diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,6426 @@
+{
+  "best_metric": 4.557027339935303,
+  "best_model_checkpoint": "/Users/bbunzeck/Documents/lexdec-llamas/small-bpe/checkpoint-8136",
+  "epoch": 0.9999723214038584,
+  "eval_steps": 904,
+  "global_step": 9032,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.001107143845664148,
+      "grad_norm": 0.9944754838943481,
+      "learning_rate": 1.4999999999999999e-05,
+      "loss": 8.2064,
+      "step": 10
+    },
+    {
+      "epoch": 0.002214287691328296,
+      "grad_norm": 0.9404617547988892,
+      "learning_rate": 2.9999999999999997e-05,
+      "loss": 8.2419,
+      "step": 20
+    },
+    {
+      "epoch": 0.0033214315369924437,
+      "grad_norm": 1.221259593963623,
+      "learning_rate": 4.4999999999999996e-05,
+      "loss": 8.186,
+      "step": 30
+    },
+    {
+      "epoch": 0.004428575382656592,
+      "grad_norm": 1.1741901636123657,
+      "learning_rate": 5.9999999999999995e-05,
+      "loss": 8.1401,
+      "step": 40
+    },
+    {
+      "epoch": 0.00553571922832074,
+      "grad_norm": 1.2530901432037354,
+      "learning_rate": 7.5e-05,
+      "loss": 7.9827,
+      "step": 50
+    },
+    {
+      "epoch": 0.006642863073984887,
+      "grad_norm": 1.1453334093093872,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 7.9182,
+      "step": 60
+    },
+    {
+      "epoch": 0.007750006919649035,
+      "grad_norm": 1.1157323122024536,
+      "learning_rate": 0.00010499999999999999,
+      "loss": 7.892,
+      "step": 70
+    },
+    {
+      "epoch": 0.008857150765313184,
+      "grad_norm": 1.0198190212249756,
+      "learning_rate": 0.00011999999999999999,
+      "loss": 7.673,
+      "step": 80
+    },
+    {
+      "epoch": 0.009964294610977332,
+      "grad_norm": 1.1610361337661743,
+      "learning_rate": 0.000135,
+      "loss": 7.5399,
+      "step": 90
+    },
+    {
+      "epoch": 0.01107143845664148,
+      "grad_norm": 1.0590828657150269,
+      "learning_rate": 0.00015,
+      "loss": 7.4137,
+      "step": 100
+    },
+    {
+      "epoch": 0.012178582302305627,
+      "grad_norm": 1.1147730350494385,
+      "learning_rate": 0.000165,
+      "loss": 7.2446,
+      "step": 110
+    },
+    {
+      "epoch": 0.013285726147969775,
+      "grad_norm": 1.0851718187332153,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 7.0368,
+      "step": 120
+    },
+    {
+      "epoch": 0.014392869993633923,
+      "grad_norm": 1.051336646080017,
+      "learning_rate": 0.000195,
+      "loss": 6.9373,
+      "step": 130
+    },
+    {
+      "epoch": 0.01550001383929807,
+      "grad_norm": 0.9345677495002747,
+      "learning_rate": 0.00020999999999999998,
+      "loss": 6.7535,
+      "step": 140
+    },
+    {
+      "epoch": 0.01660715768496222,
+      "grad_norm": 0.9441080093383789,
+      "learning_rate": 0.000225,
+      "loss": 6.5367,
+      "step": 150
+    },
+    {
+      "epoch": 0.017714301530626368,
+      "grad_norm": 0.7599549293518066,
+      "learning_rate": 0.00023999999999999998,
+      "loss": 6.5257,
+      "step": 160
+    },
+    {
+      "epoch": 0.018821445376290515,
+      "grad_norm": 0.7043230533599854,
+      "learning_rate": 0.00025499999999999996,
+      "loss": 6.2486,
+      "step": 170
+    },
+    {
+      "epoch": 0.019928589221954663,
+      "grad_norm": 0.5770099759101868,
+      "learning_rate": 0.00027,
+      "loss": 6.1883,
+      "step": 180
+    },
+    {
+      "epoch": 0.02103573306761881,
+      "grad_norm": 0.4375951886177063,
+      "learning_rate": 0.000285,
+      "loss": 6.1089,
+      "step": 190
+    },
+    {
+      "epoch": 0.02214287691328296,
+      "grad_norm": 0.49226102232933044,
+      "learning_rate": 0.0003,
+      "loss": 6.0057,
+      "step": 200
+    },
+    {
+      "epoch": 0.023250020758947106,
+      "grad_norm": 0.4686337113380432,
+      "learning_rate": 0.0002999990510519666,
+      "loss": 6.0327,
+      "step": 210
+    },
+    {
+      "epoch": 0.024357164604611254,
+      "grad_norm": 0.5040702819824219,
+      "learning_rate": 0.0002999962042198732,
+      "loss": 6.0766,
+      "step": 220
+    },
+    {
+      "epoch": 0.025464308450275402,
+      "grad_norm": 0.425012469291687,
+      "learning_rate": 0.0002999914595397396,
+      "loss": 6.0945,
+      "step": 230
+    },
+    {
+      "epoch": 0.02657145229593955,
+      "grad_norm": 0.5382273197174072,
+      "learning_rate": 0.00029998481707159863,
+      "loss": 5.9612,
+      "step": 240
+    },
+    {
+      "epoch": 0.027678596141603697,
+      "grad_norm": 0.5102465748786926,
+      "learning_rate": 0.000299976276899495,
+      "loss": 6.0474,
+      "step": 250
+    },
+    {
+      "epoch": 0.028785739987267845,
+      "grad_norm": 0.5842340588569641,
+      "learning_rate": 0.00029996583913148453,
+      "loss": 5.8334,
+      "step": 260
+    },
+    {
+      "epoch": 0.029892883832931993,
+      "grad_norm": 0.6176152229309082,
+      "learning_rate": 0.00029995350389963246,
+      "loss": 5.9422,
+      "step": 270
+    },
+    {
+      "epoch": 0.03100002767859614,
+      "grad_norm": 0.6484977006912231,
+      "learning_rate": 0.0002999392713600121,
+      "loss": 5.9173,
+      "step": 280
+    },
+    {
+      "epoch": 0.03210717152426029,
+      "grad_norm": 0.6891837120056152,
+      "learning_rate": 0.00029992314169270256,
+      "loss": 5.9265,
+      "step": 290
+    },
+    {
+      "epoch": 0.03321431536992444,
+      "grad_norm": 1.172451138496399,
+      "learning_rate": 0.0002999051151017868,
+      "loss": 5.7971,
+      "step": 300
+    },
+    {
+      "epoch": 0.03432145921558859,
+      "grad_norm": 0.9547929167747498,
+      "learning_rate": 0.0002998851918153488,
+      "loss": 5.864,
+      "step": 310
+    },
+    {
+      "epoch": 0.035428603061252735,
+      "grad_norm": 0.7213353514671326,
+      "learning_rate": 0.0002998633720854707,
+      "loss": 5.8307,
+      "step": 320
+    },
+    {
+      "epoch": 0.03653574690691688,
+      "grad_norm": 0.7909387350082397,
+      "learning_rate": 0.0002998396561882297,
+      "loss": 5.7157,
+      "step": 330
+    },
+    {
+      "epoch": 0.03764289075258103,
+      "grad_norm": 0.9111142158508301,
+      "learning_rate": 0.0002998140444236946,
+      "loss": 5.6613,
+      "step": 340
+    },
+    {
+      "epoch": 0.03875003459824518,
+      "grad_norm": 0.7047809958457947,
+      "learning_rate": 0.0002997865371159217,
+      "loss": 5.6756,
+      "step": 350
+    },
+    {
+      "epoch": 0.039857178443909326,
+      "grad_norm": 0.6649304032325745,
+      "learning_rate": 0.0002997571346129512,
+      "loss": 5.6065,
+      "step": 360
+    },
+    {
+      "epoch": 0.040964322289573474,
+      "grad_norm": 0.6929216384887695,
+      "learning_rate": 0.00029972583728680237,
+      "loss": 5.5402,
+      "step": 370
+    },
+    {
+      "epoch": 0.04207146613523762,
+      "grad_norm": 0.7586813569068909,
+      "learning_rate": 0.000299692645533469,
+      "loss": 5.6213,
+      "step": 380
+    },
+    {
+      "epoch": 0.04317860998090177,
+      "grad_norm": 0.9237579107284546,
+      "learning_rate": 0.00029965755977291443,
+      "loss": 5.5503,
+      "step": 390
+    },
+    {
+      "epoch": 0.04428575382656592,
+      "grad_norm": 0.8214355707168579,
+      "learning_rate": 0.0002996205804490661,
+      "loss": 5.53,
+      "step": 400
+    },
+    {
+      "epoch": 0.045392897672230065,
+      "grad_norm": 0.7593922019004822,
+      "learning_rate": 0.00029958170802981023,
+      "loss": 5.5515,
+      "step": 410
+    },
+    {
+      "epoch": 0.04650004151789421,
+      "grad_norm": 0.8688572645187378,
+      "learning_rate": 0.0002995409430069855,
+      "loss": 5.434,
+      "step": 420
+    },
+    {
+      "epoch": 0.04760718536355836,
+      "grad_norm": 1.099979043006897,
+      "learning_rate": 0.00029949828589637703,
+      "loss": 5.5367,
+      "step": 430
+    },
+    {
+      "epoch": 0.04871432920922251,
+      "grad_norm": 0.8518048524856567,
+      "learning_rate": 0.00029945373723771,
+      "loss": 5.3948,
+      "step": 440
+    },
+    {
+      "epoch": 0.049821473054886656,
+      "grad_norm": 1.1791898012161255,
+      "learning_rate": 0.00029940729759464247,
+      "loss": 5.3874,
+      "step": 450
+    },
+    {
+      "epoch": 0.050928616900550804,
+      "grad_norm": 0.8787040710449219,
+      "learning_rate": 0.00029935896755475865,
+      "loss": 5.4079,
+      "step": 460
+    },
+    {
+      "epoch": 0.05203576074621495,
+      "grad_norm": 0.8589553833007812,
+      "learning_rate": 0.00029930874772956113,
+      "loss": 5.3922,
+      "step": 470
+    },
+    {
+      "epoch": 0.0531429045918791,
+      "grad_norm": 0.932878851890564,
+      "learning_rate": 0.00029925663875446325,
+      "loss": 5.2226,
+      "step": 480
+    },
+    {
+      "epoch": 0.05425004843754325,
+      "grad_norm": 0.9565871953964233,
+      "learning_rate": 0.0002992026412887812,
+      "loss": 5.3662,
+      "step": 490
+    },
+    {
+      "epoch": 0.055357192283207395,
+      "grad_norm": 0.8293845057487488,
+      "learning_rate": 0.0002991467560157254,
+      "loss": 5.3237,
+      "step": 500
+    },
+    {
+      "epoch": 0.05646433612887154,
+      "grad_norm": 1.1249282360076904,
+      "learning_rate": 0.0002990889836423922,
+      "loss": 5.277,
+      "step": 510
+    },
+    {
+      "epoch": 0.05757147997453569,
+      "grad_norm": 0.9282324314117432,
+      "learning_rate": 0.00029902932489975464,
+      "loss": 5.3265,
+      "step": 520
+    },
+    {
+      "epoch": 0.05867862382019984,
+      "grad_norm": 1.0275903940200806,
+      "learning_rate": 0.0002989677805426534,
+      "loss": 5.2788,
+      "step": 530
+    },
+    {
+      "epoch": 0.059785767665863986,
+      "grad_norm": 1.0539486408233643,
+      "learning_rate": 0.000298904351349787,
+      "loss": 5.2858,
+      "step": 540
+    },
+    {
+      "epoch": 0.06089291151152813,
+      "grad_norm": 1.0935311317443848,
+      "learning_rate": 0.0002988390381237023,
+      "loss": 5.2789,
+      "step": 550
+    },
+    {
+      "epoch": 0.06200005535719228,
+      "grad_norm": 0.9337705969810486,
+      "learning_rate": 0.000298771841690784,
+      "loss": 5.1891,
+      "step": 560
+    },
+    {
+      "epoch": 0.06310719920285643,
+      "grad_norm": 0.9364931583404541,
+      "learning_rate": 0.00029870276290124443,
+      "loss": 5.2948,
+      "step": 570
+    },
+    {
+      "epoch": 0.06421434304852058,
+      "grad_norm": 0.9665646553039551,
+      "learning_rate": 0.0002986318026291127,
+      "loss": 5.1824,
+      "step": 580
+    },
+    {
+      "epoch": 0.06532148689418472,
+      "grad_norm": 1.0351765155792236,
+      "learning_rate": 0.0002985589617722236,
+      "loss": 5.1323,
+      "step": 590
+    },
+    {
+      "epoch": 0.06642863073984888,
+      "grad_norm": 0.8839932680130005,
+      "learning_rate": 0.0002984842412522063,
+      "loss": 5.3387,
+      "step": 600
+    },
+    {
+      "epoch": 0.06753577458551302,
+      "grad_norm": 1.0527654886245728,
+      "learning_rate": 0.00029840764201447266,
+      "loss": 5.2098,
+      "step": 610
+    },
+    {
+      "epoch": 0.06864291843117717,
+      "grad_norm": 1.0866479873657227,
+      "learning_rate": 0.0002983291650282053,
+      "loss": 5.1479,
+      "step": 620
+    },
+    {
+      "epoch": 0.06975006227684132,
+      "grad_norm": 0.9815286993980408,
+      "learning_rate": 0.0002982488112863453,
+      "loss": 5.1722,
+      "step": 630
+    },
+    {
+      "epoch": 0.07085720612250547,
+      "grad_norm": 1.2467665672302246,
+      "learning_rate": 0.0002981665818055797,
+      "loss": 5.0829,
+      "step": 640
+    },
+    {
+      "epoch": 0.07196434996816961,
+      "grad_norm": 1.0246318578720093,
+      "learning_rate": 0.0002980824776263286,
+      "loss": 5.1367,
+      "step": 650
+    },
+    {
+      "epoch": 0.07307149381383377,
+      "grad_norm": 1.0245940685272217,
+      "learning_rate": 0.00029799649981273186,
+      "loss": 5.1738,
+      "step": 660
+    },
+    {
+      "epoch": 0.0741786376594979,
+      "grad_norm": 1.1149893999099731,
+      "learning_rate": 0.0002979086494526358,
+      "loss": 5.2175,
+      "step": 670
+    },
+    {
+      "epoch": 0.07528578150516206,
+      "grad_norm": 1.0425163507461548,
+      "learning_rate": 0.0002978189276575795,
+      "loss": 5.1999,
+      "step": 680
+    },
+    {
+      "epoch": 0.0763929253508262,
+      "grad_norm": 1.0060694217681885,
+      "learning_rate": 0.0002977273355627806,
+      "loss": 5.2357,
+      "step": 690
+    },
+    {
+      "epoch": 0.07750006919649036,
+      "grad_norm": 1.0960465669631958,
+      "learning_rate": 0.0002976338743271209,
+      "loss": 5.1191,
+      "step": 700
+    },
+    {
+      "epoch": 0.0786072130421545,
+      "grad_norm": 0.9182215929031372,
+      "learning_rate": 0.0002975385451331318,
+      "loss": 5.0399,
+      "step": 710
+    },
+    {
+      "epoch": 0.07971435688781865,
+      "grad_norm": 1.1952297687530518,
+      "learning_rate": 0.00029744134918697937,
+      "loss": 5.1009,
+      "step": 720
+    },
+    {
+      "epoch": 0.08082150073348279,
+      "grad_norm": 1.0022062063217163,
+      "learning_rate": 0.0002973422877184489,
+      "loss": 5.1422,
+      "step": 730
+    },
+    {
+      "epoch": 0.08192864457914695,
+      "grad_norm": 1.1099451780319214,
+      "learning_rate": 0.00029724136198092967,
+      "loss": 5.1563,
+      "step": 740
+    },
+    {
+      "epoch": 0.08303578842481109,
+      "grad_norm": 0.997131884098053,
+      "learning_rate": 0.00029713857325139857,
+      "loss": 5.1601,
+      "step": 750
+    },
+    {
+      "epoch": 0.08414293227047524,
+      "grad_norm": 1.093285322189331,
+      "learning_rate": 0.00029703392283040453,
+      "loss": 5.1025,
+      "step": 760
+    },
+    {
+      "epoch": 0.08525007611613938,
+      "grad_norm": 1.0651228427886963,
+      "learning_rate": 0.0002969274120420517,
+      "loss": 5.0763,
+      "step": 770
+    },
+    {
+      "epoch": 0.08635721996180354,
+      "grad_norm": 1.1181660890579224,
+      "learning_rate": 0.00029681904223398273,
+      "loss": 5.0613,
+      "step": 780
+    },
+    {
+      "epoch": 0.08746436380746768,
+      "grad_norm": 1.138413667678833,
+      "learning_rate": 0.0002967088147773619,
+      "loss": 5.0818,
+      "step": 790
+    },
+    {
+      "epoch": 0.08857150765313183,
+      "grad_norm": 1.072251796722412,
+      "learning_rate": 0.0002965967310668576,
+      "loss": 4.872,
+      "step": 800
+    },
+    {
+      "epoch": 0.08967865149879597,
+      "grad_norm": 0.9996694922447205,
+      "learning_rate": 0.00029648279252062463,
+      "loss": 5.0797,
+      "step": 810
+    },
+    {
+      "epoch": 0.09078579534446013,
+      "grad_norm": 0.9671193957328796,
+      "learning_rate": 0.00029636700058028654,
+      "loss": 4.9804,
+      "step": 820
+    },
+    {
+      "epoch": 0.09189293919012427,
+      "grad_norm": 0.9810983538627625,
+      "learning_rate": 0.0002962493567109171,
+      "loss": 5.0741,
+      "step": 830
+    },
+    {
+      "epoch": 0.09300008303578843,
+      "grad_norm": 1.0195825099945068,
+      "learning_rate": 0.00029612986240102186,
+      "loss": 5.0606,
+      "step": 840
+    },
+    {
+      "epoch": 0.09410722688145257,
+      "grad_norm": 1.1596481800079346,
+      "learning_rate": 0.0002960085191625193,
+      "loss": 4.9347,
+      "step": 850
+    },
+    {
+      "epoch": 0.09521437072711672,
+      "grad_norm": 1.253517746925354,
+      "learning_rate": 0.00029588532853072196,
+      "loss": 5.0459,
+      "step": 860
+    },
+    {
+      "epoch": 0.09632151457278088,
+      "grad_norm": 1.123814344406128,
+      "learning_rate": 0.0002957602920643164,
+      "loss": 4.9886,
+      "step": 870
+    },
+    {
+      "epoch": 0.09742865841844502,
+      "grad_norm": 1.0790519714355469,
+      "learning_rate": 0.00029563341134534424,
+      "loss": 5.0198,
+      "step": 880
+    },
+    {
+      "epoch": 0.09853580226410917,
+      "grad_norm": 1.1297645568847656,
+      "learning_rate": 0.0002955046879791816,
+      "loss": 5.1003,
+      "step": 890
+    },
+    {
+      "epoch": 0.09964294610977331,
+      "grad_norm": 1.1308685541152954,
+      "learning_rate": 0.0002953741235945188,
+      "loss": 5.0079,
+      "step": 900
+    },
+    {
+      "epoch": 0.10008580364803897,
+      "eval_loss": 5.295600414276123,
+      "eval_runtime": 47.5395,
+      "eval_samples_per_second": 1051.757,
+      "eval_steps_per_second": 131.47,
+      "step": 904
+    },
+    {
+      "epoch": 0.10075008995543747,
+      "grad_norm": 1.1201127767562866,
+      "learning_rate": 0.0002952417198433401,
+      "loss": 5.0585,
+      "step": 910
+    },
+    {
+      "epoch": 0.10185723380110161,
+      "grad_norm": 1.123083472251892,
+      "learning_rate": 0.00029510747840090273,
+      "loss": 4.9078,
+      "step": 920
+    },
+    {
+      "epoch": 0.10296437764676576,
+      "grad_norm": 0.9523604512214661,
+      "learning_rate": 0.00029497140096571516,
+      "loss": 5.0012,
+      "step": 930
+    },
+    {
+      "epoch": 0.1040715214924299,
+      "grad_norm": 1.0448756217956543,
+      "learning_rate": 0.00029483348925951635,
+      "loss": 5.0446,
+      "step": 940
+    },
+    {
+      "epoch": 0.10517866533809406,
+      "grad_norm": 1.134711742401123,
+      "learning_rate": 0.00029469374502725355,
+      "loss": 4.9057,
+      "step": 950
+    },
+    {
+      "epoch": 0.1062858091837582,
+      "grad_norm": 1.1030372381210327,
+      "learning_rate": 0.0002945521700370603,
+      "loss": 4.9095,
+      "step": 960
+    },
+    {
+      "epoch": 0.10739295302942235,
+      "grad_norm": 1.0063588619232178,
+      "learning_rate": 0.00029440876608023394,
+      "loss": 4.9076,
+      "step": 970
+    },
+    {
+      "epoch": 0.1085000968750865,
+      "grad_norm": 1.1154258251190186,
+      "learning_rate": 0.00029426353497121334,
+      "loss": 4.9939,
+      "step": 980
+    },
+    {
+      "epoch": 0.10960724072075065,
+      "grad_norm": 1.1821072101593018,
+      "learning_rate": 0.00029411647854755534,
+      "loss": 4.8976,
+      "step": 990
+    },
+    {
+      "epoch": 0.11071438456641479,
+      "grad_norm": 1.0423575639724731,
+      "learning_rate": 0.0002939675986699121,
+      "loss": 4.8958,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11182152841207894,
+      "grad_norm": 1.0003031492233276,
+      "learning_rate": 0.0002938168972220071,
+      "loss": 4.9691,
+      "step": 1010
+    },
+    {
+      "epoch": 0.11292867225774308,
+      "grad_norm": 1.1820602416992188,
+      "learning_rate": 0.0002936643761106117,
+      "loss": 4.9697,
+      "step": 1020
+    },
+    {
+      "epoch": 0.11403581610340724,
+      "grad_norm": 1.1737552881240845,
+      "learning_rate": 0.00029351003726552056,
+      "loss": 4.9232,
+      "step": 1030
+    },
+    {
+      "epoch": 0.11514295994907138,
+      "grad_norm": 1.1008082628250122,
+      "learning_rate": 0.0002933538826395276,
+      "loss": 5.0242,
+      "step": 1040
+    },
+    {
+      "epoch": 0.11625010379473553,
+      "grad_norm": 1.2802969217300415,
+      "learning_rate": 0.00029319591420840126,
+      "loss": 4.8504,
+      "step": 1050
+    },
+    {
+      "epoch": 0.11735724764039968,
+      "grad_norm": 1.3688923120498657,
+      "learning_rate": 0.0002930361339708592,
+      "loss": 4.9035,
+      "step": 1060
+    },
+    {
+      "epoch": 0.11846439148606383,
+      "grad_norm": 1.0987589359283447,
+      "learning_rate": 0.00029287454394854337,
+      "loss": 5.0339,
+      "step": 1070
+    },
+    {
+      "epoch": 0.11957153533172797,
+      "grad_norm": 1.1078542470932007,
+      "learning_rate": 0.00029271114618599416,
+      "loss": 4.7867,
+      "step": 1080
+    },
+    {
+      "epoch": 0.12067867917739213,
+      "grad_norm": 1.1067126989364624,
+      "learning_rate": 0.0002925459427506248,
+      "loss": 4.7875,
+      "step": 1090
+    },
+    {
+      "epoch": 0.12178582302305627,
+      "grad_norm": 1.0103768110275269,
+      "learning_rate": 0.00029237893573269487,
+      "loss": 4.8739,
+      "step": 1100
+    },
+    {
+      "epoch": 0.12289296686872042,
+      "grad_norm": 1.2761735916137695,
+      "learning_rate": 0.0002922101272452842,
+      "loss": 4.8965,
+      "step": 1110
+    },
+    {
+      "epoch": 0.12400011071438456,
+      "grad_norm": 1.1619023084640503,
+      "learning_rate": 0.0002920395194242658,
+      "loss": 4.8387,
+      "step": 1120
+    },
+    {
+      "epoch": 0.1251072545600487,
