0.17039220326585056,
+      "grad_norm": 0.3857693374156952,
       "learning_rate": 0.0002,
-      "loss": 0.0811,
+      "loss": 0.413,
       "step": 1980
     },
     {
-      "epoch": 0.3763289114686236,
-      "grad_norm": 0.06599749624729156,
+      "epoch": 0.1721133366321723,
+      "grad_norm": 0.42137113213539124,
       "learning_rate": 0.0002,
-      "loss": 0.0714,
+      "loss": 0.4212,
       "step": 2000
     },
     {
-      "epoch": 0.38009220058330984,
-      "grad_norm": 0.07988911867141724,
+      "epoch": 0.173834469998494,
+      "grad_norm": 0.37976202368736267,
       "learning_rate": 0.0002,
-      "loss": 0.0874,
+      "loss": 0.3925,
       "step": 2020
     },
     {
-      "epoch": 0.38385548969799604,
-      "grad_norm": 0.07323348522186279,
+      "epoch": 0.17555560336481574,
+      "grad_norm": 0.3828701078891754,
       "learning_rate": 0.0002,
-      "loss": 0.0809,
+      "loss": 0.4004,
       "step": 2040
     },
     {
-      "epoch": 0.3876187788126823,
-      "grad_norm": 0.07455869019031525,
+      "epoch": 0.17727673673113745,
+      "grad_norm": 0.5018408894538879,
       "learning_rate": 0.0002,
-      "loss": 0.0819,
+      "loss": 0.384,
       "step": 2060
     },
     {
-      "epoch": 0.3913820679273685,
-      "grad_norm": 0.08383121341466904,
+      "epoch": 0.17899787009745918,
+      "grad_norm": 0.4595503509044647,
       "learning_rate": 0.0002,
-      "loss": 0.0811,
+      "loss": 0.4179,
       "step": 2080
     },
     {
-      "epoch": 0.3951453570420548,
-      "grad_norm": 0.07616332918405533,
+      "epoch": 0.1807190034637809,
+      "grad_norm": 0.5221086144447327,
       "learning_rate": 0.0002,
-      "loss": 0.0802,
+      "loss": 0.3855,
       "step": 2100
     },
     {
-      "epoch": 0.398908646156741,
-      "grad_norm": 0.08373293280601501,
+      "epoch": 0.18244013683010263,
+      "grad_norm": 0.5442349314689636,
       "learning_rate": 0.0002,
-      "loss": 0.0878,
+      "loss": 0.3579,
       "step": 2120
     },
     {
-      "epoch": 0.40267193527142725,
-      "grad_norm": 0.07459353655576706,
+      "epoch": 0.18416127019642434,
+      "grad_norm": 0.5430803298950195,
       "learning_rate": 0.0002,
-      "loss": 0.0806,
+      "loss": 0.3871,
       "step": 2140
     },
     {
-      "epoch": 0.40643522438611346,
-      "grad_norm": 0.08267400413751602,
+      "epoch": 0.18588240356274607,
+      "grad_norm": 0.2919999063014984,
       "learning_rate": 0.0002,
-      "loss": 0.0799,
+      "loss": 0.3947,
       "step": 2160
     },
     {
-      "epoch": 0.4101985135007997,
-      "grad_norm": 0.07844484597444534,
+      "epoch": 0.18760353692906778,
+      "grad_norm": 0.3997620940208435,
       "learning_rate": 0.0002,
-      "loss": 0.0843,
+      "loss": 0.3989,
       "step": 2180
     },
     {
-      "epoch": 0.41396180261548593,
-      "grad_norm": 0.07998470962047577,
+      "epoch": 0.18932467029538952,
+      "grad_norm": 0.38948822021484375,
       "learning_rate": 0.0002,
-      "loss": 0.0821,
+      "loss": 0.4292,
       "step": 2200
     },
     {
-      "epoch": 0.4177250917301722,
-      "grad_norm": 0.08860679715871811,
+      "epoch": 0.19104580366171123,
+      "grad_norm": 0.4080924689769745,
       "learning_rate": 0.0002,
-      "loss": 0.0819,
+      "loss": 0.398,
       "step": 2220
     },
     {
-      "epoch": 0.4214883808448584,
-      "grad_norm": 0.07845838367938995,
+      "epoch": 0.19276693702803296,
+      "grad_norm": 0.3982754647731781,
       "learning_rate": 0.0002,
-      "loss": 0.0802,
+      "loss": 0.4022,
       "step": 2240
     },
     {
-      "epoch": 0.42525166995954466,
-      "grad_norm": 0.08801402896642685,
+      "epoch": 0.19448807039435467,
+      "grad_norm": 0.41448819637298584,
       "learning_rate": 0.0002,
-      "loss": 0.0768,
+      "loss": 0.3968,
       "step": 2260
     },
     {
-      "epoch": 0.4290149590742309,
-      "grad_norm": 0.06149598956108093,
+      "epoch": 0.1962092037606764,
+      "grad_norm": 0.42457613348960876,
       "learning_rate": 0.0002,
-      "loss": 0.0806,
+      "loss": 0.4024,
       "step": 2280
     },
     {
-      "epoch": 0.43277824818891714,
-      "grad_norm": 0.07813749462366104,
+      "epoch": 0.19793033712699812,
+      "grad_norm": 0.2920919358730316,
       "learning_rate": 0.0002,
-      "loss": 0.0775,
+      "loss": 0.387,
       "step": 2300
     },
     {
-      "epoch": 0.43654153730360334,
-      "grad_norm": 0.08169027417898178,
+      "epoch": 0.19965147049331985,
+      "grad_norm": 0.4027709662914276,
       "learning_rate": 0.0002,
-      "loss": 0.0774,
+      "loss": 0.3835,
       "step": 2320
     },
     {
-      "epoch": 0.4403048264182896,
-      "grad_norm": 0.06748262792825699,
+      "epoch": 0.20137260385964156,
+      "grad_norm": 0.41888585686683655,
       "learning_rate": 0.0002,
-      "loss": 0.0721,
+      "loss": 0.3985,
       "step": 2340
     },
     {
-      "epoch": 0.4440681155329758,
-      "grad_norm": 0.07497742772102356,
+      "epoch": 0.2030937372259633,
+      "grad_norm": 0.4292152225971222,
       "learning_rate": 0.0002,
-      "loss": 0.0687,
+      "loss": 0.3823,
       "step": 2360
     },
     {
-      "epoch": 0.4478314046476621,
-      "grad_norm": 0.08851341158151627,
+      "epoch": 0.204814870592285,
+      "grad_norm": 0.4724443554878235,
       "learning_rate": 0.0002,
-      "loss": 0.0809,
+      "loss": 0.3589,
       "step": 2380
     },
     {
-      "epoch": 0.4515946937623483,
-      "grad_norm": 0.06805267184972763,
+      "epoch": 0.20653600395860675,
+      "grad_norm": 0.32000842690467834,
       "learning_rate": 0.0002,
-      "loss": 0.0783,
+      "loss": 0.383,
       "step": 2400
     },
     {
-      "epoch": 0.45535798287703455,
-      "grad_norm": 0.07998234033584595,
+      "epoch": 0.20825713732492845,
+      "grad_norm": 0.3765513300895691,
       "learning_rate": 0.0002,
-      "loss": 0.0816,
+      "loss": 0.371,
       "step": 2420
     },
     {
-      "epoch": 0.45912127199172076,
-      "grad_norm": 0.07358496636152267,
+      "epoch": 0.2099782706912502,
+      "grad_norm": 0.39066755771636963,
       "learning_rate": 0.0002,
-      "loss": 0.0715,
+      "loss": 0.3684,
       "step": 2440
     },
     {
-      "epoch": 0.462884561106407,
-      "grad_norm": 0.08105491101741791,
+      "epoch": 0.2116994040575719,
+      "grad_norm": 0.42424216866493225,
       "learning_rate": 0.0002,
-      "loss": 0.076,
+      "loss": 0.3653,
       "step": 2460
     },
     {
-      "epoch": 0.46664785022109323,
-      "grad_norm": 0.08625109493732452,
+      "epoch": 0.21342053742389364,
+      "grad_norm": 0.5066993832588196,
       "learning_rate": 0.0002,
-      "loss": 0.0731,
+      "loss": 0.3566,
       "step": 2480
     },
     {
-      "epoch": 0.4704111393357795,
-      "grad_norm": 0.07986035943031311,
+      "epoch": 0.21514167079021534,
+      "grad_norm": 0.44107547402381897,
       "learning_rate": 0.0002,
-      "loss": 0.0726,
+      "loss": 0.3748,
       "step": 2500
     },
     {
-      "epoch": 0.4741744284504657,
-      "grad_norm": 0.07247906923294067,
+      "epoch": 0.21686280415653708,
+      "grad_norm": 0.3890049457550049,
       "learning_rate": 0.0002,
-      "loss": 0.0763,
+      "loss": 0.3378,
       "step": 2520
     },
     {
-      "epoch": 0.47793771756515196,
-      "grad_norm": 0.08120191842317581,
+      "epoch": 0.2185839375228588,
+      "grad_norm": 0.3157922327518463,
       "learning_rate": 0.0002,
-      "loss": 0.0764,
+      "loss": 0.3462,
       "step": 2540
     },
     {
-      "epoch": 0.4817010066798382,
-      "grad_norm": 0.09049087017774582,
+      "epoch": 0.22030507088918053,
+      "grad_norm": 0.34473904967308044,
       "learning_rate": 0.0002,
-      "loss": 0.0799,
+      "loss": 0.3523,
       "step": 2560
     },
     {
-      "epoch": 0.48546429579452444,
-      "grad_norm": 0.06859074532985687,
+      "epoch": 0.22202620425550224,
+      "grad_norm": 0.38636985421180725,
       "learning_rate": 0.0002,
-      "loss": 0.0733,
+      "loss": 0.3694,
       "step": 2580
     },
     {
-      "epoch": 0.48922758490921064,
-      "grad_norm": 0.07872766256332397,
+      "epoch": 0.22374733762182397,
+      "grad_norm": 0.4448566436767578,
       "learning_rate": 0.0002,
-      "loss": 0.0724,
+      "loss": 0.3752,
       "step": 2600
     },
     {
-      "epoch": 0.4929908740238969,
-      "grad_norm": 0.0663181021809578,
+      "epoch": 0.22546847098814568,
+      "grad_norm": 0.3706173002719879,
       "learning_rate": 0.0002,
-      "loss": 0.0744,
+      "loss": 0.3706,
       "step": 2620
     },
     {
-      "epoch": 0.4967541631385831,
-      "grad_norm": 0.07880811393260956,
+      "epoch": 0.22718960435446742,
+      "grad_norm": 0.47221776843070984,
       "learning_rate": 0.0002,
-      "loss": 0.0759,
+      "loss": 0.3865,
       "step": 2640
     },
     {
-      "epoch": 0.5005174522532694,
-      "grad_norm": 0.12468737363815308,
+      "epoch": 0.22891073772078913,
+      "grad_norm": 0.39190933108329773,
       "learning_rate": 0.0002,
-      "loss": 0.0747,
+      "loss": 0.362,
       "step": 2660
     },
     {
-      "epoch": 0.5042807413679556,
-      "grad_norm": 0.08328507095575333,
+      "epoch": 0.23063187108711086,
+      "grad_norm": 0.36031046509742737,
       "learning_rate": 0.0002,
-      "loss": 0.0754,
+      "loss": 0.352,
       "step": 2680
     },
     {
-      "epoch": 0.5080440304826418,
-      "grad_norm": 0.06665025651454926,
+      "epoch": 0.2323530044534326,
+      "grad_norm": 0.4583648145198822,
       "learning_rate": 0.0002,
-      "loss": 0.0721,
+      "loss": 0.3343,
       "step": 2700
     },
     {
-      "epoch": 0.5118073195973281,
-      "grad_norm": 0.07980209589004517,
+      "epoch": 0.2340741378197543,
+      "grad_norm": 0.35927554965019226,
       "learning_rate": 0.0002,
-      "loss": 0.0733,
+      "loss": 0.3748,
       "step": 2720
     },
     {
-      "epoch": 0.5155706087120143,
-      "grad_norm": 0.07951670140028,
+      "epoch": 0.23579527118607604,
+      "grad_norm": 0.45593491196632385,
       "learning_rate": 0.0002,
-      "loss": 0.0792,
+      "loss": 0.356,
       "step": 2740
     },
     {
-      "epoch": 0.5193338978267006,
-      "grad_norm": 0.07515596598386765,
+      "epoch": 0.23751640455239775,
+      "grad_norm": 0.5094373226165771,
       "learning_rate": 0.0002,
-      "loss": 0.0724,
+      "loss": 0.3636,
       "step": 2760
     },
     {
-      "epoch": 0.5230971869413867,
-      "grad_norm": 0.08162270486354828,
+      "epoch": 0.2392375379187195,
+      "grad_norm": 0.47478821873664856,
       "learning_rate": 0.0002,
-      "loss": 0.0714,
+      "loss": 0.3537,
       "step": 2780
     },
     {
-      "epoch": 0.526860476056073,
-      "grad_norm": 0.08002326637506485,
+      "epoch": 0.2409586712850412,
+      "grad_norm": 0.37840044498443604,
       "learning_rate": 0.0002,
-      "loss": 0.077,
+      "loss": 0.3507,
       "step": 2800
     },
     {
-      "epoch": 0.5306237651707593,
-      "grad_norm": 0.07449716329574585,
+      "epoch": 0.24267980465136293,
+      "grad_norm": 0.41650140285491943,
       "learning_rate": 0.0002,
-      "loss": 0.0707,
+      "loss": 0.3405,
       "step": 2820
     },
     {
-      "epoch": 0.5343870542854455,
-      "grad_norm": 0.07990019023418427,
+      "epoch": 0.24440093801768464,
+      "grad_norm": 0.47170814871788025,
       "learning_rate": 0.0002,
-      "loss": 0.071,
+      "loss": 0.3564,
       "step": 2840
     },
     {
-      "epoch": 0.5381503434001317,
-      "grad_norm": 0.08538304269313812,
+      "epoch": 0.24612207138400638,
+      "grad_norm": 0.3727279305458069,
       "learning_rate": 0.0002,
-      "loss": 0.0707,
+      "loss": 0.3621,
       "step": 2860
     },
     {
-      "epoch": 0.541913632514818,
-      "grad_norm": 0.073255255818367,
+      "epoch": 0.2478432047503281,
+      "grad_norm": 0.2921255826950073,
       "learning_rate": 0.0002,
-      "loss": 0.0699,
+      "loss": 0.3504,
       "step": 2880
     },
     {
-      "epoch": 0.5456769216295042,
-      "grad_norm": 0.09101062268018723,
+      "epoch": 0.24956433811664983,
+      "grad_norm": 0.30992376804351807,
       "learning_rate": 0.0002,
-      "loss": 0.0708,
+      "loss": 0.3403,
       "step": 2900
     },
     {
-      "epoch": 0.5494402107441905,
-      "grad_norm": 0.0757337138056755,
+      "epoch": 0.25128547148297153,
+      "grad_norm": 0.3726598620414734,
       "learning_rate": 0.0002,
-      "loss": 0.0702,
+      "loss": 0.337,
       "step": 2920
     },
     {
-      "epoch": 0.5532034998588766,
-      "grad_norm": 0.06553716212511063,
+      "epoch": 0.25300660484929327,
+      "grad_norm": 0.46260905265808105,
       "learning_rate": 0.0002,
-      "loss": 0.0698,
+      "loss": 0.3428,
       "step": 2940
     },
     {
-      "epoch": 0.5569667889735629,
-      "grad_norm": 0.08524072915315628,
+      "epoch": 0.254727738215615,
+      "grad_norm": 0.4512200653553009,
       "learning_rate": 0.0002,
-      "loss": 0.0738,
+      "loss": 0.3456,
       "step": 2960
     },
     {
-      "epoch": 0.5607300780882492,
-      "grad_norm": 0.07361309975385666,
+      "epoch": 0.2564488715819367,
+      "grad_norm": 0.3799881041049957,
       "learning_rate": 0.0002,
-      "loss": 0.0695,
+      "loss": 0.3479,
       "step": 2980
     },
     {
-      "epoch": 0.5644933672029354,
-      "grad_norm": 0.07155182212591171,
+      "epoch": 0.2581700049482584,
+      "grad_norm": 0.36525946855545044,
       "learning_rate": 0.0002,
-      "loss": 0.0667,
+      "loss": 0.3324,
       "step": 3000
     },
     {
-      "epoch": 0.5682566563176216,
-      "grad_norm": 0.08088400214910507,
+      "epoch": 0.25989113831458016,
+      "grad_norm": 0.43525341153144836,
       "learning_rate": 0.0002,
-      "loss": 0.0695,
+      "loss": 0.3499,
       "step": 3020
     },
     {
-      "epoch": 0.5720199454323078,
-      "grad_norm": 0.08842818439006805,
+      "epoch": 0.2616122716809019,
+      "grad_norm": 0.39706552028656006,
       "learning_rate": 0.0002,
-      "loss": 0.0656,
+      "loss": 0.3319,
       "step": 3040
     },
     {
-      "epoch": 0.5757832345469941,
-      "grad_norm": 0.07135680317878723,
+      "epoch": 0.2633334050472236,
+      "grad_norm": 0.36725783348083496,
       "learning_rate": 0.0002,
-      "loss": 0.0673,
+      "loss": 0.3581,
       "step": 3060
     },
     {
-      "epoch": 0.5795465236616804,
-      "grad_norm": 0.08081484586000443,
+      "epoch": 0.2650545384135453,
+      "grad_norm": 0.4533792734146118,
       "learning_rate": 0.0002,
-      "loss": 0.0709,
+      "loss": 0.3377,
       "step": 3080
     },
     {
-      "epoch": 0.5833098127763665,
-      "grad_norm": 0.06999066472053528,
+      "epoch": 0.26677567177986705,
+      "grad_norm": 0.3752067983150482,
       "learning_rate": 0.0002,
-      "loss": 0.0684,
+      "loss": 0.3331,
       "step": 3100
     },
     {
-      "epoch": 0.5870731018910528,
-      "grad_norm": 0.06368447095155716,
+      "epoch": 0.2684968051461888,
+      "grad_norm": 0.31743401288986206,
       "learning_rate": 0.0002,
-      "loss": 0.0691,
+      "loss": 0.3118,
       "step": 3120
     },
     {
-      "epoch": 0.590836391005739,
-      "grad_norm": 0.08371565490961075,
+      "epoch": 0.27021793851251047,
+      "grad_norm": 0.45233970880508423,
       "learning_rate": 0.0002,
-      "loss": 0.0641,
+      "loss": 0.334,
       "step": 3140
     },
     {
-      "epoch": 0.5945996801204253,
-      "grad_norm": 0.0675625279545784,
+      "epoch": 0.2719390718788322,
+      "grad_norm": 0.32593509554862976,
       "learning_rate": 0.0002,
-      "loss": 0.069,
+      "loss": 0.3208,
       "step": 3160
     },
     {
-      "epoch": 0.5983629692351115,
-      "grad_norm": 0.08121863007545471,
+      "epoch": 0.27366020524515394,
+      "grad_norm": 0.3870694935321808,
       "learning_rate": 0.0002,
-      "loss": 0.065,
+      "loss": 0.3398,
       "step": 3180
     },
     {
-      "epoch": 0.6021262583497977,
-      "grad_norm": 0.06689095497131348,
+      "epoch": 0.2753813386114757,
+      "grad_norm": 0.39160028100013733,
       "learning_rate": 0.0002,
-      "loss": 0.0675,
+      "loss": 0.3209,
       "step": 3200
     },
     {
-      "epoch": 0.605889547464484,
-      "grad_norm": 0.06634503602981567,
+      "epoch": 0.27710247197779736,
+      "grad_norm": 0.3718438148498535,
       "learning_rate": 0.0002,
-      "loss": 0.064,
+      "loss": 0.2982,
       "step": 3220
     },
     {
-      "epoch": 0.6096528365791702,
-      "grad_norm": 0.07062681019306183,
+      "epoch": 0.2788236053441191,
+      "grad_norm": 0.4262580871582031,
       "learning_rate": 0.0002,
-      "loss": 0.0698,
+      "loss": 0.2977,
       "step": 3240
     },
     {
-      "epoch": 0.6134161256938564,
-      "grad_norm": 0.07908321917057037,
+      "epoch": 0.28054473871044083,
+      "grad_norm": 0.33147528767585754,
       "learning_rate": 0.0002,
-      "loss": 0.0674,
+      "loss": 0.3345,
       "step": 3260
     },
     {
-      "epoch": 0.6171794148085427,
-      "grad_norm": 0.08132966607809067,
+      "epoch": 0.28226587207676257,
+      "grad_norm": 0.3187743127346039,
       "learning_rate": 0.0002,
-      "loss": 0.0694,
+      "loss": 0.3325,
       "step": 3280
     },
     {
-      "epoch": 0.6209427039232289,
-      "grad_norm": 0.081205353140831,
+      "epoch": 0.28398700544308425,
+      "grad_norm": 0.4593104422092438,
       "learning_rate": 0.0002,
-      "loss": 0.0627,
+      "loss": 0.3101,
       "step": 3300
     },
     {
-      "epoch": 0.6247059930379152,
-      "grad_norm": 0.07323930412530899,
+      "epoch": 0.285708138809406,
+      "grad_norm": 0.431805819272995,
       "learning_rate": 0.0002,
-      "loss": 0.0645,
+      "loss": 0.2995,
       "step": 3320
     },
     {
-      "epoch": 0.6284692821526013,
-      "grad_norm": 0.07979920506477356,
+      "epoch": 0.2874292721757277,
+      "grad_norm": 0.3650576174259186,
       "learning_rate": 0.0002,
-      "loss": 0.0601,
+      "loss": 0.3176,
       "step": 3340
     },
     {
-      "epoch": 0.6322325712672876,
-      "grad_norm": 0.06519381701946259,
+      "epoch": 0.28915040554204946,
+      "grad_norm": 0.3000945746898651,
       "learning_rate": 0.0002,
-      "loss": 0.0677,
+      "loss": 0.3364,
       "step": 3360
     },
     {
-      "epoch": 0.6359958603819739,
-      "grad_norm": 0.07861756533384323,
+      "epoch": 0.29087153890837114,
+      "grad_norm": 0.3499468266963959,
       "learning_rate": 0.0002,
-      "loss": 0.0648,
+      "loss": 0.2956,
       "step": 3380
     },
     {
-      "epoch": 0.6397591494966601,
-      "grad_norm": 0.09498826414346695,
+      "epoch": 0.2925926722746929,
+      "grad_norm": 0.41105055809020996,
       "learning_rate": 0.0002,
-      "loss": 0.0644,
+      "loss": 0.3516,
       "step": 3400
     },
     {
-      "epoch": 0.6435224386113463,
-      "grad_norm": 0.07069331407546997,
+      "epoch": 0.2943138056410146,
+      "grad_norm": 0.4257635772228241,
       "learning_rate": 0.0002,
-      "loss": 0.0683,
+      "loss": 0.3098,
       "step": 3420
     },
     {
-      "epoch": 0.6472857277260325,
-      "grad_norm": 0.07212232798337936,
+      "epoch": 0.29603493900733635,
+      "grad_norm": 0.3407900035381317,
       "learning_rate": 0.0002,
-      "loss": 0.061,
+      "loss": 0.302,
       "step": 3440
     },
     {
-      "epoch": 0.6510490168407188,
-      "grad_norm": 0.06527985632419586,
+      "epoch": 0.29775607237365803,
+      "grad_norm": 0.3562329113483429,
       "learning_rate": 0.0002,
-      "loss": 0.063,
+      "loss": 0.3066,
       "step": 3460
     },
     {
-      "epoch": 0.6548123059554051,
-      "grad_norm": 0.09721993654966354,
+      "epoch": 0.29947720573997977,
+      "grad_norm": 0.4639800786972046,
       "learning_rate": 0.0002,
-      "loss": 0.0699,
+      "loss": 0.3108,
       "step": 3480
     },
     {
-      "epoch": 0.6585755950700912,
-      "grad_norm": 0.06381040811538696,
+      "epoch": 0.3011983391063015,
+      "grad_norm": 0.47634968161582947,
       "learning_rate": 0.0002,
-      "loss": 0.0604,
+      "loss": 0.3185,
       "step": 3500
     },
     {
-      "epoch": 0.6623388841847775,
-      "grad_norm": 0.07853078097105026,
+      "epoch": 0.30291947247262324,
+      "grad_norm": 0.42725369334220886,
       "learning_rate": 0.0002,
-      "loss": 0.0649,
+      "loss": 0.3081,
       "step": 3520
     },
     {
-      "epoch": 0.6661021732994638,
-      "grad_norm": 0.09212848544120789,
+      "epoch": 0.3046406058389449,
+      "grad_norm": 0.38221365213394165,
       "learning_rate": 0.0002,
-      "loss": 0.0588,
+      "loss": 0.3282,
       "step": 3540
     },
     {
-      "epoch": 0.66986546241415,
-      "grad_norm": 0.07097720354795456,
+      "epoch": 0.30636173920526666,
+      "grad_norm": 0.3222472369670868,
       "learning_rate": 0.0002,
-      "loss": 0.0635,
+      "loss": 0.2975,
       "step": 3560
     },
     {
-      "epoch": 0.6736287515288362,
-      "grad_norm": 0.08024760335683823,
+      "epoch": 0.3080828725715884,
+      "grad_norm": 0.3551533818244934,
       "learning_rate": 0.0002,
-      "loss": 0.068,
+      "loss": 0.308,
       "step": 3580
     },
     {
-      "epoch": 0.6773920406435224,
-      "grad_norm": 0.07371170073747635,
+      "epoch": 0.30980400593791013,
+      "grad_norm": 0.35602936148643494,
       "learning_rate": 0.0002,
-      "loss": 0.0652,
+      "loss": 0.3225,
       "step": 3600
     },
     {
-      "epoch": 0.6811553297582087,
-      "grad_norm": 0.0894666537642479,
+      "epoch": 0.3115251393042318,
+      "grad_norm": 0.36850786209106445,
       "learning_rate": 0.0002,
-      "loss": 0.0659,
+      "loss": 0.293,
       "step": 3620
     },
     {
-      "epoch": 0.684918618872895,
-      "grad_norm": 0.07533634454011917,
+      "epoch": 0.31324627267055355,
+      "grad_norm": 0.4266447126865387,
       "learning_rate": 0.0002,
-      "loss": 0.0656,
+      "loss": 0.3055,
       "step": 3640
     },
     {
-      "epoch": 0.6886819079875811,
-      "grad_norm": 0.09198255091905594,
+      "epoch": 0.3149674060368753,
+      "grad_norm": 0.3386904001235962,
       "learning_rate": 0.0002,
-      "loss": 0.0645,
+      "loss": 0.3046,
       "step": 3660
     },
     {
-      "epoch": 0.6924451971022674,
-      "grad_norm": 0.07213396579027176,
+      "epoch": 0.316688539403197,
+      "grad_norm": 0.39958760142326355,
       "learning_rate": 0.0002,
-      "loss": 0.0634,
+      "loss": 0.3177,
       "step": 3680
     },
     {
-      "epoch": 0.6962084862169536,
-      "grad_norm": 0.06533551216125488,
+      "epoch": 0.3184096727695187,
+      "grad_norm": 0.3587161600589752,
       "learning_rate": 0.0002,
-      "loss": 0.0666,
+      "loss": 0.3183,
       "step": 3700
     },
     {
-      "epoch": 0.6999717753316399,
-      "grad_norm": 0.09535999596118927,
+      "epoch": 0.32013080613584044,
+      "grad_norm": 0.37195342779159546,
       "learning_rate": 0.0002,
-      "loss": 0.0651,
+      "loss": 0.2984,
       "step": 3720
     },
     {
-      "epoch": 0.703735064446326,
-      "grad_norm": 0.07631547003984451,
+      "epoch": 0.3218519395021622,
+      "grad_norm": 0.42094019055366516,
       "learning_rate": 0.0002,
-      "loss": 0.0678,
+      "loss": 0.3019,
       "step": 3740
     },
     {
-      "epoch": 0.7074983535610123,
-      "grad_norm": 0.09607011079788208,
+      "epoch": 0.3235730728684839,
+      "grad_norm": 0.2889210879802704,
       "learning_rate": 0.0002,
-      "loss": 0.0681,
+      "loss": 0.3229,
       "step": 3760
     },
     {
-      "epoch": 0.7112616426756986,
-      "grad_norm": 0.07209835946559906,
+      "epoch": 0.3252942062348056,
+      "grad_norm": 0.4362521767616272,
       "learning_rate": 0.0002,
-      "loss": 0.068,
+      "loss": 0.2891,
       "step": 3780
     },
     {
-      "epoch": 0.7150249317903848,
-      "grad_norm": 0.09137856960296631,
+      "epoch": 0.32701533960112733,
+      "grad_norm": 0.3560026288032532,
       "learning_rate": 0.0002,
-      "loss": 0.0598,
+      "loss": 0.3042,
       "step": 3800
     },
     {
-      "epoch": 0.718788220905071,