+      "grad_norm": 1.296080470085144,
+      "learning_rate": 0.00029186711442827915,
+      "loss": 4.8743,
+      "step": 1130
+    },
+    {
+      "epoch": 0.12621439840571286,
+      "grad_norm": 1.2153007984161377,
+      "learning_rate": 0.0002916929144387027,
+      "loss": 4.7883,
+      "step": 1140
+    },
+    {
+      "epoch": 0.127321542251377,
+      "grad_norm": 1.128342866897583,
+      "learning_rate": 0.0002915169216596262,
+      "loss": 4.883,
+      "step": 1150
+    },
+    {
+      "epoch": 0.12842868609704117,
+      "grad_norm": 1.161921739578247,
+      "learning_rate": 0.00029133913831782307,
+      "loss": 4.7789,
+      "step": 1160
+    },
+    {
+      "epoch": 0.1295358299427053,
+      "grad_norm": 1.247902750968933,
+      "learning_rate": 0.00029115956666272194,
+      "loss": 4.8069,
+      "step": 1170
+    },
+    {
+      "epoch": 0.13064297378836945,
+      "grad_norm": 1.2751232385635376,
+      "learning_rate": 0.0002909782089663785,
+      "loss": 4.8252,
+      "step": 1180
+    },
+    {
+      "epoch": 0.1317501176340336,
+      "grad_norm": 1.2364708185195923,
+      "learning_rate": 0.00029079506752344636,
+      "loss": 4.7027,
+      "step": 1190
+    },
+    {
+      "epoch": 0.13285726147969776,
+      "grad_norm": 1.0895476341247559,
+      "learning_rate": 0.0002906101446511484,
+      "loss": 4.844,
+      "step": 1200
+    },
+    {
+      "epoch": 0.13396440532536188,
+      "grad_norm": 1.0647457838058472,
+      "learning_rate": 0.0002904234426892473,
+      "loss": 4.7351,
+      "step": 1210
+    },
+    {
+      "epoch": 0.13507154917102604,
+      "grad_norm": 1.2197457551956177,
+      "learning_rate": 0.00029023496400001577,
+      "loss": 4.7525,
+      "step": 1220
+    },
+    {
+      "epoch": 0.1361786930166902,
+      "grad_norm": 1.0926841497421265,
+      "learning_rate": 0.0002900447109682069,
+      "loss": 4.7707,
+      "step": 1230
+    },
+    {
+      "epoch": 0.13728583686235435,
+      "grad_norm": 1.1523051261901855,
+      "learning_rate": 0.0002898526860010239,
+      "loss": 4.7707,
+      "step": 1240
+    },
+    {
+      "epoch": 0.13839298070801848,
+      "grad_norm": 1.257418155670166,
+      "learning_rate": 0.00028965889152808967,
+      "loss": 4.8075,
+      "step": 1250
+    },
+    {
+      "epoch": 0.13950012455368263,
+      "grad_norm": 1.2135087251663208,
+      "learning_rate": 0.00028946333000141595,
+      "loss": 4.7565,
+      "step": 1260
+    },
+    {
+      "epoch": 0.14060726839934679,
+      "grad_norm": 1.1852108240127563,
+      "learning_rate": 0.00028926600389537246,
+      "loss": 4.6742,
+      "step": 1270
+    },
+    {
+      "epoch": 0.14171441224501094,
+      "grad_norm": 1.2277404069900513,
+      "learning_rate": 0.00028906691570665545,
+      "loss": 4.8706,
+      "step": 1280
+    },
+    {
+      "epoch": 0.1428215560906751,
+      "grad_norm": 1.1161786317825317,
+      "learning_rate": 0.0002888660679542562,
+      "loss": 4.8203,
+      "step": 1290
+    },
+    {
+      "epoch": 0.14392869993633922,
+      "grad_norm": 1.3227041959762573,
+      "learning_rate": 0.00028866346317942915,
+      "loss": 4.6574,
+      "step": 1300
+    },
+    {
+      "epoch": 0.14503584378200338,
+      "grad_norm": 1.3273063898086548,
+      "learning_rate": 0.00028845910394565956,
+      "loss": 4.8252,
+      "step": 1310
+    },
+    {
+      "epoch": 0.14614298762766753,
+      "grad_norm": 1.2453035116195679,
+      "learning_rate": 0.00028825299283863137,
+      "loss": 4.7731,
+      "step": 1320
+    },
+    {
+      "epoch": 0.14725013147333169,
+      "grad_norm": 1.2136322259902954,
+      "learning_rate": 0.0002880451324661944,
+      "loss": 4.7202,
+      "step": 1330
+    },
+    {
+      "epoch": 0.1483572753189958,
+      "grad_norm": 1.2811789512634277,
+      "learning_rate": 0.00028783552545833115,
+      "loss": 4.7536,
+      "step": 1340
+    },
+    {
+      "epoch": 0.14946441916465997,
+      "grad_norm": 1.458082914352417,
+      "learning_rate": 0.00028762417446712363,
+      "loss": 4.7476,
+      "step": 1350
+    },
+    {
+      "epoch": 0.15057156301032412,
+      "grad_norm": 1.1365827322006226,
+      "learning_rate": 0.00028741108216672015,
+      "loss": 4.8866,
+      "step": 1360
+    },
+    {
+      "epoch": 0.15167870685598828,
+      "grad_norm": 1.256729245185852,
+      "learning_rate": 0.00028719625125330075,
+      "loss": 4.8032,
+      "step": 1370
+    },
+    {
+      "epoch": 0.1527858507016524,
+      "grad_norm": 1.1410046815872192,
+      "learning_rate": 0.00028697968444504386,
+      "loss": 4.723,
+      "step": 1380
+    },
+    {
+      "epoch": 0.15389299454731656,
+      "grad_norm": 1.2968374490737915,
+      "learning_rate": 0.0002867613844820914,
+      "loss": 4.7044,
+      "step": 1390
+    },
+    {
+      "epoch": 0.1550001383929807,
+      "grad_norm": 1.3087129592895508,
+      "learning_rate": 0.0002865413541265143,
+      "loss": 4.799,
+      "step": 1400
+    },
+    {
+      "epoch": 0.15610728223864487,
+      "grad_norm": 1.1476242542266846,
+      "learning_rate": 0.0002863195961622776,
+      "loss": 4.7439,
+      "step": 1410
+    },
+    {
+      "epoch": 0.157214426084309,
+      "grad_norm": 1.4319286346435547,
+      "learning_rate": 0.00028609611339520496,
+      "loss": 4.721,
+      "step": 1420
+    },
+    {
+      "epoch": 0.15832156992997315,
+      "grad_norm": 1.1652687788009644,
+      "learning_rate": 0.0002858709086529436,
+      "loss": 4.6419,
+      "step": 1430
+    },
+    {
+      "epoch": 0.1594287137756373,
+      "grad_norm": 1.1913334131240845,
+      "learning_rate": 0.0002856439847849281,
+      "loss": 4.6707,
+      "step": 1440
+    },
+    {
+      "epoch": 0.16053585762130146,
+      "grad_norm": 1.176133394241333,
+      "learning_rate": 0.00028541534466234463,
+      "loss": 4.7649,
+      "step": 1450
+    },
+    {
+      "epoch": 0.16164300146696559,
+      "grad_norm": 1.2382426261901855,
+      "learning_rate": 0.0002851849911780943,
+      "loss": 4.7861,
+      "step": 1460
+    },
+    {
+      "epoch": 0.16275014531262974,
+      "grad_norm": 1.4298615455627441,
+      "learning_rate": 0.00028495292724675707,
+      "loss": 4.7134,
+      "step": 1470
+    },
+    {
+      "epoch": 0.1638572891582939,
+      "grad_norm": 1.3140671253204346,
+      "learning_rate": 0.0002847191558045544,
+      "loss": 4.6838,
+      "step": 1480
+    },
+    {
+      "epoch": 0.16496443300395805,
+      "grad_norm": 1.303039312362671,
+      "learning_rate": 0.00028448367980931223,
+      "loss": 4.6288,
+      "step": 1490
+    },
+    {
+      "epoch": 0.16607157684962218,
+      "grad_norm": 1.2238134145736694,
+      "learning_rate": 0.0002842465022404237,
+      "loss": 4.7817,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16717872069528633,
+      "grad_norm": 1.2683777809143066,
+      "learning_rate": 0.0002840076260988113,
+      "loss": 4.611,
+      "step": 1510
+    },
+    {
+      "epoch": 0.16828586454095049,
+      "grad_norm": 1.3005638122558594,
+      "learning_rate": 0.0002837670544068891,
+      "loss": 4.7329,
+      "step": 1520
+    },
+    {
+      "epoch": 0.16939300838661464,
+      "grad_norm": 1.2778176069259644,
+      "learning_rate": 0.00028352479020852395,
+      "loss": 4.7198,
+      "step": 1530
+    },
+    {
+      "epoch": 0.17050015223227877,
+      "grad_norm": 1.06227707862854,
+      "learning_rate": 0.00028328083656899784,
+      "loss": 4.6917,
+      "step": 1540
+    },
+    {
+      "epoch": 0.17160729607794292,
+      "grad_norm": 1.302010416984558,
+      "learning_rate": 0.0002830351965749684,
+      "loss": 4.6646,
+      "step": 1550
+    },
+    {
+      "epoch": 0.17271443992360708,
+      "grad_norm": 1.231354832649231,
+      "learning_rate": 0.0002827878733344301,
+      "loss": 4.721,
+      "step": 1560
+    },
+    {
+      "epoch": 0.17382158376927123,
+      "grad_norm": 1.3416827917099,
+      "learning_rate": 0.00028253886997667505,
+      "loss": 4.6533,
+      "step": 1570
+    },
+    {
+      "epoch": 0.17492872761493536,
+      "grad_norm": 1.2240086793899536,
+      "learning_rate": 0.0002822881896522532,
+      "loss": 4.6015,
+      "step": 1580
+    },
+    {
+      "epoch": 0.1760358714605995,
+      "grad_norm": 1.084755539894104,
+      "learning_rate": 0.0002820358355329326,
+      "loss": 4.7156,
+      "step": 1590
+    },
+    {
+      "epoch": 0.17714301530626367,
+      "grad_norm": 1.375949501991272,
+      "learning_rate": 0.00028178181081165907,
+      "loss": 4.6351,
+      "step": 1600
+    },
+    {
+      "epoch": 0.17825015915192782,
+      "grad_norm": 1.292930006980896,
+      "learning_rate": 0.0002815261187025162,
+      "loss": 4.6578,
+      "step": 1610
+    },
+    {
+      "epoch": 0.17935730299759195,
+      "grad_norm": 1.2997933626174927,
+      "learning_rate": 0.00028126876244068415,
+      "loss": 4.6957,
+      "step": 1620
+    },
+    {
+      "epoch": 0.1804644468432561,
+      "grad_norm": 1.2429704666137695,
+      "learning_rate": 0.00028100974528239936,
+      "loss": 4.6121,
+      "step": 1630
+    },
+    {
+      "epoch": 0.18157159068892026,
+      "grad_norm": 1.1794031858444214,
+      "learning_rate": 0.0002807490705049127,
+      "loss": 4.6287,
+      "step": 1640
+    },
+    {
+      "epoch": 0.1826787345345844,
+      "grad_norm": 1.2829527854919434,
+      "learning_rate": 0.0002804867414064483,
+      "loss": 4.6444,
+      "step": 1650
+    },
+    {
+      "epoch": 0.18378587838024854,
+      "grad_norm": 1.2679721117019653,
+      "learning_rate": 0.00028022276130616216,
+      "loss": 4.6875,
+      "step": 1660
+    },
+    {
+      "epoch": 0.1848930222259127,
+      "grad_norm": 1.3782397508621216,
+      "learning_rate": 0.0002799571335440994,
+      "loss": 4.6057,
+      "step": 1670
+    },
+    {
+      "epoch": 0.18600016607157685,
+      "grad_norm": 1.3364990949630737,
+      "learning_rate": 0.00027968986148115265,
+      "loss": 4.6526,
+      "step": 1680
+    },
+    {
+      "epoch": 0.187107309917241,
+      "grad_norm": 1.1953457593917847,
+      "learning_rate": 0.0002794209484990192,
+      "loss": 4.6452,
+      "step": 1690
+    },
+    {
+      "epoch": 0.18821445376290513,
+      "grad_norm": 1.2697588205337524,
+      "learning_rate": 0.0002791503980001583,
+      "loss": 4.5906,
+      "step": 1700
+    },
+    {
+      "epoch": 0.1893215976085693,
+      "grad_norm": 1.3566571474075317,
+      "learning_rate": 0.00027887821340774816,
+      "loss": 4.6267,
+      "step": 1710
+    },
+    {
+      "epoch": 0.19042874145423344,
+      "grad_norm": 1.3777858018875122,
+      "learning_rate": 0.0002786043981656426,
+      "loss": 4.655,
+      "step": 1720
+    },
+    {
+      "epoch": 0.1915358852998976,
+      "grad_norm": 1.2331784963607788,
+      "learning_rate": 0.00027832895573832734,
+      "loss": 4.6571,
+      "step": 1730
+    },
+    {
+      "epoch": 0.19264302914556175,
+      "grad_norm": 1.1953262090682983,
+      "learning_rate": 0.0002780518896108765,
+      "loss": 4.709,
+      "step": 1740
+    },
+    {
+      "epoch": 0.19375017299122588,
+      "grad_norm": 1.3000081777572632,
+      "learning_rate": 0.0002777732032889081,
+      "loss": 4.5183,
+      "step": 1750
+    },
+    {
+      "epoch": 0.19485731683689003,
+      "grad_norm": 1.2914705276489258,
+      "learning_rate": 0.00027749290029853993,
+      "loss": 4.5772,
+      "step": 1760
+    },
+    {
+      "epoch": 0.1959644606825542,
+      "grad_norm": 1.135388970375061,
+      "learning_rate": 0.00027721098418634495,
+      "loss": 4.7229,
+      "step": 1770
+    },
+    {
+      "epoch": 0.19707160452821834,
+      "grad_norm": 1.4572337865829468,
+      "learning_rate": 0.0002769274585193065,
+      "loss": 4.6433,
+      "step": 1780
+    },
+    {
+      "epoch": 0.19817874837388247,
+      "grad_norm": 1.5552626848220825,
+      "learning_rate": 0.0002766423268847727,
+      "loss": 4.6363,
+      "step": 1790
+    },
+    {
+      "epoch": 0.19928589221954662,
+      "grad_norm": 1.333524465560913,
+      "learning_rate": 0.0002763555928904117,
+      "loss": 4.6329,
+      "step": 1800
+    },
+    {
+      "epoch": 0.20017160729607794,
+      "eval_loss": 5.013429164886475,
+      "eval_runtime": 47.4619,
+      "eval_samples_per_second": 1053.477,
+      "eval_steps_per_second": 131.685,
+      "step": 1808
+    },
+    {
+      "epoch": 0.20039303606521078,
+      "grad_norm": 1.2497128248214722,
+      "learning_rate": 0.00027606726016416567,
+      "loss": 4.5756,
+      "step": 1810
+    },
+    {
+      "epoch": 0.20150017991087493,
+      "grad_norm": 1.3386183977127075,
+      "learning_rate": 0.00027577733235420487,
+      "loss": 4.5953,
+      "step": 1820
+    },
+    {
+      "epoch": 0.20260732375653906,
+      "grad_norm": 1.2944352626800537,
+      "learning_rate": 0.0002754858131288817,
+      "loss": 4.6047,
+      "step": 1830
+    },
+    {
+      "epoch": 0.20371446760220321,
+      "grad_norm": 1.5249584913253784,
+      "learning_rate": 0.000275192706176684,
+      "loss": 4.6308,
+      "step": 1840
+    },
+    {
+      "epoch": 0.20482161144786737,
+      "grad_norm": 1.3660956621170044,
+      "learning_rate": 0.00027489801520618867,
+      "loss": 4.6085,
+      "step": 1850
+    },
+    {
+      "epoch": 0.20592875529353152,
+      "grad_norm": 1.2916579246520996,
+      "learning_rate": 0.0002746017439460146,
+      "loss": 4.6402,
+      "step": 1860
+    },
+    {
+      "epoch": 0.20703589913919565,
+      "grad_norm": 1.3989869356155396,
+      "learning_rate": 0.0002743038961447757,
+      "loss": 4.694,
+      "step": 1870
+    },
+    {
+      "epoch": 0.2081430429848598,
+      "grad_norm": 1.320907711982727,
+      "learning_rate": 0.0002740044755710328,
+      "loss": 4.5567,
+      "step": 1880
+    },
+    {
+      "epoch": 0.20925018683052396,
+      "grad_norm": 1.3325960636138916,
+      "learning_rate": 0.000273703486013247,
+      "loss": 4.6744,
+      "step": 1890
+    },
+    {
+      "epoch": 0.21035733067618811,
+      "grad_norm": 1.3717817068099976,
+      "learning_rate": 0.0002734009312797309,
+      "loss": 4.6281,
+      "step": 1900
+    },
+    {
+      "epoch": 0.21146447452185224,
+      "grad_norm": 1.429818868637085,
+      "learning_rate": 0.0002730968151986007,
+      "loss": 4.5887,
+      "step": 1910
+    },
+    {
+      "epoch": 0.2125716183675164,
+      "grad_norm": 1.26041841506958,
+      "learning_rate": 0.0002727911416177278,
+      "loss": 4.5681,
+      "step": 1920
+    },
+    {
+      "epoch": 0.21367876221318055,
+      "grad_norm": 1.328595757484436,
+      "learning_rate": 0.0002724839144046902,
+      "loss": 4.6375,
+      "step": 1930
+    },
+    {
+      "epoch": 0.2147859060588447,
+      "grad_norm": 1.2225375175476074,
+      "learning_rate": 0.00027217513744672337,
+      "loss": 4.6839,
+      "step": 1940
+    },
+    {
+      "epoch": 0.21589304990450883,
+      "grad_norm": 1.3448318243026733,
+      "learning_rate": 0.000271864814650671,
+      "loss": 4.5533,
+      "step": 1950
+    },
+    {
+      "epoch": 0.217000193750173,
+      "grad_norm": 1.2313207387924194,
+      "learning_rate": 0.00027155294994293604,
+      "loss": 4.6695,
+      "step": 1960
+    },
+    {
+      "epoch": 0.21810733759583714,
+      "grad_norm": 1.2184146642684937,
+      "learning_rate": 0.0002712395472694304,
+      "loss": 4.5319,
+      "step": 1970
+    },
+    {
+      "epoch": 0.2192144814415013,
+      "grad_norm": 1.3842166662216187,
+      "learning_rate": 0.0002709246105955254,
+      "loss": 4.5675,
+      "step": 1980
+    },
+    {
+      "epoch": 0.22032162528716542,
+      "grad_norm": 1.3407012224197388,
+      "learning_rate": 0.0002706081439060016,
+      "loss": 4.5522,
+      "step": 1990
+    },
+    {
+      "epoch": 0.22142876913282958,
+      "grad_norm": 1.3984525203704834,
+      "learning_rate": 0.00027029015120499825,
+      "loss": 4.6108,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22253591297849373,
+      "grad_norm": 1.279784917831421,
+      "learning_rate": 0.00026997063651596257,
+      "loss": 4.6693,
+      "step": 2010
+    },
+    {
+      "epoch": 0.2236430568241579,
+      "grad_norm": 1.3760930299758911,
+      "learning_rate": 0.00026964960388159903,
+      "loss": 4.6629,
+      "step": 2020
+    },
+    {
+      "epoch": 0.22475020066982201,
+      "grad_norm": 1.3632715940475464,
+      "learning_rate": 0.00026932705736381826,
+      "loss": 4.5611,
+      "step": 2030
+    },
+    {
+      "epoch": 0.22585734451548617,
+      "grad_norm": 1.3264988660812378,
+      "learning_rate": 0.00026900300104368524,
+      "loss": 4.6813,
+      "step": 2040
+    },
+    {
+      "epoch": 0.22696448836115032,
+      "grad_norm": 1.3927732706069946,
+      "learning_rate": 0.0002686774390213682,
+      "loss": 4.5393,
+      "step": 2050
+    },
+    {
+      "epoch": 0.22807163220681448,
+      "grad_norm": 1.213679552078247,
+      "learning_rate": 0.00026835037541608623,
+      "loss": 4.6251,
+      "step": 2060
+    },
+    {
+      "epoch": 0.2291787760524786,
+      "grad_norm": 1.4621424674987793,
+      "learning_rate": 0.00026802181436605757,
+      "loss": 4.5794,
+      "step": 2070
+    },
+    {
+      "epoch": 0.23028591989814276,
+      "grad_norm": 1.398659110069275,
+      "learning_rate": 0.00026769176002844705,
+      "loss": 4.5307,
+      "step": 2080
+    },
+    {
+      "epoch": 0.23139306374380691,
+      "grad_norm": 1.3843495845794678,
+      "learning_rate": 0.0002673602165793136,
+      "loss": 4.5607,
+      "step": 2090
+    },
+    {
+      "epoch": 0.23250020758947107,
+      "grad_norm": 1.3701854944229126,
+      "learning_rate": 0.00026702718821355713,
+      "loss": 4.4907,
+      "step": 2100
+    },
+    {
+      "epoch": 0.2336073514351352,
+      "grad_norm": 1.2520064115524292,
+      "learning_rate": 0.00026669267914486596,
+      "loss": 4.5715,
+      "step": 2110
+    },
+    {
+      "epoch": 0.23471449528079935,
+      "grad_norm": 1.8549680709838867,
+      "learning_rate": 0.00026635669360566296,
+      "loss": 4.5432,
+      "step": 2120
+    },
+    {
+      "epoch": 0.2358216391264635,
+      "grad_norm": 1.323202133178711,
+      "learning_rate": 0.0002660192358470524,
+      "loss": 4.538,
+      "step": 2130
+    },
+    {
+      "epoch": 0.23692878297212766,
+      "grad_norm": 1.5440722703933716,
+      "learning_rate": 0.000265680310138766,
+      "loss": 4.5499,
+      "step": 2140
+    },
+    {
+      "epoch": 0.23803592681779182,
+      "grad_norm": 1.3992176055908203,
+      "learning_rate": 0.00026533992076910876,
+      "loss": 4.5416,
+      "step": 2150
+    },
+    {
+      "epoch": 0.23914307066345594,
+      "grad_norm": 1.3470678329467773,
+      "learning_rate": 0.00026499807204490505,
+      "loss": 4.5467,
+      "step": 2160
+    },
+    {
+      "epoch": 0.2402502145091201,
+      "grad_norm": 1.2666015625,
+      "learning_rate": 0.00026465476829144394,
+      "loss": 4.558,
+      "step": 2170
+    },
+    {
+      "epoch": 0.24135735835478425,
+      "grad_norm": 1.491350531578064,
+      "learning_rate": 0.00026431001385242433,
+      "loss": 4.5429,
+      "step": 2180
+    },
+    {
+      "epoch": 0.2424645022004484,
+      "grad_norm": 1.2662105560302734,
+      "learning_rate": 0.00026396381308990013,
+      "loss": 4.4351,
+      "step": 2190
+    },
+    {
+      "epoch": 0.24357164604611253,
+      "grad_norm": 1.2961875200271606,
+      "learning_rate": 0.00026361617038422517,
+      "loss": 4.5447,
+      "step": 2200
+    },
+    {
+      "epoch": 0.2446787898917767,
+      "grad_norm": 1.3910024166107178,
+      "learning_rate": 0.0002632670901339976,
+      "loss": 4.5399,
+      "step": 2210
+    },
+    {
+      "epoch": 0.24578593373744084,
+      "grad_norm": 1.2439371347427368,
+      "learning_rate": 0.00026291657675600426,
+      "loss": 4.4318,
+      "step": 2220
+    },
+    {
+      "epoch": 0.246893077583105,
+      "grad_norm": 1.3151521682739258,
+      "learning_rate": 0.00026256463468516495,
+      "loss": 4.491,
+      "step": 2230
+    },
+    {
+      "epoch": 0.24800022142876912,
+      "grad_norm": 1.5120826959609985,
+      "learning_rate": 0.00026221126837447617,
+      "loss": 4.5172,
+      "step": 2240
+    },
+    {
+      "epoch": 0.24910736527443328,
+      "grad_norm": 1.4124914407730103,
+      "learning_rate": 0.00026185648229495477,
+      "loss": 4.607,
+      "step": 2250
+    },
+    {
+      "epoch": 0.2502145091200974,
+      "grad_norm": 1.3351086378097534,
+      "learning_rate": 0.0002615002809355814,
+      "loss": 4.4883,
+      "step": 2260
+    },
+    {
+      "epoch": 0.2513216529657616,
+      "grad_norm": 1.389976143836975,
+      "learning_rate": 0.00026114266880324387,
+      "loss": 4.6046,
+      "step": 2270
+    },
+    {
+      "epoch": 0.2524287968114257,
+      "grad_norm": 1.510365605354309,
+      "learning_rate": 0.00026078365042267985,
+      "loss": 4.479,
+      "step": 2280
+    },
+    {
+      "epoch": 0.2535359406570899,
+      "grad_norm": 1.32261061668396,
+      "learning_rate": 0.0002604232303364198,
+      "loss": 4.5326,
+      "step": 2290
+    },
+    {
+      "epoch": 0.254643084502754,
+      "grad_norm": 1.3619463443756104,
+      "learning_rate": 0.00026006141310472953,
+      "loss": 4.5408,
+      "step": 2300
+    },
+    {
+      "epoch": 0.25575022834841815,
+      "grad_norm": 1.3347376585006714,
+      "learning_rate": 0.00025969820330555234,
+      "loss": 4.532,
+      "step": 2310
+    },
+    {
+      "epoch": 0.25685737219408233,
+      "grad_norm": 1.593105673789978,
+      "learning_rate": 0.0002593336055344513,
+      "loss": 4.5386,
+      "step": 2320
+    },
+    {
+      "epoch": 0.25796451603974646,
+      "grad_norm": 1.5122672319412231,
+      "learning_rate": 0.0002589676244045509,
+      "loss": 4.5191,
+      "step": 2330
+    },
+    {
+      "epoch": 0.2590716598854106,
+      "grad_norm": 1.1608972549438477,
+      "learning_rate": 0.0002586002645464787,
+      "loss": 4.6329,
+      "step": 2340
+    },
+    {
+      "epoch": 0.26017880373107477,
+      "grad_norm": 1.5330127477645874,
+      "learning_rate": 0.00025823153060830695,
+      "loss": 4.5624,
+      "step": 2350
+    },
+    {
+      "epoch": 0.2612859475767389,
+      "grad_norm": 1.4344722032546997,
+      "learning_rate": 0.0002578614272554936,
+      "loss": 4.458,
+      "step": 2360
+    },
+    {
+      "epoch": 0.2623930914224031,
+      "grad_norm": 1.4186034202575684,
+      "learning_rate": 0.0002574899591708233,
+      "loss": 4.5066,
+      "step": 2370
+    },
+    {
+      "epoch": 0.2635002352680672,
+      "grad_norm": 1.5190962553024292,
+      "learning_rate": 0.0002571171310543481,
+      "loss": 4.5639,
+      "step": 2380
+    },
+    {
+      "epoch": 0.26460737911373133,
+      "grad_norm": 1.4636080265045166,
+      "learning_rate": 0.00025674294762332824,
+      "loss": 4.489,
+      "step": 2390
+    },
+    {
+      "epoch": 0.2657145229593955,
+      "grad_norm": 1.430741786956787,
+      "learning_rate": 0.00025636741361217204,
+      "loss": 4.4573,
+      "step": 2400
+    },
+    {
+      "epoch": 0.26682166680505964,
+      "grad_norm": 1.2598206996917725,
+      "learning_rate": 0.0002559905337723763,
+      "loss": 4.53,
+      "step": 2410
+    },
+    {
+      "epoch": 0.26792881065072377,
+      "grad_norm": 1.387069821357727,
+      "learning_rate": 0.00025561231287246617,
+      "loss": 4.5874,
+      "step": 2420
+    },
+    {
+      "epoch": 0.26903595449638795,
+      "grad_norm": 1.3344066143035889,
+      "learning_rate": 0.00025523275569793466,
+      "loss": 4.5204,
+      "step": 2430
+    },
+    {
+      "epoch": 0.2701430983420521,
+      "grad_norm": 1.40878164768219,
+      "learning_rate": 0.00025485186705118227,
+      "loss": 4.564,
+      "step": 2440
+    },
+    {
+      "epoch": 0.27125024218771626,
+      "grad_norm": 1.4640940427780151,
+      "learning_rate": 0.00025446965175145614,
+      "loss": 4.5189,
+      "step": 2450
+    },
+    {
+      "epoch": 0.2723573860333804,
+      "grad_norm": 1.3941110372543335,
+      "learning_rate": 0.00025408611463478896,
+      "loss": 4.4818,
+      "step": 2460
+    },
+    {
+      "epoch": 0.2734645298790445,
+      "grad_norm": 1.5158843994140625,
+      "learning_rate": 0.0002537012605539379,
+      "loss": 4.4386,
+      "step": 2470
+    },
+    {
+      "epoch": 0.2745716737247087,
+      "grad_norm": 1.4512423276901245,
+      "learning_rate": 0.0002533150943783234,
+      "loss": 4.3768,
+      "step": 2480
+    },
+    {
+      "epoch": 0.2756788175703728,
+      "grad_norm": 1.5543454885482788,
+      "learning_rate": 0.0002529276209939672,
+      "loss": 4.4956,
+      "step": 2490
+    },
+    {
+      "epoch": 0.27678596141603695,
+      "grad_norm": 1.4016339778900146,
+      "learning_rate": 0.0002525388453034307,
+      "loss": 4.3915,
+      "step": 2500
+    },
+    {
+      "epoch": 0.27789310526170113,
+      "grad_norm": 1.407719373703003,
+      "learning_rate": 0.00025214877222575285,
+      "loss": 4.4623,
+      "step": 2510
+    },
+    {
+      "epoch": 0.27900024910736526,
+      "grad_norm": 1.3623199462890625,
+      "learning_rate": 0.0002517574066963882,
+      "loss": 4.5783,
+      "step": 2520
+    },
+    {
+      "epoch": 0.28010739295302944,
+      "grad_norm": 1.4061545133590698,
+      "learning_rate": 0.00025136475366714395,
+      "loss": 4.4389,
+      "step": 2530
+    },
+    {
+      "epoch": 0.28121453679869357,
+      "grad_norm": 1.5936602354049683,
+      "learning_rate": 0.0002509708181061178,
+      "loss": 4.4478,
+      "step": 2540
+    },
+    {
+      "epoch": 0.2823216806443577,
+      "grad_norm": 1.3346928358078003,
+      "learning_rate": 0.0002505756049976346,
+      "loss": 4.5374,
+      "step": 2550
+    },
+    {
+      "epoch": 0.2834288244900219,
+      "grad_norm": 1.373197317123413,
+      "learning_rate": 0.00025017911934218387,
+      "loss": 4.4188,
+      "step": 2560
+    },
+    {
+      "epoch": 0.284535968335686,
+      "grad_norm": 1.4404033422470093,
+      "learning_rate": 0.000249781366156356,
+      "loss": 4.3963,
+      "step": 2570
+    },
+    {
+      "epoch": 0.2856431121813502,
+      "grad_norm": 1.3246028423309326,
+      "learning_rate": 0.0002493823504727791,
+      "loss": 4.464,
+      "step": 2580
+    },
+    {
+      "epoch": 0.2867502560270143,
+      "grad_norm": 1.6210161447525024,
+      "learning_rate": 0.00024898207734005506,
+      "loss": 4.5101,
+      "step": 2590
+    },
+    {
+      "epoch": 0.28785739987267844,
+      "grad_norm": 1.3224226236343384,
+      "learning_rate": 0.00024858055182269586,
+      "loss": 4.426,
+      "step": 2600
+    },
+    {
+      "epoch": 0.2889645437183426,
+      "grad_norm": 1.6813453435897827,
+      "learning_rate": 0.0002481777790010596,
+      "loss": 4.3848,
+      "step": 2610
+    },
+    {
+      "epoch": 0.29007168756400675,
+      "grad_norm": 1.4079194068908691,
+      "learning_rate": 0.00024777376397128593,
+      "loss": 4.4846,
+      "step": 2620
+    },
+    {
+      "epoch": 0.2911788314096709,
+      "grad_norm": 1.4133639335632324,
+      "learning_rate": 0.00024736851184523176,
+      "loss": 4.4302,
+      "step": 2630
+    },
+    {
+      "epoch": 0.29228597525533506,
+      "grad_norm": 1.4133745431900024,
+      "learning_rate": 0.0002469620277504066,
+      "loss": 4.4695,
+      "step": 2640
+    },
+    {
+      "epoch": 0.2933931191009992,
+      "grad_norm": 1.5624943971633911,
+      "learning_rate": 0.00024655431682990743,
+      "loss": 4.3333,
+      "step": 2650
+    },
+    {
+      "epoch": 0.29450026294666337,
+      "grad_norm": 1.288777232170105,
+      "learning_rate": 0.000246145384242354,
+      "loss": 4.4505,
+      "step": 2660
+    },
+    {
+      "epoch": 0.2956074067923275,
+      "grad_norm": 1.541371464729309,
+      "learning_rate": 0.0002457352351618233,
+      "loss": 4.4481,
+      "step": 2670
+    },
+    {
+      "epoch": 0.2967145506379916,
+      "grad_norm": 1.3446531295776367,
+      "learning_rate": 0.00024532387477778415,
+      "loss": 4.5125,
+      "step": 2680
+    },
+    {
+      "epoch": 0.2978216944836558,
+      "grad_norm": 1.3117250204086304,
+      "learning_rate": 0.00024491130829503165,
+      "loss": 4.4796,
+      "step": 2690
+    },
+    {
+      "epoch": 0.29892883832931993,
+      "grad_norm": 1.5482264757156372,
+      "learning_rate": 0.0002444975409336211,
+      "loss": 4.4677,
+      "step": 2700
+    },
+    {
+      "epoch": 0.30003598217498406,
+      "grad_norm": 1.5080773830413818,
+      "learning_rate": 0.00024408257792880216,
+      "loss": 4.4094,
+      "step": 2710
+    },
+    {
+      "epoch": 0.3002574109441169,
+      "eval_loss": 4.873768329620361,
+      "eval_runtime": 46.6367,
+      "eval_samples_per_second": 1072.118,
+      "eval_steps_per_second": 134.015,
+      "step": 2712
+    },
+    {
+      "epoch": 0.30114312602064824,
+      "grad_norm": 1.5396296977996826,
+      "learning_rate": 0.00024366642453095254,
+      "loss": 4.5269,
+      "step": 2720
+    },
+    {
+      "epoch": 0.30225026986631237,
+      "grad_norm": 1.392967700958252,
+      "learning_rate": 0.00024324908600551162,
+      "loss": 4.4326,
+      "step": 2730
+    },
+    {
+      "epoch": 0.30335741371197655,
+      "grad_norm": 1.629118800163269,
+      "learning_rate": 0.00024283056763291355,
+      "loss": 4.4372,
+      "step": 2740
+    },
+    {
+      "epoch": 0.3044645575576407,
+      "grad_norm": 1.439494252204895,
+      "learning_rate": 0.00024241087470852095,
+      "loss": 4.3365,
+      "step": 2750
+    },
+    {
+      "epoch": 0.3055717014033048,
+      "grad_norm": 1.405794382095337,
+      "learning_rate": 0.00024199001254255744,
+      "loss": 4.5201,
+      "step": 2760
+    },
+    {
+      "epoch": 0.306678845248969,
+      "grad_norm": 1.4038450717926025,
+      "learning_rate": 0.00024156798646004065,
+      "loss": 4.4196,
+      "step": 2770
+    },
+    {
+      "epoch": 0.3077859890946331,
+      "grad_norm": 1.538292646408081,
+      "learning_rate": 0.00024114480180071495,
+      "loss": 4.4115,
+      "step": 2780
+    },
+    {
+      "epoch": 0.30889313294029724,
+      "grad_norm": 1.4990549087524414,
+      "learning_rate": 0.00024072046391898362,
+      "loss": 4.495,
+      "step": 2790
+    },
+    {
+      "epoch": 0.3100002767859614,
+      "grad_norm": 1.513628363609314,
+      "learning_rate": 0.0002402949781838413,
+      "loss": 4.4708,
+      "step": 2800
+    },
+    {
+      "epoch": 0.31110742063162555,
+      "grad_norm": 1.502672553062439,
+      "learning_rate": 0.000239868349978806,
+      "loss": 4.5048,
+      "step": 2810
+    },
+    {
+      "epoch": 0.31221456447728974,
+      "grad_norm": 1.4603543281555176,
+      "learning_rate": 0.00023944058470185106,
+      "loss": 4.424,
+      "step": 2820
+    },
+    {
+      "epoch": 0.31332170832295386,
+      "grad_norm": 1.561678409576416,
+      "learning_rate": 0.0002390116877653367,
+      "loss": 4.4846,
+      "step": 2830
+    },
+    {
+      "epoch": 0.314428852168618,
+      "grad_norm": 1.5144026279449463,
+      "learning_rate": 0.0002385816645959416,
+      "loss": 4.4003,
+      "step": 2840
+    },
+    {
+      "epoch": 0.31553599601428217,
+      "grad_norm": 1.4243658781051636,
+      "learning_rate": 0.00023815052063459437,
+      "loss": 4.4371,
+      "step": 2850
+    },
+    {
+      "epoch": 0.3166431398599463,
+      "grad_norm": 1.3062716722488403,
+      "learning_rate": 0.0002377182613364045,
+      "loss": 4.4265,
+      "step": 2860
+    },
+    {
+      "epoch": 0.3177502837056104,
+      "grad_norm": 1.5508390665054321,
+      "learning_rate": 0.0002372848921705934,
+      "loss": 4.3411,
+      "step": 2870
+    },
+    {
+      "epoch": 0.3188574275512746,
+      "grad_norm": 1.3499903678894043,
+      "learning_rate": 0.00023685041862042547,
+      "loss": 4.5022,
+      "step": 2880
+    },
+    {
+      "epoch": 0.31996457139693874,
+      "grad_norm": 1.3883624076843262,
+      "learning_rate": 0.00023641484618313818,
+      "loss": 4.373,
+      "step": 2890
+    },
+    {
+      "epoch": 0.3210717152426029,
+      "grad_norm": 1.437251329421997,
+      "learning_rate": 0.00023597818036987306,
+      "loss": 4.475,
+      "step": 2900
+    },
+    {
+      "epoch": 0.32217885908826704,
+      "grad_norm": 1.596030592918396,
+      "learning_rate": 0.00023554042670560554,
+      "loss": 4.5456,
+      "step": 2910
+    },
+    {
+      "epoch": 0.32328600293393117,
+      "grad_norm": 1.5421555042266846,
+      "learning_rate": 0.00023510159072907545,
+      "loss": 4.3911,
+      "step": 2920
+    },
+    {
+      "epoch": 0.32439314677959535,
+      "grad_norm": 1.3547664880752563,
+      "learning_rate": 0.0002346616779927165,
+      "loss": 4.46,
+      "step": 2930
+    },
+    {
+      "epoch": 0.3255002906252595,
+      "grad_norm": 1.508280634880066,
+      "learning_rate": 0.00023422069406258647,
+      "loss": 4.4332,
+      "step": 2940
+    },
+    {
+      "epoch": 0.3266074344709236,
+      "grad_norm": 1.4340486526489258,
+      "learning_rate": 0.00023377864451829642,
+      "loss": 4.3638,
+      "step": 2950
+    },
+    {
+      "epoch": 0.3277145783165878,
+      "grad_norm": 1.4415581226348877,
+      "learning_rate": 0.0002333355349529403,
+      "loss": 4.3998,
+      "step": 2960
+    },
+    {
+      "epoch": 0.3288217221622519,
+      "grad_norm": 1.561346411705017,
+      "learning_rate": 0.00023289137097302415,
+      "loss": 4.4799,
+      "step": 2970
+    },
+    {
+      "epoch": 0.3299288660079161,
+      "grad_norm": 1.5376802682876587,
+      "learning_rate": 0.00023244615819839506,
+      "loss": 4.4001,
+      "step": 2980
+    },
+    {
+      "epoch": 0.3310360098535802,
+      "grad_norm": 1.4146666526794434,
+      "learning_rate": 0.0002319999022621702,
+      "loss": 4.3747,
+      "step": 2990
+    },
+    {
+      "epoch": 0.33214315369924435,
+      "grad_norm": 1.449980616569519,
+      "learning_rate": 0.00023155260881066548,
+      "loss": 4.452,
+      "step": 3000
+    },
+    {
+      "epoch": 0.33325029754490854,
+      "grad_norm": 1.6006379127502441,
+      "learning_rate": 0.0002311042835033241,
+      "loss": 4.3849,
+      "step": 3010
+    },
+    {
+      "epoch": 0.33435744139057266,
+      "grad_norm": 1.5453193187713623,
+      "learning_rate": 0.00023065493201264504,
+      "loss": 4.3975,
+      "step": 3020
+    },
+    {
+      "epoch": 0.33546458523623685,
+      "grad_norm": 1.427522897720337,
+      "learning_rate": 0.0002302045600241111,
+      "loss": 4.5111,
+      "step": 3030
+    },
+    {
+      "epoch": 0.33657172908190097,
+      "grad_norm": 1.5262653827667236,
+      "learning_rate": 0.00022975317323611707,
+      "loss": 4.4219,
+      "step": 3040
+    },
+    {
+      "epoch": 0.3376788729275651,
+      "grad_norm": 1.6192231178283691,
+      "learning_rate": 0.00022930077735989777,
+      "loss": 4.4299,
+      "step": 3050
+    },
+    {
+      "epoch": 0.3387860167732293,
+      "grad_norm": 1.4724215269088745,
+      "learning_rate": 0.0002288473781194555,
+      "loss": 4.3736,
+      "step": 3060
+    },
+    {
+      "epoch": 0.3398931606188934,
+      "grad_norm": 1.611179232597351,
+      "learning_rate": 0.0002283929812514879,
+      "loss": 4.4629,
+      "step": 3070
+    },
+    {
+      "epoch": 0.34100030446455754,
+      "grad_norm": 1.5913844108581543,
+      "learning_rate": 0.00022793759250531502,
+      "loss": 4.2893,
+      "step": 3080
+    },
+    {
+      "epoch": 0.3421074483102217,
+      "grad_norm": 1.640321135520935,
+      "learning_rate": 0.00022748121764280707,
+      "loss": 4.3281,
+      "step": 3090
+    },
+    {
+      "epoch": 0.34321459215588584,
+      "grad_norm": 1.3917701244354248,
+      "learning_rate": 0.000227023862438311,
+      "loss": 4.4247,
+      "step": 3100
+    },
+    {
+      "epoch": 0.34432173600155,
+      "grad_norm": 1.4985103607177734,
+      "learning_rate": 0.0002265655326785778,
+      "loss": 4.477,
+      "step": 3110
+    },
+    {
+      "epoch": 0.34542887984721415,
+      "grad_norm": 1.4070855379104614,
+      "learning_rate": 0.0002261062341626892,
+      "loss": 4.3453,
+      "step": 3120
+    },
+    {
+      "epoch": 0.3465360236928783,
+      "grad_norm": 1.44731867313385,
+      "learning_rate": 0.00022564597270198408,
+      "loss": 4.3722,
+      "step": 3130
+    },
+    {
+      "epoch": 0.34764316753854246,
+      "grad_norm": 1.5078603029251099,
+      "learning_rate": 0.0002251847541199852,
+      "loss": 4.3694,
+      "step": 3140
+    },
+    {
+      "epoch": 0.3487503113842066,
+      "grad_norm": 1.5967936515808105,
+      "learning_rate": 0.00022472258425232554,
+      "loss": 4.3685,
+      "step": 3150
+    },
+    {
+      "epoch": 0.3498574552298707,
+      "grad_norm": 1.3859797716140747,
+      "learning_rate": 0.00022425946894667417,
+      "loss": 4.5154,
+      "step": 3160
+    },
+    {
+      "epoch": 0.3509645990755349,
+      "grad_norm": 1.6232459545135498,
+      "learning_rate": 0.0002237954140626626,
+      "loss": 4.3742,
+      "step": 3170
+    },
+    {
+      "epoch": 0.352071742921199,
+      "grad_norm": 1.5523372888565063,
+      "learning_rate": 0.0002233304254718104,
+      "loss": 4.3129,
+      "step": 3180
+    },
+    {
+      "epoch": 0.3531788867668632,
+      "grad_norm": 1.3990780115127563,
+      "learning_rate": 0.000222864509057451,
+      "loss": 4.3861,
+      "step": 3190
+    },
+    {
+      "epoch": 0.35428603061252734,
+      "grad_norm": 1.5679517984390259,
+      "learning_rate": 0.00022239767071465728,
+      "loss": 4.4173,
+      "step": 3200
+    },
+    {
+      "epoch": 0.35539317445819146,
+      "grad_norm": 1.6319177150726318,
+      "learning_rate": 0.00022192991635016697,
+      "loss": 4.3712,
+      "step": 3210
+    },
+    {
+      "epoch": 0.35650031830385565,
+      "grad_norm": 1.5475807189941406,
+      "learning_rate": 0.00022146125188230782,
+      "loss": 4.4611,
+      "step": 3220
+    },
+    {
+      "epoch": 0.3576074621495198,
+      "grad_norm": 1.6476644277572632,
+      "learning_rate": 0.00022099168324092288,
+      "loss": 4.3423,
+      "step": 3230
+    },
+    {
+      "epoch": 0.3587146059951839,
+      "grad_norm": 1.6143760681152344,
+      "learning_rate": 0.00022052121636729525,
+      "loss": 4.3829,
+      "step": 3240
+    },
+    {
+      "epoch": 0.3598217498408481,
+      "grad_norm": 1.613155484199524,
+      "learning_rate": 0.0002200498572140732,
+      "loss": 4.2611,
+      "step": 3250
+    },
+    {
+      "epoch": 0.3609288936865122,
+      "grad_norm": 1.6002566814422607,
+      "learning_rate": 0.00021957761174519458,
+      "loss": 4.3073,
+      "step": 3260
+    },
+    {
+      "epoch": 0.3620360375321764,
+      "grad_norm": 1.842801570892334,
+      "learning_rate": 0.00021910448593581155,
+      "loss": 4.3887,
+      "step": 3270
+    },
+    {
+      "epoch": 0.3631431813778405,
+      "grad_norm": 1.4907143115997314,
+      "learning_rate": 0.0002186304857722148,
+      "loss": 4.3579,
+      "step": 3280
+    },
+    {
+      "epoch": 0.36425032522350465,
+      "grad_norm": 1.5803886651992798,
+      "learning_rate": 0.00021815561725175802,
+      "loss": 4.3935,
+      "step": 3290
+    },
+    {
+      "epoch": 0.3653574690691688,
+      "grad_norm": 1.6706593036651611,
+      "learning_rate": 0.00021767988638278182,
+      "loss": 4.4003,
+      "step": 3300
+    },
+    {
+      "epoch": 0.36646461291483295,
+      "grad_norm": 1.7942376136779785,
+      "learning_rate": 0.00021720329918453786,
+      "loss": 4.4249,
+      "step": 3310
+    },
+    {
+      "epoch": 0.3675717567604971,
+      "grad_norm": 1.4756616353988647,
+      "learning_rate": 0.0002167258616871126,
+      "loss": 4.3849,
+      "step": 3320
+    },
+    {
+      "epoch": 0.36867890060616126,
+      "grad_norm": 1.4991201162338257,
+      "learning_rate": 0.00021624757993135107,
+      "loss": 4.3422,
+      "step": 3330
+    },
+    {
+      "epoch": 0.3697860444518254,
+      "grad_norm": 1.7558096647262573,
+      "learning_rate": 0.00021576845996878024,
+      "loss": 4.2834,
+      "step": 3340
+    },
+    {
+      "epoch": 0.3708931882974896,
+      "grad_norm": 1.6017481088638306,
+      "learning_rate": 0.00021528850786153283,
+      "loss": 4.343,
+      "step": 3350
+    },
+    {
+      "epoch": 0.3720003321431537,
+      "grad_norm": 1.7434948682785034,
+      "learning_rate": 0.00021480772968227025,
+      "loss": 4.295,
+      "step": 3360
+    },
+    {
+      "epoch": 0.3731074759888178,
+      "grad_norm": 1.5920963287353516,
+      "learning_rate": 0.00021432613151410593,
+      "loss": 4.3463,
+      "step": 3370
+    },
+    {
+      "epoch": 0.374214619834482,
+      "grad_norm": 1.5161998271942139,
+      "learning_rate": 0.00021384371945052836,
+      "loss": 4.2752,
+      "step": 3380
+    },
+    {
+      "epoch": 0.37532176368014614,
+      "grad_norm": 1.4809553623199463,