-      "grad_norm": 0.09964293241500854,
+      "epoch": 0.32873647296744907,
+      "grad_norm": 0.5475091338157654,
       "learning_rate": 0.0002,
-      "loss": 0.0656,
+      "loss": 0.2949,
       "step": 3820
     },
     {
-      "epoch": 0.7225515100197573,
-      "grad_norm": 0.07041902840137482,
+      "epoch": 0.3304576063337708,
+      "grad_norm": 0.3368379473686218,
       "learning_rate": 0.0002,
-      "loss": 0.0617,
+      "loss": 0.3203,
       "step": 3840
     },
     {
-      "epoch": 0.7263147991344435,
-      "grad_norm": 0.05418103560805321,
+      "epoch": 0.3321787397000925,
+      "grad_norm": 0.45856744050979614,
       "learning_rate": 0.0002,
-      "loss": 0.0686,
+      "loss": 0.3107,
       "step": 3860
     },
     {
-      "epoch": 0.7300780882491298,
-      "grad_norm": 0.07180003076791763,
+      "epoch": 0.3338998730664142,
+      "grad_norm": 0.47050952911376953,
       "learning_rate": 0.0002,
-      "loss": 0.0627,
+      "loss": 0.2882,
       "step": 3880
     },
     {
-      "epoch": 0.7338413773638159,
-      "grad_norm": 0.07342009246349335,
+      "epoch": 0.33562100643273596,
+      "grad_norm": 0.5037484169006348,
       "learning_rate": 0.0002,
-      "loss": 0.0651,
+      "loss": 0.3033,
       "step": 3900
     },
     {
-      "epoch": 0.7376046664785022,
-      "grad_norm": 0.08151030540466309,
+      "epoch": 0.3373421397990577,
+      "grad_norm": 0.34310051798820496,
       "learning_rate": 0.0002,
-      "loss": 0.0645,
+      "loss": 0.3099,
       "step": 3920
     },
     {
-      "epoch": 0.7413679555931885,
-      "grad_norm": 0.09977607429027557,
+      "epoch": 0.3390632731653794,
+      "grad_norm": 0.47114118933677673,
       "learning_rate": 0.0002,
-      "loss": 0.0606,
+      "loss": 0.2892,
       "step": 3940
     },
     {
-      "epoch": 0.7451312447078747,
-      "grad_norm": 0.06991346180438995,
+      "epoch": 0.3407844065317011,
+      "grad_norm": 0.4255194365978241,
       "learning_rate": 0.0002,
-      "loss": 0.064,
+      "loss": 0.3022,
       "step": 3960
     },
     {
-      "epoch": 0.7488945338225609,
-      "grad_norm": 0.06797617673873901,
+      "epoch": 0.34250553989802285,
+      "grad_norm": 0.5251383185386658,
       "learning_rate": 0.0002,
-      "loss": 0.0595,
+      "loss": 0.3114,
       "step": 3980
     },
     {
-      "epoch": 0.7526578229372471,
-      "grad_norm": 0.06056559830904007,
+      "epoch": 0.3442266732643446,
+      "grad_norm": 0.29961732029914856,
       "learning_rate": 0.0002,
-      "loss": 0.0595,
+      "loss": 0.3093,
       "step": 4000
     },
     {
-      "epoch": 0.7564211120519334,
-      "grad_norm": 0.06775388866662979,
+      "epoch": 0.3459478066306663,
+      "grad_norm": 0.4707966446876526,
       "learning_rate": 0.0002,
-      "loss": 0.0624,
+      "loss": 0.3139,
       "step": 4020
     },
     {
-      "epoch": 0.7601844011666197,
-      "grad_norm": 0.08091945201158524,
+      "epoch": 0.347668939996988,
+      "grad_norm": 0.33052656054496765,
       "learning_rate": 0.0002,
-      "loss": 0.0657,
+      "loss": 0.277,
       "step": 4040
     },
     {
-      "epoch": 0.7639476902813058,
-      "grad_norm": 0.1335289478302002,
+      "epoch": 0.34939007336330974,
+      "grad_norm": 0.3629648685455322,
       "learning_rate": 0.0002,
-      "loss": 0.061,
+      "loss": 0.2934,
       "step": 4060
     },
     {
-      "epoch": 0.7677109793959921,
-      "grad_norm": 0.07901336252689362,
+      "epoch": 0.3511112067296315,
+      "grad_norm": 0.3852795362472534,
       "learning_rate": 0.0002,
-      "loss": 0.0635,
+      "loss": 0.2885,
       "step": 4080
     },
     {
-      "epoch": 0.7714742685106784,
-      "grad_norm": 0.07874254137277603,
+      "epoch": 0.3528323400959532,
+      "grad_norm": 0.3913141191005707,
       "learning_rate": 0.0002,
-      "loss": 0.066,
+      "loss": 0.3009,
       "step": 4100
     },
     {
-      "epoch": 0.7752375576253646,
-      "grad_norm": 0.07547064125537872,
+      "epoch": 0.3545534734622749,
+      "grad_norm": 0.4364122152328491,
       "learning_rate": 0.0002,
-      "loss": 0.061,
+      "loss": 0.2825,
       "step": 4120
     },
     {
-      "epoch": 0.7790008467400508,
-      "grad_norm": 0.06392517685890198,
+      "epoch": 0.35627460682859663,
+      "grad_norm": 0.3607100248336792,
       "learning_rate": 0.0002,
-      "loss": 0.0595,
+      "loss": 0.2969,
       "step": 4140
     },
     {
-      "epoch": 0.782764135854737,
-      "grad_norm": 0.08115773648023605,
+      "epoch": 0.35799574019491837,
+      "grad_norm": 0.41620948910713196,
       "learning_rate": 0.0002,
-      "loss": 0.0633,
+      "loss": 0.2735,
       "step": 4160
     },
     {
-      "epoch": 0.7865274249694233,
-      "grad_norm": 0.07906658202409744,
+      "epoch": 0.3597168735612401,
+      "grad_norm": 0.36478644609451294,
       "learning_rate": 0.0002,
-      "loss": 0.06,
+      "loss": 0.2826,
       "step": 4180
     },
     {
-      "epoch": 0.7902907140841096,
-      "grad_norm": 0.8131846189498901,
+      "epoch": 0.3614380069275618,
+      "grad_norm": 0.39005762338638306,
       "learning_rate": 0.0002,
-      "loss": 0.0642,
+      "loss": 0.2945,
       "step": 4200
     },
     {
-      "epoch": 0.7940540031987957,
-      "grad_norm": 0.08816728740930557,
+      "epoch": 0.3631591402938835,
+      "grad_norm": 0.3086032271385193,
       "learning_rate": 0.0002,
-      "loss": 0.0683,
+      "loss": 0.3057,
       "step": 4220
     },
     {
-      "epoch": 0.797817292313482,
-      "grad_norm": 0.08981813490390778,
+      "epoch": 0.36488027366020526,
+      "grad_norm": 0.4346630275249481,
       "learning_rate": 0.0002,
-      "loss": 0.0654,
+      "loss": 0.2667,
       "step": 4240
     },
     {
-      "epoch": 0.8015805814281682,
-      "grad_norm": 0.08015542477369308,
+      "epoch": 0.366601407026527,
+      "grad_norm": 0.3394578993320465,
       "learning_rate": 0.0002,
-      "loss": 0.0623,
+      "loss": 0.3107,
       "step": 4260
     },
     {
-      "epoch": 0.8053438705428545,
-      "grad_norm": 0.08604927361011505,
+      "epoch": 0.3683225403928487,
+      "grad_norm": 0.44261375069618225,
       "learning_rate": 0.0002,
-      "loss": 0.0616,
+      "loss": 0.2968,
       "step": 4280
     },
     {
-      "epoch": 0.8091071596575407,
-      "grad_norm": 0.08916622400283813,
+      "epoch": 0.3700436737591704,
+      "grad_norm": 0.36678528785705566,
       "learning_rate": 0.0002,
-      "loss": 0.0651,
+      "loss": 0.3023,
       "step": 4300
     },
     {
-      "epoch": 0.8128704487722269,
-      "grad_norm": 0.06978412717580795,
+      "epoch": 0.37176480712549215,
+      "grad_norm": 0.43799522519111633,
       "learning_rate": 0.0002,
-      "loss": 0.0577,
+      "loss": 0.2781,
       "step": 4320
     },
     {
-      "epoch": 0.8166337378869132,
-      "grad_norm": 0.089443139731884,
+      "epoch": 0.3734859404918139,
+      "grad_norm": 0.3344813883304596,
       "learning_rate": 0.0002,
-      "loss": 0.0651,
+      "loss": 0.2818,
       "step": 4340
     },
     {
-      "epoch": 0.8203970270015994,
-      "grad_norm": 0.08108926564455032,
+      "epoch": 0.37520707385813556,
+      "grad_norm": 0.519148588180542,
       "learning_rate": 0.0002,
-      "loss": 0.0584,
+      "loss": 0.2897,
       "step": 4360
     },
     {
-      "epoch": 0.8241603161162856,
-      "grad_norm": 0.0913953110575676,
+      "epoch": 0.3769282072244573,
+      "grad_norm": 0.36793655157089233,
       "learning_rate": 0.0002,
-      "loss": 0.0611,
+      "loss": 0.3068,
       "step": 4380
     },
     {
-      "epoch": 0.8279236052309719,
-      "grad_norm": 0.07268033176660538,
+      "epoch": 0.37864934059077904,
+      "grad_norm": 0.37348249554634094,
       "learning_rate": 0.0002,
-      "loss": 0.0601,
+      "loss": 0.2849,
       "step": 4400
     },
     {
-      "epoch": 0.8316868943456581,
-      "grad_norm": 0.07602666318416595,
+      "epoch": 0.3803704739571008,
+      "grad_norm": 0.44873690605163574,
       "learning_rate": 0.0002,
-      "loss": 0.0642,
+      "loss": 0.2902,
       "step": 4420
     },
     {
-      "epoch": 0.8354501834603444,
-      "grad_norm": 0.06899157911539078,
+      "epoch": 0.38209160732342246,
+      "grad_norm": 0.32298997044563293,
       "learning_rate": 0.0002,
-      "loss": 0.0603,
+      "loss": 0.2892,
       "step": 4440
     },
     {
-      "epoch": 0.8392134725750305,
-      "grad_norm": 0.07310888916254044,
+      "epoch": 0.3838127406897442,
+      "grad_norm": 0.4323394298553467,
       "learning_rate": 0.0002,
-      "loss": 0.0614,
+      "loss": 0.288,
       "step": 4460
     },
     {
-      "epoch": 0.8429767616897168,
-      "grad_norm": 0.086586132645607,
+      "epoch": 0.38553387405606593,
+      "grad_norm": 0.3986116945743561,
       "learning_rate": 0.0002,
-      "loss": 0.0636,
+      "loss": 0.2962,
       "step": 4480
     },
     {
-      "epoch": 0.8467400508044031,
-      "grad_norm": 0.04938528314232826,
+      "epoch": 0.38725500742238766,
+      "grad_norm": 0.4980430603027344,
       "learning_rate": 0.0002,
-      "loss": 0.0577,
+      "loss": 0.2846,
       "step": 4500
     },
     {
-      "epoch": 0.8505033399190893,
-      "grad_norm": 0.07192698121070862,
+      "epoch": 0.38897614078870935,
+      "grad_norm": 0.4195205569267273,
       "learning_rate": 0.0002,
-      "loss": 0.0574,
+      "loss": 0.2806,
       "step": 4520
     },
     {
-      "epoch": 0.8542666290337755,
-      "grad_norm": 0.06893257051706314,
+      "epoch": 0.3906972741550311,
+      "grad_norm": 0.2984451949596405,
       "learning_rate": 0.0002,
-      "loss": 0.0604,
+      "loss": 0.2796,
       "step": 4540
     },
     {
-      "epoch": 0.8580299181484617,
-      "grad_norm": 0.06691521406173706,
+      "epoch": 0.3924184075213528,
+      "grad_norm": 0.3897689878940582,
       "learning_rate": 0.0002,
-      "loss": 0.0649,
+      "loss": 0.2842,
       "step": 4560
     },
     {
-      "epoch": 0.861793207263148,
-      "grad_norm": 0.09420821070671082,
+      "epoch": 0.39413954088767456,
+      "grad_norm": 0.40885892510414124,
       "learning_rate": 0.0002,
-      "loss": 0.0598,
+      "loss": 0.2801,
       "step": 4580
     },
     {
-      "epoch": 0.8655564963778343,
-      "grad_norm": 0.07227174192667007,
+      "epoch": 0.39586067425399624,
+      "grad_norm": 0.41099223494529724,
       "learning_rate": 0.0002,
-      "loss": 0.062,
+      "loss": 0.276,
       "step": 4600
     },
     {
-      "epoch": 0.8693197854925204,
-      "grad_norm": 0.07998435199260712,
+      "epoch": 0.397581807620318,
+      "grad_norm": 0.4116320312023163,
       "learning_rate": 0.0002,
-      "loss": 0.0585,
+      "loss": 0.2898,
       "step": 4620
     },
     {
-      "epoch": 0.8730830746072067,
-      "grad_norm": 0.07850491255521774,
+      "epoch": 0.3993029409866397,
+      "grad_norm": 0.37863150238990784,
       "learning_rate": 0.0002,
-      "loss": 0.0586,
+      "loss": 0.268,
       "step": 4640
     },
     {
-      "epoch": 0.876846363721893,
-      "grad_norm": 0.09450593590736389,
+      "epoch": 0.40102407435296145,
+      "grad_norm": 0.43875718116760254,
       "learning_rate": 0.0002,
-      "loss": 0.0625,
+      "loss": 0.277,
       "step": 4660
     },
     {
-      "epoch": 0.8806096528365792,
-      "grad_norm": 0.08073689788579941,
+      "epoch": 0.4027452077192831,
+      "grad_norm": 0.3862336277961731,
       "learning_rate": 0.0002,
-      "loss": 0.062,
+      "loss": 0.269,
       "step": 4680
     },
     {
-      "epoch": 0.8843729419512654,
-      "grad_norm": 0.06979519873857498,
+      "epoch": 0.40446634108560486,
+      "grad_norm": 0.330501914024353,
       "learning_rate": 0.0002,
-      "loss": 0.0584,
+      "loss": 0.2754,
       "step": 4700
     },
     {
-      "epoch": 0.8881362310659516,
-      "grad_norm": 0.06407754868268967,
+      "epoch": 0.4061874744519266,
+      "grad_norm": 0.3258209526538849,
       "learning_rate": 0.0002,
-      "loss": 0.0626,
+      "loss": 0.2933,
       "step": 4720
     },
     {