+      "learning_rate": 0.00021336049959532388,
+      "loss": 4.3035,
+      "step": 3390
+    },
+    {
+      "epoch": 0.37642890752581026,
+      "grad_norm": 1.578795075416565,
+      "learning_rate": 0.0002128764780624996,
+      "loss": 4.3403,
+      "step": 3400
+    },
+    {
+      "epoch": 0.37753605137147445,
+      "grad_norm": 1.590668797492981,
+      "learning_rate": 0.00021239166097620596,
+      "loss": 4.2579,
+      "step": 3410
+    },
+    {
+      "epoch": 0.3786431952171386,
+      "grad_norm": 1.6730364561080933,
+      "learning_rate": 0.00021190605447065917,
+      "loss": 4.3133,
+      "step": 3420
+    },
+    {
+      "epoch": 0.37975033906280276,
+      "grad_norm": 1.5798331499099731,
+      "learning_rate": 0.00021141966469006386,
+      "loss": 4.3727,
+      "step": 3430
+    },
+    {
+      "epoch": 0.3808574829084669,
+      "grad_norm": 1.5975507497787476,
+      "learning_rate": 0.00021093249778853494,
+      "loss": 4.3359,
+      "step": 3440
+    },
+    {
+      "epoch": 0.381964626754131,
+      "grad_norm": 1.4481183290481567,
+      "learning_rate": 0.00021044455993002003,
+      "loss": 4.3113,
+      "step": 3450
+    },
+    {
+      "epoch": 0.3830717705997952,
+      "grad_norm": 1.5482516288757324,
+      "learning_rate": 0.0002099558572882215,
+      "loss": 4.3878,
+      "step": 3460
+    },
+    {
+      "epoch": 0.3841789144454593,
+      "grad_norm": 1.4012142419815063,
+      "learning_rate": 0.00020946639604651815,
+      "loss": 4.2864,
+      "step": 3470
+    },
+    {
+      "epoch": 0.3852860582911235,
+      "grad_norm": 1.7526214122772217,
+      "learning_rate": 0.00020897618239788704,
+      "loss": 4.3149,
+      "step": 3480
+    },
+    {
+      "epoch": 0.38639320213678763,
+      "grad_norm": 1.5670076608657837,
+      "learning_rate": 0.00020848522254482527,
+      "loss": 4.3492,
+      "step": 3490
+    },
+    {
+      "epoch": 0.38750034598245175,
+      "grad_norm": 1.4638546705245972,
+      "learning_rate": 0.00020799352269927122,
+      "loss": 4.402,
+      "step": 3500
+    },
+    {
+      "epoch": 0.38860748982811594,
+      "grad_norm": 1.8438469171524048,
+      "learning_rate": 0.0002075010890825263,
+      "loss": 4.3077,
+      "step": 3510
+    },
+    {
+      "epoch": 0.38971463367378006,
+      "grad_norm": 1.5217992067337036,
+      "learning_rate": 0.0002070079279251761,
+      "loss": 4.3279,
+      "step": 3520
+    },
+    {
+      "epoch": 0.3908217775194442,
+      "grad_norm": 1.732856035232544,
+      "learning_rate": 0.00020651404546701127,
+      "loss": 4.3337,
+      "step": 3530
+    },
+    {
+      "epoch": 0.3919289213651084,
+      "grad_norm": 1.74932861328125,
+      "learning_rate": 0.00020601944795694906,
+      "loss": 4.3282,
+      "step": 3540
+    },
+    {
+      "epoch": 0.3930360652107725,
+      "grad_norm": 1.4456238746643066,
+      "learning_rate": 0.000205524141652954,
+      "loss": 4.4165,
+      "step": 3550
+    },
+    {
+      "epoch": 0.3941432090564367,
+      "grad_norm": 1.6402347087860107,
+      "learning_rate": 0.00020502813282195846,
+      "loss": 4.4085,
+      "step": 3560
+    },
+    {
+      "epoch": 0.3952503529021008,
+      "grad_norm": 1.5870882272720337,
+      "learning_rate": 0.000204531427739784,
+      "loss": 4.2847,
+      "step": 3570
+    },
+    {
+      "epoch": 0.39635749674776494,
+      "grad_norm": 1.473127007484436,
+      "learning_rate": 0.0002040340326910614,
+      "loss": 4.2992,
+      "step": 3580
+    },
+    {
+      "epoch": 0.3974646405934291,
+      "grad_norm": 1.6806960105895996,
+      "learning_rate": 0.0002035359539691514,
+      "loss": 4.2847,
+      "step": 3590
+    },
+    {
+      "epoch": 0.39857178443909325,
+      "grad_norm": 1.5456218719482422,
+      "learning_rate": 0.0002030371978760649,
+      "loss": 4.2877,
+      "step": 3600
+    },
+    {
+      "epoch": 0.3996789282847574,
+      "grad_norm": 1.7811386585235596,
+      "learning_rate": 0.00020253777072238349,
+      "loss": 4.3922,
+      "step": 3610
+    },
+    {
+      "epoch": 0.4003432145921559,
+      "eval_loss": 4.767195701599121,
+      "eval_runtime": 46.5629,
+      "eval_samples_per_second": 1073.815,
+      "eval_steps_per_second": 134.227,
+      "step": 3616
+    },
+    {
+      "epoch": 0.40078607213042156,
+      "grad_norm": 1.5370206832885742,
+      "learning_rate": 0.0002020376788271793,
+      "loss": 4.3435,
+      "step": 3620
+    },
+    {
+      "epoch": 0.4018932159760857,
+      "grad_norm": 1.6143906116485596,
+      "learning_rate": 0.00020153692851793535,
+      "loss": 4.2977,
+      "step": 3630
+    },
+    {
+      "epoch": 0.40300035982174987,
+      "grad_norm": 1.6039817333221436,
+      "learning_rate": 0.0002010355261304652,
+      "loss": 4.2706,
+      "step": 3640
+    },
+    {
+      "epoch": 0.404107503667414,
+      "grad_norm": 1.4683738946914673,
+      "learning_rate": 0.00020053347800883298,
+      "loss": 4.208,
+      "step": 3650
+    },
+    {
+      "epoch": 0.4052146475130781,
+      "grad_norm": 1.4393600225448608,
+      "learning_rate": 0.0002000307905052731,
+      "loss": 4.3913,
+      "step": 3660
+    },
+    {
+      "epoch": 0.4063217913587423,
+      "grad_norm": 1.5226246118545532,
+      "learning_rate": 0.0001995274699801098,
+      "loss": 4.3221,
+      "step": 3670
+    },
+    {
+      "epoch": 0.40742893520440643,
+      "grad_norm": 1.9454008340835571,
+      "learning_rate": 0.00019902352280167665,
+      "loss": 4.327,
+      "step": 3680
+    },
+    {
+      "epoch": 0.40853607905007056,
+      "grad_norm": 1.5962251424789429,
+      "learning_rate": 0.0001985189553462361,
+      "loss": 4.2941,
+      "step": 3690
+    },
+    {
+      "epoch": 0.40964322289573474,
+      "grad_norm": 1.6823145151138306,
+      "learning_rate": 0.00019801377399789882,
+      "loss": 4.3054,
+      "step": 3700
+    },
+    {
+      "epoch": 0.41075036674139886,
+      "grad_norm": 1.4310302734375,
+      "learning_rate": 0.00019750798514854274,
+      "loss": 4.3727,
+      "step": 3710
+    },
+    {
+      "epoch": 0.41185751058706305,
+      "grad_norm": 1.5457884073257446,
+      "learning_rate": 0.0001970015951977323,
+      "loss": 4.3687,
+      "step": 3720
+    },
+    {
+      "epoch": 0.4129646544327272,
+      "grad_norm": 1.6520789861679077,
+      "learning_rate": 0.0001964946105526375,
+      "loss": 4.3746,
+      "step": 3730
+    },
+    {
+      "epoch": 0.4140717982783913,
+      "grad_norm": 1.554999828338623,
+      "learning_rate": 0.00019598703762795273,
+      "loss": 4.3462,
+      "step": 3740
+    },
+    {
+      "epoch": 0.4151789421240555,
+      "grad_norm": 1.7624000310897827,
+      "learning_rate": 0.0001954788828458157,
+      "loss": 4.344,
+      "step": 3750
+    },
+    {
+      "epoch": 0.4162860859697196,
+      "grad_norm": 1.7389049530029297,
+      "learning_rate": 0.00019497015263572617,
+      "loss": 4.266,
+      "step": 3760
+    },
+    {
+      "epoch": 0.41739322981538374,
+      "grad_norm": 1.7113863229751587,
+      "learning_rate": 0.00019446085343446462,
+      "loss": 4.3676,
+      "step": 3770
+    },
+    {
+      "epoch": 0.4185003736610479,
+      "grad_norm": 1.5145055055618286,
+      "learning_rate": 0.00019395099168601065,
+      "loss": 4.3052,
+      "step": 3780
+    },
+    {
+      "epoch": 0.41960751750671205,
+      "grad_norm": 1.5164849758148193,
+      "learning_rate": 0.00019344057384146167,
+      "loss": 4.3953,
+      "step": 3790
+    },
+    {
+      "epoch": 0.42071466135237623,
+      "grad_norm": 1.7207586765289307,
+      "learning_rate": 0.0001929296063589512,
+      "loss": 4.3157,
+      "step": 3800
+    },
+    {
+      "epoch": 0.42182180519804036,
+      "grad_norm": 1.5681637525558472,
+      "learning_rate": 0.00019241809570356698,
+      "loss": 4.2953,
+      "step": 3810
+    },
+    {
+      "epoch": 0.4229289490437045,
+      "grad_norm": 1.6934231519699097,
+      "learning_rate": 0.00019190604834726943,
+      "loss": 4.3274,
+      "step": 3820
+    },
+    {
+      "epoch": 0.42403609288936867,
+      "grad_norm": 1.6220754384994507,
+      "learning_rate": 0.00019139347076880972,
+      "loss": 4.3144,
+      "step": 3830
+    },
+    {
+      "epoch": 0.4251432367350328,
+      "grad_norm": 1.683106780052185,
+      "learning_rate": 0.00019088036945364754,
+      "loss": 4.2909,
+      "step": 3840
+    },
+    {
+      "epoch": 0.426250380580697,
+      "grad_norm": 1.5829964876174927,
+      "learning_rate": 0.00019036675089386944,
+      "loss": 4.322,
+      "step": 3850
+    },
+    {
+      "epoch": 0.4273575244263611,
+      "grad_norm": 1.6295596361160278,
+      "learning_rate": 0.00018985262158810636,
+      "loss": 4.2721,
+      "step": 3860
+    },
+    {
+      "epoch": 0.42846466827202523,
+      "grad_norm": 1.6584826707839966,
+      "learning_rate": 0.00018933798804145148,
+      "loss": 4.3177,
+      "step": 3870
+    },
+    {
+      "epoch": 0.4295718121176894,
+      "grad_norm": 1.5818955898284912,
+      "learning_rate": 0.0001888228567653781,
+      "loss": 4.3083,
+      "step": 3880
+    },
+    {
+      "epoch": 0.43067895596335354,
+      "grad_norm": 1.7724536657333374,
+      "learning_rate": 0.000188307234277657,
+      "loss": 4.2494,
+      "step": 3890
+    },
+    {
+      "epoch": 0.43178609980901766,
+      "grad_norm": 1.8732531070709229,
+      "learning_rate": 0.00018779112710227416,
+      "loss": 4.2508,
+      "step": 3900
+    },
+    {
+      "epoch": 0.43289324365468185,
+      "grad_norm": 1.5723505020141602,
+      "learning_rate": 0.00018727454176934808,
+      "loss": 4.3309,
+      "step": 3910
+    },
+    {
+      "epoch": 0.434000387500346,
+      "grad_norm": 1.5796666145324707,
+      "learning_rate": 0.00018675748481504728,
+      "loss": 4.2512,
+      "step": 3920
+    },
+    {
+      "epoch": 0.43510753134601016,
+      "grad_norm": 1.4606853723526,
+      "learning_rate": 0.0001862399627815074,
+      "loss": 4.1902,
+      "step": 3930
+    },
+    {
+      "epoch": 0.4362146751916743,
+      "grad_norm": 1.7595901489257812,
+      "learning_rate": 0.00018572198221674868,
+      "loss": 4.2583,
+      "step": 3940
+    },
+    {
+      "epoch": 0.4373218190373384,
+      "grad_norm": 1.5277389287948608,
+      "learning_rate": 0.00018520354967459303,
+      "loss": 4.3261,
+      "step": 3950
+    },
+    {
+      "epoch": 0.4384289628830026,
+      "grad_norm": 1.672125220298767,
+      "learning_rate": 0.00018468467171458094,
+      "loss": 4.2439,
+      "step": 3960
+    },
+    {
+      "epoch": 0.4395361067286667,
+      "grad_norm": 1.8138833045959473,
+      "learning_rate": 0.00018416535490188866,
+      "loss": 4.2202,
+      "step": 3970
+    },
+    {
+      "epoch": 0.44064325057433085,
+      "grad_norm": 1.602288842201233,
+      "learning_rate": 0.0001836456058072452,
+      "loss": 4.2036,
+      "step": 3980
+    },
+    {
+      "epoch": 0.44175039441999503,
+      "grad_norm": 1.7370867729187012,
+      "learning_rate": 0.00018312543100684888,
+      "loss": 4.2621,
+      "step": 3990
+    },
+    {
+      "epoch": 0.44285753826565916,
+      "grad_norm": 1.8132164478302002,
+      "learning_rate": 0.00018260483708228438,
+      "loss": 4.2258,
+      "step": 4000
+    },
+    {
+      "epoch": 0.44396468211132334,
+      "grad_norm": 1.5764164924621582,
+      "learning_rate": 0.00018208383062043957,
+      "loss": 4.279,
+      "step": 4010
+    },
+    {
+      "epoch": 0.44507182595698747,
+      "grad_norm": 1.784719467163086,
+      "learning_rate": 0.0001815624182134218,
+      "loss": 4.2154,
+      "step": 4020
+    },
+    {
+      "epoch": 0.4461789698026516,
+      "grad_norm": 1.8702234029769897,
+      "learning_rate": 0.00018104060645847476,
+      "loss": 4.2021,
+      "step": 4030
+    },
+    {
+      "epoch": 0.4472861136483158,
+      "grad_norm": 1.6183464527130127,
+      "learning_rate": 0.00018051840195789506,
+      "loss": 4.262,
+      "step": 4040
+    },
+    {
+      "epoch": 0.4483932574939799,
+      "grad_norm": 1.5148000717163086,
+      "learning_rate": 0.00017999581131894843,
+      "loss": 4.293,
+      "step": 4050
+    },
+    {
+      "epoch": 0.44950040133964403,
+      "grad_norm": 1.7183406352996826,
+      "learning_rate": 0.00017947284115378626,
+      "loss": 4.2857,
+      "step": 4060
+    },
+    {
+      "epoch": 0.4506075451853082,
+      "grad_norm": 1.5063244104385376,
+      "learning_rate": 0.00017894949807936212,
+      "loss": 4.2994,
+      "step": 4070
+    },
+    {
+      "epoch": 0.45171468903097234,
+      "grad_norm": 1.578837275505066,
+      "learning_rate": 0.0001784257887173477,
+      "loss": 4.3426,
+      "step": 4080
+    },
+    {
+      "epoch": 0.4528218328766365,
+      "grad_norm": 1.8355525732040405,
+      "learning_rate": 0.0001779017196940493,
+      "loss": 4.3082,
+      "step": 4090
+    },
+    {
+      "epoch": 0.45392897672230065,
+      "grad_norm": 1.7444182634353638,
+      "learning_rate": 0.00017737729764032384,
+      "loss": 4.1786,
+      "step": 4100
+    },
+    {
+      "epoch": 0.4550361205679648,
+      "grad_norm": 1.6076940298080444,
+      "learning_rate": 0.00017685252919149493,
+      "loss": 4.2155,
+      "step": 4110
+    },
+    {
+      "epoch": 0.45614326441362896,
+      "grad_norm": 1.8320263624191284,
+      "learning_rate": 0.0001763274209872691,
+      "loss": 4.2537,
+      "step": 4120
+    },
+    {
+      "epoch": 0.4572504082592931,
+      "grad_norm": 1.5995029211044312,
+      "learning_rate": 0.00017580197967165167,
+      "loss": 4.2097,
+      "step": 4130
+    },
+    {
+      "epoch": 0.4583575521049572,
+      "grad_norm": 1.6503360271453857,
+      "learning_rate": 0.0001752762118928627,
+      "loss": 4.2117,
+      "step": 4140
+    },
+    {
+      "epoch": 0.4594646959506214,
+      "grad_norm": 1.750923752784729,
+      "learning_rate": 0.00017475012430325285,
+      "loss": 4.2316,
+      "step": 4150
+    },
+    {
+      "epoch": 0.4605718397962855,
+      "grad_norm": 1.576587438583374,
+      "learning_rate": 0.00017422372355921923,
+      "loss": 4.2488,
+      "step": 4160
+    },
+    {
+      "epoch": 0.4616789836419497,
+      "grad_norm": 1.720887303352356,
+      "learning_rate": 0.00017369701632112119,
+      "loss": 4.2264,
+      "step": 4170
+    },
+    {
+      "epoch": 0.46278612748761383,
+      "grad_norm": 1.7187278270721436,
+      "learning_rate": 0.00017317000925319597,
+      "loss": 4.2634,
+      "step": 4180
+    },
+    {
+      "epoch": 0.46389327133327796,
+      "grad_norm": 1.6399569511413574,
+      "learning_rate": 0.00017264270902347454,
+      "loss": 4.3008,
+      "step": 4190
+    },
+    {
+      "epoch": 0.46500041517894214,
+      "grad_norm": 1.866749882698059,
+      "learning_rate": 0.00017211512230369715,
+      "loss": 4.1805,
+      "step": 4200
+    },
+    {
+      "epoch": 0.46610755902460627,
+      "grad_norm": 1.6939700841903687,
+      "learning_rate": 0.00017158725576922888,
+      "loss": 4.2157,
+      "step": 4210
+    },
+    {
+      "epoch": 0.4672147028702704,
+      "grad_norm": 1.552985429763794,
+      "learning_rate": 0.0001710591160989752,
+      "loss": 4.245,
+      "step": 4220
+    },
+    {
+      "epoch": 0.4683218467159346,
+      "grad_norm": 1.7612532377243042,
+      "learning_rate": 0.00017053070997529737,
+      "loss": 4.2766,
+      "step": 4230
+    },
+    {
+      "epoch": 0.4694289905615987,
+      "grad_norm": 1.6406209468841553,
+      "learning_rate": 0.00017000204408392815,
+      "loss": 4.2408,
+      "step": 4240
+    },
+    {
+      "epoch": 0.4705361344072629,
+      "grad_norm": 1.714015007019043,
+      "learning_rate": 0.00016947312511388695,
+      "loss": 4.2397,
+      "step": 4250
+    },
+    {
+      "epoch": 0.471643278252927,
+      "grad_norm": 1.6452609300613403,
+      "learning_rate": 0.00016894395975739534,
+      "loss": 4.146,
+      "step": 4260
+    },
+    {
+      "epoch": 0.47275042209859114,
+      "grad_norm": 1.5308531522750854,
+      "learning_rate": 0.00016841455470979235,
+      "loss": 4.28,
+      "step": 4270
+    },
+    {
+      "epoch": 0.4738575659442553,
+      "grad_norm": 1.6363145112991333,
+      "learning_rate": 0.00016788491666944966,
+      "loss": 4.1804,
+      "step": 4280
+    },
+    {
+      "epoch": 0.47496470978991945,
+      "grad_norm": 1.776734471321106,
+      "learning_rate": 0.00016735505233768695,
+      "loss": 4.2542,
+      "step": 4290
+    },
+    {
+      "epoch": 0.47607185363558363,
+      "grad_norm": 1.6798444986343384,
+      "learning_rate": 0.0001668249684186871,
+      "loss": 4.1954,
+      "step": 4300
+    },
+    {
+      "epoch": 0.47717899748124776,
+      "grad_norm": 1.6321518421173096,
+      "learning_rate": 0.00016629467161941133,
+      "loss": 4.2106,
+      "step": 4310
+    },
+    {
+      "epoch": 0.4782861413269119,
+      "grad_norm": 1.8878740072250366,
+      "learning_rate": 0.00016576416864951443,
+      "loss": 4.2389,
+      "step": 4320
+    },
+    {
+      "epoch": 0.47939328517257607,
+      "grad_norm": 1.6510679721832275,
+      "learning_rate": 0.00016523346622125966,
+      "loss": 4.2166,
+      "step": 4330
+    },
+    {
+      "epoch": 0.4805004290182402,
+      "grad_norm": 1.6507411003112793,
+      "learning_rate": 0.0001647025710494341,
+      "loss": 4.1962,
+      "step": 4340
+    },
+    {
+      "epoch": 0.4816075728639043,
+      "grad_norm": 1.661056399345398,
+      "learning_rate": 0.0001641714898512634,
+      "loss": 4.0714,
+      "step": 4350
+    },
+    {
+      "epoch": 0.4827147167095685,
+      "grad_norm": 1.6582434177398682,
+      "learning_rate": 0.00016364022934632706,
+      "loss": 4.289,
+      "step": 4360
+    },
+    {
+      "epoch": 0.48382186055523263,
+      "grad_norm": 1.695462942123413,
+      "learning_rate": 0.0001631087962564732,
+      "loss": 4.2151,
+      "step": 4370
+    },
+    {
+      "epoch": 0.4849290044008968,
+      "grad_norm": 1.6066818237304688,
+      "learning_rate": 0.00016257719730573359,
+      "loss": 4.2921,
+      "step": 4380
+    },
+    {
+      "epoch": 0.48603614824656094,
+      "grad_norm": 1.6954655647277832,
+      "learning_rate": 0.0001620454392202387,
+      "loss": 4.2586,
+      "step": 4390
+    },
+    {
+      "epoch": 0.48714329209222507,
+      "grad_norm": 1.5290733575820923,
+      "learning_rate": 0.00016151352872813239,
+      "loss": 4.2494,
+      "step": 4400
+    },
+    {
+      "epoch": 0.48825043593788925,
+      "grad_norm": 1.599698543548584,
+      "learning_rate": 0.00016098147255948683,
+      "loss": 4.2374,
+      "step": 4410
+    },
+    {
+      "epoch": 0.4893575797835534,
+      "grad_norm": 1.6775895357131958,
+      "learning_rate": 0.0001604492774462174,
+      "loss": 4.2038,
+      "step": 4420
+    },
+    {
+      "epoch": 0.4904647236292175,
+      "grad_norm": 1.6222951412200928,
+      "learning_rate": 0.0001599169501219975,
+      "loss": 4.2022,
+      "step": 4430
+    },
+    {
+      "epoch": 0.4915718674748817,
+      "grad_norm": 1.5698493719100952,
+      "learning_rate": 0.0001593844973221734,
+      "loss": 4.2604,
+      "step": 4440
+    },
+    {
+      "epoch": 0.4926790113205458,
+      "grad_norm": 1.467875361442566,
+      "learning_rate": 0.00015885192578367894,
+      "loss": 4.2231,
+      "step": 4450
+    },
+    {
+      "epoch": 0.49378615516621,
+      "grad_norm": 1.6847503185272217,
+      "learning_rate": 0.00015831924224495028,
+      "loss": 4.1756,
+      "step": 4460
+    },
+    {
+      "epoch": 0.4948932990118741,
+      "grad_norm": 1.6515512466430664,
+      "learning_rate": 0.00015778645344584078,
+      "loss": 4.3254,
+      "step": 4470
+    },
+    {
+      "epoch": 0.49600044285753825,
+      "grad_norm": 1.5960655212402344,
+      "learning_rate": 0.00015725356612753541,
+      "loss": 4.2239,
+      "step": 4480
+    },
+    {
+      "epoch": 0.49710758670320243,
+      "grad_norm": 1.9087690114974976,
+      "learning_rate": 0.00015672058703246596,
+      "loss": 4.1995,
+      "step": 4490
+    },
+    {
+      "epoch": 0.49821473054886656,
+      "grad_norm": 1.7754182815551758,
+      "learning_rate": 0.00015618752290422512,
+      "loss": 4.2905,
+      "step": 4500
+    },
+    {
+      "epoch": 0.4993218743945307,
+      "grad_norm": 1.6775707006454468,
+      "learning_rate": 0.00015565438048748174,
+      "loss": 4.2429,
+      "step": 4510
+    },
+    {
+      "epoch": 0.5004290182401948,
+      "grad_norm": 1.5856534242630005,
+      "learning_rate": 0.00015512116652789516,
+      "loss": 4.1946,
+      "step": 4520
+    },
+    {
+      "epoch": 0.5004290182401948,
+      "eval_loss": 4.686659812927246,
+      "eval_runtime": 47.2831,
+      "eval_samples_per_second": 1057.461,
+      "eval_steps_per_second": 132.183,
+      "step": 4520
+    },
+    {
+      "epoch": 0.501536162085859,
+      "grad_norm": 1.962463617324829,
+      "learning_rate": 0.00015458788777202986,
+      "loss": 4.2388,
+      "step": 4530
+    },
+    {
+      "epoch": 0.5026433059315232,
+      "grad_norm": 1.8531359434127808,
+      "learning_rate": 0.00015405455096727014,
+      "loss": 4.2656,
+      "step": 4540
+    },
+    {
+      "epoch": 0.5037504497771873,
+      "grad_norm": 1.8523495197296143,
+      "learning_rate": 0.00015352116286173486,
+      "loss": 4.1859,
+      "step": 4550
+    },
+    {
+      "epoch": 0.5048575936228514,
+      "grad_norm": 1.6689633131027222,
+      "learning_rate": 0.00015298773020419192,
+      "loss": 4.2038,
+      "step": 4560
+    },
+    {
+      "epoch": 0.5059647374685156,
+      "grad_norm": 1.7041493654251099,
+      "learning_rate": 0.000152454259743973,
+      "loss": 4.1942,
+      "step": 4570
+    },
+    {
+      "epoch": 0.5070718813141798,
+      "grad_norm": 1.8809363842010498,
+      "learning_rate": 0.00015192075823088796,
+      "loss": 4.2054,
+      "step": 4580
+    },
+    {
+      "epoch": 0.5081790251598439,
+      "grad_norm": 1.6775367259979248,
+      "learning_rate": 0.00015138723241513968,
+      "loss": 4.2921,
+      "step": 4590
+    },
+    {
+      "epoch": 0.509286169005508,
+      "grad_norm": 1.8281633853912354,
+      "learning_rate": 0.0001508536890472384,
+      "loss": 4.2621,
+      "step": 4600
+    },
+    {
+      "epoch": 0.5103933128511722,
+      "grad_norm": 1.6567997932434082,
+      "learning_rate": 0.0001503201348779166,
+      "loss": 4.3198,
+      "step": 4610
+    },
+    {
+      "epoch": 0.5115004566968363,
+      "grad_norm": 1.784899353981018,
+      "learning_rate": 0.00014978657665804323,
+      "loss": 4.1681,
+      "step": 4620
+    },
+    {
+      "epoch": 0.5126076005425004,
+      "grad_norm": 1.7281806468963623,
+      "learning_rate": 0.00014925302113853866,
+      "loss": 4.1652,
+      "step": 4630
+    },
+    {
+      "epoch": 0.5137147443881647,
+      "grad_norm": 1.9660074710845947,
+      "learning_rate": 0.00014871947507028913,
+      "loss": 4.2037,
+      "step": 4640
+    },
+    {
+      "epoch": 0.5148218882338288,
+      "grad_norm": 1.7179691791534424,
+      "learning_rate": 0.00014818594520406105,
+      "loss": 4.1979,
+      "step": 4650
+    },
+    {
+      "epoch": 0.5159290320794929,
+      "grad_norm": 1.5476666688919067,
+      "learning_rate": 0.00014765243829041605,
+      "loss": 4.2037,
+      "step": 4660
+    },
+    {
+      "epoch": 0.517036175925157,
+      "grad_norm": 1.67542564868927,
+      "learning_rate": 0.00014711896107962534,
+      "loss": 4.1809,
+      "step": 4670
+    },
+    {
+      "epoch": 0.5181433197708212,
+      "grad_norm": 2.104140281677246,
+      "learning_rate": 0.00014658552032158412,
+      "loss": 4.1652,
+      "step": 4680
+    },
+    {
+      "epoch": 0.5192504636164854,
+      "grad_norm": 1.665068507194519,
+      "learning_rate": 0.00014605212276572665,
+      "loss": 4.3723,
+      "step": 4690
+    },
+    {
+      "epoch": 0.5203576074621495,
+      "grad_norm": 1.5425127744674683,
+      "learning_rate": 0.0001455187751609403,
+      "loss": 4.123,
+      "step": 4700
+    },
+    {
+      "epoch": 0.5214647513078137,
+      "grad_norm": 1.6816033124923706,
+      "learning_rate": 0.0001449854842554806,
+      "loss": 4.1974,
+      "step": 4710
+    },
+    {
+      "epoch": 0.5225718951534778,
+      "grad_norm": 1.7384520769119263,
+      "learning_rate": 0.00014445225679688564,
+      "loss": 4.215,
+      "step": 4720
+    },
+    {
+      "epoch": 0.5236790389991419,
+      "grad_norm": 1.768442153930664,
+      "learning_rate": 0.00014391909953189068,
+      "loss": 4.2345,
+      "step": 4730
+    },
+    {
+      "epoch": 0.5247861828448062,
+      "grad_norm": 1.8559410572052002,
+      "learning_rate": 0.00014338601920634296,
+      "loss": 4.13,
+      "step": 4740
+    },
+    {
+      "epoch": 0.5258933266904703,
+      "grad_norm": 1.7122317552566528,
+      "learning_rate": 0.0001428530225651161,
+      "loss": 4.1659,
+      "step": 4750
+    },
+    {
+      "epoch": 0.5270004705361344,
+      "grad_norm": 1.7201205492019653,
+      "learning_rate": 0.00014232011635202505,
+      "loss": 4.1604,
+      "step": 4760
+    },
+    {
+      "epoch": 0.5281076143817985,
+      "grad_norm": 1.9437607526779175,
+      "learning_rate": 0.00014178730730974042,
+      "loss": 4.1392,
+      "step": 4770
+    },
+    {
+      "epoch": 0.5292147582274627,
+      "grad_norm": 1.596734642982483,
+      "learning_rate": 0.00014125460217970345,
+      "loss": 4.211,
+      "step": 4780
+    },
+    {
+      "epoch": 0.5303219020731269,
+      "grad_norm": 1.9053328037261963,
+      "learning_rate": 0.00014072200770204072,
+      "loss": 4.2321,
+      "step": 4790
+    },
+    {
+      "epoch": 0.531429045918791,
+      "grad_norm": 1.7755322456359863,
+      "learning_rate": 0.0001401895306154785,
+      "loss": 4.1207,
+      "step": 4800
+    },
+    {
+      "epoch": 0.5325361897644552,
+      "grad_norm": 1.929548978805542,
+      "learning_rate": 0.0001396571776572581,
+      "loss": 4.2397,
+      "step": 4810
+    },
+    {
+      "epoch": 0.5336433336101193,
+      "grad_norm": 1.8717153072357178,
+      "learning_rate": 0.00013912495556304993,
+      "loss": 4.1339,
+      "step": 4820
+    },
+    {
+      "epoch": 0.5347504774557834,
+      "grad_norm": 1.6692941188812256,
+      "learning_rate": 0.00013859287106686888,
+      "loss": 4.2587,
+      "step": 4830
+    },
+    {
+      "epoch": 0.5358576213014475,
+      "grad_norm": 1.7127248048782349,
+      "learning_rate": 0.0001380609309009887,
+      "loss": 4.2461,
+      "step": 4840
+    },
+    {
+      "epoch": 0.5369647651471118,
+      "grad_norm": 1.8800731897354126,
+      "learning_rate": 0.00013752914179585707,
+      "loss": 4.247,
+      "step": 4850
+    },
+    {
+      "epoch": 0.5380719089927759,
+      "grad_norm": 1.7658569812774658,
+      "learning_rate": 0.00013699751048001037,
+      "loss": 4.2146,
+      "step": 4860
+    },
+    {
+      "epoch": 0.53917905283844,
+      "grad_norm": 1.798315167427063,
+      "learning_rate": 0.0001364660436799884,
+      "loss": 4.2983,
+      "step": 4870
+    },
+    {
+      "epoch": 0.5402861966841042,
+      "grad_norm": 1.8227429389953613,
+      "learning_rate": 0.00013593474812024966,
+      "loss": 4.1417,
+      "step": 4880
+    },
+    {
+      "epoch": 0.5413933405297683,
+      "grad_norm": 1.6355091333389282,
+      "learning_rate": 0.00013540363052308562,
+      "loss": 4.2216,
+      "step": 4890
+    },
+    {
+      "epoch": 0.5425004843754325,
+      "grad_norm": 1.5548667907714844,
+      "learning_rate": 0.0001348726976085364,
+      "loss": 4.1527,
+      "step": 4900
+    },
+    {
+      "epoch": 0.5436076282210967,
+      "grad_norm": 1.8281536102294922,
+      "learning_rate": 0.00013434195609430527,
+      "loss": 4.1626,
+      "step": 4910
+    },
+    {
+      "epoch": 0.5447147720667608,
+      "grad_norm": 1.8231089115142822,
+      "learning_rate": 0.0001338114126956737,
+      "loss": 4.1996,
+      "step": 4920
+    },
+    {
+      "epoch": 0.5458219159124249,
+      "grad_norm": 1.8852416276931763,
+      "learning_rate": 0.0001332810741254167,
+      "loss": 4.2351,
+      "step": 4930
+    },
+    {
+      "epoch": 0.546929059758089,
+      "grad_norm": 1.732532024383545,
+      "learning_rate": 0.0001327509470937173,
+      "loss": 4.1447,
+      "step": 4940
+    },
+    {
+      "epoch": 0.5480362036037533,
+      "grad_norm": 1.609029769897461,
+      "learning_rate": 0.00013222103830808245,
+      "loss": 4.2188,
+      "step": 4950
+    },
+    {
+      "epoch": 0.5491433474494174,
+      "grad_norm": 1.9934754371643066,
+      "learning_rate": 0.00013169135447325733,
+      "loss": 4.1418,
+      "step": 4960
+    },
+    {
+      "epoch": 0.5502504912950815,
+      "grad_norm": 1.727171778678894,
+      "learning_rate": 0.00013116190229114105,
+      "loss": 4.2243,
+      "step": 4970
+    },
+    {
+      "epoch": 0.5513576351407456,
+      "grad_norm": 1.692118525505066,
+      "learning_rate": 0.00013063268846070182,
+      "loss": 4.2112,
+      "step": 4980
+    },
+    {
+      "epoch": 0.5524647789864098,
+      "grad_norm": 1.7131632566452026,
+      "learning_rate": 0.00013010371967789184,
+      "loss": 4.231,
+      "step": 4990
+    },
+    {
+      "epoch": 0.5535719228320739,
+      "grad_norm": 1.615644097328186,
+      "learning_rate": 0.00012957500263556304,
+      "loss": 4.2108,
+      "step": 5000
+    },
+    {
+      "epoch": 0.5546790666777381,
+      "grad_norm": 1.747827172279358,
+      "learning_rate": 0.00012904654402338188,
+      "loss": 4.1971,
+      "step": 5010
+    },
+    {
+      "epoch": 0.5557862105234023,
+      "grad_norm": 1.6200209856033325,
+      "learning_rate": 0.00012851835052774536,
+      "loss": 4.1311,
+      "step": 5020
+    },
+    {
+      "epoch": 0.5568933543690664,
+      "grad_norm": 1.729896903038025,
+      "learning_rate": 0.00012799042883169574,
+      "loss": 4.1583,
+      "step": 5030
+    },
+    {
+      "epoch": 0.5580004982147305,
+      "grad_norm": 1.5931682586669922,
+      "learning_rate": 0.00012746278561483643,
+      "loss": 4.2101,
+      "step": 5040
+    },
+    {
+      "epoch": 0.5591076420603946,
+      "grad_norm": 1.5866844654083252,
+      "learning_rate": 0.00012693542755324745,
+      "loss": 4.1602,
+      "step": 5050
+    },
+    {
+      "epoch": 0.5602147859060589,
+      "grad_norm": 1.6571074724197388,
+      "learning_rate": 0.0001264083613194006,
+      "loss": 4.0951,
+      "step": 5060
+    },
+    {
+      "epoch": 0.561321929751723,
+      "grad_norm": 1.8744324445724487,
+      "learning_rate": 0.00012588159358207552,
+      "loss": 4.257,
+      "step": 5070
+    },
+    {
+      "epoch": 0.5624290735973871,
+      "grad_norm": 1.7426868677139282,
+      "learning_rate": 0.00012535513100627493,
+      "loss": 4.2172,
+      "step": 5080
+    },
+    {
+      "epoch": 0.5635362174430513,
+      "grad_norm": 1.730536699295044,
+      "learning_rate": 0.0001248289802531406,
+      "loss": 4.233,
+      "step": 5090
+    },
+    {
+      "epoch": 0.5646433612887154,
+      "grad_norm": 2.9089081287384033,
+      "learning_rate": 0.00012430314797986868,
+      "loss": 4.1745,
+      "step": 5100
+    },
+    {
+      "epoch": 0.5657505051343796,
+      "grad_norm": 1.5204368829727173,
+      "learning_rate": 0.00012377764083962593,
+      "loss": 4.2643,
+      "step": 5110
+    },
+    {
+      "epoch": 0.5668576489800438,
+      "grad_norm": 1.6791236400604248,
+      "learning_rate": 0.0001232524654814653,
+      "loss": 4.1622,
+      "step": 5120
+    },
+    {
+      "epoch": 0.5679647928257079,
+      "grad_norm": 1.7097837924957275,
+      "learning_rate": 0.0001227276285502417,
+      "loss": 4.1637,
+      "step": 5130
+    },
+    {
+      "epoch": 0.569071936671372,
+      "grad_norm": 1.736793041229248,
+      "learning_rate": 0.0001222031366865282,
+      "loss": 4.1484,
+      "step": 5140
+    },
+    {
+      "epoch": 0.5701790805170361,
+      "grad_norm": 1.7410223484039307,
+      "learning_rate": 0.00012167899652653166,
+      "loss": 4.2441,
+      "step": 5150
+    },
+    {
+      "epoch": 0.5712862243627004,
+      "grad_norm": 1.7958261966705322,
+      "learning_rate": 0.00012115521470200909,
+      "loss": 4.1426,
+      "step": 5160
+    },
+    {
+      "epoch": 0.5723933682083645,
+      "grad_norm": 1.6494110822677612,
+      "learning_rate": 0.00012063179784018367,
+      "loss": 4.1554,
+      "step": 5170
+    },
+    {
+      "epoch": 0.5735005120540286,
+      "grad_norm": 1.7107927799224854,
+      "learning_rate": 0.00012010875256366065,
+      "loss": 4.1758,
+      "step": 5180
+    },
+    {
+      "epoch": 0.5746076558996928,
+      "grad_norm": 1.5812790393829346,
+      "learning_rate": 0.00011958608549034393,
+      "loss": 4.1574,
+      "step": 5190
+    },
+    {
+      "epoch": 0.5757147997453569,
+      "grad_norm": 1.9988889694213867,
+      "learning_rate": 0.00011906380323335202,
+      "loss": 4.1817,
+      "step": 5200
+    },
+    {
+      "epoch": 0.576821943591021,
+      "grad_norm": 1.7576152086257935,
+      "learning_rate": 0.00011854191240093454,
+      "loss": 4.1286,
+      "step": 5210
+    },
+    {
+      "epoch": 0.5779290874366853,
+      "grad_norm": 1.8622543811798096,
+      "learning_rate": 0.00011802041959638852,
+      "loss": 4.2234,
+      "step": 5220
+    },
+    {
+      "epoch": 0.5790362312823494,
+      "grad_norm": 1.563567042350769,
+      "learning_rate": 0.00011749933141797491,
+      "loss": 4.2127,
+      "step": 5230
+    },
+    {
+      "epoch": 0.5801433751280135,
+      "grad_norm": 1.8555313348770142,
+      "learning_rate": 0.00011697865445883518,
+      "loss": 4.2174,
+      "step": 5240
+    },
+    {
+      "epoch": 0.5812505189736776,
+      "grad_norm": 1.699654221534729,
+      "learning_rate": 0.0001164583953069075,
+      "loss": 4.2326,
+      "step": 5250
+    },
+    {
+      "epoch": 0.5823576628193418,
+      "grad_norm": 1.8442728519439697,
+      "learning_rate": 0.00011593856054484402,
+      "loss": 4.1886,
+      "step": 5260
+    },
+    {
+      "epoch": 0.583464806665006,
+      "grad_norm": 1.6123567819595337,
+      "learning_rate": 0.00011541915674992698,
+      "loss": 4.2166,
+      "step": 5270
+    },
+    {
+      "epoch": 0.5845719505106701,
+      "grad_norm": 1.6981021165847778,
+      "learning_rate": 0.00011490019049398592,
+      "loss": 4.1661,
+      "step": 5280
+    },
+    {
+      "epoch": 0.5856790943563343,
+      "grad_norm": 1.945374846458435,
+      "learning_rate": 0.00011438166834331415,
+      "loss": 4.1136,
+      "step": 5290
+    },
+    {
+      "epoch": 0.5867862382019984,
+      "grad_norm": 1.8351072072982788,
+      "learning_rate": 0.00011386359685858608,
+      "loss": 4.1825,
+      "step": 5300
+    },
+    {
+      "epoch": 0.5878933820476625,
+      "grad_norm": 1.6921285390853882,
+      "learning_rate": 0.00011334598259477403,
+      "loss": 4.3186,
+      "step": 5310
+    },
+    {
+      "epoch": 0.5890005258933267,
+      "grad_norm": 1.7789745330810547,
+      "learning_rate": 0.00011282883210106502,
+      "loss": 4.174,
+      "step": 5320
+    },
+    {
+      "epoch": 0.5901076697389909,
+      "grad_norm": 1.9027832746505737,
+      "learning_rate": 0.00011231215192077837,
+      "loss": 4.1732,
+      "step": 5330
+    },
+    {
+      "epoch": 0.591214813584655,
+      "grad_norm": 1.7524372339248657,
+      "learning_rate": 0.00011179594859128261,
+      "loss": 4.1216,
+      "step": 5340
+    },
+    {
+      "epoch": 0.5923219574303191,
+      "grad_norm": 1.9307632446289062,
+      "learning_rate": 0.00011128022864391282,
+      "loss": 4.1911,
+      "step": 5350
+    },
+    {
+      "epoch": 0.5934291012759833,
+      "grad_norm": 1.544702172279358,
+      "learning_rate": 0.00011076499860388822,
+      "loss": 4.1589,
+      "step": 5360
+    },
+    {
+      "epoch": 0.5945362451216474,
+      "grad_norm": 1.8922128677368164,
+      "learning_rate": 0.00011025026499022907,
+      "loss": 4.2825,
+      "step": 5370
+    },
+    {
+      "epoch": 0.5956433889673116,
+      "grad_norm": 2.014092206954956,
+      "learning_rate": 0.00010973603431567483,
+      "loss": 4.226,
+      "step": 5380
+    },
+    {
+      "epoch": 0.5967505328129757,
+      "grad_norm": 1.8948957920074463,
+      "learning_rate": 0.00010922231308660126,
+      "loss": 4.1203,
+      "step": 5390
+    },
+    {
+      "epoch": 0.5978576766586399,
+      "grad_norm": 1.8543113470077515,
+      "learning_rate": 0.00010870910780293843,
+      "loss": 4.1794,
+      "step": 5400
+    },
+    {
+      "epoch": 0.598964820504304,
+      "grad_norm": 1.748165488243103,
+      "learning_rate": 0.00010819642495808816,
+      "loss": 4.1402,
+      "step": 5410
+    },
+    {
+      "epoch": 0.6000719643499681,
+      "grad_norm": 1.7419254779815674,
+      "learning_rate": 0.00010768427103884222,
+      "loss": 4.1379,
+      "step": 5420
+    },
+    {
+      "epoch": 0.6005148218882338,
+      "eval_loss": 4.627920150756836,
+      "eval_runtime": 47.5762,
+      "eval_samples_per_second": 1050.945,
+      "eval_steps_per_second": 131.368,
+      "step": 5424
+    },
+    {
+      "epoch": 0.6011791081956324,
+      "grad_norm": 1.6110368967056274,
+      "learning_rate": 0.00010717265252530005,
+      "loss": 4.1977,
+      "step": 5430
+    },
+    {
+      "epoch": 0.6022862520412965,
+      "grad_norm": 1.796631097793579,
+      "learning_rate": 0.00010666157589078666,
+      "loss": 4.0785,
+      "step": 5440
+    },
+    {
+      "epoch": 0.6033933958869606,
+      "grad_norm": 2.0883231163024902,
+      "learning_rate": 0.00010615104760177099,
+      "loss": 4.1797,
+      "step": 5450
+    },
+    {
+      "epoch": 0.6045005397326247,
+      "grad_norm": 1.860463261604309,
+      "learning_rate": 0.00010564107411778389,
+      "loss": 4.2259,
+      "step": 5460
+    },
+    {
+      "epoch": 0.6056076835782889,
+      "grad_norm": 1.904249668121338,
+      "learning_rate": 0.00010513166189133652,
+      "loss": 4.1542,
+      "step": 5470
+    },
+    {
+      "epoch": 0.6067148274239531,
+      "grad_norm": 1.8314825296401978,
+      "learning_rate": 0.00010462281736783863,
+      "loss": 4.1344,
+      "step": 5480
+    },
+    {
+      "epoch": 0.6078219712696172,
+      "grad_norm": 1.782535195350647,
+      "learning_rate": 0.00010411454698551695,
+      "loss": 4.2356,
+      "step": 5490
+    },
+    {
+      "epoch": 0.6089291151152814,
+      "grad_norm": 1.7429834604263306,
+      "learning_rate": 0.00010360685717533399,
+      "loss": 4.1243,
+      "step": 5500
+    },
+    {
+      "epoch": 0.6100362589609455,
+      "grad_norm": 1.6854382753372192,
+      "learning_rate": 0.00010309975436090626,
+      "loss": 4.1498,
+      "step": 5510
+    },
+    {
+      "epoch": 0.6111434028066096,
+      "grad_norm": 1.804469108581543,
+      "learning_rate": 0.00010259324495842343,
+      "loss": 4.1255,
+      "step": 5520
+    },
+    {
+      "epoch": 0.6122505466522737,
+      "grad_norm": 1.705212950706482,
+      "learning_rate": 0.00010208733537656674,
+      "loss": 4.1957,
+      "step": 5530
+    },
+    {
+      "epoch": 0.613357690497938,
+      "grad_norm": 1.9591410160064697,
+      "learning_rate": 0.00010158203201642832,
+      "loss": 4.1543,
+      "step": 5540
+    },
+    {
+      "epoch": 0.6144648343436021,
+      "grad_norm": 1.6803380250930786,
+      "learning_rate": 0.0001010773412714299,
+      "loss": 4.1026,
+      "step": 5550
+    },
+    {
+      "epoch": 0.6155719781892662,
+      "grad_norm": 1.7468843460083008,
+      "learning_rate": 0.00010057326952724198,
+      "loss": 4.1504,
+      "step": 5560
+    },
+    {
+      "epoch": 0.6166791220349304,
+      "grad_norm": 1.9714422225952148,
+      "learning_rate": 0.00010006982316170313,
+      "loss": 4.1705,
+      "step": 5570
+    },
+    {
+      "epoch": 0.6177862658805945,
+      "grad_norm": 1.8869982957839966,
+      "learning_rate": 9.956700854473915e-05,
+      "loss": 4.1569,
+      "step": 5580
+    },
+    {
+      "epoch": 0.6188934097262587,
+      "grad_norm": 1.750201940536499,
+      "learning_rate": 9.906483203828267e-05,
+      "loss": 4.14,
+      "step": 5590
+    },
+    {
+      "epoch": 0.6200005535719229,
+      "grad_norm": 1.7075363397598267,
+      "learning_rate": 9.856329999619234e-05,
+      "loss": 4.2095,
+      "step": 5600
+    },
+    {
+      "epoch": 0.621107697417587,
+      "grad_norm": 1.7756866216659546,
+      "learning_rate": 9.806241876417283e-05,
+      "loss": 4.1099,
+      "step": 5610
+    },
+    {
+      "epoch": 0.6222148412632511,
+      "grad_norm": 1.8669683933258057,
+      "learning_rate": 9.756219467969439e-05,
+      "loss": 4.0712,
+      "step": 5620
+    },
+    {
+      "epoch": 0.6233219851089152,
+      "grad_norm": 1.8651776313781738,
+      "learning_rate": 9.706263407191233e-05,
+      "loss": 4.129,
+      "step": 5630
+    },
+    {
+      "epoch": 0.6244291289545795,
+      "grad_norm": 2.035142660140991,