-      "epoch": 0.8918995201806379,
-      "grad_norm": 0.2963426411151886,
+      "epoch": 0.40790860781824834,
+      "grad_norm": 0.4297473430633545,
       "learning_rate": 0.0002,
-      "loss": 0.0625,
+      "loss": 0.2726,
       "step": 4740
     },
     {
-      "epoch": 0.8956628092953242,
-      "grad_norm": 0.08902490139007568,
+      "epoch": 0.40962974118457,
+      "grad_norm": 0.36448150873184204,
       "learning_rate": 0.0002,
-      "loss": 0.0621,
+      "loss": 0.2645,
       "step": 4760
     },
     {
-      "epoch": 0.8994260984100103,
-      "grad_norm": 0.08447249978780746,
+      "epoch": 0.41135087455089175,
+      "grad_norm": 0.4198833107948303,
       "learning_rate": 0.0002,
-      "loss": 0.0638,
+      "loss": 0.3029,
       "step": 4780
     },
     {
-      "epoch": 0.9031893875246966,
-      "grad_norm": 0.07621373981237411,
+      "epoch": 0.4130720079172135,
+      "grad_norm": 0.35666197538375854,
       "learning_rate": 0.0002,
-      "loss": 0.0618,
+      "loss": 0.2443,
       "step": 4800
     },
     {
-      "epoch": 0.9069526766393828,
-      "grad_norm": 0.07609863579273224,
+      "epoch": 0.4147931412835352,
+      "grad_norm": 0.38097846508026123,
       "learning_rate": 0.0002,
-      "loss": 0.0627,
+      "loss": 0.2706,
       "step": 4820
     },
     {
-      "epoch": 0.9107159657540691,
-      "grad_norm": 0.0785209983587265,
+      "epoch": 0.4165142746498569,
+      "grad_norm": 0.3574432134628296,
       "learning_rate": 0.0002,
-      "loss": 0.0606,
+      "loss": 0.2791,
       "step": 4840
     },
     {
-      "epoch": 0.9144792548687553,
-      "grad_norm": 0.08515089750289917,
+      "epoch": 0.41823540801617864,
+      "grad_norm": 0.46495670080184937,
       "learning_rate": 0.0002,
-      "loss": 0.0577,
+      "loss": 0.2753,
       "step": 4860
     },
     {
-      "epoch": 0.9182425439834415,
-      "grad_norm": 0.0810500979423523,
+      "epoch": 0.4199565413825004,
+      "grad_norm": 0.4777359664440155,
       "learning_rate": 0.0002,
-      "loss": 0.0578,
+      "loss": 0.2717,
       "step": 4880
     },
     {
-      "epoch": 0.9220058330981278,
-      "grad_norm": 0.08074364811182022,
+      "epoch": 0.4216776747488221,
+      "grad_norm": 0.4285059869289398,
       "learning_rate": 0.0002,
-      "loss": 0.0556,
+      "loss": 0.271,
       "step": 4900
     },
     {
-      "epoch": 0.925769122212814,
-      "grad_norm": 0.06328209489583969,
+      "epoch": 0.4233988081151438,
+      "grad_norm": 0.45529553294181824,
       "learning_rate": 0.0002,
-      "loss": 0.0555,
+      "loss": 0.268,
       "step": 4920
     },
     {
-      "epoch": 0.9295324113275002,
-      "grad_norm": 0.06556906551122665,
+      "epoch": 0.42511994148146554,
+      "grad_norm": 0.3446490466594696,
       "learning_rate": 0.0002,
-      "loss": 0.056,
+      "loss": 0.2659,
       "step": 4940
     },
     {
-      "epoch": 0.9332957004421865,
-      "grad_norm": 0.0747644379734993,
+      "epoch": 0.42684107484778727,
+      "grad_norm": 0.3138258457183838,
       "learning_rate": 0.0002,
-      "loss": 0.0608,
+      "loss": 0.2845,
       "step": 4960
     },
     {
-      "epoch": 0.9370589895568727,
-      "grad_norm": 0.07274675369262695,
+      "epoch": 0.428562208214109,
+      "grad_norm": 0.3556279242038727,
       "learning_rate": 0.0002,
-      "loss": 0.057,
+      "loss": 0.2688,
       "step": 4980
     },
     {
-      "epoch": 0.940822278671559,
-      "grad_norm": 0.061261050403118134,
+      "epoch": 0.4302833415804307,
+      "grad_norm": 0.4338746964931488,
       "learning_rate": 0.0002,
-      "loss": 0.0587,
+      "loss": 0.272,
       "step": 5000
     },
     {
-      "epoch": 0.9445855677862451,
-      "grad_norm": 0.08277834206819534,
+      "epoch": 0.4320044749467524,
+      "grad_norm": 0.3764958083629608,
       "learning_rate": 0.0002,
-      "loss": 0.0608,
+      "loss": 0.2618,
       "step": 5020
     },
     {
-      "epoch": 0.9483488569009314,
-      "grad_norm": 0.08663126826286316,
+      "epoch": 0.43372560831307416,
+      "grad_norm": 0.3759072721004486,
       "learning_rate": 0.0002,
-      "loss": 0.0595,
+      "loss": 0.2657,
       "step": 5040
     },
     {
-      "epoch": 0.9521121460156177,
-      "grad_norm": 0.06938447803258896,
+      "epoch": 0.4354467416793959,
+      "grad_norm": 0.3582591414451599,
       "learning_rate": 0.0002,
-      "loss": 0.0567,
+      "loss": 0.3033,
       "step": 5060
     },
     {
-      "epoch": 0.9558754351303039,
-      "grad_norm": 0.07865908741950989,
+      "epoch": 0.4371678750457176,
+      "grad_norm": 0.3076396584510803,
       "learning_rate": 0.0002,
-      "loss": 0.0588,
+      "loss": 0.2719,
       "step": 5080
     },
     {
-      "epoch": 0.9596387242449901,
-      "grad_norm": 0.061116304248571396,
+      "epoch": 0.4388890084120393,
+      "grad_norm": 0.4947834610939026,
       "learning_rate": 0.0002,
-      "loss": 0.0574,
+      "loss": 0.2661,
       "step": 5100
     },
     {
-      "epoch": 0.9634020133596763,
-      "grad_norm": 0.05689473822712898,
+      "epoch": 0.44061014177836105,
+      "grad_norm": 0.4582163095474243,
       "learning_rate": 0.0002,
-      "loss": 0.0533,
+      "loss": 0.2687,
       "step": 5120
     },
     {
-      "epoch": 0.9671653024743626,
-      "grad_norm": 0.07368452847003937,
+      "epoch": 0.4423312751446828,
+      "grad_norm": 0.3605941832065582,
       "learning_rate": 0.0002,
-      "loss": 0.0591,
+      "loss": 0.2692,
       "step": 5140
     },
     {
-      "epoch": 0.9709285915890489,
-      "grad_norm": 0.06758634001016617,
+      "epoch": 0.44405240851100447,
+      "grad_norm": 0.37437519431114197,
       "learning_rate": 0.0002,
-      "loss": 0.0572,
+      "loss": 0.264,
       "step": 5160
     },
     {
-      "epoch": 0.974691880703735,
-      "grad_norm": 0.07348816096782684,
+      "epoch": 0.4457735418773262,
+      "grad_norm": 0.3707126975059509,
       "learning_rate": 0.0002,
-      "loss": 0.0577,
+      "loss": 0.2535,
       "step": 5180
     },
     {
-      "epoch": 0.9784551698184213,
-      "grad_norm": 0.07613357156515121,
+      "epoch": 0.44749467524364794,
+      "grad_norm": 0.3294564485549927,
       "learning_rate": 0.0002,
-      "loss": 0.0566,
+      "loss": 0.2781,
       "step": 5200
     },
     {
-      "epoch": 0.9822184589331076,
-      "grad_norm": 0.062316812574863434,
+      "epoch": 0.4492158086099697,
+      "grad_norm": 0.3913413882255554,
       "learning_rate": 0.0002,
-      "loss": 0.0621,
+      "loss": 0.2532,
       "step": 5220
     },
     {
-      "epoch": 0.9859817480477938,
-      "grad_norm": 0.08600709587335587,
+      "epoch": 0.45093694197629136,
+      "grad_norm": 0.31665486097335815,
       "learning_rate": 0.0002,
-      "loss": 0.0576,
+      "loss": 0.2731,
       "step": 5240
     },
     {
-      "epoch": 0.98974503716248,
-      "grad_norm": 0.06687742471694946,
+      "epoch": 0.4526580753426131,
+      "grad_norm": 0.405577689409256,
       "learning_rate": 0.0002,
-      "loss": 0.0571,
+      "loss": 0.2559,
       "step": 5260
     },
     {
-      "epoch": 0.9935083262771662,
-      "grad_norm": 0.07486743479967117,
+      "epoch": 0.45437920870893483,
+      "grad_norm": 0.41928738355636597,
       "learning_rate": 0.0002,
-      "loss": 0.0594,
+      "loss": 0.2874,
       "step": 5280
     },
     {
-      "epoch": 0.9972716153918525,
-      "grad_norm": 0.08373595029115677,
+      "epoch": 0.45610034207525657,
+      "grad_norm": 0.4400196671485901,
       "learning_rate": 0.0002,
-      "loss": 0.0579,
+      "loss": 0.2797,
       "step": 5300
     },
     {
-      "epoch": 0.9999059177721329,
-      "eval_loss": 0.20169740915298462,
-      "eval_runtime": 422.4603,
-      "eval_samples_per_second": 4.1,
-      "eval_steps_per_second": 0.514,
-      "step": 5314
+      "epoch": 0.45782147544157825,
+      "grad_norm": 0.29811346530914307,
+      "learning_rate": 0.0002,
+      "loss": 0.2619,
+      "step": 5320
+    },
+    {
+      "epoch": 0.4595426088079,
+      "grad_norm": 0.3616848289966583,
+      "learning_rate": 0.0002,
+      "loss": 0.2658,
+      "step": 5340
+    },
+    {
+      "epoch": 0.4612637421742217,
+      "grad_norm": 0.2878625690937042,
+      "learning_rate": 0.0002,
+      "loss": 0.2606,
+      "step": 5360
+    },
+    {
+      "epoch": 0.46298487554054346,
+      "grad_norm": 0.40930724143981934,
+      "learning_rate": 0.0002,
+      "loss": 0.2653,
+      "step": 5380
+    },
+    {
+      "epoch": 0.4647060089068652,
+      "grad_norm": 0.29266178607940674,
+      "learning_rate": 0.0002,
+      "loss": 0.2698,
+      "step": 5400
+    },
+    {
+      "epoch": 0.4664271422731869,
+      "grad_norm": 0.4216326177120209,
+      "learning_rate": 0.0002,
+      "loss": 0.2622,
+      "step": 5420
+    },
+    {
+      "epoch": 0.4681482756395086,
+      "grad_norm": 0.4683050215244293,
+      "learning_rate": 0.0002,
+      "loss": 0.2618,
+      "step": 5440
+    },
+    {
+      "epoch": 0.46986940900583035,
+      "grad_norm": 0.41073721647262573,
+      "learning_rate": 0.0002,
+      "loss": 0.2627,
+      "step": 5460
+    },
+    {
+      "epoch": 0.4715905423721521,
+      "grad_norm": 0.3913174867630005,
+      "learning_rate": 0.0002,
+      "loss": 0.257,
+      "step": 5480
+    },
+    {
+      "epoch": 0.47331167573847377,
+      "grad_norm": 0.39269042015075684,
+      "learning_rate": 0.0002,
+      "loss": 0.2659,
+      "step": 5500
+    },
+    {
+      "epoch": 0.4750328091047955,
+      "grad_norm": 0.4052613079547882,
+      "learning_rate": 0.0002,
+      "loss": 0.2523,
+      "step": 5520
+    },
+    {
+      "epoch": 0.47675394247111724,
+      "grad_norm": 0.36656561493873596,
+      "learning_rate": 0.0002,
+      "loss": 0.2627,
+      "step": 5540
+    },
+    {
+      "epoch": 0.478475075837439,
+      "grad_norm": 0.3679077625274658,
+      "learning_rate": 0.0002,
+      "loss": 0.264,
+      "step": 5560
+    },
+    {
+      "epoch": 0.48019620920376066,
+      "grad_norm": 0.49448278546333313,
+      "learning_rate": 0.0002,
+      "loss": 0.2688,
+      "step": 5580
+    },
+    {
+      "epoch": 0.4819173425700824,
+      "grad_norm": 0.39515119791030884,
+      "learning_rate": 0.0002,
+      "loss": 0.2498,
+      "step": 5600
+    },
+    {
+      "epoch": 0.48363847593640413,
+      "grad_norm": 0.3903290033340454,
+      "learning_rate": 0.0002,
+      "loss": 0.2744,
+      "step": 5620
+    },
+    {
+      "epoch": 0.48535960930272587,
+      "grad_norm": 0.41950464248657227,
+      "learning_rate": 0.0002,
+      "loss": 0.2645,
+      "step": 5640
+    },
+    {
+      "epoch": 0.48708074266904755,
+      "grad_norm": 0.3693583607673645,
+      "learning_rate": 0.0002,
+      "loss": 0.261,
+      "step": 5660
+    },
+    {
+      "epoch": 0.4888018760353693,
+      "grad_norm": 0.3864137530326843,
+      "learning_rate": 0.0002,
+      "loss": 0.2685,
+      "step": 5680
+    },
+    {
+      "epoch": 0.490523009401691,
+      "grad_norm": 0.5029241442680359,
+      "learning_rate": 0.0002,
+      "loss": 0.2531,
+      "step": 5700
+    },
+    {
+      "epoch": 0.49224414276801276,
+      "grad_norm": 0.37595513463020325,
+      "learning_rate": 0.0002,
+      "loss": 0.2775,
+      "step": 5720
+    },
+    {
+      "epoch": 0.49396527613433444,
+      "grad_norm": 0.4152761399745941,
+      "learning_rate": 0.0002,
+      "loss": 0.2448,
+      "step": 5740
+    },
+    {
+      "epoch": 0.4956864095006562,
+      "grad_norm": 0.36774635314941406,
+      "learning_rate": 0.0002,
+      "loss": 0.2731,
+      "step": 5760
+    },
+    {
+      "epoch": 0.4974075428669779,
+      "grad_norm": 0.347969114780426,
+      "learning_rate": 0.0002,
+      "loss": 0.2689,
+      "step": 5780
+    },
+    {
+      "epoch": 0.49912867623329965,
+      "grad_norm": 0.37085452675819397,
+      "learning_rate": 0.0002,
+      "loss": 0.2545,
+      "step": 5800
+    },
+    {
+      "epoch": 0.5008498095996213,
+      "grad_norm": 0.3683365285396576,
+      "learning_rate": 0.0002,
+      "loss": 0.2755,
+      "step": 5820