+      "learning_rate": 9.656374326158751e-05,
+      "loss": 4.0953,
+      "step": 5640
+    },
+    {
+      "epoch": 0.6255362728002436,
+      "grad_norm": 1.8959888219833374,
+      "learning_rate": 9.606552856100587e-05,
+      "loss": 4.1003,
+      "step": 5650
+    },
+    {
+      "epoch": 0.6266434166459077,
+      "grad_norm": 1.856665015220642,
+      "learning_rate": 9.556799627389895e-05,
+      "loss": 4.1039,
+      "step": 5660
+    },
+    {
+      "epoch": 0.6277505604915719,
+      "grad_norm": 1.9245810508728027,
+      "learning_rate": 9.507115269536393e-05,
+      "loss": 4.1141,
+      "step": 5670
+    },
+    {
+      "epoch": 0.628857704337236,
+      "grad_norm": 2.060251235961914,
+      "learning_rate": 9.457500411178387e-05,
+      "loss": 4.0789,
+      "step": 5680
+    },
+    {
+      "epoch": 0.6299648481829002,
+      "grad_norm": 1.8615323305130005,
+      "learning_rate": 9.407955680074851e-05,
+      "loss": 4.2196,
+      "step": 5690
+    },
+    {
+      "epoch": 0.6310719920285643,
+      "grad_norm": 1.9285634756088257,
+      "learning_rate": 9.358481703097444e-05,
+      "loss": 4.1043,
+      "step": 5700
+    },
+    {
+      "epoch": 0.6321791358742285,
+      "grad_norm": 1.6034704446792603,
+      "learning_rate": 9.309079106222617e-05,
+      "loss": 4.0746,
+      "step": 5710
+    },
+    {
+      "epoch": 0.6332862797198926,
+      "grad_norm": 1.7413756847381592,
+      "learning_rate": 9.259748514523653e-05,
+      "loss": 4.192,
+      "step": 5720
+    },
+    {
+      "epoch": 0.6343934235655567,
+      "grad_norm": 1.9419269561767578,
+      "learning_rate": 9.210490552162802e-05,
+      "loss": 4.1249,
+      "step": 5730
+    },
+    {
+      "epoch": 0.6355005674112209,
+      "grad_norm": 1.8918240070343018,
+      "learning_rate": 9.16130584238335e-05,
+      "loss": 4.2222,
+      "step": 5740
+    },
+    {
+      "epoch": 0.6366077112568851,
+      "grad_norm": 2.053088903427124,
+      "learning_rate": 9.112195007501741e-05,
+      "loss": 4.1686,
+      "step": 5750
+    },
+    {
+      "epoch": 0.6377148551025492,
+      "grad_norm": 1.6693891286849976,
+      "learning_rate": 9.063158668899718e-05,
+      "loss": 4.1209,
+      "step": 5760
+    },
+    {
+      "epoch": 0.6388219989482133,
+      "grad_norm": 1.8740805387496948,
+      "learning_rate": 9.014197447016429e-05,
+      "loss": 4.1498,
+      "step": 5770
+    },
+    {
+      "epoch": 0.6399291427938775,
+      "grad_norm": 1.9773539304733276,
+      "learning_rate": 8.965311961340625e-05,
+      "loss": 4.1596,
+      "step": 5780
+    },
+    {
+      "epoch": 0.6410362866395416,
+      "grad_norm": 1.6963318586349487,
+      "learning_rate": 8.916502830402766e-05,
+      "loss": 4.2411,
+      "step": 5790
+    },
+    {
+      "epoch": 0.6421434304852058,
+      "grad_norm": 1.623247504234314,
+      "learning_rate": 8.867770671767245e-05,
+      "loss": 4.1418,
+      "step": 5800
+    },
+    {
+      "epoch": 0.64325057433087,
+      "grad_norm": 2.0673112869262695,
+      "learning_rate": 8.819116102024543e-05,
+      "loss": 4.1382,
+      "step": 5810
+    },
+    {
+      "epoch": 0.6443577181765341,
+      "grad_norm": 1.7819314002990723,
+      "learning_rate": 8.770539736783434e-05,
+      "loss": 4.1903,
+      "step": 5820
+    },
+    {
+      "epoch": 0.6454648620221982,
+      "grad_norm": 1.777214765548706,
+      "learning_rate": 8.722042190663202e-05,
+      "loss": 4.189,
+      "step": 5830
+    },
+    {
+      "epoch": 0.6465720058678623,
+      "grad_norm": 1.81984281539917,
+      "learning_rate": 8.673624077285862e-05,
+      "loss": 4.1371,
+      "step": 5840
+    },
+    {
+      "epoch": 0.6476791497135266,
+      "grad_norm": 1.8755102157592773,
+      "learning_rate": 8.625286009268392e-05,
+      "loss": 4.2596,
+      "step": 5850
+    },
+    {
+      "epoch": 0.6487862935591907,
+      "grad_norm": 1.876805067062378,
+      "learning_rate": 8.577028598214987e-05,
+      "loss": 4.0999,
+      "step": 5860
+    },
+    {
+      "epoch": 0.6498934374048548,
+      "grad_norm": 1.6826459169387817,
+      "learning_rate": 8.528852454709317e-05,
+      "loss": 4.1618,
+      "step": 5870
+    },
+    {
+      "epoch": 0.651000581250519,
+      "grad_norm": 1.8074181079864502,
+      "learning_rate": 8.480758188306806e-05,
+      "loss": 4.2433,
+      "step": 5880
+    },
+    {
+      "epoch": 0.6521077250961831,
+      "grad_norm": 1.8665210008621216,
+      "learning_rate": 8.432746407526912e-05,
+      "loss": 4.1314,
+      "step": 5890
+    },
+    {
+      "epoch": 0.6532148689418472,
+      "grad_norm": 1.8794413805007935,
+      "learning_rate": 8.384817719845437e-05,
+      "loss": 4.086,
+      "step": 5900
+    },
+    {
+      "epoch": 0.6543220127875115,
+      "grad_norm": 1.7649370431900024,
+      "learning_rate": 8.336972731686829e-05,
+      "loss": 4.1615,
+      "step": 5910
+    },
+    {
+      "epoch": 0.6554291566331756,
+      "grad_norm": 1.903421401977539,
+      "learning_rate": 8.289212048416525e-05,
+      "loss": 4.1122,
+      "step": 5920
+    },
+    {
+      "epoch": 0.6565363004788397,
+      "grad_norm": 1.7113827466964722,
+      "learning_rate": 8.241536274333274e-05,
+      "loss": 4.208,
+      "step": 5930
+    },
+    {
+      "epoch": 0.6576434443245038,
+      "grad_norm": 2.1296987533569336,
+      "learning_rate": 8.193946012661506e-05,
+      "loss": 4.0914,
+      "step": 5940
+    },
+    {
+      "epoch": 0.658750588170168,
+      "grad_norm": 1.716139316558838,
+      "learning_rate": 8.146441865543689e-05,
+      "loss": 4.1587,
+      "step": 5950
+    },
+    {
+      "epoch": 0.6598577320158322,
+      "grad_norm": 1.7411868572235107,
+      "learning_rate": 8.099024434032717e-05,
+      "loss": 4.1466,
+      "step": 5960
+    },
+    {
+      "epoch": 0.6609648758614963,
+      "grad_norm": 1.733300805091858,
+      "learning_rate": 8.051694318084302e-05,
+      "loss": 4.1343,
+      "step": 5970
+    },
+    {
+      "epoch": 0.6620720197071605,
+      "grad_norm": 1.6459636688232422,
+      "learning_rate": 8.004452116549383e-05,
+      "loss": 4.1739,
+      "step": 5980
+    },
+    {
+      "epoch": 0.6631791635528246,
+      "grad_norm": 1.8940541744232178,
+      "learning_rate": 7.95729842716655e-05,
+      "loss": 4.1174,
+      "step": 5990
+    },
+    {
+      "epoch": 0.6642863073984887,
+      "grad_norm": 1.8566889762878418,
+      "learning_rate": 7.910233846554479e-05,
+      "loss": 4.0711,
+      "step": 6000
+    },
+    {
+      "epoch": 0.665393451244153,
+      "grad_norm": 1.9669746160507202,
+      "learning_rate": 7.863258970204388e-05,
+      "loss": 4.0837,
+      "step": 6010
+    },
+    {
+      "epoch": 0.6665005950898171,
+      "grad_norm": 1.7128499746322632,
+      "learning_rate": 7.816374392472497e-05,
+      "loss": 4.1395,
+      "step": 6020
+    },
+    {
+      "epoch": 0.6676077389354812,
+      "grad_norm": 1.7379517555236816,
+      "learning_rate": 7.76958070657251e-05,
+      "loss": 4.1281,
+      "step": 6030
+    },
+    {
+      "epoch": 0.6687148827811453,
+      "grad_norm": 1.8854877948760986,
+      "learning_rate": 7.722878504568109e-05,
+      "loss": 4.0671,
+      "step": 6040
+    },
+    {
+      "epoch": 0.6698220266268095,
+      "grad_norm": 1.7916603088378906,
+      "learning_rate": 7.676268377365466e-05,
+      "loss": 4.1362,
+      "step": 6050
+    },
+    {
+      "epoch": 0.6709291704724737,
+      "grad_norm": 2.0008413791656494,
+      "learning_rate": 7.629750914705761e-05,
+      "loss": 4.1102,
+      "step": 6060
+    },
+    {
+      "epoch": 0.6720363143181378,
+      "grad_norm": 1.8542160987854004,
+      "learning_rate": 7.583326705157721e-05,
+      "loss": 4.1623,
+      "step": 6070
+    },
+    {
+      "epoch": 0.6731434581638019,
+      "grad_norm": 1.881985068321228,
+      "learning_rate": 7.53699633611018e-05,
+      "loss": 4.1393,
+      "step": 6080
+    },
+    {
+      "epoch": 0.6742506020094661,
+      "grad_norm": 1.7334316968917847,
+      "learning_rate": 7.49076039376464e-05,
+      "loss": 4.1458,
+      "step": 6090
+    },
+    {
+      "epoch": 0.6753577458551302,
+      "grad_norm": 1.8924003839492798,
+      "learning_rate": 7.444619463127853e-05,
+      "loss": 4.2553,
+      "step": 6100
+    },
+    {
+      "epoch": 0.6764648897007943,
+      "grad_norm": 1.9810236692428589,
+      "learning_rate": 7.398574128004426e-05,
+      "loss": 4.1251,
+      "step": 6110
+    },
+    {
+      "epoch": 0.6775720335464586,
+      "grad_norm": 1.8832056522369385,
+      "learning_rate": 7.352624970989424e-05,
+      "loss": 4.155,
+      "step": 6120
+    },
+    {
+      "epoch": 0.6786791773921227,
+      "grad_norm": 1.8153836727142334,
+      "learning_rate": 7.306772573461017e-05,
+      "loss": 4.0874,
+      "step": 6130
+    },
+    {
+      "epoch": 0.6797863212377868,
+      "grad_norm": 2.0987229347229004,
+      "learning_rate": 7.261017515573098e-05,
+      "loss": 4.0904,
+      "step": 6140
+    },
+    {
+      "epoch": 0.6808934650834509,
+      "grad_norm": 1.8796117305755615,
+      "learning_rate": 7.215360376247964e-05,
+      "loss": 4.15,
+      "step": 6150
+    },
+    {
+      "epoch": 0.6820006089291151,
+      "grad_norm": 1.920079231262207,
+      "learning_rate": 7.169801733168987e-05,
+      "loss": 4.2099,
+      "step": 6160
+    },
+    {
+      "epoch": 0.6831077527747793,
+      "grad_norm": 1.662898302078247,
+      "learning_rate": 7.124342162773293e-05,
+      "loss": 4.1799,
+      "step": 6170
+    },
+    {
+      "epoch": 0.6842148966204434,
+      "grad_norm": 2.019810438156128,
+      "learning_rate": 7.07898224024448e-05,
+      "loss": 4.1621,
+      "step": 6180
+    },
+    {
+      "epoch": 0.6853220404661076,
+      "grad_norm": 1.784677505493164,
+      "learning_rate": 7.033722539505341e-05,
+      "loss": 4.1441,
+      "step": 6190
+    },
+    {
+      "epoch": 0.6864291843117717,
+      "grad_norm": 1.795859932899475,
+      "learning_rate": 6.988563633210597e-05,
+      "loss": 4.0548,
+      "step": 6200
+    },
+    {
+      "epoch": 0.6875363281574358,
+      "grad_norm": 1.8101192712783813,
+      "learning_rate": 6.94350609273965e-05,
+      "loss": 4.1191,
+      "step": 6210
+    },
+    {
+      "epoch": 0.6886434720031,
+      "grad_norm": 1.8775107860565186,
+      "learning_rate": 6.89855048818936e-05,
+      "loss": 4.1886,
+      "step": 6220
+    },
+    {
+      "epoch": 0.6897506158487642,
+      "grad_norm": 1.709236979484558,
+      "learning_rate": 6.853697388366826e-05,
+      "loss": 4.1743,
+      "step": 6230
+    },
+    {
+      "epoch": 0.6908577596944283,
+      "grad_norm": 1.795945405960083,
+      "learning_rate": 6.808947360782194e-05,
+      "loss": 4.1985,
+      "step": 6240
+    },
+    {
+      "epoch": 0.6919649035400924,
+      "grad_norm": 1.6891391277313232,
+      "learning_rate": 6.764300971641474e-05,
+      "loss": 4.1698,
+      "step": 6250
+    },
+    {
+      "epoch": 0.6930720473857566,
+      "grad_norm": 2.1170356273651123,
+      "learning_rate": 6.719758785839373e-05,
+      "loss": 4.1442,
+      "step": 6260
+    },
+    {
+      "epoch": 0.6941791912314207,
+      "grad_norm": 1.8130853176116943,
+      "learning_rate": 6.675321366952153e-05,
+      "loss": 4.1854,
+      "step": 6270
+    },
+    {
+      "epoch": 0.6952863350770849,
+      "grad_norm": 1.715897798538208,
+      "learning_rate": 6.6309892772305e-05,
+      "loss": 4.2232,
+      "step": 6280
+    },
+    {
+      "epoch": 0.696393478922749,
+      "grad_norm": 1.8984107971191406,
+      "learning_rate": 6.5867630775924e-05,
+      "loss": 4.0847,
+      "step": 6290
+    },
+    {
+      "epoch": 0.6975006227684132,
+      "grad_norm": 1.8972477912902832,
+      "learning_rate": 6.542643327616062e-05,
+      "loss": 4.2082,
+      "step": 6300
+    },
+    {
+      "epoch": 0.6986077666140773,
+      "grad_norm": 1.8924113512039185,
+      "learning_rate": 6.498630585532813e-05,
+      "loss": 4.068,
+      "step": 6310
+    },
+    {
+      "epoch": 0.6997149104597414,
+      "grad_norm": 1.6712241172790527,
+      "learning_rate": 6.454725408220057e-05,
+      "loss": 4.1285,
+      "step": 6320
+    },
+    {
+      "epoch": 0.7006006255362728,
+      "eval_loss": 4.591481685638428,
+      "eval_runtime": 47.7318,
+      "eval_samples_per_second": 1047.52,
+      "eval_steps_per_second": 130.94,
+      "step": 6328
+    },
+    {
+      "epoch": 0.7008220543054057,
+      "grad_norm": 1.9414761066436768,
+      "learning_rate": 6.410928351194216e-05,
+      "loss": 4.0368,
+      "step": 6330
+    },
+    {
+      "epoch": 0.7019291981510698,
+      "grad_norm": 1.9685355424880981,
+      "learning_rate": 6.367239968603704e-05,
+      "loss": 4.1167,
+      "step": 6340
+    },
+    {
+      "epoch": 0.7030363419967339,
+      "grad_norm": 2.116163969039917,
+      "learning_rate": 6.323660813221919e-05,
+      "loss": 4.1353,
+      "step": 6350
+    },
+    {
+      "epoch": 0.704143485842398,
+      "grad_norm": 1.7894047498703003,
+      "learning_rate": 6.280191436440243e-05,
+      "loss": 4.1792,
+      "step": 6360
+    },
+    {
+      "epoch": 0.7052506296880622,
+      "grad_norm": 1.95182466506958,
+      "learning_rate": 6.236832388261071e-05,
+      "loss": 4.1151,
+      "step": 6370
+    },
+    {
+      "epoch": 0.7063577735337264,
+      "grad_norm": 2.0705292224884033,
+      "learning_rate": 6.19358421729085e-05,
+      "loss": 4.1442,
+      "step": 6380
+    },
+    {
+      "epoch": 0.7074649173793905,
+      "grad_norm": 1.8181239366531372,
+      "learning_rate": 6.150447470733139e-05,
+      "loss": 4.06,
+      "step": 6390
+    },
+    {
+      "epoch": 0.7085720612250547,
+      "grad_norm": 1.9468276500701904,
+      "learning_rate": 6.10742269438168e-05,
+      "loss": 4.1461,
+      "step": 6400
+    },
+    {
+      "epoch": 0.7096792050707188,
+      "grad_norm": 1.9070271253585815,
+      "learning_rate": 6.064510432613499e-05,
+      "loss": 4.169,
+      "step": 6410
+    },
+    {
+      "epoch": 0.7107863489163829,
+      "grad_norm": 1.8699619770050049,
+      "learning_rate": 6.0217112283820164e-05,
+      "loss": 4.1489,
+      "step": 6420
+    },
+    {
+      "epoch": 0.7118934927620472,
+      "grad_norm": 1.8163495063781738,
+      "learning_rate": 5.979025623210173e-05,
+      "loss": 4.132,
+      "step": 6430
+    },
+    {
+      "epoch": 0.7130006366077113,
+      "grad_norm": 1.9592483043670654,
+      "learning_rate": 5.936454157183582e-05,
+      "loss": 4.136,
+      "step": 6440
+    },
+    {
+      "epoch": 0.7141077804533754,
+      "grad_norm": 1.958545207977295,
+      "learning_rate": 5.8939973689436986e-05,
+      "loss": 4.1013,
+      "step": 6450
+    },
+    {
+      "epoch": 0.7152149242990395,
+      "grad_norm": 1.8481988906860352,
+      "learning_rate": 5.851655795680997e-05,
+      "loss": 4.0886,
+      "step": 6460
+    },
+    {
+      "epoch": 0.7163220681447037,
+      "grad_norm": 1.993730068206787,
+      "learning_rate": 5.809429973128188e-05,
+      "loss": 4.0543,
+      "step": 6470
+    },
+    {
+      "epoch": 0.7174292119903678,
+      "grad_norm": 2.0278005599975586,
+      "learning_rate": 5.767320435553402e-05,
+      "loss": 4.0975,
+      "step": 6480
+    },
+    {
+      "epoch": 0.718536355836032,
+      "grad_norm": 2.261789560317993,
+      "learning_rate": 5.725327715753496e-05,
+      "loss": 4.0237,
+      "step": 6490
+    },
+    {
+      "epoch": 0.7196434996816962,
+      "grad_norm": 1.8525474071502686,
+      "learning_rate": 5.683452345047249e-05,
+      "loss": 4.1494,
+      "step": 6500
+    },
+    {
+      "epoch": 0.7207506435273603,
+      "grad_norm": 2.043565034866333,
+      "learning_rate": 5.641694853268668e-05,
+      "loss": 4.0634,
+      "step": 6510
+    },
+    {
+      "epoch": 0.7218577873730244,
+      "grad_norm": 1.827444314956665,
+      "learning_rate": 5.600055768760286e-05,
+      "loss": 4.1274,
+      "step": 6520
+    },
+    {
+      "epoch": 0.7229649312186885,
+      "grad_norm": 2.0032002925872803,
+      "learning_rate": 5.5585356183664645e-05,
+      "loss": 4.1469,
+      "step": 6530
+    },
+    {
+      "epoch": 0.7240720750643528,
+      "grad_norm": 1.6478224992752075,
+      "learning_rate": 5.51713492742674e-05,
+      "loss": 4.0744,
+      "step": 6540
+    },
+    {
+      "epoch": 0.7251792189100169,
+      "grad_norm": 1.7439184188842773,
+      "learning_rate": 5.475854219769166e-05,
+      "loss": 4.2054,
+      "step": 6550
+    },
+    {
+      "epoch": 0.726286362755681,
+      "grad_norm": 1.9458317756652832,
+      "learning_rate": 5.434694017703697e-05,
+      "loss": 4.1796,
+      "step": 6560
+    },
+    {
+      "epoch": 0.7273935066013452,
+      "grad_norm": 2.0748138427734375,
+      "learning_rate": 5.3936548420155697e-05,
+      "loss": 4.1775,
+      "step": 6570
+    },
+    {
+      "epoch": 0.7285006504470093,
+      "grad_norm": 1.8868814706802368,
+      "learning_rate": 5.352737211958718e-05,
+      "loss": 4.1073,
+      "step": 6580
+    },
+    {
+      "epoch": 0.7296077942926735,
+      "grad_norm": 1.742892861366272,
+      "learning_rate": 5.311941645249202e-05,
+      "loss": 4.0817,
+      "step": 6590
+    },
+    {
+      "epoch": 0.7307149381383377,
+      "grad_norm": 1.8144102096557617,
+      "learning_rate": 5.27126865805866e-05,
+      "loss": 4.0341,
+      "step": 6600
+    },
+    {
+      "epoch": 0.7318220819840018,
+      "grad_norm": 1.926652431488037,
+      "learning_rate": 5.2307187650077754e-05,
+      "loss": 4.068,
+      "step": 6610
+    },
+    {
+      "epoch": 0.7329292258296659,
+      "grad_norm": 1.9855986833572388,
+      "learning_rate": 5.190292479159762e-05,
+      "loss": 4.1064,
+      "step": 6620
+    },
+    {
+      "epoch": 0.73403636967533,
+      "grad_norm": 2.075453758239746,
+      "learning_rate": 5.1499903120138836e-05,
+      "loss": 4.1554,
+      "step": 6630
+    },
+    {
+      "epoch": 0.7351435135209942,
+      "grad_norm": 1.9861963987350464,
+      "learning_rate": 5.109812773498967e-05,
+      "loss": 4.0293,
+      "step": 6640
+    },
+    {
+      "epoch": 0.7362506573666584,
+      "grad_norm": 1.978935956954956,
+      "learning_rate": 5.069760371966961e-05,
+      "loss": 4.2185,
+      "step": 6650
+    },
+    {
+      "epoch": 0.7373578012123225,
+      "grad_norm": 1.995675802230835,
+      "learning_rate": 5.029833614186508e-05,
+      "loss": 4.1007,
+      "step": 6660
+    },
+    {
+      "epoch": 0.7384649450579867,
+      "grad_norm": 1.9232666492462158,
+      "learning_rate": 4.9900330053365016e-05,
+      "loss": 4.0501,
+      "step": 6670
+    },
+    {
+      "epoch": 0.7395720889036508,
+      "grad_norm": 1.8937183618545532,
+      "learning_rate": 4.9503590489997526e-05,
+      "loss": 4.1335,
+      "step": 6680
+    },
+    {
+      "epoch": 0.7406792327493149,
+      "grad_norm": 1.9520975351333618,
+      "learning_rate": 4.9108122471565575e-05,
+      "loss": 4.1041,
+      "step": 6690
+    },
+    {
+      "epoch": 0.7417863765949791,
+      "grad_norm": 1.7196742296218872,
+      "learning_rate": 4.871393100178383e-05,
+      "loss": 4.0239,
+      "step": 6700
+    },
+    {
+      "epoch": 0.7428935204406433,
+      "grad_norm": 1.707877278327942,
+      "learning_rate": 4.8321021068215213e-05,
+      "loss": 4.0492,
+      "step": 6710
+    },
+    {
+      "epoch": 0.7440006642863074,
+      "grad_norm": 1.8894785642623901,
+      "learning_rate": 4.792939764220783e-05,
+      "loss": 4.1245,
+      "step": 6720
+    },
+    {
+      "epoch": 0.7451078081319715,
+      "grad_norm": 1.8639329671859741,
+      "learning_rate": 4.7539065678832124e-05,
+      "loss": 4.1391,
+      "step": 6730
+    },
+    {
+      "epoch": 0.7462149519776357,
+      "grad_norm": 1.8548616170883179,
+      "learning_rate": 4.715003011681802e-05,
+      "loss": 4.0808,
+      "step": 6740
+    },
+    {
+      "epoch": 0.7473220958232999,
+      "grad_norm": 1.9282774925231934,
+      "learning_rate": 4.676229587849266e-05,
+      "loss": 4.0539,
+      "step": 6750
+    },
+    {
+      "epoch": 0.748429239668964,
+      "grad_norm": 1.9064303636550903,
+      "learning_rate": 4.6375867869717907e-05,
+      "loss": 4.0646,
+      "step": 6760
+    },
+    {
+      "epoch": 0.7495363835146281,
+      "grad_norm": 1.8383103609085083,
+      "learning_rate": 4.599075097982845e-05,
+      "loss": 4.2169,
+      "step": 6770
+    },
+    {
+      "epoch": 0.7506435273602923,