+    },
+    {
+      "epoch": 0.5025709429659431,
+      "grad_norm": 0.2994697690010071,
+      "learning_rate": 0.0002,
+      "loss": 0.2565,
+      "step": 5840
+    },
+    {
+      "epoch": 0.5042920763322648,
+      "grad_norm": 0.3561314046382904,
+      "learning_rate": 0.0002,
+      "loss": 0.2697,
+      "step": 5860
+    },
+    {
+      "epoch": 0.5060132096985865,
+      "grad_norm": 0.47199949622154236,
+      "learning_rate": 0.0002,
+      "loss": 0.2457,
+      "step": 5880
+    },
+    {
+      "epoch": 0.5077343430649083,
+      "grad_norm": 0.3466067612171173,
+      "learning_rate": 0.0002,
+      "loss": 0.2571,
+      "step": 5900
+    },
+    {
+      "epoch": 0.50945547643123,
+      "grad_norm": 0.26111066341400146,
+      "learning_rate": 0.0002,
+      "loss": 0.2509,
+      "step": 5920
+    },
+    {
+      "epoch": 0.5111766097975516,
+      "grad_norm": 0.32691895961761475,
+      "learning_rate": 0.0002,
+      "loss": 0.2575,
+      "step": 5940
+    },
+    {
+      "epoch": 0.5128977431638734,
+      "grad_norm": 0.31905749440193176,
+      "learning_rate": 0.0002,
+      "loss": 0.2549,
+      "step": 5960
+    },
+    {
+      "epoch": 0.5146188765301951,
+      "grad_norm": 0.3610023558139801,
+      "learning_rate": 0.0002,
+      "loss": 0.263,
+      "step": 5980
+    },
+    {
+      "epoch": 0.5163400098965168,
+      "grad_norm": 0.2800861895084381,
+      "learning_rate": 0.0002,
+      "loss": 0.2517,
+      "step": 6000
+    },
+    {
+      "epoch": 0.5180611432628386,
+      "grad_norm": 0.33658984303474426,
+      "learning_rate": 0.0002,
+      "loss": 0.2633,
+      "step": 6020
+    },
+    {
+      "epoch": 0.5197822766291603,
+      "grad_norm": 0.4304841160774231,
+      "learning_rate": 0.0002,
+      "loss": 0.2759,
+      "step": 6040
+    },
+    {
+      "epoch": 0.5215034099954821,
+      "grad_norm": 0.41126030683517456,
+      "learning_rate": 0.0002,
+      "loss": 0.2719,
+      "step": 6060
+    },
+    {
+      "epoch": 0.5232245433618038,
+      "grad_norm": 0.5020288825035095,
+      "learning_rate": 0.0002,
+      "loss": 0.2532,
+      "step": 6080
+    },
+    {
+      "epoch": 0.5249456767281254,
+      "grad_norm": 0.32016468048095703,
+      "learning_rate": 0.0002,
+      "loss": 0.2601,
+      "step": 6100
+    },
+    {
+      "epoch": 0.5266668100944472,
+      "grad_norm": 0.33081743121147156,
+      "learning_rate": 0.0002,
+      "loss": 0.2544,
+      "step": 6120
+    },
+    {
+      "epoch": 0.5283879434607689,
+      "grad_norm": 0.3016170561313629,
+      "learning_rate": 0.0002,
+      "loss": 0.2454,
+      "step": 6140
+    },
+    {
+      "epoch": 0.5301090768270906,
+      "grad_norm": 0.3886088728904724,
+      "learning_rate": 0.0002,
+      "loss": 0.2693,
+      "step": 6160
+    },
+    {
+      "epoch": 0.5318302101934124,
+      "grad_norm": 0.37182730436325073,
+      "learning_rate": 0.0002,
+      "loss": 0.2599,
+      "step": 6180
+    },
+    {
+      "epoch": 0.5335513435597341,
+      "grad_norm": 0.2848903238773346,
+      "learning_rate": 0.0002,
+      "loss": 0.26,
+      "step": 6200
+    },
+    {
+      "epoch": 0.5352724769260558,
+      "grad_norm": 0.45262956619262695,
+      "learning_rate": 0.0002,
+      "loss": 0.2587,
+      "step": 6220
+    },
+    {
+      "epoch": 0.5369936102923776,
+      "grad_norm": 0.3710488975048065,
+      "learning_rate": 0.0002,
+      "loss": 0.2608,
+      "step": 6240
+    },
+    {
+      "epoch": 0.5387147436586992,
+      "grad_norm": 0.39627817273139954,
+      "learning_rate": 0.0002,
+      "loss": 0.2505,
+      "step": 6260
+    },
+    {
+      "epoch": 0.5404358770250209,
+      "grad_norm": 0.35057738423347473,
+      "learning_rate": 0.0002,
+      "loss": 0.2619,
+      "step": 6280
+    },
+    {
+      "epoch": 0.5421570103913427,
+      "grad_norm": 0.3486057221889496,
+      "learning_rate": 0.0002,
+      "loss": 0.2472,
+      "step": 6300
+    },
+    {
+      "epoch": 0.5438781437576644,
+      "grad_norm": 0.330469012260437,
+      "learning_rate": 0.0002,
+      "loss": 0.2458,
+      "step": 6320
+    },
+    {
+      "epoch": 0.5455992771239861,
+      "grad_norm": 0.34958067536354065,
+      "learning_rate": 0.0002,
+      "loss": 0.2462,
+      "step": 6340
+    },
+    {
+      "epoch": 0.5473204104903079,
+      "grad_norm": 0.3149743676185608,
+      "learning_rate": 0.0002,
+      "loss": 0.2465,
+      "step": 6360
+    },
+    {
+      "epoch": 0.5490415438566296,
+      "grad_norm": 0.33875513076782227,
+      "learning_rate": 0.0002,
+      "loss": 0.2593,
+      "step": 6380
+    },
+    {
+      "epoch": 0.5507626772229514,
+      "grad_norm": 0.8838099241256714,
+      "learning_rate": 0.0002,
+      "loss": 0.2712,
+      "step": 6400
+    },
+    {
+      "epoch": 0.552483810589273,
+      "grad_norm": 0.3521359860897064,
+      "learning_rate": 0.0002,
+      "loss": 0.2441,
+      "step": 6420
+    },
+    {
+      "epoch": 0.5542049439555947,
+      "grad_norm": 0.4659444987773895,
+      "learning_rate": 0.0002,
+      "loss": 0.2526,
+      "step": 6440
+    },
+    {
+      "epoch": 0.5559260773219165,
+      "grad_norm": 0.4129142165184021,
+      "learning_rate": 0.0002,
+      "loss": 0.2586,
+      "step": 6460
+    },
+    {
+      "epoch": 0.5576472106882382,
+      "grad_norm": 0.36697232723236084,
+      "learning_rate": 0.0002,
+      "loss": 0.2367,
+      "step": 6480
+    },
+    {
+      "epoch": 0.5593683440545599,
+      "grad_norm": 0.37690311670303345,
+      "learning_rate": 0.0002,
+      "loss": 0.253,
+      "step": 6500
+    },
+    {
+      "epoch": 0.5610894774208817,
+      "grad_norm": 0.2961209714412689,
+      "learning_rate": 0.0002,
+      "loss": 0.2441,
+      "step": 6520
+    },
+    {
+      "epoch": 0.5628106107872034,
+      "grad_norm": 0.32476598024368286,
+      "learning_rate": 0.0002,
+      "loss": 0.2552,
+      "step": 6540
+    },
+    {
+      "epoch": 0.5645317441535251,
+      "grad_norm": 0.36377888917922974,
+      "learning_rate": 0.0002,
+      "loss": 0.259,
+      "step": 6560
+    },
+    {
+      "epoch": 0.5662528775198468,
+      "grad_norm": 0.3714245557785034,
+      "learning_rate": 0.0002,
+      "loss": 0.2499,
+      "step": 6580
+    },
+    {
+      "epoch": 0.5679740108861685,
+      "grad_norm": 0.3322594463825226,
+      "learning_rate": 0.0002,
+      "loss": 0.2437,
+      "step": 6600
+    },
+    {
+      "epoch": 0.5696951442524902,
+      "grad_norm": 0.38657745718955994,
+      "learning_rate": 0.0002,
+      "loss": 0.2452,
+      "step": 6620
+    },
+    {
+      "epoch": 0.571416277618812,
+      "grad_norm": 0.34600910544395447,
+      "learning_rate": 0.0002,
+      "loss": 0.2454,
+      "step": 6640
+    },
+    {
+      "epoch": 0.5731374109851337,
+      "grad_norm": 0.3906444311141968,
+      "learning_rate": 0.0002,
+      "loss": 0.254,
+      "step": 6660
+    },
+    {
+      "epoch": 0.5748585443514554,
+      "grad_norm": 0.37352487444877625,
+      "learning_rate": 0.0002,
+      "loss": 0.2489,
+      "step": 6680
+    },
+    {
+      "epoch": 0.5765796777177772,
+      "grad_norm": 0.3105640411376953,
+      "learning_rate": 0.0002,
+      "loss": 0.2523,
+      "step": 6700
+    },
+    {
+      "epoch": 0.5783008110840989,
+      "grad_norm": 0.3537706732749939,
+      "learning_rate": 0.0002,
+      "loss": 0.2505,
+      "step": 6720
+    },
+    {
+      "epoch": 0.5800219444504207,
+      "grad_norm": 0.45570480823516846,
+      "learning_rate": 0.0002,
+      "loss": 0.2386,
+      "step": 6740
+    },
+    {
+      "epoch": 0.5817430778167423,
+      "grad_norm": 0.36528217792510986,
+      "learning_rate": 0.0002,
+      "loss": 0.2382,
+      "step": 6760
+    },
+    {
+      "epoch": 0.583464211183064,
+      "grad_norm": 0.3420511782169342,
+      "learning_rate": 0.0002,
+      "loss": 0.2456,
+      "step": 6780
+    },
+    {
+      "epoch": 0.5851853445493858,
+      "grad_norm": 0.35168027877807617,
+      "learning_rate": 0.0002,
+      "loss": 0.2578,
+      "step": 6800
+    },
+    {
+      "epoch": 0.5869064779157075,
+      "grad_norm": 0.3705766201019287,
+      "learning_rate": 0.0002,
+      "loss": 0.2481,
+      "step": 6820
+    },
+    {
+      "epoch": 0.5886276112820292,
+      "grad_norm": 0.4245602786540985,
+      "learning_rate": 0.0002,
+      "loss": 0.2603,
+      "step": 6840
+    },
+    {
+      "epoch": 0.590348744648351,
+      "grad_norm": 0.42988812923431396,
+      "learning_rate": 0.0002,
+      "loss": 0.2308,
+      "step": 6860
+    },
+    {
+      "epoch": 0.5920698780146727,
+      "grad_norm": 0.3902256190776825,
+      "learning_rate": 0.0002,
+      "loss": 0.2382,
+      "step": 6880
+    },
+    {
+      "epoch": 0.5937910113809944,
+      "grad_norm": 0.3394165337085724,
+      "learning_rate": 0.0002,
+      "loss": 0.2264,
+      "step": 6900
+    },
+    {
+      "epoch": 0.5955121447473161,
+      "grad_norm": 0.40591520071029663,
+      "learning_rate": 0.0002,
+      "loss": 0.2487,
+      "step": 6920
+    },
+    {
+      "epoch": 0.5972332781136378,
+      "grad_norm": 0.37964189052581787,
+      "learning_rate": 0.0002,
+      "loss": 0.2466,
+      "step": 6940
+    },
+    {
+      "epoch": 0.5989544114799595,
+      "grad_norm": 0.4443446099758148,
+      "learning_rate": 0.0002,
+      "loss": 0.2384,
+      "step": 6960
+    },
+    {
+      "epoch": 0.6006755448462813,
+      "grad_norm": 0.40669143199920654,
+      "learning_rate": 0.0002,
+      "loss": 0.244,
+      "step": 6980
+    },
+    {
+      "epoch": 0.602396678212603,
+      "grad_norm": 0.40179166197776794,
+      "learning_rate": 0.0002,
+      "loss": 0.2368,
+      "step": 7000
+    },
+    {
+      "epoch": 0.6041178115789247,
+      "grad_norm": 0.4410141408443451,
+      "learning_rate": 0.0002,
+      "loss": 0.2388,
+      "step": 7020
+    },
+    {
+      "epoch": 0.6058389449452465,
+      "grad_norm": 0.43160775303840637,
+      "learning_rate": 0.0002,
+      "loss": 0.2377,
+      "step": 7040
+    },
+    {
+      "epoch": 0.6075600783115682,
+      "grad_norm": 0.40169423818588257,
+      "learning_rate": 0.0002,
+      "loss": 0.2566,
+      "step": 7060
+    },
+    {
+      "epoch": 0.6092812116778898,
+      "grad_norm": 0.43439412117004395,
+      "learning_rate": 0.0002,
+      "loss": 0.2559,
+      "step": 7080
+    },
+    {
+      "epoch": 0.6110023450442116,
+      "grad_norm": 0.40532273054122925,
+      "learning_rate": 0.0002,
+      "loss": 0.242,
+      "step": 7100
+    },
+    {
+      "epoch": 0.6127234784105333,
+      "grad_norm": 0.3553959131240845,
+      "learning_rate": 0.0002,
+      "loss": 0.2497,
+      "step": 7120
+    },
+    {
+      "epoch": 0.614444611776855,
+      "grad_norm": 0.33834314346313477,
+      "learning_rate": 0.0002,
+      "loss": 0.2441,
+      "step": 7140
+    },
+    {
+      "epoch": 0.6161657451431768,
+      "grad_norm": 0.432019978761673,
+      "learning_rate": 0.0002,
+      "loss": 0.2701,
+      "step": 7160
+    },
+    {
+      "epoch": 0.6178868785094985,
+      "grad_norm": 0.35808584094047546,
+      "learning_rate": 0.0002,
+      "loss": 0.2436,
+      "step": 7180
+    },
+    {
+      "epoch": 0.6196080118758203,
+      "grad_norm": 0.5244463682174683,
+      "learning_rate": 0.0002,
+      "loss": 0.2356,
+      "step": 7200
+    },
+    {
+      "epoch": 0.621329145242142,
+      "grad_norm": 0.3416566848754883,
+      "learning_rate": 0.0002,
+      "loss": 0.2498,
+      "step": 7220
+    },
+    {
+      "epoch": 0.6230502786084636,
+      "grad_norm": 0.377354234457016,
+      "learning_rate": 0.0002,
+      "loss": 0.2361,
+      "step": 7240
+    },
+    {
+      "epoch": 0.6247714119747854,
+      "grad_norm": 0.3780601918697357,
+      "learning_rate": 0.0002,
+      "loss": 0.2414,
+      "step": 7260
+    },
+    {
+      "epoch": 0.6264925453411071,
+      "grad_norm": 0.4537530541419983,
+      "learning_rate": 0.0002,
+      "loss": 0.2553,
+      "step": 7280
+    },
+    {
+      "epoch": 0.6282136787074288,
+      "grad_norm": 0.34764549136161804,
+      "learning_rate": 0.0002,
+      "loss": 0.2294,
+      "step": 7300
+    },
+    {
+      "epoch": 0.6299348120737506,
+      "grad_norm": 0.4038061499595642,
+      "learning_rate": 0.0002,
+      "loss": 0.2267,
+      "step": 7320
+    },
+    {
+      "epoch": 0.6316559454400723,
+      "grad_norm": 0.3201640546321869,
+      "learning_rate": 0.0002,
+      "loss": 0.2407,
+      "step": 7340
+    },
+    {
+      "epoch": 0.633377078806394,
+      "grad_norm": 0.437897652387619,
+      "learning_rate": 0.0002,
+      "loss": 0.238,
+      "step": 7360
+    },
+    {
+      "epoch": 0.6350982121727158,
+      "grad_norm": 0.32471516728401184,
+      "learning_rate": 0.0002,
+      "loss": 0.2403,
+      "step": 7380
+    },
+    {
+      "epoch": 0.6368193455390374,
+      "grad_norm": 0.2739992141723633,
+      "learning_rate": 0.0002,
+      "loss": 0.2507,
+      "step": 7400
+    },
+    {
+      "epoch": 0.6385404789053591,
+      "grad_norm": 0.33521589636802673,
+      "learning_rate": 0.0002,
+      "loss": 0.2462,
+      "step": 7420
+    },
+    {
+      "epoch": 0.6402616122716809,
+      "grad_norm": 0.3575187027454376,
+      "learning_rate": 0.0002,
+      "loss": 0.2445,
+      "step": 7440
+    },
+    {
+      "epoch": 0.6419827456380026,
+      "grad_norm": 0.3016495704650879,
+      "learning_rate": 0.0002,
+      "loss": 0.2324,
+      "step": 7460
+    },
+    {
+      "epoch": 0.6437038790043244,
+      "grad_norm": 0.4337838888168335,
+      "learning_rate": 0.0002,
+      "loss": 0.2508,
+      "step": 7480
+    },
+    {
+      "epoch": 0.6454250123706461,
+      "grad_norm": 0.35893139243125916,
+      "learning_rate": 0.0002,
+      "loss": 0.2318,
+      "step": 7500
+    },
+    {
+      "epoch": 0.6471461457369678,
+      "grad_norm": 0.3144209682941437,
+      "learning_rate": 0.0002,
+      "loss": 0.2384,
+      "step": 7520
+    },
+    {
+      "epoch": 0.6488672791032896,
+      "grad_norm": 0.30142101645469666,
+      "learning_rate": 0.0002,
+      "loss": 0.2439,
+      "step": 7540
+    },
+    {
+      "epoch": 0.6505884124696112,
+      "grad_norm": 0.3650546669960022,
+      "learning_rate": 0.0002,
+      "loss": 0.2495,
+      "step": 7560
+    },
+    {
+      "epoch": 0.6523095458359329,
+      "grad_norm": 0.41195932030677795,
+      "learning_rate": 0.0002,
+      "loss": 0.2497,
+      "step": 7580
+    },
+    {
+      "epoch": 0.6540306792022547,
+      "grad_norm": 0.34587377309799194,
+      "learning_rate": 0.0002,
+      "loss": 0.2372,
+      "step": 7600
+    },
+    {
+      "epoch": 0.6557518125685764,
+      "grad_norm": 0.35166653990745544,
+      "learning_rate": 0.0002,
+      "loss": 0.2332,
+      "step": 7620
+    },
+    {
+      "epoch": 0.6574729459348981,
+      "grad_norm": 0.46442800760269165,
+      "learning_rate": 0.0002,
+      "loss": 0.232,
+      "step": 7640
+    },
+    {
+      "epoch": 0.6591940793012199,
+      "grad_norm": 0.32510513067245483,
+      "learning_rate": 0.0002,
+      "loss": 0.2286,
+      "step": 7660
+    },
+    {
+      "epoch": 0.6609152126675416,
+      "grad_norm": 0.3486369848251343,
+      "learning_rate": 0.0002,
+      "loss": 0.2311,
+      "step": 7680
+    },
+    {
+      "epoch": 0.6626363460338633,
+      "grad_norm": 0.4640563130378723,
+      "learning_rate": 0.0002,
+      "loss": 0.2386,
+      "step": 7700
+    },
+    {
+      "epoch": 0.664357479400185,
+      "grad_norm": 0.4044354557991028,
+      "learning_rate": 0.0002,
+      "loss": 0.2408,
+      "step": 7720
+    },
+    {
+      "epoch": 0.6660786127665067,
+      "grad_norm": 0.38926753401756287,
+      "learning_rate": 0.0002,
+      "loss": 0.2537,
+      "step": 7740
+    },
+    {
+      "epoch": 0.6677997461328284,
+      "grad_norm": 0.428072988986969,
+      "learning_rate": 0.0002,
+      "loss": 0.231,
+      "step": 7760
+    },
+    {
+      "epoch": 0.6695208794991502,
+      "grad_norm": 0.2939753830432892,
+      "learning_rate": 0.0002,
+      "loss": 0.2424,
+      "step": 7780
+    },
+    {
+      "epoch": 0.6712420128654719,
+      "grad_norm": 0.41152870655059814,
+      "learning_rate": 0.0002,
+      "loss": 0.2422,
+      "step": 7800
+    },
+    {
+      "epoch": 0.6729631462317937,
+      "grad_norm": 0.35313352942466736,
+      "learning_rate": 0.0002,
+      "loss": 0.2401,
+      "step": 7820
+    },
+    {
+      "epoch": 0.6746842795981154,
+      "grad_norm": 0.4280535876750946,
+      "learning_rate": 0.0002,
+      "loss": 0.2309,
+      "step": 7840
+    },
+    {
+      "epoch": 0.6764054129644371,
+      "grad_norm": 0.36120399832725525,
+      "learning_rate": 0.0002,
+      "loss": 0.2339,
+      "step": 7860
+    },
+    {
+      "epoch": 0.6781265463307588,
+      "grad_norm": 0.39646589756011963,
+      "learning_rate": 0.0002,
+      "loss": 0.2464,
+      "step": 7880
+    },
+    {
+      "epoch": 0.6798476796970805,
+      "grad_norm": 0.4060419797897339,
+      "learning_rate": 0.0002,
+      "loss": 0.2331,
+      "step": 7900
+    },
+    {
+      "epoch": 0.6815688130634022,
+      "grad_norm": 0.4130091369152069,
+      "learning_rate": 0.0002,
+      "loss": 0.2331,
+      "step": 7920
+    },
+    {
+      "epoch": 0.683289946429724,
+      "grad_norm": 0.31750190258026123,
+      "learning_rate": 0.0002,
+      "loss": 0.2371,
+      "step": 7940
+    },
+    {
+      "epoch": 0.6850110797960457,
+      "grad_norm": 0.4835965037345886,
+      "learning_rate": 0.0002,
+      "loss": 0.2292,
+      "step": 7960
+    },
+    {
+      "epoch": 0.6867322131623674,
+      "grad_norm": 0.3932303488254547,
+      "learning_rate": 0.0002,
+      "loss": 0.2243,
+      "step": 7980
+    },
+    {
+      "epoch": 0.6884533465286892,
+      "grad_norm": 0.3820064067840576,
+      "learning_rate": 0.0002,
+      "loss": 0.2288,
+      "step": 8000
+    },
+    {
+      "epoch": 0.6901744798950109,
+      "grad_norm": 0.40602800250053406,
+      "learning_rate": 0.0002,
+      "loss": 0.2259,
+      "step": 8020
+    },
+    {
+      "epoch": 0.6918956132613326,
+      "grad_norm": 0.3742893636226654,
+      "learning_rate": 0.0002,
+      "loss": 0.2395,
+      "step": 8040
+    },
+    {
+      "epoch": 0.6936167466276543,
+      "grad_norm": 0.39692652225494385,
+      "learning_rate": 0.0002,
+      "loss": 0.2296,
+      "step": 8060
+    },
+    {
+      "epoch": 0.695337879993976,
+      "grad_norm": 0.31871652603149414,
+      "learning_rate": 0.0002,
+      "loss": 0.2444,
+      "step": 8080
+    },
+    {
+      "epoch": 0.6970590133602977,
+      "grad_norm": 0.3136846125125885,
+      "learning_rate": 0.0002,
+      "loss": 0.2344,
+      "step": 8100
+    },
+    {
+      "epoch": 0.6987801467266195,
+      "grad_norm": 0.3090713918209076,
+      "learning_rate": 0.0002,
+      "loss": 0.2353,
+      "step": 8120
+    },
+    {
+      "epoch": 0.7005012800929412,
+      "grad_norm": 0.33116286993026733,
+      "learning_rate": 0.0002,
+      "loss": 0.2205,
+      "step": 8140
+    },
+    {
+      "epoch": 0.702222413459263,
+      "grad_norm": 0.2979792058467865,
+      "learning_rate": 0.0002,
+      "loss": 0.2371,
+      "step": 8160
+    },
+    {
+      "epoch": 0.7039435468255847,
+      "grad_norm": 0.3515206277370453,
+      "learning_rate": 0.0002,
+      "loss": 0.2293,
+      "step": 8180
+    },
+    {
+      "epoch": 0.7056646801919064,
+      "grad_norm": 0.2888669967651367,
+      "learning_rate": 0.0002,
+      "loss": 0.234,
+      "step": 8200
+    },
+    {
+      "epoch": 0.707385813558228,
+      "grad_norm": 0.3582359552383423,
+      "learning_rate": 0.0002,
+      "loss": 0.2291,
+      "step": 8220
+    },
+    {
+      "epoch": 0.7091069469245498,
+      "grad_norm": 0.4168799817562103,
+      "learning_rate": 0.0002,
+      "loss": 0.2241,
+      "step": 8240
+    },
+    {
+      "epoch": 0.7108280802908715,
+      "grad_norm": 0.3614865243434906,
+      "learning_rate": 0.0002,
+      "loss": 0.2323,
+      "step": 8260
+    },
+    {
+      "epoch": 0.7125492136571933,
+      "grad_norm": 0.35036012530326843,
+      "learning_rate": 0.0002,
+      "loss": 0.2285,
+      "step": 8280
+    },
+    {
+      "epoch": 0.714270347023515,
+      "grad_norm": 0.460549533367157,
+      "learning_rate": 0.0002,
+      "loss": 0.2303,
+      "step": 8300
+    },
+    {
+      "epoch": 0.7159914803898367,
+      "grad_norm": 0.44449788331985474,
+      "learning_rate": 0.0002,
+      "loss": 0.2279,
+      "step": 8320
+    },
+    {
+      "epoch": 0.7177126137561585,
+      "grad_norm": 0.29927289485931396,
+      "learning_rate": 0.0002,
+      "loss": 0.2235,
+      "step": 8340
+    },
+    {
+      "epoch": 0.7194337471224802,
+      "grad_norm": 0.3563356101512909,
+      "learning_rate": 0.0002,
+      "loss": 0.2307,
+      "step": 8360
+    },
+    {
+      "epoch": 0.7211548804888018,
+      "grad_norm": 0.42358216643333435,
+      "learning_rate": 0.0002,
+      "loss": 0.227,
+      "step": 8380
+    },
+    {
+      "epoch": 0.7228760138551236,
+      "grad_norm": 0.42043471336364746,
+      "learning_rate": 0.0002,
+      "loss": 0.2199,
+      "step": 8400
+    },
+    {
+      "epoch": 0.7245971472214453,
+      "grad_norm": 0.3534243702888489,
+      "learning_rate": 0.0002,
+      "loss": 0.2384,
+      "step": 8420
+    },
+    {
+      "epoch": 0.726318280587767,
+      "grad_norm": 0.35375455021858215,
+      "learning_rate": 0.0002,
+      "loss": 0.2337,
+      "step": 8440
+    },
+    {
+      "epoch": 0.7280394139540888,
+      "grad_norm": 0.31629830598831177,
+      "learning_rate": 0.0002,
+      "loss": 0.2326,
+      "step": 8460
+    },
+    {
+      "epoch": 0.7297605473204105,
+      "grad_norm": 0.3449802100658417,
+      "learning_rate": 0.0002,
+      "loss": 0.221,
+      "step": 8480
+    },
+    {
+      "epoch": 0.7314816806867322,
+      "grad_norm": 0.44133901596069336,
+      "learning_rate": 0.0002,
+      "loss": 0.2304,
+      "step": 8500
+    },
+    {
+      "epoch": 0.733202814053054,
+      "grad_norm": 0.3913390040397644,
+      "learning_rate": 0.0002,
+      "loss": 0.2277,
+      "step": 8520
+    },
+    {
+      "epoch": 0.7349239474193756,
+      "grad_norm": 0.3438524901866913,
+      "learning_rate": 0.0002,
+      "loss": 0.2484,
+      "step": 8540
+    },
+    {
+      "epoch": 0.7366450807856973,
+      "grad_norm": 0.34400486946105957,
+      "learning_rate": 0.0002,
+      "loss": 0.2257,
+      "step": 8560
+    },
+    {
+      "epoch": 0.7383662141520191,
+      "grad_norm": 0.40113121271133423,
+      "learning_rate": 0.0002,
+      "loss": 0.2246,
+      "step": 8580
+    },
+    {
+      "epoch": 0.7400873475183408,
+      "grad_norm": 0.3908839821815491,
+      "learning_rate": 0.0002,
+      "loss": 0.2178,
+      "step": 8600
+    },
+    {
+      "epoch": 0.7418084808846626,
+      "grad_norm": 0.3656678795814514,
+      "learning_rate": 0.0002,
+      "loss": 0.2345,
+      "step": 8620
+    },
+    {
+      "epoch": 0.7435296142509843,
+      "grad_norm": 0.3811242878437042,
+      "learning_rate": 0.0002,
+      "loss": 0.2387,
+      "step": 8640
+    },
+    {
+      "epoch": 0.745250747617306,
+      "grad_norm": 0.3093118965625763,
+      "learning_rate": 0.0002,
+      "loss": 0.2179,
+      "step": 8660
+    },
+    {
+      "epoch": 0.7469718809836278,
+      "grad_norm": 0.42958253622055054,
+      "learning_rate": 0.0002,
+      "loss": 0.2212,
+      "step": 8680
+    },
+    {
+      "epoch": 0.7486930143499494,
+      "grad_norm": 0.3202790319919586,
+      "learning_rate": 0.0002,
+      "loss": 0.2394,
+      "step": 8700
+    },
+    {
+      "epoch": 0.7504141477162711,
+      "grad_norm": 0.42061832547187805,
+      "learning_rate": 0.0002,
+      "loss": 0.2282,
+      "step": 8720
+    },
+    {
+      "epoch": 0.7521352810825929,
+      "grad_norm": 0.2962038516998291,
+      "learning_rate": 0.0002,
+      "loss": 0.2322,
+      "step": 8740
+    },
+    {
+      "epoch": 0.7538564144489146,
+      "grad_norm": 0.3707858920097351,
+      "learning_rate": 0.0002,
+      "loss": 0.2251,
+      "step": 8760
+    },
+    {