+      "grad_norm": 2.0256059169769287,
+      "learning_rate": 4.5606950081569844e-05,
+      "loss": 4.0512,
+      "step": 6780
+    },
+    {
+      "epoch": 0.7517506712059564,
+      "grad_norm": 1.736679196357727,
+      "learning_rate": 4.522447003103673e-05,
+      "loss": 4.0372,
+      "step": 6790
+    },
+    {
+      "epoch": 0.7528578150516205,
+      "grad_norm": 1.9670689105987549,
+      "learning_rate": 4.4843315667611805e-05,
+      "loss": 4.1488,
+      "step": 6800
+    },
+    {
+      "epoch": 0.7539649588972848,
+      "grad_norm": 1.8981173038482666,
+      "learning_rate": 4.446349181390415e-05,
+      "loss": 4.1287,
+      "step": 6810
+    },
+    {
+      "epoch": 0.7550721027429489,
+      "grad_norm": 1.8218042850494385,
+      "learning_rate": 4.408500327568839e-05,
+      "loss": 4.082,
+      "step": 6820
+    },
+    {
+      "epoch": 0.756179246588613,
+      "grad_norm": 2.0317442417144775,
+      "learning_rate": 4.370785484184392e-05,
+      "loss": 4.0747,
+      "step": 6830
+    },
+    {
+      "epoch": 0.7572863904342771,
+      "grad_norm": 1.9839726686477661,
+      "learning_rate": 4.3332051284294294e-05,
+      "loss": 4.1124,
+      "step": 6840
+    },
+    {
+      "epoch": 0.7583935342799413,
+      "grad_norm": 1.8812097311019897,
+      "learning_rate": 4.295759735794677e-05,
+      "loss": 4.19,
+      "step": 6850
+    },
+    {
+      "epoch": 0.7595006781256055,
+      "grad_norm": 2.1193666458129883,
+      "learning_rate": 4.258449780063224e-05,
+      "loss": 4.0732,
+      "step": 6860
+    },
+    {
+      "epoch": 0.7606078219712696,
+      "grad_norm": 1.9525375366210938,
+      "learning_rate": 4.2212757333045283e-05,
+      "loss": 4.0199,
+      "step": 6870
+    },
+    {
+      "epoch": 0.7617149658169338,
+      "grad_norm": 1.836427927017212,
+      "learning_rate": 4.184238065868437e-05,
+      "loss": 4.1303,
+      "step": 6880
+    },
+    {
+      "epoch": 0.7628221096625979,
+      "grad_norm": 1.840078592300415,
+      "learning_rate": 4.147337246379235e-05,
+      "loss": 4.1277,
+      "step": 6890
+    },
+    {
+      "epoch": 0.763929253508262,
+      "grad_norm": 1.8698256015777588,
+      "learning_rate": 4.1105737417297344e-05,
+      "loss": 4.1591,
+      "step": 6900
+    },
+    {
+      "epoch": 0.7650363973539263,
+      "grad_norm": 2.0566794872283936,
+      "learning_rate": 4.0739480170753206e-05,
+      "loss": 4.0899,
+      "step": 6910
+    },
+    {
+      "epoch": 0.7661435411995904,
+      "grad_norm": 1.9704601764678955,
+      "learning_rate": 4.0374605358281376e-05,
+      "loss": 4.1081,
+      "step": 6920
+    },
+    {
+      "epoch": 0.7672506850452545,
+      "grad_norm": 1.9555472135543823,
+      "learning_rate": 4.00111175965116e-05,
+      "loss": 4.1758,
+      "step": 6930
+    },
+    {
+      "epoch": 0.7683578288909186,
+      "grad_norm": 2.1124958992004395,
+      "learning_rate": 3.964902148452383e-05,
+      "loss": 4.0637,
+      "step": 6940
+    },
+    {
+      "epoch": 0.7694649727365828,
+      "grad_norm": 1.9690048694610596,
+      "learning_rate": 3.9288321603789986e-05,
+      "loss": 4.0652,
+      "step": 6950
+    },
+    {
+      "epoch": 0.770572116582247,
+      "grad_norm": 2.1201741695404053,
+      "learning_rate": 3.892902251811595e-05,
+      "loss": 4.1788,
+      "step": 6960
+    },
+    {
+      "epoch": 0.7716792604279111,
+      "grad_norm": 1.8210556507110596,
+      "learning_rate": 3.857112877358391e-05,
+      "loss": 4.1493,
+      "step": 6970
+    },
+    {
+      "epoch": 0.7727864042735753,
+      "grad_norm": 1.81236732006073,
+      "learning_rate": 3.8214644898494636e-05,
+      "loss": 4.1674,
+      "step": 6980
+    },
+    {
+      "epoch": 0.7738935481192394,
+      "grad_norm": 1.8724617958068848,
+      "learning_rate": 3.7859575403310534e-05,
+      "loss": 4.1305,
+      "step": 6990
+    },
+    {
+      "epoch": 0.7750006919649035,
+      "grad_norm": 2.0282046794891357,
+      "learning_rate": 3.75059247805982e-05,
+      "loss": 4.0865,
+      "step": 7000
+    },
+    {
+      "epoch": 0.7761078358105676,
+      "grad_norm": 1.8961248397827148,
+      "learning_rate": 3.715369750497182e-05,
+      "loss": 4.0857,
+      "step": 7010
+    },
+    {
+      "epoch": 0.7772149796562319,
+      "grad_norm": 2.3275086879730225,
+      "learning_rate": 3.68028980330365e-05,
+      "loss": 4.0597,
+      "step": 7020
+    },
+    {
+      "epoch": 0.778322123501896,
+      "grad_norm": 1.8095614910125732,
+      "learning_rate": 3.645353080333174e-05,
+      "loss": 4.1275,
+      "step": 7030
+    },
+    {
+      "epoch": 0.7794292673475601,
+      "grad_norm": 2.012094736099243,
+      "learning_rate": 3.610560023627555e-05,
+      "loss": 4.1378,
+      "step": 7040
+    },
+    {
+      "epoch": 0.7805364111932243,
+      "grad_norm": 1.99796462059021,
+      "learning_rate": 3.57591107341083e-05,
+      "loss": 4.0956,
+      "step": 7050
+    },
+    {
+      "epoch": 0.7816435550388884,
+      "grad_norm": 1.9282629489898682,
+      "learning_rate": 3.541406668083704e-05,
+      "loss": 4.214,
+      "step": 7060
+    },
+    {
+      "epoch": 0.7827506988845526,
+      "grad_norm": 1.991390347480774,
+      "learning_rate": 3.507047244218014e-05,
+      "loss": 4.0804,
+      "step": 7070
+    },
+    {
+      "epoch": 0.7838578427302167,
+      "grad_norm": 1.7052147388458252,
+      "learning_rate": 3.4728332365511954e-05,
+      "loss": 4.1432,
+      "step": 7080
+    },
+    {
+      "epoch": 0.7849649865758809,
+      "grad_norm": 1.8621222972869873,
+      "learning_rate": 3.43876507798079e-05,
+      "loss": 4.1015,
+      "step": 7090
+    },
+    {
+      "epoch": 0.786072130421545,
+      "grad_norm": 1.8071262836456299,
+      "learning_rate": 3.404843199558945e-05,
+      "loss": 4.1026,
+      "step": 7100
+    },
+    {
+      "epoch": 0.7871792742672091,
+      "grad_norm": 2.0991318225860596,
+      "learning_rate": 3.3710680304870056e-05,
+      "loss": 4.1409,
+      "step": 7110
+    },
+    {
+      "epoch": 0.7882864181128734,
+      "grad_norm": 1.8139592409133911,
+      "learning_rate": 3.3374399981100354e-05,
+      "loss": 4.0908,
+      "step": 7120
+    },
+    {
+      "epoch": 0.7893935619585375,
+      "grad_norm": 1.8121578693389893,
+      "learning_rate": 3.30395952791144e-05,
+      "loss": 4.1064,
+      "step": 7130
+    },
+    {
+      "epoch": 0.7905007058042016,
+      "grad_norm": 2.042020559310913,
+      "learning_rate": 3.2706270435075716e-05,
+      "loss": 4.1941,
+      "step": 7140
+    },
+    {
+      "epoch": 0.7916078496498657,
+      "grad_norm": 1.8589932918548584,
+      "learning_rate": 3.2374429666423614e-05,
+      "loss": 4.1584,
+      "step": 7150
+    },
+    {
+      "epoch": 0.7927149934955299,
+      "grad_norm": 1.8327692747116089,
+      "learning_rate": 3.2044077171820156e-05,
+      "loss": 4.1425,
+      "step": 7160
+    },
+    {
+      "epoch": 0.793822137341194,
+      "grad_norm": 1.945387601852417,
+      "learning_rate": 3.1715217131096554e-05,
+      "loss": 4.0988,
+      "step": 7170
+    },
+    {
+      "epoch": 0.7949292811868582,
+      "grad_norm": 1.9471564292907715,
+      "learning_rate": 3.138785370520073e-05,
+      "loss": 4.0443,
+      "step": 7180
+    },
+    {
+      "epoch": 0.7960364250325224,
+      "grad_norm": 2.0169224739074707,
+      "learning_rate": 3.106199103614441e-05,
+      "loss": 4.0198,
+      "step": 7190
+    },
+    {
+      "epoch": 0.7971435688781865,
+      "grad_norm": 1.8452967405319214,
+      "learning_rate": 3.073763324695078e-05,
+      "loss": 4.0925,
+      "step": 7200
+    },
+    {
+      "epoch": 0.7982507127238506,
+      "grad_norm": 1.9444732666015625,
+      "learning_rate": 3.041478444160233e-05,
+      "loss": 4.0824,
+      "step": 7210
+    },
+    {
+      "epoch": 0.7993578565695147,
+      "grad_norm": 2.1381709575653076,
+      "learning_rate": 3.0093448704988815e-05,
+      "loss": 4.0398,
+      "step": 7220
+    },
+    {
+      "epoch": 0.800465000415179,
+      "grad_norm": 1.9915144443511963,
+      "learning_rate": 2.977363010285588e-05,
+      "loss": 4.1728,
+      "step": 7230
+    },
+    {
+      "epoch": 0.8006864291843118,
+      "eval_loss": 4.569488048553467,
+      "eval_runtime": 47.645,
+      "eval_samples_per_second": 1049.428,
+      "eval_steps_per_second": 131.179,
+      "step": 7232
+    },
+    {
+      "epoch": 0.8015721442608431,
+      "grad_norm": 1.957175612449646,
+      "learning_rate": 2.945533268175326e-05,
+      "loss": 4.071,
+      "step": 7240
+    },
+    {
+      "epoch": 0.8026792881065072,
+      "grad_norm": 1.9286167621612549,
+      "learning_rate": 2.9138560468983813e-05,
+      "loss": 4.1897,
+      "step": 7250
+    },
+    {
+      "epoch": 0.8037864319521714,
+      "grad_norm": 2.1002249717712402,
+      "learning_rate": 2.882331747255242e-05,
+      "loss": 4.0542,
+      "step": 7260
+    },
+    {
+      "epoch": 0.8048935757978355,
+      "grad_norm": 1.9208799600601196,
+      "learning_rate": 2.8509607681115376e-05,
+      "loss": 4.1019,
+      "step": 7270
+    },
+    {
+      "epoch": 0.8060007196434997,
+      "grad_norm": 2.01770281791687,
+      "learning_rate": 2.8197435063929918e-05,
+      "loss": 4.189,
+      "step": 7280
+    },
+    {
+      "epoch": 0.8071078634891639,
+      "grad_norm": 1.8172143697738647,
+      "learning_rate": 2.7886803570803794e-05,
+      "loss": 4.1824,
+      "step": 7290
+    },
+    {
+      "epoch": 0.808215007334828,
+      "grad_norm": 1.9146698713302612,
+      "learning_rate": 2.757771713204575e-05,
+      "loss": 4.1174,
+      "step": 7300
+    },
+    {
+      "epoch": 0.8093221511804921,
+      "grad_norm": 1.9576022624969482,
+      "learning_rate": 2.7270179658415286e-05,
+      "loss": 4.0337,
+      "step": 7310
+    },
+    {
+      "epoch": 0.8104292950261562,
+      "grad_norm": 2.060394525527954,
+      "learning_rate": 2.6964195041073493e-05,
+      "loss": 4.1328,
+      "step": 7320
+    },
+    {
+      "epoch": 0.8115364388718205,
+      "grad_norm": 2.0036439895629883,
+      "learning_rate": 2.665976715153377e-05,
+      "loss": 4.0677,
+      "step": 7330
+    },
+    {
+      "epoch": 0.8126435827174846,
+      "grad_norm": 2.0634758472442627,
+      "learning_rate": 2.6356899841612606e-05,
+      "loss": 4.0971,
+      "step": 7340
+    },
+    {
+      "epoch": 0.8137507265631487,
+      "grad_norm": 1.9028979539871216,
+      "learning_rate": 2.6055596943381336e-05,
+      "loss": 4.1021,
+      "step": 7350
+    },
+    {
+      "epoch": 0.8148578704088129,
+      "grad_norm": 2.0446481704711914,
+      "learning_rate": 2.575586226911705e-05,
+      "loss": 4.0541,
+      "step": 7360
+    },
+    {
+      "epoch": 0.815965014254477,
+      "grad_norm": 1.9749499559402466,
+      "learning_rate": 2.5457699611254932e-05,
+      "loss": 4.0906,
+      "step": 7370
+    },
+    {
+      "epoch": 0.8170721581001411,
+      "grad_norm": 1.7769414186477661,
+      "learning_rate": 2.516111274233983e-05,
+      "loss": 4.1843,
+      "step": 7380
+    },
+    {
+      "epoch": 0.8181793019458053,
+      "grad_norm": 1.953451156616211,
+      "learning_rate": 2.486610541497876e-05,
+      "loss": 4.1701,
+      "step": 7390
+    },
+    {
+      "epoch": 0.8192864457914695,
+      "grad_norm": 2.059490442276001,
+      "learning_rate": 2.4572681361793407e-05,
+      "loss": 4.0519,
+      "step": 7400
+    },
+    {
+      "epoch": 0.8203935896371336,
+      "grad_norm": 1.713393211364746,
+      "learning_rate": 2.4280844295372688e-05,
+      "loss": 4.1224,
+      "step": 7410
+    },
+    {
+      "epoch": 0.8215007334827977,
+      "grad_norm": 2.0525457859039307,
+      "learning_rate": 2.3990597908226218e-05,
+      "loss": 4.0091,
+      "step": 7420
+    },
+    {
+      "epoch": 0.8226078773284619,
+      "grad_norm": 1.8792610168457031,
+      "learning_rate": 2.370194587273711e-05,
+      "loss": 4.1389,
+      "step": 7430
+    },
+    {
+      "epoch": 0.8237150211741261,
+      "grad_norm": 1.8851367235183716,
+      "learning_rate": 2.3414891841115786e-05,
+      "loss": 4.0522,
+      "step": 7440
+    },
+    {
+      "epoch": 0.8248221650197902,
+      "grad_norm": 1.945586919784546,
+      "learning_rate": 2.312943944535373e-05,
+      "loss": 4.0557,
+      "step": 7450
+    },
+    {
+      "epoch": 0.8259293088654543,
+      "grad_norm": 1.9544013738632202,
+      "learning_rate": 2.2845592297177385e-05,
+      "loss": 4.1416,
+      "step": 7460
+    },
+    {
+      "epoch": 0.8270364527111185,
+      "grad_norm": 2.053140640258789,
+      "learning_rate": 2.2563353988002756e-05,
+      "loss": 4.0475,
+      "step": 7470
+    },
+    {
+      "epoch": 0.8281435965567826,
+      "grad_norm": 1.8591804504394531,
+      "learning_rate": 2.2282728088889602e-05,
+      "loss": 4.0518,
+      "step": 7480
+    },
+    {
+      "epoch": 0.8292507404024468,
+      "grad_norm": 1.946296215057373,
+      "learning_rate": 2.2003718150496596e-05,
+      "loss": 4.1244,
+      "step": 7490
+    },
+    {
+      "epoch": 0.830357884248111,
+      "grad_norm": 2.015845537185669,
+      "learning_rate": 2.1726327703036133e-05,
+      "loss": 4.1237,
+      "step": 7500
+    },
+    {
+      "epoch": 0.8314650280937751,
+      "grad_norm": 1.7058475017547607,
+      "learning_rate": 2.145056025622981e-05,
+      "loss": 4.259,
+      "step": 7510
+    },
+    {
+      "epoch": 0.8325721719394392,
+      "grad_norm": 1.8366237878799438,
+      "learning_rate": 2.117641929926403e-05,
+      "loss": 4.1062,
+      "step": 7520
+    },
+    {
+      "epoch": 0.8336793157851033,
+      "grad_norm": 1.8169143199920654,
+      "learning_rate": 2.090390830074563e-05,
+      "loss": 4.2337,
+      "step": 7530
+    },
+    {
+      "epoch": 0.8347864596307675,
+      "grad_norm": 1.9458692073822021,
+      "learning_rate": 2.0633030708658395e-05,
+      "loss": 4.061,
+      "step": 7540
+    },
+    {
+      "epoch": 0.8358936034764317,
+      "grad_norm": 2.133040428161621,
+      "learning_rate": 2.036378995031908e-05,
+      "loss": 4.0249,
+      "step": 7550
+    },
+    {
+      "epoch": 0.8370007473220958,
+      "grad_norm": 2.043210983276367,
+      "learning_rate": 2.009618943233419e-05,
+      "loss": 4.0707,
+      "step": 7560
+    },
+    {
+      "epoch": 0.83810789116776,
+      "grad_norm": 1.99149751663208,
+      "learning_rate": 1.9830232540556907e-05,
+      "loss": 4.1641,
+      "step": 7570
+    },
+    {
+      "epoch": 0.8392150350134241,
+      "grad_norm": 2.227796792984009,
+      "learning_rate": 1.9565922640044023e-05,
+      "loss": 4.0605,
+      "step": 7580
+    },
+    {
+      "epoch": 0.8403221788590882,
+      "grad_norm": 2.0459980964660645,
+      "learning_rate": 1.930326307501383e-05,
+      "loss": 4.0796,
+      "step": 7590
+    },
+    {
+      "epoch": 0.8414293227047525,
+      "grad_norm": 1.6828839778900146,
+      "learning_rate": 1.9042257168803232e-05,
+      "loss": 4.0757,
+      "step": 7600
+    },
+    {
+      "epoch": 0.8425364665504166,
+      "grad_norm": 1.9480218887329102,
+      "learning_rate": 1.878290822382622e-05,
+      "loss": 4.0748,
+      "step": 7610
+    },
+    {
+      "epoch": 0.8436436103960807,
+      "grad_norm": 1.9197769165039062,
+      "learning_rate": 1.8525219521531693e-05,
+      "loss": 4.1489,
+      "step": 7620
+    },
+    {
+      "epoch": 0.8447507542417448,
+      "grad_norm": 1.9431424140930176,
+      "learning_rate": 1.8269194322362164e-05,
+      "loss": 4.1085,
+      "step": 7630
+    },
+    {
+      "epoch": 0.845857898087409,
+      "grad_norm": 1.9613072872161865,
+      "learning_rate": 1.8014835865712434e-05,
+      "loss": 4.1758,
+      "step": 7640
+    },
+    {
+      "epoch": 0.8469650419330732,
+      "grad_norm": 1.9461182355880737,
+      "learning_rate": 1.776214736988853e-05,
+      "loss": 3.9659,
+      "step": 7650
+    },
+    {
+      "epoch": 0.8480721857787373,
+      "grad_norm": 1.793887734413147,
+      "learning_rate": 1.7511132032067238e-05,
+      "loss": 4.1093,
+      "step": 7660
+    },
+    {
+      "epoch": 0.8491793296244015,
+      "grad_norm": 2.092958688735962,
+      "learning_rate": 1.7261793028255266e-05,
+      "loss": 4.0269,
+      "step": 7670
+    },
+    {
+      "epoch": 0.8502864734700656,
+      "grad_norm": 1.9379487037658691,
+      "learning_rate": 1.7014133513249474e-05,
+      "loss": 4.088,
+      "step": 7680
+    },
+    {
+      "epoch": 0.8513936173157297,
+      "grad_norm": 2.070305824279785,
+      "learning_rate": 1.676815662059663e-05,
+      "loss": 4.0454,
+      "step": 7690
+    },
+    {
+      "epoch": 0.852500761161394,
+      "grad_norm": 1.9571741819381714,
+      "learning_rate": 1.6523865462553854e-05,
+      "loss": 4.0874,
+      "step": 7700
+    },
+    {
+      "epoch": 0.8536079050070581,
+      "grad_norm": 1.7804341316223145,
+      "learning_rate": 1.6281263130049442e-05,
+      "loss": 4.0696,
+      "step": 7710
+    },
+    {
+      "epoch": 0.8547150488527222,
+      "grad_norm": 1.9219188690185547,
+      "learning_rate": 1.6040352692643356e-05,
+      "loss": 4.0852,
+      "step": 7720
+    },
+    {
+      "epoch": 0.8558221926983863,
+      "grad_norm": 2.016080379486084,
+      "learning_rate": 1.5801137198488817e-05,
+      "loss": 4.1547,
+      "step": 7730
+    },
+    {
+      "epoch": 0.8569293365440505,
+      "grad_norm": 2.0426931381225586,
+      "learning_rate": 1.5563619674293466e-05,
+      "loss": 4.1554,
+      "step": 7740
+    },
+    {
+      "epoch": 0.8580364803897146,
+      "grad_norm": 1.8984227180480957,
+      "learning_rate": 1.5327803125281103e-05,
+      "loss": 4.1468,
+      "step": 7750
+    },
+    {
+      "epoch": 0.8591436242353788,
+      "grad_norm": 1.7682805061340332,
+      "learning_rate": 1.509369053515378e-05,
+      "loss": 4.0504,
+      "step": 7760
+    },
+    {
+      "epoch": 0.860250768081043,
+      "grad_norm": 2.033978223800659,
+      "learning_rate": 1.486128486605383e-05,
+      "loss": 4.0684,
+      "step": 7770
+    },
+    {
+      "epoch": 0.8613579119267071,
+      "grad_norm": 2.0222108364105225,
+      "learning_rate": 1.4630589058526743e-05,
+      "loss": 4.0494,
+      "step": 7780
+    },
+    {
+      "epoch": 0.8624650557723712,
+      "grad_norm": 1.8937342166900635,
+      "learning_rate": 1.4401606031483497e-05,
+      "loss": 4.2374,
+      "step": 7790
+    },
+    {
+      "epoch": 0.8635721996180353,
+      "grad_norm": 1.8484954833984375,
+      "learning_rate": 1.4174338682164094e-05,
+      "loss": 4.1163,
+      "step": 7800
+    },
+    {
+      "epoch": 0.8646793434636996,
+      "grad_norm": 2.137838840484619,
+      "learning_rate": 1.3948789886100591e-05,
+      "loss": 4.1417,
+      "step": 7810
+    },
+    {
+      "epoch": 0.8657864873093637,
+      "grad_norm": 1.9695686101913452,
+      "learning_rate": 1.3724962497080732e-05,
+      "loss": 4.1018,
+      "step": 7820
+    },
+    {
+      "epoch": 0.8668936311550278,
+      "grad_norm": 2.1530511379241943,
+      "learning_rate": 1.3502859347112095e-05,
+      "loss": 4.0624,
+      "step": 7830
+    },
+    {
+      "epoch": 0.868000775000692,
+      "grad_norm": 1.9466749429702759,
+      "learning_rate": 1.3282483246385856e-05,
+      "loss": 4.0805,
+      "step": 7840
+    },
+    {
+      "epoch": 0.8691079188463561,
+      "grad_norm": 1.929768443107605,
+      "learning_rate": 1.306383698324171e-05,
+      "loss": 4.0613,
+      "step": 7850
+    },
+    {
+      "epoch": 0.8702150626920203,
+      "grad_norm": 2.0882503986358643,
+      "learning_rate": 1.284692332413209e-05,
+      "loss": 4.007,
+      "step": 7860
+    },
+    {
+      "epoch": 0.8713222065376844,
+      "grad_norm": 2.0295352935791016,
+      "learning_rate": 1.2631745013587602e-05,
+      "loss": 4.1551,
+      "step": 7870
+    },
+    {
+      "epoch": 0.8724293503833486,
+      "grad_norm": 3.1301565170288086,
+      "learning_rate": 1.2418304774182075e-05,
+      "loss": 4.1221,
+      "step": 7880
+    },
+    {
+      "epoch": 0.8735364942290127,
+      "grad_norm": 2.0402472019195557,
+      "learning_rate": 1.2206605306498003e-05,
+      "loss": 4.0936,
+      "step": 7890
+    },
+    {
+      "epoch": 0.8746436380746768,
+      "grad_norm": 1.7907408475875854,
+      "learning_rate": 1.1996649289092742e-05,
+      "loss": 4.1916,
+      "step": 7900
+    },
+    {
+      "epoch": 0.875750781920341,
+      "grad_norm": 1.8768945932388306,
+      "learning_rate": 1.1788439378464198e-05,