+      "epoch": 0.7555775478152363,
+      "grad_norm": 0.424730509519577,
+      "learning_rate": 0.0002,
+      "loss": 0.2345,
+      "step": 8780
+    },
+    {
+      "epoch": 0.7572986811815581,
+      "grad_norm": 0.34581223130226135,
+      "learning_rate": 0.0002,
+      "loss": 0.2196,
+      "step": 8800
+    },
+    {
+      "epoch": 0.7590198145478798,
+      "grad_norm": 0.436617374420166,
+      "learning_rate": 0.0002,
+      "loss": 0.2332,
+      "step": 8820
+    },
+    {
+      "epoch": 0.7607409479142015,
+      "grad_norm": 0.3432689905166626,
+      "learning_rate": 0.0002,
+      "loss": 0.231,
+      "step": 8840
+    },
+    {
+      "epoch": 0.7624620812805232,
+      "grad_norm": 0.36506009101867676,
+      "learning_rate": 0.0002,
+      "loss": 0.2299,
+      "step": 8860
+    },
+    {
+      "epoch": 0.7641832146468449,
+      "grad_norm": 0.29211345314979553,
+      "learning_rate": 0.0002,
+      "loss": 0.2169,
+      "step": 8880
+    },
+    {
+      "epoch": 0.7659043480131666,
+      "grad_norm": 0.3884661793708801,
+      "learning_rate": 0.0002,
+      "loss": 0.2205,
+      "step": 8900
+    },
+    {
+      "epoch": 0.7676254813794884,
+      "grad_norm": 0.3466391861438751,
+      "learning_rate": 0.0002,
+      "loss": 0.2249,
+      "step": 8920
+    },
+    {
+      "epoch": 0.7693466147458101,
+      "grad_norm": 0.3165854811668396,
+      "learning_rate": 0.0002,
+      "loss": 0.2436,
+      "step": 8940
+    },
+    {
+      "epoch": 0.7710677481121319,
+      "grad_norm": 0.42315399646759033,
+      "learning_rate": 0.0002,
+      "loss": 0.2207,
+      "step": 8960
+    },
+    {
+      "epoch": 0.7727888814784536,
+      "grad_norm": 0.3629299998283386,
+      "learning_rate": 0.0002,
+      "loss": 0.2356,
+      "step": 8980
+    },
+    {
+      "epoch": 0.7745100148447753,
+      "grad_norm": 0.4586748480796814,
+      "learning_rate": 0.0002,
+      "loss": 0.2152,
+      "step": 9000
+    },
+    {
+      "epoch": 0.776231148211097,
+      "grad_norm": 0.27728915214538574,
+      "learning_rate": 0.0002,
+      "loss": 0.2284,
+      "step": 9020
+    },
+    {
+      "epoch": 0.7779522815774187,
+      "grad_norm": 0.3490050435066223,
+      "learning_rate": 0.0002,
+      "loss": 0.2166,
+      "step": 9040
+    },
+    {
+      "epoch": 0.7796734149437404,
+      "grad_norm": 0.2839438021183014,
+      "learning_rate": 0.0002,
+      "loss": 0.2283,
+      "step": 9060
+    },
+    {
+      "epoch": 0.7813945483100622,
+      "grad_norm": 0.35332831740379333,
+      "learning_rate": 0.0002,
+      "loss": 0.2336,
+      "step": 9080
+    },
+    {
+      "epoch": 0.7831156816763839,
+      "grad_norm": 0.4003342390060425,
+      "learning_rate": 0.0002,
+      "loss": 0.2279,
+      "step": 9100
+    },
+    {
+      "epoch": 0.7848368150427056,
+      "grad_norm": 0.32392072677612305,
+      "learning_rate": 0.0002,
+      "loss": 0.235,
+      "step": 9120
+    },
+    {
+      "epoch": 0.7865579484090274,
+      "grad_norm": 0.3040711581707001,
+      "learning_rate": 0.0002,
+      "loss": 0.2234,
+      "step": 9140
+    },
+    {
+      "epoch": 0.7882790817753491,
+      "grad_norm": 0.35234126448631287,
+      "learning_rate": 0.0002,
+      "loss": 0.2215,
+      "step": 9160
+    },
+    {
+      "epoch": 0.7900002151416707,
+      "grad_norm": 0.3502811789512634,
+      "learning_rate": 0.0002,
+      "loss": 0.225,
+      "step": 9180
+    },
+    {
+      "epoch": 0.7917213485079925,
+      "grad_norm": 0.3539245128631592,
+      "learning_rate": 0.0002,
+      "loss": 0.2289,
+      "step": 9200
+    },
+    {
+      "epoch": 0.7934424818743142,
+      "grad_norm": 0.3154076635837555,
+      "learning_rate": 0.0002,
+      "loss": 0.2282,
+      "step": 9220
+    },
+    {
+      "epoch": 0.795163615240636,
+      "grad_norm": 0.3497600853443146,
+      "learning_rate": 0.0002,
+      "loss": 0.2159,
+      "step": 9240
+    },
+    {
+      "epoch": 0.7968847486069577,
+      "grad_norm": 0.3394204080104828,
+      "learning_rate": 0.0002,
+      "loss": 0.2175,
+      "step": 9260
+    },
+    {
+      "epoch": 0.7986058819732794,
+      "grad_norm": 0.42728209495544434,
+      "learning_rate": 0.0002,
+      "loss": 0.2316,
+      "step": 9280
+    },
+    {
+      "epoch": 0.8003270153396012,
+      "grad_norm": 0.30218786001205444,
+      "learning_rate": 0.0002,
+      "loss": 0.2268,
+      "step": 9300
+    },
+    {
+      "epoch": 0.8020481487059229,
+      "grad_norm": 0.3418307602405548,
+      "learning_rate": 0.0002,
+      "loss": 0.2315,
+      "step": 9320
+    },
+    {
+      "epoch": 0.8037692820722446,
+      "grad_norm": 0.38792407512664795,
+      "learning_rate": 0.0002,
+      "loss": 0.2302,
+      "step": 9340
+    },
+    {
+      "epoch": 0.8054904154385663,
+      "grad_norm": 0.2837861180305481,
+      "learning_rate": 0.0002,
+      "loss": 0.2193,
+      "step": 9360
+    },
+    {
+      "epoch": 0.807211548804888,
+      "grad_norm": 0.4480486214160919,
+      "learning_rate": 0.0002,
+      "loss": 0.2264,
+      "step": 9380
+    },
+    {
+      "epoch": 0.8089326821712097,
+      "grad_norm": 0.25541505217552185,
+      "learning_rate": 0.0002,
+      "loss": 0.2302,
+      "step": 9400
+    },
+    {
+      "epoch": 0.8106538155375315,
+      "grad_norm": 0.28958380222320557,
+      "learning_rate": 0.0002,
+      "loss": 0.2269,
+      "step": 9420
+    },
+    {
+      "epoch": 0.8123749489038532,
+      "grad_norm": 0.2757406532764435,
+      "learning_rate": 0.0002,
+      "loss": 0.2155,
+      "step": 9440
+    },
+    {
+      "epoch": 0.8140960822701749,
+      "grad_norm": 0.3484242558479309,
+      "learning_rate": 0.0002,
+      "loss": 0.2346,
+      "step": 9460
+    },
+    {
+      "epoch": 0.8158172156364967,
+      "grad_norm": 0.23964551091194153,
+      "learning_rate": 0.0002,
+      "loss": 0.2208,
+      "step": 9480
+    },
+    {
+      "epoch": 0.8175383490028184,
+      "grad_norm": 0.3568115234375,
+      "learning_rate": 0.0002,
+      "loss": 0.2187,
+      "step": 9500
+    },
+    {
+      "epoch": 0.81925948236914,
+      "grad_norm": 0.4668004512786865,
+      "learning_rate": 0.0002,
+      "loss": 0.2306,
+      "step": 9520
+    },
+    {
+      "epoch": 0.8209806157354618,
+      "grad_norm": 0.33082979917526245,
+      "learning_rate": 0.0002,
+      "loss": 0.2135,
+      "step": 9540
+    },
+    {
+      "epoch": 0.8227017491017835,
+      "grad_norm": 0.3717847764492035,
+      "learning_rate": 0.0002,
+      "loss": 0.2203,
+      "step": 9560
+    },
+    {
+      "epoch": 0.8244228824681052,
+      "grad_norm": 0.37038078904151917,
+      "learning_rate": 0.0002,
+      "loss": 0.2271,
+      "step": 9580
+    },
+    {
+      "epoch": 0.826144015834427,
+      "grad_norm": 0.33178070187568665,
+      "learning_rate": 0.0002,
+      "loss": 0.2222,
+      "step": 9600
+    },
+    {
+      "epoch": 0.8278651492007487,
+      "grad_norm": 0.27431583404541016,
+      "learning_rate": 0.0002,
+      "loss": 0.2304,
+      "step": 9620
+    },
+    {
+      "epoch": 0.8295862825670705,
+      "grad_norm": 0.33868762850761414,
+      "learning_rate": 0.0002,
+      "loss": 0.2197,
+      "step": 9640
+    },
+    {
+      "epoch": 0.8313074159333922,
+      "grad_norm": 0.34042665362358093,
+      "learning_rate": 0.0002,
+      "loss": 0.2194,
+      "step": 9660
+    },
+    {
+      "epoch": 0.8330285492997138,
+      "grad_norm": 0.431411474943161,
+      "learning_rate": 0.0002,
+      "loss": 0.2228,
+      "step": 9680
+    },
+    {
+      "epoch": 0.8347496826660356,
+      "grad_norm": 0.2958236634731293,
+      "learning_rate": 0.0002,
+      "loss": 0.2268,
+      "step": 9700
+    },
+    {
+      "epoch": 0.8364708160323573,
+      "grad_norm": 0.31208258867263794,
+      "learning_rate": 0.0002,
+      "loss": 0.2371,
+      "step": 9720
+    },
+    {
+      "epoch": 0.838191949398679,
+      "grad_norm": 0.3337579667568207,
+      "learning_rate": 0.0002,
+      "loss": 0.2312,
+      "step": 9740
+    },
+    {
+      "epoch": 0.8399130827650008,
+      "grad_norm": 0.5392299294471741,
+      "learning_rate": 0.0002,
+      "loss": 0.2324,
+      "step": 9760
+    },
+    {
+      "epoch": 0.8416342161313225,
+      "grad_norm": 0.3481907248497009,
+      "learning_rate": 0.0002,
+      "loss": 0.2241,
+      "step": 9780
+    },
+    {
+      "epoch": 0.8433553494976442,
+      "grad_norm": 0.29136040806770325,
+      "learning_rate": 0.0002,
+      "loss": 0.2147,
+      "step": 9800
+    },
+    {
+      "epoch": 0.845076482863966,
+      "grad_norm": 0.2877753973007202,
+      "learning_rate": 0.0002,
+      "loss": 0.1988,
+      "step": 9820
+    },
+    {
+      "epoch": 0.8467976162302876,
+      "grad_norm": 0.4024602174758911,
+      "learning_rate": 0.0002,
+      "loss": 0.2161,
+      "step": 9840
+    },
+    {
+      "epoch": 0.8485187495966093,
+      "grad_norm": 0.28684961795806885,
+      "learning_rate": 0.0002,
+      "loss": 0.2199,
+      "step": 9860
+    },
+    {
+      "epoch": 0.8502398829629311,
+      "grad_norm": 0.31743690371513367,
+      "learning_rate": 0.0002,
+      "loss": 0.2196,
+      "step": 9880
+    },
+    {
+      "epoch": 0.8519610163292528,
+      "grad_norm": 0.3548375368118286,
+      "learning_rate": 0.0002,
+      "loss": 0.2179,
+      "step": 9900
+    },
+    {
+      "epoch": 0.8536821496955745,
+      "grad_norm": 0.31174436211586,
+      "learning_rate": 0.0002,
+      "loss": 0.2236,
+      "step": 9920
+    },
+    {
+      "epoch": 0.8554032830618963,
+      "grad_norm": 0.2791374623775482,
+      "learning_rate": 0.0002,
+      "loss": 0.2257,
+      "step": 9940
+    },
+    {
+      "epoch": 0.857124416428218,
+      "grad_norm": 0.3355172574520111,
+      "learning_rate": 0.0002,
+      "loss": 0.2244,
+      "step": 9960
+    },
+    {
+      "epoch": 0.8588455497945398,
+      "grad_norm": 0.3883482813835144,
+      "learning_rate": 0.0002,
+      "loss": 0.2178,
+      "step": 9980
+    },
+    {
+      "epoch": 0.8605666831608614,
+      "grad_norm": 0.3601789176464081,
+      "learning_rate": 0.0002,
+      "loss": 0.227,
+      "step": 10000
+    },
+    {
+      "epoch": 0.8622878165271831,
+      "grad_norm": 0.24994397163391113,
+      "learning_rate": 0.0002,
+      "loss": 0.2165,
+      "step": 10020
+    },
+    {
+      "epoch": 0.8640089498935049,
+      "grad_norm": 0.40082284808158875,
+      "learning_rate": 0.0002,
+      "loss": 0.228,
+      "step": 10040
+    },
+    {
+      "epoch": 0.8657300832598266,
+      "grad_norm": 0.47781533002853394,
+      "learning_rate": 0.0002,
+      "loss": 0.2283,
+      "step": 10060
+    },
+    {
+      "epoch": 0.8674512166261483,
+      "grad_norm": 0.3652552366256714,
+      "learning_rate": 0.0002,
+      "loss": 0.2207,
+      "step": 10080
+    },
+    {
+      "epoch": 0.8691723499924701,
+      "grad_norm": 0.38674041628837585,
+      "learning_rate": 0.0002,
+      "loss": 0.2162,
+      "step": 10100
+    },
+    {
+      "epoch": 0.8708934833587918,
+      "grad_norm": 0.32635965943336487,
+      "learning_rate": 0.0002,
+      "loss": 0.2173,
+      "step": 10120
+    },
+    {
+      "epoch": 0.8726146167251135,
+      "grad_norm": 0.28276535868644714,
+      "learning_rate": 0.0002,
+      "loss": 0.2203,
+      "step": 10140
+    },
+    {
+      "epoch": 0.8743357500914352,
+      "grad_norm": 0.363016813993454,
+      "learning_rate": 0.0002,
+      "loss": 0.2355,
+      "step": 10160
+    },
+    {
+      "epoch": 0.8760568834577569,
+      "grad_norm": 0.302642822265625,
+      "learning_rate": 0.0002,
+      "loss": 0.2099,
+      "step": 10180
+    },
+    {
+      "epoch": 0.8777780168240786,
+      "grad_norm": 0.25143080949783325,
+      "learning_rate": 0.0002,
+      "loss": 0.2233,
+      "step": 10200
+    },
+    {
+      "epoch": 0.8794991501904004,
+      "grad_norm": 0.37836262583732605,
+      "learning_rate": 0.0002,
+      "loss": 0.2218,
+      "step": 10220
+    },
+    {
+      "epoch": 0.8812202835567221,
+      "grad_norm": 0.3723512887954712,
+      "learning_rate": 0.0002,