+      "loss": 4.0553,
+      "step": 7910
+    },
+    {
+      "epoch": 0.8768579257660052,
+      "grad_norm": 1.9872305393218994,
+      "learning_rate": 1.1581978209017567e-05,
+      "loss": 4.0659,
+      "step": 7920
+    },
+    {
+      "epoch": 0.8779650696116693,
+      "grad_norm": 1.8893120288848877,
+      "learning_rate": 1.1377268393031791e-05,
+      "loss": 4.0466,
+      "step": 7930
+    },
+    {
+      "epoch": 0.8790722134573334,
+      "grad_norm": 2.092783212661743,
+      "learning_rate": 1.1174312520626555e-05,
+      "loss": 4.0376,
+      "step": 7940
+    },
+    {
+      "epoch": 0.8801793573029976,
+      "grad_norm": 1.866308569908142,
+      "learning_rate": 1.0973113159729562e-05,
+      "loss": 4.1527,
+      "step": 7950
+    },
+    {
+      "epoch": 0.8812865011486617,
+      "grad_norm": 2.2415127754211426,
+      "learning_rate": 1.0773672856043914e-05,
+      "loss": 4.1154,
+      "step": 7960
+    },
+    {
+      "epoch": 0.8823936449943259,
+      "grad_norm": 1.8905448913574219,
+      "learning_rate": 1.0575994133016142e-05,
+      "loss": 4.098,
+      "step": 7970
+    },
+    {
+      "epoch": 0.8835007888399901,
+      "grad_norm": 1.9779536724090576,
+      "learning_rate": 1.0380079491803994e-05,
+      "loss": 4.1079,
+      "step": 7980
+    },
+    {
+      "epoch": 0.8846079326856542,
+      "grad_norm": 1.925347924232483,
+      "learning_rate": 1.0185931411245008e-05,
+      "loss": 4.062,
+      "step": 7990
+    },
+    {
+      "epoch": 0.8857150765313183,
+      "grad_norm": 1.9786287546157837,
+      "learning_rate": 9.993552347825078e-06,
+      "loss": 4.0755,
+      "step": 8000
+    },
+    {
+      "epoch": 0.8868222203769824,
+      "grad_norm": 2.023024082183838,
+      "learning_rate": 9.802944735647245e-06,
+      "loss": 4.1075,
+      "step": 8010
+    },
+    {
+      "epoch": 0.8879293642226467,
+      "grad_norm": 1.841030240058899,
+      "learning_rate": 9.614110986401169e-06,
+      "loss": 4.1252,
+      "step": 8020
+    },
+    {
+      "epoch": 0.8890365080683108,
+      "grad_norm": 2.0637764930725098,
+      "learning_rate": 9.427053489332304e-06,
+      "loss": 4.0517,
+      "step": 8030
+    },
+    {
+      "epoch": 0.8901436519139749,
+      "grad_norm": 1.700031042098999,
+      "learning_rate": 9.241774611211977e-06,
+      "loss": 4.067,
+      "step": 8040
+    },
+    {
+      "epoch": 0.8912507957596391,
+      "grad_norm": 1.862571358680725,
+      "learning_rate": 9.058276696307155e-06,
+      "loss": 4.1024,
+      "step": 8050
+    },
+    {
+      "epoch": 0.8923579396053032,
+      "grad_norm": 1.9017775058746338,
+      "learning_rate": 8.876562066351023e-06,
+      "loss": 4.0398,
+      "step": 8060
+    },
+    {
+      "epoch": 0.8934650834509673,
+      "grad_norm": 2.0224740505218506,
+      "learning_rate": 8.696633020513455e-06,
+      "loss": 4.0946,
+      "step": 8070
+    },
+    {
+      "epoch": 0.8945722272966316,
+      "grad_norm": 2.2029504776000977,
+      "learning_rate": 8.518491835371926e-06,
+      "loss": 4.0046,
+      "step": 8080
+    },
+    {
+      "epoch": 0.8956793711422957,
+      "grad_norm": 2.005601406097412,
+      "learning_rate": 8.342140764882876e-06,
+      "loss": 4.0866,
+      "step": 8090
+    },
+    {
+      "epoch": 0.8967865149879598,
+      "grad_norm": 2.0867319107055664,
+      "learning_rate": 8.16758204035291e-06,
+      "loss": 4.0695,
+      "step": 8100
+    },
+    {
+      "epoch": 0.8978936588336239,
+      "grad_norm": 1.722353458404541,
+      "learning_rate": 7.994817870410858e-06,
+      "loss": 4.1669,
+      "step": 8110
+    },
+    {
+      "epoch": 0.8990008026792881,
+      "grad_norm": 2.013004779815674,
+      "learning_rate": 7.823850440979645e-06,
+      "loss": 4.1026,
+      "step": 8120
+    },
+    {
+      "epoch": 0.9001079465249523,
+      "grad_norm": 2.0718815326690674,
+      "learning_rate": 7.65468191524865e-06,
+      "loss": 4.0673,
+      "step": 8130
+    },
+    {
+      "epoch": 0.9007722328323507,
+      "eval_loss": 4.557027339935303,
+      "eval_runtime": 49.1762,
+      "eval_samples_per_second": 1016.752,
+      "eval_steps_per_second": 127.094,
+      "step": 8136
+    },
+    {
+      "epoch": 0.9012150903706164,
+      "grad_norm": 1.9817490577697754,
+      "learning_rate": 7.487314433646441e-06,
+      "loss": 4.0481,
+      "step": 8140
+    },
+    {
+      "epoch": 0.9023222342162805,
+      "grad_norm": 1.7491908073425293,
+      "learning_rate": 7.321750113813546e-06,
+      "loss": 4.0959,
+      "step": 8150
+    },
+    {
+      "epoch": 0.9034293780619447,
+      "grad_norm": 1.9322396516799927,
+      "learning_rate": 7.1579910505758285e-06,
+      "loss": 4.0221,
+      "step": 8160
+    },
+    {
+      "epoch": 0.9045365219076088,
+      "grad_norm": 1.9864757061004639,
+      "learning_rate": 6.996039315917756e-06,
+      "loss": 4.1734,
+      "step": 8170
+    },
+    {
+      "epoch": 0.905643665753273,
+      "grad_norm": 1.9304691553115845,
+      "learning_rate": 6.8358969589564715e-06,
+      "loss": 4.099,
+      "step": 8180
+    },
+    {
+      "epoch": 0.9067508095989372,
+      "grad_norm": 1.7734984159469604,
+      "learning_rate": 6.6775660059156295e-06,
+      "loss": 4.0798,
+      "step": 8190
+    },
+    {
+      "epoch": 0.9078579534446013,
+      "grad_norm": 1.909828543663025,
+      "learning_rate": 6.521048460099787e-06,
+      "loss": 4.0966,
+      "step": 8200
+    },
+    {
+      "epoch": 0.9089650972902654,
+      "grad_norm": 2.0267088413238525,
+      "learning_rate": 6.366346301869252e-06,
+      "loss": 4.0239,
+      "step": 8210
+    },
+    {
+      "epoch": 0.9100722411359295,
+      "grad_norm": 1.8887717723846436,
+      "learning_rate": 6.213461488614741e-06,
+      "loss": 4.0547,
+      "step": 8220
+    },
+    {
+      "epoch": 0.9111793849815938,
+      "grad_norm": 2.1588733196258545,
+      "learning_rate": 6.062395954732879e-06,
+      "loss": 4.0832,
+      "step": 8230
+    },
+    {
+      "epoch": 0.9122865288272579,
+      "grad_norm": 1.990921974182129,
+      "learning_rate": 5.913151611601541e-06,
+      "loss": 4.0556,
+      "step": 8240
+    },
+    {
+      "epoch": 0.913393672672922,
+      "grad_norm": 2.075512409210205,
+      "learning_rate": 5.7657303475556974e-06,
+      "loss": 4.0725,
+      "step": 8250
+    },
+    {
+      "epoch": 0.9145008165185862,
+      "grad_norm": 1.87040376663208,
+      "learning_rate": 5.620134027863671e-06,
+      "loss": 4.0631,
+      "step": 8260
+    },
+    {
+      "epoch": 0.9156079603642503,
+      "grad_norm": 1.8812203407287598,
+      "learning_rate": 5.476364494703256e-06,
+      "loss": 4.0176,
+      "step": 8270
+    },
+    {
+      "epoch": 0.9167151042099144,
+      "grad_norm": 1.8779735565185547,
+      "learning_rate": 5.334423567138734e-06,
+      "loss": 4.0674,
+      "step": 8280
+    },
+    {
+      "epoch": 0.9178222480555787,
+      "grad_norm": 2.3180549144744873,
+      "learning_rate": 5.1943130410975805e-06,
+      "loss": 4.0342,
+      "step": 8290
+    },
+    {
+      "epoch": 0.9189293919012428,
+      "grad_norm": 1.9171079397201538,
+      "learning_rate": 5.056034689347927e-06,
+      "loss": 4.1577,
+      "step": 8300
+    },
+    {
+      "epoch": 0.9200365357469069,
+      "grad_norm": 1.9183543920516968,
+      "learning_rate": 4.919590261476047e-06,
+      "loss": 4.0529,
+      "step": 8310
+    },
+    {
+      "epoch": 0.921143679592571,
+      "grad_norm": 1.8293943405151367,
+      "learning_rate": 4.784981483864198e-06,
+      "loss": 4.068,
+      "step": 8320
+    },
+    {
+      "epoch": 0.9222508234382352,
+      "grad_norm": 2.3358635902404785,
+      "learning_rate": 4.652210059668893e-06,
+      "loss": 4.0771,
+      "step": 8330
+    },
+    {
+      "epoch": 0.9233579672838994,
+      "grad_norm": 1.86179518699646,
+      "learning_rate": 4.521277668799144e-06,
+      "loss": 4.1162,
+      "step": 8340
+    },
+    {
+      "epoch": 0.9244651111295635,
+      "grad_norm": 2.0704479217529297,
+      "learning_rate": 4.392185967895495e-06,
+      "loss": 4.0849,
+      "step": 8350
+    },
+    {
+      "epoch": 0.9255722549752277,
+      "grad_norm": 1.9911187887191772,
+      "learning_rate": 4.264936590308782e-06,
+      "loss": 4.0519,
+      "step": 8360
+    },
+    {
+      "epoch": 0.9266793988208918,
+      "grad_norm": 1.897602915763855,
+      "learning_rate": 4.139531146079622e-06,
+      "loss": 4.0873,
+      "step": 8370
+    },
+    {
+      "epoch": 0.9277865426665559,
+      "grad_norm": 2.2255215644836426,
+      "learning_rate": 4.015971221918046e-06,
+      "loss": 4.0185,
+      "step": 8380
+    },
+    {
+      "epoch": 0.9288936865122202,
+      "grad_norm": 1.7892693281173706,
+      "learning_rate": 3.894258381183313e-06,
+      "loss": 4.1864,
+      "step": 8390
+    },
+    {
+      "epoch": 0.9300008303578843,
+      "grad_norm": 2.5293076038360596,
+      "learning_rate": 3.774394163864258e-06,
+      "loss": 4.1182,
+      "step": 8400
+    },
+    {
+      "epoch": 0.9311079742035484,
+      "grad_norm": 1.876760482788086,
+      "learning_rate": 3.6563800865596794e-06,
+      "loss": 4.1062,
+      "step": 8410
+    },
+    {
+      "epoch": 0.9322151180492125,
+      "grad_norm": 2.0664350986480713,
+      "learning_rate": 3.540217642459331e-06,
+      "loss": 4.0941,
+      "step": 8420
+    },
+    {
+      "epoch": 0.9333222618948767,
+      "grad_norm": 1.984215259552002,
+      "learning_rate": 3.4259083013248257e-06,
+      "loss": 4.0387,
+      "step": 8430
+    },
+    {
+      "epoch": 0.9344294057405408,
+      "grad_norm": 2.0975677967071533,
+      "learning_rate": 3.3134535094711146e-06,
+      "loss": 4.0539,
+      "step": 8440
+    },
+    {
+      "epoch": 0.935536549586205,
+      "grad_norm": 1.8716676235198975,
+      "learning_rate": 3.2028546897483186e-06,
+      "loss": 4.1301,
+      "step": 8450
+    },
+    {
+      "epoch": 0.9366436934318692,
+      "grad_norm": 1.962329626083374,
+      "learning_rate": 3.0941132415234602e-06,
+      "loss": 4.1131,
+      "step": 8460
+    },
+    {
+      "epoch": 0.9377508372775333,
+      "grad_norm": 1.8207136392593384,
+      "learning_rate": 2.98723054066306e-06,
+      "loss": 4.1021,
+      "step": 8470
+    },
+    {
+      "epoch": 0.9388579811231974,
+      "grad_norm": 2.1753087043762207,
+      "learning_rate": 2.882207939515435e-06,
+      "loss": 4.0178,
+      "step": 8480
+    },
+    {
+      "epoch": 0.9399651249688615,
+      "grad_norm": 2.040548086166382,
+      "learning_rate": 2.7790467668938277e-06,
+      "loss": 4.1275,
+      "step": 8490
+    },
+    {
+      "epoch": 0.9410722688145258,
+      "grad_norm": 2.1729607582092285,
+      "learning_rate": 2.677748328059487e-06,
+      "loss": 4.1518,
+      "step": 8500
+    },
+    {
+      "epoch": 0.9421794126601899,
+      "grad_norm": 1.7103227376937866,
+      "learning_rate": 2.578313904705098e-06,
+      "loss": 4.1542,
+      "step": 8510
+    },
+    {
+      "epoch": 0.943286556505854,
+      "grad_norm": 1.8410452604293823,
+      "learning_rate": 2.480744754938696e-06,
+      "loss": 3.9978,
+      "step": 8520
+    },
+    {
+      "epoch": 0.9443937003515182,
+      "grad_norm": 1.8805840015411377,
+      "learning_rate": 2.38504211326761e-06,
+      "loss": 4.1473,
+      "step": 8530
+    },
+    {
+      "epoch": 0.9455008441971823,
+      "grad_norm": 1.840846300125122,
+      "learning_rate": 2.291207190583011e-06,
+      "loss": 4.093,
+      "step": 8540
+    },
+    {
+      "epoch": 0.9466079880428465,
+      "grad_norm": 1.9069963693618774,
+      "learning_rate": 2.1992411741443896e-06,
+      "loss": 4.1269,
+      "step": 8550
+    },
+    {
+      "epoch": 0.9477151318885106,
+      "grad_norm": 1.8701353073120117,
+      "learning_rate": 2.1091452275647013e-06,
+      "loss": 4.1305,
+      "step": 8560
+    },
+    {
+      "epoch": 0.9488222757341748,
+      "grad_norm": 2.0695340633392334,
+      "learning_rate": 2.02092049079558e-06,
+      "loss": 4.1921,
+      "step": 8570
+    },
+    {
+      "epoch": 0.9499294195798389,
+      "grad_norm": 1.9630610942840576,
+      "learning_rate": 1.9345680801128806e-06,
+      "loss": 4.069,
+      "step": 8580
+    },
+    {
+      "epoch": 0.951036563425503,
+      "grad_norm": 1.7409827709197998,
+      "learning_rate": 1.8500890881026076e-06,
+      "loss": 4.1502,
+      "step": 8590
+    },
+    {
+      "epoch": 0.9521437072711673,
+      "grad_norm": 1.9596493244171143,
+      "learning_rate": 1.7674845836470609e-06,
+      "loss": 4.1767,
+      "step": 8600
+    },
+    {
+      "epoch": 0.9532508511168314,
+      "grad_norm": 1.9688855409622192,
+      "learning_rate": 1.6867556119113612e-06,
+      "loss": 4.1115,
+      "step": 8610
+    },
+    {
+      "epoch": 0.9543579949624955,
+      "grad_norm": 1.7361944913864136,
+      "learning_rate": 1.6079031943301623e-06,
+      "loss": 4.0089,
+      "step": 8620
+    },
+    {
+      "epoch": 0.9554651388081596,
+      "grad_norm": 1.8814979791641235,
+      "learning_rate": 1.5309283285947271e-06,
+      "loss": 4.0974,
+      "step": 8630
+    },
+    {
+      "epoch": 0.9565722826538238,
+      "grad_norm": 1.8453491926193237,
+      "learning_rate": 1.4558319886403714e-06,
+      "loss": 4.0013,
+      "step": 8640
+    },
+    {
+      "epoch": 0.9576794264994879,
+      "grad_norm": 1.8325161933898926,
+      "learning_rate": 1.3826151246340566e-06,
+      "loss": 4.1164,
+      "step": 8650
+    },
+    {
+      "epoch": 0.9587865703451521,
+      "grad_norm": 2.1655688285827637,
+      "learning_rate": 1.311278662962484e-06,
+      "loss": 4.1158,
+      "step": 8660
+    },
+    {
+      "epoch": 0.9598937141908163,
+      "grad_norm": 2.060488700866699,
+      "learning_rate": 1.2418235062202187e-06,
+      "loss": 4.112,
+      "step": 8670
+    },
+    {
+      "epoch": 0.9610008580364804,
+      "grad_norm": 2.0091753005981445,
+      "learning_rate": 1.174250533198401e-06,
+      "loss": 4.0838,
+      "step": 8680
+    },
+    {
+      "epoch": 0.9621080018821445,
+      "grad_norm": 2.1044318675994873,
+      "learning_rate": 1.1085605988735536e-06,
+      "loss": 4.0997,
+      "step": 8690
+    },
+    {
+      "epoch": 0.9632151457278086,
+      "grad_norm": 2.0191855430603027,
+      "learning_rate": 1.0447545343968078e-06,
+      "loss": 4.1029,
+      "step": 8700
+    },
+    {
+      "epoch": 0.9643222895734729,
+      "grad_norm": 2.0016067028045654,
+      "learning_rate": 9.82833147083345e-07,
+      "loss": 4.0798,
+      "step": 8710
+    },
+    {
+      "epoch": 0.965429433419137,
+      "grad_norm": 1.8061256408691406,
+      "learning_rate": 9.22797220402205e-07,
+      "loss": 4.1468,
+      "step": 8720
+    },
+    {
+      "epoch": 0.9665365772648011,
+      "grad_norm": 2.0113728046417236,
+      "learning_rate": 8.646475139664266e-07,
+      "loss": 4.1141,
+      "step": 8730
+    },
+    {
+      "epoch": 0.9676437211104653,
+      "grad_norm": 2.046168804168701,
+      "learning_rate": 8.083847635232732e-07,
+      "loss": 4.0762,
+      "step": 8740
+    },
+    {
+      "epoch": 0.9687508649561294,
+      "grad_norm": 1.9780277013778687,
+      "learning_rate": 7.54009680945139e-07,
+      "loss": 4.0597,
+      "step": 8750
+    },
+    {
+      "epoch": 0.9698580088017936,
+      "grad_norm": 2.018071174621582,
+      "learning_rate": 7.015229542203904e-07,
+      "loss": 4.0509,
+      "step": 8760
+    },
+    {
+      "epoch": 0.9709651526474578,
+      "grad_norm": 2.0865936279296875,
+      "learning_rate": 6.509252474446891e-07,
+      "loss": 4.1356,
+      "step": 8770
+    },
+    {
+      "epoch": 0.9720722964931219,
+      "grad_norm": 1.813429832458496,
+      "learning_rate": 6.022172008126658e-07,
+      "loss": 4.1577,
+      "step": 8780
+    },
+    {
+      "epoch": 0.973179440338786,
+      "grad_norm": 1.8943582773208618,
+      "learning_rate": 5.553994306096765e-07,
+      "loss": 4.0896,
+      "step": 8790
+    },
+    {
+      "epoch": 0.9742865841844501,
+      "grad_norm": 1.9367841482162476,
+      "learning_rate": 5.104725292041423e-07,
+      "loss": 4.0649,
+      "step": 8800
+    },
+    {
+      "epoch": 0.9753937280301143,
+      "grad_norm": 1.9672105312347412,
+      "learning_rate": 4.674370650400217e-07,
+      "loss": 4.18,
+      "step": 8810
+    },
+    {
+      "epoch": 0.9765008718757785,
+      "grad_norm": 1.857710838317871,
+      "learning_rate": 4.2629358262953327e-07,
+      "loss": 4.0796,
+      "step": 8820
+    },
+    {
+      "epoch": 0.9776080157214426,
+      "grad_norm": 1.991847038269043,
+      "learning_rate": 3.870426025463946e-07,
+      "loss": 4.1273,
+      "step": 8830
+    },
+    {
+      "epoch": 0.9787151595671068,
+      "grad_norm": 1.8746367692947388,
+      "learning_rate": 3.496846214191107e-07,
+      "loss": 4.1366,
+      "step": 8840
+    },
+    {
+      "epoch": 0.9798223034127709,
+      "grad_norm": 1.9193933010101318,
+      "learning_rate": 3.142201119248289e-07,
+      "loss": 4.1451,
+      "step": 8850
+    },
+    {
+      "epoch": 0.980929447258435,
+      "grad_norm": 2.017108201980591,
+      "learning_rate": 2.8064952278321063e-07,
+      "loss": 4.0247,
+      "step": 8860
+    },
+    {
+      "epoch": 0.9820365911040992,
+      "grad_norm": 1.8040502071380615,
+      "learning_rate": 2.489732787508525e-07,
+      "loss": 4.0401,
+      "step": 8870
+    },
+    {
+      "epoch": 0.9831437349497634,
+      "grad_norm": 1.9902021884918213,
+      "learning_rate": 2.1919178061589048e-07,
+      "loss": 4.1217,
+      "step": 8880
+    },
+    {
+      "epoch": 0.9842508787954275,
+      "grad_norm": 1.8049163818359375,
+      "learning_rate": 1.913054051928875e-07,
+      "loss": 4.1517,
+      "step": 8890
+    },
+    {
+      "epoch": 0.9853580226410916,
+      "grad_norm": 1.8805010318756104,
+      "learning_rate": 1.653145053181537e-07,
+      "loss": 4.0831,
+      "step": 8900
+    },
+    {
+      "epoch": 0.9864651664867558,
+      "grad_norm": 1.948275089263916,
+      "learning_rate": 1.412194098451669e-07,
+      "loss": 4.0646,
+      "step": 8910
+    },
+    {
+      "epoch": 0.98757231033242,
+      "grad_norm": 1.9337522983551025,
+      "learning_rate": 1.1902042364054232e-07,
+      "loss": 4.1404,
+      "step": 8920
+    },
+    {
+      "epoch": 0.9886794541780841,
+      "grad_norm": 1.8786277770996094,
+      "learning_rate": 9.871782758005264e-08,
+      "loss": 4.1386,
+      "step": 8930
+    },
+    {
+      "epoch": 0.9897865980237482,
+      "grad_norm": 2.0808181762695312,
+      "learning_rate": 8.031187854514731e-08,
+      "loss": 4.1307,
+      "step": 8940
+    },
+    {
+      "epoch": 0.9908937418694124,
+      "grad_norm": 1.9702550172805786,
+      "learning_rate": 6.380280941968852e-08,
+      "loss": 4.0896,
+      "step": 8950
+    },
+    {
+      "epoch": 0.9920008857150765,
+      "grad_norm": 2.2030298709869385,
+      "learning_rate": 4.9190829086986904e-08,
+      "loss": 4.1014,
+      "step": 8960
+    },
+    {
+      "epoch": 0.9931080295607406,
+      "grad_norm": 1.966364860534668,
+      "learning_rate": 3.647612242717035e-08,
+      "loss": 4.1217,
+      "step": 8970
+    },
+    {
+      "epoch": 0.9942151734064049,
+      "grad_norm": 1.9965739250183105,
+      "learning_rate": 2.5658850314852486e-08,
+      "loss": 4.172,
+      "step": 8980
+    },
+    {
+      "epoch": 0.995322317252069,
+      "grad_norm": 2.0662567615509033,
+      "learning_rate": 1.673914961710099e-08,
+      "loss": 4.0832,
+      "step": 8990
+    },
+    {
+      "epoch": 0.9964294610977331,
+      "grad_norm": 1.8497223854064941,
+      "learning_rate": 9.717133191672332e-09,
+      "loss": 4.0676,
+      "step": 9000
+    },
+    {
+      "epoch": 0.9975366049433972,
+      "grad_norm": 1.814286708831787,
+      "learning_rate": 4.592889885596252e-09,
+      "loss": 4.0644,
+      "step": 9010
+    },
+    {
+      "epoch": 0.9986437487890614,
+      "grad_norm": 2.3788208961486816,
+      "learning_rate": 1.366484534109924e-09,
+      "loss": 3.8724,
+      "step": 9020
+    },
+    {
+      "epoch": 0.9997508926347256,
+      "grad_norm": 1.827810287475586,
+      "learning_rate": 3.795795975869431e-11,
+      "loss": 4.042,
+      "step": 9030
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 9032,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 904,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1268614137139200.0,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}