+      "loss": 0.219,
+      "step": 10240
+    },
+    {
+      "epoch": 0.8829414169230438,
+      "grad_norm": 0.33276501297950745,
+      "learning_rate": 0.0002,
+      "loss": 0.2153,
+      "step": 10260
+    },
+    {
+      "epoch": 0.8846625502893656,
+      "grad_norm": 0.3877188563346863,
+      "learning_rate": 0.0002,
+      "loss": 0.2213,
+      "step": 10280
+    },
+    {
+      "epoch": 0.8863836836556873,
+      "grad_norm": 0.36920130252838135,
+      "learning_rate": 0.0002,
+      "loss": 0.2099,
+      "step": 10300
+    },
+    {
+      "epoch": 0.8881048170220089,
+      "grad_norm": 0.2991655170917511,
+      "learning_rate": 0.0002,
+      "loss": 0.218,
+      "step": 10320
+    },
+    {
+      "epoch": 0.8898259503883307,
+      "grad_norm": 0.4785827100276947,
+      "learning_rate": 0.0002,
+      "loss": 0.2226,
+      "step": 10340
+    },
+    {
+      "epoch": 0.8915470837546524,
+      "grad_norm": 0.30903199315071106,
+      "learning_rate": 0.0002,
+      "loss": 0.2059,
+      "step": 10360
+    },
+    {
+      "epoch": 0.8932682171209742,
+      "grad_norm": 0.3614109754562378,
+      "learning_rate": 0.0002,
+      "loss": 0.2271,
+      "step": 10380
+    },
+    {
+      "epoch": 0.8949893504872959,
+      "grad_norm": 0.3223751485347748,
+      "learning_rate": 0.0002,
+      "loss": 0.2223,
+      "step": 10400
+    },
+    {
+      "epoch": 0.8967104838536176,
+      "grad_norm": 0.3175138533115387,
+      "learning_rate": 0.0002,
+      "loss": 0.2369,
+      "step": 10420
+    },
+    {
+      "epoch": 0.8984316172199394,
+      "grad_norm": 0.37820857763290405,
+      "learning_rate": 0.0002,
+      "loss": 0.2206,
+      "step": 10440
+    },
+    {
+      "epoch": 0.9001527505862611,
+      "grad_norm": 0.4233035445213318,
+      "learning_rate": 0.0002,
+      "loss": 0.2169,
+      "step": 10460
+    },
+    {
+      "epoch": 0.9018738839525827,
+      "grad_norm": 0.4142135977745056,
+      "learning_rate": 0.0002,
+      "loss": 0.21,
+      "step": 10480
+    },
+    {
+      "epoch": 0.9035950173189045,
+      "grad_norm": 0.43564194440841675,
+      "learning_rate": 0.0002,
+      "loss": 0.2231,
+      "step": 10500
+    },
+    {
+      "epoch": 0.9053161506852262,
+      "grad_norm": 0.3781276345252991,
+      "learning_rate": 0.0002,
+      "loss": 0.2258,
+      "step": 10520
+    },
+    {
+      "epoch": 0.9070372840515479,
+      "grad_norm": 0.4279311001300812,
+      "learning_rate": 0.0002,
+      "loss": 0.2287,
+      "step": 10540
+    },
+    {
+      "epoch": 0.9087584174178697,
+      "grad_norm": 0.3078250586986542,
+      "learning_rate": 0.0002,
+      "loss": 0.212,
+      "step": 10560
+    },
+    {
+      "epoch": 0.9104795507841914,
+      "grad_norm": 0.3039957880973816,
+      "learning_rate": 0.0002,
+      "loss": 0.2137,
+      "step": 10580
+    },
+    {
+      "epoch": 0.9122006841505131,
+      "grad_norm": 0.3359488546848297,
+      "learning_rate": 0.0002,
+      "loss": 0.2206,
+      "step": 10600
+    },
+    {
+      "epoch": 0.9139218175168349,
+      "grad_norm": 0.30003640055656433,
+      "learning_rate": 0.0002,
+      "loss": 0.213,
+      "step": 10620
+    },
+    {
+      "epoch": 0.9156429508831565,
+      "grad_norm": 0.3175880014896393,
+      "learning_rate": 0.0002,
+      "loss": 0.2099,
+      "step": 10640
+    },
+    {
+      "epoch": 0.9173640842494782,
+      "grad_norm": 0.3824600577354431,
+      "learning_rate": 0.0002,
+      "loss": 0.2126,
+      "step": 10660
+    },
+    {
+      "epoch": 0.9190852176158,
+      "grad_norm": 0.3589145839214325,
+      "learning_rate": 0.0002,
+      "loss": 0.2234,
+      "step": 10680
+    },
+    {
+      "epoch": 0.9208063509821217,
+      "grad_norm": 0.3821096420288086,
+      "learning_rate": 0.0002,
+      "loss": 0.2163,
+      "step": 10700
+    },
+    {
+      "epoch": 0.9225274843484434,
+      "grad_norm": 0.41358160972595215,
+      "learning_rate": 0.0002,
+      "loss": 0.2207,
+      "step": 10720
+    },
+    {
+      "epoch": 0.9242486177147652,
+      "grad_norm": 0.2841893434524536,
+      "learning_rate": 0.0002,
+      "loss": 0.2418,
+      "step": 10740
+    },
+    {
+      "epoch": 0.9259697510810869,
+      "grad_norm": 0.29627254605293274,
+      "learning_rate": 0.0002,
+      "loss": 0.2259,
+      "step": 10760
+    },
+    {
+      "epoch": 0.9276908844474087,
+      "grad_norm": 0.3516784906387329,
+      "learning_rate": 0.0002,
+      "loss": 0.2117,
+      "step": 10780
+    },
+    {
+      "epoch": 0.9294120178137304,
+      "grad_norm": 0.37627148628234863,
+      "learning_rate": 0.0002,
+      "loss": 0.2236,
+      "step": 10800
+    },
+    {
+      "epoch": 0.931133151180052,
+      "grad_norm": 0.3956553041934967,
+      "learning_rate": 0.0002,
+      "loss": 0.2091,
+      "step": 10820
+    },
+    {
+      "epoch": 0.9328542845463738,
+      "grad_norm": 0.3255549967288971,
+      "learning_rate": 0.0002,
+      "loss": 0.2247,
+      "step": 10840
+    },
+    {
+      "epoch": 0.9345754179126955,
+      "grad_norm": 0.3365786075592041,
+      "learning_rate": 0.0002,
+      "loss": 0.1955,
+      "step": 10860
+    },
+    {
+      "epoch": 0.9362965512790172,
+      "grad_norm": 0.3629949688911438,
+      "learning_rate": 0.0002,
+      "loss": 0.2137,
+      "step": 10880
+    },
+    {
+      "epoch": 0.938017684645339,
+      "grad_norm": 0.3127789795398712,
+      "learning_rate": 0.0002,
+      "loss": 0.2077,
+      "step": 10900
+    },
+    {
+      "epoch": 0.9397388180116607,
+      "grad_norm": 0.3616786003112793,
+      "learning_rate": 0.0002,
+      "loss": 0.2125,
+      "step": 10920
+    },
+    {
+      "epoch": 0.9414599513779824,
+      "grad_norm": 0.33411502838134766,
+      "learning_rate": 0.0002,
+      "loss": 0.2137,
+      "step": 10940
+    },
+    {
+      "epoch": 0.9431810847443042,
+      "grad_norm": 0.3387938439846039,
+      "learning_rate": 0.0002,
+      "loss": 0.2089,
+      "step": 10960
+    },
+    {
+      "epoch": 0.9449022181106258,
+      "grad_norm": 0.42915311455726624,
+      "learning_rate": 0.0002,
+      "loss": 0.2144,
+      "step": 10980
+    },
+    {
+      "epoch": 0.9466233514769475,
+      "grad_norm": 0.4593462646007538,
+      "learning_rate": 0.0002,
+      "loss": 0.2136,
+      "step": 11000
+    },
+    {
+      "epoch": 0.9483444848432693,
+      "grad_norm": 0.3778610825538635,
+      "learning_rate": 0.0002,
+      "loss": 0.2198,
+      "step": 11020
+    },
+    {
+      "epoch": 0.950065618209591,
+      "grad_norm": 0.34387239813804626,
+      "learning_rate": 0.0002,
+      "loss": 0.2215,
+      "step": 11040
+    },
+    {
+      "epoch": 0.9517867515759127,
+      "grad_norm": 0.38071706891059875,
+      "learning_rate": 0.0002,
+      "loss": 0.2159,
+      "step": 11060
+    },
+    {
+      "epoch": 0.9535078849422345,
+      "grad_norm": 0.3698328137397766,
+      "learning_rate": 0.0002,
+      "loss": 0.214,
+      "step": 11080
+    },
+    {
+      "epoch": 0.9552290183085562,
+      "grad_norm": 0.43745896220207214,
+      "learning_rate": 0.0002,
+      "loss": 0.2252,
+      "step": 11100
+    },
+    {
+      "epoch": 0.956950151674878,
+      "grad_norm": 0.2951034903526306,
+      "learning_rate": 0.0002,
+      "loss": 0.2099,
+      "step": 11120
+    },
+    {
+      "epoch": 0.9586712850411996,
+      "grad_norm": 0.35736939311027527,
+      "learning_rate": 0.0002,
+      "loss": 0.222,
+      "step": 11140
+    },
+    {
+      "epoch": 0.9603924184075213,
+      "grad_norm": 0.3355426490306854,
+      "learning_rate": 0.0002,
+      "loss": 0.2142,
+      "step": 11160
+    },
+    {
+      "epoch": 0.9621135517738431,
+      "grad_norm": 0.2925025522708893,
+      "learning_rate": 0.0002,
+      "loss": 0.2085,
+      "step": 11180
+    },
+    {
+      "epoch": 0.9638346851401648,
+      "grad_norm": 0.36077678203582764,
+      "learning_rate": 0.0002,
+      "loss": 0.2097,
+      "step": 11200
+    },
+    {
+      "epoch": 0.9655558185064865,
+      "grad_norm": 0.3234957158565521,
+      "learning_rate": 0.0002,
+      "loss": 0.2209,
+      "step": 11220
+    },
+    {
+      "epoch": 0.9672769518728083,
+      "grad_norm": 0.32046082615852356,
+      "learning_rate": 0.0002,
+      "loss": 0.2051,
+      "step": 11240
+    },
+    {
+      "epoch": 0.96899808523913,
+      "grad_norm": 0.3421325087547302,
+      "learning_rate": 0.0002,
+      "loss": 0.2288,
+      "step": 11260
+    },
+    {
+      "epoch": 0.9707192186054517,
+      "grad_norm": 0.35989081859588623,
+      "learning_rate": 0.0002,
+      "loss": 0.2061,
+      "step": 11280
+    },
+    {
+      "epoch": 0.9724403519717734,
+      "grad_norm": 0.3135656714439392,
+      "learning_rate": 0.0002,
+      "loss": 0.2238,
+      "step": 11300
+    },
+    {
+      "epoch": 0.9741614853380951,
+      "grad_norm": 0.29025885462760925,
+      "learning_rate": 0.0002,
+      "loss": 0.2111,
+      "step": 11320
+    },
+    {
+      "epoch": 0.9758826187044168,
+      "grad_norm": 0.4417757987976074,
+      "learning_rate": 0.0002,
+      "loss": 0.2,
+      "step": 11340
+    },
+    {
+      "epoch": 0.9776037520707386,
+      "grad_norm": 0.3659359812736511,
+      "learning_rate": 0.0002,
+      "loss": 0.216,
+      "step": 11360
+    },
+    {
+      "epoch": 0.9793248854370603,
+      "grad_norm": 0.3737911283969879,
+      "learning_rate": 0.0002,
+      "loss": 0.2139,
+      "step": 11380
+    },
+    {
+      "epoch": 0.981046018803382,
+      "grad_norm": 0.38346412777900696,
+      "learning_rate": 0.0002,
+      "loss": 0.2072,
+      "step": 11400
+    },
+    {
+      "epoch": 0.9827671521697038,
+      "grad_norm": 0.32871776819229126,
+      "learning_rate": 0.0002,
+      "loss": 0.2197,
+      "step": 11420
+    },
+    {
+      "epoch": 0.9844882855360255,
+      "grad_norm": 0.29058969020843506,
+      "learning_rate": 0.0002,
+      "loss": 0.211,
+      "step": 11440
+    },
+    {
+      "epoch": 0.9862094189023471,
+      "grad_norm": 0.39237943291664124,
+      "learning_rate": 0.0002,
+      "loss": 0.2053,
+      "step": 11460
+    },
+    {
+      "epoch": 0.9879305522686689,
+      "grad_norm": 0.3916817307472229,
+      "learning_rate": 0.0002,
+      "loss": 0.2177,
+      "step": 11480
+    },
+    {
+      "epoch": 0.9896516856349906,
+      "grad_norm": 0.2479276806116104,
+      "learning_rate": 0.0002,
+      "loss": 0.2128,
+      "step": 11500
+    },
+    {
+      "epoch": 0.9913728190013124,
+      "grad_norm": 0.33634355664253235,
+      "learning_rate": 0.0002,
+      "loss": 0.2097,
+      "step": 11520
+    },
+    {
+      "epoch": 0.9930939523676341,
+      "grad_norm": 0.336535781621933,
+      "learning_rate": 0.0002,
+      "loss": 0.208,
+      "step": 11540
+    },
+    {
+      "epoch": 0.9948150857339558,
+      "grad_norm": 0.4115926921367645,
+      "learning_rate": 0.0002,
+      "loss": 0.2057,
+      "step": 11560
+    },
+    {
+      "epoch": 0.9965362191002776,
+      "grad_norm": 0.3465426564216614,
+      "learning_rate": 0.0002,
+      "loss": 0.204,
+      "step": 11580
+    },
+    {
+      "epoch": 0.9982573524665993,
+      "grad_norm": 0.32643455266952515,
+      "learning_rate": 0.0002,
+      "loss": 0.2068,
+      "step": 11600
+    },
+    {
+      "epoch": 0.9999784858329209,
+      "grad_norm": 0.3390646278858185,
+      "learning_rate": 0.0002,
+      "loss": 0.2095,
+      "step": 11620
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.2866213619709015,
+      "eval_runtime": 903.225,
+      "eval_samples_per_second": 4.146,
+      "eval_steps_per_second": 0.519,
+      "step": 11621
     }
   ],
   "logging_steps": 20,
-  "max_steps": 14000,
+  "max_steps": 13000,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 3,
+  "num_train_epochs": 2,
   "save_steps": 77,
   "stateful_callbacks": {
     "TrainerControl": {
@@ -1889,7 +4101,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 3.423221588013613e+18,
+  "total_flos": 7.484935677507797e+18,
   "train_batch_size": 2,
   "trial_name": null,
   "trial_params": null